|
|
1 | /* |
|
|
2 | * libev epoll fd activity backend |
|
|
3 | * |
|
|
4 | * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libev@schmorp.de> |
|
|
5 | * All rights reserved. |
|
|
6 | * |
|
|
7 | * Redistribution and use in source and binary forms, with or without modifica- |
|
|
8 | * tion, are permitted provided that the following conditions are met: |
|
|
9 | * |
|
|
10 | * 1. Redistributions of source code must retain the above copyright notice, |
|
|
11 | * this list of conditions and the following disclaimer. |
|
|
12 | * |
|
|
13 | * 2. Redistributions in binary form must reproduce the above copyright |
|
|
14 | * notice, this list of conditions and the following disclaimer in the |
|
|
15 | * documentation and/or other materials provided with the distribution. |
|
|
16 | * |
|
|
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- |
|
|
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- |
|
|
21 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
|
|
22 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
|
|
23 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
|
|
24 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
|
|
25 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
|
|
26 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
27 | * |
|
|
28 | * Alternatively, the contents of this file may be used under the terms of |
|
|
29 | * the GNU General Public License ("GPL") version 2 or any later version, |
|
|
30 | * in which case the provisions of the GPL are applicable instead of |
|
|
31 | * the above. If you wish to allow the use of your version of this file |
|
|
32 | * only under the terms of the GPL and not to allow others to use your |
|
|
33 | * version of this file under the BSD license, indicate your decision |
|
|
34 | * by deleting the provisions above and replace them with the notice |
|
|
35 | * and other provisions required by the GPL. If you do not delete the |
|
|
36 | * provisions above, a recipient may use your version of this file under |
|
|
37 | * either the BSD or the GPL. |
|
|
38 | */ |
|
|
39 | |
|
|
40 | /* |
|
|
41 | * general notes about epoll: |
|
|
42 | * |
|
|
43 | * a) epoll silently removes fds from the fd set. as nothing tells us |
|
|
44 | * that an fd has been removed otherwise, we have to continually |
|
|
45 | * "rearm" fds that we suspect *might* have changed (same |
|
|
46 | * problem with kqueue, but much less costly there). |
|
|
47 | * b) the fact that ADD != MOD creates a lot of extra syscalls due to a) |
|
|
48 | * and seems not to have any advantage. |
|
|
49 | * c) the inability to handle fork or file descriptors (think dup) |
|
|
50 | * limits the applicability over poll, so this is not a generic |
|
|
51 | * poll replacement. |
|
|
52 | * |
|
|
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 |
|
|
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. |
|
|
58 | */ |
|
|
59 | |
1 | #include <sys/epoll.h> |
60 | #include <sys/epoll.h> |
2 | |
61 | |
3 | static int epoll_fd = -1; |
|
|
4 | |
|
|
5 | static void |
62 | static void |
6 | epoll_modify (int fd, int oev, int nev) |
63 | epoll_modify (EV_P_ int fd, int oev, int nev) |
7 | { |
64 | { |
8 | int mode = nev ? oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD : EPOLL_CTL_DEL; |
|
|
9 | |
|
|
10 | struct epoll_event ev; |
65 | struct epoll_event ev; |
11 | ev.data.fd = fd; |
66 | unsigned char oldmask; |
12 | ev.events = |
67 | |
|
|
68 | /* |
|
|
69 | * we handle EPOLL_CTL_DEL by ignoring it here |
