Improve darwin tun performance

2025-11-01 04:12:50 +08:00 · 2025-07-02 19:16:14 +08:00
parent 8763c24e49
commit a0881ada32
16 changed files with 1894 additions and 159 deletions
--- a/internal/fdbased_darwin/endpoint.go
+++ b/internal/fdbased_darwin/endpoint.go
@@ -0,0 +1,648 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package fdbased provides the implementation of data-link layer endpoints
+// backed by boundary-preserving file descriptors (e.g., TUN devices,
+// seqpacket/datagram sockets).
+//
+// FD based endpoints can be used in the networking stack by calling New() to
+// create a new endpoint, and then passing it as an argument to
+// Stack.CreateNIC().
+//
+// FD based endpoints can use more than one file descriptor to read incoming
+// packets. If there are more than one FDs specified and the underlying FD is an
+// AF_PACKET then the endpoint will enable FANOUT mode on the socket so that the
+// host kernel will consistently hash the packets to the sockets. This ensures
+// that packets for the same TCP streams are not reordered.
+//
+// Similarly if more than one FD's are specified where the underlying FD is not
+// AF_PACKET then it's the caller's responsibility to ensure that all inbound
+// packets on the descriptors are consistently 5 tuple hashed to one of the
+// descriptors to prevent TCP reordering.
+//
+// Since netstack today does not compute 5 tuple hashes for outgoing packets we
+// only use the first FD to write outbound packets. Once 5 tuple hashes for
+// all outbound packets are available we will make use of all underlying FD's to
+// write outbound packets.
+package fdbased
+
+import (
+	"fmt"
+	"runtime"
+
+	"github.com/sagernet/gvisor/pkg/buffer"
+	"github.com/sagernet/gvisor/pkg/sync"
+	"github.com/sagernet/gvisor/pkg/tcpip"
+	"github.com/sagernet/gvisor/pkg/tcpip/header"
+	"github.com/sagernet/gvisor/pkg/tcpip/stack"
+	"github.com/sagernet/sing-tun/internal/rawfile_darwin"
+	"github.com/sagernet/sing/common"
+
+	"golang.org/x/sys/unix"
+)
+
+// linkDispatcher reads packets from the link FD and dispatches them to the
+// NetworkDispatcher.
+type linkDispatcher interface {
+	Stop()
+	dispatch() (bool, tcpip.Error)
+	release()
+}
+
+// PacketDispatchMode are the various supported methods of receiving and
+// dispatching packets from the underlying FD.
+type PacketDispatchMode int
+
+// BatchSize is the number of packets to write in each syscall. It is 47
+// because when GVisorGSO is in use then a single 65KB TCP segment can get
+// split into 46 segments of 1420 bytes and a single 216 byte segment.
+const BatchSize = 47
+
+const (
+	// Readv is the default dispatch mode and is the least performant of the
+	// dispatch options but the one that is supported by all underlying FD
+	// types.
+	Readv PacketDispatchMode = iota
+)
+
+func (p PacketDispatchMode) String() string {
+	switch p {
+	case Readv:
+		return "Readv"
+	default:
+		return fmt.Sprintf("unknown packet dispatch mode '%d'", p)
+	}
+}
+
+var (
+	_ stack.LinkEndpoint = (*endpoint)(nil)
+	_ stack.GSOEndpoint  = (*endpoint)(nil)
+)
+
+// +stateify savable
+type fdInfo struct {
+	fd       int
+	isSocket bool
+}
+
+// +stateify savable
+type endpoint struct {
+	// fds is the set of file descriptors each identifying one inbound/outbound
+	// channel. The endpoint will dispatch from all inbound channels as well as
+	// hash outbound packets to specific channels based on the packet hash.
+	fds []fdInfo
+
+	// hdrSize specifies the link-layer header size. If set to 0, no header
+	// is added/removed; otherwise an ethernet header is used.
+	hdrSize int
+
+	// caps holds the endpoint capabilities.
+	caps stack.LinkEndpointCapabilities
+
+	// closed is a function to be called when the FD's peer (if any) closes
+	// its end of the communication pipe.
+	closed func(tcpip.Error) `state:"nosave"`
+
+	inboundDispatchers []linkDispatcher
+
+	mu endpointRWMutex `state:"nosave"`
+	// +checklocks:mu
+	dispatcher stack.NetworkDispatcher
+
+	// packetDispatchMode controls the packet dispatcher used by this
+	// endpoint.
+	packetDispatchMode PacketDispatchMode
+
+	// wg keeps track of running goroutines.
+	wg sync.WaitGroup `state:"nosave"`
+
+	// maxSyscallHeaderBytes has the same meaning as
+	// Options.MaxSyscallHeaderBytes.
+	maxSyscallHeaderBytes uintptr
+
+	// writevMaxIovs is the maximum number of iovecs that may be passed to
+	// rawfile.NonBlockingWriteIovec, as possibly limited by
+	// maxSyscallHeaderBytes. (No analogous limit is defined for
+	// rawfile.NonBlockingSendMMsg, since in that case the maximum number of
+	// iovecs also depends on the number of mmsghdrs. Instead, if sendBatch
+	// encounters a packet whose iovec count is limited by
+	// maxSyscallHeaderBytes, it falls back to writing the packet using writev
+	// via WritePacket.)
+	writevMaxIovs int
+
+	// addr is the address of the endpoint.
+	//
+	// +checklocks:mu
+	addr tcpip.LinkAddress
+
+	// mtu (maximum transmission unit) is the maximum size of a packet.
+	// +checklocks:mu
+	mtu uint32
+
+	batchSize int
+}
+
+// Options specify the details about the fd-based endpoint to be created.
+//
+// +stateify savable
+type Options struct {
+	// FDs is a set of FDs used to read/write packets.
+	FDs []int
+
+	// MTU is the mtu to use for this endpoint.
+	MTU uint32
+
+	// EthernetHeader if true, indicates that the endpoint should read/write
+	// ethernet frames instead of IP packets.
+	EthernetHeader bool
+
+	// ClosedFunc is a function to be called when an endpoint's peer (if
+	// any) closes its end of the communication pipe.
+	ClosedFunc func(tcpip.Error)
+
+	// Address is the link address for this endpoint. Only used if
+	// EthernetHeader is true.
+	Address tcpip.LinkAddress
+
+	// SaveRestore if true, indicates that this NIC capability set should
+	// include CapabilitySaveRestore
+	SaveRestore bool
+
+	// DisconnectOk if true, indicates that this NIC capability set should
+	// include CapabilityDisconnectOk.
+	DisconnectOk bool
+
+	// PacketDispatchMode specifies the type of inbound dispatcher to be
+	// used for this endpoint.
+	PacketDispatchMode PacketDispatchMode
+
+	// TXChecksumOffload if true, indicates that this endpoints capability
+	// set should include CapabilityTXChecksumOffload.
+	TXChecksumOffload bool
+
+	// RXChecksumOffload if true, indicates that this endpoints capability
+	// set should include CapabilityRXChecksumOffload.
+	RXChecksumOffload bool
+
+	// If MaxSyscallHeaderBytes is non-zero, it is the maximum number of bytes
+	// of struct iovec, msghdr, and mmsghdr that may be passed by each host
+	// system call.
+	MaxSyscallHeaderBytes int
+
+	// InterfaceIndex is the interface index of the underlying device.
+	InterfaceIndex int
+
+	// ProcessorsPerChannel is the number of goroutines used to handle packets
+	// from each FD.
+	ProcessorsPerChannel int
+}
+
+// New creates a new fd-based endpoint.
+//
+// Makes fd non-blocking, but does not take ownership of fd, which must remain
+// open for the lifetime of the returned endpoint (until after the endpoint has
+// stopped being using and Wait returns).
+func New(opts *Options) (stack.LinkEndpoint, error) {
+	caps := stack.LinkEndpointCapabilities(0)
+	if opts.RXChecksumOffload {
+		caps |= stack.CapabilityRXChecksumOffload
+	}
+
+	if opts.TXChecksumOffload {
+		caps |= stack.CapabilityTXChecksumOffload
+	}
+
+	hdrSize := 0
+	if opts.EthernetHeader {
+		hdrSize = header.EthernetMinimumSize
+		caps |= stack.CapabilityResolutionRequired
+	}
+
+	if opts.SaveRestore {
+		caps |= stack.CapabilitySaveRestore
+	}
+
+	if opts.DisconnectOk {
+		caps |= stack.CapabilityDisconnectOk
+	}
+
+	if len(opts.FDs) == 0 {
+		return nil, fmt.Errorf("opts.FD is empty, at least one FD must be specified")
+	}
+
+	if opts.MaxSyscallHeaderBytes < 0 {
+		return nil, fmt.Errorf("opts.MaxSyscallHeaderBytes is negative")
+	}
+
+	e := &endpoint{
+		mtu:                   opts.MTU,
+		caps:                  caps,
+		closed:                opts.ClosedFunc,
+		addr:                  opts.Address,
+		hdrSize:               hdrSize,
+		packetDispatchMode:    opts.PacketDispatchMode,
+		maxSyscallHeaderBytes: uintptr(opts.MaxSyscallHeaderBytes),
+		writevMaxIovs:         rawfile.MaxIovs,
+		batchSize:             int((512*1024)/(opts.MTU)) + 1,
+	}
+	if e.maxSyscallHeaderBytes != 0 {
+		if max := int(e.maxSyscallHeaderBytes / rawfile.SizeofIovec); max < e.writevMaxIovs {
+			e.writevMaxIovs = max
+		}
+	}
+
+	// Create per channel dispatchers.
+	for _, fd := range opts.FDs {
+		if err := unix.SetNonblock(fd, true); err != nil {
+			return nil, fmt.Errorf("unix.SetNonblock(%v) failed: %v", fd, err)
+		}
+
+		e.fds = append(e.fds, fdInfo{fd: fd, isSocket: true})
+		if opts.ProcessorsPerChannel == 0 {
+			opts.ProcessorsPerChannel = common.Max(1, runtime.GOMAXPROCS(0)/len(opts.FDs))
+		}
+
+		inboundDispatcher, err := newRecvMMsgDispatcher(fd, e, opts)
+		if err != nil {
+			return nil, fmt.Errorf("createInboundDispatcher(...) = %v", err)
+		}
+		e.inboundDispatchers = append(e.inboundDispatchers, inboundDispatcher)
+	}
+
+	return e, nil
+}
+
+func isSocketFD(fd int) (bool, error) {
+	var stat unix.Stat_t
+	if err := unix.Fstat(fd, &stat); err != nil {
