dm mpath: requeue after a small delay if blk_get_request() fails

If blk_get_request() returns ENODEV then multipath_clone_and_map()
causes a request to be requeued immediately. This can cause a kworker
thread to spend 100% of the CPU time of a single core in
__blk_mq_run_hw_queue() and also can cause device removal to never
finish.

Avoid this by only requeuing after a delay if blk_get_request() fails.
Additionally, reduce the requeue delay.

Cc: stable@vger.kernel.org # 4.9+
Signed-off-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>

drivers/md/dm-mpath.c

@@ -484,7 +484,6 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
 				   struct request **__clone)
 {
 	struct multipath *m = ti->private;
-	int r = DM_MAPIO_REQUEUE;
 	size_t nr_bytes = blk_rq_bytes(rq);
 	struct pgpath *pgpath;
 	struct block_device *bdev;
@@ -503,7 +502,7 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
 	} else if (test_bit(MPATHF_QUEUE_IO, &m->flags) ||
 		   test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
 		pg_init_all_paths(m);
-		return r;
+		return DM_MAPIO_REQUEUE;
 	}
 
 	memset(mpio, 0, sizeof(*mpio));
@@ -517,7 +516,7 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
 			GFP_ATOMIC);
 	if (IS_ERR(clone)) {
 		/* EBUSY, ENODEV or EWOULDBLOCK: requeue */
-		return r;
+		return DM_MAPIO_DELAY_REQUEUE;
 	}
 	clone->bio = clone->biotail = NULL;
 	clone->rq_disk = bdev->bd_disk;

drivers/md/dm-rq.c

@@ -280,7 +280,7 @@ static void dm_requeue_original_request(struct dm_rq_target_io *tio, bool delay_
 	if (!rq->q->mq_ops)
 		dm_old_requeue_request(rq);
 	else
-		dm_mq_delay_requeue_request(rq, delay_requeue ? 5000 : 0);
+		dm_mq_delay_requeue_request(rq, delay_requeue ? 100/*ms*/ : 0);
 
 	rq_completed(md, rw, false);
 }