Automated cherry pick of #25390: Fix/get isolated dev by addr3

pkg/apis/compute/guest_metadata.go

@@ -24,6 +24,8 @@ const (
 	DISK_CLONE_TASK_ID = "__disk_clone_task_id"
 
 	SSH_PORT = "__ssh_port"
+
+	DAEMON_GUEST_MANUAL_STOP = "daemon_guest_manual_stop"
 )
 
 const BASE_INSTANCE_SNAPSHOT_ID = "__base_instance_snapshot_id"

pkg/compute/guestdrivers/kvm.go

@@ -273,6 +273,12 @@ func (self *SKVMGuestDriver) RequestStopOnHost(ctx context.Context, guest *model
 		body.Set("is_force", jsonutils.JSONTrue)
 	}
 	body.Add(jsonutils.NewInt(timeout), "timeout")
+	if guest.IsDaemon.IsTrue() {
+		val := guest.GetMetadata(ctx, api.DAEMON_GUEST_MANUAL_STOP, task.GetUserCred())
+		if len(val) > 0 {
+			body.Set("daemon_guest_manual_stop", jsonutils.JSONTrue)
+		}
+	}
 
 	header := self.getTaskRequestHeader(task)
 

pkg/compute/models/guest_actions.go

@@ -1272,6 +1272,9 @@ func (self *SGuest) PerformStart(
 			}
 		}
 		if self.isAllDisksReady() {
+			if self.IsDaemon.IsTrue() {
+				self.SetMetadata(ctx, api.DAEMON_GUEST_MANUAL_STOP, "", userCred)
+			}
 			kwargs := jsonutils.Marshal(input).(*jsonutils.JSONDict)
 			driver, err := self.GetDriver()
 			if err != nil {
@@ -4037,6 +4040,9 @@ func (self *SGuest) PerformStatus(ctx context.Context, userCred mcclient.TokenCr
 // 关机
 func (self *SGuest) PerformStop(ctx context.Context, userCred mcclient.TokenCredential, query jsonutils.JSONObject,
 	input api.ServerStopInput) (jsonutils.JSONObject, error) {
+	if self.IsDaemon.IsTrue() {
+		self.SetMetadata(ctx, api.DAEMON_GUEST_MANUAL_STOP, true, userCred)
+	}
 	drv, err := self.GetDriver()
 	if err != nil {
 		return nil, errors.Wrap(err, "GetDriver")

pkg/hostman/guestman/guesthandlers/guesthandler.go

@@ -237,12 +237,13 @@ func guestStart(ctx context.Context, userCred mcclient.TokenCredential, sid stri
 }
 
 func guestStop(ctx context.Context, userCred mcclient.TokenCredential, sid string, body jsonutils.JSONObject) (interface{}, error) {
+	daemonGuestManualStop := jsonutils.QueryBoolean(body, "daemon_guest_manual_stop", false)
 	timeout, err := body.Int("timeout")
 	if err != nil {
 		timeout = 30
 	}
 	forceStop := jsonutils.QueryBoolean(body, "is_force", false)
-	return nil, guestman.GetGuestManager().GuestStop(ctx, sid, timeout, forceStop)
+	return nil, guestman.GetGuestManager().GuestStop(ctx, sid, timeout, forceStop, daemonGuestManualStop)
 }
 
 func guestMonitor(ctx context.Context, userCred mcclient.TokenCredential, sid string, body jsonutils.JSONObject) (interface{}, error) {

