include/boost/corosio/native/detail/epoll/epoll_scheduler.hpp

85.5% Lines (130/152) 100.0% List of functions (11/11)

[ ] prev/next file n p prev/next uncovered line

epoll_scheduler.hpp

f(x) Functions (11)

Function Calls Lines Blocks

Line	Hits	Source Code
1		//
2		// Copyright (c) 2026 Steve Gerbino
3		// Copyright (c) 2026 Michael Vandeberg
4		//
5		// Distributed under the Boost Software License, Version 1.0. (See accompanying
6		// file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
7		//
8		// Official repository: https://github.com/cppalliance/corosio
9		//
10
11		#ifndef BOOST_COROSIO_NATIVE_DETAIL_EPOLL_EPOLL_SCHEDULER_HPP
12		#define BOOST_COROSIO_NATIVE_DETAIL_EPOLL_EPOLL_SCHEDULER_HPP
13
14		#include <boost/corosio/detail/platform.hpp>
15
16		#if BOOST_COROSIO_HAS_EPOLL
17
18		#include <boost/corosio/detail/config.hpp>
19		#include <boost/capy/ex/execution_context.hpp>
20
21		#include <boost/corosio/native/detail/reactor/reactor_scheduler.hpp>
22		#include <boost/corosio/native/detail/reactor/reactor_signal_pipe.hpp>
23
24		#include <boost/corosio/native/detail/epoll/epoll_traits.hpp>
25		#include <boost/corosio/detail/timer_service.hpp>
26		#include <boost/corosio/native/detail/make_err.hpp>
27		#include <boost/corosio/native/detail/posix/posix_resolver_service.hpp>
28		#include <boost/corosio/native/detail/posix/posix_signal_service.hpp>
29		#include <boost/corosio/native/detail/posix/posix_stream_file_service.hpp>
30		#include <boost/corosio/native/detail/posix/posix_random_access_file_service.hpp>
31
32		#include <boost/corosio/detail/except.hpp>
33
34		#include <atomic>
35		#include <chrono>
36		#include <cstdint>
37		#include <mutex>
38		#include <vector>
39
40		#include <errno.h>
41		#include <sys/epoll.h>
42		#include <sys/eventfd.h>
43		#include <sys/timerfd.h>
44		#include <unistd.h>
45
46		namespace boost::corosio::detail {
47
48		/** Linux scheduler using epoll for I/O multiplexing.
49
50		This scheduler implements the scheduler interface using Linux epoll
51		for efficient I/O event notification. It uses a single reactor model
52		where one thread runs epoll_wait while other threads
53		wait on a condition variable for handler work. This design provides:
54
55		- Handler parallelism: N posted handlers can execute on N threads
56		- No thundering herd: condition_variable wakes exactly one thread
57		- IOCP parity: Behavior matches Windows I/O completion port semantics
58
59		When threads call run(), they first try to execute queued handlers.
60		If the queue is empty and no reactor is running, one thread becomes
61		the reactor and runs epoll_wait. Other threads wait on a condition
62		variable until handlers are available.
63
64		@par Thread Safety
65		All public member functions are thread-safe.
66		*/
67		class BOOST_COROSIO_DECL epoll_scheduler final : public reactor_scheduler
68		{
69		public:
70		/** Construct the scheduler.
71
72		Creates an epoll instance, eventfd for reactor interruption,
73		and timerfd for kernel-managed timer expiry.
74
75		@param ctx Reference to the owning execution_context.
76		@param concurrency_hint Hint for expected thread count (unused).
77		*/
78		epoll_scheduler(capy::execution_context& ctx, int concurrency_hint = -1);
79
80		/// Destroy the scheduler.
81		~epoll_scheduler() override;
82
83		epoll_scheduler(epoll_scheduler const&) = delete;
84		epoll_scheduler& operator=(epoll_scheduler const&) = delete;
85
86		/// Shut down the scheduler, draining pending operations.
87		void shutdown() override;
88
89		/// Apply runtime configuration, resizing the event buffer.
90		void configure_reactor(
91		unsigned max_events,
92		unsigned budget_init,
93		unsigned budget_max,
94		unsigned unassisted) override;
95
96		/** Return the epoll file descriptor.
