[PATCH 00/35] Btrfs for review

[PATCH 00/35] Btrfs for review

[Chris Mason]

Hello everyone,

Here are the btrfs files, grouped as .[ch] combo patches for review.  Many
of these were heavily contributed to or written by others, please see the git
repo for full details.

git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable.git

For actually running the sources, downloading from the git repo will
work better, but hopefully this will be easier for people to review
and ask questions about the code.

[PATCH 01/35] Btrfs xattr code

[Chris Mason]

Xattrs in btrfs are stored in the btree close to the inode item.  They use an
indexing scheme very similar to directories.

The max size of a btrfs xattr is limited by the size of a btree item,
or about 4k.  A future version will support larger items.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/xattr.c b/fs/btrfs/xattr.c
new file mode 100644
index 0000000..7f332e2
--- /dev/null
+++ b/fs/btrfs/xattr.c
@@ -0,0 +1,322 @@
+/*
+ * Copyright (C) 2007 Red Hat.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/init.h>
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/rwsem.h>
+#include <linux/xattr.h>
+#include "ctree.h"
+#include "btrfs_inode.h"
+#include "transaction.h"
+#include "xattr.h"
+#include "disk-io.h"
+
+
+ssize_t __btrfs_getxattr(struct inode *inode, const char *name,
+				void *buffer, size_t size)
+{
+	struct btrfs_dir_item *di;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	int ret = 0;
+	unsigned long data_ptr;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	/* lookup the xattr by name */
+	di = btrfs_lookup_xattr(NULL, root, path, inode->i_ino, name,
+				strlen(name), 0);
+	if (!di || IS_ERR(di)) {
+		ret = -ENODATA;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	/* if size is 0, that means we want the size of the attr */
+	if (!size) {
+		ret = btrfs_dir_data_len(leaf, di);
+		goto out;
+	}
+
+	/* now get the data out of our dir_item */
+	if (btrfs_dir_data_len(leaf, di) > size) {
+		ret = -ERANGE;
+		goto out;
+	}
+	data_ptr = (unsigned long)((char *)(di + 1) +
+				   btrfs_dir_name_len(leaf, di));
+	read_extent_buffer(leaf, buffer, data_ptr,
+			   btrfs_dir_data_len(leaf, di));
+	ret = btrfs_dir_data_len(leaf, di);
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int __btrfs_setxattr(struct inode *inode, const char *name,
+			    const void *value, size_t size, int flags)
+{
+	struct btrfs_dir_item *di;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_path *path;
+	int ret = 0, mod = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
+
+	/* first lets see if we already have this xattr */
+	di = btrfs_lookup_xattr(trans, root, path, inode->i_ino, name,
+				strlen(name), -1);
+	if (IS_ERR(di)) {
+		ret = PTR_ERR(di);
+		goto out;
+	}
+
+	/* ok we already have this xattr, lets remove it */
+	if (di) {
+		/* if we want create only exit */
+		if (flags & XATTR_CREATE) {
+			ret = -EEXIST;
+			goto out;
+		}
+
+		ret = btrfs_delete_one_dir_name(trans, root, path, di);
+		if (ret)
+			goto out;
+		btrfs_release_path(root, path);
+
+		/* if we don't have a value then we are removing the xattr */
+		if (!value) {
+			mod = 1;
+			goto out;
+		}
+	} else {
+		btrfs_release_path(root, path);
+
+		if (flags & XATTR_REPLACE) {
+			/* we couldn't find the attr to replace */
+			ret = -ENODATA;
+			goto out;
+		}
+	}
+
+	/* ok we have to create a completely new xattr */
+	ret = btrfs_insert_xattr_item(trans, root, name, strlen(name),
+				      value, size, inode->i_ino);
+	if (ret)
+		goto out;
+	mod = 1;
+
+out:
+	if (mod) {
+		inode->i_ctime = CURRENT_TIME;
+		ret = btrfs_update_inode(trans, root, inode);
+	}
+
+	btrfs_end_transaction(trans, root);
+	btrfs_free_path(path);
+	return ret;
+}
+
+ssize_t btrfs_listxattr(struct dentry *dentry, char *buffer, size_t size)
+{
+	struct btrfs_key key, found_key;
+	struct inode *inode = dentry->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_path *path;
+	struct btrfs_item *item;
+	struct extent_buffer *leaf;
+	struct btrfs_dir_item *di;
+	int ret = 0, slot, advance;
+	size_t total_size = 0, size_left = size;
+	unsigned long name_ptr;
+	size_t name_len;
+	u32 nritems;
+
+	/*
+	 * ok we want all objects associated with this id.
+	 * NOTE: we set key.offset = 0; because we want to start with the
+	 * first xattr that we find and walk forward
+	 */
+	key.objectid = inode->i_ino;
+	btrfs_set_key_type(&key, BTRFS_XATTR_ITEM_KEY);
+	key.offset = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	path->reada = 2;
+
+	/* search for our xattrs */
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto err;
+	ret = 0;
+	advance = 0;
+	while (1) {
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		slot = path->slots[0];
+
+		/* this is where we start walking through the path */
+		if (advance || slot >= nritems) {
+			/*
+			 * if we've reached the last slot in this leaf we need
+			 * to go to the next leaf and reset everything
+			 */
+			if (slot >= nritems-1) {
+				ret = btrfs_next_leaf(root, path);
+				if (ret)
+					break;
+				leaf = path->nodes[0];
+				nritems = btrfs_header_nritems(leaf);
+				slot = path->slots[0];
+			} else {
+				/*
+				 * just walking through the slots on this leaf
+				 */
+				slot++;
+				path->slots[0]++;
+			}
+		}
+		advance = 1;
+
+		item = btrfs_item_nr(leaf, slot);
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+
+		/* check to make sure this item is what we want */
+		if (found_key.objectid != key.objectid)
+			break;
+		if (btrfs_key_type(&found_key) != BTRFS_XATTR_ITEM_KEY)
+			break;
+
+		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
+
+		name_len = btrfs_dir_name_len(leaf, di);
+		total_size += name_len + 1;
+
+		/* we are just looking for how big our buffer needs to be */
+		if (!size)
+			continue;
+
+		if (!buffer || (name_len + 1) > size_left) {
+			ret = -ERANGE;
+			goto err;
+		}
+
+		name_ptr = (unsigned long)(di + 1);
+		read_extent_buffer(leaf, buffer, name_ptr, name_len);
+		buffer[name_len] = '\0';
+
+		size_left -= name_len + 1;
+		buffer += name_len + 1;
+	}
+	ret = total_size;
+
+err:
+	btrfs_free_path(path);
+
+	return ret;
+}
+
+/*
+ * List of handlers for synthetic system.* attributes.  All real ondisk
+ * attributes are handled directly.
+ */
+struct xattr_handler *btrfs_xattr_handlers[] = {
+#ifdef CONFIG_FS_POSIX_ACL
+	&btrfs_xattr_acl_access_handler,
+	&btrfs_xattr_acl_default_handler,
+#endif
+	NULL,
+};
+
+/*
+ * Check if the attribute is in a supported namespace.
+ *
+ * This applied after the check for the synthetic attributes in the system
+ * namespace.
+ */
+static bool btrfs_is_valid_xattr(const char *name)
+{
+	return !strncmp(name, XATTR_SECURITY_PREFIX,
+			XATTR_SECURITY_PREFIX_LEN) ||
+	       !strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) ||
+	       !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
+	       !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
+}
+
+ssize_t btrfs_getxattr(struct dentry *dentry, const char *name,
+		       void *buffer, size_t size)
+{
+	/*
+	 * If this is a request for a synthetic attribute in the system.*
+	 * namespace use the generic infrastructure to resolve a handler
+	 * for it via sb->s_xattr.
+	 */
+	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
+		return generic_getxattr(dentry, name, buffer, size);
+
+	if (!btrfs_is_valid_xattr(name))
+		return -EOPNOTSUPP;
+	return __btrfs_getxattr(dentry->d_inode, name, buffer, size);
+}
+
+int btrfs_setxattr(struct dentry *dentry, const char *name, const void *value,
+		   size_t size, int flags)
+{
+	/*
+	 * If this is a request for a synthetic attribute in the system.*
+	 * namespace use the generic infrastructure to resolve a handler
+	 * for it via sb->s_xattr.
+	 */
+	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
+		return generic_setxattr(dentry, name, value, size, flags);
+
+	if (!btrfs_is_valid_xattr(name))
+		return -EOPNOTSUPP;
+
+	if (size == 0)
+		value = "";  /* empty EA, do not remove */
+	return __btrfs_setxattr(dentry->d_inode, name, value, size, flags);
+}
+
+int btrfs_removexattr(struct dentry *dentry, const char *name)
+{
+	/*
+	 * If this is a request for a synthetic attribute in the system.*
+	 * namespace use the generic infrastructure to resolve a handler
+	 * for it via sb->s_xattr.
+	 */
+	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
+		return generic_removexattr(dentry, name);
+
+	if (!btrfs_is_valid_xattr(name))
+		return -EOPNOTSUPP;
+	return __btrfs_setxattr(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
+}
diff --git a/fs/btrfs/xattr.h b/fs/btrfs/xattr.h
new file mode 100644
index 0000000..5b1d08f
--- /dev/null
+++ b/fs/btrfs/xattr.h
@@ -0,0 +1,39 @@
+/*
+ * Copyright (C) 2007 Red Hat.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __XATTR__
+#define __XATTR__
+
+#include <linux/xattr.h>
+
+extern struct xattr_handler btrfs_xattr_acl_access_handler;
+extern struct xattr_handler btrfs_xattr_acl_default_handler;
+extern struct xattr_handler *btrfs_xattr_handlers[];
+
+extern ssize_t __btrfs_getxattr(struct inode *inode, const char *name,
+		void *buffer, size_t size);
+extern int __btrfs_setxattr(struct inode *inode, const char *name,
+		const void *value, size_t size, int flags);
+
+extern ssize_t btrfs_getxattr(struct dentry *dentry, const char *name,
+		void *buffer, size_t size);
+extern int btrfs_setxattr(struct dentry *dentry, const char *name,
+		const void *value, size_t size, int flags);
+extern int btrfs_removexattr(struct dentry *dentry, const char *name);
+
+#endif /* __XATTR__ */
-- 
1.6.0.2

Re: [PATCH 01/35] Btrfs xattr code

[Andrew Morton]

On Wed,  7 Jan 2009 22:56:51 -0500 Chris Mason <chris.mason@oracle.com> wrote:

> Xattrs in btrfs are stored in the btree close to the inode item.  They use an
> indexing scheme very similar to directories.
> 
> The max size of a btrfs xattr is limited by the size of a btree item,
> or about 4k.  A future version will support larger items.
> 
>
> ...
>
> +ssize_t __btrfs_getxattr(struct inode *inode, const char *name,
> +				void *buffer, size_t size)
> +{
> +	struct btrfs_dir_item *di;
> +	struct btrfs_root *root = BTRFS_I(inode)->root;
> +	struct btrfs_path *path;
> +	struct extent_buffer *leaf;
> +	int ret = 0;
> +	unsigned long data_ptr;
> +
> +	path = btrfs_alloc_path();
> +	if (!path)
> +		return -ENOMEM;
> +
> +	/* lookup the xattr by name */
> +	di = btrfs_lookup_xattr(NULL, root, path, inode->i_ino, name,
> +				strlen(name), 0);
> +	if (!di || IS_ERR(di)) {

This seems quite bad.  An error return from btrfs_lookup_xattr() can
arise from (I assume) OOM, corrupted filesystem, I/O error, etc.

Yet here we treat such errors as being the same as "not found".  This
seems grossly misleading to userspace and possible a security problem?

(I'm kind of guessing here, and might be wrong and wasting everyone's
time, because the semantics of the btrfs_lookup_xattr() and
btrfs_getxattr() return values aren't documented).

> +		ret = -ENODATA;
> +		goto out;
> +	}
> +
> +	leaf = path->nodes[0];
> +	/* if size is 0, that means we want the size of the attr */
> +	if (!size) {
> +		ret = btrfs_dir_data_len(leaf, di);
> +		goto out;
> +	}
> +
> +	/* now get the data out of our dir_item */
> +	if (btrfs_dir_data_len(leaf, di) > size) {

OK, where did we hide the btrfs_dir_data_len() definition?

> +		ret = -ERANGE;
> +		goto out;
> +	}
> +	data_ptr = (unsigned long)((char *)(di + 1) +
> +				   btrfs_dir_name_len(leaf, di));
> +	read_extent_buffer(leaf, buffer, data_ptr,
> +			   btrfs_dir_data_len(leaf, di));

Please document read_extent_buffer().

Please rename read_extent_buffer() to something more appropriate to a
kernel-wide symbol.

<spies map_private_extent_buffer as well>

Please review all kernel-wide identifiers in btrfs for appropriateness.

I'm scratching my head wondering about this `data_ptr' thing.  Is it a
disk offset?  Is it really a pointer to kernel memory?  According to
this code it is indeed a kernel pointer, but it then gets stuffed into
an unsigned long (wtf?) and then passed to the mysterious
read_extent_buffer().

<reviewer throws in the towel on this part of the code>

> +	ret = btrfs_dir_data_len(leaf, di);
> +
> +out:
> +	btrfs_free_path(path);
> +	return ret;
> +}
> +
> +int __btrfs_setxattr(struct inode *inode, const char *name,
> +			    const void *value, size_t size, int flags)
> +{
> +	struct btrfs_dir_item *di;
> +	struct btrfs_root *root = BTRFS_I(inode)->root;
> +	struct btrfs_trans_handle *trans;
> +	struct btrfs_path *path;
> +	int ret = 0, mod = 0;
> +
> +	path = btrfs_alloc_path();
> +	if (!path)
> +		return -ENOMEM;
> +
> +	trans = btrfs_start_transaction(root, 1);
> +	btrfs_set_trans_block_group(trans, inode);
> +
> +	/* first lets see if we already have this xattr */

let's :)

> +	di = btrfs_lookup_xattr(trans, root, path, inode->i_ino, name,
> +				strlen(name), -1);

<wonders what the -1 does>

<goes to the btrfs_lookup_xattr() definition site>

<towel goes flying again>

> +	if (IS_ERR(di)) {
> +		ret = PTR_ERR(di);
> +		goto out;
> +	}
> +
> +	/* ok we already have this xattr, lets remove it */
> +	if (di) {
> +		/* if we want create only exit */
> +		if (flags & XATTR_CREATE) {
> +			ret = -EEXIST;
> +			goto out;
> +		}
> +
> +		ret = btrfs_delete_one_dir_name(trans, root, path, di);
> +		if (ret)
> +			goto out;
> +		btrfs_release_path(root, path);
> +
> +		/* if we don't have a value then we are removing the xattr */
> +		if (!value) {
> +			mod = 1;
> +			goto out;
> +		}
> +	} else {
> +		btrfs_release_path(root, path);
> +
> +		if (flags & XATTR_REPLACE) {
> +			/* we couldn't find the attr to replace */
> +			ret = -ENODATA;
> +			goto out;
> +		}
> +	}
> +
> +	/* ok we have to create a completely new xattr */
> +	ret = btrfs_insert_xattr_item(trans, root, name, strlen(name),
> +				      value, size, inode->i_ino);
> +	if (ret)
> +		goto out;
> +	mod = 1;
> +
> +out:
> +	if (mod) {
> +		inode->i_ctime = CURRENT_TIME;
> +		ret = btrfs_update_inode(trans, root, inode);
> +	}
> +
> +	btrfs_end_transaction(trans, root);
> +	btrfs_free_path(path);
> +	return ret;
> +}
> +
> +ssize_t btrfs_listxattr(struct dentry *dentry, char *buffer, size_t size)
> +{
> +	struct btrfs_key key, found_key;
> +	struct inode *inode = dentry->d_inode;
> +	struct btrfs_root *root = BTRFS_I(inode)->root;
> +	struct btrfs_path *path;
> +	struct btrfs_item *item;
> +	struct extent_buffer *leaf;
> +	struct btrfs_dir_item *di;
> +	int ret = 0, slot, advance;
> +	size_t total_size = 0, size_left = size;
> +	unsigned long name_ptr;
> +	size_t name_len;
> +	u32 nritems;
> +
> +	/*
> +	 * ok we want all objects associated with this id.
> +	 * NOTE: we set key.offset = 0; because we want to start with the
> +	 * first xattr that we find and walk forward
> +	 */
> +	key.objectid = inode->i_ino;
> +	btrfs_set_key_type(&key, BTRFS_XATTR_ITEM_KEY);
> +	key.offset = 0;
> +
> +	path = btrfs_alloc_path();
> +	if (!path)
> +		return -ENOMEM;
> +	path->reada = 2;

<gets interested in btrfs_path.reada>

<greps for a while>

It's snowing towels in here!

	path->reada = 2;

gack!

<greps some more>

	int direction = path->reada;

	...

		if (direction < 0) {
			if (nr == 0)
				break;
			nr--;
		} else if (direction > 0) {
			nr++;
			if (nr >= nritems)
				break;
		}

<delicately moves on...>

> +	/* search for our xattrs */
> +	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
> +	if (ret < 0)
> +		goto err;
> +	ret = 0;

this statement isn't needed.

> +	advance = 0;
> +	while (1) {
> +		leaf = path->nodes[0];
> +		nritems = btrfs_header_nritems(leaf);
> +		slot = path->slots[0];
> +
> +		/* this is where we start walking through the path */
> +		if (advance || slot >= nritems) {
> +			/*
> +			 * if we've reached the last slot in this leaf we need
> +			 * to go to the next leaf and reset everything
> +			 */
> +			if (slot >= nritems-1) {
> +				ret = btrfs_next_leaf(root, path);
> +				if (ret)
> +					break;
> +				leaf = path->nodes[0];
> +				nritems = btrfs_header_nritems(leaf);
> +				slot = path->slots[0];
> +			} else {
> +				/*
> +				 * just walking through the slots on this leaf
> +				 */
> +				slot++;
> +				path->slots[0]++;
> +			}
> +		}
> +		advance = 1;
> +
> +		item = btrfs_item_nr(leaf, slot);
> +		btrfs_item_key_to_cpu(leaf, &found_key, slot);
> +
> +		/* check to make sure this item is what we want */
> +		if (found_key.objectid != key.objectid)
> +			break;
> +		if (btrfs_key_type(&found_key) != BTRFS_XATTR_ITEM_KEY)
> +			break;
> +
> +		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
> +
> +		name_len = btrfs_dir_name_len(leaf, di);
> +		total_size += name_len + 1;
> +
> +		/* we are just looking for how big our buffer needs to be */
> +		if (!size)
> +			continue;
> +
> +		if (!buffer || (name_len + 1) > size_left) {
> +			ret = -ERANGE;
> +			goto err;
> +		}
> +
> +		name_ptr = (unsigned long)(di + 1);
> +		read_extent_buffer(leaf, buffer, name_ptr, name_len);
> +		buffer[name_len] = '\0';
> +
> +		size_left -= name_len + 1;
> +		buffer += name_len + 1;
> +	}
> +	ret = total_size;
> +
> +err:
> +	btrfs_free_path(path);
> +
> +	return ret;
> +}
>
> ...
>

Re: [PATCH 01/35] Btrfs xattr code

[Josef Bacik]

On Mon, Jan 12, 2009 at 08:59:04PM -0800, Andrew Morton wrote:
> On Wed,  7 Jan 2009 22:56:51 -0500 Chris Mason <chris.mason@oracle.com> wrote:
> 
> > Xattrs in btrfs are stored in the btree close to the inode item.  They use an
> > indexing scheme very similar to directories.
> > 
> > The max size of a btrfs xattr is limited by the size of a btree item,
> > or about 4k.  A future version will support larger items.
> > 
> >
> > ...
> >
> > +ssize_t __btrfs_getxattr(struct inode *inode, const char *name,
> > +				void *buffer, size_t size)
> > +{
> > +	struct btrfs_dir_item *di;
> > +	struct btrfs_root *root = BTRFS_I(inode)->root;
> > +	struct btrfs_path *path;
> > +	struct extent_buffer *leaf;
> > +	int ret = 0;
> > +	unsigned long data_ptr;
> > +
> > +	path = btrfs_alloc_path();
> > +	if (!path)
> > +		return -ENOMEM;
> > +
> > +	/* lookup the xattr by name */
> > +	di = btrfs_lookup_xattr(NULL, root, path, inode->i_ino, name,
> > +				strlen(name), 0);
> > +	if (!di || IS_ERR(di)) {
> 
> This seems quite bad.  An error return from btrfs_lookup_xattr() can
> arise from (I assume) OOM, corrupted filesystem, I/O error, etc.
> 
> Yet here we treat such errors as being the same as "not found".  This
> seems grossly misleading to userspace and possible a security problem?
>

Hmm yeah sorry about that, if IS_ERR(di) we should be returning the ERR we got,
I will fix taht.
 
> (I'm kind of guessing here, and might be wrong and wasting everyone's
> time, because the semantics of the btrfs_lookup_xattr() and
> btrfs_getxattr() return values aren't documented).
> 
> > +		ret = -ENODATA;
> > +		goto out;
> > +	}
> > +
> > +	leaf = path->nodes[0];
> > +	/* if size is 0, that means we want the size of the attr */
> > +	if (!size) {
> > +		ret = btrfs_dir_data_len(leaf, di);
> > +		goto out;
> > +	}
> > +
> > +	/* now get the data out of our dir_item */
> > +	if (btrfs_dir_data_len(leaf, di) > size) {
> 
> OK, where did we hide the btrfs_dir_data_len() definition?
>

Its a macro in ctree.h I believe, just returns the data_len part of the dir item
in that leaf.
 
> > +		ret = -ERANGE;
> > +		goto out;
> > +	}
> > +	data_ptr = (unsigned long)((char *)(di + 1) +
> > +				   btrfs_dir_name_len(leaf, di));
> > +	read_extent_buffer(leaf, buffer, data_ptr,
> > +			   btrfs_dir_data_len(leaf, di));
> 
> Please document read_extent_buffer().
> 
> Please rename read_extent_buffer() to something more appropriate to a
> kernel-wide symbol.
> 
> <spies map_private_extent_buffer as well>
> 
> Please review all kernel-wide identifiers in btrfs for appropriateness.
> 
> I'm scratching my head wondering about this `data_ptr' thing.  Is it a
> disk offset?  Is it really a pointer to kernel memory?  According to
> this code it is indeed a kernel pointer, but it then gets stuffed into
> an unsigned long (wtf?) and then passed to the mysterious
> read_extent_buffer().
>

Its a pointer to kernel memory, probably should have commented that, sorry again
:).  Basically you have the dir item, which holds the name len and data len
info, then directly following that structure you have the name, and then
directly following that you have the "data", which is what is contained in the
xattr.  So this is saying read the data from the extent buffer (which is the
leaf) starting from where the data will be.
 
> <reviewer throws in the towel on this part of the code>
> 
> > +	ret = btrfs_dir_data_len(leaf, di);
> > +
> > +out:
> > +	btrfs_free_path(path);
> > +	return ret;
> > +}
> > +
> > +int __btrfs_setxattr(struct inode *inode, const char *name,
> > +			    const void *value, size_t size, int flags)
> > +{
> > +	struct btrfs_dir_item *di;
> > +	struct btrfs_root *root = BTRFS_I(inode)->root;
> > +	struct btrfs_trans_handle *trans;
> > +	struct btrfs_path *path;
> > +	int ret = 0, mod = 0;
> > +
> > +	path = btrfs_alloc_path();
> > +	if (!path)
> > +		return -ENOMEM;
> > +
> > +	trans = btrfs_start_transaction(root, 1);
> > +	btrfs_set_trans_block_group(trans, inode);
> > +
> > +	/* first lets see if we already have this xattr */
> 
> let's :)
> 
> > +	di = btrfs_lookup_xattr(trans, root, path, inode->i_ino, name,
> > +				strlen(name), -1);
> 
> <wonders what the -1 does>
> 
> <goes to the btrfs_lookup_xattr() definition site>
> 
> <towel goes flying again>
> 

Just calls btrfs_path_search with -1 for the ins_len, tells it that we may
remove the item so do the appropriate things to the tree to make it ok to do
that.

> > +	if (IS_ERR(di)) {
> > +		ret = PTR_ERR(di);
> > +		goto out;
> > +	}
> > +
> > +	/* ok we already have this xattr, lets remove it */
> > +	if (di) {
> > +		/* if we want create only exit */
> > +		if (flags & XATTR_CREATE) {
> > +			ret = -EEXIST;
> > +			goto out;
> > +		}
> > +
> > +		ret = btrfs_delete_one_dir_name(trans, root, path, di);
> > +		if (ret)
> > +			goto out;
> > +		btrfs_release_path(root, path);
> > +
> > +		/* if we don't have a value then we are removing the xattr */
> > +		if (!value) {
> > +			mod = 1;
> > +			goto out;
> > +		}
> > +	} else {
> > +		btrfs_release_path(root, path);
> > +
> > +		if (flags & XATTR_REPLACE) {
> > +			/* we couldn't find the attr to replace */
> > +			ret = -ENODATA;
> > +			goto out;
> > +		}
> > +	}
> > +
> > +	/* ok we have to create a completely new xattr */
> > +	ret = btrfs_insert_xattr_item(trans, root, name, strlen(name),
> > +				      value, size, inode->i_ino);
> > +	if (ret)
> > +		goto out;
> > +	mod = 1;
> > +
> > +out:
> > +	if (mod) {
> > +		inode->i_ctime = CURRENT_TIME;
> > +		ret = btrfs_update_inode(trans, root, inode);
> > +	}
> > +
> > +	btrfs_end_transaction(trans, root);
> > +	btrfs_free_path(path);
> > +	return ret;
> > +}
> > +
> > +ssize_t btrfs_listxattr(struct dentry *dentry, char *buffer, size_t size)
> > +{
> > +	struct btrfs_key key, found_key;
> > +	struct inode *inode = dentry->d_inode;
> > +	struct btrfs_root *root = BTRFS_I(inode)->root;
> > +	struct btrfs_path *path;
> > +	struct btrfs_item *item;
> > +	struct extent_buffer *leaf;
> > +	struct btrfs_dir_item *di;
> > +	int ret = 0, slot, advance;
> > +	size_t total_size = 0, size_left = size;
> > +	unsigned long name_ptr;
> > +	size_t name_len;
> > +	u32 nritems;
> > +
> > +	/*
> > +	 * ok we want all objects associated with this id.
> > +	 * NOTE: we set key.offset = 0; because we want to start with the
> > +	 * first xattr that we find and walk forward
> > +	 */
> > +	key.objectid = inode->i_ino;
> > +	btrfs_set_key_type(&key, BTRFS_XATTR_ITEM_KEY);
> > +	key.offset = 0;
> > +
> > +	path = btrfs_alloc_path();
> > +	if (!path)
> > +		return -ENOMEM;
> > +	path->reada = 2;
> 
> <gets interested in btrfs_path.reada>
> 
> <greps for a while>
> 
> It's snowing towels in here!
> 
> 	path->reada = 2;
> 
> gack!
> 
> <greps some more>
> 
> 	int direction = path->reada;
> 
> 	...
> 
> 		if (direction < 0) {
> 			if (nr == 0)
> 				break;
> 			nr--;
> 		} else if (direction > 0) {
> 			nr++;
> 			if (nr >= nritems)
> 				break;
> 		}
> 
> <delicately moves on...>
> 
> > +	/* search for our xattrs */
> > +	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
> > +	if (ret < 0)
> > +		goto err;
> > +	ret = 0;
> 
> this statement isn't needed.
> 

Hmm yup you are right, will fix that too.  Thanks much,

Josef

Re: [PATCH 01/35] Btrfs xattr code

[Chris Mason]

On Mon, 2009-01-12 at 20:59 -0800, Andrew Morton wrote:
> On Wed,  7 Jan 2009 22:56:51 -0500 Chris Mason
> <chris.mason@oracle.com> wrote:
> 

> > +		ret = -ENODATA;
> > +		goto out;
> > +	}
> > +
> > +	leaf = path->nodes[0];
> > +	/* if size is 0, that means we want the size of the attr */
> > +	if (!size) {
> > +		ret = btrfs_dir_data_len(leaf, di);
> > +		goto out;
> > +	}
> > +
> > +	/* now get the data out of our dir_item */
> > +	if (btrfs_dir_data_len(leaf, di) > size) {
> 
> OK, where did we hide the btrfs_dir_data_len() definition?
> 

I had this strange dream that google airlines was bombing my house with
towels....

ctree.h:
/* struct btrfs_dir_item */
BTRFS_SETGET_FUNCS(dir_data_len, struct btrfs_dir_item, data_len, 16);
BTRFS_SETGET_FUNCS(dir_type, struct btrfs_dir_item, type, 8);
BTRFS_SETGET_FUNCS(dir_name_len, struct btrfs_dir_item, name_len, 16);
BTRFS_SETGET_FUNCS(dir_transid, struct btrfs_dir_item, transid, 64);

The macros that define the SETGET_FUNCS are the ugliest part of btrfs.
They are defined and documented inside struct-funcs.c.  In general they
provide access to btree blocks that may be smaller/bigger than a single
page and hide the kmap fun.

Because the kmap fun can be a little slow, performance sensitive loops
can use map_extent_buffer to kmap one of the pages in the extent_buffer
and do their work.

> > +		ret = -ERANGE;
> > +		goto out;
> > +	}
> > +	data_ptr = (unsigned long)((char *)(di + 1) +
> > +				   btrfs_dir_name_len(leaf, di));
> > +	read_extent_buffer(leaf, buffer, data_ptr,
> > +			   btrfs_dir_data_len(leaf, di));
> 
> Please document read_extent_buffer().

Will document *extent_buffer*

> 
> Please rename read_extent_buffer() to something more appropriate to a
> kernel-wide symbol.

The long term plan was for the extent_buffers to be a generic interface,
so I gave them generic names.  Before I can finish that off I need to
fix them to properly work with blocksize != pagesize.  They have
supported multi-page btree blocks, but I turned that off for a while to
finish off other things.

If you'd rather see them renamed to btrfs_, I'll do so.

> 
> <spies map_private_extent_buffer as well>
> 
> Please review all kernel-wide identifiers in btrfs for appropriateness.
> 
> I'm scratching my head wondering about this `data_ptr' thing.  Is it a
> disk offset?  Is it really a pointer to kernel memory?  According to
> this code it is indeed a kernel pointer, but it then gets stuffed into
> an unsigned long (wtf?) and then passed to the mysterious
> read_extent_buffer().

The data_ptr is an unsigned long offset into the btree block.  If you
think of read_extent_buffer as a fancy memcpy wrapper, data_ptr is the
offset in the buffer to read from.

The code in struct-funcs.c hides that most of the time, casting the
unsigned long into a strict type so that most operations look like they
are happening against a real type.

So for: ret = btrfs_dir_data_len(leaf, di);

leaf is the extent_buffer which holds the btree block in ram, and di is
an offset from zero into that buffer where you can find a directory item
struct.  It has been cast into a struct btrfs_dir_item, but it is some
number between 0 and 4096.

(It won't actually be zero since that is the start of the header and no
items are stored in the header).

> 
> > +	ret = btrfs_dir_data_len(leaf, di);
> > +
> > +out:
> > +	btrfs_free_path(path);
> > +	return ret;
> > +}
> > +
> > +int __btrfs_setxattr(struct inode *inode, const char *name,
> > +			    const void *value, size_t size, int flags)
> > +{
> > +	struct btrfs_dir_item *di;
> > +	struct btrfs_root *root = BTRFS_I(inode)->root;
> > +	struct btrfs_trans_handle *trans;
> > +	struct btrfs_path *path;
> > +	int ret = 0, mod = 0;
> > +
> > +	path = btrfs_alloc_path();
> > +	if (!path)
> > +		return -ENOMEM;
> > +
> > +	trans = btrfs_start_transaction(root, 1);
> > +	btrfs_set_trans_block_group(trans, inode);
> > +
> > +	/* first lets see if we already have this xattr */
> 
> let's :)
> 
> > +	di = btrfs_lookup_xattr(trans, root, path, inode->i_ino, name,
> > +				strlen(name), -1);
> 
> <wonders what the -1 does>
> 
> <goes to the btrfs_lookup_xattr() definition site>
> 

I'll replace the -1 with BTRFS_SEARCH_ITEM_DEL, but as Josef said this
tells the searching code to balance for item deletion.  Nothing is
actually deleted unless btrfs_del_item is called, but the tree is
prepared for deletion/insertion from the top down, so the initial search
needs to know what the code intends to do.

> 
> > +	if (IS_ERR(di)) {
> > +		ret = PTR_ERR(di);
> > +		goto out;
> > +	}
> > +
> > +	/* ok we already have this xattr, lets remove it */
> > +	if (di) {
> > +		/* if we want create only exit */
> > +		if (flags & XATTR_CREATE) {
> > +			ret = -EEXIST;
> > +			goto out;
> > +		}
> > +
> > +		ret = btrfs_delete_one_dir_name(trans, root, path, di);
> > +		if (ret)
> > +			goto out;
> > +		btrfs_release_path(root, path);
> > +
> > +		/* if we don't have a value then we are removing the xattr */
> > +		if (!value) {
> > +			mod = 1;
> > +			goto out;
> > +		}
> > +	} else {
> > +		btrfs_release_path(root, path);
> > +
> > +		if (flags & XATTR_REPLACE) {
> > +			/* we couldn't find the attr to replace */
> > +			ret = -ENODATA;
> > +			goto out;
> > +		}
> > +	}
> > +
> > +	/* ok we have to create a completely new xattr */
> > +	ret = btrfs_insert_xattr_item(trans, root, name, strlen(name),
> > +				      value, size, inode->i_ino);
> > +	if (ret)
> > +		goto out;
> > +	mod = 1;
> > +
> > +out:
> > +	if (mod) {
> > +		inode->i_ctime = CURRENT_TIME;
> > +		ret = btrfs_update_inode(trans, root, inode);
> > +	}
> > +
> > +	btrfs_end_transaction(trans, root);
> > +	btrfs_free_path(path);
> > +	return ret;
> > +}
> > +
> > +ssize_t btrfs_listxattr(struct dentry *dentry, char *buffer, size_t size)
> > +{
> > +	struct btrfs_key key, found_key;
> > +	struct inode *inode = dentry->d_inode;
> > +	struct btrfs_root *root = BTRFS_I(inode)->root;
> > +	struct btrfs_path *path;
> > +	struct btrfs_item *item;
> > +	struct extent_buffer *leaf;
> > +	struct btrfs_dir_item *di;
> > +	int ret = 0, slot, advance;
> > +	size_t total_size = 0, size_left = size;
> > +	unsigned long name_ptr;
> > +	size_t name_len;
> > +	u32 nritems;
> > +
> > +	/*
> > +	 * ok we want all objects associated with this id.
> > +	 * NOTE: we set key.offset = 0; because we want to start with the
> > +	 * first xattr that we find and walk forward
> > +	 */
> > +	key.objectid = inode->i_ino;
> > +	btrfs_set_key_type(&key, BTRFS_XATTR_ITEM_KEY);
> > +	key.offset = 0;
> > +
> > +	path = btrfs_alloc_path();
> > +	if (!path)
> > +		return -ENOMEM;
> > +	path->reada = 2;
> 
> <gets interested in btrfs_path.reada>
> 
> <greps for a while>
> 

> 	path->reada = 2;

I'll fix this to BTRFS_SEARCH_FORCE_RA

There are three types of metadata ra done.  Some funcs need to walk
backwards from a high key to a low key, so that's backwards ra (-1)

The default is to just read blocks that are close to the target block.
The code above is forcing ra because it knows it is going to be walking
forward in the key space.

-chris

[PATCH 02/35] Btrfs multi-device code

[Chris Mason]

All block and extent pointers in btrfs use a logical address space.  This file
is responsible for mapping that logical address to a physical block on disk.

It implements raid0, raid1 and raid10.  The block groups used by extent-tree.c
are allocated from each device in chunks of ~1GB in size.  This file maintains
the chunk tree and device allocation tree that are used to record the
areas of each disk that are in use.

Each super block has a bootstrapping section that is used to translate
the extents in the chunk tree.  The chunk tree is used to translate all
of the other extents in the filesystem to a physical block on disk.

This file also provides async bio submission code.  The bios are often
sent down by async helper threads that are supposed to be working on
checksumming or other cpu intensive tasks.  We don't want those tasks
waiting for a free request, so a helper thread is used here that is willing
to actually wait on the drive.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
new file mode 100644
index 0000000..b187b53
--- /dev/null
+++ b/fs/btrfs/volumes.c
@@ -0,0 +1,3218 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+#include <linux/sched.h>
+#include <linux/bio.h>
+#include <linux/buffer_head.h>
+#include <linux/blkdev.h>
+#include <linux/random.h>
+#include <linux/version.h>
+#include <asm/div64.h>
+#include "compat.h"
+#include "ctree.h"
+#include "extent_map.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "print-tree.h"
+#include "volumes.h"
+#include "async-thread.h"
+
+struct map_lookup {
+	u64 type;
+	int io_align;
+	int io_width;
+	int stripe_len;
+	int sector_size;
+	int num_stripes;
+	int sub_stripes;
+	struct btrfs_bio_stripe stripes[];
+};
+
+static int init_first_rw_device(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct btrfs_device *device);
+static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
+
+#define map_lookup_size(n) (sizeof(struct map_lookup) + \
+			    (sizeof(struct btrfs_bio_stripe) * (n)))
+
+static DEFINE_MUTEX(uuid_mutex);
+static LIST_HEAD(fs_uuids);
+
+void btrfs_lock_volumes(void)
+{
+	mutex_lock(&uuid_mutex);
+}
+
+void btrfs_unlock_volumes(void)
+{
+	mutex_unlock(&uuid_mutex);
+}
+
+static void lock_chunks(struct btrfs_root *root)
+{
+	mutex_lock(&root->fs_info->chunk_mutex);
+}
+
+static void unlock_chunks(struct btrfs_root *root)
+{
+	mutex_unlock(&root->fs_info->chunk_mutex);
+}
+
+static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
+{
+	struct btrfs_device *device;
+	WARN_ON(fs_devices->opened);
+	while (!list_empty(&fs_devices->devices)) {
+		device = list_entry(fs_devices->devices.next,
+				    struct btrfs_device, dev_list);
+		list_del(&device->dev_list);
+		kfree(device->name);
+		kfree(device);
+	}
+	kfree(fs_devices);
+}
+
+int btrfs_cleanup_fs_uuids(void)
+{
+	struct btrfs_fs_devices *fs_devices;
+
+	while (!list_empty(&fs_uuids)) {
+		fs_devices = list_entry(fs_uuids.next,
+					struct btrfs_fs_devices, list);
+		list_del(&fs_devices->list);
+		free_fs_devices(fs_devices);
+	}
+	return 0;
+}
+
+static noinline struct btrfs_device *__find_device(struct list_head *head,
+						   u64 devid, u8 *uuid)
+{
+	struct btrfs_device *dev;
+	struct list_head *cur;
+
+	list_for_each(cur, head) {
+		dev = list_entry(cur, struct btrfs_device, dev_list);
+		if (dev->devid == devid &&
+		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
+			return dev;
+		}
+	}
+	return NULL;
+}
+
+static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
+{
+	struct list_head *cur;
+	struct btrfs_fs_devices *fs_devices;
+
+	list_for_each(cur, &fs_uuids) {
+		fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
+		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
+			return fs_devices;
+	}
+	return NULL;
+}
+
+/*
+ * we try to collect pending bios for a device so we don't get a large
+ * number of procs sending bios down to the same device.  This greatly
+ * improves the schedulers ability to collect and merge the bios.
+ *
+ * But, it also turns into a long list of bios to process and that is sure
+ * to eventually make the worker thread block.  The solution here is to
+ * make some progress and then put this work struct back at the end of
+ * the list if the block device is congested.  This way, multiple devices
+ * can make progress from a single worker thread.
+ */
+static noinline int run_scheduled_bios(struct btrfs_device *device)
+{
+	struct bio *pending;
+	struct backing_dev_info *bdi;
+	struct btrfs_fs_info *fs_info;
+	struct bio *tail;
+	struct bio *cur;
+	int again = 0;
+	unsigned long num_run = 0;
+	unsigned long limit;
+
+	bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
+	fs_info = device->dev_root->fs_info;
+	limit = btrfs_async_submit_limit(fs_info);
+	limit = limit * 2 / 3;
+
+loop:
+	spin_lock(&device->io_lock);
+
+	/* take all the bios off the list at once and process them
+	 * later on (without the lock held).  But, remember the
+	 * tail and other pointers so the bios can be properly reinserted
+	 * into the list if we hit congestion
+	 */
+	pending = device->pending_bios;
+	tail = device->pending_bio_tail;
+	WARN_ON(pending && !tail);
+	device->pending_bios = NULL;
+	device->pending_bio_tail = NULL;
+
+	/*
+	 * if pending was null this time around, no bios need processing
+	 * at all and we can stop.  Otherwise it'll loop back up again
+	 * and do an additional check so no bios are missed.
+	 *
+	 * device->running_pending is used to synchronize with the
+	 * schedule_bio code.
+	 */
+	if (pending) {
+		again = 1;
+		device->running_pending = 1;
+	} else {
+		again = 0;
+		device->running_pending = 0;
+	}
+	spin_unlock(&device->io_lock);
+
+	while (pending) {
+		cur = pending;
+		pending = pending->bi_next;
+		cur->bi_next = NULL;
+		atomic_dec(&fs_info->nr_async_bios);
+
+		if (atomic_read(&fs_info->nr_async_bios) < limit &&
+		    waitqueue_active(&fs_info->async_submit_wait))
+			wake_up(&fs_info->async_submit_wait);
+
+		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
+		bio_get(cur);
+		submit_bio(cur->bi_rw, cur);
+		bio_put(cur);
+		num_run++;
+
+		/*
+		 * we made progress, there is more work to do and the bdi
+		 * is now congested.  Back off and let other work structs
+		 * run instead
+		 */
+		if (pending && bdi_write_congested(bdi) &&
+		    fs_info->fs_devices->open_devices > 1) {
+			struct bio *old_head;
+
+			spin_lock(&device->io_lock);
+
+			old_head = device->pending_bios;
+			device->pending_bios = pending;
+			if (device->pending_bio_tail)
+				tail->bi_next = old_head;
+			else
+				device->pending_bio_tail = tail;
+
+			spin_unlock(&device->io_lock);
+			btrfs_requeue_work(&device->work);
+			goto done;
+		}
+	}
+	if (again)
+		goto loop;
+done:
+	return 0;
+}
+
+static void pending_bios_fn(struct btrfs_work *work)
+{
+	struct btrfs_device *device;
+
+	device = container_of(work, struct btrfs_device, work);
+	run_scheduled_bios(device);
+}
+
+static noinline int device_list_add(const char *path,
+			   struct btrfs_super_block *disk_super,
+			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
+{
+	struct btrfs_device *device;
+	struct btrfs_fs_devices *fs_devices;
+	u64 found_transid = btrfs_super_generation(disk_super);
+
+	fs_devices = find_fsid(disk_super->fsid);
+	if (!fs_devices) {
+		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
+		if (!fs_devices)
+			return -ENOMEM;
+		INIT_LIST_HEAD(&fs_devices->devices);
+		INIT_LIST_HEAD(&fs_devices->alloc_list);
+		list_add(&fs_devices->list, &fs_uuids);
+		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
+		fs_devices->latest_devid = devid;
+		fs_devices->latest_trans = found_transid;
+		device = NULL;
+	} else {
+		device = __find_device(&fs_devices->devices, devid,
+				       disk_super->dev_item.uuid);
+	}
+	if (!device) {
+		if (fs_devices->opened)
+			return -EBUSY;
+
+		device = kzalloc(sizeof(*device), GFP_NOFS);
+		if (!device) {
+			/* we can safely leave the fs_devices entry around */
+			return -ENOMEM;
+		}
+		device->devid = devid;
+		device->work.func = pending_bios_fn;
+		memcpy(device->uuid, disk_super->dev_item.uuid,
+		       BTRFS_UUID_SIZE);
+		device->barriers = 1;
+		spin_lock_init(&device->io_lock);
+		device->name = kstrdup(path, GFP_NOFS);
+		if (!device->name) {
+			kfree(device);
+			return -ENOMEM;
+		}
+		INIT_LIST_HEAD(&device->dev_alloc_list);
+		list_add(&device->dev_list, &fs_devices->devices);
+		device->fs_devices = fs_devices;
+		fs_devices->num_devices++;
+	}
+
+	if (found_transid > fs_devices->latest_trans) {
+		fs_devices->latest_devid = devid;
+		fs_devices->latest_trans = found_transid;
+	}
+	*fs_devices_ret = fs_devices;
+	return 0;
+}
+
+static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
+{
+	struct btrfs_fs_devices *fs_devices;
+	struct btrfs_device *device;
+	struct btrfs_device *orig_dev;
+
+	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
+	if (!fs_devices)
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&fs_devices->devices);
+	INIT_LIST_HEAD(&fs_devices->alloc_list);
+	INIT_LIST_HEAD(&fs_devices->list);
+	fs_devices->latest_devid = orig->latest_devid;
+	fs_devices->latest_trans = orig->latest_trans;
+	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
+
+	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
+		device = kzalloc(sizeof(*device), GFP_NOFS);
+		if (!device)
+			goto error;
+
+		device->name = kstrdup(orig_dev->name, GFP_NOFS);
+		if (!device->name)
+			goto error;
+
+		device->devid = orig_dev->devid;
+		device->work.func = pending_bios_fn;
+		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
+		device->barriers = 1;
+		spin_lock_init(&device->io_lock);
+		INIT_LIST_HEAD(&device->dev_list);
+		INIT_LIST_HEAD(&device->dev_alloc_list);
+
+		list_add(&device->dev_list, &fs_devices->devices);
+		device->fs_devices = fs_devices;
+		fs_devices->num_devices++;
+	}
+	return fs_devices;
+error:
+	free_fs_devices(fs_devices);
+	return ERR_PTR(-ENOMEM);
+}
+
+int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
+{
+	struct list_head *tmp;
+	struct list_head *cur;
+	struct btrfs_device *device;
+
+	mutex_lock(&uuid_mutex);
+again:
+	list_for_each_safe(cur, tmp, &fs_devices->devices) {
+		device = list_entry(cur, struct btrfs_device, dev_list);
+		if (device->in_fs_metadata)
+			continue;
+
+		if (device->bdev) {
+			close_bdev_exclusive(device->bdev, device->mode);
+			device->bdev = NULL;
+			fs_devices->open_devices--;
+		}
+		if (device->writeable) {
+			list_del_init(&device->dev_alloc_list);
+			device->writeable = 0;
+			fs_devices->rw_devices--;
+		}
+		list_del_init(&device->dev_list);
+		fs_devices->num_devices--;
+		kfree(device->name);
+		kfree(device);
+	}
+
+	if (fs_devices->seed) {
+		fs_devices = fs_devices->seed;
+		goto again;
+	}
+
+	mutex_unlock(&uuid_mutex);
+	return 0;
+}
+
+static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
+{
+	struct list_head *cur;
+	struct btrfs_device *device;
+
+	if (--fs_devices->opened > 0)
+		return 0;
+
+	list_for_each(cur, &fs_devices->devices) {
+		device = list_entry(cur, struct btrfs_device, dev_list);
+		if (device->bdev) {
+			close_bdev_exclusive(device->bdev, device->mode);
+			fs_devices->open_devices--;
+		}
+		if (device->writeable) {
+			list_del_init(&device->dev_alloc_list);
+			fs_devices->rw_devices--;
+		}
+
+		device->bdev = NULL;
+		device->writeable = 0;
+		device->in_fs_metadata = 0;
+	}
+	WARN_ON(fs_devices->open_devices);
+	WARN_ON(fs_devices->rw_devices);
+	fs_devices->opened = 0;
+	fs_devices->seeding = 0;
+
+	return 0;
+}
+
+int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
+{
+	struct btrfs_fs_devices *seed_devices = NULL;
+	int ret;
+
+	mutex_lock(&uuid_mutex);
+	ret = __btrfs_close_devices(fs_devices);
+	if (!fs_devices->opened) {
+		seed_devices = fs_devices->seed;
+		fs_devices->seed = NULL;
+	}
+	mutex_unlock(&uuid_mutex);
+
+	while (seed_devices) {
+		fs_devices = seed_devices;
+		seed_devices = fs_devices->seed;
+		__btrfs_close_devices(fs_devices);
+		free_fs_devices(fs_devices);
+	}
+	return ret;
+}
+
+static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
+				fmode_t flags, void *holder)
+{
+	struct block_device *bdev;
+	struct list_head *head = &fs_devices->devices;
+	struct list_head *cur;
+	struct btrfs_device *device;
+	struct block_device *latest_bdev = NULL;
+	struct buffer_head *bh;
+	struct btrfs_super_block *disk_super;
+	u64 latest_devid = 0;
+	u64 latest_transid = 0;
+	u64 devid;
+	int seeding = 1;
+	int ret = 0;
+
+	list_for_each(cur, head) {
+		device = list_entry(cur, struct btrfs_device, dev_list);
+		if (device->bdev)
+			continue;
+		if (!device->name)
+			continue;
+
+		bdev = open_bdev_exclusive(device->name, flags, holder);
+		if (IS_ERR(bdev)) {
+			printk(KERN_INFO "open %s failed\n", device->name);
+			goto error;
+		}
+		set_blocksize(bdev, 4096);
+
+		bh = btrfs_read_dev_super(bdev);
+		if (!bh)
+			goto error_close;
+
+		disk_super = (struct btrfs_super_block *)bh->b_data;
+		devid = le64_to_cpu(disk_super->dev_item.devid);
+		if (devid != device->devid)
+			goto error_brelse;
+
+		if (memcmp(device->uuid, disk_super->dev_item.uuid,
+			   BTRFS_UUID_SIZE))
+			goto error_brelse;
+
+		device->generation = btrfs_super_generation(disk_super);
+		if (!latest_transid || device->generation > latest_transid) {
+			latest_devid = devid;
+			latest_transid = device->generation;
+			latest_bdev = bdev;
+		}
+
+		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
+			device->writeable = 0;
+		} else {
+			device->writeable = !bdev_read_only(bdev);
+			seeding = 0;
+		}
+
+		device->bdev = bdev;
+		device->in_fs_metadata = 0;
+		device->mode = flags;
+
+		fs_devices->open_devices++;
+		if (device->writeable) {
+			fs_devices->rw_devices++;
+			list_add(&device->dev_alloc_list,
+				 &fs_devices->alloc_list);
+		}
+		continue;
+
+error_brelse:
+		brelse(bh);
+error_close:
+		close_bdev_exclusive(bdev, FMODE_READ);
+error:
+		continue;
+	}
+	if (fs_devices->open_devices == 0) {
+		ret = -EIO;
+		goto out;
+	}
+	fs_devices->seeding = seeding;
+	fs_devices->opened = 1;
+	fs_devices->latest_bdev = latest_bdev;
+	fs_devices->latest_devid = latest_devid;
+	fs_devices->latest_trans = latest_transid;
+	fs_devices->total_rw_bytes = 0;
+out:
+	return ret;
+}
+
+int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
+		       fmode_t flags, void *holder)
+{
+	int ret;
+
+	mutex_lock(&uuid_mutex);
+	if (fs_devices->opened) {
+		fs_devices->opened++;
+		ret = 0;
+	} else {
+		ret = __btrfs_open_devices(fs_devices, flags, holder);
+	}
+	mutex_unlock(&uuid_mutex);
+	return ret;
+}
+
+int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
+			  struct btrfs_fs_devices **fs_devices_ret)
+{
+	struct btrfs_super_block *disk_super;
+	struct block_device *bdev;
+	struct buffer_head *bh;
+	int ret;
+	u64 devid;
+	u64 transid;
+
+	mutex_lock(&uuid_mutex);
+
+	bdev = open_bdev_exclusive(path, flags, holder);
+
+	if (IS_ERR(bdev)) {
+		ret = PTR_ERR(bdev);
+		goto error;
+	}
+
+	ret = set_blocksize(bdev, 4096);
+	if (ret)
+		goto error_close;
+	bh = btrfs_read_dev_super(bdev);
+	if (!bh) {
+		ret = -EIO;
+		goto error_close;
+	}
+	disk_super = (struct btrfs_super_block *)bh->b_data;
+	devid = le64_to_cpu(disk_super->dev_item.devid);
+	transid = btrfs_super_generation(disk_super);
+	if (disk_super->label[0])
+		printk(KERN_INFO "device label %s ", disk_super->label);
+	else {
+		/* FIXME, make a readl uuid parser */
+		printk(KERN_INFO "device fsid %llx-%llx ",
+		       *(unsigned long long *)disk_super->fsid,
+		       *(unsigned long long *)(disk_super->fsid + 8));
+	}
+	printk(KERN_INFO "devid %llu transid %llu %s\n",
+	       (unsigned long long)devid, (unsigned long long)transid, path);
+	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
+
+	brelse(bh);
+error_close:
+	close_bdev_exclusive(bdev, flags);
+error:
+	mutex_unlock(&uuid_mutex);
+	return ret;
+}
+
+/*
+ * this uses a pretty simple search, the expectation is that it is
+ * called very infrequently and that a given device has a small number
+ * of extents
+ */
+static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
+					 struct btrfs_device *device,
+					 u64 num_bytes, u64 *start)
+{
+	struct btrfs_key key;
+	struct btrfs_root *root = device->dev_root;
+	struct btrfs_dev_extent *dev_extent = NULL;
+	struct btrfs_path *path;
+	u64 hole_size = 0;
+	u64 last_byte = 0;
+	u64 search_start = 0;
+	u64 search_end = device->total_bytes;
+	int ret;
+	int slot = 0;
+	int start_found;
+	struct extent_buffer *l;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	path->reada = 2;
+	start_found = 0;
+
+	/* FIXME use last free of some kind */
+
+	/* we don't want to overwrite the superblock on the drive,
+	 * so we make sure to start at an offset of at least 1MB
+	 */
+	search_start = max((u64)1024 * 1024, search_start);
+
+	if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
+		search_start = max(root->fs_info->alloc_start, search_start);
+
+	key.objectid = device->devid;
+	key.offset = search_start;
+	key.type = BTRFS_DEV_EXTENT_KEY;
+	ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto error;
+	ret = btrfs_previous_item(root, path, 0, key.type);
+	if (ret < 0)
+		goto error;
+	l = path->nodes[0];
+	btrfs_item_key_to_cpu(l, &key, path->slots[0]);
+	while (1) {
+		l = path->nodes[0];
+		slot = path->slots[0];
+		if (slot >= btrfs_header_nritems(l)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret == 0)
+				continue;
+			if (ret < 0)
+				goto error;
+no_more_items:
+			if (!start_found) {
+				if (search_start >= search_end) {
+					ret = -ENOSPC;
+					goto error;
+				}
+				*start = search_start;
+				start_found = 1;
+				goto check_pending;
+			}
+			*start = last_byte > search_start ?
+				last_byte : search_start;
+			if (search_end <= *start) {
+				ret = -ENOSPC;
+				goto error;
+			}
+			goto check_pending;
+		}
+		btrfs_item_key_to_cpu(l, &key, slot);
+
+		if (key.objectid < device->devid)
+			goto next;
+
+		if (key.objectid > device->devid)
+			goto no_more_items;
+
+		if (key.offset >= search_start && key.offset > last_byte &&
+		    start_found) {
+			if (last_byte < search_start)
+				last_byte = search_start;
+			hole_size = key.offset - last_byte;
+			if (key.offset > last_byte &&
+			    hole_size >= num_bytes) {
+				*start = last_byte;
+				goto check_pending;
+			}
+		}
+		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
+			goto next;
+
+		start_found = 1;
+		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
+		last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
+next:
+		path->slots[0]++;
+		cond_resched();
+	}
+check_pending:
+	/* we have to make sure we didn't find an extent that has already
+	 * been allocated by the map tree or the original allocation
+	 */
+	BUG_ON(*start < search_start);
+
+	if (*start + num_bytes > search_end) {
+		ret = -ENOSPC;
+		goto error;
+	}
+	/* check for pending inserts here */
+	ret = 0;
+
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
+			  struct btrfs_device *device,
+			  u64 start)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_root *root = device->dev_root;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf = NULL;
+	struct btrfs_dev_extent *extent = NULL;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = device->devid;
+	key.offset = start;
+	key.type = BTRFS_DEV_EXTENT_KEY;
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret > 0) {
+		ret = btrfs_previous_item(root, path, key.objectid,
+					  BTRFS_DEV_EXTENT_KEY);
+		BUG_ON(ret);
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		extent = btrfs_item_ptr(leaf, path->slots[0],
+					struct btrfs_dev_extent);
+		BUG_ON(found_key.offset > start || found_key.offset +
+		       btrfs_dev_extent_length(leaf, extent) < start);
+		ret = 0;
+	} else if (ret == 0) {
+		leaf = path->nodes[0];
+		extent = btrfs_item_ptr(leaf, path->slots[0],
+					struct btrfs_dev_extent);
+	}
+	BUG_ON(ret);
+
+	if (device->bytes_used > 0)
+		device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
+	ret = btrfs_del_item(trans, root, path);
+	BUG_ON(ret);
+
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
+			   struct btrfs_device *device,
+			   u64 chunk_tree, u64 chunk_objectid,
+			   u64 chunk_offset, u64 start, u64 num_bytes)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_root *root = device->dev_root;
+	struct btrfs_dev_extent *extent;
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+
+	WARN_ON(!device->in_fs_metadata);
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = device->devid;
+	key.offset = start;
+	key.type = BTRFS_DEV_EXTENT_KEY;
+	ret = btrfs_insert_empty_item(trans, root, path, &key,
+				      sizeof(*extent));
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	extent = btrfs_item_ptr(leaf, path->slots[0],
+				struct btrfs_dev_extent);
+	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
+	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
+	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
+
+	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
+		    (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
+		    BTRFS_UUID_SIZE);
+
+	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_free_path(path);
+	return ret;
+}
+
+static noinline int find_next_chunk(struct btrfs_root *root,
+				    u64 objectid, u64 *offset)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_chunk *chunk;
+	struct btrfs_key found_key;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	key.objectid = objectid;
+	key.offset = (u64)-1;
+	key.type = BTRFS_CHUNK_ITEM_KEY;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto error;
+
+	BUG_ON(ret == 0);
+
+	ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
+	if (ret) {
+		*offset = 0;
+	} else {
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+				      path->slots[0]);
+		if (found_key.objectid != objectid)
+			*offset = 0;
+		else {
+			chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
+					       struct btrfs_chunk);
+			*offset = found_key.offset +
+				btrfs_chunk_length(path->nodes[0], chunk);
+		}
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
+{
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct btrfs_path *path;
+
+	root = root->fs_info->chunk_root;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
+	key.type = BTRFS_DEV_ITEM_KEY;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto error;
+
+	BUG_ON(ret == 0);
+
+	ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
+				  BTRFS_DEV_ITEM_KEY);
+	if (ret) {
+		*objectid = 1;
+	} else {
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+				      path->slots[0]);
+		*objectid = found_key.offset + 1;
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * the device information is stored in the chunk root
+ * the btrfs_device struct should be fully filled in
+ */
+int btrfs_add_device(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_device *device)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_dev_item *dev_item;
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	unsigned long ptr;
+
+	root = root->fs_info->chunk_root;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
+	key.type = BTRFS_DEV_ITEM_KEY;
+	key.offset = device->devid;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key,
+				      sizeof(*dev_item));
+	if (ret)
+		goto out;
+
+	leaf = path->nodes[0];
+	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
+
+	btrfs_set_device_id(leaf, dev_item, device->devid);
+	btrfs_set_device_generation(leaf, dev_item, 0);
+	btrfs_set_device_type(leaf, dev_item, device->type);
+	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
+	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
+	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
+	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
+	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
+	btrfs_set_device_group(leaf, dev_item, 0);
+	btrfs_set_device_seek_speed(leaf, dev_item, 0);
+	btrfs_set_device_bandwidth(leaf, dev_item, 0);
+	btrfs_set_device_start_offset(leaf, dev_item, 0);
+
+	ptr = (unsigned long)btrfs_device_uuid(dev_item);
+	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
+	ptr = (unsigned long)btrfs_device_fsid(dev_item);
+	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int btrfs_rm_dev_item(struct btrfs_root *root,
+			     struct btrfs_device *device)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_trans_handle *trans;
+
+	root = root->fs_info->chunk_root;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	trans = btrfs_start_transaction(root, 1);
+	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
+	key.type = BTRFS_DEV_ITEM_KEY;
+	key.offset = device->devid;
+	lock_chunks(root);
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret < 0)
+		goto out;
+
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	ret = btrfs_del_item(trans, root, path);
+	if (ret)
+		goto out;
+out:
+	btrfs_free_path(path);
+	unlock_chunks(root);
+	btrfs_commit_transaction(trans, root);
+	return ret;
+}
+
+int btrfs_rm_device(struct btrfs_root *root, char *device_path)
+{
+	struct btrfs_device *device;
+	struct btrfs_device *next_device;
+	struct block_device *bdev;
+	struct buffer_head *bh = NULL;
+	struct btrfs_super_block *disk_super;
+	u64 all_avail;
+	u64 devid;
+	u64 num_devices;
+	u8 *dev_uuid;
+	int ret = 0;
+
+	mutex_lock(&uuid_mutex);
+	mutex_lock(&root->fs_info->volume_mutex);
+
+	all_avail = root->fs_info->avail_data_alloc_bits |
+		root->fs_info->avail_system_alloc_bits |
+		root->fs_info->avail_metadata_alloc_bits;
+
+	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
+	    root->fs_info->fs_devices->rw_devices <= 4) {
+		printk(KERN_ERR "btrfs: unable to go below four devices "
+		       "on raid10\n");
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
+	    root->fs_info->fs_devices->rw_devices <= 2) {
+		printk(KERN_ERR "btrfs: unable to go below two "
+		       "devices on raid1\n");
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (strcmp(device_path, "missing") == 0) {
+		struct list_head *cur;
+		struct list_head *devices;
+		struct btrfs_device *tmp;
+
+		device = NULL;
+		devices = &root->fs_info->fs_devices->devices;
+		list_for_each(cur, devices) {
+			tmp = list_entry(cur, struct btrfs_device, dev_list);
+			if (tmp->in_fs_metadata && !tmp->bdev) {
+				device = tmp;
+				break;
+			}
+		}
+		bdev = NULL;
+		bh = NULL;
+		disk_super = NULL;
+		if (!device) {
+			printk(KERN_ERR "btrfs: no missing devices found to "
+			       "remove\n");
+			goto out;
+		}
+	} else {
+		bdev = open_bdev_exclusive(device_path, FMODE_READ,
+				      root->fs_info->bdev_holder);
+		if (IS_ERR(bdev)) {
+			ret = PTR_ERR(bdev);
+			goto out;
+		}
+
+		set_blocksize(bdev, 4096);
+		bh = btrfs_read_dev_super(bdev);
+		if (!bh) {
+			ret = -EIO;
+			goto error_close;
+		}
+		disk_super = (struct btrfs_super_block *)bh->b_data;
+		devid = le64_to_cpu(disk_super->dev_item.devid);
+		dev_uuid = disk_super->dev_item.uuid;
+		device = btrfs_find_device(root, devid, dev_uuid,
+					   disk_super->fsid);
+		if (!device) {
+			ret = -ENOENT;
+			goto error_brelse;
+		}
+	}
+
+	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
+		printk(KERN_ERR "btrfs: unable to remove the only writeable "
+		       "device\n");
+		ret = -EINVAL;
+		goto error_brelse;
+	}
+
+	if (device->writeable) {
+		list_del_init(&device->dev_alloc_list);
+		root->fs_info->fs_devices->rw_devices--;
+	}
+
+	ret = btrfs_shrink_device(device, 0);
+	if (ret)
+		goto error_brelse;
+
+	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
+	if (ret)
+		goto error_brelse;
+
+	device->in_fs_metadata = 0;
+	list_del_init(&device->dev_list);
+	device->fs_devices->num_devices--;
+
+	next_device = list_entry(root->fs_info->fs_devices->devices.next,
+				 struct btrfs_device, dev_list);
+	if (device->bdev == root->fs_info->sb->s_bdev)
+		root->fs_info->sb->s_bdev = next_device->bdev;
+	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
+		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
+
+	if (device->bdev) {
+		close_bdev_exclusive(device->bdev, device->mode);
+		device->bdev = NULL;
+		device->fs_devices->open_devices--;
+	}
+
+	num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
+	btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
+
+	if (device->fs_devices->open_devices == 0) {
+		struct btrfs_fs_devices *fs_devices;
+		fs_devices = root->fs_info->fs_devices;
+		while (fs_devices) {
+			if (fs_devices->seed == device->fs_devices)
+				break;
+			fs_devices = fs_devices->seed;
+		}
+		fs_devices->seed = device->fs_devices->seed;
+		device->fs_devices->seed = NULL;
+		__btrfs_close_devices(device->fs_devices);
+		free_fs_devices(device->fs_devices);
+	}
+
+	/*
+	 * at this point, the device is zero sized.  We want to
+	 * remove it from the devices list and zero out the old super
+	 */
+	if (device->writeable) {
+		/* make sure this device isn't detected as part of
+		 * the FS anymore
+		 */
+		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
+		set_buffer_dirty(bh);
+		sync_dirty_buffer(bh);
+	}
+
+	kfree(device->name);
+	kfree(device);
+	ret = 0;
+
+error_brelse:
+	brelse(bh);
+error_close:
+	if (bdev)
+		close_bdev_exclusive(bdev, FMODE_READ);
+out:
+	mutex_unlock(&root->fs_info->volume_mutex);
+	mutex_unlock(&uuid_mutex);
+	return ret;
+}
+
+/*
+ * does all the dirty work required for changing file system's UUID.
+ */
+static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root)
+{
+	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
+	struct btrfs_fs_devices *old_devices;
+	struct btrfs_fs_devices *seed_devices;
+	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
+	struct btrfs_device *device;
+	u64 super_flags;
+
+	BUG_ON(!mutex_is_locked(&uuid_mutex));
+	if (!fs_devices->seeding)
+		return -EINVAL;
+
+	seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
+	if (!seed_devices)
+		return -ENOMEM;
+
+	old_devices = clone_fs_devices(fs_devices);
+	if (IS_ERR(old_devices)) {
+		kfree(seed_devices);
+		return PTR_ERR(old_devices);
+	}
+
+	list_add(&old_devices->list, &fs_uuids);
+
+	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
+	seed_devices->opened = 1;
+	INIT_LIST_HEAD(&seed_devices->devices);
+	INIT_LIST_HEAD(&seed_devices->alloc_list);
+	list_splice_init(&fs_devices->devices, &seed_devices->devices);
+	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
+	list_for_each_entry(device, &seed_devices->devices, dev_list) {
+		device->fs_devices = seed_devices;
+	}
+
+	fs_devices->seeding = 0;
+	fs_devices->num_devices = 0;
+	fs_devices->open_devices = 0;
+	fs_devices->seed = seed_devices;
+
+	generate_random_uuid(fs_devices->fsid);
+	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
+	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
+	super_flags = btrfs_super_flags(disk_super) &
+		      ~BTRFS_SUPER_FLAG_SEEDING;
+	btrfs_set_super_flags(disk_super, super_flags);
+
+	return 0;
+}
+
+/*
+ * strore the expected generation for seed devices in device items.
+ */
+static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_dev_item *dev_item;
+	struct btrfs_device *device;
+	struct btrfs_key key;
+	u8 fs_uuid[BTRFS_UUID_SIZE];
+	u8 dev_uuid[BTRFS_UUID_SIZE];
+	u64 devid;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	root = root->fs_info->chunk_root;
+	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
+	key.offset = 0;
+	key.type = BTRFS_DEV_ITEM_KEY;
+
+	while (1) {
+		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+		if (ret < 0)
+			goto error;
+
+		leaf = path->nodes[0];
+next_slot:
+		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret > 0)
+				break;
+			if (ret < 0)
+				goto error;
+			leaf = path->nodes[0];
+			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+			btrfs_release_path(root, path);
+			continue;
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
+		    key.type != BTRFS_DEV_ITEM_KEY)
+			break;
+
+		dev_item = btrfs_item_ptr(leaf, path->slots[0],
+					  struct btrfs_dev_item);
+		devid = btrfs_device_id(leaf, dev_item);
+		read_extent_buffer(leaf, dev_uuid,
+				   (unsigned long)btrfs_device_uuid(dev_item),
+				   BTRFS_UUID_SIZE);
+		read_extent_buffer(leaf, fs_uuid,
+				   (unsigned long)btrfs_device_fsid(dev_item),
+				   BTRFS_UUID_SIZE);
+		device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
+		BUG_ON(!device);
+
+		if (device->fs_devices->seeding) {
+			btrfs_set_device_generation(leaf, dev_item,
+						    device->generation);
+			btrfs_mark_buffer_dirty(leaf);
+		}
+
+		path->slots[0]++;
+		goto next_slot;
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_device *device;
+	struct block_device *bdev;
+	struct list_head *cur;
+	struct list_head *devices;
+	struct super_block *sb = root->fs_info->sb;
+	u64 total_bytes;
+	int seeding_dev = 0;
+	int ret = 0;
+
+	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
+		return -EINVAL;
+
+	bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
+	if (!bdev)
+		return -EIO;
+
+	if (root->fs_info->fs_devices->seeding) {
+		seeding_dev = 1;
+		down_write(&sb->s_umount);
+		mutex_lock(&uuid_mutex);
+	}
+
+	filemap_write_and_wait(bdev->bd_inode->i_mapping);
+	mutex_lock(&root->fs_info->volume_mutex);
+
+	devices = &root->fs_info->fs_devices->devices;
+	list_for_each(cur, devices) {
+		device = list_entry(cur, struct btrfs_device, dev_list);
+		if (device->bdev == bdev) {
+			ret = -EEXIST;
+			goto error;
+		}
+	}
+
+	device = kzalloc(sizeof(*device), GFP_NOFS);
+	if (!device) {
+		/* we can safely leave the fs_devices entry around */
+		ret = -ENOMEM;
+		goto error;
+	}
+
+	device->name = kstrdup(device_path, GFP_NOFS);
+	if (!device->name) {
+		kfree(device);
+		ret = -ENOMEM;
+		goto error;
+	}
+
+	ret = find_next_devid(root, &device->devid);
+	if (ret) {
+		kfree(device);
+		goto error;
+	}
+
+	trans = btrfs_start_transaction(root, 1);
+	lock_chunks(root);
+
+	device->barriers = 1;
+	device->writeable = 1;
+	device->work.func = pending_bios_fn;
+	generate_random_uuid(device->uuid);
+	spin_lock_init(&device->io_lock);
+	device->generation = trans->transid;
+	device->io_width = root->sectorsize;
+	device->io_align = root->sectorsize;
+	device->sector_size = root->sectorsize;
+	device->total_bytes = i_size_read(bdev->bd_inode);
+	device->dev_root = root->fs_info->dev_root;
+	device->bdev = bdev;
+	device->in_fs_metadata = 1;
+	device->mode = 0;
+	set_blocksize(device->bdev, 4096);
+
+	if (seeding_dev) {
+		sb->s_flags &= ~MS_RDONLY;
+		ret = btrfs_prepare_sprout(trans, root);
+		BUG_ON(ret);
+	}
+
+	device->fs_devices = root->fs_info->fs_devices;
+	list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
+	list_add(&device->dev_alloc_list,
+		 &root->fs_info->fs_devices->alloc_list);
+	root->fs_info->fs_devices->num_devices++;
+	root->fs_info->fs_devices->open_devices++;
+	root->fs_info->fs_devices->rw_devices++;
+	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
+
+	total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
+	btrfs_set_super_total_bytes(&root->fs_info->super_copy,
+				    total_bytes + device->total_bytes);
+
+	total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
+	btrfs_set_super_num_devices(&root->fs_info->super_copy,
+				    total_bytes + 1);
+
+	if (seeding_dev) {
+		ret = init_first_rw_device(trans, root, device);
+		BUG_ON(ret);
+		ret = btrfs_finish_sprout(trans, root);
+		BUG_ON(ret);
+	} else {
+		ret = btrfs_add_device(trans, root, device);
+	}
+
+	unlock_chunks(root);
+	btrfs_commit_transaction(trans, root);
+
+	if (seeding_dev) {
+		mutex_unlock(&uuid_mutex);
+		up_write(&sb->s_umount);
+
+		ret = btrfs_relocate_sys_chunks(root);
+		BUG_ON(ret);
+	}
+out:
+	mutex_unlock(&root->fs_info->volume_mutex);
+	return ret;
+error:
+	close_bdev_exclusive(bdev, 0);
+	if (seeding_dev) {
+		mutex_unlock(&uuid_mutex);
+		up_write(&sb->s_umount);
+	}
+	goto out;
+}
+
+static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
+					struct btrfs_device *device)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_root *root;
+	struct btrfs_dev_item *dev_item;
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+
+	root = device->dev_root->fs_info->chunk_root;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
+	key.type = BTRFS_DEV_ITEM_KEY;
+	key.offset = device->devid;
+
+	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+	if (ret < 0)
+		goto out;
+
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
+
+	btrfs_set_device_id(leaf, dev_item, device->devid);
+	btrfs_set_device_type(leaf, dev_item, device->type);
+	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
+	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
+	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
+	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
+	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
+	btrfs_mark_buffer_dirty(leaf);
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
+		      struct btrfs_device *device, u64 new_size)
+{
+	struct btrfs_super_block *super_copy =
+		&device->dev_root->fs_info->super_copy;
+	u64 old_total = btrfs_super_total_bytes(super_copy);
+	u64 diff = new_size - device->total_bytes;
+
+	if (!device->writeable)
+		return -EACCES;
+	if (new_size <= device->total_bytes)
+		return -EINVAL;
+
+	btrfs_set_super_total_bytes(super_copy, old_total + diff);
+	device->fs_devices->total_rw_bytes += diff;
+
+	device->total_bytes = new_size;
+	return btrfs_update_device(trans, device);
+}
+
+int btrfs_grow_device(struct btrfs_trans_handle *trans,
+		      struct btrfs_device *device, u64 new_size)
+{
+	int ret;
+	lock_chunks(device->dev_root);
+	ret = __btrfs_grow_device(trans, device, new_size);
+	unlock_chunks(device->dev_root);
+	return ret;
+}
+
+static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    u64 chunk_tree, u64 chunk_objectid,
+			    u64 chunk_offset)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_key key;
+
+	root = root->fs_info->chunk_root;
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = chunk_objectid;
+	key.offset = chunk_offset;
+	key.type = BTRFS_CHUNK_ITEM_KEY;
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	BUG_ON(ret);
+
+	ret = btrfs_del_item(trans, root, path);
+	BUG_ON(ret);
+
+	btrfs_free_path(path);
+	return 0;
+}
+
+static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
+			chunk_offset)
+{
+	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
+	struct btrfs_disk_key *disk_key;
+	struct btrfs_chunk *chunk;
+	u8 *ptr;
+	int ret = 0;
+	u32 num_stripes;
+	u32 array_size;
+	u32 len = 0;
+	u32 cur;
+	struct btrfs_key key;
+
+	array_size = btrfs_super_sys_array_size(super_copy);
+
+	ptr = super_copy->sys_chunk_array;
+	cur = 0;
+
+	while (cur < array_size) {
+		disk_key = (struct btrfs_disk_key *)ptr;
+		btrfs_disk_key_to_cpu(&key, disk_key);
+
+		len = sizeof(*disk_key);
+
+		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
+			chunk = (struct btrfs_chunk *)(ptr + len);
+			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
+			len += btrfs_chunk_item_size(num_stripes);
+		} else {
+			ret = -EIO;
+			break;
+		}
+		if (key.objectid == chunk_objectid &&
+		    key.offset == chunk_offset) {
+			memmove(ptr, ptr + len, array_size - (cur + len));
+			array_size -= len;
+			btrfs_set_super_sys_array_size(super_copy, array_size);
+		} else {
+			ptr += len;
+			cur += len;
+		}
+	}
+	return ret;
+}
+
+static int btrfs_relocate_chunk(struct btrfs_root *root,
+			 u64 chunk_tree, u64 chunk_objectid,
+			 u64 chunk_offset)
+{
+	struct extent_map_tree *em_tree;
+	struct btrfs_root *extent_root;
+	struct btrfs_trans_handle *trans;
+	struct extent_map *em;
+	struct map_lookup *map;
+	int ret;
+	int i;
+
+	printk(KERN_INFO "btrfs relocating chunk %llu\n",
+	       (unsigned long long)chunk_offset);
+	root = root->fs_info->chunk_root;
+	extent_root = root->fs_info->extent_root;
+	em_tree = &root->fs_info->mapping_tree.map_tree;
+
+	/* step one, relocate all the extents inside this chunk */
+	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
+	BUG_ON(ret);
+
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(!trans);
+
+	lock_chunks(root);
+
+	/*
+	 * step two, delete the device extents and the
+	 * chunk tree entries
+	 */
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
+	spin_unlock(&em_tree->lock);
+
+	BUG_ON(em->start > chunk_offset ||
+	       em->start + em->len < chunk_offset);
+	map = (struct map_lookup *)em->bdev;
+
+	for (i = 0; i < map->num_stripes; i++) {
+		ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
+					    map->stripes[i].physical);
+		BUG_ON(ret);
+
+		if (map->stripes[i].dev) {
+			ret = btrfs_update_device(trans, map->stripes[i].dev);
+			BUG_ON(ret);
+		}
+	}
+	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
+			       chunk_offset);
+
+	BUG_ON(ret);
+
+	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
+		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
+		BUG_ON(ret);
+	}
+
+	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
+	BUG_ON(ret);
+
+	spin_lock(&em_tree->lock);
+	remove_extent_mapping(em_tree, em);
+	spin_unlock(&em_tree->lock);
+
+	kfree(map);
+	em->bdev = NULL;
+
+	/* once for the tree */
+	free_extent_map(em);
+	/* once for us */
+	free_extent_map(em);
+
+	unlock_chunks(root);
+	btrfs_end_transaction(trans, root);
+	return 0;
+}
+
+static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
+{
+	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_chunk *chunk;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	u64 chunk_tree = chunk_root->root_key.objectid;
+	u64 chunk_type;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
+	key.offset = (u64)-1;
+	key.type = BTRFS_CHUNK_ITEM_KEY;
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
+		if (ret < 0)
+			goto error;
+		BUG_ON(ret == 0);
+
+		ret = btrfs_previous_item(chunk_root, path, key.objectid,
+					  key.type);
+		if (ret < 0)
+			goto error;
+		if (ret > 0)
+			break;
+
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+		chunk = btrfs_item_ptr(leaf, path->slots[0],
+				       struct btrfs_chunk);
+		chunk_type = btrfs_chunk_type(leaf, chunk);
+		btrfs_release_path(chunk_root, path);
+
+		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
+			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
+						   found_key.objectid,
+						   found_key.offset);
+			BUG_ON(ret);
+		}
+
+		if (found_key.offset == 0)
+			break;
+		key.offset = found_key.offset - 1;
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static u64 div_factor(u64 num, int factor)
+{
+	if (factor == 10)
+		return num;
+	num *= factor;
+	do_div(num, 10);
+	return num;
+}
+
+int btrfs_balance(struct btrfs_root *dev_root)
+{
+	int ret;
+	struct list_head *cur;
+	struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
+	struct btrfs_device *device;
+	u64 old_size;
+	u64 size_to_free;
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_chunk *chunk;
+	struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key found_key;
+
+	if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
+		return -EROFS;
+
+	mutex_lock(&dev_root->fs_info->volume_mutex);
+	dev_root = dev_root->fs_info->dev_root;
+
+	/* step one make some room on all the devices */
+	list_for_each(cur, devices) {
+		device = list_entry(cur, struct btrfs_device, dev_list);
+		old_size = device->total_bytes;
+		size_to_free = div_factor(old_size, 1);
+		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
+		if (!device->writeable ||
+		    device->total_bytes - device->bytes_used > size_to_free)
+			continue;
+
+		ret = btrfs_shrink_device(device, old_size - size_to_free);
+		BUG_ON(ret);
+
+		trans = btrfs_start_transaction(dev_root, 1);
+		BUG_ON(!trans);
+
+		ret = btrfs_grow_device(trans, device, old_size);
+		BUG_ON(ret);
+
+		btrfs_end_transaction(trans, dev_root);
+	}
+
+	/* step two, relocate all the chunks */
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
+	key.offset = (u64)-1;
+	key.type = BTRFS_CHUNK_ITEM_KEY;
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
+		if (ret < 0)
+			goto error;
+
+		/*
+		 * this shouldn't happen, it means the last relocate
+		 * failed
+		 */
+		if (ret == 0)
+			break;
+
+		ret = btrfs_previous_item(chunk_root, path, 0,
+					  BTRFS_CHUNK_ITEM_KEY);
+		if (ret)
+			break;
+
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+				      path->slots[0]);
+		if (found_key.objectid != key.objectid)
+			break;
+
+		chunk = btrfs_item_ptr(path->nodes[0],
+				       path->slots[0],
+				       struct btrfs_chunk);
+		key.offset = found_key.offset;
+		/* chunk zero is special */
+		if (key.offset == 0)
+			break;
+
+		btrfs_release_path(chunk_root, path);
+		ret = btrfs_relocate_chunk(chunk_root,
+					   chunk_root->root_key.objectid,
+					   found_key.objectid,
+					   found_key.offset);
+		BUG_ON(ret);
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	mutex_unlock(&dev_root->fs_info->volume_mutex);
+	return ret;
+}
+
+/*
+ * shrinking a device means finding all of the device extents past
+ * the new size, and then following the back refs to the chunks.
+ * The chunk relocation code actually frees the device extent
+ */
+int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = device->dev_root;
+	struct btrfs_dev_extent *dev_extent = NULL;
+	struct btrfs_path *path;
+	u64 length;
+	u64 chunk_tree;
+	u64 chunk_objectid;
+	u64 chunk_offset;
+	int ret;
+	int slot;
+	struct extent_buffer *l;
+	struct btrfs_key key;
+	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
+	u64 old_total = btrfs_super_total_bytes(super_copy);
+	u64 diff = device->total_bytes - new_size;
+
+	if (new_size >= device->total_bytes)
+		return -EINVAL;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	trans = btrfs_start_transaction(root, 1);
+	if (!trans) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	path->reada = 2;
+
+	lock_chunks(root);
+
+	device->total_bytes = new_size;
+	if (device->writeable)
+		device->fs_devices->total_rw_bytes -= diff;
+	ret = btrfs_update_device(trans, device);
+	if (ret) {
+		unlock_chunks(root);
+		btrfs_end_transaction(trans, root);
+		goto done;
+	}
+	WARN_ON(diff > old_total);
+	btrfs_set_super_total_bytes(super_copy, old_total - diff);
+	unlock_chunks(root);
+	btrfs_end_transaction(trans, root);
+
+	key.objectid = device->devid;
+	key.offset = (u64)-1;
+	key.type = BTRFS_DEV_EXTENT_KEY;
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+		if (ret < 0)
+			goto done;
+
+		ret = btrfs_previous_item(root, path, 0, key.type);
+		if (ret < 0)
+			goto done;
+		if (ret) {
+			ret = 0;
+			goto done;
+		}
+
+		l = path->nodes[0];
+		slot = path->slots[0];
+		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
+
+		if (key.objectid != device->devid)
+			goto done;
+
+		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
+		length = btrfs_dev_extent_length(l, dev_extent);
+
+		if (key.offset + length <= new_size)
+			goto done;
+
+		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
+		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
+		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
+		btrfs_release_path(root, path);
+
+		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
+					   chunk_offset);
+		if (ret)
+			goto done;
+	}
+
+done:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   struct btrfs_key *key,
+			   struct btrfs_chunk *chunk, int item_size)
+{
+	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
+	struct btrfs_disk_key disk_key;
+	u32 array_size;
+	u8 *ptr;
+
+	array_size = btrfs_super_sys_array_size(super_copy);
+	if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
+		return -EFBIG;
+
+	ptr = super_copy->sys_chunk_array + array_size;
+	btrfs_cpu_key_to_disk(&disk_key, key);
+	memcpy(ptr, &disk_key, sizeof(disk_key));
+	ptr += sizeof(disk_key);
+	memcpy(ptr, chunk, item_size);
+	item_size += sizeof(disk_key);
+	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
+	return 0;
+}
+
+static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
+					int num_stripes, int sub_stripes)
+{
+	if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
+		return calc_size;
+	else if (type & BTRFS_BLOCK_GROUP_RAID10)
+		return calc_size * (num_stripes / sub_stripes);
+	else
+		return calc_size * num_stripes;
+}
+
+static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *extent_root,
+			       struct map_lookup **map_ret,
+			       u64 *num_bytes, u64 *stripe_size,
+			       u64 start, u64 type)
+{
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct btrfs_device *device = NULL;
+	struct btrfs_fs_devices *fs_devices = info->fs_devices;
+	struct list_head *cur;
+	struct map_lookup *map = NULL;
+	struct extent_map_tree *em_tree;
+	struct extent_map *em;
+	struct list_head private_devs;
+	int min_stripe_size = 1 * 1024 * 1024;
+	u64 calc_size = 1024 * 1024 * 1024;
+	u64 max_chunk_size = calc_size;
+	u64 min_free;
+	u64 avail;
+	u64 max_avail = 0;
+	u64 dev_offset;
+	int num_stripes = 1;
+	int min_stripes = 1;
+	int sub_stripes = 0;
+	int looped = 0;
+	int ret;
+	int index;
+	int stripe_len = 64 * 1024;
+
+	if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
+	    (type & BTRFS_BLOCK_GROUP_DUP)) {
+		WARN_ON(1);
+		type &= ~BTRFS_BLOCK_GROUP_DUP;
+	}
+	if (list_empty(&fs_devices->alloc_list))
+		return -ENOSPC;
+
+	if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
+		num_stripes = fs_devices->rw_devices;
+		min_stripes = 2;
+	}
+	if (type & (BTRFS_BLOCK_GROUP_DUP)) {
+		num_stripes = 2;
+		min_stripes = 2;
+	}
+	if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
+		num_stripes = min_t(u64, 2, fs_devices->rw_devices);
+		if (num_stripes < 2)
+			return -ENOSPC;
+		min_stripes = 2;
+	}
+	if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
+		num_stripes = fs_devices->rw_devices;
+		if (num_stripes < 4)
+			return -ENOSPC;
+		num_stripes &= ~(u32)1;
+		sub_stripes = 2;
+		min_stripes = 4;
+	}
+
+	if (type & BTRFS_BLOCK_GROUP_DATA) {
+		max_chunk_size = 10 * calc_size;
+		min_stripe_size = 64 * 1024 * 1024;
+	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
+		max_chunk_size = 4 * calc_size;
+		min_stripe_size = 32 * 1024 * 1024;
+	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
+		calc_size = 8 * 1024 * 1024;
+		max_chunk_size = calc_size * 2;
+		min_stripe_size = 1 * 1024 * 1024;
+	}
+
+	/* we don't want a chunk larger than 10% of writeable space */
+	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
+			     max_chunk_size);
+
+again:
+	if (!map || map->num_stripes != num_stripes) {
+		kfree(map);
+		map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
+		if (!map)
+			return -ENOMEM;
+		map->num_stripes = num_stripes;
+	}
+
+	if (calc_size * num_stripes > max_chunk_size) {
+		calc_size = max_chunk_size;
+		do_div(calc_size, num_stripes);
+		do_div(calc_size, stripe_len);
+		calc_size *= stripe_len;
+	}
+	/* we don't want tiny stripes */
+	calc_size = max_t(u64, min_stripe_size, calc_size);
+
+	do_div(calc_size, stripe_len);
+	calc_size *= stripe_len;
+
+	cur = fs_devices->alloc_list.next;
+	index = 0;
+
+	if (type & BTRFS_BLOCK_GROUP_DUP)
+		min_free = calc_size * 2;
+	else
+		min_free = calc_size;
+
+	/*
+	 * we add 1MB because we never use the first 1MB of the device, unless
+	 * we've looped, then we are likely allocating the maximum amount of
+	 * space left already
+	 */
+	if (!looped)
+		min_free += 1024 * 1024;
+
+	INIT_LIST_HEAD(&private_devs);
+	while (index < num_stripes) {
+		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
+		BUG_ON(!device->writeable);
+		if (device->total_bytes > device->bytes_used)
+			avail = device->total_bytes - device->bytes_used;
+		else
+			avail = 0;
+		cur = cur->next;
+
+		if (device->in_fs_metadata && avail >= min_free) {
+			ret = find_free_dev_extent(trans, device,
+						   min_free, &dev_offset);
+			if (ret == 0) {
+				list_move_tail(&device->dev_alloc_list,
+					       &private_devs);
+				map->stripes[index].dev = device;
+				map->stripes[index].physical = dev_offset;
+				index++;
+				if (type & BTRFS_BLOCK_GROUP_DUP) {
+					map->stripes[index].dev = device;
+					map->stripes[index].physical =
+						dev_offset + calc_size;
+					index++;
+				}
+			}
+		} else if (device->in_fs_metadata && avail > max_avail)
+			max_avail = avail;
+		if (cur == &fs_devices->alloc_list)
+			break;
+	}
+	list_splice(&private_devs, &fs_devices->alloc_list);
+	if (index < num_stripes) {
+		if (index >= min_stripes) {
+			num_stripes = index;
+			if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
+				num_stripes /= sub_stripes;
+				num_stripes *= sub_stripes;
+			}
+			looped = 1;
+			goto again;
+		}
+		if (!looped && max_avail > 0) {
+			looped = 1;
+			calc_size = max_avail;
+			goto again;
+		}
+		kfree(map);
+		return -ENOSPC;
+	}
+	map->sector_size = extent_root->sectorsize;
+	map->stripe_len = stripe_len;
+	map->io_align = stripe_len;
+	map->io_width = stripe_len;
+	map->type = type;
+	map->num_stripes = num_stripes;
+	map->sub_stripes = sub_stripes;
+
+	*map_ret = map;
+	*stripe_size = calc_size;
+	*num_bytes = chunk_bytes_by_type(type, calc_size,
+					 num_stripes, sub_stripes);
+
+	em = alloc_extent_map(GFP_NOFS);
+	if (!em) {
+		kfree(map);
+		return -ENOMEM;
+	}
+	em->bdev = (struct block_device *)map;
+	em->start = start;
+	em->len = *num_bytes;
+	em->block_start = 0;
+	em->block_len = em->len;
+
+	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
+	spin_lock(&em_tree->lock);
+	ret = add_extent_mapping(em_tree, em);
+	spin_unlock(&em_tree->lock);
+	BUG_ON(ret);
+	free_extent_map(em);
+
+	ret = btrfs_make_block_group(trans, extent_root, 0, type,
+				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
+				     start, *num_bytes);
+	BUG_ON(ret);
+
+	index = 0;
+	while (index < map->num_stripes) {
+		device = map->stripes[index].dev;
+		dev_offset = map->stripes[index].physical;
+
+		ret = btrfs_alloc_dev_extent(trans, device,
+				info->chunk_root->root_key.objectid,
+				BTRFS_FIRST_CHUNK_TREE_OBJECTID,
+				start, dev_offset, calc_size);
+		BUG_ON(ret);
+		index++;
+	}
+
+	return 0;
+}
+
+static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
+				struct btrfs_root *extent_root,
+				struct map_lookup *map, u64 chunk_offset,
+				u64 chunk_size, u64 stripe_size)
+{
+	u64 dev_offset;
+	struct btrfs_key key;
+	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
+	struct btrfs_device *device;
+	struct btrfs_chunk *chunk;
+	struct btrfs_stripe *stripe;
+	size_t item_size = btrfs_chunk_item_size(map->num_stripes);
+	int index = 0;
+	int ret;
+
+	chunk = kzalloc(item_size, GFP_NOFS);
+	if (!chunk)
+		return -ENOMEM;
+
+	index = 0;
+	while (index < map->num_stripes) {
+		device = map->stripes[index].dev;
+		device->bytes_used += stripe_size;
+		ret = btrfs_update_device(trans, device);
+		BUG_ON(ret);
+		index++;
+	}
+
+	index = 0;
+	stripe = &chunk->stripe;
+	while (index < map->num_stripes) {
+		device = map->stripes[index].dev;
+		dev_offset = map->stripes[index].physical;
+
+		btrfs_set_stack_stripe_devid(stripe, device->devid);
+		btrfs_set_stack_stripe_offset(stripe, dev_offset);
+		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
+		stripe++;
+		index++;
+	}
+
+	btrfs_set_stack_chunk_length(chunk, chunk_size);
+	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
+	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
+	btrfs_set_stack_chunk_type(chunk, map->type);
+	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
+	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
+	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
+	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
+	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
+
+	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
+	key.type = BTRFS_CHUNK_ITEM_KEY;
+	key.offset = chunk_offset;
+
+	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
+	BUG_ON(ret);
+
+	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
+		ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
+					     item_size);
+		BUG_ON(ret);
+	}
+	kfree(chunk);
+	return 0;
+}
+
+/*
+ * Chunk allocation falls into two parts. The first part does works
+ * that make the new allocated chunk useable, but not do any operation
+ * that modifies the chunk tree. The second part does the works that
+ * require modifying the chunk tree. This division is important for the
+ * bootstrap process of adding storage to a seed btrfs.
+ */
+int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *extent_root, u64 type)
+{
+	u64 chunk_offset;
+	u64 chunk_size;
+	u64 stripe_size;
+	struct map_lookup *map;
+	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
+	int ret;
+
+	ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
+			      &chunk_offset);
+	if (ret)
+		return ret;
+
+	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
+				  &stripe_size, chunk_offset, type);
+	if (ret)
+		return ret;
+
+	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
+				   chunk_size, stripe_size);
+	BUG_ON(ret);
+	return 0;
+}
+
+static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
+					 struct btrfs_root *root,
+					 struct btrfs_device *device)
+{
+	u64 chunk_offset;
+	u64 sys_chunk_offset;
+	u64 chunk_size;
+	u64 sys_chunk_size;
+	u64 stripe_size;
+	u64 sys_stripe_size;
+	u64 alloc_profile;
+	struct map_lookup *map;
+	struct map_lookup *sys_map;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_root *extent_root = fs_info->extent_root;
+	int ret;
+
+	ret = find_next_chunk(fs_info->chunk_root,
+			      BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
+	BUG_ON(ret);
+
+	alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
+			(fs_info->metadata_alloc_profile &
+			 fs_info->avail_metadata_alloc_bits);
+	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
+
+	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
+				  &stripe_size, chunk_offset, alloc_profile);
+	BUG_ON(ret);
+
+	sys_chunk_offset = chunk_offset + chunk_size;
+
+	alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
+			(fs_info->system_alloc_profile &
+			 fs_info->avail_system_alloc_bits);
+	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
+
+	ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
+				  &sys_chunk_size, &sys_stripe_size,
+				  sys_chunk_offset, alloc_profile);
+	BUG_ON(ret);
+
+	ret = btrfs_add_device(trans, fs_info->chunk_root, device);
+	BUG_ON(ret);
+
+	/*
+	 * Modifying chunk tree needs allocating new blocks from both
+	 * system block group and metadata block group. So we only can
+	 * do operations require modifying the chunk tree after both
+	 * block groups were created.
+	 */
+	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
+				   chunk_size, stripe_size);
+	BUG_ON(ret);
+
+	ret = __finish_chunk_alloc(trans, extent_root, sys_map,
+				   sys_chunk_offset, sys_chunk_size,
+				   sys_stripe_size);
+	BUG_ON(ret);
+	return 0;
+}
+
+int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
+{
+	struct extent_map *em;
+	struct map_lookup *map;
+	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
+	int readonly = 0;
+	int i;
+
+	spin_lock(&map_tree->map_tree.lock);
+	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
+	spin_unlock(&map_tree->map_tree.lock);
+	if (!em)
+		return 1;
+
+	map = (struct map_lookup *)em->bdev;
+	for (i = 0; i < map->num_stripes; i++) {
+		if (!map->stripes[i].dev->writeable) {
+			readonly = 1;
+			break;
+		}
+	}
+	free_extent_map(em);
+	return readonly;
+}
+
+void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
+{
+	extent_map_tree_init(&tree->map_tree, GFP_NOFS);
+}
+
+void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
+{
+	struct extent_map *em;
+
+	while (1) {
+		spin_lock(&tree->map_tree.lock);
+		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
+		if (em)
+			remove_extent_mapping(&tree->map_tree, em);
+		spin_unlock(&tree->map_tree.lock);
+		if (!em)
+			break;
+		kfree(em->bdev);
+		/* once for us */
+		free_extent_map(em);
+		/* once for the tree */
+		free_extent_map(em);
+	}
+}
+
+int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
+{
+	struct extent_map *em;
+	struct map_lookup *map;
+	struct extent_map_tree *em_tree = &map_tree->map_tree;
+	int ret;
+
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, logical, len);
+	spin_unlock(&em_tree->lock);
+	BUG_ON(!em);
+
+	BUG_ON(em->start > logical || em->start + em->len < logical);
+	map = (struct map_lookup *)em->bdev;
+	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
+		ret = map->num_stripes;
+	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
+		ret = map->sub_stripes;
+	else
+		ret = 1;
+	free_extent_map(em);
+	return ret;
+}
+
+static int find_live_mirror(struct map_lookup *map, int first, int num,
+			    int optimal)
+{
+	int i;
+	if (map->stripes[optimal].dev->bdev)
+		return optimal;
+	for (i = first; i < first + num; i++) {
+		if (map->stripes[i].dev->bdev)
+			return i;
+	}
+	/* we couldn't find one that doesn't fail.  Just return something
+	 * and the io error handling code will clean up eventually
+	 */
+	return optimal;
+}
+
+static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
+			     u64 logical, u64 *length,
+			     struct btrfs_multi_bio **multi_ret,
+			     int mirror_num, struct page *unplug_page)
+{
+	struct extent_map *em;
+	struct map_lookup *map;
+	struct extent_map_tree *em_tree = &map_tree->map_tree;
+	u64 offset;
+	u64 stripe_offset;
+	u64 stripe_nr;
+	int stripes_allocated = 8;
+	int stripes_required = 1;
+	int stripe_index;
+	int i;
+	int num_stripes;
+	int max_errors = 0;
+	struct btrfs_multi_bio *multi = NULL;
+
+	if (multi_ret && !(rw & (1 << BIO_RW)))
+		stripes_allocated = 1;
+again:
+	if (multi_ret) {
+		multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
+				GFP_NOFS);
+		if (!multi)
+			return -ENOMEM;
+
+		atomic_set(&multi->error, 0);
+	}
+
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, logical, *length);
+	spin_unlock(&em_tree->lock);
+
+	if (!em && unplug_page)
+		return 0;
+
+	if (!em) {
+		printk(KERN_CRIT "unable to find logical %llu len %llu\n",
+		       (unsigned long long)logical,
+		       (unsigned long long)*length);
+		BUG();
+	}
+
+	BUG_ON(em->start > logical || em->start + em->len < logical);
+	map = (struct map_lookup *)em->bdev;
+	offset = logical - em->start;
+
+	if (mirror_num > map->num_stripes)
+		mirror_num = 0;
+
+	/* if our multi bio struct is too small, back off and try again */
+	if (rw & (1 << BIO_RW)) {
+		if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
+				 BTRFS_BLOCK_GROUP_DUP)) {
+			stripes_required = map->num_stripes;
+			max_errors = 1;
+		} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
+			stripes_required = map->sub_stripes;
+			max_errors = 1;
+		}
+	}
+	if (multi_ret && rw == WRITE &&
+	    stripes_allocated < stripes_required) {
+		stripes_allocated = map->num_stripes;
+		free_extent_map(em);
+		kfree(multi);
+		goto again;
+	}
+	stripe_nr = offset;
+	/*
+	 * stripe_nr counts the total number of stripes we have to stride
+	 * to get to this block
+	 */
+	do_div(stripe_nr, map->stripe_len);
+
+	stripe_offset = stripe_nr * map->stripe_len;
+	BUG_ON(offset < stripe_offset);
+
+	/* stripe_offset is the offset of this block in its stripe*/
+	stripe_offset = offset - stripe_offset;
+
+	if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
+			 BTRFS_BLOCK_GROUP_RAID10 |
+			 BTRFS_BLOCK_GROUP_DUP)) {
+		/* we limit the length of each bio to what fits in a stripe */
+		*length = min_t(u64, em->len - offset,
+			      map->stripe_len - stripe_offset);
+	} else {
+		*length = em->len - offset;
+	}
+
+	if (!multi_ret && !unplug_page)
+		goto out;
+
+	num_stripes = 1;
+	stripe_index = 0;
+	if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
+		if (unplug_page || (rw & (1 << BIO_RW)))
+			num_stripes = map->num_stripes;
+		else if (mirror_num)
+			stripe_index = mirror_num - 1;
+		else {
+			stripe_index = find_live_mirror(map, 0,
+					    map->num_stripes,
+					    current->pid % map->num_stripes);
+		}
+
+	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
+		if (rw & (1 << BIO_RW))
+			num_stripes = map->num_stripes;
+		else if (mirror_num)
+			stripe_index = mirror_num - 1;
+
+	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
+		int factor = map->num_stripes / map->sub_stripes;
+
+		stripe_index = do_div(stripe_nr, factor);
+		stripe_index *= map->sub_stripes;
+
+		if (unplug_page || (rw & (1 << BIO_RW)))
+			num_stripes = map->sub_stripes;
+		else if (mirror_num)
+			stripe_index += mirror_num - 1;
+		else {
+			stripe_index = find_live_mirror(map, stripe_index,
+					      map->sub_stripes, stripe_index +
+					      current->pid % map->sub_stripes);
+		}
+	} else {
+		/*
+		 * after this do_div call, stripe_nr is the number of stripes
+		 * on this device we have to walk to find the data, and
+		 * stripe_index is the number of our device in the stripe array
+		 */
+		stripe_index = do_div(stripe_nr, map->num_stripes);
+	}
+	BUG_ON(stripe_index >= map->num_stripes);
+
+	for (i = 0; i < num_stripes; i++) {
+		if (unplug_page) {
+			struct btrfs_device *device;
+			struct backing_dev_info *bdi;
+
+			device = map->stripes[stripe_index].dev;
+			if (device->bdev) {
+				bdi = blk_get_backing_dev_info(device->bdev);
+				if (bdi->unplug_io_fn)
+					bdi->unplug_io_fn(bdi, unplug_page);
+			}
+		} else {
+			multi->stripes[i].physical =
+				map->stripes[stripe_index].physical +
+				stripe_offset + stripe_nr * map->stripe_len;
+			multi->stripes[i].dev = map->stripes[stripe_index].dev;
+		}
+		stripe_index++;
+	}
+	if (multi_ret) {
+		*multi_ret = multi;
+		multi->num_stripes = num_stripes;
+		multi->max_errors = max_errors;
+	}
+out:
+	free_extent_map(em);
+	return 0;
+}
+
+int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
+		      u64 logical, u64 *length,
+		      struct btrfs_multi_bio **multi_ret, int mirror_num)
+{
+	return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
+				 mirror_num, NULL);
+}
+
+int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
+		     u64 chunk_start, u64 physical, u64 devid,
+		     u64 **logical, int *naddrs, int *stripe_len)
+{
+	struct extent_map_tree *em_tree = &map_tree->map_tree;
+	struct extent_map *em;
+	struct map_lookup *map;
+	u64 *buf;
+	u64 bytenr;
+	u64 length;
+	u64 stripe_nr;
+	int i, j, nr = 0;
+
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, chunk_start, 1);
+	spin_unlock(&em_tree->lock);
+
+	BUG_ON(!em || em->start != chunk_start);
+	map = (struct map_lookup *)em->bdev;
+
+	length = em->len;
+	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
+		do_div(length, map->num_stripes / map->sub_stripes);
+	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
+		do_div(length, map->num_stripes);
+
+	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
+	BUG_ON(!buf);
+
+	for (i = 0; i < map->num_stripes; i++) {
+		if (devid && map->stripes[i].dev->devid != devid)
+			continue;
+		if (map->stripes[i].physical > physical ||
+		    map->stripes[i].physical + length <= physical)
+			continue;
+
+		stripe_nr = physical - map->stripes[i].physical;
+		do_div(stripe_nr, map->stripe_len);
+
+		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
+			stripe_nr = stripe_nr * map->num_stripes + i;
+			do_div(stripe_nr, map->sub_stripes);
+		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
+			stripe_nr = stripe_nr * map->num_stripes + i;
+		}
+		bytenr = chunk_start + stripe_nr * map->stripe_len;
+		WARN_ON(nr >= map->num_stripes);
+		for (j = 0; j < nr; j++) {
+			if (buf[j] == bytenr)
+				break;
+		}
+		if (j == nr) {
+			WARN_ON(nr >= map->num_stripes);
+			buf[nr++] = bytenr;
+		}
+	}
+
+	for (i = 0; i > nr; i++) {
+		struct btrfs_multi_bio *multi;
+		struct btrfs_bio_stripe *stripe;
+		int ret;
+
+		length = 1;
+		ret = btrfs_map_block(map_tree, WRITE, buf[i],
+				      &length, &multi, 0);
+		BUG_ON(ret);
+
+		stripe = multi->stripes;
+		for (j = 0; j < multi->num_stripes; j++) {
+			if (stripe->physical >= physical &&
+			    physical < stripe->physical + length)
+				break;
+		}
+		BUG_ON(j >= multi->num_stripes);
+		kfree(multi);
+	}
+
+	*logical = buf;
+	*naddrs = nr;
+	*stripe_len = map->stripe_len;
+
+	free_extent_map(em);
+	return 0;
+}
+
+int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
+		      u64 logical, struct page *page)
+{
+	u64 length = PAGE_CACHE_SIZE;
+	return __btrfs_map_block(map_tree, READ, logical, &length,
+				 NULL, 0, page);
+}
+
+static void end_bio_multi_stripe(struct bio *bio, int err)
+{
+	struct btrfs_multi_bio *multi = bio->bi_private;
+	int is_orig_bio = 0;
+
+	if (err)
+		atomic_inc(&multi->error);
+
+	if (bio == multi->orig_bio)
+		is_orig_bio = 1;
+
+	if (atomic_dec_and_test(&multi->stripes_pending)) {
+		if (!is_orig_bio) {
+			bio_put(bio);
+			bio = multi->orig_bio;
+		}
+		bio->bi_private = multi->private;
+		bio->bi_end_io = multi->end_io;
+		/* only send an error to the higher layers if it is
+		 * beyond the tolerance of the multi-bio
+		 */
+		if (atomic_read(&multi->error) > multi->max_errors) {
+			err = -EIO;
+		} else if (err) {
+			/*
+			 * this bio is actually up to date, we didn't
+			 * go over the max number of errors
+			 */
+			set_bit(BIO_UPTODATE, &bio->bi_flags);
+			err = 0;
+		}
+		kfree(multi);
+
+		bio_endio(bio, err);
+	} else if (!is_orig_bio) {
+		bio_put(bio);
+	}
+}
+
+struct async_sched {
+	struct bio *bio;
+	int rw;
+	struct btrfs_fs_info *info;
+	struct btrfs_work work;
+};
+
+/*
+ * see run_scheduled_bios for a description of why bios are collected for
+ * async submit.
+ *
+ * This will add one bio to the pending list for a device and make sure
+ * the work struct is scheduled.
+ */
+static noinline int schedule_bio(struct btrfs_root *root,
+				 struct btrfs_device *device,
+				 int rw, struct bio *bio)
+{
+	int should_queue = 1;
+
+	/* don't bother with additional async steps for reads, right now */
+	if (!(rw & (1 << BIO_RW))) {
+		bio_get(bio);
+		submit_bio(rw, bio);
+		bio_put(bio);
+		return 0;
+	}
+
+	/*
+	 * nr_async_bios allows us to reliably return congestion to the
+	 * higher layers.  Otherwise, the async bio makes it appear we have
+	 * made progress against dirty pages when we've really just put it
+	 * on a queue for later
+	 */
+	atomic_inc(&root->fs_info->nr_async_bios);
+	WARN_ON(bio->bi_next);
+	bio->bi_next = NULL;
+	bio->bi_rw |= rw;
+
+	spin_lock(&device->io_lock);
+
+	if (device->pending_bio_tail)
+		device->pending_bio_tail->bi_next = bio;
+
+	device->pending_bio_tail = bio;
+	if (!device->pending_bios)
+		device->pending_bios = bio;
+	if (device->running_pending)
+		should_queue = 0;
+
+	spin_unlock(&device->io_lock);
+
+	if (should_queue)
+		btrfs_queue_worker(&root->fs_info->submit_workers,
+				   &device->work);
+	return 0;
+}
+
+int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
+		  int mirror_num, int async_submit)
+{
+	struct btrfs_mapping_tree *map_tree;
+	struct btrfs_device *dev;
+	struct bio *first_bio = bio;
+	u64 logical = (u64)bio->bi_sector << 9;
+	u64 length = 0;
+	u64 map_length;
+	struct btrfs_multi_bio *multi = NULL;
+	int ret;
+	int dev_nr = 0;
+	int total_devs = 1;
+
+	length = bio->bi_size;
+	map_tree = &root->fs_info->mapping_tree;
+	map_length = length;
+
+	ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
+			      mirror_num);
+	BUG_ON(ret);
+
+	total_devs = multi->num_stripes;
+	if (map_length < length) {
+		printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
+		       "len %llu\n", (unsigned long long)logical,
+		       (unsigned long long)length,
+		       (unsigned long long)map_length);
+		BUG();
+	}
+	multi->end_io = first_bio->bi_end_io;
+	multi->private = first_bio->bi_private;
+	multi->orig_bio = first_bio;
+	atomic_set(&multi->stripes_pending, multi->num_stripes);
+
+	while (dev_nr < total_devs) {
+		if (total_devs > 1) {
+			if (dev_nr < total_devs - 1) {
+				bio = bio_clone(first_bio, GFP_NOFS);
+				BUG_ON(!bio);
+			} else {
+				bio = first_bio;
+			}
+			bio->bi_private = multi;
+			bio->bi_end_io = end_bio_multi_stripe;
+		}
+		bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
+		dev = multi->stripes[dev_nr].dev;
+		BUG_ON(rw == WRITE && !dev->writeable);
+		if (dev && dev->bdev) {
+			bio->bi_bdev = dev->bdev;
+			if (async_submit)
+				schedule_bio(root, dev, rw, bio);
+			else
+				submit_bio(rw, bio);
+		} else {
+			bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
+			bio->bi_sector = logical >> 9;
+			bio_endio(bio, -EIO);
+		}
+		dev_nr++;
+	}
+	if (total_devs == 1)
+		kfree(multi);
+	return 0;
+}
+
+struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
+				       u8 *uuid, u8 *fsid)
+{
+	struct btrfs_device *device;
+	struct btrfs_fs_devices *cur_devices;
+
+	cur_devices = root->fs_info->fs_devices;
+	while (cur_devices) {
+		if (!fsid ||
+		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
+			device = __find_device(&cur_devices->devices,
+					       devid, uuid);
+			if (device)
+				return device;
+		}
+		cur_devices = cur_devices->seed;
+	}
+	return NULL;
+}
+
+static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
+					    u64 devid, u8 *dev_uuid)
+{
+	struct btrfs_device *device;
+	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
+
+	device = kzalloc(sizeof(*device), GFP_NOFS);
+	if (!device)
+		return NULL;
+	list_add(&device->dev_list,
+		 &fs_devices->devices);
+	device->barriers = 1;
+	device->dev_root = root->fs_info->dev_root;
+	device->devid = devid;
+	device->work.func = pending_bios_fn;
+	device->fs_devices = fs_devices;
+	fs_devices->num_devices++;
+	spin_lock_init(&device->io_lock);
+	INIT_LIST_HEAD(&device->dev_alloc_list);
+	memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
+	return device;
+}
+
+static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
+			  struct extent_buffer *leaf,
+			  struct btrfs_chunk *chunk)
+{
+	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
+	struct map_lookup *map;
+	struct extent_map *em;
+	u64 logical;
+	u64 length;
+	u64 devid;
+	u8 uuid[BTRFS_UUID_SIZE];
+	int num_stripes;
+	int ret;
+	int i;
+
+	logical = key->offset;
+	length = btrfs_chunk_length(leaf, chunk);
+
+	spin_lock(&map_tree->map_tree.lock);
+	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
+	spin_unlock(&map_tree->map_tree.lock);
+
+	/* already mapped? */
+	if (em && em->start <= logical && em->start + em->len > logical) {
+		free_extent_map(em);
+		return 0;
+	} else if (em) {
+		free_extent_map(em);
+	}
+
+	map = kzalloc(sizeof(*map), GFP_NOFS);
+	if (!map)
+		return -ENOMEM;
+
+	em = alloc_extent_map(GFP_NOFS);
+	if (!em)
+		return -ENOMEM;
+	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
+	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
+	if (!map) {
+		free_extent_map(em);
+		return -ENOMEM;
+	}
+
+	em->bdev = (struct block_device *)map;
+	em->start = logical;
+	em->len = length;
+	em->block_start = 0;
+	em->block_len = em->len;
+
+	map->num_stripes = num_stripes;
+	map->io_width = btrfs_chunk_io_width(leaf, chunk);
+	map->io_align = btrfs_chunk_io_align(leaf, chunk);
+	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
+	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
+	map->type = btrfs_chunk_type(leaf, chunk);
+	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
+	for (i = 0; i < num_stripes; i++) {
+		map->stripes[i].physical =
+			btrfs_stripe_offset_nr(leaf, chunk, i);
+		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
+		read_extent_buffer(leaf, uuid, (unsigned long)
+				   btrfs_stripe_dev_uuid_nr(chunk, i),
+				   BTRFS_UUID_SIZE);
+		map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
+							NULL);
+		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
+			kfree(map);
+			free_extent_map(em);
+			return -EIO;
+		}
+		if (!map->stripes[i].dev) {
+			map->stripes[i].dev =
+				add_missing_dev(root, devid, uuid);
+			if (!map->stripes[i].dev) {
+				kfree(map);
+				free_extent_map(em);
+				return -EIO;
+			}
+		}
+		map->stripes[i].dev->in_fs_metadata = 1;
+	}
+
+	spin_lock(&map_tree->map_tree.lock);
+	ret = add_extent_mapping(&map_tree->map_tree, em);
+	spin_unlock(&map_tree->map_tree.lock);
+	BUG_ON(ret);
+	free_extent_map(em);
+
+	return 0;
+}
+
+static int fill_device_from_item(struct extent_buffer *leaf,
+				 struct btrfs_dev_item *dev_item,
+				 struct btrfs_device *device)
+{
+	unsigned long ptr;
+
+	device->devid = btrfs_device_id(leaf, dev_item);
+	device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
+	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
+	device->type = btrfs_device_type(leaf, dev_item);
+	device->io_align = btrfs_device_io_align(leaf, dev_item);
+	device->io_width = btrfs_device_io_width(leaf, dev_item);
+	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
+
+	ptr = (unsigned long)btrfs_device_uuid(dev_item);
+	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
+
+	return 0;
+}
+
+static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
+{
+	struct btrfs_fs_devices *fs_devices;
+	int ret;
+
+	mutex_lock(&uuid_mutex);
+
+	fs_devices = root->fs_info->fs_devices->seed;
+	while (fs_devices) {
+		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
+			ret = 0;
+			goto out;
+		}
+		fs_devices = fs_devices->seed;
+	}
+
+	fs_devices = find_fsid(fsid);
+	if (!fs_devices) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	fs_devices = clone_fs_devices(fs_devices);
+	if (IS_ERR(fs_devices)) {
+		ret = PTR_ERR(fs_devices);
+		goto out;
+	}
+
+	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
+				   root->fs_info->bdev_holder);
+	if (ret)
+		goto out;
+
+	if (!fs_devices->seeding) {
+		__btrfs_close_devices(fs_devices);
+		free_fs_devices(fs_devices);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	fs_devices->seed = root->fs_info->fs_devices->seed;
+	root->fs_info->fs_devices->seed = fs_devices;
+out:
+	mutex_unlock(&uuid_mutex);
+	return ret;
+}
+
+static int read_one_dev(struct btrfs_root *root,
+			struct extent_buffer *leaf,
+			struct btrfs_dev_item *dev_item)
+{
+	struct btrfs_device *device;
+	u64 devid;
+	int ret;
+	u8 fs_uuid[BTRFS_UUID_SIZE];
+	u8 dev_uuid[BTRFS_UUID_SIZE];
+
+	devid = btrfs_device_id(leaf, dev_item);
+	read_extent_buffer(leaf, dev_uuid,
+			   (unsigned long)btrfs_device_uuid(dev_item),
+			   BTRFS_UUID_SIZE);
+	read_extent_buffer(leaf, fs_uuid,
+			   (unsigned long)btrfs_device_fsid(dev_item),
+			   BTRFS_UUID_SIZE);
+
+	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
+		ret = open_seed_devices(root, fs_uuid);
+		if (ret && !btrfs_test_opt(root, DEGRADED))
+			return ret;
+	}
+
+	device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
+	if (!device || !device->bdev) {
+		if (!btrfs_test_opt(root, DEGRADED))
+			return -EIO;
+
+		if (!device) {
+			printk(KERN_WARNING "warning devid %llu missing\n",
+			       (unsigned long long)devid);
+			device = add_missing_dev(root, devid, dev_uuid);
+			if (!device)
+				return -ENOMEM;
+		}
+	}
+
+	if (device->fs_devices != root->fs_info->fs_devices) {
+		BUG_ON(device->writeable);
+		if (device->generation !=
+		    btrfs_device_generation(leaf, dev_item))
+			return -EINVAL;
+	}
+
+	fill_device_from_item(leaf, dev_item, device);
+	device->dev_root = root->fs_info->dev_root;
+	device->in_fs_metadata = 1;
+	if (device->writeable)
+		device->fs_devices->total_rw_bytes += device->total_bytes;
+	ret = 0;
+	return ret;
+}
+
+int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
+{
+	struct btrfs_dev_item *dev_item;
+
+	dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
+						     dev_item);
+	return read_one_dev(root, buf, dev_item);
+}
+
+int btrfs_read_sys_array(struct btrfs_root *root)
+{
+	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
+	struct extent_buffer *sb;
+	struct btrfs_disk_key *disk_key;
+	struct btrfs_chunk *chunk;
+	u8 *ptr;
+	unsigned long sb_ptr;
+	int ret = 0;
+	u32 num_stripes;
+	u32 array_size;
+	u32 len = 0;
+	u32 cur;
+	struct btrfs_key key;
+
+	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
+					  BTRFS_SUPER_INFO_SIZE);
+	if (!sb)
+		return -ENOMEM;
+	btrfs_set_buffer_uptodate(sb);
+	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
+	array_size = btrfs_super_sys_array_size(super_copy);
+
+	ptr = super_copy->sys_chunk_array;
+	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
+	cur = 0;
+
+	while (cur < array_size) {
+		disk_key = (struct btrfs_disk_key *)ptr;
+		btrfs_disk_key_to_cpu(&key, disk_key);
+
+		len = sizeof(*disk_key); ptr += len;
+		sb_ptr += len;
+		cur += len;
+
+		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
+			chunk = (struct btrfs_chunk *)sb_ptr;
+			ret = read_one_chunk(root, &key, sb, chunk);
+			if (ret)
+				break;
+			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
+			len = btrfs_chunk_item_size(num_stripes);
+		} else {
+			ret = -EIO;
+			break;
+		}
+		ptr += len;
+		sb_ptr += len;
+		cur += len;
+	}
+	free_extent_buffer(sb);
+	return ret;
+}
+
+int btrfs_read_chunk_tree(struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	int ret;
+	int slot;
+
+	root = root->fs_info->chunk_root;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	/* first we search for all of the device items, and then we
+	 * read in all of the chunk items.  This way we can create chunk
+	 * mappings that reference all of the devices that are afound
+	 */
+	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
+	key.offset = 0;
+	key.type = 0;
+again:
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	while (1) {
+		leaf = path->nodes[0];
+		slot = path->slots[0];
+		if (slot >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret == 0)
+				continue;
+			if (ret < 0)
+				goto error;
+			break;
+		}
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+		if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
+			if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
+				break;
+			if (found_key.type == BTRFS_DEV_ITEM_KEY) {
+				struct btrfs_dev_item *dev_item;
+				dev_item = btrfs_item_ptr(leaf, slot,
+						  struct btrfs_dev_item);
+				ret = read_one_dev(root, leaf, dev_item);
+				if (ret)
+					goto error;
+			}
+		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
+			struct btrfs_chunk *chunk;
+			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
+			ret = read_one_chunk(root, &found_key, leaf, chunk);
+			if (ret)
+				goto error;
+		}
+		path->slots[0]++;
+	}
+	if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
+		key.objectid = 0;
+		btrfs_release_path(root, path);
+		goto again;
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
new file mode 100644
index 0000000..86c44e9
--- /dev/null
+++ b/fs/btrfs/volumes.h
@@ -0,0 +1,162 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_VOLUMES_
+#define __BTRFS_VOLUMES_
+
+#include <linux/bio.h>
+#include "async-thread.h"
+
+struct buffer_head;
+struct btrfs_device {
+	struct list_head dev_list;
+	struct list_head dev_alloc_list;
+	struct btrfs_fs_devices *fs_devices;
+	struct btrfs_root *dev_root;
+	struct bio *pending_bios;
+	struct bio *pending_bio_tail;
+	int running_pending;
+	u64 generation;
+
+	int barriers;
+	int writeable;
+	int in_fs_metadata;
+
+	spinlock_t io_lock;
+
+	struct block_device *bdev;
+
+	/* the mode sent to open_bdev_exclusive */
+	fmode_t mode;
+
+	char *name;
+
+	/* the internal btrfs device id */
+	u64 devid;
+
+	/* size of the device */
+	u64 total_bytes;
+
+	/* bytes used */
+	u64 bytes_used;
+
+	/* optimal io alignment for this device */
+	u32 io_align;
+
+	/* optimal io width for this device */
+	u32 io_width;
+
+	/* minimal io size for this device */
+	u32 sector_size;
+
+	/* type and info about this device */
+	u64 type;
+
+	/* physical drive uuid (or lvm uuid) */
+	u8 uuid[BTRFS_UUID_SIZE];
+
+	struct btrfs_work work;
+};
+
+struct btrfs_fs_devices {
+	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
+
+	/* the device with this id has the most recent coyp of the super */
+	u64 latest_devid;
+	u64 latest_trans;
+	u64 num_devices;
+	u64 open_devices;
+	u64 rw_devices;
+	u64 total_rw_bytes;
+	struct block_device *latest_bdev;
+	/* all of the devices in the FS */
+	struct list_head devices;
+
+	/* devices not currently being allocated */
+	struct list_head alloc_list;
+	struct list_head list;
+
+	struct btrfs_fs_devices *seed;
+	int seeding;
+
+	int opened;
+};
+
+struct btrfs_bio_stripe {
+	struct btrfs_device *dev;
+	u64 physical;
+};
+
+struct btrfs_multi_bio {
+	atomic_t stripes_pending;
+	bio_end_io_t *end_io;
+	struct bio *orig_bio;
+	void *private;
+	atomic_t error;
+	int max_errors;
+	int num_stripes;
+	struct btrfs_bio_stripe stripes[];
+};
+
+#define btrfs_multi_bio_size(n) (sizeof(struct btrfs_multi_bio) + \
+			    (sizeof(struct btrfs_bio_stripe) * (n)))
+
+int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
+			   struct btrfs_device *device,
+			   u64 chunk_tree, u64 chunk_objectid,
+			   u64 chunk_offset, u64 start, u64 num_bytes);
+int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
+		    u64 logical, u64 *length,
+		    struct btrfs_multi_bio **multi_ret, int mirror_num);
+int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
+		     u64 chunk_start, u64 physical, u64 devid,
+		     u64 **logical, int *naddrs, int *stripe_len);
+int btrfs_read_sys_array(struct btrfs_root *root);
+int btrfs_read_chunk_tree(struct btrfs_root *root);
+int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *extent_root, u64 type);
+void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
+void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
+int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
+		  int mirror_num, int async_submit);
+int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf);
+int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
+		       fmode_t flags, void *holder);
+int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
+			  struct btrfs_fs_devices **fs_devices_ret);
+int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
+int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices);
+int btrfs_add_device(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_device *device);
+int btrfs_rm_device(struct btrfs_root *root, char *device_path);
+int btrfs_cleanup_fs_uuids(void);
+int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len);
+int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
+		      u64 logical, struct page *page);
+int btrfs_grow_device(struct btrfs_trans_handle *trans,
+		      struct btrfs_device *device, u64 new_size);
+struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
+				       u8 *uuid, u8 *fsid);
+int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
+int btrfs_init_new_device(struct btrfs_root *root, char *path);
+int btrfs_balance(struct btrfs_root *dev_root);
+void btrfs_unlock_volumes(void);
+void btrfs_lock_volumes(void);
+int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset);
+#endif
-- 
1.6.0.2

Re: [PATCH 02/35] Btrfs multi-device code

[Andrew Morton]

On Wed,  7 Jan 2009 22:56:52 -0500 Chris Mason <chris.mason@oracle.com> wrote:

> All block and extent pointers in btrfs use a logical address space.  This file
> is responsible for mapping that logical address to a physical block on disk.
> 
> It implements raid0, raid1 and raid10.  The block groups used by extent-tree.c
> are allocated from each device in chunks of ~1GB in size.  This file maintains
> the chunk tree and device allocation tree that are used to record the
> areas of each disk that are in use.
> 
> Each super block has a bootstrapping section that is used to translate
> the extents in the chunk tree.  The chunk tree is used to translate all
> of the other extents in the filesystem to a physical block on disk.

Where would one go to find out about the user-visible management of
these devices?

> This file also provides async bio submission code.  The bios are often
> sent down by async helper threads that are supposed to be working on
> checksumming or other cpu intensive tasks.  We don't want those tasks
> waiting for a free request, so a helper thread is used here that is willing
> to actually wait on the drive.

Seems odd.  Why have a thread to wait on IO?  Why not just
fire-and-forget, and then only block if some caller wishes to use the
results of that IO?

>
> ...
>
> +struct map_lookup {
> +	u64 type;
> +	int io_align;
> +	int io_width;
> +	int stripe_len;
> +	int sector_size;
> +	int num_stripes;
> +	int sub_stripes;
> +	struct btrfs_bio_stripe stripes[];
> +};

this structure looks interesting.

> +static int init_first_rw_device(struct btrfs_trans_handle *trans,
> +				struct btrfs_root *root,
> +				struct btrfs_device *device);
> +static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
> +
> +#define map_lookup_size(n) (sizeof(struct map_lookup) + \
> +			    (sizeof(struct btrfs_bio_stripe) * (n)))
> +
> +static DEFINE_MUTEX(uuid_mutex);
> +static LIST_HEAD(fs_uuids);
> +
> +void btrfs_lock_volumes(void)
> +{
> +	mutex_lock(&uuid_mutex);
> +}
> +
> +void btrfs_unlock_volumes(void)
> +{
> +	mutex_unlock(&uuid_mutex);
> +}
> +
> +static void lock_chunks(struct btrfs_root *root)
> +{
> +	mutex_lock(&root->fs_info->chunk_mutex);
> +}
> +
> +static void unlock_chunks(struct btrfs_root *root)
> +{
> +	mutex_unlock(&root->fs_info->chunk_mutex);
> +}
> +
> +static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
> +{
> +	struct btrfs_device *device;
> +	WARN_ON(fs_devices->opened);
> +	while (!list_empty(&fs_devices->devices)) {
> +		device = list_entry(fs_devices->devices.next,
> +				    struct btrfs_device, dev_list);
> +		list_del(&device->dev_list);
> +		kfree(device->name);
> +		kfree(device);
> +	}
> +	kfree(fs_devices);
> +}

Called under btrfs_lock_volumes(), one assumes/hopes.  A
WARN_ON(!mutex_is_locked()) is a nice way of documenting such things.

> +int btrfs_cleanup_fs_uuids(void)
> +{
> +	struct btrfs_fs_devices *fs_devices;
> +
> +	while (!list_empty(&fs_uuids)) {
> +		fs_devices = list_entry(fs_uuids.next,
> +					struct btrfs_fs_devices, list);
> +		list_del(&fs_devices->list);
> +		free_fs_devices(fs_devices);
> +	}
> +	return 0;
> +}

Ditto.

> +static noinline struct btrfs_device *__find_device(struct list_head *head,
> +						   u64 devid, u8 *uuid)

It's not possible for the code reader to work out why all these
noinlines are here (this reader, at least).

> +{
> +	struct btrfs_device *dev;
> +	struct list_head *cur;
> +
> +	list_for_each(cur, head) {
> +		dev = list_entry(cur, struct btrfs_device, dev_list);
> +		if (dev->devid == devid &&
> +		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
> +			return dev;
> +		}
> +	}
> +	return NULL;
> +}
> +
> +static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
> +{
> +	struct list_head *cur;
> +	struct btrfs_fs_devices *fs_devices;
> +
> +	list_for_each(cur, &fs_uuids) {
> +		fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
> +		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
> +			return fs_devices;
> +	}
> +	return NULL;
> +}

Called under btrfs_lock_volumes()...

> +/*
> + * we try to collect pending bios for a device so we don't get a large
> + * number of procs sending bios down to the same device.  This greatly
> + * improves the schedulers ability to collect and merge the bios.

eh?  The block layer already implements limiting.

Why does _not_ sending bios to a device aid the IO scheduler's merging
activity?

The comment needs help, I suspect.

> + * But, it also turns into a long list of bios to process and that is sure
> + * to eventually make the worker thread block.

Why is that bad?  If there's a large amount of IO queued then blocking
seems an appropriate thing to do.

>  The solution here is to
> + * make some progress and then put this work struct back at the end of
> + * the list if the block device is congested.  This way, multiple devices
> + * can make progress from a single worker thread.

hrm, OK.  I think.

> + */
> +static noinline int run_scheduled_bios(struct btrfs_device *device)
> +{
> +	struct bio *pending;
> +	struct backing_dev_info *bdi;
> +	struct btrfs_fs_info *fs_info;
> +	struct bio *tail;
> +	struct bio *cur;
> +	int again = 0;
> +	unsigned long num_run = 0;
> +	unsigned long limit;
> +
> +	bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;

device->bdev->bd_inode_backing_dev_info is equivalent and shorter, I suspect.

I forget why we added that.

> +	fs_info = device->dev_root->fs_info;
> +	limit = btrfs_async_submit_limit(fs_info);
> +	limit = limit * 2 / 3;

??

> +loop:
> +	spin_lock(&device->io_lock);
> +
> +	/* take all the bios off the list at once and process them
> +	 * later on (without the lock held).  But, remember the
> +	 * tail and other pointers so the bios can be properly reinserted
> +	 * into the list if we hit congestion
> +	 */
> +	pending = device->pending_bios;
> +	tail = device->pending_bio_tail;
> +	WARN_ON(pending && !tail);
> +	device->pending_bios = NULL;
> +	device->pending_bio_tail = NULL;
> +
> +	/*
> +	 * if pending was null this time around, no bios need processing
> +	 * at all and we can stop.  Otherwise it'll loop back up again
> +	 * and do an additional check so no bios are missed.
> +	 *
> +	 * device->running_pending is used to synchronize with the
> +	 * schedule_bio code.
> +	 */
> +	if (pending) {
> +		again = 1;
> +		device->running_pending = 1;
> +	} else {
> +		again = 0;
> +		device->running_pending = 0;
> +	}
> +	spin_unlock(&device->io_lock);
> +
> +	while (pending) {
> +		cur = pending;
> +		pending = pending->bi_next;
> +		cur->bi_next = NULL;
> +		atomic_dec(&fs_info->nr_async_bios);
> +
> +		if (atomic_read(&fs_info->nr_async_bios) < limit &&
> +		    waitqueue_active(&fs_info->async_submit_wait))
> +			wake_up(&fs_info->async_submit_wait);
> +
> +		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
> +		bio_get(cur);
> +		submit_bio(cur->bi_rw, cur);
> +		bio_put(cur);

What do the bio_get()/bio_put() do?  Looks wrong, or unneeded.

> +		num_run++;
> +
> +		/*
> +		 * we made progress, there is more work to do and the bdi
> +		 * is now congested.  Back off and let other work structs
> +		 * run instead
> +		 */
> +		if (pending && bdi_write_congested(bdi) &&

Does this function handle only writes?  If it handles reads, why do we
test only write congestion?  (please add comments if the questions
aren't painfully stupid).

> +		    fs_info->fs_devices->open_devices > 1) {
> +			struct bio *old_head;
> +
> +			spin_lock(&device->io_lock);
> +
> +			old_head = device->pending_bios;
> +			device->pending_bios = pending;
> +			if (device->pending_bio_tail)
> +				tail->bi_next = old_head;
> +			else
> +				device->pending_bio_tail = tail;
> +
> +			spin_unlock(&device->io_lock);
> +			btrfs_requeue_work(&device->work);
> +			goto done;
> +		}
> +	}
> +	if (again)
> +		goto loop;
> +done:
> +	return 0;
> +}
> +
> +static void pending_bios_fn(struct btrfs_work *work)
> +{
> +	struct btrfs_device *device;
> +
> +	device = container_of(work, struct btrfs_device, work);
> +	run_scheduled_bios(device);
> +}
> +
> +static noinline int device_list_add(const char *path,
> +			   struct btrfs_super_block *disk_super,
> +			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
> +{
> +	struct btrfs_device *device;
> +	struct btrfs_fs_devices *fs_devices;
> +	u64 found_transid = btrfs_super_generation(disk_super);
> +
> +	fs_devices = find_fsid(disk_super->fsid);
> +	if (!fs_devices) {
> +		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);

Wondering why it's GFP_NOFS.

Presumably this function is supposed to be called under
btrfs_lock_volumes(), and we also call that function on the memory
reclaim path?

> +		if (!fs_devices)
> +			return -ENOMEM;
> +		INIT_LIST_HEAD(&fs_devices->devices);
> +		INIT_LIST_HEAD(&fs_devices->alloc_list);
> +		list_add(&fs_devices->list, &fs_uuids);
> +		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
> +		fs_devices->latest_devid = devid;
> +		fs_devices->latest_trans = found_transid;
> +		device = NULL;
> +	} else {
> +		device = __find_device(&fs_devices->devices, devid,
> +				       disk_super->dev_item.uuid);
> +	}
> +	if (!device) {
> +		if (fs_devices->opened)
> +			return -EBUSY;
> +
> +		device = kzalloc(sizeof(*device), GFP_NOFS);
> +		if (!device) {
> +			/* we can safely leave the fs_devices entry around */
> +			return -ENOMEM;
> +		}
> +		device->devid = devid;
> +		device->work.func = pending_bios_fn;
> +		memcpy(device->uuid, disk_super->dev_item.uuid,
> +		       BTRFS_UUID_SIZE);
> +		device->barriers = 1;
> +		spin_lock_init(&device->io_lock);
> +		device->name = kstrdup(path, GFP_NOFS);
> +		if (!device->name) {
> +			kfree(device);
> +			return -ENOMEM;
> +		}
> +		INIT_LIST_HEAD(&device->dev_alloc_list);
> +		list_add(&device->dev_list, &fs_devices->devices);
> +		device->fs_devices = fs_devices;
> +		fs_devices->num_devices++;
> +	}
> +
> +	if (found_transid > fs_devices->latest_trans) {
> +		fs_devices->latest_devid = devid;
> +		fs_devices->latest_trans = found_transid;
> +	}
> +	*fs_devices_ret = fs_devices;
> +	return 0;
> +}
> +
> +static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)

I wonder why this function exists.

> +{
> +	struct btrfs_fs_devices *fs_devices;
> +	struct btrfs_device *device;
> +	struct btrfs_device *orig_dev;
> +
> +	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
> +	if (!fs_devices)
> +		return ERR_PTR(-ENOMEM);
> +
> +	INIT_LIST_HEAD(&fs_devices->devices);
> +	INIT_LIST_HEAD(&fs_devices->alloc_list);
> +	INIT_LIST_HEAD(&fs_devices->list);
> +	fs_devices->latest_devid = orig->latest_devid;
> +	fs_devices->latest_trans = orig->latest_trans;
> +	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
> +
> +	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
> +		device = kzalloc(sizeof(*device), GFP_NOFS);
> +		if (!device)
> +			goto error;
> +
> +		device->name = kstrdup(orig_dev->name, GFP_NOFS);
> +		if (!device->name)
> +			goto error;

We leaked *device?

> +		device->devid = orig_dev->devid;
> +		device->work.func = pending_bios_fn;
> +		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
> +		device->barriers = 1;
> +		spin_lock_init(&device->io_lock);
> +		INIT_LIST_HEAD(&device->dev_list);
> +		INIT_LIST_HEAD(&device->dev_alloc_list);
> +
> +		list_add(&device->dev_list, &fs_devices->devices);
> +		device->fs_devices = fs_devices;
> +		fs_devices->num_devices++;
> +	}
> +	return fs_devices;
> +error:
> +	free_fs_devices(fs_devices);
> +	return ERR_PTR(-ENOMEM);
> +}
> +
> +int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)

whatdoesthisdo

> +{
> +	struct list_head *tmp;
> +	struct list_head *cur;
> +	struct btrfs_device *device;
> +
> +	mutex_lock(&uuid_mutex);

btrfs_lock_volumes()?

> +again:
> +	list_for_each_safe(cur, tmp, &fs_devices->devices) {
> +		device = list_entry(cur, struct btrfs_device, dev_list);
> +		if (device->in_fs_metadata)
> +			continue;
> +
> +		if (device->bdev) {
> +			close_bdev_exclusive(device->bdev, device->mode);
> +			device->bdev = NULL;
> +			fs_devices->open_devices--;
> +		}
> +		if (device->writeable) {
> +			list_del_init(&device->dev_alloc_list);
> +			device->writeable = 0;
> +			fs_devices->rw_devices--;
> +		}

Why are we bothering to modify *device?  It's about to be freed.

> +		list_del_init(&device->dev_list);
> +		fs_devices->num_devices--;
> +		kfree(device->name);
> +		kfree(device);
> +	}
> +
> +	if (fs_devices->seed) {
> +		fs_devices = fs_devices->seed;
> +		goto again;
> +	}
> +
> +	mutex_unlock(&uuid_mutex);

...

> +	return 0;
> +}
> +
> +static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
> +{
> +	struct list_head *cur;
> +	struct btrfs_device *device;
> +
> +	if (--fs_devices->opened > 0)
> +		return 0;
> +
> +	list_for_each(cur, &fs_devices->devices) {
> +		device = list_entry(cur, struct btrfs_device, dev_list);
> +		if (device->bdev) {

This reader is wondering what btrfs_device.bdev==NULL means.  How would
one go about determining this?

> +			close_bdev_exclusive(device->bdev, device->mode);
> +			fs_devices->open_devices--;
> +		}
> +		if (device->writeable) {
> +			list_del_init(&device->dev_alloc_list);
> +			fs_devices->rw_devices--;
> +		}
> +
> +		device->bdev = NULL;
> +		device->writeable = 0;
> +		device->in_fs_metadata = 0;

What does in_fs_metadata mean?

Perhaps those three fields should have type `bool'.

> +	}
> +	WARN_ON(fs_devices->open_devices);
> +	WARN_ON(fs_devices->rw_devices);
> +	fs_devices->opened = 0;
> +	fs_devices->seeding = 0;

What does the `seeding' field mean?  It's a boolean, I assume..

> +	return 0;
> +}
> +
> +int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
> +{
> +	struct btrfs_fs_devices *seed_devices = NULL;
> +	int ret;
> +
> +	mutex_lock(&uuid_mutex);

btrfs_lock_volumes()?  (Please check whole fs for consistency here)

Seems strange that a lock called "uuid_mutex" is protecting a pile of
things which don't seem very uuid-related.

> +	ret = __btrfs_close_devices(fs_devices);
> +	if (!fs_devices->opened) {
> +		seed_devices = fs_devices->seed;
> +		fs_devices->seed = NULL;
> +	}
> +	mutex_unlock(&uuid_mutex);
> +
> +	while (seed_devices) {
> +		fs_devices = seed_devices;
> +		seed_devices = fs_devices->seed;
> +		__btrfs_close_devices(fs_devices);
> +		free_fs_devices(fs_devices);
> +	}
> +	return ret;
> +}
> +
> +static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
> +				fmode_t flags, void *holder)
> +{
> +	struct block_device *bdev;
> +	struct list_head *head = &fs_devices->devices;
> +	struct list_head *cur;
> +	struct btrfs_device *device;
> +	struct block_device *latest_bdev = NULL;
> +	struct buffer_head *bh;
> +	struct btrfs_super_block *disk_super;
> +	u64 latest_devid = 0;
> +	u64 latest_transid = 0;
> +	u64 devid;
> +	int seeding = 1;
> +	int ret = 0;
> +
> +	list_for_each(cur, head) {
> +		device = list_entry(cur, struct btrfs_device, dev_list);

list_for_each_entry()

> +		if (device->bdev)
> +			continue;
> +		if (!device->name)
> +			continue;

What's the algorithm for understanding the above four lines?

> +		bdev = open_bdev_exclusive(device->name, flags, holder);
> +		if (IS_ERR(bdev)) {
> +			printk(KERN_INFO "open %s failed\n", device->name);
> +			goto error;
> +		}
> +		set_blocksize(bdev, 4096);

hm.  What's the situation with btrfs blocksizes?

> +		bh = btrfs_read_dev_super(bdev);
> +		if (!bh)
> +			goto error_close;
> +
> +		disk_super = (struct btrfs_super_block *)bh->b_data;
> +		devid = le64_to_cpu(disk_super->dev_item.devid);
> +		if (devid != device->devid)
> +			goto error_brelse;
> +
> +		if (memcmp(device->uuid, disk_super->dev_item.uuid,
> +			   BTRFS_UUID_SIZE))
> +			goto error_brelse;
> +
> +		device->generation = btrfs_super_generation(disk_super);
> +		if (!latest_transid || device->generation > latest_transid) {
> +			latest_devid = devid;
> +			latest_transid = device->generation;
> +			latest_bdev = bdev;
> +		}
> +
> +		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
> +			device->writeable = 0;
> +		} else {
> +			device->writeable = !bdev_read_only(bdev);
> +			seeding = 0;
> +		}
> +
> +		device->bdev = bdev;
> +		device->in_fs_metadata = 0;
> +		device->mode = flags;
> +
> +		fs_devices->open_devices++;
> +		if (device->writeable) {
> +			fs_devices->rw_devices++;
> +			list_add(&device->dev_alloc_list,
> +				 &fs_devices->alloc_list);
> +		}
> +		continue;
> +
> +error_brelse:
> +		brelse(bh);

brelse() is only needed in situations where the pointer might be NULL.

> +error_close:
> +		close_bdev_exclusive(bdev, FMODE_READ);
> +error:
> +		continue;
> +	}
> +	if (fs_devices->open_devices == 0) {
> +		ret = -EIO;
> +		goto out;
> +	}
> +	fs_devices->seeding = seeding;
> +	fs_devices->opened = 1;
> +	fs_devices->latest_bdev = latest_bdev;
> +	fs_devices->latest_devid = latest_devid;
> +	fs_devices->latest_trans = latest_transid;
> +	fs_devices->total_rw_bytes = 0;
> +out:
> +	return ret;
> +}
> +
> +int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
> +		       fmode_t flags, void *holder)
> +{
> +	int ret;
> +
> +	mutex_lock(&uuid_mutex);
> +	if (fs_devices->opened) {
> +		fs_devices->opened++;
> +		ret = 0;
> +	} else {
> +		ret = __btrfs_open_devices(fs_devices, flags, holder);
> +	}
> +	mutex_unlock(&uuid_mutex);
> +	return ret;
> +}
> +
> +int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
> +			  struct btrfs_fs_devices **fs_devices_ret)
> +{
> +	struct btrfs_super_block *disk_super;
> +	struct block_device *bdev;
> +	struct buffer_head *bh;
> +	int ret;
> +	u64 devid;
> +	u64 transid;
> +
> +	mutex_lock(&uuid_mutex);
> +
> +	bdev = open_bdev_exclusive(path, flags, holder);
> +
> +	if (IS_ERR(bdev)) {
> +		ret = PTR_ERR(bdev);
> +		goto error;
> +	}
> +
> +	ret = set_blocksize(bdev, 4096);
> +	if (ret)
> +		goto error_close;
> +	bh = btrfs_read_dev_super(bdev);
> +	if (!bh) {
> +		ret = -EIO;
> +		goto error_close;
> +	}
> +	disk_super = (struct btrfs_super_block *)bh->b_data;

<looks at struct btrfs_super_block>

It needs work, I think.  This:

	__le64 compat_ro_flags;
	__le64 incompat_flags;
	__le16 csum_type;
	u8 root_level;
	u8 chunk_root_level;
	u8 log_root_level;
	struct btrfs_dev_item dev_item;

doesn't look like it will successfully port between architectures. 
Also `struct btrfs_dev_item' has three le32's wedged between le64's.

I suppose that the __packed will help in some cases (but not the
btrfs_super_block, surely).  However it seems unwise/incautious/ugly to
rely upon __packed in this manner.


Please do s/__attribute__ ((__packed__))/__packed/g

> +	devid = le64_to_cpu(disk_super->dev_item.devid);
> +	transid = btrfs_super_generation(disk_super);
> +	if (disk_super->label[0])
> +		printk(KERN_INFO "device label %s ", disk_super->label);
> +	else {
> +		/* FIXME, make a readl uuid parser */

what's that?

> +		printk(KERN_INFO "device fsid %llx-%llx ",
> +		       *(unsigned long long *)disk_super->fsid,
> +		       *(unsigned long long *)(disk_super->fsid + 8));
> +	}
> +	printk(KERN_INFO "devid %llu transid %llu %s\n",
> +	       (unsigned long long)devid, (unsigned long long)transid, path);
> +	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
> +
> +	brelse(bh);

bh is non-NULL here

> +error_close:
> +	close_bdev_exclusive(bdev, flags);
> +error:
> +	mutex_unlock(&uuid_mutex);
> +	return ret;
> +}
> +
> +/*
> + * this uses a pretty simple search, the expectation is that it is
> + * called very infrequently and that a given device has a small number
> + * of extents
> + */
> +static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
> +					 struct btrfs_device *device,
> +					 u64 num_bytes, u64 *start)
> +{
> +	struct btrfs_key key;
> +	struct btrfs_root *root = device->dev_root;
> +	struct btrfs_dev_extent *dev_extent = NULL;
> +	struct btrfs_path *path;
> +	u64 hole_size = 0;
> +	u64 last_byte = 0;
> +	u64 search_start = 0;
> +	u64 search_end = device->total_bytes;
> +	int ret;
> +	int slot = 0;
> +	int start_found;
> +	struct extent_buffer *l;
> +
> +	path = btrfs_alloc_path();
> +	if (!path)
> +		return -ENOMEM;
> +	path->reada = 2;
> +	start_found = 0;
> +
> +	/* FIXME use last free of some kind */
> +
> +	/* we don't want to overwrite the superblock on the drive,
> +	 * so we make sure to start at an offset of at least 1MB
> +	 */
> +	search_start = max((u64)1024 * 1024, search_start);

max_t

> +	if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
> +		search_start = max(root->fs_info->alloc_start, search_start);
> +
> +	key.objectid = device->devid;
> +	key.offset = search_start;
> +	key.type = BTRFS_DEV_EXTENT_KEY;
> +	ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
> +	if (ret < 0)
> +		goto error;
> +	ret = btrfs_previous_item(root, path, 0, key.type);
> +	if (ret < 0)
> +		goto error;
> +	l = path->nodes[0];
> +	btrfs_item_key_to_cpu(l, &key, path->slots[0]);
> +	while (1) {
> +		l = path->nodes[0];
> +		slot = path->slots[0];
> +		if (slot >= btrfs_header_nritems(l)) {
> +			ret = btrfs_next_leaf(root, path);
> +			if (ret == 0)
> +				continue;
> +			if (ret < 0)
> +				goto error;
> +no_more_items:
> +			if (!start_found) {
> +				if (search_start >= search_end) {
> +					ret = -ENOSPC;

Hey.  Shouldn't that go BUG()?

> +					goto error;
> +				}
> +				*start = search_start;
> +				start_found = 1;
> +				goto check_pending;
> +			}
> +			*start = last_byte > search_start ?
> +				last_byte : search_start;

max()?

> +			if (search_end <= *start) {
> +				ret = -ENOSPC;
> +				goto error;
> +			}
> +			goto check_pending;
> +		}
> +		btrfs_item_key_to_cpu(l, &key, slot);
> +
> +		if (key.objectid < device->devid)
> +			goto next;
> +
> +		if (key.objectid > device->devid)
> +			goto no_more_items;
> +
> +		if (key.offset >= search_start && key.offset > last_byte &&
> +		    start_found) {
> +			if (last_byte < search_start)
> +				last_byte = search_start;
> +			hole_size = key.offset - last_byte;
> +			if (key.offset > last_byte &&
> +			    hole_size >= num_bytes) {
> +				*start = last_byte;
> +				goto check_pending;
> +			}
> +		}
> +		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
> +			goto next;
> +
> +		start_found = 1;
> +		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
> +		last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
> +next:
> +		path->slots[0]++;
> +		cond_resched();
> +	}
> +check_pending:
> +	/* we have to make sure we didn't find an extent that has already
> +	 * been allocated by the map tree or the original allocation
> +	 */
> +	BUG_ON(*start < search_start);

Is it triggerable by on-disk corruption?

> +	if (*start + num_bytes > search_end) {
> +		ret = -ENOSPC;
> +		goto error;
> +	}
> +	/* check for pending inserts here */
> +	ret = 0;
> +
> +error:
> +	btrfs_free_path(path);
> +	return ret;
> +}
> +
> +static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
> +			  struct btrfs_device *device,
> +			  u64 start)
> +{
> +	int ret;
> +	struct btrfs_path *path;
> +	struct btrfs_root *root = device->dev_root;
> +	struct btrfs_key key;
> +	struct btrfs_key found_key;
> +	struct extent_buffer *leaf = NULL;
> +	struct btrfs_dev_extent *extent = NULL;
> +
> +	path = btrfs_alloc_path();
> +	if (!path)
> +		return -ENOMEM;
> +
> +	key.objectid = device->devid;
> +	key.offset = start;
> +	key.type = BTRFS_DEV_EXTENT_KEY;
> +
> +	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);

<looks at btrfs_search_slot>

 * if ins_len > 0, nodes and leaves will be split as we walk down the
 * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
 * possible)

The (ab)use of the sign of formal arg `ins_len' seems a bit nasty?

> +	if (ret > 0) {
> +		ret = btrfs_previous_item(root, path, key.objectid,
> +					  BTRFS_DEV_EXTENT_KEY);
> +		BUG_ON(ret);
> +		leaf = path->nodes[0];
> +		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
> +		extent = btrfs_item_ptr(leaf, path->slots[0],
> +					struct btrfs_dev_extent);
> +		BUG_ON(found_key.offset > start || found_key.offset +
> +		       btrfs_dev_extent_length(leaf, extent) < start);
> +		ret = 0;
> +	} else if (ret == 0) {
> +		leaf = path->nodes[0];
> +		extent = btrfs_item_ptr(leaf, path->slots[0],
> +					struct btrfs_dev_extent);
> +	}
> +	BUG_ON(ret);
> +
> +	if (device->bytes_used > 0)
> +		device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
> +	ret = btrfs_del_item(trans, root, path);
> +	BUG_ON(ret);
> +
> +	btrfs_free_path(path);
> +	return ret;
> +}
> +
>
> ...
>
> +static noinline int find_next_chunk(struct btrfs_root *root,
> +				    u64 objectid, u64 *offset)
> +{
> +	struct btrfs_path *path;
> +	int ret;
> +	struct btrfs_key key;
> +	struct btrfs_chunk *chunk;
> +	struct btrfs_key found_key;
> +
> +	path = btrfs_alloc_path();
> +	BUG_ON(!path);

Stuff like this will eventually have to be weeded out and fixed.  It at
least needs a big FIXME the instant it is typed in.  Quietly slipping
it in there in this manner is just creating more work for yourselves.

> +	key.objectid = objectid;
> +	key.offset = (u64)-1;
> +	key.type = BTRFS_CHUNK_ITEM_KEY;
> +
> +	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
> +	if (ret < 0)
> +		goto error;
> +
> +	BUG_ON(ret == 0);

Is this triggerable by a corrupted fs?

> +	ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
> +	if (ret) {
> +		*offset = 0;
> +	} else {
> +		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
> +				      path->slots[0]);
> +		if (found_key.objectid != objectid)
> +			*offset = 0;
> +		else {
> +			chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
> +					       struct btrfs_chunk);
> +			*offset = found_key.offset +
> +				btrfs_chunk_length(path->nodes[0], chunk);
> +		}
> +	}
> +	ret = 0;
> +error:
> +	btrfs_free_path(path);
> +	return ret;
> +}
> +
>
> ...
>
> +
> +/*
> + * the device information is stored in the chunk root
> + * the btrfs_device struct should be fully filled in
> + */
> +int btrfs_add_device(struct btrfs_trans_handle *trans,
> +		     struct btrfs_root *root,
> +		     struct btrfs_device *device)
> +{
> +	int ret;
> +	struct btrfs_path *path;
> +	struct btrfs_dev_item *dev_item;
> +	struct extent_buffer *leaf;
> +	struct btrfs_key key;
> +	unsigned long ptr;
> +
> +	root = root->fs_info->chunk_root;
> +
> +	path = btrfs_alloc_path();
> +	if (!path)
> +		return -ENOMEM;
> +
> +	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
> +	key.type = BTRFS_DEV_ITEM_KEY;
> +	key.offset = device->devid;
> +
> +	ret = btrfs_insert_empty_item(trans, root, path, &key,
> +				      sizeof(*dev_item));
> +	if (ret)
> +		goto out;
> +
> +	leaf = path->nodes[0];
> +	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
> +
> +	btrfs_set_device_id(leaf, dev_item, device->devid);
> +	btrfs_set_device_generation(leaf, dev_item, 0);
> +	btrfs_set_device_type(leaf, dev_item, device->type);
> +	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
> +	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
> +	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
> +	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
> +	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
> +	btrfs_set_device_group(leaf, dev_item, 0);
> +	btrfs_set_device_seek_speed(leaf, dev_item, 0);
> +	btrfs_set_device_bandwidth(leaf, dev_item, 0);
> +	btrfs_set_device_start_offset(leaf, dev_item, 0);
> +
> +	ptr = (unsigned long)btrfs_device_uuid(dev_item);
> +	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
> +	ptr = (unsigned long)btrfs_device_fsid(dev_item);
> +	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
> +	btrfs_mark_buffer_dirty(leaf);

I expected a function called btrfs_mark_buffer_dirty() to be a wrapper
around mark_buffer_dirty().

> +	ret = 0;
> +out:
> +	btrfs_free_path(path);
> +	return ret;
> +}
> +
>
> ...
>
> +
> +int btrfs_rm_device(struct btrfs_root *root, char *device_path)
> +{
> +	struct btrfs_device *device;
> +	struct btrfs_device *next_device;
> +	struct block_device *bdev;
> +	struct buffer_head *bh = NULL;
> +	struct btrfs_super_block *disk_super;
> +	u64 all_avail;
> +	u64 devid;
> +	u64 num_devices;
> +	u8 *dev_uuid;
> +	int ret = 0;
> +
> +	mutex_lock(&uuid_mutex);

btrfs_lock_volumes()

> +	mutex_lock(&root->fs_info->volume_mutex);

lock_chunks() (which needs renaming)

> +	all_avail = root->fs_info->avail_data_alloc_bits |
> +		root->fs_info->avail_system_alloc_bits |
> +		root->fs_info->avail_metadata_alloc_bits;
> +
> +	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
> +	    root->fs_info->fs_devices->rw_devices <= 4) {
> +		printk(KERN_ERR "btrfs: unable to go below four devices "
> +		       "on raid10\n");
> +		ret = -EINVAL;
> +		goto out;
> +	}
> +
> +	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
> +	    root->fs_info->fs_devices->rw_devices <= 2) {
> +		printk(KERN_ERR "btrfs: unable to go below two "
> +		       "devices on raid1\n");
> +		ret = -EINVAL;
> +		goto out;
> +	}
> +
> +	if (strcmp(device_path, "missing") == 0) {

<wonders what interface all this is implementing>

> +		struct list_head *cur;
> +		struct list_head *devices;
> +		struct btrfs_device *tmp;
> +
> +		device = NULL;
> +		devices = &root->fs_info->fs_devices->devices;
> +		list_for_each(cur, devices) {
> +			tmp = list_entry(cur, struct btrfs_device, dev_list);
> +			if (tmp->in_fs_metadata && !tmp->bdev) {
> +				device = tmp;
> +				break;
> +			}
> +		}
> +		bdev = NULL;
> +		bh = NULL;
> +		disk_super = NULL;
> +		if (!device) {
> +			printk(KERN_ERR "btrfs: no missing devices found to "
> +			       "remove\n");
> +			goto out;
> +		}
> +	} else {
> +		bdev = open_bdev_exclusive(device_path, FMODE_READ,
> +				      root->fs_info->bdev_holder);
> +		if (IS_ERR(bdev)) {
> +			ret = PTR_ERR(bdev);
> +			goto out;
> +		}
> +
> +		set_blocksize(bdev, 4096);
> +		bh = btrfs_read_dev_super(bdev);
> +		if (!bh) {
> +			ret = -EIO;
> +			goto error_close;
> +		}
> +		disk_super = (struct btrfs_super_block *)bh->b_data;
> +		devid = le64_to_cpu(disk_super->dev_item.devid);
> +		dev_uuid = disk_super->dev_item.uuid;
> +		device = btrfs_find_device(root, devid, dev_uuid,
> +					   disk_super->fsid);
> +		if (!device) {
> +			ret = -ENOENT;
> +			goto error_brelse;
> +		}
> +	}
> +
> +	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
> +		printk(KERN_ERR "btrfs: unable to remove the only writeable "
> +		       "device\n");
> +		ret = -EINVAL;
> +		goto error_brelse;
> +	}
> +
> +	if (device->writeable) {
> +		list_del_init(&device->dev_alloc_list);
> +		root->fs_info->fs_devices->rw_devices--;
> +	}
> +
> +	ret = btrfs_shrink_device(device, 0);
> +	if (ret)
> +		goto error_brelse;
> +
> +	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
> +	if (ret)
> +		goto error_brelse;
> +
> +	device->in_fs_metadata = 0;
> +	list_del_init(&device->dev_list);
> +	device->fs_devices->num_devices--;
> +
> +	next_device = list_entry(root->fs_info->fs_devices->devices.next,
> +				 struct btrfs_device, dev_list);
> +	if (device->bdev == root->fs_info->sb->s_bdev)
> +		root->fs_info->sb->s_bdev = next_device->bdev;
> +	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
> +		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
> +
> +	if (device->bdev) {
> +		close_bdev_exclusive(device->bdev, device->mode);
> +		device->bdev = NULL;
> +		device->fs_devices->open_devices--;
> +	}
> +
> +	num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
> +	btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
> +
> +	if (device->fs_devices->open_devices == 0) {
> +		struct btrfs_fs_devices *fs_devices;
> +		fs_devices = root->fs_info->fs_devices;
> +		while (fs_devices) {
> +			if (fs_devices->seed == device->fs_devices)
> +				break;
> +			fs_devices = fs_devices->seed;
> +		}
> +		fs_devices->seed = device->fs_devices->seed;
> +		device->fs_devices->seed = NULL;
> +		__btrfs_close_devices(device->fs_devices);
> +		free_fs_devices(device->fs_devices);
> +	}
> +
> +	/*
> +	 * at this point, the device is zero sized.  We want to
> +	 * remove it from the devices list and zero out the old super
> +	 */
> +	if (device->writeable) {
> +		/* make sure this device isn't detected as part of
> +		 * the FS anymore
> +		 */
> +		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
> +		set_buffer_dirty(bh);
> +		sync_dirty_buffer(bh);
> +	}
> +
> +	kfree(device->name);
> +	kfree(device);
> +	ret = 0;
> +
> +error_brelse:
> +	brelse(bh);
> +error_close:
> +	if (bdev)
> +		close_bdev_exclusive(bdev, FMODE_READ);
> +out:
> +	mutex_unlock(&root->fs_info->volume_mutex);
> +	mutex_unlock(&uuid_mutex);
> +	return ret;
> +}
> +
> +/*
> + * does all the dirty work required for changing file system's UUID.
> + */
> +static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
> +				struct btrfs_root *root)
> +{
> +	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
> +	struct btrfs_fs_devices *old_devices;
> +	struct btrfs_fs_devices *seed_devices;
> +	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
> +	struct btrfs_device *device;
> +	u64 super_flags;
> +
> +	BUG_ON(!mutex_is_locked(&uuid_mutex));

Given the (regrettable?) existence of btrfs_lock_volumes() and
btrfs_unlock_volumes(), it would make sense to add another regrettable
wrapper for the above operation.

> +	if (!fs_devices->seeding)
> +		return -EINVAL;

<Again wonders what "seeding" means>

> +	seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
> +	if (!seed_devices)
> +		return -ENOMEM;
> +
> +	old_devices = clone_fs_devices(fs_devices);
> +	if (IS_ERR(old_devices)) {
> +		kfree(seed_devices);
> +		return PTR_ERR(old_devices);
> +	}
> +
> +	list_add(&old_devices->list, &fs_uuids);
> +
> +	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
> +	seed_devices->opened = 1;
> +	INIT_LIST_HEAD(&seed_devices->devices);
> +	INIT_LIST_HEAD(&seed_devices->alloc_list);
> +	list_splice_init(&fs_devices->devices, &seed_devices->devices);
> +	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
> +	list_for_each_entry(device, &seed_devices->devices, dev_list) {
> +		device->fs_devices = seed_devices;
> +	}

unneeded braces

<runs checkpatch>

<checkpatch misses this - regression?>


#2001: FILE: fs/btrfs/volumes.c:1930:
+               return calc_size * num_stripes;
                                 ^

ERROR: space prohibited after that '*' (ctx:WxW)
#2072: FILE: fs/btrfs/volumes.c:2001:
+               max_chunk_size = calc_size * 2;
                                           ^

ERROR: space prohibited after that '*' (ctx:WxW)
#2089: FILE: fs/btrfs/volumes.c:2018:
+       if (calc_size * num_stripes > max_chunk_size) {
	                      ^
	
ERROR: space prohibited after that '*' (ctx:WxW)
#2105: FILE: fs/btrfs/volumes.c:2034:
+               min_free = calc_size * 2;
	                                     ^

hm, they're all wrong.	

> +	fs_devices->seeding = 0;
> +	fs_devices->num_devices = 0;
> +	fs_devices->open_devices = 0;
> +	fs_devices->seed = seed_devices;
> +
> +	generate_random_uuid(fs_devices->fsid);
> +	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
> +	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
> +	super_flags = btrfs_super_flags(disk_super) &
> +		      ~BTRFS_SUPER_FLAG_SEEDING;
> +	btrfs_set_super_flags(disk_super, super_flags);
> +
> +	return 0;
> +}
> +
>
> ...
>
> +
> +int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
> +{
> +	struct btrfs_trans_handle *trans;
> +	struct btrfs_device *device;
> +	struct block_device *bdev;
> +	struct list_head *cur;
> +	struct list_head *devices;
> +	struct super_block *sb = root->fs_info->sb;
> +	u64 total_bytes;
> +	int seeding_dev = 0;
> +	int ret = 0;
> +
> +	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
> +		return -EINVAL;
> +
> +	bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
> +	if (!bdev)
> +		return -EIO;
> +
> +	if (root->fs_info->fs_devices->seeding) {
> +		seeding_dev = 1;
> +		down_write(&sb->s_umount);
> +		mutex_lock(&uuid_mutex);
> +	}
> +
> +	filemap_write_and_wait(bdev->bd_inode->i_mapping);

I'm surprised that this can be safely done under ->s_umount.

> +	mutex_lock(&root->fs_info->volume_mutex);
> +
> +	devices = &root->fs_info->fs_devices->devices;
> +	list_for_each(cur, devices) {
> +		device = list_entry(cur, struct btrfs_device, dev_list);
> +		if (device->bdev == bdev) {
> +			ret = -EEXIST;
> +			goto error;
> +		}
> +	}
> +
> +	device = kzalloc(sizeof(*device), GFP_NOFS);
> +	if (!device) {
> +		/* we can safely leave the fs_devices entry around */
> +		ret = -ENOMEM;
> +		goto error;
> +	}
> +
> +	device->name = kstrdup(device_path, GFP_NOFS);
> +	if (!device->name) {
> +		kfree(device);
> +		ret = -ENOMEM;
> +		goto error;
> +	}
> +
> +	ret = find_next_devid(root, &device->devid);
> +	if (ret) {
> +		kfree(device);

Leaks device->name

> +		goto error;
> +	}
> +
> +	trans = btrfs_start_transaction(root, 1);
> +	lock_chunks(root);
> +
> +	device->barriers = 1;
> +	device->writeable = 1;
> +	device->work.func = pending_bios_fn;

Should we be using INIT_WORK() here?  It has lockdep implications.
(Multiple instances).

