fix: disable the whole underload if DisableCookies is on

Thanks to @darkk for spotting and suggesting the fix

Dockerfile

@@ -5,14 +5,20 @@ RUN go mod download && \
     go mod verify && \
     go build -ldflags '-linkmode external -extldflags "-fno-PIC -static"' -v -o /usr/bin
 
-FROM alpine:3.19
-ARG AWGTOOLS_RELEASE="1.0.20250901"
+FROM alpine:3.19 as tools
+ARG AWGTOOLS_COMMIT="v3.1.20260812"
 
-RUN apk --no-cache add iproute2 iptables bash && \
-    cd /usr/bin/ && \
-    wget https://github.com/amnezia-vpn/amneziawg-tools/releases/download/v${AWGTOOLS_RELEASE}/alpine-3.19-amneziawg-tools.zip && \
-    unzip -j alpine-3.19-amneziawg-tools.zip && \
-    chmod +x /usr/bin/awg /usr/bin/awg-quick && \
+RUN apk add --no-cache git build-base linux-headers && \
+    git clone https://github.com/amnezia-vpn/amneziawg-tools.git /amneziawg-tools && \
+    cd /amneziawg-tools && git checkout ${AWGTOOLS_COMMIT} && \
+    cd src && make
+
+FROM alpine:3.19
+
+RUN apk --no-cache add iproute2 iptables bash
+COPY --from=tools /amneziawg-tools/src/wg /usr/bin/awg
+COPY --from=tools /amneziawg-tools/src/wg-quick/linux.bash /usr/bin/awg-quick
+RUN chmod +x /usr/bin/awg /usr/bin/awg-quick && \
     ln -s /usr/bin/awg /usr/bin/wg && \
     ln -s /usr/bin/awg-quick /usr/bin/wg-quick
 COPY --from=awg /usr/bin/amneziawg-go /usr/bin/amneziawg-go

README.md

@@ -78,7 +78,7 @@ Header protection is the mechanism of protecting low-entropy values of packets'
 > Use `awg genkey` to generate header protection key
 
 > [!IMPORTANT]
-> Header protection requires `S1-S4` value to be 8 at least
+> Header protection requires `S1-S4` value to be 12 at least
 
 ### Content padding [AWG 3+]
 

conn/bind_windows.go

@@ -582,6 +582,24 @@ func (bind *WinRingBind) BindSocketToInterface6(interfaceIndex uint32, blackhole
 	return nil
 }
 
+func (bind *WinRingBind) PeekLookAtSocketFd4() (fd int, err error) {
+	bind.mu.RLock()
+	defer bind.mu.RUnlock()
+	if bind.isOpen.Load() != 1 {
+		return -1, net.ErrClosed
+	}
+	return int(bind.v4.sock), nil
+}
+
+func (bind *WinRingBind) PeekLookAtSocketFd6() (fd int, err error) {
+	bind.mu.RLock()
+	defer bind.mu.RUnlock()
+	if bind.isOpen.Load() != 1 {
+		return -1, net.ErrClosed
+	}
+	return int(bind.v6.sock), nil
+}
+
 func bindSocketToInterface4(handle windows.Handle, interfaceIndex uint32) error {
 	const IP_UNICAST_IF = 31
 	/* MSDN says for IPv4 this needs to be in net byte order, so that it's like an IP address with leading zeros. */

device/constants.go

@@ -24,6 +24,7 @@ const (
 	CookieRefreshTime       = time.Second * 120
 	HandshakeInitationRate  = time.Second / 50
 	PaddingMultiple         = 16
+	DefaultUdpWindow        = 500
 )
 
 const (

device/device.go

@@ -126,6 +126,9 @@ type Device struct {
 		keepaliveTimeoutSec AtomicUintRange
 		maxHandshakeAttemps AtomicUintRange
 	}
+
+	randomTrailers atomic.Bool
+	disableCookies atomic.Bool
 }
 
 // deviceState represents the state of a Device.

device/peer.go

@@ -57,6 +57,7 @@ type Peer struct {
 	cookieGenerator             CookieGenerator
 	trieEntries                 list.List
 	persistentKeepaliveInterval AtomicUintRange
+	udpWindow                   atomic.Uint32
 }
 
 func (device *Device) NewPeer(pk NoisePublicKey) (*Peer, error) {
@@ -79,6 +80,8 @@ func (device *Device) NewPeer(pk NoisePublicKey) (*Peer, error) {
 	// create peer
 	peer := new(Peer)
 
