Skip to contentSkip to conversation

Btrfs: fix fsync not persisting changed attributes of a directory

PATCH Superseded [email protected] 3 participants 4 messages May 15, 2019 – May 28, 2019 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/3 Btrfs: fix fsync not persisting changed attributes of a directory

From: Filipe Manana <[email protected]>

While logging an inode we follow its ancestors and for each one we mark it as logged in the current transaction, even if we have not logged it. As a consequence if we change an attribute of an ancestor, such as the UID or GID for example, and then explicitly fsync it, we end up not logging the inode at all despite returning success to user space, which results in the attribute being lost if a power failure happens after the fsync.

Sample reproducer:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt
  $ mkdir /mnt/dir
  $ chown 6007:6007 /mnt/dir
  $ sync
  $ chown 9003:9003 /mnt/dir
  $ touch /mnt/dir/file
  $ xfs_io -c fsync /mnt/dir/file
  # fsync our directory after fsync'ing the new file, should persist the
  # new values for the uid and gid.
  $ xfs_io -c fsync /mnt/dir

<power failure>

  $ mount /dev/sdb /mnt
  $ stat -c %u:%g /mnt/dir
  6007:6007

--> should be 9003:9003, the uid and gid were not persisted, despite
the explicit fsync on the directory prior to the power failure

Fix this by not updating the logged_trans field of ancestor inodes when logging an inode, since we have not logged them. Let only future calls to btrfs_log_inode() to mark inodes as logged.

This could be triggered by my recent fsync fuzz tester for fstests, for which an fstests patch exists titled "fstests: generic, fsync fuzz tester with fsstress".

Fixes 12fcfd22fe5b ("Btrfs: tree logging unlink/rename fixes")
---
 fs/btrfs/tree-log.c | 11 -----------
 1 file changed, 11 deletions(-)
fs/btrfs/tree-log.c +0 -11
index 87e3e4e37606..7d13533a9620 100644
@@ -5465,7 +5465,6 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
 {
 	int ret = 0;
 	struct dentry *old_parent = NULL;
-	struct btrfs_inode *orig_inode = inode;
 
 	/*
 	 * for regular files, if its inode is already on disk, we don't
@@ -5485,16 +5484,6 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
 	}
 
 	while (1) {
-		/*
-		 * If we are logging a directory then we start with our inode,
-		 * not our parent's inode, so we need to skip setting the
-		 * logged_trans so that further down in the log code we don't
-		 * think this inode has already been logged.
-		 */
-		if (inode != orig_inode)
-			inode->logged_trans = trans->transid;
-		smp_mb();
-
 		if (btrfs_must_commit_transaction(trans, inode)) {
 			ret = 1;
 			break;
Signature · 1 line
2.11.0
#1 lore raw
Quoted 67 lines · On 15.05.19 г. 18:02 ч., [email protected] wrote:

From: Filipe Manana <[email protected]>

While logging an inode we follow its ancestors and for each one we mark it as logged in the current transaction, even if we have not logged it. As a consequence if we change an attribute of an ancestor, such as the UID or GID for example, and then explicitly fsync it, we end up not logging the inode at all despite returning success to user space, which results in the attribute being lost if a power failure happens after the fsync.

Sample reproducer:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt
  $ mkdir /mnt/dir
  $ chown 6007:6007 /mnt/dir
  $ sync
  $ chown 9003:9003 /mnt/dir
  $ touch /mnt/dir/file
  $ xfs_io -c fsync /mnt/dir/file
  # fsync our directory after fsync'ing the new file, should persist the
  # new values for the uid and gid.
  $ xfs_io -c fsync /mnt/dir

<power failure>

  $ mount /dev/sdb /mnt
  $ stat -c %u:%g /mnt/dir
  6007:6007

--> should be 9003:9003, the uid and gid were not persisted, despite
the explicit fsync on the directory prior to the power failure

Fix this by not updating the logged_trans field of ancestor inodes when logging an inode, since we have not logged them. Let only future calls to btrfs_log_inode() to mark inodes as logged.

This could be triggered by my recent fsync fuzz tester for fstests, for which an fstests patch exists titled "fstests: generic, fsync fuzz tester with fsstress".

Fixes 12fcfd22fe5b ("Btrfs: tree logging unlink/rename fixes")
---
 fs/btrfs/tree-log.c | 11 -----------
 1 file changed, 11 deletions(-)
fs/btrfs/tree-log.c +0 -1
index 87e3e4e37606..7d13533a9620 100644
@@ -5465,7 +5465,6 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
 {
 	int ret = 0;
 	struct dentry *old_parent = NULL;
-	struct btrfs_inode *orig_inode = inode;
 
 	/*
 	 * for regular files, if its inode is already on disk, we don't
@@ -5485,16 +5484,6 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
 	}
 
 	while (1) {

fs/btrfs/tree-log.c static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, · line 5497

Quoted 10 lines
fs/btrfs/tree-log.c +0 -10
-		/*
-		 * If we are logging a directory then we start with our inode,
-		 * not our parent's inode, so we need to skip setting the
-		 * logged_trans so that further down in the log code we don't
-		 * think this inode has already been logged.
-		 */
-		if (inode != orig_inode)
-			inode->logged_trans = trans->transid;
-		smp_mb();
-

By removing this memory barrier don't you also obsolete the one in btrfs_record_unlink_dir? Both of these were introduced in 12fcfd22fe5b ("Btrfs: tree logging unlink/rename fixes") and despite they are missing explicit comments about the expected pairing one can only assume they both pair against each other.

