| 1 | package cli |
| 2 | |
| 3 | import ( |
| 4 | "crypto/rand" |
| 5 | "encoding/hex" |
| 6 | "encoding/json" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "net" |
| 10 | "os" |
| 11 | "path/filepath" |
| 12 | "sort" |
| 13 | "strconv" |
| 14 | "strings" |
| 15 | "sync" |
| 16 | "time" |
| 17 | "unicode/utf8" |
| 18 | |
| 19 | "reasonix/internal/config" |
| 20 | "reasonix/internal/fileutil" |
| 21 | "reasonix/internal/store" |
| 22 | ) |
| 23 | |
| 24 | func serveConfigWithCommandDefaults(command string, authExplicit bool, cfg config.ServeConfig) config.ServeConfig { |
| 25 | if command == "web" && !authExplicit { |
| 26 | cfg.AuthMode = "token" |
| 27 | } |
| 28 | return cfg |
| 29 | } |
| 30 | |
| 31 | const ( |
| 32 | webPortRetryLimit = 100 |
| 33 | webInstanceHeartbeat = 15 * time.Second |
| 34 | maxWebSessionIDBytes = 218 // leaves room for .jsonl and session sidecars |
| 35 | webInstanceDirectoryName = "instances" |
| 36 | ) |
| 37 | |
| 38 | // listenWebWithPortRetry binds addr, walking port+1 only when a concrete port |
| 39 | // is already occupied. Port 0 remains an ordinary kernel-assigned ephemeral |
| 40 | // bind, and non-EADDRINUSE failures are returned immediately. |
| 41 | func listenWebWithPortRetry(addr string) (net.Listener, error) { |
| 42 | host, rawPort, err := net.SplitHostPort(addr) |
| 43 | if err != nil { |
| 44 | return nil, err |
| 45 | } |
| 46 | port, err := strconv.Atoi(rawPort) |
| 47 | if err != nil || port < 0 || port > 65535 { |
| 48 | return nil, fmt.Errorf("invalid listen port %q", rawPort) |
| 49 | } |
| 50 | if port == 0 { |
| 51 | return net.Listen("tcp", addr) |
| 52 | } |
| 53 | for attempt := 0; ; attempt++ { |
| 54 | candidate := net.JoinHostPort(host, strconv.Itoa(port)) |
| 55 | ln, listenErr := net.Listen("tcp", candidate) |
| 56 | if listenErr == nil { |
| 57 | return ln, nil |
| 58 | } |
| 59 | if !webAddressInUse(listenErr) || attempt >= webPortRetryLimit || port >= 65535 { |
| 60 | return nil, listenErr |
| 61 | } |
| 62 | port++ |
| 63 | } |
| 64 | } |
| 65 | |
| 66 | func requestedPort(addr string) int { |
| 67 | _, rawPort, err := net.SplitHostPort(addr) |
| 68 | if err != nil { |
| 69 | return -1 |
| 70 | } |
| 71 | port, err := strconv.Atoi(rawPort) |
| 72 | if err != nil { |
| 73 | return -1 |
| 74 | } |
| 75 | return port |
| 76 | } |
| 77 | |
| 78 | func validateWebSessionID(id string) error { |
| 79 | if strings.TrimSpace(id) == "" { |
| 80 | return errors.New("--session-id cannot be empty") |
| 81 | } |
| 82 | if !utf8.ValidString(id) || len(id) > maxWebSessionIDBytes { |
| 83 | return fmt.Errorf("invalid Web session identity %q", id) |
| 84 | } |
| 85 | if id == "." || id == ".." || strings.ContainsAny(id, `/\`) || strings.IndexByte(id, 0) >= 0 { |
| 86 | return fmt.Errorf("invalid Web session identity %q", id) |
| 87 | } |
| 88 | if !store.IsSessionTranscriptName(id + ".jsonl") { |
| 89 | return fmt.Errorf("invalid Web session identity %q", id) |
| 90 | } |
| 91 | return nil |
| 92 | } |
| 93 | |
| 94 | func freshWebSessionPath(dir, id string) (string, error) { |
| 95 | if err := validateWebSessionID(id); err != nil { |
| 96 | return "", err |
| 97 | } |
| 98 | path := filepath.Join(dir, id+".jsonl") |
| 99 | if _, err := os.Lstat(path); err == nil { |
| 100 | return "", fmt.Errorf("fresh Web session already exists: %s", path) |
| 101 | } else if !os.IsNotExist(err) { |
| 102 | return "", err |
| 103 | } |
| 104 | return path, nil |
| 105 | } |
| 106 | |
| 107 | // webInstanceRecord is stable for independent operator-facing readers. |