> +	generate_random_uuid(device->uuid);
> +	spin_lock_init(&device->io_lock);
> +	device->generation = trans->transid;
> +	device->io_width = root->sectorsize;
> +	device->io_align = root->sectorsize;
> +	device->sector_size = root->sectorsize;
> +	device->total_bytes = i_size_read(bdev->bd_inode);
> +	device->dev_root = root->fs_info->dev_root;
> +	device->bdev = bdev;
> +	device->in_fs_metadata = 1;
> +	device->mode = 0;
> +	set_blocksize(device->bdev, 4096);
> +
> +	if (seeding_dev) {
> +		sb->s_flags &= ~MS_RDONLY;
> +		ret = btrfs_prepare_sprout(trans, root);
> +		BUG_ON(ret);
> +	}
> +
> +	device->fs_devices = root->fs_info->fs_devices;
> +	list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
> +	list_add(&device->dev_alloc_list,
> +		 &root->fs_info->fs_devices->alloc_list);
> +	root->fs_info->fs_devices->num_devices++;
> +	root->fs_info->fs_devices->open_devices++;
> +	root->fs_info->fs_devices->rw_devices++;
> +	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
> +
> +	total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
> +	btrfs_set_super_total_bytes(&root->fs_info->super_copy,
> +				    total_bytes + device->total_bytes);
> +
> +	total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
> +	btrfs_set_super_num_devices(&root->fs_info->super_copy,
> +				    total_bytes + 1);
> +
> +	if (seeding_dev) {
> +		ret = init_first_rw_device(trans, root, device);
> +		BUG_ON(ret);
> +		ret = btrfs_finish_sprout(trans, root);
> +		BUG_ON(ret);
> +	} else {
> +		ret = btrfs_add_device(trans, root, device);
> +	}
> +
> +	unlock_chunks(root);
> +	btrfs_commit_transaction(trans, root);
> +
> +	if (seeding_dev) {
> +		mutex_unlock(&uuid_mutex);
> +		up_write(&sb->s_umount);
> +
> +		ret = btrfs_relocate_sys_chunks(root);
> +		BUG_ON(ret);
> +	}
> +out:
> +	mutex_unlock(&root->fs_info->volume_mutex);
> +	return ret;
> +error:
> +	close_bdev_exclusive(bdev, 0);
> +	if (seeding_dev) {
> +		mutex_unlock(&uuid_mutex);
> +		up_write(&sb->s_umount);
> +	}
> +	goto out;
> +}
> +

<attention span exceeded, sorry>

Re: [PATCH 02/35] Btrfs multi-device code

[Chris Mason]

On Mon, 2009-01-12 at 22:23 -0800, Andrew Morton wrote:
> On Wed,  7 Jan 2009 22:56:52 -0500 Chris Mason <chris.mason@oracle.com> wrote:
> 
> > All block and extent pointers in btrfs use a logical address space.  This file
> > is responsible for mapping that logical address to a physical block on disk.
> > 
> > It implements raid0, raid1 and raid10.  The block groups used by extent-tree.c
> > are allocated from each device in chunks of ~1GB in size.  This file maintains
> > the chunk tree and device allocation tree that are used to record the
> > areas of each disk that are in use.
> > 
> > Each super block has a bootstrapping section that is used to translate
> > the extents in the chunk tree.  The chunk tree is used to translate all
> > of the other extents in the filesystem to a physical block on disk.
> 
> Where would one go to find out about the user-visible management of
> these devices?

Hi Andrew,

Thanks for all the time you've spent here.

The progs are growing manpages, and the wiki has the user-visible side
documented.

http://btrfs.wiki.kernel.org/index.php/Using_Btrfs_with_Multiple_Devices
http://btrfs.wiki.kernel.org/index.php/Category:Documentation

> > This file also provides async bio submission code.  The bios are often
> > sent down by async helper threads that are supposed to be working on
> > checksumming or other cpu intensive tasks.  We don't want those tasks
> > waiting for a free request, so a helper thread is used here that is willing
> > to actually wait on the drive.
> 
> Seems odd.  Why have a thread to wait on IO?  Why not just
> fire-and-forget, and then only block if some caller wishes to use the
> results of that IO?
> 

If there was a submit_bio that never waited for a request, I'd use it ;)
This extra thread pool just makes sure the checksumming threads don't
get stuck in get_request_wait()

Throttling happens when async things are added to the checksumming queue
instead.

> >
> > ...
> >
> > +struct map_lookup {
> > +	u64 type;
> > +	int io_align;
> > +	int io_width;
> > +	int stripe_len;
> > +	int sector_size;
> > +	int num_stripes;
> > +	int sub_stripes;
> > +	struct btrfs_bio_stripe stripes[];
> > +};
> 
> this structure looks interesting.

There are a few places that need to know all of the locations a logical
block might map to on the disk.  It mainly gets used by the error
handling code so that it can loop through all the copies of a block when
checksums don't match or we get io errors.

But yes, will commit.

> > +static int init_first_rw_device(struct btrfs_trans_handle *trans,
> > +				struct btrfs_root *root,
> > +				struct btrfs_device *device);
> > +static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
> > +
> > +#define map_lookup_size(n) (sizeof(struct map_lookup) + \
> > +			    (sizeof(struct btrfs_bio_stripe) * (n)))
> > +
> > +static DEFINE_MUTEX(uuid_mutex);
> > +static LIST_HEAD(fs_uuids);
> > +
> > +void btrfs_lock_volumes(void)
> > +{
> > +	mutex_lock(&uuid_mutex);
> > +}
> > +
> > +void btrfs_unlock_volumes(void)
> > +{
> > +	mutex_unlock(&uuid_mutex);
> > +}
> > +
> > +static void lock_chunks(struct btrfs_root *root)
> > +{
> > +	mutex_lock(&root->fs_info->chunk_mutex);
> > +}
> > +
> > +static void unlock_chunks(struct btrfs_root *root)
> > +{
> > +	mutex_unlock(&root->fs_info->chunk_mutex);
> > +}
> > +
> > +static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
> > +{
> > +	struct btrfs_device *device;
> > +	WARN_ON(fs_devices->opened);
> > +	while (!list_empty(&fs_devices->devices)) {
> > +		device = list_entry(fs_devices->devices.next,
> > +				    struct btrfs_device, dev_list);
> > +		list_del(&device->dev_list);
> > +		kfree(device->name);
> > +		kfree(device);
> > +	}
> > +	kfree(fs_devices);
> > +}
> 
> Called under btrfs_lock_volumes(), one assumes/hopes.  A
> WARN_ON(!mutex_is_locked()) is a nice way of documenting such things.

Good point, thanks

> 
> > +int btrfs_cleanup_fs_uuids(void)
> > +{
> > +	struct btrfs_fs_devices *fs_devices;
> > +
> > +	while (!list_empty(&fs_uuids)) {
> > +		fs_devices = list_entry(fs_uuids.next,
> > +					struct btrfs_fs_devices, list);
> > +		list_del(&fs_devices->list);
> > +		free_fs_devices(fs_devices);
> > +	}
> > +	return 0;
> > +}
> 
> Ditto.
> 
> > +static noinline struct btrfs_device *__find_device(struct list_head *head,
> > +						   u64 devid, u8 *uuid)
> 
> It's not possible for the code reader to work out why all these
> noinlines are here (this reader, at least).

This is the "please don't make huge functions because we don't have
infinite stack size" noinline.  Eric Sandeen had a patch that gave that
a real name, I'll dig it up and switch to it.

> 
> > +{
> > +	struct btrfs_device *dev;
> > +	struct list_head *cur;
> > +
> > +	list_for_each(cur, head) {
> > +		dev = list_entry(cur, struct btrfs_device, dev_list);
> > +		if (dev->devid == devid &&
> > +		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
> > +			return dev;
> > +		}
> > +	}
> > +	return NULL;
> > +}
> > +
> > +static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
> > +{
> > +	struct list_head *cur;
> > +	struct btrfs_fs_devices *fs_devices;
> > +
> > +	list_for_each(cur, &fs_uuids) {
> > +		fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
> > +		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
> > +			return fs_devices;
> > +	}
> > +	return NULL;
> > +}
> 
> Called under btrfs_lock_volumes()...

Nod

> 
> > +/*
> > + * we try to collect pending bios for a device so we don't get a large
> > + * number of procs sending bios down to the same device.  This greatly
> > + * improves the schedulers ability to collect and merge the bios.
> 
> eh?  The block layer already implements limiting.
> 
> Why does _not_ sending bios to a device aid the IO scheduler's merging
> activity?
> 
> The comment needs help, I suspect.

run_scheduled_bios happens from the async worker thread.  The goal is to
have the fewest possible number of async threads calling submit_bio
because that generally turns nice sequential IO from the higher layers
into seeky IO.

> 
> > + * But, it also turns into a long list of bios to process and that is sure
> > + * to eventually make the worker thread block.
> 
> Why is that bad?  If there's a large amount of IO queued then blocking
> seems an appropriate thing to do.


Each btrfs device has a list of bios that are pending submission.
run_scheduled_bios is called one device at a time.  If
run_scheduled_bios notices that a given block device is congested, it
may requeue the work for the congested device and return.

This lets the async thread move on to the next device instead of getting
stuck on the congested device.

> 
> >  The solution here is to
> > + * make some progress and then put this work struct back at the end of
> > + * the list if the block device is congested.  This way, multiple devices
> > + * can make progress from a single worker thread.
> 
> hrm, OK.  I think.
> 

I'll add more comments there ;)

> > + */
> > +static noinline int run_scheduled_bios(struct btrfs_device *device)
> > +{
> > +	struct bio *pending;
> > +	struct backing_dev_info *bdi;
> > +	struct btrfs_fs_info *fs_info;
> > +	struct bio *tail;
> > +	struct bio *cur;
> > +	int again = 0;
> > +	unsigned long num_run = 0;
> > +	unsigned long limit;
> > +
> > +	bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
> 
> device->bdev->bd_inode_backing_dev_info is equivalent and shorter, I suspect.

Ok

>
> >
> 
> I forget why we added that.
> 
> > +	fs_info = device->dev_root->fs_info;
> > +	limit = btrfs_async_submit_limit(fs_info);
> > +	limit = limit * 2 / 3;
> 
> ??

There have been about 10 different schemes to throttle async requests.
I'm still working on the perfect one...

> > +
> > +	while (pending) {
> > +		cur = pending;
> > +		pending = pending->bi_next;
> > +		cur->bi_next = NULL;
> > +		atomic_dec(&fs_info->nr_async_bios);
> > +
> > +		if (atomic_read(&fs_info->nr_async_bios) < limit &&
> > +		    waitqueue_active(&fs_info->async_submit_wait))
> > +			wake_up(&fs_info->async_submit_wait);
> > +
> > +		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
> > +		bio_get(cur);
> > +		submit_bio(cur->bi_rw, cur);
> > +		bio_put(cur);
> 
> What do the bio_get()/bio_put() do?  Looks wrong, or unneeded.

Not needed, will drop.  I consistently put them everywhere while hunting
for a memory corruption.

> 
> > +		num_run++;
> > +
> > +		/*
> > +		 * we made progress, there is more work to do and the bdi
> > +		 * is now congested.  Back off and let other work structs
> > +		 * run instead
> > +		 */
> > +		if (pending && bdi_write_congested(bdi) &&
> 
> Does this function handle only writes?  If it handles reads, why do we
> test only write congestion?  (please add comments if the questions
> aren't painfully stupid).

Only writes, but since that wasn't obvious it'll get more comments as
well.

> > +
> > +static noinline int device_list_add(const char *path,
> > +			   struct btrfs_super_block *disk_super,
> > +			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
> > +{
> > +	struct btrfs_device *device;
> > +	struct btrfs_fs_devices *fs_devices;
> > +	u64 found_transid = btrfs_super_generation(disk_super);
> > +
> > +	fs_devices = find_fsid(disk_super->fsid);
> > +	if (!fs_devices) {
> > +		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
> 
> Wondering why it's GFP_NOFS.
> 
> Presumably this function is supposed to be called under
> btrfs_lock_volumes(), and we also call that function on the memory
> reclaim path?

Its mostly because my fingers can't type GFP without NOFS.  That code
doesn't need the NOFS though.

> > +
> > +static struct btrfs_fs_devices *clone_fs_devices(struct
> btrfs_fs_devices *orig)
> 
> I wonder why this function exists.

It came with Yan Zheng's seed FS support, and he'll add comments to all
the related code.  Seed filesystems let you include one filesystem as a
readonly starting point to another FS.  Basically:

* Create the perfect virtual machine image on /dev/some_disk_in_the_san
* Use /dev/sda to hold all the local modifications to the image

Other variations on readonly media like CD images would work too.

This does work today, btrfstune was added to set a filesystem as a seed.

You just mount it and add a writable device to it and you get a new FS
on the writable device that references the seed.

> > +{
> > +	struct btrfs_fs_devices *fs_devices;
> > +	struct btrfs_device *device;
> > +	struct btrfs_device *orig_dev;
> > +
> > +	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
> > +	if (!fs_devices)
> > +		return ERR_PTR(-ENOMEM);
> > +
> > +	INIT_LIST_HEAD(&fs_devices->devices);
> > +	INIT_LIST_HEAD(&fs_devices->alloc_list);
> > +	INIT_LIST_HEAD(&fs_devices->list);
> > +	fs_devices->latest_devid = orig->latest_devid;
> > +	fs_devices->latest_trans = orig->latest_trans;
> > +	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
> > +
> > +	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
> > +		device = kzalloc(sizeof(*device), GFP_NOFS);
> > +		if (!device)
> > +			goto error;
> > +
> > +		device->name = kstrdup(orig_dev->name, GFP_NOFS);
> > +		if (!device->name)
> > +			goto error;
> 
> We leaked *device?

We surely did, thanks.

> 
> > +		device->devid = orig_dev->devid;
> > +		device->work.func = pending_bios_fn;
> > +		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
> > +		device->barriers = 1;
> > +		spin_lock_init(&device->io_lock);
> > +		INIT_LIST_HEAD(&device->dev_list);
> > +		INIT_LIST_HEAD(&device->dev_alloc_list);
> > +
> > +		list_add(&device->dev_list, &fs_devices->devices);
> > +		device->fs_devices = fs_devices;
> > +		fs_devices->num_devices++;
> > +	}
> > +	return fs_devices;
> > +error:
> > +	free_fs_devices(fs_devices);
> > +	return ERR_PTR(-ENOMEM);
> > +}
> > +
> > +int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
> 
> whatdoesthisdo

Userland asks the btrfs module to probe a device, and it searches for FS
uuids and magic.  Each device has a device UUID (my name is xxxx) and an
FS uuid (I live in FS yyyy).

The kernel maintains a mapping of FS uuid to a list of devices where
that uuid was found.

At mount time, it reads the superblocks of each device in the list and
finds the most recent copy of the btree that lists all the device UUIDs
in the FS.

So, we have two lists of devices.  One in the kernel came from a
userland initiated probe.  One list in the kernel came from FS metadata.
One of three states exist:

1) The two lists match up and the mount can proceed.
2) The kernel didn't open all the devices that belong in the FS.
Mounting can't proceed.
3) The kernel opened a device that wasn't actually in the FS.  It is
extra and should be closed.

This code closes the devices that were extra in step 3.  It might happen
if the device was removed from the FS or failed etc etc.

(will add comment)

> 
> > +{
> > +	struct list_head *tmp;
> > +	struct list_head *cur;
> > +	struct btrfs_device *device;
> > +
> > +	mutex_lock(&uuid_mutex);
> 
> btrfs_lock_volumes()?

We should pick one or the other.  btrfs_lock_volumes used to do much
more, I'd rather drop it and change calls to explicitly use the
uuid_mutex.

> 
> > +again:
> > +	list_for_each_safe(cur, tmp, &fs_devices->devices) {
> > +		device = list_entry(cur, struct btrfs_device, dev_list);
> > +		if (device->in_fs_metadata)
> > +			continue;
> > +
> > +		if (device->bdev) {
> > +			close_bdev_exclusive(device->bdev, device->mode);
> > +			device->bdev = NULL;
> > +			fs_devices->open_devices--;
> > +		}
> > +		if (device->writeable) {
> > +			list_del_init(&device->dev_alloc_list);
> > +			device->writeable = 0;
> > +			fs_devices->rw_devices--;
> > +		}
> 
> Why are we bothering to modify *device?  It's about to be freed.

Good point.

> 
> > +		list_del_init(&device->dev_list);
> > +		fs_devices->num_devices--;
> > +		kfree(device->name);
> > +		kfree(device);
> > +	}
> > +
> > +	if (fs_devices->seed) {
> > +		fs_devices = fs_devices->seed;
> > +		goto again;
> > +	}
> > +
> > +	mutex_unlock(&uuid_mutex);
> 
> ...
> 
> > +	return 0;
> > +}
> > +
> > +static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
> > +{
> > +	struct list_head *cur;
> > +	struct btrfs_device *device;
> > +
> > +	if (--fs_devices->opened > 0)
> > +		return 0;
> > +
> > +	list_for_each(cur, &fs_devices->devices) {
> > +		device = list_entry(cur, struct btrfs_device, dev_list);
> > +		if (device->bdev) {
> 
> This reader is wondering what btrfs_device.bdev==NULL means.  How would
> one go about determining this?

The device list includes devices opened (bdev != NULL) and devices the
metadata says should be there but that we haven't found (bdev == NULL).

(Will add comment)

> 
> > +			close_bdev_exclusive(device->bdev, device->mode);
> > +			fs_devices->open_devices--;
> > +		}
> > +		if (device->writeable) {
> > +			list_del_init(&device->dev_alloc_list);
> > +			fs_devices->rw_devices--;
> > +		}
> > +
> > +		device->bdev = NULL;
> > +		device->writeable = 0;
> > +		device->in_fs_metadata = 0;
> 
> What does in_fs_metadata mean?

in_fs_metadata means the filesystem metadata on disk actually references
the device UUID.

> 
> Perhaps those three fields should have type `bool'.

Yes, its a bool.

> 
> > +	}
> > +	WARN_ON(fs_devices->open_devices);
> > +	WARN_ON(fs_devices->rw_devices);
> > +	fs_devices->opened = 0;
> > +	fs_devices->seeding = 0;
> 
> What does the `seeding' field mean?  It's a boolean, I assume..

Yes, seeding is a bool (see seed filesystem support above)

> 
> > +	return 0;
> > +}
> > +
> > +int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
> > +{
> > +	struct btrfs_fs_devices *seed_devices = NULL;
> > +	int ret;
> > +
> > +	mutex_lock(&uuid_mutex);
> 
> btrfs_lock_volumes()?  (Please check whole fs for consistency here)
> 
> Seems strange that a lock called "uuid_mutex" is protecting a pile of
> things which don't seem very uuid-related.
> 

The lists and mappings it protects are all keyed by FS or device
uuids...made sense at the time I suppose.  Will pick a better name.

> > +
> > +static int __btrfs_open_devices(struct btrfs_fs_devices
> *fs_devices,
> > +				fmode_t flags, void *holder)
> > +{
> > +	struct block_device *bdev;
> > +	struct list_head *head = &fs_devices->devices;
> > +	struct list_head *cur;
> > +	struct btrfs_device *device;
> > +	struct block_device *latest_bdev = NULL;
> > +	struct buffer_head *bh;
> > +	struct btrfs_super_block *disk_super;
> > +	u64 latest_devid = 0;
> > +	u64 latest_transid = 0;
> > +	u64 devid;
> > +	int seeding = 1;
> > +	int ret = 0;
> > +
> > +	list_for_each(cur, head) {
> > +		device = list_entry(cur, struct btrfs_device, dev_list);
> 
> list_for_each_entry()

ack
> 
> > +		if (device->bdev)
> > +			continue;
> > +		if (!device->name)
> > +			continue;
> 
> What's the algorithm for understanding the above four lines?

if (device->bdev) we're already open.
if (!device->name) the device uuid hasn't been discovered yet. The entry
in the device list records the device should be there but isn't.

> 
> > +		bdev = open_bdev_exclusive(device->name, flags, holder);
> > +		if (IS_ERR(bdev)) {
> > +			printk(KERN_INFO "open %s failed\n", device->name);
> > +			goto error;
> > +		}
> > +		set_blocksize(bdev, 4096);
> 
> hm.  What's the situation with btrfs blocksizes?

Basically we only need the blocksize to be set to a consistent value.
buffer_heads are only used for the super blocks.

This will be true even after we fixup the extent_buffer and extent_io
code to work with blocksize != pagesize, since the size of the super is
4096.

> > +
> > +error_brelse:
> > +		brelse(bh);
> 
> brelse() is only needed in situations where the pointer might be NULL.

put_bh doesn't give us the friendly warnings if we've messed up the
usage count, and I try to avoid the __ unless I really need the go
faster stripes.

> > +}
> > +	bh = btrfs_read_dev_super(bdev);
> > +	if (!bh) {
> > +		ret = -EIO;
> > +		goto error_close;
> > +	}
> > +	disk_super = (struct btrfs_super_block *)bh->b_data;
> 
> <looks at struct btrfs_super_block>
> 
> It needs work, I think.  This:
> 
> 	__le64 compat_ro_flags;
> 	__le64 incompat_flags;
> 	__le16 csum_type;
> 	u8 root_level;
> 	u8 chunk_root_level;
> 	u8 log_root_level;
> 	struct btrfs_dev_item dev_item;
> 
> doesn't look like it will successfully port between architectures. 
> Also `struct btrfs_dev_item' has three le32's wedged between le64's.
> 
> I suppose that the __packed will help in some cases (but not the
> btrfs_super_block, surely).  However it seems unwise/incautious/ugly to
> rely upon __packed in this manner.
> 
> 
> Please do s/__attribute__ ((__packed__))/__packed/g

Ok thanks.

> 
> > +	devid = le64_to_cpu(disk_super->dev_item.devid);
> > +	transid = btrfs_super_generation(disk_super);
> > +	if (disk_super->label[0])
> > +		printk(KERN_INFO "device label %s ", disk_super->label);
> > +	else {
> > +		/* FIXME, make a readl uuid parser */
> 
> what's that?

ack, should read a real uuid parser, in this case I just want to print
the first few bytes of the uuid.

> 
> > +		printk(KERN_INFO "device fsid %llx-%llx ",
> > +		       *(unsigned long long *)disk_super->fsid,
> > +		       *(unsigned long long *)(disk_super->fsid + 8));
> > +	}
> > +	printk(KERN_INFO "devid %llu transid %llu %s\n",
> > +	       (unsigned long long)devid, (unsigned long long)transid, path);
> > +	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
> > +
> > +	brelse(bh);

> > +	/* FIXME use last free of some kind */
> > +
> > +	/* we don't want to overwrite the superblock on the drive,
> > +	 * so we make sure to start at an offset of at least 1MB
> > +	 */
> > +	search_start = max((u64)1024 * 1024, search_start);
> 
> max_t

ack

> 
> > +	if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
> > +		search_start = max(root->fs_info->alloc_start, search_start);
> > +
> > +	key.objectid = device->devid;
> > +	key.offset = search_start;
> > +	key.type = BTRFS_DEV_EXTENT_KEY;
> > +	ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
> > +	if (ret < 0)
> > +		goto error;
> > +	ret = btrfs_previous_item(root, path, 0, key.type);
> > +	if (ret < 0)
> > +		goto error;
> > +	l = path->nodes[0];
> > +	btrfs_item_key_to_cpu(l, &key, path->slots[0]);
> > +	while (1) {
> > +		l = path->nodes[0];
> > +		slot = path->slots[0];
> > +		if (slot >= btrfs_header_nritems(l)) {
> > +			ret = btrfs_next_leaf(root, path);
> > +			if (ret == 0)
> > +				continue;
> > +			if (ret < 0)
> > +				goto error;
> > +no_more_items:
> > +			if (!start_found) {
> > +				if (search_start >= search_end) {
> > +					ret = -ENOSPC;
> 
> Hey.  Shouldn't that go BUG()?

Grin, this is the enospc we do handle.

> 
> > +					goto error;
> > +				}
> > +				*start = search_start;
> > +				start_found = 1;
> > +				goto check_pending;
> > +			}
> > +			*start = last_byte > search_start ?
> > +				last_byte : search_start;
> 
> max()?

ack

> > +check_pending:
> > +	/* we have to make sure we didn't find an extent that has already
> > +	 * been allocated by the map tree or the original allocation
> > +	 */
> > +	BUG_ON(*start < search_start);
> 
> Is it triggerable by on-disk corruption?

Yes, good point.  In general the corrupted FS checks are not yet done.

> <looks at btrfs_search_slot>
> 
>  * if ins_len > 0, nodes and leaves will be split as we walk down the
>  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
>  * possible)
> 
> The (ab)use of the sign of formal arg `ins_len' seems a bit nasty?

I'll make a real constant for the ins_len < 0, since the code uses -1
everywhere for that case.

> >
> > +static noinline int find_next_chunk(struct btrfs_root *root,
> > +				    u64 objectid, u64 *offset)
> > +{
> > +	struct btrfs_path *path;
> > +	int ret;
> > +	struct btrfs_key key;
> > +	struct btrfs_chunk *chunk;
> > +	struct btrfs_key found_key;
> > +
> > +	path = btrfs_alloc_path();
> > +	BUG_ON(!path);
> 
> Stuff like this will eventually have to be weeded out and fixed.  It at
> least needs a big FIXME the instant it is typed in.  Quietly slipping
> it in there in this manner is just creating more work for yourselves.

Andi has a big patch of fixes for many of these.

> > +	ptr = (unsigned long)btrfs_device_uuid(dev_item);
> > +	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
> > +	ptr = (unsigned long)btrfs_device_fsid(dev_item);
> > +	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
> > +	btrfs_mark_buffer_dirty(leaf);
> 
> I expected a function called btrfs_mark_buffer_dirty() to be a wrapper
> around mark_buffer_dirty.

It gets back to the name of extent_buffers I suppose...I kept the name
similar to mark_buffer_dirty because you expect to find something that
marks the extent_buffer for writeback.

But btrfs_set_extent_buffer_dirty() is fine too



> 
> > +	ret = 0;
> > +out:
> > +	btrfs_free_path(path);
> > +	return ret;
> > +}
> > +
> >
> > ...
> >
> > +
> > +int btrfs_rm_device(struct btrfs_root *root, char *device_path)
> > +{
> > +	struct btrfs_device *device;
> > +	struct btrfs_device *next_device;
> > +	struct block_device *bdev;
> > +	struct buffer_head *bh = NULL;
> > +	struct btrfs_super_block *disk_super;
> > +	u64 all_avail;
> > +	u64 devid;
> > +	u64 num_devices;
> > +	u8 *dev_uuid;
> > +	int ret = 0;
> > +
> > +	mutex_lock(&uuid_mutex);
> 
> btrfs_lock_volumes()
> 
> > +	mutex_lock(&root->fs_info->volume_mutex);
> 
> lock_chunks() (which needs renaming)

Ack to get rid of lock_chunks()

> 
> > +	all_avail = root->fs_info->avail_data_alloc_bits |
> > +		root->fs_info->avail_system_alloc_bits |
> > +		root->fs_info->avail_metadata_alloc_bits;
> > +
> > +	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
> > +	    root->fs_info->fs_devices->rw_devices <= 4) {
> > +		printk(KERN_ERR "btrfs: unable to go below four devices "
> > +		       "on raid10\n");
> > +		ret = -EINVAL;
> > +		goto out;
> > +	}
> > +
> > +	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
> > +	    root->fs_info->fs_devices->rw_devices <= 2) {
> > +		printk(KERN_ERR "btrfs: unable to go below two "
> > +		       "devices on raid1\n");
> > +		ret = -EINVAL;
> > +		goto out;
> > +	}
> > +
> > +	if (strcmp(device_path, "missing") == 0) {
> 
> <wonders what interface all this is implementing>

Will comment to reflect this is how we remove a device that is found in
the FS metadata but not present in the system.  Picture a raid1 where
one disk has been ripped out or died.

> > +static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
> > +				struct btrfs_root *root)
> > +{
> > +	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
> > +	struct btrfs_fs_devices *old_devices;
> > +	struct btrfs_fs_devices *seed_devices;
> > +	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
> > +	struct btrfs_device *device;
> > +	u64 super_flags;
> > +
> > +	BUG_ON(!mutex_is_locked(&uuid_mutex));
> 
> Given the (regrettable?) existence of btrfs_lock_volumes() and
> btrfs_unlock_volumes(), it would make sense to add another regrettable
> wrapper for the above operation.

It makes sense to drop the wrappers.

...

> > +	list_for_each_entry(device, &seed_devices->devices, dev_list) {
> > +		device->fs_devices = seed_devices;
> > +	}
> 
> unneeded braces
> 
> <runs checkpatch>
> 
> <checkpatch misses this - regression?>
> 
> 
> #2001: FILE: fs/btrfs/volumes.c:1930:
> +               return calc_size * num_stripes;
>                                  ^
> 
> ERROR: space prohibited after that '*' (ctx:WxW)

checkpatch wants me to do calc_size *num_stripes, I think because it
thinks num_stripes is a pointer?

At any rate, I missed the braces because I was busy being grumpy about
the bogus * warning...will fix the braces ;)

> #2072: FILE: fs/btrfs/volumes.c:2001:
> +               max_chunk_size = calc_size * 2;
>                                            ^
> 
> ERROR: space prohibited after that '*' (ctx:WxW)
> #2089: FILE: fs/btrfs/volumes.c:2018:
> +       if (calc_size * num_stripes > max_chunk_size) {
> 	                      ^
> 	
> ERROR: space prohibited after that '*' (ctx:WxW)
> #2105: FILE: fs/btrfs/volumes.c:2034:
> +               min_free = calc_size * 2;
> 	                                     ^
> 
> hm, they're all wrong.	
> 

Maybe its the u64 that confuses checkpatch?

> >
> > +
> > +int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
> > +{
...
> > +	bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
> > +	if (!bdev)
> > +		return -EIO;
> > +
> > +	if (root->fs_info->fs_devices->seeding) {
> > +		seeding_dev = 1;
> > +		down_write(&sb->s_umount);
> > +		mutex_lock(&uuid_mutex);
> > +	}
> > +
> > +	filemap_write_and_wait(bdev->bd_inode->i_mapping);
> 
> I'm surprised that this can be safely done under ->s_umount.

btrfs_init_new_device is only called by the add_dev ioctl.  That doesn't
happen with s_umount held.  Or do you mean we should be taking s_umount?

> > +
> > +	device->name = kstrdup(device_path, GFP_NOFS);
> > +	if (!device->name) {
> > +		kfree(device);
> > +		ret = -ENOMEM;
> > +		goto error;
> > +	}
> > +
> > +	ret = find_next_devid(root, &device->devid);
> > +	if (ret) {
> > +		kfree(device);
> 
> Leaks device->name

ack.

> 
> > +		goto error;
> > +	}
> > +
> > +	trans = btrfs_start_transaction(root, 1);
> > +	lock_chunks(root);
> > +
> > +	device->barriers = 1;
> > +	device->writeable = 1;
> > +	device->work.func = pending_bios_fn;
> 
> Should we be using INIT_WORK() here?  It has lockdep implications.
> (Multiple instances).

These are struct btrfs_work

> <attention span exceeded, sorry>

Thanks for all the comments!

Re: [PATCH 02/35] Btrfs multi-device code

[Andrew Morton]

On Tue, 13 Jan 2009 11:36:16 -0500 Chris Mason <chris.mason@oracle.com> wrote:

> > > +	bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
> > > +	if (!bdev)
> > > +		return -EIO;
> > > +
> > > +	if (root->fs_info->fs_devices->seeding) {
> > > +		seeding_dev = 1;
> > > +		down_write(&sb->s_umount);
> > > +		mutex_lock(&uuid_mutex);
> > > +	}
> > > +
> > > +	filemap_write_and_wait(bdev->bd_inode->i_mapping);
> > 
> > I'm surprised that this can be safely done under ->s_umount.
> 
> btrfs_init_new_device is only called by the add_dev ioctl.  That doesn't
> happen with s_umount held.  Or do you mean we should be taking s_umount?

I'm surprised that filemap_write_and_wait() can safely be called under
sb->s_umount.  But I guess filemap_write_and_wait() doesn't take many
locks, even though it waits a lot.

Re: [PATCH 02/35] Btrfs multi-device code

[Chris Mason]

On Tue, 2009-01-13 at 09:52 -0800, Andrew Morton wrote:
> On Tue, 13 Jan 2009 11:36:16 -0500 Chris Mason <chris.mason@oracle.com> wrote:
> 
> > > > +	bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
> > > > +	if (!bdev)
> > > > +		return -EIO;
> > > > +
> > > > +	if (root->fs_info->fs_devices->seeding) {
> > > > +		seeding_dev = 1;
> > > > +		down_write(&sb->s_umount);
> > > > +		mutex_lock(&uuid_mutex);
> > > > +	}
> > > > +
> > > > +	filemap_write_and_wait(bdev->bd_inode->i_mapping);
> > > 
> > > I'm surprised that this can be safely done under ->s_umount.
> > 
> > btrfs_init_new_device is only called by the add_dev ioctl.  That doesn't
> > happen with s_umount held.  Or do you mean we should be taking s_umount?
> 
> I'm surprised that filemap_write_and_wait() can safely be called under
> sb->s_umount.  But I guess filemap_write_and_wait() doesn't take many
> locks, even though it waits a lot.

In this case s_umount shouldn't be held.  But, aren't we running
something similar at umount time (entirely unrelated to this code path)?

At any rate, this flush is there to make sure that any mods made by
btrfs-vol -a /dev/xxx are actually on the device before the kernel tries
to add it.

-chris

Re: [PATCH 02/35] Btrfs multi-device code

[Andy Whitcroft]

On Mon, Jan 12, 2009 at 10:23:45PM -0800, Andrew Morton wrote:
> On Wed,  7 Jan 2009 22:56:52 -0500 Chris Mason <chris.mason@oracle.com> wrote:
> > +	list_for_each_entry(device, &seed_devices->devices, dev_list) {
> > +		device->fs_devices = seed_devices;
> > +	}
> 
> unneeded braces
> 
> <runs checkpatch>
> 
> <checkpatch misses this - regression?>

It never did check them for the list_for_foo things.  Will add that to
my todo list.

> #2001: FILE: fs/btrfs/volumes.c:1930:
> +               return calc_size * num_stripes;
>                                  ^
> 
> ERROR: space prohibited after that '*' (ctx:WxW)
> #2072: FILE: fs/btrfs/volumes.c:2001:
> +               max_chunk_size = calc_size * 2;
>                                            ^
> 
> ERROR: space prohibited after that '*' (ctx:WxW)
> #2089: FILE: fs/btrfs/volumes.c:2018:
> +       if (calc_size * num_stripes > max_chunk_size) {
> 	                      ^
> 	
> ERROR: space prohibited after that '*' (ctx:WxW)
> #2105: FILE: fs/btrfs/volumes.c:2034:
> +               min_free = calc_size * 2;
> 	                                     ^
> 
> hm, they're all wrong.	

Ok all the * spacing errors are the same basic fault.  That there is an
else on the previous line and that is being incomporated incorrectly
into the statement for possible type analysis.  This part will be fixed
in my next block of updates.

[PATCH 03/35] Btrfs tree logging (fsync optimizations)

[Chris Mason]

A full transaction commit in Btrfs is fairly expensive.  It doesn't
require very much IO, but it forces recow of frequently cow'd blocks,
which leads to many dirty blocks in the extent allocation tree.

To avoid a full transaction commit, btrfs will copy individual file or
directory items into a dedicated logging tree (per subvolume).  After
a crash, these items are copied out of the logging tree and back into
the subvolume.  Any extents that were referenced by the logging tree items
are re-allocated after the crash.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
new file mode 100644
index 0000000..332ec35
--- /dev/null
+++ b/fs/btrfs/tree-log.c
@@ -0,0 +1,2885 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+#include "transaction.h"
+#include "disk-io.h"
+#include "locking.h"
+#include "print-tree.h"
+#include "compat.h"
+#include "tree-log.h"
+
+/* magic values for the inode_only field in btrfs_log_inode:
+ *
+ * LOG_INODE_ALL means to log everything
+ * LOG_INODE_EXISTS means to log just enough to recreate the inode
+ * during log replay
+ */
+#define LOG_INODE_ALL 0
+#define LOG_INODE_EXISTS 1
+
+/*
+ * stages for the tree walking.  The first
+ * stage (0) is to only pin down the blocks we find
+ * the second stage (1) is to make sure that all the inodes
+ * we find in the log are created in the subvolume.
+ *
+ * The last stage is to deal with directories and links and extents
+ * and all the other fun semantics
+ */
+#define LOG_WALK_PIN_ONLY 0
+#define LOG_WALK_REPLAY_INODES 1
+#define LOG_WALK_REPLAY_ALL 2
+
+static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, struct inode *inode,
+			     int inode_only);
+static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     struct btrfs_path *path, u64 objectid);
+
+/*
+ * tree logging is a special write ahead log used to make sure that
+ * fsyncs and O_SYNCs can happen without doing full tree commits.
+ *
+ * Full tree commits are expensive because they require commonly
+ * modified blocks to be recowed, creating many dirty pages in the
+ * extent tree an 4x-6x higher write load than ext3.
+ *
+ * Instead of doing a tree commit on every fsync, we use the
+ * key ranges and transaction ids to find items for a given file or directory
+ * that have changed in this transaction.  Those items are copied into
+ * a special tree (one per subvolume root), that tree is written to disk
+ * and then the fsync is considered complete.
+ *
+ * After a crash, items are copied out of the log-tree back into the
+ * subvolume tree.  Any file data extents found are recorded in the extent
+ * allocation tree, and the log-tree freed.
+ *
+ * The log tree is read three times, once to pin down all the extents it is
+ * using in ram and once, once to create all the inodes logged in the tree
+ * and once to do all the other items.
+ */
+
+/*
+ * btrfs_add_log_tree adds a new per-subvolume log tree into the
+ * tree of log tree roots.  This must be called with a tree log transaction
+ * running (see start_log_trans).
+ */
+static int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root)
+{
+	struct btrfs_key key;
+	struct btrfs_root_item root_item;
+	struct btrfs_inode_item *inode_item;
+	struct extent_buffer *leaf;
+	struct btrfs_root *new_root = root;
+	int ret;
+	u64 objectid = root->root_key.objectid;
+
+	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
+				      BTRFS_TREE_LOG_OBJECTID,
+				      trans->transid, 0, 0, 0);
+	if (IS_ERR(leaf)) {
+		ret = PTR_ERR(leaf);
+		return ret;
+	}
+
+	btrfs_set_header_nritems(leaf, 0);
+	btrfs_set_header_level(leaf, 0);
+	btrfs_set_header_bytenr(leaf, leaf->start);
+	btrfs_set_header_generation(leaf, trans->transid);
+	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
+
+	write_extent_buffer(leaf, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(leaf),
+			    BTRFS_FSID_SIZE);
+	btrfs_mark_buffer_dirty(leaf);
+
+	inode_item = &root_item.inode;
+	memset(inode_item, 0, sizeof(*inode_item));
+	inode_item->generation = cpu_to_le64(1);
+	inode_item->size = cpu_to_le64(3);
+	inode_item->nlink = cpu_to_le32(1);
+	inode_item->nbytes = cpu_to_le64(root->leafsize);
+	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
+
+	btrfs_set_root_bytenr(&root_item, leaf->start);
+	btrfs_set_root_generation(&root_item, trans->transid);
+	btrfs_set_root_level(&root_item, 0);
+	btrfs_set_root_refs(&root_item, 0);
+	btrfs_set_root_used(&root_item, 0);
+
+	memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
+	root_item.drop_level = 0;
+
+	btrfs_tree_unlock(leaf);
+	free_extent_buffer(leaf);
+	leaf = NULL;
+
+	btrfs_set_root_dirid(&root_item, 0);
+
+	key.objectid = BTRFS_TREE_LOG_OBJECTID;
+	key.offset = objectid;
+	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+	ret = btrfs_insert_root(trans, root->fs_info->log_root_tree, &key,
+				&root_item);
+	if (ret)
+		goto fail;
+
+	new_root = btrfs_read_fs_root_no_radix(root->fs_info->log_root_tree,
+					       &key);
+	BUG_ON(!new_root);
+
+	WARN_ON(root->log_root);
+	root->log_root = new_root;
+
+	/*
+	 * log trees do not get reference counted because they go away
+	 * before a real commit is actually done.  They do store pointers
+	 * to file data extents, and those reference counts still get
+	 * updated (along with back refs to the log tree).
+	 */
+	new_root->ref_cows = 0;
+	new_root->last_trans = trans->transid;
+fail:
+	return ret;
+}
+
+/*
+ * start a sub transaction and setup the log tree
+ * this increments the log tree writer count to make the people
+ * syncing the tree wait for us to finish
+ */
+static int start_log_trans(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root)
+{
+	int ret;
+	mutex_lock(&root->fs_info->tree_log_mutex);
+	if (!root->fs_info->log_root_tree) {
+		ret = btrfs_init_log_root_tree(trans, root->fs_info);
+		BUG_ON(ret);
+	}
+	if (!root->log_root) {
+		ret = btrfs_add_log_tree(trans, root);
+		BUG_ON(ret);
+	}
+	atomic_inc(&root->fs_info->tree_log_writers);
+	root->fs_info->tree_log_batch++;
+	mutex_unlock(&root->fs_info->tree_log_mutex);
+	return 0;
+}
+
+/*
+ * returns 0 if there was a log transaction running and we were able
+ * to join, or returns -ENOENT if there were not transactions
+ * in progress
+ */
+static int join_running_log_trans(struct btrfs_root *root)
+{
+	int ret = -ENOENT;
+
+	smp_mb();
+	if (!root->log_root)
+		return -ENOENT;
+
+	mutex_lock(&root->fs_info->tree_log_mutex);
+	if (root->log_root) {
+		ret = 0;
+		atomic_inc(&root->fs_info->tree_log_writers);
+		root->fs_info->tree_log_batch++;
+	}
+	mutex_unlock(&root->fs_info->tree_log_mutex);
+	return ret;
+}
+
+/*
+ * indicate we're done making changes to the log tree
+ * and wake up anyone waiting to do a sync
+ */
+static int end_log_trans(struct btrfs_root *root)
+{
+	atomic_dec(&root->fs_info->tree_log_writers);
+	smp_mb();
+	if (waitqueue_active(&root->fs_info->tree_log_wait))
+		wake_up(&root->fs_info->tree_log_wait);
+	return 0;
+}
+
+
+/*
+ * the walk control struct is used to pass state down the chain when
+ * processing the log tree.  The stage field tells us which part
+ * of the log tree processing we are currently doing.  The others
+ * are state fields used for that specific part
+ */
+struct walk_control {
+	/* should we free the extent on disk when done?  This is used
+	 * at transaction commit time while freeing a log tree
+	 */
+	int free;
+
+	/* should we write out the extent buffer?  This is used
+	 * while flushing the log tree to disk during a sync
+	 */
+	int write;
+
+	/* should we wait for the extent buffer io to finish?  Also used
+	 * while flushing the log tree to disk for a sync
+	 */
+	int wait;
+
+	/* pin only walk, we record which extents on disk belong to the
+	 * log trees
+	 */
+	int pin;
+
+	/* what stage of the replay code we're currently in */
+	int stage;
+
+	/* the root we are currently replaying */
+	struct btrfs_root *replay_dest;
+
+	/* the trans handle for the current replay */
+	struct btrfs_trans_handle *trans;
+
+	/* the function that gets used to process blocks we find in the
+	 * tree.  Note the extent_buffer might not be up to date when it is
+	 * passed in, and it must be checked or read if you need the data
+	 * inside it
+	 */
+	int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
+			    struct walk_control *wc, u64 gen);
+};
+
+/*
+ * process_func used to pin down extents, write them or wait on them
+ */
+static int process_one_buffer(struct btrfs_root *log,
+			      struct extent_buffer *eb,
+			      struct walk_control *wc, u64 gen)
+{
+	if (wc->pin) {
+		mutex_lock(&log->fs_info->pinned_mutex);
+		btrfs_update_pinned_extents(log->fs_info->extent_root,
+					    eb->start, eb->len, 1);
+		mutex_unlock(&log->fs_info->pinned_mutex);
+	}
+
+	if (btrfs_buffer_uptodate(eb, gen)) {
+		if (wc->write)
+			btrfs_write_tree_block(eb);
+		if (wc->wait)
+			btrfs_wait_tree_block_writeback(eb);
+	}
+	return 0;
+}
+
+/*
+ * Item overwrite used by replay and tree logging.  eb, slot and key all refer
+ * to the src data we are copying out.
+ *
+ * root is the tree we are copying into, and path is a scratch
+ * path for use in this function (it should be released on entry and
+ * will be released on exit).
+ *
+ * If the key is already in the destination tree the existing item is
+ * overwritten.  If the existing item isn't big enough, it is extended.
+ * If it is too large, it is truncated.
+ *
+ * If the key isn't in the destination yet, a new item is inserted.
+ */
+static noinline int overwrite_item(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct btrfs_path *path,
+				   struct extent_buffer *eb, int slot,
+				   struct btrfs_key *key)
+{
+	int ret;
+	u32 item_size;
+	u64 saved_i_size = 0;
+	int save_old_i_size = 0;
+	unsigned long src_ptr;
+	unsigned long dst_ptr;
+	int overwrite_root = 0;
+
+	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
+		overwrite_root = 1;
+
+	item_size = btrfs_item_size_nr(eb, slot);
+	src_ptr = btrfs_item_ptr_offset(eb, slot);
+
+	/* look for the key in the destination tree */
+	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
+	if (ret == 0) {
+		char *src_copy;
+		char *dst_copy;
+		u32 dst_size = btrfs_item_size_nr(path->nodes[0],
+						  path->slots[0]);
+		if (dst_size != item_size)
+			goto insert;
+
+		if (item_size == 0) {
+			btrfs_release_path(root, path);
+			return 0;
+		}
+		dst_copy = kmalloc(item_size, GFP_NOFS);
+		src_copy = kmalloc(item_size, GFP_NOFS);
+
+		read_extent_buffer(eb, src_copy, src_ptr, item_size);
+
+		dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
+		read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
+				   item_size);
+		ret = memcmp(dst_copy, src_copy, item_size);
+
+		kfree(dst_copy);
+		kfree(src_copy);
+		/*
+		 * they have the same contents, just return, this saves
+		 * us from cowing blocks in the destination tree and doing
+		 * extra writes that may not have been done by a previous
+		 * sync
+		 */
+		if (ret == 0) {
+			btrfs_release_path(root, path);
+			return 0;
+		}
+
+	}
+insert:
+	btrfs_release_path(root, path);
+	/* try to insert the key into the destination tree */
+	ret = btrfs_insert_empty_item(trans, root, path,
+				      key, item_size);
+
+	/* make sure any existing item is the correct size */
+	if (ret == -EEXIST) {
+		u32 found_size;
+		found_size = btrfs_item_size_nr(path->nodes[0],
+						path->slots[0]);
+		if (found_size > item_size) {
+			btrfs_truncate_item(trans, root, path, item_size, 1);
+		} else if (found_size < item_size) {
+			ret = btrfs_extend_item(trans, root, path,
+						item_size - found_size);
+			BUG_ON(ret);
+		}
+	} else if (ret) {
+		BUG();
+	}
+	dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
+					path->slots[0]);
+
+	/* don't overwrite an existing inode if the generation number
+	 * was logged as zero.  This is done when the tree logging code
+	 * is just logging an inode to make sure it exists after recovery.
+	 *
+	 * Also, don't overwrite i_size on directories during replay.
+	 * log replay inserts and removes directory items based on the
+	 * state of the tree found in the subvolume, and i_size is modified
+	 * as it goes
+	 */
+	if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
+		struct btrfs_inode_item *src_item;
+		struct btrfs_inode_item *dst_item;
+
+		src_item = (struct btrfs_inode_item *)src_ptr;
+		dst_item = (struct btrfs_inode_item *)dst_ptr;
+
+		if (btrfs_inode_generation(eb, src_item) == 0)
+			goto no_copy;
+
+		if (overwrite_root &&
+		    S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
+		    S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
+			save_old_i_size = 1;
+			saved_i_size = btrfs_inode_size(path->nodes[0],
+							dst_item);
+		}
+	}
+
+	copy_extent_buffer(path->nodes[0], eb, dst_ptr,
+			   src_ptr, item_size);
+
+	if (save_old_i_size) {
+		struct btrfs_inode_item *dst_item;
+		dst_item = (struct btrfs_inode_item *)dst_ptr;
+		btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
+	}
+
+	/* make sure the generation is filled in */
+	if (key->type == BTRFS_INODE_ITEM_KEY) {
+		struct btrfs_inode_item *dst_item;
+		dst_item = (struct btrfs_inode_item *)dst_ptr;
+		if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
+			btrfs_set_inode_generation(path->nodes[0], dst_item,
+						   trans->transid);
+		}
+	}
+no_copy:
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_release_path(root, path);
+	return 0;
+}
+
+/*
+ * simple helper to read an inode off the disk from a given root
+ * This can only be called for subvolume roots and not for the log
+ */
+static noinline struct inode *read_one_inode(struct btrfs_root *root,
+					     u64 objectid)
+{
+	struct inode *inode;
+	inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
+	if (inode->i_state & I_NEW) {
+		BTRFS_I(inode)->root = root;
+		BTRFS_I(inode)->location.objectid = objectid;
+		BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
+		BTRFS_I(inode)->location.offset = 0;
+		btrfs_read_locked_inode(inode);
+		unlock_new_inode(inode);
+
+	}
+	if (is_bad_inode(inode)) {
+		iput(inode);
+		inode = NULL;
+	}
+	return inode;
+}
+
+/* replays a single extent in 'eb' at 'slot' with 'key' into the
+ * subvolume 'root'.  path is released on entry and should be released
+ * on exit.
+ *
+ * extents in the log tree have not been allocated out of the extent
+ * tree yet.  So, this completes the allocation, taking a reference
+ * as required if the extent already exists or creating a new extent
+ * if it isn't in the extent allocation tree yet.
+ *
+ * The extent is inserted into the file, dropping any existing extents
+ * from the file that overlap the new one.
+ */
+static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      struct extent_buffer *eb, int slot,
+				      struct btrfs_key *key)
+{
+	int found_type;
+	u64 mask = root->sectorsize - 1;
+	u64 extent_end;
+	u64 alloc_hint;
+	u64 start = key->offset;
+	u64 saved_nbytes;
+	struct btrfs_file_extent_item *item;
+	struct inode *inode = NULL;
+	unsigned long size;
+	int ret = 0;
+
+	item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
+	found_type = btrfs_file_extent_type(eb, item);
+
+	if (found_type == BTRFS_FILE_EXTENT_REG ||
+	    found_type == BTRFS_FILE_EXTENT_PREALLOC)
+		extent_end = start + btrfs_file_extent_num_bytes(eb, item);
+	else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+		size = btrfs_file_extent_inline_len(eb, item);
+		extent_end = (start + size + mask) & ~mask;
+	} else {
+		ret = 0;
+		goto out;
+	}
+
+	inode = read_one_inode(root, key->objectid);
+	if (!inode) {
+		ret = -EIO;
+		goto out;
+	}
+
+	/*
+	 * first check to see if we already have this extent in the
+	 * file.  This must be done before the btrfs_drop_extents run
+	 * so we don't try to drop this extent.
+	 */
+	ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
+				       start, 0);
+
+	if (ret == 0 &&
+	    (found_type == BTRFS_FILE_EXTENT_REG ||
+	     found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
+		struct btrfs_file_extent_item cmp1;
+		struct btrfs_file_extent_item cmp2;
+		struct btrfs_file_extent_item *existing;
+		struct extent_buffer *leaf;
+
+		leaf = path->nodes[0];
+		existing = btrfs_item_ptr(leaf, path->slots[0],
+					  struct btrfs_file_extent_item);
+
+		read_extent_buffer(eb, &cmp1, (unsigned long)item,
+				   sizeof(cmp1));
+		read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
+				   sizeof(cmp2));
+
+		/*
+		 * we already have a pointer to this exact extent,
+		 * we don't have to do anything
+		 */
+		if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
+			btrfs_release_path(root, path);
+			goto out;
+		}
+	}
+	btrfs_release_path(root, path);
+
+	saved_nbytes = inode_get_bytes(inode);
+	/* drop any overlapping extents */
+	ret = btrfs_drop_extents(trans, root, inode,
+			 start, extent_end, start, &alloc_hint);
+	BUG_ON(ret);
+
+	if (found_type == BTRFS_FILE_EXTENT_REG ||
+	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
+		unsigned long dest_offset;
+		struct btrfs_key ins;
+
+		ret = btrfs_insert_empty_item(trans, root, path, key,
+					      sizeof(*item));
+		BUG_ON(ret);
+		dest_offset = btrfs_item_ptr_offset(path->nodes[0],
+						    path->slots[0]);
+		copy_extent_buffer(path->nodes[0], eb, dest_offset,
+				(unsigned long)item,  sizeof(*item));
+
+		ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
+		ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
+		ins.type = BTRFS_EXTENT_ITEM_KEY;
+
+		if (ins.objectid > 0) {
+			u64 csum_start;
+			u64 csum_end;
+			LIST_HEAD(ordered_sums);
+			/*
+			 * is this extent already allocated in the extent
+			 * allocation tree?  If so, just add a reference
+			 */
+			ret = btrfs_lookup_extent(root, ins.objectid,
+						ins.offset);
+			if (ret == 0) {
+				ret = btrfs_inc_extent_ref(trans, root,
+						ins.objectid, ins.offset,
+						path->nodes[0]->start,
+						root->root_key.objectid,
+						trans->transid, key->objectid);
+			} else {
+				/*
+				 * insert the extent pointer in the extent
+				 * allocation tree
+				 */
+				ret = btrfs_alloc_logged_extent(trans, root,
+						path->nodes[0]->start,
+						root->root_key.objectid,
+						trans->transid, key->objectid,
+						&ins);
+				BUG_ON(ret);
+			}
+			btrfs_release_path(root, path);
+
+			if (btrfs_file_extent_compression(eb, item)) {
+				csum_start = ins.objectid;
+				csum_end = csum_start + ins.offset;
+			} else {
+				csum_start = ins.objectid +
+					btrfs_file_extent_offset(eb, item);
+				csum_end = csum_start +
+					btrfs_file_extent_num_bytes(eb, item);
+			}
+
+			ret = btrfs_lookup_csums_range(root->log_root,
+						csum_start, csum_end - 1,
+						&ordered_sums);
+			BUG_ON(ret);
+			while (!list_empty(&ordered_sums)) {
+				struct btrfs_ordered_sum *sums;
+				sums = list_entry(ordered_sums.next,
+						struct btrfs_ordered_sum,
+						list);
+				ret = btrfs_csum_file_blocks(trans,
+						root->fs_info->csum_root,
+						sums);
+				BUG_ON(ret);
+				list_del(&sums->list);
+				kfree(sums);
+			}
+		} else {
+			btrfs_release_path(root, path);
+		}
+	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+		/* inline extents are easy, we just overwrite them */
+		ret = overwrite_item(trans, root, path, eb, slot, key);
+		BUG_ON(ret);
+	}
+
+	inode_set_bytes(inode, saved_nbytes);
+	btrfs_update_inode(trans, root, inode);
+out:
+	if (inode)
+		iput(inode);
+	return ret;
+}
+
+/*
+ * when cleaning up conflicts between the directory names in the
+ * subvolume, directory names in the log and directory names in the
+ * inode back references, we may have to unlink inodes from directories.
+ *
+ * This is a helper function to do the unlink of a specific directory
+ * item
+ */
+static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      struct inode *dir,
+				      struct btrfs_dir_item *di)
+{
+	struct inode *inode;
+	char *name;
+	int name_len;
+	struct extent_buffer *leaf;
+	struct btrfs_key location;
+	int ret;
+
+	leaf = path->nodes[0];
+
+	btrfs_dir_item_key_to_cpu(leaf, di, &location);
+	name_len = btrfs_dir_name_len(leaf, di);
+	name = kmalloc(name_len, GFP_NOFS);
+	read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
+	btrfs_release_path(root, path);
+
+	inode = read_one_inode(root, location.objectid);
+	BUG_ON(!inode);
+
+	ret = link_to_fixup_dir(trans, root, path, location.objectid);
+	BUG_ON(ret);
+	ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
+	BUG_ON(ret);
+	kfree(name);
+
+	iput(inode);
+	return ret;
+}
+
+/*
+ * helper function to see if a given name and sequence number found
+ * in an inode back reference are already in a directory and correctly
+ * point to this inode
+ */
+static noinline int inode_in_dir(struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 u64 dirid, u64 objectid, u64 index,
+				 const char *name, int name_len)
+{
+	struct btrfs_dir_item *di;
+	struct btrfs_key location;
+	int match = 0;
+
+	di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
+					 index, name, name_len, 0);
+	if (di && !IS_ERR(di)) {
+		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
+		if (location.objectid != objectid)
+			goto out;
+	} else
+		goto out;
+	btrfs_release_path(root, path);
+
+	di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
+	if (di && !IS_ERR(di)) {
+		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
+		if (location.objectid != objectid)
+			goto out;
+	} else
+		goto out;
+	match = 1;
+out:
+	btrfs_release_path(root, path);
+	return match;
+}
+
+/*
+ * helper function to check a log tree for a named back reference in
+ * an inode.  This is used to decide if a back reference that is
+ * found in the subvolume conflicts with what we find in the log.
+ *
+ * inode backreferences may have multiple refs in a single item,
+ * during replay we process one reference at a time, and we don't
+ * want to delete valid links to a file from the subvolume if that
+ * link is also in the log.
+ */
+static noinline int backref_in_log(struct btrfs_root *log,
+				   struct btrfs_key *key,
+				   char *name, int namelen)
+{
+	struct btrfs_path *path;
+	struct btrfs_inode_ref *ref;
+	unsigned long ptr;
+	unsigned long ptr_end;
+	unsigned long name_ptr;
+	int found_name_len;
+	int item_size;
+	int ret;
+	int match = 0;
+
+	path = btrfs_alloc_path();
+	ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
+	if (ret != 0)
+		goto out;
+
+	item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
+	ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
+	ptr_end = ptr + item_size;
+	while (ptr < ptr_end) {
+		ref = (struct btrfs_inode_ref *)ptr;
+		found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
+		if (found_name_len == namelen) {
+			name_ptr = (unsigned long)(ref + 1);
+			ret = memcmp_extent_buffer(path->nodes[0], name,
+						   name_ptr, namelen);
+			if (ret == 0) {
+				match = 1;
+				goto out;
+			}
+		}
+		ptr = (unsigned long)(ref + 1) + found_name_len;
+	}
+out:
+	btrfs_free_path(path);
+	return match;
+}
+
+
+/*
+ * replay one inode back reference item found in the log tree.
+ * eb, slot and key refer to the buffer and key found in the log tree.
+ * root is the destination we are replaying into, and path is for temp
+ * use by this function.  (it should be released on return).
+ */
+static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  struct btrfs_root *log,
+				  struct btrfs_path *path,
+				  struct extent_buffer *eb, int slot,
+				  struct btrfs_key *key)
+{
+	struct inode *dir;
+	int ret;
+	struct btrfs_key location;
+	struct btrfs_inode_ref *ref;
+	struct btrfs_dir_item *di;
+	struct inode *inode;
+	char *name;
+	int namelen;
+	unsigned long ref_ptr;
+	unsigned long ref_end;
+
+	location.objectid = key->objectid;
+	location.type = BTRFS_INODE_ITEM_KEY;
+	location.offset = 0;
+
+	/*
+	 * it is possible that we didn't log all the parent directories
+	 * for a given inode.  If we don't find the dir, just don't
+	 * copy the back ref in.  The link count fixup code will take
+	 * care of the rest
+	 */
+	dir = read_one_inode(root, key->offset);
+	if (!dir)
+		return -ENOENT;
+
+	inode = read_one_inode(root, key->objectid);
+	BUG_ON(!dir);
+
+	ref_ptr = btrfs_item_ptr_offset(eb, slot);
+	ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
+
+again:
+	ref = (struct btrfs_inode_ref *)ref_ptr;
+
+	namelen = btrfs_inode_ref_name_len(eb, ref);
+	name = kmalloc(namelen, GFP_NOFS);
+	BUG_ON(!name);
+
+	read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
+
+	/* if we already have a perfect match, we're done */
+	if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
+			 btrfs_inode_ref_index(eb, ref),
+			 name, namelen)) {
+		goto out;
+	}
+
+	/*
+	 * look for a conflicting back reference in the metadata.
+	 * if we find one we have to unlink that name of the file
+	 * before we add our new link.  Later on, we overwrite any
+	 * existing back reference, and we don't want to create
+	 * dangling pointers in the directory.
+	 */
+conflict_again:
+	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
+	if (ret == 0) {
+		char *victim_name;
+		int victim_name_len;
+		struct btrfs_inode_ref *victim_ref;
+		unsigned long ptr;
+		unsigned long ptr_end;
+		struct extent_buffer *leaf = path->nodes[0];
+
+		/* are we trying to overwrite a back ref for the root directory
+		 * if so, just jump out, we're done
+		 */
+		if (key->objectid == key->offset)
+			goto out_nowrite;
+
+		/* check all the names in this back reference to see
+		 * if they are in the log.  if so, we allow them to stay
+		 * otherwise they must be unlinked as a conflict
+		 */
+		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+		ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
+		while (ptr < ptr_end) {
+			victim_ref = (struct btrfs_inode_ref *)ptr;
+			victim_name_len = btrfs_inode_ref_name_len(leaf,
+								   victim_ref);
+			victim_name = kmalloc(victim_name_len, GFP_NOFS);
+			BUG_ON(!victim_name);
+
+			read_extent_buffer(leaf, victim_name,
+					   (unsigned long)(victim_ref + 1),
+					   victim_name_len);
+
+			if (!backref_in_log(log, key, victim_name,
+					    victim_name_len)) {
+				btrfs_inc_nlink(inode);
+				btrfs_release_path(root, path);
+				ret = btrfs_unlink_inode(trans, root, dir,
+							 inode, victim_name,
+							 victim_name_len);
+				kfree(victim_name);
+				btrfs_release_path(root, path);
+				goto conflict_again;
+			}
+			kfree(victim_name);
+			ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
+		}
+		BUG_ON(ret);
+	}
+	btrfs_release_path(root, path);
+
+	/* look for a conflicting sequence number */
+	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
+					 btrfs_inode_ref_index(eb, ref),
+					 name, namelen, 0);
+	if (di && !IS_ERR(di)) {
+		ret = drop_one_dir_item(trans, root, path, dir, di);
+		BUG_ON(ret);
+	}
+	btrfs_release_path(root, path);
+
+
+	/* look for a conflicting name */
+	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
+				   name, namelen, 0);
+	if (di && !IS_ERR(di)) {
+		ret = drop_one_dir_item(trans, root, path, dir, di);
+		BUG_ON(ret);
+	}
+	btrfs_release_path(root, path);
+
+	/* insert our name */
+	ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
+			     btrfs_inode_ref_index(eb, ref));
+	BUG_ON(ret);
+
+	btrfs_update_inode(trans, root, inode);
+
+out:
+	ref_ptr = (unsigned long)(ref + 1) + namelen;
+	kfree(name);
+	if (ref_ptr < ref_end)
+		goto again;
+
+	/* finally write the back reference in the inode */
+	ret = overwrite_item(trans, root, path, eb, slot, key);
+	BUG_ON(ret);
+
+out_nowrite:
+	btrfs_release_path(root, path);
+	iput(dir);
+	iput(inode);
+	return 0;
+}
+
+/*
+ * There are a few corners where the link count of the file can't
+ * be properly maintained during replay.  So, instead of adding
+ * lots of complexity to the log code, we just scan the backrefs
+ * for any file that has been through replay.
+ *
+ * The scan will update the link count on the inode to reflect the
+ * number of back refs found.  If it goes down to zero, the iput
+ * will free the inode.
+ */
+static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
+					   struct btrfs_root *root,
+					   struct inode *inode)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_key key;
+	u64 nlink = 0;
+	unsigned long ptr;
+	unsigned long ptr_end;
+	int name_len;
+
+	key.objectid = inode->i_ino;
+	key.type = BTRFS_INODE_REF_KEY;
+	key.offset = (u64)-1;
+
+	path = btrfs_alloc_path();
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+		if (ret < 0)
+			break;
+		if (ret > 0) {
+			if (path->slots[0] == 0)
+				break;
+			path->slots[0]--;
+		}
+		btrfs_item_key_to_cpu(path->nodes[0], &key,
+				      path->slots[0]);
+		if (key.objectid != inode->i_ino ||
+		    key.type != BTRFS_INODE_REF_KEY)
+			break;
+		ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
+		ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
+						   path->slots[0]);
+		while (ptr < ptr_end) {
+			struct btrfs_inode_ref *ref;
+
+			ref = (struct btrfs_inode_ref *)ptr;
+			name_len = btrfs_inode_ref_name_len(path->nodes[0],
+							    ref);
+			ptr = (unsigned long)(ref + 1) + name_len;
+			nlink++;
+		}
+
+		if (key.offset == 0)
+			break;
+		key.offset--;
+		btrfs_release_path(root, path);
+	}
+	btrfs_free_path(path);
+	if (nlink != inode->i_nlink) {
+		inode->i_nlink = nlink;
+		btrfs_update_inode(trans, root, inode);
+	}
+	BTRFS_I(inode)->index_cnt = (u64)-1;
+
+	return 0;
+}
+
+static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
+					    struct btrfs_root *root,
+					    struct btrfs_path *path)
+{
+	int ret;
+	struct btrfs_key key;
+	struct inode *inode;
+
+	key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
+	key.type = BTRFS_ORPHAN_ITEM_KEY;
+	key.offset = (u64)-1;
+	while (1) {
+		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+		if (ret < 0)
+			break;
+
+		if (ret == 1) {
+			if (path->slots[0] == 0)
+				break;
+			path->slots[0]--;
+		}
+
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+		if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
+		    key.type != BTRFS_ORPHAN_ITEM_KEY)
+			break;
+
+		ret = btrfs_del_item(trans, root, path);
+		BUG_ON(ret);
+
+		btrfs_release_path(root, path);
+		inode = read_one_inode(root, key.offset);
+		BUG_ON(!inode);
+
+		ret = fixup_inode_link_count(trans, root, inode);
+		BUG_ON(ret);
+
+		iput(inode);
+
+		if (key.offset == 0)
+			break;
+		key.offset--;
+	}
+	btrfs_release_path(root, path);
+	return 0;
+}
+
+
+/*
+ * record a given inode in the fixup dir so we can check its link
+ * count when replay is done.  The link count is incremented here
+ * so the inode won't go away until we check it
+ */
+static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      u64 objectid)
+{
+	struct btrfs_key key;
+	int ret = 0;
+	struct inode *inode;
+
+	inode = read_one_inode(root, objectid);
+	BUG_ON(!inode);
+
+	key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
+	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
+	key.offset = objectid;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+
+	btrfs_release_path(root, path);
+	if (ret == 0) {
+		btrfs_inc_nlink(inode);
+		btrfs_update_inode(trans, root, inode);
+	} else if (ret == -EEXIST) {
+		ret = 0;
+	} else {
+		BUG();
+	}
+	iput(inode);
+
+	return ret;
+}
+
+/*
+ * when replaying the log for a directory, we only insert names
+ * for inodes that actually exist.  This means an fsync on a directory
+ * does not implicitly fsync all the new files in it
+ */
+static noinline int insert_one_name(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root,
+				    struct btrfs_path *path,
+				    u64 dirid, u64 index,
+				    char *name, int name_len, u8 type,
+				    struct btrfs_key *location)
+{
+	struct inode *inode;
+	struct inode *dir;
+	int ret;
+
+	inode = read_one_inode(root, location->objectid);
+	if (!inode)
+		return -ENOENT;
+
+	dir = read_one_inode(root, dirid);
+	if (!dir) {
+		iput(inode);
+		return -EIO;
+	}
+	ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
+
+	/* FIXME, put inode into FIXUP list */
+
+	iput(inode);
+	iput(dir);
+	return ret;
+}
+
+/*
+ * take a single entry in a log directory item and replay it into
+ * the subvolume.
+ *
+ * if a conflicting item exists in the subdirectory already,
+ * the inode it points to is unlinked and put into the link count
+ * fix up tree.
+ *
+ * If a name from the log points to a file or directory that does
+ * not exist in the FS, it is skipped.  fsyncs on directories
+ * do not force down inodes inside that directory, just changes to the
+ * names or unlinks in a directory.
+ */
+static noinline int replay_one_name(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root,
+				    struct btrfs_path *path,
+				    struct extent_buffer *eb,
+				    struct btrfs_dir_item *di,
+				    struct btrfs_key *key)
+{
+	char *name;
+	int name_len;
+	struct btrfs_dir_item *dst_di;
+	struct btrfs_key found_key;
+	struct btrfs_key log_key;
+	struct inode *dir;
+	u8 log_type;
+	int exists;
+	int ret;
+
+	dir = read_one_inode(root, key->objectid);
+	BUG_ON(!dir);
+
+	name_len = btrfs_dir_name_len(eb, di);
+	name = kmalloc(name_len, GFP_NOFS);
+	log_type = btrfs_dir_type(eb, di);
+	read_extent_buffer(eb, name, (unsigned long)(di + 1),
+		   name_len);
+
+	btrfs_dir_item_key_to_cpu(eb, di, &log_key);
+	exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
+	if (exists == 0)
+		exists = 1;
+	else
+		exists = 0;
+	btrfs_release_path(root, path);
+
+	if (key->type == BTRFS_DIR_ITEM_KEY) {
+		dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
+				       name, name_len, 1);
+	} else if (key->type == BTRFS_DIR_INDEX_KEY) {
+		dst_di = btrfs_lookup_dir_index_item(trans, root, path,
+						     key->objectid,
+						     key->offset, name,
+						     name_len, 1);
+	} else {
+		BUG();
+	}
+	if (!dst_di || IS_ERR(dst_di)) {
+		/* we need a sequence number to insert, so we only
+		 * do inserts for the BTRFS_DIR_INDEX_KEY types
+		 */
+		if (key->type != BTRFS_DIR_INDEX_KEY)
+			goto out;
+		goto insert;
+	}
+
+	btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
+	/* the existing item matches the logged item */
+	if (found_key.objectid == log_key.objectid &&
+	    found_key.type == log_key.type &&
+	    found_key.offset == log_key.offset &&
+	    btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
+		goto out;
+	}
+
+	/*
+	 * don't drop the conflicting directory entry if the inode
+	 * for the new entry doesn't exist
+	 */
+	if (!exists)
+		goto out;
+
+	ret = drop_one_dir_item(trans, root, path, dir, dst_di);
+	BUG_ON(ret);
+
+	if (key->type == BTRFS_DIR_INDEX_KEY)
+		goto insert;
+out:
+	btrfs_release_path(root, path);
+	kfree(name);
+	iput(dir);
+	return 0;
+
+insert:
+	btrfs_release_path(root, path);
+	ret = insert_one_name(trans, root, path, key->objectid, key->offset,
+			      name, name_len, log_type, &log_key);
+
+	if (ret && ret != -ENOENT)
+		BUG();
+	goto out;
+}
+
+/*
+ * find all the names in a directory item and reconcile them into
+ * the subvolume.  Only BTRFS_DIR_ITEM_KEY types will have more than
+ * one name in a directory item, but the same code gets used for
+ * both directory index types
+ */
+static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_path *path,
+					struct extent_buffer *eb, int slot,
+					struct btrfs_key *key)
+{
+	int ret;
+	u32 item_size = btrfs_item_size_nr(eb, slot);
+	struct btrfs_dir_item *di;
+	int name_len;
+	unsigned long ptr;
+	unsigned long ptr_end;
+
+	ptr = btrfs_item_ptr_offset(eb, slot);
+	ptr_end = ptr + item_size;
+	while (ptr < ptr_end) {
+		di = (struct btrfs_dir_item *)ptr;
+		name_len = btrfs_dir_name_len(eb, di);
+		ret = replay_one_name(trans, root, path, eb, di, key);
+		BUG_ON(ret);
+		ptr = (unsigned long)(di + 1);
+		ptr += name_len;
+	}
+	return 0;
+}
+
+/*
+ * directory replay has two parts.  There are the standard directory
+ * items in the log copied from the subvolume, and range items
+ * created in the log while the subvolume was logged.
+ *
+ * The range items tell us which parts of the key space the log
+ * is authoritative for.  During replay, if a key in the subvolume
+ * directory is in a logged range item, but not actually in the log
+ * that means it was deleted from the directory before the fsync
+ * and should be removed.
+ */
+static noinline int find_dir_range(struct btrfs_root *root,
+				   struct btrfs_path *path,
+				   u64 dirid, int key_type,
+				   u64 *start_ret, u64 *end_ret)
+{
+	struct btrfs_key key;
+	u64 found_end;
+	struct btrfs_dir_log_item *item;
+	int ret;
+	int nritems;
+
+	if (*start_ret == (u64)-1)
+		return 1;
+
+	key.objectid = dirid;
+	key.type = key_type;
+	key.offset = *start_ret;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		if (path->slots[0] == 0)
+			goto out;
+		path->slots[0]--;
+	}
+	if (ret != 0)
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	if (key.type != key_type || key.objectid != dirid) {
+		ret = 1;
+		goto next;
+	}
+	item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+			      struct btrfs_dir_log_item);
+	found_end = btrfs_dir_log_end(path->nodes[0], item);
+
+	if (*start_ret >= key.offset && *start_ret <= found_end) {
+		ret = 0;
+		*start_ret = key.offset;
+		*end_ret = found_end;
+		goto out;
+	}
+	ret = 1;
+next:
+	/* check the next slot in the tree to see if it is a valid item */
+	nritems = btrfs_header_nritems(path->nodes[0]);
+	if (path->slots[0] >= nritems) {
+		ret = btrfs_next_leaf(root, path);
+		if (ret)
+			goto out;
+	} else {
+		path->slots[0]++;
+	}
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	if (key.type != key_type || key.objectid != dirid) {
+		ret = 1;
+		goto out;
+	}
+	item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+			      struct btrfs_dir_log_item);
+	found_end = btrfs_dir_log_end(path->nodes[0], item);
+	*start_ret = key.offset;
+	*end_ret = found_end;
+	ret = 0;
+out:
+	btrfs_release_path(root, path);
+	return ret;
+}
+
+/*
+ * this looks for a given directory item in the log.  If the directory
+ * item is not in the log, the item is removed and the inode it points
+ * to is unlinked
+ */
+static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_root *log,
+				      struct btrfs_path *path,
+				      struct btrfs_path *log_path,
+				      struct inode *dir,
+				      struct btrfs_key *dir_key)
+{
+	int ret;
+	struct extent_buffer *eb;
+	int slot;
+	u32 item_size;
+	struct btrfs_dir_item *di;
+	struct btrfs_dir_item *log_di;
+	int name_len;
+	unsigned long ptr;
+	unsigned long ptr_end;
+	char *name;
+	struct inode *inode;
+	struct btrfs_key location;
+
+again:
+	eb = path->nodes[0];
+	slot = path->slots[0];
+	item_size = btrfs_item_size_nr(eb, slot);
+	ptr = btrfs_item_ptr_offset(eb, slot);
+	ptr_end = ptr + item_size;
+	while (ptr < ptr_end) {
+		di = (struct btrfs_dir_item *)ptr;
+		name_len = btrfs_dir_name_len(eb, di);
+		name = kmalloc(name_len, GFP_NOFS);
+		if (!name) {
+			ret = -ENOMEM;
+			goto out;
+		}
+		read_extent_buffer(eb, name, (unsigned long)(di + 1),
+				  name_len);
+		log_di = NULL;
+		if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
+			log_di = btrfs_lookup_dir_item(trans, log, log_path,
+						       dir_key->objectid,
+						       name, name_len, 0);
+		} else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
+			log_di = btrfs_lookup_dir_index_item(trans, log,
+						     log_path,
+						     dir_key->objectid,
+						     dir_key->offset,
+						     name, name_len, 0);
+		}
+		if (!log_di || IS_ERR(log_di)) {
+			btrfs_dir_item_key_to_cpu(eb, di, &location);
+			btrfs_release_path(root, path);
+			btrfs_release_path(log, log_path);
+			inode = read_one_inode(root, location.objectid);
+			BUG_ON(!inode);
+
+			ret = link_to_fixup_dir(trans, root,
+						path, location.objectid);
+			BUG_ON(ret);
+			btrfs_inc_nlink(inode);
+			ret = btrfs_unlink_inode(trans, root, dir, inode,
+						 name, name_len);
+			BUG_ON(ret);
+			kfree(name);
+			iput(inode);
+
+			/* there might still be more names under this key
+			 * check and repeat if required
+			 */
+			ret = btrfs_search_slot(NULL, root, dir_key, path,
+						0, 0);
+			if (ret == 0)
+				goto again;
+			ret = 0;
+			goto out;
+		}
+		btrfs_release_path(log, log_path);
+		kfree(name);
+
+		ptr = (unsigned long)(di + 1);
+		ptr += name_len;
+	}
+	ret = 0;
+out:
+	btrfs_release_path(root, path);
+	btrfs_release_path(log, log_path);
+	return ret;
+}
+
+/*
+ * deletion replay happens before we copy any new directory items
+ * out of the log or out of backreferences from inodes.  It
+ * scans the log to find ranges of keys that log is authoritative for,
+ * and then scans the directory to find items in those ranges that are
+ * not present in the log.
+ *
+ * Anything we don't find in the log is unlinked and removed from the
+ * directory.
+ */
+static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
+				       struct btrfs_root *root,
+				       struct btrfs_root *log,
+				       struct btrfs_path *path,
+				       u64 dirid)
+{
+	u64 range_start;
+	u64 range_end;
+	int key_type = BTRFS_DIR_LOG_ITEM_KEY;
+	int ret = 0;
+	struct btrfs_key dir_key;
+	struct btrfs_key found_key;
+	struct btrfs_path *log_path;
+	struct inode *dir;
+
+	dir_key.objectid = dirid;
+	dir_key.type = BTRFS_DIR_ITEM_KEY;
+	log_path = btrfs_alloc_path();
+	if (!log_path)
+		return -ENOMEM;
+
+	dir = read_one_inode(root, dirid);
+	/* it isn't an error if the inode isn't there, that can happen
+	 * because we replay the deletes before we copy in the inode item
+	 * from the log
+	 */
+	if (!dir) {
+		btrfs_free_path(log_path);
+		return 0;
+	}
+again:
+	range_start = 0;
+	range_end = 0;
+	while (1) {
+		ret = find_dir_range(log, path, dirid, key_type,
+				     &range_start, &range_end);
+		if (ret != 0)
+			break;
+
+		dir_key.offset = range_start;
+		while (1) {
+			int nritems;
+			ret = btrfs_search_slot(NULL, root, &dir_key, path,
+						0, 0);
+			if (ret < 0)
+				goto out;
+
+			nritems = btrfs_header_nritems(path->nodes[0]);
+			if (path->slots[0] >= nritems) {
+				ret = btrfs_next_leaf(root, path);
+				if (ret)
+					break;
+			}
+			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+					      path->slots[0]);
+			if (found_key.objectid != dirid ||
+			    found_key.type != dir_key.type)
+				goto next_type;
+
+			if (found_key.offset > range_end)
+				break;
+
+			ret = check_item_in_log(trans, root, log, path,
+						log_path, dir, &found_key);
+			BUG_ON(ret);
+			if (found_key.offset == (u64)-1)
+				break;
+			dir_key.offset = found_key.offset + 1;
+		}
+		btrfs_release_path(root, path);
+		if (range_end == (u64)-1)
+			break;
+		range_start = range_end + 1;
+	}
+
+next_type:
+	ret = 0;
+	if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
+		key_type = BTRFS_DIR_LOG_INDEX_KEY;
+		dir_key.type = BTRFS_DIR_INDEX_KEY;
+		btrfs_release_path(root, path);
+		goto again;
+	}
+out:
+	btrfs_release_path(root, path);
+	btrfs_free_path(log_path);
+	iput(dir);
+	return ret;
+}
+
+/*
+ * the process_func used to replay items from the log tree.  This
+ * gets called in two different stages.  The first stage just looks
+ * for inodes and makes sure they are all copied into the subvolume.
+ *
+ * The second stage copies all the other item types from the log into
+ * the subvolume.  The two stage approach is slower, but gets rid of
+ * lots of complexity around inodes referencing other inodes that exist
+ * only in the log (references come from either directory items or inode
+ * back refs).
+ */
+static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
+			     struct walk_control *wc, u64 gen)
+{
+	int nritems;
+	struct btrfs_path *path;
+	struct btrfs_root *root = wc->replay_dest;
+	struct btrfs_key key;
+	u32 item_size;
+	int level;
+	int i;
+	int ret;
+
+	btrfs_read_buffer(eb, gen);
+
+	level = btrfs_header_level(eb);
+
+	if (level != 0)
+		return 0;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	nritems = btrfs_header_nritems(eb);
+	for (i = 0; i < nritems; i++) {
+		btrfs_item_key_to_cpu(eb, &key, i);
+		item_size = btrfs_item_size_nr(eb, i);
+
+		/* inode keys are done during the first stage */
+		if (key.type == BTRFS_INODE_ITEM_KEY &&
+		    wc->stage == LOG_WALK_REPLAY_INODES) {
+			struct inode *inode;
+			struct btrfs_inode_item *inode_item;
+			u32 mode;
+
+			inode_item = btrfs_item_ptr(eb, i,
+					    struct btrfs_inode_item);
+			mode = btrfs_inode_mode(eb, inode_item);
+			if (S_ISDIR(mode)) {
+				ret = replay_dir_deletes(wc->trans,
+					 root, log, path, key.objectid);
+				BUG_ON(ret);
+			}
+			ret = overwrite_item(wc->trans, root, path,
+					     eb, i, &key);
+			BUG_ON(ret);
+
+			/* for regular files, truncate away
+			 * extents past the new EOF
+			 */
+			if (S_ISREG(mode)) {
+				inode = read_one_inode(root,
+						       key.objectid);
+				BUG_ON(!inode);
+
+				ret = btrfs_truncate_inode_items(wc->trans,
+					root, inode, inode->i_size,
+					BTRFS_EXTENT_DATA_KEY);
+				BUG_ON(ret);
+				iput(inode);
+			}
+			ret = link_to_fixup_dir(wc->trans, root,
+						path, key.objectid);
+			BUG_ON(ret);
+		}
+		if (wc->stage < LOG_WALK_REPLAY_ALL)
+			continue;
+
+		/* these keys are simply copied */
+		if (key.type == BTRFS_XATTR_ITEM_KEY) {
+			ret = overwrite_item(wc->trans, root, path,
+					     eb, i, &key);
+			BUG_ON(ret);
+		} else if (key.type == BTRFS_INODE_REF_KEY) {
+			ret = add_inode_ref(wc->trans, root, log, path,
+					    eb, i, &key);
+			BUG_ON(ret && ret != -ENOENT);
+		} else if (key.type == BTRFS_EXTENT_DATA_KEY) {
+			ret = replay_one_extent(wc->trans, root, path,
+						eb, i, &key);
+			BUG_ON(ret);
+		} else if (key.type == BTRFS_DIR_ITEM_KEY ||
+			   key.type == BTRFS_DIR_INDEX_KEY) {
+			ret = replay_one_dir_item(wc->trans, root, path,
+						  eb, i, &key);
+			BUG_ON(ret);
+		}
+	}
+	btrfs_free_path(path);
+	return 0;
+}
+
+static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct btrfs_path *path, int *level,
+				   struct walk_control *wc)
+{
+	u64 root_owner;
+	u64 root_gen;
+	u64 bytenr;
+	u64 ptr_gen;
+	struct extent_buffer *next;
+	struct extent_buffer *cur;
+	struct extent_buffer *parent;
+	u32 blocksize;
+	int ret = 0;
+
+	WARN_ON(*level < 0);
+	WARN_ON(*level >= BTRFS_MAX_LEVEL);
+
+	while (*level > 0) {
+		WARN_ON(*level < 0);
+		WARN_ON(*level >= BTRFS_MAX_LEVEL);
+		cur = path->nodes[*level];
+
+		if (btrfs_header_level(cur) != *level)
+			WARN_ON(1);
+
+		if (path->slots[*level] >=
+		    btrfs_header_nritems(cur))
+			break;
+
+		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
+		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
+		blocksize = btrfs_level_size(root, *level - 1);
+
+		parent = path->nodes[*level];
+		root_owner = btrfs_header_owner(parent);
+		root_gen = btrfs_header_generation(parent);
+
+		next = btrfs_find_create_tree_block(root, bytenr, blocksize);
+
+		wc->process_func(root, next, wc, ptr_gen);
+
+		if (*level == 1) {
+			path->slots[*level]++;
+			if (wc->free) {
+				btrfs_read_buffer(next, ptr_gen);
+
+				btrfs_tree_lock(next);
+				clean_tree_block(trans, root, next);
+				btrfs_wait_tree_block_writeback(next);
+				btrfs_tree_unlock(next);
+
+				ret = btrfs_drop_leaf_ref(trans, root, next);
+				BUG_ON(ret);
+
+				WARN_ON(root_owner !=
+					BTRFS_TREE_LOG_OBJECTID);
+				ret = btrfs_free_reserved_extent(root,
+							 bytenr, blocksize);
+				BUG_ON(ret);
+			}
+			free_extent_buffer(next);
+			continue;
+		}
+		btrfs_read_buffer(next, ptr_gen);
+
+		WARN_ON(*level <= 0);
+		if (path->nodes[*level-1])
+			free_extent_buffer(path->nodes[*level-1]);
+		path->nodes[*level-1] = next;
+		*level = btrfs_header_level(next);
+		path->slots[*level] = 0;
+		cond_resched();
+	}
+	WARN_ON(*level < 0);
+	WARN_ON(*level >= BTRFS_MAX_LEVEL);
+
+	if (path->nodes[*level] == root->node)
+		parent = path->nodes[*level];
+	else
+		parent = path->nodes[*level + 1];
+
+	bytenr = path->nodes[*level]->start;
+
+	blocksize = btrfs_level_size(root, *level);
+	root_owner = btrfs_header_owner(parent);
+	root_gen = btrfs_header_generation(parent);
+
+	wc->process_func(root, path->nodes[*level], wc,
+			 btrfs_header_generation(path->nodes[*level]));
+
+	if (wc->free) {
+		next = path->nodes[*level];
+		btrfs_tree_lock(next);
+		clean_tree_block(trans, root, next);
+		btrfs_wait_tree_block_writeback(next);
+		btrfs_tree_unlock(next);
+
+		if (*level == 0) {
+			ret = btrfs_drop_leaf_ref(trans, root, next);
+			BUG_ON(ret);
+		}
+		WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
+		ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
+		BUG_ON(ret);
+	}
+	free_extent_buffer(path->nodes[*level]);
+	path->nodes[*level] = NULL;
+	*level += 1;
+
+	cond_resched();
+	return 0;
+}
+
+static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path, int *level,
+				 struct walk_control *wc)
+{
+	u64 root_owner;
+	u64 root_gen;
+	int i;
+	int slot;
+	int ret;
+
+	for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
+		slot = path->slots[i];
+		if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
+			struct extent_buffer *node;
+			node = path->nodes[i];
+			path->slots[i]++;
+			*level = i;
+			WARN_ON(*level == 0);
+			return 0;
+		} else {
+			struct extent_buffer *parent;
+			if (path->nodes[*level] == root->node)
+				parent = path->nodes[*level];
+			else
+				parent = path->nodes[*level + 1];
+
+			root_owner = btrfs_header_owner(parent);
+			root_gen = btrfs_header_generation(parent);
+			wc->process_func(root, path->nodes[*level], wc,
+				 btrfs_header_generation(path->nodes[*level]));
+			if (wc->free) {
+				struct extent_buffer *next;
+
+				next = path->nodes[*level];
+
+				btrfs_tree_lock(next);
+				clean_tree_block(trans, root, next);
+				btrfs_wait_tree_block_writeback(next);
+				btrfs_tree_unlock(next);
+
+				if (*level == 0) {
+					ret = btrfs_drop_leaf_ref(trans, root,
+								  next);
+					BUG_ON(ret);
+				}
+
+				WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
+				ret = btrfs_free_reserved_extent(root,
+						path->nodes[*level]->start,
+						path->nodes[*level]->len);
+				BUG_ON(ret);
+			}
+			free_extent_buffer(path->nodes[*level]);
+			path->nodes[*level] = NULL;
+			*level = i + 1;
+		}
+	}
+	return 1;
+}
+
+/*
+ * drop the reference count on the tree rooted at 'snap'.  This traverses
+ * the tree freeing any blocks that have a ref count of zero after being
+ * decremented.
+ */
+static int walk_log_tree(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *log, struct walk_control *wc)
+{
+	int ret = 0;
+	int wret;
+	int level;
+	struct btrfs_path *path;
+	int i;
+	int orig_level;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	level = btrfs_header_level(log->node);
+	orig_level = level;
+	path->nodes[level] = log->node;
+	extent_buffer_get(log->node);
+	path->slots[level] = 0;
+
+	while (1) {
+		wret = walk_down_log_tree(trans, log, path, &level, wc);
+		if (wret > 0)
+			break;
+		if (wret < 0)
+			ret = wret;
+
+		wret = walk_up_log_tree(trans, log, path, &level, wc);
+		if (wret > 0)
+			break;
+		if (wret < 0)
+			ret = wret;
+	}
+
+	/* was the root node processed? if not, catch it here */
+	if (path->nodes[orig_level]) {
+		wc->process_func(log, path->nodes[orig_level], wc,
+			 btrfs_header_generation(path->nodes[orig_level]));
+		if (wc->free) {
+			struct extent_buffer *next;
+
+			next = path->nodes[orig_level];
+
+			btrfs_tree_lock(next);
+			clean_tree_block(trans, log, next);
+			btrfs_wait_tree_block_writeback(next);
+			btrfs_tree_unlock(next);
+
+			if (orig_level == 0) {
+				ret = btrfs_drop_leaf_ref(trans, log,
+							  next);
+				BUG_ON(ret);
+			}
+			WARN_ON(log->root_key.objectid !=
+				BTRFS_TREE_LOG_OBJECTID);
+			ret = btrfs_free_reserved_extent(log, next->start,
+							 next->len);
+			BUG_ON(ret);
+		}
+	}
+
+	for (i = 0; i <= orig_level; i++) {
+		if (path->nodes[i]) {
+			free_extent_buffer(path->nodes[i]);
+			path->nodes[i] = NULL;
+		}
+	}
+	btrfs_free_path(path);
+	if (wc->free)
+		free_extent_buffer(log->node);
+	return ret;
+}
+
+static int wait_log_commit(struct btrfs_root *log)
+{
+	DEFINE_WAIT(wait);
+	u64 transid = log->fs_info->tree_log_transid;
+
+	do {
+		prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
+				TASK_UNINTERRUPTIBLE);
+		mutex_unlock(&log->fs_info->tree_log_mutex);
+		if (atomic_read(&log->fs_info->tree_log_commit))
+			schedule();
+		finish_wait(&log->fs_info->tree_log_wait, &wait);
+		mutex_lock(&log->fs_info->tree_log_mutex);
+	} while (transid == log->fs_info->tree_log_transid &&
+		atomic_read(&log->fs_info->tree_log_commit));
+	return 0;
+}
+
+/*
+ * btrfs_sync_log does sends a given tree log down to the disk and
+ * updates the super blocks to record it.  When this call is done,
+ * you know that any inodes previously logged are safely on disk
+ */
+int btrfs_sync_log(struct btrfs_trans_handle *trans,
+		   struct btrfs_root *root)
+{
+	int ret;
+	unsigned long batch;
+	struct btrfs_root *log = root->log_root;
+
+	mutex_lock(&log->fs_info->tree_log_mutex);
+	if (atomic_read(&log->fs_info->tree_log_commit)) {
+		wait_log_commit(log);
+		goto out;
+	}
+	atomic_set(&log->fs_info->tree_log_commit, 1);
+
+	while (1) {
+		batch = log->fs_info->tree_log_batch;
+		mutex_unlock(&log->fs_info->tree_log_mutex);
+		schedule_timeout_uninterruptible(1);
+		mutex_lock(&log->fs_info->tree_log_mutex);
+
+		while (atomic_read(&log->fs_info->tree_log_writers)) {
+			DEFINE_WAIT(wait);
+			prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
+					TASK_UNINTERRUPTIBLE);
+			mutex_unlock(&log->fs_info->tree_log_mutex);
+			if (atomic_read(&log->fs_info->tree_log_writers))
+				schedule();
+			mutex_lock(&log->fs_info->tree_log_mutex);
+			finish_wait(&log->fs_info->tree_log_wait, &wait);
+		}
+		if (batch == log->fs_info->tree_log_batch)
+			break;
+	}
+
+	ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
+	BUG_ON(ret);
+	ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree,
+			       &root->fs_info->log_root_tree->dirty_log_pages);
+	BUG_ON(ret);
+
+	btrfs_set_super_log_root(&root->fs_info->super_for_commit,
+				 log->fs_info->log_root_tree->node->start);
+	btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
+		       btrfs_header_level(log->fs_info->log_root_tree->node));
+
+	write_ctree_super(trans, log->fs_info->tree_root, 2);
+	log->fs_info->tree_log_transid++;
+	log->fs_info->tree_log_batch = 0;
+	atomic_set(&log->fs_info->tree_log_commit, 0);
+	smp_mb();
+	if (waitqueue_active(&log->fs_info->tree_log_wait))
+		wake_up(&log->fs_info->tree_log_wait);
+out:
+	mutex_unlock(&log->fs_info->tree_log_mutex);
+	return 0;
+}
+
+/* * free all the extents used by the tree log.  This should be called
+ * at commit time of the full transaction
+ */
+int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
+{
+	int ret;
+	struct btrfs_root *log;
+	struct key;
+	u64 start;
+	u64 end;
+	struct walk_control wc = {
+		.free = 1,
+		.process_func = process_one_buffer
+	};
+
+	if (!root->log_root || root->fs_info->log_root_recovering)
+		return 0;
+
+	log = root->log_root;
+	ret = walk_log_tree(trans, log, &wc);
+	BUG_ON(ret);
+
+	while (1) {
+		ret = find_first_extent_bit(&log->dirty_log_pages,
+				    0, &start, &end, EXTENT_DIRTY);
+		if (ret)
+			break;
+
+		clear_extent_dirty(&log->dirty_log_pages,
+				   start, end, GFP_NOFS);
+	}
+
+	log = root->log_root;
+	ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
+			     &log->root_key);
+	BUG_ON(ret);
+	root->log_root = NULL;
+	kfree(root->log_root);
+	return 0;
+}
+
+/*
+ * helper function to update the item for a given subvolumes log root
+ * in the tree of log roots
+ */
+static int update_log_root(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *log)
+{
+	u64 bytenr = btrfs_root_bytenr(&log->root_item);
+	int ret;
+
+	if (log->node->start == bytenr)
+		return 0;
+
+	btrfs_set_root_bytenr(&log->root_item, log->node->start);
+	btrfs_set_root_generation(&log->root_item, trans->transid);
+	btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
+	ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
+				&log->root_key, &log->root_item);
+	BUG_ON(ret);
+	return ret;
+}
+
+/*
+ * If both a file and directory are logged, and unlinks or renames are
+ * mixed in, we have a few interesting corners:
+ *
+ * create file X in dir Y
+ * link file X to X.link in dir Y
+ * fsync file X
+ * unlink file X but leave X.link
+ * fsync dir Y
+ *
+ * After a crash we would expect only X.link to exist.  But file X
+ * didn't get fsync'd again so the log has back refs for X and X.link.
+ *
+ * We solve this by removing directory entries and inode backrefs from the
+ * log when a file that was logged in the current transaction is
+ * unlinked.  Any later fsync will include the updated log entries, and
+ * we'll be able to reconstruct the proper directory items from backrefs.
+ *
+ * This optimizations allows us to avoid relogging the entire inode
+ * or the entire directory.
+ */
+int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 const char *name, int name_len,
+				 struct inode *dir, u64 index)
+{
+	struct btrfs_root *log;
+	struct btrfs_dir_item *di;
+	struct btrfs_path *path;
+	int ret;
+	int bytes_del = 0;
+
+	if (BTRFS_I(dir)->logged_trans < trans->transid)
+		return 0;
+
+	ret = join_running_log_trans(root);
+	if (ret)
+		return 0;
+
+	mutex_lock(&BTRFS_I(dir)->log_mutex);
+
+	log = root->log_root;
+	path = btrfs_alloc_path();
+	di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
+				   name, name_len, -1);
+	if (di && !IS_ERR(di)) {
+		ret = btrfs_delete_one_dir_name(trans, log, path, di);
+		bytes_del += name_len;
+		BUG_ON(ret);
+	}
+	btrfs_release_path(log, path);
+	di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
+					 index, name, name_len, -1);
+	if (di && !IS_ERR(di)) {
+		ret = btrfs_delete_one_dir_name(trans, log, path, di);
+		bytes_del += name_len;
+		BUG_ON(ret);
+	}
+
+	/* update the directory size in the log to reflect the names
+	 * we have removed
+	 */
+	if (bytes_del) {
+		struct btrfs_key key;
+
+		key.objectid = dir->i_ino;
+		key.offset = 0;
+		key.type = BTRFS_INODE_ITEM_KEY;
+		btrfs_release_path(log, path);
+
+		ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
+		if (ret == 0) {
+			struct btrfs_inode_item *item;
+			u64 i_size;
+
+			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+					      struct btrfs_inode_item);
+			i_size = btrfs_inode_size(path->nodes[0], item);
+			if (i_size > bytes_del)
+				i_size -= bytes_del;
+			else
+				i_size = 0;
+			btrfs_set_inode_size(path->nodes[0], item, i_size);
+			btrfs_mark_buffer_dirty(path->nodes[0]);
+		} else
+			ret = 0;
+		btrfs_release_path(log, path);
+	}
+
+	btrfs_free_path(path);
+	mutex_unlock(&BTRFS_I(dir)->log_mutex);
+	end_log_trans(root);
+
+	return 0;
+}
+
+/* see comments for btrfs_del_dir_entries_in_log */
+int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       const char *name, int name_len,
+			       struct inode *inode, u64 dirid)
+{
+	struct btrfs_root *log;
+	u64 index;
+	int ret;
+
+	if (BTRFS_I(inode)->logged_trans < trans->transid)
+		return 0;
+
+	ret = join_running_log_trans(root);
+	if (ret)
+		return 0;
+	log = root->log_root;
+	mutex_lock(&BTRFS_I(inode)->log_mutex);
+
+	ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
+				  dirid, &index);
+	mutex_unlock(&BTRFS_I(inode)->log_mutex);
+	end_log_trans(root);
+
+	return ret;
+}
+
+/*
+ * creates a range item in the log for 'dirid'.  first_offset and
+ * last_offset tell us which parts of the key space the log should
+ * be considered authoritative for.
+ */
+static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
+				       struct btrfs_root *log,
+				       struct btrfs_path *path,
+				       int key_type, u64 dirid,
+				       u64 first_offset, u64 last_offset)
+{
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_dir_log_item *item;
+
+	key.objectid = dirid;
+	key.offset = first_offset;
+	if (key_type == BTRFS_DIR_ITEM_KEY)
+		key.type = BTRFS_DIR_LOG_ITEM_KEY;
+	else
+		key.type = BTRFS_DIR_LOG_INDEX_KEY;
+	ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
+	BUG_ON(ret);
+
+	item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+			      struct btrfs_dir_log_item);
+	btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_release_path(log, path);
+	return 0;
+}
+
+/*
+ * log all the items included in the current transaction for a given
+ * directory.  This also creates the range items in the log tree required
+ * to replay anything deleted before the fsync
+ */
+static noinline int log_dir_items(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct inode *inode,
+			  struct btrfs_path *path,
+			  struct btrfs_path *dst_path, int key_type,
+			  u64 min_offset, u64 *last_offset_ret)
+{
+	struct btrfs_key min_key;
+	struct btrfs_key max_key;
+	struct btrfs_root *log = root->log_root;
+	struct extent_buffer *src;
+	int ret;
+	int i;
+	int nritems;
+	u64 first_offset = min_offset;
+	u64 last_offset = (u64)-1;
+
+	log = root->log_root;
+	max_key.objectid = inode->i_ino;
+	max_key.offset = (u64)-1;
+	max_key.type = key_type;
+
+	min_key.objectid = inode->i_ino;
+	min_key.type = key_type;
+	min_key.offset = min_offset;
+
+	path->keep_locks = 1;
+
+	ret = btrfs_search_forward(root, &min_key, &max_key,
+				   path, 0, trans->transid);
+
+	/*
+	 * we didn't find anything from this transaction, see if there
+	 * is anything at all
+	 */
+	if (ret != 0 || min_key.objectid != inode->i_ino ||
+	    min_key.type != key_type) {
+		min_key.objectid = inode->i_ino;
+		min_key.type = key_type;
+		min_key.offset = (u64)-1;
+		btrfs_release_path(root, path);
+		ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
+		if (ret < 0) {
+			btrfs_release_path(root, path);
+			return ret;
+		}
+		ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
+
+		/* if ret == 0 there are items for this type,
+		 * create a range to tell us the last key of this type.
+		 * otherwise, there are no items in this directory after
+		 * *min_offset, and we create a range to indicate that.
+		 */
+		if (ret == 0) {
+			struct btrfs_key tmp;
+			btrfs_item_key_to_cpu(path->nodes[0], &tmp,
+					      path->slots[0]);
+			if (key_type == tmp.type)
+				first_offset = max(min_offset, tmp.offset) + 1;
+		}
+		goto done;
+	}
+
+	/* go backward to find any previous key */
+	ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
+	if (ret == 0) {
+		struct btrfs_key tmp;
+		btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
+		if (key_type == tmp.type) {
+			first_offset = tmp.offset;
+			ret = overwrite_item(trans, log, dst_path,
+					     path->nodes[0], path->slots[0],
+					     &tmp);
+		}
+	}
+	btrfs_release_path(root, path);
+
+	/* find the first key from this transaction again */
+	ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
+	if (ret != 0) {
+		WARN_ON(1);
+		goto done;
+	}
+
+	/*
+	 * we have a block from this transaction, log every item in it
+	 * from our directory
+	 */
+	while (1) {
+		struct btrfs_key tmp;
+		src = path->nodes[0];
+		nritems = btrfs_header_nritems(src);
+		for (i = path->slots[0]; i < nritems; i++) {
+			btrfs_item_key_to_cpu(src, &min_key, i);
+
+			if (min_key.objectid != inode->i_ino ||
+			    min_key.type != key_type)
+				goto done;
+			ret = overwrite_item(trans, log, dst_path, src, i,
+					     &min_key);
+			BUG_ON(ret);
+		}
+		path->slots[0] = nritems;
+
+		/*
+		 * look ahead to the next item and see if it is also
+		 * from this directory and from this transaction
+		 */
+		ret = btrfs_next_leaf(root, path);
+		if (ret == 1) {
+			last_offset = (u64)-1;
+			goto done;
+		}
+		btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
+		if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
+			last_offset = (u64)-1;
+			goto done;
+		}
+		if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
+			ret = overwrite_item(trans, log, dst_path,
+					     path->nodes[0], path->slots[0],
+					     &tmp);
+
+			BUG_ON(ret);
+			last_offset = tmp.offset;
+			goto done;
+		}
+	}
+done:
+	*last_offset_ret = last_offset;
+	btrfs_release_path(root, path);
+	btrfs_release_path(log, dst_path);
+
+	/* insert the log range keys to indicate where the log is valid */
+	ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
+				 first_offset, last_offset);
+	BUG_ON(ret);
+	return 0;
+}
+
+/*
+ * logging directories is very similar to logging inodes, We find all the items
+ * from the current transaction and write them to the log.
+ *
+ * The recovery code scans the directory in the subvolume, and if it finds a
+ * key in the range logged that is not present in the log tree, then it means
+ * that dir entry was unlinked during the transaction.
+ *
+ * In order for that scan to work, we must include one key smaller than
+ * the smallest logged by this transaction and one key larger than the largest
+ * key logged by this transaction.
+ */
+static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct inode *inode,
+			  struct btrfs_path *path,
+			  struct btrfs_path *dst_path)
+{
+	u64 min_key;
+	u64 max_key;
+	int ret;
+	int key_type = BTRFS_DIR_ITEM_KEY;
+
+again:
+	min_key = 0;
+	max_key = 0;
+	while (1) {
+		ret = log_dir_items(trans, root, inode, path,
+				    dst_path, key_type, min_key,
+				    &max_key);
+		BUG_ON(ret);
+		if (max_key == (u64)-1)
+			break;
+		min_key = max_key + 1;
+	}
+
+	if (key_type == BTRFS_DIR_ITEM_KEY) {
+		key_type = BTRFS_DIR_INDEX_KEY;
+		goto again;
+	}
+	return 0;
+}
+
+/*
+ * a helper function to drop items from the log before we relog an
+ * inode.  max_key_type indicates the highest item type to remove.
+ * This cannot be run for file data extents because it does not
+ * free the extents they point to.
+ */
+static int drop_objectid_items(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *log,
+				  struct btrfs_path *path,
+				  u64 objectid, int max_key_type)
+{
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+
+	key.objectid = objectid;
+	key.type = max_key_type;
+	key.offset = (u64)-1;
+
+	while (1) {
+		ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
+
+		if (ret != 1)
+			break;
+
+		if (path->slots[0] == 0)
+			break;
+
+		path->slots[0]--;
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+				      path->slots[0]);
+
+		if (found_key.objectid != objectid)
+			break;
+
+		ret = btrfs_del_item(trans, log, path);
+		BUG_ON(ret);
+		btrfs_release_path(log, path);
+	}
+	btrfs_release_path(log, path);
+	return 0;
+}
+
+static noinline int copy_items(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *log,
+			       struct btrfs_path *dst_path,
+			       struct extent_buffer *src,
+			       int start_slot, int nr, int inode_only)
+{
+	unsigned long src_offset;
+	unsigned long dst_offset;
+	struct btrfs_file_extent_item *extent;
+	struct btrfs_inode_item *inode_item;
+	int ret;
+	struct btrfs_key *ins_keys;
+	u32 *ins_sizes;
+	char *ins_data;
+	int i;
+	struct list_head ordered_sums;
+
+	INIT_LIST_HEAD(&ordered_sums);
+
+	ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
+			   nr * sizeof(u32), GFP_NOFS);
+	ins_sizes = (u32 *)ins_data;
+	ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
+
+	for (i = 0; i < nr; i++) {
+		ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
+		btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
+	}
+	ret = btrfs_insert_empty_items(trans, log, dst_path,
+				       ins_keys, ins_sizes, nr);
+	BUG_ON(ret);
+
+	for (i = 0; i < nr; i++) {
+		dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
+						   dst_path->slots[0]);
+
+		src_offset = btrfs_item_ptr_offset(src, start_slot + i);
+
+		copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
+				   src_offset, ins_sizes[i]);
+
+		if (inode_only == LOG_INODE_EXISTS &&
+		    ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
+			inode_item = btrfs_item_ptr(dst_path->nodes[0],
+						    dst_path->slots[0],
+						    struct btrfs_inode_item);
+			btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
+
+			/* set the generation to zero so the recover code
+			 * can tell the difference between an logging
+			 * just to say 'this inode exists' and a logging
+			 * to say 'update this inode with these values'
+			 */
+			btrfs_set_inode_generation(dst_path->nodes[0],
+						   inode_item, 0);
+		}
+		/* take a reference on file data extents so that truncates
+		 * or deletes of this inode don't have to relog the inode
+		 * again
+		 */
+		if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
+			int found_type;
+			extent = btrfs_item_ptr(src, start_slot + i,
+						struct btrfs_file_extent_item);
+
+			found_type = btrfs_file_extent_type(src, extent);
+			if (found_type == BTRFS_FILE_EXTENT_REG ||
+			    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
+				u64 ds = btrfs_file_extent_disk_bytenr(src,
+								   extent);
+				u64 dl = btrfs_file_extent_disk_num_bytes(src,
+								      extent);
+				u64 cs = btrfs_file_extent_offset(src, extent);
+				u64 cl = btrfs_file_extent_num_bytes(src,
+								     extent);;
+				if (btrfs_file_extent_compression(src,
+								  extent)) {
+					cs = 0;
+					cl = dl;
+				}
+				/* ds == 0 is a hole */
+				if (ds != 0) {
+					ret = btrfs_inc_extent_ref(trans, log,
+						   ds, dl,
+						   dst_path->nodes[0]->start,
+						   BTRFS_TREE_LOG_OBJECTID,
+						   trans->transid,
+						   ins_keys[i].objectid);
+					BUG_ON(ret);
+					ret = btrfs_lookup_csums_range(
+						   log->fs_info->csum_root,
+						   ds + cs, ds + cs + cl - 1,
+						   &ordered_sums);
+					BUG_ON(ret);
+				}
+			}
+		}
+		dst_path->slots[0]++;
+	}
+
+	btrfs_mark_buffer_dirty(dst_path->nodes[0]);
+	btrfs_release_path(log, dst_path);
+	kfree(ins_data);
+
+	/*
+	 * we have to do this after the loop above to avoid changing the
+	 * log tree while trying to change the log tree.
+	 */
+	while (!list_empty(&ordered_sums)) {
+		struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
+						   struct btrfs_ordered_sum,
+						   list);
+		ret = btrfs_csum_file_blocks(trans, log, sums);
+		BUG_ON(ret);
+		list_del(&sums->list);
+		kfree(sums);
+	}
+	return 0;
+}
+
+/* log a single inode in the tree log.
+ * At least one parent directory for this inode must exist in the tree
+ * or be logged already.
+ *
+ * Any items from this inode changed by the current transaction are copied
+ * to the log tree.  An extra reference is taken on any extents in this
+ * file, allowing us to avoid a whole pile of corner cases around logging
+ * blocks that have been removed from the tree.
+ *
+ * See LOG_INODE_ALL and related defines for a description of what inode_only
+ * does.
+ *
+ * This handles both files and directories.
+ */
+static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, struct inode *inode,
+			     int inode_only)
+{
+	struct btrfs_path *path;
+	struct btrfs_path *dst_path;
+	struct btrfs_key min_key;
+	struct btrfs_key max_key;
+	struct btrfs_root *log = root->log_root;
+	struct extent_buffer *src = NULL;
+	u32 size;
+	int ret;
+	int nritems;
+	int ins_start_slot = 0;
+	int ins_nr;
+
+	log = root->log_root;
+
+	path = btrfs_alloc_path();
+	dst_path = btrfs_alloc_path();
+
+	min_key.objectid = inode->i_ino;
+	min_key.type = BTRFS_INODE_ITEM_KEY;
+	min_key.offset = 0;
+
+	max_key.objectid = inode->i_ino;
+	if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
+		max_key.type = BTRFS_XATTR_ITEM_KEY;
+	else
+		max_key.type = (u8)-1;
+	max_key.offset = (u64)-1;
+
+	/*
+	 * if this inode has already been logged and we're in inode_only
+	 * mode, we don't want to delete the things that have already
+	 * been written to the log.
+	 *
+	 * But, if the inode has been through an inode_only log,
+	 * the logged_trans field is not set.  This allows us to catch
+	 * any new names for this inode in the backrefs by logging it
+	 * again
+	 */
+	if (inode_only == LOG_INODE_EXISTS &&
+	    BTRFS_I(inode)->logged_trans == trans->transid) {
+		btrfs_free_path(path);
+		btrfs_free_path(dst_path);
+		goto out;
+	}
+	mutex_lock(&BTRFS_I(inode)->log_mutex);
+
+	/*
+	 * a brute force approach to making sure we get the most uptodate
+	 * copies of everything.
+	 */
+	if (S_ISDIR(inode->i_mode)) {
+		int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
+
+		if (inode_only == LOG_INODE_EXISTS)
+			max_key_type = BTRFS_XATTR_ITEM_KEY;
+		ret = drop_objectid_items(trans, log, path,
+					  inode->i_ino, max_key_type);
+	} else {
+		ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
+	}
+	BUG_ON(ret);
+	path->keep_locks = 1;
+
+	while (1) {
+		ins_nr = 0;
+		ret = btrfs_search_forward(root, &min_key, &max_key,
+					   path, 0, trans->transid);
+		if (ret != 0)
+			break;
+again:
+		/* note, ins_nr might be > 0 here, cleanup outside the loop */
+		if (min_key.objectid != inode->i_ino)
+			break;
+		if (min_key.type > max_key.type)
+			break;
+
+		src = path->nodes[0];
+		size = btrfs_item_size_nr(src, path->slots[0]);
+		if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
+			ins_nr++;
+			goto next_slot;
+		} else if (!ins_nr) {
+			ins_start_slot = path->slots[0];
+			ins_nr = 1;
+			goto next_slot;
+		}
+
+		ret = copy_items(trans, log, dst_path, src, ins_start_slot,
+				 ins_nr, inode_only);
+		BUG_ON(ret);
+		ins_nr = 1;
+		ins_start_slot = path->slots[0];
+next_slot:
+
+		nritems = btrfs_header_nritems(path->nodes[0]);
+		path->slots[0]++;
+		if (path->slots[0] < nritems) {
+			btrfs_item_key_to_cpu(path->nodes[0], &min_key,
+					      path->slots[0]);
+			goto again;
+		}
+		if (ins_nr) {
+			ret = copy_items(trans, log, dst_path, src,
+					 ins_start_slot,
+					 ins_nr, inode_only);
+			BUG_ON(ret);
+			ins_nr = 0;
+		}
+		btrfs_release_path(root, path);
+
+		if (min_key.offset < (u64)-1)
+			min_key.offset++;
+		else if (min_key.type < (u8)-1)
+			min_key.type++;
+		else if (min_key.objectid < (u64)-1)
+			min_key.objectid++;
+		else
+			break;
+	}
+	if (ins_nr) {
+		ret = copy_items(trans, log, dst_path, src,
+				 ins_start_slot,
+				 ins_nr, inode_only);
+		BUG_ON(ret);
+		ins_nr = 0;
+	}
+	WARN_ON(ins_nr);
+	if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
+		btrfs_release_path(root, path);
+		btrfs_release_path(log, dst_path);
+		BTRFS_I(inode)->log_dirty_trans = 0;
+		ret = log_directory_changes(trans, root, inode, path, dst_path);
+		BUG_ON(ret);
+	}
+	BTRFS_I(inode)->logged_trans = trans->transid;
+	mutex_unlock(&BTRFS_I(inode)->log_mutex);
+
+	btrfs_free_path(path);
+	btrfs_free_path(dst_path);
+
+	mutex_lock(&root->fs_info->tree_log_mutex);
+	ret = update_log_root(trans, log);
+	BUG_ON(ret);
+	mutex_unlock(&root->fs_info->tree_log_mutex);
+out:
+	return 0;
+}
+
+int btrfs_log_inode(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, struct inode *inode,
+		    int inode_only)
+{
+	int ret;
+
+	start_log_trans(trans, root);
+	ret = __btrfs_log_inode(trans, root, inode, inode_only);
+	end_log_trans(root);
+	return ret;
+}
+
+/*
+ * helper function around btrfs_log_inode to make sure newly created
+ * parent directories also end up in the log.  A minimal inode and backref
+ * only logging is done of any parent directories that are older than
+ * the last committed transaction
+ */
+int btrfs_log_dentry(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, struct dentry *dentry)
+{
+	int inode_only = LOG_INODE_ALL;
+	struct super_block *sb;
+	int ret;
+
+	start_log_trans(trans, root);
+	sb = dentry->d_inode->i_sb;
+	while (1) {
+		ret = __btrfs_log_inode(trans, root, dentry->d_inode,
+					inode_only);
+		BUG_ON(ret);
+		inode_only = LOG_INODE_EXISTS;
+
+		dentry = dentry->d_parent;
+		if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
+			break;
+
+		if (BTRFS_I(dentry->d_inode)->generation <=
+		    root->fs_info->last_trans_committed)
+			break;
+	}
+	end_log_trans(root);
+	return 0;
+}
+
+/*
+ * it is not safe to log dentry if the chunk root has added new
+ * chunks.  This returns 0 if the dentry was logged, and 1 otherwise.
+ * If this returns 1, you must commit the transaction to safely get your
+ * data on disk.
+ */
+int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct dentry *dentry)
+{
+	u64 gen;
+	gen = root->fs_info->last_trans_new_blockgroup;
+	if (gen > root->fs_info->last_trans_committed)
+		return 1;
+	else
+		return btrfs_log_dentry(trans, root, dentry);
+}
+
+/*
+ * should be called during mount to recover any replay any log trees
+ * from the FS
+ */
+int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct btrfs_key tmp_key;
+	struct btrfs_root *log;
+	struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
+	u64 highest_inode;
+	struct walk_control wc = {
+		.process_func = process_one_buffer,
+		.stage = 0,
+	};
+
+	fs_info->log_root_recovering = 1;
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	trans = btrfs_start_transaction(fs_info->tree_root, 1);
+
+	wc.trans = trans;
+	wc.pin = 1;
+
+	walk_log_tree(trans, log_root_tree, &wc);
+
+again:
+	key.objectid = BTRFS_TREE_LOG_OBJECTID;
+	key.offset = (u64)-1;
+	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
+		if (ret < 0)
+			break;
+		if (ret > 0) {
+			if (path->slots[0] == 0)
+				break;
+			path->slots[0]--;
+		}
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+				      path->slots[0]);
+		btrfs_release_path(log_root_tree, path);
+		if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
+			break;
+
+		log = btrfs_read_fs_root_no_radix(log_root_tree,
+						  &found_key);
+		BUG_ON(!log);
+
+
+		tmp_key.objectid = found_key.offset;
+		tmp_key.type = BTRFS_ROOT_ITEM_KEY;
+		tmp_key.offset = (u64)-1;
+
+		wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
+		BUG_ON(!wc.replay_dest);
+
+		wc.replay_dest->log_root = log;
+		btrfs_record_root_in_trans(wc.replay_dest);
+		ret = walk_log_tree(trans, log, &wc);
+		BUG_ON(ret);
+
+		if (wc.stage == LOG_WALK_REPLAY_ALL) {
+			ret = fixup_inode_link_counts(trans, wc.replay_dest,
+						      path);
+			BUG_ON(ret);
+		}
+		ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
+		if (ret == 0) {
+			wc.replay_dest->highest_inode = highest_inode;
+			wc.replay_dest->last_inode_alloc = highest_inode;
+		}
+
+		key.offset = found_key.offset - 1;
+		wc.replay_dest->log_root = NULL;
+		free_extent_buffer(log->node);
+		kfree(log);
+
+		if (found_key.offset == 0)
+			break;
+	}
+	btrfs_release_path(log_root_tree, path);
+
+	/* step one is to pin it all, step two is to replay just inodes */
+	if (wc.pin) {
+		wc.pin = 0;
+		wc.process_func = replay_one_buffer;
+		wc.stage = LOG_WALK_REPLAY_INODES;
+		goto again;
+	}
+	/* step three is to replay everything */
+	if (wc.stage < LOG_WALK_REPLAY_ALL) {
+		wc.stage++;
+		goto again;
+	}
+
+	btrfs_free_path(path);
+
+	free_extent_buffer(log_root_tree->node);
+	log_root_tree->log_root = NULL;
+	fs_info->log_root_recovering = 0;
+
+	/* step 4: commit the transaction, which also unpins the blocks */
+	btrfs_commit_transaction(trans, fs_info->tree_root);
+
+	kfree(log_root_tree);
+	return 0;
+}
diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h
new file mode 100644
index 0000000..b9409b3
--- /dev/null
+++ b/fs/btrfs/tree-log.h
@@ -0,0 +1,41 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __TREE_LOG_
+#define __TREE_LOG_
+
+int btrfs_sync_log(struct btrfs_trans_handle *trans,
+		   struct btrfs_root *root);
+int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root);
+int btrfs_log_dentry(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, struct dentry *dentry);
+int btrfs_recover_log_trees(struct btrfs_root *tree_root);
+int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct dentry *dentry);
+int btrfs_log_inode(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, struct inode *inode,
+		    int inode_only);
+int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 const char *name, int name_len,
+				 struct inode *dir, u64 index);
+int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       const char *name, int name_len,
+			       struct inode *inode, u64 dirid);
+#endif
-- 
1.6.0.2

[PATCH 04/35] Btrfs transaction code

[Chris Mason]

Btrfs transactions are really just snapshots, but this provides an API
around safely coordinating them.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
new file mode 100644
index 0000000..8a08f94
--- /dev/null
+++ b/fs/btrfs/transaction.c
@@ -0,0 +1,1097 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/sched.h>
+#include <linux/writeback.h>
+#include <linux/pagemap.h>
+#include <linux/blkdev.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "locking.h"
+#include "ref-cache.h"
+#include "tree-log.h"
+
+#define BTRFS_ROOT_TRANS_TAG 0
+
+static noinline void put_transaction(struct btrfs_transaction *transaction)
+{
+	WARN_ON(transaction->use_count == 0);
+	transaction->use_count--;
+	if (transaction->use_count == 0) {
+		list_del_init(&transaction->list);
+		memset(transaction, 0, sizeof(*transaction));
+		kmem_cache_free(btrfs_transaction_cachep, transaction);
+	}
+}
+
+/*
+ * either allocate a new transaction or hop into the existing one
+ */
+static noinline int join_transaction(struct btrfs_root *root)
+{
+	struct btrfs_transaction *cur_trans;
+	cur_trans = root->fs_info->running_transaction;
+	if (!cur_trans) {
+		cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
+					     GFP_NOFS);
+		BUG_ON(!cur_trans);
+		root->fs_info->generation++;
+		root->fs_info->last_alloc = 0;
+		root->fs_info->last_data_alloc = 0;
+		cur_trans->num_writers = 1;
+		cur_trans->num_joined = 0;
+		cur_trans->transid = root->fs_info->generation;
+		init_waitqueue_head(&cur_trans->writer_wait);
+		init_waitqueue_head(&cur_trans->commit_wait);
+		cur_trans->in_commit = 0;
+		cur_trans->blocked = 0;
+		cur_trans->use_count = 1;
+		cur_trans->commit_done = 0;
+		cur_trans->start_time = get_seconds();
+		INIT_LIST_HEAD(&cur_trans->pending_snapshots);
+		list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
+		extent_io_tree_init(&cur_trans->dirty_pages,
+				     root->fs_info->btree_inode->i_mapping,
+				     GFP_NOFS);
+		spin_lock(&root->fs_info->new_trans_lock);
+		root->fs_info->running_transaction = cur_trans;
+		spin_unlock(&root->fs_info->new_trans_lock);
+	} else {
+		cur_trans->num_writers++;
+		cur_trans->num_joined++;
+	}
+
+	return 0;
+}
+
+/*
+ * this does all the record keeping required to make sure that a reference
+ * counted root is properly recorded in a given transaction.  This is required
+ * to make sure the old root from before we joined the transaction is deleted
+ * when the transaction commits
+ */
+noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
+{
+	struct btrfs_dirty_root *dirty;
+	u64 running_trans_id = root->fs_info->running_transaction->transid;
+	if (root->ref_cows && root->last_trans < running_trans_id) {
+		WARN_ON(root == root->fs_info->extent_root);
+		if (root->root_item.refs != 0) {
+			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+				   (unsigned long)root->root_key.objectid,
+				   BTRFS_ROOT_TRANS_TAG);
+
+			dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
+			BUG_ON(!dirty);
+			dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
+			BUG_ON(!dirty->root);
+			dirty->latest_root = root;
+			INIT_LIST_HEAD(&dirty->list);
+
+			root->commit_root = btrfs_root_node(root);
+
+			memcpy(dirty->root, root, sizeof(*root));
+			spin_lock_init(&dirty->root->node_lock);
+			spin_lock_init(&dirty->root->list_lock);
+			mutex_init(&dirty->root->objectid_mutex);
+			mutex_init(&dirty->root->log_mutex);
+			INIT_LIST_HEAD(&dirty->root->dead_list);
+			dirty->root->node = root->commit_root;
+			dirty->root->commit_root = NULL;
+
+			spin_lock(&root->list_lock);
+			list_add(&dirty->root->dead_list, &root->dead_list);
+			spin_unlock(&root->list_lock);
+
+			root->dirty_root = dirty;
+		} else {
+			WARN_ON(1);
+		}
+		root->last_trans = running_trans_id;
+	}
+	return 0;
+}
+
+/* wait for commit against the current transaction to become unblocked
+ * when this is done, it is safe to start a new transaction, but the current
+ * transaction might not be fully on disk.
+ */
+static void wait_current_trans(struct btrfs_root *root)
+{
+	struct btrfs_transaction *cur_trans;
+
+	cur_trans = root->fs_info->running_transaction;
+	if (cur_trans && cur_trans->blocked) {
+		DEFINE_WAIT(wait);
+		cur_trans->use_count++;
+		while (1) {
+			prepare_to_wait(&root->fs_info->transaction_wait, &wait,
+					TASK_UNINTERRUPTIBLE);
+			if (cur_trans->blocked) {
+				mutex_unlock(&root->fs_info->trans_mutex);
+				schedule();
+				mutex_lock(&root->fs_info->trans_mutex);
+				finish_wait(&root->fs_info->transaction_wait,
+					    &wait);
+			} else {
+				finish_wait(&root->fs_info->transaction_wait,
+					    &wait);
+				break;
+			}
+		}
+		put_transaction(cur_trans);
+	}
+}
+
+static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
+					     int num_blocks, int wait)
+{
+	struct btrfs_trans_handle *h =
+		kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
+	int ret;
+
+	mutex_lock(&root->fs_info->trans_mutex);
+	if (!root->fs_info->log_root_recovering &&
+	    ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
+		wait_current_trans(root);
+	ret = join_transaction(root);
+	BUG_ON(ret);
+
+	btrfs_record_root_in_trans(root);
+	h->transid = root->fs_info->running_transaction->transid;
+	h->transaction = root->fs_info->running_transaction;
+	h->blocks_reserved = num_blocks;
+	h->blocks_used = 0;
+	h->block_group = 0;
+	h->alloc_exclude_nr = 0;
+	h->alloc_exclude_start = 0;
+	root->fs_info->running_transaction->use_count++;
+	mutex_unlock(&root->fs_info->trans_mutex);
+	return h;
+}
+
+struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
+						   int num_blocks)
+{
+	return start_transaction(root, num_blocks, 1);
+}
+struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
+						   int num_blocks)
+{
+	return start_transaction(root, num_blocks, 0);
+}
+
+struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
+							 int num_blocks)
+{
+	return start_transaction(r, num_blocks, 2);
+}
+
+/* wait for a transaction commit to be fully complete */
+static noinline int wait_for_commit(struct btrfs_root *root,
+				    struct btrfs_transaction *commit)
+{
+	DEFINE_WAIT(wait);
+	mutex_lock(&root->fs_info->trans_mutex);
+	while (!commit->commit_done) {
+		prepare_to_wait(&commit->commit_wait, &wait,
+				TASK_UNINTERRUPTIBLE);
+		if (commit->commit_done)
+			break;
+		mutex_unlock(&root->fs_info->trans_mutex);
+		schedule();
+		mutex_lock(&root->fs_info->trans_mutex);
+	}
+	mutex_unlock(&root->fs_info->trans_mutex);
+	finish_wait(&commit->commit_wait, &wait);
+	return 0;
+}
+
+/*
+ * rate limit against the drop_snapshot code.  This helps to slow down new
+ * operations if the drop_snapshot code isn't able to keep up.
+ */
+static void throttle_on_drops(struct btrfs_root *root)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	int harder_count = 0;
+
+harder:
+	if (atomic_read(&info->throttles)) {
+		DEFINE_WAIT(wait);
+		int thr;
+		thr = atomic_read(&info->throttle_gen);
+
+		do {
+			prepare_to_wait(&info->transaction_throttle,
+					&wait, TASK_UNINTERRUPTIBLE);
+			if (!atomic_read(&info->throttles)) {
+				finish_wait(&info->transaction_throttle, &wait);
+				break;
+			}
+			schedule();
+			finish_wait(&info->transaction_throttle, &wait);
+		} while (thr == atomic_read(&info->throttle_gen));
+		harder_count++;
+
+		if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
+		    harder_count < 2)
+			goto harder;
+
+		if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
+		    harder_count < 10)
+			goto harder;
+
+		if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
+		    harder_count < 20)
+			goto harder;
+	}
+}
+
+void btrfs_throttle(struct btrfs_root *root)
+{
+	mutex_lock(&root->fs_info->trans_mutex);
+	if (!root->fs_info->open_ioctl_trans)
+		wait_current_trans(root);
+	mutex_unlock(&root->fs_info->trans_mutex);
+
+	throttle_on_drops(root);
+}
+
+static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, int throttle)
+{
+	struct btrfs_transaction *cur_trans;
+	struct btrfs_fs_info *info = root->fs_info;
+
+	mutex_lock(&info->trans_mutex);
+	cur_trans = info->running_transaction;
+	WARN_ON(cur_trans != trans->transaction);
+	WARN_ON(cur_trans->num_writers < 1);
+	cur_trans->num_writers--;
+
+	if (waitqueue_active(&cur_trans->writer_wait))
+		wake_up(&cur_trans->writer_wait);
+	put_transaction(cur_trans);
+	mutex_unlock(&info->trans_mutex);
+	memset(trans, 0, sizeof(*trans));
+	kmem_cache_free(btrfs_trans_handle_cachep, trans);
+
+	if (throttle)
+		throttle_on_drops(root);
+
+	return 0;
+}
+
+int btrfs_end_transaction(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root)
+{
+	return __btrfs_end_transaction(trans, root, 0);
+}
+
+int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root)
+{
+	return __btrfs_end_transaction(trans, root, 1);
+}
+
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees.  This is used to make sure all of
+ * those extents are on disk for transaction or log commit
+ */
+int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
+					struct extent_io_tree *dirty_pages)
+{
+	int ret;
+	int err = 0;
+	int werr = 0;
+	struct page *page;
+	struct inode *btree_inode = root->fs_info->btree_inode;
+	u64 start = 0;
+	u64 end;
+	unsigned long index;
+
+	while (1) {
+		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
+					    EXTENT_DIRTY);
+		if (ret)
+			break;
+		while (start <= end) {
+			cond_resched();
+
+			index = start >> PAGE_CACHE_SHIFT;
+			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
+			page = find_get_page(btree_inode->i_mapping, index);
+			if (!page)
+				continue;
+
+			btree_lock_page_hook(page);
+			if (!page->mapping) {
+				unlock_page(page);
+				page_cache_release(page);
+				continue;
+			}
+
+			if (PageWriteback(page)) {
+				if (PageDirty(page))
+					wait_on_page_writeback(page);
+				else {
+					unlock_page(page);
+					page_cache_release(page);
+					continue;
+				}
+			}
+			err = write_one_page(page, 0);
+			if (err)
+				werr = err;
+			page_cache_release(page);
+		}
+	}
+	while (1) {
+		ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
+					    EXTENT_DIRTY);
+		if (ret)
+			break;
+
+		clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
+		while (start <= end) {
+			index = start >> PAGE_CACHE_SHIFT;
+			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
+			page = find_get_page(btree_inode->i_mapping, index);
+			if (!page)
+				continue;
+			if (PageDirty(page)) {
+				btree_lock_page_hook(page);
+				wait_on_page_writeback(page);
+				err = write_one_page(page, 0);
+				if (err)
+					werr = err;
+			}
+			wait_on_page_writeback(page);
+			page_cache_release(page);
+			cond_resched();
+		}
+	}
+	if (err)
+		werr = err;
+	return werr;
+}
+
+int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root)
+{
+	if (!trans || !trans->transaction) {
+		struct inode *btree_inode;
+		btree_inode = root->fs_info->btree_inode;
+		return filemap_write_and_wait(btree_inode->i_mapping);
+	}
+	return btrfs_write_and_wait_marked_extents(root,
+					   &trans->transaction->dirty_pages);
+}
+
+/*
+ * this is used to update the root pointer in the tree of tree roots.
+ *
+ * But, in the case of the extent allocation tree, updating the root
+ * pointer may allocate blocks which may change the root of the extent
+ * allocation tree.
+ *
+ * So, this loops and repeats and makes sure the cowonly root didn't
+ * change while the root pointer was being updated in the metadata.
+ */
+static int update_cowonly_root(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root)
+{
+	int ret;
+	u64 old_root_bytenr;
+	struct btrfs_root *tree_root = root->fs_info->tree_root;
+
+	btrfs_extent_post_op(trans, root);
+	btrfs_write_dirty_block_groups(trans, root);
+	btrfs_extent_post_op(trans, root);
+
+	while (1) {
+		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
+		if (old_root_bytenr == root->node->start)
+			break;
+		btrfs_set_root_bytenr(&root->root_item,
+				       root->node->start);
+		btrfs_set_root_level(&root->root_item,
+				     btrfs_header_level(root->node));
+		btrfs_set_root_generation(&root->root_item, trans->transid);
+
+		btrfs_extent_post_op(trans, root);
+
+		ret = btrfs_update_root(trans, tree_root,
+					&root->root_key,
+					&root->root_item);
+		BUG_ON(ret);
+		btrfs_write_dirty_block_groups(trans, root);
+		btrfs_extent_post_op(trans, root);
+	}
+	return 0;
+}
+
+/*
+ * update all the cowonly tree roots on disk
+ */
+int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root)
+{
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct list_head *next;
+	struct extent_buffer *eb;
+
+	btrfs_extent_post_op(trans, fs_info->tree_root);
+
+	eb = btrfs_lock_root_node(fs_info->tree_root);
+	btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0);
+	btrfs_tree_unlock(eb);
+	free_extent_buffer(eb);
+
+	btrfs_extent_post_op(trans, fs_info->tree_root);
+
+	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
+		next = fs_info->dirty_cowonly_roots.next;
+		list_del_init(next);
+		root = list_entry(next, struct btrfs_root, dirty_list);
+
+		update_cowonly_root(trans, root);
+	}
+	return 0;
+}
+
+/*
+ * dead roots are old snapshots that need to be deleted.  This allocates
+ * a dirty root struct and adds it into the list of dead roots that need to
+ * be deleted
+ */
+int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
+{
+	struct btrfs_dirty_root *dirty;
+
+	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
+	if (!dirty)
+		return -ENOMEM;
+	dirty->root = root;
+	dirty->latest_root = latest;
+
+	mutex_lock(&root->fs_info->trans_mutex);
+	list_add(&dirty->list, &latest->fs_info->dead_roots);
+	mutex_unlock(&root->fs_info->trans_mutex);
+	return 0;
+}
+
+/*
+ * at transaction commit time we need to schedule the old roots for
+ * deletion via btrfs_drop_snapshot.  This runs through all the
+ * reference counted roots that were modified in the current
+ * transaction and puts them into the drop list
+ */
+static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
+				    struct radix_tree_root *radix,
+				    struct list_head *list)
+{
+	struct btrfs_dirty_root *dirty;
+	struct btrfs_root *gang[8];
+	struct btrfs_root *root;
+	int i;
+	int ret;
+	int err = 0;
+	u32 refs;
+
+	while (1) {
+		ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
+						 ARRAY_SIZE(gang),
+						 BTRFS_ROOT_TRANS_TAG);
+		if (ret == 0)
+			break;
+		for (i = 0; i < ret; i++) {
+			root = gang[i];
+			radix_tree_tag_clear(radix,
+				     (unsigned long)root->root_key.objectid,
+				     BTRFS_ROOT_TRANS_TAG);
+
+			BUG_ON(!root->ref_tree);
+			dirty = root->dirty_root;
+
+			btrfs_free_log(trans, root);
+			btrfs_free_reloc_root(trans, root);
+
+			if (root->commit_root == root->node) {
+				WARN_ON(root->node->start !=
+					btrfs_root_bytenr(&root->root_item));
+
+				free_extent_buffer(root->commit_root);
+				root->commit_root = NULL;
+				root->dirty_root = NULL;
+
+				spin_lock(&root->list_lock);
+				list_del_init(&dirty->root->dead_list);
+				spin_unlock(&root->list_lock);
+
+				kfree(dirty->root);
+				kfree(dirty);
+
+				/* make sure to update the root on disk
+				 * so we get any updates to the block used
+				 * counts
+				 */
+				err = btrfs_update_root(trans,
+						root->fs_info->tree_root,
+						&root->root_key,
+						&root->root_item);
+				continue;
+			}
+
+			memset(&root->root_item.drop_progress, 0,
+			       sizeof(struct btrfs_disk_key));
+			root->root_item.drop_level = 0;
+			root->commit_root = NULL;
+			root->dirty_root = NULL;
+			root->root_key.offset = root->fs_info->generation;
+			btrfs_set_root_bytenr(&root->root_item,
+					      root->node->start);
+			btrfs_set_root_level(&root->root_item,
+					     btrfs_header_level(root->node));
+			btrfs_set_root_generation(&root->root_item,
+						  root->root_key.offset);
+
+			err = btrfs_insert_root(trans, root->fs_info->tree_root,
+						&root->root_key,
+						&root->root_item);
+			if (err)
+				break;
+
+			refs = btrfs_root_refs(&dirty->root->root_item);
+			btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
+			err = btrfs_update_root(trans, root->fs_info->tree_root,
+						&dirty->root->root_key,
+						&dirty->root->root_item);
+
+			BUG_ON(err);
+			if (refs == 1) {
+				list_add(&dirty->list, list);
+			} else {
+				WARN_ON(1);
+				free_extent_buffer(dirty->root->node);
+				kfree(dirty->root);
+				kfree(dirty);
+			}
+		}
+	}
+	return err;
+}
+
+/*
+ * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
+ * otherwise every leaf in the btree is read and defragged.
+ */
+int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	int ret;
+	struct btrfs_trans_handle *trans;
+	unsigned long nr;
+
+	smp_mb();
+	if (root->defrag_running)
+		return 0;
+	trans = btrfs_start_transaction(root, 1);
+	while (1) {
+		root->defrag_running = 1;
+		ret = btrfs_defrag_leaves(trans, root, cacheonly);
+		nr = trans->blocks_used;
+		btrfs_end_transaction(trans, root);
+		btrfs_btree_balance_dirty(info->tree_root, nr);
+		cond_resched();
+
+		trans = btrfs_start_transaction(root, 1);
+		if (root->fs_info->closing || ret != -EAGAIN)
+			break;
+	}
+	root->defrag_running = 0;
+	smp_mb();
+	btrfs_end_transaction(trans, root);
+	return 0;
+}
+
+/*
+ * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
+ * all of them
+ */
+static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
+				     struct list_head *list)
+{
+	struct btrfs_dirty_root *dirty;
+	struct btrfs_trans_handle *trans;
+	unsigned long nr;
+	u64 num_bytes;
+	u64 bytes_used;
+	u64 max_useless;
+	int ret = 0;
+	int err;
+
+	while (!list_empty(list)) {
+		struct btrfs_root *root;
+
+		dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
+		list_del_init(&dirty->list);
+
+		num_bytes = btrfs_root_used(&dirty->root->root_item);
+		root = dirty->latest_root;
+		atomic_inc(&root->fs_info->throttles);
+
+		while (1) {
+			trans = btrfs_start_transaction(tree_root, 1);
+			mutex_lock(&root->fs_info->drop_mutex);
+			ret = btrfs_drop_snapshot(trans, dirty->root);
+			if (ret != -EAGAIN)
+				break;
+			mutex_unlock(&root->fs_info->drop_mutex);
+
+			err = btrfs_update_root(trans,
+					tree_root,
+					&dirty->root->root_key,
+					&dirty->root->root_item);
+			if (err)
+				ret = err;
+			nr = trans->blocks_used;
+			ret = btrfs_end_transaction(trans, tree_root);
+			BUG_ON(ret);
+
+			btrfs_btree_balance_dirty(tree_root, nr);
+			cond_resched();
+		}
+		BUG_ON(ret);
+		atomic_dec(&root->fs_info->throttles);
+		wake_up(&root->fs_info->transaction_throttle);
+
+		num_bytes -= btrfs_root_used(&dirty->root->root_item);
+		bytes_used = btrfs_root_used(&root->root_item);
+		if (num_bytes) {
+			btrfs_record_root_in_trans(root);
+			btrfs_set_root_used(&root->root_item,
+					    bytes_used - num_bytes);
+		}
+
+		ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
+		if (ret) {
+			BUG();
+			break;
+		}
+		mutex_unlock(&root->fs_info->drop_mutex);
+
+		spin_lock(&root->list_lock);
+		list_del_init(&dirty->root->dead_list);
+		if (!list_empty(&root->dead_list)) {
+			struct btrfs_root *oldest;
+			oldest = list_entry(root->dead_list.prev,
+					    struct btrfs_root, dead_list);
+			max_useless = oldest->root_key.offset - 1;
+		} else {
+			max_useless = root->root_key.offset - 1;
+		}
+		spin_unlock(&root->list_lock);
+
+		nr = trans->blocks_used;
+		ret = btrfs_end_transaction(trans, tree_root);
+		BUG_ON(ret);
+
+		ret = btrfs_remove_leaf_refs(root, max_useless, 0);
+		BUG_ON(ret);
+
+		free_extent_buffer(dirty->root->node);
+		kfree(dirty->root);
+		kfree(dirty);
+
+		btrfs_btree_balance_dirty(tree_root, nr);
+		cond_resched();
+	}
+	return ret;
+}
+
+/*
+ * new snapshots need to be created at a very specific time in the
+ * transaction commit.  This does the actual creation
+ */
+static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
+				   struct btrfs_fs_info *fs_info,
+				   struct btrfs_pending_snapshot *pending)
+{
+	struct btrfs_key key;
+	struct btrfs_root_item *new_root_item;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_root *root = pending->root;
+	struct extent_buffer *tmp;
+	struct extent_buffer *old;
+	int ret;
+	u64 objectid;
+
+	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
+	if (!new_root_item) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+	ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
+	if (ret)
+		goto fail;
+
+	btrfs_record_root_in_trans(root);
+	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
+	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
+
+	key.objectid = objectid;
+	key.offset = trans->transid;
+	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+
+	old = btrfs_lock_root_node(root);
+	btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
+
+	btrfs_copy_root(trans, root, old, &tmp, objectid);
+	btrfs_tree_unlock(old);
+	free_extent_buffer(old);
+
+	btrfs_set_root_bytenr(new_root_item, tmp->start);
+	btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
+	btrfs_set_root_generation(new_root_item, trans->transid);
+	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
+				new_root_item);
+	btrfs_tree_unlock(tmp);
+	free_extent_buffer(tmp);
+	if (ret)
+		goto fail;
+
+	key.offset = (u64)-1;
+	memcpy(&pending->root_key, &key, sizeof(key));
+fail:
+	kfree(new_root_item);
+	return ret;
+}
+
+static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
+				   struct btrfs_pending_snapshot *pending)
+{
+	int ret;
+	int namelen;
+	u64 index = 0;
+	struct btrfs_trans_handle *trans;
+	struct inode *parent_inode;
+	struct inode *inode;
+	struct btrfs_root *parent_root;
+
+	parent_inode = pending->dentry->d_parent->d_inode;
+	parent_root = BTRFS_I(parent_inode)->root;
+	trans = btrfs_join_transaction(parent_root, 1);
+
+	/*
+	 * insert the directory item
+	 */
+	namelen = strlen(pending->name);
+	ret = btrfs_set_inode_index(parent_inode, &index);
+	ret = btrfs_insert_dir_item(trans, parent_root,
+			    pending->name, namelen,
+			    parent_inode->i_ino,
+			    &pending->root_key, BTRFS_FT_DIR, index);
+
+	if (ret)
+		goto fail;
+
+	btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
+	ret = btrfs_update_inode(trans, parent_root, parent_inode);
+	BUG_ON(ret);
+
+	/* add the backref first */
+	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
+				 pending->root_key.objectid,
+				 BTRFS_ROOT_BACKREF_KEY,
+				 parent_root->root_key.objectid,
+				 parent_inode->i_ino, index, pending->name,
+				 namelen);
+
+	BUG_ON(ret);
+
+	/* now add the forward ref */
+	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
+				 parent_root->root_key.objectid,
+				 BTRFS_ROOT_REF_KEY,
+				 pending->root_key.objectid,
+				 parent_inode->i_ino, index, pending->name,
+				 namelen);
+
+	inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
+	d_instantiate(pending->dentry, inode);
+fail:
+	btrfs_end_transaction(trans, fs_info->fs_root);
+	return ret;
+}
+
+/*
+ * create all the snapshots we've scheduled for creation
+ */
+static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
+					     struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_pending_snapshot *pending;
+	struct list_head *head = &trans->transaction->pending_snapshots;
+	struct list_head *cur;
+	int ret;
+
+	list_for_each(cur, head) {
+		pending = list_entry(cur, struct btrfs_pending_snapshot, list);
+		ret = create_pending_snapshot(trans, fs_info, pending);
+		BUG_ON(ret);
+	}
+	return 0;
+}
+
+static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
+					     struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_pending_snapshot *pending;
+	struct list_head *head = &trans->transaction->pending_snapshots;
+	int ret;
+
+	while (!list_empty(head)) {
+		pending = list_entry(head->next,
+				     struct btrfs_pending_snapshot, list);
+		ret = finish_pending_snapshot(fs_info, pending);
+		BUG_ON(ret);
+		list_del(&pending->list);
+		kfree(pending->name);
+		kfree(pending);
+	}
+	return 0;
+}
+
+int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root)
+{
+	unsigned long joined = 0;
+	unsigned long timeout = 1;
+	struct btrfs_transaction *cur_trans;
+	struct btrfs_transaction *prev_trans = NULL;
+	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
+	struct list_head dirty_fs_roots;
+	struct extent_io_tree *pinned_copy;
+	DEFINE_WAIT(wait);
+	int ret;
+
+	INIT_LIST_HEAD(&dirty_fs_roots);
+	mutex_lock(&root->fs_info->trans_mutex);
+	if (trans->transaction->in_commit) {
+		cur_trans = trans->transaction;
+		trans->transaction->use_count++;
+		mutex_unlock(&root->fs_info->trans_mutex);
+		btrfs_end_transaction(trans, root);
+
+		ret = wait_for_commit(root, cur_trans);
+		BUG_ON(ret);
+
+		mutex_lock(&root->fs_info->trans_mutex);
+		put_transaction(cur_trans);
+		mutex_unlock(&root->fs_info->trans_mutex);
+
+		return 0;
+	}
+
+	pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
+	if (!pinned_copy)
+		return -ENOMEM;
+
+	extent_io_tree_init(pinned_copy,
+			     root->fs_info->btree_inode->i_mapping, GFP_NOFS);
+
+	trans->transaction->in_commit = 1;
+	trans->transaction->blocked = 1;
+	cur_trans = trans->transaction;
+	if (cur_trans->list.prev != &root->fs_info->trans_list) {
+		prev_trans = list_entry(cur_trans->list.prev,
+					struct btrfs_transaction, list);
+		if (!prev_trans->commit_done) {
+			prev_trans->use_count++;
+			mutex_unlock(&root->fs_info->trans_mutex);
+
+			wait_for_commit(root, prev_trans);
+
+			mutex_lock(&root->fs_info->trans_mutex);
+			put_transaction(prev_trans);
+		}
+	}
+
+	do {
+		int snap_pending = 0;
+		joined = cur_trans->num_joined;
+		if (!list_empty(&trans->transaction->pending_snapshots))
+			snap_pending = 1;
+
+		WARN_ON(cur_trans != trans->transaction);
+		prepare_to_wait(&cur_trans->writer_wait, &wait,
+				TASK_UNINTERRUPTIBLE);
+
+		if (cur_trans->num_writers > 1)
+			timeout = MAX_SCHEDULE_TIMEOUT;
+		else
+			timeout = 1;
+
+		mutex_unlock(&root->fs_info->trans_mutex);
+
+		if (snap_pending) {
+			ret = btrfs_wait_ordered_extents(root, 1);
+			BUG_ON(ret);
+		}
+
+		schedule_timeout(timeout);
+
+		mutex_lock(&root->fs_info->trans_mutex);
+		finish_wait(&cur_trans->writer_wait, &wait);
+	} while (cur_trans->num_writers > 1 ||
+		 (cur_trans->num_joined != joined));
+
+	ret = create_pending_snapshots(trans, root->fs_info);
+	BUG_ON(ret);
+
+	WARN_ON(cur_trans != trans->transaction);
+
+	/* btrfs_commit_tree_roots is responsible for getting the
+	 * various roots consistent with each other.  Every pointer
+	 * in the tree of tree roots has to point to the most up to date
+	 * root for every subvolume and other tree.  So, we have to keep
+	 * the tree logging code from jumping in and changing any
+	 * of the trees.
+	 *
+	 * At this point in the commit, there can't be any tree-log
+	 * writers, but a little lower down we drop the trans mutex
+	 * and let new people in.  By holding the tree_log_mutex
+	 * from now until after the super is written, we avoid races
+	 * with the tree-log code.
+	 */
+	mutex_lock(&root->fs_info->tree_log_mutex);
+	/*
+	 * keep tree reloc code from adding new reloc trees
+	 */
+	mutex_lock(&root->fs_info->tree_reloc_mutex);
+
+
+	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
+			      &dirty_fs_roots);
+	BUG_ON(ret);
+
+	/* add_dirty_roots gets rid of all the tree log roots, it is now
+	 * safe to free the root of tree log roots
+	 */
+	btrfs_free_log_root_tree(trans, root->fs_info);
+
+	ret = btrfs_commit_tree_roots(trans, root);
+	BUG_ON(ret);
+
+	cur_trans = root->fs_info->running_transaction;
+	spin_lock(&root->fs_info->new_trans_lock);
+	root->fs_info->running_transaction = NULL;
+	spin_unlock(&root->fs_info->new_trans_lock);
+	btrfs_set_super_generation(&root->fs_info->super_copy,
+				   cur_trans->transid);
+	btrfs_set_super_root(&root->fs_info->super_copy,
+			     root->fs_info->tree_root->node->start);
+	btrfs_set_super_root_level(&root->fs_info->super_copy,
+			   btrfs_header_level(root->fs_info->tree_root->node));
+
+	btrfs_set_super_chunk_root(&root->fs_info->super_copy,
+				   chunk_root->node->start);
+	btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
+					 btrfs_header_level(chunk_root->node));
+	btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
+				btrfs_header_generation(chunk_root->node));
+
+	if (!root->fs_info->log_root_recovering) {
+		btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
+		btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
+	}
+
+	memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
+	       sizeof(root->fs_info->super_copy));
+
+	btrfs_copy_pinned(root, pinned_copy);
+
+	trans->transaction->blocked = 0;
+	wake_up(&root->fs_info->transaction_throttle);
+	wake_up(&root->fs_info->transaction_wait);
+
+	mutex_unlock(&root->fs_info->trans_mutex);
+	ret = btrfs_write_and_wait_transaction(trans, root);
+	BUG_ON(ret);
+	write_ctree_super(trans, root, 0);
+
+	/*
+	 * the super is written, we can safely allow the tree-loggers
+	 * to go about their business
+	 */
+	mutex_unlock(&root->fs_info->tree_log_mutex);
+
+	btrfs_finish_extent_commit(trans, root, pinned_copy);
+	kfree(pinned_copy);
+
+	btrfs_drop_dead_reloc_roots(root);
+	mutex_unlock(&root->fs_info->tree_reloc_mutex);
+
+	/* do the directory inserts of any pending snapshot creations */
+	finish_pending_snapshots(trans, root->fs_info);
+
+	mutex_lock(&root->fs_info->trans_mutex);
+
+	cur_trans->commit_done = 1;
+	root->fs_info->last_trans_committed = cur_trans->transid;
+	wake_up(&cur_trans->commit_wait);
+
+	put_transaction(cur_trans);
+	put_transaction(cur_trans);
+
+	list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
+	if (root->fs_info->closing)
+		list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
+
+	mutex_unlock(&root->fs_info->trans_mutex);
+
+	kmem_cache_free(btrfs_trans_handle_cachep, trans);
+
+	if (root->fs_info->closing)
+		drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
+	return ret;
+}
+
+/*
+ * interface function to delete all the snapshots we have scheduled for deletion
+ */
+int btrfs_clean_old_snapshots(struct btrfs_root *root)
+{
+	struct list_head dirty_roots;
+	INIT_LIST_HEAD(&dirty_roots);
+again:
+	mutex_lock(&root->fs_info->trans_mutex);
+	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
+	mutex_unlock(&root->fs_info->trans_mutex);
+
+	if (!list_empty(&dirty_roots)) {
+		drop_dirty_roots(root, &dirty_roots);
+		goto again;
+	}
+	return 0;
+}
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
new file mode 100644
index 0000000..ea29211
--- /dev/null
+++ b/fs/btrfs/transaction.h
@@ -0,0 +1,106 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_TRANSACTION__
+#define __BTRFS_TRANSACTION__
+#include "btrfs_inode.h"
+
+struct btrfs_transaction {
+	u64 transid;
+	unsigned long num_writers;
+	unsigned long num_joined;
+	int in_commit;
+	int use_count;
+	int commit_done;
+	int blocked;
+	struct list_head list;
+	struct extent_io_tree dirty_pages;
+	unsigned long start_time;
+	wait_queue_head_t writer_wait;
+	wait_queue_head_t commit_wait;
+	struct list_head pending_snapshots;
+};
+
+struct btrfs_trans_handle {
+	u64 transid;
+	unsigned long blocks_reserved;
+	unsigned long blocks_used;
+	struct btrfs_transaction *transaction;
+	u64 block_group;
+	u64 alloc_exclude_start;
+	u64 alloc_exclude_nr;
+};
+
+struct btrfs_pending_snapshot {
+	struct dentry *dentry;
+	struct btrfs_root *root;
+	char *name;
+	struct btrfs_key root_key;
+	struct list_head list;
+};
+
+struct btrfs_dirty_root {
+	struct list_head list;
+	struct btrfs_root *root;
+	struct btrfs_root *latest_root;
+};
+
+static inline void btrfs_set_trans_block_group(struct btrfs_trans_handle *trans,
+					       struct inode *inode)
+{
+	trans->block_group = BTRFS_I(inode)->block_group;
+}
+
+static inline void btrfs_update_inode_block_group(
+					  struct btrfs_trans_handle *trans,
+					  struct inode *inode)
+{
+	BTRFS_I(inode)->block_group = trans->block_group;
+}
+
+static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans,
+					      struct inode *inode)
+{
+	BTRFS_I(inode)->last_trans = trans->transaction->transid;
+}
+
+int btrfs_end_transaction(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root);
+struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
+						   int num_blocks);
+struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
+						   int num_blocks);
+struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
+						   int num_blocks);
+int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root);
+int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root);
+
+int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest);
+int btrfs_defrag_root(struct btrfs_root *root, int cacheonly);
+int btrfs_clean_old_snapshots(struct btrfs_root *root);
+int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root);
+int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root);
+void btrfs_throttle(struct btrfs_root *root);
+int btrfs_record_root_in_trans(struct btrfs_root *root);
+int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
+					struct extent_io_tree *dirty_pages);
+#endif
-- 
1.6.0.2

[PATCH 05/35] Btrfs btree leaf reference cache

[Chris Mason]

Snapshot deletion requires walking down the btree and changing reference
counts of leaves and nodes.  This caches the extents pointed to by a leaf
so that we can avoid reading that leaf during the snapshot deletion.

Many leaves don't point to any extents at all, so this is a huge win.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/ref-cache.c b/fs/btrfs/ref-cache.c
new file mode 100644
index 0000000..6f0acc4
--- /dev/null
+++ b/fs/btrfs/ref-cache.c
@@ -0,0 +1,230 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+#include "ref-cache.h"
+#include "transaction.h"
+
+/*
+ * leaf refs are used to cache the information about which extents
+ * a given leaf has references on.  This allows us to process that leaf
+ * in btrfs_drop_snapshot without needing to read it back from disk.
+ */
+
+/*
+ * kmalloc a leaf reference struct and update the counters for the
+ * total ref cache size
+ */
+struct btrfs_leaf_ref *btrfs_alloc_leaf_ref(struct btrfs_root *root,
+					    int nr_extents)
+{
+	struct btrfs_leaf_ref *ref;
+	size_t size = btrfs_leaf_ref_size(nr_extents);
+
+	ref = kmalloc(size, GFP_NOFS);
+	if (ref) {
+		spin_lock(&root->fs_info->ref_cache_lock);
+		root->fs_info->total_ref_cache_size += size;
+		spin_unlock(&root->fs_info->ref_cache_lock);
+
+		memset(ref, 0, sizeof(*ref));
+		atomic_set(&ref->usage, 1);
+		INIT_LIST_HEAD(&ref->list);
+	}
+	return ref;
+}
+
+/*
+ * free a leaf reference struct and update the counters for the
+ * total ref cache size
+ */
+void btrfs_free_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref)
+{
+	if (!ref)
+		return;
+	WARN_ON(atomic_read(&ref->usage) == 0);
+	if (atomic_dec_and_test(&ref->usage)) {
+		size_t size = btrfs_leaf_ref_size(ref->nritems);
+
+		BUG_ON(ref->in_tree);
+		kfree(ref);
+
+		spin_lock(&root->fs_info->ref_cache_lock);
+		root->fs_info->total_ref_cache_size -= size;
+		spin_unlock(&root->fs_info->ref_cache_lock);
+	}
+}
+
+static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
+				   struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct btrfs_leaf_ref *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct btrfs_leaf_ref, rb_node);
+
+		if (bytenr < entry->bytenr)
+			p = &(*p)->rb_left;
+		else if (bytenr > entry->bytenr)
+			p = &(*p)->rb_right;
+		else
+			return parent;
+	}
+
+	entry = rb_entry(node, struct btrfs_leaf_ref, rb_node);
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
+{
+	struct rb_node *n = root->rb_node;
+	struct btrfs_leaf_ref *entry;
+
+	while (n) {
+		entry = rb_entry(n, struct btrfs_leaf_ref, rb_node);
+		WARN_ON(!entry->in_tree);
+
+		if (bytenr < entry->bytenr)
+			n = n->rb_left;
+		else if (bytenr > entry->bytenr)
+			n = n->rb_right;
+		else
+			return n;
+	}
+	return NULL;
+}
+
+int btrfs_remove_leaf_refs(struct btrfs_root *root, u64 max_root_gen,
+			   int shared)
+{
+	struct btrfs_leaf_ref *ref = NULL;
+	struct btrfs_leaf_ref_tree *tree = root->ref_tree;
+
+	if (shared)
+		tree = &root->fs_info->shared_ref_tree;
+	if (!tree)
+		return 0;
+
+	spin_lock(&tree->lock);
+	while (!list_empty(&tree->list)) {
+		ref = list_entry(tree->list.next, struct btrfs_leaf_ref, list);
+		BUG_ON(ref->tree != tree);
+		if (ref->root_gen > max_root_gen)
+			break;
+		if (!xchg(&ref->in_tree, 0)) {
+			cond_resched_lock(&tree->lock);
+			continue;
+		}
+
+		rb_erase(&ref->rb_node, &tree->root);
+		list_del_init(&ref->list);
+
+		spin_unlock(&tree->lock);
+		btrfs_free_leaf_ref(root, ref);
+		cond_resched();
+		spin_lock(&tree->lock);
+	}
+	spin_unlock(&tree->lock);
+	return 0;
+}
+
+/*
+ * find the leaf ref for a given extent.  This returns the ref struct with
+ * a usage reference incremented
+ */
+struct btrfs_leaf_ref *btrfs_lookup_leaf_ref(struct btrfs_root *root,
+					     u64 bytenr)
+{
+	struct rb_node *rb;
+	struct btrfs_leaf_ref *ref = NULL;
+	struct btrfs_leaf_ref_tree *tree = root->ref_tree;
+again:
+	if (tree) {
+		spin_lock(&tree->lock);
+		rb = tree_search(&tree->root, bytenr);
+		if (rb)
+			ref = rb_entry(rb, struct btrfs_leaf_ref, rb_node);
+		if (ref)
+			atomic_inc(&ref->usage);
+		spin_unlock(&tree->lock);
+		if (ref)
+			return ref;
+	}
+	if (tree != &root->fs_info->shared_ref_tree) {
+		tree = &root->fs_info->shared_ref_tree;
+		goto again;
+	}
+	return NULL;
+}
+
+/*
+ * add a fully filled in leaf ref struct
+ * remove all the refs older than a given root generation
+ */
+int btrfs_add_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref,
+		       int shared)
+{
+	int ret = 0;
+	struct rb_node *rb;
+	struct btrfs_leaf_ref_tree *tree = root->ref_tree;
+
+	if (shared)
+		tree = &root->fs_info->shared_ref_tree;
+
+	spin_lock(&tree->lock);
+	rb = tree_insert(&tree->root, ref->bytenr, &ref->rb_node);
+	if (rb) {
+		ret = -EEXIST;
+	} else {
+		atomic_inc(&ref->usage);
+		ref->tree = tree;
+		ref->in_tree = 1;
+		list_add_tail(&ref->list, &tree->list);
+	}
+	spin_unlock(&tree->lock);
+	return ret;
+}
+
+/*
+ * remove a single leaf ref from the tree.  This drops the ref held by the tree
+ * only
+ */
+int btrfs_remove_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref)
+{
+	struct btrfs_leaf_ref_tree *tree;
+
+	if (!xchg(&ref->in_tree, 0))
+		return 0;
+
+	tree = ref->tree;
+	spin_lock(&tree->lock);
+
+	rb_erase(&ref->rb_node, &tree->root);
+	list_del_init(&ref->list);
+
+	spin_unlock(&tree->lock);
+
+	btrfs_free_leaf_ref(root, ref);
+	return 0;
+}
diff --git a/fs/btrfs/ref-cache.h b/fs/btrfs/ref-cache.h
new file mode 100644
index 0000000..16f3183
--- /dev/null
+++ b/fs/btrfs/ref-cache.h
@@ -0,0 +1,77 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+#ifndef __REFCACHE__
+#define __REFCACHE__
+
+struct btrfs_extent_info {
+	/* bytenr and num_bytes find the extent in the extent allocation tree */
+	u64 bytenr;
+	u64 num_bytes;
+
+	/* objectid and offset find the back reference for the file */
+	u64 objectid;
+	u64 offset;
+};
+
+struct btrfs_leaf_ref {
+	struct rb_node rb_node;
+	struct btrfs_leaf_ref_tree *tree;
+	int in_tree;
+	atomic_t usage;
+
+	u64 root_gen;
+	u64 bytenr;
+	u64 owner;
+	u64 generation;
+	int nritems;
+
+	struct list_head list;
+	struct btrfs_extent_info extents[];
+};
+
+static inline size_t btrfs_leaf_ref_size(int nr_extents)
+{
+	return sizeof(struct btrfs_leaf_ref) +
+	       sizeof(struct btrfs_extent_info) * nr_extents;
+}
+
+static inline void btrfs_leaf_ref_tree_init(struct btrfs_leaf_ref_tree *tree)
+{
+	tree->root.rb_node = NULL;
+	INIT_LIST_HEAD(&tree->list);
+	spin_lock_init(&tree->lock);
+}
+
+static inline int btrfs_leaf_ref_tree_empty(struct btrfs_leaf_ref_tree *tree)
+{
+	return RB_EMPTY_ROOT(&tree->root);
+}
+
+void btrfs_leaf_ref_tree_init(struct btrfs_leaf_ref_tree *tree);
+struct btrfs_leaf_ref *btrfs_alloc_leaf_ref(struct btrfs_root *root,
+					    int nr_extents);
+void btrfs_free_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref);
+struct btrfs_leaf_ref *btrfs_lookup_leaf_ref(struct btrfs_root *root,
+					     u64 bytenr);
+int btrfs_add_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref,
+		       int shared);
+int btrfs_remove_leaf_refs(struct btrfs_root *root, u64 max_root_gen,
+			   int shared);
+int btrfs_remove_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref);
+
+#endif
-- 
1.6.0.2

[PATCH 06/35] Btrfs tree printk debugging

[Chris Mason]

When things go horribly wrong, a few places in btrfs print out the btree
leaf to help debugging.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/print-tree.c b/fs/btrfs/print-tree.c
new file mode 100644
index 0000000..5f8f218
--- /dev/null
+++ b/fs/btrfs/print-tree.c
@@ -0,0 +1,216 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+
+static void print_chunk(struct extent_buffer *eb, struct btrfs_chunk *chunk)
+{
+	int num_stripes = btrfs_chunk_num_stripes(eb, chunk);
+	int i;
+	printk(KERN_INFO "\t\tchunk length %llu owner %llu type %llu "
+	       "num_stripes %d\n",
+	       (unsigned long long)btrfs_chunk_length(eb, chunk),
+	       (unsigned long long)btrfs_chunk_owner(eb, chunk),
+	       (unsigned long long)btrfs_chunk_type(eb, chunk),
+	       num_stripes);
+	for (i = 0 ; i < num_stripes ; i++) {
+		printk(KERN_INFO "\t\t\tstripe %d devid %llu offset %llu\n", i,
+		      (unsigned long long)btrfs_stripe_devid_nr(eb, chunk, i),
+		      (unsigned long long)btrfs_stripe_offset_nr(eb, chunk, i));
+	}
+}
+static void print_dev_item(struct extent_buffer *eb,
+			   struct btrfs_dev_item *dev_item)
+{
+	printk(KERN_INFO "\t\tdev item devid %llu "
+	       "total_bytes %llu bytes used %llu\n",
+	       (unsigned long long)btrfs_device_id(eb, dev_item),
+	       (unsigned long long)btrfs_device_total_bytes(eb, dev_item),
+	       (unsigned long long)btrfs_device_bytes_used(eb, dev_item));
+}
+void btrfs_print_leaf(struct btrfs_root *root, struct extent_buffer *l)
+{
+	int i;
+	u32 nr = btrfs_header_nritems(l);
+	struct btrfs_item *item;
+	struct btrfs_extent_item *ei;
+	struct btrfs_root_item *ri;
+	struct btrfs_dir_item *di;
+	struct btrfs_inode_item *ii;
+	struct btrfs_block_group_item *bi;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct btrfs_extent_ref *ref;
+	struct btrfs_dev_extent *dev_extent;
+	u32 type;
+
+	printk(KERN_INFO "leaf %llu total ptrs %d free space %d\n",
+		(unsigned long long)btrfs_header_bytenr(l), nr,
+		btrfs_leaf_free_space(root, l));
+	for (i = 0 ; i < nr ; i++) {
+		item = btrfs_item_nr(l, i);
+		btrfs_item_key_to_cpu(l, &key, i);
+		type = btrfs_key_type(&key);
+		printk(KERN_INFO "\titem %d key (%llu %x %llu) itemoff %d "
+		       "itemsize %d\n",
+			i,
+			(unsigned long long)key.objectid, type,
+			(unsigned long long)key.offset,
+			btrfs_item_offset(l, item), btrfs_item_size(l, item));
+		switch (type) {
+		case BTRFS_INODE_ITEM_KEY:
+			ii = btrfs_item_ptr(l, i, struct btrfs_inode_item);
+			printk(KERN_INFO "\t\tinode generation %llu size %llu "
+			       "mode %o\n",
+			       (unsigned long long)
+			       btrfs_inode_generation(l, ii),
+			      (unsigned long long)btrfs_inode_size(l, ii),
+			       btrfs_inode_mode(l, ii));
+			break;
+		case BTRFS_DIR_ITEM_KEY:
+			di = btrfs_item_ptr(l, i, struct btrfs_dir_item);
+			btrfs_dir_item_key_to_cpu(l, di, &found_key);
+			printk(KERN_INFO "\t\tdir oid %llu type %u\n",
+				(unsigned long long)found_key.objectid,
+				btrfs_dir_type(l, di));
+			break;
+		case BTRFS_ROOT_ITEM_KEY:
+			ri = btrfs_item_ptr(l, i, struct btrfs_root_item);
+			printk(KERN_INFO "\t\troot data bytenr %llu refs %u\n",
+				(unsigned long long)
+				btrfs_disk_root_bytenr(l, ri),
+				btrfs_disk_root_refs(l, ri));
+			break;
+		case BTRFS_EXTENT_ITEM_KEY:
+			ei = btrfs_item_ptr(l, i, struct btrfs_extent_item);
+			printk(KERN_INFO "\t\textent data refs %u\n",
+				btrfs_extent_refs(l, ei));
+			break;
+		case BTRFS_EXTENT_REF_KEY:
+			ref = btrfs_item_ptr(l, i, struct btrfs_extent_ref);
+			printk(KERN_INFO "\t\textent back ref root %llu "
+			       "gen %llu owner %llu num_refs %lu\n",
+			       (unsigned long long)btrfs_ref_root(l, ref),
+			       (unsigned long long)btrfs_ref_generation(l, ref),
+			       (unsigned long long)btrfs_ref_objectid(l, ref),
+			       (unsigned long)btrfs_ref_num_refs(l, ref));
+			break;
+
+		case BTRFS_EXTENT_DATA_KEY:
+			fi = btrfs_item_ptr(l, i,
+					    struct btrfs_file_extent_item);
+			if (btrfs_file_extent_type(l, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE) {
+				printk(KERN_INFO "\t\tinline extent data "
+				       "size %u\n",
+				       btrfs_file_extent_inline_len(l, fi));
+				break;
+			}
+			printk(KERN_INFO "\t\textent data disk bytenr %llu "
+			       "nr %llu\n",
+			       (unsigned long long)
+			       btrfs_file_extent_disk_bytenr(l, fi),
+			       (unsigned long long)
+			       btrfs_file_extent_disk_num_bytes(l, fi));
+			printk(KERN_INFO "\t\textent data offset %llu "
+			       "nr %llu ram %llu\n",
+			       (unsigned long long)
+			       btrfs_file_extent_offset(l, fi),
+			       (unsigned long long)
+			       btrfs_file_extent_num_bytes(l, fi),
+			       (unsigned long long)
+			       btrfs_file_extent_ram_bytes(l, fi));
+			break;
+		case BTRFS_BLOCK_GROUP_ITEM_KEY:
+			bi = btrfs_item_ptr(l, i,
+					    struct btrfs_block_group_item);
+			printk(KERN_INFO "\t\tblock group used %llu\n",
+			       (unsigned long long)
+			       btrfs_disk_block_group_used(l, bi));
+			break;
+		case BTRFS_CHUNK_ITEM_KEY:
+			print_chunk(l, btrfs_item_ptr(l, i,
+						      struct btrfs_chunk));
+			break;
+		case BTRFS_DEV_ITEM_KEY:
+			print_dev_item(l, btrfs_item_ptr(l, i,
+					struct btrfs_dev_item));
+			break;
+		case BTRFS_DEV_EXTENT_KEY:
+			dev_extent = btrfs_item_ptr(l, i,
+						    struct btrfs_dev_extent);
+			printk(KERN_INFO "\t\tdev extent chunk_tree %llu\n"
+			       "\t\tchunk objectid %llu chunk offset %llu "
+			       "length %llu\n",
+			       (unsigned long long)
+			       btrfs_dev_extent_chunk_tree(l, dev_extent),
+			       (unsigned long long)
+			       btrfs_dev_extent_chunk_objectid(l, dev_extent),
+			       (unsigned long long)
+			       btrfs_dev_extent_chunk_offset(l, dev_extent),
+			       (unsigned long long)
+			       btrfs_dev_extent_length(l, dev_extent));
+		};
+	}
+}
+
+void btrfs_print_tree(struct btrfs_root *root, struct extent_buffer *c)
+{
+	int i; u32 nr;
+	struct btrfs_key key;
+	int level;
+
+	if (!c)
+		return;
+	nr = btrfs_header_nritems(c);
+	level = btrfs_header_level(c);
+	if (level == 0) {
+		btrfs_print_leaf(root, c);
+		return;
+	}
+	printk(KERN_INFO "node %llu level %d total ptrs %d free spc %u\n",
+	       (unsigned long long)btrfs_header_bytenr(c),
+	       btrfs_header_level(c), nr,
+	       (u32)BTRFS_NODEPTRS_PER_BLOCK(root) - nr);
+	for (i = 0; i < nr; i++) {
+		btrfs_node_key_to_cpu(c, &key, i);
+		printk(KERN_INFO "\tkey %d (%llu %u %llu) block %llu\n",
+		       i,
+		       (unsigned long long)key.objectid,
+		       key.type,
+		       (unsigned long long)key.offset,
+		       (unsigned long long)btrfs_node_blockptr(c, i));
+	}
+	for (i = 0; i < nr; i++) {
+		struct extent_buffer *next = read_tree_block(root,
+					btrfs_node_blockptr(c, i),
+					btrfs_level_size(root, level - 1),
+					btrfs_node_ptr_generation(c, i));
+		if (btrfs_is_leaf(next) &&
+		    btrfs_header_level(c) != 1)
+			BUG();
+		if (btrfs_header_level(next) !=
+			btrfs_header_level(c) - 1)
+			BUG();
+		btrfs_print_tree(root, next);
+		free_extent_buffer(next);
+	}
+}
diff --git a/fs/btrfs/print-tree.h b/fs/btrfs/print-tree.h
new file mode 100644
index 0000000..da75efe
--- /dev/null
+++ b/fs/btrfs/print-tree.h
@@ -0,0 +1,23 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __PRINT_TREE_
+#define __PRINT_TREE_
+void btrfs_print_leaf(struct btrfs_root *root, struct extent_buffer *l);
+void btrfs_print_tree(struct btrfs_root *root, struct extent_buffer *t);
+#endif
-- 
1.6.0.2

[PATCH 07/35] Btrfs ordered data processing

[Chris Mason]

Ordered data mode in btrfs is not optional like in ext3/4, and it works
quite differently.  Metadata for a file is not updated until after the
new data blocks are actually on disk.

