fix(security): restrict service monitors to probe tasks

cmd/dashboard/controller/oauth2.go

@@ -26,8 +26,11 @@ import (
 // code to their own origin and bind the victim's identity. A request Host is
 // trusted only when it is an operator-declared dashboard host (the same
 // allowlist that guards NAT routing). Otherwise the redirect is pinned to the
-// operator-declared DashboardHost; when DashboardHost is empty the operator has
+// operator-declared DashboardHost. Empty DashboardHost intentionally retains
-// not pinned a dashboard origin, so the request Host is passed through.
+// dynamic/multi-domain deployments by passing through request Host; those
+// deployments must validate Host at their trusted proxy and register exact
+// redirect URIs at the OAuth provider. GHSA-rf68-8gjr-36q7 documents this
+// configuration boundary and must be updated if this compatibility changes.
 func getRedirectURL(c *gin.Context) string {
 	scheme := "http://"
 	referer := c.Request.Referer()

cmd/dashboard/controller/permission_matrix_test.go

@@ -261,8 +261,8 @@ func TestShowServiceFiltersCycleTransferStatsLikeServerList(t *testing.T) {
 	assert.NoError(t, singleton.DB.Create(&model.Server{Common: model.Common{ID: 3, UserID: 200}, Name: "hidden member server", UUID: "hidden-member-server", HideForGuest: true}).Error)
 	singleton.ServerShared = singleton.NewServerClass()
 
-	assert.NoError(t, singleton.DB.Create(&model.Service{Common: model.Common{ID: 10, UserID: 1}, Name: "shown service"}).Error)
+	assert.NoError(t, singleton.DB.Create(&model.Service{Common: model.Common{ID: 10, UserID: 1}, Name: "shown service", Type: model.TaskTypeTCPPing}).Error)
-	assert.NoError(t, singleton.DB.Create(&model.Service{Common: model.Common{ID: 11, UserID: 1}, Name: "hidden service", HideForGuest: true}).Error)
+	assert.NoError(t, singleton.DB.Create(&model.Service{Common: model.Common{ID: 11, UserID: 1}, Name: "hidden service", Type: model.TaskTypeTCPPing, HideForGuest: true}).Error)
 
 	originalServiceSentinel := singleton.ServiceSentinelShared
 	serviceSentinel, err := singleton.NewServiceSentinel(make(chan *model.Service, 2))

cmd/dashboard/controller/service.go

@@ -484,6 +484,9 @@ func createService(c *gin.Context) (uint64, error) {
 	if err := c.ShouldBindJSON(&mf); err != nil {
 		return 0, err
 	}
+	if err := model.ValidateServiceMonitorType(uint64(mf.Type)); err != nil {
+		return 0, err
+	}
 
 	if !isValidServiceCover(mf.Cover) {
 		return 0, singleton.Localizer.ErrorT("permission denied")
@@ -548,6 +551,9 @@ func updateService(c *gin.Context) (any, error) {
 	if err := c.ShouldBindJSON(&mf); err != nil {
 		return nil, err
 	}
+	if err := model.ValidateServiceMonitorType(uint64(mf.Type)); err != nil {
+		return nil, err
+	}
 
 	if !isValidServiceCover(mf.Cover) {
 		return nil, singleton.Localizer.ErrorT("permission denied")

