dm: do not return early from dm_io_complete if BLK_STS_AGAIN without polling
authorMike Snitzer <snitzer@kernel.org>
Tue, 21 Jun 2022 17:37:06 +0000 (13:37 -0400)
committerMike Snitzer <snitzer@kernel.org>
Tue, 21 Jun 2022 17:48:52 +0000 (13:48 -0400)
Commit 5291984004edf ("dm: fix bio polling to handle possibile
BLK_STS_AGAIN") inadvertently introduced an early return from
dm_io_complete() without first queueing the bio to DM if BLK_STS_AGAIN
occurs and bio-polling is _not_ being used.

Fix this by only returning early from dm_io_complete() if the bio has
first been properly queued to DM. Otherwise, the bio will never finish
via bio_endio.

Fixes: 5291984004edf ("dm: fix bio polling to handle possibile BLK_STS_AGAIN")
Cc: stable@vger.kernel.org
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
drivers/md/dm.c

index b6b25d3..9ede552 100644 (file)
@@ -939,9 +939,11 @@ static void dm_io_complete(struct dm_io *io)
                        if (io_error == BLK_STS_AGAIN) {
                                /* io_uring doesn't handle BLK_STS_AGAIN (yet) */
                                queue_io(md, bio);
+                               return;
                        }
                }
-               return;
+               if (io_error == BLK_STS_DM_REQUEUE)
+                       return;
        }
 
        if (bio_is_flush_with_data(bio)) {