| 1 | package main |
| 2 | |
| 3 | import ( |
| 4 | "bufio" |
| 5 | "context" |
| 6 | "encoding/json" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "log" |
| 10 | "net" |
| 11 | "net/http" |
| 12 | "net/http/cookiejar" |
| 13 | "runtime" |
| 14 | "sort" |
| 15 | "strconv" |
| 16 | "strings" |
| 17 | "time" |
| 18 | |
| 19 | "golang.org/x/crypto/ssh" |
| 20 | |
| 21 | "reasonix/internal/config" |
| 22 | "reasonix/internal/jobs" |
| 23 | "reasonix/internal/remote/bootstrap" |
| 24 | "reasonix/internal/remote/forward" |
| 25 | "reasonix/internal/store" |
| 26 | ) |
| 27 | |
| 28 | const remoteProviderReloadTimeout = jobs.DefaultTeardownGrace + 15*time.Second |
| 29 | |
| 30 | // SwitchCredentialProxyModel stages an immutable desktop proxy route and a |
| 31 | // a complete resolver bundle, then asks a capable Serve to publish it at the |
| 32 | // existing active-work/session boundary. An unknown outcome is read back; the |
| 33 | // write is never replayed or rolled back against a possibly newer controller. |
| 34 | func (m *desktopRemoteManager) SwitchCredentialProxyModel(ctx context.Context, hostID, workspace, currentRef, nextRef, expectedPath string) error { |
| 35 | hostID, workspace = strings.TrimSpace(hostID), strings.TrimSpace(workspace) |
| 36 | currentRef, nextRef = strings.TrimSpace(currentRef), strings.TrimSpace(nextRef) |
| 37 | if hostID == "" || workspace == "" || currentRef == "" || nextRef == "" { |
| 38 | return fmt.Errorf("credential proxy model switch: host, workspace, current model, and next model are required") |
| 39 | } |
| 40 | if ctx == nil { |
| 41 | ctx = context.Background() |
| 42 | } |
| 43 | mh := m.managed(hostID) |
| 44 | if mh == nil || mh.client == nil { |
| 45 | return fmt.Errorf("host %q is not connected", hostID) |
| 46 | } |
| 47 | mh.serveMu.Lock() |
| 48 | defer mh.serveMu.Unlock() |
| 49 | if !m.isCurrent(hostID, mh) { |
| 50 | return fmt.Errorf("host %q connection was replaced", hostID) |
| 51 | } |
| 52 | m.mu.Lock() |
| 53 | serve := mh.serves[workspace] |
| 54 | m.mu.Unlock() |
| 55 | if serve == nil || serve.view.State != "ready" || serve.view.LocalURL == "" || serve.token == "" { |
| 56 | return fmt.Errorf("workspace %q has no ready Reasonix Serve", workspace) |
| 57 | } |
| 58 | app, ok := m.sink.(*App) |
| 59 | if !ok || app == nil { |
| 60 | return fmt.Errorf("credential proxy: app unavailable") |
| 61 | } |
| 62 | client, err := newServeHTTPClient(serve.view.LocalURL) |
| 63 | if err != nil { |
| 64 | return err |
| 65 | } |
| 66 | if err := serveHandshake(ctx, client, serve.view.LocalURL, serve.token); err != nil { |
| 67 | return err |
| 68 | } |
| 69 | status, err := remoteModelSettingsRequest(ctx, client, serve.view.LocalURL, expectedPath, nil) |
| 70 | if err != nil { |
| 71 | return err |
| 72 | } |
| 73 | cfg, err := config.LoadModelRuntimeSnapshot(".") |
| 74 | if err != nil { |
| 75 | return err |
| 76 | } |
| 77 | port, err := app.credentialProxyPort() |
| 78 | if err != nil { |
| 79 | return err |
| 80 | } |
| 81 | if !m.pinModelSettingsOwnership(app, hostID, workspace, mh, status) { |
| 82 | return fmt.Errorf("remote Serve must support ordered model settings ownership") |
| 83 | } |
| 84 | remotePort, err := ensureCredentialProxyForward(mh.client, hostID, port) |
| 85 | if err != nil { |
| 86 | return fmt.Errorf("credential proxy: reverse tunnel: %w", err) |
| 87 | } |
| 88 | bundle, remoteRef, err := app.buildRemoteModelSettings(hostID, workspace, nextRef, remotePort, cfg) |
| 89 | if err != nil { |
| 90 | return err |
| 91 | } |
| 92 | _, err = app.installRemoteModelSettingsSnapshot(ctx, client, serve.view.LocalURL, hostID, workspace, expectedPath, remoteRef, bundle, status) |
| 93 | if err != nil { |
| 94 | return err |
| 95 | } |
| 96 | if !m.isCurrent(hostID, mh) { |
| 97 | return fmt.Errorf("remote connection changed while applying model settings") |
| 98 | } |
| 99 | app.remoteTabMu.Lock() |
| 100 | for _, tab := range app.remoteTabs { |
| 101 | if tab != nil && tab.ref.HostID == hostID && tab.ref.Workspace == workspace && tab.routing.currentPath == expectedPath { |
| 102 | tab.settings.revision = bundle.Revision |
| 103 | tab.settings.failure = "" |
| 104 | tab.settings.generation = tab.gen |
| 105 | tab.settings.sessionPath = expectedPath |
| 106 | } |
| 107 | } |
| 108 | app.remoteTabMu.Unlock() |
| 109 | return nil |
| 110 | } |
| 111 | |