pkg/hostman/guestman/guestman.go

@@ -62,6 +62,7 @@ import (
 	"yunion.io/x/onecloud/pkg/hostman/storageman/remotefile"
 	"yunion.io/x/onecloud/pkg/httperrors"
 	"yunion.io/x/onecloud/pkg/mcclient"
+	"yunion.io/x/onecloud/pkg/mcclient/auth"
 	modules "yunion.io/x/onecloud/pkg/mcclient/modules/compute"
 	"yunion.io/x/onecloud/pkg/util/cgrouputils"
 	"yunion.io/x/onecloud/pkg/util/cgrouputils/cpuset"
@@ -503,6 +504,32 @@ func (m *SGuestManager) OnVerifyExistingGuestsSucc(servers []jsonutils.JSONObjec
 			m.RemoveCandidateServer(server)
 		}
 	}
+	timeutils2.AddTimeout(60*time.Second, func() { m.checkDaemonGuestsIsRunning() })
+}
+
+func (m *SGuestManager) checkDaemonGuestsIsRunning() {
+	m.Servers.Range(func(k, v interface{}) bool {
+		guest := v.(*SKVMGuestInstance)
+		if !guest.IsDaemon() {
+			return true
+		}
+		if guest.IsRunning() || guest.IsSuspend() {
+			return true
+		}
+
+		if guest.StartupTask != nil {
+			return true
+		}
+		if guest.isDaemonGuestManualStop() {
+			return true
+		}
+		if err := guest.StartGuest(context.Background(), auth.AdminCredential(), jsonutils.NewDict()); err != nil {
+			log.Errorf("checkDaemonGuestsIsRunning start guest %s failed: %s", guest.GetName(), err.Error())
+		}
+		return true
+	})
+
+	timeutils2.AddTimeout(60*time.Second, func() { m.checkDaemonGuestsIsRunning() })
 }
 
 func (m *SGuestManager) RemoveCandidateServer(server GuestRuntimeInstance) {
@@ -1185,9 +1212,9 @@ func (m *SGuestManager) GuestStart(ctx context.Context, userCred mcclient.TokenC
 	}
 }
 
-func (m *SGuestManager) GuestStop(ctx context.Context, sid string, timeout int64, isForce bool) error {
+func (m *SGuestManager) GuestStop(ctx context.Context, sid string, timeout int64, isForce, daemonGuestManualStop bool) error {
 	if server, ok := m.GetServer(sid); ok {
-		if err := server.HandleStop(ctx, timeout, isForce); err != nil {
+		if err := server.HandleStop(ctx, timeout, isForce, daemonGuestManualStop); err != nil {
 			return errors.Wrap(err, "Do stop")
 		}
 	} else {

pkg/hostman/guestman/pci.go

@@ -344,7 +344,13 @@ func (s *SKVMGuestInstance) initIsolatedDevices(pciRoot, pciBridge *desc.PCICont
 
 	manager := s.manager.GetHost().GetIsolatedDeviceManager()
 	for i := 0; i < len(s.Desc.IsolatedDevices); i++ {
-		dev := manager.GetDeviceByAddr(s.Desc.IsolatedDevices[i].Addr)
+		dev := manager.GetDeviceByIdent(s.Desc.IsolatedDevices[i].VendorDeviceId, s.Desc.IsolatedDevices[i].Addr, s.Desc.IsolatedDevices[i].MdevId)
+		if dev == nil {
+			log.Errorf("failed find dev by %s %s %s",
+				s.Desc.IsolatedDevices[i].VendorDeviceId, s.Desc.IsolatedDevices[i].Addr, s.Desc.IsolatedDevices[i].MdevId)
+			continue
+		}
+
 		if s.Desc.IsolatedDevices[i].DevType == compute.USB_TYPE {
 			s.Desc.IsolatedDevices[i].Usb = desc.NewUsbDevice("usb-host", dev.GetQemuId())
 			s.Desc.IsolatedDevices[i].Usb.Options = dev.GetPassthroughOptions()

pkg/hostman/guestman/pod.go

@@ -606,7 +606,7 @@ func (s *sPodGuestInstance) HandleGuestStart(ctx context.Context, userCred mccli
 	return nil, nil
 }
 
