snapshot: sagernet/gvisor v0.0.0-20250811.0-sing-box-mod.1

Содержимое пина, зафиксированного в go.mod sing-box-lx, одним коммитом
без истории. Полная история SagerNet/gvisor — 1.45 ГБ и клонируется в
каждой CI-джобе; наша дельта — одна вставка в одну функцию, история для
неё не нужна.

Module path github.com/sagernet/gvisor сохранён намеренно: на него
опирается replace-директива суперпроекта.

Патч поверх — отдельным коммитом, чтобы дельта читалась одним git show
и переносилась на новый пин копированием.

SPECS/TASKS/048-GVISOR_HANDSHAKE_NIL_CRASH
This commit is contained in:
Leadaxe 2026-08-04 15:50:08 +03:00
commit 2c4ae3b0a4
712 changed files with 185689 additions and 0 deletions

547
pkg/unet/unet.go Normal file
View file

@ -0,0 +1,547 @@
// 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 unet provides a minimal net package based on Unix Domain Sockets.
//
// This does no pooling, and should only be used for a limited number of
// connections in a Go process. Don't use this package for arbitrary servers.
package unet
import (
"errors"
"github.com/sagernet/gvisor/pkg/atomicbitops"
"github.com/sagernet/gvisor/pkg/eventfd"
"github.com/sagernet/gvisor/pkg/sync"
"golang.org/x/sys/unix"
)
// backlog is used for the listen request.
const backlog = 16
// errClosing is returned by wait if the Socket is in the process of closing.
var errClosing = errors.New("Socket is closing")
// errMessageTruncated indicates that data was lost because the provided buffer
// was too small.
var errMessageTruncated = errors.New("message truncated")
// socketType returns the appropriate type.
func socketType(packet bool) int {
if packet {
return unix.SOCK_SEQPACKET
}
return unix.SOCK_STREAM
}
// socket creates a new host socket.
func socket(packet bool) (int, error) {
// Make a new socket.
fd, err := unix.Socket(unix.AF_UNIX, socketType(packet), 0)
if err != nil {
return 0, err
}
return fd, nil
}
// Socket is a connected unix domain socket.
type Socket struct {
// gate protects use of fd.
gate sync.Gate
// fd is the bound socket.
//
// fd only remains valid if read while within gate.
fd atomicbitops.Int32
// efd is an event FD that is signaled when the socket is closing.
//
// efd is immutable and remains valid until Close/Release.
efd eventfd.Eventfd
// race is an atomic variable used to avoid triggering the race
// detector. See comment in SocketPair below.
race *atomicbitops.Int32
}
// NewSocket returns a socket from an existing FD.
//
// NewSocket takes ownership of fd.
func NewSocket(fd int) (*Socket, error) {
// fd must be non-blocking for non-blocking unix.Accept in
// ServerSocket.Accept.
if err := unix.SetNonblock(fd, true); err != nil {
return nil, err
}
efd, err := eventfd.Create()
if err != nil {
return nil, err
}
return &Socket{
fd: atomicbitops.FromInt32(int32(fd)),
efd: efd,
}, nil
}
// finish completes use of s.fd by evicting any waiters, closing the gate, and
// closing the event FD.
func (s *Socket) finish() error {
// Signal any blocked or future polls.
if err := s.efd.Notify(); err != nil {
return err
}
// Close the gate, blocking until all FD users leave.
s.gate.Close()
return s.efd.Close()
}
// Close closes the socket.
func (s *Socket) Close() error {
// Set the FD in the socket to -1, to ensure that all future calls to
// FD/Release get nothing and Close calls return immediately.
fd := int(s.fd.Swap(-1))
if fd < 0 {
// Already closed or closing.
return unix.EBADF
}
// Shutdown the socket to cancel any pending accepts.
s.shutdown(fd)
if err := s.finish(); err != nil {
return err
}
return unix.Close(fd)
}
// Release releases ownership of the socket FD.
//
// The returned FD is non-blocking.
//
// Any concurrent or future callers of Socket methods will receive EBADF.
func (s *Socket) Release() (int, error) {
// Set the FD in the socket to -1, to ensure that all future calls to
// FD/Release get nothing and Close calls return immediately.
fd := int(s.fd.Swap(-1))
if fd < 0 {
// Already closed or closing.
return -1, unix.EBADF
}
if err := s.finish(); err != nil {
return -1, err
}
return fd, nil
}
// FD returns the FD for this Socket.
//
// The FD is non-blocking and must not be made blocking.
