sync-coordinator.ts

1// Vault sync — Phase 2 manifest-diff coordinator.2//3// After reconnect Noise KK is open and PAIR_IDENTITY has been exchanged4// (`shared/sync-transport.ts`), this coordinator drives the manifest exchange:5//6//   ->  SYNC_HELLO { vaultId, deviceId, deviceManifestVersion? }7//   <-  SYNC_HELLO { vaultId, deviceId, deviceManifestVersion? }8//                                                (mismatch → SYNC_ERROR)9//   <-> SYNC_MANIFEST { entries, tombstones }  (each side sends one)10//   ->  SYNC_REQUEST { entryIds }              (entries newer on the peer)11//   <-  SYNC_ENTRIES { entries }12//   <-  SYNC_REQUEST { entryIds }              (entries newer on us)13//   ->  SYNC_ENTRIES { entries }14//   <-> SYNC_TOMBSTONES { tombstones }15//   <-> SYNC_DEVICE_MANIFEST                    (also runs on equal versions)16//   <-> SYNC_DONE17//18// Entries are opaque base64 ciphertext — the coordinator never decrypts.19// Last-write-wins at the entry level by mtime; tombstones override updates20// with a later deletion timestamp.2122import {23  DEVICE_MANIFEST_VERSION,24  SYNC_PROTOCOL_VERSION,25  type SyncMessage,26  type SignedDeviceManifest,27} from "../shared/sync-types";28import type { SyncSession } from "../shared/sync-transport";29import { SYNC_LIMITS, validateSyncManifestCounts } from "../shared/sync-limits";30import { isConnectionClosedError } from "../shared/sync-outcome";3132const SESSION_TIMEOUT_MS = 30_000;3334type SyncStage =35  | "preparing_hello"36  | "waiting_for_device_hello"37  | "building_local_manifest"38  | "waiting_for_device_manifest"39  | "waiting_for_device_request"40  | "building_requested_entries"41  | "waiting_for_device_entries"42  | "waiting_for_device_tombstones"43  | "building_device_manifest"44  | "waiting_for_device_graph"45  | "waiting_for_device_done";4647/**48 * Vault-side adapter — the coordinator's dependency on the vault layer.49 * The vault holds the canonical entry/tombstone tables; the coordinator50 * pulls a manifest, requests entries by id, applies remote entries, and51 * applies remote tombstones.52 */53export interface SyncVaultAdapter {54  computeVaultId(): Promise<string>;55  getOurDeviceId(): Promise<string>;56  /** Returns the vault's monotonic version (`peach_last_modified_at`). */57  getLocalVersion(): Promise<number>;58  /** Returns the per-entry mtimes for our active entries. */59  buildEntryManifest(): Promise<Record<string, number>>;60  /** Returns the per-entry deletion timestamps for our tombstones. */61  buildTombstoneManifest(): Promise<Record<string, number>>;62  /** Returns base64 ciphertext for each requested entry id. */63  getEncryptedEntries(entryIds: string[]): Promise<Record<string, string>>;64  /** Apply incoming encrypted entries; vault decides per-entry overwrite. */65  applyEntries(66    entries: Record<string, string>,67    remoteManifest?: Record<string, number>,68  ): Promise<void>;69  /** Apply incoming tombstones; vault retains the latest deletion ts. */70  applyTombstones(tombstones: Record<string, number>): Promise<void>;71  /**72   * Optional — v2 brief §4.3 device manifest exchange. The extension (as73   * the trust anchor for the device graph) returns a signed manifest74   * enumerating every paired device it knows about; the peer verifies it75   * against the extension's known Ed25519 signing pubkey and upserts any76   * newly-vouched devices into its own registry.77   *78   * When this method is undefined, the coordinator skips the manifest79   * step entirely. Both ends must opt in for the exchange to occur.80   */81  buildOutgoingDeviceManifest?(): Promise<SignedDeviceManifest | null>;82  /**83   * Handle a manifest received from the peer. The transport-authenticated84   * Noise static key is supplied separately so a self-asserted issuer id85   * cannot select a different paired-device verification key.86   */87  applyIncomingDeviceManifest?