| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "encoding/json" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "log/slog" |
| 8 | "os" |
| 9 | "path/filepath" |
| 10 | "strings" |
| 11 | "sync" |
| 12 | "sync/atomic" |
| 13 | "time" |
| 14 | |
| 15 | "reasonix/internal/fileutil" |
| 16 | fileencoding "reasonix/internal/fileutil/encoding" |
| 17 | "reasonix/internal/pathidentity" |
| 18 | "reasonix/internal/store" |
| 19 | ) |
| 20 | |
| 21 | var ErrSessionLeaseHeld = errors.New("session lease held by another runtime") |
| 22 | |
| 23 | // sessionLeaseOwners reserves a canonical session path for one in-process |
| 24 | // acquisition attempt or live lease. sessionLeaseActiveOwners contains only |
| 25 | // generations that have acquired the cross-process lock and written their |
| 26 | // lease metadata. Keeping the two states separate prevents ownership-sensitive |
| 27 | // repair from treating a pending or failed acquisition as proof of ownership. |
| 28 | // Storing identities instead of bare sentinels lets release and reclaim use |
| 29 | // CompareAndDelete without an old generation evicting a newer one. |
| 30 | var ( |
| 31 | sessionLeaseOwners sync.Map |
| 32 | sessionLeaseActiveOwners sync.Map |
| 33 | sessionLeaseSeq atomic.Uint64 |
| 34 | ) |
| 35 | |
| 36 | type SessionLeaseInfo struct { |
| 37 | SessionPath string `json:"session_path"` |
| 38 | WriterID string `json:"writer_id"` |
| 39 | PID int `json:"pid"` |
| 40 | Hostname string `json:"hostname,omitempty"` |
| 41 | AcquiredAt time.Time `json:"acquired_at"` |
| 42 | HandoffTo string `json:"handoff_to,omitempty"` |
| 43 | HandoffID string `json:"handoff_id,omitempty"` |
| 44 | HandoffExpiresAt time.Time `json:"handoff_expires_at,omitempty"` |
| 45 | } |
| 46 | |
| 47 | // MarshalJSON makes the zero time genuinely optional. encoding/json does not |
| 48 | // apply omitempty to a value time.Time, and writing year 1 would make legacy |
| 49 | // metadata look like an explicit (expired) reservation. Older readers still |
| 50 | // ignore these unknown fields, but they do not enforce an active reservation: |
| 51 | // every concurrent writer sharing a state directory must therefore be upgraded |
| 52 | // before takeover is used. |
| 53 | func (i SessionLeaseInfo) MarshalJSON() ([]byte, error) { |
| 54 | type wire struct { |
| 55 | SessionPath string `json:"session_path"` |
| 56 | WriterID string `json:"writer_id"` |
| 57 | PID int `json:"pid"` |
| 58 | Hostname string `json:"hostname,omitempty"` |
| 59 | AcquiredAt time.Time `json:"acquired_at"` |
| 60 | HandoffTo string `json:"handoff_to,omitempty"` |
| 61 | HandoffID string `json:"handoff_id,omitempty"` |
| 62 | HandoffExpiresAt *time.Time `json:"handoff_expires_at,omitempty"` |
| 63 | } |
| 64 | var expires *time.Time |
| 65 | if !i.HandoffExpiresAt.IsZero() { |
| 66 | value := i.HandoffExpiresAt |
| 67 | expires = &value |
| 68 | } |
| 69 | return json.Marshal(wire{ |
| 70 | SessionPath: i.SessionPath, WriterID: i.WriterID, PID: i.PID, Hostname: i.Hostname, |
| 71 | AcquiredAt: i.AcquiredAt, HandoffTo: i.HandoffTo, HandoffID: i.HandoffID, HandoffExpiresAt: expires, |
| 72 | }) |
| 73 | } |
| 74 | |
| 75 | // SessionLeaseHandoffWindow bounds how long a released lease stays reserved |
| 76 | // for its explicitly named successor. The OS lock is free during this window, |
| 77 | // but new-version callers must present the matching writer and generation. |
| 78 | const SessionLeaseHandoffWindow = 30 * time.Second |
| 79 | |
| 80 | const sessionLeaseOwnerOffset int64 = 1 |
| 81 | |
| 82 | func sessionLeaseOwnerBytes(b []byte) []byte { |
| 83 | payload := make([]byte, sessionLeaseOwnerOffset+int64(len(b))) |
| 84 | payload[0] = ' ' |
| 85 | copy(payload[sessionLeaseOwnerOffset:], b) |
| 86 | return payload |
| 87 | } |
| 88 | |
| 89 | type SessionLeaseError struct { |
| 90 | Path string |
| 91 | Info *SessionLeaseInfo |
| 92 | } |
| 93 | |
| 94 | func (e *SessionLeaseError) Error() string { |
| 95 | if e == nil { |
| 96 | return ErrSessionLeaseHeld.Error() |
| 97 | } |
| 98 | if e.Info != nil && e.Info.WriterID != "" { |
| 99 | return fmt.Sprintf("%s: %s is held by %s", ErrSessionLeaseHeld, e.Path, e.Info.WriterID) |
| 100 | } |
