| 1 | package console |
| 2 | |
| 3 | // Door keys (#175 P7): the credential an app's own processes present at a |
| 4 | // box's edge-local door. The console's database is the authority |
| 5 | // (doorkeys_store.go). The bucket holds a projection of it at |
| 6 | // doorkey.RecordKey, signed with the deployment's root release key, carrying |
| 7 | // a revision and a 72-hour expiry, and re-signed every 12 hours. |
| 8 | // |
| 9 | // Exactly one code path writes that projection: publishDoorKeys. It always |
| 10 | // rebuilds from the database and never reads a key back out of the bucket, |
| 11 | // so a record planted or copied there by a bucket writer is overwritten, not |
| 12 | // extended (Astra round 1, finding 1). |
| 13 | |
| 14 | import ( |
| 15 | "context" |
| 16 | "crypto/ed25519" |
| 17 | "errors" |
| 18 | "fmt" |
| 19 | "log" |
| 20 | "net/http" |
| 21 | "os" |
| 22 | "sync" |
| 23 | "time" |
| 24 | |
| 25 | "github.com/carlosframework/platform/internal/doorkey" |
| 26 | "github.com/carlosframework/platform/internal/release" |
| 27 | ) |
| 28 | |
| 29 | const ( |
| 30 | // doorKeyResignEvery is how often every record is re-signed with a fresh |
| 31 | // expiry. A record lives doorkey.Lifetime (72 h), so a console that is |
| 32 | // down for up to 60 hours past its last re-sign leaves no door refusing. |
| 33 | doorKeyResignEvery = 12 * time.Hour |
| 34 | // doorKeyPublishAttempts bounds publishDoorKeys' compare-and-swap loop. |
| 35 | doorKeyPublishAttempts = 3 |
| 36 | // doorKeyPublishBudget bounds one publication started from a request or |
| 37 | // from the re-sign pass. |
| 38 | doorKeyPublishBudget = 20 * time.Second |
| 39 | ) |
| 40 | |
| 41 | // Copy approved in review (plan Task 8), 2026-09-29. |
| 42 | const ( |
| 43 | doorKeyReadOnlyMsg = "no signing key on this host. the console is running read-only" |
| 44 | doorKeyNoBucketMsg = "no deployment bucket on this host. the console is running without one" |
| 45 | ) |
| 46 | |
| 47 | // doorPublishLock is the app's publication mutex, created on first use. |
| 48 | func (s *Server) doorPublishLock(sqid, app string) *sync.Mutex { |
| 49 | s.doorPublishMu.Lock() |
| 50 | defer s.doorPublishMu.Unlock() |
| 51 | if s.doorPublish == nil { |
| 52 | s.doorPublish = map[string]*sync.Mutex{} |
| 53 | } |
| 54 | k := sqid + "/" + app |
| 55 | mu := s.doorPublish[k] |
| 56 | if mu == nil { |
| 57 | mu = &sync.Mutex{} |
| 58 | s.doorPublish[k] = mu |
| 59 | } |
| 60 | return mu |
| 61 | } |
| 62 | |
| 63 | // publishDoorKeys writes the app's signed record from the database. The |
| 64 | // protocol (plan Task 2, round 2 findings 4 to 6): |
| 65 | // |
| 66 | // 1. Hold the app's publication mutex throughout. A renewal and a revoke |
| 67 | // for the same app are serialized, and each builds its record from the |
| 68 | // database inside the lock, so a renewal can never put back a key set |
| 69 | // older than one already published. |
| 70 | // 2. Read the bucket's object and its ETag. The bucket's revision is the one |
| 71 | // a valid signature vouches for (expired or not), and zero for anything |
| 72 | // absent or unverifiable: a tampered object is overwritten, never |
| 73 | // trusted. |
| 74 | // 3. Read the live keys and the revision in one transaction. If the |
| 75 | // database is below the bucket (a restored database, or a second |
