btrfs: finish ordered extent cleaning if no progress is found
authorNaohiro Aota <naohiro.aota@wdc.com>
Fri, 1 Sep 2017 08:59:07 +0000 (17:59 +0900)
committerDavid Sterba <dsterba@suse.com>
Tue, 26 Sep 2017 12:49:06 +0000 (14:49 +0200)
commit67c003f90fd68062d92a7ffade36f9b2a9098bd8
tree874bcbe06f77eb19a0e5bb5a1bb830354b0af4d9
parent63d71450c8d817649a79e37d685523f988b9cc98
btrfs: finish ordered extent cleaning if no progress is found

__endio_write_update_ordered() repeats the search until it reaches the end
of the specified range. This works well with direct IO path, because before
the function is called, it's ensured that there are ordered extents filling
whole the range. It's not the case, however, when it's called from
run_delalloc_range(): it is possible to have error in the midle of the loop
in e.g. run_delalloc_nocow(), so that there exisits the range not covered
by any ordered extents. By cleaning such "uncomplete" range,
__endio_write_update_ordered() stucks at offset where there're no ordered
extents.

Since the ordered extents are created from head to tail, we can stop the
search if there are no offset progress.

Fixes: 524272607e88 ("btrfs: Handle delalloc error correctly to avoid ordered extent hang")
Cc: <stable@vger.kernel.org> # 4.12
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Qu Wenruo <quwenruo.btrfs@gmx.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/inode.c