mirror of
https://gitclone.com/github.com/MetaCubeX/Clash.Meta
synced 2024-11-16 06:01:21 +08:00
258 lines
5.2 KiB
Go
258 lines
5.2 KiB
Go
package tuic
|
|
|
|
import (
|
|
"fmt"
|
|
"net"
|
|
"net/netip"
|
|
"sync"
|
|
"sync/atomic"
|
|
"time"
|
|
|
|
"github.com/metacubex/quic-go"
|
|
|
|
N "github.com/Dreamacro/clash/common/net"
|
|
"github.com/Dreamacro/clash/common/pool"
|
|
"github.com/Dreamacro/clash/transport/tuic/congestion"
|
|
)
|
|
|
|
const (
|
|
DefaultStreamReceiveWindow = 15728640 // 15 MB/s
|
|
DefaultConnectionReceiveWindow = 67108864 // 64 MB/s
|
|
)
|
|
|
|
func SetCongestionController(quicConn quic.Connection, cc string) {
|
|
switch cc {
|
|
case "cubic":
|
|
quicConn.SetCongestionControl(
|
|
congestion.NewCubicSender(
|
|
congestion.DefaultClock{},
|
|
congestion.GetMaxPacketSize(quicConn.RemoteAddr()),
|
|
false,
|
|
nil,
|
|
),
|
|
)
|
|
case "new_reno":
|
|
quicConn.SetCongestionControl(
|
|
congestion.NewCubicSender(
|
|
congestion.DefaultClock{},
|
|
congestion.GetMaxPacketSize(quicConn.RemoteAddr()),
|
|
true,
|
|
nil,
|
|
),
|
|
)
|
|
case "bbr":
|
|
quicConn.SetCongestionControl(
|
|
congestion.NewBBRSender(
|
|
congestion.DefaultClock{},
|
|
congestion.GetMaxPacketSize(quicConn.RemoteAddr()),
|
|
congestion.InitialCongestionWindow,
|
|
congestion.DefaultBBRMaxCongestionWindow,
|
|
),
|
|
)
|
|
}
|
|
}
|
|
|
|
type quicStreamConn struct {
|
|
quic.Stream
|
|
lock sync.Mutex
|
|
lAddr net.Addr
|
|
rAddr net.Addr
|
|
|
|
closeDeferFn func()
|
|
|
|
closeOnce sync.Once
|
|
closeErr error
|
|
}
|
|
|
|
func (q *quicStreamConn) Write(p []byte) (n int, err error) {
|
|
q.lock.Lock()
|
|
defer q.lock.Unlock()
|
|
return q.Stream.Write(p)
|
|
}
|
|
|
|
func (q *quicStreamConn) Close() error {
|
|
q.closeOnce.Do(func() {
|
|
q.closeErr = q.close()
|
|
})
|
|
return q.closeErr
|
|
}
|
|
|
|
func (q *quicStreamConn) close() error {
|
|
if q.closeDeferFn != nil {
|
|
defer q.closeDeferFn()
|
|
}
|
|
|
|
// https://github.com/cloudflare/cloudflared/commit/ed2bac026db46b239699ac5ce4fcf122d7cab2cd
|
|
// Make sure a possible writer does not block the lock forever. We need it, so we can close the writer
|
|
// side of the stream safely.
|
|
_ = q.Stream.SetWriteDeadline(time.Now())
|
|
|
|
// This lock is eventually acquired despite Write also acquiring it, because we set a deadline to writes.
|
|
q.lock.Lock()
|
|
defer q.lock.Unlock()
|
|
|
|
// We have to clean up the receiving stream ourselves since the Close in the bottom does not handle that.
