tell stupid ai not to meddle with the paths on other platforms by adding a copy

CHANGELOG.md

@@ -7,12 +7,30 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- Experimental Linux UDP offload support: enable `listen.enable_gso` and
+  `listen.enable_gro` to activate UDP_SEGMENT batching and GRO receive
+  splitting. Includes automatic capability probing, per-packet fallbacks, and
+  runtime metrics/logs for visibility.
+- Optional Linux TUN `virtio_net_hdr` support: set `tun.enable_vnet_hdr` to
+  have Nebula negotiate VNET headers and offload flags so future batches can
+  be delivered to the kernel with metadata instead of per-packet writes.
+- Linux UDP send sharding can now be tuned with `listen.send_shards`; defaults
+  to `GOMAXPROCS` but can be increased to stripe heavy peers across more
+  goroutines.
+
 ### Changed
 
 - `default_local_cidr_any` now defaults to false, meaning that any firewall rule
   intended to target an `unsafe_routes` entry must explicitly declare it via the
   `local_cidr` field. This is almost always the intended behavior. This flag is
   deprecated and will be removed in a future release.
+- UDP receive path now enqueues into per-worker lock-free rings, restoring the
+  `listen.decrypt_workers`/`listen.decrypt_queue_depth` tuning knobs while
+  eliminating the mutex contention from the old shared channel.
+- Increased replay protection window to 32k packets so high-throughput links
+  tolerate larger bursts of reordering without tripping the anti-replay logic.
 
 ## [1.9.4] - 2024-09-09
 

connection_state.go

@@ -13,7 +13,10 @@ import (
 	"github.com/slackhq/nebula/noiseutil"
 )
 
-const ReplayWindow = 1024
+// ReplayWindow controls the size of the sliding window used to detect replays.
+// High-bandwidth links with GRO/GSO can reorder more than a thousand packets in
+// flight, so keep this comfortably above the largest expected burst.
+const ReplayWindow = 32768
 
 type ConnectionState struct {
 	eKey           *NebulaCipherState

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 v0.0.0-20250521234502-f333402bd9cb
+	golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b
 	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 v0.0.0-20250503011706-39ed1f5ac29c
+	gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe
 )
 
 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 v0.7.0 // indirect
+	golang.org/x/time v0.5.0 // 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 v0.7.0 h1:ntUhktv3OPE6TgYxXWv9vKvUSJyIFJlyohwbkEwPrKQ=
+golang.org/x/time v0.5.0 h1:o7cqy6amK/52YcAKIPlM3a+Fpj35zvRj2TP+e1xFSfk=
-golang.org/x/time v0.7.0/go.mod h1:3BpzKBy/shNhVucY/MWOyx10tF3SFh9QdLuxbVysPQM=
+golang.org/x/time v0.5.0/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 v0.0.0-20250521234502-f333402bd9cb h1:whnFRlWMcXI9d+ZbWg+4sHnLp52d5yiIPUxMBSt4X9A=
+golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b h1:J1CaxgLerRR5lgx3wnr6L04cJFbWoceSK9JWBdglINo=
-golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb/go.mod h1:rpwXGsirqLqN2L0JDJQlwOboGHmptD5ZD6T2VmcqhTw=
+golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b/go.mod h1:tqur9LnfstdR9ep2LaJT4lFUl0EjlHtge+gAjmsHUG4=
 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 v0.0.0-20250503011706-39ed1f5ac29c h1:m/r7OM+Y2Ty1sgBQ7Qb27VgIMBW8ZZhT4gLnUyDIhzI=
+gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe h1:fre4i6mv4iBuz5lCMOzHD1rH1ljqHWSICFmZRbbgp3g=
-gvisor.dev/gvisor v0.0.0-20250503011706-39ed1f5ac29c/go.mod h1:3r5CMtNQMKIvBlrmM9xWUNamjKBYPOWyXOjmg5Kts3g=
+gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe/go.mod h1:sxc3Uvk/vHcd3tj7/DHVBoR5wvWT/MmRq2pj7HRJnwU=

inside.go

@@ -33,7 +33,8 @@ 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 {
-			if err := f.writeTun(q, packet); err != nil {
+			_, err := f.readers[q].Write(packet)
+			if err != nil {
 				f.l.WithError(err).Error("Failed to forward to tun")
 			}
 		}
@@ -90,7 +91,8 @@ func (f *Interface) rejectInside(packet []byte, out []byte, q int) {
 		return
 	}
 
-	if err := f.writeTun(q, out); err != nil {
+	_, err := f.readers[q].Write(out)
+	if err != nil {
 		f.l.WithError(err).Error("Failed to write to tun")
 	}
 }

interface.go

@@ -5,9 +5,11 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"math/bits"
 	"net/netip"
 	"os"
 	"runtime"
+	"sync"
 	"sync/atomic"
 	"time"
 
@@ -21,7 +23,12 @@ import (
 	"github.com/slackhq/nebula/udp"
 )
 
-const mtu = 9001
+const (
+	mtu                        = 9001
+	tunReadBufferSize          = mtu * 8
+	defaultDecryptWorkerFactor = 2
+	defaultInboundQueueDepth   = 1024
+)
 
 type InterfaceConfig struct {
 	HostMap            *HostMap
@@ -47,8 +54,9 @@ type InterfaceConfig struct {
 	reQueryWait     time.Duration
 
 	ConntrackCacheTimeout time.Duration
-	batchSize             int
 	l                     *logrus.Logger
+	DecryptWorkers        int
+	DecryptQueueDepth     int
 }
 
 type Interface struct {
@@ -85,7 +93,6 @@ type Interface struct {
 	version     string
 
 	conntrackCacheTimeout time.Duration
-	batchSize             int
 
 	writers []udp.Conn
 	readers []io.ReadWriteCloser
@@ -95,6 +102,166 @@ type Interface struct {
 	cachedPacketMetrics *cachedPacketMetrics
 
 	l              *logrus.Logger
+	ctx            context.Context
+	udpListenWG    sync.WaitGroup
+	inboundPool    sync.Pool
+	decryptWG      sync.WaitGroup
+	decryptQueues  []*inboundRing
+	decryptWorkers int
+	decryptStates  []decryptWorkerState
+	decryptCounter atomic.Uint32
+}
+
+type inboundPacket struct {
+	addr    netip.AddrPort
+	payload []byte
+	release func()
+	queue   int
+}
+
+type decryptWorkerState struct {
+	queue  *inboundRing
+	notify chan struct{}
+}
+
+type decryptContext struct {
+	ctTicker *firewall.ConntrackCacheTicker
+	plain    []byte
+	head     header.H
+	fwPacket firewall.Packet
+	light    *LightHouseHandler
+	nebula   []byte
+}
+
+type inboundCell struct {
+	seq atomic.Uint64
+	pkt *inboundPacket
+}
+
+type inboundRing struct {
+	mask       uint64
+	cells      []inboundCell
+	enqueuePos atomic.Uint64
+	dequeuePos atomic.Uint64
+}
+
+func newInboundRing(capacity int) *inboundRing {
+	if capacity < 2 {
+		capacity = 2
+	}
+	size := nextPowerOfTwo(uint32(capacity))
+	if size < 2 {
+		size = 2
+	}
+	ring := &inboundRing{
+		mask:  uint64(size - 1),
+		cells: make([]inboundCell, size),
+	}
+	for i := range ring.cells {
+		ring.cells[i].seq.Store(uint64(i))
+	}
+	return ring
+}
+
+func nextPowerOfTwo(v uint32) uint32 {
+	if v == 0 {
+		return 1
+	}
+	return 1 << (32 - bits.LeadingZeros32(v-1))
+}
+
+func (r *inboundRing) Enqueue(pkt *inboundPacket) bool {
+	var cell *inboundCell
+	pos := r.enqueuePos.Load()
+	for {
+		cell = &r.cells[pos&r.mask]
+		seq := cell.seq.Load()
+		diff := int64(seq) - int64(pos)
+		if diff == 0 {
+			if r.enqueuePos.CompareAndSwap(pos, pos+1) {
+				break
+			}
+		} else if diff < 0 {
+			return false
+		} else {
+			pos = r.enqueuePos.Load()
+		}
+	}
+	cell.pkt = pkt
+	cell.seq.Store(pos + 1)
+	return true
+}
+
+func (r *inboundRing) Dequeue() (*inboundPacket, bool) {
+	var cell *inboundCell
+	pos := r.dequeuePos.Load()
+	for {
+		cell = &r.cells[pos&r.mask]
+		seq := cell.seq.Load()
+		diff := int64(seq) - int64(pos+1)
+		if diff == 0 {
+			if r.dequeuePos.CompareAndSwap(pos, pos+1) {
+				break
+			}
+		} else if diff < 0 {
+			return nil, false
+		} else {
+			pos = r.dequeuePos.Load()
+		}
+	}
+	pkt := cell.pkt
+	cell.pkt = nil
+	cell.seq.Store(pos + r.mask + 1)
+	return pkt, true
+}
+
+func (f *Interface) getInboundPacket() *inboundPacket {
+	if pkt, ok := f.inboundPool.Get().(*inboundPacket); ok && pkt != nil {
+		return pkt
+	}
+	return &inboundPacket{}
+}
+
+func (f *Interface) putInboundPacket(pkt *inboundPacket) {
+	if pkt == nil {
+		return
+	}
+	pkt.addr = netip.AddrPort{}
+	pkt.payload = nil
+	pkt.release = nil
+	pkt.queue = 0
+	f.inboundPool.Put(pkt)
+}
+
+func newDecryptContext(f *Interface) *decryptContext {
+	return &decryptContext{
+		ctTicker: firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout),
+		plain:    make([]byte, udp.MTU),
+		head:     header.H{},
+		fwPacket: firewall.Packet{},
+		light:    f.lightHouse.NewRequestHandler(),
+		nebula:   make([]byte, 12, 12),
+	}
+}
+
+func (f *Interface) processInboundPacket(pkt *inboundPacket, ctx *decryptContext) {
+	if pkt == nil {
+		return
+	}
+	defer func() {
+		if pkt.release != nil {
+			pkt.release()
+		}
+		f.putInboundPacket(pkt)
+	}()
+
+	ctx.head = header.H{}
+	ctx.fwPacket = firewall.Packet{}
+	var cache firewall.ConntrackCache
+	if ctx.ctTicker != nil {
+		cache = ctx.ctTicker.Get(f.l)
+	}
+	f.readOutsidePackets(pkt.addr, nil, ctx.plain[:0], pkt.payload, &ctx.head, &ctx.fwPacket, ctx.light, ctx.nebula, pkt.queue, cache)
 }
 
 type EncWriter interface {
@@ -112,16 +279,6 @@ type EncWriter interface {
 	GetCertState() *CertState
 }
 
