gcc/libstdc++-v3/include/experimental/io_context
Jonathan Wakely 3084625d39 Improve docs for Networking TS
* include/experimental/bits/net.h: Fix/improve doxygen markup.
	* include/experimental/buffer: Likewise.
	* include/experimental/executor: Likewise.
	* include/experimental/internet: Likewise.
	* include/experimental/io_context: Likewise.
	* include/experimental/net: Likewise.
	* include/experimental/netfwd: Likewise.
	* include/experimental/socket: Likewise.
	* include/experimental/timer: Likewise.

From-SVN: r270810
2019-05-02 16:46:34 +01:00

873 lines
21 KiB
C++

// <experimental/io_service> -*- C++ -*-
// Copyright (C) 2015-2019 Free Software Foundation, Inc.
//
// This file is part of the GNU ISO C++ Library. This library is free
// software; you can redistribute it and/or modify it under the
// terms of the GNU General Public License as published by the
// Free Software Foundation; either version 3, or (at your option)
// any later version.
// This library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// Under Section 7 of GPL version 3, you are granted additional
// permissions described in the GCC Runtime Library Exception, version
// 3.1, as published by the Free Software Foundation.
// You should have received a copy of the GNU General Public License and
// a copy of the GCC Runtime Library Exception along with this program;
// see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
// <http://www.gnu.org/licenses/>.
/** @file experimental/io_context
* This is a TS C++ Library header.
* @ingroup networking-ts
*/
#ifndef _GLIBCXX_EXPERIMENTAL_IO_SERVICE
#define _GLIBCXX_EXPERIMENTAL_IO_SERVICE 1
#pragma GCC system_header
#if __cplusplus >= 201402L
#include <atomic>
#include <chrono>
#include <forward_list>
#include <functional>
#include <system_error>
#include <thread>
#include <experimental/netfwd>
#include <experimental/executor>
#if _GLIBCXX_HAVE_UNISTD_H
# include <unistd.h>
#endif
#ifdef _GLIBCXX_HAVE_POLL_H
# include <poll.h>
#endif
#ifdef _GLIBCXX_HAVE_FCNTL_H
# include <fcntl.h>
#endif
namespace std _GLIBCXX_VISIBILITY(default)
{
_GLIBCXX_BEGIN_NAMESPACE_VERSION
namespace experimental
{
namespace net
{
inline namespace v1
{
/** @addtogroup networking-ts
* @{
*/
class __socket_impl;
/// An ExecutionContext for I/O operations.
class io_context : public execution_context
{
public:
// types:
/// An executor for an io_context.
class executor_type
{
public:
// construct / copy / destroy:
executor_type(const executor_type& __other) noexcept = default;
executor_type(executor_type&& __other) noexcept = default;
executor_type& operator=(const executor_type& __other) noexcept = default;
executor_type& operator=(executor_type&& __other) noexcept = default;
// executor operations:
bool running_in_this_thread() const noexcept
{
lock_guard<mutex> __lock(_M_ctx->_M_mtx);
auto __end = _M_ctx->_M_call_stack.end();
return std::find(_M_ctx->_M_call_stack.begin(), __end,
this_thread::get_id()) != __end;
}
io_context& context() const noexcept { return *_M_ctx; }
void on_work_started() const noexcept { ++_M_ctx->_M_work_count; }
void on_work_finished() const noexcept { --_M_ctx->_M_work_count; }
template<typename _Func, typename _ProtoAllocator>
void
dispatch(_Func&& __f, const _ProtoAllocator& __a) const
{
if (running_in_this_thread())
decay_t<_Func>{std::forward<_Func>(__f)}();
else
post(std::forward<_Func>(__f), __a);
}
template<typename _Func, typename _ProtoAllocator>
void
post(_Func&& __f, const _ProtoAllocator& __a) const
{
lock_guard<mutex> __lock(_M_ctx->_M_mtx);
// TODO (re-use functionality in system_context)
_M_ctx->_M_reactor._M_notify();
}
template<typename _Func, typename _ProtoAllocator>
void
