claude first stab and first cleanup - ugh

go.mod

@@ -29,11 +29,11 @@ require (
 	golang.org/x/sys v0.37.0
 	golang.org/x/term v0.36.0
 	golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2
-	golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b
+	golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb
 	golang.zx2c4.com/wireguard/windows v0.5.3
 	google.golang.org/protobuf v1.36.8
 	gopkg.in/yaml.v3 v3.0.1
-	gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe
+	gvisor.dev/gvisor v0.0.0-20250503011706-39ed1f5ac29c
 )
 
 require (
@@ -49,6 +49,6 @@ require (
 	github.com/vishvananda/netns v0.0.5 // indirect
 	go.yaml.in/yaml/v2 v2.4.2 // indirect
 	golang.org/x/mod v0.24.0 // indirect
-	golang.org/x/time v0.5.0 // indirect
+	golang.org/x/time v0.7.0 // indirect
 	golang.org/x/tools v0.33.0 // indirect
 )

go.sum

@@ -215,8 +215,8 @@ golang.org/x/term v0.36.0/go.mod h1:Qu394IJq6V6dCBRgwqshf3mPF85AqzYEzofzRdZkWss=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/time v0.5.0 h1:o7cqy6amK/52YcAKIPlM3a+Fpj35zvRj2TP+e1xFSfk=
-golang.org/x/time v0.5.0/go.mod h1:3BpzKBy/shNhVucY/MWOyx10tF3SFh9QdLuxbVysPQM=
+golang.org/x/time v0.7.0 h1:ntUhktv3OPE6TgYxXWv9vKvUSJyIFJlyohwbkEwPrKQ=
+golang.org/x/time v0.7.0/go.mod h1:3BpzKBy/shNhVucY/MWOyx10tF3SFh9QdLuxbVysPQM=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/tools v0.0.0-20200130002326-2f3ba24bd6e7/go.mod h1:TB2adYChydJhpapKDTa4BR/hXlZSLoq2Wpct/0txZ28=
@@ -230,8 +230,8 @@ golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8T
 golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2 h1:B82qJJgjvYKsXS9jeunTOisW56dUokqW/FOteYJJ/yg=
 golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2/go.mod h1:deeaetjYA+DHMHg+sMSMI58GrEteJUUzzw7en6TJQcI=
-golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b h1:J1CaxgLerRR5lgx3wnr6L04cJFbWoceSK9JWBdglINo=
-golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b/go.mod h1:tqur9LnfstdR9ep2LaJT4lFUl0EjlHtge+gAjmsHUG4=
+golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb h1:whnFRlWMcXI9d+ZbWg+4sHnLp52d5yiIPUxMBSt4X9A=
+golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb/go.mod h1:rpwXGsirqLqN2L0JDJQlwOboGHmptD5ZD6T2VmcqhTw=
 golang.zx2c4.com/wireguard/windows v0.5.3 h1:On6j2Rpn3OEMXqBq00QEDC7bWSZrPIHKIus8eIuExIE=
 golang.zx2c4.com/wireguard/windows v0.5.3/go.mod h1:9TEe8TJmtwyQebdFwAkEWOPr3prrtqm+REGFifP60hI=
 google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
@@ -257,5 +257,5 @@ gopkg.in/yaml.v2 v2.3.0/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
-gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe h1:fre4i6mv4iBuz5lCMOzHD1rH1ljqHWSICFmZRbbgp3g=
-gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe/go.mod h1:sxc3Uvk/vHcd3tj7/DHVBoR5wvWT/MmRq2pj7HRJnwU=
+gvisor.dev/gvisor v0.0.0-20250503011706-39ed1f5ac29c h1:m/r7OM+Y2Ty1sgBQ7Qb27VgIMBW8ZZhT4gLnUyDIhzI=
+gvisor.dev/gvisor v0.0.0-20250503011706-39ed1f5ac29c/go.mod h1:3r5CMtNQMKIvBlrmM9xWUNamjKBYPOWyXOjmg5Kts3g=

inside.go

@@ -33,8 +33,7 @@ func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet
 		// routes packets from the Nebula addr to the Nebula addr through the Nebula
 		// TUN device.
 		if immediatelyForwardToSelf {
-			_, err := f.readers[q].Write(packet)
-			if err != nil {
+			if err := f.writeTun(q, packet); err != nil {
 				f.l.WithError(err).Error("Failed to forward to tun")
 			}
 		}
@@ -91,8 +90,7 @@ func (f *Interface) rejectInside(packet []byte, out []byte, q int) {
 		return
 	}
 
-	_, err := f.readers[q].Write(out)
-	if err != nil {
+	if err := f.writeTun(q, out); err != nil {
 		f.l.WithError(err).Error("Failed to write to tun")
 	}
 }

interface.go

@@ -110,6 +110,16 @@ type EncWriter interface {
 	GetCertState() *CertState
 }
 
+// BatchReader is an interface for readers that support vectorized packet reading
+type BatchReader interface {
+	BatchRead() ([][]byte, []int, error)
+}
+
+// BatchWriter is an interface for writers that support vectorized packet writing
+type BatchWriter interface {
+	BatchWrite([][]byte) (int, error)
+}
+
 type sendRecvErrorConfig uint8
 
 const (
@@ -279,6 +289,16 @@ func (f *Interface) listenOut(i int) {
 func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	runtime.LockOSThread()
 
+	// Check if reader supports batch operations
+	if batchReader, ok := reader.(BatchReader); ok {
+		err := f.listenInBatch(batchReader, i)
+		if err != nil {
+			f.l.WithError(err).Error("Fatal error in batch packet reader, exiting goroutine")
+		}
+		return
+	}
+
+	// Fall back to single-packet mode
 	packet := make([]byte, mtu)
 	out := make([]byte, mtu)
 	fwPacket := &firewall.Packet{}
@@ -293,15 +313,78 @@ func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 				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)
+			f.l.WithError(err).Error("Fatal error while reading outbound packet, exiting goroutine")
+			return
 		}
 
 		f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
 	}
 }
 
+// listenInBatch handles vectorized packet reading for improved performance
+func (f *Interface) listenInBatch(reader BatchReader, i int) error {
+	// Allocate per-packet state
+	fwPackets := make([]*firewall.Packet, 64) // Match batch size
+	outBuffers := make([][]byte, 64)
+	nbBuffers := make([][]byte, 64)
+
+	for j := 0; j < 64; j++ {
+		fwPackets[j] = &firewall.Packet{}
+		outBuffers[j] = make([]byte, mtu)
+		nbBuffers[j] = make([]byte, 12, 12)
+	}
+
+	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
+
+	for {
+		packets, sizes, err := reader.BatchRead()
+		if err != nil {
+			if errors.Is(err, os.ErrClosed) && f.closed.Load() {
+				return nil
+			}
+
+			return fmt.Errorf("error while batch reading outbound packets: %w", err)
+		}
+
+		// Process each packet in the batch
+		cache := conntrackCache.Get(f.l)
+		for idx := 0; idx < len(packets); idx++ {
+			if idx < len(sizes) && sizes[idx] > 0 {
+				// Use modulo to reuse fw packet state if batch is larger than our pre-allocated state
+				stateIdx := idx % len(fwPackets)
+				f.consumeInsidePacket(packets[idx][:sizes[idx]], fwPackets[stateIdx], nbBuffers[stateIdx], outBuffers[stateIdx], i, cache)
+			}
+		}
+	}
+}
+
+// writeTunBatch attempts to write multiple packets to the TUN device using batch operations if supported
+func (f *Interface) writeTunBatch(q int, packets [][]byte) error {
+	if len(packets) == 0 {
+		return nil
+	}
+
+	// Check if the reader/writer supports batch operations
+	if batchWriter, ok := f.readers[q].(BatchWriter); ok {
+		_, err := batchWriter.BatchWrite(packets)
+		return err
+	}
+
+	// Fall back to writing packets individually
+	for _, packet := range packets {
+		if _, err := f.readers[q].Write(packet); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// writeTun writes a single packet to the TUN device
+func (f *Interface) writeTun(q int, packet []byte) error {
+	_, err := f.readers[q].Write(packet)
+	return err
+}
+
 func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
 	c.RegisterReloadCallback(f.reloadFirewall)
 	c.RegisterReloadCallback(f.reloadSendRecvError)

overlay/route.go

@@ -3,7 +3,6 @@ package overlay
 import (
 	"fmt"
 	"math"
-	"net"
 	"net/netip"
 	"runtime"
 	"strconv"
@@ -305,29 +304,3 @@ func parseUnsafeRoutes(c *config.C, networks []netip.Prefix) ([]Route, error) {
 
 	return routes, nil
 }
-
-func ipWithin(o *net.IPNet, i *net.IPNet) bool {
-	// Make sure o contains the lowest form of i
-	if !o.Contains(i.IP.Mask(i.Mask)) {
-		return false
-	}
-
-	// Find the max ip in i
-	ip4 := i.IP.To4()
-	if ip4 == nil {
-		return false
-	}
-
-	last := make(net.IP, len(ip4))
-	copy(last, ip4)
-	for x := range ip4 {
-		last[x] |= ^i.Mask[x]
-	}
-
-	// Make sure o contains the max
-	if !o.Contains(last) {
-		return false
-	}
-
-	return true
-}

overlay/route_test.go

@@ -225,6 +225,7 @@ func Test_parseUnsafeRoutes(t *testing.T) {
 	// no mtu
 	c.Settings["tun"] = map[string]any{"unsafe_routes": []any{map[string]any{"via": "127.0.0.1", "route": "1.0.0.0/8"}}}
 	routes, err = parseUnsafeRoutes(c, []netip.Prefix{n})
+	require.NoError(t, err)
 	assert.Len(t, routes, 1)
 	assert.Equal(t, 0, routes[0].MTU)
 
@@ -318,7 +319,7 @@ func Test_makeRouteTree(t *testing.T) {
 
 	ip, err = netip.ParseAddr("1.1.0.1")
 	require.NoError(t, err)
-	r, ok = routeTree.Lookup(ip)
+	_, ok = routeTree.Lookup(ip)
 	assert.False(t, ok)
 }
 

overlay/tun.go

@@ -1,8 +1,6 @@
 package overlay
 
 import (
-	"fmt"
-	"net"
 	"net/netip"
 
 	"github.com/sirupsen/logrus"
@@ -72,51 +70,3 @@ func findRemovedRoutes(newRoutes, oldRoutes []Route) []Route {
 
 	return removed
 }
-
-func prefixToMask(prefix netip.Prefix) netip.Addr {
-	pLen := 128
-	if prefix.Addr().Is4() {
-		pLen = 32
-	}
-
-	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

@@ -1,5 +1,5 @@
-//go:build !ios && !e2e_testing
-// +build !ios,!e2e_testing
+//go:build darwin && !ios && !e2e_testing
+// +build darwin,!ios,!e2e_testing
 
 package overlay
 
@@ -8,48 +8,27 @@ import (
 	"fmt"
 	"io"
 	"net/netip"
-	"os"
-	"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"
+	wgtun "golang.zx2c4.com/wireguard/tun"
 )
 
 type tun struct {
-	io.ReadWriteCloser
-	Device      string
-	vpnNetworks []netip.Prefix
-	DefaultMTU  int
-	Routes      atomic.Pointer[[]Route]
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
-	linkAddr    *netroute.LinkAddr
-	l           *logrus.Logger
-
-	// cache out buffer since we need to prepend 4 bytes for tun metadata
-	out []byte
+	linkAddr *netroute.LinkAddr
 }
 
+// ioctl structures for Darwin network configuration
 type ifReq struct {
 	Name  [unix.IFNAMSIZ]byte
 	Flags uint16
 	pad   [8]byte
 }
 
-const (
-	_SIOCAIFADDR_IN6 = 2155899162
-	_UTUN_OPT_IFNAME = 2
-	_IN6_IFF_NODAD   = 0x0020
-	_IN6_IFF_SECURED = 0x0400
-	utunControlName  = "com.apple.net.utun_control"
-)
-
 type ifreqMTU struct {
 	Name [16]byte
 	MTU  int32
@@ -79,60 +58,46 @@ type ifreqAlias6 struct {
 	Lifetime   addrLifetime
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
-	name := c.GetString("tun.dev", "")
-	ifIndex := -1
-	if name != "" && name != "utun" {
-		_, err := fmt.Sscanf(name, "utun%d", &ifIndex)
-		if err != nil || ifIndex < 0 {
-			// NOTE: we don't make this error so we don't break existing
-			// configs that set a name before it was used.
-			l.Warn("interface name must be utun[0-9]+ on Darwin, ignoring")
-			ifIndex = -1
-		}
-	}
-
-	fd, err := unix.Socket(unix.AF_SYSTEM, unix.SOCK_DGRAM, unix.AF_SYS_CONTROL)
-	if err != nil {
-		return nil, fmt.Errorf("system socket: %v", err)
-	}
+const (
+	_SIOCAIFADDR_IN6 = 2155899162
+	_IN6_IFF_NODAD   = 0x0020
+)
 
-	var ctlInfo = &unix.CtlInfo{}
-	copy(ctlInfo.Name[:], utunControlName)
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
+	return nil, fmt.Errorf("newTunFromFd not supported on Darwin")
+}
 
-	err = unix.IoctlCtlInfo(fd, ctlInfo)
-	if err != nil {
-		return nil, fmt.Errorf("CTLIOCGINFO: %v", err)
-	}
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+	deviceName := c.GetString("tun.dev", "utun")
+	mtu := c.GetInt("tun.mtu", DefaultMTU)
 
-	err = unix.Connect(fd, &unix.SockaddrCtl{
-		ID:   ctlInfo.Id,
-		Unit: uint32(ifIndex) + 1,
-	})
+	// Create WireGuard TUN device
+	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, fmt.Errorf("SYS_CONNECT: %v", err)
+		return nil, fmt.Errorf("failed to create TUN device: %w", err)
 	}
 
-	name, err = unix.GetsockoptString(fd, unix.AF_SYS_CONTROL, _UTUN_OPT_IFNAME)
+	// Get the actual device name
+	actualName, err := tunDevice.Name()
 	if err != nil {
-		return nil, fmt.Errorf("failed to retrieve tun name: %w", err)
+		tunDevice.Close()
+		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
-	err = unix.SetNonblock(fd, true)
-	if err != nil {
-		return nil, fmt.Errorf("SetNonblock: %v", err)
+	t := &wgTun{
+		tunDevice:   tunDevice,
+		vpnNetworks: vpnNetworks,
+		MaxMTU:      mtu,
+		DefaultMTU:  mtu,
+		l:           l,
 	}
 
