| 1 | package repair |
| 2 | |
| 3 | import ( |
| 4 | "crypto/sha256" |
| 5 | "encoding/hex" |
| 6 | "encoding/json" |
| 7 | "fmt" |
| 8 | "os" |
| 9 | "path/filepath" |
| 10 | "reflect" |
| 11 | "strings" |
| 12 | "time" |
| 13 | |
| 14 | "reasonix/internal/config" |
| 15 | "reasonix/internal/fileutil" |
| 16 | ) |
| 17 | |
| 18 | type RepairChange struct { |
| 19 | Scope string `json:"scope,omitempty"` |
| 20 | TargetPath string `json:"targetPath"` |
| 21 | PreviousPath string `json:"previousPath,omitempty"` |
| 22 | PreviousStateID string `json:"previousStateId,omitempty"` |
| 23 | RemoveOnUndo bool `json:"removeOnUndo,omitempty"` |
| 24 | // CreatedStateID binds a remove-on-undo entry to the exact node this repair |
| 25 | // intended to create. If type, mode, or bytes later differ, undo preserves |
| 26 | // it as an unowned concurrent write. |
| 27 | CreatedStateID string `json:"createdStateId,omitempty"` |
| 28 | // Prepared is a write-ahead filesystem-mutation intent. The exact previous |
| 29 | // state may still be at TargetPath (rename not run) or at PreviousPath |
| 30 | // (rename committed). Undo resolves either state without guessing. |
| 31 | Prepared bool `json:"prepared,omitempty"` |
| 32 | // Undone marks a change already reverted by an interrupted undo, so a |
| 33 | // retry can resume with the remaining changes instead of failing the |
| 34 | // preflight on the consumed backup of a change that is already restored. |
| 35 | Undone bool `json:"undone,omitempty"` |
| 36 | } |
| 37 | |
| 38 | type RepairTransaction struct { |
| 39 | SchemaVersion int `json:"schemaVersion"` |
| 40 | ID string `json:"id"` |
| 41 | CreatedAt string `json:"createdAt"` |
| 42 | Changes []RepairChange `json:"changes"` |
| 43 | PreparedLastRepairStateID string `json:"preparedLastRepairStateId,omitempty"` |
| 44 | Undone bool `json:"undone,omitempty"` |
| 45 | UndoneAt string `json:"undoneAt,omitempty"` |
| 46 | } |
| 47 | |
| 48 | func newRepairTransaction(now time.Time) *RepairTransaction { |
| 49 | now = now.UTC() |
| 50 | return &RepairTransaction{ |
| 51 | SchemaVersion: 1, |
| 52 | ID: fmt.Sprintf("repair-%d", now.UnixNano()), |
| 53 | CreatedAt: now.Format(time.RFC3339Nano), |
| 54 | Changes: []RepairChange{}, |
| 55 | } |
| 56 | } |
| 57 | |
| 58 | func repairChangeForPrevious(scope, target, previous string) RepairChange { |
| 59 | return RepairChange{ |
| 60 | Scope: scope, |
| 61 | TargetPath: target, |
| 62 | PreviousPath: previous, |
| 63 | PreviousStateID: repairPlanReleaseNodeStateFor(previous, target), |
| 64 | } |
| 65 | } |
| 66 | |
| 67 | func preparedRepairChangeForPrevious(scope, target, previous string) RepairChange { |
| 68 | return RepairChange{ |
| 69 | Scope: scope, |
| 70 | TargetPath: target, |
| 71 | PreviousPath: previous, |
| 72 | PreviousStateID: repairPlanReleaseNodeStateFor(target, target), |
| 73 | Prepared: true, |
| 74 | } |
| 75 | } |
| 76 | |
| 77 | func preparedRepairChangeForCreate(scope, target, createdStateID string) RepairChange { |
| 78 | return RepairChange{ |
| 79 | Scope: scope, |
| 80 | TargetPath: target, |
| 81 | RemoveOnUndo: true, |
| 82 | CreatedStateID: createdStateID, |
| 83 | Prepared: true, |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | func repairTransactionPath() string { |
| 88 | if root := config.MemoryUserDir(); root != "" { |
| 89 | return filepath.Join(root, "repair", "last-repair.json") |
| 90 | } |
| 91 | return "" |
| 92 | } |
| 93 | |
| 94 | func pendingRepairTransactionPath() string { |
| 95 | if root := config.MemoryUserDir(); root != "" { |
| 96 | return filepath.Join(root, "repair", "pending-repair.json") |
| 97 | } |
| 98 | return "" |
| 99 | } |
| 100 | |
| 101 | func repairLogPath() string { |
| 102 | if root := config.MemoryUserDir(); root != "" { |
| 103 | return filepath.Join(root, "repair", "repair-log.jsonl") |
| 104 | } |
| 105 | return "" |
| 106 | } |
| 107 | |
| 108 | func saveRepairTransaction(tx *RepairTransaction) error { |
| 109 | if tx == nil || len(tx.Changes) == 0 { |
| 110 | return nil |
| 111 | } |
| 112 | if err := persistRepairTransaction(tx); err != nil { |
| 113 | return err |
| 114 | } |
