[PATCH v2] writeback: keep cleanup_offline_cgwb()'s scan out of syncs

[Liz Fong-Jones <lizf@honeycomb.io>]

On a filesystem mounted with lazytime, sync() and syncfs() can return
without persisting timestamp updates of inodes attached to a dying
cgroup wb, so a crash after the sync loses them. File data and sizes are
not affected.

Since commit f6988c90671e ("writeback: bound cleanup_offline_cgwb()
rescans by rotating scanned inodes"), isw_prepare_wbs_switch() moves the
inodes it walks to the newest end of the list. When a pass fills its isw
before reaching an inode that got its lazytime update after a sync's
cutoff, the walked inodes land past it on b_dirty_time, and the sync's
queue_io(), which stops at the first inode dirtied after the cutoff,
misses them. The late inode has to be one without a pending lazytime
update; one already on b_dirty_time keeps its position.

The window is narrow. cleanup_offline_cgwbs_workfn() skips wbs with
anything on b_dirty, b_io or b_more_io, so the scan has to land between
wb_writeback() fixing dirtied_before and taking list_lock for its first
queue_io(), unless a multi-pass drain started before the wb got dirty
IO.

Take wb_switch_rwsem for read across the scan, as
process_inode_switch_wbs() does for the switches.

Fixes: f6988c90671e ("writeback: bound cleanup_offline_cgwb() rescans by rotating scanned inodes")
Cc: stable@vger.kernel.org
Suggested-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/all/0bc5d58533f10b4841807be3025daa11@kernel.org/
Assisted-by: Claude:claude-opus-5-5 checkpatch
Assisted-by: Claude:claude-fable-5-1
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Liz Fong-Jones <lizf@honeycomb.io>
---
Changes in v2, all from Tejun's review:
- Correct the trigger and the reordering description, and note how
  narrow the window is
- Reword the code comment
- Move the test details here and trim the description

Test, with a debug-only delay holding a syncfs's work on the wb between
its cutoff and queue_io() to widen the window: 300 files on a cgroup's
wb with lazytime atime updates, one more read after the cutoff, then
the cgroup removed so the cleanup ran. Without this patch, syncfs()
queued 48 of the 300 and after a crash 252 atime updates were lost, on
ext4 (-I 4096, so neighbours' timestamps are not written along) in 3 of
3 runs and on XFS in 1 of 1; with it, none were lost. arm64 KVM guests,
7.3-rc7.

v1: https://lore.kernel.org/all/20261015114307.1041273-1-lizf@honeycomb.io/

 fs/fs-writeback.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
index ea3eb40bf828..372ac9785e03 100644
--- a/fs/fs-writeback.c
+++ b/fs/fs-writeback.c
@@ -789,6 +789,13 @@ bool cleanup_offline_cgwb(struct bdi_writeback *wb)
 		new_wb = &wb->bdi->wb; /* wb_get() is noop for bdi's wb */
 
 	nr = 0;
+	/*
+	 * The scan moves walked inodes to the newest end of b_dirty_time,
+	 * past ones it didn't reach. If one of those was dirtied after a
+	 * sync's cutoff, the sync's queue_io() stops there and misses the
+	 * moved ones. Exclude syncs as process_inode_switch_wbs() does.
+	 */
+	down_read(&wb->bdi->wb_switch_rwsem);
 	spin_lock(&wb->list_lock);
 	/*
 	 * In addition to the inodes that have completed writeback, also switch
@@ -803,6 +810,7 @@ bool cleanup_offline_cgwb(struct bdi_writeback *wb)
 		restart = isw_prepare_wbs_switch(new_wb, isw, &wb->b_dirty_time,
 						 &nr);
 	spin_unlock(&wb->list_lock);
+	up_read(&wb->bdi->wb_switch_rwsem);
 
 	/* no attached inodes? bail out */
 	if (nr == 0) {

base-commit: 24a19b953e2a3d0ea5a8c4951a0ab8f32720c907
-- 
2.53.0