-// BatchReader is an interface for readers that support vectorized packet reading
-type BatchReader interface {
-	BatchRead(buffers [][]byte, sizes []int) (int, error)
-}
-
-// BatchWriter is an interface for writers that support vectorized packet writing
-type BatchWriter interface {
-	BatchWrite([][]byte) (int, error)
-}
-
 type sendRecvErrorConfig uint8
 
 const (
@@ -174,6 +331,35 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 	}
 
 	cs := c.pki.getCertState()
+	decryptWorkers := c.DecryptWorkers
+	if decryptWorkers < 0 {
+		decryptWorkers = 0
+	}
+	if decryptWorkers == 0 {
+		decryptWorkers = c.routines * defaultDecryptWorkerFactor
+		if decryptWorkers < c.routines {
+			decryptWorkers = c.routines
+		}
+	}
+	if decryptWorkers < 0 {
+		decryptWorkers = 0
+	}
+	if runtime.GOOS != "linux" {
+		decryptWorkers = 0
+	}
+
+	queueDepth := c.DecryptQueueDepth
+	if queueDepth <= 0 {
+		queueDepth = defaultInboundQueueDepth
+	}
+	minDepth := c.routines * 64
+	if minDepth <= 0 {
+		minDepth = 64
+	}
+	if queueDepth < minDepth {
+		queueDepth = minDepth
+	}
+
 	ifce := &Interface{
 		pki:                   c.pki,
 		hostMap:               c.HostMap,
@@ -198,7 +384,6 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 		relayManager:          c.relayManager,
 		connectionManager:     c.connectionManager,
 		conntrackCacheTimeout: c.ConntrackCacheTimeout,
-		batchSize:             c.batchSize,
 
 		metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
 		messageMetrics:   c.MessageMetrics,
@@ -208,6 +393,9 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 		},
 
 		l:              c.l,
+		ctx:            ctx,
+		inboundPool:    sync.Pool{New: func() any { return &inboundPacket{} }},
+		decryptWorkers: decryptWorkers,
 	}
 
 	ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
@@ -216,6 +404,19 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 
 	ifce.connectionManager.intf = ifce
 
+	if decryptWorkers > 0 {
+		ifce.decryptQueues = make([]*inboundRing, decryptWorkers)
+		ifce.decryptStates = make([]decryptWorkerState, decryptWorkers)
+		for i := 0; i < decryptWorkers; i++ {
+			queue := newInboundRing(queueDepth)
+			ifce.decryptQueues[i] = queue
+			ifce.decryptStates[i] = decryptWorkerState{
+				queue:  queue,
+				notify: make(chan struct{}, 1),
+			}
+		}
+	}
+
 	return ifce, nil
 }
 
@@ -255,8 +456,68 @@ func (f *Interface) activate() {
 	}
 }
 
+func (f *Interface) startDecryptWorkers() {
+	if f.decryptWorkers <= 0 || len(f.decryptQueues) == 0 {
+		return
+	}
+	f.decryptWG.Add(f.decryptWorkers)
+	for i := 0; i < f.decryptWorkers; i++ {
+		go f.decryptWorker(i)
+	}
+}
+
+func (f *Interface) decryptWorker(id int) {
+	defer f.decryptWG.Done()
+	if id < 0 || id >= len(f.decryptStates) {
+		return
+	}
+	state := f.decryptStates[id]
+	if state.queue == nil {
+		return
+	}
+	ctx := newDecryptContext(f)
+	for {
+		for {
+			pkt, ok := state.queue.Dequeue()
+			if !ok {
+				break
+			}
+			f.processInboundPacket(pkt, ctx)
+		}
+		if f.closed.Load() || f.ctx.Err() != nil {
+			for {
+				pkt, ok := state.queue.Dequeue()
+				if !ok {
+					return
+				}
+				f.processInboundPacket(pkt, ctx)
+			}
+		}
+		select {
+		case <-f.ctx.Done():
+		case <-state.notify:
+		}
+	}
+}
+
+func (f *Interface) notifyDecryptWorker(idx int) {
+	if idx < 0 || idx >= len(f.decryptStates) {
+		return
+	}
+	state := f.decryptStates[idx]
+	if state.notify == nil {
+		return
+	}
+	select {
+	case state.notify <- struct{}{}:
+	default:
+	}
+}
+
 func (f *Interface) run() {
+	f.startDecryptWorkers()
 	// Launch n queues to read packets from udp
+	f.udpListenWG.Add(f.routines)
 	for i := 0; i < f.routines; i++ {
 		go f.listenOut(i)
 	}
@@ -269,6 +530,7 @@ func (f *Interface) run() {
 
 func (f *Interface) listenOut(i int) {
 	runtime.LockOSThread()
+	defer f.udpListenWG.Done()
 
 	var li udp.Conn
 	if i > 0 {
@@ -277,33 +539,78 @@ func (f *Interface) listenOut(i int) {
 		li = f.outside
 	}
 
-	ctCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
+	useWorkers := f.decryptWorkers > 0 && len(f.decryptQueues) > 0
-	lhh := f.lightHouse.NewRequestHandler()
+	var (
-	plaintext := make([]byte, udp.MTU)
+		inlineTicker  *firewall.ConntrackCacheTicker
-	h := &header.H{}
+		inlineHandler *LightHouseHandler
-	fwPacket := &firewall.Packet{}
+		inlinePlain   []byte
-	nb := make([]byte, 12)
+		inlineHeader  header.H
+		inlinePacket  firewall.Packet
+		inlineNB      []byte
+		inlineCtx     *decryptContext
+	)
 
-	li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
+	if useWorkers {
-		f.readOutsidePackets(fromUdpAddr, nil, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
+		inlineCtx = newDecryptContext(f)
+	} else {
+		inlineTicker = firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
+		inlineHandler = f.lightHouse.NewRequestHandler()
+		inlinePlain = make([]byte, udp.MTU)
+		inlineNB = make([]byte, 12, 12)
+	}
+
+	li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte, release func()) {
+		if !useWorkers {
+			if release != nil {
+				defer release()
+			}
+			select {
+			case <-f.ctx.Done():
+				return
+			default:
+			}
+			inlineHeader = header.H{}
+			inlinePacket = firewall.Packet{}
+			var cache firewall.ConntrackCache
+			if inlineTicker != nil {
+				cache = inlineTicker.Get(f.l)
+			}
+			f.readOutsidePackets(fromUdpAddr, nil, inlinePlain[:0], payload, &inlineHeader, &inlinePacket, inlineHandler, inlineNB, i, cache)
+			return
+		}
+
+		if f.ctx.Err() != nil {
+			if release != nil {
+				release()
+			}
+			return
+		}
+
+		pkt := f.getInboundPacket()
+		pkt.addr = fromUdpAddr
+		pkt.payload = payload
+		pkt.release = release
+		pkt.queue = i
+
+		queueCount := len(f.decryptQueues)
+		if queueCount == 0 {
+			f.processInboundPacket(pkt, inlineCtx)
+			return
+		}
+		w := int(f.decryptCounter.Add(1)-1) % queueCount
+		if w < 0 || w >= queueCount || !f.decryptQueues[w].Enqueue(pkt) {
+			f.processInboundPacket(pkt, inlineCtx)
+			return
+		}
+		f.notifyDecryptWorker(w)
 	})
 }
 
 func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	runtime.LockOSThread()
 
-	// Check if reader supports batch operations
+	packet := make([]byte, tunReadBufferSize)
-	if batchReader, ok := reader.(BatchReader); ok {
+	out := make([]byte, tunReadBufferSize)
-		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{}
 	nb := make([]byte, 12, 12)
 
@@ -316,85 +623,15 @@ func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 				return
 			}
 
-			f.l.WithError(err).Error("Fatal error while reading outbound packet, exiting goroutine")
+			f.l.WithError(err).Error("Error while reading outbound packet")
-			return
+			// This only seems to happen when something fatal happens to the fd, so exit.
+			os.Exit(2)
 		}
 
 		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 and buffers for batch reading
-	batchSize := f.batchSize
-	if batchSize <= 0 {
-		batchSize = 64 // Fallback to default if not configured
-	}
-	fwPackets := make([]*firewall.Packet, batchSize)
-	outBuffers := make([][]byte, batchSize)
-	nbBuffers := make([][]byte, batchSize)
-	packets := make([][]byte, batchSize)
-	sizes := make([]int, batchSize)
-
-	for j := 0; j < batchSize; j++ {
-		fwPackets[j] = &firewall.Packet{}
-		outBuffers[j] = make([]byte, mtu)
-		nbBuffers[j] = make([]byte, 12)
-		packets[j] = make([]byte, mtu)
-	}
-
-	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
-
-	for {
-		n, err := reader.BatchRead(packets, sizes)
-		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 < n; idx++ {
-			if 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)
@@ -551,6 +788,19 @@ func (f *Interface) Close() error {
 		}
 	}
 
