… | |
… | |
91 | ev.data.u64 = (uint64_t)(uint32_t)fd |
91 | ev.data.u64 = (uint64_t)(uint32_t)fd |
92 | | ((uint64_t)(uint32_t)++anfds [fd].egen << 32); |
92 | | ((uint64_t)(uint32_t)++anfds [fd].egen << 32); |
93 | ev.events = (nev & EV_READ ? EPOLLIN : 0) |
93 | ev.events = (nev & EV_READ ? EPOLLIN : 0) |
94 | | (nev & EV_WRITE ? EPOLLOUT : 0); |
94 | | (nev & EV_WRITE ? EPOLLOUT : 0); |
95 | |
95 | |
96 | if (expect_true (!epoll_ctl (backend_fd, oev && oldmask != nev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev))) |
96 | if (ecb_expect_true (!epoll_ctl (backend_fd, oev && oldmask != nev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev))) |
97 | return; |
97 | return; |
98 | |
98 | |
99 | if (expect_true (errno == ENOENT)) |
99 | if (ecb_expect_true (errno == ENOENT)) |
100 | { |
100 | { |
101 | /* if ENOENT then the fd went away, so try to do the right thing */ |
101 | /* if ENOENT then the fd went away, so try to do the right thing */ |
102 | if (!nev) |
102 | if (!nev) |
103 | goto dec_egen; |
103 | goto dec_egen; |
104 | |
104 | |
105 | if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev)) |
105 | if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev)) |
106 | return; |
106 | return; |
107 | } |
107 | } |
108 | else if (expect_true (errno == EEXIST)) |
108 | else if (ecb_expect_true (errno == EEXIST)) |
109 | { |
109 | { |
110 | /* EEXIST means we ignored a previous DEL, but the fd is still active */ |
110 | /* EEXIST means we ignored a previous DEL, but the fd is still active */ |
111 | /* if the kernel mask is the same as the new mask, we assume it hasn't changed */ |
111 | /* if the kernel mask is the same as the new mask, we assume it hasn't changed */ |
112 | if (oldmask == nev) |
112 | if (oldmask == nev) |
113 | goto dec_egen; |
113 | goto dec_egen; |
114 | |
114 | |
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 (ecb_expect_true (errno == EPERM)) |
119 | { |
119 | { |
120 | /* EPERM means the fd is always ready, but epoll is too snobbish */ |
120 | /* EPERM means the fd is always ready, but epoll is too snobbish */ |
121 | /* to handle it, unlike select or poll. */ |
121 | /* to handle it, unlike select or poll. */ |
122 | anfds [fd].emask = EV_EMASK_EPERM; |
122 | anfds [fd].emask = EV_EMASK_EPERM; |
123 | |
123 | |
… | |
… | |
128 | epoll_eperms [epoll_epermcnt++] = fd; |
128 | epoll_eperms [epoll_epermcnt++] = fd; |
129 | } |
129 | } |
130 | |
130 | |
131 | return; |
131 | return; |
132 | } |
132 | } |
|
|
133 | else |
|
|
134 | assert (("libev: I/O watcher with invalid fd found in epoll_ctl", errno != EBADF && errno != ELOOP && errno != EINVAL)); |
133 | |
135 | |
134 | fd_kill (EV_A_ fd); |
136 | fd_kill (EV_A_ fd); |
135 | |
137 | |
136 | dec_egen: |
138 | dec_egen: |
137 | /* we didn't successfully call epoll_ctl, so decrement the generation counter again */ |
139 | /* we didn't successfully call epoll_ctl, so decrement the generation counter again */ |
… | |
… | |
142 | epoll_poll (EV_P_ ev_tstamp timeout) |
144 | epoll_poll (EV_P_ ev_tstamp timeout) |
143 | { |
145 | { |
144 | int i; |
146 | int i; |
145 | int eventcnt; |
147 | int eventcnt; |
146 | |
148 | |
147 | if (expect_false (epoll_epermcnt)) |
149 | if (ecb_expect_false (epoll_epermcnt)) |
148 | timeout = 0.; |
150 | timeout = 0.; |
149 | |
151 | |
150 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
152 | /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */ |
151 | /* the default libev max wait time, however. */ |
153 | /* the default libev max wait time, however. */ |
152 | EV_RELEASE_CB; |
154 | EV_RELEASE_CB; |
153 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3); |
155 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3 + 0.9999); |
154 | EV_ACQUIRE_CB; |
156 | EV_ACQUIRE_CB; |
155 | |
157 | |
156 | if (expect_false (eventcnt < 0)) |
