blob: 3eeeb7e204c3d0040ae9277fc1eecb4fb5b684de [file] [log] [blame]
// Copyright 2012 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "net/socket/socks5_client_socket.h"
#include <stdint.h>
#include <array>
#include <utility>
#include "base/compiler_specific.h"
#include "base/format_macros.h"
#include "base/functional/bind.h"
#include "base/functional/callback_helpers.h"
#include "base/numerics/byte_conversions.h"
#include "base/numerics/safe_conversions.h"
#include "base/strings/string_util.h"
#include "base/trace_event/trace_event.h"
#include "net/base/io_buffer.h"
#include "net/base/sys_addrinfo.h"
#include "net/log/net_log.h"
#include "net/log/net_log_event_type.h"
#include "net/traffic_annotation/network_traffic_annotation.h"
namespace net {
const unsigned int SOCKS5ClientSocket::kGreetReadHeaderSize = 2;
const unsigned int SOCKS5ClientSocket::kWriteHeaderSize = 10;
const unsigned int SOCKS5ClientSocket::kReadHeaderSize = 5;
const uint8_t SOCKS5ClientSocket::kSOCKS5Version = 0x05;
const uint8_t SOCKS5ClientSocket::kTunnelCommand = 0x01;
const uint8_t SOCKS5ClientSocket::kNullByte = 0x00;
static_assert(sizeof(struct in_addr) == 4, "incorrect system size of IPv4");
static_assert(sizeof(struct in6_addr) == 16, "incorrect system size of IPv6");
SOCKS5ClientSocket::SOCKS5ClientSocket(
std::unique_ptr<StreamSocket> transport_socket,
const HostPortPair& destination,
const NetworkTrafficAnnotationTag& traffic_annotation)
: io_callback_(base::BindRepeating(&SOCKS5ClientSocket::OnIOComplete,
base::Unretained(this))),
transport_socket_(std::move(transport_socket)),
destination_(destination),
net_log_(transport_socket_->NetLog()),
traffic_annotation_(traffic_annotation) {}
SOCKS5ClientSocket::~SOCKS5ClientSocket() {
Disconnect();
}
int SOCKS5ClientSocket::Connect(CompletionOnceCallback callback) {
DCHECK(transport_socket_);
DCHECK_EQ(STATE_NONE, next_state_);
DCHECK(user_callback_.is_null());
// If already connected, then just return OK.
if (completed_handshake_)
return OK;
net_log_.BeginEvent(NetLogEventType::SOCKS5_CONNECT);
next_state_ = STATE_GREET_WRITE;
write_buf_.reset();
read_buf_.reset();
int rv = DoLoop(OK);
if (rv == ERR_IO_PENDING) {
user_callback_ = std::move(callback);
} else {
net_log_.EndEventWithNetErrorCode(NetLogEventType::SOCKS5_CONNECT, rv);
}
return rv;
}
void SOCKS5ClientSocket::Disconnect() {
completed_handshake_ = false;
transport_socket_->Disconnect();
// Reset other states to make sure they aren't mistakenly used later.
// These are the states initialized by Connect().
next_state_ = STATE_NONE;
user_callback_.Reset();
}
bool SOCKS5ClientSocket::IsConnected() const {
return completed_handshake_ && transport_socket_->IsConnected();
}
bool SOCKS5ClientSocket::IsConnectedAndIdle() const {
return completed_handshake_ && transport_socket_->IsConnectedAndIdle();
}
const NetLogWithSource& SOCKS5ClientSocket::NetLog() const {
return net_log_;
}
bool SOCKS5ClientSocket::WasEverUsed() const {
return was_ever_used_;
}
NextProto SOCKS5ClientSocket::GetNegotiatedProtocol() const {
if (transport_socket_)
return transport_socket_->GetNegotiatedProtocol();
NOTREACHED();
}
bool SOCKS5ClientSocket::GetSSLInfo(SSLInfo* ssl_info) {
if (transport_socket_)
return transport_socket_->GetSSLInfo(ssl_info);
NOTREACHED();
}
int64_t SOCKS5ClientSocket::GetTotalReceivedBytes() const {
return transport_socket_->GetTotalReceivedBytes();
}
void SOCKS5ClientSocket::ApplySocketTag(const SocketTag& tag) {
return transport_socket_->ApplySocketTag(tag);
}
// Read is called by the transport layer above to read. This can only be done
// if the SOCKS handshake is complete.
