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 svc.previewService.StartTicker(ctx, 60*time.Second, func(ctx context.Context, p domain.PreviewRoute) {
591 log := zerowrap.FromCtx(ctx)
592 for _, containerName := range p.Containers {
593 if err := svc.runtime.StopContainer(ctx, containerName); err != nil {
594 log.Warn().Err(err).Str("container", containerName).Str("preview", p.Domain).Msg("failed to stop preview container")
595 }
596 if err := svc.runtime.RemoveContainer(ctx, containerName, true); err != nil {
597 log.Warn().Err(err).Str("container", containerName).Str("preview", p.Domain).Msg("failed to remove preview container")
598 }
599 }
600 for _, volName := range p.Volumes {
601 if err := svc.runtime.RemoveVolume(ctx, volName, true); err != nil {
602 log.Warn().Err(err).Str("volume", volName).Str("preview", p.Domain).Msg("failed to remove preview volume")
603 }
604 }
605 // Remove route from config so proxy stops routing to this domain
606 if err := svc.configSvc.RemoveRoute(ctx, p.Domain); err != nil {
607 if errors.Is(err, domain.ErrRouteNotFound) {
608 log.Debug().Str("domain", p.Domain).Msg("preview route already removed from config")
609 } else {
610 log.Warn().Err(err).Str("domain", p.Domain).Msg("failed to remove preview route from config")
611 }
612 }
613 svc.proxySvc.InvalidateTarget(ctx, p.Domain)
614 })
615}
616
617// injectTelemetryMetrics creates and injects OTel metrics into services when
618// telemetry is enabled. Skipped otherwise to avoid unnecessary allocations.
619func injectTelemetryMetrics(cfg Config, svc *services, log zerowrap.Logger) {
620 if !cfg.Telemetry.Enabled || !cfg.Telemetry.Metrics {
621 return
622 }
623 gordonMetrics, err := telemetry.NewMetrics()
624 if err != nil {
625 log.Warn().Err(err).Msg("failed to create telemetry metrics, continuing without metrics")
626 return
627 }
628 svc.containerSvc.SetMetrics(gordonMetrics)
629 svc.registrySvc.SetMetrics(gordonMetrics)
630 svc.eventBus.SetMetrics(gordonMetrics)
631}
632
633func setupInternalRegistryAuth(svc *services, log zerowrap.Logger) error {
634 var err error
635 svc.internalRegUser, svc.internalRegPass, err = generateInternalRegistryAuth()
636 if err != nil {
637 return log.WrapErr(err, "failed to generate internal registry credentials")
638 }
639
640 // Persist credentials to file for CLI access (gordon auth internal)
641 if err := persistInternalCredentials(svc.internalRegUser, svc.internalRegPass); err != nil {
642 log.Warn().Err(err).Msg("failed to persist internal credentials for CLI access")
643 }
644
645 log.Debug().Msg("internal registry auth generated for loopback pulls")
646 return nil
647}
648
649func createDomainSecretStore(cfg Config, log zerowrap.Logger) (string, domain.SecretsBackend, *domainsecrets.PassStore, out.DomainSecretStore, error) {
650 envDir := resolveEnvDir(cfg)
651 backend, err := resolveSecretsBackend(cfg.Auth.SecretsBackend)
652 if err != nil {
653 return "", "", nil, nil, log.WrapErr(err, "failed to resolve secrets backend")
654 }
655
656 switch backend {
657 case domain.SecretsBackendPass:
658 passStore, err := domainsecrets.NewPassStore(log)
659 if err != nil {
660 return "", backend, nil, nil, log.WrapErr(err, "failed to create pass domain secret store")
661 }
662 if err := migrateEnvFilesToPass(envDir, passStore, log); err != nil {
663 return "", backend, nil, nil, log.WrapErr(err, "failed to migrate env files to pass")
664 }
665 return envDir, backend, passStore, passStore, nil
666 default:
667 store, err := domainsecrets.NewFileStore(envDir, log)
668 if err != nil {
669 return "", backend, nil, nil, log.WrapErr(err, "failed to create domain secret store")
670 }
671 return envDir, backend, nil, store, nil
672 }
673}
674
675// resolveLogFilePath returns the configured log file path or a default.
676func resolveLogFilePath(cfg Config) string {
677 if cfg.Logging.File.Path != "" {
678 return cfg.Logging.File.Path
679 }
680 dataDir := cfg.Server.DataDir
681 if dataDir == "" {
682 dataDir = DefaultDataDir()
683 }
684 return filepath.Join(dataDir, "logs", "gordon.log")
685}
686
687// resolveRuntimeConfig converts a server.runtime config value to a socket path.
688// "auto" or "" means auto-detect; any other value is treated as an explicit socket path.
689// URI schemes (unix://, tcp://) are stripped so callers receive a bare path.
690func resolveRuntimeConfig(value string) string {
691 if value == "" || value == "auto" {
692 return ""
693 }
694 if strings.HasPrefix(value, "unix://") {
695 return strings.TrimPrefix(value, "unix://")
696 }
697 return value
698}
699
700// createOutputAdapters creates the container runtime and event bus.
701func createOutputAdapters(ctx context.Context, log zerowrap.Logger, runtimeSocket string) (*docker.Runtime, *eventbus.InMemory, error) {
702 detection := docker.DetectRuntimeSocket(runtimeSocket)
703
704 var runtime *docker.Runtime
705 var err error
706
707 switch detection.Source {
708 case "none":
709 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")
710 case "DOCKER_HOST_passthrough":
711 detection.RuntimeName = "docker"
712 runtime, err = docker.NewRuntime()
713 default:
714 if detection.SocketPath != "" {
715 runtime, err = docker.NewRuntimeWithSocket(detection.SocketPath)
716 } else {
717 runtime, err = docker.NewRuntime()
718 }
719 }
720 if err != nil {
721 return nil, nil, log.WrapErr(err, "failed to create container runtime")
722 }
723
724 if err := runtime.Ping(ctx); err != nil {
725 return nil, nil, log.WrapErr(err, fmt.Sprintf("container runtime not available (detected: %s via %s)", detection.RuntimeName, detection.Source))
726 }
727
728 runtimeVersion, _ := runtime.Version(ctx)
729 log.Info().
730 Str("runtime", detection.RuntimeName).
731 Str("version", runtimeVersion).
732 Str("source", detection.Source).
733 Msg("container runtime initialized")
734
735 eventBus := eventbus.NewInMemory(100, log)
736
737 return runtime, eventBus, nil
738}
739
740// createStorage creates blob and manifest storage.
741func createStorage(cfg Config, log zerowrap.Logger) (*filesystem.BlobStorage, *filesystem.ManifestStorage, error) {
742 dataDir := cfg.Server.DataDir
743 if dataDir == "" {
744 dataDir = DefaultDataDir()
745 }
746
747 registryDir := filepath.Join(dataDir, "registry")
748
749 blobStorage, err := filesystem.NewBlobStorage(registryDir, log)
750 if err != nil {
751 return nil, nil, log.WrapErr(err, "failed to create blob storage")
752 }
753
754 manifestStorage, err := filesystem.NewManifestStorage(registryDir, log)
755 if err != nil {
756 return nil, nil, log.WrapErr(err, "failed to create manifest storage")
757 }
758
759 return blobStorage, manifestStorage, nil
760}
761
762// createEnvLoader creates the environment loader with secret providers.
763func createEnvLoader(backend domain.SecretsBackend, envDir string, passStore *domainsecrets.PassStore, log zerowrap.Logger) (out.EnvLoader, error) {
764 switch backend {
765 case domain.SecretsBackendPass:
766 loader, err := envloader.NewPassLoader(passStore, log)
767 if err != nil {
768 return nil, log.WrapErr(err, "failed to create pass env loader")
769 }
770 return loader, nil
771 default:
772 loader, err := envloader.NewFileLoader(envDir, log)
773 if err != nil {
774 return nil, log.WrapErr(err, "failed to create env loader")
775 }
776
777 // Register secret providers
778 passProvider := secrets.NewPassProvider(log)
779 if passProvider.IsAvailable() {
780 loader.RegisterSecretProvider(passProvider)
781 log.Debug().Msg("pass secret provider registered")
782 }
783
784 sopsProvider := secrets.NewSopsProvider(log)
785 if sopsProvider.IsAvailable() {
786 loader.RegisterSecretProvider(sopsProvider)
787 log.Debug().Msg("sops secret provider registered")
788 }
789
790 return loader, nil
791 }
792}
793
794// createLogWriter creates the container log writer.
795func createLogWriter(cfg Config, log zerowrap.Logger) (*logwriter.LogWriter, error) {
796 if !cfg.Logging.ContainerLogs.Enabled {
797 log.Debug().Msg("container log collection disabled")
798 return nil, nil
799 }
800
801 // Determine log directory
802 logDir := cfg.Logging.ContainerLogs.Dir
803 if logDir == "" {
804 dataDir := cfg.Server.DataDir
805 if dataDir == "" {
806 dataDir = DefaultDataDir()
807 }
808 logDir = filepath.Join(dataDir, "logs", "containers")
809 }
810
811 writer, err := logwriter.New(logwriter.Config{
812 Dir: logDir,
813 MaxSize: cfg.Logging.ContainerLogs.MaxSize,
814 MaxBackups: cfg.Logging.ContainerLogs.MaxBackups,
815 MaxAge: cfg.Logging.ContainerLogs.MaxAge,
816 })
817 if err != nil {
818 return nil, log.WrapErr(err, "failed to create container log writer")
819 }
820
821 log.Info().Str("dir", logDir).Msg("container log collection enabled")
822 return writer, nil
823}
824
825const (
826 internalRegistryUsername = "gordon-internal"
827 serviceTokenSubject = "gordon-service"
828 serviceTokenDefaultTTL = 30 * 24 * time.Hour
829)
830
831func generateInternalRegistryAuth() (string, string, error) {
832 password, err := randomTokenHex(32)
833 if err != nil {
834 return "", "", err
835 }
836 return internalRegistryUsername, password, nil
837}
838
839func randomTokenHex(size int) (string, error) {
840 buf := make([]byte, size)
841 if _, err := rand.Read(buf); err != nil {
842 return "", err
843 }
844 return hex.EncodeToString(buf), nil
845}
846
847// InternalCredentials holds the internal registry credentials for CLI access.
848type InternalCredentials struct {
849 Username string `json:"username"`
850 Password string `json:"password"`
851}
852
853// getSecureRuntimeDir returns a secure directory for runtime files.