+		return false, fmt.Errorf("unix.Fstat(%v,...) failed: %v", fd, err)
+	}
+	return (stat.Mode & unix.S_IFSOCK) == unix.S_IFSOCK, nil
+}
+
+// Attach launches the goroutine that reads packets from the file descriptor and
+// dispatches them via the provided dispatcher. If one is already attached,
+// then nothing happens.
+//
+// Attach implements stack.LinkEndpoint.Attach.
+func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) {
+	e.mu.Lock()
+
+	// nil means the NIC is being removed.
+	if dispatcher == nil && e.dispatcher != nil {
+		for _, dispatcher := range e.inboundDispatchers {
+			dispatcher.Stop()
+		}
+		e.dispatcher = nil
+		// NOTE(gvisor.dev/issue/11456): Unlock e.mu before e.Wait().
+		e.mu.Unlock()
+		e.Wait()
+		return
+	}
+	defer e.mu.Unlock()
+	if dispatcher != nil && e.dispatcher == nil {
+		e.dispatcher = dispatcher
+		// Link endpoints are not savable. When transportation endpoints are
+		// saved, they stop sending outgoing packets and all incoming packets
+		// are rejected.
+		for i := range e.inboundDispatchers {
+			e.wg.Add(1)
+			go func(i int) { // S/R-SAFE: See above.
+				e.dispatchLoop(e.inboundDispatchers[i])
+				e.wg.Done()
+			}(i)
+		}
+	}
+}
+
+// IsAttached implements stack.LinkEndpoint.IsAttached.
+func (e *endpoint) IsAttached() bool {
+	e.mu.RLock()
+	defer e.mu.RUnlock()
+	return e.dispatcher != nil
+}
+
+// MTU implements stack.LinkEndpoint.MTU.
+func (e *endpoint) MTU() uint32 {
+	e.mu.RLock()
+	defer e.mu.RUnlock()
+	return e.mtu
+}
+
+// SetMTU implements stack.LinkEndpoint.SetMTU.
+func (e *endpoint) SetMTU(mtu uint32) {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	e.mtu = mtu
+}
+
+// Capabilities implements stack.LinkEndpoint.Capabilities.
+func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities {
+	return e.caps
+}
+
+// MaxHeaderLength returns the maximum size of the link-layer header.
+func (e *endpoint) MaxHeaderLength() uint16 {
+	return uint16(e.hdrSize)
+}
+
+// LinkAddress returns the link address of this endpoint.
+func (e *endpoint) LinkAddress() tcpip.LinkAddress {
+	e.mu.RLock()
+	defer e.mu.RUnlock()
+	return e.addr
+}
+
+// SetLinkAddress implements stack.LinkEndpoint.SetLinkAddress.
+func (e *endpoint) SetLinkAddress(addr tcpip.LinkAddress) {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	e.addr = addr
+}
+
+// Wait implements stack.LinkEndpoint.Wait. It waits for the endpoint to stop
+// reading from its FD.
+func (e *endpoint) Wait() {
+	e.wg.Wait()
+}
+
+// AddHeader implements stack.LinkEndpoint.AddHeader.
+func (e *endpoint) AddHeader(pkt *stack.PacketBuffer) {
+	if e.hdrSize > 0 {
+		// Add ethernet header if needed.
+		eth := header.Ethernet(pkt.LinkHeader().Push(header.EthernetMinimumSize))
+		eth.Encode(&header.EthernetFields{
+			SrcAddr: pkt.EgressRoute.LocalLinkAddress,
+			DstAddr: pkt.EgressRoute.RemoteLinkAddress,
+			Type:    pkt.NetworkProtocolNumber,
+		})
+	}
+}
+
+func (e *endpoint) parseHeader(pkt *stack.PacketBuffer) (header.Ethernet, bool) {
+	if e.hdrSize <= 0 {
+		return nil, true
+	}
+	hdrBytes, ok := pkt.LinkHeader().Consume(e.hdrSize)
+	if !ok {
+		return nil, false
+	}
+	hdr := header.Ethernet(hdrBytes)
+	pkt.NetworkProtocolNumber = hdr.Type()
+	return hdr, true
+}
+
+// parseInboundHeader parses the link header of pkt and returns true if the
+// header is well-formed and sent to this endpoint's MAC or the broadcast
+// address.
+func (e *endpoint) parseInboundHeader(pkt *stack.PacketBuffer, wantAddr tcpip.LinkAddress) bool {
+	hdr, ok := e.parseHeader(pkt)
+	if !ok || e.hdrSize <= 0 {
+		return ok
+	}
+	dstAddr := hdr.DestinationAddress()
+	// Per RFC 9542 2.1 on the least significant bit of the first octet of
+	// a MAC address: "If it is zero, the MAC address is unicast. If it is
+	// a one, the address is groupcast (multicast or broadcast)." Multicast
+	// and broadcast are the same thing to ethernet; they are both sent to
+	// everyone.
+	return dstAddr == wantAddr || byte(dstAddr[0])&0x01 == 1
+}
+
+// ParseHeader implements stack.LinkEndpoint.ParseHeader.
+func (e *endpoint) ParseHeader(pkt *stack.PacketBuffer) bool {
+	_, ok := e.parseHeader(pkt)
+	return ok
+}
+
+var (
+	packetHeader4 = []byte{0x00, 0x00, 0x00, unix.AF_INET}
+	packetHeader6 = []byte{0x00, 0x00, 0x00, unix.AF_INET6}
+)
+
+// writePacket writes outbound packets to the file descriptor. If it is not
+// currently writable, the packet is dropped.
+func (e *endpoint) writePacket(pkt *stack.PacketBuffer) tcpip.Error {
+	fdInfo := e.fds[pkt.Hash%uint32(len(e.fds))]
+	fd := fdInfo.fd
+	var vnetHdrBuf []byte
+	if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber {
+		vnetHdrBuf = packetHeader4
+	} else {
+		vnetHdrBuf = packetHeader6
+	}
+	views := pkt.AsSlices()
+	numIovecs := len(views)
+	if len(vnetHdrBuf) != 0 {
+		numIovecs++
+	}
+	if numIovecs > e.writevMaxIovs {
+		numIovecs = e.writevMaxIovs
+	}
+
+	// Allocate small iovec arrays on the stack.
+	var iovecsArr [8]unix.Iovec
+	iovecs := iovecsArr[:0]
+	if numIovecs > len(iovecsArr) {
+		iovecs = make([]unix.Iovec, 0, numIovecs)
+	}
+	iovecs = rawfile.AppendIovecFromBytes(iovecs, vnetHdrBuf, numIovecs)
+	for _, v := range views {
+		iovecs = rawfile.AppendIovecFromBytes(iovecs, v, numIovecs)
+	}
+	if errno := rawfile.NonBlockingWriteIovec(fd, iovecs); errno != 0 {
+		return TranslateErrno(errno)
+	}
+	return nil
+}
+
+func (e *endpoint) sendBatch(batchFDInfo fdInfo, pkts []*stack.PacketBuffer) (int, tcpip.Error) {
+	// Degrade to writePacket if underlying fd is not a socket.
+	if !batchFDInfo.isSocket {
+		var written int
+		var err tcpip.Error
+		for written < len(pkts) {
+			if err = e.writePacket(pkts[written]); err != nil {
+				break
+			}
+			written++
+		}
+		return written, err
+	}
+
+	// Send a batch of packets through batchFD.
+	batchFD := batchFDInfo.fd
+	mmsgHdrsStorage := make([]rawfile.MsgHdrX, 0, len(pkts))
+	packets := 0
+	for packets < len(pkts) {
+		mmsgHdrs := mmsgHdrsStorage
+		batch := pkts[packets:]
+		syscallHeaderBytes := uintptr(0)
+		for _, pkt := range batch {
+			var vnetHdrBuf []byte
+			if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber {
+				vnetHdrBuf = packetHeader4
+			} else {
+				vnetHdrBuf = packetHeader6
+			}
+			views, offset := pkt.AsViewList()
+			var skipped int
+			var view *buffer.View
+			for view = views.Front(); view != nil && offset >= view.Size(); view = view.Next() {
+				offset -= view.Size()
+				skipped++
+			}
+
+			// We've made it to the usable views.
+			numIovecs := views.Len() - skipped
+			if len(vnetHdrBuf) != 0 {
+				numIovecs++
+			}
+			if numIovecs > rawfile.MaxIovs {
+				numIovecs = rawfile.MaxIovs
+			}
+			if e.maxSyscallHeaderBytes != 0 {
+				syscallHeaderBytes += rawfile.SizeofMsgHdrX + uintptr(numIovecs)*rawfile.SizeofIovec
+				if syscallHeaderBytes > e.maxSyscallHeaderBytes {
+					// We can't fit this packet into this call to sendmmsg().
+					// We could potentially do so if we reduced numIovecs
+					// further, but this might incur considerable extra
+					// copying. Leave it to the next batch instead.
+					break
+				}
+			}
+
+			// We can't easily allocate iovec arrays on the stack here since
+			// they will escape this loop iteration via mmsgHdrs.
+			iovecs := make([]unix.Iovec, 0, numIovecs)
+			iovecs = rawfile.AppendIovecFromBytes(iovecs, vnetHdrBuf, numIovecs)
+			// At most one slice has a non-zero offset.
+			iovecs = rawfile.AppendIovecFromBytes(iovecs, view.AsSlice()[offset:], numIovecs)
+			for view = view.Next(); view != nil; view = view.Next() {
+				iovecs = rawfile.AppendIovecFromBytes(iovecs, view.AsSlice(), numIovecs)
+			}
+
+			var mmsgHdr rawfile.MsgHdrX
+			mmsgHdr.Msg.Iov = &iovecs[0]
+			mmsgHdr.Msg.SetIovlen(len(iovecs))
+			// mmsgHdr.DataLen = uint32(len(iovecs))
+			mmsgHdrs = append(mmsgHdrs, mmsgHdr)
+		}
+
+		if len(mmsgHdrs) == 0 {
+			// We can't fit batch[0] into a mmsghdr while staying under
+			// e.maxSyscallHeaderBytes. Use WritePacket, which will avoid the
+			// mmsghdr (by using writev) and re-buffer iovecs more aggressively
+			// if necessary (by using e.writevMaxIovs instead of
+			// rawfile.MaxIovs).
+			pkt := batch[0]
+			if err := e.writePacket(pkt); err != nil {
+				return packets, err
+			}
+			packets++
+		} else {
+			for len(mmsgHdrs) > 0 {
+				sent, errno := rawfile.NonBlockingSendMMsg(batchFD, mmsgHdrs)
+				if errno != 0 {
+					return packets, TranslateErrno(errno)
+				}
+				packets += sent
+				mmsgHdrs = mmsgHdrs[sent:]
+			}
+		}
+	}
+
+	return packets, nil
+}
+
+// WritePackets writes outbound packets to the underlying file descriptors. If
+// one is not currently writable, the packet is dropped.
+//
+// Being a batch API, each packet in pkts should have the following
+// fields populated:
+//   - pkt.EgressRoute
+//   - pkt.GSOOptions
+//   - pkt.NetworkProtocolNumber
+func (e *endpoint) WritePackets(pkts stack.PacketBufferList) (int, tcpip.Error) {
+	// Preallocate to avoid repeated reallocation as we append to batch.
+	batch := make([]*stack.PacketBuffer, 0, e.batchSize)
+	batchFDInfo := fdInfo{fd: -1, isSocket: false}
+	sentPackets := 0
+	for _, pkt := range pkts.AsSlice() {
+		if len(batch) == 0 {