| 76 | // writer), advance its counter past the bucket first and read again. An |
| 77 | // equal revision is the ordinary renewal and is re-published as it is. |
| 78 | // 4. Sign the record with a fresh expiry and Put it against the ETag read in |
| 79 | // step 2 (create-only when there was no object). |
| 80 | // 5. A lost compare-and-swap starts again at step 2, rebuilding from the |
| 81 | // database, never reusing signed bytes. After doorKeyPublishAttempts |
| 82 | // losses the conflict is returned. |
| 83 | func (s *Server) publishDoorKeys(ctx context.Context, acct *Account, app string) error { |
| 84 | if len(s.cfg.Key) != ed25519.PrivateKeySize { |
| 85 | return errors.New("console: door keys: no signing key") |
| 86 | } |
| 87 | if s.cfg.Releases == nil { |
| 88 | return errors.New("console: door keys: no deployment bucket") |
| 89 | } |
| 90 | pub, _ := s.cfg.Key.Public().(ed25519.PublicKey) |
| 91 | mu := s.doorPublishLock(acct.Sqid, app) |
| 92 | mu.Lock() |
| 93 | defer mu.Unlock() |
| 94 | // The record read, the revision advance and the conditional write are |
| 95 | // one app mutation, so a foreign operation on the app refuses the whole |
| 96 | // publication rather than half of it. |
| 97 | return s.appMutation(ctx, acct, app, "door-keys-publish", func(ctx context.Context) error { |
| 98 | return s.publishDoorKeysLocked(ctx, acct, app, pub) |
| 99 | }) |
| 100 | } |
| 101 | |
| 102 | // publishDoorKeysLocked is steps 2 to 5 of publishDoorKeys, run with the |
| 103 | // app's publication mutex held. |
| 104 | func (s *Server) publishDoorKeysLocked(ctx context.Context, acct *Account, app string, pub ed25519.PublicKey) error { |
| 105 | key := doorkey.RecordKey(acct.Sqid, app) |
| 106 | var lost error |
| 107 | for attempt := 0; attempt < doorKeyPublishAttempts; attempt++ { |
| 108 | wire, etag, err := s.cfg.Releases.Get(ctx, key) |
| 109 | var bucketRev uint64 |
| 110 | switch { |
| 111 | case errors.Is(err, os.ErrNotExist): |
| 112 | etag = "" |
| 113 | case err != nil: |
| 114 | return fmt.Errorf("console: door keys %s/%s: read: %w", acct.Sqid, app, err) |
| 115 | default: |
| 116 | if rev, verr := doorkey.VerifyRevision(pub, wire); verr == nil { |
| 117 | bucketRev = rev |
| 118 | } else { |
| 119 | log.Printf("console: door keys %s/%s: overwriting an unverifiable object: %v", acct.Sqid, app, verr) |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | keys, rev, err := s.store.DoorKeysLiveContext(ctx, acct.ID, app) |
| 124 | if err != nil { |
| 125 | return fmt.Errorf("console: door keys %s/%s: %w", acct.Sqid, app, err) |
| 126 | } |
| 127 | if rev < bucketRev { |
| 128 | if _, err := s.store.DoorKeyAdvanceRevisionContext(ctx, acct.ID, app, bucketRev); err != nil { |
| 129 | return fmt.Errorf("console: door keys %s/%s: advance past %d: %w", acct.Sqid, app, bucketRev, err) |
| 130 | } |
| 131 | log.Printf("console: door keys %s/%s: database revision %d was below the bucket's %d; advanced past it", |
| 132 | acct.Sqid, app, rev, bucketRev) |
| 133 | if keys, rev, err = s.store.DoorKeysLiveContext(ctx, acct.ID, app); err != nil { |
| 134 | return fmt.Errorf("console: door keys %s/%s: %w", acct.Sqid, app, err) |
| 135 | } |
| 136 | } |
| 137 | if rev == 0 { |
| 138 | return nil // never minted, and nothing verifiable in the bucket: nothing to publish |
| 139 | } |
| 140 | |
| 141 | now := time.Now().UTC() |
