mirror of
https://github.com/netbirdio/netbird.git
synced 2026-07-28 08:52:37 -04:00
Removes confirm. Uses next offer to deliver confirmation/ack of previous round
We clock the next Offer initiation to the OnDataPathRekeyed, so we have 2 minutes ahead of us to do our attempts and stuff before to give up. On failure, we will know because we will not receive a new answer.. but more importantly the wg handshake will fail :D
This commit is contained in:
@@ -5,11 +5,11 @@ import (
|
||||
"time"
|
||||
)
|
||||
|
||||
// startExchange creates a fresh initiator exchange and returns the framed offer for
|
||||
// the caller to send (pushed over the data path for a rekey, or handed to the host
|
||||
// for the signalling channel when viaSignal is set). Any previous in-flight exchange
|
||||
// for the peer is cancelled.
|
||||
func (m *Manager) startExchange(remoteID string, viaSignal bool) ([]byte, error) {
|
||||
// startExchange creates a fresh initiator exchange (acknowledging ackID, zero for a
|
||||
// bootstrap) and returns the framed offer for the caller to send — pushed over the
|
||||
// data path for a chained rekey, or handed to the host for signalling when viaSignal
|
||||
// is set. Any previous in-flight exchange for the peer is cancelled.
|
||||
func (m *Manager) startExchange(remoteID string, viaSignal bool, ackID ExchangeID) ([]byte, error) {
|
||||
init, err := NewInitiator()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
@@ -18,7 +18,7 @@ func (m *Manager) startExchange(remoteID string, viaSignal bool) ([]byte, error)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
raw, err := (&OfferMsg{ExchangeID: id, KEMOffer: init.Offer()}).Encode()
|
||||
raw, err := (&OfferMsg{ExchangeID: id, AckID: ackID, KEMOffer: init.Offer()}).Encode()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
@@ -44,10 +44,15 @@ func (m *Manager) startExchange(remoteID string, viaSignal bool) ([]byte, error)
|
||||
return raw, nil
|
||||
}
|
||||
|
||||
// processOffer (responder) derives the PSK, commits it optimistically, and returns
|
||||
// the framed answer for the caller to send. A duplicate offer (same exchangeID)
|
||||
// returns the cached answer without re-deriving; a still-reserved slot returns nil.
|
||||
// processOffer (responder) first acknowledges the previous exchange the offer names
|
||||
// (that offer riding the data path under the freshly adopted key proves it worked),
|
||||
// then derives the PSK for the new offer, commits it optimistically, and returns the
|
||||
// framed answer. A duplicate offer returns the cached answer without re-deriving.
|
||||
func (m *Manager) processOffer(remoteID string, o *OfferMsg) ([]byte, error) {
|
||||
if o.AckID != (ExchangeID{}) {
|
||||
m.ackConverged(remoteID, o.AckID)
|
||||
}
|
||||
|
||||
m.mu.Lock()
|
||||
if ex := m.exchanges[remoteID]; ex != nil && ex.id == o.ExchangeID {
|
||||
state, last := ex.state, ex.lastSent
|
||||
@@ -76,7 +81,7 @@ func (m *Manager) processOffer(remoteID string, o *OfferMsg) ([]byte, error) {
|
||||
m.mu.Unlock()
|
||||
return nil, nil
|
||||
}
|
||||
ex.state = stateAwaitingConfirm
|
||||
ex.state = stateAwaitingAck
|
||||
ex.lastSent = raw
|
||||
ex.pendingPSK = psk
|
||||
m.mu.Unlock()
|
||||
@@ -88,9 +93,9 @@ func (m *Manager) processOffer(remoteID string, o *OfferMsg) ([]byte, error) {
|
||||
return raw, nil
|
||||
}
|
||||
|
||||
// processAnswer (initiator) derives and commits the PSK, then parks in
|
||||
// stateAwaitingRekey; the confirm is sent later, over the data path, from
|
||||
// OnDataPathRekeyed. Only valid in stateAwaitingAnswer; advancing the state under the
|
||||
// processAnswer (initiator) derives and commits the PSK and parks in
|
||||
// stateAwaitingRekey; the next offer (chained from OnDataPathRekeyed) will acknowledge
|
||||
// this exchange. Only valid in stateAwaitingAnswer; advancing the state under the
|
||||
// lock makes a concurrent/duplicate answer bail.
|
||||
func (m *Manager) processAnswer(remoteID string, a *AnswerMsg) error {
|
||||
m.mu.Lock()
|
||||
@@ -108,40 +113,45 @@ func (m *Manager) processAnswer(remoteID string, a *AnswerMsg) error {
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
// The initiator has converged: the responder must have derived the key to answer.
|
||||
m.mu.Lock()
|
||||
m.established[remoteID] = true
|
||||
m.failures[remoteID] = 0
|
||||
m.mu.Unlock()
|
||||
|
||||
return m.cbHandler.OnNewPSKReady(remoteID, psk)
|
||||
}
|
||||
|
||||
// processConfirm (responder) records convergence. The PSK was already committed in
|
||||
// processOffer; a confirm arriving over the data path with a matching exchangeID
|
||||
// proves we operate on the new key from this exchange. Stale/duplicate confirms find
|
||||
// the exchange gone and are ignored.
|
||||
func (m *Manager) processConfirm(remoteID string, c *ConfirmMsg) error {
|
||||
// ackConverged (responder) records convergence of the exchange named by ackID: a
|
||||
// later offer acknowledging it proves both sides operate on that exchange's key. Only
|
||||
// acts on a matching stateAwaitingAck exchange; anything else is ignored.