| 115 | appendRepairLogBestEffort(tx) |
| 116 | return nil |
| 117 | } |
| 118 | |
| 119 | // The append-only audit log is best-effort. last-repair.json is the durable |
| 120 | // undo state, so an audit failure must not turn an already committed filesystem |
| 121 | // change into a reported failure or trigger cleanup that consumes its backup. |
| 122 | func appendRepairLogBestEffort(tx *RepairTransaction) { |
| 123 | _ = appendRepairLog(tx) |
| 124 | } |
| 125 | |
| 126 | func persistRepairTransaction(tx *RepairTransaction) error { |
| 127 | if tx != nil { |
| 128 | if strings.TrimSpace(tx.PreparedLastRepairStateID) != "" { |
| 129 | return fmt.Errorf("last repair transaction contains pending-only state") |
| 130 | } |
| 131 | for _, change := range tx.Changes { |
| 132 | if change.Prepared { |
| 133 | return fmt.Errorf("last repair transaction contains a prepared change") |
| 134 | } |
| 135 | } |
| 136 | } |
| 137 | path := repairTransactionPath() |
| 138 | if path == "" { |
| 139 | return nil |
| 140 | } |
| 141 | b, err := json.MarshalIndent(tx, "", " ") |
| 142 | if err != nil { |
| 143 | return err |
| 144 | } |
| 145 | if err := fileutil.AtomicWriteFile(path, append(b, '\n'), 0o600); err != nil { |
| 146 | return err |
| 147 | } |
| 148 | repairTransactionAfterPersist(tx) |
| 149 | return nil |
| 150 | } |
| 151 | |
| 152 | var repairTransactionAfterPersist = func(*RepairTransaction) {} |
| 153 | var repairPendingAfterMove = func(string, string) {} |
| 154 | |
| 155 | func persistPreparedRepairTransaction(tx *RepairTransaction) error { |
| 156 | if tx == nil || len(tx.Changes) == 0 || !tx.Changes[len(tx.Changes)-1].Prepared { |
| 157 | return fmt.Errorf("pending repair transaction is incomplete") |
| 158 | } |
| 159 | path := pendingRepairTransactionPath() |
| 160 | if path == "" { |
| 161 | return fmt.Errorf("pending repair state path is unavailable") |
| 162 | } |
| 163 | if _, err := os.Lstat(repairTransactionPath()); err == nil { |
| 164 | if _, err := ReadLastRepair(); err != nil { |
| 165 | return fmt.Errorf("prepare repair: current undo state is invalid: %w", err) |
| 166 | } |
| 167 | } else if !os.IsNotExist(err) { |
| 168 | return fmt.Errorf("prepare repair: inspect current undo state: %w", err) |
| 169 | } |
| 170 | // Bind the journal to the undo state it is allowed to replace. A stale |
| 171 | // pending file must never promote itself over a newer completed repair. |
| 172 | tx.PreparedLastRepairStateID = repairPlanReleaseNodeState(repairTransactionPath()) |
| 173 | b, err := json.MarshalIndent(tx, "", " ") |
| 174 | if err != nil { |
| 175 | return err |
| 176 | } |
| 177 | if err := fileutil.AtomicCreateFile(path, append(b, '\n'), 0o600); err != nil { |
| 178 | return fmt.Errorf("publish pending repair transaction: %w", err) |
| 179 | } |
| 180 | return nil |
| 181 | } |
| 182 | |
| 183 | // verifyPreparedCreateOwnership proves that AtomicCreateFile published the |
| 184 | // state recorded before the mutation. It must never rebind to whatever happens |
| 185 | // to be at path afterward: an uncooperative writer could replace the file in |
| 186 | // that window and would then be incorrectly claimed and removed by undo. |
| 187 | func verifyPreparedCreateOwnership(tx *RepairTransaction, changeIndex int, path string) error { |
| 188 | if tx == nil || changeIndex < 0 || changeIndex >= len(tx.Changes) { |
| 189 | return fmt.Errorf("prepared create ownership is incomplete") |
| 190 | } |
| 191 | change := tx.Changes[changeIndex] |
| 192 | if !change.Prepared || !change.RemoveOnUndo { |
| 193 | return fmt.Errorf("prepared create ownership is incomplete") |
| 194 | } |
| 195 | expected := strings.TrimSpace(change.CreatedStateID) |
| 196 | if !validRepairStateID(expected) { |
| 197 | return fmt.Errorf("prepared create ownership is incomplete") |
| 198 | } |
| 199 | if err := verifyRepairPlanReleaseNodeStateFor(path, path, expected); err != nil { |
| 200 | return fmt.Errorf("published create ownership changed: %w", err) |
| 201 | } |
| 202 | return nil |
| 203 | } |
| 204 | |
| 205 | func clearPreparedRepairTransaction(expected *RepairTransaction) error { |
| 206 | if expected == nil { |
