1// Package app provides the application initialization and wiring.
2package app
3
4import (
5 "context"
6 "crypto/ecdsa"
7 "crypto/elliptic"
8 "crypto/rand"
9 "crypto/x509"
10 "crypto/x509/pkix"
11 "encoding/hex"
12 "encoding/json"
13 "encoding/pem"
14 "errors"
15 "fmt"
16 "math/big"
17 "net"
18 "net/http"
19 "os"
20 "os/signal"
21 "path/filepath"
22 "strconv"
23 "strings"
24 "syscall"
25 "time"
26
27 "github.com/bnema/zerowrap"
28 zerowrapotel "github.com/bnema/zerowrap/otel"
29 "github.com/fsnotify/fsnotify"
30 "github.com/spf13/viper"
31
32 // Adapters - Output
33 "github.com/bnema/gordon/internal/adapters/out/docker"
34 "github.com/bnema/gordon/internal/adapters/out/domainsecrets"
35 "github.com/bnema/gordon/internal/adapters/out/envloader"
36 "github.com/bnema/gordon/internal/adapters/out/eventbus"
37 "github.com/bnema/gordon/internal/adapters/out/filesystem"
38 "github.com/bnema/gordon/internal/adapters/out/httpprober"
39 "github.com/bnema/gordon/internal/adapters/out/logwriter"
40 "github.com/bnema/gordon/internal/adapters/out/ratelimit"
41 "github.com/bnema/gordon/internal/adapters/out/secrets"
42 "github.com/bnema/gordon/internal/adapters/out/telemetry"
43 "github.com/bnema/gordon/internal/adapters/out/tokenstore"
44
45 // OTel
46 "go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp"
47
48 // Adapters - Input
49 "github.com/bnema/gordon/internal/adapters/dto"
50 "github.com/bnema/gordon/internal/adapters/in/http/admin"
51 authhandler "github.com/bnema/gordon/internal/adapters/in/http/auth"
52 "github.com/bnema/gordon/internal/adapters/in/http/middleware"
53 proxyadapter "github.com/bnema/gordon/internal/adapters/in/http/proxy"
54 "github.com/bnema/gordon/internal/adapters/in/http/registry"
55
56 // Boundaries
57 "github.com/bnema/gordon/internal/boundaries/out"
58
59 // Domain
60 "github.com/bnema/gordon/internal/domain"
61
62 // Packages
63 "github.com/bnema/gordon/pkg/version"
64
65 // Use cases
66 "github.com/bnema/gordon/internal/usecase/auth"
67 "github.com/bnema/gordon/internal/usecase/auto"
68 "github.com/bnema/gordon/internal/usecase/auto/preview"
69 "github.com/bnema/gordon/internal/usecase/backup"
70 "github.com/bnema/gordon/internal/usecase/config"
71 "github.com/bnema/gordon/internal/usecase/container"
72 cronSvc "github.com/bnema/gordon/internal/usecase/cron"
73 "github.com/bnema/gordon/internal/usecase/health"
74 "github.com/bnema/gordon/internal/usecase/images"
75 "github.com/bnema/gordon/internal/usecase/logs"
76 "github.com/bnema/gordon/internal/usecase/proxy"
77 registrySvc "github.com/bnema/gordon/internal/usecase/registry"
78 secretsSvc "github.com/bnema/gordon/internal/usecase/secrets"
79 volumesSvc "github.com/bnema/gordon/internal/usecase/volumes"
80
81 // Pkg
82 "github.com/bnema/gordon/pkg/bytesize"
83 "github.com/bnema/gordon/pkg/duration"
84)
85
86// Config holds the application configuration.
87type Config struct {
88 Server struct {
89 Port int `mapstructure:"port"`
90 RegistryPort int `mapstructure:"registry_port"`
91 GordonDomain string `mapstructure:"gordon_domain"`
92 RegistryDomain string `mapstructure:"registry_domain"` // Deprecated: use gordon_domain
93 TLSEnabled bool `mapstructure:"tls_enabled"`
94 TLSPort int `mapstructure:"tls_port"`
95 TLSCertFile string `mapstructure:"tls_cert_file"`
96 TLSKeyFile string `mapstructure:"tls_key_file"`
97 DataDir string `mapstructure:"data_dir"`
98 MaxProxyBodySize string `mapstructure:"max_proxy_body_size"` // e.g., "512MB", "1GB"
99 MaxBlobChunkSize string `mapstructure:"max_blob_chunk_size"` // e.g., "512MB", "1GB"
100 MaxProxyResponseSize string `mapstructure:"max_proxy_response_size"` // e.g., "1GB", "0" for no limit
101 MaxConcurrentConns int `mapstructure:"max_concurrent_connections"`
102 RegistryAllowedIPs []string `mapstructure:"registry_allowed_ips"`
103 } `mapstructure:"server"`
104
105 Logging struct {
106 Level string `mapstructure:"level"`
107 Format string `mapstructure:"format"`
108 File struct {
109 Enabled bool `mapstructure:"enabled"`
110 Path string `mapstructure:"path"`
111 MaxSize int `mapstructure:"max_size"`
112 MaxBackups int `mapstructure:"max_backups"`
113 MaxAge int `mapstructure:"max_age"`
114 } `mapstructure:"file"`
115 ContainerLogs struct {
116 Enabled bool `mapstructure:"enabled"`
117 Dir string `mapstructure:"dir"`
118 MaxSize int `mapstructure:"max_size"`
119 MaxBackups int `mapstructure:"max_backups"`
120 MaxAge int `mapstructure:"max_age"`
121 } `mapstructure:"container_logs"`
122 } `mapstructure:"logging"`
123
124 Env struct {
125 Dir string `mapstructure:"dir"`
126 } `mapstructure:"env"`
127
128 Auth struct {
129 Enabled bool `mapstructure:"enabled"`
130 Type string `mapstructure:"type"` // only "token" is supported
131 SecretsBackend string `mapstructure:"secrets_backend"` // "pass", "sops", or "unsafe"
132 Username string `mapstructure:"username"`
133 TokenSecret string `mapstructure:"token_secret"` // path in secrets backend
134 TokenExpiry string `mapstructure:"token_expiry"` // e.g., "720h", "30d"
135 AccessTokenTTL string `mapstructure:"access_token_ttl"` // e.g., "15m", "30m" (default: 15m)
136 } `mapstructure:"auth"`
137
138 API struct {
139 RateLimit struct {
140 Enabled bool `mapstructure:"enabled"`
141 GlobalRPS float64 `mapstructure:"global_rps"`
142 PerIPRPS float64 `mapstructure:"per_ip_rps"`
143 Burst int `mapstructure:"burst"`
144 TrustedProxies []string `mapstructure:"trusted_proxies"`
145 } `mapstructure:"rate_limit"`
146 } `mapstructure:"api"`
147
148 Backups struct {
149 Enabled bool `mapstructure:"enabled"`
150 Schedule string `mapstructure:"schedule"`
151 StorageDir string `mapstructure:"storage_dir"`
152 Retention struct {
153 Hourly int `mapstructure:"hourly"`
154 Daily int `mapstructure:"daily"`
155 Weekly int `mapstructure:"weekly"`
156 Monthly int `mapstructure:"monthly"`
157 } `mapstructure:"retention"`
158 } `mapstructure:"backups"`
159
160 Images struct {
161 Prune struct {
162 Enabled bool `mapstructure:"enabled"`
163 Schedule string `mapstructure:"schedule"`
164 KeepLast int `mapstructure:"keep_last"`
165 } `mapstructure:"prune"`
166 } `mapstructure:"images"`
167
168 Containers struct {
169 MemoryLimit string `mapstructure:"memory_limit"` // e.g., "512MB", "1GB"
170 CPULimit float64 `mapstructure:"cpu_limit"` // CPU cores, e.g., 1.0 = 1 core
171 PidsLimit int64 `mapstructure:"pids_limit"` // e.g., 512
172 } `mapstructure:"containers"`
173
174 Telemetry telemetry.Config `mapstructure:"telemetry"`
175}
176
177// services holds all the services used by the application.
178type services struct {
179 runtime *docker.Runtime
180 eventBus *eventbus.InMemory
181 blobStorage *filesystem.BlobStorage
182 manifestStorage *filesystem.ManifestStorage
183 backupStorage *filesystem.BackupStorage
184 envLoader out.EnvLoader
185 logWriter *logwriter.LogWriter
186 tokenStore out.TokenStore
187 configSvc *config.Service
188 secretSvc *secretsSvc.Service
189 containerSvc *container.Service
190 backupSvc *backup.Service
191 registrySvc *registrySvc.Service
192 healthSvc *health.Service
193 logSvc *logs.Service
194 imageSvc *images.Service
195 volumeSvc *volumesSvc.Service
196 proxySvc *proxy.Service
197 authSvc *auth.Service
198 authHandler *authhandler.Handler
199 adminHandler *admin.Handler
200 internalRegUser string
201 internalRegPass string
202 previewStore *filesystem.PreviewStore
203 previewService *preview.Service
204 envDir string
205 maxBlobChunkSize int64
206}
207
208// Run initializes and starts the Gordon application.
209func Run(ctx context.Context, configPath string) error {
210 // Load configuration
211 v, cfg, err := initConfig(configPath)
212 if err != nil {
213 return err
214 }
215
216 // Initialize logger
217 log, cleanup, err := initLogger(cfg)
218 if err != nil {
219 return err
220 }
221 if cleanup != nil {
222 defer cleanup()
223 }
224
225 ctx = zerowrap.WithCtx(ctx, log)
226
227 // Initialize OpenTelemetry
228 telProvider, telShutdown, err := telemetry.NewProvider(ctx, cfg.Telemetry, "gordon", version.Version())
229 if err != nil {
230 log.Warn().Err(err).Msg("failed to initialize telemetry, continuing without it")
231 } else {
232 // Use a fresh context for shutdown so a canceled app ctx doesn't prevent flushing.
233 defer func() {
234 shutdownCtx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
235 defer cancel()
236 telShutdown(shutdownCtx)
237 }()
238 if cfg.Telemetry.Enabled && cfg.Telemetry.Endpoint != "" {
239 // Bridge zerowrap logs to OTel if log export is enabled
240 if cfg.Telemetry.Logs && telProvider.LogProvider != nil {
241 otelHook := zerowrapotel.NewHookWithProvider(telProvider.LogProvider, "gordon")
242 log = zerowrap.WithHook(log, otelHook)
243 ctx = zerowrap.WithCtx(ctx, log)
244 }
245 log.Info().Str("endpoint", cfg.Telemetry.Endpoint).Msg("telemetry initialized")
246 }
247 }
248
249 log.Info().Msg("Gordon starting")
250
251 if err := ensureTLSConfig(&cfg, log); err != nil {
252 return err
253 }
254
255 // Create PID file
256 pidFile := createPidFile(log)
257 if pidFile != "" {
258 defer removePidFile(pidFile, log)
259 }
260
261 // Create all services
262 svc, err := createServices(ctx, v, cfg, log)
263 if err != nil {
264 return err
265 }
266
267 // Register event handlers
268 cleanupHandlers, err := registerEventHandlers(ctx, svc, cfg)
269 if err != nil {
270 return err
271 }
272 // cleanupHandlers is passed into runServers so it can stop debounce
273 // timers before graceful shutdown, preventing deploys during drain.
274
275 // Set up config hot reload
276 setupConfigHotReload(ctx, v, svc, log)
277
278 // Start event bus
279 if err := svc.eventBus.Start(); err != nil {
280 return log.WrapErr(err, "failed to start event bus")
281 }
282 defer svc.eventBus.Stop()
283
284 // Create HTTP handlers
285 registryHandler, proxyHandler := createHTTPHandlers(svc, cfg, log)
286
287 // Start servers, wait for listeners to bind, then sync/auto-start containers.
288 return runServers(ctx, v, cfg, registryHandler, proxyHandler, svc, cleanupHandlers, log)
289}
290
291func ensureTLSConfig(cfg *Config, log zerowrap.Logger) error {
292 if !cfg.Server.TLSEnabled {
293 return nil
294 }
295
296 if cfg.Server.TLSCertFile != "" && cfg.Server.TLSKeyFile != "" {
297 return nil
298 }
299
300 dataDir := resolveDataDir(cfg.Server.DataDir)
301 tlsDir := filepath.Join(dataDir, "tls")
302 certPath := filepath.Join(tlsDir, "cert.pem")
303 keyPath := filepath.Join(tlsDir, "key.pem")
304
305 domainName := cfg.Server.GordonDomain
306 if domainName == "" {
307 domainName = cfg.Server.RegistryDomain
308 }
309 if domainName == "" {
310 domainName = "localhost"
311 }
312
313 if err := generateSelfSignedTLSCert(certPath, keyPath, domainName); err != nil {
314 return fmt.Errorf("failed to generate self-signed TLS certificate: %w", err)
315 }
316
317 cfg.Server.TLSCertFile = certPath
318 cfg.Server.TLSKeyFile = keyPath
319
320 log.Warn().
321 Str("cert_file", certPath).
322 Str("key_file", keyPath).
323 Str("domain", domainName).
324 Msg("server.tls_enabled=true with empty tls_cert_file/tls_key_file; using auto-generated self-signed certificate")
325
326 return nil
327}
328
329func generateSelfSignedTLSCert(certPath, keyPath, domainName string) error {
330 // Reuse existing pair if already generated.
