weird line breaks

* Fix incorrect CIDR construction in hostmap

* Introduce a regression test for incorrect hostmap CIDR

bits.go

@@ -5,7 +5,6 @@ import (
 	"github.com/sirupsen/logrus"
 )
 
-// TODO: Pretty sure this is just all sorts of racy now, we need it to be atomic
 type Bits struct {
 	length             uint64
 	current            uint64
@@ -44,7 +43,7 @@ func (b *Bits) Check(l logrus.FieldLogger, i uint64) bool {
 	}
 
 	// Not within the window
-	l.Error("rejected a packet (top) %d %d\n", b.current, i)
+	l.Debugf("rejected a packet (top) %d %d\n", b.current, i)
 	return false
 }
 

cmd/gso/gso.go

@@ -1,191 +0,0 @@
-package main
-
-import (
-	"encoding/binary"
-	"errors"
-	"flag"
-	"fmt"
-	"log"
-	"net"
-	"net/netip"
-	"time"
-	"unsafe"
-
-	"golang.org/x/sys/unix"
-)
-
-const (
-	// UDP_SEGMENT enables GSO segmentation
-	UDP_SEGMENT = 103
-	// Maximum GSO segment size (typical MTU - headers)
-	maxGSOSize = 1400
-)
-
-func main() {
-	destAddr := flag.String("dest", "10.4.0.16:4202", "Destination address")
-	gsoSize := flag.Int("gso", 1400, "GSO segment size")
-	totalSize := flag.Int("size", 14000, "Total payload size to send")
-	count := flag.Int("count", 1, "Number of packets to send")
-	flag.Parse()
-
-	if *gsoSize > maxGSOSize {
-		log.Fatalf("GSO size %d exceeds maximum %d", *gsoSize, maxGSOSize)
-	}
-
-	// Resolve destination address
-	_, err := net.ResolveUDPAddr("udp", *destAddr)
-	if err != nil {
-		log.Fatalf("Failed to resolve address: %v", err)
-	}
-
-	// Create a raw UDP socket with GSO support
-	fd, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_UDP)
-	if err != nil {
-		log.Fatalf("Failed to create socket: %v", err)
-	}
-	defer unix.Close(fd)
-
-	// Bind to a local address
-	localAddr := &unix.SockaddrInet4{
-		Port: 0, // Let the system choose a port
-	}
-	if err := unix.Bind(fd, localAddr); err != nil {
-		log.Fatalf("Failed to bind socket: %v", err)
-	}
-
-	fmt.Printf("Sending UDP packets with GSO enabled\n")
-	fmt.Printf("Destination: %s\n", *destAddr)
-	fmt.Printf("GSO segment size: %d bytes\n", *gsoSize)
-	fmt.Printf("Total payload size: %d bytes\n", *totalSize)
-	fmt.Printf("Number of packets: %d\n\n", *count)
-
-	// Create payload
-	payload := make([]byte, *totalSize)
-	for i := range payload {
-		payload[i] = byte(i % 256)
-	}
-
-	dest := netip.MustParseAddrPort(*destAddr)
-
-	//if err := unix.SetsockoptInt(fd, unix.SOL_UDP, unix.UDP_SEGMENT, 1400); err != nil {
-	//	panic(err)
-	//}
-
-	for i := 0; i < *count; i++ {
-		err := WriteBatch(fd, payload, dest, uint16(*gsoSize), true)
-		if err != nil {
-			log.Printf("Send error on packet %d: %v", i, err)
-			continue
-		}
-
-		if (i+1)%100 == 0 || i == *count-1 {
-			fmt.Printf("Sent %d packets\n", i+1)
-		}
-	}
-	fmt.Printf("now, let's send without the correct ctrl header\n")
-	time.Sleep(time.Second)
-	for i := 0; i < *count; i++ {
-		err := WriteBatch(fd, payload, dest, uint16(*gsoSize), false)
-		if err != nil {
-			log.Printf("Send error on packet %d: %v", i, err)
-			continue
-		}
-
-		if (i+1)%100 == 0 || i == *count-1 {
-			fmt.Printf("Sent %d packets\n", i+1)
-		}
-	}
-
-}
-
-func WriteBatch(fd int, payload []byte, addr netip.AddrPort, segSize uint16, withHeader bool) error {
-	msgs := make([]rawMessage, 0, 1)
-	iovs := make([]iovec, 0, 1)
-	names := make([][unix.SizeofSockaddrInet6]byte, 0, 1)
-
-	sent := 0
-
-	pkts := []BatchPacket{
-		{
-			Payload: payload,
-			Addr:    addr,
-		},
-	}
-
-	for _, pkt := range pkts {
-		if len(pkt.Payload) == 0 {
-			sent++
-			continue
-		}
-
-		msgs = append(msgs, rawMessage{})
-		iovs = append(iovs, iovec{})
-		names = append(names, [unix.SizeofSockaddrInet6]byte{})
-
-		idx := len(msgs) - 1
-		msg := &msgs[idx]
-		iov := &iovs[idx]
-		name := &names[idx]
-
-		setIovecSlice(iov, pkt.Payload)
-		msg.Hdr.Iov = iov
-		msg.Hdr.Iovlen = 1
-
-		if withHeader {
-			setRawMessageControl(msg, buildGSOControlMessage(segSize)) //
-		} else {
-			setRawMessageControl(msg, nil) //
-		}
-
-		msg.Hdr.Flags = 0
-
-		nameLen, err := encodeSockaddr(name[:], pkt.Addr)
-		if err != nil {
-			return err
-		}
-		msg.Hdr.Name = &name[0]
-		msg.Hdr.Namelen = nameLen
-	}
-
-	if len(msgs) == 0 {
-		return errors.New("nothing to write")
-	}
-
-	offset := 0
-	for offset < len(msgs) {
-		n, _, errno := unix.Syscall6(
-			unix.SYS_SENDMMSG,
-			uintptr(fd),
-			uintptr(unsafe.Pointer(&msgs[offset])),
-			uintptr(len(msgs)-offset),
-			0,
-			0,
-			0,
-		)
-
-		if errno != 0 {
-			if errno == unix.EINTR {
-				continue
-			}
-			return &net.OpError{Op: "sendmmsg", Err: errno}
-		}
-
-		if n == 0 {
-			break
-		}
-		offset += int(n)
-	}
-
-	return nil
-}
-
-func buildGSOControlMessage(segSize uint16) []byte {
-	control := make([]byte, unix.CmsgSpace(2))
-	hdr := (*unix.Cmsghdr)(unsafe.Pointer(&control[0]))
-	hdr.Level = unix.SOL_UDP
-	hdr.Type = unix.UDP_SEGMENT
-	setCmsgLen(hdr, unix.CmsgLen(2))
-	binary.NativeEndian.PutUint16(control[unix.CmsgLen(0):unix.CmsgLen(0)+2], uint16(segSize))
-
-	return control
-}

cmd/gso/helper.go

@@ -1,85 +0,0 @@
-package main
-
-import (
-	"encoding/binary"
-	"fmt"
-	"net/netip"
-	"unsafe"
-
-	"golang.org/x/sys/unix"
-)
-
-type iovec struct {
-	Base *byte
-	Len  uint64
-}
-
-type msghdr struct {
-	Name       *byte
-	Namelen    uint32
-	Pad0       [4]byte
-	Iov        *iovec
-	Iovlen     uint64
-	Control    *byte
-	Controllen uint64
-	Flags      int32
-	Pad1       [4]byte
-}
-
-type rawMessage struct {
-	Hdr  msghdr
-	Len  uint32
-	Pad0 [4]byte
-}
-
-type BatchPacket struct {
-	Payload []byte
-	Addr    netip.AddrPort
-}
-
-func encodeSockaddr(dst []byte, addr netip.AddrPort) (uint32, error) {
-	if addr.Addr().Is4() {
-		if !addr.Addr().Is4() {
-			return 0, fmt.Errorf("Listener is IPv4, but writing to IPv6 remote")
-		}
-		var sa unix.RawSockaddrInet4
-		sa.Family = unix.AF_INET
-		sa.Addr = addr.Addr().As4()
-		binary.BigEndian.PutUint16((*[2]byte)(unsafe.Pointer(&sa.Port))[:], addr.Port())
-		size := unix.SizeofSockaddrInet4
-		copy(dst[:size], (*(*[unix.SizeofSockaddrInet4]byte)(unsafe.Pointer(&sa)))[:])
-		return uint32(size), nil
-	}
-
-	var sa unix.RawSockaddrInet6
-	sa.Family = unix.AF_INET6
-	sa.Addr = addr.Addr().As16()
-	binary.BigEndian.PutUint16((*[2]byte)(unsafe.Pointer(&sa.Port))[:], addr.Port())
-	size := unix.SizeofSockaddrInet6
-	copy(dst[:size], (*(*[unix.SizeofSockaddrInet6]byte)(unsafe.Pointer(&sa)))[:])
-	return uint32(size), nil
-}
-
-func setRawMessageControl(msg *rawMessage, buf []byte) {
-	if len(buf) == 0 {
-		msg.Hdr.Control = nil
-		msg.Hdr.Controllen = 0
-		return
-	}
-	msg.Hdr.Control = &buf[0]
-	msg.Hdr.Controllen = uint64(len(buf))
-}
-
-func setCmsgLen(h *unix.Cmsghdr, l int) {
-	h.Len = uint64(l)
-}
-
-func setIovecSlice(iov *iovec, b []byte) {
-	if len(b) == 0 {
-		iov.Base = nil
-		iov.Len = 0
-		return
-	}
-	iov.Base = &b[0]
-	iov.Len = uint64(len(b))
-}

cmd/nebula-service/main.go

@@ -65,16 +65,8 @@ func main() {
 	}
 
 	if !*configTest {
-		wait, err := ctrl.Start()
-		if err != nil {
-			util.LogWithContextIfNeeded("Error while running", err, l)
-			os.Exit(1)
-		}
-
-		go ctrl.ShutdownBlock()
-		wait()
-
-		l.Info("Goodbye")
+		ctrl.Start()
+		ctrl.ShutdownBlock()
 	}
 
 	os.Exit(0)

cmd/nebula/main.go

@@ -3,9 +3,6 @@ package main
 import (
 	"flag"
 	"fmt"
-	"log"
-	"net/http"
-	_ "net/http/pprof"
 	"os"
 
 	"github.com/sirupsen/logrus"
@@ -61,22 +58,10 @@ func main() {
 		os.Exit(1)
 	}
 
-	go func() {
-		log.Println(http.ListenAndServe("0.0.0.0:6060", nil))
-	}()
-
 	if !*configTest {
-		wait, err := ctrl.Start()
-		if err != nil {
-			util.LogWithContextIfNeeded("Error while running", err, l)
-			os.Exit(1)
-		}
-
-		go ctrl.ShutdownBlock()
+		ctrl.Start()
 		notifyReady(l)
-		wait()
-
-		l.Info("Goodbye")
+		ctrl.ShutdownBlock()
 	}
 
 	os.Exit(0)

connection_state.go

@@ -13,9 +13,7 @@ import (
 	"github.com/slackhq/nebula/noiseutil"
 )
 
-// TODO: In a 5Gbps test, 1024 is not sufficient. With a 1400 MTU this is about 1.4Gbps of window, assuming full packets.
-// 4092 should be sufficient for 5Gbps
-const ReplayWindow = 8192
+const ReplayWindow = 1024
 
 type ConnectionState struct {
 	eKey           *NebulaCipherState

control.go

@@ -2,11 +2,9 @@ package nebula
 
 import (
 	"context"
-	"errors"
 	"net/netip"
 	"os"
 	"os/signal"
-	"sync"
 	"syscall"
 