+	peer.udpWindow.Store(DefaultUdpWindow)
+
 	peer.cookieGenerator.Init(pk)
 	peer.device = device
 	peer.queue.outbound = newAutodrainingOutboundQueue(device)
@@ -283,6 +286,9 @@ func (peer *Peer) SetEndpointFromPacket(endpoint conn.Endpoint) {
 	if peer.endpoint.disableRoaming {
 		return
 	}
+	if peer.endpoint.val != endpoint {
+		peer.udpWindow.Store(DefaultUdpWindow)
+	}
 	peer.endpoint.clearSrcOnTx = false
 	peer.endpoint.val = endpoint
 }

device/receive.go

@@ -154,8 +154,12 @@ func (device *Device) RoutineReceiveIncoming(
 			}
 
 			// get message padding and type based on information from S1-S4 and H1-H4
-			msgType, padding := device.DeterminePacketTypeAndPadding(packet, MessageUnknownType, typeHash)
+			msgSize, msgType, padding := device.DeterminePacketTypeAndPadding(packet, typeHash)
+
 			packet = packet[padding:]
+			if msgType != MessageTransportType {
+				packet = packet[:msgSize]
+			}
 
 			if cip != nil {
 				applyHash(packet[:4], packet[:4], typeHash)
@@ -365,7 +369,8 @@ func (device *Device) RoutineHandshake(id int) {
 
 			// endpoints destination address is the source of the datagram
 
-			if device.IsUnderLoad() {
+			disableCookies := device.disableCookies.Load()
+			if !disableCookies && device.IsUnderLoad() {
 
 				// verify MAC2 field
 
@@ -510,7 +515,12 @@ func (peer *Peer) RoutineSequentialReceiver(maxBatchSize int) {
 			}
 			rxBytesLen += uint64(len(elem.packet) + MinMessageSize)
 
-			if len(elem.packet) == 0 {
+			udpWindow := elem.padding + MessageTransportHeaderSize + uint32(len(elem.packet))
+			if peer.udpWindow.Load() < udpWindow {
+				peer.udpWindow.Store(udpWindow)
+			}
+
+			if len(elem.packet) == 0 || elem.packet[0] == 0 {
 				device.log.Verbosef("%v - Receiving keepalive packet", peer)
 				continue
 			}
@@ -558,7 +568,7 @@ func (peer *Peer) RoutineSequentialReceiver(maxBatchSize int) {
 				continue
 			}
 
-			bufs = append(bufs, elem.buffer[int(elem.padding):int(elem.padding)+MessageTransportOffsetContent+len(elem.packet)])
+			bufs = append(bufs, elem.buffer[int(elem.padding):int(elem.padding)+MessageTransportHeaderSize+len(elem.packet)])
 		}
 
 		peer.rxBytes.Add(rxBytesLen)
@@ -592,57 +602,58 @@ func applyHash(dst, src, hash []byte) {
 	}
 }
 
-func (device *Device) DeterminePacketTypeAndPadding(packet []byte, expectedType uint32, typeHash []byte) (uint32, uint32) {
+func (device *Device) DeterminePacketTypeAndPadding(packet []byte, typeHash []byte) (int, uint32, uint32) {
 	var headerBytes [4]byte
+	var padding uint32
+	var header UintRange
+	var expectedSize int
+
 	size := len(packet)
+	randomTrailers := device.randomTrailers.Load()
 
-	if expectedType == MessageUnknownType || expectedType == MessageInitiationType {
-		padding := device.paddings.init.Load()
-		header := device.headers.init.Load()
+	padding = device.paddings.init.Load()
+	header = device.headers.init.Load()
+	expectedSize = int(padding) + MessageInitiationSize
 
-		if size == int(padding)+MessageInitiationSize {
-			applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
-			if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
-				return MessageInitiationType, padding
-			}
+	if size == expectedSize || randomTrailers && size > expectedSize {
+		applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
+		if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
+			return MessageInitiationSize, MessageInitiationType, padding
 		}
 	}
 