| 112 | func credentialProxyBootstrapOptions(workspace string, remotePort int, info credentialProxyRouteInfo) *bootstrap.CredentialProxyOptions { |
| 113 | slug := store.RemoteWorkspaceSlug(workspace) |
| 114 | suffix := slug[len(slug)-16:] |
| 115 | if info.revision != "" { |
| 116 | suffix += "-" + info.revision[:16] |
| 117 | } |
| 118 | return &bootstrap.CredentialProxyOptions{ |
| 119 | BaseURL: fmt.Sprintf("http://127.0.0.1:%d", remotePort), |
| 120 | Token: info.token, |
| 121 | TokenEnv: "REASONIX_PROXY_TOKEN_" + strings.ToUpper(strings.ReplaceAll(suffix, "-", "_")), |
| 122 | Provider: credentialProxyProviderName + "-" + suffix, |
| 123 | Model: info.model, |
| 124 | Kind: info.kind, |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | // serveForwardName derives the per-workspace local tunnel name from the same |
| 129 | // collision-proof slug the remote state files use (store.RemoteWorkspaceSlug), |
| 130 | // so one host holds one independent forward per workspace. |
| 131 | func serveForwardName(workspace string) string { |
| 132 | return "serve-" + store.RemoteWorkspaceSlug(workspace) |
| 133 | } |
| 134 | |
| 135 | func (m *desktopRemoteManager) EnsureServer(ctx context.Context, hostID, workspace string) (RemoteServerView, string, error) { |
| 136 | hostID, workspace = strings.TrimSpace(hostID), strings.TrimSpace(workspace) |
| 137 | if hostID == "" || workspace == "" { |
| 138 | return RemoteServerView{}, "", fmt.Errorf("remote serve: host and workspace are required") |
| 139 | } |
| 140 | if ctx == nil { |
| 141 | ctx = context.Background() |
| 142 | } |
| 143 | mh := m.managed(hostID) |
| 144 | if mh == nil || mh.client == nil { |
| 145 | return RemoteServerView{}, "", fmt.Errorf("host %q is not connected", hostID) |
| 146 | } |
| 147 | // Serialize per-host so two concurrent EnsureServer calls cannot both miss |
| 148 | // the state and launch duplicate/orphan serve processes. |
| 149 | mh.serveMu.Lock() |
| 150 | defer mh.serveMu.Unlock() |
| 151 | m.mu.Lock() |
| 152 | if m.hosts[hostID] != mh { |
| 153 | m.mu.Unlock() |
| 154 | return RemoteServerView{}, "", fmt.Errorf("host %q connection was replaced", hostID) |
| 155 | } |
| 156 | var previousServer RemoteServerView |
| 157 | previousToken := "" |
| 158 | previousAddr := "" |
| 159 | if e := mh.serves[workspace]; e != nil { |
| 160 | previousServer, previousToken, previousAddr = e.view, e.token, e.addr |
| 161 | } |
| 162 | m.mu.Unlock() |
| 163 | c := mh.client |
| 164 | if m.readyServeReusable(ctx, c, mh, hostID, workspace, previousServer, previousToken, previousAddr) { |
| 165 | m.startCredentialWatchdogIfEnabled(mh, hostID, workspace) |
| 166 | return previousServer, previousToken, nil |
| 167 | } |
| 168 | opCtx, cancel := managedOperationContext(ctx, mh) |
| 169 | defer cancel() |
| 170 | |
| 171 | entry, err := configuredRemoteHost(hostID) |
| 172 | if err != nil { |
| 173 | return RemoteServerView{}, "", err |
| 174 | } |
| 175 | starting := RemoteServerView{HostID: hostID, Workspace: workspace, State: "starting"} |
| 176 | if !m.publishServerIfCurrent(hostID, mh, starting, "", "") { |
| 177 | return RemoteServerView{}, "", fmt.Errorf("host %q connection was replaced", hostID) |
| 178 | } |
| 179 | // Local-proxy credential mode: start the desktop key holder, open the |
| 180 | // reverse tunnel, and hand the bootstrap the virtual token + provider |
| 181 | // entry to install on the remote. The real key never leaves this machine. |
| 182 | credOpts, err := m.credentialOptions(c, hostID, workspace, entry) |
| 183 | if err != nil { |
| 184 | view := RemoteServerView{HostID: hostID, Workspace: workspace, State: "error", Error: err.Error()} |
| 185 | m.publishFailedServeStart(hostID, mh, previousServer, previousToken, previousAddr, view) |
| 186 | return view, "", err |
| 187 | } |
| 188 | res, err := m.ensureServe(opCtx, c, bootstrap.Options{ |
| 189 | Workspace: workspace, |
| 190 | Install: entry.ServeInstallMode(), |
| 191 | LocalBinary: m.localBinary(), |
| 192 | LocalGOOS: runtime.GOOS, |
| 193 | LocalGOARCH: runtime.GOARCH, |
| 194 | ProductVersion: version, |
| 195 | FetchBinary: m.fetchRemoteBinary, |
| 196 | MinVersion: bootstrap.MinServeVersion, |
| 197 | CredentialProxy: credOpts, |
| 198 | BrowserBroker: m.browserBrokerCallback(c, hostID, mh), |
| 199 | Progress: func(step, detail string) { |
| 200 | view := RemoteServerView{HostID: hostID, Workspace: workspace, State: step, Message: detail} |
| 201 | m.publishServerIfCurrent(hostID, mh, view, "", "") |