 	"github.com/sirupsen/logrus"
@@ -15,16 +13,6 @@ import (
 	"github.com/slackhq/nebula/overlay"
 )
 
-type RunState int
-
-const (
-	Stopped  RunState = 0 // The control has yet to be started
-	Started  RunState = 1 // The control has been started
-	Stopping RunState = 2 // The control is stopping
-)
-
-var ErrAlreadyStarted = errors.New("nebula is already started")
-
 // Every interaction here needs to take extra care to copy memory and not return or use arguments "as is" when touching
 // core. This means copying IP objects, slices, de-referencing pointers and taking the actual value, etc
 
@@ -38,9 +26,6 @@ type controlHostLister interface {
 }
 
 type Control struct {
-	stateLock sync.Mutex
-	state     RunState
-
 	f                      *Interface
 	l                      *logrus.Logger
 	ctx                    context.Context
@@ -64,21 +49,10 @@ type ControlHostInfo struct {
 	CurrentRelaysThroughMe []netip.Addr     `json:"currentRelaysThroughMe"`
 }
 
-// Start actually runs nebula, this is a nonblocking call.
-// The returned function can be used to wait for nebula to fully stop.
-func (c *Control) Start() (func(), error) {
-	c.stateLock.Lock()
-	if c.state != Stopped {
-		c.stateLock.Unlock()
-		return nil, ErrAlreadyStarted
-	}
-
+// Start actually runs nebula, this is a nonblocking call. To block use Control.ShutdownBlock()
+func (c *Control) Start() {
 	// Activate the interface
-	err := c.f.activate()
-	if err != nil {
-		c.stateLock.Unlock()
-		return nil, err
-	}
+	c.f.activate()
 
 	// Call all the delayed funcs that waited patiently for the interface to be created.
 	if c.sshStart != nil {
@@ -98,33 +72,15 @@ func (c *Control) Start() (func(), error) {
 	}
 
 	// Start reading packets.
-	c.state = Started
-	c.stateLock.Unlock()
-	return c.f.run(c.ctx)
-}
-
-func (c *Control) State() RunState {
-	c.stateLock.Lock()
-	defer c.stateLock.Unlock()
-	return c.state
+	c.f.run()
 }
 
 func (c *Control) Context() context.Context {
 	return c.ctx
 }
 
-// Stop is a non-blocking call that signals nebula to close all tunnels and shut down
+// Stop signals nebula to shutdown and close all tunnels, returns after the shutdown is complete
 func (c *Control) Stop() {
-	c.stateLock.Lock()
-	if c.state != Started {
-		c.stateLock.Unlock()
-		// We are stopping or stopped already
-		return
-	}
-
-	c.state = Stopping
-	c.stateLock.Unlock()
-
 	// Stop the handshakeManager (and other services), to prevent new tunnels from
 	// being created while we're shutting them all down.
 	c.cancel()
@@ -133,7 +89,7 @@ func (c *Control) Stop() {
 	if err := c.f.Close(); err != nil {
 		c.l.WithError(err).Error("Close interface failed")
 	}
-	c.state = Stopped
+	c.l.Info("Goodbye")
 }
 
 // ShutdownBlock will listen for and block on term and interrupt signals, calling Control.Stop() once signalled

e2e/helpers_test.go

@@ -29,8 +29,6 @@ type m = map[string]any
 
 // newSimpleServer creates a nebula instance with many assumptions
 func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name string, sVpnNetworks string, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
-	l := NewTestLogger()
-
 	var vpnNetworks []netip.Prefix
 	for _, sn := range strings.Split(sVpnNetworks, ",") {
 		vpnIpNet, err := netip.ParsePrefix(strings.TrimSpace(sn))
@@ -56,6 +54,25 @@ func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name
 		budpIp[3] = 239
 		udpAddr = netip.AddrPortFrom(netip.AddrFrom16(budpIp), 4242)
 	}
+	return newSimpleServerWithUdp(v, caCrt, caKey, name, sVpnNetworks, udpAddr, overrides)
+}
+
+func newSimpleServerWithUdp(v cert.Version, caCrt cert.Certificate, caKey []byte, name string, sVpnNetworks string, udpAddr netip.AddrPort, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
+	l := NewTestLogger()
+
+	var vpnNetworks []netip.Prefix
+	for _, sn := range strings.Split(sVpnNetworks, ",") {
+		vpnIpNet, err := netip.ParsePrefix(strings.TrimSpace(sn))
+		if err != nil {
+			panic(err)
+		}
+		vpnNetworks = append(vpnNetworks, vpnIpNet)
+	}
+
+	if len(vpnNetworks) == 0 {
+		panic("no vpn networks")
+	}
+
 	_, _, myPrivKey, myPEM := cert_test.NewTestCert(v, cert.Curve_CURVE25519, caCrt, caKey, name, time.Now(), time.Now().Add(5*time.Minute), vpnNetworks, nil, []string{})
 
 	caB, err := caCrt.MarshalPEM()

e2e/tunnels_test.go

@@ -4,6 +4,7 @@
 package e2e
 
 import (
+	"net/netip"
 	"testing"
 	"time"
 
@@ -55,3 +56,50 @@ func TestDropInactiveTunnels(t *testing.T) {
 	myControl.Stop()
 	theirControl.Stop()
 }
+
+func TestCrossStackRelaysWork(t *testing.T) {
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version2, ca, caKey, "me     ", "10.128.0.1/24,fc00::1/64", m{"relay": m{"use_relays": true}})
+	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "relay  ", "10.128.0.128/24,fc00::128/64", m{"relay": m{"am_relay": true}})
+	theirUdp := netip.MustParseAddrPort("10.0.0.2:4242")
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServerWithUdp(cert.Version2, ca, caKey, "them   ", "fc00::2/64", theirUdp, m{"relay": m{"use_relays": true}})
+
+	//myVpnV4 := myVpnIpNet[0]
+	myVpnV6 := myVpnIpNet[1]
+	relayVpnV4 := relayVpnIpNet[0]
+	relayVpnV6 := relayVpnIpNet[1]
+	theirVpnV6 := theirVpnIpNet[0]
+
+	// Teach my how to get to the relay and that their can be reached via the relay
+	myControl.InjectLightHouseAddr(relayVpnV4.Addr(), relayUdpAddr)
+	myControl.InjectLightHouseAddr(relayVpnV6.Addr(), relayUdpAddr)
+	myControl.InjectRelays(theirVpnV6.Addr(), []netip.Addr{relayVpnV6.Addr()})
+	relayControl.InjectLightHouseAddr(theirVpnV6.Addr(), theirUdpAddr)
+
+	// Build a router so we don't have to reason who gets which packet
+	r := router.NewR(t, myControl, relayControl, theirControl)
+	defer r.RenderFlow()
+
+	// Start the servers
+	myControl.Start()
+	relayControl.Start()
+	theirControl.Start()
+
+	t.Log("Trigger a handshake from me to them via the relay")
+	myControl.InjectTunUDPPacket(theirVpnV6.Addr(), 80, myVpnV6.Addr(), 80, []byte("Hi from me"))
+
+	p := r.RouteForAllUntilTxTun(theirControl)
+	r.Log("Assert the tunnel works")
+	assertUdpPacket(t, []byte("Hi from me"), p, myVpnV6.Addr(), theirVpnV6.Addr(), 80, 80)
+
+	t.Log("reply?")
+	theirControl.InjectTunUDPPacket(myVpnV6.Addr(), 80, theirVpnV6.Addr(), 80, []byte("Hi from them"))
+	p = r.RouteForAllUntilTxTun(myControl)
+	assertUdpPacket(t, []byte("Hi from them"), p, theirVpnV6.Addr(), myVpnV6.Addr(), 80, 80)
+
+	r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
+	//t.Log("finish up")
+	//myControl.Stop()
+	//theirControl.Stop()
+	//relayControl.Stop()
+}

firewall_test.go

@@ -692,6 +692,50 @@ func TestFirewall_DropConntrackReload(t *testing.T) {
 	assert.Equal(t, fw.Drop(p, false, &h, cp, nil), ErrNoMatchingRule)
 }
 
+func TestFirewall_DropIPSpoofing(t *testing.T) {
+	l := test.NewLogger()
+	ob := &bytes.Buffer{}
+	l.SetOutput(ob)
+
+	c := cert.CachedCertificate{
+		Certificate: &dummyCert{
+			name:     "host-owner",
+			networks: []netip.Prefix{netip.MustParsePrefix("192.0.2.1/24")},
+		},
+	}
+
+	c1 := cert.CachedCertificate{
+		Certificate: &dummyCert{
+			name:           "host",
+			networks:       []netip.Prefix{netip.MustParsePrefix("192.0.2.2/24")},
+			unsafeNetworks: []netip.Prefix{netip.MustParsePrefix("198.51.100.0/24")},
+		},
+	}
+	h1 := HostInfo{
+		ConnectionState: &ConnectionState{
+			peerCert: &c1,
+		},
+		vpnAddrs: []netip.Addr{c1.Certificate.Networks()[0].Addr()},
+	}
+	h1.buildNetworks(c1.Certificate.Networks(), c1.Certificate.UnsafeNetworks())
+
+	fw := NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
+
+	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 1, 1, []string{}, "", netip.Prefix{}, netip.Prefix{}, "", ""))
+	cp := cert.NewCAPool()
+
+	// Packet spoofed by `c1`. Note that the remote addr is not a valid one.
+	p := firewall.Packet{
+		LocalAddr:  netip.MustParseAddr("192.0.2.1"),
+		RemoteAddr: netip.MustParseAddr("192.0.2.3"),
+		LocalPort:  1,
+		RemotePort: 1,
+		Protocol:   firewall.ProtoUDP,
+		Fragment:   false,
+	}
+	assert.Equal(t, fw.Drop(p, true, &h1, cp, nil), ErrInvalidRemoteIP)
+}
+
 func BenchmarkLookup(b *testing.B) {
 	ml := func(m map[string]struct{}, a [][]string) {
 		for n := 0; n < b.N; n++ {

handshake_manager.go

@@ -292,6 +292,7 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 					idx, err := AddRelay(hm.l, relayHostInfo, hm.mainHostMap, vpnIp, nil, TerminalType, Requested)
 					if err != nil {
 						hostinfo.logger(hm.l).WithField("relay", relay.String()).WithError(err).Info("Failed to add relay to hostmap")
+						continue
 					}
 
 					m := NebulaControl{
@@ -301,37 +302,25 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 
 					switch relayHostInfo.GetCert().Certificate.Version() {
 					case cert.Version1:
-						if !hm.f.myVpnAddrs[0].Is4() {
-							hostinfo.logger(hm.l).Error("can not establish v1 relay with a v6 network because the relay is not running a current nebula version")
-							continue
-						}
-
-						if !vpnIp.Is4() {
-							hostinfo.logger(hm.l).Error("can not establish v1 relay with a v6 remote network because the relay is not running a current nebula version")
-							continue
-						}
-
-						b := hm.f.myVpnAddrs[0].As4()
-						m.OldRelayFromAddr = binary.BigEndian.Uint32(b[:])
-						b = vpnIp.As4()
-						m.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
+						err = buildRelayInfoCertV1(&m, hm.f.myVpnNetworks, vpnIp)
 					case cert.Version2:
-						m.RelayFromAddr = netAddrToProtoAddr(hm.f.myVpnAddrs[0])
-						m.RelayToAddr = netAddrToProtoAddr(vpnIp)
+						err = buildRelayInfoCertV2(&m, hm.f.myVpnNetworks, vpnIp)
 					default:
-						hostinfo.logger(hm.l).Error("Unknown certificate version found while creating relay")
+						err = errors.New("unknown certificate version found while creating relay")
+					}
+					if err != nil {
+						hostinfo.logger(hm.l).WithError(err).Error("Refusing to relay")
 						continue
 					}
 
 					msg, err := m.Marshal()
 					if err != nil {
-						hostinfo.logger(hm.l).
-							WithError(err).
+						hostinfo.logger(hm.l).WithError(err).
 							Error("Failed to marshal Control message to create relay")
 					} else {
 						hm.f.SendMessageToHostInfo(header.Control, 0, relayHostInfo, msg, make([]byte, 12), make([]byte, mtu))
 						hm.l.WithFields(logrus.Fields{
-							"relayFrom":           hm.f.myVpnAddrs[0],
+							"relayFrom":           m.GetRelayFrom(),
 							"relayTo":             vpnIp,
 							"initiatorRelayIndex": idx,
 							"relay":               relay}).
@@ -357,39 +346,27 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 					InitiatorRelayIndex: existingRelay.LocalIndex,
 				}
 