| 101 | return fmt.Sprintf("%s: %s", ErrSessionLeaseHeld, e.Path) |
| 102 | } |
| 103 | |
| 104 | func (e *SessionLeaseError) Unwrap() error { |
| 105 | return ErrSessionLeaseHeld |
| 106 | } |
| 107 | |
| 108 | type SessionLease struct { |
| 109 | path string // filesystem identity key |
| 110 | accessPath string // physical path used for file access |
| 111 | legacyAccessPath string // frozen v1 path used by older runtime locks |
| 112 | ownerID uint64 |
| 113 | mu sync.Mutex |
| 114 | leaseLock *sessionLockFile |
| 115 | legacyLeaseLock *sessionLockFile |
| 116 | released bool |
| 117 | writeGeneration uint64 |
| 118 | // activeSaves counts authority-guarded save cycles still inside path/file |
| 119 | // locks. Release waits for this to reach zero so a rebind cannot revoke |
| 120 | // mid-write and create an ABA ownership hole. |
| 121 | activeSaves int |
| 122 | // releaseWait is closed when activeSaves drains to zero while a Release |
| 123 | // is waiting. At most one waiter is parked. |
| 124 | releaseWait chan struct{} |
| 125 | // writerOnce caches the SessionWriter facade for this lease. One writer |
| 126 | // per lease keeps the writer's save serialization meaningful across |
| 127 | // controller rebinds. |
| 128 | writerOnce sync.Once |
| 129 | writer *SessionWriter |
| 130 | // beforeReleaseLock is a test hook for the registry-before-unlock invariant. |
| 131 | beforeReleaseLock func() |
| 132 | // beforeReleaseWait is a test hook reached only after Release observes an |
| 133 | // in-flight authority-guarded save and before it parks. |
| 134 | beforeReleaseWait func() |
| 135 | // beforeHandoffWrite is a test hook for reservation persistence failures. |
| 136 | beforeHandoffWrite func() error |
| 137 | } |
| 138 | |
| 139 | // Writer returns the single SessionWriter facade bound to this lease. The |
| 140 | // first call creates it; every authority minted for a controller rebind goes |
| 141 | // through the same writer, so all of the lease's saves serialize together. |
| 142 | func (l *SessionLease) Writer() *SessionWriter { |
| 143 | if l == nil { |
| 144 | return nil |
| 145 | } |
| 146 | l.writerOnce.Do(func() { |
| 147 | info, err := LoadSessionLeaseInfo(l.accessPath) |
| 148 | if err != nil || info == nil { |
| 149 | info = &SessionLeaseInfo{} |
| 150 | } |
| 151 | l.writer = &SessionWriter{lease: l, info: *info} |
| 152 | }) |
| 153 | return l.writer |
| 154 | } |
| 155 | |
| 156 | func TryAcquireSessionLease(path string) (*SessionLease, error) { |
| 157 | if strings.TrimSpace(path) == "" { |
| 158 | return nil, fmt.Errorf("empty session path") |
| 159 | } |
| 160 | identity, err := resolveSessionPathIdentity(path) |
| 161 | if err != nil { |
| 162 | return nil, err |
| 163 | } |
| 164 | path, accessPath := identity.Key, identity.PhysicalPath |
| 165 | if err := os.MkdirAll(filepath.Dir(accessPath), 0o755); err != nil { |
| 166 | return nil, err |
| 167 | } |
| 168 | ownerID := sessionLeaseSeq.Add(1) |
| 169 | if _, loaded := sessionLeaseOwners.LoadOrStore(path, ownerID); loaded { |
| 170 | info, _ := LoadSessionLeaseInfo(accessPath) |
| 171 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 172 | } |
| 173 | legacyAccessPath := pathidentity.Canonical(identity.AccessPath) |
| 174 | leaseLock, legacyLeaseLock, err := tryTakeCompatibleSessionLeaseLocks(accessPath, legacyAccessPath) |
| 175 | if err != nil { |
| 176 | sessionLeaseOwners.CompareAndDelete(path, ownerID) |
| 177 | if errors.Is(err, ErrSessionLeaseHeld) { |
| 178 | info, _ := LoadSessionLeaseInfo(accessPath) |
| 179 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 180 | } |
| 181 | return nil, err |
| 182 | } |
| 183 | // A handoff reservation is stored inside the lock file before the previous |
| 184 | // holder unlocks it. Re-check only after acquiring the OS lock so a plain |
| 185 | // contender cannot race between reservation publication and unlock. |
| 186 | if info, infoErr := LoadSessionLeaseInfo(accessPath); infoErr == nil && handoffReservationActive(info, time.Now().UTC()) { |
| 187 | unlockSessionLeaseLocks(leaseLock, legacyLeaseLock) |
| 188 | sessionLeaseOwners.CompareAndDelete(path, ownerID) |
| 189 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 190 | } |
