fix(initiator): persist SubsystemNQN after NVMe/TCP connect

Persist the discovered SubsystemNQN immediately after a successful
discover-and-connect operation.

This keeps initiator state aligned with the actual connected NVMe/TCP
target even when controllerName validation fails afterwards.

Without this update, reload/load/cleanup paths may use a stale
SubsystemNQN and fail to locate or disconnect the correct NVMe device.

Longhorn 7124

Signed-off-by: Derek Su <derek.su@suse.com>

Author: derekbit

pkg/initiator/initiator.go

@@ -423,11 +423,9 @@ func (i *Initiator) StartNvmeTCPInitiator(transportAddress, transportServiceID s
 	if err != nil {
 		return dmDeviceIsBusy, errors.Wrapf(err, "failed to discover and connect NVMe/TCP target %s:%s", transportAddress, transportServiceID)
 	}
-	if controllerName == "" {
-		return dmDeviceIsBusy, fmt.Errorf("controller name is empty after discovering and connecting NVMe/TCP target %s:%s", transportAddress, transportServiceID)
+	if err := i.recordConnectedNVMeTCPInfo(subsystemNQN, controllerName); err != nil {
+		return dmDeviceIsBusy, errors.Wrapf(err, "invalid connected NVMe/TCP target state %s:%s", transportAddress, transportServiceID)
 	}
-	i.NVMeTCPInfo.SubsystemNQN = subsystemNQN
-	i.NVMeTCPInfo.ControllerName = controllerName
 
 	i.logger.Info("Loading NVMe/TCP initiator device info")
 	err = i.waitAndLoadNVMeDeviceInfoWithoutLock(transportAddress, transportServiceID)
@@ -639,6 +637,25 @@ func (i *Initiator) waitAndLoadNVMeDeviceInfoWithoutLock(transportAddress, trans
 	return nil
 }
 
+func (i *Initiator) recordConnectedNVMeTCPInfo(subsystemNQN, controllerName string) error {
+	if i.NVMeTCPInfo == nil {
+		return fmt.Errorf("nvmeTCPInfo is nil")
+	}
+
+	// Persist the discovered subsystem immediately after a successful connect.
+	// Later reload/load/cleanup paths use SubsystemNQN to locate or disconnect
+	// the NVMe device, so it must stay in sync even if controllerName validation fails.
+	if subsystemNQN != "" {
+		i.NVMeTCPInfo.SubsystemNQN = subsystemNQN
+	}
+	if controllerName == "" {
+		return fmt.Errorf("controller name is empty")
+	}
+
+	i.NVMeTCPInfo.ControllerName = controllerName
+	return nil
+}
+
 func (i *Initiator) discoverAndConnectNVMeTCPTarget(transportAddress, transportServiceID string, maxRetries int, retryInterval time.Duration) (subsystemNQN, controllerName string, err error) {
 	if i.NVMeTCPInfo == nil {
 		return "", "", fmt.Errorf("nvmeTCPInfo is nil")

pkg/initiator/initiator_test.go

@@ -179,6 +179,29 @@ func (s *InitiatorTestSuite) TestLoadEndpointForNvmeTcpFrontendNilInfo(c *C) {
 	c.Assert(err.Error(), Matches, ".*nvmeTCPInfo is nil.*")
 }
 
+func (s *InitiatorTestSuite) TestRecordConnectedNVMeTCPInfoUpdatesSubsystemBeforeControllerValidation(c *C) {
+	i := &Initiator{
+		NVMeTCPInfo: &NVMeTCPInfo{
+			SubsystemNQN:   "nqn.old",
+			ControllerName: "nvme-old",
+		},
+	}
+
+	err := i.recordConnectedNVMeTCPInfo("nqn.new", "")
+	c.Assert(err, NotNil)
+	c.Assert(err.Error(), Equals, "controller name is empty")
+	c.Assert(i.NVMeTCPInfo.SubsystemNQN, Equals, "nqn.new")
+	c.Assert(i.NVMeTCPInfo.ControllerName, Equals, "nvme-old")
+}
+
+func (s *InitiatorTestSuite) TestRecordConnectedNVMeTCPInfoNilInfo(c *C) {
+	i := &Initiator{}
+
+	err := i.recordConnectedNVMeTCPInfo("nqn.new", "nvme1")
+	c.Assert(err, NotNil)
+	c.Assert(err.Error(), Equals, "nvmeTCPInfo is nil")
+}
+
 func (s *InitiatorTestSuite) TestIsNamespaceExist(c *C) {
 	i := &Initiator{}
 	c.Assert(i.isNamespaceExist([]string{"nvme0n1"}), Equals, false)