331 if _, err := os.Stat(certPath); err == nil {
332 if _, keyErr := os.Stat(keyPath); keyErr == nil {
333 return nil
334 }
335 }
336
337 if err := os.MkdirAll(filepath.Dir(certPath), 0700); err != nil {
338 return err
339 }
340 if err := os.MkdirAll(filepath.Dir(keyPath), 0700); err != nil {
341 return err
342 }
343
344 priv, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
345 if err != nil {
346 return err
347 }
348
349 now := time.Now()
350 serialLimit := new(big.Int).Lsh(big.NewInt(1), 128)
351 serialNumber, err := rand.Int(rand.Reader, serialLimit)
352 if err != nil {
353 return err
354 }
355
356 template := x509.Certificate{
357 SerialNumber: serialNumber,
358 Subject: pkix.Name{
359 CommonName: domainName,
360 Organization: []string{"Gordon Auto-Generated TLS"},
361 },
362 NotBefore: now.Add(-5 * time.Minute),
363 NotAfter: now.AddDate(1, 0, 0),
364 KeyUsage: x509.KeyUsageDigitalSignature | x509.KeyUsageKeyEncipherment,
365 ExtKeyUsage: []x509.ExtKeyUsage{x509.ExtKeyUsageServerAuth},
366 BasicConstraintsValid: true,
367 }
368
369 if ip := net.ParseIP(domainName); ip != nil {
370 template.IPAddresses = []net.IP{ip}
371 } else {
372 template.DNSNames = []string{domainName}
373 }
374
375 derBytes, err := x509.CreateCertificate(rand.Reader, &template, &template, &priv.PublicKey, priv)
376 if err != nil {
377 return err
378 }
379
380 certOut, err := os.OpenFile(certPath, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0600)
381 if err != nil {
382 return err
383 }
384 defer certOut.Close()
385
386 if err := pem.Encode(certOut, &pem.Block{Type: "CERTIFICATE", Bytes: derBytes}); err != nil {
387 return err
388 }
389
390 keyBytes, err := x509.MarshalECPrivateKey(priv)
391 if err != nil {
392 return err
393 }
394
395 keyOut, err := os.OpenFile(keyPath, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0600)
396 if err != nil {
397 return err
398 }
399 defer keyOut.Close()
400
401 if err := pem.Encode(keyOut, &pem.Block{Type: "EC PRIVATE KEY", Bytes: keyBytes}); err != nil {
402 return err
403 }
404
405 return nil
406}
407
408// initConfig loads configuration from file.
409func initConfig(configPath string) (*viper.Viper, Config, error) {
410 v := viper.New()
411 if err := loadConfig(v, configPath); err != nil {
412 return nil, Config{}, fmt.Errorf("failed to load config: %w", err)
413 }
414
415 var cfg Config
416 if err := v.Unmarshal(&cfg); err != nil {
417 return nil, Config{}, fmt.Errorf("failed to unmarshal config: %w", err)
418 }
419
420 // Normalize domain config: prefer gordon_domain over registry_domain
421 if cfg.Server.GordonDomain != "" {
422 cfg.Server.RegistryDomain = cfg.Server.GordonDomain
423 }
424
425 return v, cfg, nil
426}
427
428// initLogger initializes the zerowrap logger.
429func initLogger(cfg Config) (zerowrap.Logger, func(), error) {
430 logConfig := zerowrap.Config{
431 Level: cfg.Logging.Level,
432 Format: cfg.Logging.Format,
433 }
434
435 // Always enable file logging so admin API can read process logs
436 // (Config flag is respected for backward-compatibility)
437 logPath := cfg.Logging.File.Path
438 if logPath == "" {
439 dataDir := cfg.Server.DataDir
440 if dataDir == "" {
441 dataDir = DefaultDataDir()
442 }
443 logPath = filepath.Join(dataDir, "logs", "gordon.log")
444 }
445
446 log, cleanup, err := zerowrap.NewWithFile(logConfig, zerowrap.FileConfig{
447 Enabled: cfg.Logging.File.Enabled,
448 Path: logPath,
449 MaxSize: cfg.Logging.File.MaxSize,
450 MaxBackups: cfg.Logging.File.MaxBackups,
451 MaxAge: cfg.Logging.File.MaxAge,
452 Compress: true,
453 })
454 if err != nil {
455 return zerowrap.Default(), nil, fmt.Errorf("failed to create logger with file: %w", err)
456 }
457 return log, cleanup, nil
458}
459
460// createServices creates all the application services.
461func createServices(ctx context.Context, v *viper.Viper, cfg Config, log zerowrap.Logger) (*services, error) {
462 svc := &services{}
463 var err error
464
465 // Create output adapters
466 runtimeSocket := resolveRuntimeConfig(v.GetString("server.runtime"))
467 if svc.runtime, svc.eventBus, err = createOutputAdapters(ctx, log, runtimeSocket); err != nil {
468 return nil, err
469 }
470
471 // Create storage
472 if svc.blobStorage, svc.manifestStorage, err = createStorage(cfg, log); err != nil {
473 return nil, err
474 }
475
476 // Create log writer
477 if svc.logWriter, err = createLogWriter(cfg, log); err != nil {
478 return nil, err
479 }
480
481 // Create auth service (if enabled)
482 if svc.tokenStore, svc.authSvc, err = createAuthService(ctx, cfg, log); err != nil {
483 return nil, err
484 }
485
486 if err := setupInternalRegistryAuth(svc, log); err != nil {
487 return nil, err
488 }
489
490 // Create use case services
491 svc.configSvc = config.NewService(v, svc.eventBus)
492 if err := svc.configSvc.Load(ctx); err != nil {
493 return nil, log.WrapErr(err, "failed to load configuration")
494 }
495
496 var backend domain.SecretsBackend
497 var passStore *domainsecrets.PassStore
498 var domainSecretStore out.DomainSecretStore
499
500 svc.envDir, backend, passStore, domainSecretStore, err = createDomainSecretStore(cfg, log)
501 if err != nil {
502 return nil, err
503 }
504
505 if svc.envLoader, err = createEnvLoader(backend, svc.envDir, passStore, log); err != nil {
506 return nil, err
507 }
508
509 svc.secretSvc = secretsSvc.NewService(domainSecretStore, log, svc.eventBus)
510
511 if svc.containerSvc, err = createContainerService(ctx, v, cfg, svc, log); err != nil {
512 return nil, err
513 }
514
515 if svc.backupStorage, svc.backupSvc, err = createBackupService(cfg, svc, log); err != nil {
516 return nil, err
517 }
518
519 svc.registrySvc = registrySvc.NewService(svc.blobStorage, svc.manifestStorage, svc.eventBus)
520 svc.imageSvc = images.NewService(svc.runtime, svc.manifestStorage, svc.blobStorage, log)
521 svc.volumeSvc = volumesSvc.NewService(svc.runtime)
522
523 injectTelemetryMetrics(cfg, svc, log)
524
525 proxyCfg, err := buildProxyConfig(cfg, log)
526 if err != nil {
527 return nil, err
528 }
529 svc.maxBlobChunkSize = proxyCfg.maxBlobChunkSize
530 svc.proxySvc = proxy.NewService(svc.runtime, svc.containerSvc, svc.configSvc, proxyCfg.proxyConfig)
531
532 // Wire synchronous proxy cache invalidation for zero-downtime deployments.
533 // The proxy service implements out.ProxyCacheInvalidator via InvalidateTarget().
534 svc.containerSvc.SetProxyCacheInvalidator(svc.proxySvc)
535 svc.containerSvc.SetProxyDrainWaiter(svc.proxySvc)
536
537 // Create token handler for registry token endpoint
538 if svc.authSvc != nil {
539 internalAuth := authhandler.InternalAuth{
540 Username: svc.internalRegUser,
541 Password: svc.internalRegPass,
542 }
543 svc.authHandler = authhandler.NewHandler(svc.authSvc, internalAuth, log)
544 }
545
546 // Create health service for route health checking
547 prober := httpprober.New()
548 svc.healthSvc = health.NewService(svc.configSvc, svc.containerSvc, prober, log)
549
550 // Create log service for accessing logs via admin API
551 svc.logSvc = logs.NewService(resolveLogFilePath(cfg), cfg.Logging.File.Enabled, svc.containerSvc, svc.runtime, log)
552
553 initPreviewService(ctx, cfg, svc, log)
554
555 // Create admin handler for admin API
556 svc.adminHandler = admin.NewHandler(admin.HandlerDeps{
557 ConfigSvc: svc.configSvc,
558 AuthSvc: svc.authSvc,
559 ContainerSvc: svc.containerSvc,
560 HealthSvc: svc.healthSvc,
561 SecretSvc: svc.secretSvc,
562 LogSvc: svc.logSvc,
563 RegistrySvc: svc.registrySvc,
564 EventBus: svc.eventBus,
565 Log: log,
566 BackupSvc: svc.backupSvc,
567 PreviewSvc: svc.previewService,
568 ImageSvc: svc.imageSvc,
569 VolumeSvc: svc.volumeSvc,
570 })
571
572 return svc, nil
573}
574
575// initPreviewService sets up the preview store, service, and TTL ticker.
576func initPreviewService(ctx context.Context, cfg Config, svc *services, log zerowrap.Logger) {
577 previewStorePath := filepath.Join(resolveDataDir(cfg.Server.DataDir), "previews.json")
578 svc.previewStore = filesystem.NewPreviewStore(previewStorePath)
579 previewConfig := svc.configSvc.GetPreviewConfig()
580 svc.previewService = preview.NewService(svc.previewStore, previewConfig.TTL).
581 WithDeployer(svc.containerSvc).
582 WithRouteManager(svc.configSvc).
583 WithVolumeCloner(svc.runtime).
584 WithRegistryDomain(svc.configSvc.GetRegistryDomain()).
585 WithEnvLoader(svc.envLoader)
586 if err := svc.previewService.Load(ctx); err != nil {
587 log.Warn().Err(err).Msg("failed to load previews")
588 }
589
590 // Derive sweep interval from TTL: half the TTL, capped at 1 hour, minimum 1 minute.
591 sweepInterval := previewConfig.TTL / 2
592 if sweepInterval > 1*time.Hour {
593 sweepInterval = 1 * time.Hour
594 }
595 if sweepInterval < 1*time.Minute {
596 sweepInterval = 1 * time.Minute
597 }
598
599 svc.previewService.StartTicker(ctx, sweepInterval, func(ctx context.Context, p domain.PreviewRoute) {
600 teardownTrackedPreview(ctx, svc, p)
601 }, func(ctx context.Context) {
602 gcOrphanedPreviews(ctx, svc, previewConfig)
603 })
604}
605
606// teardownTrackedPreview removes all resources for a tracked preview that has expired.
607func teardownTrackedPreview(ctx context.Context, svc *services, p domain.PreviewRoute) {
608 log := zerowrap.FromCtx(ctx)
609 for _, containerName := range p.Containers {
610 if err := svc.runtime.StopContainer(ctx, containerName); err != nil {
611 log.Warn().Err(err).Str("container", containerName).Str("preview", p.Domain).Msg("failed to stop preview container")
612 }
613 if err := svc.runtime.RemoveContainer(ctx, containerName, true); err != nil {
614 log.Warn().Err(err).Str("container", containerName).Str("preview", p.Domain).Msg("failed to remove preview container")
615 }
616 }
617 for _, volName := range p.Volumes {
618 if err := svc.runtime.RemoveVolume(ctx, volName, true); err != nil {
619 log.Warn().Err(err).Str("volume", volName).Str("preview", p.Domain).Msg("failed to remove preview volume")
620 }
621 }
622 // Remove network (naming convention: {networkPrefix}-{domain-sanitized}).
623 networkPrefix := svc.configSvc.GetNetworkPrefix()
624 networkName := networkPrefix + "-" + strings.ReplaceAll(p.Domain, ".", "-")
625 if err := svc.runtime.RemoveNetwork(ctx, networkName); err != nil {
626 log.Warn().Err(err).Str("network", networkName).Str("preview", p.Domain).Msg("failed to remove preview network")
627 }
628 // Remove route from config so proxy stops routing to this domain.
629 if err := svc.configSvc.RemoveRoute(ctx, p.Domain); err != nil {
630 if errors.Is(err, domain.ErrRouteNotFound) {
631 log.Debug().Str("domain", p.Domain).Msg("preview route already removed from config")
632 } else {
633 log.Warn().Err(err).Str("domain", p.Domain).Msg("failed to remove preview route from config")
634 }
635 }
636 svc.proxySvc.InvalidateTarget(ctx, p.Domain)
637}
638
639// gcOrphanedPreviews finds and tears down untracked preview containers.
640func gcOrphanedPreviews(ctx context.Context, svc *services, previewConfig domain.PreviewConfig) {
641 log := zerowrap.FromCtx(ctx)
642 orphans := svc.previewService.CollectOrphans(ctx, svc.runtime, previewConfig.TagPatterns, previewConfig.Separator)
643 for _, c := range orphans {
644 orphanDomain := c.Labels[domain.LabelDomain]
645 log.Warn().Str("container", c.Name).Str("image", c.Image).Str("domain", orphanDomain).
646 Time("created", c.Created).Msg("orphaned preview container detected, cleaning up")
647
648 // Stop container first, then clean up domain resources while container
649 // still exists (so future scans can rediscover it on failure), then remove.
650 if err := svc.runtime.StopContainer(ctx, c.Name); err != nil {
651 log.Warn().Err(err).Str("container", c.Name).Msg("failed to stop orphan container")
652 }
653
654 if orphanDomain != "" {
655 if err := cleanupOrphanDomainResources(ctx, svc, orphanDomain); err != nil {
656 log.Warn().Err(err).Str("container", c.Name).Str("domain", orphanDomain).
657 Msg("orphan resource cleanup failed, deferring container removal to next scan")
658 continue
659 }
660 }
661
662 if err := svc.runtime.RemoveContainer(ctx, c.Name, true); err != nil {
663 log.Warn().Err(err).Str("container", c.Name).Msg("failed to remove orphan container")
664 }
665
666 log.Info().Str("container", c.Name).Str("domain", orphanDomain).Msg("orphaned preview container cleaned up")
667 }
668}
669
670// cleanupOrphanDomainResources removes volumes, network, and route for an
671// orphaned preview domain. Returns an error if any step fails so the caller
672// can defer container removal until the next scan.
673func cleanupOrphanDomainResources(ctx context.Context, svc *services, orphanDomain string) error {
674 log := zerowrap.FromCtx(ctx)
675 domainSanitized := strings.ReplaceAll(orphanDomain, ".", "-")
676 var errs []error
677
678 errs = append(errs, removeOrphanVolumes(ctx, svc, domainSanitized, log)...)
679 errs = append(errs, removeOrphanNetwork(ctx, svc, domainSanitized, log))
680 errs = append(errs, removeOrphanRoute(ctx, svc, orphanDomain, log))
681 svc.proxySvc.InvalidateTarget(ctx, orphanDomain)
682
683 return errors.Join(errs...)