int SOCKS5ClientSocket::Read(IOBuffer* buf,
int buf_len,
CompletionOnceCallback callback) {
DCHECK(completed_handshake_);
DCHECK_EQ(STATE_NONE, next_state_);
DCHECK(user_callback_.is_null());
DCHECK(!callback.is_null());
int rv = transport_socket_->Read(
buf, buf_len,
base::BindOnce(&SOCKS5ClientSocket::OnReadWriteComplete,
base::Unretained(this), std::move(callback)));
if (rv > 0)
was_ever_used_ = true;
return rv;
}
// Write is called by the transport layer. This can only be done if the
// SOCKS handshake is complete.
int SOCKS5ClientSocket::Write(
IOBuffer* buf,
int buf_len,
CompletionOnceCallback callback,
const NetworkTrafficAnnotationTag& traffic_annotation) {
DCHECK(completed_handshake_);
DCHECK_EQ(STATE_NONE, next_state_);
DCHECK(user_callback_.is_null());
DCHECK(!callback.is_null());
int rv = transport_socket_->Write(
buf, buf_len,
base::BindOnce(&SOCKS5ClientSocket::OnReadWriteComplete,
base::Unretained(this), std::move(callback)),
traffic_annotation);
if (rv > 0)
was_ever_used_ = true;
return rv;
}
int SOCKS5ClientSocket::SetReceiveBufferSize(int32_t size) {
return transport_socket_->SetReceiveBufferSize(size);
}
int SOCKS5ClientSocket::SetSendBufferSize(int32_t size) {
return transport_socket_->SetSendBufferSize(size);
}
void SOCKS5ClientSocket::DoCallback(int result) {
DCHECK_NE(ERR_IO_PENDING, result);
DCHECK(!user_callback_.is_null());
// Since Run() may result in Read being called,
// clear user_callback_ up front.
std::move(user_callback_).Run(result);
}
void SOCKS5ClientSocket::OnIOComplete(int result) {
DCHECK_NE(STATE_NONE, next_state_);
int rv = DoLoop(result);
if (rv != ERR_IO_PENDING) {
net_log_.EndEvent(NetLogEventType::SOCKS5_CONNECT);
DoCallback(rv);
}
}
void SOCKS5ClientSocket::OnReadWriteComplete(CompletionOnceCallback callback,
int result) {
DCHECK_NE(ERR_IO_PENDING, result);
DCHECK(!callback.is_null());
if (result > 0)
was_ever_used_ = true;
std::move(callback).Run(result);
}
int SOCKS5ClientSocket::DoLoop(int last_io_result) {
DCHECK_NE(next_state_, STATE_NONE);
int rv = last_io_result;
do {
State state = next_state_;
next_state_ = STATE_NONE;
switch (state) {
case STATE_GREET_WRITE:
DCHECK_EQ(OK, rv);
net_log_.BeginEvent(NetLogEventType::SOCKS5_GREET_WRITE);
rv = DoGreetWrite();
break;
case STATE_GREET_WRITE_COMPLETE:
rv = DoGreetWriteComplete(rv);
net_log_.EndEventWithNetErrorCode(NetLogEventType::SOCKS5_GREET_WRITE,
rv);
break;
case STATE_GREET_READ:
DCHECK_EQ(OK, rv);
net_log_.BeginEvent(NetLogEventType::SOCKS5_GREET_READ);
rv = DoGreetRead();
break;
case STATE_GREET_READ_COMPLETE:
rv = DoGreetReadComplete(rv);
net_log_.EndEventWithNetErrorCode(NetLogEventType::SOCKS5_GREET_READ,
rv);
break;
case STATE_HANDSHAKE_WRITE:
DCHECK_EQ(OK, rv);
net_log_.BeginEvent(NetLogEventType::SOCKS5_HANDSHAKE_WRITE);
rv = DoHandshakeWrite();
break;
case STATE_HANDSHAKE_WRITE_COMPLETE:
rv = DoHandshakeWriteComplete(rv);
net_log_.EndEventWithNetErrorCode(
NetLogEventType::SOCKS5_HANDSHAKE_WRITE, rv);
break;
case STATE_HANDSHAKE_READ:
DCHECK_EQ(OK, rv);
net_log_.BeginEvent(NetLogEventType::SOCKS5_HANDSHAKE_READ);
rv = DoHandshakeRead();
break;
case STATE_HANDSHAKE_READ_COMPLETE:
rv = DoHandshakeReadComplete(rv);
net_log_.EndEventWithNetErrorCode(
NetLogEventType::SOCKS5_HANDSHAKE_READ, rv);
break;
default:
NOTREACHED() << "bad state";
}
} while (rv != ERR_IO_PENDING && next_state_ != STATE_NONE);
return rv;
}
static constexpr std::array<uint8_t, 3> kSOCKS5GreetWriteData{
0x05, 0x01, 0x00}; // no authentication
int SOCKS5ClientSocket::DoGreetWrite() {
// Since we only have 1 byte to send the hostname length in, if the
// URL has a hostname longer than 255 characters we can't send it.