| 202 | }, |
| 203 | }) |
| 204 | if err != nil { |
| 205 | view := RemoteServerView{HostID: hostID, Workspace: workspace, State: "error", Error: err.Error()} |
| 206 | m.publishFailedServeStart(hostID, mh, previousServer, previousToken, previousAddr, view) |
| 207 | return view, "", err |
| 208 | } |
| 209 | if !m.isCurrent(hostID, mh) { |
| 210 | return RemoteServerView{}, "", fmt.Errorf("host %q connection was replaced", hostID) |
| 211 | } |
| 212 | if res.Reused && previousServer.State == "ready" && previousServer.Workspace == workspace && |
| 213 | hasUsableServeForward(c.Forwards().List(), serveForwardName(workspace), res.State.Addr, previousServer.LocalURL) { |
| 214 | previousServer.InstanceID = remoteServeInstanceID(res.State) |
| 215 | if !m.publishServerIfCurrent(hostID, mh, previousServer, res.Token, res.State.Addr) { |
| 216 | return RemoteServerView{}, "", fmt.Errorf("host %q connection was replaced", hostID) |
| 217 | } |
| 218 | // A reused process still carries the previous generation's broker |
| 219 | // environment; rotate the route and rebind it to this connection. |
| 220 | m.rebindBrowserBrokerBestEffort(opCtx, c, mh, hostID, previousServer, res.Token) |
| 221 | return m.finishCredentialServe(opCtx, c, mh, hostID, workspace, previousServer, res.Token, res, entry.CredentialProxyEnabled()) |
| 222 | } |
| 223 | // Start the replacement before retiring the old tunnel. If binding fails, |
| 224 | // the previous ready server stays usable instead of leaving a dead gap. |
| 225 | bound, ferr := c.Forwards().Replace(forward.Spec{ |
| 226 | Name: serveForwardName(workspace), Direction: forward.Local, BindAddr: "127.0.0.1:0", TargetAddr: res.State.Addr, |
| 227 | }) |
| 228 | if ferr != nil { |
| 229 | if !res.Reused { |
| 230 | cleanupCtx, cleanupCancel := context.WithTimeout(context.Background(), 5*time.Second) |
| 231 | _ = m.stopServe(cleanupCtx, c, workspace) |
| 232 | cleanupCancel() |
| 233 | } |
| 234 | view := RemoteServerView{HostID: hostID, Workspace: workspace, State: "error", Error: ferr.Error()} |
| 235 | m.publishFailedServeStart(hostID, mh, previousServer, previousToken, previousAddr, view) |
| 236 | return view, "", ferr |
| 237 | } |
| 238 | localURL := fmt.Sprintf("http://%s/", bound) |
| 239 | view := RemoteServerView{HostID: hostID, Workspace: workspace, State: "ready", LocalURL: localURL, InstanceID: remoteServeInstanceID(res.State)} |
| 240 | if !m.publishServerIfCurrent(hostID, mh, view, res.Token, res.State.Addr) { |
| 241 | _ = c.Forwards().Remove(serveForwardName(workspace)) |
| 242 | return RemoteServerView{}, "", fmt.Errorf("host %q connection was replaced", hostID) |
| 243 | } |
| 244 | return m.finishCredentialServe(opCtx, c, mh, hostID, workspace, view, res.Token, res, entry.CredentialProxyEnabled()) |
| 245 | } |
| 246 | |
| 247 | func remoteServeInstanceID(state bootstrap.ServeState) string { |
| 248 | if state.PID == 0 && state.StartedAt == 0 && strings.TrimSpace(state.Addr) == "" { |
| 249 | return "" |
| 250 | } |
| 251 | return fmt.Sprintf("%d:%d:%s", state.PID, state.StartedAt, strings.TrimSpace(state.Addr)) |
| 252 | } |
| 253 | |
| 254 | func configuredRemoteHost(hostID string) (config.RemoteHostEntry, error) { |
| 255 | cfg, err := config.Load() |
| 256 | if err != nil { |
| 257 | return config.RemoteHostEntry{}, err |
| 258 | } |
| 259 | entry, ok := cfg.RemoteHost(hostID) |
| 260 | if !ok { |
| 261 | return config.RemoteHostEntry{}, fmt.Errorf("remote host %q is no longer configured", hostID) |
| 262 | } |
| 263 | return entry, nil |
| 264 | } |
| 265 | |
| 266 | func (m *desktopRemoteManager) credentialOptions(c desktopSSHClient, hostID, workspace string, entry config.RemoteHostEntry) (*bootstrap.CredentialProxyOptions, error) { |
| 267 | if !entry.CredentialProxyEnabled() { |
| 268 | return nil, nil |
| 269 | } |
| 270 | return m.credentialProxySetup(c, hostID, workspace) |
| 271 | } |
| 272 | |
| 273 | // readyServeReusable validates the serve forward and credential channel. |
| 274 | // Config read failures fail closed into the full ensure path. |
| 275 | func (m *desktopRemoteManager) readyServeReusable(ctx context.Context, c desktopSSHClient, mh *managedHost, hostID, workspace string, view RemoteServerView, token, addr string) bool { |
| 276 | if view.State != "ready" || view.LocalURL == "" || token == "" || addr == "" || |
| 277 | !hasUsableServeForward(c.Forwards().List(), serveForwardName(workspace), addr, view.LocalURL) || |
