Btrfs: fix a dio write regression
authorLiu Bo <bo.li.liu@oracle.com>
Thu, 23 Aug 2012 02:10:38 +0000 (20:10 -0600)
committerChris Mason <chris.mason@oracle.com>
Tue, 28 Aug 2012 20:53:41 +0000 (16:53 -0400)
commit24c03fa5cf3d02c327cf9f2fc39f72664b1bd7e1
treeb0e14cbc951996a4b33fa81e7a19d092e08398d5
parentbd7de2c9a449e26a5493d918618eb20ae60d56bd
Btrfs: fix a dio write regression

This bug is introduced by commit 3b8bde746f6f9bd36a9f05f5f3b6e334318176a9
(Btrfs: lock extents as we map them in DIO).

In dio write, we should unlock the section which we didn't do IO on in case that
we fall back to buffered write.  But we need to not only unlock the section
but also cleanup reserved space for the section.

This bug was found while running xfstests 133, with this 133 no longer complains.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
fs/btrfs/inode.c