-	t := &tun{
-		ReadWriteCloser: os.NewFile(uintptr(fd), ""),
-		Device:          name,
-		vpnNetworks:     vpnNetworks,
-		DefaultMTU:      c.GetInt("tun.mtu", DefaultMTU),
-		l:               l,
-	}
+	// Create Darwin-specific route manager
+	t.routeManager = &tun{}
 
 	err = t.reload(c, true)
 	if err != nil {
+		tunDevice.Close()
 		return nil, err
 	}
 
@@ -143,215 +108,252 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 		}
 	})
 
+	l.WithField("name", actualName).Info("Created WireGuard TUN device")
+
 	return t, nil
 }
 
-func (t *tun) deviceBytes() (o [16]byte) {
-	for i, c := range t.Device {
-		o[i] = byte(c)
+func (rm *tun) Activate(t *wgTun) error {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
-	return
-}
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
-	return nil, fmt.Errorf("newTunFromFd not supported in Darwin")
-}
+	// Set the MTU
+	rm.SetMTU(t, t.MaxMTU)
 
-func (t *tun) Close() error {
-	if t.ReadWriteCloser != nil {
-		return t.ReadWriteCloser.Close()
+	// Add IP addresses
+	for _, network := range t.vpnNetworks {
+		if err := rm.addIP(t, name, network); err != nil {
+			return err
+		}
 	}
-	return nil
-}
 
-func (t *tun) Activate() error {
-	devName := t.deviceBytes()
+	// Bring up the interface using ioctl
+	if err := rm.bringUpInterface(name); err != nil {
+		return fmt.Errorf("failed to bring up interface: %w", err)
+	}
 
-	s, err := unix.Socket(
-		unix.AF_INET,
-		unix.SOCK_DGRAM,
-		unix.IPPROTO_IP,
-	)
+	// Get the link address for routing
+	linkAddr, err := getLinkAddr(name)
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to get link address: %w", err)
 	}
-	defer unix.Close(s)
-
-	fd := uintptr(s)
-
-	// Set the MTU on the device
-	ifm := ifreqMTU{Name: devName, MTU: int32(t.DefaultMTU)}
-	if err = ioctl(fd, unix.SIOCSIFMTU, uintptr(unsafe.Pointer(&ifm))); err != nil {
-		return fmt.Errorf("failed to set tun mtu: %v", err)
+	if linkAddr == nil {
+		return fmt.Errorf("unable to discover link_addr for tun interface")
 	}
+	rm.linkAddr = linkAddr
 
-	// Get the device flags
-	ifrf := ifReq{Name: devName}
-	if err = ioctl(fd, unix.SIOCGIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
-		return fmt.Errorf("failed to get tun flags: %s", err)
+	// Set the routes
+	if err := rm.AddRoutes(t, false); err != nil {
+		return err
 	}
 
-	linkAddr, err := getLinkAddr(t.Device)
+	return nil
+}
+
+func (rm *tun) bringUpInterface(name string) error {
+	// Open a socket for ioctl
+	fd, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, 0)
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to create socket: %w", err)
 	}
-	if linkAddr == nil {
-		return fmt.Errorf("unable to discover link_addr for tun interface")
-	}
-	t.linkAddr = linkAddr
+	defer unix.Close(fd)
 
-	for _, network := range t.vpnNetworks {
-		if network.Addr().Is4() {
-			err = t.activate4(network)
-			if err != nil {
-				return err
-			}
-		} else {
-			err = t.activate6(network)
-			if err != nil {
-				return err
-			}
-		}
+	// Get current flags
+	var ifrf ifReq
+	copy(ifrf.Name[:], name)
+
+	if err := ioctl(uintptr(fd), unix.SIOCGIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
+		return fmt.Errorf("failed to get interface flags: %w", err)
 	}
 
-	// Run the interface
+	// Set IFF_UP and IFF_RUNNING flags
 	ifrf.Flags = ifrf.Flags | unix.IFF_UP | unix.IFF_RUNNING
-	if err = ioctl(fd, unix.SIOCSIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
-		return fmt.Errorf("failed to run tun device: %s", err)
+
+	if err := ioctl(uintptr(fd), unix.SIOCSIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
+		return fmt.Errorf("failed to set interface flags: %w", err)
 	}
 
-	// Unsafe path routes
-	return t.addRoutes(false)
+	return nil
 }
 
-func (t *tun) activate4(network netip.Prefix) error {
-	s, err := unix.Socket(
-		unix.AF_INET,
-		unix.SOCK_DGRAM,
-		unix.IPPROTO_IP,
-	)
+func (rm *tun) SetMTU(t *wgTun, mtu int) {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return err
+		t.l.WithError(err).Error("Failed to get device name for MTU set")
+		return
 	}
-	defer unix.Close(s)
 
-	ifr := ifreqAlias4{
-		Name: t.deviceBytes(),
-		Addr: unix.RawSockaddrInet4{
-			Len:    unix.SizeofSockaddrInet4,
-			Family: unix.AF_INET,
-			Addr:   network.Addr().As4(),
-		},
-		DstAddr: unix.RawSockaddrInet4{
-			Len:    unix.SizeofSockaddrInet4,
-			Family: unix.AF_INET,
-			Addr:   network.Addr().As4(),
-		},
-		MaskAddr: unix.RawSockaddrInet4{
-			Len:    unix.SizeofSockaddrInet4,
-			Family: unix.AF_INET,
-			Addr:   prefixToMask(network).As4(),
-		},
+	// Open a socket for ioctl
+	fd, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, 0)
+	if err != nil {
+		t.l.WithError(err).Error("Failed to create socket for MTU set")
+		return
 	}
+	defer unix.Close(fd)
 
-	if err := ioctl(uintptr(s), unix.SIOCAIFADDR, uintptr(unsafe.Pointer(&ifr))); err != nil {
-		return fmt.Errorf("failed to set tun v4 address: %s", err)
+	// Prepare the ioctl request
+	var ifr ifreqMTU
+	copy(ifr.Name[:], name)
+	ifr.MTU = int32(mtu)
+
+	// Set the MTU using ioctl
+	if err := ioctl(uintptr(fd), unix.SIOCSIFMTU, uintptr(unsafe.Pointer(&ifr))); err != nil {
+		t.l.WithError(err).Error("Failed to set tun mtu via ioctl")
 	}
+}
 
-	err = addRoute(network, t.linkAddr)
-	if err != nil {
-		return err
+func (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
+	// On Darwin, routes are set via ifconfig and route commands
+	return nil
+}
+
+func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
+	routes := *t.Routes.Load()
+	for _, r := range routes {
+		if !r.Install {
+			continue
+		}
+
+		err := rm.addRoute(r.Cidr)
+		if err != nil {
+			if errors.Is(err, unix.EEXIST) {
+				t.l.WithField("route", r.Cidr).
+					Warnf("unable to add unsafe_route, identical route already exists")
+			} else {
+				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
 }
 
-func (t *tun) activate6(network netip.Prefix) error {
-	s, err := unix.Socket(
-		unix.AF_INET6,
-		unix.SOCK_DGRAM,
-		unix.IPPROTO_IP,
-	)
+func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
+	for _, r := range routes {
+		if !r.Install {
+			continue
+		}
+
+		err := rm.delRoute(r.Cidr)
+		if err != nil {
+			t.l.WithError(err).WithField("route", r).Error("Failed to remove route")
+		} else {
+			t.l.WithField("route", r).Info("Removed route")
+		}
+	}
+}
+
+func (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+	// Darwin doesn't support multi-queue TUN devices in the same way as Linux
+	// Return a reader that wraps the same device
+	return &wgTunReader{
+		parent:    t,
+		tunDevice: t.tunDevice,
+		batchSize: 64,
+		offset:    0,
+		l:         t.l,
+	}, nil
+}
+
+func (rm *tun) addIP(t *wgTun, name string, network netip.Prefix) error {
+	addr := network.Addr()
+
+	if addr.Is4() {
+		return rm.addIPv4(name, network)
+	} else {
+		return rm.addIPv6(name, network)
+	}
+}
+
+func (rm *tun) addIPv4(name string, network netip.Prefix) error {
+	// Open an IPv4 socket for ioctl
+	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to create IPv4 socket: %w", err)
 	}
 	defer unix.Close(s)
 
-	ifr := ifreqAlias6{
-		Name: t.deviceBytes(),
-		Addr: unix.RawSockaddrInet6{
-			Len:    unix.SizeofSockaddrInet6,
-			Family: unix.AF_INET6,
-			Addr:   network.Addr().As16(),
-		},
-		PrefixMask: unix.RawSockaddrInet6{
-			Len:    unix.SizeofSockaddrInet6,
-			Family: unix.AF_INET6,
-			Addr:   prefixToMask(network).As16(),
-		},
-		Lifetime: addrLifetime{
-			// never expires
-			Vltime: 0xffffffff,
-			Pltime: 0xffffffff,
-		},
-		Flags: _IN6_IFF_NODAD,
+	var ifr ifreqAlias4
+	copy(ifr.Name[:], name)
+
+	// Set the address
+	ifr.Addr = unix.RawSockaddrInet4{
+		Len:    unix.SizeofSockaddrInet4,
+		Family: unix.AF_INET,
+		Addr:   network.Addr().As4(),
 	}
 
-	if err := ioctl(uintptr(s), _SIOCAIFADDR_IN6, uintptr(unsafe.Pointer(&ifr))); err != nil {
-		return fmt.Errorf("failed to set tun address: %s", err)
+	// Set the destination address (same as address for point-to-point)
+	ifr.DstAddr = unix.RawSockaddrInet4{
+		Len:    unix.SizeofSockaddrInet4,
+		Family: unix.AF_INET,
+		Addr:   network.Addr().As4(),
+	}
+
+	// Set the netmask
+	ifr.MaskAddr = unix.RawSockaddrInet4{
+		Len:    unix.SizeofSockaddrInet4,
+		Family: unix.AF_INET,
+		Addr:   prefixToMask(network).As4(),
+	}
+
+	if err := ioctl(uintptr(s), unix.SIOCAIFADDR, uintptr(unsafe.Pointer(&ifr))); err != nil {
+		return fmt.Errorf("failed to set IPv4 address via ioctl: %w", err)
 	}
 
 	return nil
 }
 
-func (t *tun) reload(c *config.C, initial bool) error {
-	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
+func (rm *tun) addIPv6(name string, network netip.Prefix) error {
+	// Open an IPv6 socket for ioctl
+	s, err := unix.Socket(unix.AF_INET6, unix.SOCK_DGRAM, unix.IPPROTO_IP)
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to create IPv6 socket: %w", err)
 	}
+	defer unix.Close(s)
+
+	var ifr ifreqAlias6
+	copy(ifr.Name[:], name)
 
-	if !initial && !change {
-		return nil
+	// Set the address
+	ifr.Addr = unix.RawSockaddrInet6{
+		Len:    unix.SizeofSockaddrInet6,
+		Family: unix.AF_INET6,
+		Addr:   network.Addr().As16(),
 	}
 
-	routeTree, err := makeRouteTree(t.l, routes, false)
-	if err != nil {
-		return err
+	// Set the prefix mask
+	ifr.PrefixMask = unix.RawSockaddrInet6{
+		Len:    unix.SizeofSockaddrInet6,
+		Family: unix.AF_INET6,
+		Addr:   prefixToMask(network).As16(),
 	}
 
-	// Teach nebula how to handle the routes before establishing them in the system table
-	oldRoutes := t.Routes.Swap(&routes)
-	t.routeTree.Store(routeTree)
+	// Set lifetime (never expires)
+	ifr.Lifetime = addrLifetime{
+		Vltime: 0xffffffff,
+		Pltime: 0xffffffff,
+	}
 
-	if !initial {
-		// Remove first, if the system removes a wanted route hopefully it will be re-added next
-		err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
-		if err != nil {
-			util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
-		}
+	// Set flags (no DAD - Duplicate Address Detection)
+	ifr.Flags = _IN6_IFF_NODAD
 
-		// Ensure any routes we actually want are installed
-		err = t.addRoutes(true)
-		if err != nil {
-			// Catch any stray logs
-			util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
-		}
+	if err := ioctl(uintptr(s), _SIOCAIFADDR_IN6, uintptr(unsafe.Pointer(&ifr))); err != nil {
+		return fmt.Errorf("failed to set IPv6 address via ioctl: %w", err)
 	}
 
 	return nil
 }
 
-func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
-	r, ok := t.routeTree.Load().Lookup(ip)
-	if ok {
-		return r
-	}
-	return routing.Gateways{}
-}
-
-// Get the LinkAddr for the interface of the given name
-// Is there an easier way to fetch this when we create the interface?
-// Maybe SIOCGIFINDEX? but this doesn't appear to exist in the darwin headers.
 func getLinkAddr(name string) (*netroute.LinkAddr, error) {
 	rib, err := netroute.FetchRIB(unix.AF_UNSPEC, unix.NET_RT_IFLIST, 0)
 	if err != nil {
@@ -377,53 +379,7 @@ func getLinkAddr(name string) (*netroute.LinkAddr, error) {
 	return nil, nil
 }
 
-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
-		}
-
-		err := addRoute(r.Cidr, t.linkAddr)
-		if err != nil {
-			if errors.Is(err, unix.EEXIST) {
-				t.l.WithField("route", r.Cidr).
-					Warnf("unable to add unsafe_route, identical route already exists")
-			} else {
-				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
-}
-
-func (t *tun) removeRoutes(routes []Route) error {
-	for _, r := range routes {
-		if !r.Install {
-			continue
-		}
-
-		err := delRoute(r.Cidr, t.linkAddr)
-		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 addRoute(prefix netip.Prefix, gateway netroute.Addr) error {
+func (rm *tun) addRoute(prefix 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)
@@ -441,13 +397,13 @@ func addRoute(prefix netip.Prefix, gateway netroute.Addr) error {
 		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: gateway,
+			unix.RTAX_GATEWAY: rm.linkAddr,
 		}
 	} else {
 		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: gateway,
+			unix.RTAX_GATEWAY: rm.linkAddr,
 		}
 	}
 
@@ -464,7 +420,7 @@ func addRoute(prefix netip.Prefix, gateway netroute.Addr) error {
 	return nil
 }
 
-func delRoute(prefix netip.Prefix, gateway netroute.Addr) error {
+func (rm *tun) delRoute(prefix 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)
@@ -481,13 +437,13 @@ func delRoute(prefix netip.Prefix, gateway netroute.Addr) error {
 		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: gateway,
+			unix.RTAX_GATEWAY: rm.linkAddr,
 		}
 	} else {
 		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: gateway,
+			unix.RTAX_GATEWAY: rm.linkAddr,
 		}
 	}
 
@@ -495,6 +451,7 @@ func delRoute(prefix netip.Prefix, gateway netroute.Addr) error {
 	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)
@@ -503,52 +460,34 @@ func delRoute(prefix netip.Prefix, gateway netroute.Addr) error {
 	return nil
 }
 