+				var err error
 				switch relayHostInfo.GetCert().Certificate.Version() {
 				case cert.Version1:
-					if !hm.f.myVpnAddrs[0].Is4() {
-						hostinfo.logger(hm.l).Error("can not establish v1 relay with a v6 network because the relay is not running a current nebula version")
-						continue
-					}
-
-					if !vpnIp.Is4() {
-						hostinfo.logger(hm.l).Error("can not establish v1 relay with a v6 remote network because the relay is not running a current nebula version")
-						continue
-					}
-
-					b := hm.f.myVpnAddrs[0].As4()
-					m.OldRelayFromAddr = binary.BigEndian.Uint32(b[:])
-					b = vpnIp.As4()
-					m.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
+					err = buildRelayInfoCertV1(&m, hm.f.myVpnNetworks, vpnIp)
 				case cert.Version2:
-					m.RelayFromAddr = netAddrToProtoAddr(hm.f.myVpnAddrs[0])
-					m.RelayToAddr = netAddrToProtoAddr(vpnIp)
+					err = buildRelayInfoCertV2(&m, hm.f.myVpnNetworks, vpnIp)
 				default:
-					hostinfo.logger(hm.l).Error("Unknown certificate version found while creating relay")
+					err = errors.New("unknown certificate version found while creating relay")
+				}
+				if err != nil {
+					hostinfo.logger(hm.l).WithError(err).Error("Refusing to relay")
 					continue
 				}
 				msg, err := m.Marshal()
 				if err != nil {
-					hostinfo.logger(hm.l).
-						WithError(err).
-						Error("Failed to marshal Control message to create relay")
+					hostinfo.logger(hm.l).WithError(err).Error("Failed to marshal Control message to create relay")
 				} else {
 					// This must send over the hostinfo, not over hm.Hosts[ip]
 					hm.f.SendMessageToHostInfo(header.Control, 0, relayHostInfo, msg, make([]byte, 12), make([]byte, mtu))
 					hm.l.WithFields(logrus.Fields{
-						"relayFrom":           hm.f.myVpnAddrs[0],
+						"relayFrom":           m.GetRelayFrom(),
 						"relayTo":             vpnIp,
 						"initiatorRelayIndex": existingRelay.LocalIndex,
 						"relay":               relay}).
@@ -724,3 +701,32 @@ func generateIndex(l *logrus.Logger) (uint32, error) {
 func hsTimeout(tries int64, interval time.Duration) time.Duration {
 	return time.Duration(tries / 2 * ((2 * int64(interval)) + (tries-1)*int64(interval)))
 }
+
+var errNoRelayTooOld = errors.New("can not establish v1 relay with a v6 network because the relay is not running a current nebula version")
+
+func buildRelayInfoCertV1(m *NebulaControl, myVpnNetworks []netip.Prefix, peerVpnIp netip.Addr) error {
+	relayFrom := myVpnNetworks[0].Addr()
+	if !relayFrom.Is4() {
+		return errNoRelayTooOld
+	}
+	if !peerVpnIp.Is4() {
+		return errNoRelayTooOld
+	}
+
+	b := relayFrom.As4()
+	m.OldRelayFromAddr = binary.BigEndian.Uint32(b[:])
+	b = peerVpnIp.As4()
+	m.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
+	return nil
+}
+
+func buildRelayInfoCertV2(m *NebulaControl, myVpnNetworks []netip.Prefix, peerVpnIp netip.Addr) error {
+	for i := range myVpnNetworks {
+		if myVpnNetworks[i].Contains(peerVpnIp) {
+			m.RelayFromAddr = netAddrToProtoAddr(myVpnNetworks[i].Addr())
+			m.RelayToAddr = netAddrToProtoAddr(peerVpnIp)
+			return nil
+		}
+	}
+	return errors.New("cannot establish relay, no networks in common")
+}

hostmap.go

@@ -512,13 +512,16 @@ func (hm *HostMap) QueryVpnAddr(vpnIp netip.Addr) *HostInfo {
 	return hm.queryVpnAddr(vpnIp, nil)
 }
 
+var errUnableToFindHost = errors.New("unable to find host")
+var errUnableToFindHostWithRelay = errors.New("unable to find host with relay")
+
 func (hm *HostMap) QueryVpnAddrsRelayFor(targetIps []netip.Addr, relayHostIp netip.Addr) (*HostInfo, *Relay, error) {
 	hm.RLock()
 	defer hm.RUnlock()
 
 	h, ok := hm.Hosts[relayHostIp]
 	if !ok {
-		return nil, nil, errors.New("unable to find host")
+		return nil, nil, errUnableToFindHost
 	}
 
 	for h != nil {
@@ -531,7 +534,7 @@ func (hm *HostMap) QueryVpnAddrsRelayFor(targetIps []netip.Addr, relayHostIp net
 		h = h.next
 	}
 
-	return nil, nil, errors.New("unable to find host with relay")
+	return nil, nil, errUnableToFindHostWithRelay
 }
 
 func (hm *HostMap) unlockedDisestablishVpnAddrRelayFor(hi *HostInfo) {
@@ -738,7 +741,8 @@ func (i *HostInfo) buildNetworks(networks, unsafeNetworks []netip.Prefix) {
 
 	i.networks = new(bart.Lite)
 	for _, network := range networks {
-		i.networks.Insert(network)
+		nprefix := netip.PrefixFrom(network.Addr(), network.Addr().BitLen())
+		i.networks.Insert(nprefix)
 	}
 
 	for _, network := range unsafeNetworks {

interface.go

@@ -6,8 +6,8 @@ import (
 	"fmt"
 	"io"
 	"net/netip"
+	"os"
 	"runtime"
-	"sync"
 	"sync/atomic"
 	"time"
 
@@ -18,7 +18,6 @@ import (
 	"github.com/slackhq/nebula/firewall"
 	"github.com/slackhq/nebula/header"
 	"github.com/slackhq/nebula/overlay"
-	"github.com/slackhq/nebula/packet"
 	"github.com/slackhq/nebula/udp"
 )
 
@@ -88,17 +87,12 @@ type Interface struct {
 
 	writers []udp.Conn
 	readers []io.ReadWriteCloser
-	wg      sync.WaitGroup
 
 	metricHandshakes    metrics.Histogram
 	messageMetrics      *MessageMetrics
 	cachedPacketMetrics *cachedPacketMetrics
 
 	l *logrus.Logger
-
-	pktPool  *packet.Pool
-	inbound  chan *packet.Packet
-	outbound chan *packet.Packet
 }
 
 type EncWriter interface {
@@ -200,15 +194,9 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 			dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
 		},
 
-		//TODO: configurable size
-		inbound:  make(chan *packet.Packet, 2048),
-		outbound: make(chan *packet.Packet, 2048),
-
 		l: c.l,
 	}
 
-	ifce.pktPool = packet.GetPool()
-
 	ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
 	ifce.reQueryEvery.Store(c.reQueryEvery)
 	ifce.reQueryWait.Store(int64(c.reQueryWait))
@@ -221,7 +209,7 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 // activate creates the interface on the host. After the interface is created, any
 // other services that want to bind listeners to its IP may do so successfully. However,
 // the interface isn't going to process anything until run() is called.
-func (f *Interface) activate() error {
+func (f *Interface) activate() {
 	// actually turn on tun dev
 
 	addr, err := f.outside.LocalAddr()
@@ -242,46 +230,33 @@ func (f *Interface) activate() error {
 		if i > 0 {
 			reader, err = f.inside.NewMultiQueueReader()
 			if err != nil {
-				return err
+				f.l.Fatal(err)
 			}
 		}
 		f.readers[i] = reader
 	}
 
-	if err = f.inside.Activate(); err != nil {
+	if err := f.inside.Activate(); err != nil {
 		f.inside.Close()
-		return err
+		f.l.Fatal(err)
 	}
-
-	return nil
 }
 
-func (f *Interface) run(c context.Context) (func(), error) {
+func (f *Interface) run() {
+	// Launch n queues to read packets from udp
 	for i := 0; i < f.routines; i++ {
-		// read packets from udp and queue to f.inbound
-		f.wg.Add(1)
 		go f.listenOut(i)
-
-		// Launch n queues to read packets from inside tun dev and queue to f.outbound
-		//todo this never stops f.wg.Add(1)
-		go f.listenIn(f.readers[i], i)
-
-		// Launch n workers to process traffic from f.inbound and smash it onto the inside of the tun
-		f.wg.Add(1)
-		go f.workerIn(i, c)
-		f.wg.Add(1)
-		go f.workerIn(i, c)
-
-		// read from f.outbound and write to UDP (outside the tun)
-		f.wg.Add(1)
-		go f.workerOut(i, c)
 	}
 
-	return f.wg.Wait, nil
+	// Launch n queues to read packets from tun dev
+	for i := 0; i < f.routines; i++ {
+		go f.listenIn(f.readers[i], i)
+	}
 }
 
 func (f *Interface) listenOut(i int) {
 	runtime.LockOSThread()
+
 	var li udp.Conn
 	if i > 0 {
 		li = f.writers[i]
@@ -289,90 +264,41 @@ func (f *Interface) listenOut(i int) {
 		li = f.outside
 	}
 
-	err := li.ListenOut(f.pktPool.Get, f.inbound)
-	if err != nil && !f.closed.Load() {
-		f.l.WithError(err).Error("Error while reading packet inbound packet, closing")
-		//TODO: Trigger Control to close
-	}
+	ctCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
+	lhh := f.lightHouse.NewRequestHandler()
+	plaintext := make([]byte, udp.MTU)
+	h := &header.H{}
+	fwPacket := &firewall.Packet{}
+	nb := make([]byte, 12, 12)
 
-	f.l.Debugf("underlay reader %v is done", i)
-	f.wg.Done()
+	li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
+		f.readOutsidePackets(fromUdpAddr, nil, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
+	})
 }
 
 func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	runtime.LockOSThread()
 