854// Priority: XDG_RUNTIME_DIR > ~/.gordon/run
855func getSecureRuntimeDir() (string, error) {
856 // Try XDG_RUNTIME_DIR first (typically /run/user/<uid> on Linux)
857 if runtimeDir := os.Getenv("XDG_RUNTIME_DIR"); runtimeDir != "" {
858 gordonDir := filepath.Join(runtimeDir, "gordon")
859 if err := os.MkdirAll(gordonDir, 0700); err == nil {
860 return gordonDir, nil
861 }
862 }
863
864 // Fall back to ~/.gordon/run
865 homeDir, err := os.UserHomeDir()
866 if err != nil {
867 return "", fmt.Errorf("failed to get home directory: %w", err)
868 }
869
870 gordonDir := filepath.Join(homeDir, ".gordon", "run")
871 if err := os.MkdirAll(gordonDir, 0700); err != nil {
872 return "", fmt.Errorf("failed to create runtime directory: %w", err)
873 }
874
875 return gordonDir, nil
876}
877
878// getInternalCredentialsFile returns the path to the internal credentials file.
879// SECURITY: Credentials are stored in a secure location with restricted permissions.
880func getInternalCredentialsFile() string {
881 runtimeDir, err := getSecureRuntimeDir()
882 if err != nil {
883 // Fall back to temp dir if we can't get secure dir (shouldn't happen)
884 return filepath.Join(os.TempDir(), "gordon-internal-creds.json")
885 }
886 return filepath.Join(runtimeDir, "internal-creds.json")
887}
888
889// persistInternalCredentials saves the internal registry credentials to a secure file.
890// SECURITY: Credentials are stored in XDG_RUNTIME_DIR or ~/.gordon/run with 0600 permissions.
891// The file is cleaned up on graceful shutdown but may persist if Gordon crashes.
892// These credentials are for internal loopback communication only and are regenerated on each start.
893func persistInternalCredentials(username, password string) error {
894 creds := InternalCredentials{
895 Username: username,
896 Password: password,
897 }
898 data, err := json.Marshal(creds)
899 if err != nil {
900 return fmt.Errorf("failed to marshal credentials: %w", err)
901 }
902
903 credFile := getInternalCredentialsFile()
904
905 // Ensure parent directory exists with secure permissions
906 if err := os.MkdirAll(filepath.Dir(credFile), 0700); err != nil {
907 return fmt.Errorf("failed to create credentials directory: %w", err)
908 }
909
910 // Write file with restrictive permissions (owner read/write only)
911 if err := os.WriteFile(credFile, data, 0600); err != nil {
912 return fmt.Errorf("failed to write credentials file: %w", err)
913 }
914 return nil
915}
916
917// cleanupInternalCredentials removes the internal credentials file.
918func cleanupInternalCredentials() {
919 _ = os.Remove(getInternalCredentialsFile())
920}
921
922// getInternalCredentialsCandidates returns candidate file paths in priority order:
923// 1. XDG_RUNTIME_DIR/gordon/ (set by systemd for the daemon)
924// 2. /run/user/<uid>/gordon/ (well-known systemd default, for CLI in shells without XDG_RUNTIME_DIR)
925// 3. ~/.gordon/run/ (fallback for non-systemd environments)
926// 4. os.TempDir() (last resort, matches getInternalCredentialsFile fallback path)
927func getInternalCredentialsCandidates() []string {
928 var candidates []string
929
930 // 1. XDG_RUNTIME_DIR (set in daemon's environment)
931 if runtimeDir := os.Getenv("XDG_RUNTIME_DIR"); runtimeDir != "" {
932 candidates = append(candidates, filepath.Join(runtimeDir, "gordon", "internal-creds.json"))
933 }
934
935 // 2. /run/user/<uid>/gordon/ (systemd default, may not be in CLI's env)
936 uid := os.Getuid()
937 sysRuntime := filepath.Join("/run/user", fmt.Sprintf("%d", uid), "gordon", "internal-creds.json")
938 // Avoid duplicate if XDG_RUNTIME_DIR already points here
939 if len(candidates) == 0 || candidates[0] != sysRuntime {
940 candidates = append(candidates, sysRuntime)
941 }
942
943 // 3. ~/.gordon/run/ fallback
944 if homeDir, err := os.UserHomeDir(); err == nil {
945 candidates = append(candidates, filepath.Join(homeDir, ".gordon", "run", "internal-creds.json"))
946 }
947
948 // 4. os.TempDir() last resort — matches the fallback path in getInternalCredentialsFile,
949 // ensuring GetInternalCredentials can find credentials even when getSecureRuntimeDir fails.
950 candidates = append(candidates, filepath.Join(os.TempDir(), "gordon-internal-creds.json"))
951
952 return candidates
953}
954
955// GetInternalCredentialsFromCandidates reads credentials from the first candidate file that exists.
956// Exported for testing.
957func GetInternalCredentialsFromCandidates(candidates []string) (*InternalCredentials, error) {
958 var lastErr error
959 for _, path := range candidates {
960 data, err := os.ReadFile(path)
961 if os.IsNotExist(err) {
962 continue
963 }
964 if err != nil {
965 // Non-permission errors (e.g. EACCES) may be transient or path-specific;
966 // record and try the next candidate rather than failing immediately.
967 lastErr = fmt.Errorf("failed to read credentials file %s: %w", path, err)
968 continue
969 }
970 var creds InternalCredentials
971 if err := json.Unmarshal(data, &creds); err != nil {
972 // Corrupt file — record and fall through to lower-priority candidates.
973 lastErr = fmt.Errorf("failed to parse credentials at %s: %w", path, err)
974 continue
975 }
976 return &creds, nil
977 }
978 if lastErr != nil {
979 return nil, lastErr
980 }
981 return nil, fmt.Errorf("no credentials file found (is Gordon running?): checked %v", candidates)
982}
983
984// GetInternalCredentials reads the internal registry credentials from file.
985// Probes all candidate runtime directories so CLI works regardless of whether
986// XDG_RUNTIME_DIR is set in the current shell environment.
987func GetInternalCredentials() (*InternalCredentials, error) {
988 return GetInternalCredentialsFromCandidates(getInternalCredentialsCandidates())
989}
990
991// createAuthService creates the authentication service and token store.
992func createAuthService(ctx context.Context, cfg Config, log zerowrap.Logger) (out.TokenStore, *auth.Service, error) {
993 if !cfg.Auth.Enabled {
994 log.Warn().Msg("auth.enabled=false detected: running in local-only mode (registry loopback-only, admin API disabled)")
995 return nil, nil, nil
996 }
997
998 authType, err := resolveAuthType(cfg)
999 if err != nil {
1000 return nil, nil, fmt.Errorf("resolve auth type: %w", err)
1001 }
1002 backend, err := resolveSecretsBackend(cfg.Auth.SecretsBackend)
1003 if err != nil {
1004 return nil, nil, log.WrapErr(err, "failed to resolve secrets backend")
1005 }
1006 dataDir := resolveDataDir(cfg.Server.DataDir)
1007
1008 store, err := createTokenStore(backend, dataDir, log)
1009 if err != nil {
1010 return nil, nil, err
1011 }
1012
1013 authConfig, err := buildAuthConfig(ctx, cfg, authType, backend, dataDir, log)
1014 if err != nil {
1015 return nil, nil, err
1016 }
1017
1018 authSvc := auth.NewService(authConfig, store, log)
1019
1020 log.Info().
1021 Str("type", string(authType)).
1022 Str("backend", string(backend)).
1023 Msg("registry authentication enabled")
1024
1025 return store, authSvc, nil
1026}
1027
1028// resolveAuthType determines the auth type from config.
1029// Token-only authentication is the only supported mode.
1030func resolveAuthType(cfg Config) (domain.AuthType, error) {
1031 if cfg.Auth.Type != "" && cfg.Auth.Type != "token" {
1032 return "", fmt.Errorf("unsupported auth.type %q; only \"token\" is supported", cfg.Auth.Type)
1033 }
1034 return domain.AuthTypeToken, nil
1035}
1036
1037func resolveSecretsBackend(backend string) (domain.SecretsBackend, error) {
1038 switch backend {
1039 case "pass":
1040 return domain.SecretsBackendPass, nil
1041 case "sops":
1042 return domain.SecretsBackendSops, nil
1043 case "unsafe":
1044 return domain.SecretsBackendUnsafe, nil
1045 case "":
1046 return "", fmt.Errorf("auth.secrets_backend is required")
1047 default:
1048 return "", fmt.Errorf("unsupported auth.secrets_backend %q", backend)
1049 }
1050}
1051
1052func resolveDataDir(dataDir string) string {
1053 if dataDir == "" {
1054 return DefaultDataDir()
1055 }
1056 return dataDir
1057}
1058
1059func resolveEnvDir(cfg Config) string {
1060 dataDir := resolveDataDir(cfg.Server.DataDir)
1061 envDir := cfg.Env.Dir
1062 if envDir == "" {
1063 envDir = filepath.Join(dataDir, "env")
1064 }
1065 return envDir
1066}
1067
1068func createTokenStore(backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) (out.TokenStore, error) {
1069 // Token store is always created since tokens work in both auth modes
1070 store, err := tokenstore.NewStore(backend, dataDir, log)
1071 if err != nil {
1072 return nil, log.WrapErr(err, "failed to create token store")
1073 }
1074 return store, nil
1075}
1076
1077func buildAuthConfig(ctx context.Context, cfg Config, authType domain.AuthType, backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) (auth.Config, error) {
1078 authConfig := auth.Config{
1079 Enabled: cfg.Auth.Enabled,
1080 AuthType: authType,
1081 Username: cfg.Auth.Username,
1082 }
1083
1084 // Token config is always required (tokens work in all auth modes)
1085 secret, expiry, err := loadTokenConfig(ctx, cfg, backend, dataDir, log)
1086 if err != nil {
1087 return auth.Config{}, err
1088 }
1089 authConfig.TokenSecret = secret
1090 authConfig.TokenExpiry = expiry
1091
1092 accessTokenTTL := 15 * time.Minute // default
1093 if cfg.Auth.AccessTokenTTL != "" {
1094 parsed, err := time.ParseDuration(cfg.Auth.AccessTokenTTL)
1095 if err != nil {
1096 return auth.Config{}, fmt.Errorf("invalid auth.access_token_ttl %q: %w", cfg.Auth.AccessTokenTTL, err)
1097 }
1098 if parsed <= 0 {
1099 return auth.Config{}, fmt.Errorf("auth.access_token_ttl must be positive")
1100 }
1101 if parsed > auth.MaxAccessTokenLifetime {
1102 return auth.Config{}, fmt.Errorf("auth.access_token_ttl must not exceed %v", auth.MaxAccessTokenLifetime)
1103 }
1104 accessTokenTTL = parsed
1105 }
1106 authConfig.AccessTokenTTL = accessTokenTTL
1107
1108 return authConfig, nil
1109}
1110
1111func loadTokenConfig(ctx context.Context, cfg Config, backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) ([]byte, time.Duration, error) {
1112 secret, err := loadTokenSecret(ctx, cfg, backend, dataDir, log)
1113 if err != nil {
1114 return nil, 0, err
1115 }
1116
1117 expiry, err := parseTokenExpiry(cfg.Auth.TokenExpiry)
1118 if err != nil {
1119 return nil, 0, err
1120 }
1121
1122 return secret, expiry, nil
1123}
1124
1125// TokenSecretEnvVar is the environment variable for the JWT signing secret.
