| 1 | package repair |
| 2 | |
| 3 | import ( |
| 4 | "crypto/sha256" |
| 5 | "encoding/hex" |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "os" |
| 9 | "path/filepath" |
| 10 | "slices" |
| 11 | "strings" |
| 12 | "sync" |
| 13 | "time" |
| 14 | |
| 15 | "reasonix/internal/config" |
| 16 | "reasonix/internal/pathidentity" |
| 17 | ) |
| 18 | |
| 19 | const repairMutationLockTimeout = 5 * time.Second |
| 20 | |
| 21 | // repairMutationBeforeLock is a test seam for forcing competing repair |
| 22 | // operations to overlap before one waits on the shared file lock. |
| 23 | var repairMutationBeforeLock = func([]string) {} |
| 24 | |
| 25 | // repairMutationBeforeRename is a test seam for changing a target in the |
| 26 | // narrow interval between its final state check and quarantine rename. |
| 27 | var repairMutationBeforeRename = func(string) {} |
| 28 | |
| 29 | // repairMutationAfterRename is a test seam for forcing an uncooperative writer |
| 30 | // to create a new target after the confirmed node has been quarantined. |
| 31 | var repairMutationAfterRename = func(string) {} |
| 32 | |
| 33 | // repairMutationAfterPrepare is a test seam for simulating process exit after |
| 34 | // the write-ahead repair intent is durable but before the filesystem rename. |
| 35 | var repairMutationAfterPrepare = func(string) {} |
| 36 | |
| 37 | var repairPathCaseInsensitive = platformRepairPathCaseInsensitive |
| 38 | |
| 39 | func lockRepairTransaction() (func(), error) { |
| 40 | expectedPendingState := repairPlanReleaseNodeState(pendingRepairTransactionPath()) |
| 41 | unlock, err := lockRepairMutationProtocolFile(repairTransactionPath()) |
| 42 | if err != nil { |
| 43 | return nil, fmt.Errorf("lock repair transaction: %w", err) |
| 44 | } |
| 45 | if actual := repairPlanReleaseNodeState(pendingRepairTransactionPath()); actual != expectedPendingState { |
| 46 | unlock() |
| 47 | return nil, fmt.Errorf("lock repair transaction: pending repair transaction changed while waiting") |
| 48 | } |
| 49 | return unlock, nil |
| 50 | } |
| 51 | |
| 52 | func restoreRepairNodeIfAbsent(backup, target string) error { |
| 53 | // Every backup passed here was produced by renaming the target to a sibling |
| 54 | // or to a same-filesystem repair directory. A no-replace rename restores the |
| 55 | // exact node and consumes the backup in one operation. Recreating a link/file |
| 56 | // and then removing backup would let another writer replace backup between |
| 57 | // those syscalls and have its node deleted. |
| 58 | if err := renameRepairNodeNoReplace(backup, target); err != nil { |
| 59 | return fmt.Errorf("restore repair target: %w", err) |
| 60 | } |
| 61 | return nil |
| 62 | } |
| 63 | |
| 64 | // removeRepairNodeIfMatching displaces and verifies a transaction backup. |
| 65 | // Replaced paths are restored or retained, never unlinked as transaction-owned. |
| 66 | func removeRepairNodeIfMatching(path, identityPath, expectedStateID string) error { |
| 67 | expectedStateID = strings.TrimSpace(expectedStateID) |
| 68 | if expectedStateID == "" { |
| 69 | // Legacy transactions did not persist backup identity. Leaving a stale |
| 70 | // backup is safer than deleting a path whose ownership cannot be proven. |
| 71 | return nil |
| 72 | } |
| 73 | cleanup, err := moveRepairNodeToUniqueCleanup(path) |
| 74 | if err != nil { |
| 75 | return err |
| 76 | } |
| 77 | if cleanup == "" { |
| 78 | return nil |
| 79 | } |
| 80 | if err := verifyRepairPlanReleaseNodeStateFor(cleanup, identityPath, expectedStateID); err != nil { |
| 81 | if restoreErr := renameRepairNodeNoReplace(cleanup, path); restoreErr != nil { |
| 82 | return errors.Join(err, fmt.Errorf("preserve changed repair backup at %s: %w", cleanup, restoreErr)) |
| 83 | } |
| 84 | return err |
| 85 | } |
| 86 | info, err := os.Lstat(cleanup) |
| 87 | if err != nil { |
| 88 | return err |
| 89 | } |
| 90 | if info.IsDir() { |
| 91 | if restoreErr := renameRepairNodeNoReplace(cleanup, path); restoreErr != nil { |
| 92 | return errors.Join( |
