Files
3x-ui/internal/amneziawgnet/netstack.go
T
Kuzz007 3d4fda9a0d Drain buffered TUN packets in one Read call, not just one at a time
stackTun.Read always returned exactly one packet per call regardless
of how many the caller's buf could hold. amneziawg-go's
RoutineReadFromTUN sizes its buffers to device.BatchSize(), which on
Linux is the UDP bind's own batch size (128, conn.IdealBatchSize) --
so real batch capacity was already there and going unused on the
download path.

The UDP bind's Send/Receive both genuinely batch via recvmmsg/
sendmmsg (conn/bind_std.go, confirmed in vendored source). The upload
path exploits this end to end: bind.Receive returns up to 128
datagrams per syscall, decrypt processes them as a batch, and
stackTun.Write already loops over its whole buf. The download path
never reached that batching at all: capped to 1 packet at the TUN
read step, every downstream stage (peer lookup, per-peer staging,
eventual UDP send) paid a full cycle per packet instead of amortizing
it across up to 128.

Have Read block for the first packet, then opportunistically drain
whatever's already buffered (non-blocking), up to len(buf). This is
the second half of the throughput-asymmetry fix (see 6436fd9c, which
fixed the channel being fully unbuffered and blocking the producer on
every packet) -- confirmed live on a real test connection: download
went from 30-40 Mbit/s to 130-250 after the channel-buffering fix,
with a real-network-plausible sequential-speedtest gap remaining
against upload's 300+. This closes the remaining structural gap
between the two directions' per-packet processing cost.

Regression tests confirm the drain behavior directly (not just "it
doesn't crash"): stashed this fix alone and re-ran both new tests to
confirm they fail with the exact expected message first.
2026-08-04 02:29:00 +03:00