97
98		Used by socket services to register file descriptors
99		for I/O event notification.
100
101		@return The epoll file descriptor.
102		*/
103		int epoll_fd() const noexcept
104		{
105		return epoll_fd_;
106		}
107
108		/** Register a descriptor for persistent monitoring.
109
110		The fd is registered once and stays registered until explicitly
111		deregistered. Events are dispatched via reactor_descriptor_state which
112		tracks pending read/write/connect operations.
113
114		@param fd The file descriptor to register.
115		@param desc Pointer to descriptor data (stored in epoll_event.data.ptr).
116		*/
117		void register_descriptor(int fd, reactor_descriptor_state* desc) const;
118
119		/** Deregister a persistently registered descriptor.
120
121		@param fd The file descriptor to deregister.
122		*/
123		void deregister_descriptor(int fd) const;
124
125		/// Watch the read end of the POSIX signal self-pipe (see scheduler.hpp).
126	41x	void register_signal_reader(int read_fd) override
127		{
128	41x	register_descriptor(read_fd, signal_pipe_reader_.arm());
129	41x	}
130
131		private:
132		void
133		run_task(lock_type& lock, context_type* ctx,
134		long timeout_us) override;
135		void interrupt_reactor() const override;
136		void update_timerfd() const;
137
138		int epoll_fd_;
139		int event_fd_;
140		int timer_fd_;
141
142		// Watches the global signal self-pipe's read end (armed lazily by
143		// register_signal_reader on the first signal registration).
144		reactor_signal_pipe_reader signal_pipe_reader_;
145
146		// Edge-triggered eventfd state
147		mutable std::atomic<bool> eventfd_armed_{false};
148
149		// Set when the earliest timer changes; flushed before epoll_wait
150		mutable std::atomic<bool> timerfd_stale_{false};
151
152		// Event buffer sized from max_events_per_poll_ (set at construction,
153		// resized by configure_reactor via io_context_options).
154		std::vector<epoll_event> event_buffer_;
155		};
156
157	642x	inline epoll_scheduler::epoll_scheduler(capy::execution_context& ctx, int)
158	642x	: epoll_fd_(-1)
159	642x	, event_fd_(-1)
160	642x	, timer_fd_(-1)
161	1284x	, event_buffer_(max_events_per_poll_)
162		{
163	642x	epoll_fd_ = ::epoll_create1(EPOLL_CLOEXEC);
164	642x	if (epoll_fd_ < 0)
165	✗	detail::throw_system_error(make_err(errno), "epoll_create1");
166
167	642x	event_fd_ = ::eventfd(0, EFD_NONBLOCK \| EFD_CLOEXEC);
168	642x	if (event_fd_ < 0)
169		{
170	✗	int errn = errno;
171	✗	::close(epoll_fd_);
172	✗	detail::throw_system_error(make_err(errn), "eventfd");
173		}
174
175	642x	timer_fd_ = ::timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK \| TFD_CLOEXEC);
176	642x	if (timer_fd_ < 0)
177		{
178	✗	int errn = errno;
179	✗	::close(event_fd_);
180	✗	::close(epoll_fd_);
181	✗	detail::throw_system_error(make_err(errn), "timerfd_create");
182		}
183
184	642x	epoll_event ev{};
185	642x	ev.events = EPOLLIN \| EPOLLET;
186	642x	ev.data.ptr = nullptr;
187	642x	if (::epoll_ctl(epoll_fd_, EPOLL_CTL_ADD, event_fd_, &ev) < 0)
188		{
189	✗	int errn = errno;
190	✗	::close(timer_fd_);
191	✗	::close(event_fd_);
192	✗	::close(epoll_fd_);
193	✗	detail::throw_system_error(make_err(errn), "epoll_ctl");
194		}
195
196	642x	epoll_event timer_ev{};
197	642x	timer_ev.events = EPOLLIN \| EPOLLERR;
