1 | /* |
1 | /* |
2 | * libev epoll fd activity backend |
2 | * libev epoll fd activity backend |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libev@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010,2011,2016,2017,2019 Marc Alexander Lehmann <libev@schmorp.de> |
5 | * All rights reserved. |
5 | * All rights reserved. |
6 | * |
6 | * |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * tion, are permitted provided that the following conditions are met: |
8 | * tion, are permitted provided that the following conditions are met: |
9 | * |
9 | * |
10 | * 1. Redistributions of source code must retain the above copyright notice, |
10 | * 1. Redistributions of source code must retain the above copyright notice, |
11 | * this list of conditions and the following disclaimer. |
11 | * this list of conditions and the following disclaimer. |
12 | * |
12 | * |
13 | * 2. Redistributions in binary form must reproduce the above copyright |
13 | * 2. Redistributions in binary form must reproduce the above copyright |
14 | * notice, this list of conditions and the following disclaimer in the |
14 | * notice, this list of conditions and the following disclaimer in the |
15 | * documentation and/or other materials provided with the distribution. |
15 | * documentation and/or other materials provided with the distribution. |
16 | * |
16 | * |
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED |
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED |
18 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
18 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
19 | * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
19 | * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
20 | * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- |
20 | * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- |
21 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
21 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
… | |
… | |
122 | anfds [fd].emask = EV_EMASK_EPERM; |
122 | anfds [fd].emask = EV_EMASK_EPERM; |
123 | |
123 | |
124 | /* add fd to epoll_eperms, if not already inside */ |
124 | /* add fd to epoll_eperms, if not already inside */ |
125 | if (!(oldmask & EV_EMASK_EPERM)) |
125 | if (!(oldmask & EV_EMASK_EPERM)) |
126 | { |
126 | { |
127 | array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, EMPTY2); |
127 | array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, array_needsize_noinit); |
128 | epoll_eperms [epoll_epermcnt++] = fd; |
128 | epoll_eperms [epoll_epermcnt++] = fd; |
129 | } |
129 | } |
130 | |
130 | |
131 | return; |
131 | return; |
132 | } |
132 | } |
|
|
133 | else |
|
|
134 | assert (("libev: I/O watcher with invalid fd found in epoll_ctl", errno != EBADF && errno != ELOOP && errno != EINVAL)); |
133 | |
135 | |
134 | fd_kill (EV_A_ fd); |
136 | fd_kill (EV_A_ fd); |
135 | |
137 | |
136 | dec_egen: |
138 | dec_egen: |
137 | /* we didn't successfully call epoll_ctl, so decrement the generation counter again */ |
139 | /* we didn't successfully call epoll_ctl, so decrement the generation counter again */ |
… | |
… | |
141 | static void |
143 | static void |
142 | epoll_poll (EV_P_ ev_tstamp timeout) |
144 | epoll_poll (EV_P_ ev_tstamp timeout) |
143 | { |
145 | { |
144 | int i; |
146 | int i; |
145 | int eventcnt; |
147 | int eventcnt; |
146 | |
148 | |
|
|
149 | if (expect_false (epoll_epermcnt)) |
|
|
150 | timeout = 0.; |
|
|
151 | |
147 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
152 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
148 | /* the default libev max wait time, however. */ |
153 | /* the default libev max wait time, however. */ |
149 | EV_RELEASE_CB; |
154 | EV_RELEASE_CB; |
150 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.)); |
155 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3); |
151 | EV_ACQUIRE_CB; |
156 | EV_ACQUIRE_CB; |
152 | |
157 | |
153 | if (expect_false (eventcnt < 0)) |
158 | if (expect_false (eventcnt < 0)) |
154 | { |
159 | { |
155 | if (errno != EINTR) |
160 | if (errno != EINTR) |
… | |
… | |
165 | int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */ |
170 | int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */ |
166 | int want = anfds [fd].events; |
171 | int want = anfds [fd].events; |
167 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
172 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
168 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
173 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
169 | |
174 | |
|
|
175 | /* |
170 | /* check for spurious notification */ |
176 | * check for spurious notification. |
|
|
177 | * this only finds spurious notifications on egen updates |
|
|
178 | * other spurious notifications will be found by epoll_ctl, below |
171 | /* we assume that fd is always in range, as we never shrink the anfds array */ |
179 | * we assume that fd is always in range, as we never shrink the anfds array |
|
|
180 | */ |
172 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
181 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
173 | { |
182 | { |
174 | /* recreate kernel state */ |
183 | /* recreate kernel state */ |
