config tweaks for batching

cert/pem.go

@@ -1,8 +1,10 @@
 package cert
 
 import (
+	"encoding/hex"
 	"encoding/pem"
 	"fmt"
+	"time"
 
 	"golang.org/x/crypto/ed25519"
 )
@@ -138,6 +140,101 @@ func MarshalSigningPrivateKeyToPEM(curve Curve, b []byte) []byte {
 	}
 }
 
+// Backward compatibility functions for older API
+func MarshalX25519PublicKey(b []byte) []byte {
+	return MarshalPublicKeyToPEM(Curve_CURVE25519, b)
+}
+
+func MarshalX25519PrivateKey(b []byte) []byte {
+	return MarshalPrivateKeyToPEM(Curve_CURVE25519, b)
+}
+
+func MarshalPublicKey(curve Curve, b []byte) []byte {
+	return MarshalPublicKeyToPEM(curve, b)
+}
+
+func MarshalPrivateKey(curve Curve, b []byte) []byte {
+	return MarshalPrivateKeyToPEM(curve, b)
+}
+
+// NebulaCertificate is a compatibility wrapper for the old API
+type NebulaCertificate struct {
+	Details   NebulaCertificateDetails
+	Signature []byte
+	cert      Certificate
+}
+
+// NebulaCertificateDetails is a compatibility wrapper for certificate details
+type NebulaCertificateDetails struct {
+	Name      string
+	NotBefore time.Time
+	NotAfter  time.Time
+	PublicKey []byte
+	IsCA      bool
+	Issuer    []byte
+	Curve     Curve
+}
+
+// UnmarshalNebulaCertificateFromPEM provides backward compatibility with the old API
+func UnmarshalNebulaCertificateFromPEM(b []byte) (*NebulaCertificate, []byte, error) {
+	c, rest, err := UnmarshalCertificateFromPEM(b)
+	if err != nil {
+		return nil, rest, err
+	}
+
+	issuerBytes, err := func() ([]byte, error) {
+		issuer := c.Issuer()
+		if issuer == "" {
+			return nil, nil
+		}
+		decoded, err := hex.DecodeString(issuer)
+		if err != nil {
+			return nil, fmt.Errorf("failed to decode issuer fingerprint: %w", err)
+		}
+		return decoded, nil
+	}()
+	if err != nil {
+		return nil, rest, err
+	}
+
+	pubKey := c.PublicKey()
+	if pubKey != nil {
+		pubKey = append([]byte(nil), pubKey...)
+	}
+
+	sig := c.Signature()
+	if sig != nil {
+		sig = append([]byte(nil), sig...)
+	}
+
+	return &NebulaCertificate{
+		Details: NebulaCertificateDetails{
+			Name:      c.Name(),
+			NotBefore: c.NotBefore(),
+			NotAfter:  c.NotAfter(),
+			PublicKey: pubKey,
+			IsCA:      c.IsCA(),
+			Issuer:    issuerBytes,
+			Curve:     c.Curve(),
+		},
+		Signature: sig,
+		cert:      c,
+	}, rest, nil
+}
+
+// IssuerString returns the issuer in hex format for compatibility
+func (n *NebulaCertificate) IssuerString() string {
+	if n.Details.Issuer == nil {
+		return ""
+	}
+	return hex.EncodeToString(n.Details.Issuer)
+}
+
+// Certificate returns the underlying certificate (read-only)
+func (n *NebulaCertificate) Certificate() Certificate {
+	return n.cert
+}
+
 // UnmarshalPrivateKeyFromPEM will try to unmarshal the first pem block in a byte array, returning any non
 // consumed data or an error on failure
 func UnmarshalPrivateKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {

examples/config.yml

@@ -132,6 +132,13 @@ listen:
   # Sets the max number of packets to pull from the kernel for each syscall (under systems that support recvmmsg)
   # default is 64, does not support reload
   #batch: 64
+
+# Control batching between UDP and TUN pipelines
+#batch:
+#  inbound_size: 32        # packets to queue from UDP before handing to workers
+#  outbound_size: 32       # packets to queue from TUN before handing to workers
+#  flush_interval: 50us    # flush partially filled batches after this duration
+#  max_outstanding: 1028   # batches buffered per routine on each channel
   # Configure socket buffers for the udp side (outside), leave unset to use the system defaults. Values will be doubled by the kernel
   # Default is net.core.rmem_default and net.core.wmem_default (/proc/sys/net/core/rmem_default and /proc/sys/net/core/rmem_default)
   # Maximum is limited by memory in the system, SO_RCVBUFFORCE and SO_SNDBUFFORCE is used to avoid having to raise the system wide

interface.go

@@ -22,7 +22,14 @@ import (
 	"github.com/slackhq/nebula/udp"
 )
 