+	f.udpListenWG.Wait()
+	if f.decryptWorkers > 0 {
+		for _, state := range f.decryptStates {
+			if state.notify != nil {
+				select {
+				case state.notify <- struct{}{}:
+				default:
+				}
+			}
+		}
+		f.decryptWG.Wait()
+	}
+
 	// Release the tun device
 	return f.inside.Close()
 }

main.go

@@ -120,6 +120,8 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		l.WithField("duration", conntrackCacheTimeout).Info("Using routine-local conntrack cache")
 	}
 
+	udp.SetDisableUDPCsum(c.GetBool("listen.disable_udp_checksum", false))
+
 	var tun overlay.Device
 	if !configTest {
 		c.CatchHUP(ctx)
@@ -221,6 +223,9 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		}
 	}
 
+	decryptWorkers := c.GetInt("listen.decrypt_workers", 0)
+	decryptQueueDepth := c.GetInt("listen.decrypt_queue_depth", 0)
+
 	ifConfig := &InterfaceConfig{
 		HostMap:               hostMap,
 		Inside:                tun,
@@ -242,8 +247,9 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		relayManager:          NewRelayManager(ctx, l, hostMap, c),
 		punchy:                punchy,
 		ConntrackCacheTimeout: conntrackCacheTimeout,
-		batchSize:             c.GetInt("tun.batch_size", 64),
 		l:                     l,
+		DecryptWorkers:        decryptWorkers,
+		DecryptQueueDepth:     decryptQueueDepth,
 	}
 
 	var ifce *Interface

outside.go

@@ -333,13 +333,12 @@ func parseV6(data []byte, incoming bool, fp *firewall.Packet) error {
 			}
 
 			fp.Protocol = uint8(proto)
-			ports := data[offset : offset+4]
 			if incoming {
-				fp.RemotePort = binary.BigEndian.Uint16(ports[0:2])
+				fp.RemotePort = binary.BigEndian.Uint16(data[offset : offset+2])
-				fp.LocalPort = binary.BigEndian.Uint16(ports[2:4])
+				fp.LocalPort = binary.BigEndian.Uint16(data[offset+2 : offset+4])
 			} else {
-				fp.LocalPort = binary.BigEndian.Uint16(ports[0:2])
+				fp.LocalPort = binary.BigEndian.Uint16(data[offset : offset+2])
-				fp.RemotePort = binary.BigEndian.Uint16(ports[2:4])
+				fp.RemotePort = binary.BigEndian.Uint16(data[offset+2 : offset+4])
 			}
 
 			fp.Fragment = false
@@ -471,7 +470,13 @@ func (f *Interface) decryptToTun(hostinfo *HostInfo, messageCounter uint64, out
 
 	out, err = hostinfo.ConnectionState.dKey.DecryptDanger(out, packet[:header.Len], packet[header.Len:], messageCounter, nb)
 	if err != nil {
-		hostinfo.logger(f.l).WithError(err).Error("Failed to decrypt packet")
+		hostinfo.logger(f.l).
+			WithError(err).
+			WithField("tag", "decrypt-debug").
+			WithField("remoteIndexLocal", hostinfo.localIndexId).
+			WithField("messageCounter", messageCounter).
+			WithField("packet_len", len(packet)).
+			Error("Failed to decrypt packet")
 		return false
 	}
 

overlay/route.go

@@ -3,6 +3,7 @@ package overlay
 import (
 	"fmt"
 	"math"
+	"net"
 	"net/netip"
 	"runtime"
 	"strconv"
@@ -304,3 +305,29 @@ 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

@@ -225,7 +225,6 @@ 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)
 
@@ -319,7 +318,7 @@ func Test_makeRouteTree(t *testing.T) {
 
 	ip, err = netip.ParseAddr("1.1.0.1")
 	require.NoError(t, err)
-	_, ok = routeTree.Lookup(ip)
+	r, ok = routeTree.Lookup(ip)
 	assert.False(t, ok)
 }
 

overlay/tun.go

@@ -1,6 +1,8 @@
 package overlay
 
 import (
+	"fmt"
+	"net"
 	"net/netip"
 
 	"github.com/sirupsen/logrus"
@@ -70,3 +72,51 @@ 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 darwin && !ios && !e2e_testing
+//go:build !ios && !e2e_testing
-// +build darwin,!ios,!e2e_testing
+// +build !ios,!e2e_testing
 
 package overlay
 
@@ -8,27 +8,48 @@ 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
 }
 
-// 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
@@ -58,61 +79,60 @@ type ifreqAlias6 struct {
 	Lifetime   addrLifetime
 }
 
-const (
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
-	_SIOCAIFADDR_IN6 = 2155899162
-	_IN6_IFF_NODAD   = 0x0020
-)
-
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
-	return nil, fmt.Errorf("newTunFromFd not supported on Darwin")
-}
-
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
 	name := c.GetString("tun.dev", "")
-	deviceName := "utun"
-
-	// Parse device name to handle utun[0-9]+ format
-	if name != "" && name != "utun" {
 	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")
-		} else {
+			ifIndex = -1
-			deviceName = name
 		}
 	}
 
-	mtu := c.GetInt("tun.mtu", DefaultMTU)
+	fd, err := unix.Socket(unix.AF_SYSTEM, unix.SOCK_DGRAM, unix.AF_SYS_CONTROL)
-
-	// Create WireGuard TUN device
-	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, fmt.Errorf("failed to create TUN device: %w", err)
+		return nil, fmt.Errorf("system socket: %v", err)
 	}
 
-	// Get the actual device name
+	var ctlInfo = &unix.CtlInfo{}
-	actualName, err := tunDevice.Name()
+	copy(ctlInfo.Name[:], utunControlName)
+
+	err = unix.IoctlCtlInfo(fd, ctlInfo)
 	if err != nil {
-		tunDevice.Close()
+		return nil, fmt.Errorf("CTLIOCGINFO: %v", err)
-		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
-	t := &wgTun{
+	err = unix.Connect(fd, &unix.SockaddrCtl{
-		tunDevice:   tunDevice,
+		ID:   ctlInfo.Id,
+		Unit: uint32(ifIndex) + 1,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("SYS_CONNECT: %v", err)
+	}
+
+	name, err = unix.GetsockoptString(fd, unix.AF_SYS_CONTROL, _UTUN_OPT_IFNAME)
+	if err != nil {
+		return nil, fmt.Errorf("failed to retrieve tun name: %w", err)
+	}
+
+	err = unix.SetNonblock(fd, true)
+	if err != nil {
+		return nil, fmt.Errorf("SetNonblock: %v", err)
+	}
+
+	t := &tun{
+		ReadWriteCloser: os.NewFile(uintptr(fd), ""),
+		Device:          name,
 		vpnNetworks:     vpnNetworks,
-		MaxMTU:      mtu,
+		DefaultMTU:      c.GetInt("tun.mtu", DefaultMTU),
-		DefaultMTU:  mtu,
 		l:               l,
 	}
 
-	// Create Darwin-specific route manager
-	t.routeManager = &tun{}
-
 	err = t.reload(c, true)
 	if err != nil {
-		tunDevice.Close()
 		return nil, err
 	}
 
@@ -123,251 +143,215 @@ 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 (rm *tun) Activate(t *wgTun) error {
+func (t *tun) deviceBytes() (o [16]byte) {
-	name, err := t.tunDevice.Name()
+	for i, c := range t.Device {
-	if err != nil {
+		o[i] = byte(c)
-		return fmt.Errorf("failed to get device name: %w", err)
+	}
+	return
 }
 
-	// Set the MTU
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
-	rm.SetMTU(t, t.MaxMTU)
+	return nil, fmt.Errorf("newTunFromFd not supported in Darwin")
+}
 
-	// Add IP addresses
+func (t *tun) Close() error {
-	for _, network := range t.vpnNetworks {
+	if t.ReadWriteCloser != nil {
-		if err := rm.addIP(t, name, network); err != nil {
+		return t.ReadWriteCloser.Close()
+	}
+	return nil
+}
+
+func (t *tun) Activate() error {
+	devName := t.deviceBytes()
+
+	s, err := unix.Socket(
+		unix.AF_INET,
+		unix.SOCK_DGRAM,
+		unix.IPPROTO_IP,
+	)
+	if err != nil {
 		return 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)
 	}
 
-	// Bring up the interface using ioctl
+	// Get the device flags
-	if err := rm.bringUpInterface(name); err != nil {
+	ifrf := ifReq{Name: devName}
-		return fmt.Errorf("failed to bring up interface: %w", err)
+	if err = ioctl(fd, unix.SIOCGIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
+		return fmt.Errorf("failed to get tun flags: %s", err)
 	}
 
-	// Get the link address for routing
+	linkAddr, err := getLinkAddr(t.Device)
-	linkAddr, err := getLinkAddr(name)
 	if err != nil {
-		return fmt.Errorf("failed to get link address: %w", err)
+		return err
 	}
 	if linkAddr == nil {
 		return fmt.Errorf("unable to discover link_addr for tun interface")
 	}
-	rm.linkAddr = linkAddr
+	t.linkAddr = linkAddr
 
