[0/3] Rework fs error handling

Message

Josef Bacik May 20, 2021, 3:21 p.m. UTC

Hello,

This series brings together 3 patches I had sent separately, because they mostly
depend on eachother.  The first is

	btrfs: always abort the transaction if we abort a trans handle

Which changes our behavior for trans handles that haven't modified any metadata.
Our dependency chain can be complex, and thus we may rely on a particular async
action happening and rely on the transaction actually aborting if it fails.
Previously if nothing occurred we'd just let the transaction conitue, but this
could put us in a bad state.  We need to simply always abort the transaction for
safety reasons.

That patch allows us to use the next patch

	btrfs: add a btrfs_has_fs_error helper

Now that we can rely on FS_STATE_ERROR always being set if something is wrong,
we can wrap that check in a helper and convert all the open coded checks for
FS_STATE_ERROR and FS_STATE_ABORTED to the new helper.

And finally an actual fix

	btrfs: do not infinite loop in data reclaim if we aborted

The async data path could infinite loop if we aborted because it only broke in
the case of space_info->full, which would never happen if we weren't full and
had aborted the transaction.  This is fine stand alone, but I took advantage of
the helpers here so I want to make sure the fix goes along with its
dependencies.  Thanks,

Josef

Josef Bacik (3):
  btrfs: always abort the transaction if we abort a trans handle
  btrfs: add a btrfs_has_fs_error helper
  btrfs: do not infinite loop in data reclaim if we aborted

 fs/btrfs/ctree.c       |  5 +----
 fs/btrfs/ctree.h       |  5 +++++
 fs/btrfs/disk-io.c     |  8 +++-----
 fs/btrfs/extent-tree.c |  1 -
 fs/btrfs/extent_io.c   |  2 +-
 fs/btrfs/file.c        |  2 +-
 fs/btrfs/inode.c       |  6 +++---
 fs/btrfs/scrub.c       |  2 +-
 fs/btrfs/space-info.c  | 29 ++++++++++++++++++++++++-----
 fs/btrfs/super.c       | 13 +------------
 fs/btrfs/transaction.c | 19 +++++--------------
 fs/btrfs/transaction.h |  1 -
 fs/btrfs/tree-log.c    |  2 +-
 13 files changed, 46 insertions(+), 49 deletions(-)

Comments

David Sterba May 25, 2021, 3:39 p.m. UTC | #1

On Thu, May 20, 2021 at 11:21:30AM -0400, Josef Bacik wrote:
> Josef Bacik (3):
>   btrfs: always abort the transaction if we abort a trans handle

This one now added to misc-next.

>   btrfs: add a btrfs_has_fs_error helper
>   btrfs: do not infinite loop in data reclaim if we aborted

I can fix up the 2nd patch locally, it's basically only a rename but the
fs and transaction bits getting out of sync as mentioned could be a
problem (or not). Patch 3 depends on the 2nd so I'll keep both in
for-next for the time being.