-func (s *sPodGuestInstance) HandleStop(ctx context.Context, timeout int64, isForce bool) error {
+func (s *sPodGuestInstance) HandleStop(ctx context.Context, timeout int64, isForce, daemonGuestManualStop bool) error {
 	hostutils.DelayTask(ctx, func(ctx context.Context, params interface{}) (jsonutils.JSONObject, error) {
 		err := s.stopPod(ctx, timeout)
 		if err != nil {

pkg/hostman/guestman/qemu-kvm.go

@@ -758,6 +758,7 @@ func (s *SKVMGuestInstance) asyncScriptStart(ctx context.Context, params interfa
 	}
 
 	if err != nil {
+		log.Errorf("asyncScriptStart init desc failed %s", err)
 		if ctx != nil && len(appctx.AppContextTaskId(ctx)) >= 0 {
 			hostutils.TaskFailed(ctx, fmt.Sprintf("Async start server failed: %s", err))
 		}
@@ -1988,11 +1989,15 @@ func (s *SKVMGuestInstance) StartGuest(ctx context.Context, userCred mcclient.To
 	return nil
 }
 
-func (s *SKVMGuestInstance) HandleStop(ctx context.Context, timeout int64, isForce bool) error {
+func (s *SKVMGuestInstance) HandleStop(ctx context.Context, timeout int64, isForce, daemonGuestManualStop bool) error {
 	params := &SGuestStopParams{
 		IsForce: isForce,
 		Timeout: timeout,
 	}
+	if daemonGuestManualStop {
+		s.Desc.Metadata[api.DAEMON_GUEST_MANUAL_STOP] = "true"
+		SaveLiveDesc(s, s.Desc)
+	}
 	hostutils.DelayTaskWithoutReqctx(ctx, s.ExecStopTask, params)
 	return nil
 }

pkg/hostman/guestman/qemu-kvmhelper.go

@@ -270,6 +270,10 @@ func (s *SKVMGuestInstance) isDisableAutoMergeSnapshots() bool {
 	return s.Desc.Metadata[api.VM_METADATA_DISABLE_AUTO_MERGE_SNAPSHOT] == "true"
 }
 
+func (s *SKVMGuestInstance) isDaemonGuestManualStop() bool {
+	return s.Desc.Metadata[api.DAEMON_GUEST_MANUAL_STOP] == "true"
+}
+
 func (s *SKVMGuestInstance) getMachine() string {
 	machine := s.Desc.Machine
 	if machine == "" {
@@ -1162,7 +1166,12 @@ func (s *SKVMGuestInstance) startMemCleaner() error {
 
 func (s *SKVMGuestInstance) gpusHasVga() bool {
 	for i := 0; i < len(s.Desc.IsolatedDevices); i++ {
-		if s.Desc.IsolatedDevices[i].GpuType == api.GPU_VGA {
+		manager := s.manager.GetHost().GetIsolatedDeviceManager()
+		dev := manager.GetDeviceByAddr(s.Desc.IsolatedDevices[i].Addr)
+		if dev == nil {
+			continue
+		}
+		if dev.GetDeviceType() == api.GPU_VGA {
 			return true
 		}
 	}
@@ -1173,6 +1182,9 @@ func (s *SKVMGuestInstance) hasGPU() bool {
 	manager := s.manager.GetHost().GetIsolatedDeviceManager()
 	for i := 0; i < len(s.Desc.IsolatedDevices); i++ {
 		dev := manager.GetDeviceByAddr(s.Desc.IsolatedDevices[i].Addr)
+		if dev == nil {
+			continue
+		}
 		if dev.GetDeviceType() == api.GPU_TYPE {
 			return true
 		}

pkg/hostman/guestman/runtime.go

@@ -72,7 +72,7 @@ type GuestRuntimeInstance interface {
 	PostUploadStatus(resp *computeapi.HostUploadGuestStatusInput, reason string)
 	HandleGuestStart(ctx context.Context, userCred mcclient.TokenCredential, body jsonutils.JSONObject) (jsonutils.JSONObject, error)
 
-	HandleStop(ctx context.Context, timeout int64, isForce bool) error
+	HandleStop(ctx context.Context, timeout int64, isForce, daemonGuestManualStop bool) error
 
 	LoadDesc() error
 	PostLoad(m *SGuestManager) error