[PATCH v2 0/3] btrfs: make qgroup rescan work to handle fs freezing

[Qu Wenruo <wqu@suse.com>]

[CHANGELOG]
v2:
- Add a new patch to slightly change when RESCAN bit should be cleared

- Add a new small patch to always output the rescan ending message
  This is split from the original patch.

- Save if the rescan is cancelled or not when still holding
  qgroup_rescan_lock
  To reduce possible races affecting the output message

- Also clear the RESCAN bit if btrfs_try_start_transaction() failed

There is a user report about a running qgroup rescan failed the pm
suspension.

It turns out that if we sleep inside a work item, the workqueue still
counts as active during suspension, thus suspension can not continue.

The fix is to:

- Check if the fs is frozen in rescan_should_stop()
  To cover the common case.

- Use a trylock version o btrfs_start_transaction()
  So it won't sleep on sb_start_intwrite(). This is to cover
  the minor cases where fs is frozen between rescan_should_stop() and
  btrfs_start_transaction().

And two small patches to clean up the rescan code before the proper fix.

Qu Wenruo (3):
  btrfs: let rescan_should_stop() decide if the rescan flag is cleared
  btrfs: always show the message when qgroup rescan ended
  btrfs: make qgroup rescan work handle fs freezing

 fs/btrfs/qgroup.c      | 59 ++++++++++++++++++++++++++++++------------
 fs/btrfs/transaction.c | 22 +++++++++++++---
 fs/btrfs/transaction.h |  4 +++
 3 files changed, 65 insertions(+), 20 deletions(-)

-- 
2.55.0