| 207 | return fmt.Errorf("pending repair transaction identity is incomplete") |
| 208 | } |
| 209 | path := pendingRepairTransactionPath() |
| 210 | if path == "" { |
| 211 | return nil |
| 212 | } |
| 213 | cleanup, err := moveRepairNodeToUniqueCleanup(path) |
| 214 | if err != nil || cleanup == "" { |
| 215 | return err |
| 216 | } |
| 217 | repairPendingAfterMove(path, cleanup) |
| 218 | restoreUnknown := func(cause error) error { |
| 219 | if restoreErr := renameRepairNodeNoReplace(cleanup, path); restoreErr != nil { |
| 220 | return fmt.Errorf("%w; displaced pending repair retained at %s: %v", cause, cleanup, restoreErr) |
| 221 | } |
| 222 | return cause |
| 223 | } |
| 224 | b, err := os.ReadFile(cleanup) |
| 225 | if err != nil { |
| 226 | return restoreUnknown(fmt.Errorf("read displaced pending repair: %w", err)) |
| 227 | } |
| 228 | var actual RepairTransaction |
| 229 | if err := json.Unmarshal(b, &actual); err != nil { |
| 230 | return restoreUnknown(fmt.Errorf("pending repair transaction changed before cleanup: %w", err)) |
| 231 | } |
| 232 | if !reflect.DeepEqual(&actual, expected) { |
| 233 | return restoreUnknown(fmt.Errorf("pending repair transaction changed before cleanup")) |
| 234 | } |
| 235 | // Do not unlink the moved journal. A path-based remove has an unavoidable |
| 236 | // check/remove race: an uncooperative writer could replace cleanup after the |
| 237 | // equality check and have its node deleted. The uniquely named, inactive |
| 238 | // journal is small and doubles as crash-recovery evidence. |
| 239 | if _, err := os.Lstat(path); err == nil { |
| 240 | return fmt.Errorf("a new pending repair transaction appeared during cleanup") |
| 241 | } else if !os.IsNotExist(err) { |
| 242 | return err |
| 243 | } |
| 244 | return nil |
| 245 | } |
| 246 | |
| 247 | // reconcilePreparedRepairTransaction runs while the repair transaction lock is |
| 248 | // held. It promotes a crashed post-rename intent into last-repair.json, or |
| 249 | // discards an intent whose exact source is still live and therefore never |
| 250 | // committed. Ambiguous state remains fail closed for manual recovery. |
| 251 | func reconcilePreparedRepairTransaction() error { |
| 252 | path := pendingRepairTransactionPath() |
| 253 | if path == "" { |
| 254 | return nil |
| 255 | } |
| 256 | b, err := os.ReadFile(path) |
| 257 | if err != nil { |
| 258 | if os.IsNotExist(err) { |
| 259 | return nil |
| 260 | } |
| 261 | return err |
| 262 | } |
| 263 | var tx RepairTransaction |
| 264 | if err := json.Unmarshal(b, &tx); err != nil { |
| 265 | return fmt.Errorf("pending repair transaction is invalid: %w", err) |
| 266 | } |
| 267 | if tx.SchemaVersion != 1 || tx.ID == "" || len(tx.Changes) == 0 || |
| 268 | !validRepairStateID(tx.PreparedLastRepairStateID) { |
| 269 | return fmt.Errorf("pending repair transaction is incomplete") |
| 270 | } |
| 271 | for i, change := range tx.Changes { |
| 272 | if err := validateRepairChange(change); err != nil { |
| 273 | return err |
| 274 | } |
| 275 | if change.Prepared != (i == len(tx.Changes)-1) { |
| 276 | return fmt.Errorf("pending repair transaction has an invalid prepared prefix") |
| 277 | } |
| 278 | } |
| 279 | last := tx.Changes[len(tx.Changes)-1] |
| 280 | unlockTargets, err := lockRepairMutations(last.TargetPath, last.PreviousPath) |
| 281 | if err != nil { |
| 282 | return err |
| 283 | } |
| 284 | defer unlockTargets() |
| 285 | |
| 286 | // The target lock may have blocked behind another cooperative process. |
| 287 | // Re-read the journal so that process cannot substitute a different intent |
| 288 | // between validation and promotion. |
| 289 | currentPendingBytes, err := os.ReadFile(path) |
| 290 | if err != nil { |
| 291 | return err |
| 292 | } |
| 293 | var currentPending RepairTransaction |
| 294 | if err := json.Unmarshal(currentPendingBytes, ¤tPending); err != nil { |
| 295 | return fmt.Errorf("pending repair transaction changed while waiting: %w", err) |
| 296 | } |
| 297 | if !reflect.DeepEqual(¤tPending, &tx) { |
| 298 | return fmt.Errorf("pending repair transaction changed while waiting") |
| 299 | } |
| 300 | |
| 301 | committed := tx |
