Skip to contentSkip to conversation

blk-mq/nvme: cancel request synchronously

PATCH Superseded Ming Lei 2 participants 6 messages Mar 27, 2019 – Apr 2, 2019

Subsystems (2)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Ming Lei

author
#0 lore raw

v2 cover blk-mq/nvme: cancel request synchronously

Hi,

This patchset introduces blk_mq_complete_request_sync() for canceling request synchronously in error handler context, then one race between completing request remotely and destroying contoller/queues can be fixed.

V2:
- export via EXPORT_SYMBOL_GPL
- minor commit log change

Ming Lei (2):
blk-mq: introduce blk_mq_complete_request_sync()
nvme: cancel request synchronously

 block/blk-mq.c           | 20 ++++++++++++++++----
 drivers/nvme/host/core.c |  2 +-
 include/linux/blk-mq.h   |  1 +
 3 files changed, 18 insertions(+), 5 deletions(-)
Signature · 1 line
2.9.5
author
#1 lore raw

PATCH v2 1/2 blk-mq: introduce blk_mq_complete_request_sync()

In NVMe's error handler, follows the typical steps of tearing down hardware for recovering controller:

1) stop blk_mq hw queues
2) stop the real hw queues
3) cancel in-flight requests via blk_mq_tagset_busy_iter(tags, cancel_request, ...)
cancel_request():

	mark the request as abort
	blk_mq_complete_request(req);

4) destroy real hw queues

However, there may be race between #3 and #4, because blk_mq_complete_request() may run q->mq_ops->complete(rq) remotelly and asynchronously, and ->complete(rq) may be run after #4.

This patch introduces blk_mq_complete_request_sync() for fixing the above race.

Cc Keith Busch <kbusch at kernel.org>
Cc Sagi Grimberg <sagi at grimberg.me>
Cc Bart Van Assche <bvanassche at acm.org>
Cc James Smart <james.smart at broadcom.com>
Cc Christoph Hellwig <hch at lst.de>
Cc linux-nvme at lists.infradead.org
Reviewed-by Christoph Hellwig <hch at lst.de>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 block/blk-mq.c         | 20 ++++++++++++++++----
 include/linux/blk-mq.h |  1 +
 2 files changed, 17 insertions(+), 4 deletions(-)
block/blk-mq.c +16 -4
index a9c181603cbd..bc3524428b96 100644
@@ -569,7 +569,7 @@ static void __blk_mq_complete_request_remote(void *data)
 	q->mq_ops->complete(rq);
 }
 
-static void __blk_mq_complete_request(struct request *rq)
+static void __blk_mq_complete_request(struct request *rq, bool sync)
 {
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
 	struct request_queue *q = rq->q;
@@ -586,7 +586,7 @@ static void __blk_mq_complete_request(struct request *rq)
 	 * So complete IO reqeust in softirq context in case of single queue
 	 * for not degrading IO performance by irqsoff latency.
 	 */
-	if (q->nr_hw_queues == 1) {
+	if (q->nr_hw_queues == 1 && !sync) {
 		__blk_complete_request(rq);
 		return;
 	}
@@ -594,8 +594,11 @@ static void __blk_mq_complete_request(struct request *rq)
 	/*
 	 * For a polled request, always complete locallly, it's pointless
 	 * to redirect the completion.
+	 *
+	 * If driver requires to complete the request synchronously,
+	 * complete it locally, and it is usually done in error handler.
 	 */
-	if ((rq->cmd_flags & REQ_HIPRI) ||
+	if ((rq->cmd_flags & REQ_HIPRI) || sync ||
 	    !test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) {
 		q->mq_ops->complete(rq);
 		return;
@@ -648,11 +651,20 @@ bool blk_mq_complete_request(struct request *rq)
 {
 	if (unlikely(blk_should_fake_timeout(rq->q)))
 		return false;
-	__blk_mq_complete_request(rq);
+	__blk_mq_complete_request(rq, false);
 	return true;
 }
 EXPORT_SYMBOL(blk_mq_complete_request);
 
+bool blk_mq_complete_request_sync(struct request *rq)
+{
+	if (unlikely(blk_should_fake_timeout(rq->q)))
+		return false;
+	__blk_mq_complete_request(rq, true);
+	return true;
+}
+EXPORT_SYMBOL_GPL(blk_mq_complete_request_sync);
+
 int blk_mq_request_started(struct request *rq)
 {
 	return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
include/linux/blk-mq.h +1 -0
index b0c814bcc7e3..6a514e5136f4 100644
@@ -305,6 +305,7 @@ void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
 void blk_mq_kick_requeue_list(struct request_queue *q);
 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
 bool blk_mq_complete_request(struct request *rq);
+bool blk_mq_complete_request_sync(struct request *rq);
 bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list,
 			   struct bio *bio);
 bool blk_mq_queue_stopped(struct request_queue *q);
Signature · 1 line
2.9.5
maintainer
#2 lore raw

