… | |
… | |
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, |
… | |
… | |
115 | if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) |
115 | if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) |
116 | return; |
116 | return; |
117 | } |
117 | } |
118 | else if (expect_true (errno == EPERM)) |
118 | else if (expect_true (errno == EPERM)) |
119 | { |
119 | { |
|
|
120 | /* EPERM means the fd is always ready, but epoll is too snobbish */ |
|
|
121 | /* to handle it, unlike select or poll. */ |
120 | anfds [fd].emask = EV_EMASK_EPERM; |
122 | anfds [fd].emask = EV_EMASK_EPERM; |
121 | |
123 | |
122 | /* add fd to epoll_eperms, if not already inside */ |
124 | /* add fd to epoll_eperms, if not already inside */ |
123 | if (!(oldmask & EV_EMASK_EPERM)) |
125 | if (!(oldmask & EV_EMASK_EPERM)) |
124 | { |
126 | { |
… | |
… | |
139 | static void |
141 | static void |
140 | epoll_poll (EV_P_ ev_tstamp timeout) |
142 | epoll_poll (EV_P_ ev_tstamp timeout) |
141 | { |
143 | { |
142 | int i; |
144 | int i; |
143 | int eventcnt; |
145 | int eventcnt; |
144 | |
146 | |
|
|
147 | if (expect_false (epoll_epermcnt)) |
|
|
148 | timeout = 0.; |
|
|
149 | |
145 | /* 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 */ |
146 | /* the default libev max wait time, however. */ |
151 | /* the default libev max wait time, however. */ |
147 | EV_RELEASE_CB; |
152 | EV_RELEASE_CB; |
148 | 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); |
149 | EV_ACQUIRE_CB; |
154 | EV_ACQUIRE_CB; |
150 | |
155 | |
151 | if (expect_false (eventcnt < 0)) |
156 | if (expect_false (eventcnt < 0)) |
152 | { |
157 | { |
153 | if (errno != EINTR) |
158 | if (errno != EINTR) |
… | |
… | |
163 | 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 */ |
164 | int want = anfds [fd].events; |
169 | int want = anfds [fd].events; |
165 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
170 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
166 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
171 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
167 | |
172 | |
|
|
173 | /* |
168 | /* 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 |
169 | /* 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 | */ |
170 | 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))) |
171 | { |
180 | { |
172 | /* recreate kernel state */ |
181 | /* recreate kernel state */ |
173 | postfork = 1; |
182 | postfork = 1; |
174 | continue; |
183 | continue; |
… | |
… | |
176 | |
185 | |
177 | if (expect_false (got & ~want)) |
186 | if (expect_false (got & ~want)) |
178 | { |
187 | { |
179 | anfds [fd].emask = want; |
188 | anfds [fd].emask = want; |
180 | |
189 | |
|
|
190 | /* |
181 | /* 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 |
182 | /* 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 | */ |
183 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
199 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
184 | | (want & EV_WRITE ? EPOLLOUT : 0); |
200 | | (want & EV_WRITE ? EPOLLOUT : 0); |
185 | |
201 | |
186 | /* 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, */ |
187 | /* which is fortunately easy to do for us. */ |
203 | /* which is fortunately easy to do for us. */ |
… | |
… | |
220 | epoll_init (EV_P_ int flags) |
236 | epoll_init (EV_P_ int flags) |
221 | { |
237 | { |
222 | #ifdef EPOLL_CLOEXEC |
238 | #ifdef EPOLL_CLOEXEC |
223 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
239 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
224 | |
240 | |
225 | if (backend_fd <= 0) |
241 | if (backend_fd < 0 && (errno == EINVAL || errno == ENOSYS)) |
226 | #endif |
242 | #endif |
227 | backend_fd = epoll_create (256); |
243 | backend_fd = epoll_create (256); |
228 | |
244 | |
229 | if (backend_fd < 0) |
245 | if (backend_fd < 0) |
230 | return 0; |
246 | return 0; |
231 | |
247 | |
232 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
248 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
233 | |
249 | |
234 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
250 | backend_mintime = 1e-3; /* epoll does sometimes return early, this is just to avoid the worst */ |
235 | backend_modify = epoll_modify; |
251 | backend_modify = epoll_modify; |
236 | backend_poll = epoll_poll; |
252 | backend_poll = epoll_poll; |
237 | |
253 | |
238 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
254 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
239 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
255 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
240 | |
256 | |
241 | return EVBACKEND_EPOLL; |
257 | return EVBACKEND_EPOLL; |