+	packet := make([]byte, mtu)
+	out := make([]byte, mtu)
+	fwPacket := &firewall.Packet{}
+	nb := make([]byte, 12, 12)
+
+	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
+
 	for {
-		p := f.pktPool.Get()
-		n, err := reader.Read(p.Payload)
+		n, err := reader.Read(packet)
 		if err != nil {
-			if !f.closed.Load() {
-				f.l.WithError(err).Error("Error while reading outbound packet, closing")
-				//TODO: Trigger Control to close
-			}
-			break
-		}
-
-		p.Payload = (p.Payload)[:n]
-		//TODO: nonblocking channel write
-		f.outbound <- p
-		//select {
-		//case f.outbound <- p:
-		//default:
-		//	f.l.Error("Dropped packet from outbound channel")
-		//}
-	}
-
-	f.l.Debugf("overlay reader %v is done", i)
-	f.wg.Done()
-}
-
-func (f *Interface) workerIn(i int, ctx context.Context) {
-	lhh := f.lightHouse.NewRequestHandler()
-	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
-	fwPacket2 := &firewall.Packet{}
-	nb2 := make([]byte, 12, 12)
-	result2 := make([]byte, mtu)
-	h := &header.H{}
-
-	for {
-		select {
-		case p := <-f.inbound:
-			if p.SegSize > 0 && p.SegSize < len(p.Payload) {
-				for offset := 0; offset < len(p.Payload); offset += p.SegSize {
-					end := offset + p.SegSize
-					if end > len(p.Payload) {
-						end = len(p.Payload)
-					}
-					f.readOutsidePackets(p.Addr, nil, result2[:0], p.Payload[offset:end], h, fwPacket2, lhh, nb2, i, conntrackCache.Get(f.l))
-				}
-			} else {
-				f.readOutsidePackets(p.Addr, nil, result2[:0], p.Payload, h, fwPacket2, lhh, nb2, i, conntrackCache.Get(f.l))
+			if errors.Is(err, os.ErrClosed) && f.closed.Load() {
+				return
 			}
 
-			f.pktPool.Put(p)
-		case <-ctx.Done():
-			f.wg.Done()
-			return
+			f.l.WithError(err).Error("Error while reading outbound packet")
+			// This only seems to happen when something fatal happens to the fd, so exit.
+			os.Exit(2)
 		}
-	}
-}
 
-func (f *Interface) workerOut(i int, ctx context.Context) {
-	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
-	fwPacket1 := &firewall.Packet{}
-	nb1 := make([]byte, 12, 12)
-	result1 := make([]byte, mtu)
-
-	for {
-		select {
-		case data := <-f.outbound:
-			f.consumeInsidePacket(data.Payload, fwPacket1, nb1, result1, i, conntrackCache.Get(f.l))
-			f.pktPool.Put(data)
-		case <-ctx.Done():
-			f.wg.Done()
-			return
-		}
+		f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
 	}
 }
 
@@ -525,7 +451,6 @@ func (f *Interface) GetCertState() *CertState {
 func (f *Interface) Close() error {
 	f.closed.Store(true)
 
-	// Release the udp readers
 	for _, u := range f.writers {
 		err := u.Close()
 		if err != nil {
@@ -533,13 +458,6 @@ func (f *Interface) Close() error {
 		}
 	}
 
-	// Release the tun readers
-	for _, u := range f.readers {
-		err := u.Close()
-		if err != nil {
-			f.l.WithError(err).Error("Error while closing tun device")
-		}
-	}
-
-	return nil
+	// Release the tun device
+	return f.inside.Close()
 }

lighthouse.go

@@ -1425,7 +1425,7 @@ func (d *NebulaMetaDetails) GetRelays() []netip.Addr {
 	return relays
 }
 
-// FindNetworkUnion returns the first netip.Addr contained in the list of provided netip.Prefix, if able
+// findNetworkUnion returns the first netip.Addr of addrs contained in the list of provided netip.Prefix, if able
 func findNetworkUnion(prefixes []netip.Prefix, addrs []netip.Addr) (netip.Addr, bool) {
 	for i := range prefixes {
 		for j := range addrs {
@@ -1450,3 +1450,13 @@ func (d *NebulaMetaDetails) GetVpnAddrAndVersion() (netip.Addr, cert.Version, er
 		return netip.Addr{}, cert.Version1, ErrBadDetailsVpnAddr
 	}
 }
+
+func (d *NebulaControl) GetRelayFrom() netip.Addr {
+	if d.OldRelayFromAddr != 0 {
+		b := [4]byte{}
+		binary.BigEndian.PutUint32(b[:], d.OldRelayFromAddr)
+		return netip.AddrFrom4(b)
+	} else {
+		return protoAddrToNetAddr(d.RelayFromAddr)
+	}
+}

main.go

@@ -284,14 +284,14 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 	}
 
 	return &Control{
-		f:                      ifce,
-		l:                      l,
-		ctx:                    ctx,
-		cancel:                 cancel,
-		sshStart:               sshStart,
-		statsStart:             statsStart,
-		dnsStart:               dnsStart,
-		lighthouseStart:        lightHouse.StartUpdateWorker,
-		connectionManagerStart: connManager.Start,
+		ifce,
+		l,
+		ctx,
+		cancel,
+		sshStart,
+		statsStart,
+		dnsStart,
+		lightHouse.StartUpdateWorker,
+		connManager.Start,
 	}, nil
 }

outside.go

@@ -29,7 +29,7 @@ func (f *Interface) readOutsidePackets(ip netip.AddrPort, via *ViaSender, out []
 		return
 	}
 
-	//f.l.Error("in packet ", h)
+	//l.Error("in packet ", header, packet[HeaderLen:])
 	if ip.IsValid() {
 		if f.myVpnNetworksTable.Contains(ip.Addr()) {
 			if f.l.Level >= logrus.DebugLevel {
@@ -245,7 +245,6 @@ func (f *Interface) handleHostRoaming(hostinfo *HostInfo, udpAddr netip.AddrPort
 			return
 		}
 
-		//TODO: Seems we have a bunch of stuff racing here, since we don't have a lock on hostinfo anymore we announce roaming in bursts
 		hostinfo.logger(f.l).WithField("udpAddr", hostinfo.remote).WithField("newAddr", udpAddr).
 			Info("Host roamed to new udp ip/port.")
 		hostinfo.lastRoam = time.Now()
@@ -471,7 +470,7 @@ func (f *Interface) decryptToTun(hostinfo *HostInfo, messageCounter uint64, out
 
 	out, err = hostinfo.ConnectionState.dKey.DecryptDanger(out, packet[:header.Len], packet[header.Len:], messageCounter, nb)
 	if err != nil {
-		hostinfo.logger(f.l).WithError(err).WithField("fwPacket", fwPacket).Error("Failed to decrypt packet")
+		hostinfo.logger(f.l).WithError(err).Error("Failed to decrypt packet")
 		return false
 	}
 

overlay/tun.go

@@ -1,6 +1,7 @@
 package overlay
 
 import (
+	"fmt"
 	"net"
 	"net/netip"
 
@@ -81,3 +82,41 @@ func prefixToMask(prefix netip.Prefix) netip.Addr {
 	addr, _ := netip.AddrFromSlice(net.CIDRMask(prefix.Bits(), pLen))
 	return addr
 }
+
+func flipBytes(b []byte) []byte {
+	for i := 0; i < len(b); i++ {
+		b[i] ^= 0xFF
+	}
+	return b
+}
+func orBytes(a []byte, b []byte) []byte {
+	ret := make([]byte, len(a))
+	for i := 0; i < len(a); i++ {
+		ret[i] = a[i] | b[i]
+	}
+	return ret
+}
+
+func getBroadcast(cidr netip.Prefix) netip.Addr {
+	broadcast, _ := netip.AddrFromSlice(
+		orBytes(
+			cidr.Addr().AsSlice(),
+			flipBytes(prefixToMask(cidr).AsSlice()),
+		),
+	)
+	return broadcast
+}
+
+func selectGateway(dest netip.Prefix, gateways []netip.Prefix) (netip.Prefix, error) {
+	for _, gateway := range gateways {
+		if dest.Addr().Is4() && gateway.Addr().Is4() {
+			return gateway, nil
+		}
+
+		if dest.Addr().Is6() && gateway.Addr().Is6() {
+			return gateway, nil
+		}
+	}
+
+	return netip.Prefix{}, fmt.Errorf("no gateway found for %v in the list of vpn networks", dest)
+}

overlay/tun_darwin.go

@@ -294,7 +294,6 @@ func (t *tun) activate6(network netip.Prefix) error {
 			Vltime: 0xffffffff,
 			Pltime: 0xffffffff,
 		},
-		//TODO: CERT-V2 should we disable DAD (duplicate address detection) and mark this as a secured address?
 		Flags: _IN6_IFF_NODAD,
 	}
 

overlay/tun_freebsd.go

@@ -501,30 +501,6 @@ func (t *tun) deviceBytes() (o [16]byte) {
 	return
 }
 
-func flipBytes(b []byte) []byte {
-	for i := 0; i < len(b); i++ {
-		b[i] ^= 0xFF
-	}
-	return b
-}
-func orBytes(a []byte, b []byte) []byte {
-	ret := make([]byte, len(a))
-	for i := 0; i < len(a); i++ {
-		ret[i] = a[i] | b[i]
-	}
-	return ret
-}
-
-func getBroadcast(cidr netip.Prefix) netip.Addr {
-	broadcast, _ := netip.AddrFromSlice(
-		orBytes(
-			cidr.Addr().AsSlice(),
-			flipBytes(prefixToMask(cidr).AsSlice()),
-		),
-	)
-	return broadcast
-}
-
 func addRoute(prefix netip.Prefix, gateway netroute.Addr) error {
 	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
 	if err != nil {

overlay/tun_netbsd.go

@@ -4,13 +4,12 @@
 package overlay
 
 import (
+	"errors"
 	"fmt"
 	"io"
 	"net/netip"
 	"os"
-	"os/exec"
 	"regexp"
-	"strconv"
 	"sync/atomic"
 	"syscall"
 	"unsafe"
@@ -20,11 +19,42 @@ import (
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/routing"
 	"github.com/slackhq/nebula/util"
+	netroute "golang.org/x/net/route"
+	"golang.org/x/sys/unix"
 )
 
-type ifreqDestroy struct {
-	Name [16]byte
-	pad  [16]byte
+const (
+	SIOCAIFADDR_IN6 = 0x8080696b
+	TUNSIFHEAD      = 0x80047442
+	TUNSIFMODE      = 0x80047458
+)
+
+type ifreqAlias4 struct {
+	Name     [unix.IFNAMSIZ]byte
+	Addr     unix.RawSockaddrInet4
+	DstAddr  unix.RawSockaddrInet4
+	MaskAddr unix.RawSockaddrInet4
+}
+
+type ifreqAlias6 struct {
+	Name       [unix.IFNAMSIZ]byte
+	Addr       unix.RawSockaddrInet6
+	DstAddr    unix.RawSockaddrInet6
+	PrefixMask unix.RawSockaddrInet6
+	Flags      uint32
+	Lifetime   addrLifetime
+}
+
+type ifreq struct {
+	Name [unix.IFNAMSIZ]byte
+	data int
+}
+
+type addrLifetime struct {
+	Expire    uint64
+	Preferred uint64
+	Vltime    uint32
+	Pltime    uint32
 }
 
 type tun struct {
@@ -34,40 +64,18 @@ type tun struct {
 	Routes      atomic.Pointer[[]Route]
 	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
 	l           *logrus.Logger
-
-	io.ReadWriteCloser
+	f           *os.File
+	fd          int
 }
 