(88    manifest: SignedDeviceManifest,89    authenticatedRemoteStaticPublicKey: Uint8Array,90  ): Promise<void>;91}9293export interface SyncOutcome {94  status: "completed" | "vault_mismatch" | "errored" | "timeout";95  entriesReceived: number;96  entriesSent: number;97  tombstonesReceived: number;98  tombstonesSent: number;99  error?: string;100}101102/**103 * Run one sync round as the lower-key initiator. Device type does not decide104 * the role; the higher-key peer runs [runSyncResponder].105 */106export async function runSyncInitiator(107  session: SyncSession,108  vault: SyncVaultAdapter,109): Promise<SyncOutcome> {110  const result: SyncOutcome = {111    status: "errored",112    entriesReceived: 0,113    entriesSent: 0,114    tombstonesReceived: 0,115    tombstonesSent: 0,116  };117118  const timer = scheduleTimeout(SESSION_TIMEOUT_MS);119  // Privacy-safe progress only: stages contain no device identifiers, relay120  // capabilities, keys, entry ids/counts, or vault content. Keeping the last121  // stage turns an otherwise opaque 30-second timeout into an actionable122  // cross-device diagnostic.123  let stage: SyncStage = "preparing_hello";124  try {125    const ourVaultId = await vault.computeVaultId();126    const ourDeviceId = await vault.getOurDeviceId();127    const localVersion = await vault.getLocalVersion();128    const localTime = Date.now();129    const supportsDeviceManifest =130      vault.buildOutgoingDeviceManifest !== undefined &&131      vault.applyIncomingDeviceManifest !== undefined;132133    await sendMessage(session, {134      type: "SYNC_HELLO",135      vaultId: ourVaultId,136      deviceId: ourDeviceId,137      version: localVersion,138      deviceTime: localTime,139      protocolVersion: SYNC_PROTOCOL_VERSION,140      deviceManifestVersion: supportsDeviceManifest141        ? DEVICE_MANIFEST_VERSION142        : undefined,143    });144    stage = "waiting_for_device_hello";145    const helloIn = await receiveMessage(session, timer);146    if (helloIn.type !== "SYNC_HELLO") {147      return errorOutcome(result, `expected SYNC_HELLO, got ${helloIn.type}`);148    }149    if ((helloIn.protocolVersion ?? 1) < SYNC_PROTOCOL_VERSION) {150      await sendMessage(session, { type: "SYNC_ERROR", error: "PEER_PROTOCOL_OUTDATED" });151      return errorOutcome(result, "PEER_PROTOCOL_OUTDATED");152    }153    if (helloIn.vaultId !== ourVaultId) {154      await sendMessage(session, { type: "SYNC_ERROR", error: "vault_id_mismatch" });155      result.status = "vault_mismatch";156      return result;157    }158159    if (Math.abs(localTime - (helloIn.deviceTime ?? 0)) > 120_000) {160      console.warn(`[sync] clock skew detected: local=${localTime} remote=${helloIn.deviceTime ?? "unknown"}`);161    }162163    const exchangeDeviceManifest =164      supportsDeviceManifest &&165      helloIn.deviceManifestVersion === DEVICE_MANIFEST_VERSION;166167    stage = "building_local_manifest";168    const ourEntries = await vault.buildEntryManifest();169    const ourTombstones = await vault.buildTombstoneManifest();170    await sendMessage(session, {171      type: "SYNC_MANIFEST",172      entries: ourEntries,173      tombstones: ourTombstones,174    });175    stage = "waiting_for_device_manifest";176    const manifestIn = await receiveMessage(session, timer);177    if (manifestIn.type !== "SYNC_MANIFEST") {178      return errorOutcome(result, `expected SYNC_MANIFEST, got ${manifestIn.type}`);179    }180    validateSyncManifestCounts(manifestIn.entries, manifestIn.tombstones);181182    const wantFromPeer = diffNewer(manifestIn.entries, ourEntries);183    await sendMessage(session, { type: "SYNC_REQUEST", entryIds: wantFromPeer });184185    stage = "waiting_for_device_request";186    const requestIn = await receiveMessage(session, timer);187    if (requestIn.type !== "SYNC_REQUEST") {188      return errorOutcome(result, `expected SYNC_REQUEST, got ${requestIn.type}`);189    }190    if (requestIn.entryIds.length > SYNC_LIMITS.MAX_REQUEST_ENTRY_COUNT) {191      return errorOutcome(result, "sync request entry count exceeds limit");192    }193    // Sanity: the peer's request must be a subset of what its manifest194    // declared as newer. A malformed/over-eager request is a protocol195    // violation but not a data-safety problem — we just send what they ask.196197    stage = "building_requested_entries";198    const ourPayload = await vault.getEncryptedEntries(requestIn.entryIds);199    await sendMessage(session, { type: "SYNC_ENTRIES", entries: ourPayload });200    result.entriesSent = Object.keys(ourPayload).length;201202    if (wantFromPeer.length > 0) {203      stage = "waiting_for_device_entries";204      const entriesIn = await receiveMessage(session, timer);205      if (entriesIn.type !