Fix ordinary Noise handshake races

This commit is contained in:
jack
2026-07-26 12:21:38 +02:00
committed by jack
parent 3b7c590df4
commit 52e753eee2
10 changed files with 1850 additions and 129 deletions
@@ -37,9 +37,29 @@ enum NoiseSecurityConstants {
// Noise XX message 1 contains only the initiator's 32-byte ephemeral key.
static let xxInitialMessageSize = 32
// Bounds an ordinary initiator whose message 1 or 2 is lost.
static let ordinaryHandshakeTimeout: TimeInterval = 10
// Bounds the receive-only rollback quarantine created by an unauthenticated
// inbound message 1. A lost message 3 must not strand outbound traffic.
static let ordinaryResponderHandshakeTimeout: TimeInterval = 20
// A released client may immediately retry after both crossed initiators
// yielded. Give that unilateral retry a brief head start before the
// patched side spends its one bounded recovery.
static let handshakeCollisionRecoveryDelay: TimeInterval = 0.2
// Rate-limited recovery remains actionable without spinning.
static let handshakeRateLimitRecoveryDelay: TimeInterval = 60
// Covers only reordering between a winning message 3 and the losing
// crossed message 1.
static let recentInitiatorCompletionGracePeriod: TimeInterval = 1
// After unauthenticated responder rollback, reject another attempt long
// enough that paced message 1 traffic cannot keep outbound paused. A
// legitimate peer converges through the one manager-owned local retry.
static let ordinaryReconnectRollbackCooldown: TimeInterval = 60
// Session timeout - sessions older than this should be renegotiated
static let sessionTimeout: TimeInterval = 86400 // 24 hours
+6
View File
@@ -13,3 +13,9 @@ enum NoiseSessionError: Error, Equatable {
case alreadyEstablished
case peerIdentityMismatch
}
/// The manager owns the exact attempt's one bounded recovery. Packet handling
/// must not launch its historical second, immediate restart for this failure.
struct NoiseManagedHandshakeFailure: Error {
let underlying: Error
}
File diff suppressed because it is too large Load Diff
@@ -112,6 +112,14 @@ final class BLENoisePacketHandler {
didEstablishAuthenticatedSession:
result.didEstablishAuthenticatedSession
)
} catch let managedFailure as NoiseManagedHandshakeFailure {
SecureLogger.error(
"Failed to process handshake; manager owns recovery: \(managedFailure.underlying)"
)
return BLENoiseHandshakeHandlingResult(
processed: false,
didEstablishAuthenticatedSession: false
)
} catch NoiseSessionError.peerIdentityMismatch {
// The responder was already discarded by the session manager.
// Do not let a spoofed claimed ID trigger a fresh outbound
+113 -19
View File
@@ -2955,14 +2955,21 @@ extension BLEService: CBCentralManagerDelegate {
startScanning()
case .poweredOff:
// Bluetooth was turned off - stop scanning and clean up connection state
// CoreBluetooth has already transitioned out of poweredOn. Do
// not issue stop/cancel commands now; they are rejected as API
// misuse. Retire our link state locally instead.
