| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/ecdh" |
| 6 | "encoding/json" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "io" |
| 10 | "io/fs" |
| 11 | "log" |
| 12 | "os" |
| 13 | "path/filepath" |
| 14 | "strings" |
| 15 | |
| 16 | "github.com/carlosframework/platform/internal/activator" |
| 17 | "github.com/carlosframework/platform/internal/appconfig" |
| 18 | "github.com/carlosframework/platform/internal/hostbox" |
| 19 | "github.com/carlosframework/platform/internal/hostclaim" |
| 20 | "github.com/carlosframework/platform/internal/hostguard" |
| 21 | "github.com/carlosframework/platform/internal/lease" |
| 22 | "github.com/carlosframework/platform/internal/moves" |
| 23 | "github.com/carlosframework/platform/internal/moves/archive" |
| 24 | "github.com/carlosframework/platform/internal/provision" |
| 25 | "github.com/carlosframework/platform/internal/registry" |
| 26 | ) |
| 27 | |
| 28 | // The installer's helpers (moves spec §7.6 local, restored; shared by the |
| 29 | // landing and the dirty return, movesinstall.go): the sealed archive read by |
| 30 | // exact version, verified and extracted into the install's staging, then |
| 31 | // placed at the live paths inside the host guard; the sealed timers; the |
| 32 | // database install from the build scratch with its marker, inside the |
| 33 | // guard; the build again when the journal lost it; and the fsync and |
| 34 | // atomic-write helpers. The phases are in movesinstall.go, the landing's |
| 35 | // checking and pass in movesland.go. |
| 36 | |
| 37 | // errLandingChanged is a guarded install's refusal when the claim, the |
| 38 | // record, the installation record or the destination is no longer this |
| 39 | // install's (a landing's or a dirty return's): nothing is written, and the |
| 40 | // pass ends with the error. |
| 41 | var errLandingChanged = errors.New("the installation is no longer this install's") |
| 42 | |
| 43 | // errFilesOccupied is the files install's refusal when the files directory |
| 44 | // holds anything, or <host>.env exists, that this install did not write: |
| 45 | // it is not the install's to clear, so the install's refusal is written |
| 46 | // instead (a landing's rollback marker, a return's stuck mark). |
| 47 | var errFilesOccupied = errors.New("the files paths hold what this install did not write") |
| 48 | |
| 49 | // capture reads the installation record and requires it at the holder's |
| 50 | // incarnation and stamp: the identity every guarded install runs under. |
| 51 | func (ir *installRun) capture(holder hostclaim.Holder) (hostguard.Captured, error) { |
| 52 | captured, err := hostguard.Capture(ir.cfg().DataDir, ir.rec.Host) |
| 53 | if err != nil { |
| 54 | return captured, fmt.Errorf("read the installation record: %w", err) |
| 55 | } |
| 56 | if captured.Incarnation != holder.Incarnation || captured.Stamp != holder.Stamp { |
| 57 | return captured, fmt.Errorf("move %s: the installation record is at stamp %d of %s, not this %s's %d", ir.g.ID, captured.Stamp, captured.Incarnation, ir.src.Name, holder.Stamp) |
| 58 | } |
| 59 | return captured, nil |
| 60 | } |
| 61 | |
| 62 | // guarded runs fn inside the host guard for this install's installation |
| 63 | // (final review, finding 2; hostbox ColdRestore.Install's pattern): under |
| 64 | // the host's lock, the installation record still active at the captured |
| 65 | // stamp, no fence over it, the claim read fresh and still passing the |
| 66 | // install's claim check at phase (a landing's op at phase with this box |
| 67 | // executing; a return's restoring/claimed with this box executing) with the |
| 68 | // holder at the captured stamp, and the record read fresh and still this |
| 69 | // install at phase (installSource.RecordPhase: a landing with no rollback |
| 70 | // reason; a dirty return). fn re-checks its destination and makes the |
| 71 | // install; it must not call anything that takes the guard itself |
| 72 | // (park.Local's retirement and inventory), nor wait on the network. |
| 73 | func (ir *installRun) guarded(ctx context.Context, captured hostguard.Captured, phase string, fn func() error) error { |
| 74 | return ir.guardedWith(ctx, captured, phase, nil, fn) |
| 75 | } |
| 76 | |
| 77 | // guardedWith is guarded with one more check of the fresh claim (nil for |
| 78 | // none), run after the install's own. |
| 79 | func (ir *installRun) guardedWith(ctx context.Context, captured hostguard.Captured, phase string, extra func(*hostclaim.Claim) error, fn func() error) error { |