684}
685
686func removeOrphanVolumes(ctx context.Context, svc *services, domainSanitized string, log zerowrap.Logger) []error {
687 _, volPrefix, _ := svc.configSvc.GetVolumeConfig()
688 prefix := volPrefix + "-" + domainSanitized + "-"
689
690 volumes, err := svc.runtime.ListVolumes(ctx)
691 if err != nil {
692 log.Warn().Err(err).Msg("failed to list volumes for orphan cleanup")
693 return []error{err}
694 }
695
696 var errs []error
697 for _, v := range volumes {
698 if !strings.HasPrefix(v.Name, prefix) {
699 continue
700 }
701 if err := svc.runtime.RemoveVolume(ctx, v.Name, true); err != nil {
702 log.Warn().Err(err).Str("volume", v.Name).Msg("failed to remove orphan volume")
703 errs = append(errs, err)
704 }
705 }
706 return errs
707}
708
709func removeOrphanNetwork(ctx context.Context, svc *services, domainSanitized string, log zerowrap.Logger) error {
710 name := svc.configSvc.GetNetworkPrefix() + "-" + domainSanitized
711 if err := svc.runtime.RemoveNetwork(ctx, name); err != nil {
712 log.Warn().Err(err).Str("network", name).Msg("failed to remove orphan network")
713 return err
714 }
715 return nil
716}
717
718func removeOrphanRoute(ctx context.Context, svc *services, orphanDomain string, log zerowrap.Logger) error {
719 err := svc.configSvc.RemoveRoute(ctx, orphanDomain)
720 if err == nil || errors.Is(err, domain.ErrRouteNotFound) {
721 return nil
722 }
723 log.Warn().Err(err).Str("domain", orphanDomain).Msg("failed to remove orphan route")
724 return err
725}
726
727// injectTelemetryMetrics creates and injects OTel metrics into services when
728// telemetry is enabled. Skipped otherwise to avoid unnecessary allocations.
729func injectTelemetryMetrics(cfg Config, svc *services, log zerowrap.Logger) {
730 if !cfg.Telemetry.Enabled || !cfg.Telemetry.Metrics {
731 return
732 }
733 gordonMetrics, err := telemetry.NewMetrics()
734 if err != nil {
735 log.Warn().Err(err).Msg("failed to create telemetry metrics, continuing without metrics")
736 return
737 }
738 svc.containerSvc.SetMetrics(gordonMetrics)
739 svc.registrySvc.SetMetrics(gordonMetrics)
740 svc.eventBus.SetMetrics(gordonMetrics)
741}
742
743func setupInternalRegistryAuth(svc *services, log zerowrap.Logger) error {
744 var err error
745 svc.internalRegUser, svc.internalRegPass, err = generateInternalRegistryAuth()
746 if err != nil {
747 return log.WrapErr(err, "failed to generate internal registry credentials")
748 }
749
750 // Persist credentials to file for CLI access (gordon auth internal)
751 if err := persistInternalCredentials(svc.internalRegUser, svc.internalRegPass); err != nil {
752 log.Warn().Err(err).Msg("failed to persist internal credentials for CLI access")
753 }
754
755 log.Debug().Msg("internal registry auth generated for loopback pulls")
756 return nil
757}
758
759func createDomainSecretStore(cfg Config, log zerowrap.Logger) (string, domain.SecretsBackend, *domainsecrets.PassStore, out.DomainSecretStore, error) {
760 envDir := resolveEnvDir(cfg)
761 backend, err := resolveSecretsBackend(cfg.Auth.SecretsBackend)
762 if err != nil {
763 return "", "", nil, nil, log.WrapErr(err, "failed to resolve secrets backend")
764 }
765
766 switch backend {
767 case domain.SecretsBackendPass:
768 passStore, err := domainsecrets.NewPassStore(log)
769 if err != nil {
770 return "", backend, nil, nil, log.WrapErr(err, "failed to create pass domain secret store")
771 }
772 if err := migrateEnvFilesToPass(envDir, passStore, log); err != nil {
773 return "", backend, nil, nil, log.WrapErr(err, "failed to migrate env files to pass")
774 }
775 return envDir, backend, passStore, passStore, nil
776 default:
777 store, err := domainsecrets.NewFileStore(envDir, log)
778 if err != nil {
779 return "", backend, nil, nil, log.WrapErr(err, "failed to create domain secret store")
780 }
781 return envDir, backend, nil, store, nil
782 }
783}
784
785// resolveLogFilePath returns the configured log file path or a default.
786func resolveLogFilePath(cfg Config) string {
787 if cfg.Logging.File.Path != "" {
788 return cfg.Logging.File.Path
789 }
790 dataDir := cfg.Server.DataDir
791 if dataDir == "" {
792 dataDir = DefaultDataDir()
793 }
794 return filepath.Join(dataDir, "logs", "gordon.log")
795}
796
797// resolveRuntimeConfig converts a server.runtime config value to a socket path.
798// "auto" or "" means auto-detect; any other value is treated as an explicit socket path.
799// URI schemes (unix://, tcp://) are stripped so callers receive a bare path.
800func resolveRuntimeConfig(value string) string {
801 if value == "" || value == "auto" {
802 return ""
803 }
804 if strings.HasPrefix(value, "unix://") {
805 return strings.TrimPrefix(value, "unix://")
806 }
807 return value
808}
809
810// createOutputAdapters creates the container runtime and event bus.
811func createOutputAdapters(ctx context.Context, log zerowrap.Logger, runtimeSocket string) (*docker.Runtime, *eventbus.InMemory, error) {
812 detection := docker.DetectRuntimeSocket(runtimeSocket)
813
814 var runtime *docker.Runtime
815 var err error
816
817 switch detection.Source {
818 case "none":
819 return nil, nil, fmt.Errorf("no container runtime found: checked Docker socket, Podman socket, DOCKER_HOST env var. Install Docker or Podman, or set server.runtime in config")
820 case "DOCKER_HOST_passthrough":
821 detection.RuntimeName = "docker"
822 runtime, err = docker.NewRuntime()
823 default:
824 if detection.SocketPath != "" {
825 runtime, err = docker.NewRuntimeWithSocket(detection.SocketPath)
826 } else {
827 runtime, err = docker.NewRuntime()
828 }
829 }
830 if err != nil {
831 return nil, nil, log.WrapErr(err, "failed to create container runtime")
832 }
833
834 if err := runtime.Ping(ctx); err != nil {
835 return nil, nil, log.WrapErr(err, fmt.Sprintf("container runtime not available (detected: %s via %s)", detection.RuntimeName, detection.Source))
836 }
837
838 runtimeVersion, _ := runtime.Version(ctx)
839 log.Info().
840 Str("runtime", detection.RuntimeName).
841 Str("version", runtimeVersion).
842 Str("source", detection.Source).
843 Msg("container runtime initialized")
844
845 eventBus := eventbus.NewInMemory(100, log)
846
847 return runtime, eventBus, nil
848}
849
850// createStorage creates blob and manifest storage.
851func createStorage(cfg Config, log zerowrap.Logger) (*filesystem.BlobStorage, *filesystem.ManifestStorage, error) {
852 dataDir := cfg.Server.DataDir
853 if dataDir == "" {
854 dataDir = DefaultDataDir()
855 }
856
857 registryDir := filepath.Join(dataDir, "registry")
858
859 blobStorage, err := filesystem.NewBlobStorage(registryDir, log)
860 if err != nil {
861 return nil, nil, log.WrapErr(err, "failed to create blob storage")
862 }
863
864 manifestStorage, err := filesystem.NewManifestStorage(registryDir, log)
865 if err != nil {
866 return nil, nil, log.WrapErr(err, "failed to create manifest storage")
867 }
868
869 return blobStorage, manifestStorage, nil
870}
871
872// createEnvLoader creates the environment loader with secret providers.
873func createEnvLoader(backend domain.SecretsBackend, envDir string, passStore *domainsecrets.PassStore, log zerowrap.Logger) (out.EnvLoader, error) {
874 switch backend {
875 case domain.SecretsBackendPass:
876 loader, err := envloader.NewPassLoader(passStore, log)
877 if err != nil {
878 return nil, log.WrapErr(err, "failed to create pass env loader")
879 }
880 return loader, nil
881 default:
882 loader, err := envloader.NewFileLoader(envDir, log)
883 if err != nil {
884 return nil, log.WrapErr(err, "failed to create env loader")
885 }
886
887 // Register secret providers
888 passProvider := secrets.NewPassProvider(log)
889 if passProvider.IsAvailable() {
890 loader.RegisterSecretProvider(passProvider)
891 log.Debug().Msg("pass secret provider registered")
892 }
893
894 sopsProvider := secrets.NewSopsProvider(log)
895 if sopsProvider.IsAvailable() {
896 loader.RegisterSecretProvider(sopsProvider)
897 log.Debug().Msg("sops secret provider registered")
898 }
899
900 return loader, nil
901 }
902}
903
904// createLogWriter creates the container log writer.
905func createLogWriter(cfg Config, log zerowrap.Logger) (*logwriter.LogWriter, error) {
906 if !cfg.Logging.ContainerLogs.Enabled {
907 log.Debug().Msg("container log collection disabled")
908 return nil, nil
909 }
910
911 // Determine log directory
912 logDir := cfg.Logging.ContainerLogs.Dir
913 if logDir == "" {
914 dataDir := cfg.Server.DataDir
915 if dataDir == "" {
916 dataDir = DefaultDataDir()
917 }
918 logDir = filepath.Join(dataDir, "logs", "containers")
919 }
920
921 writer, err := logwriter.New(logwriter.Config{
922 Dir: logDir,
923 MaxSize: cfg.Logging.ContainerLogs.MaxSize,
924 MaxBackups: cfg.Logging.ContainerLogs.MaxBackups,
925 MaxAge: cfg.Logging.ContainerLogs.MaxAge,
926 })
927 if err != nil {
928 return nil, log.WrapErr(err, "failed to create container log writer")
929 }
930
931 log.Info().Str("dir", logDir).Msg("container log collection enabled")
932 return writer, nil
933}
934
935const (
936 internalRegistryUsername = "gordon-internal"
937 serviceTokenSubject = "gordon-service"
938 serviceTokenDefaultTTL = 30 * 24 * time.Hour
939)
940
941func generateInternalRegistryAuth() (string, string, error) {
942 password, err := randomTokenHex(32)
943 if err != nil {
944 return "", "", err
945 }
946 return internalRegistryUsername, password, nil
947}
948
949func randomTokenHex(size int) (string, error) {
950 buf := make([]byte, size)
951 if _, err := rand.Read(buf); err != nil {
952 return "", err
953 }
954 return hex.EncodeToString(buf), nil
955}
956
957// InternalCredentials holds the internal registry credentials for CLI access.
958type InternalCredentials struct {
959 Username string `json:"username"`
960 Password string `json:"password"`
961}
962
963// getSecureRuntimeDir returns a secure directory for runtime files.
964// Priority: XDG_RUNTIME_DIR > ~/.gordon/run
965func getSecureRuntimeDir() (string, error) {
966 // Try XDG_RUNTIME_DIR first (typically /run/user/<uid> on Linux)
967 if runtimeDir := os.Getenv("XDG_RUNTIME_DIR"); runtimeDir != "" {
968 gordonDir := filepath.Join(runtimeDir, "gordon")
969 if err := os.MkdirAll(gordonDir, 0700); err == nil {
970 return gordonDir, nil
971 }
972 }
973
974 // Fall back to ~/.gordon/run
975 homeDir, err := os.UserHomeDir()
976 if err != nil {
977 return "", fmt.Errorf("failed to get home directory: %w", err)
978 }
979
980 gordonDir := filepath.Join(homeDir, ".gordon", "run")
981 if err := os.MkdirAll(gordonDir, 0700); err != nil {
982 return "", fmt.Errorf("failed to create runtime directory: %w", err)
983 }
984
985 return gordonDir, nil
986}
987
988// getInternalCredentialsFile returns the path to the internal credentials file.
989// SECURITY: Credentials are stored in a secure location with restricted permissions.
990func getInternalCredentialsFile() string {
991 runtimeDir, err := getSecureRuntimeDir()
992 if err != nil {
993 // Fall back to temp dir if we can't get secure dir (shouldn't happen)
994 return filepath.Join(os.TempDir(), "gordon-internal-creds.json")
995 }
996 return filepath.Join(runtimeDir, "internal-creds.json")
997}
998
999// persistInternalCredentials saves the internal registry credentials to a secure file.
1000// SECURITY: Credentials are stored in XDG_RUNTIME_DIR or ~/.gordon/run with 0600 permissions.
1001// The file is cleaned up on graceful shutdown but may persist if Gordon crashes.
1002// These credentials are for internal loopback communication only and are regenerated on each start.
1003func persistInternalCredentials(username, password string) error {
1004 creds := InternalCredentials{
1005 Username: username,
1006 Password: password,
1007 }
1008 data, err := json.Marshal(creds)
1009 if err != nil {
1010 return fmt.Errorf("failed to marshal credentials: %w", err)
1011 }
1012
1013 credFile := getInternalCredentialsFile()
1014
1015 // Ensure parent directory exists with secure permissions
1016 if err := os.MkdirAll(filepath.Dir(credFile), 0700); err != nil {
1017 return fmt.Errorf("failed to create credentials directory: %w", err)
1018 }
1019
1020 // Write file with restrictive permissions (owner read/write only)
1021 if err := os.WriteFile(credFile, data, 0600); err != nil {
1022 return fmt.Errorf("failed to write credentials file: %w", err)
1023 }
1024 return nil
1025}
1026
1027// cleanupInternalCredentials removes the internal credentials file.
