ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev_epoll.c
(Generate patch)

Comparing libev/ev_epoll.c (file contents):
Revision 1.34 by root, Fri May 23 16:37:38 2008 UTC vs.
Revision 1.48 by root, Sat Jul 25 10:14:35 2009 UTC

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 61
60static void 62static void
61epoll_modify (EV_P_ int fd, int oev, int nev) 63epoll_modify (EV_P_ int fd, int oev, int nev)
62{ 64{
63 struct epoll_event ev; 65 struct epoll_event ev;
66 unsigned char oldmask;
64 67
65 /* 68 /*
66 * we handle EPOLL_CTL_DEL by ignoring it here 69 * we handle EPOLL_CTL_DEL by ignoring it here
67 * on the assumption that the fd is gone anyways 70 * on the assumption that the fd is gone anyways
68 * if that is wrong, we have to handle the spurious 71 * if that is wrong, we have to handle the spurious
69 * event in epoll_poll. 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.
70 */ 75 */
71 if (!nev) 76 if (!nev)
72 return; 77 return;
73 78
74 ev.data.u64 = fd; /* use u64 to fully initialise the struct, for nicer strace etc. */ 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);
75 ev.events = (nev & EV_READ ? EPOLLIN : 0) 85 ev.events = (nev & EV_READ ? EPOLLIN : 0)
76 | (nev & EV_WRITE ? EPOLLOUT : 0); 86 | (nev & EV_WRITE ? EPOLLOUT : 0);
77 87
78 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)))
79 return; 89 return;
80 90
81 if (expect_true (errno == ENOENT)) 91 if (expect_true (errno == ENOENT))
82 { 92 {
83 /* on ENOENT the fd went away, so try to do the right thing */ 93 /* if ENOENT then the fd went away, so try to do the right thing */
84 if (!nev) 94 if (!nev)
85 return; 95 goto dec_egen;
86 96
87 if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev)) 97 if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
88 return; 98 return;
89 } 99 }
90 else if (expect_true (errno == EEXIST)) 100 else if (expect_true (errno == EEXIST))
91 { 101 {
92 /* on EEXIST we ignored a previous DEL */ 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;
106
93 if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) 107 if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
94 return; 108 return;
95 } 109 }
96 110
97 fd_kill (EV_A_ fd); 111 fd_kill (EV_A_ fd);
112
113dec_egen:
114 /* we didn't successfully call epoll_ctl, so decrement the generation counter again */
115 --anfds [fd].egen;
98} 116}
99 117
100static void 118static void
101epoll_poll (EV_P_ ev_tstamp timeout) 119epoll_poll (EV_P_ ev_tstamp timeout)
102{ 120{
103 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;
104 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;
105 129
106 if (expect_false (eventcnt < 0)) 130 if (expect_false (eventcnt < 0))
107 { 131 {
108 if (errno != EINTR) 132 if (errno != EINTR)
109 syserr ("(libev) epoll_wait"); 133 ev_syserr ("(libev) epoll_wait");
110 134
111 return; 135 return;
112 } 136 }
113 137
114 for (i = 0; i < eventcnt; ++i) 138 for (i = 0; i < eventcnt; ++i)
115 { 139 {
116 struct epoll_event *ev = epoll_events + i; 140 struct epoll_event *ev = epoll_events + i;
117 141
118 int fd = ev->data.u64; 142 int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */
143 int want = anfds [fd].events;
119 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) 144 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
120 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); 145 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0);
121 int want = anfds [fd].events; 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 }
122 154
123 if (expect_false (got & ~want)) 155 if (expect_false (got & ~want))
124 { 156 {
157 anfds [fd].emask = want;
158
125 /* 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 */
160 /* I don't think we ever need MOD, but let's handle it anyways */
126 ev->events = (want & EV_READ ? EPOLLIN : 0) 161 ev->events = (want & EV_READ ? EPOLLIN : 0)
127 | (want & EV_WRITE ? EPOLLOUT : 0); 162 | (want & EV_WRITE ? EPOLLOUT : 0);
128 163
129 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 }
130 } 169 }
131 170
132 fd_event (EV_A_ fd, got); 171 fd_event (EV_A_ fd, got);
133 } 172 }
134 173
… …
142} 181}
143 182
144int inline_size 183int inline_size
145epoll_init (EV_P_ int flags) 184epoll_init (EV_P_ int flags)
146{ 185{
186#ifdef EPOLL_CLOEXEC
187 backend_fd = epoll_create1 (EPOLL_CLOEXEC);
188
189 if (backend_fd <= 0)
190#endif
147 backend_fd = epoll_create (256); 191 backend_fd = epoll_create (256);
148 192
149 if (backend_fd < 0) 193 if (backend_fd < 0)
150 return 0; 194 return 0;
151 195
152 fcntl (backend_fd, F_SETFD, FD_CLOEXEC); 196 fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
153 197
154 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 */
155 backend_modify = epoll_modify; 199 backend_modify = epoll_modify;
156 backend_poll = epoll_poll; 200 backend_poll = epoll_poll;
157 201
158 epoll_eventmax = 64; /* intiial number of events receivable per poll */ 202 epoll_eventmax = 64; /* initial number of events receivable per poll */
159 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);
160 204
161 return EVBACKEND_EPOLL; 205 return EVBACKEND_EPOLL;
162} 206}
163 207
… …
171epoll_fork (EV_P) 215epoll_fork (EV_P)
172{ 216{
173 close (backend_fd); 217 close (backend_fd);
174 218
175 while ((backend_fd = epoll_create (256)) < 0) 219 while ((backend_fd = epoll_create (256)) < 0)
176 syserr ("(libev) epoll_create"); 220 ev_syserr ("(libev) epoll_create");
177 221
178 fcntl (backend_fd, F_SETFD, FD_CLOEXEC); 222 fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
179 223
180 fd_rearm_all (EV_A); 224 fd_rearm_all (EV_A);
181} 225}

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines