feat(build): kaniko/trivy 与扫描库改用 registry 内的 mirror 副本，定时刷新并在过期时告警

cmd/felis/api.go

@@ -466,22 +466,18 @@ func buildConfig(cfg *config.Config) build.Config {
 	return build.Config{
 		Namespace:   cfg.Registry.BuildNamespace,
 		RegistryURL: cfg.Registry.URL,
-		// Empty overrides fall back to the build package's defaults, so an
-		// install that has not imported kaniko/trivy keeps the compiled-in refs
-		// (and fails loudly on pull rather than silently building with the wrong
-		// image).
+		// Empty overrides fall back to the registry's copies of the tools
+		// (build.Tools), which felis mirror-build-tools keeps current.
 		KanikoImage: cfg.Registry.KanikoImage,
 		TrivyImage:  cfg.Registry.TrivyImage,
 		CPULimit:    cfg.Registry.BuildCPULimit,
 		MemLimit:    cfg.Registry.BuildMemLimit,
 		DiskLimit:   cfg.Registry.BuildDiskLimit,
 		// "auto" follows the startup probe (see probeBuildUserNamespaces).
-		UserNamespaces:      cfg.Registry.BuildUserNamespaces,
-		UserNamespacesProbe: new(atomic.Bool),
-		RuntimeClass:        cfg.Registry.BuildRuntimeClass,
-		MaxConcurrent:       cfg.Registry.MaxConcurrentBuilds,
-		// Empty keeps Trivy's own default; an install with builds points this at
-		// the internal DB mirror (see config.RegistryConfig.TrivyDBRepository).
+		UserNamespaces:        cfg.Registry.BuildUserNamespaces,
+		UserNamespacesProbe:   new(atomic.Bool),
+		RuntimeClass:          cfg.Registry.BuildRuntimeClass,
+		MaxConcurrent:         cfg.Registry.MaxConcurrentBuilds,
 		TrivyDBRepository:     cfg.Registry.TrivyDBRepository,
 		TrivyJavaDBRepository: cfg.Registry.TrivyJavaDBRepository,
 		// The submit lane's derived context URLs live here; the fetch step's
@@ -669,11 +665,7 @@ func registryPruner(cfg *config.Config, store imageRefStore, servers serverListe
 		fmt.Fprintf(stderr, "felis api: registry pruner disabled (%s unset) — images nothing uses are never deleted from the registry\n", platform.RegistryPruneTokenEnv)
 		return nil
 	}
-	static := []string{
-		os.Getenv("FELIS_IMAGE"),
-		cfg.Registry.KanikoImage, cfg.Registry.TrivyImage,
-		cfg.Registry.TrivyDBRepository, cfg.Registry.TrivyJavaDBRepository,
-	}
+	static := append([]string{os.Getenv("FELIS_IMAGE")}, buildConfig(cfg).ToolRefs()...)
 	return &registryprune.Pruner{
 		Registry: &registryprune.Client{Endpoint: "http://" + cfg.Registry.URL, Token: token},
 		Host:     cfg.Registry.URL,