-const mtu = 9001
+const (
+	mtu = 9001
+
+	inboundBatchSizeDefault      = 32
+	outboundBatchSizeDefault     = 32
+	batchFlushIntervalDefault    = 50 * time.Microsecond
+	maxOutstandingBatchesDefault = 1028
+)
 
 type InterfaceConfig struct {
 	HostMap            *HostMap
@@ -48,9 +55,17 @@ type InterfaceConfig struct {
 	reQueryWait     time.Duration
 
 	ConntrackCacheTimeout time.Duration
+	BatchConfig           BatchConfig
 	l                     *logrus.Logger
 }
 
+type BatchConfig struct {
+	InboundBatchSize      int
+	OutboundBatchSize     int
+	FlushInterval         time.Duration
+	MaxOutstandingPerChan int
+}
+
 type Interface struct {
 	hostMap               *HostMap
 	outside               udp.Conn
@@ -97,10 +112,86 @@ type Interface struct {
 	l *logrus.Logger
 
 	inPool  sync.Pool
-	inbound chan *packet.Packet
+	inbound []chan *packetBatch
 
 	outPool  sync.Pool
-	outbound chan *[]byte
+	outbound []chan *outboundBatch
+
+	packetBatchPool   sync.Pool
+	outboundBatchPool sync.Pool
+
+	inboundBatchSize      int
+	outboundBatchSize     int
+	batchFlushInterval    time.Duration
+	maxOutstandingPerChan int
+}
+
+type packetBatch struct {
+	packets []*packet.Packet
+}
+
+func newPacketBatch(capacity int) *packetBatch {
+	return &packetBatch{
+		packets: make([]*packet.Packet, 0, capacity),
+	}
+}
+
+func (b *packetBatch) add(p *packet.Packet) {
+	b.packets = append(b.packets, p)
+}
+
+func (b *packetBatch) reset() {
+	for i := range b.packets {
+		b.packets[i] = nil
+	}
+	b.packets = b.packets[:0]
+}
+
+func (f *Interface) getPacketBatch() *packetBatch {
+	if v := f.packetBatchPool.Get(); v != nil {
+		b := v.(*packetBatch)
+		b.reset()
+		return b
+	}
+	return newPacketBatch(f.inboundBatchSize)
+}
+
+func (f *Interface) releasePacketBatch(b *packetBatch) {
+	b.reset()
+	f.packetBatchPool.Put(b)
+}
+
+type outboundBatch struct {
+	payloads []*[]byte
+}
+
+func newOutboundBatch(capacity int) *outboundBatch {
+	return &outboundBatch{payloads: make([]*[]byte, 0, capacity)}
+}
+
+func (b *outboundBatch) add(buf *[]byte) {
+	b.payloads = append(b.payloads, buf)
+}
+
+func (b *outboundBatch) reset() {
+	for i := range b.payloads {
+		b.payloads[i] = nil
+	}
+	b.payloads = b.payloads[:0]
+}
+
+func (f *Interface) getOutboundBatch() *outboundBatch {
+	if v := f.outboundBatchPool.Get(); v != nil {
+		b := v.(*outboundBatch)
+		b.reset()
+		return b
+	}
+	return newOutboundBatch(f.outboundBatchSize)
+}
+
+func (f *Interface) releaseOutboundBatch(b *outboundBatch) {
+	b.reset()
+	f.outboundBatchPool.Put(b)
 }
 
 type EncWriter interface {
@@ -170,6 +261,20 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 	}
 
 	cs := c.pki.getCertState()
+
+	bc := c.BatchConfig
+	if bc.InboundBatchSize <= 0 {
+		bc.InboundBatchSize = inboundBatchSizeDefault
+	}
+	if bc.OutboundBatchSize <= 0 {
+		bc.OutboundBatchSize = outboundBatchSizeDefault
+	}
+	if bc.FlushInterval <= 0 {
+		bc.FlushInterval = batchFlushIntervalDefault
+	}
+	if bc.MaxOutstandingPerChan <= 0 {
+		bc.MaxOutstandingPerChan = maxOutstandingBatchesDefault
+	}
 	ifce := &Interface{
 		pki:                   c.pki,
 		hostMap:               c.HostMap,
@@ -202,11 +307,20 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 			dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
 		},
 
-		//TODO: configurable size
+		inbound:  make([]chan *packetBatch, c.routines),
-		inbound:  make(chan *packet.Packet, 1028),
+		outbound: make([]chan *outboundBatch, c.routines),
-		outbound: make(chan *[]byte, 1028),
 
 		l: c.l,
+
+		inboundBatchSize:      bc.InboundBatchSize,
+		outboundBatchSize:     bc.OutboundBatchSize,
+		batchFlushInterval:    bc.FlushInterval,
+		maxOutstandingPerChan: bc.MaxOutstandingPerChan,
+	}
+
+	for i := 0; i < c.routines; i++ {
+		ifce.inbound[i] = make(chan *packetBatch, ifce.maxOutstandingPerChan)
+		ifce.outbound[i] = make(chan *outboundBatch, ifce.maxOutstandingPerChan)
 	}
 
 	ifce.inPool = sync.Pool{New: func() any {
@@ -218,6 +332,14 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 		return &t
 	}}
 