defer(_Func&& __f, const _ProtoAllocator& __a) const
{ post(std::forward<_Func>(__f), __a); }
private:
friend io_context;
explicit
executor_type(io_context& __ctx) : _M_ctx(std::addressof(__ctx)) { }
io_context* _M_ctx;
};
using count_type = size_t;
// construct / copy / destroy:
io_context() : _M_work_count(0) { }
explicit
io_context(int __concurrency_hint) : _M_work_count(0) { }
io_context(const io_context&) = delete;
io_context& operator=(const io_context&) = delete;
// io_context operations:
executor_type get_executor() noexcept { return executor_type(*this); }
count_type
run()
{
count_type __n = 0;
while (run_one())
if (__n != numeric_limits<count_type>::max())
++__n;
return __n;
}
template<typename _Rep, typename _Period>
count_type
run_for(const chrono::duration<_Rep, _Period>& __rel_time)
{ return run_until(chrono::steady_clock::now() + __rel_time); }
template<typename _Clock, typename _Duration>
count_type
run_until(const chrono::time_point<_Clock, _Duration>& __abs_time)
{
count_type __n = 0;
while (run_one_until(__abs_time))
if (__n != numeric_limits<count_type>::max())
++__n;
return __n;
}
count_type
run_one()
{ return _M_do_one(chrono::milliseconds{-1}); }
template<typename _Rep, typename _Period>
count_type
run_one_for(const chrono::duration<_Rep, _Period>& __rel_time)
{ return run_one_until(chrono::steady_clock::now() + __rel_time); }
template<typename _Clock, typename _Duration>
count_type
run_one_until(const chrono::time_point<_Clock, _Duration>& __abs_time)
{
auto __now = _Clock::now();
while (__now < __abs_time)
{
using namespace std::chrono;
auto __ms = duration_cast<milliseconds>(__abs_time - __now);
if (_M_do_one(__ms))
return 1;
__now = _Clock::now();
}
return 0;
}
count_type
poll()
{
count_type __n = 0;
while (poll_one())
if (__n != numeric_limits<count_type>::max())
++__n;
return __n;
}
count_type
poll_one()
{ return _M_do_one(chrono::milliseconds{0}); }
void stop()
{
lock_guard<mutex> __lock(_M_mtx);
_M_stopped = true;
_M_reactor._M_notify();
}
bool stopped() const noexcept
{
lock_guard<mutex> __lock(_M_mtx);
return _M_stopped;
}
void restart()
{
_M_stopped = false;
}
private:
template<typename _Clock, typename _WaitTraits>
friend class basic_waitable_timer;
friend __socket_impl;
template<typename _Protocol>
friend class __basic_socket_impl;
template<typename _Protocol>
friend class basic_socket;
template<typename _Protocol>
friend class basic_datagram_socket;
template<typename _Protocol>
friend class basic_stream_socket;
template<typename _Protocol>
friend class basic_socket_acceptor;
count_type
_M_outstanding_work() const
{ return _M_work_count + !_M_ops.empty(); }
struct __timer_queue_base : execution_context::service
{
// return milliseconds until next timer expires, or milliseconds::max()
virtual chrono::milliseconds _M_next() const = 0;
virtual bool run_one() = 0;
protected:
explicit
__timer_queue_base(execution_context& __ctx) : service(__ctx)
{
auto& __ioc = static_cast<io_context&>(__ctx);
lock_guard<mutex> __lock(__ioc._M_mtx);
__ioc._M_timers.push_back(this);
}
mutable mutex _M_qmtx;
};
template<typename _Timer, typename _Key = typename _Timer::_Key>
struct __timer_queue : __timer_queue_base
{
using key_type = __timer_queue;
explicit
__timer_queue(execution_context& __ctx) : __timer_queue_base(__ctx)
{ }
void shutdown() noexcept { }
io_context& context() noexcept
{ return static_cast<io_context&>(service::context()); }
// Start an asynchronous wait.
void
push(const _Timer& __t, function<void(error_code)> __h)
{
context().get_executor().on_work_started();
lock_guard<mutex> __lock(_M_qmtx);
_M_queue.emplace(__t, _M_next_id++, std::move(__h));
// no need to notify reactor unless this timer went to the front?