| 142 | body, err := doorkey.Sign(s.cfg.Key, doorkey.Record{V: 1, Account: acct.Sqid, App: app, |
| 143 | Revision: rev, Keys: keys, SignedAt: now, ExpiresAt: now.Add(doorkey.Lifetime)}) |
| 144 | if err != nil { |
| 145 | return fmt.Errorf("console: door keys %s/%s: sign: %w", acct.Sqid, app, err) |
| 146 | } |
| 147 | _, err = s.cfg.Releases.Put(ctx, key, body, etag) |
| 148 | if err == nil { |
| 149 | return nil |
| 150 | } |
| 151 | if !errors.Is(err, release.ErrCASConflict) { |
| 152 | return fmt.Errorf("console: door keys %s/%s: write: %w", acct.Sqid, app, err) |
| 153 | } |
| 154 | lost = err |
| 155 | } |
| 156 | return fmt.Errorf("console: door keys %s/%s: %d attempts: %w", acct.Sqid, app, doorKeyPublishAttempts, lost) |
| 157 | } |
| 158 | |
| 159 | // publishDoorKeysDetached runs publishDoorKeys for a request whose database |
| 160 | // write has already landed: a client that hangs up must not cancel the |
| 161 | // publication of a change that is already the truth. |
| 162 | func (s *Server) publishDoorKeysDetached(r *http.Request, acct *Account, app string) error { |
| 163 | ctx, cancel := context.WithTimeout(context.WithoutCancel(r.Context()), doorKeyPublishBudget) |
| 164 | defer cancel() |
| 165 | return s.publishDoorKeys(ctx, acct, app) |
| 166 | } |
| 167 | |
| 168 | // republishDoorKeys is publishDoorKeysDetached for a change made elsewhere |
| 169 | // that revoked an app name's keys in passing (purge, move): best-effort, |
| 170 | // because that change has already landed and is not undone by a publish |
| 171 | // that fails. The re-sign pass publishes within 12 hours regardless. A |
| 172 | // console with nothing to sign with or nowhere to write skips it. |
| 173 | func (s *Server) republishDoorKeys(r *http.Request, acct *Account, app string) { |
| 174 | if len(s.cfg.Key) != ed25519.PrivateKeySize || s.cfg.Releases == nil { |
| 175 | return |
| 176 | } |
| 177 | if err := s.publishDoorKeysDetached(r, acct, app); err != nil { |
| 178 | log.Printf("console: door keys %s/%s: publish after revoking (the re-sign pass retries): %v", acct.Sqid, app, err) |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | // doorKeyWritable answers the 503 a console that cannot publish gives, before |
| 183 | // anything is written. |
| 184 | func (s *Server) doorKeyWritable(w http.ResponseWriter) bool { |
| 185 | if len(s.cfg.Key) == 0 { |
| 186 | jsonError(w, http.StatusServiceUnavailable, doorKeyReadOnlyMsg) |
| 187 | return false |
| 188 | } |
| 189 | if s.cfg.Releases == nil { |
| 190 | jsonError(w, http.StatusServiceUnavailable, doorKeyNoBucketMsg) |
| 191 | return false |
| 192 | } |
| 193 | return true |
| 194 | } |
| 195 | |
| 196 | // handleAPIDoorKeyMint mints one door key for the app. Person-only, the |
| 197 | // instance-token rule: a credential may never mint a credential. The zero |
| 198 | // Set at the registration refuses every service role (authcli.go), and |
| 199 | // apiPerson is the inner wall. |
| 200 | // |
| 201 | // The key is shown exactly once. Only its sha256 is stored. |
| 202 | func (s *Server) handleAPIDoorKeyMint(w http.ResponseWriter, r *http.Request) { |
| 203 | if !s.apiPerson(w, r) { |
| 204 | return |
| 205 | } |
| 206 | addr, acct, app, ok := s.apiAccountApp(w, r) |
| 207 | if !ok { |
| 208 | return |
| 209 | } |
| 210 | p, _ := principalFrom(r.Context()) |
