[RFC 2/5] implement metadata_incore in btrfs

[RFC 2/5] implement metadata_incore in btrfs

[Shaohua Li]

Implement btrfs specific .metadata_incore.
In btrfs, all metadata pages are in a special btree_inode, we take pages from it.
we only account updated and referenced pages here. Say we collect metadata info
in one boot, do metadata readahead in next boot and we might collect metadata
again. The readahead could read garbage data in as metadata could be changed
from first run. If we only account updated pages, the metadata info collected
by userspace will increase every run. Btrfs alloc_extent_buffer will do
mark_page_accessed() for pages which will be used soon, so we could use
referenced bit to filter some garbage pages.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>

---
 fs/btrfs/super.c |   48 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 48 insertions(+)

Index: linux/fs/btrfs/super.c
===================================================================
--- linux.orig/fs/btrfs/super.c	2010-12-07 10:10:20.000000000 +0800
+++ linux/fs/btrfs/super.c	2010-12-07 13:25:20.000000000 +0800
@@ -39,6 +39,7 @@
 #include <linux/miscdevice.h>
 #include <linux/magic.h>
 #include <linux/slab.h>
+#include <linux/pagevec.h>
 #include "compat.h"
 #include "ctree.h"
 #include "disk-io.h"
@@ -845,6 +846,52 @@ static int btrfs_unfreeze(struct super_b
 	return 0;
 }
 
+static int btrfs_metadata_incore(struct super_block *sb, loff_t *offset,
+	ssize_t *size)
+{
+	struct btrfs_root *tree_root = btrfs_sb(sb);
+	struct inode *btree_inode = tree_root->fs_info->btree_inode;
+	struct pagevec pvec;
+	loff_t index = (*offset) >> PAGE_CACHE_SHIFT;
+	int i, nr_pages;
+
+	*size = 0;
+retry:
+	pagevec_init(&pvec, 0);
+	nr_pages = pagevec_lookup(&pvec, btree_inode->i_mapping, index,
+		PAGEVEC_SIZE);
+	if (nr_pages == 0)
+		goto out;
+	for (i = 0; i < nr_pages; i++) {
+		struct page *page = pvec.pages[i];
+
+		/* Only take pages with 'referenced' bit set */
+		if (PageUptodate(page) && PageReferenced(page)) {
+			if (*size == 0) {
+				*size += PAGE_CACHE_SIZE;
+				*offset = page->index << PAGE_CACHE_SHIFT;
+				continue;
+			}
+			if (page->index !=
+			    (*offset + *size) >> PAGE_CACHE_SHIFT)
+				break;
+			*size += PAGE_CACHE_SIZE;
+		} else if (*size > 0)
+			break;
+		else
+			index = page->index + 1;
+	}
+	pagevec_release(&pvec);
+
+	if (nr_pages > 0 && *size == 0)
+		goto retry;
+out:
+	if (*size > 0)
+		return 0;
+	else
+		return -ENOENT;
+}
+
 static const struct super_operations btrfs_super_ops = {
 	.drop_inode	= btrfs_drop_inode,
 	.evict_inode	= btrfs_evict_inode,
@@ -859,6 +906,7 @@ static const struct super_operations btr
 	.remount_fs	= btrfs_remount,
 	.freeze_fs	= btrfs_freeze,
 	.unfreeze_fs	= btrfs_unfreeze,
+	.metadata_incore = btrfs_metadata_incore,
 };
 
 static const struct file_operations btrfs_ctl_fops = {

Re: [RFC 2/5] implement metadata_incore in btrfs

[Andrew Morton]

On Mon, 13 Dec 2010 15:22:11 +0800
Shaohua Li <shaohua.li@intel.com> wrote:

> Implement btrfs specific .metadata_incore.
> In btrfs, all metadata pages are in a special btree_inode, we take pages from it.
> we only account updated and referenced pages here. Say we collect metadata info
> in one boot, do metadata readahead in next boot and we might collect metadata
> again. The readahead could read garbage data in as metadata could be changed
> from first run. If we only account updated pages, the metadata info collected
> by userspace will increase every run. Btrfs alloc_extent_buffer will do
> mark_page_accessed() for pages which will be used soon, so we could use
> referenced bit to filter some garbage pages.
> 
> Signed-off-by: Shaohua Li <shaohua.li@intel.com>
> 
> ---
>  fs/btrfs/super.c |   48 ++++++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 48 insertions(+)
> 
> Index: linux/fs/btrfs/super.c
> ===================================================================
> --- linux.orig/fs/btrfs/super.c	2010-12-07 10:10:20.000000000 +0800
> +++ linux/fs/btrfs/super.c	2010-12-07 13:25:20.000000000 +0800
> @@ -39,6 +39,7 @@
>  #include <linux/miscdevice.h>
>  #include <linux/magic.h>
>  #include <linux/slab.h>
> +#include <linux/pagevec.h>
>  #include "compat.h"
>  #include "ctree.h"
>  #include "disk-io.h"
> @@ -845,6 +846,52 @@ static int btrfs_unfreeze(struct super_b
>  	return 0;
>  }
>  
> +static int btrfs_metadata_incore(struct super_block *sb, loff_t *offset,
> +	ssize_t *size)
> +{
> +	struct btrfs_root *tree_root = btrfs_sb(sb);
> +	struct inode *btree_inode = tree_root->fs_info->btree_inode;
> +	struct pagevec pvec;
> +	loff_t index = (*offset) >> PAGE_CACHE_SHIFT;

pgoff_t would be a more appropriate type.

> +	int i, nr_pages;
> +
> +	*size = 0;
> +retry:
> +	pagevec_init(&pvec, 0);
> +	nr_pages = pagevec_lookup(&pvec, btree_inode->i_mapping, index,
> +		PAGEVEC_SIZE);
> +	if (nr_pages == 0)
> +		goto out;
> +	for (i = 0; i < nr_pages; i++) {
> +		struct page *page = pvec.pages[i];
> +
> +		/* Only take pages with 'referenced' bit set */

The comment describes the utterly obvious and doesn't explain the
utterly unobvious: "why?".

> +		if (PageUptodate(page) && PageReferenced(page)) {
> +			if (*size == 0) {
> +				*size += PAGE_CACHE_SIZE;
> +				*offset = page->index << PAGE_CACHE_SHIFT;
> +				continue;
> +			}
> +			if (page->index !=
> +			    (*offset + *size) >> PAGE_CACHE_SHIFT)
> +				break;
> +			*size += PAGE_CACHE_SIZE;
> +		} else if (*size > 0)
> +			break;
> +		else
> +			index = page->index + 1;
> +	}
> +	pagevec_release(&pvec);
> +
> +	if (nr_pages > 0 && *size == 0)
> +		goto retry;

I don't think I know why this retry loop exists.  A comment would be
nice.

> +out:
> +	if (*size > 0)
> +		return 0;
> +	else
> +		return -ENOENT;
> +}
> +
>  static const struct super_operations btrfs_super_ops = {
>  	.drop_inode	= btrfs_drop_inode,
>  	.evict_inode	= btrfs_evict_inode,
> @@ -859,6 +906,7 @@ static const struct super_operations btr
>  	.remount_fs	= btrfs_remount,
>  	.freeze_fs	= btrfs_freeze,
>  	.unfreeze_fs	= btrfs_unfreeze,
> +	.metadata_incore = btrfs_metadata_incore,
>  };
>  
>  static const struct file_operations btrfs_ctl_fops = {
> 
> 
> --
> 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