1 | /* |
1 | /* |
2 | * libev epoll fd activity backend |
2 | * libev epoll fd activity backend |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008 Marc Alexander Lehmann <libev@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009 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 | * |
… | |
… | |
50 | * limits the applicability over poll, so this is not a generic |
50 | * limits the applicability over poll, so this is not a generic |
51 | * poll replacement. |
51 | * poll replacement. |
52 | * |
52 | * |
53 | * lots of "weird code" and complication handling in this file is due |
53 | * lots of "weird code" and complication handling in this file is due |
54 | * to these design problems with epoll, as we try very hard to avoid |
54 | * to these design problems with epoll, as we try very hard to avoid |
55 | * epoll_ctl syscalls for common usage patterns. |
55 | * epoll_ctl syscalls for common usage patterns and handle the breakage |
|
|
56 | * ensuing from receiving events for closed and otherwise long gone |
|
|
57 | * file descriptors. |
56 | */ |
58 | */ |
57 | |
59 | |
58 | #include <sys/epoll.h> |
60 | #include <sys/epoll.h> |
59 | |
|
|
60 | void inline_size |
|
|
61 | unsigned_char_init (unsigned char *base, int count) |
|
|
62 | { |
|
|
63 | /* memset might be overkill */ |
|
|
64 | while (count--) |
|
|
65 | *base++ = 0; |
|
|
66 | } |
|
|
67 | |
61 | |
68 | static void |
62 | static void |
69 | epoll_modify (EV_P_ int fd, int oev, int nev) |
63 | epoll_modify (EV_P_ int fd, int oev, int nev) |
70 | { |
64 | { |
71 | struct epoll_event ev; |
65 | struct epoll_event ev; |
… | |
… | |
74 | /* |
68 | /* |
75 | * we handle EPOLL_CTL_DEL by ignoring it here |
69 | * we handle EPOLL_CTL_DEL by ignoring it here |
76 | * on the assumption that the fd is gone anyways |
70 | * on the assumption that the fd is gone anyways |
77 | * if that is wrong, we have to handle the spurious |
71 | * if that is wrong, we have to handle the spurious |
78 | * event in epoll_poll. |
72 | * event in epoll_poll. |
79 | * the fd is later added, we try to ADD it, and, if that |
73 | * if the fd is added again, we try to ADD it, and, if that |
80 | * fails, we assume it still has the same eventmask. |
74 | * fails, we assume it still has the same eventmask. |
81 | */ |
75 | */ |
82 | if (!nev) |
76 | if (!nev) |
83 | return; |
77 | return; |
84 | |
78 | |
85 | oldmask = anfds [fd].emask; |
79 | oldmask = anfds [fd].emask; |
86 | anfds [fd].emask = nev; |
80 | anfds [fd].emask = nev; |
87 | |
81 | |
88 | /* store the generation counter in the upper 32 bits */ |
82 | /* store the generation counter in the upper 32 bits, the fd in the lower 32 bits */ |
89 | ev.data.u64 = fd | ((uint64_t)++anfds [fd].egen << 32); |
83 | ev.data.u64 = (uint64_t)(uint32_t)fd |
|
|
84 | | ((uint64_t)(uint32_t)++anfds [fd].egen << 32); |
90 | ev.events = (nev & EV_READ ? EPOLLIN : 0) |
85 | ev.events = (nev & EV_READ ? EPOLLIN : 0) |
91 | | (nev & EV_WRITE ? EPOLLOUT : 0); |
86 | | (nev & EV_WRITE ? EPOLLOUT : 0); |
92 | |
87 | |
93 | if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev))) |
88 | if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev))) |
94 | return; |
89 | return; |
… | |
… | |
122 | |
117 | |
123 | static void |
118 | static void |
124 | epoll_poll (EV_P_ ev_tstamp timeout) |
119 | epoll_poll (EV_P_ ev_tstamp timeout) |
125 | { |
120 | { |
126 | int i; |
121 | int i; |
|
|
122 | int eventcnt; |
|
|
123 | |
|
|
124 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
|
|
125 | /* the default libev max wait time, however. */ |
|
|
126 | EV_RELEASE_CB; |
127 | int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.)); |
127 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.)); |
|
|
128 | EV_ACQUIRE_CB; |
128 | |
129 | |
129 | if (expect_false (eventcnt < 0)) |
130 | if (expect_false (eventcnt < 0)) |
130 | { |
131 | { |
131 | if (errno != EINTR) |
132 | if (errno != EINTR) |
132 | syserr ("(libev) epoll_wait"); |
133 | ev_syserr ("(libev) epoll_wait"); |
133 | |
134 | |
134 | return; |
135 | return; |
135 | } |
136 | } |
136 | |
137 | |
