cursed

go.mod

@@ -6,6 +6,7 @@ require (
 	dario.cat/mergo v1.0.2
 	github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be
 	github.com/armon/go-radix v1.0.0
+	github.com/cilium/ebpf v0.12.3
 	github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432
 	github.com/flynn/noise v1.1.0
 	github.com/gaissmai/bart v0.25.0

go.sum

@@ -17,6 +17,8 @@ github.com/beorn7/perks v1.0.1/go.mod h1:G2ZrVWU2WbWT9wwq4/hrbKbnv/1ERSJQ0ibhJ6r
 github.com/cespare/xxhash/v2 v2.1.1/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
 github.com/cespare/xxhash/v2 v2.3.0 h1:UL815xU9SqsFlibzuggzjXhog7bL6oX9BbNZnL2UFvs=
 github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
+github.com/cilium/ebpf v0.12.3 h1:8ht6F9MquybnY97at+VDZb3eQQr8ev79RueWeVaEcG4=
+github.com/cilium/ebpf v0.12.3/go.mod h1:TctK1ivibvI3znr66ljgi4hqOT8EYQjz1KWBfb1UVgM=
 github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432 h1:M5QgkYacWj0Xs8MhpIK/5uwU02icXpEoSo9sM2aRCps=
 github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432/go.mod h1:xwIwAxMvYnVrGJPe2FKx5prTrnAjGOD8zvDOnxnrrkM=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -24,6 +26,8 @@ github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/flynn/noise v1.1.0 h1:KjPQoQCEFdZDiP03phOvGi11+SVVhBG2wOWAorLsstg=
 github.com/flynn/noise v1.1.0/go.mod h1:xbMo+0i6+IGbYdJhF31t2eR1BIU0CYc12+BNAKwUTag=
+github.com/frankban/quicktest v1.14.5 h1:dfYrrRyLtiqT9GyKXgdh+k4inNeTvmGbuSgZ3lx3GhA=
+github.com/frankban/quicktest v1.14.5/go.mod h1:4ptaffx2x8+WTWXmUCuVU6aPUX1/Mz7zb5vbUoiM6w0=
 github.com/gaissmai/bart v0.25.0 h1:eqiokVPqM3F94vJ0bTHXHtH91S8zkKL+bKh+BsGOsJM=
 github.com/gaissmai/bart v0.25.0/go.mod h1:GREWQfTLRWz/c5FTOsIw+KkscuFkIV5t8Rp7Nd1Td5c=
 github.com/go-kit/kit v0.8.0/go.mod h1:xBxKIO96dXMWWy0MnWVtmwkA9/13aqxPnvrjFYMA2as=
@@ -78,8 +82,9 @@ github.com/kr/pretty v0.2.1/go.mod h1:ipq/a2n7PKx3OHsz4KJII5eveXtPO4qwEXGdVfWzfn
 github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
 github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
 github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ=
-github.com/kr/text v0.1.0 h1:45sCR5RtlFHMR4UwH9sdQ5TC8v0qDQCHnXt+kaKSTVE=
 github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
+github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
+github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/kylelemons/godebug v1.1.0 h1:RPNrshWIDI6G2gRW9EHilWtl7Z6Sb1BR0xunSBf0SNc=
 github.com/kylelemons/godebug v1.1.0/go.mod h1:9/0rRGxNHcop5bhtWyNeEfOS8JIWk580+fNqagV/RAw=
 github.com/matttproud/golang_protobuf_extensions v1.0.1/go.mod h1:D8He9yQNgCq6Z5Ld7szi9bcBfOoFv/3dc6xSMkL2PC0=

main.go

@@ -179,7 +179,7 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 
 		useWGDefault := runtime.GOOS == "linux"
 		useWG := c.GetBool("listen.use_wireguard_stack", useWGDefault)
-		var mkListener func(*logrus.Logger, netip.Addr, int, bool, int) (udp.Conn, error)
+		var mkListener func(*logrus.Logger, netip.Addr, int, bool, int, int) (udp.Conn, error)
 		if useWG {
 			mkListener = udp.NewWireguardListener
 		} else {
@@ -188,10 +188,11 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 
 		for i := 0; i < routines; i++ {
 			l.Infof("listening on %v", netip.AddrPortFrom(listenHost, uint16(port)))
-			udpServer, err := mkListener(l, listenHost, port, routines > 1, c.GetInt("listen.batch", 64))
+			udpServer, err := mkListener(l, listenHost, port, routines > 1, c.GetInt("listen.batch", 64), i)
 			if err != nil {
 				return nil, util.NewContextualError("Failed to open udp listener", m{"queue": i}, err)
 			}
+			//todo set bpf on zeroth socket
 			udpServer.ReloadConfig(c)
 			if cfg, ok := udpServer.(interface {
 				ConfigureOffload(bool, bool, int)

udp/udp_linux.go

@@ -32,7 +32,7 @@ func maybeIPV4(ip net.IP) (net.IP, bool) {
 	return ip, false
 }
 
-func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int) (Conn, error) {
+func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int, q int) (Conn, error) {
 	af := unix.AF_INET6
 	if ip.Is4() {
 		af = unix.AF_INET

udp/wireguard_conn_linux.go

@@ -27,13 +27,13 @@ type WGConn struct {
 	enableGRO bool
 	gsoMaxSeg int
 	closed    atomic.Bool
-
+	q         int
 	closeOnce sync.Once
 }
 
 // NewWireguardListener creates a UDP listener backed by WireGuard's StdNetBind.
-func NewWireguardListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int) (Conn, error) {
+func NewWireguardListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int, q int) (Conn, error) {
-	bind := wgconn.NewStdNetBindForAddr(ip, multi)
+	bind := wgconn.NewStdNetBindForAddr(ip, multi, q)
 	recvers, actualPort, err := bind.Open(uint16(port))
 	if err != nil {
 		return nil, err
@@ -51,6 +51,7 @@ func NewWireguardListener(l *logrus.Logger, ip netip.Addr, port int, multi bool,
 		reqBatch:  batch,
 		localIP:   ip,
 		localPort: actualPort,
+		q:         q,
 	}, nil
 }
 
@@ -71,7 +72,7 @@ func (c *WGConn) listen(fn wgconn.ReceiveFunc, r EncReader) {
 	batchSize := c.batch
 	packets := make([][]byte, batchSize)
 	for i := range packets {
-		packets[i] = make([]byte, MTU)
+		packets[i] = make([]byte, 0xffff)
 	}
 	sizes := make([]int, batchSize)
 	endpoints := make([]wgconn.Endpoint, batchSize)

wgstack/conn/bind_std.go

@@ -1,12 +1,14 @@
-// SPDX-License-Identifier: MIT
+/* SPDX-License-Identifier: MIT
-//
+ *
-// Copyright (C) 2017-2023 WireGuard LLC. All Rights Reserved.
+ * Copyright (C) 2017-2023 WireGuard LLC. All Rights Reserved.
+ */
 
 package conn
 
 import (
 	"context"
 	"errors"
+	"fmt"
 	"net"
 	"net/netip"
 	"runtime"
@@ -16,7 +18,6 @@ import (
 
