| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "github.com/carlosframework/platform/internal/hostactors" |
| 9 | execlease "github.com/carlosframework/platform/internal/hostexec" |
| 10 | "github.com/carlosframework/platform/internal/hostfloor" |
| 11 | "github.com/carlosframework/platform/internal/release" |
| 12 | "github.com/carlosframework/platform/internal/rootparticipation" |
| 13 | "os" |
| 14 | "path/filepath" |
| 15 | "slices" |
| 16 | "time" |
| 17 | ) |
| 18 | |
| 19 | type guardTransaction struct { |
| 20 | Roster json.RawMessage `json:"roster"` |
| 21 | ID string `json:"id"` |
| 22 | Actor string `json:"actor"` |
| 23 | Enrollment string `json:"enrollment"` |
| 24 | } |
| 25 | |
| 26 | func (a *systemApplier) guarded() bool { return a.floor != nil } |
| 27 | func (a *systemApplier) activeFloor() (bool, error) { |
| 28 | if !a.guarded() { |
| 29 | return false, nil |
| 30 | } |
| 31 | p, _, err := a.floor.Load() |
| 32 | return p.Mode != "unpublished", err |
| 33 | } |
| 34 | func (a *systemApplier) guardImage(path string) error { |
| 35 | _, err := a.guardImageHash(path) |
| 36 | return err |
| 37 | } |
| 38 | |
| 39 | // guardImageHash is guardImage, returning the verified image's SHA-256 |
| 40 | // ("" when no floor guards this box). |
| 41 | func (a *systemApplier) guardImageHash(path string) (string, error) { |
| 42 | if !a.guarded() { |
| 43 | return "", nil |
| 44 | } |
| 45 | open := hostfloor.OpenProtected |
| 46 | if a.guardOpen != nil { |
| 47 | open = a.guardOpen |
| 48 | } |
| 49 | f, err := open(path) |
| 50 | if err != nil { |
| 51 | return "", err |
| 52 | } |
| 53 | defer f.Close() |
| 54 | hash, err := a.floor.VerifyFile(f, a.platform) |
| 55 | if err != nil { |
| 56 | return "", err |
| 57 | } |
| 58 | p, _, err := a.floor.Load() |
| 59 | if err != nil || p.Mode == "unpublished" { |
| 60 | return hash, err |
| 61 | } |
| 62 | tx, err := a.guardTransaction() |
| 63 | if err != nil { |
| 64 | if os.IsNotExist(err) { |
| 65 | if j, e := readSystemJournal(a.optDir); e == nil && j == nil && path == systemBinaryPath(a.optDir) { |
| 66 | return hash, nil |
| 67 | } |
| 68 | } |
| 69 | return "", err |
| 70 | } |
| 71 | var roster hostactors.Roster |
| 72 | if release.VerifyRootEnvelope(a.floor.PublicKey, tx.Roster, &roster) != nil || roster.Validate(a.floor.Bucket) != nil { |
| 73 | return "", hostfloor.ErrFloor |
| 74 | } |
| 75 | reservation, ok := roster.Activations[tx.ID] |
| 76 | _, floor, err := a.floor.Load() |
| 77 | if err != nil { |
| 78 | return "", err |
| 79 | } |
| 80 | if !ok || reservation.FloorSHA256 != floor.Digest || reservation.Actors[tx.Actor] != tx.Enrollment || !slices.Contains(reservation.Hashes[tx.Actor], hash) { |
| 81 | return "", hostfloor.ErrFloor |
| 82 | } |
| 83 | return hash, nil |
| 84 | } |
| 85 | |
| 86 | // guardTarget is the transaction's target image: the middle of the three |
| 87 | // hashes prepareGuard selected the reservation by, read from the signed |
| 88 | // roster root persisted with the transaction. |
| 89 | func (a *systemApplier) guardTarget(tx guardTransaction) (string, error) { |
| 90 | var roster hostactors.Roster |
| 91 | if release.VerifyRootEnvelope(a.floor.PublicKey, tx.Roster, &roster) != nil || roster.Validate(a.floor.Bucket) != nil { |