cmd/felis/mirrortools.go

@@ -0,0 +1,125 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"os"
+	"os/signal"
+	"strings"
+	"syscall"
+	"time"
+
+	"felis.lolicon.best/internal/build"
+	"felis.lolicon.best/internal/imagepush"
+	"felis.lolicon.best/internal/registrygate"
+)
+
+// defaultBuildToolsStatus is where mirror-build-tools records its last run; the
+// watchdog reads it to tell a vulnerability DB that stopped refreshing.
+const defaultBuildToolsStatus = "/var/lib/felis/build-tools/status.json"
+
+// cmdMirrorBuildTools copies the build lane's tools (build.Tools: the kaniko and
+// trivy images, Trivy's vulnerability and Java DBs) from upstream into the
+// platform registry, where build Jobs pull them. deploy/bootstrap.sh runs it at
+// install and from felis-build-tools.timer twice a day, which is what keeps the
+// DBs fresh; a root shell can run it the same way to refresh now.
+//
+// It writes as the platform principal through the node's loopback hostPort, the
+// same way the installer pushes, reading the token from the environment or from
+// /etc/felis/secrets.env.
+func cmdMirrorBuildTools(args []string, stdout, stderr io.Writer) int {
+	fs := flag.NewFlagSet("mirror-build-tools", flag.ContinueOnError)
+	fs.SetOutput(stderr)
+	endpoint := fs.String("endpoint", "127.0.0.1:5000", "host[:port] of the registry to write to (plain HTTP)")
+	only := fs.String("only", "", "comma-separated tool names to copy (default: all of "+toolNames()+")")
+	status := fs.String("status", defaultBuildToolsStatus, `file to record the run in ("" records nothing)`)
+	secrets := fs.String("secrets-env", "/etc/felis/secrets.env", "installer secrets file holding REGISTRY_PLATFORM_TOKEN, read when FELIS_REGISTRY_PASSWORD is unset")
+	platformFlag := fs.String("platform", "", "os/arch of the images to copy (default: this machine's)")
+	if err := fs.Parse(args); err != nil {
+		if errors.Is(err, flag.ErrHelp) {
+			return 0
+		}
+		return 2
+	}
+	tools, err := selectTools(*only)
+	if err != nil {
+		fmt.Fprintf(stderr, "felis mirror-build-tools: %v\n", err)
+		return 2
+	}
+	if err := loadEnvFile(*secrets); err != nil {
+		fmt.Fprintf(stderr, "felis mirror-build-tools: read %s: %v\n", *secrets, err)
+		return 1
+	}
+	user, pass := os.Getenv("FELIS_REGISTRY_USERNAME"), os.Getenv("FELIS_REGISTRY_PASSWORD")
+	if pass == "" {
+		user, pass = registrygate.PrincipalPlatform, os.Getenv("REGISTRY_PLATFORM_TOKEN")
+	}
+	if pass == "" {
+		fmt.Fprintln(stderr, "felis mirror-build-tools: no registry credential: set FELIS_REGISTRY_PASSWORD or run as root on the node (REGISTRY_PLATFORM_TOKEN in /etc/felis/secrets.env)")
+		return 2
+	}
+
+	ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
+	defer stop()
+	p := &imagepush.Pusher{Scheme: "http", Username: user, Password: pass, Log: stdout}
+	src := &imagepush.Source{Platform: *platformFlag}
+	started := time.Now()
+	var failed []string
+	for _, t := range tools {
+		dst := strings.TrimSuffix(*endpoint, "/") + "/" + t.Mirror
+		if _, err := p.Mirror(ctx, src, t.Source, dst); err != nil {
+			fmt.Fprintf(stderr, "felis mirror-build-tools: %s: %v\n", t.Name, err)
+			failed = append(failed, t.Name+": "+err.Error())
+		}
+	}
+	if *status != "" {
+		st, err := imagepush.ReadMirrorStatus(*status)
+		if err != nil || st == nil {
+			st = &imagepush.MirrorStatus{}
+		}
+		st.LastAttempt = started
+		st.LastError = strings.Join(failed, "; ")
+		if len(failed) == 0 {
+			st.LastSuccess = started
+		}
+		if err := imagepush.WriteMirrorStatus(*status, *st); err != nil {
+			fmt.Fprintf(stderr, "felis mirror-build-tools: record %s: %v\n", *status, err)
+		}
+	}
+	if len(failed) > 0 {
+		return 1
+	}
+	return 0
+}
+
+func toolNames() string {
+	var names []string
+	for _, t := range build.Tools {
+		names = append(names, t.Name)
+	}
+	return strings.Join(names, ",")
+}
+
+func selectTools(only string) ([]build.Tool, error) {
+	if only == "" {
+		return build.Tools, nil
+	}
+	var out []build.Tool
+	for _, name := range strings.Split(only, ",") {
+		name = strings.TrimSpace(name)
+		found := false
+		for _, t := range build.Tools {
+			if t.Name == name {
+				out = append(out, t)
+				found = true
+			}
+		}
+		if !found {
+			return nil, fmt.Errorf("unknown tool %q (known: %s)", name, toolNames())
+		}
+	}
+	return out, nil
+}

cmd/felis/offsite.go

@@ -90,10 +90,10 @@ type offsiteEnv struct {
 	key    []byte
 }
 
-// loadOffsiteEnvFile sets each KEY=VALUE of path that is not already in the
-// environment, so a root shell reaches the bucket the same way the unit does.
+// loadEnvFile sets each KEY=VALUE of path that is not already in the
+// environment, so a root shell runs a command the same way its unit does.
 // A missing file is not an error.
-func loadOffsiteEnvFile(path string) error {
+func loadEnvFile(path string) error {
 	if path == "" {
 		return nil
 	}
@@ -167,7 +167,7 @@ func resolveOffsite(c config.OffsiteConfig) (*offsiteEnv, error) {
 
