transmission/libtransmission/peer-io.cc

793 lines
21 KiB
C++

// This file Copyright © Mnemosyne LLC.
// It may be used under GPLv2 (SPDX: GPL-2.0-only), GPLv3 (SPDX: GPL-3.0-only),
// or any future license endorsed by Mnemosyne LLC.
// License text can be found in the licenses/ folder.
#include <cerrno>
#include <cstdint>
#include <functional>
#include <type_traits> // std::underlying_type_t
#ifdef _WIN32
#include <ws2tcpip.h>
#else
#include <arpa/inet.h> // ntohl, ntohs
#endif
#include <event2/event.h>
#include <libutp/utp.h>
#include <fmt/core.h>
#include <small/map.hpp>
#include "libtransmission/transmission.h"
#include "libtransmission/bandwidth.h"
#include "libtransmission/block-info.h" // tr_block_info
#include "libtransmission/error.h"
#include "libtransmission/log.h"
#include "libtransmission/net.h"
#include "libtransmission/peer-io.h"
#include "libtransmission/peer-socket.h" // tr_peer_socket, tr_netOpen...
#include "libtransmission/session.h"
#include "libtransmission/tr-assert.h"
#include "libtransmission/utils.h" // for _()
struct sockaddr;
#define tr_logAddErrorIo(io, msg) tr_logAddError(msg, (io)->display_name())
#define tr_logAddWarnIo(io, msg) tr_logAddWarn(msg, (io)->display_name())
#define tr_logAddDebugIo(io, msg) tr_logAddDebug(msg, (io)->display_name())
#define tr_logAddTraceIo(io, msg) tr_logAddTrace(msg, (io)->display_name())
namespace
{
// Helps us to ignore errors that say "try again later"
// since that's what peer-io does by default anyway.
[[nodiscard]] constexpr auto can_retry_from_error(int error_code) noexcept
{
#ifdef _WIN32
return error_code == 0 || error_code == WSAEWOULDBLOCK || error_code == WSAEINTR || error_code == WSAEINPROGRESS;
#else
return error_code == 0 || error_code == EAGAIN || error_code == EWOULDBLOCK || error_code == EINTR ||
error_code == EINPROGRESS;
#endif
}
size_t get_desired_output_buffer_size(tr_peerIo const* io, uint64_t now)
{
// this is all kind of arbitrary, but what seems to work well is
// being large enough to hold the next 20 seconds' worth of input,
// or a few blocks, whichever is bigger. OK to tweak this as needed.
static auto constexpr PeriodSecs = uint64_t{ 15U };
// the 3 is an arbitrary number of blocks;
// the .5 is to leave room for protocol messages
static auto constexpr Floor = static_cast<uint64_t>(tr_block_info::BlockSize * 3.5);
auto const current_speed = io->get_piece_speed(now, TR_UP);
return std::max(Floor, current_speed.base_quantity() * PeriodSecs);
}
} // namespace
// ---
tr_peerIo::tr_peerIo(
tr_session* session,
tr_sha1_digest_t const* info_hash,
bool is_incoming,
bool is_seed,
tr_bandwidth* parent_bandwidth)
: bandwidth_{ parent_bandwidth }
, info_hash_{ info_hash != nullptr ? *info_hash : tr_sha1_digest_t{} }
, session_{ session }
, is_seed_{ is_seed }
, is_incoming_{ is_incoming }
{
}
std::shared_ptr<tr_peerIo> tr_peerIo::create(
tr_session* session,
tr_bandwidth* parent,
tr_sha1_digest_t const* info_hash,
bool is_incoming,
bool is_seed)
{
TR_ASSERT(session != nullptr);
auto lock = session->unique_lock();
auto io = std::make_shared<tr_peerIo>(session, info_hash, is_incoming, is_seed, parent);
io->bandwidth().set_peer(io);
tr_logAddTraceIo(io, fmt::format("bandwidth is {}; its parent is {}", fmt::ptr(&io->bandwidth()), fmt::ptr(parent)));
return io;
}
std::shared_ptr<tr_peerIo> tr_peerIo::new_incoming(tr_session* session, tr_bandwidth* parent, tr_peer_socket socket)
{
TR_ASSERT(session != nullptr);
auto peer_io = tr_peerIo::create(session, parent, nullptr, true, false);