137 | for (i = 0; i < eventcnt; ++i) |
138 | for (i = 0; i < eventcnt; ++i) |
138 | { |
139 | { |
139 | struct epoll_event *ev = epoll_events + i; |
140 | struct epoll_event *ev = epoll_events + i; |
140 | |
141 | |
141 | int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */ |
142 | int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */ |
|
|
143 | int want = anfds [fd].events; |
142 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
144 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
143 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
145 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
144 | int want = anfds [fd].events; |
|
|
145 | |
146 | |
146 | if (anfds [fd].egen != (unsigned char)(ev->data.u64 >> 32)) |
147 | /* check for spurious notification */ |
147 | /*fprintf (stderr, "spurious notification fd %d, %d vs %d\n", fd, (int)(ev->data.u64 >> 32), anfds [fd].egen);*/ |
148 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
|
|
149 | { |
|
|
150 | /* recreate kernel state */ |
|
|
151 | postfork = 1; |
148 | continue; |
152 | continue; |
|
|
153 | } |
149 | |
154 | |
150 | if (expect_false (got & ~want)) |
155 | if (expect_false (got & ~want)) |
151 | { |
156 | { |
152 | anfds [fd].emask = want; |
157 | anfds [fd].emask = want; |
153 | |
158 | |
154 | /* we received an event but are not interested in it, try mod or del */ |
159 | /* we received an event but are not interested in it, try mod or del */ |
155 | /* I don't think we ever need MOD, but let's handle it anyways */ |
160 | /* I don't think we ever need MOD, but let's handle it anyways */ |
156 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
161 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
157 | | (want & EV_WRITE ? EPOLLOUT : 0); |
162 | | (want & EV_WRITE ? EPOLLOUT : 0); |
158 | |
163 | |
159 | epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev); |
164 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
|
|
165 | { |
|
|
166 | postfork = 1; /* an error occured, recreate kernel state */ |
|
|
167 | continue; |
|
|
168 | } |
160 | } |
169 | } |
161 | |
170 | |
162 | fd_event (EV_A_ fd, got); |
171 | fd_event (EV_A_ fd, got); |
163 | } |
172 | } |
164 | |
173 | |
… | |
… | |
172 | } |
181 | } |
173 | |
182 | |
174 | int inline_size |
183 | int inline_size |
175 | epoll_init (EV_P_ int flags) |
184 | epoll_init (EV_P_ int flags) |
176 | { |
185 | { |
|
|
186 | #ifdef EPOLL_CLOEXEC |
|
|
187 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
|
|
188 | |
|
|
189 | if (backend_fd <= 0) |
|
|
190 | #endif |
177 | backend_fd = epoll_create (256); |
191 | backend_fd = epoll_create (256); |
178 | |
192 | |
179 | if (backend_fd < 0) |
193 | if (backend_fd < 0) |
180 | return 0; |
194 | return 0; |
181 | |
195 | |
182 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
196 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
183 | |
197 | |
184 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
198 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
185 | backend_modify = epoll_modify; |
199 | backend_modify = epoll_modify; |
186 | backend_poll = epoll_poll; |
200 | backend_poll = epoll_poll; |
187 | |
201 | |
188 | epoll_eventmax = 64; /* intiial number of events receivable per poll */ |
202 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
189 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
203 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
190 | |
204 | |
191 | return EVBACKEND_EPOLL; |
205 | return EVBACKEND_EPOLL; |
192 | } |
206 | } |
193 | |
207 | |
… | |
… | |
201 | epoll_fork (EV_P) |
215 | epoll_fork (EV_P) |
202 | { |
216 | { |
203 | close (backend_fd); |
217 | close (backend_fd); |
204 | |
218 | |
205 | while ((backend_fd = epoll_create (256)) < 0) |
219 | while ((backend_fd = epoll_create (256)) < 0) |
206 | syserr ("(libev) epoll_create"); |
220 | ev_syserr ("(libev) epoll_create"); |
207 | |
221 | |
208 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
222 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
209 | |
223 | |
210 | fd_rearm_all (EV_A); |
224 | fd_rearm_all (EV_A); |
211 | } |
225 | } |