158 | if (ecb_expect_false (eventcnt < 0)) |
157 | { |
159 | { |
158 | if (errno != EINTR) |
160 | if (errno != EINTR) |
159 | ev_syserr ("(libev) epoll_wait"); |
161 | ev_syserr ("(libev) epoll_wait"); |
160 | |
162 | |
161 | return; |
163 | return; |
… | |
… | |
174 | * check for spurious notification. |
176 | * check for spurious notification. |
175 | * this only finds spurious notifications on egen updates |
177 | * this only finds spurious notifications on egen updates |
176 | * other spurious notifications will be found by epoll_ctl, below |
178 | * other spurious notifications will be found by epoll_ctl, below |
177 | * we assume that fd is always in range, as we never shrink the anfds array |
179 | * we assume that fd is always in range, as we never shrink the anfds array |
178 | */ |
180 | */ |
179 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
181 | if (ecb_expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
180 | { |
182 | { |
181 | /* recreate kernel state */ |
183 | /* recreate kernel state */ |
182 | postfork |= 2; |
184 | postfork |= 2; |
183 | continue; |
185 | continue; |
184 | } |
186 | } |
185 | |
187 | |
186 | if (expect_false (got & ~want)) |
188 | if (ecb_expect_false (got & ~want)) |
187 | { |
189 | { |
188 | anfds [fd].emask = want; |
190 | anfds [fd].emask = want; |
189 | |
191 | |
190 | /* |
192 | /* |
191 | * we received an event but are not interested in it, try mod or del |
193 | * we received an event but are not interested in it, try mod or del |
… | |
… | |
193 | * when we are no longer interested in them, but also when we get spurious |
195 | * 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 |
196 | * notifications for fds from another process. this is partially handled |
195 | * above with the gencounter check (== our fd is not the event fd), and |
197 | * 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 |
198 | * partially here, when epoll_ctl returns an error (== a child has the fd |
197 | * but we closed it). |
199 | * but we closed it). |
|
|
200 | * note: for events such as POLLHUP, where we can't know whether it refers |
|
|
201 | * to EV_READ or EV_WRITE, we might issue redundant EPOLL_CTL_MOD calls. |
198 | */ |
202 | */ |
199 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
203 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
200 | | (want & EV_WRITE ? EPOLLOUT : 0); |
204 | | (want & EV_WRITE ? EPOLLOUT : 0); |
201 | |
205 | |
202 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
206 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
… | |
… | |
210 | |
214 | |
211 | fd_event (EV_A_ fd, got); |
215 | fd_event (EV_A_ fd, got); |
212 | } |
216 | } |
213 | |
217 | |
214 | /* if the receive array was full, increase its size */ |
218 | /* if the receive array was full, increase its size */ |
215 | if (expect_false (eventcnt == epoll_eventmax)) |
219 | if (ecb_expect_false (eventcnt == epoll_eventmax)) |
216 | { |
220 | { |
217 | ev_free (epoll_events); |
221 | ev_free (epoll_events); |
218 | epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1); |
222 | epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1); |
219 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
223 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
220 | } |
224 | } |
… | |
… | |
278 | { |
282 | { |
279 | ev_free (epoll_events); |
283 | ev_free (epoll_events); |
280 | array_free (epoll_eperm, EMPTY); |
284 | array_free (epoll_eperm, EMPTY); |
281 | } |
285 | } |
282 | |
286 | |
283 | inline_size |
287 | ecb_cold |
284 | void |
288 | static void |
285 | epoll_fork (EV_P) |
289 | epoll_fork (EV_P) |
286 | { |
290 | { |
287 | close (backend_fd); |
291 | close (backend_fd); |
288 | |
292 | |
289 | while ((backend_fd = epoll_epoll_create ()) < 0) |
293 | while ((backend_fd = epoll_epoll_create ()) < 0) |