| 191 | // The OS lock proves any active-registry entry left without its reservation |
| 192 | // is stale. Clear it before publishing this generation. |
| 193 | sessionLeaseActiveOwners.Delete(path) |
| 194 | lease := &SessionLease{ |
| 195 | path: path, accessPath: accessPath, legacyAccessPath: legacyAccessPath, |
| 196 | ownerID: ownerID, leaseLock: leaseLock, legacyLeaseLock: legacyLeaseLock, |
| 197 | } |
| 198 | if err := lease.publishOwner(); err != nil { |
| 199 | lease.Release() |
| 200 | return nil, err |
| 201 | } |
| 202 | sessionLeaseActiveOwners.Store(path, ownerID) |
| 203 | return lease, nil |
| 204 | } |
| 205 | |
| 206 | // TryReclaimCurrentProcessSessionLease re-acquires a lease whose in-process |
| 207 | // owner entry was orphaned (a lease dropped without Release). The OS lease |
| 208 | // lock is the arbiter: an active holder keeps its lock file locked for the |
| 209 | // whole hold, so reclaiming from one fails with ErrSessionLeaseHeld without |
| 210 | // touching the holder's entry. Holding the lock proves nobody does, which |
| 211 | // also covers metadata-damage states — a missing or unreadable lease info |
| 212 | // (deleted by the user, quarantined by AV, torn by a crash) with a free lock |
| 213 | // is a leftover, not a holder, and must not wedge the session as busy. |
| 214 | func TryReclaimCurrentProcessSessionLease(path string) (*SessionLease, error) { |
| 215 | identity, err := resolveSessionPathIdentity(path) |
| 216 | if err != nil { |
| 217 | return nil, err |
| 218 | } |
| 219 | path, accessPath := identity.Key, identity.PhysicalPath |
| 220 | info, err := LoadSessionLeaseInfo(accessPath) |
| 221 | switch { |
| 222 | case err == nil: |
| 223 | if handoffReservationActive(info, time.Now().UTC()) { |
| 224 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 225 | } |
| 226 | if info == nil || info.PID != os.Getpid() || info.WriterID != SessionWriterID() { |
| 227 | // A readable info naming another live runtime: never steal it. |
| 228 | // (A crashed foreign leftover is separated from a live holder by |
| 229 | // the lock probe in SessionLeaseHeldByOtherRuntime; reclaim is |
| 230 | // only for leases this process lost track of.) |
| 231 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 232 | } |
| 233 | case os.IsNotExist(err): |
| 234 | // The holder finished releasing (info removed first) or the sidecar |
| 235 | // was deleted out from under an orphaned entry. Either way the lock |
| 236 | // probe below decides; info identity has nothing left to say. |
| 237 | info = nil |
| 238 | default: |
| 239 | // Unreadable info hides the holder's identity, but the lock still |
| 240 | // tells the truth: a live holder keeps it locked. Fall through to the |
| 241 | // probe instead of wedging on metadata damage. |
| 242 | info = nil |
| 243 | } |
| 244 | legacyAccessPath := pathidentity.Canonical(identity.AccessPath) |
| 245 | leaseLock, legacyLeaseLock, err := tryTakeCompatibleSessionLeaseLocks(accessPath, legacyAccessPath) |
| 246 | if err != nil { |
| 247 | if errors.Is(err, ErrSessionLeaseHeld) { |
| 248 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 249 | } |
| 250 | return nil, err |
| 251 | } |
| 252 | // Holding the OS lock proves no live lease owns this path right now, so |
| 253 | // overwriting the stale owner entry is safe; concurrent reclaimers fail |
| 254 | // the lock above and never reach this store, and a stale lease released |
| 255 | // later misses its CompareAndDelete against the new owner id. |
| 256 | ownerID := sessionLeaseSeq.Add(1) |
| 257 | lease := &SessionLease{ |
| 258 | path: path, accessPath: accessPath, legacyAccessPath: legacyAccessPath, |
| 259 | ownerID: ownerID, leaseLock: leaseLock, legacyLeaseLock: legacyLeaseLock, |
| 260 | } |
| 261 | sessionLeaseActiveOwners.Delete(path) |
| 262 | sessionLeaseOwners.Store(path, ownerID) |
| 263 | if err := lease.publishOwner(); err != nil { |
| 264 | lease.Release() |
| 265 | return nil, err |
| 266 | } |
| 267 | sessionLeaseActiveOwners.Store(path, ownerID) |
| 268 | return lease, nil |
| 269 | } |
| 270 | |
| 271 | // SessionLeaseHeldByOtherRuntime reports whether path's session lease is held |
| 272 | // by a live runtime other than the calling process. Callers use it to keep |