-	// Set the routes
+	for _, network := range t.vpnNetworks {
-	if err := rm.AddRoutes(t, false); err != nil {
+		if network.Addr().Is4() {
+			err = t.activate4(network)
+			if err != nil {
+				return err
+			}
+		} else {
+			err = t.activate6(network)
+			if err != nil {
+				return err
+			}
+		}
+	}
+
+	// Run the interface
+	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)
+	}
+
+	// Unsafe path routes
+	return t.addRoutes(false)
+}
+
+func (t *tun) activate4(network netip.Prefix) error {
+	s, err := unix.Socket(
+		unix.AF_INET,
+		unix.SOCK_DGRAM,
+		unix.IPPROTO_IP,
+	)
+	if err != nil {
+		return err
+	}
+	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(),
+		},
+	}
+
+	if err := ioctl(uintptr(s), unix.SIOCAIFADDR, uintptr(unsafe.Pointer(&ifr))); err != nil {
+		return fmt.Errorf("failed to set tun v4 address: %s", err)
+	}
+
+	err = addRoute(network, t.linkAddr)
+	if err != nil {
 		return err
 	}
 
 	return nil
 }
 
-func (rm *tun) bringUpInterface(name string) error {
+func (t *tun) activate6(network netip.Prefix) error {
-	// Open a socket for ioctl
+	s, err := unix.Socket(
-	fd, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, 0)
+		unix.AF_INET6,
+		unix.SOCK_DGRAM,
+		unix.IPPROTO_IP,
+	)
 	if err != nil {
-		return fmt.Errorf("failed to create socket: %w", err)
+		return err
-	}
-	defer unix.Close(fd)
-
-	// 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)
-	}
-
-	// Set IFF_UP and IFF_RUNNING flags
-	ifrf.Flags = ifrf.Flags | unix.IFF_UP | unix.IFF_RUNNING
-
-	if err := ioctl(uintptr(fd), unix.SIOCSIFFLAGS, uintptr(unsafe.Pointer(&ifrf))); err != nil {
-		return fmt.Errorf("failed to set interface flags: %w", err)
-	}
-
-	return nil
-}
-
-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
-	}
-
-	// 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)
-
-	// 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")
-	}
-}
-
-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 (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,
-		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 fmt.Errorf("failed to create IPv4 socket: %w", err)
 	}
 	defer unix.Close(s)
 
-	var ifr ifreqAlias4
+	ifr := ifreqAlias6{
-	copy(ifr.Name[:], name)
+		Name: t.deviceBytes(),
-
+		Addr: unix.RawSockaddrInet6{
-	// Set the address
-	ifr.Addr = unix.RawSockaddrInet4{
-		Len:    unix.SizeofSockaddrInet4,
-		Family: unix.AF_INET,
-		Addr:   network.Addr().As4(),
-	}
-
-	// 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 (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 fmt.Errorf("failed to create IPv6 socket: %w", err)
-	}
-	defer unix.Close(s)
-
-	var ifr ifreqAlias6
-	copy(ifr.Name[:], name)
-
-	// Set the address
-	ifr.Addr = unix.RawSockaddrInet6{
 			Len:    unix.SizeofSockaddrInet6,
 			Family: unix.AF_INET6,
 			Addr:   network.Addr().As16(),
-	}
+		},
-
+		PrefixMask: unix.RawSockaddrInet6{
-	// Set the prefix mask
-	ifr.PrefixMask = unix.RawSockaddrInet6{
 			Len:    unix.SizeofSockaddrInet6,
 			Family: unix.AF_INET6,
 			Addr:   prefixToMask(network).As16(),
-	}
+		},
-
+		Lifetime: addrLifetime{
-	// Set lifetime (never expires)
+			// never expires
-	ifr.Lifetime = addrLifetime{
 			Vltime: 0xffffffff,
 			Pltime: 0xffffffff,
+		},
+		Flags: _IN6_IFF_NODAD,
 	}
 
-	// Set flags (no DAD - Duplicate Address Detection)
-	ifr.Flags = _IN6_IFF_NODAD
-
 	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 fmt.Errorf("failed to set tun address: %s", err)
 	}
 
 	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)
+	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)
+		}
+	}
+
+	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 {
@@ -393,7 +377,53 @@ func getLinkAddr(name string) (*netroute.LinkAddr, error) {
 	return nil, nil
 }
 
-func (rm *tun) addRoute(prefix netip.Prefix) error {
+func (t *tun) addRoutes(logErrors bool) error {
+	routes := *t.Routes.Load()
+
+	for _, r := range routes {
+		if len(r.Via) == 0 || !r.Install {
+			// We don't allow route MTUs so only install routes with a via
+			continue
+		}
+
+		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 {
 	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)
@@ -411,13 +441,13 @@ func (rm *tun) addRoute(prefix netip.Prefix) 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: rm.linkAddr,
+			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: rm.linkAddr,
+			unix.RTAX_GATEWAY: gateway,
 		}
 	}
 
@@ -434,7 +464,7 @@ func (rm *tun) addRoute(prefix netip.Prefix) error {
 	return nil
 }
 
-func (rm *tun) delRoute(prefix netip.Prefix) error {
+func delRoute(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)
@@ -451,13 +481,13 @@ func (rm *tun) delRoute(prefix netip.Prefix) 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: rm.linkAddr,
+			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: rm.linkAddr,
+			unix.RTAX_GATEWAY: gateway,
 		}
 	}
 
@@ -465,7 +495,6 @@ func (rm *tun) delRoute(prefix netip.Prefix) 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)
@@ -474,34 +503,52 @@ func (rm *tun) delRoute(prefix netip.Prefix) error {
 	return nil
 }
 
-func ioctl(a1, a2, a3 uintptr) error {
+func (t *tun) Read(to []byte) (int, error) {
-	_, _, errno := unix.Syscall(unix.SYS_IOCTL, a1, a2, a3)
+	buf := make([]byte, len(to)+4)
-	if errno != 0 {
+
-		return errno
+	n, err := t.ReadWriteCloser.Read(buf)
-	}
+
-	return nil
+	copy(to, buf[4:])
+	return n - 4, err
 }
 
-func prefixToMask(prefix netip.Prefix) netip.Addr {
+// Write is only valid for single threaded use
-	bits := prefix.Bits()
+func (t *tun) Write(from []byte) (int, error) {
-	if prefix.Addr().Is4() {
+	buf := t.out
-		// Create IPv4 netmask from prefix length
+	if cap(buf) < len(from)+4 {
-		mask := ^uint32(0) << (32 - bits)
+		buf = make([]byte, len(from)+4)
-		return netip.AddrFrom4([4]byte{
+		t.out = buf
-			byte(mask >> 24),
+	}
-			byte(mask >> 16),
+	buf = buf[:len(from)+4]
-			byte(mask >> 8),
+
-			byte(mask),
+	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 {
-		// Create IPv6 netmask from prefix length
+		return 0, fmt.Errorf("unable to determine IP version from packet")
-		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)
+	copy(buf[4:], from)
+
+	n, err := t.ReadWriteCloser.Write(buf)
+	return n - 4, err
 }
-		return netip.AddrFrom16(mask)
+
+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

@@ -1,77 +1,284 @@
-//go:build freebsd && !e2e_testing
+//go:build !e2e_testing
-// +build freebsd,!e2e_testing
+// +build !e2e_testing
 
 package overlay
 
 import (
+	"bytes"
+	"errors"
 	"fmt"
 	"io"
+	"io/fs"
 	"net/netip"
-	"os/exec"
+	"sync/atomic"
-	"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"
 )
 
-type tun struct{}
+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
+}
 
-// ifreqRename is used for renaming network interfaces on FreeBSD
 type ifreqRename struct {
 	Name [unix.IFNAMSIZ]byte
 	Data uintptr
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
+type ifreqDestroy struct {
-	return nil, fmt.Errorf("newTunFromFd not supported on FreeBSD")
+	Name [unix.IFNAMSIZ]byte
+	pad  [16]byte
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+type ifReq struct {
-	deviceName := c.GetString("tun.dev", "tun")
+	Name  [unix.IFNAMSIZ]byte
-	mtu := c.GetInt("tun.mtu", DefaultMTU)
+	Flags uint16
-
-	// Create WireGuard TUN device
-	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
-	if err != nil {
-		return nil, fmt.Errorf("failed to create TUN device: %w", err)
 }
 
-	// Get the actual device name
+type ifreqMTU struct {
-	actualName, err := tunDevice.Name()
+	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 {
-		tunDevice.Close()
+			t.l.WithError(err).Error("Error closing device")
-		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
+		}
+		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) (*tun, error) {
+	return nil, fmt.Errorf("newTunFromFd not supported in FreeBSD")
+}
+
+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)
+	}
+	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)))
+	}
+
+	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
 	}
 
 	// If the name doesn't match the desired interface name, rename it now
-	if actualName != deviceName && deviceName != "" && deviceName != "tun" {
+	if ifName != deviceName {
-		if err := renameInterface(actualName, deviceName); err != nil {
+		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
-			tunDevice.Close()
+		if err != nil {
-			return nil, fmt.Errorf("failed to rename interface from %s to %s: %w", actualName, deviceName, err)
+			return nil, err
 		}
-		actualName = deviceName
+		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)),
 		}
 
-	t := &wgTun{
+		// Set the device name
-		tunDevice:   tunDevice,
+		ioctl(fd, syscall.SIOCSIFNAME, uintptr(unsafe.Pointer(&ifrr)))
+	}
+
+	t := &tun{
+		Device:      deviceName,
 		vpnNetworks: vpnNetworks,
-		MaxMTU:      mtu,
+		MTU:         c.GetInt("tun.mtu", DefaultMTU),
-		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
 	}
 
@@ -82,86 +289,180 @@ 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 (rm *tun) Activate(t *wgTun) error {
+func (t *tun) addIp(cidr netip.Prefix) error {
-	name, err := t.tunDevice.Name()
+	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 fmt.Errorf("failed to get device name: %w", err)
+			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
 	}
 