-func (t *tun) Close() error {
-	if t.ReadWriteCloser != nil {
-		if err := t.ReadWriteCloser.Close(); err != nil {
-			return err
-		}
-
-		s, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_DGRAM, syscall.IPPROTO_IP)
-		if err != nil {
-			return err
-		}
-		defer syscall.Close(s)
-
-		ifreq := ifreqDestroy{Name: t.deviceBytes()}
-
-		err = ioctl(uintptr(s), syscall.SIOCIFDESTROY, uintptr(unsafe.Pointer(&ifreq)))
-
-		return err
-	}
-	return nil
-}
+var deviceNameRE = regexp.MustCompile(`^tun[0-9]+$`)
 
 func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in NetBSD")
 }
 
-var deviceNameRE = regexp.MustCompile(`^tun[0-9]+$`)
-
 func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
 	// Try to open tun device
-	var file *os.File
 	var err error
 	deviceName := c.GetString("tun.dev", "")
 	if deviceName == "" {
@@ -77,17 +85,23 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 		return nil, fmt.Errorf("a device name in the format of /dev/tunN must be specified")
 	}
 
-	file, err = os.OpenFile("/dev/"+deviceName, os.O_RDWR, 0)
+	fd, err := unix.Open("/dev/"+deviceName, os.O_RDWR, 0)
 	if err != nil {
 		return nil, err
 	}
 
+	err = unix.SetNonblock(fd, true)
+	if err != nil {
+		l.WithError(err).Warn("Failed to set the tun device as nonblocking")
+	}
+
 	t := &tun{
-		ReadWriteCloser: file,
-		Device:          deviceName,
-		vpnNetworks:     vpnNetworks,
-		MTU:             c.GetInt("tun.mtu", DefaultMTU),
-		l:               l,
+		f:           os.NewFile(uintptr(fd), ""),
+		fd:          fd,
+		Device:      deviceName,
+		vpnNetworks: vpnNetworks,
+		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		l:           l,
 	}
 
 	err = t.reload(c, true)
@@ -105,40 +119,225 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 	return t, nil
 }
 
+func (t *tun) Close() error {
+	if t.f != nil {
+		if err := t.f.Close(); err != nil {
+			return fmt.Errorf("error closing tun file: %w", err)
+		}
+
+		// t.f.Close should have handled it for us but let's be extra sure
+		_ = unix.Close(t.fd)
+
+		s, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_DGRAM, syscall.IPPROTO_IP)
+		if err != nil {
+			return err
+		}
+		defer syscall.Close(s)
+
+		ifr := ifreq{Name: t.deviceBytes()}
+		err = ioctl(uintptr(s), syscall.SIOCIFDESTROY, uintptr(unsafe.Pointer(&ifr)))
+		return err
+	}
+	return nil
+}
+
+func (t *tun) Read(to []byte) (int, error) {
+	rc, err := t.f.SyscallConn()
+	if err != nil {
+		return 0, fmt.Errorf("failed to get syscall conn for tun: %w", err)
+	}
+
+	var errno syscall.Errno
+	var n uintptr
+	err = rc.Read(func(fd uintptr) bool {
+		// first 4 bytes is protocol family, in network byte order
+		head := [4]byte{}
+		iovecs := []syscall.Iovec{
+			{&head[0], 4},
+			{&to[0], uint64(len(to))},
+		}
+
+		n, _, errno = syscall.Syscall(syscall.SYS_READV, fd, uintptr(unsafe.Pointer(&iovecs[0])), uintptr(2))
+		if errno.Temporary() {
+			// We got an EAGAIN, EINTR, or EWOULDBLOCK, go again
+			return false
+		}
+		return true
+	})
+	if err != nil {
+		if err == syscall.EBADF || err.Error() == "use of closed file" {
+			// Go doesn't export poll.ErrFileClosing but happily reports it to us so here we are
+			// https://github.com/golang/go/blob/master/src/internal/poll/fd_poll_runtime.go#L121
+			return 0, os.ErrClosed
+		}
+		return 0, fmt.Errorf("failed to make read call for tun: %w", err)
+	}
+
+	if errno != 0 {
+		return 0, fmt.Errorf("failed to make inner read call for tun: %w", errno)
+	}
+
+	// fix bytes read number to exclude header
+	bytesRead := int(n)
+	if bytesRead < 0 {
+		return bytesRead, nil
+	} else if bytesRead < 4 {
+		return 0, nil
+	} else {
+		return bytesRead - 4, nil
+	}
+}
+
+// Write is only valid for single threaded use
+func (t *tun) Write(from []byte) (int, error) {
+	if len(from) <= 1 {
+		return 0, syscall.EIO
+	}
+
+	ipVer := from[0] >> 4
+	var head [4]byte
+	// first 4 bytes is protocol family, in network byte order
+	if ipVer == 4 {
+		head[3] = syscall.AF_INET
+	} else if ipVer == 6 {
+		head[3] = syscall.AF_INET6
+	} else {
+		return 0, fmt.Errorf("unable to determine IP version from packet")
+	}
+
+	rc, err := t.f.SyscallConn()
+	if err != nil {
+		return 0, err
+	}
+
+	var errno syscall.Errno
+	var n uintptr
+	err = rc.Write(func(fd uintptr) bool {
+		iovecs := []syscall.Iovec{
+			{&head[0], 4},
+			{&from[0], uint64(len(from))},
+		}
+
+		n, _, errno = syscall.Syscall(syscall.SYS_WRITEV, fd, uintptr(unsafe.Pointer(&iovecs[0])), uintptr(2))
+		// According to NetBSD documentation for TUN, writes will only return errors in which
+		// this packet will never be delivered so just go on living life.
+		return true
+	})
+	if err != nil {
+		return 0, err
+	}
+
+	if errno != 0 {
+		return 0, errno
+	}
+
+	return int(n) - 4, err
+}
+
 func (t *tun) addIp(cidr netip.Prefix) error {
-	var err error
+	if cidr.Addr().Is4() {
+		var req ifreqAlias4
+		req.Name = t.deviceBytes()
+		req.Addr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   cidr.Addr().As4(),
+		}
+		req.DstAddr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   cidr.Addr().As4(),
+		}
+		req.MaskAddr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   prefixToMask(cidr).As4(),
+		}
 
-	// TODO use syscalls instead of exec.Command
-	cmd := exec.Command("/sbin/ifconfig", t.Device, cidr.String(), cidr.Addr().String())
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'ifconfig': %s", err)
+		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+		if err != nil {
+			return err
+		}
+		defer syscall.Close(s)
+
+		if err := ioctl(uintptr(s), unix.SIOCAIFADDR, uintptr(unsafe.Pointer(&req))); err != nil {
+			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr(), err)
+		}
+
+		return nil
 	}
 
-	cmd = exec.Command("/sbin/route", "-n", "add", "-net", cidr.String(), cidr.Addr().String())
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'route add': %s", err)
+	if cidr.Addr().Is6() {
+		var req ifreqAlias6
+		req.Name = t.deviceBytes()
+		req.Addr = unix.RawSockaddrInet6{
+			Len:    unix.SizeofSockaddrInet6,
+			Family: unix.AF_INET6,
+			Addr:   cidr.Addr().As16(),
+		}
+		req.PrefixMask = unix.RawSockaddrInet6{
+			Len:    unix.SizeofSockaddrInet6,
+			Family: unix.AF_INET6,
+			Addr:   prefixToMask(cidr).As16(),
+		}
+		req.Lifetime = addrLifetime{
+			Vltime: 0xffffffff,
+			Pltime: 0xffffffff,
+		}
+
+		s, err := unix.Socket(unix.AF_INET6, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+		if err != nil {
+			return err
+		}
+		defer syscall.Close(s)
+
+		if err := ioctl(uintptr(s), SIOCAIFADDR_IN6, uintptr(unsafe.Pointer(&req))); err != nil {
+			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr().String(), err)
+		}
+		return nil
 	}
 
-	cmd = exec.Command("/sbin/ifconfig", t.Device, "mtu", strconv.Itoa(t.MTU))
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'ifconfig': %s", err)
-	}
-
-	// Unsafe path routes
-	return t.addRoutes(false)
+	return fmt.Errorf("unknown address type %v", cidr)
 }
 
 func (t *tun) Activate() error {
+	mode := int32(unix.IFF_BROADCAST)
+	err := ioctl(uintptr(t.fd), TUNSIFMODE, uintptr(unsafe.Pointer(&mode)))
+	if err != nil {
+		return fmt.Errorf("failed to set tun device mode: %w", err)
+	}
+
+	v := 1
+	err = ioctl(uintptr(t.fd), TUNSIFHEAD, uintptr(unsafe.Pointer(&v)))
+	if err != nil {
+		return fmt.Errorf("failed to set tun device head: %w", err)
+	}
+
+	err = t.doIoctlByName(unix.SIOCSIFMTU, uint32(t.MTU))
+	if err != nil {
+		return fmt.Errorf("failed to set tun mtu: %w", err)
+	}
+
 	for i := range t.vpnNetworks {
-		err := t.addIp(t.vpnNetworks[i])
+		err = t.addIp(t.vpnNetworks[i])
 		if err != nil {
 			return err
 		}
 	}
-	return nil
+
+	return t.addRoutes(false)
+}
+
+func (t *tun) doIoctlByName(ctl uintptr, value uint32) error {
+	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(s)
+
+	ir := ifreq{Name: t.deviceBytes(), data: int(value)}
+	err = ioctl(uintptr(s), ctl, uintptr(unsafe.Pointer(&ir)))
+	return err
 }
 
 func (t *tun) reload(c *config.C, initial bool) error {
@@ -197,21 +396,23 @@ func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 
 func (t *tun) addRoutes(logErrors bool) error {
 	routes := *t.Routes.Load()
+
 	for _, r := range routes {
 		if len(r.Via) == 0 || !r.Install {
 			// We don't allow route MTUs so only install routes with a via
 			continue
 		}
 
-		cmd := exec.Command("/sbin/route", "-n", "add", "-net", r.Cidr.String(), t.vpnNetworks[0].Addr().String())
-		t.l.Debug("command: ", cmd.String())
-		if err := cmd.Run(); err != nil {
-			retErr := util.NewContextualError("failed to run 'route add' for unsafe_route", map[string]any{"route": r}, err)
+		err := addRoute(r.Cidr, t.vpnNetworks)
+		if err != nil {
+			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
 				retErr.Log(t.l)
 			} else {
 				return retErr
 			}
+		} else {
+			t.l.WithField("route", r).Info("Added route")
 		}
 	}
 
@@ -224,10 +425,8 @@ func (t *tun) removeRoutes(routes []Route) error {
 			continue
 		}
 
