Re: [PATCH v4] blk-mq: Fix race conditions in request timeout handling

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hello,

On Tue, Apr 10, 2018 at 08:40:43AM -0700, tj@xxxxxxxxxx wrote:
> On Tue, Apr 10, 2018 at 11:38:18PM +0800, Ming Lei wrote:
> > I agree, the issue should be in driver's irq handler and .timeout in
> > theory.
> > 
> > For example, even though one request has been done by irq handler, .timeout
> > still may return RESET_TIMER.
> 
> blk-mq can use a separate flag to track normal completions during
> timeout and complete the request normally on RESET_TIMER if the flag
> is set while EH was in progress.  With a bit of care, we'd be able to
> plug the race completely.

So, something like the following.  Just compile tested.  The extra
dedicated field is kinda unfortunate but we need something like that
no matter how we do this because the involved completion marking
violates the ownership (we want to record the normail completion while
timeout handling is in progress).  Alternatively, we can add an atomic
flags field.  The addition doesn't change the size of struct request
because it fits in a hole but that hole seems removeable, so...

Anyways, it'd be great if someone can test this combined with the
previous two rcu sync patches.

Thanks.
---
 block/blk-mq.c         |   44 +++++++++++++++++++++++++++++++-------------
 include/linux/blkdev.h |    2 ++
 2 files changed, 33 insertions(+), 13 deletions(-)

--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -642,6 +642,8 @@ void blk_mq_complete_request(struct requ
 	hctx_lock(hctx, &srcu_idx);
 	if (blk_mq_rq_aborted_gstate(rq) != rq->gstate)
 		__blk_mq_complete_request(rq);
+	else
+		rq->missed_completion = true;
 	hctx_unlock(hctx, srcu_idx);
 }
 EXPORT_SYMBOL(blk_mq_complete_request);
@@ -881,7 +883,7 @@ static void blk_mq_check_expired(struct
 	}
 }
 
-static void blk_mq_timeout_sync_rcu(struct request_queue *q)
+static void blk_mq_timeout_sync_rcu(struct request_queue *q, bool clear)
 {
 	struct blk_mq_hw_ctx *hctx;
 	bool has_rcu = false;
@@ -896,7 +898,8 @@ static void blk_mq_timeout_sync_rcu(stru
 		else
 			synchronize_srcu(hctx->srcu);
 
-		hctx->need_sync_rcu = false;
+		if (clear)
+			hctx->need_sync_rcu = false;
 	}
 	if (has_rcu)
 		synchronize_rcu();
@@ -917,25 +920,37 @@ static void blk_mq_terminate_expired(str
 		blk_mq_rq_timed_out(hctx, rq, priv, reserved);
 }
 
-static void blk_mq_finish_timeout_reset(struct blk_mq_hw_ctx *hctx,
+static void blk_mq_timeout_reset_return(struct blk_mq_hw_ctx *hctx,
 		struct request *rq, void *priv, bool reserved)
 {
 	/*
 	 * @rq's timer reset has gone through rcu synchronization and is
 	 * visible now.  Allow normal completions again by resetting
 	 * ->aborted_gstate.  Don't clear RQF_MQ_TIMEOUT_RESET here as
-	 * there's no memory ordering around ->aborted_gstate making it the
-	 * only field safe to update.  Let blk_add_timer() clear it later
-	 * when the request is recycled or times out again.
-	 *
-	 * As nothing prevents from completion happening while
-	 * ->aborted_gstate is set, this may lead to ignored completions
-	 * and further spurious timeouts.
+	 * blk_mq_timeout_reset_cleanup() needs it again and there's no
+	 * memory ordering around ->aborted_gstate making it the only field
+	 * safe to update.  Let blk_add_timer() clear it later when the
+	 * request is recycled or times out again.
 	 */
 	if (rq->rq_flags & RQF_MQ_TIMEOUT_RESET)
 		blk_mq_rq_update_aborted_gstate(rq, 0);
 }
 
+static void blk_mq_timeout_reset_cleanup(struct blk_mq_hw_ctx *hctx,
+		struct request *rq, void *priv, bool reserved)
+{
+	/*
+	 * @rq is now fully returned to the normal path.  If normal
+	 * completion raced timeout handling, execute the missed completion
+	 * here.  This is safe because 1. ->missed_completion can no longer
+	 * be asserted because nothing is timing out right now and 2. if
+	 * ->missed_completion is set, @rq is safe from recycling because
+	 * nobody could have completed it.
+	 */
+	if ((rq->rq_flags & RQF_MQ_TIMEOUT_RESET) && rq->missed_completion)
+		blk_mq_complete_request(rq);
+}
+
 static void blk_mq_timeout_work(struct work_struct *work)
 {
 	struct request_queue *q =
@@ -976,7 +991,7 @@ static void blk_mq_timeout_work(struct w
 		 * becomes a problem, we can add per-hw_ctx rcu_head and
 		 * wait in parallel.
 		 */
-		blk_mq_timeout_sync_rcu(q);
+		blk_mq_timeout_sync_rcu(q, true);
 
 		/* terminate the ones we won */
 		blk_mq_queue_tag_busy_iter(q, blk_mq_terminate_expired,
@@ -988,9 +1003,12 @@ static void blk_mq_timeout_work(struct w
 		 * reset racing against recycling.
 		 */
 		if (nr_resets) {
-			blk_mq_timeout_sync_rcu(q);
+			blk_mq_timeout_sync_rcu(q, false);
+			blk_mq_queue_tag_busy_iter(q,
+					blk_mq_timeout_reset_return, NULL);
+			blk_mq_timeout_sync_rcu(q, true);
 			blk_mq_queue_tag_busy_iter(q,
-					blk_mq_finish_timeout_reset, NULL);
+					blk_mq_timeout_reset_cleanup, NULL);
 		}
 	}
 
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -227,6 +227,8 @@ struct request {
 
 	unsigned int extra_len;	/* length of alignment and padding */
 
+	bool missed_completion;
+
 	/*
 	 * On blk-mq, the lower bits of ->gstate (generation number and
 	 * state) carry the MQ_RQ_* state value and the upper bits the



[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]