175 | postfork = 1; |
184 | postfork |= 2; |
176 | continue; |
185 | continue; |
177 | } |
186 | } |
178 | |
187 | |
179 | if (expect_false (got & ~want)) |
188 | if (expect_false (got & ~want)) |
180 | { |
189 | { |
181 | anfds [fd].emask = want; |
190 | anfds [fd].emask = want; |
182 | |
191 | |
|
|
192 | /* |
183 | /* we received an event but are not interested in it, try mod or del */ |
193 | * we received an event but are not interested in it, try mod or del |
184 | /* I don't think we ever need MOD, but let's handle it anyways */ |
194 | * this often happens because we optimistically do not unregister fds |
|
|
195 | * when we are no longer interested in them, but also when we get spurious |
|
|
196 | * notifications for fds from another process. this is partially handled |
|
|
197 | * above with the gencounter check (== our fd is not the event fd), and |
|
|
198 | * partially here, when epoll_ctl returns an error (== a child has the fd |
|
|
199 | * but we closed it). |
|
|
200 | */ |
185 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
201 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
186 | | (want & EV_WRITE ? EPOLLOUT : 0); |
202 | | (want & EV_WRITE ? EPOLLOUT : 0); |
187 | |
203 | |
188 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
204 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
189 | /* which is fortunately easy to do for us. */ |
205 | /* which is fortunately easy to do for us. */ |
190 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
206 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
191 | { |
207 | { |
192 | postfork = 1; /* an error occurred, recreate kernel state */ |
208 | postfork |= 2; /* an error occurred, recreate kernel state */ |
193 | continue; |
209 | continue; |
194 | } |
210 | } |
195 | } |
211 | } |
196 | |
212 | |
197 | fd_event (EV_A_ fd, got); |
213 | fd_event (EV_A_ fd, got); |
… | |
… | |
212 | unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE); |
228 | unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE); |
213 | |
229 | |
214 | if (anfds [fd].emask & EV_EMASK_EPERM && events) |
230 | if (anfds [fd].emask & EV_EMASK_EPERM && events) |
215 | fd_event (EV_A_ fd, events); |
231 | fd_event (EV_A_ fd, events); |
216 | else |
232 | else |
|
|
233 | { |
217 | epoll_eperms [i] = epoll_eperms [--epoll_epermcnt]; |
234 | epoll_eperms [i] = epoll_eperms [--epoll_epermcnt]; |
|
|
235 | anfds [fd].emask = 0; |
|
|
236 | } |
|
|
237 | } |
|
|
238 | } |
|
|
239 | |
|
|
240 | static int |
|
|
241 | epoll_epoll_create (void) |
|
|
242 | { |
|
|
243 | int fd; |
|
|
244 | |
|
|
245 | #if defined EPOLL_CLOEXEC && !defined __ANDROID__ |
|
|
246 | fd = epoll_create1 (EPOLL_CLOEXEC); |
|
|
247 | |
|
|
248 | if (fd < 0 && (errno == EINVAL || errno == ENOSYS)) |
|
|
249 | #endif |
218 | } |
250 | { |
219 | } |
251 | fd = epoll_create (256); |
220 | |
252 | |
|
|
253 | if (fd >= 0) |
|
|
254 | fcntl (fd, F_SETFD, FD_CLOEXEC); |
|
|
255 | } |
|
|
256 | |
|
|
257 | return fd; |
|
|
258 | } |
|
|
259 | |
221 | int inline_size |
260 | inline_size |
|
|
261 | int |
222 | epoll_init (EV_P_ int flags) |
262 | epoll_init (EV_P_ int flags) |
223 | { |
263 | { |
224 | #ifdef EPOLL_CLOEXEC |
264 | if ((backend_fd = epoll_epoll_create ()) < 0) |
225 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
|
|
226 | |
|
|
227 | if (backend_fd <= 0) |
|
|
228 | #endif |
|
|
229 | backend_fd = epoll_create (256); |
|
|
230 | |
|
|
231 | if (backend_fd < 0) |
|
|
232 | return 0; |
265 | return 0; |
233 | |
266 | |
234 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
267 | backend_mintime = 1e-3; /* epoll does sometimes return early, this is just to avoid the worst */ |
235 | |
|
|
236 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
|
|
237 | backend_modify = epoll_modify; |
268 | backend_modify = epoll_modify; |
238 | backend_poll = epoll_poll; |
269 | backend_poll = epoll_poll; |
239 | |
270 | |
240 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
271 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
241 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
272 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
242 | |
273 | |
243 | return EVBACKEND_EPOLL; |
274 | return EVBACKEND_EPOLL; |
244 | } |
275 | } |
245 | |
276 | |
246 | void inline_size |
277 | inline_size |
|
|
278 | void |
247 | epoll_destroy (EV_P) |
279 | epoll_destroy (EV_P) |
248 | { |
280 | { |
249 | ev_free (epoll_events); |
281 | ev_free (epoll_events); |
250 | array_free (epoll_eperm, EMPTY); |
282 | array_free (epoll_eperm, EMPTY); |
251 | } |
283 | } |
252 | |
284 | |
253 | void inline_size |
285 | inline_size |
|
|
286 | void |
254 | epoll_fork (EV_P) |
287 | epoll_fork (EV_P) |
255 | { |
288 | { |
256 | close (backend_fd); |
289 | close (backend_fd); |
257 | |
290 | |
258 | while ((backend_fd = epoll_create (256)) < 0) |
291 | while ((backend_fd = epoll_epoll_create ()) < 0) |
259 | ev_syserr ("(libev) epoll_create"); |
292 | ev_syserr ("(libev) epoll_create"); |
260 | |
293 | |
261 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
|
|
262 | |
|
|
263 | fd_rearm_all (EV_A); |
294 | fd_rearm_all (EV_A); |
264 | } |
295 | } |
265 | |
296 | |