| 93 | fmt.Errorf("remove repair backup: directories are unsupported"), |
| 94 | fmt.Errorf("preserve repair backup at %s: %w", cleanup, restoreErr), |
| 95 | ) |
| 96 | } |
| 97 | return fmt.Errorf("remove repair backup: directories are unsupported") |
| 98 | } |
| 99 | return os.Remove(cleanup) |
| 100 | } |
| 101 | |
| 102 | func moveRepairNodeToUniqueCleanup(path string) (string, error) { |
| 103 | for attempt := range 16 { |
| 104 | cleanup := fmt.Sprintf("%s.reasonix-cleanup-%d-%d", path, time.Now().UTC().UnixNano(), attempt) |
| 105 | err := renameRepairNodeNoReplace(path, cleanup) |
| 106 | if err == nil { |
| 107 | return cleanup, nil |
| 108 | } |
| 109 | if os.IsNotExist(err) { |
| 110 | return "", nil |
| 111 | } |
| 112 | if os.IsExist(err) { |
| 113 | continue |
| 114 | } |
| 115 | return "", err |
| 116 | } |
| 117 | return "", fmt.Errorf("remove repair node: cannot allocate cleanup path") |
| 118 | } |
| 119 | |
| 120 | // canonicalRepairPath resolves a repair target to a stable key shared by |
| 121 | // mutation locks and preview identity. Parent-directory symlinks are followed |
| 122 | // so alias paths converge, but the leaf name is never resolved: repair mutates |
| 123 | // the leaf node itself via Lstat/Rename (including when the leaf is a symlink). |
| 124 | // Case-insensitive filesystems fold case so /Project and /project cannot take |
| 125 | // different locks. The decision is made from the target's actual parent |
| 126 | // directory: macOS and Windows can both host case-sensitive directories. |
| 127 | func canonicalRepairPath(path string) string { |
| 128 | identity, err := pathidentity.Resolve(path, pathidentity.Options{FollowLeaf: false}) |
| 129 | if err != nil { |
| 130 | return "" |
| 131 | } |
| 132 | return identity.Key |
| 133 | } |
| 134 | |
| 135 | // legacyCanonicalRepairPath freezes the lock and persisted-target identity used |
| 136 | // before path identity v2. New writers acquire both names during the supported |
| 137 | // cross-version window. |
| 138 | func legacyCanonicalRepairPath(path string) string { |
| 139 | path = strings.TrimSpace(path) |
| 140 | if path == "" { |
| 141 | return "" |
| 142 | } |
| 143 | absolute, err := filepath.Abs(filepath.Clean(path)) |
| 144 | if err != nil { |
| 145 | absolute = filepath.Clean(path) |
| 146 | } |
| 147 | absolute = resolveParentSymlinkPath(absolute) |
| 148 | absolute = filepath.Clean(absolute) |
| 149 | caseInsensitive := repairPathCaseInsensitive(absolute) |
| 150 | absolute = platformRepairPathUnicodeNormalized(absolute) |
| 151 | if caseInsensitive { |
| 152 | return strings.ToLower(filepath.ToSlash(absolute)) |
| 153 | } |
| 154 | return absolute |
| 155 | } |
| 156 | |
| 157 | // resolveParentSymlinkPath resolves symlink parents of path and re-attaches the |
| 158 | // original leaf base name. The leaf is intentionally not EvalSymlinks'd: two |
| 159 | // different symlink leaves that share a referent must stay distinct targets. |
| 160 | func resolveParentSymlinkPath(path string) string { |
| 161 | if path == "" { |
| 162 | return "" |
| 163 | } |
| 164 | parent := filepath.Dir(path) |
| 165 | base := filepath.Base(path) |
| 166 | if parent == path { |
| 167 | // Root or volume path: nothing to resolve above the leaf. |
| 168 | return path |
| 169 | } |
| 170 | if resolved, err := filepath.EvalSymlinks(parent); err == nil { |
| 171 | return filepath.Join(resolved, base) |
| 172 | } |
| 173 | // Parent may not exist yet (create-only targets). Resolve the longest |
| 174 | // existing ancestor and rejoin the missing components including the leaf. |
| 175 | var missing []string |
| 176 | dir := parent |
| 177 | missing = append(missing, base) |
| 178 | for { |
| 179 | if resolved, err := filepath.EvalSymlinks(dir); err == nil { |
| 180 | parts := make([]string, 0, 1+len(missing)) |
| 181 | parts = append(parts, resolved) |
| 182 | for _, v := range slices.Backward(missing) { |
| 183 | parts = append(parts, v) |
| 184 | } |
| 185 | return filepath.Join(parts...) |
| 186 | } |
| 187 | next := filepath.Dir(dir) |