-		//TODO: CERT-V2 is this right?
-		cmd := exec.Command("/sbin/route", "-n", "delete", "-net", r.Cidr.String(), t.vpnNetworks[0].Addr().String())
-		t.l.Debug("command: ", cmd.String())
-		if err := cmd.Run(); err != nil {
+		err := delRoute(r.Cidr, t.vpnNetworks)
+		if err != nil {
 			t.l.WithError(err).WithField("route", r).Error("Failed to remove route")
 		} else {
 			t.l.WithField("route", r).Info("Removed route")
@@ -242,3 +441,109 @@ func (t *tun) deviceBytes() (o [16]byte) {
 	}
 	return
 }
+
+func addRoute(prefix netip.Prefix, gateways []netip.Prefix) error {
+	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
+	}
+	defer unix.Close(sock)
+
+	route := &netroute.RouteMessage{
+		Version: unix.RTM_VERSION,
+		Type:    unix.RTM_ADD,
+		Flags:   unix.RTF_UP | unix.RTF_GATEWAY,
+		Seq:     1,
+	}
+
+	if prefix.Addr().Is4() {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: &netroute.Inet4Addr{IP: gw.Addr().As4()},
+		}
+	} else {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: &netroute.Inet6Addr{IP: gw.Addr().As16()},
+		}
+	}
+
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
+
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		if errors.Is(err, unix.EEXIST) {
+			// Try to do a change
+			route.Type = unix.RTM_CHANGE
+			data, err = route.Marshal()
+			if err != nil {
+				return fmt.Errorf("failed to create route.RouteMessage for change: %w", err)
+			}
+			_, err = unix.Write(sock, data[:])
+			return err
+		}
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
+	return nil
+}
+
+func delRoute(prefix netip.Prefix, gateways []netip.Prefix) error {
+	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
+	}
+	defer unix.Close(sock)
+
+	route := netroute.RouteMessage{
+		Version: unix.RTM_VERSION,
+		Type:    unix.RTM_DELETE,
+		Seq:     1,
+	}
+
+	if prefix.Addr().Is4() {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: &netroute.Inet4Addr{IP: gw.Addr().As4()},
+		}
+	} else {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: &netroute.Inet6Addr{IP: gw.Addr().As16()},
+		}
+	}
+
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
+	return nil
+}

overlay/tun_openbsd.go

@@ -4,23 +4,50 @@
 package overlay
 
 import (
+	"errors"
 	"fmt"
 	"io"
 	"net/netip"
 	"os"
-	"os/exec"
 	"regexp"
-	"strconv"
 	"sync/atomic"
 	"syscall"
+	"unsafe"
 
 	"github.com/gaissmai/bart"
 	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/routing"
 	"github.com/slackhq/nebula/util"
+	netroute "golang.org/x/net/route"
+	"golang.org/x/sys/unix"
 )
 
+const (
+	SIOCAIFADDR_IN6 = 0x8080691a
+)
+
+type ifreqAlias4 struct {
+	Name     [unix.IFNAMSIZ]byte
+	Addr     unix.RawSockaddrInet4
+	DstAddr  unix.RawSockaddrInet4
+	MaskAddr unix.RawSockaddrInet4
+}
+
+type ifreqAlias6 struct {
+	Name       [unix.IFNAMSIZ]byte
+	Addr       unix.RawSockaddrInet6
+	DstAddr    unix.RawSockaddrInet6
+	PrefixMask unix.RawSockaddrInet6
+	Flags      uint32
+	Lifetime   [2]uint32
+}
+
+type ifreq struct {
+	Name [unix.IFNAMSIZ]byte
+	data int
+}
+
 type tun struct {
 	Device      string
 	vpnNetworks []netip.Prefix
@@ -28,48 +55,46 @@ type tun struct {
 	Routes      atomic.Pointer[[]Route]
 	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
 	l           *logrus.Logger
-
-	io.ReadWriteCloser
-
+	f           *os.File
+	fd          int
 	// cache out buffer since we need to prepend 4 bytes for tun metadata
 	out []byte
 }
 
-func (t *tun) Close() error {
-	if t.ReadWriteCloser != nil {
-		return t.ReadWriteCloser.Close()
-	}
-
-	return nil
-}
-
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
-	return nil, fmt.Errorf("newTunFromFd not supported in OpenBSD")
-}
-
 var deviceNameRE = regexp.MustCompile(`^tun[0-9]+$`)
 
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
+	return nil, fmt.Errorf("newTunFromFd not supported in openbsd")
+}
+
 func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
+	// Try to open tun device
+	var err error
 	deviceName := c.GetString("tun.dev", "")
 	if deviceName == "" {
-		return nil, fmt.Errorf("a device name in the format of tunN must be specified")
+		return nil, fmt.Errorf("a device name in the format of /dev/tunN must be specified")
 	}
-
 	if !deviceNameRE.MatchString(deviceName) {
-		return nil, fmt.Errorf("a device name in the format of tunN must be specified")
+		return nil, fmt.Errorf("a device name in the format of /dev/tunN must be specified")
 	}
 
-	file, err := os.OpenFile("/dev/"+deviceName, os.O_RDWR, 0)
+	fd, err := unix.Open("/dev/"+deviceName, os.O_RDWR, 0)
 	if err != nil {
 		return nil, err
 	}
 
+	err = unix.SetNonblock(fd, true)
+	if err != nil {
+		l.WithError(err).Warn("Failed to set the tun device as nonblocking")
+	}
+
 	t := &tun{
-		ReadWriteCloser: file,
-		Device:          deviceName,
-		vpnNetworks:     vpnNetworks,
-		MTU:             c.GetInt("tun.mtu", DefaultMTU),
-		l:               l,
+		f:           os.NewFile(uintptr(fd), ""),
+		fd:          fd,
+		Device:      deviceName,
+		vpnNetworks: vpnNetworks,
+		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		l:           l,
 	}
 
 	err = t.reload(c, true)
@@ -87,6 +112,154 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 	return t, nil
 }
 
+func (t *tun) Close() error {
+	if t.f != nil {
+		if err := t.f.Close(); err != nil {
+			return fmt.Errorf("error closing tun file: %w", err)
+		}
+
+		// t.f.Close should have handled it for us but let's be extra sure
+		_ = unix.Close(t.fd)
+	}
+	return nil
+}
+
+func (t *tun) Read(to []byte) (int, error) {
+	buf := make([]byte, len(to)+4)
+
+	n, err := t.f.Read(buf)
+
+	copy(to, buf[4:])
+	return n - 4, err
+}
+
+// Write is only valid for single threaded use
+func (t *tun) Write(from []byte) (int, error) {
+	buf := t.out
+	if cap(buf) < len(from)+4 {
+		buf = make([]byte, len(from)+4)
+		t.out = buf
+	}
+	buf = buf[:len(from)+4]
+
+	if len(from) == 0 {
+		return 0, syscall.EIO
+	}
+
+	// Determine the IP Family for the NULL L2 Header
+	ipVer := from[0] >> 4
+	if ipVer == 4 {
+		buf[3] = syscall.AF_INET
+	} else if ipVer == 6 {
+		buf[3] = syscall.AF_INET6
+	} else {
+		return 0, fmt.Errorf("unable to determine IP version from packet")
+	}
+
+	copy(buf[4:], from)
+
+	n, err := t.f.Write(buf)
+	return n - 4, err
+}
+
+func (t *tun) addIp(cidr netip.Prefix) error {
+	if cidr.Addr().Is4() {
+		var req ifreqAlias4
+		req.Name = t.deviceBytes()
+		req.Addr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   cidr.Addr().As4(),
+		}
+		req.DstAddr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   cidr.Addr().As4(),
+		}
+		req.MaskAddr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   prefixToMask(cidr).As4(),
+		}
+
+		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+		if err != nil {
+			return err
+		}
+		defer syscall.Close(s)
+
+		if err := ioctl(uintptr(s), unix.SIOCAIFADDR, uintptr(unsafe.Pointer(&req))); err != nil {
+			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr(), err)
+		}
+
+		err = addRoute(cidr, t.vpnNetworks)
+		if err != nil {
+			return fmt.Errorf("failed to set route for vpn network %v: %w", cidr, err)
+		}
+
+		return nil
+	}
+
+	if cidr.Addr().Is6() {
+		var req ifreqAlias6
+		req.Name = t.deviceBytes()
+		req.Addr = unix.RawSockaddrInet6{
+			Len:    unix.SizeofSockaddrInet6,
+			Family: unix.AF_INET6,
+			Addr:   cidr.Addr().As16(),
+		}
+		req.PrefixMask = unix.RawSockaddrInet6{
+			Len:    unix.SizeofSockaddrInet6,
+			Family: unix.AF_INET6,
+			Addr:   prefixToMask(cidr).As16(),
+		}
+		req.Lifetime[0] = 0xffffffff
+		req.Lifetime[1] = 0xffffffff
+
+		s, err := unix.Socket(unix.AF_INET6, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+		if err != nil {
+			return err
+		}
+		defer syscall.Close(s)
+
+		if err := ioctl(uintptr(s), SIOCAIFADDR_IN6, uintptr(unsafe.Pointer(&req))); err != nil {
+			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr().String(), err)
+		}
+
+		return nil
+	}
+
+	return fmt.Errorf("unknown address type %v", cidr)
+}
+
+func (t *tun) Activate() error {
+	err := t.doIoctlByName(unix.SIOCSIFMTU, uint32(t.MTU))
+	if err != nil {
+		return fmt.Errorf("failed to set tun mtu: %w", err)
+	}
+
+	for i := range t.vpnNetworks {
+		err = t.addIp(t.vpnNetworks[i])
+		if err != nil {
+			return err
+		}
+	}
+
+	return t.addRoutes(false)
+}
+
+func (t *tun) doIoctlByName(ctl uintptr, value uint32) error {
+	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(s)
+
+	ir := ifreq{Name: t.deviceBytes(), data: int(value)}
+	err = ioctl(uintptr(s), ctl, uintptr(unsafe.Pointer(&ir)))
+	return err
+}
+
 func (t *tun) reload(c *config.C, initial bool) error {
 	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
 	if err != nil {
@@ -124,86 +297,11 @@ func (t *tun) reload(c *config.C, initial bool) error {
 	return nil
 }
 
-func (t *tun) addIp(cidr netip.Prefix) error {
-	var err error
-	// TODO use syscalls instead of exec.Command
-	cmd := exec.Command("/sbin/ifconfig", t.Device, cidr.String(), cidr.Addr().String())
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'ifconfig': %s", err)
-	}
-
-	cmd = exec.Command("/sbin/ifconfig", t.Device, "mtu", strconv.Itoa(t.MTU))
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'ifconfig': %s", err)
-	}
-
-	cmd = exec.Command("/sbin/route", "-n", "add", "-inet", cidr.String(), cidr.Addr().String())
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'route add': %s", err)
-	}
-
-	// Unsafe path routes
-	return t.addRoutes(false)
-}
-
-func (t *tun) Activate() error {
-	for i := range t.vpnNetworks {
-		err := t.addIp(t.vpnNetworks[i])
-		if err != nil {
-			return err
-		}
-	}
-	return nil
-}
-
 func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
 	r, _ := t.routeTree.Load().Lookup(ip)
 	return r
 }
 
-func (t *tun) addRoutes(logErrors bool) error {
-	routes := *t.Routes.Load()
-	for _, r := range routes {
-		if len(r.Via) == 0 || !r.Install {
-			// We don't allow route MTUs so only install routes with a via
-			continue
-		}
-		//TODO: CERT-V2 is this right?
-		cmd := exec.Command("/sbin/route", "-n", "add", "-inet", r.Cidr.String(), t.vpnNetworks[0].Addr().String())
-		t.l.Debug("command: ", cmd.String())
-		if err := cmd.Run(); err != nil {
-			retErr := util.NewContextualError("failed to run 'route add' for unsafe_route", map[string]any{"route": r}, err)
-			if logErrors {
-				retErr.Log(t.l)
-			} else {
-				return retErr
-			}
-		}
-	}
-
-	return nil
-}
-
-func (t *tun) removeRoutes(routes []Route) error {
-	for _, r := range routes {
-		if !r.Install {
-			continue
-		}
-		//TODO: CERT-V2 is this right?
-		cmd := exec.Command("/sbin/route", "-n", "delete", "-inet", r.Cidr.String(), t.vpnNetworks[0].Addr().String())
-		t.l.Debug("command: ", cmd.String())
-		if err := cmd.Run(); err != nil {
-			t.l.WithError(err).WithField("route", r).Error("Failed to remove route")
-		} else {
-			t.l.WithField("route", r).Info("Removed route")
-		}
-	}
-	return nil
-}
-
 func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
@@ -213,43 +311,159 @@ func (t *tun) Name() string {
 }
 
 func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	return nil, fmt.Errorf("TODO: multiqueue not implemented for freebsd")
+	return nil, fmt.Errorf("TODO: multiqueue not implemented for openbsd")
 }
 