cmd/dashboard/controller/service_type_security_test.go

@@ -0,0 +1,91 @@
+package controller
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+
+	"github.com/nezhahq/nezha/model"
+	"github.com/nezhahq/nezha/service/singleton"
+)
+
+func serviceTypeSecurityRouter() *gin.Engine {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+	r.Use(func(c *gin.Context) {
+		setAuthUser(c, 100, model.RoleMember)
+		c.Next()
+	})
+	r.POST("/api/v1/service", commonHandler(createService))
+	r.PATCH("/api/v1/service/:id", commonHandler(updateService))
+	return r
+}
+
+func serviceTypeSecurityBody(taskType uint8) []byte {
+	body, _ := json.Marshal(model.ServiceForm{
+		Name:        "service-type-security",
+		Target:      "example.invalid:443",
+		Type:        taskType,
+		Cover:       model.ServiceCoverIgnoreAll,
+		SkipServers: map[uint64]bool{1: true},
+		Duration:    30,
+	})
+	return body
+}
+
+func TestCreateServiceRejectsNonProbeTaskTypes(t *testing.T) {
+	setupCoverPATFixture(t)
+	r := serviceTypeSecurityRouter()
+
+	for _, taskType := range []uint8{0, model.TaskTypeCommand, model.TaskTypeApplyConfig, model.TaskTypeExec, 255} {
+		w := httptest.NewRecorder()
+		req := httptest.NewRequest(http.MethodPost, "/api/v1/service", bytes.NewReader(serviceTypeSecurityBody(taskType)))
+		req.Header.Set("Content-Type", "application/json")
+		r.ServeHTTP(w, req)
+
+		success, errMsg := decodeCommonResponseError(t, w.Body.Bytes())
+		require.False(t, success, "type %d must be rejected", taskType)
+		require.Contains(t, errMsg, "invalid service monitor type")
+	}
+
+	var count int64
+	require.NoError(t, singleton.DB.Model(&model.Service{}).Count(&count).Error)
+	require.Zero(t, count, "rejected task types must not reach persistence")
+}
+
+func TestUpdateServiceRejectsNonProbeTaskTypes(t *testing.T) {
+	setupCoverPATFixture(t)
+	r := serviceTypeSecurityRouter()
+	service := &model.Service{
+		Common:      model.Common{UserID: 100},
+		Name:        "valid-service",
+		Target:      "example.invalid:443",
+		Type:        model.TaskTypeTCPPing,
+		Cover:       model.ServiceCoverIgnoreAll,
+		SkipServers: map[uint64]bool{1: true},
+		Duration:    30,
+	}
+	require.NoError(t, singleton.DB.Create(service).Error)
+
+	for _, taskType := range []uint8{model.TaskTypeCommand, model.TaskTypeApplyConfig, model.TaskTypeExec, 255} {
+		w := httptest.NewRecorder()
+		path := fmt.Sprintf("/api/v1/service/%d", service.ID)
+		req := httptest.NewRequest(http.MethodPatch, path, bytes.NewReader(serviceTypeSecurityBody(taskType)))
+		req.Header.Set("Content-Type", "application/json")
+		r.ServeHTTP(w, req)
+
+		success, errMsg := decodeCommonResponseError(t, w.Body.Bytes())
+		require.False(t, success, "type %d must be rejected", taskType)
+		require.Contains(t, errMsg, "invalid service monitor type")
+
+		var persisted model.Service
+		require.NoError(t, singleton.DB.First(&persisted, service.ID).Error)
+		require.Equal(t, uint8(model.TaskTypeTCPPing), persisted.Type)
+	}
+}

cmd/dashboard/main.go

@@ -46,6 +46,12 @@ func initSystem(bus chan<- *model.Service) error {
 	if err := singleton.DB.Model(&model.User{}).Count(&usersCount).Error; err != nil {
 		return err
 	}
+	// Backward-compatible bootstrap state: existing installers and recovery
+	// procedures expect the first login on an empty database to be admin/admin.
+	// This is not a permanent credential or an authentication-bypass fallback;
+	// operators must complete initialization and change it before exposing the
+	// Dashboard. Replacing it requires a coordinated installer/migration flow so
+	// existing unattended installations are not locked out.
 	if usersCount == 0 {
 		hash, err := bcrypt.GenerateFromPassword([]byte("admin"), bcrypt.DefaultCost)
 		if err != nil {

cmd/dashboard/rpc/service_dispatch.go

@@ -14,6 +14,17 @@ func DispatchTask(serviceSentinelDispatchBus <-chan *model.Service) {
 		if task == nil {
 			continue
 		}
+		if err := model.ValidateServiceMonitorType(uint64(task.Type)); err != nil {
+			// Defense in depth for stale database rows and future internal callers:
+			// Service.Type shares its integer namespace with command/config tasks.
+			log.Printf("NEZHA>> DispatchTask rejected service %d: %v", task.ID, err)
+			continue
+		}
+		probe := task.PB()
+		if probe == nil {
+			log.Printf("NEZHA>> DispatchTask rejected service %d: invalid probe", task.ID)
+			continue
+		}
 