| 92 | return "", hostfloor.ErrFloor |
| 93 | } |
| 94 | hashes := roster.Activations[tx.ID].Hashes[tx.Actor] |
| 95 | if len(hashes) != 3 || hashes[1] == "" { |
| 96 | return "", hostfloor.ErrFloor |
| 97 | } |
| 98 | return hashes[1], nil |
| 99 | } |
| 100 | |
| 101 | // guardFinished reports whether a transaction with no journal open has |
| 102 | // finished: the verified image on disk is its target, or root's marker |
| 103 | // records that it rolled back (Opus re-review M1-4). A transaction that is |
| 104 | // neither never changed the box: it was refused or deferred before its |
| 105 | // first destructive step, or abandoned at the rename. |
| 106 | func (a *systemApplier) guardFinished(tx guardTransaction) (bool, error) { |
| 107 | hash, err := a.guardImageHash(systemBinaryPath(a.optDir)) |
| 108 | if err != nil { |
| 109 | return false, err |
| 110 | } |
| 111 | begun, err := hostfloor.ReadBegun(a.floor.Dir) |
| 112 | if err == nil { |
| 113 | if begun.ID != tx.ID { |
| 114 | return false, hostfloor.ErrFloor |
| 115 | } |
| 116 | if begun.RolledBack { |
| 117 | return true, nil |
| 118 | } |
| 119 | } else if !errors.Is(err, os.ErrNotExist) { |
| 120 | return false, err |
| 121 | } |
| 122 | target, err := a.guardTarget(tx) |
| 123 | if err != nil { |
| 124 | return false, err |
| 125 | } |
| 126 | return hash == target, nil |
| 127 | } |
| 128 | |
| 129 | // settleGuard closes a transaction with no journal open. A finished one is |
| 130 | // completed, once no controller is active, as before. One that never |
| 131 | // changed the box records no completion: root withdraws its own marker and |
| 132 | // transaction, so neither wedges the box nor reports the old image as the |
| 133 | // result, and the console's reservation stays held for the agent's retry |
| 134 | // or an operator. The withdrawal does not wait for controllers: it gives up |
| 135 | // only root's intent, on a box it did not change. |
| 136 | func (a *systemApplier) settleGuard(ctx context.Context, tx guardTransaction, beforeCompletion func() error) error { |
| 137 | finished, err := a.guardFinished(tx) |
| 138 | if err != nil { |
| 139 | return err |
| 140 | } |
| 141 | if !finished { |
| 142 | // The marker goes inside the settlement, after root's census, so a |
| 143 | // census that refuses leaves it with the rest of the transaction. |
| 144 | return a.floor.EndParticipation(ctx, tx.ID, func() error { |
| 145 | if err := hostfloor.RemoveDurable(filepath.Join(a.floor.Dir, "update-begun.json")); err != nil { |
| 146 | return err |
| 147 | } |
| 148 | return hostfloor.RemoveDurable(filepath.Join(a.floor.Dir, "activation.json")) |
| 149 | }) |
| 150 | } |
| 151 | if beforeCompletion != nil { |
| 152 | if err := beforeCompletion(); err != nil { |
| 153 | return err |
| 154 | } |
| 155 | } |
| 156 | // Completion is durable before the marker disappears. Replaying either |
| 157 | // write is harmless; readers never report completion with a marker present. |
| 158 | if err = hostfloor.Durable(filepath.Join(a.floor.Dir, "completed.json"), []byte(tx.ID), 0644); err != nil { |
| 159 | return err |
| 160 | } |
| 161 | return a.floor.EndParticipation(ctx, tx.ID, func() error { return hostfloor.RemoveDurable(filepath.Join(a.floor.Dir, "update-begun.json")) }) |
