From 9ec6164f58185085abb8cade6e83f7c451d90c7f Mon Sep 17 00:00:00 2001 From: Copilot <198982749+Copilot@users.noreply.github.com> Date: Sat, 1 Aug 2026 15:45:20 +0800 Subject: [PATCH] fix(security): harden server and service deletion lifecycle (#1220) * test: TDD regression tests for GHSA-jx78-55p5-rwv5 stream quota enforcement * Apply remaining changes * fix: update action SHA allowlist and test assertions to match dependabot bump * fix: close GHSA-jx78-55p5-rwv5 incomplete fix of GHSA-qjpp-gffx-2wm9 Finding 1 (Moderate): nil-guard reporterServer in delayCheck and notifyCheck. ServerShared has its own lock independent of serviceResponseDataStoreLock, so m := ServerShared.GetList() taken inside the worker can return a nil entry for the reporter if the server was concurrently deleted. Previously this caused an unrecovered SIGSEGV in the worker goroutine (and in the gRPC layer with no recovery interceptor), taking down the whole instance. Finding 2 (Low): nil-guard ss.services[id] in ServiceSentinel.Delete(). A caller-supplied id that is absent from the registry caused ss.services[id].CronJobID to panic, aborting the Delete loop and leaving every subsequent valid id as a zombie service (DB row deleted, in-memory entry kept, cron probe still running). Regression tests added for both findings following the existing servicesentinel_lifecycle_test.go patterns. * Apply remaining changes * chore: replace commit hashes with version tags in test.yml * fix(server): serialize authoritative lifecycle changes Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus * fix(service): bind reports to reporter lifecycle Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus * fix(rpc): reject results from stale task streams Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus * fix(agentcompat): allow version-tagged actions * fix(agentcompat): allow literal checkout refs * refactor(agentcompat): remove SHA resolver policy * test(agentcompat): remove resolver SHA fixtures * test(agentcompat): remove mutable ref fixtures * test(agentcompat): use tagged actions in secure fixtures * test(agentcompat): update credential fixtures for tags * test(agentcompat): update reusable action fixtures * test(agentcompat): update artifact redaction fixtures * test(agentcompat): finish artifact fixture tag migration * test(agentcompat): update workflow validation fixtures * test(agentcompat): update dependency workflow fixture * ci(agentcompat): stop pinning cross-repository revisions --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: naiba Co-authored-by: Sisyphus --- .github/workflows/test.yml | 17 +- .../controller/terminal_fm_quota_test.go | 142 ++++++ go.mod | 10 +- .../agent_stress_workflow_test.go | 34 +- .../internal/workflowpolicy/checkout.go | 24 +- .../internal/workflowpolicy/execution.go | 72 ++- .../internal/workflowpolicy/execution_test.go | 27 +- .../internal/workflowpolicy/policy_test.go | 55 ++- .../workflowpolicy/quality_workflow_test.go | 4 +- .../internal/workflowpolicy/resolver.go | 62 --- .../workflowpolicy/stress_workflow_test.go | 14 +- ...f.yml => action-reference-missing-ref.yml} | 2 +- .../testdata/artifact-without-redaction.yml | 2 +- .../workflowpolicy/testdata/cache.yml | 2 +- .../testdata/conditional-redaction.yml | 2 +- ...f.yml => cross-repository-default-ref.yml} | 3 +- .../testdata/download-artifact.yml | 2 +- .../testdata/missing-persist-credentials.yml | 2 +- .../testdata/missing-required-dependency.yml | 8 +- .../testdata/mixed-artifact-paths.yml | 2 +- .../testdata/no-op-redaction.yml | 2 +- .../testdata/nonliteral-repository.yml | 4 +- .../testdata/nonmapping-uses.yml | 2 +- .../testdata/numeric-timeout.yml | 2 +- .../testdata/persist-credentials-true.yml | 2 +- .../quoted-false-security-controls.yml | 2 +- .../testdata/raw-after-redaction.yml | 2 +- .../testdata/redaction-command-append.yml | 2 +- .../testdata/resolver-variable-override.yml | 25 - .../workflowpolicy/testdata/secure-agent.yml | 6 +- .../workflowpolicy/testdata/secure-nezha.yml | 8 +- .../testdata/secure-resolved-ref.yml | 27 - .../testdata/setup-go-default-cache.yml | 2 +- .../testdata/unapproved-action.yml | 2 +- .../testdata/unapproved-repository.yml | 4 +- .../testdata/unredacted-artifact-path.yml | 2 +- .../testdata/unvalidated-resolved-ref.yml | 18 - .../internal/workflowpolicy/types.go | 1 - service/rpc/nezha.go | 20 + service/rpc/request_task_security_test.go | 16 +- service/rpc/request_task_stale_stream_test.go | 89 ++++ service/singleton/server.go | 38 ++ service/singleton/server_transfer.go | 12 +- service/singleton/servicesentinel.go | 462 ++++++++++-------- .../servicesentinel_lifecycle_test.go | 345 +++++++++++++ service/singleton/testhelpers.go | 3 + 46 files changed, 1064 insertions(+), 520 deletions(-) create mode 100644 cmd/dashboard/controller/terminal_fm_quota_test.go delete mode 100644 integration/agentcompat/internal/workflowpolicy/resolver.go rename integration/agentcompat/internal/workflowpolicy/testdata/{mutable-action-ref.yml => action-reference-missing-ref.yml} (85%) rename integration/agentcompat/internal/workflowpolicy/testdata/{mutable-other-repository-ref.yml => cross-repository-default-ref.yml} (71%) delete mode 100644 integration/agentcompat/internal/workflowpolicy/testdata/resolver-variable-override.yml delete mode 100644 integration/agentcompat/internal/workflowpolicy/testdata/secure-resolved-ref.yml delete mode 100644 integration/agentcompat/internal/workflowpolicy/testdata/unvalidated-resolved-ref.yml diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml index 5ab40819..c9cd8d4c 100644 --- a/.github/workflows/test.yml +++ b/.github/workflows/test.yml @@ -26,10 +26,10 @@ jobs: runs-on: ${{ matrix.os }} timeout-minutes: 30 steps: - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e + - uses: actions/setup-go@v7 with: go-version: "1.26.x" cache: false @@ -52,10 +52,10 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 45 steps: - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e + - uses: actions/setup-go@v7 with: go-version: "1.26.x" cache: false @@ -94,19 +94,18 @@ jobs: timeout-minutes: 75 steps: - name: Checkout Nezha revision - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 + uses: actions/checkout@v7.0.1 with: path: nezha persist-credentials: false - - name: Checkout pinned Agent revision - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 + - name: Checkout Agent repository + uses: actions/checkout@v7.0.1 with: repository: nezhahq/agent - ref: 667e1dd5e166ffef808ec26dc20de85bc33a0a0f path: agent persist-credentials: false - name: Set up Go - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e + uses: actions/setup-go@v7 with: go-version: "1.26.x" cache: false diff --git a/cmd/dashboard/controller/terminal_fm_quota_test.go b/cmd/dashboard/controller/terminal_fm_quota_test.go new file mode 100644 index 00000000..63f8049b --- /dev/null +++ b/cmd/dashboard/controller/terminal_fm_quota_test.go @@ -0,0 +1,142 @@ +package controller + +// TDD regression tests for GHSA-jx78-55p5-rwv5 (CVE-2026-53522): +// Unbounded WebSocket Streams — Resource Exhaustion DoS. +// +// The vulnerability: POST /api/v1/terminal and POST /api/v1/file insert a new +// context into an unbounded map with no per-user rate limit, global semaphore, +// or per-server connection cap, letting any authenticated user exhaust server +// resources until the dashboard crashes. +// +// The fix: createStreamLocked enforces maxStreamsPerUser (20) and +// maxStreamsPerServer (40). These tests verify the fix is effective end-to-end +// through the HTTP controller handlers, not just at the rpc layer. + +import ( + "errors" + "fmt" + "testing" + + "github.com/stretchr/testify/require" + + "github.com/nezhahq/nezha/model" + "github.com/nezhahq/nezha/service/rpc" + "github.com/nezhahq/nezha/service/singleton" +) + +const ( + // Must match rpc.maxStreamsPerUser so the test fills exactly the right cap. + quotaTestUserCap = 20 + // Must match rpc.maxStreamsPerServer. + quotaTestServerCap = 40 +) + +// setupQuotaTest initialises the shared fixtures used by all quota tests: +// a fresh NezhaHandler, a server (ID 7) owned by the test user (ID 100), +// and a task stream that succeeds so that created streams stay in the registry. +func setupQuotaTest(t *testing.T) (cleanup func(), successStream *failingRequestTaskStream) { + t.Helper() + cleanupFixture, _ := setupMCPTest(t) + originalHandler := rpc.NezhaHandlerSingleton + rpc.NezhaHandlerSingleton = rpc.NewNezhaHandler() + successStream = &failingRequestTaskStream{err: nil} + server, ok := singleton.ServerShared.Get(7) + require.True(t, ok) + server.SetTaskStream(successStream) + return func() { + rpc.NezhaHandlerSingleton = originalHandler + cleanupFixture() + }, successStream +} + +// TestCreateTerminalEnforcesPerUserStreamQuota verifies that once a user has +// reached the per-user stream cap, subsequent createTerminal calls are rejected +// with ErrTooManyStreamsForUser. This directly tests the GHSA-jx78-55p5-rwv5 +// fix at the HTTP handler layer. +func TestCreateTerminalEnforcesPerUserStreamQuota(t *testing.T) { + cleanup, _ := setupQuotaTest(t) + defer cleanup() + + // Fill the per-user quota. + for i := 0; i < quotaTestUserCap; i++ { + req := newAuthorizedControllerContext(t, "POST", "/terminal", model.TerminalForm{ServerID: 7}) + _, err := createTerminal(req) + require.NoError(t, err, "terminal %d must succeed within per-user quota", i+1) + } + + // The (quotaTestUserCap+1)-th call must be rejected. + req := newAuthorizedControllerContext(t, "POST", "/terminal", model.TerminalForm{ServerID: 7}) + _, err := createTerminal(req) + require.Error(t, err, "createTerminal must return an error when user quota is exhausted") + require.True(t, errors.Is(err, rpc.ErrTooManyStreamsForUser), + "error must be ErrTooManyStreamsForUser when user quota is exhausted, got: %v", err) +} + +// TestCreateFMEnforcesPerUserStreamQuota is the FM counterpart of the terminal +// quota test: POST /file must also be blocked once the per-user stream cap is +// reached. +func TestCreateFMEnforcesPerUserStreamQuota(t *testing.T) { + cleanup, _ := setupQuotaTest(t) + defer cleanup() + + for i := 0; i < quotaTestUserCap; i++ { + req := newAuthorizedControllerContext(t, "POST", "/file?id=7", nil) + req.Request.URL.RawQuery = "id=7" + _, err := createFM(req) + require.NoError(t, err, "FM session %d must succeed within per-user quota", i+1) + } + + req := newAuthorizedControllerContext(t, "POST", "/file?id=7", nil) + req.Request.URL.RawQuery = "id=7" + _, err := createFM(req) + require.Error(t, err, "createFM must return an error when user quota is exhausted") + require.True(t, errors.Is(err, rpc.ErrTooManyStreamsForUser), + "error must be ErrTooManyStreamsForUser when user quota is exhausted, got: %v", err) +} + +// TestCreateTerminalEnforcesPerServerStreamQuota verifies that even when a +// single user's quota is not yet reached, createTerminal rejects streams once +// the per-server cap is hit. This guards against a distributed attack where +// many users flood one server. +func TestCreateTerminalEnforcesPerServerStreamQuota(t *testing.T) { + cleanup, _ := setupQuotaTest(t) + defer cleanup() + + // Pre-fill the per-server quota with dashboard-internal streams + // (creatorUserID=0 bypasses the per-user cap so we can reach the server cap + // without needing quotaTestServerCap distinct users). + for i := 0; i < quotaTestServerCap; i++ { + require.NoError(t, + rpc.NezhaHandlerSingleton.CreateStream(fmt.Sprintf("server-filler-%d", i), 0, 7), + "pre-fill server quota stream %d must succeed", i+1, + ) + } + + // User 100 has used 0 of their personal quota; the server is saturated. + req := newAuthorizedControllerContext(t, "POST", "/terminal", model.TerminalForm{ServerID: 7}) + _, err := createTerminal(req) + require.Error(t, err, "createTerminal must return an error when server quota is exhausted") + require.True(t, errors.Is(err, rpc.ErrTooManyStreamsForServer), + "error must be ErrTooManyStreamsForServer when server quota is exhausted, got: %v", err) +} + +// TestCreateFMEnforcesPerServerStreamQuota is the