1126// SECURITY: This takes priority over config file to allow secure secret injection.
1127const TokenSecretEnvVar = "GORDON_AUTH_TOKEN_SECRET" //nolint:gosec // This is an env var name, not a credential
1128
1129func loadTokenSecret(ctx context.Context, cfg Config, backend domain.SecretsBackend, dataDir string, log zerowrap.Logger) ([]byte, error) {
1130 // SECURITY: Priority order for token secret:
1131 // 1. Environment variable (most secure - no disk exposure)
1132 // 2. Secrets backend (pass/sops - encrypted)
1133 // 3. Config file path (least preferred)
1134
1135 const minTokenSecretLength = 32
1136
1137 // Check environment variable first
1138 if envSecret := os.Getenv(TokenSecretEnvVar); envSecret != "" {
1139 if len(envSecret) < minTokenSecretLength {
1140 return nil, fmt.Errorf("token secret from %s must be at least %d bytes (got %d)", TokenSecretEnvVar, minTokenSecretLength, len(envSecret))
1141 }
1142 log.Debug().Msg("using token secret from environment variable")
1143 return []byte(envSecret), nil
1144 }
1145
1146 // Fall back to config-specified path via secrets backend
1147 if cfg.Auth.TokenSecret == "" {
1148 return nil, fmt.Errorf("token_secret is required for JWT token generation; set %s environment variable or configure auth.token_secret", TokenSecretEnvVar)
1149 }
1150
1151 secret, err := loadSecret(ctx, backend, cfg.Auth.TokenSecret, dataDir, log)
1152 if err != nil {
1153 return nil, log.WrapErr(err, "failed to load token secret")
1154 }
1155
1156 if len(secret) < minTokenSecretLength {
1157 return nil, fmt.Errorf("token_secret must be at least %d bytes (got %d); use a strong random secret", minTokenSecretLength, len(secret))
1158 }
1159
1160 return []byte(secret), nil
1161}
1162
1163func parseTokenExpiry(expiry string) (time.Duration, error) {
1164 if expiry == "" {
1165 return 0, nil
1166 }
1167
1168 parsed, err := duration.Parse(expiry)
1169 if err != nil {
1170 return 0, fmt.Errorf("invalid token_expiry: %w", err)
1171 }
1172
1173 return parsed, nil
1174}
1175
1176func resolveServiceTokenExpiry(cfg Config) (time.Duration, error) {
1177 expiry, err := parseTokenExpiry(cfg.Auth.TokenExpiry)
1178 if err != nil {
1179 return 0, err
1180 }
1181 if expiry <= 0 {
1182 return serviceTokenDefaultTTL, nil
1183 }
1184 return expiry, nil
1185}
1186
1187// loadSecret loads a secret from the configured backend.
1188func loadSecret(ctx context.Context, backend domain.SecretsBackend, path, dataDir string, log zerowrap.Logger) (string, error) {
1189 switch backend {
1190 case domain.SecretsBackendPass:
1191 provider := secrets.NewPassProvider(log)
1192 return provider.GetSecret(ctx, path)
1193 case domain.SecretsBackendSops:
1194 provider := secrets.NewSopsProvider(log)
1195 return provider.GetSecret(ctx, path)
1196 case domain.SecretsBackendUnsafe:
1197 // For unsafe backend, path is relative to dataDir/secrets/
1198 secretFile := filepath.Join(dataDir, "secrets", path)
1199 data, err := os.ReadFile(secretFile)
1200 if err != nil {
1201 return "", fmt.Errorf("failed to read secret file: %w", err)
1202 }
1203 return string(data), nil
1204 default:
1205 return "", fmt.Errorf("unknown secrets backend: %s", backend)
1206 }
1207}
1208
1209// proxyConfigResult holds parsed proxy and blob chunk size config.
1210type proxyConfigResult struct {
1211 proxyConfig proxy.Config
1212 maxBlobChunkSize int64
1213}
1214
1215// buildProxyConfig parses size-related config fields and builds the proxy config.
1216func buildProxyConfig(cfg Config, log zerowrap.Logger) (*proxyConfigResult, error) {
1217 maxProxyBodySize := int64(512 << 20) // 512MB default
1218 if cfg.Server.MaxProxyBodySize != "" {
1219 parsedSize, err := bytesize.Parse(cfg.Server.MaxProxyBodySize)
1220 if err != nil {
1221 return nil, log.WrapErrWithFields(err, "invalid server.max_proxy_body_size configuration", map[string]any{"value": cfg.Server.MaxProxyBodySize})
1222 }
1223 maxProxyBodySize = parsedSize
1224 }
1225
1226 maxBlobChunkSize := int64(registry.DefaultMaxBlobChunkSize)
1227 if cfg.Server.MaxBlobChunkSize != "" {
1228 parsedSize, err := bytesize.Parse(cfg.Server.MaxBlobChunkSize)
1229 if err != nil {
1230 return nil, log.WrapErrWithFields(err, "invalid server.max_blob_chunk_size configuration", map[string]any{"value": cfg.Server.MaxBlobChunkSize})
1231 }
1232 maxBlobChunkSize = parsedSize
1233 }
1234
1235 maxProxyResponseSize := int64(1 << 30) // 1GB default
1236 if cfg.Server.MaxProxyResponseSize != "" {
1237 parsedSize, err := bytesize.Parse(cfg.Server.MaxProxyResponseSize)
1238 if err != nil {
1239 return nil, log.WrapErrWithFields(err, "invalid server.max_proxy_response_size configuration", map[string]any{"value": cfg.Server.MaxProxyResponseSize})
1240 }
1241 maxProxyResponseSize = parsedSize
1242 }
1243
1244 maxConcurrentConns := cfg.Server.MaxConcurrentConns
1245 if maxConcurrentConns < 0 {
1246 maxConcurrentConns = 10000 // default when explicitly set to -1
1247 }
1248 // 0 means no limit (as documented in proxy.Config)
1249
1250 return &proxyConfigResult{
1251 proxyConfig: proxy.Config{
1252 RegistryDomain: cfg.Server.RegistryDomain,
1253 RegistryPort: cfg.Server.RegistryPort,
1254 MaxBodySize: maxProxyBodySize,
1255 MaxResponseSize: maxProxyResponseSize,
1256 MaxConcurrentConns: maxConcurrentConns,
1257 },
1258 maxBlobChunkSize: maxBlobChunkSize,
1259 }, nil
1260}
1261
1262// createContainerService creates the container service with configuration.
1263func createContainerService(ctx context.Context, v *viper.Viper, cfg Config, svc *services, log zerowrap.Logger) (*container.Service, error) {
1264 // Parse and validate container resource limits from config
1265 if cfg.Containers.CPULimit < 0 {
1266 return nil, fmt.Errorf("containers.cpu_limit must be >= 0 (got %f)", cfg.Containers.CPULimit)
1267 }
1268 if cfg.Containers.PidsLimit < 0 {
1269 return nil, fmt.Errorf("containers.pids_limit must be >= 0 (got %d)", cfg.Containers.PidsLimit)
1270 }
1271 var defaultMemoryLimit int64
1272 if cfg.Containers.MemoryLimit != "" {
1273 parsed, err := bytesize.Parse(cfg.Containers.MemoryLimit)
1274 if err != nil {
1275 return nil, fmt.Errorf("invalid containers.memory_limit %q: %w", cfg.Containers.MemoryLimit, err)
1276 }
1277 if parsed <= 0 {
1278 return nil, fmt.Errorf("containers.memory_limit must be positive (got %q)", cfg.Containers.MemoryLimit)
1279 }
1280 defaultMemoryLimit = parsed
1281 }
1282 var defaultNanoCPUs int64
1283 if cfg.Containers.CPULimit > 0 {
1284 defaultNanoCPUs = int64(cfg.Containers.CPULimit * 1e9)
1285 }
1286
1287 containerConfig := container.Config{
1288 RegistryAuthEnabled: cfg.Auth.Enabled,
1289 RegistryDomain: cfg.Server.RegistryDomain,
1290 RegistryPort: cfg.Server.RegistryPort,
1291 InternalRegistryUsername: svc.internalRegUser,
1292 InternalRegistryPassword: svc.internalRegPass,
1293 PullPolicy: v.GetString("deploy.pull_policy"),
1294 VolumeAutoCreate: v.GetBool("volumes.auto_create"),
1295 VolumePrefix: v.GetString("volumes.prefix"),
1296 VolumePreserve: v.GetBool("volumes.preserve"),
1297 NetworkIsolation: v.GetBool("network_isolation.enabled"),
1298 NetworkPrefix: v.GetString("network_isolation.network_prefix"),
1299 NetworkGroups: svc.configSvc.GetNetworkGroups(),
1300 Attachments: svc.configSvc.GetAttachments(),
1301 ReadinessDelay: v.GetDuration("deploy.readiness_delay"),
1302 ReadinessMode: v.GetString("deploy.readiness_mode"),
1303 HealthTimeout: v.GetDuration("deploy.health_timeout"),
1304 StabilizationDelay: v.GetDuration("deploy.stabilization_delay"),
1305 TCPProbeTimeout: v.GetDuration("deploy.tcp_probe_timeout"),
1306 HTTPProbeTimeout: v.GetDuration("deploy.http_probe_timeout"),
1307 DrainDelay: v.GetDuration("deploy.drain_delay"),
1308 DrainMode: v.GetString("deploy.drain_mode"),
1309 DrainTimeout: v.GetDuration("deploy.drain_timeout"),
1310 DefaultMemoryLimit: defaultMemoryLimit,
1311 DefaultNanoCPUs: defaultNanoCPUs,
1312 DefaultPidsLimit: cfg.Containers.PidsLimit,
1313 AttachmentReadinessTimeout: v.GetDuration("deploy.attachment_readiness_timeout"),
1314 }
1315 if v.IsSet("deploy.drain_delay") {
1316 containerConfig.DrainDelayConfigured = true
1317 containerConfig.DrainDelay = v.GetDuration("deploy.drain_delay")
1318 }
1319
1320 if containerConfig.RegistryAuthEnabled {
1321 if svc.authSvc == nil {
1322 return nil, fmt.Errorf("authentication service unavailable: cannot generate registry service token")
1323 }
1324 expiry, err := resolveServiceTokenExpiry(cfg)
1325 if err != nil {
1326 return nil, log.WrapErr(err, "failed to resolve service token expiry")
1327 }
1328 // Note: Service tokens are not auto-refreshed. If the token expires during
1329 // container runtime, the container will need to be recreated to get a new token.