 		switch task.Cover {
 		case model.ServiceCoverIgnoreAll:
@@ -29,7 +40,7 @@ func DispatchTask(serviceSentinelDispatchBus <-chan *model.Service) {
 				if !canSendTaskToServer(task, server) {
 					continue
 				}
-				if err := server.SendTask(task.PB()); err != nil && !errors.Is(err, model.ErrTaskStreamOffline) {
+				if err := server.SendTask(probe); err != nil && !errors.Is(err, model.ErrTaskStreamOffline) {
 					log.Printf("NEZHA>> DispatchTask send error (server=%d): %v", id, err)
 				}
 			}
@@ -41,7 +52,7 @@ func DispatchTask(serviceSentinelDispatchBus <-chan *model.Service) {
 				if !canSendTaskToServer(task, server) {
 					continue
 				}
-				if err := server.SendTask(task.PB()); err != nil && !errors.Is(err, model.ErrTaskStreamOffline) {
+				if err := server.SendTask(probe); err != nil && !errors.Is(err, model.ErrTaskStreamOffline) {
 					log.Printf("NEZHA>> DispatchTask send error (server=%d): %v", id, err)
 				}
 			}

cmd/dashboard/rpc/service_dispatch_type_security_test.go

@@ -0,0 +1,59 @@
+package rpc
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/nezhahq/nezha/model"
+	"github.com/nezhahq/nezha/service/singleton"
+)
+
+func TestDispatchTaskSendsOnlyProbeTypes(t *testing.T) {
+	originalServerShared := singleton.ServerShared
+	originalUserInfo := singleton.UserInfoMap
+	t.Cleanup(func() {
+		singleton.ServerShared = originalServerShared
+		singleton.UserLock.Lock()
+		singleton.UserInfoMap = originalUserInfo
+		singleton.UserLock.Unlock()
+	})
+
+	server := &model.Server{Common: model.Common{ID: 1, UserID: 100}}
+	stream := &serveNATTaskStream{}
+	server.SetTaskStream(stream)
+	serverShared := singleton.NewEmptyServerClassForTest()
+	serverShared.InsertForTest(server)
+	singleton.ServerShared = serverShared
+	singleton.UserLock.Lock()
+	singleton.UserInfoMap = map[uint64]model.UserInfo{100: {Role: model.RoleMember}}
+	singleton.UserLock.Unlock()
+
+	bus := make(chan *model.Service, 8)
+	done := make(chan struct{})
+	go func() {
+		DispatchTask(bus)
+		close(done)
+	}()
+	for _, taskType := range []uint8{model.TaskTypeCommand, model.TaskTypeApplyConfig, model.TaskTypeExec, 255} {
+		bus <- &model.Service{
+			Common:      model.Common{ID: uint64(taskType), UserID: 100},
+			Type:        taskType,
+			Cover:       model.ServiceCoverIgnoreAll,
+			SkipServers: map[uint64]bool{1: true},
+		}
+	}
+	bus <- &model.Service{
+		Common:      model.Common{ID: 1000, UserID: 100},
+		Type:        model.TaskTypeTCPPing,
+		Target:      "example.invalid:443",
+		Cover:       model.ServiceCoverIgnoreAll,
+		SkipServers: map[uint64]bool{1: true},
+	}
+	close(bus)
+	<-done
+
+	require.Len(t, stream.sent, 1)
+	require.Equal(t, uint64(model.TaskTypeTCPPing), stream.sent[0].GetType())
+	require.Equal(t, uint64(1000), stream.sent[0].GetId())
+}

model/config.go

@@ -43,7 +43,12 @@ type ConfigForGuests struct {
 
 type ConfigDashboard struct {
 	InstallHost string `koanf:"install_host" json:"install_host,omitempty"`
-	AgentTLS    bool   `koanf:"tls" json:"tls,omitempty"` // 用于前端判断生成的安装命令是否启用 TLS
+	// AgentTLS controls the transport emitted by Agent installation commands.
+	// false intentionally supports trusted private networks and does not provide
+	// Dashboard peer authentication; Internet-facing control planes must use
+	// verified TLS. Changing this compatibility default belongs in the installer
+	// migration path, not in the gRPC task authorization model.
+	AgentTLS bool `koanf:"tls" json:"tls,omitempty"`
 