| 302 | committed.Changes = append([]RepairChange(nil), tx.Changes...) |
| 303 | committed.Changes[len(committed.Changes)-1].Prepared = false |
| 304 | committed.PreparedLastRepairStateID = "" |
| 305 | currentLast, readLastErr := ReadLastRepair() |
| 306 | if readLastErr == nil && reflect.DeepEqual(currentLast, &committed) { |
| 307 | // last-repair.json was already published and only journal cleanup was |
| 308 | // interrupted. Its backup must not be compensated or reclassified. |
| 309 | return clearPreparedRepairTransaction(&tx) |
| 310 | } |
| 311 | if readLastErr != nil && !os.IsNotExist(readLastErr) { |
| 312 | return fmt.Errorf("pending repair transaction cannot replace invalid undo state: %w", readLastErr) |
| 313 | } |
| 314 | if actual := repairPlanReleaseNodeState(repairTransactionPath()); actual != tx.PreparedLastRepairStateID { |
| 315 | return fmt.Errorf("pending repair transaction no longer matches the previous undo state") |
| 316 | } |
| 317 | applied, err := preparedRepairChangeApplied(last) |
| 318 | if err != nil { |
| 319 | return err |
| 320 | } |
| 321 | if applied { |
| 322 | if err := persistRepairTransaction(&committed); err != nil { |
| 323 | return err |
| 324 | } |
| 325 | } |
| 326 | return clearPreparedRepairTransaction(&tx) |
| 327 | } |
| 328 | |
| 329 | // commitPreparedRepairTransaction reports whether last-repair.json became |
| 330 | // durable separately from journal-cleanup errors. Once durable is true callers |
| 331 | // must retain the backup referenced by the undo record and never compensate the |
| 332 | // already committed rename. |
| 333 | func commitPreparedRepairTransaction(tx *RepairTransaction, changeIndex int) (durable bool, err error) { |
| 334 | if tx == nil || changeIndex < 0 || changeIndex >= len(tx.Changes) || !tx.Changes[changeIndex].Prepared { |
| 335 | return false, fmt.Errorf("prepared repair transaction is incomplete") |
| 336 | } |
| 337 | pending := *tx |
| 338 | pending.Changes = append([]RepairChange(nil), tx.Changes...) |
| 339 | tx.Changes[changeIndex].Prepared = false |
| 340 | tx.PreparedLastRepairStateID = "" |
| 341 | if err := persistRepairTransaction(tx); err != nil { |
| 342 | tx.Changes[changeIndex].Prepared = true |
| 343 | tx.PreparedLastRepairStateID = pending.PreparedLastRepairStateID |
| 344 | return false, err |
| 345 | } |
| 346 | if err := clearPreparedRepairTransaction(&pending); err != nil { |
| 347 | return true, err |
| 348 | } |
| 349 | return true, nil |
| 350 | } |
| 351 | |
| 352 | func validRepairStateID(value string) bool { |
| 353 | value = strings.TrimSpace(value) |
| 354 | if len(value) != sha256.Size*2 || value != strings.ToLower(value) { |
| 355 | return false |
| 356 | } |
| 357 | _, err := hex.DecodeString(value) |
| 358 | return err == nil |
| 359 | } |
| 360 | |
| 361 | func appendRepairLog(tx *RepairTransaction) error { |
| 362 | path := repairLogPath() |
| 363 | if path == "" { |
| 364 | return nil |
| 365 | } |
| 366 | if err := os.MkdirAll(filepath.Dir(path), 0o700); err != nil { |
| 367 | return err |
| 368 | } |
| 369 | b, err := json.Marshal(tx) |
| 370 | if err != nil { |
| 371 | return err |
| 372 | } |
| 373 | f, err := os.OpenFile(path, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0o600) |
| 374 | if err != nil { |
| 375 | return err |
| 376 | } |
| 377 | defer f.Close() |
| 378 | _, err = f.Write(append(b, '\n')) |
| 379 | return err |
| 380 | } |
| 381 | |
| 382 | func ReadLastRepair() (*RepairTransaction, error) { |
| 383 | path := repairTransactionPath() |
| 384 | if path == "" { |
| 385 | return nil, os.ErrNotExist |
| 386 | } |
| 387 | b, err := os.ReadFile(path) |
| 388 | if err != nil { |
| 389 | return nil, err |
| 390 | } |
| 391 | var tx RepairTransaction |
| 392 | if err := json.Unmarshal(b, &tx); err != nil { |
| 393 | return nil, err |
| 394 | } |
| 395 | if tx.SchemaVersion != 1 || tx.ID == "" || len(tx.Changes) == 0 { |
| 396 | return nil, fmt.Errorf("last repair transaction is incomplete") |
| 397 | } |
| 398 | if strings.TrimSpace(tx.PreparedLastRepairStateID) != "" { |
| 399 | return nil, fmt.Errorf("last repair transaction contains pending-only state") |
| 400 | } |