-func (t *tun) Read(to []byte) (int, error) {
-	buf := make([]byte, len(to)+4)
-
-	n, err := t.ReadWriteCloser.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
+func ioctl(a1, a2, a3 uintptr) error {
+	_, _, errno := unix.Syscall(unix.SYS_IOCTL, a1, a2, a3)
+	if errno != 0 {
+		return errno
 	}
+	return nil
+}
 
-	// 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
+func prefixToMask(prefix netip.Prefix) netip.Addr {
+	bits := prefix.Bits()
+	if prefix.Addr().Is4() {
+		// Create IPv4 netmask from prefix length
+		mask := ^uint32(0) << (32 - bits)
+		return netip.AddrFrom4([4]byte{
+			byte(mask >> 24),
+			byte(mask >> 16),
+			byte(mask >> 8),
+			byte(mask),
+		})
 	} else {
-		return 0, fmt.Errorf("unable to determine IP version from packet")
+		// Create IPv6 netmask from prefix length
+		var mask [16]byte
+		for i := 0; i < bits/8; i++ {
+			mask[i] = 0xff
+		}
+		if bits%8 != 0 {
+			mask[bits/8] = ^byte(0) << (8 - bits%8)
+		}
+		return netip.AddrFrom16(mask)
 	}
-
-	copy(buf[4:], from)
-
-	n, err := t.ReadWriteCloser.Write(buf)
-	return n - 4, err
-}
-
-func (t *tun) Networks() []netip.Prefix {
-	return t.vpnNetworks
-}
-
-func (t *tun) Name() string {
-	return t.Device
-}
-
-func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	return nil, fmt.Errorf("TODO: multiqueue not implemented for darwin")
 }

overlay/tun_freebsd.go

@@ -1,284 +1,77 @@
-//go:build !e2e_testing
-// +build !e2e_testing
+//go:build freebsd && !e2e_testing
+// +build freebsd,!e2e_testing
 
 package overlay
 
 import (
-	"bytes"
-	"errors"
 	"fmt"
 	"io"
-	"io/fs"
 	"net/netip"
-	"sync/atomic"
+	"os/exec"
+	"strconv"
+	"strings"
 	"syscall"
-	"time"
 	"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"
+	wgtun "golang.zx2c4.com/wireguard/tun"
 )
 
-const (
-	// FIODGNAME is defined in sys/sys/filio.h on FreeBSD
-	// For 32-bit systems, use FIODGNAME_32 (not defined in this file: 0x80086678)
-	FIODGNAME        = 0x80106678
-	TUNSIFMODE       = 0x8004745e
-	TUNSIFHEAD       = 0x80047460
-	OSIOCAIFADDR_IN6 = 0x8088691b
-	IN6_IFF_NODAD    = 0x0020
-)
-
-type fiodgnameArg struct {
-	length int32
-	pad    [4]byte
-	buf    unsafe.Pointer
-}
+type tun struct{}
 
+// ifreqRename is used for renaming network interfaces on FreeBSD
 type ifreqRename struct {
 	Name [unix.IFNAMSIZ]byte
 	Data uintptr
 }
 
-type ifreqDestroy struct {
-	Name [unix.IFNAMSIZ]byte
-	pad  [16]byte
-}
-
-type ifReq struct {
-	Name  [unix.IFNAMSIZ]byte
-	Flags uint16
-}
-
-type ifreqMTU struct {
-	Name [unix.IFNAMSIZ]byte
-	MTU  int32
-}
-
-type addrLifetime struct {
-	Expire    uint64
-	Preferred uint64
-	Vltime    uint32
-	Pltime    uint32
-}
-
-type ifreqAlias4 struct {
-	Name     [unix.IFNAMSIZ]byte
-	Addr     unix.RawSockaddrInet4
-	DstAddr  unix.RawSockaddrInet4
-	MaskAddr unix.RawSockaddrInet4
-	VHid     uint32
-}
-
-type ifreqAlias6 struct {
-	Name       [unix.IFNAMSIZ]byte
-	Addr       unix.RawSockaddrInet6
-	DstAddr    unix.RawSockaddrInet6
-	PrefixMask unix.RawSockaddrInet6
-	Flags      uint32
-	Lifetime   addrLifetime
-	VHid       uint32
-}
-
-type tun struct {
-	Device      string
-	vpnNetworks []netip.Prefix
-	MTU         int
-	Routes      atomic.Pointer[[]Route]
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
-	linkAddr    *netroute.LinkAddr
-	l           *logrus.Logger
-	devFd       int
-}
-
-func (t *tun) Read(to []byte) (int, error) {
-	// use readv() to read from the tunnel device, to eliminate the need for copying the buffer
-	if t.devFd < 0 {
-		return -1, syscall.EINVAL
-	}
-
-	// first 4 bytes is protocol family, in network byte order
-	head := make([]byte, 4)
-
-	iovecs := []syscall.Iovec{
-		{&head[0], 4},
-		{&to[0], uint64(len(to))},
-	}
-
-	n, _, errno := syscall.Syscall(syscall.SYS_READV, uintptr(t.devFd), uintptr(unsafe.Pointer(&iovecs[0])), uintptr(2))
-
-	var err error
-	if errno != 0 {
-		err = syscall.Errno(errno)
-	} else {
-		err = nil
-	}
-	// fix bytes read number to exclude header
-	bytesRead := int(n)
-	if bytesRead < 0 {
-		return bytesRead, err
-	} else if bytesRead < 4 {
-		return 0, err
-	} else {
-		return bytesRead - 4, err
-	}
-}
-
-// Write is only valid for single threaded use
-func (t *tun) Write(from []byte) (int, error) {
-	// use writev() to write to the tunnel device, to eliminate the need for copying the buffer
-	if t.devFd < 0 {
-		return -1, syscall.EINVAL
-	}
-
-	if len(from) <= 1 {
-		return 0, syscall.EIO
-	}
-	ipVer := from[0] >> 4
-	var head []byte
-	// first 4 bytes is protocol family, in network byte order
-	if ipVer == 4 {
-		head = []byte{0, 0, 0, syscall.AF_INET}
-	} else if ipVer == 6 {
-		head = []byte{0, 0, 0, syscall.AF_INET6}
-	} else {
-		return 0, fmt.Errorf("unable to determine IP version from packet")
-	}
-	iovecs := []syscall.Iovec{
-		{&head[0], 4},
-		{&from[0], uint64(len(from))},
-	}
-
-	n, _, errno := syscall.Syscall(syscall.SYS_WRITEV, uintptr(t.devFd), uintptr(unsafe.Pointer(&iovecs[0])), uintptr(2))
-
-	var err error
-	if errno != 0 {
-		err = syscall.Errno(errno)
-	} else {
-		err = nil
-	}
-
-	return int(n) - 4, err
-}
-
-func (t *tun) Close() error {
-	if t.devFd >= 0 {
-		err := syscall.Close(t.devFd)
-		if err != nil {
-			t.l.WithError(err).Error("Error closing device")
-		}
-		t.devFd = -1
-
-		c := make(chan struct{})
-		go func() {
-			// destroying the interface can block if a read() is still pending. Do this asynchronously.
-			defer close(c)
-			s, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_DGRAM, syscall.IPPROTO_IP)
-			if err == nil {
-				defer syscall.Close(s)
-				ifreq := ifreqDestroy{Name: t.deviceBytes()}
-				err = ioctl(uintptr(s), syscall.SIOCIFDESTROY, uintptr(unsafe.Pointer(&ifreq)))
-			}
-			if err != nil {
-				t.l.WithError(err).Error("Error destroying tunnel")
-			}
-		}()
-
-		// wait up to 1 second so we start blocking at the ioctl
-		select {
-		case <-c:
-		case <-time.After(1 * time.Second):
-		}
-	}
-
-	return nil
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
+	return nil, fmt.Errorf("newTunFromFd not supported on FreeBSD")
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
-	return nil, fmt.Errorf("newTunFromFd not supported in FreeBSD")
-}
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+	deviceName := c.GetString("tun.dev", "tun")
+	mtu := c.GetInt("tun.mtu", DefaultMTU)
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
-	// Try to open existing tun device
-	var fd int
-	var err error
-	deviceName := c.GetString("tun.dev", "")
-	if deviceName != "" {
-		fd, err = syscall.Open("/dev/"+deviceName, syscall.O_RDWR, 0)
-	}
-	if errors.Is(err, fs.ErrNotExist) || deviceName == "" {
-		// If the device doesn't already exist, request a new one and rename it
-		fd, err = syscall.Open("/dev/tun", syscall.O_RDWR, 0)
-	}
+	// Create WireGuard TUN device
+	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, err
-	}
-
-	// Read the name of the interface
-	var name [16]byte
-	arg := fiodgnameArg{length: 16, buf: unsafe.Pointer(&name)}
-	ctrlErr := ioctl(uintptr(fd), FIODGNAME, uintptr(unsafe.Pointer(&arg)))
-
-	if ctrlErr == nil {
-		// set broadcast mode and multicast
-		ifmode := uint32(unix.IFF_BROADCAST | unix.IFF_MULTICAST)
-		ctrlErr = ioctl(uintptr(fd), TUNSIFMODE, uintptr(unsafe.Pointer(&ifmode)))
+		return nil, fmt.Errorf("failed to create TUN device: %w", err)
 	}
 
-	if ctrlErr == nil {
-		// turn on link-layer mode, to support ipv6
-		ifhead := uint32(1)
-		ctrlErr = ioctl(uintptr(fd), TUNSIFHEAD, uintptr(unsafe.Pointer(&ifhead)))
-	}
-
-	if ctrlErr != nil {
-		return nil, err
-	}
-
-	ifName := string(bytes.TrimRight(name[:], "\x00"))
-	if deviceName == "" {
-		deviceName = ifName
+	// Get the actual device name
+	actualName, err := tunDevice.Name()
+	if err != nil {
+		tunDevice.Close()
+		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
 	// If the name doesn't match the desired interface name, rename it now
-	if ifName != deviceName {
-		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
-		if err != nil {
-			return nil, err
-		}
-		defer syscall.Close(s)
-
-		fd := uintptr(s)
-
-		var fromName [16]byte
-		var toName [16]byte
-		copy(fromName[:], ifName)
-		copy(toName[:], deviceName)
-
-		ifrr := ifreqRename{
-			Name: fromName,
-			Data: uintptr(unsafe.Pointer(&toName)),
+	if actualName != deviceName && deviceName != "" && deviceName != "tun" {
+		if err := renameInterface(actualName, deviceName); err != nil {
+			tunDevice.Close()
+			return nil, fmt.Errorf("failed to rename interface from %s to %s: %w", actualName, deviceName, err)
 		}
-
-		// Set the device name
-		ioctl(fd, syscall.SIOCSIFNAME, uintptr(unsafe.Pointer(&ifrr)))
+		actualName = deviceName
 	}
 
-	t := &tun{
-		Device:      deviceName,
+	t := &wgTun{
+		tunDevice:   tunDevice,
 		vpnNetworks: vpnNetworks,
-		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		MaxMTU:      mtu,
+		DefaultMTU:  mtu,
 		l:           l,
-		devFd:       fd,
 	}
 
+	// Create FreeBSD-specific route manager
+	t.routeManager = &tun{}
+
 	err = t.reload(c, true)
 	if err != nil {
+		tunDevice.Close()
 		return nil, err
 	}
 
@@ -289,180 +82,86 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 		}
 	})
 
-	return t, nil
-}
-
-func (t *tun) addIp(cidr netip.Prefix) error {
-	if cidr.Addr().Is4() {
-		ifr := ifreqAlias4{
-			Name: t.deviceBytes(),
-			Addr: unix.RawSockaddrInet4{
-				Len:    unix.SizeofSockaddrInet4,
-				Family: unix.AF_INET,
-				Addr:   cidr.Addr().As4(),
-			},
-			DstAddr: unix.RawSockaddrInet4{
-				Len:    unix.SizeofSockaddrInet4,
-				Family: unix.AF_INET,
-				Addr:   getBroadcast(cidr).As4(),
-			},
-			MaskAddr: unix.RawSockaddrInet4{
-				Len:    unix.SizeofSockaddrInet4,
-				Family: unix.AF_INET,
-				Addr:   prefixToMask(cidr).As4(),
-			},
-			VHid: 0,
-		}
-		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
-		if err != nil {
-			return err
-		}
-		defer syscall.Close(s)
-		// Note: unix.SIOCAIFADDR corresponds to FreeBSD's OSIOCAIFADDR
-		if err := ioctl(uintptr(s), unix.SIOCAIFADDR, uintptr(unsafe.Pointer(&ifr))); err != nil {
-			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr().String(), err)
-		}
-		return nil
-	}
-
-	if cidr.Addr().Is6() {
-		ifr := ifreqAlias6{
-			Name: t.deviceBytes(),
-			Addr: unix.RawSockaddrInet6{
-				Len:    unix.SizeofSockaddrInet6,
-				Family: unix.AF_INET6,
-				Addr:   cidr.Addr().As16(),
-			},
-			PrefixMask: unix.RawSockaddrInet6{
-				Len:    unix.SizeofSockaddrInet6,
-				Family: unix.AF_INET6,
-				Addr:   prefixToMask(cidr).As16(),
-			},
-			Lifetime: addrLifetime{
-				Expire:    0,
-				Preferred: 0,
-				Vltime:    0xffffffff,
-				Pltime:    0xffffffff,
-			},
-			Flags: IN6_IFF_NODAD,
-		}
-		s, err := syscall.Socket(syscall.AF_INET6, syscall.SOCK_DGRAM, syscall.IPPROTO_IP)
-		if err != nil {
-			return err
-		}
-		defer syscall.Close(s)
-
-		if err := ioctl(uintptr(s), OSIOCAIFADDR_IN6, uintptr(unsafe.Pointer(&ifr))); err != nil {
-			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr().String(), err)
-		}
-		return nil
-	}
+	l.WithField("name", actualName).Info("Created WireGuard TUN device")
 
-	return fmt.Errorf("unknown address type %v", cidr)
+	return t, nil
 }
 
-func (t *tun) Activate() error {
-	// Setup our default MTU
-	err := t.setMTU()
+func (rm *tun) Activate(t *wgTun) error {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
 
-	linkAddr, err := getLinkAddr(t.Device)
-	if err != nil {
-		return err
-	}
-	if linkAddr == nil {
-		return fmt.Errorf("unable to discover link_addr for tun interface")
-	}
-	t.linkAddr = linkAddr
+	// Set the MTU
+	rm.SetMTU(t, t.MaxMTU)
 
-	for i := range t.vpnNetworks {
-		err := t.addIp(t.vpnNetworks[i])
-		if err != nil {
+	// Add IP addresses
+	for _, network := range t.vpnNetworks {
+		if err := rm.addIP(t, name, network); err != nil {
 			return err
 		}
 	}
 