-	if expectedType == MessageUnknownType || expectedType == MessageResponseType {
-		padding := device.paddings.response.Load()
-		header := device.headers.response.Load()
+	padding = device.paddings.response.Load()
+	header = device.headers.response.Load()
+	expectedSize = int(padding) + MessageResponseSize
 
-		if size == int(padding)+MessageResponseSize {
-			applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
-			if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
-				return MessageResponseType, padding
-			}
+	if size == expectedSize || randomTrailers && size > expectedSize {
+		applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
+		if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
+			return MessageResponseSize, MessageResponseType, padding
 		}
 	}
 
-	if expectedType == MessageUnknownType || expectedType == MessageCookieReplyType {
-		padding := device.paddings.cookie.Load()
-		header := device.headers.cookie.Load()
+	padding = device.paddings.cookie.Load()
+	header = device.headers.cookie.Load()
+	expectedSize = int(padding) + MessageCookieReplySize
 
-		if size == int(padding)+MessageCookieReplySize {
-			applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
-			if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
-				return MessageCookieReplyType, padding
-			}
+	if size == expectedSize || randomTrailers && size > expectedSize {
+		applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
+		if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
+			return MessageCookieReplySize, MessageCookieReplyType, padding
 		}
 	}
 
-	if expectedType == MessageUnknownType || expectedType == MessageTransportType {
-		padding := device.paddings.transport.Load()
-		header := device.headers.transport.Load()
+	padding = device.paddings.transport.Load()
+	header = device.headers.transport.Load()
+	expectedSize = int(padding) + MessageTransportSize
 
-		if size >= int(padding)+MessageTransportHeaderSize {
-			applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
-			if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
-				return MessageTransportType, padding
-			}
+	if size >= expectedSize {
+		applyHash(headerBytes[:], packet[padding:padding+4], typeHash)
+		if header.Contains(binary.LittleEndian.Uint32(headerBytes[:])) {
+			return MessageTransportSize, MessageTransportType, padding
 		}
 	}
 
-	return MessageUnknownType, 0
+	return 0, MessageUnknownType, 0
 }

device/send.go

@@ -48,12 +48,13 @@ import (
  */
 
 type QueueOutboundElement struct {
-	buffer  *[MaxMessageSize]byte // slice holding the packet data
-	packet  []byte                // slice of "buffer" (always!)
-	nonce   uint64                // nonce for encryption
-	keypair *Keypair              // keypair for encryption
-	peer    *Peer                 // related peer
-	padding uint32
+	buffer      *[MaxMessageSize]byte // slice holding the packet data
+	packet      []byte                // slice of "buffer" (always!)
+	nonce       uint64                // nonce for encryption
+	keypair     *Keypair              // keypair for encryption
+	peer        *Peer                 // related peer
+	padding     uint32
+	isKeepalive bool
 }
 
 type QueueOutboundElementsContainer struct {
@@ -66,6 +67,7 @@ func (device *Device) NewOutboundElement() *QueueOutboundElement {
 	elem.buffer = device.GetMessageBuffer()
 	elem.nonce = 0
 	elem.padding = device.paddings.transport.Load()
+	elem.isKeepalive = false
 	// keypair and peer were cleared (if necessary) by clearPointers.
 	return elem
 }
@@ -86,6 +88,7 @@ func (elem *QueueOutboundElement) clearPointers() {
 func (peer *Peer) SendKeepalive() {
 	if len(peer.queue.staged) == 0 && peer.isRunning.Load() {
 		elem := peer.device.NewOutboundElement()
+		elem.isKeepalive = true
 		elemsContainer := peer.device.GetOutboundElementsContainer()
 		elemsContainer.elems = append(elemsContainer.elems, elem)
 		select {
@@ -143,8 +146,10 @@ func (peer *Peer) SendHandshakeInitiation(isRetry bool) error {
 
 	sendBuffer = append(sendBuffer, peer.device.JunkPackets()...)
 