+			batchFDInfo = e.fds[pkt.Hash%uint32(len(e.fds))]
+		}
+		pktFDInfo := e.fds[pkt.Hash%uint32(len(e.fds))]
+		if sendNow := pktFDInfo != batchFDInfo; !sendNow {
+			batch = append(batch, pkt)
+			continue
+		}
+		n, err := e.sendBatch(batchFDInfo, batch)
+		sentPackets += n
+		if err != nil {
+			return sentPackets, err
+		}
+		batch = batch[:0]
+		batch = append(batch, pkt)
+		batchFDInfo = pktFDInfo
+	}
+
+	if len(batch) != 0 {
+		n, err := e.sendBatch(batchFDInfo, batch)
+		sentPackets += n
+		if err != nil {
+			return sentPackets, err
+		}
+	}
+	return sentPackets, nil
+}
+
+// dispatchLoop reads packets from the file descriptor in a loop and dispatches
+// them to the network stack.
+func (e *endpoint) dispatchLoop(inboundDispatcher linkDispatcher) tcpip.Error {
+	for {
+		cont, err := inboundDispatcher.dispatch()
+		if err != nil || !cont {
+			if e.closed != nil {
+				e.closed(err)
+			}
+			inboundDispatcher.release()
+			return err
+		}
+	}
+}
+
+// GSOMaxSize implements stack.GSOEndpoint.
+func (e *endpoint) GSOMaxSize() uint32 {
+	return 0
+}
+
+// SupportedGSO implements stack.GSOEndpoint.
+func (e *endpoint) SupportedGSO() stack.SupportedGSO {
+	return stack.GSONotSupported
+}
+
+// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType.
+func (e *endpoint) ARPHardwareType() header.ARPHardwareType {
+	if e.hdrSize > 0 {
+		return header.ARPHardwareEther
+	}
+	return header.ARPHardwareNone
+}
+
+// Close implements stack.LinkEndpoint.
+func (e *endpoint) Close() {}
+
+// SetOnCloseAction implements stack.LinkEndpoint.
+func (*endpoint) SetOnCloseAction(func()) {}
--- a/internal/fdbased_darwin/endpoint_mutex.go
+++ b/internal/fdbased_darwin/endpoint_mutex.go
@@ -0,0 +1,96 @@
+package fdbased
+
+import (
+	"reflect"
+
+	"github.com/sagernet/gvisor/pkg/sync"
+	"github.com/sagernet/gvisor/pkg/sync/locking"
+)
+
+// RWMutex is sync.RWMutex with the correctness validator.
+type endpointRWMutex struct {
+	mu sync.RWMutex
+}
+
+// lockNames is a list of user-friendly lock names.
+// Populated in init.
+var endpointlockNames []string
+
+// lockNameIndex is used as an index passed to NestedLock and NestedUnlock,
+// referring to an index within lockNames.
+// Values are specified using the "consts" field of go_template_instance.
+type endpointlockNameIndex int
+
+// DO NOT REMOVE: The following function automatically replaced with lock index constants.
+// LOCK_NAME_INDEX_CONSTANTS
+const ()
+
+// Lock locks m.
+// +checklocksignore
+func (m *endpointRWMutex) Lock() {
+	locking.AddGLock(endpointprefixIndex, -1)
+	m.mu.Lock()
+}
+
+// NestedLock locks m knowing that another lock of the same type is held.
+// +checklocksignore
+func (m *endpointRWMutex) NestedLock(i endpointlockNameIndex) {
+	locking.AddGLock(endpointprefixIndex, int(i))
+	m.mu.Lock()
+}
+
+// Unlock unlocks m.
+// +checklocksignore
+func (m *endpointRWMutex) Unlock() {
+	m.mu.Unlock()
+	locking.DelGLock(endpointprefixIndex, -1)
+}
+
+// NestedUnlock unlocks m knowing that another lock of the same type is held.
+// +checklocksignore
+func (m *endpointRWMutex) NestedUnlock(i endpointlockNameIndex) {
+	m.mu.Unlock()
+	locking.DelGLock(endpointprefixIndex, int(i))
+}
+
+// RLock locks m for reading.
+// +checklocksignore
+func (m *endpointRWMutex) RLock() {
+	locking.AddGLock(endpointprefixIndex, -1)
+	m.mu.RLock()
+}
+
+// RUnlock undoes a single RLock call.
+// +checklocksignore
+func (m *endpointRWMutex) RUnlock() {
+	m.mu.RUnlock()
+	locking.DelGLock(endpointprefixIndex, -1)
+}
+
+// RLockBypass locks m for reading without executing the validator.
+// +checklocksignore
+func (m *endpointRWMutex) RLockBypass() {
+	m.mu.RLock()
+}
+
+// RUnlockBypass undoes a single RLockBypass call.
+// +checklocksignore
+func (m *endpointRWMutex) RUnlockBypass() {
+	m.mu.RUnlock()
+}
+
+// DowngradeLock atomically unlocks rw for writing and locks it for reading.
+// +checklocksignore
+func (m *endpointRWMutex) DowngradeLock() {
+	m.mu.DowngradeLock()
+}
+
+var endpointprefixIndex *locking.MutexClass
+
+// DO NOT REMOVE: The following function is automatically replaced.
+func endpointinitLockNames() {}
+
+func init() {
+	endpointinitLockNames()
+	endpointprefixIndex = locking.NewMutexClass(reflect.TypeOf(endpointRWMutex{}), endpointlockNames)
+}
--- a/internal/fdbased_darwin/errno.go
+++ b/internal/fdbased_darwin/errno.go
@@ -0,0 +1,54 @@
+package fdbased
+
+import (
+	"github.com/sagernet/gvisor/pkg/tcpip"
+
+	"golang.org/x/sys/unix"
+)
+
+func TranslateErrno(e unix.Errno) tcpip.Error {
+	switch e {
+	case unix.EEXIST:
+		return &tcpip.ErrDuplicateAddress{}
+	case unix.ENETUNREACH:
+		return &tcpip.ErrHostUnreachable{}
+	case unix.EINVAL:
+		return &tcpip.ErrInvalidEndpointState{}
+	case unix.EALREADY:
+		return &tcpip.ErrAlreadyConnecting{}
+	case unix.EISCONN:
+		return &tcpip.ErrAlreadyConnected{}
+	case unix.EADDRINUSE:
+		return &tcpip.ErrPortInUse{}
+	case unix.EADDRNOTAVAIL:
+		return &tcpip.ErrBadLocalAddress{}
+	case unix.EPIPE:
+		return &tcpip.ErrClosedForSend{}
+	case unix.EWOULDBLOCK:
+		return &tcpip.ErrWouldBlock{}
+	case unix.ECONNREFUSED:
+		return &tcpip.ErrConnectionRefused{}
+	case unix.ETIMEDOUT:
+		return &tcpip.ErrTimeout{}
+	case unix.EINPROGRESS:
+		return &tcpip.ErrConnectStarted{}
+	case unix.EDESTADDRREQ:
+		return &tcpip.ErrDestinationRequired{}
+	case unix.ENOTSUP:
+		return &tcpip.ErrNotSupported{}
+	case unix.ENOTTY:
+		return &tcpip.ErrQueueSizeNotSupported{}
+	case unix.ENOTCONN:
+		return &tcpip.ErrNotConnected{}
+	case unix.ECONNRESET:
+		return &tcpip.ErrConnectionReset{}
+	case unix.ECONNABORTED:
+		return &tcpip.ErrConnectionAborted{}
+	case unix.EMSGSIZE:
+		return &tcpip.ErrMessageTooLong{}
+	case unix.ENOBUFS:
+		return &tcpip.ErrNoBufferSpace{}
+	default:
+		return &tcpip.ErrInvalidEndpointState{}
+	}
+}
--- a/internal/fdbased_darwin/packet_dispatchers.go
+++ b/internal/fdbased_darwin/packet_dispatchers.go
@@ -0,0 +1,229 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package fdbased
+
+import (
+	"github.com/sagernet/gvisor/pkg/buffer"
+	"github.com/sagernet/gvisor/pkg/tcpip"
+	"github.com/sagernet/gvisor/pkg/tcpip/stack"
+	"github.com/sagernet/gvisor/pkg/tcpip/stack/gro"
+	"github.com/sagernet/sing-tun/internal/rawfile_darwin"
+	"github.com/sagernet/sing-tun/internal/stopfd_darwin"
+
+	"golang.org/x/sys/unix"
+)
+
+// BufConfig defines the shape of the buffer used to read packets from the NIC.
+var BufConfig = []int{4, 128, 256, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768}
+
+// +stateify savable
+type iovecBuffer struct {
+	// buffer is the actual buffer that holds the packet contents. Some contents
+	// are reused across calls to pullBuffer if number of requested bytes is
+	// smaller than the number of bytes allocated in the buffer.
+	views []*buffer.View
+
+	// iovecs are initialized with base pointers/len of the corresponding
+	// entries in the views defined above, except when GSO is enabled
+	// (skipsVnetHdr) then the first iovec points to a buffer for the vnet header
+	// which is stripped before the views are passed up the stack for further
+	// processing.
+	iovecs []unix.Iovec `state:"nosave"`
+
+	// sizes is an array of buffer sizes for the underlying views. sizes is
+	// immutable.
+	sizes []int
+
+	// pulledIndex is the index of the last []byte buffer pulled from the
+	// underlying buffer storage during a call to pullBuffers. It is -1
+	// if no buffer is pulled.
+	pulledIndex int
+}
+
+func newIovecBuffer(sizes []int) *iovecBuffer {
+	b := &iovecBuffer{
+		views:  make([]*buffer.View, len(sizes)),
+		iovecs: make([]unix.Iovec, len(sizes)),
+		sizes:  sizes,
+	}
+	return b
+}
+
+func (b *iovecBuffer) nextIovecs() []unix.Iovec {
+	for i := range b.views {
+		if b.views[i] != nil {
+			break
+		}
+		v := buffer.NewViewSize(b.sizes[i])
+		b.views[i] = v
+		b.iovecs[i] = unix.Iovec{Base: v.BasePtr()}
+		b.iovecs[i].SetLen(v.Size())
+	}
+	return b.iovecs
+}
+
+// pullBuffer extracts the enough underlying storage from b.buffer to hold n
+// bytes. It removes this storage from b.buffer, returns a new buffer
+// that holds the storage, and updates pulledIndex to indicate which part
+// of b.buffer's storage must be reallocated during the next call to
+// nextIovecs.
+func (b *iovecBuffer) pullBuffer(n int) buffer.Buffer {
+	var views []*buffer.View
+	c := 0
+	// Remove the used views from the buffer.
+	for i, v := range b.views {
+		c += v.Size()
+		if c >= n {
+			b.views[i].CapLength(v.Size() - (c - n))
+			views = append(views, b.views[:i+1]...)
+			break
+		}
+	}
+	for i := range views {
+		b.views[i] = nil
+	}
+	pulled := buffer.Buffer{}
+	for _, v := range views {
+		pulled.Append(v)
+	}
+	pulled.Truncate(int64(n))
+	return pulled
+}
+
+func (b *iovecBuffer) release() {
+	for _, v := range b.views {
+		if v != nil {
+			v.Release()
+			v = nil
+		}
+	}
+}
+
+// recvMMsgDispatcher uses the recvmmsg system call to read inbound packets and
+// dispatches them.
+//
+// +stateify savable
+type recvMMsgDispatcher struct {
+	stopfd.StopFD
+	// fd is the file descriptor used to send and receive packets.
+	fd int
+
+	// e is the endpoint this dispatcher is attached to.
+	e *endpoint
+