| 273 | // destructive operations away from sessions another process may be writing; |
| 274 | // leases held by this process report false because callers tear their own |
| 275 | // runtimes down before acting. The lock file is only probed when a foreign |
| 276 | // lease info file exists, so the common uncontended case never touches the |
| 277 | // lock; a probe cannot steal a live lease because holders keep the lock held |
| 278 | // for their whole lifetime. |
| 279 | func SessionLeaseHeldByOtherRuntime(path string) bool { |
| 280 | if strings.TrimSpace(path) == "" { |
| 281 | return false |
| 282 | } |
| 283 | identity, identityErr := resolveSessionPathIdentity(path) |
| 284 | if identityErr != nil { |
| 285 | return true |
| 286 | } |
| 287 | key, accessPath := identity.Key, identity.PhysicalPath |
| 288 | legacyAccessPath := pathidentity.Canonical(identity.AccessPath) |
| 289 | if _, ok := sessionLeaseActiveOwners.Load(key); ok { |
| 290 | // Held by this process; no need to touch the lock file. |
| 291 | return false |
| 292 | } |
| 293 | info, err := loadCompatibleSessionLeaseInfo(accessPath, legacyAccessPath) |
| 294 | if err != nil { |
| 295 | if os.IsNotExist(err) { |
| 296 | // No info file means no holder: live holders keep it present for |
| 297 | // their whole hold. |
| 298 | return false |
| 299 | } |
| 300 | unlock, lockErr := tryLockCompatibleSessionLeaseFiles(accessPath, legacyAccessPath) |
| 301 | if lockErr == nil { |
| 302 | // Corrupt/empty info with a free lock is a crash leftover. Remove the |
| 303 | // bad metadata so future probes do not keep reporting a ghost owner. |
| 304 | _ = os.Remove(sessionLeaseInfoPath(accessPath)) |
| 305 | unlock() |
| 306 | return false |
| 307 | } |
| 308 | // An unreadable info file with a live lock still hides the holder's |
| 309 | // identity, so err on the side of treating the session as busy. |
| 310 | return true |
| 311 | } |
| 312 | if info != nil && info.PID == os.Getpid() && info.WriterID == SessionWriterID() { |
| 313 | return false |
| 314 | } |
| 315 | unlock, err := tryLockCompatibleSessionLeaseFiles(accessPath, legacyAccessPath) |
| 316 | if err == nil { |
| 317 | if handoffReservationActive(info, time.Now().UTC()) { |
| 318 | unlock() |
| 319 | return false |
| 320 | } |
| 321 | // Foreign info but a free lock: leftover from a crashed process. |
| 322 | _ = os.Remove(sessionLeaseInfoPath(accessPath)) |
| 323 | unlock() |
| 324 | return false |
| 325 | } |
| 326 | return true |
| 327 | } |
| 328 | |
| 329 | // InspectSessionLease reports the published owner and whether the OS lock is |
| 330 | // currently held. It never acquires ownership and preserves live handoff |
| 331 | // reservations. Serve uses it to prove that /adopt callers really own the |
| 332 | // session they claim. |
| 333 | func InspectSessionLease(path string) (*SessionLeaseInfo, bool, error) { |
| 334 | if strings.TrimSpace(path) == "" { |
| 335 | return nil, false, fmt.Errorf("empty session path") |
| 336 | } |
| 337 | identity, err := resolveSessionPathIdentity(path) |
| 338 | if err != nil { |
| 339 | return nil, false, err |
| 340 | } |
| 341 | key, accessPath := identity.Key, identity.PhysicalPath |
| 342 | legacyAccessPath := pathidentity.Canonical(identity.AccessPath) |
| 343 | info, err := loadCompatibleSessionLeaseInfo(accessPath, legacyAccessPath) |
| 344 | if err != nil { |
| 345 | return nil, false, err |
| 346 | } |
| 347 | if _, ok := sessionLeaseActiveOwners.Load(key); ok { |
| 348 | return info, true, nil |
| 349 | } |
| 350 | unlock, lockErr := tryLockCompatibleSessionLeaseFiles(accessPath, legacyAccessPath) |
| 351 | if lockErr != nil { |
| 352 | if errors.Is(lockErr, ErrSessionLeaseHeld) { |
| 353 | return info, true, nil |
| 354 | } |
| 355 | return info, false, lockErr |
| 356 | } |
| 357 | unlock() |
| 358 | return info, false, nil |
| 359 | } |
| 360 | |
| 361 | // SessionLeaseHeldByCurrentRuntime reports whether this process has completed |
| 362 | // acquisition of path's session lease. Pending reservations and generations |
| 363 | // already retiring report false, so callers cannot authorize destructive repair |
| 364 | // before the OS lock is held or after release has begun. |