-	return t.addRoutes(false)
-}
+	// Bring up the interface
+	if err := runCommandBSD("ifconfig", name, "up"); err != nil {
+		return fmt.Errorf("failed to bring up interface: %w", err)
+	}
 
-func (t *tun) setMTU() error {
-	// Set the MTU on the device
-	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
-	if err != nil {
+	// Set the routes
+	if err := rm.AddRoutes(t, false); err != nil {
 		return err
 	}
-	defer syscall.Close(s)
 
-	ifm := ifreqMTU{Name: t.deviceBytes(), MTU: int32(t.MTU)}
-	err = ioctl(uintptr(s), unix.SIOCSIFMTU, uintptr(unsafe.Pointer(&ifm)))
-	return err
+	return nil
 }
 
-func (t *tun) reload(c *config.C, initial bool) error {
-	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
+func (rm *tun) SetMTU(t *wgTun, mtu int) {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return err
+		t.l.WithError(err).Error("Failed to get device name for MTU set")
+		return
 	}
 
-	if !initial && !change {
-		return nil
+	if err := runCommandBSD("ifconfig", name, "mtu", strconv.Itoa(mtu)); err != nil {
+		t.l.WithError(err).Error("Failed to set tun mtu")
 	}
+}
 
-	routeTree, err := makeRouteTree(t.l, routes, false)
+func (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
+	// On FreeBSD, routes are set via ifconfig and route commands
+	return nil
+}
+
+func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
 
-	// Teach nebula how to handle the routes before establishing them in the system table
-	oldRoutes := t.Routes.Swap(&routes)
-	t.routeTree.Store(routeTree)
-
-	if !initial {
-		// Remove first, if the system removes a wanted route hopefully it will be re-added next
-		err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
-		if err != nil {
-			util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
-		}
-
-		// Ensure any routes we actually want are installed
-		err = t.addRoutes(true)
-		if err != nil {
-			// Catch any stray logs
-			util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
+	routes := *t.Routes.Load()
+	for _, r := range routes {
+		if !r.Install {
+			continue
 		}
-	}
-
-	return nil
-}
 
-func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
-	r, _ := t.routeTree.Load().Lookup(ip)
-	return r
-}
+		// Add route using route command
+		args := []string{"add"}
 
-func (t *tun) Networks() []netip.Prefix {
-	return t.vpnNetworks
-}
+		if r.Cidr.Addr().Is6() {
+			args = append(args, "-inet6")
+		} else {
+			args = append(args, "-inet")
+		}
 
-func (t *tun) Name() string {
-	return t.Device
-}
+		args = append(args, r.Cidr.String(), "-interface", name)
 
-func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	return nil, fmt.Errorf("TODO: multiqueue not implemented for freebsd")
-}
-
-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
+		if r.Metric > 0 {
+			// FreeBSD doesn't support route metrics directly like Linux
+			t.l.WithField("route", r).Warn("Route metrics are not fully supported on FreeBSD")
 		}
 
-		err := addRoute(r.Cidr, t.linkAddr)
+		err := runCommandBSD("route", args...)
 		if err != nil {
 			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
@@ -478,142 +177,100 @@ func (t *tun) addRoutes(logErrors bool) error {
 	return nil
 }
 
-func (t *tun) removeRoutes(routes []Route) error {
+func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		t.l.WithError(err).Error("Failed to get device name for route removal")
+		return
+	}
+
 	for _, r := range routes {
 		if !r.Install {
 			continue
 		}
 
-		err := delRoute(r.Cidr, t.linkAddr)
+		args := []string{"delete"}
+
+		if r.Cidr.Addr().Is6() {
+			args = append(args, "-inet6")
+		} else {
+			args = append(args, "-inet")
+		}
+
+		args = append(args, r.Cidr.String(), "-interface", name)
+
+		err := runCommandBSD("route", args...)
 		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 (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+	// FreeBSD doesn't support multi-queue TUN devices in the same way as Linux
+	// Return a reader that wraps the same device
+	return &wgTunReader{
+		parent:    t,
+		tunDevice: t.tunDevice,
+		batchSize: 64,
+		offset:    0,
+		l:         t.l,
+	}, nil
 }
 
-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 {
-		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
-	}
-	defer unix.Close(sock)
+func (rm *tun) addIP(t *wgTun, name string, network netip.Prefix) error {
+	addr := network.Addr()
 
-	route := &netroute.RouteMessage{
-		Version: unix.RTM_VERSION,
-		Type:    unix.RTM_ADD,
-		Flags:   unix.RTF_UP,
-		Seq:     1,
-	}
-
-	if prefix.Addr().Is4() {
-		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: gateway,
+	if addr.Is4() {
+		// For IPv4: ifconfig tun0 10.0.0.1/24
+		if err := runCommandBSD("ifconfig", name, network.String()); err != nil {
+			return fmt.Errorf("failed to add IPv4 address: %w", err)
 		}
 	} else {
-		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: gateway,
+		// For IPv6: ifconfig tun0 inet6 add 2001:db8::1/64
+		if err := runCommandBSD("ifconfig", name, "inet6", "add", network.String()); err != nil {
+			return fmt.Errorf("failed to add IPv6 address: %w", err)
 		}
 	}
 
-	data, err := route.Marshal()
-	if err != nil {
-		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
-	}
+	return nil
+}
 
-	_, err = unix.Write(sock, data[:])
+func runCommandBSD(name string, args ...string) error {
+	cmd := exec.Command(name, args...)
+	output, err := cmd.CombinedOutput()
 	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[:])
-			fmt.Println("DOING CHANGE")
-			return err
-		}
-		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+		return fmt.Errorf("%s %s failed: %w\nOutput: %s", name, strings.Join(args, " "), err, string(output))
 	}
-
 	return nil
 }
 
-func delRoute(prefix netip.Prefix, gateway netroute.Addr) error {
-	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+func renameInterface(fromName, toName string) error {
+	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
 	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() {
-		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: gateway,
-		}
-	} else {
-		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: gateway,
-		}
+		return fmt.Errorf("failed to create socket: %w", err)
 	}
+	defer syscall.Close(s)
 
-	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)
-	}
+	fd := uintptr(s)
 
-	return nil
-}
+	var fromNameBytes [unix.IFNAMSIZ]byte
+	var toNameBytes [unix.IFNAMSIZ]byte
+	copy(fromNameBytes[:], fromName)
+	copy(toNameBytes[:], toName)
 
-// getLinkAddr Gets the link address for the interface of the given name
-func getLinkAddr(name string) (*netroute.LinkAddr, error) {
-	rib, err := netroute.FetchRIB(unix.AF_UNSPEC, unix.NET_RT_IFLIST, 0)
-	if err != nil {
-		return nil, err
-	}
-	msgs, err := netroute.ParseRIB(unix.NET_RT_IFLIST, rib)
-	if err != nil {
-		return nil, err
+	ifrr := ifreqRename{
+		Name: fromNameBytes,
+		Data: uintptr(unsafe.Pointer(&toNameBytes)),
 	}
 
-	for _, m := range msgs {
-		switch m := m.(type) {
-		case *netroute.InterfaceMessage:
-			if m.Name == name {
-				sa, ok := m.Addrs[unix.RTAX_IFP].(*netroute.LinkAddr)
-				if ok {
-					return sa, nil
-				}
-			}
-		}
+	// Set the device name using SIOCSIFNAME ioctl
+	_, _, errno := syscall.Syscall(syscall.SYS_IOCTL, fd, syscall.SIOCSIFNAME, uintptr(unsafe.Pointer(&ifrr)))
+	if errno != 0 {
+		return fmt.Errorf("SIOCSIFNAME ioctl failed: %w", errno)
 	}
 
-	return nil, nil
+	return nil
 }

overlay/tun_linux.go

@@ -1,5 +1,5 @@
-//go:build !android && !e2e_testing
-// +build !android,!e2e_testing
+//go:build linux && !android && !e2e_testing
+// +build linux,!android,!e2e_testing
 
 package overlay
 
@@ -9,133 +9,60 @@ import (
 	"net"
 	"net/netip"
 	"os"
-	"strings"
-	"sync/atomic"
 	"time"
 	"unsafe"
 
-	"github.com/gaissmai/bart"
 	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/routing"
 	"github.com/slackhq/nebula/util"
 	"github.com/vishvananda/netlink"
 	"golang.org/x/sys/unix"
+	wgtun "golang.zx2c4.com/wireguard/tun"
 )
 
 type tun struct {
-	io.ReadWriteCloser
-	fd          int
-	Device      string
-	vpnNetworks []netip.Prefix
-	MaxMTU      int
-	DefaultMTU  int
-	TXQueueLen  int
-	deviceIndex int
-	ioctlFd     uintptr
-
-	Routes                    atomic.Pointer[[]Route]
-	routeTree                 atomic.Pointer[bart.Table[routing.Gateways]]
-	routeChan                 chan struct{}
+	deviceIndex               int
+	ioctlFd                   uintptr
 	useSystemRoutes           bool
 	useSystemRoutesBufferSize int
-
-	l *logrus.Logger
-}
-
-func (t *tun) Networks() []netip.Prefix {
-	return t.vpnNetworks
-}
-
-type ifReq struct {
-	Name  [16]byte
-	Flags uint16
-	pad   [8]byte
 }
 
-type ifreqMTU struct {
-	Name [16]byte
-	MTU  int32
-	pad  [8]byte
-}
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, multiqueue bool) (*wgTun, error) {
+	deviceName := c.GetString("tun.dev", "")
+	mtu := c.GetInt("tun.mtu", DefaultMTU)
 
-type ifreqQLEN struct {
-	Name  [16]byte
-	Value int32
-	pad   [8]byte
-}
-
-func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix) (*tun, error) {
-	file := os.NewFile(uintptr(deviceFd), "/dev/net/tun")
-
-	t, err := newTunGeneric(c, l, file, vpnNetworks)
+	// Create WireGuard TUN device
+	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("failed to create TUN device: %w", err)
 	}
 
-	t.Device = "tun0"
-
-	return t, nil
-}
-
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, multiqueue bool) (*tun, error) {
-	fd, err := unix.Open("/dev/net/tun", os.O_RDWR, 0)
+	// Get the actual device name
+	actualName, err := tunDevice.Name()
 	if err != nil {
-		// If /dev/net/tun doesn't exist, try to create it (will happen in docker)
-		if os.IsNotExist(err) {
-			err = os.MkdirAll("/dev/net", 0755)
-			if err != nil {
-				return nil, fmt.Errorf("/dev/net/tun doesn't exist, failed to mkdir -p /dev/net: %w", err)
-			}
-			err = unix.Mknod("/dev/net/tun", unix.S_IFCHR|0600, int(unix.Mkdev(10, 200)))
-			if err != nil {
-				return nil, fmt.Errorf("failed to create /dev/net/tun: %w", err)
-			}
-
-			fd, err = unix.Open("/dev/net/tun", os.O_RDWR, 0)
-			if err != nil {
-				return nil, fmt.Errorf("created /dev/net/tun, but still failed: %w", err)
-			}
-		} else {
-			return nil, err
-		}
-	}
-
-	var req ifReq
-	req.Flags = uint16(unix.IFF_TUN | unix.IFF_NO_PI)
-	if multiqueue {
-		req.Flags |= unix.IFF_MULTI_QUEUE
-	}
-	copy(req.Name[:], c.GetString("tun.dev", ""))
-	if err = ioctl(uintptr(fd), uintptr(unix.TUNSETIFF), uintptr(unsafe.Pointer(&req))); err != nil {
-		return nil, err
+		tunDevice.Close()
+		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
-	name := strings.Trim(string(req.Name[:]), "\x00")
 
-	file := os.NewFile(uintptr(fd), "/dev/net/tun")
-	t, err := newTunGeneric(c, l, file, vpnNetworks)
-	if err != nil {
-		return nil, err
+	t := &wgTun{
+		tunDevice:   tunDevice,
+		vpnNetworks: vpnNetworks,
+		MaxMTU:      mtu,
+		DefaultMTU:  mtu,
+		l:           l,
 	}
 
-	t.Device = name
-
-	return t, nil
-}
-
-func newTunGeneric(c *config.C, l *logrus.Logger, file *os.File, vpnNetworks []netip.Prefix) (*tun, error) {
-	t := &tun{
-		ReadWriteCloser:           file,
-		fd:                        int(file.Fd()),
-		vpnNetworks:               vpnNetworks,
-		TXQueueLen:                c.GetInt("tun.tx_queue", 500),
+	// Create Linux-specific route manager
+	routeManager := &tun{
 		useSystemRoutes:           c.GetBool("tun.use_system_route_table", false),
 		useSystemRoutesBufferSize: c.GetInt("tun.use_system_route_table_buffer_size", 0),
-		l:                         l,
 	}
+	t.routeManager = routeManager
 
-	err := t.reload(c, true)
+	err = t.reload(c, true)
 	if err != nil {
+		tunDevice.Close()
 		return nil, err
 	}
 
@@ -146,276 +73,151 @@ func newTunGeneric(c *config.C, l *logrus.Logger, file *os.File, vpnNetworks []n
 		}
 	})
 
+	l.WithField("name", actualName).Info("Created WireGuard TUN device")
+
 	return t, nil
 }
 
-func (t *tun) reload(c *config.C, initial bool) error {
-	routeChange, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
-	if err != nil {
-		return err
-	}
-
-	if !initial && !routeChange && !c.HasChanged("tun.mtu") {
-		return nil
-	}
-
-	routeTree, err := makeRouteTree(t.l, routes, true)
+func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix) (*wgTun, error) {
+	// Create TUN device from file descriptor
+	file := os.NewFile(uintptr(deviceFd), "/dev/net/tun")
+	mtu := c.GetInt("tun.mtu", DefaultMTU)
+	tunDevice, err := wgtun.CreateTUNFromFile(file, mtu)
 	if err != nil {
-		return err
+		return nil, fmt.Errorf("failed to create TUN device from fd: %w", err)
 	}
 
-	oldDefaultMTU := t.DefaultMTU
-	oldMaxMTU := t.MaxMTU
-	newDefaultMTU := c.GetInt("tun.mtu", DefaultMTU)
-	newMaxMTU := newDefaultMTU
-	for i, r := range routes {
-		if r.MTU == 0 {
-			routes[i].MTU = newDefaultMTU
-		}
-
-		if r.MTU > t.MaxMTU {
-			newMaxMTU = r.MTU
-		}
+	t := &wgTun{
+		tunDevice:   tunDevice,
+		vpnNetworks: vpnNetworks,
+		MaxMTU:      mtu,
+		DefaultMTU:  mtu,
+		l:           l,
 	}
 