== "SYNC_ENTRIES") {206        return errorOutcome(result, `expected SYNC_ENTRIES, got ${entriesIn.type}`);207      }208      await vault.applyEntries(entriesIn.entries, manifestIn.entries);209      result.entriesReceived = Object.keys(entriesIn.entries).length;210    }211212    // Tombstones: each side ships the tombstones that the other does not213    // yet have a newer record for.214    const newerTombstonesForPeer: Record<string, number> = {};215    for (const [id, ts] of Object.entries(ourTombstones)) {216      const peerTs = manifestIn.tombstones[id] ?? 0;217      if (ts > peerTs) newerTombstonesForPeer[id] = ts;218    }219    await sendMessage(session, { type: "SYNC_TOMBSTONES", tombstones: newerTombstonesForPeer });220    result.tombstonesSent = Object.keys(newerTombstonesForPeer).length;221222    stage = "waiting_for_device_tombstones";223    const tombstonesIn = await receiveMessage(session, timer);224    if (tombstonesIn.type !== "SYNC_TOMBSTONES") {225      return errorOutcome(result, `expected SYNC_TOMBSTONES, got ${tombstonesIn.type}`);226    }227    await vault.applyTombstones(tombstonesIn.tombstones);228    result.tombstonesReceived = Object.keys(tombstonesIn.tombstones).length;229230    // v2 brief §5.6 / §4.3 — device manifest exchange. Optional: both ends231    // must expose the adapter methods. Disabling on either side cleanly232    // bypasses the step. The exchange is symmetric on the wire; the233    // production adapter accepts only a directly paired issuer whose stored234    // long-term key matches the Noise-authenticated remote static key.235    if (exchangeDeviceManifest) {236      stage = "building_device_manifest";237      const ours = await vault.buildOutgoingDeviceManifest!();238      await sendMessage(session, { type: "SYNC_DEVICE_MANIFEST", manifest: ours });239      stage = "waiting_for_device_graph";240      const manifestIn = await receiveMessage(session, timer);241      if (manifestIn.type !== "SYNC_DEVICE_MANIFEST") {242        return errorOutcome(result, `expected SYNC_DEVICE_MANIFEST, got ${manifestIn.type}`);243      }244      if (manifestIn.manifest) {245        await vault.applyIncomingDeviceManifest!(246          manifestIn.manifest,247          session.remoteStaticPublicKey,248        );249      }250    }251252    await sendMessage(session, {253      type: "SYNC_DONE",254      entriesReceived: result.entriesReceived,255      entriesSent: result.entriesSent,256    });257    stage = "waiting_for_device_done";258    const closing = await awaitClosingDone(session, timer);259    if (closing.error) return errorOutcome(result, closing.error);260    result.status = "completed";261    return result;262  } catch (err) {263    if ((err as Error).message === "sync_timeout") {264      result.status = "timeout";265      result.error = `sync_timeout:${stage}`;266      return result;267    }268    return errorOutcome(result, (err as Error).message);269  } finally {270    // Session lifetime is owned by the runner (sync-runner.ts), which271    // closes it in its own finally block. The coordinator must not close it272    // here — the runner still needs to record the outcome after we return.273    timer.cancel();274  }275}276277/**278 * Exact complement of [runSyncInitiator] for the higher-key peer holding a279 * relay waiter room. Keeping this in the browser lets two extensions form a280 * direct mesh edge; the device that introduced them is not in the data path.281 */282export async function runSyncResponder(283  session: SyncSession,284  vault: SyncVaultAdapter,285): Promise<SyncOutcome> {286  const result: SyncOutcome = {287    status: "errored",288    entriesReceived: 0,289    entriesSent: 0,290    tombstonesReceived: 0,291    tombstonesSent: 0,292  };293  const timer = scheduleTimeout(SESSION_TIMEOUT_MS);294  let stage: SyncStage = "waiting_for_device_hello";295  try {296    const ourVaultId = await vault.computeVaultId();297    const ourDeviceId = await vault.getOurDeviceId();298    const localVersion = await vault.getLocalVersion();299    const localTime = Date.now();300    const supportsDeviceManifest =301      vault.buildOutgoingDeviceManifest !