| 278 | !serveTunnelAlive(ctx, view.LocalURL) { |
| 279 | return false |
| 280 | } |
| 281 | cfg, err := config.Load() |
| 282 | if err != nil { |
| 283 | return false |
| 284 | } |
| 285 | host, ok := cfg.RemoteHost(hostID) |
| 286 | if !ok || !host.CredentialProxyEnabled() { |
| 287 | return true |
| 288 | } |
| 289 | port, hasForward := credentialForwardPort(c, hostID) |
| 290 | healedPort := int(mh.credPort.Load()) |
| 291 | if !hasForward || healedPort != port || probeReverseTunnel(c, port) != nil { |
| 292 | log.Printf("[remote] EnsureServer: FAST-REUSE blocked (credential channel) host=%s ws=%s port=%d has=%v healedPort=%d", hostID, workspace, port, hasForward, healedPort) |
| 293 | return false |
| 294 | } |
| 295 | return true |
| 296 | } |
| 297 | |
| 298 | // healCredentialChannel runs after ensure when the forward is live. A reconnect |
| 299 | // can move its remote port, so heal every tracked workspace config before any |
| 300 | // Serve reloads providers, then verify the channel before recording the port. |
| 301 | func (m *desktopRemoteManager) healCredentialChannel(ctx context.Context, c desktopSSHClient, mh *managedHost, hostID, workspace, base, token string, res bootstrap.Result) error { |
| 302 | port, has := credentialForwardPort(c, hostID) |
| 303 | if !has { |
| 304 | return fmt.Errorf("credential proxy: reverse tunnel is not available") |
| 305 | } |
| 306 | // The tunnel secret rotates on every SSH reconnection even when the remote |
| 307 | // port is reused, and a running serve keeps validating the previous token |
| 308 | // until its providers are rebuilt from the healed disk config. So heal the |
| 309 | // tracked configs and reload unconditionally: both steps are idempotent, |
| 310 | // and skipping the config heal on a same-port rebind leaves every model |
| 311 | // call failing with "invalid credential proxy token". |
| 312 | if int(mh.credPort.Load()) != port || res.CredentialConfigChanged { |
| 313 | log.Printf("[remote] EnsureServer: cred port drift host=%s old=%d new=%d configChanged=%v -> healing serve credentials", hostID, mh.credPort.Load(), port, res.CredentialConfigChanged) |
| 314 | } else { |
| 315 | log.Printf("[remote] EnsureServer: same cred port %d -> healing serve credentials (fresh tunnel secret)", port) |
| 316 | } |
| 317 | workspaces := m.trackedCredentialWorkspaces(hostID, workspace) |
| 318 | // base+token covers the Serve ensured by this round even if it has not |
| 319 | // reached the registry yet; tracked peers are reloaded alongside it. |
| 320 | if err := healCredentialConfigsBeforeReload(ctx, workspaces, |
| 321 | func(workspace string) (*bootstrap.CredentialProxyOptions, error) { |
| 322 | return m.credentialProxySetup(c, hostID, workspace) |
| 323 | }, |
| 324 | func(ctx context.Context, opts *bootstrap.CredentialProxyOptions) error { |
| 325 | _, err := bootstrap.HealCredentialProvider(ctx, c, opts) |
| 326 | return err |
| 327 | }, |
| 328 | func() bool { return m.reloadServeProviders(ctx, mh, hostID, workspace, base, token) }, |
| 329 | ); err != nil { |
| 330 | return err |
| 331 | } |
| 332 | if perr := probeReverseTunnel(c, port); perr != nil { |
| 333 | log.Printf("[remote] EnsureServer: reverse probe FAILED host=%s ws=%s port=%d err=%v", hostID, workspace, port, perr) |
| 334 | return fmt.Errorf("credential proxy: reverse tunnel health check: %w", perr) |
| 335 | } |
| 336 | m.mu.Lock() |
| 337 | defer m.mu.Unlock() |
| 338 | if m.hosts[hostID] != mh { |
| 339 | return fmt.Errorf("host %q connection was replaced", hostID) |
| 340 | } |
| 341 | mh.credPort.Store(int64(port)) |
| 342 | return nil |
| 343 | } |
| 344 | |
| 345 | // serveTunnelAlive probes the local serve forward: any HTTP response — |
| 346 | // the remote serve behind it are answering. |
| 347 | func serveTunnelAlive(ctx context.Context, localURL string) bool { |
| 348 | probeCtx, cancel := context.WithTimeout(ctx, 2*time.Second) |
| 349 | defer cancel() |
| 350 | req, err := http.NewRequestWithContext(probeCtx, http.MethodGet, serveURL(localURL, "/"), nil) |
| 351 | if err != nil { |
| 352 | return false |
| 353 | } |
| 354 | resp, err := http.DefaultClient.Do(req) |
| 355 | if err != nil { |
| 356 | return false |
| 357 | } |
| 358 | _ = resp.Body.Close() |
| 359 | return true |
| 360 | } |
| 361 | |
| 362 | func (m *desktopRemoteManager) StopServer(hostID, workspace string) error { |