|
||||
func (m *Manager) ackConverged(remoteID string, ackID ExchangeID) {
|
||||
m.mu.Lock()
|
||||
ex := m.exchanges[remoteID]
|
||||
if ex == nil || ex.id != c.ExchangeID || ex.state != stateAwaitingConfirm {
|
||||
if ex == nil || ex.id != ackID || ex.state != stateAwaitingAck {
|
||||
m.mu.Unlock()
|
||||
return nil
|
||||
return
|
||||
}
|
||||
delete(m.exchanges, remoteID)
|
||||
m.established[remoteID] = true
|
||||
m.failures[remoteID] = 0
|
||||
_ = time.Since(ex.startedAt) // convergence latency (metrics hook, later step)
|
||||
m.mu.Unlock()
|
||||
return nil
|
||||
}
|
||||
|
||||
// initiatorLoop enforces the convergence deadline and retransmits the initiator's
|
||||
// outstanding DATA-PATH message: it resends the offer while awaiting the answer only
|
||||
// when the offer went over the data path (a signalling-bootstrapped offer is
|
||||
// retransmitted by the host with its own negotiation); it waits, counting toward the
|
||||
// deadline, while awaiting the data-path rekey; it resends the confirm a few times
|
||||
// once sent. Exhausting the deadline before the confirm phase is a failure.
|
||||
// outstanding data-path offer while awaiting the answer (a signalling-bootstrapped
|
||||
// offer is retransmitted by the host, so it is not resent here). It then waits,
|
||||
// counting toward the deadline, in stateAwaitingRekey until OnDataPathRekeyed chains
|
||||
// the next exchange (which supersedes and cancels this loop). Exhausting the deadline
|
||||
// is a failure.
|
||||
func (m *Manager) initiatorLoop(ctx context.Context, remoteID string, id ExchangeID) {
|
||||
defer m.wait.Done()
|
||||
t := time.NewTicker(m.retryInterval)
|
||||
defer t.Stop()
|
||||
|
||||
attempts, confirmsSent := 0, 0
|
||||
attempts := 0
|
||||
for {
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
@@ -174,8 +184,8 @@ func (m *Manager) initiatorLoop(ctx context.Context, remoteID string, id Exchang
|
||||
}
|
||||
|
||||
case stateAwaitingRekey:
|
||||
// Waiting for OnDataPathRekeyed; nothing to send, but the deadline
|
||||
// still applies (the data path may never come up with the new key).
|
||||
// Waiting for OnDataPathRekeyed to chain the next exchange; the
|
||||
// deadline still applies (the data path may never adopt the new key).
|
||||
if attempts >= m.maxRetries {
|
||||
delete(m.exchanges, remoteID)
|
||||
fail := m.registerFailureLocked(remoteID)
|
||||
@@ -186,19 +196,6 @@ func (m *Manager) initiatorLoop(ctx context.Context, remoteID string, id Exchang
|
||||
attempts++
|
||||
m.mu.Unlock()
|
||||
|
||||
case stateConfirming:
|
||||
if confirmsSent >= confirmRetransmits {
|
||||
delete(m.exchanges, remoteID)
|
||||
m.mu.Unlock()
|
||||
return
|
||||
}
|
||||
msg := ex.lastSent
|
||||
confirmsSent++
|
||||
m.mu.Unlock()
|
||||
if err := m.pushDataPath(remoteID, msg); err != nil {
|
||||
m.logger.Warn("pqkem confirm retransmit failed", "peer", remoteID, "err", err)
|
||||
}
|
||||
|
||||
default:
|
||||
m.mu.Unlock()
|
||||
return
|
||||
|
||||
@@ -30,10 +30,9 @@ func (g *gate) SendDataPath(remoteID string, msg []byte) error {
|
||||
|
||||
func TestManager_InitialTimeoutFailsImmediately(t *testing.T) {
|
||||
wg := newFakeWG()
|
||||
d := NewManager("bbbb", dropTransport{}, wg, time.Hour, nil) // bbbb > aaaa -> initiator
|
||||
d := NewManager("bbbb", dropTransport{}, wg, nil) // bbbb > aaaa -> initiator
|
||||
d.retryInterval = 5 * time.Millisecond
|
||||
d.maxRetries = 3
|
||||
d.AddPeer("aaaa")
|
||||
defer d.Stop()
|
||||
|
||||
// Bootstrap offer is produced for signalling; no answer ever comes back -> the
|
||||
@@ -55,41 +54,38 @@ func TestManager_RekeyToleratesKFailures(t *testing.T) {
|
||||
wgA := newFakeWG()
|
||||
wgB := newFakeWG()
|
||||
|
||||
dA := NewManager("aaaa", gA, wgA, time.Hour, nil)
|
||||
dB := NewManager("bbbb", gB, wgB, time.Hour, nil)
|
||||
dA := NewManager("aaaa", gA, wgA, nil)
|
||||
dB := NewManager("bbbb", gB, wgB, nil)
|
||||
gA.peer = dB
|
||||
gB.peer = dA
|
||||
dB.retryInterval = 5 * time.Millisecond
|
||||
dB.maxRetries = 2
|
||||
dA.AddPeer("bbbb")
|
||||
dB.AddPeer("aaaa")
|
||||
defer dA.Stop()
|
||||
defer dB.Stop()
|
||||
|
||||
// Establish via a signalling bootstrap + data-path-rekeyed, so B becomes
|
||||
// established and its data path is up.
|
||||
// Bootstrap over signalling -> B becomes established.
|
||||
offer, err := dB.SignalOffer("aaaa")
|
||||
require.NoError(t, err)
|
||||
answer, err := dA.SignalOnOffer("bbbb", offer)
|
||||
require.NoError(t, err)
|
||||
require.NoError(t, dB.SignalOnAnswer("aaaa", answer))
|
||||
dA.OnDataPathRekeyed("bbbb")
|
||||
dB.OnDataPathRekeyed("aaaa")
|
||||
require.NotEqual(t, PSK{}, wgB.psk("aaaa"))
|
||||
|
||||
// Now drop B's data-path delivery: rekeys can no longer converge.
|
||||
// Bring the data path up on both, then drop B's delivery so rekeys can't converge.