-	padding := peer.device.paddings.init.Load()
-	buf := make([]byte, padding+MessageInitiationSize)
+	padding := int(peer.device.paddings.init.Load())
+	trailerLen := max(peer.randomTrailer(padding+MessageInitiationSize), 0)
+
+	buf := make([]byte, padding+MessageInitiationSize+trailerLen)
 
 	crypt := buf[:padding]
 	rand.Read(crypt)
@@ -165,6 +170,9 @@ func (peer *Peer) SendHandshakeInitiation(isRetry bool) error {
 		cip.XORKeyStream(packet, packet)
 	}
 
+	trailer := buf[padding+MessageInitiationSize:]
+	rand.Read(trailer)
+
 	sendBuffer = append(sendBuffer, buf)
 	err = peer.SendBuffers(sendBuffer)
 	if err != nil {
@@ -188,8 +196,10 @@ func (peer *Peer) SendHandshakeResponse() error {
 		return err
 	}
 
-	padding := peer.device.paddings.response.Load()
-	buf := make([]byte, padding+MessageResponseSize)
+	padding := int(peer.device.paddings.response.Load())
+	trailerLen := max(peer.randomTrailer(padding+MessageResponseSize), 0)
+
+	buf := make([]byte, padding+MessageResponseSize+trailerLen)
 
 	crypt := buf[:padding]
 	rand.Read(crypt)
@@ -217,6 +227,9 @@ func (peer *Peer) SendHandshakeResponse() error {
 		cip.XORKeyStream(packet, packet)
 	}
 
+	trailer := buf[padding+MessageResponseSize:]
+	rand.Read(trailer)
+
 	// TODO: allocation could be avoided
 	err = peer.SendBuffers([][]byte{buf})
 	if err != nil {
@@ -242,8 +255,10 @@ func (device *Device) SendHandshakeCookie(initiatingElem *QueueHandshakeElement)
 		return err
 	}
 
-	padding := device.paddings.cookie.Load()
-	buf := make([]byte, padding+MessageCookieReplySize)
+	padding := int(device.paddings.cookie.Load())
+	trailerLen := max(device.randomTrailer(padding+MessageCookieReplySize), 0)
+
+	buf := make([]byte, padding+MessageCookieReplySize+trailerLen)
 
 	crypt := buf[:padding]
 	rand.Read(crypt)
@@ -260,6 +275,9 @@ func (device *Device) SendHandshakeCookie(initiatingElem *QueueHandshakeElement)
 		cip.XORKeyStream(packet, packet)
 	}
 
+	trailer := buf[padding+MessageCookieReplySize:]
+	rand.Read(trailer)
+
 	// TODO: allocation could be avoided
 	device.net.bind.Send([][]byte{buf}, initiatingElem.endpoint)
 	return nil
@@ -507,27 +525,49 @@ func calculatePaddingSize(packetSize, mtu int) int {
 	return paddedSize - lastUnit
 }
 
-func (device *Device) randomPaddingAddition(packetSize, mtu int) int {
-	addition := device.contentPaddingAddition.Load()
+func (peer *Peer) randomPaddingAddition(packetSize int) int {
+	addition := peer.device.contentPaddingAddition.Load()
 
 	if addition.IsZero() {
 		return -1
 	}
 
-	add := int(addition.PickOne())
-	if mtu != 0 {
-		if packetSize > mtu {
-			packetSize %= mtu
-		}
+	udpWindow := int(peer.udpWindow.Load())
+	if udpWindow < packetSize {
+		return 0
+	}
 
-		space := mtu - packetSize
-		if add > space {
-			add = space
-		}
+	add := int(addition.PickOne())
+	space := udpWindow - packetSize
+	if add > space {
+		add = space
 	}
 	return add
 }
 
+func (device *Device) randomTrailer(packetSize int) int {
+	if !device.randomTrailers.Load() {
+		return -1
+	}
+
+	if DefaultUdpWindow < packetSize {
+		return 0
+	}
+	return int(fastrandn(uint32(DefaultUdpWindow - packetSize)))
+}
+
+func (peer *Peer) randomTrailer(packetSize int) int {
+	if !peer.device.randomTrailers.Load() {
+		return -1
+	}
+
+	udpWindow := int(peer.udpWindow.Load())
+	if udpWindow < packetSize {
+		return 0
+	}
+	return int(fastrandn(uint32(udpWindow - packetSize)))
+}
+
 /* Encrypts the elements in the queue
  * and marks them for sequential consumption (by releasing the mutex)
  *
@@ -541,6 +581,11 @@ func (device *Device) RoutineEncryption(id int) {
 
 	for elemsContainer := range device.queue.encryption.c {
 		for _, elem := range elemsContainer.elems {
+			udpWindow := elem.padding + MinMessageSize + uint32(len(elem.packet))
+			if elem.peer.udpWindow.Load() < udpWindow {
+				elem.peer.udpWindow.Store(udpWindow)
+			}
+
 			// fill crypto padding
 			crypt := elem.buffer[:elem.padding]
 			rand.Read(crypt)
@@ -556,13 +601,16 @@ func (device *Device) RoutineEncryption(id int) {
 			binary.LittleEndian.PutUint32(fieldReceiver, elem.keypair.remoteIndex)
 			binary.LittleEndian.PutUint64(fieldNonce, elem.nonce)
 