if (0xFF < destination_.host().size()) {
net_log_.AddEvent(NetLogEventType::SOCKS_HOSTNAME_TOO_BIG);
return ERR_SOCKS_CONNECTION_FAILED;
}
if (!write_buf_) {
auto greet_buffer =
base::MakeRefCounted<WrappedIOBuffer>(kSOCKS5GreetWriteData);
write_buf_ = base::MakeRefCounted<DrainableIOBuffer>(
std::move(greet_buffer), greet_buffer->size());
}
next_state_ = STATE_GREET_WRITE_COMPLETE;
return transport_socket_->Write(write_buf_.get(),
write_buf_->BytesRemaining(), io_callback_,
traffic_annotation_);
}
int SOCKS5ClientSocket::DoGreetWriteComplete(int result) {
if (result < 0)
return result;
write_buf_->DidConsume(result);
if (write_buf_->BytesRemaining() == 0) {
write_buf_.reset();
next_state_ = STATE_GREET_READ;
} else {
next_state_ = STATE_GREET_WRITE;
}
return OK;
}
int SOCKS5ClientSocket::DoGreetRead() {
next_state_ = STATE_GREET_READ_COMPLETE;
if (!read_buf_) {
read_buf_ = base::MakeRefCounted<GrowableIOBuffer>();
read_buf_->SetCapacity(kGreetReadHeaderSize);
}
return transport_socket_->Read(read_buf_.get(),
read_buf_->RemainingCapacity(), io_callback_);
}
int SOCKS5ClientSocket::DoGreetReadComplete(int result) {
if (result < 0)
return result;
if (result == 0) {
net_log_.AddEvent(
NetLogEventType::SOCKS_UNEXPECTEDLY_CLOSED_DURING_GREETING);
return ERR_SOCKS_CONNECTION_FAILED;
}
read_buf_->set_offset(read_buf_->offset() + result);
if (read_buf_->RemainingCapacity() > 0) {
next_state_ = STATE_GREET_READ;
return OK;
}
// Got the greet data.
base::span<uint8_t> read_data = read_buf_->span_before_offset();
if (read_data[0] != kSOCKS5Version) {
net_log_.AddEventWithIntParams(NetLogEventType::SOCKS_UNEXPECTED_VERSION,
"version", read_data[0]);
return ERR_SOCKS_CONNECTION_FAILED;
}
if (read_data[1] != 0x00) {
net_log_.AddEventWithIntParams(NetLogEventType::SOCKS_UNEXPECTED_AUTH,
"method", read_data[1]);
return ERR_SOCKS_CONNECTION_FAILED;
}
read_buf_.reset();
next_state_ = STATE_HANDSHAKE_WRITE;
return OK;
}
scoped_refptr<DrainableIOBuffer> SOCKS5ClientSocket::BuildHandshakeWriteBuffer()
const {
std::vector<uint8_t> handshake;
handshake.reserve(7 + destination_.host().size());
handshake.push_back(kSOCKS5Version);
handshake.push_back(kTunnelCommand); // Connect command
handshake.push_back(kNullByte); // Reserved null
handshake.push_back(kEndPointDomain); // The type of the address.
// First add the size of the hostname, followed by the hostname. The length of
// the hostname must fit within one byte.
const auto& host = destination_.host();
handshake.push_back(base::checked_cast<uint8_t>(host.size()));
handshake.insert(handshake.end(), host.begin(), host.end());
auto nw_port = base::U16ToBigEndian(destination_.port());
handshake.insert(handshake.end(), nw_port.begin(), nw_port.end());
auto base_buffer = base::MakeRefCounted<VectorIOBuffer>(std::move(handshake));
return base::MakeRefCounted<DrainableIOBuffer>(std::move(base_buffer),
base_buffer->size());
}
// Writes the SOCKS handshake data to the underlying socket connection.
