Stian/Shade
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(session): remember where aliasSession moved a session

Browse Source 
aliasSession knew that two labels name the same peer, then threw that
knowledge away. The binding lived only in the caller's memory, so a
restart lost it — and the peer could not repair it from its side.

First contact forces the receiver to label a session by the only sender
hint a relay surfaces, an 8-byte signing-key fingerprint (`fp:<hex>`).
Once the peer announces its canonical address, aliasSession moves the
session there. But the peer keeps sending under `fp:<hex>`, because its
transport derives the same label from the same hint every time. After a
restart the session sat under the canonical address, inbound frames
resolved to `fp:<hex>`, and nothing matched. The peer held a valid
session so it never re-ran X3DH: the failure was permanent, and only a
manual re-link cleared it.

Observed in Prism as `No session for address: fp:579c3b335d66e2c0` on
every receive for three days, with a phone whose every RPC timed out.

StorageProvider gains saveSessionAlias / getSessionAlias /
removeSessionAliasesFor, optional so third-party implementations keep
compiling, and implemented across all seven backends. Lookups resolve
through resolveLabel(), which runs BEFORE the peer mutex — locking the
alias while mutating the canonical session would let an aliased and a
canonical caller ratchet the same state concurrently.

A live session under a label always wins over an alias, and prekey
envelopes never resolve: both keep a re-link establishing a fresh
session instead of being redirected into the stale one. Aliases are
dropped in resetSession and acceptIdentityChange, and memoized so the
hot path costs no extra read.

The sdk.test.ts case that asserted a dead fp-label encoded the old
behaviour; it now pins the new contract.

Verified: 1166 tests pass (from 1160). With alias persistence disabled
as a negative control, 5 of the 6 new tests fail, including both
restart cases.

Also drops `baseUrl` from the consumer-strict tsconfig — removed in
TS 6.0, and it was failing the typecheck that gates publishing.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Sterister
2026-08-13 19:36:31 +02:00
parent b44acf867b
commit 96c20cb4b2
39 changed files with 536 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
packages/shade-core/package.json
View File

@@ -1,6 +1,6 @@
{
"name": "@shade/core",
"version": "4.11.1",
"version": "4.12.0",
"type": "module",
"main": "src/index.ts",
"types": "src/index.ts",