+	ifce.packetBatchPool = sync.Pool{New: func() any {
+		return newPacketBatch(ifce.inboundBatchSize)
+	}}
+
+	ifce.outboundBatchPool = sync.Pool{New: func() any {
+		return newOutboundBatch(ifce.outboundBatchSize)
+	}}
+
 	ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
 	ifce.reQueryEvery.Store(c.reQueryEvery)
 	ifce.reQueryWait.Store(int64(c.reQueryWait))
@@ -296,22 +418,41 @@ func (f *Interface) listenOut(i int) {
 		li = f.outside
 	}
 
+	batch := f.getPacketBatch()
+	lastFlush := time.Now()
+
+	flush := func(force bool) {
+		if len(batch.packets) == 0 {
+			if force {
+				f.releasePacketBatch(batch)
+			}
+			return
+		}
+
+		f.inbound[i] <- batch
+		batch = f.getPacketBatch()
+		lastFlush = time.Now()
+	}
+
 	err := li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
 		p := f.inPool.Get().(*packet.Packet)
-		//TODO: have the listener store this in the msgs array after a read instead of doing a copy
-
 		p.Payload = p.Payload[:mtu]
 		copy(p.Payload, payload)
 		p.Payload = p.Payload[:len(payload)]
 		p.Addr = fromUdpAddr
-		f.inbound <- p
+		batch.add(p)
-		//select {
+
-		//case f.inbound <- p:
+		if len(batch.packets) >= f.inboundBatchSize || time.Since(lastFlush) >= f.batchFlushInterval {
-		//default:
+			flush(false)
-		//	f.l.Error("Dropped packet from inbound channel")
+		}
-		//}
 	})
 
+	if len(batch.packets) > 0 {
+		f.inbound[i] <- batch
+	} else {
+		f.releasePacketBatch(batch)
+	}
+
 	if err != nil && !f.closed.Load() {
 		f.l.WithError(err).Error("Error while reading packet inbound packet, closing")
 		//TODO: Trigger Control to close
@@ -324,6 +465,22 @@ func (f *Interface) listenOut(i int) {
 func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	runtime.LockOSThread()
 
+	batch := f.getOutboundBatch()
+	lastFlush := time.Now()
+
+	flush := func(force bool) {
+		if len(batch.payloads) == 0 {
+			if force {
+				f.releaseOutboundBatch(batch)
+			}
+			return
+		}
+
+		f.outbound[i] <- batch
+		batch = f.getOutboundBatch()
+		lastFlush = time.Now()
+	}
+
 	for {
 		p := f.outPool.Get().(*[]byte)
 		*p = (*p)[:mtu]
@@ -337,13 +494,17 @@ func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 		}
 
 		*p = (*p)[:n]
-		//TODO: nonblocking channel write
+		batch.add(p)
-		f.outbound <- p
+
-		//select {
+		if len(batch.payloads) >= f.outboundBatchSize || time.Since(lastFlush) >= f.batchFlushInterval {
-		//case f.outbound <- p:
+			flush(false)
-		//default:
+		}
-		//	f.l.Error("Dropped packet from outbound channel")
+	}
-		//}
+
+	if len(batch.payloads) > 0 {
+		f.outbound[i] <- batch
+	} else {
+		f.releaseOutboundBatch(batch)
 	}
 
 	f.l.Debugf("overlay reader %v is done", i)
@@ -360,10 +521,13 @@ func (f *Interface) workerIn(i int, ctx context.Context) {
 
 	for {
 		select {
-		case p := <-f.inbound:
+		case batch := <-f.inbound[i]:
-			f.readOutsidePackets(p.Addr, nil, result2[:0], p.Payload, h, fwPacket2, lhh, nb2, i, conntrackCache.Get(f.l))
+			for _, p := range batch.packets {
-			p.Payload = p.Payload[:mtu]
+				f.readOutsidePackets(p.Addr, nil, result2[:0], p.Payload, h, fwPacket2, lhh, nb2, i, conntrackCache.Get(f.l))
-			f.inPool.Put(p)
+				p.Payload = p.Payload[:mtu]
+				f.inPool.Put(p)
+			}
+			f.releasePacketBatch(batch)
 		case <-ctx.Done():
 			f.wg.Done()
 			return
@@ -379,10 +543,13 @@ func (f *Interface) workerOut(i int, ctx context.Context) {
 
 	for {
 		select {
-		case data := <-f.outbound:
+		case batch := <-f.outbound[i]:
-			f.consumeInsidePacket(*data, fwPacket1, nb1, result1, i, conntrackCache.Get(f.l))
+			for _, data := range batch.payloads {
-			*data = (*data)[:mtu]
+				f.consumeInsidePacket(*data, fwPacket1, nb1, result1, i, conntrackCache.Get(f.l))
-			f.outPool.Put(data)
+				*data = (*data)[:mtu]
+				f.outPool.Put(data)
+			}
+			f.releaseOutboundBatch(batch)
 		case <-ctx.Done():
 			f.wg.Done()
 			return