SecureLogger.info("📴 Bluetooth powered off - cleaning up central state", category: .session)
central.stopScan()
// Mark all peripheral connections as disconnected (they are now invalid)
let peripheralStates = linkStateStore.peripheralStates
let peerIDs: [PeerID] = peripheralStates.compactMap(\.peerID)
for state in peripheralStates {
central.cancelPeripheralConnection(state.peripheral)
let peripheralID = state.peripheral.identifier.uuidString
collectionsQueue.sync(flags: .barrier) {
pendingPeripheralWrites.discardAll(for: peripheralID)
}
noiseAuthenticatedLinkOwners.removeValue(
forKey: .peripheral(peripheralID)
)
noiseReconnectPolicy.endLinkEpoch(.peripheral(peripheralID))
}
_ = linkStateStore.clearPeripherals()
// Notify UI of disconnections
@@ -2975,7 +2982,6 @@ extension BLEService: CBCentralManagerDelegate {
case .unauthorized:
// User denied Bluetooth permission
SecureLogger.warning("🚫 Bluetooth unauthorized - user denied permission", category: .session)
central.stopScan()
_ = linkStateStore.clearPeripherals()
case .unsupported:
@@ -3867,8 +3873,19 @@ extension BLEService: CBPeripheralManagerDelegate {
case .poweredOff:
// Bluetooth was turned off - clean up peripheral state
SecureLogger.info("📴 Bluetooth powered off - cleaning up peripheral state", category: .session)
peripheral.stopAdvertising()
// Clear subscribed centrals (they are now invalid)
let centralSnapshot = linkStateStore.subscribedCentralSnapshot
for central in centralSnapshot.centrals {
let centralID = central.identifier.uuidString
noiseAuthenticatedLinkOwners.removeValue(
forKey: .central(centralID)
)
noiseReconnectPolicy.endLinkEpoch(.central(centralID))
}
collectionsQueue.sync(flags: .barrier) {
pendingNotifications.removeAll()
pendingWriteBuffers.removeAll()
}
let centralPeerIDs = linkStateStore.clearCentrals()
subscriptionAnnounceLimiter.removeAll()
characteristic = nil
@@ -3882,7 +3899,6 @@ extension BLEService: CBPeripheralManagerDelegate {
case .unauthorized:
// User denied Bluetooth permission
SecureLogger.warning("🚫 Bluetooth unauthorized for peripheral role", category: .session)
peripheral.stopAdvertising()
_ = linkStateStore.clearCentrals()
subscriptionAnnounceLimiter.removeAll()
characteristic = nil
@@ -4659,13 +4675,71 @@ extension BLEService {
)
}
}
service.onRekeyHandshakeReady = { [weak self] peerID, message in
self?.messageQueue.async { [weak self] in
guard let self else { return }
service.onRekeyHandshakeReady = {
[weak self, weak service] peerID, initiation in
self?.messageQueue.async(flags: .barrier) {
[weak self, weak service] in
guard let self,
let service,
self.noiseService === service else {
return
}
self.noteNoiseSessionCleared(for: peerID)
guard let message = service.claimHandshakeInitiation(
initiation,
for: peerID
) else {
return
}
self.broadcastNoiseHandshake(message, to: peerID)
}
}
service.onHandshakeRecoveryRequired = {
[weak self, weak service] request in
guard let self, let service else { return }
self.messageQueue.async(flags: .barrier) {
[weak self, weak service] in
guard let self,
let service,
self.noiseService === service else {
return
}
let peerID = request.peerID
guard self.isPeerReachable(peerID) else {
service.cancelHandshakeRecovery(request)
return
}
do {
guard let preparation =
try service.prepareHandshakeRecovery(request) else {
return
}
switch preparation {
case .ordinary(let initiation):
self.noteNoiseSessionCleared(for: peerID)
guard let handshakeData =
service.claimHandshakeInitiation(
initiation,
for: peerID
) else {
return
}
self.broadcastNoiseHandshake(
handshakeData,
to: peerID
)
case .transferred:
return
}
} catch {
SecureLogger.error(
"Failed to prepare handshake recovery with \(peerID.id.prefix(8))…: \(error)",
category: .session
)
}
}
}
service.onSessionRestoredWithGeneration = { [weak self, weak service] peerID, generation in
guard let self, let service else { return }
self.messageQueue.async { [weak self, weak service] in
@@ -5941,12 +6015,27 @@ extension BLEService {
}
private func initiateNoiseHandshake(with peerID: PeerID) {
// Use NoiseEncryptionService for handshake
guard !noiseService.hasSession(with: peerID) else { return }
let service = noiseService
do {
let handshakeData = try noiseService.initiateHandshake(with: peerID)
broadcastNoiseHandshake(handshakeData, to: peerID)