-	t.MaxMTU = newMaxMTU
-	t.DefaultMTU = newDefaultMTU
-
-	// Teach nebula how to handle the routes before establishing them in the system table
-	oldRoutes := t.Routes.Swap(&routes)
-	t.routeTree.Store(routeTree)
-
-	if !initial {
-		if oldMaxMTU != newMaxMTU {
-			t.setMTU()
-			t.l.Infof("Set max MTU to %v was %v", t.MaxMTU, oldMaxMTU)
-		}
-
-		if oldDefaultMTU != newDefaultMTU {
-			for i := range t.vpnNetworks {
-				err := t.setDefaultRoute(t.vpnNetworks[i])
-				if err != nil {
-					t.l.Warn(err)
-				} else {
-					t.l.Infof("Set default MTU to %v was %v", t.DefaultMTU, oldDefaultMTU)
-				}
-			}
-		}
-
-		// Remove first, if the system removes a wanted route hopefully it will be re-added next
-		t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
-
-		// Ensure any routes we actually want are installed
-		err = t.addRoutes(true)
-		if err != nil {
-			// This should never be called since addRoutes should log its own errors in a reload condition
-			util.LogWithContextIfNeeded("Failed to refresh routes", err, t.l)
-		}
+	// Create Linux-specific route manager
+	routeManager := &tun{
+		useSystemRoutes:           c.GetBool("tun.use_system_route_table", false),
+		useSystemRoutesBufferSize: c.GetInt("tun.use_system_route_table_buffer_size", 0),
 	}
+	t.routeManager = routeManager
 
-	return nil
-}
-
-func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	fd, err := unix.Open("/dev/net/tun", os.O_RDWR, 0)
+	err = t.reload(c, true)
 	if err != nil {
+		tunDevice.Close()
 		return nil, err
 	}
 
-	var req ifReq
-	req.Flags = uint16(unix.IFF_TUN | unix.IFF_NO_PI | unix.IFF_MULTI_QUEUE)
-	copy(req.Name[:], t.Device)
-	if err = ioctl(uintptr(fd), uintptr(unix.TUNSETIFF), uintptr(unsafe.Pointer(&req))); err != nil {
-		return nil, err
-	}
-
-	file := os.NewFile(uintptr(fd), "/dev/net/tun")
-
-	return file, nil
-}
-
-func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
-	r, _ := t.routeTree.Load().Lookup(ip)
-	return r
-}
-
-func (t *tun) Write(b []byte) (int, error) {
-	var nn int
-	maximum := len(b)
-
-	for {
-		n, err := unix.Write(t.fd, b[nn:maximum])
-		if n > 0 {
-			nn += n
-		}
-		if nn == len(b) {
-			return nn, err
-		}
-
+	c.RegisterReloadCallback(func(c *config.C) {
+		err := t.reload(c, false)
 		if err != nil {
-			return nn, err
-		}
-
-		if n == 0 {
-			return nn, io.ErrUnexpectedEOF
+			util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
 		}
-	}
-}
-
-func (t *tun) deviceBytes() (o [16]byte) {
-	for i, c := range t.Device {
-		o[i] = byte(c)
-	}
-	return
-}
+	})
 
-func hasNetlinkAddr(al []*netlink.Addr, x netlink.Addr) bool {
-	for i := range al {
-		if al[i].Equal(x) {
-			return true
-		}
-	}
-	return false
+	return t, nil
 }
 
-// addIPs uses netlink to add all addresses that don't exist, then it removes ones that should not be there
-func (t *tun) addIPs(link netlink.Link) error {
-	newAddrs := make([]*netlink.Addr, len(t.vpnNetworks))
-	for i := range t.vpnNetworks {
-		newAddrs[i] = &netlink.Addr{
-			IPNet: &net.IPNet{
-				IP:   t.vpnNetworks[i].Addr().AsSlice(),
-				Mask: net.CIDRMask(t.vpnNetworks[i].Bits(), t.vpnNetworks[i].Addr().BitLen()),
-			},
-			Label: t.vpnNetworks[i].Addr().Zone(),
-		}
+func (rm *tun) Activate(t *wgTun) error {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
 
-	//add all new addresses
-	for i := range newAddrs {
-		//AddrReplace still adds new IPs, but if their properties change it will change them as well
-		if err := netlink.AddrReplace(link, newAddrs[i]); err != nil {
-			return err
-		}
+	if t.routeManager.useSystemRoutes {
+		t.watchRoutes()
 	}
 
-	//iterate over remainder, remove whoever shouldn't be there
-	al, err := netlink.AddrList(link, netlink.FAMILY_ALL)
+	// Get the netlink device
+	link, err := netlink.LinkByName(name)
 	if err != nil {
-		return fmt.Errorf("failed to get tun address list: %s", err)
-	}
-
-	for i := range al {
-		if hasNetlinkAddr(newAddrs, al[i]) {
-			continue
-		}
-		err = netlink.AddrDel(link, &al[i])
-		if err != nil {
-			t.l.WithError(err).Error("failed to remove address from tun address list")
-		} else {
-			t.l.WithField("removed", al[i].String()).Info("removed address not listed in cert(s)")
-		}
+		return fmt.Errorf("failed to get tun device link: %s", err)
 	}
 
-	return nil
-}
-
-func (t *tun) Activate() error {
-	devName := t.deviceBytes()
-
-	if t.useSystemRoutes {
-		t.watchRoutes()
-	}
+	rm.deviceIndex = link.Attrs().Index
 
+	// Open socket for ioctl operations
 	s, err := unix.Socket(
-		unix.AF_INET, //because everything we use t.ioctlFd for is address family independent, this is fine
+		unix.AF_INET,
 		unix.SOCK_DGRAM,
 		unix.IPPROTO_IP,
 	)
 	if err != nil {
 		return err
 	}
-	t.ioctlFd = uintptr(s)
-
-	// Set the device name
-	ifrf := ifReq{Name: devName}
-	if err = ioctl(t.ioctlFd, unix.SIOCGIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
-		return fmt.Errorf("failed to set tun device name: %s", err)
-	}
+	rm.ioctlFd = uintptr(s)
 
-	link, err := netlink.LinkByName(t.Device)
-	if err != nil {
-		return fmt.Errorf("failed to get tun device link: %s", err)
-	}
-
-	t.deviceIndex = link.Attrs().Index
-
-	// Setup our default MTU
-	t.setMTU()
+	// Set the MTU
+	rm.SetMTU(t, t.MaxMTU)
 
 	// Set the transmit queue length
-	ifrq := ifreqQLEN{Name: devName, Value: int32(t.TXQueueLen)}
-	if err = ioctl(t.ioctlFd, unix.SIOCSIFTXQLEN, uintptr(unsafe.Pointer(&ifrq))); err != nil {
-		// If we can't set the queue length nebula will still work but it may lead to packet loss
+	txQueueLen := 500 // default
+	devName := deviceBytes(name)
+	ifrq := ifreqQLEN{Name: devName, Value: int32(txQueueLen)}
+	if err = ioctl(t.routeManager.ioctlFd, unix.SIOCSIFTXQLEN, uintptr(unsafe.Pointer(&ifrq))); err != nil {
 		t.l.WithError(err).Error("Failed to set tun tx queue length")
 	}
 
+	// Disable IPv6 link-local address generation
 	const modeNone = 1
 	if err = netlink.LinkSetIP6AddrGenMode(link, modeNone); err != nil {
 		t.l.WithError(err).Warn("Failed to disable link local address generation")
 	}
 
-	if err = t.addIPs(link); err != nil {
+	// Add IP addresses
+	if err = t.routeManager.addIPs(t, link); err != nil {
 		return err
 	}
 
 	// Bring up the interface
-	ifrf.Flags = ifrf.Flags | unix.IFF_UP
-	if err = ioctl(t.ioctlFd, unix.SIOCSIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
+	if err = netlink.LinkSetUp(link); err != nil {
 		return fmt.Errorf("failed to bring the tun device up: %s", err)
 	}
 
-	//set route MTU
+	// Set route MTU
 	for i := range t.vpnNetworks {
-		if err = t.setDefaultRoute(t.vpnNetworks[i]); err != nil {
+		if err = t.routeManager.SetDefaultRoute(t, t.vpnNetworks[i]); err != nil {
 			return fmt.Errorf("failed to set default route MTU: %w", err)
 		}
 	}
 
 	// Set the routes
-	if err = t.addRoutes(false); err != nil {
+	if err = t.routeManager.AddRoutes(t, false); err != nil {
 		return err
 	}
 
-	// Run the interface
-	ifrf.Flags = ifrf.Flags | unix.IFF_UP | unix.IFF_RUNNING
-	if err = ioctl(t.ioctlFd, unix.SIOCSIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
-		return fmt.Errorf("failed to run tun device: %s", err)
-	}
-
 	return nil
 }
 
-func (t *tun) setMTU() {
-	// Set the MTU on the device
-	ifm := ifreqMTU{Name: t.deviceBytes(), MTU: int32(t.MaxMTU)}
-	if err := ioctl(t.ioctlFd, unix.SIOCSIFMTU, uintptr(unsafe.Pointer(&ifm))); err != nil {
-		// This is currently a non fatal condition because the route table must have the MTU set appropriately as well
+func (rm *tun) SetMTU(t *wgTun, mtu int) {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		t.l.WithError(err).Error("Failed to get device name for MTU set")
+		return
+	}
+
+	link, err := netlink.LinkByName(name)
+	if err != nil {
+		t.l.WithError(err).Error("Failed to get link for MTU set")
+		return
+	}
+
+	if err := netlink.LinkSetMTU(link, mtu); err != nil {
 		t.l.WithError(err).Error("Failed to set tun mtu")
 	}
 }
 
-func (t *tun) setDefaultRoute(cidr netip.Prefix) error {
+func (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
 	dr := &net.IPNet{
 		IP:   cidr.Masked().Addr().AsSlice(),
 		Mask: net.CIDRMask(cidr.Bits(), cidr.Addr().BitLen()),
 	}
 
 	nr := netlink.Route{
-		LinkIndex: t.deviceIndex,
+		LinkIndex: t.routeManager.deviceIndex,
 		Dst:       dr,
 		MTU:       t.DefaultMTU,
-		AdvMSS:    t.advMSS(Route{}),
+		AdvMSS:    advMSS(Route{}, t.DefaultMTU, t.MaxMTU),
 		Scope:     unix.RT_SCOPE_LINK,
 		Src:       net.IP(cidr.Addr().AsSlice()),
 		Protocol:  unix.RTPROT_KERNEL,
@@ -425,7 +227,7 @@ func (t *tun) setDefaultRoute(cidr netip.Prefix) error {
 	err := netlink.RouteReplace(&nr)
 	if err != nil {
 		t.l.WithError(err).WithField("cidr", cidr).Warn("Failed to set default route MTU, retrying")
-		//retry twice more -- on some systems there appears to be a race condition where if we set routes too soon, netlink says `invalid argument`
+		// Retry twice more
 		for i := 0; i < 2; i++ {
 			time.Sleep(100 * time.Millisecond)
 			err = netlink.RouteReplace(&nr)
@@ -443,8 +245,7 @@ func (t *tun) setDefaultRoute(cidr netip.Prefix) error {
 	return nil
 }
 
-func (t *tun) addRoutes(logErrors bool) error {
-	// Path routes
+func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
 	routes := *t.Routes.Load()
 	for _, r := range routes {
 		if !r.Install {
@@ -457,10 +258,10 @@ func (t *tun) addRoutes(logErrors bool) error {
 		}
 
 		nr := netlink.Route{
-			LinkIndex: t.deviceIndex,
+			LinkIndex: t.routeManager.deviceIndex,
 			Dst:       dr,
 			MTU:       r.MTU,
-			AdvMSS:    t.advMSS(r),
+			AdvMSS:    advMSS(r, t.DefaultMTU, t.MaxMTU),
 			Scope:     unix.RT_SCOPE_LINK,
 		}
 
@@ -484,7 +285,7 @@ func (t *tun) addRoutes(logErrors bool) error {
 	return nil
 }
 
-func (t *tun) removeRoutes(routes []Route) {
+func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
 	for _, r := range routes {
 		if !r.Install {
 			continue
@@ -496,10 +297,10 @@ func (t *tun) removeRoutes(routes []Route) {
 		}
 
 		nr := netlink.Route{
-			LinkIndex: t.deviceIndex,
+			LinkIndex: t.routeManager.deviceIndex,
 			Dst:       dr,
 			MTU:       r.MTU,
-			AdvMSS:    t.advMSS(r),
+			AdvMSS:    advMSS(r, t.DefaultMTU, t.MaxMTU),
 			Scope:     unix.RT_SCOPE_LINK,
 		}
 
@@ -516,30 +317,108 @@ func (t *tun) removeRoutes(routes []Route) {
 	}
 }
 
-func (t *tun) Name() string {
-	return t.Device
+func (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+	// For Linux with WireGuard TUN, we can reuse the same device
+	// The vectorized I/O will handle batching
+	return &wgTunReader{
+		parent:    t,
+		tunDevice: t.tunDevice,
+		batchSize: 64, // Default batch size
+		offset:    0,
+		l:         t.l,
+	}, nil
+}
+
+// Helper functions
+
+func deviceBytes(name string) [16]byte {
+	var o [16]byte
+	for i, c := range name {
+		if i >= 16 {
+			break
+		}
+		o[i] = byte(c)
+	}
+	return o
 }
 
-func (t *tun) advMSS(r Route) int {
+func advMSS(r Route, defaultMTU, maxMTU int) int {
 	mtu := r.MTU
 	if r.MTU == 0 {
-		mtu = t.DefaultMTU
+		mtu = defaultMTU
 	}
 
 	// We only need to set advmss if the route MTU does not match the device MTU
-	if mtu != t.MaxMTU {
+	if mtu != maxMTU {
 		return mtu - 40
 	}
 	return 0
 }
 