main.go

@@ -221,6 +221,13 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		}
 	}
 
+	batchCfg := BatchConfig{
+		InboundBatchSize:      c.GetInt("batch.inbound_size", inboundBatchSizeDefault),
+		OutboundBatchSize:     c.GetInt("batch.outbound_size", outboundBatchSizeDefault),
+		FlushInterval:         c.GetDuration("batch.flush_interval", batchFlushIntervalDefault),
+		MaxOutstandingPerChan: c.GetInt("batch.max_outstanding", maxOutstandingBatchesDefault),
+	}
+
 	ifConfig := &InterfaceConfig{
 		HostMap:               hostMap,
 		Inside:                tun,
@@ -242,6 +249,7 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		relayManager:          NewRelayManager(ctx, l, hostMap, c),
 		punchy:                punchy,
 		ConntrackCacheTimeout: conntrackCacheTimeout,
+		BatchConfig:           batchCfg,
 		l:                     l,
 	}
 

service/service.go

@@ -9,13 +9,10 @@ import (
 	"math"
 	"net"
 	"net/netip"
-	"os"
 	"strings"
 	"sync"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
-	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/overlay"
 	"golang.org/x/sync/errgroup"
 	"gvisor.dev/gvisor/pkg/buffer"
@@ -46,15 +43,7 @@ type Service struct {
 	}
 }
 
-func New(config *config.C) (*Service, error) {
+func New(control *nebula.Control) (*Service, error) {
-	logger := logrus.New()
-	logger.Out = os.Stdout
-
-	control, err := nebula.Main(config, false, "custom-app", logger, overlay.NewUserDeviceFromConfig)
-	if err != nil {
-		return nil, err
-	}
-
 	wait, err := control.Start()
 	if err != nil {
 		return nil, err

udp/conn.go

@@ -30,8 +30,8 @@ func (NoopConn) Rebind() error {
 func (NoopConn) LocalAddr() (netip.AddrPort, error) {
 	return netip.AddrPort{}, nil
 }
-func (NoopConn) ListenOut(_ EncReader) {
+func (NoopConn) ListenOut(_ EncReader) error {
-	return
+	return nil
 }
 func (NoopConn) WriteTo(_ []byte, _ netip.AddrPort) error {
 	return nil

udp/udp_darwin.go

@@ -165,7 +165,7 @@ func NewUDPStatsEmitter(udpConns []Conn) func() {
 	return func() {}
 }
 
-func (u *StdConn) ListenOut(r EncReader) {
+func (u *StdConn) ListenOut(r EncReader) error {
 	buffer := make([]byte, MTU)
 
 	for {
@@ -174,14 +174,17 @@ func (u *StdConn) ListenOut(r EncReader) {
 		if err != nil {
 			if errors.Is(err, net.ErrClosed) {
 				u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-				return
+				return err
 			}
 
 			u.l.WithError(err).Error("unexpected udp socket receive error")
+			continue
 		}
 
 		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n])
 	}
+
+	return nil
 }
 
 func (u *StdConn) Rebind() error {

udp/udp_linux.go

@@ -5,9 +5,11 @@ package udp
 
 import (
 	"encoding/binary"
+	"errors"
 	"fmt"
 	"net"
 	"net/netip"
+	"sync"
 	"syscall"
 	"time"
 	"unsafe"
@@ -20,11 +22,38 @@ import (
 
 var readTimeout = unix.NsecToTimeval(int64(time.Millisecond * 500))
 
+const (
+	defaultGSOMaxSegments    = 8
+	defaultGSOFlushTimeout   = 150 * time.Microsecond
+	defaultGROReadBufferSize = MTU * defaultGSOMaxSegments
+	maxGSOBatchBytes         = 0xFFFF
+)
+
+var (
+	errGSOFallback = errors.New("udp gso fallback")
+	errGSODisabled = errors.New("udp gso disabled")
+)
+
 type StdConn struct {
 	sysFd int
 	isV4  bool
 	l     *logrus.Logger
 	batch int
+
+	enableGRO bool
+	enableGSO bool
+
+	gsoMu           sync.Mutex
+	gsoBuf          []byte
+	gsoAddr         netip.AddrPort
+	gsoSegSize      int
+	gsoSegments     int
+	gsoMaxSegments  int
+	gsoMaxBytes     int
+	gsoFlushTimeout time.Duration
+	gsoTimer        *time.Timer
+
+	groBufSize int
 }
 
 func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int) (Conn, error) {
@@ -69,7 +98,16 @@ func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch in
 		return nil, fmt.Errorf("unable to bind to socket: %s", err)
 	}
 