| 401 | for _, change := range tx.Changes { |
| 402 | if change.Prepared { |
| 403 | return nil, fmt.Errorf("last repair transaction contains a prepared change") |
| 404 | } |
| 405 | if err := validateRepairChange(change); err != nil { |
| 406 | return nil, err |
| 407 | } |
| 408 | } |
| 409 | return &tx, nil |
| 410 | } |
| 411 | |
| 412 | func validateRepairChange(change RepairChange) error { |
| 413 | target := filepath.Clean(change.TargetPath) |
| 414 | switch { |
| 415 | case change.Scope == "global": |
| 416 | if target != filepath.Clean(config.UserConfigPath()) { |
| 417 | return fmt.Errorf("repair transaction global target is invalid") |
| 418 | } |
| 419 | case change.Scope == "project": |
| 420 | if filepath.Base(target) != "reasonix.toml" { |
| 421 | return fmt.Errorf("repair transaction project target is invalid") |
| 422 | } |
| 423 | case strings.HasPrefix(change.Scope, "derived:"): |
| 424 | name := change.Scope[len("derived:"):] |
| 425 | want, ok := derivedStatePaths()[name] |
| 426 | if !ok || target != filepath.Clean(want) { |
| 427 | return fmt.Errorf("repair transaction derived-state target is invalid") |
| 428 | } |
| 429 | default: |
| 430 | return fmt.Errorf("repair transaction scope is invalid") |
| 431 | } |
| 432 | if change.RemoveOnUndo { |
| 433 | if change.Scope != "global" || change.PreviousPath != "" { |
| 434 | return fmt.Errorf("repair transaction remove-on-undo action is invalid") |
| 435 | } |
| 436 | value := strings.TrimSpace(change.CreatedStateID) |
| 437 | if change.Prepared && value == "" { |
| 438 | return fmt.Errorf("repair transaction prepared create state is invalid") |
| 439 | } |
| 440 | if value != "" && !validRepairStateID(value) { |
| 441 | return fmt.Errorf("repair transaction created state is invalid") |
| 442 | } |
| 443 | return nil |
| 444 | } |
| 445 | if change.Prepared && strings.TrimSpace(change.PreviousStateID) == "" { |
| 446 | return fmt.Errorf("repair transaction prepared state is invalid") |
| 447 | } |
| 448 | previous := filepath.Clean(change.PreviousPath) |
| 449 | if filepath.Dir(previous) == filepath.Dir(target) && strings.HasPrefix(filepath.Base(previous), filepath.Base(target)+".reasonix-") { |
| 450 | return nil |
| 451 | } |
| 452 | if config.MemoryUserDir() == "" { |
| 453 | return fmt.Errorf("repair transaction state directory is unavailable") |
| 454 | } |
| 455 | restoreRoot := filepath.Join(config.MemoryUserDir(), "repair", "restore-backups") |
| 456 | if repairNodeInsideResolvedRoot(restoreRoot, previous) { |
| 457 | return nil |
| 458 | } |
| 459 | return fmt.Errorf("repair transaction previous path is invalid") |
| 460 | } |
| 461 | |
| 462 | // repairNodeInsideResolvedRoot follows directory symlinks but deliberately |
| 463 | // leaves the leaf unresolved. A backed-up config may itself be a symlink whose |
| 464 | // referent is outside the repair directory; cleanup owns that link node, not |
| 465 | // its target. A symlinked parent, however, would let a forged transaction move |
| 466 | // and delete an unrelated node outside the repair root. |
| 467 | func repairNodeInsideResolvedRoot(root, path string) bool { |
| 468 | root = filepath.Clean(strings.TrimSpace(root)) |
| 469 | path = filepath.Clean(strings.TrimSpace(path)) |
| 470 | if root == "" || path == "" || root == "." || path == "." { |
| 471 | return false |
| 472 | } |
| 473 | rootInfo, err := os.Lstat(root) |
| 474 | if err != nil || !rootInfo.IsDir() { |
| 475 | return false |
| 476 | } |
| 477 | resolvedRoot, err := filepath.EvalSymlinks(root) |
| 478 | if err != nil { |
| 479 | return false |
| 480 | } |
| 481 | resolvedParent, err := filepath.EvalSymlinks(filepath.Dir(path)) |
| 482 | if err != nil { |
| 483 | return false |
| 484 | } |
| 485 | resolvedPath := filepath.Join(resolvedParent, filepath.Base(path)) |
| 486 | rel, err := filepath.Rel(resolvedRoot, resolvedPath) |
| 487 | return err == nil && rel != "." && rel != ".." && |
| 488 | !strings.HasPrefix(rel, ".."+string(filepath.Separator)) |
| 489 | } |
| 490 | |
| 491 | var ( |
| 492 | readRepairPreviousFile = os.ReadFile |
| 493 | readRepairPreviousLink = os.Readlink |
| 494 | ) |
| 495 | |
