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  type SyncMessage,25  type SignedDeviceManifest,26} from "../shared/sync-types";27import type { SyncSession } from "../shared/sync-transport";28import { SYNC_LIMITS } from "../shared/sync-limits";29import { isConnectionClosedError } from "../shared/sync-outcome";3031const SESSION_TIMEOUT_MS = 30_000;3233type SyncStage =34  | "preparing_hello"35  | "waiting_for_device_hello"36  | "building_local_manifest"37  | "waiting_for_device_manifest"38  | "waiting_for_device_request"39  | "building_requested_entries"40  | "waiting_for_device_entries"41  | "waiting_for_device_tombstones"42  | "building_device_manifest"43  | "waiting_for_device_graph"44  | "waiting_for_device_done";4546/**47 * Vault-side adapter — the coordinator's dependency on the vault layer.48 * The vault holds the canonical entry/tombstone tables; the coordinator49 * pulls a manifest, requests entries by id, applies remote entries, and50 * applies remote tombstones.51 */52export interface SyncVaultAdapter {53  computeVaultId(): Promise<string>;54  getOurDeviceId(): Promise<string>;55  /** Returns the vault's monotonic version (`peach_last_modified_at`). */56  getLocalVersion(): Promise<number>;57  /** Returns the per-entry mtimes for our active entries. */58  buildEntryManifest(): Promise<Record<string, number>>;59  /** Returns the per-entry deletion timestamps for our tombstones. */60  buildTombstoneManifest(): Promise<Record<string, number>>;61  /** Returns base64 ciphertext for each requested entry id. */62  getEncryptedEntries(entryIds: string[]): Promise<Record<string, string>>;63  /** Apply incoming encrypted entries; vault decides per-entry overwrite. */64  applyEntries(65    entries: Record<string, string>,66    remoteManifest?: Record<string, number>,67  ): Promise<void>;68  /** Apply incoming tombstones; vault retains the latest deletion ts. */69  applyTombstones(tombstones: Record<string, number>): Promise<void>;70  /**71   * Optional — v2 brief §4.3 device manifest exchange. The extension (as72   * the trust anchor for the device graph) returns a signed manifest73   * enumerating every paired device it knows about; the peer verifies it74   * against the extension's known Ed25519 signing pubkey and upserts any75   * newly-vouched devices into its own registry.76   *77   * When this method is undefined, the coordinator skips the manifest78   * step entirely. Both ends must opt in for the exchange to occur.79   */80  buildOutgoingDeviceManifest?(): Promise<SignedDeviceManifest | null>;81  /**82   * Handle a manifest received from the peer. The transport-authenticated83   * Noise static key is supplied separately so a self-asserted issuer id84   * cannot select a different paired-device verification key.85   */86  applyIncomingDeviceManifest?(87    manifest: SignedDeviceManifest,88    authenticatedRemoteStaticPublicKey: Uint8Array,89  ): Promise<void>;90}9192export interface SyncOutcome {93  status: "completed" | "vault_mismatch" | "errored" | "timeout";94  entriesReceived: number;95  entriesSent: number;96  tombstonesReceived: number;97  tombstonesSent: number;98  error?: string;99}100101/**102 * Run one sync round as the lower-key initiator. Device type does not decide103 * the role; the higher-key peer runs [runSyncResponder].104 */105export async function runSyncInitiator(106  session: SyncSession,107  vault: SyncVaultAdapter,108): Promise<SyncOutcome> {109  const result: SyncOutcome = {110    status: "errored",111    entriesReceived: 0,112    entriesSent: 0,113    tombstonesReceived: 0,114    tombstonesSent: 0,115  };116117  const timer = scheduleTimeout(SESSION_TIMEOUT_MS);118  // Privacy-safe progress only: stages contain no device identifiers, relay119  // capabilities, keys, entry ids/counts, or vault content. Keeping the last120  // stage turns an otherwise opaque 30-second timeout into an actionable121  // cross-device diagnostic.122  let stage: SyncStage = "preparing_hello";123  try {124    const ourVaultId = await vault.computeVaultId();125    const ourDeviceId = await vault.getOurDeviceId();126    const localVersion = await vault.getLocalVersion();127    const localTime = Date.now();128    const supportsDeviceManifest =129      vault.buildOutgoingDeviceManifest !== undefined &&130      vault.applyIncomingDeviceManifest !