-func (t *tun) Read(to []byte) (int, error) {
-	buf := make([]byte, len(to)+4)
+func (t *tun) addRoutes(logErrors bool) error {
+	routes := *t.Routes.Load()
 
-	n, err := t.ReadWriteCloser.Read(buf)
+	for _, r := range routes {
+		if len(r.Via) == 0 || !r.Install {
+			// We don't allow route MTUs so only install routes with a via
+			continue
+		}
 
-	copy(to, buf[4:])
-	return n - 4, err
+		err := addRoute(r.Cidr, t.vpnNetworks)
+		if err != nil {
+			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
+			if logErrors {
+				retErr.Log(t.l)
+			} else {
+				return retErr
+			}
+		} else {
+			t.l.WithField("route", r).Info("Added route")
+		}
+	}
+
+	return nil
 }
 
-// Write is only valid for single threaded use
-func (t *tun) Write(from []byte) (int, error) {
-	buf := t.out
-	if cap(buf) < len(from)+4 {
-		buf = make([]byte, len(from)+4)
-		t.out = buf
-	}
-	buf = buf[:len(from)+4]
+func (t *tun) removeRoutes(routes []Route) error {
+	for _, r := range routes {
+		if !r.Install {
+			continue
+		}
 
-	if len(from) == 0 {
-		return 0, syscall.EIO
+		err := delRoute(r.Cidr, t.vpnNetworks)
+		if err != nil {
+			t.l.WithError(err).WithField("route", r).Error("Failed to remove route")
+		} else {
+			t.l.WithField("route", r).Info("Removed route")
+		}
+	}
+	return nil
+}
+
+func (t *tun) deviceBytes() (o [16]byte) {
+	for i, c := range t.Device {
+		o[i] = byte(c)
+	}
+	return
+}
+
+func addRoute(prefix netip.Prefix, gateways []netip.Prefix) error {
+	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
+	}
+	defer unix.Close(sock)
+
+	route := &netroute.RouteMessage{
+		Version: unix.RTM_VERSION,
+		Type:    unix.RTM_ADD,
+		Flags:   unix.RTF_UP | unix.RTF_GATEWAY,
+		Seq:     1,
 	}
 
-	// Determine the IP Family for the NULL L2 Header
-	ipVer := from[0] >> 4
-	if ipVer == 4 {
-		buf[3] = syscall.AF_INET
-	} else if ipVer == 6 {
-		buf[3] = syscall.AF_INET6
+	if prefix.Addr().Is4() {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: &netroute.Inet4Addr{IP: gw.Addr().As4()},
+		}
 	} else {
-		return 0, fmt.Errorf("unable to determine IP version from packet")
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: &netroute.Inet6Addr{IP: gw.Addr().As16()},
+		}
 	}
 
-	copy(buf[4:], from)
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
 
-	n, err := t.ReadWriteCloser.Write(buf)
-	return n - 4, err
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		if errors.Is(err, unix.EEXIST) {
+			// Try to do a change
+			route.Type = unix.RTM_CHANGE
+			data, err = route.Marshal()
+			if err != nil {
+				return fmt.Errorf("failed to create route.RouteMessage for change: %w", err)
+			}
+			_, err = unix.Write(sock, data[:])
+			return err
+		}
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
+	return nil
+}
+
+func delRoute(prefix netip.Prefix, gateways []netip.Prefix) error {
+	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
+	}
+	defer unix.Close(sock)
+
+	route := netroute.RouteMessage{
+		Version: unix.RTM_VERSION,
+		Type:    unix.RTM_DELETE,
+		Seq:     1,
+	}
+
+	if prefix.Addr().Is4() {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: &netroute.Inet4Addr{IP: gw.Addr().As4()},
+		}
+	} else {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: &netroute.Inet6Addr{IP: gw.Addr().As16()},
+		}
+	}
+
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
+	return nil
 }

packet/packet.go

@@ -1,45 +0,0 @@
-package packet
-
-import (
-	"net/netip"
-	"sync"
-
-	"golang.org/x/sys/unix"
-)
-
-const Size = 0xffff
-
-type Packet struct {
-	Payload []byte
-	Control []byte
-	SegSize int
-	Addr    netip.AddrPort
-}
-
-func New() *Packet {
-	return &Packet{
-		Payload: make([]byte, Size),
-		Control: make([]byte, unix.CmsgSpace(2)),
-	}
-}
-
-type Pool struct {
-	pool sync.Pool
-}
-
-var bigPool = &Pool{
-	pool: sync.Pool{New: func() any { return New() }},
-}
-
-func GetPool() *Pool {
-	return bigPool
-}
-
-func (p *Pool) Get() *Packet {
-	return p.pool.Get().(*Packet)
-}
-
-func (p *Pool) Put(x *Packet) {
-	x.Payload = x.Payload[:Size]
-	p.pool.Put(x)
-}

relay_manager.go

@@ -155,6 +155,8 @@ func (rm *relayManager) handleCreateRelayResponse(v cert.Version, h *HostInfo, f
 		"vpnAddrs":            h.vpnAddrs}).
 		Info("handleCreateRelayResponse")
 
+	//peer == relayFrom
+	//target == relayTo
 	target := m.RelayToAddr
 	targetAddr := protoAddrToNetAddr(target)
 
@@ -190,11 +192,12 @@ func (rm *relayManager) handleCreateRelayResponse(v cert.Version, h *HostInfo, f
 			InitiatorRelayIndex: peerRelay.RemoteIndex,
 		}
 
+		relayFrom := h.vpnAddrs[0]
 		if v == cert.Version1 {
 			peer := peerHostInfo.vpnAddrs[0]
 			if !peer.Is4() {
 				rm.l.WithField("relayFrom", peer).
-					WithField("relayTo", target).
+					WithField("relayTo", targetAddr).
 					WithField("initiatorRelayIndex", resp.InitiatorRelayIndex).
 					WithField("responderRelayIndex", resp.ResponderRelayIndex).
 					WithField("vpnAddrs", peerHostInfo.vpnAddrs).
@@ -207,7 +210,22 @@ func (rm *relayManager) handleCreateRelayResponse(v cert.Version, h *HostInfo, f
 			b = targetAddr.As4()
 			resp.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
 		} else {
-			resp.RelayFromAddr = netAddrToProtoAddr(peerHostInfo.vpnAddrs[0])
+			ok = false
+			peerNetworks := h.GetCert().Certificate.Networks()
+			for i := range peerNetworks {
+				if peerNetworks[i].Contains(targetAddr) {
+					relayFrom = peerNetworks[i].Addr()
+					ok = true
+					break
+				}
+			}
+			if !ok {
+				rm.l.WithFields(logrus.Fields{"from": f.myVpnNetworks, "to": targetAddr}).
+					Error("cannot establish relay, no networks in common")
+				return
+			}
+
+			resp.RelayFromAddr = netAddrToProtoAddr(relayFrom)
 			resp.RelayToAddr = target
 		}
 
@@ -218,8 +236,8 @@ func (rm *relayManager) handleCreateRelayResponse(v cert.Version, h *HostInfo, f
 		} else {
 			f.SendMessageToHostInfo(header.Control, 0, peerHostInfo, msg, make([]byte, 12), make([]byte, mtu))
 			rm.l.WithFields(logrus.Fields{
-				"relayFrom":           resp.RelayFromAddr,
-				"relayTo":             resp.RelayToAddr,
+				"relayFrom":           relayFrom,
+				"relayTo":             targetAddr,
 				"initiatorRelayIndex": resp.InitiatorRelayIndex,
 				"responderRelayIndex": resp.ResponderRelayIndex,
 				"vpnAddrs":            peerHostInfo.vpnAddrs}).
@@ -313,8 +331,7 @@ func (rm *relayManager) handleCreateRelayRequest(v cert.Version, h *HostInfo, f
 
 		msg, err := resp.Marshal()
 		if err != nil {
-			logMsg.
-				WithError(err).Error("relayManager Failed to marshal Control CreateRelayResponse message to create relay")
+			logMsg.WithError(err).Error("relayManager Failed to marshal Control CreateRelayResponse message to create relay")
 		} else {
 			f.SendMessageToHostInfo(header.Control, 0, h, msg, make([]byte, 12), make([]byte, mtu))
 			rm.l.WithFields(logrus.Fields{
@@ -360,10 +377,10 @@ func (rm *relayManager) handleCreateRelayRequest(v cert.Version, h *HostInfo, f
 			Type:                NebulaControl_CreateRelayRequest,
 			InitiatorRelayIndex: index,
 		}
-
+		relayFrom := h.vpnAddrs[0]
 		if v == cert.Version1 {
-			if !h.vpnAddrs[0].Is4() {
-				rm.l.WithField("relayFrom", h.vpnAddrs[0]).
+			if !relayFrom.Is4() {
+				rm.l.WithField("relayFrom", relayFrom).
 					WithField("relayTo", target).
 					WithField("initiatorRelayIndex", req.InitiatorRelayIndex).
 					WithField("responderRelayIndex", req.ResponderRelayIndex).
@@ -372,23 +389,37 @@ func (rm *relayManager) handleCreateRelayRequest(v cert.Version, h *HostInfo, f
 				return
 			}
 
-			b := h.vpnAddrs[0].As4()
+			b := relayFrom.As4()
 			req.OldRelayFromAddr = binary.BigEndian.Uint32(b[:])
 			b = target.As4()
 			req.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
 		} else {
-			req.RelayFromAddr = netAddrToProtoAddr(h.vpnAddrs[0])
+			ok = false
+			peerNetworks := h.GetCert().Certificate.Networks()
+			for i := range peerNetworks {
+				if peerNetworks[i].Contains(target) {
+					relayFrom = peerNetworks[i].Addr()
+					ok = true
+					break
+				}
+			}
+			if !ok {
+				rm.l.WithFields(logrus.Fields{"from": f.myVpnNetworks, "to": target}).
+					Error("cannot establish relay, no networks in common")
+				return
+			}
+
+			req.RelayFromAddr = netAddrToProtoAddr(relayFrom)
 			req.RelayToAddr = netAddrToProtoAddr(target)
 		}
 
 		msg, err := req.Marshal()
 		if err != nil {
-			logMsg.
-				WithError(err).Error("relayManager Failed to marshal Control message to create relay")
+			logMsg.WithError(err).Error("relayManager Failed to marshal Control message to create relay")
 		} else {
 			f.SendMessageToHostInfo(header.Control, 0, peer, msg, make([]byte, 12), make([]byte, mtu))
 			rm.l.WithFields(logrus.Fields{
-				"relayFrom":           h.vpnAddrs[0],
+				"relayFrom":           relayFrom,
 				"relayTo":             target,
 				"initiatorRelayIndex": req.InitiatorRelayIndex,
 				"responderRelayIndex": req.ResponderRelayIndex,
@@ -401,8 +432,7 @@ func (rm *relayManager) handleCreateRelayRequest(v cert.Version, h *HostInfo, f
 		if !ok {
 			_, err := AddRelay(rm.l, h, f.hostMap, target, &m.InitiatorRelayIndex, ForwardingType, PeerRequested)
 			if err != nil {
-				logMsg.
-					WithError(err).Error("relayManager Failed to allocate a local index for relay")
+				logMsg.WithError(err).Error("relayManager Failed to allocate a local index for relay")
 				return
 			}
 		}

service/service.go

@@ -9,13 +9,10 @@ import (
 	"math"
 	"net"
 	"net/netip"
-	"os"
 	"strings"
 	"sync"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
-	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/overlay"
 	"golang.org/x/sync/errgroup"
 	"gvisor.dev/gvisor/pkg/buffer"
@@ -46,19 +43,8 @@ type Service struct {
 	}
 }
 