| 80 | check := func(c *hostclaim.Claim) error { |
| 81 | if err := ir.src.ClaimCheck(phase)(c); err != nil { |
| 82 | return err |
| 83 | } |
| 84 | if extra != nil { |
| 85 | if err := extra(c); err != nil { |
| 86 | return err |
| 87 | } |
| 88 | } |
| 89 | if c.Holder == nil || c.Holder.Stamp != captured.Stamp { |
| 90 | return fmt.Errorf("%w: the claim's holder is not at the installation stamp %d", errLandingChanged, captured.Stamp) |
| 91 | } |
| 92 | rec, err := provision.GetRecord(ctx, ir.s, ir.rec.Account, ir.rec.App, ir.rec.Host) |
| 93 | if err != nil { |
| 94 | return fmt.Errorf("re-read the record: %w", err) |
| 95 | } |
| 96 | if at, err := ir.src.RecordPhase(rec); err != nil || at != phase { |
| 97 | return fmt.Errorf("%w: the record is no longer the %s of %s at %s", errLandingChanged, ir.src.Name, ir.g.ID, phase) |
| 98 | } |
| 99 | return nil |
| 100 | } |
| 101 | return ir.guard.With(ctx, ir.rec.Host, captured, check, fn) |
| 102 | } |
| 103 | |
| 104 | // stagedFiles names the staged extraction by identity (device and inode): |
| 105 | // the files directory's and the env file's, Env "" when the archive |
| 106 | // carries none. A rename keeps an identity, so the same record, journalled |
| 107 | // as the install receipt once the tree is placed, is what tells this |
| 108 | // landing's own tree at the live paths from anything else there. |
| 109 | type stagedFiles struct { |
| 110 | Files string `json:"files"` |
| 111 | Env string `json:"env,omitempty"` |
| 112 | } |
| 113 | |
| 114 | // stageDir is the op's own extraction staging, <JournalDir>/extract/: the |
| 115 | // one place a retry clears (final review, finding 1). |
| 116 | func (ir *installRun) stageDir() string { return filepath.Join(ir.journalDir(), "extract") } |
| 117 | |
| 118 | // journalledStaged reads a stagedFiles journal step. |
| 119 | func journalledStaged(j *moves.Journal, step string) (stagedFiles, error) { |
| 120 | var st stagedFiles |
| 121 | raw, ok := j.Steps[step] |
| 122 | if !ok { |
| 123 | return st, fmt.Errorf("the journal has no %s step", step) |
| 124 | } |
| 125 | if err := json.Unmarshal([]byte(raw), &st); err != nil || st.Files == "" { |
| 126 | return st, fmt.Errorf("the journal's %s step %q does not name the staged files (%v)", step, raw, err) |
| 127 | } |
| 128 | return st, nil |
| 129 | } |
| 130 | |
| 131 | // installFiles is local's sub-step (3) (final review, finding 1). The |
| 132 | // archive is extracted into the op's staging, never at the live paths, and |
| 133 | // its identities journalled as `staged`; a retry before that clears only |
| 134 | // the staging and extracts again. Then, inside the host guard, the staged |
| 135 | // tree and env are moved to the files directory and <host>.env when those |
| 136 | // hold nothing this install did not write (absent, or an empty directory, |
| 137 | // or this install's own tree by identity after a crash mid-move). Anything |
| 138 | // else there is not the install's to clear: nothing is moved and the |
| 139 | // rollback is marked. The install receipt (`files_placed`) is journalled, |
| 140 | // the empty staging removed, and only then `extracted`. |
| 141 | func (ir *installRun) installFiles(ctx context.Context, j *moves.Journal, mark func(step, value string) error, captured hostguard.Captured, seal moves.Seal, want provision.MoveFilesDigest, dir, envPath string) (restoreOutcome, error) { |
| 142 | stage := ir.stageDir() |
| 143 | stFiles, stEnv := filepath.Join(stage, "files"), filepath.Join(stage, "env") |
| 144 | var staged stagedFiles |
| 145 | if j.Done(landStepStaged) { |
| 146 | var err error |
| 147 | if staged, err = journalledStaged(j, landStepStaged); err != nil { |
| 148 | return restoreWaiting, err |
| 149 | } |
| 150 | } else { |
| 151 | // Only the op's own staging is cleared: nothing at the live paths |
| 152 | // is touched before the guarded move. |
| 153 | if err := removeTree(stage); err != nil { |
| 154 | return restoreWaiting, fmt.Errorf("clear the extraction staging: %w", err) |
| 155 | } |
| 156 | if err := os.MkdirAll(stage, 0o700); err != nil { |
| 157 | return restoreWaiting, err |
| 158 | } |
| 159 | if err := ir.extractArchive(ctx, seal, want, stFiles, stEnv); err != nil { |
| 160 | return restoreWaiting, fmt.Errorf("extract the archive: %w", err) |
| 161 | } |
| 162 | var err error |
| 163 | if staged.Files, err = pathIdentity(stFiles); err != nil { |
| 164 | return restoreWaiting, err |