| 365 | func SessionLeaseHeldByCurrentRuntime(path string) bool { |
| 366 | if strings.TrimSpace(path) == "" { |
| 367 | return false |
| 368 | } |
| 369 | _, ok := sessionLeaseActiveOwners.Load(CanonicalSessionPath(path)) |
| 370 | return ok |
| 371 | } |
| 372 | |
| 373 | func (l *SessionLease) Path() string { |
| 374 | if l == nil { |
| 375 | return "" |
| 376 | } |
| 377 | return l.path |
| 378 | } |
| 379 | |
| 380 | // ReleaseForHandoff publishes a target-writer reservation while the current |
| 381 | // lease lock is still held, then releases the OS lock without deleting the |
| 382 | // metadata. A persistence failure leaves the current lease fully active. |
| 383 | func (l *SessionLease) ReleaseForHandoff(targetWriterID, handoffID string) error { |
| 384 | if l == nil { |
| 385 | return nil |
| 386 | } |
| 387 | targetWriterID = strings.TrimSpace(targetWriterID) |
| 388 | handoffID = strings.TrimSpace(handoffID) |
| 389 | if targetWriterID == "" || handoffID == "" { |
| 390 | return fmt.Errorf("handoff target writer id and generation are required") |
| 391 | } |
| 392 | for { |
| 393 | l.mu.Lock() |
| 394 | if l.released { |
| 395 | l.mu.Unlock() |
| 396 | return ErrSessionLeaseHeld |
| 397 | } |
| 398 | if l.activeSaves == 0 { |
| 399 | break |
| 400 | } |
| 401 | if l.releaseWait == nil { |
| 402 | l.releaseWait = make(chan struct{}) |
| 403 | } |
| 404 | wait := l.releaseWait |
| 405 | beforeReleaseWait := l.beforeReleaseWait |
| 406 | l.mu.Unlock() |
| 407 | if beforeReleaseWait != nil { |
| 408 | beforeReleaseWait() |
| 409 | } |
| 410 | <-wait |
| 411 | } |
| 412 | if l.beforeHandoffWrite != nil { |
| 413 | if err := l.beforeHandoffWrite(); err != nil { |
| 414 | l.mu.Unlock() |
| 415 | return err |
| 416 | } |
| 417 | } |
| 418 | info := newSessionLeaseInfo(l.accessPath) |
| 419 | info.HandoffTo = targetWriterID |
| 420 | info.HandoffID = handoffID |
| 421 | info.HandoffExpiresAt = time.Now().UTC().Add(SessionLeaseHandoffWindow) |
| 422 | if err := l.writeOwnerInfo(info); err != nil { |
| 423 | l.mu.Unlock() |
| 424 | return err |
| 425 | } |
| 426 | l.released = true |
| 427 | leaseLock := l.leaseLock |
| 428 | legacyLeaseLock := l.legacyLeaseLock |
| 429 | l.leaseLock = nil |
| 430 | l.legacyLeaseLock = nil |
| 431 | l.mu.Unlock() |
| 432 | |
| 433 | sessionLeaseActiveOwners.CompareAndDelete(l.path, l.ownerID) |
| 434 | sessionLeaseOwners.CompareAndDelete(l.path, l.ownerID) |
| 435 | unlockSessionLeaseLocks(leaseLock, legacyLeaseLock) |
| 436 | _ = os.Remove(sessionLeaseInfoPath(l.accessPath)) |
| 437 | _ = removeStaleSessionLockSidecar(l.accessPath, store.SessionLockFile(l.accessPath)) |
| 438 | return nil |
| 439 | } |
| 440 | |
| 441 | // TryAcquireSessionLeaseWithHandoff consumes one unexpired reservation for the |
| 442 | // current process writer. The reservation is checked again while holding the |
| 443 | // OS lock, fencing stale grants and check-then-use races. |
| 444 | func TryAcquireSessionLeaseWithHandoff(path, sourceWriterID, handoffID string) (*SessionLease, error) { |
| 445 | if strings.TrimSpace(path) == "" { |
| 446 | return nil, fmt.Errorf("empty session path") |
| 447 | } |
| 448 | identity, err := resolveSessionPathIdentity(path) |
| 449 | if err != nil { |
| 450 | return nil, err |
| 451 | } |
| 452 | path, accessPath := identity.Key, identity.PhysicalPath |
| 453 | if err := os.MkdirAll(filepath.Dir(accessPath), 0o755); err != nil { |
| 454 | return nil, err |
| 455 | } |
| 456 | ownerID := sessionLeaseSeq.Add(1) |
| 457 | if _, loaded := sessionLeaseOwners.LoadOrStore(path, ownerID); loaded { |
| 458 | info, _ := LoadSessionLeaseInfo(accessPath) |
| 459 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 460 | } |
| 461 | legacyAccessPath := pathidentity.Canonical(identity.AccessPath) |
| 462 | leaseLock, legacyLeaseLock, err := tryTakeCompatibleSessionLeaseLocks(accessPath, legacyAccessPath) |
| 463 | if err != nil { |
| 464 | sessionLeaseOwners.CompareAndDelete(path, ownerID) |
| 465 | info, _ := LoadSessionLeaseInfo(accessPath) |
| 466 | if errors.Is(err, ErrSessionLeaseHeld) { |
| 467 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 468 | } |
| 469 | return nil, err |
| 470 | } |
| 471 | info, infoErr := LoadSessionLeaseInfo(accessPath) |
