lib/nvme: handle qpair state in transport layer.

The state should be changed and checked by the transport
layer. All transports should follow the same list of steps
when disconnecting/reconnecting.

Signed-off-by: Seth Howell <seth.howell@intel.com>
Change-Id: If2647624345f2c70f78a20bba4e2206d2762f120
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1853
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>

lib/nvme/nvme_ctrlr.c

@@ -472,9 +472,6 @@ spdk_nvme_ctrlr_reconnect_io_qpair(struct spdk_nvme_qpair *qpair)
 		goto out;
 	}
 
-	/* We have to confirm that any old memory is cleaned up. */
-	nvme_transport_ctrlr_disconnect_qpair(ctrlr, qpair);
-
 	rc = nvme_transport_ctrlr_connect_qpair(ctrlr, qpair);
 	if (rc) {
 		rc = -EAGAIN;
@@ -1165,9 +1162,7 @@ spdk_nvme_ctrlr_reset(struct spdk_nvme_ctrlr *ctrlr)
 	/* Disable all queues before disabling the controller hardware. */
 	TAILQ_FOREACH(qpair, &ctrlr->active_io_qpairs, tailq) {
 		qpair->transport_failure_reason = SPDK_NVME_QPAIR_FAILURE_LOCAL;
-		nvme_qpair_set_state(qpair, NVME_QPAIR_DISCONNECTED);
 	}
-	nvme_qpair_set_state(ctrlr->adminq, NVME_QPAIR_DISCONNECTED);
 	nvme_qpair_complete_error_reqs(ctrlr->adminq);
 	nvme_transport_qpair_abort_reqs(ctrlr->adminq, 0 /* retry */);
 	ctrlr->adminq->transport_failure_reason = SPDK_NVME_QPAIR_FAILURE_LOCAL;

lib/nvme/nvme_internal.h

@@ -357,6 +357,7 @@ struct nvme_async_event_request {
 
 enum nvme_qpair_state {
 	NVME_QPAIR_DISCONNECTED,
+	NVME_QPAIR_DISCONNECTING,
 	NVME_QPAIR_CONNECTING,
 	NVME_QPAIR_CONNECTED,
 	NVME_QPAIR_ENABLING,

lib/nvme/nvme_rdma.c

@@ -330,18 +330,15 @@ nvme_rdma_qpair_process_cm_event(struct nvme_rdma_qpair *rqpair)
 			break;
 		case RDMA_CM_EVENT_DISCONNECTED:
 			rqpair->qpair.transport_failure_reason = SPDK_NVME_QPAIR_FAILURE_REMOTE;
-			nvme_qpair_set_state(&rqpair->qpair, NVME_QPAIR_DISCONNECTED);
 			break;
 		case RDMA_CM_EVENT_DEVICE_REMOVAL:
 			rqpair->qpair.transport_failure_reason = SPDK_NVME_QPAIR_FAILURE_LOCAL;
-			nvme_qpair_set_state(&rqpair->qpair, NVME_QPAIR_DISCONNECTED);
 			break;
 		case RDMA_CM_EVENT_MULTICAST_JOIN:
 		case RDMA_CM_EVENT_MULTICAST_ERROR:
 			break;
 		case RDMA_CM_EVENT_ADDR_CHANGE:
 			rqpair->qpair.transport_failure_reason = SPDK_NVME_QPAIR_FAILURE_LOCAL;
-			nvme_qpair_set_state(&rqpair->qpair, NVME_QPAIR_DISCONNECTED);
 			break;
 		case RDMA_CM_EVENT_TIMEWAIT_EXIT:
 			break;
@@ -1631,7 +1628,6 @@ nvme_rdma_ctrlr_disconnect_qpair(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme
 	struct nvme_rdma_ctrlr *rctrlr;
 	struct nvme_rdma_cm_event_entry *entry, *tmp;
 
-	nvme_qpair_set_state(qpair, NVME_QPAIR_DISCONNECTED);
 	nvme_rdma_unregister_mem(rqpair);
 	nvme_rdma_unregister_reqs(rqpair);
 	nvme_rdma_unregister_rsps(rqpair);
@@ -2007,7 +2003,7 @@ nvme_rdma_qpair_process_completions(struct spdk_nvme_qpair *qpair,
 	}
 	nvme_rdma_qpair_process_cm_event(rqpair);
 
-	if (spdk_unlikely(nvme_qpair_get_state(qpair) == NVME_QPAIR_DISCONNECTED)) {
+	if (spdk_unlikely(qpair->transport_failure_reason != SPDK_NVME_QPAIR_FAILURE_NONE)) {
 		goto fail;
 	}
 

lib/nvme/nvme_tcp.c

@@ -235,12 +235,6 @@ nvme_tcp_ctrlr_disconnect_qpair(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_
 	struct nvme_tcp_qpair *tqpair = nvme_tcp_qpair(qpair);
 	struct nvme_tcp_pdu *pdu;
 
-	if (nvme_qpair_get_state(qpair) == NVME_QPAIR_DISCONNECTED) {
-		/* Already disconnecting */
-		return;
-	}
-
-	nvme_qpair_set_state(qpair, NVME_QPAIR_DISCONNECTED);
 	spdk_sock_close(&tqpair->sock);
 
 	/* clear the send_queue */

lib/nvme/nvme_transport.c

@@ -280,12 +280,17 @@ int
 nvme_transport_ctrlr_connect_qpair(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_qpair *qpair)
 {
 	const struct spdk_nvme_transport *transport = nvme_get_transport(ctrlr->trid.trstring);
+	uint8_t transport_failure_reason;
 	int rc;
 
 	assert(transport != NULL);
 	if (!nvme_qpair_is_admin_queue(qpair)) {
 		qpair->transport = transport;
 	}
+
+	transport_failure_reason = qpair->transport_failure_reason;
+	qpair->transport_failure_reason = SPDK_NVME_QPAIR_FAILURE_NONE;
+
 	nvme_qpair_set_state(qpair, NVME_QPAIR_CONNECTING);
 	rc = transport->ops.ctrlr_connect_qpair(ctrlr, qpair);
 	if (rc != 0) {
@@ -300,10 +305,11 @@ nvme_transport_ctrlr_connect_qpair(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nv
 		}
 	}
 
-	qpair->transport_failure_reason = SPDK_NVME_QPAIR_FAILURE_NONE;
 	return rc;
 
 err:
+	/* If the qpair was unable to reconnect, restore the original failure reason. */
+	qpair->transport_failure_reason = transport_failure_reason;
 	nvme_transport_ctrlr_disconnect_qpair(ctrlr, qpair);
 	nvme_qpair_set_state(qpair, NVME_QPAIR_DISCONNECTED);
 	return rc;
@@ -314,11 +320,18 @@ nvme_transport_ctrlr_disconnect_qpair(struct spdk_nvme_ctrlr *ctrlr, struct spdk
 {
 	const struct spdk_nvme_transport *transport = nvme_get_transport(ctrlr->trid.trstring);
 
+	if (nvme_qpair_get_state(qpair) == NVME_QPAIR_DISCONNECTING ||
+	    nvme_qpair_get_state(qpair) == NVME_QPAIR_DISCONNECTED) {
+		return;
+	}
+
+	nvme_qpair_set_state(qpair, NVME_QPAIR_DISCONNECTING);
 	assert(transport != NULL);
 	if (qpair->poll_group) {
 		nvme_poll_group_deactivate_qpair(qpair);
 	}
 	transport->ops.ctrlr_disconnect_qpair(ctrlr, qpair);
+	nvme_qpair_set_state(qpair, NVME_QPAIR_DISCONNECTED);
 }
 
 void