+	// bufs is an array of iovec buffers that contain packet contents.
+	bufs []*iovecBuffer
+
+	// msgHdrs is an array of MMsgHdr objects where each MMsghdr is used to
+	// reference an array of iovecs in the iovecs field defined above.  This
+	// array is passed as the parameter to recvmmsg call to retrieve
+	// potentially more than 1 packet per unix.
+	msgHdrs []rawfile.MsgHdrX `state:"nosave"`
+
+	// pkts is reused to avoid allocations.
+	pkts stack.PacketBufferList
+
+	// gro coalesces incoming packets to increase throughput.
+	gro gro.GRO
+
+	// mgr is the processor goroutine manager.
+	mgr *processorManager
+}
+
+func newRecvMMsgDispatcher(fd int, e *endpoint, opts *Options) (linkDispatcher, error) {
+	stopFD, err := stopfd.New()
+	if err != nil {
+		return nil, err
+	}
+	batchSize := int((512*1024)/(opts.MTU)) + 1
+	d := &recvMMsgDispatcher{
+		StopFD:  stopFD,
+		fd:      fd,
+		e:       e,
+		bufs:    make([]*iovecBuffer, batchSize),
+		msgHdrs: make([]rawfile.MsgHdrX, batchSize),
+	}
+	bufConfig := []int{4, int(opts.MTU)}
+	for i := range d.bufs {
+		d.bufs[i] = newIovecBuffer(bufConfig)
+	}
+	d.gro.Init(false)
+	d.mgr = newProcessorManager(opts, e)
+	d.mgr.start()
+
+	return d, nil
+}
+
+func (d *recvMMsgDispatcher) release() {
+	for _, iov := range d.bufs {
+		iov.release()
+	}
+	d.mgr.close()
+}
+
+// recvMMsgDispatch reads more than one packet at a time from the file
+// descriptor and dispatches it.
+func (d *recvMMsgDispatcher) dispatch() (bool, tcpip.Error) {
+	// Fill message headers.
+	for k := range d.msgHdrs {
+		if d.msgHdrs[k].Msg.Iovlen > 0 {
+			break
+		}
+		iovecs := d.bufs[k].nextIovecs()
+		iovLen := len(iovecs)
+		d.msgHdrs[k].DataLen = 0
+		d.msgHdrs[k].Msg.Iov = &iovecs[0]
+		d.msgHdrs[k].Msg.SetIovlen(iovLen)
+	}
+
+	nMsgs, errno := rawfile.BlockingRecvMMsgUntilStopped(d.ReadFD, d.fd, d.msgHdrs)
+	if errno != 0 {
+		return false, TranslateErrno(errno)
+	}
+	if nMsgs == -1 {
+		return false, nil
+	}
+
+	// Process each of received packets.
+
+	d.e.mu.RLock()
+	addr := d.e.addr
+	dsp := d.e.dispatcher
+	d.e.mu.RUnlock()
+
+	d.gro.Dispatcher = dsp
+	defer d.pkts.Reset()
+
+	for k := 0; k < nMsgs; k++ {
+		n := int(d.msgHdrs[k].DataLen)
+		payload := d.bufs[k].pullBuffer(n)
+		payload.TrimFront(4)
+		pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+			Payload: payload,
+		})
+		d.pkts.PushBack(pkt)
+
+		// Mark that this iovec has been processed.
+		d.msgHdrs[k].Msg.Iovlen = 0
+
+		if d.e.parseInboundHeader(pkt, addr) {
+			pkt.RXChecksumValidated = d.e.caps&stack.CapabilityRXChecksumOffload != 0
+			d.mgr.queuePacket(pkt, d.e.hdrSize > 0)
+		}
+	}
+	d.mgr.wakeReady()
+
+	return true, nil
+}
--- a/internal/fdbased_darwin/processor_mutex.go
+++ b/internal/fdbased_darwin/processor_mutex.go
@@ -0,0 +1,64 @@
+package fdbased
+
+import (
+	"reflect"
+
+	"github.com/sagernet/gvisor/pkg/sync"
+	"github.com/sagernet/gvisor/pkg/sync/locking"
+)
+
+// Mutex is sync.Mutex with the correctness validator.
+type processorMutex struct {
+	mu sync.Mutex
+}
+
+var processorprefixIndex *locking.MutexClass
+
+// lockNames is a list of user-friendly lock names.
+// Populated in init.
+var processorlockNames []string
+
+// lockNameIndex is used as an index passed to NestedLock and NestedUnlock,
+// referring to an index within lockNames.
+// Values are specified using the "consts" field of go_template_instance.
+type processorlockNameIndex int
+
+// DO NOT REMOVE: The following function automatically replaced with lock index constants.
+// LOCK_NAME_INDEX_CONSTANTS
+const ()
+
+// Lock locks m.
+// +checklocksignore
+func (m *processorMutex) Lock() {
+	locking.AddGLock(processorprefixIndex, -1)
+	m.mu.Lock()
+}
+
+// NestedLock locks m knowing that another lock of the same type is held.
+// +checklocksignore
+func (m *processorMutex) NestedLock(i processorlockNameIndex) {
+	locking.AddGLock(processorprefixIndex, int(i))
+	m.mu.Lock()
+}
+
+// Unlock unlocks m.
+// +checklocksignore
+func (m *processorMutex) Unlock() {
+	locking.DelGLock(processorprefixIndex, -1)
+	m.mu.Unlock()
+}
+
+// NestedUnlock unlocks m knowing that another lock of the same type is held.
+// +checklocksignore
+func (m *processorMutex) NestedUnlock(i processorlockNameIndex) {
+	locking.DelGLock(processorprefixIndex, int(i))
+	m.mu.Unlock()
+}
+
+// DO NOT REMOVE: The following function is automatically replaced.
+func processorinitLockNames() {}
+
+func init() {
+	processorinitLockNames()
+	processorprefixIndex = locking.NewMutexClass(reflect.TypeOf(processorMutex{}), processorlockNames)
+}
--- a/internal/fdbased_darwin/processors.go
+++ b/internal/fdbased_darwin/processors.go
@@ -0,0 +1,275 @@
+// Copyright 2024 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package fdbased
+
+import (
+	"context"
+	"encoding/binary"
+
+	"github.com/sagernet/gvisor/pkg/rand"
+	"github.com/sagernet/gvisor/pkg/sleep"
+	"github.com/sagernet/gvisor/pkg/sync"
+	"github.com/sagernet/gvisor/pkg/tcpip"
+	"github.com/sagernet/gvisor/pkg/tcpip/hash/jenkins"
+	"github.com/sagernet/gvisor/pkg/tcpip/header"
+	"github.com/sagernet/gvisor/pkg/tcpip/stack"
+	"github.com/sagernet/gvisor/pkg/tcpip/stack/gro"
+)
+
+// +stateify savable
+type processor struct {
+	mu processorMutex `state:"nosave"`
+	// +checklocks:mu
+	pkts stack.PacketBufferList
+
+	e           *endpoint
+	gro         gro.GRO
+	sleeper     sleep.Sleeper
+	packetWaker sleep.Waker
+	closeWaker  sleep.Waker
+}
+
+func (p *processor) start(wg *sync.WaitGroup) {
+	defer wg.Done()
+	defer p.sleeper.Done()
+	for {
+		switch w := p.sleeper.Fetch(true); {
+		case w == &p.packetWaker:
+			p.deliverPackets()
+		case w == &p.closeWaker:
+			p.mu.Lock()
+			p.pkts.Reset()
+			p.mu.Unlock()
+			return
+		}
+	}
+}
+
+func (p *processor) deliverPackets() {
+	p.e.mu.RLock()
+	p.gro.Dispatcher = p.e.dispatcher
+	p.e.mu.RUnlock()
+	if p.gro.Dispatcher == nil {
+		p.mu.Lock()
+		p.pkts.Reset()
+		p.mu.Unlock()
+		return
+	}
+
+	p.mu.Lock()
+	for p.pkts.Len() > 0 {
+		pkt := p.pkts.PopFront()
+		p.mu.Unlock()
+		p.gro.Enqueue(pkt)
+		pkt.DecRef()
+		p.mu.Lock()
+	}
+	p.mu.Unlock()
+	p.gro.Flush()
+}
+
+// processorManager handles starting, closing, and queuing packets on processor
+// goroutines.
+//
+// +stateify savable
+type processorManager struct {
+	processors []processor
+	seed       uint32
+	wg         sync.WaitGroup `state:"nosave"`
+	e          *endpoint
+	ready      []bool
+}
+
+// newProcessorManager creates a new processor manager.
+func newProcessorManager(opts *Options, e *endpoint) *processorManager {
+	m := &processorManager{}
+	m.seed = rand.Uint32()
+	m.ready = make([]bool, opts.ProcessorsPerChannel)
+	m.processors = make([]processor, opts.ProcessorsPerChannel)
+	m.e = e
+	m.wg.Add(opts.ProcessorsPerChannel)
+
+	for i := range m.processors {
+		p := &m.processors[i]
+		p.sleeper.AddWaker(&p.packetWaker)
+		p.sleeper.AddWaker(&p.closeWaker)
+		p.gro.Init(false)
+		p.e = e
+	}
+
+	return m
+}
+
+// start starts the processor goroutines if the processor manager is configured
+// with more than one processor.
+func (m *processorManager) start() {
+	for i := range m.processors {
+		p := &m.processors[i]
+		// Only start processor in a separate goroutine if we have multiple of them.
+		if len(m.processors) > 1 {
+			go p.start(&m.wg)
+		}
+	}
+}
+
+// afterLoad is invoked by stateify.
+func (m *processorManager) afterLoad(context.Context) {
+	m.wg.Add(len(m.processors))
+	m.start()
+}
+
+func (m *processorManager) connectionHash(cid *connectionID) uint32 {
+	var payload [4]byte
+	binary.LittleEndian.PutUint16(payload[0:], cid.srcPort)
+	binary.LittleEndian.PutUint16(payload[2:], cid.dstPort)
+
+	h := jenkins.Sum32(m.seed)
+	h.Write(payload[:])
+	h.Write(cid.srcAddr)
+	h.Write(cid.dstAddr)
+	return h.Sum32()
+}
+
+// queuePacket queues a packet to be delivered to the appropriate processor.
+func (m *processorManager) queuePacket(pkt *stack.PacketBuffer, hasEthHeader bool) {
+	var pIdx uint32
+	cid, nonConnectionPkt := tcpipConnectionID(pkt)
+	if !hasEthHeader {
+		if nonConnectionPkt {
+			// If there's no eth header this should be a standard tcpip packet. If
+			// it isn't the packet is invalid so drop it.
+			return
+		}
+		pkt.NetworkProtocolNumber = cid.proto
+	}
+	if len(m.processors) == 1 || nonConnectionPkt {
+		// If the packet is not associated with an active connection, use the
+		// first processor.
+		pIdx = 0
+	} else {
+		pIdx = m.connectionHash(&cid) % uint32(len(m.processors))
+	}
+	p := &m.processors[pIdx]
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	p.pkts.PushBack(pkt.IncRef())
+	m.ready[pIdx] = true
+}
+
+type connectionID struct {
+	srcAddr, dstAddr []byte
+	srcPort, dstPort uint16
+	proto            tcpip.NetworkProtocolNumber
+}
+
+// tcpipConnectionID returns a tcpip connection id tuple based on the data found
+// in the packet. It returns true if the packet is not associated with an active
+// connection (e.g ARP, NDP, etc). The method assumes link headers have already
+// been processed if they were present.
+func tcpipConnectionID(pkt *stack.PacketBuffer) (connectionID, bool) {
+	var cid connectionID
+	h, ok := pkt.Data().PullUp(1)
+	if !ok {
+		// Skip this packet.
+		return cid, true
+	}
+