1028func cleanupInternalCredentials() {
1029 _ = os.Remove(getInternalCredentialsFile())
1030}
1031
1032// getInternalCredentialsCandidates returns candidate file paths in priority order:
1033// 1. XDG_RUNTIME_DIR/gordon/ (set by systemd for the daemon)
1034// 2. /run/user/<uid>/gordon/ (well-known systemd default, for CLI in shells without XDG_RUNTIME_DIR)
1035// 3. ~/.gordon/run/ (fallback for non-systemd environments)
1036// 4. os.TempDir() (last resort, matches getInternalCredentialsFile fallback path)
1037func getInternalCredentialsCandidates() []string {
1038 var candidates []string
1039
1040 // 1. XDG_RUNTIME_DIR (set in daemon's environment)
1041 if runtimeDir := os.Getenv("XDG_RUNTIME_DIR"); runtimeDir != "" {
1042 candidates = append(candidates, filepath.Join(runtimeDir, "gordon", "internal-creds.json"))
1043 }
1044
1045 // 2. /run/user/<uid>/gordon/ (systemd default, may not be in CLI's env)
1046 uid := os.Getuid()
1047 sysRuntime := filepath.Join("/run/user", fmt.Sprintf("%d", uid), "gordon", "internal-creds.json")
1048 // Avoid duplicate if XDG_RUNTIME_DIR already points here
1049 if len(candidates) == 0 || candidates[0] != sysRuntime {
1050 candidates = append(candidates, sysRuntime)
1051 }
1052
1053 // 3. ~/.gordon/run/ fallback
1054 if homeDir, err := os.UserHomeDir(); err == nil {
1055 candidates = append(candidates, filepath.Join(homeDir, ".gordon", "run", "internal-creds.json"))
1056 }
1057
1058 // 4. os.TempDir() last resort — matches the fallback path in getInternalCredentialsFile,
1059 // ensuring GetInternalCredentials can find credentials even when getSecureRuntimeDir fails.
1060 candidates = append(candidates, filepath.Join(os.TempDir(), "gordon-internal-creds.json"))
1061
1062 return candidates
1063}
1064
1065// GetInternalCredentialsFromCandidates reads credentials from the first candidate file that exists.
1066// Exported for testing.
1067func GetInternalCredentialsFromCandidates(candidates []string) (*InternalCredentials, error) {
1068 var lastErr error
1069 for _, path := range candidates {
1070 data, err := os.ReadFile(path)
1071 if os.IsNotExist(err) {
1072 continue
1073 }
1074 if err != nil {
1075 // Non-permission errors (e.g. EACCES) may be transient or path-specific;
1076 // record and try the next candidate rather than failing immediately.
1077 lastErr = fmt.Errorf("failed to read credentials file %s: %w", path, err)
1078 continue
1079 }
1080 var creds InternalCredentials
1081 if err := json.Unmarshal(data, &creds); err != nil {
1082 // Corrupt file — record and fall through to lower-priority candidates.
1083 lastErr = fmt.Errorf("failed to parse credentials at %s: %w", path, err)
1084 continue
1085 }
1086 return &creds, nil
1087 }
1088 if lastErr != nil {
1089 return nil, lastErr
1090 }
1091 return nil, fmt.Errorf("no credentials file found (is Gordon running?): checked %v", candidates)
1092}
1093
1094// GetInternalCredentials reads the internal registry credentials from file.
1095// Probes all candidate runtime directories so CLI works regardless of whether
1096// XDG_RUNTIME_DIR is set in the current shell environment.
1097func GetInternalCredentials() (*InternalCredentials, error) {
1098 return GetInternalCredentialsFromCandidates(getInternalCredentialsCandidates())
1099}
1100
1101// createAuthService creates the authentication service and token store.
1102func createAuthService(ctx context.Context, cfg Config, log zerowrap.Logger) (out.TokenStore, *auth.Service, error) {
1103 if !cfg.Auth.Enabled {
1104 log.Warn().Msg("auth.enabled=false detected: running in local-only mode (registry loopback-only, admin API disabled)")
1105 return nil, nil, nil
1106 }
1107
1108 authType, err := resolveAuthType(cfg)
1109 if err != nil {
1110 return nil, nil, fmt.Errorf("resolve auth type: %w", err)
1111 }
1112 backend, err := resolveSecretsBackend(cfg.Auth.SecretsBackend)
1113 if err != nil {
1114 return nil, nil, log.WrapErr(err, "failed to resolve secrets backend")
1115 }
1116 dataDir := resolveDataDir(cfg.Server.DataDir)
1117
1118 store, err := createTokenStore(backend, dataDir, log)
1119 if err != nil {
1120 return nil, nil, err
1121 }
1122
1123 authConfig, err := buildAuthConfig(ctx, cfg, authType, backend, dataDir, log)
1124 if err != nil {
1125 return nil, nil, err
1126 }
1127
1128 authSvc := auth.NewService(authConfig, store, log)
1129
1130 log.Info().
1131 Str("type", string(authType)).
1132 Str("backend", string(backend)).
1133 Msg("registry authentication enabled")
1134
1135 return store, authSvc, nil
1136}
1137
1138// resolveAuthType determines the auth type from config.
1139// Token-only authentication is the only supported mode.
1140func resolveAuthType(cfg Config) (domain.AuthType, error) {
1141 if cfg.Auth.Type != "" && cfg.Auth.Type != "token" {
1142 return "", fmt.Errorf("unsupported auth.type %q; only \"token\" is supported", cfg.Auth.Type)
1143 }
1144 return domain.AuthTypeToken, nil
1145}
1146
1147func resolveSecretsBackend(backend string) (domain.SecretsBackend, error) {
1148 switch backend {
1149 case "pass":
1150 return domain.SecretsBackendPass, nil
1151 case "sops":
1152 return domain.SecretsBackendSops, nil
1153 case "unsafe":
1154 return domain.SecretsBackendUnsafe, nil
1155 case "":
1156 return "", fmt.Errorf("auth.secrets_backend is required")
1157 default:
1158 return "", fmt.Errorf("unsupported auth.secrets_backend %q", backend)
1159 }
1160}
1161
1162func resolveDataDir(dataDir string) string {
1163 if dataDir == "" {
1164 return DefaultDataDir()
1165 }
1166 return dataDir
1167}
1168
1169func resolveEnvDir(cfg Config) string {
1170 dataDir := resolveDataDir(cfg.Server.DataDir)
1171 envDir := cfg.Env.Dir
1172 if envDir == "" {
1173 envDir = filepath.Join(dataDir, "env")
1174 }
1175 return envDir
1176}
1177
1178func createTokenStore(backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) (out.TokenStore, error) {
1179 // Token store is always created since tokens work in both auth modes
1180 store, err := tokenstore.NewStore(backend, dataDir, log)
1181 if err != nil {
1182 return nil, log.WrapErr(err, "failed to create token store")
1183 }
1184 return store, nil
1185}
1186
1187func buildAuthConfig(ctx context.Context, cfg Config, authType domain.AuthType, backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) (auth.Config, error) {
1188 authConfig := auth.Config{
1189 Enabled: cfg.Auth.Enabled,
1190 AuthType: authType,
1191 Username: cfg.Auth.Username,
1192 }
1193
1194 // Token config is always required (tokens work in all auth modes)
1195 secret, expiry, err := loadTokenConfig(ctx, cfg, backend, dataDir, log)
1196 if err != nil {
1197 return auth.Config{}, err
1198 }
1199 authConfig.TokenSecret = secret
1200 authConfig.TokenExpiry = expiry
1201
1202 accessTokenTTL := 15 * time.Minute // default
1203 if cfg.Auth.AccessTokenTTL != "" {
1204 parsed, err := time.ParseDuration(cfg.Auth.AccessTokenTTL)
1205 if err != nil {
1206 return auth.Config{}, fmt.Errorf("invalid auth.access_token_ttl %q: %w", cfg.Auth.AccessTokenTTL, err)
1207 }
1208 if parsed <= 0 {
1209 return auth.Config{}, fmt.Errorf("auth.access_token_ttl must be positive")
1210 }
1211 if parsed > auth.MaxAccessTokenLifetime {
1212 return auth.Config{}, fmt.Errorf("auth.access_token_ttl must not exceed %v", auth.MaxAccessTokenLifetime)
1213 }
1214 accessTokenTTL = parsed
1215 }
1216 authConfig.AccessTokenTTL = accessTokenTTL
1217
1218 return authConfig, nil
1219}
1220
1221func loadTokenConfig(ctx context.Context, cfg Config, backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) ([]byte, time.Duration, error) {
1222 secret, err := loadTokenSecret(ctx, cfg, backend, dataDir, log)
1223 if err != nil {
1224 return nil, 0, err
1225 }
1226
1227 expiry, err := parseTokenExpiry(cfg.Auth.TokenExpiry)
1228 if err != nil {
1229 return nil, 0, err
1230 }
1231
1232 return secret, expiry, nil
1233}
1234
1235// TokenSecretEnvVar is the environment variable for the JWT signing secret.
1236// SECURITY: This takes priority over config file to allow secure secret injection.
1237const TokenSecretEnvVar = "GORDON_AUTH_TOKEN_SECRET" //nolint:gosec // This is an env var name, not a credential
1238
1239func loadTokenSecret(ctx context.Context, cfg Config, backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) ([]byte, error) {
1240 // SECURITY: Priority order for token secret:
1241 // 1. Environment variable (most secure - no disk exposure)
1242 // 2. Secrets backend (pass/sops - encrypted)
1243 // 3. Config file path (least preferred)
1244
1245 const minTokenSecretLength = 32
1246
1247 // Check environment variable first
1248 if envSecret := os.Getenv(TokenSecretEnvVar); envSecret != "" {
1249 if len(envSecret) < minTokenSecretLength {
1250 return nil, fmt.Errorf("token secret from %s must be at least %d bytes (got %d)", TokenSecretEnvVar, minTokenSecretLength, len(envSecret))
1251 }
1252 log.Debug().Msg("using token secret from environment variable")
1253 return []byte(envSecret), nil
1254 }
1255
1256 // Fall back to config-specified path via secrets backend
1257 if cfg.Auth.TokenSecret == "" {
1258 return nil, fmt.Errorf("token_secret is required for JWT token generation; set %s environment variable or configure auth.token_secret", TokenSecretEnvVar)
1259 }
1260
1261 secret, err := loadSecret(ctx, backend, cfg.Auth.TokenSecret, dataDir, log)
1262 if err != nil {
1263 return nil, log.WrapErr(err, "failed to load token secret")
1264 }
1265
1266 if len(secret) < minTokenSecretLength {
1267 return nil, fmt.Errorf("token_secret must be at least %d bytes (got %d); use a strong random secret", minTokenSecretLength, len(secret))
1268 }
1269
1270 return []byte(secret), nil
1271}
1272
1273func parseTokenExpiry(expiry string) (time.Duration, error) {
1274 if expiry == "" {
1275 return 0, nil
1276 }
1277
1278 parsed, err := duration.Parse(expiry)
1279 if err != nil {
1280 return 0, fmt.Errorf("invalid token_expiry: %w", err)
1281 }
1282
1283 return parsed, nil
1284}
1285
1286func resolveServiceTokenExpiry(cfg Config) (time.Duration, error) {
1287 expiry, err := parseTokenExpiry(cfg.Auth.TokenExpiry)
1288 if err != nil {
1289 return 0, err
1290 }
1291 if expiry <= 0 {
1292 return serviceTokenDefaultTTL, nil
1293 }
1294 return expiry, nil
1295}
1296
1297// loadSecret loads a secret from the configured backend.
1298func loadSecret(ctx context.Context, backend domain.SecretsBackend, path, dataDir string, log zerowrap.Logger) (string, error) {
1299 switch backend {
1300 case domain.SecretsBackendPass:
1301 provider := secrets.NewPassProvider(log)
1302 return provider.GetSecret(ctx, path)
1303 case domain.SecretsBackendSops:
1304 provider := secrets.NewSopsProvider(log)
1305 return provider.GetSecret(ctx, path)
1306 case domain.SecretsBackendUnsafe:
1307 // For unsafe backend, path is relative to dataDir/secrets/
1308 secretFile := filepath.Join(dataDir, "secrets", path)
1309 data, err := os.ReadFile(secretFile)
1310 if err != nil {
1311 return "", fmt.Errorf("failed to read secret file: %w", err)
1312 }
1313 return string(data), nil
1314 default:
1315 return "", fmt.Errorf("unknown secrets backend: %s", backend)
1316 }
1317}
1318
1319// proxyConfigResult holds parsed proxy and blob chunk size config.
1320type proxyConfigResult struct {
1321 proxyConfig proxy.Config
1322 maxBlobChunkSize int64
1323}
1324
1325// buildProxyConfig parses size-related config fields and builds the proxy config.
1326func buildProxyConfig(cfg Config, log zerowrap.Logger) (*proxyConfigResult, error) {
1327 maxProxyBodySize := int64(512 << 20) // 512MB default
1328 if cfg.Server.MaxProxyBodySize != "" {
1329 parsedSize, err := bytesize.Parse(cfg.Server.MaxProxyBodySize)
1330 if err != nil {
1331 return nil, log.WrapErrWithFields(err, "invalid server.max_proxy_body_size configuration", map[string]any{"value": cfg.Server.MaxProxyBodySize})
1332 }
1333 maxProxyBodySize = parsedSize
1334 }
1335
1336 maxBlobChunkSize := int64(registry.DefaultMaxBlobChunkSize)
1337 if cfg.Server.MaxBlobChunkSize != "" {
1338 parsedSize, err := bytesize.Parse(cfg.Server.MaxBlobChunkSize)
1339 if err != nil {
1340 return nil, log.WrapErrWithFields(err, "invalid server.max_blob_chunk_size configuration", map[string]any{"value": cfg.Server.MaxBlobChunkSize})
1341 }
1342 maxBlobChunkSize = parsedSize
1343 }
1344
1345 maxProxyResponseSize := int64(1 << 30) // 1GB default
1346 if cfg.Server.MaxProxyResponseSize != "" {
1347 parsedSize, err := bytesize.Parse(cfg.Server.MaxProxyResponseSize)
1348 if err != nil {
1349 return nil, log.WrapErrWithFields(err, "invalid server.max_proxy_response_size configuration", map[string]any{"value": cfg.Server.MaxProxyResponseSize})
1350 }
1351 maxProxyResponseSize = parsedSize
1352 }
1353
1354 maxConcurrentConns := cfg.Server.MaxConcurrentConns
1355 if maxConcurrentConns < 0 {
1356 maxConcurrentConns = 10000 // default when explicitly set to -1
1357 }
1358 // 0 means no limit (as documented in proxy.Config)
1359
1360 return &proxyConfigResult{
1361 proxyConfig: proxy.Config{
1362 RegistryDomain: cfg.Server.RegistryDomain,
1363 RegistryPort: cfg.Server.RegistryPort,
1364 MaxBodySize: maxProxyBodySize,
1365 MaxResponseSize: maxProxyResponseSize,
1366 MaxConcurrentConns: maxConcurrentConns,
1367 },
1368 maxBlobChunkSize: maxBlobChunkSize,
1369 }, nil
1370}
1371
1372// createContainerService creates the container service with configuration.