 	"golang.org/x/net/ipv4"
 	"golang.org/x/net/ipv6"
-	"golang.org/x/sys/unix"
 )
 
 var (
@@ -29,28 +30,55 @@ var (
 // methods for sending and receiving multiple datagrams per-syscall. See the
 // proposal in https://github.com/golang/go/issues/45886#issuecomment-1218301564.
 type StdNetBind struct {
-	mu     sync.Mutex // protects all fields except as specified
+	mu            sync.Mutex // protects all fields except as specified
-	ipv4   *net.UDPConn
+	ipv4          *net.UDPConn
-	ipv6   *net.UDPConn
+	ipv6          *net.UDPConn
-	ipv4PC *ipv4.PacketConn // will be nil on non-Linux
+	ipv4PC        *ipv4.PacketConn // will be nil on non-Linux
-	ipv6PC *ipv6.PacketConn // will be nil on non-Linux
+	ipv6PC        *ipv6.PacketConn // will be nil on non-Linux
+	ipv4TxOffload bool
+	ipv4RxOffload bool
+	ipv6TxOffload bool
+	ipv6RxOffload bool
 
-	// these three fields are not guarded by mu
+	// these two fields are not guarded by mu
-	udpAddrPool  sync.Pool
+	udpAddrPool sync.Pool
-	ipv4MsgsPool sync.Pool
+	msgsPool    sync.Pool
-	ipv6MsgsPool sync.Pool
 
 	blackhole4 bool
 	blackhole6 bool
-
+	q          int
-	listenAddr4 string
-	listenAddr6 string
-	bindV4      bool
-	bindV6      bool
-	reusePort   bool
 }
 
-func newStdNetBind() *StdNetBind {
+// NewStdNetBind creates a bind that listens on all interfaces.
+func NewStdNetBind() *StdNetBind {
+	return newStdNetBind().(*StdNetBind)
+}
+
+// NewStdNetBindForAddr creates a bind that listens on a specific address.
+// If addr is IPv4, only the IPv4 socket will be created. For IPv6, only the
+// IPv6 socket will be created.
+func NewStdNetBindForAddr(addr netip.Addr, reusePort bool, q int) *StdNetBind {
+	b := NewStdNetBind()
+	b.q = q
+	//if addr.IsValid() {
+	//	if addr.IsUnspecified() {
+	//		// keep dual-stack defaults with empty listen addresses
+	//	} else if addr.Is4() {
+	//		b.listenAddr4 = addr.Unmap().String()
+	//		b.bindV4 = true
+	//		b.bindV6 = false
+	//	} else {
+	//		b.listenAddr6 = addr.Unmap().String()
+	//		b.bindV6 = true
+	//		b.bindV4 = false
+	//	}
+	//}
+	//b.reusePort = reusePort
+
+	return b
+}
+
+func newStdNetBind() Bind {
 	return &StdNetBind{
 		udpAddrPool: sync.Pool{
 			New: func() any {
@@ -60,68 +88,28 @@ func newStdNetBind() *StdNetBind {
 			},
 		},
 
-		ipv4MsgsPool: sync.Pool{
+		msgsPool: sync.Pool{
-			New: func() any {
-				msgs := make([]ipv4.Message, IdealBatchSize)
-				for i := range msgs {
-					msgs[i].Buffers = make(net.Buffers, 1)
-					msgs[i].OOB = make([]byte, srcControlSize)
-				}
-				return &msgs
-			},
-		},
-
-		ipv6MsgsPool: sync.Pool{
 			New: func() any {
+				// ipv6.Message and ipv4.Message are interchangeable as they are
+				// both aliases for x/net/internal/socket.Message.
 				msgs := make([]ipv6.Message, IdealBatchSize)
 				for i := range msgs {
 					msgs[i].Buffers = make(net.Buffers, 1)
-					msgs[i].OOB = make([]byte, srcControlSize)
+					msgs[i].OOB = make([]byte, 0, stickyControlSize+gsoControlSize)
 				}
 				return &msgs
 			},
 		},
-		bindV4:    true,
-		bindV6:    true,
-		reusePort: false,
 	}
 }
 
-// NewStdNetBind creates a bind that listens on all interfaces.
-func NewStdNetBind() *StdNetBind {
-	return newStdNetBind()
-}
-
-// NewStdNetBindForAddr creates a bind that listens on a specific address.
-// If addr is IPv4, only the IPv4 socket will be created. For IPv6, only the
-// IPv6 socket will be created.
-func NewStdNetBindForAddr(addr netip.Addr, reusePort bool) *StdNetBind {
-	b := newStdNetBind()
-	if addr.IsValid() {
-		if addr.IsUnspecified() {
-			// keep dual-stack defaults with empty listen addresses
-		} else if addr.Is4() {
-			b.listenAddr4 = addr.Unmap().String()
-			b.bindV4 = true
-			b.bindV6 = false
-		} else {
-			b.listenAddr6 = addr.Unmap().String()
-			b.bindV6 = true
-			b.bindV4 = false
-		}
-	}
-	b.reusePort = reusePort
-	return b
-}
-
 type StdNetEndpoint struct {
 	// AddrPort is the endpoint destination.
 	netip.AddrPort
-	// src is the current sticky source address and interface index, if supported.
+	// src is the current sticky source address and interface index, if
-	src struct {
+	// supported. Typically this is a PKTINFO structure from/for control
-		netip.Addr
+	// messages, see unix.PKTINFO for an example.
-		ifidx int32
+	src []byte
-	}
 }
 
 var (
@@ -140,21 +128,17 @@ func (*StdNetBind) ParseEndpoint(s string) (Endpoint, error) {
 }
 
 func (e *StdNetEndpoint) ClearSrc() {
-	e.src.ifidx = 0
+	if e.src != nil {
-	e.src.Addr = netip.Addr{}
+		// Truncate src, no need to reallocate.
+		e.src = e.src[:0]
+	}
 }
 
 func (e *StdNetEndpoint) DstIP() netip.Addr {
 	return e.AddrPort.Addr()
 }
 
-func (e *StdNetEndpoint) SrcIP() netip.Addr {
+// See control_default,linux, etc for implementations of SrcIP and SrcIfidx.
-	return e.src.Addr
-}
-
-func (e *StdNetEndpoint) SrcIfidx() int32 {
-	return e.src.ifidx
-}
 
 func (e *StdNetEndpoint) DstToBytes() []byte {
 	b, _ := e.AddrPort.MarshalBinary()
@@ -165,36 +149,24 @@ func (e *StdNetEndpoint) DstToString() string {
 	return e.AddrPort.String()
 }
 
-func (e *StdNetEndpoint) SrcToString() string {
+func listenNet(network string, port int, q int) (*net.UDPConn, int, error) {
-	return e.src.Addr.String()
+	lc := listenConfig(q)
-}
 