This tracks all of the extents that are pending for a file in an rbtree.
The checksums of all the extents are stored as well so the pages
that correspond to an extent can be freed before the ordered extent is
actually completed.

This provides routines to wait for all of the ordered extents on a given
file to be done, which much be done before syncing the metadata for the
file to disk in an fsync operation (among others).

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
new file mode 100644
index 0000000..a209401
--- /dev/null
+++ b/fs/btrfs/ordered-data.c
@@ -0,0 +1,730 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/gfp.h>
+#include <linux/slab.h>
+#include <linux/blkdev.h>
+#include <linux/writeback.h>
+#include <linux/pagevec.h>
+#include "ctree.h"
+#include "transaction.h"
+#include "btrfs_inode.h"
+#include "extent_io.h"
+
+static u64 entry_end(struct btrfs_ordered_extent *entry)
+{
+	if (entry->file_offset + entry->len < entry->file_offset)
+		return (u64)-1;
+	return entry->file_offset + entry->len;
+}
+
+/* returns NULL if the insertion worked, or it returns the node it did find
+ * in the tree
+ */
+static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
+				   struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct btrfs_ordered_extent *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
+
+		if (file_offset < entry->file_offset)
+			p = &(*p)->rb_left;
+		else if (file_offset >= entry_end(entry))
+			p = &(*p)->rb_right;
+		else
+			return parent;
+	}
+
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+/*
+ * look for a given offset in the tree, and if it can't be found return the
+ * first lesser offset
+ */
+static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
+				     struct rb_node **prev_ret)
+{
+	struct rb_node *n = root->rb_node;
+	struct rb_node *prev = NULL;
+	struct rb_node *test;
+	struct btrfs_ordered_extent *entry;
+	struct btrfs_ordered_extent *prev_entry = NULL;
+
+	while (n) {
+		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
+		prev = n;
+		prev_entry = entry;
+
+		if (file_offset < entry->file_offset)
+			n = n->rb_left;
+		else if (file_offset >= entry_end(entry))
+			n = n->rb_right;
+		else
+			return n;
+	}
+	if (!prev_ret)
+		return NULL;
+
+	while (prev && file_offset >= entry_end(prev_entry)) {
+		test = rb_next(prev);
+		if (!test)
+			break;
+		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
+				      rb_node);
+		if (file_offset < entry_end(prev_entry))
+			break;
+
+		prev = test;
+	}
+	if (prev)
+		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
+				      rb_node);
+	while (prev && file_offset < entry_end(prev_entry)) {
+		test = rb_prev(prev);
+		if (!test)
+			break;
+		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
+				      rb_node);
+		prev = test;
+	}
+	*prev_ret = prev;
+	return NULL;
+}
+
+/*
+ * helper to check if a given offset is inside a given entry
+ */
+static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
+{
+	if (file_offset < entry->file_offset ||
+	    entry->file_offset + entry->len <= file_offset)
+		return 0;
+	return 1;
+}
+
+/*
+ * look find the first ordered struct that has this offset, otherwise
+ * the first one less than this offset
+ */
+static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
+					  u64 file_offset)
+{
+	struct rb_root *root = &tree->tree;
+	struct rb_node *prev;
+	struct rb_node *ret;
+	struct btrfs_ordered_extent *entry;
+
+	if (tree->last) {
+		entry = rb_entry(tree->last, struct btrfs_ordered_extent,
+				 rb_node);
+		if (offset_in_entry(entry, file_offset))
+			return tree->last;
+	}
+	ret = __tree_search(root, file_offset, &prev);
+	if (!ret)
+		ret = prev;
+	if (ret)
+		tree->last = ret;
+	return ret;
+}
+
+/* allocate and add a new ordered_extent into the per-inode tree.
+ * file_offset is the logical offset in the file
+ *
+ * start is the disk block number of an extent already reserved in the
+ * extent allocation tree
+ *
+ * len is the length of the extent
+ *
+ * This also sets the EXTENT_ORDERED bit on the range in the inode.
+ *
+ * The tree is given a single reference on the ordered extent that was
+ * inserted.
+ */
+int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
+			     u64 start, u64 len, u64 disk_len, int type)
+{
+	struct btrfs_ordered_inode_tree *tree;
+	struct rb_node *node;
+	struct btrfs_ordered_extent *entry;
+
+	tree = &BTRFS_I(inode)->ordered_tree;
+	entry = kzalloc(sizeof(*entry), GFP_NOFS);
+	if (!entry)
+		return -ENOMEM;
+
+	mutex_lock(&tree->mutex);
+	entry->file_offset = file_offset;
+	entry->start = start;
+	entry->len = len;
+	entry->disk_len = disk_len;
+	entry->inode = inode;
+	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
+		set_bit(type, &entry->flags);
+
+	/* one ref for the tree */
+	atomic_set(&entry->refs, 1);
+	init_waitqueue_head(&entry->wait);
+	INIT_LIST_HEAD(&entry->list);
+	INIT_LIST_HEAD(&entry->root_extent_list);
+
+	node = tree_insert(&tree->tree, file_offset,
+			   &entry->rb_node);
+	BUG_ON(node);
+
+	set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
+			   entry_end(entry) - 1, GFP_NOFS);
+
+	spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+	list_add_tail(&entry->root_extent_list,
+		      &BTRFS_I(inode)->root->fs_info->ordered_extents);
+	spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+
+	mutex_unlock(&tree->mutex);
+	BUG_ON(node);
+	return 0;
+}
+
+/*
+ * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
+ * when an ordered extent is finished.  If the list covers more than one
+ * ordered extent, it is split across multiples.
+ */
+int btrfs_add_ordered_sum(struct inode *inode,
+			  struct btrfs_ordered_extent *entry,
+			  struct btrfs_ordered_sum *sum)
+{
+	struct btrfs_ordered_inode_tree *tree;
+
+	tree = &BTRFS_I(inode)->ordered_tree;
+	mutex_lock(&tree->mutex);
+	list_add_tail(&sum->list, &entry->list);
+	mutex_unlock(&tree->mutex);
+	return 0;
+}
+
+/*
+ * this is used to account for finished IO across a given range
+ * of the file.  The IO should not span ordered extents.  If
+ * a given ordered_extent is completely done, 1 is returned, otherwise
+ * 0.
+ *
+ * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
+ * to make sure this function only returns 1 once for a given ordered extent.
+ */
+int btrfs_dec_test_ordered_pending(struct inode *inode,
+				   u64 file_offset, u64 io_size)
+{
+	struct btrfs_ordered_inode_tree *tree;
+	struct rb_node *node;
+	struct btrfs_ordered_extent *entry;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	int ret;
+
+	tree = &BTRFS_I(inode)->ordered_tree;
+	mutex_lock(&tree->mutex);
+	clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
+			     GFP_NOFS);
+	node = tree_search(tree, file_offset);
+	if (!node) {
+		ret = 1;
+		goto out;
+	}
+
+	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+	if (!offset_in_entry(entry, file_offset)) {
+		ret = 1;
+		goto out;
+	}
+
+	ret = test_range_bit(io_tree, entry->file_offset,
+			     entry->file_offset + entry->len - 1,
+			     EXTENT_ORDERED, 0);
+	if (ret == 0)
+		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
+out:
+	mutex_unlock(&tree->mutex);
+	return ret == 0;
+}
+
+/*
+ * used to drop a reference on an ordered extent.  This will free
+ * the extent if the last reference is dropped
+ */
+int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
+{
+	struct list_head *cur;
+	struct btrfs_ordered_sum *sum;
+
+	if (atomic_dec_and_test(&entry->refs)) {
+		while (!list_empty(&entry->list)) {
+			cur = entry->list.next;
+			sum = list_entry(cur, struct btrfs_ordered_sum, list);
+			list_del(&sum->list);
+			kfree(sum);
+		}
+		kfree(entry);
+	}
+	return 0;
+}
+
+/*
+ * remove an ordered extent from the tree.  No references are dropped
+ * but, anyone waiting on this extent is woken up.
+ */
+int btrfs_remove_ordered_extent(struct inode *inode,
+				struct btrfs_ordered_extent *entry)
+{
+	struct btrfs_ordered_inode_tree *tree;
+	struct rb_node *node;
+
+	tree = &BTRFS_I(inode)->ordered_tree;
+	mutex_lock(&tree->mutex);
+	node = &entry->rb_node;
+	rb_erase(node, &tree->tree);
+	tree->last = NULL;
+	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
+
+	spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+	list_del_init(&entry->root_extent_list);
+	spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+
+	mutex_unlock(&tree->mutex);
+	wake_up(&entry->wait);
+	return 0;
+}
+
+/*
+ * wait for all the ordered extents in a root.  This is done when balancing
+ * space between drives.
+ */
+int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
+{
+	struct list_head splice;
+	struct list_head *cur;
+	struct btrfs_ordered_extent *ordered;
+	struct inode *inode;
+
+	INIT_LIST_HEAD(&splice);
+
+	spin_lock(&root->fs_info->ordered_extent_lock);
+	list_splice_init(&root->fs_info->ordered_extents, &splice);
+	while (!list_empty(&splice)) {
+		cur = splice.next;
+		ordered = list_entry(cur, struct btrfs_ordered_extent,
+				     root_extent_list);
+		if (nocow_only &&
+		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
+		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
+			list_move(&ordered->root_extent_list,
+				  &root->fs_info->ordered_extents);
+			cond_resched_lock(&root->fs_info->ordered_extent_lock);
+			continue;
+		}
+
+		list_del_init(&ordered->root_extent_list);
+		atomic_inc(&ordered->refs);
+
+		/*
+		 * the inode may be getting freed (in sys_unlink path).
+		 */
+		inode = igrab(ordered->inode);
+
+		spin_unlock(&root->fs_info->ordered_extent_lock);
+
+		if (inode) {
+			btrfs_start_ordered_extent(inode, ordered, 1);
+			btrfs_put_ordered_extent(ordered);
+			iput(inode);
+		} else {
+			btrfs_put_ordered_extent(ordered);
+		}
+
+		spin_lock(&root->fs_info->ordered_extent_lock);
+	}
+	spin_unlock(&root->fs_info->ordered_extent_lock);
+	return 0;
+}
+
+/*
+ * Used to start IO or wait for a given ordered extent to finish.
+ *
+ * If wait is one, this effectively waits on page writeback for all the pages
+ * in the extent, and it waits on the io completion code to insert
+ * metadata into the btree corresponding to the extent
+ */
+void btrfs_start_ordered_extent(struct inode *inode,
+				       struct btrfs_ordered_extent *entry,
+				       int wait)
+{
+	u64 start = entry->file_offset;
+	u64 end = start + entry->len - 1;
+
+	/*
+	 * pages in the range can be dirty, clean or writeback.  We
+	 * start IO on any dirty ones so the wait doesn't stall waiting
+	 * for pdflush to find them
+	 */
+	btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_ALL);
+	if (wait) {
+		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
+						 &entry->flags));
+	}
+}
+
+/*
+ * Used to wait on ordered extents across a large range of bytes.
+ */
+int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
+{
+	u64 end;
+	u64 orig_end;
+	u64 wait_end;
+	struct btrfs_ordered_extent *ordered;
+
+	if (start + len < start) {
+		orig_end = INT_LIMIT(loff_t);
+	} else {
+		orig_end = start + len - 1;
+		if (orig_end > INT_LIMIT(loff_t))
+			orig_end = INT_LIMIT(loff_t);
+	}
+	wait_end = orig_end;
+again:
+	/* start IO across the range first to instantiate any delalloc
+	 * extents
+	 */
+	btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
+
+	/* The compression code will leave pages locked but return from
+	 * writepage without setting the page writeback.  Starting again
+	 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
+	 */
+	btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
+
+	btrfs_wait_on_page_writeback_range(inode->i_mapping,
+					   start >> PAGE_CACHE_SHIFT,
+					   orig_end >> PAGE_CACHE_SHIFT);
+
+	end = orig_end;
+	while (1) {
+		ordered = btrfs_lookup_first_ordered_extent(inode, end);
+		if (!ordered)
+			break;
+		if (ordered->file_offset > orig_end) {
+			btrfs_put_ordered_extent(ordered);
+			break;
+		}
+		if (ordered->file_offset + ordered->len < start) {
+			btrfs_put_ordered_extent(ordered);
+			break;
+		}
+		btrfs_start_ordered_extent(inode, ordered, 1);
+		end = ordered->file_offset;
+		btrfs_put_ordered_extent(ordered);
+		if (end == 0 || end == start)
+			break;
+		end--;
+	}
+	if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
+			   EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
+		schedule_timeout(1);
+		goto again;
+	}
+	return 0;
+}
+
+/*
+ * find an ordered extent corresponding to file_offset.  return NULL if
+ * nothing is found, otherwise take a reference on the extent and return it
+ */
+struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
+							 u64 file_offset)
+{
+	struct btrfs_ordered_inode_tree *tree;
+	struct rb_node *node;
+	struct btrfs_ordered_extent *entry = NULL;
+
+	tree = &BTRFS_I(inode)->ordered_tree;
+	mutex_lock(&tree->mutex);
+	node = tree_search(tree, file_offset);
+	if (!node)
+		goto out;
+
+	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+	if (!offset_in_entry(entry, file_offset))
+		entry = NULL;
+	if (entry)
+		atomic_inc(&entry->refs);
+out:
+	mutex_unlock(&tree->mutex);
+	return entry;
+}
+
+/*
+ * lookup and return any extent before 'file_offset'.  NULL is returned
+ * if none is found
+ */
+struct btrfs_ordered_extent *
+btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
+{
+	struct btrfs_ordered_inode_tree *tree;
+	struct rb_node *node;
+	struct btrfs_ordered_extent *entry = NULL;
+
+	tree = &BTRFS_I(inode)->ordered_tree;
+	mutex_lock(&tree->mutex);
+	node = tree_search(tree, file_offset);
+	if (!node)
+		goto out;
+
+	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+	atomic_inc(&entry->refs);
+out:
+	mutex_unlock(&tree->mutex);
+	return entry;
+}
+
+/*
+ * After an extent is done, call this to conditionally update the on disk
+ * i_size.  i_size is updated to cover any fully written part of the file.
+ */
+int btrfs_ordered_update_i_size(struct inode *inode,
+				struct btrfs_ordered_extent *ordered)
+{
+	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	u64 disk_i_size;
+	u64 new_i_size;
+	u64 i_size_test;
+	struct rb_node *node;
+	struct btrfs_ordered_extent *test;
+
+	mutex_lock(&tree->mutex);
+	disk_i_size = BTRFS_I(inode)->disk_i_size;
+
+	/*
+	 * if the disk i_size is already at the inode->i_size, or
+	 * this ordered extent is inside the disk i_size, we're done
+	 */
+	if (disk_i_size >= inode->i_size ||
+	    ordered->file_offset + ordered->len <= disk_i_size) {
+		goto out;
+	}
+
+	/*
+	 * we can't update the disk_isize if there are delalloc bytes
+	 * between disk_i_size and  this ordered extent
+	 */
+	if (test_range_bit(io_tree, disk_i_size,
+			   ordered->file_offset + ordered->len - 1,
+			   EXTENT_DELALLOC, 0)) {
+		goto out;
+	}
+	/*
+	 * walk backward from this ordered extent to disk_i_size.
+	 * if we find an ordered extent then we can't update disk i_size
+	 * yet
+	 */
+	node = &ordered->rb_node;
+	while (1) {
+		node = rb_prev(node);
+		if (!node)
+			break;
+		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+		if (test->file_offset + test->len <= disk_i_size)
+			break;
+		if (test->file_offset >= inode->i_size)
+			break;
+		if (test->file_offset >= disk_i_size)
+			goto out;
+	}
+	new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
+
+	/*
+	 * at this point, we know we can safely update i_size to at least
+	 * the offset from this ordered extent.  But, we need to
+	 * walk forward and see if ios from higher up in the file have
+	 * finished.
+	 */
+	node = rb_next(&ordered->rb_node);
+	i_size_test = 0;
+	if (node) {
+		/*
+		 * do we have an area where IO might have finished
+		 * between our ordered extent and the next one.
+		 */
+		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+		if (test->file_offset > entry_end(ordered))
+			i_size_test = test->file_offset;
+	} else {
+		i_size_test = i_size_read(inode);
+	}
+
+	/*
+	 * i_size_test is the end of a region after this ordered
+	 * extent where there are no ordered extents.  As long as there
+	 * are no delalloc bytes in this area, it is safe to update
+	 * disk_i_size to the end of the region.
+	 */
+	if (i_size_test > entry_end(ordered) &&
+	    !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
+			   EXTENT_DELALLOC, 0)) {
+		new_i_size = min_t(u64, i_size_test, i_size_read(inode));
+	}
+	BTRFS_I(inode)->disk_i_size = new_i_size;
+out:
+	mutex_unlock(&tree->mutex);
+	return 0;
+}
+
+/*
+ * search the ordered extents for one corresponding to 'offset' and
+ * try to find a checksum.  This is used because we allow pages to
+ * be reclaimed before their checksum is actually put into the btree
+ */
+int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
+			   u32 *sum)
+{
+	struct btrfs_ordered_sum *ordered_sum;
+	struct btrfs_sector_sum *sector_sums;
+	struct btrfs_ordered_extent *ordered;
+	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
+	struct list_head *cur;
+	unsigned long num_sectors;
+	unsigned long i;
+	u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
+	int ret = 1;
+
+	ordered = btrfs_lookup_ordered_extent(inode, offset);
+	if (!ordered)
+		return 1;
+
+	mutex_lock(&tree->mutex);
+	list_for_each_prev(cur, &ordered->list) {
+		ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
+		if (disk_bytenr >= ordered_sum->bytenr) {
+			num_sectors = ordered_sum->len / sectorsize;
+			sector_sums = ordered_sum->sums;
+			for (i = 0; i < num_sectors; i++) {
+				if (sector_sums[i].bytenr == disk_bytenr) {
+					*sum = sector_sums[i].sum;
+					ret = 0;
+					goto out;
+				}
+			}
+		}
+	}
+out:
+	mutex_unlock(&tree->mutex);
+	btrfs_put_ordered_extent(ordered);
+	return ret;
+}
+
+
+/**
+ * taken from mm/filemap.c because it isn't exported
+ *
+ * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
+ * @mapping:	address space structure to write
+ * @start:	offset in bytes where the range starts
+ * @end:	offset in bytes where the range ends (inclusive)
+ * @sync_mode:	enable synchronous operation
+ *
+ * Start writeback against all of a mapping's dirty pages that lie
+ * within the byte offsets <start, end> inclusive.
+ *
+ * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
+ * opposed to a regular memory cleansing writeback.  The difference between
+ * these two operations is that if a dirty page/buffer is encountered, it must
+ * be waited upon, and not just skipped over.
+ */
+int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
+			   loff_t end, int sync_mode)
+{
+	struct writeback_control wbc = {
+		.sync_mode = sync_mode,
+		.nr_to_write = mapping->nrpages * 2,
+		.range_start = start,
+		.range_end = end,
+		.for_writepages = 1,
+	};
+	return btrfs_writepages(mapping, &wbc);
+}
+
+/**
+ * taken from mm/filemap.c because it isn't exported
+ *
+ * wait_on_page_writeback_range - wait for writeback to complete
+ * @mapping:	target address_space
+ * @start:	beginning page index
+ * @end:	ending page index
+ *
+ * Wait for writeback to complete against pages indexed by start->end
+ * inclusive
+ */
+int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
+				       pgoff_t start, pgoff_t end)
+{
+	struct pagevec pvec;
+	int nr_pages;
+	int ret = 0;
+	pgoff_t index;
+
+	if (end < start)
+		return 0;
+
+	pagevec_init(&pvec, 0);
+	index = start;
+	while ((index <= end) &&
+			(nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
+			PAGECACHE_TAG_WRITEBACK,
+			min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
+		unsigned i;
+
+		for (i = 0; i < nr_pages; i++) {
+			struct page *page = pvec.pages[i];
+
+			/* until radix tree lookup accepts end_index */
+			if (page->index > end)
+				continue;
+
+			wait_on_page_writeback(page);
+			if (PageError(page))
+				ret = -EIO;
+		}
+		pagevec_release(&pvec);
+		cond_resched();
+	}
+
+	/* Check for outstanding write errors */
+	if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
+		ret = -ENOSPC;
+	if (test_and_clear_bit(AS_EIO, &mapping->flags))
+		ret = -EIO;
+
+	return ret;
+}
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
new file mode 100644
index 0000000..ab66d5e
--- /dev/null
+++ b/fs/btrfs/ordered-data.h
@@ -0,0 +1,158 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_ORDERED_DATA__
+#define __BTRFS_ORDERED_DATA__
+
+/* one of these per inode */
+struct btrfs_ordered_inode_tree {
+	struct mutex mutex;
+	struct rb_root tree;
+	struct rb_node *last;
+};
+
+/*
+ * these are used to collect checksums done just before bios submission.
+ * They are attached via a list into the ordered extent, and
+ * checksum items are inserted into the tree after all the blocks in
+ * the ordered extent are on disk
+ */
+struct btrfs_sector_sum {
+	/* bytenr on disk */
+	u64 bytenr;
+	u32 sum;
+};
+
+struct btrfs_ordered_sum {
+	/* bytenr is the start of this extent on disk */
+	u64 bytenr;
+
+	/*
+	 * this is the length in bytes covered by the sums array below.
+	 */
+	unsigned long len;
+	struct list_head list;
+	/* last field is a variable length array of btrfs_sector_sums */
+	struct btrfs_sector_sum sums[];
+};
+
+/*
+ * bits for the flags field:
+ *
+ * BTRFS_ORDERED_IO_DONE is set when all of the blocks are written.
+ * It is used to make sure metadata is inserted into the tree only once
+ * per extent.
+ *
+ * BTRFS_ORDERED_COMPLETE is set when the extent is removed from the
+ * rbtree, just before waking any waiters.  It is used to indicate the
+ * IO is done and any metadata is inserted into the tree.
+ */
+#define BTRFS_ORDERED_IO_DONE 0 /* set when all the pages are written */
+
+#define BTRFS_ORDERED_COMPLETE 1 /* set when removed from the tree */
+
+#define BTRFS_ORDERED_NOCOW 2 /* set when we want to write in place */
+
+#define BTRFS_ORDERED_COMPRESSED 3 /* writing a compressed extent */
+
+#define BTRFS_ORDERED_PREALLOC 4 /* set when writing to prealloced extent */
+
+struct btrfs_ordered_extent {
+	/* logical offset in the file */
+	u64 file_offset;
+
+	/* disk byte number */
+	u64 start;
+
+	/* ram length of the extent in bytes */
+	u64 len;
+
+	/* extent length on disk */
+	u64 disk_len;
+
+	/* flags (described above) */
+	unsigned long flags;
+
+	/* reference count */
+	atomic_t refs;
+
+	/* the inode we belong to */
+	struct inode *inode;
+
+	/* list of checksums for insertion when the extent io is done */
+	struct list_head list;
+
+	/* used to wait for the BTRFS_ORDERED_COMPLETE bit */
+	wait_queue_head_t wait;
+
+	/* our friendly rbtree entry */
+	struct rb_node rb_node;
+
+	/* a per root list of all the pending ordered extents */
+	struct list_head root_extent_list;
+};
+
+
+/*
+ * calculates the total size you need to allocate for an ordered sum
+ * structure spanning 'bytes' in the file
+ */
+static inline int btrfs_ordered_sum_size(struct btrfs_root *root,
+					 unsigned long bytes)
+{
+	unsigned long num_sectors = (bytes + root->sectorsize - 1) /
+		root->sectorsize;
+	num_sectors++;
+	return sizeof(struct btrfs_ordered_sum) +
+		num_sectors * sizeof(struct btrfs_sector_sum);
+}
+
+static inline void
+btrfs_ordered_inode_tree_init(struct btrfs_ordered_inode_tree *t)
+{
+	mutex_init(&t->mutex);
+	t->tree.rb_node = NULL;
+	t->last = NULL;
+}
+
+int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry);
+int btrfs_remove_ordered_extent(struct inode *inode,
+				struct btrfs_ordered_extent *entry);
+int btrfs_dec_test_ordered_pending(struct inode *inode,
+				       u64 file_offset, u64 io_size);
+int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
+			     u64 start, u64 len, u64 disk_len, int tyep);
+int btrfs_add_ordered_sum(struct inode *inode,
+			  struct btrfs_ordered_extent *entry,
+			  struct btrfs_ordered_sum *sum);
+struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
+							 u64 file_offset);
+void btrfs_start_ordered_extent(struct inode *inode,
+				struct btrfs_ordered_extent *entry, int wait);
+int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len);
+struct btrfs_ordered_extent *
+btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset);
+int btrfs_ordered_update_i_size(struct inode *inode,
+				struct btrfs_ordered_extent *ordered);
+int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, u32 *sum);
+int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
+				       pgoff_t start, pgoff_t end);
+int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
+			   loff_t end, int sync_mode);
+int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only);
+#endif
-- 
1.6.0.2

[PATCH 08/35] Btrfs btree locking helpers

[Chris Mason]

These are simple wrappers around mutexes that take extent_buffers.  The
btree lock currently does an adaptive spin, but that should go away soon.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
new file mode 100644
index 0000000..39bae77
--- /dev/null
+++ b/fs/btrfs/locking.c
@@ -0,0 +1,88 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+#include <linux/sched.h>
+#include <linux/gfp.h>
+#include <linux/pagemap.h>
+#include <linux/spinlock.h>
+#include <linux/page-flags.h>
+#include <asm/bug.h>
+#include "ctree.h"
+#include "extent_io.h"
+#include "locking.h"
+
+/*
+ * locks the per buffer mutex in an extent buffer.  This uses adaptive locks
+ * and the spin is not tuned very extensively.  The spinning does make a big
+ * difference in almost every workload, but spinning for the right amount of
+ * time needs some help.
+ *
+ * In general, we want to spin as long as the lock holder is doing btree
+ * searches, and we should give up if they are in more expensive code.
+ */
+
+int btrfs_tree_lock(struct extent_buffer *eb)
+{
+	int i;
+
+	if (mutex_trylock(&eb->mutex))
+		return 0;
+	for (i = 0; i < 512; i++) {
+		cpu_relax();
+		if (mutex_trylock(&eb->mutex))
+			return 0;
+	}
+	cpu_relax();
+	mutex_lock_nested(&eb->mutex, BTRFS_MAX_LEVEL - btrfs_header_level(eb));
+	return 0;
+}
+
+int btrfs_try_tree_lock(struct extent_buffer *eb)
+{
+	return mutex_trylock(&eb->mutex);
+}
+
+int btrfs_tree_unlock(struct extent_buffer *eb)
+{
+	mutex_unlock(&eb->mutex);
+	return 0;
+}
+
+int btrfs_tree_locked(struct extent_buffer *eb)
+{
+	return mutex_is_locked(&eb->mutex);
+}
+
+/*
+ * btrfs_search_slot uses this to decide if it should drop its locks
+ * before doing something expensive like allocating free blocks for cow.
+ */
+int btrfs_path_lock_waiting(struct btrfs_path *path, int level)
+{
+	int i;
+	struct extent_buffer *eb;
+	for (i = level; i <= level + 1 && i < BTRFS_MAX_LEVEL; i++) {
+		eb = path->nodes[i];
+		if (!eb)
+			break;
+		smp_mb();
+		if (!list_empty(&eb->mutex.wait_list))
+			return 1;
+	}
+	return 0;
+}
+
diff --git a/fs/btrfs/locking.h b/fs/btrfs/locking.h
new file mode 100644
index 0000000..bc1faef
--- /dev/null
+++ b/fs/btrfs/locking.h
@@ -0,0 +1,27 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_LOCKING_
+#define __BTRFS_LOCKING_
+
+int btrfs_tree_lock(struct extent_buffer *eb);
+int btrfs_tree_unlock(struct extent_buffer *eb);
+int btrfs_tree_locked(struct extent_buffer *eb);
+int btrfs_try_tree_lock(struct extent_buffer *eb);
+int btrfs_path_lock_waiting(struct btrfs_path *path, int level);
+#endif
-- 
1.6.0.2

[PATCH 09/35] Btrfs zlib helpers

[Chris Mason]

These are helpers to send pages down to the zlib API.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/zlib.c b/fs/btrfs/zlib.c
new file mode 100644
index 0000000..ecfbce8
--- /dev/null
+++ b/fs/btrfs/zlib.c
@@ -0,0 +1,632 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ * Based on jffs2 zlib code:
+ * Copyright © 2001-2007 Red Hat, Inc.
+ * Created by David Woodhouse <dwmw2@infradead.org>
+ */
+
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/zlib.h>
+#include <linux/zutil.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/err.h>
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include <linux/bio.h>
+#include "compression.h"
+
+/* Plan: call deflate() with avail_in == *sourcelen,
+	avail_out = *dstlen - 12 and flush == Z_FINISH.
+	If it doesn't manage to finish,	call it again with
+	avail_in == 0 and avail_out set to the remaining 12
+	bytes for it to clean up.
+   Q: Is 12 bytes sufficient?
+*/
+#define STREAM_END_SPACE 12
+
+struct workspace {
+	z_stream inf_strm;
+	z_stream def_strm;
+	char *buf;
+	struct list_head list;
+};
+
+static LIST_HEAD(idle_workspace);
+static DEFINE_SPINLOCK(workspace_lock);
+static unsigned long num_workspace;
+static atomic_t alloc_workspace = ATOMIC_INIT(0);
+static DECLARE_WAIT_QUEUE_HEAD(workspace_wait);
+
+/*
+ * this finds an available zlib workspace or allocates a new one
+ * NULL or an ERR_PTR is returned if things go bad.
+ */
+static struct workspace *find_zlib_workspace(void)
+{
+	struct workspace *workspace;
+	int ret;
+	int cpus = num_online_cpus();
+
+again:
+	spin_lock(&workspace_lock);
+	if (!list_empty(&idle_workspace)) {
+		workspace = list_entry(idle_workspace.next, struct workspace,
+				       list);
+		list_del(&workspace->list);
+		num_workspace--;
+		spin_unlock(&workspace_lock);
+		return workspace;
+
+	}
+	spin_unlock(&workspace_lock);
+	if (atomic_read(&alloc_workspace) > cpus) {
+		DEFINE_WAIT(wait);
+		prepare_to_wait(&workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
+		if (atomic_read(&alloc_workspace) > cpus)
+			schedule();
+		finish_wait(&workspace_wait, &wait);
+		goto again;
+	}
+	atomic_inc(&alloc_workspace);
+	workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
+	if (!workspace) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	workspace->def_strm.workspace = vmalloc(zlib_deflate_workspacesize());
+	if (!workspace->def_strm.workspace) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+	workspace->inf_strm.workspace = vmalloc(zlib_inflate_workspacesize());
+	if (!workspace->inf_strm.workspace) {
+		ret = -ENOMEM;
+		goto fail_inflate;
+	}
+	workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
+	if (!workspace->buf) {
+		ret = -ENOMEM;
+		goto fail_kmalloc;
+	}
+	return workspace;
+
+fail_kmalloc:
+	vfree(workspace->inf_strm.workspace);
+fail_inflate:
+	vfree(workspace->def_strm.workspace);
+fail:
+	kfree(workspace);
+	atomic_dec(&alloc_workspace);
+	wake_up(&workspace_wait);
+	return ERR_PTR(ret);
+}
+
+/*
+ * put a workspace struct back on the list or free it if we have enough
+ * idle ones sitting around
+ */
+static int free_workspace(struct workspace *workspace)
+{
+	spin_lock(&workspace_lock);
+	if (num_workspace < num_online_cpus()) {
+		list_add_tail(&workspace->list, &idle_workspace);
+		num_workspace++;
+		spin_unlock(&workspace_lock);
+		if (waitqueue_active(&workspace_wait))
+			wake_up(&workspace_wait);
+		return 0;
+	}
+	spin_unlock(&workspace_lock);
+	vfree(workspace->def_strm.workspace);
+	vfree(workspace->inf_strm.workspace);
+	kfree(workspace->buf);
+	kfree(workspace);
+
+	atomic_dec(&alloc_workspace);
+	if (waitqueue_active(&workspace_wait))
+		wake_up(&workspace_wait);
+	return 0;
+}
+
+/*
+ * cleanup function for module exit
+ */
+static void free_workspaces(void)
+{
+	struct workspace *workspace;
+	while (!list_empty(&idle_workspace)) {
+		workspace = list_entry(idle_workspace.next, struct workspace,
+				       list);
+		list_del(&workspace->list);
+		vfree(workspace->def_strm.workspace);
+		vfree(workspace->inf_strm.workspace);
+		kfree(workspace->buf);
+		kfree(workspace);
+		atomic_dec(&alloc_workspace);
+	}
+}
+
+/*
+ * given an address space and start/len, compress the bytes.
+ *
+ * pages are allocated to hold the compressed result and stored
+ * in 'pages'
+ *
+ * out_pages is used to return the number of pages allocated.  There
+ * may be pages allocated even if we return an error
+ *
+ * total_in is used to return the number of bytes actually read.  It
+ * may be smaller then len if we had to exit early because we
+ * ran out of room in the pages array or because we cross the
+ * max_out threshold.
+ *
+ * total_out is used to return the total number of compressed bytes
+ *
+ * max_out tells us the max number of bytes that we're allowed to
+ * stuff into pages
+ */
+int btrfs_zlib_compress_pages(struct address_space *mapping,
+			      u64 start, unsigned long len,
+			      struct page **pages,
+			      unsigned long nr_dest_pages,
+			      unsigned long *out_pages,
+			      unsigned long *total_in,
+			      unsigned long *total_out,
+			      unsigned long max_out)
+{
+	int ret;
+	struct workspace *workspace;
+	char *data_in;
+	char *cpage_out;
+	int nr_pages = 0;
+	struct page *in_page = NULL;
+	struct page *out_page = NULL;
+	int out_written = 0;
+	int in_read = 0;
+	unsigned long bytes_left;
+
+	*out_pages = 0;
+	*total_out = 0;
+	*total_in = 0;
+
+	workspace = find_zlib_workspace();
+	if (!workspace)
+		return -1;
+
+	if (Z_OK != zlib_deflateInit(&workspace->def_strm, 3)) {
+		printk(KERN_WARNING "deflateInit failed\n");
+		ret = -1;
+		goto out;
+	}
+
+	workspace->def_strm.total_in = 0;
+	workspace->def_strm.total_out = 0;
+
+	in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
+	data_in = kmap(in_page);
+
+	out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
+	cpage_out = kmap(out_page);
+	pages[0] = out_page;
+	nr_pages = 1;
+
+	workspace->def_strm.next_in = data_in;
+	workspace->def_strm.next_out = cpage_out;
+	workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
+	workspace->def_strm.avail_in = min(len, PAGE_CACHE_SIZE);
+
+	out_written = 0;
+	in_read = 0;
+
+	while (workspace->def_strm.total_in < len) {
+		ret = zlib_deflate(&workspace->def_strm, Z_SYNC_FLUSH);
+		if (ret != Z_OK) {
+			printk(KERN_DEBUG "btrfs deflate in loop returned %d\n",
+			       ret);
+			zlib_deflateEnd(&workspace->def_strm);
+			ret = -1;
+			goto out;
+		}
+
+		/* we're making it bigger, give up */
+		if (workspace->def_strm.total_in > 8192 &&
+		    workspace->def_strm.total_in <
+		    workspace->def_strm.total_out) {
+			ret = -1;
+			goto out;
+		}
+		/* we need another page for writing out.  Test this
+		 * before the total_in so we will pull in a new page for
+		 * the stream end if required
+		 */
+		if (workspace->def_strm.avail_out == 0) {
+			kunmap(out_page);
+			if (nr_pages == nr_dest_pages) {
+				out_page = NULL;
+				ret = -1;
+				goto out;
+			}
+			out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
+			cpage_out = kmap(out_page);
+			pages[nr_pages] = out_page;
+			nr_pages++;
+			workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
+			workspace->def_strm.next_out = cpage_out;
+		}
+		/* we're all done */
+		if (workspace->def_strm.total_in >= len)
+			break;
+
+		/* we've read in a full page, get a new one */
+		if (workspace->def_strm.avail_in == 0) {
+			if (workspace->def_strm.total_out > max_out)
+				break;
+
+			bytes_left = len - workspace->def_strm.total_in;
+			kunmap(in_page);
+			page_cache_release(in_page);
+
+			start += PAGE_CACHE_SIZE;
+			in_page = find_get_page(mapping,
+						start >> PAGE_CACHE_SHIFT);
+			data_in = kmap(in_page);
+			workspace->def_strm.avail_in = min(bytes_left,
+							   PAGE_CACHE_SIZE);
+			workspace->def_strm.next_in = data_in;
+		}
+	}
+	workspace->def_strm.avail_in = 0;
+	ret = zlib_deflate(&workspace->def_strm, Z_FINISH);
+	zlib_deflateEnd(&workspace->def_strm);
+
+	if (ret != Z_STREAM_END) {
+		ret = -1;
+		goto out;
+	}
+
+	if (workspace->def_strm.total_out >= workspace->def_strm.total_in) {
+		ret = -1;
+		goto out;
+	}
+
+	ret = 0;
+	*total_out = workspace->def_strm.total_out;
+	*total_in = workspace->def_strm.total_in;
+out:
+	*out_pages = nr_pages;
+	if (out_page)
+		kunmap(out_page);
+
+	if (in_page) {
+		kunmap(in_page);
+		page_cache_release(in_page);
+	}
+	free_workspace(workspace);
+	return ret;
+}
+
+/*
+ * pages_in is an array of pages with compressed data.
+ *
+ * disk_start is the starting logical offset of this array in the file
+ *
+ * bvec is a bio_vec of pages from the file that we want to decompress into
+ *
+ * vcnt is the count of pages in the biovec
+ *
+ * srclen is the number of bytes in pages_in
+ *
+ * The basic idea is that we have a bio that was created by readpages.
+ * The pages in the bio are for the uncompressed data, and they may not
+ * be contiguous.  They all correspond to the range of bytes covered by
+ * the compressed extent.
+ */
+int btrfs_zlib_decompress_biovec(struct page **pages_in,
+			      u64 disk_start,
+			      struct bio_vec *bvec,
+			      int vcnt,
+			      size_t srclen)
+{
+	int ret = 0;
+	int wbits = MAX_WBITS;
+	struct workspace *workspace;
+	char *data_in;
+	size_t total_out = 0;
+	unsigned long page_bytes_left;
+	unsigned long page_in_index = 0;
+	unsigned long page_out_index = 0;
+	struct page *page_out;
+	unsigned long total_pages_in = (srclen + PAGE_CACHE_SIZE - 1) /
+					PAGE_CACHE_SIZE;
+	unsigned long buf_start;
+	unsigned long buf_offset;
+	unsigned long bytes;
+	unsigned long working_bytes;
+	unsigned long pg_offset;
+	unsigned long start_byte;
+	unsigned long current_buf_start;
+	char *kaddr;
+
+	workspace = find_zlib_workspace();
+	if (!workspace)
+		return -ENOMEM;
+
+	data_in = kmap(pages_in[page_in_index]);
+	workspace->inf_strm.next_in = data_in;
+	workspace->inf_strm.avail_in = min_t(size_t, srclen, PAGE_CACHE_SIZE);
+	workspace->inf_strm.total_in = 0;
+
+	workspace->inf_strm.total_out = 0;
+	workspace->inf_strm.next_out = workspace->buf;
+	workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
+	page_out = bvec[page_out_index].bv_page;
+	page_bytes_left = PAGE_CACHE_SIZE;
+	pg_offset = 0;
+
+	/* If it's deflate, and it's got no preset dictionary, then
+	   we can tell zlib to skip the adler32 check. */
+	if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
+	    ((data_in[0] & 0x0f) == Z_DEFLATED) &&
+	    !(((data_in[0]<<8) + data_in[1]) % 31)) {
+
+		wbits = -((data_in[0] >> 4) + 8);
+		workspace->inf_strm.next_in += 2;
+		workspace->inf_strm.avail_in -= 2;
+	}
+
+	if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
+		printk(KERN_WARNING "inflateInit failed\n");
+		ret = -1;
+		goto out;
+	}
+	while (workspace->inf_strm.total_in < srclen) {
+		ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
+		if (ret != Z_OK && ret != Z_STREAM_END)
+			break;
+		/*
+		 * buf start is the byte offset we're of the start of
+		 * our workspace buffer
+		 */
+		buf_start = total_out;
+
+		/* total_out is the last byte of the workspace buffer */
+		total_out = workspace->inf_strm.total_out;
+
+		working_bytes = total_out - buf_start;
+
+		/*
+		 * start byte is the first byte of the page we're currently
+		 * copying into relative to the start of the compressed data.
+		 */
+		start_byte = page_offset(page_out) - disk_start;
+
+		if (working_bytes == 0) {
+			/* we didn't make progress in this inflate
+			 * call, we're done
+			 */
+			if (ret != Z_STREAM_END)
+				ret = -1;
+			break;
+		}
+
+		/* we haven't yet hit data corresponding to this page */
+		if (total_out <= start_byte)
+			goto next;
+
+		/*
+		 * the start of the data we care about is offset into
+		 * the middle of our working buffer
+		 */
+		if (total_out > start_byte && buf_start < start_byte) {
+			buf_offset = start_byte - buf_start;
+			working_bytes -= buf_offset;
+		} else {
+			buf_offset = 0;
+		}
+		current_buf_start = buf_start;
+
+		/* copy bytes from the working buffer into the pages */
+		while (working_bytes > 0) {
+			bytes = min(PAGE_CACHE_SIZE - pg_offset,
+				    PAGE_CACHE_SIZE - buf_offset);
+			bytes = min(bytes, working_bytes);
+			kaddr = kmap_atomic(page_out, KM_USER0);
+			memcpy(kaddr + pg_offset, workspace->buf + buf_offset,
+			       bytes);
+			kunmap_atomic(kaddr, KM_USER0);
+			flush_dcache_page(page_out);
+
+			pg_offset += bytes;
+			page_bytes_left -= bytes;
+			buf_offset += bytes;
+			working_bytes -= bytes;
+			current_buf_start += bytes;
+
+			/* check if we need to pick another page */
+			if (page_bytes_left == 0) {
+				page_out_index++;
+				if (page_out_index >= vcnt) {
+					ret = 0;
+					goto done;
+				}
+
+				page_out = bvec[page_out_index].bv_page;
+				pg_offset = 0;
+				page_bytes_left = PAGE_CACHE_SIZE;
+				start_byte = page_offset(page_out) - disk_start;
+
+				/*
+				 * make sure our new page is covered by this
+				 * working buffer
+				 */
+				if (total_out <= start_byte)
+					goto next;
+
+				/* the next page in the biovec might not
+				 * be adjacent to the last page, but it
+				 * might still be found inside this working
+				 * buffer.  bump our offset pointer
+				 */
+				if (total_out > start_byte &&
+				    current_buf_start < start_byte) {
+					buf_offset = start_byte - buf_start;
+					working_bytes = total_out - start_byte;
+					current_buf_start = buf_start +
+						buf_offset;
+				}
+			}
+		}
+next:
+		workspace->inf_strm.next_out = workspace->buf;
+		workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
+
+		if (workspace->inf_strm.avail_in == 0) {
+			unsigned long tmp;
+			kunmap(pages_in[page_in_index]);
+			page_in_index++;
+			if (page_in_index >= total_pages_in) {
+				data_in = NULL;
+				break;
+			}
+			data_in = kmap(pages_in[page_in_index]);
+			workspace->inf_strm.next_in = data_in;
+			tmp = srclen - workspace->inf_strm.total_in;
+			workspace->inf_strm.avail_in = min(tmp,
+							   PAGE_CACHE_SIZE);
+		}
+	}
+	if (ret != Z_STREAM_END)
+		ret = -1;
+	else
+		ret = 0;
+done:
+	zlib_inflateEnd(&workspace->inf_strm);
+	if (data_in)
+		kunmap(pages_in[page_in_index]);
+out:
+	free_workspace(workspace);
+	return ret;
+}
+
+/*
+ * a less complex decompression routine.  Our compressed data fits in a
+ * single page, and we want to read a single page out of it.
+ * start_byte tells us the offset into the compressed data we're interested in
+ */
+int btrfs_zlib_decompress(unsigned char *data_in,
+			  struct page *dest_page,
+			  unsigned long start_byte,
+			  size_t srclen, size_t destlen)
+{
+	int ret = 0;
+	int wbits = MAX_WBITS;
+	struct workspace *workspace;
+	unsigned long bytes_left = destlen;
+	unsigned long total_out = 0;
+	char *kaddr;
+
+	if (destlen > PAGE_CACHE_SIZE)
+		return -ENOMEM;
+
+	workspace = find_zlib_workspace();
+	if (!workspace)
+		return -ENOMEM;
+
+	workspace->inf_strm.next_in = data_in;
+	workspace->inf_strm.avail_in = srclen;
+	workspace->inf_strm.total_in = 0;
+
+	workspace->inf_strm.next_out = workspace->buf;
+	workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
+	workspace->inf_strm.total_out = 0;
+	/* If it's deflate, and it's got no preset dictionary, then
+	   we can tell zlib to skip the adler32 check. */
+	if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
+	    ((data_in[0] & 0x0f) == Z_DEFLATED) &&
+	    !(((data_in[0]<<8) + data_in[1]) % 31)) {
+
+		wbits = -((data_in[0] >> 4) + 8);
+		workspace->inf_strm.next_in += 2;
+		workspace->inf_strm.avail_in -= 2;
+	}
+
+	if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
+		printk(KERN_WARNING "inflateInit failed\n");
+		ret = -1;
+		goto out;
+	}
+
+	while (bytes_left > 0) {
+		unsigned long buf_start;
+		unsigned long buf_offset;
+		unsigned long bytes;
+		unsigned long pg_offset = 0;
+
+		ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
+		if (ret != Z_OK && ret != Z_STREAM_END)
+			break;
+
+		buf_start = total_out;
+		total_out = workspace->inf_strm.total_out;
+
+		if (total_out == buf_start) {
+			ret = -1;
+			break;
+		}
+
+		if (total_out <= start_byte)
+			goto next;
+
+		if (total_out > start_byte && buf_start < start_byte)
+			buf_offset = start_byte - buf_start;
+		else
+			buf_offset = 0;
+
+		bytes = min(PAGE_CACHE_SIZE - pg_offset,
+			    PAGE_CACHE_SIZE - buf_offset);
+		bytes = min(bytes, bytes_left);
+
+		kaddr = kmap_atomic(dest_page, KM_USER0);
+		memcpy(kaddr + pg_offset, workspace->buf + buf_offset, bytes);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		pg_offset += bytes;
+		bytes_left -= bytes;
+next:
+		workspace->inf_strm.next_out = workspace->buf;
+		workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
+	}
+
+	if (ret != Z_STREAM_END && bytes_left != 0)
+		ret = -1;
+	else
+		ret = 0;
+
+	zlib_inflateEnd(&workspace->inf_strm);
+out:
+	free_workspace(workspace);
+	return ret;
+}
+
+void btrfs_zlib_exit(void)
+{
+    free_workspaces();
+}
-- 
1.6.0.2

--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 10/35] Btrfs online btree defragging

[Chris Mason]

This walks a btree and reallocates all of the leaves in a more optimal disk
order.  It is only triggered by the defrag ioctl today.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
new file mode 100644
index 0000000..3e8358c
--- /dev/null
+++ b/fs/btrfs/tree-defrag.c
@@ -0,0 +1,147 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+#include "transaction.h"
+#include "locking.h"
+
+/* defrag all the leaves in a given btree.  If cache_only == 1, don't read
+ * things from disk, otherwise read all the leaves and try to get key order to
+ * better reflect disk order
+ */
+
+int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root, int cache_only)
+{
+	struct btrfs_path *path = NULL;
+	struct btrfs_key key;
+	int ret = 0;
+	int wret;
+	int level;
+	int orig_level;
+	int is_extent = 0;
+	int next_key_ret = 0;
+	u64 last_ret = 0;
+	u64 min_trans = 0;
+
+	if (cache_only)
+		goto out;
+
+	if (root->fs_info->extent_root == root) {
+		/*
+		 * there's recursion here right now in the tree locking,
+		 * we can't defrag the extent root without deadlock
+		 */
+		goto out;
+	}
+
+	if (root->ref_cows == 0 && !is_extent)
+		goto out;
+
+	if (btrfs_test_opt(root, SSD))
+		goto out;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	level = btrfs_header_level(root->node);
+	orig_level = level;
+
+	if (level == 0)
+		goto out;
+
+	if (root->defrag_progress.objectid == 0) {
+		struct extent_buffer *root_node;
+		u32 nritems;
+
+		root_node = btrfs_lock_root_node(root);
+		nritems = btrfs_header_nritems(root_node);
+		root->defrag_max.objectid = 0;
+		/* from above we know this is not a leaf */
+		btrfs_node_key_to_cpu(root_node, &root->defrag_max,
+				      nritems - 1);
+		btrfs_tree_unlock(root_node);
+		free_extent_buffer(root_node);
+		memset(&key, 0, sizeof(key));
+	} else {
+		memcpy(&key, &root->defrag_progress, sizeof(key));
+	}
+
+	path->keep_locks = 1;
+	if (cache_only)
+		min_trans = root->defrag_trans_start;
+
+	ret = btrfs_search_forward(root, &key, NULL, path,
+				   cache_only, min_trans);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = 0;
+		goto out;
+	}
+	btrfs_release_path(root, path);
+	wret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+
+	if (wret < 0) {
+		ret = wret;
+		goto out;
+	}
+	if (!path->nodes[1]) {
+		ret = 0;
+		goto out;
+	}
+	path->slots[1] = btrfs_header_nritems(path->nodes[1]);
+	next_key_ret = btrfs_find_next_key(root, path, &key, 1, cache_only,
+					   min_trans);
+	ret = btrfs_realloc_node(trans, root,
+				 path->nodes[1], 0,
+				 cache_only, &last_ret,
+				 &root->defrag_progress);
+	WARN_ON(ret && ret != -EAGAIN);
+	if (next_key_ret == 0) {
+		memcpy(&root->defrag_progress, &key, sizeof(key));
+		ret = -EAGAIN;
+	}
+
+	btrfs_release_path(root, path);
+	if (is_extent)
+		btrfs_extent_post_op(trans, root);
+out:
+	if (path)
+		btrfs_free_path(path);
+	if (ret == -EAGAIN) {
+		if (root->defrag_max.objectid > root->defrag_progress.objectid)
+			goto done;
+		if (root->defrag_max.type > root->defrag_progress.type)
+			goto done;
+		if (root->defrag_max.offset > root->defrag_progress.offset)
+			goto done;
+		ret = 0;
+	}
+done:
+	if (ret != -EAGAIN) {
+		memset(&root->defrag_progress, 0,
+		       sizeof(root->defrag_progress));
+		root->defrag_trans_start = trans->transid;
+	}
+	return ret;
+}
-- 
1.6.0.2

[PATCH 11/35] Btrfs sysfs routines

[Chris Mason]

Btrfs had plans to export key information sysfs.  In general this
isn't used today and may soon go away.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
new file mode 100644
index 0000000..a240b6f
--- /dev/null
+++ b/fs/btrfs/sysfs.c
@@ -0,0 +1,269 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/completion.h>
+#include <linux/buffer_head.h>
+#include <linux/module.h>
+#include <linux/kobject.h>
+
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+
+static ssize_t root_blocks_used_show(struct btrfs_root *root, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%llu\n",
+		(unsigned long long)btrfs_root_used(&root->root_item));
+}
+
+static ssize_t root_block_limit_show(struct btrfs_root *root, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%llu\n",
+		(unsigned long long)btrfs_root_limit(&root->root_item));
+}
+
+static ssize_t super_blocks_used_show(struct btrfs_fs_info *fs, char *buf)
+{
+
+	return snprintf(buf, PAGE_SIZE, "%llu\n",
+		(unsigned long long)btrfs_super_bytes_used(&fs->super_copy));
+}
+
+static ssize_t super_total_blocks_show(struct btrfs_fs_info *fs, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%llu\n",
+		(unsigned long long)btrfs_super_total_bytes(&fs->super_copy));
+}
+
+static ssize_t super_blocksize_show(struct btrfs_fs_info *fs, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%llu\n",
+		(unsigned long long)btrfs_super_sectorsize(&fs->super_copy));
+}
+
+/* this is for root attrs (subvols/snapshots) */
+struct btrfs_root_attr {
+	struct attribute attr;
+	ssize_t (*show)(struct btrfs_root *, char *);
+	ssize_t (*store)(struct btrfs_root *, const char *, size_t);
+};
+
+#define ROOT_ATTR(name, mode, show, store) \
+static struct btrfs_root_attr btrfs_root_attr_##name = __ATTR(name, mode, \
+							      show, store)
+
+ROOT_ATTR(blocks_used,	0444,	root_blocks_used_show,	NULL);
+ROOT_ATTR(block_limit,	0644,	root_block_limit_show,	NULL);
+
+static struct attribute *btrfs_root_attrs[] = {
+	&btrfs_root_attr_blocks_used.attr,
+	&btrfs_root_attr_block_limit.attr,
+	NULL,
+};
+
+/* this is for super attrs (actual full fs) */
+struct btrfs_super_attr {
+	struct attribute attr;
+	ssize_t (*show)(struct btrfs_fs_info *, char *);
+	ssize_t (*store)(struct btrfs_fs_info *, const char *, size_t);
+};
+
+#define SUPER_ATTR(name, mode, show, store) \
+static struct btrfs_super_attr btrfs_super_attr_##name = __ATTR(name, mode, \
+								show, store)
+
+SUPER_ATTR(blocks_used,		0444,	super_blocks_used_show,		NULL);
+SUPER_ATTR(total_blocks,	0444,	super_total_blocks_show,	NULL);
+SUPER_ATTR(blocksize,		0444,	super_blocksize_show,		NULL);
+
+static struct attribute *btrfs_super_attrs[] = {
+	&btrfs_super_attr_blocks_used.attr,
+	&btrfs_super_attr_total_blocks.attr,
+	&btrfs_super_attr_blocksize.attr,
+	NULL,
+};
+
+static ssize_t btrfs_super_attr_show(struct kobject *kobj,
+				    struct attribute *attr, char *buf)
+{
+	struct btrfs_fs_info *fs = container_of(kobj, struct btrfs_fs_info,
+						super_kobj);
+	struct btrfs_super_attr *a = container_of(attr,
+						  struct btrfs_super_attr,
+						  attr);
+
+	return a->show ? a->show(fs, buf) : 0;
+}
+
+static ssize_t btrfs_super_attr_store(struct kobject *kobj,
+				     struct attribute *attr,
+				     const char *buf, size_t len)
+{
+	struct btrfs_fs_info *fs = container_of(kobj, struct btrfs_fs_info,
+						super_kobj);
+	struct btrfs_super_attr *a = container_of(attr,
+						  struct btrfs_super_attr,
+						  attr);
+
+	return a->store ? a->store(fs, buf, len) : 0;
+}
+
+static ssize_t btrfs_root_attr_show(struct kobject *kobj,
+				    struct attribute *attr, char *buf)
+{
+	struct btrfs_root *root = container_of(kobj, struct btrfs_root,
+						root_kobj);
+	struct btrfs_root_attr *a = container_of(attr,
+						 struct btrfs_root_attr,
+						 attr);
+
+	return a->show ? a->show(root, buf) : 0;
+}
+
+static ssize_t btrfs_root_attr_store(struct kobject *kobj,
+				     struct attribute *attr,
+				     const char *buf, size_t len)
+{
+	struct btrfs_root *root = container_of(kobj, struct btrfs_root,
+						root_kobj);
+	struct btrfs_root_attr *a = container_of(attr,
+						 struct btrfs_root_attr,
+						 attr);
+	return a->store ? a->store(root, buf, len) : 0;
+}
+
+static void btrfs_super_release(struct kobject *kobj)
+{
+	struct btrfs_fs_info *fs = container_of(kobj, struct btrfs_fs_info,
+						super_kobj);
+	complete(&fs->kobj_unregister);
+}
+
+static void btrfs_root_release(struct kobject *kobj)
+{
+	struct btrfs_root *root = container_of(kobj, struct btrfs_root,
+						root_kobj);
+	complete(&root->kobj_unregister);
+}
+
+static struct sysfs_ops btrfs_super_attr_ops = {
+	.show	= btrfs_super_attr_show,
+	.store	= btrfs_super_attr_store,
+};
+
+static struct sysfs_ops btrfs_root_attr_ops = {
+	.show	= btrfs_root_attr_show,
+	.store	= btrfs_root_attr_store,
+};
+
+static struct kobj_type btrfs_root_ktype = {
+	.default_attrs	= btrfs_root_attrs,
+	.sysfs_ops	= &btrfs_root_attr_ops,
+	.release	= btrfs_root_release,
+};
+
+static struct kobj_type btrfs_super_ktype = {
+	.default_attrs	= btrfs_super_attrs,
+	.sysfs_ops	= &btrfs_super_attr_ops,
+	.release	= btrfs_super_release,
+};
+
+/* /sys/fs/btrfs/ entry */
+static struct kset *btrfs_kset;
+
+int btrfs_sysfs_add_super(struct btrfs_fs_info *fs)
+{
+	int error;
+	char *name;
+	char c;
+	int len = strlen(fs->sb->s_id) + 1;
+	int i;
+
+	name = kmalloc(len, GFP_NOFS);
+	if (!name) {
+		error = -ENOMEM;
+		goto fail;
+	}
+
+	for (i = 0; i < len; i++) {
+		c = fs->sb->s_id[i];
+		if (c == '/' || c == '\\')
+			c = '!';
+		name[i] = c;
+	}
+	name[len] = '\0';
+
+	fs->super_kobj.kset = btrfs_kset;
+	error = kobject_init_and_add(&fs->super_kobj, &btrfs_super_ktype,
+				     NULL, "%s", name);
+	kfree(name);
+	if (error)
+		goto fail;
+
+	return 0;
+
+fail:
+	printk(KERN_ERR "btrfs: sysfs creation for super failed\n");
+	return error;
+}
+
+int btrfs_sysfs_add_root(struct btrfs_root *root)
+{
+	int error;
+
+	error = kobject_init_and_add(&root->root_kobj, &btrfs_root_ktype,
+				     &root->fs_info->super_kobj,
+				     "%s", root->name);
+	if (error)
+		goto fail;
+
+	return 0;
+
+fail:
+	printk(KERN_ERR "btrfs: sysfs creation for root failed\n");
+	return error;
+}
+
+void btrfs_sysfs_del_root(struct btrfs_root *root)
+{
+	kobject_put(&root->root_kobj);
+	wait_for_completion(&root->kobj_unregister);
+}
+
+void btrfs_sysfs_del_super(struct btrfs_fs_info *fs)
+{
+	kobject_put(&fs->super_kobj);
+	wait_for_completion(&fs->kobj_unregister);
+}
+
+int btrfs_init_sysfs(void)
+{
+	btrfs_kset = kset_create_and_add("btrfs", NULL, fs_kobj);
+	if (!btrfs_kset)
+		return -ENOMEM;
+	return 0;
+}
+
+void btrfs_exit_sysfs(void)
+{
+	kset_unregister(btrfs_kset);
+}
+
-- 
1.6.0.2

[PATCH 12/35] Btrfs super operations

[Chris Mason]

This provides the normal super operations for btrfs.

It also has the ioctl used to scan devices for btrfs filesystems.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
new file mode 100644
index 0000000..b4c101d
--- /dev/null
+++ b/fs/btrfs/super.c
@@ -0,0 +1,720 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/blkdev.h>
+#include <linux/module.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/highmem.h>
+#include <linux/time.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/smp_lock.h>
+#include <linux/backing-dev.h>
+#include <linux/mount.h>
+#include <linux/mpage.h>
+#include <linux/swap.h>
+#include <linux/writeback.h>
+#include <linux/statfs.h>
+#include <linux/compat.h>
+#include <linux/parser.h>
+#include <linux/ctype.h>
+#include <linux/namei.h>
+#include <linux/miscdevice.h>
+#include <linux/version.h>
+#include "compat.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "btrfs_inode.h"
+#include "ioctl.h"
+#include "print-tree.h"
+#include "xattr.h"
+#include "volumes.h"
+#include "version.h"
+#include "export.h"
+#include "compression.h"
+
+#define BTRFS_SUPER_MAGIC 0x9123683E
+
+static struct super_operations btrfs_super_ops;
+
+static void btrfs_put_super(struct super_block *sb)
+{
+	struct btrfs_root *root = btrfs_sb(sb);
+	int ret;
+
+	ret = close_ctree(root);
+	sb->s_fs_info = NULL;
+}
+
+enum {
+	Opt_degraded, Opt_subvol, Opt_device, Opt_nodatasum, Opt_nodatacow,
+	Opt_max_extent, Opt_max_inline, Opt_alloc_start, Opt_nobarrier,
+	Opt_ssd, Opt_thread_pool, Opt_noacl,  Opt_compress, Opt_err,
+};
+
+static match_table_t tokens = {
+	{Opt_degraded, "degraded"},
+	{Opt_subvol, "subvol=%s"},
+	{Opt_device, "device=%s"},
+	{Opt_nodatasum, "nodatasum"},
+	{Opt_nodatacow, "nodatacow"},
+	{Opt_nobarrier, "nobarrier"},
+	{Opt_max_extent, "max_extent=%s"},
+	{Opt_max_inline, "max_inline=%s"},
+	{Opt_alloc_start, "alloc_start=%s"},
+	{Opt_thread_pool, "thread_pool=%d"},
+	{Opt_compress, "compress"},
+	{Opt_ssd, "ssd"},
+	{Opt_noacl, "noacl"},
+	{Opt_err, NULL},
+};
+
+u64 btrfs_parse_size(char *str)
+{
+	u64 res;
+	int mult = 1;
+	char *end;
+	char last;
+
+	res = simple_strtoul(str, &end, 10);
+
+	last = end[0];
+	if (isalpha(last)) {
+		last = tolower(last);
+		switch (last) {
+		case 'g':
+			mult *= 1024;
+		case 'm':
+			mult *= 1024;
+		case 'k':
+			mult *= 1024;
+		}
+		res = res * mult;
+	}
+	return res;
+}
+
+/*
+ * Regular mount options parser.  Everything that is needed only when
+ * reading in a new superblock is parsed here.
+ */
+int btrfs_parse_options(struct btrfs_root *root, char *options)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	substring_t args[MAX_OPT_ARGS];
+	char *p, *num;
+	int intarg;
+
+	if (!options)
+		return 0;
+
+	/*
+	 * strsep changes the string, duplicate it because parse_options
+	 * gets called twice
+	 */
+	options = kstrdup(options, GFP_NOFS);
+	if (!options)
+		return -ENOMEM;
+
+
+	while ((p = strsep(&options, ",")) != NULL) {
+		int token;
+		if (!*p)
+			continue;
+
+		token = match_token(p, tokens, args);
+		switch (token) {
+		case Opt_degraded:
+			printk(KERN_INFO "btrfs: allowing degraded mounts\n");
+			btrfs_set_opt(info->mount_opt, DEGRADED);
+			break;
+		case Opt_subvol:
+		case Opt_device:
+			/*
+			 * These are parsed by btrfs_parse_early_options
+			 * and can be happily ignored here.
+			 */
+			break;
+		case Opt_nodatasum:
+			printk(KERN_INFO "btrfs: setting nodatacsum\n");
+			btrfs_set_opt(info->mount_opt, NODATASUM);
+			break;
+		case Opt_nodatacow:
+			printk(KERN_INFO "btrfs: setting nodatacow\n");
+			btrfs_set_opt(info->mount_opt, NODATACOW);
+			btrfs_set_opt(info->mount_opt, NODATASUM);
+			break;
+		case Opt_compress:
+			printk(KERN_INFO "btrfs: use compression\n");
+			btrfs_set_opt(info->mount_opt, COMPRESS);
+			break;
+		case Opt_ssd:
+			printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
+			btrfs_set_opt(info->mount_opt, SSD);
+			break;
+		case Opt_nobarrier:
+			printk(KERN_INFO "btrfs: turning off barriers\n");
+			btrfs_set_opt(info->mount_opt, NOBARRIER);
+			break;
+		case Opt_thread_pool:
+			intarg = 0;
+			match_int(&args[0], &intarg);
+			if (intarg) {
+				info->thread_pool_size = intarg;
+				printk(KERN_INFO "btrfs: thread pool %d\n",
+				       info->thread_pool_size);
+			}
+			break;
+		case Opt_max_extent:
+			num = match_strdup(&args[0]);
+			if (num) {
+				info->max_extent = btrfs_parse_size(num);
+				kfree(num);
+
+				info->max_extent = max_t(u64,
+					info->max_extent, root->sectorsize);
+				printk(KERN_INFO "btrfs: max_extent at %llu\n",
+				       info->max_extent);
+			}
+			break;
+		case Opt_max_inline:
+			num = match_strdup(&args[0]);
+			if (num) {
+				info->max_inline = btrfs_parse_size(num);
+				kfree(num);
+
+				if (info->max_inline) {
+					info->max_inline = max_t(u64,
+						info->max_inline,
+						root->sectorsize);
+				}
+				printk(KERN_INFO "btrfs: max_inline at %llu\n",
+					info->max_inline);
+			}
+			break;
+		case Opt_alloc_start:
+			num = match_strdup(&args[0]);
+			if (num) {
+				info->alloc_start = btrfs_parse_size(num);
+				kfree(num);
+				printk(KERN_INFO
+					"btrfs: allocations start at %llu\n",
+					info->alloc_start);
+			}
+			break;
+		case Opt_noacl:
+			root->fs_info->sb->s_flags &= ~MS_POSIXACL;
+			break;
+		default:
+			break;
+		}
+	}
+	kfree(options);
+	return 0;
+}
+
+/*
+ * Parse mount options that are required early in the mount process.
+ *
+ * All other options will be parsed on much later in the mount process and
+ * only when we need to allocate a new super block.
+ */
+static int btrfs_parse_early_options(const char *options, fmode_t flags,
+		void *holder, char **subvol_name,
+		struct btrfs_fs_devices **fs_devices)
+{
+	substring_t args[MAX_OPT_ARGS];
+	char *opts, *p;
+	int error = 0;
+
+	if (!options)
+		goto out;
+
+	/*
+	 * strsep changes the string, duplicate it because parse_options
+	 * gets called twice
+	 */
+	opts = kstrdup(options, GFP_KERNEL);
+	if (!opts)
+		return -ENOMEM;
+
+	while ((p = strsep(&opts, ",")) != NULL) {
+		int token;
+		if (!*p)
+			continue;
+
+		token = match_token(p, tokens, args);
+		switch (token) {
+		case Opt_subvol:
+			*subvol_name = match_strdup(&args[0]);
+			break;
+		case Opt_device:
+			error = btrfs_scan_one_device(match_strdup(&args[0]),
+					flags, holder, fs_devices);
+			if (error)
+				goto out_free_opts;
+			break;
+		default:
+			break;
+		}
+	}
+
+ out_free_opts:
+	kfree(opts);
+ out:
+	/*
+	 * If no subvolume name is specified we use the default one.  Allocate
+	 * a copy of the string "." here so that code later in the
+	 * mount path doesn't care if it's the default volume or another one.
+	 */
+	if (!*subvol_name) {
+		*subvol_name = kstrdup(".", GFP_KERNEL);
+		if (!*subvol_name)
+			return -ENOMEM;
+	}
+	return error;
+}
+
+static int btrfs_fill_super(struct super_block *sb,
+			    struct btrfs_fs_devices *fs_devices,
+			    void *data, int silent)
+{
+	struct inode *inode;
+	struct dentry *root_dentry;
+	struct btrfs_super_block *disk_super;
+	struct btrfs_root *tree_root;
+	struct btrfs_inode *bi;
+	int err;
+
+	sb->s_maxbytes = MAX_LFS_FILESIZE;
+	sb->s_magic = BTRFS_SUPER_MAGIC;
+	sb->s_op = &btrfs_super_ops;
+	sb->s_export_op = &btrfs_export_ops;
+	sb->s_xattr = btrfs_xattr_handlers;
+	sb->s_time_gran = 1;
+	sb->s_flags |= MS_POSIXACL;
+
+	tree_root = open_ctree(sb, fs_devices, (char *)data);
+
+	if (IS_ERR(tree_root)) {
+		printk("btrfs: open_ctree failed\n");
+		return PTR_ERR(tree_root);
+	}
+	sb->s_fs_info = tree_root;
+	disk_super = &tree_root->fs_info->super_copy;
+	inode = btrfs_iget_locked(sb, BTRFS_FIRST_FREE_OBJECTID,
+				  tree_root->fs_info->fs_root);
+	bi = BTRFS_I(inode);
+	bi->location.objectid = inode->i_ino;
+	bi->location.offset = 0;
+	bi->root = tree_root->fs_info->fs_root;
+
+	btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
+
+	if (!inode) {
+		err = -ENOMEM;
+		goto fail_close;
+	}
+	if (inode->i_state & I_NEW) {
+		btrfs_read_locked_inode(inode);
+		unlock_new_inode(inode);
+	}
+
+	root_dentry = d_alloc_root(inode);
+	if (!root_dentry) {
+		iput(inode);
+		err = -ENOMEM;
+		goto fail_close;
+	}
+#if 0
+	/* this does the super kobj at the same time */
+	err = btrfs_sysfs_add_super(tree_root->fs_info);
+	if (err)
+		goto fail_close;
+#endif
+
+	sb->s_root = root_dentry;
+
+	save_mount_options(sb, data);
+	return 0;
+
+fail_close:
+	close_ctree(tree_root);
+	return err;
+}
+
+int btrfs_sync_fs(struct super_block *sb, int wait)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root;
+	int ret;
+	root = btrfs_sb(sb);
+
+	if (sb->s_flags & MS_RDONLY)
+		return 0;
+
+	sb->s_dirt = 0;
+	if (!wait) {
+		filemap_flush(root->fs_info->btree_inode->i_mapping);
+		return 0;
+	}
+
+	btrfs_start_delalloc_inodes(root);
+	btrfs_wait_ordered_extents(root, 0);
+
+	btrfs_clean_old_snapshots(root);
+	trans = btrfs_start_transaction(root, 1);
+	ret = btrfs_commit_transaction(trans, root);
+	sb->s_dirt = 0;
+	return ret;
+}
+
+static void btrfs_write_super(struct super_block *sb)
+{
+	sb->s_dirt = 0;
+}
+
+static int btrfs_test_super(struct super_block *s, void *data)
+{
+	struct btrfs_fs_devices *test_fs_devices = data;
+	struct btrfs_root *root = btrfs_sb(s);
+
+	return root->fs_info->fs_devices == test_fs_devices;
+}
+
+/*
+ * Find a superblock for the given device / mount point.
+ *
+ * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
+ *	  for multiple device setup.  Make sure to keep it in sync.
+ */
+static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
+		const char *dev_name, void *data, struct vfsmount *mnt)
+{
+	char *subvol_name = NULL;
+	struct block_device *bdev = NULL;
+	struct super_block *s;
+	struct dentry *root;
+	struct btrfs_fs_devices *fs_devices = NULL;
+	fmode_t mode = FMODE_READ;
+	int error = 0;
+
+	if (!(flags & MS_RDONLY))
+		mode |= FMODE_WRITE;
+
+	error = btrfs_parse_early_options(data, mode, fs_type,
+					  &subvol_name, &fs_devices);
+	if (error)
+		return error;
+
+	error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
+	if (error)
+		goto error_free_subvol_name;
+
+	error = btrfs_open_devices(fs_devices, mode, fs_type);
+	if (error)
+		goto error_free_subvol_name;
+
+	if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
+		error = -EACCES;
+		goto error_close_devices;
+	}
+
+	bdev = fs_devices->latest_bdev;
+	s = sget(fs_type, btrfs_test_super, set_anon_super, fs_devices);
+	if (IS_ERR(s))
+		goto error_s;
+
+	if (s->s_root) {
+		if ((flags ^ s->s_flags) & MS_RDONLY) {
+			up_write(&s->s_umount);
+			deactivate_super(s);
+			error = -EBUSY;
+			goto error_close_devices;
+		}
+
+		btrfs_close_devices(fs_devices);
+	} else {
+		char b[BDEVNAME_SIZE];
+
+		s->s_flags = flags;
+		strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
+		error = btrfs_fill_super(s, fs_devices, data,
+					 flags & MS_SILENT ? 1 : 0);
+		if (error) {
+			up_write(&s->s_umount);
+			deactivate_super(s);
+			goto error_free_subvol_name;
+		}
+
+		btrfs_sb(s)->fs_info->bdev_holder = fs_type;
+		s->s_flags |= MS_ACTIVE;
+	}
+
+	if (!strcmp(subvol_name, "."))
+		root = dget(s->s_root);
+	else {
+		mutex_lock(&s->s_root->d_inode->i_mutex);
+		root = lookup_one_len(subvol_name, s->s_root,
+				      strlen(subvol_name));
+		mutex_unlock(&s->s_root->d_inode->i_mutex);
+
+		if (IS_ERR(root)) {
+			up_write(&s->s_umount);
+			deactivate_super(s);
+			error = PTR_ERR(root);
+			goto error_free_subvol_name;
+		}
+		if (!root->d_inode) {
+			dput(root);
+			up_write(&s->s_umount);
+			deactivate_super(s);
+			error = -ENXIO;
+			goto error_free_subvol_name;
+		}
+	}
+
+	mnt->mnt_sb = s;
+	mnt->mnt_root = root;
+
+	kfree(subvol_name);
+	return 0;
+
+error_s:
+	error = PTR_ERR(s);
+error_close_devices:
+	btrfs_close_devices(fs_devices);
+error_free_subvol_name:
+	kfree(subvol_name);
+	return error;
+}
+
+static int btrfs_remount(struct super_block *sb, int *flags, char *data)
+{
+	struct btrfs_root *root = btrfs_sb(sb);
+	int ret;
+
+	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
+		return 0;
+
+	if (*flags & MS_RDONLY) {
+		sb->s_flags |= MS_RDONLY;
+
+		ret =  btrfs_commit_super(root);
+		WARN_ON(ret);
+	} else {
+		if (root->fs_info->fs_devices->rw_devices == 0)
+			return -EACCES;
+
+		if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
+			return -EINVAL;
+
+		ret = btrfs_cleanup_reloc_trees(root);
+		WARN_ON(ret);
+
+		ret = btrfs_cleanup_fs_roots(root->fs_info);
+		WARN_ON(ret);
+
+		sb->s_flags &= ~MS_RDONLY;
+	}
+
+	return 0;
+}
+
+static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
+{
+	struct btrfs_root *root = btrfs_sb(dentry->d_sb);
+	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
+	int bits = dentry->d_sb->s_blocksize_bits;
+	__be32 *fsid = (__be32 *)root->fs_info->fsid;
+
+	buf->f_namelen = BTRFS_NAME_LEN;
+	buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
+	buf->f_bfree = buf->f_blocks -
+		(btrfs_super_bytes_used(disk_super) >> bits);
+	buf->f_bavail = buf->f_bfree;
+	buf->f_bsize = dentry->d_sb->s_blocksize;
+	buf->f_type = BTRFS_SUPER_MAGIC;
+
+	/* We treat it as constant endianness (it doesn't matter _which_)
+	   because we want the fsid to come out the same whether mounted
+	   on a big-endian or little-endian host */
+	buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
+	buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
+	/* Mask in the root object ID too, to disambiguate subvols */
+	buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
+	buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
+
+	return 0;
+}
+
+static struct file_system_type btrfs_fs_type = {
+	.owner		= THIS_MODULE,
+	.name		= "btrfs",
+	.get_sb		= btrfs_get_sb,
+	.kill_sb	= kill_anon_super,
+	.fs_flags	= FS_REQUIRES_DEV,
+};
+
+/*
+ * used by btrfsctl to scan devices when no FS is mounted
+ */
+static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
+				unsigned long arg)
+{
+	struct btrfs_ioctl_vol_args *vol;
+	struct btrfs_fs_devices *fs_devices;
+	int ret = 0;
+	int len;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	vol = kmalloc(sizeof(*vol), GFP_KERNEL);
+	if (copy_from_user(vol, (void __user *)arg, sizeof(*vol))) {
+		ret = -EFAULT;
+		goto out;
+	}
+	len = strnlen(vol->name, BTRFS_PATH_NAME_MAX);
+	switch (cmd) {
+	case BTRFS_IOC_SCAN_DEV:
+		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
+					    &btrfs_fs_type, &fs_devices);
+		break;
+	}
+out:
+	kfree(vol);
+	return ret;
+}
+
+static void btrfs_write_super_lockfs(struct super_block *sb)
+{
+	struct btrfs_root *root = btrfs_sb(sb);
+	mutex_lock(&root->fs_info->transaction_kthread_mutex);
+	mutex_lock(&root->fs_info->cleaner_mutex);
+}
+
+static void btrfs_unlockfs(struct super_block *sb)
+{
+	struct btrfs_root *root = btrfs_sb(sb);
+	mutex_unlock(&root->fs_info->cleaner_mutex);
+	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
+}
+
+static struct super_operations btrfs_super_ops = {
+	.delete_inode	= btrfs_delete_inode,
+	.put_super	= btrfs_put_super,
+	.write_super	= btrfs_write_super,
+	.sync_fs	= btrfs_sync_fs,
+	.show_options	= generic_show_options,
+	.write_inode	= btrfs_write_inode,
+	.dirty_inode	= btrfs_dirty_inode,
+	.alloc_inode	= btrfs_alloc_inode,
+	.destroy_inode	= btrfs_destroy_inode,
+	.statfs		= btrfs_statfs,
+	.remount_fs	= btrfs_remount,
+	.write_super_lockfs = btrfs_write_super_lockfs,
+	.unlockfs	= btrfs_unlockfs,
+};
+
+static const struct file_operations btrfs_ctl_fops = {
+	.unlocked_ioctl	 = btrfs_control_ioctl,
+	.compat_ioctl = btrfs_control_ioctl,
+	.owner	 = THIS_MODULE,
+};
+
+static struct miscdevice btrfs_misc = {
+	.minor		= MISC_DYNAMIC_MINOR,
+	.name		= "btrfs-control",
+	.fops		= &btrfs_ctl_fops
+};
+
+static int btrfs_interface_init(void)
+{
+	return misc_register(&btrfs_misc);
+}
+
+static void btrfs_interface_exit(void)
+{
+	if (misc_deregister(&btrfs_misc) < 0)
+		printk(KERN_INFO "misc_deregister failed for control device");
+}
+
+static int __init init_btrfs_fs(void)
+{
+	int err;
+
+	err = btrfs_init_sysfs();
+	if (err)
+		return err;
+
+	err = btrfs_init_cachep();
+	if (err)
+		goto free_sysfs;
+
+	err = extent_io_init();
+	if (err)
+		goto free_cachep;
+
+	err = extent_map_init();
+	if (err)
+		goto free_extent_io;
+
+	err = btrfs_interface_init();
+	if (err)
+		goto free_extent_map;
+
+	err = register_filesystem(&btrfs_fs_type);
+	if (err)
+		goto unregister_ioctl;
+
+	printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
+	return 0;
+
+unregister_ioctl:
+	btrfs_interface_exit();
+free_extent_map:
+	extent_map_exit();
+free_extent_io:
+	extent_io_exit();
+free_cachep:
+	btrfs_destroy_cachep();
+free_sysfs:
+	btrfs_exit_sysfs();
+	return err;
+}
+
+static void __exit exit_btrfs_fs(void)
+{
+	btrfs_destroy_cachep();
+	extent_map_exit();
+	extent_io_exit();
+	btrfs_interface_exit();
+	unregister_filesystem(&btrfs_fs_type);
+	btrfs_exit_sysfs();
+	btrfs_cleanup_fs_uuids();
+	btrfs_zlib_exit();
+}
+
+module_init(init_btrfs_fs)
+module_exit(exit_btrfs_fs)
+
+MODULE_LICENSE("GPL");
-- 
1.6.0.2

[PATCH 13/35] Btrfs metadata access routines

[Chris Mason]

This is some cpp abuse that is used to generate functions.
to read and modify the extent buffers.  Look for 'deeply nasty code' below
for a longer description of how it all works.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/struct-funcs.c b/fs/btrfs/struct-funcs.c
new file mode 100644
index 0000000..c0f7eca
--- /dev/null
+++ b/fs/btrfs/struct-funcs.c
@@ -0,0 +1,139 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/highmem.h>
+
+/* this is some deeply nasty code.  ctree.h has a different
+ * definition for this BTRFS_SETGET_FUNCS macro, behind a #ifndef
+ *
+ * The end result is that anyone who #includes ctree.h gets a
+ * declaration for the btrfs_set_foo functions and btrfs_foo functions
+ *
+ * This file declares the macros and then #includes ctree.h, which results
+ * in cpp creating the function here based on the template below.
+ *
+ * These setget functions do all the extent_buffer related mapping
+ * required to efficiently read and write specific fields in the extent
+ * buffers.  Every pointer to metadata items in btrfs is really just
+ * an unsigned long offset into the extent buffer which has been
+ * cast to a specific type.  This gives us all the gcc type checking.
+ *
+ * The extent buffer api is used to do all the kmapping and page
+ * spanning work required to get extent buffers in highmem and have
+ * a metadata blocksize different from the page size.
+ *
+ * The macro starts with a simple function prototype declaration so that
+ * sparse won't complain about it being static.
+ */
+
+#define BTRFS_SETGET_FUNCS(name, type, member, bits)			\
+u##bits btrfs_##name(struct extent_buffer *eb, type *s);		\
+void btrfs_set_##name(struct extent_buffer *eb, type *s, u##bits val);	\
+u##bits btrfs_##name(struct extent_buffer *eb,				\
+				   type *s)				\
+{									\
+	unsigned long part_offset = (unsigned long)s;			\
+	unsigned long offset = part_offset + offsetof(type, member);	\
+	type *p;							\
+	/* ugly, but we want the fast path here */			\
+	if (eb->map_token && offset >= eb->map_start &&			\
+	    offset + sizeof(((type *)0)->member) <= eb->map_start +	\
+	    eb->map_len) {						\
+		p = (type *)(eb->kaddr + part_offset - eb->map_start);	\
+		return le##bits##_to_cpu(p->member);			\
+	}								\
+	{								\
+		int err;						\
+		char *map_token;					\
+		char *kaddr;						\
+		int unmap_on_exit = (eb->map_token == NULL);		\
+		unsigned long map_start;				\
+		unsigned long map_len;					\
+		u##bits res;						\
+		err = map_extent_buffer(eb, offset,			\
+				sizeof(((type *)0)->member),		\
+				&map_token, &kaddr,			\
+				&map_start, &map_len, KM_USER1);	\
+		if (err) {						\
+			__le##bits leres;				\
+			read_eb_member(eb, s, type, member, &leres);	\
+			return le##bits##_to_cpu(leres);		\
+		}							\
+		p = (type *)(kaddr + part_offset - map_start);		\
+		res = le##bits##_to_cpu(p->member);			\
+		if (unmap_on_exit)					\
+			unmap_extent_buffer(eb, map_token, KM_USER1);	\
+		return res;						\
+	}								\
+}									\
+void btrfs_set_##name(struct extent_buffer *eb,				\
+				    type *s, u##bits val)		\
+{									\
+	unsigned long part_offset = (unsigned long)s;			\
+	unsigned long offset = part_offset + offsetof(type, member);	\
+	type *p;							\
+	/* ugly, but we want the fast path here */			\
+	if (eb->map_token && offset >= eb->map_start &&			\
+	    offset + sizeof(((type *)0)->member) <= eb->map_start +	\
+	    eb->map_len) {						\
+		p = (type *)(eb->kaddr + part_offset - eb->map_start);	\
+		p->member = cpu_to_le##bits(val);			\
+		return;							\
+	}								\
+	{								\
+		int err;						\
+		char *map_token;					\
+		char *kaddr;						\
+		int unmap_on_exit = (eb->map_token == NULL);		\
+		unsigned long map_start;				\
+		unsigned long map_len;					\
+		err = map_extent_buffer(eb, offset,			\
+				sizeof(((type *)0)->member),		\
+				&map_token, &kaddr,			\
+				&map_start, &map_len, KM_USER1);	\
+		if (err) {						\
+			__le##bits val2;				\
+			val2 = cpu_to_le##bits(val);			\
+			write_eb_member(eb, s, type, member, &val2);	\
+			return;						\
+		}							\
+		p = (type *)(kaddr + part_offset - map_start);		\
+		p->member = cpu_to_le##bits(val);			\
+		if (unmap_on_exit)					\
+			unmap_extent_buffer(eb, map_token, KM_USER1);	\
+	}								\
+}
+
+#include "ctree.h"
+
+void btrfs_node_key(struct extent_buffer *eb,
+		    struct btrfs_disk_key *disk_key, int nr)
+{
+	unsigned long ptr = btrfs_node_key_ptr_offset(nr);
+	if (eb->map_token && ptr >= eb->map_start &&
+	    ptr + sizeof(*disk_key) <= eb->map_start + eb->map_len) {
+		memcpy(disk_key, eb->kaddr + ptr - eb->map_start,
+			sizeof(*disk_key));
+		return;
+	} else if (eb->map_token) {
+		unmap_extent_buffer(eb, eb->map_token, KM_USER1);
+		eb->map_token = NULL;
+	}
+	read_eb_member(eb, (struct btrfs_key_ptr *)ptr,
+		       struct btrfs_key_ptr, key, disk_key);
+}
-- 
1.6.0.2

[PATCH 14/35] Btrfs tree of tree roots code

[Chris Mason]

The tree of tree roots stores pointers to all of the subvolumes and snapshots
on the filesystem.  It may have multiple pointers to each one if there
are snapshots that were in the process of being deleted when a system crashed.

Subvolumes actually point to these items in regular directories.  In general
a subvolume will only point to the latest generation of a snapshot because
any other version is in the process of being deleted.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
new file mode 100644
index 0000000..b48650d
--- /dev/null
+++ b/fs/btrfs/root-tree.c
@@ -0,0 +1,366 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include "ctree.h"
+#include "transaction.h"
+#include "disk-io.h"
+#include "print-tree.h"
+
+/*
+ *  search forward for a root, starting with objectid 'search_start'
+ *  if a root key is found, the objectid we find is filled into 'found_objectid'
+ *  and 0 is returned.  < 0 is returned on error, 1 if there is nothing
+ *  left in the tree.
+ */
+int btrfs_search_root(struct btrfs_root *root, u64 search_start,
+		      u64 *found_objectid)
+{
+	struct btrfs_path *path;
+	struct btrfs_key search_key;
+	int ret;
+
+	root = root->fs_info->tree_root;
+	search_key.objectid = search_start;
+	search_key.type = (u8)-1;
+	search_key.offset = (u64)-1;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+again:
+	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	if (ret == 0) {
+		ret = 1;
+		goto out;
+	}
+	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
+		ret = btrfs_next_leaf(root, path);
+		if (ret)
+			goto out;
+	}
+	btrfs_item_key_to_cpu(path->nodes[0], &search_key, path->slots[0]);
+	if (search_key.type != BTRFS_ROOT_ITEM_KEY) {
+		search_key.offset++;
+		btrfs_release_path(root, path);
+		goto again;
+	}
+	ret = 0;
+	*found_objectid = search_key.objectid;
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * lookup the root with the highest offset for a given objectid.  The key we do
+ * find is copied into 'key'.  If we find something return 0, otherwise 1, < 0
+ * on error.
+ */
+int btrfs_find_last_root(struct btrfs_root *root, u64 objectid,
+			struct btrfs_root_item *item, struct btrfs_key *key)
+{
+	struct btrfs_path *path;
+	struct btrfs_key search_key;
+	struct btrfs_key found_key;
+	struct extent_buffer *l;
+	int ret;
+	int slot;
+
+	search_key.objectid = objectid;
+	search_key.type = BTRFS_ROOT_ITEM_KEY;
+	search_key.offset = (u64)-1;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+
+	BUG_ON(ret == 0);
+	l = path->nodes[0];
+	BUG_ON(path->slots[0] == 0);
+	slot = path->slots[0] - 1;
+	btrfs_item_key_to_cpu(l, &found_key, slot);
+	if (found_key.objectid != objectid) {
+		ret = 1;
+		goto out;
+	}
+	read_extent_buffer(l, item, btrfs_item_ptr_offset(l, slot),
+			   sizeof(*item));
+	memcpy(key, &found_key, sizeof(found_key));
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * copy the data in 'item' into the btree
+ */
+int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, struct btrfs_root_item
+		      *item)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *l;
+	int ret;
+	int slot;
+	unsigned long ptr;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_search_slot(trans, root, key, path, 0, 1);
+	if (ret < 0)
+		goto out;
+
+	if (ret != 0) {
+		btrfs_print_leaf(root, path->nodes[0]);
+		printk(KERN_CRIT "unable to update root key %llu %u %llu\n",
+		       (unsigned long long)key->objectid, key->type,
+		       (unsigned long long)key->offset);
+		BUG_ON(1);
+	}
+
+	l = path->nodes[0];
+	slot = path->slots[0];
+	ptr = btrfs_item_ptr_offset(l, slot);
+	write_extent_buffer(l, item, ptr, sizeof(*item));
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+out:
+	btrfs_release_path(root, path);
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, struct btrfs_root_item
+		      *item)
+{
+	int ret;
+	ret = btrfs_insert_item(trans, root, key, item, sizeof(*item));
+	return ret;
+}
+
+/*
+ * at mount time we want to find all the old transaction snapshots that were in
+ * the process of being deleted if we crashed.  This is any root item with an
+ * offset lower than the latest root.  They need to be queued for deletion to
+ * finish what was happening when we crashed.
+ */
+int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid,
+			  struct btrfs_root *latest)
+{
+	struct btrfs_root *dead_root;
+	struct btrfs_item *item;
+	struct btrfs_root_item *ri;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct btrfs_path *path;
+	int ret;
+	u32 nritems;
+	struct extent_buffer *leaf;
+	int slot;
+
+	key.objectid = objectid;
+	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+	key.offset = 0;
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+again:
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto err;
+	while (1) {
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		slot = path->slots[0];
+		if (slot >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret)
+				break;
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+			slot = path->slots[0];
+		}
+		item = btrfs_item_nr(leaf, slot);
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+		if (btrfs_key_type(&key) != BTRFS_ROOT_ITEM_KEY)
+			goto next;
+
+		if (key.objectid < objectid)
+			goto next;
+
+		if (key.objectid > objectid)
+			break;
+
+		ri = btrfs_item_ptr(leaf, slot, struct btrfs_root_item);
+		if (btrfs_disk_root_refs(leaf, ri) != 0)
+			goto next;
+
+		memcpy(&found_key, &key, sizeof(key));
+		key.offset++;
+		btrfs_release_path(root, path);
+		dead_root =
+			btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
+						    &found_key);
+		if (IS_ERR(dead_root)) {
+			ret = PTR_ERR(dead_root);
+			goto err;
+		}
+
+		if (objectid == BTRFS_TREE_RELOC_OBJECTID)
+			ret = btrfs_add_dead_reloc_root(dead_root);
+		else
+			ret = btrfs_add_dead_root(dead_root, latest);
+		if (ret)
+			goto err;
+		goto again;
+next:
+		slot++;
+		path->slots[0]++;
+	}
+	ret = 0;
+err:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/* drop the root item for 'key' from 'root' */
+int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_key *key)
+{
+	struct btrfs_path *path;
+	int ret;
+	u32 refs;
+	struct btrfs_root_item *ri;
+	struct extent_buffer *leaf;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_search_slot(trans, root, key, path, -1, 1);
+	if (ret < 0)
+		goto out;
+
+	BUG_ON(ret != 0);
+	leaf = path->nodes[0];
+	ri = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_item);
+
+	refs = btrfs_disk_root_refs(leaf, ri);
+	BUG_ON(refs != 0);
+	ret = btrfs_del_item(trans, root, path);
+out:
+	btrfs_release_path(root, path);
+	btrfs_free_path(path);
+	return ret;
+}
+
+#if 0 /* this will get used when snapshot deletion is implemented */
+int btrfs_del_root_ref(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *tree_root,
+		       u64 root_id, u8 type, u64 ref_id)
+{
+	struct btrfs_key key;
+	int ret;
+	struct btrfs_path *path;
+
+	path = btrfs_alloc_path();
+
+	key.objectid = root_id;
+	key.type = type;
+	key.offset = ref_id;
+
+	ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
+	BUG_ON(ret);
+
+	ret = btrfs_del_item(trans, tree_root, path);
+	BUG_ON(ret);
+
+	btrfs_free_path(path);
+	return ret;
+}
+#endif
+
+int btrfs_find_root_ref(struct btrfs_root *tree_root,
+		   struct btrfs_path *path,
+		   u64 root_id, u64 ref_id)
+{
+	struct btrfs_key key;
+	int ret;
+
+	key.objectid = root_id;
+	key.type = BTRFS_ROOT_REF_KEY;
+	key.offset = ref_id;
+
+	ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
+	return ret;
+}
+
+
+/*
+ * add a btrfs_root_ref item.  type is either BTRFS_ROOT_REF_KEY
+ * or BTRFS_ROOT_BACKREF_KEY.
+ *
+ * The dirid, sequence, name and name_len refer to the directory entry
+ * that is referencing the root.
+ *
+ * For a forward ref, the root_id is the id of the tree referencing
+ * the root and ref_id is the id of the subvol  or snapshot.
+ *
+ * For a back ref the root_id is the id of the subvol or snapshot and
+ * ref_id is the id of the tree referencing it.
+ */
+int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *tree_root,
+		       u64 root_id, u8 type, u64 ref_id,
+		       u64 dirid, u64 sequence,
+		       const char *name, int name_len)
+{
+	struct btrfs_key key;
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_root_ref *ref;
+	struct extent_buffer *leaf;
+	unsigned long ptr;
+
+
+	path = btrfs_alloc_path();
+
+	key.objectid = root_id;
+	key.type = type;
+	key.offset = ref_id;
+
+	ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
+				      sizeof(*ref) + name_len);
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
+	btrfs_set_root_ref_dirid(leaf, ref, dirid);
+	btrfs_set_root_ref_sequence(leaf, ref, sequence);
+	btrfs_set_root_ref_name_len(leaf, ref, name_len);
+	ptr = (unsigned long)(ref + 1);
+	write_extent_buffer(leaf, name, ptr, name_len);
+	btrfs_mark_buffer_dirty(leaf);
+
+	btrfs_free_path(path);
+	return ret;
+}
-- 
1.6.0.2

[PATCH 15/35] Btrfs Orphan prevention

[Chris Mason]

Lost files are prevented by creating an orphan item in the subvolume
for the inode.  This handles lost file prevention and finishes truncates
that were interrupted by a crash.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/orphan.c b/fs/btrfs/orphan.c
new file mode 100644
index 0000000..3c0d52a
--- /dev/null
+++ b/fs/btrfs/orphan.c
@@ -0,0 +1,67 @@
+/*
+ * Copyright (C) 2008 Red Hat.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include "ctree.h"
+#include "disk-io.h"
+
+int btrfs_insert_orphan_item(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 offset)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int ret = 0;
+
+	key.objectid = BTRFS_ORPHAN_OBJECTID;
+	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
+	key.offset = offset;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_del_orphan_item(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, u64 offset)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int ret = 0;
+
+	key.objectid = BTRFS_ORPHAN_OBJECTID;
+	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
+	key.offset = offset;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	ret = btrfs_del_item(trans, root, path);
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
-- 
1.6.0.2

[PATCH 16/35] Btrfs free inode allocation

[Chris Mason]

Today this is much too simple, it just allocates the highest inode + 1.
Longer term something similar to the caching in extent-tree.c needs to
be done.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/inode-map.c b/fs/btrfs/inode-map.c
new file mode 100644
index 0000000..2aa7987
--- /dev/null
+++ b/fs/btrfs/inode-map.c
@@ -0,0 +1,144 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+
+int btrfs_find_highest_inode(struct btrfs_root *root, u64 *objectid)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct extent_buffer *l;
+	struct btrfs_key search_key;
+	struct btrfs_key found_key;
+	int slot;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
+	search_key.type = -1;
+	search_key.offset = (u64)-1;
+	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
+	if (ret < 0)
+		goto error;
+	BUG_ON(ret == 0);
+	if (path->slots[0] > 0) {
+		slot = path->slots[0] - 1;
+		l = path->nodes[0];
+		btrfs_item_key_to_cpu(l, &found_key, slot);
+		*objectid = found_key.objectid;
+	} else {
+		*objectid = BTRFS_FIRST_FREE_OBJECTID;
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * walks the btree of allocated inodes and find a hole.
+ */
+int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     u64 dirid, u64 *objectid)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int ret;
+	int slot = 0;
+	u64 last_ino = 0;
+	int start_found;
+	struct extent_buffer *l;
+	struct btrfs_key search_key;
+	u64 search_start = dirid;
+
+	mutex_lock(&root->objectid_mutex);
+	if (root->last_inode_alloc >= BTRFS_FIRST_FREE_OBJECTID &&
+	    root->last_inode_alloc < BTRFS_LAST_FREE_OBJECTID) {
+		*objectid = ++root->last_inode_alloc;
+		mutex_unlock(&root->objectid_mutex);
+		return 0;
+	}
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	search_start = max(search_start, BTRFS_FIRST_FREE_OBJECTID);
+	search_key.objectid = search_start;
+	search_key.type = 0;
+	search_key.offset = 0;
+
+	btrfs_init_path(path);
+	start_found = 0;
+	ret = btrfs_search_slot(trans, root, &search_key, path, 0, 0);
+	if (ret < 0)
+		goto error;
+
+	while (1) {
+		l = path->nodes[0];
+		slot = path->slots[0];
+		if (slot >= btrfs_header_nritems(l)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret == 0)
+				continue;
+			if (ret < 0)
+				goto error;
+			if (!start_found) {
+				*objectid = search_start;
+				start_found = 1;
+				goto found;
+			}
+			*objectid = last_ino > search_start ?
+				last_ino : search_start;
+			goto found;
+		}
+		btrfs_item_key_to_cpu(l, &key, slot);
+		if (key.objectid >= search_start) {
+			if (start_found) {
+				if (last_ino < search_start)
+					last_ino = search_start;
+				if (key.objectid > last_ino) {
+					*objectid = last_ino;
+					goto found;
+				}
+			} else if (key.objectid > search_start) {
+				*objectid = search_start;
+				goto found;
+			}
+		}
+		if (key.objectid >= BTRFS_LAST_FREE_OBJECTID)
+			break;
+
+		start_found = 1;
+		last_ino = key.objectid + 1;
+		path->slots[0]++;
+	}
+	BUG_ON(1);
+found:
+	btrfs_release_path(root, path);
+	btrfs_free_path(path);
+	BUG_ON(*objectid < search_start);
+	mutex_unlock(&root->objectid_mutex);
+	return 0;
+error:
+	btrfs_release_path(root, path);
+	btrfs_free_path(path);
+	mutex_unlock(&root->objectid_mutex);
+	return ret;
+}
-- 
1.6.0.2

[PATCH 17/35] Btrfs inode item routines

[Chris Mason]

Inode items hold all of the traditional data returned by stat in
the btree.  There is a back reference item stored in the btree for
each directory item that points to an inode.

This has routines to maintain the inode items and their backreferences.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/inode-item.c b/fs/btrfs/inode-item.c
new file mode 100644
index 0000000..3d46fa1
--- /dev/null
+++ b/fs/btrfs/inode-item.c
@@ -0,0 +1,206 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+
+static int find_name_in_backref(struct btrfs_path *path, const char *name,
+			 int name_len, struct btrfs_inode_ref **ref_ret)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_inode_ref *ref;
+	unsigned long ptr;
+	unsigned long name_ptr;
+	u32 item_size;
+	u32 cur_offset = 0;
+	int len;
+
+	leaf = path->nodes[0];
+	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	while (cur_offset < item_size) {
+		ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
+		len = btrfs_inode_ref_name_len(leaf, ref);
+		name_ptr = (unsigned long)(ref + 1);
+		cur_offset += len + sizeof(*ref);
+		if (len != name_len)
+			continue;
+		if (memcmp_extent_buffer(leaf, name, name_ptr, name_len) == 0) {
+			*ref_ret = ref;
+			return 1;
+		}
+	}
+	return 0;
+}
+
+int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   const char *name, int name_len,
+			   u64 inode_objectid, u64 ref_objectid, u64 *index)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_inode_ref *ref;
+	struct extent_buffer *leaf;
+	unsigned long ptr;
+	unsigned long item_start;
+	u32 item_size;
+	u32 sub_item_len;
+	int ret;
+	int del_len = name_len + sizeof(*ref);
+
+	key.objectid = inode_objectid;
+	key.offset = ref_objectid;
+	btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	} else if (ret < 0) {
+		goto out;
+	}
+	if (!find_name_in_backref(path, name, name_len, &ref)) {
+		ret = -ENOENT;
+		goto out;
+	}
+	leaf = path->nodes[0];
+	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+
+	if (index)
+		*index = btrfs_inode_ref_index(leaf, ref);
+
+	if (del_len == item_size) {
+		ret = btrfs_del_item(trans, root, path);
+		goto out;
+	}
+	ptr = (unsigned long)ref;
+	sub_item_len = name_len + sizeof(*ref);
+	item_start = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	memmove_extent_buffer(leaf, ptr, ptr + sub_item_len,
+			      item_size - (ptr + sub_item_len - item_start));
+	ret = btrfs_truncate_item(trans, root, path,
+				  item_size - sub_item_len, 1);
+	BUG_ON(ret);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   const char *name, int name_len,
+			   u64 inode_objectid, u64 ref_objectid, u64 index)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_inode_ref *ref;
+	unsigned long ptr;
+	int ret;
+	int ins_len = name_len + sizeof(*ref);
+
+	key.objectid = inode_objectid;
+	key.offset = ref_objectid;
+	btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key,
+				      ins_len);
+	if (ret == -EEXIST) {
+		u32 old_size;
+
+		if (find_name_in_backref(path, name, name_len, &ref))
+			goto out;
+
+		old_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
+		ret = btrfs_extend_item(trans, root, path, ins_len);
+		BUG_ON(ret);
+		ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
+				     struct btrfs_inode_ref);
+		ref = (struct btrfs_inode_ref *)((unsigned long)ref + old_size);
+		btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
+		btrfs_set_inode_ref_index(path->nodes[0], ref, index);
+		ptr = (unsigned long)(ref + 1);
+		ret = 0;
+	} else if (ret < 0) {
+		goto out;
+	} else {
+		ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
+				     struct btrfs_inode_ref);
+		btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
+		btrfs_set_inode_ref_index(path->nodes[0], ref, index);
+		ptr = (unsigned long)(ref + 1);
+	}
+	write_extent_buffer(path->nodes[0], name, ptr, name_len);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_insert_empty_inode(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     struct btrfs_path *path, u64 objectid)
+{
+	struct btrfs_key key;
+	int ret;
+	key.objectid = objectid;
+	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
+	key.offset = 0;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key,
+				      sizeof(struct btrfs_inode_item));
+	if (ret == 0 && objectid > root->highest_inode)
+		root->highest_inode = objectid;
+	return ret;
+}
+
+int btrfs_lookup_inode(struct btrfs_trans_handle *trans, struct btrfs_root
+		       *root, struct btrfs_path *path,
+		       struct btrfs_key *location, int mod)
+{
+	int ins_len = mod < 0 ? -1 : 0;
+	int cow = mod != 0;
+	int ret;
+	int slot;
+	struct extent_buffer *leaf;
+	struct btrfs_key found_key;
+
+	ret = btrfs_search_slot(trans, root, location, path, ins_len, cow);
+	if (ret > 0 && btrfs_key_type(location) == BTRFS_ROOT_ITEM_KEY &&
+	    location->offset == (u64)-1 && path->slots[0] != 0) {
+		slot = path->slots[0] - 1;
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+		if (found_key.objectid == location->objectid &&
+		    btrfs_key_type(&found_key) == btrfs_key_type(location)) {
+			path->slots[0]--;
+			return 0;
+		}
+	}
+	return ret;
+}
-- 
1.6.0.2

[PATCH 19/35] Btrfs directory name hashing

[Chris Mason]

This used to do more, but today it is just crc32c.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/hash.h b/fs/btrfs/hash.h
new file mode 100644
index 0000000..2a020b2
--- /dev/null
+++ b/fs/btrfs/hash.h
@@ -0,0 +1,27 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __HASH__
+#define __HASH__
+
+#include "crc32c.h"
+static inline u64 btrfs_name_hash(const char *name, int len)
+{
+	return btrfs_crc32c((u32)~1, name, len);
+}
+#endif
-- 
1.6.0.2

[PATCH 20/35] Btrfs free space caching

[Chris Mason]

The free space in btrfs is cached in ram to more efficiently search for
free extents on disk.  This implements the cache for use by extent-tree.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
new file mode 100644
index 0000000..d1e5f0e
--- /dev/null
+++ b/fs/btrfs/free-space-cache.c
@@ -0,0 +1,495 @@
+/*
+ * Copyright (C) 2008 Red Hat.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+
+static int tree_insert_offset(struct rb_root *root, u64 offset,
+			      struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct btrfs_free_space *info;
+
+	while (*p) {
+		parent = *p;
+		info = rb_entry(parent, struct btrfs_free_space, offset_index);
+
+		if (offset < info->offset)
+			p = &(*p)->rb_left;
+		else if (offset > info->offset)
+			p = &(*p)->rb_right;
+		else
+			return -EEXIST;
+	}
+
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+
+	return 0;
+}
+
+static int tree_insert_bytes(struct rb_root *root, u64 bytes,
+			     struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct btrfs_free_space *info;
+
+	while (*p) {
+		parent = *p;
+		info = rb_entry(parent, struct btrfs_free_space, bytes_index);
+
+		if (bytes < info->bytes)
+			p = &(*p)->rb_left;
+		else
+			p = &(*p)->rb_right;
+	}
+
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+
+	return 0;
+}
+
+/*
+ * searches the tree for the given offset.  If contains is set we will return
+ * the free space that contains the given offset.  If contains is not set we
+ * will return the free space that starts at or after the given offset and is
+ * at least bytes long.
+ */
+static struct btrfs_free_space *tree_search_offset(struct rb_root *root,
+						   u64 offset, u64 bytes,
+						   int contains)
+{
+	struct rb_node *n = root->rb_node;
+	struct btrfs_free_space *entry, *ret = NULL;
+
+	while (n) {
+		entry = rb_entry(n, struct btrfs_free_space, offset_index);
+
+		if (offset < entry->offset) {
+			if (!contains &&
+			    (!ret || entry->offset < ret->offset) &&
+			    (bytes <= entry->bytes))
+				ret = entry;
+			n = n->rb_left;
+		} else if (offset > entry->offset) {
+			if ((entry->offset + entry->bytes - 1) >= offset &&
+			    bytes <= entry->bytes) {
+				ret = entry;
+				break;
+			}
+			n = n->rb_right;
+		} else {
+			if (bytes > entry->bytes) {
+				n = n->rb_right;
+				continue;
+			}
+			ret = entry;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * return a chunk at least bytes size, as close to offset that we can get.
+ */
+static struct btrfs_free_space *tree_search_bytes(struct rb_root *root,
+						  u64 offset, u64 bytes)
+{
+	struct rb_node *n = root->rb_node;
+	struct btrfs_free_space *entry, *ret = NULL;
+
+	while (n) {
+		entry = rb_entry(n, struct btrfs_free_space, bytes_index);
+
+		if (bytes < entry->bytes) {
+			/*
+			 * We prefer to get a hole size as close to the size we
+			 * are asking for so we don't take small slivers out of
+			 * huge holes, but we also want to get as close to the
+			 * offset as possible so we don't have a whole lot of
+			 * fragmentation.
+			 */
+			if (offset <= entry->offset) {
+				if (!ret)
+					ret = entry;
+				else if (entry->bytes < ret->bytes)
+					ret = entry;
+				else if (entry->offset < ret->offset)
+					ret = entry;
+			}
+			n = n->rb_left;
+		} else if (bytes > entry->bytes) {
+			n = n->rb_right;
+		} else {
+			/*
+			 * Ok we may have multiple chunks of the wanted size,
+			 * so we don't want to take the first one we find, we
+			 * want to take the one closest to our given offset, so
+			 * keep searching just in case theres a better match.
+			 */
+			n = n->rb_right;
+			if (offset > entry->offset)
+				continue;
+			else if (!ret || entry->offset < ret->offset)
+				ret = entry;
+		}
+	}
+
+	return ret;
+}
+
+static void unlink_free_space(struct btrfs_block_group_cache *block_group,
+			      struct btrfs_free_space *info)
+{
+	rb_erase(&info->offset_index, &block_group->free_space_offset);
+	rb_erase(&info->bytes_index, &block_group->free_space_bytes);
+}
+
+static int link_free_space(struct btrfs_block_group_cache *block_group,
+			   struct btrfs_free_space *info)
+{
+	int ret = 0;
+
+
+	ret = tree_insert_offset(&block_group->free_space_offset, info->offset,
+				 &info->offset_index);
+	if (ret)
+		return ret;
+
+	ret = tree_insert_bytes(&block_group->free_space_bytes, info->bytes,
+				&info->bytes_index);
+	if (ret)
+		return ret;
+
+	return ret;
+}
+
+static int __btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
+				  u64 offset, u64 bytes)
+{
+	struct btrfs_free_space *right_info;
+	struct btrfs_free_space *left_info;
+	struct btrfs_free_space *info = NULL;
+	struct btrfs_free_space *alloc_info;
+	int ret = 0;
+
+	alloc_info = kzalloc(sizeof(struct btrfs_free_space), GFP_NOFS);
+	if (!alloc_info)
+		return -ENOMEM;
+
+	/*
+	 * first we want to see if there is free space adjacent to the range we
+	 * are adding, if there is remove that struct and add a new one to
+	 * cover the entire range
+	 */
+	right_info = tree_search_offset(&block_group->free_space_offset,
+					offset+bytes, 0, 1);
+	left_info = tree_search_offset(&block_group->free_space_offset,
+				       offset-1, 0, 1);
+
+	if (right_info && right_info->offset == offset+bytes) {
+		unlink_free_space(block_group, right_info);
+		info = right_info;
+		info->offset = offset;
+		info->bytes += bytes;
+	} else if (right_info && right_info->offset != offset+bytes) {
+		printk(KERN_ERR "btrfs adding space in the middle of an "
+		       "existing free space area. existing: "
+		       "offset=%llu, bytes=%llu. new: offset=%llu, "
+		       "bytes=%llu\n", (unsigned long long)right_info->offset,
+		       (unsigned long long)right_info->bytes,
+		       (unsigned long long)offset,
+		       (unsigned long long)bytes);
+		BUG();
+	}
+
+	if (left_info) {
+		unlink_free_space(block_group, left_info);
+
+		if (unlikely((left_info->offset + left_info->bytes) !=
+			     offset)) {
+			printk(KERN_ERR "btrfs free space to the left "
+			       "of new free space isn't "
+			       "quite right. existing: offset=%llu, "
+			       "bytes=%llu. new: offset=%llu, bytes=%llu\n",
+			       (unsigned long long)left_info->offset,
+			       (unsigned long long)left_info->bytes,
+			       (unsigned long long)offset,
+			       (unsigned long long)bytes);
+			BUG();
+		}
+
+		if (info) {
+			info->offset = left_info->offset;
+			info->bytes += left_info->bytes;
+			kfree(left_info);
+		} else {
+			info = left_info;
+			info->bytes += bytes;
+		}
+	}
+
+	if (info) {
+		ret = link_free_space(block_group, info);
+		if (!ret)
+			info = NULL;
+		goto out;
+	}
+
+	info = alloc_info;
+	alloc_info = NULL;
+	info->offset = offset;
+	info->bytes = bytes;
+
+	ret = link_free_space(block_group, info);
+	if (ret)
+		kfree(info);
+out:
+	if (ret) {
+		printk(KERN_ERR "btrfs: unable to add free space :%d\n", ret);
+		if (ret == -EEXIST)
+			BUG();
+	}
+
+	kfree(alloc_info);
+
+	return ret;
+}
+
+static int
+__btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
+			  u64 offset, u64 bytes)
+{
+	struct btrfs_free_space *info;
+	int ret = 0;
+
+	info = tree_search_offset(&block_group->free_space_offset, offset, 0,
+				  1);
+
+	if (info && info->offset == offset) {
+		if (info->bytes < bytes) {
+			printk(KERN_ERR "Found free space at %llu, size %llu,"
+			       "trying to use %llu\n",
+			       (unsigned long long)info->offset,
+			       (unsigned long long)info->bytes,
+			       (unsigned long long)bytes);
+			WARN_ON(1);
+			ret = -EINVAL;
+			goto out;
+		}
+		unlink_free_space(block_group, info);
+
+		if (info->bytes == bytes) {
+			kfree(info);
+			goto out;
+		}
+
+		info->offset += bytes;
+		info->bytes -= bytes;
+
+		ret = link_free_space(block_group, info);
+		BUG_ON(ret);
+	} else if (info && info->offset < offset &&
+		   info->offset + info->bytes >= offset + bytes) {
+		u64 old_start = info->offset;
+		/*
+		 * we're freeing space in the middle of the info,
+		 * this can happen during tree log replay
+		 *
+		 * first unlink the old info and then
+		 * insert it again after the hole we're creating
+		 */
+		unlink_free_space(block_group, info);
+		if (offset + bytes < info->offset + info->bytes) {
+			u64 old_end = info->offset + info->bytes;
+
+			info->offset = offset + bytes;
+			info->bytes = old_end - info->offset;
+			ret = link_free_space(block_group, info);
+			BUG_ON(ret);
+		} else {
+			/* the hole we're creating ends at the end
+			 * of the info struct, just free the info
+			 */
+			kfree(info);
+		}
+
+		/* step two, insert a new info struct to cover anything
+		 * before the hole
+		 */
+		ret = __btrfs_add_free_space(block_group, old_start,
+					     offset - old_start);
+		BUG_ON(ret);
+	} else {
+		WARN_ON(1);
+	}
+out:
+	return ret;
+}
+
+int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
+			 u64 offset, u64 bytes)
+{
+	int ret;
+	struct btrfs_free_space *sp;
+
+	mutex_lock(&block_group->alloc_mutex);
+	ret = __btrfs_add_free_space(block_group, offset, bytes);
+	sp = tree_search_offset(&block_group->free_space_offset, offset, 0, 1);
+	BUG_ON(!sp);
+	mutex_unlock(&block_group->alloc_mutex);
+
+	return ret;
+}
+
+int btrfs_add_free_space_lock(struct btrfs_block_group_cache *block_group,
+			      u64 offset, u64 bytes)
+{
+	int ret;
+	struct btrfs_free_space *sp;
+
+	ret = __btrfs_add_free_space(block_group, offset, bytes);
+	sp = tree_search_offset(&block_group->free_space_offset, offset, 0, 1);
+	BUG_ON(!sp);
+
+	return ret;
+}
+
+int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
+			    u64 offset, u64 bytes)
+{
+	int ret = 0;
+
+	mutex_lock(&block_group->alloc_mutex);
+	ret = __btrfs_remove_free_space(block_group, offset, bytes);
+	mutex_unlock(&block_group->alloc_mutex);
+
+	return ret;
+}
+
+int btrfs_remove_free_space_lock(struct btrfs_block_group_cache *block_group,
+				 u64 offset, u64 bytes)
+{
+	int ret;
+
+	ret = __btrfs_remove_free_space(block_group, offset, bytes);
+
+	return ret;
+}
+
+void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
+			   u64 bytes)
+{
+	struct btrfs_free_space *info;
+	struct rb_node *n;
+	int count = 0;
+
+	for (n = rb_first(&block_group->free_space_offset); n; n = rb_next(n)) {
+		info = rb_entry(n, struct btrfs_free_space, offset_index);
+		if (info->bytes >= bytes)
+			count++;
+	}
+	printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
+	       "\n", count);
+}
+
+u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_free_space *info;
+	struct rb_node *n;
+	u64 ret = 0;
+
+	for (n = rb_first(&block_group->free_space_offset); n;
+	     n = rb_next(n)) {
+		info = rb_entry(n, struct btrfs_free_space, offset_index);
+		ret += info->bytes;
+	}
+
+	return ret;
+}
+
+void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_free_space *info;
+	struct rb_node *node;
+
+	mutex_lock(&block_group->alloc_mutex);
+	while ((node = rb_last(&block_group->free_space_bytes)) != NULL) {
+		info = rb_entry(node, struct btrfs_free_space, bytes_index);
+		unlink_free_space(block_group, info);
+		kfree(info);
+		if (need_resched()) {
+			mutex_unlock(&block_group->alloc_mutex);
+			cond_resched();
+			mutex_lock(&block_group->alloc_mutex);
+		}
+	}
+	mutex_unlock(&block_group->alloc_mutex);
+}
+
+#if 0
+static struct btrfs_free_space *btrfs_find_free_space_offset(struct
+						      btrfs_block_group_cache
+						      *block_group, u64 offset,
+						      u64 bytes)
+{
+	struct btrfs_free_space *ret;
+
+	mutex_lock(&block_group->alloc_mutex);
+	ret = tree_search_offset(&block_group->free_space_offset, offset,
+				 bytes, 0);
+	mutex_unlock(&block_group->alloc_mutex);
+
+	return ret;
+}
+
+static struct btrfs_free_space *btrfs_find_free_space_bytes(struct
+						     btrfs_block_group_cache
+						     *block_group, u64 offset,
+						     u64 bytes)
+{
+	struct btrfs_free_space *ret;
+
+	mutex_lock(&block_group->alloc_mutex);
+
+	ret = tree_search_bytes(&block_group->free_space_bytes, offset, bytes);
+	mutex_unlock(&block_group->alloc_mutex);
+
+	return ret;
+}
+#endif
+
+struct btrfs_free_space *btrfs_find_free_space(struct btrfs_block_group_cache
+					       *block_group, u64 offset,
+					       u64 bytes)
+{
+	struct btrfs_free_space *ret = NULL;
+
+	ret = tree_search_offset(&block_group->free_space_offset, offset,
+				 bytes, 0);
+	if (!ret)
+		ret = tree_search_bytes(&block_group->free_space_bytes,
+					offset, bytes);
+
+	return ret;
+}
-- 
1.6.0.2

[PATCH 21/35] Btrfs file-item and checksumming routines

[Chris Mason]

This provides helpers to insert, access and modify file extent pointers and
data checksums.

Checksums are stored in a single btree and indexed by the starting offset
of the extent.

File extents are stored in the subvolumes and indexed by inode number and
offset.  They can either point to extents on disk or directly include
an inline extent in the btree.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/file-item.c b/fs/btrfs/file-item.c
new file mode 100644
index 0000000..9646524
--- /dev/null
+++ b/fs/btrfs/file-item.c
@@ -0,0 +1,831 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/bio.h>
+#include <linux/pagemap.h>
+#include <linux/highmem.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "print-tree.h"
+
+#define MAX_CSUM_ITEMS(r, size) ((((BTRFS_LEAF_DATA_SIZE(r) - \
+				   sizeof(struct btrfs_item) * 2) / \
+				  size) - 1))
+
+#define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
+				   sizeof(struct btrfs_ordered_sum)) / \
+				   sizeof(struct btrfs_sector_sum) * \
+				   (r)->sectorsize - (r)->sectorsize)
+
+int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     u64 objectid, u64 pos,
+			     u64 disk_offset, u64 disk_num_bytes,
+			     u64 num_bytes, u64 offset, u64 ram_bytes,
+			     u8 compression, u8 encryption, u16 other_encoding)
+{
+	int ret = 0;
+	struct btrfs_file_extent_item *item;
+	struct btrfs_key file_key;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	file_key.objectid = objectid;
+	file_key.offset = pos;
+	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
+
+	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
+				      sizeof(*item));
+	if (ret < 0)
+		goto out;
+	BUG_ON(ret);
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0],
+			      struct btrfs_file_extent_item);
+	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
+	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
+	btrfs_set_file_extent_offset(leaf, item, offset);
+	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
+	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
+	btrfs_set_file_extent_generation(leaf, item, trans->transid);
+	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
+	btrfs_set_file_extent_compression(leaf, item, compression);
+	btrfs_set_file_extent_encryption(leaf, item, encryption);
+	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
+
+	btrfs_mark_buffer_dirty(leaf);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, int cow)
+{
+	int ret;
+	struct btrfs_key file_key;
+	struct btrfs_key found_key;
+	struct btrfs_csum_item *item;
+	struct extent_buffer *leaf;
+	u64 csum_offset = 0;
+	u16 csum_size =
+		btrfs_super_csum_size(&root->fs_info->super_copy);
+	int csums_in_item;
+
+	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
+	file_key.offset = bytenr;
+	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
+	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
+	if (ret < 0)
+		goto fail;
+	leaf = path->nodes[0];
+	if (ret > 0) {
+		ret = 1;
+		if (path->slots[0] == 0)
+			goto fail;
+		path->slots[0]--;
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
+			goto fail;
+
+		csum_offset = (bytenr - found_key.offset) >>
+				root->fs_info->sb->s_blocksize_bits;
+		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
+		csums_in_item /= csum_size;
+
+		if (csum_offset >= csums_in_item) {
+			ret = -EFBIG;
+			goto fail;
+		}
+	}
+	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
+	item = (struct btrfs_csum_item *)((unsigned char *)item +
+					  csum_offset * csum_size);
+	return item;
+fail:
+	if (ret > 0)
+		ret = -ENOENT;
+	return ERR_PTR(ret);
+}
+
+
+int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     struct btrfs_path *path, u64 objectid,
+			     u64 offset, int mod)
+{
+	int ret;
+	struct btrfs_key file_key;
+	int ins_len = mod < 0 ? -1 : 0;
+	int cow = mod != 0;
+
+	file_key.objectid = objectid;
+	file_key.offset = offset;
+	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
+	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
+	return ret;
+}
+
+
+int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
+			  struct bio *bio, u32 *dst)
+{
+	u32 sum;
+	struct bio_vec *bvec = bio->bi_io_vec;
+	int bio_index = 0;
+	u64 offset;
+	u64 item_start_offset = 0;
+	u64 item_last_offset = 0;
+	u64 disk_bytenr;
+	u32 diff;
+	u16 csum_size =
+		btrfs_super_csum_size(&root->fs_info->super_copy);
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_csum_item *item = NULL;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+
+	path = btrfs_alloc_path();
+	if (bio->bi_size > PAGE_CACHE_SIZE * 8)
+		path->reada = 2;
+
+	WARN_ON(bio->bi_vcnt <= 0);
+
+	disk_bytenr = (u64)bio->bi_sector << 9;
+	while (bio_index < bio->bi_vcnt) {
+		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
+		ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
+		if (ret == 0)
+			goto found;
+
+		if (!item || disk_bytenr < item_start_offset ||
+		    disk_bytenr >= item_last_offset) {
+			struct btrfs_key found_key;
+			u32 item_size;
+
+			if (item)
+				btrfs_release_path(root, path);
+			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
+						 path, disk_bytenr, 0);
+			if (IS_ERR(item)) {
+				ret = PTR_ERR(item);
+				if (ret == -ENOENT || ret == -EFBIG)
+					ret = 0;
+				sum = 0;
+				if (BTRFS_I(inode)->root->root_key.objectid ==
+				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
+					set_extent_bits(io_tree, offset,
+						offset + bvec->bv_len - 1,
+						EXTENT_NODATASUM, GFP_NOFS);
+				} else {
+					printk(KERN_INFO "btrfs no csum found "
+					       "for inode %lu start %llu\n",
+					       inode->i_ino,
+					       (unsigned long long)offset);
+				}
+				item = NULL;
+				btrfs_release_path(root, path);
+				goto found;
+			}
+			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+					      path->slots[0]);
+
+			item_start_offset = found_key.offset;
+			item_size = btrfs_item_size_nr(path->nodes[0],
+						       path->slots[0]);
+			item_last_offset = item_start_offset +
+				(item_size / csum_size) *
+				root->sectorsize;
+			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+					      struct btrfs_csum_item);
+		}
+		/*
+		 * this byte range must be able to fit inside
+		 * a single leaf so it will also fit inside a u32
+		 */
+		diff = disk_bytenr - item_start_offset;
+		diff = diff / root->sectorsize;
+		diff = diff * csum_size;
+
+		read_extent_buffer(path->nodes[0], &sum,
+				   ((unsigned long)item) + diff,
+				   csum_size);
+found:
+		if (dst)
+			*dst++ = sum;
+		else
+			set_state_private(io_tree, offset, sum);
+		disk_bytenr += bvec->bv_len;
+		bio_index++;
+		bvec++;
+	}
+	btrfs_free_path(path);
+	return 0;
+}
+
+int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
+			     struct list_head *list)
+{
+	struct btrfs_key key;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_ordered_sum *sums;
+	struct btrfs_sector_sum *sector_sum;
+	struct btrfs_csum_item *item;
+	unsigned long offset;
+	int ret;
+	size_t size;
+	u64 csum_end;
+	u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
+	key.offset = start;
+	key.type = BTRFS_EXTENT_CSUM_KEY;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto fail;
+	if (ret > 0 && path->slots[0] > 0) {
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
+		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
+		    key.type == BTRFS_EXTENT_CSUM_KEY) {
+			offset = (start - key.offset) >>
+				 root->fs_info->sb->s_blocksize_bits;
+			if (offset * csum_size <
+			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
+				path->slots[0]--;
+		}
+	}
+
+	while (start <= end) {
+		leaf = path->nodes[0];
+		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto fail;
+			if (ret > 0)
+				break;
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
+		    key.type != BTRFS_EXTENT_CSUM_KEY)
+			break;
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.offset > end)
+			break;
+
+		if (key.offset > start)
+			start = key.offset;
+
+		size = btrfs_item_size_nr(leaf, path->slots[0]);
+		csum_end = key.offset + (size / csum_size) * root->sectorsize;
+		if (csum_end <= start) {
+			path->slots[0]++;
+			continue;
+		}
+
+		csum_end = min(csum_end, end + 1);
+		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+				      struct btrfs_csum_item);
+		while (start < csum_end) {
+			size = min_t(size_t, csum_end - start,
+					MAX_ORDERED_SUM_BYTES(root));
+			sums = kzalloc(btrfs_ordered_sum_size(root, size),
+					GFP_NOFS);
+			BUG_ON(!sums);
+
+			sector_sum = sums->sums;
+			sums->bytenr = start;
+			sums->len = size;
+
+			offset = (start - key.offset) >>
+				root->fs_info->sb->s_blocksize_bits;
+			offset *= csum_size;
+
+			while (size > 0) {
+				read_extent_buffer(path->nodes[0],
+						&sector_sum->sum,
+						((unsigned long)item) +
+						offset, csum_size);
+				sector_sum->bytenr = start;
+
+				size -= root->sectorsize;
+				start += root->sectorsize;
+				offset += csum_size;
+				sector_sum++;
+			}
+			list_add_tail(&sums->list, list);
+		}
+		path->slots[0]++;
+	}
+	ret = 0;
+fail:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
+		       struct bio *bio, u64 file_start, int contig)
+{
+	struct btrfs_ordered_sum *sums;
+	struct btrfs_sector_sum *sector_sum;
+	struct btrfs_ordered_extent *ordered;
+	char *data;
+	struct bio_vec *bvec = bio->bi_io_vec;
+	int bio_index = 0;
+	unsigned long total_bytes = 0;
+	unsigned long this_sum_bytes = 0;
+	u64 offset;
+	u64 disk_bytenr;
+
+	WARN_ON(bio->bi_vcnt <= 0);
+	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
+	if (!sums)
+		return -ENOMEM;
+
+	sector_sum = sums->sums;
+	disk_bytenr = (u64)bio->bi_sector << 9;
+	sums->len = bio->bi_size;
+	INIT_LIST_HEAD(&sums->list);
+
+	if (contig)
+		offset = file_start;
+	else
+		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
+
+	ordered = btrfs_lookup_ordered_extent(inode, offset);
+	BUG_ON(!ordered);
+	sums->bytenr = ordered->start;
+
+	while (bio_index < bio->bi_vcnt) {
+		if (!contig)
+			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
+
+		if (!contig && (offset >= ordered->file_offset + ordered->len ||
+		    offset < ordered->file_offset)) {
+			unsigned long bytes_left;
+			sums->len = this_sum_bytes;
+			this_sum_bytes = 0;
+			btrfs_add_ordered_sum(inode, ordered, sums);
+			btrfs_put_ordered_extent(ordered);
+
+			bytes_left = bio->bi_size - total_bytes;
+
+			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
+				       GFP_NOFS);
+			BUG_ON(!sums);
+			sector_sum = sums->sums;
+			sums->len = bytes_left;
+			ordered = btrfs_lookup_ordered_extent(inode, offset);
+			BUG_ON(!ordered);
+			sums->bytenr = ordered->start;
+		}
+
+		data = kmap_atomic(bvec->bv_page, KM_USER0);
+		sector_sum->sum = ~(u32)0;
+		sector_sum->sum = btrfs_csum_data(root,
+						  data + bvec->bv_offset,
+						  sector_sum->sum,
+						  bvec->bv_len);
+		kunmap_atomic(data, KM_USER0);
+		btrfs_csum_final(sector_sum->sum,
+				 (char *)&sector_sum->sum);
+		sector_sum->bytenr = disk_bytenr;
+
+		sector_sum++;
+		bio_index++;
+		total_bytes += bvec->bv_len;
+		this_sum_bytes += bvec->bv_len;
+		disk_bytenr += bvec->bv_len;
+		offset += bvec->bv_len;
+		bvec++;
+	}
+	this_sum_bytes = 0;
+	btrfs_add_ordered_sum(inode, ordered, sums);
+	btrfs_put_ordered_extent(ordered);
+	return 0;
+}
+
+/*
+ * helper function for csum removal, this expects the
+ * key to describe the csum pointed to by the path, and it expects
+ * the csum to overlap the range [bytenr, len]
+ *
+ * The csum should not be entirely contained in the range and the
+ * range should not be entirely contained in the csum.
+ *
+ * This calls btrfs_truncate_item with the correct args based on the
+ * overlap, and fixes up the key as required.
+ */
+static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      struct btrfs_key *key,
+				      u64 bytenr, u64 len)
+{
+	struct extent_buffer *leaf;
+	u16 csum_size =
+		btrfs_super_csum_size(&root->fs_info->super_copy);
+	u64 csum_end;
+	u64 end_byte = bytenr + len;
+	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
+	int ret;
+
+	leaf = path->nodes[0];
+	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
+	csum_end <<= root->fs_info->sb->s_blocksize_bits;
+	csum_end += key->offset;
+
+	if (key->offset < bytenr && csum_end <= end_byte) {
+		/*
+		 *         [ bytenr - len ]
+		 *         [   ]
+		 *   [csum     ]
+		 *   A simple truncate off the end of the item
+		 */
+		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
+		new_size *= csum_size;
+		ret = btrfs_truncate_item(trans, root, path, new_size, 1);
+		BUG_ON(ret);
+	} else if (key->offset >= bytenr && csum_end > end_byte &&
+		   end_byte > key->offset) {
+		/*
+		 *         [ bytenr - len ]
+		 *                 [ ]
+		 *                 [csum     ]
+		 * we need to truncate from the beginning of the csum
+		 */
+		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
+		new_size *= csum_size;
+
+		ret = btrfs_truncate_item(trans, root, path, new_size, 0);
+		BUG_ON(ret);
+
+		key->offset = end_byte;
+		ret = btrfs_set_item_key_safe(trans, root, path, key);
+		BUG_ON(ret);
+	} else {
+		BUG();
+	}
+	return 0;
+}
+
+/*
+ * deletes the csum items from the csum tree for a given
+ * range of bytes.
+ */
+int btrfs_del_csums(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, u64 bytenr, u64 len)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	u64 end_byte = bytenr + len;
+	u64 csum_end;
+	struct extent_buffer *leaf;
+	int ret;
+	u16 csum_size =
+		btrfs_super_csum_size(&root->fs_info->super_copy);
+	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
+
+	root = root->fs_info->csum_root;
+
+	path = btrfs_alloc_path();
+
+	while (1) {
+		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
+		key.offset = end_byte - 1;
+		key.type = BTRFS_EXTENT_CSUM_KEY;
+
+		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+		if (ret > 0) {
+			if (path->slots[0] == 0)
+				goto out;
+			path->slots[0]--;
+		}
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+
+		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
+		    key.type != BTRFS_EXTENT_CSUM_KEY) {
+			break;
+		}
+
+		if (key.offset >= end_byte)
+			break;
+
+		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
+		csum_end <<= blocksize_bits;
+		csum_end += key.offset;
+
+		/* this csum ends before we start, we're done */
+		if (csum_end <= bytenr)
+			break;
+
+		/* delete the entire item, it is inside our range */
+		if (key.offset >= bytenr && csum_end <= end_byte) {
+			ret = btrfs_del_item(trans, root, path);
+			BUG_ON(ret);
+			if (key.offset == bytenr)
+				break;
+		} else if (key.offset < bytenr && csum_end > end_byte) {
+			unsigned long offset;
+			unsigned long shift_len;
+			unsigned long item_offset;
+			/*
+			 *        [ bytenr - len ]
+			 *     [csum                ]
+			 *
+			 * Our bytes are in the middle of the csum,
+			 * we need to split this item and insert a new one.
+			 *
+			 * But we can't drop the path because the
+			 * csum could change, get removed, extended etc.
+			 *
+			 * The trick here is the max size of a csum item leaves
+			 * enough room in the tree block for a single
+			 * item header.  So, we split the item in place,
+			 * adding a new header pointing to the existing
+			 * bytes.  Then we loop around again and we have
+			 * a nicely formed csum item that we can neatly
+			 * truncate.
+			 */
+			offset = (bytenr - key.offset) >> blocksize_bits;
+			offset *= csum_size;
+
+			shift_len = (len >> blocksize_bits) * csum_size;
+
+			item_offset = btrfs_item_ptr_offset(leaf,
+							    path->slots[0]);
+
+			memset_extent_buffer(leaf, 0, item_offset + offset,
+					     shift_len);
+			key.offset = bytenr;
+
+			/*
+			 * btrfs_split_item returns -EAGAIN when the
+			 * item changed size or key
+			 */
+			ret = btrfs_split_item(trans, root, path, &key, offset);
+			BUG_ON(ret && ret != -EAGAIN);
+
+			key.offset = end_byte - 1;
+		} else {
+			ret = truncate_one_csum(trans, root, path,
+						&key, bytenr, len);
+			BUG_ON(ret);
+			if (key.offset < bytenr)
+				break;
+		}
+		btrfs_release_path(root, path);
+	}
+out:
+	btrfs_free_path(path);
+	return 0;
+}
+
+int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   struct btrfs_ordered_sum *sums)
+{
+	u64 bytenr;
+	int ret;
+	struct btrfs_key file_key;
+	struct btrfs_key found_key;
+	u64 next_offset;
+	u64 total_bytes = 0;
+	int found_next;
+	struct btrfs_path *path;
+	struct btrfs_csum_item *item;
+	struct btrfs_csum_item *item_end;
+	struct extent_buffer *leaf = NULL;
+	u64 csum_offset;
+	struct btrfs_sector_sum *sector_sum;
+	u32 nritems;
+	u32 ins_size;
+	char *eb_map;
+	char *eb_token;
+	unsigned long map_len;
+	unsigned long map_start;
+	u16 csum_size =
+		btrfs_super_csum_size(&root->fs_info->super_copy);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	sector_sum = sums->sums;
+again:
+	next_offset = (u64)-1;
+	found_next = 0;
+	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
+	file_key.offset = sector_sum->bytenr;
+	bytenr = sector_sum->bytenr;
+	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
+
+	item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
+	if (!IS_ERR(item)) {
+		leaf = path->nodes[0];
+		ret = 0;
+		goto found;
+	}
+	ret = PTR_ERR(item);
+	if (ret == -EFBIG) {
+		u32 item_size;
+		/* we found one, but it isn't big enough yet */
+		leaf = path->nodes[0];
+		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+		if ((item_size / csum_size) >=
+		    MAX_CSUM_ITEMS(root, csum_size)) {
+			/* already at max size, make a new one */
+			goto insert;
+		}
+	} else {
+		int slot = path->slots[0] + 1;
+		/* we didn't find a csum item, insert one */
+		nritems = btrfs_header_nritems(path->nodes[0]);
+		if (path->slots[0] >= nritems - 1) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret == 1)
+				found_next = 1;
+			if (ret != 0)
+				goto insert;
+			slot = 0;
+		}
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
+		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
+		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
+			found_next = 1;
+			goto insert;
+		}
+		next_offset = found_key.offset;
+		found_next = 1;
+		goto insert;
+	}
+
+	/*
+	 * at this point, we know the tree has an item, but it isn't big
+	 * enough yet to put our csum in.  Grow it
+	 */
+	btrfs_release_path(root, path);
+	ret = btrfs_search_slot(trans, root, &file_key, path,
+				csum_size, 1);
+	if (ret < 0)
+		goto fail_unlock;
+
+	if (ret > 0) {
+		if (path->slots[0] == 0)
+			goto insert;
+		path->slots[0]--;
+	}
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+	csum_offset = (bytenr - found_key.offset) >>
+			root->fs_info->sb->s_blocksize_bits;
+
+	if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
+	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
+	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
+		goto insert;
+	}
+
+	if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
+	    csum_size) {
+		u32 diff = (csum_offset + 1) * csum_size;
+
+		/*
+		 * is the item big enough already?  we dropped our lock
+		 * before and need to recheck
+		 */
+		if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
+			goto csum;
+
+		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
+		if (diff != csum_size)
+			goto insert;
+
+		ret = btrfs_extend_item(trans, root, path, diff);
+		BUG_ON(ret);
+		goto csum;
+	}
+
+insert:
+	btrfs_release_path(root, path);
+	csum_offset = 0;
+	if (found_next) {
+		u64 tmp = total_bytes + root->sectorsize;
+		u64 next_sector = sector_sum->bytenr;
+		struct btrfs_sector_sum *next = sector_sum + 1;
+
+		while (tmp < sums->len) {
+			if (next_sector + root->sectorsize != next->bytenr)
+				break;
+			tmp += root->sectorsize;
+			next_sector = next->bytenr;
+			next++;
+		}
+		tmp = min(tmp, next_offset - file_key.offset);
+		tmp >>= root->fs_info->sb->s_blocksize_bits;
+		tmp = max((u64)1, tmp);
+		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
+		ins_size = csum_size * tmp;
+	} else {
+		ins_size = csum_size;
+	}
+	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
+				      ins_size);
+	if (ret < 0)
+		goto fail_unlock;
+	if (ret != 0) {
+		WARN_ON(1);
+		goto fail_unlock;
+	}
+csum:
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
+	ret = 0;
+	item = (struct btrfs_csum_item *)((unsigned char *)item +
+					  csum_offset * csum_size);
+found:
+	item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
+	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
+				      btrfs_item_size_nr(leaf, path->slots[0]));
+	eb_token = NULL;
+	cond_resched();
+next_sector:
+
+	if (!eb_token ||
+	   (unsigned long)item + csum_size >= map_start + map_len) {
+		int err;
+
+		if (eb_token)
+			unmap_extent_buffer(leaf, eb_token, KM_USER1);
+		eb_token = NULL;
+		err = map_private_extent_buffer(leaf, (unsigned long)item,
+						csum_size,
+						&eb_token, &eb_map,
+						&map_start, &map_len, KM_USER1);
+		if (err)
+			eb_token = NULL;
+	}
+	if (eb_token) {
+		memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
+		       &sector_sum->sum, csum_size);
+	} else {
+		write_extent_buffer(leaf, &sector_sum->sum,
+				    (unsigned long)item, csum_size);
+	}
+
+	total_bytes += root->sectorsize;
+	sector_sum++;
+	if (total_bytes < sums->len) {
+		item = (struct btrfs_csum_item *)((char *)item +
+						  csum_size);
+		if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
+		    sector_sum->bytenr) {
+			bytenr = sector_sum->bytenr;
+			goto next_sector;
+		}
+	}
+	if (eb_token) {
+		unmap_extent_buffer(leaf, eb_token, KM_USER1);
+		eb_token = NULL;
+	}
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	cond_resched();
+	if (total_bytes < sums->len) {
+		btrfs_release_path(root, path);
+		goto again;
+	}
+out:
+	btrfs_free_path(path);
+	return ret;
+
+fail_unlock:
+	goto out;
+}
-- 
1.6.0.2

[PATCH 22/35] Btrfs file write routines

[Chris Mason]

Btrfs file write is mostly and engine to organize pages and set them
up for delayed allocation.  It could be moved into generic helpers instead.

One difficult part of COW for file data is the need to overwrite part of
an extent.  btrfs_drop_extents does the dirty work of trimming and
splitting extents to deal with a COW.