1373func createContainerService(ctx context.Context, v *viper.Viper, cfg Config, svc *services, log zerowrap.Logger) (*container.Service, error) {
1374 // Parse and validate container resource limits from config
1375 if cfg.Containers.CPULimit < 0 {
1376 return nil, fmt.Errorf("containers.cpu_limit must be >= 0 (got %f)", cfg.Containers.CPULimit)
1377 }
1378 if cfg.Containers.PidsLimit < 0 {
1379 return nil, fmt.Errorf("containers.pids_limit must be >= 0 (got %d)", cfg.Containers.PidsLimit)
1380 }
1381 var defaultMemoryLimit int64
1382 if cfg.Containers.MemoryLimit != "" {
1383 parsed, err := bytesize.Parse(cfg.Containers.MemoryLimit)
1384 if err != nil {
1385 return nil, fmt.Errorf("invalid containers.memory_limit %q: %w", cfg.Containers.MemoryLimit, err)
1386 }
1387 if parsed <= 0 {
1388 return nil, fmt.Errorf("containers.memory_limit must be positive (got %q)", cfg.Containers.MemoryLimit)
1389 }
1390 defaultMemoryLimit = parsed
1391 }
1392 var defaultNanoCPUs int64
1393 if cfg.Containers.CPULimit > 0 {
1394 defaultNanoCPUs = int64(cfg.Containers.CPULimit * 1e9)
1395 }
1396
1397 containerConfig := container.Config{
1398 RegistryAuthEnabled: cfg.Auth.Enabled,
1399 RegistryDomain: cfg.Server.RegistryDomain,
1400 RegistryPort: cfg.Server.RegistryPort,
1401 InternalRegistryUsername: svc.internalRegUser,
1402 InternalRegistryPassword: svc.internalRegPass,
1403 PullPolicy: v.GetString("deploy.pull_policy"),
1404 VolumeAutoCreate: v.GetBool("volumes.auto_create"),
1405 VolumePrefix: v.GetString("volumes.prefix"),
1406 VolumePreserve: v.GetBool("volumes.preserve"),
1407 NetworkIsolation: v.GetBool("network_isolation.enabled"),
1408 NetworkPrefix: v.GetString("network_isolation.network_prefix"),
1409 NetworkGroups: svc.configSvc.GetNetworkGroups(),
1410 Attachments: svc.configSvc.GetAttachments(),
1411 ReadinessDelay: v.GetDuration("deploy.readiness_delay"),
1412 ReadinessMode: v.GetString("deploy.readiness_mode"),
1413 HealthTimeout: v.GetDuration("deploy.health_timeout"),
1414 StabilizationDelay: v.GetDuration("deploy.stabilization_delay"),
1415 TCPProbeTimeout: v.GetDuration("deploy.tcp_probe_timeout"),
1416 HTTPProbeTimeout: v.GetDuration("deploy.http_probe_timeout"),
1417 DrainDelay: v.GetDuration("deploy.drain_delay"),
1418 DrainMode: v.GetString("deploy.drain_mode"),
1419 DrainTimeout: v.GetDuration("deploy.drain_timeout"),
1420 DefaultMemoryLimit: defaultMemoryLimit,
1421 DefaultNanoCPUs: defaultNanoCPUs,
1422 DefaultPidsLimit: cfg.Containers.PidsLimit,
1423 AttachmentReadinessTimeout: v.GetDuration("deploy.attachment_readiness_timeout"),
1424 }
1425 if v.IsSet("deploy.drain_delay") {
1426 containerConfig.DrainDelayConfigured = true
1427 containerConfig.DrainDelay = v.GetDuration("deploy.drain_delay")
1428 }
1429
1430 if containerConfig.RegistryAuthEnabled {
1431 if svc.authSvc == nil {
1432 return nil, fmt.Errorf("authentication service unavailable: cannot generate registry service token")
1433 }
1434 expiry, err := resolveServiceTokenExpiry(cfg)
1435 if err != nil {
1436 return nil, log.WrapErr(err, "failed to resolve service token expiry")
1437 }
1438 // Note: Service tokens are not auto-refreshed. If the token expires during
1439 // container runtime, the container will need to be recreated to get a new token.
1440 serviceToken, err := svc.authSvc.GenerateToken(ctx, serviceTokenSubject, []string{"pull"}, expiry)
1441 if err != nil {
1442 return nil, log.WrapErr(err, "failed to generate registry service token")
1443 }
1444 log.Info().
1445 Str("subject", serviceTokenSubject).
1446 Str("expiry", expiry.String()).
1447 Msg("generated service token for container registry access")
1448 containerConfig.ServiceTokenUsername = serviceTokenSubject
1449 containerConfig.ServiceToken = serviceToken
1450 } else {
1451 log.Warn().Msg("registry auth disabled; container image pulls will use unauthenticated mode")
1452 }
1453
1454 return container.NewService(svc.runtime, svc.envLoader, svc.eventBus, svc.logWriter, containerConfig, svc.configSvc), nil
1455}
1456
1457func createBackupService(cfg Config, svc *services, log zerowrap.Logger) (*filesystem.BackupStorage, *backup.Service, error) {
1458 if !cfg.Backups.Enabled {
1459 return nil, nil, nil
1460 }
1461
1462 storageDir := cfg.Backups.StorageDir
1463 if storageDir == "" {
1464 dataDir := resolveDataDir(cfg.Server.DataDir)
1465 storageDir = filepath.Join(dataDir, "backups")
1466 }
1467
1468 backupStorage, err := filesystem.NewBackupStorage(storageDir, log)
1469 if err != nil {
1470 return nil, nil, log.WrapErr(err, "failed to create backup storage")
1471 }
1472
1473 retention, err := validateBackupRetention(cfg)
1474 if err != nil {
1475 return nil, nil, log.WrapErr(err, "invalid backup retention policy")
1476 }
1477
1478 backupCfg := domain.BackupConfig{
1479 Enabled: cfg.Backups.Enabled,
1480 StorageDir: storageDir,
1481 Retention: retention,
1482 }
1483
1484 backupSvc := backup.NewService(svc.runtime, backupStorage, svc.containerSvc, backupCfg, log)
1485
1486 log.Info().
1487 Str("storage_dir", storageDir).
1488 Msg("backup service initialized")
1489
1490 return backupStorage, backupSvc, nil
1491}
1492
1493func validateBackupRetention(cfg Config) (domain.RetentionPolicy, error) {
1494 if cfg.Backups.Retention.Hourly < 0 {
1495 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.hourly cannot be negative")
1496 }
1497 if cfg.Backups.Retention.Daily < 0 {
1498 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.daily cannot be negative")
1499 }
1500 if cfg.Backups.Retention.Weekly < 0 {
1501 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.weekly cannot be negative")
1502 }
1503 if cfg.Backups.Retention.Monthly < 0 {
1504 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.monthly cannot be negative")
1505 }
1506
1507 return domain.RetentionPolicy{
1508 Hourly: cfg.Backups.Retention.Hourly,
1509 Daily: cfg.Backups.Retention.Daily,
1510 Weekly: cfg.Backups.Retention.Weekly,
1511 Monthly: cfg.Backups.Retention.Monthly,
1512 }, nil
1513}
1514
1515// registerEventHandlers registers all event handlers.
1516func registerEventHandlers(ctx context.Context, svc *services, cfg Config) (func(), error) {
1517 imagePushedHandler := container.NewImagePushedHandler(ctx, svc.containerSvc, svc.configSvc)
1518 if err := svc.eventBus.Subscribe(imagePushedHandler); err != nil {
1519 return nil, fmt.Errorf("failed to subscribe image pushed handler: %w", err)
1520 }
1521
1522 // Auto-route handler for creating routes from image labels
1523 autoRouteHandler := container.NewAutoRouteHandler(ctx, svc.configSvc, svc.containerSvc, svc.blobStorage, cfg.Server.RegistryDomain).
1524 WithEnvExtractor(svc.runtime, svc.envDir)
1525
1526 // Preview handler for creating preview environments from tagged images
1527 autoPreviewHandler := preview.NewAutoPreviewHandler(
1528 ctx,
1529 svc.configSvc,
1530 svc.blobStorage,
1531 svc.previewService,
1532 )
1533
1534 // Dispatcher routes image push events to either auto-route or preview handler
1535 dispatcher := auto.NewImagePushDispatcher(svc.configSvc, autoRouteHandler, autoPreviewHandler)
1536 if err := svc.eventBus.Subscribe(dispatcher); err != nil {
1537 return nil, fmt.Errorf("subscribe image push dispatcher: %w", err)
1538 }
1539
1540 configReloadHandler := container.NewConfigReloadHandler(ctx, svc.containerSvc, svc.configSvc)
1541 if err := svc.eventBus.Subscribe(configReloadHandler); err != nil {
1542 return nil, fmt.Errorf("failed to subscribe config reload handler: %w", err)
1543 }
1544
1545 manualReloadHandler := container.NewManualReloadHandler(ctx, svc.containerSvc, svc.configSvc)
1546 if err := svc.eventBus.Subscribe(manualReloadHandler); err != nil {
1547 return nil, fmt.Errorf("failed to subscribe manual reload handler: %w", err)
1548 }
1549
1550 manualDeployHandler := container.NewManualDeployHandler(ctx, svc.containerSvc, svc.configSvc)
1551 if err := svc.eventBus.Subscribe(manualDeployHandler); err != nil {
1552 return nil, fmt.Errorf("failed to subscribe manual deploy handler: %w", err)
1553 }
1554
1555 secretsChangedHandler := container.NewSecretsChangedHandler(ctx, svc.containerSvc, svc.configSvc, container.DefaultSecretsDebounce)
1556 if err := svc.eventBus.Subscribe(secretsChangedHandler); err != nil {
1557 return nil, fmt.Errorf("failed to subscribe secrets changed handler: %w", err)
1558 }
1559
1560 // Proxy cache invalidation on config reload (clears stale targets for removed routes)
1561 configReloadProxyHandler := proxy.NewConfigReloadProxyHandler(ctx, svc.proxySvc)
1562 if err := svc.eventBus.Subscribe(configReloadProxyHandler); err != nil {
1563 return nil, fmt.Errorf("failed to subscribe config reload proxy handler: %w", err)
1564 }
1565
1566 cleanup := func() {
1567 secretsChangedHandler.Stop()
1568 }
1569
1570 return cleanup, nil
1571}
1572
1573// setupConfigHotReload sets up Viper config hot reload.
1574// NOTE: This does NOT reload routes into memory. Routes are managed via API
1575// (AddRoute/UpdateRoute/RemoveRoute) and memory is the source of truth.
1576// The file watcher only updates proxy config and refreshes targets.
1577// Manual config file edits to routes require a server restart.
1578func setupConfigHotReload(ctx context.Context, v *viper.Viper, svc *services, log zerowrap.Logger) {
1579 v.OnConfigChange(func(e fsnotify.Event) {
1580 log.Info().Str("file", e.Name).Msg("config file changed")
1581
1582 if err := v.ReadInConfig(); err != nil {
1583 log.Error().Err(err).Msg("failed to reload config")
1584 return
1585 }
1586
1587 // Re-unmarshal config and rebuild proxy config to pick up all changes
1588 var reloadCfg Config
1589 if err := v.Unmarshal(&reloadCfg); err != nil {
1590 log.Error().Err(err).Msg("failed to unmarshal config on reload")
1591 return
1592 }
1593 if reloadCfg.Server.GordonDomain != "" {
1594 reloadCfg.Server.RegistryDomain = reloadCfg.Server.GordonDomain
1595 }
1596 reloadedProxy, err := buildProxyConfig(reloadCfg, log)
1597 if err != nil {
1598 log.Error().Err(err).Msg("failed to parse proxy config on reload")
1599 return
1600 }
1601 svc.proxySvc.UpdateConfig(reloadedProxy.proxyConfig)
1602
1603 // Clear proxy target cache to pick up external route changes
1604 if err := svc.proxySvc.RefreshTargets(ctx); err != nil {
1605 log.Warn().Err(err).Msg("failed to refresh proxy targets")
1606 }
1607
1608 log.Debug().Msg("config hot reload complete (routes unchanged, memory is source of truth)")
1609 })
1610 v.WatchConfig()
1611}
1612
1613// syncAndAutoStart syncs existing containers and auto-starts if configured.
1614func syncAndAutoStart(ctx context.Context, svc *services, log zerowrap.Logger) {
1615 if err := svc.containerSvc.SyncContainers(ctx); err != nil {
1616 log.Warn().Err(err).Msg("failed to sync existing containers")
1617 }
1618
1619 if svc.configSvc.IsAutoRouteEnabled() {
1620 routes := svc.configSvc.GetRoutes(ctx)
1621 if err := svc.containerSvc.AutoStart(domain.WithInternalDeploy(ctx), routes); err != nil {
1622 log.Warn().Err(err).Msg("failed to auto-start containers")
1623 }
1624 }
1625
1626 // Start background monitor to restart crashed containers.
1627 svc.containerSvc.StartMonitor(ctx)
1628}
1629func loopbackOnly(next http.Handler, log zerowrap.Logger) http.Handler {
1630 return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
1631 host, _, err := net.SplitHostPort(r.RemoteAddr)
1632 if err != nil {
1633 host = r.RemoteAddr
1634 }
1635
1636 ip := net.ParseIP(host)
1637 if ip == nil || !ip.IsLoopback() {
1638 log.Warn().
1639 Str("path", r.URL.Path).
1640 Str("remote_addr", r.RemoteAddr).
1641 Msg("blocked non-loopback access on internal admin route")
1642 w.Header().Set("Content-Type", "application/json")
1643 w.WriteHeader(http.StatusForbidden)
1644 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "Forbidden"})
1645 return
1646 }
1647
1648 next.ServeHTTP(w, r)
1649 })
1650}
1651
1652// createHTTPHandlers creates HTTP handlers with middleware.
1653func createHTTPHandlers(svc *services, cfg Config, log zerowrap.Logger) (http.Handler, http.Handler) {
1654 // Parse trusted proxies once for all middleware chains.
1655 // This ensures consistent IP extraction across logging, rate limiting, and auth.