| 162 | } |
| 163 | func (a *systemApplier) guardTransaction() (guardTransaction, error) { |
| 164 | raw, err := os.ReadFile(filepath.Join(a.floor.Dir, "activation.json")) |
| 165 | if err != nil { |
| 166 | return guardTransaction{}, err |
| 167 | } |
| 168 | var tx guardTransaction |
| 169 | if json.Unmarshal(raw, &tx) != nil || tx.ID == "" || tx.Enrollment == "" || tx.Actor == "" { |
| 170 | return guardTransaction{}, hostfloor.ErrFloor |
| 171 | } |
| 172 | return tx, nil |
| 173 | } |
| 174 | func (a *systemApplier) beginGuard(ctx context.Context, target *release.SystemPointer) error { |
| 175 | if !a.guarded() { |
| 176 | return nil |
| 177 | } |
| 178 | p, _, err := a.floor.Load() |
| 179 | if err != nil { |
| 180 | return err |
| 181 | } |
| 182 | if p.Mode == "unpublished" { |
| 183 | return nil |
| 184 | } |
| 185 | if err := a.prepareGuard(ctx, target); err != nil { |
| 186 | return err |
| 187 | } |
| 188 | tx, err := a.guardTransaction() |
| 189 | if err != nil { |
| 190 | return err |
| 191 | } |
| 192 | // The signed reservation is persisted by the root approval step, never by |
| 193 | // the worker's request. A new transaction must still find it in the roster. |
| 194 | store, pub, err := a.openStore(ctx) |
| 195 | if err != nil { |
| 196 | return err |
| 197 | } |
| 198 | _, local, err := a.floor.Load() |
| 199 | if err != nil { |
| 200 | return err |
| 201 | } |
| 202 | remote, err := hostfloor.Read(ctx, store, pub, a.floor.Bucket) |
| 203 | if err != nil || remote.Digest != local.Digest { |
| 204 | return hostfloor.ErrFloor |
| 205 | } |
| 206 | d := hostactors.Directory{Store: store, PublicKey: pub, Bucket: a.floor.Bucket} |
| 207 | live, err := d.Read(ctx) |
| 208 | if err != nil { |
| 209 | return err |
| 210 | } |
| 211 | h, ok := live.Roster.Activations[tx.ID] |
| 212 | if !ok || h.Actors[tx.Actor] != tx.Enrollment { |
| 213 | return hostfloor.ErrFloor |
| 214 | } |
| 215 | var saved hostactors.Roster |
| 216 | if release.VerifyRootEnvelope(pub, tx.Roster, &saved) != nil { |
| 217 | return hostfloor.ErrFloor |
| 218 | } |
| 219 | x, _ := json.Marshal(h) |
| 220 | y, _ := json.Marshal(saved.Activations[tx.ID]) |
| 221 | if string(x) != string(y) { |
| 222 | return hostfloor.ErrFloor |
| 223 | } |
| 224 | return a.floor.Begin(ctx, tx.ID, a.dataDir) |
| 225 | } |
| 226 | |
| 227 | // checkGuardRecovery runs with the apply lock held. No-journal markers can |
| 228 | // only precede the first destructive step or follow durable journal removal. |
| 229 | // Both are terminal after the selected current image has been verified. |
| 230 | func (a *systemApplier) checkGuardRecovery(ctx context.Context) error { |
| 231 | if !a.guarded() { |
| 232 | return a.checkUnarmedParticipationRecovery(ctx) |
| 233 | } |
| 234 | p, _, err := a.floor.Load() |
| 235 | if err != nil { |
| 236 | return err |
| 237 | } |
| 238 | if p.Mode == "unpublished" { |
| 239 | return a.checkUnarmedParticipationRecovery(ctx) |
| 240 | } |
| 241 | j, err := readSystemJournal(a.optDir) |
| 242 | if err != nil { |
| 243 | return err |
| 244 | } |
| 245 | marker := filepath.Join(a.floor.Dir, "update-begun.json") |
| 246 | raw, err := os.ReadFile(marker) |
| 247 | if errors.Is(err, os.ErrNotExist) && j == nil { |