peer_io->set_socket(std::move(socket));
return peer_io;
}
std::shared_ptr<tr_peerIo> tr_peerIo::new_outgoing(
tr_session* session,
tr_bandwidth* parent,
tr_socket_address const& socket_address,
tr_sha1_digest_t const& info_hash,
bool is_seed,
bool utp)
{
using preferred_key_t = std::underlying_type_t<tr_preferred_transport>;
auto const preferred = session->preferred_transport();
TR_ASSERT(!tr_peer_socket::limit_reached(session));
TR_ASSERT(session != nullptr);
TR_ASSERT(socket_address.is_valid());
TR_ASSERT(utp || session->allowsTCP());
if (!socket_address.is_valid_for_peers())
{
return {};
}
auto peer_io = tr_peerIo::create(session, parent, &info_hash, false, is_seed);
auto const func = small::max_size_map<preferred_key_t, std::function<bool()>, TR_NUM_PREFERRED_TRANSPORT>{
{ TR_PREFER_UTP,
[&]()
{
#ifdef WITH_UTP
if (utp)
{
auto* const sock = utp_create_socket(session->utp_context);
utp_set_userdata(sock, peer_io.get());
peer_io->set_socket(tr_peer_socket{ socket_address, sock });
auto const [ss, sslen] = socket_address.to_sockaddr();
if (utp_connect(sock, reinterpret_cast<sockaddr const*>(&ss), sslen) == 0)
{
return true;
}
}
#endif
return false;
} },
{ TR_PREFER_TCP,
[&]()
{
if (!peer_io->socket_.is_valid())
{
if (auto sock = tr_netOpenPeerSocket(session, socket_address, is_seed); sock.is_valid())
{
peer_io->set_socket(std::move(sock));
return true;
}
}
return false;
} }
};
if (func.at(preferred)())
{
return peer_io;
}
for (preferred_key_t i = 0U; i < TR_NUM_PREFERRED_TRANSPORT; ++i)
{
if (i != preferred && func.at(i)())
{
return peer_io;
}
}
return {};
}
tr_peerIo::~tr_peerIo()
{
auto const lock = session_->unique_lock();
clear_callbacks();
tr_logAddTraceIo(this, "in tr_peerIo destructor");
event_disable(EV_READ | EV_WRITE);
close();
}
// ---
void tr_peerIo::set_socket(tr_peer_socket socket_in)
{
close(); // tear down the previous socket, if any
socket_ = std::move(socket_in);
if (socket_.is_tcp())
{
event_read_.reset(event_new(session_->event_base(), socket_.handle.tcp, EV_READ, &tr_peerIo::event_read_cb, this));
event_write_.reset(event_new(session_->event_base(), socket_.handle.tcp, EV_WRITE, &tr_peerIo::event_write_cb, this));
}
#ifdef WITH_UTP
else if (socket_.is_utp())
{
utp_set_userdata(socket_.handle.utp, this);
}
#endif
else
{
TR_ASSERT_MSG(false, "unsupported peer socket type");
}
}
void tr_peerIo::close()
{
socket_.close();
event_write_.reset();
event_read_.reset();
}
void tr_peerIo::clear()
{
clear_callbacks();
set_enabled(TR_UP, false);
set_enabled(TR_DOWN, false);
close();
}
bool tr_peerIo::reconnect()
{
TR_ASSERT(!this->is_incoming());
TR_ASSERT(this->session_->allowsTCP());
short int const pending_events = this->pending_events_;
event_disable(EV_READ | EV_WRITE);
close();
if (tr_peer_socket::limit_reached(session_))
{
return false;
}
socket_ = tr_netOpenPeerSocket(session_, socket_address(), is_seed());
if (!socket_.is_tcp())
{
return false;
}
this->event_read_.reset(event_new(session_->event_base(), socket_.handle.tcp, EV_READ, event_read_cb, this));
this->event_write_.reset(event_new(session_->event_base(), socket_.handle.tcp, EV_WRITE, event_write_cb, this));
event_enable(pending_events);
return true;
}
// ---
void tr_peerIo::did_write_wrapper(size_t bytes_transferred)
{
auto const keep_alive = shared_from_this();
while (bytes_transferred != 0U && !std::empty(outbuf_info_))
{
auto& [n_bytes_left, is_piece_data] = outbuf_info_.front();
size_t const payload = std::min(uint64_t{ n_bytes_left }, uint64_t{ bytes_transferred });