|
|
q.Stream.CancelRead(0)
|
|
return q.Stream.Close()
|
|
}
|
|
|
|
func (q *quicStreamConn) LocalAddr() net.Addr {
|
|
return q.lAddr
|
|
}
|
|
|
|
func (q *quicStreamConn) RemoteAddr() net.Addr {
|
|
return q.rAddr
|
|
}
|
|
|
|
var _ net.Conn = &quicStreamConn{}
|
|
|
|
type quicStreamPacketConn struct {
|
|
connId uint32
|
|
quicConn quic.Connection
|
|
lAddr net.Addr
|
|
inputConn *N.BufferedConn
|
|
|
|
udpRelayMode string
|
|
maxUdpRelayPacketSize int
|
|
|
|
deferQuicConnFn func(quicConn quic.Connection, err error)
|
|
closeDeferFn func()
|
|
writeClosed *atomic.Bool
|
|
|
|
closeOnce sync.Once
|
|
closeErr error
|
|
closed bool
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) Close() error {
|
|
q.closeOnce.Do(func() {
|
|
q.closed = true
|
|
q.closeErr = q.close()
|
|
})
|
|
return q.closeErr
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) close() (err error) {
|
|
if q.closeDeferFn != nil {
|
|
defer q.closeDeferFn()
|
|
}
|
|
if q.deferQuicConnFn != nil {
|
|
defer func() {
|
|
q.deferQuicConnFn(q.quicConn, err)
|
|
}()
|
|
}
|
|
if q.inputConn != nil {
|
|
_ = q.inputConn.Close()
|
|
q.inputConn = nil
|
|
|
|
buf := pool.GetBuffer()
|
|
defer pool.PutBuffer(buf)
|
|
err = NewDissociate(q.connId).WriteTo(buf)
|
|
if err != nil {
|
|
return
|
|
}
|
|
var stream quic.SendStream
|
|
stream, err = q.quicConn.OpenUniStream()
|
|
if err != nil {
|
|
return
|
|
}
|
|
_, err = buf.WriteTo(stream)
|
|
if err != nil {
|
|
return
|
|
}
|
|
err = stream.Close()
|
|
if err != nil {
|
|
return
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) SetDeadline(t time.Time) error {
|
|
//TODO implement me
|
|
return nil
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) SetReadDeadline(t time.Time) error {
|
|
if q.inputConn != nil {
|
|
return q.inputConn.SetReadDeadline(t)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) SetWriteDeadline(t time.Time) error {
|
|
//TODO implement me
|
|
return nil
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) ReadFrom(p []byte) (n int, addr net.Addr, err error) {
|
|
if q.inputConn != nil {
|
|
var packet Packet
|
|
packet, err = ReadPacket(q.inputConn)
|
|
if err != nil {
|
|
return
|
|
}
|
|
n = copy(p, packet.DATA)
|
|
addr = packet.ADDR.UDPAddr()
|
|
} else {
|
|
err = net.ErrClosed
|
|
}
|
|
return
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) WriteTo(p []byte, addr net.Addr) (n int, err error) {
|
|
if len(p) > q.maxUdpRelayPacketSize {
|
|
return 0, fmt.Errorf("udp packet too large(%d > %d)", len(p), q.maxUdpRelayPacketSize)
|
|
}
|
|
if q.closed {
|
|
return 0, net.ErrClosed
|
|
}
|
|
if q.writeClosed != nil && q.writeClosed.Load() {
|
|
_ = q.Close()
|
|
return 0, net.ErrClosed
|
|
}
|
|
if q.deferQuicConnFn != nil {
|
|
defer func() {
|
|
q.deferQuicConnFn(q.quicConn, err)
|
|
}()
|
|
}
|
|
addr.String()
|
|
buf := pool.GetBuffer()
|
|
defer pool.PutBuffer(buf)
|
|
addrPort, err := netip.ParseAddrPort(addr.String())
|
|
if err != nil {
|
|
return
|
|
}
|
|
err = NewPacket(q.connId, uint16(len(p)), NewAddressAddrPort(addrPort), p).WriteTo(buf)
|
|
if err != nil {
|
|
return
|
|
}
|
|
switch q.udpRelayMode {
|
|
case "quic":
|
|
var stream quic.SendStream
|
|
stream, err = q.quicConn.OpenUniStream()
|
|
if err != nil {
|
|
return
|
|
}
|
|
defer stream.Close()
|
|
_, err = buf.WriteTo(stream)
|
|
if err != nil {
|
|
return
|
|
}
|
|
default: // native
|
|
err = q.quicConn.SendMessage(buf.Bytes())
|
|
if err != nil {
|
|
return
|
|
}
|
|
}
|
|
n = len(p)
|
|
|
|
return
|
|
}
|
|
|
|
func (q *quicStreamPacketConn) LocalAddr() net.Addr {
|
|
return q.lAddr
|
|
}
|
|
|
|
var _ net.PacketConn = &quicStreamPacketConn{}
|