 	// DashboardHost 是 dashboard 对外访问的主机名，专用于 OAuth2 回调地址。
 	// 它与 InstallHost（agent 连接用主机名）解耦：两者可以是不同域名。

model/server_api.go

@@ -21,8 +21,12 @@ type StreamServerData struct {
 }
 
 type ServerForm struct {
-	Name                string              `json:"name,omitempty"`
+	Name string `json:"name,omitempty"`
-	Note                string              `json:"note,omitempty" validate:"optional"`           // 管理员可见备注
+	Note string `json:"note,omitempty" validate:"optional"` // 管理员可见备注
+	// PublicNote is opaque public metadata consumed by independently maintained
+	// user themes. The Dashboard stores/transports it but never renders it as
+	// HTML or navigates URL-like fields. Themes must validate schemes before
+	// using nested values such as customData.orderLink in href/window.open.
 	PublicNote          string              `json:"public_note,omitempty" validate:"optional"`    // 公开备注
 	DisplayIndex        int                 `json:"display_index,omitempty" default:"0"`          // 展示排序，越大越靠前
 	HideForGuest        bool                `json:"hide_for_guest,omitempty" validate:"optional"` // 对游客隐藏

model/service.go

@@ -39,6 +39,30 @@ const (
 	TaskTypeFsTransfer
 )
 
+// IsServiceMonitorType reports whether t is a passive service probe. Service
+// monitors and privileged Agent-control tasks share the protobuf Task.Type
+// namespace, so every path that persists, schedules, or dispatches a Service
+// must use this allowlist instead of accepting an arbitrary task integer.
+func IsServiceMonitorType(t uint64) bool {
+	switch t {
+	case TaskTypeHTTPGet, TaskTypeICMPPing, TaskTypeTCPPing:
+		return true
+	default:
+		return false
+	}
+}
+
+// ValidateServiceMonitorType returns an actionable error at API, model, and
+// scheduler boundaries. Keeping the check in model avoids a future caller
+// accidentally turning a monitor-only capability into Agent command/config
+// execution by copying Service.Type into pb.Task.Type.
+func ValidateServiceMonitorType(t uint64) error {
+	if !IsServiceMonitorType(t) {
+		return fmt.Errorf("invalid service monitor type %d: allowed types are 1 (HTTP GET), 2 (ICMP ping), and 3 (TCP ping)", t)
+	}
+	return nil
+}
+
 // IsMCPRPCResult 判定一个 TaskResult.Type 是否属于 MCP 走 RequestTask 通道的
 // 一次性 RPC 类型。dashboard 的 RequestTask 接收循环用它把这些回包路由到
 // Server.inflightRPC 等待方，而不是走 ServiceSentinel。
@@ -232,6 +256,9 @@ type Service struct {
 }
 
 func (m *Service) PB() *pb.Task {
+	if m == nil || !IsServiceMonitorType(uint64(m.Type)) {
+		return nil
+	}
 	return &pb.Task{
 		Id:   m.ID,
 		Type: uint64(m.Type),
@@ -300,6 +327,9 @@ func (m *Service) CronSpec() string {
 }
 
 func (m *Service) BeforeSave(tx *gorm.DB) error {
+	if err := ValidateServiceMonitorType(uint64(m.Type)); err != nil {
+		return err
+	}
 	if data, err := json.Marshal(m.SkipServers); err != nil {
 		return err
 	} else {
@@ -336,15 +366,9 @@ func (m *Service) AfterFind(tx *gorm.DB) error {
 	return nil
 }
 
