Index/include / boost / corosio / native / detail / select / select_op.hpp

include/boost/corosio/native/detail/select/select_op.hpp

74.4% Lines (96/129) 84.2% Functions (16/19)
include/boost/corosio/native/detail/select/select_op.hpp
Prev Index Next
Line TLA Hits Source Code
1 //
2 // Copyright (c) 2026 Steve Gerbino
3 //
4 // Distributed under the Boost Software License, Version 1.0. (See accompanying
5 // file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
6 //
7 // Official repository: https://github.com/cppalliance/corosio
8 //
9
10 #ifndef BOOST_COROSIO_NATIVE_DETAIL_SELECT_SELECT_OP_HPP
11 #define BOOST_COROSIO_NATIVE_DETAIL_SELECT_SELECT_OP_HPP
12
13 #include <boost/corosio/detail/platform.hpp>
14
15 #if BOOST_COROSIO_HAS_SELECT
16
17 #include <boost/corosio/detail/config.hpp>
18 #include <boost/corosio/io/io_object.hpp>
19 #include <boost/corosio/endpoint.hpp>
20 #include <boost/capy/ex/executor_ref.hpp>
21 #include <coroutine>
22 #include <boost/capy/error.hpp>
23 #include <system_error>
24
25 #include <boost/corosio/detail/make_err.hpp>
26 #include <boost/corosio/detail/dispatch_coro.hpp>
27 #include <boost/corosio/detail/scheduler_op.hpp>
28 #include <boost/corosio/detail/endpoint_convert.hpp>
29
30 #include <unistd.h>
31 #include <errno.h>
32 #include <fcntl.h>
33
34 #include <atomic>
35 #include <cstddef>
36 #include <memory>
37 #include <optional>
38 #include <stop_token>
39
40 #include <netinet/in.h>
41 #include <sys/select.h>
42 #include <sys/socket.h>
43 #include <sys/uio.h>
44
45 /*
46 select Operation State
47 ======================
48
49 Each async I/O operation has a corresponding select_op-derived struct that
50 holds the operation's state while it's in flight. The socket impl owns
51 fixed slots for each operation type (conn_, rd_, wr_), so only one
52 operation of each type can be pending per socket at a time.
53
54 This mirrors the epoll_op design for consistency across backends.
55
56 Completion vs Cancellation Race
57 -------------------------------
58 The `registered` atomic uses a tri-state (unregistered, registering,
59 registered) to handle two races: (1) between register_fd() and the
60 reactor seeing an event, and (2) between reactor completion and cancel().
61
62 The registering state closes the window where an event could arrive
63 after register_fd() but before the boolean was set. The reactor and
64 cancel() both treat registering the same as registered when claiming.
65
66 Whoever atomically exchanges to unregistered "claims" the operation
67 and is responsible for completing it. The loser sees unregistered and
68 does nothing. The initiating thread uses compare_exchange to transition
69 from registering to registered; if this fails, the reactor or cancel
70 already claimed the op.
71
72 Impl Lifetime Management
73 ------------------------
74 When cancel() posts an op to the scheduler's ready queue, the socket impl
75 might be destroyed before the scheduler processes the op. The `impl_ptr`
76 member holds a shared_ptr to the impl, keeping it alive until the op
77 completes.
78
79 EOF Detection
80 -------------
81 For reads, 0 bytes with no error means EOF. But an empty user buffer also
82 returns 0 bytes. The `empty_buffer_read` flag distinguishes these cases.
83
84 SIGPIPE Prevention
85 ------------------
86 Writes use sendmsg() with MSG_NOSIGNAL instead of writev() to prevent
87 SIGPIPE when the peer has closed.
88 */
89
90 namespace boost::corosio::detail {
91
92 // Forward declarations for cancellation support
93 class select_socket;
94 class select_acceptor;
95
96 /** Registration state for async operations.
97
98 Tri-state enum to handle the race between register_fd() and
99 run_reactor() seeing an event. Setting REGISTERING before
100 calling register_fd() ensures events delivered during the
101 registration window are not dropped.
102 */
103 enum class select_registration_state : std::uint8_t
104 {
105 unregistered, ///< Not registered with reactor
106 registering, ///< register_fd() called, not yet confirmed
107 registered ///< Fully registered, ready for events
108 };
109
110 struct select_op : scheduler_op
111 {
112 struct canceller
113 {
114 select_op* op;
115 void operator()() const noexcept;
116 };
117
118 std::coroutine_handle<> h;
119 capy::executor_ref ex;
120 std::error_code* ec_out = nullptr;
121 std::size_t* bytes_out = nullptr;
122
123 int fd = -1;
124 int errn = 0;
125 std::size_t bytes_transferred = 0;
126
127 std::atomic<bool> cancelled{false};
128 std::atomic<select_registration_state> registered{
129 select_registration_state::unregistered};
130 std::optional<std::stop_callback<canceller>> stop_cb;
131
132 // Prevents use-after-free when socket is closed with pending ops.
