[patch 20/35 xattr-rework] reiserfs: eliminate per-super xattr lock

[Jeff Mahoney <jeffm@suse.com>]

[-- Attachment #1: patches.suse/reiserfs-eliminate-per-super-xattr-lock.diff --]
[-- Type: text/plain, Size: 17558 bytes --]

 With the switch to using inode->i_mutex locking during lookups/creation in
 the xattr root, the per-super xattr lock is no longer needed.

 This patch removes it.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
--
 fs/reiserfs/inode.c            |   14 -------
 fs/reiserfs/namei.c            |   29 ----------------
 fs/reiserfs/super.c            |    4 --
 fs/reiserfs/xattr.c            |   70 +++++++++++++++++++-------------------
 fs/reiserfs/xattr_acl.c        |   74 ++++++++++++++++++-----------------------
 include/linux/reiserfs_fs.h    |    3 -
 include/linux/reiserfs_fs_sb.h |    3 -
 include/linux/reiserfs_xattr.h |   28 ++-------------
 8 files changed, 74 insertions(+), 151 deletions(-)

--- a/fs/reiserfs/inode.c
+++ b/fs/reiserfs/inode.c
@@ -1957,19 +1957,7 @@ int reiserfs_new_inode(struct reiserfs_t
 	inode->i_nlink = 0;
 	th->t_trans_id = 0;	/* so the caller can't use this handle later */
 	unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
-
-	/* If we were inheriting an ACL, we need to release the lock so that
-	 * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
-	 * code really needs to be reworked, but this will take care of it
-	 * for now. -jeffm */
-#ifdef CONFIG_REISERFS_FS_POSIX_ACL
-	if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
-		reiserfs_write_unlock_xattrs(dir->i_sb);
-		iput(inode);
-		reiserfs_write_lock_xattrs(dir->i_sb);
-	} else
-#endif
-		iput(inode);
+	iput(inode);
 	return err;
 }
 
--- a/fs/reiserfs/namei.c
+++ b/fs/reiserfs/namei.c
@@ -609,9 +609,6 @@ static int reiserfs_create(struct inode
 
 	reiserfs_write_lock(dir->i_sb);
 
-	if (locked)
-		reiserfs_write_lock_xattrs(dir->i_sb);
-
 	retval = journal_begin(&th, dir->i_sb, jbegin_count);
 	if (retval) {
 		drop_new_inode(inode);
@@ -624,11 +621,6 @@ static int reiserfs_create(struct inode
 	if (retval)
 		goto out_failed;
 
-	if (locked) {
-		reiserfs_write_unlock_xattrs(dir->i_sb);
-		locked = 0;
-	}
-
 	inode->i_op = &reiserfs_file_inode_operations;
 	inode->i_fop = &reiserfs_file_operations;
 	inode->i_mapping->a_ops = &reiserfs_address_space_operations;
@@ -655,8 +647,6 @@ static int reiserfs_create(struct inode
 	retval = journal_end(&th, dir->i_sb, jbegin_count);
 
       out_failed:
-	if (locked)
-		reiserfs_write_unlock_xattrs(dir->i_sb);
 	reiserfs_write_unlock(dir->i_sb);
 	return retval;
 }
@@ -686,9 +676,6 @@ static int reiserfs_mknod(struct inode *
 
 	reiserfs_write_lock(dir->i_sb);
 
-	if (locked)
-		reiserfs_write_lock_xattrs(dir->i_sb);
-
 	retval = journal_begin(&th, dir->i_sb, jbegin_count);
 	if (retval) {
 		drop_new_inode(inode);
@@ -702,11 +689,6 @@ static int reiserfs_mknod(struct inode *
 		goto out_failed;
 	}
 
-	if (locked) {
-		reiserfs_write_unlock_xattrs(dir->i_sb);
-		locked = 0;
-	}
-
 	inode->i_op = &reiserfs_special_inode_operations;
 	init_special_inode(inode, inode->i_mode, rdev);
 