-	return &StdConn{sysFd: fd, isV4: ip.Is4(), l: l, batch: batch}, err
+	return &StdConn{
+		sysFd:           fd,
+		isV4:            ip.Is4(),
+		l:               l,
+		batch:           batch,
+		gsoMaxSegments:  defaultGSOMaxSegments,
+		gsoMaxBytes:     MTU * defaultGSOMaxSegments,
+		gsoFlushTimeout: defaultGSOFlushTimeout,
+		groBufSize:      MTU,
+	}, err
 }
 
 func (u *StdConn) Rebind() error {
@@ -119,15 +157,42 @@ func (u *StdConn) LocalAddr() (netip.AddrPort, error) {
 }
 
 func (u *StdConn) ListenOut(r EncReader) error {
-	var ip netip.Addr
+	var (
+		ip       netip.Addr
+		controls [][]byte
+	)
 
-	msgs, buffers, names := u.PrepareRawMessages(u.batch)
+	bufSize := u.readBufferSize()
+	msgs, buffers, names := u.PrepareRawMessages(u.batch, bufSize)
 	read := u.ReadMulti
 	if u.batch == 1 {
 		read = u.ReadSingle
 	}
 
 	for {
+		desired := u.readBufferSize()
+		if len(buffers) == 0 || cap(buffers[0]) < desired {
+			msgs, buffers, names = u.PrepareRawMessages(u.batch, desired)
+			controls = nil
+		}
+
+		if u.enableGRO {
+			if controls == nil {
+				controls = make([][]byte, len(msgs))
+				for i := range controls {
+					controls[i] = make([]byte, unix.CmsgSpace(4))
+				}
+			}
+			for i := range msgs {
+				setRawMessageControl(&msgs[i], controls[i])
+			}
+		} else if controls != nil {
+			for i := range msgs {
+				setRawMessageControl(&msgs[i], nil)
+			}
+			controls = nil
+		}
+
 		n, err := read(msgs)
 		if err != nil {
 			return err
@@ -140,11 +205,82 @@ func (u *StdConn) ListenOut(r EncReader) error {
 			} else {
 				ip, _ = netip.AddrFromSlice(names[i][8:24])
 			}
-			r(netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4])), buffers[i][:msgs[i].Len])
+			addr := netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4]))
+			payload := buffers[i][:msgs[i].Len]
+
+			if u.enableGRO && u.l.IsLevelEnabled(logrus.DebugLevel) {
+				ctrlLen := getRawMessageControlLen(&msgs[i])
+				msgFlags := getRawMessageFlags(&msgs[i])
+				u.l.WithFields(logrus.Fields{
+					"tag":         "gro-debug",
+					"stage":       "recv",
+					"payload_len": len(payload),
+					"ctrl_len":    ctrlLen,
+					"msg_flags":   msgFlags,
+				}).Debug("gro batch data")
+				if controls != nil && ctrlLen > 0 {
+					maxDump := ctrlLen
+					if maxDump > 16 {
+						maxDump = 16
+					}
+					u.l.WithFields(logrus.Fields{
+						"tag":         "gro-debug",
+						"stage":       "control-bytes",
+						"control_hex": fmt.Sprintf("%x", controls[i][:maxDump]),
+						"datalen":     ctrlLen,
+					}).Debug("gro control dump")
+				}
+			}
+
+			sawControl := false
+			if controls != nil {
+				if ctrlLen := getRawMessageControlLen(&msgs[i]); ctrlLen > 0 {
+					if segSize, segCount := parseGROControl(controls[i][:ctrlLen]); segSize > 0 {
+						sawControl = true
+						if u.l.IsLevelEnabled(logrus.DebugLevel) {
+							u.l.WithFields(logrus.Fields{
+								"tag":        "gro-debug",
+								"stage":      "control",
+								"seg_size":   segSize,
+								"seg_count":  segCount,
+								"payloadLen": len(payload),
+							}).Debug("gro control parsed")
+						}
+						segSize = normalizeGROSegSize(segSize, segCount, len(payload))
+						if segSize > 0 && segSize < len(payload) {
+							if u.emitGROSegments(r, addr, payload, segSize) {
+								continue
+							}
+						}
+					}
+				}
+			}
+
+			if u.enableGRO && len(payload) > MTU {
+				if !sawControl && u.l.IsLevelEnabled(logrus.DebugLevel) {
+					u.l.WithFields(logrus.Fields{
+						"tag":         "gro-debug",
+						"stage":       "fallback",
+						"payload_len": len(payload),
+					}).Debug("gro control missing; splitting payload by MTU")
+				}
+				if u.emitGROSegments(r, addr, payload, MTU) {
+					continue
+				}
+			}
+
+			r(addr, payload)
 		}
 	}
 }
 