1656 trustedNets := middleware.ParseTrustedProxies(cfg.API.RateLimit.TrustedProxies)
1657
1658 registryHandler := registry.NewHandler(svc.registrySvc, log, svc.maxBlobChunkSize)
1659 registryWithMiddleware, cidrAllowlistMiddleware, rateLimitMiddleware := buildRegistryHandlerWithMiddleware(
1660 svc,
1661 cfg,
1662 trustedNets,
1663 registryHandler,
1664 log,
1665 )
1666
1667 registryMux := http.NewServeMux()
1668 registerAuthRoutes(registryMux, svc, trustedNets, cidrAllowlistMiddleware, rateLimitMiddleware, cfg, log)
1669 registryMux.Handle("/v2/", wrapRegistryForLocalMode(registryWithMiddleware, cfg, log))
1670
1671 // SECURITY: No CORS middleware on the proxy chain. Backend applications
1672 // should control their own CORS policies. A blanket Access-Control-Allow-Origin: *
1673 // would override backend CORS settings and allow any website to make
1674 // cross-origin authenticated requests to proxied applications.
1675 proxyMiddlewares := []func(http.Handler) http.Handler{
1676 middleware.PanicRecovery(log),
1677 middleware.RequestLogger(log, trustedNets),
1678 middleware.SecurityHeaders,
1679 }
1680
1681 proxyHandler := proxyadapter.NewHandler(svc.proxySvc, log)
1682 proxyWithMiddleware := otelhttp.NewHandler(
1683 middleware.Chain(proxyMiddlewares...)(proxyHandler),
1684 "gordon.proxy",
1685 )
1686 proxyMux := http.NewServeMux()
1687 proxyMux.Handle("/", proxyWithMiddleware)
1688
1689 registerAdminRoutes(registryMux, svc, cfg, trustedNets, log)
1690
1691 return registryMux, proxyMux
1692}
1693
1694func buildRegistryHandlerWithMiddleware(
1695 svc *services,
1696 cfg Config,
1697 trustedNets []*net.IPNet,
1698 registryHandler http.Handler,
1699 log zerowrap.Logger,
1700) (http.Handler, func(http.Handler) http.Handler, func(http.Handler) http.Handler) {
1701 registryMiddlewares := []func(http.Handler) http.Handler{
1702 middleware.PanicRecovery(log),
1703 middleware.RequestLogger(log, trustedNets),
1704 middleware.SecurityHeaders,
1705 }
1706
1707 cidrAllowlistMiddleware := buildRegistryCIDRAllowlistMiddleware(cfg, trustedNets, log)
1708 if cidrAllowlistMiddleware != nil {
1709 registryMiddlewares = append(registryMiddlewares, cidrAllowlistMiddleware)
1710 }
1711
1712 rateLimitMiddleware := buildRegistryRateLimitMiddleware(cfg, log)
1713 registryMiddlewares = append(registryMiddlewares, rateLimitMiddleware)
1714
1715 appendRegistryAuthMiddleware(®istryMiddlewares, svc, cfg, log)
1716
1717 registryWithOtel := otelhttp.NewHandler(
1718 middleware.Chain(registryMiddlewares...)(registryHandler),
1719 "gordon.registry",
1720 )
1721 return registryWithOtel, cidrAllowlistMiddleware, rateLimitMiddleware
1722}
1723
1724func buildRegistryCIDRAllowlistMiddleware(cfg Config, trustedNets []*net.IPNet, log zerowrap.Logger) func(http.Handler) http.Handler {
1725 if len(cfg.Server.RegistryAllowedIPs) == 0 {
1726 return nil
1727 }
1728
1729 allowedNets := middleware.ParseTrustedProxies(cfg.Server.RegistryAllowedIPs)
1730 if len(allowedNets) != len(cfg.Server.RegistryAllowedIPs) {
1731 for _, entry := range cfg.Server.RegistryAllowedIPs {
1732 if nets := middleware.ParseTrustedProxies([]string{entry}); len(nets) == 0 {
1733 log.Warn().Str("entry", entry).Msg("ignoring invalid registry_allowed_ips entry")
1734 }
1735 }
1736 }
1737
1738 if len(allowedNets) == 0 {
1739 log.Error().
1740 Strs("registry_allowed_ips", cfg.Server.RegistryAllowedIPs).
1741 Msg("registry_allowed_ips is set but no valid entries were parsed; registry will deny all traffic (fail-closed)")
1742 return func(next http.Handler) http.Handler {
1743 return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
1744 log.Warn().
1745 Str(zerowrap.FieldPath, r.URL.Path).
1746 Str(zerowrap.FieldClientIP, middleware.GetClientIP(r, trustedNets)).
1747 Msg("registry access denied due to invalid registry_allowed_ips configuration")
1748 w.Header().Set("Content-Type", "application/json")
1749 w.WriteHeader(http.StatusForbidden)
1750 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "Forbidden"})
1751 })
1752 }
1753 }
1754
1755 return middleware.RegistryCIDRAllowlist(allowedNets, trustedNets, log)
1756}
1757
1758func buildRegistryRateLimitMiddleware(cfg Config, log zerowrap.Logger) func(http.Handler) http.Handler {
1759 if cfg.API.RateLimit.Enabled {
1760 globalLimiter := ratelimit.NewMemoryStore(cfg.API.RateLimit.GlobalRPS, cfg.API.RateLimit.Burst, log)
1761 ipLimiter := ratelimit.NewMemoryStore(cfg.API.RateLimit.PerIPRPS, cfg.API.RateLimit.Burst, log)
1762 return registry.RateLimitMiddleware(
1763 globalLimiter,
1764 ipLimiter,
1765 cfg.API.RateLimit.TrustedProxies,
1766 log,
1767 )
1768 }
1769
1770 return registry.RateLimitMiddleware(nil, nil, nil, log)
1771}
1772
1773func appendRegistryAuthMiddleware(registryMiddlewares *[]func(http.Handler) http.Handler, svc *services, cfg Config, log zerowrap.Logger) {
1774 if svc.authSvc != nil {
1775 internalAuth := middleware.InternalRegistryAuth{
1776 Username: svc.internalRegUser,
1777 Password: svc.internalRegPass,
1778 }
1779 *registryMiddlewares = append(*registryMiddlewares, middleware.RegistryAuthV2(svc.authSvc, internalAuth, log))
1780 return
1781 }
1782
1783 if cfg.Auth.Enabled {
1784 log.Error().Msg("authentication service unavailable; registry requests will be denied")
1785 *registryMiddlewares = append(*registryMiddlewares, func(next http.Handler) http.Handler {
1786 return http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
1787 w.Header().Set("Content-Type", "application/json")
1788 w.WriteHeader(http.StatusServiceUnavailable)
1789 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "authentication service unavailable"})
1790 })
1791 })
1792 }
1793}
1794
1795func registerAuthRoutes(
1796 registryMux *http.ServeMux,
1797 svc *services,
1798 trustedNets []*net.IPNet,
1799 cidrAllowlistMiddleware func(http.Handler) http.Handler,
1800 rateLimitMiddleware func(http.Handler) http.Handler,
1801 cfg Config,
1802 log zerowrap.Logger,
1803) {
1804 if svc.authHandler == nil {
1805 return
1806 }
1807
1808 // Auth endpoints always get rate limiting, even if global rate limiting is disabled.
1809 // This prevents brute-force attacks against password/token endpoints.
1810 authRateLimitMiddleware := rateLimitMiddleware
1811 if !cfg.API.RateLimit.Enabled {
1812 authGlobalLimiter := ratelimit.NewMemoryStore(50, 100, log)
1813 authIPLimiter := ratelimit.NewMemoryStore(5, 10, log)
1814 authRateLimitMiddleware = registry.RateLimitMiddleware(authGlobalLimiter, authIPLimiter, cfg.API.RateLimit.TrustedProxies, log)
1815 }
1816
1817 // Auth endpoints are NOT protected by auth - they're where clients authenticate
1818 // but still need rate limiting to prevent brute force attacks.
1819 authMiddlewares := []func(http.Handler) http.Handler{
1820 middleware.PanicRecovery(log),
1821 middleware.RequestLogger(log, trustedNets),
1822 middleware.SecurityHeaders,
1823 }
1824 if cidrAllowlistMiddleware != nil {
1825 authMiddlewares = append(authMiddlewares, cidrAllowlistMiddleware)
1826 }
1827 authMiddlewares = append(authMiddlewares, authRateLimitMiddleware)
1828 authWithMiddleware := otelhttp.NewHandler(
1829 middleware.Chain(authMiddlewares...)(svc.authHandler),
1830 "gordon.auth",
1831 )
1832 registryMux.Handle("/auth/", authWithMiddleware)
1833}
1834
1835func wrapRegistryForLocalMode(registryWithMiddleware http.Handler, cfg Config, log zerowrap.Logger) http.Handler {
1836 if !cfg.Auth.Enabled {
1837 return loopbackOnly(registryWithMiddleware, log)
1838 }
1839 return registryWithMiddleware
1840}
1841
1842func registerAdminRoutes(registryMux *http.ServeMux, svc *services, cfg Config, trustedNets []*net.IPNet, log zerowrap.Logger) {
1843 if svc.adminHandler == nil {
1844 return
1845 }
1846
1847 if !cfg.Auth.Enabled {
1848 log.Warn().Msg("auth disabled: admin API endpoints are not registered")
1849 return
1850 }
1851
1852 adminMiddlewares := []func(http.Handler) http.Handler{
1853 middleware.PanicRecovery(log),
1854 middleware.RequestLogger(log, trustedNets),
1855 middleware.SecurityHeaders,
1856 }
1857
1858 if svc.authSvc != nil {
1859 // Create rate limiters for admin API - uses same config as registry.
1860 var globalLimiter, ipLimiter out.RateLimiter
1861 if cfg.API.RateLimit.Enabled {
1862 globalLimiter = ratelimit.NewMemoryStore(cfg.API.RateLimit.GlobalRPS, cfg.API.RateLimit.Burst, log)
1863 ipLimiter = ratelimit.NewMemoryStore(cfg.API.RateLimit.PerIPRPS, cfg.API.RateLimit.Burst, log)
1864 }
1865 adminMiddlewares = append(adminMiddlewares, admin.AuthMiddleware(svc.authSvc, globalLimiter, ipLimiter, trustedNets, log))
1866 } else {
1867 adminMiddlewares = append(adminMiddlewares, func(next http.Handler) http.Handler {
1868 return http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
1869 w.Header().Set("Content-Type", "application/json")
1870 w.WriteHeader(http.StatusServiceUnavailable)
1871 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "authentication service unavailable"})
1872 })
1873 })
1874 }
1875
1876 adminWithMiddleware := otelhttp.NewHandler(
1877 middleware.Chain(adminMiddlewares...)(svc.adminHandler),
1878 "gordon.admin",
1879 )
1880 registryMux.Handle("/admin/", loopbackOnly(adminWithMiddleware, log))
1881}
1882
1883// runServers starts the HTTP servers and waits for shutdown.
1884// Signal handling notes:
1885// - SIGINT/SIGTERM: Triggers graceful shutdown via signal.NotifyContext
1886// - SIGUSR1: Triggers config reload without restart
1887// - SIGUSR2: Triggers manual deploy for a specific route
1888// The deferred signal.Stop calls ensure signal handlers are properly
1889// cleaned up before program exit, preventing signal handler leaks.
1890func runServers(ctx context.Context, v *viper.Viper, cfg Config, registryHandler, proxyHandler http.Handler, svc *services, cleanupHandlers func(), log zerowrap.Logger) error {
1891 ctx, cancel := signal.NotifyContext(ctx, os.Interrupt, syscall.SIGTERM)
1892 defer cancel()
1893
1894 // Set up SIGUSR1 for reload.
1895 // Note: signal.Stop must be called (via defer) to release the channel
1896 // and prevent signal handler leaks when the function returns.
1897 reloadChan := make(chan os.Signal, 1)
1898 signal.Notify(reloadChan, syscall.SIGUSR1)
1899 defer signal.Stop(reloadChan)
1900
1901 // Set up SIGUSR2 for manual deploy.
1902 deployChan := make(chan os.Signal, 1)
1903 signal.Notify(deployChan, syscall.SIGUSR2)
1904 defer signal.Stop(deployChan)
1905
1906 errChan := make(chan error, 3)
1907
1908 registrySrv, registryReady := startServer(fmt.Sprintf(":%d", cfg.Server.RegistryPort), registryHandler, "registry", nil, errChan, log)
1909
1910 proxySrv, proxyReady, tlsSrv, tlsReady, err := startProxyServers(cfg, proxyHandler, errChan, log)
1911 if err != nil {
1912 return err
1913 }
1914
1915 // Wait for all enabled servers to bind their ports before auto-starting containers.
1916 // This prevents the race where auto-start pulls from the registry before it's listening.
1917 if err := waitForServerReady(registryReady, errChan); err != nil {
1918 return err
1919 }
1920 if err := waitForServerReady(proxyReady, errChan); err != nil {
1921 return err
1922 }
1923 if cfg.Server.TLSEnabled {
1924 if err := waitForServerReady(tlsReady, errChan); err != nil {
1925 return err
1926 }
1927 }
1928
1929 logEvent := log.Info().
1930 Int("proxy_port", cfg.Server.Port).
1931 Int("registry_port", cfg.Server.RegistryPort)
1932 if cfg.Server.TLSEnabled {
1933 logEvent = logEvent.Int("tls_port", cfg.Server.TLSPort)
1934 }
1935 logEvent.Msg("Gordon is running")
1936
1937 schedulerCleanup, err := startOptionalSchedulers(ctx, cfg, svc, log, v)
1938 if err != nil {
1939 return err
1940 }
1941 if schedulerCleanup != nil {
1942 defer schedulerCleanup()
1943 }
1944
1945 // Auto-start after servers are listening (registry port is now bound).