|
||||
dA.OnDataPathRekeyed("bbbb")
|
||||
dB.OnDataPathRekeyed("aaaa")
|
||||
gB.drop.Store(true)
|
||||
|
||||
// K-1 data-path rekeys must NOT raise OnRekeyFailed.
|
||||
for i := 0; i < DefaultMaxRekeyFailures-1; i++ {
|
||||
_, err := dB.startExchange("aaaa", false)
|
||||
_, err := dB.startExchange("aaaa", false, ExchangeID{})
|
||||
require.NoError(t, err)
|
||||
time.Sleep(50 * time.Millisecond)
|
||||
}
|
||||
require.Equal(t, 0, failedCount(wgB), "no failure before K attempts")
|
||||
|
||||
// The K-th failure raises it once.
|
||||
_, err = dB.startExchange("aaaa", false)
|
||||
_, err = dB.startExchange("aaaa", false, ExchangeID{})
|
||||
require.NoError(t, err)
|
||||
require.Eventually(t, func() bool { return failedCount(wgB) == 1 }, time.Second, 5*time.Millisecond)
|
||||
}
|
||||
|
||||
@@ -10,12 +10,8 @@ import (
|
||||
)
|
||||
|
||||
const (
|
||||
// DefaultRekeyInterval is the default PSK rotation cadence (~2 min), chosen so a
|
||||
// rotated PSK is adopted by the consumer's next transport handshake without
|
||||
// forcing one.
|
||||
DefaultRekeyInterval = 2 * time.Minute
|
||||
// DefaultRetryInterval is how often the initiator retransmits its outstanding
|
||||
// data-path message while an exchange is in flight.
|
||||
// data-path offer while awaiting the answer.
|
||||
DefaultRetryInterval = 2 * time.Second
|
||||
// DefaultMaxRetries bounds how many ticks an exchange may run before it is
|
||||
// declared failed. The convergence deadline is thus MaxRetries * RetryInterval.
|
||||
@@ -23,16 +19,12 @@ const (
|
||||
// DefaultMaxRekeyFailures is how many consecutive rekey (non-initial) failures
|
||||
// are tolerated before OnRekeyFailed. The initial exchange fails immediately.
|
||||
DefaultMaxRekeyFailures = 3
|
||||
// confirmRetransmits is how many times the initiator best-effort resends the
|
||||
// confirm over the data path, to cover its loss.
|
||||
confirmRetransmits = 3
|
||||
)
|
||||
|
||||
// Transport pushes a message over the peer's data path (e.g. a WireGuard tunnel).
|
||||
// It is the library's only outbound send: the control-plane (signalling) channel is
|
||||
// host-driven — the library hands the host offer/answer payloads to piggyback on the
|
||||
// host's own offer/answer, it never pushes there itself (that would drive the host's
|
||||
// connection negotiation, which is not the library's to control).
|
||||
// host's own negotiation, it never pushes there itself.
|
||||
type Transport interface {
|
||||
SendDataPath(remoteID string, msg []byte) error
|
||||
}
|
||||
@@ -41,20 +33,19 @@ type Transport interface {
|
||||
type exchangeState uint8
|
||||
|
||||
const (
|
||||
stateReserved exchangeState = iota // responder: deriving the answer
|
||||
stateAwaitingAnswer // initiator: offer sent, awaiting the answer
|
||||
stateAwaitingRekey // initiator: PSK derived+set, awaiting OnDataPathRekeyed to send the confirm
|
||||
stateConfirming // initiator: confirm sent over the data path, best-effort retransmit
|
||||
stateAwaitingConfirm // responder: answer sent, awaiting the confirm over the data path
|
||||
stateReserved exchangeState = iota // responder: deriving the answer
|
||||
stateAwaitingAnswer // initiator: offer sent, awaiting the answer
|
||||
stateAwaitingRekey // initiator: PSK derived+set, awaiting OnDataPathRekeyed to chain the next offer
|
||||
stateAwaitingAck // responder: answer sent, awaiting the next offer that acks this exchange
|
||||
)
|
||||
|
||||
// exchangeCtl holds all state for one in-flight exchange with a peer, under the
|
||||
// Manager's single lock. state drives every decision. lastSent is the current
|
||||
// data-path retransmit payload. initiator is the ephemeral handle used at Finish;
|
||||
// pendingPSK is the responder's derived key. viaSignal records that the offer was
|
||||
// handed to the host for the signalling channel (bootstrap), so the loop does not
|
||||
// retransmit it on the data path (the host retransmits it with its own negotiation).
|
||||
// Only the initiator runs a retransmit loop, so only it sets cancel.
|
||||
// data-path retransmit payload (the offer, for the initiator). initiator is the
|
||||
// ephemeral handle used at Finish; pendingPSK is the responder's derived key.
|
||||
// viaSignal records that the offer went to the host for the signalling channel, so
|
||||
// the loop does not retransmit it on the data path. Only the initiator runs a
|
||||
// retransmit loop, so only it sets cancel.
|
||||
type exchangeCtl struct {
|
||||
id ExchangeID
|
||||
state exchangeState
|
||||
@@ -67,16 +58,17 @@ type exchangeCtl struct {
|
||||
}
|
||||
|
||||
// Manager is the stateful orchestrator — the analogue of go-rosenpass's Server. It
|
||||
// runs the per-peer rekey timer, drives the X25519MLKEM768 exchange, and surfaces the
|
||||
// derived PSK and convergence to the host via CallbackHandler. The cryptography is
|
||||
// the pure kem.go primitives; all state lives here under one lock.
|
||||
// drives the X25519MLKEM768 exchange and surfaces the derived PSK and convergence to
|
||||
// the host via CallbackHandler. It is fully event-driven: the bootstrap is triggered
|
||||
// by the host (SignalOffer) and each rotation is clocked by OnDataPathRekeyed (the
|
||||
// consumer's transport rekey). The cryptography is the pure kem.go primitives; all
|
||||
// state lives here under one lock.