Btrfs does not do read/modify/write cycles to COW extents. It adds
a reference to the existing extent and creates a second pointer to it.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
new file mode 100644
index 0000000..9026833
--- /dev/null
+++ b/fs/btrfs/file.c
@@ -0,0 +1,1288 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/highmem.h>
+#include <linux/time.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/smp_lock.h>
+#include <linux/backing-dev.h>
+#include <linux/mpage.h>
+#include <linux/swap.h>
+#include <linux/writeback.h>
+#include <linux/statfs.h>
+#include <linux/compat.h>
+#include <linux/version.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "btrfs_inode.h"
+#include "ioctl.h"
+#include "print-tree.h"
+#include "tree-log.h"
+#include "locking.h"
+#include "compat.h"
+
+
+/* simple helper to fault in pages and copy.  This should go away
+ * and be replaced with calls into generic code.
+ */
+static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
+					 int write_bytes,
+					 struct page **prepared_pages,
+					 const char __user *buf)
+{
+	long page_fault = 0;
+	int i;
+	int offset = pos & (PAGE_CACHE_SIZE - 1);
+
+	for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
+		size_t count = min_t(size_t,
+				     PAGE_CACHE_SIZE - offset, write_bytes);
+		struct page *page = prepared_pages[i];
+		fault_in_pages_readable(buf, count);
+
+		/* Copy data from userspace to the current page */
+		kmap(page);
+		page_fault = __copy_from_user(page_address(page) + offset,
+					      buf, count);
+		/* Flush processor's dcache for this page */
+		flush_dcache_page(page);
+		kunmap(page);
+		buf += count;
+		write_bytes -= count;
+
+		if (page_fault)
+			break;
+	}
+	return page_fault ? -EFAULT : 0;
+}
+
+/*
+ * unlocks pages after btrfs_file_write is done with them
+ */
+static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
+{
+	size_t i;
+	for (i = 0; i < num_pages; i++) {
+		if (!pages[i])
+			break;
+		/* page checked is some magic around finding pages that
+		 * have been modified without going through btrfs_set_page_dirty
+		 * clear it here
+		 */
+		ClearPageChecked(pages[i]);
+		unlock_page(pages[i]);
+		mark_page_accessed(pages[i]);
+		page_cache_release(pages[i]);
+	}
+}
+
+/*
+ * after copy_from_user, pages need to be dirtied and we need to make
+ * sure holes are created between the current EOF and the start of
+ * any next extents (if required).
+ *
+ * this also makes the decision about creating an inline extent vs
+ * doing real data extents, marking pages dirty and delalloc as required.
+ */
+static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct file *file,
+				   struct page **pages,
+				   size_t num_pages,
+				   loff_t pos,
+				   size_t write_bytes)
+{
+	int err = 0;
+	int i;
+	struct inode *inode = fdentry(file)->d_inode;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	u64 hint_byte;
+	u64 num_bytes;
+	u64 start_pos;
+	u64 end_of_last_block;
+	u64 end_pos = pos + write_bytes;
+	loff_t isize = i_size_read(inode);
+
+	start_pos = pos & ~((u64)root->sectorsize - 1);
+	num_bytes = (write_bytes + pos - start_pos +
+		    root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
+
+	end_of_last_block = start_pos + num_bytes - 1;
+
+	lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
+	trans = btrfs_join_transaction(root, 1);
+	if (!trans) {
+		err = -ENOMEM;
+		goto out_unlock;
+	}
+	btrfs_set_trans_block_group(trans, inode);
+	hint_byte = 0;
+
+	set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
+
+	/* check for reserved extents on each page, we don't want
+	 * to reset the delalloc bit on things that already have
+	 * extents reserved.
+	 */
+	btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
+	for (i = 0; i < num_pages; i++) {
+		struct page *p = pages[i];
+		SetPageUptodate(p);
+		ClearPageChecked(p);
+		set_page_dirty(p);
+	}
+	if (end_pos > isize) {
+		i_size_write(inode, end_pos);
+		btrfs_update_inode(trans, root, inode);
+	}
+	err = btrfs_end_transaction(trans, root);
+out_unlock:
+	unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
+	return err;
+}
+
+/*
+ * this drops all the extents in the cache that intersect the range
+ * [start, end].  Existing extents are split as required.
+ */
+int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
+			    int skip_pinned)
+{
+	struct extent_map *em;
+	struct extent_map *split = NULL;
+	struct extent_map *split2 = NULL;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	u64 len = end - start + 1;
+	int ret;
+	int testend = 1;
+	unsigned long flags;
+	int compressed = 0;
+
+	WARN_ON(end < start);
+	if (end == (u64)-1) {
+		len = (u64)-1;
+		testend = 0;
+	}
+	while (1) {
+		if (!split)
+			split = alloc_extent_map(GFP_NOFS);
+		if (!split2)
+			split2 = alloc_extent_map(GFP_NOFS);
+
+		spin_lock(&em_tree->lock);
+		em = lookup_extent_mapping(em_tree, start, len);
+		if (!em) {
+			spin_unlock(&em_tree->lock);
+			break;
+		}
+		flags = em->flags;
+		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
+			spin_unlock(&em_tree->lock);
+			if (em->start <= start &&
+			    (!testend || em->start + em->len >= start + len)) {
+				free_extent_map(em);
+				break;
+			}
+			if (start < em->start) {
+				len = em->start - start;
+			} else {
+				len = start + len - (em->start + em->len);
+				start = em->start + em->len;
+			}
+			free_extent_map(em);
+			continue;
+		}
+		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
+		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
+		remove_extent_mapping(em_tree, em);
+
+		if (em->block_start < EXTENT_MAP_LAST_BYTE &&
+		    em->start < start) {
+			split->start = em->start;
+			split->len = start - em->start;
+			split->orig_start = em->orig_start;
+			split->block_start = em->block_start;
+
+			if (compressed)
+				split->block_len = em->block_len;
+			else
+				split->block_len = split->len;
+
+			split->bdev = em->bdev;
+			split->flags = flags;
+			ret = add_extent_mapping(em_tree, split);
+			BUG_ON(ret);
+			free_extent_map(split);
+			split = split2;
+			split2 = NULL;
+		}
+		if (em->block_start < EXTENT_MAP_LAST_BYTE &&
+		    testend && em->start + em->len > start + len) {
+			u64 diff = start + len - em->start;
+
+			split->start = start + len;
+			split->len = em->start + em->len - (start + len);
+			split->bdev = em->bdev;
+			split->flags = flags;
+
+			if (compressed) {
+				split->block_len = em->block_len;
+				split->block_start = em->block_start;
+				split->orig_start = em->orig_start;
+			} else {
+				split->block_len = split->len;
+				split->block_start = em->block_start + diff;
+				split->orig_start = split->start;
+			}
+
+			ret = add_extent_mapping(em_tree, split);
+			BUG_ON(ret);
+			free_extent_map(split);
+			split = NULL;
+		}
+		spin_unlock(&em_tree->lock);
+
+		/* once for us */
+		free_extent_map(em);
+		/* once for the tree*/
+		free_extent_map(em);
+	}
+	if (split)
+		free_extent_map(split);
+	if (split2)
+		free_extent_map(split2);
+	return 0;
+}
+
+int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
+{
+	return 0;
+#if 0
+	struct btrfs_path *path;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+	struct btrfs_file_extent_item *extent;
+	u64 last_offset = 0;
+	int nritems;
+	int slot;
+	int found_type;
+	int ret;
+	int err = 0;
+	u64 extent_end = 0;
+
+	path = btrfs_alloc_path();
+	ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
+				       last_offset, 0);
+	while (1) {
+		nritems = btrfs_header_nritems(path->nodes[0]);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret)
+				goto out;
+			nritems = btrfs_header_nritems(path->nodes[0]);
+		}
+		slot = path->slots[0];
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+		if (found_key.objectid != inode->i_ino)
+			break;
+		if (found_key.type != BTRFS_EXTENT_DATA_KEY)
+			goto out;
+
+		if (found_key.offset < last_offset) {
+			WARN_ON(1);
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_ERR "inode %lu found offset %llu "
+			       "expected %llu\n", inode->i_ino,
+			       (unsigned long long)found_key.offset,
+			       (unsigned long long)last_offset);
+			err = 1;
+			goto out;
+		}
+		extent = btrfs_item_ptr(leaf, slot,
+					struct btrfs_file_extent_item);
+		found_type = btrfs_file_extent_type(leaf, extent);
+		if (found_type == BTRFS_FILE_EXTENT_REG) {
+			extent_end = found_key.offset +
+			     btrfs_file_extent_num_bytes(leaf, extent);
+		} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+			struct btrfs_item *item;
+			item = btrfs_item_nr(leaf, slot);
+			extent_end = found_key.offset +
+			     btrfs_file_extent_inline_len(leaf, extent);
+			extent_end = (extent_end + root->sectorsize - 1) &
+				~((u64)root->sectorsize - 1);
+		}
+		last_offset = extent_end;
+		path->slots[0]++;
+	}
+	if (0 && last_offset < inode->i_size) {
+		WARN_ON(1);
+		btrfs_print_leaf(root, leaf);
+		printk(KERN_ERR "inode %lu found offset %llu size %llu\n",
+		       inode->i_ino, (unsigned long long)last_offset,
+		       (unsigned long long)inode->i_size);
+		err = 1;
+
+	}
+out:
+	btrfs_free_path(path);
+	return err;
+#endif
+}
+
+/*
+ * this is very complex, but the basic idea is to drop all extents
+ * in the range start - end.  hint_block is filled in with a block number
+ * that would be a good hint to the block allocator for this file.
+ *
+ * If an extent intersects the range but is not entirely inside the range
+ * it is either truncated or split.  Anything entirely inside the range
+ * is deleted from the tree.
+ *
+ * inline_limit is used to tell this code which offsets in the file to keep
+ * if they contain inline extents.
+ */
+noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root, struct inode *inode,
+		       u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
+{
+	u64 extent_end = 0;
+	u64 locked_end = end;
+	u64 search_start = start;
+	u64 leaf_start;
+	u64 ram_bytes = 0;
+	u64 orig_parent = 0;
+	u64 disk_bytenr = 0;
+	u8 compression;
+	u8 encryption;
+	u16 other_encoding = 0;
+	u64 root_gen;
+	u64 root_owner;
+	struct extent_buffer *leaf;
+	struct btrfs_file_extent_item *extent;
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item old;
+	int keep;
+	int slot;
+	int bookend;
+	int found_type = 0;
+	int found_extent;
+	int found_inline;
+	int recow;
+	int ret;
+
+	inline_limit = 0;
+	btrfs_drop_extent_cache(inode, start, end - 1, 0);
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	while (1) {
+		recow = 0;
+		btrfs_release_path(root, path);
+		ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
+					       search_start, -1);
+		if (ret < 0)
+			goto out;
+		if (ret > 0) {
+			if (path->slots[0] == 0) {
+				ret = 0;
+				goto out;
+			}
+			path->slots[0]--;
+		}
+next_slot:
+		keep = 0;
+		bookend = 0;
+		found_extent = 0;
+		found_inline = 0;
+		leaf_start = 0;
+		root_gen = 0;
+		root_owner = 0;
+		compression = 0;
+		encryption = 0;
+		extent = NULL;
+		leaf = path->nodes[0];
+		slot = path->slots[0];
+		ret = 0;
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
+		    key.offset >= end) {
+			goto out;
+		}
+		if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
+		    key.objectid != inode->i_ino) {
+			goto out;
+		}
+		if (recow) {
+			search_start = max(key.offset, start);
+			continue;
+		}
+		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
+			extent = btrfs_item_ptr(leaf, slot,
+						struct btrfs_file_extent_item);
+			found_type = btrfs_file_extent_type(leaf, extent);
+			compression = btrfs_file_extent_compression(leaf,
+								    extent);
+			encryption = btrfs_file_extent_encryption(leaf,
+								  extent);
+			other_encoding = btrfs_file_extent_other_encoding(leaf,
+								  extent);
+			if (found_type == BTRFS_FILE_EXTENT_REG ||
+			    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
+				extent_end =
+				     btrfs_file_extent_disk_bytenr(leaf,
+								   extent);
+				if (extent_end)
+					*hint_byte = extent_end;
+
+				extent_end = key.offset +
+				     btrfs_file_extent_num_bytes(leaf, extent);
+				ram_bytes = btrfs_file_extent_ram_bytes(leaf,
+								extent);
+				found_extent = 1;
+			} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+				found_inline = 1;
+				extent_end = key.offset +
+				     btrfs_file_extent_inline_len(leaf, extent);
+			}
+		} else {
+			extent_end = search_start;
+		}
+
+		/* we found nothing we can drop */
+		if ((!found_extent && !found_inline) ||
+		    search_start >= extent_end) {
+			int nextret;
+			u32 nritems;
+			nritems = btrfs_header_nritems(leaf);
+			if (slot >= nritems - 1) {
+				nextret = btrfs_next_leaf(root, path);
+				if (nextret)
+					goto out;
+				recow = 1;
+			} else {
+				path->slots[0]++;
+			}
+			goto next_slot;
+		}
+
+		if (end <= extent_end && start >= key.offset && found_inline)
+			*hint_byte = EXTENT_MAP_INLINE;
+
+		if (found_extent) {
+			read_extent_buffer(leaf, &old, (unsigned long)extent,
+					   sizeof(old));
+			root_gen = btrfs_header_generation(leaf);
+			root_owner = btrfs_header_owner(leaf);
+			leaf_start = leaf->start;
+		}
+
+		if (end < extent_end && end >= key.offset) {
+			bookend = 1;
+			if (found_inline && start <= key.offset)
+				keep = 1;
+		}
+
+		if (bookend && found_extent) {
+			if (locked_end < extent_end) {
+				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
+						locked_end, extent_end - 1,
+						GFP_NOFS);
+				if (!ret) {
+					btrfs_release_path(root, path);
+					lock_extent(&BTRFS_I(inode)->io_tree,
+						locked_end, extent_end - 1,
+						GFP_NOFS);
+					locked_end = extent_end;
+					continue;
+				}
+				locked_end = extent_end;
+			}
+			orig_parent = path->nodes[0]->start;
+			disk_bytenr = le64_to_cpu(old.disk_bytenr);
+			if (disk_bytenr != 0) {
+				ret = btrfs_inc_extent_ref(trans, root,
+					   disk_bytenr,
+					   le64_to_cpu(old.disk_num_bytes),
+					   orig_parent, root->root_key.objectid,
+					   trans->transid, inode->i_ino);
+				BUG_ON(ret);
+			}
+		}
+
+		if (found_inline) {
+			u64 mask = root->sectorsize - 1;
+			search_start = (extent_end + mask) & ~mask;
+		} else
+			search_start = extent_end;
+
+		/* truncate existing extent */
+		if (start > key.offset) {
+			u64 new_num;
+			u64 old_num;
+			keep = 1;
+			WARN_ON(start & (root->sectorsize - 1));
+			if (found_extent) {
+				new_num = start - key.offset;
+				old_num = btrfs_file_extent_num_bytes(leaf,
+								      extent);
+				*hint_byte =
+					btrfs_file_extent_disk_bytenr(leaf,
+								      extent);
+				if (btrfs_file_extent_disk_bytenr(leaf,
+								  extent)) {
+					inode_sub_bytes(inode, old_num -
+							new_num);
+				}
+				btrfs_set_file_extent_num_bytes(leaf,
+							extent, new_num);
+				btrfs_mark_buffer_dirty(leaf);
+			} else if (key.offset < inline_limit &&
+				   (end > extent_end) &&
+				   (inline_limit < extent_end)) {
+				u32 new_size;
+				new_size = btrfs_file_extent_calc_inline_size(
+						   inline_limit - key.offset);
+				inode_sub_bytes(inode, extent_end -
+						inline_limit);
+				btrfs_set_file_extent_ram_bytes(leaf, extent,
+							new_size);
+				if (!compression && !encryption) {
+					btrfs_truncate_item(trans, root, path,
+							    new_size, 1);
+				}
+			}
+		}
+		/* delete the entire extent */
+		if (!keep) {
+			if (found_inline)
+				inode_sub_bytes(inode, extent_end -
+						key.offset);
+			ret = btrfs_del_item(trans, root, path);
+			/* TODO update progress marker and return */
+			BUG_ON(ret);
+			extent = NULL;
+			btrfs_release_path(root, path);
+			/* the extent will be freed later */
+		}
+		if (bookend && found_inline && start <= key.offset) {
+			u32 new_size;
+			new_size = btrfs_file_extent_calc_inline_size(
+						   extent_end - end);
+			inode_sub_bytes(inode, end - key.offset);
+			btrfs_set_file_extent_ram_bytes(leaf, extent,
+							new_size);
+			if (!compression && !encryption)
+				ret = btrfs_truncate_item(trans, root, path,
+							  new_size, 0);
+			BUG_ON(ret);
+		}
+		/* create bookend, splitting the extent in two */
+		if (bookend && found_extent) {
+			struct btrfs_key ins;
+			ins.objectid = inode->i_ino;
+			ins.offset = end;
+			btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
+
+			btrfs_release_path(root, path);
+			ret = btrfs_insert_empty_item(trans, root, path, &ins,
+						      sizeof(*extent));
+			BUG_ON(ret);
+
+			leaf = path->nodes[0];
+			extent = btrfs_item_ptr(leaf, path->slots[0],
+						struct btrfs_file_extent_item);
+			write_extent_buffer(leaf, &old,
+					    (unsigned long)extent, sizeof(old));
+
+			btrfs_set_file_extent_compression(leaf, extent,
+							  compression);
+			btrfs_set_file_extent_encryption(leaf, extent,
+							 encryption);
+			btrfs_set_file_extent_other_encoding(leaf, extent,
+							     other_encoding);
+			btrfs_set_file_extent_offset(leaf, extent,
+				    le64_to_cpu(old.offset) + end - key.offset);
+			WARN_ON(le64_to_cpu(old.num_bytes) <
+				(extent_end - end));
+			btrfs_set_file_extent_num_bytes(leaf, extent,
+							extent_end - end);
+
+			/*
+			 * set the ram bytes to the size of the full extent
+			 * before splitting.  This is a worst case flag,
+			 * but its the best we can do because we don't know
+			 * how splitting affects compression
+			 */
+			btrfs_set_file_extent_ram_bytes(leaf, extent,
+							ram_bytes);
+			btrfs_set_file_extent_type(leaf, extent, found_type);
+
+			btrfs_mark_buffer_dirty(path->nodes[0]);
+
+			if (disk_bytenr != 0) {
+				ret = btrfs_update_extent_ref(trans, root,
+						disk_bytenr, orig_parent,
+						leaf->start,
+						root->root_key.objectid,
+						trans->transid, ins.objectid);
+
+				BUG_ON(ret);
+			}
+			btrfs_release_path(root, path);
+			if (disk_bytenr != 0)
+				inode_add_bytes(inode, extent_end - end);
+		}
+
+		if (found_extent && !keep) {
+			u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
+
+			if (old_disk_bytenr != 0) {
+				inode_sub_bytes(inode,
+						le64_to_cpu(old.num_bytes));
+				ret = btrfs_free_extent(trans, root,
+						old_disk_bytenr,
+						le64_to_cpu(old.disk_num_bytes),
+						leaf_start, root_owner,
+						root_gen, key.objectid, 0);
+				BUG_ON(ret);
+				*hint_byte = old_disk_bytenr;
+			}
+		}
+
+		if (search_start >= end) {
+			ret = 0;
+			goto out;
+		}
+	}
+out:
+	btrfs_free_path(path);
+	if (locked_end > end) {
+		unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
+			      GFP_NOFS);
+	}
+	btrfs_check_file(root, inode);
+	return ret;
+}
+
+static int extent_mergeable(struct extent_buffer *leaf, int slot,
+			    u64 objectid, u64 bytenr, u64 *start, u64 *end)
+{
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_key key;
+	u64 extent_end;
+
+	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
+		return 0;
+
+	btrfs_item_key_to_cpu(leaf, &key, slot);
+	if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
+		return 0;
+
+	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
+	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
+	    btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
+	    btrfs_file_extent_compression(leaf, fi) ||
+	    btrfs_file_extent_encryption(leaf, fi) ||
+	    btrfs_file_extent_other_encoding(leaf, fi))
+		return 0;
+
+	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
+	if ((*start && *start != key.offset) || (*end && *end != extent_end))
+		return 0;
+
+	*start = key.offset;
+	*end = extent_end;
+	return 1;
+}
+
+/*
+ * Mark extent in the range start - end as written.
+ *
+ * This changes extent type from 'pre-allocated' to 'regular'. If only
+ * part of extent is marked as written, the extent will be split into
+ * two or three.
+ */
+int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct inode *inode, u64 start, u64 end)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_path *path;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_key key;
+	u64 bytenr;
+	u64 num_bytes;
+	u64 extent_end;
+	u64 extent_offset;
+	u64 other_start;
+	u64 other_end;
+	u64 split = start;
+	u64 locked_end = end;
+	u64 orig_parent;
+	int extent_type;
+	int split_end = 1;
+	int ret;
+
+	btrfs_drop_extent_cache(inode, start, end - 1, 0);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+again:
+	key.objectid = inode->i_ino;
+	key.type = BTRFS_EXTENT_DATA_KEY;
+	if (split == start)
+		key.offset = split;
+	else
+		key.offset = split - 1;
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret > 0 && path->slots[0] > 0)
+		path->slots[0]--;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+	BUG_ON(key.objectid != inode->i_ino ||
+	       key.type != BTRFS_EXTENT_DATA_KEY);
+	fi = btrfs_item_ptr(leaf, path->slots[0],
+			    struct btrfs_file_extent_item);
+	extent_type = btrfs_file_extent_type(leaf, fi);
+	BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
+	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
+	BUG_ON(key.offset > start || extent_end < end);
+
+	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
+	extent_offset = btrfs_file_extent_offset(leaf, fi);
+
+	if (key.offset == start)
+		split = end;
+
+	if (key.offset == start && extent_end == end) {
+		int del_nr = 0;
+		int del_slot = 0;
+		u64 leaf_owner = btrfs_header_owner(leaf);
+		u64 leaf_gen = btrfs_header_generation(leaf);
+		other_start = end;
+		other_end = 0;
+		if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
+				     bytenr, &other_start, &other_end)) {
+			extent_end = other_end;
+			del_slot = path->slots[0] + 1;
+			del_nr++;
+			ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
+						leaf->start, leaf_owner,
+						leaf_gen, inode->i_ino, 0);
+			BUG_ON(ret);
+		}
+		other_start = 0;
+		other_end = start;
+		if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
+				     bytenr, &other_start, &other_end)) {
+			key.offset = other_start;
+			del_slot = path->slots[0];
+			del_nr++;
+			ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
+						leaf->start, leaf_owner,
+						leaf_gen, inode->i_ino, 0);
+			BUG_ON(ret);
+		}
+		split_end = 0;
+		if (del_nr == 0) {
+			btrfs_set_file_extent_type(leaf, fi,
+						   BTRFS_FILE_EXTENT_REG);
+			goto done;
+		}
+
+		fi = btrfs_item_ptr(leaf, del_slot - 1,
+				    struct btrfs_file_extent_item);
+		btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
+		btrfs_set_file_extent_num_bytes(leaf, fi,
+						extent_end - key.offset);
+		btrfs_mark_buffer_dirty(leaf);
+
+		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
+		BUG_ON(ret);
+		goto done;
+	} else if (split == start) {
+		if (locked_end < extent_end) {
+			ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
+					locked_end, extent_end - 1, GFP_NOFS);
+			if (!ret) {
+				btrfs_release_path(root, path);
+				lock_extent(&BTRFS_I(inode)->io_tree,
+					locked_end, extent_end - 1, GFP_NOFS);
+				locked_end = extent_end;
+				goto again;
+			}
+			locked_end = extent_end;
+		}
+		btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
+		extent_offset += split - key.offset;
+	} else  {
+		BUG_ON(key.offset != start);
+		btrfs_set_file_extent_offset(leaf, fi, extent_offset +
+					     split - key.offset);
+		btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
+		key.offset = split;
+		btrfs_set_item_key_safe(trans, root, path, &key);
+		extent_end = split;
+	}
+
+	if (extent_end == end) {
+		split_end = 0;
+		extent_type = BTRFS_FILE_EXTENT_REG;
+	}
+	if (extent_end == end && split == start) {
+		other_start = end;
+		other_end = 0;
+		if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
+				     bytenr, &other_start, &other_end)) {
+			path->slots[0]++;
+			fi = btrfs_item_ptr(leaf, path->slots[0],
+					    struct btrfs_file_extent_item);
+			key.offset = split;
+			btrfs_set_item_key_safe(trans, root, path, &key);
+			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
+			btrfs_set_file_extent_num_bytes(leaf, fi,
+							other_end - split);
+			goto done;
+		}
+	}
+	if (extent_end == end && split == end) {
+		other_start = 0;
+		other_end = start;
+		if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
+				     bytenr, &other_start, &other_end)) {
+			path->slots[0]--;
+			fi = btrfs_item_ptr(leaf, path->slots[0],
+					    struct btrfs_file_extent_item);
+			btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
+							other_start);
+			goto done;
+		}
+	}
+
+	btrfs_mark_buffer_dirty(leaf);
+
+	orig_parent = leaf->start;
+	ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
+				   orig_parent, root->root_key.objectid,
+				   trans->transid, inode->i_ino);
+	BUG_ON(ret);
+	btrfs_release_path(root, path);
+
+	key.offset = start;
+	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	fi = btrfs_item_ptr(leaf, path->slots[0],
+			    struct btrfs_file_extent_item);
+	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
+	btrfs_set_file_extent_type(leaf, fi, extent_type);
+	btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
+	btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
+	btrfs_set_file_extent_offset(leaf, fi, extent_offset);
+	btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
+	btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
+	btrfs_set_file_extent_compression(leaf, fi, 0);
+	btrfs_set_file_extent_encryption(leaf, fi, 0);
+	btrfs_set_file_extent_other_encoding(leaf, fi, 0);
+
+	if (orig_parent != leaf->start) {
+		ret = btrfs_update_extent_ref(trans, root, bytenr,
+					      orig_parent, leaf->start,
+					      root->root_key.objectid,
+					      trans->transid, inode->i_ino);
+		BUG_ON(ret);
+	}
+done:
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(root, path);
+	if (split_end && split == start) {
+		split = end;
+		goto again;
+	}
+	if (locked_end > end) {
+		unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
+			      GFP_NOFS);
+	}
+	btrfs_free_path(path);
+	return 0;
+}
+
+/*
+ * this gets pages into the page cache and locks them down, it also properly
+ * waits for data=ordered extents to finish before allowing the pages to be
+ * modified.
+ */
+static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
+			 struct page **pages, size_t num_pages,
+			 loff_t pos, unsigned long first_index,
+			 unsigned long last_index, size_t write_bytes)
+{
+	int i;
+	unsigned long index = pos >> PAGE_CACHE_SHIFT;
+	struct inode *inode = fdentry(file)->d_inode;
+	int err = 0;
+	u64 start_pos;
+	u64 last_pos;
+
+	start_pos = pos & ~((u64)root->sectorsize - 1);
+	last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
+
+	if (start_pos > inode->i_size) {
+		err = btrfs_cont_expand(inode, start_pos);
+		if (err)
+			return err;
+	}
+
+	memset(pages, 0, num_pages * sizeof(struct page *));
+again:
+	for (i = 0; i < num_pages; i++) {
+		pages[i] = grab_cache_page(inode->i_mapping, index + i);
+		if (!pages[i]) {
+			err = -ENOMEM;
+			BUG_ON(1);
+		}
+		wait_on_page_writeback(pages[i]);
+	}
+	if (start_pos < inode->i_size) {
+		struct btrfs_ordered_extent *ordered;
+		lock_extent(&BTRFS_I(inode)->io_tree,
+			    start_pos, last_pos - 1, GFP_NOFS);
+		ordered = btrfs_lookup_first_ordered_extent(inode,
+							    last_pos - 1);
+		if (ordered &&
+		    ordered->file_offset + ordered->len > start_pos &&
+		    ordered->file_offset < last_pos) {
+			btrfs_put_ordered_extent(ordered);
+			unlock_extent(&BTRFS_I(inode)->io_tree,
+				      start_pos, last_pos - 1, GFP_NOFS);
+			for (i = 0; i < num_pages; i++) {
+				unlock_page(pages[i]);
+				page_cache_release(pages[i]);
+			}
+			btrfs_wait_ordered_range(inode, start_pos,
+						 last_pos - start_pos);
+			goto again;
+		}
+		if (ordered)
+			btrfs_put_ordered_extent(ordered);
+
+		clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
+				  last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
+				  GFP_NOFS);
+		unlock_extent(&BTRFS_I(inode)->io_tree,
+			      start_pos, last_pos - 1, GFP_NOFS);
+	}
+	for (i = 0; i < num_pages; i++) {
+		clear_page_dirty_for_io(pages[i]);
+		set_page_extent_mapped(pages[i]);
+		WARN_ON(!PageLocked(pages[i]));
+	}
+	return 0;
+}
+
+static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
+				size_t count, loff_t *ppos)
+{
+	loff_t pos;
+	loff_t start_pos;
+	ssize_t num_written = 0;
+	ssize_t err = 0;
+	int ret = 0;
+	struct inode *inode = fdentry(file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct page **pages = NULL;
+	int nrptrs;
+	struct page *pinned[2];
+	unsigned long first_index;
+	unsigned long last_index;
+	int will_write;
+
+	will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
+		      (file->f_flags & O_DIRECT));
+
+	nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
+		     PAGE_CACHE_SIZE / (sizeof(struct page *)));
+	pinned[0] = NULL;
+	pinned[1] = NULL;
+
+	pos = *ppos;
+	start_pos = pos;
+
+	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
+	current->backing_dev_info = inode->i_mapping->backing_dev_info;
+	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
+	if (err)
+		goto out_nolock;
+	if (count == 0)
+		goto out_nolock;
+
+	err = file_remove_suid(file);
+	if (err)
+		goto out_nolock;
+	file_update_time(file);
+
+	pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
+
+	mutex_lock(&inode->i_mutex);
+	BTRFS_I(inode)->sequence++;
+	first_index = pos >> PAGE_CACHE_SHIFT;
+	last_index = (pos + count) >> PAGE_CACHE_SHIFT;
+
+	/*
+	 * there are lots of better ways to do this, but this code
+	 * makes sure the first and last page in the file range are
+	 * up to date and ready for cow
+	 */
+	if ((pos & (PAGE_CACHE_SIZE - 1))) {
+		pinned[0] = grab_cache_page(inode->i_mapping, first_index);
+		if (!PageUptodate(pinned[0])) {
+			ret = btrfs_readpage(NULL, pinned[0]);
+			BUG_ON(ret);
+			wait_on_page_locked(pinned[0]);
+		} else {
+			unlock_page(pinned[0]);
+		}
+	}
+	if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
+		pinned[1] = grab_cache_page(inode->i_mapping, last_index);
+		if (!PageUptodate(pinned[1])) {
+			ret = btrfs_readpage(NULL, pinned[1]);
+			BUG_ON(ret);
+			wait_on_page_locked(pinned[1]);
+		} else {
+			unlock_page(pinned[1]);
+		}
+	}
+
+	while (count > 0) {
+		size_t offset = pos & (PAGE_CACHE_SIZE - 1);
+		size_t write_bytes = min(count, nrptrs *
+					(size_t)PAGE_CACHE_SIZE -
+					 offset);
+		size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
+					PAGE_CACHE_SHIFT;
+
+		WARN_ON(num_pages > nrptrs);
+		memset(pages, 0, sizeof(struct page *) * nrptrs);
+
+		ret = btrfs_check_free_space(root, write_bytes, 0);
+		if (ret)
+			goto out;
+
+		ret = prepare_pages(root, file, pages, num_pages,
+				    pos, first_index, last_index,
+				    write_bytes);
+		if (ret)
+			goto out;
+
+		ret = btrfs_copy_from_user(pos, num_pages,
+					   write_bytes, pages, buf);
+		if (ret) {
+			btrfs_drop_pages(pages, num_pages);
+			goto out;
+		}
+
+		ret = dirty_and_release_pages(NULL, root, file, pages,
+					      num_pages, pos, write_bytes);
+		btrfs_drop_pages(pages, num_pages);
+		if (ret)
+			goto out;
+
+		if (will_write) {
+			btrfs_fdatawrite_range(inode->i_mapping, pos,
+					       pos + write_bytes - 1,
+					       WB_SYNC_NONE);
+		} else {
+			balance_dirty_pages_ratelimited_nr(inode->i_mapping,
+							   num_pages);
+			if (num_pages <
+			    (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
+				btrfs_btree_balance_dirty(root, 1);
+			btrfs_throttle(root);
+		}
+
+		buf += write_bytes;
+		count -= write_bytes;
+		pos += write_bytes;
+		num_written += write_bytes;
+
+		cond_resched();
+	}
+out:
+	mutex_unlock(&inode->i_mutex);
+
+out_nolock:
+	kfree(pages);
+	if (pinned[0])
+		page_cache_release(pinned[0]);
+	if (pinned[1])
+		page_cache_release(pinned[1]);
+	*ppos = pos;
+
+	if (num_written > 0 && will_write) {
+		struct btrfs_trans_handle *trans;
+
+		err = btrfs_wait_ordered_range(inode, start_pos, num_written);
+		if (err)
+			num_written = err;
+
+		if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
+			trans = btrfs_start_transaction(root, 1);
+			ret = btrfs_log_dentry_safe(trans, root,
+						    file->f_dentry);
+			if (ret == 0) {
+				btrfs_sync_log(trans, root);
+				btrfs_end_transaction(trans, root);
+			} else {
+				btrfs_commit_transaction(trans, root);
+			}
+		}
+		if (file->f_flags & O_DIRECT) {
+			invalidate_mapping_pages(inode->i_mapping,
+			      start_pos >> PAGE_CACHE_SHIFT,
+			     (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
+		}
+	}
+	current->backing_dev_info = NULL;
+	return num_written ? num_written : err;
+}
+
+int btrfs_release_file(struct inode *inode, struct file *filp)
+{
+	if (filp->private_data)
+		btrfs_ioctl_trans_end(filp);
+	return 0;
+}
+
+/*
+ * fsync call for both files and directories.  This logs the inode into
+ * the tree log instead of forcing full commits whenever possible.
+ *
+ * It needs to call filemap_fdatawait so that all ordered extent updates are
+ * in the metadata btree are up to date for copying to the log.
+ *
+ * It drops the inode mutex before doing the tree log commit.  This is an
+ * important optimization for directories because holding the mutex prevents
+ * new operations on the dir while we write to disk.
+ */
+int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
+{
+	struct inode *inode = dentry->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret = 0;
+	struct btrfs_trans_handle *trans;
+
+	/*
+	 * check the transaction that last modified this inode
+	 * and see if its already been committed
+	 */
+	if (!BTRFS_I(inode)->last_trans)
+		goto out;
+
+	mutex_lock(&root->fs_info->trans_mutex);
+	if (BTRFS_I(inode)->last_trans <=
+	    root->fs_info->last_trans_committed) {
+		BTRFS_I(inode)->last_trans = 0;
+		mutex_unlock(&root->fs_info->trans_mutex);
+		goto out;
+	}
+	mutex_unlock(&root->fs_info->trans_mutex);
+
+	root->fs_info->tree_log_batch++;
+	filemap_fdatawrite(inode->i_mapping);
+	btrfs_wait_ordered_range(inode, 0, (u64)-1);
+	root->fs_info->tree_log_batch++;
+
+	/*
+	 * ok we haven't committed the transaction yet, lets do a commit
+	 */
+	if (file->private_data)
+		btrfs_ioctl_trans_end(file);
+
+	trans = btrfs_start_transaction(root, 1);
+	if (!trans) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
+	if (ret < 0)
+		goto out;
+
+	/* we've logged all the items and now have a consistent
+	 * version of the file in the log.  It is possible that
+	 * someone will come in and modify the file, but that's
+	 * fine because the log is consistent on disk, and we
+	 * have references to all of the file's extents
+	 *
+	 * It is possible that someone will come in and log the
+	 * file again, but that will end up using the synchronization
+	 * inside btrfs_sync_log to keep things safe.
+	 */
+	mutex_unlock(&file->f_dentry->d_inode->i_mutex);
+
+	if (ret > 0) {
+		ret = btrfs_commit_transaction(trans, root);
+	} else {
+		btrfs_sync_log(trans, root);
+		ret = btrfs_end_transaction(trans, root);
+	}
+	mutex_lock(&file->f_dentry->d_inode->i_mutex);
+out:
+	return ret > 0 ? EIO : ret;
+}
+
+static struct vm_operations_struct btrfs_file_vm_ops = {
+	.fault		= filemap_fault,
+	.page_mkwrite	= btrfs_page_mkwrite,
+};
+
+static int btrfs_file_mmap(struct file	*filp, struct vm_area_struct *vma)
+{
+	vma->vm_ops = &btrfs_file_vm_ops;
+	file_accessed(filp);
+	return 0;
+}
+
+struct file_operations btrfs_file_operations = {
+	.llseek		= generic_file_llseek,
+	.read		= do_sync_read,
+	.aio_read       = generic_file_aio_read,
+	.splice_read	= generic_file_splice_read,
+	.write		= btrfs_file_write,
+	.mmap		= btrfs_file_mmap,
+	.open		= generic_file_open,
+	.release	= btrfs_release_file,
+	.fsync		= btrfs_sync_file,
+	.unlocked_ioctl	= btrfs_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl	= btrfs_ioctl,
+#endif
+};
-- 
1.6.0.2

[PATCH 24/35] Btrfs directory item routines

[Chris Mason]

Directories are indexed in two ways.  The first index is based on a
hash of the file name (crc32c right now) and the second index is based
on a per-directory sequence number.

The first index gives us fast lookups of file names, and the second index
gives us a stable index for readdir/telldir and NFS.  Since the
second index is roughly similar to inode order on disk, it allows
readdir to return things in a good order for reading the files returned.

xattrs are indexed by a hash of the name in a very similar fashion to the
directories, and so their helpers are implemented here as well.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/dir-item.c b/fs/btrfs/dir-item.c
new file mode 100644
index 0000000..926a0b2
--- /dev/null
+++ b/fs/btrfs/dir-item.c
@@ -0,0 +1,386 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include "ctree.h"
+#include "disk-io.h"
+#include "hash.h"
+#include "transaction.h"
+
+/*
+ * insert a name into a directory, doing overflow properly if there is a hash
+ * collision.  data_size indicates how big the item inserted should be.  On
+ * success a struct btrfs_dir_item pointer is returned, otherwise it is
+ * an ERR_PTR.
+ *
+ * The name is not copied into the dir item, you have to do that yourself.
+ */
+static struct btrfs_dir_item *insert_with_overflow(struct btrfs_trans_handle
+						   *trans,
+						   struct btrfs_root *root,
+						   struct btrfs_path *path,
+						   struct btrfs_key *cpu_key,
+						   u32 data_size,
+						   const char *name,
+						   int name_len)
+{
+	int ret;
+	char *ptr;
+	struct btrfs_item *item;
+	struct extent_buffer *leaf;
+
+	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
+	if (ret == -EEXIST) {
+		struct btrfs_dir_item *di;
+		di = btrfs_match_dir_item_name(root, path, name, name_len);
+		if (di)
+			return ERR_PTR(-EEXIST);
+		ret = btrfs_extend_item(trans, root, path, data_size);
+		WARN_ON(ret > 0);
+	}
+	if (ret < 0)
+		return ERR_PTR(ret);
+	WARN_ON(ret > 0);
+	leaf = path->nodes[0];
+	item = btrfs_item_nr(leaf, path->slots[0]);
+	ptr = btrfs_item_ptr(leaf, path->slots[0], char);
+	BUG_ON(data_size > btrfs_item_size(leaf, item));
+	ptr += btrfs_item_size(leaf, item) - data_size;
+	return (struct btrfs_dir_item *)ptr;
+}
+
+/*
+ * xattrs work a lot like directories, this inserts an xattr item
+ * into the tree
+ */
+int btrfs_insert_xattr_item(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, const char *name,
+			    u16 name_len, const void *data, u16 data_len,
+			    u64 dir)
+{
+	int ret = 0;
+	struct btrfs_path *path;
+	struct btrfs_dir_item *dir_item;
+	unsigned long name_ptr, data_ptr;
+	struct btrfs_key key, location;
+	struct btrfs_disk_key disk_key;
+	struct extent_buffer *leaf;
+	u32 data_size;
+
+	key.objectid = dir;
+	btrfs_set_key_type(&key, BTRFS_XATTR_ITEM_KEY);
+	key.offset = btrfs_name_hash(name, name_len);
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	if (name_len + data_len + sizeof(struct btrfs_dir_item) >
+	    BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item))
+		return -ENOSPC;
+
+	data_size = sizeof(*dir_item) + name_len + data_len;
+	dir_item = insert_with_overflow(trans, root, path, &key, data_size,
+					name, name_len);
+	/*
+	 * FIXME: at some point we should handle xattr's that are larger than
+	 * what we can fit in our leaf.  We set location to NULL b/c we arent
+	 * pointing at anything else, that will change if we store the xattr
+	 * data in a separate inode.
+	 */
+	BUG_ON(IS_ERR(dir_item));
+	memset(&location, 0, sizeof(location));
+
+	leaf = path->nodes[0];
+	btrfs_cpu_key_to_disk(&disk_key, &location);
+	btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
+	btrfs_set_dir_type(leaf, dir_item, BTRFS_FT_XATTR);
+	btrfs_set_dir_name_len(leaf, dir_item, name_len);
+	btrfs_set_dir_transid(leaf, dir_item, trans->transid);
+	btrfs_set_dir_data_len(leaf, dir_item, data_len);
+	name_ptr = (unsigned long)(dir_item + 1);
+	data_ptr = (unsigned long)((char *)name_ptr + name_len);
+
+	write_extent_buffer(leaf, name, name_ptr, name_len);
+	write_extent_buffer(leaf, data, data_ptr, data_len);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * insert a directory item in the tree, doing all the magic for
+ * both indexes. 'dir' indicates which objectid to insert it into,
+ * 'location' is the key to stuff into the directory item, 'type' is the
+ * type of the inode we're pointing to, and 'index' is the sequence number
+ * to use for the second index (if one is created).
+ */
+int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
+			  *root, const char *name, int name_len, u64 dir,
+			  struct btrfs_key *location, u8 type, u64 index)
+{
+	int ret = 0;
+	int ret2 = 0;
+	struct btrfs_path *path;
+	struct btrfs_dir_item *dir_item;
+	struct extent_buffer *leaf;
+	unsigned long name_ptr;
+	struct btrfs_key key;
+	struct btrfs_disk_key disk_key;
+	u32 data_size;
+
+	key.objectid = dir;
+	btrfs_set_key_type(&key, BTRFS_DIR_ITEM_KEY);
+	key.offset = btrfs_name_hash(name, name_len);
+	path = btrfs_alloc_path();
+	data_size = sizeof(*dir_item) + name_len;
+	dir_item = insert_with_overflow(trans, root, path, &key, data_size,
+					name, name_len);
+	if (IS_ERR(dir_item)) {
+		ret = PTR_ERR(dir_item);
+		if (ret == -EEXIST)
+			goto second_insert;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	btrfs_cpu_key_to_disk(&disk_key, location);
+	btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
+	btrfs_set_dir_type(leaf, dir_item, type);
+	btrfs_set_dir_data_len(leaf, dir_item, 0);
+	btrfs_set_dir_name_len(leaf, dir_item, name_len);
+	btrfs_set_dir_transid(leaf, dir_item, trans->transid);
+	name_ptr = (unsigned long)(dir_item + 1);
+
+	write_extent_buffer(leaf, name, name_ptr, name_len);
+	btrfs_mark_buffer_dirty(leaf);
+
+second_insert:
+	/* FIXME, use some real flag for selecting the extra index */
+	if (root == root->fs_info->tree_root) {
+		ret = 0;
+		goto out;
+	}
+	btrfs_release_path(root, path);
+
+	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
+	key.offset = index;
+	dir_item = insert_with_overflow(trans, root, path, &key, data_size,
+					name, name_len);
+	if (IS_ERR(dir_item)) {
+		ret2 = PTR_ERR(dir_item);
+		goto out;
+	}
+	leaf = path->nodes[0];
+	btrfs_cpu_key_to_disk(&disk_key, location);
+	btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
+	btrfs_set_dir_type(leaf, dir_item, type);
+	btrfs_set_dir_data_len(leaf, dir_item, 0);
+	btrfs_set_dir_name_len(leaf, dir_item, name_len);
+	btrfs_set_dir_transid(leaf, dir_item, trans->transid);
+	name_ptr = (unsigned long)(dir_item + 1);
+	write_extent_buffer(leaf, name, name_ptr, name_len);
+	btrfs_mark_buffer_dirty(leaf);
+out:
+	btrfs_free_path(path);
+	if (ret)
+		return ret;
+	if (ret2)
+		return ret2;
+	return 0;
+}
+
+/*
+ * lookup a directory item based on name.  'dir' is the objectid
+ * we're searching in, and 'mod' tells us if you plan on deleting the
+ * item (use mod < 0) or changing the options (use mod > 0)
+ */
+struct btrfs_dir_item *btrfs_lookup_dir_item(struct btrfs_trans_handle *trans,
+					     struct btrfs_root *root,
+					     struct btrfs_path *path, u64 dir,
+					     const char *name, int name_len,
+					     int mod)
+{
+	int ret;
+	struct btrfs_key key;
+	int ins_len = mod < 0 ? -1 : 0;
+	int cow = mod != 0;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+
+	key.objectid = dir;
+	btrfs_set_key_type(&key, BTRFS_DIR_ITEM_KEY);
+
+	key.offset = btrfs_name_hash(name, name_len);
+
+	ret = btrfs_search_slot(trans, root, &key, path, ins_len, cow);
+	if (ret < 0)
+		return ERR_PTR(ret);
+	if (ret > 0) {
+		if (path->slots[0] == 0)
+			return NULL;
+		path->slots[0]--;
+	}
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+	if (found_key.objectid != dir ||
+	    btrfs_key_type(&found_key) != BTRFS_DIR_ITEM_KEY ||
+	    found_key.offset != key.offset)
+		return NULL;
+
+	return btrfs_match_dir_item_name(root, path, name, name_len);
+}
+
+/*
+ * lookup a directory item based on index.  'dir' is the objectid
+ * we're searching in, and 'mod' tells us if you plan on deleting the
+ * item (use mod < 0) or changing the options (use mod > 0)
+ *
+ * The name is used to make sure the index really points to the name you were
+ * looking for.
+ */
+struct btrfs_dir_item *
+btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path, u64 dir,
+			    u64 objectid, const char *name, int name_len,
+			    int mod)
+{
+	int ret;
+	struct btrfs_key key;
+	int ins_len = mod < 0 ? -1 : 0;
+	int cow = mod != 0;
+
+	key.objectid = dir;
+	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
+	key.offset = objectid;
+
+	ret = btrfs_search_slot(trans, root, &key, path, ins_len, cow);
+	if (ret < 0)
+		return ERR_PTR(ret);
+	if (ret > 0)
+		return ERR_PTR(-ENOENT);
+	return btrfs_match_dir_item_name(root, path, name, name_len);
+}
+
+struct btrfs_dir_item *btrfs_lookup_xattr(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path, u64 dir,
+					  const char *name, u16 name_len,
+					  int mod)
+{
+	int ret;
+	struct btrfs_key key;
+	int ins_len = mod < 0 ? -1 : 0;
+	int cow = mod != 0;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+
+	key.objectid = dir;
+	btrfs_set_key_type(&key, BTRFS_XATTR_ITEM_KEY);
+	key.offset = btrfs_name_hash(name, name_len);
+	ret = btrfs_search_slot(trans, root, &key, path, ins_len, cow);
+	if (ret < 0)
+		return ERR_PTR(ret);
+	if (ret > 0) {
+		if (path->slots[0] == 0)
+			return NULL;
+		path->slots[0]--;
+	}
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+	if (found_key.objectid != dir ||
+	    btrfs_key_type(&found_key) != BTRFS_XATTR_ITEM_KEY ||
+	    found_key.offset != key.offset)
+		return NULL;
+
+	return btrfs_match_dir_item_name(root, path, name, name_len);
+}
+
+/*
+ * helper function to look at the directory item pointed to by 'path'
+ * this walks through all the entries in a dir item and finds one
+ * for a specific name.
+ */
+struct btrfs_dir_item *btrfs_match_dir_item_name(struct btrfs_root *root,
+			      struct btrfs_path *path,
+			      const char *name, int name_len)
+{
+	struct btrfs_dir_item *dir_item;
+	unsigned long name_ptr;
+	u32 total_len;
+	u32 cur = 0;
+	u32 this_len;
+	struct extent_buffer *leaf;
+
+	leaf = path->nodes[0];
+	dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
+	total_len = btrfs_item_size_nr(leaf, path->slots[0]);
+	while (cur < total_len) {
+		this_len = sizeof(*dir_item) +
+			btrfs_dir_name_len(leaf, dir_item) +
+			btrfs_dir_data_len(leaf, dir_item);
+		name_ptr = (unsigned long)(dir_item + 1);
+
+		if (btrfs_dir_name_len(leaf, dir_item) == name_len &&
+		    memcmp_extent_buffer(leaf, name, name_ptr, name_len) == 0)
+			return dir_item;
+
+		cur += this_len;
+		dir_item = (struct btrfs_dir_item *)((char *)dir_item +
+						     this_len);
+	}
+	return NULL;
+}
+
+/*
+ * given a pointer into a directory item, delete it.  This
+ * handles items that have more than one entry in them.
+ */
+int btrfs_delete_one_dir_name(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct btrfs_path *path,
+			      struct btrfs_dir_item *di)
+{
+
+	struct extent_buffer *leaf;
+	u32 sub_item_len;
+	u32 item_len;
+	int ret = 0;
+
+	leaf = path->nodes[0];
+	sub_item_len = sizeof(*di) + btrfs_dir_name_len(leaf, di) +
+		btrfs_dir_data_len(leaf, di);
+	item_len = btrfs_item_size_nr(leaf, path->slots[0]);
+	if (sub_item_len == item_len) {
+		ret = btrfs_del_item(trans, root, path);
+	} else {
+		/* MARKER */
+		unsigned long ptr = (unsigned long)di;
+		unsigned long start;
+
+		start = btrfs_item_ptr_offset(leaf, path->slots[0]);
+		memmove_extent_buffer(leaf, ptr, ptr + sub_item_len,
+			item_len - (ptr + sub_item_len - start));
+		ret = btrfs_truncate_item(trans, root, path,
+					  item_len - sub_item_len, 1);
+	}
+	return 0;
+}
-- 
1.6.0.2

[PATCH 25/35] Btrfs main metadata header file

[Chris Mason]

This file collects most of the function declarations used for
btrfs metadata.

Most of the on-disk data structures are also here.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
new file mode 100644
index 0000000..eee060f
--- /dev/null
+++ b/fs/btrfs/ctree.h
@@ -0,0 +1,2129 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_CTREE__
+#define __BTRFS_CTREE__
+
+#include <linux/version.h>
+#include <linux/mm.h>
+#include <linux/highmem.h>
+#include <linux/fs.h>
+#include <linux/completion.h>
+#include <linux/backing-dev.h>
+#include <linux/wait.h>
+#include <asm/kmap_types.h>
+#include "extent_io.h"
+#include "extent_map.h"
+#include "async-thread.h"
+
+struct btrfs_trans_handle;
+struct btrfs_transaction;
+extern struct kmem_cache *btrfs_trans_handle_cachep;
+extern struct kmem_cache *btrfs_transaction_cachep;
+extern struct kmem_cache *btrfs_bit_radix_cachep;
+extern struct kmem_cache *btrfs_path_cachep;
+struct btrfs_ordered_sum;
+
+#define BTRFS_MAGIC "_BHRfS_M"
+
+#define BTRFS_ACL_NOT_CACHED    ((void *)-1)
+
+#ifdef CONFIG_LOCKDEP
+# define BTRFS_MAX_LEVEL 7
+#else
+# define BTRFS_MAX_LEVEL 8
+#endif
+
+/* holds pointers to all of the tree roots */
+#define BTRFS_ROOT_TREE_OBJECTID 1ULL
+
+/* stores information about which extents are in use, and reference counts */
+#define BTRFS_EXTENT_TREE_OBJECTID 2ULL
+
+/*
+ * chunk tree stores translations from logical -> physical block numbering
+ * the super block points to the chunk tree
+ */
+#define BTRFS_CHUNK_TREE_OBJECTID 3ULL
+
+/*
+ * stores information about which areas of a given device are in use.
+ * one per device.  The tree of tree roots points to the device tree
+ */
+#define BTRFS_DEV_TREE_OBJECTID 4ULL
+
+/* one per subvolume, storing files and directories */
+#define BTRFS_FS_TREE_OBJECTID 5ULL
+
+/* directory objectid inside the root tree */
+#define BTRFS_ROOT_TREE_DIR_OBJECTID 6ULL
+
+/* holds checksums of all the data extents */
+#define BTRFS_CSUM_TREE_OBJECTID 7ULL
+
+/* orhpan objectid for tracking unlinked/truncated files */
+#define BTRFS_ORPHAN_OBJECTID -5ULL
+
+/* does write ahead logging to speed up fsyncs */
+#define BTRFS_TREE_LOG_OBJECTID -6ULL
+#define BTRFS_TREE_LOG_FIXUP_OBJECTID -7ULL
+
+/* for space balancing */
+#define BTRFS_TREE_RELOC_OBJECTID -8ULL
+#define BTRFS_DATA_RELOC_TREE_OBJECTID -9ULL
+
+/*
+ * extent checksums all have this objectid
+ * this allows them to share the logging tree
+ * for fsyncs
+ */
+#define BTRFS_EXTENT_CSUM_OBJECTID -10ULL
+
+/* dummy objectid represents multiple objectids */
+#define BTRFS_MULTIPLE_OBJECTIDS -255ULL
+
+/*
+ * All files have objectids in this range.
+ */
+#define BTRFS_FIRST_FREE_OBJECTID 256ULL
+#define BTRFS_LAST_FREE_OBJECTID -256ULL
+#define BTRFS_FIRST_CHUNK_TREE_OBJECTID 256ULL
+
+
+/*
+ * the device items go into the chunk tree.  The key is in the form
+ * [ 1 BTRFS_DEV_ITEM_KEY device_id ]
+ */
+#define BTRFS_DEV_ITEMS_OBJECTID 1ULL
+
+/*
+ * we can actually store much bigger names, but lets not confuse the rest
+ * of linux
+ */
+#define BTRFS_NAME_LEN 255
+
+/* 32 bytes in various csum fields */
+#define BTRFS_CSUM_SIZE 32
+
+/* csum types */
+#define BTRFS_CSUM_TYPE_CRC32	0
+
+static int btrfs_csum_sizes[] = { 4, 0 };
+
+/* four bytes for CRC32 */
+#define BTRFS_EMPTY_DIR_SIZE 0
+
+#define BTRFS_FT_UNKNOWN	0
+#define BTRFS_FT_REG_FILE	1
+#define BTRFS_FT_DIR		2
+#define BTRFS_FT_CHRDEV		3
+#define BTRFS_FT_BLKDEV		4
+#define BTRFS_FT_FIFO		5
+#define BTRFS_FT_SOCK		6
+#define BTRFS_FT_SYMLINK	7
+#define BTRFS_FT_XATTR		8
+#define BTRFS_FT_MAX		9
+
+/*
+ * the key defines the order in the tree, and so it also defines (optimal)
+ * block layout.  objectid corresonds to the inode number.  The flags
+ * tells us things about the object, and is a kind of stream selector.
+ * so for a given inode, keys with flags of 1 might refer to the inode
+ * data, flags of 2 may point to file data in the btree and flags == 3
+ * may point to extents.
+ *
+ * offset is the starting byte offset for this key in the stream.
+ *
+ * btrfs_disk_key is in disk byte order.  struct btrfs_key is always
+ * in cpu native order.  Otherwise they are identical and their sizes
+ * should be the same (ie both packed)
+ */
+struct btrfs_disk_key {
+	__le64 objectid;
+	u8 type;
+	__le64 offset;
+} __attribute__ ((__packed__));
+
+struct btrfs_key {
+	u64 objectid;
+	u8 type;
+	u64 offset;
+} __attribute__ ((__packed__));
+
+struct btrfs_mapping_tree {
+	struct extent_map_tree map_tree;
+};
+
+#define BTRFS_UUID_SIZE 16
+struct btrfs_dev_item {
+	/* the internal btrfs device id */
+	__le64 devid;
+
+	/* size of the device */
+	__le64 total_bytes;
+
+	/* bytes used */
+	__le64 bytes_used;
+
+	/* optimal io alignment for this device */
+	__le32 io_align;
+
+	/* optimal io width for this device */
+	__le32 io_width;
+
+	/* minimal io size for this device */
+	__le32 sector_size;
+
+	/* type and info about this device */
+	__le64 type;
+
+	/* expected generation for this device */
+	__le64 generation;
+
+	/*
+	 * starting byte of this partition on the device,
+	 * to allowr for stripe alignment in the future
+	 */
+	__le64 start_offset;
+
+	/* grouping information for allocation decisions */
+	__le32 dev_group;
+
+	/* seek speed 0-100 where 100 is fastest */
+	u8 seek_speed;
+
+	/* bandwidth 0-100 where 100 is fastest */
+	u8 bandwidth;
+
+	/* btrfs generated uuid for this device */
+	u8 uuid[BTRFS_UUID_SIZE];
+
+	/* uuid of FS who owns this device */
+	u8 fsid[BTRFS_UUID_SIZE];
+} __attribute__ ((__packed__));
+
+struct btrfs_stripe {
+	__le64 devid;
+	__le64 offset;
+	u8 dev_uuid[BTRFS_UUID_SIZE];
+} __attribute__ ((__packed__));
+
+struct btrfs_chunk {
+	/* size of this chunk in bytes */
+	__le64 length;
+
+	/* objectid of the root referencing this chunk */
+	__le64 owner;
+
+	__le64 stripe_len;
+	__le64 type;
+
+	/* optimal io alignment for this chunk */
+	__le32 io_align;
+
+	/* optimal io width for this chunk */
+	__le32 io_width;
+
+	/* minimal io size for this chunk */
+	__le32 sector_size;
+
+	/* 2^16 stripes is quite a lot, a second limit is the size of a single
+	 * item in the btree
+	 */
+	__le16 num_stripes;
+
+	/* sub stripes only matter for raid10 */
+	__le16 sub_stripes;
+	struct btrfs_stripe stripe;
+	/* additional stripes go here */
+} __attribute__ ((__packed__));
+
+static inline unsigned long btrfs_chunk_item_size(int num_stripes)
+{
+	BUG_ON(num_stripes == 0);
+	return sizeof(struct btrfs_chunk) +
+		sizeof(struct btrfs_stripe) * (num_stripes - 1);
+}
+
+#define BTRFS_FSID_SIZE 16
+#define BTRFS_HEADER_FLAG_WRITTEN (1 << 0)
+
+/*
+ * every tree block (leaf or node) starts with this header.
+ */
+struct btrfs_header {
+	/* these first four must match the super block */
+	u8 csum[BTRFS_CSUM_SIZE];
+	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
+	__le64 bytenr; /* which block this node is supposed to live in */
+	__le64 flags;
+
+	/* allowed to be different from the super from here on down */
+	u8 chunk_tree_uuid[BTRFS_UUID_SIZE];
+	__le64 generation;
+	__le64 owner;
+	__le32 nritems;
+	u8 level;
+} __attribute__ ((__packed__));
+
+#define BTRFS_NODEPTRS_PER_BLOCK(r) (((r)->nodesize - \
+				      sizeof(struct btrfs_header)) / \
+				     sizeof(struct btrfs_key_ptr))
+#define __BTRFS_LEAF_DATA_SIZE(bs) ((bs) - sizeof(struct btrfs_header))
+#define BTRFS_LEAF_DATA_SIZE(r) (__BTRFS_LEAF_DATA_SIZE(r->leafsize))
+#define BTRFS_MAX_INLINE_DATA_SIZE(r) (BTRFS_LEAF_DATA_SIZE(r) - \
+					sizeof(struct btrfs_item) - \
+					sizeof(struct btrfs_file_extent_item))
+
+#define BTRFS_SUPER_FLAG_SEEDING (1ULL << 32)
+
+/*
+ * this is a very generous portion of the super block, giving us
+ * room to translate 14 chunks with 3 stripes each.
+ */
+#define BTRFS_SYSTEM_CHUNK_ARRAY_SIZE 2048
+#define BTRFS_LABEL_SIZE 256
+
+/*
+ * the super block basically lists the main trees of the FS
+ * it currently lacks any block count etc etc
+ */
+struct btrfs_super_block {
+	u8 csum[BTRFS_CSUM_SIZE];
+	/* the first 4 fields must match struct btrfs_header */
+	u8 fsid[BTRFS_FSID_SIZE];    /* FS specific uuid */
+	__le64 bytenr; /* this block number */
+	__le64 flags;
+
+	/* allowed to be different from the btrfs_header from here own down */
+	__le64 magic;
+	__le64 generation;
+	__le64 root;
+	__le64 chunk_root;
+	__le64 log_root;
+
+	/* this will help find the new super based on the log root */
+	__le64 log_root_transid;
+	__le64 total_bytes;
+	__le64 bytes_used;
+	__le64 root_dir_objectid;
+	__le64 num_devices;
+	__le32 sectorsize;
+	__le32 nodesize;
+	__le32 leafsize;
+	__le32 stripesize;
+	__le32 sys_chunk_array_size;
+	__le64 chunk_root_generation;
+	__le64 compat_flags;
+	__le64 compat_ro_flags;
+	__le64 incompat_flags;
+	__le16 csum_type;
+	u8 root_level;
+	u8 chunk_root_level;
+	u8 log_root_level;
+	struct btrfs_dev_item dev_item;
+
+	char label[BTRFS_LABEL_SIZE];
+
+	/* future expansion */
+	__le64 reserved[32];
+	u8 sys_chunk_array[BTRFS_SYSTEM_CHUNK_ARRAY_SIZE];
+} __attribute__ ((__packed__));
+
+/*
+ * Compat flags that we support.  If any incompat flags are set other than the
+ * ones specified below then we will fail to mount
+ */
+#define BTRFS_FEATURE_COMPAT_SUPP	0x0
+#define BTRFS_FEATURE_COMPAT_RO_SUPP	0x0
+#define BTRFS_FEATURE_INCOMPAT_SUPP	0x0
+
+/*
+ * A leaf is full of items. offset and size tell us where to find
+ * the item in the leaf (relative to the start of the data area)
+ */
+struct btrfs_item {
+	struct btrfs_disk_key key;
+	__le32 offset;
+	__le32 size;
+} __attribute__ ((__packed__));
+
+/*
+ * leaves have an item area and a data area:
+ * [item0, item1....itemN] [free space] [dataN...data1, data0]
+ *
+ * The data is separate from the items to get the keys closer together
+ * during searches.
+ */
+struct btrfs_leaf {
+	struct btrfs_header header;
+	struct btrfs_item items[];
+} __attribute__ ((__packed__));
+
+/*
+ * all non-leaf blocks are nodes, they hold only keys and pointers to
+ * other blocks
+ */
+struct btrfs_key_ptr {
+	struct btrfs_disk_key key;
+	__le64 blockptr;
+	__le64 generation;
+} __attribute__ ((__packed__));
+
+struct btrfs_node {
+	struct btrfs_header header;
+	struct btrfs_key_ptr ptrs[];
+} __attribute__ ((__packed__));
+
+/*
+ * btrfs_paths remember the path taken from the root down to the leaf.
+ * level 0 is always the leaf, and nodes[1...BTRFS_MAX_LEVEL] will point
+ * to any other levels that are present.
+ *
+ * The slots array records the index of the item or block pointer
+ * used while walking the tree.
+ */
+struct btrfs_path {
+	struct extent_buffer *nodes[BTRFS_MAX_LEVEL];
+	int slots[BTRFS_MAX_LEVEL];
+	/* if there is real range locking, this locks field will change */
+	int locks[BTRFS_MAX_LEVEL];
+	int reada;
+	/* keep some upper locks as we walk down */
+	int keep_locks;
+	int skip_locking;
+	int lowest_level;
+
+	/*
+	 * set by btrfs_split_item, tells search_slot to keep all locks
+	 * and to force calls to keep space in the nodes
+	 */
+	int search_for_split;
+};
+
+/*
+ * items in the extent btree are used to record the objectid of the
+ * owner of the block and the number of references
+ */
+struct btrfs_extent_item {
+	__le32 refs;
+} __attribute__ ((__packed__));
+
+struct btrfs_extent_ref {
+	__le64 root;
+	__le64 generation;
+	__le64 objectid;
+	__le32 num_refs;
+} __attribute__ ((__packed__));
+
+/* dev extents record free space on individual devices.  The owner
+ * field points back to the chunk allocation mapping tree that allocated
+ * the extent.  The chunk tree uuid field is a way to double check the owner
+ */
+struct btrfs_dev_extent {
+	__le64 chunk_tree;
+	__le64 chunk_objectid;
+	__le64 chunk_offset;
+	__le64 length;
+	u8 chunk_tree_uuid[BTRFS_UUID_SIZE];
+} __attribute__ ((__packed__));
+
+struct btrfs_inode_ref {
+	__le64 index;
+	__le16 name_len;
+	/* name goes here */
+} __attribute__ ((__packed__));
+
+struct btrfs_timespec {
+	__le64 sec;
+	__le32 nsec;
+} __attribute__ ((__packed__));
+
+typedef enum {
+	BTRFS_COMPRESS_NONE = 0,
+	BTRFS_COMPRESS_ZLIB = 1,
+	BTRFS_COMPRESS_LAST = 2,
+} btrfs_compression_type;
+
+/* we don't understand any encryption methods right now */
+typedef enum {
+	BTRFS_ENCRYPTION_NONE = 0,
+	BTRFS_ENCRYPTION_LAST = 1,
+} btrfs_encryption_type;
+
+struct btrfs_inode_item {
+	/* nfs style generation number */
+	__le64 generation;
+	/* transid that last touched this inode */
+	__le64 transid;
+	__le64 size;
+	__le64 nbytes;
+	__le64 block_group;
+	__le32 nlink;
+	__le32 uid;
+	__le32 gid;
+	__le32 mode;
+	__le64 rdev;
+	__le64 flags;
+
+	/* modification sequence number for NFS */
+	__le64 sequence;
+
+	/*
+	 * a little future expansion, for more than this we can
+	 * just grow the inode item and version it
+	 */
+	__le64 reserved[4];
+	struct btrfs_timespec atime;
+	struct btrfs_timespec ctime;
+	struct btrfs_timespec mtime;
+	struct btrfs_timespec otime;
+} __attribute__ ((__packed__));
+
+struct btrfs_dir_log_item {
+	__le64 end;
+} __attribute__ ((__packed__));
+
+struct btrfs_dir_item {
+	struct btrfs_disk_key location;
+	__le64 transid;
+	__le16 data_len;
+	__le16 name_len;
+	u8 type;
+} __attribute__ ((__packed__));
+
+struct btrfs_root_item {
+	struct btrfs_inode_item inode;
+	__le64 generation;
+	__le64 root_dirid;
+	__le64 bytenr;
+	__le64 byte_limit;
+	__le64 bytes_used;
+	__le64 last_snapshot;
+	__le64 flags;
+	__le32 refs;
+	struct btrfs_disk_key drop_progress;
+	u8 drop_level;
+	u8 level;
+} __attribute__ ((__packed__));
+
+/*
+ * this is used for both forward and backward root refs
+ */
+struct btrfs_root_ref {
+	__le64 dirid;
+	__le64 sequence;
+	__le16 name_len;
+} __attribute__ ((__packed__));
+
+#define BTRFS_FILE_EXTENT_INLINE 0
+#define BTRFS_FILE_EXTENT_REG 1
+#define BTRFS_FILE_EXTENT_PREALLOC 2
+
+struct btrfs_file_extent_item {
+	/*
+	 * transaction id that created this extent
+	 */
+	__le64 generation;
+	/*
+	 * max number of bytes to hold this extent in ram
+	 * when we split a compressed extent we can't know how big
+	 * each of the resulting pieces will be.  So, this is
+	 * an upper limit on the size of the extent in ram instead of
+	 * an exact limit.
+	 */
+	__le64 ram_bytes;
+
+	/*
+	 * 32 bits for the various ways we might encode the data,
+	 * including compression and encryption.  If any of these
+	 * are set to something a given disk format doesn't understand
+	 * it is treated like an incompat flag for reading and writing,
+	 * but not for stat.
+	 */
+	u8 compression;
+	u8 encryption;
+	__le16 other_encoding; /* spare for later use */
+
+	/* are we inline data or a real extent? */
+	u8 type;
+
+	/*
+	 * disk space consumed by the extent, checksum blocks are included
+	 * in these numbers
+	 */
+	__le64 disk_bytenr;
+	__le64 disk_num_bytes;
+	/*
+	 * the logical offset in file blocks (no csums)
+	 * this extent record is for.  This allows a file extent to point
+	 * into the middle of an existing extent on disk, sharing it
+	 * between two snapshots (useful if some bytes in the middle of the
+	 * extent have changed
+	 */
+	__le64 offset;
+	/*
+	 * the logical number of file blocks (no csums included).  This
+	 * always reflects the size uncompressed and without encoding.
+	 */
+	__le64 num_bytes;
+
+} __attribute__ ((__packed__));
+
+struct btrfs_csum_item {
+	u8 csum;
+} __attribute__ ((__packed__));
+
+/* different types of block groups (and chunks) */
+#define BTRFS_BLOCK_GROUP_DATA     (1 << 0)
+#define BTRFS_BLOCK_GROUP_SYSTEM   (1 << 1)
+#define BTRFS_BLOCK_GROUP_METADATA (1 << 2)
+#define BTRFS_BLOCK_GROUP_RAID0    (1 << 3)
+#define BTRFS_BLOCK_GROUP_RAID1    (1 << 4)
+#define BTRFS_BLOCK_GROUP_DUP	   (1 << 5)
+#define BTRFS_BLOCK_GROUP_RAID10   (1 << 6)
+
+struct btrfs_block_group_item {
+	__le64 used;
+	__le64 chunk_objectid;
+	__le64 flags;
+} __attribute__ ((__packed__));
+
+struct btrfs_space_info {
+	u64 flags;
+	u64 total_bytes;
+	u64 bytes_used;
+	u64 bytes_pinned;
+	u64 bytes_reserved;
+	u64 bytes_readonly;
+	int full;
+	int force_alloc;
+	struct list_head list;
+
+	/* for block groups in our same type */
+	struct list_head block_groups;
+	spinlock_t lock;
+	struct rw_semaphore groups_sem;
+};
+
+struct btrfs_free_space {
+	struct rb_node bytes_index;
+	struct rb_node offset_index;
+	u64 offset;
+	u64 bytes;
+};
+
+struct btrfs_block_group_cache {
+	struct btrfs_key key;
+	struct btrfs_block_group_item item;
+	spinlock_t lock;
+	struct mutex alloc_mutex;
+	struct mutex cache_mutex;
+	u64 pinned;
+	u64 reserved;
+	u64 flags;
+	int cached;
+	int ro;
+	int dirty;
+
+	struct btrfs_space_info *space_info;
+
+	/* free space cache stuff */
+	struct rb_root free_space_bytes;
+	struct rb_root free_space_offset;
+
+	/* block group cache stuff */
+	struct rb_node cache_node;
+
+	/* for block groups in the same raid type */
+	struct list_head list;
+
+	/* usage count */
+	atomic_t count;
+};
+
+struct btrfs_leaf_ref_tree {
+	struct rb_root root;
+	struct list_head list;
+	spinlock_t lock;
+};
+
+struct btrfs_device;
+struct btrfs_fs_devices;
+struct btrfs_fs_info {
+	u8 fsid[BTRFS_FSID_SIZE];
+	u8 chunk_tree_uuid[BTRFS_UUID_SIZE];
+	struct btrfs_root *extent_root;
+	struct btrfs_root *tree_root;
+	struct btrfs_root *chunk_root;
+	struct btrfs_root *dev_root;
+	struct btrfs_root *fs_root;
+	struct btrfs_root *csum_root;
+
+	/* the log root tree is a directory of all the other log roots */
+	struct btrfs_root *log_root_tree;
+	struct radix_tree_root fs_roots_radix;
+
+	/* block group cache stuff */
+	spinlock_t block_group_cache_lock;
+	struct rb_root block_group_cache_tree;
+
+	struct extent_io_tree pinned_extents;
+	struct extent_io_tree pending_del;
+	struct extent_io_tree extent_ins;
+
+	/* logical->physical extent mapping */
+	struct btrfs_mapping_tree mapping_tree;
+
+	u64 generation;
+	u64 last_trans_committed;
+	u64 last_trans_new_blockgroup;
+	u64 open_ioctl_trans;
+	unsigned long mount_opt;
+	u64 max_extent;
+	u64 max_inline;
+	u64 alloc_start;
+	struct btrfs_transaction *running_transaction;
+	wait_queue_head_t transaction_throttle;
+	wait_queue_head_t transaction_wait;
+
+	wait_queue_head_t async_submit_wait;
+	wait_queue_head_t tree_log_wait;
+
+	struct btrfs_super_block super_copy;
+	struct btrfs_super_block super_for_commit;
+	struct block_device *__bdev;
+	struct super_block *sb;
+	struct inode *btree_inode;
+	struct backing_dev_info bdi;
+	spinlock_t hash_lock;
+	struct mutex trans_mutex;
+	struct mutex tree_log_mutex;
+	struct mutex transaction_kthread_mutex;
+	struct mutex cleaner_mutex;
+	struct mutex extent_ins_mutex;
+	struct mutex pinned_mutex;
+	struct mutex chunk_mutex;
+	struct mutex drop_mutex;
+	struct mutex volume_mutex;
+	struct mutex tree_reloc_mutex;
+	struct list_head trans_list;
+	struct list_head hashers;
+	struct list_head dead_roots;
+
+	atomic_t nr_async_submits;
+	atomic_t async_submit_draining;
+	atomic_t nr_async_bios;
+	atomic_t async_delalloc_pages;
+	atomic_t tree_log_writers;
+	atomic_t tree_log_commit;
+	unsigned long tree_log_batch;
+	u64 tree_log_transid;
+
+	/*
+	 * this is used by the balancing code to wait for all the pending
+	 * ordered extents
+	 */
+	spinlock_t ordered_extent_lock;
+	struct list_head ordered_extents;
+	struct list_head delalloc_inodes;
+
+	/*
+	 * there is a pool of worker threads for checksumming during writes
+	 * and a pool for checksumming after reads.  This is because readers
+	 * can run with FS locks held, and the writers may be waiting for
+	 * those locks.  We don't want ordering in the pending list to cause
+	 * deadlocks, and so the two are serviced separately.
+	 *
+	 * A third pool does submit_bio to avoid deadlocking with the other
+	 * two
+	 */
+	struct btrfs_workers workers;
+	struct btrfs_workers delalloc_workers;
+	struct btrfs_workers endio_workers;
+	struct btrfs_workers endio_meta_workers;
+	struct btrfs_workers endio_meta_write_workers;
+	struct btrfs_workers endio_write_workers;
+	struct btrfs_workers submit_workers;
+	/*
+	 * fixup workers take dirty pages that didn't properly go through
+	 * the cow mechanism and make them safe to write.  It happens
+	 * for the sys_munmap function call path
+	 */
+	struct btrfs_workers fixup_workers;
+	struct task_struct *transaction_kthread;
+	struct task_struct *cleaner_kthread;
+	int thread_pool_size;
+
+	/* tree relocation relocated fields */
+	struct list_head dead_reloc_roots;
+	struct btrfs_leaf_ref_tree reloc_ref_tree;
+	struct btrfs_leaf_ref_tree shared_ref_tree;
+
+	struct kobject super_kobj;
+	struct completion kobj_unregister;
+	int do_barriers;
+	int closing;
+	int log_root_recovering;
+	atomic_t throttles;
+	atomic_t throttle_gen;
+
+	u64 total_pinned;
+	struct list_head dirty_cowonly_roots;
+
+	struct btrfs_fs_devices *fs_devices;
+	struct list_head space_info;
+	spinlock_t delalloc_lock;
+	spinlock_t new_trans_lock;
+	u64 delalloc_bytes;
+	u64 last_alloc;
+	u64 last_data_alloc;
+
+	spinlock_t ref_cache_lock;
+	u64 total_ref_cache_size;
+
+	u64 avail_data_alloc_bits;
+	u64 avail_metadata_alloc_bits;
+	u64 avail_system_alloc_bits;
+	u64 data_alloc_profile;
+	u64 metadata_alloc_profile;
+	u64 system_alloc_profile;
+
+	void *bdev_holder;
+};
+
+/*
+ * in ram representation of the tree.  extent_root is used for all allocations
+ * and for the extent tree extent_root root.
+ */
+struct btrfs_dirty_root;
+struct btrfs_root {
+	struct extent_buffer *node;
+
+	/* the node lock is held while changing the node pointer */
+	spinlock_t node_lock;
+
+	struct extent_buffer *commit_root;
+	struct btrfs_leaf_ref_tree *ref_tree;
+	struct btrfs_leaf_ref_tree ref_tree_struct;
+	struct btrfs_dirty_root *dirty_root;
+	struct btrfs_root *log_root;
+	struct btrfs_root *reloc_root;
+
+	struct btrfs_root_item root_item;
+	struct btrfs_key root_key;
+	struct btrfs_fs_info *fs_info;
+	struct extent_io_tree dirty_log_pages;
+
+	struct kobject root_kobj;
+	struct completion kobj_unregister;
+	struct mutex objectid_mutex;
+	struct mutex log_mutex;
+
+	u64 objectid;
+	u64 last_trans;
+
+	/* data allocations are done in sectorsize units */
+	u32 sectorsize;
+
+	/* node allocations are done in nodesize units */
+	u32 nodesize;
+
+	/* leaf allocations are done in leafsize units */
+	u32 leafsize;
+
+	u32 stripesize;
+
+	u32 type;
+	u64 highest_inode;
+	u64 last_inode_alloc;
+	int ref_cows;
+	int track_dirty;
+	u64 defrag_trans_start;
+	struct btrfs_key defrag_progress;
+	struct btrfs_key defrag_max;
+	int defrag_running;
+	int defrag_level;
+	char *name;
+	int in_sysfs;
+
+	/* the dirty list is only used by non-reference counted roots */
+	struct list_head dirty_list;
+
+	spinlock_t list_lock;
+	struct list_head dead_list;
+	struct list_head orphan_list;
+
+	/*
+	 * right now this just gets used so that a root has its own devid
+	 * for stat.  It may be used for more later
+	 */
+	struct super_block anon_super;
+};
+
+/*
+
+ * inode items have the data typically returned from stat and store other
+ * info about object characteristics.  There is one for every file and dir in
+ * the FS
+ */
+#define BTRFS_INODE_ITEM_KEY		1
+#define BTRFS_INODE_REF_KEY		12
+#define BTRFS_XATTR_ITEM_KEY		24
+#define BTRFS_ORPHAN_ITEM_KEY		48
+/* reserve 2-15 close to the inode for later flexibility */
+
+/*
+ * dir items are the name -> inode pointers in a directory.  There is one
+ * for every name in a directory.
+ */
+#define BTRFS_DIR_LOG_ITEM_KEY  60
+#define BTRFS_DIR_LOG_INDEX_KEY 72
+#define BTRFS_DIR_ITEM_KEY	84
+#define BTRFS_DIR_INDEX_KEY	96
+/*
+ * extent data is for file data
+ */
+#define BTRFS_EXTENT_DATA_KEY	108
+
+/*
+ * extent csums are stored in a separate tree and hold csums for
+ * an entire extent on disk.
+ */
+#define BTRFS_EXTENT_CSUM_KEY	128
+
+/*
+ * root items point to tree roots.  There are typically in the root
+ * tree used by the super block to find all the other trees
+ */
+#define BTRFS_ROOT_ITEM_KEY	132
+
+/*
+ * root backrefs tie subvols and snapshots to the directory entries that
+ * reference them
+ */
+#define BTRFS_ROOT_BACKREF_KEY	144
+
+/*
+ * root refs make a fast index for listing all of the snapshots and
+ * subvolumes referenced by a given root.  They point directly to the
+ * directory item in the root that references the subvol
+ */
+#define BTRFS_ROOT_REF_KEY	156
+
+/*
+ * extent items are in the extent map tree.  These record which blocks
+ * are used, and how many references there are to each block
+ */
+#define BTRFS_EXTENT_ITEM_KEY	168
+#define BTRFS_EXTENT_REF_KEY	180
+
+/*
+ * block groups give us hints into the extent allocation trees.  Which
+ * blocks are free etc etc
+ */
+#define BTRFS_BLOCK_GROUP_ITEM_KEY 192
+
+#define BTRFS_DEV_EXTENT_KEY	204
+#define BTRFS_DEV_ITEM_KEY	216
+#define BTRFS_CHUNK_ITEM_KEY	228
+
+/*
+ * string items are for debugging.  They just store a short string of
+ * data in the FS
+ */
+#define BTRFS_STRING_ITEM_KEY	253
+
+#define BTRFS_MOUNT_NODATASUM		(1 << 0)
+#define BTRFS_MOUNT_NODATACOW		(1 << 1)
+#define BTRFS_MOUNT_NOBARRIER		(1 << 2)
+#define BTRFS_MOUNT_SSD			(1 << 3)
+#define BTRFS_MOUNT_DEGRADED		(1 << 4)
+#define BTRFS_MOUNT_COMPRESS		(1 << 5)
+
+#define btrfs_clear_opt(o, opt)		((o) &= ~BTRFS_MOUNT_##opt)
+#define btrfs_set_opt(o, opt)		((o) |= BTRFS_MOUNT_##opt)
+#define btrfs_test_opt(root, opt)	((root)->fs_info->mount_opt & \
+					 BTRFS_MOUNT_##opt)
+/*
+ * Inode flags
+ */
+#define BTRFS_INODE_NODATASUM		(1 << 0)
+#define BTRFS_INODE_NODATACOW		(1 << 1)
+#define BTRFS_INODE_READONLY		(1 << 2)
+#define BTRFS_INODE_NOCOMPRESS		(1 << 3)
+#define BTRFS_INODE_PREALLOC		(1 << 4)
+#define btrfs_clear_flag(inode, flag)	(BTRFS_I(inode)->flags &= \
+					 ~BTRFS_INODE_##flag)
+#define btrfs_set_flag(inode, flag)	(BTRFS_I(inode)->flags |= \
+					 BTRFS_INODE_##flag)
+#define btrfs_test_flag(inode, flag)	(BTRFS_I(inode)->flags & \
+					 BTRFS_INODE_##flag)
+/* some macros to generate set/get funcs for the struct fields.  This
+ * assumes there is a lefoo_to_cpu for every type, so lets make a simple
+ * one for u8:
+ */
+#define le8_to_cpu(v) (v)
+#define cpu_to_le8(v) (v)
+#define __le8 u8
+
+#define read_eb_member(eb, ptr, type, member, result) (			\
+	read_extent_buffer(eb, (char *)(result),			\
+			   ((unsigned long)(ptr)) +			\
+			    offsetof(type, member),			\
+			   sizeof(((type *)0)->member)))
+
+#define write_eb_member(eb, ptr, type, member, result) (		\
+	write_extent_buffer(eb, (char *)(result),			\
+			   ((unsigned long)(ptr)) +			\
+			    offsetof(type, member),			\
+			   sizeof(((type *)0)->member)))
+
+#ifndef BTRFS_SETGET_FUNCS
+#define BTRFS_SETGET_FUNCS(name, type, member, bits)			\
+u##bits btrfs_##name(struct extent_buffer *eb, type *s);		\
+void btrfs_set_##name(struct extent_buffer *eb, type *s, u##bits val);
+#endif
+
+#define BTRFS_SETGET_HEADER_FUNCS(name, type, member, bits)		\
+static inline u##bits btrfs_##name(struct extent_buffer *eb)		\
+{									\
+	type *p = kmap_atomic(eb->first_page, KM_USER0);		\
+	u##bits res = le##bits##_to_cpu(p->member);			\
+	kunmap_atomic(p, KM_USER0);					\
+	return res;							\
+}									\
+static inline void btrfs_set_##name(struct extent_buffer *eb,		\
+				    u##bits val)			\
+{									\
+	type *p = kmap_atomic(eb->first_page, KM_USER0);		\
+	p->member = cpu_to_le##bits(val);				\
+	kunmap_atomic(p, KM_USER0);					\
+}
+
+#define BTRFS_SETGET_STACK_FUNCS(name, type, member, bits)		\
+static inline u##bits btrfs_##name(type *s)				\
+{									\
+	return le##bits##_to_cpu(s->member);				\
+}									\
+static inline void btrfs_set_##name(type *s, u##bits val)		\
+{									\
+	s->member = cpu_to_le##bits(val);				\
+}
+
+BTRFS_SETGET_FUNCS(device_type, struct btrfs_dev_item, type, 64);
+BTRFS_SETGET_FUNCS(device_total_bytes, struct btrfs_dev_item, total_bytes, 64);
+BTRFS_SETGET_FUNCS(device_bytes_used, struct btrfs_dev_item, bytes_used, 64);
+BTRFS_SETGET_FUNCS(device_io_align, struct btrfs_dev_item, io_align, 32);
+BTRFS_SETGET_FUNCS(device_io_width, struct btrfs_dev_item, io_width, 32);
+BTRFS_SETGET_FUNCS(device_start_offset, struct btrfs_dev_item,
+		   start_offset, 64);
+BTRFS_SETGET_FUNCS(device_sector_size, struct btrfs_dev_item, sector_size, 32);
+BTRFS_SETGET_FUNCS(device_id, struct btrfs_dev_item, devid, 64);
+BTRFS_SETGET_FUNCS(device_group, struct btrfs_dev_item, dev_group, 32);
+BTRFS_SETGET_FUNCS(device_seek_speed, struct btrfs_dev_item, seek_speed, 8);
+BTRFS_SETGET_FUNCS(device_bandwidth, struct btrfs_dev_item, bandwidth, 8);
+BTRFS_SETGET_FUNCS(device_generation, struct btrfs_dev_item, generation, 64);
+
+BTRFS_SETGET_STACK_FUNCS(stack_device_type, struct btrfs_dev_item, type, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_total_bytes, struct btrfs_dev_item,
+			 total_bytes, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_bytes_used, struct btrfs_dev_item,
+			 bytes_used, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_io_align, struct btrfs_dev_item,
+			 io_align, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_device_io_width, struct btrfs_dev_item,
+			 io_width, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_device_sector_size, struct btrfs_dev_item,
+			 sector_size, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_device_id, struct btrfs_dev_item, devid, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_group, struct btrfs_dev_item,
+			 dev_group, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_device_seek_speed, struct btrfs_dev_item,
+			 seek_speed, 8);
+BTRFS_SETGET_STACK_FUNCS(stack_device_bandwidth, struct btrfs_dev_item,
+			 bandwidth, 8);
+BTRFS_SETGET_STACK_FUNCS(stack_device_generation, struct btrfs_dev_item,
+			 generation, 64);
+
+static inline char *btrfs_device_uuid(struct btrfs_dev_item *d)
+{
+	return (char *)d + offsetof(struct btrfs_dev_item, uuid);
+}
+
+static inline char *btrfs_device_fsid(struct btrfs_dev_item *d)
+{
+	return (char *)d + offsetof(struct btrfs_dev_item, fsid);
+}
+
+BTRFS_SETGET_FUNCS(chunk_length, struct btrfs_chunk, length, 64);
+BTRFS_SETGET_FUNCS(chunk_owner, struct btrfs_chunk, owner, 64);
+BTRFS_SETGET_FUNCS(chunk_stripe_len, struct btrfs_chunk, stripe_len, 64);
+BTRFS_SETGET_FUNCS(chunk_io_align, struct btrfs_chunk, io_align, 32);
+BTRFS_SETGET_FUNCS(chunk_io_width, struct btrfs_chunk, io_width, 32);
+BTRFS_SETGET_FUNCS(chunk_sector_size, struct btrfs_chunk, sector_size, 32);
+BTRFS_SETGET_FUNCS(chunk_type, struct btrfs_chunk, type, 64);
+BTRFS_SETGET_FUNCS(chunk_num_stripes, struct btrfs_chunk, num_stripes, 16);
+BTRFS_SETGET_FUNCS(chunk_sub_stripes, struct btrfs_chunk, sub_stripes, 16);
+BTRFS_SETGET_FUNCS(stripe_devid, struct btrfs_stripe, devid, 64);
+BTRFS_SETGET_FUNCS(stripe_offset, struct btrfs_stripe, offset, 64);
+
+static inline char *btrfs_stripe_dev_uuid(struct btrfs_stripe *s)
+{
+	return (char *)s + offsetof(struct btrfs_stripe, dev_uuid);
+}
+
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_length, struct btrfs_chunk, length, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_owner, struct btrfs_chunk, owner, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_stripe_len, struct btrfs_chunk,
+			 stripe_len, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_io_align, struct btrfs_chunk,
+			 io_align, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_io_width, struct btrfs_chunk,
+			 io_width, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_sector_size, struct btrfs_chunk,
+			 sector_size, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_type, struct btrfs_chunk, type, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_num_stripes, struct btrfs_chunk,
+			 num_stripes, 16);
+BTRFS_SETGET_STACK_FUNCS(stack_chunk_sub_stripes, struct btrfs_chunk,
+			 sub_stripes, 16);
+BTRFS_SETGET_STACK_FUNCS(stack_stripe_devid, struct btrfs_stripe, devid, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_stripe_offset, struct btrfs_stripe, offset, 64);
+
+static inline struct btrfs_stripe *btrfs_stripe_nr(struct btrfs_chunk *c,
+						   int nr)
+{
+	unsigned long offset = (unsigned long)c;
+	offset += offsetof(struct btrfs_chunk, stripe);
+	offset += nr * sizeof(struct btrfs_stripe);
+	return (struct btrfs_stripe *)offset;
+}
+
+static inline char *btrfs_stripe_dev_uuid_nr(struct btrfs_chunk *c, int nr)
+{
+	return btrfs_stripe_dev_uuid(btrfs_stripe_nr(c, nr));
+}
+
+static inline u64 btrfs_stripe_offset_nr(struct extent_buffer *eb,
+					 struct btrfs_chunk *c, int nr)
+{
+	return btrfs_stripe_offset(eb, btrfs_stripe_nr(c, nr));
+}
+
+static inline void btrfs_set_stripe_offset_nr(struct extent_buffer *eb,
+					     struct btrfs_chunk *c, int nr,
+					     u64 val)
+{
+	btrfs_set_stripe_offset(eb, btrfs_stripe_nr(c, nr), val);
+}
+
+static inline u64 btrfs_stripe_devid_nr(struct extent_buffer *eb,
+					 struct btrfs_chunk *c, int nr)
+{
+	return btrfs_stripe_devid(eb, btrfs_stripe_nr(c, nr));
+}
+
+static inline void btrfs_set_stripe_devid_nr(struct extent_buffer *eb,
+					     struct btrfs_chunk *c, int nr,
+					     u64 val)
+{
+	btrfs_set_stripe_devid(eb, btrfs_stripe_nr(c, nr), val);
+}
+
+/* struct btrfs_block_group_item */
+BTRFS_SETGET_STACK_FUNCS(block_group_used, struct btrfs_block_group_item,
+			 used, 64);
+BTRFS_SETGET_FUNCS(disk_block_group_used, struct btrfs_block_group_item,
+			 used, 64);
+BTRFS_SETGET_STACK_FUNCS(block_group_chunk_objectid,
+			struct btrfs_block_group_item, chunk_objectid, 64);
+
+BTRFS_SETGET_FUNCS(disk_block_group_chunk_objectid,
+		   struct btrfs_block_group_item, chunk_objectid, 64);
+BTRFS_SETGET_FUNCS(disk_block_group_flags,
+		   struct btrfs_block_group_item, flags, 64);
+BTRFS_SETGET_STACK_FUNCS(block_group_flags,
+			struct btrfs_block_group_item, flags, 64);
+
+/* struct btrfs_inode_ref */
+BTRFS_SETGET_FUNCS(inode_ref_name_len, struct btrfs_inode_ref, name_len, 16);
+BTRFS_SETGET_FUNCS(inode_ref_index, struct btrfs_inode_ref, index, 64);
+
+/* struct btrfs_inode_item */
+BTRFS_SETGET_FUNCS(inode_generation, struct btrfs_inode_item, generation, 64);
+BTRFS_SETGET_FUNCS(inode_sequence, struct btrfs_inode_item, sequence, 64);
+BTRFS_SETGET_FUNCS(inode_transid, struct btrfs_inode_item, transid, 64);
+BTRFS_SETGET_FUNCS(inode_size, struct btrfs_inode_item, size, 64);
+BTRFS_SETGET_FUNCS(inode_nbytes, struct btrfs_inode_item, nbytes, 64);
+BTRFS_SETGET_FUNCS(inode_block_group, struct btrfs_inode_item, block_group, 64);
+BTRFS_SETGET_FUNCS(inode_nlink, struct btrfs_inode_item, nlink, 32);
+BTRFS_SETGET_FUNCS(inode_uid, struct btrfs_inode_item, uid, 32);
+BTRFS_SETGET_FUNCS(inode_gid, struct btrfs_inode_item, gid, 32);
+BTRFS_SETGET_FUNCS(inode_mode, struct btrfs_inode_item, mode, 32);
+BTRFS_SETGET_FUNCS(inode_rdev, struct btrfs_inode_item, rdev, 64);
+BTRFS_SETGET_FUNCS(inode_flags, struct btrfs_inode_item, flags, 64);
+
+static inline struct btrfs_timespec *
+btrfs_inode_atime(struct btrfs_inode_item *inode_item)
+{
+	unsigned long ptr = (unsigned long)inode_item;
+	ptr += offsetof(struct btrfs_inode_item, atime);
+	return (struct btrfs_timespec *)ptr;
+}
+
+static inline struct btrfs_timespec *
+btrfs_inode_mtime(struct btrfs_inode_item *inode_item)
+{
+	unsigned long ptr = (unsigned long)inode_item;
+	ptr += offsetof(struct btrfs_inode_item, mtime);
+	return (struct btrfs_timespec *)ptr;
+}
+
+static inline struct btrfs_timespec *
+btrfs_inode_ctime(struct btrfs_inode_item *inode_item)
+{
+	unsigned long ptr = (unsigned long)inode_item;
+	ptr += offsetof(struct btrfs_inode_item, ctime);
+	return (struct btrfs_timespec *)ptr;
+}
+
+static inline struct btrfs_timespec *
+btrfs_inode_otime(struct btrfs_inode_item *inode_item)
+{
+	unsigned long ptr = (unsigned long)inode_item;
+	ptr += offsetof(struct btrfs_inode_item, otime);
+	return (struct btrfs_timespec *)ptr;
+}
+
+BTRFS_SETGET_FUNCS(timespec_sec, struct btrfs_timespec, sec, 64);
+BTRFS_SETGET_FUNCS(timespec_nsec, struct btrfs_timespec, nsec, 32);
+
+/* struct btrfs_dev_extent */
+BTRFS_SETGET_FUNCS(dev_extent_chunk_tree, struct btrfs_dev_extent,
+		   chunk_tree, 64);
+BTRFS_SETGET_FUNCS(dev_extent_chunk_objectid, struct btrfs_dev_extent,
+		   chunk_objectid, 64);
+BTRFS_SETGET_FUNCS(dev_extent_chunk_offset, struct btrfs_dev_extent,
+		   chunk_offset, 64);
+BTRFS_SETGET_FUNCS(dev_extent_length, struct btrfs_dev_extent, length, 64);
+
+static inline u8 *btrfs_dev_extent_chunk_tree_uuid(struct btrfs_dev_extent *dev)
+{
+	unsigned long ptr = offsetof(struct btrfs_dev_extent, chunk_tree_uuid);
+	return (u8 *)((unsigned long)dev + ptr);
+}
+
+/* struct btrfs_extent_ref */
+BTRFS_SETGET_FUNCS(ref_root, struct btrfs_extent_ref, root, 64);
+BTRFS_SETGET_FUNCS(ref_generation, struct btrfs_extent_ref, generation, 64);
+BTRFS_SETGET_FUNCS(ref_objectid, struct btrfs_extent_ref, objectid, 64);
+BTRFS_SETGET_FUNCS(ref_num_refs, struct btrfs_extent_ref, num_refs, 32);
+
+BTRFS_SETGET_STACK_FUNCS(stack_ref_root, struct btrfs_extent_ref, root, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_ref_generation, struct btrfs_extent_ref,
+			 generation, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_ref_objectid, struct btrfs_extent_ref,
+			 objectid, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_ref_num_refs, struct btrfs_extent_ref,
+			 num_refs, 32);
+
+/* struct btrfs_extent_item */
+BTRFS_SETGET_FUNCS(extent_refs, struct btrfs_extent_item, refs, 32);
+BTRFS_SETGET_STACK_FUNCS(stack_extent_refs, struct btrfs_extent_item,
+			 refs, 32);
+
+/* struct btrfs_node */
+BTRFS_SETGET_FUNCS(key_blockptr, struct btrfs_key_ptr, blockptr, 64);
+BTRFS_SETGET_FUNCS(key_generation, struct btrfs_key_ptr, generation, 64);
+
+static inline u64 btrfs_node_blockptr(struct extent_buffer *eb, int nr)
+{
+	unsigned long ptr;
+	ptr = offsetof(struct btrfs_node, ptrs) +
+		sizeof(struct btrfs_key_ptr) * nr;
+	return btrfs_key_blockptr(eb, (struct btrfs_key_ptr *)ptr);
+}
+
+static inline void btrfs_set_node_blockptr(struct extent_buffer *eb,
+					   int nr, u64 val)
+{
+	unsigned long ptr;
+	ptr = offsetof(struct btrfs_node, ptrs) +
+		sizeof(struct btrfs_key_ptr) * nr;
+	btrfs_set_key_blockptr(eb, (struct btrfs_key_ptr *)ptr, val);
+}
+
+static inline u64 btrfs_node_ptr_generation(struct extent_buffer *eb, int nr)
+{
+	unsigned long ptr;
+	ptr = offsetof(struct btrfs_node, ptrs) +
+		sizeof(struct btrfs_key_ptr) * nr;
+	return btrfs_key_generation(eb, (struct btrfs_key_ptr *)ptr);
+}
+
+static inline void btrfs_set_node_ptr_generation(struct extent_buffer *eb,
+						 int nr, u64 val)
+{
+	unsigned long ptr;
+	ptr = offsetof(struct btrfs_node, ptrs) +
+		sizeof(struct btrfs_key_ptr) * nr;
+	btrfs_set_key_generation(eb, (struct btrfs_key_ptr *)ptr, val);
+}
+
+static inline unsigned long btrfs_node_key_ptr_offset(int nr)
+{
+	return offsetof(struct btrfs_node, ptrs) +
+		sizeof(struct btrfs_key_ptr) * nr;
+}
+
+void btrfs_node_key(struct extent_buffer *eb,
+		    struct btrfs_disk_key *disk_key, int nr);
+
+static inline void btrfs_set_node_key(struct extent_buffer *eb,
+				      struct btrfs_disk_key *disk_key, int nr)
+{
+	unsigned long ptr;
+	ptr = btrfs_node_key_ptr_offset(nr);
+	write_eb_member(eb, (struct btrfs_key_ptr *)ptr,
+		       struct btrfs_key_ptr, key, disk_key);
+}
+
+/* struct btrfs_item */
+BTRFS_SETGET_FUNCS(item_offset, struct btrfs_item, offset, 32);
+BTRFS_SETGET_FUNCS(item_size, struct btrfs_item, size, 32);
+
+static inline unsigned long btrfs_item_nr_offset(int nr)
+{
+	return offsetof(struct btrfs_leaf, items) +
+		sizeof(struct btrfs_item) * nr;
+}
+
+static inline struct btrfs_item *btrfs_item_nr(struct extent_buffer *eb,
+					       int nr)
+{
+	return (struct btrfs_item *)btrfs_item_nr_offset(nr);
+}
+
+static inline u32 btrfs_item_end(struct extent_buffer *eb,
+				 struct btrfs_item *item)
+{
+	return btrfs_item_offset(eb, item) + btrfs_item_size(eb, item);
+}
+
+static inline u32 btrfs_item_end_nr(struct extent_buffer *eb, int nr)
+{
+	return btrfs_item_end(eb, btrfs_item_nr(eb, nr));
+}
+
+static inline u32 btrfs_item_offset_nr(struct extent_buffer *eb, int nr)
+{
+	return btrfs_item_offset(eb, btrfs_item_nr(eb, nr));
+}
+
+static inline u32 btrfs_item_size_nr(struct extent_buffer *eb, int nr)
+{
+	return btrfs_item_size(eb, btrfs_item_nr(eb, nr));
+}
+
+static inline void btrfs_item_key(struct extent_buffer *eb,
+			   struct btrfs_disk_key *disk_key, int nr)
+{
+	struct btrfs_item *item = btrfs_item_nr(eb, nr);
+	read_eb_member(eb, item, struct btrfs_item, key, disk_key);
+}
+
+static inline void btrfs_set_item_key(struct extent_buffer *eb,
+			       struct btrfs_disk_key *disk_key, int nr)
+{
+	struct btrfs_item *item = btrfs_item_nr(eb, nr);
+	write_eb_member(eb, item, struct btrfs_item, key, disk_key);
+}
+
+BTRFS_SETGET_FUNCS(dir_log_end, struct btrfs_dir_log_item, end, 64);
+
+/*
+ * struct btrfs_root_ref
+ */
+BTRFS_SETGET_FUNCS(root_ref_dirid, struct btrfs_root_ref, dirid, 64);
+BTRFS_SETGET_FUNCS(root_ref_sequence, struct btrfs_root_ref, sequence, 64);
+BTRFS_SETGET_FUNCS(root_ref_name_len, struct btrfs_root_ref, name_len, 16);
+
+/* struct btrfs_dir_item */
+BTRFS_SETGET_FUNCS(dir_data_len, struct btrfs_dir_item, data_len, 16);
+BTRFS_SETGET_FUNCS(dir_type, struct btrfs_dir_item, type, 8);
+BTRFS_SETGET_FUNCS(dir_name_len, struct btrfs_dir_item, name_len, 16);
+BTRFS_SETGET_FUNCS(dir_transid, struct btrfs_dir_item, transid, 64);
+
+static inline void btrfs_dir_item_key(struct extent_buffer *eb,
+				      struct btrfs_dir_item *item,
+				      struct btrfs_disk_key *key)
+{
+	read_eb_member(eb, item, struct btrfs_dir_item, location, key);
+}
+
+static inline void btrfs_set_dir_item_key(struct extent_buffer *eb,
+					  struct btrfs_dir_item *item,
+					  struct btrfs_disk_key *key)
+{
+	write_eb_member(eb, item, struct btrfs_dir_item, location, key);
+}
+
+/* struct btrfs_disk_key */
+BTRFS_SETGET_STACK_FUNCS(disk_key_objectid, struct btrfs_disk_key,
+			 objectid, 64);
+BTRFS_SETGET_STACK_FUNCS(disk_key_offset, struct btrfs_disk_key, offset, 64);
+BTRFS_SETGET_STACK_FUNCS(disk_key_type, struct btrfs_disk_key, type, 8);
+
+static inline void btrfs_disk_key_to_cpu(struct btrfs_key *cpu,
+					 struct btrfs_disk_key *disk)
+{
+	cpu->offset = le64_to_cpu(disk->offset);
+	cpu->type = disk->type;
+	cpu->objectid = le64_to_cpu(disk->objectid);
+}
+
+static inline void btrfs_cpu_key_to_disk(struct btrfs_disk_key *disk,
+					 struct btrfs_key *cpu)
+{
+	disk->offset = cpu_to_le64(cpu->offset);
+	disk->type = cpu->type;
+	disk->objectid = cpu_to_le64(cpu->objectid);
+}
+
+static inline void btrfs_node_key_to_cpu(struct extent_buffer *eb,
+				  struct btrfs_key *key, int nr)
+{
+	struct btrfs_disk_key disk_key;
+	btrfs_node_key(eb, &disk_key, nr);
+	btrfs_disk_key_to_cpu(key, &disk_key);
+}
+
+static inline void btrfs_item_key_to_cpu(struct extent_buffer *eb,
+				  struct btrfs_key *key, int nr)
+{
+	struct btrfs_disk_key disk_key;
+	btrfs_item_key(eb, &disk_key, nr);
+	btrfs_disk_key_to_cpu(key, &disk_key);
+}
+
+static inline void btrfs_dir_item_key_to_cpu(struct extent_buffer *eb,
+				      struct btrfs_dir_item *item,
+				      struct btrfs_key *key)
+{
+	struct btrfs_disk_key disk_key;
+	btrfs_dir_item_key(eb, item, &disk_key);
+	btrfs_disk_key_to_cpu(key, &disk_key);
+}
+
+
+static inline u8 btrfs_key_type(struct btrfs_key *key)
+{
+	return key->type;
+}
+
+static inline void btrfs_set_key_type(struct btrfs_key *key, u8 val)
+{
+	key->type = val;
+}
+
+/* struct btrfs_header */
+BTRFS_SETGET_HEADER_FUNCS(header_bytenr, struct btrfs_header, bytenr, 64);
+BTRFS_SETGET_HEADER_FUNCS(header_generation, struct btrfs_header,
+			  generation, 64);
+BTRFS_SETGET_HEADER_FUNCS(header_owner, struct btrfs_header, owner, 64);
+BTRFS_SETGET_HEADER_FUNCS(header_nritems, struct btrfs_header, nritems, 32);
+BTRFS_SETGET_HEADER_FUNCS(header_flags, struct btrfs_header, flags, 64);
+BTRFS_SETGET_HEADER_FUNCS(header_level, struct btrfs_header, level, 8);
+
+static inline int btrfs_header_flag(struct extent_buffer *eb, u64 flag)
+{
+	return (btrfs_header_flags(eb) & flag) == flag;
+}
+
+static inline int btrfs_set_header_flag(struct extent_buffer *eb, u64 flag)
+{
+	u64 flags = btrfs_header_flags(eb);
+	btrfs_set_header_flags(eb, flags | flag);
+	return (flags & flag) == flag;
+}
+
+static inline int btrfs_clear_header_flag(struct extent_buffer *eb, u64 flag)
+{
+	u64 flags = btrfs_header_flags(eb);
+	btrfs_set_header_flags(eb, flags & ~flag);
+	return (flags & flag) == flag;
+}
+
+static inline u8 *btrfs_header_fsid(struct extent_buffer *eb)
+{
+	unsigned long ptr = offsetof(struct btrfs_header, fsid);
+	return (u8 *)ptr;
+}
+
+static inline u8 *btrfs_header_chunk_tree_uuid(struct extent_buffer *eb)
+{
+	unsigned long ptr = offsetof(struct btrfs_header, chunk_tree_uuid);
+	return (u8 *)ptr;
+}
+
+static inline u8 *btrfs_super_fsid(struct extent_buffer *eb)
+{
+	unsigned long ptr = offsetof(struct btrfs_super_block, fsid);
+	return (u8 *)ptr;
+}
+
+static inline u8 *btrfs_header_csum(struct extent_buffer *eb)
+{
+	unsigned long ptr = offsetof(struct btrfs_header, csum);
+	return (u8 *)ptr;
+}
+
+static inline struct btrfs_node *btrfs_buffer_node(struct extent_buffer *eb)
+{
+	return NULL;
+}
+
+static inline struct btrfs_leaf *btrfs_buffer_leaf(struct extent_buffer *eb)
+{
+	return NULL;
+}
+
+static inline struct btrfs_header *btrfs_buffer_header(struct extent_buffer *eb)
+{
+	return NULL;
+}
+
+static inline int btrfs_is_leaf(struct extent_buffer *eb)
+{
+	return btrfs_header_level(eb) == 0;
+}
+
+/* struct btrfs_root_item */
+BTRFS_SETGET_FUNCS(disk_root_generation, struct btrfs_root_item,
+		   generation, 64);
+BTRFS_SETGET_FUNCS(disk_root_refs, struct btrfs_root_item, refs, 32);
+BTRFS_SETGET_FUNCS(disk_root_bytenr, struct btrfs_root_item, bytenr, 64);
+BTRFS_SETGET_FUNCS(disk_root_level, struct btrfs_root_item, level, 8);
+
+BTRFS_SETGET_STACK_FUNCS(root_generation, struct btrfs_root_item,
+			 generation, 64);
+BTRFS_SETGET_STACK_FUNCS(root_bytenr, struct btrfs_root_item, bytenr, 64);
+BTRFS_SETGET_STACK_FUNCS(root_level, struct btrfs_root_item, level, 8);
+BTRFS_SETGET_STACK_FUNCS(root_dirid, struct btrfs_root_item, root_dirid, 64);
+BTRFS_SETGET_STACK_FUNCS(root_refs, struct btrfs_root_item, refs, 32);
+BTRFS_SETGET_STACK_FUNCS(root_flags, struct btrfs_root_item, flags, 64);
+BTRFS_SETGET_STACK_FUNCS(root_used, struct btrfs_root_item, bytes_used, 64);
+BTRFS_SETGET_STACK_FUNCS(root_limit, struct btrfs_root_item, byte_limit, 64);
+BTRFS_SETGET_STACK_FUNCS(root_last_snapshot, struct btrfs_root_item,
+			 last_snapshot, 64);
+
+/* struct btrfs_super_block */
+
+BTRFS_SETGET_STACK_FUNCS(super_bytenr, struct btrfs_super_block, bytenr, 64);
+BTRFS_SETGET_STACK_FUNCS(super_flags, struct btrfs_super_block, flags, 64);
+BTRFS_SETGET_STACK_FUNCS(super_generation, struct btrfs_super_block,
+			 generation, 64);
+BTRFS_SETGET_STACK_FUNCS(super_root, struct btrfs_super_block, root, 64);
+BTRFS_SETGET_STACK_FUNCS(super_sys_array_size,
+			 struct btrfs_super_block, sys_chunk_array_size, 32);
+BTRFS_SETGET_STACK_FUNCS(super_chunk_root_generation,
+			 struct btrfs_super_block, chunk_root_generation, 64);
+BTRFS_SETGET_STACK_FUNCS(super_root_level, struct btrfs_super_block,
+			 root_level, 8);
+BTRFS_SETGET_STACK_FUNCS(super_chunk_root, struct btrfs_super_block,
+			 chunk_root, 64);
+BTRFS_SETGET_STACK_FUNCS(super_chunk_root_level, struct btrfs_super_block,
+			 chunk_root_level, 8);
+BTRFS_SETGET_STACK_FUNCS(super_log_root, struct btrfs_super_block,
+			 log_root, 64);
+BTRFS_SETGET_STACK_FUNCS(super_log_root_transid, struct btrfs_super_block,
+			 log_root_transid, 64);
+BTRFS_SETGET_STACK_FUNCS(super_log_root_level, struct btrfs_super_block,
+			 log_root_level, 8);
+BTRFS_SETGET_STACK_FUNCS(super_total_bytes, struct btrfs_super_block,
+			 total_bytes, 64);
+BTRFS_SETGET_STACK_FUNCS(super_bytes_used, struct btrfs_super_block,
+			 bytes_used, 64);
+BTRFS_SETGET_STACK_FUNCS(super_sectorsize, struct btrfs_super_block,
+			 sectorsize, 32);
+BTRFS_SETGET_STACK_FUNCS(super_nodesize, struct btrfs_super_block,
+			 nodesize, 32);
+BTRFS_SETGET_STACK_FUNCS(super_leafsize, struct btrfs_super_block,
+			 leafsize, 32);
+BTRFS_SETGET_STACK_FUNCS(super_stripesize, struct btrfs_super_block,
+			 stripesize, 32);
+BTRFS_SETGET_STACK_FUNCS(super_root_dir, struct btrfs_super_block,
+			 root_dir_objectid, 64);
+BTRFS_SETGET_STACK_FUNCS(super_num_devices, struct btrfs_super_block,
+			 num_devices, 64);
+BTRFS_SETGET_STACK_FUNCS(super_compat_flags, struct btrfs_super_block,
+			 compat_flags, 64);
+BTRFS_SETGET_STACK_FUNCS(super_compat_ro_flags, struct btrfs_super_block,
+			 compat_flags, 64);
+BTRFS_SETGET_STACK_FUNCS(super_incompat_flags, struct btrfs_super_block,
+			 incompat_flags, 64);
+BTRFS_SETGET_STACK_FUNCS(super_csum_type, struct btrfs_super_block,
+			 csum_type, 16);
+
+static inline int btrfs_super_csum_size(struct btrfs_super_block *s)
+{
+	int t = btrfs_super_csum_type(s);
+	BUG_ON(t >= ARRAY_SIZE(btrfs_csum_sizes));
+	return btrfs_csum_sizes[t];
+}
+
+static inline unsigned long btrfs_leaf_data(struct extent_buffer *l)
+{
+	return offsetof(struct btrfs_leaf, items);
+}
+
+/* struct btrfs_file_extent_item */
+BTRFS_SETGET_FUNCS(file_extent_type, struct btrfs_file_extent_item, type, 8);
+
+static inline unsigned long
+btrfs_file_extent_inline_start(struct btrfs_file_extent_item *e)
+{
+	unsigned long offset = (unsigned long)e;
+	offset += offsetof(struct btrfs_file_extent_item, disk_bytenr);
+	return offset;
+}
+
+static inline u32 btrfs_file_extent_calc_inline_size(u32 datasize)
+{
+	return offsetof(struct btrfs_file_extent_item, disk_bytenr) + datasize;
+}
+
+BTRFS_SETGET_FUNCS(file_extent_disk_bytenr, struct btrfs_file_extent_item,
+		   disk_bytenr, 64);
+BTRFS_SETGET_FUNCS(file_extent_generation, struct btrfs_file_extent_item,
+		   generation, 64);
+BTRFS_SETGET_FUNCS(file_extent_disk_num_bytes, struct btrfs_file_extent_item,
+		   disk_num_bytes, 64);
+BTRFS_SETGET_FUNCS(file_extent_offset, struct btrfs_file_extent_item,
+		  offset, 64);
+BTRFS_SETGET_FUNCS(file_extent_num_bytes, struct btrfs_file_extent_item,
+		   num_bytes, 64);
+BTRFS_SETGET_FUNCS(file_extent_ram_bytes, struct btrfs_file_extent_item,
+		   ram_bytes, 64);
+BTRFS_SETGET_FUNCS(file_extent_compression, struct btrfs_file_extent_item,
+		   compression, 8);
+BTRFS_SETGET_FUNCS(file_extent_encryption, struct btrfs_file_extent_item,
+		   encryption, 8);
+BTRFS_SETGET_FUNCS(file_extent_other_encoding, struct btrfs_file_extent_item,
+		   other_encoding, 16);
+
+/* this returns the number of file bytes represented by the inline item.
+ * If an item is compressed, this is the uncompressed size
+ */
+static inline u32 btrfs_file_extent_inline_len(struct extent_buffer *eb,
+					       struct btrfs_file_extent_item *e)
+{
+	return btrfs_file_extent_ram_bytes(eb, e);
+}
+
+/*
+ * this returns the number of bytes used by the item on disk, minus the
+ * size of any extent headers.  If a file is compressed on disk, this is
+ * the compressed size
+ */
+static inline u32 btrfs_file_extent_inline_item_len(struct extent_buffer *eb,
+						    struct btrfs_item *e)
+{
+	unsigned long offset;
+	offset = offsetof(struct btrfs_file_extent_item, disk_bytenr);
+	return btrfs_item_size(eb, e) - offset;
+}
+
+static inline struct btrfs_root *btrfs_sb(struct super_block *sb)
+{
+	return sb->s_fs_info;
+}
+
+static inline int btrfs_set_root_name(struct btrfs_root *root,
+				      const char *name, int len)
+{
+	/* if we already have a name just free it */
+	kfree(root->name);
+
+	root->name = kmalloc(len+1, GFP_KERNEL);
+	if (!root->name)
+		return -ENOMEM;
+
+	memcpy(root->name, name, len);
+	root->name[len] = '\0';
+
+	return 0;
+}
+
+static inline u32 btrfs_level_size(struct btrfs_root *root, int level)
+{
+	if (level == 0)
+		return root->leafsize;
+	return root->nodesize;
+}
+
+/* helper function to cast into the data area of the leaf. */
+#define btrfs_item_ptr(leaf, slot, type) \
+	((type *)(btrfs_leaf_data(leaf) + \
+	btrfs_item_offset_nr(leaf, slot)))
+
+#define btrfs_item_ptr_offset(leaf, slot) \
+	((unsigned long)(btrfs_leaf_data(leaf) + \
+	btrfs_item_offset_nr(leaf, slot)))
+
+static inline struct dentry *fdentry(struct file *file)
+{
+	return file->f_path.dentry;
+}
+
+/* extent-tree.c */
+int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len);
+int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, u64 bytenr,
+			    u64 num_bytes, u32 *refs);
+int btrfs_update_pinned_extents(struct btrfs_root *root,
+				u64 bytenr, u64 num, int pin);
+int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root, struct extent_buffer *leaf);
+int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, u64 objectid, u64 bytenr);
+int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root);
+int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy);
+struct btrfs_block_group_cache *btrfs_lookup_block_group(
+						 struct btrfs_fs_info *info,
+						 u64 bytenr);
+u64 btrfs_find_block_group(struct btrfs_root *root,
+			   u64 search_start, u64 search_hint, int owner);
+struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
+					     struct btrfs_root *root,
+					     u32 blocksize, u64 parent,
+					     u64 root_objectid,
+					     u64 ref_generation,
+					     int level,
+					     u64 hint,
+					     u64 empty_size);
+struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
+					    struct btrfs_root *root,
+					    u64 bytenr, u32 blocksize);
+int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root,
+		       u64 num_bytes, u64 parent, u64 min_bytes,
+		       u64 root_objectid, u64 ref_generation,
+		       u64 owner, u64 empty_size, u64 hint_byte,
+		       u64 search_end, struct btrfs_key *ins, u64 data);
+int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root, u64 parent,
+				u64 root_objectid, u64 ref_generation,
+				u64 owner, struct btrfs_key *ins);
+int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root, u64 parent,
+				u64 root_objectid, u64 ref_generation,
+				u64 owner, struct btrfs_key *ins);
+int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  u64 num_bytes, u64 min_alloc_size,
+				  u64 empty_size, u64 hint_byte,
+				  u64 search_end, struct btrfs_key *ins,
+				  u64 data);
+int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		  struct extent_buffer *orig_buf, struct extent_buffer *buf,
+		  u32 *nr_extents);
+int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		    struct extent_buffer *buf, u32 nr_extents);
+int btrfs_update_ref(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root, struct extent_buffer *orig_buf,
+		     struct extent_buffer *buf, int start_slot, int nr);
+int btrfs_free_extent(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root,
+		      u64 bytenr, u64 num_bytes, u64 parent,
+		      u64 root_objectid, u64 ref_generation,
+		      u64 owner_objectid, int pin);
+int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len);
+int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct extent_io_tree *unpin);
+int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 u64 bytenr, u64 num_bytes, u64 parent,
+			 u64 root_objectid, u64 ref_generation,
+			 u64 owner_objectid);
+int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, u64 bytenr,
+			    u64 orig_parent, u64 parent,
+			    u64 root_objectid, u64 ref_generation,
+			    u64 owner_objectid);
+int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root);
+int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr);
+int btrfs_free_block_groups(struct btrfs_fs_info *info);
+int btrfs_read_block_groups(struct btrfs_root *root);
+int btrfs_make_block_group(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, u64 bytes_used,
+			   u64 type, u64 chunk_objectid, u64 chunk_offset,
+			   u64 size);
+int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 group_start);
+int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start);
+int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root);
+int btrfs_drop_dead_reloc_roots(struct btrfs_root *root);
+int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct extent_buffer *buf, u64 orig_start);
+int btrfs_add_dead_reloc_root(struct btrfs_root *root);
+int btrfs_cleanup_reloc_trees(struct btrfs_root *root);
+int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len);
+u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags);
+/* ctree.c */
+int btrfs_previous_item(struct btrfs_root *root,
+			struct btrfs_path *path, u64 min_objectid,
+			int type);
+int btrfs_merge_path(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_key *node_keys,
+		     u64 *nodes, int lowest_level);
+int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, struct btrfs_path *path,
+			    struct btrfs_key *new_key);
+struct extent_buffer *btrfs_root_node(struct btrfs_root *root);
+struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root);
+int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
+			struct btrfs_key *key, int lowest_level,
+			int cache_only, u64 min_trans);
+int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
+			 struct btrfs_key *max_key,
+			 struct btrfs_path *path, int cache_only,
+			 u64 min_trans);
+int btrfs_cow_block(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, struct extent_buffer *buf,
+		    struct extent_buffer *parent, int parent_slot,
+		    struct extent_buffer **cow_ret, u64 prealloc_dest);
+int btrfs_copy_root(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root,
+		      struct extent_buffer *buf,
+		      struct extent_buffer **cow_ret, u64 new_root_objectid);
+int btrfs_extend_item(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_path *path, u32 data_size);
+int btrfs_truncate_item(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_path *path,
+			u32 new_size, int from_end);
+int btrfs_split_item(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_path *path,
+		     struct btrfs_key *new_key,
+		     unsigned long split_offset);
+int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, struct btrfs_path *p, int
+		      ins_len, int cow);
+int btrfs_realloc_node(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root, struct extent_buffer *parent,
+		       int start_slot, int cache_only, u64 *last_ret,
+		       struct btrfs_key *progress);
+void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p);
+struct btrfs_path *btrfs_alloc_path(void);
+void btrfs_free_path(struct btrfs_path *p);
+void btrfs_init_path(struct btrfs_path *p);
+int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_path *path, int slot, int nr);
+int btrfs_del_leaf(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path, u64 bytenr);
+static inline int btrfs_del_item(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path)
+{
+	return btrfs_del_items(trans, root, path, path->slots[0], 1);
+}
+
+int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, void *data, u32 data_size);
+int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path,
+			    struct btrfs_key *cpu_key, u32 *data_size,
+			    int nr);
+int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     struct btrfs_path *path,
+			     struct btrfs_key *cpu_key, u32 *data_size, int nr);
+
+static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  struct btrfs_key *key,
+					  u32 data_size)
+{
+	return btrfs_insert_empty_items(trans, root, path, key, &data_size, 1);
+}
+
+int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path);
+int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
+int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf);
+int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
+			*root);
+int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct extent_buffer *node,
+			struct extent_buffer *parent);
+/* root-item.c */
+int btrfs_find_root_ref(struct btrfs_root *tree_root,
+		   struct btrfs_path *path,
+		   u64 root_id, u64 ref_id);
+int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *tree_root,
+		       u64 root_id, u8 type, u64 ref_id,
+		       u64 dirid, u64 sequence,
+		       const char *name, int name_len);
+int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_key *key);
+int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, struct btrfs_root_item
+		      *item);
+int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, struct btrfs_root_item
+		      *item);
+int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct
+			 btrfs_root_item *item, struct btrfs_key *key);
+int btrfs_search_root(struct btrfs_root *root, u64 search_start,
+		      u64 *found_objectid);
+int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid,
+			  struct btrfs_root *latest_root);
+/* dir-item.c */
+int btrfs_insert_dir_item(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, const char *name,
+			  int name_len, u64 dir,
+			  struct btrfs_key *location, u8 type, u64 index);
+struct btrfs_dir_item *btrfs_lookup_dir_item(struct btrfs_trans_handle *trans,
+					     struct btrfs_root *root,
+					     struct btrfs_path *path, u64 dir,
+					     const char *name, int name_len,
+					     int mod);
+struct btrfs_dir_item *
+btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path, u64 dir,
+			    u64 objectid, const char *name, int name_len,
+			    int mod);
+struct btrfs_dir_item *btrfs_match_dir_item_name(struct btrfs_root *root,
+			      struct btrfs_path *path,
+			      const char *name, int name_len);
+int btrfs_delete_one_dir_name(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct btrfs_path *path,
+			      struct btrfs_dir_item *di);
+int btrfs_insert_xattr_item(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, const char *name,
+			    u16 name_len, const void *data, u16 data_len,
+			    u64 dir);
+struct btrfs_dir_item *btrfs_lookup_xattr(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path, u64 dir,
+					  const char *name, u16 name_len,
+					  int mod);
+
+/* orphan.c */
+int btrfs_insert_orphan_item(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 offset);
+int btrfs_del_orphan_item(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, u64 offset);
+
+/* inode-map.c */
+int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *fs_root,
+			     u64 dirid, u64 *objectid);
+int btrfs_find_highest_inode(struct btrfs_root *fs_root, u64 *objectid);
+
+/* inode-item.c */
+int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   const char *name, int name_len,
+			   u64 inode_objectid, u64 ref_objectid, u64 index);
+int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   const char *name, int name_len,
+			   u64 inode_objectid, u64 ref_objectid, u64 *index);
+int btrfs_insert_empty_inode(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     struct btrfs_path *path, u64 objectid);
+int btrfs_lookup_inode(struct btrfs_trans_handle *trans, struct btrfs_root
+		       *root, struct btrfs_path *path,
+		       struct btrfs_key *location, int mod);
+
+/* file-item.c */
+int btrfs_del_csums(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, u64 bytenr, u64 len);
+int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
+			  struct bio *bio, u32 *dst);
+int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     u64 objectid, u64 pos,
+			     u64 disk_offset, u64 disk_num_bytes,
+			     u64 num_bytes, u64 offset, u64 ram_bytes,
+			     u8 compression, u8 encryption, u16 other_encoding);
+int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     struct btrfs_path *path, u64 objectid,
+			     u64 bytenr, int mod);
+int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   struct btrfs_ordered_sum *sums);
+int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
+		       struct bio *bio, u64 file_start, int contig);
+int btrfs_csum_file_bytes(struct btrfs_root *root, struct inode *inode,
+			  u64 start, unsigned long len);
+struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, int cow);
+int btrfs_csum_truncate(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root, struct btrfs_path *path,
+			u64 isize);
+int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start,
+			     u64 end, struct list_head *list);
+/* inode.c */
+
+/* RHEL and EL kernels have a patch that renames PG_checked to FsMisc */
+#if defined(ClearPageFsMisc) && !defined(ClearPageChecked)
+#define ClearPageChecked ClearPageFsMisc
+#define SetPageChecked SetPageFsMisc
+#define PageChecked PageFsMisc
+#endif
+
+struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry);
+int btrfs_set_inode_index(struct inode *dir, u64 *index);
+int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root,
+		       struct inode *dir, struct inode *inode,
+		       const char *name, int name_len);
+int btrfs_add_link(struct btrfs_trans_handle *trans,
+		   struct inode *parent_inode, struct inode *inode,
+		   const char *name, int name_len, int add_backref, u64 index);
+int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct inode *inode, u64 new_size,
+			       u32 min_type);
+
+int btrfs_start_delalloc_inodes(struct btrfs_root *root);
+int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end);
+int btrfs_writepages(struct address_space *mapping,
+		     struct writeback_control *wbc);
+int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *new_root, struct dentry *dentry,
+			     u64 new_dirid, u64 alloc_hint);
+int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
+			 size_t size, struct bio *bio, unsigned long bio_flags);
+
+unsigned long btrfs_force_ra(struct address_space *mapping,
+			      struct file_ra_state *ra, struct file *file,
+			      pgoff_t offset, pgoff_t last_index);
+int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
+			   int for_del);
+int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page);
+int btrfs_readpage(struct file *file, struct page *page);
+void btrfs_delete_inode(struct inode *inode);
+void btrfs_put_inode(struct inode *inode);
+void btrfs_read_locked_inode(struct inode *inode);
+int btrfs_write_inode(struct inode *inode, int wait);
+void btrfs_dirty_inode(struct inode *inode);
+struct inode *btrfs_alloc_inode(struct super_block *sb);
+void btrfs_destroy_inode(struct inode *inode);
+int btrfs_init_cachep(void);
+void btrfs_destroy_cachep(void);
+long btrfs_ioctl_trans_end(struct file *file);
+struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
+			    struct btrfs_root *root, int wait);
+struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
+				struct btrfs_root *root);
+struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
+			 struct btrfs_root *root, int *is_new);
+int btrfs_commit_write(struct file *file, struct page *page,
+		       unsigned from, unsigned to);
+struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
+				    size_t page_offset, u64 start, u64 end,
+				    int create);
+int btrfs_update_inode(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct inode *inode);
+int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode);
+int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode);
+void btrfs_orphan_cleanup(struct btrfs_root *root);
+int btrfs_cont_expand(struct inode *inode, loff_t size);
+
+/* ioctl.c */
+long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
+
+/* file.c */
+int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync);
+int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
+			    int skip_pinned);
+int btrfs_check_file(struct btrfs_root *root, struct inode *inode);
+extern struct file_operations btrfs_file_operations;
+int btrfs_drop_extents(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root, struct inode *inode,
+		       u64 start, u64 end, u64 inline_limit, u64 *hint_block);
+int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct inode *inode, u64 start, u64 end);
+int btrfs_release_file(struct inode *inode, struct file *file);
+
+/* tree-defrag.c */
+int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root, int cache_only);
+
+/* sysfs.c */
+int btrfs_init_sysfs(void);
+void btrfs_exit_sysfs(void);
+int btrfs_sysfs_add_super(struct btrfs_fs_info *fs);
+int btrfs_sysfs_add_root(struct btrfs_root *root);
+void btrfs_sysfs_del_root(struct btrfs_root *root);
+void btrfs_sysfs_del_super(struct btrfs_fs_info *root);
+
+/* xattr.c */
+ssize_t btrfs_listxattr(struct dentry *dentry, char *buffer, size_t size);
+
+/* super.c */
+u64 btrfs_parse_size(char *str);
+int btrfs_parse_options(struct btrfs_root *root, char *options);
+int btrfs_sync_fs(struct super_block *sb, int wait);
+
+/* acl.c */
+int btrfs_check_acl(struct inode *inode, int mask);
+int btrfs_init_acl(struct inode *inode, struct inode *dir);
+int btrfs_acl_chmod(struct inode *inode);
+
+/* free-space-cache.c */
+int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
+			 u64 bytenr, u64 size);
+int btrfs_add_free_space_lock(struct btrfs_block_group_cache *block_group,
+			      u64 offset, u64 bytes);
+int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
+			    u64 bytenr, u64 size);
+int btrfs_remove_free_space_lock(struct btrfs_block_group_cache *block_group,
+				 u64 offset, u64 bytes);
+void btrfs_remove_free_space_cache(struct btrfs_block_group_cache
+				   *block_group);
+struct btrfs_free_space *btrfs_find_free_space(struct btrfs_block_group_cache
+					       *block_group, u64 offset,
+					       u64 bytes);
+void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
+			   u64 bytes);
+u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group);
+#endif
-- 
1.6.0.2

[PATCH 27/35] Btrfs in-memory inode.h

[Chris Mason]

This defines the btrfs in-memory inode.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
new file mode 100644
index 0000000..a8c9693
--- /dev/null
+++ b/fs/btrfs/btrfs_inode.h
@@ -0,0 +1,131 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_I__
+#define __BTRFS_I__
+
+#include "extent_map.h"
+#include "extent_io.h"
+#include "ordered-data.h"
+
+/* in memory btrfs inode */
+struct btrfs_inode {
+	/* which subvolume this inode belongs to */
+	struct btrfs_root *root;
+
+	/* key used to find this inode on disk.  This is used by the code
+	 * to read in roots of subvolumes
+	 */
+	struct btrfs_key location;
+
+	/* the extent_tree has caches of all the extent mappings to disk */
+	struct extent_map_tree extent_tree;
+
+	/* the io_tree does range state (DIRTY, LOCKED etc) */
+	struct extent_io_tree io_tree;
+
+	/* special utility tree used to record which mirrors have already been
+	 * tried when checksums fail for a given block
+	 */
+	struct extent_io_tree io_failure_tree;
+
+	/* held while inesrting or deleting extents from files */
+	struct mutex extent_mutex;
+
+	/* held while logging the inode in tree-log.c */
+	struct mutex log_mutex;
+
+	/* used to order data wrt metadata */
+	struct btrfs_ordered_inode_tree ordered_tree;
+
+	/* standard acl pointers */
+	struct posix_acl *i_acl;
+	struct posix_acl *i_default_acl;
+
+	/* for keeping track of orphaned inodes */
+	struct list_head i_orphan;
+
+	/* list of all the delalloc inodes in the FS.  There are times we need
+	 * to write all the delalloc pages to disk, and this list is used
+	 * to walk them all.
+	 */
+	struct list_head delalloc_inodes;
+
+	/* full 64 bit generation number, struct vfs_inode doesn't have a big
+	 * enough field for this.
+	 */
+	u64 generation;
+
+	/* sequence number for NFS changes */
+	u64 sequence;
+
+	/*
+	 * transid of the trans_handle that last modified this inode
+	 */
+	u64 last_trans;
+	/*
+	 * transid that last logged this inode
+	 */
+	u64 logged_trans;
+
+	/*
+	 * trans that last made a change that should be fully fsync'd.  This
+	 * gets reset to zero each time the inode is logged
+	 */
+	u64 log_dirty_trans;
+
+	/* total number of bytes pending delalloc, used by stat to calc the
+	 * real block usage of the file
+	 */
+	u64 delalloc_bytes;
+
+	/*
+	 * the size of the file stored in the metadata on disk.  data=ordered
+	 * means the in-memory i_size might be larger than the size on disk
+	 * because not all the blocks are written yet.
+	 */
+	u64 disk_i_size;
+
+	/* flags field from the on disk inode */
+	u32 flags;
+
+	/*
+	 * if this is a directory then index_cnt is the counter for the index
+	 * number for new files that are created
+	 */
+	u64 index_cnt;
+
+	/* the start of block group preferred for allocations. */
+	u64 block_group;
+
+	struct inode vfs_inode;
+};
+
+static inline struct btrfs_inode *BTRFS_I(struct inode *inode)
+{
+	return container_of(inode, struct btrfs_inode, vfs_inode);
+}
+
+static inline void btrfs_i_size_write(struct inode *inode, u64 size)
+{
+	inode->i_size = size;
+	BTRFS_I(inode)->disk_i_size = size;
+}
+
+
+#endif
-- 
1.6.0.2

[PATCH 28/35] Btrfs acl implementation

[Chris Mason]

This is a fairly plain acl implementation that stores the posix acls
into xattrs

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/acl.c b/fs/btrfs/acl.c
new file mode 100644
index 0000000..1d53b62
--- /dev/null
+++ b/fs/btrfs/acl.c
@@ -0,0 +1,351 @@
+/*
+ * Copyright (C) 2007 Red Hat.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/string.h>
+#include <linux/xattr.h>
+#include <linux/posix_acl_xattr.h>
+#include <linux/posix_acl.h>
+#include <linux/sched.h>
+
+#include "ctree.h"
+#include "btrfs_inode.h"
+#include "xattr.h"
+
+#ifdef CONFIG_FS_POSIX_ACL
+
+static void btrfs_update_cached_acl(struct inode *inode,
+				    struct posix_acl **p_acl,
+				    struct posix_acl *acl)
+{
+	spin_lock(&inode->i_lock);
+	if (*p_acl && *p_acl != BTRFS_ACL_NOT_CACHED)
+		posix_acl_release(*p_acl);
+	*p_acl = posix_acl_dup(acl);
+	spin_unlock(&inode->i_lock);
+}
+
+static struct posix_acl *btrfs_get_acl(struct inode *inode, int type)
+{
+	int size;
+	const char *name;
+	char *value = NULL;
+	struct posix_acl *acl = NULL, **p_acl;
+
+	switch (type) {
+	case ACL_TYPE_ACCESS:
+		name = POSIX_ACL_XATTR_ACCESS;
+		p_acl = &BTRFS_I(inode)->i_acl;
+		break;
+	case ACL_TYPE_DEFAULT:
+		name = POSIX_ACL_XATTR_DEFAULT;
+		p_acl = &BTRFS_I(inode)->i_default_acl;
+		break;
+	default:
+		return ERR_PTR(-EINVAL);
+	}
+
+	spin_lock(&inode->i_lock);
+	if (*p_acl != BTRFS_ACL_NOT_CACHED)
+		acl = posix_acl_dup(*p_acl);
+	spin_unlock(&inode->i_lock);
+
+	if (acl)
+		return acl;
+
+
+	size = __btrfs_getxattr(inode, name, "", 0);
+	if (size > 0) {
+		value = kzalloc(size, GFP_NOFS);
+		if (!value)
+			return ERR_PTR(-ENOMEM);
+		size = __btrfs_getxattr(inode, name, value, size);
+		if (size > 0) {
+			acl = posix_acl_from_xattr(value, size);
+			btrfs_update_cached_acl(inode, p_acl, acl);
+		}
+		kfree(value);
+	} else if (size == -ENOENT) {
+		acl = NULL;
+		btrfs_update_cached_acl(inode, p_acl, acl);
+	}
+
+	return acl;
+}
+
+static int btrfs_xattr_get_acl(struct inode *inode, int type,
+			       void *value, size_t size)
+{
+	struct posix_acl *acl;
+	int ret = 0;
+
+	acl = btrfs_get_acl(inode, type);
+
+	if (IS_ERR(acl))
+		return PTR_ERR(acl);
+	if (acl == NULL)
+		return -ENODATA;
+	ret = posix_acl_to_xattr(acl, value, size);
+	posix_acl_release(acl);
+
+	return ret;
+}
+
+/*
+ * Needs to be called with fs_mutex held
+ */
+static int btrfs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
+{
+	int ret, size = 0;
+	const char *name;
+	struct posix_acl **p_acl;
+	char *value = NULL;
+	mode_t mode;
+
+	if (acl) {
+		ret = posix_acl_valid(acl);
+		if (ret < 0)
+			return ret;
+		ret = 0;
+	}
+
+	switch (type) {
+	case ACL_TYPE_ACCESS:
+		mode = inode->i_mode;
+		ret = posix_acl_equiv_mode(acl, &mode);
+		if (ret < 0)
+			return ret;
+		ret = 0;
+		inode->i_mode = mode;
+		name = POSIX_ACL_XATTR_ACCESS;
+		p_acl = &BTRFS_I(inode)->i_acl;
+		break;
+	case ACL_TYPE_DEFAULT:
+		if (!S_ISDIR(inode->i_mode))
+			return acl ? -EINVAL : 0;
+		name = POSIX_ACL_XATTR_DEFAULT;
+		p_acl = &BTRFS_I(inode)->i_default_acl;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (acl) {
+		size = posix_acl_xattr_size(acl->a_count);
+		value = kmalloc(size, GFP_NOFS);
+		if (!value) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		ret = posix_acl_to_xattr(acl, value, size);
+		if (ret < 0)
+			goto out;
+	}
+
+	ret = __btrfs_setxattr(inode, name, value, size, 0);
+
+out:
+	kfree(value);
+
+	if (!ret)
+		btrfs_update_cached_acl(inode, p_acl, acl);
+
+	return ret;
+}
+
+static int btrfs_xattr_set_acl(struct inode *inode, int type,
+			       const void *value, size_t size)
+{
+	int ret = 0;
+	struct posix_acl *acl = NULL;
+
+	if (value) {
+		acl = posix_acl_from_xattr(value, size);
+		if (acl == NULL) {
+			value = NULL;
+			size = 0;
+		} else if (IS_ERR(acl)) {
+			return PTR_ERR(acl);
+		}
+	}
+
+	ret = btrfs_set_acl(inode, acl, type);
+
+	posix_acl_release(acl);
+
+	return ret;
+}
+
+
+static int btrfs_xattr_acl_access_get(struct inode *inode, const char *name,
+				      void *value, size_t size)
+{
+	return btrfs_xattr_get_acl(inode, ACL_TYPE_ACCESS, value, size);
+}
+
+static int btrfs_xattr_acl_access_set(struct inode *inode, const char *name,
+				      const void *value, size_t size, int flags)
+{
+	return btrfs_xattr_set_acl(inode, ACL_TYPE_ACCESS, value, size);
+}
+
+static int btrfs_xattr_acl_default_get(struct inode *inode, const char *name,
+				       void *value, size_t size)
+{
+	return btrfs_xattr_get_acl(inode, ACL_TYPE_DEFAULT, value, size);
+}
+
+static int btrfs_xattr_acl_default_set(struct inode *inode, const char *name,
+			       const void *value, size_t size, int flags)
+{
+	return btrfs_xattr_set_acl(inode, ACL_TYPE_DEFAULT, value, size);
+}
+
+int btrfs_check_acl(struct inode *inode, int mask)
+{
+	struct posix_acl *acl;
+	int error = -EAGAIN;
+
+	acl = btrfs_get_acl(inode, ACL_TYPE_ACCESS);
+
+	if (IS_ERR(acl))
+		return PTR_ERR(acl);
+	if (acl) {
+		error = posix_acl_permission(inode, acl, mask);
+		posix_acl_release(acl);
+	}
+
+	return error;
+}
+
+/*
+ * btrfs_init_acl is already generally called under fs_mutex, so the locking
+ * stuff has been fixed to work with that.  If the locking stuff changes, we
+ * need to re-evaluate the acl locking stuff.
+ */
+int btrfs_init_acl(struct inode *inode, struct inode *dir)
+{
+	struct posix_acl *acl = NULL;
+	int ret = 0;
+
+	/* this happens with subvols */
+	if (!dir)
+		return 0;
+
+	if (!S_ISLNK(inode->i_mode)) {
+		if (IS_POSIXACL(dir)) {
+			acl = btrfs_get_acl(dir, ACL_TYPE_DEFAULT);
+			if (IS_ERR(acl))
+				return PTR_ERR(acl);
+		}
+
+		if (!acl)
+			inode->i_mode &= ~current->fs->umask;
+	}
+
+	if (IS_POSIXACL(dir) && acl) {
+		struct posix_acl *clone;
+		mode_t mode;
+
+		if (S_ISDIR(inode->i_mode)) {
+			ret = btrfs_set_acl(inode, acl, ACL_TYPE_DEFAULT);
+			if (ret)
+				goto failed;
+		}
+		clone = posix_acl_clone(acl, GFP_NOFS);
+		ret = -ENOMEM;
+		if (!clone)
+			goto failed;
+
+		mode = inode->i_mode;
+		ret = posix_acl_create_masq(clone, &mode);
+		if (ret >= 0) {
+			inode->i_mode = mode;
+			if (ret > 0) {
+				/* we need an acl */
+				ret = btrfs_set_acl(inode, clone,
+						    ACL_TYPE_ACCESS);
+			}
+		}
+	}
+failed:
+	posix_acl_release(acl);
+
+	return ret;
+}
+
+int btrfs_acl_chmod(struct inode *inode)
+{
+	struct posix_acl *acl, *clone;
+	int ret = 0;
+
+	if (S_ISLNK(inode->i_mode))
+		return -EOPNOTSUPP;
+
+	if (!IS_POSIXACL(inode))
+		return 0;
+
+	acl = btrfs_get_acl(inode, ACL_TYPE_ACCESS);
+	if (IS_ERR(acl) || !acl)
+		return PTR_ERR(acl);
+
+	clone = posix_acl_clone(acl, GFP_KERNEL);
+	posix_acl_release(acl);
+	if (!clone)
+		return -ENOMEM;
+
+	ret = posix_acl_chmod_masq(clone, inode->i_mode);
+	if (!ret)
+		ret = btrfs_set_acl(inode, clone, ACL_TYPE_ACCESS);
+
+	posix_acl_release(clone);
+
+	return ret;
+}
+
+struct xattr_handler btrfs_xattr_acl_default_handler = {
+	.prefix = POSIX_ACL_XATTR_DEFAULT,
+	.get	= btrfs_xattr_acl_default_get,
+	.set	= btrfs_xattr_acl_default_set,
+};
+
+struct xattr_handler btrfs_xattr_acl_access_handler = {
+	.prefix = POSIX_ACL_XATTR_ACCESS,
+	.get	= btrfs_xattr_acl_access_get,
+	.set	= btrfs_xattr_acl_access_set,
+};
+
+#else /* CONFIG_FS_POSIX_ACL */
+
+int btrfs_acl_chmod(struct inode *inode)
+{
+	return 0;
+}
+
+int btrfs_init_acl(struct inode *inode, struct inode *dir)
+{
+	return 0;
+}
+
+int btrfs_check_acl(struct inode *inode, int mask)
+{
+	return 0;
+}
+
+#endif /* CONFIG_FS_POSIX_ACL */
-- 
1.6.0.2

[PATCH 29/35] Btrfs ioctl code

[Chris Mason]

Btrfs ioctls for:

subvolume and snapshot creation
defragmentation
Resizing
Device add/remove
Single file cloning (COW copies)
Balancing extents across drives (restriping)

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
new file mode 100644
index 0000000..c2aa33e
--- /dev/null
+++ b/fs/btrfs/ioctl.c
@@ -0,0 +1,1132 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/kernel.h>
+#include <linux/bio.h>
+#include <linux/buffer_head.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/fsnotify.h>
+#include <linux/pagemap.h>
+#include <linux/highmem.h>
+#include <linux/time.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/smp_lock.h>
+#include <linux/backing-dev.h>
+#include <linux/mount.h>
+#include <linux/mpage.h>
+#include <linux/namei.h>
+#include <linux/swap.h>
+#include <linux/writeback.h>
+#include <linux/statfs.h>
+#include <linux/compat.h>
+#include <linux/bit_spinlock.h>
+#include <linux/security.h>
+#include <linux/version.h>
+#include <linux/xattr.h>
+#include <linux/vmalloc.h>
+#include "compat.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "btrfs_inode.h"
+#include "ioctl.h"
+#include "print-tree.h"
+#include "volumes.h"
+#include "locking.h"
+
+
+
+static noinline int create_subvol(struct btrfs_root *root,
+				  struct dentry *dentry,
+				  char *name, int namelen)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key key;
+	struct btrfs_root_item root_item;
+	struct btrfs_inode_item *inode_item;
+	struct extent_buffer *leaf;
+	struct btrfs_root *new_root = root;
+	struct inode *dir;
+	int ret;
+	int err;
+	u64 objectid;
+	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
+	u64 index = 0;
+	unsigned long nr = 1;
+
+	ret = btrfs_check_free_space(root, 1, 0);
+	if (ret)
+		goto fail_commit;
+
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(!trans);
+
+	ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
+				       0, &objectid);
+	if (ret)
+		goto fail;
+
+	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
+				      objectid, trans->transid, 0, 0, 0);
+	if (IS_ERR(leaf)) {
+		ret = PTR_ERR(leaf);
+		goto fail;
+	}
+
+	btrfs_set_header_nritems(leaf, 0);
+	btrfs_set_header_level(leaf, 0);
+	btrfs_set_header_bytenr(leaf, leaf->start);
+	btrfs_set_header_generation(leaf, trans->transid);
+	btrfs_set_header_owner(leaf, objectid);
+
+	write_extent_buffer(leaf, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(leaf),
+			    BTRFS_FSID_SIZE);
+	btrfs_mark_buffer_dirty(leaf);
+
+	inode_item = &root_item.inode;
+	memset(inode_item, 0, sizeof(*inode_item));
+	inode_item->generation = cpu_to_le64(1);
+	inode_item->size = cpu_to_le64(3);
+	inode_item->nlink = cpu_to_le32(1);
+	inode_item->nbytes = cpu_to_le64(root->leafsize);
+	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
+
+	btrfs_set_root_bytenr(&root_item, leaf->start);
+	btrfs_set_root_generation(&root_item, trans->transid);
+	btrfs_set_root_level(&root_item, 0);
+	btrfs_set_root_refs(&root_item, 1);
+	btrfs_set_root_used(&root_item, 0);
+	btrfs_set_root_last_snapshot(&root_item, 0);
+
+	memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
+	root_item.drop_level = 0;
+
+	btrfs_tree_unlock(leaf);
+	free_extent_buffer(leaf);
+	leaf = NULL;
+
+	btrfs_set_root_dirid(&root_item, new_dirid);
+
+	key.objectid = objectid;
+	key.offset = 1;
+	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
+				&root_item);
+	if (ret)
+		goto fail;
+
+	/*
+	 * insert the directory item
+	 */
+	key.offset = (u64)-1;
+	dir = dentry->d_parent->d_inode;
+	ret = btrfs_set_inode_index(dir, &index);
+	BUG_ON(ret);
+
+	ret = btrfs_insert_dir_item(trans, root,
+				    name, namelen, dir->i_ino, &key,
+				    BTRFS_FT_DIR, index);
+	if (ret)
+		goto fail;
+
+	btrfs_i_size_write(dir, dir->i_size + namelen * 2);
+	ret = btrfs_update_inode(trans, root, dir);
+	BUG_ON(ret);
+
+	/* add the backref first */
+	ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
+				 objectid, BTRFS_ROOT_BACKREF_KEY,
+				 root->root_key.objectid,
+				 dir->i_ino, index, name, namelen);
+
+	BUG_ON(ret);
+
+	/* now add the forward ref */
+	ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
+				 root->root_key.objectid, BTRFS_ROOT_REF_KEY,
+				 objectid,
+				 dir->i_ino, index, name, namelen);
+
+	BUG_ON(ret);
+
+	ret = btrfs_commit_transaction(trans, root);
+	if (ret)
+		goto fail_commit;
+
+	new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
+	BUG_ON(!new_root);
+
+	trans = btrfs_start_transaction(new_root, 1);
+	BUG_ON(!trans);
+
+	ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
+				       BTRFS_I(dir)->block_group);
+	if (ret)
+		goto fail;
+
+fail:
+	nr = trans->blocks_used;
+	err = btrfs_commit_transaction(trans, new_root);
+	if (err && !ret)
+		ret = err;
+fail_commit:
+	btrfs_btree_balance_dirty(root, nr);
+	return ret;
+}
+
+static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
+			   char *name, int namelen)
+{
+	struct btrfs_pending_snapshot *pending_snapshot;
+	struct btrfs_trans_handle *trans;
+	int ret = 0;
+	int err;
+	unsigned long nr = 0;
+
+	if (!root->ref_cows)
+		return -EINVAL;
+
+	ret = btrfs_check_free_space(root, 1, 0);
+	if (ret)
+		goto fail_unlock;
+
+	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
+	if (!pending_snapshot) {
+		ret = -ENOMEM;
+		goto fail_unlock;
+	}
+	pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
+	if (!pending_snapshot->name) {
+		ret = -ENOMEM;
+		kfree(pending_snapshot);
+		goto fail_unlock;
+	}
+	memcpy(pending_snapshot->name, name, namelen);
+	pending_snapshot->name[namelen] = '\0';
+	pending_snapshot->dentry = dentry;
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(!trans);
+	pending_snapshot->root = root;
+	list_add(&pending_snapshot->list,
+		 &trans->transaction->pending_snapshots);
+	err = btrfs_commit_transaction(trans, root);
+
+fail_unlock:
+	btrfs_btree_balance_dirty(root, nr);
+	return ret;
+}
+
+/* copy of may_create in fs/namei.c() */
+static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
+{
+	if (child->d_inode)
+		return -EEXIST;
+	if (IS_DEADDIR(dir))
+		return -ENOENT;
+	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
+}
+
+/*
+ * Create a new subvolume below @parent.  This is largely modeled after
+ * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
+ * inside this filesystem so it's quite a bit simpler.
+ */
+static noinline int btrfs_mksubvol(struct path *parent, char *name,
+				   int mode, int namelen,
+				   struct btrfs_root *snap_src)
+{
+	struct dentry *dentry;
+	int error;
+
+	mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+
+	dentry = lookup_one_len(name, parent->dentry, namelen);
+	error = PTR_ERR(dentry);
+	if (IS_ERR(dentry))
+		goto out_unlock;
+
+	error = -EEXIST;
+	if (dentry->d_inode)
+		goto out_dput;
+
+	if (!IS_POSIXACL(parent->dentry->d_inode))
+		mode &= ~current->fs->umask;
+
+	error = mnt_want_write(parent->mnt);
+	if (error)
+		goto out_dput;
+
+	error = btrfs_may_create(parent->dentry->d_inode, dentry);
+	if (error)
+		goto out_drop_write;
+
+	/*
+	 * Actually perform the low-level subvolume creation after all
+	 * this VFS fuzz.
+	 *
+	 * Eventually we want to pass in an inode under which we create this
+	 * subvolume, but for now all are under the filesystem root.
+	 *
+	 * Also we should pass on the mode eventually to allow creating new
+	 * subvolume with specific mode bits.
+	 */
+	if (snap_src) {
+		struct dentry *dir = dentry->d_parent;
+		struct dentry *test = dir->d_parent;
+		struct btrfs_path *path = btrfs_alloc_path();
+		int ret;
+		u64 test_oid;
+		u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
+
+		test_oid = snap_src->root_key.objectid;
+
+		ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
+					  path, parent_oid, test_oid);
+		if (ret == 0)
+			goto create;
+		btrfs_release_path(snap_src->fs_info->tree_root, path);
+
+		/* we need to make sure we aren't creating a directory loop
+		 * by taking a snapshot of something that has our current
+		 * subvol in its directory tree.  So, this loops through
+		 * the dentries and checks the forward refs for each subvolume
+		 * to see if is references the subvolume where we are
+		 * placing this new snapshot.
+		 */
+		while (1) {
+			if (!test ||
+			    dir == snap_src->fs_info->sb->s_root ||
+			    test == snap_src->fs_info->sb->s_root ||
+			    test->d_inode->i_sb != snap_src->fs_info->sb) {
+				break;
+			}
+			if (S_ISLNK(test->d_inode->i_mode)) {
+				printk(KERN_INFO "Btrfs symlink in snapshot "
+				       "path, failed\n");
+				error = -EMLINK;
+				btrfs_free_path(path);
+				goto out_drop_write;
+			}
+			test_oid =
+				BTRFS_I(test->d_inode)->root->root_key.objectid;
+			ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
+				  path, test_oid, parent_oid);
+			if (ret == 0) {
+				printk(KERN_INFO "Btrfs snapshot creation "
+				       "failed, looping\n");
+				error = -EMLINK;
+				btrfs_free_path(path);
+				goto out_drop_write;
+			}
+			btrfs_release_path(snap_src->fs_info->tree_root, path);
+			test = test->d_parent;
+		}
+create:
+		btrfs_free_path(path);
+		error = create_snapshot(snap_src, dentry, name, namelen);
+	} else {
+		error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
+				      dentry, name, namelen);
+	}
+	if (error)
+		goto out_drop_write;
+
+	fsnotify_mkdir(parent->dentry->d_inode, dentry);
+out_drop_write:
+	mnt_drop_write(parent->mnt);
+out_dput:
+	dput(dentry);
+out_unlock:
+	mutex_unlock(&parent->dentry->d_inode->i_mutex);
+	return error;
+}
+
+
+static int btrfs_defrag_file(struct file *file)
+{
+	struct inode *inode = fdentry(file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_ordered_extent *ordered;
+	struct page *page;
+	unsigned long last_index;
+	unsigned long ra_pages = root->fs_info->bdi.ra_pages;
+	unsigned long total_read = 0;
+	u64 page_start;
+	u64 page_end;
+	unsigned long i;
+	int ret;
+
+	ret = btrfs_check_free_space(root, inode->i_size, 0);
+	if (ret)
+		return -ENOSPC;
+
+	mutex_lock(&inode->i_mutex);
+	last_index = inode->i_size >> PAGE_CACHE_SHIFT;
+	for (i = 0; i <= last_index; i++) {
+		if (total_read % ra_pages == 0) {
+			btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
+				       min(last_index, i + ra_pages - 1));
+		}
+		total_read++;
+again:
+		page = grab_cache_page(inode->i_mapping, i);
+		if (!page)
+			goto out_unlock;
+		if (!PageUptodate(page)) {
+			btrfs_readpage(NULL, page);
+			lock_page(page);
+			if (!PageUptodate(page)) {
+				unlock_page(page);
+				page_cache_release(page);
+				goto out_unlock;
+			}
+		}
+
+		wait_on_page_writeback(page);
+
+		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
+		page_end = page_start + PAGE_CACHE_SIZE - 1;
+		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
+
+		ordered = btrfs_lookup_ordered_extent(inode, page_start);
+		if (ordered) {
+			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			btrfs_start_ordered_extent(inode, ordered, 1);
+			btrfs_put_ordered_extent(ordered);
+			goto again;
+		}
+		set_page_extent_mapped(page);
+
+		/*
+		 * this makes sure page_mkwrite is called on the
+		 * page if it is dirtied again later
+		 */
+		clear_page_dirty_for_io(page);
+
+		btrfs_set_extent_delalloc(inode, page_start, page_end);
+
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+		set_page_dirty(page);
+		unlock_page(page);
+		page_cache_release(page);
+		balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
+	}
+
+out_unlock:
+	mutex_unlock(&inode->i_mutex);
+	return 0;
+}
+
+/*
+ * Called inside transaction, so use GFP_NOFS
+ */
+
+static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
+{
+	u64 new_size;
+	u64 old_size;
+	u64 devid = 1;
+	struct btrfs_ioctl_vol_args *vol_args;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_device *device = NULL;
+	char *sizestr;
+	char *devstr = NULL;
+	int ret = 0;
+	int namelen;
+	int mod = 0;
+
+	if (root->fs_info->sb->s_flags & MS_RDONLY)
+		return -EROFS;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
+
+	if (!vol_args)
+		return -ENOMEM;
+
+	if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
+	namelen = strlen(vol_args->name);
+
+	mutex_lock(&root->fs_info->volume_mutex);
+	sizestr = vol_args->name;
+	devstr = strchr(sizestr, ':');
+	if (devstr) {
+		char *end;
+		sizestr = devstr + 1;
+		*devstr = '\0';
+		devstr = vol_args->name;
+		devid = simple_strtoull(devstr, &end, 10);
+		printk(KERN_INFO "resizing devid %llu\n", devid);
+	}
+	device = btrfs_find_device(root, devid, NULL, NULL);
+	if (!device) {
+		printk(KERN_INFO "resizer unable to find device %llu\n", devid);
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+	if (!strcmp(sizestr, "max"))
+		new_size = device->bdev->bd_inode->i_size;
+	else {
+		if (sizestr[0] == '-') {
+			mod = -1;
+			sizestr++;
+		} else if (sizestr[0] == '+') {
+			mod = 1;
+			sizestr++;
+		}
+		new_size = btrfs_parse_size(sizestr);
+		if (new_size == 0) {
+			ret = -EINVAL;
+			goto out_unlock;
+		}
+	}
+
+	old_size = device->total_bytes;
+
+	if (mod < 0) {
+		if (new_size > old_size) {
+			ret = -EINVAL;
+			goto out_unlock;
+		}
+		new_size = old_size - new_size;
+	} else if (mod > 0) {
+		new_size = old_size + new_size;
+	}
+
+	if (new_size < 256 * 1024 * 1024) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+	if (new_size > device->bdev->bd_inode->i_size) {
+		ret = -EFBIG;
+		goto out_unlock;
+	}
+
+	do_div(new_size, root->sectorsize);
+	new_size *= root->sectorsize;
+
+	printk(KERN_INFO "new size for %s is %llu\n",
+		device->name, (unsigned long long)new_size);
+
+	if (new_size > old_size) {
+		trans = btrfs_start_transaction(root, 1);
+		ret = btrfs_grow_device(trans, device, new_size);
+		btrfs_commit_transaction(trans, root);
+	} else {
+		ret = btrfs_shrink_device(device, new_size);
+	}
+
+out_unlock:
+	mutex_unlock(&root->fs_info->volume_mutex);
+out:
+	kfree(vol_args);
+	return ret;
+}
+
+static noinline int btrfs_ioctl_snap_create(struct file *file,
+					    void __user *arg, int subvol)
+{
+	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
+	struct btrfs_ioctl_vol_args *vol_args;
+	struct btrfs_dir_item *di;
+	struct btrfs_path *path;
+	struct file *src_file;
+	u64 root_dirid;
+	int namelen;
+	int ret = 0;
+
+	if (root->fs_info->sb->s_flags & MS_RDONLY)
+		return -EROFS;
+
+	vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
+
+	if (!vol_args)
+		return -ENOMEM;
+
+	if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
+	namelen = strlen(vol_args->name);
+	if (strchr(vol_args->name, '/')) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
+	di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
+			    path, root_dirid,
+			    vol_args->name, namelen, 0);
+	btrfs_free_path(path);
+
+	if (di && !IS_ERR(di)) {
+		ret = -EEXIST;
+		goto out;
+	}
+
+	if (IS_ERR(di)) {
+		ret = PTR_ERR(di);
+		goto out;
+	}
+
+	if (subvol) {
+		ret = btrfs_mksubvol(&file->f_path, vol_args->name,
+				     file->f_path.dentry->d_inode->i_mode,
+				     namelen, NULL);
+	} else {
+		struct inode *src_inode;
+		src_file = fget(vol_args->fd);
+		if (!src_file) {
+			ret = -EINVAL;
+			goto out;
+		}
+
+		src_inode = src_file->f_path.dentry->d_inode;
+		if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
+			printk(KERN_INFO "btrfs: Snapshot src from "
+			       "another FS\n");
+			ret = -EINVAL;
+			fput(src_file);
+			goto out;
+		}
+		ret = btrfs_mksubvol(&file->f_path, vol_args->name,
+			     file->f_path.dentry->d_inode->i_mode,
+			     namelen, BTRFS_I(src_inode)->root);
+		fput(src_file);
+	}
+
+out:
+	kfree(vol_args);
+	return ret;
+}
+
+static int btrfs_ioctl_defrag(struct file *file)
+{
+	struct inode *inode = fdentry(file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret;
+
+	ret = mnt_want_write(file->f_path.mnt);
+	if (ret)
+		return ret;
+
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFDIR:
+		if (!capable(CAP_SYS_ADMIN)) {
+			ret = -EPERM;
+			goto out;
+		}
+		btrfs_defrag_root(root, 0);
+		btrfs_defrag_root(root->fs_info->extent_root, 0);
+		break;
+	case S_IFREG:
+		if (!(file->f_mode & FMODE_WRITE)) {
+			ret = -EINVAL;
+			goto out;
+		}
+		btrfs_defrag_file(file);
+		break;
+	}
+out:
+	mnt_drop_write(file->f_path.mnt);
+	return ret;
+}
+
+static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
+{
+	struct btrfs_ioctl_vol_args *vol_args;
+	int ret;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
+
+	if (!vol_args)
+		return -ENOMEM;
+
+	if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
+		ret = -EFAULT;
+		goto out;
+	}
+	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
+	ret = btrfs_init_new_device(root, vol_args->name);
+
+out:
+	kfree(vol_args);
+	return ret;
+}
+
+static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
+{
+	struct btrfs_ioctl_vol_args *vol_args;
+	int ret;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (root->fs_info->sb->s_flags & MS_RDONLY)
+		return -EROFS;
+
+	vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
+
+	if (!vol_args)
+		return -ENOMEM;
+
+	if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
+		ret = -EFAULT;
+		goto out;
+	}
+	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
+	ret = btrfs_rm_device(root, vol_args->name);
+
+out:
+	kfree(vol_args);
+	return ret;
+}
+
+static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
+		u64 off, u64 olen, u64 destoff)
+{
+	struct inode *inode = fdentry(file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct file *src_file;
+	struct inode *src;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	char *buf;
+	struct btrfs_key key;
+	u32 nritems;
+	int slot;
+	int ret;
+	u64 len = olen;
+	u64 bs = root->fs_info->sb->s_blocksize;
+	u64 hint_byte;
+
+	/*
+	 * TODO:
+	 * - split compressed inline extents.  annoying: we need to
+	 *   decompress into destination's address_space (the file offset
+	 *   may change, so source mapping won't do), then recompress (or
+	 *   otherwise reinsert) a subrange.
+	 * - allow ranges within the same file to be cloned (provided
+	 *   they don't overlap)?
+	 */
+
+	/* the destination must be opened for writing */
+	if (!(file->f_mode & FMODE_WRITE))
+		return -EINVAL;
+
+	ret = mnt_want_write(file->f_path.mnt);
+	if (ret)
+		return ret;
+
+	src_file = fget(srcfd);
+	if (!src_file) {
+		ret = -EBADF;
+		goto out_drop_write;
+	}
+	src = src_file->f_dentry->d_inode;
+
+	ret = -EINVAL;
+	if (src == inode)
+		goto out_fput;
+
+	ret = -EISDIR;
+	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
+		goto out_fput;
+
+	ret = -EXDEV;
+	if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
+		goto out_fput;
+
+	ret = -ENOMEM;
+	buf = vmalloc(btrfs_level_size(root, 0));
+	if (!buf)
+		goto out_fput;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		vfree(buf);
+		goto out_fput;
+	}
+	path->reada = 2;
+
+	if (inode < src) {
+		mutex_lock(&inode->i_mutex);
+		mutex_lock(&src->i_mutex);
+	} else {
+		mutex_lock(&src->i_mutex);
+		mutex_lock(&inode->i_mutex);
+	}
+
+	/* determine range to clone */
+	ret = -EINVAL;
+	if (off >= src->i_size || off + len > src->i_size)
+		goto out_unlock;
+	if (len == 0)
+		olen = len = src->i_size - off;
+	/* if we extend to eof, continue to block boundary */
+	if (off + len == src->i_size)
+		len = ((src->i_size + bs-1) & ~(bs-1))
+			- off;
+
+	/* verify the end result is block aligned */
+	if ((off & (bs-1)) ||
+	    ((off + len) & (bs-1)))
+		goto out_unlock;
+
+	/* do any pending delalloc/csum calc on src, one way or
+	   another, and lock file content */
+	while (1) {
+		struct btrfs_ordered_extent *ordered;
+		lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
+		ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
+		if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
+			break;
+		unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
+		if (ordered)
+			btrfs_put_ordered_extent(ordered);
+		btrfs_wait_ordered_range(src, off, off+len);
+	}
+
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(!trans);
+
+	/* punch hole in destination first */
+	btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
+
+	/* clone data */
+	key.objectid = src->i_ino;
+	key.type = BTRFS_EXTENT_DATA_KEY;
+	key.offset = 0;
+
+	while (1) {
+		/*
+		 * note the key will change type as we walk through the
+		 * tree.
+		 */
+		ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
+		if (ret < 0)
+			goto out;
+
+		nritems = btrfs_header_nritems(path->nodes[0]);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0)
+				break;
+			nritems = btrfs_header_nritems(path->nodes[0]);
+		}
+		leaf = path->nodes[0];
+		slot = path->slots[0];
+
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+		if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
+		    key.objectid != src->i_ino)
+			break;
+
+		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
+			struct btrfs_file_extent_item *extent;
+			int type;
+			u32 size;
+			struct btrfs_key new_key;
+			u64 disko = 0, diskl = 0;
+			u64 datao = 0, datal = 0;
+			u8 comp;
+
+			size = btrfs_item_size_nr(leaf, slot);
+			read_extent_buffer(leaf, buf,
+					   btrfs_item_ptr_offset(leaf, slot),
+					   size);
+
+			extent = btrfs_item_ptr(leaf, slot,
+						struct btrfs_file_extent_item);
+			comp = btrfs_file_extent_compression(leaf, extent);
+			type = btrfs_file_extent_type(leaf, extent);
+			if (type == BTRFS_FILE_EXTENT_REG) {
+				disko = btrfs_file_extent_disk_bytenr(leaf,
+								      extent);
+				diskl = btrfs_file_extent_disk_num_bytes(leaf,
+								 extent);
+				datao = btrfs_file_extent_offset(leaf, extent);
+				datal = btrfs_file_extent_num_bytes(leaf,
+								    extent);
+			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
+				/* take upper bound, may be compressed */
+				datal = btrfs_file_extent_ram_bytes(leaf,
+								    extent);
+			}
+			btrfs_release_path(root, path);
+
+			if (key.offset + datal < off ||
+			    key.offset >= off+len)
+				goto next;
+
+			memcpy(&new_key, &key, sizeof(new_key));
+			new_key.objectid = inode->i_ino;
+			new_key.offset = key.offset + destoff - off;
+
+			if (type == BTRFS_FILE_EXTENT_REG) {
+				ret = btrfs_insert_empty_item(trans, root, path,
+							      &new_key, size);
+				if (ret)
+					goto out;
+
+				leaf = path->nodes[0];
+				slot = path->slots[0];
+				write_extent_buffer(leaf, buf,
+					    btrfs_item_ptr_offset(leaf, slot),
+					    size);
+
+				extent = btrfs_item_ptr(leaf, slot,
+						struct btrfs_file_extent_item);
+
+				if (off > key.offset) {
+					datao += off - key.offset;
+					datal -= off - key.offset;
+				}
+				if (key.offset + datao + datal + key.offset >
+				    off + len)
+					datal = off + len - key.offset - datao;
+				/* disko == 0 means it's a hole */
+				if (!disko)
+					datao = 0;
+
+				btrfs_set_file_extent_offset(leaf, extent,
+							     datao);
+				btrfs_set_file_extent_num_bytes(leaf, extent,
+								datal);
+				if (disko) {
+					inode_add_bytes(inode, datal);
+					ret = btrfs_inc_extent_ref(trans, root,
+						   disko, diskl, leaf->start,
+						   root->root_key.objectid,
+						   trans->transid,
+						   inode->i_ino);
+					BUG_ON(ret);
+				}
+			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
+				u64 skip = 0;
+				u64 trim = 0;
+				if (off > key.offset) {
+					skip = off - key.offset;
+					new_key.offset += skip;
+				}
+
+				if (key.offset + datal > off+len)
+					trim = key.offset + datal - (off+len);
+
+				if (comp && (skip || trim)) {
+					ret = -EINVAL;
+					goto out;
+				}
+				size -= skip + trim;
+				datal -= skip + trim;
+				ret = btrfs_insert_empty_item(trans, root, path,
+							      &new_key, size);
+				if (ret)
+					goto out;
+
+				if (skip) {
+					u32 start =
+					  btrfs_file_extent_calc_inline_size(0);
+					memmove(buf+start, buf+start+skip,
+						datal);
+				}
+
+				leaf = path->nodes[0];
+				slot = path->slots[0];
+				write_extent_buffer(leaf, buf,
+					    btrfs_item_ptr_offset(leaf, slot),
+					    size);
+				inode_add_bytes(inode, datal);
+			}
+
+			btrfs_mark_buffer_dirty(leaf);
+		}
+
+next:
+		btrfs_release_path(root, path);
+		key.offset++;
+	}
+	ret = 0;
+out:
+	btrfs_release_path(root, path);
+	if (ret == 0) {
+		inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+		if (destoff + olen > inode->i_size)
+			btrfs_i_size_write(inode, destoff + olen);
+		BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
+		ret = btrfs_update_inode(trans, root, inode);
+	}
+	btrfs_end_transaction(trans, root);
+	unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
+	if (ret)
+		vmtruncate(inode, 0);
+out_unlock:
+	mutex_unlock(&src->i_mutex);
+	mutex_unlock(&inode->i_mutex);
+	vfree(buf);
+	btrfs_free_path(path);
+out_fput:
+	fput(src_file);
+out_drop_write:
+	mnt_drop_write(file->f_path.mnt);
+	return ret;
+}
+
+static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
+{
+	struct btrfs_ioctl_clone_range_args args;
+
+	if (copy_from_user(&args, argp, sizeof(args)))
+		return -EFAULT;
+	return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
+				 args.src_length, args.dest_offset);
+}
+
+/*
+ * there are many ways the trans_start and trans_end ioctls can lead
+ * to deadlocks.  They should only be used by applications that
+ * basically own the machine, and have a very in depth understanding
+ * of all the possible deadlocks and enospc problems.
+ */
+static long btrfs_ioctl_trans_start(struct file *file)
+{
+	struct inode *inode = fdentry(file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	int ret = 0;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (file->private_data) {
+		ret = -EINPROGRESS;
+		goto out;
+	}
+
+	ret = mnt_want_write(file->f_path.mnt);
+	if (ret)
+		goto out;
+
+	mutex_lock(&root->fs_info->trans_mutex);
+	root->fs_info->open_ioctl_trans++;
+	mutex_unlock(&root->fs_info->trans_mutex);
+
+	trans = btrfs_start_ioctl_transaction(root, 0);
+	if (trans)
+		file->private_data = trans;
+	else
+		ret = -ENOMEM;
+	/*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
+out:
+	return ret;
+}
+
+/*
+ * there are many ways the trans_start and trans_end ioctls can lead
+ * to deadlocks.  They should only be used by applications that
+ * basically own the machine, and have a very in depth understanding
+ * of all the possible deadlocks and enospc problems.
+ */
+long btrfs_ioctl_trans_end(struct file *file)
+{
+	struct inode *inode = fdentry(file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	int ret = 0;
+
+	trans = file->private_data;
+	if (!trans) {
+		ret = -EINVAL;
+		goto out;
+	}
+	btrfs_end_transaction(trans, root);
+	file->private_data = NULL;
+
+	mutex_lock(&root->fs_info->trans_mutex);
+	root->fs_info->open_ioctl_trans--;
+	mutex_unlock(&root->fs_info->trans_mutex);
+
+	mnt_drop_write(file->f_path.mnt);
+
+out:
+	return ret;
+}
+
+long btrfs_ioctl(struct file *file, unsigned int
+		cmd, unsigned long arg)
+{
+	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
+	void __user *argp = (void __user *)arg;
+
+	switch (cmd) {
+	case BTRFS_IOC_SNAP_CREATE:
+		return btrfs_ioctl_snap_create(file, argp, 0);
+	case BTRFS_IOC_SUBVOL_CREATE:
+		return btrfs_ioctl_snap_create(file, argp, 1);
+	case BTRFS_IOC_DEFRAG:
+		return btrfs_ioctl_defrag(file);
+	case BTRFS_IOC_RESIZE:
+		return btrfs_ioctl_resize(root, argp);
+	case BTRFS_IOC_ADD_DEV:
+		return btrfs_ioctl_add_dev(root, argp);
+	case BTRFS_IOC_RM_DEV:
+		return btrfs_ioctl_rm_dev(root, argp);
+	case BTRFS_IOC_BALANCE:
+		return btrfs_balance(root->fs_info->dev_root);
+	case BTRFS_IOC_CLONE:
+		return btrfs_ioctl_clone(file, arg, 0, 0, 0);
+	case BTRFS_IOC_CLONE_RANGE:
+		return btrfs_ioctl_clone_range(file, argp);
+	case BTRFS_IOC_TRANS_START:
+		return btrfs_ioctl_trans_start(file);
+	case BTRFS_IOC_TRANS_END:
+		return btrfs_ioctl_trans_end(file);
+	case BTRFS_IOC_SYNC:
+		btrfs_sync_fs(file->f_dentry->d_sb, 1);
+		return 0;
+	}
+
+	return -ENOTTY;
+}
diff --git a/fs/btrfs/ioctl.h b/fs/btrfs/ioctl.h
new file mode 100644
index 0000000..78049ea
--- /dev/null
+++ b/fs/btrfs/ioctl.h
@@ -0,0 +1,67 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __IOCTL_
+#define __IOCTL_
+#include <linux/ioctl.h>
+
+#define BTRFS_IOCTL_MAGIC 0x94
+#define BTRFS_VOL_NAME_MAX 255
+#define BTRFS_PATH_NAME_MAX 3072
+
+struct btrfs_ioctl_vol_args {
+	__s64 fd;
+	char name[BTRFS_PATH_NAME_MAX + 1];
+};
+
+#define BTRFS_IOC_SNAP_CREATE _IOW(BTRFS_IOCTL_MAGIC, 1, \
+				   struct btrfs_ioctl_vol_args)
+#define BTRFS_IOC_DEFRAG _IOW(BTRFS_IOCTL_MAGIC, 2, \
+				   struct btrfs_ioctl_vol_args)
+#define BTRFS_IOC_RESIZE _IOW(BTRFS_IOCTL_MAGIC, 3, \
+				   struct btrfs_ioctl_vol_args)
+#define BTRFS_IOC_SCAN_DEV _IOW(BTRFS_IOCTL_MAGIC, 4, \
+				   struct btrfs_ioctl_vol_args)
+/* trans start and trans end are dangerous, and only for
+ * use by applications that know how to avoid the
+ * resulting deadlocks
+ */
+#define BTRFS_IOC_TRANS_START  _IO(BTRFS_IOCTL_MAGIC, 6)
+#define BTRFS_IOC_TRANS_END    _IO(BTRFS_IOCTL_MAGIC, 7)
+#define BTRFS_IOC_SYNC         _IO(BTRFS_IOCTL_MAGIC, 8)
+
+#define BTRFS_IOC_CLONE        _IOW(BTRFS_IOCTL_MAGIC, 9, int)
+#define BTRFS_IOC_ADD_DEV _IOW(BTRFS_IOCTL_MAGIC, 10, \
+				   struct btrfs_ioctl_vol_args)
+#define BTRFS_IOC_RM_DEV _IOW(BTRFS_IOCTL_MAGIC, 11, \
+				   struct btrfs_ioctl_vol_args)
+#define BTRFS_IOC_BALANCE _IOW(BTRFS_IOCTL_MAGIC, 12, \
+				   struct btrfs_ioctl_vol_args)
+struct btrfs_ioctl_clone_range_args {
+  __s64 src_fd;
+  __u64 src_offset, src_length;
+  __u64 dest_offset;
+};
+
+#define BTRFS_IOC_CLONE_RANGE _IOW(BTRFS_IOCTL_MAGIC, 13, \
+				  struct btrfs_ioctl_clone_range_args)
+
+#define BTRFS_IOC_SUBVOL_CREATE _IOW(BTRFS_IOCTL_MAGIC, 14, \
+				   struct btrfs_ioctl_vol_args)
+
+#endif
-- 
1.6.0.2

Re: [PATCH 29/35] Btrfs ioctl code

[Ryusuke Konishi]

Hi Chris,
On Wed,  7 Jan 2009 22:57:19 -0500, Chris Mason wrote:
> Btrfs ioctls for:
> 
> subvolume and snapshot creation
> defragmentation
> Resizing
> Device add/remove
> Single file cloning (COW copies)
> Balancing extents across drives (restriping)
<snip>
> +
> +#define BTRFS_IOCTL_MAGIC 0x94
> +#define BTRFS_VOL_NAME_MAX 255
> +#define BTRFS_PATH_NAME_MAX 3072
> +
> +struct btrfs_ioctl_vol_args {
> +	__s64 fd;
> +	char name[BTRFS_PATH_NAME_MAX + 1];
> +};
> +
> +#define BTRFS_IOC_SNAP_CREATE _IOW(BTRFS_IOCTL_MAGIC, 1, \
> +				   struct btrfs_ioctl_vol_args)
> +#define BTRFS_IOC_DEFRAG _IOW(BTRFS_IOCTL_MAGIC, 2, \
> +				   struct btrfs_ioctl_vol_args)
> +#define BTRFS_IOC_RESIZE _IOW(BTRFS_IOCTL_MAGIC, 3, \
> +				   struct btrfs_ioctl_vol_args)
> +#define BTRFS_IOC_SCAN_DEV _IOW(BTRFS_IOCTL_MAGIC, 4, \
> +				   struct btrfs_ioctl_vol_args)

The size of btrfs_ioctl_vol_args seems to depend on architectures.

For example, the size was 3084 bytes on a x86-32 machine, but was 3088
bytes on a x86-64 machine.

Doesn't it cause mismatch of IOC values between the kernel and
userland in some environments ?

Regards,
Ryusuke Konishi

Re: [PATCH 29/35] Btrfs ioctl code

[Linus Torvalds]

On Fri, 16 Jan 2009, Ryusuke Konishi wrote:

> Hi Chris,
> On Wed,  7 Jan 2009 22:57:19 -0500, Chris Mason wrote:
> > +
> > +struct btrfs_ioctl_vol_args {
> > +	__s64 fd;
> > +	char name[BTRFS_PATH_NAME_MAX + 1];
> > +};
> 
> The size of btrfs_ioctl_vol_args seems to depend on architectures.
> 
> For example, the size was 3084 bytes on a x86-32 machine, but was 3088
> bytes on a x86-64 machine.