}
// Cancel all outstanding waits for __t
size_t
cancel(const _Timer& __t)
{
lock_guard<mutex> __lock(_M_qmtx);
size_t __count = 0;
auto __last = _M_queue.end();
for (auto __it = _M_queue.begin(), __end = __last; __it != __end;
++__it)
{
if (__it->_M_key == __t._M_key.get())
{
__it->cancel();
__last = __it;
++__count;
}
}
if (__count)
_M_queue._M_sort_to(__last);
return __count;
}
// Cancel oldest outstanding wait for __t
bool
cancel_one(const _Timer& __t)
{
lock_guard<mutex> __lock(_M_qmtx);
const auto __end = _M_queue.end();
auto __oldest = __end;
for (auto __it = _M_queue.begin(); __it != __end; ++__it)
if (__it->_M_key == __t._M_key.get())
if (__oldest == __end || __it->_M_id < __oldest->_M_id)
__oldest = __it;
if (__oldest == __end)
return false;
__oldest->cancel();
_M_queue._M_sort_to(__oldest);
return true;
}
chrono::milliseconds
_M_next() const override
{
typename _Timer::time_point __exp;
{
lock_guard<mutex> __lock(_M_qmtx);
if (_M_queue.empty())
return chrono::milliseconds::max(); // no pending timers
if (_M_queue.top()._M_key == nullptr)
return chrono::milliseconds::zero(); // cancelled, run now
__exp = _M_queue.top()._M_expiry;
}
auto __dur = _Timer::traits_type::to_wait_duration(__exp);
if (__dur < __dur.zero())
__dur = __dur.zero();
return chrono::duration_cast<chrono::milliseconds>(__dur);
}
private:
bool run_one() override
{
auto __now = _Timer::clock_type::now();
function<void(error_code)> __h;
error_code __ec;
{
lock_guard<mutex> __lock(_M_qmtx);
if (_M_queue.top()._M_key == nullptr) // cancelled
{
__h = std::move(_M_queue.top()._M_h);
__ec = std::make_error_code(errc::operation_canceled);
_M_queue.pop();
}
else if (_M_queue.top()._M_expiry <= _Timer::clock_type::now())
{
__h = std::move(_M_queue.top()._M_h);
_M_queue.pop();
}
}
if (__h)
{
__h(__ec);
context().get_executor().on_work_finished();
return true;
}
return false;
}
using __timer_id_type = uint64_t;
struct __pending_timer
{
__pending_timer(const _Timer& __t, uint64_t __id,
function<void(error_code)> __h)
: _M_expiry(__t.expiry()), _M_key(__t._M_key.get()), _M_id(__id),
_M_h(std::move(__h))
{ }
typename _Timer::time_point _M_expiry;
_Key* _M_key;
__timer_id_type _M_id;
function<void(error_code)> _M_h;
void cancel() { _M_expiry = _M_expiry.min(); _M_key = nullptr; }
bool
operator<(const __pending_timer& __rhs) const
{ return _M_expiry < __rhs._M_expiry; }
};
struct __queue : priority_queue<__pending_timer>
{
using iterator =
typename priority_queue<__pending_timer>::container_type::iterator;
// expose begin/end/erase for direct access to underlying container
iterator begin() { return this->c.begin(); }
iterator end() { return this->c.end(); }
iterator erase(iterator __it) { return this->c.erase(__it); }
void
_M_sort_to(iterator __it)
{ std::stable_sort(this->c.begin(), ++__it); }
};
__queue _M_queue;
__timer_id_type _M_next_id = 0;
};
template<typename _Timer, typename _CompletionHandler>
void
async_wait(const _Timer& __timer, _CompletionHandler&& __h)
{
auto& __queue = use_service<__timer_queue<_Timer>>(*this);
__queue.push(__timer, std::move(__h));
_M_reactor._M_notify();
}
// Cancel all wait operations initiated by __timer.
template<typename _Timer>
size_t
cancel(const _Timer& __timer)
{
if (!has_service<__timer_queue<_Timer>>(*this))
return 0;
auto __c = use_service<__timer_queue<_Timer>>(*this).cancel(__timer);
if (__c != 0)
_M_reactor._M_notify();
return __c;
}
// Cancel the oldest wait operation initiated by __timer.
