1 | /* |
1 | /* |
2 | * libev epoll fd activity backend |
2 | * libev epoll fd activity backend |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008,2009,2010 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, |
… | |
… | |
51 | * poll replacement. |
51 | * poll replacement. |
52 | * d) epoll doesn't work the same as select with many file descriptors |
52 | * d) epoll doesn't work the same as select with many file descriptors |
53 | * (such as files). while not critical, no other advanced interface |
53 | * (such as files). while not critical, no other advanced interface |
54 | * seems to share this (rather non-unixy) limitation. |
54 | * seems to share this (rather non-unixy) limitation. |
55 | * e) epoll claims to be embeddable, but in practise you never get |
55 | * e) epoll claims to be embeddable, but in practise you never get |
56 | * a ready event for the epoll fd. |
56 | * a ready event for the epoll fd (broken: <=2.6.26, working: >=2.6.32). |
|
|
57 | * f) epoll_ctl returning EPERM means the fd is always ready. |
57 | * |
58 | * |
58 | * lots of "weird code" and complication handling in this file is due |
59 | * lots of "weird code" and complication handling in this file is due |
59 | * to these design problems with epoll, as we try very hard to avoid |
60 | * to these design problems with epoll, as we try very hard to avoid |
60 | * epoll_ctl syscalls for common usage patterns and handle the breakage |
61 | * epoll_ctl syscalls for common usage patterns and handle the breakage |
61 | * ensuing from receiving events for closed and otherwise long gone |
62 | * ensuing from receiving events for closed and otherwise long gone |
62 | * file descriptors. |
63 | * file descriptors. |
63 | */ |
64 | */ |
64 | |
65 | |
65 | #include <sys/epoll.h> |
66 | #include <sys/epoll.h> |
|
|
67 | |
|
|
68 | #define EV_EMASK_EPERM 0x80 |
66 | |
69 | |
67 | static void |
70 | static void |
68 | epoll_modify (EV_P_ int fd, int oev, int nev) |
71 | epoll_modify (EV_P_ int fd, int oev, int nev) |
69 | { |
72 | { |
70 | struct epoll_event ev; |
73 | struct epoll_event ev; |
… | |
… | |
110 | goto dec_egen; |
113 | goto dec_egen; |
111 | |
114 | |
112 | if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) |
115 | if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) |
113 | return; |
116 | return; |
114 | } |
117 | } |
|
|
118 | else if (expect_true (errno == EPERM)) |
|
|
119 | { |
|
|
120 | /* EPERM means the fd is always ready, but epoll is too snobbish */ |
|
|
121 | /* to handle it, unlike select or poll. */ |
|
|
122 | anfds [fd].emask = EV_EMASK_EPERM; |
|
|
123 | |
|
|
124 | /* add fd to epoll_eperms, if not already inside */ |
|
|
125 | if (!(oldmask & EV_EMASK_EPERM)) |
|
|
126 | { |
|
|
127 | array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, EMPTY2); |
|
|
128 | epoll_eperms [epoll_epermcnt++] = fd; |
|
|
129 | } |
|
|
130 | |
|
|
131 | return; |
|
|
132 | } |
115 | |
133 | |
116 | fd_kill (EV_A_ fd); |
134 | fd_kill (EV_A_ fd); |
117 | |
135 | |
118 | dec_egen: |
136 | dec_egen: |
119 | /* we didn't successfully call epoll_ctl, so decrement the generation counter again */ |
137 | /* we didn't successfully call epoll_ctl, so decrement the generation counter again */ |
… | |
… | |
123 | static void |
141 | static void |
124 | epoll_poll (EV_P_ ev_tstamp timeout) |
142 | epoll_poll (EV_P_ ev_tstamp timeout) |
125 | { |
143 | { |
126 | int i; |
144 | int i; |
127 | int eventcnt; |
145 | int eventcnt; |
128 | |
146 | |
129 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
147 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
130 | /* the default libev max wait time, however. */ |
148 | /* the default libev max wait time, however. */ |
131 | EV_RELEASE_CB; |
149 | EV_RELEASE_CB; |
132 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.)); |
150 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, |
|
|
151 | epoll_epermcnt ? 0 : ev_timeout_to_ms (timeout)); |