| 165 | } |
| 166 | if present, err := regularFile(stEnv); err != nil { |
| 167 | return restoreWaiting, err |
| 168 | } else if present { |
| 169 | if staged.Env, err = pathIdentity(stEnv); err != nil { |
| 170 | return restoreWaiting, err |
| 171 | } |
| 172 | } |
| 173 | body, err := json.Marshal(staged) |
| 174 | if err != nil { |
| 175 | return restoreWaiting, err |
| 176 | } |
| 177 | if err := mark(landStepStaged, string(body)); err != nil { |
| 178 | return restoreWaiting, err |
| 179 | } |
| 180 | } |
| 181 | err := ir.guarded(ctx, captured, ir.src.Phases[3], func() error { |
| 182 | return ir.placeStaged(staged, stFiles, stEnv, dir, envPath) |
| 183 | }) |
| 184 | if errors.Is(err, errFilesOccupied) { |
| 185 | log.Printf("moves: %s: %v; nothing was moved or removed", ir.rec.Host, err) |
| 186 | ok, err := ir.refuse(ctx, refuseFilesOccupied) |
| 187 | if err != nil || !ok { |
| 188 | return restoreWaiting, err |
| 189 | } |
| 190 | return restoreMarked, nil |
| 191 | } |
| 192 | if err != nil { |
| 193 | return restoreWaiting, fmt.Errorf("install the files: %w", err) |
| 194 | } |
| 195 | if err := mark(landStepFilesPlaced, j.Steps[landStepStaged]); err != nil { |
| 196 | return restoreWaiting, err |
| 197 | } |
| 198 | if err := removeTree(stage); err != nil { |
| 199 | return restoreWaiting, fmt.Errorf("remove the extraction staging: %w", err) |
| 200 | } |
| 201 | if err := mark(landStepExtracted, ""); err != nil { |
| 202 | return restoreWaiting, err |
| 203 | } |
| 204 | return restoreInstalled, nil |
| 205 | } |
| 206 | |
| 207 | // placeStaged runs inside the guard. Every destination is decided before |
| 208 | // anything moves, so a refusal changes nothing: the files directory must be |
| 209 | // this install's tree already (by identity: a crash after its move) or |
| 210 | // absent or an empty directory, and <host>.env this install's env already, |
| 211 | // or absent. An empty directory gives its mode to the tree and is replaced, |
| 212 | // as Extract has always done. |
| 213 | func (ir *installRun) placeStaged(staged stagedFiles, stFiles, stEnv, dir, envPath string) error { |
| 214 | filesDone, err := sameIdentity(dir, staged.Files) |
| 215 | if err != nil { |
| 216 | return err |
| 217 | } |
| 218 | var emptyDir bool |
| 219 | var mode fs.FileMode |
| 220 | if !filesDone { |
| 221 | if ok, err := sameIdentity(stFiles, staged.Files); err != nil || !ok { |
| 222 | return fmt.Errorf("the staged files at %s are not the ones journalled (%v)", stFiles, err) |
| 223 | } |
| 224 | if emptyDir, mode, err = emptyOrAbsentDir(dir); err != nil { |
| 225 | return err |
| 226 | } |
| 227 | } |
| 228 | envDone := false |
| 229 | if staged.Env != "" { |
| 230 | if envDone, err = sameIdentity(envPath, staged.Env); err != nil { |
| 231 | return err |
| 232 | } |
| 233 | if !envDone { |
| 234 | if ok, err := sameIdentity(stEnv, staged.Env); err != nil || !ok { |
| 235 | return fmt.Errorf("the staged env at %s is not the one journalled (%v)", stEnv, err) |
| 236 | } |
| 237 | } |
| 238 | } |
| 239 | if !envDone { |
| 240 | if _, err := os.Lstat(envPath); err == nil { |
| 241 | return fmt.Errorf("%w: %s exists", errFilesOccupied, envPath) |
| 242 | } else if !errors.Is(err, os.ErrNotExist) { |
| 243 | return err |
| 244 | } |
| 245 | } |
| 246 | if !filesDone { |
| 247 | if emptyDir { |
| 248 | if err := os.Chmod(stFiles, mode); err != nil { |
| 249 | return err |
| 250 | } |
| 251 | if err := syncPath(stFiles); err != nil { |
| 252 | return err |
| 253 | } |
| 254 | // os.Remove refuses a directory that is no longer empty. |
| 255 | if err := os.Remove(dir); err != nil { |
| 256 | return err |
| 257 | } |
| 258 | } |
| 259 | if err := os.Rename(stFiles, dir); err != nil { |
| 260 | return fmt.Errorf("place the files: %w", err) |
| 261 | } |
| 262 | if err := syncPath(filepath.Dir(dir)); err != nil { |
| 263 | return err |
| 264 | } |
| 265 | if err := ir.fault("after-files-rename"); err != nil { |
| 266 | return err |
| 267 | } |
| 268 | } |
| 269 | if staged.Env != "" && !envDone { |
| 270 | if err := os.Rename(stEnv, envPath); err != nil { |
| 271 | return fmt.Errorf("place the env file: %w", err) |
| 272 | } |
| 273 | if err := syncPath(filepath.Dir(envPath)); err != nil { |
| 274 | return err |
| 275 | } |
| 276 | } |
| 277 | return nil |
| 278 | } |
| 279 | |
| 280 | // emptyOrAbsentDir reports whether path is an empty directory (and its |
| 281 | // mode); absent is false with no error; anything else (a file, a symlink, |