== undefined;131132    await sendMessage(session, {133      type: "SYNC_HELLO",134      vaultId: ourVaultId,135      deviceId: ourDeviceId,136      version: localVersion,137      deviceTime: localTime,138      deviceManifestVersion: supportsDeviceManifest139        ? DEVICE_MANIFEST_VERSION140        : undefined,141    });142    stage = "waiting_for_device_hello";143    const helloIn = await receiveMessage(session, timer);144    if (helloIn.type !== "SYNC_HELLO") {145      return errorOutcome(result, `expected SYNC_HELLO, got ${helloIn.type}`);146    }147    if (helloIn.vaultId !== ourVaultId) {148      await sendMessage(session, { type: "SYNC_ERROR", error: "vault_id_mismatch" });149      result.status = "vault_mismatch";150      return result;151    }152153    if (Math.abs(localTime - (helloIn.deviceTime ?? 0)) > 120_000) {154      console.warn(`[sync] clock skew detected: local=${localTime} remote=${helloIn.deviceTime ?? "unknown"}`);155    }156157    // Equal entry versions do not imply equal device graphs: paired-device158    // introductions and revocations live outside the vault version counter.159    // Exchange the signed graph when both peers advertise support, while an160    // older peer that omits the capability retains the legacy DONE fast path.161    if (typeof helloIn.version === "number" && helloIn.version === localVersion) {162      if (163        supportsDeviceManifest &&164        helloIn.deviceManifestVersion === DEVICE_MANIFEST_VERSION165      ) {166        stage = "building_device_manifest";167        const ours = await vault.buildOutgoingDeviceManifest!();168        await sendMessage(session, {169          type: "SYNC_DEVICE_MANIFEST",170          manifest: ours,171        });172        stage = "waiting_for_device_graph";173        const graphIn = await receiveMessage(session, timer);174        if (graphIn.type !== "SYNC_DEVICE_MANIFEST") {175          return errorOutcome(176            result,177            `expected SYNC_DEVICE_MANIFEST, got ${graphIn.type}`,178          );179        }180        if (graphIn.manifest) {181          await vault.applyIncomingDeviceManifest!(182            graphIn.manifest,183            session.remoteStaticPublicKey,184          );185        }186      }187      await sendMessage(session, { type: "SYNC_DONE", entriesReceived: 0, entriesSent: 0 });188      stage = "waiting_for_device_done";189      const closing = await awaitClosingDone(session, timer);190      if (closing.error) return errorOutcome(result, closing.error);191      result.status = "completed";192      return result;193    }194195    stage = "building_local_manifest";196    const ourEntries = await vault.buildEntryManifest();197    const ourTombstones = await vault.buildTombstoneManifest();198    await sendMessage(session, {199      type: "SYNC_MANIFEST",200      entries: ourEntries,201      tombstones: ourTombstones,202    });203    stage = "waiting_for_device_manifest";204    const manifestIn = await receiveMessage(session, timer);205    if (manifestIn.type !== "SYNC_MANIFEST") {206      return errorOutcome(result, `expected SYNC_MANIFEST, got ${manifestIn.type}`);207    }208209    const wantFromPeer = diffNewer(manifestIn.entries, ourEntries);210    await sendMessage(session, { type: "SYNC_REQUEST", entryIds: wantFromPeer });211212    stage = "waiting_for_device_request";213    const requestIn = await receiveMessage(session, timer);214    if (requestIn.type !== "SYNC_REQUEST") {215      return errorOutcome(result, `expected SYNC_REQUEST, got ${requestIn.type}`);216    }217    // Sanity: the peer's request must be a subset of what its manifest218    // declared as newer. A malformed/over-eager request is a protocol219    // violation but not a data-safety problem — we just send what they ask.220221    stage = "building_requested_entries";222    const ourPayload = await vault.getEncryptedEntries(requestIn.entryIds);223    await sendMessage(session, { type: "SYNC_ENTRIES", entries: ourPayload });224    result.entriesSent = Object.keys(ourPayload).length;225226    if (wantFromPeer.length > 0) {227      stage = "waiting_for_device_entries";228      const entriesIn = await receiveMessage(session, timer);229      if (entriesIn.type !