-func (t *tun) watchRoutes() {
+type ifreqQLEN struct {
+	Name  [16]byte
+	Value int32
+	pad   [8]byte
+}
+
+func hasNetlinkAddr(al []*netlink.Addr, x netlink.Addr) bool {
+	for i := range al {
+		if al[i].Equal(x) {
+			return true
+		}
+	}
+	return false
+}
+
+func (rm *tun) addIPs(t *wgTun, link netlink.Link) error {
+	newAddrs := make([]*netlink.Addr, len(t.vpnNetworks))
+	for i := range t.vpnNetworks {
+		newAddrs[i] = &netlink.Addr{
+			IPNet: &net.IPNet{
+				IP:   t.vpnNetworks[i].Addr().AsSlice(),
+				Mask: net.CIDRMask(t.vpnNetworks[i].Bits(), t.vpnNetworks[i].Addr().BitLen()),
+			},
+			Label: t.vpnNetworks[i].Addr().Zone(),
+		}
+	}
+
+	// Add all new addresses
+	for i := range newAddrs {
+		if err := netlink.AddrReplace(link, newAddrs[i]); err != nil {
+			return err
+		}
+	}
+
+	// Iterate over remainder, remove whoever shouldn't be there
+	al, err := netlink.AddrList(link, netlink.FAMILY_ALL)
+	if err != nil {
+		return fmt.Errorf("failed to get tun address list: %s", err)
+	}
+
+	for i := range al {
+		if hasNetlinkAddr(newAddrs, al[i]) {
+			continue
+		}
+		err = netlink.AddrDel(link, &al[i])
+		if err != nil {
+			t.l.WithError(err).Error("failed to remove address from tun address list")
+		} else {
+			t.l.WithField("removed", al[i].String()).Info("removed address not listed in cert(s)")
+		}
+	}
+
+	return nil
+}
+
+// watchRoutes monitors system route changes
+func (t *wgTun) watchRoutes() {
+
 	rch := make(chan netlink.RouteUpdate)
 	doneChan := make(chan struct{})
 
 	netlinkOptions := netlink.RouteSubscribeOptions{
-		ReceiveBufferSize:      t.useSystemRoutesBufferSize,
-		ReceiveBufferForceSize: t.useSystemRoutesBufferSize != 0,
+		ReceiveBufferSize:      t.routeManager.useSystemRoutesBufferSize,
+		ReceiveBufferForceSize: t.routeManager.useSystemRoutesBufferSize != 0,
 		ErrorCallback:          func(e error) { t.l.WithError(e).Errorf("netlink error") },
 	}
 
@@ -557,37 +436,61 @@ func (t *tun) watchRoutes() {
 				if ok {
 					t.updateRoutes(r)
 				} else {
-					// may be should do something here as
-					// netlink stops sending updates
 					return
 				}
 			case <-doneChan:
-				// netlink.RouteSubscriber will close the rch for us
 				return
 			}
 		}
 	}()
 }
 
-func (t *tun) isGatewayInVpnNetworks(gwAddr netip.Addr) bool {
-	withinNetworks := false
-	for i := range t.vpnNetworks {
-		if t.vpnNetworks[i].Contains(gwAddr) {
-			withinNetworks = true
-			break
-		}
+func (t *wgTun) updateRoutes(r netlink.RouteUpdate) {
+	gateways := t.getGatewaysFromRoute(&r.Route, t.routeManager.deviceIndex)
+
+	if len(gateways) == 0 {
+		t.l.WithField("route", r).Debug("Ignoring route update, no gateways")
+		return
 	}
 
-	return withinNetworks
-}
+	if r.Dst == nil {
+		t.l.WithField("route", r).Debug("Ignoring route update, no destination address")
+		return
+	}
+
+	dstAddr, ok := netip.AddrFromSlice(r.Dst.IP)
+	if !ok {
+		t.l.WithField("route", r).Debug("Ignoring route update, invalid destination address")
+		return
+	}
+
+	ones, _ := r.Dst.Mask.Size()
+	dst := netip.PrefixFrom(dstAddr, ones)
+
+	newTree := t.routeTree.Load().Clone()
 
-func (t *tun) getGatewaysFromRoute(r *netlink.Route) routing.Gateways {
+	if r.Type == unix.RTM_NEWROUTE {
+		t.l.WithField("destination", dst).WithField("via", gateways).Info("Adding route")
+		newTree.Insert(dst, gateways)
+	} else {
+		t.l.WithField("destination", dst).WithField("via", gateways).Info("Removing route")
+		newTree.Delete(dst)
+	}
+	t.routeTree.Store(newTree)
+}
 
+func (t *wgTun) getGatewaysFromRoute(r *netlink.Route, deviceIndex int) routing.Gateways {
 	var gateways routing.Gateways
 
-	link, err := netlink.LinkByName(t.Device)
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		t.l.WithField("Devicename", t.Device).Error("Ignoring route update: failed to get link by name")
+		t.l.Error("Ignoring route update: failed to get device name")
+		return gateways
+	}
+
+	link, err := netlink.LinkByName(name)
+	if err != nil {
+		t.l.WithField("DeviceName", name).Error("Ignoring route update: failed to get link by name")
 		return gateways
 	}
 
@@ -600,7 +503,6 @@ func (t *tun) getGatewaysFromRoute(r *netlink.Route) routing.Gateways {
 			gwAddr = gwAddr.Unmap()
 
 			if !t.isGatewayInVpnNetworks(gwAddr) {
-				// Gateway isn't in our overlay network, ignore
 				t.l.WithField("route", r).Debug("Ignoring route update, not in our network")
 			} else {
 				gateways = append(gateways, routing.NewGateway(gwAddr, 1))
@@ -609,7 +511,6 @@ func (t *tun) getGatewaysFromRoute(r *netlink.Route) routing.Gateways {
 	}
 
 	for _, p := range r.MultiPath {
-		// If this route is relevant to our interface and there is a gateway then add it
 		if p.LinkIndex == link.Attrs().Index && len(p.Gw) > 0 {
 			gwAddr, ok := netip.AddrFromSlice(p.Gw)
 			if !ok {
@@ -618,10 +519,8 @@ func (t *tun) getGatewaysFromRoute(r *netlink.Route) routing.Gateways {
 				gwAddr = gwAddr.Unmap()
 
 				if !t.isGatewayInVpnNetworks(gwAddr) {
-					// Gateway isn't in our overlay network, ignore
 					t.l.WithField("route", r).Debug("Ignoring route update, not in our network")
 				} else {
-					// p.Hops+1 = weight of the route
 					gateways = append(gateways, routing.NewGateway(gwAddr, p.Hops+1))
 				}
 			}
@@ -632,55 +531,19 @@ func (t *tun) getGatewaysFromRoute(r *netlink.Route) routing.Gateways {
 	return gateways
 }
 
-func (t *tun) updateRoutes(r netlink.RouteUpdate) {
-
-	gateways := t.getGatewaysFromRoute(&r.Route)
-
-	if len(gateways) == 0 {
-		// No gateways relevant to our network, no routing changes required.
-		t.l.WithField("route", r).Debug("Ignoring route update, no gateways")
-		return
-	}
-
-	if r.Dst == nil {
-		t.l.WithField("route", r).Debug("Ignoring route update, no destination address")
-		return
-	}
-
-	dstAddr, ok := netip.AddrFromSlice(r.Dst.IP)
-	if !ok {
-		t.l.WithField("route", r).Debug("Ignoring route update, invalid destination address")
-		return
-	}
-
-	ones, _ := r.Dst.Mask.Size()
-	dst := netip.PrefixFrom(dstAddr, ones)
-
-	newTree := t.routeTree.Load().Clone()
-
-	if r.Type == unix.RTM_NEWROUTE {
-		t.l.WithField("destination", dst).WithField("via", gateways).Info("Adding route")
-		newTree.Insert(dst, gateways)
-
-	} else {
-		t.l.WithField("destination", dst).WithField("via", gateways).Info("Removing route")
-		newTree.Delete(dst)
+func (t *wgTun) isGatewayInVpnNetworks(gwAddr netip.Addr) bool {
+	for i := range t.vpnNetworks {
+		if t.vpnNetworks[i].Contains(gwAddr) {
+			return true
+		}
 	}
-	t.routeTree.Store(newTree)
+	return false
 }
 
-func (t *tun) Close() error {
-	if t.routeChan != nil {
-		close(t.routeChan)
+func ioctl(a1, a2, a3 uintptr) error {
+	_, _, errno := unix.Syscall(unix.SYS_IOCTL, a1, a2, a3)
+	if errno != 0 {
+		return errno
 	}
-
-	if t.ReadWriteCloser != nil {
-		_ = t.ReadWriteCloser.Close()
-	}
-
-	if t.ioctlFd > 0 {
-		_ = os.NewFile(t.ioctlFd, "ioctlFd").Close()
-	}
-
 	return nil
 }

overlay/tun_netbsd.go

@@ -1,111 +1,59 @@
-//go:build !e2e_testing
-// +build !e2e_testing
+//go:build netbsd && !e2e_testing
+// +build netbsd,!e2e_testing
 
 package overlay
 
 import (
-	"errors"
 	"fmt"
 	"io"
 	"net/netip"
-	"os"
-	"regexp"
-	"sync/atomic"
-	"syscall"
-	"unsafe"
+	"os/exec"
+	"strconv"
+	"strings"
 
-	"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"
+	wgtun "golang.zx2c4.com/wireguard/tun"
 )
 
-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 {
-	Device      string
-	vpnNetworks []netip.Prefix
-	MTU         int
-	Routes      atomic.Pointer[[]Route]
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
-	l           *logrus.Logger
-	f           *os.File
-	fd          int
-}
+type tun struct{}
 
-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")
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
+	return nil, fmt.Errorf("newTunFromFd not supported on NetBSD")
 }
 
-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 /dev/tunN must be specified")
-	}
-	if !deviceNameRE.MatchString(deviceName) {
-		return nil, fmt.Errorf("a device name in the format of /dev/tunN must be specified")
-	}
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+	deviceName := c.GetString("tun.dev", "tun")
+	mtu := c.GetInt("tun.mtu", DefaultMTU)
 
-	fd, err := unix.Open("/dev/"+deviceName, os.O_RDWR, 0)
+	// Create WireGuard TUN device
+	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("failed to create TUN device: %w", err)
 	}
 
-	err = unix.SetNonblock(fd, true)
+	// Get the actual device name
+	actualName, err := tunDevice.Name()
 	if err != nil {
-		l.WithError(err).Warn("Failed to set the tun device as nonblocking")
+		tunDevice.Close()
+		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
-	t := &tun{
-		f:           os.NewFile(uintptr(fd), ""),
-		fd:          fd,
-		Device:      deviceName,
+	t := &wgTun{
+		tunDevice:   tunDevice,
 		vpnNetworks: vpnNetworks,
-		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		MaxMTU:      mtu,
+		DefaultMTU:  mtu,
 		l:           l,
 	}
 
+	// Create NetBSD-specific route manager
+	t.routeManager = &tun{}
+
 	err = t.reload(c, true)
 	if err != nil {
+		tunDevice.Close()
 		return nil, err
 	}
 
@@ -116,294 +64,86 @@ 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)
-	}
+	l.WithField("name", actualName).Info("Created WireGuard TUN device")
 
-	// 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
-	}
+	return t, 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
-	})
+func (rm *tun) Activate(t *wgTun) error {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return 0, err
-	}
-
-	if errno != 0 {
-		return 0, errno
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
 
-	return int(n) - 4, err
-}
+	// Set the MTU
+	rm.SetMTU(t, t.MaxMTU)
 
-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 {
+	// Add IP addresses
+	for _, network := range t.vpnNetworks {
+		if err := rm.addIP(t, name, network); 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
 	}
 
-	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)
+	// Bring up the interface
+	if err := runCommandBSD("ifconfig", name, "up"); err != nil {
+		return fmt.Errorf("failed to bring up interface: %w", err)
+	}
 
-		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
+	// Set the routes
+	if err := rm.AddRoutes(t, false); err != nil {
+		return err
 	}
 
-	return fmt.Errorf("unknown address type %v", cidr)
+	return nil
 }
 
-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)))
+func (rm *tun) SetMTU(t *wgTun, mtu int) {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return fmt.Errorf("failed to set tun device head: %w", err)
+		t.l.WithError(err).Error("Failed to get device name for MTU set")
+		return
 	}
 
-	err = t.doIoctlByName(unix.SIOCSIFMTU, uint32(t.MTU))
-	if err != nil {
-		return fmt.Errorf("failed to set tun mtu: %w", err)
+	if err := runCommandBSD("ifconfig", name, "mtu", strconv.Itoa(mtu)); err != nil {
+		t.l.WithError(err).Error("Failed to set tun mtu")
 	}
-
-	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 (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
+	// On NetBSD, routes are set via ifconfig and route commands
+	return nil
 }
 
-func (t *tun) reload(c *config.C, initial bool) error {
-	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
-	if err != nil {
-		return err
-	}
-
-	if !initial && !change {
-		return nil
-	}
-
-	routeTree, err := makeRouteTree(t.l, routes, false)
+func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
 
-	// Teach nebula how to handle the routes before establishing them in the system table
-	oldRoutes := t.Routes.Swap(&routes)
-	t.routeTree.Store(routeTree)
-
-	if !initial {
-		// Remove first, if the system removes a wanted route hopefully it will be re-added next
-		err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
-		if err != nil {
-			util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
-		}
-
-		// Ensure any routes we actually want are installed
-		err = t.addRoutes(true)
-		if err != nil {
-			// Catch any stray logs
-			util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
+	routes := *t.Routes.Load()
+	for _, r := range routes {
+		if !r.Install {
+			continue
 		}
-	}
-
-	return nil
-}
 
-func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
-	r, _ := t.routeTree.Load().Lookup(ip)
-	return r
-}
+		// Add route using route command
+		args := []string{"add"}
 
-func (t *tun) Networks() []netip.Prefix {
-	return t.vpnNetworks
-}
-
-func (t *tun) Name() string {
-	return t.Device
-}
-
-func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	return nil, fmt.Errorf("TODO: multiqueue not implemented for netbsd")
-}
+		if r.Cidr.Addr().Is6() {
+			args = append(args, "-inet6")
+		} else {
+			args = append(args, "-inet")
+		}
 
-func (t *tun) addRoutes(logErrors bool) error {
-	routes := *t.Routes.Load()
+		args = append(args, r.Cidr.String(), "-interface", name)
 
-	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
+		if r.Metric > 0 {
+			// NetBSD doesn't support route metrics directly like Linux
+			t.l.WithField("route", r).Warn("Route metrics are not fully supported on NetBSD")
 		}
 
-		err := addRoute(r.Cidr, t.vpnNetworks)
+		err := runCommandBSD("route", args...)
 		if err != nil {
 			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
@@ -419,131 +159,72 @@ func (t *tun) addRoutes(logErrors bool) error {
 	return nil
 }
 
-func (t *tun) removeRoutes(routes []Route) error {
+func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		t.l.WithError(err).Error("Failed to get device name for route removal")
+		return
+	}
+
 	for _, r := range routes {
 		if !r.Install {
 			continue
 		}
 
-		err := delRoute(r.Cidr, t.vpnNetworks)
+		args := []string{"delete"}
+
+		if r.Cidr.Addr().Is6() {
+			args = append(args, "-inet6")
+		} else {
+			args = append(args, "-inet")
+		}
+
+		args = append(args, r.Cidr.String(), "-interface", name)
+
+		err := runCommandBSD("route", args...)
 		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 (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+	// NetBSD doesn't support multi-queue TUN devices in the same way as Linux
+	// Return a reader that wraps the same device
+	return &wgTunReader{
+		parent:    t,
+		tunDevice: t.tunDevice,
+		batchSize: 64,
+		offset:    0,
+		l:         t.l,
+	}, nil
 }
 