//
// N.B. os.File.Fd makes the FD blocking. Use of Release instead of FD is
// strongly preferred.
//
// The returned FD cannot be used safely if there may be concurrent callers to
// Close or Release.
//
// Use Release to take ownership of the FD.
func (s *Socket) FD() int {
return int(s.fd.Load())
}
// enterFD enters the FD gate and returns the FD value.
//
// If enterFD returns ok, s.gate.Leave must be called when done with the FD.
// Callers may only block while within the gate using s.wait.
//
// The returned FD is guaranteed to remain valid until s.gate.Leave.
func (s *Socket) enterFD() (int, bool) {
if !s.gate.Enter() {
return -1, false
}
fd := int(s.fd.Load())
if fd < 0 {
s.gate.Leave()
return -1, false
}
return fd, true
}
// SocketPair creates a pair of connected sockets.
func SocketPair(packet bool) (*Socket, *Socket, error) {
// Make a new pair.
fds, err := unix.Socketpair(unix.AF_UNIX, socketType(packet)|unix.SOCK_CLOEXEC, 0)
if err != nil {
return nil, nil, err
}
// race is an atomic variable used to avoid triggering the race
// detector. We have to fool TSAN into thinking there is a race
// variable between our two sockets. We only use SocketPair in tests
// anyway.
//
// NOTE(b/27107811): This is purely due to the fact that the raw
// syscall does not serve as a boundary for the sanitizer.
a, err := NewSocket(fds[0])
if err != nil {
unix.Close(fds[0])
unix.Close(fds[1])
return nil, nil, err
}
var race atomicbitops.Int32
a.race = &race
b, err := NewSocket(fds[1])
if err != nil {
a.Close()
unix.Close(fds[1])
return nil, nil, err
}
b.race = &race
return a, b, nil
}
// Connect connects to a server.
func Connect(addr string, packet bool) (*Socket, error) {
fd, err := socket(packet)
if err != nil {
return nil, err
}
// Connect the socket.
usa := &unix.SockaddrUnix{Name: addr}
if err := unix.Connect(fd, usa); err != nil {
unix.Close(fd)
return nil, err
}
return NewSocket(fd)
}
// ControlMessage wraps around a byte array and provides functions for parsing
// as a Unix Domain Socket control message.
type ControlMessage []byte
// EnableFDs enables receiving FDs via control message.
//
// This guarantees only a MINIMUM number of FDs received. You may receive MORE
// than this due to the way FDs are packed. To be specific, the number of
// receivable buffers will be rounded up to the nearest even number.
//
// This must be called prior to ReadVec if you want to receive FDs.
func (c *ControlMessage) EnableFDs(count int) {
*c = make([]byte, unix.CmsgSpace(count*4))
}
// ExtractFDs returns the list of FDs in the control message.
//
// Either this or CloseFDs should be used after EnableFDs.
func (c *ControlMessage) ExtractFDs() ([]int, error) {
msgs, err := unix.ParseSocketControlMessage(*c)
if err != nil {
return nil, err
}
var fds []int
for _, msg := range msgs {
thisFds, err := unix.ParseUnixRights(&msg)
if err != nil {
// Different control message.
return nil, err
}
for _, fd := range thisFds {
if fd >= 0 {
fds = append(fds, fd)
}
}
}
return fds, nil
}
// CloseFDs closes the list of FDs in the control message.
//
// Either this or ExtractFDs should be used after EnableFDs.
func (c *ControlMessage) CloseFDs() {
fds, _ := c.ExtractFDs()
for _, fd := range fds {
if fd >= 0 {
unix.Close(fd)
}
}
}
// PackFDs packs the given list of FDs in the control message.
//
// This must be used prior to WriteVec.
func (c *ControlMessage) PackFDs(fds ...int) {
*c = ControlMessage(unix.UnixRights(fds...))
}
// UnpackFDs clears the control message.
func (c *ControlMessage) UnpackFDs() {
*c = nil
}
// SocketWriter wraps an individual send operation.
//
// The normal entrypoint is WriteVec.
type SocketWriter struct {
socket *Socket
to []byte
blocking bool
race *atomicbitops.Int32
ControlMessage
}
// Writer returns a writer for this socket.
func (s *Socket) Writer(blocking bool) SocketWriter {
return SocketWriter{socket: s, blocking: blocking, race: s.race}
}
// Write implements io.Writer.Write.
func (s *Socket) Write(p []byte) (int, error) {
r := s.Writer(true)
return r.WriteVec([][]byte{p})
}
// GetSockOpt gets the given socket option.
func (s *Socket) GetSockOpt(level int, name int, b []byte) (uint32, error) {