| 363 | mh := m.managed(hostID) |
| 364 | if mh == nil || mh.client == nil { |
| 365 | return fmt.Errorf("host %q is not connected", hostID) |
| 366 | } |
| 367 | mh.serveMu.Lock() |
| 368 | defer mh.serveMu.Unlock() |
| 369 | if !m.isCurrent(hostID, mh) { |
| 370 | return fmt.Errorf("host %q connection was replaced", hostID) |
| 371 | } |
| 372 | c := mh.client |
| 373 | m.mu.Lock() |
| 374 | _, tracked := mh.serves[workspace] |
| 375 | m.mu.Unlock() |
| 376 | if !tracked { |
| 377 | return fmt.Errorf("host %q has no managed server for workspace %q", hostID, workspace) |
| 378 | } |
| 379 | opCtx, cancel := managedOperationContext(context.Background(), mh) |
| 380 | defer cancel() |
| 381 | if err := m.stopServe(opCtx, c, workspace); err != nil { |
| 382 | return err |
| 383 | } |
| 384 | // Tear down the local serve tunnel so a stale forward can't linger. |
| 385 | _ = c.Forwards().Remove(serveForwardName(workspace)) |
| 386 | // A stopped serve keeps its broker environment; drop the host's route so |
| 387 | // the token dies with the process that held it. |
| 388 | m.revokeBrowserBroker(hostID) |
| 389 | view := RemoteServerView{HostID: hostID, Workspace: workspace, State: "stopped"} |
| 390 | m.publishServerIfCurrent(hostID, mh, view, "", "") |
| 391 | return nil |
| 392 | } |
| 393 | |
| 394 | // managed returns the managed host record for hostID, or nil. |
| 395 | func (m *desktopRemoteManager) managed(hostID string) *managedHost { |
| 396 | m.mu.Lock() |
| 397 | defer m.mu.Unlock() |
| 398 | return m.hosts[hostID] |
| 399 | } |
| 400 | |
| 401 | func (m *desktopRemoteManager) ServerStatus(hostID, workspace string) RemoteServerView { |
| 402 | m.mu.Lock() |
| 403 | defer m.mu.Unlock() |
| 404 | if mh := m.hosts[hostID]; mh != nil { |
| 405 | if e := mh.serves[workspace]; e != nil { |
| 406 | return e.view |
| 407 | } |
| 408 | } |
| 409 | return RemoteServerView{HostID: hostID, Workspace: workspace, State: "stopped"} |
| 410 | } |
| 411 | |
| 412 | // ServeSnapshot is the read-only resolution for callers that want to talk to |
| 413 | // an already-running serve without waking one: it returns the registry's view |
| 414 | // and token only when the recorded state is ready with a usable URL. Query |
| 415 | // paths (session listing) must go through this — a full EnsureServer from a |
| 416 | // poll serializes behind tab bootstraps on the per-host lock and starves them. |
| 417 | func (m *desktopRemoteManager) ServeSnapshot(hostID, workspace string) (RemoteServerView, string, bool) { |
| 418 | m.mu.Lock() |
| 419 | defer m.mu.Unlock() |
| 420 | mh := m.hosts[hostID] |
| 421 | if mh == nil { |
| 422 | return RemoteServerView{}, "", false |
| 423 | } |
| 424 | e := mh.serves[workspace] |
| 425 | if e == nil || e.view.State != "ready" || e.view.LocalURL == "" || e.token == "" { |
| 426 | return RemoteServerView{}, "", false |
| 427 | } |
| 428 | return e.view, e.token, true |
| 429 | } |
| 430 | |
| 431 | func (m *desktopRemoteManager) ServerLogs(ctx context.Context, hostID, workspace string, tailLines int) (string, error) { |
| 432 | m.mu.Lock() |
| 433 | mh := m.hosts[hostID] |
| 434 | tracked := false |
| 435 | if mh != nil { |
| 436 | _, tracked = mh.serves[workspace] |
| 437 | } |
| 438 | m.mu.Unlock() |
| 439 | if mh == nil || mh.client == nil { |
| 440 | return "", fmt.Errorf("host %q is not connected", hostID) |
| 441 | } |
| 442 | if !tracked { |
| 443 | return "", fmt.Errorf("host %q has no managed server for workspace %q", hostID, workspace) |
| 444 | } |
| 445 | opCtx, cancel := managedOperationContext(ctx, mh) |
| 446 | defer cancel() |
| 447 | var sb strings.Builder |
| 448 | if err := m.serveLogs(opCtx, mh.client, workspace, tailLines, &sb); err != nil { |
| 449 | return "", err |
| 450 | } |
| 451 | if !m.isCurrent(hostID, mh) { |
| 452 | return "", fmt.Errorf("host %q connection was replaced", hostID) |
| 453 | } |
| 454 | return sb.String(), nil |
| 455 | } |
| 456 | |
| 457 | // credentialProxySetup prepares local-proxy credential mode for one |
| 458 | // workspace: starts the desktop key holder, registers every tracked workspace |
| 459 | // against the desktop's current default model, and opens the reverse tunnel the |
| 460 | // remote serve will call through. The returned options are ready to hand to |
| 461 | // the bootstrap (BaseURL points at the tunnel's remote loopback port). |
