[1/4] btrfs: fix non-empty delayed iputs list on unmount due to endio workers

Commit Message

Filipe Manana March 5, 2025, 5:45 p.m. UTC

From: Filipe Manana <[e-mail]>

At close_ctree() after we have ran delayed iputs either through explicitly
calling btrfs_run_delayed_iputs() or later during the call to
btrfs_commit_super() or btrfs_error_commit_super(), we assert that the
delayed iputs list is empty.

Sometimes this assertion may fail because delayed iputs may have been
added to the list after we last ran delayed iputs, and this happens due
to workers in the endio_workers workqueue still running. These workers can
do a final put on an ordered extent attached to a data bio, which results
in adding a delayed iput. This is done at btrfs_bio_end_io() and its
helper __btrfs_bio_end_io().

Fix this by flushing the endio_workers workqueue before running delayed
iputs at close_ctree().

David reported this when running generic/648.

Reported-by: David Sterba <[e-mail]>
Fixes: ec63b84d4611 ("btrfs: add an ordered_extent pointer to struct btrfs_bio")
Signed-off-by: Filipe Manana <[e-mail]>
---
 fs/btrfs/disk-io.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

Patch

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index d96ea974ef73..df8e075e69a3 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4340,6 +4340,16 @@  void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	 */
 	btrfs_flush_workqueue(fs_info->delalloc_workers);
 
+	/*
+	 * We can also have ordered extents getting their last reference dropped
+	 * from the endio_workers workqueue because for data bios we keep a
+	 * reference on an ordered extent which gets dropped when running
+	 * btrfs_bio_end_io() in that workqueue, and that final drop results in
+	 * adding a delayed iput for the inode.
+	 */
+	if (fs_info->endio_workers)
+		flush_workqueue(fs_info->endio_workers);
+
 	/*
 	 * After we parked the cleaner kthread, ordered extents may have
 	 * completed and created new delayed iputs. If one of the async reclaim