| 472 | if infoErr != nil || !handoffReservationMatches(info, sourceWriterID, SessionWriterID(), handoffID, time.Now().UTC()) { |
| 473 | unlockSessionLeaseLocks(leaseLock, legacyLeaseLock) |
| 474 | sessionLeaseOwners.CompareAndDelete(path, ownerID) |
| 475 | return nil, &SessionLeaseError{Path: path, Info: info} |
| 476 | } |
| 477 | sessionLeaseActiveOwners.Delete(path) |
| 478 | lease := &SessionLease{ |
| 479 | path: path, accessPath: accessPath, legacyAccessPath: legacyAccessPath, |
| 480 | ownerID: ownerID, leaseLock: leaseLock, legacyLeaseLock: legacyLeaseLock, |
| 481 | } |
| 482 | if err := lease.publishOwner(); err != nil { |
| 483 | lease.Release() |
| 484 | return nil, err |
| 485 | } |
| 486 | sessionLeaseActiveOwners.Store(path, ownerID) |
| 487 | _ = os.Remove(sessionLeaseInfoPath(accessPath)) |
| 488 | return lease, nil |
| 489 | } |
| 490 | |
| 491 | func handoffReservationActive(info *SessionLeaseInfo, now time.Time) bool { |
| 492 | if info == nil || strings.TrimSpace(info.HandoffTo) == "" || strings.TrimSpace(info.HandoffID) == "" { |
| 493 | return false |
| 494 | } |
| 495 | return info.HandoffExpiresAt.IsZero() || now.Before(info.HandoffExpiresAt) |
| 496 | } |
| 497 | |
| 498 | func handoffReservationMatches(info *SessionLeaseInfo, sourceWriterID, targetWriterID, handoffID string, now time.Time) bool { |
| 499 | if !handoffReservationActive(info, now) { |
| 500 | return false |
| 501 | } |
| 502 | return strings.TrimSpace(info.WriterID) == strings.TrimSpace(sourceWriterID) && |
| 503 | strings.TrimSpace(info.HandoffTo) == strings.TrimSpace(targetWriterID) && |
| 504 | strings.TrimSpace(info.HandoffID) == strings.TrimSpace(handoffID) |
| 505 | } |
| 506 | |
| 507 | func (l *SessionLease) Release() { |
| 508 | if l == nil { |
| 509 | return |
| 510 | } |
| 511 | // Wait for authority-guarded saves to finish before revoking ownership. |
| 512 | // Without this, a concurrent save that already passed Valid() can finish |
| 513 | // after a successor lease is issued for the same path (ABA). |
| 514 | for { |
| 515 | l.mu.Lock() |
| 516 | if l.released { |
| 517 | l.mu.Unlock() |
| 518 | return |
| 519 | } |
| 520 | if l.activeSaves == 0 { |
| 521 | break |
| 522 | } |
| 523 | if l.releaseWait == nil { |
| 524 | l.releaseWait = make(chan struct{}) |
| 525 | } |
| 526 | wait := l.releaseWait |
| 527 | beforeReleaseWait := l.beforeReleaseWait |
| 528 | l.mu.Unlock() |
| 529 | if beforeReleaseWait != nil { |
| 530 | beforeReleaseWait() |
| 531 | } |
| 532 | <-wait |
| 533 | } |
| 534 | l.released = true |
| 535 | leaseLock := l.leaseLock |
| 536 | legacyLeaseLock := l.legacyLeaseLock |
| 537 | l.leaseLock = nil |
| 538 | l.legacyLeaseLock = nil |
| 539 | beforeReleaseLock := l.beforeReleaseLock |
| 540 | l.mu.Unlock() |
| 541 | |
| 542 | // Revoke ownership-sensitive repair before the OS lock becomes available |
| 543 | // to a successor. CompareAndDelete keeps a stale generation from |
| 544 | // deauthorizing a newer reclaimed lease. |
| 545 | sessionLeaseActiveOwners.CompareAndDelete(l.path, l.ownerID) |
| 546 | _ = os.Remove(sessionLeaseInfoPath(l.accessPath)) |
| 547 | // Only remove the entry this lease owns: after a reclaim the map may |
| 548 | // already point at a newer lease for the same path. |
| 549 | sessionLeaseOwners.CompareAndDelete(l.path, l.ownerID) |
| 550 | if beforeReleaseLock != nil { |
| 551 | beforeReleaseLock() |
| 552 | } |
| 553 | if leaseLock != nil { |
| 554 | // Delete the exact lock file while its lock is still held. Besides |
| 555 | // retiring the sidecar, retaining the lock object enables an atomic |
| 556 | // handoff to SessionRemovalGuard without an unlock/reacquire window. |
| 557 | _ = leaseLock.RemoveAndUnlock() |
| 558 | } |
| 559 | if legacyLeaseLock != nil { |
| 560 | _ = legacyLeaseLock.RemoveAndUnlock() |
| 561 | } |
| 562 | _ = removeStaleSessionLockSidecar(l.accessPath, store.SessionLockFile(l.accessPath)) |
| 563 | } |
| 564 | |
| 565 | func newSessionLeaseInfo(path string) SessionLeaseInfo { |
| 566 | host, _ := os.Hostname() |
| 567 | return SessionLeaseInfo{ |
| 568 | SessionPath: path, |
| 569 | WriterID: SessionWriterID(), |
| 570 | PID: os.Getpid(), |
| 571 | Hostname: host, |
| 572 | AcquiredAt: time.Now().UTC(), |