-	// Set the MTU
+	if cidr.Addr().Is6() {
-	rm.SetMTU(t, t.MaxMTU)
+		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)
 
-	// Add IP addresses
+		if err := ioctl(uintptr(s), OSIOCAIFADDR_IN6, uintptr(unsafe.Pointer(&ifr))); err != nil {
-	for _, network := range t.vpnNetworks {
+			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr().String(), err)
-		if err := rm.addIP(t, name, network); err != nil {
+		}
+		return nil
+	}
+
+	return fmt.Errorf("unknown address type %v", cidr)
+}
+
+func (t *tun) Activate() error {
+	// Setup our default MTU
+	err := t.setMTU()
+	if err != nil {
+		return 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
+
+	for i := range t.vpnNetworks {
+		err := t.addIp(t.vpnNetworks[i])
+		if err != nil {
 			return err
 		}
 	}
 
-	// Bring up the interface
+	return t.addRoutes(false)
-	if err := runCommandBSD("ifconfig", name, "up"); err != nil {
-		return fmt.Errorf("failed to bring up interface: %w", err)
 }
 
-	// Set the routes
+func (t *tun) setMTU() error {
-	if err := rm.AddRoutes(t, false); err != nil {
+	// Set the MTU on the device
+	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+	if 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
+}
+
+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)
+	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)
+		}
+	}
 
 	return nil
 }
 
-func (rm *tun) SetMTU(t *wgTun, mtu int) {
+func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
-	name, err := t.tunDevice.Name()
+	r, _ := t.routeTree.Load().Lookup(ip)
-	if err != nil {
+	return r
-		t.l.WithError(err).Error("Failed to get device name for MTU set")
-		return
 }
 
-	if err := runCommandBSD("ifconfig", name, "mtu", strconv.Itoa(mtu)); err != nil {
+func (t *tun) Networks() []netip.Prefix {
-		t.l.WithError(err).Error("Failed to set tun mtu")
+	return t.vpnNetworks
-	}
 }
 
-func (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
+func (t *tun) Name() string {
-	// On FreeBSD, routes are set via ifconfig and route commands
+	return t.Device
-	return nil
 }
 
-func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
-	name, err := t.tunDevice.Name()
+	return nil, fmt.Errorf("TODO: multiqueue not implemented for freebsd")
-	if err != nil {
-		return fmt.Errorf("failed to get device name: %w", err)
 }
 
+func (t *tun) addRoutes(logErrors bool) error {
 	routes := *t.Routes.Load()
 	for _, r := range routes {
-		if !r.Install {
+		if len(r.Via) == 0 || !r.Install {
+			// We don't allow route MTUs so only install routes with a via
 			continue
 		}
 
-		// Add route using route command
+		err := addRoute(r.Cidr, t.linkAddr)
-		args := []string{"add"}
-
-		if r.Cidr.Addr().Is6() {
-			args = append(args, "-inet6")
-		} else {
-			args = append(args, "-inet")
-		}
-
-		args = append(args, r.Cidr.String(), "-interface", name)
-
-		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 := runCommandBSD("route", args...)
 		if err != nil {
 			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
@@ -177,99 +478,142 @@ func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
 	return nil
 }
 
-func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
+func (t *tun) removeRoutes(routes []Route) error {
-	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
 		}
 
-		args := []string{"delete"}
+		err := delRoute(r.Cidr, t.linkAddr)
-
-		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 (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+func (t *tun) deviceBytes() (o [16]byte) {
-	// FreeBSD doesn't support multi-queue TUN devices in the same way as Linux
+	for i, c := range t.Device {
-	// Return a reader that wraps the same device
+		o[i] = byte(c)
-	return &wgTunReader{
+	}
-		parent:    t,
+	return
-		tunDevice: t.tunDevice,
-		offset:    0,
-		l:         t.l,
-	}, nil
 }
 
-func (rm *tun) addIP(t *wgTun, name string, network netip.Prefix) error {
+func addRoute(prefix netip.Prefix, gateway netroute.Addr) error {
-	addr := network.Addr()
+	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)
 
-	if addr.Is4() {
+	route := &netroute.RouteMessage{
-		// For IPv4: ifconfig tun0 10.0.0.1/24
+		Version: unix.RTM_VERSION,
-		if err := runCommandBSD("ifconfig", name, network.String()); err != nil {
+		Type:    unix.RTM_ADD,
-			return fmt.Errorf("failed to add IPv4 address: %w", err)
+		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,
 		}
 	} else {
-		// For IPv6: ifconfig tun0 inet6 add 2001:db8::1/64
+		route.Addrs = []netroute.Addr{
-		if err := runCommandBSD("ifconfig", name, "inet6", "add", network.String()); err != nil {
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
-			return fmt.Errorf("failed to add IPv6 address: %w", err)
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: gateway,
 		}
 	}
 
-	return nil
+	data, err := route.Marshal()
-}
-
-func runCommandBSD(name string, args ...string) error {
-	cmd := exec.Command(name, args...)
-	output, err := cmd.CombinedOutput()
 	if err != nil {
-		return fmt.Errorf("%s %s failed: %w\nOutput: %s", name, strings.Join(args, " "), err, string(output))
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
-	}
-	return nil
 	}
 
-func renameInterface(fromName, toName string) error {
+	_, err = unix.Write(sock, data[:])
-	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
 	if err != nil {
-		return fmt.Errorf("failed to create socket: %w", err)
+		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)
 			}
-	defer syscall.Close(s)
+			_, err = unix.Write(sock, data[:])
-
+			fmt.Println("DOING CHANGE")
-	fd := uintptr(s)
+			return err
-
-	var fromNameBytes [unix.IFNAMSIZ]byte
-	var toNameBytes [unix.IFNAMSIZ]byte
-	copy(fromNameBytes[:], fromName)
-	copy(toNameBytes[:], toName)
-
-	ifrr := ifreqRename{
-		Name: fromNameBytes,
-		Data: uintptr(unsafe.Pointer(&toNameBytes)),
 		}
-
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
-	// 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
 }
+
+func delRoute(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)
+
+	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,
+		}
+	}
+
+	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
+}
+
+// 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
+	}
+
+	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
+				}
+			}
+		}
+	}
+
+	return nil, nil
+}

overlay/tun_linux_test.go

@@ -7,26 +7,25 @@ import "testing"
 
 var runAdvMSSTests = []struct {
 	name     string
-	defaultMTU int
+	tun      *tun
-	maxMTU     int
 	r        Route
 	expected int
 }{
 	// Standard case, default MTU is the device max MTU
-	{"default", 1440, 1440, Route{}, 0},
+	{"default", &tun{DefaultMTU: 1440, MaxMTU: 1440}, Route{}, 0},
-	{"default-min", 1440, 1440, Route{MTU: 1440}, 0},
+	{"default-min", &tun{DefaultMTU: 1440, MaxMTU: 1440}, Route{MTU: 1440}, 0},
-	{"default-low", 1440, 1440, Route{MTU: 1200}, 1160},
+	{"default-low", &tun{DefaultMTU: 1440, MaxMTU: 1440}, Route{MTU: 1200}, 1160},
 
 	// Case where we have a route MTU set higher than the default
-	{"route", 1440, 8941, Route{}, 1400},
+	{"route", &tun{DefaultMTU: 1440, MaxMTU: 8941}, Route{}, 1400},
-	{"route-min", 1440, 8941, Route{MTU: 1440}, 1400},
+	{"route-min", &tun{DefaultMTU: 1440, MaxMTU: 8941}, Route{MTU: 1440}, 1400},
-	{"route-high", 1440, 8941, Route{MTU: 8941}, 0},
+	{"route-high", &tun{DefaultMTU: 1440, MaxMTU: 8941}, Route{MTU: 8941}, 0},
 }
 
 func TestTunAdvMSS(t *testing.T) {
 	for _, tt := range runAdvMSSTests {
 		t.Run(tt.name, func(t *testing.T) {
-			o := advMSS(tt.r, tt.defaultMTU, tt.maxMTU)
+			o := tt.tun.advMSS(tt.r)
 			if o != tt.expected {
 				t.Errorf("got %d, want %d", o, tt.expected)
 			}

overlay/tun_netbsd.go

@@ -547,41 +547,3 @@ func delRoute(prefix netip.Prefix, gateways []netip.Prefix) error {
 
 	return nil
 }
-
-func ioctl(a1, a2, a3 uintptr) error {
-	_, _, errno := syscall.Syscall(syscall.SYS_IOCTL, a1, a2, a3)
-	if errno != 0 {
-		return errno
-	}
-	return nil
-}
-
-func prefixToMask(prefix netip.Prefix) netip.Addr {
-	bits := prefix.Bits()
-	if prefix.Addr().Is4() {
-		mask := ^uint32(0) << (32 - bits)
-		return netip.AddrFrom4([4]byte{
-			byte(mask >> 24),
-			byte(mask >> 16),
-			byte(mask >> 8),
-			byte(mask),
-		})
-	}
-	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)
-}
-
-func selectGateway(prefix netip.Prefix, gateways []netip.Prefix) (netip.Prefix, error) {
-	for _, gw := range gateways {
-		if prefix.Addr().Is4() == gw.Addr().Is4() {
-			return gw, nil
-		}
-	}
-	return netip.Prefix{}, fmt.Errorf("no suitable gateway found for prefix %v", prefix)
-}

overlay/tun_notwin.go

@@ -0,0 +1,14 @@
+//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

@@ -1,59 +1,104 @@
-//go:build openbsd && !e2e_testing
+//go:build !e2e_testing
-// +build openbsd,!e2e_testing
+// +build !e2e_testing
 
 package overlay
 
 import (
+	"errors"
 	"fmt"
 	"io"
 	"net/netip"
-	"os/exec"
+	"os"
-	"strconv"
+	"regexp"
-	"strings"
+	"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"
-	wgtun "golang.zx2c4.com/wireguard/tun"
+	netroute "golang.org/x/net/route"
+	"golang.org/x/sys/unix"
 )
 