-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,
-	}
+func (rm *tun) addIP(t *wgTun, name string, network netip.Prefix) error {
+	addr := network.Addr()
 
-	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()},
+	if addr.Is4() {
+		// For IPv4: ifconfig tun0 10.0.0.1/24
+		if err := runCommandBSD("ifconfig", name, network.String()); err != nil {
+			return fmt.Errorf("failed to add IPv4 address: %w", err)
 		}
 	} 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()},
+		// For IPv6: ifconfig tun0 inet6 add 2001:db8::1/64
+		if err := runCommandBSD("ifconfig", name, "inet6", "add", network.String()); err != nil {
+			return fmt.Errorf("failed to add IPv6 address: %w", err)
 		}
 	}
 
-	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)
+func runCommandBSD(name string, args ...string) error {
+	cmd := exec.Command(name, args...)
+	output, err := cmd.CombinedOutput()
 	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,
+		return fmt.Errorf("%s %s failed: %w\nOutput: %s", name, strings.Join(args, " "), err, string(output))
 	}
-
-	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_notwin.go

@@ -1,14 +0,0 @@
-//go:build !windows
-// +build !windows
-
-package overlay
-
-import "syscall"
-
-func ioctl(a1, a2, a3 uintptr) error {
-	_, _, errno := syscall.Syscall(syscall.SYS_IOCTL, a1, a2, a3)
-	if errno != 0 {
-		return errno
-	}
-	return nil
-}

overlay/tun_openbsd.go

@@ -1,104 +1,59 @@
-//go:build !e2e_testing
-// +build !e2e_testing
+//go:build openbsd && !e2e_testing
+// +build openbsd,!e2e_testing
 
 package overlay
 
 import (
-	"errors"
 	"fmt"
 	"io"
 	"net/netip"
-	"os"
-	"regexp"
-	"sync/atomic"
-	"syscall"
-	"unsafe"
+	"os/exec"
+	"strconv"
+	"strings"
 
-	"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"
+	wgtun "golang.zx2c4.com/wireguard/tun"
 )
 
-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
-	MTU         int
-	Routes      atomic.Pointer[[]Route]
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
-	l           *logrus.Logger
-	f           *os.File
-	fd          int
-	// cache out buffer since we need to prepend 4 bytes for tun metadata
-	out []byte
-}
+type tun struct{}
 
-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 newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
+	return nil, fmt.Errorf("newTunFromFd not supported on 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 /dev/tunN must be specified")
-	}
-	if !deviceNameRE.MatchString(deviceName) {
-		return nil, fmt.Errorf("a device name in the format of /dev/tunN must be specified")
-	}
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+	deviceName := c.GetString("tun.dev", "tun")
+	mtu := c.GetInt("tun.mtu", DefaultMTU)
 
-	fd, err := unix.Open("/dev/"+deviceName, os.O_RDWR, 0)
+	// Create WireGuard TUN device
+	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("failed to create TUN device: %w", err)
 	}
 
-	err = unix.SetNonblock(fd, true)
+	// Get the actual device name
+	actualName, err := tunDevice.Name()
 	if err != nil {
-		l.WithError(err).Warn("Failed to set the tun device as nonblocking")
+		tunDevice.Close()
+		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
-	t := &tun{
-		f:           os.NewFile(uintptr(fd), ""),
-		fd:          fd,
-		Device:      deviceName,
+	t := &wgTun{
+		tunDevice:   tunDevice,
 		vpnNetworks: vpnNetworks,
-		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		MaxMTU:      mtu,
+		DefaultMTU:  mtu,
 		l:           l,
 	}
 
+	// Create OpenBSD-specific route manager
+	t.routeManager = &tun{}
+
 	err = t.reload(c, true)
 	if err != nil {
+		tunDevice.Close()
 		return nil, err
 	}
 
@@ -109,221 +64,86 @@ 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)
+	l.WithField("name", actualName).Info("Created WireGuard TUN device")
 
-	n, err := t.f.Read(buf)
-
-	copy(to, buf[4:])
-	return n - 4, err
+	return t, 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]
-
-	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")
+func (rm *tun) Activate(t *wgTun) error {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
 
-	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(),
-		}
+	// Set the MTU
+	rm.SetMTU(t, t.MaxMTU)
 
-		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
-		if err != nil {
+	// Add IP addresses
+	for _, network := range t.vpnNetworks {
+		if err := rm.addIP(t, name, network); 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)
-		}
+	// Bring up the interface
+	if err := runCommandBSD("ifconfig", name, "up"); err != nil {
+		return fmt.Errorf("failed to bring up interface: %w", err)
+	}
 
-		return nil
+	// Set the routes
+	if err := rm.AddRoutes(t, false); err != nil {
+		return err
 	}
 
-	return fmt.Errorf("unknown address type %v", cidr)
+	return nil
 }
 
-func (t *tun) Activate() error {
-	err := t.doIoctlByName(unix.SIOCSIFMTU, uint32(t.MTU))
+func (rm *tun) SetMTU(t *wgTun, mtu int) {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return fmt.Errorf("failed to set tun mtu: %w", err)
+		t.l.WithError(err).Error("Failed to get device name for MTU set")
+		return
 	}
 
-	for i := range t.vpnNetworks {
-		err = t.addIp(t.vpnNetworks[i])
-		if err != nil {
-			return err
-		}
+	if err := runCommandBSD("ifconfig", name, "mtu", strconv.Itoa(mtu)); err != nil {
+		t.l.WithError(err).Error("Failed to set tun mtu")
 	}
-
-	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 (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
+	// On OpenBSD, routes are set via ifconfig and route commands
+	return nil
 }
 
-func (t *tun) reload(c *config.C, initial bool) error {
-	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
+func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
+	name, err := t.tunDevice.Name()
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to get device name: %w", err)
 	}
 
-	if !initial && !change {
-		return nil
-	}
-
-	routeTree, err := makeRouteTree(t.l, routes, false)
-	if err != nil {
-		return err
-	}
-
-	// Teach nebula how to handle the routes before establishing them in the system table
-	oldRoutes := t.Routes.Swap(&routes)
-	t.routeTree.Store(routeTree)
-
-	if !initial {
-		// Remove first, if the system removes a wanted route hopefully it will be re-added next
-		err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
-		if err != nil {
-			util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
-		}
-
-		// Ensure any routes we actually want are installed
-		err = t.addRoutes(true)
-		if err != nil {
-			// Catch any stray logs
-			util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
+	routes := *t.Routes.Load()
+	for _, r := range routes {
+		if !r.Install {
+			continue
 		}
-	}
 
-	return nil
-}
+		// Add route using route command
+		args := []string{"add"}
 
-func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
-	r, _ := t.routeTree.Load().Lookup(ip)
-	return r
-}
-
-func (t *tun) Networks() []netip.Prefix {
-	return t.vpnNetworks
-}
-
-func (t *tun) Name() string {
-	return t.Device
-}
-
-func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	return nil, fmt.Errorf("TODO: multiqueue not implemented for openbsd")
-}
+		if r.Cidr.Addr().Is6() {
+			args = append(args, "-inet6")
+		} else {
+			args = append(args, "-inet")
+		}
 
-func (t *tun) addRoutes(logErrors bool) error {
-	routes := *t.Routes.Load()
+		args = append(args, r.Cidr.String(), "-interface", name)
 
-	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
+		if r.Metric > 0 {
+			// OpenBSD doesn't support route metrics directly like Linux
+			t.l.WithField("route", r).Warn("Route metrics are not fully supported on OpenBSD")
 		}
 
-		err := addRoute(r.Cidr, t.vpnNetworks)
+		err := runCommandBSD("route", args...)
 		if err != nil {
 			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
@@ -339,131 +159,72 @@ func (t *tun) addRoutes(logErrors bool) error {
 	return nil
 }
 
-func (t *tun) removeRoutes(routes []Route) error {
+func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		t.l.WithError(err).Error("Failed to get device name for route removal")
+		return
+	}
+
 	for _, r := range routes {
 		if !r.Install {
 			continue
 		}
 
-		err := delRoute(r.Cidr, t.vpnNetworks)
+		args := []string{"delete"}
+
+		if r.Cidr.Addr().Is6() {
+			args = append(args, "-inet6")
+		} else {
+			args = append(args, "-inet")
+		}
+
+		args = append(args, r.Cidr.String(), "-interface", name)
+
+		err := runCommandBSD("route", args...)
 		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 (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+	// OpenBSD doesn't support multi-queue TUN devices in the same way as Linux
+	// Return a reader that wraps the same device
+	return &wgTunReader{
+		parent:    t,
+		tunDevice: t.tunDevice,
+		batchSize: 64,
+		offset:    0,
+		l:         t.l,
+	}, nil
 }
 
-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)
+func (rm *tun) addIP(t *wgTun, name string, network netip.Prefix) error {
+	addr := network.Addr()
 
-	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()},
+	if addr.Is4() {
+		// For IPv4: ifconfig tun0 10.0.0.1/24
+		if err := runCommandBSD("ifconfig", name, network.String()); err != nil {
+			return fmt.Errorf("failed to add IPv4 address: %w", err)
 		}
 	} 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
+		// For IPv6: ifconfig tun0 inet6 add 2001:db8::1/64
+		if err := runCommandBSD("ifconfig", name, "inet6", "add", network.String()); err != nil {
+			return fmt.Errorf("failed to add IPv6 address: %w", 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)
+func runCommandBSD(name string, args ...string) error {
+	cmd := exec.Command(name, args...)
+	output, err := cmd.CombinedOutput()
 	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,
+		return fmt.Errorf("%s %s failed: %w\nOutput: %s", name, strings.Join(args, " "), err, string(output))
 	}
-
-	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_wg.go

@@ -0,0 +1,279 @@
+//go:build !android && !e2e_testing
+// +build !android,!e2e_testing
+
+package overlay
+
+import (
+	"fmt"
+	"io"
+	"net/netip"
+	"sync"
+	"sync/atomic"
+
+	"github.com/gaissmai/bart"
+	"github.com/sirupsen/logrus"
+	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/routing"
+	"github.com/slackhq/nebula/util"
+	wgtun "golang.zx2c4.com/wireguard/tun"
+)
+
+// wgTun wraps a WireGuard TUN device and implements the overlay.Device interface
+type wgTun struct {
+	tunDevice   wgtun.Device
+	vpnNetworks []netip.Prefix
+	MaxMTU      int
+	DefaultMTU  int
+
+	Routes      atomic.Pointer[[]Route]
+	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	routeChan   chan struct{}
+
+	// Platform-specific route management
+	routeManager *tun
+
+	l *logrus.Logger
+}
+
+// BatchReader interface for readers that support vectorized I/O
+type BatchReader interface {
+	BatchRead() ([][]byte, []int, error)
+}
+
+// BatchWriter interface for writers that support vectorized I/O
+type BatchWriter interface {
+	BatchWrite(packets [][]byte) (int, error)
+}
+
+// wgTunReader wraps a single TUN queue for multi-queue support
+type wgTunReader struct {
+	parent     *wgTun
+	tunDevice  wgtun.Device
+	buffers    [][]byte
+	sizes      []int
+	offset     int
+	batchSize  int
+	l          *logrus.Logger
+}
+
+var (
+	bufferPool = sync.Pool{
+		New: func() interface{} {
+			buf := make([]byte, 9001) // MTU size
+			return &buf
+		},
+	}
+)
+
+func (t *wgTun) Networks() []netip.Prefix {
+	return t.vpnNetworks
+}
+
+func (t *wgTun) Name() string {
+	name, err := t.tunDevice.Name()
+	if err != nil {
+		t.l.WithError(err).Error("Failed to get TUN device name")
+		return "unknown"
+	}
+	return name
+}
+
+func (t *wgTun) RoutesFor(ip netip.Addr) routing.Gateways {
+	r, _ := t.routeTree.Load().Lookup(ip)
+	return r
+}
+
+func (t *wgTun) Activate() error {
+	if t.routeManager == nil {
+		return fmt.Errorf("route manager not initialized")
+	}
+	return t.routeManager.Activate(t)
+}
+
+// Read implements single-packet read for backward compatibility
+func (t *wgTun) Read(b []byte) (int, error) {
+	bufs := [][]byte{b}
+	sizes := []int{0}
+	n, err := t.tunDevice.Read(bufs, sizes, 0)
+	if err != nil {
+		return 0, err
+	}
+	if n == 0 {
+		return 0, io.ErrNoProgress
+	}
+	return sizes[0], nil
+}
+
+// Write implements single-packet write for backward compatibility
+func (t *wgTun) Write(b []byte) (int, error) {
+	bufs := [][]byte{b}
+	offset := 0
+
+	// WireGuard TUN expects the packet data to start at offset 0
+	n, err := t.tunDevice.Write(bufs, offset)
+	if err != nil {
+		return 0, err
+	}
+	if n == 0 {
+		return 0, io.ErrShortWrite
+	}
+	return len(b), nil
+}
+
+func (t *wgTun) Close() error {
+	if t.routeChan != nil {
+		close(t.routeChan)
+	}
+
+	if t.tunDevice != nil {
+		return t.tunDevice.Close()
+	}
+
+	return nil
+}
+
+func (t *wgTun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
+	// For WireGuard TUN, we need to create separate TUN device instances for multi-queue
+	// The platform-specific implementation will handle this
+	if t.routeManager == nil {
+		return nil, fmt.Errorf("route manager not initialized for multi-queue reader")
+	}
+
+	return t.routeManager.NewMultiQueueReader(t)
+}
+
+func (t *wgTun) reload(c *config.C, initial bool) error {
+	routeChange, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
+	if err != nil {
+		return err
+	}
+
+	if !initial && !routeChange && !c.HasChanged("tun.mtu") {
+		return nil
+	}
+
+	routeTree, err := makeRouteTree(t.l, routes, true)
+	if err != nil {
+		return err
+	}
+
+	oldDefaultMTU := t.DefaultMTU
+	oldMaxMTU := t.MaxMTU
+	newDefaultMTU := c.GetInt("tun.mtu", DefaultMTU)
+	newMaxMTU := newDefaultMTU
+	for i, r := range routes {
+		if r.MTU == 0 {
+			routes[i].MTU = newDefaultMTU
+		}
+
+		if r.MTU > t.MaxMTU {
+			newMaxMTU = r.MTU
+		}
+	}
+
+	t.MaxMTU = newMaxMTU
+	t.DefaultMTU = newDefaultMTU
+
+	// Teach nebula how to handle the routes before establishing them in the system table
+	oldRoutes := t.Routes.Swap(&routes)
+	t.routeTree.Store(routeTree)
+
+	if !initial && t.routeManager != nil {
+		if oldMaxMTU != newMaxMTU {
+			t.routeManager.SetMTU(t, t.MaxMTU)
+			t.l.Infof("Set max MTU to %v was %v", t.MaxMTU, oldMaxMTU)
+		}
+
+		if oldDefaultMTU != newDefaultMTU {
+			for i := range t.vpnNetworks {
+				err := t.routeManager.SetDefaultRoute(t, t.vpnNetworks[i])
+				if err != nil {
+					t.l.Warn(err)
+				} else {
+					t.l.Infof("Set default MTU to %v was %v", t.DefaultMTU, oldDefaultMTU)
+				}
+			}
+		}
+
+		// Remove first, if the system removes a wanted route hopefully it will be re-added next
+		t.routeManager.RemoveRoutes(t, findRemovedRoutes(routes, *oldRoutes))
+
+		// Ensure any routes we actually want are installed
+		err = t.routeManager.AddRoutes(t, true)
+		if err != nil {
+			// This should never be called since AddRoutes should log its own errors in a reload condition
+			util.LogWithContextIfNeeded("Failed to refresh routes", err, t.l)
+		}
+	}
+
+	return nil
+}
+
+// BatchRead reads multiple packets from the TUN device using vectorized I/O
+func (r *wgTunReader) BatchRead() ([][]byte, []int, error) {
+	// Reuse buffers from pool
+	if len(r.buffers) == 0 {
+		r.buffers = make([][]byte, r.batchSize)
+		r.sizes = make([]int, r.batchSize)
+		for i := 0; i < r.batchSize; i++ {
+			buf := bufferPool.Get().(*[]byte)
+			r.buffers[i] = (*buf)[:cap(*buf)]
+		}
+	}
+
+	n, err := r.tunDevice.Read(r.buffers, r.sizes, r.offset)
+	if err != nil {
+		return nil, nil, err
+	}
+
+	return r.buffers[:n], r.sizes[:n], nil
+}
+
+// Read implements io.Reader for wgTunReader (single packet for compatibility)
+func (r *wgTunReader) Read(b []byte) (int, error) {
+	bufs := [][]byte{b}
+	sizes := []int{0}
+	n, err := r.tunDevice.Read(bufs, sizes, r.offset)
+	if err != nil {
+		return 0, err
+	}
+	if n == 0 {
+		return 0, io.ErrNoProgress
+	}
+	return sizes[0], nil
+}
+
+// Write implements io.Writer for wgTunReader
+func (r *wgTunReader) Write(b []byte) (int, error) {
+	bufs := [][]byte{b}
+	n, err := r.tunDevice.Write(bufs, r.offset)
+	if err != nil {
+		return 0, err
+	}
+	if n == 0 {
+		return 0, io.ErrShortWrite
+	}
+	return len(b), nil
+}
+
+// BatchWrite writes multiple packets to the TUN device using vectorized I/O
+func (r *wgTunReader) BatchWrite(packets [][]byte) (int, error) {
+	return r.tunDevice.Write(packets, r.offset)
+}
+
+func (r *wgTunReader) Close() error {
+	// Return buffers to pool
+	for i := range r.buffers {
+		if r.buffers[i] != nil {
+			bufferPool.Put(&r.buffers[i])
+			r.buffers[i] = nil
+		}
+	}
+	r.buffers = nil
+	r.sizes = nil
+
+	if r.tunDevice != nil {
+		return r.tunDevice.Close()
+	}
+	return nil
+}