98
packages/shade-core/src/session.ts
View File

@@ -86,6 +86,14 @@ export class ShadeSessionManager {
* fully concurrent.
*/
private readonly peerOpChains = new Map<string, Promise<unknown>>();
/**
* Memoized `alias → canonical` lookups; `null` records "no alias" so a
* label without one costs a single storage read for the life of the
* process instead of one per encrypt/decrypt. Aliases change only via
* the mutators in this class, each of which clears the whole map — it
* holds at most one entry per peer, so rebuilding is cheap.
*/
private readonly aliasCache = new Map<string, string | null>();
constructor(
private readonly crypto: CryptoProvider,
@@ -151,6 +159,46 @@ export class ShadeSessionManager {
}
}
/**
* Map a session label onto the label its state actually lives under.
*
* `aliasSession` moves a session from a first-contact label (typically
* `fp:<hex>`) to the peer's canonical address, but the peer keeps
* sending under the old one. The persisted alias lets us follow that
* move across restarts — see the alias block in `StorageProvider`.
*
* A live session under `label` always wins: after a re-link the peer
* re-runs X3DH and a fresh session is established under the
* first-contact label again, and that new session — not the stale
* alias target — is the one that can decrypt what follows.
*
* Resolves one hop only. Aliases are always written pointing at a
* canonical label, so a chain would mean corrupt state; following it
* would risk a loop for no legitimate gain.
*
* MUST be called before taking the peer mutex: locking the alias while
* mutating the canonical session would let an aliased caller and a
* canonical caller ratchet the same state concurrently.
*/
private async resolveLabel(label: string): Promise<string> {
let canonical = this.aliasCache.get(label);
if (canonical === undefined) {
canonical = (await this.storage.getSessionAlias?.(label)) ?? null;
this.aliasCache.set(label, canonical);
}
if (canonical === null || canonical === label) return label;
if (await this.storage.getSession(label)) return label;
return canonical;
}
/**
* Public label resolution for callers that need to know where a
* peer's state lives (e.g. a transport routing an inbound frame).
*/
async resolveSessionLabel(label: string): Promise<string> {
return this.resolveLabel(label);
}
/** Get the event emitter (if observability is enabled) */
getEvents(): ShadeEventEmitter | undefined {
return this.events;
@@ -268,6 +316,11 @@ export class ShadeSessionManager {
*/
async resetSession(address: string): Promise<void> {
await this.storage.removeSession(address);
// Aliases pointing here are now dangling — drop them so the next
// first-contact frame resolves to its own label and establishes the
// fresh session this reset exists to force.
await this.storage.removeSessionAliasesFor?.(address);
this.aliasCache.clear();
this.events?.emit('session.removed', { address });
// Note: we keep the trusted identity; new session will verify against it.
}
@@ -336,6 +389,14 @@ export class ShadeSessionManager {
await this.storage.bumpPeerIdentityVersion(newLabel);
}
await this.storage.removeSession(oldLabel);
// Remember the move durably. The peer goes on sending under
// `oldLabel` — its transport derives the same first-contact label
// from our relay hint every time — so without this record every
// inbound frame after a restart resolves to a label whose session
// we just removed, and the peer (holding a valid session, never
// re-running X3DH) can never recover on its own.
await this.storage.saveSessionAlias?.(oldLabel, newLabel);
this.aliasCache.clear();
this.events?.emit('session.aliased', { oldLabel, newLabel });
}
@@ -349,6 +410,10 @@ export class ShadeSessionManager {
// because isTrustedIdentity() compares not retrieves; we just emit the new hash)
await this.storage.saveTrustedIdentity(address, newIdentityKey);
await this.storage.removeSession(address);
// The peer rotated identity — any alias into the old session is
// dangling and must not redirect frames meant for the new one.
await this.storage.removeSessionAliasesFor?.(address);
this.aliasCache.clear();
if (this.events) {
const newHash = await shortHash(this.crypto, newIdentityKey);
@@ -516,14 +581,19 @@ export class ShadeSessionManager {
* Subsequent messages are standard RatchetMessages.
*/
async encrypt(address: string, plaintext: string): Promise<ShadeEnvelope> {
return this.withSpan('encrypt', address, async () => {
const session = await this.storage.getSession(address);
if (!session) throw new NoSessionError(address);
// Follow a persisted alias so a caller still holding a first-contact
// label reaches the session that alias moved to — without this, a
// restarted host can decrypt a peer's frames but not reply to them.
// Resolved before the mutex so the lock lands on the canonical label.
const target = await this.resolveLabel(address);
return this.withSpan('encrypt', target, async () => {
const session = await this.storage.getSession(target);
if (!session) throw new NoSessionError(target);
const ratchetMsg = await ratchetEncrypt(this.crypto, session, enc.encode(plaintext));
this.events?.emit('message.encrypted', {
address,
address: target,
counter: ratchetMsg.counter,
ciphertextSize: ratchetMsg.ciphertext.length,
});
@@ -532,7 +602,7 @@ export class ShadeSessionManager {
const x3dh = (session as any).__x3dh;
if (x3dh) {
delete (session as any).__x3dh;
await this.storage.saveSession(address, session);
await this.storage.saveSession(target, session);
const preKeyMsg: PreKeyMessage = {
registrationId: x3dh.registrationId,
@@ -546,16 +616,16 @@ export class ShadeSessionManager {
type: 'prekey',
content: preKeyMsg,
timestamp: Date.now(),
senderAddress: address,
senderAddress: target,
};
}
await this.storage.saveSession(address, session);
await this.storage.saveSession(target, session);
return {
type: 'ratchet',
content: ratchetMsg,
timestamp: Date.now(),
senderAddress: address,
senderAddress: target,
};
});
}
@@ -564,11 +634,17 @@ export class ShadeSessionManager {
* Decrypt a message from a peer. Handles both PreKeyMessage and RatchetMessage.
*/
async decrypt(address: string, envelope: ShadeEnvelope): Promise<string> {
return this.withSpan('decrypt', address, async () => {
// A prekey envelope carries its own X3DH material and establishes a
// fresh session, which must land under the label it arrived on —
// that is exactly what a re-link looks like. Only ratchet envelopes,
// which need state that already exists, follow an alias.
const target =
envelope.type === 'prekey' ? address : await this.resolveLabel(address);
return this.withSpan('decrypt', target, async () => {
if (envelope.type === 'prekey') {
return this.decryptPreKeyMessage(address, envelope.content as PreKeyMessage);
return this.decryptPreKeyMessage(target, envelope.content as PreKeyMessage);
}
return this.decryptRatchetMessage(address, envelope.content as RatchetMessage);
return this.decryptRatchetMessage(target, envelope.content as RatchetMessage);
});
}