|
|
70 | * on the assumption that the fd is gone anyways |
|
|
71 | * if that is wrong, we have to handle the spurious |
|
|
72 | * event in epoll_poll. |
|
|
73 | * if the fd is added again, we try to ADD it, and, if that |
|
|
74 | * fails, we assume it still has the same eventmask. |
|
|
75 | */ |
|
|
76 | if (!nev) |
|
|
77 | return; |
|
|
78 | |
|
|
79 | oldmask = anfds [fd].emask; |
|
|
80 | anfds [fd].emask = nev; |
|
|
81 | |
|
|
82 | /* store the generation counter in the upper 32 bits, the fd in the lower 32 bits */ |
|
|
83 | ev.data.u64 = (uint64_t)(uint32_t)fd |
|
|
84 | | ((uint64_t)(uint32_t)++anfds [fd].egen << 32); |
13 | (nev & EV_READ ? EPOLLIN : 0) |
85 | ev.events = (nev & EV_READ ? EPOLLIN : 0) |
14 | | (nev & EV_WRITE ? EPOLLOUT : 0); |
86 | | (nev & EV_WRITE ? EPOLLOUT : 0); |
15 | |
87 | |
16 | fprintf (stderr, "reify %d,%d,%d m%d (r=%d)\n", fd, oev, nev, mode,//D |
88 | if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev))) |
17 | epoll_ctl (epoll_fd, mode, fd, &ev) |
89 | return; |
18 | );//D |
|
|
19 | } |
|
|
20 | |
90 | |
21 | void epoll_postfork_child (void) |
91 | if (expect_true (errno == ENOENT)) |
22 | { |
92 | { |
23 | int fd; |
93 | /* if ENOENT then the fd went away, so try to do the right thing */ |
|
|
94 | if (!nev) |
|
|
95 | goto dec_egen; |
24 | |
96 | |
25 | epoll_fd = epoll_create (256); |
97 | if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev)) |
26 | fcntl (epoll_fd, F_SETFD, FD_CLOEXEC); |
98 | return; |
|
|
99 | } |
|
|
100 | else if (expect_true (errno == EEXIST)) |
|
|
101 | { |
|
|
102 | /* EEXIST means we ignored a previous DEL, but the fd is still active */ |
|
|
103 | /* if the kernel mask is the same as the new mask, we assume it hasn't changed */ |
|
|
104 | if (oldmask == nev) |
|
|
105 | goto dec_egen; |
27 | |
106 | |
28 | /* re-register interest in fds */ |
107 | if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) |
29 | for (fd = 0; fd < anfdmax; ++fd) |
108 | return; |
30 | if (anfds [fd].wev) |
109 | } |
31 | epoll_modify (fd, EV_NONE, anfds [fd].wev); |
|
|
32 | } |
|
|
33 | |
110 | |
34 | static struct epoll_event *events; |
111 | fd_kill (EV_A_ fd); |
35 | static int eventmax; |
|
|
36 | |
112 | |
|
|
113 | dec_egen: |
|
|
114 | /* we didn't successfully call epoll_ctl, so decrement the generation counter again */ |
|
|
115 | --anfds [fd].egen; |
|
|
116 | } |
|
|
117 | |
|
|
118 | static void |
37 | static void epoll_poll (ev_tstamp timeout) |
119 | epoll_poll (EV_P_ ev_tstamp timeout) |
38 | { |
120 | { |
39 | int eventcnt = epoll_wait (epoll_fd, events, eventmax, ceil (timeout * 1000.)); |
|
|
40 | 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 | eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.)); |
|
|
128 | EV_ACQUIRE_CB; |
41 | |
129 | |
42 | if (eventcnt < 0) |
130 | if (expect_false (eventcnt < 0)) |
|
|
131 | { |
|
|
132 | if (errno != EINTR) |
|
|
133 | ev_syserr ("(libev) epoll_wait"); |
|
|
134 | |
43 | return; |
135 | return; |
|
|
136 | } |
44 | |
137 | |
45 | for (i = 0; i < eventcnt; ++i) |
138 | for (i = 0; i < eventcnt; ++i) |
46 | fd_event ( |
139 | { |
47 | events [i].data.fd, |
140 | struct epoll_event *ev = epoll_events + i; |
|
|
141 | |
|
|
142 | int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */ |
|
|
143 | int want = anfds [fd].events; |
48 | (events [i].events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
144 | int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) |
49 | | (events [i].events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0) |
145 | | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); |
50 | ); |
146 | |
|
|
147 | /* check for spurious notification */ |
|
|
148 | if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) |
|
|
149 | { |
|
|
150 | /* recreate kernel state */ |
|
|
151 | postfork = 1; |
|
|
152 | continue; |
|
|
153 | } |
|
|
154 | |
|
|
155 | if (expect_false (got & ~want)) |
|
|
156 | { |
|
|
157 | anfds [fd].emask = want; |
|
|
158 | |
|
|
159 | /* we received an event but are not interested in it, try mod or del */ |
|
|
160 | /* I don't think we ever need MOD, but let's handle it anyways */ |
|
|
161 | ev->events = (want & EV_READ ? EPOLLIN : 0) |
|
|
162 | | (want & EV_WRITE ? EPOLLOUT : 0); |
|
|
163 | |
|
|
164 | /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ |
|
|
165 | /* which is fortunately easy to do for us. */ |
|
|
166 | if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) |
|
|
167 | { |
|
|
168 | postfork = 1; /* an error occured, recreate kernel state */ |
|
|
169 | continue; |
|
|
170 | } |
|
|
171 | } |
|
|
172 | |
|
|
173 | fd_event (EV_A_ fd, got); |
|
|
174 | } |
51 | |
175 | |
52 | /* if the receive array was full, increase its size */ |
176 | /* if the receive array was full, increase its size */ |
53 | if (eventcnt == eventmax) |
177 | if (expect_false (eventcnt == epoll_eventmax)) |
54 | { |
178 | { |
55 | free (events); |
179 | ev_free (epoll_events); |
56 | eventmax += eventmax >> 1; |
180 | epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1); |
57 | events = malloc (sizeof (struct epoll_event) * eventmax); |
181 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
58 | } |
182 | } |
59 | } |
183 | } |
60 | |
184 | |
|
|
185 | int inline_size |
61 | void epoll_init (int flags) |
186 | epoll_init (EV_P_ int flags) |
62 | { |
187 | { |
|
|
188 | #ifdef EPOLL_CLOEXEC |
|
|
189 | backend_fd = epoll_create1 (EPOLL_CLOEXEC); |
|
|
190 | |
|
|
191 | if (backend_fd <= 0) |
|
|
192 | #endif |
63 | epoll_fd = epoll_create (256); |
193 | backend_fd = epoll_create (256); |
64 | |
194 | |
65 | if (epoll_fd < 0) |
195 | if (backend_fd < 0) |
66 | return; |
196 | return 0; |
67 | |
197 | |
68 | fcntl (epoll_fd, F_SETFD, FD_CLOEXEC); |
198 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
69 | |
199 | |
70 | ev_method = EVMETHOD_EPOLL; |
200 | backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ |
71 | method_fudge = 1e-3; /* needed to compensate for epoll returning early */ |
|
|
72 | method_modify = epoll_modify; |
201 | backend_modify = epoll_modify; |
73 | method_poll = epoll_poll; |
202 | backend_poll = epoll_poll; |
74 | |
203 | |
75 | eventmax = 64; /* intiial number of events receivable per poll */ |
204 | epoll_eventmax = 64; /* initial number of events receivable per poll */ |
76 | events = malloc (sizeof (struct epoll_event) * eventmax); |
205 | epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); |
|
|
206 | |
|
|
207 | return EVBACKEND_EPOLL; |
77 | } |
208 | } |
|
|
209 | |
|
|
210 | void inline_size |
|
|
211 | epoll_destroy (EV_P) |
|
|
212 | { |
|
|
213 | ev_free (epoll_events); |
|
|
214 | } |
|
|
215 | |
|
|
216 | void inline_size |
|
|
217 | epoll_fork (EV_P) |
|
|
218 | { |
|
|
219 | close (backend_fd); |
|
|
220 | |
|
|
221 | while ((backend_fd = epoll_create (256)) < 0) |
|
|
222 | ev_syserr ("(libev) epoll_create"); |
|
|
223 | |
|
|
224 | fcntl (backend_fd, F_SETFD, FD_CLOEXEC); |
|
|
225 | |
|
|
226 | fd_rearm_all (EV_A); |
|
|
227 | } |
|
|
228 | |