+func (u *StdConn) readBufferSize() int {
+	if u.enableGRO && u.groBufSize > MTU {
+		return u.groBufSize
+	}
+	return MTU
+}
+
 func (u *StdConn) ReadSingle(msgs []rawMessage) (int, error) {
 	for {
 		n, _, err := unix.Syscall6(
@@ -193,6 +329,14 @@ func (u *StdConn) ReadMulti(msgs []rawMessage) (int, error) {
 }
 
 func (u *StdConn) WriteTo(b []byte, ip netip.AddrPort) error {
+	if u.enableGSO && ip.IsValid() {
+		if err := u.queueGSOPacket(b, ip); err == nil {
+			return nil
+		} else if !errors.Is(err, errGSOFallback) {
+			return err
+		}
+	}
+
 	if u.isV4 {
 		return u.writeTo4(b, ip)
 	}
@@ -299,6 +443,94 @@ func (u *StdConn) ReloadConfig(c *config.C) {
 			u.l.WithError(err).Error("Failed to set listen.so_mark")
 		}
 	}
+
+	u.configureGRO(c)
+	u.configureGSO(c)
+}
+
+func (u *StdConn) configureGRO(c *config.C) {
+	if c == nil {
+		return
+	}
+
+	enable := c.GetBool("listen.enable_gro", false)
+	if enable == u.enableGRO {
+		if enable {
+			if size := c.GetInt("listen.gro_read_buffer", 0); size > 0 {
+				u.setGROBufferSize(size)
+			}
+		}
+		return
+	}
+
+	if enable {
+		if err := unix.SetsockoptInt(u.sysFd, unix.SOL_UDP, unix.UDP_GRO, 1); err != nil {
+			u.l.WithError(err).Warn("Failed to enable UDP GRO")
+			return
+		}
+		u.enableGRO = true
+		u.setGROBufferSize(c.GetInt("listen.gro_read_buffer", defaultGROReadBufferSize))
+		u.l.WithField("buffer_size", u.groBufSize).Info("UDP GRO enabled")
+		return
+	}
+
+	if err := unix.SetsockoptInt(u.sysFd, unix.SOL_UDP, unix.UDP_GRO, 0); err != nil && err != unix.ENOPROTOOPT {
+		u.l.WithError(err).Warn("Failed to disable UDP GRO")
+	}
+	u.enableGRO = false
+	u.groBufSize = MTU
+}
+
+func (u *StdConn) configureGSO(c *config.C) {
+	enable := c.GetBool("listen.enable_gso", false)
+	if !enable {
+		u.disableGSO()
+	} else {
+		u.enableGSO = true
+	}
+
+	segments := c.GetInt("listen.gso_max_segments", defaultGSOMaxSegments)
+	if segments < 1 {
+		segments = 1
+	}
+	u.gsoMaxSegments = segments
+
+	maxBytes := c.GetInt("listen.gso_max_bytes", 0)
+	if maxBytes <= 0 {
+		maxBytes = MTU * segments
+	}
+	if maxBytes > maxGSOBatchBytes {
+		u.l.WithField("requested", maxBytes).Warn("listen.gso_max_bytes larger than UDP limit; clamping")
+		maxBytes = maxGSOBatchBytes
+	}
+	u.gsoMaxBytes = maxBytes
+
+	timeout := c.GetDuration("listen.gso_flush_timeout", defaultGSOFlushTimeout)
+	if timeout < 0 {
+		timeout = 0
+	}
+	u.gsoFlushTimeout = timeout
+}
+
+func (u *StdConn) setGROBufferSize(size int) {
+	if size < MTU {
+		size = defaultGROReadBufferSize
+	}
+	if size > maxGSOBatchBytes {
+		size = maxGSOBatchBytes
+	}
+	u.groBufSize = size
+}
+
+func (u *StdConn) disableGSO() {
+	u.gsoMu.Lock()
+	defer u.gsoMu.Unlock()
+	u.enableGSO = false
+	_ = u.flushGSOlocked()
+	u.gsoBuf = nil
+	u.gsoSegments = 0
+	u.gsoSegSize = 0
+	u.stopGSOTimerLocked()
 }
 
 func (u *StdConn) getMemInfo(meminfo *[unix.SK_MEMINFO_VARS]uint32) error {
@@ -310,7 +542,239 @@ func (u *StdConn) getMemInfo(meminfo *[unix.SK_MEMINFO_VARS]uint32) error {
 	return nil
 }
 