 // loadOffsite loads felis.toml and the env file and resolves [offsite].
 func loadOffsite(cfgPath, envFile string) (*config.Config, *offsiteEnv, error) {
-	if err := loadOffsiteEnvFile(envFile); err != nil {
+	if err := loadEnvFile(envFile); err != nil {
 		return nil, nil, fmt.Errorf("read %s: %w", envFile, err)
 	}
 	cfg, err := config.Load(cfgPath)
@@ -454,7 +454,7 @@ func offsiteFetchDB(fs *flag.FlagSet, args []string, stdout, stderr io.Writer) i
 		fmt.Fprint(stderr, offsiteUsage)
 		return 2
 	}
-	if err := loadOffsiteEnvFile(*envFile); err != nil {
+	if err := loadEnvFile(*envFile); err != nil {
 		fmt.Fprintf(stderr, "felis offsite fetch-db: read %s: %v\n", *envFile, err)
 		return 1
 	}

cmd/felis/offsite_test.go

@@ -26,7 +26,7 @@ not a line
 	t.Setenv("FELIS_OFFSITE_ACCESS_KEY", "from-the-shell")
 	t.Setenv("FELIS_OFFSITE_SECRET_KEY", "")
 	t.Setenv("FELIS_OFFSITE_KEY", "")
-	if err := loadOffsiteEnvFile(path); err != nil {
+	if err := loadEnvFile(path); err != nil {
 		t.Fatal(err)
 	}
 	for k, want := range map[string]string{
@@ -38,7 +38,7 @@ not a line
 			t.Errorf("%s = %q, want %q", k, got, want)
 		}
 	}
-	if err := loadOffsiteEnvFile(filepath.Join(t.TempDir(), "absent")); err != nil {
+	if err := loadEnvFile(filepath.Join(t.TempDir(), "absent")); err != nil {
 		t.Errorf("a missing env file is not an error: %v", err)
 	}
 }

cmd/felis/run.go

@@ -24,6 +24,7 @@ Commands:
   egress-gate       Hold a build pod until its egress NetworkPolicy is enforced (internal Job entrypoint)
   fetch-context     Fetch and extract a submission's build context (internal Job entrypoint)
   push-image        Push a scanned image tarball to the registry (internal Job entrypoint)
+  mirror-build-tools Copy kaniko, trivy and Trivy's DBs into the registry (run by felis-build-tools.timer)
   registry-gate     Authorize registry writes in front of registry:2 (internal sidecar entrypoint)
   manifests         Render the control-plane RBAC + NetworkPolicy install bundle as YAML
   apply             Create a MinecraftServer CRD (direct K8s write; use -f server.json)
@@ -47,32 +48,33 @@ Run "felis <command> -h" for command-specific flags.
 // The help aliases are deliberately NOT entries: they print usage rather than run a
 // subcommand, and listing them would make the table disagree with the command list.
 var commands = map[string]func(args []string, stdout, stderr io.Writer) int{
-	"migrate":          cmdMigrate,
-	"db":               cmdDB,
-	"offsite":          cmdOffsite,
-	"operator":         cmdOperator,
-	"api":              cmdAPI,
-	"nano":             cmdNano,
-	"reaper":           cmdReaper,
-	"restore":          cmdRestore,
-	"backup":           cmdBackup,
-	"files":            cmdFiles,
-	"egress-gate":      cmdEgressGate,
-	"fetch-context":    cmdFetchContext,
-	"push-image":       cmdPushImage,
-	"registry-gate":    cmdRegistryGate,
-	"manifests":        cmdManifests,
-	"apply":            cmdApply,
-	"setup":            cmdSetup,
-	"converge":         cmdConverge,
-	"breakGlass":       cmdBreakGlass,
-	"bootstrap-assets": cmdBootstrapAssets,
-	"init-forwarding":  cmdInitForwarding,
-	"init-volume":      cmdInitVolume,
-	"pin-images":       cmdPinImages,
-	"version":          cmdVersion,
-	"update":           cmdUpdate,
-	"watchdog":         cmdWatchdog,
+	"migrate":            cmdMigrate,
+	"db":                 cmdDB,
+	"offsite":            cmdOffsite,
+	"operator":           cmdOperator,
+	"api":                cmdAPI,
+	"nano":               cmdNano,
+	"reaper":             cmdReaper,
+	"restore":            cmdRestore,
+	"backup":             cmdBackup,
+	"files":              cmdFiles,
+	"egress-gate":        cmdEgressGate,
+	"fetch-context":      cmdFetchContext,
+	"push-image":         cmdPushImage,
+	"mirror-build-tools": cmdMirrorBuildTools,
+	"registry-gate":      cmdRegistryGate,
+	"manifests":          cmdManifests,
+	"apply":              cmdApply,
+	"setup":              cmdSetup,
+	"converge":           cmdConverge,
+	"breakGlass":         cmdBreakGlass,
+	"bootstrap-assets":   cmdBootstrapAssets,
+	"init-forwarding":    cmdInitForwarding,
+	"init-volume":        cmdInitVolume,
+	"pin-images":         cmdPinImages,
+	"version":            cmdVersion,
+	"update":             cmdUpdate,
+	"watchdog":           cmdWatchdog,
 }
 