| 282 | // a directory with anything in it) is errFilesOccupied. |
| 283 | func emptyOrAbsentDir(path string) (bool, fs.FileMode, error) { |
| 284 | fi, err := os.Lstat(path) |
| 285 | if errors.Is(err, os.ErrNotExist) { |
| 286 | return false, 0, nil |
| 287 | } |
| 288 | if err != nil { |
| 289 | return false, 0, err |
| 290 | } |
| 291 | if !fi.IsDir() { |
| 292 | return false, 0, fmt.Errorf("%w: %s is not a directory", errFilesOccupied, path) |
| 293 | } |
| 294 | f, err := os.Open(path) |
| 295 | if err != nil { |
| 296 | return false, 0, err |
| 297 | } |
| 298 | names, err := f.Readdirnames(1) |
| 299 | f.Close() |
| 300 | if len(names) > 0 { |
| 301 | return false, 0, fmt.Errorf("%w: %s holds %s", errFilesOccupied, path, names[0]) |
| 302 | } |
| 303 | if err != nil && err != io.EOF { |
| 304 | return false, 0, err |
| 305 | } |
| 306 | return true, fi.Mode().Perm(), nil |
| 307 | } |
| 308 | |
| 309 | // pathIdentity is the device and inode of path itself (not a symlink's |
| 310 | // target), "dev:ino". Off unix there is none, and the landing refuses. |
| 311 | func pathIdentity(path string) (string, error) { |
| 312 | fi, err := os.Lstat(path) |
| 313 | if err != nil { |
| 314 | return "", err |
| 315 | } |
| 316 | dev, ino, ok := fileInode(fi) |
| 317 | if !ok { |
| 318 | return "", fmt.Errorf("%s: this platform cannot identify files", path) |
| 319 | } |
| 320 | return fmt.Sprintf("%d:%d", dev, ino), nil |
| 321 | } |
| 322 | |
| 323 | // sameIdentity reports whether path exists with identity id. |
| 324 | func sameIdentity(path, id string) (bool, error) { |
| 325 | got, err := pathIdentity(path) |
| 326 | if errors.Is(err, os.ErrNotExist) { |
| 327 | return false, nil |
| 328 | } |
| 329 | if err != nil { |
| 330 | return false, err |
| 331 | } |
| 332 | return id != "" && got == id, nil |
| 333 | } |
| 334 | |
| 335 | // removeTree removes a tree this install made under its own journal |
| 336 | // directory, whose directories may carry the archive's restrictive modes: |
| 337 | // each is made writable before its contents are removed. Absent is nothing |
| 338 | // to do. |
| 339 | func removeTree(path string) error { |
| 340 | if _, err := os.Lstat(path); errors.Is(err, os.ErrNotExist) { |
| 341 | return nil |
| 342 | } |
| 343 | filepath.WalkDir(path, func(p string, d fs.DirEntry, err error) error { |
| 344 | if err == nil && d.IsDir() { |
| 345 | os.Chmod(p, 0o700) |
| 346 | } |
| 347 | return nil |
| 348 | }) |
| 349 | return os.RemoveAll(path) |
| 350 | } |
| 351 | |
| 352 | // filesGenMarker is the .files-gen body (§5.1): the op and the incarnation. |
| 353 | type filesGenMarker struct { |
| 354 | Op string `json:"op"` |
| 355 | Incarnation string `json:"incarnation"` |
| 356 | } |
| 357 | |
| 358 | // extractArchive reads the seal's archive and manifest by exact version from |
| 359 | // this box's own bucket, verifies the archive against the manifest and the |
| 360 | // record's digests (archive.Verify: the sealed digest, every entry), and |
| 361 | // extracts it into dir and envPath (archive.Open under the box's secrets |
| 362 | // key, then archive.Extract, which admits nothing it did not read whole). |
| 363 | func (ir *installRun) extractArchive(ctx context.Context, seal moves.Seal, want provision.MoveFilesDigest, dir, envPath string) error { |
| 364 | cfg := ir.cfg() |
| 365 | var sealedObj, manifestObj *moves.SealedObject |
| 366 | for i := range seal.Files { |
| 367 | switch o := &seal.Files[i]; { |
| 368 | case strings.HasSuffix(o.Key, ".sealed"): |
| 369 | sealedObj = o |
| 370 | case strings.HasSuffix(o.Key, ".json"): |
| 371 | manifestObj = o |
| 372 | } |
| 373 | } |
| 374 | if sealedObj == nil || manifestObj == nil { |
| 375 | return errors.New("the seal does not name the archive and its manifest") |
| 376 | } |
| 377 | vs, err := ir.versioned(ctx) |
| 378 | if err != nil { |
| 379 | return err |
| 380 | } |
| 381 | raw, err := vs.GetVersion(ctx, cfg.OwnBucket, manifestObj.Key, manifestObj.VersionID) |
| 382 | if err != nil { |
| 383 | return fmt.Errorf("read the archive manifest: %w", err) |
| 384 | } |
| 385 | var m archive.Manifest |
| 386 | if err := json.Unmarshal(raw, &m); err != nil { |
| 387 | return fmt.Errorf("the archive manifest: %w", err) |
| 388 | } |
| 389 | if m.Op != ir.src.OpID || m.Incarnation != ir.rec.Move.Incarnation || m.Plain != want.Plain || m.Sealed != want.Sealed { |