| 496 | // UndoLastRepair restores the exact files moved aside by the latest repair. Any |
| 497 | // currently repaired file is retained as a timestamped redo candidate. |
| 498 | func UndoLastRepair() (*RepairTransaction, error) { |
| 499 | invocationLastState := repairPlanReleaseNodeState(repairTransactionPath()) |
| 500 | invocationPendingState := repairPlanReleaseNodeState(pendingRepairTransactionPath()) |
| 501 | unlockTransaction, err := lockRepairTransaction() |
| 502 | if err != nil { |
| 503 | return nil, err |
| 504 | } |
| 505 | defer unlockTransaction() |
| 506 | if repairPlanReleaseNodeState(repairTransactionPath()) != invocationLastState || |
| 507 | repairPlanReleaseNodeState(pendingRepairTransactionPath()) != invocationPendingState { |
| 508 | return nil, fmt.Errorf("undo repair: repair transaction changed while waiting") |
| 509 | } |
| 510 | if err := reconcilePreparedRepairTransaction(); err != nil { |
| 511 | return nil, fmt.Errorf("undo repair: reconcile pending mutation: %w", err) |
| 512 | } |
| 513 | tx, err := ReadLastRepair() |
| 514 | if err != nil { |
| 515 | return nil, err |
| 516 | } |
| 517 | if tx.Undone { |
| 518 | return nil, fmt.Errorf("repair %s was already undone", tx.ID) |
| 519 | } |
| 520 | if err := verifyUndoRepairBackups(tx); err != nil { |
| 521 | return nil, err |
| 522 | } |
| 523 | invocationID := repairPlanStateID(tx) |
| 524 | targets := make([]string, 0, len(tx.Changes)) |
| 525 | for _, change := range tx.Changes { |
| 526 | targets = append(targets, change.TargetPath) |
| 527 | } |
| 528 | unlockTargets, err := lockRepairMutations(targets...) |
| 529 | if err != nil { |
| 530 | return nil, err |
| 531 | } |
| 532 | defer unlockTargets() |
| 533 | current, err := ReadLastRepair() |
| 534 | if err != nil { |
| 535 | return nil, fmt.Errorf("undo repair: re-read last repair transaction: %w", err) |
| 536 | } |
| 537 | if repairPlanStateID(current) != invocationID { |
| 538 | return nil, fmt.Errorf("undo repair: last repair transaction changed while waiting") |
| 539 | } |
| 540 | tx = current |
| 541 | if err := verifyUndoRepairBackups(tx); err != nil { |
| 542 | return nil, err |
| 543 | } |
| 544 | now := time.Now().UTC() |
| 545 | // markUndone persists per-change progress so a failure partway through a |
| 546 | // multi-change undo leaves a transaction the next undo can resume. |
| 547 | markUndone := func(i int) error { |
| 548 | tx.Changes[i].Undone = true |
| 549 | return persistRepairTransaction(tx) |
| 550 | } |
| 551 | for i := len(tx.Changes) - 1; i >= 0; i-- { |
| 552 | change := tx.Changes[i] |
| 553 | if change.Undone { |
| 554 | // Progress was persisted but the backup removal may have been cut |
| 555 | // short by a crash; finish the cleanup the completed step owed. |
| 556 | if !change.RemoveOnUndo && change.PreviousPath != "" { |
| 557 | _ = removeRepairNodeIfMatching(change.PreviousPath, change.TargetPath, change.PreviousStateID) |
| 558 | } |
| 559 | continue |
| 560 | } |
| 561 | if change.Prepared { |
| 562 | applied, resolveErr := preparedRepairChangeApplied(change) |
| 563 | if resolveErr != nil { |
| 564 | return nil, fmt.Errorf("undo repair: resolve prepared change: %w", resolveErr) |
| 565 | } |
| 566 | if !applied { |
| 567 | if err := markUndone(i); err != nil { |
| 568 | return nil, err |
| 569 | } |
| 570 | continue |
| 571 | } |
| 572 | } |
| 573 | previousStateID := strings.TrimSpace(change.PreviousStateID) |
| 574 | redo := "" |
| 575 | if change.RemoveOnUndo && strings.TrimSpace(change.CreatedStateID) != "" { |
| 576 | info, statErr := os.Lstat(change.TargetPath) |
| 577 | if os.IsNotExist(statErr) { |
| 578 | if err := markUndone(i); err != nil { |
| 579 | return nil, err |
| 580 | } |
| 581 | continue |
| 582 | } |
| 583 | if statErr != nil { |
| 584 | return nil, fmt.Errorf("undo repair: inspect created file: %w", statErr) |
| 585 | } |
| 586 | owned := info.Mode().IsRegular() && |
| 587 | verifyRepairPlanReleaseNodeStateFor( |
| 588 | change.TargetPath, |
| 589 | change.TargetPath, |
| 590 | change.CreatedStateID, |
| 591 | ) == nil |
| 592 | if !owned { |
| 593 | // A failed/crashed create intent may be followed by an |