| 248 | tx, txErr := a.guardTransaction() |
| 249 | if errors.Is(txErr, os.ErrNotExist) { |
| 250 | return nil |
| 251 | } |
| 252 | if txErr != nil { |
| 253 | return txErr |
| 254 | } |
| 255 | completed, readErr := os.ReadFile(filepath.Join(a.floor.Dir, "completed.json")) |
| 256 | if readErr == nil && string(completed) == tx.ID { |
| 257 | // Completion recorded and root's marker gone. With root's update |
| 258 | // record gone too the transaction is settled: a replay has |
| 259 | // nothing left to close, and runs no census. |
| 260 | open, err := a.updateRecordOpen(ctx) |
| 261 | if err != nil || !open { |
| 262 | return err |
| 263 | } |
| 264 | return a.settleWithoutJournal(ctx, tx.ID, a.floor.EndParticipation(ctx, tx.ID, nil)) |
| 265 | } |
| 266 | if readErr != nil && !errors.Is(readErr, os.ErrNotExist) { |
| 267 | return readErr |
| 268 | } |
| 269 | return a.settleWithoutJournal(ctx, tx.ID, a.settleGuard(ctx, tx, func() error { return a.controllersRecovered(ctx) })) |
| 270 | } |
| 271 | if err != nil { |
| 272 | return err |
| 273 | } |
| 274 | tx, err := a.guardTransaction() |
| 275 | if err != nil { |
| 276 | return err |
| 277 | } |
| 278 | var begun struct { |
| 279 | ID string `json:"id"` |
| 280 | } |
| 281 | if json.Unmarshal(raw, &begun) != nil || begun.ID != tx.ID { |
| 282 | return hostfloor.ErrFloor |
| 283 | } |
| 284 | if j == nil { |
| 285 | return a.settleWithoutJournal(ctx, tx.ID, a.settleGuard(ctx, tx, func() error { return a.controllersRecovered(ctx) })) |
| 286 | } |
| 287 | if err := a.controllersRecovered(ctx); err != nil { |
| 288 | return err |
| 289 | } |
| 290 | if j.Phase == systemPhaseBackingUp { |
| 291 | return a.guardImage(systemBinaryPath(a.optDir)) |
| 292 | } |
| 293 | return a.guardImage(systemPrevPath(a.optDir)) |
| 294 | } |
| 295 | |
| 296 | // errSettlementDeferred marks a settlement that root's generation census |
| 297 | // refused with no journal open. No apply is half done and the files are |
| 298 | // whole, so there is nothing for recovery to restore and no reason to keep |
| 299 | // the edge down (its drop-in Requires= the recover unit). The transaction |
| 300 | // evidence stays as it was: the reservation, root's marker and update |
| 301 | // record, so every new claim acquisition stays refused, and the next apply, |
| 302 | // or recover on the next edge start, settles it. Run under --recover logs |
| 303 | // it and exits 0; the apply returns it as before. beginExecutableChange |
| 304 | // treats a refused census as a deferral too (systemparticipation.go). |
| 305 | var errSettlementDeferred = errors.New("participation settlement deferred") |
| 306 | |
| 307 | // errAcquisitionBarrier marks a deferral that could not hold claim |
| 308 | // acquisitions refused (Ledger.HoldUpdate failed). It is never swallowed: |
| 309 | // recover must not exit 0 with acquisitions open (Astra rounds 2 and 3). |
| 310 | var errAcquisitionBarrier = errors.New("acquisition barrier") |
| 311 | |
| 312 | // settleWithoutJournal is the result of one settlement made with no |
| 313 | // journal open: the pending-controller recovery, the completion or |
| 314 | // withdrawal, and the end of root's update record, whose census runs again |
| 315 | // inside EndParticipation; and, after an apply's journal is durably |