1330 serviceToken, err := svc.authSvc.GenerateToken(ctx, serviceTokenSubject, []string{"pull"}, expiry)
1331 if err != nil {
1332 return nil, log.WrapErr(err, "failed to generate registry service token")
1333 }
1334 log.Info().
1335 Str("subject", serviceTokenSubject).
1336 Str("expiry", expiry.String()).
1337 Msg("generated service token for container registry access")
1338 containerConfig.ServiceTokenUsername = serviceTokenSubject
1339 containerConfig.ServiceToken = serviceToken
1340 } else {
1341 log.Warn().Msg("registry auth disabled; container image pulls will use unauthenticated mode")
1342 }
1343
1344 return container.NewService(svc.runtime, svc.envLoader, svc.eventBus, svc.logWriter, containerConfig, svc.configSvc), nil
1345}
1346
1347func createBackupService(cfg Config, svc *services, log zerowrap.Logger) (*filesystem.BackupStorage, *backup.Service, error) {
1348 if !cfg.Backups.Enabled {
1349 return nil, nil, nil
1350 }
1351
1352 storageDir := cfg.Backups.StorageDir
1353 if storageDir == "" {
1354 dataDir := resolveDataDir(cfg.Server.DataDir)
1355 storageDir = filepath.Join(dataDir, "backups")
1356 }
1357
1358 backupStorage, err := filesystem.NewBackupStorage(storageDir, log)
1359 if err != nil {
1360 return nil, nil, log.WrapErr(err, "failed to create backup storage")
1361 }
1362
1363 retention, err := validateBackupRetention(cfg)
1364 if err != nil {
1365 return nil, nil, log.WrapErr(err, "invalid backup retention policy")
1366 }
1367
1368 backupCfg := domain.BackupConfig{
1369 Enabled: cfg.Backups.Enabled,
1370 StorageDir: storageDir,
1371 Retention: retention,
1372 }
1373
1374 backupSvc := backup.NewService(svc.runtime, backupStorage, svc.containerSvc, backupCfg, log)
1375
1376 log.Info().
1377 Str("storage_dir", storageDir).
1378 Msg("backup service initialized")
1379
1380 return backupStorage, backupSvc, nil
1381}
1382
1383func validateBackupRetention(cfg Config) (domain.RetentionPolicy, error) {
1384 if cfg.Backups.Retention.Hourly < 0 {
1385 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.hourly cannot be negative")
1386 }
1387 if cfg.Backups.Retention.Daily < 0 {
1388 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.daily cannot be negative")
1389 }
1390 if cfg.Backups.Retention.Weekly < 0 {
1391 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.weekly cannot be negative")
1392 }
1393 if cfg.Backups.Retention.Monthly < 0 {
1394 return domain.RetentionPolicy{}, fmt.Errorf("backups.retention.monthly cannot be negative")
1395 }
1396
1397 return domain.RetentionPolicy{
1398 Hourly: cfg.Backups.Retention.Hourly,
1399 Daily: cfg.Backups.Retention.Daily,
1400 Weekly: cfg.Backups.Retention.Weekly,
1401 Monthly: cfg.Backups.Retention.Monthly,
1402 }, nil
1403}
1404
1405// registerEventHandlers registers all event handlers.
1406func registerEventHandlers(ctx context.Context, svc *services, cfg Config) (func(), error) {
1407 imagePushedHandler := container.NewImagePushedHandler(ctx, svc.containerSvc, svc.configSvc)
1408 if err := svc.eventBus.Subscribe(imagePushedHandler); err != nil {
1409 return nil, fmt.Errorf("failed to subscribe image pushed handler: %w", err)
1410 }
1411
1412 // Auto-route handler for creating routes from image labels
1413 autoRouteHandler := container.NewAutoRouteHandler(ctx, svc.configSvc, svc.containerSvc, svc.blobStorage, cfg.Server.RegistryDomain).
1414 WithEnvExtractor(svc.runtime, svc.envDir)
1415
1416 // Preview handler for creating preview environments from tagged images
1417 autoPreviewHandler := preview.NewAutoPreviewHandler(
1418 ctx,
1419 svc.configSvc,
1420 svc.blobStorage,
1421 svc.previewService,
1422 )
1423
1424 // Dispatcher routes image push events to either auto-route or preview handler
1425 dispatcher := auto.NewImagePushDispatcher(svc.configSvc, autoRouteHandler, autoPreviewHandler)
1426 if err := svc.eventBus.Subscribe(dispatcher); err != nil {
1427 return nil, fmt.Errorf("subscribe image push dispatcher: %w", err)
1428 }
1429
1430 configReloadHandler := container.NewConfigReloadHandler(ctx, svc.containerSvc, svc.configSvc)
1431 if err := svc.eventBus.Subscribe(configReloadHandler); err != nil {
1432 return nil, fmt.Errorf("failed to subscribe config reload handler: %w", err)
1433 }
1434
1435 manualReloadHandler := container.NewManualReloadHandler(ctx, svc.containerSvc, svc.configSvc)
1436 if err := svc.eventBus.Subscribe(manualReloadHandler); err != nil {
1437 return nil, fmt.Errorf("failed to subscribe manual reload handler: %w", err)
1438 }
1439
1440 manualDeployHandler := container.NewManualDeployHandler(ctx, svc.containerSvc, svc.configSvc)
1441 if err := svc.eventBus.Subscribe(manualDeployHandler); err != nil {
1442 return nil, fmt.Errorf("failed to subscribe manual deploy handler: %w", err)
1443 }
1444
1445 secretsChangedHandler := container.NewSecretsChangedHandler(ctx, svc.containerSvc, svc.configSvc, container.DefaultSecretsDebounce)
1446 if err := svc.eventBus.Subscribe(secretsChangedHandler); err != nil {
1447 return nil, fmt.Errorf("failed to subscribe secrets changed handler: %w", err)
1448 }
1449
1450 // Proxy cache invalidation on config reload (clears stale targets for removed routes)
1451 configReloadProxyHandler := proxy.NewConfigReloadProxyHandler(ctx, svc.proxySvc)
1452 if err := svc.eventBus.Subscribe(configReloadProxyHandler); err != nil {
1453 return nil, fmt.Errorf("failed to subscribe config reload proxy handler: %w", err)
1454 }
1455
1456 cleanup := func() {
1457 secretsChangedHandler.Stop()
1458 }
1459
1460 return cleanup, nil
1461}
1462
1463// setupConfigHotReload sets up Viper config hot reload.
1464// NOTE: This does NOT reload routes into memory. Routes are managed via API
1465// (AddRoute/UpdateRoute/RemoveRoute) and memory is the source of truth.
1466// The file watcher only updates proxy config and refreshes targets.
1467// Manual config file edits to routes require a server restart.
1468func setupConfigHotReload(ctx context.Context, v *viper.Viper, svc *services, log zerowrap.Logger) {
1469 v.OnConfigChange(func(e fsnotify.Event) {
1470 log.Info().Str("file", e.Name).Msg("config file changed")
1471
1472 if err := v.ReadInConfig(); err != nil {
1473 log.Error().Err(err).Msg("failed to reload config")
1474 return
1475 }
1476
1477 // Re-unmarshal config and rebuild proxy config to pick up all changes
1478 var reloadCfg Config
1479 if err := v.Unmarshal(&reloadCfg); err != nil {
1480 log.Error().Err(err).Msg("failed to unmarshal config on reload")
1481 return
1482 }
1483 if reloadCfg.Server.GordonDomain != "" {
1484 reloadCfg.Server.RegistryDomain = reloadCfg.Server.GordonDomain
1485 }
1486 reloadedProxy, err := buildProxyConfig(reloadCfg, log)
1487 if err != nil {
1488 log.Error().Err(err).Msg("failed to parse proxy config on reload")
1489 return
1490 }
1491 svc.proxySvc.UpdateConfig(reloadedProxy.proxyConfig)
1492
1493 // Clear proxy target cache to pick up external route changes
1494 if err := svc.proxySvc.RefreshTargets(ctx); err != nil {
1495 log.Warn().Err(err).Msg("failed to refresh proxy targets")
1496 }
1497
1498 log.Debug().Msg("config hot reload complete (routes unchanged, memory is source of truth)")
1499 })
1500 v.WatchConfig()
1501}
1502
1503// syncAndAutoStart syncs existing containers and auto-starts if configured.
1504func syncAndAutoStart(ctx context.Context, svc *services, log zerowrap.Logger) {
1505 if err := svc.containerSvc.SyncContainers(ctx); err != nil {
1506 log.Warn().Err(err).Msg("failed to sync existing containers")
1507 }
1508
1509 if svc.configSvc.IsAutoRouteEnabled() {
1510 routes := svc.configSvc.GetRoutes(ctx)
1511 if err := svc.containerSvc.AutoStart(domain.WithInternalDeploy(ctx), routes); err != nil {
1512 log.Warn().Err(err).Msg("failed to auto-start containers")
1513 }
1514 }
1515
1516 // Start background monitor to restart crashed containers.
1517 svc.containerSvc.StartMonitor(ctx)
1518}
1519func loopbackOnly(next http.Handler, log zerowrap.Logger) http.Handler {
1520 return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
1521 host, _, err := net.SplitHostPort(r.RemoteAddr)
1522 if err != nil {
1523 host = r.RemoteAddr
1524 }
1525
1526 ip := net.ParseIP(host)
1527 if ip == nil || !ip.IsLoopback() {
1528 log.Warn().
1529 Str("path", r.URL.Path).
1530 Str("remote_addr", r.RemoteAddr).
1531 Msg("blocked non-loopback access on internal admin route")
1532 w.Header().Set("Content-Type", "application/json")
1533 w.WriteHeader(http.StatusForbidden)
1534 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "Forbidden"})
1535 return
1536 }
1537
1538 next.ServeHTTP(w, r)
1539 })
1540}
1541
1542// createHTTPHandlers creates HTTP handlers with middleware.
1543func createHTTPHandlers(svc *services, cfg Config, log zerowrap.Logger) (http.Handler, http.Handler) {
1544 // Parse trusted proxies once for all middleware chains.
1545 // This ensures consistent IP extraction across logging, rate limiting, and auth.