-func (s *StdNetBind) listenNet(network string, host string, port int) (*net.UDPConn, int, error) {
+	conn, err := lc.ListenPacket(context.Background(), network, ":"+strconv.Itoa(port))
-	lc := listenConfig()
-	if s.reusePort {
-		base := lc.Control
-		lc.Control = func(network, address string, c syscall.RawConn) error {
-			if base != nil {
-				if err := base(network, address, c); err != nil {
-					return err
-				}
-			}
-			return c.Control(func(fd uintptr) {
-				_ = unix.SetsockoptInt(int(fd), unix.SOL_SOCKET, unix.SO_REUSEPORT, 1)
-			})
-		}
-	}
-
-	addr := ":" + strconv.Itoa(port)
-	if host != "" {
-		addr = net.JoinHostPort(host, strconv.Itoa(port))
-	}
-
-	conn, err := lc.ListenPacket(context.Background(), network, addr)
 	if err != nil {
 		return nil, 0, err
 	}
 
+	if q == 0 {
+		if EvilFdZero == 0 {
+			panic("fuck")
+		}
+		err = reusePortHax(EvilFdZero)
+		if err != nil {
+			return nil, 0, fmt.Errorf("reuse port hax: %v", err)
+		}
+	}
+
 	// Retrieve port.
 	laddr := conn.LocalAddr()
 	uaddr, err := net.ResolveUDPAddr(
@@ -204,47 +176,10 @@ func (s *StdNetBind) listenNet(network string, host string, port int) (*net.UDPC
 	if err != nil {
 		return nil, 0, err
 	}
+
 	return conn.(*net.UDPConn), uaddr.Port, nil
 }
 
-func (s *StdNetBind) openIPv4(port int) (*net.UDPConn, *ipv4.PacketConn, int, error) {
-	if !s.bindV4 {
-		return nil, nil, port, nil
-	}
-	host := s.listenAddr4
-	conn, actualPort, err := s.listenNet("udp4", host, port)
-	if err != nil {
-		if errors.Is(err, syscall.EAFNOSUPPORT) {
-			return nil, nil, port, nil
-		}
-		return nil, nil, port, err
-	}
-	if runtime.GOOS != "linux" {
-		return conn, nil, actualPort, nil
-	}
-	pc := ipv4.NewPacketConn(conn)
-	return conn, pc, actualPort, nil
-}
-
-func (s *StdNetBind) openIPv6(port int) (*net.UDPConn, *ipv6.PacketConn, int, error) {
-	if !s.bindV6 {
-		return nil, nil, port, nil
-	}
-	host := s.listenAddr6
-	conn, actualPort, err := s.listenNet("udp6", host, port)
-	if err != nil {
-		if errors.Is(err, syscall.EAFNOSUPPORT) {
-			return nil, nil, port, nil
-		}
-		return nil, nil, port, err
-	}
-	if runtime.GOOS != "linux" {
-		return conn, nil, actualPort, nil
-	}
-	pc := ipv6.NewPacketConn(conn)
-	return conn, pc, actualPort, nil
-}
-
 func (s *StdNetBind) Open(uport uint16) ([]ReceiveFunc, uint16, error) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
@@ -260,46 +195,44 @@ func (s *StdNetBind) Open(uport uint16) ([]ReceiveFunc, uint16, error) {
 	// If uport is 0, we can retry on failure.
 again:
 	port := int(uport)
-	var v4conn *net.UDPConn
+	var v4conn, v6conn *net.UDPConn
-	var v6conn *net.UDPConn
 	var v4pc *ipv4.PacketConn
 	var v6pc *ipv6.PacketConn
 
-	v4conn, v4pc, port, err = s.openIPv4(port)
+	v4conn, port, err = listenNet("udp4", port, s.q)
-	if err != nil {
+	if err != nil && !errors.Is(err, syscall.EAFNOSUPPORT) {
 		return nil, 0, err
 	}
 
 	// Listen on the same port as we're using for ipv4.
-	v6conn, v6pc, port, err = s.openIPv6(port)
+	v6conn, port, err = listenNet("udp6", port, s.q)
 	if uport == 0 && errors.Is(err, syscall.EADDRINUSE) && tries < 100 {
-		if v4conn != nil {
+		v4conn.Close()
-			v4conn.Close()
-		}
 		tries++
 		goto again
 	}
-	if err != nil {
+	if err != nil && !errors.Is(err, syscall.EAFNOSUPPORT) {
-		if v4conn != nil {
+		v4conn.Close()
-			v4conn.Close()
-		}
 		return nil, 0, err
 	}
-
 	var fns []ReceiveFunc
 	if v4conn != nil {
-		s.ipv4 = v4conn
+		s.ipv4TxOffload, s.ipv4RxOffload = supportsUDPOffload(v4conn)
-		if v4pc != nil {
+		if runtime.GOOS == "linux" || runtime.GOOS == "android" {
+			v4pc = ipv4.NewPacketConn(v4conn)
 			s.ipv4PC = v4pc
 		}
-		fns = append(fns, s.makeReceiveIPv4(v4pc, v4conn))
+		fns = append(fns, s.makeReceiveIPv4(v4pc, v4conn, s.ipv4RxOffload))
+		s.ipv4 = v4conn
 	}
 	if v6conn != nil {
-		s.ipv6 = v6conn
+		s.ipv6TxOffload, s.ipv6RxOffload = supportsUDPOffload(v6conn)
-		if v6pc != nil {
+		if runtime.GOOS == "linux" || runtime.GOOS == "android" {
+			v6pc = ipv6.NewPacketConn(v6conn)
 			s.ipv6PC = v6pc
 		}
-		fns = append(fns, s.makeReceiveIPv6(v6pc, v6conn))
+		fns = append(fns, s.makeReceiveIPv6(v6pc, v6conn, s.ipv6RxOffload))
+		s.ipv6 = v6conn
 	}
 	if len(fns) == 0 {
 		return nil, 0, syscall.EAFNOSUPPORT
@@ -308,76 +241,101 @@ again:
 	return fns, uint16(port), nil
 }
 