/* For µTP sockets, the overhead is computed in utp_on_overhead. */
size_t const overhead = socket_.guess_packet_overhead(payload);
uint64_t const now = tr_time_msec();
bandwidth().notify_bandwidth_consumed(TR_UP, payload, is_piece_data, now);
if (overhead > 0U)
{
bandwidth().notify_bandwidth_consumed(TR_UP, overhead, false, now);
}
if (did_write_ != nullptr)
{
did_write_(this, payload, is_piece_data, user_data_);
}
bytes_transferred -= payload;
n_bytes_left -= payload;
if (n_bytes_left == 0U)
{
outbuf_info_.pop_front();
}
}
}
size_t tr_peerIo::try_write(size_t max)
{
static auto constexpr Dir = TR_UP;
if (max == 0U)
{
return {};
}
auto& buf = outbuf_;
max = std::min(max, std::size(buf));
max = bandwidth().clamp(Dir, max);
if (max == 0U)
{
set_enabled(Dir, false);
return {};
}
auto error = tr_error{};
auto const n_written = socket_.try_write(buf, max, &error);
// enable further writes if there's more data to write
set_enabled(Dir, !std::empty(buf) && (!error || can_retry_from_error(error.code())));
if (error)
{
if (!can_retry_from_error(error.code()))
{
tr_logAddTraceIo(
this,
fmt::format("try_write err: wrote:{}, errno:{} ({})", n_written, error.code(), error.message()));
call_error_callback(error);
}
}
else if (n_written > 0U)
{
did_write_wrapper(n_written);
}
return n_written;
}
void tr_peerIo::event_write_cb([[maybe_unused]] evutil_socket_t fd, short /*event*/, void* vio)
{
auto* const io = static_cast<tr_peerIo*>(vio);
tr_logAddTraceIo(io, "libevent says this peer socket is ready for writing");
TR_ASSERT(io->socket_.is_tcp());
TR_ASSERT(io->socket_.handle.tcp == fd);
io->pending_events_ &= ~EV_WRITE;
// Write as much as possible. Since the socket is non-blocking,
// write() will return if it can't write any more without blocking
io->try_write(SIZE_MAX);
}
// ---
void tr_peerIo::can_read_wrapper()
{
// try to consume the input buffer
if (can_read_ == nullptr)
{
return;
}
auto const lock = session_->unique_lock();
auto const keep_alive = shared_from_this();
auto const now = tr_time_msec();
auto done = bool{ false };
auto err = bool{ false };
// In normal conditions, only continue processing if we still have bandwidth
// quota for it.
//
// The read buffer will grow indefinitely if libutp or the TCP stack keeps buffering
// data faster than the bandwidth limit allows. To safeguard against that, we keep
// processing if the read buffer is more than twice as large as the target size.
while (!done && !err && (read_buffer_size() > RcvBuf * 2U || bandwidth().clamp(TR_DOWN, read_buffer_size()) != 0U))
{
size_t piece = 0U;
auto const old_len = read_buffer_size();
auto const read_state = can_read_ != nullptr ? can_read_(this, user_data_, &piece) : READ_ERR;
auto const used = old_len - read_buffer_size();
auto const overhead = socket_.guess_packet_overhead(used);
if (piece != 0U)
{
bandwidth().notify_bandwidth_consumed(TR_DOWN, piece, true, now);
}
if (used != piece)
{
bandwidth().notify_bandwidth_consumed(TR_DOWN, used - piece, false, now);
}
if (overhead > 0U)
{
bandwidth().notify_bandwidth_consumed(TR_DOWN, overhead, false, now);
}
switch (read_state)
{
case READ_NOW:
if (!std::empty(inbuf_))
{
continue;
}
done = true;
break;
case READ_LATER:
done = true;
break;
case READ_ERR:
err = true;
break;
}
}
}
size_t tr_peerIo::try_read(size_t max)
{
static auto constexpr Dir = TR_DOWN;
if (max == 0U)
{
return {};
}
// Do not write more than the bandwidth allows.
// If there is no bandwidth left available, disable writes.