 // run dispatches a subcommand. It is separate from main so the router is

cmd/felis/watchdog.go

@@ -42,6 +42,7 @@ func cmdWatchdog(args []string, stdout, stderr io.Writer) int {
 	proxyAddr := fs.String("proxy-addr", "", `game proxy address to dial, e.g. 127.0.0.1:25565 ("" skips the check)`)
 	controlNS := fs.String("control-namespace", platform.DefaultControlNamespace, "namespace of the control plane")
 	offsiteStatus := fs.String("offsite-status", offsite.DefaultStatusFile, "the record `felis offsite sync` leaves, checked when [offsite] is configured")
+	toolsStatus := fs.String("build-tools-status", defaultBuildToolsStatus, "the record `felis mirror-build-tools` leaves, checked when builds scan against the registry's DB copy")
 	dryRun := fs.Bool("dry-run", false, "print every finding and the mail that is due; send nothing and keep the state as it was")
 	if err := fs.Parse(args); err != nil {
 		if errors.Is(err, flag.ErrHelp) {
@@ -107,6 +108,9 @@ func cmdWatchdog(args []string, stdout, stderr io.Writer) int {
 	if cfg.Offsite.Enabled() {
 		add(watchdog.OffsiteFinding(*offsiteStatus, now))
 	}
+	if usesMirroredScanDB(cfg) {
+		add(watchdog.ScanDBFinding(*toolsStatus, now))
+	}
 	report.Findings = append(report.Findings, watchdog.DiskFindings(splitList(*diskPaths))...)
 	add(watchdog.MemoryFinding("/proc/meminfo"))
 
@@ -161,6 +165,17 @@ func cmdWatchdog(args []string, stdout, stderr io.Writer) int {
 	return save()
 }
 
+// usesMirroredScanDB reports whether build scans read the vulnerability DB copy
+// felis mirror-build-tools keeps in the platform registry: the default, or an
+// explicit trivy_db_repository under the registry's mirror/.
+func usesMirroredScanDB(cfg *config.Config) bool {
+	if cfg.Registry.URL == "" {
+		return false
+	}
+	repo := cfg.Registry.TrivyDBRepository
+	return repo == "" || strings.HasPrefix(repo, cfg.Registry.URL+"/mirror/")
+}
+
 // refreshSMTPPassword caches the relay password from the felis-smtp Secret, or
 // forgets it when the Secret is gone (a relay without AUTH). An env var named by
 // [smtp] password_ref, when set, wins at send time instead.

deploy/bootstrap.sh

@@ -279,6 +279,9 @@ WATCHDOG_STATE="/var/lib/felis/watchdog/state.json"
 OFFSITE_ENV="${STATE_DIR}/offsite.env"
 OFFSITE_SERVICE="/etc/systemd/system/felis-offsite.service"
 OFFSITE_TIMER="/etc/systemd/system/felis-offsite.timer"
+BUILD_TOOLS_SERVICE="/etc/systemd/system/felis-build-tools.service"
+BUILD_TOOLS_TIMER="/etc/systemd/system/felis-build-tools.timer"
+BUILD_TOOLS_STATUS="/var/lib/felis/build-tools/status.json"
 # While this marker holds a future Unix time, felis watchdog mails nothing: an install
 # restarts the control plane and the system servers on purpose. cleanup removes it; the
 # time in it is the backstop for an installer killed before its EXIT trap runs.
@@ -2679,6 +2682,47 @@ EOF
   fi
 }
 
+# The build lane's tools: kaniko and trivy (pinned by digest in internal/build/tools.go)
+# and Trivy's vulnerability and Java DBs, copied into the registry's mirror/ where build
+# Jobs pull them; the build namespace has no internet egress. The timer refreshes the DBs
+# twice a day (upstream publishes every six hours) and the watchdog warns when three days
+# pass without a clean run. The first copy starts now in the background: the Java DB
+# alone is several hundred MB.
+install_build_tools_timer() {
+  install -d -m 0755 "$(dirname "$BUILD_TOOLS_STATUS")"
+  cat > "$BUILD_TOOLS_SERVICE" <<EOF
+[Unit]
+Description=Copy the Felis build tools and Trivy's vulnerability DBs into the registry
+After=network-online.target k3s.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+ExecStart=${HOST_BIN} mirror-build-tools -endpoint ${REGISTRY_PUSH_HOST} -status ${BUILD_TOOLS_STATUS} -secrets-env ${SECRETS_ENV}
+TimeoutStartSec=1h
+Nice=10
+PrivateTmp=yes
+NoNewPrivileges=yes
+ProtectSystem=full
+EOF
+  cat > "$BUILD_TOOLS_TIMER" <<EOF
+[Unit]
+Description=Refresh the Felis build tools and Trivy DBs twice a day
+
+[Timer]
+OnCalendar=*-*-* 04,16:00:00
+RandomizedDelaySec=1h
+Persistent=true
+
+[Install]
+WantedBy=timers.target
+EOF
+  systemctl daemon-reload
+  systemctl enable --now felis-build-tools.timer
+  systemctl start --no-block felis-build-tools.service
+  ok "build tools: kaniko, trivy and the Trivy DBs are being copied into the registry (journalctl -u felis-build-tools); refreshed twice a day"
+}
+
 # The daily database backup. The first run happens now, so a broken pipeline (pg_dump
 # missing, directory unwritable) shows up in this install rather than in the first
 # restore someone needs.
@@ -3426,6 +3470,8 @@ main() {
   # AFTER deploy_bundle: the registry the built images are mirrored into is part
   # of that bundle.
   push_images_to_registry
+  # After deploy_bundle, like the pushes: it writes into the registry.
+  install_build_tools_timer
   restart_existing_system_servers
   # After deploy_bundle: the proxy dials felis-api's internal ClusterIP, which does not
   # exist until the bundle is applied.