guard let initiation = try service.initiateHandshakeIfNeeded(
with: peerID,
retryOnTimeout: true
) else {
return
}
messageQueue.async(flags: .barrier) {
[weak self, weak service] in
guard let self,
let service,
self.noiseService === service,
let handshakeData = service.claimHandshakeInitiation(
initiation,
for: peerID
) else {
return
}
self.broadcastNoiseHandshake(handshakeData, to: peerID)
}
} catch {
SecureLogger.error("Failed to initiate handshake: \(error)")
}
@@ -5972,13 +6061,18 @@ extension BLEService {
private func initiateNoiseReconnectHandshake(with peerID: PeerID) {
let service = noiseService
do {
let initiation = try service.initiateReconnectHandshake(with: peerID)
noteNoiseSessionCleared(for: peerID)
let initiation = try service.initiateReconnectHandshake(
with: peerID,
retryOnTimeout: true
)
messageQueue.async(flags: .barrier) { [weak self, weak service] in
guard let self,
let service,
self.noiseService === service,
let handshakeData = service.claimHandshakeInitiation(
self.noiseService === service else {
return
}
self.noteNoiseSessionCleared(for: peerID)
guard let handshakeData = service.claimHandshakeInitiation(
initiation,
for: peerID
) else {
+80 -10
View File
@@ -184,11 +184,13 @@ final class NoiseEncryptionService {
private var onPeerAuthenticatedHandlers: [((PeerID, String) -> Void)] = [] // Array of handlers for peer authentication
private var onPeerAuthenticatedWithGenerationHandlers: [((PeerID, String, UUID) -> Void)] = []
var onHandshakeRequired: ((PeerID) -> Void)? // peerID needs handshake
/// Automatic rekey removed the old session and produced XX message 1.
/// The transport must clear session-scoped state and put these exact bytes
/// on the wire; merely reporting "handshake required" strands the partial
/// initiator session because a second initiate call sees it already exists.
var onRekeyHandshakeReady: ((_ peerID: PeerID, _ message: Data) -> Void)?
/// Automatic rekey prepared XX message 1. The transport must claim the
/// exact attempt at its actual BLE handoff; a crossed inbound initiation
/// can invalidate the token before that point.
var onRekeyHandshakeReady:
((_ peerID: PeerID, _ initiation: NoiseHandshakeInitiation) -> Void)?
var onHandshakeRecoveryRequired:
((_ request: NoiseHandshakeRecoveryRequest) -> Void)?
/// An unauthenticated reconnect attempt failed or timed out and the
/// receive-only rollback session became the active transport again.
/// Transport queues must be drained for this exact restored generation.
@@ -225,8 +227,14 @@ final class NoiseEncryptionService {
init(
keychain: KeychainManagerProtocol,
ordinaryHandshakeTimeout: TimeInterval =
NoiseSecurityConstants.ordinaryHandshakeTimeout,
ordinaryResponderHandshakeTimeout: TimeInterval =
NoiseSecurityConstants.ordinaryResponderHandshakeTimeout
NoiseSecurityConstants.ordinaryResponderHandshakeTimeout,
recentInitiatorCompletionGracePeriod: TimeInterval =
NoiseSecurityConstants.recentInitiatorCompletionGracePeriod,
ordinaryReconnectRollbackCooldown: TimeInterval =
NoiseSecurityConstants.ordinaryReconnectRollbackCooldown
) {
self.keychain = keychain
self.localPrekeys = LocalPrekeyStore(keychain: keychain)
@@ -320,7 +328,13 @@ final class NoiseEncryptionService {
self.sessionManager = NoiseSessionManager(
localStaticKey: staticIdentityKey,
keychain: keychain,
ordinaryResponderHandshakeTimeout: ordinaryResponderHandshakeTimeout
ordinaryHandshakeTimeout: ordinaryHandshakeTimeout,
ordinaryResponderHandshakeTimeout:
ordinaryResponderHandshakeTimeout,
recentInitiatorCompletionGracePeriod:
recentInitiatorCompletionGracePeriod,
ordinaryReconnectRollbackCooldown:
ordinaryReconnectRollbackCooldown
)
// Set up session callbacks
@@ -334,6 +348,9 @@ final class NoiseEncryptionService {
sessionManager.onSessionRestored = { [weak self] peerID, generation in
self?.onSessionRestoredWithGeneration?(peerID, generation)
}
sessionManager.onHandshakeRecoveryRequired = { [weak self] request in
self?.onHandshakeRecoveryRequired?(request)
}
// Start session maintenance timer
startRekeyTimer()
@@ -698,11 +715,41 @@ final class NoiseEncryptionService {
return handshakeData
}
/// Atomically admits and prepares one initial ordinary handshake. Returns
/// nil when another discovery callback already created a session.
