nvmet: fix false keep-alive timeout when a controller is torn down

[ Upstream commit aaeadd7 ]

Controller teardown flow may take some time in case it has many I/O
queues, and the host may not send us keep-alive during this period.
Hence reset the traffic based keep-alive timer so we don't trigger
a controller teardown as a result of a keep-alive expiration.

Reported-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>

drivers/nvme/target/core.c

@@ -388,10 +388,10 @@ static void nvmet_keep_alive_timer(struct work_struct *work)
 {
 	struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
 			struct nvmet_ctrl, ka_work);
-	bool cmd_seen = ctrl->cmd_seen;
+	bool reset_tbkas = ctrl->reset_tbkas;
 
-	ctrl->cmd_seen = false;
-	if (cmd_seen) {
+	ctrl->reset_tbkas = false;
+	if (reset_tbkas) {
 		pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
 			ctrl->cntlid);
 		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
@@ -804,6 +804,13 @@ void nvmet_sq_destroy(struct nvmet_sq *sq)
 	percpu_ref_exit(&sq->ref);
 
 	if (ctrl) {
+		/*
+		 * The teardown flow may take some time, and the host may not
+		 * send us keep-alive during this period, hence reset the
+		 * traffic based keep-alive timer so we don't trigger a
+		 * controller teardown as a result of a keep-alive expiration.
+		 */
+		ctrl->reset_tbkas = true;
 		nvmet_ctrl_put(ctrl);
 		sq->ctrl = NULL; /* allows reusing the queue later */
 	}
@@ -953,7 +960,7 @@ bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
 	}
 
 	if (sq->ctrl)
-		sq->ctrl->cmd_seen = true;
+		sq->ctrl->reset_tbkas = true;
 
 	return true;
 

drivers/nvme/target/nvmet.h

@@ -166,7 +166,7 @@ struct nvmet_ctrl {
 	struct nvmet_subsys	*subsys;
 	struct nvmet_sq		**sqs;
 
-	bool			cmd_seen;
+	bool			reset_tbkas;
 
 	struct mutex		lock;
 	u64			cap;