fd, ok := s.enterFD()
if !ok {
return 0, unix.EBADF
}
defer s.gate.Leave()
return getsockopt(fd, level, name, b)
}
// SetSockOpt sets the given socket option.
func (s *Socket) SetSockOpt(level, name int, b []byte) error {
fd, ok := s.enterFD()
if !ok {
return unix.EBADF
}
defer s.gate.Leave()
return setsockopt(fd, level, name, b)
}
// GetSockName returns the socket name.
func (s *Socket) GetSockName() ([]byte, error) {
fd, ok := s.enterFD()
if !ok {
return nil, unix.EBADF
}
defer s.gate.Leave()
var buf []byte
l := unix.SizeofSockaddrAny
for {
// If the buffer is not large enough, allocate a new one with the hint.
buf = make([]byte, l)
l, err := getsockname(fd, buf)
if err != nil {
return nil, err
}
if l <= uint32(len(buf)) {
return buf[:l], nil
}
}
}
// GetPeerName returns the peer name.
func (s *Socket) GetPeerName() ([]byte, error) {
fd, ok := s.enterFD()
if !ok {
return nil, unix.EBADF
}
defer s.gate.Leave()
var buf []byte
l := unix.SizeofSockaddrAny
for {
// See above.
buf = make([]byte, l)
l, err := getpeername(fd, buf)
if err != nil {
return nil, err
}
if l <= uint32(len(buf)) {
return buf[:l], nil
}
}
}
// SocketReader wraps an individual receive operation.
//
// This may be used for doing vectorized reads and/or sending additional
// control messages (e.g. FDs). The normal entrypoint is ReadVec.
//
// One of ExtractFDs or DisposeFDs must be called if EnableFDs is used.
type SocketReader struct {
socket *Socket
source []byte
blocking bool
race *atomicbitops.Int32
ControlMessage
}
// Reader returns a reader for this socket.
func (s *Socket) Reader(blocking bool) SocketReader {
return SocketReader{socket: s, blocking: blocking, race: s.race}
}
// Read implements io.Reader.Read.
func (s *Socket) Read(p []byte) (int, error) {
r := s.Reader(true)
return r.ReadVec([][]byte{p})
}
func (s *Socket) shutdown(fd int) error {
// Shutdown the socket to cancel any pending accepts.
return unix.Shutdown(fd, unix.SHUT_RDWR)
}
// Shutdown closes the socket for read and write.
func (s *Socket) Shutdown() error {
fd, ok := s.enterFD()
if !ok {
return unix.EBADF
}
defer s.gate.Leave()
return s.shutdown(fd)
}
// ServerSocket is a bound unix domain socket.
type ServerSocket struct {
socket *Socket
}
// NewServerSocket returns a socket from an existing FD.
func NewServerSocket(fd int) (*ServerSocket, error) {
s, err := NewSocket(fd)
if err != nil {
return nil, err
}
return &ServerSocket{socket: s}, nil
}
// Bind creates and binds a new socket.
func Bind(addr string, packet bool) (*ServerSocket, error) {
fd, err := socket(packet)
if err != nil {
return nil, err
}
// Do the bind.
usa := &unix.SockaddrUnix{Name: addr}
if err := unix.Bind(fd, usa); err != nil {
unix.Close(fd)
return nil, err
}
return NewServerSocket(fd)
}
// BindAndListen creates, binds and listens on a new socket.
func BindAndListen(addr string, packet bool) (*ServerSocket, error) {
s, err := Bind(addr, packet)
if err != nil {
return nil, err
}
// Start listening.
if err := s.Listen(); err != nil {
s.Close()
return nil, err
}
return s, nil
}
// Listen starts listening on the socket.
func (s *ServerSocket) Listen() error {
fd, ok := s.socket.enterFD()
if !ok {
return unix.EBADF
}
defer s.socket.gate.Leave()
return unix.Listen(fd, backlog)
}
// Accept accepts a new connection.
//
// This is always blocking.
//
// Preconditions:
// - ServerSocket is listening (Listen called).
func (s *ServerSocket) Accept() (*Socket, error) {
fd, ok := s.socket.enterFD()
if !ok {
return nil, unix.EBADF
}
defer s.socket.gate.Leave()
for {
nfd, _, err := unix.Accept(fd)
switch err {
case nil:
return NewSocket(nfd)
case unix.EAGAIN:
err = s.socket.wait(false)
if err == errClosing {
err = unix.EBADF
}
}
if err != nil {
return nil, err
}
}
}
// Close closes the server socket.
//
// This must only be called once.
func (s *ServerSocket) Close() error {
return s.socket.Close()
}
// FD returns the socket's file descriptor.
//
// See Socket.FD.
func (s *ServerSocket) FD() int {
return s.socket.FD()
}
// Release releases ownership of the socket's file descriptor.
//
// See Socket.Release.
func (s *ServerSocket) Release() (int, error) {
return s.socket.Release()
}

