… | |
… | |
4 | * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libev@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010,2011 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, |
… | |
… | |
141 | static void |
141 | static void |
142 | epoll_poll (EV_P_ ev_tstamp timeout) |
142 | epoll_poll (EV_P_ ev_tstamp timeout) |
143 | { |
143 | { |
144 | int i; |
144 | int i; |
145 | int eventcnt; |
145 | int eventcnt; |
146 | |
146 | |
|
|
147 | if (expect_false (epoll_epermcnt)) |
|
|
148 | timeout = 0.; |
|
|
149 | |
147 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
150 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
148 | /* the default libev max wait time, however. */ |
151 | /* the default libev max wait time, however. */ |
149 | EV_RELEASE_CB; |
152 | EV_RELEASE_CB; |
150 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.)); |
153 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3); |
151 | EV_ACQUIRE_CB; |
154 | EV_ACQUIRE_CB; |
152 | |
155 | |
153 | if (expect_false (eventcnt < 0)) |
156 | if (expect_false (eventcnt < 0)) |
154 | { |
157 | { |
155 | if (errno != EINTR) |
158 | if (errno != EINTR) |
… | |
… | |
165 | int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */ |
168 | int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */ |
166 | int want = anfds [fd].events; |
169 | int want = anfds [fd].events; |
167 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
170 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
168 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
171 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
169 | |
172 | |
|
|
173 | /* |
170 | /* check for spurious notification */ |
174 | * check for spurious notification. |
|
|
175 | * this only finds spurious notifications on egen updates |
|
|
176 | * 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 */ |
177 | * we assume that fd is always in range, as we never shrink the anfds array |
|
|
178 | */ |
172 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
179 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
173 | { |
180 | { |
174 | /* recreate kernel state */ |
181 | /* recreate kernel state */ |
175 | postfork = 1; |
182 | postfork |= 2; |
176 | continue; |
183 | continue; |
177 | } |
184 | } |
178 | |
185 | |
179 | if (expect_false (got & ~want)) |
186 | if (expect_false (got & ~want)) |
180 | { |
187 | { |
181 | anfds [fd].emask = want; |
188 | anfds [fd].emask = want; |
182 | |
189 | |
|
|
190 | /* |
183 | /* we received an event but are not interested in it, try mod or del */ |
191 | * 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 */ |
192 | * this often happens because we optimistically do not unregister fds |
|
|
193 | * when we are no longer interested in them, but also when we get spurious |
|
|
194 | * notifications for fds from another process. this is partially handled |
|
|
195 | * above with the gencounter check (== our fd is not the event fd), and |
|
|
196 | * partially here, when epoll_ctl returns an error (== a child has the fd |
|
|
197 | * but we closed it). |
|
|
198 | */ |
185 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
199 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
186 | | (want & EV_WRITE ? EPOLLOUT : 0); |
200 | | (want & EV_WRITE ? EPOLLOUT : 0); |
187 | |
201 | |
188 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
202 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
189 | /* which is fortunately easy to do for us. */ |
203 | /* which is fortunately easy to do for us. */ |
190 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
204 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
191 | { |
205 | { |
192 | postfork = 1; /* an error occurred, recreate kernel state */ |
206 | postfork |= 2; /* an error occurred, recreate kernel state */ |
193 | continue; |
207 | continue; |
194 | } |
208 | } |
195 | } |
209 | } |
196 | |
210 | |
197 | fd_event (EV_A_ fd, got); |
211 | fd_event (EV_A_ fd, got); |
… | |
… | |
212 | unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE); |
226 | unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE); |
213 | |
227 | |
214 | if (anfds [fd].emask & EV_EMASK_EPERM && events) |
228 | if (anfds [fd].emask & EV_EMASK_EPERM && events) |
215 | fd_event (EV_A_ fd, events); |
229 | fd_event (EV_A_ fd, events); |
216 | else |
230 | else |
|
|
231 | { |
217 | epoll_eperms [i] = epoll_eperms [--epoll_epermcnt]; |
232 | epoll_eperms [i] = epoll_eperms [--epoll_epermcnt]; |
|
|
233 | anfds [fd].emask = 0; |
|
|
234 | } |
218 | } |
235 | } |
219 | } |
236 | } |
220 | |
237 | |
221 | int inline_size |
238 | inline_size |
|
|
239 | int |
222 | epoll_init (EV_P_ int flags) |
240 | epoll_init (EV_P_ int flags) |
223 | { |
241 | { |
224 | #ifdef EPOLL_CLOEXEC |
242 | #if defined EPOLL_CLOEXEC && !defined __ANDROID__ |
225 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
243 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
226 | |
244 | |
227 | if (backend_fd <= 0) |
245 | if (backend_fd < 0 && (errno == EINVAL || errno == ENOSYS)) |
228 | #endif |
246 | #endif |
229 | backend_fd = epoll_create (256); |
247 | backend_fd = epoll_create (256); |
230 | |
248 | |
231 | if (backend_fd < 0) |
249 | if (backend_fd < 0) |
232 | return 0; |
250 | return 0; |
233 | |
251 | |
234 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
252 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
235 | |
253 | |
236 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
254 | backend_mintime = 1e-3; /* epoll does sometimes return early, this is just to avoid the worst */ |
237 | backend_modify = epoll_modify; |
255 | backend_modify = epoll_modify; |
238 | backend_poll = epoll_poll; |
256 | backend_poll = epoll_poll; |
239 | |
257 | |
240 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
258 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
241 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
259 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
242 | |
260 | |
243 | return EVBACKEND_EPOLL; |
261 | return EVBACKEND_EPOLL; |
244 | } |
262 | } |
245 | |
263 | |
246 | void inline_size |
264 | inline_size |
|
|
265 | void |
247 | epoll_destroy (EV_P) |
266 | epoll_destroy (EV_P) |
248 | { |
267 | { |
249 | ev_free (epoll_events); |
268 | ev_free (epoll_events); |
250 | array_free (epoll_eperm, EMPTY); |
269 | array_free (epoll_eperm, EMPTY); |
251 | } |
270 | } |
252 | |
271 | |
253 | void inline_size |
272 | inline_size |
|
|
273 | void |
254 | epoll_fork (EV_P) |
274 | epoll_fork (EV_P) |
255 | { |
275 | { |
256 | close (backend_fd); |
276 | close (backend_fd); |
257 | |
277 | |
258 | while ((backend_fd = epoll_create (256)) < 0) |
278 | while ((backend_fd = epoll_create (256)) < 0) |