| 462 | func (m *desktopRemoteManager) credentialProxySetup(c desktopSSHClient, hostID, workspace string) (*bootstrap.CredentialProxyOptions, error) { |
| 463 | app, ok := m.sink.(*App) |
| 464 | if !ok || app == nil { |
| 465 | return nil, fmt.Errorf("credential proxy: app unavailable") |
| 466 | } |
| 467 | info, err := m.registerTrackedCredentialRoutes(app, hostID, workspace) |
| 468 | if err != nil { |
| 469 | return nil, err |
| 470 | } |
| 471 | remotePort, err := ensureCredentialProxyForward(c, hostID, info.port) |
| 472 | if err != nil { |
| 473 | return nil, fmt.Errorf("credential proxy: reverse tunnel: %w", err) |
| 474 | } |
| 475 | return credentialProxyBootstrapOptions(workspace, remotePort, info), nil |
| 476 | } |
| 477 | |
| 478 | func (m *desktopRemoteManager) registerTrackedCredentialRoutes(app *App, hostID, workspace string) (credentialProxyRouteInfo, error) { |
| 479 | workspaces := m.trackedCredentialWorkspaces(hostID, workspace) |
| 480 | var info credentialProxyRouteInfo |
| 481 | for index, trackedWorkspace := range workspaces { |
| 482 | registered, err := app.registerCredentialProxyRoute(hostID, trackedWorkspace) |
| 483 | if err != nil { |
| 484 | return credentialProxyRouteInfo{}, err |
| 485 | } |
| 486 | if index == 0 { |
| 487 | info = registered |
| 488 | } |
| 489 | } |
| 490 | return info, nil |
| 491 | } |
| 492 | |
| 493 | func (m *desktopRemoteManager) trackedCredentialWorkspaces(hostID, workspace string) []string { |
| 494 | workspaces := []string{workspace} |
| 495 | m.mu.Lock() |
| 496 | if managed := m.hosts[hostID]; managed != nil { |
| 497 | peers := make([]string, 0, len(managed.serves)) |
| 498 | for peer := range managed.serves { |
| 499 | if peer != workspace { |
| 500 | peers = append(peers, peer) |
| 501 | } |
| 502 | } |
| 503 | sort.Strings(peers) |
| 504 | workspaces = append(workspaces, peers...) |
| 505 | } |
| 506 | m.mu.Unlock() |
| 507 | return workspaces |
| 508 | } |
| 509 | |
| 510 | // ensureCredentialProxyForward opens (idempotently) the reverse tunnel: the |
| 511 | // REMOTE binds an ephemeral loopback port (avoids conflicts with stale |
| 512 | // listeners from half-dead sessions) and forwards back through SSH to the |
| 513 | // desktop proxy. Returns the actually bound remote port. |
| 514 | func ensureCredentialProxyForward(c desktopSSHClient, hostID string, desktopPort int) (int, error) { |
| 515 | name := "cred-proxy:" + hostID |
| 516 | target := fmt.Sprintf("127.0.0.1:%d", desktopPort) |
| 517 | for _, f := range c.Forwards().List() { |
| 518 | if f.Spec.Name == name && f.Spec.TargetAddr == target && f.Up { |
| 519 | if port, ok := portOfAddr(f.BoundAddr); ok { |
| 520 | return port, nil |
| 521 | } |
| 522 | } |
| 523 | } |
| 524 | bound, err := c.Forwards().Replace(forward.Spec{ |
| 525 | Name: name, |
| 526 | Direction: forward.Remote, |
| 527 | BindAddr: "127.0.0.1:0", |
| 528 | TargetAddr: target, |
| 529 | }) |
| 530 | if err != nil { |
| 531 | return 0, err |
| 532 | } |
| 533 | port, ok := portOfAddr(bound) |
| 534 | if !ok { |
| 535 | return 0, fmt.Errorf("reverse tunnel bound unexpected address %q", bound) |
| 536 | } |
| 537 | return port, nil |
| 538 | } |
| 539 | |
| 540 | // credentialForwardPort reports the remote-side port of the host's reverse |
| 541 | // credential forward, when one is up. This is the port the remote serve's |
| 542 | // provider config must point at. |
| 543 | func credentialForwardPort(c desktopSSHClient, hostID string) (int, bool) { |
| 544 | name := "cred-proxy:" + hostID |
| 545 | for _, f := range c.Forwards().List() { |
| 546 | if f.Spec.Name == name && f.Up { |
| 547 | if port, ok := portOfAddr(f.BoundAddr); ok { |
| 548 | return port, true |
| 549 | } |
| 550 | } |
| 551 | } |
| 552 | return 0, false |
| 553 | } |
| 554 | |
| 555 | // probeReverseTunnel verifies the reverse credential channel end to end. The |
| 556 | // desktop cannot dial the remote-side loopback listener itself, so the probe |
| 557 | // rides the same SSH connection as a direct-tcpip channel — the remote sshd |
| 558 | // connects to its own 127.0.0.1:<port>, which forwards back through the |
| 559 | // tunnel to the desktop credential proxy. Any HTTP response to /healthz |
| 560 | // proves the whole chain; a refused dial or timeout means the channel the |
| 561 | // serve depends on is dead. |
| 562 | func probeReverseTunnel(c desktopSSHClient, port int) error { |
| 563 | type sshDialer interface{ SSH() (*ssh.Client, error) } |
| 564 | d, ok := c.(sshDialer) |