+	const tcpSrcDstPortLen = 4
+	switch header.IPVersion(h) {
+	case header.IPv4Version:
+		hdrLen := header.IPv4(h).HeaderLength()
+		h, ok = pkt.Data().PullUp(int(hdrLen) + tcpSrcDstPortLen)
+		if !ok {
+			return cid, true
+		}
+		ipHdr := header.IPv4(h[:hdrLen])
+		tcpHdr := header.TCP(h[hdrLen:][:tcpSrcDstPortLen])
+
+		cid.srcAddr = ipHdr.SourceAddressSlice()
+		cid.dstAddr = ipHdr.DestinationAddressSlice()
+		// All fragment packets need to be processed by the same goroutine, so
+		// only record the TCP ports if this is not a fragment packet.
+		if ipHdr.IsValid(pkt.Data().Size()) && !ipHdr.More() && ipHdr.FragmentOffset() == 0 {
+			cid.srcPort = tcpHdr.SourcePort()
+			cid.dstPort = tcpHdr.DestinationPort()
+		}
+		cid.proto = header.IPv4ProtocolNumber
+	case header.IPv6Version:
+		h, ok = pkt.Data().PullUp(header.IPv6FixedHeaderSize + tcpSrcDstPortLen)
+		if !ok {
+			return cid, true
+		}
+		ipHdr := header.IPv6(h)
+
+		var tcpHdr header.TCP
+		if tcpip.TransportProtocolNumber(ipHdr.NextHeader()) == header.TCPProtocolNumber {
+			tcpHdr = header.TCP(h[header.IPv6FixedHeaderSize:][:tcpSrcDstPortLen])
+		} else {
+			// Slow path for IPv6 extension headers :(.
+			dataBuf := pkt.Data().ToBuffer()
+			dataBuf.TrimFront(header.IPv6MinimumSize)
+			it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(ipHdr.NextHeader()), dataBuf)
+			defer it.Release()
+			for {
+				hdr, done, err := it.Next()
+				if done || err != nil {
+					break
+				}
+				hdr.Release()
+			}
+			h, ok = pkt.Data().PullUp(int(it.HeaderOffset()) + tcpSrcDstPortLen)
+			if !ok {
+				return cid, true
+			}
+			tcpHdr = header.TCP(h[it.HeaderOffset():][:tcpSrcDstPortLen])
+		}
+		cid.srcAddr = ipHdr.SourceAddressSlice()
+		cid.dstAddr = ipHdr.DestinationAddressSlice()
+		cid.srcPort = tcpHdr.SourcePort()
+		cid.dstPort = tcpHdr.DestinationPort()
+		cid.proto = header.IPv6ProtocolNumber
+	default:
+		return cid, true
+	}
+	return cid, false
+}
+
+func (m *processorManager) close() {
+	if len(m.processors) < 2 {
+		return
+	}
+	for i := range m.processors {
+		p := &m.processors[i]
+		p.closeWaker.Assert()
+	}
+}
+
+// wakeReady wakes up all processors that have a packet queued. If there is only
+// one processor, the method delivers the packet inline without waking a
+// goroutine.
+func (m *processorManager) wakeReady() {
+	for i, ready := range m.ready {
+		if !ready {
+			continue
+		}
+		p := &m.processors[i]
+		if len(m.processors) > 1 {
+			p.packetWaker.Assert()
+		} else {
+			p.deliverPackets()
+		}
+		m.ready[i] = false
+	}
+}
--- a/internal/rawfile_darwin/rawfile.go
+++ b/internal/rawfile_darwin/rawfile.go
@@ -0,0 +1,188 @@
+package rawfile
+
+import (
+	"reflect"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// SizeofIovec is the size of a unix.Iovec in bytes.
+const SizeofIovec = unsafe.Sizeof(unix.Iovec{})
+
+// MaxIovs is UIO_MAXIOV, the maximum number of iovecs that may be passed to a
+// host system call in a single array.
+const MaxIovs = 1024
+
+// IovecFromBytes returns a unix.Iovec representing bs.
+//
+// Preconditions: len(bs) > 0.
+func IovecFromBytes(bs []byte) unix.Iovec {
+	iov := unix.Iovec{
+		Base: &bs[0],
+	}
+	iov.SetLen(len(bs))
+	return iov
+}
+
+func bytesFromIovec(iov unix.Iovec) (bs []byte) {
+	sh := (*reflect.SliceHeader)(unsafe.Pointer(&bs))
+	sh.Data = uintptr(unsafe.Pointer(iov.Base))
+	sh.Len = int(iov.Len)
+	sh.Cap = int(iov.Len)
+	return
+}
+
+// AppendIovecFromBytes returns append(iovs, IovecFromBytes(bs)). If len(bs) ==
+// 0, AppendIovecFromBytes returns iovs without modification. If len(iovs) >=
+// max, AppendIovecFromBytes replaces the final iovec in iovs with one that
+// also includes the contents of bs. Note that this implies that
+// AppendIovecFromBytes is only usable when the returned iovec slice is used as
+// the source of a write.
+func AppendIovecFromBytes(iovs []unix.Iovec, bs []byte, max int) []unix.Iovec {
+	if len(bs) == 0 {
+		return iovs
+	}
+	if len(iovs) < max {
+		return append(iovs, IovecFromBytes(bs))
+	}
+	iovs[len(iovs)-1] = IovecFromBytes(append(bytesFromIovec(iovs[len(iovs)-1]), bs...))
+	return iovs
+}
+
+type MsgHdrX struct {
+	Msg     unix.Msghdr
+	DataLen uint32
+}
+
+func NonBlockingSendMMsg(fd int, msgHdrs []MsgHdrX) (int, unix.Errno) {
+	n, _, e := unix.RawSyscall6(unix.SYS_SENDMSG_X, uintptr(fd), uintptr(unsafe.Pointer(&msgHdrs[0])), uintptr(len(msgHdrs)), unix.MSG_DONTWAIT, 0, 0)
+	return int(n), e
+}
+
+const SizeofMsgHdrX = unsafe.Sizeof(MsgHdrX{})
+
+// NonBlockingWriteIovec writes iovec to a file descriptor in a single unix.
+// It fails if partial data is written.
+func NonBlockingWriteIovec(fd int, iovec []unix.Iovec) unix.Errno {
+	iovecLen := uintptr(len(iovec))
+	_, _, e := unix.RawSyscall(unix.SYS_WRITEV, uintptr(fd), uintptr(unsafe.Pointer(&iovec[0])), iovecLen)
+	return e
+}
+
+func BlockingReadvUntilStopped(efd int, fd int, iovecs []unix.Iovec) (int, unix.Errno) {
+	for {
+		n, _, e := unix.RawSyscall(unix.SYS_READV, uintptr(fd), uintptr(unsafe.Pointer(&iovecs[0])), uintptr(len(iovecs)))
+		if e == 0 {
+			return int(n), 0
+		}
+		if e != 0 && e != unix.EWOULDBLOCK {
+			return 0, e
+		}
+		stopped, e := BlockingPollUntilStopped(efd, fd, unix.POLLIN)
+		if stopped {
+			return -1, e
+		}
+		if e != 0 && e != unix.EINTR {
+			return 0, e
+		}
+	}
+}
+
+func BlockingRecvMMsgUntilStopped(efd int, fd int, msgHdrs []MsgHdrX) (int, unix.Errno) {
+	for {
+		n, _, e := unix.RawSyscall6(unix.SYS_RECVMSG_X, uintptr(fd), uintptr(unsafe.Pointer(&msgHdrs[0])), uintptr(len(msgHdrs)), unix.MSG_DONTWAIT, 0, 0)
+		if e == 0 {
+			return int(n), e
+		}
+
+		if e != 0 && e != unix.EWOULDBLOCK {
+			return 0, e
+		}
+
+		stopped, e := BlockingPollUntilStopped(efd, fd, unix.POLLIN)
+		if stopped {
+			return -1, e
+		}
+		if e != 0 && e != unix.EINTR {
+			return 0, e
+		}
+	}
+}
+
+func BlockingPollUntilStopped(efd int, fd int, events int16) (bool, unix.Errno) {
+	// Create kqueue
+	kq, err := unix.Kqueue()
+	if err != nil {
+		return false, unix.Errno(err.(unix.Errno))
+	}
+	defer unix.Close(kq)
+
+	// Prepare kevents for registration
+	var kevents []unix.Kevent_t
+
+	// Always monitor efd for read events
+	kevents = append(kevents, unix.Kevent_t{
+		Ident:  uint64(efd),
+		Filter: unix.EVFILT_READ,
+		Flags:  unix.EV_ADD | unix.EV_ENABLE,
+	})
+
+	// Monitor fd based on requested events
+	// Convert poll events to kqueue filters
+	if events&unix.POLLIN != 0 {
+		kevents = append(kevents, unix.Kevent_t{
+			Ident:  uint64(fd),
+			Filter: unix.EVFILT_READ,
+			Flags:  unix.EV_ADD | unix.EV_ENABLE,
+		})
+	}
+	if events&unix.POLLOUT != 0 {
+		kevents = append(kevents, unix.Kevent_t{
+			Ident:  uint64(fd),
+			Filter: unix.EVFILT_WRITE,
+			Flags:  unix.EV_ADD | unix.EV_ENABLE,
+		})
+	}
+
+	// Register events
+	_, err = unix.Kevent(kq, kevents, nil, nil)
+	if err != nil {
+		return false, unix.Errno(err.(unix.Errno))
+	}
+
+	// Wait for events (blocking)
+	revents := make([]unix.Kevent_t, len(kevents))
+	n, err := unix.Kevent(kq, nil, revents, nil)
+	if err != nil {
+		return false, unix.Errno(err.(unix.Errno))
+	}
+
+	// Check results
+	var efdHasData bool
+	var errno unix.Errno
+
+	for i := 0; i < n; i++ {
+		ev := &revents[i]
+
+		if int(ev.Ident) == efd && ev.Filter == unix.EVFILT_READ {
+			efdHasData = true
+		}
+
+		if int(ev.Ident) == fd {
+			// Check for errors or EOF
+			if ev.Flags&unix.EV_EOF != 0 {
+				errno = unix.ECONNRESET
+			} else if ev.Flags&unix.EV_ERROR != 0 {
+				// Extract error from Data field
+				if ev.Data != 0 {
+					errno = unix.Errno(ev.Data)
+				} else {
+					errno = unix.ECONNRESET
+				}
+			}
+		}
+	}
+
+	return efdHasData, errno
+}
--- a/internal/stopfd_darwin/stopfd.go
+++ b/internal/stopfd_darwin/stopfd.go
@@ -0,0 +1,61 @@
+package stopfd
+
+import (
+	"fmt"
+
+	"golang.org/x/sys/unix"
+)
+
+type StopFD struct {
+	ReadFD  int
+	WriteFD int
+}
+
+func New() (StopFD, error) {
+	fds := make([]int, 2)
+	err := unix.Pipe(fds)
+	if err != nil {
+		return StopFD{ReadFD: -1, WriteFD: -1}, fmt.Errorf("failed to create pipe: %w", err)
+	}
+
+	if err := unix.SetNonblock(fds[0], true); err != nil {
+		unix.Close(fds[0])
+		unix.Close(fds[1])
+		return StopFD{ReadFD: -1, WriteFD: -1}, fmt.Errorf("failed to set read end non-blocking: %w", err)
+	}
+
+	if err := unix.SetNonblock(fds[1], true); err != nil {
+		unix.Close(fds[0])
+		unix.Close(fds[1])
+		return StopFD{ReadFD: -1, WriteFD: -1}, fmt.Errorf("failed to set write end non-blocking: %w", err)
+	}
+
+	return StopFD{ReadFD: fds[0], WriteFD: fds[1]}, nil
+}
+
+func (sf *StopFD) Stop() {
+	signal := []byte{1}
+	if n, err := unix.Write(sf.WriteFD, signal); n != len(signal) || err != nil {
+		panic(fmt.Sprintf("write(WriteFD) = (%d, %s), want (%d, nil)", n, err, len(signal)))
+	}
+}
+
+func (sf *StopFD) Close() error {
+	var err1, err2 error
+	if sf.ReadFD != -1 {
+		err1 = unix.Close(sf.ReadFD)
+		sf.ReadFD = -1
+	}
+	if sf.WriteFD != -1 {
+		err2 = unix.Close(sf.WriteFD)
+		sf.WriteFD = -1
+	}
+	if err1 != nil {
+		return err1
+	}
+	return err2
+}
+
+func (sf *StopFD) EFD() int {
+	return sf.ReadFD
+}
--- a/stack_gvisor.go
+++ b/stack_gvisor.go
@@ -40,7 +40,7 @@ type GVisor struct {

