Btrfs: fix transaction leak during fsync call
authorFilipe Manana <fdmanana@gmail.com>
Thu, 29 May 2014 22:31:39 +0000 (23:31 +0100)
committerChris Mason <clm@fb.com>
Tue, 10 Jun 2014 00:21:06 +0000 (17:21 -0700)
If btrfs_log_dentry_safe() returns an error, we set ret to 1 and
fall through with the goal of committing the transaction. However,
in the case where the inode doesn't need a full sync, we would call
btrfs_wait_ordered_range() against the target range for our inode,
and if it returned an error, we would return without commiting or
ending the transaction.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/file.c

index eb3f2708a01da909b7c6a27521cf48d9ac9b34e9..ad7c05909a49b387b72e5ce02dd8d6bbc65fb925 100644 (file)
@@ -2025,8 +2025,10 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
                if (!full_sync) {
                        ret = btrfs_wait_ordered_range(inode, start,
                                                       end - start + 1);
-                       if (ret)
+                       if (ret) {
+                               btrfs_end_transaction(trans, root);
                                goto out;
+                       }
                }
                ret = btrfs_commit_transaction(trans, root);
        } else {