246 lines
8.9 KiB
Go

// Package amneziawgnet embeds amneziawg-go (a userspace AmneziaWG
// implementation, https://github.com/amnezia-vpn/amneziawg-go) directly in
// the panel process, as an alternative to internal/amneziawg's
// kernel-module (DKMS) + awg-quick approach. A gVisor userspace network
// stack (gvisor.dev/gvisor/pkg/tcpip -- already an indirect dependency via
// xray-core's own proxy/wireguard support) terminates each tunnel, and a
// forwarder recovers each connection's real, dynamically-arbitrary
// destination for the caller to relay onward (see Phase 2 of the migration
// plan: a loopback SOCKS5 dial into Xray, giving native stats/routing/
// sniffing for free).
package amneziawgnet
import (
"fmt"
"net/netip"
"os"
"syscall"
awgtun "github.com/amnezia-vpn/amneziawg-go/v3/tun"
"gvisor.dev/gvisor/pkg/buffer"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/tcpip/link/channel"
"gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
"gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
"gvisor.dev/gvisor/pkg/tcpip/stack"
"gvisor.dev/gvisor/pkg/tcpip/transport/icmp"
"gvisor.dev/gvisor/pkg/tcpip/transport/tcp"
"gvisor.dev/gvisor/pkg/tcpip/transport/udp"
)
// tunQueueDepth is the outbound packet queue depth for both the gVisor
// channel endpoint and the handoff channel to amneziawg-go's TUN reader
// (see the stackTun literal in createNetTUNWithStack for why both need it).
const tunQueueDepth = 1024
// stackTun implements amneziawg-go's tun.Device directly against a gVisor
// channel endpoint, the same approach amneziawg-go's own tun/netstack
// package and xray-core's proxy/wireguard/netstack.go both take. Neither of
// those exposes the raw *stack.Stack a forwarder needs (amneziawg-go's Net
// type keeps it unexported), so this is a local, from-source reimplementation
// rather than a wrapper -- adapted from amneziawg-go v3.0.3's
// tun/netstack/tun.go (MIT licensed), trimmed to the constructor this
// package needs.
type stackTun struct {
ep *channel.Endpoint
stack *stack.Stack
events chan awgtun.Event
notifyHandle *channel.NotificationHandle
incomingPacket chan *buffer.View
mtu int
}
// createNetTUNWithStack builds a gVisor-backed tun.Device for the given
// local addresses (interface address(es), one per family) and returns the
// underlying *stack.Stack alongside it so a caller can attach a forwarder
// (see forwarder.go / udp.go).
func createNetTUNWithStack(localAddresses []netip.Addr, mtu int) (awgtun.Device, *stack.Stack, error) {
opts := stack.Options{
NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol},
TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol, udp.NewProtocol, icmp.NewProtocol6, icmp.NewProtocol4},
// HandleLocal must stay false: promiscuous+spoofing mode (see
// forwarder.go) is what lets a destination other than the stack's
// own configured address reach the forwarder at all.
HandleLocal: false,
}
dev := &stackTun{
// tunQueueDepth matches channel.New's own outbound queue depth
// below. WriteNotify (called synchronously from whatever gVisor
// goroutine is sending TCP data for the download/server->client
// direction) pushes into incomingPacket; RoutineReadFromTUN (a
// single amneziawg-go goroutine that encrypts and sends each
// packet over UDP) is the only reader. With no buffer, every
// outbound packet forced a full synchronous handoff between the
// two -- gVisor's sender blocked until the encrypt loop was ready
// for the next one, one packet at a time, no pipelining. The
// upload/client->server direction has no equivalent stall:
// Write->InjectInbound->DeliverNetworkPacket hands off into
// gVisor's own ~1MB per-connection TCP receive buffer and returns
// immediately. Buffering this channel gives the download
// direction the same slack the upload direction already had.
ep: channel.New(tunQueueDepth, uint32(mtu), ""),
stack: stack.New(opts),
events: make(chan awgtun.Event, 10),
incomingPacket: make(chan *buffer.View, tunQueueDepth),
mtu: mtu,
}
sackEnabledOpt := tcpip.TCPSACKEnabled(true)
if err := dev.stack.SetTransportProtocolOption(tcp.ProtocolNumber, &sackEnabledOpt); err != nil {
return nil, nil, fmt.Errorf("amneziawgnet: enable TCP SACK: %s", err)
}
dev.notifyHandle = dev.ep.AddNotify(dev)
if err := dev.stack.CreateNIC(1, dev.ep); err != nil {
return nil, nil, fmt.Errorf("amneziawgnet: CreateNIC: %s", err)
}
var hasV4, hasV6 bool
for _, ip := range localAddresses {
var protoNumber tcpip.NetworkProtocolNumber
switch {
case ip.Is4():
protoNumber = ipv4.ProtocolNumber
hasV4 = true
case ip.Is6():
protoNumber = ipv6.ProtocolNumber
hasV6 = true
default:
continue
}
protoAddr := tcpip.ProtocolAddress{
Protocol: protoNumber,
AddressWithPrefix: tcpip.AddrFromSlice(ip.AsSlice()).WithPrefix(),
}
if err := dev.stack.AddProtocolAddress(1, protoAddr, stack.AddressProperties{}); err != nil {
return nil, nil, fmt.Errorf("amneziawgnet: AddProtocolAddress(%v): %s", ip, err)
}
}
if hasV4 {
dev.stack.AddRoute(tcpip.Route{Destination: header.IPv4EmptySubnet, NIC: 1})
}
if hasV6 {
dev.stack.AddRoute(tcpip.Route{Destination: header.IPv6EmptySubnet, NIC: 1})
}
dev.events <- awgtun.EventUp
return dev, dev.stack, nil
}
func (t *stackTun) Name() (string, error) { return "amneziawgnet", nil }
func (t *stackTun) File() *os.File { return nil }
func (t *stackTun) Events() <-chan awgtun.Event { return t.events }
func (t *stackTun) MTU() (int, error) { return t.mtu, nil }
func (t *stackTun) BatchSize() int { return 1 }
// Read blocks for the first packet, then opportunistically drains any more
// that are already buffered (non-blocking), up to len(buf). amneziawg-go's
// caller (RoutineReadFromTUN) sizes buf/sizes to device.BatchSize(), which
// is the UDP bind's own batch size (128 on Linux, see conn.IdealBatchSize)
// since that's larger than BatchSize()'s 1 below -- so real buffer capacity
// for a batch is already there. Without this drain loop, Read always
// returned exactly one packet no matter how many buf could hold, so every
// downstream step (peer lookup, per-peer staging, and ultimately the UDP
// bind's own genuinely batched Send/sendmmsg) processed the download
// direction one packet at a time while the upload direction's equivalent
// (bind.Receive/recvmmsg -> decrypt -> stackTun.Write, which already loops
// over its whole buf) processed up to 128 per cycle. That asymmetry is
// real, not gVisor/amneziawg-go's -- both the receive and send paths on the
// UDP bind support batching identically, only this Read implementation
// didn't use it.
func (t *stackTun) Read(buf [][]byte, sizes []int, offset int) (int, error) {
view, ok := <-t.incomingPacket
if !ok {
return 0, os.ErrClosed
}
n, err := view.Read(buf[0][offset:])
if err != nil {
return 0, err
}
sizes[0] = n
count := 1
for count < len(buf) {
select {
case view, ok := <-t.incomingPacket:
if !ok {
return count, nil
}
n, err := view.Read(buf[count][offset:])
if err != nil {
return count, nil
}
sizes[count] = n
count++
default:
return count, nil
}
}
return count, nil
}
func (t *stackTun) Write(buf [][]byte, offset int) (int, error) {
for _, b := range buf {
packet := b[offset:]
if len(packet) == 0 {
continue
}
pkb := stack.NewPacketBuffer(stack.PacketBufferOptions{Payload: buffer.MakeWithData(packet)})
switch packet[0] >> 4 {
case 4:
t.ep.InjectInbound(header.IPv4ProtocolNumber, pkb)
case 6:
t.ep.InjectInbound(header.IPv6ProtocolNumber, pkb)
default:
return 0, syscall.EAFNOSUPPORT
}
}
return len(buf), nil
}
func (t *stackTun) WriteNotify() {
pkt := t.ep.Read()
if pkt == nil {
return
}
view := pkt.ToView()
pkt.DecRef()
t.incomingPacket <- view
}
func (t *stackTun) Close() error {
t.stack.RemoveNIC(1)
t.stack.Close()
t.ep.RemoveNotify(t.notifyHandle)
t.ep.Close()
if t.events != nil {
close(t.events)
}
if t.incomingPacket != nil {
close(t.incomingPacket)
}
return nil
}
// enablePromiscuousRouting puts the NIC into promiscuous + spoofing mode,
// the precondition both AttachTCPForwarder and AttachUDPHandler need to see
// packets addressed to a destination other than the stack's own configured
// local address. Safe to call from both (and more than once): gVisor's
// SetPromiscuousMode/SetSpoofing just set a bool on the NIC, not something
// that accumulates or needs undoing between calls.
func enablePromiscuousRouting(gstack *stack.Stack) {
gstack.SetPromiscuousMode(1, true)
gstack.SetSpoofing(1, true)
}
// addrFromTcpip converts a gVisor tcpip.Address (4 or 16 raw bytes) to the
// stdlib netip.Addr type the rest of this package and its callers use.
func addrFromTcpip(a tcpip.Address) netip.Addr {
if a.Len() == 4 {
var b [4]byte
copy(b[:], a.AsSlice())
return netip.AddrFrom4(b)
}
var b [16]byte
copy(b[:], a.AsSlice())
return netip.AddrFrom16(b)
}