Don't rotate PQ keys if data path is idle for ~90s (less than a WG handhshake time

This commit is contained in:
riccardom
2026-07-31 10:49:27 +02:00
parent 46c12f3d01
commit ba2d9abbdc
6 changed files with 79 additions and 11 deletions
+2 -2
View File
@@ -52,8 +52,8 @@ func TestManager_RekeyToleratesKFailures(t *testing.T) {
// Establish: bootstrap + data-path-rekeyed so B becomes established and its data
// path is usable.
bootstrap(t, dA, dB)
dA.OnDataPathRekeyed("bbbb")
dB.OnDataPathRekeyed("aaaa")
dA.OnDataPathRekeyed("bbbb", 0)
dB.OnDataPathRekeyed("aaaa", 0)
require.NotEqual(t, PSK{}, wgB.psk("aaaa"))
// Drop B's outbound so rekeys can no longer converge.
+19 -1
View File
@@ -20,6 +20,14 @@ const (
// DefaultMaxRekeyFailures is how many consecutive rekey (non-initial) failures
// are tolerated before OnRekeyFailed. The initial exchange fails immediately.
DefaultMaxRekeyFailures = 3
// rotationActivityWindow gates rotation on recent real-data activity: a rekey
// clocks a rotation only if the peer exchanged user data within this window. It
// must stay shorter than the data path's rekey interval (WireGuard
// REKEY_AFTER_TIME ~120s) so the rotation's own traffic — which itself renews the
// activity signal — ages out before the next rekey, letting an idle tunnel stop
// rotating instead of self-sustaining.
rotationActivityWindow = 90 * time.Second
)
// LocalID and RemoteID are peer identity keys (e.g. WireGuard public keys). They are
@@ -319,7 +327,17 @@ func (m *Manager) OnDataPathMessage(remoteID RemoteID, raw []byte) error {
// initiator that just derived a PSK, it 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 RemoteID) {
// OnDataPathRekeyed clocks the next chained PSK rotation on a fresh data-path rekey.
// sinceActivity is how long ago the peer last exchanged real user data; when it
// exceeds rotationActivityWindow the tunnel is treated as idle and rotation is
// skipped — an idle tunnel has nothing to protect, and rotating would emit data-path
// traffic that keeps the peer artificially active (see conn.onWGCheckSuccess).
func (m *Manager) OnDataPathRekeyed(remoteID RemoteID, sinceActivity time.Duration) {
if sinceActivity >= rotationActivityWindow {
m.trace("pqkem: peer idle, skipping data-path rotation", "peer", remoteID, "since_activity", sinceActivity)
return
}
m.mu.Lock()
ex := m.exchanges[remoteID]
chain := ex != nil && ex.state == stateAwaitingRekey
+26 -2
View File
@@ -143,8 +143,8 @@ func TestManager_ChainRotatesAndAcks(t *testing.T) {
// Data path up: B (initiator) chains the next offer over the data path, which
// rotates both to a fresh PSK and acknowledges A.
dA.OnDataPathRekeyed("bbbb")
dB.OnDataPathRekeyed("aaaa")
dA.OnDataPathRekeyed("bbbb", 0)
dB.OnDataPathRekeyed("aaaa", 0)
psk2A := wgA.psk("bbbb")
psk2B := wgB.psk("aaaa")
@@ -152,6 +152,30 @@ func TestManager_ChainRotatesAndAcks(t *testing.T) {
require.NotEqual(t, psk1, psk2B, "the chain rotated to a new PSK")
}
func TestManager_RotationSkippedWhenIdle(t *testing.T) {
dA, dB, wgA, wgB, _ := pair(t)
defer dA.Stop()
defer dB.Stop()
bootstrap(t, dA, dB)
psk1 := wgB.psk("aaaa")
require.NotEqual(t, PSK{}, psk1)
// Idle: the peer's last real-data activity is older than the window, so a rekey
// must NOT clock a rotation.
dA.OnDataPathRekeyed("bbbb", rotationActivityWindow)
dB.OnDataPathRekeyed("aaaa", rotationActivityWindow)
require.Equal(t, psk1, wgB.psk("aaaa"), "idle peer must not rotate the PSK")
require.Equal(t, psk1, wgA.psk("bbbb"), "idle peer must not rotate the PSK")
// Active: activity within the window clocks the rotation as usual.
dA.OnDataPathRekeyed("bbbb", rotationActivityWindow-1)
dB.OnDataPathRekeyed("aaaa", rotationActivityWindow-1)
psk2 := wgB.psk("aaaa")
require.NotEqual(t, psk1, psk2, "recent activity must clock a rotation")
require.Equal(t, psk2, wgA.psk("bbbb"), "both sides converge on the rotated PSK")
}
func TestManager_NonInitiatorReturnsNoOffer(t *testing.T) {
dA := NewManager("aaaa", newFakeWG(), nil)
defer dA.Stop()