Skip to contentSkip to conversation

md: fix deadlock between mddev_suspend and flush bio

PATCH Andrew Kanner 1 participant 1 message May 22, 2025 Stable branch

lore
Share & tools
lore

Andrew Kanner

author
#0 lore raw
Quoted 20 lines

[...]

Additionally, the only difference between fixing the issue and before is that there is no return error handling of make_request(). But after previous patch cleaned md_write_start(), make_requst() only return error in raid5_make_request() by dm-raid, see commit 41425f96d7aa ("dm-raid456, md/raid456: fix a deadlock for dm-raid456 while io concurrent with reshape)". Since dm always splits data and flush operation into two separate io, io size of flush submitted by dm always is 0, make_request() will not be called in md_submit_flush_data(). To prevent future modifications from introducing issues, add WARN_ON to ensure make_request() no error is returned in this context.

[...]

(quoted diff) +1 -1
@@ -560,8 +552,20 @@ static void md_submit_flush_data(struct work_struct *ws)
 		bio_endio(bio);
 	} else {
 		bio->bi_opf &= ~REQ_PREFLUSH;
-		md_handle_request(mddev, bio);
+

Review static void md_submit_flush_data(struct work_struct *ws) · line 565

Quoted 10 lines
(quoted diff) +9 -0
+		/*
+		 * make_requst() will never return error here, it only
+		 * returns error in raid5_make_request() by dm-raid.
+		 * Since dm always splits data and flush operation into
+		 * two separate io, io size of flush submitted by dm
+		 * always is 0, make_request() will not be called here.
+		 */
+		if (WARN_ON_ONCE(!mddev->pers->make_request(mddev, bio)))
+			bio_io_error(bio);;
 	}

Hello,

It looks we can hit this WARN_ON_ONCE() after which rootfs is switching to read-only:

May 20 15:13:35 hostname kernel: WARNING: CPU: 35 PID: 1517323 at drivers/md/md.c:621 md_submit_flush_data+0x9b/0xe0
...
May 20 15:13:35 hostname kernel: XFS (md125): log I/O error -5
May 20 15:13:35 hostname kernel: XFS (md125): Filesystem has been shut down due to log error (0x2).
May 20 15:13:35 hostname kernel: XFS (md125): Please unmount the filesystem and rectify the problem(s).

Can you double check if the following regression is actual?

Since both stable/linux-6.1.y and stable/linux-6.6.y branches don't have b75197e86e6d ("md: Remove flush handling") there is a minor issue with this backport.

Statement "previous patch cleaned md_write_start(), make_requst() only return error in raid5_make_request() by dm-raid" will not work for both branches since 03e792eaf18e ("md: change the return value type of md_write_start to void") was not backported.

So we should either backport it, or do error handling, not the WARN_ON_ONCE().

Signature · 1 line
Andrew Kanner