-func (s *StdNetBind) makeReceiveIPv4(pc *ipv4.PacketConn, conn *net.UDPConn) ReceiveFunc {
+func (s *StdNetBind) putMessages(msgs *[]ipv6.Message) {
-	return func(bufs [][]byte, sizes []int, eps []Endpoint) (n int, err error) {
+	for i := range *msgs {
-		msgs := s.ipv4MsgsPool.Get().(*[]ipv4.Message)
+		(*msgs)[i].OOB = (*msgs)[i].OOB[:0]
-		defer s.ipv4MsgsPool.Put(msgs)
+		(*msgs)[i] = ipv6.Message{Buffers: (*msgs)[i].Buffers, OOB: (*msgs)[i].OOB}
-		for i := range bufs {
+	}
-			(*msgs)[i].Buffers[0] = bufs[i]
+	s.msgsPool.Put(msgs)
-		}
+}
-		var numMsgs int
+
-		if runtime.GOOS == "linux" && pc != nil {
+func (s *StdNetBind) getMessages() *[]ipv6.Message {
-			numMsgs, err = pc.ReadBatch(*msgs, 0)
+	return s.msgsPool.Get().(*[]ipv6.Message)
+}
+
+var (
+	// If compilation fails here these are no longer the same underlying type.
+	_ ipv6.Message = ipv4.Message{}
+)
+
+type batchReader interface {
+	ReadBatch([]ipv6.Message, int) (int, error)
+}
+
+type batchWriter interface {
+	WriteBatch([]ipv6.Message, int) (int, error)
+}
+
+func (s *StdNetBind) receiveIP(
+	br batchReader,
+	conn *net.UDPConn,
+	rxOffload bool,
+	bufs [][]byte,
+	sizes []int,
+	eps []Endpoint,
+) (n int, err error) {
+	msgs := s.getMessages()
+	for i := range bufs {
+		(*msgs)[i].Buffers[0] = bufs[i]
+		(*msgs)[i].OOB = (*msgs)[i].OOB[:cap((*msgs)[i].OOB)]
+	}
+	defer s.putMessages(msgs)
+	var numMsgs int
+	if runtime.GOOS == "linux" || runtime.GOOS == "android" {
+		if rxOffload {
+			readAt := len(*msgs) - (IdealBatchSize / udpSegmentMaxDatagrams)
+			numMsgs, err = br.ReadBatch((*msgs)[readAt:], 0)
+			if err != nil {
+				return 0, err
+			}
+			numMsgs, err = splitCoalescedMessages(*msgs, readAt, getGSOSize)
 			if err != nil {
 				return 0, err
 			}
 		} else {
-			msg := &(*msgs)[0]
+			numMsgs, err = br.ReadBatch(*msgs, 0)
-			msg.N, msg.NN, _, msg.Addr, err = conn.ReadMsgUDP(msg.Buffers[0], msg.OOB)
 			if err != nil {
 				return 0, err
 			}
-			numMsgs = 1
 		}
-		for i := 0; i < numMsgs; i++ {
+	} else {
-			msg := &(*msgs)[i]
+		msg := &(*msgs)[0]
-			sizes[i] = msg.N
+		msg.N, msg.NN, _, msg.Addr, err = conn.ReadMsgUDP(msg.Buffers[0], msg.OOB)
-			addrPort := msg.Addr.(*net.UDPAddr).AddrPort()
+		if err != nil {
-			ep := &StdNetEndpoint{AddrPort: addrPort} // TODO: remove allocation
+			return 0, err
-			getSrcFromControl(msg.OOB[:msg.NN], ep)
-			eps[i] = ep
 		}
-		return numMsgs, nil
+		numMsgs = 1
+	}
+	for i := 0; i < numMsgs; i++ {
+		msg := &(*msgs)[i]
+		sizes[i] = msg.N
+		if sizes[i] == 0 {
+			continue
+		}
+		addrPort := msg.Addr.(*net.UDPAddr).AddrPort()
+		ep := &StdNetEndpoint{AddrPort: addrPort} // TODO: remove allocation
+		getSrcFromControl(msg.OOB[:msg.NN], ep)
+		eps[i] = ep
+	}
+	return numMsgs, nil
+}
+
+func (s *StdNetBind) makeReceiveIPv4(pc *ipv4.PacketConn, conn *net.UDPConn, rxOffload bool) ReceiveFunc {
+	return func(bufs [][]byte, sizes []int, eps []Endpoint) (n int, err error) {
+		return s.receiveIP(pc, conn, rxOffload, bufs, sizes, eps)
 	}
 }
 
-func (s *StdNetBind) makeReceiveIPv6(pc *ipv6.PacketConn, conn *net.UDPConn) ReceiveFunc {
+func (s *StdNetBind) makeReceiveIPv6(pc *ipv6.PacketConn, conn *net.UDPConn, rxOffload bool) ReceiveFunc {
 	return func(bufs [][]byte, sizes []int, eps []Endpoint) (n int, err error) {
-		msgs := s.ipv6MsgsPool.Get().(*[]ipv6.Message)
+		return s.receiveIP(pc, conn, rxOffload, bufs, sizes, eps)
-		defer s.ipv6MsgsPool.Put(msgs)
-		for i := range bufs {
-			(*msgs)[i].Buffers[0] = bufs[i]
-		}
-		var numMsgs int
-		if runtime.GOOS == "linux" && pc != nil {
-			numMsgs, err = pc.ReadBatch(*msgs, 0)
-			if err != nil {
-				return 0, err
-			}
-		} else {
-			msg := &(*msgs)[0]
-			msg.N, msg.NN, _, msg.Addr, err = conn.ReadMsgUDP(msg.Buffers[0], msg.OOB)
-			if err != nil {
-				return 0, err
-			}
-			numMsgs = 1
-		}
-		for i := 0; i < numMsgs; i++ {
-			msg := &(*msgs)[i]
-			sizes[i] = msg.N
-			addrPort := msg.Addr.(*net.UDPAddr).AddrPort()
-			ep := &StdNetEndpoint{AddrPort: addrPort} // TODO: remove allocation
-			getSrcFromControl(msg.OOB[:msg.NN], ep)
-			eps[i] = ep
-		}
-		return numMsgs, nil
 	}
 }
 
 // TODO: When all Binds handle IdealBatchSize, remove this dynamic function and
 // rename the IdealBatchSize constant to BatchSize.
 func (s *StdNetBind) BatchSize() int {
-	if runtime.GOOS == "linux" {
+	if runtime.GOOS == "linux" || runtime.GOOS == "android" {
 		return IdealBatchSize
 	}
 	return 1
@@ -400,28 +358,42 @@ func (s *StdNetBind) Close() error {
 	}
 	s.blackhole4 = false
 	s.blackhole6 = false
+	s.ipv4TxOffload = false
+	s.ipv4RxOffload = false
+	s.ipv6TxOffload = false
+	s.ipv6RxOffload = false
 	if err1 != nil {
 		return err1
 	}
 	return err2
 }
 
+type ErrUDPGSODisabled struct {
+	onLaddr  string
+	RetryErr error
+}
+
+func (e ErrUDPGSODisabled) Error() string {
+	return fmt.Sprintf("disabled UDP GSO on %s, NIC(s) may not support checksum offload", e.onLaddr)
+}
+
+func (e ErrUDPGSODisabled) Unwrap() error {
+	return e.RetryErr
+}
+
 func (s *StdNetBind) Send(bufs [][]byte, endpoint Endpoint) error {
 	s.mu.Lock()
 	blackhole := s.blackhole4
 	conn := s.ipv4
-	var (
+	offload := s.ipv4TxOffload
-		pc4 *ipv4.PacketConn
+	br := batchWriter(s.ipv4PC)
-		pc6 *ipv6.PacketConn
-	)
 	is6 := false
 	if endpoint.DstIP().Is6() {
 		blackhole = s.blackhole6
 		conn = s.ipv6
-		pc6 = s.ipv6PC
+		br = s.ipv6PC
 		is6 = true
-	} else {
+		offload = s.ipv6TxOffload
-		pc4 = s.ipv4PC
 	}
 	s.mu.Unlock()
 