|
||||
type Manager struct {
|
||||
localID string
|
||||
transport Transport
|
||||
cbHandler CallbackHandler
|
||||
logger *slog.Logger
|
||||
|
||||
rekeyInterval time.Duration
|
||||
retryInterval time.Duration
|
||||
maxRetries int
|
||||
maxRekeyFailures int
|
||||
@@ -85,24 +77,20 @@ type Manager struct {
|
||||
rootCancel context.CancelFunc
|
||||
|
||||
mu sync.Mutex
|
||||
peers map[string]context.CancelFunc // per-peer rekey loop
|
||||
exchanges map[string]*exchangeCtl // in-flight exchange per peer
|
||||
established map[string]bool // peer has completed at least one exchange
|
||||
failures map[string]int // consecutive rekey failures per peer
|
||||
// dataSend holds the peer's data-path sender when the data path is up; nil (absent)
|
||||
// means it is down. Toggled by OnDataPathRekeyed / OnDataPathDown. Its presence is
|
||||
// the "a data path exists" signal.
|
||||
exchanges map[string]*exchangeCtl // in-flight exchange per peer
|
||||
established map[string]bool // peer has completed at least one exchange
|
||||
failures map[string]int // consecutive rekey failures per peer
|
||||
// dataSend holds the peer's data-path sender when the data path is up; nil
|
||||
// (absent) means it is down. Toggled by OnDataPathRekeyed / OnDataPathDown.
|
||||
dataSend map[string]func(string, []byte) error
|
||||
wait sync.WaitGroup
|
||||
}
|
||||
|
||||
// NewManager builds a manager for the local peer identified by its peer identity key
|
||||
// (used for the deterministic initiator role and the identity binding). A zero
|
||||
// interval falls back to DefaultRekeyInterval; a nil logger to slog.Default().
|
||||
func NewManager(localID string, t Transport, h CallbackHandler, interval time.Duration, logger *slog.Logger) *Manager {
|
||||
if interval <= 0 {
|
||||
interval = DefaultRekeyInterval
|
||||
}
|
||||
// (used for the deterministic initiator role and the identity binding). A nil logger
|
||||
// falls back to slog.Default(). Retry/retries/K use their defaults and can be
|
||||
// overridden before use.
|
||||
func NewManager(localID string, t Transport, h CallbackHandler, logger *slog.Logger) *Manager {
|
||||
if logger == nil {
|
||||
logger = slog.Default()
|
||||
}
|
||||
@@ -112,13 +100,11 @@ func NewManager(localID string, t Transport, h CallbackHandler, interval time.Du
|
||||
transport: t,
|
||||
cbHandler: h,
|
||||
logger: logger,
|
||||
rekeyInterval: interval,
|
||||
retryInterval: DefaultRetryInterval,
|
||||
maxRetries: DefaultMaxRetries,
|
||||
maxRekeyFailures: DefaultMaxRekeyFailures,
|
||||
rootCtx: ctx,
|
||||
rootCancel: cancel,
|
||||
peers: make(map[string]context.CancelFunc),
|
||||
exchanges: make(map[string]*exchangeCtl),
|
||||
established: make(map[string]bool),
|
||||
failures: make(map[string]int),
|
||||
@@ -133,26 +119,9 @@ func (m *Manager) IsInitiator(remoteID string) bool {
|
||||
return m.localID > remoteID
|
||||
}
|
||||
|
||||
// AddPeer registers a remote peer and starts its rekey timer. Re-adding is a no-op.
|
||||
func (m *Manager) AddPeer(remoteID string) {
|
||||
m.mu.Lock()
|
||||
defer m.mu.Unlock()
|
||||
if _, ok := m.peers[remoteID]; ok {
|
||||
return
|
||||
}
|
||||
ctx, cancel := context.WithCancel(m.rootCtx)
|
||||
m.peers[remoteID] = cancel
|
||||
m.wait.Add(1)
|
||||
go m.rekeyLoop(ctx, remoteID)
|
||||
}
|
||||
|
||||
// RemovePeer stops a peer's rekey timer and any in-flight exchange, and drops state.
|
||||
// RemovePeer stops any in-flight exchange for a peer and drops its state.
|
||||
func (m *Manager) RemovePeer(remoteID string) {
|
||||
m.mu.Lock()
|
||||
if cancel, ok := m.peers[remoteID]; ok {
|
||||
cancel()
|
||||
delete(m.peers, remoteID)
|
||||
}
|
||||
if ex, ok := m.exchanges[remoteID]; ok {
|
||||
if ex.cancel != nil {
|
||||
ex.cancel()
|
||||
@@ -165,23 +134,21 @@ func (m *Manager) RemovePeer(remoteID string) {
|
||||
m.mu.Unlock()
|
||||
}
|
||||
|
||||
// Stop cancels all timers and in-flight exchanges and waits for goroutines to exit.
|
||||
// Stop cancels all in-flight exchanges and waits for their goroutines to exit.
|
||||
func (m *Manager) Stop() {
|
||||
m.rootCancel()
|
||||
m.wait.Wait()
|
||||
m.mu.Lock()
|
||||
m.peers = make(map[string]context.CancelFunc)
|
||||
m.exchanges = make(map[string]*exchangeCtl)
|
||||
m.mu.Unlock()
|
||||
}
|
||||
|
||||
// ---- Signalling channel (host-driven; rides the host's offer/answer) ----
|
||||
// ---- Signalling channel (host-driven; rides the host's negotiation) ----
|
||||
|
||||
// SignalOffer returns the KEM offer for the host to embed in its outgoing offer to
|
||||
// remoteID (bootstrap). It returns (nil, nil) when the local peer is not the
|
||||
// initiator. It is idempotent for an in-flight bootstrap: a repeat call (e.g. the
|
||||
// host retransmitting its offer) returns the same offer rather than starting a new
|
||||
// exchange.