| 316 | // removed, the settlement that follows it (closeGuardJournal). A result |
| 317 | // that is only a census refusal, on a box with root participation, is |
| 318 | // wrapped in errSettlementDeferred, once acquisitions are held refused: |
| 319 | // root's update record is kept, or, where the refused census was the one |
| 320 | // that would have begun it, written under barrier's id (Ledger.HoldUpdate). |
| 321 | // Anything else is returned unchanged: an image that does not verify, |
| 322 | // malformed state, a live or unreadable controller, a durable write that |
| 323 | // failed, a barrier that could not be written, and every refusal on a box |
| 324 | // without a census, which sees exactly main's recovery (ruling 6). |
| 325 | func (a *systemApplier) settleWithoutJournal(ctx context.Context, barrier string, err error) error { |
| 326 | if err == nil || !onlyCensusRefusal(err) || ctx.Err() != nil { |
| 327 | return err |
| 328 | } |
| 329 | required, checkErr := a.participationDisk().ParticipationRequired(ctx) |
| 330 | if checkErr != nil { |
| 331 | // Whether a barrier is needed could not be established (the store |
| 332 | // did not open, a retained census did not read): that is a barrier |
| 333 | // failure too, never a plain refusal a caller may log (Astra |
| 334 | // round 4). Both errors are kept. |
| 335 | return errors.Join(err, fmt.Errorf("%w: participation required: %w", errAcquisitionBarrier, checkErr)) |
| 336 | } |
| 337 | if !required { |
| 338 | return err |
| 339 | } |
| 340 | if holdErr := a.participationDisk().WithParticipation(ctx, func(ctx context.Context, l rootparticipation.Ledger) error { |
| 341 | return l.HoldUpdate(ctx, barrier) |
| 342 | }, func() error { return rootparticipation.ErrUnavailable }); holdErr != nil { |
| 343 | return errors.Join(err, fmt.Errorf("%w: %w", errAcquisitionBarrier, holdErr)) |
| 344 | } |
| 345 | return fmt.Errorf("%w: no journal is open, so the files are whole; root's update record holds claim acquisitions refused until the settlement passes: %w", errSettlementDeferred, err) |
| 346 | } |
| 347 | |
| 348 | // updateRecordOpen reports whether root's update record is there, on a box |
| 349 | // with root participation. Without one it is false: main's legacy path. |
| 350 | func (a *systemApplier) updateRecordOpen(ctx context.Context) (bool, error) { |
| 351 | open := false |
| 352 | err := a.participationDisk().WithParticipation(ctx, func(_ context.Context, l rootparticipation.Ledger) error { |
| 353 | _, err := l.UpdateID() |
| 354 | if errors.Is(err, os.ErrNotExist) { |
| 355 | return nil |
| 356 | } |
| 357 | open = err == nil |
| 358 | return err |
| 359 | }, func() error { open = true; return nil }) |
| 360 | return open, err |
| 361 | } |
| 362 | |
| 363 | // deferSettlement reports whether Run, under --recover, leaves err's |
| 364 | // settlement to a later run instead of failing: only an errSettlementDeferred. |
| 365 | // It logs why. |
| 366 | func (a *systemApplier) deferSettlement(recover bool, err error) bool { |
| 367 | if !recover || !errors.Is(err, errSettlementDeferred) { |
| 368 | return false |
| 369 | } |
| 370 | a.ops.logf("system apply: recover: %v; the edge may start, and the next apply or recover settles it", err) |