| 565 | if !ok { |
| 566 | return errors.New("reverse probe: ssh client does not expose a raw connection") |
| 567 | } |
| 568 | cl, err := d.SSH() |
| 569 | if err != nil { |
| 570 | return fmt.Errorf("reverse probe: no ssh connection: %w", err) |
| 571 | } |
| 572 | conn, err := cl.Dial("tcp", fmt.Sprintf("127.0.0.1:%d", port)) |
| 573 | if err != nil { |
| 574 | return fmt.Errorf("reverse probe: remote dial 127.0.0.1:%d: %w", port, err) |
| 575 | } |
| 576 | defer conn.Close() |
| 577 | _ = conn.SetDeadline(time.Now().Add(5 * time.Second)) |
| 578 | if _, err := conn.Write([]byte("GET /healthz HTTP/1.1\r\nHost: 127.0.0.1\r\nConnection: close\r\n\r\n")); err != nil { |
| 579 | return fmt.Errorf("reverse probe: write: %w", err) |
| 580 | } |
| 581 | statusLine, err := bufio.NewReader(conn).ReadString('\n') |
| 582 | if err != nil { |
| 583 | return fmt.Errorf("reverse probe: read: %w", err) |
| 584 | } |
| 585 | if !strings.Contains(statusLine, "HTTP/") { |
| 586 | return fmt.Errorf("reverse probe: non-HTTP response %q", strings.TrimSpace(statusLine)) |
| 587 | } |
| 588 | return nil |
| 589 | } |
| 590 | |
| 591 | // reloadServeProviders asks every running serve on the host to rebuild its |
| 592 | // providers: POST /providers/reload rebinds the controller with the CURRENT |
| 593 | // model, re-reading the healed config. Needed after a reverse credential |
| 594 | // tunnel rebind — a reused serve otherwise keeps dialing the dead old port. |
| 595 | // workspace+extraBase+extraToken target the serve this ensure round just |
| 596 | // produced (the registry may not list it yet); registry entries cover the |
| 597 | // host's other workspaces, which share the same healed config. Returns false |
| 598 | // when any serve could not reload (busy turn, or a serve too old to know the |
| 599 | // endpoint): callers keep their heal gate closed so the next ensure retries. |
| 600 | func (m *desktopRemoteManager) reloadServeProviders(ctx context.Context, generation *managedHost, hostID, workspace, extraBase, extraToken string) bool { |
| 601 | type target struct{ base, token string } |
| 602 | m.mu.Lock() |
| 603 | mh := m.hosts[hostID] |
| 604 | if mh == nil || mh != generation { |
| 605 | m.mu.Unlock() |
| 606 | return false |
| 607 | } |
| 608 | targets := make(map[string]target, len(mh.serves)+1) |
| 609 | for ws, e := range mh.serves { |
| 610 | if e != nil && e.view.LocalURL != "" && e.token != "" { |
| 611 | targets[ws] = target{e.view.LocalURL, e.token} |
| 612 | } |
| 613 | } |
| 614 | m.mu.Unlock() |
| 615 | if extraBase != "" && extraToken != "" && workspace != "" { |
| 616 | // The just-ensured serve takes precedence under its REAL workspace |
| 617 | // key; drop any registry entry pointing at the same base so it is |
| 618 | // not reloaded twice. |
| 619 | targets[workspace] = target{extraBase, extraToken} |
| 620 | for ws, t := range targets { |
| 621 | if ws != workspace && t.base == extraBase { |
| 622 | delete(targets, ws) |
| 623 | } |
| 624 | } |
| 625 | } |
| 626 | allOK := true |
| 627 | for ws, t := range targets { |
| 628 | jar, err := cookiejar.New(nil) |
| 629 | if err != nil { |
| 630 | allOK = false |
| 631 | continue |
| 632 | } |
| 633 | client := &http.Client{Jar: jar} |
| 634 | callCtx, cancel := context.WithTimeout(ctx, remoteProviderReloadTimeout) |
| 635 | err = serveHandshake(callCtx, client, t.base, t.token) |
| 636 | if err == nil { |
| 637 | err = servePost(callCtx, client, serveURL(t.base, "/providers/reload"), nil) |
| 638 | } |
| 639 | cancel() |
| 640 | if err != nil && strings.Contains(err.Error(), "status 409") { |
| 641 | err = m.cancelThenReload(ctx, client, t.base, t.token) |
| 642 | } |
| 643 | if err != nil { |
| 644 | log.Printf("[remote] reloadServeProviders: FAILED host=%s ws=%s err=%v", hostID, ws, err) |
| 645 | allOK = false |
| 646 | // Legacy serves lack this route; replace them asynchronously because |
| 647 | // EnsureServer currently holds serveMu. |
| 648 | if (strings.Contains(err.Error(), "status 404") || strings.Contains(err.Error(), "status 405")) && m.markCredFallback(hostID, ws) { |
| 649 | log.Printf("[remote] reloadServeProviders: legacy serve -> replacing host=%s ws=%s", hostID, ws) |
| 650 | go func(hostID, ws string) { |
| 651 | if err := m.StopServer(hostID, ws); err != nil { |
| 652 | log.Printf("[remote] reloadServeProviders: stop legacy serve failed host=%s ws=%s err=%v", hostID, ws, err) |