-type tun struct{}
+const (
+	SIOCAIFADDR_IN6 = 0x8080691a
+)
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
+type ifreqAlias4 struct {
-	return nil, fmt.Errorf("newTunFromFd not supported on OpenBSD")
+	Name     [unix.IFNAMSIZ]byte
+	Addr     unix.RawSockaddrInet4
+	DstAddr  unix.RawSockaddrInet4
+	MaskAddr unix.RawSockaddrInet4
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+type ifreqAlias6 struct {
-	deviceName := c.GetString("tun.dev", "tun")
+	Name       [unix.IFNAMSIZ]byte
-	mtu := c.GetInt("tun.mtu", DefaultMTU)
+	Addr       unix.RawSockaddrInet6
+	DstAddr    unix.RawSockaddrInet6
+	PrefixMask unix.RawSockaddrInet6
+	Flags      uint32
+	Lifetime   [2]uint32
+}
 
-	// Create WireGuard TUN device
+type ifreq struct {
-	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
+	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
+}
+
+var deviceNameRE = regexp.MustCompile(`^tun[0-9]+$`)
+
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
+	return nil, fmt.Errorf("newTunFromFd not supported in openbsd")
+}
+
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
+	// Try to open tun device
+	var err error
+	deviceName := c.GetString("tun.dev", "")
+	if deviceName == "" {
+		return nil, fmt.Errorf("a device name in the format of /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")
+	}
+
+	fd, err := unix.Open("/dev/"+deviceName, os.O_RDWR, 0)
 	if err != nil {
-		return nil, fmt.Errorf("failed to create TUN device: %w", err)
+		return nil, err
 	}
 
-	// Get the actual device name
+	err = unix.SetNonblock(fd, true)
-	actualName, err := tunDevice.Name()
 	if err != nil {
-		tunDevice.Close()
+		l.WithError(err).Warn("Failed to set the tun device as nonblocking")
-		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
-	t := &wgTun{
+	t := &tun{
-		tunDevice:   tunDevice,
+		f:           os.NewFile(uintptr(fd), ""),
+		fd:          fd,
+		Device:      deviceName,
 		vpnNetworks: vpnNetworks,
-		MaxMTU:      mtu,
+		MTU:         c.GetInt("tun.mtu", DefaultMTU),
-		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
 	}
 
@@ -64,86 +109,221 @@ 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 (rm *tun) Activate(t *wgTun) error {
+func (t *tun) Close() error {
-	name, err := t.tunDevice.Name()
+	if t.f != nil {
+		if err := t.f.Close(); err != nil {
+			return fmt.Errorf("error closing tun file: %w", err)
+		}
+
+		// t.f.Close should have handled it for us but let's be extra sure
+		_ = unix.Close(t.fd)
+	}
+	return nil
+}
+
+func (t *tun) Read(to []byte) (int, error) {
+	buf := make([]byte, len(to)+4)
+
+	n, err := t.f.Read(buf)
+
+	copy(to, buf[4:])
+	return n - 4, err
+}
+
+// Write is only valid for single threaded use
+func (t *tun) Write(from []byte) (int, error) {
+	buf := t.out
+	if cap(buf) < len(from)+4 {
+		buf = make([]byte, len(from)+4)
+		t.out = buf
+	}
+	buf = buf[:len(from)+4]
+
+	if len(from) == 0 {
+		return 0, syscall.EIO
+	}
+
+	// Determine the IP Family for the NULL L2 Header
+	ipVer := from[0] >> 4
+	if ipVer == 4 {
+		buf[3] = syscall.AF_INET
+	} else if ipVer == 6 {
+		buf[3] = syscall.AF_INET6
+	} else {
+		return 0, fmt.Errorf("unable to determine IP version from packet")
+	}
+
+	copy(buf[4:], from)
+
+	n, err := t.f.Write(buf)
+	return n - 4, err
+}
+
+func (t *tun) addIp(cidr netip.Prefix) error {
+	if cidr.Addr().Is4() {
+		var req ifreqAlias4
+		req.Name = t.deviceBytes()
+		req.Addr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   cidr.Addr().As4(),
+		}
+		req.DstAddr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   cidr.Addr().As4(),
+		}
+		req.MaskAddr = unix.RawSockaddrInet4{
+			Len:    unix.SizeofSockaddrInet4,
+			Family: unix.AF_INET,
+			Addr:   prefixToMask(cidr).As4(),
+		}
+
+		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
 		if err != nil {
-		return fmt.Errorf("failed to get device name: %w", err)
-	}
-
-	// Set the MTU
-	rm.SetMTU(t, t.MaxMTU)
-
-	// 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)
 		}
 
-	// Bring up the interface
+		err = addRoute(cidr, t.vpnNetworks)
-	if err := runCommandBSD("ifconfig", name, "up"); err != nil {
+		if err != nil {
-		return fmt.Errorf("failed to bring up interface: %w", err)
+			return fmt.Errorf("failed to set route for vpn network %v: %w", cidr, err)
-	}
-
-	// Set the routes
-	if err := rm.AddRoutes(t, false); err != nil {
-		return err
 		}
 
 		return nil
 	}
 
-func (rm *tun) SetMTU(t *wgTun, mtu int) {
+	if cidr.Addr().Is6() {
-	name, err := t.tunDevice.Name()
+		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 {
-		t.l.WithError(err).Error("Failed to get device name for MTU set")
+			return err
-		return
+		}
+		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)
 		}
 
-	if err := runCommandBSD("ifconfig", name, "mtu", strconv.Itoa(mtu)); err != nil {
-		t.l.WithError(err).Error("Failed to set tun mtu")
-	}
-}
-
-func (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
-	// On OpenBSD, routes are set via ifconfig and route commands
 		return nil
 	}
 
-func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
+	return fmt.Errorf("unknown address type %v", cidr)
-	name, err := t.tunDevice.Name()
-	if err != nil {
-		return fmt.Errorf("failed to get device name: %w", err)
 }
 
+func (t *tun) Activate() error {
+	err := t.doIoctlByName(unix.SIOCSIFMTU, uint32(t.MTU))
+	if err != nil {
+		return fmt.Errorf("failed to set tun mtu: %w", err)
+	}
+
+	for i := range t.vpnNetworks {
+		err = t.addIp(t.vpnNetworks[i])
+		if err != nil {
+			return err
+		}
+	}
+
+	return t.addRoutes(false)
+}
+
+func (t *tun) doIoctlByName(ctl uintptr, value uint32) error {
+	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(s)
+
+	ir := ifreq{Name: t.deviceBytes(), data: int(value)}
+	err = ioctl(uintptr(s), ctl, uintptr(unsafe.Pointer(&ir)))
+	return err
+}
+
+func (t *tun) reload(c *config.C, initial bool) error {
+	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
+	if err != nil {
+		return 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)
+		}
+	}
+
+	return nil
+}
+
+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")
+}
+
+func (t *tun) addRoutes(logErrors bool) error {
 	routes := *t.Routes.Load()
+
 	for _, r := range routes {
-		if !r.Install {
+		if len(r.Via) == 0 || !r.Install {
+			// We don't allow route MTUs so only install routes with a via
 			continue
 		}
 
-		// Add route using route command
+		err := addRoute(r.Cidr, t.vpnNetworks)
-		args := []string{"add"}
-
-		if r.Cidr.Addr().Is6() {
-			args = append(args, "-inet6")
-		} else {
-			args = append(args, "-inet")
-		}
-
-		args = append(args, r.Cidr.String(), "-interface", name)
-
-		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 := runCommandBSD("route", args...)
 		if err != nil {
 			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
@@ -159,71 +339,131 @@ func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
 	return nil
 }
 
-func (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
+func (t *tun) removeRoutes(routes []Route) error {
-	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
 		}
 
-		args := []string{"delete"}
+		err := delRoute(r.Cidr, t.vpnNetworks)
-
-		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 (rm *tun) NewMultiQueueReader(t *wgTun) (io.ReadWriteCloser, error) {
+func (t *tun) deviceBytes() (o [16]byte) {
-	// OpenBSD doesn't support multi-queue TUN devices in the same way as Linux
+	for i, c := range t.Device {
-	// Return a reader that wraps the same device
+		o[i] = byte(c)
-	return &wgTunReader{
+	}
-		parent:    t,
+	return
-		tunDevice: t.tunDevice,
-		offset:    0,
-		l:         t.l,
-	}, nil
 }
 
-func (rm *tun) addIP(t *wgTun, name string, network netip.Prefix) error {
+func addRoute(prefix netip.Prefix, gateways []netip.Prefix) error {
-	addr := network.Addr()
+	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)
 