| 371 | return true |
| 372 | } |
| 373 | |
| 374 | // settleMarker settles the update-begun marker at the end of a run |
| 375 | // (settleUpdateBegun) and decides what its outcome means for the run: a |
| 376 | // deferral under --recover is logged and left to a later run; a barrier |
| 377 | // that could not be written joins the run's error, whatever the mode, |
| 378 | // because acquisitions would otherwise be open with the settlement |
| 379 | // unmade; a deferral under the apply is logged, as every marker failure |
| 380 | // was on main, with acquisitions held refused by the barrier. |
| 381 | func (a *systemApplier) settleMarker(ctx context.Context, recover bool, retErr *error) { |
| 382 | err := a.settleUpdateBegun(ctx) |
| 383 | if err == nil || a.deferSettlement(recover, err) { |
| 384 | return |
| 385 | } |
| 386 | if errors.Is(err, errAcquisitionBarrier) { |
| 387 | *retErr = errors.Join(*retErr, err) |
| 388 | return |
| 389 | } |
| 390 | a.ops.logf("system apply: %v", err) |
| 391 | } |
| 392 | |
| 393 | func (a *systemApplier) endGuard(ctx context.Context) error { |
| 394 | if !a.guarded() { |
| 395 | return nil |
| 396 | } |
| 397 | p, _, err := a.floor.Load() |
| 398 | if err != nil || p.Mode == "unpublished" { |
| 399 | return err |
| 400 | } |
| 401 | tx, err := a.guardTransaction() |
| 402 | if err != nil { |
| 403 | return err |
| 404 | } |
| 405 | if j, err := readSystemJournal(a.optDir); err != nil { |
| 406 | return err |
| 407 | } else if j != nil { |
| 408 | return hostfloor.ErrFloor |
| 409 | } |
| 410 | return a.settleGuard(ctx, tx, nil) |
| 411 | } |
| 412 | |
| 413 | func (a *systemApplier) controllersIdle(ctx context.Context) error { |
| 414 | return a.participationDisk().ControllersIdle(ctx, a.dataDir) |
| 415 | } |
| 416 | |
| 417 | var recoverPendingControllers = recoverRootExecution |
| 418 | |
| 419 | // controllersRecovered lets root recover a provably dead pending controller |
| 420 | // before refusing, whether or not a journal is open; a live or unreadable one |
| 421 | // still refuses. |
| 422 | func (a *systemApplier) controllersRecovered(ctx context.Context) error { |
| 423 | err := a.controllersIdle(ctx) |
| 424 | if err == nil { |
| 425 | return nil |
| 426 | } |
| 427 | guardDir := systemGuardDir(a.optDir) |
| 428 | if a.floor != nil { |
| 429 | guardDir = a.floor.Dir |
| 430 | } |
| 431 | required, checkErr := a.participationDisk().ParticipationRequired(ctx) |
| 432 | if checkErr != nil { |
| 433 | return errors.Join(err, checkErr) |
| 434 | } |
| 435 | recover := recoverPendingControllers |
| 436 | if !required { |
| 437 | recover = execlease.RecoverPending |
| 438 | } |
| 439 | if recoveryErr := recover(ctx, a.dataDir, guardDir); recoveryErr != nil { |
| 440 | return errors.Join(err, recoveryErr) |
| 441 | } |
| 442 | return a.controllersIdle(ctx) |
| 443 | } |
| 444 | func (a *systemApplier) closeGuardJournal(ctx context.Context) error { |
| 445 | required, err := a.participationDisk().ParticipationRequired(ctx) |
| 446 | if err != nil { |
| 447 | return err |
| 448 | } |
| 449 | if !required { |
| 450 | if !a.guarded() { |
| 451 | a.removeJournal() |
| 452 | return nil |
| 453 | } |
| 454 | if err := hostfloor.RemoveDurable(a.file("journal.json")); err != nil { |