internal/build/build.go

@@ -234,16 +234,15 @@ type Config struct {
 	// shape); an install that never builds user submissions can leave it empty.
 	FelisImage string
 	// TrivyDBRepository overrides where Trivy fetches its vulnerability DB
-	// (--db-repository). Empty keeps Trivy's upstream default, which the build
-	// egress lock denies — an install with builds must point this at an internal
-	// mirror (see config.RegistryConfig.TrivyDBRepository).
+	// (--db-repository). Empty means the platform registry's copy (Tools); with no
+	// RegistryURL it stays empty, which keeps Trivy's upstream default.
 	TrivyDBRepository string
 	// TrivyJavaDBRepository overrides where Trivy fetches its Java DB
 	// (--java-db-repository), downloaded lazily for images that contain Java
-	// artifacts — i.e. every real modpack. Same egress story as the
-	// vulnerability DB (see config.RegistryConfig.TrivyJavaDBRepository).
+	// artifacts — i.e. every real modpack. Defaults like TrivyDBRepository.
 	TrivyJavaDBRepository string
-	// KanikoImage / TrivyImage are the executor images.
+	// KanikoImage / TrivyImage are the executor images. Empty means the platform
+	// registry's copies (Tools).
 	KanikoImage string
 	TrivyImage  string
 	// Deadline caps a build's wall-clock (spec §16: activeDeadlineSeconds).
@@ -301,8 +300,6 @@ func (c Config) userNamespaces() bool {
 const (
 	defaultNamespace      = "felis-build"
 	defaultServiceAccount = "felis-build"
-	defaultKanikoImage    = "gcr.io/kaniko-project/executor:latest"
-	defaultTrivyImage     = "aquasec/trivy:latest"
 	defaultDeadline       = 30 * time.Minute
 	defaultMaxDockerfile  = 256 * 1024 // 256 KiB
 	defaultCPULimit       = "2"
@@ -325,11 +322,19 @@ func (c Config) withDefaults() Config {
 	if c.ServiceAccount == "" {
 		c.ServiceAccount = defaultServiceAccount
 	}
+	// The executor images and the scan DBs default to the platform registry's
+	// copies (Tools); an explicit felis.toml value wins.
 	if c.KanikoImage == "" {
-		c.KanikoImage = defaultKanikoImage
+		c.KanikoImage = toolRef(c.RegistryURL, "kaniko")
 	}
 	if c.TrivyImage == "" {
-		c.TrivyImage = defaultTrivyImage
+		c.TrivyImage = toolRef(c.RegistryURL, "trivy")
+	}
+	if c.TrivyDBRepository == "" && c.RegistryURL != "" {
+		c.TrivyDBRepository = toolRef(c.RegistryURL, "trivy-db")
+	}
+	if c.TrivyJavaDBRepository == "" && c.RegistryURL != "" {
+		c.TrivyJavaDBRepository = toolRef(c.RegistryURL, "trivy-java-db")
 	}
 	if c.Deadline <= 0 {
 		c.Deadline = defaultDeadline

internal/build/jobspec_test.go

@@ -18,8 +18,8 @@ func sampleJobParams() JobParams {
 		ServiceAccount: defaultServiceAccount,
 		RegistryURL:    "registry.felis.svc:5000",
 		FelisImage:     "felis:test",
-		KanikoImage:    defaultKanikoImage,
-		TrivyImage:     defaultTrivyImage,
+		KanikoImage:    toolRef("", "kaniko"),
+		TrivyImage:     toolRef("", "trivy"),
 		Deadline:       30 * time.Minute,
 		CPULimit:       "2",
 		MemLimit:       "4Gi",