Grr. Different alignment for 64-bit values, coupled with "name" not 
having a size that is a multiple-of-eight.

Chris, I think we want to fix this.  Even if it breaks binary 
compatibility with previous tools. I don't think we want to have compat 
ioctls for something that is so close to already working.

		Linus

Re: [PATCH 29/35] Btrfs ioctl code

[Chris Mason]

On Thu, 2009-01-15 at 12:17 -0800, Linus Torvalds wrote:
> 
> On Fri, 16 Jan 2009, Ryusuke Konishi wrote:
> 
> > Hi Chris,
> > On Wed,  7 Jan 2009 22:57:19 -0500, Chris Mason wrote:
> > > +
> > > +struct btrfs_ioctl_vol_args {
> > > +	__s64 fd;
> > > +	char name[BTRFS_PATH_NAME_MAX + 1];
> > > +};
> > 
> > The size of btrfs_ioctl_vol_args seems to depend on architectures.
> > 
> > For example, the size was 3084 bytes on a x86-32 machine, but was 3088
> > bytes on a x86-64 machine.
> 
> Grr. Different alignment for 64-bit values, coupled with "name" not 
> having a size that is a multiple-of-eight.
> 

Will the fd and the start of the name get aligned differently?  I'll fix
it up

-chris

Re: [PATCH 29/35] Btrfs ioctl code

[Linus Torvalds]

On Thu, 15 Jan 2009, Chris Mason wrote:
> > 
> > Grr. Different alignment for 64-bit values, coupled with "name" not 
> > having a size that is a multiple-of-eight.
> 
> Will the fd and the start of the name get aligned differently?  I'll fix
> it up

No, since the name is byte-aligned, the only thing that happens is that 
the total alignment (which comes from the alignment of __s64) affects the 
final size - think about having an array of those things, and the 
"sizeof()" gives you what the offset to the next entry in the array would 
be.

So they'll look _exactly_ the same in all ways, except for padding at the 
end.

			Linus

Re: [PATCH 29/35] Btrfs ioctl code

[Chris Mason]

On Thu, 2009-01-15 at 12:44 -0800, Linus Torvalds wrote:
> 
> On Thu, 15 Jan 2009, Chris Mason wrote:
> > > 
> > > Grr. Different alignment for 64-bit values, coupled with "name" not 
> > > having a size that is a multiple-of-eight.
> > 
> > Will the fd and the start of the name get aligned differently?  I'll fix
> > it up
> 
> No, since the name is byte-aligned, the only thing that happens is that 
> the total alignment (which comes from the alignment of __s64) affects the 
> final size - think about having an array of those things, and the 
> "sizeof()" gives you what the offset to the next entry in the array would 
> be.
> 
> So they'll look _exactly_ the same in all ways, except for padding at the 
> end.

Ok, that makes more sense thanks.  The kernel does:

if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
	ret = -EFAULT;
	goto out;
}

So I'll try to get something that works as expected and just trims some
bytes off the end to align things.

-chris

Re: [PATCH 29/35] Btrfs ioctl code

[Chris Mason]

On Thu, 2009-01-15 at 12:17 -0800, Linus Torvalds wrote:
> 
> On Fri, 16 Jan 2009, Ryusuke Konishi wrote:
> 
> > Hi Chris,
> > On Wed,  7 Jan 2009 22:57:19 -0500, Chris Mason wrote:
> > > +
> > > +struct btrfs_ioctl_vol_args {
> > > +	__s64 fd;
> > > +	char name[BTRFS_PATH_NAME_MAX + 1];
> > > +};
> > 
> > The size of btrfs_ioctl_vol_args seems to depend on architectures.
> > 
> > For example, the size was 3084 bytes on a x86-32 machine, but was 3088
> > bytes on a x86-64 machine.
> 
> Grr. Different alignment for 64-bit values, coupled with "name" not 
> having a size that is a multiple-of-eight.
> 
> Chris, I think we want to fix this.  Even if it breaks binary 
> compatibility with previous tools. I don't think we want to have compat 
> ioctls for something that is so close to already working.

I've fixed both the kernel and the progs to use a 4k struct, and tested
all the btrfs_ioctl_var_arg ioctls on x86-64 with a program compiled on
x86-32.

Please pull from:

git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable.git

And I'll roll out v0.18 of btrfs-progs with a notice about the ioctl
interface change.

Ryusuke, thank you very much for finding this now.

-chris

Re: [PATCH 29/35] Btrfs ioctl code

[Ryusuke Konishi]

On Fri, 16 Jan 2009 12:23:03 -0500, Chris Mason wrote:
> On Thu, 2009-01-15 at 12:17 -0800, Linus Torvalds wrote:
> > 
> > On Fri, 16 Jan 2009, Ryusuke Konishi wrote:
> > 
> > > Hi Chris,
> > > On Wed,  7 Jan 2009 22:57:19 -0500, Chris Mason wrote:
> > > > +
> > > > +struct btrfs_ioctl_vol_args {
> > > > +	__s64 fd;
> > > > +	char name[BTRFS_PATH_NAME_MAX + 1];
> > > > +};
> > > 
> > > The size of btrfs_ioctl_vol_args seems to depend on architectures.
> > > 
> > > For example, the size was 3084 bytes on a x86-32 machine, but was 3088
> > > bytes on a x86-64 machine.
> > 
> > Grr. Different alignment for 64-bit values, coupled with "name" not 
> > having a size that is a multiple-of-eight.
> > 
> > Chris, I think we want to fix this.  Even if it breaks binary 
> > compatibility with previous tools. I don't think we want to have compat 
> > ioctls for something that is so close to already working.
> 
> I've fixed both the kernel and the progs to use a 4k struct, and tested
> all the btrfs_ioctl_var_arg ioctls on x86-64 with a program compiled on
> x86-32.
> 
> Ryusuke, thank you very much for finding this now.

You're welcome.
I'm glad to return the favor.

Ryusuke

[PATCH 30/35] Btrfs extent_map: per-file extent lookup cache

[Chris Mason]

Instead of storing the block numbers for file offsets inside of a buffer
head, btrfs uses an rbtree cache of extents.

This is a fairly simple API to maintain the cache, and it is primarily
used by the btrfs_get_extent() call in fs/btrfs/inode.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
new file mode 100644
index 0000000..4a83e33
--- /dev/null
+++ b/fs/btrfs/extent_map.c
@@ -0,0 +1,351 @@
+#include <linux/err.h>
+#include <linux/gfp.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/version.h>
+#include <linux/hardirq.h>
+#include "extent_map.h"
+
+/* temporary define until extent_map moves out of btrfs */
+struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
+				       unsigned long extra_flags,
+				       void (*ctor)(void *, struct kmem_cache *,
+						    unsigned long));
+
+static struct kmem_cache *extent_map_cache;
+
+int __init extent_map_init(void)
+{
+	extent_map_cache = btrfs_cache_create("extent_map",
+					    sizeof(struct extent_map), 0,
+					    NULL);
+	if (!extent_map_cache)
+		return -ENOMEM;
+	return 0;
+}
+
+void extent_map_exit(void)
+{
+	if (extent_map_cache)
+		kmem_cache_destroy(extent_map_cache);
+}
+
+/**
+ * extent_map_tree_init - initialize extent map tree
+ * @tree:		tree to initialize
+ * @mask:		flags for memory allocations during tree operations
+ *
+ * Initialize the extent tree @tree.  Should be called for each new inode
+ * or other user of the extent_map interface.
+ */
+void extent_map_tree_init(struct extent_map_tree *tree, gfp_t mask)
+{
+	tree->map.rb_node = NULL;
+	spin_lock_init(&tree->lock);
+}
+EXPORT_SYMBOL(extent_map_tree_init);
+
+/**
+ * alloc_extent_map - allocate new extent map structure
+ * @mask:	memory allocation flags
+ *
+ * Allocate a new extent_map structure.  The new structure is
+ * returned with a reference count of one and needs to be
+ * freed using free_extent_map()
+ */
+struct extent_map *alloc_extent_map(gfp_t mask)
+{
+	struct extent_map *em;
+	em = kmem_cache_alloc(extent_map_cache, mask);
+	if (!em || IS_ERR(em))
+		return em;
+	em->in_tree = 0;
+	em->flags = 0;
+	atomic_set(&em->refs, 1);
+	return em;
+}
+EXPORT_SYMBOL(alloc_extent_map);
+
+/**
+ * free_extent_map - drop reference count of an extent_map
+ * @em:		extent map beeing releasead
+ *
+ * Drops the reference out on @em by one and free the structure
+ * if the reference count hits zero.
+ */
+void free_extent_map(struct extent_map *em)
+{
+	if (!em)
+		return;
+	WARN_ON(atomic_read(&em->refs) == 0);
+	if (atomic_dec_and_test(&em->refs)) {
+		WARN_ON(em->in_tree);
+		kmem_cache_free(extent_map_cache, em);
+	}
+}
+EXPORT_SYMBOL(free_extent_map);
+
+static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
+				   struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct extent_map *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct extent_map, rb_node);
+
+		WARN_ON(!entry->in_tree);
+
+		if (offset < entry->start)
+			p = &(*p)->rb_left;
+		else if (offset >= extent_map_end(entry))
+			p = &(*p)->rb_right;
+		else
+			return parent;
+	}
+
+	entry = rb_entry(node, struct extent_map, rb_node);
+	entry->in_tree = 1;
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+/*
+ * search through the tree for an extent_map with a given offset.  If
+ * it can't be found, try to find some neighboring extents
+ */
+static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
+				     struct rb_node **prev_ret,
+				     struct rb_node **next_ret)
+{
+	struct rb_node *n = root->rb_node;
+	struct rb_node *prev = NULL;
+	struct rb_node *orig_prev = NULL;
+	struct extent_map *entry;
+	struct extent_map *prev_entry = NULL;
+
+	while (n) {
+		entry = rb_entry(n, struct extent_map, rb_node);
+		prev = n;
+		prev_entry = entry;
+
+		WARN_ON(!entry->in_tree);
+
+		if (offset < entry->start)
+			n = n->rb_left;
+		else if (offset >= extent_map_end(entry))
+			n = n->rb_right;
+		else
+			return n;
+	}
+
+	if (prev_ret) {
+		orig_prev = prev;
+		while (prev && offset >= extent_map_end(prev_entry)) {
+			prev = rb_next(prev);
+			prev_entry = rb_entry(prev, struct extent_map, rb_node);
+		}
+		*prev_ret = prev;
+		prev = orig_prev;
+	}
+
+	if (next_ret) {
+		prev_entry = rb_entry(prev, struct extent_map, rb_node);
+		while (prev && offset < prev_entry->start) {
+			prev = rb_prev(prev);
+			prev_entry = rb_entry(prev, struct extent_map, rb_node);
+		}
+		*next_ret = prev;
+	}
+	return NULL;
+}
+
+/*
+ * look for an offset in the tree, and if it can't be found, return
+ * the first offset we can find smaller than 'offset'.
+ */
+static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
+{
+	struct rb_node *prev;
+	struct rb_node *ret;
+	ret = __tree_search(root, offset, &prev, NULL);
+	if (!ret)
+		return prev;
+	return ret;
+}
+
+/* check to see if two extent_map structs are adjacent and safe to merge */
+static int mergable_maps(struct extent_map *prev, struct extent_map *next)
+{
+	if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
+		return 0;
+
+	/*
+	 * don't merge compressed extents, we need to know their
+	 * actual size
+	 */
+	if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
+		return 0;
+
+	if (extent_map_end(prev) == next->start &&
+	    prev->flags == next->flags &&
+	    prev->bdev == next->bdev &&
+	    ((next->block_start == EXTENT_MAP_HOLE &&
+	      prev->block_start == EXTENT_MAP_HOLE) ||
+	     (next->block_start == EXTENT_MAP_INLINE &&
+	      prev->block_start == EXTENT_MAP_INLINE) ||
+	     (next->block_start == EXTENT_MAP_DELALLOC &&
+	      prev->block_start == EXTENT_MAP_DELALLOC) ||
+	     (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
+	      next->block_start == extent_map_block_end(prev)))) {
+		return 1;
+	}
+	return 0;
+}
+
+/**
+ * add_extent_mapping - add new extent map to the extent tree
+ * @tree:	tree to insert new map in
+ * @em:		map to insert
+ *
+ * Insert @em into @tree or perform a simple forward/backward merge with
+ * existing mappings.  The extent_map struct passed in will be inserted
+ * into the tree directly, with an additional reference taken, or a
+ * reference dropped if the merge attempt was sucessfull.
+ */
+int add_extent_mapping(struct extent_map_tree *tree,
+		       struct extent_map *em)
+{
+	int ret = 0;
+	struct extent_map *merge = NULL;
+	struct rb_node *rb;
+	struct extent_map *exist;
+
+	exist = lookup_extent_mapping(tree, em->start, em->len);
+	if (exist) {
+		free_extent_map(exist);
+		ret = -EEXIST;
+		goto out;
+	}
+	assert_spin_locked(&tree->lock);
+	rb = tree_insert(&tree->map, em->start, &em->rb_node);
+	if (rb) {
+		ret = -EEXIST;
+		free_extent_map(merge);
+		goto out;
+	}
+	atomic_inc(&em->refs);
+	if (em->start != 0) {
+		rb = rb_prev(&em->rb_node);
+		if (rb)
+			merge = rb_entry(rb, struct extent_map, rb_node);
+		if (rb && mergable_maps(merge, em)) {
+			em->start = merge->start;
+			em->len += merge->len;
+			em->block_len += merge->block_len;
+			em->block_start = merge->block_start;
+			merge->in_tree = 0;
+			rb_erase(&merge->rb_node, &tree->map);
+			free_extent_map(merge);
+		}
+	 }
+	rb = rb_next(&em->rb_node);
+	if (rb)
+		merge = rb_entry(rb, struct extent_map, rb_node);
+	if (rb && mergable_maps(em, merge)) {
+		em->len += merge->len;
+		em->block_len += merge->len;
+		rb_erase(&merge->rb_node, &tree->map);
+		merge->in_tree = 0;
+		free_extent_map(merge);
+	}
+out:
+	return ret;
+}
+EXPORT_SYMBOL(add_extent_mapping);
+
+/* simple helper to do math around the end of an extent, handling wrap */
+static u64 range_end(u64 start, u64 len)
+{
+	if (start + len < start)
+		return (u64)-1;
+	return start + len;
+}
+
+/**
+ * lookup_extent_mapping - lookup extent_map
+ * @tree:	tree to lookup in
+ * @start:	byte offset to start the search
+ * @len:	length of the lookup range
+ *
+ * Find and return the first extent_map struct in @tree that intersects the
+ * [start, len] range.  There may be additional objects in the tree that
+ * intersect, so check the object returned carefully to make sure that no
+ * additional lookups are needed.
+ */
+struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
+					 u64 start, u64 len)
+{
+	struct extent_map *em;
+	struct rb_node *rb_node;
+	struct rb_node *prev = NULL;
+	struct rb_node *next = NULL;
+	u64 end = range_end(start, len);
+
+	assert_spin_locked(&tree->lock);
+	rb_node = __tree_search(&tree->map, start, &prev, &next);
+	if (!rb_node && prev) {
+		em = rb_entry(prev, struct extent_map, rb_node);
+		if (end > em->start && start < extent_map_end(em))
+			goto found;
+	}
+	if (!rb_node && next) {
+		em = rb_entry(next, struct extent_map, rb_node);
+		if (end > em->start && start < extent_map_end(em))
+			goto found;
+	}
+	if (!rb_node) {
+		em = NULL;
+		goto out;
+	}
+	if (IS_ERR(rb_node)) {
+		em = ERR_PTR(PTR_ERR(rb_node));
+		goto out;
+	}
+	em = rb_entry(rb_node, struct extent_map, rb_node);
+	if (end > em->start && start < extent_map_end(em))
+		goto found;
+
+	em = NULL;
+	goto out;
+
+found:
+	atomic_inc(&em->refs);
+out:
+	return em;
+}
+EXPORT_SYMBOL(lookup_extent_mapping);
+
+/**
+ * remove_extent_mapping - removes an extent_map from the extent tree
+ * @tree:	extent tree to remove from
+ * @em:		extent map beeing removed
+ *
+ * Removes @em from @tree.  No reference counts are dropped, and no checks
+ * are done to see if the range is in use
+ */
+int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
+{
+	int ret = 0;
+
+	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
+	assert_spin_locked(&tree->lock);
+	rb_erase(&em->rb_node, &tree->map);
+	em->in_tree = 0;
+	return ret;
+}
+EXPORT_SYMBOL(remove_extent_mapping);
diff --git a/fs/btrfs/extent_map.h b/fs/btrfs/extent_map.h
new file mode 100644
index 0000000..fb6eeef
--- /dev/null
+++ b/fs/btrfs/extent_map.h
@@ -0,0 +1,62 @@
+#ifndef __EXTENTMAP__
+#define __EXTENTMAP__
+
+#include <linux/rbtree.h>
+
+#define EXTENT_MAP_LAST_BYTE (u64)-4
+#define EXTENT_MAP_HOLE (u64)-3
+#define EXTENT_MAP_INLINE (u64)-2
+#define EXTENT_MAP_DELALLOC (u64)-1
+
+/* bits for the flags field */
+#define EXTENT_FLAG_PINNED 0 /* this entry not yet on disk, don't free it */
+#define EXTENT_FLAG_COMPRESSED 1
+#define EXTENT_FLAG_VACANCY 2 /* no file extent item found */
+#define EXTENT_FLAG_PREALLOC 3 /* pre-allocated extent */
+
+struct extent_map {
+	struct rb_node rb_node;
+
+	/* all of these are in bytes */
+	u64 start;
+	u64 len;
+	u64 orig_start;
+	u64 block_start;
+	u64 block_len;
+	unsigned long flags;
+	struct block_device *bdev;
+	atomic_t refs;
+	int in_tree;
+};
+
+struct extent_map_tree {
+	struct rb_root map;
+	spinlock_t lock;
+};
+
+static inline u64 extent_map_end(struct extent_map *em)
+{
+	if (em->start + em->len < em->start)
+		return (u64)-1;
+	return em->start + em->len;
+}
+
+static inline u64 extent_map_block_end(struct extent_map *em)
+{
+	if (em->block_start + em->block_len < em->block_start)
+		return (u64)-1;
+	return em->block_start + em->block_len;
+}
+
+void extent_map_tree_init(struct extent_map_tree *tree, gfp_t mask);
+struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
+					 u64 start, u64 len);
+int add_extent_mapping(struct extent_map_tree *tree,
+		       struct extent_map *em);
+int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em);
+
+struct extent_map *alloc_extent_map(gfp_t mask);
+void free_extent_map(struct extent_map *em);
+int __init extent_map_init(void);
+void extent_map_exit(void);
+#endif
-- 
1.6.0.2

[PATCH 32/35] Btrfs NFS exporting routines

[Chris Mason]

Btrfs inode numbers are only unique inside a subvolume, and so NFS needs
helper routines to send along the subvolume id along with the inode number.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/export.c b/fs/btrfs/export.c
new file mode 100644
index 0000000..85315d2
--- /dev/null
+++ b/fs/btrfs/export.c
@@ -0,0 +1,203 @@
+#include <linux/fs.h>
+#include <linux/types.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "btrfs_inode.h"
+#include "print-tree.h"
+#include "export.h"
+#include "compat.h"
+
+#define BTRFS_FID_SIZE_NON_CONNECTABLE (offsetof(struct btrfs_fid, \
+						 parent_objectid) / 4)
+#define BTRFS_FID_SIZE_CONNECTABLE (offsetof(struct btrfs_fid, \
+					     parent_root_objectid) / 4)
+#define BTRFS_FID_SIZE_CONNECTABLE_ROOT (sizeof(struct btrfs_fid) / 4)
+
+static int btrfs_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
+			   int connectable)
+{
+	struct btrfs_fid *fid = (struct btrfs_fid *)fh;
+	struct inode *inode = dentry->d_inode;
+	int len = *max_len;
+	int type;
+
+	if ((len < BTRFS_FID_SIZE_NON_CONNECTABLE) ||
+	    (connectable && len < BTRFS_FID_SIZE_CONNECTABLE))
+		return 255;
+
+	len  = BTRFS_FID_SIZE_NON_CONNECTABLE;
+	type = FILEID_BTRFS_WITHOUT_PARENT;
+
+	fid->objectid = BTRFS_I(inode)->location.objectid;
+	fid->root_objectid = BTRFS_I(inode)->root->objectid;
+	fid->gen = inode->i_generation;
+
+	if (connectable && !S_ISDIR(inode->i_mode)) {
+		struct inode *parent;
+		u64 parent_root_id;
+
+		spin_lock(&dentry->d_lock);
+
+		parent = dentry->d_parent->d_inode;
+		fid->parent_objectid = BTRFS_I(parent)->location.objectid;
+		fid->parent_gen = parent->i_generation;
+		parent_root_id = BTRFS_I(parent)->root->objectid;
+
+		spin_unlock(&dentry->d_lock);
+
+		if (parent_root_id != fid->root_objectid) {
+			fid->parent_root_objectid = parent_root_id;
+			len = BTRFS_FID_SIZE_CONNECTABLE_ROOT;
+			type = FILEID_BTRFS_WITH_PARENT_ROOT;
+		} else {
+			len = BTRFS_FID_SIZE_CONNECTABLE;
+			type = FILEID_BTRFS_WITH_PARENT;
+		}
+	}
+
+	*max_len = len;
+	return type;
+}
+
+static struct dentry *btrfs_get_dentry(struct super_block *sb, u64 objectid,
+				       u64 root_objectid, u32 generation)
+{
+	struct btrfs_root *root;
+	struct inode *inode;
+	struct btrfs_key key;
+
+	key.objectid = root_objectid;
+	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+	key.offset = (u64)-1;
+
+	root = btrfs_read_fs_root_no_name(btrfs_sb(sb)->fs_info, &key);
+	if (IS_ERR(root))
+		return ERR_CAST(root);
+
+	key.objectid = objectid;
+	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
+	key.offset = 0;
+
+	inode = btrfs_iget(sb, &key, root, NULL);
+	if (IS_ERR(inode))
+		return (void *)inode;
+
+	if (generation != inode->i_generation) {
+		iput(inode);
+		return ERR_PTR(-ESTALE);
+	}
+
+	return d_obtain_alias(inode);
+}
+
+static struct dentry *btrfs_fh_to_parent(struct super_block *sb, struct fid *fh,
+					 int fh_len, int fh_type)
+{
+	struct btrfs_fid *fid = (struct btrfs_fid *) fh;
+	u64 objectid, root_objectid;
+	u32 generation;
+
+	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
+		if (fh_len !=  BTRFS_FID_SIZE_CONNECTABLE)
+			return NULL;
+		root_objectid = fid->root_objectid;
+	} else if (fh_type == FILEID_BTRFS_WITH_PARENT_ROOT) {
+		if (fh_len != BTRFS_FID_SIZE_CONNECTABLE_ROOT)
+			return NULL;
+		root_objectid = fid->parent_root_objectid;
+	} else
+		return NULL;
+
+	objectid = fid->parent_objectid;
+	generation = fid->parent_gen;
+
+	return btrfs_get_dentry(sb, objectid, root_objectid, generation);
+}
+
+static struct dentry *btrfs_fh_to_dentry(struct super_block *sb, struct fid *fh,
+					 int fh_len, int fh_type)
+{
+	struct btrfs_fid *fid = (struct btrfs_fid *) fh;
+	u64 objectid, root_objectid;
+	u32 generation;
+
+	if ((fh_type != FILEID_BTRFS_WITH_PARENT ||
+	     fh_len != BTRFS_FID_SIZE_CONNECTABLE) &&
+	    (fh_type != FILEID_BTRFS_WITH_PARENT_ROOT ||
+	     fh_len != BTRFS_FID_SIZE_CONNECTABLE_ROOT) &&
+	    (fh_type != FILEID_BTRFS_WITHOUT_PARENT ||
+	     fh_len != BTRFS_FID_SIZE_NON_CONNECTABLE))
+		return NULL;
+
+	objectid = fid->objectid;
+	root_objectid = fid->root_objectid;
+	generation = fid->gen;
+
+	return btrfs_get_dentry(sb, objectid, root_objectid, generation);
+}
+
+static struct dentry *btrfs_get_parent(struct dentry *child)
+{
+	struct inode *dir = child->d_inode;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct btrfs_key key;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	int slot;
+	u64 objectid;
+	int ret;
+
+	path = btrfs_alloc_path();
+
+	key.objectid = dir->i_ino;
+	btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0) {
+		/* Error */
+		btrfs_free_path(path);
+		return ERR_PTR(ret);
+	}
+	leaf = path->nodes[0];
+	slot = path->slots[0];
+	if (ret) {
+		/* btrfs_search_slot() returns the slot where we'd want to
+		   insert a backref for parent inode #0xFFFFFFFFFFFFFFFF.
+		   The _real_ backref, telling us what the parent inode
+		   _actually_ is, will be in the slot _before_ the one
+		   that btrfs_search_slot() returns. */
+		if (!slot) {
+			/* Unless there is _no_ key in the tree before... */
+			btrfs_free_path(path);
+			return ERR_PTR(-EIO);
+		}
+		slot--;
+	}
+
+	btrfs_item_key_to_cpu(leaf, &key, slot);
+	btrfs_free_path(path);
+
+	if (key.objectid != dir->i_ino || key.type != BTRFS_INODE_REF_KEY)
+		return ERR_PTR(-EINVAL);
+
+	objectid = key.offset;
+
+	/* If we are already at the root of a subvol, return the real root */
+	if (objectid == dir->i_ino)
+		return dget(dir->i_sb->s_root);
+
+	/* Build a new key for the inode item */
+	key.objectid = objectid;
+	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
+	key.offset = 0;
+
+	return d_obtain_alias(btrfs_iget(root->fs_info->sb, &key, root, NULL));
+}
+
+const struct export_operations btrfs_export_ops = {
+	.encode_fh	= btrfs_encode_fh,
+	.fh_to_dentry	= btrfs_fh_to_dentry,
+	.fh_to_parent	= btrfs_fh_to_parent,
+	.get_parent	= btrfs_get_parent,
+};
diff --git a/fs/btrfs/export.h b/fs/btrfs/export.h
new file mode 100644
index 0000000..074348a
--- /dev/null
+++ b/fs/btrfs/export.h
@@ -0,0 +1,19 @@
+#ifndef BTRFS_EXPORT_H
+#define BTRFS_EXPORT_H
+
+#include <linux/exportfs.h>
+
+extern const struct export_operations btrfs_export_ops;
+
+struct btrfs_fid {
+	u64 objectid;
+	u64 root_objectid;
+	u32 gen;
+
+	u64 parent_objectid;
+	u32 parent_gen;
+
+	u64 parent_root_objectid;
+} __attribute__ ((packed));
+
+#endif
-- 
1.6.0.2

[PATCH 33/35] Btrfs metadata disk-io routines

[Chris Mason]

Btrfs metadata lives in the page cache on a private inode.  These
routines provide helpers to allocate, read, write and dirty the
extent_buffers that are used to access all the metadata.

disk-io.c also provides helpers, for async IO submission and completion,
which are basically wrappers around the async-thread calls.

Finally the code to read and initialize the filesystem trees lives here.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
new file mode 100644
index 0000000..81a3138
--- /dev/null
+++ b/fs/btrfs/disk-io.c
@@ -0,0 +1,2343 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/version.h>
+#include <linux/fs.h>
+#include <linux/blkdev.h>
+#include <linux/scatterlist.h>
+#include <linux/swap.h>
+#include <linux/radix-tree.h>
+#include <linux/writeback.h>
+#include <linux/buffer_head.h>
+#include <linux/workqueue.h>
+#include <linux/kthread.h>
+#include <linux/freezer.h>
+#include "compat.h"
+#include "crc32c.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "btrfs_inode.h"
+#include "volumes.h"
+#include "print-tree.h"
+#include "async-thread.h"
+#include "locking.h"
+#include "ref-cache.h"
+#include "tree-log.h"
+
+static struct extent_io_ops btree_extent_io_ops;
+static void end_workqueue_fn(struct btrfs_work *work);
+
+/*
+ * end_io_wq structs are used to do processing in task context when an IO is
+ * complete.  This is used during reads to verify checksums, and it is used
+ * by writes to insert metadata for new file extents after IO is complete.
+ */
+struct end_io_wq {
+	struct bio *bio;
+	bio_end_io_t *end_io;
+	void *private;
+	struct btrfs_fs_info *info;
+	int error;
+	int metadata;
+	struct list_head list;
+	struct btrfs_work work;
+};
+
+/*
+ * async submit bios are used to offload expensive checksumming
+ * onto the worker threads.  They checksum file and metadata bios
+ * just before they are sent down the IO stack.
+ */
+struct async_submit_bio {
+	struct inode *inode;
+	struct bio *bio;
+	struct list_head list;
+	extent_submit_bio_hook_t *submit_bio_start;
+	extent_submit_bio_hook_t *submit_bio_done;
+	int rw;
+	int mirror_num;
+	unsigned long bio_flags;
+	struct btrfs_work work;
+};
+
+/*
+ * extents on the btree inode are pretty simple, there's one extent
+ * that covers the entire device
+ */
+static struct extent_map *btree_get_extent(struct inode *inode,
+		struct page *page, size_t page_offset, u64 start, u64 len,
+		int create)
+{
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct extent_map *em;
+	int ret;
+
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, start, len);
+	if (em) {
+		em->bdev =
+			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
+		spin_unlock(&em_tree->lock);
+		goto out;
+	}
+	spin_unlock(&em_tree->lock);
+
+	em = alloc_extent_map(GFP_NOFS);
+	if (!em) {
+		em = ERR_PTR(-ENOMEM);
+		goto out;
+	}
+	em->start = 0;
+	em->len = (u64)-1;
+	em->block_len = (u64)-1;
+	em->block_start = 0;
+	em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
+
+	spin_lock(&em_tree->lock);
+	ret = add_extent_mapping(em_tree, em);
+	if (ret == -EEXIST) {
+		u64 failed_start = em->start;
+		u64 failed_len = em->len;
+
+		free_extent_map(em);
+		em = lookup_extent_mapping(em_tree, start, len);
+		if (em) {
+			ret = 0;
+		} else {
+			em = lookup_extent_mapping(em_tree, failed_start,
+						   failed_len);
+			ret = -EIO;
+		}
+	} else if (ret) {
+		free_extent_map(em);
+		em = NULL;
+	}
+	spin_unlock(&em_tree->lock);
+
+	if (ret)
+		em = ERR_PTR(ret);
+out:
+	return em;
+}
+
+u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
+{
+	return btrfs_crc32c(seed, data, len);
+}
+
+void btrfs_csum_final(u32 crc, char *result)
+{
+	*(__le32 *)result = ~cpu_to_le32(crc);
+}
+
+/*
+ * compute the csum for a btree block, and either verify it or write it
+ * into the csum field of the block.
+ */
+static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
+			   int verify)
+{
+	u16 csum_size =
+		btrfs_super_csum_size(&root->fs_info->super_copy);
+	char *result = NULL;
+	unsigned long len;
+	unsigned long cur_len;
+	unsigned long offset = BTRFS_CSUM_SIZE;
+	char *map_token = NULL;
+	char *kaddr;
+	unsigned long map_start;
+	unsigned long map_len;
+	int err;
+	u32 crc = ~(u32)0;
+	unsigned long inline_result;
+
+	len = buf->len - offset;
+	while (len > 0) {
+		err = map_private_extent_buffer(buf, offset, 32,
+					&map_token, &kaddr,
+					&map_start, &map_len, KM_USER0);
+		if (err)
+			return 1;
+		cur_len = min(len, map_len - (offset - map_start));
+		crc = btrfs_csum_data(root, kaddr + offset - map_start,
+				      crc, cur_len);
+		len -= cur_len;
+		offset += cur_len;
+		unmap_extent_buffer(buf, map_token, KM_USER0);
+	}
+	if (csum_size > sizeof(inline_result)) {
+		result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
+		if (!result)
+			return 1;
+	} else {
+		result = (char *)&inline_result;
+	}
+
+	btrfs_csum_final(crc, result);
+
+	if (verify) {
+		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
+			u32 val;
+			u32 found = 0;
+			memcpy(&found, result, csum_size);
+
+			read_extent_buffer(buf, &val, 0, csum_size);
+			printk(KERN_INFO "btrfs: %s checksum verify failed "
+			       "on %llu wanted %X found %X level %d\n",
+			       root->fs_info->sb->s_id,
+			       buf->start, val, found, btrfs_header_level(buf));
+			if (result != (char *)&inline_result)
+				kfree(result);
+			return 1;
+		}
+	} else {
+		write_extent_buffer(buf, result, 0, csum_size);
+	}
+	if (result != (char *)&inline_result)
+		kfree(result);
+	return 0;
+}
+
+/*
+ * we can't consider a given block up to date unless the transid of the
+ * block matches the transid in the parent node's pointer.  This is how we
+ * detect blocks that either didn't get written at all or got written
+ * in the wrong place.
+ */
+static int verify_parent_transid(struct extent_io_tree *io_tree,
+				 struct extent_buffer *eb, u64 parent_transid)
+{
+	int ret;
+
+	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
+		return 0;
+
+	lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
+	if (extent_buffer_uptodate(io_tree, eb) &&
+	    btrfs_header_generation(eb) == parent_transid) {
+		ret = 0;
+		goto out;
+	}
+	printk("parent transid verify failed on %llu wanted %llu found %llu\n",
+	       (unsigned long long)eb->start,
+	       (unsigned long long)parent_transid,
+	       (unsigned long long)btrfs_header_generation(eb));
+	ret = 1;
+	clear_extent_buffer_uptodate(io_tree, eb);
+out:
+	unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
+		      GFP_NOFS);
+	return ret;
+}
+
+/*
+ * helper to read a given tree block, doing retries as required when
+ * the checksums don't match and we have alternate mirrors to try.
+ */
+static int btree_read_extent_buffer_pages(struct btrfs_root *root,
+					  struct extent_buffer *eb,
+					  u64 start, u64 parent_transid)
+{
+	struct extent_io_tree *io_tree;
+	int ret;
+	int num_copies = 0;
+	int mirror_num = 0;
+
+	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
+	while (1) {
+		ret = read_extent_buffer_pages(io_tree, eb, start, 1,
+					       btree_get_extent, mirror_num);
+		if (!ret &&
+		    !verify_parent_transid(io_tree, eb, parent_transid))
+			return ret;
+
+		num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
+					      eb->start, eb->len);
+		if (num_copies == 1)
+			return ret;
+
+		mirror_num++;
+		if (mirror_num > num_copies)
+			return ret;
+	}
+	return -EIO;
+}
+
+/*
+ * checksum a dirty tree block before IO.  This has extra checks to make sure
+ * we only fill in the checksum field in the first page of a multi-page block
+ */
+
+static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
+{
+	struct extent_io_tree *tree;
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 found_start;
+	int found_level;
+	unsigned long len;
+	struct extent_buffer *eb;
+	int ret;
+
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+	if (page->private == EXTENT_PAGE_PRIVATE)
+		goto out;
+	if (!page->private)
+		goto out;
+	len = page->private >> 2;
+	WARN_ON(len == 0);
+
+	eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
+	ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
+					     btrfs_header_generation(eb));
+	BUG_ON(ret);
+	found_start = btrfs_header_bytenr(eb);
+	if (found_start != start) {
+		WARN_ON(1);
+		goto err;
+	}
+	if (eb->first_page != page) {
+		WARN_ON(1);
+		goto err;
+	}
+	if (!PageUptodate(page)) {
+		WARN_ON(1);
+		goto err;
+	}
+	found_level = btrfs_header_level(eb);
+
+	csum_tree_block(root, eb, 0);
+err:
+	free_extent_buffer(eb);
+out:
+	return 0;
+}
+
+static int check_tree_block_fsid(struct btrfs_root *root,
+				 struct extent_buffer *eb)
+{
+	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
+	u8 fsid[BTRFS_UUID_SIZE];
+	int ret = 1;
+
+	read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
+			   BTRFS_FSID_SIZE);
+	while (fs_devices) {
+		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
+			ret = 0;
+			break;
+		}
+		fs_devices = fs_devices->seed;
+	}
+	return ret;
+}
+
+static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
+			       struct extent_state *state)
+{
+	struct extent_io_tree *tree;
+	u64 found_start;
+	int found_level;
+	unsigned long len;
+	struct extent_buffer *eb;
+	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
+	int ret = 0;
+
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	if (page->private == EXTENT_PAGE_PRIVATE)
+		goto out;
+	if (!page->private)
+		goto out;
+
+	len = page->private >> 2;
+	WARN_ON(len == 0);
+
+	eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
+
+	found_start = btrfs_header_bytenr(eb);
+	if (found_start != start) {
+		printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
+		       (unsigned long long)found_start,
+		       (unsigned long long)eb->start);
+		ret = -EIO;
+		goto err;
+	}
+	if (eb->first_page != page) {
+		printk(KERN_INFO "btrfs bad first page %lu %lu\n",
+		       eb->first_page->index, page->index);
+		WARN_ON(1);
+		ret = -EIO;
+		goto err;
+	}
+	if (check_tree_block_fsid(root, eb)) {
+		printk(KERN_INFO "btrfs bad fsid on block %llu\n",
+		       (unsigned long long)eb->start);
+		ret = -EIO;
+		goto err;
+	}
+	found_level = btrfs_header_level(eb);
+
+	ret = csum_tree_block(root, eb, 1);
+	if (ret)
+		ret = -EIO;
+
+	end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
+	end = eb->start + end - 1;
+err:
+	free_extent_buffer(eb);
+out:
+	return ret;
+}
+
+static void end_workqueue_bio(struct bio *bio, int err)
+{
+	struct end_io_wq *end_io_wq = bio->bi_private;
+	struct btrfs_fs_info *fs_info;
+
+	fs_info = end_io_wq->info;
+	end_io_wq->error = err;
+	end_io_wq->work.func = end_workqueue_fn;
+	end_io_wq->work.flags = 0;
+
+	if (bio->bi_rw & (1 << BIO_RW)) {
+		if (end_io_wq->metadata)
+			btrfs_queue_worker(&fs_info->endio_meta_write_workers,
+					   &end_io_wq->work);
+		else
+			btrfs_queue_worker(&fs_info->endio_write_workers,
+					   &end_io_wq->work);
+	} else {
+		if (end_io_wq->metadata)
+			btrfs_queue_worker(&fs_info->endio_meta_workers,
+					   &end_io_wq->work);
+		else
+			btrfs_queue_worker(&fs_info->endio_workers,
+					   &end_io_wq->work);
+	}
+}
+
+int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
+			int metadata)
+{
+	struct end_io_wq *end_io_wq;
+	end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
+	if (!end_io_wq)
+		return -ENOMEM;
+
+	end_io_wq->private = bio->bi_private;
+	end_io_wq->end_io = bio->bi_end_io;
+	end_io_wq->info = info;
+	end_io_wq->error = 0;
+	end_io_wq->bio = bio;
+	end_io_wq->metadata = metadata;
+
+	bio->bi_private = end_io_wq;
+	bio->bi_end_io = end_workqueue_bio;
+	return 0;
+}
+
+unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
+{
+	unsigned long limit = min_t(unsigned long,
+				    info->workers.max_workers,
+				    info->fs_devices->open_devices);
+	return 256 * limit;
+}
+
+int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
+{
+	return atomic_read(&info->nr_async_bios) >
+		btrfs_async_submit_limit(info);
+}
+
+static void run_one_async_start(struct btrfs_work *work)
+{
+	struct btrfs_fs_info *fs_info;
+	struct async_submit_bio *async;
+
+	async = container_of(work, struct  async_submit_bio, work);
+	fs_info = BTRFS_I(async->inode)->root->fs_info;
+	async->submit_bio_start(async->inode, async->rw, async->bio,
+			       async->mirror_num, async->bio_flags);
+}
+
+static void run_one_async_done(struct btrfs_work *work)
+{
+	struct btrfs_fs_info *fs_info;
+	struct async_submit_bio *async;
+	int limit;
+
+	async = container_of(work, struct  async_submit_bio, work);
+	fs_info = BTRFS_I(async->inode)->root->fs_info;
+
+	limit = btrfs_async_submit_limit(fs_info);
+	limit = limit * 2 / 3;
+
+	atomic_dec(&fs_info->nr_async_submits);
+
+	if (atomic_read(&fs_info->nr_async_submits) < limit &&
+	    waitqueue_active(&fs_info->async_submit_wait))
+		wake_up(&fs_info->async_submit_wait);
+
+	async->submit_bio_done(async->inode, async->rw, async->bio,
+			       async->mirror_num, async->bio_flags);
+}
+
+static void run_one_async_free(struct btrfs_work *work)
+{
+	struct async_submit_bio *async;
+
+	async = container_of(work, struct  async_submit_bio, work);
+	kfree(async);
+}
+
+int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
+			int rw, struct bio *bio, int mirror_num,
+			unsigned long bio_flags,
+			extent_submit_bio_hook_t *submit_bio_start,
+			extent_submit_bio_hook_t *submit_bio_done)
+{
+	struct async_submit_bio *async;
+
+	async = kmalloc(sizeof(*async), GFP_NOFS);
+	if (!async)
+		return -ENOMEM;
+
+	async->inode = inode;
+	async->rw = rw;
+	async->bio = bio;
+	async->mirror_num = mirror_num;
+	async->submit_bio_start = submit_bio_start;
+	async->submit_bio_done = submit_bio_done;
+
+	async->work.func = run_one_async_start;
+	async->work.ordered_func = run_one_async_done;
+	async->work.ordered_free = run_one_async_free;
+
+	async->work.flags = 0;
+	async->bio_flags = bio_flags;
+
+	atomic_inc(&fs_info->nr_async_submits);
+	btrfs_queue_worker(&fs_info->workers, &async->work);
+#if 0
+	int limit = btrfs_async_submit_limit(fs_info);
+	if (atomic_read(&fs_info->nr_async_submits) > limit) {
+		wait_event_timeout(fs_info->async_submit_wait,
+			   (atomic_read(&fs_info->nr_async_submits) < limit),
+			   HZ/10);
+
+		wait_event_timeout(fs_info->async_submit_wait,
+			   (atomic_read(&fs_info->nr_async_bios) < limit),
+			   HZ/10);
+	}
+#endif
+	while (atomic_read(&fs_info->async_submit_draining) &&
+	      atomic_read(&fs_info->nr_async_submits)) {
+		wait_event(fs_info->async_submit_wait,
+			   (atomic_read(&fs_info->nr_async_submits) == 0));
+	}
+
+	return 0;
+}
+
+static int btree_csum_one_bio(struct bio *bio)
+{
+	struct bio_vec *bvec = bio->bi_io_vec;
+	int bio_index = 0;
+	struct btrfs_root *root;
+
+	WARN_ON(bio->bi_vcnt <= 0);
+	while (bio_index < bio->bi_vcnt) {
+		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
+		csum_dirty_buffer(root, bvec->bv_page);
+		bio_index++;
+		bvec++;
+	}
+	return 0;
+}
+
+static int __btree_submit_bio_start(struct inode *inode, int rw,
+				    struct bio *bio, int mirror_num,
+				    unsigned long bio_flags)
+{
+	/*
+	 * when we're called for a write, we're already in the async
+	 * submission context.  Just jump into btrfs_map_bio
+	 */
+	btree_csum_one_bio(bio);
+	return 0;
+}
+
+static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
+				 int mirror_num, unsigned long bio_flags)
+{
+	/*
+	 * when we're called for a write, we're already in the async
+	 * submission context.  Just jump into btrfs_map_bio
+	 */
+	return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
+}
+
+static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
+				 int mirror_num, unsigned long bio_flags)
+{
+	int ret;
+
+	ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
+					  bio, 1);
+	BUG_ON(ret);
+
+	if (!(rw & (1 << BIO_RW))) {
+		/*
+		 * called for a read, do the setup so that checksum validation
+		 * can happen in the async kernel threads
+		 */
+		return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
+				     mirror_num, 0);
+	}
+	/*
+	 * kthread helpers are used to submit writes so that checksumming
+	 * can happen in parallel across all CPUs
+	 */
+	return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
+				   inode, rw, bio, mirror_num, 0,
+				   __btree_submit_bio_start,
+				   __btree_submit_bio_done);
+}
+
+static int btree_writepage(struct page *page, struct writeback_control *wbc)
+{
+	struct extent_io_tree *tree;
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+	if (current->flags & PF_MEMALLOC) {
+		redirty_page_for_writepage(wbc, page);
+		unlock_page(page);
+		return 0;
+	}
+	return extent_write_full_page(tree, page, btree_get_extent, wbc);
+}
+
+static int btree_writepages(struct address_space *mapping,
+			    struct writeback_control *wbc)
+{
+	struct extent_io_tree *tree;
+	tree = &BTRFS_I(mapping->host)->io_tree;
+	if (wbc->sync_mode == WB_SYNC_NONE) {
+		u64 num_dirty;
+		u64 start = 0;
+		unsigned long thresh = 32 * 1024 * 1024;
+
+		if (wbc->for_kupdate)
+			return 0;
+
+		num_dirty = count_range_bits(tree, &start, (u64)-1,
+					     thresh, EXTENT_DIRTY);
+		if (num_dirty < thresh)
+			return 0;
+	}
+	return extent_writepages(tree, mapping, btree_get_extent, wbc);
+}
+
+static int btree_readpage(struct file *file, struct page *page)
+{
+	struct extent_io_tree *tree;
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	return extent_read_full_page(tree, page, btree_get_extent);
+}
+
+static int btree_releasepage(struct page *page, gfp_t gfp_flags)
+{
+	struct extent_io_tree *tree;
+	struct extent_map_tree *map;
+	int ret;
+
+	if (PageWriteback(page) || PageDirty(page))
+		return 0;
+
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	map = &BTRFS_I(page->mapping->host)->extent_tree;
+
+	ret = try_release_extent_state(map, tree, page, gfp_flags);
+	if (!ret)
+		return 0;
+
+	ret = try_release_extent_buffer(tree, page);
+	if (ret == 1) {
+		ClearPagePrivate(page);
+		set_page_private(page, 0);
+		page_cache_release(page);
+	}
+
+	return ret;
+}
+
+static void btree_invalidatepage(struct page *page, unsigned long offset)
+{
+	struct extent_io_tree *tree;
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	extent_invalidatepage(tree, page, offset);
+	btree_releasepage(page, GFP_NOFS);
+	if (PagePrivate(page)) {
+		printk(KERN_WARNING "btrfs warning page private not zero "
+		       "on page %llu\n", (unsigned long long)page_offset(page));
+		ClearPagePrivate(page);
+		set_page_private(page, 0);
+		page_cache_release(page);
+	}
+}
+
+#if 0
+static int btree_writepage(struct page *page, struct writeback_control *wbc)
+{
+	struct buffer_head *bh;
+	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
+	struct buffer_head *head;
+	if (!page_has_buffers(page)) {
+		create_empty_buffers(page, root->fs_info->sb->s_blocksize,
+					(1 << BH_Dirty)|(1 << BH_Uptodate));
+	}
+	head = page_buffers(page);
+	bh = head;
+	do {
+		if (buffer_dirty(bh))
+			csum_tree_block(root, bh, 0);
+		bh = bh->b_this_page;
+	} while (bh != head);
+	return block_write_full_page(page, btree_get_block, wbc);
+}
+#endif
+
+static struct address_space_operations btree_aops = {
+	.readpage	= btree_readpage,
+	.writepage	= btree_writepage,
+	.writepages	= btree_writepages,
+	.releasepage	= btree_releasepage,
+	.invalidatepage = btree_invalidatepage,
+	.sync_page	= block_sync_page,
+};
+
+int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
+			 u64 parent_transid)
+{
+	struct extent_buffer *buf = NULL;
+	struct inode *btree_inode = root->fs_info->btree_inode;
+	int ret = 0;
+
+	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
+	if (!buf)
+		return 0;
+	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
+				 buf, 0, 0, btree_get_extent, 0);
+	free_extent_buffer(buf);
+	return ret;
+}
+
+struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
+					    u64 bytenr, u32 blocksize)
+{
+	struct inode *btree_inode = root->fs_info->btree_inode;
+	struct extent_buffer *eb;
+	eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
+				bytenr, blocksize, GFP_NOFS);
+	return eb;
+}
+
+struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
+						 u64 bytenr, u32 blocksize)
+{
+	struct inode *btree_inode = root->fs_info->btree_inode;
+	struct extent_buffer *eb;
+
+	eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
+				 bytenr, blocksize, NULL, GFP_NOFS);
+	return eb;
+}
+
+
+int btrfs_write_tree_block(struct extent_buffer *buf)
+{
+	return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
+				      buf->start + buf->len - 1, WB_SYNC_ALL);
+}
+
+int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
+{
+	return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
+				  buf->start, buf->start + buf->len - 1);
+}
+
+struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
+				      u32 blocksize, u64 parent_transid)
+{
+	struct extent_buffer *buf = NULL;
+	struct inode *btree_inode = root->fs_info->btree_inode;
+	struct extent_io_tree *io_tree;
+	int ret;
+
+	io_tree = &BTRFS_I(btree_inode)->io_tree;
+
+	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
+	if (!buf)
+		return NULL;
+
+	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
+
+	if (ret == 0)
+		buf->flags |= EXTENT_UPTODATE;
+	else
+		WARN_ON(1);
+	return buf;
+
+}
+
+int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		     struct extent_buffer *buf)
+{
+	struct inode *btree_inode = root->fs_info->btree_inode;
+	if (btrfs_header_generation(buf) ==
+	    root->fs_info->running_transaction->transid) {
+		WARN_ON(!btrfs_tree_locked(buf));
+		clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
+					  buf);
+	}
+	return 0;
+}
+
+static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
+			u32 stripesize, struct btrfs_root *root,
+			struct btrfs_fs_info *fs_info,
+			u64 objectid)
+{
+	root->node = NULL;
+	root->commit_root = NULL;
+	root->ref_tree = NULL;
+	root->sectorsize = sectorsize;
+	root->nodesize = nodesize;
+	root->leafsize = leafsize;
+	root->stripesize = stripesize;
+	root->ref_cows = 0;
+	root->track_dirty = 0;
+
+	root->fs_info = fs_info;
+	root->objectid = objectid;
+	root->last_trans = 0;
+	root->highest_inode = 0;
+	root->last_inode_alloc = 0;
+	root->name = NULL;
+	root->in_sysfs = 0;
+
+	INIT_LIST_HEAD(&root->dirty_list);
+	INIT_LIST_HEAD(&root->orphan_list);
+	INIT_LIST_HEAD(&root->dead_list);
+	spin_lock_init(&root->node_lock);
+	spin_lock_init(&root->list_lock);
+	mutex_init(&root->objectid_mutex);
+	mutex_init(&root->log_mutex);
+	extent_io_tree_init(&root->dirty_log_pages,
+			     fs_info->btree_inode->i_mapping, GFP_NOFS);
+
+	btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
+	root->ref_tree = &root->ref_tree_struct;
+
+	memset(&root->root_key, 0, sizeof(root->root_key));
+	memset(&root->root_item, 0, sizeof(root->root_item));
+	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
+	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
+	root->defrag_trans_start = fs_info->generation;
+	init_completion(&root->kobj_unregister);
+	root->defrag_running = 0;
+	root->defrag_level = 0;
+	root->root_key.objectid = objectid;
+	root->anon_super.s_root = NULL;
+	root->anon_super.s_dev = 0;
+	INIT_LIST_HEAD(&root->anon_super.s_list);
+	INIT_LIST_HEAD(&root->anon_super.s_instances);
+	init_rwsem(&root->anon_super.s_umount);
+
+	return 0;
+}
+
+static int find_and_setup_root(struct btrfs_root *tree_root,
+			       struct btrfs_fs_info *fs_info,
+			       u64 objectid,
+			       struct btrfs_root *root)
+{
+	int ret;
+	u32 blocksize;
+	u64 generation;
+
+	__setup_root(tree_root->nodesize, tree_root->leafsize,
+		     tree_root->sectorsize, tree_root->stripesize,
+		     root, fs_info, objectid);
+	ret = btrfs_find_last_root(tree_root, objectid,
+				   &root->root_item, &root->root_key);
+	BUG_ON(ret);
+
+	generation = btrfs_root_generation(&root->root_item);
+	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
+	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
+				     blocksize, generation);
+	BUG_ON(!root->node);
+	return 0;
+}
+
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info)
+{
+	struct extent_buffer *eb;
+	struct btrfs_root *log_root_tree = fs_info->log_root_tree;
+	u64 start = 0;
+	u64 end = 0;
+	int ret;
+
+	if (!log_root_tree)
+		return 0;
+
+	while (1) {
+		ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
+				    0, &start, &end, EXTENT_DIRTY);
+		if (ret)
+			break;
+
+		clear_extent_dirty(&log_root_tree->dirty_log_pages,
+				   start, end, GFP_NOFS);
+	}
+	eb = fs_info->log_root_tree->node;
+
+	WARN_ON(btrfs_header_level(eb) != 0);
+	WARN_ON(btrfs_header_nritems(eb) != 0);
+
+	ret = btrfs_free_reserved_extent(fs_info->tree_root,
+				eb->start, eb->len);
+	BUG_ON(ret);
+
+	free_extent_buffer(eb);
+	kfree(fs_info->log_root_tree);
+	fs_info->log_root_tree = NULL;
+	return 0;
+}
+
+int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_root *root;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+
+	root = kzalloc(sizeof(*root), GFP_NOFS);
+	if (!root)
+		return -ENOMEM;
+
+	__setup_root(tree_root->nodesize, tree_root->leafsize,
+		     tree_root->sectorsize, tree_root->stripesize,
+		     root, fs_info, BTRFS_TREE_LOG_OBJECTID);
+
+	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
+	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
+	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
+	root->ref_cows = 0;
+
+	root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
+					    0, BTRFS_TREE_LOG_OBJECTID,
+					    trans->transid, 0, 0, 0);
+
+	btrfs_set_header_nritems(root->node, 0);
+	btrfs_set_header_level(root->node, 0);
+	btrfs_set_header_bytenr(root->node, root->node->start);
+	btrfs_set_header_generation(root->node, trans->transid);
+	btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
+
+	write_extent_buffer(root->node, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(root->node),
+			    BTRFS_FSID_SIZE);
+	btrfs_mark_buffer_dirty(root->node);
+	btrfs_tree_unlock(root->node);
+	fs_info->log_root_tree = root;
+	return 0;
+}
+
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
+					       struct btrfs_key *location)
+{
+	struct btrfs_root *root;
+	struct btrfs_fs_info *fs_info = tree_root->fs_info;
+	struct btrfs_path *path;
+	struct extent_buffer *l;
+	u64 highest_inode;
+	u64 generation;
+	u32 blocksize;
+	int ret = 0;
+
+	root = kzalloc(sizeof(*root), GFP_NOFS);
+	if (!root)
+		return ERR_PTR(-ENOMEM);
+	if (location->offset == (u64)-1) {
+		ret = find_and_setup_root(tree_root, fs_info,
+					  location->objectid, root);
+		if (ret) {
+			kfree(root);
+			return ERR_PTR(ret);
+		}
+		goto insert;
+	}
+
+	__setup_root(tree_root->nodesize, tree_root->leafsize,
+		     tree_root->sectorsize, tree_root->stripesize,
+		     root, fs_info, location->objectid);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
+	if (ret != 0) {
+		if (ret > 0)
+			ret = -ENOENT;
+		goto out;
+	}
+	l = path->nodes[0];
+	read_extent_buffer(l, &root->root_item,
+	       btrfs_item_ptr_offset(l, path->slots[0]),
+	       sizeof(root->root_item));
+	memcpy(&root->root_key, location, sizeof(*location));
+	ret = 0;
+out:
+	btrfs_release_path(root, path);
+	btrfs_free_path(path);
+	if (ret) {
+		kfree(root);
+		return ERR_PTR(ret);
+	}
+	generation = btrfs_root_generation(&root->root_item);
+	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
+	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
+				     blocksize, generation);
+	BUG_ON(!root->node);
+insert:
+	if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
+		root->ref_cows = 1;
+		ret = btrfs_find_highest_inode(root, &highest_inode);
+		if (ret == 0) {
+			root->highest_inode = highest_inode;
+			root->last_inode_alloc = highest_inode;
+		}
+	}
+	return root;
+}
+
+struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
+					u64 root_objectid)
+{
+	struct btrfs_root *root;
+
+	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
+		return fs_info->tree_root;
+	if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
+		return fs_info->extent_root;
+
+	root = radix_tree_lookup(&fs_info->fs_roots_radix,
+				 (unsigned long)root_objectid);
+	return root;
+}
+
+struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
+					      struct btrfs_key *location)
+{
+	struct btrfs_root *root;
+	int ret;
+
+	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
+		return fs_info->tree_root;
+	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
+		return fs_info->extent_root;
+	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
+		return fs_info->chunk_root;
+	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
+		return fs_info->dev_root;
+	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
+		return fs_info->csum_root;
+
+	root = radix_tree_lookup(&fs_info->fs_roots_radix,
+				 (unsigned long)location->objectid);
+	if (root)
+		return root;
+
+	root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
+	if (IS_ERR(root))
+		return root;
+
+	set_anon_super(&root->anon_super, NULL);
+
+	ret = radix_tree_insert(&fs_info->fs_roots_radix,
+				(unsigned long)root->root_key.objectid,
+				root);
+	if (ret) {
+		free_extent_buffer(root->node);
+		kfree(root);
+		return ERR_PTR(ret);
+	}
+	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
+		ret = btrfs_find_dead_roots(fs_info->tree_root,
+					    root->root_key.objectid, root);
+		BUG_ON(ret);
+		btrfs_orphan_cleanup(root);
+	}
+	return root;
+}
+
+struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
+				      struct btrfs_key *location,
+				      const char *name, int namelen)
+{
+	struct btrfs_root *root;
+	int ret;
+
+	root = btrfs_read_fs_root_no_name(fs_info, location);
+	if (!root)
+		return NULL;
+
+	if (root->in_sysfs)
+		return root;
+
+	ret = btrfs_set_root_name(root, name, namelen);
+	if (ret) {
+		free_extent_buffer(root->node);
+		kfree(root);
+		return ERR_PTR(ret);
+	}
+#if 0
+	ret = btrfs_sysfs_add_root(root);
+	if (ret) {
+		free_extent_buffer(root->node);
+		kfree(root->name);
+		kfree(root);
+		return ERR_PTR(ret);
+	}
+#endif
+	root->in_sysfs = 1;
+	return root;
+}
+
+static int btrfs_congested_fn(void *congested_data, int bdi_bits)
+{
+	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
+	int ret = 0;
+	struct list_head *cur;
+	struct btrfs_device *device;
+	struct backing_dev_info *bdi;
+#if 0
+	if ((bdi_bits & (1 << BDI_write_congested)) &&
+	    btrfs_congested_async(info, 0))
+		return 1;
+#endif
+	list_for_each(cur, &info->fs_devices->devices) {
+		device = list_entry(cur, struct btrfs_device, dev_list);
+		if (!device->bdev)
+			continue;
+		bdi = blk_get_backing_dev_info(device->bdev);
+		if (bdi && bdi_congested(bdi, bdi_bits)) {
+			ret = 1;
+			break;
+		}
+	}
+	return ret;
+}
+
+/*
+ * this unplugs every device on the box, and it is only used when page
+ * is null
+ */
+static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
+{
+	struct list_head *cur;
+	struct btrfs_device *device;
+	struct btrfs_fs_info *info;
+
+	info = (struct btrfs_fs_info *)bdi->unplug_io_data;
+	list_for_each(cur, &info->fs_devices->devices) {
+		device = list_entry(cur, struct btrfs_device, dev_list);
+		if (!device->bdev)
+			continue;
+
+		bdi = blk_get_backing_dev_info(device->bdev);
+		if (bdi->unplug_io_fn)
+			bdi->unplug_io_fn(bdi, page);
+	}
+}
+
+static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
+{
+	struct inode *inode;
+	struct extent_map_tree *em_tree;
+	struct extent_map *em;
+	struct address_space *mapping;
+	u64 offset;
+
+	/* the generic O_DIRECT read code does this */
+	if (1 || !page) {
+		__unplug_io_fn(bdi, page);
+		return;
+	}
+
+	/*
+	 * page->mapping may change at any time.  Get a consistent copy
+	 * and use that for everything below
+	 */
+	smp_mb();
+	mapping = page->mapping;
+	if (!mapping)
+		return;
+
+	inode = mapping->host;
+
+	/*
+	 * don't do the expensive searching for a small number of
+	 * devices
+	 */
+	if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
+		__unplug_io_fn(bdi, page);
+		return;
+	}
+
+	offset = page_offset(page);
+
+	em_tree = &BTRFS_I(inode)->extent_tree;
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
+	spin_unlock(&em_tree->lock);
+	if (!em) {
+		__unplug_io_fn(bdi, page);
+		return;
+	}
+
+	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
+		free_extent_map(em);
+		__unplug_io_fn(bdi, page);
+		return;
+	}
+	offset = offset - em->start;
+	btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
+			  em->block_start + offset, page);
+	free_extent_map(em);
+}
+
+static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
+{
+	bdi_init(bdi);
+	bdi->ra_pages	= default_backing_dev_info.ra_pages;
+	bdi->state		= 0;
+	bdi->capabilities	= default_backing_dev_info.capabilities;
+	bdi->unplug_io_fn	= btrfs_unplug_io_fn;
+	bdi->unplug_io_data	= info;
+	bdi->congested_fn	= btrfs_congested_fn;
+	bdi->congested_data	= info;
+	return 0;
+}
+
+static int bio_ready_for_csum(struct bio *bio)
+{
+	u64 length = 0;
+	u64 buf_len = 0;
+	u64 start = 0;
+	struct page *page;
+	struct extent_io_tree *io_tree = NULL;
+	struct btrfs_fs_info *info = NULL;
+	struct bio_vec *bvec;
+	int i;
+	int ret;
+
+	bio_for_each_segment(bvec, bio, i) {
+		page = bvec->bv_page;
+		if (page->private == EXTENT_PAGE_PRIVATE) {
+			length += bvec->bv_len;
+			continue;
+		}
+		if (!page->private) {
+			length += bvec->bv_len;
+			continue;
+		}
+		length = bvec->bv_len;
+		buf_len = page->private >> 2;
+		start = page_offset(page) + bvec->bv_offset;
+		io_tree = &BTRFS_I(page->mapping->host)->io_tree;
+		info = BTRFS_I(page->mapping->host)->root->fs_info;
+	}
+	/* are we fully contained in this bio? */
+	if (buf_len <= length)
+		return 1;
+
+	ret = extent_range_uptodate(io_tree, start + length,
+				    start + buf_len - 1);
+	if (ret == 1)
+		return ret;
+	return ret;
+}
+
+/*
+ * called by the kthread helper functions to finally call the bio end_io
+ * functions.  This is where read checksum verification actually happens
+ */
+static void end_workqueue_fn(struct btrfs_work *work)
+{
+	struct bio *bio;
+	struct end_io_wq *end_io_wq;
+	struct btrfs_fs_info *fs_info;
+	int error;
+
+	end_io_wq = container_of(work, struct end_io_wq, work);
+	bio = end_io_wq->bio;
+	fs_info = end_io_wq->info;
+
+	/* metadata bio reads are special because the whole tree block must
+	 * be checksummed at once.  This makes sure the entire block is in
+	 * ram and up to date before trying to verify things.  For
+	 * blocksize <= pagesize, it is basically a noop
+	 */
+	if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
+	    !bio_ready_for_csum(bio)) {
+		btrfs_queue_worker(&fs_info->endio_meta_workers,
+				   &end_io_wq->work);
+		return;
+	}
+	error = end_io_wq->error;
+	bio->bi_private = end_io_wq->private;
+	bio->bi_end_io = end_io_wq->end_io;
+	kfree(end_io_wq);
+	bio_endio(bio, error);
+}
+
+static int cleaner_kthread(void *arg)
+{
+	struct btrfs_root *root = arg;
+
+	do {
+		smp_mb();
+		if (root->fs_info->closing)
+			break;
+
+		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
+		mutex_lock(&root->fs_info->cleaner_mutex);
+		btrfs_clean_old_snapshots(root);
+		mutex_unlock(&root->fs_info->cleaner_mutex);
+
+		if (freezing(current)) {
+			refrigerator();
+		} else {
+			smp_mb();
+			if (root->fs_info->closing)
+				break;
+			set_current_state(TASK_INTERRUPTIBLE);
+			schedule();
+			__set_current_state(TASK_RUNNING);
+		}
+	} while (!kthread_should_stop());
+	return 0;
+}
+
+static int transaction_kthread(void *arg)
+{
+	struct btrfs_root *root = arg;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_transaction *cur;
+	unsigned long now;
+	unsigned long delay;
+	int ret;
+
+	do {
+		smp_mb();
+		if (root->fs_info->closing)
+			break;
+
+		delay = HZ * 30;
+		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
+		mutex_lock(&root->fs_info->transaction_kthread_mutex);
+
+		if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
+			printk(KERN_INFO "btrfs: total reference cache "
+			       "size %llu\n",
+			       root->fs_info->total_ref_cache_size);
+		}
+
+		mutex_lock(&root->fs_info->trans_mutex);
+		cur = root->fs_info->running_transaction;
+		if (!cur) {
+			mutex_unlock(&root->fs_info->trans_mutex);
+			goto sleep;
+		}
+
+		now = get_seconds();
+		if (now < cur->start_time || now - cur->start_time < 30) {
+			mutex_unlock(&root->fs_info->trans_mutex);
+			delay = HZ * 5;
+			goto sleep;
+		}
+		mutex_unlock(&root->fs_info->trans_mutex);
+		trans = btrfs_start_transaction(root, 1);
+		ret = btrfs_commit_transaction(trans, root);
+sleep:
+		wake_up_process(root->fs_info->cleaner_kthread);
+		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
+
+		if (freezing(current)) {
+			refrigerator();
+		} else {
+			if (root->fs_info->closing)
+				break;
+			set_current_state(TASK_INTERRUPTIBLE);
+			schedule_timeout(delay);
+			__set_current_state(TASK_RUNNING);
+		}
+	} while (!kthread_should_stop());
+	return 0;
+}
+
+struct btrfs_root *open_ctree(struct super_block *sb,
+			      struct btrfs_fs_devices *fs_devices,
+			      char *options)
+{
+	u32 sectorsize;
+	u32 nodesize;
+	u32 leafsize;
+	u32 blocksize;
+	u32 stripesize;
+	u64 generation;
+	u64 features;
+	struct btrfs_key location;
+	struct buffer_head *bh;
+	struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
+						 GFP_NOFS);
+	struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
+						 GFP_NOFS);
+	struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
+					       GFP_NOFS);
+	struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
+						GFP_NOFS);
+	struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
+						GFP_NOFS);
+	struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
+					      GFP_NOFS);
+	struct btrfs_root *log_tree_root;
+
+	int ret;
+	int err = -EINVAL;
+
+	struct btrfs_super_block *disk_super;
+
+	if (!extent_root || !tree_root || !fs_info ||
+	    !chunk_root || !dev_root || !csum_root) {
+		err = -ENOMEM;
+		goto fail;
+	}
+	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
+	INIT_LIST_HEAD(&fs_info->trans_list);
+	INIT_LIST_HEAD(&fs_info->dead_roots);
+	INIT_LIST_HEAD(&fs_info->hashers);
+	INIT_LIST_HEAD(&fs_info->delalloc_inodes);
+	spin_lock_init(&fs_info->hash_lock);
+	spin_lock_init(&fs_info->delalloc_lock);
+	spin_lock_init(&fs_info->new_trans_lock);
+	spin_lock_init(&fs_info->ref_cache_lock);
+
+	init_completion(&fs_info->kobj_unregister);
+	fs_info->tree_root = tree_root;
+	fs_info->extent_root = extent_root;
+	fs_info->csum_root = csum_root;
+	fs_info->chunk_root = chunk_root;
+	fs_info->dev_root = dev_root;
+	fs_info->fs_devices = fs_devices;
+	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
+	INIT_LIST_HEAD(&fs_info->space_info);
+	btrfs_mapping_init(&fs_info->mapping_tree);
+	atomic_set(&fs_info->nr_async_submits, 0);
+	atomic_set(&fs_info->async_delalloc_pages, 0);
+	atomic_set(&fs_info->async_submit_draining, 0);
+	atomic_set(&fs_info->nr_async_bios, 0);
+	atomic_set(&fs_info->throttles, 0);
+	atomic_set(&fs_info->throttle_gen, 0);
+	fs_info->sb = sb;
+	fs_info->max_extent = (u64)-1;
+	fs_info->max_inline = 8192 * 1024;
+	setup_bdi(fs_info, &fs_info->bdi);
+	fs_info->btree_inode = new_inode(sb);
+	fs_info->btree_inode->i_ino = 1;
+	fs_info->btree_inode->i_nlink = 1;
+
+	fs_info->thread_pool_size = min_t(unsigned long,
+					  num_online_cpus() + 2, 8);
+
+	INIT_LIST_HEAD(&fs_info->ordered_extents);
+	spin_lock_init(&fs_info->ordered_extent_lock);
+
+	sb->s_blocksize = 4096;
+	sb->s_blocksize_bits = blksize_bits(4096);
+
+	/*
+	 * we set the i_size on the btree inode to the max possible int.
+	 * the real end of the address space is determined by all of
+	 * the devices in the system
+	 */
+	fs_info->btree_inode->i_size = OFFSET_MAX;
+	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
+	fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
+
+	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
+			     fs_info->btree_inode->i_mapping,
+			     GFP_NOFS);
+	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
+			     GFP_NOFS);
+
+	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
+
+	spin_lock_init(&fs_info->block_group_cache_lock);
+	fs_info->block_group_cache_tree.rb_node = NULL;
+
+	extent_io_tree_init(&fs_info->pinned_extents,
+			     fs_info->btree_inode->i_mapping, GFP_NOFS);
+	extent_io_tree_init(&fs_info->pending_del,
+			     fs_info->btree_inode->i_mapping, GFP_NOFS);
+	extent_io_tree_init(&fs_info->extent_ins,
+			     fs_info->btree_inode->i_mapping, GFP_NOFS);
+	fs_info->do_barriers = 1;
+
+	INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
+	btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
+	btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
+
+	BTRFS_I(fs_info->btree_inode)->root = tree_root;
+	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
+	       sizeof(struct btrfs_key));
+	insert_inode_hash(fs_info->btree_inode);
+
+	mutex_init(&fs_info->trans_mutex);
+	mutex_init(&fs_info->tree_log_mutex);
+	mutex_init(&fs_info->drop_mutex);
+	mutex_init(&fs_info->extent_ins_mutex);
+	mutex_init(&fs_info->pinned_mutex);
+	mutex_init(&fs_info->chunk_mutex);
+	mutex_init(&fs_info->transaction_kthread_mutex);
+	mutex_init(&fs_info->cleaner_mutex);
+	mutex_init(&fs_info->volume_mutex);
+	mutex_init(&fs_info->tree_reloc_mutex);
+	init_waitqueue_head(&fs_info->transaction_throttle);
+	init_waitqueue_head(&fs_info->transaction_wait);
+	init_waitqueue_head(&fs_info->async_submit_wait);
+	init_waitqueue_head(&fs_info->tree_log_wait);
+	atomic_set(&fs_info->tree_log_commit, 0);
+	atomic_set(&fs_info->tree_log_writers, 0);
+	fs_info->tree_log_transid = 0;
+
+	__setup_root(4096, 4096, 4096, 4096, tree_root,
+		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
+
+
+	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
+	if (!bh)
+		goto fail_iput;
+
+	memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
+	memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
+	       sizeof(fs_info->super_for_commit));
+	brelse(bh);
+
+	memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
+
+	disk_super = &fs_info->super_copy;
+	if (!btrfs_super_root(disk_super))
+		goto fail_iput;
+
+	ret = btrfs_parse_options(tree_root, options);
+	if (ret) {
+		err = ret;
+		goto fail_iput;
+	}
+
+	features = btrfs_super_incompat_flags(disk_super) &
+		~BTRFS_FEATURE_INCOMPAT_SUPP;
+	if (features) {
+		printk(KERN_ERR "BTRFS: couldn't mount because of "
+		       "unsupported optional features (%Lx).\n",
+		       features);
+		err = -EINVAL;
+		goto fail_iput;
+	}
+
+	features = btrfs_super_compat_ro_flags(disk_super) &
+		~BTRFS_FEATURE_COMPAT_RO_SUPP;
+	if (!(sb->s_flags & MS_RDONLY) && features) {
+		printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
+		       "unsupported option features (%Lx).\n",
+		       features);
+		err = -EINVAL;
+		goto fail_iput;
+	}
+
+	/*
+	 * we need to start all the end_io workers up front because the
+	 * queue work function gets called at interrupt time, and so it
+	 * cannot dynamically grow.
+	 */
+	btrfs_init_workers(&fs_info->workers, "worker",
+			   fs_info->thread_pool_size);
+
+	btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
+			   fs_info->thread_pool_size);
+
+	btrfs_init_workers(&fs_info->submit_workers, "submit",
+			   min_t(u64, fs_devices->num_devices,
+			   fs_info->thread_pool_size));
+
+	/* a higher idle thresh on the submit workers makes it much more
+	 * likely that bios will be send down in a sane order to the
+	 * devices
+	 */
+	fs_info->submit_workers.idle_thresh = 64;
+
+	fs_info->workers.idle_thresh = 16;
+	fs_info->workers.ordered = 1;
+
+	fs_info->delalloc_workers.idle_thresh = 2;
+	fs_info->delalloc_workers.ordered = 1;
+
+	btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
+	btrfs_init_workers(&fs_info->endio_workers, "endio",
+			   fs_info->thread_pool_size);
+	btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
+			   fs_info->thread_pool_size);
+	btrfs_init_workers(&fs_info->endio_meta_write_workers,
+			   "endio-meta-write", fs_info->thread_pool_size);
+	btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
+			   fs_info->thread_pool_size);
+
+	/*
+	 * endios are largely parallel and should have a very
+	 * low idle thresh
+	 */
+	fs_info->endio_workers.idle_thresh = 4;
+	fs_info->endio_write_workers.idle_thresh = 64;
+	fs_info->endio_meta_write_workers.idle_thresh = 64;
+
+	btrfs_start_workers(&fs_info->workers, 1);
+	btrfs_start_workers(&fs_info->submit_workers, 1);
+	btrfs_start_workers(&fs_info->delalloc_workers, 1);
+	btrfs_start_workers(&fs_info->fixup_workers, 1);
+	btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
+	btrfs_start_workers(&fs_info->endio_meta_workers,
+			    fs_info->thread_pool_size);
+	btrfs_start_workers(&fs_info->endio_meta_write_workers,
+			    fs_info->thread_pool_size);
+	btrfs_start_workers(&fs_info->endio_write_workers,
+			    fs_info->thread_pool_size);
+
+	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
+	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
+				    4 * 1024 * 1024 / PAGE_CACHE_SIZE);
+
+	nodesize = btrfs_super_nodesize(disk_super);
+	leafsize = btrfs_super_leafsize(disk_super);
+	sectorsize = btrfs_super_sectorsize(disk_super);
+	stripesize = btrfs_super_stripesize(disk_super);
+	tree_root->nodesize = nodesize;
+	tree_root->leafsize = leafsize;
+	tree_root->sectorsize = sectorsize;
+	tree_root->stripesize = stripesize;
+
+	sb->s_blocksize = sectorsize;
+	sb->s_blocksize_bits = blksize_bits(sectorsize);
+
+	if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
+		    sizeof(disk_super->magic))) {
+		printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
+		goto fail_sb_buffer;
+	}
+
+	mutex_lock(&fs_info->chunk_mutex);
+	ret = btrfs_read_sys_array(tree_root);
+	mutex_unlock(&fs_info->chunk_mutex);
+	if (ret) {
+		printk(KERN_WARNING "btrfs: failed to read the system "
+		       "array on %s\n", sb->s_id);
+		goto fail_sys_array;
+	}
+
+	blocksize = btrfs_level_size(tree_root,
+				     btrfs_super_chunk_root_level(disk_super));
+	generation = btrfs_super_chunk_root_generation(disk_super);
+
+	__setup_root(nodesize, leafsize, sectorsize, stripesize,
+		     chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
+
+	chunk_root->node = read_tree_block(chunk_root,
+					   btrfs_super_chunk_root(disk_super),
+					   blocksize, generation);
+	BUG_ON(!chunk_root->node);
+
+	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
+	   (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
+	   BTRFS_UUID_SIZE);
+
+	mutex_lock(&fs_info->chunk_mutex);
+	ret = btrfs_read_chunk_tree(chunk_root);
+	mutex_unlock(&fs_info->chunk_mutex);
+	if (ret) {
+		printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
+		       sb->s_id);
+		goto fail_chunk_root;
+	}
+
+	btrfs_close_extra_devices(fs_devices);
+
+	blocksize = btrfs_level_size(tree_root,
+				     btrfs_super_root_level(disk_super));
+	generation = btrfs_super_generation(disk_super);
+
+	tree_root->node = read_tree_block(tree_root,
+					  btrfs_super_root(disk_super),
+					  blocksize, generation);
+	if (!tree_root->node)
+		goto fail_chunk_root;
+
+
+	ret = find_and_setup_root(tree_root, fs_info,
+				  BTRFS_EXTENT_TREE_OBJECTID, extent_root);
+	if (ret)
+		goto fail_tree_root;
+	extent_root->track_dirty = 1;
+
+	ret = find_and_setup_root(tree_root, fs_info,
+				  BTRFS_DEV_TREE_OBJECTID, dev_root);
+	dev_root->track_dirty = 1;
+
+	if (ret)
+		goto fail_extent_root;
+
+	ret = find_and_setup_root(tree_root, fs_info,
+				  BTRFS_CSUM_TREE_OBJECTID, csum_root);
+	if (ret)
+		goto fail_extent_root;
+
+	csum_root->track_dirty = 1;
+
+	btrfs_read_block_groups(extent_root);
+
+	fs_info->generation = generation;
+	fs_info->last_trans_committed = generation;
+	fs_info->data_alloc_profile = (u64)-1;
+	fs_info->metadata_alloc_profile = (u64)-1;
+	fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
+	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
+					       "btrfs-cleaner");
+	if (!fs_info->cleaner_kthread)
+		goto fail_csum_root;
+
+	fs_info->transaction_kthread = kthread_run(transaction_kthread,
+						   tree_root,
+						   "btrfs-transaction");
+	if (!fs_info->transaction_kthread)
+		goto fail_cleaner;
+
+	if (btrfs_super_log_root(disk_super) != 0) {
+		u64 bytenr = btrfs_super_log_root(disk_super);
+
+		if (fs_devices->rw_devices == 0) {
+			printk(KERN_WARNING "Btrfs log replay required "
+			       "on RO media\n");
+			err = -EIO;
+			goto fail_trans_kthread;
+		}
+		blocksize =
+		     btrfs_level_size(tree_root,
+				      btrfs_super_log_root_level(disk_super));
+
+		log_tree_root = kzalloc(sizeof(struct btrfs_root),
+						      GFP_NOFS);
+
+		__setup_root(nodesize, leafsize, sectorsize, stripesize,
+			     log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
+
+		log_tree_root->node = read_tree_block(tree_root, bytenr,
+						      blocksize,
+						      generation + 1);
+		ret = btrfs_recover_log_trees(log_tree_root);
+		BUG_ON(ret);
+
+		if (sb->s_flags & MS_RDONLY) {
+			ret =  btrfs_commit_super(tree_root);
+			BUG_ON(ret);
+		}
+	}
+
+	if (!(sb->s_flags & MS_RDONLY)) {
+		ret = btrfs_cleanup_reloc_trees(tree_root);
+		BUG_ON(ret);
+	}
+
+	location.objectid = BTRFS_FS_TREE_OBJECTID;
+	location.type = BTRFS_ROOT_ITEM_KEY;
+	location.offset = (u64)-1;
+
+	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
+	if (!fs_info->fs_root)
+		goto fail_trans_kthread;
+	return tree_root;
+
+fail_trans_kthread:
+	kthread_stop(fs_info->transaction_kthread);
+fail_cleaner:
+	kthread_stop(fs_info->cleaner_kthread);
+
+	/*
+	 * make sure we're done with the btree inode before we stop our
+	 * kthreads
+	 */
+	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
+	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
+
+fail_csum_root:
+	free_extent_buffer(csum_root->node);
+fail_extent_root:
+	free_extent_buffer(extent_root->node);
+fail_tree_root:
+	free_extent_buffer(tree_root->node);
+fail_chunk_root:
+	free_extent_buffer(chunk_root->node);
+fail_sys_array:
+	free_extent_buffer(dev_root->node);
+fail_sb_buffer:
+	btrfs_stop_workers(&fs_info->fixup_workers);
+	btrfs_stop_workers(&fs_info->delalloc_workers);
+	btrfs_stop_workers(&fs_info->workers);
+	btrfs_stop_workers(&fs_info->endio_workers);
+	btrfs_stop_workers(&fs_info->endio_meta_workers);
+	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
+	btrfs_stop_workers(&fs_info->endio_write_workers);
+	btrfs_stop_workers(&fs_info->submit_workers);
+fail_iput:
+	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
+	iput(fs_info->btree_inode);
+fail:
+	btrfs_close_devices(fs_info->fs_devices);
+	btrfs_mapping_tree_free(&fs_info->mapping_tree);
+
+	kfree(extent_root);
+	kfree(tree_root);
+	bdi_destroy(&fs_info->bdi);
+	kfree(fs_info);
+	kfree(chunk_root);
+	kfree(dev_root);
+	kfree(csum_root);
+	return ERR_PTR(err);
+}
+
+static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
+{
+	char b[BDEVNAME_SIZE];
+
+	if (uptodate) {
+		set_buffer_uptodate(bh);
+	} else {
+		if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
+			printk(KERN_WARNING "lost page write due to "
+					"I/O error on %s\n",
+				       bdevname(bh->b_bdev, b));
+		}
+		/* note, we dont' set_buffer_write_io_error because we have
+		 * our own ways of dealing with the IO errors
+		 */
+		clear_buffer_uptodate(bh);
+	}
+	unlock_buffer(bh);
+	put_bh(bh);
+}
+
+struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
+{
+	struct buffer_head *bh;
+	struct buffer_head *latest = NULL;
+	struct btrfs_super_block *super;
+	int i;
+	u64 transid = 0;
+	u64 bytenr;
+
+	/* we would like to check all the supers, but that would make
+	 * a btrfs mount succeed after a mkfs from a different FS.
+	 * So, we need to add a special mount option to scan for
+	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
+	 */
+	for (i = 0; i < 1; i++) {
+		bytenr = btrfs_sb_offset(i);
+		if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
+			break;
+		bh = __bread(bdev, bytenr / 4096, 4096);
+		if (!bh)
+			continue;
+
+		super = (struct btrfs_super_block *)bh->b_data;
+		if (btrfs_super_bytenr(super) != bytenr ||
+		    strncmp((char *)(&super->magic), BTRFS_MAGIC,
+			    sizeof(super->magic))) {
+			brelse(bh);
+			continue;
+		}
+
+		if (!latest || btrfs_super_generation(super) > transid) {
+			brelse(latest);
+			latest = bh;
+			transid = btrfs_super_generation(super);
+		} else {
+			brelse(bh);
+		}
+	}
+	return latest;
+}
+
+static int write_dev_supers(struct btrfs_device *device,
+			    struct btrfs_super_block *sb,
+			    int do_barriers, int wait, int max_mirrors)
+{
+	struct buffer_head *bh;
+	int i;
+	int ret;
+	int errors = 0;
+	u32 crc;
+	u64 bytenr;
+	int last_barrier = 0;
+
+	if (max_mirrors == 0)
+		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
+
+	/* make sure only the last submit_bh does a barrier */
+	if (do_barriers) {
+		for (i = 0; i < max_mirrors; i++) {
+			bytenr = btrfs_sb_offset(i);
+			if (bytenr + BTRFS_SUPER_INFO_SIZE >=
+			    device->total_bytes)
+				break;
+			last_barrier = i;
+		}
+	}
+
+	for (i = 0; i < max_mirrors; i++) {
+		bytenr = btrfs_sb_offset(i);
+		if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
+			break;
+
+		if (wait) {
+			bh = __find_get_block(device->bdev, bytenr / 4096,
+					      BTRFS_SUPER_INFO_SIZE);
+			BUG_ON(!bh);
+			brelse(bh);
+			wait_on_buffer(bh);
+			if (buffer_uptodate(bh)) {
+				brelse(bh);
+				continue;
+			}
+		} else {
+			btrfs_set_super_bytenr(sb, bytenr);
+
+			crc = ~(u32)0;
+			crc = btrfs_csum_data(NULL, (char *)sb +
+					      BTRFS_CSUM_SIZE, crc,
+					      BTRFS_SUPER_INFO_SIZE -
+					      BTRFS_CSUM_SIZE);
+			btrfs_csum_final(crc, sb->csum);
+
+			bh = __getblk(device->bdev, bytenr / 4096,
+				      BTRFS_SUPER_INFO_SIZE);
+			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
+
+			set_buffer_uptodate(bh);
+			get_bh(bh);
+			lock_buffer(bh);
+			bh->b_end_io = btrfs_end_buffer_write_sync;
+		}
+
+		if (i == last_barrier && do_barriers && device->barriers) {
+			ret = submit_bh(WRITE_BARRIER, bh);
+			if (ret == -EOPNOTSUPP) {
+				printk("btrfs: disabling barriers on dev %s\n",
+				       device->name);
+				set_buffer_uptodate(bh);
+				device->barriers = 0;
+				get_bh(bh);
+				lock_buffer(bh);
+				ret = submit_bh(WRITE, bh);
+			}
+		} else {
+			ret = submit_bh(WRITE, bh);
+		}
+
+		if (!ret && wait) {
+			wait_on_buffer(bh);
+			if (!buffer_uptodate(bh))
+				errors++;
+		} else if (ret) {
+			errors++;
+		}
+		if (wait)
+			brelse(bh);
+	}
+	return errors < i ? 0 : -1;
+}
+
+int write_all_supers(struct btrfs_root *root, int max_mirrors)
+{
+	struct list_head *cur;
+	struct list_head *head = &root->fs_info->fs_devices->devices;
+	struct btrfs_device *dev;
+	struct btrfs_super_block *sb;
+	struct btrfs_dev_item *dev_item;
+	int ret;
+	int do_barriers;
+	int max_errors;
+	int total_errors = 0;
+	u64 flags;
+
+	max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
+	do_barriers = !btrfs_test_opt(root, NOBARRIER);
+
+	sb = &root->fs_info->super_for_commit;
+	dev_item = &sb->dev_item;
+	list_for_each(cur, head) {
+		dev = list_entry(cur, struct btrfs_device, dev_list);
+		if (!dev->bdev) {
+			total_errors++;
+			continue;
+		}
+		if (!dev->in_fs_metadata || !dev->writeable)
+			continue;
+
+		btrfs_set_stack_device_generation(dev_item, 0);
+		btrfs_set_stack_device_type(dev_item, dev->type);
+		btrfs_set_stack_device_id(dev_item, dev->devid);
+		btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
+		btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
+		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
+		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
+		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
+		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
+		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
+
+		flags = btrfs_super_flags(sb);
+		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
+
+		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
+		if (ret)
+			total_errors++;
+	}
+	if (total_errors > max_errors) {
+		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
+		       total_errors);
+		BUG();
+	}
+
+	total_errors = 0;
+	list_for_each(cur, head) {
+		dev = list_entry(cur, struct btrfs_device, dev_list);
+		if (!dev->bdev)
+			continue;
+		if (!dev->in_fs_metadata || !dev->writeable)
+			continue;
+
+		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
+		if (ret)
+			total_errors++;
+	}
+	if (total_errors > max_errors) {
+		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
+		       total_errors);
+		BUG();
+	}
+	return 0;
+}
+
+int write_ctree_super(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root, int max_mirrors)
+{
+	int ret;
+
+	ret = write_all_supers(root, max_mirrors);
+	return ret;
+}
+
+int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
+{
+	radix_tree_delete(&fs_info->fs_roots_radix,
+			  (unsigned long)root->root_key.objectid);
+	if (root->anon_super.s_dev) {
+		down_write(&root->anon_super.s_umount);
+		kill_anon_super(&root->anon_super);
+	}
+	if (root->node)
+		free_extent_buffer(root->node);
+	if (root->commit_root)
+		free_extent_buffer(root->commit_root);
+	kfree(root->name);
+	kfree(root);
+	return 0;
+}
+
+static int del_fs_roots(struct btrfs_fs_info *fs_info)
+{
+	int ret;
+	struct btrfs_root *gang[8];
+	int i;
+
+	while (1) {
+		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
+					     (void **)gang, 0,
+					     ARRAY_SIZE(gang));
+		if (!ret)
+			break;
+		for (i = 0; i < ret; i++)
+			btrfs_free_fs_root(fs_info, gang[i]);
+	}
+	return 0;
+}
+
+int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
+{
+	u64 root_objectid = 0;
+	struct btrfs_root *gang[8];
+	int i;
+	int ret;
+
+	while (1) {
+		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
+					     (void **)gang, root_objectid,
+					     ARRAY_SIZE(gang));
+		if (!ret)
+			break;
+		for (i = 0; i < ret; i++) {
+			root_objectid = gang[i]->root_key.objectid;
+			ret = btrfs_find_dead_roots(fs_info->tree_root,
+						    root_objectid, gang[i]);
+			BUG_ON(ret);
+			btrfs_orphan_cleanup(gang[i]);
+		}
+		root_objectid++;
+	}
+	return 0;
+}
+
+int btrfs_commit_super(struct btrfs_root *root)
+{
+	struct btrfs_trans_handle *trans;
+	int ret;
+
+	mutex_lock(&root->fs_info->cleaner_mutex);
+	btrfs_clean_old_snapshots(root);
+	mutex_unlock(&root->fs_info->cleaner_mutex);
+	trans = btrfs_start_transaction(root, 1);
+	ret = btrfs_commit_transaction(trans, root);
+	BUG_ON(ret);
+	/* run commit again to drop the original snapshot */
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_commit_transaction(trans, root);
+	ret = btrfs_write_and_wait_transaction(NULL, root);
+	BUG_ON(ret);
+
+	ret = write_ctree_super(NULL, root, 0);
+	return ret;
+}
+
+int close_ctree(struct btrfs_root *root)
+{
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	int ret;
+
+	fs_info->closing = 1;
+	smp_mb();
+
+	kthread_stop(root->fs_info->transaction_kthread);
+	kthread_stop(root->fs_info->cleaner_kthread);
+
+	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
+		ret =  btrfs_commit_super(root);
+		if (ret)
+			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
+	}
+
+	if (fs_info->delalloc_bytes) {
+		printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
+		       fs_info->delalloc_bytes);
+	}
+	if (fs_info->total_ref_cache_size) {
+		printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
+		       (unsigned long long)fs_info->total_ref_cache_size);
+	}
+
+	if (fs_info->extent_root->node)
+		free_extent_buffer(fs_info->extent_root->node);
+
+	if (fs_info->tree_root->node)
+		free_extent_buffer(fs_info->tree_root->node);
+
+	if (root->fs_info->chunk_root->node)
+		free_extent_buffer(root->fs_info->chunk_root->node);
+
+	if (root->fs_info->dev_root->node)
+		free_extent_buffer(root->fs_info->dev_root->node);
+
+	if (root->fs_info->csum_root->node)
+		free_extent_buffer(root->fs_info->csum_root->node);
+
+	btrfs_free_block_groups(root->fs_info);
+
+	del_fs_roots(fs_info);
+
+	iput(fs_info->btree_inode);
+
+	btrfs_stop_workers(&fs_info->fixup_workers);
+	btrfs_stop_workers(&fs_info->delalloc_workers);
+	btrfs_stop_workers(&fs_info->workers);
+	btrfs_stop_workers(&fs_info->endio_workers);
+	btrfs_stop_workers(&fs_info->endio_meta_workers);
+	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
+	btrfs_stop_workers(&fs_info->endio_write_workers);
+	btrfs_stop_workers(&fs_info->submit_workers);
+
+#if 0
+	while (!list_empty(&fs_info->hashers)) {
+		struct btrfs_hasher *hasher;
+		hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
+				    hashers);
+		list_del(&hasher->hashers);
+		crypto_free_hash(&fs_info->hash_tfm);
+		kfree(hasher);
+	}
+#endif
+	btrfs_close_devices(fs_info->fs_devices);
+	btrfs_mapping_tree_free(&fs_info->mapping_tree);
+
+	bdi_destroy(&fs_info->bdi);
+
+	kfree(fs_info->extent_root);
+	kfree(fs_info->tree_root);
+	kfree(fs_info->chunk_root);
+	kfree(fs_info->dev_root);
+	kfree(fs_info->csum_root);
+	return 0;
+}
+
+int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
+{
+	int ret;
+	struct inode *btree_inode = buf->first_page->mapping->host;
+
+	ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
+	if (!ret)
+		return ret;
+
+	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
+				    parent_transid);
+	return !ret;
+}
+
+int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
+{
+	struct inode *btree_inode = buf->first_page->mapping->host;
+	return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
+					  buf);
+}
+
+void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
+{
+	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
+	u64 transid = btrfs_header_generation(buf);
+	struct inode *btree_inode = root->fs_info->btree_inode;
+
+	WARN_ON(!btrfs_tree_locked(buf));
+	if (transid != root->fs_info->generation) {
+		printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
+		       "found %llu running %llu\n",
+			(unsigned long long)buf->start,
+			(unsigned long long)transid,
+			(unsigned long long)root->fs_info->generation);
+		WARN_ON(1);
+	}
+	set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
+}
+
+void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
+{
+	/*
+	 * looks as though older kernels can get into trouble with
+	 * this code, they end up stuck in balance_dirty_pages forever
+	 */
+	struct extent_io_tree *tree;
+	u64 num_dirty;
+	u64 start = 0;
+	unsigned long thresh = 32 * 1024 * 1024;
+	tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
+
+	if (current_is_pdflush() || current->flags & PF_MEMALLOC)
+		return;
+
+	num_dirty = count_range_bits(tree, &start, (u64)-1,
+				     thresh, EXTENT_DIRTY);
+	if (num_dirty > thresh) {
+		balance_dirty_pages_ratelimited_nr(
+				   root->fs_info->btree_inode->i_mapping, 1);
+	}
+	return;
+}
+
+int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
+{
+	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
+	int ret;
+	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
+	if (ret == 0)
+		buf->flags |= EXTENT_UPTODATE;
+	return ret;
+}
+
+int btree_lock_page_hook(struct page *page)
+{
+	struct inode *inode = page->mapping->host;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct extent_buffer *eb;
+	unsigned long len;
+	u64 bytenr = page_offset(page);
+
+	if (page->private == EXTENT_PAGE_PRIVATE)
+		goto out;
+
+	len = page->private >> 2;
+	eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
+	if (!eb)
+		goto out;
+
+	btrfs_tree_lock(eb);
+	spin_lock(&root->fs_info->hash_lock);
+	btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
+	spin_unlock(&root->fs_info->hash_lock);
+	btrfs_tree_unlock(eb);
+	free_extent_buffer(eb);
+out:
+	lock_page(page);
+	return 0;
+}
+
+static struct extent_io_ops btree_extent_io_ops = {
+	.write_cache_pages_lock_hook = btree_lock_page_hook,
+	.readpage_end_io_hook = btree_readpage_end_io_hook,
+	.submit_bio_hook = btree_submit_bio_hook,
+	/* note we're sharing with inode.c for the merge bio hook */
+	.merge_bio_hook = btrfs_merge_bio_hook,
+};
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
new file mode 100644
index 0000000..c0ff404
--- /dev/null
+++ b/fs/btrfs/disk-io.h
@@ -0,0 +1,102 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __DISKIO__
+#define __DISKIO__
+
+#define BTRFS_SUPER_INFO_OFFSET (64 * 1024)
+#define BTRFS_SUPER_INFO_SIZE 4096
+
+#define BTRFS_SUPER_MIRROR_MAX	 3
+#define BTRFS_SUPER_MIRROR_SHIFT 12
+
+static inline u64 btrfs_sb_offset(int mirror)
+{
+	u64 start = 16 * 1024;
+	if (mirror)
+		return start << (BTRFS_SUPER_MIRROR_SHIFT * mirror);
+	return BTRFS_SUPER_INFO_OFFSET;
+}
+
+struct btrfs_device;
+struct btrfs_fs_devices;
+
+struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
+				      u32 blocksize, u64 parent_transid);
+int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
+			 u64 parent_transid);
+struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
+						   u64 bytenr, u32 blocksize);
+int clean_tree_block(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root, struct extent_buffer *buf);
+struct btrfs_root *open_ctree(struct super_block *sb,
+			      struct btrfs_fs_devices *fs_devices,
+			      char *options);
+int close_ctree(struct btrfs_root *root);
+int write_ctree_super(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root, int max_mirrors);
+struct buffer_head *btrfs_read_dev_super(struct block_device *bdev);
+int btrfs_commit_super(struct btrfs_root *root);
+struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
+					    u64 bytenr, u32 blocksize);
+struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
+					u64 root_objectid);
+struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
+				      struct btrfs_key *location,
+				      const char *name, int namelen);
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
+					       struct btrfs_key *location);
+struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
+					      struct btrfs_key *location);
+int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info);
+int btrfs_insert_dev_radix(struct btrfs_root *root,
+			   struct block_device *bdev,
+			   u64 device_id,
+			   u64 block_start,
+			   u64 num_blocks);
+void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr);
+int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root);
+void btrfs_mark_buffer_dirty(struct extent_buffer *buf);
+int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid);
+int btrfs_set_buffer_uptodate(struct extent_buffer *buf);
+int wait_on_tree_block_writeback(struct btrfs_root *root,
+				 struct extent_buffer *buf);
+int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid);
+u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len);
+void btrfs_csum_final(u32 crc, char *result);
+int btrfs_open_device(struct btrfs_device *dev);
+int btrfs_verify_block_csum(struct btrfs_root *root,
+			    struct extent_buffer *buf);
+int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
+			int metadata);
+int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
+			int rw, struct bio *bio, int mirror_num,
+			unsigned long bio_flags,
+			extent_submit_bio_hook_t *submit_bio_start,
+			extent_submit_bio_hook_t *submit_bio_done);
+
+int btrfs_congested_async(struct btrfs_fs_info *info, int iodone);
+unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info);
+int btrfs_write_tree_block(struct extent_buffer *buf);
+int btrfs_wait_tree_block_writeback(struct extent_buffer *buf);
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info);
+int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info);
+int btree_lock_page_hook(struct page *page);
+#endif
-- 
1.6.0.2

[PATCH 34/35] Btrfs compression routines

[Chris Mason]

Btrfs can optionally compress all file data transparently during reads
and writes.  This code takes pages and sends them down to zlib.  Working
pages are allocated to hold the compressed/uncompressed result from zlib
and bios are submitted to send the compressed IOs.

The interfaces here allow the compression code to limit the max size
of the compressed result.  This is used to make sure a compressed extent
isn't too large to decompress later to satisfy random reads.

The compression code includes checks to make sure it is actually making
the data smaller, and it bails early if it is not being effective.

Checksumming in btrfs is done based on the data on disk (the compressed data),
so the bios are sent through the checksumming routines as appropriate.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
new file mode 100644
index 0000000..ee848d8
--- /dev/null
+++ b/fs/btrfs/compression.c
@@ -0,0 +1,709 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/kernel.h>
+#include <linux/bio.h>
+#include <linux/buffer_head.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/highmem.h>
+#include <linux/time.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/smp_lock.h>
+#include <linux/backing-dev.h>
+#include <linux/mpage.h>
+#include <linux/swap.h>
+#include <linux/writeback.h>
+#include <linux/bit_spinlock.h>
+#include <linux/version.h>
+#include <linux/pagevec.h>
+#include "compat.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "btrfs_inode.h"
+#include "volumes.h"
+#include "ordered-data.h"
+#include "compression.h"
+#include "extent_io.h"
+#include "extent_map.h"
+
+struct compressed_bio {
+	/* number of bios pending for this compressed extent */
+	atomic_t pending_bios;
+
+	/* the pages with the compressed data on them */
+	struct page **compressed_pages;
+
+	/* inode that owns this data */
+	struct inode *inode;
+
+	/* starting offset in the inode for our pages */
+	u64 start;
+
+	/* number of bytes in the inode we're working on */
+	unsigned long len;
+
+	/* number of bytes on disk */
+	unsigned long compressed_len;
+
+	/* number of compressed pages in the array */
+	unsigned long nr_pages;
+
+	/* IO errors */
+	int errors;
+	int mirror_num;
+
+	/* for reads, this is the bio we are copying the data into */
+	struct bio *orig_bio;
+
+	/*
+	 * the start of a variable length array of checksums only
+	 * used by reads
+	 */
+	u32 sums;
+};
+
+static inline int compressed_bio_size(struct btrfs_root *root,
+				      unsigned long disk_size)
+{
+	u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
+	return sizeof(struct compressed_bio) +
+		((disk_size + root->sectorsize - 1) / root->sectorsize) *
+		csum_size;
+}
+
+static struct bio *compressed_bio_alloc(struct block_device *bdev,
+					u64 first_byte, gfp_t gfp_flags)
+{
+	struct bio *bio;
+	int nr_vecs;
+
+	nr_vecs = bio_get_nr_vecs(bdev);
+	bio = bio_alloc(gfp_flags, nr_vecs);
+
+	if (bio == NULL && (current->flags & PF_MEMALLOC)) {
+		while (!bio && (nr_vecs /= 2))
+			bio = bio_alloc(gfp_flags, nr_vecs);
+	}
+
+	if (bio) {
+		bio->bi_size = 0;
+		bio->bi_bdev = bdev;
+		bio->bi_sector = first_byte >> 9;
+	}
+	return bio;
+}
+
+static int check_compressed_csum(struct inode *inode,
+				 struct compressed_bio *cb,
+				 u64 disk_start)
+{
+	int ret;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct page *page;
+	unsigned long i;
+	char *kaddr;
+	u32 csum;
+	u32 *cb_sum = &cb->sums;
+
+	if (btrfs_test_flag(inode, NODATASUM))
+		return 0;
+
+	for (i = 0; i < cb->nr_pages; i++) {
+		page = cb->compressed_pages[i];
+		csum = ~(u32)0;
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		csum = btrfs_csum_data(root, kaddr, csum, PAGE_CACHE_SIZE);
+		btrfs_csum_final(csum, (char *)&csum);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		if (csum != *cb_sum) {
+			printk(KERN_INFO "btrfs csum failed ino %lu "
+			       "extent %llu csum %u "
+			       "wanted %u mirror %d\n", inode->i_ino,
+			       (unsigned long long)disk_start,
+			       csum, *cb_sum, cb->mirror_num);
+			ret = -EIO;
+			goto fail;
+		}
+		cb_sum++;
+
+	}
+	ret = 0;
+fail:
+	return ret;
+}
+
+/* when we finish reading compressed pages from the disk, we
+ * decompress them and then run the bio end_io routines on the
+ * decompressed pages (in the inode address space).
+ *
+ * This allows the checksumming and other IO error handling routines
+ * to work normally
+ *
+ * The compressed pages are freed here, and it must be run
+ * in process context
+ */
+static void end_compressed_bio_read(struct bio *bio, int err)
+{
+	struct extent_io_tree *tree;
+	struct compressed_bio *cb = bio->bi_private;
+	struct inode *inode;
+	struct page *page;
+	unsigned long index;
+	int ret;
+
+	if (err)
+		cb->errors = 1;
+
+	/* if there are more bios still pending for this compressed
+	 * extent, just exit
+	 */
+	if (!atomic_dec_and_test(&cb->pending_bios))
+		goto out;
+
+	inode = cb->inode;
+	ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9);
+	if (ret)
+		goto csum_failed;
+
+	/* ok, we're the last bio for this extent, lets start
+	 * the decompression.
+	 */
+	tree = &BTRFS_I(inode)->io_tree;
+	ret = btrfs_zlib_decompress_biovec(cb->compressed_pages,
+					cb->start,
+					cb->orig_bio->bi_io_vec,
+					cb->orig_bio->bi_vcnt,
+					cb->compressed_len);
+csum_failed:
+	if (ret)
+		cb->errors = 1;
+
+	/* release the compressed pages */
+	index = 0;
+	for (index = 0; index < cb->nr_pages; index++) {
+		page = cb->compressed_pages[index];
+		page->mapping = NULL;
+		page_cache_release(page);
+	}
+
+	/* do io completion on the original bio */
+	if (cb->errors) {
+		bio_io_error(cb->orig_bio);
+	} else {
+		int bio_index = 0;
+		struct bio_vec *bvec = cb->orig_bio->bi_io_vec;
+
+		/*
+		 * we have verified the checksum already, set page
+		 * checked so the end_io handlers know about it
+		 */
+		while (bio_index < cb->orig_bio->bi_vcnt) {
+			SetPageChecked(bvec->bv_page);
+			bvec++;
+			bio_index++;
+		}
+		bio_endio(cb->orig_bio, 0);
+	}
+
+	/* finally free the cb struct */
+	kfree(cb->compressed_pages);
+	kfree(cb);
+out:
+	bio_put(bio);
+}
+
+/*
+ * Clear the writeback bits on all of the file
+ * pages for a compressed write
+ */
+static noinline int end_compressed_writeback(struct inode *inode, u64 start,
+					     unsigned long ram_size)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
+	struct page *pages[16];
+	unsigned long nr_pages = end_index - index + 1;
+	int i;
+	int ret;
+
+	while (nr_pages > 0) {
+		ret = find_get_pages_contig(inode->i_mapping, index,
+				     min_t(unsigned long,
+				     nr_pages, ARRAY_SIZE(pages)), pages);
+		if (ret == 0) {
+			nr_pages -= 1;
+			index += 1;
+			continue;
+		}
+		for (i = 0; i < ret; i++) {
+			end_page_writeback(pages[i]);
+			page_cache_release(pages[i]);
+		}
+		nr_pages -= ret;
+		index += ret;
+	}
+	/* the inode may be gone now */
+	return 0;
+}
+
+/*
+ * do the cleanup once all the compressed pages hit the disk.
+ * This will clear writeback on the file pages and free the compressed
+ * pages.
+ *
+ * This also calls the writeback end hooks for the file pages so that
+ * metadata and checksums can be updated in the file.
+ */
+static void end_compressed_bio_write(struct bio *bio, int err)
+{
+	struct extent_io_tree *tree;
+	struct compressed_bio *cb = bio->bi_private;
+	struct inode *inode;
+	struct page *page;
+	unsigned long index;
+
+	if (err)
+		cb->errors = 1;
+
+	/* if there are more bios still pending for this compressed
+	 * extent, just exit
+	 */
+	if (!atomic_dec_and_test(&cb->pending_bios))
+		goto out;
+
+	/* ok, we're the last bio for this extent, step one is to
+	 * call back into the FS and do all the end_io operations
+	 */
+	inode = cb->inode;
+	tree = &BTRFS_I(inode)->io_tree;
+	cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
+	tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
+					 cb->start,
+					 cb->start + cb->len - 1,
+					 NULL, 1);
+	cb->compressed_pages[0]->mapping = NULL;
+
+	end_compressed_writeback(inode, cb->start, cb->len);
+	/* note, our inode could be gone now */
+
+	/*
+	 * release the compressed pages, these came from alloc_page and
+	 * are not attached to the inode at all
+	 */
+	index = 0;
+	for (index = 0; index < cb->nr_pages; index++) {
+		page = cb->compressed_pages[index];
+		page->mapping = NULL;
+		page_cache_release(page);
+	}
+
+	/* finally free the cb struct */
+	kfree(cb->compressed_pages);
+	kfree(cb);
+out:
+	bio_put(bio);
+}
+
+/*
+ * worker function to build and submit bios for previously compressed pages.
+ * The corresponding pages in the inode should be marked for writeback
+ * and the compressed pages should have a reference on them for dropping
+ * when the IO is complete.
+ *
+ * This also checksums the file bytes and gets things ready for
+ * the end io hooks.
+ */
+int btrfs_submit_compressed_write(struct inode *inode, u64 start,
+				 unsigned long len, u64 disk_start,
+				 unsigned long compressed_len,
+				 struct page **compressed_pages,
+				 unsigned long nr_pages)
+{
+	struct bio *bio = NULL;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct compressed_bio *cb;
+	unsigned long bytes_left;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	int page_index = 0;
+	struct page *page;
+	u64 first_byte = disk_start;
+	struct block_device *bdev;
+	int ret;
+
+	WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
+	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
+	atomic_set(&cb->pending_bios, 0);
+	cb->errors = 0;
+	cb->inode = inode;
+	cb->start = start;
+	cb->len = len;
+	cb->mirror_num = 0;
+	cb->compressed_pages = compressed_pages;
+	cb->compressed_len = compressed_len;
+	cb->orig_bio = NULL;
+	cb->nr_pages = nr_pages;
+
+	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
+
+	bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
+	bio->bi_private = cb;
+	bio->bi_end_io = end_compressed_bio_write;
+	atomic_inc(&cb->pending_bios);
+
+	/* create and submit bios for the compressed pages */
+	bytes_left = compressed_len;
+	for (page_index = 0; page_index < cb->nr_pages; page_index++) {
+		page = compressed_pages[page_index];
+		page->mapping = inode->i_mapping;
+		if (bio->bi_size)
+			ret = io_tree->ops->merge_bio_hook(page, 0,
+							   PAGE_CACHE_SIZE,
+							   bio, 0);
+		else
+			ret = 0;
+
+		page->mapping = NULL;
+		if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
+		    PAGE_CACHE_SIZE) {
+			bio_get(bio);
+
+			/*
+			 * inc the count before we submit the bio so
+			 * we know the end IO handler won't happen before
+			 * we inc the count.  Otherwise, the cb might get
+			 * freed before we're done setting it up
+			 */
+			atomic_inc(&cb->pending_bios);
+			ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
+			BUG_ON(ret);
+
+			ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
+			BUG_ON(ret);
+
+			ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
+			BUG_ON(ret);
+
+			bio_put(bio);
+
+			bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
+			bio->bi_private = cb;
+			bio->bi_end_io = end_compressed_bio_write;
+			bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
+		}
+		if (bytes_left < PAGE_CACHE_SIZE) {
+			printk("bytes left %lu compress len %lu nr %lu\n",
+			       bytes_left, cb->compressed_len, cb->nr_pages);
+		}
+		bytes_left -= PAGE_CACHE_SIZE;
+		first_byte += PAGE_CACHE_SIZE;
+		cond_resched();
+	}
+	bio_get(bio);
+
+	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
+	BUG_ON(ret);
+
+	ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
+	BUG_ON(ret);
+
+	ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
+	BUG_ON(ret);
+
+	bio_put(bio);
+	return 0;
+}
+
+static noinline int add_ra_bio_pages(struct inode *inode,
+				     u64 compressed_end,
+				     struct compressed_bio *cb)
+{
+	unsigned long end_index;
+	unsigned long page_index;
+	u64 last_offset;
+	u64 isize = i_size_read(inode);
+	int ret;
+	struct page *page;
+	unsigned long nr_pages = 0;
+	struct extent_map *em;
+	struct address_space *mapping = inode->i_mapping;
+	struct pagevec pvec;
+	struct extent_map_tree *em_tree;
+	struct extent_io_tree *tree;
+	u64 end;
+	int misses = 0;
+
+	page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
+	last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
+	em_tree = &BTRFS_I(inode)->extent_tree;
+	tree = &BTRFS_I(inode)->io_tree;
+
+	if (isize == 0)
+		return 0;
+
+	end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
+
+	pagevec_init(&pvec, 0);
+	while (last_offset < compressed_end) {
+		page_index = last_offset >> PAGE_CACHE_SHIFT;
+
+		if (page_index > end_index)
+			break;
+
+		rcu_read_lock();
+		page = radix_tree_lookup(&mapping->page_tree, page_index);
+		rcu_read_unlock();
+		if (page) {
+			misses++;
+			if (misses > 4)
+				break;
+			goto next;
+		}
+
+		page = alloc_page(mapping_gfp_mask(mapping) | GFP_NOFS);
+		if (!page)
+			break;
+
+		page->index = page_index;
+		/*
+		 * what we want to do here is call add_to_page_cache_lru,
+		 * but that isn't exported, so we reproduce it here
+		 */
+		if (add_to_page_cache(page, mapping,
+				      page->index, GFP_NOFS)) {
+			page_cache_release(page);
+			goto next;
+		}
+
+		/* open coding of lru_cache_add, also not exported */
+		page_cache_get(page);
+		if (!pagevec_add(&pvec, page))
+			__pagevec_lru_add_file(&pvec);
+
+		end = last_offset + PAGE_CACHE_SIZE - 1;
+		/*
+		 * at this point, we have a locked page in the page cache
+		 * for these bytes in the file.  But, we have to make
+		 * sure they map to this compressed extent on disk.
+		 */
+		set_page_extent_mapped(page);
+		lock_extent(tree, last_offset, end, GFP_NOFS);
+		spin_lock(&em_tree->lock);
+		em = lookup_extent_mapping(em_tree, last_offset,
+					   PAGE_CACHE_SIZE);
+		spin_unlock(&em_tree->lock);
+
+		if (!em || last_offset < em->start ||
+		    (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
+		    (em->block_start >> 9) != cb->orig_bio->bi_sector) {
+			free_extent_map(em);
+			unlock_extent(tree, last_offset, end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			break;
+		}
+		free_extent_map(em);
+
+		if (page->index == end_index) {
+			char *userpage;
+			size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
+
+			if (zero_offset) {
+				int zeros;
+				zeros = PAGE_CACHE_SIZE - zero_offset;
+				userpage = kmap_atomic(page, KM_USER0);
+				memset(userpage + zero_offset, 0, zeros);
+				flush_dcache_page(page);
+				kunmap_atomic(userpage, KM_USER0);
+			}
+		}
+
+		ret = bio_add_page(cb->orig_bio, page,
+				   PAGE_CACHE_SIZE, 0);
+
+		if (ret == PAGE_CACHE_SIZE) {
+			nr_pages++;
+			page_cache_release(page);
+		} else {
+			unlock_extent(tree, last_offset, end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			break;
+		}
+next:
+		last_offset += PAGE_CACHE_SIZE;
+	}
+	if (pagevec_count(&pvec))
+		__pagevec_lru_add_file(&pvec);
+	return 0;
+}
+
+/*
+ * for a compressed read, the bio we get passed has all the inode pages
+ * in it.  We don't actually do IO on those pages but allocate new ones
+ * to hold the compressed pages on disk.
+ *
+ * bio->bi_sector points to the compressed extent on disk
+ * bio->bi_io_vec points to all of the inode pages
+ * bio->bi_vcnt is a count of pages
+ *
+ * After the compressed pages are read, we copy the bytes into the
+ * bio we were passed and then call the bio end_io calls
+ */
+int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
+				 int mirror_num, unsigned long bio_flags)
+{
+	struct extent_io_tree *tree;
+	struct extent_map_tree *em_tree;
+	struct compressed_bio *cb;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
+	unsigned long compressed_len;
+	unsigned long nr_pages;
+	unsigned long page_index;
+	struct page *page;
+	struct block_device *bdev;
+	struct bio *comp_bio;
+	u64 cur_disk_byte = (u64)bio->bi_sector << 9;
+	u64 em_len;
+	u64 em_start;
+	struct extent_map *em;
+	int ret;
+	u32 *sums;
+
+	tree = &BTRFS_I(inode)->io_tree;
+	em_tree = &BTRFS_I(inode)->extent_tree;
+
+	/* we need the actual starting offset of this extent in the file */
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree,
+				   page_offset(bio->bi_io_vec->bv_page),
+				   PAGE_CACHE_SIZE);
+	spin_unlock(&em_tree->lock);
+
+	compressed_len = em->block_len;
+	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
+	atomic_set(&cb->pending_bios, 0);
+	cb->errors = 0;
+	cb->inode = inode;
+	cb->mirror_num = mirror_num;
+	sums = &cb->sums;
+
+	cb->start = em->orig_start;
+	em_len = em->len;
+	em_start = em->start;
+
+	free_extent_map(em);
+	em = NULL;
+
+	cb->len = uncompressed_len;
+	cb->compressed_len = compressed_len;
+	cb->orig_bio = bio;
+
+	nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
+				 PAGE_CACHE_SIZE;
+	cb->compressed_pages = kmalloc(sizeof(struct page *) * nr_pages,
+				       GFP_NOFS);
+	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
+
+	for (page_index = 0; page_index < nr_pages; page_index++) {
+		cb->compressed_pages[page_index] = alloc_page(GFP_NOFS |
+							      __GFP_HIGHMEM);
+	}
+	cb->nr_pages = nr_pages;
+
+	add_ra_bio_pages(inode, em_start + em_len, cb);
+
+	/* include any pages we added in add_ra-bio_pages */
+	uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
+	cb->len = uncompressed_len;
+
+	comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
+	comp_bio->bi_private = cb;
+	comp_bio->bi_end_io = end_compressed_bio_read;
+	atomic_inc(&cb->pending_bios);
+
+	for (page_index = 0; page_index < nr_pages; page_index++) {
+		page = cb->compressed_pages[page_index];
+		page->mapping = inode->i_mapping;
+		page->index = em_start >> PAGE_CACHE_SHIFT;
+
+		if (comp_bio->bi_size)
+			ret = tree->ops->merge_bio_hook(page, 0,
+							PAGE_CACHE_SIZE,
+							comp_bio, 0);
+		else
+			ret = 0;
+
+		page->mapping = NULL;
+		if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
+		    PAGE_CACHE_SIZE) {
+			bio_get(comp_bio);
+
+			ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
+			BUG_ON(ret);
+
+			/*
+			 * inc the count before we submit the bio so
+			 * we know the end IO handler won't happen before
+			 * we inc the count.  Otherwise, the cb might get
+			 * freed before we're done setting it up
+			 */
+			atomic_inc(&cb->pending_bios);
+
+			if (!btrfs_test_flag(inode, NODATASUM)) {
+				btrfs_lookup_bio_sums(root, inode, comp_bio,
+						      sums);
+			}
+			sums += (comp_bio->bi_size + root->sectorsize - 1) /
+				root->sectorsize;
+
+			ret = btrfs_map_bio(root, READ, comp_bio,
+					    mirror_num, 0);
+			BUG_ON(ret);
+
+			bio_put(comp_bio);
+
+			comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
+							GFP_NOFS);
+			comp_bio->bi_private = cb;
+			comp_bio->bi_end_io = end_compressed_bio_read;
+
+			bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
+		}
+		cur_disk_byte += PAGE_CACHE_SIZE;
+	}
+	bio_get(comp_bio);
+
+	ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
+	BUG_ON(ret);
+
+	if (!btrfs_test_flag(inode, NODATASUM))
+		btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
+
+	ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
+	BUG_ON(ret);
+
+	bio_put(comp_bio);
+	return 0;
+}
diff --git a/fs/btrfs/compression.h b/fs/btrfs/compression.h
new file mode 100644
index 0000000..421f5b4
--- /dev/null
+++ b/fs/btrfs/compression.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_COMPRESSION_
+#define __BTRFS_COMPRESSION_
+
+int btrfs_zlib_decompress(unsigned char *data_in,
+			  struct page *dest_page,
+			  unsigned long start_byte,
+			  size_t srclen, size_t destlen);
+int btrfs_zlib_compress_pages(struct address_space *mapping,
+			      u64 start, unsigned long len,
+			      struct page **pages,
+			      unsigned long nr_dest_pages,
+			      unsigned long *out_pages,
+			      unsigned long *total_in,
+			      unsigned long *total_out,
+			      unsigned long max_out);
+int btrfs_zlib_decompress_biovec(struct page **pages_in,
+			      u64 disk_start,
+			      struct bio_vec *bvec,
+			      int vcnt,
+			      size_t srclen);
+void btrfs_zlib_exit(void);
+int btrfs_submit_compressed_write(struct inode *inode, u64 start,
+				  unsigned long len, u64 disk_start,
+				  unsigned long compressed_len,
+				  struct page **compressed_pages,
+				  unsigned long nr_pages);
+int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
+				 int mirror_num, unsigned long bio_flags);
+#endif
-- 
1.6.0.2

[PATCH 35/35] Async thread routines

[Chris Mason]

Btrfs users a number of helper threads to do cpu intensive tasks.  This
includes checksumming, compression and metadata updates that happen
after IO is complete.

This implementation uses kthreads and is intended to resize the thread
pool as it goes along.  Right now the pool grows but does not shrink.

A given thread pool can be flagged as ordered, which breaks the work
up into two phases.  The first phase is unordered work that can happen
in parallel (like compression) and the second phase is done in the
same order entries were put on the queue.

The ordered queues are used for submitting IO.  They allow asynchronous
work to be done while also preserving IO ordering and avoiding extra
seeks from parallel bio submission.

To limit context switches, each kthread tries to maintain a work list
of a given size (default is 64, some pools use other values).  The goal
is to keep a given thread busy without ping-ponging a small amount of work
between a large number of threads.

Btrfs uses a number of thread pools, mostly to prevent deadlocking between
them.  For example, the write IO completion functions may need to read
a metadata block, and so they must be done by separate thread pools.

Except for the word btrfs, these are fairly generic, and could be moved
outside of the btrfs directory.  The locking in them if fairly simple
and has room for optimization.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
new file mode 100644
index 0000000..8e2fec0
--- /dev/null
+++ b/fs/btrfs/async-thread.c
@@ -0,0 +1,419 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/version.h>
+#include <linux/kthread.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+# include <linux/freezer.h>
+#include "async-thread.h"
+
+#define WORK_QUEUED_BIT 0
+#define WORK_DONE_BIT 1
+#define WORK_ORDER_DONE_BIT 2
+
+/*
+ * container for the kthread task pointer and the list of pending work
+ * One of these is allocated per thread.
+ */
+struct btrfs_worker_thread {
+	/* pool we belong to */
+	struct btrfs_workers *workers;
+
+	/* list of struct btrfs_work that are waiting for service */
+	struct list_head pending;
+
+	/* list of worker threads from struct btrfs_workers */
+	struct list_head worker_list;
+
+	/* kthread */
+	struct task_struct *task;
+
+	/* number of things on the pending list */
+	atomic_t num_pending;
+
+	unsigned long sequence;
+
+	/* protects the pending list. */
+	spinlock_t lock;
+
+	/* set to non-zero when this thread is already awake and kicking */
+	int working;
+
+	/* are we currently idle */
+	int idle;
+};
+
+/*
+ * helper function to move a thread onto the idle list after it
+ * has finished some requests.
+ */
+static void check_idle_worker(struct btrfs_worker_thread *worker)
+{
+	if (!worker->idle && atomic_read(&worker->num_pending) <
+	    worker->workers->idle_thresh / 2) {
+		unsigned long flags;
+		spin_lock_irqsave(&worker->workers->lock, flags);
+		worker->idle = 1;
+		list_move(&worker->worker_list, &worker->workers->idle_list);
+		spin_unlock_irqrestore(&worker->workers->lock, flags);
+	}
+}
+
+/*
+ * helper function to move a thread off the idle list after new
+ * pending work is added.
+ */
+static void check_busy_worker(struct btrfs_worker_thread *worker)
+{
+	if (worker->idle && atomic_read(&worker->num_pending) >=
+	    worker->workers->idle_thresh) {
+		unsigned long flags;
+		spin_lock_irqsave(&worker->workers->lock, flags);
+		worker->idle = 0;
+		list_move_tail(&worker->worker_list,
+			       &worker->workers->worker_list);
+		spin_unlock_irqrestore(&worker->workers->lock, flags);
+	}
+}
+
+static noinline int run_ordered_completions(struct btrfs_workers *workers,
+					    struct btrfs_work *work)
+{
+	unsigned long flags;
+
+	if (!workers->ordered)
+		return 0;
+
+	set_bit(WORK_DONE_BIT, &work->flags);
+
+	spin_lock_irqsave(&workers->lock, flags);
+
+	while (!list_empty(&workers->order_list)) {
+		work = list_entry(workers->order_list.next,
+				  struct btrfs_work, order_list);
+
+		if (!test_bit(WORK_DONE_BIT, &work->flags))
+			break;
+
+		/* we are going to call the ordered done function, but
+		 * we leave the work item on the list as a barrier so
+		 * that later work items that are done don't have their
+		 * functions called before this one returns
+		 */
+		if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
+			break;
+
+		spin_unlock_irqrestore(&workers->lock, flags);
+
+		work->ordered_func(work);
+
+		/* now take the lock again and call the freeing code */
+		spin_lock_irqsave(&workers->lock, flags);
+		list_del(&work->order_list);
+		work->ordered_free(work);
+	}
+
+	spin_unlock_irqrestore(&workers->lock, flags);
+	return 0;
+}
+
+/*
+ * main loop for servicing work items
+ */
+static int worker_loop(void *arg)
+{
+	struct btrfs_worker_thread *worker = arg;
+	struct list_head *cur;
+	struct btrfs_work *work;
+	do {
+		spin_lock_irq(&worker->lock);
+		while (!list_empty(&worker->pending)) {
+			cur = worker->pending.next;
+			work = list_entry(cur, struct btrfs_work, list);
+			list_del(&work->list);
+			clear_bit(WORK_QUEUED_BIT, &work->flags);
+
+			work->worker = worker;
+			spin_unlock_irq(&worker->lock);
+
+			work->func(work);
+
+			atomic_dec(&worker->num_pending);
+			/*
+			 * unless this is an ordered work queue,
+			 * 'work' was probably freed by func above.
+			 */
+			run_ordered_completions(worker->workers, work);
+
+			spin_lock_irq(&worker->lock);
+			check_idle_worker(worker);
+
+		}
+		worker->working = 0;
+		if (freezing(current)) {
+			refrigerator();
+		} else {
+			set_current_state(TASK_INTERRUPTIBLE);
+			spin_unlock_irq(&worker->lock);
+			if (!kthread_should_stop())
+				schedule();
+			__set_current_state(TASK_RUNNING);
+		}
+	} while (!kthread_should_stop());
+	return 0;
+}
+
+/*
+ * this will wait for all the worker threads to shutdown
+ */
+int btrfs_stop_workers(struct btrfs_workers *workers)
+{
+	struct list_head *cur;
+	struct btrfs_worker_thread *worker;
+
+	list_splice_init(&workers->idle_list, &workers->worker_list);
+	while (!list_empty(&workers->worker_list)) {
+		cur = workers->worker_list.next;
+		worker = list_entry(cur, struct btrfs_worker_thread,
+				    worker_list);
+		kthread_stop(worker->task);
+		list_del(&worker->worker_list);
+		kfree(worker);
+	}
+	return 0;
+}
+
+/*
+ * simple init on struct btrfs_workers
+ */
+void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
+{
+	workers->num_workers = 0;
+	INIT_LIST_HEAD(&workers->worker_list);
+	INIT_LIST_HEAD(&workers->idle_list);
+	INIT_LIST_HEAD(&workers->order_list);
+	spin_lock_init(&workers->lock);
+	workers->max_workers = max;
+	workers->idle_thresh = 32;
+	workers->name = name;
+	workers->ordered = 0;
+}
+
+/*
+ * starts new worker threads.  This does not enforce the max worker
+ * count in case you need to temporarily go past it.
+ */
+int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
+{
+	struct btrfs_worker_thread *worker;
+	int ret = 0;
+	int i;
+
+	for (i = 0; i < num_workers; i++) {
+		worker = kzalloc(sizeof(*worker), GFP_NOFS);
+		if (!worker) {
+			ret = -ENOMEM;
+			goto fail;
+		}
+
+		INIT_LIST_HEAD(&worker->pending);
+		INIT_LIST_HEAD(&worker->worker_list);
+		spin_lock_init(&worker->lock);
+		atomic_set(&worker->num_pending, 0);
+		worker->task = kthread_run(worker_loop, worker,
+					   "btrfs-%s-%d", workers->name,
+					   workers->num_workers + i);
+		worker->workers = workers;
+		if (IS_ERR(worker->task)) {
+			kfree(worker);
+			ret = PTR_ERR(worker->task);
+			goto fail;
+		}
+
+		spin_lock_irq(&workers->lock);
+		list_add_tail(&worker->worker_list, &workers->idle_list);
+		worker->idle = 1;
+		workers->num_workers++;
+		spin_unlock_irq(&workers->lock);
+	}
+	return 0;
+fail:
+	btrfs_stop_workers(workers);
+	return ret;
+}
+
+/*
+ * run through the list and find a worker thread that doesn't have a lot
+ * to do right now.  This can return null if we aren't yet at the thread
+ * count limit and all of the threads are busy.
+ */
+static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
+{
+	struct btrfs_worker_thread *worker;
+	struct list_head *next;
+	int enforce_min = workers->num_workers < workers->max_workers;
+
+	/*
+	 * if we find an idle thread, don't move it to the end of the
+	 * idle list.  This improves the chance that the next submission
+	 * will reuse the same thread, and maybe catch it while it is still
+	 * working
+	 */
+	if (!list_empty(&workers->idle_list)) {
+		next = workers->idle_list.next;
+		worker = list_entry(next, struct btrfs_worker_thread,
+				    worker_list);
+		return worker;
+	}
+	if (enforce_min || list_empty(&workers->worker_list))
+		return NULL;
+
+	/*
+	 * if we pick a busy task, move the task to the end of the list.
+	 * hopefully this will keep things somewhat evenly balanced.
+	 * Do the move in batches based on the sequence number.  This groups
+	 * requests submitted at roughly the same time onto the same worker.
+	 */
+	next = workers->worker_list.next;
+	worker = list_entry(next, struct btrfs_worker_thread, worker_list);
+	atomic_inc(&worker->num_pending);
+	worker->sequence++;
+
+	if (worker->sequence % workers->idle_thresh == 0)
+		list_move_tail(next, &workers->worker_list);
+	return worker;
+}
+
+/*
+ * selects a worker thread to take the next job.  This will either find
+ * an idle worker, start a new worker up to the max count, or just return
+ * one of the existing busy workers.
+ */
+static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
+{
+	struct btrfs_worker_thread *worker;
+	unsigned long flags;
+
+again:
+	spin_lock_irqsave(&workers->lock, flags);
+	worker = next_worker(workers);
+	spin_unlock_irqrestore(&workers->lock, flags);
+
+	if (!worker) {
+		spin_lock_irqsave(&workers->lock, flags);
+		if (workers->num_workers >= workers->max_workers) {
+			struct list_head *fallback = NULL;
+			/*
+			 * we have failed to find any workers, just
+			 * return the force one
+			 */
+			if (!list_empty(&workers->worker_list))
+				fallback = workers->worker_list.next;
+			if (!list_empty(&workers->idle_list))
+				fallback = workers->idle_list.next;
+			BUG_ON(!fallback);
+			worker = list_entry(fallback,
+				  struct btrfs_worker_thread, worker_list);
+			spin_unlock_irqrestore(&workers->lock, flags);
+		} else {
+			spin_unlock_irqrestore(&workers->lock, flags);
+			/* we're below the limit, start another worker */
+			btrfs_start_workers(workers, 1);
+			goto again;
+		}
+	}
+	return worker;
+}
+
+/*
+ * btrfs_requeue_work just puts the work item back on the tail of the list
+ * it was taken from.  It is intended for use with long running work functions
+ * that make some progress and want to give the cpu up for others.
+ */
+int btrfs_requeue_work(struct btrfs_work *work)
+{
+	struct btrfs_worker_thread *worker = work->worker;
+	unsigned long flags;
+
+	if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags))
+		goto out;
+
+	spin_lock_irqsave(&worker->lock, flags);
+	atomic_inc(&worker->num_pending);
+	list_add_tail(&work->list, &worker->pending);
+
+	/* by definition we're busy, take ourselves off the idle
+	 * list
+	 */
+	if (worker->idle) {
+		spin_lock_irqsave(&worker->workers->lock, flags);
+		worker->idle = 0;
+		list_move_tail(&worker->worker_list,
+			       &worker->workers->worker_list);
+		spin_unlock_irqrestore(&worker->workers->lock, flags);
+	}
+
+	spin_unlock_irqrestore(&worker->lock, flags);
+
+out:
+	return 0;
+}
+
+/*
+ * places a struct btrfs_work into the pending queue of one of the kthreads
+ */
+int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
+{
+	struct btrfs_worker_thread *worker;
+	unsigned long flags;
+	int wake = 0;
+
+	/* don't requeue something already on a list */
+	if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags))
+		goto out;
+
+	worker = find_worker(workers);
+	if (workers->ordered) {
+		spin_lock_irqsave(&workers->lock, flags);
+		list_add_tail(&work->order_list, &workers->order_list);
+		spin_unlock_irqrestore(&workers->lock, flags);
+	} else {
+		INIT_LIST_HEAD(&work->order_list);
+	}
+
+	spin_lock_irqsave(&worker->lock, flags);
+	atomic_inc(&worker->num_pending);
+	check_busy_worker(worker);
+	list_add_tail(&work->list, &worker->pending);
+
+	/*
+	 * avoid calling into wake_up_process if this thread has already
+	 * been kicked
+	 */
+	if (!worker->working)
+		wake = 1;
+	worker->working = 1;
+
+	spin_unlock_irqrestore(&worker->lock, flags);
+
+	if (wake)
+		wake_up_process(worker->task);
+out:
+	return 0;
+}
diff --git a/fs/btrfs/async-thread.h b/fs/btrfs/async-thread.h
new file mode 100644
index 0000000..31be4ed
--- /dev/null
+++ b/fs/btrfs/async-thread.h
@@ -0,0 +1,101 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_ASYNC_THREAD_
+#define __BTRFS_ASYNC_THREAD_
+
+struct btrfs_worker_thread;
+
+/*
+ * This is similar to a workqueue, but it is meant to spread the operations
+ * across all available cpus instead of just the CPU that was used to
+ * queue the work.  There is also some batching introduced to try and
+ * cut down on context switches.
+ *
+ * By default threads are added on demand up to 2 * the number of cpus.
+ * Changing struct btrfs_workers->max_workers is one way to prevent
+ * demand creation of kthreads.
+ *
+ * the basic model of these worker threads is to embed a btrfs_work
+ * structure in your own data struct, and use container_of in a
+ * work function to get back to your data struct.
+ */
+struct btrfs_work {
+	/*
+	 * func should be set to the function you want called
+	 * your work struct is passed as the only arg
+	 *
+	 * ordered_func must be set for work sent to an ordered work queue,
+	 * and it is called to complete a given work item in the same
+	 * order they were sent to the queue.
+	 */
+	void (*func)(struct btrfs_work *work);
+	void (*ordered_func)(struct btrfs_work *work);
+	void (*ordered_free)(struct btrfs_work *work);
+
+	/*
+	 * flags should be set to zero.  It is used to make sure the
+	 * struct is only inserted once into the list.
+	 */
+	unsigned long flags;
+
+	/* don't touch these */
+	struct btrfs_worker_thread *worker;
+	struct list_head list;
+	struct list_head order_list;
+};
+
+struct btrfs_workers {
+	/* current number of running workers */
+	int num_workers;
+
+	/* max number of workers allowed.  changed by btrfs_start_workers */
+	int max_workers;
+
+	/* once a worker has this many requests or fewer, it is idle */
+	int idle_thresh;
+
+	/* force completions in the order they were queued */
+	int ordered;
+
+	/* list with all the work threads.  The workers on the idle thread
+	 * may be actively servicing jobs, but they haven't yet hit the
+	 * idle thresh limit above.
+	 */
+	struct list_head worker_list;
+	struct list_head idle_list;
+
+	/*
+	 * when operating in ordered mode, this maintains the list
+	 * of work items waiting for completion
+	 */
+	struct list_head order_list;
+
+	/* lock for finding the next worker thread to queue on */
+	spinlock_t lock;
+
+	/* extra name for this worker, used for current->name */
+	char *name;
+};
+
+int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work);
+int btrfs_start_workers(struct btrfs_workers *workers, int num_workers);
+int btrfs_stop_workers(struct btrfs_workers *workers);
+void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max);
+int btrfs_requeue_work(struct btrfs_work *work);
+#endif
-- 
1.6.0.2

[PATCH 18.1/35] Btrfs inode operations

[Chris Mason]