|
||||
// initiator. It is idempotent for an in-flight bootstrap: a repeat call returns the
|
||||
// same offer rather than starting a new exchange.
|
||||
func (m *Manager) SignalOffer(remoteID string) ([]byte, error) {
|
||||
if !m.IsInitiator(remoteID) {
|
||||
return nil, nil
|
||||
@@ -193,7 +160,8 @@ func (m *Manager) SignalOffer(remoteID string) ([]byte, error) {
|
||||
return last, nil
|
||||
}
|
||||
m.mu.Unlock()
|
||||
return m.startExchange(remoteID, true)
|
||||
// bootstrap offer acknowledges nothing (zero AckID).
|
||||
return m.startExchange(remoteID, true, ExchangeID{})
|
||||
}
|
||||
|
||||
// SignalOnOffer processes a KEM offer the host extracted from an incoming offer and
|
||||
@@ -210,7 +178,7 @@ func (m *Manager) SignalOnOffer(remoteID string, offer []byte) ([]byte, error) {
|
||||
}
|
||||
|
||||
// SignalOnAnswer processes a KEM answer the host extracted from an incoming answer.
|
||||
// There is no reply: the confirm rides the data path after OnDataPathRekeyed.
|
||||
// There is no reply: the next offer (over the data path) acknowledges this exchange.
|
||||
func (m *Manager) SignalOnAnswer(remoteID string, answer []byte) error {
|
||||
typ, msg, err := Decode(answer)
|
||||
if err != nil {
|
||||
@@ -243,46 +211,42 @@ func (m *Manager) OnDataPathMessage(remoteID string, raw []byte) error {
|
||||
return m.pushDataPath(remoteID, answer)
|
||||
case MsgAnswer:
|
||||
return m.processAnswer(remoteID, msg.(*AnswerMsg))
|
||||
case MsgConfirm:
|
||||
return m.processConfirm(remoteID, msg.(*ConfirmMsg))
|
||||
default:
|
||||
return fmt.Errorf("unhandled data-path message type %d from %s", typ, remoteID)
|
||||
}
|
||||
}
|
||||
|
||||
// OnDataPathRekeyed notifies that the peer's data path is up and freshly keyed with
|
||||
// the latest PSK (fired on first establishment AND every rekey — the same event). It
|
||||
// marks the data path usable and, if we are the initiator waiting to confirm, sends
|
||||
// the confirm over the data path (its arrival proves to the responder that we operate
|
||||
// on the new key, correlated by exchangeID).
|
||||
// the latest PSK (fired on first establishment AND every rekey). It marks the data
|
||||
// path usable and, if we are the initiator that just derived a PSK, chains the next
|
||||
// exchange: a fresh offer over the data path that acknowledges the just-completed one
|
||||
// (its arrival under the new key proves to the responder that the key works).
|
||||
func (m *Manager) OnDataPathRekeyed(remoteID string) {
|
||||
m.mu.Lock()
|
||||
m.dataSend[remoteID] = m.transport.SendDataPath
|
||||
ex := m.exchanges[remoteID]
|
||||
if ex == nil || ex.state != stateAwaitingRekey {
|
||||
m.mu.Unlock()
|
||||
return
|
||||
chain := ex != nil && ex.state == stateAwaitingRekey
|
||||
var ackID ExchangeID
|
||||
if chain {
|
||||
ackID = ex.id
|
||||
}
|
||||
confirm, err := (&ConfirmMsg{ExchangeID: ex.id}).Encode()
|
||||
if err != nil {
|
||||
m.mu.Unlock()
|
||||
m.logger.Error("pqkem encode confirm", "peer", remoteID, "err", err)
|
||||
return
|
||||
}
|
||||
ex.state = stateConfirming
|
||||
ex.lastSent = confirm
|
||||
m.established[remoteID] = true
|
||||
m.failures[remoteID] = 0
|
||||
_ = time.Since(ex.startedAt) // convergence latency (metrics hook, later step)
|
||||
m.mu.Unlock()
|
||||
|
||||
if err := m.pushDataPath(remoteID, confirm); err != nil {
|
||||
m.logger.Warn("pqkem send confirm failed", "peer", remoteID, "err", err)
|
||||
if !chain {
|
||||
return
|
||||
}
|
||||
offer, err := m.startExchange(remoteID, false, ackID)
|
||||
if err != nil {
|
||||
m.logger.Error("pqkem chain offer failed to start", "peer", remoteID, "err", err)
|
||||
return
|
||||
}
|
||||
if err := m.pushDataPath(remoteID, offer); err != nil {
|
||||
m.logger.Warn("pqkem send chain offer failed", "peer", remoteID, "err", err)
|
||||
}
|
||||
}
|
||||
|
||||
// OnDataPathDown notifies that the peer's data path went down; further rekeys wait
|
||||
// until it is up again (the host re-bootstraps over signalling on reconnect).
|
||||
// OnDataPathDown notifies that the peer's data path went down; rotations pause until
|
||||
// it is up again (the host re-bootstraps over signalling on reconnect).
|
||||
func (m *Manager) OnDataPathDown(remoteID string) {
|
||||
m.mu.Lock()
|
||||
delete(m.dataSend, remoteID)
|
||||
@@ -291,36 +255,6 @@ func (m *Manager) OnDataPathDown(remoteID string) {
|
||||
|
||||
// ---- internals ----
|
||||
|
||||
func (m *Manager) rekeyLoop(ctx context.Context, remoteID string) {
|
||||
defer m.wait.Done()
|
||||
t := time.NewTicker(m.rekeyInterval)
|
||||
defer t.Stop()
|
||||
for {
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return
|
||||
case <-t.C:
|
||||
m.mu.Lock()
|
||||
_, dpUp := m.dataSend[remoteID]
|
||||
_, inFlight := m.exchanges[remoteID]
|
||||
m.mu.Unlock()
|
||||
// Rekeys ride the data path only; skip when it is down (the host will
|
||||
// re-bootstrap over signalling on reconnect) or an exchange is in flight.