-func New(config *config.C) (*Service, error) {
-	logger := logrus.New()
-	logger.Out = os.Stdout
-
-	control, err := nebula.Main(config, false, "custom-app", logger, overlay.NewUserDeviceFromConfig)
-	if err != nil {
-		return nil, err
-	}
-
-	wait, err := control.Start()
-	if err != nil {
-		return nil, err
-	}
+func New(control *nebula.Control) (*Service, error) {
+	control.Start()
 
 	ctx := control.Context()
 	eg, ctx := errgroup.WithContext(ctx)
@@ -155,12 +141,6 @@ func New(config *config.C) (*Service, error) {
 		}
 	})
 
-	// Add the nebula wait function to the group
-	eg.Go(func() error {
-		wait()
-		return nil
-	})
-
 	return &s, nil
 }
 

udp/conn.go

@@ -4,19 +4,19 @@ import (
 	"net/netip"
 
 	"github.com/slackhq/nebula/config"
-	"github.com/slackhq/nebula/packet"
 )
 
 const MTU = 9001
 
-type EncReader func(*packet.Packet)
-
-type PacketBufferGetter func() *packet.Packet
+type EncReader func(
+	addr netip.AddrPort,
+	payload []byte,
+)
 
 type Conn interface {
 	Rebind() error
 	LocalAddr() (netip.AddrPort, error)
-	ListenOut(pg PacketBufferGetter, pc chan *packet.Packet) error
+	ListenOut(r EncReader)
 	WriteTo(b []byte, addr netip.AddrPort) error
 	ReloadConfig(c *config.C)
 	Close() error

udp/udp_generic.go

@@ -71,14 +71,15 @@ type rawMessage struct {
 	Len uint32
 }
 
-func (u *GenericConn) ListenOut(r EncReader) error {
+func (u *GenericConn) ListenOut(r EncReader) {
 	buffer := make([]byte, MTU)
 
 	for {
 		// Just read one packet at a time
 		n, rua, err := u.ReadFromUDPAddrPort(buffer)
 		if err != nil {
-			return err
+			u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
+			return
 		}
 
 		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n])

udp/udp_linux.go

@@ -9,26 +9,27 @@ import (
 	"net"
 	"net/netip"
 	"syscall"
-	"time"
 	"unsafe"
 
 	"github.com/rcrowley/go-metrics"
 	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/config"
-	"github.com/slackhq/nebula/packet"
 	"golang.org/x/sys/unix"
 )
 
-var readTimeout = unix.NsecToTimeval(int64(time.Millisecond * 500))
-
 type StdConn struct {
-	sysFd     int
-	isV4      bool
-	l         *logrus.Logger
-	batch     int
-	enableGRO bool
-	enableGSO bool
-	//gso       gsoState
+	sysFd int
+	isV4  bool
+	l     *logrus.Logger
+	batch int
+}
+
+func maybeIPV4(ip net.IP) (net.IP, bool) {
+	ip4 := ip.To4()
+	if ip4 != nil {
+		return ip4, true
+	}
+	return ip, false
 }
 
 func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int) (Conn, error) {
@@ -54,11 +55,6 @@ func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch in
 		}
 	}
 
-	// Set a read timeout
-	if err = unix.SetsockoptTimeval(fd, unix.SOL_SOCKET, unix.SO_RCVTIMEO, &readTimeout); err != nil {
-		return nil, fmt.Errorf("unable to set SO_RCVTIMEO: %s", err)
-	}
-
 	var sa unix.Sockaddr
 	if ip.Is4() {
 		sa4 := &unix.SockaddrInet4{Port: port}
@@ -122,10 +118,10 @@ func (u *StdConn) LocalAddr() (netip.AddrPort, error) {
 	}
 }
 
-func (u *StdConn) ListenOut(pg PacketBufferGetter, pc chan *packet.Packet) error {
+func (u *StdConn) ListenOut(r EncReader) {
 	var ip netip.Addr
 
-	msgs, packets, names := u.PrepareRawMessages(u.batch, pg)
+	msgs, buffers, names := u.PrepareRawMessages(u.batch)
 	read := u.ReadMulti
 	if u.batch == 1 {
 		read = u.ReadSingle
@@ -134,61 +130,22 @@ func (u *StdConn) ListenOut(pg PacketBufferGetter, pc chan *packet.Packet) error
 	for {
 		n, err := read(msgs)
 		if err != nil {
-			return err
+			u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
+			return
 		}
 
 		for i := 0; i < n; i++ {
-			out := packets[i]
-			out.Payload = out.Payload[:msgs[i].Len]
-
 			// Its ok to skip the ok check here, the slicing is the only error that can occur and it will panic
 			if u.isV4 {
 				ip, _ = netip.AddrFromSlice(names[i][4:8])
 			} else {
 				ip, _ = netip.AddrFromSlice(names[i][8:24])
 			}
-			out.Addr = netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4]))
-			ctrlLen := getRawMessageControlLen(&msgs[i])
-			if ctrlLen > 0 {
-				packets[i].SegSize = parseGROControl(packets[i].Control[:ctrlLen])
-			} else {
-				packets[i].SegSize = 0
-			}
-
-			pc <- out
-
-			//rotate this packet out so we don't overwrite it
-			packets[i] = pg()
-			msgs[i].Hdr.Iov.Base = &packets[i].Payload[0]
-			if u.enableGRO {
-				msgs[i].Hdr.Control = &packets[i].Control[0]
-				msgs[i].Hdr.Controllen = uint64(cap(packets[i].Control))
-			}
-
+			r(netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4])), buffers[i][:msgs[i].Len])
 		}
 	}
 }
 
-func parseGROControl(control []byte) int {
-	if len(control) == 0 {
-		return 0
-	}
-
-	cmsgs, err := unix.ParseSocketControlMessage(control)
-	if err != nil {
-		return 0
-	}
-
-	for _, c := range cmsgs {
-		if c.Header.Level == unix.SOL_UDP && c.Header.Type == unix.UDP_GRO && len(c.Data) >= 2 {
-			segSize := int(binary.LittleEndian.Uint16(c.Data[:2]))
-			return segSize
-		}
-	}
-
-	return 0
-}
-
 func (u *StdConn) ReadSingle(msgs []rawMessage) (int, error) {
 	for {
 		n, _, err := unix.Syscall6(
@@ -202,9 +159,6 @@ func (u *StdConn) ReadSingle(msgs []rawMessage) (int, error) {
 		)
 
 		if err != 0 {
-			if err == unix.EAGAIN || err == unix.EINTR {
-				continue
-			}
 			return 0, &net.OpError{Op: "recvmsg", Err: err}
 		}
 
@@ -226,9 +180,6 @@ func (u *StdConn) ReadMulti(msgs []rawMessage) (int, error) {
 		)
 
 		if err != 0 {
-			if err == unix.EAGAIN || err == unix.EINTR {
-				continue
-			}
 			return 0, &net.OpError{Op: "recvmmsg", Err: err}
 		}
 
@@ -270,7 +221,7 @@ func (u *StdConn) writeTo6(b []byte, ip netip.AddrPort) error {
 
 func (u *StdConn) writeTo4(b []byte, ip netip.AddrPort) error {
 	if !ip.Addr().Is4() {
-		return fmt.Errorf("Listener is IPv4, but writing to IPv6 remote")
+		return ErrInvalidIPv6RemoteForSocket
 	}
 
 	var rsa unix.RawSockaddrInet4
@@ -343,28 +294,6 @@ func (u *StdConn) ReloadConfig(c *config.C) {
 			u.l.WithError(err).Error("Failed to set listen.so_mark")
 		}
 	}
-	u.configureGRO(true)
-}
-
-func (u *StdConn) configureGRO(enable bool) {
-	if enable == u.enableGRO {
-		return
-	}
-
-	if enable {
-		if err := unix.SetsockoptInt(u.sysFd, unix.SOL_UDP, unix.UDP_GRO, 1); err != nil {
-			u.l.WithError(err).Warn("Failed to enable UDP GRO")
-			return
-		}
-		u.enableGRO = true
-		u.l.Info("UDP GRO enabled")
-		return
-	}
-
-	if err := unix.SetsockoptInt(u.sysFd, unix.SOL_UDP, unix.UDP_GRO, 0); err != nil && err != unix.ENOPROTOOPT {
-		u.l.WithError(err).Warn("Failed to disable UDP GRO")
-	}
-	u.enableGRO = false
 }
 
 func (u *StdConn) getMemInfo(meminfo *[unix.SK_MEMINFO_VARS]uint32) error {

udp/udp_linux_64.go

@@ -7,7 +7,6 @@
 package udp
 
 import (
-	"github.com/slackhq/nebula/packet"
 	"golang.org/x/sys/unix"
 )
 
@@ -34,39 +33,17 @@ type rawMessage struct {
 	Pad0 [4]byte
 }
 
-func setRawMessageControl(msg *rawMessage, buf []byte) {
-	if len(buf) == 0 {
-		msg.Hdr.Control = nil
-		msg.Hdr.Controllen = 0
-		return
-	}
-	msg.Hdr.Control = &buf[0]
-	msg.Hdr.Controllen = uint64(len(buf))
-}
-
-func getRawMessageControlLen(msg *rawMessage) int {
-	return int(msg.Hdr.Controllen)
-}
-
-func setCmsgLen(h *unix.Cmsghdr, l int) {
-	h.Len = uint64(l)
-}
-
-func (u *StdConn) PrepareRawMessages(n int, pg PacketBufferGetter) ([]rawMessage, []*packet.Packet, [][]byte) {
+func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 	msgs := make([]rawMessage, n)
+	buffers := make([][]byte, n)
 	names := make([][]byte, n)
 
-	packets := make([]*packet.Packet, n)
-	for i := range packets {
-		packets[i] = pg()
-	}
-	//todo?
-
 	for i := range msgs {
+		buffers[i] = make([]byte, MTU)
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
-			{Base: &packets[i].Payload[0], Len: uint64(packet.Size)},
+			{Base: &buffers[i][0], Len: uint64(len(buffers[i]))},
 		}
 
 		msgs[i].Hdr.Iov = &vs[0]
@@ -74,14 +51,7 @@ func (u *StdConn) PrepareRawMessages(n int, pg PacketBufferGetter) ([]rawMessage
 
 		msgs[i].Hdr.Name = &names[i][0]
 		msgs[i].Hdr.Namelen = uint32(len(names[i]))
-		if u.enableGRO {
-			msgs[i].Hdr.Control = &packets[i].Control[0]
-			msgs[i].Hdr.Controllen = uint64(len(packets[i].Control))
-		} else {
-			msgs[i].Hdr.Control = nil
-			msgs[i].Hdr.Controllen = 0
-		}
 	}
 
-	return msgs, packets, names
+	return msgs, buffers, names
 }

udp/udp_rio_windows.go

@@ -134,7 +134,7 @@ func (u *RIOConn) bind(sa windows.Sockaddr) error {
 	return nil
 }
 
-func (u *RIOConn) ListenOut(r EncReader) error {
+func (u *RIOConn) ListenOut(r EncReader) {
 	buffer := make([]byte, MTU)
 
 	for {