| 653 | } |
| 654 | if _, _, err := m.EnsureServer(context.Background(), hostID, ws); err != nil { |
| 655 | // EnsureServer errors can carry credential-configuration context. |
| 656 | // Keep logs useful for correlation without persisting that detail. |
| 657 | log.Printf("[remote] reloadServeProviders: restart legacy serve failed host=%s ws=%s", hostID, ws) |
| 658 | } |
| 659 | }(hostID, ws) |
| 660 | } |
| 661 | continue |
| 662 | } |
| 663 | } |
| 664 | return allOK |
| 665 | } |
| 666 | |
| 667 | // cancelThenReload breaks the credential-heal deadlock: after tunnel drift a |
| 668 | // busy turn is already doomed against the stale port, so cancel it and retry |
| 669 | // the provider rebuild with bounded backoff. |
| 670 | func (m *desktopRemoteManager) cancelThenReload(ctx context.Context, client *http.Client, base, token string) error { |
| 671 | log.Printf("[remote] reloadServeProviders: busy turn blocks reload -> canceling turn base=%s", base) |
| 672 | cancelCtx, cancel := context.WithTimeout(ctx, 5*time.Second) |
| 673 | cancelErr := servePost(cancelCtx, client, serveURL(base, "/cancel"), nil) |
| 674 | cancel() |
| 675 | if cancelErr != nil { |
| 676 | log.Printf("[remote] reloadServeProviders: cancel busy turn FAILED err=%v", cancelErr) |
| 677 | } |
| 678 | jobsCtx, jobsCancel := context.WithTimeout(ctx, 5*time.Second) |
| 679 | jobsErr := cancelServeBackgroundJobs(jobsCtx, client, base) |
| 680 | jobsCancel() |
| 681 | if jobsErr != nil { |
| 682 | log.Printf("[remote] reloadServeProviders: cancel background jobs FAILED err=%v", jobsErr) |
| 683 | } |
| 684 | var err error |
| 685 | for attempt := 1; attempt <= 3; attempt++ { |
| 686 | timer := time.NewTimer(time.Duration(attempt) * time.Second) |
| 687 | select { |
| 688 | case <-ctx.Done(): |
| 689 | if !timer.Stop() { |
| 690 | <-timer.C |
| 691 | } |
| 692 | return ctx.Err() |
| 693 | case <-timer.C: |
| 694 | } |
| 695 | callCtx, callCancel := context.WithTimeout(ctx, remoteProviderReloadTimeout) |
| 696 | err = serveHandshake(callCtx, client, base, token) |
| 697 | if err == nil { |
| 698 | err = servePost(callCtx, client, serveURL(base, "/providers/reload"), nil) |
| 699 | } |
| 700 | callCancel() |
| 701 | if err == nil || !strings.Contains(err.Error(), "status 409") { |
| 702 | return err |
| 703 | } |
| 704 | } |
| 705 | return err |
| 706 | } |
| 707 | |
| 708 | func cancelServeBackgroundJobs(ctx context.Context, client *http.Client, base string) error { |
| 709 | status, err := serveGet(ctx, client, serveURL(base, "/status?runtime=1")) |
| 710 | if err != nil { |
| 711 | return err |
| 712 | } |
| 713 | var payload struct { |
| 714 | Jobs []struct { |
| 715 | ID string `json:"id"` |
| 716 | } `json:"jobs"` |
| 717 | } |
| 718 | if err := json.Unmarshal(status, &payload); err != nil { |
| 719 | return fmt.Errorf("decode serve jobs: %w", err) |
| 720 | } |
| 721 | ids := make([]string, 0, len(payload.Jobs)) |
| 722 | for _, job := range payload.Jobs { |
| 723 | if id := strings.TrimSpace(job.ID); id != "" { |
| 724 | ids = append(ids, id) |
| 725 | } |
| 726 | } |
| 727 | if len(ids) == 0 { |
| 728 | return nil |
| 729 | } |
| 730 | body, err := json.Marshal(map[string]any{"ids": ids}) |
| 731 | if err != nil { |
| 732 | return err |
| 733 | } |
| 734 | return servePost(ctx, client, serveURL(base, "/jobs/cancel"), body) |
| 735 | } |
| 736 | |
| 737 | // markCredFallback rate-limits the legacy-serve replacement to at most one |
| 738 | // attempt every two minutes per host and workspace. |
| 739 | func (m *desktopRemoteManager) markCredFallback(hostID, workspace string) bool { |
| 740 | m.mu.Lock() |
| 741 | defer m.mu.Unlock() |
| 742 | mh := m.hosts[hostID] |
| 743 | if mh == nil { |
| 744 | return false |
| 745 | } |
| 746 | now := time.Now().Unix() |
| 747 | if mh.credFallbackAt == nil { |
| 748 | mh.credFallbackAt = map[string]int64{} |
| 749 | } |
| 750 | last := mh.credFallbackAt[workspace] |
| 751 | if last != 0 && now-last < 120 { |
| 752 | return false |
| 753 | } |
| 754 | mh.credFallbackAt[workspace] = now |
| 755 | return true |
| 756 | } |
| 757 | |
| 758 | func portOfAddr(addr string) (int, bool) { |
| 759 | _, portStr, err := net.SplitHostPort(addr) |
| 760 | if err != nil { |
| 761 | return 0, false |
| 762 | } |
| 763 | port, err := strconv.Atoi(portStr) |
| 764 | if err != nil || port <= 0 { |
| 765 | return 0, false |
| 766 | } |
| 767 | return port, true |
| 768 | } |
| 769 |