== "SYNC_ENTRIES") {230        return errorOutcome(result, `expected SYNC_ENTRIES, got ${entriesIn.type}`);231      }232      await vault.applyEntries(entriesIn.entries, manifestIn.entries);233      result.entriesReceived = Object.keys(entriesIn.entries).length;234    }235236    // Tombstones: each side ships the tombstones that the other does not237    // yet have a newer record for.238    const newerTombstonesForPeer: Record<string, number> = {};239    for (const [id, ts] of Object.entries(ourTombstones)) {240      const peerTs = manifestIn.tombstones[id] ?? 0;241      if (ts > peerTs) newerTombstonesForPeer[id] = ts;242    }243    await sendMessage(session, { type: "SYNC_TOMBSTONES", tombstones: newerTombstonesForPeer });244    result.tombstonesSent = Object.keys(newerTombstonesForPeer).length;245246    stage = "waiting_for_device_tombstones";247    const tombstonesIn = await receiveMessage(session, timer);248    if (tombstonesIn.type !== "SYNC_TOMBSTONES") {249      return errorOutcome(result, `expected SYNC_TOMBSTONES, got ${tombstonesIn.type}`);250    }251    await vault.applyTombstones(tombstonesIn.tombstones);252    result.tombstonesReceived = Object.keys(tombstonesIn.tombstones).length;253254    // v2 brief §5.6 / §4.3 — device manifest exchange. Optional: both ends255    // must expose the adapter methods. Disabling on either side cleanly256    // bypasses the step. The exchange is symmetric on the wire; the257    // production adapter accepts only a directly paired issuer whose stored258    // long-term key matches the Noise-authenticated remote static key.259    if (vault.buildOutgoingDeviceManifest && vault.applyIncomingDeviceManifest) {260      stage = "building_device_manifest";261      const ours = await vault.buildOutgoingDeviceManifest();262      await sendMessage(session, { type: "SYNC_DEVICE_MANIFEST", manifest: ours });263      stage = "waiting_for_device_graph";264      const manifestIn = await receiveMessage(session, timer);265      if (manifestIn.type !== "SYNC_DEVICE_MANIFEST") {266        return errorOutcome(result, `expected SYNC_DEVICE_MANIFEST, got ${manifestIn.type}`);267      }268      if (manifestIn.manifest) {269        await vault.applyIncomingDeviceManifest(270          manifestIn.manifest,271          session.remoteStaticPublicKey,272        );273      }274    }275276    await sendMessage(session, {277      type: "SYNC_DONE",278      entriesReceived: result.entriesReceived,279      entriesSent: result.entriesSent,280    });281    stage = "waiting_for_device_done";282    const closing = await awaitClosingDone(session, timer);283    if (closing.error) return errorOutcome(result, closing.error);284    result.status = "completed";285    return result;286  } catch (err) {287    if ((err as Error).message === "sync_timeout") {288      result.status = "timeout";289      result.error = `sync_timeout:${stage}`;290      return result;291    }292    return errorOutcome(result, (err as Error).message);293  } finally {294    // Session lifetime is owned by the runner (sync-runner.ts), which295    // closes it in its own finally block. The coordinator must not close it296    // here — the runner still needs to record the outcome after we return.297    timer.cancel();298  }299}300301/**302 * Exact complement of [runSyncInitiator] for the higher-key peer holding a303 * relay waiter room. Keeping this in the browser lets two extensions form a304 * direct mesh edge; the device that introduced them is not in the data path.305 */306export async function runSyncResponder(307  session: SyncSession,308  vault: SyncVaultAdapter,309): Promise<SyncOutcome> {310  const result: SyncOutcome = {311    status: "errored",312    entriesReceived: 0,313    entriesSent: 0,314    tombstonesReceived: 0,315    tombstonesSent: 0,316  };317  const timer = scheduleTimeout(SESSION_TIMEOUT_MS);318  let stage: SyncStage = "waiting_for_device_hello";319  try {320    const ourVaultId = await vault.computeVaultId();321    const ourDeviceId = await vault.getOurDeviceId();322    const localVersion = await vault.getLocalVersion();323    const localTime = Date.now();324    const supportsDeviceManifest =325      vault.buildOutgoingDeviceManifest !== undefined &&326      vault.applyIncomingDeviceManifest !== undefined;327328    const helloIn = await receiveMessage(session, timer);329    if (helloIn.type !== "SYNC_HELLO") {330      return errorOutcome(result, `expected SYNC_HELLO, got ${helloIn.type}`);331    }332    await sendMessage(session, {333      type: "SYNC_HELLO",334      vaultId: ourVaultId,335      deviceId: ourDeviceId,336      version: localVersion,337      deviceTime: localTime,338      deviceManifestVersion: supportsDeviceManifest339        ? DEVICE_MANIFEST_VERSION340        : undefined,341    });342    if (helloIn.vaultId !