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 1.35 by root, Thu Oct 23 04:56:49 2008 UTC vs.
Revision 1.40 by root, Wed Oct 29 07:09:37 2008 UTC

… …
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
60void inline_size
61unsigned_char_init (unsigned char *base, int count)
62{
63 /* memset might be overkill */
64 while (count--)
65 *base++ = 0;
66}
67 61
68static void 62static void
69epoll_modify (EV_P_ int fd, int oev, int nev) 63epoll_modify (EV_P_ int fd, int oev, int nev)
70{ 64{
71 struct epoll_event ev; 65 struct epoll_event ev;
… …
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 ev.data.u64 = fd; /* use u64 to fully initialise the struct, for nicer strace etc. */ 82 /* store the generation counter in the upper 32 bits */
83 ev.data.u64 = fd | ((uint64_t)(uint32_t)++anfds [fd].egen << 32);
89 ev.events = (nev & EV_READ ? EPOLLIN : 0) 84 ev.events = (nev & EV_READ ? EPOLLIN : 0)
90 | (nev & EV_WRITE ? EPOLLOUT : 0); 85 | (nev & EV_WRITE ? EPOLLOUT : 0);
91 86
92 if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev))) 87 if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
93 return; 88 return;
94 89
95 if (expect_true (errno == ENOENT)) 90 if (expect_true (errno == ENOENT))
96 { 91 {
97 /* if ENOENT then the fd went away, so try to do the right thing */ 92 /* if ENOENT then the fd went away, so try to do the right thing */
98 if (!nev) 93 if (!nev)
99 return; 94 goto dec_egen;
100 95
101 if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev)) 96 if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
102 return; 97 return;
103 } 98 }
104 else if (expect_true (errno == EEXIST)) 99 else if (expect_true (errno == EEXIST))
105 { 100 {
106 /* EEXIST means we ignored a previous DEL, but the fd is still active */ 101 /* EEXIST means we ignored a previous DEL, but the fd is still active */
107 /* if the kernel mask is the same as the new mask, we assume it hasn't changed */ 102 /* if the kernel mask is the same as the new mask, we assume it hasn't changed */
103 if (oldmask == nev)
104 goto dec_egen;
105
108 if (oldmask == nev || !epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) 106 if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
109 return; 107 return;
110 } 108 }
111 109
112 fd_kill (EV_A_ fd); 110 fd_kill (EV_A_ fd);
111
112dec_egen:
113 /* we didn't successfully call epoll_ctl, so decrement the generation counter again */
114 --anfds [fd].egen;
113} 115}
114 116
115static void 117static void
116epoll_poll (EV_P_ ev_tstamp timeout) 118epoll_poll (EV_P_ ev_tstamp timeout)
117{ 119{
… …
119 int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.)); 121 int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
120 122
121 if (expect_false (eventcnt < 0)) 123 if (expect_false (eventcnt < 0))
122 { 124 {
123 if (errno != EINTR) 125 if (errno != EINTR)
124 syserr ("(libev) epoll_wait"); 126 ev_syserr ("(libev) epoll_wait");
125 127
126 return; 128 return;
127 } 129 }
128 130
129 for (i = 0; i < eventcnt; ++i) 131 for (i = 0; i < eventcnt; ++i)
130 { 132 {
131 struct epoll_event *ev = epoll_events + i; 133 struct epoll_event *ev = epoll_events + i;
132 134
133 int fd = ev->data.u64; 135 int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */
136 int want = anfds [fd].events;
134 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) 137 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
135 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); 138 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0);
136 int want = anfds [fd].events; 139
140 /* check for spurious notification */
141 if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32)))
142 {
143 /* recreate kernel state */
144 postfork = 1;
145 continue;
146 }
137 147
138 if (expect_false (got & ~want)) 148 if (expect_false (got & ~want))
139 { 149 {
140 anfds [fd].emask = want; 150 anfds [fd].emask = want;
141 151
142 /* we received an event but are not interested in it, try mod or del */ 152 /* we received an event but are not interested in it, try mod or del */
143 /* I don't think we ever need MOD, but let's handle it anyways */ 153 /* I don't think we ever need MOD, but let's handle it anyways */
144 ev->events = (want & EV_READ ? EPOLLIN : 0) 154 ev->events = (want & EV_READ ? EPOLLIN : 0)
145 | (want & EV_WRITE ? EPOLLOUT : 0); 155 | (want & EV_WRITE ? EPOLLOUT : 0);
146 156
147 epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev); 157 if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev))
158 {
159 postfork = 1; /* an error occured, recreate kernel state */
160 continue;
161 }
148 } 162 }
149 163
150 fd_event (EV_A_ fd, got); 164 fd_event (EV_A_ fd, got);
151 } 165 }
152 166
… …
171 185
172 backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */ 186 backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */
173 backend_modify = epoll_modify; 187 backend_modify = epoll_modify;
174 backend_poll = epoll_poll; 188 backend_poll = epoll_poll;
175 189
176 epoll_eventmax = 64; /* intiial number of events receivable per poll */ 190 epoll_eventmax = 64; /* initial number of events receivable per poll */
177 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); 191 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
178 192
179 return EVBACKEND_EPOLL; 193 return EVBACKEND_EPOLL;
180} 194}
181 195
… …
189epoll_fork (EV_P) 203epoll_fork (EV_P)
190{ 204{
191 close (backend_fd); 205 close (backend_fd);
192 206
193 while ((backend_fd = epoll_create (256)) < 0) 207 while ((backend_fd = epoll_create (256)) < 0)
194 syserr ("(libev) epoll_create"); 208 ev_syserr ("(libev) epoll_create");
195 209
196 fcntl (backend_fd, F_SETFD, FD_CLOEXEC); 210 fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
197 211
198 fd_rearm_all (EV_A); 212 fd_rearm_all (EV_A);
199} 213}

Diff Legend

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