== undefined &&302      vault.applyIncomingDeviceManifest !== undefined;303304    const helloIn = await receiveMessage(session, timer);305    if (helloIn.type !== "SYNC_HELLO") {306      return errorOutcome(result, `expected SYNC_HELLO, got ${helloIn.type}`);307    }308    if ((helloIn.protocolVersion ?? 1) < SYNC_PROTOCOL_VERSION) {309      await sendMessage(session, { type: "SYNC_ERROR", error: "PEER_PROTOCOL_OUTDATED" });310      return errorOutcome(result, "PEER_PROTOCOL_OUTDATED");311    }312    await sendMessage(session, {313      type: "SYNC_HELLO",314      vaultId: ourVaultId,315      deviceId: ourDeviceId,316      version: localVersion,317      deviceTime: localTime,318      protocolVersion: SYNC_PROTOCOL_VERSION,319      deviceManifestVersion: supportsDeviceManifest320        ? DEVICE_MANIFEST_VERSION321        : undefined,322    });323    if (helloIn.vaultId !== ourVaultId) {324      result.status = "vault_mismatch";325      return result;326    }327    if (Math.abs(localTime - (helloIn.deviceTime ?? 0)) > 120_000) {328      console.warn(329        `[sync] clock skew detected: local=${localTime} remote=${helloIn.deviceTime ?? "unknown"}`,330      );331    }332333    const exchangeDeviceManifest =334      supportsDeviceManifest &&335      helloIn.deviceManifestVersion === DEVICE_MANIFEST_VERSION;336337    const ourEntries = await vault.buildEntryManifest();338    const ourTombstones = await vault.buildTombstoneManifest();339    stage = "waiting_for_device_manifest";340    const manifestIn = await receiveMessage(session, timer);341    if (manifestIn.type !== "SYNC_MANIFEST") {342      return errorOutcome(result, `expected SYNC_MANIFEST, got ${manifestIn.type}`);343    }344    validateSyncManifestCounts(manifestIn.entries, manifestIn.tombstones);345    await sendMessage(session, {346      type: "SYNC_MANIFEST",347      entries: ourEntries,348      tombstones: ourTombstones,349    });350351    stage = "waiting_for_device_request";352    const requestIn = await receiveMessage(session, timer);353    if (requestIn.type !== "SYNC_REQUEST") {354      return errorOutcome(result, `expected SYNC_REQUEST, got ${requestIn.type}`);355    }356    if (requestIn.entryIds.length > SYNC_LIMITS.MAX_REQUEST_ENTRY_COUNT) {357      return errorOutcome(result, "sync request entry count exceeds limit");358    }359    const wantFromPeer = diffNewer(manifestIn.entries, ourEntries);360    await sendMessage(session, { type: "SYNC_REQUEST", entryIds: wantFromPeer });361362    stage = "waiting_for_device_entries";363    const entriesIn = await receiveMessage(session, timer);364    if (entriesIn.type !== "SYNC_ENTRIES") {365      return errorOutcome(result, `expected SYNC_ENTRIES, got ${entriesIn.type}`);366    }367    if (wantFromPeer.length > 0) {368      await vault.applyEntries(entriesIn.entries, manifestIn.entries);369      result.entriesReceived = Object.keys(entriesIn.entries).length;370    }371    if (requestIn.entryIds.length > 0) {372      const ourPayload = await vault.getEncryptedEntries(requestIn.entryIds);373      await sendMessage(session, { type: "SYNC_ENTRIES", entries: ourPayload });374      result.entriesSent = Object.keys(ourPayload).length;375    }376377    stage = "waiting_for_device_tombstones";378    const tombstonesIn = await receiveMessage(session, timer);379    if (tombstonesIn.type !== "SYNC_TOMBSTONES") {380      return errorOutcome(result, `expected SYNC_TOMBSTONES, got ${tombstonesIn.type}`);381    }382    await vault.applyTombstones(tombstonesIn.tombstones);383    result.tombstonesReceived = Object.keys(tombstonesIn.tombstones).length;384385    const newerTombstonesForPeer: Record<string, number> = {};386    for (const [id, timestamp] of Object.entries(ourTombstones)) {387      if (timestamp > (manifestIn.tombstones[id] ?? 0)) {388        newerTombstonesForPeer[id] = timestamp;389      }390    }391    await sendMessage(session, {392      type: "SYNC_TOMBSTONES",393      tombstones: newerTombstonesForPeer,394    });395    result.tombstonesSent = Object.keys(newerTombstonesForPeer).length;396397    if (exchangeDeviceManifest) {398      stage = "waiting_for_device_graph";399      const graphIn = await receiveMessage(session, timer);400      if (graphIn.type !