[PATCH 1/2] blk-mq: fix race between timeout and queue_rq

[Ming Lei <ming.lei@canonical.com>]

Either the request is from requeue or just being allocated from
tag pool, its REQ_ATOM_STARTED flag has been cleared already, so
don't test it in blk_mq_start_request().

One memory barrier is needed between writing rq->deadline and
setting REQ_ATOM_STARTED so that timeout can't happen too early
if timeout handler reads obsolete rq->deadline caused by out
of order of the two writes.

Signed-off-by: Ming Lei <ming.lei@canonical.com>
---
 block/blk-mq.c |   11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 47f3938..957815e 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -377,13 +377,20 @@ static void blk_mq_start_request(struct request *rq, bool last)
 	blk_add_timer(rq);
 
 	/*
+	 * Order writing rq->deadline and setting the rq's
+	 * REQ_ATOM_STARTED flag so that timeout handler can see
+	 * correct rq->deadline once the rq's REQ_ATOM_STARTED flag
+	 * is observed.
+	 */
+	smp_mb__before_atomic();
+
+	/*
 	 * Mark us as started and clear complete. Complete might have been
 	 * set if requeue raced with timeout, which then marked it as
 	 * complete. So be sure to clear complete again when we start
 	 * the request, otherwise we'll ignore the completion event.
 	 */
-	if (!test_bit(REQ_ATOM_STARTED, &rq->atomic_flags))
-		set_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
+	set_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
 	if (test_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags))
 		clear_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags);
 
-- 
1.7.9.5