| 573 | } |
| 574 | } |
| 575 | |
| 576 | func (l *SessionLease) publishOwner() error { |
| 577 | info := newSessionLeaseInfo(l.accessPath) |
| 578 | return l.writeOwnerInfo(info) |
| 579 | } |
| 580 | |
| 581 | func (l *SessionLease) writeOwnerInfo(info SessionLeaseInfo) error { |
| 582 | if err := writeSessionLeaseInfo(l.leaseLock, info); err != nil { |
| 583 | return err |
| 584 | } |
| 585 | if l.legacyLeaseLock != nil { |
| 586 | if err := writeSessionLeaseInfo(l.legacyLeaseLock, info); err != nil { |
| 587 | return err |
| 588 | } |
| 589 | } |
| 590 | return nil |
| 591 | } |
| 592 | |
| 593 | // tryTakeCompatibleSessionLeaseLocks preserves the v1.38.10 lock name while |
| 594 | // making the physical-path lock authoritative. The compatibility lock is only |
| 595 | // opened when its directory entry has a distinct identity, avoiding a second |
| 596 | // lock attempt on the same file on case-insensitive filesystems. |
| 597 | func tryTakeCompatibleSessionLeaseLocks(accessPath, legacyAccessPath string) (*sessionLockFile, *sessionLockFile, error) { |
| 598 | primary, err := tryTakeSessionLeaseLock(accessPath) |
| 599 | if err != nil { |
| 600 | return nil, nil, err |
| 601 | } |
| 602 | if sameSessionLeaseLockIdentity(accessPath, legacyAccessPath) { |
| 603 | return primary, nil, nil |
| 604 | } |
| 605 | legacy, err := tryTakeSessionLeaseLock(legacyAccessPath) |
| 606 | if err != nil { |
| 607 | // Preserve any reservation metadata already stored in the primary |
| 608 | // lock file. Stale empty lock files are harmless and cleaned later. |
| 609 | primary.Unlock() |
| 610 | return nil, nil, err |
| 611 | } |
| 612 | return primary, legacy, nil |
| 613 | } |
| 614 | |
| 615 | func sameSessionLeaseLockIdentity(a, b string) bool { |
| 616 | if strings.TrimSpace(a) == "" || strings.TrimSpace(b) == "" { |
| 617 | return true |
| 618 | } |
| 619 | left, leftErr := pathidentity.Resolve(store.SessionLeaseLock(a), pathidentity.Options{FollowLeaf: false}) |
| 620 | right, rightErr := pathidentity.Resolve(store.SessionLeaseLock(b), pathidentity.Options{FollowLeaf: false}) |
| 621 | return leftErr == nil && rightErr == nil && left.Key == right.Key |
| 622 | } |
| 623 | |
| 624 | func unlockSessionLeaseLocks(primary, legacy *sessionLockFile) { |
| 625 | if legacy != nil { |
| 626 | legacy.Unlock() |
| 627 | } |
| 628 | if primary != nil { |
| 629 | primary.Unlock() |
| 630 | } |
| 631 | } |
| 632 | |
| 633 | func tryLockCompatibleSessionLeaseFiles(accessPath, legacyAccessPath string) (func(), error) { |
| 634 | primary, legacy, err := tryTakeCompatibleSessionLeaseLocks(accessPath, legacyAccessPath) |
| 635 | if err != nil { |
| 636 | return nil, err |
| 637 | } |
| 638 | return func() { unlockSessionLeaseLocks(primary, legacy) }, nil |
| 639 | } |
| 640 | |
| 641 | func loadCompatibleSessionLeaseInfo(accessPath, legacyAccessPath string) (*SessionLeaseInfo, error) { |
| 642 | info, err := LoadSessionLeaseInfo(accessPath) |
| 643 | if err == nil || !os.IsNotExist(err) || sameSessionLeaseLockIdentity(accessPath, legacyAccessPath) { |
| 644 | return info, err |
| 645 | } |
| 646 | return LoadSessionLeaseInfo(legacyAccessPath) |
| 647 | } |
| 648 | |
| 649 | func writeSessionLeaseInfo(leaseLock *sessionLockFile, info SessionLeaseInfo) error { |
| 650 | b, err := json.MarshalIndent(info, "", " ") |
| 651 | if err != nil { |
| 652 | return err |
| 653 | } |
| 654 | b = append(b, '\n') |
| 655 | if leaseLock == nil { |
| 656 | return errors.New("session lease lock not held") |
| 657 | } |
| 658 | return leaseLock.writeOwnerInfo(b) |
| 659 | } |
| 660 | |
| 661 | // errSessionLeaseInfoCorrupt marks a present-but-undecodable lease-info |
| 662 | // source (empty or invalid bytes). Readers treat it as "identity hidden, |
| 663 | // let the lock decide" instead of "no holder". |
| 664 | var errSessionLeaseInfoCorrupt = errors.New("session lease info corrupt") |
| 665 | |
| 666 | // LoadSessionLeaseInfo reports the holder identity for path. New writers |
| 667 | // publish it inside .lease.lock; the .lease.json sidecar is a read-only |
| 668 | // compatibility source for sessions last held by older builds. An empty or |
| 669 | // undecodable source reads as corrupt (the live lock is the truth); only the |
