[PATCH 3/4] btrfs: zoned: drop stranded dirty metadata buffers at unmount

On a zoned filesystem a freed tree block is kept dirty and flagged
EXTENT_BUFFER_ZONED_ZEROOUT so a later writeback zeroes it out and advances the
zone write pointer. Unsynced tree-log updates (e.g. from rename or link) leave
such buffers behind when the log is freed at commit, and across log generations
they can end up ahead of the write pointer behind a hole, so btree_writepages()
can never write them. During normal operation the space is later reclaimed by a
zone reset; at unmount it is not, and the buffers survive to the final iput() of
the btree inode, which hangs in folio_wait_writeback() once the endio
workqueues are stopped.

They cannot be written back from where they are freed (free_log_tree(), inside
the committing transaction) without deadlocking against that commit, and they
are stale anyway, not referenced by the committed superblock. Drop their dirty
state in close_ctree(), before btrfs_stop_all_workers().

Fixes: aa6313e6ff2b ("btrfs: zoned: don't clear dirty flag of extent buffer")
Signed-off-by: Johannes Thumshirn <[email protected]>
---
 fs/btrfs/disk-io.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index e84ee395ac7f..48374cec77d8 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4539,6 +4539,13 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	free_root_pointers(fs_info, true);
 	btrfs_free_fs_roots(fs_info);
 
+	/*
+	 * Drop metadata left stranded ahead of a zone write pointer while the
+	 * endio workqueues are still up, so the final iput() of the btree inode
+	 * below does not hang submitting a write that can no longer complete.
+	 */
+	btrfs_zoned_release_dirty_metadata(fs_info);
+
 	/*
 	 * We must make sure there is not any read request to
 	 * submit after we stop all workers.
-- 
2.54.0