30
packages/shade-core/src/storage.ts
View File

@@ -165,6 +165,36 @@ export interface StorageProvider {
/** Remove session for a peer */
removeSession(address: string): Promise<void>;
// ─── Session label aliases (V4.12) ────────────────────────
//
// First contact forces the receiver to label a session by the only
// sender hint the relay surfaces — an 8-byte signing-key fingerprint
// (`fp:<hex>`). A later in-band announcement reveals the peer's
// canonical address and `aliasSession` moves the session there.
//
// The peer, however, keeps sending under whatever label its own
// transport derives — which for a fingerprint-hinted relay is still
// `fp:<hex>`. Before V4.12 that binding lived only in the consumer's
// memory: after a restart the alias was gone, inbound ratchet frames
// resolved to `fp:<hex>`, found no session there, and failed forever
// (the peer holds a valid session so it never re-runs X3DH).
//
// Persisting the alias makes the binding survive restarts, so
// `getSession` can follow it. Optional so third-party storage
// implementations keep compiling — they simply lose alias recovery.
/**
* Record that `alias` names the same peer session as `canonical`.
* Idempotent upsert on `alias`.
*/
saveSessionAlias?(alias: string, canonical: string): Promise<void>;
/** Resolve an alias to its canonical label (null when unaliased). */
getSessionAlias?(alias: string): Promise<string | null>;
/** Drop every alias pointing at `canonical` (session teardown). */
removeSessionAliasesFor?(canonical: string): Promise<void>;
/** Check if we trust a remote identity key (for TOFU or pinned keys) */
isTrustedIdentity(address: string, identityKey: Uint8Array): Promise<boolean>;