@@ -431,109 +403,185 @@ func (s *StdNetBind) Send(bufs [][]byte, endpoint Endpoint) error {
 	if conn == nil {
 		return syscall.EAFNOSUPPORT
 	}
+
+	msgs := s.getMessages()
+	defer s.putMessages(msgs)
+	ua := s.udpAddrPool.Get().(*net.UDPAddr)
+	defer s.udpAddrPool.Put(ua)
 	if is6 {
-		return s.send6(conn, pc6, endpoint, bufs)
+		as16 := endpoint.DstIP().As16()
+		copy(ua.IP, as16[:])
+		ua.IP = ua.IP[:16]
 	} else {
-		return s.send4(conn, pc4, endpoint, bufs)
+		as4 := endpoint.DstIP().As4()
+		copy(ua.IP, as4[:])
+		ua.IP = ua.IP[:4]
 	}
+	ua.Port = int(endpoint.(*StdNetEndpoint).Port())
+	var (
+		retried bool
+		err     error
+	)
+retry:
+	if offload {
+		n := coalesceMessages(ua, endpoint.(*StdNetEndpoint), bufs, *msgs, setGSOSize)
+		err = s.send(conn, br, (*msgs)[:n])
+		if err != nil && offload && errShouldDisableUDPGSO(err) {
+			offload = false
+			s.mu.Lock()
+			if is6 {
+				s.ipv6TxOffload = false
+			} else {
+				s.ipv4TxOffload = false
+			}
+			s.mu.Unlock()
+			retried = true
+			goto retry
+		}
+	} else {
+		for i := range bufs {
+			(*msgs)[i].Addr = ua
+			(*msgs)[i].Buffers[0] = bufs[i]
+			setSrcControl(&(*msgs)[i].OOB, endpoint.(*StdNetEndpoint))
+		}
+		err = s.send(conn, br, (*msgs)[:len(bufs)])
+	}
+	if retried {
+		return ErrUDPGSODisabled{onLaddr: conn.LocalAddr().String(), RetryErr: err}
+	}
+	return err
 }
 
-func (s *StdNetBind) send4(conn *net.UDPConn, pc *ipv4.PacketConn, ep Endpoint, bufs [][]byte) error {
+func (s *StdNetBind) send(conn *net.UDPConn, pc batchWriter, msgs []ipv6.Message) error {
-	ua := s.udpAddrPool.Get().(*net.UDPAddr)
-	as4 := ep.DstIP().As4()
-	copy(ua.IP, as4[:])
-	ua.IP = ua.IP[:4]
-	ua.Port = int(ep.(*StdNetEndpoint).Port())
-	msgs := s.ipv4MsgsPool.Get().(*[]ipv4.Message)
-	for i, buf := range bufs {
-		(*msgs)[i].Buffers[0] = buf
-		(*msgs)[i].Addr = ua
-		setSrcControl(&(*msgs)[i].OOB, ep.(*StdNetEndpoint))
-	}
 	var (
 		n     int
 		err   error
 		start int
 	)
-	if runtime.GOOS == "linux" && pc != nil {
+	if runtime.GOOS == "linux" || runtime.GOOS == "android" {
 		for {
-			n, err = pc.WriteBatch((*msgs)[start:len(bufs)], 0)
+			n, err = pc.WriteBatch(msgs[start:], 0)
-			if err != nil {
+			if err != nil || n == len(msgs[start:]) {
-				if errors.Is(err, syscall.EAFNOSUPPORT) {
-					for j := start; j < len(bufs); j++ {
-						_, _, werr := conn.WriteMsgUDP(bufs[j], (*msgs)[j].OOB, ua)
-						if werr != nil {
-							err = werr
-							break
-						}
-					}
-				}
-				break
-			}
-			if n == len((*msgs)[start:len(bufs)]) {
 				break
 			}
 			start += n
 		}
 	} else {
-		for i, buf := range bufs {
+		for _, msg := range msgs {
-			_, _, err = conn.WriteMsgUDP(buf, (*msgs)[i].OOB, ua)
+			_, _, err = conn.WriteMsgUDP(msg.Buffers[0], msg.OOB, msg.Addr.(*net.UDPAddr))
 			if err != nil {
 				break
 			}
 		}
 	}
-	s.udpAddrPool.Put(ua)
-	s.ipv4MsgsPool.Put(msgs)
 	return err
 }
 