| 108 | // Unknown fields remain forward-compatible with independent readers. |
| 109 | type webInstanceRecord struct { |
| 110 | ServerID string `json:"server_id"` |
| 111 | PID int `json:"pid"` |
| 112 | Host string `json:"host"` |
| 113 | Port int `json:"port"` |
| 114 | StartedAt int64 `json:"started_at"` |
| 115 | HeartbeatAt int64 `json:"heartbeat_at"` |
| 116 | } |
| 117 | |
| 118 | type webInstanceRegistry struct { |
| 119 | dir string |
| 120 | now func() time.Time |
| 121 | heartbeatInterval time.Duration |
| 122 | processAlive func(int) bool |
| 123 | } |
| 124 | |
| 125 | type webInstanceRegistration struct { |
| 126 | path string |
| 127 | record webInstanceRecord |
| 128 | registry *webInstanceRegistry |
| 129 | stop chan struct{} |
| 130 | done chan struct{} |
| 131 | releaseOne sync.Once |
| 132 | } |
| 133 | |
| 134 | func registerWebInstance(reasonixHome, addr string) (*webInstanceRegistration, error) { |
| 135 | if strings.TrimSpace(reasonixHome) == "" { |
| 136 | return nil, errors.New("cannot register Web instance: Reasonix home is empty") |
| 137 | } |
| 138 | registry := &webInstanceRegistry{ |
| 139 | dir: filepath.Join(reasonixHome, "server", webInstanceDirectoryName), |
| 140 | now: time.Now, |
| 141 | heartbeatInterval: webInstanceHeartbeat, |
| 142 | processAlive: webInstanceProcessAlive, |
| 143 | } |
| 144 | return registry.register(addr, os.Getpid()) |
| 145 | } |
| 146 | |
| 147 | func (r *webInstanceRegistry) register(addr string, pid int) (*webInstanceRegistration, error) { |
| 148 | if strings.TrimSpace(r.dir) == "" { |
| 149 | return nil, errors.New("cannot register Web instance: Reasonix home is empty") |
| 150 | } |
| 151 | host, rawPort, err := net.SplitHostPort(addr) |
| 152 | if err != nil { |
| 153 | return nil, fmt.Errorf("register Web instance: %w", err) |
| 154 | } |
| 155 | port, err := strconv.Atoi(rawPort) |
| 156 | if err != nil || port <= 0 || port > 65535 { |
| 157 | return nil, fmt.Errorf("register Web instance: invalid bound port %q", rawPort) |
| 158 | } |
| 159 | if err := os.MkdirAll(r.dir, 0o700); err != nil { |
| 160 | return nil, fmt.Errorf("create Web instance registry: %w", err) |
| 161 | } |
| 162 | if err := r.sweepStale(); err != nil { |
| 163 | return nil, fmt.Errorf("sweep Web instance registry: %w", err) |
| 164 | } |
| 165 | |
| 166 | now := r.now().UnixMilli() |
| 167 | for range 8 { |
| 168 | serverID, err := randomWebInstanceID() |
| 169 | if err != nil { |
| 170 | return nil, fmt.Errorf("generate Web instance id: %w", err) |
| 171 | } |
| 172 | record := webInstanceRecord{ |
| 173 | ServerID: serverID, |
| 174 | PID: pid, |
| 175 | Host: host, |
| 176 | Port: port, |
| 177 | StartedAt: now, |
| 178 | HeartbeatAt: now, |
| 179 | } |
| 180 | path := filepath.Join(r.dir, serverID+".json") |
| 181 | data, err := json.Marshal(record) |
| 182 | if err != nil { |
| 183 | return nil, err |
| 184 | } |
| 185 | if err := fileutil.AtomicCreateFile(path, data, 0o600); err != nil { |
| 186 | if errors.Is(err, os.ErrExist) { |
| 187 | continue |
| 188 | } |
| 189 | return nil, fmt.Errorf("register Web instance: %w", err) |
| 190 | } |
| 191 | reg := &webInstanceRegistration{ |
| 192 | path: path, |
| 193 | record: record, |
| 194 | registry: r, |
| 195 | stop: make(chan struct{}), |
| 196 | done: make(chan struct{}), |
| 197 | } |
| 198 | go reg.heartbeat() |
| 199 | return reg, nil |
| 200 | } |
| 201 | return nil, errors.New("register Web instance: could not allocate a unique id") |
| 202 | } |
| 203 | |
| 204 | func randomWebInstanceID() (string, error) { |