+func (u *StdConn) queueGSOPacket(b []byte, addr netip.AddrPort) error {
+	if len(b) == 0 {
+		return nil
+	}
+
+	u.gsoMu.Lock()
+	defer u.gsoMu.Unlock()
+
+	if !u.enableGSO || !addr.IsValid() || len(b) > u.gsoMaxBytes {
+		if err := u.flushGSOlocked(); err != nil {
+			return err
+		}
+		return errGSOFallback
+	}
+
+	if u.gsoSegments == 0 {
+		if cap(u.gsoBuf) < u.gsoMaxBytes {
+			u.gsoBuf = make([]byte, 0, u.gsoMaxBytes)
+		}
+		u.gsoAddr = addr
+		u.gsoSegSize = len(b)
+	} else if addr != u.gsoAddr || len(b) != u.gsoSegSize {
+		if err := u.flushGSOlocked(); err != nil {
+			return err
+		}
+		if cap(u.gsoBuf) < u.gsoMaxBytes {
+			u.gsoBuf = make([]byte, 0, u.gsoMaxBytes)
+		}
+		u.gsoAddr = addr
+		u.gsoSegSize = len(b)
+	}
+
+	if len(u.gsoBuf)+len(b) > u.gsoMaxBytes {
+		if err := u.flushGSOlocked(); err != nil {
+			return err
+		}
+		if cap(u.gsoBuf) < u.gsoMaxBytes {
+			u.gsoBuf = make([]byte, 0, u.gsoMaxBytes)
+		}
+		u.gsoAddr = addr
+		u.gsoSegSize = len(b)
+	}
+
+	u.gsoBuf = append(u.gsoBuf, b...)
+	u.gsoSegments++
+
+	if u.gsoSegments >= u.gsoMaxSegments || u.gsoFlushTimeout <= 0 {
+		return u.flushGSOlocked()
+	}
+
+	u.scheduleGSOFlushLocked()
+	return nil
+}
+
+func (u *StdConn) flushGSOlocked() error {
+	if u.gsoSegments == 0 {
+		u.stopGSOTimerLocked()
+		return nil
+	}
+
+	payload := append([]byte(nil), u.gsoBuf...)
+	addr := u.gsoAddr
+	segSize := u.gsoSegSize
+
+	u.gsoBuf = u.gsoBuf[:0]
+	u.gsoSegments = 0
+	u.gsoSegSize = 0
+	u.stopGSOTimerLocked()
+
+	if segSize <= 0 {
+		return errGSOFallback
+	}
+
+	err := u.sendSegmented(payload, addr, segSize)
+	if errors.Is(err, errGSODisabled) {
+		u.l.WithField("addr", addr).Warn("UDP GSO disabled by kernel, falling back to sendto")
+		u.enableGSO = false
+		return u.sendSegmentsIndividually(payload, addr, segSize)
+	}
+
+	return err
+}
+
+func (u *StdConn) sendSegmented(payload []byte, addr netip.AddrPort, segSize int) error {
+	if len(payload) == 0 {
+		return nil
+	}
+
+	control := make([]byte, unix.CmsgSpace(2))
+	hdr := (*unix.Cmsghdr)(unsafe.Pointer(&control[0]))
+	hdr.Level = unix.SOL_UDP
+	hdr.Type = unix.UDP_SEGMENT
+	setCmsgLen(hdr, unix.CmsgLen(2))
+	binary.NativeEndian.PutUint16(control[unix.CmsgLen(0):unix.CmsgLen(0)+2], uint16(segSize))
+
+	var sa unix.Sockaddr
+	if addr.Addr().Is4() {
+		var sa4 unix.SockaddrInet4
+		sa4.Port = int(addr.Port())
+		sa4.Addr = addr.Addr().As4()
+		sa = &sa4
+	} else {
+		var sa6 unix.SockaddrInet6
+		sa6.Port = int(addr.Port())
+		sa6.Addr = addr.Addr().As16()
+		sa = &sa6
+	}
+
+	if _, err := unix.SendmsgN(u.sysFd, payload, control, sa, 0); err != nil {
+		if errno, ok := err.(syscall.Errno); ok && (errno == unix.EINVAL || errno == unix.ENOTSUP || errno == unix.EOPNOTSUPP) {
+			return errGSODisabled
+		}
+		return &net.OpError{Op: "sendmsg", Err: err}
+	}
+	return nil
+}
+
+func (u *StdConn) sendSegmentsIndividually(buf []byte, addr netip.AddrPort, segSize int) error {
+	if segSize <= 0 {
+		return errGSOFallback
+	}
+
+	for offset := 0; offset < len(buf); offset += segSize {
+		end := offset + segSize
+		if end > len(buf) {
+			end = len(buf)
+		}
+		var err error
+		if u.isV4 {
+			err = u.writeTo4(buf[offset:end], addr)
+		} else {
+			err = u.writeTo6(buf[offset:end], addr)
+		}
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (u *StdConn) scheduleGSOFlushLocked() {
+	if u.gsoTimer == nil {
+		u.gsoTimer = time.AfterFunc(u.gsoFlushTimeout, u.gsoFlushTimer)
+		return
+	}
+	u.gsoTimer.Reset(u.gsoFlushTimeout)
+}
+
+func (u *StdConn) stopGSOTimerLocked() {
+	if u.gsoTimer != nil {
+		u.gsoTimer.Stop()
+		u.gsoTimer = nil
+	}
+}
+
+func (u *StdConn) gsoFlushTimer() {
+	u.gsoMu.Lock()
+	defer u.gsoMu.Unlock()
+	_ = u.flushGSOlocked()
+}
+
+func parseGROControl(control []byte) (int, int) {
+	if len(control) == 0 {
+		return 0, 0
+	}
+
+	cmsgs, err := unix.ParseSocketControlMessage(control)
+	if err != nil {
+		return 0, 0
+	}
+
+	for _, c := range cmsgs {
+		if c.Header.Level == unix.SOL_UDP && c.Header.Type == unix.UDP_GRO && len(c.Data) >= 2 {
+			segSize := int(binary.NativeEndian.Uint16(c.Data[:2]))
+			segCount := 0
+			if len(c.Data) >= 4 {
+				segCount = int(binary.NativeEndian.Uint16(c.Data[2:4]))
+			}
+			return segSize, segCount
+		}
+	}
+
+	return 0, 0
+}
+
+func (u *StdConn) emitGROSegments(r EncReader, addr netip.AddrPort, payload []byte, segSize int) bool {
+	if segSize <= 0 {
+		return false
+	}
+
+	for offset := 0; offset < len(payload); offset += segSize {
+		end := offset + segSize
+		if end > len(payload) {
+			end = len(payload)
+		}
+		segment := make([]byte, end-offset)
+		copy(segment, payload[offset:end])
+		r(addr, segment)
+	}
+	return true
+}
+
+func normalizeGROSegSize(segSize, segCount, total int) int {
+	if segSize <= 0 || total <= 0 {
+		return segSize
+	}
+
+	if segSize > total && segCount > 0 {
+		segSize = total / segCount
+		if segSize == 0 {
+			segSize = total
+		}
+	}
+
+	if segCount <= 1 && segSize > 0 && total > segSize {
+		calculated := total / segSize
+		if calculated <= 1 {
+			calculated = (total + segSize - 1) / segSize
+		}
+		if calculated > 1 {
+			segCount = calculated
+		}
+	}
+
+	if segSize > MTU {
+		return MTU
+	}
+
+	return segSize
+}
+
 func (u *StdConn) Close() error {
+	u.disableGSO()
 	return syscall.Close(u.sysFd)
 }
 