|
||||
if !dpUp || inFlight || !m.IsInitiator(remoteID) {
|
||||
continue
|
||||
}
|
||||
offer, err := m.startExchange(remoteID, false)
|
||||
if err != nil {
|
||||
m.logger.Error("pqkem rekey failed to start", "peer", remoteID, "err", err)
|
||||
continue
|
||||
}
|
||||
if err := m.pushDataPath(remoteID, offer); err != nil {
|
||||
m.logger.Warn("pqkem send rekey offer failed", "peer", remoteID, "err", err)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// pushDataPath sends over the peer's data path, erroring if it is down.
|
||||
func (m *Manager) pushDataPath(remoteID string, msg []byte) error {
|
||||
m.mu.Lock()
|
||||
|
||||
@@ -3,14 +3,13 @@ package pqkem
|
||||
import (
|
||||
"sync"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/stretchr/testify/require"
|
||||
)
|
||||
|
||||
// loopback is a data-path transport: a SendDataPath delivers synchronously to the
|
||||
// peer manager's OnDataPathMessage, attributing it to localID (the sender). The
|
||||
// signalling channel is driven by the test directly via the SignalX methods.
|
||||
// loopback is a data-path transport: SendDataPath delivers synchronously to the peer
|
||||
// manager's OnDataPathMessage, attributing it to localID (the sender). The signalling
|
||||
// channel is driven by the test directly via the SignalX methods.
|
||||
type loopback struct {
|
||||
localID string
|
||||
peer *Manager
|
||||
@@ -49,60 +48,75 @@ func (f *fakeWG) psk(peer string) PSK {
|
||||
return f.psks[peer]
|
||||
}
|
||||
|
||||
func TestManager_ExchangeConverges(t *testing.T) {
|
||||
// pair builds two wired managers (B is the initiator, "bbbb" > "aaaa").
|
||||
func pair(t *testing.T) (dA, dB *Manager, wgA, wgB *fakeWG) {
|
||||
t.Helper()
|
||||
lbA := &loopback{localID: "aaaa"}
|
||||
lbB := &loopback{localID: "bbbb"}
|
||||
wgA := newFakeWG()
|
||||
wgB := newFakeWG()
|
||||
wgA = newFakeWG()
|
||||
wgB = newFakeWG()
|
||||
dA = NewManager("aaaa", lbA, wgA, nil)
|
||||
dB = NewManager("bbbb", lbB, wgB, nil)
|
||||
lbA.peer = dB
|
||||
lbB.peer = dA
|
||||
return dA, dB, wgA, wgB
|
||||
}
|
||||
|
||||
dA := NewManager("aaaa", lbA, wgA, time.Hour, nil)
|
||||
dB := NewManager("bbbb", lbB, wgB, time.Hour, nil)
|
||||
lbA.peer = dB // A's data-path sends -> B receives
|
||||
lbB.peer = dA // B's data-path sends -> A receives
|
||||
|
||||
dA.AddPeer("bbbb")
|
||||
dB.AddPeer("aaaa")
|
||||
defer dA.Stop()
|
||||
defer dB.Stop()
|
||||
|
||||
// Bootstrap over the signalling channel (the test plays the host carrying bytes).
|
||||
// B is the initiator ("bbbb" > "aaaa").
|
||||
// bootstrap runs the signalling offer/answer (the test plays the host carrying bytes).
|
||||
func bootstrap(t *testing.T, dA, dB *Manager) {
|
||||
t.Helper()
|
||||
offer, err := dB.SignalOffer("aaaa")
|
||||
require.NoError(t, err)
|
||||
require.NotNil(t, offer)
|
||||
|
||||
answer, err := dA.SignalOnOffer("bbbb", offer)
|
||||
require.NoError(t, err)
|
||||
require.NotNil(t, answer)
|
||||
|
||||
require.NoError(t, dB.SignalOnAnswer("aaaa", answer))
|
||||
}
|
||||
|
||||
// Data path comes up on both sides; this makes B send the confirm over the data
|
||||
// path, converging A.
|
||||
func TestManager_BootstrapDerivesSamePSK(t *testing.T) {
|
||||
dA, dB, wgA, wgB := pair(t)
|
||||
defer dA.Stop()
|
||||
defer dB.Stop()
|
||||
|
||||
bootstrap(t, dA, dB)
|
||||
|
||||
pskA := wgA.psk("bbbb")
|
||||
pskB := wgB.psk("aaaa")
|
||||
require.NotEqual(t, PSK{}, pskA)
|
||||
require.Equal(t, pskB, pskA, "both sides derive the same PSK from the bootstrap exchange")
|
||||
}
|
||||
|
||||
func TestManager_ChainRotatesAndAcks(t *testing.T) {
|
||||
dA, dB, wgA, wgB := pair(t)
|
||||
defer dA.Stop()
|
||||
defer dB.Stop()
|
||||
|
||||
bootstrap(t, dA, dB)
|
||||
psk1 := wgB.psk("aaaa")
|
||||
|
||||
// Data path up on both sides; B chains the next offer (acking exchange 1) over the
|
||||
// data path, which rotates both to a fresh PSK and acknowledges A.