198	642x	timer_ev.data.ptr = &timer_fd_;
199	642x	if (::epoll_ctl(epoll_fd_, EPOLL_CTL_ADD, timer_fd_, &timer_ev) < 0)
200		{
201	✗	int errn = errno;
202	✗	::close(timer_fd_);
203	✗	::close(event_fd_);
204	✗	::close(epoll_fd_);
205	✗	detail::throw_system_error(make_err(errn), "epoll_ctl (timerfd)");
206		}
207
208	642x	timer_svc_ = &get_timer_service(ctx, *this);
209	642x	timer_svc_->set_on_earliest_changed(
210	3488x	timer_service::callback(this, [](void* p) {
211	2846x	auto* self = static_cast<epoll_scheduler*>(p);
212	2846x	self->timerfd_stale_.store(true, std::memory_order_release);
213	2846x	self->interrupt_reactor();
214	2846x	}));
215
216	642x	get_resolver_service(ctx, *this);
217	642x	get_signal_service(ctx, *this);
218	642x	get_stream_file_service(ctx, *this);
219	642x	get_random_access_file_service(ctx, *this);
220
221	642x	completed_ops_.push(&task_op_);
222	642x	}
223
224	1284x	inline epoll_scheduler::~epoll_scheduler()
225		{
226	642x	if (timer_fd_ >= 0)
227	642x	::close(timer_fd_);
228	642x	if (event_fd_ >= 0)
229	642x	::close(event_fd_);
230	642x	if (epoll_fd_ >= 0)
231	642x	::close(epoll_fd_);
232	1284x	}
233
234		inline void
235	642x	epoll_scheduler::shutdown()
236		{
237	642x	shutdown_drain();
238
239	642x	if (event_fd_ >= 0)
240	642x	interrupt_reactor();
241	642x	}
242
243		inline void
244	16x	epoll_scheduler::configure_reactor(
245		unsigned max_events,
246		unsigned budget_init,
247		unsigned budget_max,
248		unsigned unassisted)
249		{
250	16x	reactor_scheduler::configure_reactor(
251		max_events, budget_init, budget_max, unassisted);
252	15x	event_buffer_.resize(max_events_per_poll_);
253	15x	}
254
255		inline void
256	4367x	epoll_scheduler::register_descriptor(int fd, reactor_descriptor_state* desc) const
257		{
258	4367x	epoll_event ev{};
259	4367x	ev.events = EPOLLIN \| EPOLLOUT \| EPOLLET \| EPOLLERR \| EPOLLHUP;
260	4367x	ev.data.ptr = desc;
261
262	4367x	if (::epoll_ctl(epoll_fd_, EPOLL_CTL_ADD, fd, &ev) < 0)
263	✗	detail::throw_system_error(make_err(errno), "epoll_ctl (register)");
264
265	4367x	desc->registered_events = ev.events;
266	4367x	desc->fd = fd;
267	4367x	desc->scheduler_ = this;
268	4367x	desc->mutex.set_enabled(reactor_io_locking_);
269	4367x	desc->ready_events_.store(0, std::memory_order_relaxed);
270
271	4367x	conditionally_enabled_mutex::scoped_lock lock(desc->mutex);
272	4367x	desc->impl_ref_.reset();
273	4367x	desc->read_ready = false;
274	4367x	desc->write_ready = false;
275	4367x	}
276
277		inline void
278	4326x	epoll_scheduler::deregister_descriptor(int fd) const
279		{
280	4326x	::epoll_ctl(epoll_fd_, EPOLL_CTL_DEL, fd, nullptr);
281	4326x	}
282
283		inline void
284	4011x	epoll_scheduler::interrupt_reactor() const
285		{
286	4011x	bool expected = false;
287	4011x	if (eventfd_armed_.compare_exchange_strong(
288		expected, true, std::memory_order_release,
289		std::memory_order_relaxed))
290		{
291	2979x	std::uint64_t val = 1;
292	2979x	[[maybe_unused]] auto r = ::write(event_fd_, &val, sizeof(val));
293		}
294	4011x	}
295
296		inline void
297	4351x	epoll_scheduler::update_timerfd() const
298		{
299	4351x	auto nearest = timer_svc_->nearest_expiry();
300
301	4351x	itimerspec ts{};
302	4351x	int flags = 0;
303