| 594 | // uncooperative writer. It is not this repair's file, so leave |
| 595 | // it live while allowing earlier plan changes to be undone. |
| 596 | if err := markUndone(i); err != nil { |
| 597 | return nil, err |
| 598 | } |
| 599 | continue |
| 600 | } |
| 601 | } |
| 602 | // Lstat so a dangling symlink at the target is still moved aside |
| 603 | // instead of being clobbered by the restore below. |
| 604 | if _, err := os.Lstat(change.TargetPath); err == nil { |
| 605 | // Index suffix keeps redo names unique when one undo touches the |
| 606 | // same target twice (e.g. quarantine + snapshot restore): a shared |
| 607 | // name would silently overwrite the earlier redo copy. |
| 608 | redo = fmt.Sprintf("%s.reasonix-redo-%s-%d", change.TargetPath, now.Format("20060102T150405.000000000Z"), i) |
| 609 | if err := renameRepairNodeNoReplace(change.TargetPath, redo); err != nil { |
| 610 | return nil, fmt.Errorf("undo repair: retain current file: %w", err) |
| 611 | } |
| 612 | repairMutationAfterRename(change.TargetPath) |
| 613 | } |
| 614 | if change.RemoveOnUndo { |
| 615 | if err := markUndone(i); err != nil { |
| 616 | return nil, err |
| 617 | } |
| 618 | continue |
| 619 | } |
| 620 | // Restore by copy and keep the backup until the progress record is on |
| 621 | // disk: consuming the backup first (rename or delete) would leave an |
| 622 | // unresumable transaction if the process died before markUndone, since |
| 623 | // the retry's preflight requires the backup of every un-undone change. |
| 624 | restoreErr := func() error { |
| 625 | if expected := previousStateID; expected != "" { |
| 626 | if err := verifyRepairPlanReleaseNodeStateFor(change.PreviousPath, change.TargetPath, expected); err != nil { |
| 627 | return fmt.Errorf("previous state changed: %w", err) |
| 628 | } |
| 629 | } |
| 630 | info, err := os.Lstat(change.PreviousPath) |
| 631 | if err != nil { |
| 632 | return err |
| 633 | } |
| 634 | if info.Mode()&os.ModeSymlink != 0 { |
| 635 | // A quarantined symlink (e.g. a dotfiles-managed config.toml) |
| 636 | // must come back as a symlink: ReadFile would follow it and |
| 637 | // materialize a regular file, permanently severing the link. |
| 638 | // Recreating a symlink is a single atomic syscall, so the |
| 639 | // crash-safety of the copy path is preserved. |
| 640 | linkTarget, err := readRepairPreviousLink(change.PreviousPath) |
| 641 | if err != nil { |
| 642 | return err |
| 643 | } |
| 644 | linkContent, linkContentErr := readRepairPreviousFile(change.PreviousPath) |
| 645 | exactStateID := repairPlanReadStateIDFor( |
| 646 | change.TargetPath, |
| 647 | info.Mode(), |
| 648 | "symlink", |
| 649 | linkTarget, |
| 650 | linkContent, |
| 651 | linkContentErr == nil, |
| 652 | ) |
| 653 | if previousStateID != "" && exactStateID != previousStateID { |
| 654 | return fmt.Errorf("previous link read changed since it was verified") |
| 655 | } |
| 656 | if expected := previousStateID; expected != "" { |
| 657 | if err := verifyRepairPlanReleaseNodeStateFor(change.PreviousPath, change.TargetPath, expected); err != nil { |
| 658 | return fmt.Errorf("previous state changed while reading link: %w", err) |
| 659 | } |
| 660 | } |
| 661 | return os.Symlink(linkTarget, change.TargetPath) |
| 662 | } |
| 663 | b, err := readRepairPreviousFile(change.PreviousPath) |
| 664 | if err != nil { |
| 665 | return err |
| 666 | } |
| 667 | exactStateID := repairPlanReadStateIDFor( |
| 668 | change.TargetPath, |
| 669 | info.Mode(), |
| 670 | "file", |
| 671 | "", |
| 672 | b, |
| 673 | true, |
| 674 | ) |
| 675 | if previousStateID != "" && exactStateID != previousStateID { |
| 676 | return fmt.Errorf("previous file bytes changed since they were verified") |
| 677 | } |
| 678 | if expected := previousStateID; expected != "" { |
| 679 | if err := verifyRepairPlanReleaseNodeStateFor(change.PreviousPath, change.TargetPath, expected); err != nil { |
| 680 | return fmt.Errorf("previous state changed while reading file: %w", err) |
| 681 | } |
| 682 | } |
| 683 | return fileutil.AtomicCreateFile(change.TargetPath, b, info.Mode().Perm()) |
| 684 | }() |