== ourVaultId) {343      result.status = "vault_mismatch";344      return result;345    }346    if (Math.abs(localTime - (helloIn.deviceTime ?? 0)) > 120_000) {347      console.warn(348        `[sync] clock skew detected: local=${localTime} remote=${helloIn.deviceTime ?? "unknown"}`,349      );350    }351352    if (helloIn.version === localVersion) {353      if (354        supportsDeviceManifest &&355        helloIn.deviceManifestVersion === DEVICE_MANIFEST_VERSION356      ) {357        stage = "waiting_for_device_graph";358        const graphIn = await receiveMessage(session, timer);359        if (graphIn.type !== "SYNC_DEVICE_MANIFEST") {360          return errorOutcome(361            result,362            `expected SYNC_DEVICE_MANIFEST, got ${graphIn.type}`,363          );364        }365        const ours = await vault.buildOutgoingDeviceManifest!();366        await sendMessage(session, {367          type: "SYNC_DEVICE_MANIFEST",368          manifest: ours,369        });370        if (graphIn.manifest) {371          await vault.applyIncomingDeviceManifest!(372            graphIn.manifest,373            session.remoteStaticPublicKey,374          );375        }376      }377      stage = "waiting_for_device_done";378      const done = await receiveMessage(session, timer);379      if (done.type !== "SYNC_DONE") {380        return errorOutcome(result, `expected SYNC_DONE, got ${done.type}`);381      }382      await sendMessage(session, {383        type: "SYNC_DONE",384        entriesReceived: 0,385        entriesSent: 0,386      });387      result.status = "completed";388      return result;389    }390391    const ourEntries = await vault.buildEntryManifest();392    const ourTombstones = await vault.buildTombstoneManifest();393    stage = "waiting_for_device_manifest";394    const manifestIn = await receiveMessage(session, timer);395    if (manifestIn.type !== "SYNC_MANIFEST") {396      return errorOutcome(result, `expected SYNC_MANIFEST, got ${manifestIn.type}`);397    }398    await sendMessage(session, {399      type: "SYNC_MANIFEST",400      entries: ourEntries,401      tombstones: ourTombstones,402    });403404    stage = "waiting_for_device_request";405    const requestIn = await receiveMessage(session, timer);406    if (requestIn.type !== "SYNC_REQUEST") {407      return errorOutcome(result, `expected SYNC_REQUEST, got ${requestIn.type}`);408    }409    const wantFromPeer = diffNewer(manifestIn.entries, ourEntries);410    await sendMessage(session, { type: "SYNC_REQUEST", entryIds: wantFromPeer });411412    stage = "waiting_for_device_entries";413    const entriesIn = await receiveMessage(session, timer);414    if (entriesIn.type !== "SYNC_ENTRIES") {415      return errorOutcome(result, `expected SYNC_ENTRIES, got ${entriesIn.type}`);416    }417    if (wantFromPeer.length > 0) {418      await vault.applyEntries(entriesIn.entries, manifestIn.entries);419      result.entriesReceived = Object.keys(entriesIn.entries).length;420    }421    if (requestIn.entryIds.length > 0) {422      const ourPayload = await vault.getEncryptedEntries(requestIn.entryIds);423      await sendMessage(session, { type: "SYNC_ENTRIES", entries: ourPayload });424      result.entriesSent = Object.keys(ourPayload).length;425    }426427    stage = "waiting_for_device_tombstones";428    const tombstonesIn = await receiveMessage(session, timer);429    if (tombstonesIn.type !== "SYNC_TOMBSTONES") {430      return errorOutcome(result, `expected SYNC_TOMBSTONES, got ${tombstonesIn.type}`);431    }432    await vault.applyTombstones(tombstonesIn.tombstones);433    result.tombstonesReceived = Object.keys(tombstonesIn.tombstones).length;434435    const newerTombstonesForPeer: Record<string, number> = {};436    for (const [id, timestamp] of Object.entries(ourTombstones)) {437      if (timestamp > (manifestIn.tombstones[id] ?? 0)) {438        newerTombstonesForPeer[id] = timestamp;439      }440    }441    await sendMessage(session, {442      type: "SYNC_TOMBSTONES",443      tombstones: newerTombstonesForPeer,444    });445    result.tombstonesSent = Object.keys(newerTombstonesForPeer).length;446447    if (vault.buildOutgoingDeviceManifest && vault.applyIncomingDeviceManifest) {448      stage = "waiting_for_device_graph";449      const graphIn = await receiveMessage(session, timer);450      if (graphIn.type !