The basic inode and directory inode operations live here.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- /dev/null	2008-06-10 15:56:50.000000000 -0400
+++ b/fs/btrfs/inode.c	2009-01-08 09:43:29.596585385 -0500
@@ -0,0 +1,2907 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/kernel.h>
+#include <linux/bio.h>
+#include <linux/buffer_head.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/highmem.h>
+#include <linux/time.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/smp_lock.h>
+#include <linux/backing-dev.h>
+#include <linux/mpage.h>
+#include <linux/swap.h>
+#include <linux/writeback.h>
+#include <linux/statfs.h>
+#include <linux/compat.h>
+#include <linux/bit_spinlock.h>
+#include <linux/version.h>
+#include <linux/xattr.h>
+#include <linux/posix_acl.h>
+#include <linux/falloc.h>
+#include "compat.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "btrfs_inode.h"
+#include "ioctl.h"
+#include "print-tree.h"
+#include "volumes.h"
+#include "ordered-data.h"
+#include "xattr.h"
+#include "tree-log.h"
+#include "ref-cache.h"
+#include "compression.h"
+
+struct btrfs_iget_args {
+	u64 ino;
+	struct btrfs_root *root;
+};
+
+static struct inode_operations btrfs_dir_inode_operations;
+static struct inode_operations btrfs_symlink_inode_operations;
+static struct inode_operations btrfs_dir_ro_inode_operations;
+static struct inode_operations btrfs_special_inode_operations;
+static struct inode_operations btrfs_file_inode_operations;
+static struct address_space_operations btrfs_aops;
+static struct address_space_operations btrfs_symlink_aops;
+static struct file_operations btrfs_dir_file_operations;
+static struct extent_io_ops btrfs_extent_io_ops;
+
+static struct kmem_cache *btrfs_inode_cachep;
+struct kmem_cache *btrfs_trans_handle_cachep;
+struct kmem_cache *btrfs_transaction_cachep;
+struct kmem_cache *btrfs_bit_radix_cachep;
+struct kmem_cache *btrfs_path_cachep;
+
+#define S_SHIFT 12
+static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
+	[S_IFREG >> S_SHIFT]	= BTRFS_FT_REG_FILE,
+	[S_IFDIR >> S_SHIFT]	= BTRFS_FT_DIR,
+	[S_IFCHR >> S_SHIFT]	= BTRFS_FT_CHRDEV,
+	[S_IFBLK >> S_SHIFT]	= BTRFS_FT_BLKDEV,
+	[S_IFIFO >> S_SHIFT]	= BTRFS_FT_FIFO,
+	[S_IFSOCK >> S_SHIFT]	= BTRFS_FT_SOCK,
+	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
+};
+
+static void btrfs_truncate(struct inode *inode);
+static noinline int cow_file_range(struct inode *inode,
+				   struct page *locked_page,
+				   u64 start, u64 end, int *page_started,
+				   unsigned long *nr_written, int unlock);
+
+/*
+ * a very lame attempt at stopping writes when the FS is 85% full.  There
+ * are countless ways this is incorrect, but it is better than nothing.
+ */
+int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
+			   int for_del)
+{
+	u64 total;
+	u64 used;
+	u64 thresh;
+	int ret = 0;
+
+	spin_lock(&root->fs_info->delalloc_lock);
+	total = btrfs_super_total_bytes(&root->fs_info->super_copy);
+	used = btrfs_super_bytes_used(&root->fs_info->super_copy);
+	if (for_del)
+		thresh = total * 90;
+	else
+		thresh = total * 85;
+
+	do_div(thresh, 100);
+
+	if (used + root->fs_info->delalloc_bytes + num_required > thresh)
+		ret = -ENOSPC;
+	spin_unlock(&root->fs_info->delalloc_lock);
+	return ret;
+}
+
+static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
+				       struct inode *inode, u64 file_pos,
+				       u64 disk_bytenr, u64 disk_num_bytes,
+				       u64 num_bytes, u64 ram_bytes,
+				       u8 compression, u8 encryption,
+				       u16 other_encoding, int extent_type)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_key ins;
+	u64 hint;
+	int ret;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	ret = btrfs_drop_extents(trans, root, inode, file_pos,
+				 file_pos + num_bytes, file_pos, &hint);
+	BUG_ON(ret);
+
+	ins.objectid = inode->i_ino;
+	ins.offset = file_pos;
+	ins.type = BTRFS_EXTENT_DATA_KEY;
+	ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
+	BUG_ON(ret);
+	leaf = path->nodes[0];
+	fi = btrfs_item_ptr(leaf, path->slots[0],
+			    struct btrfs_file_extent_item);
+	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
+	btrfs_set_file_extent_type(leaf, fi, extent_type);
+	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
+	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
+	btrfs_set_file_extent_offset(leaf, fi, 0);
+	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
+	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
+	btrfs_set_file_extent_compression(leaf, fi, compression);
+	btrfs_set_file_extent_encryption(leaf, fi, encryption);
+	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
+	btrfs_mark_buffer_dirty(leaf);
+
+	inode_add_bytes(inode, num_bytes);
+	btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
+
+	ins.objectid = disk_bytenr;
+	ins.offset = disk_num_bytes;
+	ins.type = BTRFS_EXTENT_ITEM_KEY;
+	ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
+					  root->root_key.objectid,
+					  trans->transid, inode->i_ino, &ins);
+	BUG_ON(ret);
+
+	btrfs_free_path(path);
+	return 0;
+}
+
+/*
+ * This creates an orphan entry for the given inode in case something goes
+ * wrong in the middle of an unlink/truncate.
+ */
+int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret = 0;
+
+	spin_lock(&root->list_lock);
+
+	/* already on the orphan list, we're good */
+	if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
+		spin_unlock(&root->list_lock);
+		return 0;
+	}
+
+	list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
+
+	spin_unlock(&root->list_lock);
+
+	/*
+	 * insert an orphan item to track this unlinked/truncated file
+	 */
+	ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
+
+	return ret;
+}
+
+/*
+ * We have done the truncate/delete so we can go ahead and remove the orphan
+ * item for this particular inode.
+ */
+int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret = 0;
+
+	spin_lock(&root->list_lock);
+
+	if (list_empty(&BTRFS_I(inode)->i_orphan)) {
+		spin_unlock(&root->list_lock);
+		return 0;
+	}
+
+	list_del_init(&BTRFS_I(inode)->i_orphan);
+	if (!trans) {
+		spin_unlock(&root->list_lock);
+		return 0;
+	}
+
+	spin_unlock(&root->list_lock);
+
+	ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
+
+	return ret;
+}
+
+/*
+ * this cleans up any orphans that may be left on the list from the last use
+ * of this root.
+ */
+void btrfs_orphan_cleanup(struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	struct btrfs_key key, found_key;
+	struct btrfs_trans_handle *trans;
+	struct inode *inode;
+	int ret = 0, nr_unlink = 0, nr_truncate = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return;
+	path->reada = -1;
+
+	key.objectid = BTRFS_ORPHAN_OBJECTID;
+	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
+	key.offset = (u64)-1;
+
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+		if (ret < 0) {
+			printk(KERN_ERR "Error searching slot for orphan: %d"
+			       "\n", ret);
+			break;
+		}
+
+		/*
+		 * if ret == 0 means we found what we were searching for, which
+		 * is weird, but possible, so only screw with path if we didnt
+		 * find the key and see if we have stuff that matches
+		 */
+		if (ret > 0) {
+			if (path->slots[0] == 0)
+				break;
+			path->slots[0]--;
+		}
+
+		/* pull out the item */
+		leaf = path->nodes[0];
+		item = btrfs_item_nr(leaf, path->slots[0]);
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+		/* make sure the item matches what we want */
+		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
+			break;
+		if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
+			break;
+
+		/* release the path since we're done with it */
+		btrfs_release_path(root, path);
+
+		/*
+		 * this is where we are basically btrfs_lookup, without the
+		 * crossing root thing.  we store the inode number in the
+		 * offset of the orphan item.
+		 */
+		inode = btrfs_iget_locked(root->fs_info->sb,
+					  found_key.offset, root);
+		if (!inode)
+			break;
+
+		if (inode->i_state & I_NEW) {
+			BTRFS_I(inode)->root = root;
+
+			/* have to set the location manually */
+			BTRFS_I(inode)->location.objectid = inode->i_ino;
+			BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
+			BTRFS_I(inode)->location.offset = 0;
+
+			btrfs_read_locked_inode(inode);
+			unlock_new_inode(inode);
+		}
+
+		/*
+		 * add this inode to the orphan list so btrfs_orphan_del does
+		 * the proper thing when we hit it
+		 */
+		spin_lock(&root->list_lock);
+		list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
+		spin_unlock(&root->list_lock);
+
+		/*
+		 * if this is a bad inode, means we actually succeeded in
+		 * removing the inode, but not the orphan record, which means
+		 * we need to manually delete the orphan since iput will just
+		 * do a destroy_inode
+		 */
+		if (is_bad_inode(inode)) {
+			trans = btrfs_start_transaction(root, 1);
+			btrfs_orphan_del(trans, inode);
+			btrfs_end_transaction(trans, root);
+			iput(inode);
+			continue;
+		}
+
+		/* if we have links, this was a truncate, lets do that */
+		if (inode->i_nlink) {
+			nr_truncate++;
+			btrfs_truncate(inode);
+		} else {
+			nr_unlink++;
+		}
+
+		/* this will do delete_inode and everything for us */
+		iput(inode);
+	}
+
+	if (nr_unlink)
+		printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
+	if (nr_truncate)
+		printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
+
+	btrfs_free_path(path);
+}
+
+/*
+ * read an inode from the btree into the in-memory inode
+ */
+void btrfs_read_locked_inode(struct inode *inode)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_inode_item *inode_item;
+	struct btrfs_timespec *tspec;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_key location;
+	u64 alloc_group_block;
+	u32 rdev;
+	int ret;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
+
+	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
+	if (ret)
+		goto make_bad;
+
+	leaf = path->nodes[0];
+	inode_item = btrfs_item_ptr(leaf, path->slots[0],
+				    struct btrfs_inode_item);
+
+	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
+	inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
+	inode->i_uid = btrfs_inode_uid(leaf, inode_item);
+	inode->i_gid = btrfs_inode_gid(leaf, inode_item);
+	btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
+
+	tspec = btrfs_inode_atime(inode_item);
+	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
+	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
+
+	tspec = btrfs_inode_mtime(inode_item);
+	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
+	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
+
+	tspec = btrfs_inode_ctime(inode_item);
+	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
+	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
+
+	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
+	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
+	BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
+	inode->i_generation = BTRFS_I(inode)->generation;
+	inode->i_rdev = 0;
+	rdev = btrfs_inode_rdev(leaf, inode_item);
+
+	BTRFS_I(inode)->index_cnt = (u64)-1;
+	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
+
+	alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
+	BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
+						alloc_group_block, 0);
+	btrfs_free_path(path);
+	inode_item = NULL;
+
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFREG:
+		inode->i_mapping->a_ops = &btrfs_aops;
+		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
+		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
+		inode->i_fop = &btrfs_file_operations;
+		inode->i_op = &btrfs_file_inode_operations;
+		break;
+	case S_IFDIR:
+		inode->i_fop = &btrfs_dir_file_operations;
+		if (root == root->fs_info->tree_root)
+			inode->i_op = &btrfs_dir_ro_inode_operations;
+		else
+			inode->i_op = &btrfs_dir_inode_operations;
+		break;
+	case S_IFLNK:
+		inode->i_op = &btrfs_symlink_inode_operations;
+		inode->i_mapping->a_ops = &btrfs_symlink_aops;
+		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
+		break;
+	default:
+		init_special_inode(inode, inode->i_mode, rdev);
+		break;
+	}
+	return;
+
+make_bad:
+	btrfs_free_path(path);
+	make_bad_inode(inode);
+}
+
+/*
+ * given a leaf and an inode, copy the inode fields into the leaf
+ */
+static void fill_inode_item(struct btrfs_trans_handle *trans,
+			    struct extent_buffer *leaf,
+			    struct btrfs_inode_item *item,
+			    struct inode *inode)
+{
+	btrfs_set_inode_uid(leaf, item, inode->i_uid);
+	btrfs_set_inode_gid(leaf, item, inode->i_gid);
+	btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
+	btrfs_set_inode_mode(leaf, item, inode->i_mode);
+	btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
+
+	btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
+			       inode->i_atime.tv_sec);
+	btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
+				inode->i_atime.tv_nsec);
+
+	btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
+			       inode->i_mtime.tv_sec);
+	btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
+				inode->i_mtime.tv_nsec);
+
+	btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
+			       inode->i_ctime.tv_sec);
+	btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
+				inode->i_ctime.tv_nsec);
+
+	btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
+	btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
+	btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
+	btrfs_set_inode_transid(leaf, item, trans->transid);
+	btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
+	btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
+	btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
+}
+
+/*
+ * copy everything in the in-memory inode into the btree.
+ */
+noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root, struct inode *inode)
+{
+	struct btrfs_inode_item *inode_item;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	int ret;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_lookup_inode(trans, root, path,
+				 &BTRFS_I(inode)->location, 1);
+	if (ret) {
+		if (ret > 0)
+			ret = -ENOENT;
+		goto failed;
+	}
+
+	leaf = path->nodes[0];
+	inode_item = btrfs_item_ptr(leaf, path->slots[0],
+				  struct btrfs_inode_item);
+
+	fill_inode_item(trans, leaf, inode_item, inode);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_set_inode_last_trans(trans, inode);
+	ret = 0;
+failed:
+	btrfs_free_path(path);
+	return ret;
+}
+
+
+/*
+ * unlink helper that gets used here in inode.c and in the tree logging
+ * recovery code.  It remove a link in a directory with a given name, and
+ * also drops the back refs in the inode to the directory
+ */
+int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root,
+		       struct inode *dir, struct inode *inode,
+		       const char *name, int name_len)
+{
+	struct btrfs_path *path;
+	int ret = 0;
+	struct extent_buffer *leaf;
+	struct btrfs_dir_item *di;
+	struct btrfs_key key;
+	u64 index;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
+				    name, name_len, -1);
+	if (IS_ERR(di)) {
+		ret = PTR_ERR(di);
+		goto err;
+	}
+	if (!di) {
+		ret = -ENOENT;
+		goto err;
+	}
+	leaf = path->nodes[0];
+	btrfs_dir_item_key_to_cpu(leaf, di, &key);
+	ret = btrfs_delete_one_dir_name(trans, root, path, di);
+	if (ret)
+		goto err;
+	btrfs_release_path(root, path);
+
+	ret = btrfs_del_inode_ref(trans, root, name, name_len,
+				  inode->i_ino,
+				  dir->i_ino, &index);
+	if (ret) {
+		printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
+		       "inode %lu parent %lu\n", name_len, name,
+		       inode->i_ino, dir->i_ino);
+		goto err;
+	}
+
+	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
+					 index, name, name_len, -1);
+	if (IS_ERR(di)) {
+		ret = PTR_ERR(di);
+		goto err;
+	}
+	if (!di) {
+		ret = -ENOENT;
+		goto err;
+	}
+	ret = btrfs_delete_one_dir_name(trans, root, path, di);
+	btrfs_release_path(root, path);
+
+	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
+					 inode, dir->i_ino);
+	BUG_ON(ret != 0 && ret != -ENOENT);
+	if (ret != -ENOENT)
+		BTRFS_I(dir)->log_dirty_trans = trans->transid;
+
+	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
+					   dir, index);
+	BUG_ON(ret);
+err:
+	btrfs_free_path(path);
+	if (ret)
+		goto out;
+
+	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
+	inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+	btrfs_update_inode(trans, root, dir);
+	btrfs_drop_nlink(inode);
+	ret = btrfs_update_inode(trans, root, inode);
+	dir->i_sb->s_dirt = 1;
+out:
+	return ret;
+}
+
+static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
+{
+	struct btrfs_root *root;
+	struct btrfs_trans_handle *trans;
+	struct inode *inode = dentry->d_inode;
+	int ret;
+	unsigned long nr = 0;
+
+	root = BTRFS_I(dir)->root;
+
+	ret = btrfs_check_free_space(root, 1, 1);
+	if (ret)
+		goto fail;
+
+	trans = btrfs_start_transaction(root, 1);
+
+	btrfs_set_trans_block_group(trans, dir);
+	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
+				 dentry->d_name.name, dentry->d_name.len);
+
+	if (inode->i_nlink == 0)
+		ret = btrfs_orphan_add(trans, inode);
+
+	nr = trans->blocks_used;
+
+	btrfs_end_transaction_throttle(trans, root);
+fail:
+	btrfs_btree_balance_dirty(root, nr);
+	return ret;
+}
+
+static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
+{
+	struct inode *inode = dentry->d_inode;
+	int err = 0;
+	int ret;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct btrfs_trans_handle *trans;
+	unsigned long nr = 0;
+
+	/*
+	 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
+	 * the root of a subvolume or snapshot
+	 */
+	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
+	    inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
+		return -ENOTEMPTY;
+	}
+
+	ret = btrfs_check_free_space(root, 1, 1);
+	if (ret)
+		goto fail;
+
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
+
+	err = btrfs_orphan_add(trans, inode);
+	if (err)
+		goto fail_trans;
+
+	/* now the directory is empty */
+	err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
+				 dentry->d_name.name, dentry->d_name.len);
+	if (!err)
+		btrfs_i_size_write(inode, 0);
+
+fail_trans:
+	nr = trans->blocks_used;
+	ret = btrfs_end_transaction_throttle(trans, root);
+fail:
+	btrfs_btree_balance_dirty(root, nr);
+
+	if (ret && !err)
+		err = ret;
+	return err;
+}
+
+/*
+ * this can truncate away extent items, csum items and directory items.
+ * It starts at a high offset and removes keys until it can't find
+ * any higher than new_size
+ *
+ * csum items that cross the new i_size are truncated to the new size
+ * as well.
+ *
+ * min_type is the minimum key type to truncate down to.  If set to 0, this
+ * will kill all the items on this inode, including the INODE_ITEM_KEY.
+ */
+noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct inode *inode,
+					u64 new_size, u32 min_type)
+{
+	int ret;
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	u32 found_type;
+	struct extent_buffer *leaf;
+	struct btrfs_file_extent_item *fi;
+	u64 extent_start = 0;
+	u64 extent_num_bytes = 0;
+	u64 item_end = 0;
+	u64 root_gen = 0;
+	u64 root_owner = 0;
+	int found_extent;
+	int del_item;
+	int pending_del_nr = 0;
+	int pending_del_slot = 0;
+	int extent_type = -1;
+	int encoding;
+	u64 mask = root->sectorsize - 1;
+
+	if (root->ref_cows)
+		btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
+	path = btrfs_alloc_path();
+	path->reada = -1;
+	BUG_ON(!path);
+
+	/* FIXME, add redo link to tree so we don't leak on crash */
+	key.objectid = inode->i_ino;
+	key.offset = (u64)-1;
+	key.type = (u8)-1;
+
+	btrfs_init_path(path);
+
+search_again:
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret < 0)
+		goto error;
+
+	if (ret > 0) {
+		/* there are no items in the tree for us to truncate, we're
+		 * done
+		 */
+		if (path->slots[0] == 0) {
+			ret = 0;
+			goto error;
+		}
+		path->slots[0]--;
+	}
+
+	while (1) {
+		fi = NULL;
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		found_type = btrfs_key_type(&found_key);
+		encoding = 0;
+
+		if (found_key.objectid != inode->i_ino)
+			break;
+
+		if (found_type < min_type)
+			break;
+
+		item_end = found_key.offset;
+		if (found_type == BTRFS_EXTENT_DATA_KEY) {
+			fi = btrfs_item_ptr(leaf, path->slots[0],
+					    struct btrfs_file_extent_item);
+			extent_type = btrfs_file_extent_type(leaf, fi);
+			encoding = btrfs_file_extent_compression(leaf, fi);
+			encoding |= btrfs_file_extent_encryption(leaf, fi);
+			encoding |= btrfs_file_extent_other_encoding(leaf, fi);
+
+			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
+				item_end +=
+				    btrfs_file_extent_num_bytes(leaf, fi);
+			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+				item_end += btrfs_file_extent_inline_len(leaf,
+									 fi);
+			}
+			item_end--;
+		}
+		if (item_end < new_size) {
+			if (found_type == BTRFS_DIR_ITEM_KEY)
+				found_type = BTRFS_INODE_ITEM_KEY;
+			else if (found_type == BTRFS_EXTENT_ITEM_KEY)
+				found_type = BTRFS_EXTENT_DATA_KEY;
+			else if (found_type == BTRFS_EXTENT_DATA_KEY)
+				found_type = BTRFS_XATTR_ITEM_KEY;
+			else if (found_type == BTRFS_XATTR_ITEM_KEY)
+				found_type = BTRFS_INODE_REF_KEY;
+			else if (found_type)
+				found_type--;
+			else
+				break;
+			btrfs_set_key_type(&key, found_type);
+			goto next;
+		}
+		if (found_key.offset >= new_size)
+			del_item = 1;
+		else
+			del_item = 0;
+		found_extent = 0;
+
+		/* FIXME, shrink the extent if the ref count is only 1 */
+		if (found_type != BTRFS_EXTENT_DATA_KEY)
+			goto delete;
+
+		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
+			u64 num_dec;
+			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
+			if (!del_item && !encoding) {
+				u64 orig_num_bytes =
+					btrfs_file_extent_num_bytes(leaf, fi);
+				extent_num_bytes = new_size -
+					found_key.offset + root->sectorsize - 1;
+				extent_num_bytes = extent_num_bytes &
+					~((u64)root->sectorsize - 1);
+				btrfs_set_file_extent_num_bytes(leaf, fi,
+							 extent_num_bytes);
+				num_dec = (orig_num_bytes -
+					   extent_num_bytes);
+				if (root->ref_cows && extent_start != 0)
+					inode_sub_bytes(inode, num_dec);
+				btrfs_mark_buffer_dirty(leaf);
+			} else {
+				extent_num_bytes =
+					btrfs_file_extent_disk_num_bytes(leaf,
+									 fi);
+				/* FIXME blocksize != 4096 */
+				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
+				if (extent_start != 0) {
+					found_extent = 1;
+					if (root->ref_cows)
+						inode_sub_bytes(inode, num_dec);
+				}
+				root_gen = btrfs_header_generation(leaf);
+				root_owner = btrfs_header_owner(leaf);
+			}
+		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+			/*
+			 * we can't truncate inline items that have had
+			 * special encodings
+			 */
+			if (!del_item &&
+			    btrfs_file_extent_compression(leaf, fi) == 0 &&
+			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
+			    btrfs_file_extent_other_encoding(leaf, fi) == 0) {
+				u32 size = new_size - found_key.offset;
+
+				if (root->ref_cows) {
+					inode_sub_bytes(inode, item_end + 1 -
+							new_size);
+				}
+				size =
+				    btrfs_file_extent_calc_inline_size(size);
+				ret = btrfs_truncate_item(trans, root, path,
+							  size, 1);
+				BUG_ON(ret);
+			} else if (root->ref_cows) {
+				inode_sub_bytes(inode, item_end + 1 -
+						found_key.offset);
+			}
+		}
+delete:
+		if (del_item) {
+			if (!pending_del_nr) {
+				/* no pending yet, add ourselves */
+				pending_del_slot = path->slots[0];
+				pending_del_nr = 1;
+			} else if (pending_del_nr &&
+				   path->slots[0] + 1 == pending_del_slot) {
+				/* hop on the pending chunk */
+				pending_del_nr++;
+				pending_del_slot = path->slots[0];
+			} else {
+				BUG();
+			}
+		} else {
+			break;
+		}
+		if (found_extent) {
+			ret = btrfs_free_extent(trans, root, extent_start,
+						extent_num_bytes,
+						leaf->start, root_owner,
+						root_gen, inode->i_ino, 0);
+			BUG_ON(ret);
+		}
+next:
+		if (path->slots[0] == 0) {
+			if (pending_del_nr)
+				goto del_pending;
+			btrfs_release_path(root, path);
+			goto search_again;
+		}
+
+		path->slots[0]--;
+		if (pending_del_nr &&
+		    path->slots[0] + 1 != pending_del_slot) {
+			struct btrfs_key debug;
+del_pending:
+			btrfs_item_key_to_cpu(path->nodes[0], &debug,
+					      pending_del_slot);
+			ret = btrfs_del_items(trans, root, path,
+					      pending_del_slot,
+					      pending_del_nr);
+			BUG_ON(ret);
+			pending_del_nr = 0;
+			btrfs_release_path(root, path);
+			goto search_again;
+		}
+	}
+	ret = 0;
+error:
+	if (pending_del_nr) {
+		ret = btrfs_del_items(trans, root, path, pending_del_slot,
+				      pending_del_nr);
+	}
+	btrfs_free_path(path);
+	inode->i_sb->s_dirt = 1;
+	return ret;
+}
+
+/*
+ * taken from block_truncate_page, but does cow as it zeros out
+ * any bytes left in the last page in the file.
+ */
+static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
+{
+	struct inode *inode = mapping->host;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_ordered_extent *ordered;
+	char *kaddr;
+	u32 blocksize = root->sectorsize;
+	pgoff_t index = from >> PAGE_CACHE_SHIFT;
+	unsigned offset = from & (PAGE_CACHE_SIZE-1);
+	struct page *page;
+	int ret = 0;
+	u64 page_start;
+	u64 page_end;
+
+	if ((offset & (blocksize - 1)) == 0)
+		goto out;
+
+	ret = -ENOMEM;
+again:
+	page = grab_cache_page(mapping, index);
+	if (!page)
+		goto out;
+
+	page_start = page_offset(page);
+	page_end = page_start + PAGE_CACHE_SIZE - 1;
+
+	if (!PageUptodate(page)) {
+		ret = btrfs_readpage(NULL, page);
+		lock_page(page);
+		if (page->mapping != mapping) {
+			unlock_page(page);
+			page_cache_release(page);
+			goto again;
+		}
+		if (!PageUptodate(page)) {
+			ret = -EIO;
+			goto out_unlock;
+		}
+	}
+	wait_on_page_writeback(page);
+
+	lock_extent(io_tree, page_start, page_end, GFP_NOFS);
+	set_page_extent_mapped(page);
+
+	ordered = btrfs_lookup_ordered_extent(inode, page_start);
+	if (ordered) {
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+		unlock_page(page);
+		page_cache_release(page);
+		btrfs_start_ordered_extent(inode, ordered, 1);
+		btrfs_put_ordered_extent(ordered);
+		goto again;
+	}
+
+	btrfs_set_extent_delalloc(inode, page_start, page_end);
+	ret = 0;
+	if (offset != PAGE_CACHE_SIZE) {
+		kaddr = kmap(page);
+		memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
+		flush_dcache_page(page);
+		kunmap(page);
+	}
+	ClearPageChecked(page);
+	set_page_dirty(page);
+	unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+
+out_unlock:
+	unlock_page(page);
+	page_cache_release(page);
+out:
+	return ret;
+}
+
+int btrfs_cont_expand(struct inode *inode, loff_t size)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct extent_map *em;
+	u64 mask = root->sectorsize - 1;
+	u64 hole_start = (inode->i_size + mask) & ~mask;
+	u64 block_end = (size + mask) & ~mask;
+	u64 last_byte;
+	u64 cur_offset;
+	u64 hole_size;
+	int err;
+
+	if (size <= hole_start)
+		return 0;
+
+	err = btrfs_check_free_space(root, 1, 0);
+	if (err)
+		return err;
+
+	btrfs_truncate_page(inode->i_mapping, inode->i_size);
+
+	while (1) {
+		struct btrfs_ordered_extent *ordered;
+		btrfs_wait_ordered_range(inode, hole_start,
+					 block_end - hole_start);
+		lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
+		ordered = btrfs_lookup_ordered_extent(inode, hole_start);
+		if (!ordered)
+			break;
+		unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
+		btrfs_put_ordered_extent(ordered);
+	}
+
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
+
+	cur_offset = hole_start;
+	while (1) {
+		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
+				block_end - cur_offset, 0);
+		BUG_ON(IS_ERR(em) || !em);
+		last_byte = min(extent_map_end(em), block_end);
+		last_byte = (last_byte + mask) & ~mask;
+		if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
+			u64 hint_byte = 0;
+			hole_size = last_byte - cur_offset;
+			err = btrfs_drop_extents(trans, root, inode,
+						 cur_offset,
+						 cur_offset + hole_size,
+						 cur_offset, &hint_byte);
+			if (err)
+				break;
+			err = btrfs_insert_file_extent(trans, root,
+					inode->i_ino, cur_offset, 0,
+					0, hole_size, 0, hole_size,
+					0, 0, 0);
+			btrfs_drop_extent_cache(inode, hole_start,
+					last_byte - 1, 0);
+		}
+		free_extent_map(em);
+		cur_offset = last_byte;
+		if (err || cur_offset >= block_end)
+			break;
+	}
+
+	btrfs_end_transaction(trans, root);
+	unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
+	return err;
+}
+
+static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
+{
+	struct inode *inode = dentry->d_inode;
+	int err;
+
+	err = inode_change_ok(inode, attr);
+	if (err)
+		return err;
+
+	if (S_ISREG(inode->i_mode) &&
+	    attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
+		err = btrfs_cont_expand(inode, attr->ia_size);
+		if (err)
+			return err;
+	}
+
+	err = inode_setattr(inode, attr);
+
+	if (!err && ((attr->ia_valid & ATTR_MODE)))
+		err = btrfs_acl_chmod(inode);
+	return err;
+}
+
+void btrfs_delete_inode(struct inode *inode)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	unsigned long nr;
+	int ret;
+
+	truncate_inode_pages(&inode->i_data, 0);
+	if (is_bad_inode(inode)) {
+		btrfs_orphan_del(NULL, inode);
+		goto no_delete;
+	}
+	btrfs_wait_ordered_range(inode, 0, (u64)-1);
+
+	btrfs_i_size_write(inode, 0);
+	trans = btrfs_join_transaction(root, 1);
+
+	btrfs_set_trans_block_group(trans, inode);
+	ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
+	if (ret) {
+		btrfs_orphan_del(NULL, inode);
+		goto no_delete_lock;
+	}
+
+	btrfs_orphan_del(trans, inode);
+
+	nr = trans->blocks_used;
+	clear_inode(inode);
+
+	btrfs_end_transaction(trans, root);
+	btrfs_btree_balance_dirty(root, nr);
+	return;
+
+no_delete_lock:
+	nr = trans->blocks_used;
+	btrfs_end_transaction(trans, root);
+	btrfs_btree_balance_dirty(root, nr);
+no_delete:
+	clear_inode(inode);
+}
+
+/*
+ * this returns the key found in the dir entry in the location pointer.
+ * If no dir entries were found, location->objectid is 0.
+ */
+static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
+			       struct btrfs_key *location)
+{
+	const char *name = dentry->d_name.name;
+	int namelen = dentry->d_name.len;
+	struct btrfs_dir_item *di;
+	struct btrfs_path *path;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	int ret = 0;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
+				    namelen, 0);
+	if (IS_ERR(di))
+		ret = PTR_ERR(di);
+
+	if (!di || IS_ERR(di))
+		goto out_err;
+
+	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
+out:
+	btrfs_free_path(path);
+	return ret;
+out_err:
+	location->objectid = 0;
+	goto out;
+}
+
+/*
+ * when we hit a tree root in a directory, the btrfs part of the inode
+ * needs to be changed to reflect the root directory of the tree root.  This
+ * is kind of like crossing a mount point.
+ */
+static int fixup_tree_root_location(struct btrfs_root *root,
+			     struct btrfs_key *location,
+			     struct btrfs_root **sub_root,
+			     struct dentry *dentry)
+{
+	struct btrfs_root_item *ri;
+
+	if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
+		return 0;
+	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
+		return 0;
+
+	*sub_root = btrfs_read_fs_root(root->fs_info, location,
+					dentry->d_name.name,
+					dentry->d_name.len);
+	if (IS_ERR(*sub_root))
+		return PTR_ERR(*sub_root);
+
+	ri = &(*sub_root)->root_item;
+	location->objectid = btrfs_root_dirid(ri);
+	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
+	location->offset = 0;
+
+	return 0;
+}
+
+static noinline void init_btrfs_i(struct inode *inode)
+{
+	struct btrfs_inode *bi = BTRFS_I(inode);
+
+	bi->i_acl = NULL;
+	bi->i_default_acl = NULL;
+
+	bi->generation = 0;
+	bi->sequence = 0;
+	bi->last_trans = 0;
+	bi->logged_trans = 0;
+	bi->delalloc_bytes = 0;
+	bi->disk_i_size = 0;
+	bi->flags = 0;
+	bi->index_cnt = (u64)-1;
+	bi->log_dirty_trans = 0;
+	extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
+	extent_io_tree_init(&BTRFS_I(inode)->io_tree,
+			     inode->i_mapping, GFP_NOFS);
+	extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
+			     inode->i_mapping, GFP_NOFS);
+	INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
+	btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
+	mutex_init(&BTRFS_I(inode)->extent_mutex);
+	mutex_init(&BTRFS_I(inode)->log_mutex);
+}
+
+static int btrfs_init_locked_inode(struct inode *inode, void *p)
+{
+	struct btrfs_iget_args *args = p;
+	inode->i_ino = args->ino;
+	init_btrfs_i(inode);
+	BTRFS_I(inode)->root = args->root;
+	return 0;
+}
+
+static int btrfs_find_actor(struct inode *inode, void *opaque)
+{
+	struct btrfs_iget_args *args = opaque;
+	return args->ino == inode->i_ino &&
+		args->root == BTRFS_I(inode)->root;
+}
+
+struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
+			    struct btrfs_root *root, int wait)
+{
+	struct inode *inode;
+	struct btrfs_iget_args args;
+	args.ino = objectid;
+	args.root = root;
+
+	if (wait) {
+		inode = ilookup5(s, objectid, btrfs_find_actor,
+				 (void *)&args);
+	} else {
+		inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
+					(void *)&args);
+	}
+	return inode;
+}
+
+struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
+				struct btrfs_root *root)
+{
+	struct inode *inode;
+	struct btrfs_iget_args args;
+	args.ino = objectid;
+	args.root = root;
+
+	inode = iget5_locked(s, objectid, btrfs_find_actor,
+			     btrfs_init_locked_inode,
+			     (void *)&args);
+	return inode;
+}
+
+/* Get an inode object given its location and corresponding root.
+ * Returns in *is_new if the inode was read from disk
+ */
+struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
+			 struct btrfs_root *root, int *is_new)
+{
+	struct inode *inode;
+
+	inode = btrfs_iget_locked(s, location->objectid, root);
+	if (!inode)
+		return ERR_PTR(-EACCES);
+
+	if (inode->i_state & I_NEW) {
+		BTRFS_I(inode)->root = root;
+		memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
+		btrfs_read_locked_inode(inode);
+		unlock_new_inode(inode);
+		if (is_new)
+			*is_new = 1;
+	} else {
+		if (is_new)
+			*is_new = 0;
+	}
+
+	return inode;
+}
+
+struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
+{
+	struct inode *inode;
+	struct btrfs_inode *bi = BTRFS_I(dir);
+	struct btrfs_root *root = bi->root;
+	struct btrfs_root *sub_root = root;
+	struct btrfs_key location;
+	int ret, new;
+
+	if (dentry->d_name.len > BTRFS_NAME_LEN)
+		return ERR_PTR(-ENAMETOOLONG);
+
+	ret = btrfs_inode_by_name(dir, dentry, &location);
+
+	if (ret < 0)
+		return ERR_PTR(ret);
+
+	inode = NULL;
+	if (location.objectid) {
+		ret = fixup_tree_root_location(root, &location, &sub_root,
+						dentry);
+		if (ret < 0)
+			return ERR_PTR(ret);
+		if (ret > 0)
+			return ERR_PTR(-ENOENT);
+		inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
+		if (IS_ERR(inode))
+			return ERR_CAST(inode);
+	}
+	return inode;
+}
+
+static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
+				   struct nameidata *nd)
+{
+	struct inode *inode;
+
+	if (dentry->d_name.len > BTRFS_NAME_LEN)
+		return ERR_PTR(-ENAMETOOLONG);
+
+	inode = btrfs_lookup_dentry(dir, dentry);
+	if (IS_ERR(inode))
+		return ERR_CAST(inode);
+
+	return d_splice_alias(inode, dentry);
+}
+
+static unsigned char btrfs_filetype_table[] = {
+	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
+};
+
+static int btrfs_real_readdir(struct file *filp, void *dirent,
+			      filldir_t filldir)
+{
+	struct inode *inode = filp->f_dentry->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_item *item;
+	struct btrfs_dir_item *di;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct btrfs_path *path;
+	int ret;
+	u32 nritems;
+	struct extent_buffer *leaf;
+	int slot;
+	int advance;
+	unsigned char d_type;
+	int over = 0;
+	u32 di_cur;
+	u32 di_total;
+	u32 di_len;
+	int key_type = BTRFS_DIR_INDEX_KEY;
+	char tmp_name[32];
+	char *name_ptr;
+	int name_len;
+
+	/* FIXME, use a real flag for deciding about the key type */
+	if (root->fs_info->tree_root == root)
+		key_type = BTRFS_DIR_ITEM_KEY;
+
+	/* special case for "." */
+	if (filp->f_pos == 0) {
+		over = filldir(dirent, ".", 1,
+			       1, inode->i_ino,
+			       DT_DIR);
+		if (over)
+			return 0;
+		filp->f_pos = 1;
+	}
+	/* special case for .., just use the back ref */
+	if (filp->f_pos == 1) {
+		u64 pino = parent_ino(filp->f_path.dentry);
+		over = filldir(dirent, "..", 2,
+			       2, pino, DT_DIR);
+		if (over)
+			return 0;
+		filp->f_pos = 2;
+	}
+	path = btrfs_alloc_path();
+	path->reada = 2;
+
+	btrfs_set_key_type(&key, key_type);
+	key.offset = filp->f_pos;
+	key.objectid = inode->i_ino;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto err;
+	advance = 0;
+
+	while (1) {
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		slot = path->slots[0];
+		if (advance || slot >= nritems) {
+			if (slot >= nritems - 1) {
+				ret = btrfs_next_leaf(root, path);
+				if (ret)
+					break;
+				leaf = path->nodes[0];
+				nritems = btrfs_header_nritems(leaf);
+				slot = path->slots[0];
+			} else {
+				slot++;
+				path->slots[0]++;
+			}
+		}
+
+		advance = 1;
+		item = btrfs_item_nr(leaf, slot);
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+
+		if (found_key.objectid != key.objectid)
+			break;
+		if (btrfs_key_type(&found_key) != key_type)
+			break;
+		if (found_key.offset < filp->f_pos)
+			continue;
+
+		filp->f_pos = found_key.offset;
+
+		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
+		di_cur = 0;
+		di_total = btrfs_item_size(leaf, item);
+
+		while (di_cur < di_total) {
+			struct btrfs_key location;
+
+			name_len = btrfs_dir_name_len(leaf, di);
+			if (name_len <= sizeof(tmp_name)) {
+				name_ptr = tmp_name;
+			} else {
+				name_ptr = kmalloc(name_len, GFP_NOFS);
+				if (!name_ptr) {
+					ret = -ENOMEM;
+					goto err;
+				}
+			}
+			read_extent_buffer(leaf, name_ptr,
+					   (unsigned long)(di + 1), name_len);
+
+			d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
+			btrfs_dir_item_key_to_cpu(leaf, di, &location);
+
+			/* is this a reference to our own snapshot? If so
+			 * skip it
+			 */
+			if (location.type == BTRFS_ROOT_ITEM_KEY &&
+			    location.objectid == root->root_key.objectid) {
+				over = 0;
+				goto skip;
+			}
+			over = filldir(dirent, name_ptr, name_len,
+				       found_key.offset, location.objectid,
+				       d_type);
+
+skip:
+			if (name_ptr != tmp_name)
+				kfree(name_ptr);
+
+			if (over)
+				goto nopos;
+			di_len = btrfs_dir_name_len(leaf, di) +
+				 btrfs_dir_data_len(leaf, di) + sizeof(*di);
+			di_cur += di_len;
+			di = (struct btrfs_dir_item *)((char *)di + di_len);
+		}
+	}
+
+	/* Reached end of directory/root. Bump pos past the last item. */
+	if (key_type == BTRFS_DIR_INDEX_KEY)
+		filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
+	else
+		filp->f_pos++;
+nopos:
+	ret = 0;
+err:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_write_inode(struct inode *inode, int wait)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	int ret = 0;
+
+	if (root->fs_info->btree_inode == inode)
+		return 0;
+
+	if (wait) {
+		trans = btrfs_join_transaction(root, 1);
+		btrfs_set_trans_block_group(trans, inode);
+		ret = btrfs_commit_transaction(trans, root);
+	}
+	return ret;
+}
+
+/*
+ * This is somewhat expensive, updating the tree every time the
+ * inode changes.  But, it is most likely to find the inode in cache.
+ * FIXME, needs more benchmarking...there are no reasons other than performance
+ * to keep or drop this code.
+ */
+void btrfs_dirty_inode(struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+
+	trans = btrfs_join_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
+	btrfs_update_inode(trans, root, inode);
+	btrfs_end_transaction(trans, root);
+}
+
+/*
+ * find the highest existing sequence number in a directory
+ * and then set the in-memory index_cnt variable to reflect
+ * free sequence numbers
+ */
+static int btrfs_set_inode_index_count(struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_key key, found_key;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	int ret;
+
+	key.objectid = inode->i_ino;
+	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
+	key.offset = (u64)-1;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	/* FIXME: we should be able to handle this */
+	if (ret == 0)
+		goto out;
+	ret = 0;
+
+	/*
+	 * MAGIC NUMBER EXPLANATION:
+	 * since we search a directory based on f_pos we have to start at 2
+	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
+	 * else has to start at 2
+	 */
+	if (path->slots[0] == 0) {
+		BTRFS_I(inode)->index_cnt = 2;
+		goto out;
+	}
+
+	path->slots[0]--;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+	if (found_key.objectid != inode->i_ino ||
+	    btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
+		BTRFS_I(inode)->index_cnt = 2;
+		goto out;
+	}
+
+	BTRFS_I(inode)->index_cnt = found_key.offset + 1;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * helper to find a free sequence number in a given directory.  This current
+ * code is very simple, later versions will do smarter things in the btree
+ */
+int btrfs_set_inode_index(struct inode *dir, u64 *index)
+{
+	int ret = 0;
+
+	if (BTRFS_I(dir)->index_cnt == (u64)-1) {
+		ret = btrfs_set_inode_index_count(dir);
+		if (ret)
+			return ret;
+	}
+
+	*index = BTRFS_I(dir)->index_cnt;
+	BTRFS_I(dir)->index_cnt++;
+
+	return ret;
+}
+
+static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root,
+				     struct inode *dir,
+				     const char *name, int name_len,
+				     u64 ref_objectid, u64 objectid,
+				     u64 alloc_hint, int mode, u64 *index)
+{
+	struct inode *inode;
+	struct btrfs_inode_item *inode_item;
+	struct btrfs_key *location;
+	struct btrfs_path *path;
+	struct btrfs_inode_ref *ref;
+	struct btrfs_key key[2];
+	u32 sizes[2];
+	unsigned long ptr;
+	int ret;
+	int owner;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	inode = new_inode(root->fs_info->sb);
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+
+	if (dir) {
+		ret = btrfs_set_inode_index(dir, index);
+		if (ret)
+			return ERR_PTR(ret);
+	}
+	/*
+	 * index_cnt is ignored for everything but a dir,
+	 * btrfs_get_inode_index_count has an explanation for the magic
+	 * number
+	 */
+	init_btrfs_i(inode);
+	BTRFS_I(inode)->index_cnt = 2;
+	BTRFS_I(inode)->root = root;
+	BTRFS_I(inode)->generation = trans->transid;
+
+	if (mode & S_IFDIR)
+		owner = 0;
+	else
+		owner = 1;
+	BTRFS_I(inode)->block_group =
+			btrfs_find_block_group(root, 0, alloc_hint, owner);
+	if ((mode & S_IFREG)) {
+		if (btrfs_test_opt(root, NODATASUM))
+			btrfs_set_flag(inode, NODATASUM);
+		if (btrfs_test_opt(root, NODATACOW))
+			btrfs_set_flag(inode, NODATACOW);
+	}
+
+	key[0].objectid = objectid;
+	btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
+	key[0].offset = 0;
+
+	key[1].objectid = objectid;
+	btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
+	key[1].offset = ref_objectid;
+
+	sizes[0] = sizeof(struct btrfs_inode_item);
+	sizes[1] = name_len + sizeof(*ref);
+
+	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
+	if (ret != 0)
+		goto fail;
+
+	if (objectid > root->highest_inode)
+		root->highest_inode = objectid;
+
+	inode->i_uid = current_fsuid();
+	inode->i_gid = current_fsgid();
+	inode->i_mode = mode;
+	inode->i_ino = objectid;
+	inode_set_bytes(inode, 0);
+	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
+	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+				  struct btrfs_inode_item);
+	fill_inode_item(trans, path->nodes[0], inode_item, inode);
+
+	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
+			     struct btrfs_inode_ref);
+	btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
+	btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
+	ptr = (unsigned long)(ref + 1);
+	write_extent_buffer(path->nodes[0], name, ptr, name_len);
+
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_free_path(path);
+
+	location = &BTRFS_I(inode)->location;
+	location->objectid = objectid;
+	location->offset = 0;
+	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
+
+	insert_inode_hash(inode);
+	return inode;
+fail:
+	if (dir)
+		BTRFS_I(dir)->index_cnt--;
+	btrfs_free_path(path);
+	return ERR_PTR(ret);
+}
+
+static inline u8 btrfs_inode_type(struct inode *inode)
+{
+	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
+}
+
+/*
+ * utility function to add 'inode' into 'parent_inode' with
+ * a give name and a given sequence number.
+ * if 'add_backref' is true, also insert a backref from the
+ * inode to the parent directory.
+ */
+int btrfs_add_link(struct btrfs_trans_handle *trans,
+		   struct inode *parent_inode, struct inode *inode,
+		   const char *name, int name_len, int add_backref, u64 index)
+{
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_root *root = BTRFS_I(parent_inode)->root;
+
+	key.objectid = inode->i_ino;
+	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
+	key.offset = 0;
+
+	ret = btrfs_insert_dir_item(trans, root, name, name_len,
+				    parent_inode->i_ino,
+				    &key, btrfs_inode_type(inode),
+				    index);
+	if (ret == 0) {
+		if (add_backref) {
+			ret = btrfs_insert_inode_ref(trans, root,
+						     name, name_len,
+						     inode->i_ino,
+						     parent_inode->i_ino,
+						     index);
+		}
+		btrfs_i_size_write(parent_inode, parent_inode->i_size +
+				   name_len * 2);
+		parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
+		ret = btrfs_update_inode(trans, root, parent_inode);
+	}
+	return ret;
+}
+
+static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
+			    struct dentry *dentry, struct inode *inode,
+			    int backref, u64 index)
+{
+	int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
+				 inode, dentry->d_name.name,
+				 dentry->d_name.len, backref, index);
+	if (!err) {
+		d_instantiate(dentry, inode);
+		return 0;
+	}
+	if (err > 0)
+		err = -EEXIST;
+	return err;
+}
+
+static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
+			int mode, dev_t rdev)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct inode *inode = NULL;
+	int err;
+	int drop_inode = 0;
+	u64 objectid;
+	unsigned long nr = 0;
+	u64 index = 0;
+
+	if (!new_valid_dev(rdev))
+		return -EINVAL;
+
+	err = btrfs_check_free_space(root, 1, 0);
+	if (err)
+		goto fail;
+
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
+
+	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
+	if (err) {
+		err = -ENOSPC;
+		goto out_unlock;
+	}
+
+	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
+				dentry->d_name.len,
+				dentry->d_parent->d_inode->i_ino, objectid,
+				BTRFS_I(dir)->block_group, mode, &index);
+	err = PTR_ERR(inode);
+	if (IS_ERR(inode))
+		goto out_unlock;
+
+	err = btrfs_init_acl(inode, dir);
+	if (err) {
+		drop_inode = 1;
+		goto out_unlock;
+	}
+
+	btrfs_set_trans_block_group(trans, inode);
+	err = btrfs_add_nondir(trans, dentry, inode, 0, index);
+	if (err)
+		drop_inode = 1;
+	else {
+		inode->i_op = &btrfs_special_inode_operations;
+		init_special_inode(inode, inode->i_mode, rdev);
+		btrfs_update_inode(trans, root, inode);
+	}
+	dir->i_sb->s_dirt = 1;
+	btrfs_update_inode_block_group(trans, inode);
+	btrfs_update_inode_block_group(trans, dir);
+out_unlock:
+	nr = trans->blocks_used;
+	btrfs_end_transaction_throttle(trans, root);
+fail:
+	if (drop_inode) {
+		inode_dec_link_count(inode);
+		iput(inode);
+	}
+	btrfs_btree_balance_dirty(root, nr);
+	return err;
+}
+
+static int btrfs_create(struct inode *dir, struct dentry *dentry,
+			int mode, struct nameidata *nd)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct inode *inode = NULL;
+	int err;
+	int drop_inode = 0;
+	unsigned long nr = 0;
+	u64 objectid;
+	u64 index = 0;
+
+	err = btrfs_check_free_space(root, 1, 0);
+	if (err)
+		goto fail;
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
+
+	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
+	if (err) {
+		err = -ENOSPC;
+		goto out_unlock;
+	}
+
+	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
+				dentry->d_name.len,
+				dentry->d_parent->d_inode->i_ino,
+				objectid, BTRFS_I(dir)->block_group, mode,
+				&index);
+	err = PTR_ERR(inode);
+	if (IS_ERR(inode))
+		goto out_unlock;
+
+	err = btrfs_init_acl(inode, dir);
+	if (err) {
+		drop_inode = 1;
+		goto out_unlock;
+	}
+
+	btrfs_set_trans_block_group(trans, inode);
+	err = btrfs_add_nondir(trans, dentry, inode, 0, index);
+	if (err)
+		drop_inode = 1;
+	else {
+		inode->i_mapping->a_ops = &btrfs_aops;
+		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
+		inode->i_fop = &btrfs_file_operations;
+		inode->i_op = &btrfs_file_inode_operations;
+		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
+	}
+	dir->i_sb->s_dirt = 1;
+	btrfs_update_inode_block_group(trans, inode);
+	btrfs_update_inode_block_group(trans, dir);
+out_unlock:
+	nr = trans->blocks_used;
+	btrfs_end_transaction_throttle(trans, root);
+fail:
+	if (drop_inode) {
+		inode_dec_link_count(inode);
+		iput(inode);
+	}
+	btrfs_btree_balance_dirty(root, nr);
+	return err;
+}
+
+static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
+		      struct dentry *dentry)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct inode *inode = old_dentry->d_inode;
+	u64 index;
+	unsigned long nr = 0;
+	int err;
+	int drop_inode = 0;
+
+	if (inode->i_nlink == 0)
+		return -ENOENT;
+
+	btrfs_inc_nlink(inode);
+	err = btrfs_check_free_space(root, 1, 0);
+	if (err)
+		goto fail;
+	err = btrfs_set_inode_index(dir, &index);
+	if (err)
+		goto fail;
+
+	trans = btrfs_start_transaction(root, 1);
+
+	btrfs_set_trans_block_group(trans, dir);
+	atomic_inc(&inode->i_count);
+
+	err = btrfs_add_nondir(trans, dentry, inode, 1, index);
+
+	if (err)
+		drop_inode = 1;
+
+	dir->i_sb->s_dirt = 1;
+	btrfs_update_inode_block_group(trans, dir);
+	err = btrfs_update_inode(trans, root, inode);
+
+	if (err)
+		drop_inode = 1;
+
+	nr = trans->blocks_used;
+	btrfs_end_transaction_throttle(trans, root);
+fail:
+	if (drop_inode) {
+		inode_dec_link_count(inode);
+		iput(inode);
+	}
+	btrfs_btree_balance_dirty(root, nr);
+	return err;
+}
+
+static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
+{
+	struct inode *inode = NULL;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	int err = 0;
+	int drop_on_err = 0;
+	u64 objectid = 0;
+	u64 index = 0;
+	unsigned long nr = 1;
+
+	err = btrfs_check_free_space(root, 1, 0);
+	if (err)
+		goto out_unlock;
+
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
+
+	if (IS_ERR(trans)) {
+		err = PTR_ERR(trans);
+		goto out_unlock;
+	}
+
+	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
+	if (err) {
+		err = -ENOSPC;
+		goto out_unlock;
+	}
+
+	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
+				dentry->d_name.len,
+				dentry->d_parent->d_inode->i_ino, objectid,
+				BTRFS_I(dir)->block_group, S_IFDIR | mode,
+				&index);
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
+		goto out_fail;
+	}
+
+	drop_on_err = 1;
+
+	err = btrfs_init_acl(inode, dir);
+	if (err)
+		goto out_fail;
+
+	inode->i_op = &btrfs_dir_inode_operations;
+	inode->i_fop = &btrfs_dir_file_operations;
+	btrfs_set_trans_block_group(trans, inode);
+
+	btrfs_i_size_write(inode, 0);
+	err = btrfs_update_inode(trans, root, inode);
+	if (err)
+		goto out_fail;
+
+	err = btrfs_add_link(trans, dentry->d_parent->d_inode,
+				 inode, dentry->d_name.name,
+				 dentry->d_name.len, 0, index);
+	if (err)
+		goto out_fail;
+
+	d_instantiate(dentry, inode);
+	drop_on_err = 0;
+	dir->i_sb->s_dirt = 1;
+	btrfs_update_inode_block_group(trans, inode);
+	btrfs_update_inode_block_group(trans, dir);
+
+out_fail:
+	nr = trans->blocks_used;
+	btrfs_end_transaction_throttle(trans, root);
+
+out_unlock:
+	if (drop_on_err)
+		iput(inode);
+	btrfs_btree_balance_dirty(root, nr);
+	return err;
+}
+
+static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
+			const struct iovec *iov, loff_t offset,
+			unsigned long nr_segs)
+{
+	return -EINVAL;
+}
+
+static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
+{
+	return extent_bmap(mapping, iblock, btrfs_get_extent);
+}
+
+int btrfs_readpage(struct file *file, struct page *page)
+{
+	struct extent_io_tree *tree;
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	return extent_read_full_page(tree, page, btrfs_get_extent);
+}
+
+static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
+{
+	struct extent_io_tree *tree;
+
+
+	if (current->flags & PF_MEMALLOC) {
+		redirty_page_for_writepage(wbc, page);
+		unlock_page(page);
+		return 0;
+	}
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
+}
+
+int btrfs_writepages(struct address_space *mapping,
+		     struct writeback_control *wbc)
+{
+	struct extent_io_tree *tree;
+
+	tree = &BTRFS_I(mapping->host)->io_tree;
+	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
+}
+
+static int
+btrfs_readpages(struct file *file, struct address_space *mapping,
+		struct list_head *pages, unsigned nr_pages)
+{
+	struct extent_io_tree *tree;
+	tree = &BTRFS_I(mapping->host)->io_tree;
+	return extent_readpages(tree, mapping, pages, nr_pages,
+				btrfs_get_extent);
+}
+static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
+{
+	struct extent_io_tree *tree;
+	struct extent_map_tree *map;
+	int ret;
+
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	map = &BTRFS_I(page->mapping->host)->extent_tree;
+	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
+	if (ret == 1) {
+		ClearPagePrivate(page);
+		set_page_private(page, 0);
+		page_cache_release(page);
+	}
+	return ret;
+}
+
+static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
+{
+	if (PageWriteback(page) || PageDirty(page))
+		return 0;
+	return __btrfs_releasepage(page, gfp_flags);
+}
+
+static void btrfs_invalidatepage(struct page *page, unsigned long offset)
+{
+	struct extent_io_tree *tree;
+	struct btrfs_ordered_extent *ordered;
+	u64 page_start = page_offset(page);
+	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
+
+	wait_on_page_writeback(page);
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	if (offset) {
+		btrfs_releasepage(page, GFP_NOFS);
+		return;
+	}
+
+	lock_extent(tree, page_start, page_end, GFP_NOFS);
+	ordered = btrfs_lookup_ordered_extent(page->mapping->host,
+					   page_offset(page));
+	if (ordered) {
+		/*
+		 * IO on this page will never be started, so we need
+		 * to account for any ordered extents now
+		 */
+		clear_extent_bit(tree, page_start, page_end,
+				 EXTENT_DIRTY | EXTENT_DELALLOC |
+				 EXTENT_LOCKED, 1, 0, GFP_NOFS);
+		btrfs_finish_ordered_io(page->mapping->host,
+					page_start, page_end);
+		btrfs_put_ordered_extent(ordered);
+		lock_extent(tree, page_start, page_end, GFP_NOFS);
+	}
+	clear_extent_bit(tree, page_start, page_end,
+		 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
+		 EXTENT_ORDERED,
+		 1, 1, GFP_NOFS);
+	__btrfs_releasepage(page, GFP_NOFS);
+
+	ClearPageChecked(page);
+	if (PagePrivate(page)) {
+		ClearPagePrivate(page);
+		set_page_private(page, 0);
+		page_cache_release(page);
+	}
+}
+
+/*
+ * btrfs_page_mkwrite() is not allowed to change the file size as it gets
+ * called from a page fault handler when a page is first dirtied. Hence we must
+ * be careful to check for EOF conditions here. We set the page up correctly
+ * for a written page which means we get ENOSPC checking when writing into
+ * holes and correct delalloc and unwritten extent mapping on filesystems that
+ * support these features.
+ *
+ * We are not allowed to take the i_mutex here so we have to play games to
+ * protect against truncate races as the page could now be beyond EOF.  Because
+ * vmtruncate() writes the inode size before removing pages, once we have the
+ * page lock we can determine safely if the page is beyond EOF. If it is not
+ * beyond EOF, then the page is guaranteed safe against truncation until we
+ * unlock the page.
+ */
+int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
+{
+	struct inode *inode = fdentry(vma->vm_file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_ordered_extent *ordered;
+	char *kaddr;
+	unsigned long zero_start;
+	loff_t size;
+	int ret;
+	u64 page_start;
+	u64 page_end;
+
+	ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
+	if (ret)
+		goto out;
+
+	ret = -EINVAL;
+again:
+	lock_page(page);
+	size = i_size_read(inode);
+	page_start = page_offset(page);
+	page_end = page_start + PAGE_CACHE_SIZE - 1;
+
+	if ((page->mapping != inode->i_mapping) ||
+	    (page_start >= size)) {
+		/* page got truncated out from underneath us */
+		goto out_unlock;
+	}
+	wait_on_page_writeback(page);
+
+	lock_extent(io_tree, page_start, page_end, GFP_NOFS);
+	set_page_extent_mapped(page);
+
+	/*
+	 * we can't set the delalloc bits if there are pending ordered
+	 * extents.  Drop our locks and wait for them to finish
+	 */
+	ordered = btrfs_lookup_ordered_extent(inode, page_start);
+	if (ordered) {
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+		unlock_page(page);
+		btrfs_start_ordered_extent(inode, ordered, 1);
+		btrfs_put_ordered_extent(ordered);
+		goto again;
+	}
+
+	btrfs_set_extent_delalloc(inode, page_start, page_end);
+	ret = 0;
+
+	/* page is wholly or partially inside EOF */
+	if (page_start + PAGE_CACHE_SIZE > size)
+		zero_start = size & ~PAGE_CACHE_MASK;
+	else
+		zero_start = PAGE_CACHE_SIZE;
+
+	if (zero_start != PAGE_CACHE_SIZE) {
+		kaddr = kmap(page);
+		memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
+		flush_dcache_page(page);
+		kunmap(page);
+	}
+	ClearPageChecked(page);
+	set_page_dirty(page);
+	unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+
+out_unlock:
+	unlock_page(page);
+out:
+	return ret;
+}
+
+static void btrfs_truncate(struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret;
+	struct btrfs_trans_handle *trans;
+	unsigned long nr;
+	u64 mask = root->sectorsize - 1;
+
+	if (!S_ISREG(inode->i_mode))
+		return;
+	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
+		return;
+
+	btrfs_truncate_page(inode->i_mapping, inode->i_size);
+	btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
+
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
+	btrfs_i_size_write(inode, inode->i_size);
+
+	ret = btrfs_orphan_add(trans, inode);
+	if (ret)
+		goto out;
+	/* FIXME, add redo link to tree so we don't leak on crash */
+	ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
+				      BTRFS_EXTENT_DATA_KEY);
+	btrfs_update_inode(trans, root, inode);
+
+	ret = btrfs_orphan_del(trans, inode);
+	BUG_ON(ret);
+
+out:
+	nr = trans->blocks_used;
+	ret = btrfs_end_transaction_throttle(trans, root);
+	BUG_ON(ret);
+	btrfs_btree_balance_dirty(root, nr);
+}
+
+/*
+ * create a new subvolume directory/inode (helper for the ioctl).
+ */
+int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *new_root, struct dentry *dentry,
+			     u64 new_dirid, u64 alloc_hint)
+{
+	struct inode *inode;
+	int error;
+	u64 index = 0;
+
+	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
+				new_dirid, alloc_hint, S_IFDIR | 0700, &index);
+	if (IS_ERR(inode))
+		return PTR_ERR(inode);
+	inode->i_op = &btrfs_dir_inode_operations;
+	inode->i_fop = &btrfs_dir_file_operations;
+
+	inode->i_nlink = 1;
+	btrfs_i_size_write(inode, 0);
+
+	error = btrfs_update_inode(trans, new_root, inode);
+	if (error)
+		return error;
+
+	d_instantiate(dentry, inode);
+	return 0;
+}
+
+/* helper function for file defrag and space balancing.  This
+ * forces readahead on a given range of bytes in an inode
+ */
+unsigned long btrfs_force_ra(struct address_space *mapping,
+			      struct file_ra_state *ra, struct file *file,
+			      pgoff_t offset, pgoff_t last_index)
+{
+	pgoff_t req_size = last_index - offset + 1;
+
+	page_cache_sync_readahead(mapping, ra, file, offset, req_size);
+	return offset + req_size;
+}
+
+struct inode *btrfs_alloc_inode(struct super_block *sb)
+{
+	struct btrfs_inode *ei;
+
+	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
+	if (!ei)
+		return NULL;
+	ei->last_trans = 0;
+	ei->logged_trans = 0;
+	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
+	ei->i_acl = BTRFS_ACL_NOT_CACHED;
+	ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
+	INIT_LIST_HEAD(&ei->i_orphan);
+	return &ei->vfs_inode;
+}
+
+void btrfs_destroy_inode(struct inode *inode)
+{
+	struct btrfs_ordered_extent *ordered;
+	WARN_ON(!list_empty(&inode->i_dentry));
+	WARN_ON(inode->i_data.nrpages);
+
+	if (BTRFS_I(inode)->i_acl &&
+	    BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
+		posix_acl_release(BTRFS_I(inode)->i_acl);
+	if (BTRFS_I(inode)->i_default_acl &&
+	    BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
+		posix_acl_release(BTRFS_I(inode)->i_default_acl);
+
+	spin_lock(&BTRFS_I(inode)->root->list_lock);
+	if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
+		printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
+		       " list\n", inode->i_ino);
+		dump_stack();
+	}
+	spin_unlock(&BTRFS_I(inode)->root->list_lock);
+
+	while (1) {
+		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
+		if (!ordered)
+			break;
+		else {
+			printk(KERN_ERR "btrfs found ordered "
+			       "extent %llu %llu on inode cleanup\n",
+			       (unsigned long long)ordered->file_offset,
+			       (unsigned long long)ordered->len);
+			btrfs_remove_ordered_extent(inode, ordered);
+			btrfs_put_ordered_extent(ordered);
+			btrfs_put_ordered_extent(ordered);
+		}
+	}
+	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
+	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
+}
+
+static void init_once(void *foo)
+{
+	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
+
+	inode_init_once(&ei->vfs_inode);
+}
+
+void btrfs_destroy_cachep(void)
+{
+	if (btrfs_inode_cachep)
+		kmem_cache_destroy(btrfs_inode_cachep);
+	if (btrfs_trans_handle_cachep)
+		kmem_cache_destroy(btrfs_trans_handle_cachep);
+	if (btrfs_transaction_cachep)
+		kmem_cache_destroy(btrfs_transaction_cachep);
+	if (btrfs_bit_radix_cachep)
+		kmem_cache_destroy(btrfs_bit_radix_cachep);
+	if (btrfs_path_cachep)
+		kmem_cache_destroy(btrfs_path_cachep);
+}
+
+struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
+				       unsigned long extra_flags,
+				       void (*ctor)(void *))
+{
+	return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
+				 SLAB_MEM_SPREAD | extra_flags), ctor);
+}
+
+int btrfs_init_cachep(void)
+{
+	btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
+					  sizeof(struct btrfs_inode),
+					  0, init_once);
+	if (!btrfs_inode_cachep)
+		goto fail;
+	btrfs_trans_handle_cachep =
+			btrfs_cache_create("btrfs_trans_handle_cache",
+					   sizeof(struct btrfs_trans_handle),
+					   0, NULL);
+	if (!btrfs_trans_handle_cachep)
+		goto fail;
+	btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
+					     sizeof(struct btrfs_transaction),
+					     0, NULL);
+	if (!btrfs_transaction_cachep)
+		goto fail;
+	btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
+					 sizeof(struct btrfs_path),
+					 0, NULL);
+	if (!btrfs_path_cachep)
+		goto fail;
+	btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
+					      SLAB_DESTROY_BY_RCU, NULL);
+	if (!btrfs_bit_radix_cachep)
+		goto fail;
+	return 0;
+fail:
+	btrfs_destroy_cachep();
+	return -ENOMEM;
+}
+
+static int btrfs_getattr(struct vfsmount *mnt,
+			 struct dentry *dentry, struct kstat *stat)
+{
+	struct inode *inode = dentry->d_inode;
+	generic_fillattr(inode, stat);
+	stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
+	stat->blksize = PAGE_CACHE_SIZE;
+	stat->blocks = (inode_get_bytes(inode) +
+			BTRFS_I(inode)->delalloc_bytes) >> 9;
+	return 0;
+}
+
+static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
+			   struct inode *new_dir, struct dentry *new_dentry)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(old_dir)->root;
+	struct inode *new_inode = new_dentry->d_inode;
+	struct inode *old_inode = old_dentry->d_inode;
+	struct timespec ctime = CURRENT_TIME;
+	u64 index = 0;
+	int ret;
+
+	/* we're not allowed to rename between subvolumes */
+	if (BTRFS_I(old_inode)->root->root_key.objectid !=
+	    BTRFS_I(new_dir)->root->root_key.objectid)
+		return -EXDEV;
+
+	if (S_ISDIR(old_inode->i_mode) && new_inode &&
+	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
+		return -ENOTEMPTY;
+	}
+
+	/* to rename a snapshot or subvolume, we need to juggle the
+	 * backrefs.  This isn't coded yet
+	 */
+	if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
+		return -EXDEV;
+
+	ret = btrfs_check_free_space(root, 1, 0);
+	if (ret)
+		goto out_unlock;
+
+	trans = btrfs_start_transaction(root, 1);
+
+	btrfs_set_trans_block_group(trans, new_dir);
+
+	btrfs_inc_nlink(old_dentry->d_inode);
+	old_dir->i_ctime = old_dir->i_mtime = ctime;
+	new_dir->i_ctime = new_dir->i_mtime = ctime;
+	old_inode->i_ctime = ctime;
+
+	ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
+				 old_dentry->d_name.name,
+				 old_dentry->d_name.len);
+	if (ret)
+		goto out_fail;
+
+	if (new_inode) {
+		new_inode->i_ctime = CURRENT_TIME;
+		ret = btrfs_unlink_inode(trans, root, new_dir,
+					 new_dentry->d_inode,
+					 new_dentry->d_name.name,
+					 new_dentry->d_name.len);
+		if (ret)
+			goto out_fail;
+		if (new_inode->i_nlink == 0) {
+			ret = btrfs_orphan_add(trans, new_dentry->d_inode);
+			if (ret)
+				goto out_fail;
+		}
+
+	}
+	ret = btrfs_set_inode_index(new_dir, &index);
+	if (ret)
+		goto out_fail;
+
+	ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
+			     old_inode, new_dentry->d_name.name,
+			     new_dentry->d_name.len, 1, index);
+	if (ret)
+		goto out_fail;
+
+out_fail:
+	btrfs_end_transaction_throttle(trans, root);
+out_unlock:
+	return ret;
+}
+
+/*
+ * some fairly slow code that needs optimization. This walks the list
+ * of all the inodes with pending delalloc and forces them to disk.
+ */
+int btrfs_start_delalloc_inodes(struct btrfs_root *root)
+{
+	struct list_head *head = &root->fs_info->delalloc_inodes;
+	struct btrfs_inode *binode;
+	struct inode *inode;
+
+	if (root->fs_info->sb->s_flags & MS_RDONLY)
+		return -EROFS;
+
+	spin_lock(&root->fs_info->delalloc_lock);
+	while (!list_empty(head)) {
+		binode = list_entry(head->next, struct btrfs_inode,
+				    delalloc_inodes);
+		inode = igrab(&binode->vfs_inode);
+		if (!inode)
+			list_del_init(&binode->delalloc_inodes);
+		spin_unlock(&root->fs_info->delalloc_lock);
+		if (inode) {
+			filemap_flush(inode->i_mapping);
+			iput(inode);
+		}
+		cond_resched();
+		spin_lock(&root->fs_info->delalloc_lock);
+	}
+	spin_unlock(&root->fs_info->delalloc_lock);
+
+	/* the filemap_flush will queue IO into the worker threads, but
+	 * we have to make sure the IO is actually started and that
+	 * ordered extents get created before we return
+	 */
+	atomic_inc(&root->fs_info->async_submit_draining);
+	while (atomic_read(&root->fs_info->nr_async_submits) ||
+	      atomic_read(&root->fs_info->async_delalloc_pages)) {
+		wait_event(root->fs_info->async_submit_wait,
+		   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
+		    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
+	}
+	atomic_dec(&root->fs_info->async_submit_draining);
+	return 0;
+}
+
+static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
+			 const char *symname)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct inode *inode = NULL;
+	int err;
+	int drop_inode = 0;
+	u64 objectid;
+	u64 index = 0 ;
+	int name_len;
+	int datasize;
+	unsigned long ptr;
+	struct btrfs_file_extent_item *ei;
+	struct extent_buffer *leaf;
+	unsigned long nr = 0;
+
+	name_len = strlen(symname) + 1;
+	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
+		return -ENAMETOOLONG;
+
+	err = btrfs_check_free_space(root, 1, 0);
+	if (err)
+		goto out_fail;
+
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
+
+	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
+	if (err) {
+		err = -ENOSPC;
+		goto out_unlock;
+	}
+
+	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
+				dentry->d_name.len,
+				dentry->d_parent->d_inode->i_ino, objectid,
+				BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
+				&index);
+	err = PTR_ERR(inode);
+	if (IS_ERR(inode))
+		goto out_unlock;
+
+	err = btrfs_init_acl(inode, dir);
+	if (err) {
+		drop_inode = 1;
+		goto out_unlock;
+	}
+
+	btrfs_set_trans_block_group(trans, inode);
+	err = btrfs_add_nondir(trans, dentry, inode, 0, index);
+	if (err)
+		drop_inode = 1;
+	else {
+		inode->i_mapping->a_ops = &btrfs_aops;
+		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
+		inode->i_fop = &btrfs_file_operations;
+		inode->i_op = &btrfs_file_inode_operations;
+		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
+	}
+	dir->i_sb->s_dirt = 1;
+	btrfs_update_inode_block_group(trans, inode);
+	btrfs_update_inode_block_group(trans, dir);
+	if (drop_inode)
+		goto out_unlock;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	key.objectid = inode->i_ino;
+	key.offset = 0;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
+	datasize = btrfs_file_extent_calc_inline_size(name_len);
+	err = btrfs_insert_empty_item(trans, root, path, &key,
+				      datasize);
+	if (err) {
+		drop_inode = 1;
+		goto out_unlock;
+	}
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, path->slots[0],
+			    struct btrfs_file_extent_item);
+	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
+	btrfs_set_file_extent_type(leaf, ei,
+				   BTRFS_FILE_EXTENT_INLINE);
+	btrfs_set_file_extent_encryption(leaf, ei, 0);
+	btrfs_set_file_extent_compression(leaf, ei, 0);
+	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
+	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
+
+	ptr = btrfs_file_extent_inline_start(ei);
+	write_extent_buffer(leaf, symname, ptr, name_len);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_free_path(path);
+
+	inode->i_op = &btrfs_symlink_inode_operations;
+	inode->i_mapping->a_ops = &btrfs_symlink_aops;
+	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
+	inode_set_bytes(inode, name_len);
+	btrfs_i_size_write(inode, name_len - 1);
+	err = btrfs_update_inode(trans, root, inode);
+	if (err)
+		drop_inode = 1;
+
+out_unlock:
+	nr = trans->blocks_used;
+	btrfs_end_transaction_throttle(trans, root);
+out_fail:
+	if (drop_inode) {
+		inode_dec_link_count(inode);
+		iput(inode);
+	}
+	btrfs_btree_balance_dirty(root, nr);
+	return err;
+}
+
+static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
+			       u64 alloc_hint, int mode)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_key ins;
+	u64 alloc_size;
+	u64 cur_offset = start;
+	u64 num_bytes = end - start;
+	int ret = 0;
+
+	trans = btrfs_join_transaction(root, 1);
+	BUG_ON(!trans);
+	btrfs_set_trans_block_group(trans, inode);
+
+	while (num_bytes > 0) {
+		alloc_size = min(num_bytes, root->fs_info->max_extent);
+		ret = btrfs_reserve_extent(trans, root, alloc_size,
+					   root->sectorsize, 0, alloc_hint,
+					   (u64)-1, &ins, 1);
+		if (ret) {
+			WARN_ON(1);
+			goto out;
+		}
+		ret = insert_reserved_file_extent(trans, inode,
+						  cur_offset, ins.objectid,
+						  ins.offset, ins.offset,
+						  ins.offset, 0, 0, 0,
+						  BTRFS_FILE_EXTENT_PREALLOC);
+		BUG_ON(ret);
+		num_bytes -= ins.offset;
+		cur_offset += ins.offset;
+		alloc_hint = ins.objectid + ins.offset;
+	}
+out:
+	if (cur_offset > start) {
+		inode->i_ctime = CURRENT_TIME;
+		btrfs_set_flag(inode, PREALLOC);
+		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
+		    cur_offset > i_size_read(inode))
+			btrfs_i_size_write(inode, cur_offset);
+		ret = btrfs_update_inode(trans, root, inode);
+		BUG_ON(ret);
+	}
+
+	btrfs_end_transaction(trans, root);
+	return ret;
+}
+
+static long btrfs_fallocate(struct inode *inode, int mode,
+			    loff_t offset, loff_t len)
+{
+	u64 cur_offset;
+	u64 last_byte;
+	u64 alloc_start;
+	u64 alloc_end;
+	u64 alloc_hint = 0;
+	u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
+	struct extent_map *em;
+	int ret;
+
+	alloc_start = offset & ~mask;
+	alloc_end =  (offset + len + mask) & ~mask;
+
+	mutex_lock(&inode->i_mutex);
+	if (alloc_start > inode->i_size) {
+		ret = btrfs_cont_expand(inode, alloc_start);
+		if (ret)
+			goto out;
+	}
+
+	while (1) {
+		struct btrfs_ordered_extent *ordered;
+		lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
+			    alloc_end - 1, GFP_NOFS);
+		ordered = btrfs_lookup_first_ordered_extent(inode,
+							    alloc_end - 1);
+		if (ordered &&
+		    ordered->file_offset + ordered->len > alloc_start &&
+		    ordered->file_offset < alloc_end) {
+			btrfs_put_ordered_extent(ordered);
+			unlock_extent(&BTRFS_I(inode)->io_tree,
+				      alloc_start, alloc_end - 1, GFP_NOFS);
+			btrfs_wait_ordered_range(inode, alloc_start,
+						 alloc_end - alloc_start);
+		} else {
+			if (ordered)
+				btrfs_put_ordered_extent(ordered);
+			break;
+		}
+	}
+
+	cur_offset = alloc_start;
+	while (1) {
+		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
+				      alloc_end - cur_offset, 0);
+		BUG_ON(IS_ERR(em) || !em);
+		last_byte = min(extent_map_end(em), alloc_end);
+		last_byte = (last_byte + mask) & ~mask;
+		if (em->block_start == EXTENT_MAP_HOLE) {
+			ret = prealloc_file_range(inode, cur_offset,
+					last_byte, alloc_hint, mode);
+			if (ret < 0) {
+				free_extent_map(em);
+				break;
+			}
+		}
+		if (em->block_start <= EXTENT_MAP_LAST_BYTE)
+			alloc_hint = em->block_start;
+		free_extent_map(em);
+
+		cur_offset = last_byte;
+		if (cur_offset >= alloc_end) {
+			ret = 0;
+			break;
+		}
+	}
+	unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
+		      GFP_NOFS);
+out:
+	mutex_unlock(&inode->i_mutex);
+	return ret;
+}
+
+static int btrfs_set_page_dirty(struct page *page)
+{
+	return __set_page_dirty_nobuffers(page);
+}
+
+static int btrfs_permission(struct inode *inode, int mask)
+{
+	if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
+		return -EACCES;
+	return generic_permission(inode, mask, btrfs_check_acl);
+}
+
+static struct inode_operations btrfs_dir_inode_operations = {
+	.getattr	= btrfs_getattr,
+	.lookup		= btrfs_lookup,
+	.create		= btrfs_create,
+	.unlink		= btrfs_unlink,
+	.link		= btrfs_link,
+	.mkdir		= btrfs_mkdir,
+	.rmdir		= btrfs_rmdir,
+	.rename		= btrfs_rename,
+	.symlink	= btrfs_symlink,
+	.setattr	= btrfs_setattr,
+	.mknod		= btrfs_mknod,
+	.setxattr	= btrfs_setxattr,
+	.getxattr	= btrfs_getxattr,
+	.listxattr	= btrfs_listxattr,
+	.removexattr	= btrfs_removexattr,
+	.permission	= btrfs_permission,
+};
+static struct inode_operations btrfs_dir_ro_inode_operations = {
+	.lookup		= btrfs_lookup,
+	.permission	= btrfs_permission,
+};
+static struct file_operations btrfs_dir_file_operations = {
+	.llseek		= generic_file_llseek,
+	.read		= generic_read_dir,
+	.readdir	= btrfs_real_readdir,
+	.unlocked_ioctl	= btrfs_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl	= btrfs_ioctl,
+#endif
+	.release        = btrfs_release_file,
+	.fsync		= btrfs_sync_file,
+};
+
+static struct address_space_operations btrfs_aops = {
+	.readpage	= btrfs_readpage,
+	.writepage	= btrfs_writepage,
+	.writepages	= btrfs_writepages,
+	.readpages	= btrfs_readpages,
+	.sync_page	= block_sync_page,
+	.bmap		= btrfs_bmap,
+	.direct_IO	= btrfs_direct_IO,
+	.invalidatepage = btrfs_invalidatepage,
+	.releasepage	= btrfs_releasepage,
+	.set_page_dirty	= btrfs_set_page_dirty,
+};
+
+static struct address_space_operations btrfs_symlink_aops = {
+	.readpage	= btrfs_readpage,
+	.writepage	= btrfs_writepage,
+	.invalidatepage = btrfs_invalidatepage,
+	.releasepage	= btrfs_releasepage,
+};
+
+static struct inode_operations btrfs_file_inode_operations = {
+	.truncate	= btrfs_truncate,
+	.getattr	= btrfs_getattr,
+	.setattr	= btrfs_setattr,
+	.setxattr	= btrfs_setxattr,
+	.getxattr	= btrfs_getxattr,
+	.listxattr      = btrfs_listxattr,
+	.removexattr	= btrfs_removexattr,
+	.permission	= btrfs_permission,
+	.fallocate	= btrfs_fallocate,
+};
+static struct inode_operations btrfs_special_inode_operations = {
+	.getattr	= btrfs_getattr,
+	.setattr	= btrfs_setattr,
+	.permission	= btrfs_permission,
+	.setxattr	= btrfs_setxattr,
+	.getxattr	= btrfs_getxattr,
+	.listxattr	= btrfs_listxattr,
+	.removexattr	= btrfs_removexattr,
+};
+static struct inode_operations btrfs_symlink_inode_operations = {
+	.readlink	= generic_readlink,
+	.follow_link	= page_follow_link_light,
+	.put_link	= page_put_link,
+	.permission	= btrfs_permission,
+};

[PATCH 18.2/35] Btrfs inode extent_io hooks

[Chris Mason]

extent_io hooks for data IO

This code handles most of the data IO processing.  It provides the
hooks used by extent_io.c to do readpages/writepages and other basic
page operations.

It also does delayed allocation processing and metadata updates when
IO is complete for data=ordered.

Compression is started done when delayed allocation is filled.  The
pages are sent from extent_io.c locked with the extent_state locked as well.
This code will try to compress the pages via the async work queues and
start IO either on compressed extents or as plain data.

Inline extents are inserted as delayed allocation time as well.  Doing it
this late allows us to avoid inefficient inline extent manipulation at
file write time (for example appending to the file one byte at a time).

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- a/fs/btrfs/inode.c	2009-01-08 09:43:29.596585385 -0500
+++ b/fs/btrfs/inode.c	2009-01-08 09:46:16.272224440 -0500
@@ -85,6 +85,7 @@
 };
 
 static void btrfs_truncate(struct inode *inode);
+static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
 static noinline int cow_file_range(struct inode *inode,
 				   struct page *locked_page,
 				   u64 start, u64 end, int *page_started,
@@ -118,6 +119,1323 @@
 	return ret;
 }
 
+/*
+ * this does all the hard work for inserting an inline extent into
+ * the btree.  The caller should have done a btrfs_drop_extents so that
+ * no overlapping inline items exist in the btree
+ */
+static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root, struct inode *inode,
+				u64 start, size_t size, size_t compressed_size,
+				struct page **compressed_pages)
+{
+	struct btrfs_key key;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct page *page = NULL;
+	char *kaddr;
+	unsigned long ptr;
+	struct btrfs_file_extent_item *ei;
+	int err = 0;
+	int ret;
+	size_t cur_size = size;
+	size_t datasize;
+	unsigned long offset;
+	int use_compress = 0;
+
+	if (compressed_size && compressed_pages) {
+		use_compress = 1;
+		cur_size = compressed_size;
+	}
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	btrfs_set_trans_block_group(trans, inode);
+
+	key.objectid = inode->i_ino;
+	key.offset = start;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
+	datasize = btrfs_file_extent_calc_inline_size(cur_size);
+
+	inode_add_bytes(inode, size);
+	ret = btrfs_insert_empty_item(trans, root, path, &key,
+				      datasize);
+	BUG_ON(ret);
+	if (ret) {
+		err = ret;
+		goto fail;
+	}
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, path->slots[0],
+			    struct btrfs_file_extent_item);
+	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
+	btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
+	btrfs_set_file_extent_encryption(leaf, ei, 0);
+	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
+	btrfs_set_file_extent_ram_bytes(leaf, ei, size);
+	ptr = btrfs_file_extent_inline_start(ei);
+
+	if (use_compress) {
+		struct page *cpage;
+		int i = 0;
+		while (compressed_size > 0) {
+			cpage = compressed_pages[i];
+			cur_size = min_t(unsigned long, compressed_size,
+				       PAGE_CACHE_SIZE);
+
+			kaddr = kmap(cpage);
+			write_extent_buffer(leaf, kaddr, ptr, cur_size);
+			kunmap(cpage);
+
+			i++;
+			ptr += cur_size;
+			compressed_size -= cur_size;
+		}
+		btrfs_set_file_extent_compression(leaf, ei,
+						  BTRFS_COMPRESS_ZLIB);
+	} else {
+		page = find_get_page(inode->i_mapping,
+				     start >> PAGE_CACHE_SHIFT);
+		btrfs_set_file_extent_compression(leaf, ei, 0);
+		kaddr = kmap_atomic(page, KM_USER0);
+		offset = start & (PAGE_CACHE_SIZE - 1);
+		write_extent_buffer(leaf, kaddr + offset, ptr, size);
+		kunmap_atomic(kaddr, KM_USER0);
+		page_cache_release(page);
+	}
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_free_path(path);
+
+	BTRFS_I(inode)->disk_i_size = inode->i_size;
+	btrfs_update_inode(trans, root, inode);
+	return 0;
+fail:
+	btrfs_free_path(path);
+	return err;
+}
+
+
+/*
+ * conditionally insert an inline extent into the file.  This
+ * does the checks required to make sure the data is small enough
+ * to fit as an inline extent.
+ */
+static int cow_file_range_inline(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct inode *inode, u64 start, u64 end,
+				 size_t compressed_size,
+				 struct page **compressed_pages)
+{
+	u64 isize = i_size_read(inode);
+	u64 actual_end = min(end + 1, isize);
+	u64 inline_len = actual_end - start;
+	u64 aligned_end = (end + root->sectorsize - 1) &
+			~((u64)root->sectorsize - 1);
+	u64 hint_byte;
+	u64 data_len = inline_len;
+	int ret;
+
+	if (compressed_size)
+		data_len = compressed_size;
+
+	if (start > 0 ||
+	    actual_end >= PAGE_CACHE_SIZE ||
+	    data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
+	    (!compressed_size &&
+	    (actual_end & (root->sectorsize - 1)) == 0) ||
+	    end + 1 < isize ||
+	    data_len > root->fs_info->max_inline) {
+		return 1;
+	}
+
+	ret = btrfs_drop_extents(trans, root, inode, start,
+				 aligned_end, start, &hint_byte);
+	BUG_ON(ret);
+
+	if (isize > actual_end)
+		inline_len = min_t(u64, isize, actual_end);
+	ret = insert_inline_extent(trans, root, inode, start,
+				   inline_len, compressed_size,
+				   compressed_pages);
+	BUG_ON(ret);
+	btrfs_drop_extent_cache(inode, start, aligned_end, 0);
+	return 0;
+}
+
+struct async_extent {
+	u64 start;
+	u64 ram_size;
+	u64 compressed_size;
+	struct page **pages;
+	unsigned long nr_pages;
+	struct list_head list;
+};
+
+struct async_cow {
+	struct inode *inode;
+	struct btrfs_root *root;
+	struct page *locked_page;
+	u64 start;
+	u64 end;
+	struct list_head extents;
+	struct btrfs_work work;
+};
+
+static noinline int add_async_extent(struct async_cow *cow,
+				     u64 start, u64 ram_size,
+				     u64 compressed_size,
+				     struct page **pages,
+				     unsigned long nr_pages)
+{
+	struct async_extent *async_extent;
+
+	async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
+	async_extent->start = start;
+	async_extent->ram_size = ram_size;
+	async_extent->compressed_size = compressed_size;
+	async_extent->pages = pages;
+	async_extent->nr_pages = nr_pages;
+	list_add_tail(&async_extent->list, &cow->extents);
+	return 0;
+}
+
+/*
+ * we create compressed extents in two phases.  The first
+ * phase compresses a range of pages that have already been
+ * locked (both pages and state bits are locked).
+ *
+ * This is done inside an ordered work queue, and the compression
+ * is spread across many cpus.  The actual IO submission is step
+ * two, and the ordered work queue takes care of making sure that
+ * happens in the same order things were put onto the queue by
+ * writepages and friends.
+ *
+ * If this code finds it can't get good compression, it puts an
+ * entry onto the work queue to write the uncompressed bytes.  This
+ * makes sure that both compressed inodes and uncompressed inodes
+ * are written in the same order that pdflush sent them down.
+ */
+static noinline int compress_file_range(struct inode *inode,
+					struct page *locked_page,
+					u64 start, u64 end,
+					struct async_cow *async_cow,
+					int *num_added)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	u64 num_bytes;
+	u64 orig_start;
+	u64 disk_num_bytes;
+	u64 blocksize = root->sectorsize;
+	u64 actual_end;
+	u64 isize = i_size_read(inode);
+	int ret = 0;
+	struct page **pages = NULL;
+	unsigned long nr_pages;
+	unsigned long nr_pages_ret = 0;
+	unsigned long total_compressed = 0;
+	unsigned long total_in = 0;
+	unsigned long max_compressed = 128 * 1024;
+	unsigned long max_uncompressed = 128 * 1024;
+	int i;
+	int will_compress;
+
+	orig_start = start;
+
+	actual_end = min_t(u64, isize, end + 1);
+again:
+	will_compress = 0;
+	nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
+	nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
+
+	total_compressed = actual_end - start;
+
+	/* we want to make sure that amount of ram required to uncompress
+	 * an extent is reasonable, so we limit the total size in ram
+	 * of a compressed extent to 128k.  This is a crucial number
+	 * because it also controls how easily we can spread reads across
+	 * cpus for decompression.
+	 *
+	 * We also want to make sure the amount of IO required to do
+	 * a random read is reasonably small, so we limit the size of
+	 * a compressed extent to 128k.
+	 */
+	total_compressed = min(total_compressed, max_uncompressed);
+	num_bytes = (end - start + blocksize) & ~(blocksize - 1);
+	num_bytes = max(blocksize,  num_bytes);
+	disk_num_bytes = num_bytes;
+	total_in = 0;
+	ret = 0;
+
+	/*
+	 * we do compression for mount -o compress and when the
+	 * inode has not been flagged as nocompress.  This flag can
+	 * change at any time if we discover bad compression ratios.
+	 */
+	if (!btrfs_test_flag(inode, NOCOMPRESS) &&
+	    btrfs_test_opt(root, COMPRESS)) {
+		WARN_ON(pages);
+		pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
+
+		ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
+						total_compressed, pages,
+						nr_pages, &nr_pages_ret,
+						&total_in,
+						&total_compressed,
+						max_compressed);
+
+		if (!ret) {
+			unsigned long offset = total_compressed &
+				(PAGE_CACHE_SIZE - 1);
+			struct page *page = pages[nr_pages_ret - 1];
+			char *kaddr;
+
+			/* zero the tail end of the last page, we might be
+			 * sending it down to disk
+			 */
+			if (offset) {
+				kaddr = kmap_atomic(page, KM_USER0);
+				memset(kaddr + offset, 0,
+				       PAGE_CACHE_SIZE - offset);
+				kunmap_atomic(kaddr, KM_USER0);
+			}
+			will_compress = 1;
+		}
+	}
+	if (start == 0) {
+		trans = btrfs_join_transaction(root, 1);
+		BUG_ON(!trans);
+		btrfs_set_trans_block_group(trans, inode);
+
+		/* lets try to make an inline extent */
+		if (ret || total_in < (actual_end - start)) {
+			/* we didn't compress the entire range, try
+			 * to make an uncompressed inline extent.
+			 */
+			ret = cow_file_range_inline(trans, root, inode,
+						    start, end, 0, NULL);
+		} else {
+			/* try making a compressed inline extent */
+			ret = cow_file_range_inline(trans, root, inode,
+						    start, end,
+						    total_compressed, pages);
+		}
+		btrfs_end_transaction(trans, root);
+		if (ret == 0) {
+			/*
+			 * inline extent creation worked, we don't need
+			 * to create any more async work items.  Unlock
+			 * and free up our temp pages.
+			 */
+			extent_clear_unlock_delalloc(inode,
+						     &BTRFS_I(inode)->io_tree,
+						     start, end, NULL, 1, 0,
+						     0, 1, 1, 1);
+			ret = 0;
+			goto free_pages_out;
+		}
+	}
+
+	if (will_compress) {
+		/*
+		 * we aren't doing an inline extent round the compressed size
+		 * up to a block size boundary so the allocator does sane
+		 * things
+		 */
+		total_compressed = (total_compressed + blocksize - 1) &
+			~(blocksize - 1);
+
+		/*
+		 * one last check to make sure the compression is really a
+		 * win, compare the page count read with the blocks on disk
+		 */
+		total_in = (total_in + PAGE_CACHE_SIZE - 1) &
+			~(PAGE_CACHE_SIZE - 1);
+		if (total_compressed >= total_in) {
+			will_compress = 0;
+		} else {
+			disk_num_bytes = total_compressed;
+			num_bytes = total_in;
+		}
+	}
+	if (!will_compress && pages) {
+		/*
+		 * the compression code ran but failed to make things smaller,
+		 * free any pages it allocated and our page pointer array
+		 */
+		for (i = 0; i < nr_pages_ret; i++) {
+			WARN_ON(pages[i]->mapping);
+			page_cache_release(pages[i]);
+		}
+		kfree(pages);
+		pages = NULL;
+		total_compressed = 0;
+		nr_pages_ret = 0;
+
+		/* flag the file so we don't compress in the future */
+		btrfs_set_flag(inode, NOCOMPRESS);
+	}
+	if (will_compress) {
+		*num_added += 1;
+
+		/* the async work queues will take care of doing actual
+		 * allocation on disk for these compressed pages,
+		 * and will submit them to the elevator.
+		 */
+		add_async_extent(async_cow, start, num_bytes,
+				 total_compressed, pages, nr_pages_ret);
+
+		if (start + num_bytes < end && start + num_bytes < actual_end) {
+			start += num_bytes;
+			pages = NULL;
+			cond_resched();
+			goto again;
+		}
+	} else {
+		/*
+		 * No compression, but we still need to write the pages in
+		 * the file we've been given so far.  redirty the locked
+		 * page if it corresponds to our extent and set things up
+		 * for the async work queue to run cow_file_range to do
+		 * the normal delalloc dance
+		 */
+		if (page_offset(locked_page) >= start &&
+		    page_offset(locked_page) <= end) {
+			__set_page_dirty_nobuffers(locked_page);
+			/* unlocked later on in the async handlers */
+		}
+		add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
+		*num_added += 1;
+	}
+
+out:
+	return 0;
+
+free_pages_out:
+	for (i = 0; i < nr_pages_ret; i++) {
+		WARN_ON(pages[i]->mapping);
+		page_cache_release(pages[i]);
+	}
+	kfree(pages);
+
+	goto out;
+}
+
+/*
+ * phase two of compressed writeback.  This is the ordered portion
+ * of the code, which only gets called in the order the work was
+ * queued.  We walk all the async extents created by compress_file_range
+ * and send them down to the disk.
+ */
+static noinline int submit_compressed_extents(struct inode *inode,
+					      struct async_cow *async_cow)
+{
+	struct async_extent *async_extent;
+	u64 alloc_hint = 0;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key ins;
+	struct extent_map *em;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct extent_io_tree *io_tree;
+	int ret;
+
+	if (list_empty(&async_cow->extents))
+		return 0;
+
+	trans = btrfs_join_transaction(root, 1);
+
+	while (!list_empty(&async_cow->extents)) {
+		async_extent = list_entry(async_cow->extents.next,
+					  struct async_extent, list);
+		list_del(&async_extent->list);
+
+		io_tree = &BTRFS_I(inode)->io_tree;
+
+		/* did the compression code fall back to uncompressed IO? */
+		if (!async_extent->pages) {
+			int page_started = 0;
+			unsigned long nr_written = 0;
+
+			lock_extent(io_tree, async_extent->start,
+				    async_extent->start +
+				    async_extent->ram_size - 1, GFP_NOFS);
+
+			/* allocate blocks */
+			cow_file_range(inode, async_cow->locked_page,
+				       async_extent->start,
+				       async_extent->start +
+				       async_extent->ram_size - 1,
+				       &page_started, &nr_written, 0);
+
+			/*
+			 * if page_started, cow_file_range inserted an
+			 * inline extent and took care of all the unlocking
+			 * and IO for us.  Otherwise, we need to submit
+			 * all those pages down to the drive.
+			 */
+			if (!page_started)
+				extent_write_locked_range(io_tree,
+						  inode, async_extent->start,
+						  async_extent->start +
+						  async_extent->ram_size - 1,
+						  btrfs_get_extent,
+						  WB_SYNC_ALL);
+			kfree(async_extent);
+			cond_resched();
+			continue;
+		}
+
+		lock_extent(io_tree, async_extent->start,
+			    async_extent->start + async_extent->ram_size - 1,
+			    GFP_NOFS);
+		/*
+		 * here we're doing allocation and writeback of the
+		 * compressed pages
+		 */
+		btrfs_drop_extent_cache(inode, async_extent->start,
+					async_extent->start +
+					async_extent->ram_size - 1, 0);
+
+		ret = btrfs_reserve_extent(trans, root,
+					   async_extent->compressed_size,
+					   async_extent->compressed_size,
+					   0, alloc_hint,
+					   (u64)-1, &ins, 1);
+		BUG_ON(ret);
+		em = alloc_extent_map(GFP_NOFS);
+		em->start = async_extent->start;
+		em->len = async_extent->ram_size;
+		em->orig_start = em->start;
+
+		em->block_start = ins.objectid;
+		em->block_len = ins.offset;
+		em->bdev = root->fs_info->fs_devices->latest_bdev;
+		set_bit(EXTENT_FLAG_PINNED, &em->flags);
+		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
+
+		while (1) {
+			spin_lock(&em_tree->lock);
+			ret = add_extent_mapping(em_tree, em);
+			spin_unlock(&em_tree->lock);
+			if (ret != -EEXIST) {
+				free_extent_map(em);
+				break;
+			}
+			btrfs_drop_extent_cache(inode, async_extent->start,
+						async_extent->start +
+						async_extent->ram_size - 1, 0);
+		}
+
+		ret = btrfs_add_ordered_extent(inode, async_extent->start,
+					       ins.objectid,
+					       async_extent->ram_size,
+					       ins.offset,
+					       BTRFS_ORDERED_COMPRESSED);
+		BUG_ON(ret);
+
+		btrfs_end_transaction(trans, root);
+
+		/*
+		 * clear dirty, set writeback and unlock the pages.
+		 */
+		extent_clear_unlock_delalloc(inode,
+					     &BTRFS_I(inode)->io_tree,
+					     async_extent->start,
+					     async_extent->start +
+					     async_extent->ram_size - 1,
+					     NULL, 1, 1, 0, 1, 1, 0);
+
+		ret = btrfs_submit_compressed_write(inode,
+				    async_extent->start,
+				    async_extent->ram_size,
+				    ins.objectid,
+				    ins.offset, async_extent->pages,
+				    async_extent->nr_pages);
+
+		BUG_ON(ret);
+		trans = btrfs_join_transaction(root, 1);
+		alloc_hint = ins.objectid + ins.offset;
+		kfree(async_extent);
+		cond_resched();
+	}
+
+	btrfs_end_transaction(trans, root);
+	return 0;
+}
+
+/*
+ * when extent_io.c finds a delayed allocation range in the file,
+ * the call backs end up in this code.  The basic idea is to
+ * allocate extents on disk for the range, and create ordered data structs
+ * in ram to track those extents.
+ *
+ * locked_page is the page that writepage had locked already.  We use
+ * it to make sure we don't do extra locks or unlocks.
+ *
+ * *page_started is set to one if we unlock locked_page and do everything
+ * required to start IO on it.  It may be clean and already done with
+ * IO when we return.
+ */
+static noinline int cow_file_range(struct inode *inode,
+				   struct page *locked_page,
+				   u64 start, u64 end, int *page_started,
+				   unsigned long *nr_written,
+				   int unlock)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	u64 alloc_hint = 0;
+	u64 num_bytes;
+	unsigned long ram_size;
+	u64 disk_num_bytes;
+	u64 cur_alloc_size;
+	u64 blocksize = root->sectorsize;
+	u64 actual_end;
+	u64 isize = i_size_read(inode);
+	struct btrfs_key ins;
+	struct extent_map *em;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	int ret = 0;
+
+	trans = btrfs_join_transaction(root, 1);
+	BUG_ON(!trans);
+	btrfs_set_trans_block_group(trans, inode);
+
+	actual_end = min_t(u64, isize, end + 1);
+
+	num_bytes = (end - start + blocksize) & ~(blocksize - 1);
+	num_bytes = max(blocksize,  num_bytes);
+	disk_num_bytes = num_bytes;
+	ret = 0;
+
+	if (start == 0) {
+		/* lets try to make an inline extent */
+		ret = cow_file_range_inline(trans, root, inode,
+					    start, end, 0, NULL);
+		if (ret == 0) {
+			extent_clear_unlock_delalloc(inode,
+						     &BTRFS_I(inode)->io_tree,
+						     start, end, NULL, 1, 1,
+						     1, 1, 1, 1);
+			*nr_written = *nr_written +
+			     (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
+			*page_started = 1;
+			ret = 0;
+			goto out;
+		}
+	}
+
+	BUG_ON(disk_num_bytes >
+	       btrfs_super_total_bytes(&root->fs_info->super_copy));
+
+	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
+
+	while (disk_num_bytes > 0) {
+		cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
+		ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
+					   root->sectorsize, 0, alloc_hint,
+					   (u64)-1, &ins, 1);
+		BUG_ON(ret);
+
+		em = alloc_extent_map(GFP_NOFS);
+		em->start = start;
+		em->orig_start = em->start;
+
+		ram_size = ins.offset;
+		em->len = ins.offset;
+
+		em->block_start = ins.objectid;
+		em->block_len = ins.offset;
+		em->bdev = root->fs_info->fs_devices->latest_bdev;
+		set_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+		while (1) {
+			spin_lock(&em_tree->lock);
+			ret = add_extent_mapping(em_tree, em);
+			spin_unlock(&em_tree->lock);
+			if (ret != -EEXIST) {
+				free_extent_map(em);
+				break;
+			}
+			btrfs_drop_extent_cache(inode, start,
+						start + ram_size - 1, 0);
+		}
+
+		cur_alloc_size = ins.offset;
+		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
+					       ram_size, cur_alloc_size, 0);
+		BUG_ON(ret);
+
+		if (root->root_key.objectid ==
+		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
+			ret = btrfs_reloc_clone_csums(inode, start,
+						      cur_alloc_size);
+			BUG_ON(ret);
+		}
+
+		if (disk_num_bytes < cur_alloc_size)
+			break;
+
+		/* we're not doing compressed IO, don't unlock the first
+		 * page (which the caller expects to stay locked), don't
+		 * clear any dirty bits and don't set any writeback bits
+		 */
+		extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
+					     start, start + ram_size - 1,
+					     locked_page, unlock, 1,
+					     1, 0, 0, 0);
+		disk_num_bytes -= cur_alloc_size;
+		num_bytes -= cur_alloc_size;
+		alloc_hint = ins.objectid + ins.offset;
+		start += cur_alloc_size;
+	}
+out:
+	ret = 0;
+	btrfs_end_transaction(trans, root);
+
+	return ret;
+}
+
+/*
+ * work queue call back to started compression on a file and pages
+ */
+static noinline void async_cow_start(struct btrfs_work *work)
+{
+	struct async_cow *async_cow;
+	int num_added = 0;
+	async_cow = container_of(work, struct async_cow, work);
+
+	compress_file_range(async_cow->inode, async_cow->locked_page,
+			    async_cow->start, async_cow->end, async_cow,
+			    &num_added);
+	if (num_added == 0)
+		async_cow->inode = NULL;
+}
+
+/*
+ * work queue call back to submit previously compressed pages
+ */
+static noinline void async_cow_submit(struct btrfs_work *work)
+{
+	struct async_cow *async_cow;
+	struct btrfs_root *root;
+	unsigned long nr_pages;
+
+	async_cow = container_of(work, struct async_cow, work);
+
+	root = async_cow->root;
+	nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
+		PAGE_CACHE_SHIFT;
+
+	atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
+
+	if (atomic_read(&root->fs_info->async_delalloc_pages) <
+	    5 * 1042 * 1024 &&
+	    waitqueue_active(&root->fs_info->async_submit_wait))
+		wake_up(&root->fs_info->async_submit_wait);
+
+	if (async_cow->inode)
+		submit_compressed_extents(async_cow->inode, async_cow);
+}
+
+static noinline void async_cow_free(struct btrfs_work *work)
+{
+	struct async_cow *async_cow;
+	async_cow = container_of(work, struct async_cow, work);
+	kfree(async_cow);
+}
+
+static int cow_file_range_async(struct inode *inode, struct page *locked_page,
+				u64 start, u64 end, int *page_started,
+				unsigned long *nr_written)
+{
+	struct async_cow *async_cow;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	unsigned long nr_pages;
+	u64 cur_end;
+	int limit = 10 * 1024 * 1042;
+
+	if (!btrfs_test_opt(root, COMPRESS)) {
+		return cow_file_range(inode, locked_page, start, end,
+				      page_started, nr_written, 1);
+	}
+
+	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
+			 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
+	while (start < end) {
+		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
+		async_cow->inode = inode;
+		async_cow->root = root;
+		async_cow->locked_page = locked_page;
+		async_cow->start = start;
+
+		if (btrfs_test_flag(inode, NOCOMPRESS))
+			cur_end = end;
+		else
+			cur_end = min(end, start + 512 * 1024 - 1);
+
+		async_cow->end = cur_end;
+		INIT_LIST_HEAD(&async_cow->extents);
+
+		async_cow->work.func = async_cow_start;
+		async_cow->work.ordered_func = async_cow_submit;
+		async_cow->work.ordered_free = async_cow_free;
+		async_cow->work.flags = 0;
+
+		nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
+			PAGE_CACHE_SHIFT;
+		atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
+
+		btrfs_queue_worker(&root->fs_info->delalloc_workers,
+				   &async_cow->work);
+
+		if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
+			wait_event(root->fs_info->async_submit_wait,
+			   (atomic_read(&root->fs_info->async_delalloc_pages) <
+			    limit));
+		}
+
+		while (atomic_read(&root->fs_info->async_submit_draining) &&
+		      atomic_read(&root->fs_info->async_delalloc_pages)) {
+			wait_event(root->fs_info->async_submit_wait,
+			  (atomic_read(&root->fs_info->async_delalloc_pages) ==
+			   0));
+		}
+
+		*nr_written += nr_pages;
+		start = cur_end + 1;
+	}
+	*page_started = 1;
+	return 0;
+}
+
+static noinline int csum_exist_in_range(struct btrfs_root *root,
+					u64 bytenr, u64 num_bytes)
+{
+	int ret;
+	struct btrfs_ordered_sum *sums;
+	LIST_HEAD(list);
+
+	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
+				       bytenr + num_bytes - 1, &list);
+	if (ret == 0 && list_empty(&list))
+		return 0;
+
+	while (!list_empty(&list)) {
+		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
+		list_del(&sums->list);
+		kfree(sums);
+	}
+	return 1;
+}
+
+/*
+ * when nowcow writeback call back.  This checks for snapshots or COW copies
+ * of the extents that exist in the file, and COWs the file as required.
+ *
+ * If no cow copies or snapshots exist, we write directly to the existing
+ * blocks on disk
+ */
+static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
+			      u64 start, u64 end, int *page_started, int force,
+			      unsigned long *nr_written)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	struct extent_buffer *leaf;
+	struct btrfs_path *path;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_key found_key;
+	u64 cow_start;
+	u64 cur_offset;
+	u64 extent_end;
+	u64 disk_bytenr;
+	u64 num_bytes;
+	int extent_type;
+	int ret;
+	int type;
+	int nocow;
+	int check_prev = 1;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	trans = btrfs_join_transaction(root, 1);
+	BUG_ON(!trans);
+
+	cow_start = (u64)-1;
+	cur_offset = start;
+	while (1) {
+		ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
+					       cur_offset, 0);
+		BUG_ON(ret < 0);
+		if (ret > 0 && path->slots[0] > 0 && check_prev) {
+			leaf = path->nodes[0];
+			btrfs_item_key_to_cpu(leaf, &found_key,
+					      path->slots[0] - 1);
+			if (found_key.objectid == inode->i_ino &&
+			    found_key.type == BTRFS_EXTENT_DATA_KEY)
+				path->slots[0]--;
+		}
+		check_prev = 0;
+next_slot:
+		leaf = path->nodes[0];
+		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				BUG_ON(1);
+			if (ret > 0)
+				break;
+			leaf = path->nodes[0];
+		}
+
+		nocow = 0;
+		disk_bytenr = 0;
+		num_bytes = 0;
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+		if (found_key.objectid > inode->i_ino ||
+		    found_key.type > BTRFS_EXTENT_DATA_KEY ||
+		    found_key.offset > end)
+			break;
+
+		if (found_key.offset > cur_offset) {
+			extent_end = found_key.offset;
+			goto out_check;
+		}
+
+		fi = btrfs_item_ptr(leaf, path->slots[0],
+				    struct btrfs_file_extent_item);
+		extent_type = btrfs_file_extent_type(leaf, fi);
+
+		if (extent_type == BTRFS_FILE_EXTENT_REG ||
+		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
+			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+			extent_end = found_key.offset +
+				btrfs_file_extent_num_bytes(leaf, fi);
+			if (extent_end <= start) {
+				path->slots[0]++;
+				goto next_slot;
+			}
+			if (disk_bytenr == 0)
+				goto out_check;
+			if (btrfs_file_extent_compression(leaf, fi) ||
+			    btrfs_file_extent_encryption(leaf, fi) ||
+			    btrfs_file_extent_other_encoding(leaf, fi))
+				goto out_check;
+			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
+				goto out_check;
+			if (btrfs_extent_readonly(root, disk_bytenr))
+				goto out_check;
+			if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
+						  disk_bytenr))
+				goto out_check;
+			disk_bytenr += btrfs_file_extent_offset(leaf, fi);
+			disk_bytenr += cur_offset - found_key.offset;
+			num_bytes = min(end + 1, extent_end) - cur_offset;
+			/*
+			 * force cow if csum exists in the range.
+			 * this ensure that csum for a given extent are
+			 * either valid or do not exist.
+			 */
+			if (csum_exist_in_range(root, disk_bytenr, num_bytes))
+				goto out_check;
+			nocow = 1;
+		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+			extent_end = found_key.offset +
+				btrfs_file_extent_inline_len(leaf, fi);
+			extent_end = ALIGN(extent_end, root->sectorsize);
+		} else {
+			BUG_ON(1);
+		}
+out_check:
+		if (extent_end <= start) {
+			path->slots[0]++;
+			goto next_slot;
+		}
+		if (!nocow) {
+			if (cow_start == (u64)-1)
+				cow_start = cur_offset;
+			cur_offset = extent_end;
+			if (cur_offset > end)
+				break;
+			path->slots[0]++;
+			goto next_slot;
+		}
+
+		btrfs_release_path(root, path);
+		if (cow_start != (u64)-1) {
+			ret = cow_file_range(inode, locked_page, cow_start,
+					found_key.offset - 1, page_started,
+					nr_written, 1);
+			BUG_ON(ret);
+			cow_start = (u64)-1;
+		}
+
+		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
+			struct extent_map *em;
+			struct extent_map_tree *em_tree;
+			em_tree = &BTRFS_I(inode)->extent_tree;
+			em = alloc_extent_map(GFP_NOFS);
+			em->start = cur_offset;
+			em->orig_start = em->start;
+			em->len = num_bytes;
+			em->block_len = num_bytes;
+			em->block_start = disk_bytenr;
+			em->bdev = root->fs_info->fs_devices->latest_bdev;
+			set_bit(EXTENT_FLAG_PINNED, &em->flags);
+			while (1) {
+				spin_lock(&em_tree->lock);
+				ret = add_extent_mapping(em_tree, em);
+				spin_unlock(&em_tree->lock);
+				if (ret != -EEXIST) {
+					free_extent_map(em);
+					break;
+				}
+				btrfs_drop_extent_cache(inode, em->start,
+						em->start + em->len - 1, 0);
+			}
+			type = BTRFS_ORDERED_PREALLOC;
+		} else {
+			type = BTRFS_ORDERED_NOCOW;
+		}
+
+		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
+					       num_bytes, num_bytes, type);
+		BUG_ON(ret);
+
+		extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
+					cur_offset, cur_offset + num_bytes - 1,
+					locked_page, 1, 1, 1, 0, 0, 0);
+		cur_offset = extent_end;
+		if (cur_offset > end)
+			break;
+	}
+	btrfs_release_path(root, path);
+
+	if (cur_offset <= end && cow_start == (u64)-1)
+		cow_start = cur_offset;
+	if (cow_start != (u64)-1) {
+		ret = cow_file_range(inode, locked_page, cow_start, end,
+				     page_started, nr_written, 1);
+		BUG_ON(ret);
+	}
+
+	ret = btrfs_end_transaction(trans, root);
+	BUG_ON(ret);
+	btrfs_free_path(path);
+	return 0;
+}
+
+/*
+ * extent_io.c call back to do delayed allocation processing
+ */
+static int run_delalloc_range(struct inode *inode, struct page *locked_page,
+			      u64 start, u64 end, int *page_started,
+			      unsigned long *nr_written)
+{
+	int ret;
+
+	if (btrfs_test_flag(inode, NODATACOW))
+		ret = run_delalloc_nocow(inode, locked_page, start, end,
+					 page_started, 1, nr_written);
+	else if (btrfs_test_flag(inode, PREALLOC))
+		ret = run_delalloc_nocow(inode, locked_page, start, end,
+					 page_started, 0, nr_written);
+	else
+		ret = cow_file_range_async(inode, locked_page, start, end,
+					   page_started, nr_written);
+
+	return ret;
+}
+
+/*
+ * extent_io.c set_bit_hook, used to track delayed allocation
+ * bytes in this file, and to maintain the list of inodes that
+ * have pending delalloc work to be done.
+ */
+static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
+		       unsigned long old, unsigned long bits)
+{
+	/*
+	 * set_bit and clear bit hooks normally require _irqsave/restore
+	 * but in this case, we are only testeing for the DELALLOC
+	 * bit, which is only set or cleared with irqs on
+	 */
+	if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
+		struct btrfs_root *root = BTRFS_I(inode)->root;
+		spin_lock(&root->fs_info->delalloc_lock);
+		BTRFS_I(inode)->delalloc_bytes += end - start + 1;
+		root->fs_info->delalloc_bytes += end - start + 1;
+		if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
+			list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
+				      &root->fs_info->delalloc_inodes);
+		}
+		spin_unlock(&root->fs_info->delalloc_lock);
+	}
+	return 0;
+}
+
+/*
+ * extent_io.c clear_bit_hook, see set_bit_hook for why
+ */
+static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
+			 unsigned long old, unsigned long bits)
+{
+	/*
+	 * set_bit and clear bit hooks normally require _irqsave/restore
+	 * but in this case, we are only testeing for the DELALLOC
+	 * bit, which is only set or cleared with irqs on
+	 */
+	if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
+		struct btrfs_root *root = BTRFS_I(inode)->root;
+
+		spin_lock(&root->fs_info->delalloc_lock);
+		if (end - start + 1 > root->fs_info->delalloc_bytes) {
+			printk(KERN_INFO "btrfs warning: delalloc account "
+			       "%llu %llu\n",
+			       (unsigned long long)end - start + 1,
+			       (unsigned long long)
+			       root->fs_info->delalloc_bytes);
+			root->fs_info->delalloc_bytes = 0;
+			BTRFS_I(inode)->delalloc_bytes = 0;
+		} else {
+			root->fs_info->delalloc_bytes -= end - start + 1;
+			BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
+		}
+		if (BTRFS_I(inode)->delalloc_bytes == 0 &&
+		    !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
+			list_del_init(&BTRFS_I(inode)->delalloc_inodes);
+		}
+		spin_unlock(&root->fs_info->delalloc_lock);
+	}
+	return 0;
+}
+
+/*
+ * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
+ * we don't create bios that span stripes or chunks
+ */
+int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
+			 size_t size, struct bio *bio,
+			 unsigned long bio_flags)
+{
+	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
+	struct btrfs_mapping_tree *map_tree;
+	u64 logical = (u64)bio->bi_sector << 9;
+	u64 length = 0;
+	u64 map_length;
+	int ret;
+
+	if (bio_flags & EXTENT_BIO_COMPRESSED)
+		return 0;
+
+	length = bio->bi_size;
+	map_tree = &root->fs_info->mapping_tree;
+	map_length = length;
+	ret = btrfs_map_block(map_tree, READ, logical,
+			      &map_length, NULL, 0);
+
+	if (map_length < length + size)
+		return 1;
+	return 0;
+}
+
+/*
+ * in order to insert checksums into the metadata in large chunks,
+ * we wait until bio submission time.   All the pages in the bio are
+ * checksummed and sums are attached onto the ordered extent record.
+ *
+ * At IO completion time the cums attached on the ordered extent record
+ * are inserted into the btree
+ */
+static int __btrfs_submit_bio_start(struct inode *inode, int rw,
+				    struct bio *bio, int mirror_num,
+				    unsigned long bio_flags)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret = 0;
+
+	ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
+	BUG_ON(ret);
+	return 0;
+}
+
+/*
+ * in order to insert checksums into the metadata in large chunks,
+ * we wait until bio submission time.   All the pages in the bio are
+ * checksummed and sums are attached onto the ordered extent record.
+ *
+ * At IO completion time the cums attached on the ordered extent record
+ * are inserted into the btree
+ */
+static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
+			  int mirror_num, unsigned long bio_flags)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	return btrfs_map_bio(root, rw, bio, mirror_num, 1);
+}
+
+/*
+ * extent_io.c submission hook. This does the right thing for csum calculation
+ * on write, or reading the csums from the tree before a read
+ */
+static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
+			  int mirror_num, unsigned long bio_flags)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret = 0;
+	int skip_sum;
+
+	skip_sum = btrfs_test_flag(inode, NODATASUM);
+
+	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
+	BUG_ON(ret);
+
+	if (!(rw & (1 << BIO_RW))) {
+		if (bio_flags & EXTENT_BIO_COMPRESSED) {
+			return btrfs_submit_compressed_read(inode, bio,
+						    mirror_num, bio_flags);
+		} else if (!skip_sum)
+			btrfs_lookup_bio_sums(root, inode, bio, NULL);
+		goto mapit;
+	} else if (!skip_sum) {
+		/* csum items have already been cloned */
+		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
+			goto mapit;
+		/* we're doing a write, do the async checksumming */
+		return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
+				   inode, rw, bio, mirror_num,
+				   bio_flags, __btrfs_submit_bio_start,
+				   __btrfs_submit_bio_done);
+	}
+
+mapit:
+	return btrfs_map_bio(root, rw, bio, mirror_num, 0);
+}
+
+/*
+ * given a list of ordered sums record them in the inode.  This happens
+ * at IO completion time based on sums calculated at bio submission time.
+ */
+static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
+			     struct inode *inode, u64 file_offset,
+			     struct list_head *list)
+{
+	struct list_head *cur;
+	struct btrfs_ordered_sum *sum;
+
+	btrfs_set_trans_block_group(trans, inode);
+	list_for_each(cur, list) {
+		sum = list_entry(cur, struct btrfs_ordered_sum, list);
+		btrfs_csum_file_blocks(trans,
+		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
+	}
+	return 0;
+}
+
+int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
+{
+	if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
+		WARN_ON(1);
+	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
+				   GFP_NOFS);
+}
+
+/* see btrfs_writepage_start_hook for details on why this is required */
+struct btrfs_writepage_fixup {
+	struct page *page;
+	struct btrfs_work work;
+};
+
+static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
+{
+	struct btrfs_writepage_fixup *fixup;
+	struct btrfs_ordered_extent *ordered;
+	struct page *page;
+	struct inode *inode;
+	u64 page_start;
+	u64 page_end;
+
+	fixup = container_of(work, struct btrfs_writepage_fixup, work);
+	page = fixup->page;
+again:
+	lock_page(page);
+	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
+		ClearPageChecked(page);
+		goto out_page;
+	}
+
+	inode = page->mapping->host;
+	page_start = page_offset(page);
+	page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
+
+	lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
+
+	/* already ordered? We're done */
+	if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
+			     EXTENT_ORDERED, 0)) {
+		goto out;
+	}
+
+	ordered = btrfs_lookup_ordered_extent(inode, page_start);
+	if (ordered) {
+		unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
+			      page_end, GFP_NOFS);
+		unlock_page(page);
+		btrfs_start_ordered_extent(inode, ordered, 1);
+		goto again;
+	}
+
+	btrfs_set_extent_delalloc(inode, page_start, page_end);
+	ClearPageChecked(page);
+out:
+	unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
+out_page:
+	unlock_page(page);
+	page_cache_release(page);
+}
+
+/*
+ * There are a few paths in the higher layers of the kernel that directly
+ * set the page dirty bit without asking the filesystem if it is a
+ * good idea.  This causes problems because we want to make sure COW
+ * properly happens and the data=ordered rules are followed.
+ *
+ * In our case any range that doesn't have the ORDERED bit set
+ * hasn't been properly setup for IO.  We kick off an async process
+ * to fix it up.  The async helper will wait for ordered extents, set
+ * the delalloc bit and make it safe to write the page.
+ */
+static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
+{
+	struct inode *inode = page->mapping->host;
+	struct btrfs_writepage_fixup *fixup;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret;
+
+	ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
+			     EXTENT_ORDERED, 0);
+	if (ret)
+		return 0;
+
+	if (PageChecked(page))
+		return -EAGAIN;
+
+	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
+	if (!fixup)
+		return -EAGAIN;
+
+	SetPageChecked(page);
+	page_cache_get(page);
+	fixup->work.func = btrfs_writepage_fixup_worker;
+	fixup->page = page;
+	btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
+	return -EAGAIN;
+}
+
 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 				       struct inode *inode, u64 file_pos,
 				       u64 disk_bytenr, u64 disk_num_bytes,
@@ -175,6 +1493,291 @@
 	return 0;
 }
 
+/* as ordered data IO finishes, this gets called so we can finish
+ * an ordered extent if the range of bytes in the file it covers are
+ * fully written.
+ */
+static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_ordered_extent *ordered_extent;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	int compressed = 0;
+	int ret;
+
+	ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
+	if (!ret)
+		return 0;
+
+	trans = btrfs_join_transaction(root, 1);
+
+	ordered_extent = btrfs_lookup_ordered_extent(inode, start);
+	BUG_ON(!ordered_extent);
+	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
+		goto nocow;
+
+	lock_extent(io_tree, ordered_extent->file_offset,
+		    ordered_extent->file_offset + ordered_extent->len - 1,
+		    GFP_NOFS);
+
+	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
+		compressed = 1;
+	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
+		BUG_ON(compressed);
+		ret = btrfs_mark_extent_written(trans, root, inode,
+						ordered_extent->file_offset,
+						ordered_extent->file_offset +
+						ordered_extent->len);
+		BUG_ON(ret);
+	} else {
+		ret = insert_reserved_file_extent(trans, inode,
+						ordered_extent->file_offset,
+						ordered_extent->start,
+						ordered_extent->disk_len,
+						ordered_extent->len,
+						ordered_extent->len,
+						compressed, 0, 0,
+						BTRFS_FILE_EXTENT_REG);
+		BUG_ON(ret);
+	}
+	unlock_extent(io_tree, ordered_extent->file_offset,
+		    ordered_extent->file_offset + ordered_extent->len - 1,
+		    GFP_NOFS);
+nocow:
+	add_pending_csums(trans, inode, ordered_extent->file_offset,
+			  &ordered_extent->list);
+
+	mutex_lock(&BTRFS_I(inode)->extent_mutex);
+	btrfs_ordered_update_i_size(inode, ordered_extent);
+	btrfs_update_inode(trans, root, inode);
+	btrfs_remove_ordered_extent(inode, ordered_extent);
+	mutex_unlock(&BTRFS_I(inode)->extent_mutex);
+
+	/* once for us */
+	btrfs_put_ordered_extent(ordered_extent);
+	/* once for the tree */
+	btrfs_put_ordered_extent(ordered_extent);
+
+	btrfs_end_transaction(trans, root);
+	return 0;
+}
+
+static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
+				struct extent_state *state, int uptodate)
+{
+	return btrfs_finish_ordered_io(page->mapping->host, start, end);
+}
+
+/*
+ * When IO fails, either with EIO or csum verification fails, we
+ * try other mirrors that might have a good copy of the data.  This
+ * io_failure_record is used to record state as we go through all the
+ * mirrors.  If another mirror has good data, the page is set up to date
+ * and things continue.  If a good mirror can't be found, the original
+ * bio end_io callback is called to indicate things have failed.
+ */
+struct io_failure_record {
+	struct page *page;
+	u64 start;
+	u64 len;
+	u64 logical;
+	unsigned long bio_flags;
+	int last_mirror;
+};
+
+static int btrfs_io_failed_hook(struct bio *failed_bio,
+			 struct page *page, u64 start, u64 end,
+			 struct extent_state *state)
+{
+	struct io_failure_record *failrec = NULL;
+	u64 private;
+	struct extent_map *em;
+	struct inode *inode = page->mapping->host;
+	struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct bio *bio;
+	int num_copies;
+	int ret;
+	int rw;
+	u64 logical;
+
+	ret = get_state_private(failure_tree, start, &private);
+	if (ret) {
+		failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
+		if (!failrec)
+			return -ENOMEM;
+		failrec->start = start;
+		failrec->len = end - start + 1;
+		failrec->last_mirror = 0;
+		failrec->bio_flags = 0;
+
+		spin_lock(&em_tree->lock);
+		em = lookup_extent_mapping(em_tree, start, failrec->len);
+		if (em->start > start || em->start + em->len < start) {
+			free_extent_map(em);
+			em = NULL;
+		}
+		spin_unlock(&em_tree->lock);
+
+		if (!em || IS_ERR(em)) {
+			kfree(failrec);
+			return -EIO;
+		}
+		logical = start - em->start;
+		logical = em->block_start + logical;
+		if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
+			logical = em->block_start;
+			failrec->bio_flags = EXTENT_BIO_COMPRESSED;
+		}
+		failrec->logical = logical;
+		free_extent_map(em);
+		set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
+				EXTENT_DIRTY, GFP_NOFS);
+		set_state_private(failure_tree, start,
+				 (u64)(unsigned long)failrec);
+	} else {
+		failrec = (struct io_failure_record *)(unsigned long)private;
+	}
+	num_copies = btrfs_num_copies(
+			      &BTRFS_I(inode)->root->fs_info->mapping_tree,
+			      failrec->logical, failrec->len);
+	failrec->last_mirror++;
+	if (!state) {
+		spin_lock(&BTRFS_I(inode)->io_tree.lock);
+		state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
+						    failrec->start,
+						    EXTENT_LOCKED);
+		if (state && state->start != failrec->start)
+			state = NULL;
+		spin_unlock(&BTRFS_I(inode)->io_tree.lock);
+	}
+	if (!state || failrec->last_mirror > num_copies) {
+		set_state_private(failure_tree, failrec->start, 0);
+		clear_extent_bits(failure_tree, failrec->start,
+				  failrec->start + failrec->len - 1,
+				  EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
+		kfree(failrec);
+		return -EIO;
+	}
+	bio = bio_alloc(GFP_NOFS, 1);
+	bio->bi_private = state;
+	bio->bi_end_io = failed_bio->bi_end_io;
+	bio->bi_sector = failrec->logical >> 9;
+	bio->bi_bdev = failed_bio->bi_bdev;
+	bio->bi_size = 0;
+
+	bio_add_page(bio, page, failrec->len, start - page_offset(page));
+	if (failed_bio->bi_rw & (1 << BIO_RW))
+		rw = WRITE;
+	else
+		rw = READ;
+
+	BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
+						      failrec->last_mirror,
+						      failrec->bio_flags);
+	return 0;
+}
+
+/*
+ * each time an IO finishes, we do a fast check in the IO failure tree
+ * to see if we need to process or clean up an io_failure_record
+ */
+static int btrfs_clean_io_failures(struct inode *inode, u64 start)
+{
+	u64 private;
+	u64 private_failure;
+	struct io_failure_record *failure;
+	int ret;
+
+	private = 0;
+	if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
+			     (u64)-1, 1, EXTENT_DIRTY)) {
+		ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
+					start, &private_failure);
+		if (ret == 0) {
+			failure = (struct io_failure_record *)(unsigned long)
+				   private_failure;
+			set_state_private(&BTRFS_I(inode)->io_failure_tree,
+					  failure->start, 0);
+			clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
+					  failure->start,
+					  failure->start + failure->len - 1,
+					  EXTENT_DIRTY | EXTENT_LOCKED,
+					  GFP_NOFS);
+			kfree(failure);
+		}
+	}
+	return 0;
+}
+
+/*
+ * when reads are done, we need to check csums to verify the data is correct
+ * if there's a match, we allow the bio to finish.  If not, we go through
+ * the io_failure_record routines to find good copies
+ */
+static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
+			       struct extent_state *state)
+{
+	size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
+	struct inode *inode = page->mapping->host;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	char *kaddr;
+	u64 private = ~(u32)0;
+	int ret;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	u32 csum = ~(u32)0;
+
+	if (PageChecked(page)) {
+		ClearPageChecked(page);
+		goto good;
+	}
+	if (btrfs_test_flag(inode, NODATASUM))
+		return 0;
+
+	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
+	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
+		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
+				  GFP_NOFS);
+		return 0;
+	}
+
+	if (state && state->start == start) {
+		private = state->private;
+		ret = 0;
+	} else {
+		ret = get_state_private(io_tree, start, &private);
+	}
+	kaddr = kmap_atomic(page, KM_USER0);
+	if (ret)
+		goto zeroit;
+
+	csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
+	btrfs_csum_final(csum, (char *)&csum);
+	if (csum != private)
+		goto zeroit;
+
+	kunmap_atomic(kaddr, KM_USER0);
+good:
+	/* if the io failure tree for this inode is non-empty,
+	 * check to see if we've recovered from a failed IO
+	 */
+	btrfs_clean_io_failures(inode, start);
+	return 0;
+
+zeroit:
+	printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
+	       "private %llu\n", page->mapping->host->i_ino,
+	       (unsigned long long)start, csum,
+	       (unsigned long long)private);
+	memset(kaddr + offset, 1, end - start + 1);
+	flush_dcache_page(page);
+	kunmap_atomic(kaddr, KM_USER0);
+	if (private == 0)
+		return 0;
+	return -EIO;
+}
+
 /*
  * This creates an orphan entry for the given inode in case something goes
  * wrong in the middle of an unlink/truncate.
@@ -2035,6 +3638,349 @@
 	return err;
 }
 
+/* helper for btfs_get_extent.  Given an existing extent in the tree,
+ * and an extent that you want to insert, deal with overlap and insert
+ * the new extent into the tree.
+ */
+static int merge_extent_mapping(struct extent_map_tree *em_tree,
+				struct extent_map *existing,
+				struct extent_map *em,
+				u64 map_start, u64 map_len)
+{
+	u64 start_diff;
+
+	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
+	start_diff = map_start - em->start;
+	em->start = map_start;
+	em->len = map_len;
+	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
+	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
+		em->block_start += start_diff;
+		em->block_len -= start_diff;
+	}
+	return add_extent_mapping(em_tree, em);
+}
+
+static noinline int uncompress_inline(struct btrfs_path *path,
+				      struct inode *inode, struct page *page,
+				      size_t pg_offset, u64 extent_offset,
+				      struct btrfs_file_extent_item *item)
+{
+	int ret;
+	struct extent_buffer *leaf = path->nodes[0];
+	char *tmp;
+	size_t max_size;
+	unsigned long inline_size;
+	unsigned long ptr;
+
+	WARN_ON(pg_offset != 0);
+	max_size = btrfs_file_extent_ram_bytes(leaf, item);
+	inline_size = btrfs_file_extent_inline_item_len(leaf,
+					btrfs_item_nr(leaf, path->slots[0]));
+	tmp = kmalloc(inline_size, GFP_NOFS);
+	ptr = btrfs_file_extent_inline_start(item);
+
+	read_extent_buffer(leaf, tmp, ptr, inline_size);
+
+	max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
+	ret = btrfs_zlib_decompress(tmp, page, extent_offset,
+				    inline_size, max_size);
+	if (ret) {
+		char *kaddr = kmap_atomic(page, KM_USER0);
+		unsigned long copy_size = min_t(u64,
+				  PAGE_CACHE_SIZE - pg_offset,
+				  max_size - extent_offset);
+		memset(kaddr + pg_offset, 0, copy_size);
+		kunmap_atomic(kaddr, KM_USER0);
+	}
+	kfree(tmp);
+	return 0;
+}
+
+/*
+ * a bit scary, this does extent mapping from logical file offset to the disk.
+ * the ugly parts come from merging extents from the disk with the in-ram
+ * representation.  This gets more complex because of the data=ordered code,
+ * where the in-ram extents might be locked pending data=ordered completion.
+ *
+ * This also copies inline extents directly into the page.
+ */
+
+struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
+				    size_t pg_offset, u64 start, u64 len,
+				    int create)
+{
+	int ret;
+	int err = 0;
+	u64 bytenr;
+	u64 extent_start = 0;
+	u64 extent_end = 0;
+	u64 objectid = inode->i_ino;
+	u32 found_type;
+	struct btrfs_path *path = NULL;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_file_extent_item *item;
+	struct extent_buffer *leaf;
+	struct btrfs_key found_key;
+	struct extent_map *em = NULL;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_trans_handle *trans = NULL;
+	int compressed;
+
+again:
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, start, len);
+	if (em)
+		em->bdev = root->fs_info->fs_devices->latest_bdev;
+	spin_unlock(&em_tree->lock);
+
+	if (em) {
+		if (em->start > start || em->start + em->len <= start)
+			free_extent_map(em);
+		else if (em->block_start == EXTENT_MAP_INLINE && page)
+			free_extent_map(em);
+		else
+			goto out;
+	}
+	em = alloc_extent_map(GFP_NOFS);
+	if (!em) {
+		err = -ENOMEM;
+		goto out;
+	}
+	em->bdev = root->fs_info->fs_devices->latest_bdev;
+	em->start = EXTENT_MAP_HOLE;
+	em->orig_start = EXTENT_MAP_HOLE;
+	em->len = (u64)-1;
+	em->block_len = (u64)-1;
+
+	if (!path) {
+		path = btrfs_alloc_path();
+		BUG_ON(!path);
+	}
+
+	ret = btrfs_lookup_file_extent(trans, root, path,
+				       objectid, start, trans != NULL);
+	if (ret < 0) {
+		err = ret;
+		goto out;
+	}
+
+	if (ret != 0) {
+		if (path->slots[0] == 0)
+			goto not_found;
+		path->slots[0]--;
+	}
+
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0],
+			      struct btrfs_file_extent_item);
+	/* are we inside the extent that was found? */
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+	found_type = btrfs_key_type(&found_key);
+	if (found_key.objectid != objectid ||
+	    found_type != BTRFS_EXTENT_DATA_KEY) {
+		goto not_found;
+	}
+
+	found_type = btrfs_file_extent_type(leaf, item);
+	extent_start = found_key.offset;
+	compressed = btrfs_file_extent_compression(leaf, item);
+	if (found_type == BTRFS_FILE_EXTENT_REG ||
+	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
+		extent_end = extent_start +
+		       btrfs_file_extent_num_bytes(leaf, item);
+	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+		size_t size;
+		size = btrfs_file_extent_inline_len(leaf, item);
+		extent_end = (extent_start + size + root->sectorsize - 1) &
+			~((u64)root->sectorsize - 1);
+	}
+
+	if (start >= extent_end) {
+		path->slots[0]++;
+		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0) {
+				err = ret;
+				goto out;
+			}
+			if (ret > 0)
+				goto not_found;
+			leaf = path->nodes[0];
+		}
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.objectid != objectid ||
+		    found_key.type != BTRFS_EXTENT_DATA_KEY)
+			goto not_found;
+		if (start + len <= found_key.offset)
+			goto not_found;
+		em->start = start;
+		em->len = found_key.offset - start;
+		goto not_found_em;
+	}
+
+	if (found_type == BTRFS_FILE_EXTENT_REG ||
+	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
+		em->start = extent_start;
+		em->len = extent_end - extent_start;
+		em->orig_start = extent_start -
+				 btrfs_file_extent_offset(leaf, item);
+		bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
+		if (bytenr == 0) {
+			em->block_start = EXTENT_MAP_HOLE;
+			goto insert;
+		}
+		if (compressed) {
+			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
+			em->block_start = bytenr;
+			em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
+									 item);
+		} else {
+			bytenr += btrfs_file_extent_offset(leaf, item);
+			em->block_start = bytenr;
+			em->block_len = em->len;
+			if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
+				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
+		}
+		goto insert;
+	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+		unsigned long ptr;
+		char *map;
+		size_t size;
+		size_t extent_offset;
+		size_t copy_size;
+
+		em->block_start = EXTENT_MAP_INLINE;
+		if (!page || create) {
+			em->start = extent_start;
+			em->len = extent_end - extent_start;
+			goto out;
+		}
+
+		size = btrfs_file_extent_inline_len(leaf, item);
+		extent_offset = page_offset(page) + pg_offset - extent_start;
+		copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
+				size - extent_offset);
+		em->start = extent_start + extent_offset;
+		em->len = (copy_size + root->sectorsize - 1) &
+			~((u64)root->sectorsize - 1);
+		em->orig_start = EXTENT_MAP_INLINE;
+		if (compressed)
+			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
+		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
+		if (create == 0 && !PageUptodate(page)) {
+			if (btrfs_file_extent_compression(leaf, item) ==
+			    BTRFS_COMPRESS_ZLIB) {
+				ret = uncompress_inline(path, inode, page,
+							pg_offset,
+							extent_offset, item);
+				BUG_ON(ret);
+			} else {
+				map = kmap(page);
+				read_extent_buffer(leaf, map + pg_offset, ptr,
+						   copy_size);
+				kunmap(page);
+			}
+			flush_dcache_page(page);
+		} else if (create && PageUptodate(page)) {
+			if (!trans) {
+				kunmap(page);
+				free_extent_map(em);
+				em = NULL;
+				btrfs_release_path(root, path);
+				trans = btrfs_join_transaction(root, 1);
+				goto again;
+			}
+			map = kmap(page);
+			write_extent_buffer(leaf, map + pg_offset, ptr,
+					    copy_size);
+			kunmap(page);
+			btrfs_mark_buffer_dirty(leaf);
+		}
+		set_extent_uptodate(io_tree, em->start,
+				    extent_map_end(em) - 1, GFP_NOFS);
+		goto insert;
+	} else {
+		printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
+		WARN_ON(1);
+	}
+not_found:
+	em->start = start;
+	em->len = len;
+not_found_em:
+	em->block_start = EXTENT_MAP_HOLE;
+	set_bit(EXTENT_FLAG_VACANCY, &em->flags);
+insert:
+	btrfs_release_path(root, path);
+	if (em->start > start || extent_map_end(em) <= start) {
+		printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
+		       "[%llu %llu]\n", (unsigned long long)em->start,
+		       (unsigned long long)em->len,
+		       (unsigned long long)start,
+		       (unsigned long long)len);
+		err = -EIO;
+		goto out;
+	}
+
+	err = 0;
+	spin_lock(&em_tree->lock);
+	ret = add_extent_mapping(em_tree, em);
+	/* it is possible that someone inserted the extent into the tree
+	 * while we had the lock dropped.  It is also possible that
+	 * an overlapping map exists in the tree
+	 */
+	if (ret == -EEXIST) {
+		struct extent_map *existing;
+
+		ret = 0;
+
+		existing = lookup_extent_mapping(em_tree, start, len);
+		if (existing && (existing->start > start ||
+		    existing->start + existing->len <= start)) {
+			free_extent_map(existing);
+			existing = NULL;
+		}
+		if (!existing) {
+			existing = lookup_extent_mapping(em_tree, em->start,
+							 em->len);
+			if (existing) {
+				err = merge_extent_mapping(em_tree, existing,
+							   em, start,
+							   root->sectorsize);
+				free_extent_map(existing);
+				if (err) {
+					free_extent_map(em);
+					em = NULL;
+				}
+			} else {
+				err = -EIO;
+				free_extent_map(em);
+				em = NULL;
+			}
+		} else {
+			free_extent_map(em);
+			em = existing;
+			err = 0;
+		}
+	}
+	spin_unlock(&em_tree->lock);
+out:
+	if (path)
+		btrfs_free_path(path);
+	if (trans) {
+		ret = btrfs_end_transaction(trans, root);
+		if (!err)
+			err = ret;
+	}
+	if (err) {
+		free_extent_map(em);
+		WARN_ON(1);
+		return ERR_PTR(err);
+	}
+	return em;
+}
+
 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
 			const struct iovec *iov, loff_t offset,
 			unsigned long nr_segs)
@@ -2859,6 +4805,18 @@
 	.fsync		= btrfs_sync_file,
 };
 
+static struct extent_io_ops btrfs_extent_io_ops = {
+	.fill_delalloc = run_delalloc_range,
+	.submit_bio_hook = btrfs_submit_bio_hook,
+	.merge_bio_hook = btrfs_merge_bio_hook,
+	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
+	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
+	.writepage_start_hook = btrfs_writepage_start_hook,
+	.readpage_io_failed_hook = btrfs_io_failed_hook,
+	.set_bit_hook = btrfs_set_bit_hook,
+	.clear_bit_hook = btrfs_clear_bit_hook,
+};
+
 static struct address_space_operations btrfs_aops = {
 	.readpage	= btrfs_readpage,
 	.writepage	= btrfs_writepage,

[PATCH 23.1/35] Btrfs extent allocation code

[Chris Mason]

All metadata and data extents are allocated here.  A single btree holds
records for all of the extents allocated on the FS.  It also records
backreferences from those extents to the objects that have references on
them.

Btree blocks for the extent allocation tree are allocated from the
extent allocation tree.  This leads to some ugly recursion, which is
avoided by building up a list of pending allocations/deletions for the
extent allocation tree and processing them at safe times.

--- /dev/null	2008-06-10 15:56:50.000000000 -0400
+++ a/fs/btrfs/extent-tree.c	2009-01-08 09:56:31.803215082 -0500
@@ -0,0 +1,3336 @@
+/*
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include <linux/writeback.h>
+#include <linux/blkdev.h>
+#include <linux/version.h>
+#include "compat.h"
+#include "hash.h"
+#include "crc32c.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+#include "transaction.h"
+#include "volumes.h"
+#include "locking.h"
+#include "ref-cache.h"
+#include "compat.h"
+
+#define PENDING_EXTENT_INSERT 0
+#define PENDING_EXTENT_DELETE 1
+#define PENDING_BACKREF_UPDATE 2
+
+struct pending_extent_op {
+	int type;
+	u64 bytenr;
+	u64 num_bytes;
+	u64 parent;
+	u64 orig_parent;
+	u64 generation;
+	u64 orig_generation;
+	int level;
+	struct list_head list;
+	int del;
+};
+
+static int finish_current_insert(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *extent_root, int all);
+static int del_pending_extents(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *extent_root, int all);
+static int pin_down_bytes(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  u64 bytenr, u64 num_bytes, int is_data);
+static int update_block_group(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      u64 bytenr, u64 num_bytes, int alloc,
+			      int mark_free);
+
+static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
+{
+	return (cache->flags & bits) == bits;
+}
+
+/*
+ * this adds the block group to the fs_info rb tree for the block group
+ * cache
+ */
+static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
+				struct btrfs_block_group_cache *block_group)
+{
+	struct rb_node **p;
+	struct rb_node *parent = NULL;
+	struct btrfs_block_group_cache *cache;
+
+	spin_lock(&info->block_group_cache_lock);
+	p = &info->block_group_cache_tree.rb_node;
+
+	while (*p) {
+		parent = *p;
+		cache = rb_entry(parent, struct btrfs_block_group_cache,
+				 cache_node);
+		if (block_group->key.objectid < cache->key.objectid) {
+			p = &(*p)->rb_left;
+		} else if (block_group->key.objectid > cache->key.objectid) {
+			p = &(*p)->rb_right;
+		} else {
+			spin_unlock(&info->block_group_cache_lock);
+			return -EEXIST;
+		}
+	}
+
+	rb_link_node(&block_group->cache_node, parent, p);
+	rb_insert_color(&block_group->cache_node,
+			&info->block_group_cache_tree);
+	spin_unlock(&info->block_group_cache_lock);
+
+	return 0;
+}
+
+/*
+ * This will return the block group at or after bytenr if contains is 0, else
+ * it will return the block group that contains the bytenr
+ */
+static struct btrfs_block_group_cache *
+block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
+			      int contains)
+{
+	struct btrfs_block_group_cache *cache, *ret = NULL;
+	struct rb_node *n;
+	u64 end, start;
+
+	spin_lock(&info->block_group_cache_lock);
+	n = info->block_group_cache_tree.rb_node;
+
+	while (n) {
+		cache = rb_entry(n, struct btrfs_block_group_cache,
+				 cache_node);
+		end = cache->key.objectid + cache->key.offset - 1;
+		start = cache->key.objectid;
+
+		if (bytenr < start) {
+			if (!contains && (!ret || start < ret->key.objectid))
+				ret = cache;
+			n = n->rb_left;
+		} else if (bytenr > start) {
+			if (contains && bytenr <= end) {
+				ret = cache;
+				break;
+			}
+			n = n->rb_right;
+		} else {
+			ret = cache;
+			break;
+		}
+	}
+	if (ret)
+		atomic_inc(&ret->count);
+	spin_unlock(&info->block_group_cache_lock);
+
+	return ret;
+}
+
+/*
+ * this is only called by cache_block_group, since we could have freed extents
+ * we need to check the pinned_extents for any extents that can't be used yet
+ * since their free space will be released as soon as the transaction commits.
+ */
+static int add_new_free_space(struct btrfs_block_group_cache *block_group,
+			      struct btrfs_fs_info *info, u64 start, u64 end)
+{
+	u64 extent_start, extent_end, size;
+	int ret;
+
+	mutex_lock(&info->pinned_mutex);
+	while (start < end) {
+		ret = find_first_extent_bit(&info->pinned_extents, start,
+					    &extent_start, &extent_end,
+					    EXTENT_DIRTY);
+		if (ret)
+			break;
+
+		if (extent_start == start) {
+			start = extent_end + 1;
+		} else if (extent_start > start && extent_start < end) {
+			size = extent_start - start;
+			ret = btrfs_add_free_space(block_group, start,
+						   size);
+			BUG_ON(ret);
+			start = extent_end + 1;
+		} else {
+			break;
+		}
+	}
+
+	if (start < end) {
+		size = end - start;
+		ret = btrfs_add_free_space(block_group, start, size);
+		BUG_ON(ret);
+	}
+	mutex_unlock(&info->pinned_mutex);
+
+	return 0;
+}
+
+static int remove_sb_from_cache(struct btrfs_root *root,
+				struct btrfs_block_group_cache *cache)
+{
+	u64 bytenr;
+	u64 *logical;
+	int stripe_len;
+	int i, nr, ret;
+
+	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
+		bytenr = btrfs_sb_offset(i);
+		ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
+				       cache->key.objectid, bytenr, 0,
+				       &logical, &nr, &stripe_len);
+		BUG_ON(ret);
+		while (nr--) {
+			btrfs_remove_free_space(cache, logical[nr],
+						stripe_len);
+		}
+		kfree(logical);
+	}
+	return 0;
+}
+
+static int cache_block_group(struct btrfs_root *root,
+			     struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_path *path;
+	int ret = 0;
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	int slot;
+	u64 last;
+
+	if (!block_group)
+		return 0;
+
+	root = root->fs_info->extent_root;
+
+	if (block_group->cached)
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->reada = 2;
+	/*
+	 * we get into deadlocks with paths held by callers of this function.
+	 * since the alloc_mutex is protecting things right now, just
+	 * skip the locking here
+	 */
+	path->skip_locking = 1;
+	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
+	key.objectid = last;
+	key.offset = 0;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto err;
+
+	while (1) {
+		leaf = path->nodes[0];
+		slot = path->slots[0];
+		if (slot >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto err;
+			if (ret == 0)
+				continue;
+			else
+				break;
+		}
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+		if (key.objectid < block_group->key.objectid)
+			goto next;
+
+		if (key.objectid >= block_group->key.objectid +
+		    block_group->key.offset)
+			break;
+
+		if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
+			add_new_free_space(block_group, root->fs_info, last,
+					   key.objectid);
+
+			last = key.objectid + key.offset;
+		}
+next:
+		path->slots[0]++;
+	}
+
+	add_new_free_space(block_group, root->fs_info, last,
+			   block_group->key.objectid +
+			   block_group->key.offset);
+
+	remove_sb_from_cache(root, block_group);
+	block_group->cached = 1;
+	ret = 0;
+err:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * return the block group that starts at or after bytenr
+ */
+static struct btrfs_block_group_cache *
+btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
+{
+	struct btrfs_block_group_cache *cache;
+
+	cache = block_group_cache_tree_search(info, bytenr, 0);
+
+	return cache;
+}
+
+/*
+ * return the block group that contains teh given bytenr
+ */
+struct btrfs_block_group_cache *btrfs_lookup_block_group(
+						 struct btrfs_fs_info *info,
+						 u64 bytenr)
+{
+	struct btrfs_block_group_cache *cache;
+
+	cache = block_group_cache_tree_search(info, bytenr, 1);
+
+	return cache;
+}
+
+static inline void put_block_group(struct btrfs_block_group_cache *cache)
+{
+	if (atomic_dec_and_test(&cache->count))
+		kfree(cache);
+}
+
+static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
+						  u64 flags)
+{
+	struct list_head *head = &info->space_info;
+	struct list_head *cur;
+	struct btrfs_space_info *found;
+	list_for_each(cur, head) {
+		found = list_entry(cur, struct btrfs_space_info, list);
+		if (found->flags == flags)
+			return found;
+	}
+	return NULL;
+}
+
+static u64 div_factor(u64 num, int factor)
+{
+	if (factor == 10)
+		return num;
+	num *= factor;
+	do_div(num, 10);
+	return num;
+}
+
+u64 btrfs_find_block_group(struct btrfs_root *root,
+			   u64 search_start, u64 search_hint, int owner)
+{
+	struct btrfs_block_group_cache *cache;
+	u64 used;
+	u64 last = max(search_hint, search_start);
+	u64 group_start = 0;
+	int full_search = 0;
+	int factor = 9;
+	int wrapped = 0;
+again:
+	while (1) {
+		cache = btrfs_lookup_first_block_group(root->fs_info, last);
+		if (!cache)
+			break;
+
+		spin_lock(&cache->lock);
+		last = cache->key.objectid + cache->key.offset;
+		used = btrfs_block_group_used(&cache->item);
+
+		if ((full_search || !cache->ro) &&
+		    block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
+			if (used + cache->pinned + cache->reserved <
+			    div_factor(cache->key.offset, factor)) {
+				group_start = cache->key.objectid;
+				spin_unlock(&cache->lock);
+				put_block_group(cache);
+				goto found;
+			}
+		}
+		spin_unlock(&cache->lock);
+		put_block_group(cache);
+		cond_resched();
+	}
+	if (!wrapped) {
+		last = search_start;
+		wrapped = 1;
+		goto again;
+	}
+	if (!full_search && factor < 10) {
+		last = search_start;
+		full_search = 1;
+		factor = 10;
+		goto again;
+	}
+found:
+	return group_start;
+}
+
+/* simple helper to search for an existing extent at a given offset */
+int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
+{
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_path *path;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	key.objectid = start;
+	key.offset = len;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
+				0, 0);
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * Back reference rules.  Back refs have three main goals:
+ *
+ * 1) differentiate between all holders of references to an extent so that
+ *    when a reference is dropped we can make sure it was a valid reference
+ *    before freeing the extent.
+ *
+ * 2) Provide enough information to quickly find the holders of an extent
+ *    if we notice a given block is corrupted or bad.
+ *
+ * 3) Make it easy to migrate blocks for FS shrinking or storage pool
+ *    maintenance.  This is actually the same as #2, but with a slightly
+ *    different use case.
+ *
+ * File extents can be referenced by:
+ *
+ * - multiple snapshots, subvolumes, or different generations in one subvol
+ * - different files inside a single subvolume
+ * - different offsets inside a file (bookend extents in file.c)
+ *
+ * The extent ref structure has fields for:
+ *
+ * - Objectid of the subvolume root
+ * - Generation number of the tree holding the reference
+ * - objectid of the file holding the reference
+ * - number of references holding by parent node (alway 1 for tree blocks)
+ *
+ * Btree leaf may hold multiple references to a file extent. In most cases,
+ * these references are from same file and the corresponding offsets inside
+ * the file are close together.
+ *
+ * When a file extent is allocated the fields are filled in:
+ *     (root_key.objectid, trans->transid, inode objectid, 1)
+ *
+ * When a leaf is cow'd new references are added for every file extent found
+ * in the leaf.  It looks similar to the create case, but trans->transid will
+ * be different when the block is cow'd.
+ *
+ *     (root_key.objectid, trans->transid, inode objectid,
+ *      number of references in the leaf)
+ *
+ * When a file extent is removed either during snapshot deletion or
+ * file truncation, we find the corresponding back reference and check
+ * the following fields:
+ *
+ *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
+ *      inode objectid)
+ *
+ * Btree extents can be referenced by:
+ *
+ * - Different subvolumes
+ * - Different generations of the same subvolume
+ *
+ * When a tree block is created, back references are inserted:
+ *
+ * (root->root_key.objectid, trans->transid, level, 1)
+ *
+ * When a tree block is cow'd, new back references are added for all the
+ * blocks it points to. If the tree block isn't in reference counted root,
+ * the old back references are removed. These new back references are of
+ * the form (trans->transid will have increased since creation):
+ *
+ * (root->root_key.objectid, trans->transid, level, 1)
+ *
+ * When a backref is in deleting, the following fields are checked:
+ *
+ * if backref was for a tree root:
+ *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
+ * else
+ *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
+ *
+ * Back Reference Key composing:
+ *
+ * The key objectid corresponds to the first byte in the extent, the key
+ * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
+ * byte of parent extent. If a extent is tree root, the key offset is set
+ * to the key objectid.
+ */
+
+static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, u64 parent,
+					  u64 ref_root, u64 ref_generation,
+					  u64 owner_objectid, int del)
+{
+	struct btrfs_key key;
+	struct btrfs_extent_ref *ref;
+	struct extent_buffer *leaf;
+	u64 ref_objectid;
+	int ret;
+
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	key.offset = parent;
+
+	ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+	ref_objectid = btrfs_ref_objectid(leaf, ref);
+	if (btrfs_ref_root(leaf, ref) != ref_root ||
+	    btrfs_ref_generation(leaf, ref) != ref_generation ||
+	    (ref_objectid != owner_objectid &&
+	     ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
+		ret = -EIO;
+		WARN_ON(1);
+		goto out;
+	}
+	ret = 0;
+out:
+	return ret;
+}
+
+/*
+ * updates all the backrefs that are pending on update_list for the
+ * extent_root
+ */
+static noinline int update_backrefs(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *extent_root,
+				    struct btrfs_path *path,
+				    struct list_head *update_list)
+{
+	struct btrfs_key key;
+	struct btrfs_extent_ref *ref;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct pending_extent_op *op;
+	struct extent_buffer *leaf;
+	int ret = 0;
+	struct list_head *cur = update_list->next;
+	u64 ref_objectid;
+	u64 ref_root = extent_root->root_key.objectid;
+
+	op = list_entry(cur, struct pending_extent_op, list);
+
+search:
+	key.objectid = op->bytenr;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	key.offset = op->orig_parent;
+
+	ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+
+loop:
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+
+	ref_objectid = btrfs_ref_objectid(leaf, ref);
+
+	if (btrfs_ref_root(leaf, ref) != ref_root ||
+	    btrfs_ref_generation(leaf, ref) != op->orig_generation ||
+	    (ref_objectid != op->level &&
+	     ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
+		printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
+		       "root %llu, owner %u\n",
+		       (unsigned long long)op->bytenr,
+		       (unsigned long long)op->orig_parent,
+		       (unsigned long long)ref_root, op->level);
+		btrfs_print_leaf(extent_root, leaf);
+		BUG();
+	}
+
+	key.objectid = op->bytenr;
+	key.offset = op->parent;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
+	BUG_ON(ret);
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+	btrfs_set_ref_generation(leaf, ref, op->generation);
+
+	cur = cur->next;
+
+	list_del_init(&op->list);
+	unlock_extent(&info->extent_ins, op->bytenr,
+		      op->bytenr + op->num_bytes - 1, GFP_NOFS);
+	kfree(op);
+
+	if (cur == update_list) {
+		btrfs_mark_buffer_dirty(path->nodes[0]);
+		btrfs_release_path(extent_root, path);
+		goto out;
+	}
+
+	op = list_entry(cur, struct pending_extent_op, list);
+
+	path->slots[0]++;
+	while (path->slots[0] < btrfs_header_nritems(leaf)) {
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid == op->bytenr &&
+		    key.type == BTRFS_EXTENT_REF_KEY)
+			goto loop;
+		path->slots[0]++;
+	}
+
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_release_path(extent_root, path);
+	goto search;
+
+out:
+	return 0;
+}
+
+static noinline int insert_extents(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *extent_root,
+				   struct btrfs_path *path,
+				   struct list_head *insert_list, int nr)
+{
+	struct btrfs_key *keys;
+	u32 *data_size;
+	struct pending_extent_op *op;
+	struct extent_buffer *leaf;
+	struct list_head *cur = insert_list->next;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	u64 ref_root = extent_root->root_key.objectid;
+	int i = 0, last = 0, ret;
+	int total = nr * 2;
+
+	if (!nr)
+		return 0;
+
+	keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
+	if (!keys)
+		return -ENOMEM;
+
+	data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
+	if (!data_size) {
+		kfree(keys);
+		return -ENOMEM;
+	}
+
+	list_for_each_entry(op, insert_list, list) {
+		keys[i].objectid = op->bytenr;
+		keys[i].offset = op->num_bytes;
+		keys[i].type = BTRFS_EXTENT_ITEM_KEY;
+		data_size[i] = sizeof(struct btrfs_extent_item);
+		i++;
+
+		keys[i].objectid = op->bytenr;
+		keys[i].offset = op->parent;
+		keys[i].type = BTRFS_EXTENT_REF_KEY;
+		data_size[i] = sizeof(struct btrfs_extent_ref);
+		i++;
+	}
+
+	op = list_entry(cur, struct pending_extent_op, list);
+	i = 0;
+	while (i < total) {
+		int c;
+		ret = btrfs_insert_some_items(trans, extent_root, path,
+					      keys+i, data_size+i, total-i);
+		BUG_ON(ret < 0);
+
+		if (last && ret > 1)
+			BUG();
+
+		leaf = path->nodes[0];
+		for (c = 0; c < ret; c++) {
+			int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
+
+			/*
+			 * if the first item we inserted was a backref, then
+			 * the EXTENT_ITEM will be the odd c's, else it will
+			 * be the even c's
+			 */
+			if ((ref_first && (c % 2)) ||
+			    (!ref_first && !(c % 2))) {
+				struct btrfs_extent_item *itm;
+
+				itm = btrfs_item_ptr(leaf, path->slots[0] + c,
+						     struct btrfs_extent_item);
+				btrfs_set_extent_refs(path->nodes[0], itm, 1);
+				op->del++;
+			} else {
+				struct btrfs_extent_ref *ref;
+
+				ref = btrfs_item_ptr(leaf, path->slots[0] + c,
+						     struct btrfs_extent_ref);
+				btrfs_set_ref_root(leaf, ref, ref_root);
+				btrfs_set_ref_generation(leaf, ref,
+							 op->generation);
+				btrfs_set_ref_objectid(leaf, ref, op->level);
+				btrfs_set_ref_num_refs(leaf, ref, 1);
+				op->del++;
+			}
+
+			/*
+			 * using del to see when its ok to free up the
+			 * pending_extent_op.  In the case where we insert the
+			 * last item on the list in order to help do batching
+			 * we need to not free the extent op until we actually
+			 * insert the extent_item
+			 */
+			if (op->del == 2) {
+				unlock_extent(&info->extent_ins, op->bytenr,
+					      op->bytenr + op->num_bytes - 1,
+					      GFP_NOFS);
+				cur = cur->next;
+				list_del_init(&op->list);
+				kfree(op);
+				if (cur != insert_list)
+					op = list_entry(cur,
+						struct pending_extent_op,
+						list);
+			}
+		}
+		btrfs_mark_buffer_dirty(leaf);
+		btrfs_release_path(extent_root, path);
+
+		/*
+		 * Ok backref's and items usually go right next to eachother,
+		 * but if we could only insert 1 item that means that we
+		 * inserted on the end of a leaf, and we have no idea what may
+		 * be on the next leaf so we just play it safe.  In order to
+		 * try and help this case we insert the last thing on our
+		 * insert list so hopefully it will end up being the last
+		 * thing on the leaf and everything else will be before it,
+		 * which will let us insert a whole bunch of items at the same
+		 * time.
+		 */
+		if (ret == 1 && !last && (i + ret < total)) {
+			/*
+			 * last: where we will pick up the next time around
+			 * i: our current key to insert, will be total - 1
+			 * cur: the current op we are screwing with
+			 * op: duh
+			 */
+			last = i + ret;
+			i = total - 1;
+			cur = insert_list->prev;
+			op = list_entry(cur, struct pending_extent_op, list);
+		} else if (last) {
+			/*
+			 * ok we successfully inserted the last item on the
+			 * list, lets reset everything
+			 *
+			 * i: our current key to insert, so where we left off
+			 *    last time
+			 * last: done with this
+			 * cur: the op we are messing with
+			 * op: duh
+			 * total: since we inserted the last key, we need to
+			 *        decrement total so we dont overflow
+			 */
+			i = last;
+			last = 0;
+			total--;
+			if (i < total) {
+				cur = insert_list->next;
+				op = list_entry(cur, struct pending_extent_op,
+						list);
+			}
+		} else {
+			i += ret;
+		}
+
+		cond_resched();
+	}
+	ret = 0;
+	kfree(keys);
+	kfree(data_size);
+	return ret;
+}
+
+static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, u64 parent,
+					  u64 ref_root, u64 ref_generation,
+					  u64 owner_objectid)
+{
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_ref *ref;
+	u32 num_refs;
+	int ret;
+
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	key.offset = parent;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
+	if (ret == 0) {
+		leaf = path->nodes[0];
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_ref);
+		btrfs_set_ref_root(leaf, ref, ref_root);
+		btrfs_set_ref_generation(leaf, ref, ref_generation);
+		btrfs_set_ref_objectid(leaf, ref, owner_objectid);
+		btrfs_set_ref_num_refs(leaf, ref, 1);
+	} else if (ret == -EEXIST) {
+		u64 existing_owner;
+		BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
+		leaf = path->nodes[0];
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_ref);
+		if (btrfs_ref_root(leaf, ref) != ref_root ||
+		    btrfs_ref_generation(leaf, ref) != ref_generation) {
+			ret = -EIO;
+			WARN_ON(1);
+			goto out;
+		}
+
+		num_refs = btrfs_ref_num_refs(leaf, ref);
+		BUG_ON(num_refs == 0);
+		btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
+
+		existing_owner = btrfs_ref_objectid(leaf, ref);
+		if (existing_owner != owner_objectid &&
+		    existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
+			btrfs_set_ref_objectid(leaf, ref,
+					BTRFS_MULTIPLE_OBJECTIDS);
+		}
+		ret = 0;
+	} else {
+		goto out;
+	}
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+out:
+	btrfs_release_path(root, path);
+	return ret;
+}
+
+static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_extent_ref *ref;
+	u32 num_refs;
+	int ret = 0;
+
+	leaf = path->nodes[0];
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+	num_refs = btrfs_ref_num_refs(leaf, ref);
+	BUG_ON(num_refs == 0);
+	num_refs -= 1;
+	if (num_refs == 0) {
+		ret = btrfs_del_item(trans, root, path);
+	} else {
+		btrfs_set_ref_num_refs(leaf, ref, num_refs);
+		btrfs_mark_buffer_dirty(leaf);
+	}
+	btrfs_release_path(root, path);
+	return ret;
+}
+
+#ifdef BIO_RW_DISCARD
+static void btrfs_issue_discard(struct block_device *bdev,
+				u64 start, u64 len)
+{
+	blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
+}
+#endif
+
+static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
+				u64 num_bytes)
+{
+#ifdef BIO_RW_DISCARD
+	int ret;
+	u64 map_length = num_bytes;
+	struct btrfs_multi_bio *multi = NULL;
+
+	/* Tell the block device(s) that the sectors can be discarded */
+	ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
+			      bytenr, &map_length, &multi, 0);
+	if (!ret) {
+		struct btrfs_bio_stripe *stripe = multi->stripes;
+		int i;
+
+		if (map_length > num_bytes)
+			map_length = num_bytes;
+
+		for (i = 0; i < multi->num_stripes; i++, stripe++) {
+			btrfs_issue_discard(stripe->dev->bdev,
+					    stripe->physical,
+					    map_length);
+		}
+		kfree(multi);
+	}
+
+	return ret;
+#else
+	return 0;
+#endif
+}
+
+static noinline int free_extents(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *extent_root,
+				 struct list_head *del_list)
+{
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct btrfs_path *path;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	struct list_head *cur;
+	struct pending_extent_op *op;
+	struct btrfs_extent_item *ei;
+	int ret, num_to_del, extent_slot = 0, found_extent = 0;
+	u32 refs;
+	u64 bytes_freed = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	path->reada = 1;
+
+search:
+	/* search for the backref for the current ref we want to delete */
+	cur = del_list->next;
+	op = list_entry(cur, struct pending_extent_op, list);
+	ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
+				    op->orig_parent,
+				    extent_root->root_key.objectid,
+				    op->orig_generation, op->level, 1);
+	if (ret) {
+		printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
+		       "root %llu gen %llu owner %u\n",
+		       (unsigned long long)op->bytenr,
+		       (unsigned long long)extent_root->root_key.objectid,
+		       (unsigned long long)op->orig_generation, op->level);
+		btrfs_print_leaf(extent_root, path->nodes[0]);
+		WARN_ON(1);
+		goto out;
+	}
+
+	extent_slot = path->slots[0];
+	num_to_del = 1;
+	found_extent = 0;
+
+	/*
+	 * if we aren't the first item on the leaf we can move back one and see
+	 * if our ref is right next to our extent item
+	 */
+	if (likely(extent_slot)) {
+		extent_slot--;
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+				      extent_slot);
+		if (found_key.objectid == op->bytenr &&
+		    found_key.type == BTRFS_EXTENT_ITEM_KEY &&
+		    found_key.offset == op->num_bytes) {
+			num_to_del++;
+			found_extent = 1;
+		}
+	}
+
+	/*
+	 * if we didn't find the extent we need to delete the backref and then
+	 * search for the extent item key so we can update its ref count
+	 */
+	if (!found_extent) {
+		key.objectid = op->bytenr;
+		key.type = BTRFS_EXTENT_ITEM_KEY;
+		key.offset = op->num_bytes;
+
+		ret = remove_extent_backref(trans, extent_root, path);
+		BUG_ON(ret);
+		btrfs_release_path(extent_root, path);
+		ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
+		BUG_ON(ret);
+		extent_slot = path->slots[0];
+	}
+
+	/* this is where we update the ref count for the extent */
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
+	refs = btrfs_extent_refs(leaf, ei);
+	BUG_ON(refs == 0);
+	refs--;
+	btrfs_set_extent_refs(leaf, ei, refs);
+
+	btrfs_mark_buffer_dirty(leaf);
+
+	/*
+	 * This extent needs deleting.  The reason cur_slot is extent_slot +
+	 * num_to_del is because extent_slot points to the slot where the extent
+	 * is, and if the backref was not right next to the extent we will be
+	 * deleting at least 1 item, and will want to start searching at the
+	 * slot directly next to extent_slot.  However if we did find the
+	 * backref next to the extent item them we will be deleting at least 2
+	 * items and will want to start searching directly after the ref slot
+	 */
+	if (!refs) {
+		struct list_head *pos, *n, *end;
+		int cur_slot = extent_slot+num_to_del;
+		u64 super_used;
+		u64 root_used;
+
+		path->slots[0] = extent_slot;
+		bytes_freed = op->num_bytes;
+
+		mutex_lock(&info->pinned_mutex);
+		ret = pin_down_bytes(trans, extent_root, op->bytenr,
+				     op->num_bytes, op->level >=
+				     BTRFS_FIRST_FREE_OBJECTID);
+		mutex_unlock(&info->pinned_mutex);
+		BUG_ON(ret < 0);
+		op->del = ret;
+
+		/*
+		 * we need to see if we can delete multiple things at once, so
+		 * start looping through the list of extents we are wanting to
+		 * delete and see if their extent/backref's are right next to
+		 * eachother and the extents only have 1 ref
+		 */
+		for (pos = cur->next; pos != del_list; pos = pos->next) {
+			struct pending_extent_op *tmp;
+
+			tmp = list_entry(pos, struct pending_extent_op, list);
+
+			/* we only want to delete extent+ref at this stage */
+			if (cur_slot >= btrfs_header_nritems(leaf) - 1)
+				break;
+
+			btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
+			if (found_key.objectid != tmp->bytenr ||
+			    found_key.type != BTRFS_EXTENT_ITEM_KEY ||
+			    found_key.offset != tmp->num_bytes)
+				break;
+
+			/* check to make sure this extent only has one ref */
+			ei = btrfs_item_ptr(leaf, cur_slot,
+					    struct btrfs_extent_item);
+			if (btrfs_extent_refs(leaf, ei) != 1)
+				break;
+
+			btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
+			if (found_key.objectid != tmp->bytenr ||
+			    found_key.type != BTRFS_EXTENT_REF_KEY ||
+			    found_key.offset != tmp->orig_parent)
+				break;
+
+			/*
+			 * the ref is right next to the extent, we can set the
+			 * ref count to 0 since we will delete them both now
+			 */
+			btrfs_set_extent_refs(leaf, ei, 0);
+
+			/* pin down the bytes for this extent */
+			mutex_lock(&info->pinned_mutex);
+			ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
+					     tmp->num_bytes, tmp->level >=
+					     BTRFS_FIRST_FREE_OBJECTID);
+			mutex_unlock(&info->pinned_mutex);
+			BUG_ON(ret < 0);
+
+			/*
+			 * use the del field to tell if we need to go ahead and
+			 * free up the extent when we delete the item or not.
+			 */
+			tmp->del = ret;
+			bytes_freed += tmp->num_bytes;
+
+			num_to_del += 2;
+			cur_slot += 2;
+		}
+		end = pos;
+
+		/* update the free space counters */
+		spin_lock(&info->delalloc_lock);
+		super_used = btrfs_super_bytes_used(&info->super_copy);
+		btrfs_set_super_bytes_used(&info->super_copy,
+					   super_used - bytes_freed);
+
+		root_used = btrfs_root_used(&extent_root->root_item);
+		btrfs_set_root_used(&extent_root->root_item,
+				    root_used - bytes_freed);
+		spin_unlock(&info->delalloc_lock);
+
+		/* delete the items */
+		ret = btrfs_del_items(trans, extent_root, path,
+				      path->slots[0], num_to_del);
+		BUG_ON(ret);
+
+		/*
+		 * loop through the extents we deleted and do the cleanup work
+		 * on them
+		 */
+		for (pos = cur, n = pos->next; pos != end;
+		     pos = n, n = pos->next) {
+			struct pending_extent_op *tmp;
+			tmp = list_entry(pos, struct pending_extent_op, list);
+
+			/*
+			 * remember tmp->del tells us wether or not we pinned
+			 * down the extent
+			 */
+			ret = update_block_group(trans, extent_root,
+						 tmp->bytenr, tmp->num_bytes, 0,
+						 tmp->del);
+			BUG_ON(ret);
+
+			list_del_init(&tmp->list);
+			unlock_extent(&info->extent_ins, tmp->bytenr,
+				      tmp->bytenr + tmp->num_bytes - 1,
+				      GFP_NOFS);
+			kfree(tmp);
+		}
+	} else if (refs && found_extent) {
+		/*
+		 * the ref and extent were right next to eachother, but the
+		 * extent still has a ref, so just free the backref and keep
+		 * going
+		 */
+		ret = remove_extent_backref(trans, extent_root, path);
+		BUG_ON(ret);
+
+		list_del_init(&op->list);
+		unlock_extent(&info->extent_ins, op->bytenr,
+			      op->bytenr + op->num_bytes - 1, GFP_NOFS);
+		kfree(op);
+	} else {
+		/*
+		 * the extent has multiple refs and the backref we were looking
+		 * for was not right next to it, so just unlock and go next,
+		 * we're good to go
+		 */
+		list_del_init(&op->list);
+		unlock_extent(&info->extent_ins, op->bytenr,
+			      op->bytenr + op->num_bytes - 1, GFP_NOFS);
+		kfree(op);
+	}
+
+	btrfs_release_path(extent_root, path);
+	if (!list_empty(del_list))
+		goto search;
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root, u64 bytenr,
+				     u64 orig_parent, u64 parent,
+				     u64 orig_root, u64 ref_root,
+				     u64 orig_generation, u64 ref_generation,
+				     u64 owner_objectid)
+{
+	int ret;
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	struct btrfs_path *path;
+
+	if (root == root->fs_info->extent_root) {
+		struct pending_extent_op *extent_op;
+		u64 num_bytes;
+
+		BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
+		num_bytes = btrfs_level_size(root, (int)owner_objectid);
+		mutex_lock(&root->fs_info->extent_ins_mutex);
+		if (test_range_bit(&root->fs_info->extent_ins, bytenr,
+				bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
+			u64 priv;
+			ret = get_state_private(&root->fs_info->extent_ins,
+						bytenr, &priv);
+			BUG_ON(ret);
+			extent_op = (struct pending_extent_op *)
+							(unsigned long)priv;
+			BUG_ON(extent_op->parent != orig_parent);
+			BUG_ON(extent_op->generation != orig_generation);
+
+			extent_op->parent = parent;
+			extent_op->generation = ref_generation;
+		} else {
+			extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+			BUG_ON(!extent_op);
+
+			extent_op->type = PENDING_BACKREF_UPDATE;
+			extent_op->bytenr = bytenr;
+			extent_op->num_bytes = num_bytes;
+			extent_op->parent = parent;
+			extent_op->orig_parent = orig_parent;
+			extent_op->generation = ref_generation;
+			extent_op->orig_generation = orig_generation;
+			extent_op->level = (int)owner_objectid;
+			INIT_LIST_HEAD(&extent_op->list);
+			extent_op->del = 0;
+
+			set_extent_bits(&root->fs_info->extent_ins,
+					bytenr, bytenr + num_bytes - 1,
+					EXTENT_WRITEBACK, GFP_NOFS);
+			set_state_private(&root->fs_info->extent_ins,
+					  bytenr, (unsigned long)extent_op);
+		}
+		mutex_unlock(&root->fs_info->extent_ins_mutex);
+		return 0;
+	}
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	ret = lookup_extent_backref(trans, extent_root, path,
+				    bytenr, orig_parent, orig_root,
+				    orig_generation, owner_objectid, 1);
+	if (ret)
+		goto out;
+	ret = remove_extent_backref(trans, extent_root, path);
+	if (ret)
+		goto out;
+	ret = insert_extent_backref(trans, extent_root, path, bytenr,
+				    parent, ref_root, ref_generation,
+				    owner_objectid);
+	BUG_ON(ret);
+	finish_current_insert(trans, extent_root, 0);
+	del_pending_extents(trans, extent_root, 0);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, u64 bytenr,
+			    u64 orig_parent, u64 parent,
+			    u64 ref_root, u64 ref_generation,
+			    u64 owner_objectid)
+{
+	int ret;
+	if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
+	    owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
+		return 0;
+	ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
+					parent, ref_root, ref_root,
+					ref_generation, ref_generation,
+					owner_objectid);
+	return ret;
+}
+
+static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root, u64 bytenr,
+				  u64 orig_parent, u64 parent,
+				  u64 orig_root, u64 ref_root,
+				  u64 orig_generation, u64 ref_generation,
+				  u64 owner_objectid)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_key key;
+	struct extent_buffer *l;
+	struct btrfs_extent_item *item;
+	u32 refs;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->reada = 1;
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
+				0, 1);
+	if (ret < 0)
+		return ret;
+	BUG_ON(ret == 0 || path->slots[0] == 0);
+
+	path->slots[0]--;
+	l = path->nodes[0];
+
+	btrfs_item_key_to_cpu(l, &key, path->slots[0]);
+	if (key.objectid != bytenr) {
+		btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
+		printk(KERN_ERR "btrfs wanted %llu found %llu\n",
+		       (unsigned long long)bytenr,
+		       (unsigned long long)key.objectid);
+		BUG();
+	}
+	BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
+
+	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
+	refs = btrfs_extent_refs(l, item);
+	btrfs_set_extent_refs(l, item, refs + 1);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+
+	btrfs_release_path(root->fs_info->extent_root, path);
+
+	path->reada = 1;
+	ret = insert_extent_backref(trans, root->fs_info->extent_root,
+				    path, bytenr, parent,
+				    ref_root, ref_generation,
+				    owner_objectid);
+	BUG_ON(ret);
+	finish_current_insert(trans, root->fs_info->extent_root, 0);
+	del_pending_extents(trans, root->fs_info->extent_root, 0);
+
+	btrfs_free_path(path);
+	return 0;
+}
+
+int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 u64 bytenr, u64 num_bytes, u64 parent,
+			 u64 ref_root, u64 ref_generation,
+			 u64 owner_objectid)
+{
+	int ret;
+	if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
+	    owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
+		return 0;
+	ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
+				     0, ref_root, 0, ref_generation,
+				     owner_objectid);
+	return ret;
+}
+
+int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root)
+{
+	finish_current_insert(trans, root->fs_info->extent_root, 1);
+	del_pending_extents(trans, root->fs_info->extent_root, 1);
+	return 0;
+}
+
+int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, u64 bytenr,
+			    u64 num_bytes, u32 *refs)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_key key;
+	struct extent_buffer *l;
+	struct btrfs_extent_item *item;
+
+	WARN_ON(num_bytes < root->sectorsize);
+	path = btrfs_alloc_path();
+	path->reada = 1;
+	key.objectid = bytenr;
+	key.offset = num_bytes;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
+				0, 0);
+	if (ret < 0)
+		goto out;
+	if (ret != 0) {
+		btrfs_print_leaf(root, path->nodes[0]);
+		printk(KERN_INFO "btrfs failed to find block number %llu\n",
+		       (unsigned long long)bytenr);
+		BUG();
+	}
+	l = path->nodes[0];
+	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
+	*refs = btrfs_extent_refs(l, item);
+out:
+	btrfs_free_path(path);
+	return 0;
+}
+
+int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, u64 objectid, u64 bytenr)
+{
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_ref *ref_item;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	u64 ref_root;
+	u64 last_snapshot;
+	u32 nritems;
+	int ret;
+
+	key.objectid = bytenr;
+	key.offset = (u64)-1;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+
+	path = btrfs_alloc_path();
+	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	BUG_ON(ret == 0);
+
+	ret = -ENOENT;
+	if (path->slots[0] == 0)
+		goto out;
+
+	path->slots[0]--;
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+	if (found_key.objectid != bytenr ||
+	    found_key.type != BTRFS_EXTENT_ITEM_KEY)
+		goto out;
+
+	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
+	while (1) {
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(extent_root, path);
+			if (ret < 0)
+				goto out;
+			if (ret == 0)
+				continue;
+			break;
+		}
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.objectid != bytenr)
+			break;
+
+		if (found_key.type != BTRFS_EXTENT_REF_KEY) {
+			path->slots[0]++;
+			continue;
+		}
+
+		ref_item = btrfs_item_ptr(leaf, path->slots[0],
+					  struct btrfs_extent_ref);
+		ref_root = btrfs_ref_root(leaf, ref_item);
+		if ((ref_root != root->root_key.objectid &&
+		     ref_root != BTRFS_TREE_LOG_OBJECTID) ||
+		     objectid != btrfs_ref_objectid(leaf, ref_item)) {
+			ret = 1;
+			goto out;
+		}
+		if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
+			ret = 1;
+			goto out;
+		}
+
+		path->slots[0]++;
+	}
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		    struct extent_buffer *buf, u32 nr_extents)
+{
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	u64 root_gen;
+	u32 nritems;
+	int i;
+	int level;
+	int ret = 0;
+	int shared = 0;
+
+	if (!root->ref_cows)
+		return 0;
+
+	if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
+		shared = 0;
+		root_gen = root->root_key.offset;
+	} else {
+		shared = 1;
+		root_gen = trans->transid - 1;
+	}
+
+	level = btrfs_header_level(buf);
+	nritems = btrfs_header_nritems(buf);
+
+	if (level == 0) {
+		struct btrfs_leaf_ref *ref;
+		struct btrfs_extent_info *info;
+
+		ref = btrfs_alloc_leaf_ref(root, nr_extents);
+		if (!ref) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		ref->root_gen = root_gen;
+		ref->bytenr = buf->start;
+		ref->owner = btrfs_header_owner(buf);
+		ref->generation = btrfs_header_generation(buf);
+		ref->nritems = nr_extents;
+		info = ref->extents;
+
+		for (i = 0; nr_extents > 0 && i < nritems; i++) {
+			u64 disk_bytenr;
+			btrfs_item_key_to_cpu(buf, &key, i);
+			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+				continue;
+			fi = btrfs_item_ptr(buf, i,
+					    struct btrfs_file_extent_item);
+			if (btrfs_file_extent_type(buf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE)
+				continue;
+			disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
+			if (disk_bytenr == 0)
+				continue;
+
+			info->bytenr = disk_bytenr;
+			info->num_bytes =
+				btrfs_file_extent_disk_num_bytes(buf, fi);
+			info->objectid = key.objectid;
+			info->offset = key.offset;
+			info++;
+		}
+
+		ret = btrfs_add_leaf_ref(root, ref, shared);
+		if (ret == -EEXIST && shared) {
+			struct btrfs_leaf_ref *old;
+			old = btrfs_lookup_leaf_ref(root, ref->bytenr);
+			BUG_ON(!old);
+			btrfs_remove_leaf_ref(root, old);
+			btrfs_free_leaf_ref(root, old);
+			ret = btrfs_add_leaf_ref(root, ref, shared);
+		}
+		WARN_ON(ret);
+		btrfs_free_leaf_ref(root, ref);
+	}
+out:
+	return ret;
+}
+
+int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		  struct extent_buffer *orig_buf, struct extent_buffer *buf,
+		  u32 *nr_extents)
+{
+	u64 bytenr;
+	u64 ref_root;
+	u64 orig_root;
+	u64 ref_generation;
+	u64 orig_generation;
+	u32 nritems;
+	u32 nr_file_extents = 0;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	int i;
+	int level;
+	int ret = 0;
+	int faili = 0;
+	int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
+			    u64, u64, u64, u64, u64, u64, u64, u64);
+
+	ref_root = btrfs_header_owner(buf);
+	ref_generation = btrfs_header_generation(buf);
+	orig_root = btrfs_header_owner(orig_buf);
+	orig_generation = btrfs_header_generation(orig_buf);
+
+	nritems = btrfs_header_nritems(buf);
+	level = btrfs_header_level(buf);
+
+	if (root->ref_cows) {
+		process_func = __btrfs_inc_extent_ref;
+	} else {
+		if (level == 0 &&
+		    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
+			goto out;
+		if (level != 0 &&
+		    root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
+			goto out;
+		process_func = __btrfs_update_extent_ref;
+	}
+
+	for (i = 0; i < nritems; i++) {
+		cond_resched();
+		if (level == 0) {
+			btrfs_item_key_to_cpu(buf, &key, i);
+			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+				continue;
+			fi = btrfs_item_ptr(buf, i,
+					    struct btrfs_file_extent_item);
+			if (btrfs_file_extent_type(buf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE)
+				continue;
+			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
+			if (bytenr == 0)
+				continue;
+
+			nr_file_extents++;
+
+			ret = process_func(trans, root, bytenr,
+					   orig_buf->start, buf->start,
+					   orig_root, ref_root,
+					   orig_generation, ref_generation,
+					   key.objectid);
+
+			if (ret) {
+				faili = i;
+				WARN_ON(1);
+				goto fail;
+			}
+		} else {
+			bytenr = btrfs_node_blockptr(buf, i);
+			ret = process_func(trans, root, bytenr,
+					   orig_buf->start, buf->start,
+					   orig_root, ref_root,
+					   orig_generation, ref_generation,
+					   level - 1);
+			if (ret) {
+				faili = i;
+				WARN_ON(1);
+				goto fail;
+			}
+		}
+	}
+out:
+	if (nr_extents) {
+		if (level == 0)
+			*nr_extents = nr_file_extents;
+		else
+			*nr_extents = nritems;
+	}
+	return 0;
+fail:
+	WARN_ON(1);
+	return ret;
+}
+
+int btrfs_update_ref(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root, struct extent_buffer *orig_buf,
+		     struct extent_buffer *buf, int start_slot, int nr)
+
+{
+	u64 bytenr;
+	u64 ref_root;
+	u64 orig_root;
+	u64 ref_generation;
+	u64 orig_generation;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	int i;
+	int ret;
+	int slot;
+	int level;
+
+	BUG_ON(start_slot < 0);
+	BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
+
+	ref_root = btrfs_header_owner(buf);
+	ref_generation = btrfs_header_generation(buf);
+	orig_root = btrfs_header_owner(orig_buf);
+	orig_generation = btrfs_header_generation(orig_buf);
+	level = btrfs_header_level(buf);
+
+	if (!root->ref_cows) {
+		if (level == 0 &&
+		    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
+			return 0;
+		if (level != 0 &&
+		    root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
+			return 0;
+	}
+
+	for (i = 0, slot = start_slot; i < nr; i++, slot++) {
+		cond_resched();
+		if (level == 0) {
+			btrfs_item_key_to_cpu(buf, &key, slot);
+			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+				continue;
+			fi = btrfs_item_ptr(buf, slot,
+					    struct btrfs_file_extent_item);
+			if (btrfs_file_extent_type(buf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE)
+				continue;
+			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
+			if (bytenr == 0)
+				continue;
+			ret = __btrfs_update_extent_ref(trans, root, bytenr,
+					    orig_buf->start, buf->start,
+					    orig_root, ref_root,
+					    orig_generation, ref_generation,
+					    key.objectid);
+			if (ret)
+				goto fail;
+		} else {
+			bytenr = btrfs_node_blockptr(buf, slot);
+			ret = __btrfs_update_extent_ref(trans, root, bytenr,
+					    orig_buf->start, buf->start,
+					    orig_root, ref_root,
+					    orig_generation, ref_generation,
+					    level - 1);
+			if (ret)
+				goto fail;
+		}
+	}
+	return 0;
+fail:
+	WARN_ON(1);
+	return -1;
+}
+
+static int write_one_cache_group(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 struct btrfs_block_group_cache *cache)
+{
+	int ret;
+	int pending_ret;
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	unsigned long bi;
+	struct extent_buffer *leaf;
+
+	ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
+	if (ret < 0)
+		goto fail;
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(extent_root, path);
+fail:
+	finish_current_insert(trans, extent_root, 0);
+	pending_ret = del_pending_extents(trans, extent_root, 0);
+	if (ret)
+		return ret;
+	if (pending_ret)
+		return pending_ret;
+	return 0;
+
+}
+
+int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root)
+{
+	struct btrfs_block_group_cache *cache, *entry;
+	struct rb_node *n;
+	int err = 0;
+	int werr = 0;
+	struct btrfs_path *path;
+	u64 last = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	while (1) {
+		cache = NULL;
+		spin_lock(&root->fs_info->block_group_cache_lock);
+		for (n = rb_first(&root->fs_info->block_group_cache_tree);
+		     n; n = rb_next(n)) {
+			entry = rb_entry(n, struct btrfs_block_group_cache,
+					 cache_node);
+			if (entry->dirty) {
+				cache = entry;
+				break;
+			}
+		}
+		spin_unlock(&root->fs_info->block_group_cache_lock);
+
+		if (!cache)
+			break;
+
+		cache->dirty = 0;
+		last += cache->key.offset;
+
+		err = write_one_cache_group(trans, root,
+					    path, cache);
+		/*
+		 * if we fail to write the cache group, we want
+		 * to keep it marked dirty in hopes that a later
+		 * write will work
+		 */
+		if (err) {
+			werr = err;
+			continue;
+		}
+	}
+	btrfs_free_path(path);
+	return werr;
+}
+
+int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
+{
+	struct btrfs_block_group_cache *block_group;
+	int readonly = 0;
+
+	block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
+	if (!block_group || block_group->ro)
+		readonly = 1;
+	if (block_group)
+		put_block_group(block_group);
+	return readonly;
+}
+
+static int update_space_info(struct btrfs_fs_info *info, u64 flags,
+			     u64 total_bytes, u64 bytes_used,
+			     struct btrfs_space_info **space_info)
+{
+	struct btrfs_space_info *found;
+
+	found = __find_space_info(info, flags);
+	if (found) {
+		spin_lock(&found->lock);
+		found->total_bytes += total_bytes;
+		found->bytes_used += bytes_used;
+		found->full = 0;
+		spin_unlock(&found->lock);
+		*space_info = found;
+		return 0;
+	}
+	found = kzalloc(sizeof(*found), GFP_NOFS);
+	if (!found)
+		return -ENOMEM;
+
+	list_add(&found->list, &info->space_info);
+	INIT_LIST_HEAD(&found->block_groups);
+	init_rwsem(&found->groups_sem);
+	spin_lock_init(&found->lock);
+	found->flags = flags;
+	found->total_bytes = total_bytes;
+	found->bytes_used = bytes_used;
+	found->bytes_pinned = 0;
+	found->bytes_reserved = 0;
+	found->bytes_readonly = 0;
+	found->full = 0;
+	found->force_alloc = 0;
+	*space_info = found;
+	return 0;
+}
+
+static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
+{
+	u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
+				   BTRFS_BLOCK_GROUP_RAID1 |
+				   BTRFS_BLOCK_GROUP_RAID10 |
+				   BTRFS_BLOCK_GROUP_DUP);
+	if (extra_flags) {
+		if (flags & BTRFS_BLOCK_GROUP_DATA)
+			fs_info->avail_data_alloc_bits |= extra_flags;
+		if (flags & BTRFS_BLOCK_GROUP_METADATA)
+			fs_info->avail_metadata_alloc_bits |= extra_flags;
+		if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
+			fs_info->avail_system_alloc_bits |= extra_flags;
+	}
+}
+
+static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
+{
+	spin_lock(&cache->space_info->lock);
+	spin_lock(&cache->lock);
+	if (!cache->ro) {
+		cache->space_info->bytes_readonly += cache->key.offset -
+					btrfs_block_group_used(&cache->item);
+		cache->ro = 1;
+	}
+	spin_unlock(&cache->lock);
+	spin_unlock(&cache->space_info->lock);
+}
+
+u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
+{
+	u64 num_devices = root->fs_info->fs_devices->rw_devices;
+
+	if (num_devices == 1)
+		flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
+	if (num_devices < 4)
+		flags &= ~BTRFS_BLOCK_GROUP_RAID10;
+
+	if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
+	    (flags & (BTRFS_BLOCK_GROUP_RAID1 |
+		      BTRFS_BLOCK_GROUP_RAID10))) {
+		flags &= ~BTRFS_BLOCK_GROUP_DUP;
+	}
+
+	if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
+	    (flags & BTRFS_BLOCK_GROUP_RAID10)) {
+		flags &= ~BTRFS_BLOCK_GROUP_RAID1;
+	}
+
+	if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
+	    ((flags & BTRFS_BLOCK_GROUP_RAID1) |
+	     (flags & BTRFS_BLOCK_GROUP_RAID10) |
+	     (flags & BTRFS_BLOCK_GROUP_DUP)))
+		flags &= ~BTRFS_BLOCK_GROUP_RAID0;
+	return flags;
+}
+
+static int update_block_group(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      u64 bytenr, u64 num_bytes, int alloc,
+			      int mark_free)
+{
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *info = root->fs_info;
+	u64 total = num_bytes;
+	u64 old_val;
+	u64 byte_in_group;
+
+	while (total) {
+		cache = btrfs_lookup_block_group(info, bytenr);
+		if (!cache)
+			return -1;
+		byte_in_group = bytenr - cache->key.objectid;
+		WARN_ON(byte_in_group > cache->key.offset);
+
+		spin_lock(&cache->space_info->lock);
+		spin_lock(&cache->lock);
+		cache->dirty = 1;
+		old_val = btrfs_block_group_used(&cache->item);
+		num_bytes = min(total, cache->key.offset - byte_in_group);
+		if (alloc) {
+			old_val += num_bytes;
+			cache->space_info->bytes_used += num_bytes;
+			if (cache->ro)
+				cache->space_info->bytes_readonly -= num_bytes;
+			btrfs_set_block_group_used(&cache->item, old_val);
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+		} else {
+			old_val -= num_bytes;
+			cache->space_info->bytes_used -= num_bytes;
+			if (cache->ro)
+				cache->space_info->bytes_readonly += num_bytes;
+			btrfs_set_block_group_used(&cache->item, old_val);
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+			if (mark_free) {
+				int ret;
+
+				ret = btrfs_discard_extent(root, bytenr,
+							   num_bytes);
+				WARN_ON(ret);
+
+				ret = btrfs_add_free_space(cache, bytenr,
+							   num_bytes);
+				WARN_ON(ret);
+			}
+		}
+		put_block_group(cache);
+		total -= num_bytes;
+		bytenr += num_bytes;
+	}
+	return 0;
+}
+
+static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
+{
+	struct btrfs_block_group_cache *cache;
+	u64 bytenr;
+
+	cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
+	if (!cache)
+		return 0;
+
+	bytenr = cache->key.objectid;
+	put_block_group(cache);
+
+	return bytenr;
+}
+
+int btrfs_update_pinned_extents(struct btrfs_root *root,
+				u64 bytenr, u64 num, int pin)
+{
+	u64 len;
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+
+	WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
+	if (pin) {
+		set_extent_dirty(&fs_info->pinned_extents,
+				bytenr, bytenr + num - 1, GFP_NOFS);
+	} else {
+		clear_extent_dirty(&fs_info->pinned_extents,
+				bytenr, bytenr + num - 1, GFP_NOFS);
+	}
+	while (num > 0) {
+		cache = btrfs_lookup_block_group(fs_info, bytenr);
+		BUG_ON(!cache);
+		len = min(num, cache->key.offset -
+			  (bytenr - cache->key.objectid));
+		if (pin) {
+			spin_lock(&cache->space_info->lock);
+			spin_lock(&cache->lock);
+			cache->pinned += len;
+			cache->space_info->bytes_pinned += len;
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+			fs_info->total_pinned += len;
+		} else {
+			spin_lock(&cache->space_info->lock);
+			spin_lock(&cache->lock);
+			cache->pinned -= len;
+			cache->space_info->bytes_pinned -= len;
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+			fs_info->total_pinned -= len;
+			if (cache->cached)
+				btrfs_add_free_space(cache, bytenr, len);
+		}
+		put_block_group(cache);
+		bytenr += len;
+		num -= len;
+	}
+	return 0;
+}
+
+static int update_reserved_extents(struct btrfs_root *root,
+				   u64 bytenr, u64 num, int reserve)
+{
+	u64 len;
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+
+	while (num > 0) {
+		cache = btrfs_lookup_block_group(fs_info, bytenr);
+		BUG_ON(!cache);
+		len = min(num, cache->key.offset -
+			  (bytenr - cache->key.objectid));
+
+		spin_lock(&cache->space_info->lock);
+		spin_lock(&cache->lock);
+		if (reserve) {
+			cache->reserved += len;
+			cache->space_info->bytes_reserved += len;
+		} else {
+			cache->reserved -= len;
+			cache->space_info->bytes_reserved -= len;
+		}
+		spin_unlock(&cache->lock);
+		spin_unlock(&cache->space_info->lock);
+		put_block_group(cache);
+		bytenr += len;
+		num -= len;
+	}
+	return 0;
+}
+
+int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
+{
+	u64 last = 0;
+	u64 start;
+	u64 end;
+	struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
+	int ret;
+
+	mutex_lock(&root->fs_info->pinned_mutex);
+	while (1) {
+		ret = find_first_extent_bit(pinned_extents, last,
+					    &start, &end, EXTENT_DIRTY);
+		if (ret)
+			break;
+		set_extent_dirty(copy, start, end, GFP_NOFS);
+		last = end + 1;
+	}
+	mutex_unlock(&root->fs_info->pinned_mutex);
+	return 0;
+}
+
+int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct extent_io_tree *unpin)
+{
+	u64 start;
+	u64 end;
+	int ret;
+
+	mutex_lock(&root->fs_info->pinned_mutex);
+	while (1) {
+		ret = find_first_extent_bit(unpin, 0, &start, &end,
+					    EXTENT_DIRTY);
+		if (ret)
+			break;
+
+		ret = btrfs_discard_extent(root, start, end + 1 - start);
+
+		btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
+		clear_extent_dirty(unpin, start, end, GFP_NOFS);
+
+		if (need_resched()) {
+			mutex_unlock(&root->fs_info->pinned_mutex);
+			cond_resched();
+			mutex_lock(&root->fs_info->pinned_mutex);
+		}
+	}
+	mutex_unlock(&root->fs_info->pinned_mutex);
+	return ret;
+}
+
+static int finish_current_insert(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *extent_root, int all)
+{
+	u64 start;
+	u64 end;
+	u64 priv;
+	u64 search = 0;
+	u64 skipped = 0;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct btrfs_path *path;
+	struct pending_extent_op *extent_op, *tmp;
+	struct list_head insert_list, update_list;
+	int ret;
+	int num_inserts = 0, max_inserts;
+
+	path = btrfs_alloc_path();
+	INIT_LIST_HEAD(&insert_list);
+	INIT_LIST_HEAD(&update_list);
+
+	max_inserts = extent_root->leafsize /
+		(2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
+		 sizeof(struct btrfs_extent_ref) +
+		 sizeof(struct btrfs_extent_item));
+again:
+	mutex_lock(&info->extent_ins_mutex);
+	while (1) {
+		ret = find_first_extent_bit(&info->extent_ins, search, &start,
+					    &end, EXTENT_WRITEBACK);
+		if (ret) {
+			if (skipped && all && !num_inserts) {
+				skipped = 0;
+				search = 0;
+				continue;
+			}
+			mutex_unlock(&info->extent_ins_mutex);
+			break;
+		}
+
+		ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
+		if (!ret) {
+			skipped = 1;
+			search = end + 1;
+			if (need_resched()) {
+				mutex_unlock(&info->extent_ins_mutex);
+				cond_resched();
+				mutex_lock(&info->extent_ins_mutex);
+			}
+			continue;
+		}
+
+		ret = get_state_private(&info->extent_ins, start, &priv);
+		BUG_ON(ret);
+		extent_op = (struct pending_extent_op *)(unsigned long) priv;
+
+		if (extent_op->type == PENDING_EXTENT_INSERT) {
+			num_inserts++;
+			list_add_tail(&extent_op->list, &insert_list);
+			search = end + 1;
+			if (num_inserts == max_inserts) {
+				mutex_unlock(&info->extent_ins_mutex);
+				break;
+			}
+		} else if (extent_op->type == PENDING_BACKREF_UPDATE) {
+			list_add_tail(&extent_op->list, &update_list);
+			search = end + 1;
+		} else {
+			BUG();
+		}
+	}
+
+	/*
+	 * process the update list, clear the writeback bit for it, and if
+	 * somebody marked this thing for deletion then just unlock it and be
+	 * done, the free_extents will handle it
+	 */
+	mutex_lock(&info->extent_ins_mutex);
+	list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
+		clear_extent_bits(&info->extent_ins, extent_op->bytenr,
+				  extent_op->bytenr + extent_op->num_bytes - 1,
+				  EXTENT_WRITEBACK, GFP_NOFS);
+		if (extent_op->del) {
+			list_del_init(&extent_op->list);
+			unlock_extent(&info->extent_ins, extent_op->bytenr,
+				      extent_op->bytenr + extent_op->num_bytes
+				      - 1, GFP_NOFS);
+			kfree(extent_op);
+		}
+	}
+	mutex_unlock(&info->extent_ins_mutex);
+
+	/*
+	 * still have things left on the update list, go ahead an update
+	 * everything
+	 */
+	if (!list_empty(&update_list)) {
+		ret = update_backrefs(trans, extent_root, path, &update_list);
+		BUG_ON(ret);
+	}
+
+	/*
+	 * if no inserts need to be done, but we skipped some extents and we
+	 * need to make sure everything is cleaned then reset everything and
+	 * go back to the beginning
+	 */
+	if (!num_inserts && all && skipped) {
+		search = 0;
+		skipped = 0;
+		INIT_LIST_HEAD(&update_list);
+		INIT_LIST_HEAD(&insert_list);
+		goto again;
+	} else if (!num_inserts) {
+		goto out;
+	}
+
+	/*
+	 * process the insert extents list.  Again if we are deleting this
+	 * extent, then just unlock it, pin down the bytes if need be, and be
+	 * done with it.  Saves us from having to actually insert the extent
+	 * into the tree and then subsequently come along and delete it
+	 */
+	mutex_lock(&info->extent_ins_mutex);
+	list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
+		clear_extent_bits(&info->extent_ins, extent_op->bytenr,
+				  extent_op->bytenr + extent_op->num_bytes - 1,
+				  EXTENT_WRITEBACK, GFP_NOFS);
+		if (extent_op->del) {
+			u64 used;
+			list_del_init(&extent_op->list);
+			unlock_extent(&info->extent_ins, extent_op->bytenr,
+				      extent_op->bytenr + extent_op->num_bytes
+				      - 1, GFP_NOFS);
+
+			mutex_lock(&extent_root->fs_info->pinned_mutex);
+			ret = pin_down_bytes(trans, extent_root,
+					     extent_op->bytenr,
+					     extent_op->num_bytes, 0);
+			mutex_unlock(&extent_root->fs_info->pinned_mutex);
+
+			spin_lock(&info->delalloc_lock);
+			used = btrfs_super_bytes_used(&info->super_copy);
+			btrfs_set_super_bytes_used(&info->super_copy,
+					used - extent_op->num_bytes);
+			used = btrfs_root_used(&extent_root->root_item);
+			btrfs_set_root_used(&extent_root->root_item,
+					used - extent_op->num_bytes);
+			spin_unlock(&info->delalloc_lock);
+
+			ret = update_block_group(trans, extent_root,
+						 extent_op->bytenr,
+						 extent_op->num_bytes,
+						 0, ret > 0);
+			BUG_ON(ret);
+			kfree(extent_op);
+			num_inserts--;
+		}
+	}
+	mutex_unlock(&info->extent_ins_mutex);
+
+	ret = insert_extents(trans, extent_root, path, &insert_list,
+			     num_inserts);
+	BUG_ON(ret);
+
+	/*
+	 * if we broke out of the loop in order to insert stuff because we hit
+	 * the maximum number of inserts at a time we can handle, then loop
+	 * back and pick up where we left off
+	 */
+	if (num_inserts == max_inserts) {
+		INIT_LIST_HEAD(&insert_list);
+		INIT_LIST_HEAD(&update_list);
+		num_inserts = 0;
+		goto again;
+	}
+
+	/*
+	 * again, if we need to make absolutely sure there are no more pending
+	 * extent operations left and we know that we skipped some, go back to
+	 * the beginning and do it all again
+	 */
+	if (all && skipped) {
+		INIT_LIST_HEAD(&insert_list);
+		INIT_LIST_HEAD(&update_list);
+		search = 0;
+		skipped = 0;
+		num_inserts = 0;
+		goto again;
+	}
+out:
+	btrfs_free_path(path);
+	return 0;
+}
+
+static int pin_down_bytes(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  u64 bytenr, u64 num_bytes, int is_data)
+{
+	int err = 0;
+	struct extent_buffer *buf;
+
+	if (is_data)
+		goto pinit;
+
+	buf = btrfs_find_tree_block(root, bytenr, num_bytes);
+	if (!buf)
+		goto pinit;
+
+	/* we can reuse a block if it hasn't been written
+	 * and it is from this transaction.  We can't
+	 * reuse anything from the tree log root because
+	 * it has tiny sub-transactions.
+	 */
+	if (btrfs_buffer_uptodate(buf, 0) &&
+	    btrfs_try_tree_lock(buf)) {
+		u64 header_owner = btrfs_header_owner(buf);
+		u64 header_transid = btrfs_header_generation(buf);
+		if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
+		    header_owner != BTRFS_TREE_RELOC_OBJECTID &&
+		    header_transid == trans->transid &&
+		    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
+			clean_tree_block(NULL, root, buf);
+			btrfs_tree_unlock(buf);
+			free_extent_buffer(buf);
+			return 1;
+		}
+		btrfs_tree_unlock(buf);
+	}
+	free_extent_buffer(buf);
+pinit:
+	btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
+
+	BUG_ON(err < 0);
+	return 0;
+}
+
+/*
+ * remove an extent from the root, returns 0 on success
+ */
+static int __free_extent(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 u64 bytenr, u64 num_bytes, u64 parent,
+			 u64 root_objectid, u64 ref_generation,
+			 u64 owner_objectid, int pin, int mark_free)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_root *extent_root = info->extent_root;
+	struct extent_buffer *leaf;
+	int ret;
+	int extent_slot = 0;
+	int found_extent = 0;
+	int num_to_del = 1;
+	struct btrfs_extent_item *ei;
+	u32 refs;
+
+	key.objectid = bytenr;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	key.offset = num_bytes;
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->reada = 1;
+	ret = lookup_extent_backref(trans, extent_root, path,
+				    bytenr, parent, root_objectid,
+				    ref_generation, owner_objectid, 1);
+	if (ret == 0) {
+		struct btrfs_key found_key;
+		extent_slot = path->slots[0];
+		while (extent_slot > 0) {
+			extent_slot--;
+			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+					      extent_slot);
+			if (found_key.objectid != bytenr)
+				break;
+			if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
+			    found_key.offset == num_bytes) {
+				found_extent = 1;
+				break;
+			}
+			if (path->slots[0] - extent_slot > 5)
+				break;
+		}
+		if (!found_extent) {
+			ret = remove_extent_backref(trans, extent_root, path);
+			BUG_ON(ret);
+			btrfs_release_path(extent_root, path);
+			ret = btrfs_search_slot(trans, extent_root,
+						&key, path, -1, 1);
+			if (ret) {
+				printk(KERN_ERR "umm, got %d back from search"
+				       ", was looking for %llu\n", ret,
+				       (unsigned long long)bytenr);
+				btrfs_print_leaf(extent_root, path->nodes[0]);
+			}
+			BUG_ON(ret);
+			extent_slot = path->slots[0];
+		}
+	} else {
+		btrfs_print_leaf(extent_root, path->nodes[0]);
+		WARN_ON(1);
+		printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
+		       "root %llu gen %llu owner %llu\n",
+		       (unsigned long long)bytenr,
+		       (unsigned long long)root_objectid,
+		       (unsigned long long)ref_generation,
+		       (unsigned long long)owner_objectid);
+	}
+
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, extent_slot,
+			    struct btrfs_extent_item);
+	refs = btrfs_extent_refs(leaf, ei);
+	BUG_ON(refs == 0);
+	refs -= 1;
+	btrfs_set_extent_refs(leaf, ei, refs);
+
+	btrfs_mark_buffer_dirty(leaf);
+
+	if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
+		struct btrfs_extent_ref *ref;
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_ref);
+		BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
+		/* if the back ref and the extent are next to each other
+		 * they get deleted below in one shot
+		 */
+		path->slots[0] = extent_slot;
+		num_to_del = 2;
+	} else if (found_extent) {
+		/* otherwise delete the extent back ref */
+		ret = remove_extent_backref(trans, extent_root, path);
+		BUG_ON(ret);
+		/* if refs are 0, we need to setup the path for deletion */
+		if (refs == 0) {
+			btrfs_release_path(extent_root, path);
+			ret = btrfs_search_slot(trans, extent_root, &key, path,
+						-1, 1);
+			BUG_ON(ret);
+		}
+	}
+
+	if (refs == 0) {
+		u64 super_used;
+		u64 root_used;
+
+		if (pin) {
+			mutex_lock(&root->fs_info->pinned_mutex);
+			ret = pin_down_bytes(trans, root, bytenr, num_bytes,
+				owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
+			mutex_unlock(&root->fs_info->pinned_mutex);
+			if (ret > 0)
+				mark_free = 1;
+			BUG_ON(ret < 0);
+		}
+		/* block accounting for super block */
+		spin_lock(&info->delalloc_lock);
+		super_used = btrfs_super_bytes_used(&info->super_copy);
+		btrfs_set_super_bytes_used(&info->super_copy,
+					   super_used - num_bytes);
+
+		/* block accounting for root item */
+		root_used = btrfs_root_used(&root->root_item);
+		btrfs_set_root_used(&root->root_item,
+					   root_used - num_bytes);
+		spin_unlock(&info->delalloc_lock);
+		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
+				      num_to_del);
+		BUG_ON(ret);
+		btrfs_release_path(extent_root, path);
+
+		if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
+			BUG_ON(ret);
+		}
+
+		ret = update_block_group(trans, root, bytenr, num_bytes, 0,
+					 mark_free);
+		BUG_ON(ret);
+	}
+	btrfs_free_path(path);
+	finish_current_insert(trans, extent_root, 0);
+	return ret;
+}
+
+/*
+ * find all the blocks marked as pending in the radix tree and remove
+ * them from the extent map
+ */
+static int del_pending_extents(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *extent_root, int all)
+{
+	int ret;
+	int err = 0;
+	u64 start;
+	u64 end;
+	u64 priv;
+	u64 search = 0;
+	int nr = 0, skipped = 0;
+	struct extent_io_tree *pending_del;
+	struct extent_io_tree *extent_ins;
+	struct pending_extent_op *extent_op;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct list_head delete_list;
+
+	INIT_LIST_HEAD(&delete_list);
+	extent_ins = &extent_root->fs_info->extent_ins;
+	pending_del = &extent_root->fs_info->pending_del;
+
+again:
+	mutex_lock(&info->extent_ins_mutex);
+	while (1) {
+		ret = find_first_extent_bit(pending_del, search, &start, &end,
+					    EXTENT_WRITEBACK);
+		if (ret) {
+			if (all && skipped && !nr) {
+				search = 0;
+				continue;
+			}
+			mutex_unlock(&info->extent_ins_mutex);
+			break;
+		}
+
+		ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
+		if (!ret) {
+			search = end+1;
+			skipped = 1;
+
+			if (need_resched()) {
+				mutex_unlock(&info->extent_ins_mutex);
+				cond_resched();
+				mutex_lock(&info->extent_ins_mutex);
+			}
+
+			continue;
+		}
+		BUG_ON(ret < 0);
+
+		ret = get_state_private(pending_del, start, &priv);
+		BUG_ON(ret);
+		extent_op = (struct pending_extent_op *)(unsigned long)priv;
+
+		clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
+				  GFP_NOFS);
+		if (!test_range_bit(extent_ins, start, end,
+				    EXTENT_WRITEBACK, 0)) {
+			list_add_tail(&extent_op->list, &delete_list);
+			nr++;
+		} else {
+			kfree(extent_op);
+
+			ret = get_state_private(&info->extent_ins, start,
+						&priv);
+			BUG_ON(ret);
+			extent_op = (struct pending_extent_op *)
+						(unsigned long)priv;
+
+			clear_extent_bits(&info->extent_ins, start, end,
+					  EXTENT_WRITEBACK, GFP_NOFS);
+
+			if (extent_op->type == PENDING_BACKREF_UPDATE) {
+				list_add_tail(&extent_op->list, &delete_list);
+				search = end + 1;
+				nr++;
+				continue;
+			}
+
+			mutex_lock(&extent_root->fs_info->pinned_mutex);
+			ret = pin_down_bytes(trans, extent_root, start,
+					     end + 1 - start, 0);
+			mutex_unlock(&extent_root->fs_info->pinned_mutex);
+
+			ret = update_block_group(trans, extent_root, start,
+						end + 1 - start, 0, ret > 0);
+
+			unlock_extent(extent_ins, start, end, GFP_NOFS);
+			BUG_ON(ret);
+			kfree(extent_op);
+		}
+		if (ret)
+			err = ret;
+
+		search = end + 1;
+
+		if (need_resched()) {
+			mutex_unlock(&info->extent_ins_mutex);
+			cond_resched();
+			mutex_lock(&info->extent_ins_mutex);
+		}
+	}
+
+	if (nr) {
+		ret = free_extents(trans, extent_root, &delete_list);
+		BUG_ON(ret);
+	}
+
+	if (all && skipped) {
+		INIT_LIST_HEAD(&delete_list);
+		search = 0;
+		nr = 0;
+		goto again;
+	}
+
+	return err;
+}
+
+/*
+ * remove an extent from the root, returns 0 on success
+ */
+static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       u64 bytenr, u64 num_bytes, u64 parent,
+			       u64 root_objectid, u64 ref_generation,
+			       u64 owner_objectid, int pin)
+{
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	int pending_ret;
+	int ret;
+
+	WARN_ON(num_bytes < root->sectorsize);
+	if (root == extent_root) {
+		struct pending_extent_op *extent_op = NULL;
+
+		mutex_lock(&root->fs_info->extent_ins_mutex);
+		if (test_range_bit(&root->fs_info->extent_ins, bytenr,
+				bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
+			u64 priv;
+			ret = get_state_private(&root->fs_info->extent_ins,
+						bytenr, &priv);
+			BUG_ON(ret);
+			extent_op = (struct pending_extent_op *)
+						(unsigned long)priv;
+
+			extent_op->del = 1;
+			if (extent_op->type == PENDING_EXTENT_INSERT) {
+				mutex_unlock(&root->fs_info->extent_ins_mutex);
+				return 0;
+			}
+		}
+
+		if (extent_op) {
+			ref_generation = extent_op->orig_generation;
+			parent = extent_op->orig_parent;
+		}
+
+		extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+		BUG_ON(!extent_op);
+
+		extent_op->type = PENDING_EXTENT_DELETE;
+		extent_op->bytenr = bytenr;
+		extent_op->num_bytes = num_bytes;
+		extent_op->parent = parent;
+		extent_op->orig_parent = parent;
+		extent_op->generation = ref_generation;
+		extent_op->orig_generation = ref_generation;
+		extent_op->level = (int)owner_objectid;
+		INIT_LIST_HEAD(&extent_op->list);
+		extent_op->del = 0;
+
+		set_extent_bits(&root->fs_info->pending_del,
+				bytenr, bytenr + num_bytes - 1,
+				EXTENT_WRITEBACK, GFP_NOFS);
+		set_state_private(&root->fs_info->pending_del,
+				  bytenr, (unsigned long)extent_op);
+		mutex_unlock(&root->fs_info->extent_ins_mutex);
+		return 0;
+	}
+	/* if metadata always pin */
+	if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
+		if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
+			struct btrfs_block_group_cache *cache;
+
+			/* btrfs_free_reserved_extent */
+			cache = btrfs_lookup_block_group(root->fs_info, bytenr);
+			BUG_ON(!cache);
+			btrfs_add_free_space(cache, bytenr, num_bytes);
+			put_block_group(cache);
+			update_reserved_extents(root, bytenr, num_bytes, 0);
+			return 0;
+		}
+		pin = 1;
+	}
+
+	/* if data pin when any transaction has committed this */
+	if (ref_generation != trans->transid)
+		pin = 1;
+
+	ret = __free_extent(trans, root, bytenr, num_bytes, parent,
+			    root_objectid, ref_generation,
+			    owner_objectid, pin, pin == 0);
+
+	finish_current_insert(trans, root->fs_info->extent_root, 0);
+	pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
+	return ret ? ret : pending_ret;
+}
+
+int btrfs_free_extent(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root,
+		      u64 bytenr, u64 num_bytes, u64 parent,
+		      u64 root_objectid, u64 ref_generation,
+		      u64 owner_objectid, int pin)
+{
+	int ret;
+
+	ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
+				  root_objectid, ref_generation,
+				  owner_objectid, pin);
+	return ret;
+}
+
+static u64 stripe_align(struct btrfs_root *root, u64 val)
+{
+	u64 mask = ((u64)root->stripesize - 1);
+	u64 ret = (val + mask) & ~mask;
+	return ret;
+}
+
+/*
+ * walks the btree of allocated extents and find a hole of a given size.
+ * The key ins is changed to record the hole:
+ * ins->objectid == block start
+ * ins->flags = BTRFS_EXTENT_ITEM_KEY
+ * ins->offset == number of blocks
+ * Any available blocks before search_start are skipped.
+ */
+static noinline int find_free_extent(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *orig_root,
+				     u64 num_bytes, u64 empty_size,
+				     u64 search_start, u64 search_end,
+				     u64 hint_byte, struct btrfs_key *ins,
+				     u64 exclude_start, u64 exclude_nr,
+				     int data)
+{
+	int ret = 0;
+	struct btrfs_root *root = orig_root->fs_info->extent_root;
+	u64 total_needed = num_bytes;
+	u64 *last_ptr = NULL;
+	u64 last_wanted = 0;
+	struct btrfs_block_group_cache *block_group = NULL;
+	int chunk_alloc_done = 0;
+	int empty_cluster = 2 * 1024 * 1024;
+	int allowed_chunk_alloc = 0;
+	struct list_head *head = NULL, *cur = NULL;
+	int loop = 0;
+	int extra_loop = 0;
+	struct btrfs_space_info *space_info;
+
+	WARN_ON(num_bytes < root->sectorsize);
+	btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
+	ins->objectid = 0;
+	ins->offset = 0;
+
+	if (orig_root->ref_cows || empty_size)
+		allowed_chunk_alloc = 1;
+
+	if (data & BTRFS_BLOCK_GROUP_METADATA) {
+		last_ptr = &root->fs_info->last_alloc;
+		empty_cluster = 64 * 1024;
+	}
+
+	if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
+		last_ptr = &root->fs_info->last_data_alloc;
+
+	if (last_ptr) {
+		if (*last_ptr) {
+			hint_byte = *last_ptr;
+			last_wanted = *last_ptr;
+		} else
+			empty_size += empty_cluster;
+	} else {
+		empty_cluster = 0;
+	}
+	search_start = max(search_start, first_logical_byte(root, 0));
+	search_start = max(search_start, hint_byte);
+
+	if (last_wanted && search_start != last_wanted) {
+		last_wanted = 0;
+		empty_size += empty_cluster;
+	}
+
+	total_needed += empty_size;
+	block_group = btrfs_lookup_block_group(root->fs_info, search_start);
+	if (!block_group)
+		block_group = btrfs_lookup_first_block_group(root->fs_info,
+							     search_start);
+	space_info = __find_space_info(root->fs_info, data);
+
+	down_read(&space_info->groups_sem);
+	while (1) {
+		struct btrfs_free_space *free_space;
+		/*
+		 * the only way this happens if our hint points to a block
+		 * group thats not of the proper type, while looping this
+		 * should never happen
+		 */
+		if (empty_size)
+			extra_loop = 1;
+
+		if (!block_group)
+			goto new_group_no_lock;
+
+		if (unlikely(!block_group->cached)) {
+			mutex_lock(&block_group->cache_mutex);
+			ret = cache_block_group(root, block_group);
+			mutex_unlock(&block_group->cache_mutex);
+			if (ret)
+				break;
+		}
+
+		mutex_lock(&block_group->alloc_mutex);
+		if (unlikely(!block_group_bits(block_group, data)))
+			goto new_group;
+
+		if (unlikely(block_group->ro))
+			goto new_group;
+
+		free_space = btrfs_find_free_space(block_group, search_start,
+						   total_needed);
+		if (free_space) {
+			u64 start = block_group->key.objectid;
+			u64 end = block_group->key.objectid +
+				block_group->key.offset;
+
+			search_start = stripe_align(root, free_space->offset);
+
+			/* move on to the next group */
+			if (search_start + num_bytes >= search_end)
+				goto new_group;
+
+			/* move on to the next group */
+			if (search_start + num_bytes > end)
+				goto new_group;
+
+			if (last_wanted && search_start != last_wanted) {
+				total_needed += empty_cluster;
+				empty_size += empty_cluster;
+				last_wanted = 0;
+				/*
+				 * if search_start is still in this block group
+				 * then we just re-search this block group
+				 */
+				if (search_start >= start &&
+				    search_start < end) {
+					mutex_unlock(&block_group->alloc_mutex);
+					continue;
+				}
+
+				/* else we go to the next block group */
+				goto new_group;
+			}
+
+			if (exclude_nr > 0 &&
+			    (search_start + num_bytes > exclude_start &&
+			     search_start < exclude_start + exclude_nr)) {
+				search_start = exclude_start + exclude_nr;
+				/*
+				 * if search_start is still in this block group
+				 * then we just re-search this block group
+				 */
+				if (search_start >= start &&
+				    search_start < end) {
+					mutex_unlock(&block_group->alloc_mutex);
+					last_wanted = 0;
+					continue;
+				}
+
+				/* else we go to the next block group */
+				goto new_group;
+			}
+
+			ins->objectid = search_start;
+			ins->offset = num_bytes;
+
+			btrfs_remove_free_space_lock(block_group, search_start,
+						     num_bytes);
+			/* we are all good, lets return */
+			mutex_unlock(&block_group->alloc_mutex);
+			break;
+		}
+new_group:
+		mutex_unlock(&block_group->alloc_mutex);
+		put_block_group(block_group);
+		block_group = NULL;
+new_group_no_lock:
+		/* don't try to compare new allocations against the
+		 * last allocation any more
+		 */
+		last_wanted = 0;
+
+		/*
+		 * Here's how this works.
+		 * loop == 0: we were searching a block group via a hint
+		 *		and didn't find anything, so we start at
+		 *		the head of the block groups and keep searching
+		 * loop == 1: we're searching through all of the block groups
+		 *		if we hit the head again we have searched
+		 *		all of the block groups for this space and we
+		 *		need to try and allocate, if we cant error out.
+		 * loop == 2: we allocated more space and are looping through
+		 *		all of the block groups again.
+		 */
+		if (loop == 0) {
+			head = &space_info->block_groups;
+			cur = head->next;
+			loop++;
+		} else if (loop == 1 && cur == head) {
+			int keep_going;
+
+			/* at this point we give up on the empty_size
+			 * allocations and just try to allocate the min
+			 * space.
+			 *
+			 * The extra_loop field was set if an empty_size
+			 * allocation was attempted above, and if this
+			 * is try we need to try the loop again without
+			 * the additional empty_size.
+			 */
+			total_needed -= empty_size;
+			empty_size = 0;
+			keep_going = extra_loop;
+			loop++;
+
+			if (allowed_chunk_alloc && !chunk_alloc_done) {
+				up_read(&space_info->groups_sem);
+				ret = do_chunk_alloc(trans, root, num_bytes +
+						     2 * 1024 * 1024, data, 1);
+				down_read(&space_info->groups_sem);
+				if (ret < 0)
+					goto loop_check;
+				head = &space_info->block_groups;
+				/*
+				 * we've allocated a new chunk, keep
+				 * trying
+				 */
+				keep_going = 1;
+				chunk_alloc_done = 1;
+			} else if (!allowed_chunk_alloc) {
+				space_info->force_alloc = 1;
+			}
+loop_check:
+			if (keep_going) {
+				cur = head->next;
+				extra_loop = 0;
+			} else {
+				break;
+			}
+		} else if (cur == head) {
+			break;
+		}
+
+		block_group = list_entry(cur, struct btrfs_block_group_cache,
+					 list);
+		atomic_inc(&block_group->count);
+
+		search_start = block_group->key.objectid;
+		cur = cur->next;
+	}
+
+	/* we found what we needed */
+	if (ins->objectid) {
+		if (!(data & BTRFS_BLOCK_GROUP_DATA))
+			trans->block_group = block_group->key.objectid;
+
+		if (last_ptr)
+			*last_ptr = ins->objectid + ins->offset;
+		ret = 0;
+	} else if (!ret) {
+		printk(KERN_ERR "btrfs searching for %llu bytes, "
+		       "num_bytes %llu, loop %d, allowed_alloc %d\n",
+		       (unsigned long long)total_needed,
+		       (unsigned long long)num_bytes,
+		       loop, allowed_chunk_alloc);
+		ret = -ENOSPC;
+	}
+	if (block_group)
+		put_block_group(block_group);
+
+	up_read(&space_info->groups_sem);
+	return ret;
+}
+
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
+{
+	struct btrfs_block_group_cache *cache;
+	struct list_head *l;
+
+	printk(KERN_INFO "space_info has %llu free, is %sfull\n",
+	       (unsigned long long)(info->total_bytes - info->bytes_used -
+				    info->bytes_pinned - info->bytes_reserved),
+	       (info->full) ? "" : "not ");
+
+	down_read(&info->groups_sem);
+	list_for_each(l, &info->block_groups) {
+		cache = list_entry(l, struct btrfs_block_group_cache, list);
+		spin_lock(&cache->lock);
+		printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
+		       "%llu pinned %llu reserved\n",
+		       (unsigned long long)cache->key.objectid,
+		       (unsigned long long)cache->key.offset,
+		       (unsigned long long)btrfs_block_group_used(&cache->item),
+		       (unsigned long long)cache->pinned,
+		       (unsigned long long)cache->reserved);
+		btrfs_dump_free_space(cache, bytes);
+		spin_unlock(&cache->lock);
+	}
+	up_read(&info->groups_sem);
+}
+
+static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  u64 num_bytes, u64 min_alloc_size,
+				  u64 empty_size, u64 hint_byte,
+				  u64 search_end, struct btrfs_key *ins,
+				  u64 data)
+{
+	int ret;
+	u64 search_start = 0;
+	u64 alloc_profile;
+	struct btrfs_fs_info *info = root->fs_info;
+
+	if (data) {
+		alloc_profile = info->avail_data_alloc_bits &
+			info->data_alloc_profile;
+		data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
+	} else if (root == root->fs_info->chunk_root) {
+		alloc_profile = info->avail_system_alloc_bits &
+			info->system_alloc_profile;
+		data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
+	} else {
+		alloc_profile = info->avail_metadata_alloc_bits &
+			info->metadata_alloc_profile;
+		data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
+	}
+again:
+	data = btrfs_reduce_alloc_profile(root, data);
+	/*
+	 * the only place that sets empty_size is btrfs_realloc_node, which
+	 * is not called recursively on allocations
+	 */
+	if (empty_size || root->ref_cows) {
+		if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
+			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+				     2 * 1024 * 1024,
+				     BTRFS_BLOCK_GROUP_METADATA |
+				     (info->metadata_alloc_profile &
+				      info->avail_metadata_alloc_bits), 0);
+		}
+		ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+				     num_bytes + 2 * 1024 * 1024, data, 0);
+	}
+
+	WARN_ON(num_bytes < root->sectorsize);
+	ret = find_free_extent(trans, root, num_bytes, empty_size,
+			       search_start, search_end, hint_byte, ins,
+			       trans->alloc_exclude_start,
+			       trans->alloc_exclude_nr, data);
+
+	if (ret == -ENOSPC && num_bytes > min_alloc_size) {
+		num_bytes = num_bytes >> 1;
+		num_bytes = num_bytes & ~(root->sectorsize - 1);
+		num_bytes = max(num_bytes, min_alloc_size);
+		do_chunk_alloc(trans, root->fs_info->extent_root,
+			       num_bytes, data, 1);
+		goto again;
+	}
+	if (ret) {
+		struct btrfs_space_info *sinfo;
+
+		sinfo = __find_space_info(root->fs_info, data);
+		printk(KERN_ERR "btrfs allocation failed flags %llu, "
+		       "wanted %llu\n", (unsigned long long)data,
+		       (unsigned long long)num_bytes);
+		dump_space_info(sinfo, num_bytes);
+		BUG();
+	}
+
+	return ret;
+}
+
+int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
+{
+	struct btrfs_block_group_cache *cache;
+	int ret = 0;
+
+	cache = btrfs_lookup_block_group(root->fs_info, start);
+	if (!cache) {
+		printk(KERN_ERR "Unable to find block group for %llu\n",
+		       (unsigned long long)start);
+		return -ENOSPC;
+	}
+
+	ret = btrfs_discard_extent(root, start, len);
+
+	btrfs_add_free_space(cache, start, len);
+	put_block_group(cache);
+	update_reserved_extents(root, start, len, 0);
+
+	return ret;
+}
+
+int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  u64 num_bytes, u64 min_alloc_size,
+				  u64 empty_size, u64 hint_byte,
+				  u64 search_end, struct btrfs_key *ins,
+				  u64 data)
+{
+	int ret;
+	ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
+				     empty_size, hint_byte, search_end, ins,
+				     data);
+	update_reserved_extents(root, ins->objectid, ins->offset, 1);
+	return ret;
+}
+
+static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
+					 struct btrfs_root *root, u64 parent,
+					 u64 root_objectid, u64 ref_generation,
+					 u64 owner, struct btrfs_key *ins)
+{
+	int ret;
+	int pending_ret;
+	u64 super_used;
+	u64 root_used;
+	u64 num_bytes = ins->offset;
+	u32 sizes[2];
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_root *extent_root = info->extent_root;
+	struct btrfs_extent_item *extent_item;
+	struct btrfs_extent_ref *ref;
+	struct btrfs_path *path;
+	struct btrfs_key keys[2];
+
+	if (parent == 0)
+		parent = ins->objectid;
+
+	/* block accounting for super block */
+	spin_lock(&info->delalloc_lock);
+	super_used = btrfs_super_bytes_used(&info->super_copy);
+	btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
+
+	/* block accounting for root item */
+	root_used = btrfs_root_used(&root->root_item);
+	btrfs_set_root_used(&root->root_item, root_used + num_bytes);
+	spin_unlock(&info->delalloc_lock);
+
+	if (root == extent_root) {
+		struct pending_extent_op *extent_op;
+
+		extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+		BUG_ON(!extent_op);
+
+		extent_op->type = PENDING_EXTENT_INSERT;
+		extent_op->bytenr = ins->objectid;
+		extent_op->num_bytes = ins->offset;
+		extent_op->parent = parent;
+		extent_op->orig_parent = 0;
+		extent_op->generation = ref_generation;
+		extent_op->orig_generation = 0;
+		extent_op->level = (int)owner;
+		INIT_LIST_HEAD(&extent_op->list);
+		extent_op->del = 0;
+
+		mutex_lock(&root->fs_info->extent_ins_mutex);
+		set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
+				ins->objectid + ins->offset - 1,
+				EXTENT_WRITEBACK, GFP_NOFS);
+		set_state_private(&root->fs_info->extent_ins,
+				  ins->objectid, (unsigned long)extent_op);
+		mutex_unlock(&root->fs_info->extent_ins_mutex);
+		goto update_block;
+	}
+
+	memcpy(&keys[0], ins, sizeof(*ins));
+	keys[1].objectid = ins->objectid;
+	keys[1].type = BTRFS_EXTENT_REF_KEY;
+	keys[1].offset = parent;
+	sizes[0] = sizeof(*extent_item);
+	sizes[1] = sizeof(*ref);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
+				       sizes, 2);
+	BUG_ON(ret);
+
+	extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+				     struct btrfs_extent_item);
+	btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
+	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
+			     struct btrfs_extent_ref);
+
+	btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
+	btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
+	btrfs_set_ref_objectid(path->nodes[0], ref, owner);
+	btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
+
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+
+	trans->alloc_exclude_start = 0;
+	trans->alloc_exclude_nr = 0;
+	btrfs_free_path(path);
+	finish_current_insert(trans, extent_root, 0);
+	pending_ret = del_pending_extents(trans, extent_root, 0);
+
+	if (ret)
+		goto out;
+	if (pending_ret) {
+		ret = pending_ret;
+		goto out;
+	}
+
+update_block:
+	ret = update_block_group(trans, root, ins->objectid,
+				 ins->offset, 1, 0);
+	if (ret) {
+		printk(KERN_ERR "btrfs update block group failed for %llu "
+		       "%llu\n", (unsigned long long)ins->objectid,
+		       (unsigned long long)ins->offset);
+		BUG();
+	}
+out:
+	return ret;
+}
+
+int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root, u64 parent,
+				u64 root_objectid, u64 ref_generation,
+				u64 owner, struct btrfs_key *ins)
+{
+	int ret;
+
+	if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
+		return 0;
+	ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
+					    ref_generation, owner, ins);
+	update_reserved_extents(root, ins->objectid, ins->offset, 0);
+	return ret;
+}
+
+/*
+ * this is used by the tree logging recovery code.  It records that
+ * an extent has been allocated and makes sure to clear the free
+ * space cache bits as well
+ */
+int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root, u64 parent,
+				u64 root_objectid, u64 ref_generation,
+				u64 owner, struct btrfs_key *ins)
+{
+	int ret;
+	struct btrfs_block_group_cache *block_group;
+
+	block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
+	mutex_lock(&block_group->cache_mutex);
+	cache_block_group(root, block_group);
+	mutex_unlock(&block_group->cache_mutex);
+
+	ret = btrfs_remove_free_space(block_group, ins->objectid,
+				      ins->offset);
+	BUG_ON(ret);
+	put_block_group(block_group);
+	ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
+					    ref_generation, owner, ins);
+	return ret;
+}
+
+/*
+ * finds a free extent and does all the dirty work required for allocation
+ * returns the key for the extent through ins, and a tree buffer for
+ * the first block of the extent through buf.
+ *
+ * returns 0 if everything worked, non-zero otherwise.
+ */
+int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root,
+		       u64 num_bytes, u64 parent, u64 min_alloc_size,
+		       u64 root_objectid, u64 ref_generation,
+		       u64 owner_objectid, u64 empty_size, u64 hint_byte,
+		       u64 search_end, struct btrfs_key *ins, u64 data)
+{
+	int ret;
+
+	ret = __btrfs_reserve_extent(trans, root, num_bytes,
+				     min_alloc_size, empty_size, hint_byte,
+				     search_end, ins, data);
+	BUG_ON(ret);
+	if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
+		ret = __btrfs_alloc_reserved_extent(trans, root, parent,
+					root_objectid, ref_generation,
+					owner_objectid, ins);
+		BUG_ON(ret);
+
+	} else {
+		update_reserved_extents(root, ins->objectid, ins->offset, 1);
+	}
+	return ret;
+}
+
+struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
+					    struct btrfs_root *root,
+					    u64 bytenr, u32 blocksize)
+{
+	struct extent_buffer *buf;
+
+	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
+	if (!buf)
+		return ERR_PTR(-ENOMEM);
+	btrfs_set_header_generation(buf, trans->transid);
+	btrfs_tree_lock(buf);
+	clean_tree_block(trans, root, buf);
+	btrfs_set_buffer_uptodate(buf);
+	if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
+		set_extent_dirty(&root->dirty_log_pages, buf->start,
+			 buf->start + buf->len - 1, GFP_NOFS);
+	} else {
+		set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
+			 buf->start + buf->len - 1, GFP_NOFS);
+	}
+	trans->blocks_used++;
+	return buf;
+}
+
+/*
+ * helper function to allocate a block for a given tree
+ * returns the tree buffer or NULL.
+ */
+struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
+					     struct btrfs_root *root,
+					     u32 blocksize, u64 parent,
+					     u64 root_objectid,
+					     u64 ref_generation,
+					     int level,
+					     u64 hint,
+					     u64 empty_size)
+{
+	struct btrfs_key ins;
+	int ret;
+	struct extent_buffer *buf;
+
+	ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
+				 root_objectid, ref_generation, level,
+				 empty_size, hint, (u64)-1, &ins, 0);
+	if (ret) {
+		BUG_ON(ret > 0);
+		return ERR_PTR(ret);
+	}
+
+	buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
+	return buf;
+}