-	if addr.Is4() {
+	route := &netroute.RouteMessage{
-		// For IPv4: ifconfig tun0 10.0.0.1/24
+		Version: unix.RTM_VERSION,
-		if err := runCommandBSD("ifconfig", name, network.String()); err != nil {
+		Type:    unix.RTM_ADD,
-			return fmt.Errorf("failed to add IPv4 address: %w", err)
+		Flags:   unix.RTF_UP | unix.RTF_GATEWAY,
+		Seq:     1,
+	}
+
+	if prefix.Addr().Is4() {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: &netroute.Inet4Addr{IP: gw.Addr().As4()},
 		}
 	} else {
-		// For IPv6: ifconfig tun0 inet6 add 2001:db8::1/64
+		gw, err := selectGateway(prefix, gateways)
-		if err := runCommandBSD("ifconfig", name, "inet6", "add", network.String()); err != nil {
-			return fmt.Errorf("failed to add IPv6 address: %w", err)
-		}
-	}
-
-	return nil
-}
-
-func runCommandBSD(name string, args ...string) error {
-	cmd := exec.Command(name, args...)
-	output, err := cmd.CombinedOutput()
 		if err != nil {
-		return fmt.Errorf("%s %s failed: %w\nOutput: %s", name, strings.Join(args, " "), err, string(output))
+			return err
 		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: &netroute.Inet6Addr{IP: gw.Addr().As16()},
+		}
+	}
+
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
+
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		if errors.Is(err, unix.EEXIST) {
+			// Try to do a change
+			route.Type = unix.RTM_CHANGE
+			data, err = route.Marshal()
+			if err != nil {
+				return fmt.Errorf("failed to create route.RouteMessage for change: %w", err)
+			}
+			_, err = unix.Write(sock, data[:])
+			return err
+		}
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
+	return nil
+}
+
+func delRoute(prefix netip.Prefix, gateways []netip.Prefix) error {
+	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
+	}
+	defer unix.Close(sock)
+
+	route := netroute.RouteMessage{
+		Version: unix.RTM_VERSION,
+		Type:    unix.RTM_DELETE,
+		Seq:     1,
+	}
+
+	if prefix.Addr().Is4() {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: &netroute.Inet4Addr{IP: gw.Addr().As4()},
+		}
+	} else {
+		gw, err := selectGateway(prefix, gateways)
+		if err != nil {
+			return err
+		}
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: &netroute.Inet6Addr{IP: gw.Addr().As16()},
+		}
+	}
+
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
 	return nil
 }

overlay/tun_wg.go

@@ -1,242 +0,0 @@
-//go:build !android && !netbsd && !e2e_testing
-// +build !android,!netbsd,!e2e_testing
-
-package overlay
-
-import (
-	"fmt"
-	"io"
-	"net/netip"
-	"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(buffers [][]byte, sizes []int) (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
-	offset    int
-	l         *logrus.Logger
-}
-
-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
-// The caller provides buffers and sizes slices, and this function returns the number of packets read.
-func (r *wgTunReader) BatchRead(buffers [][]byte, sizes []int) (int, error) {
-	return r.tunDevice.Read(buffers, sizes, r.offset)
-}
-
-// 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 {
-	if r.tunDevice != nil {
-		return r.tunDevice.Close()
-	}
-	return nil
-}

overlay/tun_windows.go

@@ -1,77 +1,84 @@
-//go:build windows && !e2e_testing
+//go:build !e2e_testing
-// +build windows,!e2e_testing
+// +build !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 tun struct {
+type winTun struct {
-	luid winipcfg.LUID
+	Device      string
+	vpnNetworks []netip.Prefix
+	MTU         int
+	Routes      atomic.Pointer[[]Route]
+	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	l           *logrus.Logger
+
+	tun *wintun.NativeTun
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*wgTun, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (Device, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in Windows")
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*wgTun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*winTun, error) {
-	deviceName := c.GetString("tun.dev", "Nebula")
+	err := checkWinTunExists()
-	mtu := c.GetInt("tun.mtu", DefaultMTU)
-
-	// Create WireGuard TUN device
-	tunDevice, err := wgtun.CreateTUN(deviceName, mtu)
 	if err != nil {
-		return nil, fmt.Errorf("failed to create TUN device: %w", err)
+		return nil, fmt.Errorf("can not load the wintun driver: %w", err)
 	}
 
-	// Get the actual device name
+	deviceName := c.GetString("tun.dev", "")
-	actualName, err := tunDevice.Name()
+	guid, err := generateGUIDByDeviceName(deviceName)
 	if err != nil {
-		tunDevice.Close()
+		return nil, fmt.Errorf("generate GUID failed: %w", err)
-		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
 	}
 
-	t := &wgTun{
+	t := &winTun{
-		tunDevice:   tunDevice,
+		Device:      deviceName,
 		vpnNetworks: vpnNetworks,
-		MaxMTU:      mtu,
+		MTU:         c.GetInt("tun.mtu", DefaultMTU),
-		DefaultMTU:  mtu,
 		l:           l,
 	}
 
-	// Create Windows-specific route manager
-	rm := &tun{}
-
-	// Get LUID from the TUN device
-	// The WireGuard TUN device on Windows should provide a LUID() method
-	if nativeTun, ok := tunDevice.(interface{ LUID() uint64 }); ok {
-		rm.luid = winipcfg.LUID(nativeTun.LUID())
-	} else {
-		tunDevice.Close()
-		return nil, fmt.Errorf("failed to get LUID from TUN device")
-	}
-	t.routeManager = rm
-
 	err = t.reload(c, true)
 	if err != nil {
-		tunDevice.Close()
 		return nil, err
 	}
 
+	var tunDevice wintun.Device
+	tunDevice, err = wintun.CreateTUNWithRequestedGUID(deviceName, guid, t.MTU)
+	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)
+		}
+	}
+	t.tun = tunDevice.(*wintun.NativeTun)
+
 	c.RegisterReloadCallback(func(c *config.C) {
 		err := t.reload(c, false)
 		if err != nil {
@@ -79,140 +86,206 @@ 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 (rm *tun) Activate(t *wgTun) error {
+func (t *winTun) reload(c *config.C, initial bool) error {
-	// Set MTU
+	change, routes, err := getAllRoutesFromConfig(c, t.vpnNetworks, initial)
-	err := rm.setMTU(t, t.MaxMTU)
 	if err != nil {
-		return fmt.Errorf("failed to set MTU: %w", err)
-	}
-
-	// Add IP addresses
-	for _, network := range t.vpnNetworks {
-		if err := rm.addIP(t, network); err != nil {
 		return err
 	}
+
+	if !initial && !change {
+		return nil
 	}
 
-	// Add routes
+	routeTree, err := makeRouteTree(t.l, routes, false)
-	if err := rm.AddRoutes(t, false); err != nil {
+	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)
+		}
+	}
+
+	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)
+	}
+
+	err = t.addRoutes(false)
+	if err != nil {
 		return err
 	}
 
 	return nil
 }
 
-func (rm *tun) SetMTU(t *wgTun, mtu int) {
+func (t *winTun) addRoutes(logErrors bool) error {
-	if err := rm.setMTU(t, mtu); err != nil {
+	luid := winipcfg.LUID(t.tun.LUID())
-		t.l.WithError(err).Error("Failed to set MTU")
-	}
-}
-
-func (rm *tun) setMTU(t *wgTun, mtu int) error {
-	// Set MTU using winipcfg
-	// Note: MTU setting on Windows TUN devices may be handled by the driver
-	// For now, we'll skip explicit MTU setting as the WireGuard TUN handles it
-	return nil
-}
-
-func (rm *tun) SetDefaultRoute(t *wgTun, cidr netip.Prefix) error {
-	// On Windows, routes are managed differently
-	return nil
-}
-
-func (rm *tun) AddRoutes(t *wgTun, logErrors bool) error {
 	routes := *t.Routes.Load()
+	foundDefault4 := false
+
 	for _, r := range routes {
-		if !r.Install {
+		if len(r.Via) == 0 || !r.Install {
+			// We don't allow route MTUs so only install routes with a via
 			continue
 		}
 
-		if r.MTU > 0 {
+		// Add our unsafe route
-			// Windows route MTU is not directly supported
+		// Windows does not support multipath routes natively, so we install only a single route.
-			t.l.WithField("route", r).Debug("Route MTU is not supported on Windows")
+		// 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))
-		// Use winipcfg to add the route
-		// The rm.luid should have the AddRoute method from winipcfg
-		if len(r.Via) == 0 {
-			t.l.WithField("route", r).Warn("Route has no via address, skipping")
-			continue
-		}
-
-		err := rm.luid.AddRoute(r.Cidr, r.Via[0].Addr(), uint32(r.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 (rm *tun) RemoveRoutes(t *wgTun, routes []Route) {
+func (t *winTun) removeRoutes(routes []Route) error {
+	luid := winipcfg.LUID(t.tun.LUID())
+
 	for _, r := range routes {
 		if !r.Install {
 			continue
 		}
 
-		if len(r.Via) == 0 {
+		// See comment on luid.AddRoute
-			continue
+		err := luid.DeleteRoute(r.Cidr, r.Via[0].Addr())
-		}
-
-		err := rm.luid.DeleteRoute(r.Cidr, r.Via[0].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")
 		}
 	}
-}
-
-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,
-		offset:    0,
-		l:         t.l,
-	}, nil
-}
-
-func (rm *tun) addIP(t *wgTun, network netip.Prefix) error {
-	// Add IP address using winipcfg
-	// SetIPAddresses expects a slice of prefixes
-	err := rm.luid.SetIPAddresses([]netip.Prefix{network})
-	if err != nil {
-		return fmt.Errorf("failed to add IP address %s: %w", network, err)
-	}
 	return nil
 }
 
-// generateGUIDByDeviceName generates a GUID based on the device name
+func (t *winTun) RoutesFor(ip netip.Addr) routing.Gateways {
-func generateGUIDByDeviceName(deviceName string) (*windows.GUID, error) {
+	r, _ := t.routeTree.Load().Lookup(ip)
-	// Hash the device name to create a deterministic GUID
+	return r
-	h := crypto.SHA256.New()
-	h.Write([]byte(tunGUIDLabel))
-	h.Write([]byte(deviceName))
-	sum := h.Sum(nil)
-
-	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])
 