@@ -736,8 +718,6 @@ static int reiserfs_mknod(struct inode *
 	retval = journal_end(&th, dir->i_sb, jbegin_count);
 
       out_failed:
-	if (locked)
-		reiserfs_write_unlock_xattrs(dir->i_sb);
 	reiserfs_write_unlock(dir->i_sb);
 	return retval;
 }
@@ -767,8 +747,6 @@ static int reiserfs_mkdir(struct inode *
 	locked = reiserfs_cache_default_acl(dir);
 
 	reiserfs_write_lock(dir->i_sb);
-	if (locked)
-		reiserfs_write_lock_xattrs(dir->i_sb);
 
 	retval = journal_begin(&th, dir->i_sb, jbegin_count);
 	if (retval) {
@@ -790,11 +768,6 @@ static int reiserfs_mkdir(struct inode *
 		goto out_failed;
 	}
 
-	if (locked) {
-		reiserfs_write_unlock_xattrs(dir->i_sb);
-		locked = 0;
-	}
-
 	reiserfs_update_inode_transaction(inode);
 	reiserfs_update_inode_transaction(dir);
 
@@ -824,8 +797,6 @@ static int reiserfs_mkdir(struct inode *
 	unlock_new_inode(inode);
 	retval = journal_end(&th, dir->i_sb, jbegin_count);
       out_failed:
-	if (locked)
-		reiserfs_write_unlock_xattrs(dir->i_sb);
 	reiserfs_write_unlock(dir->i_sb);
 	return retval;
 }
--- a/fs/reiserfs/super.c
+++ b/fs/reiserfs/super.c
@@ -1646,10 +1646,6 @@ static int reiserfs_fill_super(struct su
 	REISERFS_SB(s)->s_alloc_options.preallocmin = 0;
 	/* Preallocate by 16 blocks (17-1) at once */
 	REISERFS_SB(s)->s_alloc_options.preallocsize = 17;
-#ifdef CONFIG_REISERFS_FS_XATTR
-	/* Initialize the rwsem for xattr dir */
-	init_rwsem(&REISERFS_SB(s)->xattr_dir_sem);
-#endif
 	/* setup default block allocator options */
 	reiserfs_init_alloc_options(s);
 
--- a/fs/reiserfs/xattr.c
+++ b/fs/reiserfs/xattr.c
@@ -27,6 +27,12 @@
  * these are special cases for filesystem ACLs, they are interpreted by the
  * kernel, in addition, they are negatively and positively cached and attached
  * to the inode so that unnecessary lookups are avoided.
+ *
+ * Locking works like so:
+ * The xattr root (/.reiserfs_priv/xattrs) is protected by its i_mutex.
+ * The xattr dir (/.reiserfs_priv/xattrs/<oid>.<gen>) is protected by
+ * inode->xattr_sem.
+ * The xattrs themselves are likewise protected by the xattr_sem.
  */
 
 #include <linux/reiserfs_fs.h>
@@ -392,16 +398,17 @@ reiserfs_delete_xattrs_filler(void *buf,
 /* This is called w/ inode->i_mutex downed */
 int reiserfs_delete_xattrs(struct inode *inode)
 {
-	struct dentry *dir, *root;
 	int err = 0;
+	struct dentry *dir, *root;
+	struct reiserfs_transaction_handle th;
+	int blocks = JOURNAL_PER_BALANCE_CNT * 2 + 2 +
+		     4 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 
 	/* Skip out, an xattr has no xattrs associated with it */
 	if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1)
 		return 0;
 
-	reiserfs_read_lock_xattrs(inode->i_sb);
 	dir = open_xa_dir(inode, XATTR_REPLACE);
-	reiserfs_read_unlock_xattrs(inode->i_sb);
 	if (IS_ERR(dir)) {
 		err = PTR_ERR(dir);
 		goto out;
@@ -416,18 +423,26 @@ int reiserfs_delete_xattrs(struct inode
 	if (err)
 		goto out_dir;
 