| 390 | return fmt.Errorf("the archive manifest is for %s/%s (%s, %s), not this move's %s (%s, %s)", m.Op, m.Incarnation, m.Plain, m.Sealed, ir.src.OpID, want.Plain, want.Sealed) |
| 391 | } |
| 392 | if err := os.MkdirAll(ir.journalDir(), 0o700); err != nil { |
| 393 | return err |
| 394 | } |
| 395 | sealedPath := filepath.Join(ir.journalDir(), "archive.sealed") |
| 396 | if err := os.Remove(sealedPath); err != nil && !errors.Is(err, os.ErrNotExist) { |
| 397 | return err |
| 398 | } |
| 399 | defer os.Remove(sealedPath) |
| 400 | if err := downloadVersion(ctx, vs, cfg.OwnBucket, sealedObj.Key, sealedObj.VersionID, sealedPath); err != nil { |
| 401 | return fmt.Errorf("read the archive: %w", err) |
| 402 | } |
| 403 | f, err := os.Open(sealedPath) |
| 404 | if err != nil { |
| 405 | return err |
| 406 | } |
| 407 | defer f.Close() |
| 408 | bind := appconfig.Binding{Account: ir.rec.Account, App: ir.rec.App} |
| 409 | if err := archive.Verify(f, m, cfg.SecretsPriv, bind); err != nil { |
| 410 | return fmt.Errorf("verify the archive: %w", err) |
| 411 | } |
| 412 | return openAndExtract(f, m, cfg.SecretsPriv, bind, dir, envPath) |
| 413 | } |
| 414 | |
| 415 | // openAndExtract pipes the opened archive into Extract, as carriedEnv does. |
| 416 | func openAndExtract(f *os.File, m archive.Manifest, priv *ecdh.PrivateKey, bind appconfig.Binding, dir, envPath string) error { |
| 417 | if _, err := f.Seek(0, io.SeekStart); err != nil { |
| 418 | return err |
| 419 | } |
| 420 | pr, pw := io.Pipe() |
| 421 | done := make(chan error, 1) |
| 422 | go func() { |
| 423 | err := archive.Open(f, m.SealedDataKey, priv, bind, pw) |
| 424 | pw.CloseWithError(err) |
| 425 | done <- err |
| 426 | }() |
| 427 | xerr := archive.Extract(pr, dir, envPath) |
| 428 | pr.CloseWithError(errors.New("extract stopped reading")) |
| 429 | if oerr := <-done; oerr != nil && xerr == nil { |
| 430 | xerr = oerr |
| 431 | } |
| 432 | return xerr |
| 433 | } |
| 434 | |
| 435 | // sealedTimers reads the sealed timers snapshot by exact version and checks |
| 436 | // its digest against the record's; every row is this host's. |
| 437 | func (ir *installRun) sealedTimers(ctx context.Context, seal moves.Seal, wantDigest string) ([]registry.Timer, error) { |
| 438 | cfg := ir.cfg() |
| 439 | if len(seal.Timers) != 1 { |
| 440 | return nil, fmt.Errorf("the seal names %d timers objects, want 1", len(seal.Timers)) |
| 441 | } |
| 442 | vs, err := ir.versioned(ctx) |
| 443 | if err != nil { |
| 444 | return nil, err |
| 445 | } |
| 446 | raw, err := vs.GetVersion(ctx, cfg.OwnBucket, seal.Timers[0].Key, seal.Timers[0].VersionID) |
| 447 | if err != nil { |
| 448 | return nil, fmt.Errorf("read the timers snapshot: %w", err) |
| 449 | } |
| 450 | snap, err := moves.DecodeTimers(raw) |
| 451 | if err != nil { |
| 452 | return nil, err |
| 453 | } |
| 454 | if snap.Host != ir.rec.Host { |
| 455 | return nil, fmt.Errorf("the timers snapshot is %s's, not %s's", snap.Host, ir.rec.Host) |
| 456 | } |
| 457 | if got := moves.TimersDigest(snap.Rows); got != wantDigest { |
| 458 | return nil, fmt.Errorf("the timers snapshot digests to %s, the record says %s", got, wantDigest) |
| 459 | } |
| 460 | for _, r := range snap.Rows { |
| 461 | if r.Host != ir.rec.Host { |
| 462 | return nil, fmt.Errorf("timer %q is %s's, not %s's", r.Name, r.Host, ir.rec.Host) |
| 463 | } |
| 464 | } |
| 465 | return snap.Rows, nil |
| 466 | } |
| 467 | |
| 468 | // sameBundle compares two bundle etag maps. |
| 469 | func sameBundle(a, b map[string]string) bool { |
| 470 | if len(a) != len(b) { |
| 471 | return false |
| 472 | } |
| 473 | for k, v := range a { |
| 474 | if bv, ok := b[k]; !ok || bv != v { |
| 475 | return false |
| 476 | } |
| 477 | } |
| 478 | return true |
| 479 | } |
| 480 | |
| 481 | // restoreOutcome is what restore (or local's files install) ended with: the |
| 482 | // database installed and its markers written (or the files placed); the |
| 483 | // record changed instead (the rollback marker, fix round 1, item 2: a |
| 484 | // marker written is never read as an install); or the phase waiting for |
| 485 | // the next pass (the location's lease held elsewhere). |
| 486 | type restoreOutcome int |
| 487 | |
| 488 | const ( |
| 489 | restoreInstalled restoreOutcome = iota |
| 490 | restoreMarked |
| 491 | restoreWaiting |
| 492 | ) |
| 493 | |
| 494 | // restore is the install of the built database (§7.6 restored, R6-3): |
| 495 | // its sha256 must be the manifest's and CheckDB must pass, else the |
| 496 | // rollback marker; then, inside the host guard (final review, finding 2; |