| 211 | if !s.allowWrite(p, time.Now()) { |
| 212 | jsonError(w, http.StatusTooManyRequests, "too many requests. wait a minute and try again") |
| 213 | return |
| 214 | } |
| 215 | if !s.doorKeyWritable(w) { |
| 216 | return |
| 217 | } |
| 218 | var req struct { |
| 219 | Label string `json:"label"` |
| 220 | } |
| 221 | if !readJSON(w, r, &req) { |
| 222 | return |
| 223 | } |
| 224 | raw, entry, err := doorkey.NewToken(acct.Sqid, app) |
| 225 | if err != nil { |
| 226 | log.Printf("console: door key mint %s/%s: %v", acct.Sqid, app, err) |
| 227 | jsonError(w, http.StatusInternalServerError, "could not mint the door key. try again") |
| 228 | return |
| 229 | } |
| 230 | entry.Label = cleanLabel(req.Label) |
| 231 | if _, err := s.store.DoorKeyInsertContext(r.Context(), acct.ID, app, entry); err != nil { |
| 232 | if errors.Is(err, errDoorKeysFull) { |
| 233 | jsonError(w, http.StatusConflict, fmt.Sprintf( |
| 234 | "this app already has %d door keys. revoke one before you mint another", doorkey.MaxKeys)) |
| 235 | return |
| 236 | } |
| 237 | log.Printf("console: door key mint %s/%s: %v", acct.Sqid, app, err) |
| 238 | jsonError(w, http.StatusInternalServerError, "could not mint the door key. try again") |
| 239 | return |
| 240 | } |
| 241 | if err := s.publishDoorKeysDetached(r, acct, app); err != nil { |
| 242 | // The secret is about to be lost (it is never shown), so the row it |
| 243 | // hashes is revoked rather than left holding one of the app's slots |
| 244 | // until the next re-sign publishes a key nobody has. |
| 245 | log.Printf("console: door key mint %s/%s: publish: %v", acct.Sqid, app, err) |
| 246 | if _, rerr := s.store.DoorKeyRevokeContext(r.Context(), acct.ID, app, entry.ID); rerr != nil { |
| 247 | log.Printf("console: door key mint %s/%s: withdraw unpublished key %s: %v", acct.Sqid, app, entry.ID, rerr) |
| 248 | } |
| 249 | jsonError(w, http.StatusBadGateway, "could not publish the door key. try again") |
| 250 | return |
| 251 | } |
| 252 | s.audit(acct, app, addr, "door-key", "minted door key "+entry.ID+" for "+app+" (cli)") |
| 253 | s.bumpSerial(r.Context()) |
| 254 | writeJSON(w, http.StatusCreated, map[string]any{ |
| 255 | "key": raw, "id": entry.ID, "account": acct.Sqid, "app": app, |
| 256 | }) |
| 257 | } |
| 258 | |
| 259 | // handleAPIDoorKeyList lists the app's live keys: id, label and when each |
| 260 | // was minted. Never a digest. |
| 261 | func (s *Server) handleAPIDoorKeyList(w http.ResponseWriter, r *http.Request) { |
| 262 | if !s.apiPerson(w, r) { |
| 263 | return |
| 264 | } |
| 265 | _, acct, app, ok := s.apiAccountApp(w, r) |
| 266 | if !ok { |
| 267 | return |
| 268 | } |
| 269 | keys, _, err := s.store.DoorKeysLiveContext(r.Context(), acct.ID, app) |
| 270 | if err != nil { |
| 271 | log.Printf("console: door key list %s/%s: %v", acct.Sqid, app, err) |
| 272 | jsonError(w, http.StatusInternalServerError, "could not list the door keys. try again") |
| 273 | return |
| 274 | } |
| 275 | type listed struct { |
| 276 | ID string `json:"id"` |
| 277 | Label string `json:"label"` |
| 278 | CreatedAt time.Time `json:"created_at"` |
| 279 | } |
| 280 | out := make([]listed, 0, len(keys)) |
| 281 | for _, k := range keys { |
| 282 | out = append(out, listed{ID: k.ID, Label: k.Label, CreatedAt: k.CreatedAt}) |
| 283 | } |
| 284 | writeJSON(w, http.StatusOK, map[string]any{"keys": out}) |