1546 trustedNets := middleware.ParseTrustedProxies(cfg.API.RateLimit.TrustedProxies)
1547
1548 registryHandler := registry.NewHandler(svc.registrySvc, log, svc.maxBlobChunkSize)
1549 registryWithMiddleware, cidrAllowlistMiddleware, rateLimitMiddleware := buildRegistryHandlerWithMiddleware(
1550 svc,
1551 cfg,
1552 trustedNets,
1553 registryHandler,
1554 log,
1555 )
1556
1557 registryMux := http.NewServeMux()
1558 registerAuthRoutes(registryMux, svc, trustedNets, cidrAllowlistMiddleware, rateLimitMiddleware, cfg, log)
1559 registryMux.Handle("/v2/", wrapRegistryForLocalMode(registryWithMiddleware, cfg, log))
1560
1561 // SECURITY: No CORS middleware on the proxy chain. Backend applications
1562 // should control their own CORS policies. A blanket Access-Control-Allow-Origin: *
1563 // would override backend CORS settings and allow any website to make
1564 // cross-origin authenticated requests to proxied applications.
1565 proxyMiddlewares := []func(http.Handler) http.Handler{
1566 middleware.PanicRecovery(log),
1567 middleware.RequestLogger(log, trustedNets),
1568 middleware.SecurityHeaders,
1569 }
1570
1571 proxyHandler := proxyadapter.NewHandler(svc.proxySvc, log)
1572 proxyWithMiddleware := otelhttp.NewHandler(
1573 middleware.Chain(proxyMiddlewares...)(proxyHandler),
1574 "gordon.proxy",
1575 )
1576 proxyMux := http.NewServeMux()
1577 proxyMux.Handle("/", proxyWithMiddleware)
1578
1579 registerAdminRoutes(registryMux, svc, cfg, trustedNets, log)
1580
1581 return registryMux, proxyMux
1582}
1583
1584func buildRegistryHandlerWithMiddleware(
1585 svc *services,
1586 cfg Config,
1587 trustedNets []*net.IPNet,
1588 registryHandler http.Handler,
1589 log zerowrap.Logger,
1590) (http.Handler, func(http.Handler) http.Handler, func(http.Handler) http.Handler) {
1591 registryMiddlewares := []func(http.Handler) http.Handler{
1592 middleware.PanicRecovery(log),
1593 middleware.RequestLogger(log, trustedNets),
1594 middleware.SecurityHeaders,
1595 }
1596
1597 cidrAllowlistMiddleware := buildRegistryCIDRAllowlistMiddleware(cfg, trustedNets, log)
1598 if cidrAllowlistMiddleware != nil {
1599 registryMiddlewares = append(registryMiddlewares, cidrAllowlistMiddleware)
1600 }
1601
1602 rateLimitMiddleware := buildRegistryRateLimitMiddleware(cfg, log)
1603 registryMiddlewares = append(registryMiddlewares, rateLimitMiddleware)
1604
1605 appendRegistryAuthMiddleware(®istryMiddlewares, svc, cfg, log)
1606
1607 registryWithOtel := otelhttp.NewHandler(
1608 middleware.Chain(registryMiddlewares...)(registryHandler),
1609 "gordon.registry",
1610 )
1611 return registryWithOtel, cidrAllowlistMiddleware, rateLimitMiddleware
1612}
1613
1614func buildRegistryCIDRAllowlistMiddleware(cfg Config, trustedNets []*net.IPNet, log zerowrap.Logger) func(http.Handler) http.Handler {
1615 if len(cfg.Server.RegistryAllowedIPs) == 0 {
1616 return nil
1617 }
1618
1619 allowedNets := middleware.ParseTrustedProxies(cfg.Server.RegistryAllowedIPs)
1620 if len(allowedNets) != len(cfg.Server.RegistryAllowedIPs) {
1621 for _, entry := range cfg.Server.RegistryAllowedIPs {
1622 if nets := middleware.ParseTrustedProxies([]string{entry}); len(nets) == 0 {
1623 log.Warn().Str("entry", entry).Msg("ignoring invalid registry_allowed_ips entry")
1624 }
1625 }
1626 }
1627
1628 if len(allowedNets) == 0 {
1629 log.Error().
1630 Strs("registry_allowed_ips", cfg.Server.RegistryAllowedIPs).
1631 Msg("registry_allowed_ips is set but no valid entries were parsed; registry will deny all traffic (fail-closed)")
1632 return func(next http.Handler) http.Handler {
1633 return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
1634 log.Warn().
1635 Str(zerowrap.FieldPath, r.URL.Path).
1636 Str(zerowrap.FieldClientIP, middleware.GetClientIP(r, trustedNets)).
1637 Msg("registry access denied due to invalid registry_allowed_ips configuration")
1638 w.Header().Set("Content-Type", "application/json")
1639 w.WriteHeader(http.StatusForbidden)
1640 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "Forbidden"})
1641 })
1642 }
1643 }
1644
1645 return middleware.RegistryCIDRAllowlist(allowedNets, trustedNets, log)
1646}
1647
1648func buildRegistryRateLimitMiddleware(cfg Config, log zerowrap.Logger) func(http.Handler) http.Handler {
1649 if cfg.API.RateLimit.Enabled {
1650 globalLimiter := ratelimit.NewMemoryStore(cfg.API.RateLimit.GlobalRPS, cfg.API.RateLimit.Burst, log)
1651 ipLimiter := ratelimit.NewMemoryStore(cfg.API.RateLimit.PerIPRPS, cfg.API.RateLimit.Burst, log)
1652 return registry.RateLimitMiddleware(
1653 globalLimiter,
1654 ipLimiter,
1655 cfg.API.RateLimit.TrustedProxies,
1656 log,
1657 )
1658 }
1659
1660 return registry.RateLimitMiddleware(nil, nil, nil, log)
1661}
1662
1663func appendRegistryAuthMiddleware(registryMiddlewares *[]func(http.Handler) http.Handler, svc *services, cfg Config, log zerowrap.Logger) {
1664 if svc.authSvc != nil {
1665 internalAuth := middleware.InternalRegistryAuth{
1666 Username: svc.internalRegUser,
1667 Password: svc.internalRegPass,
1668 }
1669 *registryMiddlewares = append(*registryMiddlewares, middleware.RegistryAuthV2(svc.authSvc, internalAuth, log))
1670 return
1671 }
1672
1673 if cfg.Auth.Enabled {
1674 log.Error().Msg("authentication service unavailable; registry requests will be denied")
1675 *registryMiddlewares = append(*registryMiddlewares, func(next http.Handler) http.Handler {
1676 return http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
1677 w.Header().Set("Content-Type", "application/json")
1678 w.WriteHeader(http.StatusServiceUnavailable)
1679 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "authentication service unavailable"})
1680 })
1681 })
1682 }
1683}
1684
1685func registerAuthRoutes(
1686 registryMux *http.ServeMux,
1687 svc *services,
1688 trustedNets []*net.IPNet,
1689 cidrAllowlistMiddleware func(http.Handler) http.Handler,
1690 rateLimitMiddleware func(http.Handler) http.Handler,
1691 cfg Config,
1692 log zerowrap.Logger,
1693) {
1694 if svc.authHandler == nil {
1695 return
1696 }
1697
1698 // Auth endpoints always get rate limiting, even if global rate limiting is disabled.
1699 // This prevents brute-force attacks against password/token endpoints.
1700 authRateLimitMiddleware := rateLimitMiddleware
1701 if !cfg.API.RateLimit.Enabled {
1702 authGlobalLimiter := ratelimit.NewMemoryStore(50, 100, log)
1703 authIPLimiter := ratelimit.NewMemoryStore(5, 10, log)
1704 authRateLimitMiddleware = registry.RateLimitMiddleware(authGlobalLimiter, authIPLimiter, cfg.API.RateLimit.TrustedProxies, log)
1705 }
1706
1707 // Auth endpoints are NOT protected by auth - they're where clients authenticate
1708 // but still need rate limiting to prevent brute force attacks.
1709 authMiddlewares := []func(http.Handler) http.Handler{
1710 middleware.PanicRecovery(log),
1711 middleware.RequestLogger(log, trustedNets),
1712 middleware.SecurityHeaders,
1713 }
1714 if cidrAllowlistMiddleware != nil {
1715 authMiddlewares = append(authMiddlewares, cidrAllowlistMiddleware)
1716 }
1717 authMiddlewares = append(authMiddlewares, authRateLimitMiddleware)
1718 authWithMiddleware := otelhttp.NewHandler(
1719 middleware.Chain(authMiddlewares...)(svc.authHandler),
1720 "gordon.auth",
1721 )
1722 registryMux.Handle("/auth/", authWithMiddleware)
1723}
1724
1725func wrapRegistryForLocalMode(registryWithMiddleware http.Handler, cfg Config, log zerowrap.Logger) http.Handler {
1726 if !cfg.Auth.Enabled {
1727 return loopbackOnly(registryWithMiddleware, log)
1728 }
1729 return registryWithMiddleware
1730}
1731
1732func registerAdminRoutes(registryMux *http.ServeMux, svc *services, cfg Config, trustedNets []*net.IPNet, log zerowrap.Logger) {
1733 if svc.adminHandler == nil {
1734 return
1735 }
1736
1737 if !cfg.Auth.Enabled {
1738 log.Warn().Msg("auth disabled: admin API endpoints are not registered")
1739 return
1740 }
1741
1742 adminMiddlewares := []func(http.Handler) http.Handler{
1743 middleware.PanicRecovery(log),
1744 middleware.RequestLogger(log, trustedNets),
1745 middleware.SecurityHeaders,
1746 }
1747
1748 if svc.authSvc != nil {
1749 // Create rate limiters for admin API - uses same config as registry.
1750 var globalLimiter, ipLimiter out.RateLimiter
1751 if cfg.API.RateLimit.Enabled {
1752 globalLimiter = ratelimit.NewMemoryStore(cfg.API.RateLimit.GlobalRPS, cfg.API.RateLimit.Burst, log)
1753 ipLimiter = ratelimit.NewMemoryStore(cfg.API.RateLimit.PerIPRPS, cfg.API.RateLimit.Burst, log)
1754 }
1755 adminMiddlewares = append(adminMiddlewares, admin.AuthMiddleware(svc.authSvc, globalLimiter, ipLimiter, trustedNets, log))
1756 } else {
1757 adminMiddlewares = append(adminMiddlewares, func(next http.Handler) http.Handler {
1758 return http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
1759 w.Header().Set("Content-Type", "application/json")
1760 w.WriteHeader(http.StatusServiceUnavailable)
1761 _ = json.NewEncoder(w).Encode(dto.ErrorResponse{Error: "authentication service unavailable"})
1762 })
1763 })
1764 }
1765
1766 adminWithMiddleware := otelhttp.NewHandler(
1767 middleware.Chain(adminMiddlewares...)(svc.adminHandler),
1768 "gordon.admin",
1769 )
1770 registryMux.Handle("/admin/", loopbackOnly(adminWithMiddleware, log))
1771}
1772
1773// runServers starts the HTTP servers and waits for shutdown.