| 685 | if restoreErr != nil { |
| 686 | if redo != "" { |
| 687 | if compensateErr := restoreRepairNodeIfAbsent(redo, change.TargetPath); compensateErr != nil { |
| 688 | return nil, fmt.Errorf("undo repair: restore %s: %w (current state retained at %s: %v)", change.TargetPath, restoreErr, redo, compensateErr) |
| 689 | } |
| 690 | } |
| 691 | return nil, fmt.Errorf("undo repair: restore %s: %w", change.TargetPath, restoreErr) |
| 692 | } |
| 693 | if err := markUndone(i); err != nil { |
| 694 | return nil, err |
| 695 | } |
| 696 | _ = removeRepairNodeIfMatching(change.PreviousPath, change.TargetPath, previousStateID) |
| 697 | } |
| 698 | tx.Undone = true |
| 699 | tx.UndoneAt = now.Format(time.RFC3339Nano) |
| 700 | if err := saveRepairTransaction(tx); err != nil { |
| 701 | return nil, err |
| 702 | } |
| 703 | return tx, nil |
| 704 | } |
| 705 | |
| 706 | func verifyUndoRepairBackups(tx *RepairTransaction) error { |
| 707 | if tx == nil { |
| 708 | return fmt.Errorf("undo repair: transaction identity is incomplete") |
| 709 | } |
| 710 | for _, change := range tx.Changes { |
| 711 | if change.RemoveOnUndo { |
| 712 | continue |
| 713 | } |
| 714 | expected := strings.TrimSpace(change.PreviousStateID) |
| 715 | if expected == "" { |
| 716 | return fmt.Errorf("undo repair: previous state identity is missing; legacy transaction cannot be undone safely") |
| 717 | } |
| 718 | if change.Undone { |
| 719 | continue |
| 720 | } |
| 721 | if change.Prepared { |
| 722 | applied, err := preparedRepairChangeApplied(change) |
| 723 | if err != nil { |
| 724 | return fmt.Errorf("undo repair: resolve prepared change: %w", err) |
| 725 | } |
| 726 | if !applied { |
| 727 | continue |
| 728 | } |
| 729 | } |
| 730 | // Lstat: a quarantined symlink counts as present even when its link |
| 731 | // target is gone. Undo restores the link node, not its referent. |
| 732 | if _, err := os.Lstat(change.PreviousPath); err != nil { |
| 733 | return fmt.Errorf("undo repair: previous file %s: %w", change.PreviousPath, err) |
| 734 | } |
| 735 | if err := verifyRepairPlanReleaseNodeStateFor(change.PreviousPath, change.TargetPath, expected); err != nil { |
| 736 | return fmt.Errorf("undo repair: previous state changed: %w", err) |
| 737 | } |
| 738 | } |
| 739 | return nil |
| 740 | } |
| 741 | |
| 742 | func preparedRepairChangeApplied(change RepairChange) (bool, error) { |
| 743 | if !change.Prepared { |
| 744 | return true, nil |
| 745 | } |
| 746 | if change.RemoveOnUndo { |
| 747 | if _, err := os.Lstat(change.TargetPath); err == nil { |
| 748 | // The target was absent when the intent was prepared. Any node now |
| 749 | // present makes this the newest repair transaction, but undo removes |
| 750 | // it only when CreatedStateID proves the repair owns the exact node. |
| 751 | return true, nil |
| 752 | } else if os.IsNotExist(err) { |
| 753 | return false, nil |
| 754 | } else { |
| 755 | return false, err |
| 756 | } |
| 757 | } |
| 758 | expected := strings.TrimSpace(change.PreviousStateID) |
| 759 | if expected == "" { |
| 760 | return false, fmt.Errorf("prepared previous state identity is missing") |
| 761 | } |
| 762 | targetExists := true |
| 763 | if _, err := os.Lstat(change.TargetPath); err != nil { |
| 764 | if !os.IsNotExist(err) { |
| 765 | return false, err |
| 766 | } |
| 767 | targetExists = false |
| 768 | } |
| 769 | if targetExists && verifyRepairPlanReleaseNodeStateFor(change.TargetPath, change.TargetPath, expected) == nil { |
| 770 | // The exact source is still live, so the no-replace rename did not |
| 771 | // commit. Ignore an unrelated/colliding node at PreviousPath. |
| 772 | return false, nil |
| 773 | } |
| 774 | if _, err := os.Lstat(change.PreviousPath); err == nil { |
| 775 | if err := verifyRepairPlanReleaseNodeStateFor(change.PreviousPath, change.TargetPath, expected); err != nil { |
| 776 | return false, fmt.Errorf("prepared previous state changed: %w", err) |
| 777 | } |
| 778 | return true, nil |
| 779 | } else if !os.IsNotExist(err) { |
| 780 | return false, err |
| 781 | } |
| 782 | return false, fmt.Errorf("prepared state is not present at its target or previous path") |
| 783 | } |
| 784 |