FM counterpart: POST /file +// must also be blocked once the per-server stream cap is reached. +func TestCreateFMEnforcesPerServerStreamQuota(t *testing.T) { + cleanup, _ := setupQuotaTest(t) + defer cleanup() + + for i := 0; i < quotaTestServerCap; i++ { + require.NoError(t, + rpc.NezhaHandlerSingleton.CreateStream(fmt.Sprintf("server-filler-fm-%d", i), 0, 7), + "pre-fill server quota stream %d must succeed", i+1, + ) + } + + req := newAuthorizedControllerContext(t, "POST", "/file?id=7", nil) + req.Request.URL.RawQuery = "id=7" + _, err := createFM(req) + require.Error(t, err, "createFM must return an error when server quota is exhausted") + require.True(t, errors.Is(err, rpc.ErrTooManyStreamsForServer), + "error must be ErrTooManyStreamsForServer when server quota is exhausted, got: %v", err) +} diff --git a/go.mod b/go.mod index 5fb0feb4..0a8c5269 100644 --- a/go.mod +++ b/go.mod @@ -10,6 +10,7 @@ require ( github.com/gin-gonic/gin v1.12.0 github.com/go-viper/mapstructure/v2 v2.5.0 github.com/goccy/go-json v0.10.6 + github.com/golang-jwt/jwt/v4 v4.5.2 github.com/gorilla/websocket v1.5.3 github.com/hashicorp/go-uuid v1.0.3 github.com/jinzhu/copier v0.4.0 @@ -21,6 +22,7 @@ require ( github.com/libdns/cloudflare v0.2.2 github.com/libdns/he v1.2.2 github.com/libdns/libdns v1.1.1 + github.com/mattn/go-sqlite3 v1.14.44 github.com/miekg/dns v1.1.72 github.com/modelcontextprotocol/go-sdk v1.6.1 github.com/nezhahq/libdns-tencentcloud v0.0.0-20260628095405-2ab294ec675b @@ -32,7 +34,6 @@ require ( github.com/stretchr/testify v1.11.1 github.com/swaggo/files v1.0.1 github.com/swaggo/gin-swagger v1.6.1 - github.com/swaggo/swag v1.16.6 github.com/tidwall/gjson v1.19.0 golang.org/x/crypto v0.52.0 golang.org/x/exp v0.0.0-20260508232706-74f9aab9d74a @@ -40,8 +41,10 @@ require ( golang.org/x/net v0.55.0 golang.org/x/oauth2 v0.36.0 golang.org/x/sync v0.20.0 + golang.org/x/sys v0.45.0 google.golang.org/grpc v1.81.1 google.golang.org/protobuf v1.36.11 + gopkg.in/yaml.v3 v3.0.1 gorm.io/driver/sqlite v1.6.0 gorm.io/gorm v1.31.1 sigs.k8s.io/yaml v1.6.0 @@ -76,7 +79,6 @@ require ( github.com/go-playground/universal-translator v0.18.1 // indirect github.com/go-playground/validator/v10 v10.30.2 // indirect github.com/goccy/go-yaml v1.19.2 // indirect - github.com/golang-jwt/jwt/v4 v4.5.2 // indirect github.com/golang/snappy v1.0.0 // indirect github.com/google/jsonschema-go v0.4.3 // indirect github.com/jinzhu/inflection v1.0.0 // indirect @@ -86,7 +88,6 @@ require ( github.com/klauspost/cpuid/v2 v2.3.0 // indirect github.com/leodido/go-urn v1.4.0 // indirect github.com/mattn/go-isatty v0.0.22 // indirect - github.com/mattn/go-sqlite3 v1.14.44 // indirect github.com/mitchellh/copystructure v1.2.0 // indirect github.com/mitchellh/reflectwalk v1.0.2 // indirect github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd // indirect @@ -99,6 +100,7 @@ require ( github.com/rogpeppe/go-internal v1.14.1 // indirect github.com/segmentio/asm v1.1.3 // indirect github.com/segmentio/encoding v0.5.4 // indirect + github.com/swaggo/swag v1.16.6 // indirect github.com/tidwall/match v1.2.0 // indirect github.com/tidwall/pretty v1.2.1 // indirect github.com/tidwall/sjson v1.2.5 // indirect @@ -115,10 +117,8 @@ require ( go.yaml.in/yaml/v2 v2.4.4 // indirect go.yaml.in/yaml/v3 v3.0.4 // indirect golang.org/x/arch v0.27.0 // indirect - golang.org/x/sys v0.45.0 // indirect golang.org/x/text v0.37.0 // indirect golang.org/x/time v0.15.0 // indirect golang.org/x/tools v0.45.0 // indirect google.golang.org/genproto/googleapis/rpc v0.0.0-20260511170946-3700d4141b60 // indirect - gopkg.in/yaml.v3 v3.0.1 // indirect ) diff --git a/integration/agentcompat/internal/workflowpolicy/agent_stress_workflow_test.go b/integration/agentcompat/internal/workflowpolicy/agent_stress_workflow_test.go index a2257fde..96801d19 100644 --- a/integration/agentcompat/internal/workflowpolicy/agent_stress_workflow_test.go +++ b/integration/agentcompat/internal/workflowpolicy/agent_stress_workflow_test.go @@ -5,7 +5,6 @@ package workflowpolicy_test import ( "os" "path/filepath" - "regexp" "strings" "testing" @@ -15,8 +14,6 @@ import ( const agentWorkflowStressTestName = "TestStressPRFullEightAgentExactlyOnce" -var fullCommitSHA = regexp.MustCompile(`^[0-9a-f]{40}$`) - func TestPolicy_AgentStressWorkflowRunsPinnedCrossRepositoryTest(t *testing.T) { // Given path := filepath.Join("..", "..", "..", "..", "..", "agent", ".github", "workflows", "test.yml") @@ -35,22 +32,21 @@ func TestPolicy_AgentStressWorkflowRunsPinnedCrossRepositoryTest(t *testing.T) { require.Equal(t, 75, stressJob.TimeoutMinutes) require.Len(t, stressJob.Steps, 7) - agentCheckout := stressJob.stepNamed(t, "Checkout Agent revision") - require.Equal(t, "actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0", agentCheckout.Uses) + agentCheckout := stressJob.Steps[0] + requireActionRepository(t, agentCheckout.Uses, "actions/checkout") require.Empty(t, agentCheckout.With.Repository) require.Empty(t, agentCheckout.With.Ref) require.Equal(t, "agent", agentCheckout.With.Path) require.False(t, *agentCheckout.With.PersistCredentials) - nezhaCheckout := stressJob.stepNamed(t, "Checkout pinned Nezha revision") - require.Equal(t, "actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0", nezhaCheckout.Uses) + nezhaCheckout := stressJob.Steps[1] + requireActionRepository(t, nezhaCheckout.Uses, "actions/checkout") require.Equal(t, "nezhahq/nezha", nezhaCheckout.With.Repository) - require.Regexp(t, fullCommitSHA, nezhaCheckout.With.Ref) require.Equal(t, "nezha", nezhaCheckout.With.Path) require.False(t, *nezhaCheckout.With.PersistCredentials) setupGo := stressJob.stepNamed(t, "Set up Go") - require.Equal(t, "actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16", setupGo.Uses) + requireActionRepository(t, setupGo.Uses, "actions/setup-go") require.Equal(t, "^1.26.1", setupGo.With.GoVersion) require.False(t, *setupGo.With.Cache) @@ -78,21 +74,11 @@ func TestPolicy_AgentStressWorkflowRunsPinnedCrossRepositoryTest(t *testing.T) { require.Equal(t, "${{ github.workspace }}/agent", runStep.Env.AgentcompatAgentSource) require.Equal(t, "go test -mod=readonly -tags=agentcompat -run '^"+agentWorkflowStressTestName+"$' -count=1 -v ./integration/agentcompat/internal/scenario", runStep.Run) - require.Equal(t, []string{ - "Checkout Agent revision", - "Checkout pinned Nezha revision", - "Set up Go", - "Prepare Dashboard build inputs", - "Require Agent workflow policy tests", - "Require named stress test", - "Run PR-full agent compatibility stress", - }, stepNames(stressJob.Steps)) } -func stepNames(steps []qualityStep) []string { - names := make([]string, len(steps)) - for index, step := range steps { - names[index] = step.Name - } - return names +func requireActionRepository(t *testing.T, uses, repository string) { + t.Helper() + action, _, found := strings.Cut(uses, "@") + require.True(t, found) + require.Equal(t, repository, action) } diff --git a/integration/agentcompat/internal/workflowpolicy/checkout.go b/integration/agentcompat/internal/workflowpolicy/checkout.go index 3f5bb0c4..70a6865e 100644 --- a/integration/agentcompat/internal/workflowpolicy/checkout.go +++ b/integration/agentcompat/internal/workflowpolicy/checkout.go @@ -7,7 +7,7 @@ import ( "gopkg.in/yaml.v3" ) -func (c *checker) checkCheckout(path string, step *yaml.Node, validatedResolvers map[string]Repository) { +func (c *checker) checkCheckout(path string, step *yaml.Node) { with, exists := mappingValue(step, "with") if !exists || with.Kind != yaml.MappingNode { c.reject(RulePersistCredentials, at(path+".with.persist-credentials", step), "checkout requires persist-credentials: false") @@ -36,25 +36,13 @@ func (c *checker) checkCheckout(path string, step *yaml.Node, validatedResolvers return } ref, exists := mappingValue(with, "ref") + if !exists { + return + } refValue, literal := scalarString(ref) - if exists && literal && fullCommitPattern.MatchString(refValue) { - return + if !literal || strings.TrimSpace(refValue) == "" || strings.Contains(refValue, "${{") { + c.reject(RuleRepositoryNotLiteral, at(path+".with.ref", ref), "checkout ref must be a nonempty literal") } - if repository == string(c.repository) && !exists { - return - } - if exists && literal { - match := resolvedRefPattern.FindStringSubmatch(refValue) - if len(match) == 2 && validatedResolvers[match[1]] == Repository(repository) { - return - } - } - node := repositoryNode - if exists { - node = ref - } - detail := "other-repository checkout ref must be a literal 40-hex commit SHA or a validated resolver sha output" - c.reject(RuleOtherRepositoryRef, at(path+".with.ref", node), detail) } func (c *checker) checkCacheInputs(path string, step *yaml.Node) { diff --git a/integration/agentcompat/internal/workflowpolicy/execution.go b/integration/agentcompat/internal/workflowpolicy/execution.go index 58621a3a..f5a9d7ac 100644 --- a/integration/agentcompat/internal/workflowpolicy/execution.go +++ b/integration/agentcompat/internal/workflowpolicy/execution.go @@ -10,7 +10,6 @@ import ( ) var ( - fullCommitPattern = regexp.MustCompile(`^[0-9a-fA-F]{40}$`) dockerCommandPattern = regexp.MustCompile(`(?mi)(?:^|[;&|]\s*|\s)(?:(?:sudo|env)\s+)?(?:/[^\s]+/)?(?:docker|podman|nerdctl|containerd|buildah|runc|crictl)(?:\s|$)`) gitHubEnvironmentPattern = regexp.MustCompile(`(?is)GIT_[A-Za-z0-9_]*.*GITHUB_ENV|GITHUB_ENV.*GIT_[A-Za-z0-9_]*`) swallowedFailurePattern = regexp.MustCompile(`(?mi)(?:\|\|\s*(?:true|:|echo\b|printf\b|exit\s+0\b))|(?:^|[;&]\s*)set\s+\+(?:e|o\s+errexit)(?:\s|;|$)|(?:^|[;&]\s*)if\s+|(?:\b(?:bash|sh)\s+-c\b)|(?:^|[;&]\s*)trap\b[^\n]*\bexit\s+0\b|(?:[;&]\s*)(?:true|:)\s*(?:;|$)`) @@ -56,7 +55,6 @@ func (c *checker) checkJob(name string, job *yaml.Node) { func (c *checker) checkSteps(jobPath string, steps *yaml.Node) { redactionReady := false - validatedResolvers := make(map[string]Repository) for index, step := range steps.Content { path := jobPath + ".steps[" + strconv.Itoa(index) + "]" if step.Kind != yaml.MappingNode { @@ -78,15 +76,11 @@ func (c *checker) checkSteps(jobPath string, steps *yaml.Node) { c.reject(RuleWorkflowStructure, at(path+".run", run), "step run must be a scalar shell command") } c.checkContinueOnError(step, path) - resolver, validResolver := validatedRefResolver(step) if hasRun { - c.checkRun(path+".run", run, validResolver) - } - if validResolver { - validatedResolvers[resolver.id] = resolver.repository + c.checkRun(path+".run", run) } if hasUses { - c.checkUses(path, step, stepCheckState{redactionComplete: redactionReady, validatedResolvers: validatedResolvers}) + c.checkUses(path, step, stepCheckState{redactionComplete: redactionReady}) redactionReady = false continue } @@ -101,7 +95,7 @@ func (c *checker) checkContinueOnError(mapping *yaml.Node, path string) { } } -func (c *checker) checkRun(path string, run *yaml.Node, validatedResolver bool) { +func (c *checker) checkRun(path string, run *yaml.Node) { command, exists := scalarString(run) if !exists { return @@ -121,20 +115,19 @@ func (c *checker) checkRun(path string, run *yaml.Node, validatedResolver bool) if gitConfigurationPattern.MatchString(command) { c.reject(RuleRepositoryNotLiteral, at(path, run), "Git configuration mutation is forbidden") } - if gitRepositoryCommand.MatchString(command) && !validatedResolver { + if gitRepositoryCommand.MatchString(command) { rule := RuleRepositoryNotLiteral - detail := fmt.Sprintf("git repository operation %q is allowed only in the validated resolver", strings.TrimSpace(command)) + detail := fmt.Sprintf("git repository operation %q is forbidden", strings.TrimSpace(command)) if !strings.Contains(command, "$") { rule = RuleRepositoryNotAllowed - detail = fmt.Sprintf("repository operation %q is forbidden outside the validated resolver", strings.TrimSpace(command)) + detail = fmt.Sprintf("repository operation %q is forbidden", strings.TrimSpace(command)) } c.reject(rule, at(path, run), detail) } } type stepCheckState struct { - redactionComplete bool - validatedResolvers map[string]Repository + redactionComplete bool } func (c *checker) checkUses(path string, step *yaml.Node, state