max = bandwidth().clamp(Dir, max);
if (max == 0U)
{
set_enabled(Dir, false);
return {};
}
auto& buf = inbuf_;
auto error = tr_error{};
auto const n_read = socket_.try_read(buf, max, std::empty(buf), &error);
set_enabled(Dir, !error || can_retry_from_error(error.code()));
if (error)
{
if (!can_retry_from_error(error.code()))
{
tr_logAddTraceIo(this, fmt::format("try_read err: n_read:{} errno:{} ({})", n_read, error.code(), error.message()));
call_error_callback(error);
}
}
else if (!std::empty(buf))
{
can_read_wrapper();
}
return n_read;
}
void tr_peerIo::event_read_cb([[maybe_unused]] evutil_socket_t fd, short /*event*/, void* vio)
{
static auto constexpr MaxLen = RcvBuf;
auto* const io = static_cast<tr_peerIo*>(vio);
tr_logAddTraceIo(io, "libevent says this peer socket is ready for reading");
TR_ASSERT(io->socket_.is_tcp());
TR_ASSERT(io->socket_.handle.tcp == fd);
io->pending_events_ &= ~EV_READ;
// if we don't have any bandwidth left, stop reading
auto const n_used = std::size(io->inbuf_);
auto const n_left = n_used >= MaxLen ? 0U : MaxLen - n_used;
io->try_read(n_left);
}
// ---
void tr_peerIo::event_enable(short event)
{
TR_ASSERT(session_ != nullptr);
bool const need_events = socket_.is_tcp();
TR_ASSERT(!need_events || event_read_);
TR_ASSERT(!need_events || event_write_);
if ((event & EV_READ) != 0 && (pending_events_ & EV_READ) == 0)
{
tr_logAddTraceIo(this, "enabling ready-to-read polling");
if (need_events)
{
event_add(event_read_.get(), nullptr);
}
pending_events_ |= EV_READ;
}
if ((event & EV_WRITE) != 0 && (pending_events_ & EV_WRITE) == 0)
{
tr_logAddTraceIo(this, "enabling ready-to-write polling");
if (need_events)
{
event_add(event_write_.get(), nullptr);
}
pending_events_ |= EV_WRITE;
}
}
void tr_peerIo::event_disable(short event)
{
bool const need_events = socket_.is_tcp();
TR_ASSERT(!need_events || event_read_);
TR_ASSERT(!need_events || event_write_);
if ((event & EV_READ) != 0 && (pending_events_ & EV_READ) != 0)
{
tr_logAddTraceIo(this, "disabling ready-to-read polling");
if (need_events)
{
event_del(event_read_.get());
}
pending_events_ &= ~EV_READ;
}
if ((event & EV_WRITE) != 0 && (pending_events_ & EV_WRITE) != 0)
{
tr_logAddTraceIo(this, "disabling ready-to-write polling");
if (need_events)
{
event_del(event_write_.get());
}
pending_events_ &= ~EV_WRITE;
}
}
void tr_peerIo::set_enabled(tr_direction dir, bool is_enabled)
{
TR_ASSERT(tr_isDirection(dir));
short const event = dir == TR_UP ? EV_WRITE : EV_READ;
if (is_enabled)
{
event_enable(event);
}
else
{
event_disable(event);
}
}
size_t tr_peerIo::flush(tr_direction dir, size_t limit)
{
TR_ASSERT(tr_isDirection(dir));
return dir == TR_DOWN ? try_read(limit) : try_write(limit);
}
size_t tr_peerIo::flush_outgoing_protocol_msgs()
{
size_t byte_count = 0U;
/* count up how many bytes are used by non-piece-data messages
at the front of our outbound queue */
for (auto const& [n_bytes, is_piece_data] : outbuf_info_)
{
if (is_piece_data)
{
break;
}
byte_count += n_bytes;
}
return flush(TR_UP, byte_count);
}
// ---
size_t tr_peerIo::get_write_buffer_space(uint64_t now) const noexcept
{
size_t const desired_len = get_desired_output_buffer_size(this, now);
size_t const current_len = std::size(outbuf_);
return desired_len > current_len ? desired_len - current_len : 0U;
}
// ---
void tr_peerIo::read_bytes(void* bytes, size_t n_bytes)
{
auto walk = reinterpret_cast<std::byte*>(bytes);
n_bytes = std::min(n_bytes, std::size(inbuf_));
if (decrypt_remain_len_)
{
if (*decrypt_remain_len_ <= n_bytes)
{
filter_.decrypt(std::data(inbuf_), *decrypt_remain_len_, walk);
inbuf_.drain(*decrypt_remain_len_);
if (walk != nullptr)
{