1774// Signal handling notes:
1775// - SIGINT/SIGTERM: Triggers graceful shutdown via signal.NotifyContext
1776// - SIGUSR1: Triggers config reload without restart
1777// - SIGUSR2: Triggers manual deploy for a specific route
1778// The deferred signal.Stop calls ensure signal handlers are properly
1779// cleaned up before program exit, preventing signal handler leaks.
1780func runServers(ctx context.Context, v *viper.Viper, cfg Config, registryHandler, proxyHandler http.Handler, svc *services, cleanupHandlers func(), log zerowrap.Logger) error {
1781 ctx, cancel := signal.NotifyContext(ctx, os.Interrupt, syscall.SIGTERM)
1782 defer cancel()
1783
1784 // Set up SIGUSR1 for reload.
1785 // Note: signal.Stop must be called (via defer) to release the channel
1786 // and prevent signal handler leaks when the function returns.
1787 reloadChan := make(chan os.Signal, 1)
1788 signal.Notify(reloadChan, syscall.SIGUSR1)
1789 defer signal.Stop(reloadChan)
1790
1791 // Set up SIGUSR2 for manual deploy.
1792 deployChan := make(chan os.Signal, 1)
1793 signal.Notify(deployChan, syscall.SIGUSR2)
1794 defer signal.Stop(deployChan)
1795
1796 errChan := make(chan error, 3)
1797
1798 registrySrv, registryReady := startServer(fmt.Sprintf(":%d", cfg.Server.RegistryPort), registryHandler, "registry", nil, errChan, log)
1799
1800 proxySrv, proxyReady, tlsSrv, tlsReady, err := startProxyServers(cfg, proxyHandler, errChan, log)
1801 if err != nil {
1802 return err
1803 }
1804
1805 // Wait for all enabled servers to bind their ports before auto-starting containers.
1806 // This prevents the race where auto-start pulls from the registry before it's listening.
1807 if err := waitForServerReady(registryReady, errChan); err != nil {
1808 return err
1809 }
1810 if err := waitForServerReady(proxyReady, errChan); err != nil {
1811 return err
1812 }
1813 if cfg.Server.TLSEnabled {
1814 if err := waitForServerReady(tlsReady, errChan); err != nil {
1815 return err
1816 }
1817 }
1818
1819 logEvent := log.Info().
1820 Int("proxy_port", cfg.Server.Port).
1821 Int("registry_port", cfg.Server.RegistryPort)
1822 if cfg.Server.TLSEnabled {
1823 logEvent = logEvent.Int("tls_port", cfg.Server.TLSPort)
1824 }
1825 logEvent.Msg("Gordon is running")
1826
1827 schedulerCleanup, err := startOptionalSchedulers(ctx, cfg, svc, log, v)
1828 if err != nil {
1829 return err
1830 }
1831 if schedulerCleanup != nil {
1832 defer schedulerCleanup()
1833 }
1834
1835 // Auto-start after servers are listening (registry port is now bound).
1836 syncAndAutoStart(ctx, svc, log)
1837
1838 waitForShutdown(ctx, errChan, reloadChan, deployChan, svc.eventBus, log)
1839 cleanupHandlers() // Stop debounce timers before draining containers
1840 gracefulShutdown(registrySrv, proxySrv, tlsSrv, svc.containerSvc, svc.proxySvc, log)
1841 return nil
1842}
1843
1844func waitForServerReady(ready <-chan struct{}, errChan <-chan error) error {
1845 select {
1846 case <-ready:
1847 return nil
1848 case err := <-errChan:
1849 return err
1850 }
1851}
1852
1853func startOptionalSchedulers(ctx context.Context, cfg Config, svc *services, log zerowrap.Logger, v *viper.Viper) (func(), error) {
1854 schedulers := make([]*cronSvc.Scheduler, 0, 2)
1855
1856 backupScheduler, err := startBackupScheduler(ctx, cfg, svc, log)
1857 if err != nil {
1858 return nil, err
1859 }
1860 if backupScheduler != nil {
1861 schedulers = append(schedulers, backupScheduler)
1862 }
1863
1864 imageScheduler, err := startImagePruneScheduler(ctx, cfg, svc, log, func() int {
1865 return v.GetInt("images.prune.keep_last")
1866 })
1867 if err != nil {
1868 return nil, err
1869 }
1870 if imageScheduler != nil {
1871 schedulers = append(schedulers, imageScheduler)
1872 }
1873
1874 if len(schedulers) == 0 {
1875 return nil, nil
1876 }
1877
1878 return func() {
1879 for i := len(schedulers) - 1; i >= 0; i-- {
1880 schedulers[i].Stop()
1881 }
1882 }, nil
1883}
1884
1885func startBackupScheduler(ctx context.Context, cfg Config, svc *services, log zerowrap.Logger) (*cronSvc.Scheduler, error) {
1886 if !cfg.Backups.Enabled || svc == nil || svc.backupSvc == nil {
1887 return nil, nil
1888 }
1889
1890 preset, err := resolveBackupSchedule(cfg.Backups.Schedule)
1891 if err != nil {
1892 return nil, err
1893 }
1894
1895 scheduler := cronSvc.NewScheduler(log)
1896 err = scheduler.Add(
1897 "backup-scheduler",
1898 "Backups",
1899 domain.CronSchedule{Preset: preset},
1900 func(jobCtx context.Context) error {
1901 if err := svc.backupSvc.RunForSchedule(jobCtx, preset); err != nil {
1902 return err
1903 }
1904 log.Info().
1905 Str("schedule", string(preset)).
1906 Msg("scheduled backup run complete")
1907 return nil
1908 },
1909 )
1910 if err != nil {
1911 return nil, log.WrapErr(err, "failed to register backup schedule")
1912 }
1913
1914 scheduler.Start(ctx)
1915 log.Info().
1916 Str("schedule", string(preset)).
1917 Msg("backup scheduler enabled")
1918
1919 return scheduler, nil
1920}
1921
1922func resolveBackupSchedule(raw string) (domain.BackupSchedule, error) {
1923 return resolveSchedulePreset(raw, "backups.schedule", domain.ScheduleDaily)
1924}
1925
1926func startImagePruneScheduler(ctx context.Context, cfg Config, svc *services, log zerowrap.Logger, keepLastGetter func() int) (*cronSvc.Scheduler, error) {
1927 if !cfg.Images.Prune.Enabled || svc == nil || svc.imageSvc == nil {
1928 return nil, nil
1929 }
1930 if keepLastGetter == nil {
1931 keepLastGetter = func() int { return cfg.Images.Prune.KeepLast }
1932 }
1933 if keepLastGetter() < 0 {
1934 return nil, fmt.Errorf("images.prune.keep_last must be >= 0")
1935 }
1936
1937 preset, err := resolveImagePruneSchedule(cfg.Images.Prune.Schedule)
1938 if err != nil {
1939 return nil, err
1940 }
1941
1942 scheduler := cronSvc.NewScheduler(log)
1943 err = scheduler.Add(
1944 "image-prune",
1945 "Image prune",
1946 domain.CronSchedule{Preset: preset},
1947 func(jobCtx context.Context) error {
1948 keepLast := keepLastGetter()
1949 if keepLast < 0 {
1950 log.Warn().
1951 Int("configured_keep_last", keepLast).
1952 Int("fallback_keep_last", domain.DefaultImagePruneKeepLast).
1953 Msg("invalid images.prune.keep_last; using default")
1954 keepLast = domain.DefaultImagePruneKeepLast
1955 }
1956
1957 report, err := svc.imageSvc.Prune(jobCtx, domain.ImagePruneOptions{
1958 KeepLast: keepLast,
1959 PruneDangling: true,
1960 PruneRegistry: true,
1961 })
1962 if err != nil {
1963 return err
1964 }
1965
1966 log.Info().
1967 Int("keep_last", keepLast).
1968 Int("runtime_deleted", report.Runtime.DeletedCount).
1969 Int64("runtime_reclaimed_bytes", report.Runtime.SpaceReclaimed).
1970 Int("registry_tags_removed", report.Registry.TagsRemoved).
1971 Int("registry_blobs_removed", report.Registry.BlobsRemoved).
1972 Int("registry_uploads_removed", report.Registry.UploadsRemoved).
1973 Int64("registry_upload_bytes_reclaimed", report.Registry.UploadSpaceReclaimed).
1974 Msg("scheduled image prune complete")
1975 return nil
1976 },
1977 )
1978 if err != nil {
1979 return nil, log.WrapErr(err, "failed to register image prune schedule")
1980 }
1981
1982 scheduler.Start(ctx)
1983 log.Info().
1984 Str("schedule", string(preset)).
1985 Int("keep_last", keepLastGetter()).
1986 Msg("image prune scheduler enabled")
1987
1988 return scheduler, nil
1989}
1990
1991func resolveImagePruneSchedule(raw string) (domain.BackupSchedule, error) {
1992 return resolveSchedulePreset(raw, "images.prune.schedule", domain.ScheduleDaily)
1993}
1994
1995func resolveSchedulePreset(raw, name string, defaultVal domain.BackupSchedule) (domain.BackupSchedule, error) {
1996 schedule := domain.BackupSchedule(strings.ToLower(strings.TrimSpace(raw)))
1997 if schedule == "" {
1998 schedule = defaultVal
1999 }
2000
2001 switch schedule {
2002 case domain.ScheduleHourly, domain.ScheduleDaily, domain.ScheduleWeekly, domain.ScheduleMonthly:
2003 return schedule, nil
2004 default:
2005 return "", fmt.Errorf("%s must be one of: hourly, daily, weekly, monthly", name)
2006 }
2007}
2008
2009// waitForShutdown blocks on the event loop, handling server errors and
2010// Unix signals (reload, deploy, shutdown) until the context is cancelled.
2011func waitForShutdown(ctx context.Context, errChan <-chan error, reloadChan, deployChan <-chan os.Signal, eventBus out.EventBus, log zerowrap.Logger) {
2012 for {
2013 select {
2014 case err := <-errChan:
2015 log.Error().Err(err).Msg("server error")
2016 return
2017 case <-reloadChan:
2018 log.Info().Msg("reload signal received (SIGUSR1)")
2019 if err := eventBus.Publish(domain.EventManualReload, nil); err != nil {
2020 log.Error().Err(err).Msg("failed to publish manual reload event")
2021 }
2022 case <-deployChan:
2023 log.Info().Msg("deploy signal received (SIGUSR2)")
2024 domainName, err := readDeployRequest()
2025 if err != nil {
2026 log.Error().Err(err).Msg("failed to read deploy request")
2027 continue
2028 }
2029 payload := &domain.ManualDeployPayload{Domain: domainName}
2030 if err := eventBus.Publish(domain.EventManualDeploy, payload); err != nil {
2031 log.Error().Err(err).Str("domain", domainName).Msg("failed to publish manual deploy event")
2032 }
2033 case <-ctx.Done():
2034 log.Info().Msg("shutdown signal received")
2035 return
2036 }
2037 }
2038}
2039
2040// gracefulShutdown stops HTTP servers with a 30s timeout, then shuts down
2041// the container service and cleans up runtime files.