| 285 | } |
| 286 | |
| 287 | // handleAPIDoorKeyRevoke revokes one live key and publishes the record |
| 288 | // without it. |
| 289 | func (s *Server) handleAPIDoorKeyRevoke(w http.ResponseWriter, r *http.Request) { |
| 290 | if !s.apiPerson(w, r) { |
| 291 | return |
| 292 | } |
| 293 | addr, acct, app, ok := s.apiAccountApp(w, r) |
| 294 | if !ok { |
| 295 | return |
| 296 | } |
| 297 | p, _ := principalFrom(r.Context()) |
| 298 | if !s.allowWrite(p, time.Now()) { |
| 299 | jsonError(w, http.StatusTooManyRequests, "too many requests. wait a minute and try again") |
| 300 | return |
| 301 | } |
| 302 | if !s.doorKeyWritable(w) { |
| 303 | return |
| 304 | } |
| 305 | id := r.PathValue("id") |
| 306 | if _, err := s.store.DoorKeyRevokeContext(r.Context(), acct.ID, app, id); err != nil { |
| 307 | if errors.Is(err, os.ErrNotExist) { |
| 308 | jsonError(w, http.StatusNotFound, "no live door key with that id") |
| 309 | return |
| 310 | } |
| 311 | log.Printf("console: door key revoke %s/%s %s: %v", acct.Sqid, app, id, err) |
| 312 | jsonError(w, http.StatusInternalServerError, "could not revoke the door key. try again") |
| 313 | return |
| 314 | } |
| 315 | s.audit(acct, app, addr, "door-key-revoke", "revoked door key "+id+" for "+app+" (cli)") |
| 316 | if err := s.publishDoorKeysDetached(r, acct, app); err != nil { |
| 317 | // The database is the truth and the key is revoked there; the |
| 318 | // re-sign pass publishes it on its next run. |
| 319 | log.Printf("console: door key revoke %s/%s %s: publish: %v", acct.Sqid, app, id, err) |
| 320 | jsonError(w, http.StatusBadGateway, |
| 321 | "the door key is revoked but not yet published. boxes keep admitting it until the console publishes again, within 12 hours") |
| 322 | return |
| 323 | } |
| 324 | s.bumpSerial(r.Context()) |
| 325 | writeJSON(w, http.StatusOK, map[string]any{"revoked": id}) |
| 326 | } |
| 327 | |
| 328 | // resignDoorKeys re-publishes every app's record with a fresh expiry. A |
| 329 | // failure for one app is logged and does not stop the others. |
| 330 | func (s *Server) resignDoorKeys(ctx context.Context) { |
| 331 | apps, err := s.store.DoorKeyApps() |
| 332 | if err != nil { |
| 333 | log.Printf("console: door keys re-sign: %v", err) |
| 334 | return |
| 335 | } |
| 336 | for _, a := range apps { |
| 337 | pctx, cancel := context.WithTimeout(ctx, doorKeyPublishBudget) |
| 338 | err := s.publishDoorKeys(pctx, &Account{ID: a.AccountID, Sqid: a.Sqid}, a.App) |
| 339 | cancel() |
| 340 | if err != nil { |
| 341 | log.Printf("console: door keys re-sign %s/%s: %v", a.Sqid, a.App, err) |
| 342 | } |
| 343 | } |
| 344 | } |
| 345 | |
| 346 | // StartDoorKeyResign re-signs every door-key record now and then every 12 |
| 347 | // hours until ctx is done. Started from cmd/console beside the other passes, |
| 348 | // and only on a console that holds the signing key: a read-only console has |
| 349 | // nothing to sign with. |
| 350 | func (s *Server) StartDoorKeyResign(ctx context.Context) { |
| 351 | if len(s.cfg.Key) == 0 || s.cfg.Releases == nil { |
| 352 | return |
| 353 | } |
| 354 | t := time.NewTicker(doorKeyResignEvery) |
| 355 | defer t.Stop() |
| 356 | for { |
| 357 | s.resignDoorKeys(ctx) |
| 358 | select { |
| 359 | case <-ctx.Done(): |
| 360 | return |
| 361 | case <-t.C: |
| 362 | } |
| 363 | } |
| 364 | } |
| 365 | |
amadan