View file

@ -0,0 +1,3 @@
// automatically generated by stateify.
package unet

286
pkg/unet/unet_unsafe.go Normal file
View file

@ -0,0 +1,286 @@
// 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 unet
import (
"io"
"unsafe"
"golang.org/x/sys/unix"
)
// wait blocks until the socket FD is ready for reading or writing, depending
// on the value of write.
//
// Returns errClosing if the Socket is in the process of closing.
func (s *Socket) wait(write bool) error {
for {
// Checking the FD on each loop is not strictly necessary, it
// just avoids an extra poll call.
fd := s.fd.Load()
if fd < 0 {
return errClosing
}
events := []unix.PollFd{
{
// The actual socket FD.
Fd: fd,
Events: unix.POLLIN,
},
{
// The eventfd, signaled when we are closing.
Fd: int32(s.efd.FD()),
Events: unix.POLLIN,
},
}
if write {
events[0].Events = unix.POLLOUT
}
_, _, e := unix.Syscall6(unix.SYS_PPOLL, uintptr(unsafe.Pointer(&events[0])), 2, 0, 0, 0, 0)
if e == unix.EINTR {
continue
}
if e != 0 {
return e
}
if events[1].Revents&unix.POLLIN == unix.POLLIN {
// eventfd signaled, we're closing.
return errClosing
}
return nil
}
}
// buildIovec builds an iovec slice from the given []byte slice.
//
// iovecs is used as an initial slice, to avoid excessive allocations.
func buildIovec(bufs [][]byte, iovecs []unix.Iovec) ([]unix.Iovec, int) {
var length int
for i := range bufs {
if l := len(bufs[i]); l > 0 {
iovecs = append(iovecs, unix.Iovec{
Base: &bufs[i][0],
Len: uint64(l),
})
length += l
}
}
return iovecs, length
}
// ReadVec reads into the pre-allocated bufs. Returns bytes read.
//
// The pre-allocatted space used by ReadVec is based upon slice lengths.
//
// This function is not guaranteed to read all available data, it
// returns as soon as a single recvmsg call succeeds.
func (r *SocketReader) ReadVec(bufs [][]byte) (int, error) {
iovecs, length := buildIovec(bufs, make([]unix.Iovec, 0, 2))
var msg unix.Msghdr
if len(r.source) != 0 {
msg.Name = &r.source[0]
msg.Namelen = uint32(len(r.source))
}
if len(r.ControlMessage) != 0 {
msg.Control = &r.ControlMessage[0]
msg.Controllen = uint64(len(r.ControlMessage))
}
if len(iovecs) != 0 {
msg.Iov = &iovecs[0]
msg.Iovlen = uint64(len(iovecs))
}
// n is the bytes received.
var n uintptr
fd, ok := r.socket.enterFD()
if !ok {
return 0, unix.EBADF
}
// Leave on returns below.
for {
var e unix.Errno
// Try a non-blocking recv first, so we don't give up the go runtime M.
n, _, e = unix.RawSyscall(unix.SYS_RECVMSG, uintptr(fd), uintptr(unsafe.Pointer(&msg)), unix.MSG_DONTWAIT|unix.MSG_TRUNC)
if e == 0 {
break
}
if e == unix.EINTR {
continue
}
if !r.blocking {
r.socket.gate.Leave()
return 0, e
}
if e != unix.EAGAIN && e != unix.EWOULDBLOCK {
r.socket.gate.Leave()
return 0, e
}
// Wait for the socket to become readable.
err := r.socket.wait(false)
if err == errClosing {
err = unix.EBADF
}
if err != nil {
r.socket.gate.Leave()
return 0, err
}
}
r.socket.gate.Leave()
if msg.Controllen < uint64(len(r.ControlMessage)) {
r.ControlMessage = r.ControlMessage[:msg.Controllen]
}
if msg.Namelen < uint32(len(r.source)) {