internal/build/tools.go

@@ -0,0 +1,59 @@
+package build
+
+import "strings"
+
+// Tool is an image or OCI artifact a build Job runs or reads: where it comes
+// from upstream, and the repository:tag the platform registry keeps its copy
+// under. A build pulls only the copy. The build namespace has no internet egress,
+// so Trivy cannot reach the upstream DBs, and the node pulls the executor images
+// from the in-cluster registry like any other image, which also survives an image
+// GC. `felis mirror-build-tools` (deploy/bootstrap.sh runs it at install and from
+// felis-build-tools.timer twice a day) copies each Source to its Mirror.
+type Tool struct {
+	Name string
+	// Source is the upstream reference. The executor images are pinned by the
+	// digest of their multi-platform index, so a moved or re-pushed upstream tag
+	// cannot change what a build runs; the DBs follow their tag, since a fresh
+	// vulnerability DB is the point of refreshing them.
+	Source string
+	// Mirror is repository:tag inside the platform registry.
+	Mirror string
+}
+
+// Tools lists every tool a build needs. Kaniko is archived upstream (June 2025)
+// and v1.24.0 is its final release; docs/troubleshooting.md §8e covers moving to a
+// maintained fork.
+var Tools = []Tool{
+	{Name: "kaniko", Source: "gcr.io/kaniko-project/executor:v1.24.0@sha256:4e7a52dd1f14872430652bb3b027405b8dfd17c4538751c620ac005741ef9698", Mirror: "mirror/kaniko-executor:v1.24.0"},
+	{Name: "trivy", Source: "ghcr.io/aquasecurity/trivy:0.74.0@sha256:62b1e65e8869bc4b4c6aa4fa2b21595256c7c2f6018a9d9ad61caf87187c1969", Mirror: "mirror/trivy:0.74.0"},
+	{Name: "trivy-db", Source: "mirror.gcr.io/aquasec/trivy-db:2", Mirror: "mirror/trivy-db:2"},
+	{Name: "trivy-java-db", Source: "mirror.gcr.io/aquasec/trivy-java-db:1", Mirror: "mirror/trivy-java-db:1"},
+}
+
+// tool returns the named entry of Tools.
+func tool(name string) Tool {
+	for _, t := range Tools {
+		if t.Name == name {
+			return t
+		}
+	}
+	panic("build: unknown tool " + name)
+}
+
+// toolRef is where a build reads tool name: its copy in registry, or the upstream
+// source when there is no platform registry (tests, a bare Config).
+func toolRef(registry, name string) string {
+	t := tool(name)
+	if registry == "" {
+		return t.Source
+	}
+	return strings.TrimSuffix(registry, "/") + "/" + t.Mirror
+}
+
+// ToolRefs returns the four references a build of this Config reads, after
+// defaults: the kaniko and trivy images and the two Trivy DB repositories. The
+// registry pruner keeps each of them.
+func (c Config) ToolRefs() []string {
+	c = c.withDefaults()
+	return []string{c.KanikoImage, c.TrivyImage, c.TrivyDBRepository, c.TrivyJavaDBRepository}
+}

internal/build/tools_test.go

@@ -0,0 +1,46 @@
+package build
+
+import (
+	"strings"
+	"testing"
+)
+
+// A build reads every tool from the platform registry's copy by default, and
+// each copy's repository lives under mirror/, which only the platform principal
+// may write and the pruner keeps.
+func TestToolRefsDefaultToTheRegistryCopies(t *testing.T) {
+	got := Config{RegistryURL: "registry.felis.svc:5000"}.ToolRefs()
+	want := []string{
+		"registry.felis.svc:5000/mirror/kaniko-executor:v1.24.0",
+		"registry.felis.svc:5000/mirror/trivy:0.74.0",
+		"registry.felis.svc:5000/mirror/trivy-db:2",
+		"registry.felis.svc:5000/mirror/trivy-java-db:1",
+	}
+	if strings.Join(got, " ") != strings.Join(want, " ") {
+		t.Errorf("ToolRefs = %v, want %v", got, want)
+	}
+
+	explicit := Config{RegistryURL: "r:5000", KanikoImage: "r:5000/mirror/kaniko-fork:v2", TrivyDBRepository: "r:5000/mirror/db:9"}.ToolRefs()
+	if explicit[0] != "r:5000/mirror/kaniko-fork:v2" || explicit[2] != "r:5000/mirror/db:9" {
+		t.Errorf("explicit values lost: %v", explicit)
+	}
+
+	// Without a registry the images are the pinned upstream sources and Trivy
+	// keeps its own DB default.
+	bare := Config{}.ToolRefs()
+	if !strings.Contains(bare[0], "@sha256:") || !strings.Contains(bare[1], "@sha256:") || bare[2] != "" || bare[3] != "" {
+		t.Errorf("bare ToolRefs = %v", bare)
+	}
+}
+
+func TestToolsArePinnedAndMirroredUnderMirror(t *testing.T) {
+	for _, tl := range Tools {
+		if !strings.HasPrefix(tl.Mirror, "mirror/") || !strings.Contains(tl.Mirror, ":") {
+			t.Errorf("%s: mirror %q must be mirror/<repo>:<tag>", tl.Name, tl.Mirror)
+		}
+		isDB := strings.HasSuffix(tl.Name, "-db")
+		if pinned := strings.Contains(tl.Source, "@sha256:"); pinned == isDB {
+			t.Errorf("%s: source %q: executor images must be pinned by digest, DBs follow their tag", tl.Name, tl.Source)
+		}
+	}
+}