udp/udp_linux_32.go

@@ -30,13 +30,16 @@ type rawMessage struct {
 	Len uint32
 }
 
-func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
+func (u *StdConn) PrepareRawMessages(n int, bufSize int) ([]rawMessage, [][]byte, [][]byte) {
+	if bufSize <= 0 {
+		bufSize = MTU
+	}
 	msgs := make([]rawMessage, n)
 	buffers := make([][]byte, n)
 	names := make([][]byte, n)
 
 	for i := range msgs {
-		buffers[i] = make([]byte, MTU)
+		buffers[i] = make([]byte, bufSize)
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
@@ -52,3 +55,25 @@ func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 
 	return msgs, buffers, names
 }
+
+func setRawMessageControl(msg *rawMessage, buf []byte) {
+	if len(buf) == 0 {
+		msg.Hdr.Control = nil
+		msg.Hdr.Controllen = 0
+		return
+	}
+	msg.Hdr.Control = &buf[0]
+	msg.Hdr.Controllen = uint32(len(buf))
+}
+
+func getRawMessageControlLen(msg *rawMessage) int {
+	return int(msg.Hdr.Controllen)
+}
+
+func getRawMessageFlags(msg *rawMessage) int {
+	return int(msg.Hdr.Flags)
+}
+
+func setCmsgLen(h *unix.Cmsghdr, l int) {
+	h.Len = uint32(l)
+}

udp/udp_linux_64.go

@@ -33,13 +33,16 @@ type rawMessage struct {
 	Pad0 [4]byte
 }
 
-func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
+func (u *StdConn) PrepareRawMessages(n int, bufSize int) ([]rawMessage, [][]byte, [][]byte) {
+	if bufSize <= 0 {
+		bufSize = MTU
+	}
 	msgs := make([]rawMessage, n)
 	buffers := make([][]byte, n)
 	names := make([][]byte, n)
 
 	for i := range msgs {
-		buffers[i] = make([]byte, MTU)
+		buffers[i] = make([]byte, bufSize)
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
@@ -55,3 +58,25 @@ func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 
 	return msgs, buffers, names
 }
+
+func setRawMessageControl(msg *rawMessage, buf []byte) {
+	if len(buf) == 0 {
+		msg.Hdr.Control = nil
+		msg.Hdr.Controllen = 0
+		return
+	}
+	msg.Hdr.Control = &buf[0]
+	msg.Hdr.Controllen = uint64(len(buf))
+}
+
+func getRawMessageControlLen(msg *rawMessage) int {
+	return int(msg.Hdr.Controllen)
+}
+
+func getRawMessageFlags(msg *rawMessage) int {
+	return int(msg.Hdr.Flags)
+}
+
+func setCmsgLen(h *unix.Cmsghdr, l int) {
+	h.Len = uint64(l)
+}