1946 syncAndAutoStart(ctx, svc, log)
1947
1948 waitForShutdown(ctx, errChan, reloadChan, deployChan, svc.eventBus, log)
1949 cleanupHandlers() // Stop debounce timers before draining containers
1950 gracefulShutdown(registrySrv, proxySrv, tlsSrv, svc.containerSvc, svc.proxySvc, log)
1951 return nil
1952}
1953
1954func waitForServerReady(ready <-chan struct{}, errChan <-chan error) error {
1955 select {
1956 case <-ready:
1957 return nil
1958 case err := <-errChan:
1959 return err
1960 }
1961}
1962
1963func startOptionalSchedulers(ctx context.Context, cfg Config, svc *services, log zerowrap.Logger, v *viper.Viper) (func(), error) {
1964 schedulers := make([]*cronSvc.Scheduler, 0, 2)
1965
1966 backupScheduler, err := startBackupScheduler(ctx, cfg, svc, log)
1967 if err != nil {
1968 return nil, err
1969 }
1970 if backupScheduler != nil {
1971 schedulers = append(schedulers, backupScheduler)
1972 }
1973
1974 imageScheduler, err := startImagePruneScheduler(ctx, cfg, svc, log, func() int {
1975 return v.GetInt("images.prune.keep_last")
1976 })
1977 if err != nil {
1978 return nil, err
1979 }
1980 if imageScheduler != nil {
1981 schedulers = append(schedulers, imageScheduler)
1982 }
1983
1984 if len(schedulers) == 0 {
1985 return nil, nil
1986 }
1987
1988 return func() {
1989 for i := len(schedulers) - 1; i >= 0; i-- {
1990 schedulers[i].Stop()
1991 }
1992 }, nil
1993}
1994
1995func startBackupScheduler(ctx context.Context, cfg Config, svc *services, log zerowrap.Logger) (*cronSvc.Scheduler, error) {
1996 if !cfg.Backups.Enabled || svc == nil || svc.backupSvc == nil {
1997 return nil, nil
1998 }
1999
2000 preset, err := resolveBackupSchedule(cfg.Backups.Schedule)
2001 if err != nil {
2002 return nil, err
2003 }
2004
2005 scheduler := cronSvc.NewScheduler(log)
2006 err = scheduler.Add(
2007 "backup-scheduler",
2008 "Backups",
2009 domain.CronSchedule{Preset: preset},
2010 func(jobCtx context.Context) error {
2011 if err := svc.backupSvc.RunForSchedule(jobCtx, preset); err != nil {
2012 return err
2013 }
2014 log.Info().
2015 Str("schedule", string(preset)).
2016 Msg("scheduled backup run complete")
2017 return nil
2018 },
2019 )
2020 if err != nil {
2021 return nil, log.WrapErr(err, "failed to register backup schedule")
2022 }
2023
2024 scheduler.Start(ctx)
2025 log.Info().
2026 Str("schedule", string(preset)).
2027 Msg("backup scheduler enabled")
2028
2029 return scheduler, nil
2030}
2031
2032func resolveBackupSchedule(raw string) (domain.BackupSchedule, error) {
2033 return resolveSchedulePreset(raw, "backups.schedule", domain.ScheduleDaily)
2034}
2035
2036func startImagePruneScheduler(ctx context.Context, cfg Config, svc *services, log zerowrap.Logger, keepLastGetter func() int) (*cronSvc.Scheduler, error) {
2037 if !cfg.Images.Prune.Enabled || svc == nil || svc.imageSvc == nil {
2038 return nil, nil
2039 }
2040 if keepLastGetter == nil {
2041 keepLastGetter = func() int { return cfg.Images.Prune.KeepLast }
2042 }
2043 if keepLastGetter() < 0 {
2044 return nil, fmt.Errorf("images.prune.keep_last must be >= 0")
2045 }
2046
2047 preset, err := resolveImagePruneSchedule(cfg.Images.Prune.Schedule)
2048 if err != nil {
2049 return nil, err
2050 }
2051
2052 scheduler := cronSvc.NewScheduler(log)
2053 err = scheduler.Add(
2054 "image-prune",
2055 "Image prune",
2056 domain.CronSchedule{Preset: preset},
2057 func(jobCtx context.Context) error {
2058 keepLast := keepLastGetter()
2059 if keepLast < 0 {
2060 log.Warn().
2061 Int("configured_keep_last", keepLast).
2062 Int("fallback_keep_last", domain.DefaultImagePruneKeepLast).
2063 Msg("invalid images.prune.keep_last; using default")
2064 keepLast = domain.DefaultImagePruneKeepLast
2065 }
2066
2067 report, err := svc.imageSvc.Prune(jobCtx, domain.ImagePruneOptions{
2068 KeepLast: keepLast,
2069 PruneDangling: true,
2070 PruneRegistry: true,
2071 })
2072 if err != nil {
2073 return err
2074 }
2075
2076 log.Info().
2077 Int("keep_last", keepLast).
2078 Int("runtime_deleted", report.Runtime.DeletedCount).
2079 Int64("runtime_reclaimed_bytes", report.Runtime.SpaceReclaimed).
2080 Int("registry_tags_removed", report.Registry.TagsRemoved).
2081 Int("registry_blobs_removed", report.Registry.BlobsRemoved).
2082 Int("registry_uploads_removed", report.Registry.UploadsRemoved).
2083 Int64("registry_upload_bytes_reclaimed", report.Registry.UploadSpaceReclaimed).
2084 Msg("scheduled image prune complete")
2085 return nil
2086 },
2087 )
2088 if err != nil {
2089 return nil, log.WrapErr(err, "failed to register image prune schedule")
2090 }
2091
2092 scheduler.Start(ctx)
2093 log.Info().
2094 Str("schedule", string(preset)).
2095 Int("keep_last", keepLastGetter()).
2096 Msg("image prune scheduler enabled")
2097
2098 return scheduler, nil
2099}
2100
2101func resolveImagePruneSchedule(raw string) (domain.BackupSchedule, error) {
2102 return resolveSchedulePreset(raw, "images.prune.schedule", domain.ScheduleDaily)
2103}
2104
2105func resolveSchedulePreset(raw, name string, defaultVal domain.BackupSchedule) (domain.BackupSchedule, error) {
2106 schedule := domain.BackupSchedule(strings.ToLower(strings.TrimSpace(raw)))
2107 if schedule == "" {
2108 schedule = defaultVal
2109 }
2110
2111 switch schedule {
2112 case domain.ScheduleHourly, domain.ScheduleDaily, domain.ScheduleWeekly, domain.ScheduleMonthly:
2113 return schedule, nil
2114 default:
2115 return "", fmt.Errorf("%s must be one of: hourly, daily, weekly, monthly", name)
2116 }
2117}
2118
2119// waitForShutdown blocks on the event loop, handling server errors and
2120// Unix signals (reload, deploy, shutdown) until the context is cancelled.
2121func waitForShutdown(ctx context.Context, errChan <-chan error, reloadChan, deployChan <-chan os.Signal, eventBus out.EventBus, log zerowrap.Logger) {
2122 for {
2123 select {
2124 case err := <-errChan:
2125 log.Error().Err(err).Msg("server error")
2126 return
2127 case <-reloadChan:
2128 log.Info().Msg("reload signal received (SIGUSR1)")
2129 if err := eventBus.Publish(domain.EventManualReload, nil); err != nil {
2130 log.Error().Err(err).Msg("failed to publish manual reload event")
2131 }
2132 case <-deployChan:
2133 log.Info().Msg("deploy signal received (SIGUSR2)")
2134 domainName, err := readDeployRequest()
2135 if err != nil {
2136 log.Error().Err(err).Msg("failed to read deploy request")
2137 continue
2138 }
2139 payload := &domain.ManualDeployPayload{Domain: domainName}
2140 if err := eventBus.Publish(domain.EventManualDeploy, payload); err != nil {
2141 log.Error().Err(err).Str("domain", domainName).Msg("failed to publish manual deploy event")
2142 }
2143 case <-ctx.Done():
2144 log.Info().Msg("shutdown signal received")
2145 return
2146 }
2147 }
2148}
2149
2150// gracefulShutdown stops HTTP servers with a 30s timeout, then shuts down
2151// the container service and cleans up runtime files.
2152func gracefulShutdown(registrySrv, proxySrv, tlsSrv *http.Server, containerSvc *container.Service, proxySvc *proxy.Service, log zerowrap.Logger) {
2153 log.Info().Msg("shutting down Gordon...")
2154
2155 shutdownCtx, shutdownCancel := context.WithTimeout(context.Background(), 30*time.Second)
2156 defer shutdownCancel()
2157
2158 // Phase 1: Stop ingress frontends (TLS, then proxy) — no new traffic accepted
2159 for _, srv := range []*http.Server{tlsSrv, proxySrv} {
2160 if srv == nil {
2161 continue
2162 }
2163 if err := srv.Shutdown(shutdownCtx); err != nil {
2164 log.Warn().Err(err).Str("addr", srv.Addr).Msg("server shutdown error")
2165 }
2166 }
2167
2168 // Phase 2: Drain in-flight registry push sessions before stopping the backend
2169 if proxySvc != nil {
2170 log.Info().Msg("draining in-flight registry requests...")
2171 if drained := proxySvc.DrainRegistryInFlight(25 * time.Second); !drained {
2172 log.Warn().Int64("in_flight", proxySvc.RegistryInFlight()).Msg("registry drain timed out; some in-flight pushes may be interrupted")
2173 }
2174 }
2175
2176 // Phase 3: Stop the registry backend
2177 if registrySrv != nil {
2178 if err := registrySrv.Shutdown(shutdownCtx); err != nil {
2179 log.Warn().Err(err).Str("addr", registrySrv.Addr).Msg("server shutdown error")
2180 }
2181 }
2182
2183 containerSvc.StopMonitor()
2184
2185 if err := containerSvc.Shutdown(shutdownCtx); err != nil {
2186 log.Warn().Err(err).Msg("error during container shutdown")
2187 }
2188
2189 cleanupInternalCredentials()
2190 log.Info().Msg("Gordon stopped")
2191}
2192
2193// startProxyServers sets up proxy server(s) depending on TLS configuration.
2194// With TLS enabled, the HTTP port becomes a redirect-only server and a separate TLS server handles traffic.
2195// Without TLS, a single server handles both HTTP/1.1 and cleartext HTTP/2.
2196func startProxyServers(cfg Config, proxyHandler http.Handler, errChan chan<- error, log zerowrap.Logger) (*http.Server, <-chan struct{}, *http.Server, <-chan struct{}, error) {
2197 if !cfg.Server.TLSEnabled {
2198 var proxyProtos http.Protocols
2199 proxyProtos.SetHTTP1(true)
2200 proxyProtos.SetUnencryptedHTTP2(true)
2201 srv, ready := startServer(fmt.Sprintf(":%d", cfg.Server.Port), proxyHandler, "proxy", &proxyProtos, errChan, log)
2202 return srv, ready, nil, nil, nil
2203 }
2204
2205 if cfg.Server.TLSCertFile == "" || cfg.Server.TLSKeyFile == "" {
2206 return nil, nil, nil, nil, fmt.Errorf("server.tls_enabled=true requires both server.tls_cert_file and server.tls_key_file")
2207 }
2208
2209 tlsPort := cfg.Server.TLSPort
2210 redirectHandler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
2211 host := r.Host
2212 if h, _, err := net.SplitHostPort(host); err == nil {
2213 host = h
2214 } else if strings.HasPrefix(host, "[") && strings.HasSuffix(host, "]") {
2215 host = strings.TrimSuffix(strings.TrimPrefix(host, "["), "]")
2216 }
2217 if tlsPort != 443 {
2218 host = net.JoinHostPort(host, strconv.Itoa(tlsPort))
2219 } else if strings.Contains(host, ":") {
2220 host = "[" + host + "]"
2221 }
2222 target := "https://" + host + r.RequestURI
2223 http.Redirect(w, r, target, http.StatusPermanentRedirect)
2224 })
2225
2226 proxySrv, proxyReady := startServer(fmt.Sprintf(":%d", cfg.Server.Port), redirectHandler, "proxy-redirect", nil, errChan, log)
2227 tlsSrv, tlsReady := startTLSServer(
2228 fmt.Sprintf(":%d", cfg.Server.TLSPort),
2229 proxyHandler,
2230 "proxy-tls",
2231 cfg.Server.TLSCertFile,
2232 cfg.Server.TLSKeyFile,
2233 errChan,
2234 log,
2235 )
2236 return proxySrv, proxyReady, tlsSrv, tlsReady, nil
2237}
2238
2239// startServer starts an HTTP server, returning the server instance and a channel
2240// that closes once the listening socket is bound. This lets callers wait for the
2241// port to be ready before taking actions that depend on it (e.g. auto-start
2242// pulling from the local registry). The returned *http.Server can be used for
2243// graceful shutdown.
2244func startServer(addr string, handler http.Handler, name string, protocols *http.Protocols, errChan chan<- error, log zerowrap.Logger) (*http.Server, <-chan struct{}) {
2245 ready := make(chan struct{})
2246
2247 server := &http.Server{
2248 Addr: addr,
2249 Handler: handler,
2250 Protocols: protocols,
2251 ReadHeaderTimeout: 10 * time.Second,
2252 ReadTimeout: 5 * time.Minute,
2253 WriteTimeout: 5 * time.Minute,
2254 IdleTimeout: 120 * time.Second,
2255 MaxHeaderBytes: 1 << 20,
2256 }
2257
2258 go func() {
2259 log.Info().Str("address", addr).Msgf("%s server starting", name)
2260
2261 ln, err := net.Listen("tcp", addr)
2262 if err != nil {
2263 errChan <- fmt.Errorf("%s server error: %w", name, err)
2264 return
2265 }
2266 close(ready) // signal: port is bound and accepting connections
2267
2268 if err := server.Serve(ln); err != nil && err != http.ErrServerClosed {
2269 errChan <- fmt.Errorf("%s server error: %w", name, err)
2270 }
2271 }()
2272
2273 return server, ready
2274}
2275
2276// startTLSServer starts an HTTPS server with the provided certificate and key.
2277// It returns the server instance (for graceful shutdown) and a channel that closes once the port is bound.
2278func startTLSServer(addr string, handler http.Handler, name, certFile, keyFile string, errChan chan<- error, log zerowrap.Logger) (*http.Server, <-chan struct{}) {
2279 ready := make(chan struct{})
2280
2281 server := &http.Server{
2282 Addr: addr,
2283 Handler: handler,
2284 ReadHeaderTimeout: 10 * time.Second,
2285 ReadTimeout: 5 * time.Minute,
2286 WriteTimeout: 5 * time.Minute,
2287 IdleTimeout: 120 * time.Second,
2288 MaxHeaderBytes: 1 << 20,
2289 }
2290
2291 go func() {
2292 log.Info().