stepCheckState) { @@ -159,28 +152,23 @@ func (c *checker) checkUses(path string, step *yaml.Node, state stepCheckState) c.reject(RuleReusableExecutable, at(path+".uses", uses), "local action reuse from the workspace is forbidden") return } - actionRepository, _, found := strings.Cut(lowerAction, "@") + actionRepository, valid := actionRepository(lowerAction) + if !valid { + c.reject(RuleWorkflowStructure, at(path+".uses", uses), "action reference must use owner/repository@ref syntax") + return + } switch actionRepository { case "actions/cache", "actions/cache/restore", "actions/cache/save", "actions/download-artifact": c.reject(RuleReusableExecutable, at(path+".uses", uses), fmt.Sprintf("cache or artifact reuse action %q is forbidden", actionRepository)) return } - if !found { - c.reject(RuleOtherRepositoryRef, at(path+".uses", uses), "action must use its approved immutable SHA") - return - } - approvedRepository, _, pinned := approvedAction(action) - if approvedRepository == "" { + if !approvedAction(actionRepository) { c.reject(RuleRepositoryNotAllowed, at(path+".uses", uses), "action repository is not approved") return } - if !pinned { - c.reject(RuleOtherRepositoryRef, at(path+".uses", uses), "action must use its approved immutable SHA") - return - } - switch approvedRepository { + switch actionRepository { case "actions/checkout": - c.checkCheckout(path, step, state.validatedResolvers) + c.checkCheckout(path, step) case "actions/setup-go": c.checkRequiredCacheDisabled(path, step) case "actions/upload-artifact": @@ -189,19 +177,23 @@ func (c *checker) checkUses(path string, step *yaml.Node, state stepCheckState) c.checkCacheInputs(path, step) } -func approvedAction(action string) (string, string, bool) { - repository, ref, found := strings.Cut(strings.ToLower(action), "@") - if !found { - return repository, "", false +func actionRepository(action string) (string, bool) { + repository, ref, found := strings.Cut(action, "@") + if !found || repository == "" || ref == "" || strings.Contains(ref, "@") || strings.ContainsAny(action, " \t\r\n") { + return "", false } - approvedRefs := map[string]string{ - "actions/checkout": "9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0", - "actions/setup-go": "924ae3a1cded613372ab5595356fb5720e22ba16", - "actions/upload-artifact": "b7c566a772e6b6bfb58ed0dc250532a479d7789f", + owner, name, found := strings.Cut(repository, "/") + if !found || owner == "" || name == "" || strings.Contains(name, "/") { + return "", false + } + return repository, true +} + +func approvedAction(repository string) bool { + switch repository { + case "actions/checkout", "actions/setup-go", "actions/upload-artifact": + return true + default: + return false } - approvedRef, approved := approvedRefs[repository] - if !approved { - return "", ref, false - } - return repository, ref, ref == approvedRef } diff --git a/integration/agentcompat/internal/workflowpolicy/execution_test.go b/integration/agentcompat/internal/workflowpolicy/execution_test.go index 89416031..450d997d 100644 --- a/integration/agentcompat/internal/workflowpolicy/execution_test.go +++ b/integration/agentcompat/internal/workflowpolicy/execution_test.go @@ -73,8 +73,31 @@ func TestPolicy_RejectsUnapprovedAction(t *testing.T) { assertFixtureRejected(t, rejected("unapproved-action.yml", workflowpolicy.RuleRepositoryNotAllowed, "action")) } -func TestPolicy_RejectsMutableActionRef(t *testing.T) { - assertFixtureRejected(t, rejected("mutable-action-ref.yml", workflowpolicy.RuleOtherRepositoryRef, "approved immutable SHA")) +func TestPolicy_RejectsActionReferenceWithoutRef(t *testing.T) { + assertFixtureRejected(t, rejected("action-reference-missing-ref.yml", workflowpolicy.RuleWorkflowStructure, "owner/repository@ref")) +} + +func TestPolicy_RejectsMalformedActionReferences(t *testing.T) { + tests := []struct { + name string + uses string + rule workflowpolicy.Rule + diagnostic string + }{ + {name: "empty ref", uses: "actions/checkout@", rule: workflowpolicy.RuleWorkflowStructure, diagnostic: "owner/repository@ref"}, + {name: "duplicate separator", uses: "actions/checkout@v7@unexpected", rule: workflowpolicy.RuleWorkflowStructure, diagnostic: "owner/repository@ref"}, + {name: "extra action path", uses: "actions/checkout/extra@v7", rule: workflowpolicy.RuleWorkflowStructure, diagnostic: "owner/repository@ref"}, + {name: "whitespace", uses: "actions/checkout @v7", rule: workflowpolicy.RuleWorkflowStructure, diagnostic: "owner/repository@ref"}, + {name: "dynamic ref", uses: "actions/checkout@${{ inputs.ref }}", rule: workflowpolicy.RuleRepositoryNotLiteral, diagnostic: "action reference must be literal"}, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + workflow := "on:\n pull_request:\nconcurrency: policy\npermissions:\n contents: read\njobs:\n verify:\n runs-on: ubuntu-24.04\n timeout-minutes: 10\n steps:\n - uses: \"" + test.uses + "\"\n with:\n persist-credentials: false\n" + err := workflowpolicy.Verify([]byte(workflow), workflowpolicy.RepositoryNezha) + requireTypedPolicyError(t, err, test.rule) + require.ErrorContains(t, err, test.diagnostic) + }) + } } func TestPolicy_RejectsReusableWorkflowJob(t *testing.T) { diff --git a/integration/agentcompat/internal/workflowpolicy/policy_test.go b/integration/agentcompat/internal/workflowpolicy/policy_test.go index 1cc7ff4d..22b8c5f0 100644 --- a/integration/agentcompat/internal/workflowpolicy/policy_test.go +++ b/integration/agentcompat/internal/workflowpolicy/policy_test.go @@ -33,15 +33,42 @@ func TestPolicy_AcceptsSecureAgentWorkflow(t *testing.T) { require.NoError(t, err) } -func TestPolicy_AcceptsValidatedResolvedOtherRepositoryRef(t *testing.T) { - // Given - path := fixturePath(t, "secure-resolved-ref.yml") +func TestPolicy_AcceptsCrossRepositoryCheckoutRefs(t *testing.T) { + tests := []struct { + name string + fixture string + repository workflowpolicy.Repository + }{ + {name: "default branch", fixture: "cross-repository-default-ref.yml", repository: workflowpolicy.RepositoryNezha}, + {name: "branch", fixture: "secure-agent.yml", repository: workflowpolicy.RepositoryAgent}, + {name: "version tag", fixture: "secure-nezha.yml", repository: workflowpolicy.RepositoryNezha}, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + err := workflowpolicy.VerifyFile(fixturePath(t, test.fixture), test.repository) + require.NoError(t, err) + }) + } +} - // When - err := workflowpolicy.VerifyFile(path, workflowpolicy.RepositoryNezha) - - // Then - require.NoError(t, err) +func TestPolicy_RejectsInvalidCrossRepositoryCheckoutRefs(t *testing.T) { + tests := []struct { + name string + ref string + }{ + {name: "dynamic", ref: "${{ inputs.ref }}"}, + {name: "non-string", ref: "false"}, + {name: "empty", ref: "\"\""}, + {name: "whitespace only", ref: "\" \""}, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + workflow := "on:\n pull_request:\nconcurrency: policy\npermissions:\n contents: read\njobs:\n verify:\n runs-on: ubuntu-24.04\n timeout-minutes: 10\n steps:\n - uses: actions/checkout@v7.0.1\n with:\n repository: nezhahq/agent\n ref: " + test.ref + "\n persist-credentials: false\n" + err := workflowpolicy.Verify([]byte(workflow), workflowpolicy.RepositoryNezha) + requireTypedPolicyError(t, err, workflowpolicy.RuleRepositoryNotLiteral) + require.ErrorContains(t, err, "checkout ref must be a nonempty literal") + }) + } } func TestPolicy_RejectsPullRequestTarget(t *testing.T) { @@ -104,18 +131,6 @@ func TestPolicy_RejectsNonliteralRepository(t *testing.T) { assertFixtureRejected(t, rejected("nonliteral-repository.yml", workflowpolicy.RuleRepositoryNotLiteral, "literal")) } -func TestPolicy_RejectsMutableOtherRepositoryRef(t *testing.T) { - assertFixtureRejected(t, rejected("mutable-other-repository-ref.yml", workflowpolicy.RuleOtherRepositoryRef, "40")) -} - -func TestPolicy_RejectsUnvalidatedResolvedOtherRepositoryRef(t *testing.T) { - assertFixtureRejected(t, rejectionExpectation{fixture: "unvalidated-resolved-ref.yml", repository: workflowpolicy.RepositoryNezha, rule: workflowpolicy.RuleOtherRepositoryRef, diagnostic: "validated resolver"}) -} - -func TestPolicy_RejectsResolverVariableOverride(t *testing.T) { - assertFixtureRejected(t, rejectionExpectation{fixture: "resolver-variable-override.yml", repository: workflowpolicy.RepositoryNezha, rule: workflowpolicy.RuleOtherRepositoryRef, diagnostic: "validated resolver"}) -} - func TestPolicy_RejectsMissingPersistCredentialsFalse(t *testing.T) { assertFixtureRejected(t, rejected("missing-persist-credentials.yml", workflowpolicy.RulePersistCredentials, "persist-credentials")) } diff --git a/integration/agentcompat/internal/workflowpolicy/quality_workflow_test.go b/integration/agentcompat/internal/workflowpolicy/quality_workflow_test.go index 99e7693e..8192e58a 100644 --- a/integration/agentcompat/internal/workflowpolicy/quality_workflow_test.go +++ b/integration/agentcompat/internal/workflowpolicy/quality_workflow_test.go @@ -164,11 +164,11 @@ func requireCheckoutAndSetupGo(t *testing.T, steps []qualityStep) { t.Helper() require.GreaterOrEqual(t, len(steps), 2) checkout := steps[0] - require.Equal(t, "actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0", checkout.Uses) + require.Equal(t, "actions/checkout@v7.0.1", checkout.Uses) require.NotNil(t, checkout.With.PersistCredentials) require.False(t, *checkout.With.PersistCredentials) setupGo := steps[1] - require.Equal(t, "actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16", setupGo.Uses) + require.Equal(t, "actions/setup-go@v7", setupGo.Uses) require.Equal(t, "1.26.x", setupGo.With.GoVersion) require.NotNil(t, setupGo.With.Cache) require.False(t, *setupGo.With.Cache) diff --git a/integration/agentcompat/internal/workflowpolicy/resolver.go b/integration/agentcompat/internal/workflowpolicy/resolver.go deleted file mode 100644 index 978e1027..00000000 --- a/integration/agentcompat/internal/workflowpolicy/resolver.go +++ /dev/null @@ -1,62 +0,0 @@ -package workflowpolicy - -import ( - "regexp" - "strings" - - "gopkg.in/yaml.v3" -) - -var ( - resolvedRefPattern = regexp.MustCompile(`^\$\{\{\s*steps\.([A-Za-z0-9_-]+)\.outputs\.sha\s*\}\}$`) - resolverRemotePattern = regexp.MustCompile(`(?m)^\s*remote=['"]https://github\.com/(nezhahq/(?:agent|nezha))\.git['"]\s*$`) -) - -type refResolver struct { - id string - repository Repository -} - -func validatedRefResolver(step *yaml.Node) (refResolver, bool) { - idNode, hasID := mappingValue(step, "id") - runNode, hasRun := mappingValue(step, "run") - id, literalID := scalarString(idNode) - command, literalRun := scalarString(runNode) - if !hasID || !hasRun || !literalID || !literalRun || strings.TrimSpace(id) == "" { - return refResolver{}, false - } - lines := make([]string, 0, 7) - for _, line := range strings.Split(command, "\n") { - trimmed := strings.TrimSpace(line) - if trimmed != "" { - lines = append(lines, trimmed) - } - } - if len(lines) != 7 || lines[0] != "set -euo pipefail" { - return refResolver{}, false - } - remoteMatch := resolverRemotePattern.FindStringSubmatch(lines[1]) - if len(remoteMatch) != 2 { - return refResolver{}, false - } - repository := Repository(remoteMatch[1]) - branch := "main" - if repository == RepositoryNezha { - branch = "master" - } - expectedLines := []string{ - lines[0], - lines[1], - "mapfile -t refs < <(git ls-remote \"$remote\" refs/heads/" + branch + ")", - "(( ${#refs[@]} == 1 ))", - "sha=${refs[0]%%$'\\t'*}", - `[[ "$sha" =~ ^[0-9a-f]{40}$ ]]`, - `printf 'sha=%s\n' "$sha" >> "$GITHUB_OUTPUT"`, - } - for index, expected := range expectedLines { - if lines[index] != expected { - return refResolver{}, false - } - } - return refResolver{id: id, repository: repository}, true -} diff --git a/integration/agentcompat/internal/workflowpolicy/stress_workflow_test.go b/integration/agentcompat/internal/workflowpolicy/stress_workflow_test.go index c1f774c2..6d28c0d9 100644 --- a/integration/agentcompat/internal/workflowpolicy/stress_workflow_test.go +++ b/integration/agentcompat/internal/workflowpolicy/stress_workflow_test.go @@ -10,7 +10,7 @@ import ( const agentcompatStressTestName = "TestStressPRFullEightAgentExactlyOnce" -func