| 497 | // hostbox ColdRestore.Install's pattern), the destination checked again, |
| 498 | // the rename into DBPath and the generation marker. The returned sha |
| 499 | // carries the record's "sha256:" prefix, as Source.DBSHA256 does. |
| 500 | // |
| 501 | // The build, the verification and the rollback marker run outside the |
| 502 | // guard (they read the network or write the record); the observation they |
| 503 | // start from is only a hint, and the guarded block decides: a database at |
| 504 | // DBPath that is not this build, a marker naming another generation, or a |
| 505 | // WAL beside an absent database is a successor's or someone else's, and |
| 506 | // is never replaced. |
| 507 | func (ir *installRun) restore(ctx context.Context, obs hostclaim.Observed) (sha string, outcome restoreOutcome, err error) { |
| 508 | cfg := ir.cfg() |
| 509 | host, id := ir.rec.Host, ir.src.GenerationOp |
| 510 | m, waiting, err := ir.installManifest(ctx, obs) |
| 511 | if err != nil || waiting { |
| 512 | return "", restoreWaiting, err |
| 513 | } |
| 514 | want := "sha256:" + m.DBSHA256 |
| 515 | dbPath := ir.dbPath() |
| 516 | captured, err := ir.capture(*obs.Claim.Holder) |
| 517 | if err != nil { |
| 518 | return "", restoreWaiting, err |
| 519 | } |
| 520 | present, err := regularFile(dbPath) |
| 521 | if err != nil { |
| 522 | return "", restoreWaiting, err |
| 523 | } |
| 524 | built := filepath.Join(ir.buildDir(), host+".db") |
| 525 | if !present { |
| 526 | if ok, err := regularFile(built); err != nil { |
| 527 | return "", restoreWaiting, err |
| 528 | } else if !ok { |
| 529 | // The scratch was lost (a rebuilt data directory): build again |
| 530 | // from the pinned set, which reproduces the same bytes, under |
| 531 | // the location's lease. |
| 532 | if _, waiting, err := ir.rebuild(ctx, obs); err != nil || waiting { |
| 533 | return "", restoreWaiting, err |
| 534 | } |
| 535 | } |
| 536 | if err := ir.fault("before-db-verify"); err != nil { |
| 537 | return "", restoreWaiting, err |
| 538 | } |
| 539 | got, err := fileSHA256(built) |
| 540 | if err != nil { |
| 541 | return "", restoreWaiting, err |
| 542 | } |
| 543 | if got != want { |
| 544 | log.Printf("moves: %s: the built database is %s, the manifest says %s", host, got, want) |
| 545 | return ir.refuseInstall(ctx) |
| 546 | } |
| 547 | if err := lease.CheckDB(built); err != nil { |
| 548 | log.Printf("moves: %s: the built database does not pass its check: %v", host, err) |
| 549 | return ir.refuseInstall(ctx) |
| 550 | } |
| 551 | for _, side := range []string{"-wal", "-shm"} { |
| 552 | os.Remove(built + side) |
| 553 | } |
| 554 | } |
| 555 | err = ir.guarded(ctx, captured, ir.src.Phases[4], func() error { |
| 556 | // Every destination is Lstat-checked, never read through: a symlink |
| 557 | // or any other non-regular file at <dbPath>.gen, DBPath or the WAL |
| 558 | // paths, dangling or not, is not this install's and is never |
| 559 | // replaced (final re-review residual). ReadGenerationMarker follows |
| 560 | // a symlink and reads a dangling one as absent. |
| 561 | if fi, err := os.Lstat(dbPath + ".gen"); err == nil && !fi.Mode().IsRegular() { |
| 562 | return fmt.Errorf("%w: %s.gen is not a regular file (%s)", errLandingChanged, dbPath, fi.Mode()) |
| 563 | } else if err != nil && !errors.Is(err, os.ErrNotExist) { |
| 564 | return err |
| 565 | } |
| 566 | gen, err := hostbox.ReadGenerationMarker(dbPath + ".gen") |
| 567 | if err != nil { |
| 568 | return fmt.Errorf("read %s.gen: %w", dbPath, err) |
| 569 | } |
| 570 | if gen != "" && gen != id { |
| 571 | return fmt.Errorf("%w: %s.gen names generation %s", errLandingChanged, dbPath, gen) |
| 572 | } |
| 573 | now, err := regularFile(dbPath) // Lstat: a non-regular DBPath is refused |
| 574 | if err != nil { |
| 575 | return fmt.Errorf("%w: %v", errLandingChanged, err) |
| 576 | } |
| 577 | switch { |
| 578 | case now: |
| 579 | // This landing's own install (a replay after the rename) is |
| 580 | // the build's bytes; anything else is not to be replaced. |
| 581 | got, err := fileSHA256(dbPath) |
| 582 | if err != nil { |
| 583 | return err |
| 584 | } |
| 585 | if got != want { |
| 586 | return fmt.Errorf("%w: %s holds a database that is not this install's build (%s, want %s)", errLandingChanged, dbPath, got, want) |
| 587 | } |
| 588 | case present: |
| 589 | return fmt.Errorf("%w: the database observed at %s is gone", errLandingChanged, dbPath) |