2042func gracefulShutdown(registrySrv, proxySrv, tlsSrv *http.Server, containerSvc *container.Service, proxySvc *proxy.Service, log zerowrap.Logger) {
2043 log.Info().Msg("shutting down Gordon...")
2044
2045 shutdownCtx, shutdownCancel := context.WithTimeout(context.Background(), 30*time.Second)
2046 defer shutdownCancel()
2047
2048 // Phase 1: Stop ingress frontends (TLS, then proxy) — no new traffic accepted
2049 for _, srv := range []*http.Server{tlsSrv, proxySrv} {
2050 if srv == nil {
2051 continue
2052 }
2053 if err := srv.Shutdown(shutdownCtx); err != nil {
2054 log.Warn().Err(err).Str("addr", srv.Addr).Msg("server shutdown error")
2055 }
2056 }
2057
2058 // Phase 2: Drain in-flight registry push sessions before stopping the backend
2059 if proxySvc != nil {
2060 log.Info().Msg("draining in-flight registry requests...")
2061 if drained := proxySvc.DrainRegistryInFlight(25 * time.Second); !drained {
2062 log.Warn().Int64("in_flight", proxySvc.RegistryInFlight()).Msg("registry drain timed out; some in-flight pushes may be interrupted")
2063 }
2064 }
2065
2066 // Phase 3: Stop the registry backend
2067 if registrySrv != nil {
2068 if err := registrySrv.Shutdown(shutdownCtx); err != nil {
2069 log.Warn().Err(err).Str("addr", registrySrv.Addr).Msg("server shutdown error")
2070 }
2071 }
2072
2073 containerSvc.StopMonitor()
2074
2075 if err := containerSvc.Shutdown(shutdownCtx); err != nil {
2076 log.Warn().Err(err).Msg("error during container shutdown")
2077 }
2078
2079 cleanupInternalCredentials()
2080 log.Info().Msg("Gordon stopped")
2081}
2082
2083// startProxyServers sets up proxy server(s) depending on TLS configuration.
2084// With TLS enabled, the HTTP port becomes a redirect-only server and a separate TLS server handles traffic.
2085// Without TLS, a single server handles both HTTP/1.1 and cleartext HTTP/2.
2086func startProxyServers(cfg Config, proxyHandler http.Handler, errChan chan<- error, log zerowrap.Logger) (*http.Server, <-chan struct{}, *http.Server, <-chan struct{}, error) {
2087 if !cfg.Server.TLSEnabled {
2088 var proxyProtos http.Protocols
2089 proxyProtos.SetHTTP1(true)
2090 proxyProtos.SetUnencryptedHTTP2(true)
2091 srv, ready := startServer(fmt.Sprintf(":%d", cfg.Server.Port), proxyHandler, "proxy", &proxyProtos, errChan, log)
2092 return srv, ready, nil, nil, nil
2093 }
2094
2095 if cfg.Server.TLSCertFile == "" || cfg.Server.TLSKeyFile == "" {
2096 return nil, nil, nil, nil, fmt.Errorf("server.tls_enabled=true requires both server.tls_cert_file and server.tls_key_file")
2097 }
2098
2099 tlsPort := cfg.Server.TLSPort
2100 redirectHandler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
2101 host := r.Host
2102 if h, _, err := net.SplitHostPort(host); err == nil {
2103 host = h
2104 } else if strings.HasPrefix(host, "[") && strings.HasSuffix(host, "]") {
2105 host = strings.TrimSuffix(strings.TrimPrefix(host, "["), "]")
2106 }
2107 if tlsPort != 443 {
2108 host = net.JoinHostPort(host, strconv.Itoa(tlsPort))
2109 } else if strings.Contains(host, ":") {
2110 host = "[" + host + "]"
2111 }
2112 target := "https://" + host + r.RequestURI
2113 http.Redirect(w, r, target, http.StatusPermanentRedirect)
2114 })
2115
2116 proxySrv, proxyReady := startServer(fmt.Sprintf(":%d", cfg.Server.Port), redirectHandler, "proxy-redirect", nil, errChan, log)
2117 tlsSrv, tlsReady := startTLSServer(
2118 fmt.Sprintf(":%d", cfg.Server.TLSPort),
2119 proxyHandler,
2120 "proxy-tls",
2121 cfg.Server.TLSCertFile,
2122 cfg.Server.TLSKeyFile,
2123 errChan,
2124 log,
2125 )
2126 return proxySrv, proxyReady, tlsSrv, tlsReady, nil
2127}
2128
2129// startServer starts an HTTP server, returning the server instance and a channel
2130// that closes once the listening socket is bound. This lets callers wait for the
2131// port to be ready before taking actions that depend on it (e.g. auto-start
2132// pulling from the local registry). The returned *http.Server can be used for
2133// graceful shutdown.
2134func startServer(addr string, handler http.Handler, name string, protocols *http.Protocols, errChan chan<- error, log zerowrap.Logger) (*http.Server, <-chan struct{}) {
2135 ready := make(chan struct{})
2136
2137 server := &http.Server{
2138 Addr: addr,
2139 Handler: handler,
2140 Protocols: protocols,
2141 ReadHeaderTimeout: 10 * time.Second,
2142 ReadTimeout: 5 * time.Minute,
2143 WriteTimeout: 5 * time.Minute,
2144 IdleTimeout: 120 * time.Second,
2145 MaxHeaderBytes: 1 << 20,
2146 }
2147
2148 go func() {
2149 log.Info().Str("address", addr).Msgf("%s server starting", name)
2150
2151 ln, err := net.Listen("tcp", addr)
2152 if err != nil {
2153 errChan <- fmt.Errorf("%s server error: %w", name, err)
2154 return
2155 }
2156 close(ready) // signal: port is bound and accepting connections
2157
2158 if err := server.Serve(ln); err != nil && err != http.ErrServerClosed {
2159 errChan <- fmt.Errorf("%s server error: %w", name, err)
2160 }
2161 }()
2162
2163 return server, ready
2164}
2165
2166// startTLSServer starts an HTTPS server with the provided certificate and key.
2167// It returns the server instance (for graceful shutdown) and a channel that closes once the port is bound.
2168func startTLSServer(addr string, handler http.Handler, name, certFile, keyFile string, errChan chan<- error, log zerowrap.Logger) (*http.Server, <-chan struct{}) {
2169 ready := make(chan struct{})
2170
2171 server := &http.Server{
2172 Addr: addr,
2173 Handler: handler,
2174 ReadHeaderTimeout: 10 * time.Second,
2175 ReadTimeout: 5 * time.Minute,
2176 WriteTimeout: 5 * time.Minute,
2177 IdleTimeout: 120 * time.Second,
2178 MaxHeaderBytes: 1 << 20,
2179 }
2180
2181 go func() {
2182 log.Info().
2183 Str("address", addr).
2184 Str("cert_file", certFile).
2185 Str("key_file", keyFile).
2186 Msgf("%s server starting", name)
2187
2188 ln, err := net.Listen("tcp", addr)
2189 if err != nil {
2190 errChan <- fmt.Errorf("%s server error: %w", name, err)
2191 return
2192 }
2193 close(ready)
2194
2195 if err := server.ServeTLS(ln, certFile, keyFile); err != nil && err != http.ErrServerClosed {
2196 errChan <- fmt.Errorf("%s server error: %w", name, err)
2197 }
2198 }()
2199
2200 return server, ready
2201}
2202
2203// SendReloadSignal sends SIGUSR1 to the running Gordon process.
2204func SendReloadSignal() error {
2205 process, _, err := findRunningProcess()
2206 if err != nil {
2207 return err
2208 }
2209
2210 if err := process.Signal(syscall.SIGUSR1); err != nil {
2211 return fmt.Errorf("failed to send reload signal: %w", err)
2212 }
2213
2214 return nil
2215}
2216
2217// getDeployRequestFile returns the path to the deploy request file.
2218func getDeployRequestFile() string {
2219 return filepath.Join(os.TempDir(), "gordon-deploy-request")
2220}
2221
2222// writeDeployRequestFile creates the deploy request file exclusively with retry.
2223// This prevents race conditions when multiple deploy commands run simultaneously.
2224func writeDeployRequestFile(path string, data []byte, timeout time.Duration) error {
2225 deadline := time.Now().Add(timeout)
2226
2227 for {
2228 f, err := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0600)
2229 if err == nil {
2230 _, writeErr := f.Write(data)
2231 f.Close()
2232 if writeErr != nil {
2233 _ = os.Remove(path)
2234 return writeErr
2235 }
2236 return nil
2237 }
2238
2239 if os.IsExist(err) {
2240 if time.Now().After(deadline) {
2241 return fmt.Errorf("deploy request file still present after timeout; another deploy may be in progress")
2242 }
2243 time.Sleep(100 * time.Millisecond)
2244 continue
2245 }
2246
2247 return err
2248 }
2249}
2250
2251// SendDeploySignal triggers a manual deploy for a specific route via SIGUSR2.
2252// Returns the domain name on success for the caller to display.
2253func SendDeploySignal(domain string) (string, error) {
2254 deployFile := getDeployRequestFile()
2255 if err := writeDeployRequestFile(deployFile, []byte(domain), 5*time.Second); err != nil {
2256 return "", fmt.Errorf("failed to write deploy request: %w", err)
2257 }
2258
2259 // Find PID and send SIGUSR2
2260 process, _, err := findRunningProcess()
2261 if err != nil {
2262 _ = os.Remove(deployFile)
2263 return "", err
2264 }
2265
2266 if err := process.Signal(syscall.SIGUSR2); err != nil {
2267 _ = os.Remove(deployFile)
2268 return "", fmt.Errorf("failed to send deploy signal: %w", err)
2269 }
2270
2271 return domain, nil
2272}
2273
2274// readDeployRequest reads and removes the deploy request file atomically.
2275// Returns empty string if file doesn't exist (may have been consumed by another handler).