-	/* Leftovers besides . and .. -- that's not good. */
-	if (dir->d_inode->i_nlink <= 2) {
-		root = open_xa_root(inode->i_sb, XATTR_REPLACE);
-		reiserfs_write_lock_xattrs(inode->i_sb);
+	/* We start a transaction here to avoid a ABBA situation
+	 * between the xattr root's i_mutex and the journal lock.
+	 * Inode creation will inherit an ACL, which requires a
+	 * lookup. The lookup locks the xattr root i_mutex with a
+	 * transaction open.  Inode deletion takes teh xattr root
+	 * i_mutex to delete the directory and then starts a
+	 * transaction inside it. Boom. This doesn't incur much
+	 * additional overhead since the reiserfs_rmdir transaction
+	 * will just nest inside the outer transaction. */
+	err = journal_begin(&th, inode->i_sb, blocks);
+	if (!err) {
+		int jerror;
+		root = dget(dir->d_parent);
 		mutex_lock_nested(&root->d_inode->i_mutex, I_MUTEX_XATTR);
 		err = xattr_rmdir(root->d_inode, dir);
+		jerror = journal_end(&th, inode->i_sb, blocks);
 		mutex_unlock(&root->d_inode->i_mutex);
-		reiserfs_write_unlock_xattrs(inode->i_sb);
 		dput(root);
-	} else {
-		reiserfs_warning(inode->i_sb, "jdm-20006",
-				 "Couldn't remove all entries in directory");
+
+		err = jerror ?: err;
 	}
 
 out_dir:
@@ -437,6 +452,9 @@ out:
 	if (!err)
 		REISERFS_I(inode)->i_flags =
 		    REISERFS_I(inode)->i_flags & ~i_has_xattr_dir;
+	else
+		reiserfs_warning(inode->i_sb, "jdm-20004",
+				 "Couldn't remove all xattrs (%d)\n", err);
 	return err;
 }
 
@@ -485,9 +503,7 @@ int reiserfs_chown_xattrs(struct inode *
 	if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1)
 		return 0;
 
-	reiserfs_read_lock_xattrs(inode->i_sb);
 	dir = open_xa_dir(inode, XATTR_REPLACE);
-	reiserfs_read_unlock_xattrs(inode->i_sb);
 	if (IS_ERR(dir)) {
 		if (PTR_ERR(dir) != -ENODATA)
 			err = PTR_ERR(dir);
@@ -731,6 +747,11 @@ reiserfs_xattr_get(const struct inode *i
 		goto out;
 	}
 
+	/* protect against concurrent access. xattrs are backed by
+	 * regular files, but they're not regular files. The updates
+	 * must be atomic from the perspective of the user. */
+	mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_XATTR);
+
 	isize = i_size_read(dentry->d_inode);
 	REISERFS_I(inode)->i_flags |= i_has_xattr_dir;
 
@@ -798,6 +819,7 @@ reiserfs_xattr_get(const struct inode *i
 	}
 
 out_dput:
+	mutex_unlock(&dentry->d_inode->i_mutex);
 	dput(dentry);
 
 out:
@@ -834,7 +856,6 @@ int reiserfs_xattr_del(struct inode *ino
 static struct reiserfs_xattr_handler *find_xattr_handler_prefix(const char *);
 /*
  * Inode operation getxattr()
- * Preliminary locking: we down dentry->d_inode->i_mutex
  */
 ssize_t
 reiserfs_getxattr(struct dentry * dentry, const char *name, void *buffer,
@@ -848,9 +869,7 @@ reiserfs_getxattr(struct dentry * dentry
 		return -EOPNOTSUPP;
 
 	reiserfs_read_lock_xattr_i(dentry->d_inode);
-	reiserfs_read_lock_xattrs(dentry->d_sb);
 	err = xah->get(dentry->d_inode, name, buffer, size);
-	reiserfs_read_unlock_xattrs(dentry->d_sb);
 	reiserfs_read_unlock_xattr_i(dentry->d_inode);
 	return err;
 }
@@ -866,23 +885,13 @@ reiserfs_setxattr(struct dentry *dentry,
 {
 	struct reiserfs_xattr_handler *xah = find_xattr_handler_prefix(name);
 	int err;
-	int lock;
 
 	if (!xah || !reiserfs_xattrs(dentry->d_sb) ||
 	    get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
 		return -EOPNOTSUPP;
 