-	return guid, nil
+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 (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))
+	if err != nil {
+		return nil, err
+	}
+
+	sum := hash.Sum(nil)
+
+	return (*windows.GUID)(unsafe.Pointer(&sum[0])), nil
+}
+
+func checkWinTunExists() error {
+	myPath, err := os.Executable()
+	if err != nil {
+		return err
+	}
+
+	arch := runtime.GOARCH
+	switch arch {
+	case "386":
+		//NOTE: wintun bundles 386 as x86
+		arch = "x86"
+	}
+
+	_, err = syscall.LoadDLL(filepath.Join(filepath.Dir(myPath), "dist", "windows", "wintun", "bin", arch, "wintun.dll"))
+	return err
 }

udp/config.go

@@ -0,0 +1,16 @@
+package udp
+
+import "sync/atomic"
+
+var disableUDPCsum atomic.Bool
+
+// SetDisableUDPCsum controls whether IPv4 UDP sockets opt out of kernel
+// checksum calculation via SO_NO_CHECK. Only applicable on platforms that
+// support the option (Linux). IPv6 always keeps the checksum enabled.
+func SetDisableUDPCsum(disable bool) {
+	disableUDPCsum.Store(disable)
+}
+
+func udpChecksumDisabled() bool {
+	return disableUDPCsum.Load()
+}

udp/conn.go

@@ -11,6 +11,7 @@ const MTU = 9001
 type EncReader func(
 	addr netip.AddrPort,
 	payload []byte,
+	release func(),
 )
 
 type Conn interface {

udp/msghdr_helper_linux_32.go

@@ -0,0 +1,25 @@
+//go:build linux && (386 || amd64p32 || arm || mips || mipsle) && !android && !e2e_testing
+// +build linux
+// +build 386 amd64p32 arm mips mipsle
+// +build !android
+// +build !e2e_testing
+
+package udp
+
+import "golang.org/x/sys/unix"
+
+func controllen(n int) uint32 {
+	return uint32(n)
+}
+
+func setCmsgLen(h *unix.Cmsghdr, n int) {
+	h.Len = uint32(unix.CmsgLen(n))
+}
+
+func setIovecLen(v *unix.Iovec, n int) {
+	v.Len = uint32(n)
+}
+
+func setMsghdrIovlen(m *unix.Msghdr, n int) {
+	m.Iovlen = uint32(n)
+}

udp/msghdr_helper_linux_64.go

@@ -0,0 +1,25 @@
+//go:build linux && (amd64 || arm64 || ppc64 || ppc64le || mips64 || mips64le || s390x || riscv64 || loong64) && !android && !e2e_testing
+// +build linux
+// +build amd64 arm64 ppc64 ppc64le mips64 mips64le s390x riscv64 loong64
+// +build !android
+// +build !e2e_testing
+
+package udp
+
+import "golang.org/x/sys/unix"
+
+func controllen(n int) uint64 {
+	return uint64(n)
+}
+
+func setCmsgLen(h *unix.Cmsghdr, n int) {
+	h.Len = uint64(unix.CmsgLen(n))
+}
+
+func setIovecLen(v *unix.Iovec, n int) {
+	v.Len = uint64(n)
+}
+
+func setMsghdrIovlen(m *unix.Msghdr, n int) {
+	m.Iovlen = uint64(n)
+}

udp/sendmmsg_linux_32.go

@@ -0,0 +1,25 @@
+//go:build linux && (386 || amd64p32 || arm || mips || mipsle) && !android && !e2e_testing
+
+package udp
+
+import (
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+type linuxMmsgHdr struct {
+	Hdr unix.Msghdr
+	Len uint32
+}
+
+func sendmmsg(fd int, hdrs []linuxMmsgHdr, flags int) (int, error) {
+	if len(hdrs) == 0 {
+		return 0, nil
+	}
+	n, _, errno := unix.Syscall6(unix.SYS_SENDMMSG, uintptr(fd), uintptr(unsafe.Pointer(&hdrs[0])), uintptr(len(hdrs)), uintptr(flags), 0, 0)
+	if errno != 0 {
+		return int(n), errno
+	}
+	return int(n), nil
+}

udp/sendmmsg_linux_64.go

@@ -0,0 +1,26 @@
+//go:build linux && (amd64 || arm64 || ppc64 || ppc64le || mips64 || mips64le || s390x || riscv64 || loong64) && !android && !e2e_testing
+
+package udp
+
+import (
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+type linuxMmsgHdr struct {
+	Hdr unix.Msghdr
+	Len uint32
+	_   uint32
+}
+
+func sendmmsg(fd int, hdrs []linuxMmsgHdr, flags int) (int, error) {
+	if len(hdrs) == 0 {
+		return 0, nil
+	}
+	n, _, errno := unix.Syscall6(unix.SYS_SENDMMSG, uintptr(fd), uintptr(unsafe.Pointer(&hdrs[0])), uintptr(len(hdrs)), uintptr(flags), 0, 0)
+	if errno != 0 {
+		return int(n), errno
+	}
+	return int(n), nil
+}

udp/udp_darwin.go

@@ -180,7 +180,7 @@ func (u *StdConn) ListenOut(r EncReader) {
 			u.l.WithError(err).Error("unexpected udp socket receive error")
 		}
 
-		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n])
+		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n], nil)
 	}
 }
 

udp/udp_generic.go

@@ -82,6 +82,6 @@ func (u *GenericConn) ListenOut(r EncReader) {
 			return
 		}
 
-		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n])
+		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n], nil)
 	}
 }

udp/udp_linux_32.go

@@ -30,17 +30,29 @@ type rawMessage struct {
 	Len uint32
 }
 
-func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
+func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte, [][]byte) {
+	controlLen := int(u.controlLen.Load())
+
 	msgs := make([]rawMessage, n)
 	buffers := make([][]byte, n)
 	names := make([][]byte, n)
 
+	var controls [][]byte
+	if controlLen > 0 {
+		controls = make([][]byte, n)
+	}
+
 	for i := range msgs {
-		buffers[i] = make([]byte, MTU)
+		size := int(u.groBufSize.Load())
+		if size < MTU {
+			size = MTU
+		}
+		buf := u.borrowRxBuffer(size)
+		buffers[i] = buf
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
-			{Base: &buffers[i][0], Len: uint32(len(buffers[i]))},
+			{Base: &buf[0], Len: uint32(len(buf))},
 		}
 
 		msgs[i].Hdr.Iov = &vs[0]
@@ -48,7 +60,22 @@ func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 
 		msgs[i].Hdr.Name = &names[i][0]
 		msgs[i].Hdr.Namelen = uint32(len(names[i]))
+
+		if controlLen > 0 {
+			controls[i] = make([]byte, controlLen)
+			msgs[i].Hdr.Control = &controls[i][0]
+			msgs[i].Hdr.Controllen = controllen(len(controls[i]))
+		} else {
+			msgs[i].Hdr.Control = nil
+			msgs[i].Hdr.Controllen = controllen(0)
+		}
 	}
 
-	return msgs, buffers, names
+	return msgs, buffers, names, controls
+}
+
+func setIovecBase(msg *rawMessage, buf []byte) {
+	iov := (*iovec)(msg.Hdr.Iov)
+	iov.Base = &buf[0]
+	iov.Len = uint32(len(buf))
 }

udp/udp_linux_64.go

@@ -33,25 +33,50 @@ type rawMessage struct {
 	Pad0 [4]byte
 }
 
-func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
+func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte, [][]byte) {
+	controlLen := int(u.controlLen.Load())
+
 	msgs := make([]rawMessage, n)
 	buffers := make([][]byte, n)
 	names := make([][]byte, n)
 
+	var controls [][]byte
+	if controlLen > 0 {
+		controls = make([][]byte, n)
+	}
+
 	for i := range msgs {
-		buffers[i] = make([]byte, MTU)
+		size := int(u.groBufSize.Load())
+		if size < MTU {
+			size = MTU
+		}
+		buf := u.borrowRxBuffer(size)
+		buffers[i] = buf
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
-		vs := []iovec{
+		vs := []iovec{{Base: &buf[0], Len: uint64(len(buf))}}
-			{Base: &buffers[i][0], Len: uint64(len(buffers[i]))},
-		}
 
 		msgs[i].Hdr.Iov = &vs[0]
 		msgs[i].Hdr.Iovlen = uint64(len(vs))
 
 		msgs[i].Hdr.Name = &names[i][0]
 		msgs[i].Hdr.Namelen = uint32(len(names[i]))
+
+		if controlLen > 0 {
+			controls[i] = make([]byte, controlLen)
+			msgs[i].Hdr.Control = &controls[i][0]
+			msgs[i].Hdr.Controllen = controllen(len(controls[i]))
+		} else {
+			msgs[i].Hdr.Control = nil
+			msgs[i].Hdr.Controllen = controllen(0)
+		}
 	}
 
-	return msgs, buffers, names
+	return msgs, buffers, names, controls
+}
+
+func setIovecBase(msg *rawMessage, buf []byte) {
+	iov := (*iovec)(msg.Hdr.Iov)
+	iov.Base = &buf[0]
+	iov.Len = uint64(len(buf))
 }

udp/udp_rio_windows.go

@@ -149,7 +149,7 @@ func (u *RIOConn) ListenOut(r EncReader) {
 			continue
 		}
 
-		r(netip.AddrPortFrom(netip.AddrFrom16(rua.Addr).Unmap(), (rua.Port>>8)|((rua.Port&0xff)<<8)), buffer[:n])
+		r(netip.AddrPortFrom(netip.AddrFrom16(rua.Addr).Unmap(), (rua.Port>>8)|((rua.Port&0xff)<<8)), buffer[:n], nil)
 	}
 }
 

udp/udp_tester.go

@@ -112,7 +112,7 @@ func (u *TesterConn) ListenOut(r EncReader) {
 		if !ok {
 			return
 		}
-		r(p.From, p.Data)
+		r(p.From, p.Data, func() {})
 	}
 }