133 std::shared_ptr<void> impl_ptr;
134
135 // For stop_token cancellation - pointer to owning socket/acceptor impl.
136 select_socket* socket_impl_ = nullptr;
137 select_acceptor* acceptor_impl_ = nullptr;
138
139 26155 select_op() = default;
140
141 177993 void reset() noexcept
142 {
143 177993 fd = -1;
144 177993 errn = 0;
145 177993 bytes_transferred = 0;
146 177993 cancelled.store(false, std::memory_order_relaxed);
147 177993 registered.store(
148 select_registration_state::unregistered, std::memory_order_relaxed);
149 177993 impl_ptr.reset();
150 177993 socket_impl_ = nullptr;
151 177993 acceptor_impl_ = nullptr;
152 177993 }
153
154 172200 void operator()() override
155 {
156 172200 stop_cb.reset();
157
158 172200 if (ec_out)
159 {
160 172200 if (cancelled.load(std::memory_order_acquire))
161 194 *ec_out = capy::error::canceled;
162 172006 else if (errn != 0)
163 1 *ec_out = make_err(errn);
164 172005 else if (is_read_operation() && bytes_transferred == 0)
165 5 *ec_out = capy::error::eof;
166 else
167 172000 *ec_out = {};
168 }
169
170 172200 if (bytes_out)
171 172200 *bytes_out = bytes_transferred;
172
173 // Move to stack before destroying the frame
174 172200 capy::executor_ref saved_ex(ex);
175 172200 std::coroutine_handle<> saved_h(h);
176 172200 impl_ptr.reset();
177 172200 dispatch_coro(saved_ex, saved_h).resume();
178 172200 }
179
180 86016 virtual bool is_read_operation() const noexcept
181 {
182 86016 return false;
183 }
184 virtual void cancel() noexcept = 0;
185
186 ✗ void destroy() override
187 {
188 ✗ stop_cb.reset();
189 ✗ impl_ptr.reset();
190 ✗ }
191
192 79127 void request_cancel() noexcept
193 {
194 79127 cancelled.store(true, std::memory_order_release);
195 79127 }
196
197 // NOLINTNEXTLINE(performance-unnecessary-value-param)
198 void start(std::stop_token token)
199 {
200 cancelled.store(false, std::memory_order_release);
201 stop_cb.reset();
202 socket_impl_ = nullptr;
203 acceptor_impl_ = nullptr;
204
205 if (token.stop_possible())
206 stop_cb.emplace(token, canceller{this});
207 }
208
209 // NOLINTNEXTLINE(performance-unnecessary-value-param)
210 175096 void start(std::stop_token token, select_socket* impl)
211 {
212 175096 cancelled.store(false, std::memory_order_release);
213 175096 stop_cb.reset();
214 175096 socket_impl_ = impl;
215 175096 acceptor_impl_ = nullptr;
216
217 175096 if (token.stop_possible())
218 98 stop_cb.emplace(token, canceller{this});
219 175096 }
220
221 // NOLINTNEXTLINE(performance-unnecessary-value-param)
222 2897 void start(std::stop_token token, select_acceptor* impl)
223 {
224 2897 cancelled.store(false, std::memory_order_release);
225 2897 stop_cb.reset();
226 2897 socket_impl_ = nullptr;
227 2897 acceptor_impl_ = impl;
228
229 2897 if (token.stop_possible())
230 ✗ stop_cb.emplace(token, canceller{this});
231 2897 }
232
233 177836 void complete(int err, std::size_t bytes) noexcept
234 {
235 177836 errn = err;
236 177836 bytes_transferred = bytes;
237 177836 }
238
239 ✗ virtual void perform_io() noexcept {}
240 };
241
242 struct select_connect_op final : select_op
243 {
244 endpoint target_endpoint;
245
246 2896 void reset() noexcept
247 {
248 2896 select_op::reset();
249 2896 target_endpoint = endpoint{};
250 2896 }
251
252 2896 void perform_io() noexcept override
253 {
254 // connect() completion status is retrieved via SO_ERROR, not return value
255 2896 int err = 0;
256 2896 socklen_t len = sizeof(err);
257 2896 if (::getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &len) < 0)
258 ✗ err = errno;
259 2896 complete(err, 0);
260 2896 }