internal/config/config.go

@@ -148,13 +148,10 @@ type RegistryConfig struct {
 	URL            string `toml:"url"`
 	BuildNamespace string `toml:"build_namespace"`
 	// KanikoImage / TrivyImage / BuildCPULimit / BuildMemLimit override the
-	// build subsystem's compiled-in defaults (gcr.io/kaniko-project/executor and
-	// aquasec/trivy, 2 CPU / 4Gi per build container). The kubelet pulls the
-	// executor images over the node's network, so the defaults need a node that
-	// can reach those registries; an air-gapped or mirrored install points these
-	// at images mirrored into the in-cluster registry (docs/troubleshooting.md
-	// §8e). A bare node-containerd import does not survive an image GC, there is
-	// no pull source for it. Empty keeps the default.
+	// build subsystem's defaults: the kaniko and trivy copies the installer keeps
+	// in this registry under mirror/ (build.Tools), 2 CPU / 4Gi per build
+	// container. Set the images only to run another build of the tools
+	// (docs/troubleshooting.md §8e). Empty keeps the default.
 	KanikoImage   string `toml:"kaniko_image"`
 	TrivyImage    string `toml:"trivy_image"`
 	BuildCPULimit string `toml:"build_cpu_limit"`
@@ -173,25 +170,18 @@ type RegistryConfig struct {
 	// Zero keeps 2; at most 6 (the build namespace's pod quota).
 	MaxConcurrentBuilds int `toml:"max_concurrent_builds"`
 	// TrivyDBRepository points Trivy at an OCI repository holding the
-	// vulnerability DB (--db-repository). Trivy's default fetches from
-	// mirror.gcr.io/ghcr.io, which the build egress lock denies — so on a
-	// default install the scan step fails closed and no build ever completes.
-	// The supported shape is an internal mirror: copy
-	// mirror.gcr.io/aquasec/trivy-db:2 into this cluster's registry (recipe in
-	// docs/troubleshooting.md §8) and set this to
-	// registry.<ns>.svc:5000/mirror/trivy-db:2. The scan runs with --insecure,
-	// so the plain-HTTP internal registry works. Empty keeps Trivy's own
-	// default (only usable on an install that deliberately opens internet
-	// egress to the DB hosts).
+	// vulnerability DB (--db-repository). Trivy's own default fetches from
+	// mirror.gcr.io/ghcr.io, which the build egress lock denies, so the default
+	// here is the copy felis-build-tools.timer refreshes in this registry,
+	// <url>/mirror/trivy-db:2 (build.Tools). The scan runs with --insecure, so
+	// the plain-HTTP internal registry works. Empty keeps that default.
 	TrivyDBRepository string `toml:"trivy_db_repository"`
 	// TrivyJavaDBRepository points Trivy at an OCI repository holding the Java
 	// DB (--java-db-repository). Trivy fetches it lazily whenever the scanned
 	// image contains Java artifacts — every real modpack image does — so on an
-	// egress-locked box the scan fails closed without this mirror exactly like
-	// the vulnerability DB. The supported shape is an internal mirror: copy
-	// mirror.gcr.io/aquasec/trivy-java-db:1 into this cluster's registry and
-	// set this to registry.<ns>.svc:5000/mirror/trivy-java-db:1 (recipe in
-	// docs/troubleshooting.md §8e). Empty keeps Trivy's own default.
+	// egress-locked box it comes from this registry exactly like the
+	// vulnerability DB: <url>/mirror/trivy-java-db:1 by default. Empty keeps that
+	// default.
 	TrivyJavaDBRepository string `toml:"trivy_java_db_repository"`
 	// UserUploadsContext is the object-store base under which a user-submitted
 	// modpack's Kaniko build context is pinned. It belongs to the §16 build

internal/watchdog/probes.go

@@ -12,6 +12,7 @@ import (
 
 	"felis.lolicon.best/internal/apis/felis/v1alpha1"
 	"felis.lolicon.best/internal/dbbackup"
+	"felis.lolicon.best/internal/imagepush"
 	"felis.lolicon.best/internal/naming"
 	"felis.lolicon.best/internal/offsite"
 	"felis.lolicon.best/internal/platform"
@@ -45,6 +46,11 @@ const (
 	maxBackupAge = 26 * time.Hour
 	// maxReaperAge is the same for the daily reaper CronJob.
 	maxReaperAge = 26 * time.Hour
+	// maxScanDBAge is how old the registry's copy of Trivy's vulnerability DB may
+	// grow. felis-build-tools.timer refreshes it twice a day and upstream publishes
+	// every six hours; three days of failed refreshes means scans are passing
+	// images against advisories that are no longer current.
+	maxScanDBAge = 72 * time.Hour
 
 	diskLowRatio      = 0.15
 	diskCriticalRatio = 0.05
@@ -371,6 +377,42 @@ func OffsiteFinding(statusFile string, now time.Time) *Finding {
 	return f
 }
 
