Files
netbird/client/internal/peer/endpoint.go
riccardom b5a72eca65 Introduces a forced WG handshake on initial MLKEM bootstrap.
To ensure two peers agree on a key, we need asymmetry. one peer is
the controller ("initiator") the other is the "responder".

Otherwise imagine two offers in parallel driving two answers at the same time

   A                   B
   | <----B-OFFER----- |
   | -----A-OFFER----> |
   |                   |
   |                   |
   ---------------------------------
  |****** ICE + WG Handshake ****** |
   ---------------------------------
   |                   |
   | <----B-ANSWER---- |
   | -----A-ANSWER---> |

PSK is derived on receive of offer, so A and B derive different PSKs.
When WG handshake takes place it picks misaligned PSKs.

So we impair the two nodes and only the offer of one of the two (the controller/initiator)
carries the KEM material.

This means that if the responder OFFER/ANSWER comes first, when the controller/initiator's one
completes (and the genuine PSK is shared between A and B, we need to force a new WG handshake with
the proper keys.
2026-08-06 14:42:07 +02:00

172 lines
4.6 KiB
Go

package peer
import (
"context"
"net"
"sync"
"time"
"github.com/sirupsen/logrus"
"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
)
const (
defaultWgKeepAlive = 25 * time.Second
fallbackDelay = 5 * time.Second
)
type EndpointUpdater struct {
log *logrus.Entry
wgConfig WgConfig
initiator bool
// mu protects cancelFunc and lastEndpoint
mu sync.Mutex
cancelFunc func()
updateWg sync.WaitGroup
// lastEndpoint is the most recent non-nil endpoint applied to the peer, used to
// re-add it on a forced re-handshake (ForceRehandshake).
lastEndpoint *net.UDPAddr
}
func NewEndpointUpdater(log *logrus.Entry, wgConfig WgConfig, initiator bool) *EndpointUpdater {
return &EndpointUpdater{
log: log,
wgConfig: wgConfig,
initiator: initiator,
}
}
func (e *EndpointUpdater) ConfigureWGEndpoint(addr *net.UDPAddr, presharedKey *wgtypes.Key) error {
e.mu.Lock()
defer e.mu.Unlock()
if addr != nil {
e.lastEndpoint = addr
}
if e.initiator {
e.log.Debugf("configure up WireGuard as initiator")
return e.configureAsInitiator(addr, presharedKey)
}
e.log.Debugf("configure up WireGuard as responder")
return e.configureAsResponder(addr, presharedKey)
}
func (e *EndpointUpdater) SwitchWGEndpoint(addr *net.UDPAddr, presharedKey *wgtypes.Key) error {
e.mu.Lock()
defer e.mu.Unlock()
if addr != nil {
e.lastEndpoint = addr
}
// prevent to run new update while cancel the previous update
e.waitForCloseTheDelayedUpdate()
return e.updateWireGuardPeer(addr, presharedKey)
}
// ForceRehandshake removes and re-adds the peer so WireGuard drops the current session
// and negotiates a fresh one with the given PSK. Used when a post-quantum PSK is
// bootstrapped after the session already came up on a pre-PQ key. No-op if no endpoint
// has been applied yet (the pending config will pull the PSK itself).
func (e *EndpointUpdater) ForceRehandshake(presharedKey *wgtypes.Key) error {
e.mu.Lock()
defer e.mu.Unlock()
if e.lastEndpoint == nil {
return nil
}
// Cancel any pending delayed responder update: it carries the stale pre-PQ key and
// would otherwise re-poison the session after we reset it.
e.waitForCloseTheDelayedUpdate()
if err := e.wgConfig.WgInterface.RemovePeer(e.wgConfig.RemoteKey); err != nil {
return err
}
return e.updateWireGuardPeer(e.lastEndpoint, presharedKey)
}
func (e *EndpointUpdater) RemoveWgPeer() error {
e.mu.Lock()
defer e.mu.Unlock()
e.waitForCloseTheDelayedUpdate()
return e.wgConfig.WgInterface.RemovePeer(e.wgConfig.RemoteKey)
}
func (e *EndpointUpdater) RemoveEndpointAddress() error {
e.mu.Lock()
defer e.mu.Unlock()
e.waitForCloseTheDelayedUpdate()
return e.wgConfig.WgInterface.RemoveEndpointAddress(e.wgConfig.RemoteKey)
}
func (e *EndpointUpdater) configureAsInitiator(addr *net.UDPAddr, presharedKey *wgtypes.Key) error {
if err := e.updateWireGuardPeer(addr, presharedKey); err != nil {
return err
}
return nil
}
func (e *EndpointUpdater) configureAsResponder(addr *net.UDPAddr, presharedKey *wgtypes.Key) error {
// prevent to run new update while cancel the previous update
e.waitForCloseTheDelayedUpdate()
e.log.Debugf("configure up WireGuard and wait for handshake")
var ctx context.Context
ctx, e.cancelFunc = context.WithCancel(context.Background())
e.updateWg.Add(1)
go e.scheduleDelayedUpdate(ctx, addr, presharedKey)
if err := e.updateWireGuardPeer(nil, presharedKey); err != nil {
e.waitForCloseTheDelayedUpdate()
return err
}
return nil
}
func (e *EndpointUpdater) waitForCloseTheDelayedUpdate() {
if e.cancelFunc == nil {
return
}
e.cancelFunc()
e.cancelFunc = nil
e.updateWg.Wait()
}
// scheduleDelayedUpdate waits for the fallback period before updating the endpoint
func (e *EndpointUpdater) scheduleDelayedUpdate(ctx context.Context, addr *net.UDPAddr, presharedKey *wgtypes.Key) {
defer e.updateWg.Done()
t := time.NewTimer(fallbackDelay)
defer t.Stop()
select {
case <-ctx.Done():
return
case <-t.C:
if err := e.updateWireGuardPeer(addr, presharedKey); err != nil {
e.log.Errorf("failed to update WireGuard peer, address: %s, error: %v", addr, err)
}
}
}
func (e *EndpointUpdater) updateWireGuardPeer(endpoint *net.UDPAddr, presharedKey *wgtypes.Key) error {
return e.wgConfig.WgInterface.UpdatePeer(
e.wgConfig.RemoteKey,
e.wgConfig.AllowedIps,
defaultWgKeepAlive,
endpoint,
presharedKey,
)
}
// wgConfigWorkaround is a workaround for the issue with WireGuard configuration update
// When update a peer configuration in near to each other time, the second update can be ignored by WireGuard
func wgConfigWorkaround() {
time.Sleep(100 * time.Millisecond)
}