261
262 // Defined in sockets.cpp where select_socket is complete
263 void operator()() override;
264 void cancel() noexcept override;
265 };
266
267 struct select_read_op final : select_op
268 {
269 static constexpr std::size_t max_buffers = 16;
270 iovec iovecs[max_buffers];
271 int iovec_count = 0;
272 bool empty_buffer_read = false;
273
274 85989 bool is_read_operation() const noexcept override
275 {
276 85989 return !empty_buffer_read;
277 }
278
279 86179 void reset() noexcept
280 {
281 86179 select_op::reset();
282 86179 iovec_count = 0;
283 86179 empty_buffer_read = false;
284 86179 }
285
286 119 void perform_io() noexcept override
287 {
288 119 ssize_t n = ::readv(fd, iovecs, iovec_count);
289 119 if (n >= 0)
290 119 complete(0, static_cast<std::size_t>(n));
291 else
292 ✗ complete(errno, 0);
293 119 }
294
295 void cancel() noexcept override;
296 };
297
298 struct select_write_op final : select_op
299 {
300 static constexpr std::size_t max_buffers = 16;
301 iovec iovecs[max_buffers];
302 int iovec_count = 0;
303
304 86021 void reset() noexcept
305 {
306 86021 select_op::reset();
307 86021 iovec_count = 0;
308 86021 }
309
310 ✗ void perform_io() noexcept override
311 {
312 ✗ msghdr msg{};
313 ✗ msg.msg_iov = iovecs;
314 ✗ msg.msg_iovlen = static_cast<std::size_t>(iovec_count);
315
316 ✗ ssize_t n = ::sendmsg(fd, &msg, MSG_NOSIGNAL);
317 ✗ if (n >= 0)
318 ✗ complete(0, static_cast<std::size_t>(n));
319 else
320 ✗ complete(errno, 0);
321 ✗ }
322
323 void cancel() noexcept override;
324 };
325
326 struct select_accept_op final : select_op
327 {
328 int accepted_fd = -1;
329 io_object::implementation* peer_impl = nullptr;
330 io_object::implementation** impl_out = nullptr;
331
332 2897 void reset() noexcept
333 {
334 2897 select_op::reset();
335 2897 accepted_fd = -1;
336 2897 peer_impl = nullptr;
337 2897 impl_out = nullptr;
338 2897 }
339
340 2892 void perform_io() noexcept override
341 {
342 2892 sockaddr_in addr{};
343 2892 socklen_t addrlen = sizeof(addr);
344
345 // Note: select backend uses accept() + fcntl instead of accept4()
346 // for broader POSIX compatibility
347 2892 int new_fd = ::accept(fd, reinterpret_cast<sockaddr*>(&addr), &addrlen);
348
349 2892 if (new_fd >= 0)
350 {
351 // Reject fds that exceed select()'s FD_SETSIZE limit.
352 // Better to fail now than during later async operations.
353 2892 if (new_fd >= FD_SETSIZE)
354 {
355 ✗ ::close(new_fd);
356 ✗ complete(EINVAL, 0);
357 ✗ return;
358 }
359
360 // Set non-blocking and close-on-exec flags.
361 // A non-blocking socket is essential for the async reactor;
362 // if we can't configure it, fail rather than risk blocking.
363 2892 int flags = ::fcntl(new_fd, F_GETFL, 0);
364 2892 if (flags == -1)
365 {
366 ✗ int err = errno;
367 ✗ ::close(new_fd);
368 ✗ complete(err, 0);
369 ✗ return;
370 }
371
372 2892 if (::fcntl(new_fd, F_SETFL, flags | O_NONBLOCK) == -1)
373 {
374 ✗ int err = errno;
375 ✗ ::close(new_fd);
376 ✗ complete(err, 0);
377 ✗ return;
378 }
379
380 2892 if (::fcntl(new_fd, F_SETFD, FD_CLOEXEC) == -1)
381 {
382 ✗ int err = errno;
383 ✗ ::close(new_fd);
384 ✗ complete(err, 0);
385 ✗ return;
386 }
387
388 2892 accepted_fd = new_fd;
389 2892 complete(0, 0);
390 }
391 else
392 {
393 ✗ complete(errno, 0);
394 }
395 }
396
397 // Defined in acceptors.cpp where select_acceptor is complete
398 void operator()() override;
399 void cancel() noexcept override;
400 };
401
402 } // namespace boost::corosio::detail
403
404 #endif // BOOST_COROSIO_HAS_SELECT
405
406 #endif // BOOST_COROSIO_NATIVE_DETAIL_SELECT_SELECT_OP_HPP
407