== "SYNC_DEVICE_MANIFEST") {401        return errorOutcome(402          result,403          `expected SYNC_DEVICE_MANIFEST, got ${graphIn.type}`,404        );405      }406      const ours = await vault.buildOutgoingDeviceManifest!();407      await sendMessage(session, {408        type: "SYNC_DEVICE_MANIFEST",409        manifest: ours,410      });411      if (graphIn.manifest) {412        await vault.applyIncomingDeviceManifest!(413          graphIn.manifest,414          session.remoteStaticPublicKey,415        );416      }417    }418419    stage = "waiting_for_device_done";420    const done = await receiveMessage(session, timer);421    if (done.type !== "SYNC_DONE") {422      return errorOutcome(result, `expected SYNC_DONE, got ${done.type}`);423    }424    await sendMessage(session, {425      type: "SYNC_DONE",426      entriesReceived: result.entriesReceived,427      entriesSent: result.entriesSent,428    });429    result.status = "completed";430    return result;431  } catch (error) {432    if ((error as Error).message === "sync_timeout") {433      result.status = "timeout";434      result.error = `sync_timeout:${stage}`;435      return result;436    }437    return errorOutcome(result, (error as Error).message);438  } finally {439    timer.cancel();440  }441}442443/**444 * Wait for the peer's closing SYNC_DONE echo, treating a hang-up as success.445 *446 * By the time we send our own SYNC_DONE, every entry and tombstone has already447 * been exchanged and applied on both sides — the echo is a courtesy, not a448 * commit. Android closes the socket as soon as it has our DONE, so the parked449 * receive() rejects with "WebSocket closed"; counting that as `run_failed`450 * reported a red indicator for syncs that had fully succeeded and whose data451 * had demonstrably landed on the peer.452 *453 * A wrong *message* here is still a real protocol error — only a closed454 * transport is forgiven.455 */456async function awaitClosingDone(457  session: SyncSession,458  timer: Timer,459): Promise<{ error?: string }> {460  try {461    const done = await receiveMessage(session, timer);462    if (done.type !== "SYNC_DONE") {463      return { error: `expected SYNC_DONE, got ${done.type}` };464    }465    return {};466  } catch (err) {467    if ((err as Error).message === "sync_timeout") throw err;468    if (isConnectionClosedError(err)) return {};469    return { error: (err as Error).message };470  }471}472473function diffNewer(remote: Record<string, number>, local: Record<string, number>): string[] {474  const want: string[] = [];475  for (const [id, ts] of Object.entries(remote)) {476    const ours = local[id];477    if (ours === undefined || ts > ours) want.push(id);478  }479  return want;480}481482interface Timer {483  signal: AbortSignal;484  cancel(): void;485}486487function scheduleTimeout(ms: number): Timer {488  const ctrl = new AbortController();489  const t = setTimeout(() => ctrl.abort(new Error("sync_timeout")), ms);490  return {491    signal: ctrl.signal,492    cancel: () => clearTimeout(t),493  };494}495496async function sendMessage(session: SyncSession, msg: SyncMessage): Promise<void> {497  session.send(new TextEncoder().encode(JSON.stringify(msg)));498}499500async function receiveMessage(session: SyncSession, timer: Timer): Promise<SyncMessage> {501  const pending = session.receive();502  const aborted = new Promise<never>((_, reject) => {503    if (timer.signal.aborted) reject(new Error("sync_timeout"));504    timer.signal.addEventListener("abort", () => reject(new Error("sync_timeout")));505  });506  const bytes = await Promise.race([pending, aborted]);507  // MV-3: decrypted plaintext can exceed the wire frame (and a non-WS508  // transport may not have enforced anything) — re-check before JSON.parse509  // materialises a multi-GB string/object graph.510  if (bytes.length > SYNC_LIMITS.MAX_FRAME_BYTES) {511    throw new Error(`sync frame exceeds max size (${SYNC_LIMITS.MAX_FRAME_BYTES} bytes)`);512  }513  const parsed = JSON.parse(new TextDecoder().decode(bytes));514  if (typeof parsed?.type !== "string") throw new Error("malformed SyncMessage");515  return parsed as SyncMessage;516}517518function errorOutcome(outcome: SyncOutcome, error: string): SyncOutcome {519  outcome.status = "errored";520  outcome.error = error;521  return outcome;522}