 	reiserfs_write_lock_xattr_i(dentry->d_inode);
-	lock = !has_xattr_dir(dentry->d_inode);
-	if (lock)
-		reiserfs_write_lock_xattrs(dentry->d_sb);
-	else
-		reiserfs_read_lock_xattrs(dentry->d_sb);
 	err = xah->set(dentry->d_inode, name, value, size, flags);
-	if (lock)
-		reiserfs_write_unlock_xattrs(dentry->d_sb);
-	else
-		reiserfs_read_unlock_xattrs(dentry->d_sb);
 	reiserfs_write_unlock_xattr_i(dentry->d_inode);
 	return err;
 }
@@ -902,8 +911,6 @@ int reiserfs_removexattr(struct dentry *
 		return -EOPNOTSUPP;
 
 	reiserfs_write_lock_xattr_i(dentry->d_inode);
-	reiserfs_read_lock_xattrs(dentry->d_sb);
-
 	/* Deletion pre-operation */
 	if (xah->del) {
 		err = xah->del(dentry->d_inode, name);
@@ -917,7 +924,6 @@ int reiserfs_removexattr(struct dentry *
 	mark_inode_dirty(dentry->d_inode);
 
       out:
-	reiserfs_read_unlock_xattrs(dentry->d_sb);
 	reiserfs_write_unlock_xattr_i(dentry->d_inode);
 	return err;
 }
@@ -966,8 +972,6 @@ reiserfs_listxattr_filler(void *buf, con
 
 /*
  * Inode operation listxattr()
- *
- * Preliminary locking: we down dentry->d_inode->i_mutex
  */
 ssize_t reiserfs_listxattr(struct dentry * dentry, char *buffer, size_t size)
 {
@@ -983,9 +987,7 @@ ssize_t reiserfs_listxattr(struct dentry
 		return -EOPNOTSUPP;
 
 	reiserfs_read_lock_xattr_i(dentry->d_inode);
-	reiserfs_read_lock_xattrs(dentry->d_sb);
 	dir = open_xa_dir(dentry->d_inode, XATTR_REPLACE);
-	reiserfs_read_unlock_xattrs(dentry->d_sb);
 	if (IS_ERR(dir)) {
 		err = PTR_ERR(dir);
 		if (err == -ENODATA)
@@ -1114,11 +1116,9 @@ static int reiserfs_check_acl(struct ino
 	int error = -EAGAIN; /* do regular unix permission checks by default */
 
 	reiserfs_read_lock_xattr_i(inode);
-	reiserfs_read_lock_xattrs(inode->i_sb);
 
 	acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS);
 
-	reiserfs_read_unlock_xattrs(inode->i_sb);
 	reiserfs_read_unlock_xattr_i(inode);
 
 	if (acl) {
--- a/fs/reiserfs/xattr_acl.c
+++ b/fs/reiserfs/xattr_acl.c
@@ -172,6 +172,29 @@ static void *posix_acl_to_disk(const str
 	return ERR_PTR(-EINVAL);
 }
 
+static inline void iset_acl(struct inode *inode, struct posix_acl **i_acl,
+			    struct posix_acl *acl)
+{
+	spin_lock(&inode->i_lock);
+	if (*i_acl != ERR_PTR(-ENODATA))
+		posix_acl_release(*i_acl);
+	*i_acl = posix_acl_dup(acl);
+	spin_unlock(&inode->i_lock);
+}
+
+static inline struct posix_acl *iget_acl(struct inode *inode,
+					 struct posix_acl **i_acl)
+{
+	struct posix_acl *acl = ERR_PTR(-ENODATA);
+
+	spin_lock(&inode->i_lock);
+	if (*i_acl != ERR_PTR(-ENODATA))
+		acl = posix_acl_dup(*i_acl);
+	spin_unlock(&inode->i_lock);
+
+	return acl;
+}
+
 /*
  * Inode operation get_posix_acl().
  *
@@ -199,11 +222,11 @@ struct posix_acl *reiserfs_get_acl(struc
 		return ERR_PTR(-EINVAL);
 	}
 