Quoted 4 lines
fs/btrfs/tree-log.c +0 -0
 		if (btrfs_must_commit_transaction(trans, inode)) {
 			ret = 1;
 			break;
#2 lore raw

PATCH v2 1/3 Btrfs: fix fsync not persisting changed attributes of a directory

In reply to [email protected]

From: Filipe Manana <[email protected]>

While logging an inode we follow its ancestors and for each one we mark it as logged in the current transaction, even if we have not logged it. As a consequence if we change an attribute of an ancestor, such as the UID or GID for example, and then explicitly fsync it, we end up not logging the inode at all despite returning success to user space, which results in the attribute being lost if a power failure happens after the fsync.

Sample reproducer:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt
  $ mkdir /mnt/dir
  $ chown 6007:6007 /mnt/dir
  $ sync
  $ chown 9003:9003 /mnt/dir
  $ touch /mnt/dir/file
  $ xfs_io -c fsync /mnt/dir/file
  # fsync our directory after fsync'ing the new file, should persist the
  # new values for the uid and gid.
  $ xfs_io -c fsync /mnt/dir

<power failure>

  $ mount /dev/sdb /mnt
  $ stat -c %u:%g /mnt/dir
  6007:6007

--> should be 9003:9003, the uid and gid were not persisted, despite
the explicit fsync on the directory prior to the power failure

Fix this by not updating the logged_trans field of ancestor inodes when logging an inode, since we have not logged them. Let only future calls to btrfs_log_inode() to mark inodes as logged.

This could be triggered by my recent fsync fuzz tester for fstests, for which an fstests patch exists titled "fstests: generic, fsync fuzz tester with fsstress".

Fixes 12fcfd22fe5b ("Btrfs: tree logging unlink/rename fixes")

---

V2: Removed no longer needed smp_mb() at btrfs_record_unlink_dir().

 fs/btrfs/tree-log.c | 12 ------------
 1 file changed, 12 deletions(-)
fs/btrfs/tree-log.c +0 -12
index 87e3e4e37606..64b74f0a48d8 100644
@@ -5465,7 +5465,6 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
 {
 	int ret = 0;
 	struct dentry *old_parent = NULL;
-	struct btrfs_inode *orig_inode = inode;
 
 	/*
 	 * for regular files, if its inode is already on disk, we don't
@@ -5485,16 +5484,6 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
 	}
 
 	while (1) {
-		/*
-		 * If we are logging a directory then we start with our inode,
-		 * not our parent's inode, so we need to skip setting the
-		 * logged_trans so that further down in the log code we don't
-		 * think this inode has already been logged.
-		 */
-		if (inode != orig_inode)
-			inode->logged_trans = trans->transid;
-		smp_mb();
-
 		if (btrfs_must_commit_transaction(trans, inode)) {
 			ret = 1;
 			break;
@@ -6371,7 +6360,6 @@ void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
 	 * if this directory was already logged any new
 	 * names for this file/dir will get recorded
 	 */
-	smp_mb();
 	if (dir->logged_trans == trans->transid)
 		return;
 
Signature · 1 line
2.11.0
#3 lore raw
Quoted 41 lines · On Thu, May 16, 2019 at 03:48:55PM +0100, [email protected] wrote:

From: Filipe Manana <[email protected]>

While logging an inode we follow its ancestors and for each one we mark it as logged in the current transaction, even if we have not logged it. As a consequence if we change an attribute of an ancestor, such as the UID or GID for example, and then explicitly fsync it, we end up not logging the inode at all despite returning success to user space, which results in the attribute being lost if a power failure happens after the fsync.

Sample reproducer:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt
  $ mkdir /mnt/dir
  $ chown 6007:6007 /mnt/dir
  $ sync
  $ chown 9003:9003 /mnt/dir
  $ touch /mnt/dir/file
  $ xfs_io -c fsync /mnt/dir/file
  # fsync our directory after fsync'ing the new file, should persist the
  # new values for the uid and gid.
  $ xfs_io -c fsync /mnt/dir

<power failure>

  $ mount /dev/sdb /mnt
  $ stat -c %u:%g /mnt/dir
  6007:6007

--> should be 9003:9003, the uid and gid were not persisted, despite
the explicit fsync on the directory prior to the power failure

Fix this by not updating the logged_trans field of ancestor inodes when logging an inode, since we have not logged them. Let only future calls to btrfs_log_inode() to mark inodes as logged.

Quoted 7 lines

This could be triggered by my recent fsync fuzz tester for fstests, for which an fstests patch exists titled "fstests: generic, fsync fuzz tester with fsstress".

Fixes 12fcfd22fe5b ("Btrfs: tree logging unlink/rename fixes")

---

Added to 5.2-rc queue, thanks.