walk += *decrypt_remain_len_;
}
n_bytes -= *decrypt_remain_len_;
filter_.decrypt_disable();
decrypt_remain_len_.reset();
}
else
{
*decrypt_remain_len_ -= n_bytes;
}
}
filter_.decrypt(std::data(inbuf_), n_bytes, walk);
inbuf_.drain(n_bytes);
}
void tr_peerIo::read_uint16(uint16_t* setme)
{
auto tmp = uint16_t{};
read_bytes(&tmp, sizeof(tmp));
*setme = ntohs(tmp);
}
void tr_peerIo::read_uint32(uint32_t* setme)
{
auto tmp = uint32_t{};
read_bytes(&tmp, sizeof(tmp));
*setme = ntohl(tmp);
}
// --- UTP
#ifdef WITH_UTP
void tr_peerIo::on_utp_state_change(int state)
{
if (state == UTP_STATE_CONNECT)
{
tr_logAddTraceIo(this, "utp_on_state_change -- changed to connected");
utp_supported_ = true;
}
else if (state == UTP_STATE_WRITABLE)
{
tr_logAddTraceIo(this, "utp_on_state_change -- changed to writable");
if ((pending_events_ & EV_WRITE) != 0)
{
try_write(SIZE_MAX);
}
}
else if (state == UTP_STATE_EOF)
{
auto error = tr_error{};
error.set_from_errno(ENOTCONN);
call_error_callback(error);
}
else if (state == UTP_STATE_DESTROYING)
{
tr_logAddErrorIo(this, "Impossible state UTP_STATE_DESTROYING");
}
else
{
tr_logAddErrorIo(this, fmt::format(_("Unknown state: {state}"), fmt::arg("state", state)));
}
}
void tr_peerIo::on_utp_error(int errcode)
{
tr_logAddTraceIo(this, fmt::format("utp_on_error -- {}", utp_error_code_names[errcode]));
if (got_error_ == nullptr)
{
return;
}
auto error = tr_error{};
switch (errcode)
{
case UTP_ECONNREFUSED:
error.set_from_errno(ECONNREFUSED);
break;
case UTP_ECONNRESET:
error.set_from_errno(ECONNRESET);
break;
case UTP_ETIMEDOUT:
error.set_from_errno(ETIMEDOUT);
break;
default:
error.set(errcode, utp_error_code_names[errcode]);
break;
}
call_error_callback(error);
}
#endif /* #ifdef WITH_UTP */
void tr_peerIo::utp_init([[maybe_unused]] struct_utp_context* ctx)
{
#ifdef WITH_UTP
utp_context_set_option(ctx, UTP_RCVBUF, RcvBuf);
// note: all the callback handlers here need to check `userdata` for nullptr
// because libutp can fire callbacks on a socket after utp_close() is called
utp_set_callback(
ctx,
UTP_ON_READ,
[](utp_callback_arguments* args) -> uint64
{
if (auto* const io = static_cast<tr_peerIo*>(utp_get_userdata(args->socket)); io != nullptr)
{
io->inbuf_.add(args->buf, args->len);
io->set_enabled(TR_DOWN, true);
io->can_read_wrapper();
// utp_read_drained() notifies libutp that this read buffer is empty.
// It opens up the congestion window by sending an ACK (soonish) if
// one was not going to be sent.
if (std::empty(io->inbuf_))
{
utp_read_drained(args->socket);
}
}
return {};
});
utp_set_callback(
ctx,
UTP_GET_READ_BUFFER_SIZE,
[](utp_callback_arguments* args) -> uint64
{
if (auto const* const io = static_cast<tr_peerIo*>(utp_get_userdata(args->socket)); io != nullptr)
{
return io->read_buffer_size();
}
return {};
});
utp_set_callback(
ctx,
UTP_ON_ERROR,
[](utp_callback_arguments* args) -> uint64
{
if (auto* const io = static_cast<tr_peerIo*>(utp_get_userdata(args->socket)); io != nullptr)
{
io->on_utp_error(args->error_code);
}
return {};
});
utp_set_callback(
ctx,
UTP_ON_OVERHEAD_STATISTICS,
[](utp_callback_arguments* args) -> uint64
{
if (auto* const io = static_cast<tr_peerIo*>(utp_get_userdata(args->socket)); io != nullptr)
{
tr_logAddTraceIo(io, fmt::format("{:d} overhead bytes via utp", args->len));
io->bandwidth().notify_bandwidth_consumed(args->send != 0 ? TR_UP : TR_DOWN, args->len, false, tr_time_msec());
}
return {};
});
utp_set_callback(
ctx,
UTP_ON_STATE_CHANGE,
[](utp_callback_arguments* args) -> uint64
{
if (auto* const io = static_cast<tr_peerIo*>(utp_get_userdata(args->socket)); io != nullptr)
{
io->on_utp_state_change(args->state);
}
return {};
});
#endif
}