-	if (IS_ERR(*p_acl)) {
-		if (PTR_ERR(*p_acl) == -ENODATA)
-			return NULL;
-	} else if (*p_acl != NULL)
-		return posix_acl_dup(*p_acl);
+	acl = iget_acl(inode, p_acl);
+	if (acl && !IS_ERR(acl))
+		return acl;
+	else if (PTR_ERR(acl) == -ENODATA)
+		return NULL;
 
 	size = reiserfs_xattr_get(inode, name, NULL, 0);
 	if (size < 0) {
@@ -229,7 +252,7 @@ struct posix_acl *reiserfs_get_acl(struc
 	} else {
 		acl = posix_acl_from_disk(value, retval);
 		if (!IS_ERR(acl))
-			*p_acl = posix_acl_dup(acl);
+			iset_acl(inode, p_acl, acl);
 	}
 
 	kfree(value);
@@ -300,16 +323,8 @@ reiserfs_set_acl(struct inode *inode, in
 
 	kfree(value);
 
-	if (!error) {
-		/* Release the old one */
-		if (!IS_ERR(*p_acl) && *p_acl)
-			posix_acl_release(*p_acl);
-
-		if (acl == NULL)
-			*p_acl = ERR_PTR(-ENODATA);
-		else
-			*p_acl = posix_acl_dup(acl);
-	}
+	if (!error)
+		iset_acl(inode, p_acl, acl);
 
 	return error;
 }
@@ -404,9 +419,7 @@ int reiserfs_cache_default_acl(struct in
 	if (reiserfs_posixacl(inode->i_sb) && !IS_PRIVATE(inode)) {
 		struct posix_acl *acl;
 		reiserfs_read_lock_xattr_i(inode);
-		reiserfs_read_lock_xattrs(inode->i_sb);
 		acl = reiserfs_get_acl(inode, ACL_TYPE_DEFAULT);
-		reiserfs_read_unlock_xattrs(inode->i_sb);
 		reiserfs_read_unlock_xattr_i(inode);
 		ret = (acl && !IS_ERR(acl));
 		if (ret)
@@ -429,9 +442,7 @@ int reiserfs_acl_chmod(struct inode *ino
 		return 0;
 	}
 
-	reiserfs_read_lock_xattrs(inode->i_sb);
 	acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS);
-	reiserfs_read_unlock_xattrs(inode->i_sb);
 	if (!acl)
 		return 0;
 	if (IS_ERR(acl))
@@ -442,17 +453,8 @@ int reiserfs_acl_chmod(struct inode *ino
 		return -ENOMEM;
 	error = posix_acl_chmod_masq(clone, inode->i_mode);
 	if (!error) {
-		int lock = !has_xattr_dir(inode);
 		reiserfs_write_lock_xattr_i(inode);
-		if (lock)
-			reiserfs_write_lock_xattrs(inode->i_sb);
-		else
-			reiserfs_read_lock_xattrs(inode->i_sb);
 		error = reiserfs_set_acl(inode, ACL_TYPE_ACCESS, clone);
-		if (lock)
-			reiserfs_write_unlock_xattrs(inode->i_sb);
-		else
-			reiserfs_read_unlock_xattrs(inode->i_sb);
 		reiserfs_write_unlock_xattr_i(inode);
 	}
 	posix_acl_release(clone);
@@ -480,14 +482,9 @@ posix_acl_access_set(struct inode *inode
 static int posix_acl_access_del(struct inode *inode, const char *name)
 {
 	struct reiserfs_inode_info *reiserfs_i = REISERFS_I(inode);
-	struct posix_acl **acl = &reiserfs_i->i_acl_access;
 	if (strlen(name) != sizeof(POSIX_ACL_XATTR_ACCESS) - 1)
 		return -EINVAL;
-	if (!IS_ERR(*acl) && *acl) {
-		posix_acl_release(*acl);
-		*acl = ERR_PTR(-ENODATA);
-	}
-
+	iset_acl(inode, &reiserfs_i->i_acl_access, ERR_PTR(-ENODATA));
 	return 0;
 }
 