 type GVisorTun interface {
 	Tun
-	NewEndpoint() (stack.LinkEndpoint, error)
+	NewEndpoint() (stack.LinkEndpoint, stack.NICOptions, error)
 }

 func NewGVisor(
@@ -65,12 +65,12 @@ func NewGVisor(
 }

 func (t *GVisor) Start() error {
-	linkEndpoint, err := t.tun.NewEndpoint()
+	linkEndpoint, nicOptions, err := t.tun.NewEndpoint()
 	if err != nil {
 		return err
 	}
 	linkEndpoint = &LinkEndpointFilter{linkEndpoint, t.broadcastAddr, t.tun}
-	ipStack, err := NewGVisorStack(linkEndpoint)
+	ipStack, err := NewGVisorStackWithOptions(linkEndpoint, nicOptions)
 	if err != nil {
 		return err
 	}
@@ -110,6 +110,10 @@ func AddrFromAddress(address tcpip.Address) netip.Addr {
 }

 func NewGVisorStack(ep stack.LinkEndpoint) (*stack.Stack, error) {
+	return NewGVisorStackWithOptions(ep, stack.NICOptions{})
+}
+
+func NewGVisorStackWithOptions(ep stack.LinkEndpoint, opts stack.NICOptions) (*stack.Stack, error) {
 	ipStack := stack.New(stack.Options{
 		NetworkProtocols: []stack.NetworkProtocolFactory{
 			ipv4.NewProtocol,
@@ -122,7 +126,7 @@ func NewGVisorStack(ep stack.LinkEndpoint) (*stack.Stack, error) {
 			icmp.NewProtocol6,
 		},
 	})
-	err := ipStack.CreateNIC(DefaultNIC, ep)
+	err := ipStack.CreateNICWithOptions(DefaultNIC, ep, opts)
 	if err != nil {
 		return nil, gonet.TranslateNetstackError(err)
 	}
--- a/stack_mixed.go
+++ b/stack_mixed.go
@@ -10,6 +10,7 @@ import (
 	"github.com/sagernet/gvisor/pkg/tcpip/stack"
 	"github.com/sagernet/gvisor/pkg/tcpip/transport/udp"
 	"github.com/sagernet/sing-tun/internal/gtcpip/header"
+	"github.com/sagernet/sing/common/buf"
 	"github.com/sagernet/sing/common/bufio"
 	E "github.com/sagernet/sing/common/exceptions"
 )
@@ -72,10 +73,14 @@ func (m *Mixed) tunLoop() {
 		m.txChecksumOffload = linuxTUN.TXChecksumOffload()
 		batchSize := linuxTUN.BatchSize()
 		if batchSize > 1 {
-			m.batchLoop(linuxTUN, batchSize)
+			m.batchLoopLinux(linuxTUN, batchSize)
 			return
 		}
 	}
+	if darwinTUN, isDarwinTUN := m.tun.(DarwinTUN); isDarwinTUN {
+		m.batchLoopDarwin(darwinTUN)
+		return
+	}
 	packetBuffer := make([]byte, m.mtu+PacketOffset)
 	for {
 		n, err := m.tun.Read(packetBuffer)
@@ -119,12 +124,12 @@ func (m *Mixed) wintunLoop(winTun WinTun) {
 	}
 }

-func (m *Mixed) batchLoop(linuxTUN LinuxTUN, batchSize int) {
+func (m *Mixed) batchLoopLinux(linuxTUN LinuxTUN, batchSize int) {
 	packetBuffers := make([][]byte, batchSize)
 	writeBuffers := make([][]byte, batchSize)
 	packetSizes := make([]int, batchSize)
 	for i := range packetBuffers {
-		packetBuffers[i] = make([]byte, m.mtu+m.frontHeadroom)
+		packetBuffers[i] = make([]byte, m.mtu+PacketOffset+m.frontHeadroom)
 	}
 	for {
 		n, err := linuxTUN.BatchRead(packetBuffers, m.frontHeadroom, packetSizes)
@@ -158,6 +163,40 @@ func (m *Mixed) batchLoop(linuxTUN LinuxTUN, batchSize int) {
 	}
 }

