[PATCH 0/5] btrfs: zoned: fix two deadlocks when runing fstests

Fix two distinct deadlocks when running fstests:

- The first one happens primarily (but not only) on btrfs/187, Where we
  bunch up dirty metadata and then on umount try to submit it. But at
  the time of submission the block-group is already gone leaving stale
  folios behind that can't be written out. Patch 2 fixes it by clearing
  the stale buffers, patch 1 is a preparation patch for patch 2.

- The second one is a double lock of fs_info->ordered_extent_mutex when
  doing zone finish. It creates a copy of the roots that need to be
  finished and then does an unlocked version of
  btrfs_wait_ordered_roots() btrfs_wait_ordered_roots_nolock() on that
  copy.

For both issues debugging was done LLM assisted and I attributed it in
the respective fix patches (not the prep patches).

Johannes Thumshirn (5):
  btrfs: factor out dirty-clearing part of btrfs_clear_buffer_dirty
  btrfs: zoned: drop stale dirty btree buffers at close_ctree()
  btrfs: walk waited ordered extents in place
  btrfs: walk waited ordered roots in place
  btrfs: zoned: avoid ordered_operations_mutex when finishing a zone

 fs/btrfs/disk-io.c      |   4 ++
 fs/btrfs/extent_io.c    |  57 ++++++++++++------
 fs/btrfs/extent_io.h    |   1 +
 fs/btrfs/ordered-data.c | 129 +++++++++++++++++++++++++++++++++-------
 fs/btrfs/ordered-data.h |   4 ++
 fs/btrfs/zoned.c        |   8 ++-
 6 files changed, 159 insertions(+), 44 deletions(-)

-- 
2.54.0