-func (s *StdNetBind) send6(conn *net.UDPConn, pc *ipv6.PacketConn, ep Endpoint, bufs [][]byte) error {
+const (
-	ua := s.udpAddrPool.Get().(*net.UDPAddr)
+	// Exceeding these values results in EMSGSIZE. They account for layer3 and
-	as16 := ep.DstIP().As16()
+	// layer4 headers. IPv6 does not need to account for itself as the payload
-	copy(ua.IP, as16[:])
+	// length field is self excluding.
-	ua.IP = ua.IP[:16]
+	maxIPv4PayloadLen = 1<<16 - 1 - 20 - 8
-	ua.Port = int(ep.(*StdNetEndpoint).Port())
+	maxIPv6PayloadLen = 1<<16 - 1 - 8
-	msgs := s.ipv6MsgsPool.Get().(*[]ipv6.Message)
+
-	for i, buf := range bufs {
+	// This is a hard limit imposed by the kernel.
-		(*msgs)[i].Buffers[0] = buf
+	udpSegmentMaxDatagrams = 64
-		(*msgs)[i].Addr = ua
+)
-		setSrcControl(&(*msgs)[i].OOB, ep.(*StdNetEndpoint))
+
-	}
+type setGSOFunc func(control *[]byte, gsoSize uint16)
+
+func coalesceMessages(addr *net.UDPAddr, ep *StdNetEndpoint, bufs [][]byte, msgs []ipv6.Message, setGSO setGSOFunc) int {
 	var (
-		n     int
+		base     = -1 // index of msg we are currently coalescing into
-		err   error
+		gsoSize  int  // segmentation size of msgs[base]
-		start int
+		dgramCnt int  // number of dgrams coalesced into msgs[base]
+		endBatch bool // tracking flag to start a new batch on next iteration of bufs
 	)
-	if runtime.GOOS == "linux" && pc != nil {
+	maxPayloadLen := maxIPv4PayloadLen
-		for {
+	if ep.DstIP().Is6() {
-			n, err = pc.WriteBatch((*msgs)[start:len(bufs)], 0)
+		maxPayloadLen = maxIPv6PayloadLen
-			if err != nil {
+	}
-				if errors.Is(err, syscall.EAFNOSUPPORT) {
+	for i, buf := range bufs {
-					for j := start; j < len(bufs); j++ {
+		if i > 0 {
-						_, _, werr := conn.WriteMsgUDP(bufs[j], (*msgs)[j].OOB, ua)
+			msgLen := len(buf)
-						if werr != nil {
+			baseLenBefore := len(msgs[base].Buffers[0])
-							err = werr
+			freeBaseCap := cap(msgs[base].Buffers[0]) - baseLenBefore
-							break
+			if msgLen+baseLenBefore <= maxPayloadLen &&
-						}
+				msgLen <= gsoSize &&
-					}
+				msgLen <= freeBaseCap &&
+				dgramCnt < udpSegmentMaxDatagrams &&
+				!endBatch {
+				msgs[base].Buffers[0] = append(msgs[base].Buffers[0], buf...)
+				if i == len(bufs)-1 {
+					setGSO(&msgs[base].OOB, uint16(gsoSize))
 				}
-				break
+				dgramCnt++
+				if msgLen < gsoSize {
+					// A smaller than gsoSize packet on the tail is legal, but
+					// it must end the batch.
+					endBatch = true
+				}
+				continue
 			}
-			if n == len((*msgs)[start:len(bufs)]) {
-				break
-			}
-			start += n
 		}
-	} else {
+		if dgramCnt > 1 {
-		for i, buf := range bufs {
+			setGSO(&msgs[base].OOB, uint16(gsoSize))
-			_, _, err = conn.WriteMsgUDP(buf, (*msgs)[i].OOB, ua)
+		}
-			if err != nil {
+		// Reset prior to incrementing base since we are preparing to start a
-				break
+		// new potential batch.
+		endBatch = false
+		base++
+		gsoSize = len(buf)
+		setSrcControl(&msgs[base].OOB, ep)
+		msgs[base].Buffers[0] = buf
+		msgs[base].Addr = addr
+		dgramCnt = 1
+	}
+	return base + 1
+}
+
+type getGSOFunc func(control []byte) (int, error)
+
+func splitCoalescedMessages(msgs []ipv6.Message, firstMsgAt int, getGSO getGSOFunc) (n int, err error) {
+	for i := firstMsgAt; i < len(msgs); i++ {
+		msg := &msgs[i]
+		if msg.N == 0 {
+			return n, err
+		}
+		var (
+			gsoSize    int
+			start      int
+			end        = msg.N
+			numToSplit = 1
+		)
+		gsoSize, err = getGSO(msg.OOB[:msg.NN])
+		if err != nil {
+			return n, err
+		}
+		if gsoSize > 0 {
+			numToSplit = (msg.N + gsoSize - 1) / gsoSize
+			end = gsoSize
+		}
+		for j := 0; j < numToSplit; j++ {
+			if n > i {
+				return n, errors.New("splitting coalesced packet resulted in overflow")
 			}
+			copied := copy(msgs[n].Buffers[0], msg.Buffers[0][start:end])
+			msgs[n].N = copied
+			msgs[n].Addr = msg.Addr
+			start = end
+			end += gsoSize
+			if end > msg.N {
+				end = msg.N
+			}
+			n++
+		}
+		if i != n-1 {
+			// It is legal for bytes to move within msg.Buffers[0] as a result
+			// of splitting, so we only zero the source msg len when it is not
+			// the destination of the last split operation above.
+			msg.N = 0
 		}
 	}
-	s.udpAddrPool.Put(ua)
+	return n, nil
-	s.ipv6MsgsPool.Put(msgs)
-	return err
 }

wgstack/conn/controlfns.go

@@ -5,8 +5,12 @@
 package conn
 
 import (
+	"fmt"
 	"net"
 	"syscall"
+
+	"github.com/cilium/ebpf"
+	"github.com/cilium/ebpf/asm"
 )
 
 // UDP socket read/write buffer size (7MB). The value of 7MB is chosen as it is
@@ -25,10 +29,169 @@ type controlFn func(network, address string, c syscall.RawConn) error
 // that can apply socket options.
 var controlFns = []controlFn{}
 