165
packages/shade-core/tests/session-aliases.test.ts Normal file
View File

@@ -0,0 +1,165 @@
import { describe, test, expect, beforeEach } from 'bun:test';
import { SubtleCryptoProvider, MemoryStorage } from '@shade/crypto-web';
import { ShadeSessionManager } from '../src/index.js';
const crypto = new SubtleCryptoProvider();
/**
* Durable session-label aliases (V4.12).
*
* THE BUG THIS FILE EXISTS TO KILL — diagnosed live in Prism:
*
* A phone pairs with a host. First contact forces the host to label
* the session by the only sender hint the relay surfaces, an 8-byte
* signing-key fingerprint (`fp:<hex>`). The pair handshake then
* announces the phone's canonical address and the host calls
* `aliasSession(fp:<hex> → device:<addr>)`, which moved the session
* on disk and dropped the binding.
*
* The phone, however, keeps sending under `fp:<hex>` — its transport
* derives the same label from the same relay hint every time. While
* the host process lived, an in-memory map papered over the gap.
* After a restart that map was empty, every inbound ratchet frame
* resolved to `fp:<hex>`, found no session, and failed. The phone
* held a perfectly valid session so it never re-ran X3DH — meaning
* the failure was permanent and self-inflicted, not transient.
*
* Observed as `No session for address: fp:579c3b335d66e2c0` on every
* receive for three days, with the phone's RPCs timing out forever.
*
* The fix: `aliasSession` persists the binding, and session lookup
* follows it. These tests pin the restart behaviour specifically —
* a same-process test cannot fail the way production did.
*/
describe('session label aliases', () => {
let alice: ShadeSessionManager;
let bob: ShadeSessionManager;
let aliceStorage: MemoryStorage;
let bobStorage: MemoryStorage;
/** The first-contact label Alice is forced to use for Bob. */
const FP = 'fp:579c3b335d66e2c0';
beforeEach(async () => {
aliceStorage = new MemoryStorage();
bobStorage = new MemoryStorage();
alice = new ShadeSessionManager(crypto, aliceStorage);
bob = new ShadeSessionManager(crypto, bobStorage);
await alice.initialize();
await bob.initialize();
});
/**
* Bob initiates X3DH against Alice, exactly like a phone reaching a
* host it just scanned. Returns Bob's first (prekey) envelope.
*/
async function bobInitiates(target: ShadeSessionManager, initiator: ShadeSessionManager) {
const otpks = await target.generateOneTimePreKeys(10);
const bundle = await target.createPreKeyBundle();
const otpk = otpks[0]!;
bundle.oneTimePreKey = { keyId: otpk.keyId, publicKey: otpk.keyPair.publicKey };
await initiator.initSessionFromBundle('alice', bundle);
}
/** Simulate a host restart: fresh manager, same durable storage. */
async function restartAlice(): Promise<ShadeSessionManager> {
const revived = new ShadeSessionManager(crypto, aliceStorage);
await revived.initialize();
return revived;
}
test('an aliased session still decrypts under the old label after a restart', async () => {
await bobInitiates(alice, bob);
// First contact lands under the fingerprint label.
const env1 = await bob.encrypt('alice', 'hello, my address is bob');
expect(await alice.decrypt(FP, env1)).toBe('hello, my address is bob');
// Alice canonicalizes to Bob's announced address.
await alice.aliasSession(FP, 'bob');
// The host restarts. Storage survives; every in-memory map does not.
const alice2 = await restartAlice();
// Bob has a valid session and keeps sending under the same label he
// always has. Before the fix this threw NoSessionError forever.
const env2 = await bob.encrypt('alice', 'still here after restart');
expect(await alice2.decrypt(FP, env2)).toBe('still here after restart');
});
test('the host can reply under the old label after a restart', async () => {
await bobInitiates(alice, bob);
const env1 = await bob.encrypt('alice', 'hi');
await alice.decrypt(FP, env1);
await alice.aliasSession(FP, 'bob');
const alice2 = await restartAlice();
// Decrypting is only half of it — a host that cannot encrypt back
// leaves every RPC hanging just the same.
const reply = await alice2.encrypt(FP, 'reply from the host');
expect(await bob.decrypt('alice', reply)).toBe('reply from the host');
});
test('a live session under the label wins over an alias (re-link)', async () => {
await bobInitiates(alice, bob);
const env1 = await bob.encrypt('alice', 'first pairing');
await alice.decrypt(FP, env1);
await alice.aliasSession(FP, 'bob');
const alice2 = await restartAlice();
// Bob reinstalls: brand-new identity, same relay fingerprint label.
const bob2Storage = new MemoryStorage();
const bob2 = new ShadeSessionManager(crypto, bob2Storage);
await bob2.initialize();
await bobInitiates(alice2, bob2);
// The prekey envelope must establish a FRESH session under FP rather
// than being redirected into the stale aliased one.
const fresh1 = await bob2.encrypt('alice', 'fresh contact');
expect(await alice2.decrypt(FP, fresh1)).toBe('fresh contact');
// And subsequent ratchet frames must keep using that new session.
const fresh2 = await bob2.encrypt('alice', 'second message');
expect(await alice2.decrypt(FP, fresh2)).toBe('second message');
});
test('resolveSessionLabel reports where the state actually lives', async () => {
await bobInitiates(alice, bob);
const env1 = await bob.encrypt('alice', 'hi');
await alice.decrypt(FP, env1);
expect(await alice.resolveSessionLabel(FP)).toBe(FP);
await alice.aliasSession(FP, 'bob');
const alice2 = await restartAlice();
expect(await alice2.resolveSessionLabel(FP)).toBe('bob');
// An unaliased label resolves to itself.
expect(await alice2.resolveSessionLabel('carol')).toBe('carol');
});
test('resetSession drops aliases pointing at the cleared session', async () => {
await bobInitiates(alice, bob);
const env1 = await bob.encrypt('alice', 'hi');
await alice.decrypt(FP, env1);
await alice.aliasSession(FP, 'bob');
expect(await aliceStorage.getSessionAlias(FP)).toBe('bob');
await alice.resetSession('bob');
// A dangling alias would redirect the next first-contact frame into
// a session that no longer exists, defeating the reset.
expect(await aliceStorage.getSessionAlias(FP)).toBeNull();
expect(await alice.resolveSessionLabel(FP)).toBe(FP);
});
test('aliasing persists the binding to storage', async () => {
await bobInitiates(alice, bob);
const env1 = await bob.encrypt('alice', 'hi');
await alice.decrypt(FP, env1);
expect(await aliceStorage.getSessionAlias(FP)).toBeNull();
await alice.aliasSession(FP, 'bob');
expect(await aliceStorage.getSessionAlias(FP)).toBe('bob');
});
});