template<typename _Timer>
size_t
cancel_one(const _Timer& __timer)
{
if (!has_service<__timer_queue<_Timer>>(*this))
return 0;
if (use_service<__timer_queue<_Timer>>(*this).cancel_one(__timer))
{
_M_reactor._M_notify();
return 1;
}
return 0;
}
template<typename _Op>
void
async_wait(int __fd, int __w, _Op&& __op)
{
lock_guard<mutex> __lock(_M_mtx);
// TODO need push_back, use std::list not std::forward_list
auto __tail = _M_ops.before_begin(), __it = _M_ops.begin();
while (__it != _M_ops.end())
{
++__it;
++__tail;
}
using __type = __async_operation_impl<_Op>;
_M_ops.emplace_after(__tail,
make_unique<__type>(std::move(__op), __fd, __w));
_M_reactor._M_fd_interest(__fd, __w);
}
void _M_add_fd(int __fd) { _M_reactor._M_add_fd(__fd); }
void _M_remove_fd(int __fd) { _M_reactor._M_remove_fd(__fd); }
void cancel(int __fd, error_code&)
{
lock_guard<mutex> __lock(_M_mtx);
const auto __end = _M_ops.end();
auto __it = _M_ops.begin();
auto __prev = _M_ops.before_begin();
while (__it != __end && (*__it)->_M_is_cancelled())
{
++__it;
++__prev;
}
auto __cancelled = __prev;
while (__it != __end)
{
if ((*__it)->_M_fd == __fd)
{
(*__it)->cancel();
++__it;
_M_ops.splice_after(__cancelled, _M_ops, __prev);
++__cancelled;
}
else
{
++__it;
++__prev;
}
}
_M_reactor._M_not_interested(__fd);
}
struct __async_operation
{
__async_operation(int __fd, int __ev) : _M_fd(__fd), _M_ev(__ev) { }
virtual ~__async_operation() = default;
int _M_fd;
short _M_ev;
void cancel() { _M_fd = -1; }
bool _M_is_cancelled() const { return _M_fd == -1; }
virtual void run(io_context&) = 0;
};
template<typename _Op>
struct __async_operation_impl : __async_operation
{
__async_operation_impl(_Op&& __op, int __fd, int __ev)
: __async_operation{__fd, __ev}, _M_op(std::move(__op)) { }
_Op _M_op;
void run(io_context& __ctx)
{
if (_M_is_cancelled())
_M_op(std::make_error_code(errc::operation_canceled));
else
_M_op(error_code{});
}
};
atomic<count_type> _M_work_count;
mutable mutex _M_mtx;
queue<function<void()>> _M_op;
bool _M_stopped = false;
struct __monitor
{
__monitor(io_context& __c) : _M_ctx(__c)
{
lock_guard<mutex> __lock(_M_ctx._M_mtx);
_M_ctx._M_call_stack.push_back(this_thread::get_id());
}
~__monitor()
{
lock_guard<mutex> __lock(_M_ctx._M_mtx);
_M_ctx._M_call_stack.pop_back();
if (_M_ctx._M_outstanding_work() == 0)
{
_M_ctx._M_stopped = true;
_M_ctx._M_reactor._M_notify();
}
}
__monitor(__monitor&&) = delete;
io_context& _M_ctx;
};
bool
_M_do_one(chrono::milliseconds __timeout)
{
const bool __block = __timeout != chrono::milliseconds::zero();
__reactor::__fdvec __fds;
__monitor __mon{*this};
__timer_queue_base* __timerq = nullptr;
unique_ptr<__async_operation> __async_op;
while (true)
{
if (__timerq)
{
if (__timerq->run_one())
return true;
else
__timerq = nullptr;
}
if (__async_op)
{
__async_op->run(*this);
// TODO need to unregister __async_op
return true;
}
chrono::milliseconds __ms{0};
{
lock_guard<mutex> __lock(_M_mtx);
if (_M_stopped)
return false;
// find first timer with something to do
for (auto __q : _M_timers)
{
auto __next = __q->_M_next();
if (__next == __next.zero()) // ready to run immediately
{
__timerq = __q;
__ms = __next;
break;
}
else if (__next != __next.max() && __block
&& (__next < __ms || __timerq == nullptr))
{
__timerq = __q;
__ms = __next;
}
}
if (__timerq && __ms == __ms.zero())
continue; // restart loop to run a timer immediately
if (!_M_ops.empty() && _M_ops.front()->_M_is_cancelled())
{
_M_ops.front().swap(__async_op);
_M_ops.pop_front();
continue;
}
// TODO run any posted items
if (__block)
{
if (__timerq == nullptr)
__ms = __timeout;
else if (__ms.zero() <= __timeout && __timeout < __ms)
__ms = __timeout;
else if (__ms.count() > numeric_limits<int>::max())
__ms = chrono::milliseconds{numeric_limits<int>::max()};
}
// else __ms == 0 and poll() will return immediately
}
auto __res = _M_reactor.wait(__fds, __ms);
if (__res == __reactor::_S_retry)
continue;
if (__res == __reactor::_S_timeout)
if (__timerq == nullptr)
return false;
else
continue; // timed out, so restart loop and process the timer
__timerq = nullptr;
if (__fds.empty()) // nothing to do
return false;
lock_guard<mutex> __lock(_M_mtx);
for (auto __it = _M_ops.begin(), __end = _M_ops.end(),
__prev = _M_ops.before_begin(); __it != __end; ++__it, ++__prev)
{
auto& __op = **__it;
auto __pos = std::lower_bound(__fds.begin(), __fds.end(),
__op._M_fd,
[](const auto& __p, int __fd) { return __p.fd < __fd; });
if (__pos != __fds.end() && __pos->fd == __op._M_fd
&& __pos->revents & __op._M_ev)
{
__it->swap(__async_op);
_M_ops.erase_after(__prev);
break; // restart loop and run op
}
}
}
}
struct __reactor
{
__reactor() : _M_fds(1)
{
int __pipe[2];
if (::pipe(__pipe) == -1)
__throw_system_error(errno);
if (::fcntl(__pipe[0], F_SETFL, O_NONBLOCK) == -1
|| ::fcntl(__pipe[1], F_SETFL, O_NONBLOCK) == -1)
{
int __e = errno;
::close(__pipe[0]);
::close(__pipe[1]);
__throw_system_error(__e);
}
_M_fds.back().events = POLLIN;
_M_fds.back().fd = __pipe[0];
_M_notify_wr = __pipe[1];
}
~__reactor()
{
::close(_M_fds.back().fd);
::close(_M_notify_wr);
}
// write a notification byte to the pipe (ignoring errors)
void _M_notify()
{
int __n;
do {
__n = ::write(_M_notify_wr, "", 1);
} while (__n == -1 && errno == EINTR);
}
// read all notification bytes from the pipe
void _M_on_notify()
{
// Drain the pipe.