133 | EV_ACQUIRE_CB; |
152 | EV_ACQUIRE_CB; |
134 | |
153 | |
135 | if (expect_false (eventcnt < 0)) |
154 | if (expect_false (eventcnt < 0)) |
136 | { |
155 | { |
137 | if (errno != EINTR) |
156 | if (errno != EINTR) |
… | |
… | |
148 | int want = anfds [fd].events; |
167 | int want = anfds [fd].events; |
149 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
168 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
150 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
169 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
151 | |
170 | |
152 | /* check for spurious notification */ |
171 | /* check for spurious notification */ |
|
|
172 | /* we assume that fd is always in range, as we never shrink the anfds array */ |
153 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
173 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
154 | { |
174 | { |
155 | /* recreate kernel state */ |
175 | /* recreate kernel state */ |
156 | postfork = 1; |
176 | postfork = 1; |
157 | continue; |
177 | continue; |
… | |
… | |
168 | |
188 | |
169 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
189 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
170 | /* which is fortunately easy to do for us. */ |
190 | /* which is fortunately easy to do for us. */ |
171 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
191 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
172 | { |
192 | { |
173 | postfork = 1; /* an error occured, recreate kernel state */ |
193 | postfork = 1; /* an error occurred, recreate kernel state */ |
174 | continue; |
194 | continue; |
175 | } |
195 | } |
176 | } |
196 | } |
177 | |
197 | |
178 | fd_event (EV_A_ fd, got); |
198 | fd_event (EV_A_ fd, got); |
… | |
… | |
183 | { |
203 | { |
184 | ev_free (epoll_events); |
204 | ev_free (epoll_events); |
185 | epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1); |
205 | epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1); |
186 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
206 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
187 | } |
207 | } |
|
|
208 | |
|
|
209 | /* now synthesize events for all fds where epoll fails, while select works... */ |
|
|
210 | for (i = epoll_epermcnt; i--; ) |
|
|
211 | { |
|
|
212 | int fd = epoll_eperms [i]; |
|
|
213 | unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE); |
|
|
214 | |
|
|
215 | if (anfds [fd].emask & EV_EMASK_EPERM && events) |
|
|
216 | fd_event (EV_A_ fd, events); |
|
|
217 | else |
|
|
218 | epoll_eperms [i] = epoll_eperms [--epoll_epermcnt]; |
|
|
219 | } |
188 | } |
220 | } |
189 | |
221 | |
190 | int inline_size |
222 | int inline_size |
191 | epoll_init (EV_P_ int flags) |
223 | epoll_init (EV_P_ int flags) |
192 | { |
224 | { |
… | |
… | |
200 | if (backend_fd < 0) |
232 | if (backend_fd < 0) |
201 | return 0; |
233 | return 0; |
202 | |
234 | |
203 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
235 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
204 | |
236 | |
205 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
237 | backend_fudge = 1./1024.; /* epoll does sometimes return early, this is just to avoid the worst */ |
206 | backend_modify = epoll_modify; |
238 | backend_modify = epoll_modify; |
207 | backend_poll = epoll_poll; |
239 | backend_poll = epoll_poll; |
208 | |
240 | |
209 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
241 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
210 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
242 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
… | |
… | |
214 | |
246 | |
215 | void inline_size |
247 | void inline_size |
216 | epoll_destroy (EV_P) |
248 | epoll_destroy (EV_P) |
217 | { |
249 | { |
218 | ev_free (epoll_events); |
250 | ev_free (epoll_events); |
|
|
251 | array_free (epoll_eperm, EMPTY); |
219 | } |
252 | } |
220 | |
253 | |
221 | void inline_size |
254 | void inline_size |
222 | epoll_fork (EV_P) |
255 | epoll_fork (EV_P) |
223 | { |
256 | { |