btrfs: fix double inode unlock for direct IO sync writes

If we do a direct IO sync write, at btrfs_sync_file(), and we need to skip
inode logging or we get an error starting a transaction or an error when
flushing delalloc, we end up unlocking the inode when we shouldn't under
the 'out_release_extents' label, and then unlock it again at
btrfs_direct_write().

Fix that by checking if we have to skip inode unlocking under that label.

Reported-by: syzbot+7dbbb74af6291b5a5a8b@syzkaller.appspotmail.com
Link: https://lore.kernel.org/linux-btrfs/000000000000dfd631061eaeb4bc@google.com/
Fixes: 939b656bc8ab ("btrfs: fix corruption after buffer fault in during direct IO append write")
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 9f10a9f23fcc8b91e9cd929f3365a9bc0b0f4a99..9914419f3b7d6105cd16ac013b62d0c831fce4e8 100644 (file)
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -1868,7 +1868,10 @@ out:
 
 out_release_extents:
        btrfs_release_log_ctx_extents(&ctx);
-       btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
+       if (skip_ilock)
+               up_write(&inode->i_mmap_lock);
+       else
+               btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
        goto out;
 }