[PATCH 23.2/35] Btrfs snapshot deletion

[Chris Mason]

Snapshot deletion is done by walking down the btree and dropping references
as you go.  For a given node or leaf, if dropping the reference count will
free the block, all of the extents that block points to will also have
their reference counts dropped.

If dropping the reference count will not free a block, the entire
subtree pointed to by that block is skipped.

Since most of the metadata lives in leaves, a reference cache is maintained
in ram of the extents a given leaf has references on.  This limits the
number of btree leaves that need to be read while deleting old transactions
at run time.

Snapshot deletion happens incrementally.  The metadata for each root
indicates how far the last delete run got into the tree.  This allows
smaller atomic units for snapshot deletion, and it doesn't stall transaction
commit with long running operations.  It also means that once snapshot
deletion starts, that tree must not be accessed for reading or writing
outside of the snapshot deletion code.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- a/fs/btrfs/extent-tree.c	2009-01-08 09:56:31.803215082 -0500
+++ b/fs/btrfs/extent-tree.c	2009-01-08 09:56:07.321458970 -0500
@@ -3334,3 +3334,498 @@
 	buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
 	return buf;
 }
+
+int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root, struct extent_buffer *leaf)
+{
+	u64 leaf_owner;
+	u64 leaf_generation;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	int i;
+	int nritems;
+	int ret;
+
+	BUG_ON(!btrfs_is_leaf(leaf));
+	nritems = btrfs_header_nritems(leaf);
+	leaf_owner = btrfs_header_owner(leaf);
+	leaf_generation = btrfs_header_generation(leaf);
+
+	for (i = 0; i < nritems; i++) {
+		u64 disk_bytenr;
+		cond_resched();
+
+		btrfs_item_key_to_cpu(leaf, &key, i);
+		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+			continue;
+		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) ==
+		    BTRFS_FILE_EXTENT_INLINE)
+			continue;
+		/*
+		 * FIXME make sure to insert a trans record that
+		 * repeats the snapshot del on crash
+		 */
+		disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+		if (disk_bytenr == 0)
+			continue;
+
+		ret = __btrfs_free_extent(trans, root, disk_bytenr,
+				btrfs_file_extent_disk_num_bytes(leaf, fi),
+				leaf->start, leaf_owner, leaf_generation,
+				key.objectid, 0);
+		BUG_ON(ret);
+
+		atomic_inc(&root->fs_info->throttle_gen);
+		wake_up(&root->fs_info->transaction_throttle);
+		cond_resched();
+	}
+	return 0;
+}
+
+static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_leaf_ref *ref)
+{
+	int i;
+	int ret;
+	struct btrfs_extent_info *info = ref->extents;
+
+	for (i = 0; i < ref->nritems; i++) {
+		ret = __btrfs_free_extent(trans, root, info->bytenr,
+					  info->num_bytes, ref->bytenr,
+					  ref->owner, ref->generation,
+					  info->objectid, 0);
+
+		atomic_inc(&root->fs_info->throttle_gen);
+		wake_up(&root->fs_info->transaction_throttle);
+		cond_resched();
+
+		BUG_ON(ret);
+		info++;
+	}
+
+	return 0;
+}
+
+static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
+				     u64 len, u32 *refs)
+{
+	int ret;
+
+	ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
+	BUG_ON(ret);
+
+#if 0 /* some debugging code in case we see problems here */
+	/* if the refs count is one, it won't get increased again.  But
+	 * if the ref count is > 1, someone may be decreasing it at
+	 * the same time we are.
+	 */
+	if (*refs != 1) {
+		struct extent_buffer *eb = NULL;
+		eb = btrfs_find_create_tree_block(root, start, len);
+		if (eb)
+			btrfs_tree_lock(eb);
+
+		mutex_lock(&root->fs_info->alloc_mutex);
+		ret = lookup_extent_ref(NULL, root, start, len, refs);
+		BUG_ON(ret);
+		mutex_unlock(&root->fs_info->alloc_mutex);
+
+		if (eb) {
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
+		}
+		if (*refs == 1) {
+			printk(KERN_ERR "btrfs block %llu went down to one "
+			       "during drop_snap\n", (unsigned long long)start);
+		}
+
+	}
+#endif
+
+	cond_resched();
+	return ret;
+}
+
+/*
+ * helper function for drop_snapshot, this walks down the tree dropping ref
+ * counts as it goes.
+ */
+static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct btrfs_path *path, int *level)
+{
+	u64 root_owner;
+	u64 root_gen;
+	u64 bytenr;
+	u64 ptr_gen;
+	struct extent_buffer *next;
+	struct extent_buffer *cur;
+	struct extent_buffer *parent;
+	struct btrfs_leaf_ref *ref;
+	u32 blocksize;
+	int ret;
+	u32 refs;
+
+	WARN_ON(*level < 0);
+	WARN_ON(*level >= BTRFS_MAX_LEVEL);
+	ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
+				path->nodes[*level]->len, &refs);
+	BUG_ON(ret);
+	if (refs > 1)
+		goto out;
+
+	/*
+	 * walk down to the last node level and free all the leaves
+	 */
+	while (*level >= 0) {
+		WARN_ON(*level < 0);
+		WARN_ON(*level >= BTRFS_MAX_LEVEL);
+		cur = path->nodes[*level];
+
+		if (btrfs_header_level(cur) != *level)
+			WARN_ON(1);
+
+		if (path->slots[*level] >=
+		    btrfs_header_nritems(cur))
+			break;
+		if (*level == 0) {
+			ret = btrfs_drop_leaf_ref(trans, root, cur);
+			BUG_ON(ret);
+			break;
+		}
+		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
+		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
+		blocksize = btrfs_level_size(root, *level - 1);
+
+		ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
+		BUG_ON(ret);
+		if (refs != 1) {
+			parent = path->nodes[*level];
+			root_owner = btrfs_header_owner(parent);
+			root_gen = btrfs_header_generation(parent);
+			path->slots[*level]++;
+
+			ret = __btrfs_free_extent(trans, root, bytenr,
+						blocksize, parent->start,
+						root_owner, root_gen,
+						*level - 1, 1);
+			BUG_ON(ret);
+
+			atomic_inc(&root->fs_info->throttle_gen);
+			wake_up(&root->fs_info->transaction_throttle);
+			cond_resched();
+
+			continue;
+		}
+		/*
+		 * at this point, we have a single ref, and since the
+		 * only place referencing this extent is a dead root
+		 * the reference count should never go higher.
+		 * So, we don't need to check it again
+		 */
+		if (*level == 1) {
+			ref = btrfs_lookup_leaf_ref(root, bytenr);
+			if (ref && ref->generation != ptr_gen) {
+				btrfs_free_leaf_ref(root, ref);
+				ref = NULL;
+			}
+			if (ref) {
+				ret = cache_drop_leaf_ref(trans, root, ref);
+				BUG_ON(ret);
+				btrfs_remove_leaf_ref(root, ref);
+				btrfs_free_leaf_ref(root, ref);
+				*level = 0;
+				break;
+			}
+		}
+		next = btrfs_find_tree_block(root, bytenr, blocksize);
+		if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
+			free_extent_buffer(next);
+
+			next = read_tree_block(root, bytenr, blocksize,
+					       ptr_gen);
+			cond_resched();
+#if 0
+			/*
+			 * this is a debugging check and can go away
+			 * the ref should never go all the way down to 1
+			 * at this point
+			 */
+			ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
+						&refs);
+			BUG_ON(ret);
+			WARN_ON(refs != 1);
+#endif
+		}
+		WARN_ON(*level <= 0);
+		if (path->nodes[*level-1])
+			free_extent_buffer(path->nodes[*level-1]);
+		path->nodes[*level-1] = next;
+		*level = btrfs_header_level(next);
+		path->slots[*level] = 0;
+		cond_resched();
+	}
+out:
+	WARN_ON(*level < 0);
+	WARN_ON(*level >= BTRFS_MAX_LEVEL);
+
+	if (path->nodes[*level] == root->node) {
+		parent = path->nodes[*level];
+		bytenr = path->nodes[*level]->start;
+	} else {
+		parent = path->nodes[*level + 1];
+		bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
+	}
+
+	blocksize = btrfs_level_size(root, *level);
+	root_owner = btrfs_header_owner(parent);
+	root_gen = btrfs_header_generation(parent);
+
+	ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
+				  parent->start, root_owner, root_gen,
+				  *level, 1);
+	free_extent_buffer(path->nodes[*level]);
+	path->nodes[*level] = NULL;
+	*level += 1;
+	BUG_ON(ret);
+
+	cond_resched();
+	return 0;
+}
+
+/*
+ * helper function for drop_subtree, this function is similar to
+ * walk_down_tree. The main difference is that it checks reference
+ * counts while tree blocks are locked.
+ */
+static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path, int *level)
+{
+	struct extent_buffer *next;
+	struct extent_buffer *cur;
+	struct extent_buffer *parent;
+	u64 bytenr;
+	u64 ptr_gen;
+	u32 blocksize;
+	u32 refs;
+	int ret;
+
+	cur = path->nodes[*level];
+	ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
+				      &refs);
+	BUG_ON(ret);
+	if (refs > 1)
+		goto out;
+
+	while (*level >= 0) {
+		cur = path->nodes[*level];
+		if (*level == 0) {
+			ret = btrfs_drop_leaf_ref(trans, root, cur);
+			BUG_ON(ret);
+			clean_tree_block(trans, root, cur);
+			break;
+		}
+		if (path->slots[*level] >= btrfs_header_nritems(cur)) {
+			clean_tree_block(trans, root, cur);
+			break;
+		}
+
+		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
+		blocksize = btrfs_level_size(root, *level - 1);
+		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
+
+		next = read_tree_block(root, bytenr, blocksize, ptr_gen);
+		btrfs_tree_lock(next);
+
+		ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
+					      &refs);
+		BUG_ON(ret);
+		if (refs > 1) {
+			parent = path->nodes[*level];
+			ret = btrfs_free_extent(trans, root, bytenr,
+					blocksize, parent->start,
+					btrfs_header_owner(parent),
+					btrfs_header_generation(parent),
+					*level - 1, 1);
+			BUG_ON(ret);
+			path->slots[*level]++;
+			btrfs_tree_unlock(next);
+			free_extent_buffer(next);
+			continue;
+		}
+
+		*level = btrfs_header_level(next);
+		path->nodes[*level] = next;
+		path->slots[*level] = 0;
+		path->locks[*level] = 1;
+		cond_resched();
+	}
+out:
+	parent = path->nodes[*level + 1];
+	bytenr = path->nodes[*level]->start;
+	blocksize = path->nodes[*level]->len;
+
+	ret = btrfs_free_extent(trans, root, bytenr, blocksize,
+			parent->start, btrfs_header_owner(parent),
+			btrfs_header_generation(parent), *level, 1);
+	BUG_ON(ret);
+
+	if (path->locks[*level]) {
+		btrfs_tree_unlock(path->nodes[*level]);
+		path->locks[*level] = 0;
+	}
+	free_extent_buffer(path->nodes[*level]);
+	path->nodes[*level] = NULL;
+	*level += 1;
+	cond_resched();
+	return 0;
+}
+
+/*
+ * helper for dropping snapshots.  This walks back up the tree in the path
+ * to find the first node higher up where we haven't yet gone through
+ * all the slots
+ */
+static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 int *level, int max_level)
+{
+	u64 root_owner;
+	u64 root_gen;
+	struct btrfs_root_item *root_item = &root->root_item;
+	int i;
+	int slot;
+	int ret;
+
+	for (i = *level; i < max_level && path->nodes[i]; i++) {
+		slot = path->slots[i];
+		if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
+			struct extent_buffer *node;
+			struct btrfs_disk_key disk_key;
+			node = path->nodes[i];
+			path->slots[i]++;
+			*level = i;
+			WARN_ON(*level == 0);
+			btrfs_node_key(node, &disk_key, path->slots[i]);
+			memcpy(&root_item->drop_progress,
+			       &disk_key, sizeof(disk_key));
+			root_item->drop_level = i;
+			return 0;
+		} else {
+			struct extent_buffer *parent;
+			if (path->nodes[*level] == root->node)
+				parent = path->nodes[*level];
+			else
+				parent = path->nodes[*level + 1];
+
+			root_owner = btrfs_header_owner(parent);
+			root_gen = btrfs_header_generation(parent);
+
+			clean_tree_block(trans, root, path->nodes[*level]);
+			ret = btrfs_free_extent(trans, root,
+						path->nodes[*level]->start,
+						path->nodes[*level]->len,
+						parent->start, root_owner,
+						root_gen, *level, 1);
+			BUG_ON(ret);
+			if (path->locks[*level]) {
+				btrfs_tree_unlock(path->nodes[*level]);
+				path->locks[*level] = 0;
+			}
+			free_extent_buffer(path->nodes[*level]);
+			path->nodes[*level] = NULL;
+			*level = i + 1;
+		}
+	}
+	return 1;
+}
+
+/*
+ * drop the reference count on the tree rooted at 'snap'.  This traverses
+ * the tree freeing any blocks that have a ref count of zero after being
+ * decremented.
+ */
+int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
+			*root)
+{
+	int ret = 0;
+	int wret;
+	int level;
+	struct btrfs_path *path;
+	int i;
+	int orig_level;
+	struct btrfs_root_item *root_item = &root->root_item;
+
+	WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	level = btrfs_header_level(root->node);
+	orig_level = level;
+	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
+		path->nodes[level] = root->node;
+		extent_buffer_get(root->node);
+		path->slots[level] = 0;
+	} else {
+		struct btrfs_key key;
+		struct btrfs_disk_key found_key;
+		struct extent_buffer *node;
+
+		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
+		level = root_item->drop_level;
+		path->lowest_level = level;
+		wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+		if (wret < 0) {
+			ret = wret;
+			goto out;
+		}
+		node = path->nodes[level];
+		btrfs_node_key(node, &found_key, path->slots[level]);
+		WARN_ON(memcmp(&found_key, &root_item->drop_progress,
+			       sizeof(found_key)));
+		/*
+		 * unlock our path, this is safe because only this
+		 * function is allowed to delete this snapshot
+		 */
+		for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+			if (path->nodes[i] && path->locks[i]) {
+				path->locks[i] = 0;
+				btrfs_tree_unlock(path->nodes[i]);
+			}
+		}
+	}
+	while (1) {
+		wret = walk_down_tree(trans, root, path, &level);
+		if (wret > 0)
+			break;
+		if (wret < 0)
+			ret = wret;
+
+		wret = walk_up_tree(trans, root, path, &level,
+				    BTRFS_MAX_LEVEL);
+		if (wret > 0)
+			break;
+		if (wret < 0)
+			ret = wret;
+		if (trans->transaction->in_commit) {
+			ret = -EAGAIN;
+			break;
+		}
+		atomic_inc(&root->fs_info->throttle_gen);
+		wake_up(&root->fs_info->transaction_throttle);
+	}
+	for (i = 0; i <= orig_level; i++) {
+		if (path->nodes[i]) {
+			free_extent_buffer(path->nodes[i]);
+			path->nodes[i] = NULL;
+		}
+	}
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+

[PATCH 23.3/35] Btrfs extent relocation

[Chris Mason]

Block groups can be balanced and relocated.  Today the relocation happens
by following back references and forcing COW on all of the extents
that live inside a given block group.

Other the long term relocation code will be added that blindly shifts
extents around without changing their logical addresses.  That way it won't
have to follow the backreferences at all.

This code has the advantage of heavily testing the backrefs, and it
can restripe into new RAID levels.  When two or more snapshots
have references on an extent, the extent is COW'd only once and all
of the reference holders are updated to point to the new location.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- a/fs/btrfs/extent-tree.c	2009-01-08 09:56:07.321458970 -0500
+++ b/fs/btrfs/extent-tree.c	2009-01-08 09:50:24.557586920 -0500
@@ -1881,6 +1881,54 @@
 	return flags;
 }
 
+static int do_chunk_alloc(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *extent_root, u64 alloc_bytes,
+			  u64 flags, int force)
+{
+	struct btrfs_space_info *space_info;
+	u64 thresh;
+	int ret = 0;
+
+	mutex_lock(&extent_root->fs_info->chunk_mutex);
+
+	flags = btrfs_reduce_alloc_profile(extent_root, flags);
+
+	space_info = __find_space_info(extent_root->fs_info, flags);
+	if (!space_info) {
+		ret = update_space_info(extent_root->fs_info, flags,
+					0, 0, &space_info);
+		BUG_ON(ret);
+	}
+	BUG_ON(!space_info);
+
+	spin_lock(&space_info->lock);
+	if (space_info->force_alloc) {
+		force = 1;
+		space_info->force_alloc = 0;
+	}
+	if (space_info->full) {
+		spin_unlock(&space_info->lock);
+		goto out;
+	}
+
+	thresh = space_info->total_bytes - space_info->bytes_readonly;
+	thresh = div_factor(thresh, 6);
+	if (!force &&
+	   (space_info->bytes_used + space_info->bytes_pinned +
+	    space_info->bytes_reserved + alloc_bytes) < thresh) {
+		spin_unlock(&space_info->lock);
+		goto out;
+	}
+	spin_unlock(&space_info->lock);
+
+	ret = btrfs_alloc_chunk(trans, extent_root, flags);
+	if (ret)
+		space_info->full = 1;
+out:
+	mutex_unlock(&extent_root->fs_info->chunk_mutex);
+	return ret;
+}
+
 static int update_block_group(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
 			      u64 bytenr, u64 num_bytes, int alloc,
@@ -3829,3 +3877,2110 @@
 	return ret;
 }
 
+int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct extent_buffer *node,
+			struct extent_buffer *parent)
+{
+	struct btrfs_path *path;
+	int level;
+	int parent_level;
+	int ret = 0;
+	int wret;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	BUG_ON(!btrfs_tree_locked(parent));
+	parent_level = btrfs_header_level(parent);
+	extent_buffer_get(parent);
+	path->nodes[parent_level] = parent;
+	path->slots[parent_level] = btrfs_header_nritems(parent);
+
+	BUG_ON(!btrfs_tree_locked(node));
+	level = btrfs_header_level(node);
+	extent_buffer_get(node);
+	path->nodes[level] = node;
+	path->slots[level] = 0;
+
+	while (1) {
+		wret = walk_down_subtree(trans, root, path, &level);
+		if (wret < 0)
+			ret = wret;
+		if (wret != 0)
+			break;
+
+		wret = walk_up_tree(trans, root, path, &level, parent_level);
+		if (wret < 0)
+			ret = wret;
+		if (wret != 0)
+			break;
+	}
+
+	btrfs_free_path(path);
+	return ret;
+}
+
+static unsigned long calc_ra(unsigned long start, unsigned long last,
+			     unsigned long nr)
+{
+	return min(last, start + nr - 1);
+}
+
+static noinline int relocate_inode_pages(struct inode *inode, u64 start,
+					 u64 len)
+{
+	u64 page_start;
+	u64 page_end;
+	unsigned long first_index;
+	unsigned long last_index;
+	unsigned long i;
+	struct page *page;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct file_ra_state *ra;
+	struct btrfs_ordered_extent *ordered;
+	unsigned int total_read = 0;
+	unsigned int total_dirty = 0;
+	int ret = 0;
+
+	ra = kzalloc(sizeof(*ra), GFP_NOFS);
+
+	mutex_lock(&inode->i_mutex);
+	first_index = start >> PAGE_CACHE_SHIFT;
+	last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
+
+	/* make sure the dirty trick played by the caller work */
+	ret = invalidate_inode_pages2_range(inode->i_mapping,
+					    first_index, last_index);
+	if (ret)
+		goto out_unlock;
+
+	file_ra_state_init(ra, inode->i_mapping);
+
+	for (i = first_index ; i <= last_index; i++) {
+		if (total_read % ra->ra_pages == 0) {
+			btrfs_force_ra(inode->i_mapping, ra, NULL, i,
+				       calc_ra(i, last_index, ra->ra_pages));
+		}
+		total_read++;
+again:
+		if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
+			BUG_ON(1);
+		page = grab_cache_page(inode->i_mapping, i);
+		if (!page) {
+			ret = -ENOMEM;
+			goto out_unlock;
+		}
+		if (!PageUptodate(page)) {
+			btrfs_readpage(NULL, page);
+			lock_page(page);
+			if (!PageUptodate(page)) {
+				unlock_page(page);
+				page_cache_release(page);
+				ret = -EIO;
+				goto out_unlock;
+			}
+		}
+		wait_on_page_writeback(page);
+
+		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
+		page_end = page_start + PAGE_CACHE_SIZE - 1;
+		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
+
+		ordered = btrfs_lookup_ordered_extent(inode, page_start);
+		if (ordered) {
+			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			btrfs_start_ordered_extent(inode, ordered, 1);
+			btrfs_put_ordered_extent(ordered);
+			goto again;
+		}
+		set_page_extent_mapped(page);
+
+		if (i == first_index)
+			set_extent_bits(io_tree, page_start, page_end,
+					EXTENT_BOUNDARY, GFP_NOFS);
+		btrfs_set_extent_delalloc(inode, page_start, page_end);
+
+		set_page_dirty(page);
+		total_dirty++;
+
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+		unlock_page(page);
+		page_cache_release(page);
+	}
+
+out_unlock:
+	kfree(ra);
+	mutex_unlock(&inode->i_mutex);
+	balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
+	return ret;
+}
+
+static noinline int relocate_data_extent(struct inode *reloc_inode,
+					 struct btrfs_key *extent_key,
+					 u64 offset)
+{
+	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
+	struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
+	struct extent_map *em;
+	u64 start = extent_key->objectid - offset;
+	u64 end = start + extent_key->offset - 1;
+
+	em = alloc_extent_map(GFP_NOFS);
+	BUG_ON(!em || IS_ERR(em));
+
+	em->start = start;
+	em->len = extent_key->offset;
+	em->block_len = extent_key->offset;
+	em->block_start = extent_key->objectid;
+	em->bdev = root->fs_info->fs_devices->latest_bdev;
+	set_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+	/* setup extent map to cheat btrfs_readpage */
+	lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
+	while (1) {
+		int ret;
+		spin_lock(&em_tree->lock);
+		ret = add_extent_mapping(em_tree, em);
+		spin_unlock(&em_tree->lock);
+		if (ret != -EEXIST) {
+			free_extent_map(em);
+			break;
+		}
+		btrfs_drop_extent_cache(reloc_inode, start, end, 0);
+	}
+	unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
+
+	return relocate_inode_pages(reloc_inode, start, extent_key->offset);
+}
+
+struct btrfs_ref_path {
+	u64 extent_start;
+	u64 nodes[BTRFS_MAX_LEVEL];
+	u64 root_objectid;
+	u64 root_generation;
+	u64 owner_objectid;
+	u32 num_refs;
+	int lowest_level;
+	int current_level;
+	int shared_level;
+
+	struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
+	u64 new_nodes[BTRFS_MAX_LEVEL];
+};
+
+struct disk_extent {
+	u64 ram_bytes;
+	u64 disk_bytenr;
+	u64 disk_num_bytes;
+	u64 offset;
+	u64 num_bytes;
+	u8 compression;
+	u8 encryption;
+	u16 other_encoding;
+};
+
+static int is_cowonly_root(u64 root_objectid)
+{
+	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
+	    root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
+	    root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
+	    root_objectid == BTRFS_DEV_TREE_OBJECTID ||
+	    root_objectid == BTRFS_TREE_LOG_OBJECTID ||
+	    root_objectid == BTRFS_CSUM_TREE_OBJECTID)
+		return 1;
+	return 0;
+}
+
+static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *extent_root,
+				    struct btrfs_ref_path *ref_path,
+				    int first_time)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_path *path;
+	struct btrfs_extent_ref *ref;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	u64 bytenr;
+	u32 nritems;
+	int level;
+	int ret = 1;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	if (first_time) {
+		ref_path->lowest_level = -1;
+		ref_path->current_level = -1;
+		ref_path->shared_level = -1;
+		goto walk_up;
+	}
+walk_down:
+	level = ref_path->current_level - 1;
+	while (level >= -1) {
+		u64 parent;
+		if (level < ref_path->lowest_level)
+			break;
+
+		if (level >= 0)
+			bytenr = ref_path->nodes[level];
+		else
+			bytenr = ref_path->extent_start;
+		BUG_ON(bytenr == 0);
+
+		parent = ref_path->nodes[level + 1];
+		ref_path->nodes[level + 1] = 0;
+		ref_path->current_level = level;
+		BUG_ON(parent == 0);
+
+		key.objectid = bytenr;
+		key.offset = parent + 1;
+		key.type = BTRFS_EXTENT_REF_KEY;
+
+		ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
+		if (ret < 0)
+			goto out;
+		BUG_ON(ret == 0);
+
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(extent_root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0)
+				goto next;
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.objectid == bytenr &&
+		    found_key.type == BTRFS_EXTENT_REF_KEY) {
+			if (level < ref_path->shared_level)
+				ref_path->shared_level = level;
+			goto found;
+		}
+next:
+		level--;
+		btrfs_release_path(extent_root, path);
+		cond_resched();
+	}
+	/* reached lowest level */
+	ret = 1;
+	goto out;
+walk_up:
+	level = ref_path->current_level;
+	while (level < BTRFS_MAX_LEVEL - 1) {
+		u64 ref_objectid;
+
+		if (level >= 0)
+			bytenr = ref_path->nodes[level];
+		else
+			bytenr = ref_path->extent_start;
+
+		BUG_ON(bytenr == 0);
+
+		key.objectid = bytenr;
+		key.offset = 0;
+		key.type = BTRFS_EXTENT_REF_KEY;
+
+		ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
+		if (ret < 0)
+			goto out;
+
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(extent_root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0) {
+				/* the extent was freed by someone */
+				if (ref_path->lowest_level == level)
+					goto out;
+				btrfs_release_path(extent_root, path);
+				goto walk_down;
+			}
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.objectid != bytenr ||
+				found_key.type != BTRFS_EXTENT_REF_KEY) {
+			/* the extent was freed by someone */
+			if (ref_path->lowest_level == level) {
+				ret = 1;
+				goto out;
+			}
+			btrfs_release_path(extent_root, path);
+			goto walk_down;
+		}
+found:
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				struct btrfs_extent_ref);
+		ref_objectid = btrfs_ref_objectid(leaf, ref);
+		if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
+			if (first_time) {
+				level = (int)ref_objectid;
+				BUG_ON(level >= BTRFS_MAX_LEVEL);
+				ref_path->lowest_level = level;
+				ref_path->current_level = level;
+				ref_path->nodes[level] = bytenr;
+			} else {
+				WARN_ON(ref_objectid != level);
+			}
+		} else {
+			WARN_ON(level != -1);
+		}
+		first_time = 0;
+
+		if (ref_path->lowest_level == level) {
+			ref_path->owner_objectid = ref_objectid;
+			ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
+		}
+
+		/*
+		 * the block is tree root or the block isn't in reference
+		 * counted tree.
+		 */
+		if (found_key.objectid == found_key.offset ||
+		    is_cowonly_root(btrfs_ref_root(leaf, ref))) {
+			ref_path->root_objectid = btrfs_ref_root(leaf, ref);
+			ref_path->root_generation =
+				btrfs_ref_generation(leaf, ref);
+			if (level < 0) {
+				/* special reference from the tree log */
+				ref_path->nodes[0] = found_key.offset;
+				ref_path->current_level = 0;
+			}
+			ret = 0;
+			goto out;
+		}
+
+		level++;
+		BUG_ON(ref_path->nodes[level] != 0);
+		ref_path->nodes[level] = found_key.offset;
+		ref_path->current_level = level;
+
+		/*
+		 * the reference was created in the running transaction,
+		 * no need to continue walking up.
+		 */
+		if (btrfs_ref_generation(leaf, ref) == trans->transid) {
+			ref_path->root_objectid = btrfs_ref_root(leaf, ref);
+			ref_path->root_generation =
+				btrfs_ref_generation(leaf, ref);
+			ret = 0;
+			goto out;
+		}
+
+		btrfs_release_path(extent_root, path);
+		cond_resched();
+	}
+	/* reached max tree level, but no tree root found. */
+	BUG();
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
+				struct btrfs_root *extent_root,
+				struct btrfs_ref_path *ref_path,
+				u64 extent_start)
+{
+	memset(ref_path, 0, sizeof(*ref_path));
+	ref_path->extent_start = extent_start;
+
+	return __next_ref_path(trans, extent_root, ref_path, 1);
+}
+
+static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *extent_root,
+			       struct btrfs_ref_path *ref_path)
+{
+	return __next_ref_path(trans, extent_root, ref_path, 0);
+}
+
+static noinline int get_new_locations(struct inode *reloc_inode,
+				      struct btrfs_key *extent_key,
+				      u64 offset, int no_fragment,
+				      struct disk_extent **extents,
+				      int *nr_extents)
+{
+	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
+	struct btrfs_path *path;
+	struct btrfs_file_extent_item *fi;
+	struct extent_buffer *leaf;
+	struct disk_extent *exts = *extents;
+	struct btrfs_key found_key;
+	u64 cur_pos;
+	u64 last_byte;
+	u32 nritems;
+	int nr = 0;
+	int max = *nr_extents;
+	int ret;
+
+	WARN_ON(!no_fragment && *extents);
+	if (!exts) {
+		max = 1;
+		exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
+		if (!exts)
+			return -ENOMEM;
+	}
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	cur_pos = extent_key->objectid - offset;
+	last_byte = extent_key->objectid + extent_key->offset;
+	ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
+				       cur_pos, 0);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	while (1) {
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0)
+				break;
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.offset != cur_pos ||
+		    found_key.type != BTRFS_EXTENT_DATA_KEY ||
+		    found_key.objectid != reloc_inode->i_ino)
+			break;
+
+		fi = btrfs_item_ptr(leaf, path->slots[0],
+				    struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) !=
+		    BTRFS_FILE_EXTENT_REG ||
+		    btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
+			break;
+
+		if (nr == max) {
+			struct disk_extent *old = exts;
+			max *= 2;
+			exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
+			memcpy(exts, old, sizeof(*exts) * nr);
+			if (old != *extents)
+				kfree(old);
+		}
+
+		exts[nr].disk_bytenr =
+			btrfs_file_extent_disk_bytenr(leaf, fi);
+		exts[nr].disk_num_bytes =
+			btrfs_file_extent_disk_num_bytes(leaf, fi);
+		exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
+		exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
+		exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
+		exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
+		exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
+		exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
+									   fi);
+		BUG_ON(exts[nr].offset > 0);
+		BUG_ON(exts[nr].compression || exts[nr].encryption);
+		BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
+
+		cur_pos += exts[nr].num_bytes;
+		nr++;
+
+		if (cur_pos + offset >= last_byte)
+			break;
+
+		if (no_fragment) {
+			ret = 1;
+			goto out;
+		}
+		path->slots[0]++;
+	}
+
+	BUG_ON(cur_pos + offset > last_byte);
+	if (cur_pos + offset < last_byte) {
+		ret = -ENOENT;
+		goto out;
+	}
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	if (ret) {
+		if (exts != *extents)
+			kfree(exts);
+	} else {
+		*extents = exts;
+		*nr_extents = nr;
+	}
+	return ret;
+}
+
+static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_path *path,
+					struct btrfs_key *extent_key,
+					struct btrfs_key *leaf_key,
+					struct btrfs_ref_path *ref_path,
+					struct disk_extent *new_extents,
+					int nr_extents)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_file_extent_item *fi;
+	struct inode *inode = NULL;
+	struct btrfs_key key;
+	u64 lock_start = 0;
+	u64 lock_end = 0;
+	u64 num_bytes;
+	u64 ext_offset;
+	u64 first_pos;
+	u32 nritems;
+	int nr_scaned = 0;
+	int extent_locked = 0;
+	int extent_type;
+	int ret;
+
+	memcpy(&key, leaf_key, sizeof(key));
+	first_pos = INT_LIMIT(loff_t) - extent_key->offset;
+	if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
+		if (key.objectid < ref_path->owner_objectid ||
+		    (key.objectid == ref_path->owner_objectid &&
+		     key.type < BTRFS_EXTENT_DATA_KEY)) {
+			key.objectid = ref_path->owner_objectid;
+			key.type = BTRFS_EXTENT_DATA_KEY;
+			key.offset = 0;
+		}
+	}
+
+	while (1) {
+		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+		if (ret < 0)
+			goto out;
+
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+next:
+		if (extent_locked && ret > 0) {
+			/*
+			 * the file extent item was modified by someone
+			 * before the extent got locked.
+			 */
+			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				      lock_end, GFP_NOFS);
+			extent_locked = 0;
+		}
+
+		if (path->slots[0] >= nritems) {
+			if (++nr_scaned > 2)
+				break;
+
+			BUG_ON(extent_locked);
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0)
+				break;
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+
+		if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
+			if ((key.objectid > ref_path->owner_objectid) ||
+			    (key.objectid == ref_path->owner_objectid &&
+			     key.type > BTRFS_EXTENT_DATA_KEY) ||
+			    (key.offset >= first_pos + extent_key->offset))
+				break;
+		}
+
+		if (inode && key.objectid != inode->i_ino) {
+			BUG_ON(extent_locked);
+			btrfs_release_path(root, path);
+			mutex_unlock(&inode->i_mutex);
+			iput(inode);
+			inode = NULL;
+			continue;
+		}
+
+		if (key.type != BTRFS_EXTENT_DATA_KEY) {
+			path->slots[0]++;
+			ret = 1;
+			goto next;
+		}
+		fi = btrfs_item_ptr(leaf, path->slots[0],
+				    struct btrfs_file_extent_item);
+		extent_type = btrfs_file_extent_type(leaf, fi);
+		if ((extent_type != BTRFS_FILE_EXTENT_REG &&
+		     extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
+		    (btrfs_file_extent_disk_bytenr(leaf, fi) !=
+		     extent_key->objectid)) {
+			path->slots[0]++;
+			ret = 1;
+			goto next;
+		}
+
+		num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
+		ext_offset = btrfs_file_extent_offset(leaf, fi);
+
+		if (first_pos > key.offset - ext_offset)
+			first_pos = key.offset - ext_offset;
+
+		if (!extent_locked) {
+			lock_start = key.offset;
+			lock_end = lock_start + num_bytes - 1;
+		} else {
+			if (lock_start > key.offset ||
+			    lock_end + 1 < key.offset + num_bytes) {
+				unlock_extent(&BTRFS_I(inode)->io_tree,
+					      lock_start, lock_end, GFP_NOFS);
+				extent_locked = 0;
+			}
+		}
+
+		if (!inode) {
+			btrfs_release_path(root, path);
+
+			inode = btrfs_iget_locked(root->fs_info->sb,
+						  key.objectid, root);
+			if (inode->i_state & I_NEW) {
+				BTRFS_I(inode)->root = root;
+				BTRFS_I(inode)->location.objectid =
+					key.objectid;
+				BTRFS_I(inode)->location.type =
+					BTRFS_INODE_ITEM_KEY;
+				BTRFS_I(inode)->location.offset = 0;
+				btrfs_read_locked_inode(inode);
+				unlock_new_inode(inode);
+			}
+			/*
+			 * some code call btrfs_commit_transaction while
+			 * holding the i_mutex, so we can't use mutex_lock
+			 * here.
+			 */
+			if (is_bad_inode(inode) ||
+			    !mutex_trylock(&inode->i_mutex)) {
+				iput(inode);
+				inode = NULL;
+				key.offset = (u64)-1;
+				goto skip;
+			}
+		}
+
+		if (!extent_locked) {
+			struct btrfs_ordered_extent *ordered;
+
+			btrfs_release_path(root, path);
+
+			lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				    lock_end, GFP_NOFS);
+			ordered = btrfs_lookup_first_ordered_extent(inode,
+								    lock_end);
+			if (ordered &&
+			    ordered->file_offset <= lock_end &&
+			    ordered->file_offset + ordered->len > lock_start) {
+				unlock_extent(&BTRFS_I(inode)->io_tree,
+					      lock_start, lock_end, GFP_NOFS);
+				btrfs_start_ordered_extent(inode, ordered, 1);
+				btrfs_put_ordered_extent(ordered);
+				key.offset += num_bytes;
+				goto skip;
+			}
+			if (ordered)
+				btrfs_put_ordered_extent(ordered);
+
+			extent_locked = 1;
+			continue;
+		}
+
+		if (nr_extents == 1) {
+			/* update extent pointer in place */
+			btrfs_set_file_extent_disk_bytenr(leaf, fi,
+						new_extents[0].disk_bytenr);
+			btrfs_set_file_extent_disk_num_bytes(leaf, fi,
+						new_extents[0].disk_num_bytes);
+			btrfs_mark_buffer_dirty(leaf);
+
+			btrfs_drop_extent_cache(inode, key.offset,
+						key.offset + num_bytes - 1, 0);
+
+			ret = btrfs_inc_extent_ref(trans, root,
+						new_extents[0].disk_bytenr,
+						new_extents[0].disk_num_bytes,
+						leaf->start,
+						root->root_key.objectid,
+						trans->transid,
+						key.objectid);
+			BUG_ON(ret);
+
+			ret = btrfs_free_extent(trans, root,
+						extent_key->objectid,
+						extent_key->offset,
+						leaf->start,
+						btrfs_header_owner(leaf),
+						btrfs_header_generation(leaf),
+						key.objectid, 0);
+			BUG_ON(ret);
+
+			btrfs_release_path(root, path);
+			key.offset += num_bytes;
+		} else {
+			BUG_ON(1);
+#if 0
+			u64 alloc_hint;
+			u64 extent_len;
+			int i;
+			/*
+			 * drop old extent pointer at first, then insert the
+			 * new pointers one bye one
+			 */
+			btrfs_release_path(root, path);
+			ret = btrfs_drop_extents(trans, root, inode, key.offset,
+						 key.offset + num_bytes,
+						 key.offset, &alloc_hint);
+			BUG_ON(ret);
+
+			for (i = 0; i < nr_extents; i++) {
+				if (ext_offset >= new_extents[i].num_bytes) {
+					ext_offset -= new_extents[i].num_bytes;
+					continue;
+				}
+				extent_len = min(new_extents[i].num_bytes -
+						 ext_offset, num_bytes);
+
+				ret = btrfs_insert_empty_item(trans, root,
+							      path, &key,
+							      sizeof(*fi));
+				BUG_ON(ret);
+
+				leaf = path->nodes[0];
+				fi = btrfs_item_ptr(leaf, path->slots[0],
+						struct btrfs_file_extent_item);
+				btrfs_set_file_extent_generation(leaf, fi,
+							trans->transid);
+				btrfs_set_file_extent_type(leaf, fi,
+							BTRFS_FILE_EXTENT_REG);
+				btrfs_set_file_extent_disk_bytenr(leaf, fi,
+						new_extents[i].disk_bytenr);
+				btrfs_set_file_extent_disk_num_bytes(leaf, fi,
+						new_extents[i].disk_num_bytes);
+				btrfs_set_file_extent_ram_bytes(leaf, fi,
+						new_extents[i].ram_bytes);
+
+				btrfs_set_file_extent_compression(leaf, fi,
+						new_extents[i].compression);
+				btrfs_set_file_extent_encryption(leaf, fi,
+						new_extents[i].encryption);
+				btrfs_set_file_extent_other_encoding(leaf, fi,
+						new_extents[i].other_encoding);
+
+				btrfs_set_file_extent_num_bytes(leaf, fi,
+							extent_len);
+				ext_offset += new_extents[i].offset;
+				btrfs_set_file_extent_offset(leaf, fi,
+							ext_offset);
+				btrfs_mark_buffer_dirty(leaf);
+
+				btrfs_drop_extent_cache(inode, key.offset,
+						key.offset + extent_len - 1, 0);
+
+				ret = btrfs_inc_extent_ref(trans, root,
+						new_extents[i].disk_bytenr,
+						new_extents[i].disk_num_bytes,
+						leaf->start,
+						root->root_key.objectid,
+						trans->transid, key.objectid);
+				BUG_ON(ret);
+				btrfs_release_path(root, path);
+
+				inode_add_bytes(inode, extent_len);
+
+				ext_offset = 0;
+				num_bytes -= extent_len;
+				key.offset += extent_len;
+
+				if (num_bytes == 0)
+					break;
+			}
+			BUG_ON(i >= nr_extents);
+#endif
+		}
+
+		if (extent_locked) {
+			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				      lock_end, GFP_NOFS);
+			extent_locked = 0;
+		}
+skip:
+		if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
+		    key.offset >= first_pos + extent_key->offset)
+			break;
+
+		cond_resched();
+	}
+	ret = 0;
+out:
+	btrfs_release_path(root, path);
+	if (inode) {
+		mutex_unlock(&inode->i_mutex);
+		if (extent_locked) {
+			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				      lock_end, GFP_NOFS);
+		}
+		iput(inode);
+	}
+	return ret;
+}
+
+int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct extent_buffer *buf, u64 orig_start)
+{
+	int level;
+	int ret;
+
+	BUG_ON(btrfs_header_generation(buf) != trans->transid);
+	BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
+
+	level = btrfs_header_level(buf);
+	if (level == 0) {
+		struct btrfs_leaf_ref *ref;
+		struct btrfs_leaf_ref *orig_ref;
+
+		orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
+		if (!orig_ref)
+			return -ENOENT;
+
+		ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
+		if (!ref) {
+			btrfs_free_leaf_ref(root, orig_ref);
+			return -ENOMEM;
+		}
+
+		ref->nritems = orig_ref->nritems;
+		memcpy(ref->extents, orig_ref->extents,
+			sizeof(ref->extents[0]) * ref->nritems);
+
+		btrfs_free_leaf_ref(root, orig_ref);
+
+		ref->root_gen = trans->transid;
+		ref->bytenr = buf->start;
+		ref->owner = btrfs_header_owner(buf);
+		ref->generation = btrfs_header_generation(buf);
+		ret = btrfs_add_leaf_ref(root, ref, 0);
+		WARN_ON(ret);
+		btrfs_free_leaf_ref(root, ref);
+	}
+	return 0;
+}
+
+static noinline int invalidate_extent_cache(struct btrfs_root *root,
+					struct extent_buffer *leaf,
+					struct btrfs_block_group_cache *group,
+					struct btrfs_root *target_root)
+{
+	struct btrfs_key key;
+	struct inode *inode = NULL;
+	struct btrfs_file_extent_item *fi;
+	u64 num_bytes;
+	u64 skip_objectid = 0;
+	u32 nritems;
+	u32 i;
+
+	nritems = btrfs_header_nritems(leaf);
+	for (i = 0; i < nritems; i++) {
+		btrfs_item_key_to_cpu(leaf, &key, i);
+		if (key.objectid == skip_objectid ||
+		    key.type != BTRFS_EXTENT_DATA_KEY)
+			continue;
+		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) ==
+		    BTRFS_FILE_EXTENT_INLINE)
+			continue;
+		if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
+			continue;
+		if (!inode || inode->i_ino != key.objectid) {
+			iput(inode);
+			inode = btrfs_ilookup(target_root->fs_info->sb,
+					      key.objectid, target_root, 1);
+		}
+		if (!inode) {
+			skip_objectid = key.objectid;
+			continue;
+		}
+		num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
+
+		lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
+			    key.offset + num_bytes - 1, GFP_NOFS);
+		btrfs_drop_extent_cache(inode, key.offset,
+					key.offset + num_bytes - 1, 1);
+		unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
+			      key.offset + num_bytes - 1, GFP_NOFS);
+		cond_resched();
+	}
+	iput(inode);
+	return 0;
+}
+
+static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct extent_buffer *leaf,
+					struct btrfs_block_group_cache *group,
+					struct inode *reloc_inode)
+{
+	struct btrfs_key key;
+	struct btrfs_key extent_key;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_leaf_ref *ref;
+	struct disk_extent *new_extent;
+	u64 bytenr;
+	u64 num_bytes;
+	u32 nritems;
+	u32 i;
+	int ext_index;
+	int nr_extent;
+	int ret;
+
+	new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
+	BUG_ON(!new_extent);
+
+	ref = btrfs_lookup_leaf_ref(root, leaf->start);
+	BUG_ON(!ref);
+
+	ext_index = -1;
+	nritems = btrfs_header_nritems(leaf);
+	for (i = 0; i < nritems; i++) {
+		btrfs_item_key_to_cpu(leaf, &key, i);
+		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+			continue;
+		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) ==
+		    BTRFS_FILE_EXTENT_INLINE)
+			continue;
+		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
+		if (bytenr == 0)
+			continue;
+
+		ext_index++;
+		if (bytenr >= group->key.objectid + group->key.offset ||
+		    bytenr + num_bytes <= group->key.objectid)
+			continue;
+
+		extent_key.objectid = bytenr;
+		extent_key.offset = num_bytes;
+		extent_key.type = BTRFS_EXTENT_ITEM_KEY;
+		nr_extent = 1;
+		ret = get_new_locations(reloc_inode, &extent_key,
+					group->key.objectid, 1,
+					&new_extent, &nr_extent);
+		if (ret > 0)
+			continue;
+		BUG_ON(ret < 0);
+
+		BUG_ON(ref->extents[ext_index].bytenr != bytenr);
+		BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
+		ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
+		ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
+
+		btrfs_set_file_extent_disk_bytenr(leaf, fi,
+						new_extent->disk_bytenr);
+		btrfs_set_file_extent_disk_num_bytes(leaf, fi,
+						new_extent->disk_num_bytes);
+		btrfs_mark_buffer_dirty(leaf);
+
+		ret = btrfs_inc_extent_ref(trans, root,
+					new_extent->disk_bytenr,
+					new_extent->disk_num_bytes,
+					leaf->start,
+					root->root_key.objectid,
+					trans->transid, key.objectid);
+		BUG_ON(ret);
+		ret = btrfs_free_extent(trans, root,
+					bytenr, num_bytes, leaf->start,
+					btrfs_header_owner(leaf),
+					btrfs_header_generation(leaf),
+					key.objectid, 0);
+		BUG_ON(ret);
+		cond_resched();
+	}
+	kfree(new_extent);
+	BUG_ON(ext_index + 1 != ref->nritems);
+	btrfs_free_leaf_ref(root, ref);
+	return 0;
+}
+
+int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	int ret;
+
+	if (root->reloc_root) {
+		reloc_root = root->reloc_root;
+		root->reloc_root = NULL;
+		list_add(&reloc_root->dead_list,
+			 &root->fs_info->dead_reloc_roots);
+
+		btrfs_set_root_bytenr(&reloc_root->root_item,
+				      reloc_root->node->start);
+		btrfs_set_root_level(&root->root_item,
+				     btrfs_header_level(reloc_root->node));
+		memset(&reloc_root->root_item.drop_progress, 0,
+			sizeof(struct btrfs_disk_key));
+		reloc_root->root_item.drop_level = 0;
+
+		ret = btrfs_update_root(trans, root->fs_info->tree_root,
+					&reloc_root->root_key,
+					&reloc_root->root_item);
+		BUG_ON(ret);
+	}
+	return 0;
+}
+
+int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *reloc_root;
+	struct btrfs_root *prev_root = NULL;
+	struct list_head dead_roots;
+	int ret;
+	unsigned long nr;
+
+	INIT_LIST_HEAD(&dead_roots);
+	list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
+
+	while (!list_empty(&dead_roots)) {
+		reloc_root = list_entry(dead_roots.prev,
+					struct btrfs_root, dead_list);
+		list_del_init(&reloc_root->dead_list);
+
+		BUG_ON(reloc_root->commit_root != NULL);
+		while (1) {
+			trans = btrfs_join_transaction(root, 1);
+			BUG_ON(!trans);
+
+			mutex_lock(&root->fs_info->drop_mutex);
+			ret = btrfs_drop_snapshot(trans, reloc_root);
+			if (ret != -EAGAIN)
+				break;
+			mutex_unlock(&root->fs_info->drop_mutex);
+
+			nr = trans->blocks_used;
+			ret = btrfs_end_transaction(trans, root);
+			BUG_ON(ret);
+			btrfs_btree_balance_dirty(root, nr);
+		}
+
+		free_extent_buffer(reloc_root->node);
+
+		ret = btrfs_del_root(trans, root->fs_info->tree_root,
+				     &reloc_root->root_key);
+		BUG_ON(ret);
+		mutex_unlock(&root->fs_info->drop_mutex);
+
+		nr = trans->blocks_used;
+		ret = btrfs_end_transaction(trans, root);
+		BUG_ON(ret);
+		btrfs_btree_balance_dirty(root, nr);
+
+		kfree(prev_root);
+		prev_root = reloc_root;
+	}
+	if (prev_root) {
+		btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
+		kfree(prev_root);
+	}
+	return 0;
+}
+
+int btrfs_add_dead_reloc_root(struct btrfs_root *root)
+{
+	list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
+	return 0;
+}
+
+int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key location;
+	int found;
+	int ret;
+
+	mutex_lock(&root->fs_info->tree_reloc_mutex);
+	ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
+	BUG_ON(ret);
+	found = !list_empty(&root->fs_info->dead_reloc_roots);
+	mutex_unlock(&root->fs_info->tree_reloc_mutex);
+
+	if (found) {
+		trans = btrfs_start_transaction(root, 1);
+		BUG_ON(!trans);
+		ret = btrfs_commit_transaction(trans, root);
+		BUG_ON(ret);
+	}
+
+	location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
+	location.offset = (u64)-1;
+	location.type = BTRFS_ROOT_ITEM_KEY;
+
+	reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
+	BUG_ON(!reloc_root);
+	btrfs_orphan_cleanup(reloc_root);
+	return 0;
+}
+
+static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	struct extent_buffer *eb;
+	struct btrfs_root_item *root_item;
+	struct btrfs_key root_key;
+	int ret;
+
+	BUG_ON(!root->ref_cows);
+	if (root->reloc_root)
+		return 0;
+
+	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
+	BUG_ON(!root_item);
+
+	ret = btrfs_copy_root(trans, root, root->commit_root,
+			      &eb, BTRFS_TREE_RELOC_OBJECTID);
+	BUG_ON(ret);
+
+	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
+	root_key.offset = root->root_key.objectid;
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+
+	memcpy(root_item, &root->root_item, sizeof(root_item));
+	btrfs_set_root_refs(root_item, 0);
+	btrfs_set_root_bytenr(root_item, eb->start);
+	btrfs_set_root_level(root_item, btrfs_header_level(eb));
+	btrfs_set_root_generation(root_item, trans->transid);
+
+	btrfs_tree_unlock(eb);
+	free_extent_buffer(eb);
+
+	ret = btrfs_insert_root(trans, root->fs_info->tree_root,
+				&root_key, root_item);
+	BUG_ON(ret);
+	kfree(root_item);
+
+	reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
+						 &root_key);
+	BUG_ON(!reloc_root);
+	reloc_root->last_trans = trans->transid;
+	reloc_root->commit_root = NULL;
+	reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
+
+	root->reloc_root = reloc_root;
+	return 0;
+}
+
+/*
+ * Core function of space balance.
+ *
+ * The idea is using reloc trees to relocate tree blocks in reference
+ * counted roots. There is one reloc tree for each subvol, and all
+ * reloc trees share same root key objectid. Reloc trees are snapshots
+ * of the latest committed roots of subvols (root->commit_root).
+ *
+ * To relocate a tree block referenced by a subvol, there are two steps.
+ * COW the block through subvol's reloc tree, then update block pointer
+ * in the subvol to point to the new block. Since all reloc trees share
+ * same root key objectid, doing special handing for tree blocks owned
+ * by them is easy. Once a tree block has been COWed in one reloc tree,
+ * we can use the resulting new block directly when the same block is
+ * required to COW again through other reloc trees. By this way, relocated
+ * tree blocks are shared between reloc trees, so they are also shared
+ * between subvols.
+ */
+static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      struct btrfs_key *first_key,
+				      struct btrfs_ref_path *ref_path,
+				      struct btrfs_block_group_cache *group,
+				      struct inode *reloc_inode)
+{
+	struct btrfs_root *reloc_root;
+	struct extent_buffer *eb = NULL;
+	struct btrfs_key *keys;
+	u64 *nodes;
+	int level;
+	int shared_level;
+	int lowest_level = 0;
+	int ret;
+
+	if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
+		lowest_level = ref_path->owner_objectid;
+
+	if (!root->ref_cows) {
+		path->lowest_level = lowest_level;
+		ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
+		BUG_ON(ret < 0);
+		path->lowest_level = 0;
+		btrfs_release_path(root, path);
+		return 0;
+	}
+
+	mutex_lock(&root->fs_info->tree_reloc_mutex);
+	ret = init_reloc_tree(trans, root);
+	BUG_ON(ret);
+	reloc_root = root->reloc_root;
+
+	shared_level = ref_path->shared_level;
+	ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
+
+	keys = ref_path->node_keys;
+	nodes = ref_path->new_nodes;
+	memset(&keys[shared_level + 1], 0,
+	       sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
+	memset(&nodes[shared_level + 1], 0,
+	       sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
+
+	if (nodes[lowest_level] == 0) {
+		path->lowest_level = lowest_level;
+		ret = btrfs_search_slot(trans, reloc_root, first_key, path,
+					0, 1);
+		BUG_ON(ret);
+		for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
+			eb = path->nodes[level];
+			if (!eb || eb == reloc_root->node)
+				break;
+			nodes[level] = eb->start;
+			if (level == 0)
+				btrfs_item_key_to_cpu(eb, &keys[level], 0);
+			else
+				btrfs_node_key_to_cpu(eb, &keys[level], 0);
+		}
+		if (nodes[0] &&
+		    ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			eb = path->nodes[0];
+			ret = replace_extents_in_leaf(trans, reloc_root, eb,
+						      group, reloc_inode);
+			BUG_ON(ret);
+		}
+		btrfs_release_path(reloc_root, path);
+	} else {
+		ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
+				       lowest_level);
+		BUG_ON(ret);
+	}
+
+	/*
+	 * replace tree blocks in the fs tree with tree blocks in
+	 * the reloc tree.
+	 */
+	ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
+	BUG_ON(ret < 0);
+
+	if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+		ret = btrfs_search_slot(trans, reloc_root, first_key, path,
+					0, 0);
+		BUG_ON(ret);
+		extent_buffer_get(path->nodes[0]);
+		eb = path->nodes[0];
+		btrfs_release_path(reloc_root, path);
+		ret = invalidate_extent_cache(reloc_root, eb, group, root);
+		BUG_ON(ret);
+		free_extent_buffer(eb);
+	}
+
+	mutex_unlock(&root->fs_info->tree_reloc_mutex);
+	path->lowest_level = 0;
+	return 0;
+}
+
+static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_path *path,
+					struct btrfs_key *first_key,
+					struct btrfs_ref_path *ref_path)
+{
+	int ret;
+
+	ret = relocate_one_path(trans, root, path, first_key,
+				ref_path, NULL, NULL);
+	BUG_ON(ret);
+
+	if (root == root->fs_info->extent_root)
+		btrfs_extent_post_op(trans, root);
+
+	return 0;
+}
+
+static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *extent_root,
+				    struct btrfs_path *path,
+				    struct btrfs_key *extent_key)
+{
+	int ret;
+
+	ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
+	if (ret)
+		goto out;
+	ret = btrfs_del_item(trans, extent_root, path);
+out:
+	btrfs_release_path(extent_root, path);
+	return ret;
+}
+
+static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
+						struct btrfs_ref_path *ref_path)
+{
+	struct btrfs_key root_key;
+
+	root_key.objectid = ref_path->root_objectid;
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+	if (is_cowonly_root(ref_path->root_objectid))
+		root_key.offset = 0;
+	else
+		root_key.offset = (u64)-1;
+
+	return btrfs_read_fs_root_no_name(fs_info, &root_key);
+}
+
+static noinline int relocate_one_extent(struct btrfs_root *extent_root,
+					struct btrfs_path *path,
+					struct btrfs_key *extent_key,
+					struct btrfs_block_group_cache *group,
+					struct inode *reloc_inode, int pass)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *found_root;
+	struct btrfs_ref_path *ref_path = NULL;
+	struct disk_extent *new_extents = NULL;
+	int nr_extents = 0;
+	int loops;
+	int ret;
+	int level;
+	struct btrfs_key first_key;
+	u64 prev_block = 0;
+
+
+	trans = btrfs_start_transaction(extent_root, 1);
+	BUG_ON(!trans);
+
+	if (extent_key->objectid == 0) {
+		ret = del_extent_zero(trans, extent_root, path, extent_key);
+		goto out;
+	}
+
+	ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
+	if (!ref_path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	for (loops = 0; ; loops++) {
+		if (loops == 0) {
+			ret = btrfs_first_ref_path(trans, extent_root, ref_path,
+						   extent_key->objectid);
+		} else {
+			ret = btrfs_next_ref_path(trans, extent_root, ref_path);
+		}
+		if (ret < 0)
+			goto out;
+		if (ret > 0)
+			break;
+
+		if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
+		    ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
+			continue;
+
+		found_root = read_ref_root(extent_root->fs_info, ref_path);
+		BUG_ON(!found_root);
+		/*
+		 * for reference counted tree, only process reference paths
+		 * rooted at the latest committed root.
+		 */
+		if (found_root->ref_cows &&
+		    ref_path->root_generation != found_root->root_key.offset)
+			continue;
+
+		if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			if (pass == 0) {
+				/*
+				 * copy data extents to new locations
+				 */
+				u64 group_start = group->key.objectid;
+				ret = relocate_data_extent(reloc_inode,
+							   extent_key,
+							   group_start);
+				if (ret < 0)
+					goto out;
+				break;
+			}
+			level = 0;
+		} else {
+			level = ref_path->owner_objectid;
+		}
+
+		if (prev_block != ref_path->nodes[level]) {
+			struct extent_buffer *eb;
+			u64 block_start = ref_path->nodes[level];
+			u64 block_size = btrfs_level_size(found_root, level);
+
+			eb = read_tree_block(found_root, block_start,
+					     block_size, 0);
+			btrfs_tree_lock(eb);
+			BUG_ON(level != btrfs_header_level(eb));
+
+			if (level == 0)
+				btrfs_item_key_to_cpu(eb, &first_key, 0);
+			else
+				btrfs_node_key_to_cpu(eb, &first_key, 0);
+
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
+			prev_block = block_start;
+		}
+
+		btrfs_record_root_in_trans(found_root);
+		if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			/*
+			 * try to update data extent references while
+			 * keeping metadata shared between snapshots.
+			 */
+			if (pass == 1) {
+				ret = relocate_one_path(trans, found_root,
+						path, &first_key, ref_path,
+						group, reloc_inode);
+				if (ret < 0)
+					goto out;
+				continue;
+			}
+			/*
+			 * use fallback method to process the remaining
+			 * references.
+			 */
+			if (!new_extents) {
+				u64 group_start = group->key.objectid;
+				new_extents = kmalloc(sizeof(*new_extents),
+						      GFP_NOFS);
+				nr_extents = 1;
+				ret = get_new_locations(reloc_inode,
+							extent_key,
+							group_start, 1,
+							&new_extents,
+							&nr_extents);
+				if (ret)
+					goto out;
+			}
+			ret = replace_one_extent(trans, found_root,
+						path, extent_key,
+						&first_key, ref_path,
+						new_extents, nr_extents);
+		} else {
+			ret = relocate_tree_block(trans, found_root, path,
+						  &first_key, ref_path);
+		}
+		if (ret < 0)
+			goto out;
+	}
+	ret = 0;
+out:
+	btrfs_end_transaction(trans, extent_root);
+	kfree(new_extents);
+	kfree(ref_path);
+	return ret;
+}
+
+static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
+{
+	u64 num_devices;
+	u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
+		BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
+
+	num_devices = root->fs_info->fs_devices->rw_devices;
+	if (num_devices == 1) {
+		stripped |= BTRFS_BLOCK_GROUP_DUP;
+		stripped = flags & ~stripped;
+
+		/* turn raid0 into single device chunks */
+		if (flags & BTRFS_BLOCK_GROUP_RAID0)
+			return stripped;
+
+		/* turn mirroring into duplication */
+		if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
+			     BTRFS_BLOCK_GROUP_RAID10))
+			return stripped | BTRFS_BLOCK_GROUP_DUP;
+		return flags;
+	} else {
+		/* they already had raid on here, just return */
+		if (flags & stripped)
+			return flags;
+
+		stripped |= BTRFS_BLOCK_GROUP_DUP;
+		stripped = flags & ~stripped;
+
+		/* switch duplicated blocks with raid1 */
+		if (flags & BTRFS_BLOCK_GROUP_DUP)
+			return stripped | BTRFS_BLOCK_GROUP_RAID1;
+
+		/* turn single device chunks into raid0 */
+		return stripped | BTRFS_BLOCK_GROUP_RAID0;
+	}
+	return flags;
+}
+
+static int __alloc_chunk_for_shrink(struct btrfs_root *root,
+		     struct btrfs_block_group_cache *shrink_block_group,
+		     int force)
+{
+	struct btrfs_trans_handle *trans;
+	u64 new_alloc_flags;
+	u64 calc;
+
+	spin_lock(&shrink_block_group->lock);
+	if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
+		spin_unlock(&shrink_block_group->lock);
+
+		trans = btrfs_start_transaction(root, 1);
+		spin_lock(&shrink_block_group->lock);
+
+		new_alloc_flags = update_block_group_flags(root,
+						   shrink_block_group->flags);
+		if (new_alloc_flags != shrink_block_group->flags) {
+			calc =
+			     btrfs_block_group_used(&shrink_block_group->item);
+		} else {
+			calc = shrink_block_group->key.offset;
+		}
+		spin_unlock(&shrink_block_group->lock);
+
+		do_chunk_alloc(trans, root->fs_info->extent_root,
+			       calc + 2 * 1024 * 1024, new_alloc_flags, force);
+
+		btrfs_end_transaction(trans, root);
+	} else
+		spin_unlock(&shrink_block_group->lock);
+	return 0;
+}
+
+static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 u64 objectid, u64 size)
+{
+	struct btrfs_path *path;
+	struct btrfs_inode_item *item;
+	struct extent_buffer *leaf;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
+	if (ret)
+		goto out;
+
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
+	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
+	btrfs_set_inode_generation(leaf, item, 1);
+	btrfs_set_inode_size(leaf, item, size);
+	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
+	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(root, path);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
+					struct btrfs_block_group_cache *group)
+{
+	struct inode *inode = NULL;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root;
+	struct btrfs_key root_key;
+	u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
+	int err = 0;
+
+	root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+	root_key.offset = (u64)-1;
+	root = btrfs_read_fs_root_no_name(fs_info, &root_key);
+	if (IS_ERR(root))
+		return ERR_CAST(root);
+
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(!trans);
+
+	err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
+	if (err)
+		goto out;
+
+	err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
+	BUG_ON(err);
+
+	err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
+				       group->key.offset, 0, group->key.offset,
+				       0, 0, 0);
+	BUG_ON(err);
+
+	inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
+	if (inode->i_state & I_NEW) {
+		BTRFS_I(inode)->root = root;
+		BTRFS_I(inode)->location.objectid = objectid;
+		BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
+		BTRFS_I(inode)->location.offset = 0;
+		btrfs_read_locked_inode(inode);
+		unlock_new_inode(inode);
+		BUG_ON(is_bad_inode(inode));
+	} else {
+		BUG_ON(1);
+	}
+	BTRFS_I(inode)->index_cnt = group->key.objectid;
+
+	err = btrfs_orphan_add(trans, inode);
+out:
+	btrfs_end_transaction(trans, root);
+	if (err) {
+		if (inode)
+			iput(inode);
+		inode = ERR_PTR(err);
+	}
+	return inode;
+}
+
+int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
+{
+
+	struct btrfs_ordered_sum *sums;
+	struct btrfs_sector_sum *sector_sum;
+	struct btrfs_ordered_extent *ordered;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct list_head list;
+	size_t offset;
+	int ret;
+	u64 disk_bytenr;
+
+	INIT_LIST_HEAD(&list);
+
+	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
+	BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
+
+	disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
+	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
+				       disk_bytenr + len - 1, &list);
+
+	while (!list_empty(&list)) {
+		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
+		list_del_init(&sums->list);
+
+		sector_sum = sums->sums;
+		sums->bytenr = ordered->start;
+
+		offset = 0;
+		while (offset < sums->len) {
+			sector_sum->bytenr += ordered->start - disk_bytenr;
+			sector_sum++;
+			offset += root->sectorsize;
+		}
+
+		btrfs_add_ordered_sum(inode, ordered, sums);
+	}
+	btrfs_put_ordered_extent(ordered);
+	return 0;
+}
+
+int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_path *path;
+	struct btrfs_fs_info *info = root->fs_info;
+	struct extent_buffer *leaf;
+	struct inode *reloc_inode;
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_key key;
+	u64 skipped;
+	u64 cur_byte;
+	u64 total_found;
+	u32 nritems;
+	int ret;
+	int progress;
+	int pass = 0;
+
+	root = root->fs_info->extent_root;
+
+	block_group = btrfs_lookup_block_group(info, group_start);
+	BUG_ON(!block_group);
+
+	printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
+	       (unsigned long long)block_group->key.objectid,
+	       (unsigned long long)block_group->flags);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	reloc_inode = create_reloc_inode(info, block_group);
+	BUG_ON(IS_ERR(reloc_inode));
+
+	__alloc_chunk_for_shrink(root, block_group, 1);
+	set_block_group_readonly(block_group);
+
+	btrfs_start_delalloc_inodes(info->tree_root);
+	btrfs_wait_ordered_extents(info->tree_root, 0);
+again:
+	skipped = 0;
+	total_found = 0;
+	progress = 0;
+	key.objectid = block_group->key.objectid;
+	key.offset = 0;
+	key.type = 0;
+	cur_byte = key.objectid;
+
+	trans = btrfs_start_transaction(info->tree_root, 1);
+	btrfs_commit_transaction(trans, info->tree_root);
+
+	mutex_lock(&root->fs_info->cleaner_mutex);
+	btrfs_clean_old_snapshots(info->tree_root);
+	btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
+	mutex_unlock(&root->fs_info->cleaner_mutex);
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+		if (ret < 0)
+			goto out;
+next:
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto out;
+			if (ret == 1) {
+				ret = 0;
+				break;
+			}
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+
+		if (key.objectid >= block_group->key.objectid +
+		    block_group->key.offset)
+			break;
+
+		if (progress && need_resched()) {
+			btrfs_release_path(root, path);
+			cond_resched();
+			progress = 0;
+			continue;
+		}
+		progress = 1;
+
+		if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
+		    key.objectid + key.offset <= cur_byte) {
+			path->slots[0]++;
+			goto next;
+		}
+
+		total_found++;
+		cur_byte = key.objectid + key.offset;
+		btrfs_release_path(root, path);
+
+		__alloc_chunk_for_shrink(root, block_group, 0);
+		ret = relocate_one_extent(root, path, &key, block_group,
+					  reloc_inode, pass);
+		BUG_ON(ret < 0);
+		if (ret > 0)
+			skipped++;
+
+		key.objectid = cur_byte;
+		key.type = 0;
+		key.offset = 0;
+	}
+
+	btrfs_release_path(root, path);
+
+	if (pass == 0) {
+		btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
+		invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
+	}
+
+	if (total_found > 0) {
+		printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
+		       (unsigned long long)total_found, pass);
+		pass++;
+		if (total_found == skipped && pass > 2) {
+			iput(reloc_inode);
+			reloc_inode = create_reloc_inode(info, block_group);
+			pass = 0;
+		}
+		goto again;
+	}
+
+	/* delete reloc_inode */
+	iput(reloc_inode);
+
+	/* unpin extents in this range */
+	trans = btrfs_start_transaction(info->tree_root, 1);
+	btrfs_commit_transaction(trans, info->tree_root);
+
+	spin_lock(&block_group->lock);
+	WARN_ON(block_group->pinned > 0);
+	WARN_ON(block_group->reserved > 0);
+	WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
+	spin_unlock(&block_group->lock);
+	put_block_group(block_group);
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int find_first_block_group(struct btrfs_root *root,
+		struct btrfs_path *path, struct btrfs_key *key)
+{
+	int ret = 0;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+	int slot;
+
+	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+
+	while (1) {
+		slot = path->slots[0];
+		leaf = path->nodes[0];
+		if (slot >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret == 0)
+				continue;
+			if (ret < 0)
+				goto out;
+			break;
+		}
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+
+		if (found_key.objectid >= key->objectid &&
+		    found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
+			ret = 0;
+			goto out;
+		}
+		path->slots[0]++;
+	}
+	ret = -ENOENT;
+out:
+	return ret;
+}
+
+int btrfs_free_block_groups(struct btrfs_fs_info *info)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct rb_node *n;
+
+	spin_lock(&info->block_group_cache_lock);
+	while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
+		block_group = rb_entry(n, struct btrfs_block_group_cache,
+				       cache_node);
+		rb_erase(&block_group->cache_node,
+			 &info->block_group_cache_tree);
+		spin_unlock(&info->block_group_cache_lock);
+
+		btrfs_remove_free_space_cache(block_group);
+		down_write(&block_group->space_info->groups_sem);
+		list_del(&block_group->list);
+		up_write(&block_group->space_info->groups_sem);
+
+		WARN_ON(atomic_read(&block_group->count) != 1);
+		kfree(block_group);
+
+		spin_lock(&info->block_group_cache_lock);
+	}
+	spin_unlock(&info->block_group_cache_lock);
+	return 0;
+}
+
+int btrfs_read_block_groups(struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_space_info *space_info;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+
+	root = info->extent_root;
+	key.objectid = 0;
+	key.offset = 0;
+	btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	while (1) {
+		ret = find_first_block_group(root, path, &key);
+		if (ret > 0) {
+			ret = 0;
+			goto error;
+		}
+		if (ret != 0)
+			goto error;
+
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		cache = kzalloc(sizeof(*cache), GFP_NOFS);
+		if (!cache) {
+			ret = -ENOMEM;
+			break;
+		}
+
+		atomic_set(&cache->count, 1);
+		spin_lock_init(&cache->lock);
+		mutex_init(&cache->alloc_mutex);
+		mutex_init(&cache->cache_mutex);
+		INIT_LIST_HEAD(&cache->list);
+		read_extent_buffer(leaf, &cache->item,
+				   btrfs_item_ptr_offset(leaf, path->slots[0]),
+				   sizeof(cache->item));
+		memcpy(&cache->key, &found_key, sizeof(found_key));
+
+		key.objectid = found_key.objectid + found_key.offset;
+		btrfs_release_path(root, path);
+		cache->flags = btrfs_block_group_flags(&cache->item);
+
+		ret = update_space_info(info, cache->flags, found_key.offset,
+					btrfs_block_group_used(&cache->item),
+					&space_info);
+		BUG_ON(ret);
+		cache->space_info = space_info;
+		down_write(&space_info->groups_sem);
+		list_add_tail(&cache->list, &space_info->block_groups);
+		up_write(&space_info->groups_sem);
+
+		ret = btrfs_add_block_group_cache(root->fs_info, cache);
+		BUG_ON(ret);
+
+		set_avail_alloc_bits(root->fs_info, cache->flags);
+		if (btrfs_chunk_readonly(root, cache->key.objectid))
+			set_block_group_readonly(cache);
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_make_block_group(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, u64 bytes_used,
+			   u64 type, u64 chunk_objectid, u64 chunk_offset,
+			   u64 size)
+{
+	int ret;
+	struct btrfs_root *extent_root;
+	struct btrfs_block_group_cache *cache;
+
+	extent_root = root->fs_info->extent_root;
+
+	root->fs_info->last_trans_new_blockgroup = trans->transid;
+
+	cache = kzalloc(sizeof(*cache), GFP_NOFS);
+	if (!cache)
+		return -ENOMEM;
+
+	cache->key.objectid = chunk_offset;
+	cache->key.offset = size;
+	cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
+	atomic_set(&cache->count, 1);
+	spin_lock_init(&cache->lock);
+	mutex_init(&cache->alloc_mutex);
+	mutex_init(&cache->cache_mutex);
+	INIT_LIST_HEAD(&cache->list);
+
+	btrfs_set_block_group_used(&cache->item, bytes_used);
+	btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
+	cache->flags = type;
+	btrfs_set_block_group_flags(&cache->item, type);
+
+	ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
+				&cache->space_info);
+	BUG_ON(ret);
+	down_write(&cache->space_info->groups_sem);
+	list_add_tail(&cache->list, &cache->space_info->block_groups);
+	up_write(&cache->space_info->groups_sem);
+
+	ret = btrfs_add_block_group_cache(root->fs_info, cache);
+	BUG_ON(ret);
+
+	ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
+				sizeof(cache->item));
+	BUG_ON(ret);
+
+	finish_current_insert(trans, extent_root, 0);
+	ret = del_pending_extents(trans, extent_root, 0);
+	BUG_ON(ret);
+	set_avail_alloc_bits(extent_root->fs_info, type);
+
+	return 0;
+}
+
+int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 group_start)
+{
+	struct btrfs_path *path;
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_key key;
+	int ret;
+
+	root = root->fs_info->extent_root;
+
+	block_group = btrfs_lookup_block_group(root->fs_info, group_start);
+	BUG_ON(!block_group);
+	BUG_ON(!block_group->ro);
+
+	memcpy(&key, &block_group->key, sizeof(key));
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	btrfs_remove_free_space_cache(block_group);
+	rb_erase(&block_group->cache_node,
+		 &root->fs_info->block_group_cache_tree);
+	down_write(&block_group->space_info->groups_sem);
+	list_del(&block_group->list);
+	up_write(&block_group->space_info->groups_sem);
+
+	spin_lock(&block_group->space_info->lock);
+	block_group->space_info->total_bytes -= block_group->key.offset;
+	block_group->space_info->bytes_readonly -= block_group->key.offset;
+	spin_unlock(&block_group->space_info->lock);
+	block_group->space_info->full = 0;
+
+	put_block_group(block_group);
+	put_block_group(block_group);
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret > 0)
+		ret = -EIO;
+	if (ret < 0)
+		goto out;
+
+	ret = btrfs_del_item(trans, root, path);
+out:
+	btrfs_free_path(path);
+	return ret;
+}

[PATCH 26.1/35] Btrfs btree core

[Chris Mason]

This implements ctrees, which are checksummed btrees.  The btrfs
btree stores all data in the leaves and only has pointers to leaves
or nodes in the nodes.  Data stored in the leaves is broken up
into two areas.

* Fixed sized item headers contain the keys, the offsets, and size of the
item data in the leaf

* Variable sized data

These two areas grow toward each other.  Searches are done on the fixed
sized item headers, but all the useful data lives in the variable sized
data portion of the leaf.

Btrfs keys have three parts: [ u64 objectid, u8 type, u64 offset ]

The type field controls what the offset field means, and the objectid
field is generally the inode number, but this varies by the metadata
being stored.

All metadata in btrfs is indexed in one of these trees.  The fields of
the key are used in different ways depending on which btree it is.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- /dev/null	2008-06-10 15:56:50.000000000 -0400
+++ b/fs/btrfs/ctree.c	2009-01-08 10:04:14.659535797 -0500
@@ -0,0 +1,2441 @@
+/*
+ * Copyright (C) 2007,2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "print-tree.h"
+#include "locking.h"
+
+static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_path *path, int level);
+static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *ins_key,
+		      struct btrfs_path *path, int data_size, int extend);
+static int push_node_left(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct extent_buffer *dst,
+			  struct extent_buffer *src, int empty);
+static int balance_node_right(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct extent_buffer *dst_buf,
+			      struct extent_buffer *src_buf);
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_path *path, int level, int slot);
+
+inline void btrfs_init_path(struct btrfs_path *p)
+{
+	memset(p, 0, sizeof(*p));
+}
+
+struct btrfs_path *btrfs_alloc_path(void)
+{
+	struct btrfs_path *path;
+	path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
+	if (path) {
+		btrfs_init_path(path);
+		path->reada = 1;
+	}
+	return path;
+}
+
+/* this also releases the path */
+void btrfs_free_path(struct btrfs_path *p)
+{
+	btrfs_release_path(NULL, p);
+	kmem_cache_free(btrfs_path_cachep, p);
+}
+
+/*
+ * path release drops references on the extent buffers in the path
+ * and it drops any locks held by this path
+ *
+ * It is safe to call this on paths that no locks or extent buffers held.
+ */
+noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
+{
+	int i;
+
+	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+		p->slots[i] = 0;
+		if (!p->nodes[i])
+			continue;
+		if (p->locks[i]) {
+			btrfs_tree_unlock(p->nodes[i]);
+			p->locks[i] = 0;
+		}
+		free_extent_buffer(p->nodes[i]);
+		p->nodes[i] = NULL;
+	}
+}
+
+/*
+ * safely gets a reference on the root node of a tree.  A lock
+ * is not taken, so a concurrent writer may put a different node
+ * at the root of the tree.  See btrfs_lock_root_node for the
+ * looping required.
+ *
+ * The extent buffer returned by this has a reference taken, so
+ * it won't disappear.  It may stop being the root of the tree
+ * at any time because there are no locks held.
+ */
+struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
+{
+	struct extent_buffer *eb;
+	spin_lock(&root->node_lock);
+	eb = root->node;
+	extent_buffer_get(eb);
+	spin_unlock(&root->node_lock);
+	return eb;
+}
+
+/* loop around taking references on and locking the root node of the
+ * tree until you end up with a lock on the root.  A locked buffer
+ * is returned, with a reference held.
+ */
+struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
+{
+	struct extent_buffer *eb;
+
+	while (1) {
+		eb = btrfs_root_node(root);
+		btrfs_tree_lock(eb);
+
+		spin_lock(&root->node_lock);
+		if (eb == root->node) {
+			spin_unlock(&root->node_lock);
+			break;
+		}
+		spin_unlock(&root->node_lock);
+
+		btrfs_tree_unlock(eb);
+		free_extent_buffer(eb);
+	}
+	return eb;
+}
+
+/* cowonly root (everything not a reference counted cow subvolume), just get
+ * put onto a simple dirty list.  transaction.c walks this to make sure they
+ * get properly updated on disk.
+ */
+static void add_root_to_dirty_list(struct btrfs_root *root)
+{
+	if (root->track_dirty && list_empty(&root->dirty_list)) {
+		list_add(&root->dirty_list,
+			 &root->fs_info->dirty_cowonly_roots);
+	}
+}
+
+/*
+ * used by snapshot creation to make a copy of a root for a tree with
+ * a given objectid.  The buffer with the new root node is returned in
+ * cow_ret, and this func returns zero on success or a negative error code.
+ */
+int btrfs_copy_root(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root,
+		      struct extent_buffer *buf,
+		      struct extent_buffer **cow_ret, u64 new_root_objectid)
+{
+	struct extent_buffer *cow;
+	u32 nritems;
+	int ret = 0;
+	int level;
+	struct btrfs_root *new_root;
+
+	new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
+	if (!new_root)
+		return -ENOMEM;
+
+	memcpy(new_root, root, sizeof(*new_root));
+	new_root->root_key.objectid = new_root_objectid;
+
+	WARN_ON(root->ref_cows && trans->transid !=
+		root->fs_info->running_transaction->transid);
+	WARN_ON(root->ref_cows && trans->transid != root->last_trans);
+
+	level = btrfs_header_level(buf);
+	nritems = btrfs_header_nritems(buf);
+
+	cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
+				     new_root_objectid, trans->transid,
+				     level, buf->start, 0);
+	if (IS_ERR(cow)) {
+		kfree(new_root);
+		return PTR_ERR(cow);
+	}
+
+	copy_extent_buffer(cow, buf, 0, 0, cow->len);
+	btrfs_set_header_bytenr(cow, cow->start);
+	btrfs_set_header_generation(cow, trans->transid);
+	btrfs_set_header_owner(cow, new_root_objectid);
+	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
+
+	write_extent_buffer(cow, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(cow),
+			    BTRFS_FSID_SIZE);
+
+	WARN_ON(btrfs_header_generation(buf) > trans->transid);
+	ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
+	kfree(new_root);
+
+	if (ret)
+		return ret;
+
+	btrfs_mark_buffer_dirty(cow);
+	*cow_ret = cow;
+	return 0;
+}
+
+/*
+ * does the dirty work in cow of a single block.  The parent block (if
+ * supplied) is updated to point to the new cow copy.  The new buffer is marked
+ * dirty and returned locked.  If you modify the block it needs to be marked
+ * dirty again.
+ *
+ * search_start -- an allocation hint for the new block
+ *
+ * empty_size -- a hint that you plan on doing more cow.  This is the size in
+ * bytes the allocator should try to find free next to the block it returns.
+ * This is just a hint and may be ignored by the allocator.
+ *
+ * prealloc_dest -- if you have already reserved a destination for the cow,
+ * this uses that block instead of allocating a new one.
+ * btrfs_alloc_reserved_extent is used to finish the allocation.
+ */
+static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
+			     struct extent_buffer *buf,
+			     struct extent_buffer *parent, int parent_slot,
+			     struct extent_buffer **cow_ret,
+			     u64 search_start, u64 empty_size,
+			     u64 prealloc_dest)
+{
+	u64 parent_start;
+	struct extent_buffer *cow;
+	u32 nritems;
+	int ret = 0;
+	int level;
+	int unlock_orig = 0;
+
+	if (*cow_ret == buf)
+		unlock_orig = 1;
+
+	WARN_ON(!btrfs_tree_locked(buf));
+
+	if (parent)
+		parent_start = parent->start;
+	else
+		parent_start = 0;
+
+	WARN_ON(root->ref_cows && trans->transid !=
+		root->fs_info->running_transaction->transid);
+	WARN_ON(root->ref_cows && trans->transid != root->last_trans);
+
+	level = btrfs_header_level(buf);
+	nritems = btrfs_header_nritems(buf);
+
+	if (prealloc_dest) {
+		struct btrfs_key ins;
+
+		ins.objectid = prealloc_dest;
+		ins.offset = buf->len;
+		ins.type = BTRFS_EXTENT_ITEM_KEY;
+
+		ret = btrfs_alloc_reserved_extent(trans, root, parent_start,
+						  root->root_key.objectid,
+						  trans->transid, level, &ins);
+		BUG_ON(ret);
+		cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
+					    buf->len);
+	} else {
+		cow = btrfs_alloc_free_block(trans, root, buf->len,
+					     parent_start,
+					     root->root_key.objectid,
+					     trans->transid, level,
+					     search_start, empty_size);
+	}
+	if (IS_ERR(cow))
+		return PTR_ERR(cow);
+
+	copy_extent_buffer(cow, buf, 0, 0, cow->len);
+	btrfs_set_header_bytenr(cow, cow->start);
+	btrfs_set_header_generation(cow, trans->transid);
+	btrfs_set_header_owner(cow, root->root_key.objectid);
+	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
+
+	write_extent_buffer(cow, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(cow),
+			    BTRFS_FSID_SIZE);
+
+	WARN_ON(btrfs_header_generation(buf) > trans->transid);
+	if (btrfs_header_generation(buf) != trans->transid) {
+		u32 nr_extents;
+		ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
+		if (ret)
+			return ret;
+
+		ret = btrfs_cache_ref(trans, root, buf, nr_extents);
+		WARN_ON(ret);
+	} else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
+		/*
+		 * There are only two places that can drop reference to
+		 * tree blocks owned by living reloc trees, one is here,
+		 * the other place is btrfs_drop_subtree. In both places,
+		 * we check reference count while tree block is locked.
+		 * Furthermore, if reference count is one, it won't get
+		 * increased by someone else.
+		 */
+		u32 refs;
+		ret = btrfs_lookup_extent_ref(trans, root, buf->start,
+					      buf->len, &refs);
+		BUG_ON(ret);
+		if (refs == 1) {
+			ret = btrfs_update_ref(trans, root, buf, cow,
+					       0, nritems);
+			clean_tree_block(trans, root, buf);
+		} else {
+			ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
+		}
+		BUG_ON(ret);
+	} else {
+		ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
+		if (ret)
+			return ret;
+		clean_tree_block(trans, root, buf);
+	}
+
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
+		ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
+		WARN_ON(ret);
+	}
+
+	if (buf == root->node) {
+		WARN_ON(parent && parent != buf);
+
+		spin_lock(&root->node_lock);
+		root->node = cow;
+		extent_buffer_get(cow);
+		spin_unlock(&root->node_lock);
+
+		if (buf != root->commit_root) {
+			btrfs_free_extent(trans, root, buf->start,
+					  buf->len, buf->start,
+					  root->root_key.objectid,
+					  btrfs_header_generation(buf),
+					  level, 1);
+		}
+		free_extent_buffer(buf);
+		add_root_to_dirty_list(root);
+	} else {
+		btrfs_set_node_blockptr(parent, parent_slot,
+					cow->start);
+		WARN_ON(trans->transid == 0);
+		btrfs_set_node_ptr_generation(parent, parent_slot,
+					      trans->transid);
+		btrfs_mark_buffer_dirty(parent);
+		WARN_ON(btrfs_header_generation(parent) != trans->transid);
+		btrfs_free_extent(trans, root, buf->start, buf->len,
+				  parent_start, btrfs_header_owner(parent),
+				  btrfs_header_generation(parent), level, 1);
+	}
+	if (unlock_orig)
+		btrfs_tree_unlock(buf);
+	free_extent_buffer(buf);
+	btrfs_mark_buffer_dirty(cow);
+	*cow_ret = cow;
+	return 0;
+}
+
+/*
+ * cows a single block, see __btrfs_cow_block for the real work.
+ * This version of it has extra checks so that a block isn't cow'd more than
+ * once per transaction, as long as it hasn't been written yet
+ */
+noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, struct extent_buffer *buf,
+		    struct extent_buffer *parent, int parent_slot,
+		    struct extent_buffer **cow_ret, u64 prealloc_dest)
+{
+	u64 search_start;
+	int ret;
+
+	if (trans->transaction != root->fs_info->running_transaction) {
+		printk(KERN_CRIT "trans %llu running %llu\n",
+		       (unsigned long long)trans->transid,
+		       (unsigned long long)
+		       root->fs_info->running_transaction->transid);
+		WARN_ON(1);
+	}
+	if (trans->transid != root->fs_info->generation) {
+		printk(KERN_CRIT "trans %llu running %llu\n",
+		       (unsigned long long)trans->transid,
+		       (unsigned long long)root->fs_info->generation);
+		WARN_ON(1);
+	}
+
+	spin_lock(&root->fs_info->hash_lock);
+	if (btrfs_header_generation(buf) == trans->transid &&
+	    btrfs_header_owner(buf) == root->root_key.objectid &&
+	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
+		*cow_ret = buf;
+		spin_unlock(&root->fs_info->hash_lock);
+		WARN_ON(prealloc_dest);
+		return 0;
+	}
+	spin_unlock(&root->fs_info->hash_lock);
+	search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
+	ret = __btrfs_cow_block(trans, root, buf, parent,
+				 parent_slot, cow_ret, search_start, 0,
+				 prealloc_dest);
+	return ret;
+}
+
+/*
+ * helper function for defrag to decide if two blocks pointed to by a
+ * node are actually close by
+ */
+static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
+{
+	if (blocknr < other && other - (blocknr + blocksize) < 32768)
+		return 1;
+	if (blocknr > other && blocknr - (other + blocksize) < 32768)
+		return 1;
+	return 0;
+}
+
+/*
+ * compare two keys in a memcmp fashion
+ */
+static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
+{
+	struct btrfs_key k1;
+
+	btrfs_disk_key_to_cpu(&k1, disk);
+
+	if (k1.objectid > k2->objectid)
+		return 1;
+	if (k1.objectid < k2->objectid)
+		return -1;
+	if (k1.type > k2->type)
+		return 1;
+	if (k1.type < k2->type)
+		return -1;
+	if (k1.offset > k2->offset)
+		return 1;
+	if (k1.offset < k2->offset)
+		return -1;
+	return 0;
+}
+
+/*
+ * same as comp_keys only with two btrfs_key's
+ */
+static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
+{
+	if (k1->objectid > k2->objectid)
+		return 1;
+	if (k1->objectid < k2->objectid)
+		return -1;
+	if (k1->type > k2->type)
+		return 1;
+	if (k1->type < k2->type)
+		return -1;
+	if (k1->offset > k2->offset)
+		return 1;
+	if (k1->offset < k2->offset)
+		return -1;
+	return 0;
+}
+
+/*
+ * The leaf data grows from end-to-front in the node.
+ * this returns the address of the start of the last item,
+ * which is the stop of the leaf data stack
+ */
+static inline unsigned int leaf_data_end(struct btrfs_root *root,
+					 struct extent_buffer *leaf)
+{
+	u32 nr = btrfs_header_nritems(leaf);
+	if (nr == 0)
+		return BTRFS_LEAF_DATA_SIZE(root);
+	return btrfs_item_offset_nr(leaf, nr - 1);
+}
+
+/*
+ * extra debugging checks to make sure all the items in a key are
+ * well formed and in the proper order
+ */
+static int check_node(struct btrfs_root *root, struct btrfs_path *path,
+		      int level)
+{
+	struct extent_buffer *parent = NULL;
+	struct extent_buffer *node = path->nodes[level];
+	struct btrfs_disk_key parent_key;
+	struct btrfs_disk_key node_key;
+	int parent_slot;
+	int slot;
+	struct btrfs_key cpukey;
+	u32 nritems = btrfs_header_nritems(node);
+
+	if (path->nodes[level + 1])
+		parent = path->nodes[level + 1];
+
+	slot = path->slots[level];
+	BUG_ON(nritems == 0);
+	if (parent) {
+		parent_slot = path->slots[level + 1];
+		btrfs_node_key(parent, &parent_key, parent_slot);
+		btrfs_node_key(node, &node_key, 0);
+		BUG_ON(memcmp(&parent_key, &node_key,
+			      sizeof(struct btrfs_disk_key)));
+		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
+		       btrfs_header_bytenr(node));
+	}
+	BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
+	if (slot != 0) {
+		btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
+		btrfs_node_key(node, &node_key, slot);
+		BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
+	}
+	if (slot < nritems - 1) {
+		btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
+		btrfs_node_key(node, &node_key, slot);
+		BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
+	}
+	return 0;
+}
+
+/*
+ * extra checking to make sure all the items in a leaf are
+ * well formed and in the proper order
+ */
+static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
+		      int level)
+{
+	struct extent_buffer *leaf = path->nodes[level];
+	struct extent_buffer *parent = NULL;
+	int parent_slot;
+	struct btrfs_key cpukey;
+	struct btrfs_disk_key parent_key;
+	struct btrfs_disk_key leaf_key;
+	int slot = path->slots[0];
+
+	u32 nritems = btrfs_header_nritems(leaf);
+
+	if (path->nodes[level + 1])
+		parent = path->nodes[level + 1];
+
+	if (nritems == 0)
+		return 0;
+
+	if (parent) {
+		parent_slot = path->slots[level + 1];
+		btrfs_node_key(parent, &parent_key, parent_slot);
+		btrfs_item_key(leaf, &leaf_key, 0);
+
+		BUG_ON(memcmp(&parent_key, &leaf_key,
+		       sizeof(struct btrfs_disk_key)));
+		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
+		       btrfs_header_bytenr(leaf));
+	}
+	if (slot != 0 && slot < nritems - 1) {
+		btrfs_item_key(leaf, &leaf_key, slot);
+		btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
+		if (comp_keys(&leaf_key, &cpukey) <= 0) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d offset bad key\n", slot);
+			BUG_ON(1);
+		}
+		if (btrfs_item_offset_nr(leaf, slot - 1) !=
+		       btrfs_item_end_nr(leaf, slot)) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d offset bad\n", slot);
+			BUG_ON(1);
+		}
+	}
+	if (slot < nritems - 1) {
+		btrfs_item_key(leaf, &leaf_key, slot);
+		btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
+		BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
+		if (btrfs_item_offset_nr(leaf, slot) !=
+			btrfs_item_end_nr(leaf, slot + 1)) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d offset bad\n", slot);
+			BUG_ON(1);
+		}
+	}
+	BUG_ON(btrfs_item_offset_nr(leaf, 0) +
+	       btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
+	return 0;
+}
+
+static noinline int check_block(struct btrfs_root *root,
+				struct btrfs_path *path, int level)
+{
+	return 0;
+	if (level == 0)
+		return check_leaf(root, path, level);
+	return check_node(root, path, level);
+}
+
+/*
+ * search for key in the extent_buffer.  The items start at offset p,
+ * and they are item_size apart.  There are 'max' items in p.
+ *
+ * the slot in the array is returned via slot, and it points to
+ * the place where you would insert key if it is not found in
+ * the array.
+ *
+ * slot may point to max if the key is bigger than all of the keys
+ */
+static noinline int generic_bin_search(struct extent_buffer *eb,
+				       unsigned long p,
+				       int item_size, struct btrfs_key *key,
+				       int max, int *slot)
+{
+	int low = 0;
+	int high = max;
+	int mid;
+	int ret;
+	struct btrfs_disk_key *tmp = NULL;
+	struct btrfs_disk_key unaligned;
+	unsigned long offset;
+	char *map_token = NULL;
+	char *kaddr = NULL;
+	unsigned long map_start = 0;
+	unsigned long map_len = 0;
+	int err;
+
+	while (low < high) {
+		mid = (low + high) / 2;
+		offset = p + mid * item_size;
+
+		if (!map_token || offset < map_start ||
+		    (offset + sizeof(struct btrfs_disk_key)) >
+		    map_start + map_len) {
+			if (map_token) {
+				unmap_extent_buffer(eb, map_token, KM_USER0);
+				map_token = NULL;
+			}
+
+			err = map_private_extent_buffer(eb, offset,
+						sizeof(struct btrfs_disk_key),
+						&map_token, &kaddr,
+						&map_start, &map_len, KM_USER0);
+
+			if (!err) {
+				tmp = (struct btrfs_disk_key *)(kaddr + offset -
+							map_start);
+			} else {
+				read_extent_buffer(eb, &unaligned,
+						   offset, sizeof(unaligned));
+				tmp = &unaligned;
+			}
+
+		} else {
+			tmp = (struct btrfs_disk_key *)(kaddr + offset -
+							map_start);
+		}
+		ret = comp_keys(tmp, key);
+
+		if (ret < 0)
+			low = mid + 1;
+		else if (ret > 0)
+			high = mid;
+		else {
+			*slot = mid;
+			if (map_token)
+				unmap_extent_buffer(eb, map_token, KM_USER0);
+			return 0;
+		}
+	}
+	*slot = low;
+	if (map_token)
+		unmap_extent_buffer(eb, map_token, KM_USER0);
+	return 1;
+}
+
+/*
+ * simple bin_search frontend that does the right thing for
+ * leaves vs nodes
+ */
+static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
+		      int level, int *slot)
+{
+	if (level == 0) {
+		return generic_bin_search(eb,
+					  offsetof(struct btrfs_leaf, items),
+					  sizeof(struct btrfs_item),
+					  key, btrfs_header_nritems(eb),
+					  slot);
+	} else {
+		return generic_bin_search(eb,
+					  offsetof(struct btrfs_node, ptrs),
+					  sizeof(struct btrfs_key_ptr),
+					  key, btrfs_header_nritems(eb),
+					  slot);
+	}
+	return -1;
+}
+
+/* given a node and slot number, this reads the blocks it points to.  The
+ * extent buffer is returned with a reference taken (but unlocked).
+ * NULL is returned on error.
+ */
+static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
+				   struct extent_buffer *parent, int slot)
+{
+	int level = btrfs_header_level(parent);
+	if (slot < 0)
+		return NULL;
+	if (slot >= btrfs_header_nritems(parent))
+		return NULL;
+
+	BUG_ON(level == 0);
+
+	return read_tree_block(root, btrfs_node_blockptr(parent, slot),
+		       btrfs_level_size(root, level - 1),
+		       btrfs_node_ptr_generation(parent, slot));
+}
+
+/*
+ * node level balancing, used to make sure nodes are in proper order for
+ * item deletion.  We balance from the top down, so we have to make sure
+ * that a deletion won't leave an node completely empty later on.
+ */
+static noinline int balance_level(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 struct btrfs_path *path, int level)
+{
+	struct extent_buffer *right = NULL;
+	struct extent_buffer *mid;
+	struct extent_buffer *left = NULL;
+	struct extent_buffer *parent = NULL;
+	int ret = 0;
+	int wret;
+	int pslot;
+	int orig_slot = path->slots[level];
+	int err_on_enospc = 0;
+	u64 orig_ptr;
+
+	if (level == 0)
+		return 0;
+
+	mid = path->nodes[level];
+	WARN_ON(!path->locks[level]);
+	WARN_ON(btrfs_header_generation(mid) != trans->transid);
+
+	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
+
+	if (level < BTRFS_MAX_LEVEL - 1)
+		parent = path->nodes[level + 1];
+	pslot = path->slots[level + 1];
+
+	/*
+	 * deal with the case where there is only one pointer in the root
+	 * by promoting the node below to a root
+	 */
+	if (!parent) {
+		struct extent_buffer *child;
+
+		if (btrfs_header_nritems(mid) != 1)
+			return 0;
+
+		/* promote the child to a root */
+		child = read_node_slot(root, mid, 0);
+		btrfs_tree_lock(child);
+		BUG_ON(!child);
+		ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
+		BUG_ON(ret);
+
+		spin_lock(&root->node_lock);
+		root->node = child;
+		spin_unlock(&root->node_lock);
+
+		ret = btrfs_update_extent_ref(trans, root, child->start,
+					      mid->start, child->start,
+					      root->root_key.objectid,
+					      trans->transid, level - 1);
+		BUG_ON(ret);
+
+		add_root_to_dirty_list(root);
+		btrfs_tree_unlock(child);
+		path->locks[level] = 0;
+		path->nodes[level] = NULL;
+		clean_tree_block(trans, root, mid);
+		btrfs_tree_unlock(mid);
+		/* once for the path */
+		free_extent_buffer(mid);
+		ret = btrfs_free_extent(trans, root, mid->start, mid->len,
+					mid->start, root->root_key.objectid,
+					btrfs_header_generation(mid),
+					level, 1);
+		/* once for the root ptr */
+		free_extent_buffer(mid);
+		return ret;
+	}
+	if (btrfs_header_nritems(mid) >
+	    BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
+		return 0;
+
+	if (btrfs_header_nritems(mid) < 2)
+		err_on_enospc = 1;
+
+	left = read_node_slot(root, parent, pslot - 1);
+	if (left) {
+		btrfs_tree_lock(left);
+		wret = btrfs_cow_block(trans, root, left,
+				       parent, pslot - 1, &left, 0);
+		if (wret) {
+			ret = wret;
+			goto enospc;
+		}
+	}
+	right = read_node_slot(root, parent, pslot + 1);
+	if (right) {
+		btrfs_tree_lock(right);
+		wret = btrfs_cow_block(trans, root, right,
+				       parent, pslot + 1, &right, 0);
+		if (wret) {
+			ret = wret;
+			goto enospc;
+		}
+	}
+
+	/* first, try to make some room in the middle buffer */
+	if (left) {
+		orig_slot += btrfs_header_nritems(left);
+		wret = push_node_left(trans, root, left, mid, 1);
+		if (wret < 0)
+			ret = wret;
+		if (btrfs_header_nritems(mid) < 2)
+			err_on_enospc = 1;
+	}
+
+	/*
+	 * then try to empty the right most buffer into the middle
+	 */
+	if (right) {
+		wret = push_node_left(trans, root, mid, right, 1);
+		if (wret < 0 && wret != -ENOSPC)
+			ret = wret;
+		if (btrfs_header_nritems(right) == 0) {
+			u64 bytenr = right->start;
+			u64 generation = btrfs_header_generation(parent);
+			u32 blocksize = right->len;
+
+			clean_tree_block(trans, root, right);
+			btrfs_tree_unlock(right);
+			free_extent_buffer(right);
+			right = NULL;
+			wret = del_ptr(trans, root, path, level + 1, pslot +
+				       1);
+			if (wret)
+				ret = wret;
+			wret = btrfs_free_extent(trans, root, bytenr,
+						 blocksize, parent->start,
+						 btrfs_header_owner(parent),
+						 generation, level, 1);
+			if (wret)
+				ret = wret;
+		} else {
+			struct btrfs_disk_key right_key;
+			btrfs_node_key(right, &right_key, 0);
+			btrfs_set_node_key(parent, &right_key, pslot + 1);
+			btrfs_mark_buffer_dirty(parent);
+		}
+	}
+	if (btrfs_header_nritems(mid) == 1) {
+		/*
+		 * we're not allowed to leave a node with one item in the
+		 * tree during a delete.  A deletion from lower in the tree
+		 * could try to delete the only pointer in this node.
+		 * So, pull some keys from the left.
+		 * There has to be a left pointer at this point because
+		 * otherwise we would have pulled some pointers from the
+		 * right
+		 */
+		BUG_ON(!left);
+		wret = balance_node_right(trans, root, mid, left);
+		if (wret < 0) {
+			ret = wret;
+			goto enospc;
+		}
+		if (wret == 1) {
+			wret = push_node_left(trans, root, left, mid, 1);
+			if (wret < 0)
+				ret = wret;
+		}
+		BUG_ON(wret == 1);
+	}
+	if (btrfs_header_nritems(mid) == 0) {
+		/* we've managed to empty the middle node, drop it */
+		u64 root_gen = btrfs_header_generation(parent);
+		u64 bytenr = mid->start;
+		u32 blocksize = mid->len;
+
+		clean_tree_block(trans, root, mid);
+		btrfs_tree_unlock(mid);
+		free_extent_buffer(mid);
+		mid = NULL;
+		wret = del_ptr(trans, root, path, level + 1, pslot);
+		if (wret)
+			ret = wret;
+		wret = btrfs_free_extent(trans, root, bytenr, blocksize,
+					 parent->start,
+					 btrfs_header_owner(parent),
+					 root_gen, level, 1);
+		if (wret)
+			ret = wret;
+	} else {
+		/* update the parent key to reflect our changes */
+		struct btrfs_disk_key mid_key;
+		btrfs_node_key(mid, &mid_key, 0);
+		btrfs_set_node_key(parent, &mid_key, pslot);
+		btrfs_mark_buffer_dirty(parent);
+	}
+
+	/* update the path */
+	if (left) {
+		if (btrfs_header_nritems(left) > orig_slot) {
+			extent_buffer_get(left);
+			/* left was locked after cow */
+			path->nodes[level] = left;
+			path->slots[level + 1] -= 1;
+			path->slots[level] = orig_slot;
+			if (mid) {
+				btrfs_tree_unlock(mid);
+				free_extent_buffer(mid);
+			}
+		} else {
+			orig_slot -= btrfs_header_nritems(left);
+			path->slots[level] = orig_slot;
+		}
+	}
+	/* double check we haven't messed things up */
+	check_block(root, path, level);
+	if (orig_ptr !=
+	    btrfs_node_blockptr(path->nodes[level], path->slots[level]))
+		BUG();
+enospc:
+	if (right) {
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+	if (left) {
+		if (path->nodes[level] != left)
+			btrfs_tree_unlock(left);
+		free_extent_buffer(left);
+	}
+	return ret;
+}
+
+/* Node balancing for insertion.  Here we only split or push nodes around
+ * when they are completely full.  This is also done top down, so we
+ * have to be pessimistic.
+ */
+static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path, int level)
+{
+	struct extent_buffer *right = NULL;
+	struct extent_buffer *mid;
+	struct extent_buffer *left = NULL;
+	struct extent_buffer *parent = NULL;
+	int ret = 0;
+	int wret;
+	int pslot;
+	int orig_slot = path->slots[level];
+	u64 orig_ptr;
+
+	if (level == 0)
+		return 1;
+
+	mid = path->nodes[level];
+	WARN_ON(btrfs_header_generation(mid) != trans->transid);
+	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
+
+	if (level < BTRFS_MAX_LEVEL - 1)
+		parent = path->nodes[level + 1];
+	pslot = path->slots[level + 1];
+
+	if (!parent)
+		return 1;
+
+	left = read_node_slot(root, parent, pslot - 1);
+
+	/* first, try to make some room in the middle buffer */
+	if (left) {
+		u32 left_nr;
+
+		btrfs_tree_lock(left);
+		left_nr = btrfs_header_nritems(left);
+		if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
+			wret = 1;
+		} else {
+			ret = btrfs_cow_block(trans, root, left, parent,
+					      pslot - 1, &left, 0);
+			if (ret)
+				wret = 1;
+			else {
+				wret = push_node_left(trans, root,
+						      left, mid, 0);
+			}
+		}
+		if (wret < 0)
+			ret = wret;
+		if (wret == 0) {
+			struct btrfs_disk_key disk_key;
+			orig_slot += left_nr;
+			btrfs_node_key(mid, &disk_key, 0);
+			btrfs_set_node_key(parent, &disk_key, pslot);
+			btrfs_mark_buffer_dirty(parent);
+			if (btrfs_header_nritems(left) > orig_slot) {
+				path->nodes[level] = left;
+				path->slots[level + 1] -= 1;
+				path->slots[level] = orig_slot;
+				btrfs_tree_unlock(mid);
+				free_extent_buffer(mid);
+			} else {
+				orig_slot -=
+					btrfs_header_nritems(left);
+				path->slots[level] = orig_slot;
+				btrfs_tree_unlock(left);
+				free_extent_buffer(left);
+			}
+			return 0;
+		}
+		btrfs_tree_unlock(left);
+		free_extent_buffer(left);
+	}
+	right = read_node_slot(root, parent, pslot + 1);
+
+	/*
+	 * then try to empty the right most buffer into the middle
+	 */
+	if (right) {
+		u32 right_nr;
+		btrfs_tree_lock(right);
+		right_nr = btrfs_header_nritems(right);
+		if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
+			wret = 1;
+		} else {
+			ret = btrfs_cow_block(trans, root, right,
+					      parent, pslot + 1,
+					      &right, 0);
+			if (ret)
+				wret = 1;
+			else {
+				wret = balance_node_right(trans, root,
+							  right, mid);
+			}
+		}
+		if (wret < 0)
+			ret = wret;
+		if (wret == 0) {
+			struct btrfs_disk_key disk_key;
+
+			btrfs_node_key(right, &disk_key, 0);
+			btrfs_set_node_key(parent, &disk_key, pslot + 1);
+			btrfs_mark_buffer_dirty(parent);
+
+			if (btrfs_header_nritems(mid) <= orig_slot) {
+				path->nodes[level] = right;
+				path->slots[level + 1] += 1;
+				path->slots[level] = orig_slot -
+					btrfs_header_nritems(mid);
+				btrfs_tree_unlock(mid);
+				free_extent_buffer(mid);
+			} else {
+				btrfs_tree_unlock(right);
+				free_extent_buffer(right);
+			}
+			return 0;
+		}
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+	return 1;
+}
+
+/*
+ * readahead one full node of leaves, finding things that are close
+ * to the block in 'slot', and triggering ra on them.
+ */
+static noinline void reada_for_search(struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      int level, int slot, u64 objectid)
+{
+	struct extent_buffer *node;
+	struct btrfs_disk_key disk_key;
+	u32 nritems;
+	u64 search;
+	u64 lowest_read;
+	u64 highest_read;
+	u64 nread = 0;
+	int direction = path->reada;
+	struct extent_buffer *eb;
+	u32 nr;
+	u32 blocksize;
+	u32 nscan = 0;
+
+	if (level != 1)
+		return;
+
+	if (!path->nodes[level])
+		return;
+
+	node = path->nodes[level];
+
+	search = btrfs_node_blockptr(node, slot);
+	blocksize = btrfs_level_size(root, level - 1);
+	eb = btrfs_find_tree_block(root, search, blocksize);
+	if (eb) {
+		free_extent_buffer(eb);
+		return;
+	}
+
+	highest_read = search;
+	lowest_read = search;
+
+	nritems = btrfs_header_nritems(node);
+	nr = slot;
+	while (1) {
+		if (direction < 0) {
+			if (nr == 0)
+				break;
+			nr--;
+		} else if (direction > 0) {
+			nr++;
+			if (nr >= nritems)
+				break;
+		}
+		if (path->reada < 0 && objectid) {
+			btrfs_node_key(node, &disk_key, nr);
+			if (btrfs_disk_key_objectid(&disk_key) != objectid)
+				break;
+		}
+		search = btrfs_node_blockptr(node, nr);
+		if ((search >= lowest_read && search <= highest_read) ||
+		    (search < lowest_read && lowest_read - search <= 16384) ||
+		    (search > highest_read && search - highest_read <= 16384)) {
+			readahead_tree_block(root, search, blocksize,
+				     btrfs_node_ptr_generation(node, nr));
+			nread += blocksize;
+		}
+		nscan++;
+		if (path->reada < 2 && (nread > (64 * 1024) || nscan > 32))
+			break;
+
+		if (nread > (256 * 1024) || nscan > 128)
+			break;
+
+		if (search < lowest_read)
+			lowest_read = search;
+		if (search > highest_read)
+			highest_read = search;
+	}
+}
+
+/*
+ * when we walk down the tree, it is usually safe to unlock the higher layers
+ * in the tree.  The exceptions are when our path goes through slot 0, because
+ * operations on the tree might require changing key pointers higher up in the
+ * tree.
+ *
+ * callers might also have set path->keep_locks, which tells this code to keep
+ * the lock if the path points to the last slot in the block.  This is part of
+ * walking through the tree, and selecting the next slot in the higher block.
+ *
+ * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
+ * if lowest_unlock is 1, level 0 won't be unlocked
+ */
+static noinline void unlock_up(struct btrfs_path *path, int level,
+			       int lowest_unlock)
+{
+	int i;
+	int skip_level = level;
+	int no_skips = 0;
+	struct extent_buffer *t;
+
+	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
+		if (!path->nodes[i])
+			break;
+		if (!path->locks[i])
+			break;
+		if (!no_skips && path->slots[i] == 0) {
+			skip_level = i + 1;
+			continue;
+		}
+		if (!no_skips && path->keep_locks) {
+			u32 nritems;
+			t = path->nodes[i];
+			nritems = btrfs_header_nritems(t);
+			if (nritems < 1 || path->slots[i] >= nritems - 1) {
+				skip_level = i + 1;
+				continue;
+			}
+		}
+		if (skip_level < i && i >= lowest_unlock)
+			no_skips = 1;
+
+		t = path->nodes[i];
+		if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
+			btrfs_tree_unlock(t);
+			path->locks[i] = 0;
+		}
+	}
+}
+
+/*
+ * look for key in the tree.  path is filled in with nodes along the way
+ * if key is found, we return zero and you can find the item in the leaf
+ * level of the path (level 0)
+ *
+ * If the key isn't found, the path points to the slot where it should
+ * be inserted, and 1 is returned.  If there are other errors during the
+ * search a negative error number is returned.
+ *
+ * if ins_len > 0, nodes and leaves will be split as we walk down the
+ * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
+ * possible)
+ */
+int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, struct btrfs_path *p, int
+		      ins_len, int cow)
+{
+	struct extent_buffer *b;
+	struct extent_buffer *tmp;
+	int slot;
+	int ret;
+	int level;
+	int should_reada = p->reada;
+	int lowest_unlock = 1;
+	int blocksize;
+	u8 lowest_level = 0;
+	u64 blocknr;
+	u64 gen;
+	struct btrfs_key prealloc_block;
+
+	lowest_level = p->lowest_level;
+	WARN_ON(lowest_level && ins_len > 0);
+	WARN_ON(p->nodes[0] != NULL);
+
+	if (ins_len < 0)
+		lowest_unlock = 2;
+
+	prealloc_block.objectid = 0;
+
+again:
+	if (p->skip_locking)
+		b = btrfs_root_node(root);
+	else
+		b = btrfs_lock_root_node(root);
+
+	while (b) {
+		level = btrfs_header_level(b);
+
+		/*
+		 * setup the path here so we can release it under lock
+		 * contention with the cow code
+		 */
+		p->nodes[level] = b;
+		if (!p->skip_locking)
+			p->locks[level] = 1;
+
+		if (cow) {
+			int wret;
+
+			/* is a cow on this block not required */
+			spin_lock(&root->fs_info->hash_lock);
+			if (btrfs_header_generation(b) == trans->transid &&
+			    btrfs_header_owner(b) == root->root_key.objectid &&
+			    !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
+				spin_unlock(&root->fs_info->hash_lock);
+				goto cow_done;
+			}
+			spin_unlock(&root->fs_info->hash_lock);
+
+			/* ok, we have to cow, is our old prealloc the right
+			 * size?
+			 */
+			if (prealloc_block.objectid &&
+			    prealloc_block.offset != b->len) {
+				btrfs_free_reserved_extent(root,
+					   prealloc_block.objectid,
+					   prealloc_block.offset);
+				prealloc_block.objectid = 0;
+			}
+
+			/*
+			 * for higher level blocks, try not to allocate blocks
+			 * with the block and the parent locks held.
+			 */
+			if (level > 1 && !prealloc_block.objectid &&
+			    btrfs_path_lock_waiting(p, level)) {
+				u32 size = b->len;
+				u64 hint = b->start;
+
+				btrfs_release_path(root, p);
+				ret = btrfs_reserve_extent(trans, root,
+							   size, size, 0,
+							   hint, (u64)-1,
+							   &prealloc_block, 0);
+				BUG_ON(ret);
+				goto again;
+			}
+
+			wret = btrfs_cow_block(trans, root, b,
+					       p->nodes[level + 1],
+					       p->slots[level + 1],
+					       &b, prealloc_block.objectid);
+			prealloc_block.objectid = 0;
+			if (wret) {
+				free_extent_buffer(b);
+				ret = wret;
+				goto done;
+			}
+		}
+cow_done:
+		BUG_ON(!cow && ins_len);
+		if (level != btrfs_header_level(b))
+			WARN_ON(1);
+		level = btrfs_header_level(b);
+
+		p->nodes[level] = b;
+		if (!p->skip_locking)
+			p->locks[level] = 1;
+
+		ret = check_block(root, p, level);
+		if (ret) {
+			ret = -1;
+			goto done;
+		}
+
+		ret = bin_search(b, key, level, &slot);
+		if (level != 0) {
+			if (ret && slot > 0)
+				slot -= 1;
+			p->slots[level] = slot;
+			if ((p->search_for_split || ins_len > 0) &&
+			    btrfs_header_nritems(b) >=
+			    BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
+				int sret = split_node(trans, root, p, level);
+				BUG_ON(sret > 0);
+				if (sret) {
+					ret = sret;
+					goto done;
+				}
+				b = p->nodes[level];
+				slot = p->slots[level];
+			} else if (ins_len < 0) {
+				int sret = balance_level(trans, root, p,
+							 level);
+				if (sret) {
+					ret = sret;
+					goto done;
+				}
+				b = p->nodes[level];
+				if (!b) {
+					btrfs_release_path(NULL, p);
+					goto again;
+				}
+				slot = p->slots[level];
+				BUG_ON(btrfs_header_nritems(b) == 1);
+			}
+			unlock_up(p, level, lowest_unlock);
+
+			/* this is only true while dropping a snapshot */
+			if (level == lowest_level) {
+				ret = 0;
+				goto done;
+			}
+
+			blocknr = btrfs_node_blockptr(b, slot);
+			gen = btrfs_node_ptr_generation(b, slot);
+			blocksize = btrfs_level_size(root, level - 1);
+
+			tmp = btrfs_find_tree_block(root, blocknr, blocksize);
+			if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
+				b = tmp;
+			} else {
+				/*
+				 * reduce lock contention at high levels
+				 * of the btree by dropping locks before
+				 * we read.
+				 */
+				if (level > 1) {
+					btrfs_release_path(NULL, p);
+					if (tmp)
+						free_extent_buffer(tmp);
+					if (should_reada)
+						reada_for_search(root, p,
+								 level, slot,
+								 key->objectid);
+
+					tmp = read_tree_block(root, blocknr,
+							 blocksize, gen);
+					if (tmp)
+						free_extent_buffer(tmp);
+					goto again;
+				} else {
+					if (tmp)
+						free_extent_buffer(tmp);
+					if (should_reada)
+						reada_for_search(root, p,
+								 level, slot,
+								 key->objectid);
+					b = read_node_slot(root, b, slot);
+				}
+			}
+			if (!p->skip_locking)
+				btrfs_tree_lock(b);
+		} else {
+			p->slots[level] = slot;
+			if (ins_len > 0 &&
+			    btrfs_leaf_free_space(root, b) < ins_len) {
+				int sret = split_leaf(trans, root, key,
+						      p, ins_len, ret == 0);
+				BUG_ON(sret > 0);
+				if (sret) {
+					ret = sret;
+					goto done;
+				}
+			}
+			if (!p->search_for_split)
+				unlock_up(p, level, lowest_unlock);
+			goto done;
+		}
+	}
+	ret = 1;
+done:
+	if (prealloc_block.objectid) {
+		btrfs_free_reserved_extent(root,
+			   prealloc_block.objectid,
+			   prealloc_block.offset);
+	}
+
+	return ret;
+}
+
+/*
+ * adjust the pointers going up the tree, starting at level
+ * making sure the right key of each node is points to 'key'.
+ * This is used after shifting pointers to the left, so it stops
+ * fixing up pointers when a given leaf/node is not in slot 0 of the
+ * higher levels
+ *
+ * If this fails to write a tree block, it returns -1, but continues
+ * fixing up the blocks in ram so the tree is consistent.
+ */
+static int fixup_low_keys(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct btrfs_path *path,
+			  struct btrfs_disk_key *key, int level)
+{
+	int i;
+	int ret = 0;
+	struct extent_buffer *t;
+
+	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
+		int tslot = path->slots[i];
+		if (!path->nodes[i])
+			break;
+		t = path->nodes[i];
+		btrfs_set_node_key(t, key, tslot);
+		btrfs_mark_buffer_dirty(path->nodes[i]);
+		if (tslot != 0)
+			break;
+	}
+	return ret;
+}
+
+
+/*
+ * try to push data from one node into the next node left in the
+ * tree.
+ *
+ * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
+ * error, and > 0 if there was no room in the left hand block.
+ */
+static int push_node_left(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct extent_buffer *dst,
+			  struct extent_buffer *src, int empty)
+{
+	int push_items = 0;
+	int src_nritems;
+	int dst_nritems;
+	int ret = 0;
+
+	src_nritems = btrfs_header_nritems(src);
+	dst_nritems = btrfs_header_nritems(dst);
+	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
+	WARN_ON(btrfs_header_generation(src) != trans->transid);
+	WARN_ON(btrfs_header_generation(dst) != trans->transid);
+
+	if (!empty && src_nritems <= 8)
+		return 1;
+
+	if (push_items <= 0)
+		return 1;
+
+	if (empty) {
+		push_items = min(src_nritems, push_items);
+		if (push_items < src_nritems) {
+			/* leave at least 8 pointers in the node if
+			 * we aren't going to empty it
+			 */
+			if (src_nritems - push_items < 8) {
+				if (push_items <= 8)
+					return 1;
+				push_items -= 8;
+			}
+		}
+	} else
+		push_items = min(src_nritems - 8, push_items);
+
+	copy_extent_buffer(dst, src,
+			   btrfs_node_key_ptr_offset(dst_nritems),
+			   btrfs_node_key_ptr_offset(0),
+			   push_items * sizeof(struct btrfs_key_ptr));
+
+	if (push_items < src_nritems) {
+		memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
+				      btrfs_node_key_ptr_offset(push_items),
+				      (src_nritems - push_items) *
+				      sizeof(struct btrfs_key_ptr));
+	}
+	btrfs_set_header_nritems(src, src_nritems - push_items);
+	btrfs_set_header_nritems(dst, dst_nritems + push_items);
+	btrfs_mark_buffer_dirty(src);
+	btrfs_mark_buffer_dirty(dst);
+
+	ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
+	BUG_ON(ret);
+
+	return ret;
+}
+
+/*
+ * try to push data from one node into the next node right in the
+ * tree.
+ *
+ * returns 0 if some ptrs were pushed, < 0 if there was some horrible
+ * error, and > 0 if there was no room in the right hand block.
+ *
+ * this will  only push up to 1/2 the contents of the left node over
+ */
+static int balance_node_right(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct extent_buffer *dst,
+			      struct extent_buffer *src)
+{
+	int push_items = 0;
+	int max_push;
+	int src_nritems;
+	int dst_nritems;
+	int ret = 0;
+
+	WARN_ON(btrfs_header_generation(src) != trans->transid);
+	WARN_ON(btrfs_header_generation(dst) != trans->transid);
+
+	src_nritems = btrfs_header_nritems(src);
+	dst_nritems = btrfs_header_nritems(dst);
+	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
+	if (push_items <= 0)
+		return 1;
+
+	if (src_nritems < 4)
+		return 1;
+
+	max_push = src_nritems / 2 + 1;
+	/* don't try to empty the node */
+	if (max_push >= src_nritems)
+		return 1;
+
+	if (max_push < push_items)
+		push_items = max_push;
+
+	memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
+				      btrfs_node_key_ptr_offset(0),
+				      (dst_nritems) *
+				      sizeof(struct btrfs_key_ptr));
+
+	copy_extent_buffer(dst, src,
+			   btrfs_node_key_ptr_offset(0),
+			   btrfs_node_key_ptr_offset(src_nritems - push_items),
+			   push_items * sizeof(struct btrfs_key_ptr));
+
+	btrfs_set_header_nritems(src, src_nritems - push_items);
+	btrfs_set_header_nritems(dst, dst_nritems + push_items);
+
+	btrfs_mark_buffer_dirty(src);
+	btrfs_mark_buffer_dirty(dst);
+
+	ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
+	BUG_ON(ret);
+
+	return ret;
+}
+
+/*
+ * worker function to insert a single pointer in a node.
+ * the node should have enough room for the pointer already
+ *
+ * slot and level indicate where you want the key to go, and
+ * blocknr is the block the key points to.
+ *
+ * returns zero on success and < 0 on any error
+ */
+static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_path *path, struct btrfs_disk_key
+		      *key, u64 bytenr, int slot, int level)
+{
+	struct extent_buffer *lower;
+	int nritems;
+
+	BUG_ON(!path->nodes[level]);
+	lower = path->nodes[level];
+	nritems = btrfs_header_nritems(lower);
+	if (slot > nritems)
+		BUG();
+	if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
+		BUG();
+	if (slot != nritems) {
+		memmove_extent_buffer(lower,
+			      btrfs_node_key_ptr_offset(slot + 1),
+			      btrfs_node_key_ptr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_key_ptr));
+	}
+	btrfs_set_node_key(lower, key, slot);
+	btrfs_set_node_blockptr(lower, slot, bytenr);
+	WARN_ON(trans->transid == 0);
+	btrfs_set_node_ptr_generation(lower, slot, trans->transid);
+	btrfs_set_header_nritems(lower, nritems + 1);
+	btrfs_mark_buffer_dirty(lower);
+	return 0;
+}
+
+/*
+ * split the node at the specified level in path in two.
+ * The path is corrected to point to the appropriate node after the split
+ *
+ * Before splitting this tries to make some room in the node by pushing
+ * left and right, if either one works, it returns right away.
+ *
+ * returns 0 on success and < 0 on failure
+ */
+static noinline int split_node(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct btrfs_path *path, int level)
+{
+	struct extent_buffer *c;
+	struct extent_buffer *split;
+	struct btrfs_disk_key disk_key;
+	int mid;
+	int ret;
+	int wret;
+	u32 c_nritems;
+
+	c = path->nodes[level];
+	WARN_ON(btrfs_header_generation(c) != trans->transid);
+	if (c == root->node) {
+		/* trying to split the root, lets make a new one */
+		ret = insert_new_root(trans, root, path, level + 1);
+		if (ret)
+			return ret;
+	} else {
+		ret = push_nodes_for_insert(trans, root, path, level);
+		c = path->nodes[level];
+		if (!ret && btrfs_header_nritems(c) <
+		    BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
+			return 0;
+		if (ret < 0)
+			return ret;
+	}
+
+	c_nritems = btrfs_header_nritems(c);
+
+	split = btrfs_alloc_free_block(trans, root, root->nodesize,
+					path->nodes[level + 1]->start,
+					root->root_key.objectid,
+					trans->transid, level, c->start, 0);
+	if (IS_ERR(split))
+		return PTR_ERR(split);
+
+	btrfs_set_header_flags(split, btrfs_header_flags(c));
+	btrfs_set_header_level(split, btrfs_header_level(c));
+	btrfs_set_header_bytenr(split, split->start);
+	btrfs_set_header_generation(split, trans->transid);
+	btrfs_set_header_owner(split, root->root_key.objectid);
+	btrfs_set_header_flags(split, 0);
+	write_extent_buffer(split, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(split),
+			    BTRFS_FSID_SIZE);
+	write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
+			    (unsigned long)btrfs_header_chunk_tree_uuid(split),
+			    BTRFS_UUID_SIZE);
+
+	mid = (c_nritems + 1) / 2;
+
+	copy_extent_buffer(split, c,
+			   btrfs_node_key_ptr_offset(0),
+			   btrfs_node_key_ptr_offset(mid),
+			   (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
+	btrfs_set_header_nritems(split, c_nritems - mid);
+	btrfs_set_header_nritems(c, mid);
+	ret = 0;
+
+	btrfs_mark_buffer_dirty(c);
+	btrfs_mark_buffer_dirty(split);
+
+	btrfs_node_key(split, &disk_key, 0);
+	wret = insert_ptr(trans, root, path, &disk_key, split->start,
+			  path->slots[level + 1] + 1,
+			  level + 1);
+	if (wret)
+		ret = wret;
+
+	ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
+	BUG_ON(ret);
+
+	if (path->slots[level] >= mid) {
+		path->slots[level] -= mid;
+		btrfs_tree_unlock(c);
+		free_extent_buffer(c);
+		path->nodes[level] = split;
+		path->slots[level + 1] += 1;
+	} else {
+		btrfs_tree_unlock(split);
+		free_extent_buffer(split);
+	}
+	return ret;
+}
+
+/*
+ * how many bytes are required to store the items in a leaf.  start
+ * and nr indicate which items in the leaf to check.  This totals up the
+ * space used both by the item structs and the item data
+ */
+static int leaf_space_used(struct extent_buffer *l, int start, int nr)
+{
+	int data_len;
+	int nritems = btrfs_header_nritems(l);
+	int end = min(nritems, start + nr) - 1;
+
+	if (!nr)
+		return 0;
+	data_len = btrfs_item_end_nr(l, start);
+	data_len = data_len - btrfs_item_offset_nr(l, end);
+	data_len += sizeof(struct btrfs_item) * nr;
+	WARN_ON(data_len < 0);
+	return data_len;
+}
+
+/*
+ * The space between the end of the leaf items and
+ * the start of the leaf data.  IOW, how much room
+ * the leaf has left for both items and data
+ */
+noinline int btrfs_leaf_free_space(struct btrfs_root *root,
+				   struct extent_buffer *leaf)
+{
+	int nritems = btrfs_header_nritems(leaf);
+	int ret;
+	ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
+	if (ret < 0) {
+		printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
+		       "used %d nritems %d\n",
+		       ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
+		       leaf_space_used(leaf, 0, nritems), nritems);
+	}
+	return ret;
+}
+
+/*
+ * push some data in the path leaf to the right, trying to free up at
+ * least data_size bytes.  returns zero if the push worked, nonzero otherwise
+ *
+ * returns 1 if the push failed because the other node didn't have enough
+ * room, 0 if everything worked out and < 0 if there were major errors.
+ */
+static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
+			   *root, struct btrfs_path *path, int data_size,
+			   int empty)
+{
+	struct extent_buffer *left = path->nodes[0];
+	struct extent_buffer *right;
+	struct extent_buffer *upper;
+	struct btrfs_disk_key disk_key;
+	int slot;
+	u32 i;
+	int free_space;
+	int push_space = 0;
+	int push_items = 0;
+	struct btrfs_item *item;
+	u32 left_nritems;
+	u32 nr;
+	u32 right_nritems;
+	u32 data_end;
+	u32 this_item_size;
+	int ret;
+
+	slot = path->slots[1];
+	if (!path->nodes[1])
+		return 1;
+
+	upper = path->nodes[1];
+	if (slot >= btrfs_header_nritems(upper) - 1)
+		return 1;
+
+	WARN_ON(!btrfs_tree_locked(path->nodes[1]));
+
+	right = read_node_slot(root, upper, slot + 1);
+	btrfs_tree_lock(right);
+	free_space = btrfs_leaf_free_space(root, right);
+	if (free_space < data_size)
+		goto out_unlock;
+
+	/* cow and double check */
+	ret = btrfs_cow_block(trans, root, right, upper,
+			      slot + 1, &right, 0);
+	if (ret)
+		goto out_unlock;
+
+	free_space = btrfs_leaf_free_space(root, right);
+	if (free_space < data_size)
+		goto out_unlock;
+
+	left_nritems = btrfs_header_nritems(left);
+	if (left_nritems == 0)
+		goto out_unlock;
+
+	if (empty)
+		nr = 0;
+	else
+		nr = 1;
+
+	if (path->slots[0] >= left_nritems)
+		push_space += data_size;
+
+	i = left_nritems - 1;
+	while (i >= nr) {
+		item = btrfs_item_nr(left, i);
+
+		if (!empty && push_items > 0) {
+			if (path->slots[0] > i)
+				break;
+			if (path->slots[0] == i) {
+				int space = btrfs_leaf_free_space(root, left);
+				if (space + push_space * 2 > free_space)
+					break;
+			}
+		}
+
+		if (path->slots[0] == i)
+			push_space += data_size;
+
+		if (!left->map_token) {
+			map_extent_buffer(left, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&left->map_token, &left->kaddr,
+					&left->map_start, &left->map_len,
+					KM_USER1);
+		}
+
+		this_item_size = btrfs_item_size(left, item);
+		if (this_item_size + sizeof(*item) + push_space > free_space)
+			break;
+
+		push_items++;
+		push_space += this_item_size + sizeof(*item);
+		if (i == 0)
+			break;
+		i--;
+	}
+	if (left->map_token) {
+		unmap_extent_buffer(left, left->map_token, KM_USER1);
+		left->map_token = NULL;
+	}
+
+	if (push_items == 0)
+		goto out_unlock;
+
+	if (!empty && push_items == left_nritems)
+		WARN_ON(1);
+
+	/* push left to right */
+	right_nritems = btrfs_header_nritems(right);
+
+	push_space = btrfs_item_end_nr(left, left_nritems - push_items);
+	push_space -= leaf_data_end(root, left);
+
+	/* make room in the right data area */
+	data_end = leaf_data_end(root, right);
+	memmove_extent_buffer(right,
+			      btrfs_leaf_data(right) + data_end - push_space,
+			      btrfs_leaf_data(right) + data_end,
+			      BTRFS_LEAF_DATA_SIZE(root) - data_end);
+
+	/* copy from the left data area */
+	copy_extent_buffer(right, left, btrfs_leaf_data(right) +
+		     BTRFS_LEAF_DATA_SIZE(root) - push_space,
+		     btrfs_leaf_data(left) + leaf_data_end(root, left),
+		     push_space);
+
+	memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
+			      btrfs_item_nr_offset(0),
+			      right_nritems * sizeof(struct btrfs_item));
+
+	/* copy the items from left to right */
+	copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
+		   btrfs_item_nr_offset(left_nritems - push_items),
+		   push_items * sizeof(struct btrfs_item));
+
+	/* update the item pointers */
+	right_nritems += push_items;
+	btrfs_set_header_nritems(right, right_nritems);
+	push_space = BTRFS_LEAF_DATA_SIZE(root);
+	for (i = 0; i < right_nritems; i++) {
+		item = btrfs_item_nr(right, i);
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+		push_space -= btrfs_item_size(right, item);
+		btrfs_set_item_offset(right, item, push_space);
+	}
+
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+	left_nritems -= push_items;
+	btrfs_set_header_nritems(left, left_nritems);
+
+	if (left_nritems)
+		btrfs_mark_buffer_dirty(left);
+	btrfs_mark_buffer_dirty(right);
+
+	ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
+	BUG_ON(ret);
+
+	btrfs_item_key(right, &disk_key, 0);
+	btrfs_set_node_key(upper, &disk_key, slot + 1);
+	btrfs_mark_buffer_dirty(upper);
+
+	/* then fixup the leaf pointer in the path */
+	if (path->slots[0] >= left_nritems) {
+		path->slots[0] -= left_nritems;
+		if (btrfs_header_nritems(path->nodes[0]) == 0)
+			clean_tree_block(trans, root, path->nodes[0]);
+		btrfs_tree_unlock(path->nodes[0]);
+		free_extent_buffer(path->nodes[0]);
+		path->nodes[0] = right;
+		path->slots[1] += 1;
+	} else {
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+	return 0;
+
+out_unlock:
+	btrfs_tree_unlock(right);
+	free_extent_buffer(right);
+	return 1;
+}
+
+/*
+ * push some data in the path leaf to the left, trying to free up at
+ * least data_size bytes.  returns zero if the push worked, nonzero otherwise
+ */
+static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
+			  *root, struct btrfs_path *path, int data_size,
+			  int empty)
+{
+	struct btrfs_disk_key disk_key;
+	struct extent_buffer *right = path->nodes[0];
+	struct extent_buffer *left;
+	int slot;
+	int i;
+	int free_space;
+	int push_space = 0;
+	int push_items = 0;
+	struct btrfs_item *item;
+	u32 old_left_nritems;
+	u32 right_nritems;
+	u32 nr;
+	int ret = 0;
+	int wret;
+	u32 this_item_size;
+	u32 old_left_item_size;
+
+	slot = path->slots[1];
+	if (slot == 0)
+		return 1;
+	if (!path->nodes[1])
+		return 1;
+
+	right_nritems = btrfs_header_nritems(right);
+	if (right_nritems == 0)
+		return 1;
+
+	WARN_ON(!btrfs_tree_locked(path->nodes[1]));
+
+	left = read_node_slot(root, path->nodes[1], slot - 1);
+	btrfs_tree_lock(left);
+	free_space = btrfs_leaf_free_space(root, left);
+	if (free_space < data_size) {
+		ret = 1;
+		goto out;
+	}
+
+	/* cow and double check */
+	ret = btrfs_cow_block(trans, root, left,
+			      path->nodes[1], slot - 1, &left, 0);
+	if (ret) {
+		/* we hit -ENOSPC, but it isn't fatal here */
+		ret = 1;
+		goto out;
+	}
+
+	free_space = btrfs_leaf_free_space(root, left);
+	if (free_space < data_size) {
+		ret = 1;
+		goto out;
+	}
+
+	if (empty)
+		nr = right_nritems;
+	else
+		nr = right_nritems - 1;
+
+	for (i = 0; i < nr; i++) {
+		item = btrfs_item_nr(right, i);
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+
+		if (!empty && push_items > 0) {
+			if (path->slots[0] < i)
+				break;
+			if (path->slots[0] == i) {
+				int space = btrfs_leaf_free_space(root, right);
+				if (space + push_space * 2 > free_space)
+					break;
+			}
+		}
+
+		if (path->slots[0] == i)
+			push_space += data_size;
+
+		this_item_size = btrfs_item_size(right, item);
+		if (this_item_size + sizeof(*item) + push_space > free_space)
+			break;
+
+		push_items++;
+		push_space += this_item_size + sizeof(*item);
+	}
+
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+
+	if (push_items == 0) {
+		ret = 1;
+		goto out;
+	}
+	if (!empty && push_items == btrfs_header_nritems(right))
+		WARN_ON(1);
+
+	/* push data from right to left */
+	copy_extent_buffer(left, right,
+			   btrfs_item_nr_offset(btrfs_header_nritems(left)),
+			   btrfs_item_nr_offset(0),
+			   push_items * sizeof(struct btrfs_item));
+
+	push_space = BTRFS_LEAF_DATA_SIZE(root) -
+		     btrfs_item_offset_nr(right, push_items - 1);
+
+	copy_extent_buffer(left, right, btrfs_leaf_data(left) +
+		     leaf_data_end(root, left) - push_space,
+		     btrfs_leaf_data(right) +
+		     btrfs_item_offset_nr(right, push_items - 1),
+		     push_space);
+	old_left_nritems = btrfs_header_nritems(left);
+	BUG_ON(old_left_nritems <= 0);
+
+	old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
+	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
+		u32 ioff;
+
+		item = btrfs_item_nr(left, i);
+		if (!left->map_token) {
+			map_extent_buffer(left, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&left->map_token, &left->kaddr,
+					&left->map_start, &left->map_len,
+					KM_USER1);
+		}
+
+		ioff = btrfs_item_offset(left, item);
+		btrfs_set_item_offset(left, item,
+		      ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
+	}
+	btrfs_set_header_nritems(left, old_left_nritems + push_items);
+	if (left->map_token) {
+		unmap_extent_buffer(left, left->map_token, KM_USER1);
+		left->map_token = NULL;
+	}
+
+	/* fixup right node */
+	if (push_items > right_nritems) {
+		printk(KERN_CRIT "push items %d nr %u\n", push_items,
+		       right_nritems);
+		WARN_ON(1);
+	}
+
+	if (push_items < right_nritems) {
+		push_space = btrfs_item_offset_nr(right, push_items - 1) -
+						  leaf_data_end(root, right);
+		memmove_extent_buffer(right, btrfs_leaf_data(right) +
+				      BTRFS_LEAF_DATA_SIZE(root) - push_space,
+				      btrfs_leaf_data(right) +
+				      leaf_data_end(root, right), push_space);
+
+		memmove_extent_buffer(right, btrfs_item_nr_offset(0),
+			      btrfs_item_nr_offset(push_items),
+			     (btrfs_header_nritems(right) - push_items) *
+			     sizeof(struct btrfs_item));
+	}
+	right_nritems -= push_items;
+	btrfs_set_header_nritems(right, right_nritems);
+	push_space = BTRFS_LEAF_DATA_SIZE(root);
+	for (i = 0; i < right_nritems; i++) {
+		item = btrfs_item_nr(right, i);
+
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+
+		push_space = push_space - btrfs_item_size(right, item);
+		btrfs_set_item_offset(right, item, push_space);
+	}
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+
+	btrfs_mark_buffer_dirty(left);
+	if (right_nritems)
+		btrfs_mark_buffer_dirty(right);
+
+	ret = btrfs_update_ref(trans, root, right, left,
+			       old_left_nritems, push_items);
+	BUG_ON(ret);
+
+	btrfs_item_key(right, &disk_key, 0);
+	wret = fixup_low_keys(trans, root, path, &disk_key, 1);
+	if (wret)
+		ret = wret;
+
+	/* then fixup the leaf pointer in the path */
+	if (path->slots[0] < push_items) {
+		path->slots[0] += old_left_nritems;
+		if (btrfs_header_nritems(path->nodes[0]) == 0)
+			clean_tree_block(trans, root, path->nodes[0]);
+		btrfs_tree_unlock(path->nodes[0]);
+		free_extent_buffer(path->nodes[0]);
+		path->nodes[0] = left;
+		path->slots[1] -= 1;
+	} else {
+		btrfs_tree_unlock(left);
+		free_extent_buffer(left);
+		path->slots[0] -= push_items;
+	}
+	BUG_ON(path->slots[0] < 0);
+	return ret;
+out:
+	btrfs_tree_unlock(left);
+	free_extent_buffer(left);
+	return ret;
+}
+
+/*
+ * split the path's leaf in two, making sure there is at least data_size
+ * available for the resulting leaf level of the path.
+ *
+ * returns 0 if all went well and < 0 on failure.
+ */
+static noinline int split_leaf(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct btrfs_key *ins_key,
+			       struct btrfs_path *path, int data_size,
+			       int extend)
+{
+	struct extent_buffer *l;
+	u32 nritems;
+	int mid;
+	int slot;
+	struct extent_buffer *right;
+	int data_copy_size;
+	int rt_data_off;
+	int i;
+	int ret = 0;
+	int wret;
+	int double_split;
+	int num_doubles = 0;
+	struct btrfs_disk_key disk_key;
+
+	/* first try to make some room by pushing left and right */
+	if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
+		wret = push_leaf_right(trans, root, path, data_size, 0);
+		if (wret < 0)
+			return wret;
+		if (wret) {
+			wret = push_leaf_left(trans, root, path, data_size, 0);
+			if (wret < 0)
+				return wret;
+		}
+		l = path->nodes[0];
+
+		/* did the pushes work? */
+		if (btrfs_leaf_free_space(root, l) >= data_size)
+			return 0;
+	}
+
+	if (!path->nodes[1]) {
+		ret = insert_new_root(trans, root, path, 1);
+		if (ret)
+			return ret;
+	}
+again:
+	double_split = 0;
+	l = path->nodes[0];
+	slot = path->slots[0];
+	nritems = btrfs_header_nritems(l);
+	mid = (nritems + 1) / 2;
+
+	right = btrfs_alloc_free_block(trans, root, root->leafsize,
+					path->nodes[1]->start,
+					root->root_key.objectid,
+					trans->transid, 0, l->start, 0);
+	if (IS_ERR(right)) {
+		BUG_ON(1);
+		return PTR_ERR(right);
+	}
+
+	memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
+	btrfs_set_header_bytenr(right, right->start);
+	btrfs_set_header_generation(right, trans->transid);
+	btrfs_set_header_owner(right, root->root_key.objectid);
+	btrfs_set_header_level(right, 0);
+	write_extent_buffer(right, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(right),
+			    BTRFS_FSID_SIZE);
+
+	write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
+			    (unsigned long)btrfs_header_chunk_tree_uuid(right),
+			    BTRFS_UUID_SIZE);
+	if (mid <= slot) {
+		if (nritems == 1 ||
+		    leaf_space_used(l, mid, nritems - mid) + data_size >
+			BTRFS_LEAF_DATA_SIZE(root)) {
+			if (slot >= nritems) {
+				btrfs_cpu_key_to_disk(&disk_key, ins_key);
+				btrfs_set_header_nritems(right, 0);
+				wret = insert_ptr(trans, root, path,
+						  &disk_key, right->start,
+						  path->slots[1] + 1, 1);
+				if (wret)
+					ret = wret;
+
+				btrfs_tree_unlock(path->nodes[0]);
+				free_extent_buffer(path->nodes[0]);
+				path->nodes[0] = right;
+				path->slots[0] = 0;
+				path->slots[1] += 1;
+				btrfs_mark_buffer_dirty(right);
+				return ret;
+			}
+			mid = slot;
+			if (mid != nritems &&
+			    leaf_space_used(l, mid, nritems - mid) +
+			    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
+				double_split = 1;
+			}
+		}
+	} else {
+		if (leaf_space_used(l, 0, mid) + data_size >
+			BTRFS_LEAF_DATA_SIZE(root)) {
+			if (!extend && data_size && slot == 0) {
+				btrfs_cpu_key_to_disk(&disk_key, ins_key);
+				btrfs_set_header_nritems(right, 0);
+				wret = insert_ptr(trans, root, path,
+						  &disk_key,
+						  right->start,
+						  path->slots[1], 1);
+				if (wret)
+					ret = wret;
+				btrfs_tree_unlock(path->nodes[0]);
+				free_extent_buffer(path->nodes[0]);
+				path->nodes[0] = right;
+				path->slots[0] = 0;
+				if (path->slots[1] == 0) {
+					wret = fixup_low_keys(trans, root,
+						      path, &disk_key, 1);
+					if (wret)
+						ret = wret;
+				}
+				btrfs_mark_buffer_dirty(right);
+				return ret;
+			} else if ((extend || !data_size) && slot == 0) {
+				mid = 1;
+			} else {
+				mid = slot;
+				if (mid != nritems &&
+				    leaf_space_used(l, mid, nritems - mid) +
+				    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
+					double_split = 1;
+				}
+			}
+		}
+	}
+	nritems = nritems - mid;
+	btrfs_set_header_nritems(right, nritems);
+	data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
+
+	copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
+			   btrfs_item_nr_offset(mid),
+			   nritems * sizeof(struct btrfs_item));
+
+	copy_extent_buffer(right, l,
+		     btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
+		     data_copy_size, btrfs_leaf_data(l) +
+		     leaf_data_end(root, l), data_copy_size);
+
+	rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
+		      btrfs_item_end_nr(l, mid);
+
+	for (i = 0; i < nritems; i++) {
+		struct btrfs_item *item = btrfs_item_nr(right, i);
+		u32 ioff;
+
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+
+		ioff = btrfs_item_offset(right, item);
+		btrfs_set_item_offset(right, item, ioff + rt_data_off);
+	}
+
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+
+	btrfs_set_header_nritems(l, mid);
+	ret = 0;
+	btrfs_item_key(right, &disk_key, 0);
+	wret = insert_ptr(trans, root, path, &disk_key, right->start,
+			  path->slots[1] + 1, 1);
+	if (wret)
+		ret = wret;
+
+	btrfs_mark_buffer_dirty(right);
+	btrfs_mark_buffer_dirty(l);
+	BUG_ON(path->slots[0] != slot);
+
+	ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
+	BUG_ON(ret);
+
+	if (mid <= slot) {
+		btrfs_tree_unlock(path->nodes[0]);
+		free_extent_buffer(path->nodes[0]);
+		path->nodes[0] = right;
+		path->slots[0] -= mid;
+		path->slots[1] += 1;
+	} else {
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+
+	BUG_ON(path->slots[0] < 0);
+
+	if (double_split) {
+		BUG_ON(num_doubles != 0);
+		num_doubles++;
+		goto again;
+	}
+	return ret;
+}
+
+/*
+ * delete the pointer from a given node.
+ *
+ * the tree should have been previously balanced so the deletion does not
+ * empty a node.
+ */
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_path *path, int level, int slot)
+{
+	struct extent_buffer *parent = path->nodes[level];
+	u32 nritems;
+	int ret = 0;
+	int wret;
+
+	nritems = btrfs_header_nritems(parent);
+	if (slot != nritems - 1) {
+		memmove_extent_buffer(parent,
+			      btrfs_node_key_ptr_offset(slot),
+			      btrfs_node_key_ptr_offset(slot + 1),
+			      sizeof(struct btrfs_key_ptr) *
+			      (nritems - slot - 1));
+	}
+	nritems--;
+	btrfs_set_header_nritems(parent, nritems);
+	if (nritems == 0 && parent == root->node) {
+		BUG_ON(btrfs_header_level(root->node) != 1);
+		/* just turn the root into a leaf and break */
+		btrfs_set_header_level(root->node, 0);
+	} else if (slot == 0) {
+		struct btrfs_disk_key disk_key;
+
+		btrfs_node_key(parent, &disk_key, 0);
+		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
+		if (wret)
+			ret = wret;
+	}
+	btrfs_mark_buffer_dirty(parent);
+	return ret;
+}

[PATCH 26.2/35] Btrfs btree item routines

[Chris Mason]

Btrfs btree item manipulation.