+func (m *Mixed) batchLoopDarwin(darwinTUN DarwinTUN) {
+	var writeBuffers []*buf.Buffer
+	for {
+		buffers, err := darwinTUN.BatchRead()
+		if err != nil {
+			if E.IsClosed(err) {
+				return
+			}
+			m.logger.Error(E.Cause(err, "batch read packet"))
+		}
+		if len(buffers) == 0 {
+			continue
+		}
+		writeBuffers = writeBuffers[:0]
+		for _, buffer := range buffers {
+			packetSize := buffer.Len()
+			if packetSize < header.IPv4MinimumSize {
+				continue
+			}
+			if m.processPacket(buffer.Bytes()) {
+				writeBuffers = append(writeBuffers, buffer)
+			} else {
+				buffer.Release()
+			}
+		}
+		if len(writeBuffers) > 0 {
+			err = darwinTUN.BatchWrite(writeBuffers)
+			if err != nil {
+				m.logger.Trace(E.Cause(err, "batch write packet"))
+			}
+		}
+	}
+}
+
 func (m *Mixed) processPacket(packet []byte) bool {
 	var (
 		writeBack bool
--- a/stack_system.go
+++ b/stack_system.go
@@ -170,10 +170,14 @@ func (s *System) tunLoop() {
 		s.txChecksumOffload = linuxTUN.TXChecksumOffload()
 		batchSize := linuxTUN.BatchSize()
 		if batchSize > 1 {
-			s.batchLoop(linuxTUN, batchSize)
+			s.batchLoopLinux(linuxTUN, batchSize)
 			return
 		}
 	}
+	if darwinTUN, isDarwinTUN := s.tun.(DarwinTUN); isDarwinTUN {
+		s.batchLoopDarwin(darwinTUN)
+		return
+	}
 	packetBuffer := make([]byte, s.mtu+PacketOffset)
 	for {
 		n, err := s.tun.Read(packetBuffer)
@@ -217,7 +221,7 @@ func (s *System) wintunLoop(winTun WinTun) {
 	}
 }

-func (s *System) batchLoop(linuxTUN LinuxTUN, batchSize int) {
+func (s *System) batchLoopLinux(linuxTUN LinuxTUN, batchSize int) {
 	packetBuffers := make([][]byte, batchSize)
 	writeBuffers := make([][]byte, batchSize)
 	packetSizes := make([]int, batchSize)
@@ -256,6 +260,40 @@ func (s *System) batchLoop(linuxTUN LinuxTUN, batchSize int) {
 	}
 }

+func (s *System) batchLoopDarwin(darwinTUN DarwinTUN) {
+	var writeBuffers []*buf.Buffer
+	for {
+		buffers, err := darwinTUN.BatchRead()
+		if err != nil {
+			if E.IsClosed(err) {
+				return
+			}
+			s.logger.Error(E.Cause(err, "batch read packet"))
+		}
+		if len(buffers) == 0 {
+			continue
+		}
+		writeBuffers = writeBuffers[:0]
+		for _, buffer := range buffers {
+			packetSize := buffer.Len()
+			if packetSize < header.IPv4MinimumSize {
+				continue
+			}
+			if s.processPacket(buffer.Bytes()) {
+				writeBuffers = append(writeBuffers, buffer)
+			} else {
+				buffer.Release()
+			}
+		}
+		if len(writeBuffers) > 0 {
+			err = darwinTUN.BatchWrite(writeBuffers)
+			if err != nil {
+				s.logger.Trace(E.Cause(err, "batch write packet"))
+			}
+		}
+	}
+}
+
 func (s *System) processPacket(packet []byte) bool {
 	var (
 		writeBack bool
--- a/tun.go
+++ b/tun.go
@@ -8,6 +8,7 @@ import (
 	"strconv"
 	"strings"

+	"github.com/sagernet/sing/common/buf"
 	"github.com/sagernet/sing/common/control"
 	F "github.com/sagernet/sing/common/format"
 	"github.com/sagernet/sing/common/logger"
@@ -45,6 +46,12 @@ type LinuxTUN interface {
 	TXChecksumOffload() bool
 }

+type DarwinTUN interface {
+	Tun
+	BatchRead() ([]*buf.Buffer, error)
+	BatchWrite(buffers []*buf.Buffer) error
+}
+
 const (
 	DefaultIPRoute2TableIndex = 2022
 	DefaultIPRoute2RuleIndex  = 9000
--- a/tun_darwin.go
+++ b/tun_darwin.go
@@ -10,6 +10,8 @@ import (
 	"unsafe"

 	"github.com/sagernet/sing-tun/internal/gtcpip/header"
+	"github.com/sagernet/sing-tun/internal/rawfile_darwin"
+	"github.com/sagernet/sing-tun/internal/stopfd_darwin"
 	"github.com/sagernet/sing/common"
 	"github.com/sagernet/sing/common/buf"
 	"github.com/sagernet/sing/common/bufio"
@@ -21,15 +23,64 @@ import (
 	"golang.org/x/sys/unix"
 )

+var _ DarwinTUN = (*NativeTun)(nil)
+
 const PacketOffset = 4

 type NativeTun struct {
-	tunFile      *os.File
-	tunWriter    N.VectorisedWriter
-	options      Options
-	inet4Address [4]byte
-	inet6Address [16]byte
-	routeSet     bool
+	tunFd         int
+	tunFile       *os.File
+	batchSize     int
+	iovecs        []iovecBuffer
+	iovecsOutput  []iovecBuffer
+	msgHdrs       []rawfile.MsgHdrX
+	msgHdrsOutput []rawfile.MsgHdrX
+	buffers       []*buf.Buffer
+	stopFd        stopfd.StopFD
+	tunWriter     N.VectorisedWriter
+	options       Options
+	inet4Address  [4]byte
+	inet6Address  [16]byte
+	routeSet      bool
+}
+
+type iovecBuffer struct {
+	mtu    int
+	buffer *buf.Buffer
+	iovecs []unix.Iovec
+}
+
+func newIovecBuffer(mtu int) iovecBuffer {
+	return iovecBuffer{
+		mtu:    mtu,
+		iovecs: make([]unix.Iovec, 2),
+	}
+}
+
+func (b *iovecBuffer) nextIovecs() []unix.Iovec {
+	if b.iovecs[0].Len == 0 {
+		headBuffer := make([]byte, PacketOffset)
+		b.iovecs[0].Base = &headBuffer[0]
+		b.iovecs[0].SetLen(PacketOffset)
+	}
+	if b.buffer == nil {
+		b.buffer = buf.NewSize(b.mtu)
+		b.iovecs[1].Base = &b.buffer.FreeBytes()[0]
+		b.iovecs[1].SetLen(b.mtu)
+	}
+	return b.iovecs
+}
+
+func (b *iovecBuffer) nextIovecsOutput(buffer *buf.Buffer) []unix.Iovec {
+	switch header.IPVersion(buffer.Bytes()) {
+	case header.IPv4Version:
+		b.iovecs[0] = packetHeaderVec4
+	case header.IPv6Version:
+		b.iovecs[0] = packetHeaderVec6
+	}
+	b.iovecs[1].Base = &buffer.Bytes()[0]
+	b.iovecs[1].SetLen(buffer.Len())
+	return b.iovecs
 }

 func (t *NativeTun) Name() (string, error) {
@@ -42,6 +93,7 @@ func (t *NativeTun) Name() (string, error) {

 func New(options Options) (Tun, error) {
 	var tunFd int
+	batchSize := ((512 * 1024) / int(options.MTU)) + 1
 	if options.FileDescriptor == 0 {
 		ifIndex := -1
 		_, err := fmt.Sscanf(options.Name, "utun%d", &ifIndex)
@@ -54,18 +106,37 @@ func New(options Options) (Tun, error) {
 			return nil, err
 		}

-		err = configure(tunFd, ifIndex, options.Name, options)
+		err = create(tunFd, ifIndex, options.Name, options)
+		if err != nil {
+			unix.Close(tunFd)
+			return nil, err
+		}
+		err = configure(tunFd, batchSize)
 		if err != nil {
 			unix.Close(tunFd)
 			return nil, err
 		}
 	} else {
 		tunFd = options.FileDescriptor
+		err := configure(tunFd, batchSize)
+		if err != nil {
+			return nil, err
+		}
 	}
-
 	nativeTun := &NativeTun{
-		tunFile: os.NewFile(uintptr(tunFd), "utun"),
-		options: options,
+		tunFd:         tunFd,
+		tunFile:       os.NewFile(uintptr(tunFd), "utun"),
+		options:       options,
+		batchSize:     batchSize,
+		iovecs:        make([]iovecBuffer, batchSize),
+		iovecsOutput:  make([]iovecBuffer, batchSize),
+		msgHdrs:       make([]rawfile.MsgHdrX, batchSize),
+		msgHdrsOutput: make([]rawfile.MsgHdrX, batchSize),
+		stopFd:        common.Must1(stopfd.New()),
+	}
+	for i := 0; i < batchSize; i++ {
+		nativeTun.iovecs[i] = newIovecBuffer(int(options.MTU))
+		nativeTun.iovecsOutput[i] = newIovecBuffer(int(options.MTU))
 	}
 	if len(options.Inet4Address) > 0 {
 		nativeTun.inet4Address = options.Inet4Address[0].Addr().As4()
@@ -100,10 +171,17 @@ func (t *NativeTun) Write(p []byte) (n int, err error) {
 }

 var (
-	packetHeader4 = [4]byte{0x00, 0x00, 0x00, unix.AF_INET}
-	packetHeader6 = [4]byte{0x00, 0x00, 0x00, unix.AF_INET6}
+	packetHeader4    = []byte{0x00, 0x00, 0x00, unix.AF_INET}
+	packetHeader6    = []byte{0x00, 0x00, 0x00, unix.AF_INET6}
+	packetHeaderVec4 = unix.Iovec{Base: &packetHeader4[0]}
+	packetHeaderVec6 = unix.Iovec{Base: &packetHeader6[0]}
 )

+func init() {
+	packetHeaderVec4.SetLen(4)
+	packetHeaderVec6.SetLen(4)
+}
+
 func (t *NativeTun) WriteVectorised(buffers []*buf.Buffer) error {
 	var packetHeader []byte
 	switch header.IPVersion(buffers[0].Bytes()) {
@@ -147,7 +225,7 @@ type addrLifetime6 struct {
 	Pltime    uint32
 }

-func configure(tunFd int, ifIndex int, name string, options Options) error {
+func create(tunFd int, ifIndex int, name string, options Options) error {
 	ctlInfo := &unix.CtlInfo{}
 	copy(ctlInfo.Name[:], utunControlName)
 	err := unix.IoctlCtlInfo(tunFd, ctlInfo)
@@ -163,11 +241,6 @@ func configure(tunFd int, ifIndex int, name string, options Options) error {
 		return os.NewSyscallError("Connect", err)
 	}