char __buf[64];
ssize_t __n;
do {
__n = ::read(_M_fds.back().fd, __buf, sizeof(__buf));
} while (__n != -1 || errno == EINTR);
}
void
_M_add_fd(int __fd)
{
auto __pos = _M_lower_bound(__fd);
if (__pos->fd == __fd)
__throw_system_error((int)errc::invalid_argument);
_M_fds.insert(__pos, __fdvec::value_type{})->fd = __fd;
_M_notify();
}
void
_M_remove_fd(int __fd)
{
auto __pos = _M_lower_bound(__fd);
if (__pos->fd == __fd)
_M_fds.erase(__pos);
// else bug!
_M_notify();
}
void
_M_fd_interest(int __fd, int __w)
{
auto __pos = _M_lower_bound(__fd);
if (__pos->fd == __fd)
__pos->events |= __w;
// else bug!
_M_notify();
}
void
_M_not_interested(int __fd)
{
auto __pos = _M_lower_bound(__fd);
if (__pos->fd == __fd)
__pos->events = 0;
_M_notify();
}
# ifdef _GLIBCXX_HAVE_POLL_H
using __fdvec = vector<::pollfd>;
// Find first element p such that !(p.fd < __fd)
// N.B. always returns a dereferencable iterator.
__fdvec::iterator
_M_lower_bound(int __fd)
{
return std::lower_bound(_M_fds.begin(), _M_fds.end() - 1,
__fd, [](const auto& __p, int __fd) { return __p.fd < __fd; });
}
enum __status { _S_retry, _S_timeout, _S_ok, _S_error };
__status
wait(__fdvec& __fds, chrono::milliseconds __timeout)
{
// XXX not thread-safe!
__fds = _M_fds; // take snapshot to pass to poll()
int __res = ::poll(__fds.data(), __fds.size(), __timeout.count());
if (__res == -1)
{
__fds.clear();
if (errno == EINTR)
return _S_retry;
return _S_error; // XXX ???
}
else if (__res == 0)
{
__fds.clear();
return _S_timeout;
}
else if (__fds.back().revents != 0) // something changed, restart
{
__fds.clear();
_M_on_notify();
return _S_retry;
}
auto __part = std::stable_partition(__fds.begin(), __fds.end() - 1,
[](const __fdvec::value_type& __p) { return __p.revents != 0; });
__fds.erase(__part, __fds.end());
return _S_ok;
}
__fdvec _M_fds; // _M_fds.back() is the read end of the self-pipe
#endif
int _M_notify_wr; // write end of the self-pipe
};
__reactor _M_reactor;
vector<__timer_queue_base*> _M_timers;
forward_list<unique_ptr<__async_operation>> _M_ops;
vector<thread::id> _M_call_stack;
};
inline bool
operator==(const io_context::executor_type& __a,
const io_context::executor_type& __b) noexcept
{
// https://github.com/chriskohlhoff/asio-tr2/issues/201
using executor_type = io_context::executor_type;
return std::addressof(executor_type(__a).context())
== std::addressof(executor_type(__b).context());
}
inline bool
operator!=(const io_context::executor_type& __a,
const io_context::executor_type& __b) noexcept
{ return !(__a == __b); }
template<> struct is_executor<io_context::executor_type> : true_type {};
/// @}
} // namespace v1
} // namespace net
} // namespace experimental
_GLIBCXX_END_NAMESPACE_VERSION
} // namespace std
#endif // C++14
#endif // _GLIBCXX_EXPERIMENTAL_IO_SERVICE