This includes the helpers to actually insert, remove, extent, truncate
and do other basic operations on the btree items.

It also has helpers for walking forward and backward in the btree.

All metadata updates eventually hit this code.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- a/fs/btrfs/ctree.c	2009-01-08 10:04:14.659535797 -0500
+++ b/fs/btrfs/ctree.c	2009-01-08 09:59:41.600461904 -0500
@@ -463,6 +463,131 @@
 }
 
 /*
+ * this is used by the defrag code to go through all the
+ * leaves pointed to by a node and reallocate them so that
+ * disk order is close to key order
+ */
+int btrfs_realloc_node(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root, struct extent_buffer *parent,
+		       int start_slot, int cache_only, u64 *last_ret,
+		       struct btrfs_key *progress)
+{
+	struct extent_buffer *cur;
+	u64 blocknr;
+	u64 gen;
+	u64 search_start = *last_ret;
+	u64 last_block = 0;
+	u64 other;
+	u32 parent_nritems;
+	int end_slot;
+	int i;
+	int err = 0;
+	int parent_level;
+	int uptodate;
+	u32 blocksize;
+	int progress_passed = 0;
+	struct btrfs_disk_key disk_key;
+
+	parent_level = btrfs_header_level(parent);
+	if (cache_only && parent_level != 1)
+		return 0;
+
+	if (trans->transaction != root->fs_info->running_transaction)
+		WARN_ON(1);
+	if (trans->transid != root->fs_info->generation)
+		WARN_ON(1);
+
+	parent_nritems = btrfs_header_nritems(parent);
+	blocksize = btrfs_level_size(root, parent_level - 1);
+	end_slot = parent_nritems;
+
+	if (parent_nritems == 1)
+		return 0;
+
+	for (i = start_slot; i < end_slot; i++) {
+		int close = 1;
+
+		if (!parent->map_token) {
+			map_extent_buffer(parent,
+					btrfs_node_key_ptr_offset(i),
+					sizeof(struct btrfs_key_ptr),
+					&parent->map_token, &parent->kaddr,
+					&parent->map_start, &parent->map_len,
+					KM_USER1);
+		}
+		btrfs_node_key(parent, &disk_key, i);
+		if (!progress_passed && comp_keys(&disk_key, progress) < 0)
+			continue;
+
+		progress_passed = 1;
+		blocknr = btrfs_node_blockptr(parent, i);
+		gen = btrfs_node_ptr_generation(parent, i);
+		if (last_block == 0)
+			last_block = blocknr;
+
+		if (i > 0) {
+			other = btrfs_node_blockptr(parent, i - 1);
+			close = close_blocks(blocknr, other, blocksize);
+		}
+		if (!close && i < end_slot - 2) {
+			other = btrfs_node_blockptr(parent, i + 1);
+			close = close_blocks(blocknr, other, blocksize);
+		}
+		if (close) {
+			last_block = blocknr;
+			continue;
+		}
+		if (parent->map_token) {
+			unmap_extent_buffer(parent, parent->map_token,
+					    KM_USER1);
+			parent->map_token = NULL;
+		}
+
+		cur = btrfs_find_tree_block(root, blocknr, blocksize);
+		if (cur)
+			uptodate = btrfs_buffer_uptodate(cur, gen);
+		else
+			uptodate = 0;
+		if (!cur || !uptodate) {
+			if (cache_only) {
+				free_extent_buffer(cur);
+				continue;
+			}
+			if (!cur) {
+				cur = read_tree_block(root, blocknr,
+							 blocksize, gen);
+			} else if (!uptodate) {
+				btrfs_read_buffer(cur, gen);
+			}
+		}
+		if (search_start == 0)
+			search_start = last_block;
+
+		btrfs_tree_lock(cur);
+		err = __btrfs_cow_block(trans, root, cur, parent, i,
+					&cur, search_start,
+					min(16 * blocksize,
+					    (end_slot - i) * blocksize), 0);
+		if (err) {
+			btrfs_tree_unlock(cur);
+			free_extent_buffer(cur);
+			break;
+		}
+		search_start = cur->start;
+		last_block = cur->start;
+		*last_ret = search_start;
+		btrfs_tree_unlock(cur);
+		free_extent_buffer(cur);
+	}
+	if (parent->map_token) {
+		unmap_extent_buffer(parent, parent->map_token,
+				    KM_USER1);
+		parent->map_token = NULL;
+	}
+	return err;
+}
+
+/*
  * The leaf data grows from end-to-front in the node.
  * this returns the address of the start of the last item,
  * which is the stop of the leaf data stack
@@ -1433,6 +1558,134 @@
 	return ret;
 }
 
+int btrfs_merge_path(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_key *node_keys,
+		     u64 *nodes, int lowest_level)
+{
+	struct extent_buffer *eb;
+	struct extent_buffer *parent;
+	struct btrfs_key key;
+	u64 bytenr;
+	u64 generation;
+	u32 blocksize;
+	int level;
+	int slot;
+	int key_match;
+	int ret;
+
+	eb = btrfs_lock_root_node(root);
+	ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0);
+	BUG_ON(ret);
+
+	parent = eb;
+	while (1) {
+		level = btrfs_header_level(parent);
+		if (level == 0 || level <= lowest_level)
+			break;
+
+		ret = bin_search(parent, &node_keys[lowest_level], level,
+				 &slot);
+		if (ret && slot > 0)
+			slot--;
+
+		bytenr = btrfs_node_blockptr(parent, slot);
+		if (nodes[level - 1] == bytenr)
+			break;
+
+		blocksize = btrfs_level_size(root, level - 1);
+		generation = btrfs_node_ptr_generation(parent, slot);
+		btrfs_node_key_to_cpu(eb, &key, slot);
+		key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
+
+		if (generation == trans->transid) {
+			eb = read_tree_block(root, bytenr, blocksize,
+					     generation);
+			btrfs_tree_lock(eb);
+		}
+
+		/*
+		 * if node keys match and node pointer hasn't been modified
+		 * in the running transaction, we can merge the path. for
+		 * blocks owened by reloc trees, the node pointer check is
+		 * skipped, this is because these blocks are fully controlled
+		 * by the space balance code, no one else can modify them.
+		 */
+		if (!nodes[level - 1] || !key_match ||
+		    (generation == trans->transid &&
+		     btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
+			if (level == 1 || level == lowest_level + 1) {
+				if (generation == trans->transid) {
+					btrfs_tree_unlock(eb);
+					free_extent_buffer(eb);
+				}
+				break;
+			}
+
+			if (generation != trans->transid) {
+				eb = read_tree_block(root, bytenr, blocksize,
+						generation);
+				btrfs_tree_lock(eb);
+			}
+
+			ret = btrfs_cow_block(trans, root, eb, parent, slot,
+					      &eb, 0);
+			BUG_ON(ret);
+
+			if (root->root_key.objectid ==
+			    BTRFS_TREE_RELOC_OBJECTID) {
+				if (!nodes[level - 1]) {
+					nodes[level - 1] = eb->start;
+					memcpy(&node_keys[level - 1], &key,
+					       sizeof(node_keys[0]));
+				} else {
+					WARN_ON(1);
+				}
+			}
+
+			btrfs_tree_unlock(parent);
+			free_extent_buffer(parent);
+			parent = eb;
+			continue;
+		}
+
+		btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
+		btrfs_set_node_ptr_generation(parent, slot, trans->transid);
+		btrfs_mark_buffer_dirty(parent);
+
+		ret = btrfs_inc_extent_ref(trans, root,
+					nodes[level - 1],
+					blocksize, parent->start,
+					btrfs_header_owner(parent),
+					btrfs_header_generation(parent),
+					level - 1);
+		BUG_ON(ret);
+
+		/*
+		 * If the block was created in the running transaction,
+		 * it's possible this is the last reference to it, so we
+		 * should drop the subtree.
+		 */
+		if (generation == trans->transid) {
+			ret = btrfs_drop_subtree(trans, root, eb, parent);
+			BUG_ON(ret);
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
+		} else {
+			ret = btrfs_free_extent(trans, root, bytenr,
+					blocksize, parent->start,
+					btrfs_header_owner(parent),
+					btrfs_header_generation(parent),
+					level - 1, 1);
+			BUG_ON(ret);
+		}
+		break;
+	}
+	btrfs_tree_unlock(parent);
+	free_extent_buffer(parent);
+	return 0;
+}
+
 /*
  * adjust the pointers going up the tree, starting at level
  * making sure the right key of each node is points to 'key'.
@@ -1464,6 +1717,40 @@
 	return ret;
 }
 
+/*
+ * update item key.
+ *
+ * This function isn't completely safe. It's the caller's responsibility
+ * that the new key won't break the order
+ */
+int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, struct btrfs_path *path,
+			    struct btrfs_key *new_key)
+{
+	struct btrfs_disk_key disk_key;
+	struct extent_buffer *eb;
+	int slot;
+
+	eb = path->nodes[0];
+	slot = path->slots[0];
+	if (slot > 0) {
+		btrfs_item_key(eb, &disk_key, slot - 1);
+		if (comp_keys(&disk_key, new_key) >= 0)
+			return -1;
+	}
+	if (slot < btrfs_header_nritems(eb) - 1) {
+		btrfs_item_key(eb, &disk_key, slot + 1);
+		if (comp_keys(&disk_key, new_key) <= 0)
+			return -1;
+	}
+
+	btrfs_cpu_key_to_disk(&disk_key, new_key);
+	btrfs_set_item_key(eb, &disk_key, slot);
+	btrfs_mark_buffer_dirty(eb);
+	if (slot == 0)
+		fixup_low_keys(trans, root, path, &disk_key, 1);
+	return 0;
+}
 
 /*
  * try to push data from one node into the next node left in the
@@ -1593,6 +1880,85 @@
 }
 
 /*
+ * helper function to insert a new root level in the tree.
+ * A new node is allocated, and a single item is inserted to
+ * point to the existing root
+ *
+ * returns zero on success or < 0 on failure.
+ */
+static noinline int insert_new_root(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   struct btrfs_path *path, int level)
+{
+	u64 lower_gen;
+	struct extent_buffer *lower;
+	struct extent_buffer *c;
+	struct extent_buffer *old;
+	struct btrfs_disk_key lower_key;
+	int ret;
+
+	BUG_ON(path->nodes[level]);
+	BUG_ON(path->nodes[level-1] != root->node);
+
+	lower = path->nodes[level-1];
+	if (level == 1)
+		btrfs_item_key(lower, &lower_key, 0);
+	else
+		btrfs_node_key(lower, &lower_key, 0);
+
+	c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
+				   root->root_key.objectid, trans->transid,
+				   level, root->node->start, 0);
+	if (IS_ERR(c))
+		return PTR_ERR(c);
+
+	memset_extent_buffer(c, 0, 0, root->nodesize);
+	btrfs_set_header_nritems(c, 1);
+	btrfs_set_header_level(c, level);
+	btrfs_set_header_bytenr(c, c->start);
+	btrfs_set_header_generation(c, trans->transid);
+	btrfs_set_header_owner(c, root->root_key.objectid);
+
+	write_extent_buffer(c, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(c),
+			    BTRFS_FSID_SIZE);
+
+	write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
+			    (unsigned long)btrfs_header_chunk_tree_uuid(c),
+			    BTRFS_UUID_SIZE);
+
+	btrfs_set_node_key(c, &lower_key, 0);
+	btrfs_set_node_blockptr(c, 0, lower->start);
+	lower_gen = btrfs_header_generation(lower);
+	WARN_ON(lower_gen != trans->transid);
+
+	btrfs_set_node_ptr_generation(c, 0, lower_gen);
+
+	btrfs_mark_buffer_dirty(c);
+
+	spin_lock(&root->node_lock);
+	old = root->node;
+	root->node = c;
+	spin_unlock(&root->node_lock);
+
+	ret = btrfs_update_extent_ref(trans, root, lower->start,
+				      lower->start, c->start,
+				      root->root_key.objectid,
+				      trans->transid, level - 1);
+	BUG_ON(ret);
+
+	/* the super has an extra ref to root->node */
+	free_extent_buffer(old);
+
+	add_root_to_dirty_list(root);
+	extent_buffer_get(c);
+	path->nodes[level] = c;
+	path->locks[level] = 1;
+	path->slots[level] = 0;
+	return 0;
+}
+
+/*
  * worker function to insert a single pointer in a node.
  * the node should have enough room for the pointer already
  *
@@ -2401,41 +2767,1187 @@
 }
 
 /*
- * delete the pointer from a given node.
+ * This function splits a single item into two items,
+ * giving 'new_key' to the new item and splitting the
+ * old one at split_offset (from the start of the item).
  *
- * the tree should have been previously balanced so the deletion does not
- * empty a node.
+ * The path may be released by this operation.  After
+ * the split, the path is pointing to the old item.  The
+ * new item is going to be in the same node as the old one.
+ *
+ * Note, the item being split must be smaller enough to live alone on
+ * a tree block with room for one extra struct btrfs_item
+ *
+ * This allows us to split the item in place, keeping a lock on the
+ * leaf the entire time.
  */
-static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		   struct btrfs_path *path, int level, int slot)
-{
-	struct extent_buffer *parent = path->nodes[level];
+int btrfs_split_item(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_path *path,
+		     struct btrfs_key *new_key,
+		     unsigned long split_offset)
+{
+	u32 item_size;
+	struct extent_buffer *leaf;
+	struct btrfs_key orig_key;
+	struct btrfs_item *item;
+	struct btrfs_item *new_item;
+	int ret = 0;
+	int slot;
 	u32 nritems;
+	u32 orig_offset;
+	struct btrfs_disk_key disk_key;
+	char *buf;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
+	if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
+		goto split;
+
+	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+	btrfs_release_path(root, path);
+
+	path->search_for_split = 1;
+	path->keep_locks = 1;
+
+	ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
+	path->search_for_split = 0;
+
+	/* if our item isn't there or got smaller, return now */
+	if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
+							path->slots[0])) {
+		path->keep_locks = 0;
+		return -EAGAIN;
+	}
+
+	ret = split_leaf(trans, root, &orig_key, path,
+			 sizeof(struct btrfs_item), 1);
+	path->keep_locks = 0;
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
+
+split:
+	item = btrfs_item_nr(leaf, path->slots[0]);
+	orig_offset = btrfs_item_offset(leaf, item);
+	item_size = btrfs_item_size(leaf, item);
+
+
+	buf = kmalloc(item_size, GFP_NOFS);
+	read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
+			    path->slots[0]), item_size);
+	slot = path->slots[0] + 1;
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+
+	if (slot != nritems) {
+		/* shift the items */
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
+			      btrfs_item_nr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_item));
+
+	}
+
+	btrfs_cpu_key_to_disk(&disk_key, new_key);
+	btrfs_set_item_key(leaf, &disk_key, slot);
+
+	new_item = btrfs_item_nr(leaf, slot);
+
+	btrfs_set_item_offset(leaf, new_item, orig_offset);
+	btrfs_set_item_size(leaf, new_item, item_size - split_offset);
+
+	btrfs_set_item_offset(leaf, item,
+			      orig_offset + item_size - split_offset);
+	btrfs_set_item_size(leaf, item, split_offset);
+
+	btrfs_set_header_nritems(leaf, nritems + 1);
+
+	/* write the data for the start of the original item */
+	write_extent_buffer(leaf, buf,
+			    btrfs_item_ptr_offset(leaf, path->slots[0]),
+			    split_offset);
+
+	/* write the data for the new item */
+	write_extent_buffer(leaf, buf + split_offset,
+			    btrfs_item_ptr_offset(leaf, slot),
+			    item_size - split_offset);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+	kfree(buf);
+	return ret;
+}
+
+/*
+ * make the item pointed to by the path smaller.  new_size indicates
+ * how small to make it, and from_end tells us if we just chop bytes
+ * off the end of the item or if we shift the item to chop bytes off
+ * the front.
+ */
+int btrfs_truncate_item(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_path *path,
+			u32 new_size, int from_end)
+{
 	int ret = 0;
-	int wret;
+	int slot;
+	int slot_orig;
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	u32 nritems;
+	unsigned int data_end;
+	unsigned int old_data_start;
+	unsigned int old_size;
+	unsigned int size_diff;
+	int i;
 
-	nritems = btrfs_header_nritems(parent);
-	if (slot != nritems - 1) {
-		memmove_extent_buffer(parent,
-			      btrfs_node_key_ptr_offset(slot),
-			      btrfs_node_key_ptr_offset(slot + 1),
-			      sizeof(struct btrfs_key_ptr) *
-			      (nritems - slot - 1));
+	slot_orig = path->slots[0];
+	leaf = path->nodes[0];
+	slot = path->slots[0];
+
+	old_size = btrfs_item_size_nr(leaf, slot);
+	if (old_size == new_size)
+		return 0;
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	old_data_start = btrfs_item_offset_nr(leaf, slot);
+
+	size_diff = old_size - new_size;
+
+	BUG_ON(slot < 0);
+	BUG_ON(slot >= nritems);
+
+	/*
+	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+	 */
+	/* first correct the data pointers */
+	for (i = slot; i < nritems; i++) {
+		u32 ioff;
+		item = btrfs_item_nr(leaf, i);
+
+		if (!leaf->map_token) {
+			map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+		}
+
+		ioff = btrfs_item_offset(leaf, item);
+		btrfs_set_item_offset(leaf, item, ioff + size_diff);
 	}
-	nritems--;
-	btrfs_set_header_nritems(parent, nritems);
-	if (nritems == 0 && parent == root->node) {
-		BUG_ON(btrfs_header_level(root->node) != 1);
-		/* just turn the root into a leaf and break */
-		btrfs_set_header_level(root->node, 0);
-	} else if (slot == 0) {
+
+	if (leaf->map_token) {
+		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+		leaf->map_token = NULL;
+	}
+
+	/* shift the data */
+	if (from_end) {
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + size_diff, btrfs_leaf_data(leaf) +
+			      data_end, old_data_start + new_size - data_end);
+	} else {
 		struct btrfs_disk_key disk_key;
+		u64 offset;
 
-		btrfs_node_key(parent, &disk_key, 0);
-		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
-		if (wret)
-			ret = wret;
+		btrfs_item_key(leaf, &disk_key, slot);
+
+		if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
+			unsigned long ptr;
+			struct btrfs_file_extent_item *fi;
+
+			fi = btrfs_item_ptr(leaf, slot,
+					    struct btrfs_file_extent_item);
+			fi = (struct btrfs_file_extent_item *)(
+			     (unsigned long)fi - size_diff);
+
+			if (btrfs_file_extent_type(leaf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE) {
+				ptr = btrfs_item_ptr_offset(leaf, slot);
+				memmove_extent_buffer(leaf, ptr,
+				      (unsigned long)fi,
+				      offsetof(struct btrfs_file_extent_item,
+						 disk_bytenr));
+			}
+		}
+
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + size_diff, btrfs_leaf_data(leaf) +
+			      data_end, old_data_start - data_end);
+
+		offset = btrfs_disk_key_offset(&disk_key);
+		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
+		btrfs_set_item_key(leaf, &disk_key, slot);
+		if (slot == 0)
+			fixup_low_keys(trans, root, path, &disk_key, 1);
+	}
+
+	item = btrfs_item_nr(leaf, slot);
+	btrfs_set_item_size(leaf, item, new_size);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+	return ret;
+}
+
+/*
+ * make the item pointed to by the path bigger, data_size is the new size.
+ */
+int btrfs_extend_item(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root, struct btrfs_path *path,
+		      u32 data_size)
+{
+	int ret = 0;
+	int slot;
+	int slot_orig;
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	u32 nritems;
+	unsigned int data_end;
+	unsigned int old_data;
+	unsigned int old_size;
+	int i;
+
+	slot_orig = path->slots[0];
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	if (btrfs_leaf_free_space(root, leaf) < data_size) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+	slot = path->slots[0];
+	old_data = btrfs_item_end_nr(leaf, slot);
+
+	BUG_ON(slot < 0);
+	if (slot >= nritems) {
+		btrfs_print_leaf(root, leaf);
+		printk(KERN_CRIT "slot %d too large, nritems %d\n",
+		       slot, nritems);
+		BUG_ON(1);
+	}
+
+	/*
+	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+	 */
+	/* first correct the data pointers */
+	for (i = slot; i < nritems; i++) {
+		u32 ioff;
+		item = btrfs_item_nr(leaf, i);
+
+		if (!leaf->map_token) {
+			map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+		}
+		ioff = btrfs_item_offset(leaf, item);
+		btrfs_set_item_offset(leaf, item, ioff - data_size);
+	}
+
+	if (leaf->map_token) {
+		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+		leaf->map_token = NULL;
+	}
+
+	/* shift the data */
+	memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+		      data_end - data_size, btrfs_leaf_data(leaf) +
+		      data_end, old_data - data_end);
+
+	data_end = old_data;
+	old_size = btrfs_item_size_nr(leaf, slot);
+	item = btrfs_item_nr(leaf, slot);
+	btrfs_set_item_size(leaf, item, old_size + data_size);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
 	}
-	btrfs_mark_buffer_dirty(parent);
 	return ret;
 }
+
+/*
+ * Given a key and some data, insert items into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ * Returns the number of keys that were inserted.
+ */
+int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path,
+			    struct btrfs_key *cpu_key, u32 *data_size,
+			    int nr)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	int ret = 0;
+	int slot;
+	int i;
+	u32 nritems;
+	u32 total_data = 0;
+	u32 total_size = 0;
+	unsigned int data_end;
+	struct btrfs_disk_key disk_key;
+	struct btrfs_key found_key;
+
+	for (i = 0; i < nr; i++) {
+		if (total_size + data_size[i] + sizeof(struct btrfs_item) >
+		    BTRFS_LEAF_DATA_SIZE(root)) {
+			break;
+			nr = i;
+		}
+		total_data += data_size[i];
+		total_size += data_size[i] + sizeof(struct btrfs_item);
+	}
+	BUG_ON(nr == 0);
+
+	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
+	if (ret == 0)
+		return -EEXIST;
+	if (ret < 0)
+		goto out;
+
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	if (btrfs_leaf_free_space(root, leaf) < total_size) {
+		for (i = nr; i >= 0; i--) {
+			total_data -= data_size[i];
+			total_size -= data_size[i] + sizeof(struct btrfs_item);
+			if (total_size < btrfs_leaf_free_space(root, leaf))
+				break;
+		}
+		nr = i;
+	}
+
+	slot = path->slots[0];
+	BUG_ON(slot < 0);
+
+	if (slot != nritems) {
+		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
+
+		item = btrfs_item_nr(leaf, slot);
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+
+		/* figure out how many keys we can insert in here */
+		total_data = data_size[0];
+		for (i = 1; i < nr; i++) {
+			if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
+				break;
+			total_data += data_size[i];
+		}
+		nr = i;
+
+		if (old_data < data_end) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
+			       slot, old_data, data_end);
+			BUG_ON(1);
+		}
+		/*
+		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+		 */
+		/* first correct the data pointers */
+		WARN_ON(leaf->map_token);
+		for (i = slot; i < nritems; i++) {
+			u32 ioff;
+
+			item = btrfs_item_nr(leaf, i);
+			if (!leaf->map_token) {
+				map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+			}
+
+			ioff = btrfs_item_offset(leaf, item);
+			btrfs_set_item_offset(leaf, item, ioff - total_data);
+		}
+		if (leaf->map_token) {
+			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+			leaf->map_token = NULL;
+		}
+
+		/* shift the items */
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
+			      btrfs_item_nr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_item));
+
+		/* shift the data */
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end - total_data, btrfs_leaf_data(leaf) +
+			      data_end, old_data - data_end);
+		data_end = old_data;
+	} else {
+		/*
+		 * this sucks but it has to be done, if we are inserting at
+		 * the end of the leaf only insert 1 of the items, since we
+		 * have no way of knowing whats on the next leaf and we'd have
+		 * to drop our current locks to figure it out
+		 */
+		nr = 1;
+	}
+
+	/* setup the item for the new data */
+	for (i = 0; i < nr; i++) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
+		btrfs_set_item_key(leaf, &disk_key, slot + i);
+		item = btrfs_item_nr(leaf, slot + i);
+		btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
+		data_end -= data_size[i];
+		btrfs_set_item_size(leaf, item, data_size[i]);
+	}
+	btrfs_set_header_nritems(leaf, nritems + nr);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (slot == 0) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
+		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
+	}
+
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+out:
+	if (!ret)
+		ret = nr;
+	return ret;
+}
+
+/*
+ * Given a key and some data, insert items into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ */
+int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path,
+			    struct btrfs_key *cpu_key, u32 *data_size,
+			    int nr)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	int ret = 0;
+	int slot;
+	int slot_orig;
+	int i;
+	u32 nritems;
+	u32 total_size = 0;
+	u32 total_data = 0;
+	unsigned int data_end;
+	struct btrfs_disk_key disk_key;
+
+	for (i = 0; i < nr; i++)
+		total_data += data_size[i];
+
+	total_size = total_data + (nr * sizeof(struct btrfs_item));
+	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
+	if (ret == 0)
+		return -EEXIST;
+	if (ret < 0)
+		goto out;
+
+	slot_orig = path->slots[0];
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	if (btrfs_leaf_free_space(root, leaf) < total_size) {
+		btrfs_print_leaf(root, leaf);
+		printk(KERN_CRIT "not enough freespace need %u have %d\n",
+		       total_size, btrfs_leaf_free_space(root, leaf));
+		BUG();
+	}
+
+	slot = path->slots[0];
+	BUG_ON(slot < 0);
+
+	if (slot != nritems) {
+		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
+
+		if (old_data < data_end) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
+			       slot, old_data, data_end);
+			BUG_ON(1);
+		}
+		/*
+		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+		 */
+		/* first correct the data pointers */
+		WARN_ON(leaf->map_token);
+		for (i = slot; i < nritems; i++) {
+			u32 ioff;
+
+			item = btrfs_item_nr(leaf, i);
+			if (!leaf->map_token) {
+				map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+			}
+
+			ioff = btrfs_item_offset(leaf, item);
+			btrfs_set_item_offset(leaf, item, ioff - total_data);
+		}
+		if (leaf->map_token) {
+			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+			leaf->map_token = NULL;
+		}
+
+		/* shift the items */
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
+			      btrfs_item_nr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_item));
+
+		/* shift the data */
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end - total_data, btrfs_leaf_data(leaf) +
+			      data_end, old_data - data_end);
+		data_end = old_data;
+	}
+
+	/* setup the item for the new data */
+	for (i = 0; i < nr; i++) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
+		btrfs_set_item_key(leaf, &disk_key, slot + i);
+		item = btrfs_item_nr(leaf, slot + i);
+		btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
+		data_end -= data_size[i];
+		btrfs_set_item_size(leaf, item, data_size[i]);
+	}
+	btrfs_set_header_nritems(leaf, nritems + nr);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (slot == 0) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
+		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
+	}
+
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+out:
+	return ret;
+}
+
+/*
+ * Given a key and some data, insert an item into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ */
+int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *cpu_key, void *data, u32
+		      data_size)
+{
+	int ret = 0;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	unsigned long ptr;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
+	if (!ret) {
+		leaf = path->nodes[0];
+		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+		write_extent_buffer(leaf, data, ptr, data_size);
+		btrfs_mark_buffer_dirty(leaf);
+	}
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * delete the pointer from a given node.
+ *
+ * the tree should have been previously balanced so the deletion does not
+ * empty a node.
+ */
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_path *path, int level, int slot)
+{
+	struct extent_buffer *parent = path->nodes[level];
+	u32 nritems;
+	int ret = 0;
+	int wret;
+
+	nritems = btrfs_header_nritems(parent);
+	if (slot != nritems - 1) {
+		memmove_extent_buffer(parent,
+			      btrfs_node_key_ptr_offset(slot),
+			      btrfs_node_key_ptr_offset(slot + 1),
+			      sizeof(struct btrfs_key_ptr) *
+			      (nritems - slot - 1));
+	}
+	nritems--;
+	btrfs_set_header_nritems(parent, nritems);
+	if (nritems == 0 && parent == root->node) {
+		BUG_ON(btrfs_header_level(root->node) != 1);
+		/* just turn the root into a leaf and break */
+		btrfs_set_header_level(root->node, 0);
+	} else if (slot == 0) {
+		struct btrfs_disk_key disk_key;
+
+		btrfs_node_key(parent, &disk_key, 0);
+		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
+		if (wret)
+			ret = wret;
+	}
+	btrfs_mark_buffer_dirty(parent);
+	return ret;
+}
+
+/*
+ * a helper function to delete the leaf pointed to by path->slots[1] and
+ * path->nodes[1].  bytenr is the node block pointer, but since the callers
+ * already know it, it is faster to have them pass it down than to
+ * read it out of the node again.
+ *
+ * This deletes the pointer in path->nodes[1] and frees the leaf
+ * block extent.  zero is returned if it all worked out, < 0 otherwise.
+ *
+ * The path must have already been setup for deleting the leaf, including
+ * all the proper balancing.  path->nodes[1] must be locked.
+ */
+noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path, u64 bytenr)
+{
+	int ret;
+	u64 root_gen = btrfs_header_generation(path->nodes[1]);
+
+	ret = del_ptr(trans, root, path, 1, path->slots[1]);
+	if (ret)
+		return ret;
+
+	ret = btrfs_free_extent(trans, root, bytenr,
+				btrfs_level_size(root, 0),
+				path->nodes[1]->start,
+				btrfs_header_owner(path->nodes[1]),
+				root_gen, 0, 1);
+	return ret;
+}
+/*
+ * delete the item at the leaf level in path.  If that empties
+ * the leaf, remove it from the tree
+ */
+int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		    struct btrfs_path *path, int slot, int nr)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	int last_off;
+	int dsize = 0;
+	int ret = 0;
+	int wret;
+	int i;
+	u32 nritems;
+
+	leaf = path->nodes[0];
+	last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
+
+	for (i = 0; i < nr; i++)
+		dsize += btrfs_item_size_nr(leaf, slot + i);
+
+	nritems = btrfs_header_nritems(leaf);
+
+	if (slot + nr != nritems) {
+		int data_end = leaf_data_end(root, leaf);
+
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + dsize,
+			      btrfs_leaf_data(leaf) + data_end,
+			      last_off - data_end);
+
+		for (i = slot + nr; i < nritems; i++) {
+			u32 ioff;
+
+			item = btrfs_item_nr(leaf, i);
+			if (!leaf->map_token) {
+				map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+			}
+			ioff = btrfs_item_offset(leaf, item);
+			btrfs_set_item_offset(leaf, item, ioff + dsize);
+		}
+
+		if (leaf->map_token) {
+			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+			leaf->map_token = NULL;
+		}
+
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
+			      btrfs_item_nr_offset(slot + nr),
+			      sizeof(struct btrfs_item) *
+			      (nritems - slot - nr));
+	}
+	btrfs_set_header_nritems(leaf, nritems - nr);
+	nritems -= nr;
+
+	/* delete the leaf if we've emptied it */
+	if (nritems == 0) {
+		if (leaf == root->node) {
+			btrfs_set_header_level(leaf, 0);
+		} else {
+			ret = btrfs_del_leaf(trans, root, path, leaf->start);
+			BUG_ON(ret);
+		}
+	} else {
+		int used = leaf_space_used(leaf, 0, nritems);
+		if (slot == 0) {
+			struct btrfs_disk_key disk_key;
+
+			btrfs_item_key(leaf, &disk_key, 0);
+			wret = fixup_low_keys(trans, root, path,
+					      &disk_key, 1);
+			if (wret)
+				ret = wret;
+		}
+
+		/* delete the leaf if it is mostly empty */
+		if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
+			/* push_leaf_left fixes the path.
+			 * make sure the path still points to our leaf
+			 * for possible call to del_ptr below
+			 */
+			slot = path->slots[1];
+			extent_buffer_get(leaf);
+
+			wret = push_leaf_left(trans, root, path, 1, 1);
+			if (wret < 0 && wret != -ENOSPC)
+				ret = wret;
+
+			if (path->nodes[0] == leaf &&
+			    btrfs_header_nritems(leaf)) {
+				wret = push_leaf_right(trans, root, path, 1, 1);
+				if (wret < 0 && wret != -ENOSPC)
+					ret = wret;
+			}
+
+			if (btrfs_header_nritems(leaf) == 0) {
+				path->slots[1] = slot;
+				ret = btrfs_del_leaf(trans, root, path,
+						     leaf->start);
+				BUG_ON(ret);
+				free_extent_buffer(leaf);
+			} else {
+				/* if we're still in the path, make sure
+				 * we're dirty.  Otherwise, one of the
+				 * push_leaf functions must have already
+				 * dirtied this buffer
+				 */
+				if (path->nodes[0] == leaf)
+					btrfs_mark_buffer_dirty(leaf);
+				free_extent_buffer(leaf);
+			}
+		} else {
+			btrfs_mark_buffer_dirty(leaf);
+		}
+	}
+	return ret;
+}
+
+/*
+ * search the tree again to find a leaf with lesser keys
+ * returns 0 if it found something or 1 if there are no lesser leaves.
+ * returns < 0 on io errors.
+ *
+ * This may release the path, and so you may lose any locks held at the
+ * time you call it.
+ */
+int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
+{
+	struct btrfs_key key;
+	struct btrfs_disk_key found_key;
+	int ret;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
+
+	if (key.offset > 0)
+		key.offset--;
+	else if (key.type > 0)
+		key.type--;
+	else if (key.objectid > 0)
+		key.objectid--;
+	else
+		return 1;
+
+	btrfs_release_path(root, path);
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		return ret;
+	btrfs_item_key(path->nodes[0], &found_key, 0);
+	ret = comp_keys(&found_key, &key);
+	if (ret < 0)
+		return 0;
+	return 1;
+}
+
+/*
+ * A helper function to walk down the tree starting at min_key, and looking
+ * for nodes or leaves that are either in cache or have a minimum
+ * transaction id.  This is used by the btree defrag code, and tree logging
+ *
+ * This does not cow, but it does stuff the starting key it finds back
+ * into min_key, so you can call btrfs_search_slot with cow=1 on the
+ * key and get a writable path.
+ *
+ * This does lock as it descends, and path->keep_locks should be set
+ * to 1 by the caller.
+ *
+ * This honors path->lowest_level to prevent descent past a given level
+ * of the tree.
+ *
+ * min_trans indicates the oldest transaction that you are interested
+ * in walking through.  Any nodes or leaves older than min_trans are
+ * skipped over (without reading them).
+ *
+ * returns zero if something useful was found, < 0 on error and 1 if there
+ * was nothing in the tree that matched the search criteria.
+ */
+int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
+			 struct btrfs_key *max_key,
+			 struct btrfs_path *path, int cache_only,
+			 u64 min_trans)
+{
+	struct extent_buffer *cur;
+	struct btrfs_key found_key;
+	int slot;
+	int sret;
+	u32 nritems;
+	int level;
+	int ret = 1;
+
+	WARN_ON(!path->keep_locks);
+again:
+	cur = btrfs_lock_root_node(root);
+	level = btrfs_header_level(cur);
+	WARN_ON(path->nodes[level]);
+	path->nodes[level] = cur;
+	path->locks[level] = 1;
+
+	if (btrfs_header_generation(cur) < min_trans) {
+		ret = 1;
+		goto out;
+	}
+	while (1) {
+		nritems = btrfs_header_nritems(cur);
+		level = btrfs_header_level(cur);
+		sret = bin_search(cur, min_key, level, &slot);
+
+		/* at the lowest level, we're done, setup the path and exit */
+		if (level == path->lowest_level) {
+			if (slot >= nritems)
+				goto find_next_key;
+			ret = 0;
+			path->slots[level] = slot;
+			btrfs_item_key_to_cpu(cur, &found_key, slot);
+			goto out;
+		}
+		if (sret && slot > 0)
+			slot--;
+		/*
+		 * check this node pointer against the cache_only and
+		 * min_trans parameters.  If it isn't in cache or is too
+		 * old, skip to the next one.
+		 */
+		while (slot < nritems) {
+			u64 blockptr;
+			u64 gen;
+			struct extent_buffer *tmp;
+			struct btrfs_disk_key disk_key;
+
+			blockptr = btrfs_node_blockptr(cur, slot);
+			gen = btrfs_node_ptr_generation(cur, slot);
+			if (gen < min_trans) {
+				slot++;
+				continue;
+			}
+			if (!cache_only)
+				break;
+
+			if (max_key) {
+				btrfs_node_key(cur, &disk_key, slot);
+				if (comp_keys(&disk_key, max_key) >= 0) {
+					ret = 1;
+					goto out;
+				}
+			}
+
+			tmp = btrfs_find_tree_block(root, blockptr,
+					    btrfs_level_size(root, level - 1));
+
+			if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
+				free_extent_buffer(tmp);
+				break;
+			}
+			if (tmp)
+				free_extent_buffer(tmp);
+			slot++;
+		}
+find_next_key:
+		/*
+		 * we didn't find a candidate key in this node, walk forward
+		 * and find another one
+		 */
+		if (slot >= nritems) {
+			path->slots[level] = slot;
+			sret = btrfs_find_next_key(root, path, min_key, level,
+						  cache_only, min_trans);
+			if (sret == 0) {
+				btrfs_release_path(root, path);
+				goto again;
+			} else {
+				goto out;
+			}
+		}
+		/* save our key for returning back */
+		btrfs_node_key_to_cpu(cur, &found_key, slot);
+		path->slots[level] = slot;
+		if (level == path->lowest_level) {
+			ret = 0;
+			unlock_up(path, level, 1);
+			goto out;
+		}
+		cur = read_node_slot(root, cur, slot);
+
+		btrfs_tree_lock(cur);
+		path->locks[level - 1] = 1;
+		path->nodes[level - 1] = cur;
+		unlock_up(path, level, 1);
+	}
+out:
+	if (ret == 0)
+		memcpy(min_key, &found_key, sizeof(found_key));
+	return ret;
+}
+
+/*
+ * this is similar to btrfs_next_leaf, but does not try to preserve
+ * and fixup the path.  It looks for and returns the next key in the
+ * tree based on the current path and the cache_only and min_trans
+ * parameters.
+ *
+ * 0 is returned if another key is found, < 0 if there are any errors
+ * and 1 is returned if there are no higher keys in the tree
+ *
+ * path->keep_locks should be set to 1 on the search made before
+ * calling this function.
+ */
+int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
+			struct btrfs_key *key, int lowest_level,
+			int cache_only, u64 min_trans)
+{
+	int level = lowest_level;
+	int slot;
+	struct extent_buffer *c;
+
+	WARN_ON(!path->keep_locks);
+	while (level < BTRFS_MAX_LEVEL) {
+		if (!path->nodes[level])
+			return 1;
+
+		slot = path->slots[level] + 1;
+		c = path->nodes[level];
+next:
+		if (slot >= btrfs_header_nritems(c)) {
+			level++;
+			if (level == BTRFS_MAX_LEVEL)
+				return 1;
+			continue;
+		}
+		if (level == 0)
+			btrfs_item_key_to_cpu(c, key, slot);
+		else {
+			u64 blockptr = btrfs_node_blockptr(c, slot);
+			u64 gen = btrfs_node_ptr_generation(c, slot);
+
+			if (cache_only) {
+				struct extent_buffer *cur;
+				cur = btrfs_find_tree_block(root, blockptr,
+					    btrfs_level_size(root, level - 1));
+				if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
+					slot++;
+					if (cur)
+						free_extent_buffer(cur);
+					goto next;
+				}
+				free_extent_buffer(cur);
+			}
+			if (gen < min_trans) {
+				slot++;
+				goto next;
+			}
+			btrfs_node_key_to_cpu(c, key, slot);
+		}
+		return 0;
+	}
+	return 1;
+}
+
+/*
+ * search the tree again to find a leaf with greater keys
+ * returns 0 if it found something or 1 if there are no greater leaves.
+ * returns < 0 on io errors.
+ */
+int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
+{
+	int slot;
+	int level = 1;
+	struct extent_buffer *c;
+	struct extent_buffer *next = NULL;
+	struct btrfs_key key;
+	u32 nritems;
+	int ret;
+
+	nritems = btrfs_header_nritems(path->nodes[0]);
+	if (nritems == 0)
+		return 1;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
+
+	btrfs_release_path(root, path);
+	path->keep_locks = 1;
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	path->keep_locks = 0;
+
+	if (ret < 0)
+		return ret;
+
+	nritems = btrfs_header_nritems(path->nodes[0]);
+	/*
+	 * by releasing the path above we dropped all our locks.  A balance
+	 * could have added more items next to the key that used to be
+	 * at the very end of the block.  So, check again here and
+	 * advance the path if there are now more items available.
+	 */
+	if (nritems > 0 && path->slots[0] < nritems - 1) {
+		path->slots[0]++;
+		goto done;
+	}
+
+	while (level < BTRFS_MAX_LEVEL) {
+		if (!path->nodes[level])
+			return 1;
+
+		slot = path->slots[level] + 1;
+		c = path->nodes[level];
+		if (slot >= btrfs_header_nritems(c)) {
+			level++;
+			if (level == BTRFS_MAX_LEVEL)
+				return 1;
+			continue;
+		}
+
+		if (next) {
+			btrfs_tree_unlock(next);
+			free_extent_buffer(next);
+		}
+
+		if (level == 1 && (path->locks[1] || path->skip_locking) &&
+		    path->reada)
+			reada_for_search(root, path, level, slot, 0);
+
+		next = read_node_slot(root, c, slot);
+		if (!path->skip_locking) {
+			WARN_ON(!btrfs_tree_locked(c));
+			btrfs_tree_lock(next);
+		}
+		break;
+	}
+	path->slots[level] = slot;
+	while (1) {
+		level--;
+		c = path->nodes[level];
+		if (path->locks[level])
+			btrfs_tree_unlock(c);
+		free_extent_buffer(c);
+		path->nodes[level] = next;
+		path->slots[level] = 0;
+		if (!path->skip_locking)
+			path->locks[level] = 1;
+		if (!level)
+			break;
+		if (level == 1 && path->locks[1] && path->reada)
+			reada_for_search(root, path, level, slot, 0);
+		next = read_node_slot(root, next, 0);
+		if (!path->skip_locking) {
+			WARN_ON(!btrfs_tree_locked(path->nodes[level]));
+			btrfs_tree_lock(next);
+		}
+	}
+done:
+	unlock_up(path, 0, 1);
+	return 0;
+}
+
+/*
+ * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
+ * searching until it gets past min_objectid or finds an item of 'type'
+ *
+ * returns 0 if something is found, 1 if nothing was found and < 0 on error
+ */
+int btrfs_previous_item(struct btrfs_root *root,
+			struct btrfs_path *path, u64 min_objectid,
+			int type)
+{
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+	u32 nritems;
+	int ret;
+
+	while (1) {
+		if (path->slots[0] == 0) {
+			ret = btrfs_prev_leaf(root, path);
+			if (ret != 0)
+				return ret;
+		} else {
+			path->slots[0]--;
+		}
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (nritems == 0)
+			return 1;
+		if (path->slots[0] == nritems)
+			path->slots[0]--;
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.type == type)
+			return 0;
+		if (found_key.objectid < min_objectid)
+			break;
+		if (found_key.objectid == min_objectid &&
+		    found_key.type < type)
+			break;
+	}
+	return 1;
+}

[PATCH 31.1/35] Btrfs extent io operations

[Chris Mason]

extent_io includes two interfaces.  Each one should live in its own
file, but for now they share a single one.

extent_state: These track state bits across a range of bytes instead of
a single page.  So, btrfs can say bytes 0-1MB are marked for delalloc, and
do searches based on the ranges.  An rbtree is used to track state per
inode.

This code also implements readpages and writepages based on the extent_state
code.  No buffer heads are used, even as temporary containers for IO.  This
only works with page sized IO today.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- /dev/null	2008-06-10 15:56:50.000000000 -0400
+++ b/fs/btrfs/extent_io.c	2009-01-08 10:10:26.609514565 -0500
@@ -0,0 +1,2795 @@
+#include <linux/bitops.h>
+#include <linux/slab.h>
+#include <linux/bio.h>
+#include <linux/mm.h>
+#include <linux/gfp.h>
+#include <linux/pagemap.h>
+#include <linux/page-flags.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/blkdev.h>
+#include <linux/swap.h>
+#include <linux/version.h>
+#include <linux/writeback.h>
+#include <linux/pagevec.h>
+#include "extent_io.h"
+#include "extent_map.h"
+#include "compat.h"
+#include "ctree.h"
+#include "btrfs_inode.h"
+
+/* temporary define until extent_map moves out of btrfs */
+struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
+				       unsigned long extra_flags,
+				       void (*ctor)(void *, struct kmem_cache *,
+						    unsigned long));
+
+static struct kmem_cache *extent_state_cache;
+
+static LIST_HEAD(buffers);
+static LIST_HEAD(states);
+
+#define LEAK_DEBUG 0
+#ifdef LEAK_DEBUG
+static DEFINE_SPINLOCK(leak_lock);
+#endif
+
+#define BUFFER_LRU_MAX 64
+
+struct tree_entry {
+	u64 start;
+	u64 end;
+	struct rb_node rb_node;
+};
+
+struct extent_page_data {
+	struct bio *bio;
+	struct extent_io_tree *tree;
+	get_extent_t *get_extent;
+
+	/* tells writepage not to lock the state bits for this range
+	 * it still does the unlocking
+	 */
+	int extent_locked;
+};
+
+int __init extent_io_init(void)
+{
+	extent_state_cache = btrfs_cache_create("extent_state",
+					    sizeof(struct extent_state), 0,
+					    NULL);
+	if (!extent_state_cache)
+		return -ENOMEM;
+
+	return 0;
+
+free_state_cache:
+	kmem_cache_destroy(extent_state_cache);
+	return -ENOMEM;
+}
+
+void extent_io_exit(void)
+{
+	struct extent_state *state;
+
+	while (!list_empty(&states)) {
+		state = list_entry(states.next, struct extent_state, leak_list);
+		printk(KERN_ERR "btrfs state leak: start %llu end %llu "
+		       "state %lu in tree %p refs %d\n",
+		       (unsigned long long)state->start,
+		       (unsigned long long)state->end,
+		       state->state, state->tree, atomic_read(&state->refs));
+		list_del(&state->leak_list);
+		kmem_cache_free(extent_state_cache, state);
+
+	}
+
+	if (extent_state_cache)
+		kmem_cache_destroy(extent_state_cache);
+}
+
+void extent_io_tree_init(struct extent_io_tree *tree,
+			  struct address_space *mapping, gfp_t mask)
+{
+	tree->state.rb_node = NULL;
+	tree->buffer.rb_node = NULL;
+	tree->ops = NULL;
+	tree->dirty_bytes = 0;
+	spin_lock_init(&tree->lock);
+	spin_lock_init(&tree->buffer_lock);
+	tree->mapping = mapping;
+}
+
+static struct extent_state *alloc_extent_state(gfp_t mask)
+{
+	struct extent_state *state;
+#ifdef LEAK_DEBUG
+	unsigned long flags;
+#endif
+
+	state = kmem_cache_alloc(extent_state_cache, mask);
+	if (!state)
+		return state;
+	state->state = 0;
+	state->private = 0;
+	state->tree = NULL;
+#ifdef LEAK_DEBUG
+	spin_lock_irqsave(&leak_lock, flags);
+	list_add(&state->leak_list, &states);
+	spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+	atomic_set(&state->refs, 1);
+	init_waitqueue_head(&state->wq);
+	return state;
+}
+
+static void free_extent_state(struct extent_state *state)
+{
+	if (!state)
+		return;
+	if (atomic_dec_and_test(&state->refs)) {
+#ifdef LEAK_DEBUG
+		unsigned long flags;
+#endif
+		WARN_ON(state->tree);
+#ifdef LEAK_DEBUG
+		spin_lock_irqsave(&leak_lock, flags);
+		list_del(&state->leak_list);
+		spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+		kmem_cache_free(extent_state_cache, state);
+	}
+}
+
+static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
+				   struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct tree_entry *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct tree_entry, rb_node);
+
+		if (offset < entry->start)
+			p = &(*p)->rb_left;
+		else if (offset > entry->end)
+			p = &(*p)->rb_right;
+		else
+			return parent;
+	}
+
+	entry = rb_entry(node, struct tree_entry, rb_node);
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
+				     struct rb_node **prev_ret,
+				     struct rb_node **next_ret)
+{
+	struct rb_root *root = &tree->state;
+	struct rb_node *n = root->rb_node;
+	struct rb_node *prev = NULL;
+	struct rb_node *orig_prev = NULL;
+	struct tree_entry *entry;
+	struct tree_entry *prev_entry = NULL;
+
+	while (n) {
+		entry = rb_entry(n, struct tree_entry, rb_node);
+		prev = n;
+		prev_entry = entry;
+
+		if (offset < entry->start)
+			n = n->rb_left;
+		else if (offset > entry->end)
+			n = n->rb_right;
+		else
+			return n;
+	}
+
+	if (prev_ret) {
+		orig_prev = prev;
+		while (prev && offset > prev_entry->end) {
+			prev = rb_next(prev);
+			prev_entry = rb_entry(prev, struct tree_entry, rb_node);
+		}
+		*prev_ret = prev;
+		prev = orig_prev;
+	}
+
+	if (next_ret) {
+		prev_entry = rb_entry(prev, struct tree_entry, rb_node);
+		while (prev && offset < prev_entry->start) {
+			prev = rb_prev(prev);
+			prev_entry = rb_entry(prev, struct tree_entry, rb_node);
+		}
+		*next_ret = prev;
+	}
+	return NULL;
+}
+
+static inline struct rb_node *tree_search(struct extent_io_tree *tree,
+					  u64 offset)
+{
+	struct rb_node *prev = NULL;
+	struct rb_node *ret;
+
+	ret = __etree_search(tree, offset, &prev, NULL);
+	if (!ret)
+		return prev;
+	return ret;
+}
+
+/*
+ * utility function to look for merge candidates inside a given range.
+ * Any extents with matching state are merged together into a single
+ * extent in the tree.  Extents with EXTENT_IO in their state field
+ * are not merged because the end_io handlers need to be able to do
+ * operations on them without sleeping (or doing allocations/splits).
+ *
+ * This should be called with the tree lock held.
+ */
+static int merge_state(struct extent_io_tree *tree,
+		       struct extent_state *state)
+{
+	struct extent_state *other;
+	struct rb_node *other_node;
+
+	if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
+		return 0;
+
+	other_node = rb_prev(&state->rb_node);
+	if (other_node) {
+		other = rb_entry(other_node, struct extent_state, rb_node);
+		if (other->end == state->start - 1 &&
+		    other->state == state->state) {
+			state->start = other->start;
+			other->tree = NULL;
+			rb_erase(&other->rb_node, &tree->state);
+			free_extent_state(other);
+		}
+	}
+	other_node = rb_next(&state->rb_node);
+	if (other_node) {
+		other = rb_entry(other_node, struct extent_state, rb_node);
+		if (other->start == state->end + 1 &&
+		    other->state == state->state) {
+			other->start = state->start;
+			state->tree = NULL;
+			rb_erase(&state->rb_node, &tree->state);
+			free_extent_state(state);
+		}
+	}
+	return 0;
+}
+
+static void set_state_cb(struct extent_io_tree *tree,
+			 struct extent_state *state,
+			 unsigned long bits)
+{
+	if (tree->ops && tree->ops->set_bit_hook) {
+		tree->ops->set_bit_hook(tree->mapping->host, state->start,
+					state->end, state->state, bits);
+	}
+}
+
+static void clear_state_cb(struct extent_io_tree *tree,
+			   struct extent_state *state,
+			   unsigned long bits)
+{
+	if (tree->ops && tree->ops->clear_bit_hook) {
+		tree->ops->clear_bit_hook(tree->mapping->host, state->start,
+					  state->end, state->state, bits);
+	}
+}
+
+/*
+ * insert an extent_state struct into the tree.  'bits' are set on the
+ * struct before it is inserted.
+ *
+ * This may return -EEXIST if the extent is already there, in which case the
+ * state struct is freed.
+ *
+ * The tree lock is not taken internally.  This is a utility function and
+ * probably isn't what you want to call (see set/clear_extent_bit).
+ */
+static int insert_state(struct extent_io_tree *tree,
+			struct extent_state *state, u64 start, u64 end,
+			int bits)
+{
+	struct rb_node *node;
+
+	if (end < start) {
+		printk(KERN_ERR "btrfs end < start %llu %llu\n",
+		       (unsigned long long)end,
+		       (unsigned long long)start);
+		WARN_ON(1);
+	}
+	if (bits & EXTENT_DIRTY)
+		tree->dirty_bytes += end - start + 1;
+	set_state_cb(tree, state, bits);
+	state->state |= bits;
+	state->start = start;
+	state->end = end;
+	node = tree_insert(&tree->state, end, &state->rb_node);
+	if (node) {
+		struct extent_state *found;
+		found = rb_entry(node, struct extent_state, rb_node);
+		printk(KERN_ERR "btrfs found node %llu %llu on insert of "
+		       "%llu %llu\n", (unsigned long long)found->start,
+		       (unsigned long long)found->end,
+		       (unsigned long long)start, (unsigned long long)end);
+		free_extent_state(state);
+		return -EEXIST;
+	}
+	state->tree = tree;
+	merge_state(tree, state);
+	return 0;
+}
+
+/*
+ * split a given extent state struct in two, inserting the preallocated
+ * struct 'prealloc' as the newly created second half.  'split' indicates an
+ * offset inside 'orig' where it should be split.
+ *
+ * Before calling,
+ * the tree has 'orig' at [orig->start, orig->end].  After calling, there
+ * are two extent state structs in the tree:
+ * prealloc: [orig->start, split - 1]
+ * orig: [ split, orig->end ]
+ *
+ * The tree locks are not taken by this function. They need to be held
+ * by the caller.
+ */
+static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
+		       struct extent_state *prealloc, u64 split)
+{
+	struct rb_node *node;
+	prealloc->start = orig->start;
+	prealloc->end = split - 1;
+	prealloc->state = orig->state;
+	orig->start = split;
+
+	node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
+	if (node) {
+		struct extent_state *found;
+		found = rb_entry(node, struct extent_state, rb_node);
+		free_extent_state(prealloc);
+		return -EEXIST;
+	}
+	prealloc->tree = tree;
+	return 0;
+}
+
+/*
+ * utility function to clear some bits in an extent state struct.
+ * it will optionally wake up any one waiting on this state (wake == 1), or
+ * forcibly remove the state from the tree (delete == 1).
+ *
+ * If no bits are set on the state struct after clearing things, the
+ * struct is freed and removed from the tree
+ */
+static int clear_state_bit(struct extent_io_tree *tree,
+			    struct extent_state *state, int bits, int wake,
+			    int delete)
+{
+	int ret = state->state & bits;
+
+	if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
+		u64 range = state->end - state->start + 1;
+		WARN_ON(range > tree->dirty_bytes);
+		tree->dirty_bytes -= range;
+	}
+	clear_state_cb(tree, state, bits);
+	state->state &= ~bits;
+	if (wake)
+		wake_up(&state->wq);
+	if (delete || state->state == 0) {
+		if (state->tree) {
+			clear_state_cb(tree, state, state->state);
+			rb_erase(&state->rb_node, &tree->state);
+			state->tree = NULL;
+			free_extent_state(state);
+		} else {
+			WARN_ON(1);
+		}
+	} else {
+		merge_state(tree, state);
+	}
+	return ret;
+}
+
+/*
+ * clear some bits on a range in the tree.  This may require splitting
+ * or inserting elements in the tree, so the gfp mask is used to
+ * indicate which allocations or sleeping are allowed.
+ *
+ * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
+ * the given range from the tree regardless of state (ie for truncate).
+ *
+ * the range [start, end] is inclusive.
+ *
+ * This takes the tree lock, and returns < 0 on error, > 0 if any of the
+ * bits were already set, or zero if none of the bits were already set.
+ */
+int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
+		     int bits, int wake, int delete, gfp_t mask)
+{
+	struct extent_state *state;
+	struct extent_state *prealloc = NULL;
+	struct rb_node *node;
+	int err;
+	int set = 0;
+
+again:
+	if (!prealloc && (mask & __GFP_WAIT)) {
+		prealloc = alloc_extent_state(mask);
+		if (!prealloc)
+			return -ENOMEM;
+	}
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find the extents that end after
+	 * our range starts
+	 */
+	node = tree_search(tree, start);
+	if (!node)
+		goto out;
+	state = rb_entry(node, struct extent_state, rb_node);
+	if (state->start > end)
+		goto out;
+	WARN_ON(state->end < start);
+
+	/*
+	 *     | ---- desired range ---- |
+	 *  | state | or
+	 *  | ------------- state -------------- |
+	 *
+	 * We need to split the extent we found, and may flip
+	 * bits on second half.
+	 *
+	 * If the extent we found extends past our range, we
+	 * just split and search again.  It'll get split again
+	 * the next time though.
+	 *
+	 * If the extent we found is inside our range, we clear
+	 * the desired bit on it.
+	 */
+
+	if (state->start < start) {
+		if (!prealloc)
+			prealloc = alloc_extent_state(GFP_ATOMIC);
+		err = split_state(tree, state, prealloc, start);
+		BUG_ON(err == -EEXIST);
+		prealloc = NULL;
+		if (err)
+			goto out;
+		if (state->end <= end) {
+			start = state->end + 1;
+			set |= clear_state_bit(tree, state, bits,
+					wake, delete);
+		} else {
+			start = state->start;
+		}
+		goto search_again;
+	}
+	/*
+	 * | ---- desired range ---- |
+	 *                        | state |
+	 * We need to split the extent, and clear the bit
+	 * on the first half
+	 */
+	if (state->start <= end && state->end > end) {
+		if (!prealloc)
+			prealloc = alloc_extent_state(GFP_ATOMIC);
+		err = split_state(tree, state, prealloc, end + 1);
+		BUG_ON(err == -EEXIST);
+
+		if (wake)
+			wake_up(&state->wq);
+		set |= clear_state_bit(tree, prealloc, bits,
+				       wake, delete);
+		prealloc = NULL;
+		goto out;
+	}
+
+	start = state->end + 1;
+	set |= clear_state_bit(tree, state, bits, wake, delete);
+	goto search_again;
+
+out:
+	spin_unlock(&tree->lock);
+	if (prealloc)
+		free_extent_state(prealloc);
+
+	return set;
+
+search_again:
+	if (start > end)
+		goto out;
+	spin_unlock(&tree->lock);
+	if (mask & __GFP_WAIT)
+		cond_resched();
+	goto again;
+}
+
+static int wait_on_state(struct extent_io_tree *tree,
+			 struct extent_state *state)
+		__releases(tree->lock)
+		__acquires(tree->lock)
+{
+	DEFINE_WAIT(wait);
+	prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
+	spin_unlock(&tree->lock);
+	schedule();
+	spin_lock(&tree->lock);
+	finish_wait(&state->wq, &wait);
+	return 0;
+}
+
+/*
+ * waits for one or more bits to clear on a range in the state tree.
+ * The range [start, end] is inclusive.
+ * The tree lock is taken by this function
+ */
+int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
+{
+	struct extent_state *state;
+	struct rb_node *node;
+
+	spin_lock(&tree->lock);
+again:
+	while (1) {
+		/*
+		 * this search will find all the extents that end after
+		 * our range starts
+		 */
+		node = tree_search(tree, start);
+		if (!node)
+			break;
+
+		state = rb_entry(node, struct extent_state, rb_node);
+
+		if (state->start > end)
+			goto out;
+
+		if (state->state & bits) {
+			start = state->start;
+			atomic_inc(&state->refs);
+			wait_on_state(tree, state);
+			free_extent_state(state);
+			goto again;
+		}
+		start = state->end + 1;
+
+		if (start > end)
+			break;
+
+		if (need_resched()) {
+			spin_unlock(&tree->lock);
+			cond_resched();
+			spin_lock(&tree->lock);
+		}
+	}
+out:
+	spin_unlock(&tree->lock);
+	return 0;
+}
+
+static void set_state_bits(struct extent_io_tree *tree,
+			   struct extent_state *state,
+			   int bits)
+{
+	if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
+		u64 range = state->end - state->start + 1;
+		tree->dirty_bytes += range;
+	}
+	set_state_cb(tree, state, bits);
+	state->state |= bits;
+}
+
+/*
+ * set some bits on a range in the tree.  This may require allocations
+ * or sleeping, so the gfp mask is used to indicate what is allowed.
+ *
+ * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
+ * range already has the desired bits set.  The start of the existing
+ * range is returned in failed_start in this case.
+ *
+ * [start, end] is inclusive
+ * This takes the tree lock.
+ */
+static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
+			  int bits, int exclusive, u64 *failed_start,
+			  gfp_t mask)
+{
+	struct extent_state *state;
+	struct extent_state *prealloc = NULL;
+	struct rb_node *node;
+	int err = 0;
+	int set;
+	u64 last_start;
+	u64 last_end;
+again:
+	if (!prealloc && (mask & __GFP_WAIT)) {
+		prealloc = alloc_extent_state(mask);
+		if (!prealloc)
+			return -ENOMEM;
+	}
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node) {
+		err = insert_state(tree, prealloc, start, end, bits);
+		prealloc = NULL;
+		BUG_ON(err == -EEXIST);
+		goto out;
+	}
+
+	state = rb_entry(node, struct extent_state, rb_node);
+	last_start = state->start;
+	last_end = state->end;
+
+	/*
+	 * | ---- desired range ---- |
+	 * | state |
+	 *
+	 * Just lock what we found and keep going
+	 */
+	if (state->start == start && state->end <= end) {
+		set = state->state & bits;
+		if (set && exclusive) {
+			*failed_start = state->start;
+			err = -EEXIST;
+			goto out;
+		}
+		set_state_bits(tree, state, bits);
+		start = state->end + 1;
+		merge_state(tree, state);
+		goto search_again;
+	}
+
+	/*
+	 *     | ---- desired range ---- |
+	 * | state |
+	 *   or
+	 * | ------------- state -------------- |
+	 *
+	 * We need to split the extent we found, and may flip bits on
+	 * second half.
+	 *
+	 * If the extent we found extends past our
+	 * range, we just split and search again.  It'll get split
+	 * again the next time though.
+	 *
+	 * If the extent we found is inside our range, we set the
+	 * desired bit on it.
+	 */
+	if (state->start < start) {
+		set = state->state & bits;
+		if (exclusive && set) {
+			*failed_start = start;
+			err = -EEXIST;
+			goto out;
+		}
+		err = split_state(tree, state, prealloc, start);
+		BUG_ON(err == -EEXIST);
+		prealloc = NULL;
+		if (err)
+			goto out;
+		if (state->end <= end) {
+			set_state_bits(tree, state, bits);
+			start = state->end + 1;
+			merge_state(tree, state);
+		} else {
+			start = state->start;
+		}
+		goto search_again;
+	}
+	/*
+	 * | ---- desired range ---- |
+	 *     | state | or               | state |
+	 *
+	 * There's a hole, we need to insert something in it and
+	 * ignore the extent we found.
+	 */
+	if (state->start > start) {
+		u64 this_end;
+		if (end < last_start)
+			this_end = end;
+		else
+			this_end = last_start - 1;
+		err = insert_state(tree, prealloc, start, this_end,
+				   bits);
+		prealloc = NULL;
+		BUG_ON(err == -EEXIST);
+		if (err)
+			goto out;
+		start = this_end + 1;
+		goto search_again;
+	}
+	/*
+	 * | ---- desired range ---- |
+	 *                        | state |
+	 * We need to split the extent, and set the bit
+	 * on the first half
+	 */
+	if (state->start <= end && state->end > end) {
+		set = state->state & bits;
+		if (exclusive && set) {
+			*failed_start = start;
+			err = -EEXIST;
+			goto out;
+		}
+		err = split_state(tree, state, prealloc, end + 1);
+		BUG_ON(err == -EEXIST);
+
+		set_state_bits(tree, prealloc, bits);
+		merge_state(tree, prealloc);
+		prealloc = NULL;
+		goto out;
+	}
+
+	goto search_again;
+
+out:
+	spin_unlock(&tree->lock);
+	if (prealloc)
+		free_extent_state(prealloc);
+
+	return err;
+
+search_again:
+	if (start > end)
+		goto out;
+	spin_unlock(&tree->lock);
+	if (mask & __GFP_WAIT)
+		cond_resched();
+	goto again;
+}
+
+/* wrappers around set/clear extent bit */
+int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
+		     gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
+			      mask);
+}
+
+int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
+		       gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
+}
+
+int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
+		    int bits, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, bits, 0, NULL,
+			      mask);
+}
+
+int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
+		      int bits, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
+}
+
+int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
+		     gfp_t mask)
+{
+	return set_extent_bit(tree, start, end,
+			      EXTENT_DELALLOC | EXTENT_DIRTY,
+			      0, NULL, mask);
+}
+
+int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
+		       gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end,
+				EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
+}
+
+int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
+			 gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
+}
+
+int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
+		     gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
+			      mask);
+}
+
+static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
+		       gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
+}
+
+int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
+			gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
+			      mask);
+}
+
+static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
+				 u64 end, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
+}
+
+static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
+			 gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
+			      0, NULL, mask);
+}
+
+static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
+				  u64 end, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
+}
+
+int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
+}
+
+/*
+ * either insert or lock state struct between start and end use mask to tell
+ * us if waiting is desired.
+ */
+int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
+{
+	int err;
+	u64 failed_start;
+	while (1) {
+		err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
+				     &failed_start, mask);
+		if (err == -EEXIST && (mask & __GFP_WAIT)) {
+			wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
+			start = failed_start;
+		} else {
+			break;
+		}
+		WARN_ON(start > end);
+	}
+	return err;
+}
+
+int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
+		    gfp_t mask)
+{
+	int err;
+	u64 failed_start;
+
+	err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
+			     &failed_start, mask);
+	if (err == -EEXIST) {
+		if (failed_start > start)
+			clear_extent_bit(tree, start, failed_start - 1,
+					 EXTENT_LOCKED, 1, 0, mask);
+		return 0;
+	}
+	return 1;
+}
+
+int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
+		  gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
+}
+
+/*
+ * helper function to set pages and extents in the tree dirty
+ */
+int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+
+	while (index <= end_index) {
+		page = find_get_page(tree->mapping, index);
+		BUG_ON(!page);
+		__set_page_dirty_nobuffers(page);
+		page_cache_release(page);
+		index++;
+	}
+	set_extent_dirty(tree, start, end, GFP_NOFS);
+	return 0;
+}
+
+/*
+ * helper function to set both pages and extents in the tree writeback
+ */
+static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+
+	while (index <= end_index) {
+		page = find_get_page(tree->mapping, index);
+		BUG_ON(!page);
+		set_page_writeback(page);
+		page_cache_release(page);
+		index++;
+	}
+	set_extent_writeback(tree, start, end, GFP_NOFS);
+	return 0;
+}
+
+/*
+ * find the first offset in the io tree with 'bits' set. zero is
+ * returned if we find something, and *start_ret and *end_ret are
+ * set to reflect the state struct that was found.
+ *
+ * If nothing was found, 1 is returned, < 0 on error
+ */
+int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
+			  u64 *start_ret, u64 *end_ret, int bits)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	int ret = 1;
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node)
+		goto out;
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->end >= start && (state->state & bits)) {
+			*start_ret = state->start;
+			*end_ret = state->end;
+			ret = 0;
+			break;
+		}
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	spin_unlock(&tree->lock);
+	return ret;
+}
+
+/* find the first state struct with 'bits' set after 'start', and
+ * return it.  tree->lock must be held.  NULL will returned if
+ * nothing was found after 'start'
+ */
+struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
+						 u64 start, int bits)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node)
+		goto out;
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->end >= start && (state->state & bits))
+			return state;
+
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	return NULL;
+}
+
+/*
+ * find a contiguous range of bytes in the file marked as delalloc, not
+ * more than 'max_bytes'.  start and end are used to return the range,
+ *
+ * 1 is returned if we find something, 0 if nothing was in the tree
+ */
+static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
+					u64 *start, u64 *end, u64 max_bytes)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	u64 cur_start = *start;
+	u64 found = 0;
+	u64 total_bytes = 0;
+
+	spin_lock(&tree->lock);
+
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, cur_start);
+	if (!node) {
+		if (!found)
+			*end = (u64)-1;
+		goto out;
+	}
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (found && (state->start != cur_start ||
+			      (state->state & EXTENT_BOUNDARY))) {
+			goto out;
+		}
+		if (!(state->state & EXTENT_DELALLOC)) {
+			if (!found)
+				*end = state->end;
+			goto out;
+		}
+		if (!found)
+			*start = state->start;
+		found++;
+		*end = state->end;
+		cur_start = state->end + 1;
+		node = rb_next(node);
+		if (!node)
+			break;
+		total_bytes += state->end - state->start + 1;
+		if (total_bytes >= max_bytes)
+			break;
+	}
+out:
+	spin_unlock(&tree->lock);
+	return found;
+}
+
+static noinline int __unlock_for_delalloc(struct inode *inode,
+					  struct page *locked_page,
+					  u64 start, u64 end)
+{
+	int ret;
+	struct page *pages[16];
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	unsigned long nr_pages = end_index - index + 1;
+	int i;
+
+	if (index == locked_page->index && end_index == index)
+		return 0;
+
+	while (nr_pages > 0) {
+		ret = find_get_pages_contig(inode->i_mapping, index,
+				     min_t(unsigned long, nr_pages,
+				     ARRAY_SIZE(pages)), pages);
+		for (i = 0; i < ret; i++) {
+			if (pages[i] != locked_page)
+				unlock_page(pages[i]);
+			page_cache_release(pages[i]);
+		}
+		nr_pages -= ret;
+		index += ret;
+		cond_resched();
+	}
+	return 0;
+}
+
+static noinline int lock_delalloc_pages(struct inode *inode,
+					struct page *locked_page,
+					u64 delalloc_start,
+					u64 delalloc_end)
+{
+	unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
+	unsigned long start_index = index;
+	unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
+	unsigned long pages_locked = 0;
+	struct page *pages[16];
+	unsigned long nrpages;
+	int ret;
+	int i;
+
+	/* the caller is responsible for locking the start index */
+	if (index == locked_page->index && index == end_index)
+		return 0;
+
+	/* skip the page at the start index */
+	nrpages = end_index - index + 1;
+	while (nrpages > 0) {
+		ret = find_get_pages_contig(inode->i_mapping, index,
+				     min_t(unsigned long,
+				     nrpages, ARRAY_SIZE(pages)), pages);
+		if (ret == 0) {
+			ret = -EAGAIN;
+			goto done;
+		}
+		/* now we have an array of pages, lock them all */
+		for (i = 0; i < ret; i++) {
+			/*
+			 * the caller is taking responsibility for
+			 * locked_page
+			 */
+			if (pages[i] != locked_page) {
+				lock_page(pages[i]);
+				if (!PageDirty(pages[i]) ||
+				    pages[i]->mapping != inode->i_mapping) {
+					ret = -EAGAIN;
+					unlock_page(pages[i]);
+					page_cache_release(pages[i]);
+					goto done;
+				}
+			}
+			page_cache_release(pages[i]);
+			pages_locked++;
+		}
+		nrpages -= ret;
+		index += ret;
+		cond_resched();
+	}
+	ret = 0;
+done:
+	if (ret && pages_locked) {
+		__unlock_for_delalloc(inode, locked_page,
+			      delalloc_start,
+			      ((u64)(start_index + pages_locked - 1)) <<
+			      PAGE_CACHE_SHIFT);
+	}
+	return ret;
+}
+
+/*
+ * find a contiguous range of bytes in the file marked as delalloc, not
+ * more than 'max_bytes'.  start and end are used to return the range,
+ *
+ * 1 is returned if we find something, 0 if nothing was in the tree
+ */
+static noinline u64 find_lock_delalloc_range(struct inode *inode,
+					     struct extent_io_tree *tree,
+					     struct page *locked_page,
+					     u64 *start, u64 *end,
+					     u64 max_bytes)
+{
+	u64 delalloc_start;
+	u64 delalloc_end;
+	u64 found;
+	int ret;
+	int loops = 0;
+
+again:
+	/* step one, find a bunch of delalloc bytes starting at start */
+	delalloc_start = *start;
+	delalloc_end = 0;
+	found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
+				    max_bytes);
+	if (!found || delalloc_end <= *start) {
+		*start = delalloc_start;
+		*end = delalloc_end;
+		return found;
+	}
+
+	/*
+	 * start comes from the offset of locked_page.  We have to lock
+	 * pages in order, so we can't process delalloc bytes before
+	 * locked_page
+	 */
+	if (delalloc_start < *start)
+		delalloc_start = *start;
+
+	/*
+	 * make sure to limit the number of pages we try to lock down
+	 * if we're looping.
+	 */
+	if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
+		delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
+
+	/* step two, lock all the pages after the page that has start */
+	ret = lock_delalloc_pages(inode, locked_page,
+				  delalloc_start, delalloc_end);
+	if (ret == -EAGAIN) {
+		/* some of the pages are gone, lets avoid looping by
+		 * shortening the size of the delalloc range we're searching
+		 */
+		if (!loops) {
+			unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
+			max_bytes = PAGE_CACHE_SIZE - offset;
+			loops = 1;
+			goto again;
+		} else {
+			found = 0;
+			goto out_failed;
+		}
+	}
+	BUG_ON(ret);
+
+	/* step three, lock the state bits for the whole range */
+	lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
+
+	/* then test to make sure it is all still delalloc */
+	ret = test_range_bit(tree, delalloc_start, delalloc_end,
+			     EXTENT_DELALLOC, 1);
+	if (!ret) {
+		unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
+		__unlock_for_delalloc(inode, locked_page,
+			      delalloc_start, delalloc_end);
+		cond_resched();
+		goto again;
+	}
+	*start = delalloc_start;
+	*end = delalloc_end;
+out_failed:
+	return found;
+}
+
+int extent_clear_unlock_delalloc(struct inode *inode,
+				struct extent_io_tree *tree,
+				u64 start, u64 end, struct page *locked_page,
+				int unlock_pages,
+				int clear_unlock,
+				int clear_delalloc, int clear_dirty,
+				int set_writeback,
+				int end_writeback)
+{
+	int ret;
+	struct page *pages[16];
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	unsigned long nr_pages = end_index - index + 1;
+	int i;
+	int clear_bits = 0;
+
+	if (clear_unlock)
+		clear_bits |= EXTENT_LOCKED;
+	if (clear_dirty)
+		clear_bits |= EXTENT_DIRTY;
+
+	if (clear_delalloc)
+		clear_bits |= EXTENT_DELALLOC;
+
+	clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
+	if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
+		return 0;
+
+	while (nr_pages > 0) {
+		ret = find_get_pages_contig(inode->i_mapping, index,
+				     min_t(unsigned long,
+				     nr_pages, ARRAY_SIZE(pages)), pages);
+		for (i = 0; i < ret; i++) {
+			if (pages[i] == locked_page) {
+				page_cache_release(pages[i]);
+				continue;
+			}
+			if (clear_dirty)
+				clear_page_dirty_for_io(pages[i]);
+			if (set_writeback)
+				set_page_writeback(pages[i]);
+			if (end_writeback)
+				end_page_writeback(pages[i]);
+			if (unlock_pages)
+				unlock_page(pages[i]);
+			page_cache_release(pages[i]);
+		}
+		nr_pages -= ret;
+		index += ret;
+		cond_resched();
+	}
+	return 0;
+}
+
+/*
+ * count the number of bytes in the tree that have a given bit(s)
+ * set.  This can be fairly slow, except for EXTENT_DIRTY which is
+ * cached.  The total number found is returned.
+ */
+u64 count_range_bits(struct extent_io_tree *tree,
+		     u64 *start, u64 search_end, u64 max_bytes,
+		     unsigned long bits)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	u64 cur_start = *start;
+	u64 total_bytes = 0;
+	int found = 0;
+
+	if (search_end <= cur_start) {
+		WARN_ON(1);
+		return 0;
+	}
+
+	spin_lock(&tree->lock);
+	if (cur_start == 0 && bits == EXTENT_DIRTY) {
+		total_bytes = tree->dirty_bytes;
+		goto out;
+	}
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, cur_start);
+	if (!node)
+		goto out;
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->start > search_end)
+			break;
+		if (state->end >= cur_start && (state->state & bits)) {
+			total_bytes += min(search_end, state->end) + 1 -
+				       max(cur_start, state->start);
+			if (total_bytes >= max_bytes)
+				break;
+			if (!found) {
+				*start = state->start;
+				found = 1;
+			}
+		}
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	spin_unlock(&tree->lock);
+	return total_bytes;
+}
+
+#if 0
+/*
+ * helper function to lock both pages and extents in the tree.
+ * pages must be locked first.
+ */
+static int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+	int err;
+
+	while (index <= end_index) {
+		page = grab_cache_page(tree->mapping, index);
+		if (!page) {
+			err = -ENOMEM;
+			goto failed;
+		}
+		if (IS_ERR(page)) {
+			err = PTR_ERR(page);
+			goto failed;
+		}
+		index++;
+	}
+	lock_extent(tree, start, end, GFP_NOFS);
+	return 0;
+
+failed:
+	/*
+	 * we failed above in getting the page at 'index', so we undo here
+	 * up to but not including the page at 'index'
+	 */
+	end_index = index;
+	index = start >> PAGE_CACHE_SHIFT;
+	while (index < end_index) {
+		page = find_get_page(tree->mapping, index);
+		unlock_page(page);
+		page_cache_release(page);
+		index++;
+	}
+	return err;
+}
+
+/*
+ * helper function to unlock both pages and extents in the tree.
+ */
+static int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+
+	while (index <= end_index) {
+		page = find_get_page(tree->mapping, index);
+		unlock_page(page);
+		page_cache_release(page);
+		index++;
+	}
+	unlock_extent(tree, start, end, GFP_NOFS);
+	return 0;
+}
+#endif
+
+/*
+ * set the private field for a given byte offset in the tree.  If there isn't
+ * an extent_state there already, this does nothing.
+ */
+int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	int ret = 0;
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node) {
+		ret = -ENOENT;
+		goto out;
+	}
+	state = rb_entry(node, struct extent_state, rb_node);
+	if (state->start != start) {
+		ret = -ENOENT;
+		goto out;
+	}
+	state->private = private;
+out:
+	spin_unlock(&tree->lock);
+	return ret;
+}
+
+int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	int ret = 0;
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node) {
+		ret = -ENOENT;
+		goto out;
+	}
+	state = rb_entry(node, struct extent_state, rb_node);
+	if (state->start != start) {
+		ret = -ENOENT;
+		goto out;
+	}
+	*private = state->private;
+out:
+	spin_unlock(&tree->lock);
+	return ret;
+}
+
+/*
+ * searches a range in the state tree for a given mask.
+ * If 'filled' == 1, this returns 1 only if every extent in the tree
+ * has the bits set.  Otherwise, 1 is returned if any bit in the
+ * range is found set.
+ */
+int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
+		   int bits, int filled)
+{
+	struct extent_state *state = NULL;
+	struct rb_node *node;
+	int bitset = 0;
+
+	spin_lock(&tree->lock);
+	node = tree_search(tree, start);
+	while (node && start <= end) {
+		state = rb_entry(node, struct extent_state, rb_node);
+
+		if (filled && state->start > start) {
+			bitset = 0;
+			break;
+		}
+
+		if (state->start > end)
+			break;
+
+		if (state->state & bits) {
+			bitset = 1;
+			if (!filled)
+				break;
+		} else if (filled) {
+			bitset = 0;
+			break;
+		}
+		start = state->end + 1;
+		if (start > end)
+			break;
+		node = rb_next(node);
+		if (!node) {
+			if (filled)
+				bitset = 0;
+			break;
+		}
+	}
+	spin_unlock(&tree->lock);
+	return bitset;
+}
+
+/*
+ * helper function to set a given page up to date if all the
+ * extents in the tree for that page are up to date
+ */
+static int check_page_uptodate(struct extent_io_tree *tree,
+			       struct page *page)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
+		SetPageUptodate(page);
+	return 0;
+}
+
+/*
+ * helper function to unlock a page if all the extents in the tree
+ * for that page are unlocked
+ */
+static int check_page_locked(struct extent_io_tree *tree,
+			     struct page *page)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
+		unlock_page(page);
+	return 0;
+}
+
+/*
+ * helper function to end page writeback if all the extents
+ * in the tree for that page are done with writeback
+ */
+static int check_page_writeback(struct extent_io_tree *tree,
+			     struct page *page)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
+		end_page_writeback(page);
+	return 0;
+}
+
+/* lots and lots of room for performance fixes in the end_bio funcs */
+
+/*
+ * after a writepage IO is done, we need to:
+ * clear the uptodate bits on error
+ * clear the writeback bits in the extent tree for this IO
+ * end_page_writeback if the page has no more pending IO
+ *
+ * Scheduling is not allowed, so the extent state tree is expected
+ * to have one and only one object corresponding to this IO.
+ */
+static void end_bio_extent_writepage(struct bio *bio, int err)
+{
+	int uptodate = err == 0;
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct extent_io_tree *tree;
+	u64 start;
+	u64 end;
+	int whole_page;
+	int ret;
+
+	do {
+		struct page *page = bvec->bv_page;
+		tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+		start = ((u64)page->index << PAGE_CACHE_SHIFT) +
+			 bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
+
+		if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
+			whole_page = 1;
+		else
+			whole_page = 0;
+
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+		if (tree->ops && tree->ops->writepage_end_io_hook) {
+			ret = tree->ops->writepage_end_io_hook(page, start,
+						       end, NULL, uptodate);
+			if (ret)
+				uptodate = 0;
+		}
+
+		if (!uptodate && tree->ops &&
+		    tree->ops->writepage_io_failed_hook) {
+			ret = tree->ops->writepage_io_failed_hook(bio, page,
+							 start, end, NULL);
+			if (ret == 0) {
+				uptodate = (err == 0);
+				continue;
+			}
+		}
+
+		if (!uptodate) {
+			clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
+			ClearPageUptodate(page);
+			SetPageError(page);
+		}
+
+		clear_extent_writeback(tree, start, end, GFP_ATOMIC);
+
+		if (whole_page)
+			end_page_writeback(page);
+		else
+			check_page_writeback(tree, page);
+	} while (bvec >= bio->bi_io_vec);
+
+	bio_put(bio);
+}
+
+/*
+ * after a readpage IO is done, we need to:
+ * clear the uptodate bits on error
+ * set the uptodate bits if things worked
+ * set the page up to date if all extents in the tree are uptodate
+ * clear the lock bit in the extent tree
+ * unlock the page if there are no other extents locked for it
+ *
+ * Scheduling is not allowed, so the extent state tree is expected
+ * to have one and only one object corresponding to this IO.
+ */
+static void end_bio_extent_readpage(struct bio *bio, int err)
+{
+	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct extent_io_tree *tree;
+	u64 start;
+	u64 end;
+	int whole_page;
+	int ret;
+
+	if (err)
+		uptodate = 0;
+
+	do {
+		struct page *page = bvec->bv_page;
+		tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+		start = ((u64)page->index << PAGE_CACHE_SHIFT) +
+			bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
+
+		if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
+			whole_page = 1;
+		else
+			whole_page = 0;
+
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+
+		if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
+			ret = tree->ops->readpage_end_io_hook(page, start, end,
+							      NULL);
+			if (ret)
+				uptodate = 0;
+		}
+		if (!uptodate && tree->ops &&
+		    tree->ops->readpage_io_failed_hook) {
+			ret = tree->ops->readpage_io_failed_hook(bio, page,
+							 start, end, NULL);
+			if (ret == 0) {
+				uptodate =
+					test_bit(BIO_UPTODATE, &bio->bi_flags);
+				if (err)
+					uptodate = 0;
+				continue;
+			}
+		}
+
+		if (uptodate) {
+			set_extent_uptodate(tree, start, end,
+					    GFP_ATOMIC);
+		}
+		unlock_extent(tree, start, end, GFP_ATOMIC);
+
+		if (whole_page) {
+			if (uptodate) {
+				SetPageUptodate(page);
+			} else {
+				ClearPageUptodate(page);
+				SetPageError(page);
+			}
+			unlock_page(page);
+		} else {
+			if (uptodate) {
+				check_page_uptodate(tree, page);
+			} else {
+				ClearPageUptodate(page);
+				SetPageError(page);
+			}
+			check_page_locked(tree, page);
+		}
+	} while (bvec >= bio->bi_io_vec);
+
+	bio_put(bio);
+}
+
+/*
+ * IO done from prepare_write is pretty simple, we just unlock
+ * the structs in the extent tree when done, and set the uptodate bits
+ * as appropriate.
+ */
+static void end_bio_extent_preparewrite(struct bio *bio, int err)
+{
+	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct extent_io_tree *tree;
+	u64 start;
+	u64 end;
+
+	do {
+		struct page *page = bvec->bv_page;
+		tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+		start = ((u64)page->index << PAGE_CACHE_SHIFT) +
+			bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
+
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+
+		if (uptodate) {
+			set_extent_uptodate(tree, start, end, GFP_ATOMIC);
+		} else {
+			ClearPageUptodate(page);
+			SetPageError(page);
+		}
+
+		unlock_extent(tree, start, end, GFP_ATOMIC);
+
+	} while (bvec >= bio->bi_io_vec);
+
+	bio_put(bio);
+}
+
+static struct bio *
+extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
+		 gfp_t gfp_flags)
+{
+	struct bio *bio;
+
+	bio = bio_alloc(gfp_flags, nr_vecs);
+
+	if (bio == NULL && (current->flags & PF_MEMALLOC)) {
+		while (!bio && (nr_vecs /= 2))
+			bio = bio_alloc(gfp_flags, nr_vecs);
+	}
+
+	if (bio) {
+		bio->bi_size = 0;
+		bio->bi_bdev = bdev;
+		bio->bi_sector = first_sector;
+	}
+	return bio;
+}
+
+static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
+			  unsigned long bio_flags)
+{
+	int ret = 0;
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct page *page = bvec->bv_page;
+	struct extent_io_tree *tree = bio->bi_private;
+	u64 start;
+	u64 end;
+
+	start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
+	end = start + bvec->bv_len - 1;
+
+	bio->bi_private = NULL;
+
+	bio_get(bio);
+
+	if (tree->ops && tree->ops->submit_bio_hook)
+		tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
+					   mirror_num, bio_flags);
+	else
+		submit_bio(rw, bio);
+	if (bio_flagged(bio, BIO_EOPNOTSUPP))
+		ret = -EOPNOTSUPP;
+	bio_put(bio);
+	return ret;
+}
+
+static int submit_extent_page(int rw, struct extent_io_tree *tree,
+			      struct page *page, sector_t sector,
+			      size_t size, unsigned long offset,
+			      struct block_device *bdev,
+			      struct bio **bio_ret,
+			      unsigned long max_pages,
+			      bio_end_io_t end_io_func,
+			      int mirror_num,
+			      unsigned long prev_bio_flags,
+			      unsigned long bio_flags)
+{
+	int ret = 0;
+	struct bio *bio;
+	int nr;
+	int contig = 0;
+	int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
+	int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
+	size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
+
+	if (bio_ret && *bio_ret) {
+		bio = *bio_ret;
+		if (old_compressed)
+			contig = bio->bi_sector == sector;
+		else
+			contig = bio->bi_sector + (bio->bi_size >> 9) ==
+				sector;
+
+		if (prev_bio_flags != bio_flags || !contig ||
+		    (tree->ops && tree->ops->merge_bio_hook &&
+		     tree->ops->merge_bio_hook(page, offset, page_size, bio,
+					       bio_flags)) ||
+		    bio_add_page(bio, page, page_size, offset) < page_size) {
+			ret = submit_one_bio(rw, bio, mirror_num,
+					     prev_bio_flags);
+			bio = NULL;
+		} else {
+			return 0;
+		}
+	}
+	if (this_compressed)
+		nr = BIO_MAX_PAGES;
+	else
+		nr = bio_get_nr_vecs(bdev);
+
+	bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
+
+	bio_add_page(bio, page, page_size, offset);
+	bio->bi_end_io = end_io_func;
+	bio->bi_private = tree;
+
+	if (bio_ret)
+		*bio_ret = bio;
+	else
+		ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
+
+	return ret;
+}
+
+void set_page_extent_mapped(struct page *page)
+{
+	if (!PagePrivate(page)) {
+		SetPagePrivate(page);
+		page_cache_get(page);
+		set_page_private(page, EXTENT_PAGE_PRIVATE);
+	}
+}
+
+static void set_page_extent_head(struct page *page, unsigned long len)
+{
+	set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
+}
+
+/*
+ * basic readpage implementation.  Locked extent state structs are inserted
+ * into the tree that are removed when the IO is done (by the end_io
+ * handlers)
+ */
+static int __extent_read_full_page(struct extent_io_tree *tree,
+				   struct page *page,
+				   get_extent_t *get_extent,
+				   struct bio **bio, int mirror_num,
+				   unsigned long *bio_flags)
+{
+	struct inode *inode = page->mapping->host;
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 page_end = start + PAGE_CACHE_SIZE - 1;
+	u64 end;
+	u64 cur = start;
+	u64 extent_offset;
+	u64 last_byte = i_size_read(inode);
+	u64 block_start;
+	u64 cur_end;
+	sector_t sector;
+	struct extent_map *em;
+	struct block_device *bdev;
+	int ret;
+	int nr = 0;
+	size_t page_offset = 0;
+	size_t iosize;
+	size_t disk_io_size;
+	size_t blocksize = inode->i_sb->s_blocksize;
+	unsigned long this_bio_flag = 0;
+
+	set_page_extent_mapped(page);
+
+	end = page_end;
+	lock_extent(tree, start, end, GFP_NOFS);
+
+	if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
+		char *userpage;
+		size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
+
+		if (zero_offset) {
+			iosize = PAGE_CACHE_SIZE - zero_offset;
+			userpage = kmap_atomic(page, KM_USER0);
+			memset(userpage + zero_offset, 0, iosize);
+			flush_dcache_page(page);
+			kunmap_atomic(userpage, KM_USER0);
+		}
+	}
+	while (cur <= end) {
+		if (cur >= last_byte) {
+			char *userpage;
+			iosize = PAGE_CACHE_SIZE - page_offset;
+			userpage = kmap_atomic(page, KM_USER0);
+			memset(userpage + page_offset, 0, iosize);
+			flush_dcache_page(page);
+			kunmap_atomic(userpage, KM_USER0);
+			set_extent_uptodate(tree, cur, cur + iosize - 1,
+					    GFP_NOFS);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			break;
+		}
+		em = get_extent(inode, page, page_offset, cur,
+				end - cur + 1, 0);
+		if (IS_ERR(em) || !em) {
+			SetPageError(page);
+			unlock_extent(tree, cur, end, GFP_NOFS);
+			break;
+		}
+		extent_offset = cur - em->start;
+		BUG_ON(extent_map_end(em) <= cur);
+		BUG_ON(end < cur);
+
+		if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
+			this_bio_flag = EXTENT_BIO_COMPRESSED;
+
+		iosize = min(extent_map_end(em) - cur, end - cur + 1);
+		cur_end = min(extent_map_end(em) - 1, end);
+		iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
+		if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
+			disk_io_size = em->block_len;
+			sector = em->block_start >> 9;
+		} else {
+			sector = (em->block_start + extent_offset) >> 9;
+			disk_io_size = iosize;
+		}
+		bdev = em->bdev;
+		block_start = em->block_start;
+		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
+			block_start = EXTENT_MAP_HOLE;
+		free_extent_map(em);
+		em = NULL;
+
+		/* we've found a hole, just zero and go on */
+		if (block_start == EXTENT_MAP_HOLE) {
+			char *userpage;
+			userpage = kmap_atomic(page, KM_USER0);
+			memset(userpage + page_offset, 0, iosize);
+			flush_dcache_page(page);
+			kunmap_atomic(userpage, KM_USER0);
+
+			set_extent_uptodate(tree, cur, cur + iosize - 1,
+					    GFP_NOFS);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			cur = cur + iosize;
+			page_offset += iosize;
+			continue;
+		}
+		/* the get_extent function already copied into the page */
+		if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
+			check_page_uptodate(tree, page);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			cur = cur + iosize;
+			page_offset += iosize;
+			continue;
+		}
+		/* we have an inline extent but it didn't get marked up
+		 * to date.  Error out
+		 */
+		if (block_start == EXTENT_MAP_INLINE) {
+			SetPageError(page);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			cur = cur + iosize;
+			page_offset += iosize;
+			continue;
+		}
+
+		ret = 0;
+		if (tree->ops && tree->ops->readpage_io_hook) {
+			ret = tree->ops->readpage_io_hook(page, cur,
+							  cur + iosize - 1);
+		}
+		if (!ret) {
+			unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
+			pnr -= page->index;
+			ret = submit_extent_page(READ, tree, page,
+					 sector, disk_io_size, page_offset,
+					 bdev, bio, pnr,
+					 end_bio_extent_readpage, mirror_num,
+					 *bio_flags,
+					 this_bio_flag);
+			nr++;
+			*bio_flags = this_bio_flag;
+		}
+		if (ret)
+			SetPageError(page);
+		cur = cur + iosize;
+		page_offset += iosize;
+	}
+	if (!nr) {
+		if (!PageError(page))
+			SetPageUptodate(page);
+		unlock_page(page);
+	}
+	return 0;
+}
+
+int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
+			    get_extent_t *get_extent)
+{
+	struct bio *bio = NULL;
+	unsigned long bio_flags = 0;
+	int ret;
+
+	ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
+				      &bio_flags);
+	if (bio)
+		submit_one_bio(READ, bio, 0, bio_flags);
+	return ret;
+}
+
+/*
+ * the writepage semantics are similar to regular writepage.  extent
+ * records are inserted to lock ranges in the tree, and as dirty areas
+ * are found, they are marked writeback.  Then the lock bits are removed
+ * and the end_io handler clears the writeback ranges
+ */
+static int __extent_writepage(struct page *page, struct writeback_control *wbc,
+			      void *data)
+{
+	struct inode *inode = page->mapping->host;
+	struct extent_page_data *epd = data;
+	struct extent_io_tree *tree = epd->tree;
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 delalloc_start;
+	u64 page_end = start + PAGE_CACHE_SIZE - 1;
+	u64 end;
+	u64 cur = start;
+	u64 extent_offset;
+	u64 last_byte = i_size_read(inode);
+	u64 block_start;
+	u64 iosize;
+	u64 unlock_start;
+	sector_t sector;
+	struct extent_map *em;
+	struct block_device *bdev;
+	int ret;
+	int nr = 0;
+	size_t pg_offset = 0;
+	size_t blocksize;
+	loff_t i_size = i_size_read(inode);
+	unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
+	u64 nr_delalloc;
+	u64 delalloc_end;
+	int page_started;
+	int compressed;
+	unsigned long nr_written = 0;
+
+	WARN_ON(!PageLocked(page));
+	pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
+	if (page->index > end_index ||
+	   (page->index == end_index && !pg_offset)) {
+		page->mapping->a_ops->invalidatepage(page, 0);
+		unlock_page(page);
+		return 0;
+	}
+
+	if (page->index == end_index) {
+		char *userpage;
+
+		userpage = kmap_atomic(page, KM_USER0);
+		memset(userpage + pg_offset, 0,
+		       PAGE_CACHE_SIZE - pg_offset);
+		kunmap_atomic(userpage, KM_USER0);
+		flush_dcache_page(page);
+	}
+	pg_offset = 0;
+
+	set_page_extent_mapped(page);
+
+	delalloc_start = start;
+	delalloc_end = 0;
+	page_started = 0;
+	if (!epd->extent_locked) {
+		while (delalloc_end < page_end) {
+			nr_delalloc = find_lock_delalloc_range(inode, tree,
+						       page,
+						       &delalloc_start,
+						       &delalloc_end,
+						       128 * 1024 * 1024);
+			if (nr_delalloc == 0) {
+				delalloc_start = delalloc_end + 1;
+				continue;
+			}
+			tree->ops->fill_delalloc(inode, page, delalloc_start,
+						 delalloc_end, &page_started,
+						 &nr_written);
+			delalloc_start = delalloc_end + 1;
+		}
+
+		/* did the fill delalloc function already unlock and start
+		 * the IO?
+		 */
+		if (page_started) {
+			ret = 0;
+			goto update_nr_written;
+		}
+	}
+	lock_extent(tree, start, page_end, GFP_NOFS);
+
+	unlock_start = start;
+
+	if (tree->ops && tree->ops->writepage_start_hook) {
+		ret = tree->ops->writepage_start_hook(page, start,
+						      page_end);
+		if (ret == -EAGAIN) {
+			unlock_extent(tree, start, page_end, GFP_NOFS);
+			redirty_page_for_writepage(wbc, page);
+			unlock_page(page);
+			ret = 0;
+			goto update_nr_written;
+		}
+	}
+
+	nr_written++;
+
+	end = page_end;
+	if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
+		printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
+
+	if (last_byte <= start) {
+		clear_extent_dirty(tree, start, page_end, GFP_NOFS);
+		unlock_extent(tree, start, page_end, GFP_NOFS);
+		if (tree->ops && tree->ops->writepage_end_io_hook)
+			tree->ops->writepage_end_io_hook(page, start,
+							 page_end, NULL, 1);
+		unlock_start = page_end + 1;
+		goto done;
+	}
+
+	set_extent_uptodate(tree, start, page_end, GFP_NOFS);
+	blocksize = inode->i_sb->s_blocksize;
+
+	while (cur <= end) {
+		if (cur >= last_byte) {
+			clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
+			unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
+			if (tree->ops && tree->ops->writepage_end_io_hook)
+				tree->ops->writepage_end_io_hook(page, cur,
+							 page_end, NULL, 1);
+			unlock_start = page_end + 1;
+			break;
+		}
+		em = epd->get_extent(inode, page, pg_offset, cur,
+				     end - cur + 1, 1);
+		if (IS_ERR(em) || !em) {
+			SetPageError(page);
+			break;
+		}
+
+		extent_offset = cur - em->start;
+		BUG_ON(extent_map_end(em) <= cur);
+		BUG_ON(end < cur);
+		iosize = min(extent_map_end(em) - cur, end - cur + 1);
+		iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
+		sector = (em->block_start + extent_offset) >> 9;
+		bdev = em->bdev;
+		block_start = em->block_start;
+		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
+		free_extent_map(em);
+		em = NULL;
+
+		/*
+		 * compressed and inline extents are written through other
+		 * paths in the FS
+		 */
+		if (compressed || block_start == EXTENT_MAP_HOLE ||
+		    block_start == EXTENT_MAP_INLINE) {
+			clear_extent_dirty(tree, cur,
+					   cur + iosize - 1, GFP_NOFS);
+
+			unlock_extent(tree, unlock_start, cur + iosize - 1,
+				      GFP_NOFS);
+
+			/*
+			 * end_io notification does not happen here for
+			 * compressed extents
+			 */
+			if (!compressed && tree->ops &&
+			    tree->ops->writepage_end_io_hook)
+				tree->ops->writepage_end_io_hook(page, cur,
+							 cur + iosize - 1,
+							 NULL, 1);
+			else if (compressed) {
+				/* we don't want to end_page_writeback on
+				 * a compressed extent.  this happens
+				 * elsewhere
+				 */
+				nr++;
+			}
+
+			cur += iosize;
+			pg_offset += iosize;
+			unlock_start = cur;
+			continue;
+		}
+		/* leave this out until we have a page_mkwrite call */
+		if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
+				   EXTENT_DIRTY, 0)) {
+			cur = cur + iosize;
+			pg_offset += iosize;
+			continue;
+		}
+
+		clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
+		if (tree->ops && tree->ops->writepage_io_hook) {
+			ret = tree->ops->writepage_io_hook(page, cur,
+						cur + iosize - 1);
+		} else {
+			ret = 0;
+		}
+		if (ret) {
+			SetPageError(page);
+		} else {
+			unsigned long max_nr = end_index + 1;
+
+			set_range_writeback(tree, cur, cur + iosize - 1);
+			if (!PageWriteback(page)) {
+				printk(KERN_ERR "btrfs warning page %lu not "
+				       "writeback, cur %llu end %llu\n",
+				       page->index, (unsigned long long)cur,
+				       (unsigned long long)end);
+			}
+
+			ret = submit_extent_page(WRITE, tree, page, sector,
+						 iosize, pg_offset, bdev,
+						 &epd->bio, max_nr,
+						 end_bio_extent_writepage,
+						 0, 0, 0);
+			if (ret)
+				SetPageError(page);
+		}
+		cur = cur + iosize;
+		pg_offset += iosize;
+		nr++;
+	}
+done:
+	if (nr == 0) {
+		/* make sure the mapping tag for page dirty gets cleared */
+		set_page_writeback(page);
+		end_page_writeback(page);
+	}
+	if (unlock_start <= page_end)
+		unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
+	unlock_page(page);
+
+update_nr_written:
+	wbc->nr_to_write -= nr_written;
+	if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
+	    wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
+		page->mapping->writeback_index = page->index + nr_written;
+	return 0;
+}
+
+/**
+ * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
+ * @mapping: address space structure to write
+ * @wbc: subtract the number of written pages from *@wbc->nr_to_write
+ * @writepage: function called for each page
+ * @data: data passed to writepage function
+ *
+ * If a page is already under I/O, write_cache_pages() skips it, even
+ * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
+ * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
+ * and msync() need to guarantee that all the data which was dirty at the time
+ * the call was made get new I/O started against them.  If wbc->sync_mode is
+ * WB_SYNC_ALL then we were called for data integrity and we must wait for
+ * existing IO to complete.
+ */
+static int extent_write_cache_pages(struct extent_io_tree *tree,
+			     struct address_space *mapping,
+			     struct writeback_control *wbc,
+			     writepage_t writepage, void *data,
+			     void (*flush_fn)(void *))
+{
+	struct backing_dev_info *bdi = mapping->backing_dev_info;
+	int ret = 0;
+	int done = 0;
+	struct pagevec pvec;
+	int nr_pages;
+	pgoff_t index;
+	pgoff_t end;		/* Inclusive */
+	int scanned = 0;
+	int range_whole = 0;
+
+	if (wbc->nonblocking && bdi_write_congested(bdi)) {
+		wbc->encountered_congestion = 1;
+		return 0;
+	}
+
+	pagevec_init(&pvec, 0);
+	if (wbc->range_cyclic) {
+		index = mapping->writeback_index; /* Start from prev offset */
+		end = -1;
+	} else {
+		index = wbc->range_start >> PAGE_CACHE_SHIFT;
+		end = wbc->range_end >> PAGE_CACHE_SHIFT;
+		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+			range_whole = 1;
+		scanned = 1;
+	}
+retry:
+	while (!done && (index <= end) &&
+	       (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
+			      PAGECACHE_TAG_DIRTY, min(end - index,
+				  (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
+		unsigned i;
+
+		scanned = 1;
+		for (i = 0; i < nr_pages; i++) {
+			struct page *page = pvec.pages[i];
+
+			/*
+			 * At this point we hold neither mapping->tree_lock nor
+			 * lock on the page itself: the page may be truncated or
+			 * invalidated (changing page->mapping to NULL), or even
+			 * swizzled back from swapper_space to tmpfs file
+			 * mapping
+			 */
+			if (tree->ops && tree->ops->write_cache_pages_lock_hook)
+				tree->ops->write_cache_pages_lock_hook(page);
+			else
+				lock_page(page);
+
+			if (unlikely(page->mapping != mapping)) {
+				unlock_page(page);
+				continue;
+			}
+
+			if (!wbc->range_cyclic && page->index > end) {
+				done = 1;
+				unlock_page(page);
+				continue;
+			}
+
+			if (wbc->sync_mode != WB_SYNC_NONE) {
+				if (PageWriteback(page))
+					flush_fn(data);
+				wait_on_page_writeback(page);
+			}
+
+			if (PageWriteback(page) ||
+			    !clear_page_dirty_for_io(page)) {
+				unlock_page(page);
+				continue;
+			}
+
+			ret = (*writepage)(page, wbc, data);
+
+			if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
+				unlock_page(page);
+				ret = 0;
+			}
+			if (ret || wbc->nr_to_write <= 0)
+				done = 1;
+			if (wbc->nonblocking && bdi_write_congested(bdi)) {
+				wbc->encountered_congestion = 1;
+				done = 1;
+			}
+		}
+		pagevec_release(&pvec);
+		cond_resched();
+	}
+	if (!scanned && !done) {
+		/*
+		 * We hit the last page and there is more work to be done: wrap
+		 * back to the start of the file
+		 */
+		scanned = 1;
+		index = 0;
+		goto retry;
+	}
+	return ret;
+}
+
+static noinline void flush_write_bio(void *data)
+{
+	struct extent_page_data *epd = data;
+	if (epd->bio) {
+		submit_one_bio(WRITE, epd->bio, 0, 0);
+		epd->bio = NULL;
+	}
+}
+
+int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
+			  get_extent_t *get_extent,
+			  struct writeback_control *wbc)
+{
+	int ret;
+	struct address_space *mapping = page->mapping;
+	struct extent_page_data epd = {
+		.bio = NULL,
+		.tree = tree,
+		.get_extent = get_extent,
+		.extent_locked = 0,
+	};
+	struct writeback_control wbc_writepages = {
+		.bdi		= wbc->bdi,
+		.sync_mode	= WB_SYNC_NONE,
+		.older_than_this = NULL,
+		.nr_to_write	= 64,
+		.range_start	= page_offset(page) + PAGE_CACHE_SIZE,
+		.range_end	= (loff_t)-1,
+	};
+
+
+	ret = __extent_writepage(page, wbc, &epd);
+
+	extent_write_cache_pages(tree, mapping, &wbc_writepages,
+				 __extent_writepage, &epd, flush_write_bio);
+	if (epd.bio)
+		submit_one_bio(WRITE, epd.bio, 0, 0);
+	return ret;
+}
+
+int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
+			      u64 start, u64 end, get_extent_t *get_extent,
+			      int mode)
+{
+	int ret = 0;
+	struct address_space *mapping = inode->i_mapping;
+	struct page *page;
+	unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
+		PAGE_CACHE_SHIFT;
+
+	struct extent_page_data epd = {
+		.bio = NULL,
+		.tree = tree,
+		.get_extent = get_extent,
+		.extent_locked = 1,
+	};
+	struct writeback_control wbc_writepages = {
+		.bdi		= inode->i_mapping->backing_dev_info,
+		.sync_mode	= mode,
+		.older_than_this = NULL,
+		.nr_to_write	= nr_pages * 2,
+		.range_start	= start,
+		.range_end	= end + 1,
+	};
+
+	while (start <= end) {
+		page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
+		if (clear_page_dirty_for_io(page))
+			ret = __extent_writepage(page, &wbc_writepages, &epd);
+		else {
+			if (tree->ops && tree->ops->writepage_end_io_hook)
+				tree->ops->writepage_end_io_hook(page, start,
+						 start + PAGE_CACHE_SIZE - 1,
+						 NULL, 1);
+			unlock_page(page);
+		}
+		page_cache_release(page);
+		start += PAGE_CACHE_SIZE;
+	}
+
+	if (epd.bio)
+		submit_one_bio(WRITE, epd.bio, 0, 0);
+	return ret;
+}
+
+int extent_writepages(struct extent_io_tree *tree,
+		      struct address_space *mapping,
+		      get_extent_t *get_extent,
+		      struct writeback_control *wbc)
+{
+	int ret = 0;
+	struct extent_page_data epd = {
+		.bio = NULL,
+		.tree = tree,
+		.get_extent = get_extent,
+		.extent_locked = 0,
+	};
+
+	ret = extent_write_cache_pages(tree, mapping, wbc,
+				       __extent_writepage, &epd,
+				       flush_write_bio);
+	if (epd.bio)
+		submit_one_bio(WRITE, epd.bio, 0, 0);
+	return ret;
+}
+
+int extent_readpages(struct extent_io_tree *tree,
+		     struct address_space *mapping,
+		     struct list_head *pages, unsigned nr_pages,
+		     get_extent_t get_extent)
+{
+	struct bio *bio = NULL;
+	unsigned page_idx;
+	struct pagevec pvec;
+	unsigned long bio_flags = 0;
+
+	pagevec_init(&pvec, 0);
+	for (page_idx = 0; page_idx < nr_pages; page_idx++) {
+		struct page *page = list_entry(pages->prev, struct page, lru);
+
+		prefetchw(&page->flags);
+		list_del(&page->lru);
+		/*
+		 * what we want to do here is call add_to_page_cache_lru,
+		 * but that isn't exported, so we reproduce it here
+		 */
+		if (!add_to_page_cache(page, mapping,
+					page->index, GFP_KERNEL)) {
+
+			/* open coding of lru_cache_add, also not exported */
+			page_cache_get(page);
+			if (!pagevec_add(&pvec, page))
+				__pagevec_lru_add_file(&pvec);
+			__extent_read_full_page(tree, page, get_extent,
+						&bio, 0, &bio_flags);
+		}
+		page_cache_release(page);
+	}
+	if (pagevec_count(&pvec))
+		__pagevec_lru_add_file(&pvec);
+	BUG_ON(!list_empty(pages));
+	if (bio)
+		submit_one_bio(READ, bio, 0, bio_flags);
+	return 0;
+}
+
+/*
+ * basic invalidatepage code, this waits on any locked or writeback
+ * ranges corresponding to the page, and then deletes any extent state
+ * records from the tree
+ */
+int extent_invalidatepage(struct extent_io_tree *tree,
+			  struct page *page, unsigned long offset)
+{
+	u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	size_t blocksize = page->mapping->host->i_sb->s_blocksize;
+
+	start += (offset + blocksize - 1) & ~(blocksize - 1);
+	if (start > end)
+		return 0;
+
+	lock_extent(tree, start, end, GFP_NOFS);
+	wait_on_extent_writeback(tree, start, end);
+	clear_extent_bit(tree, start, end,
+			 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
+			 1, 1, GFP_NOFS);
+	return 0;
+}
+
+/*
+ * simple commit_write call, set_range_dirty is used to mark both
+ * the pages and the extent records as dirty
+ */
+int extent_commit_write(struct extent_io_tree *tree,
+			struct inode *inode, struct page *page,
+			unsigned from, unsigned to)
+{
+	loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
+
+	set_page_extent_mapped(page);
+	set_page_dirty(page);
+
+	if (pos > inode->i_size) {
+		i_size_write(inode, pos);
+		mark_inode_dirty(inode);
+	}
+	return 0;
+}
+
+int extent_prepare_write(struct extent_io_tree *tree,
+			 struct inode *inode, struct page *page,
+			 unsigned from, unsigned to, get_extent_t *get_extent)
+{
+	u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
+	u64 block_start;
+	u64 orig_block_start;
+	u64 block_end;
+	u64 cur_end;
+	struct extent_map *em;
+	unsigned blocksize = 1 << inode->i_blkbits;
+	size_t page_offset = 0;
+	size_t block_off_start;
+	size_t block_off_end;
+	int err = 0;
+	int iocount = 0;
+	int ret = 0;
+	int isnew;
+
+	set_page_extent_mapped(page);
+
+	block_start = (page_start + from) & ~((u64)blocksize - 1);
+	block_end = (page_start + to - 1) | (blocksize - 1);
+	orig_block_start = block_start;
+
+	lock_extent(tree, page_start, page_end, GFP_NOFS);
+	while (block_start <= block_end) {
+		em = get_extent(inode, page, page_offset, block_start,
+				block_end - block_start + 1, 1);
+		if (IS_ERR(em) || !em)
+			goto err;
+
+		cur_end = min(block_end, extent_map_end(em) - 1);
+		block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
+		block_off_end = block_off_start + blocksize;
+		isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
+
+		if (!PageUptodate(page) && isnew &&
+		    (block_off_end > to || block_off_start < from)) {
+			void *kaddr;
+
+			kaddr = kmap_atomic(page, KM_USER0);
+			if (block_off_end > to)
+				memset(kaddr + to, 0, block_off_end - to);
+			if (block_off_start < from)
+				memset(kaddr + block_off_start, 0,
+				       from - block_off_start);
+			flush_dcache_page(page);
+			kunmap_atomic(kaddr, KM_USER0);
+		}
+		if ((em->block_start != EXTENT_MAP_HOLE &&
+		     em->block_start != EXTENT_MAP_INLINE) &&
+		    !isnew && !PageUptodate(page) &&
+		    (block_off_end > to || block_off_start < from) &&
+		    !test_range_bit(tree, block_start, cur_end,
+				    EXTENT_UPTODATE, 1)) {
+			u64 sector;
+			u64 extent_offset = block_start - em->start;
+			size_t iosize;
+			sector = (em->block_start + extent_offset) >> 9;
+			iosize = (cur_end - block_start + blocksize) &
+				~((u64)blocksize - 1);
+			/*
+			 * we've already got the extent locked, but we
+			 * need to split the state such that our end_bio
+			 * handler can clear the lock.
+			 */
+			set_extent_bit(tree, block_start,
+				       block_start + iosize - 1,
+				       EXTENT_LOCKED, 0, NULL, GFP_NOFS);
+			ret = submit_extent_page(READ, tree, page,
+					 sector, iosize, page_offset, em->bdev,
+					 NULL, 1,
+					 end_bio_extent_preparewrite, 0,
+					 0, 0);
+			iocount++;
+			block_start = block_start + iosize;
+		} else {
+			set_extent_uptodate(tree, block_start, cur_end,
+					    GFP_NOFS);
+			unlock_extent(tree, block_start, cur_end, GFP_NOFS);
+			block_start = cur_end + 1;
+		}
+		page_offset = block_start & (PAGE_CACHE_SIZE - 1);
+		free_extent_map(em);
+	}
+	if (iocount) {
+		wait_extent_bit(tree, orig_block_start,
+				block_end, EXTENT_LOCKED);
+	}
+	check_page_uptodate(tree, page);
+err:
+	/* FIXME, zero out newly allocated blocks on error */
+	return err;
+}
+
+/*
+ * a helper for releasepage, this tests for areas of the page that
+ * are locked or under IO and drops the related state bits if it is safe
+ * to drop the page.
+ */
+int try_release_extent_state(struct extent_map_tree *map,
+			     struct extent_io_tree *tree, struct page *page,
+			     gfp_t mask)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	int ret = 1;
+
+	if (test_range_bit(tree, start, end,
+			   EXTENT_IOBITS | EXTENT_ORDERED, 0))
+		ret = 0;
+	else {
+		if ((mask & GFP_NOFS) == GFP_NOFS)
+			mask = GFP_NOFS;
+		clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
+				 1, 1, mask);
+	}
+	return ret;
+}
+
+/*
+ * a helper for releasepage.  As long as there are no locked extents
+ * in the range corresponding to the page, both state records and extent
+ * map records are removed
+ */
+int try_release_extent_mapping(struct extent_map_tree *map,
+			       struct extent_io_tree *tree, struct page *page,
+			       gfp_t mask)
+{
+	struct extent_map *em;
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+
+	if ((mask & __GFP_WAIT) &&
+	    page->mapping->host->i_size > 16 * 1024 * 1024) {
+		u64 len;
+		while (start <= end) {
+			len = end - start + 1;
+			spin_lock(&map->lock);
+			em = lookup_extent_mapping(map, start, len);
+			if (!em || IS_ERR(em)) {
+				spin_unlock(&map->lock);
+				break;
+			}
+			if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
+			    em->start != start) {
+				spin_unlock(&map->lock);
+				free_extent_map(em);
+				break;
+			}
+			if (!test_range_bit(tree, em->start,
+					    extent_map_end(em) - 1,
+					    EXTENT_LOCKED | EXTENT_WRITEBACK |
+					    EXTENT_ORDERED,
+					    0)) {
+				remove_extent_mapping(map, em);
+				/* once for the rb tree */
+				free_extent_map(em);
+			}
+			start = extent_map_end(em);
+			spin_unlock(&map->lock);
+
+			/* once for us */
+			free_extent_map(em);
+		}
+	}
+	return try_release_extent_state(map, tree, page, mask);
+}
+
+sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
+		get_extent_t *get_extent)
+{
+	struct inode *inode = mapping->host;
+	u64 start = iblock << inode->i_blkbits;
+	sector_t sector = 0;
+	size_t blksize = (1 << inode->i_blkbits);
+	struct extent_map *em;
+
+	lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
+		    GFP_NOFS);
+	em = get_extent(inode, NULL, 0, start, blksize, 0);
+	unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
+		      GFP_NOFS);
+	if (!em || IS_ERR(em))
+		return 0;
+
+	if (em->block_start > EXTENT_MAP_LAST_BYTE)
+		goto out;
+
+	sector = (em->block_start + start - em->start) >> inode->i_blkbits;
+out:
+	free_extent_map(em);
+	return sector;
+}

[PATCH 31.2/35] Btrfs extent_buffer code

[Chris Mason]

Extent buffers provide an API to access pages or collections of pages.  They
are meant to store the btrfs metadata and to handle cases where the block
size is different from the OS page size.

Right now, it only works for the OS page size, but it does work with highmem
pages.

The core usage here is an explicit read and write function call to
access the contents of a tree block.  The contents may be in a highmem page
or may span multiple different pages, this is supposed to do the right thing.

Btrfs defines helper functions to access almost all of the on disk structures
which also handle endian translation.  Most of the btrfs code only knows
about the helpers and does not directly call the extent_buffer read/write
functions.

See ctree.h and struct-funcs.c for more details on the helpers themselves.

This also contains the code for doing IO on extent buffers.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

--- a/fs/btrfs/extent_io.c	2009-01-08 10:10:26.609514565 -0500
+++ b/fs/btrfs/extent_io.c	2009-01-08 10:09:03.239672687 -0500
@@ -25,6 +25,7 @@
 						    unsigned long));
 
 static struct kmem_cache *extent_state_cache;
+static struct kmem_cache *extent_buffer_cache;
 
 static LIST_HEAD(buffers);
 static LIST_HEAD(states);
@@ -61,6 +62,11 @@
 	if (!extent_state_cache)
 		return -ENOMEM;
 
+	extent_buffer_cache = btrfs_cache_create("extent_buffers",
+					    sizeof(struct extent_buffer), 0,
+					    NULL);
+	if (!extent_buffer_cache)
+		goto free_state_cache;
 	return 0;
 
 free_state_cache:
@@ -71,6 +77,7 @@
 void extent_io_exit(void)
 {
 	struct extent_state *state;
+	struct extent_buffer *eb;
 
 	while (!list_empty(&states)) {
 		state = list_entry(states.next, struct extent_state, leak_list);
@@ -84,8 +91,18 @@
 
 	}
 
+	while (!list_empty(&buffers)) {
+		eb = list_entry(buffers.next, struct extent_buffer, leak_list);
+		printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
+		       "refs %d\n", (unsigned long long)eb->start,
+		       eb->len, atomic_read(&eb->refs));
+		list_del(&eb->leak_list);
+		kmem_cache_free(extent_buffer_cache, eb);
+	}
 	if (extent_state_cache)
 		kmem_cache_destroy(extent_state_cache);
+	if (extent_buffer_cache)
+		kmem_cache_destroy(extent_buffer_cache);
 }
 
 void extent_io_tree_init(struct extent_io_tree *tree,
@@ -223,6 +240,50 @@
 	return ret;
 }
 
+static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
+					  u64 offset, struct rb_node *node)
+{
+	struct rb_root *root = &tree->buffer;
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct extent_buffer *eb;
+
+	while (*p) {
+		parent = *p;
+		eb = rb_entry(parent, struct extent_buffer, rb_node);
+
+		if (offset < eb->start)
+			p = &(*p)->rb_left;
+		else if (offset > eb->start)
+			p = &(*p)->rb_right;
+		else
+			return eb;
+	}
+
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
+					   u64 offset)
+{
+	struct rb_root *root = &tree->buffer;
+	struct rb_node *n = root->rb_node;
+	struct extent_buffer *eb;
+
+	while (n) {
+		eb = rb_entry(n, struct extent_buffer, rb_node);
+		if (offset < eb->start)
+			n = n->rb_left;
+		else if (offset > eb->start)
+			n = n->rb_right;
+		else
+			return eb;
+	}
+	return NULL;
+}
+
 /*
  * utility function to look for merge candidates inside a given range.
  * Any extents with matching state are merged together into a single
@@ -2793,3 +2854,864 @@
 	free_extent_map(em);
 	return sector;
 }
+
+static inline struct page *extent_buffer_page(struct extent_buffer *eb,
+					      unsigned long i)
+{
+	struct page *p;
+	struct address_space *mapping;
+
+	if (i == 0)
+		return eb->first_page;
+	i += eb->start >> PAGE_CACHE_SHIFT;
+	mapping = eb->first_page->mapping;
+	if (!mapping)
+		return NULL;
+
+	/*
+	 * extent_buffer_page is only called after pinning the page
+	 * by increasing the reference count.  So we know the page must
+	 * be in the radix tree.
+	 */
+	rcu_read_lock();
+	p = radix_tree_lookup(&mapping->page_tree, i);
+	rcu_read_unlock();
+
+	return p;
+}
+
+static inline unsigned long num_extent_pages(u64 start, u64 len)
+{
+	return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
+		(start >> PAGE_CACHE_SHIFT);
+}
+
+static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
+						   u64 start,
+						   unsigned long len,
+						   gfp_t mask)
+{
+	struct extent_buffer *eb = NULL;
+#ifdef LEAK_DEBUG
+	unsigned long flags;
+#endif
+
+	eb = kmem_cache_zalloc(extent_buffer_cache, mask);
+	eb->start = start;
+	eb->len = len;
+	mutex_init(&eb->mutex);
+#ifdef LEAK_DEBUG
+	spin_lock_irqsave(&leak_lock, flags);
+	list_add(&eb->leak_list, &buffers);
+	spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+	atomic_set(&eb->refs, 1);
+
+	return eb;
+}
+
+static void __free_extent_buffer(struct extent_buffer *eb)
+{
+#ifdef LEAK_DEBUG
+	unsigned long flags;
+	spin_lock_irqsave(&leak_lock, flags);
+	list_del(&eb->leak_list);
+	spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+	kmem_cache_free(extent_buffer_cache, eb);
+}
+
+struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
+					  u64 start, unsigned long len,
+					  struct page *page0,
+					  gfp_t mask)
+{
+	unsigned long num_pages = num_extent_pages(start, len);
+	unsigned long i;
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	struct extent_buffer *eb;
+	struct extent_buffer *exists = NULL;
+	struct page *p;
+	struct address_space *mapping = tree->mapping;
+	int uptodate = 1;
+
+	spin_lock(&tree->buffer_lock);
+	eb = buffer_search(tree, start);
+	if (eb) {
+		atomic_inc(&eb->refs);
+		spin_unlock(&tree->buffer_lock);
+		mark_page_accessed(eb->first_page);
+		return eb;
+	}
+	spin_unlock(&tree->buffer_lock);
+
+	eb = __alloc_extent_buffer(tree, start, len, mask);
+	if (!eb)
+		return NULL;
+
+	if (page0) {
+		eb->first_page = page0;
+		i = 1;
+		index++;
+		page_cache_get(page0);
+		mark_page_accessed(page0);
+		set_page_extent_mapped(page0);
+		set_page_extent_head(page0, len);
+		uptodate = PageUptodate(page0);
+	} else {
+		i = 0;
+	}
+	for (; i < num_pages; i++, index++) {
+		p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
+		if (!p) {
+			WARN_ON(1);
+			goto free_eb;
+		}
+		set_page_extent_mapped(p);
+		mark_page_accessed(p);
+		if (i == 0) {
+			eb->first_page = p;
+			set_page_extent_head(p, len);
+		} else {
+			set_page_private(p, EXTENT_PAGE_PRIVATE);
+		}
+		if (!PageUptodate(p))
+			uptodate = 0;
+		unlock_page(p);
+	}
+	if (uptodate)
+		eb->flags |= EXTENT_UPTODATE;
+	eb->flags |= EXTENT_BUFFER_FILLED;
+
+	spin_lock(&tree->buffer_lock);
+	exists = buffer_tree_insert(tree, start, &eb->rb_node);
+	if (exists) {
+		/* add one reference for the caller */
+		atomic_inc(&exists->refs);
+		spin_unlock(&tree->buffer_lock);
+		goto free_eb;
+	}
+	spin_unlock(&tree->buffer_lock);
+
+	/* add one reference for the tree */
+	atomic_inc(&eb->refs);
+	return eb;
+
+free_eb:
+	if (!atomic_dec_and_test(&eb->refs))
+		return exists;
+	for (index = 1; index < i; index++)
+		page_cache_release(extent_buffer_page(eb, index));
+	page_cache_release(extent_buffer_page(eb, 0));
+	__free_extent_buffer(eb);
+	return exists;
+}
+
+struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
+					 u64 start, unsigned long len,
+					  gfp_t mask)
+{
+	struct extent_buffer *eb;
+
+	spin_lock(&tree->buffer_lock);
+	eb = buffer_search(tree, start);
+	if (eb)
+		atomic_inc(&eb->refs);
+	spin_unlock(&tree->buffer_lock);
+
+	if (eb)
+		mark_page_accessed(eb->first_page);
+
+	return eb;
+}
+
+void free_extent_buffer(struct extent_buffer *eb)
+{
+	if (!eb)
+		return;
+
+	if (!atomic_dec_and_test(&eb->refs))
+		return;
+
+	WARN_ON(1);
+}
+
+int clear_extent_buffer_dirty(struct extent_io_tree *tree,
+			      struct extent_buffer *eb)
+{
+	int set;
+	unsigned long i;
+	unsigned long num_pages;
+	struct page *page;
+
+	u64 start = eb->start;
+	u64 end = start + eb->len - 1;
+
+	set = clear_extent_dirty(tree, start, end, GFP_NOFS);
+	num_pages = num_extent_pages(eb->start, eb->len);
+
+	for (i = 0; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (!set && !PageDirty(page))
+			continue;
+
+		lock_page(page);
+		if (i == 0)
+			set_page_extent_head(page, eb->len);
+		else
+			set_page_private(page, EXTENT_PAGE_PRIVATE);
+
+		/*
+		 * if we're on the last page or the first page and the
+		 * block isn't aligned on a page boundary, do extra checks
+		 * to make sure we don't clean page that is partially dirty
+		 */
+		if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
+		    ((i == num_pages - 1) &&
+		     ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
+			start = (u64)page->index << PAGE_CACHE_SHIFT;
+			end  = start + PAGE_CACHE_SIZE - 1;
+			if (test_range_bit(tree, start, end,
+					   EXTENT_DIRTY, 0)) {
+				unlock_page(page);
+				continue;
+			}
+		}
+		clear_page_dirty_for_io(page);
+		spin_lock_irq(&page->mapping->tree_lock);
+		if (!PageDirty(page)) {
+			radix_tree_tag_clear(&page->mapping->page_tree,
+						page_index(page),
+						PAGECACHE_TAG_DIRTY);
+		}
+		spin_unlock_irq(&page->mapping->tree_lock);
+		unlock_page(page);
+	}
+	return 0;
+}
+
+int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
+				    struct extent_buffer *eb)
+{
+	return wait_on_extent_writeback(tree, eb->start,
+					eb->start + eb->len - 1);
+}
+
+int set_extent_buffer_dirty(struct extent_io_tree *tree,
+			     struct extent_buffer *eb)
+{
+	unsigned long i;
+	unsigned long num_pages;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	for (i = 0; i < num_pages; i++) {
+		struct page *page = extent_buffer_page(eb, i);
+		/* writepage may need to do something special for the
+		 * first page, we have to make sure page->private is
+		 * properly set.  releasepage may drop page->private
+		 * on us if the page isn't already dirty.
+		 */
+		lock_page(page);
+		if (i == 0) {
+			set_page_extent_head(page, eb->len);
+		} else if (PagePrivate(page) &&
+			   page->private != EXTENT_PAGE_PRIVATE) {
+			set_page_extent_mapped(page);
+		}
+		__set_page_dirty_nobuffers(extent_buffer_page(eb, i));
+		set_extent_dirty(tree, page_offset(page),
+				 page_offset(page) + PAGE_CACHE_SIZE - 1,
+				 GFP_NOFS);
+		unlock_page(page);
+	}
+	return 0;
+}
+
+int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
+				struct extent_buffer *eb)
+{
+	unsigned long i;
+	struct page *page;
+	unsigned long num_pages;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	eb->flags &= ~EXTENT_UPTODATE;
+
+	clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
+			      GFP_NOFS);
+	for (i = 0; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (page)
+			ClearPageUptodate(page);
+	}
+	return 0;
+}
+
+int set_extent_buffer_uptodate(struct extent_io_tree *tree,
+				struct extent_buffer *eb)
+{
+	unsigned long i;
+	struct page *page;
+	unsigned long num_pages;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+
+	set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
+			    GFP_NOFS);
+	for (i = 0; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
+		    ((i == num_pages - 1) &&
+		     ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
+			check_page_uptodate(tree, page);
+			continue;
+		}
+		SetPageUptodate(page);
+	}
+	return 0;
+}
+
+int extent_range_uptodate(struct extent_io_tree *tree,
+			  u64 start, u64 end)
+{
+	struct page *page;
+	int ret;
+	int pg_uptodate = 1;
+	int uptodate;
+	unsigned long index;
+
+	ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
+	if (ret)
+		return 1;
+	while (start <= end) {
+		index = start >> PAGE_CACHE_SHIFT;
+		page = find_get_page(tree->mapping, index);
+		uptodate = PageUptodate(page);
+		page_cache_release(page);
+		if (!uptodate) {
+			pg_uptodate = 0;
+			break;
+		}
+		start += PAGE_CACHE_SIZE;
+	}
+	return pg_uptodate;
+}
+
+int extent_buffer_uptodate(struct extent_io_tree *tree,
+			   struct extent_buffer *eb)
+{
+	int ret = 0;
+	unsigned long num_pages;
+	unsigned long i;
+	struct page *page;
+	int pg_uptodate = 1;
+
+	if (eb->flags & EXTENT_UPTODATE)
+		return 1;
+
+	ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
+			   EXTENT_UPTODATE, 1);
+	if (ret)
+		return ret;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	for (i = 0; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (!PageUptodate(page)) {
+			pg_uptodate = 0;
+			break;
+		}
+	}
+	return pg_uptodate;
+}
+
+int read_extent_buffer_pages(struct extent_io_tree *tree,
+			     struct extent_buffer *eb,
+			     u64 start, int wait,
+			     get_extent_t *get_extent, int mirror_num)
+{
+	unsigned long i;
+	unsigned long start_i;
+	struct page *page;
+	int err;
+	int ret = 0;
+	int locked_pages = 0;
+	int all_uptodate = 1;
+	int inc_all_pages = 0;
+	unsigned long num_pages;
+	struct bio *bio = NULL;
+	unsigned long bio_flags = 0;
+
+	if (eb->flags & EXTENT_UPTODATE)
+		return 0;
+
+	if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
+			   EXTENT_UPTODATE, 1)) {
+		return 0;
+	}
+
+	if (start) {
+		WARN_ON(start < eb->start);
+		start_i = (start >> PAGE_CACHE_SHIFT) -
+			(eb->start >> PAGE_CACHE_SHIFT);
+	} else {
+		start_i = 0;
+	}
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	for (i = start_i; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (!wait) {
+			if (!trylock_page(page))
+				goto unlock_exit;
+		} else {
+			lock_page(page);
+		}
+		locked_pages++;
+		if (!PageUptodate(page))
+			all_uptodate = 0;
+	}
+	if (all_uptodate) {
+		if (start_i == 0)
+			eb->flags |= EXTENT_UPTODATE;
+		goto unlock_exit;
+	}
+
+	for (i = start_i; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (inc_all_pages)
+			page_cache_get(page);
+		if (!PageUptodate(page)) {
+			if (start_i == 0)
+				inc_all_pages = 1;
+			ClearPageError(page);
+			err = __extent_read_full_page(tree, page,
+						      get_extent, &bio,
+						      mirror_num, &bio_flags);
+			if (err)
+				ret = err;
+		} else {
+			unlock_page(page);
+		}
+	}
+
+	if (bio)
+		submit_one_bio(READ, bio, mirror_num, bio_flags);
+
+	if (ret || !wait)
+		return ret;
+
+	for (i = start_i; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		wait_on_page_locked(page);
+		if (!PageUptodate(page))
+			ret = -EIO;
+	}
+
+	if (!ret)
+		eb->flags |= EXTENT_UPTODATE;
+	return ret;
+
+unlock_exit:
+	i = start_i;
+	while (locked_pages > 0) {
+		page = extent_buffer_page(eb, i);
+		i++;
+		unlock_page(page);
+		locked_pages--;
+	}
+	return ret;
+}
+
+void read_extent_buffer(struct extent_buffer *eb, void *dstv,
+			unsigned long start,
+			unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	char *dst = (char *)dstv;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+
+		cur = min(len, (PAGE_CACHE_SIZE - offset));
+		kaddr = kmap_atomic(page, KM_USER1);
+		memcpy(dst, kaddr + offset, cur);
+		kunmap_atomic(kaddr, KM_USER1);
+
+		dst += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
+			       unsigned long min_len, char **token, char **map,
+			       unsigned long *map_start,
+			       unsigned long *map_len, int km)
+{
+	size_t offset = start & (PAGE_CACHE_SIZE - 1);
+	char *kaddr;
+	struct page *p;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+	unsigned long end_i = (start_offset + start + min_len - 1) >>
+		PAGE_CACHE_SHIFT;
+
+	if (i != end_i)
+		return -EINVAL;
+
+	if (i == 0) {
+		offset = start_offset;
+		*map_start = 0;
+	} else {
+		offset = 0;
+		*map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
+	}
+
+	if (start + min_len > eb->len) {
+		printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
+		       "wanted %lu %lu\n", (unsigned long long)eb->start,
+		       eb->len, start, min_len);
+		WARN_ON(1);
+	}
+
+	p = extent_buffer_page(eb, i);
+	kaddr = kmap_atomic(p, km);
+	*token = kaddr;
+	*map = kaddr + offset;
+	*map_len = PAGE_CACHE_SIZE - offset;
+	return 0;
+}
+
+int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
+		      unsigned long min_len,
+		      char **token, char **map,
+		      unsigned long *map_start,
+		      unsigned long *map_len, int km)
+{
+	int err;
+	int save = 0;
+	if (eb->map_token) {
+		unmap_extent_buffer(eb, eb->map_token, km);
+		eb->map_token = NULL;
+		save = 1;
+		WARN_ON(!mutex_is_locked(&eb->mutex));
+	}
+	err = map_private_extent_buffer(eb, start, min_len, token, map,
+				       map_start, map_len, km);
+	if (!err && save) {
+		eb->map_token = *token;
+		eb->kaddr = *map;
+		eb->map_start = *map_start;
+		eb->map_len = *map_len;
+	}
+	return err;
+}
+
+void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
+{
+	kunmap_atomic(token, km);
+}
+
+int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
+			  unsigned long start,
+			  unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	char *ptr = (char *)ptrv;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+	int ret = 0;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+
+		cur = min(len, (PAGE_CACHE_SIZE - offset));
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		ret = memcmp(ptr, kaddr + offset, cur);
+		kunmap_atomic(kaddr, KM_USER0);
+		if (ret)
+			break;
+
+		ptr += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+	return ret;
+}
+
+void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
+			 unsigned long start, unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	char *src = (char *)srcv;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+		WARN_ON(!PageUptodate(page));
+
+		cur = min(len, PAGE_CACHE_SIZE - offset);
+		kaddr = kmap_atomic(page, KM_USER1);
+		memcpy(kaddr + offset, src, cur);
+		kunmap_atomic(kaddr, KM_USER1);
+
+		src += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+void memset_extent_buffer(struct extent_buffer *eb, char c,
+			  unsigned long start, unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+		WARN_ON(!PageUptodate(page));
+
+		cur = min(len, PAGE_CACHE_SIZE - offset);
+		kaddr = kmap_atomic(page, KM_USER0);
+		memset(kaddr + offset, c, cur);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
+			unsigned long dst_offset, unsigned long src_offset,
+			unsigned long len)
+{
+	u64 dst_len = dst->len;
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(src->len != dst_len);
+
+	offset = (start_offset + dst_offset) &
+		((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(dst, i);
+		WARN_ON(!PageUptodate(page));
+
+		cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		read_extent_buffer(src, kaddr + offset, src_offset, cur);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		src_offset += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+static void move_pages(struct page *dst_page, struct page *src_page,
+		       unsigned long dst_off, unsigned long src_off,
+		       unsigned long len)
+{
+	char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
+	if (dst_page == src_page) {
+		memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
+	} else {
+		char *src_kaddr = kmap_atomic(src_page, KM_USER1);
+		char *p = dst_kaddr + dst_off + len;
+		char *s = src_kaddr + src_off + len;
+
+		while (len--)
+			*--p = *--s;
+
+		kunmap_atomic(src_kaddr, KM_USER1);
+	}
+	kunmap_atomic(dst_kaddr, KM_USER0);
+}
+
+static void copy_pages(struct page *dst_page, struct page *src_page,
+		       unsigned long dst_off, unsigned long src_off,
+		       unsigned long len)
+{
+	char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
+	char *src_kaddr;
+
+	if (dst_page != src_page)
+		src_kaddr = kmap_atomic(src_page, KM_USER1);
+	else
+		src_kaddr = dst_kaddr;
+
+	memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
+	kunmap_atomic(dst_kaddr, KM_USER0);
+	if (dst_page != src_page)
+		kunmap_atomic(src_kaddr, KM_USER1);
+}
+
+void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
+			   unsigned long src_offset, unsigned long len)
+{
+	size_t cur;
+	size_t dst_off_in_page;
+	size_t src_off_in_page;
+	size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long dst_i;
+	unsigned long src_i;
+
+	if (src_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
+		       "len %lu dst len %lu\n", src_offset, len, dst->len);
+		BUG_ON(1);
+	}
+	if (dst_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
+		       "len %lu dst len %lu\n", dst_offset, len, dst->len);
+		BUG_ON(1);
+	}
+
+	while (len > 0) {
+		dst_off_in_page = (start_offset + dst_offset) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+		src_off_in_page = (start_offset + src_offset) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+
+		dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
+		src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
+
+		cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
+					       src_off_in_page));
+		cur = min_t(unsigned long, cur,
+			(unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
+
+		copy_pages(extent_buffer_page(dst, dst_i),
+			   extent_buffer_page(dst, src_i),
+			   dst_off_in_page, src_off_in_page, cur);
+
+		src_offset += cur;
+		dst_offset += cur;
+		len -= cur;
+	}
+}
+
+void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
+			   unsigned long src_offset, unsigned long len)
+{
+	size_t cur;
+	size_t dst_off_in_page;
+	size_t src_off_in_page;
+	unsigned long dst_end = dst_offset + len - 1;
+	unsigned long src_end = src_offset + len - 1;
+	size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long dst_i;
+	unsigned long src_i;
+
+	if (src_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
+		       "len %lu len %lu\n", src_offset, len, dst->len);
+		BUG_ON(1);
+	}
+	if (dst_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
+		       "len %lu len %lu\n", dst_offset, len, dst->len);
+		BUG_ON(1);
+	}
+	if (dst_offset < src_offset) {
+		memcpy_extent_buffer(dst, dst_offset, src_offset, len);
+		return;
+	}
+	while (len > 0) {
+		dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
+		src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
+
+		dst_off_in_page = (start_offset + dst_end) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+		src_off_in_page = (start_offset + src_end) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+
+		cur = min_t(unsigned long, len, src_off_in_page + 1);
+		cur = min(cur, dst_off_in_page + 1);
+		move_pages(extent_buffer_page(dst, dst_i),
+			   extent_buffer_page(dst, src_i),
+			   dst_off_in_page - cur + 1,
+			   src_off_in_page - cur + 1, cur);
+
+		dst_end -= cur;
+		src_end -= cur;
+		len -= cur;
+	}
+}
+
+int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
+{
+	u64 start = page_offset(page);
+	struct extent_buffer *eb;
+	int ret = 1;
+	unsigned long i;
+	unsigned long num_pages;
+
+	spin_lock(&tree->buffer_lock);
+	eb = buffer_search(tree, start);
+	if (!eb)
+		goto out;
+
+	if (atomic_read(&eb->refs) > 1) {
+		ret = 0;
+		goto out;
+	}
+	/* at this point we can safely release the extent buffer */
+	num_pages = num_extent_pages(eb->start, eb->len);
+	for (i = 0; i < num_pages; i++)
+		page_cache_release(extent_buffer_page(eb, i));
+	rb_erase(&eb->rb_node, &tree->buffer);
+	__free_extent_buffer(eb);
+out:
+	spin_unlock(&tree->buffer_lock);
+	return ret;
+}

Re: [PATCH 00/35] Btrfs for review

[Andi Kleen]

Chris Mason <chris.mason@oracle.com> writes:
>
> For actually running the sources, downloading from the git repo will
> work better, but hopefully this will be easier for people to review
> and ask questions about the code.

Could you perhaps put the descriptions you wrote for each .c file patch
as comments into the respective source files? 

That would address one of the items in my earlier list.

-Andi
-- 
ak@linux.intel.com

Re: [PATCH 00/35] Btrfs for review

[Chris Mason]

On Fri, 2009-01-09 at 03:28 +0100, Andi Kleen wrote:
> Chris Mason <chris.mason@oracle.com> writes:
> >
> > For actually running the sources, downloading from the git repo will
> > work better, but hopefully this will be easier for people to review
> > and ask questions about the code.
> 
> Could you perhaps put the descriptions you wrote for each .c file patch
> as comments into the respective source files? 
> 
> That would address one of the items in my earlier list.

Good point, will do.

-chris