|
||||
dA.OnDataPathRekeyed("bbbb")
|
||||
dB.OnDataPathRekeyed("aaaa")
|
||||
|
||||
pskB := wgB.psk("aaaa")
|
||||
pskA := wgA.psk("bbbb")
|
||||
require.NotEqual(t, PSK{}, pskA, "responder A must have a PSK")
|
||||
require.NotEqual(t, PSK{}, pskB, "initiator B must have a PSK")
|
||||
require.Equal(t, pskB, pskA, "both sides converge on the same PSK")
|
||||
psk2A := wgA.psk("bbbb")
|
||||
psk2B := wgB.psk("aaaa")
|
||||
require.Equal(t, psk2B, psk2A, "both sides converge on the rotated PSK")
|
||||
require.NotEqual(t, psk1, psk2B, "the chain rotated to a new PSK")
|
||||
}
|
||||
|
||||
func TestManager_NonInitiatorReturnsNoOffer(t *testing.T) {
|
||||
dA := NewManager("aaaa", &loopback{localID: "aaaa"}, newFakeWG(), time.Hour, nil)
|
||||
dA.AddPeer("bbbb")
|
||||
dA := NewManager("aaaa", &loopback{localID: "aaaa"}, newFakeWG(), nil)
|
||||
defer dA.Stop()
|
||||
|
||||
// A is NOT the initiator vs "bbbb" -> no offer to send.
|
||||
offer, err := dA.SignalOffer("bbbb")
|
||||
offer, err := dA.SignalOffer("bbbb") // not the initiator vs "bbbb"
|
||||
require.NoError(t, err)
|
||||
require.Nil(t, offer)
|
||||
}
|
||||
|
||||
func TestManager_StopIsIdempotent(t *testing.T) {
|
||||
dA := NewManager("aaaa", &loopback{localID: "aaaa"}, newFakeWG(), time.Hour, nil)
|
||||
dA.AddPeer("bbbb")
|
||||
dA := NewManager("aaaa", &loopback{localID: "aaaa"}, newFakeWG(), nil)
|
||||
dA.Stop()
|
||||
dA.Stop() // must not panic or hang
|
||||
}
|
||||
|
||||
@@ -6,39 +6,45 @@ import (
|
||||
)
|
||||
|
||||
// Wire framing for the PQ-KEM exchange. Messages are self-contained, versioned,
|
||||
// transport-agnostic byte blobs: the same bytes ride the Signal offer/answer
|
||||
// (initial, pre-tunnel) or a data-tunnel packet (rekey). They are NOT a gRPC
|
||||
// service — the network layer only sees opaque []byte.
|
||||
// transport-agnostic byte blobs: the same bytes ride the signalling channel
|
||||
// (initial bootstrap) or a data-tunnel packet (rekey). The library only ever sees
|
||||
// opaque []byte at the transport seam.
|
||||
//
|
||||
// Layout (all messages): [type:1][version:1][exchangeID:16][payload...]
|
||||
//
|
||||
// There is no confirm message: an exchange is acknowledged by the NEXT offer, which
|
||||
// carries the acked exchange's id (see OfferMsg.AckID) and — riding the data path
|
||||
// under the freshly adopted key — proves that key works.
|
||||
|
||||
const (
|
||||
// ProtocolVersion is bumped on any wire-incompatible change; a peer rejects
|
||||
// messages it does not understand rather than misparsing them.
|
||||
ProtocolVersion uint8 = 1
|
||||
|
||||
// ExchangeIDSize identifies one offer/answer/confirm round so stale answers
|
||||
// (e.g. an answer to a pre-restart offer) are dropped instead of applied.
|
||||
// ExchangeIDSize identifies one exchange so answers/acks correlate and stale
|
||||
// messages are dropped.
|
||||
ExchangeIDSize = 16
|
||||
|
||||
headerSize = 1 + 1 + ExchangeIDSize
|
||||
)
|
||||
|
||||
// MsgType tags the three message kinds of the exchange.
|
||||
// MsgType tags the two message kinds of the exchange.
|
||||
type MsgType uint8
|
||||
|
||||
const (
|
||||
MsgOffer MsgType = iota + 1
|
||||
MsgAnswer
|
||||
MsgConfirm
|
||||
)
|
||||
|
||||
// ExchangeID is the per-round correlator echoed by the answer and the confirm.
|
||||
// ExchangeID is the per-exchange correlator. The zero value means "none" (an offer
|
||||
// that acknowledges nothing, i.e. the first exchange of a connection).
|
||||
type ExchangeID [ExchangeIDSize]byte
|
||||
|
||||
// OfferMsg carries the initiator's public material (X25519 pub ‖ ML-KEM encap key).
|
||||
// OfferMsg carries the initiator's public material (X25519 pub ‖ ML-KEM encap key)
|
||||
// and AckID, the id of the previous exchange this offer acknowledges (zero if none).
|
||||
type OfferMsg struct {
|
||||
ExchangeID ExchangeID
|
||||
AckID ExchangeID
|
||||
// KEMOffer is the raw Initiator.Offer() blob (OfferSize bytes).
|
||||
KEMOffer []byte
|
||||
}
|
||||
@@ -51,27 +57,15 @@ type AnswerMsg struct {
|
||||
KEMAnswer []byte
|
||||
}
|
||||
|
||||
// ConfirmMsg is sent ONE WAY, initiator -> responder, over the tunnel under the
|
||||
// NEW PSK, for the round identified by ExchangeID. It carries no key material.
|
||||
//
|
||||
// It is one-way because of the knowledge asymmetry of a KEM: the initiator learns
|
||||
// the responder holds the key simply by receiving the answer (the responder had to
|
||||
// derive it to encapsulate), so the initiator needs no confirmation. Only the
|
||||
// responder is left unsure whether the initiator received the answer and committed
|
||||
// the key — this message resolves that. As a bonus, being sent under the new PSK it
|
||||
// exercises the consumer's channel handshake with the new key: the responder
|
||||
// converges on receiving it, and the initiator converges by observing that handshake
|
||||
// succeed (which fails on a PSK mismatch, so success proves the responder committed too).
|
||||
type ConfirmMsg struct {
|
||||
ExchangeID ExchangeID
|
||||
}
|
||||
|
||||
// Encode serialises the offer with its framed header.
|
||||
// Encode serialises the offer with its framed header (payload = AckID ‖ KEMOffer).
