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 | * |
… | |
… | |
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 | } |
… | |
… | |
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, |
153 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3); |
151 | epoll_epermcnt ? 0 : ev_timeout_to_ms (timeout)); |
|
|
152 | EV_ACQUIRE_CB; |
154 | EV_ACQUIRE_CB; |
153 | |
155 | |
154 | if (expect_false (eventcnt < 0)) |
156 | if (expect_false (eventcnt < 0)) |
155 | { |
157 | { |
156 | if (errno != EINTR) |
158 | if (errno != EINTR) |
… | |
… | |
166 | 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 */ |
167 | int want = anfds [fd].events; |
169 | int want = anfds [fd].events; |
168 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
170 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
169 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
171 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
170 | |
172 | |
|
|
173 | /* |
171 | /* 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 |
172 | /* 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 | */ |
173 | 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))) |
174 | { |
180 | { |
175 | /* recreate kernel state */ |
181 | /* recreate kernel state */ |
176 | postfork = 1; |
182 | postfork |= 2; |
177 | continue; |
183 | continue; |
178 | } |
184 | } |
179 | |
185 | |
180 | if (expect_false (got & ~want)) |
186 | if (expect_false (got & ~want)) |
181 | { |
187 | { |
182 | anfds [fd].emask = want; |
188 | anfds [fd].emask = want; |
183 | |
189 | |
|
|
190 | /* |
184 | /* 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 |
185 | /* 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 | */ |
186 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
199 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
187 | | (want & EV_WRITE ? EPOLLOUT : 0); |
200 | | (want & EV_WRITE ? EPOLLOUT : 0); |
188 | |
201 | |
189 | /* 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, */ |
190 | /* which is fortunately easy to do for us. */ |
203 | /* which is fortunately easy to do for us. */ |
191 | 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)) |
192 | { |
205 | { |
193 | postfork = 1; /* an error occurred, recreate kernel state */ |
206 | postfork |= 2; /* an error occurred, recreate kernel state */ |
194 | continue; |
207 | continue; |
195 | } |
208 | } |
196 | } |
209 | } |
197 | |
210 | |
198 | fd_event (EV_A_ fd, got); |
211 | fd_event (EV_A_ fd, got); |
… | |
… | |
213 | unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE); |
226 | unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE); |
214 | |
227 | |
215 | if (anfds [fd].emask & EV_EMASK_EPERM && events) |
228 | if (anfds [fd].emask & EV_EMASK_EPERM && events) |
216 | fd_event (EV_A_ fd, events); |
229 | fd_event (EV_A_ fd, events); |
217 | else |
230 | else |
|
|
231 | { |
218 | epoll_eperms [i] = epoll_eperms [--epoll_epermcnt]; |
232 | epoll_eperms [i] = epoll_eperms [--epoll_epermcnt]; |
|
|
233 | anfds [fd].emask = 0; |
|
|
234 | } |
|
|
235 | } |
|
|
236 | } |
|
|
237 | |
|
|
238 | static int |
|
|
239 | ev_epoll_create (void) |
|
|
240 | { |
|
|
241 | int fd; |
|
|
242 | |
|
|
243 | #if defined EPOLL_CLOEXEC && !defined __ANDROID__ |
|
|
244 | fd = epoll_create1 (EPOLL_CLOEXEC); |
|
|
245 | |
|
|
246 | if (fd < 0 && (errno == EINVAL || errno == ENOSYS)) |
|
|
247 | #endif |
219 | } |
248 | { |
220 | } |
249 | fd = epoll_create (256); |
221 | |
250 | |
|
|
251 | if (fd >= 0) |
|
|
252 | fcntl (fd, F_SETFD, FD_CLOEXEC); |
|
|
253 | } |
|
|
254 | |
|
|
255 | return fd; |
|
|
256 | } |
|
|
257 | |
222 | int inline_size |
258 | inline_size |
|
|
259 | int |
223 | epoll_init (EV_P_ int flags) |
260 | epoll_init (EV_P_ int flags) |
224 | { |
261 | { |
225 | #ifdef EPOLL_CLOEXEC |
|
|
226 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
|
|
227 | |
|
|
228 | if (backend_fd <= 0) |
|
|
229 | #endif |
|
|
230 | backend_fd = epoll_create (256); |
262 | if ((backend_fd = ev_epoll_create ()) < 0) |
231 | |
|
|
232 | if (backend_fd < 0) |
|
|
233 | return 0; |
263 | return 0; |
234 | |
264 | |
235 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
265 | backend_mintime = 1e-3; /* epoll does sometimes return early, this is just to avoid the worst */ |
236 | |
|
|
237 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
|
|
238 | backend_modify = epoll_modify; |
266 | backend_modify = epoll_modify; |
239 | backend_poll = epoll_poll; |
267 | backend_poll = epoll_poll; |
240 | |
268 | |
241 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
269 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
242 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
270 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
243 | |
271 | |
244 | return EVBACKEND_EPOLL; |
272 | return EVBACKEND_EPOLL; |
245 | } |
273 | } |
246 | |
274 | |
247 | void inline_size |
275 | inline_size |
|
|
276 | void |
248 | epoll_destroy (EV_P) |
277 | epoll_destroy (EV_P) |
249 | { |
278 | { |
250 | ev_free (epoll_events); |
279 | ev_free (epoll_events); |
251 | array_free (epoll_eperm, EMPTY); |
280 | array_free (epoll_eperm, EMPTY); |
252 | } |
281 | } |
253 | |
282 | |
254 | void inline_size |
283 | inline_size |
|
|
284 | void |
255 | epoll_fork (EV_P) |
285 | epoll_fork (EV_P) |
256 | { |
286 | { |
257 | close (backend_fd); |
287 | close (backend_fd); |
258 | |
288 | |
259 | while ((backend_fd = epoll_create (256)) < 0) |
289 | while ((backend_fd = ev_epoll_create ()) < 0) |
260 | ev_syserr ("(libev) epoll_create"); |
290 | ev_syserr ("(libev) epoll_create"); |
261 | |
291 | |
262 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
|
|
263 | |
|
|
264 | fd_rearm_all (EV_A); |
292 | fd_rearm_all (EV_A); |
265 | } |
293 | } |
266 | |
294 | |