| 188 | if next == dir { |
| 189 | return path |
| 190 | } |
| 191 | missing = append(missing, filepath.Base(dir)) |
| 192 | dir = next |
| 193 | } |
| 194 | } |
| 195 | |
| 196 | // LockRepairMutations is the exported form of lockRepairMutations for desktop |
| 197 | // handoff helpers that replace release-unit paths outside ApplyRepairPlan. |
| 198 | func LockRepairMutations(paths ...string) (func(), error) { |
| 199 | return lockRepairMutations(paths...) |
| 200 | } |
| 201 | |
| 202 | // LockRepairMutationsTimeout is like LockRepairMutations but waits up to |
| 203 | // timeout for competing repair or update holders. |
| 204 | func LockRepairMutationsTimeout(timeout time.Duration, paths ...string) (func(), error) { |
| 205 | if timeout <= 0 { |
| 206 | timeout = repairMutationLockTimeout |
| 207 | } |
| 208 | return lockRepairMutationsTimeout(timeout, paths...) |
| 209 | } |
| 210 | |
| 211 | // repairPlanTargetIdentity is a non-reversible identity for a filesystem |
| 212 | // target. It is embedded in preview state IDs so confirmation cannot be |
| 213 | // reused against a different real path that happens to have the same content. |
| 214 | func repairPlanTargetIdentity(path string) string { |
| 215 | key := canonicalRepairPath(path) |
| 216 | if key == "" { |
| 217 | return "" |
| 218 | } |
| 219 | sum := sha256.Sum256([]byte(key)) |
| 220 | return hex.EncodeToString(sum[:]) |
| 221 | } |
| 222 | |
| 223 | // lockRepairMutations serializes repair read-check-write cycles by canonical |
| 224 | // target path. Lock files live in Reasonix state rather than beside project or |
| 225 | // configuration files, and paths are sorted so multi-target actions cannot |
| 226 | // deadlock each other. |
| 227 | func lockRepairMutations(paths ...string) (func(), error) { |
| 228 | return lockRepairMutationsTimeoutMode(repairMutationLockTimeout, true, paths...) |
| 229 | } |
| 230 | |
| 231 | func lockRepairMutationsTimeout(timeout time.Duration, paths ...string) (func(), error) { |
| 232 | return lockRepairMutationsTimeoutMode(timeout, true, paths...) |
| 233 | } |
| 234 | |
| 235 | // Protocol files are expected to be atomically replaced by the previous lock |
| 236 | // holder. Their callers compare content state after acquisition, so only the |
| 237 | // lock domain is shared here; ordinary repair targets still require native |
| 238 | // file identity to remain unchanged while waiting. |
| 239 | func lockRepairMutationProtocolFile(path string) (func(), error) { |
| 240 | return lockRepairMutationsTimeoutMode(repairMutationLockTimeout, false, path) |
| 241 | } |
| 242 | |
| 243 | func lockRepairMutationsTimeoutMode(timeout time.Duration, revalidateTargets bool, paths ...string) (func(), error) { |
| 244 | lockDir := config.RepairMutationLockDir() |
| 245 | if lockDir == "" { |
| 246 | return nil, fmt.Errorf("lock repair mutations: OS user cache directory is unavailable") |
| 247 | } |
| 248 | if err := os.MkdirAll(lockDir, 0o700); err != nil { |
| 249 | return nil, fmt.Errorf("lock repair mutations: create lock directory: %w", err) |
| 250 | } |
| 251 | |
| 252 | targets, primaryKeys, lockKeys, err := repairMutationTargets(paths) |
| 253 | if err != nil { |
| 254 | return nil, err |
| 255 | } |
| 256 | if len(targets) == 0 { |
| 257 | return func() {}, nil |
| 258 | } |
| 259 | repairMutationBeforeLock(append([]string(nil), primaryKeys...)) |
| 260 | domains, err := repairMutationLockDomains(lockDir, lockKeys) |
| 261 | if err != nil { |
| 262 | return nil, err |
| 263 | } |
| 264 | releases, err := acquireRepairMutationLocks(timeout, domains) |
| 265 | if err != nil { |
| 266 | return nil, err |
| 267 | } |
| 268 | if revalidateTargets { |
| 269 | if err := revalidateRepairMutationTargets(targets); err != nil { |
| 270 | releaseRepairMutationLocks(releases) |
| 271 | return nil, err |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | var once sync.Once |
| 276 | return func() { |
| 277 | once.Do(func() { |
| 278 | releaseRepairMutationLocks(releases) |
| 279 | }) |
| 280 | }, nil |
| 281 | } |
| 282 |