|
||||
func (m *OfferMsg) Encode() ([]byte, error) {
|
||||
if len(m.KEMOffer) != OfferSize {
|
||||
return nil, fmt.Errorf("offer payload: got %d, want %d", len(m.KEMOffer), OfferSize)
|
||||
}
|
||||
return frame(MsgOffer, m.ExchangeID, m.KEMOffer), nil
|
||||
payload := make([]byte, 0, ExchangeIDSize+OfferSize)
|
||||
payload = append(payload, m.AckID[:]...)
|
||||
payload = append(payload, m.KEMOffer...)
|
||||
return frame(MsgOffer, m.ExchangeID, payload), nil
|
||||
}
|
||||
|
||||
// Encode serialises the answer with its framed header.
|
||||
@@ -82,13 +76,7 @@ func (m *AnswerMsg) Encode() ([]byte, error) {
|
||||
return frame(MsgAnswer, m.ExchangeID, m.KEMAnswer), nil
|
||||
}
|
||||
|
||||
// Encode serialises the confirm with its framed header. It returns an error only
|
||||
// for signature uniformity with the other messages; it never actually fails.
|
||||
func (m *ConfirmMsg) Encode() ([]byte, error) {
|
||||
return frame(MsgConfirm, m.ExchangeID, nil), nil
|
||||
}
|
||||
|
||||
// Decode parses a framed message into one of *OfferMsg / *AnswerMsg / *ConfirmMsg.
|
||||
// Decode parses a framed message into one of *OfferMsg / *AnswerMsg.
|
||||
func Decode(buf []byte) (MsgType, any, error) {
|
||||
if len(buf) < headerSize {
|
||||
return 0, nil, fmt.Errorf("message too short: %d bytes", len(buf))
|
||||
@@ -104,20 +92,17 @@ func Decode(buf []byte) (MsgType, any, error) {
|
||||
|
||||
switch typ {
|
||||
case MsgOffer:
|
||||
if len(payload) != OfferSize {
|
||||
return typ, nil, fmt.Errorf("offer payload: got %d, want %d", len(payload), OfferSize)
|
||||
if len(payload) != ExchangeIDSize+OfferSize {
|
||||
return typ, nil, fmt.Errorf("offer payload: got %d, want %d", len(payload), ExchangeIDSize+OfferSize)
|
||||
}
|
||||
return typ, &OfferMsg{ExchangeID: id, KEMOffer: payload}, nil
|
||||
var ack ExchangeID
|
||||
copy(ack[:], payload[:ExchangeIDSize])
|
||||
return typ, &OfferMsg{ExchangeID: id, AckID: ack, KEMOffer: payload[ExchangeIDSize:]}, nil
|
||||
case MsgAnswer:
|
||||
if len(payload) != AnswerSize {
|
||||
return typ, nil, fmt.Errorf("answer payload: got %d, want %d", len(payload), AnswerSize)
|
||||
}
|
||||
return typ, &AnswerMsg{ExchangeID: id, KEMAnswer: payload}, nil
|
||||
case MsgConfirm:
|
||||
if len(payload) != 0 {
|
||||
return typ, nil, fmt.Errorf("confirm payload must be empty, got %d bytes", len(payload))
|
||||
}
|
||||
return typ, &ConfirmMsg{ExchangeID: id}, nil
|
||||
default:
|
||||
return typ, nil, fmt.Errorf("unknown message type %d", typ)
|
||||
}
|
||||
|
||||
@@ -13,13 +13,15 @@ func TestMessageRoundTrip(t *testing.T) {
|
||||
require.NoError(t, err)
|
||||
|
||||
id := ExchangeID{1, 2, 3, 4}
|
||||
ack := ExchangeID{9, 9, 9}
|
||||
|
||||
offBytes, err := (&OfferMsg{ExchangeID: id, KEMOffer: init.Offer()}).Encode()
|
||||
offBytes, err := (&OfferMsg{ExchangeID: id, AckID: ack, KEMOffer: init.Offer()}).Encode()
|
||||
require.NoError(t, err)
|
||||
typ, decoded, err := Decode(offBytes)
|
||||
require.NoError(t, err)
|
||||
require.Equal(t, MsgOffer, typ)
|
||||
require.Equal(t, id, decoded.(*OfferMsg).ExchangeID)
|
||||
require.Equal(t, ack, decoded.(*OfferMsg).AckID)
|
||||
require.Equal(t, init.Offer(), decoded.(*OfferMsg).KEMOffer)
|
||||
|
||||
ansBytes, err := (&AnswerMsg{ExchangeID: id, KEMAnswer: answer}).Encode()
|
||||
@@ -28,13 +30,6 @@ func TestMessageRoundTrip(t *testing.T) {
|
||||
require.NoError(t, err)
|
||||
require.Equal(t, MsgAnswer, typ)
|
||||
require.Equal(t, answer, decoded.(*AnswerMsg).KEMAnswer)
|
||||
|
||||
confBytes, err := (&ConfirmMsg{ExchangeID: id}).Encode()
|
||||
require.NoError(t, err)
|
||||
typ, decoded, err = Decode(confBytes)
|
||||
require.NoError(t, err)
|
||||
require.Equal(t, MsgConfirm, typ)
|
||||
require.Equal(t, id, decoded.(*ConfirmMsg).ExchangeID)
|
||||
}
|
||||
|
||||
func TestDecodeRejects(t *testing.T) {
|
||||
@@ -43,7 +38,7 @@ func TestDecodeRejects(t *testing.T) {
|
||||
require.Error(t, err)
|
||||
|
||||
// wrong version
|
||||
bad := make([]byte, headerSize+OfferSize)
|
||||
bad := make([]byte, headerSize+ExchangeIDSize+OfferSize)
|
||||
bad[0] = byte(MsgOffer)
|
||||
bad[1] = ProtocolVersion + 1
|
||||
_, _, err = Decode(bad)
|
||||
|
||||
Reference in New Issue
Block a user