func initiateHandshakeIfNeeded(
with peerID: PeerID,
retryOnTimeout: Bool = false
) throws -> NoiseHandshakeInitiation? {
guard peerID.isValid else {
SecureLogger.warning(.authenticationFailed(peerID: peerID.id))
throw NoiseSecurityError.invalidPeerID
}
guard let initiation = try sessionManager.initiateHandshakeIfAbsent(
with: peerID,
notifyOnTimeout: retryOnTimeout,
authorize: { [rateLimiter] in
guard rateLimiter.allowHandshake(from: peerID) else {
SecureLogger.warning(
.authenticationFailed(peerID: "Rate limited: \(peerID)")
)
throw NoiseSecurityError.rateLimitExceeded
}
}
) else {
return nil
}
SecureLogger.info(.handshakeStarted(peerID: peerID.id))
return initiation
}
/// Atomically prepares an ordinary reconnect for a peer whose cached
/// transport belongs to an earlier physical link. Failed authorization or
/// handshake setup preserves the established session.
func initiateReconnectHandshake(
with peerID: PeerID
with peerID: PeerID,
retryOnTimeout: Bool = false
) throws -> NoiseHandshakeInitiation {
guard peerID.isValid else {
SecureLogger.warning(.authenticationFailed(peerID: peerID.id))
@@ -711,6 +758,7 @@ final class NoiseEncryptionService {
return try sessionManager.initiateReconnectHandshake(
with: peerID,
notifyOnTimeout: retryOnTimeout,
authorize: { [rateLimiter] in
guard rateLimiter.allowHandshake(from: peerID) else {
SecureLogger.warning(
@@ -722,6 +770,28 @@ final class NoiseEncryptionService {
)
}
func prepareHandshakeRecovery(
_ request: NoiseHandshakeRecoveryRequest
) throws -> NoiseHandshakeRecoveryPreparation? {
try sessionManager.prepareHandshakeRecovery(
request,
authorizeAttempt: { [rateLimiter] in
guard rateLimiter.allowHandshake(from: request.peerID) else {
SecureLogger.warning(
.authenticationFailed(
peerID: "Rate limited: \(request.peerID)"
)
)
throw NoiseSecurityError.rateLimitExceeded
}
}
)
}
func cancelHandshakeRecovery(_ request: NoiseHandshakeRecoveryRequest) {
sessionManager.cancelHandshakeRecovery(request)
}
func claimHandshakeInitiation(
_ initiation: NoiseHandshakeInitiation,
for peerID: PeerID
@@ -991,9 +1061,9 @@ final class NoiseEncryptionService {
}
private func initiateAutomaticRekey(for peerID: PeerID) throws {
let handshakeMessage = try sessionManager.initiateRekey(for: peerID)
let initiation = try sessionManager.initiateRekey(for: peerID)
SecureLogger.debug("Key rotation initiated for peer: \(peerID)", category: .security)
onRekeyHandshakeReady?(peerID, handshakeMessage)
onRekeyHandshakeReady?(peerID, initiation)
onHandshakeRequired?(peerID)
}