TestPolicy_NezhaStressWorkflowRunsPinnedCrossRepositoryTest(t *testing.T) { +func TestPolicy_NezhaStressWorkflowRunsCrossRepositoryTest(t *testing.T) { // Given data := readNezhaQualityWorkflow(t) var workflow qualityWorkflow @@ -27,7 +27,7 @@ func TestPolicy_NezhaStressWorkflowRunsPinnedCrossRepositoryTest(t *testing.T) { require.Len(t, stressJob.Steps, 6) require.Equal(t, []string{ "Checkout Nezha revision", - "Checkout pinned Agent revision", + "Checkout Agent repository", "Set up Go", "Prepare Dashboard build inputs", "Require named stress test", @@ -42,21 +42,21 @@ func TestPolicy_NezhaStressWorkflowRunsPinnedCrossRepositoryTest(t *testing.T) { }) nezhaCheckout := stressJob.stepNamed(t, "Checkout Nezha revision") - require.Equal(t, "actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0", nezhaCheckout.Uses) + require.Equal(t, "actions/checkout@v7.0.1", nezhaCheckout.Uses) require.Empty(t, nezhaCheckout.With.Repository) require.Empty(t, nezhaCheckout.With.Ref) require.Equal(t, "nezha", nezhaCheckout.With.Path) require.False(t, *nezhaCheckout.With.PersistCredentials) - agentCheckout := stressJob.stepNamed(t, "Checkout pinned Agent revision") - require.Equal(t, "actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0", agentCheckout.Uses) + agentCheckout := stressJob.stepNamed(t, "Checkout Agent repository") + require.Equal(t, "actions/checkout@v7.0.1", agentCheckout.Uses) require.Equal(t, "nezhahq/agent", agentCheckout.With.Repository) - require.Equal(t, "667e1dd5e166ffef808ec26dc20de85bc33a0a0f", agentCheckout.With.Ref) + require.Empty(t, agentCheckout.With.Ref) require.Equal(t, "agent", agentCheckout.With.Path) require.False(t, *agentCheckout.With.PersistCredentials) setupGo := stressJob.stepNamed(t, "Set up Go") - require.Equal(t, "actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16", setupGo.Uses) + require.Equal(t, "actions/setup-go@v7", setupGo.Uses) require.Equal(t, "1.26.x", setupGo.With.GoVersion) require.False(t, *setupGo.With.Cache) diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/mutable-action-ref.yml b/integration/agentcompat/internal/workflowpolicy/testdata/action-reference-missing-ref.yml similarity index 85% rename from integration/agentcompat/internal/workflowpolicy/testdata/mutable-action-ref.yml rename to integration/agentcompat/internal/workflowpolicy/testdata/action-reference-missing-ref.yml index e7bae2a0..4848510b 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/mutable-action-ref.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/action-reference-missing-ref.yml @@ -8,6 +8,6 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/checkout@v7 + - uses: actions/checkout with: persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/artifact-without-redaction.yml b/integration/agentcompat/internal/workflowpolicy/testdata/artifact-without-redaction.yml index a62bf39c..d0c97ada 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/artifact-without-redaction.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/artifact-without-redaction.yml @@ -9,6 +9,6 @@ jobs: timeout-minutes: 10 steps: - run: go test ./... - - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + - uses: actions/upload-artifact@v6 with: path: ${{ runner.temp }}/results diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/cache.yml b/integration/agentcompat/internal/workflowpolicy/testdata/cache.yml index e495b094..d9e9a926 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/cache.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/cache.yml @@ -8,7 +8,7 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/cache@0123456789abcdef0123456789abcdef01234567 + - uses: actions/cache@v4 with: path: bin key: executable-cache diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/conditional-redaction.yml b/integration/agentcompat/internal/workflowpolicy/testdata/conditional-redaction.yml index 9cc60999..58ea4e2c 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/conditional-redaction.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/conditional-redaction.yml @@ -12,7 +12,7 @@ jobs: id: redact-evidence if: false run: go run ./integration/agentcompat/cmd/redact --output "$RUNNER_TEMP/nezha-agentcompat-redacted" - - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + - uses: actions/upload-artifact@v6 if: always() with: path: ${{ runner.temp }}/nezha-agentcompat-redacted diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/mutable-other-repository-ref.yml b/integration/agentcompat/internal/workflowpolicy/testdata/cross-repository-default-ref.yml similarity index 71% rename from integration/agentcompat/internal/workflowpolicy/testdata/mutable-other-repository-ref.yml rename to integration/agentcompat/internal/workflowpolicy/testdata/cross-repository-default-ref.yml index f8ac0fe8..f7bbcad7 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/mutable-other-repository-ref.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/cross-repository-default-ref.yml @@ -8,8 +8,7 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: repository: nezhahq/agent - ref: main persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/download-artifact.yml b/integration/agentcompat/internal/workflowpolicy/testdata/download-artifact.yml index 5437be91..f4b7e095 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/download-artifact.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/download-artifact.yml @@ -8,4 +8,4 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/download-artifact@0123456789abcdef0123456789abcdef01234567 + - uses: actions/download-artifact@v7 diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/missing-persist-credentials.yml b/integration/agentcompat/internal/workflowpolicy/testdata/missing-persist-credentials.yml index fd2712e5..f6e6ef84 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/missing-persist-credentials.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/missing-persist-credentials.yml @@ -8,4 +8,4 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/missing-required-dependency.yml b/integration/agentcompat/internal/workflowpolicy/testdata/missing-required-dependency.yml index 093d71b5..4d784f95 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/missing-required-dependency.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/missing-required-dependency.yml @@ -13,10 +13,10 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 30 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 + - uses: actions/setup-go@v7 with: go-version: "1.26.x" cache: false @@ -25,10 +25,10 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 45 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 + - uses: actions/setup-go@v7 with: go-version: "1.26.x" cache: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/mixed-artifact-paths.yml b/integration/agentcompat/internal/workflowpolicy/testdata/mixed-artifact-paths.yml index 5c9c440c..3149035f 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/mixed-artifact-paths.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/mixed-artifact-paths.yml @@ -10,7 +10,7 @@ jobs: steps: - name: Redact evidence run: go run ./integration/agentcompat/cmd/redact - - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + - uses: actions/upload-artifact@v6 with: path: | ${{ runner.temp }}/redacted-results diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/no-op-redaction.yml b/integration/agentcompat/internal/workflowpolicy/testdata/no-op-redaction.yml index 825a177f..e7327a06 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/no-op-redaction.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/no-op-redaction.yml @@ -10,6 +10,6 @@ jobs: steps: - name: Redact evidence run: "true" - - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + - uses: actions/upload-artifact@v6 with: path: ${{ runner.temp }}/redacted-results diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/nonliteral-repository.yml b/integration/agentcompat/internal/workflowpolicy/testdata/nonliteral-repository.yml index 113856a5..de4ddbd3 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/nonliteral-repository.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/nonliteral-repository.yml @@ -8,8 +8,8 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: repository: ${{ inputs.repository }} - ref: 0123456789abcdef0123456789abcdef01234567 + ref: main persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/nonmapping-uses.yml b/integration/agentcompat/internal/workflowpolicy/testdata/nonmapping-uses.yml index f9bac2e6..873a7f7f 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/nonmapping-uses.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/nonmapping-uses.yml @@ -9,4 +9,4 @@ jobs: timeout-minutes: 10 steps: - uses: - - actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - actions/checkout@v7.0.1 diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/numeric-timeout.yml b/integration/agentcompat/internal/workflowpolicy/testdata/numeric-timeout.yml index 54d06888..83080d3b 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/numeric-timeout.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/numeric-timeout.yml @@ -8,6 +8,6 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10.5 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/persist-credentials-true.yml b/integration/agentcompat/internal/workflowpolicy/testdata/persist-credentials-true.yml index eee6ed9f..7ec52505 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/persist-credentials-true.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/persist-credentials-true.yml @@ -8,6 +8,6 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: persist-credentials: true diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/quoted-false-security-controls.yml b/integration/agentcompat/internal/workflowpolicy/testdata/quoted-false-security-controls.yml index 4d81da3b..b8a81955 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/quoted-false-security-controls.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/quoted-false-security-controls.yml @@ -8,6 +8,6 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: persist-credentials: "false" diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/raw-after-redaction.yml b/integration/agentcompat/internal/workflowpolicy/testdata/raw-after-redaction.yml index a3c84295..6a21836b 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/raw-after-redaction.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/raw-after-redaction.yml @@ -13,7 +13,7 @@ jobs: if: always() run: go run ./integration/agentcompat/cmd/redact --output "$RUNNER_TEMP/nezha-agentcompat-redacted" - run: cp raw-secret "$RUNNER_TEMP/nezha-agentcompat-redacted/raw-secret" - - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + - uses: actions/upload-artifact@v6 if: always() with: path: ${{ runner.temp }}/nezha-agentcompat-redacted diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/redaction-command-append.yml b/integration/agentcompat/internal/workflowpolicy/testdata/redaction-command-append.yml index 68519f6d..81e0d7bf 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/redaction-command-append.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/redaction-command-append.yml @@ -14,7 +14,7 @@ jobs: run: | go run ./integration/agentcompat/cmd/redact --output "$RUNNER_TEMP/nezha-agentcompat-redacted" cp raw-secret "$RUNNER_TEMP/nezha-agentcompat-redacted/raw-secret" - - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + - uses: actions/upload-artifact@v6 if: always() with: path: ${{ runner.temp }}/nezha-agentcompat-redacted diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/resolver-variable-override.yml b/integration/agentcompat/internal/workflowpolicy/testdata/resolver-variable-override.yml deleted file mode 100644 index c6173ad1..00000000 --- a/integration/agentcompat/internal/workflowpolicy/testdata/resolver-variable-override.yml +++ /dev/null @@ -1,25 +0,0 @@ -on: - pull_request: -concurrency: policy -permissions: - contents: read -jobs: - verify: - runs-on: ubuntu-24.04 - timeout-minutes: 10 - steps: - - id: resolve-agent - run: | - set -euo pipefail - remote='https://github.com/nezhahq/agent.git' - mapfile -t refs < <(git ls-remote "$remote" refs/heads/main) - (( ${#refs[@]} == 1 )) - sha=${refs[0]%%$'\t'*} - [[ "$sha" =~ ^[0-9a-f]{40}$ ]] - sha=attacker-controlled - printf 'sha=%s\n' "$sha" >> "$GITHUB_OUTPUT" - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 - with: - repository: nezhahq/agent - ref: ${{ steps.resolve-agent.outputs.sha }} - persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/secure-agent.yml b/integration/agentcompat/internal/workflowpolicy/testdata/secure-agent.yml index ac94147f..079a4877 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/secure-agent.