int SOCKS5ClientSocket::DoHandshakeWrite() {
next_state_ = STATE_HANDSHAKE_WRITE_COMPLETE;
if (!write_buf_) {
write_buf_ = BuildHandshakeWriteBuffer();
}
return transport_socket_->Write(write_buf_.get(),
write_buf_->BytesRemaining(), io_callback_,
traffic_annotation_);
}
int SOCKS5ClientSocket::DoHandshakeWriteComplete(int result) {
if (result < 0)
return result;
// We ignore the case when result is 0, since the underlying Write
// may return spurious writes while waiting on the socket.
write_buf_->DidConsume(result);
if (write_buf_->BytesRemaining() == 0) {
write_buf_.reset();
next_state_ = STATE_HANDSHAKE_READ;
} else {
next_state_ = STATE_HANDSHAKE_WRITE;
}
return OK;
}
int SOCKS5ClientSocket::DoHandshakeRead() {
next_state_ = STATE_HANDSHAKE_READ_COMPLETE;
if (!read_buf_) {
read_buf_ = base::MakeRefCounted<GrowableIOBuffer>();
read_buf_->SetCapacity(kReadHeaderSize);
}
return transport_socket_->Read(read_buf_.get(),
read_buf_->RemainingCapacity(), io_callback_);
}
int SOCKS5ClientSocket::DoHandshakeReadComplete(int result) {
if (result < 0)
return result;
// The underlying socket closed unexpectedly.
if (result == 0) {
net_log_.AddEvent(
NetLogEventType::SOCKS_UNEXPECTEDLY_CLOSED_DURING_HANDSHAKE);
return ERR_SOCKS_CONNECTION_FAILED;
}
read_buf_->set_offset(read_buf_->offset() + result);
// When the first few bytes are read, check how many more are required
// and accordingly increase them
if (read_buf_->offset() == kReadHeaderSize) {
base::span<uint8_t> read_data = read_buf_->span_before_offset();
if (read_data[0] != kSOCKS5Version || read_data[2] != kNullByte) {
net_log_.AddEventWithIntParams(NetLogEventType::SOCKS_UNEXPECTED_VERSION,
"version", read_data[0]);
return ERR_SOCKS_CONNECTION_FAILED;
}
if (read_data[1] != 0x00) {
net_log_.AddEventWithIntParams(NetLogEventType::SOCKS_SERVER_ERROR,
"error_code", read_data[1]);
return ERR_SOCKS_CONNECTION_FAILED;
}
// We check the type of IP/Domain the server returns and accordingly
// increase the size of the response. For domains, we need to read the
// size of the domain, so the initial request size is upto the domain
// size. Since for IPv4/IPv6 the size is fixed and hence no 'size' is
// read, we substract 1 byte from the additional request size.
SocksEndPointAddressType address_type =
static_cast<SocksEndPointAddressType>(read_data[3]);
int additional_bytes_expected = 0;
if (address_type == kEndPointDomain) {
additional_bytes_expected += read_data[4];
} else if (address_type == kEndPointResolvedIPv4) {
additional_bytes_expected += sizeof(struct in_addr) - 1;
} else if (address_type == kEndPointResolvedIPv6) {
additional_bytes_expected += sizeof(struct in6_addr) - 1;
} else {
net_log_.AddEventWithIntParams(
NetLogEventType::SOCKS_UNKNOWN_ADDRESS_TYPE, "address_type",
read_data[3]);
return ERR_SOCKS_CONNECTION_FAILED;
}
additional_bytes_expected += 2; // for the port.
// Update capacity.
read_buf_->SetCapacity(kReadHeaderSize + additional_bytes_expected);
next_state_ = STATE_HANDSHAKE_READ;
return OK;
}
// When the final bytes are read, setup handshake. We ignore the rest
// of the response since they represent the SOCKSv5 endpoint and have
// no use when doing a tunnel connection.
if (read_buf_->RemainingCapacity() == 0) {
completed_handshake_ = true;
read_buf_.reset();
next_state_ = STATE_NONE;
return OK;
}
next_state_ = STATE_HANDSHAKE_READ;
return OK;
}
int SOCKS5ClientSocket::GetPeerAddress(IPEndPoint* address) const {
return transport_socket_->GetPeerAddress(address);
}
int SOCKS5ClientSocket::GetLocalAddress(IPEndPoint* address) const {
return transport_socket_->GetLocalAddress(address);
}
} // namespace net