r.source = r.source[:msg.Namelen]
}
// All unet sockets are SOCK_STREAM or SOCK_SEQPACKET, both of which
// indicate that the other end is closed by returning a 0 length read
// with no error.
if n == 0 {
return 0, io.EOF
}
if r.race != nil {
// See comments on Socket.race.
r.race.Add(1)
}
if int(n) > length {
return length, errMessageTruncated
}
return int(n), nil
}
// WriteVec writes the bufs to the socket. Returns bytes written.
//
// This function is not guaranteed to send all data, it returns
// as soon as a single sendmsg call succeeds.
func (w *SocketWriter) WriteVec(bufs [][]byte) (int, error) {
iovecs, _ := buildIovec(bufs, make([]unix.Iovec, 0, 2))
if w.race != nil {
// See comments on Socket.race.
w.race.Add(1)
}
var msg unix.Msghdr
if len(w.to) != 0 {
msg.Name = &w.to[0]
msg.Namelen = uint32(len(w.to))
}
if len(w.ControlMessage) != 0 {
msg.Control = &w.ControlMessage[0]
msg.Controllen = uint64(len(w.ControlMessage))
}
if len(iovecs) > 0 {
msg.Iov = &iovecs[0]
msg.Iovlen = uint64(len(iovecs))
}
fd, ok := w.socket.enterFD()
if !ok {
return 0, unix.EBADF
}
// Leave on returns below.
for {
// Try a non-blocking send first, so we don't give up the go runtime M.
n, _, e := unix.RawSyscall(unix.SYS_SENDMSG, uintptr(fd), uintptr(unsafe.Pointer(&msg)), unix.MSG_DONTWAIT|unix.MSG_NOSIGNAL)
if e == 0 {
w.socket.gate.Leave()
return int(n), nil
}
if e == unix.EINTR {
continue
}
if !w.blocking {
w.socket.gate.Leave()
return 0, e
}
if e != unix.EAGAIN && e != unix.EWOULDBLOCK {
w.socket.gate.Leave()
return 0, e
}
// Wait for the socket to become writeable.
err := w.socket.wait(true)
if err == errClosing {
err = unix.EBADF
}
if err != nil {
w.socket.gate.Leave()
return 0, err
}
}
// Unreachable, no s.gate.Leave needed.
}
// getsockopt issues a getsockopt unix.
func getsockopt(fd int, level int, optname int, buf []byte) (uint32, error) {
l := uint32(len(buf))
_, _, e := unix.RawSyscall6(unix.SYS_GETSOCKOPT, uintptr(fd), uintptr(level), uintptr(optname), uintptr(unsafe.Pointer(&buf[0])), uintptr(unsafe.Pointer(&l)), 0)
if e != 0 {
return 0, e
}
return l, nil
}
// setsockopt issues a setsockopt unix.
func setsockopt(fd int, level int, optname int, buf []byte) error {
_, _, e := unix.RawSyscall6(unix.SYS_SETSOCKOPT, uintptr(fd), uintptr(level), uintptr(optname), uintptr(unsafe.Pointer(&buf[0])), uintptr(len(buf)), 0)
if e != 0 {
return e
}
return nil
}
// getsockname issues a getsockname unix.
func getsockname(fd int, buf []byte) (uint32, error) {
l := uint32(len(buf))
_, _, e := unix.RawSyscall(unix.SYS_GETSOCKNAME, uintptr(fd), uintptr(unsafe.Pointer(&buf[0])), uintptr(unsafe.Pointer(&l)))
if e != 0 {
return 0, e
}
return l, nil
}
// getpeername issues a getpeername unix.
func getpeername(fd int, buf []byte) (uint32, error) {
l := uint32(len(buf))
_, _, e := unix.RawSyscall(unix.SYS_GETPEERNAME, uintptr(fd), uintptr(unsafe.Pointer(&buf[0])), uintptr(unsafe.Pointer(&l)))
if e != 0 {
return 0, e
}
return l, nil
}

View file

@ -0,0 +1,3 @@
// automatically generated by stateify.
package unet