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 1.50 by root, Tue Mar 23 23:45:12 2010 UTC vs.
Revision 1.57 by root, Sat Jan 8 10:25:16 2011 UTC

1/* 1/*
2 * libev epoll fd activity backend 2 * libev epoll fd activity backend
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libev@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 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 *
… …
51 * poll replacement. 51 * poll replacement.
52 * d) epoll doesn't work the same as select with many file descriptors 52 * d) epoll doesn't work the same as select with many file descriptors
53 * (such as files). while not critical, no other advanced interface 53 * (such as files). while not critical, no other advanced interface
54 * seems to share this (rather non-unixy) limitation. 54 * seems to share this (rather non-unixy) limitation.
55 * e) epoll claims to be embeddable, but in practise you never get 55 * e) epoll claims to be embeddable, but in practise you never get
56 * a ready event for the epoll fd. 56 * a ready event for the epoll fd (broken: <=2.6.26, working: >=2.6.32).
57 * f) epoll_ctl returning EPERM means the fd is always ready.
57 * 58 *
58 * lots of "weird code" and complication handling in this file is due 59 * lots of "weird code" and complication handling in this file is due
59 * to these design problems with epoll, as we try very hard to avoid 60 * to these design problems with epoll, as we try very hard to avoid
60 * epoll_ctl syscalls for common usage patterns and handle the breakage 61 * epoll_ctl syscalls for common usage patterns and handle the breakage
61 * ensuing from receiving events for closed and otherwise long gone 62 * ensuing from receiving events for closed and otherwise long gone
62 * file descriptors. 63 * file descriptors.
63 */ 64 */
64 65
65#include <sys/epoll.h> 66#include <sys/epoll.h>
67
68#define EV_EMASK_EPERM 0x80
66 69
67static void 70static void
68epoll_modify (EV_P_ int fd, int oev, int nev) 71epoll_modify (EV_P_ int fd, int oev, int nev)
69{ 72{
70 struct epoll_event ev; 73 struct epoll_event ev;
… …
88 ev.data.u64 = (uint64_t)(uint32_t)fd 91 ev.data.u64 = (uint64_t)(uint32_t)fd
89 | ((uint64_t)(uint32_t)++anfds [fd].egen << 32); 92 | ((uint64_t)(uint32_t)++anfds [fd].egen << 32);
90 ev.events = (nev & EV_READ ? EPOLLIN : 0) 93 ev.events = (nev & EV_READ ? EPOLLIN : 0)
91 | (nev & EV_WRITE ? EPOLLOUT : 0); 94 | (nev & EV_WRITE ? EPOLLOUT : 0);
92 95
93 if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev))) 96 if (expect_true (!epoll_ctl (backend_fd, oev && oldmask != nev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
94 return; 97 return;
95 98
96 if (expect_true (errno == ENOENT)) 99 if (expect_true (errno == ENOENT))
97 { 100 {
98 /* 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 */
… …
109 if (oldmask == nev) 112 if (oldmask == nev)
110 goto dec_egen; 113 goto dec_egen;
111 114
112 if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev)) 115 if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
113 return; 116 return;
117 }
118 else if (expect_true (errno == EPERM))
119 {
120 /* EPERM means the fd is always ready, but epoll is too snobbish */
121 /* to handle it, unlike select or poll. */
122 anfds [fd].emask = EV_EMASK_EPERM;
123
124 /* add fd to epoll_eperms, if not already inside */
125 if (!(oldmask & EV_EMASK_EPERM))
126 {
127 array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, EMPTY2);
128 epoll_eperms [epoll_epermcnt++] = fd;
129 }
130
131 return;
114 } 132 }
115 133
116 fd_kill (EV_A_ fd); 134 fd_kill (EV_A_ fd);
117 135
118dec_egen: 136dec_egen:
… …
148 int want = anfds [fd].events; 166 int want = anfds [fd].events;
149 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) 167 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
150 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0); 168 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0);
151 169
152 /* check for spurious notification */ 170 /* check for spurious notification */
171 /* we assume that fd is always in range, as we never shrink the anfds array */
153 if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32))) 172 if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32)))
154 { 173 {
155 /* recreate kernel state */ 174 /* recreate kernel state */
156 postfork = 1; 175 postfork = 1;
157 continue; 176 continue;
… …
168 187
169 /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */ 188 /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */
170 /* which is fortunately easy to do for us. */ 189 /* which is fortunately easy to do for us. */
171 if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev)) 190 if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev))
172 { 191 {
173 postfork = 1; /* an error occured, recreate kernel state */ 192 postfork = 1; /* an error occurred, recreate kernel state */
174 continue; 193 continue;
175 } 194 }
176 } 195 }
177 196
178 fd_event (EV_A_ fd, got); 197 fd_event (EV_A_ fd, got);
… …
183 { 202 {
184 ev_free (epoll_events); 203 ev_free (epoll_events);
185 epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1); 204 epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1);
186 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax); 205 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
187 } 206 }
207
208 /* now synthesize events for all fds where epoll fails, while select works... */
209 for (i = epoll_epermcnt; i--; )
210 {
211 int fd = epoll_eperms [i];
212 unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE);
213
214 if (anfds [fd].emask & EV_EMASK_EPERM && events)
215 fd_event (EV_A_ fd, events);
216 else
217 epoll_eperms [i] = epoll_eperms [--epoll_epermcnt];
218 }
188} 219}
189 220
190int inline_size 221int inline_size
191epoll_init (EV_P_ int flags) 222epoll_init (EV_P_ int flags)
192{ 223{
… …
214 245
215void inline_size 246void inline_size
216epoll_destroy (EV_P) 247epoll_destroy (EV_P)
217{ 248{
218 ev_free (epoll_events); 249 ev_free (epoll_events);
250 array_free (epoll_eperm, EMPTY);
219} 251}
220 252
221void inline_size 253void inline_size
222epoll_fork (EV_P) 254epoll_fork (EV_P)
223{ 255{

Diff Legend

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