-	err = unix.SetNonblock(tunFd, true)
-	if err != nil {
-		return os.NewSyscallError("SetNonblock", err)
-	}
-
 	err = useSocket(unix.AF_INET, unix.SOCK_DGRAM, 0, func(socketFd int) error {
 		var ifr unix.IfreqMTU
 		copy(ifr.Name[:], name)
@@ -259,6 +332,65 @@ func configure(tunFd int, ifIndex int, name string, options Options) error {
 	return nil
 }

+func configure(tunFd int, batchSize int) error {
+	err := unix.SetNonblock(tunFd, true)
+	if err != nil {
+		return os.NewSyscallError("SetNonblock", err)
+	}
+	const UTUN_OPT_MAX_PENDING_PACKETS = 16
+	err = unix.SetsockoptInt(tunFd, 2, UTUN_OPT_MAX_PENDING_PACKETS, batchSize)
+	if err != nil {
+		return os.NewSyscallError("SetsockoptInt UTUN_OPT_MAX_PENDING_PACKETS", err)
+	}
+	return nil
+}
+
+func (t *NativeTun) BatchSize() int {
+	return t.batchSize
+}
+
+func (t *NativeTun) BatchRead() ([]*buf.Buffer, error) {
+	for i := 0; i < t.batchSize; i++ {
+		iovecs := t.iovecs[i].nextIovecs()
+		t.msgHdrs[i].DataLen = 0
+		t.msgHdrs[i].Msg.Iov = &iovecs[0]
+		t.msgHdrs[i].Msg.Iovlen = 2
+	}
+	n, errno := rawfile.BlockingRecvMMsgUntilStopped(t.stopFd.ReadFD, t.tunFd, t.msgHdrs)
+	if errno != 0 {
+		return nil, errno
+	}
+	if n < 1 {
+		return nil, nil
+	}
+	buffers := t.buffers
+	for k := 0; k < n; k++ {
+		buffer := t.iovecs[k].buffer
+		t.iovecs[k].buffer = nil
+		buffer.Truncate(int(t.msgHdrs[k].DataLen) - PacketOffset)
+		buffers = append(buffers, buffer)
+	}
+	t.buffers = buffers[:0]
+	return buffers, nil
+}
+
+func (t *NativeTun) BatchWrite(buffers []*buf.Buffer) error {
+	for i, buffer := range buffers {
+		iovecs := t.iovecsOutput[i].nextIovecsOutput(buffer)
+		t.msgHdrsOutput[i].Msg.Iov = &iovecs[0]
+		t.msgHdrsOutput[i].Msg.Iovlen = 2
+	}
+	_, errno := rawfile.NonBlockingSendMMsg(t.tunFd, t.msgHdrsOutput[:len(buffers)])
+	if errno != 0 {
+		return errno
+	}
+	return nil
+}
+
+func (t *NativeTun) TXChecksumOffload() bool {
+	return false
+}
+
 func (t *NativeTun) UpdateRouteOptions(tunOptions Options) error {
 	err := t.unsetRoutes()
 	if err != nil {
--- a/tun_darwin_gvisor.go
+++ b/tun_darwin_gvisor.go
@@ -3,132 +3,23 @@
 package tun

 import (
-	"github.com/sagernet/gvisor/pkg/buffer"
-	"github.com/sagernet/gvisor/pkg/tcpip"
-	"github.com/sagernet/gvisor/pkg/tcpip/header"
+	"github.com/sagernet/gvisor/pkg/tcpip/link/qdisc/fifo"
 	"github.com/sagernet/gvisor/pkg/tcpip/stack"
-	"github.com/sagernet/sing/common/bufio"
+	"github.com/sagernet/sing-tun/internal/fdbased_darwin"
 )

 var _ GVisorTun = (*NativeTun)(nil)

-func (t *NativeTun) NewEndpoint() (stack.LinkEndpoint, error) {
-	return &DarwinEndpoint{tun: t}, nil
-}
-
-var _ stack.LinkEndpoint = (*DarwinEndpoint)(nil)
-
-type DarwinEndpoint struct {
-	tun        *NativeTun
-	dispatcher stack.NetworkDispatcher
-}
-
-func (e *DarwinEndpoint) MTU() uint32 {
-	return e.tun.options.MTU
-}
-
-func (e *DarwinEndpoint) SetMTU(mtu uint32) {
-}
-
-func (e *DarwinEndpoint) MaxHeaderLength() uint16 {
-	return 0
-}
-
-func (e *DarwinEndpoint) LinkAddress() tcpip.LinkAddress {
-	return ""
-}
-
-func (e *DarwinEndpoint) SetLinkAddress(addr tcpip.LinkAddress) {
-}
-
-func (e *DarwinEndpoint) Capabilities() stack.LinkEndpointCapabilities {
-	return stack.CapabilityRXChecksumOffload
-}
-
-func (e *DarwinEndpoint) Attach(dispatcher stack.NetworkDispatcher) {
-	if dispatcher == nil && e.dispatcher != nil {
-		e.dispatcher = nil
-		return
-	}
-	if dispatcher != nil && e.dispatcher == nil {
-		e.dispatcher = dispatcher
-		go e.dispatchLoop()
+func (t *NativeTun) NewEndpoint() (stack.LinkEndpoint, stack.NICOptions, error) {
+	ep, err := fdbased.New(&fdbased.Options{
+		FDs:               []int{t.tunFd},
+		MTU:               t.options.MTU,
+		RXChecksumOffload: true,
+	})
+	if err != nil {
+		return nil, stack.NICOptions{}, err
 	}
-}
-
-func (e *DarwinEndpoint) dispatchLoop() {
-	packetBuffer := make([]byte, e.tun.options.MTU+PacketOffset)
-	for {
-		n, err := e.tun.tunFile.Read(packetBuffer)
-		if err != nil {
-			break
-		}
-		packet := packetBuffer[PacketOffset:n]
-		var networkProtocol tcpip.NetworkProtocolNumber
-		switch header.IPVersion(packet) {
-		case header.IPv4Version:
-			networkProtocol = header.IPv4ProtocolNumber
-			if header.IPv4(packet).DestinationAddress().As4() == e.tun.inet4Address {
-				e.tun.tunFile.Write(packetBuffer[:n])
-				continue
-			}
-		case header.IPv6Version:
-			networkProtocol = header.IPv6ProtocolNumber
-			if header.IPv6(packet).DestinationAddress().As16() == e.tun.inet6Address {
-				e.tun.tunFile.Write(packetBuffer[:n])
-				continue
-			}
-		default:
-			e.tun.tunFile.Write(packetBuffer[:n])
-			continue
-		}
-		pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
-			Payload:           buffer.MakeWithData(packetBuffer[4:n]),
-			IsForwardedPacket: true,
-		})
-		pkt.NetworkProtocolNumber = networkProtocol
-		dispatcher := e.dispatcher
-		if dispatcher == nil {
-			pkt.DecRef()
-			return
-		}
-		dispatcher.DeliverNetworkPacket(networkProtocol, pkt)
-		pkt.DecRef()
-	}
-}
-
-func (e *DarwinEndpoint) IsAttached() bool {
-	return e.dispatcher != nil
-}
-
-func (e *DarwinEndpoint) Wait() {
-}
-
-func (e *DarwinEndpoint) ARPHardwareType() header.ARPHardwareType {
-	return header.ARPHardwareNone
-}
-
-func (e *DarwinEndpoint) AddHeader(buffer *stack.PacketBuffer) {
-}
-
-func (e *DarwinEndpoint) ParseHeader(ptr *stack.PacketBuffer) bool {
-	return true
-}
-
-func (e *DarwinEndpoint) WritePackets(packetBufferList stack.PacketBufferList) (int, tcpip.Error) {
-	var n int
-	for _, packet := range packetBufferList.AsSlice() {
-		_, err := bufio.WriteVectorised(e.tun, packet.AsSlices())
-		if err != nil {
-			return n, &tcpip.ErrAborted{}
-		}
-		n++
-	}
-	return n, nil
-}
-
-func (e *DarwinEndpoint) Close() {
-}
-
-func (e *DarwinEndpoint) SetOnCloseAction(f func()) {
+	return ep, stack.NICOptions{
+		QDisc: fifo.New(ep, 1, 1000),
+	}, nil
 }
--- a/tun_linux_gvisor.go
+++ b/tun_linux_gvisor.go
@@ -9,9 +9,9 @@ import (

 var _ GVisorTun = (*NativeTun)(nil)

-func (t *NativeTun) NewEndpoint() (stack.LinkEndpoint, error) {
+func (t *NativeTun) NewEndpoint() (stack.LinkEndpoint, stack.NICOptions, error) {
 	if t.vnetHdr {
-		return fdbased.New(&fdbased.Options{
+		ep, err := fdbased.New(&fdbased.Options{
 			FDs:               []int{t.tunFd},
 			MTU:               t.options.MTU,
 			GSOMaxSize:        gsoMaxSize,
@@ -19,11 +19,20 @@ func (t *NativeTun) NewEndpoint() (stack.LinkEndpoint, error) {
 			RXChecksumOffload: true,
 			TXChecksumOffload: t.txChecksumOffload,
 		})
+		if err != nil {
+			return nil, stack.NICOptions{}, err
+		}
+		return ep, stack.NICOptions{}, nil
+	} else {
+		ep, err := fdbased.New(&fdbased.Options{
+			FDs:               []int{t.tunFd},
+			MTU:               t.options.MTU,
+			RXChecksumOffload: true,
+			TXChecksumOffload: t.txChecksumOffload,
+		})
+		if err != nil {
+			return nil, stack.NICOptions{}, err
+		}
+		return ep, stack.NICOptions{}, nil
 	}
-	return fdbased.New(&fdbased.Options{
-		FDs:               []int{t.tunFd},
-		MTU:               t.options.MTU,
-		RXChecksumOffload: true,
-		TXChecksumOffload: t.txChecksumOffload,
-	})
 }
--- a/tun_windows_gvisor.go
+++ b/tun_windows_gvisor.go
@@ -11,8 +11,8 @@ import (

 var _ GVisorTun = (*NativeTun)(nil)

-func (t *NativeTun) NewEndpoint() (stack.LinkEndpoint, error) {
-	return &WintunEndpoint{tun: t}, nil
+func (t *NativeTun) NewEndpoint() (stack.LinkEndpoint, stack.NICOptions, error) {
+	return &WintunEndpoint{tun: t}, stack.NICOptions{}, nil
 }

 var _ stack.LinkEndpoint = (*WintunEndpoint)(nil)