-			packetSize := len(elem.packet)
+			packetSize := len(elem.packet) + MinMessageSize + int(elem.padding)
 			mtu := int(device.tun.mtu.Load())
 
-			paddingSize := device.randomPaddingAddition(packetSize, mtu)
+			paddingSize := elem.peer.randomPaddingAddition(packetSize)
+			if paddingSize < 0 {
+				paddingSize = elem.peer.randomTrailer(packetSize)
+			}
 			if paddingSize < 0 {
 				// pad content to multiple of 16
-				paddingSize = calculatePaddingSize(packetSize, mtu)
+				paddingSize = calculatePaddingSize(len(elem.packet), mtu)
 			}
 
 			// append trailing zeroes
@@ -628,7 +676,7 @@ func (peer *Peer) RoutineSequentialSender(maxBatchSize int) {
 		dataSent := false
 		elemsContainer.Lock()
 		for _, elem := range elemsContainer.elems {
-			if len(elem.packet) != MessageKeepaliveSize {
+			if !elem.isKeepalive {
 				dataSent = true
 			}
 

device/uapi.go

@@ -70,6 +70,18 @@ func (device *Device) IpcGetOperation(w io.Writer) error {
 		}
 		buf.WriteByte('\n')
 	}
+	boolf := func(prefix string, val bool) {
+		buf.Grow(3 + len(prefix))
+		buf.WriteString(prefix)
+		buf.WriteByte('=')
+		if val {
+			buf.WriteByte('1')
+		} else {
+			buf.WriteByte('0')
+		}
+		buf.WriteByte('\n')
+
+	}
 
 	func() {
 		// lock required resources
@@ -173,6 +185,8 @@ func (device *Device) IpcGetOperation(w io.Writer) error {
 		if rang := device.timings.maxHandshakeAttemps.Load(); !rang.IsZero() {
 			sendf("max_handshake_attempts=%s", rang.ToString())
 		}
+		boolf("random_trailers", device.randomTrailers.Load())
+		boolf("disable_cookies", device.disableCookies.Load())
 
 		for _, peer := range device.peers.keyMap {
 			// Serialize peer state.
@@ -513,6 +527,22 @@ func (device *Device) handleDeviceLine(ipcDev *ipcSetDevice, key, value string)
 		device.log.Verbosef("UAPI: Updating max handshake attempts")
 		device.timings.maxHandshakeAttemps.Store(rang)
 
+	case "random_trailers":
+		val, err := strconv.ParseBool(value)
+		if err != nil {
+			return ipcErrorf(ipc.IpcErrorInvalid, "failed to parse random trailers: %w", err)
+		}
+		device.log.Verbosef("UAPI: Updating random trailers")
+		device.randomTrailers.Store(val)
+
+	case "disable_cookies":
+		val, err := strconv.ParseBool(value)
+		if err != nil {
+			return ipcErrorf(ipc.IpcErrorInvalid, "failed to parse disable cookies: %w", err)
+		}
+		device.log.Verbosef("UAPI: Updating disable cookies")
+		device.disableCookies.Store(val)
+
 	default:
 		return ipcErrorf(ipc.IpcErrorInvalid, "invalid UAPI device key: %v", key)
 	}
@@ -823,7 +853,7 @@ func (d *ipcSetDevice) mergeWithDevice(device *Device) error {
 		paddings := []uint32{d.paddings.init, d.paddings.response, d.paddings.cookie, d.paddings.transport}
 		for i, padding := range paddings {
 			if padding < HeaderCipherNonceSize {
-				return fmt.Errorf("S%d must be more then 8 to use headerProtection", i)
+				return fmt.Errorf("S%d must be more then %d to use headerProtection", i, HeaderCipherNonceSize)
 			}
 		}
 	}