== "SYNC_DEVICE_MANIFEST") {451        return errorOutcome(452          result,453          `expected SYNC_DEVICE_MANIFEST, got ${graphIn.type}`,454        );455      }456      const ours = await vault.buildOutgoingDeviceManifest();457      await sendMessage(session, {458        type: "SYNC_DEVICE_MANIFEST",459        manifest: ours,460      });461      if (graphIn.manifest) {462        await vault.applyIncomingDeviceManifest(463          graphIn.manifest,464          session.remoteStaticPublicKey,465        );466      }467    }468469    stage = "waiting_for_device_done";470    const done = await receiveMessage(session, timer);471    if (done.type !== "SYNC_DONE") {472      return errorOutcome(result, `expected SYNC_DONE, got ${done.type}`);473    }474    await sendMessage(session, {475      type: "SYNC_DONE",476      entriesReceived: result.entriesReceived,477      entriesSent: result.entriesSent,478    });479    result.status = "completed";480    return result;481  } catch (error) {482    if ((error as Error).message === "sync_timeout") {483      result.status = "timeout";484      result.error = `sync_timeout:${stage}`;485      return result;486    }487    return errorOutcome(result, (error as Error).message);488  } finally {489    timer.cancel();490  }491}492493/**494 * Wait for the peer's closing SYNC_DONE echo, treating a hang-up as success.495 *496 * By the time we send our own SYNC_DONE, every entry and tombstone has already497 * been exchanged and applied on both sides — the echo is a courtesy, not a498 * commit. Android closes the socket as soon as it has our DONE, so the parked499 * receive() rejects with "WebSocket closed"; counting that as `run_failed`500 * reported a red indicator for syncs that had fully succeeded and whose data501 * had demonstrably landed on the peer.502 *503 * A wrong *message* here is still a real protocol error — only a closed504 * transport is forgiven.505 */506async function awaitClosingDone(507  session: SyncSession,508  timer: Timer,509): Promise<{ error?: string }> {510  try {511    const done = await receiveMessage(session, timer);512    if (done.type !== "SYNC_DONE") {513      return { error: `expected SYNC_DONE, got ${done.type}` };514    }515    return {};516  } catch (err) {517    if ((err as Error).message === "sync_timeout") throw err;518    if (isConnectionClosedError(err)) return {};519    return { error: (err as Error).message };520  }521}522523function diffNewer(remote: Record<string, number>, local: Record<string, number>): string[] {524  const want: string[] = [];525  for (const [id, ts] of Object.entries(remote)) {526    const ours = local[id];527    if (ours === undefined || ts > ours) want.push(id);528  }529  return want;530}531532interface Timer {533  signal: AbortSignal;534  cancel(): void;535}536537function scheduleTimeout(ms: number): Timer {538  const ctrl = new AbortController();539  const t = setTimeout(() => ctrl.abort(new Error("sync_timeout")), ms);540  return {541    signal: ctrl.signal,542    cancel: () => clearTimeout(t),543  };544}545546async function sendMessage(session: SyncSession, msg: SyncMessage): Promise<void> {547  session.send(new TextEncoder().encode(JSON.stringify(msg)));548}549550async function receiveMessage(session: SyncSession, timer: Timer): Promise<SyncMessage> {551  const pending = session.receive();552  const aborted = new Promise<never>((_, reject) => {553    if (timer.signal.aborted) reject(new Error("sync_timeout"));554    timer.signal.addEventListener("abort", () => reject(new Error("sync_timeout")));555  });556  const bytes = await Promise.race([pending, aborted]);557  // MV-3: decrypted plaintext can exceed the wire frame (and a non-WS558  // transport may not have enforced anything) — re-check before JSON.parse559  // materialises a multi-GB string/object graph.560  if (bytes.length > SYNC_LIMITS.MAX_FRAME_BYTES) {561    throw new Error(`sync frame exceeds max size (${SYNC_LIMITS.MAX_FRAME_BYTES} bytes)`);562  }563  const parsed = JSON.parse(new TextDecoder().decode(bytes));564  if (typeof parsed?.type !== "string") throw new Error("malformed SyncMessage");565  return parsed as SyncMessage;566}567568function errorOutcome(outcome: SyncOutcome, error: string): SyncOutcome {569  outcome.status = "errored";570  outcome.error = error;571  return outcome;572}