| 590 | default: |
| 591 | for _, side := range []string{"-wal", "-shm"} { |
| 592 | if _, err := os.Lstat(dbPath + side); err == nil { |
| 593 | return fmt.Errorf("%w: %s%s exists beside an absent database", errLandingChanged, dbPath, side) |
| 594 | } else if !errors.Is(err, os.ErrNotExist) { |
| 595 | return err |
| 596 | } |
| 597 | } |
| 598 | // No seed marker may stand beside the database this landing |
| 599 | // installs: it would name another seed's leftover for it |
| 600 | // (re-review finding 8). |
| 601 | if err := os.Remove(dbPath + activator.SeedMarkerSuffix); err != nil && !errors.Is(err, os.ErrNotExist) { |
| 602 | return err |
| 603 | } |
| 604 | if err := ir.notePreInstall(); err != nil { |
| 605 | return err |
| 606 | } |
| 607 | if err := os.Rename(built, dbPath); err != nil { |
| 608 | return fmt.Errorf("install the database: %w", err) |
| 609 | } |
| 610 | if err := syncPath(filepath.Dir(dbPath)); err != nil { |
| 611 | return err |
| 612 | } |
| 613 | } |
| 614 | if err := ir.fault("after-db-rename"); err != nil { |
| 615 | return err |
| 616 | } |
| 617 | if err := cfg.WriteGen(dbPath+".gen", id); err != nil { |
| 618 | return fmt.Errorf("write the generation marker: %w", err) |
| 619 | } |
| 620 | return syncPath(filepath.Dir(dbPath)) |
| 621 | }) |
| 622 | if err != nil { |
| 623 | return "", restoreWaiting, err |
| 624 | } |
| 625 | log.Printf("moves: %s: database installed at %s (%s), generation %s", host, dbPath, want, id) |
| 626 | return want, restoreInstalled, nil |
| 627 | } |
| 628 | |
| 629 | // preInstall is the landStepPreInstall record: which of the database's |
| 630 | // unchecked siblings (installSiblingPaths) were absent, relative to the |
| 631 | // data directory, when the install was about to put the database in place. |
| 632 | // Only a sibling recorded absent can later count as this op's (the target's |
| 633 | // cleaning, rollbackOwnership; the final re-review's residual): one already |
| 634 | // there was not made by anything of this install's. |
| 635 | type preInstall struct { |
| 636 | Absent []string `json:"absent"` |
| 637 | } |
| 638 | |
| 639 | // installSiblingPaths are the database's siblings the install does not |
| 640 | // refuse beside an absent database (it refuses a WAL and shared memory): |
| 641 | // the quarantine and the litestream shadow, relative to the data directory. |
| 642 | func installSiblingPaths(host string) []string { |
| 643 | db := host + ".db" |
| 644 | return []string{db + ".corrupt", ".activator-" + db + "-litestream"} |
| 645 | } |
| 646 | |
| 647 | // notePreInstall journals preInstall inside the database install's guard, |
| 648 | // right before the rename: each sibling Lstat'ed, never followed. A replay |
| 649 | // before the rename (a crash after the step) looks again and journals what |
| 650 | // it sees, so the record is always the look that immediately preceded the |
| 651 | // rename, under the same guard. A replay that finds the database already |
| 652 | // installed never reaches here, so a step lost that way is missing, and its |
| 653 | // siblings are never counted as the op's. |
| 654 | func (ir *installRun) notePreInstall() error { |
| 655 | j, err := ir.journal() |
| 656 | if err != nil { |
| 657 | return err |
| 658 | } |
| 659 | raw, replay := j.Steps[landStepPreInstall] |
| 660 | rec := preInstall{Absent: []string{}} |
| 661 | for _, p := range installSiblingPaths(ir.rec.Host) { |
| 662 | _, err := os.Lstat(filepath.Join(ir.cfg().DataDir, p)) |
| 663 | switch { |
| 664 | case errors.Is(err, os.ErrNotExist): |
| 665 | rec.Absent = append(rec.Absent, p) |
| 666 | case err != nil: |
| 667 | return err |
| 668 | default: |
| 669 | log.Printf("moves: %s: %s is already there before the database is installed; it is not the %s's", ir.rec.Host, p, ir.src.Name) |
| 670 | } |
| 671 | } |
| 672 | body, err := json.Marshal(rec) |
| 673 | if err != nil { |
| 674 | return err |
| 675 | } |
| 676 | if replay && string(body) == raw { |
| 677 | return nil |
| 678 | } |
| 679 | j.Mark(landStepPreInstall, string(body)) |
| 680 | if err := j.Commit(ir.cfg().DataDir); err != nil { |
| 681 | return err |
| 682 | } |
| 683 | return ir.fault("after-" + landStepPreInstall) |
| 684 | } |
| 685 | |
| 686 | // refuseInstall writes the install's refusal for a build that does not |
| 687 | // verify. A refusal that landed is restoreMarked; one refused (the record |
| 688 | // moved on) is restoreWaiting, and nothing is installed either way. |