2276func readDeployRequest() (string, error) {
2277 deployFile := getDeployRequestFile()
2278
2279 // Rename to a temp file first to make the read-and-delete atomic
2280 tmpFile := deployFile + ".processing"
2281 if err := os.Rename(deployFile, tmpFile); err != nil {
2282 if os.IsNotExist(err) {
2283 return "", fmt.Errorf("deploy request file not found (may have been processed already)")
2284 }
2285 return "", fmt.Errorf("failed to acquire deploy request: %w", err)
2286 }
2287
2288 data, err := os.ReadFile(tmpFile)
2289 _ = os.Remove(tmpFile) // Always clean up
2290 if err != nil {
2291 return "", fmt.Errorf("failed to read deploy request: %w", err)
2292 }
2293
2294 return string(data), nil
2295}
2296
2297// createPidFile creates a PID file for the Gordon process.
2298// SECURITY: Prefers secure locations (XDG_RUNTIME_DIR, ~/.gordon/run) over /tmp
2299// to prevent symlink attacks and unauthorized access.
2300func createPidFile(log zerowrap.Logger) string {
2301 pid := os.Getpid()
2302
2303 // SECURITY: Prioritize secure locations over /tmp
2304 var locations []string
2305
2306 // Try secure runtime directory first
2307 if runtimeDir, err := getSecureRuntimeDir(); err == nil {
2308 locations = append(locations, filepath.Join(runtimeDir, "gordon.pid"))
2309 }
2310
2311 // Fall back to home directory
2312 if homeDir, err := os.UserHomeDir(); err == nil {
2313 locations = append(locations, filepath.Join(homeDir, ".gordon", "gordon.pid"))
2314 }
2315
2316 // Last resort: /tmp (least secure due to world-writable)
2317 locations = append(locations, filepath.Join(os.TempDir(), "gordon.pid"))
2318
2319 for _, location := range locations {
2320 // Ensure parent directory exists with secure permissions
2321 if err := os.MkdirAll(filepath.Dir(location), 0700); err != nil {
2322 continue
2323 }
2324 if err := os.WriteFile(location, []byte(fmt.Sprintf("%d", pid)), 0600); err == nil {
2325 log.Debug().Str("pid_file", location).Int("pid", pid).Msg("created PID file")
2326 return location
2327 }
2328 }
2329
2330 log.Warn().Int("pid", pid).Msg("failed to create PID file in any location")
2331 return ""
2332}
2333
2334// removePidFile removes the PID file.
2335func removePidFile(pidFile string, log zerowrap.Logger) {
2336 if err := os.Remove(pidFile); err != nil {
2337 log.Warn().Err(err).Str("pid_file", pidFile).Msg("failed to remove PID file")
2338 } else {
2339 log.Debug().Str("pid_file", pidFile).Msg("removed PID file")
2340 }
2341}
2342
2343func pidFileLocations() []string {
2344 var locations []string
2345 seen := make(map[string]struct{})
2346 add := func(path string) {
2347 if path == "" {
2348 return
2349 }
2350 if _, ok := seen[path]; ok {
2351 return
2352 }
2353 seen[path] = struct{}{}
2354 locations = append(locations, path)
2355 }
2356
2357 // Check secure runtime directory first
2358 if runtimeDir, err := getSecureRuntimeDir(); err == nil {
2359 add(filepath.Join(runtimeDir, "gordon.pid"))
2360 }
2361
2362 // Also check canonical /run/user/<uid> runtime path. This handles cases where
2363 // Gordon started under systemd user services with runtime dir available, but
2364 // CLI invocations (e.g. non-interactive SSH) don't have XDG_RUNTIME_DIR set.
2365 runtimeByUID := filepath.Join("/run/user", strconv.Itoa(os.Getuid()), "gordon", "gordon.pid")
2366 add(runtimeByUID)
2367
2368 // Check explicit XDG_RUNTIME_DIR if present in this process env.
2369 if runtimeDir := os.Getenv("XDG_RUNTIME_DIR"); runtimeDir != "" {
2370 add(filepath.Join(runtimeDir, "gordon.pid"))
2371 }
2372
2373 // Check home directory
2374 if homeDir, err := os.UserHomeDir(); err == nil {
2375 add(filepath.Join(homeDir, ".gordon", "gordon.pid"))
2376 // Legacy location for backward compatibility
2377 add(filepath.Join(homeDir, ".gordon.pid"))
2378 }
2379
2380 // Legacy /tmp locations for backward compatibility
2381 add(filepath.Join(os.TempDir(), "gordon.pid"))
2382 add("/tmp/gordon.pid")
2383
2384 return locations
2385}
2386
2387// findRunningPidFile returns the first PID file whose PID belongs to a live process.
2388// Stale/invalid PID files are ignored and removed when possible.
2389func findRunningPidFile() (string, int, error) {
2390 return findRunningPidFileInLocations(pidFileLocations())
2391}
2392
2393func findRunningPidFileInLocations(locations []string) (string, int, error) {
2394 foundAny := false
2395
2396 for _, location := range locations {
2397 pidBytes, err := os.ReadFile(location)
2398 if err != nil {
2399 continue
2400 }
2401
2402 foundAny = true
2403
2404 var pid int
2405 if _, err := fmt.Sscanf(string(pidBytes), "%d", &pid); err != nil || pid <= 0 {
2406 _ = os.Remove(location)
2407 continue
2408 }
2409
2410 if isProcessAlive(pid) {
2411 return location, pid, nil
2412 }
2413
2414 _ = os.Remove(location)
2415 }
2416
2417 if foundAny {
2418 return "", 0, fmt.Errorf("found stale gordon PID file(s), is Gordon running?")
2419 }
2420
2421 return "", 0, fmt.Errorf("gordon PID file not found, is Gordon running?")
2422}
2423
2424func findRunningProcess() (*os.Process, int, error) {
2425 _, pid, err := findRunningPidFile()
2426 if err != nil {
2427 return nil, 0, err
2428 }
2429
2430 process, err := os.FindProcess(pid)
2431 if err != nil {
2432 return nil, 0, fmt.Errorf("failed to find process: %w", err)
2433 }
2434
2435 return process, pid, nil
2436}
2437
2438func isProcessAlive(pid int) bool {
2439 process, err := os.FindProcess(pid)
2440 if err != nil {
2441 return false
2442 }
2443
2444 err = process.Signal(syscall.Signal(0))
2445 if err == nil {
2446 return true
2447 }
2448
2449 return errors.Is(err, syscall.EPERM)
2450}
2451
2452// loadConfig loads configuration from file and sets defaults.
2453func loadConfig(v *viper.Viper, configPath string) error {
2454 v.SetDefault("server.port", 80)
2455 v.SetDefault("server.registry_port", 5000)
2456 v.SetDefault("server.tls_enabled", false)
2457 v.SetDefault("server.tls_port", 443)
2458 v.SetDefault("server.tls_cert_file", "")
2459 v.SetDefault("server.tls_key_file", "")
2460 v.SetDefault("server.data_dir", DefaultDataDir())
2461 v.SetDefault("server.runtime", "auto")
2462 v.SetDefault("logging.level", "info")
2463 v.SetDefault("logging.format", "console")
2464 v.SetDefault("logging.file.enabled", false)
2465 v.SetDefault("logging.file.max_size", 100)
2466 v.SetDefault("logging.file.max_backups", 3)
2467 v.SetDefault("logging.file.max_age", 28)
2468 v.SetDefault("logging.container_logs.enabled", true)
2469 v.SetDefault("logging.container_logs.dir", "")
2470 v.SetDefault("logging.container_logs.max_size", 100)
2471 v.SetDefault("logging.container_logs.max_backups", 3)
2472 v.SetDefault("logging.container_logs.max_age", 28)
2473 v.SetDefault("env.dir", "") // defaults to {data_dir}/env when empty
2474 v.SetDefault("auth.enabled", true)
2475 // Note: auth.type defaults to "token" (the only supported mode)
2476 v.SetDefault("auth.secrets_backend", "")
2477 v.SetDefault("auth.token_expiry", "720h")
2478 v.SetDefault("api.rate_limit.enabled", true)
2479 v.SetDefault("api.rate_limit.global_rps", 500)
2480 v.SetDefault("api.rate_limit.per_ip_rps", 50)
2481 v.SetDefault("api.rate_limit.burst", 100)
2482 v.SetDefault("auto_route.enabled", false)
2483 v.SetDefault("network_isolation.enabled", false)
2484 v.SetDefault("network_isolation.network_prefix", "gordon")
2485 v.SetDefault("volumes.auto_create", true)
2486 v.SetDefault("volumes.prefix", "gordon")
2487 v.SetDefault("volumes.preserve", true)
2488 v.SetDefault("deploy.pull_policy", container.PullPolicyIfTagChanged)
2489 v.SetDefault("backups.enabled", false)
2490 v.SetDefault("backups.schedule", string(domain.ScheduleDaily))
2491 v.SetDefault("backups.storage_dir", "")
2492 v.SetDefault("backups.retention.hourly", 0)
2493 v.SetDefault("backups.retention.daily", 0)
2494 v.SetDefault("backups.retention.weekly", 0)
2495 v.SetDefault("backups.retention.monthly", 0)
2496 v.SetDefault("images.prune.enabled", false)
2497 v.SetDefault("images.prune.schedule", string(domain.ScheduleDaily))
2498 v.SetDefault("images.prune.keep_last", domain.DefaultImagePruneKeepLast)
2499 v.SetDefault("telemetry.enabled", false)
2500 v.SetDefault("telemetry.endpoint", "")
2501 v.SetDefault("telemetry.auth_token", "")
2502 v.SetDefault("telemetry.traces", true)
2503 v.SetDefault("telemetry.metrics", true)
2504 v.SetDefault("telemetry.logs", true)
2505 v.SetDefault("telemetry.trace_sample_rate", 1.0)
2506
2507 v.SetDefault("server.max_concurrent_connections", -1) // -1 = use default (10000), 0 = no limit
2508 v.SetDefault("server.registry_allowed_ips", []string{})
2509 v.SetDefault("deploy.readiness_delay", "5s")
2510 v.SetDefault("deploy.readiness_mode", "auto")
2511 v.SetDefault("deploy.health_timeout", "90s")
2512 v.SetDefault("deploy.stabilization_delay", "2s")
2513 v.SetDefault("deploy.tcp_probe_timeout", "30s")
2514 v.SetDefault("deploy.http_probe_timeout", "60s")
2515 v.SetDefault("deploy.attachment_readiness_timeout", "30s")
2516 v.SetDefault("deploy.drain_mode", "auto")
2517 v.SetDefault("deploy.drain_timeout", "30s")
2518
2519 ConfigureViper(v, configPath)
2520
2521 if err := v.ReadInConfig(); err != nil {
2522 if _, ok := err.(viper.ConfigFileNotFoundError); !ok {
2523 return fmt.Errorf("failed to read config file: %w", err)
2524 }
2525 }
2526
2527 v.SetEnvPrefix("GORDON")
2528 v.SetEnvKeyReplacer(strings.NewReplacer(".", "_"))
2529 v.AutomaticEnv()
2530
2531 return nil
2532}