Re: [PATCH V2 1/2] blk-mq: introduce blk_mq_complete_request_sync()

block/blk-mq.c static void __blk_mq_complete_request(struct request *rq) · line 604

Quoted 13 lines · Ming Lei
block/blk-mq.c +4 -1
@@ -594,8 +594,11 @@ static void __blk_mq_complete_request(struct request *rq)
 	/*
 	 * For a polled request, always complete locallly, it's pointless
 	 * to redirect the completion.
+	 *
+	 * If driver requires to complete the request synchronously,
+	 * complete it locally, and it is usually done in error handler.
 	 */
-	if ((rq->cmd_flags & REQ_HIPRI) ||
+	if ((rq->cmd_flags & REQ_HIPRI) || sync ||
 	    !test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) {
 		q->mq_ops->complete(rq);
 		return;

I think this is fine and agree this is fixing a problem, but just considering an alternative: would it be okay to just temporarily clear QUEUE_FLAG_SAME_COMP in the necessary error handling that wants a sync completion so that we're not introducing an additional parameter?

author
#3 lore raw

Re: [PATCH V2 1/2] blk-mq: introduce blk_mq_complete_request_sync()

Quoted 11 lines · Keith Busch
Quoted 10 lines · Ming Lei
block/blk-mq.c +4 -1
@@ -594,8 +594,11 @@ static void __blk_mq_complete_request(struct request *rq)
 	/*
 	 * For a polled request, always complete locallly, it's pointless
 	 * to redirect the completion.
+	 *
+	 * If driver requires to complete the request synchronously,
+	 * complete it locally, and it is usually done in error handler.
 	 */
-	if ((rq->cmd_flags & REQ_HIPRI) ||
+	if ((rq->cmd_flags & REQ_HIPRI) || sync ||
Quoted 8 lines

block/blk-mq.c static void __blk_mq_complete_request(struct request *rq) · line 604

Quoted 3 lines
block/blk-mq.c +0 -0
 	    !test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) {
 		q->mq_ops->complete(rq);
 		return;

I think this is fine and agree this is fixing a problem, but just considering an alternative: would it be okay to just temporarily clear QUEUE_FLAG_SAME_COMP in the necessary error handling that wants a sync completion so that we're not introducing an additional parameter?

But the flag can be changed by sysfs too.

Thanks,
Ming

author
#4 lore raw

PATCH v2 2/2 nvme: cancel request synchronously

In reply to Ming Lei

nvme_cancel_request() is used in error handler, and it is always reliable to cancel request synchronously, and avoids possible race in which request may be completed after real hw queue is destroyed.

One issue is reported by our customer on NVMe RDMA, in which freed ib queue pair may be used in nvme_rdma_complete_rq().

Cc Keith Busch <kbusch at kernel.org>
Cc Sagi Grimberg <sagi at grimberg.me>
Cc Bart Van Assche <bvanassche at acm.org>
Cc James Smart <james.smart at broadcom.com>
Cc Christoph Hellwig <hch at lst.de>
Cc linux-nvme at lists.infradead.org
Reviewed-by Christoph Hellwig <hch at lst.de>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/nvme/host/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/core.c +1 -1
index 470601980794..2c43e12b70af 100644
@@ -288,7 +288,7 @@ bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 				"Cancelling I/O %d", req->tag);
 
 	nvme_req(req)->status = NVME_SC_ABORT_REQ;
-	blk_mq_complete_request(req);
+	blk_mq_complete_request_sync(req);
 	return true;
 }
 EXPORT_SYMBOL_GPL(nvme_cancel_request);
Signature · 1 line
2.9.5
author
#5 lore raw
In reply to Ming Lei
Quoted 11 lines · Ming Lei

Hi,

This patchset introduces blk_mq_complete_request_sync() for canceling request synchronously in error handler context, then one race between completing request remotely and destroying contoller/queues can be fixed.

V2:
- export via EXPORT_SYMBOL_GPL
- minor commit log change

Quoted 8 lines

Ming Lei (2):
blk-mq: introduce blk_mq_complete_request_sync()
nvme: cancel request synchronously

 block/blk-mq.c           | 20 ++++++++++++++++----
 drivers/nvme/host/core.c |  2 +-
 include/linux/blk-mq.h   |  1 +
 3 files changed, 18 insertions(+), 5 deletions(-)

Hi Jens,

Gentle ping on this fix...

Thanks,
Ming