overlay/tun_windows.go

@@ -1,83 +1,75 @@
-//go:build !e2e_testing
-// +build !e2e_testing
+//go:build windows && !e2e_testing
+// +build windows,!e2e_testing
 
 package overlay
 
 import (
 	"crypto"
+	"encoding/binary"
 	"fmt"
 	"io"
 	"net/netip"
-	"os"
-	"path/filepath"
-	"runtime"
-	"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"
-	"github.com/slackhq/nebula/wintun"
 	"golang.org/x/sys/windows"
+	wgtun "golang.zx2c4.com/wireguard/tun"
 	"golang.zx2c4.com/wireguard/windows/tunnel/winipcfg"
 )
 
 const tunGUIDLabel = "Fixed Nebula Windows GUID v1"
 
-type winTun struct {
-	Device      string
-	vpnNetworks []netip.Prefix
-	MTU         int
-	Routes      atomic.Pointer[[]Route]
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
-	l           *logrus.Logger
-
-	tun *wintun.NativeTun
+type tun struct {
+	luid windows.LUID
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (Device, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in Windows")
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*winTun, error) {
-	err := checkWinTunExists()
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+	deviceName := c.GetString("tun.dev", "Nebula")
+	mtu := c.GetInt("tun.mtu", DefaultMTU)
+
+	// Create WireGuard TUN device
+	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, fmt.Errorf("can not load the wintun driver: %w", err)
+		return nil, fmt.Errorf("failed to create TUN device: %w", err)
 	}
 
-	deviceName := c.GetString("tun.dev", "")
-	guid, err := generateGUIDByDeviceName(deviceName)
+	// Get the actual device name
+	actualName, err := tunDevice.Name()
 	if err != nil {
-		return nil, fmt.Errorf("generate GUID failed: %w", err)
+		tunDevice.Close()
+		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
-	t := &winTun{
-		Device:      deviceName,
+	t := &wgTun{
+		tunDevice:   tunDevice,
 		vpnNetworks: vpnNetworks,
-		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		MaxMTU:      mtu,
+		DefaultMTU:  mtu,
 		l:           l,
 	}
 
-	err = t.reload(c, true)
+	// Create Windows-specific route manager
+	rm := &tun{}
+
+	// Get LUID from the device name
+	luid, err := winipcfg.LUIDFromAlias(actualName)
 	if err != nil {
-		return nil, err
+		tunDevice.Close()
+		return nil, fmt.Errorf("failed to get LUID: %w", err)
 	}
+	rm.luid = luid
+	t.routeManager = rm
 
-	var tunDevice wintun.Device
-	tunDevice, err = wintun.CreateTUNWithRequestedGUID(deviceName, guid, t.MTU)
+	err = t.reload(c, true)
 	if err != nil {
-		// Windows 10 has an issue with unclean shutdowns not fully cleaning up the wintun device.
-		// Trying a second time resolves the issue.
-		l.WithError(err).Debug("Failed to create wintun device, retrying")
-		tunDevice, err = wintun.CreateTUNWithRequestedGUID(deviceName, guid, t.MTU)
-		if err != nil {
-			return nil, fmt.Errorf("create TUN device failed: %w", err)
-		}
+		tunDevice.Close()
+		return nil, err
 	}
-	t.tun = tunDevice.(*wintun.NativeTun)
 
 	c.RegisterReloadCallback(func(c *config.C) {
 		err := t.reload(c, false)
@@ -86,206 +78,132 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 		}
 	})
 
+	l.WithField("name", actualName).Info("Created WireGuard TUN device")
+
 	return t, nil
 }
 
-func (t *winTun) reload(c *config.C, initial bool) error {
-	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
+func (rm *tun) Activate(t *wgTun) error {
+	// Set MTU
+	err := rm.setMTU(t, t.MaxMTU)
 	if err != nil {
-		return err
+		return fmt.Errorf("failed to set MTU: %w", err)
 	}
 
-	if !initial && !change {
-		return nil
+	// Add IP addresses
+	for _, network := range t.vpnNetworks {
+		if err := rm.addIP(t, network); err != nil {
+			return err
+		}
 	}
 
-	routeTree, err := makeRouteTree(t.l, routes, false)
-	if err != nil {
+	// Add routes
+	if err := rm.AddRoutes(t, false); err != nil {
 		return err
 	}
 
-	// Teach nebula how to handle the routes before establishing them in the system table
-	oldRoutes := t.Routes.Swap(&routes)
-	t.routeTree.Store(routeTree)
-
-	if !initial {
-		// Remove first, if the system removes a wanted route hopefully it will be re-added next
-		err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
-		if err != nil {
-			util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
-		}
-
-		// Ensure any routes we actually want are installed
-		err = t.addRoutes(true)
-		if err != nil {
-			// Catch any stray logs
-			util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
-		}
-	}
-
 	return nil
 }
 
-func (t *winTun) Activate() error {
-	luid := winipcfg.LUID(t.tun.LUID())
-
-	err := luid.SetIPAddresses(t.vpnNetworks)
-	if err != nil {
-		return fmt.Errorf("failed to set address: %w", err)
+func (rm *tun) SetMTU(t *wgTun, mtu int) {
+	if err := rm.setMTU(t, mtu); err != nil {
+		t.l.WithError(err).Error("Failed to set MTU")
 	}
+}
 
-	err = t.addRoutes(false)
-	if err != nil {
-		return err
-	}
+func (rm *tun) setMTU(t *wgTun, mtu int) error {
+	// Set MTU using winipcfg
+	return rm.luid.SetIPInterfaceMTU(uint32(mtu))
+}
 
+func (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
+	// On Windows, routes are managed differently
 	return nil
 }
 
-func (t *winTun) addRoutes(logErrors bool) error {
-	luid := winipcfg.LUID(t.tun.LUID())
+func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
 	routes := *t.Routes.Load()
-	foundDefault4 := false
-
 	for _, r := range routes {
-		if len(r.Via) == 0 || !r.Install {
-			// We don't allow route MTUs so only install routes with a via
+		if !r.Install {
 			continue
 		}
 
-		// Add our unsafe route
-		// Windows does not support multipath routes natively, so we install only a single route.
-		// This is not a problem as traffic will always be sent to Nebula which handles the multipath routing internally.
-		// In effect this provides multipath routing support to windows supporting loadbalancing and redundancy.
-		err := luid.AddRoute(r.Cidr, r.Via[0].Addr(), uint32(r.Metric))
+		route := winipcfg.RouteData{
+			Destination: r.Cidr,
+			Metric:      uint32(r.Metric),
+		}
+
+		if r.MTU > 0 {
+			// Windows route MTU is not directly supported
+			t.l.WithField("route", r).Debug("Route MTU is not supported on Windows")
+		}
+
+		err := rm.luid.AddRoute(route.Destination, route.Destination.Addr(), route.Metric)
 		if err != nil {
 			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
 				retErr.Log(t.l)
-				continue
 			} else {
 				return retErr
 			}
 		} else {
 			t.l.WithField("route", r).Info("Added route")
 		}
-
-		if !foundDefault4 {
-			if r.Cidr.Bits() == 0 && r.Cidr.Addr().BitLen() == 32 {
-				foundDefault4 = true
-			}
-		}
 	}
 
-	ipif, err := luid.IPInterface(windows.AF_INET)
-	if err != nil {
-		return fmt.Errorf("failed to get ip interface: %w", err)
-	}
-
-	ipif.NLMTU = uint32(t.MTU)
-	if foundDefault4 {
-		ipif.UseAutomaticMetric = false
-		ipif.Metric = 0
-	}
-
-	if err := ipif.Set(); err != nil {
-		return fmt.Errorf("failed to set ip interface: %w", err)
-	}
 	return nil
 }
 
-func (t *winTun) removeRoutes(routes []Route) error {
-	luid := winipcfg.LUID(t.tun.LUID())
-
+func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
 	for _, r := range routes {
 		if !r.Install {
 			continue
 		}
 
-		// See comment on luid.AddRoute
-		err := luid.DeleteRoute(r.Cidr, r.Via[0].Addr())
+		err := rm.luid.DeleteRoute(r.Cidr, r.Cidr.Addr())
 		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 *winTun) RoutesFor(ip netip.Addr) routing.Gateways {
-	r, _ := t.routeTree.Load().Lookup(ip)
-	return r
-}
-
-func (t *winTun) Networks() []netip.Prefix {
-	return t.vpnNetworks
-}
-
-func (t *winTun) Name() string {
-	return t.Device
-}
-
-func (t *winTun) Read(b []byte) (int, error) {
-	return t.tun.Read(b, 0)
-}
-
-func (t *winTun) Write(b []byte) (int, error) {
-	return t.tun.Write(b, 0)
 }
 
-func (t *winTun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	return nil, fmt.Errorf("TODO: multiqueue not implemented for windows")
+func (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+	// Windows doesn't support multi-queue TUN devices
+	// Return a reader that wraps the same device
+	return &wgTunReader{
+		parent:    t,
+		tunDevice: t.tunDevice,
+		batchSize: 64,
+		offset:    0,
+		l:         t.l,
+	}, nil
 }
 
-func (t *winTun) Close() error {
-	// It seems that the Windows networking stack doesn't like it when we destroy interfaces that have active routes,
-	// so to be certain, just remove everything before destroying.
-	luid := winipcfg.LUID(t.tun.LUID())
-	_ = luid.FlushRoutes(windows.AF_INET)
-	_ = luid.FlushIPAddresses(windows.AF_INET)
-
-	_ = luid.FlushRoutes(windows.AF_INET6)
-	_ = luid.FlushIPAddresses(windows.AF_INET6)
-
-	_ = luid.FlushDNS(windows.AF_INET)
-	_ = luid.FlushDNS(windows.AF_INET6)
-
-	return t.tun.Close()
-}
-
-func generateGUIDByDeviceName(name string) (*windows.GUID, error) {
-	// GUID is 128 bit
-	hash := crypto.MD5.New()
-
-	_, err := hash.Write([]byte(tunGUIDLabel))
-	if err != nil {
-		return nil, err
-	}
-
-	_, err = hash.Write([]byte(name))
+func (rm *tun) addIP(t *wgTun, network netip.Prefix) error {
+	// Add IP address using winipcfg
+	err := rm.luid.AddIPAddress(network)
 	if err != nil {
-		return nil, err
+		return fmt.Errorf("failed to add IP address %s: %w", network, err)
 	}
-
-	sum := hash.Sum(nil)
-
-	return (*windows.GUID)(unsafe.Pointer(&sum[0])), nil
+	return nil
 }
 
-func checkWinTunExists() error {
-	myPath, err := os.Executable()
-	if err != nil {
-		return err
-	}
+// generateGUIDByDeviceName generates a GUID based on the device name
+func generateGUIDByDeviceName(deviceName string) (*windows.GUID, error) {
+	// Hash the device name to create a deterministic GUID
+	h := crypto.SHA256.New()
+	h.Write([]byte(tunGUIDLabel))
+	h.Write([]byte(deviceName))
+	sum := h.Sum(nil)
 
-	arch := runtime.GOARCH
-	switch arch {
-	case "386":
-		//NOTE: wintun bundles 386 as x86
-		arch = "x86"
+	guid := &windows.GUID{
+		Data1: binary.LittleEndian.Uint32(sum[0:4]),
+		Data2: binary.LittleEndian.Uint16(sum[4:6]),
+		Data3: binary.LittleEndian.Uint16(sum[6:8]),
 	}
+	copy(guid.Data4[:], sum[8:16])
 
-	_, err = syscall.LoadDLL(filepath.Join(filepath.Dir(myPath), "dist", "windows", "wintun", "bin", arch, "wintun.dll"))
-	return err
+	return guid, nil
 }