@@ -533,14 +530,9 @@ posix_acl_default_set(struct inode *inod
 static int posix_acl_default_del(struct inode *inode, const char *name)
 {
 	struct reiserfs_inode_info *reiserfs_i = REISERFS_I(inode);
-	struct posix_acl **acl = &reiserfs_i->i_acl_default;
 	if (strlen(name) != sizeof(POSIX_ACL_XATTR_DEFAULT) - 1)
 		return -EINVAL;
-	if (!IS_ERR(*acl) && *acl) {
-		posix_acl_release(*acl);
-		*acl = ERR_PTR(-ENODATA);
-	}
-
+	iset_acl(inode, &reiserfs_i->i_acl_default, ERR_PTR(-ENODATA));
 	return 0;
 }
 
--- a/include/linux/reiserfs_fs.h
+++ b/include/linux/reiserfs_fs.h
@@ -2224,7 +2224,4 @@ int reiserfs_unpack(struct inode *inode,
 #define reiserfs_write_lock( sb ) lock_kernel()
 #define reiserfs_write_unlock( sb ) unlock_kernel()
 
-/* xattr stuff */
-#define REISERFS_XATTR_DIR_SEM(s) (REISERFS_SB(s)->xattr_dir_sem)
-
 #endif				/* _LINUX_REISER_FS_H */
--- a/include/linux/reiserfs_fs_sb.h
+++ b/include/linux/reiserfs_fs_sb.h
@@ -403,9 +403,6 @@ struct reiserfs_sb_info {
 	spinlock_t bitmap_lock;	/* this lock on now only used to protect reserved_blocks variable */
 	struct dentry *priv_root;	/* root of /.reiserfs_priv */
 	struct dentry *xattr_root;	/* root of /.reiserfs_priv/.xa */
-#ifdef CONFIG_REISERFS_FS_XATTR
-	struct rw_semaphore xattr_dir_sem;
-#endif
 	int j_errno;
 #ifdef CONFIG_QUOTA
 	char *s_qf_names[MAXQUOTAS];
--- a/include/linux/reiserfs_xattr.h
+++ b/include/linux/reiserfs_xattr.h
@@ -67,45 +67,27 @@ extern struct reiserfs_xattr_handler use
 extern struct reiserfs_xattr_handler trusted_handler;
 extern struct reiserfs_xattr_handler security_handler;
 
-static inline void reiserfs_write_lock_xattrs(struct super_block *sb)
-{
-	down_write(&REISERFS_XATTR_DIR_SEM(sb));
-}
-static inline void reiserfs_write_unlock_xattrs(struct super_block *sb)
-{
-	up_write(&REISERFS_XATTR_DIR_SEM(sb));
-}
-static inline void reiserfs_read_lock_xattrs(struct super_block *sb)
-{
-	down_read(&REISERFS_XATTR_DIR_SEM(sb));
-}
-
-static inline void reiserfs_read_unlock_xattrs(struct super_block *sb)
-{
-	up_read(&REISERFS_XATTR_DIR_SEM(sb));
-}
-
 static inline void reiserfs_write_lock_xattr_i(struct inode *inode)
 {
-	down_write(&REISERFS_I(inode)->xattr_sem);
+	down_write(&REISERFS_I(inode)->i_xattr_sem);
 }
 static inline void reiserfs_write_unlock_xattr_i(struct inode *inode)
 {
-	up_write(&REISERFS_I(inode)->xattr_sem);
+	up_write(&REISERFS_I(inode)->i_xattr_sem);
 }
 static inline void reiserfs_read_lock_xattr_i(struct inode *inode)
 {
-	down_read(&REISERFS_I(inode)->xattr_sem);
+	down_read(&REISERFS_I(inode)->i_xattr_sem);
 }
 
 static inline void reiserfs_read_unlock_xattr_i(struct inode *inode)
 {
-	up_read(&REISERFS_I(inode)->xattr_sem);
+	up_read(&REISERFS_I(inode)->i_xattr_sem);
 }
 
 static inline void reiserfs_init_xattr_rwsem(struct inode *inode)
 {
-	init_rwsem(&REISERFS_I(inode)->xattr_sem);
+	init_rwsem(&REISERFS_I(inode)->i_xattr_sem);
 }
 
 #else