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/secure-agent.yml @@ -9,13 +9,13 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 30 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: repository: nezhahq/nezha - ref: fedcba9876543210fedcba9876543210fedcba98 + ref: master path: nezha persist-credentials: false - run: go test ./integration/agentcompat/... diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/secure-nezha.yml b/integration/agentcompat/internal/workflowpolicy/testdata/secure-nezha.yml index a123d0c6..99ad07c3 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/secure-nezha.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/secure-nezha.yml @@ -15,14 +15,14 @@ jobs: timeout-minutes: 30 steps: - name: Checkout Nezha - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + uses: actions/checkout@v7.0.1 with: persist-credentials: false - name: Checkout Agent - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + uses: actions/checkout@v7.0.1 with: repository: nezhahq/agent - ref: 0123456789abcdef0123456789abcdef01234567 + ref: v1.2.3 path: agent persist-credentials: false - name: Test @@ -34,7 +34,7 @@ jobs: if: always() run: go run ./integration/agentcompat/cmd/redact --output "$RUNNER_TEMP/nezha-agentcompat-redacted" - name: Upload redacted evidence - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + uses: actions/upload-artifact@v6 if: always() with: path: ${{ runner.temp }}/nezha-agentcompat-redacted diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/secure-resolved-ref.yml b/integration/agentcompat/internal/workflowpolicy/testdata/secure-resolved-ref.yml deleted file mode 100644 index 92c7824c..00000000 --- a/integration/agentcompat/internal/workflowpolicy/testdata/secure-resolved-ref.yml +++ /dev/null @@ -1,27 +0,0 @@ -name: Secure resolved ref -on: - pull_request: -concurrency: resolved-${{ github.ref }} -permissions: - contents: read -jobs: - verify: - runs-on: ubuntu-24.04 - timeout-minutes: 30 - steps: - - id: resolve-agent - name: Resolve Agent commit - shell: bash - run: | - set -euo pipefail - remote='https://github.com/nezhahq/agent.git' - mapfile -t refs < <(git ls-remote "$remote" refs/heads/main) - (( ${#refs[@]} == 1 )) - sha=${refs[0]%%$'\t'*} - [[ "$sha" =~ ^[0-9a-f]{40}$ ]] - printf 'sha=%s\n' "$sha" >> "$GITHUB_OUTPUT" - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 - with: - repository: nezhahq/agent - ref: ${{ steps.resolve-agent.outputs.sha }} - persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/setup-go-default-cache.yml b/integration/agentcompat/internal/workflowpolicy/testdata/setup-go-default-cache.yml index 5e220263..5b2cf1c2 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/setup-go-default-cache.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/setup-go-default-cache.yml @@ -8,6 +8,6 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 + - uses: actions/setup-go@v7 with: go-version: 1.26.3 diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-action.yml b/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-action.yml index 258a32ac..33790dec 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-action.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-action.yml @@ -8,4 +8,4 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: attacker/exfiltrate@0123456789abcdef0123456789abcdef01234567 + - uses: attacker/exfiltrate@v1 diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-repository.yml b/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-repository.yml index 082668ee..88651cf6 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-repository.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/unapproved-repository.yml @@ -8,8 +8,8 @@ jobs: runs-on: ubuntu-24.04 timeout-minutes: 10 steps: - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + - uses: actions/checkout@v7.0.1 with: repository: attacker/fork - ref: 0123456789abcdef0123456789abcdef01234567 + ref: main persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/unredacted-artifact-path.yml b/integration/agentcompat/internal/workflowpolicy/testdata/unredacted-artifact-path.yml index aaffc06c..91b12f1d 100644 --- a/integration/agentcompat/internal/workflowpolicy/testdata/unredacted-artifact-path.yml +++ b/integration/agentcompat/internal/workflowpolicy/testdata/unredacted-artifact-path.yml @@ -11,6 +11,6 @@ jobs: - name: Redact evidence id: redact-evidence run: go run ./integration/agentcompat/cmd/redact - - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f + - uses: actions/upload-artifact@v6 with: path: ${{ runner.temp }}/results diff --git a/integration/agentcompat/internal/workflowpolicy/testdata/unvalidated-resolved-ref.yml b/integration/agentcompat/internal/workflowpolicy/testdata/unvalidated-resolved-ref.yml deleted file mode 100644 index abd2d7ae..00000000 --- a/integration/agentcompat/internal/workflowpolicy/testdata/unvalidated-resolved-ref.yml +++ /dev/null @@ -1,18 +0,0 @@ -on: - pull_request: -concurrency: policy -permissions: - contents: read -jobs: - verify: - runs-on: ubuntu-24.04 - timeout-minutes: 10 - steps: - - id: resolve-agent - run: | - echo "sha=$(git ls-remote https://github.com/nezhahq/agent.git refs/heads/main | cut -f1)" >> "$GITHUB_OUTPUT" - - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 - with: - repository: nezhahq/agent - ref: ${{ steps.resolve-agent.outputs.sha }} - persist-credentials: false diff --git a/integration/agentcompat/internal/workflowpolicy/types.go b/integration/agentcompat/internal/workflowpolicy/types.go index 75ca1aa5..f9564971 100644 --- a/integration/agentcompat/internal/workflowpolicy/types.go +++ b/integration/agentcompat/internal/workflowpolicy/types.go @@ -24,7 +24,6 @@ const ( RuleContainerizedExecution Rule = "containerized-execution" RuleRepositoryNotAllowed Rule = "repository-not-allowed" RuleRepositoryNotLiteral Rule = "repository-not-literal" - RuleOtherRepositoryRef Rule = "other-repository-ref" RulePersistCredentials Rule = "persist-credentials" // #nosec G101 -- GitHub Actions configuration key, not a credential. RuleReusableExecutable Rule = "reusable-executable" RuleContinueOnError Rule = "continue-on-error" diff --git a/service/rpc/nezha.go b/service/rpc/nezha.go index 5ad70c67..e332aa74 100644 --- a/service/rpc/nezha.go +++ b/service/rpc/nezha.go @@ -20,6 +20,10 @@ var _ pb.NezhaServiceServer = (*NezhaHandler)(nil) var NezhaHandlerSingleton *NezhaHandler +// ErrRequestTaskStreamSuperseded is returned when a RequestTask result arrives +// after its stream is no longer the live stream for the authenticated server. +var ErrRequestTaskStreamSuperseded = errors.New("request task stream superseded") + type NezhaHandler struct { Auth *authHandler ioStreams map[string]*ioStreamContext @@ -79,6 +83,18 @@ func clearRequestTaskStream(clientID uint64, captured *model.Server, stream pb.N captured.ClearTaskStreamIfCurrent(stream) } +// currentRequestTaskServer authorizes a received result against the live +// ServerShared entry. Server pointer replacement is valid when it inherited +// the same task stream holder; only a missing entry or different stream makes +// a received result stale. +func currentRequestTaskServer(clientID uint64, stream pb.NezhaService_RequestTaskServer) (*model.Server, error) { + current, ok := singleton.ServerShared.Get(clientID) + if !ok || current == nil || current.GetTaskStream() != stream { + return nil, ErrRequestTaskStreamSuperseded + } + return current, nil +} + func (s *NezhaHandler) RequestTask(stream pb.NezhaService_RequestTaskServer) error { var clientID uint64 var err error @@ -106,6 +122,10 @@ func (s *NezhaHandler) RequestTask(stream pb.NezhaService_RequestTaskServer) err log.Printf("NEZHA>> RequestTask error: %v, clientID: %d\n", err, clientID) return err } + server, err = currentRequestTaskServer(clientID, stream) + if err != nil { + return err + } switch result.GetType() { case model.TaskTypeCommand: // 处理上报的计划任务 diff --git a/service/rpc/request_task_security_test.go b/service/rpc/request_task_security_test.go index 3a86c4e2..3b57016d 100644 --- a/service/rpc/request_task_security_test.go +++ b/service/rpc/request_task_security_test.go @@ -17,11 +17,12 @@ import ( ) type requestTaskSecurityStream struct { - ctx context.Context - results []*pb.TaskResult - onRecv func() - onSend func(*pb.Task) - sendErr error + ctx context.Context + results []*pb.TaskResult + onRecv func() + onResult func() + onSend func(*pb.Task) + sendErr error } func (s *requestTaskSecurityStream) Send(task *pb.Task) error { @@ -40,6 +41,11 @@ func (s *requestTaskSecurityStream) Recv() (*pb.TaskResult, error) { } result := s.results[0] s.results = s.results[1:] + if s.onResult != nil { + onResult := s.onResult + s.onResult = nil + onResult() + } return result, nil } diff --git a/service/rpc/request_task_stale_stream_test.go b/service/rpc/request_task_stale_stream_test.go index 8924c3d9..95885255 100644 --- a/service/rpc/request_task_stale_stream_test.go +++ b/service/rpc/request_task_stale_stream_test.go @@ -6,6 +6,7 @@ import ( "testing" "github.com/nezhahq/nezha/model" + pb "github.com/nezhahq/nezha/proto" "github.com/nezhahq/nezha/service/singleton" ) @@ -44,3 +45,91 @@ func TestRequestTaskCleanupDetachesStreamFromCurrentServerAfterEdit(t *testing.T t.Fatalf("edited server must report offline after the agent stream dropped, got %T", got) } } + +func TestRequestTaskRejectsResultWhenServerDeletedAfterRecv(t *testing.T) { + reporter := requestTaskSecurityServer(7, 200, "10101010-1010-1010-1010-101010101010") + cronTask := requestTaskSecurityCron(42, 200, model.CronCoverAll, nil) + setupRequestTaskSecurityFixture(t, []*model.Server{reporter}, []*model.Cron{cronTask}, map[uint64]model.UserInfo{ + 200: {Role: model.RoleMember}, + }, map[string]uint64{"reporter-secret": 200}) + + stream := requestTaskSecurityAuthedStream("reporter-secret", reporter.UUID) + stream.results = []*pb.TaskResult{cronTaskResult(cronTask.ID, true)} + stream.onResult = func() { + singleton.ServerShared.Delete([]uint64{reporter.ID}) + } + + err := NewNezhaHandler().RequestTask(stream) + if !errors.Is(err, ErrRequestTaskStreamSuperseded) { + t.Fatalf("expected stale RequestTask stream error, got %v", err) + } + assertCronResultNotUpdated(t, cronTask.ID) +} + +func TestRequestTaskRejectsResultWhenNewerStreamSupersedesOld(t *testing.T) { + reporter := requestTaskSecurityServer(7, 200, "20202020-2020-2020-2020-202020202020") + cronTask := requestTaskSecurityCron(42, 200, model.CronCoverAll, nil) + setupRequestTaskSecurityFixture(t, []*model.Server{reporter}, []*model.Cron{cronTask}, map[uint64]model.UserInfo{ + 200: {Role: model.RoleMember}, + }, map[string]uint64{"reporter-secret": 200}) + + current, ok := singleton.ServerShared.Get(reporter.ID) + if !ok { + t.Fatalf("server %d not found", reporter.ID) + } + newer := &requestTaskSecurityStream{ctx: context.Background()} + stream := requestTaskSecurityAuthedStream("reporter-secret", reporter.UUID) + stream.results = []*pb.TaskResult{cronTaskResult(cronTask.ID, true)} + stream.onResult = func() { + current.SetTaskStream(newer) + } + + err := NewNezhaHandler().RequestTask(stream) + if !errors.Is(err, ErrRequestTaskStreamSuperseded) { + t.Fatalf("expected superseded RequestTask stream error, got %v", err) + } + if got := current.GetTaskStream(); got != newer { + t.Fatalf("old stream cleanup must preserve newer stream, got %T", got) + } + assertCronResultNotUpdated(t, cronTask.ID) +} + +func