-// IsServiceSentinelNeeded 判断该任务类型是否需要进行服务监控 需要则返回true
+// IsServiceSentinelNeeded accepts results only for the three probe types. An
+// unknown or privileged task type must never enter ServiceSentinel merely
+// because it was not listed in a denylist.
 func IsServiceSentinelNeeded(t uint64) bool {
-	switch t {
+	return IsServiceMonitorType(t)
-	case TaskTypeCommand, TaskTypeTerminalGRPC, TaskTypeUpgrade,
-		TaskTypeKeepalive, TaskTypeNAT, TaskTypeFM,
-		TaskTypeReportConfig, TaskTypeApplyConfig,
-		TaskTypeServerTransferApply:
-		return false
-	default:
-		return true
-	}
 }

model/service_type_security_test.go

@@ -0,0 +1,45 @@
+package model
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestServiceMonitorTypeAllowlist(t *testing.T) {
+	for _, taskType := range []uint64{TaskTypeHTTPGet, TaskTypeICMPPing, TaskTypeTCPPing} {
+		require.True(t, IsServiceMonitorType(taskType), "probe type %d must remain allowed", taskType)
+		require.NoError(t, ValidateServiceMonitorType(taskType))
+		require.True(t, IsServiceSentinelNeeded(taskType))
+	}
+
+	for _, taskType := range []uint64{
+		0,
+		TaskTypeCommand,
+		TaskTypeApplyConfig,
+		TaskTypeServerTransferApply,
+		TaskTypeExec,
+		TaskTypeFsTransfer,
+		255,
+	} {
+		require.False(t, IsServiceMonitorType(taskType), "privileged/unknown type %d must be rejected", taskType)
+		require.Error(t, ValidateServiceMonitorType(taskType))
+		require.False(t, IsServiceSentinelNeeded(taskType))
+	}
+}
+
+func TestServicePersistenceAndPBRejectPrivilegedTaskTypes(t *testing.T) {
+	for _, taskType := range []uint8{0, TaskTypeCommand, TaskTypeApplyConfig, TaskTypeExec, 255} {
+		service := &Service{Type: taskType}
+		require.Error(t, service.BeforeSave(nil), "type %d must not be persisted", taskType)
+		require.Nil(t, service.PB(), "type %d must not become an Agent task", taskType)
+	}
+
+	service := &Service{Common: Common{ID: 7}, Type: TaskTypeTCPPing, Target: "example.invalid:443"}
+	require.NoError(t, service.BeforeSave(nil))
+	task := service.PB()
+	require.NotNil(t, task)
+	require.Equal(t, uint64(7), task.GetId())
+	require.Equal(t, uint64(TaskTypeTCPPing), task.GetType())
+	require.Equal(t, service.Target, task.GetData())
+}

service/rpc/auth.go

@@ -222,6 +222,9 @@ func authorizeAgentForUUID(userId uint64, clientUUID string) (clientID uint64, h
 	if userId == 0 {
 		// The legacy global agent secret maps to user 0. It predates per-user
 		// agent secrets, so keep it compatible by allowing any existing UUID.
+		// Possession of this deployment-wide master credential is therefore not
+		// a tenant-scoped authorization claim. Removal must follow an inventory and
+		// credential-rotation migration or legacy Agents will be locked out.
 		return cid, true, nil
 	}
 	if server.GetUserID() == userId {

service/singleton/frontend-templates.yaml

@@ -16,6 +16,10 @@
   repository: "https://github.com/karllao/nezha-pixel"
   author: "karllao"
   version: "v1.6.0"
+# Third-party user themes consume the opaque Server.PublicNote field. Theme
+# maintainers must validate URL schemes (including after decoding) before using
+# values such as customData.orderLink in href or window.open; the Dashboard
+# backend and admin frontend do not execute those fields.
 - path: "nazhua-dist"
   name: "Nazhua"
   repository: "https://github.com/hi2shark/nazhua"