| 670 | // absence of both sources reads as no-holder. |
| 671 | func LoadSessionLeaseInfo(path string) (*SessionLeaseInfo, error) { |
| 672 | lockPath := store.SessionLeaseLock(canonicalSessionSavePath(path)) |
| 673 | if raw, err := readSessionLeaseLockFile(lockPath); err == nil { |
| 674 | b := fileencoding.DecodeToUTF8(raw) |
| 675 | if info, decodeErr := decodeSessionLeaseInfo(b); decodeErr == nil { |
| 676 | return info, nil |
| 677 | } else if !errors.Is(decodeErr, os.ErrNotExist) { |
| 678 | return nil, decodeErr |
| 679 | } |
| 680 | } else if !os.IsNotExist(err) { |
| 681 | // An unreadable lock file (permission damage, torn disk) still hides |
| 682 | // the holder identity; fail on the read rather than silently falling |
| 683 | // back to the legacy sidecar. |
| 684 | return nil, err |
| 685 | } |
| 686 | b, err := fileencoding.ReadFileUTF8(sessionLeaseInfoPath(path)) |
| 687 | if err != nil { |
| 688 | return nil, err |
| 689 | } |
| 690 | return decodeSessionLeaseInfo(b) |
| 691 | } |
| 692 | |
| 693 | func decodeSessionLeaseInfo(b []byte) (*SessionLeaseInfo, error) { |
| 694 | if len(strings.TrimSpace(string(b))) == 0 { |
| 695 | return nil, errSessionLeaseInfoCorrupt |
| 696 | } |
| 697 | var info SessionLeaseInfo |
| 698 | if err := json.Unmarshal(b, &info); err != nil { |
| 699 | return nil, fmt.Errorf("%w: %w", errSessionLeaseInfoCorrupt, err) |
| 700 | } |
| 701 | return &info, nil |
| 702 | } |
| 703 | |
| 704 | // SaveSessionLeaseInfo writes the legacy .lease.json sidecar. Production |
| 705 | // writers publish owner identity inside .lease.lock instead; this remains for |
| 706 | // tests and tooling that need to stage the compatibility read path. |
| 707 | func SaveSessionLeaseInfo(path string, info SessionLeaseInfo) error { |
| 708 | leasePath := sessionLeaseInfoPath(path) |
| 709 | if err := os.MkdirAll(filepath.Dir(leasePath), 0o755); err != nil { |
| 710 | return err |
| 711 | } |
| 712 | b, err := json.MarshalIndent(info, "", " ") |
| 713 | if err != nil { |
| 714 | return err |
| 715 | } |
| 716 | b = append(b, '\n') |
| 717 | tmp, err := os.CreateTemp(filepath.Dir(leasePath), ".lease.*.tmp") |
| 718 | if err != nil { |
| 719 | return err |
| 720 | } |
| 721 | tmpPath := tmp.Name() |
| 722 | if _, err := tmp.Write(b); err != nil { |
| 723 | tmp.Close() |
| 724 | os.Remove(tmpPath) |
| 725 | return err |
| 726 | } |
| 727 | if err := tmp.Close(); err != nil { |
| 728 | os.Remove(tmpPath) |
| 729 | return err |
| 730 | } |
| 731 | if err := fileutil.ReplaceFile(tmpPath, leasePath); err != nil { |
| 732 | os.Remove(tmpPath) |
| 733 | return err |
| 734 | } |
| 735 | return nil |
| 736 | } |
| 737 | |
| 738 | func sessionLeaseInfoPath(path string) string { |
| 739 | return store.SessionLeaseInfo(canonicalSessionSavePath(path)) |
| 740 | } |
| 741 | |
| 742 | // unleasedWriteObserved dedupes the write-authority probe below to one report |
| 743 | // per canonical path per process. |
| 744 | var unleasedWriteObserved sync.Map |
| 745 | |
| 746 | // observeUnleasedSessionWrite is the store-P2 write-authority probe: the target |
| 747 | // model is "the lease holder is the only writer of a session's content", but |
| 748 | // enforcement can't land before we know every writer that currently saves |
| 749 | // without holding the lease (fresh-session creation saves before the first |
| 750 | // Rebind, headless runs, recovery tooling, ...). Until then this only records |
| 751 | // evidence: one structured warning per path per process, never a failure. The |
| 752 | // snapshot-conflict machinery stays the safety net for the writers this |
| 753 | // surfaces. |
| 754 | func observeUnleasedSessionWrite(path string, mode sessionSaveMode) { |
| 755 | canonical := CanonicalSessionPath(path) |
| 756 | if _, ok := sessionLeaseOwners.Load(canonical); ok { |
| 757 | return |
| 758 | } |
| 759 | if _, seen := unleasedWriteObserved.LoadOrStore(canonical, struct{}{}); seen { |
| 760 | return |
| 761 | } |
| 762 | slog.Warn("session: save without a held lease (write-authority probe, store P2)", |
| 763 | "path", filepath.Base(path), |
| 764 | "mode", int(mode), |
| 765 | "writer", SessionWriterID(), |
| 766 | ) |
| 767 | } |
| 768 |