+const SO_ATTACH_REUSEPORT_EBPF = 52
+
+//Create eBPF program that returns a hash to distribute packets
+
+func createReuseportProgram() (*ebpf.Program, error) {
+	// This program uses the packet's hash and returns it modulo number of sockets
+	// Simple version: just return a counter-based distribution
+	//instructions := asm.Instructions{
+	//	// Load the skb->hash value (already computed by kernel)
+	//	asm.LoadMem(asm.R0, asm.R1, int16(unsafe.Offsetof(unix.XDPMd{}.RxQueueIndex)), asm.Word),
+	//	asm.Return(),
+	//}
+	//
+	//// Alternative: simpler round-robin approach
+	//// This returns the CPU number, effectively round-robin
+	//instructions := asm.Instructions{
+	//	asm.Mov.Reg(asm.R0, asm.R1),              // Move ctx to R0
+	//	asm.LoadMem(asm.R0, asm.R1, 0, asm.Word), // Load some field
+	//	asm.Return(),
+	//}
+
+	// Better: Use BPF helper to get random/hash value
+	//instructions := asm.Instructions{
+	//	// Call get_prandom_u32() to get random value for distribution
+	//	asm.Mov.Imm(asm.R0, 0),
+	//	asm.Call.Label("get_prandom_u32"),
+	//	asm.Return(),
+	//}
+	//
+	//prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
+	//	Type:         ebpf.SocketFilter,
+	//	Instructions: instructions,
+	//	License:      "GPL",
+	//})
+
+	//instructions := asm.Instructions{
+	//	// R1 contains pointer to skb
+	//	// Load skb->hash at offset 0x20 (may vary by kernel, but 0x20 is common)
+	//	asm.LoadMem(asm.R0, asm.R1, 0x20, asm.Word),
+	//
+	//	// If hash is 0, use rxhash instead (fallback)
+	//	asm.JEq.Imm(asm.R0, 0, "use_rxhash"),
+	//	asm.Return().Sym("return"),
+	//
+	//	// Fallback: load rxhash
+	//	asm.LoadMem(asm.R0, asm.R1, 0x24, asm.Word).Sym("use_rxhash"),
+	//	asm.Return(),
+	//}
+	//
+	//prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
+	//	Type:         ebpf.SkReuseport,
+	//	Instructions: instructions,
+	//	License:      "GPL",
+	//})
+
+	//instructions := asm.Instructions{
+	//	// R1 = ctx (sk_reuseport_md)
+	//	// R2 = sk_reuseport map (we'll use NULL/0 for default behavior)
+	//	// R3 = key (select socket index)
+	//	// R4 = flags
+	//
+	//	// Simple approach: use the hash field from sk_reuseport_md
+	//	// struct sk_reuseport_md { ... __u32 hash; ... } at offset 24
+	//	asm.Mov.Reg(asm.R6, asm.R1), // Save ctx
+	//
+	//	// Load the hash value at offset 24
+	//	asm.LoadMem(asm.R2, asm.R6, 24, asm.Word),
+	//
+	//	// Call bpf_sk_select_reuseport(ctx, map, key, flags)
+	//	asm.Mov.Reg(asm.R1, asm.R6), // ctx
+	//	asm.Mov.Imm(asm.R2, 0),      // map (NULL = use default)
+	//	asm.Mov.Reg(asm.R3, asm.R2), // key = hash we loaded (in R2)
+	//	asm.Mov.Imm(asm.R4, 0),      // flags
+	//	asm.Call.Label("sk_select_reuseport"),
+	//
+	//	// Return 0
+	//	asm.Mov.Imm(asm.R0, 0),
+	//	asm.Return(),
+	//}
+	//
+	//prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
+	//	Type:         ebpf.SkReuseport,
+	//	Instructions: instructions,
+	//	License:      "GPL",
+	//})
+
+	instructions := asm.Instructions{
+		// R1 = ctx (sk_reuseport_md pointer)
+		// Load hash from sk_reuseport_md at offset 24
+		//asm.LoadMem(asm.R0, asm.R1, 20, asm.Word),
+
+		// R1 = ctx (save it)
+		asm.Mov.Reg(asm.R6, asm.R1),
+
+		// Prepare string on stack: "BPF called!\n"
+		// We need to build the format string on the stack
+		asm.Mov.Reg(asm.R1, asm.R10), // R1 = frame pointer
+		asm.Add.Imm(asm.R1, -16),     // R1 = stack location for string
+
+		// Write "BPF called!\n" to stack (we'll use a simpler version)
+		// Store immediate 64-bit values
+		asm.StoreImm(asm.R1, 0, 0x2066706220, asm.DWord), // "bpf "
+		asm.StoreImm(asm.R1, 8, 0x0a21, asm.DWord),       // "!\n"
+
+		// Call bpf_trace_printk(fmt, fmt_size)
+		// R1 already points to format string
+		asm.Mov.Imm(asm.R2, 16), // R2 = format size
+		asm.Call.Label("bpf_printk"),
+
+		// Return 0 (send to socket 0 for testing)
+		asm.Mov.Imm(asm.R0, 0),
+		asm.Return(),
+
+		//asm.Mov.Imm(asm.R0, 0),
+		//// Just return the hash directly
+		//// The kernel will automatically modulo by number of sockets
+		//asm.Return(),
+	}
+
+	prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
+		Type:         ebpf.SkReuseport,
+		Instructions: instructions,
+		License:      "GPL",
+	})
+
+	return prog, err
+}
+
+//func createReuseportProgram() (*ebpf.Program, error) {
+//	// Try offset 20 (common in newer kernels)
+//	instructions := asm.Instructions{
+//		asm.LoadMem(asm.R0, asm.R1, 20, asm.Word),
+//		asm.Return(),
+//	}
+//
+//	prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
+//		Type:         ebpf.SkReuseport,
+//		Instructions: instructions,
+//		License:      "GPL",
+//	})
+//
+//	return prog, err
+//}
+
+func reusePortHax(fd uintptr) error {
+	prog, err := createReuseportProgram()
+	if err != nil {
+		return fmt.Errorf("failed to create eBPF program: %w", err)
+	}
+	//defer prog.Close()
+	sockErr := syscall.SetsockoptInt(int(fd), syscall.SOL_SOCKET, SO_ATTACH_REUSEPORT_EBPF, prog.FD())
+	if sockErr != nil {
+		return sockErr
+	}
+	return nil
+}
+
+var EvilFdZero uintptr
+
 // listenConfig returns a net.ListenConfig that applies the controlFns to the
 // socket prior to bind. This is used to apply socket buffer sizing and packet
 // information OOB configuration for sticky sockets.
-func listenConfig() *net.ListenConfig {
+func listenConfig(q int) *net.ListenConfig {
 	return &net.ListenConfig{
 		Control: func(network, address string, c syscall.RawConn) error {
 			for _, fn := range controlFns {
@@ -36,6 +199,23 @@ func listenConfig() *net.ListenConfig {
 					return err
 				}
 			}
+
+			if q == 0 {
+				c.Control(func(fd uintptr) {
+					EvilFdZero = fd
+				})
+				//	var e error
+				//	err := c.Control(func(fd uintptr) {
+				//		e = reusePortHax(fd)
+				//	})
+				//	if err != nil {
+				//		return err
+				//	}
+				//	if e != nil {
+				//		return e
+				//	}
+			}
+
 			return nil
 		},
 	}

wgstack/conn/controlfns_linux.go

@@ -29,6 +29,10 @@ func init() {
 				// Set beyond *mem_max if CAP_NET_ADMIN
 				_ = unix.SetsockoptInt(int(fd), unix.SOL_SOCKET, unix.SO_RCVBUFFORCE, socketBufferSize)
 				_ = unix.SetsockoptInt(int(fd), unix.SOL_SOCKET, unix.SO_SNDBUFFORCE, socketBufferSize)
+				_ = unix.SetsockoptInt(int(fd), unix.SOL_SOCKET, unix.SO_REUSEPORT, 1)  //todo!!!
+				_ = unix.SetsockoptInt(int(fd), unix.IPPROTO_UDP, unix.UDP_GRO, 1)      //todo!!!
+				_ = unix.SetsockoptInt(int(fd), unix.SOL_UDP, unix.UDP_SEGMENT, 0xffff) //todo!!!
+				//print(err.Error())
 			})
 		},
 

wgstack/conn/features_linux.go

@@ -6,6 +6,7 @@
 package conn
 
 import (
+	"fmt"
 	"net"
 
 	"golang.org/x/sys/unix"
@@ -16,12 +17,15 @@ func supportsUDPOffload(conn *net.UDPConn) (txOffload, rxOffload bool) {
 	if err != nil {
 		return
 	}
+	a := 0
 	err = rc.Control(func(fd uintptr) {
-		_, errSyscall := unix.GetsockoptInt(int(fd), unix.IPPROTO_UDP, unix.UDP_SEGMENT)
+		a, err = unix.GetsockoptInt(int(fd), unix.IPPROTO_UDP, unix.UDP_SEGMENT)
-		txOffload = errSyscall == nil
+
+		txOffload = err == nil
 		opt, errSyscall := unix.GetsockoptInt(int(fd), unix.IPPROTO_UDP, unix.UDP_GRO)
 		rxOffload = errSyscall == nil && opt == 1
 	})
+	fmt.Printf("%d", a)
 	if err != nil {
 		return false, false
 	}

wgstack/conn/sticky_linux.go

@@ -1,9 +1,3 @@
-//go:build linux && !android
-
-// SPDX-License-Identifier: MIT
-//
-// Copyright (C) 2017-2023 WireGuard LLC. All Rights Reserved.
-
 package conn
 