service/singleton/service_type_security_test.go

@@ -0,0 +1,37 @@
+package singleton
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/nezhahq/nezha/model"
+)
+
+func TestServiceSentinelUpdateRejectsNonProbeTaskTypes(t *testing.T) {
+	ss := &ServiceSentinel{}
+	require.Error(t, ss.Update(nil))
+	for _, taskType := range []uint8{0, model.TaskTypeCommand, model.TaskTypeApplyConfig, model.TaskTypeExec, 255} {
+		require.Error(t, ss.Update(&model.Service{Type: taskType}), "type %d must not be scheduled", taskType)
+	}
+}
+
+func TestServiceSentinelQuarantinesInvalidPersistedTypes(t *testing.T) {
+	ss := newServiceMonitorSecurityHarness(t)
+
+	insert := `INSERT INTO services
+		(id, user_id, name, type, target, duration, cover, skip_servers_raw, fail_trigger_tasks_raw, recover_trigger_tasks_raw)
+		VALUES (?, 100, ?, ?, 'example.invalid:443', 3600, ?, '{}', '[]', '[]')`
+	require.NoError(t, DB.Exec(insert, 91, "legacy-command", model.TaskTypeCommand, model.ServiceCoverIgnoreAll).Error)
+	require.NoError(t, DB.Exec(insert, 92, "legacy-apply-config", model.TaskTypeApplyConfig, model.ServiceCoverIgnoreAll).Error)
+	require.NoError(t, DB.Exec(insert, 93, "valid-probe", model.TaskTypeTCPPing, model.ServiceCoverIgnoreAll).Error)
+
+	require.NoError(t, ss.loadServiceHistory())
+	_, commandLoaded := ss.Get(91)
+	_, applyConfigLoaded := ss.Get(92)
+	valid, validLoaded := ss.Get(93)
+	require.False(t, commandLoaded)
+	require.False(t, applyConfigLoaded)
+	require.True(t, validLoaded)
+	require.Equal(t, uint8(model.TaskTypeTCPPing), valid.Type)
+}

service/singleton/servicesentinel.go

@@ -209,7 +209,15 @@ func (ss *ServiceSentinel) loadServiceHistory() error {
 		return err
 	}
 
+	validServices := services[:0]
 	for _, service := range services {
+		if err := model.ValidateServiceMonitorType(uint64(service.Type)); err != nil {
+			// Existing databases may contain values written before Service.Type was
+			// constrained. Quarantine them in the database for operator review, but
+			// never register a cron job that could dispatch a privileged Agent task.
+			log.Printf("NEZHA>> quarantining service %d: %v", service.ID, err)
+			continue
+		}
 		task := service
 		// 通过cron定时将服务监控任务传递给任务调度管道
 		service.CronJobID, err = CronShared.AddFunc(task.CronSpec(), func() {
@@ -222,7 +230,9 @@ func (ss *ServiceSentinel) loadServiceHistory() error {
 		ss.serviceCurrentStatusData[service.ID] = new(serviceTaskStatus)
 		ss.serviceCurrentStatusData[service.ID].result = make([]*pb.TaskResult, 0, _CurrentStatusSize)
 		ss.serviceStatusToday[service.ID] = &_TodayStatsOfService{}
+		validServices = append(validServices, service)
 	}
+	services = validServices
 	ss.serviceList = services
 	sortServices(ss.serviceList)
 
@@ -339,6 +349,13 @@ func (ss *ServiceSentinel) loadTodayStats(today time.Time) {
 }
 
 func (ss *ServiceSentinel) Update(m *model.Service) error {
+	if m == nil {
+		return fmt.Errorf("service is nil")
+	}
+	if err := model.ValidateServiceMonitorType(uint64(m.Type)); err != nil {
+		return err
+	}
+
 	ss.serviceResponseDataStoreLock.Lock()
 	defer ss.serviceResponseDataStoreLock.Unlock()
 	ss.monthlyStatusLock.Lock()

service/singleton/user.go

@@ -23,7 +23,10 @@ func initUser() {
 	var users []model.User
 	DB.Find(&users)
 
-	// for backward compatibility
+	// Backward compatibility for pre-user-scoped Agents. AgentSecretKey is a
+	// deployment-wide migration/master credential, so user 0 is intentionally
+	// not tenant-scoped. Do not remove this mapping until every legacy Agent has
+	// rotated to a per-user/per-Agent credential; doing so would disconnect them.
 	UserInfoMap[0] = model.UserInfo{
 		Role:        model.RoleAdmin,
 		AgentSecret: Conf.AgentSecretKey,