| 455 | return err |
| 456 | } |
| 457 | return a.endGuard(ctx) |
| 458 | } |
| 459 | active, err := a.activeFloor() |
| 460 | if err != nil { |
| 461 | return err |
| 462 | } |
| 463 | if err := hostfloor.RemoveDurable(a.file("journal.json")); err != nil { |
| 464 | return err |
| 465 | } |
| 466 | // The journal is gone and the files are whole: from here a settlement |
| 467 | // the census refuses is deferred as with no journal open. |
| 468 | if !active { |
| 469 | return a.settleWithoutJournal(ctx, "recover-legacy", a.clearUpdateBegun(ctx)) |
| 470 | } |
| 471 | err = a.endGuard(ctx) |
| 472 | if err == nil || !onlyCensusRefusal(err) { |
| 473 | return err |
| 474 | } |
| 475 | tx, txErr := a.guardTransaction() |
| 476 | if txErr != nil { |
| 477 | return errors.Join(err, txErr) |
| 478 | } |
| 479 | return a.settleWithoutJournal(ctx, tx.ID, err) |
| 480 | } |
| 481 | func (a *systemApplier) guardRun(ctx context.Context, wait time.Duration, name string, args ...string) (string, error) { |
| 482 | if id, required, err := a.rootMoves.read(); err != nil { |
| 483 | return "", err |
| 484 | } else if required { |
| 485 | binary, err := readRootFloorBinary(a.optDir, name) |
| 486 | if err != nil { |
| 487 | return "", err |
| 488 | } |
| 489 | floor, err := readEffectiveMovesFloor(a.rootMoves, a.optDir, a.dataDir) |
| 490 | if err != nil { |
| 491 | return "", err |
| 492 | } |
| 493 | if !binary.supports(floor) { |
| 494 | repair := name == systemBinaryPath(a.optDir) || name == filepath.Join(systemGuardDir(a.optDir), "carlos") |
| 495 | if name == systemPrevPath(a.optDir) && rootFloorAllowsPrevIdentity(a.optDir, id) == nil { |
| 496 | repair = true |
| 497 | } |
| 498 | if !binary.supports(nil) || !repair { |
| 499 | return "", errRootFloorHistory |
| 500 | } |
| 501 | rootFloorAlert() |
| 502 | } |
| 503 | } |
| 504 | |
| 505 | if !a.guarded() { |
| 506 | return a.ops.run(ctx, wait, name, args...) |
| 507 | } |
| 508 | if err := a.guardImage(name); err != nil { |
| 509 | return "", err |
| 510 | } |
| 511 | bounded, cancel := context.WithTimeout(ctx, wait) |
| 512 | defer cancel() |
| 513 | return a.floor.Run(bounded, name, a.platform, args...) |
| 514 | } |
| 515 | func installFloorHook(a *systemApplier) error { |
| 516 | dir := systemGuardDir(a.optDir) |
| 517 | fi, err := os.Lstat(dir) |
| 518 | if errors.Is(err, os.ErrNotExist) { |
| 519 | return nil |
| 520 | } |
| 521 | if err != nil { |
| 522 | return err |
| 523 | } |
| 524 | if !fi.IsDir() { |
| 525 | return hostfloor.ErrFloor |
| 526 | } |
| 527 | entries, err := os.ReadDir(dir) |
| 528 | if err != nil { |
| 529 | return err |
| 530 | } |
| 531 | // Main's participation marker predates floor arming, is written through |
| 532 | // update-begun.json.tmp, and leaves an empty directory after completion. |
| 533 | // A torn or leftover write of it is still legacy state; only a marker |
| 534 | // naming a floor transaction ("id") or a floor artifact needs a policy. |
| 535 | if legacyGuardEntries(dir, entries) { |
| 536 | return nil |
| 537 | } |
| 538 | d, err := floorDisk(a.optDir) |
| 539 | if err != nil { |
| 540 | return err |
| 541 | } |
| 542 | a.floor = &d |
| 543 | _, _, err = d.Load() |
| 544 | return err |
| 545 | } |
| 546 | |