| 205 | var raw [16]byte |
| 206 | if _, err := rand.Read(raw[:]); err != nil { |
| 207 | return "", err |
| 208 | } |
| 209 | return hex.EncodeToString(raw[:]), nil |
| 210 | } |
| 211 | |
| 212 | func (r *webInstanceRegistry) sweepStale() error { |
| 213 | entries, err := os.ReadDir(r.dir) |
| 214 | if os.IsNotExist(err) { |
| 215 | return nil |
| 216 | } |
| 217 | if err != nil { |
| 218 | return err |
| 219 | } |
| 220 | for _, entry := range entries { |
| 221 | if entry.IsDir() || filepath.Ext(entry.Name()) != ".json" { |
| 222 | continue |
| 223 | } |
| 224 | path := filepath.Join(r.dir, entry.Name()) |
| 225 | record, ok := readWebInstanceRecord(path) |
| 226 | // Malformed files may belong to a newer/live writer. Only delete an entry |
| 227 | // that can be positively identified as owned by a dead process. |
| 228 | if !ok || r.processAlive(record.PID) { |
| 229 | continue |
| 230 | } |
| 231 | if err := os.Remove(path); err != nil && !os.IsNotExist(err) { |
| 232 | return err |
| 233 | } |
| 234 | } |
| 235 | return nil |
| 236 | } |
| 237 | |
| 238 | func (r *webInstanceRegistry) listLive() ([]webInstanceRecord, error) { |
| 239 | if err := r.sweepStale(); err != nil { |
| 240 | return nil, err |
| 241 | } |
| 242 | entries, err := os.ReadDir(r.dir) |
| 243 | if os.IsNotExist(err) { |
| 244 | return nil, nil |
| 245 | } |
| 246 | if err != nil { |
| 247 | return nil, err |
| 248 | } |
| 249 | live := make([]webInstanceRecord, 0, len(entries)) |
| 250 | for _, entry := range entries { |
| 251 | if entry.IsDir() || filepath.Ext(entry.Name()) != ".json" { |
| 252 | continue |
| 253 | } |
| 254 | if record, ok := readWebInstanceRecord(filepath.Join(r.dir, entry.Name())); ok && r.processAlive(record.PID) { |
| 255 | live = append(live, record) |
| 256 | } |
| 257 | } |
| 258 | sort.Slice(live, func(i, j int) bool { return live[i].StartedAt < live[j].StartedAt }) |
| 259 | return live, nil |
| 260 | } |
| 261 | |
| 262 | func readWebInstanceRecord(path string) (webInstanceRecord, bool) { |
| 263 | data, err := os.ReadFile(path) |
| 264 | if err != nil { |
| 265 | return webInstanceRecord{}, false |
| 266 | } |
| 267 | var record webInstanceRecord |
| 268 | if json.Unmarshal(data, &record) != nil || record.ServerID == "" || record.PID <= 0 || record.Host == "" || record.Port <= 0 || record.StartedAt <= 0 || record.HeartbeatAt <= 0 { |
| 269 | return webInstanceRecord{}, false |
| 270 | } |
| 271 | return record, true |
| 272 | } |
| 273 | |
| 274 | func (r *webInstanceRegistration) heartbeat() { |
| 275 | defer close(r.done) |
| 276 | interval := r.registry.heartbeatInterval |
| 277 | if interval <= 0 { |
| 278 | <-r.stop |
| 279 | return |
| 280 | } |
| 281 | ticker := time.NewTicker(interval) |
| 282 | defer ticker.Stop() |
| 283 | for { |
| 284 | select { |
| 285 | case <-ticker.C: |
| 286 | r.record.HeartbeatAt = r.registry.now().UnixMilli() |
| 287 | if data, err := json.Marshal(r.record); err == nil { |
| 288 | _ = fileutil.AtomicWriteFile(r.path, data, 0o600) |
| 289 | } |
| 290 | case <-r.stop: |
| 291 | return |
| 292 | } |
| 293 | } |
| 294 | } |
| 295 | |
| 296 | // Release stops heartbeats before removing the single-writer instance file, so |
| 297 | // an in-flight atomic rename cannot recreate a supposedly released entry. |
| 298 | func (r *webInstanceRegistration) Release() { |
| 299 | if r == nil { |
| 300 | return |
| 301 | } |
| 302 | r.releaseOne.Do(func() { |
| 303 | close(r.stop) |
| 304 | <-r.done |
| 305 | _ = os.Remove(r.path) |
| 306 | }) |
| 307 | } |
| 308 |