+// ScanDBFinding reports the build lane's tool copies (the vulnerability DBs in
+// particular) not refreshed within maxScanDBAge, going by the record
+// `felis mirror-build-tools` leaves in statusFile. A build still runs and its scan
+// still gates, but against an old DB: a warning.
+func ScanDBFinding(statusFile string, now time.Time) *Finding {
+	const hint = "journalctl -u felis-build-tools -n 50; refresh now with `sudo felis mirror-build-tools` (docs/troubleshooting.md §8e)"
+	st, err := imagepush.ReadMirrorStatus(statusFile)
+	if err != nil {
+		return &Finding{
+			Key: "scan-db", Severity: Warning, For: backupFor,
+			Summary:   fmt.Sprintf("无法读取构建工具镜像状态 %s", statusFile),
+			SummaryEN: fmt.Sprintf("cannot read the build tools status %s: %v", statusFile, err),
+			Hint:      hint,
+		}
+	}
+	if st != nil && !st.LastSuccess.IsZero() && now.Sub(st.LastSuccess) <= maxScanDBAge {
+		return nil
+	}
+	f := &Finding{
+		Key: "scan-db", Severity: Warning, For: backupFor,
+		Summary:   "漏洞库从未复制进内置 registry，构建的漏洞扫描无法运行",
+		SummaryEN: "the vulnerability DB was never copied into the registry; build scans cannot run",
+		Hint:      hint,
+	}
+	if st != nil && !st.LastSuccess.IsZero() {
+		age := roundHours(now.Sub(st.LastSuccess))
+		f.Summary = fmt.Sprintf("漏洞库已有 %s 没有刷新，构建扫描用的是过期数据", age)
+		f.SummaryEN = fmt.Sprintf("the vulnerability DB was last refreshed %s ago; build scans use stale advisories", age)
+	}
+	if st != nil && st.LastError != "" {
+		f.Summary += "（最近一次错误：" + st.LastError + "）"
+		f.SummaryEN += " (last error: " + st.LastError + ")"
+	}
+	return f
+}
+
 // DiskFindings reports each filesystem under paths that is running out of
 // space. Paths on one filesystem are reported once, under the first of them; a
 // path that does not exist is skipped (a feature that is not in use).

internal/watchdog/probes_test.go

@@ -12,6 +12,7 @@ import (
 
 	"felis.lolicon.best/internal/apis/felis/v1alpha1"
 	"felis.lolicon.best/internal/dbbackup"
+	"felis.lolicon.best/internal/imagepush"
 	"felis.lolicon.best/internal/offsite"
 	appsv1 "k8s.io/api/apps/v1"
 	batchv1 "k8s.io/api/batch/v1"
@@ -210,3 +211,24 @@ func TestDiskFindingsDedupAndSkip(t *testing.T) {
 		t.Fatalf("findings = %v, want at most one for one filesystem", findingKeys(got))
 	}
 }
+
+func TestScanDBFinding(t *testing.T) {
+	now := time.Date(2026, 9, 24, 12, 0, 0, 0, time.UTC)
+	path := filepath.Join(t.TempDir(), "status.json")
+	if f := ScanDBFinding(path, now); f == nil || !strings.Contains(f.SummaryEN, "never copied") {
+		t.Errorf("no status file: %+v", f)
+	}
+	if err := imagepush.WriteMirrorStatus(path, imagepush.MirrorStatus{LastAttempt: now, LastSuccess: now.Add(-24 * time.Hour)}); err != nil {
+		t.Fatal(err)
+	}
+	if f := ScanDBFinding(path, now); f != nil {
+		t.Errorf("a day-old DB: %+v", f)
+	}
+	if err := imagepush.WriteMirrorStatus(path, imagepush.MirrorStatus{LastAttempt: now, LastSuccess: now.Add(-100 * time.Hour), LastError: "trivy-db: dial tcp: timeout"}); err != nil {
+		t.Fatal(err)
+	}
+	f := ScanDBFinding(path, now)
+	if f == nil || f.Severity != Warning || !strings.Contains(f.SummaryEN, "100h") || !strings.Contains(f.SummaryEN, "dial tcp") {
+		t.Errorf("stale DB: %+v", f)
+	}
+}