2293 Str("address", addr).
2294 Str("cert_file", certFile).
2295 Str("key_file", keyFile).
2296 Msgf("%s server starting", name)
2297
2298 ln, err := net.Listen("tcp", addr)
2299 if err != nil {
2300 errChan <- fmt.Errorf("%s server error: %w", name, err)
2301 return
2302 }
2303 close(ready)
2304
2305 if err := server.ServeTLS(ln, certFile, keyFile); err != nil && err != http.ErrServerClosed {
2306 errChan <- fmt.Errorf("%s server error: %w", name, err)
2307 }
2308 }()
2309
2310 return server, ready
2311}
2312
2313// SendReloadSignal sends SIGUSR1 to the running Gordon process.
2314func SendReloadSignal() error {
2315 process, _, err := findRunningProcess()
2316 if err != nil {
2317 return err
2318 }
2319
2320 if err := process.Signal(syscall.SIGUSR1); err != nil {
2321 return fmt.Errorf("failed to send reload signal: %w", err)
2322 }
2323
2324 return nil
2325}
2326
2327// getDeployRequestFile returns the path to the deploy request file.
2328func getDeployRequestFile() string {
2329 return filepath.Join(os.TempDir(), "gordon-deploy-request")
2330}
2331
2332// writeDeployRequestFile creates the deploy request file exclusively with retry.
2333// This prevents race conditions when multiple deploy commands run simultaneously.
2334func writeDeployRequestFile(path string, data []byte, timeout time.Duration) error {
2335 deadline := time.Now().Add(timeout)
2336
2337 for {
2338 f, err := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0600)
2339 if err == nil {
2340 _, writeErr := f.Write(data)
2341 f.Close()
2342 if writeErr != nil {
2343 _ = os.Remove(path)
2344 return writeErr
2345 }
2346 return nil
2347 }
2348
2349 if os.IsExist(err) {
2350 if time.Now().After(deadline) {
2351 return fmt.Errorf("deploy request file still present after timeout; another deploy may be in progress")
2352 }
2353 time.Sleep(100 * time.Millisecond)
2354 continue
2355 }
2356
2357 return err
2358 }
2359}
2360
2361// SendDeploySignal triggers a manual deploy for a specific route via SIGUSR2.
2362// Returns the domain name on success for the caller to display.
2363func SendDeploySignal(domain string) (string, error) {
2364 deployFile := getDeployRequestFile()
2365 if err := writeDeployRequestFile(deployFile, []byte(domain), 5*time.Second); err != nil {
2366 return "", fmt.Errorf("failed to write deploy request: %w", err)
2367 }
2368
2369 // Find PID and send SIGUSR2
2370 process, _, err := findRunningProcess()
2371 if err != nil {
2372 _ = os.Remove(deployFile)
2373 return "", err
2374 }
2375
2376 if err := process.Signal(syscall.SIGUSR2); err != nil {
2377 _ = os.Remove(deployFile)
2378 return "", fmt.Errorf("failed to send deploy signal: %w", err)
2379 }
2380
2381 return domain, nil
2382}
2383
2384// readDeployRequest reads and removes the deploy request file atomically.
2385// Returns empty string if file doesn't exist (may have been consumed by another handler).
2386func readDeployRequest() (string, error) {
2387 deployFile := getDeployRequestFile()
2388
2389 // Rename to a temp file first to make the read-and-delete atomic
2390 tmpFile := deployFile + ".processing"
2391 if err := os.Rename(deployFile, tmpFile); err != nil {
2392 if os.IsNotExist(err) {
2393 return "", fmt.Errorf("deploy request file not found (may have been processed already)")
2394 }
2395 return "", fmt.Errorf("failed to acquire deploy request: %w", err)
2396 }
2397
2398 data, err := os.ReadFile(tmpFile)
2399 _ = os.Remove(tmpFile) // Always clean up
2400 if err != nil {
2401 return "", fmt.Errorf("failed to read deploy request: %w", err)
2402 }
2403
2404 return string(data), nil
2405}
2406
2407// createPidFile creates a PID file for the Gordon process.
2408// SECURITY: Prefers secure locations (XDG_RUNTIME_DIR, ~/.gordon/run) over /tmp
2409// to prevent symlink attacks and unauthorized access.
2410func createPidFile(log zerowrap.Logger) string {
2411 pid := os.Getpid()
2412
2413 // SECURITY: Prioritize secure locations over /tmp
2414 var locations []string
2415
2416 // Try secure runtime directory first
2417 if runtimeDir, err := getSecureRuntimeDir(); err == nil {
2418 locations = append(locations, filepath.Join(runtimeDir, "gordon.pid"))
2419 }
2420
2421 // Fall back to home directory
2422 if homeDir, err := os.UserHomeDir(); err == nil {
2423 locations = append(locations, filepath.Join(homeDir, ".gordon", "gordon.pid"))
2424 }
2425
2426 // Last resort: /tmp (least secure due to world-writable)
2427 locations = append(locations, filepath.Join(os.TempDir(), "gordon.pid"))
2428
2429 for _, location := range locations {
2430 // Ensure parent directory exists with secure permissions
2431 if err := os.MkdirAll(filepath.Dir(location), 0700); err != nil {
2432 continue
2433 }
2434 if err := os.WriteFile(location, []byte(fmt.Sprintf("%d", pid)), 0600); err == nil {
2435 log.Debug().Str("pid_file", location).Int("pid", pid).Msg("created PID file")
2436 return location
2437 }
2438 }
2439
2440 log.Warn().Int("pid", pid).Msg("failed to create PID file in any location")
2441 return ""
2442}
2443
2444// removePidFile removes the PID file.
2445func removePidFile(pidFile string, log zerowrap.Logger) {
2446 if err := os.Remove(pidFile); err != nil {
2447 log.Warn().Err(err).Str("pid_file", pidFile).Msg("failed to remove PID file")
2448 } else {
2449 log.Debug().Str("pid_file", pidFile).Msg("removed PID file")
2450 }
2451}
2452
2453func pidFileLocations() []string {
2454 var locations []string
2455 seen := make(map[string]struct{})
2456 add := func(path string) {
2457 if path == "" {
2458 return
2459 }
2460 if _, ok := seen[path]; ok {
2461 return
2462 }
2463 seen[path] = struct{}{}
2464 locations = append(locations, path)
2465 }
2466
2467 // Check secure runtime directory first
2468 if runtimeDir, err := getSecureRuntimeDir(); err == nil {
2469 add(filepath.Join(runtimeDir, "gordon.pid"))
2470 }
2471
2472 // Also check canonical /run/user/<uid> runtime path. This handles cases where
2473 // Gordon started under systemd user services with runtime dir available, but
2474 // CLI invocations (e.g. non-interactive SSH) don't have XDG_RUNTIME_DIR set.
2475 runtimeByUID := filepath.Join("/run/user", strconv.Itoa(os.Getuid()), "gordon", "gordon.pid")
2476 add(runtimeByUID)
2477
2478 // Check explicit XDG_RUNTIME_DIR if present in this process env.
2479 if runtimeDir := os.Getenv("XDG_RUNTIME_DIR"); runtimeDir != "" {
2480 add(filepath.Join(runtimeDir, "gordon.pid"))
2481 }
2482
2483 // Check home directory
2484 if homeDir, err := os.UserHomeDir(); err == nil {
2485 add(filepath.Join(homeDir, ".gordon", "gordon.pid"))
2486 // Legacy location for backward compatibility
2487 add(filepath.Join(homeDir, ".gordon.pid"))
2488 }
2489
2490 // Legacy /tmp locations for backward compatibility
2491 add(filepath.Join(os.TempDir(), "gordon.pid"))
2492 add("/tmp/gordon.pid")
2493
2494 return locations
2495}
2496
2497// findRunningPidFile returns the first PID file whose PID belongs to a live process.
2498// Stale/invalid PID files are ignored and removed when possible.
2499func findRunningPidFile() (string, int, error) {
2500 return findRunningPidFileInLocations(pidFileLocations())
2501}
2502
2503func findRunningPidFileInLocations(locations []string) (string, int, error) {
2504 foundAny := false
2505
2506 for _, location := range locations {
2507 pidBytes, err := os.ReadFile(location)
2508 if err != nil {
2509 continue
2510 }
2511
2512 foundAny = true
2513
2514 var pid int
2515 if _, err := fmt.Sscanf(string(pidBytes), "%d", &pid); err != nil || pid <= 0 {
2516 _ = os.Remove(location)
2517 continue
2518 }
2519
2520 if isProcessAlive(pid) {
2521 return location, pid, nil
2522 }
2523
2524 _ = os.Remove(location)
2525 }
2526
2527 if foundAny {
2528 return "", 0, fmt.Errorf("found stale gordon PID file(s), is Gordon running?")
2529 }
2530
2531 return "", 0, fmt.Errorf("gordon PID file not found, is Gordon running?")
2532}
2533
2534func findRunningProcess() (*os.Process, int, error) {
2535 _, pid, err := findRunningPidFile()
2536 if err != nil {
2537 return nil, 0, err
2538 }
2539
2540 process, err := os.FindProcess(pid)
2541 if err != nil {
2542 return nil, 0, fmt.Errorf("failed to find process: %w", err)
2543 }
2544
2545 return process, pid, nil
2546}
2547
2548func isProcessAlive(pid int) bool {
2549 process, err := os.FindProcess(pid)
2550 if err != nil {
2551 return false
2552 }
2553
2554 err = process.Signal(syscall.Signal(0))
2555 if err == nil {
2556 return true
2557 }
2558
2559 return errors.Is(err, syscall.EPERM)
2560}
2561
2562// loadConfig loads configuration from file and sets defaults.
2563func loadConfig(v *viper.Viper, configPath string) error {
2564 v.SetDefault("server.port", 80)
2565 v.SetDefault("server.registry_port", 5000)
2566 v.SetDefault("server.tls_enabled", false)
2567 v.SetDefault("server.tls_port", 443)
2568 v.SetDefault("server.tls_cert_file", "")
2569 v.SetDefault("server.tls_key_file", "")
2570 v.SetDefault("server.data_dir", DefaultDataDir())
2571 v.SetDefault("server.runtime", "auto")
2572 v.SetDefault("logging.level", "info")
2573 v.SetDefault("logging.format", "console")
2574 v.SetDefault("logging.file.enabled", false)
2575 v.SetDefault("logging.file.max_size", 100)
2576 v.SetDefault("logging.file.max_backups", 3)
2577 v.SetDefault("logging.file.max_age", 28)
2578 v.SetDefault("logging.container_logs.enabled", true)
2579 v.SetDefault("logging.container_logs.dir", "")
2580 v.SetDefault("logging.container_logs.max_size", 100)
2581 v.SetDefault("logging.container_logs.max_backups", 3)
2582 v.SetDefault("logging.container_logs.max_age", 28)
2583 v.SetDefault("env.dir", "") // defaults to {data_dir}/env when empty
2584 v.SetDefault("auth.enabled", true)
2585 // Note: auth.type defaults to "token" (the only supported mode)
2586 v.SetDefault("auth.secrets_backend", "")
2587 v.SetDefault("auth.token_expiry", "720h")
2588 v.SetDefault("api.rate_limit.enabled", true)
2589 v.SetDefault("api.rate_limit.global_rps", 500)
2590 v.SetDefault("api.rate_limit.per_ip_rps", 50)
2591 v.SetDefault("api.rate_limit.burst", 100)
2592 v.SetDefault("auto_route.enabled", false)
2593 v.SetDefault("network_isolation.enabled", false)
2594 v.SetDefault("network_isolation.network_prefix", "gordon")
2595 v.SetDefault("volumes.auto_create", true)
2596 v.SetDefault("volumes.prefix", "gordon")
2597 v.SetDefault("volumes.preserve", true)
2598 v.SetDefault("deploy.pull_policy", container.PullPolicyIfTagChanged)
2599 v.SetDefault("backups.enabled", false)
2600 v.SetDefault("backups.schedule", string(domain.ScheduleDaily))
2601 v.SetDefault("backups.storage_dir", "")
2602 v.SetDefault("backups.retention.hourly", 0)
2603 v.SetDefault("backups.retention.daily", 0)
2604 v.SetDefault("backups.retention.weekly", 0)
2605 v.SetDefault("backups.retention.monthly", 0)
2606 v.SetDefault("images.prune.enabled", false)
2607 v.SetDefault("images.prune.schedule", string(domain.ScheduleDaily))
2608 v.SetDefault("images.prune.keep_last", domain.DefaultImagePruneKeepLast)
2609 v.SetDefault("telemetry.enabled", false)
2610 v.SetDefault("telemetry.endpoint", "")
2611 v.SetDefault("telemetry.auth_token", "")
2612 v.SetDefault("telemetry.traces", true)
2613 v.SetDefault("telemetry.metrics", true)
2614 v.SetDefault("telemetry.logs", true)
2615 v.SetDefault("telemetry.trace_sample_rate", 1.0)
2616
2617 v.SetDefault("server.max_concurrent_connections", -1) // -1 = use default (10000), 0 = no limit
2618 v.SetDefault("server.registry_allowed_ips", []string{})
2619 v.SetDefault("deploy.readiness_delay", "5s")
2620 v.SetDefault("deploy.readiness_mode", "auto")
2621 v.SetDefault("deploy.health_timeout", "90s")
2622 v.SetDefault("deploy.stabilization_delay", "2s")
2623 v.SetDefault("deploy.tcp_probe_timeout", "30s")
2624 v.SetDefault("deploy.http_probe_timeout", "60s")
2625 v.SetDefault("deploy.attachment_readiness_timeout", "30s")
2626 v.SetDefault("deploy.drain_mode", "auto")
2627 v.SetDefault("deploy.drain_timeout", "30s")
2628
2629 ConfigureViper(v, configPath)
2630
2631 if err := v.ReadInConfig(); err != nil {
2632 if _, ok := err.(viper.ConfigFileNotFoundError); !ok {
2633 return fmt.Errorf("failed to read config file: %w", err)
2634 }
2635 }
2636
2637 v.SetEnvPrefix("GORDON")
2638 v.SetEnvKeyReplacer(strings.NewReplacer(".", "_"))
2639 v.AutomaticEnv()
2640
2641 return nil
2642}