| 547 | func (a *systemApplier) prepareGuard(ctx context.Context, target *release.SystemPointer) error { |
| 548 | enrollment, err := hostfloor.ReadEnrollment(filepath.Join(a.etcDir, "carlos", "host-actor.json")) |
| 549 | if err != nil { |
| 550 | return err |
| 551 | } |
| 552 | store, pub, err := a.openStore(ctx) |
| 553 | if err != nil { |
| 554 | return err |
| 555 | } |
| 556 | raw, _, err := store.Get(ctx, hostactors.Key) |
| 557 | if err != nil { |
| 558 | return err |
| 559 | } |
| 560 | var roster hostactors.Roster |
| 561 | if release.VerifyRootEnvelope(pub, raw, &roster) != nil || roster.Validate(a.floor.Bucket) != nil { |
| 562 | return hostfloor.ErrFloor |
| 563 | } |
| 564 | b, ok := roster.Boxes[enrollment.Box] |
| 565 | if !ok || b.Enrollment != enrollment.Enrollment || b.Cloud.InstanceID != enrollment.InstanceID { |
| 566 | return hostfloor.ErrFloor |
| 567 | } |
| 568 | actor := "box:" + enrollment.Box |
| 569 | id := "" |
| 570 | for name, h := range roster.Activations { |
| 571 | if h.Publication != nil && target.ActivationID != name { |
| 572 | continue |
| 573 | } |
| 574 | hashes := h.Hashes[actor] |
| 575 | if h.Actors[actor] == enrollment.Enrollment && len(hashes) == 3 && hashes[1] == target.Current.Artifacts[a.platform].SHA256 { |
| 576 | if id != "" { |
| 577 | return hostfloor.ErrFloor |
| 578 | } |
| 579 | id = name |
| 580 | } |
| 581 | } |
| 582 | if id == "" { |
| 583 | return hostfloor.ErrFloor |
| 584 | } |
| 585 | identityKey, err := release.IdentityKey(target.Current.ReleaseTag) |
| 586 | if err != nil { |
| 587 | return err |
| 588 | } |
| 589 | identity, _, err := store.Get(ctx, identityKey) |
| 590 | if err != nil { |
| 591 | return err |
| 592 | } |
| 593 | _, floor, err := a.floor.Load() |
| 594 | if err != nil { |
| 595 | return err |
| 596 | } |
| 597 | if err = hostfloor.CheckImage(pub, floor, identity, target.Current.Version, a.platform, target.Current.Artifacts[a.platform].SHA256); err != nil { |
| 598 | return err |
| 599 | } |
| 600 | if err = hostfloor.Durable(filepath.Join(a.floor.Dir, "identities", target.Current.ReleaseTag+".json"), identity, 0644); err != nil { |
| 601 | return err |
| 602 | } |
| 603 | txRaw, _ := json.Marshal(guardTransaction{raw, id, actor, enrollment.Enrollment}) |
| 604 | return hostfloor.Durable(filepath.Join(a.floor.Dir, "activation.json"), txRaw, 0644) |
| 605 | } |
| 606 | |
| 607 | // legacyGuardEntries reports whether the guard directory holds nothing but |
| 608 | // main's region participation marker, its temporary file, or both, with no |
| 609 | // floor transaction ID in the marker. |
| 610 | func legacyGuardEntries(dir string, entries []os.DirEntry) bool { |
| 611 | for _, e := range entries { |
| 612 | if !e.Type().IsRegular() || (e.Name() != updateBegunFile && e.Name() != updateBegunFile+".tmp") { |
| 613 | return false |
| 614 | } |
| 615 | if e.Name() == updateBegunFile { |
| 616 | raw, err := os.ReadFile(filepath.Join(dir, e.Name())) |
| 617 | if err != nil { |
| 618 | return false |
| 619 | } |
| 620 | var marker map[string]json.RawMessage |
| 621 | if json.Unmarshal(raw, &marker) == nil { |
| 622 | if _, ok := marker["id"]; ok { |
| 623 | return false |
| 624 | } |
| 625 | } |
| 626 | } |
| 627 | } |
| 628 | return true |
| 629 | } |
| 630 | |
amadan