| 689 | func (ir *installRun) refuseInstall(ctx context.Context) (string, restoreOutcome, error) { |
| 690 | ok, err := ir.refuse(ctx, refuseInstallMismatch) |
| 691 | if err != nil || !ok { |
| 692 | return "", restoreWaiting, err |
| 693 | } |
| 694 | return "", restoreMarked, nil |
| 695 | } |
| 696 | |
| 697 | // installManifest is the journal's, or the build's again when the journal |
| 698 | // lost it. waiting is true when the rebuild could not take the lease. |
| 699 | func (ir *installRun) installManifest(ctx context.Context, obs hostclaim.Observed) (moves.InstallManifest, bool, error) { |
| 700 | j, err := ir.journal() |
| 701 | if err != nil { |
| 702 | return moves.InstallManifest{}, false, err |
| 703 | } |
| 704 | var m moves.InstallManifest |
| 705 | if raw, ok := j.Steps[landStepInstall]; ok && json.Unmarshal([]byte(raw), &m) == nil && m.Op == ir.src.GenerationOp && m.Host == ir.rec.Host && m.DBSHA256 != "" { |
| 706 | return m, false, nil |
| 707 | } |
| 708 | return ir.rebuild(ctx, obs) |
| 709 | } |
| 710 | |
| 711 | // rebuild runs BuildGeneration again over the seal: a replay that reuses |
| 712 | // every object it finds and restores the same bytes into the scratch. It |
| 713 | // takes the location's lease first (published released it), runs every |
| 714 | // generation request under it, checks it again before recording the result |
| 715 | // and releases it after (fix round 1, item 3). waiting is true when another |
| 716 | // holder has the lease; nothing is written then. |
| 717 | func (ir *installRun) rebuild(ctx context.Context, obs hostclaim.Observed) (moves.InstallManifest, bool, error) { |
| 718 | cfg := ir.cfg() |
| 719 | seal, err := ir.seal(obs.Claim) |
| 720 | if err != nil { |
| 721 | return moves.InstallManifest{}, false, err |
| 722 | } |
| 723 | vs, err := ir.versioned(ctx) |
| 724 | if err != nil { |
| 725 | return moves.InstallManifest{}, false, err |
| 726 | } |
| 727 | holder := *obs.Claim.Holder |
| 728 | held, waiting, err := ir.lease(ctx, holder) |
| 729 | if waiting || err != nil { |
| 730 | return moves.InstallManifest{}, waiting, err |
| 731 | } |
| 732 | defer ir.releaseLease() |
| 733 | m, err := moves.BuildGeneration(leaseContext(ctx, held), moves.BuildInputs{Store: leasedStore{VersionedStore: vs, check: held.Check}, Bucket: cfg.OwnBucket, Source: seal.Replica, N: seal.N, |
| 734 | GenerationPrefix: moves.GenerationPrefix(holder.Location, ir.src.GenerationOp), Op: ir.src.GenerationOp, Host: ir.rec.Host, WorkDir: ir.buildDir(), Now: cfg.Now}) |
| 735 | if err != nil { |
| 736 | return moves.InstallManifest{}, false, fmt.Errorf("rebuild the generation: %w", err) |
| 737 | } |
| 738 | if err := held.Check(ctx); err != nil { |
| 739 | return moves.InstallManifest{}, false, fmt.Errorf("the location's lease after the rebuild: %w", err) |
| 740 | } |
| 741 | j, err := ir.journal() |
| 742 | if err != nil { |
| 743 | return m, false, err |
| 744 | } |
| 745 | body, err := json.Marshal(m) |
| 746 | if err != nil { |
| 747 | return m, false, err |
| 748 | } |
| 749 | j.Mark(landStepInstall, string(body)) |
| 750 | return m, false, j.Commit(cfg.DataDir) |
| 751 | } |
| 752 | |
| 753 | // writeFileAtomic writes body to path through a temporary file in the same |
| 754 | // directory, synced and renamed, then syncs the directory. |
| 755 | func writeFileAtomic(path string, body []byte, mode os.FileMode) error { |
| 756 | dir := filepath.Dir(path) |
| 757 | f, err := os.CreateTemp(dir, "."+filepath.Base(path)+".*") |
| 758 | if err != nil { |
| 759 | return err |
| 760 | } |
| 761 | tmp := f.Name() |
| 762 | if _, err := f.Write(body); err != nil { |
| 763 | f.Close() |
| 764 | os.Remove(tmp) |
| 765 | return err |
| 766 | } |
| 767 | if err := f.Chmod(mode); err != nil { |
| 768 | f.Close() |
| 769 | os.Remove(tmp) |
| 770 | return err |
| 771 | } |
| 772 | if err := f.Sync(); err != nil { |
| 773 | f.Close() |
| 774 | os.Remove(tmp) |
| 775 | return err |
| 776 | } |
| 777 | if err := f.Close(); err != nil { |
| 778 | os.Remove(tmp) |
| 779 | return err |
| 780 | } |
| 781 | if err := os.Rename(tmp, path); err != nil { |
| 782 | os.Remove(tmp) |
| 783 | return err |
| 784 | } |
| 785 | return syncPath(dir) |
| 786 | } |
| 787 | |
| 788 | // syncPath fsyncs a file or directory. |
| 789 | func syncPath(path string) error { |
| 790 | f, err := os.Open(path) |
| 791 | if err != nil { |
| 792 | return err |
| 793 | } |
| 794 | defer f.Close() |
| 795 | return f.Sync() |
| 796 | } |
| 797 | |
amadan