fix: align Bad MAC retry receipt with WA Desktop behavior

Three issues corrected in sendRetryRequest (receiver side):

1. BUG — error code hardcoded as '0' (UnknownError) in retry receipt.
   The peer (especially another InfiniteAPI instance) could not detect
   the failure type and would fall back to the 1-hour session recreation
   timeout instead of recreating immediately.
   Fix: derive error code from the actual libsignal decryption error
   message stored in messageStubParameters[0].

2. BUG — shouldRecreateSession block was reading the wrong node: it
   called getBinaryNodeChild(node, 'retry') on the incoming bad-MAC
   message, which never has a <retry> child. errorCode was always
   undefined, so MAC_ERROR_CODES never matched — the logic was dead.

3. BUG (consequence of #2) — when shouldRecreateSession accidentally
   did fire (no-session or 1-hour timeout), it deleted the receiver's
   session BEFORE the sender's pkmsg arrived. This opened a race window
   where concurrent messages would fail with "No Session".
   Fix: remove the entire session-deletion block from sendRetryRequest.
   The Signal Protocol pkmsg automatically overwrites the corrupted
   session — no explicit delete needed on the receiver side.

WA Desktop reference (CDP capture 2026-03-19):
- Retry receipt sent ~99ms after Bad MAC with specific error code
- pkmsg arrives ~304ms later, new session established automatically
- Old session is never deleted; pkmsg overwrites it implicitly
- Full recovery in ~1.3s without any race window

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Renato Alcara
2026-03-19 23:10:59 -03:00
parent 876af2e96c
commit ba1c1b0fdb
+26 -39
View File
@@ -1213,7 +1213,7 @@ export const makeMessagesRecvSocket = (config: SocketConfig) => {
return await query(stanza)
}
const sendRetryRequest = async (node: BinaryNode, forceIncludeKeys = false) => {
const sendRetryRequest = async (node: BinaryNode, forceIncludeKeys = false, decryptionError?: string) => {
const { fullMessage } = decodeMessageNode(node, authState.creds.me!.id, authState.creds.me!.lid || '')
const { key: msgKey } = fullMessage
const msgId = msgKey.id!
@@ -1311,39 +1311,27 @@ export const makeMessagesRecvSocket = (config: SocketConfig) => {
const { account, signedPreKey, signedIdentityKey: identityKey } = authState.creds
const fromJid = node.attrs.from!
// Check if we should recreate the session
let shouldRecreateSession = false
let recreateReason = ''
if (enableAutoSessionRecreation && messageRetryManager && retryCount >= 1) {
try {
// Check if we have a session with this JID
const sessionId = signalRepository.jidToSignalProtocolAddress(fromJid)
const hasSession = await signalRepository.validateSession(fromJid)
// Extract error code from retry node if present (for MAC error detection)
const retryNode = getBinaryNodeChild(node, 'retry')
const errorAttr = retryNode?.attrs?.error
const errorCode = messageRetryManager.parseRetryErrorCode(errorAttr)
const result = messageRetryManager.shouldRecreateSession(fromJid, hasSession.exists, errorCode)
shouldRecreateSession = result.recreate
recreateReason = result.reason
if (shouldRecreateSession) {
logger.debug({ fromJid, retryCount, reason: recreateReason, errorCode }, 'recreating session for retry')
// Delete existing session to force recreation
// CRITICAL: Use same transaction key as encrypt/decrypt operations to prevent race
// Using meId ensures this delete serializes with sendMessage() and other session operations
await authState.keys.transaction(async () => {
await authState.keys.set({ session: { [sessionId]: null } })
}, authState.creds.me?.id || 'session-operation')
forceIncludeKeys = true
}
} catch (error) {
logger.warn({ error, fromJid }, 'failed to check session recreation')
}
}
// Derive the Signal error code from the actual decryption failure message.
// This is sent in the retry receipt so the peer (even another InfiniteAPI instance)
// knows the exact reason and can recreate the session immediately instead of waiting
// for the 1-hour timeout fallback.
//
// Codes mirror RetryReason enum in message-retry-manager.ts:
// 0 = UnknownError | 1 = NoSession | 2 = InvalidKey
// 3 = InvalidKeyId | 7 = BadMac (= SignalErrorInvalidMessage/InvalidCipherKey)
//
// NOTE: We do NOT delete the session here (receiver side). The Signal Protocol
// recovers automatically when the sender's pkmsg arrives — it overwrites the
// corrupted session. Deleting prematurely creates a race window where no session
// exists, which can cause "No Session" errors on concurrent messages.
const retryErrorCode = (() => {
if (!decryptionError) return 0
if (/bad\s*mac/i.test(decryptionError)) return 7 // SignalErrorBadMac
if (/no\s*session/i.test(decryptionError)) return 1 // SignalErrorNoSession
if (/pre\s*key/i.test(decryptionError)) return 3 // SignalErrorInvalidKeyId
if (/invalid\s*key/i.test(decryptionError)) return 2 // SignalErrorInvalidKey
return 0
})()
if (retryCount <= 2) {
// Use new retry manager for phone requests if available
@@ -1384,8 +1372,7 @@ export const makeMessagesRecvSocket = (config: SocketConfig) => {
id: node.attrs.id!,
t: node.attrs.t!,
v: '1',
// ADD ERROR FIELD
error: '0'
error: retryErrorCode.toString()
}
},
{
@@ -1404,7 +1391,7 @@ export const makeMessagesRecvSocket = (config: SocketConfig) => {
receipt.attrs.participant = node.attrs.participant
}
if (retryCount > 1 || forceIncludeKeys || shouldRecreateSession) {
if (retryCount > 1 || forceIncludeKeys) {
const { update, preKeys } = await getNextPreKeys(authState, 1)
const [keyId] = Object.keys(preKeys)
@@ -2507,7 +2494,7 @@ export const makeMessagesRecvSocket = (config: SocketConfig) => {
}
const encNode = getBinaryNodeChild(node, 'enc')
await sendRetryRequest(node, !encNode)
await sendRetryRequest(node, !encNode, errorMessage)
if (retryRequestDelayMs) {
await delay(retryRequestDelayMs)
}
@@ -2516,7 +2503,7 @@ export const makeMessagesRecvSocket = (config: SocketConfig) => {
// Still attempt retry even if pre-key upload failed
try {
const encNode = getBinaryNodeChild(node, 'enc')
await sendRetryRequest(node, !encNode)
await sendRetryRequest(node, !encNode, errorMessage)
} catch (retryErr) {
logger.error({ retryErr }, 'Failed to send retry after error handling')
}