 import (
@@ -13,6 +7,37 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+func (e *StdNetEndpoint) SrcIP() netip.Addr {
+	switch len(e.src) {
+	case unix.CmsgSpace(unix.SizeofInet4Pktinfo):
+		info := (*unix.Inet4Pktinfo)(unsafe.Pointer(&e.src[unix.CmsgLen(0)]))
+		return netip.AddrFrom4(info.Spec_dst)
+	case unix.CmsgSpace(unix.SizeofInet6Pktinfo):
+		info := (*unix.Inet6Pktinfo)(unsafe.Pointer(&e.src[unix.CmsgLen(0)]))
+		// TODO: set zone. in order to do so we need to check if the address is
+		// link local, and if it is perform a syscall to turn the ifindex into a
+		// zone string because netip uses string zones.
+		return netip.AddrFrom16(info.Addr)
+	}
+	return netip.Addr{}
+}
+
+func (e *StdNetEndpoint) SrcIfidx() int32 {
+	switch len(e.src) {
+	case unix.CmsgSpace(unix.SizeofInet4Pktinfo):
+		info := (*unix.Inet4Pktinfo)(unsafe.Pointer(&e.src[unix.CmsgLen(0)]))
+		return info.Ifindex
+	case unix.CmsgSpace(unix.SizeofInet6Pktinfo):
+		info := (*unix.Inet6Pktinfo)(unsafe.Pointer(&e.src[unix.CmsgLen(0)]))
+		return int32(info.Ifindex)
+	}
+	return 0
+}
+
+func (e *StdNetEndpoint) SrcToString() string {
+	return e.SrcIP().String()
+}
+
 // getSrcFromControl parses the control for PKTINFO and if found updates ep with
 // the source information found.
 func getSrcFromControl(control []byte, ep *StdNetEndpoint) {
@@ -34,83 +59,47 @@ func getSrcFromControl(control []byte, ep *StdNetEndpoint) {
 		if hdr.Level == unix.IPPROTO_IP &&
 			hdr.Type == unix.IP_PKTINFO {
 
-			info := pktInfoFromBuf[unix.Inet4Pktinfo](data)
+			if ep.src == nil || cap(ep.src) < unix.CmsgSpace(unix.SizeofInet4Pktinfo) {
-			ep.src.Addr = netip.AddrFrom4(info.Spec_dst)
+				ep.src = make([]byte, 0, unix.CmsgSpace(unix.SizeofInet4Pktinfo))
-			ep.src.ifidx = info.Ifindex
+			}
+			ep.src = ep.src[:unix.CmsgSpace(unix.SizeofInet4Pktinfo)]
 
+			hdrBuf := unsafe.Slice((*byte)(unsafe.Pointer(&hdr)), unix.SizeofCmsghdr)
+			copy(ep.src, hdrBuf)
+			copy(ep.src[unix.CmsgLen(0):], data)
 			return
 		}
 
 		if hdr.Level == unix.IPPROTO_IPV6 &&
 			hdr.Type == unix.IPV6_PKTINFO {
 
-			info := pktInfoFromBuf[unix.Inet6Pktinfo](data)
+			if ep.src == nil || cap(ep.src) < unix.CmsgSpace(unix.SizeofInet6Pktinfo) {
-			ep.src.Addr = netip.AddrFrom16(info.Addr)
+				ep.src = make([]byte, 0, unix.CmsgSpace(unix.SizeofInet6Pktinfo))
-			ep.src.ifidx = int32(info.Ifindex)
+			}
 
+			ep.src = ep.src[:unix.CmsgSpace(unix.SizeofInet6Pktinfo)]
+
+			hdrBuf := unsafe.Slice((*byte)(unsafe.Pointer(&hdr)), unix.SizeofCmsghdr)
+			copy(ep.src, hdrBuf)
+			copy(ep.src[unix.CmsgLen(0):], data)
 			return
 		}
 	}
 }
 
-// pktInfoFromBuf returns type T populated from the provided buf via copy(). It
-// panics if buf is of insufficient size.
-func pktInfoFromBuf[T unix.Inet4Pktinfo | unix.Inet6Pktinfo](buf []byte) (t T) {
-	size := int(unsafe.Sizeof(t))
-	if len(buf) < size {
-		panic("pktInfoFromBuf: buffer too small")
-	}
-	copy(unsafe.Slice((*byte)(unsafe.Pointer(&t)), size), buf)
-	return t
-}
-
 // setSrcControl sets an IP{V6}_PKTINFO in control based on the source address
 // and source ifindex found in ep. control's len will be set to 0 in the event
 // that ep is a default value.
 func setSrcControl(control *[]byte, ep *StdNetEndpoint) {
-	*control = (*control)[:cap(*control)]
+	if cap(*control) < len(ep.src) {
-	if len(*control) < int(unsafe.Sizeof(unix.Cmsghdr{})) {
-		*control = (*control)[:0]
 		return
 	}
-
+	*control = (*control)[:0]
-	if ep.src.ifidx == 0 && !ep.SrcIP().IsValid() {
+	*control = append(*control, ep.src...)
-		*control = (*control)[:0]
-		return
-	}
-
-	if len(*control) < srcControlSize {
-		*control = (*control)[:0]
-		return
-	}
-
-	hdr := (*unix.Cmsghdr)(unsafe.Pointer(&(*control)[0]))
-	if ep.SrcIP().Is4() {
-		hdr.Level = unix.IPPROTO_IP
-		hdr.Type = unix.IP_PKTINFO
-		hdr.SetLen(unix.CmsgLen(unix.SizeofInet4Pktinfo))
-
-		info := (*unix.Inet4Pktinfo)(unsafe.Pointer(&(*control)[unix.SizeofCmsghdr]))
-		info.Ifindex = ep.src.ifidx
-		if ep.SrcIP().IsValid() {
-			info.Spec_dst = ep.SrcIP().As4()
-		}
-		*control = (*control)[:unix.CmsgSpace(unix.SizeofInet4Pktinfo)]
-	} else {
-		hdr.Level = unix.IPPROTO_IPV6
-		hdr.Type = unix.IPV6_PKTINFO
-		hdr.SetLen(unix.CmsgLen(unix.SizeofInet6Pktinfo))
-
-		info := (*unix.Inet6Pktinfo)(unsafe.Pointer(&(*control)[unix.SizeofCmsghdr]))
-		info.Ifindex = uint32(ep.src.ifidx)
-		if ep.SrcIP().IsValid() {
-			info.Addr = ep.SrcIP().As16()
-		}
-		*control = (*control)[:unix.CmsgSpace(unix.SizeofInet6Pktinfo)]
-	}
-
 }
 
-var srcControlSize = unix.CmsgSpace(unix.SizeofInet6Pktinfo)
+// stickyControlSize returns the recommended buffer size for pooling sticky
+// offloading control data.
+var stickyControlSize = unix.CmsgSpace(unix.SizeofInet6Pktinfo)
 
 const StdNetSupportsStickySockets = true