TestRequestTaskAcceptsResultAfterServerPointerReplacementWithSameStream(t *testing.T) { + reporter := requestTaskSecurityServer(7, 200, "30303030-3030-3030-3030-303030303030") + cronTask := requestTaskSecurityCron(42, 200, model.CronCoverAll, nil) + setupRequestTaskSecurityFixture(t, []*model.Server{reporter}, []*model.Cron{cronTask}, map[uint64]model.UserInfo{ + 200: {Role: model.RoleMember}, + }, map[string]uint64{"reporter-secret": 200}) + + old, ok := singleton.ServerShared.Get(reporter.ID) + if !ok { + t.Fatalf("server %d not found", reporter.ID) + } + stream := requestTaskSecurityAuthedStream("reporter-secret", reporter.UUID) + stream.results = []*pb.TaskResult{cronTaskResult(cronTask.ID, true)} + stream.onResult = func() { + replacement := &model.Server{Common: model.Common{ID: old.ID, UserID: old.UserID}, UUID: old.UUID, Name: "replacement"} + replacement.CopyFromRunningServer(old) + singleton.ServerShared.Update(replacement, "") + } + + err := NewNezhaHandler().RequestTask(stream) + if !errors.Is(err, context.Canceled) { + t.Fatalf("expected RequestTask to finish after accepted result, got %v", err) + } + if !cronLastResult(t, cronTask.ID) { + t.Fatal("result on a replacement server that inherited the stream must be accepted") + } +} + +func assertCronResultNotUpdated(t *testing.T, cronID uint64) { + t.Helper() + + var cronTask model.Cron + if err := singleton.DB.First(&cronTask, cronID).Error; err != nil { + t.Fatal(err) + } + if cronTask.LastResult || !cronTask.LastExecutedAt.IsZero() { + t.Fatalf("stale RequestTask result must not mutate cron, got last_result=%t last_executed_at=%s", cronTask.LastResult, cronTask.LastExecutedAt) + } +} diff --git a/service/singleton/server.go b/service/singleton/server.go index ea758ea8..a74a7f62 100644 --- a/service/singleton/server.go +++ b/service/singleton/server.go @@ -6,6 +6,7 @@ import ( "log" "slices" "strings" + "sync" "github.com/nezhahq/nezha/model" "github.com/nezhahq/nezha/pkg/ddns" @@ -15,6 +16,10 @@ import ( type ServerClass struct { class[uint64, *model.Server] + // lifecycleMu serializes changes to the authoritative server entries with + // synchronous ServiceSentinel report processing. + lifecycleMu sync.RWMutex + uuidToID map[string]uint64 sortedListForGuest []*model.Server @@ -71,7 +76,26 @@ func ownerIsAdmin(ownerUID uint64) bool { return userIsAdmin(ownerUID) } +func (c *ServerClass) lockLifecycleRead() { + c.lifecycleMu.RLock() +} + +func (c *ServerClass) unlockLifecycleRead() { + c.lifecycleMu.RUnlock() +} + +func (c *ServerClass) lockLifecycleWrite() { + c.lifecycleMu.Lock() +} + +func (c *ServerClass) unlockLifecycleWrite() { + c.lifecycleMu.Unlock() +} + func (c *ServerClass) Update(s *model.Server, uuid string) { + c.lockLifecycleWrite() + defer c.unlockLifecycleWrite() + c.listMu.Lock() c.list[s.ID] = s @@ -91,6 +115,9 @@ func (c *ServerClass) Update(s *model.Server, uuid string) { } func (c *ServerClass) Delete(idList []uint64) { + c.lockLifecycleWrite() + defer c.unlockLifecycleWrite() + c.listMu.Lock() for _, id := range idList { @@ -107,6 +134,17 @@ func (c *ServerClass) Delete(idList []uint64) { c.sortList() } +// setUserID updates in-memory ownership under the server lifecycle lock so a +// transfer cannot change authorization during synchronous report processing. +func (c *ServerClass) setUserID(id, userID uint64) { + c.lockLifecycleWrite() + defer c.unlockLifecycleWrite() + + if s, ok := c.Get(id); ok && s != nil { + s.SetUserID(userID) + } +} + func (c *ServerClass) GetSortedListForGuest() []*model.Server { c.sortedListMu.RLock() defer c.sortedListMu.RUnlock() diff --git a/service/singleton/server_transfer.go b/service/singleton/server_transfer.go index fe87aef8..cef41476 100644 --- a/service/singleton/server_transfer.go +++ b/service/singleton/server_transfer.go @@ -663,11 +663,9 @@ func (c *ServerTransferClass) Initiate(tx *gorm.DB, serverID, fromUserID, toUser // and admits the old AgentSecret on the happy "owner match" path — // bypassing the bounded pending-tolerance contract. func (c *ServerTransferClass) Register(t *model.ServerTransfer) { - if s, ok := ServerShared.Get(t.ServerID); ok && s != nil { - // SetUserID over atomic write — auth.go hot path concurrently - // reads this field; a plain assignment would be a data race. - s.SetUserID(t.ToUserID) - } + // SetUserID uses an atomic write because auth.go reads this hot-path field + // concurrently; ServerClass also serializes it with service reports. + ServerShared.setUserID(t.ServerID, t.ToUserID) c.mu.Lock() c.pending[t.ServerID] = t @@ -1163,9 +1161,7 @@ func (c *ServerTransferClass) revertTransition(transferID uint64, newStatus mode // no longer admits the destination user's global AgentSecret via // ServerShared.GetUserID() == userId on the happy "owner match" path. if transitionedByThisCall { - if s, ok := ServerShared.Get(t.ServerID); ok && s != nil { - s.SetUserID(t.FromUserID) - } + ServerShared.setUserID(t.ServerID, t.FromUserID) } // Self-heal: any non-Pending DB status invalidates the in-memory entry — diff --git a/service/singleton/servicesentinel.go b/service/singleton/servicesentinel.go index 2a98aab6..dd8882ff 100644 --- a/service/singleton/servicesentinel.go +++ b/service/singleton/servicesentinel.go @@ -394,7 +394,16 @@ func (ss *ServiceSentinel) Delete(ids []uint64) { delete(ss.serviceStatusToday, id) // 停掉定时任务 - CronShared.Remove(ss.services[id].CronJobID) + // GHSA-jx78-55p5-rwv5 (Finding 2): guard against a caller supplying an id + // that does not exist in the in-memory registry. CheckPermission returns + // vacuously true for unknown ids, so the controller layer cannot prevent + // this. Without the guard, ss.services[id] is nil and the .CronJobID + // field access panics, aborting the Delete loop before the remaining valid + // ids are cleaned from memory — their service records were already deleted + // from the database, producing zombie services. + if svc := ss.services[id]; svc != nil { + CronShared.Remove(svc.CronJobID) + } delete(ss.services, id) delete(ss.monthlyStatus, id) @@ -536,238 +545,252 @@ func (ss *ServiceSentinel) Close() { func (ss *ServiceSentinel) worker() { // 从服务状态汇报管道获取汇报的服务数据 for r := range ss.serviceReportChannel { - cs, _ := ss.Get(r.Data.GetId()) - reporter, _ := ServerShared.Get(r.Reporter) - // 入站结果必须匹配出站任务派发边界，避免 agent 伪造其他服务 ID 写入监控状态。 - if !canReportServiceResult(cs, reporter, r.Data.GetType()) { - log.Printf("NEZHA>> Incorrect service monitor report %+v", r) - continue - } - if ss.serviceReportValidatedHook != nil { - ss.serviceReportValidatedHook(r.Data.GetId()) - } - - mh := r.Data - // Serialize Delete and Update before this accepted report causes any side effect. - ss.serviceResponseDataStoreLock.Lock() - serviceStatusToday := ss.serviceStatusToday[mh.GetId()] - serviceCurrentStatusData := ss.serviceCurrentStatusData[mh.GetId()] - currentService, serviceExists := ss.Get(mh.GetId()) - if serviceStatusToday == nil || serviceCurrentStatusData == nil || !serviceExists || - !canReportServiceResult(currentService, reporter, mh.GetType()) { - ss.serviceResponseDataStoreLock.Unlock() - continue - } - cs = currentService - - if mh.Type == model.TaskTypeTCPPing || mh.Type == model.TaskTypeICMPPing { - // TCP/ICMP Ping 使用平均值计算后再写入 - serviceTcpMap, ok := ss.serviceResponsePing[mh.GetId()] - if !ok { - serviceTcpMap = make(map[uint64]*pingStore) - ss.serviceResponsePing[mh.GetId()] = serviceTcpMap - } - ts, ok := serviceTcpMap[r.Reporter] - if !ok { - ts = &pingStore{} - } - ts.count++ - ts.ping = (ts.ping*float64(ts.count-1) + float64(mh.Delay)) / float64(ts.count) - if mh.Successful { - ts.successCount++ - } - if ts.count == Conf.AvgPingCount { - if TSDBEnabled() { - if err := TSDBShared.WriteServiceMetrics(&tsdb.ServiceMetrics{ - ServiceID: mh.GetId(), - ServerID: r.Reporter, - Timestamp: time.Now(), - Delay: ts.ping, - Successful: ts.successCount*2 >= ts.count, - }); err != nil { - log.Printf("NEZHA>> Failed to save service monitor metrics to TSDB: %v", err) - } - } else { - if err := DB.Create(&model.ServiceHistory{ - ServiceID: mh.GetId(), - AvgDelay: ts.ping, - Data: mh.Data, - ServerID: r.Reporter, - }).Error; err != nil { - log.Printf("NEZHA>> Failed to save service monitor metrics: %v", err) - } + serverShared := ServerShared + func() { + defer func() { + if recovered := recover(); recovered != nil { + log.Printf("NEZHA>> Service monitor report processing panicked: %v", recovered) } - ts.count = 0 - ts.ping = 0 - ts.successCount = 0 - } - serviceTcpMap[r.Reporter] = ts - } else { + }() + ss.processReport(r, serverShared) + }() + } +} + +func (ss *ServiceSentinel) processReport(r ReportData, serverShared *ServerClass) { + serverShared.lockLifecycleRead() + defer serverShared.unlockLifecycleRead() + + cs, _ := ss.Get(r.Data.GetId()) + reporter, _ := serverShared.Get(r.Reporter) + // 入站结果必须匹配出站任务派发边界，避免 agent 伪造其他服务 ID 写入监控状态。 + if !canReportServiceResult(cs, reporter, r.Data.GetType()) { + log.Printf("NEZHA>> Incorrect service monitor report %+v", r) + return + } + if ss.serviceReportValidatedHook != nil { + ss.serviceReportValidatedHook(r.Data.GetId()) + } + + mh := r.Data + m := serverShared.GetList() + // Serialize Delete and Update before this accepted report causes any side effect. + ss.serviceResponseDataStoreLock.Lock() + defer ss.serviceResponseDataStoreLock.Unlock() + serviceStatusToday := ss.serviceStatusToday[mh.GetId()] + serviceCurrentStatusData := ss.serviceCurrentStatusData[mh.GetId()] + currentService, serviceExists := ss.Get(mh.GetId()) + if serviceStatusToday == nil || serviceCurrentStatusData == nil || !serviceExists || + !canReportServiceResult(currentService, reporter, mh.GetType()) { + return + } + cs = currentService + + if mh.Type == model.TaskTypeTCPPing || mh.Type == model.TaskTypeICMPPing { + // TCP/ICMP Ping 使用平均值计算后再写入 + serviceTcpMap, ok := ss.serviceResponsePing[mh.GetId()] + if !ok { + serviceTcpMap = make(map[uint64]*pingStore) + ss.serviceResponsePing[mh.GetId()] = serviceTcpMap + } + ts, ok := serviceTcpMap[r.Reporter] + if !ok { + ts = &pingStore{} + } + ts.count++ + ts.ping = (ts.ping*float64(ts.count-1) + float64(mh.Delay)) / float64(ts.count) + if mh.Successful { + ts.successCount++ + } + if ts.count == Conf.AvgPingCount { if TSDBEnabled() { if err := TSDBShared.WriteServiceMetrics(&tsdb.ServiceMetrics{ ServiceID: mh.GetId(), ServerID: r.Reporter, Timestamp: time.Now(), - Delay: float64(mh.Delay), - Successful: mh.Successful, + Delay: ts.ping, + Successful: ts.successCount*2 >= ts.count, }); err != nil { log.Printf("NEZHA>> Failed to save service monitor metrics to TSDB: %v", err) } - } - } - - // 写入当天状态 - if mh.Successful { - serviceStatusToday.Delay = (serviceStatusToday.Delay*float64(serviceStatusToday.Up) + - float64(mh.Delay)) / float64(serviceStatusToday.Up+1) - serviceStatusToday.Up++ - } else { - serviceStatusToday.Down++ - } - - currentTime := time.Now() - if serviceCurrentStatusData.t.IsZero() { - serviceCurrentStatusData.t = currentTime - } - - // 写入当前数据 - if serviceCurrentStatusData.t.Before(currentTime) { - serviceCurrentStatusData.t = currentTime.Add(30 * time.Second) - serviceCurrentStatusData.result = append(serviceCurrentStatusData.result, mh) - } - - // 更新当前状态 - ss.serviceResponseDataStore[mh.GetId()] = serviceResponseData{} - - // 永远是最新的 30 个数据的状态 [01:00, 02:00, 03:00] -> [04:00, 02:00, 03: 00] - for _, cs := range serviceCurrentStatusData.result { - if cs.GetId() > 0 { - rd := ss.serviceResponseDataStore[mh.GetId()] - if cs.Successful { - rd.Up++ - rd.Delay = (rd.Delay*float64(rd.Up-1) + float64(cs.Delay)) / float64(rd.Up) - } else { - rd.Down++ - } - ss.serviceResponseDataStore[mh.GetId()] = rd - } - } - - // 计算在线率， - var stateCode uint8 - { - upPercent := uint64(0) - rd := ss.serviceResponseDataStore[mh.GetId()] - if rd.Down+rd.Up > 0 { - upPercent = rd.Up * 100 / (rd.Down + rd.Up) - } - stateCode = GetStatusCode(upPercent) - } - - if len(serviceCurrentStatusData.result) == _CurrentStatusSize { - serviceCurrentStatusData.t = currentTime - if !TSDBEnabled() { - rd := ss.serviceResponseDataStore[mh.GetId()] + } else { if err := DB.Create(&model.ServiceHistory{ ServiceID: mh.GetId(), - AvgDelay: rd.Delay, + AvgDelay: ts.ping, Data: mh.Data, - Up: rd.Up, - Down: rd.Down, + ServerID: r.Reporter, }).Error; err != nil { log.Printf("NEZHA>> Failed to save service monitor metrics: %v", err) } } - serviceCurrentStatusData.result = serviceCurrentStatusData.result[:0] + ts.count = 0 + ts.ping = 0 + ts.successCount = 0 } - - m := ServerShared.GetList() - // 延迟报警 - if mh.Delay > 0 { - delayCheck(&r, m, cs, mh) - } - - // 状态变更报警+触发任务执行 - if stateCode == StatusDown || stateCode != serviceCurrentStatusData.lastStatus { - lastStatus := serviceCurrentStatusData.lastStatus - // 存储新的状态值 - serviceCurrentStatusData.lastStatus = stateCode - - notifyCheck(&r, m, cs, mh, lastStatus, stateCode) - } - - // TLS 证书报警 - if ss.serviceReportBeforeTLSSideEffectsHook != nil { - ss.serviceReportBeforeTLSSideEffectsHook(mh.GetId()) - } - var errMsg string - if strings.HasPrefix(mh.Data, "SSL证书错误：") { - // i/o timeout、connection timeout、EOF 错误 - if !strings.HasSuffix(mh.Data, "timeout") && - !strings.HasSuffix(mh.Data, "EOF") && - !strings.HasSuffix(mh.Data, "timed out") { - errMsg = mh.Data - if cs.Notify { - muteLabel := NotificationMuteLabel.ServiceTLS(mh.GetId(), "network") - go NotificationShared.SendNotification(cs.NotificationGroupID, Localizer.Tf("[TLS] Fetch cert info failed, Reporter: %s, Error: %s", cs.Name, errMsg), muteLabel) - } + serviceTcpMap[r.Reporter] = ts + } else { + if TSDBEnabled() { + if err := TSDBShared.WriteServiceMetrics(&tsdb.ServiceMetrics{ + ServiceID: mh.GetId(), + ServerID: r.Reporter, + Timestamp: time.Now(), + Delay: float64(mh.Delay), + Successful: mh.Successful, + }); err != nil { + log.Printf("NEZHA>> Failed to save service monitor metrics to TSDB: %v", err) } - } else { - // 清除网络错误静音缓存 - NotificationShared.UnMuteNotification(cs.NotificationGroupID, NotificationMuteLabel.ServiceTLS(mh.GetId(), "network")) + } + } - var newCert = strings.Split(mh.Data, "|") - if len(newCert) > 1 { - enableNotify := cs.Notify + // 写入当天状态 + if mh.Successful { + serviceStatusToday.Delay = (serviceStatusToday.Delay*float64(serviceStatusToday.Up) + + float64(mh.Delay)) / float64(serviceStatusToday.Up+1) + serviceStatusToday.Up++ + } else { + serviceStatusToday.Down++ + } - // 首次获取证书信息时，缓存证书信息 - if ss.tlsCertCache[mh.GetId()] == "" { - ss.tlsCertCache[mh.GetId()] = mh.Data + currentTime := time.Now() + if serviceCurrentStatusData.t.IsZero() { + serviceCurrentStatusData.t = currentTime + } + + // 写入当前数据 + if serviceCurrentStatusData.t.Before(currentTime) { + serviceCurrentStatusData.t = currentTime.Add(30 * time.Second) + serviceCurrentStatusData.result = append(serviceCurrentStatusData.result, mh) + } + + // 更新当前状态 + ss.serviceResponseDataStore[mh.GetId()] = serviceResponseData{} + + // 永远是最新的 30 个数据的状态 [01:00, 02:00, 03:00] -> [04:00, 02:00, 03: 00] + for _, cs := range serviceCurrentStatusData.result { + if cs.GetId() > 0 { + rd := ss.serviceResponseDataStore[mh.GetId()] + if cs.Successful { + rd.Up++ + rd.Delay = (rd.Delay*float64(rd.Up-1) + float64(cs.Delay)) / float64(rd.Up) + } else { + rd.Down++ + } + ss.serviceResponseDataStore[mh.GetId()] = rd + } + } + + // 计算在线率， + var stateCode uint8 + { + upPercent := uint64(0) + rd := ss.serviceResponseDataStore[mh.GetId()] + if rd.Down+rd.Up > 0 { + upPercent = rd.Up * 100 / (rd.Down + rd.Up) + } + stateCode = GetStatusCode(upPercent) + } + + if len(serviceCurrentStatusData.result) == _CurrentStatusSize { + serviceCurrentStatusData.t = currentTime + if !TSDBEnabled() { + rd := ss.serviceResponseDataStore[mh.GetId()] + if err := DB.Create(&model.ServiceHistory{ + ServiceID: mh.GetId(), + AvgDelay: rd.Delay, + Data: mh.Data, + Up: rd.Up, + Down: rd.Down, + }).Error; err != nil { + log.Printf("NEZHA>> Failed to save service monitor metrics: %v", err) + } + } + serviceCurrentStatusData.result = serviceCurrentStatusData.result[:0] + } + + // 延迟报警 + if mh.Delay > 0 { + delayCheck(&r, m, cs, mh) + } + + // 状态变更报警+触发任务执行 + if stateCode == StatusDown || stateCode != serviceCurrentStatusData.lastStatus { + lastStatus := serviceCurrentStatusData.lastStatus + // 存储新的状态值 + serviceCurrentStatusData.lastStatus = stateCode + + notifyCheck(&r, m, cs, mh, lastStatus, stateCode) + } + + // TLS 证书报警 + if ss.serviceReportBeforeTLSSideEffectsHook != nil { + ss.serviceReportBeforeTLSSideEffectsHook(mh.GetId()) + } + var errMsg string + if strings.HasPrefix(mh.Data, "SSL证书错误：") { + // i/o timeout、connection timeout、EOF 错误 + if !strings.HasSuffix(mh.Data, "timeout") && + !strings.HasSuffix(mh.Data, "EOF") && + !strings.HasSuffix(mh.Data, "timed out") { + errMsg = mh.Data + if cs.Notify { + muteLabel := NotificationMuteLabel.ServiceTLS(mh.GetId(), "network") + go NotificationShared.SendNotification(cs.NotificationGroupID, Localizer.Tf("[TLS] Fetch cert info failed, Reporter: %s, Error: %s", cs.Name, errMsg), muteLabel) + } + } + } else { + // 清除网络错误静音缓存 + NotificationShared.UnMuteNotification(cs.NotificationGroupID, NotificationMuteLabel.ServiceTLS(mh.GetId(), "network")) + + var newCert = strings.Split(mh.Data, "|") + if len(newCert) > 1 { + enableNotify := cs.Notify + + // 首次获取证书信息时，缓存证书信息 + if ss.tlsCertCache[mh.GetId()] == "" { + ss.tlsCertCache[mh.GetId()] = mh.Data + } + + oldCert := strings.Split(ss.tlsCertCache[mh.GetId()], "|") + isCertChanged := false + expiresOld, _ := time.Parse("2006-01-02 15:04:05 -0700 MST", oldCert[1]) + expiresNew, _ := time.Parse("2006-01-02 15:04:05 -0700 MST", newCert[1]) + + // 证书变更时，更新缓存 + if oldCert[0] != newCert[0] && !expiresNew.Equal(expiresOld) { + isCertChanged = true + ss.tlsCertCache[mh.GetId()] = mh.Data + } + + notificationGroupID := cs.NotificationGroupID + serviceName := cs.Name + + // 需要发送提醒 + if enableNotify { + // 证书过期提醒 + if expiresNew.Before(time.Now().AddDate(0, 0, 7)) { + expiresTimeStr := expiresNew.Format("2006-01-02 15:04:05") + errMsg = Localizer.Tf( + "The TLS certificate will expire within seven days. Expiration time: %s", + expiresTimeStr, + ) + + // 静音规则： 服务id+证书过期时间 + // 用于避免多个监测点对相同证书同时报警 + muteLabel := NotificationMuteLabel.ServiceTLS(mh.GetId(), fmt.Sprintf("expire_%s", expiresTimeStr)) + go NotificationShared.SendNotification(notificationGroupID, fmt.Sprintf("[TLS] %s %s", serviceName, errMsg), muteLabel) } - oldCert := strings.Split(ss.tlsCertCache[mh.GetId()], "|") - isCertChanged := false - expiresOld, _ := time.Parse("2006-01-02 15:04:05 -0700 MST", oldCert[1]) - expiresNew, _ := time.Parse("2006-01-02 15:04:05 -0700 MST", newCert[1]) + // 证书变更提醒 + if isCertChanged { + errMsg = Localizer.Tf( + "TLS certificate changed, old: issuer %s, expires at %s; new: issuer %s, expires at %s", + oldCert[0], expiresOld.Format("2006-01-02 15:04:05"), newCert[0], expiresNew.Format("2006-01-02 15:04:05")) - // 证书变更时，更新缓存 - if oldCert[0] != newCert[0] && !expiresNew.Equal(expiresOld) { - isCertChanged = true - ss.tlsCertCache[mh.GetId()] = mh.Data - } - - notificationGroupID := cs.NotificationGroupID - serviceName := cs.Name - - // 需要发送提醒 - if enableNotify { - // 证书过期提醒 - if expiresNew.Before(time.Now().AddDate(0, 0, 7)) { - expiresTimeStr := expiresNew.Format("2006-01-02 15:04:05") - errMsg = Localizer.Tf( - "The TLS certificate will expire within seven days. Expiration time: %s", - expiresTimeStr, - ) - - // 静音规则： 服务id+证书过期时间 - // 用于避免多个监测点对相同证书同时报警 - muteLabel := NotificationMuteLabel.ServiceTLS(mh.GetId(), fmt.Sprintf("expire_%s", expiresTimeStr)) - go NotificationShared.SendNotification(notificationGroupID, fmt.Sprintf("[TLS] %s %s", serviceName, errMsg), muteLabel) - } - - // 证书变更提醒 - if isCertChanged { - errMsg = Localizer.Tf( - "TLS certificate changed, old: issuer %s, expires at %s; new: issuer %s, expires at %s", - oldCert[0], expiresOld.Format("2006-01-02 15:04:05"), newCert[0], expiresNew.Format("2006-01-02 15:04:05")) - - // 证书变更后会自动更新缓存，所以不需要静音 - go NotificationShared.SendNotification(notificationGroupID, fmt.Sprintf("[TLS] %s %s", serviceName, errMsg), "") - } + // 证书变更后会自动更新缓存，所以不需要静音 + go NotificationShared.SendNotification(notificationGroupID, fmt.Sprintf("[TLS] %s %s", serviceName, errMsg), "") } } } - ss.serviceResponseDataStoreLock.Unlock() } } @@ -776,17 +799,25 @@ func delayCheck(r *ReportData, m map[uint64]*model.Server, ss *model.Service, mh return } + // GHSA-jx78-55p5-rwv5 (incomplete fix of GHSA-qjpp-gffx-2wm9): the server + // map snapshot m is taken outside serviceResponseDataStoreLock and + // ServerShared has its own independent lock, so a concurrent batch-delete of + // the reporter's server can remove the entry between the pre-lock validation + // and this point. Guard against the nil pointer before using the server. + reporterServer := m[r.Reporter] + if reporterServer == nil { + return + } + notificationGroupID := ss.NotificationGroupID minMuteLabel := NotificationMuteLabel.ServiceLatencyMin(mh.GetId()) maxMuteLabel := NotificationMuteLabel.ServiceLatencyMax(mh.GetId()) if mh.Delay > ss.MaxLatency { // 延迟超过最大值 - reporterServer := m[r.Reporter] msg := Localizer.Tf("[Latency] %s %2f > %2f, Reporter: %s", ss.Name, mh.Delay, ss.MaxLatency, reporterServer.Name) go NotificationShared.SendNotification(notificationGroupID, msg, minMuteLabel) } else if mh.Delay < ss.MinLatency { // 延迟低于最小值 - reporterServer := m[r.Reporter] msg := Localizer.Tf("[Latency] %s %2f < %2f, Reporter: %s", ss.Name, mh.Delay, ss.MinLatency, reporterServer.Name) go NotificationShared.SendNotification(notificationGroupID, msg, maxMuteLabel) } else { @@ -798,10 +829,16 @@ func delayCheck(r *ReportData, m map[uint64]*model.Server, ss *model.Service, mh func notifyCheck(r *ReportData, m map[uint64]*model.Server, ss *model.Service, mh *pb.TaskResult, lastStatus, stateCode uint8) { + // GHSA-jx78-55p5-rwv5: guard against concurrent server deletion (same TOCTOU + // class as the 2026-07-21 fix, a few dozen lines lower in the same worker). + // ServerShared has its own lock; m is a snapshot taken outside + // serviceResponseDataStoreLock, so the server may have been removed between + // the pre-lock validation and here. + reporterServer := m[r.Reporter] + // 判断是否需要发送通知 isNeedSendNotification := ss.Notify && (lastStatus != 0 || stateCode == StatusDown) - if isNeedSendNotification { - reporterServer := m[r.Reporter] + if isNeedSendNotification && reporterServer != nil { notificationGroupID := ss.NotificationGroupID notificationMsg := Localizer.Tf("[%s] %s Reporter: %s, Error: %s", StatusCodeToString(stateCode), ss.Name, reporterServer.Name, mh.Data) muteLabel := NotificationMuteLabel.ServiceStateChanged(mh.GetId()) @@ -816,8 +853,7 @@ func notifyCheck(r *ReportData, m map[uint64]*model.Server, // 判断是否需要触发任务 isNeedTriggerTask := ss.EnableTriggerTask && lastStatus != 0 - if isNeedTriggerTask { - reporterServer := m[r.Reporter] + if isNeedTriggerTask && reporterServer != nil { if stateCode == StatusGood && lastStatus != stateCode { // 当前状态正常 前序状态非正常时 触发恢复任务 go CronShared.SendTriggerTasks(ss.RecoverTriggerTasks, reporterServer.ID, ss.UserID) diff --git a/service/singleton/servicesentinel_lifecycle_test.go b/service/singleton/servicesentinel_lifecycle_test.go index 0941b453..af8a24ac 100644 --- a/service/singleton/servicesentinel_lifecycle_test.go +++ b/service/singleton/servicesentinel_lifecycle_test.go @@ -13,8 +13,183 @@ import ( "github.com/nezhahq/nezha/model" ) +// Regression markers for Finding 1 and Finding 2 of GHSA-jx78-55p5-rwv5 +// (incomplete fix of GHSA-qjpp-gffx-2wm9). +const ( + concurrentServerDeleteSuccessMarker = "ghsa-jx78-55p5-rwv5-finding1-no-crash" + deleteUnknownIDSuccessMarker = "ghsa-jx78-55p5-rwv5-finding2-no-zombie" +) + const serviceSentinelLifecycleSuccessMarker = "service-sentinel-stale-report-lifecycle-success" +func TestServiceSentinelReporterDeleteWaitsForSynchronousReportProcessing(t *testing.T) { + ctx, cancel := context.WithTimeout(t.Context(), time.Second) + defer cancel() + ss := newServiceMonitorSecurityHarness(t, + &model.Server{Common: model.Common{ID: 1, UserID: 1}, Name: "reporter"}, + ) + service := &model.Service{ + Common: model.Common{ID: 10, UserID: 1}, + Name: "lifecycle-service", + Type: model.TaskTypeTCPPing, + Target: "lifecycle.example.invalid:443", + Duration: 3600, + Cover: model.ServiceCoverIgnoreAll, + SkipServers: map[uint64]bool{1: true}, + } + addServiceMonitorSecurityService(t, ss, service) + + reportValidated := make(chan struct{}) + releaseReport := make(chan struct{}) + var releaseOnce sync.Once + release := func() { releaseOnce.Do(func() { close(releaseReport) }) } + ss.serviceReportValidatedHook = func(serviceID uint64) { + if serviceID == service.ID { + close(reportValidated) + <-releaseReport + } + } + t.Cleanup(func() { + release() + ss.Close() + }) + + ss.Dispatch(serviceMonitorResult(1, service.ID, model.TaskTypeTCPPing, true)) + select { + case <-reportValidated: + case <-ctx.Done(): + t.Fatal(ctx.Err()) + } + + deleteDone := make(chan struct{}) + go func() { + ServerShared.Delete([]uint64{1}) + close(deleteDone) + }() + select { + case <-deleteDone: + t.Fatal("server deletion returned before the accepted report completed") + case <-time.After(25 * time.Millisecond): + } + + release() + select { + case <-deleteDone: + case <-ctx.Done(): + t.Fatal(ctx.Err()) + } + + var historyCount int64 + if err := DB.Model(&model.ServiceHistory{}). + Where("service_id = ? AND server_id = ?", service.ID, 1). + Count(&historyCount).Error; err != nil { + t.Fatal(err) + } + if historyCount != 1 { + t.Fatalf("expected report side effects before deletion returned, got %d history rows", historyCount) + } + if _, ok := ServerShared.Get(1); ok { + t.Fatal("expected reporter to be deleted after the report completed") + } +} + +func TestServiceSentinelWorkerRejectsReportAfterReporterDeletion(t *testing.T) { + ss := newServiceMonitorSecurityHarness(t, + &model.Server{Common: model.Common{ID: 1, UserID: 1}, Name: "reporter"}, + ) + service := &model.Service{ + Common: model.Common{ID: 10, UserID: 1}, + Name: "deleted-reporter-service", + Type: model.TaskTypeTCPPing, + Target: "deleted-reporter.example.invalid:443", + Duration: 3600, + Cover: model.ServiceCoverIgnoreAll, + SkipServers: map[uint64]bool{1: true}, + } + addServiceMonitorSecurityService(t, ss, service) + + ServerShared.Delete([]uint64{1}) + ss.Dispatch(serviceMonitorResult(1, service.ID, model.TaskTypeTCPPing, true)) + ss.Close() + + var historyCount int64 + if err := DB.Model(&model.ServiceHistory{}). + Where("service_id = ? AND server_id = ?", service.ID, 1). + Count(&historyCount).Error; err != nil { + t.Fatal(err) + } + if historyCount != 0 { + t.Fatalf("expected no history after reporter deletion, got %d rows", historyCount) + } + ss.serviceResponseDataStoreLock.RLock() + _, pingCached := ss.serviceResponsePing[service.ID] + _, responseCached := ss.serviceResponseDataStore[service.ID] + stats := ss.serviceStatusToday[service.ID] + ss.serviceResponseDataStoreLock.RUnlock() + if pingCached { + t.Fatal("expected no ping cache side effect after reporter deletion") + } + if responseCached { + t.Fatal("expected no response cache side effect after reporter deletion") + } + if stats == nil || stats.Up != 0 || stats.Down != 0 { + t.Fatalf("expected no stats side effect after reporter deletion, got %+v", stats) + } +} + +func TestServiceSentinelWorkerRecoversPerReportPanic(t *testing.T) { + ctx, cancel := context.WithTimeout(t.Context(), time.Second) + defer cancel() + ss := newServiceMonitorSecurityHarness(t, + &model.Server{Common: model.Common{ID: 1, UserID: 1}, Name: "reporter"}, + ) + panicService := &model.Service{ + Common: model.Common{ID: 10, UserID: 1}, + Name: "panic-service", + Type: model.TaskTypeHTTPGet, + Target: "https://panic.example.invalid", + Duration: 3600, + Cover: model.ServiceCoverIgnoreAll, + SkipServers: map[uint64]bool{1: true}, + } + validService := &model.Service{ + Common: model.Common{ID: 20, UserID: 1}, + Name: "valid-service", + Type: model.TaskTypeTCPPing, + Target: "valid.example.invalid:443", + Duration: 3600, + Cover: model.ServiceCoverIgnoreAll, + SkipServers: map[uint64]bool{1: true}, + } + addServiceMonitorSecurityService(t, ss, panicService) + addServiceMonitorSecurityService(t, ss, validService) + ss.serviceReportBeforeTLSSideEffectsHook = func(serviceID uint64) { + if serviceID == panicService.ID { + panic("test service report panic") + } + } + + ss.Dispatch(serviceMonitorResult(1, panicService.ID, model.TaskTypeHTTPGet, true)) + ss.Dispatch(serviceMonitorResult(1, validService.ID, model.TaskTypeTCPPing, true)) + waitForServiceHistory(t, validService.ID, 1) + ss.Close() + if !ss.serviceResponseDataStoreLock.TryLock() { + t.Fatal("panic leaked the service response lock") + } + ss.serviceResponseDataStoreLock.Unlock() + + deleteDone := make(chan struct{}) + go func() { + ServerShared.Delete([]uint64{1}) + close(deleteDone) + }() + select { + case <-deleteDone: + case <-ctx.Done(): + t.Fatal("panic leaked a lifecycle lock: " + ctx.Err().Error()) + } +} + func TestServiceSentinelWorkerIgnoresStaleReportAfterDeletion(t *testing.T) { if os.Getenv("NEZHA_SERVICE_SENTINEL_LIFECYCLE_CHILD") == "1" { testServiceSentinelWorkerIgnoresStaleReportAfterDeletionChild(t) @@ -299,3 +474,173 @@ func TestServiceSentinelWorkerHoldsResponseLockDuringTLSSideEffects(t *testing.T t.Fatalf("expected TLS cache %q, got %q", report.Data.Data, cachedCertificate) } } + +// TestServiceSentinelWorkerSurvivesConcurrentReporterServerDelete is a +// regression test for GHSA-jx78-55p5-rwv5 Finding 1 (incomplete fix of +// GHSA-qjpp-gffx-2wm9). +// +// The vulnerability: after the 2026-07-21 fix, the worker re-validates the +// service under serviceResponseDataStoreLock, but then takes a fresh snapshot +// m := ServerShared.GetList() with no guard. A concurrent batch-delete of the +// reporter's own server removes it between the pre-lock validation and the +// GetList call, so m[r.Reporter] is nil. delayCheck and notifyCheck then +// dereference m[r.Reporter].Name unconditionally — SIGSEGV. +// +// The subprocess-isolation pattern is used because the pre-fix code path +// panicked (nil pointer dereference in an unrecovered goroutine), which would +// crash the whole test binary rather than simply failing a single test. +func TestServiceSentinelWorkerSurvivesConcurrentReporterServerDelete(t *testing.T) { + if os.Getenv("NEZHA_SENTINEL_CONCURRENT_DELETE_CHILD") == "1" { + testServiceSentinelWorkerSurvivesConcurrentReporterServerDeleteChild(t) + return + } + + ctx, cancel := context.WithTimeout(t.Context(), 5*time.Second) + defer cancel() + child := exec.CommandContext(ctx, os.Args[0], + "-test.run=^TestServiceSentinelWorkerSurvivesConcurrentReporterServerDelete$", + "-test.v", + ) + child.Env = append(os.Environ(), "NEZHA_SENTINEL_CONCURRENT_DELETE_CHILD=1") + + output, err := child.CombinedOutput() + if ctx.Err() != nil { + t.Fatalf("child process timed out: %v\n%s", ctx.Err(), output) + } + if err != nil { + t.Fatalf("child process crashed (likely nil deref in delayCheck/notifyCheck): %v\n%s", err, output) + } + if !strings.Contains(string(output), concurrentServerDeleteSuccessMarker) { + t.Fatalf("child did not print success marker:\n%s", output) + } +} + +func testServiceSentinelWorkerSurvivesConcurrentReporterServerDeleteChild(t *testing.T) { + // Given: a reporter server and a service with latency-alerting enabled so + // that delayCheck (the vulnerable sink at line 785) is exercised on every + // dispatch. MaxLatency=1 ensures delay=12 always exceeds the threshold and + // the notification branch (not just the mute-clear branch) is taken. + ss := newServiceMonitorSecurityHarness(t, + &model.Server{Common: model.Common{ID: 1, UserID: 1}, Name: "reporter"}, + ) + service := &model.Service{ + Common: model.Common{ID: 10, UserID: 1}, + Name: "latency-service", + Type: model.TaskTypeTCPPing, + Target: "example.invalid:443", + Duration: 3600, + Cover: model.ServiceCoverIgnoreAll, + SkipServers: map[uint64]bool{1: true}, + LatencyNotify: true, + MaxLatency: 1, + } + addServiceMonitorSecurityService(t, ss, service) + + reportProcessing := make(chan struct{}) + releaseWorker := make(chan struct{}) + var releaseOnce sync.Once + releaseWorkerFn := func() { releaseOnce.Do(func() { close(releaseWorker) }) } + + // serviceReportValidatedHook runs while the report holds the lifecycle read + // lock. Deletion must therefore run in another goroutine and wait until this + // hook releases; attempting Delete here would try to upgrade the RWMutex. + ss.serviceReportValidatedHook = func(serviceID uint64) { + if serviceID == service.ID { + close(reportProcessing) + <-releaseWorker + } + } + t.Cleanup(func() { + releaseWorkerFn() + ss.Close() + }) + + // When + ss.Dispatch(serviceMonitorResult(1, service.ID, model.TaskTypeTCPPing, true)) + select { + case <-reportProcessing: + case <-t.Context().Done(): + t.Fatal(t.Context().Err()) + } + deleteDone := make(chan struct{}) + go func() { + ServerShared.Delete([]uint64{1}) + close(deleteDone) + }() + releaseWorkerFn() + select { + case <-deleteDone: + case <-t.Context().Done(): + t.Fatal(t.Context().Err()) + } + ss.Close() + + // Then: no crash; the worker handled the nil reporter gracefully. + if _, err := fmt.Fprintln(os.Stdout, concurrentServerDeleteSuccessMarker); err != nil { + t.Fatal(err) + } +} + +// TestServiceSentinelDeleteWithUnknownIDDoesNotLeaveZombies is a regression +// test for GHSA-jx78-55p5-rwv5 Finding 2 (low severity). +// +// The vulnerability: ServiceSentinel.Delete iterates the caller-supplied id +// slice and does CronShared.Remove(ss.services[id].CronJobID) without checking +// whether id is present in ss.services. CheckPermission returns vacuously +// true for unknown ids, so the controller layer cannot block this path. +// ss.services[unknownID] returns nil, and .CronJobID panics. Because the +// panic aborts the loop, every id ordered AFTER the bogus one is never removed +// from the in-memory registry even though its database row was already deleted, +// producing zombie services that keep dispatching cron probes. +func TestServiceSentinelDeleteWithUnknownIDDoesNotLeaveZombies(t *testing.T) { + // Given: one legitimate service (ID 10) registered in the sentinel. + ss := newServiceMonitorSecurityHarness(t, + &model.Server{Common: model.Common{ID: 1, UserID: 1}, Name: "reporter"}, + ) + service := &model.Service{ + Common: model.Common{ID: 10, UserID: 1}, + Name: "real-service", + Type: model.TaskTypeTCPPing, + Target: "example.invalid:443", + Duration: 3600, + Cover: model.ServiceCoverIgnoreAll, + SkipServers: map[uint64]bool{1: true}, + } + addServiceMonitorSecurityService(t, ss, service) + + // When: Delete is called with a bogus ID first, then the real service ID. + // Before the fix this panicked on ss.services[99999].CronJobID and left + // service 10 as a zombie. + ss.Delete([]uint64{99999, service.ID}) + + // Then: the real service must be fully removed from every in-memory map. + ss.serviceResponseDataStoreLock.RLock() + _, todayPresent := ss.serviceStatusToday[service.ID] + _, pingPresent := ss.serviceResponsePing[service.ID] + ss.serviceResponseDataStoreLock.RUnlock() + + ss.servicesLock.RLock() + _, servicePresent := ss.services[service.ID] + ss.servicesLock.RUnlock() + + ss.monthlyStatusLock.Lock() + _, monthlyPresent := ss.monthlyStatus[service.ID] + ss.monthlyStatusLock.Unlock() + + if todayPresent { + t.Error("zombie: serviceStatusToday still contains the deleted service") + } + if pingPresent { + t.Error("zombie: serviceResponsePing still contains the deleted service") + } + if servicePresent { + t.Error("zombie: services map still contains the deleted service") + } + if monthlyPresent { + t.Error("zombie: monthlyStatus still contains the deleted service") + } + + if _, err := fmt.Fprintln(os.Stdout, deleteUnknownIDSuccessMarker); err != nil { + t.Fatal(err) + } +} diff --git a/service/singleton/testhelpers.go b/service/singleton/testhelpers.go index eb60b9c6..8e889a27 100644 --- a/service/singleton/testhelpers.go +++ b/service/singleton/testhelpers.go @@ -20,6 +20,9 @@ func NewEmptyServerClassForTest() *ServerClass { // InsertForTest 把一个 server 直接塞进内存表与排序快照，跳过 DB & InitServer 逻辑。 // 调用方需保证 server.ID 已经设置。 func (c *ServerClass) InsertForTest(s *model.Server) { + c.lockLifecycleWrite() + defer c.unlockLifecycleWrite() + c.listMu.Lock() c.list[s.ID] = s if s.UUID != "" {