304	4351x	if (nearest == timer_service::time_point::max())
305		{
306		// No timers — disarm by setting to 0 (relative)
307		}
308		else
309		{
310	4213x	auto now = std::chrono::steady_clock::now();
311	4213x	if (nearest <= now)
312		{
313		// Use 1ns instead of 0 — zero disarms the timerfd
314	524x	ts.it_value.tv_nsec = 1;
315		}
316		else
317		{
318	3689x	auto nsec = std::chrono::duration_cast<std::chrono::nanoseconds>(
319	3689x	nearest - now)
320	3689x	.count();
321	3689x	ts.it_value.tv_sec = nsec / 1000000000;
322	3689x	ts.it_value.tv_nsec = nsec % 1000000000;
323	3689x	if (ts.it_value.tv_sec == 0 && ts.it_value.tv_nsec == 0)
324	✗	ts.it_value.tv_nsec = 1;
325		}
326		}
327
328	4351x	if (::timerfd_settime(timer_fd_, flags, &ts, nullptr) < 0)
329	✗	detail::throw_system_error(make_err(errno), "timerfd_settime");
330	4351x	}
331
332		inline void
333	31256x	epoll_scheduler::run_task(
334		lock_type& lock, context_type* ctx, long timeout_us)
335		{
336		int timeout_ms;
337	31256x	if (task_interrupted_)
338	25285x	timeout_ms = 0;
339	5971x	else if (timeout_us < 0)
340	5967x	timeout_ms = -1;
341		else
342	4x	timeout_ms = static_cast<int>((timeout_us + 999) / 1000);
343
344	31256x	if (lock.owns_lock())
345	5973x	lock.unlock();
346
347	31256x	task_cleanup on_exit{this, &lock, ctx};
348
349		// Flush deferred timerfd programming before blocking
350	31256x	if (timerfd_stale_.exchange(false, std::memory_order_acquire))
351	2180x	update_timerfd();
352
353	31256x	int nfds = ::epoll_wait(
354		epoll_fd_, event_buffer_.data(),
355	31256x	static_cast<int>(event_buffer_.size()), timeout_ms);
356
357	31256x	if (nfds < 0 && errno != EINTR)
358	✗	detail::throw_system_error(make_err(errno), "epoll_wait");
359
360	31256x	bool check_timers = false;
361	31256x	ready_queue local_ops;
362
363	72692x	for (int i = 0; i < nfds; ++i)
364		{
365	41436x	if (event_buffer_[i].data.ptr == nullptr)
366		{
367		std::uint64_t val;
368		// NOLINTNEXTLINE(clang-analyzer-unix.BlockInCriticalSection)
369	2337x	[[maybe_unused]] auto r = ::read(event_fd_, &val, sizeof(val));
370	2337x	eventfd_armed_.store(false, std::memory_order_relaxed);
371	2337x	continue;
372	2337x	}
373
374	39099x	if (event_buffer_[i].data.ptr == &timer_fd_)
375		{
376		std::uint64_t expirations;
377		// NOLINTNEXTLINE(clang-analyzer-unix.BlockInCriticalSection)
378		[[maybe_unused]] auto r =
379	2171x	::read(timer_fd_, &expirations, sizeof(expirations));
380	2171x	check_timers = true;
381	2171x	continue;
382	2171x	}
383
384		auto* desc =
385	36928x	static_cast<reactor_descriptor_state*>(event_buffer_[i].data.ptr);
386	36928x	desc->add_ready_events(event_buffer_[i].events);
387
388	36928x	bool expected = false;
389	36928x	if (desc->is_enqueued_.compare_exchange_strong(
390		expected, true, std::memory_order_release,
391		std::memory_order_relaxed))
392		{
393	36928x	local_ops.push(desc);
394		}
395		}
396
397	31256x	if (check_timers)
398		{
399	2171x	timer_svc_->process_expired();
400	2171x	update_timerfd();
401		}
402
403	31256x	lock.lock();
404
405	31256x	completed_ops_.splice(local_ops);
406	31256x	}
407
408		} // namespace boost::corosio::detail
409
410		#endif // BOOST_COROSIO_HAS_EPOLL
411
412		#endif // BOOST_COROSIO_NATIVE_DETAIL_EPOLL_EPOLL_SCHEDULER_HPP
413