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 1.5 by root, Wed Oct 31 14:44:15 2007 UTC vs.
Revision 1.32 by root, Thu Dec 13 16:52:50 2007 UTC

1/* 1/*
2 * libev epoll fd activity backend
3 *
2 * Copyright (c) 2007 Marc Alexander Lehmann <libev@schmorp.de> 4 * Copyright (c) 2007 Marc Alexander Lehmann <libev@schmorp.de>
3 * All rights reserved. 5 * All rights reserved.
4 * 6 *
5 * Redistribution and use in source and binary forms, with or without 7 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are 8 * modification, are permitted provided that the following conditions are
… …
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 28 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 29 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */ 30 */
29 31
32/*
33 * general notes about epoll:
34 *
35 * a) epoll silently removes fds from the fd set. as nothing tells us
36 * that an fd has been removed otherwise, we have to continually
37 * "rearm" fds that we suspect *might* have changed (same
38 * problem with kqueue, but much less costly there).
39 * b) the fact that ADD != MOD creates a lot of extra syscalls due to a)
40 * and seems not to have any advantage.
41 * c) the inability to handle fork or file descriptors (think dup)
42 * limits the applicability over poll, so this is not a generic
43 * poll replacement.
44 *
45 * lots of "weird code" and complication handling in this file is due
46 * to these design problems with epoll, as we try very hard to avoid
47 * epoll_ctl syscalls for common usage patterns.
48 */
49
30#include <sys/epoll.h> 50#include <sys/epoll.h>
31 51
32static int epoll_fd = -1; 52static void
53epoll_modify (EV_P_ int fd, int oev, int nev)
54{
55 struct epoll_event ev;
56
57 /*
58 * we handle EPOLL_CTL_DEL by ignoring it here
59 * on the assumption that the fd is gone anyways
60 * if that is wrong, we have to handle the spurious
61 * event in epoll_poll.
62 */
63 if (!nev)
64 return;
65
66 ev.data.u64 = fd; /* use u64 to fully initialise the struct, for nicer strace etc. */
67 ev.events = (nev & EV_READ ? EPOLLIN : 0)
68 | (nev & EV_WRITE ? EPOLLOUT : 0);
69
70 if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
71 return;
72
73 if (expect_true (errno == ENOENT))
74 {
75 /* on ENOENT the fd went away, so try to do the right thing */
76 if (!nev)
77 return;
78
79 if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
80 return;
81 }
82 else if (expect_true (errno == EEXIST))
83 {
84 /* on EEXIST we ignored a previous DEL */
85 if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
86 return;
87 }
88
89 fd_kill (EV_A_ fd);
90}
33 91
34static void 92static void
35epoll_modify (int fd, int oev, int nev) 93epoll_poll (EV_P_ ev_tstamp timeout)
36{ 94{
37 int mode = nev ? oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD : EPOLL_CTL_DEL; 95 int i;
96 int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
38 97
39 struct epoll_event ev; 98 if (expect_false (eventcnt < 0))
40 ev.data.fd = fd; 99 {
41 ev.events = 100 if (errno != EINTR)
42 (nev & EV_READ ? EPOLLIN : 0) 101 syserr ("(libev) epoll_wait");
43 | (nev & EV_WRITE ? EPOLLOUT : 0);
44 102
45 epoll_ctl (epoll_fd, mode, fd, &ev);
46}
47
48void epoll_postfork_child (void)
49{
50 int fd;
51
52 epoll_fd = epoll_create (256);
53 fcntl (epoll_fd, F_SETFD, FD_CLOEXEC);
54
55 /* re-register interest in fds */
56 for (fd = 0; fd < anfdmax; ++fd)
57 if (anfds [fd].wev)
58 epoll_modify (fd, EV_NONE, anfds [fd].wev);
59}
60
61static struct epoll_event *events;
62static int eventmax;
63
64static void epoll_poll (ev_tstamp timeout)
65{
66 int eventcnt = epoll_wait (epoll_fd, events, eventmax, ceil (timeout * 1000.));
67 int i;
68
69 if (eventcnt < 0)
70 return; 103 return;
104 }
71 105
72 for (i = 0; i < eventcnt; ++i) 106 for (i = 0; i < eventcnt; ++i)
73 fd_event ( 107 {
74 events [i].data.fd, 108 struct epoll_event *ev = epoll_events + i;
109
110 int fd = ev->data.u64;
75 (events [i].events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0) 111 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
76 | (events [i].events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0) 112 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0);
77 ); 113 int want = anfds [fd].events;
114
115 if (expect_false (got & ~want))
116 {
117 /* we received an event but are not interested in it, try mod or del */
118 ev->events = (want & EV_READ ? EPOLLIN : 0)
119 | (want & EV_WRITE ? EPOLLOUT : 0);
120
121 epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev);
122 }
123
124 fd_event (EV_A_ fd, got);
125 }
78 126
79 /* if the receive array was full, increase its size */ 127 /* if the receive array was full, increase its size */
80 if (eventcnt == eventmax) 128 if (expect_false (eventcnt == epoll_eventmax))
81 { 129 {
82 free (events); 130 ev_free (epoll_events);
83 eventmax += eventmax >> 1; 131 epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1);
84 events = malloc (sizeof (struct epoll_event) * eventmax); 132 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
85 } 133 }
86} 134}
87 135
136int inline_size
88void epoll_init (int flags) 137epoll_init (EV_P_ int flags)
89{ 138{
90 epoll_fd = epoll_create (256); 139 backend_fd = epoll_create (256);
91 140
92 if (epoll_fd < 0) 141 if (backend_fd < 0)
93 return; 142 return 0;
94 143
95 fcntl (epoll_fd, F_SETFD, FD_CLOEXEC); 144 fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
96 145
97 ev_method = EVMETHOD_EPOLL; 146 backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */
98 method_fudge = 1e-3; /* needed to compensate for epoll returning early */
99 method_modify = epoll_modify; 147 backend_modify = epoll_modify;
100 method_poll = epoll_poll; 148 backend_poll = epoll_poll;
101 149
102 eventmax = 64; /* intiial number of events receivable per poll */ 150 epoll_eventmax = 64; /* intiial number of events receivable per poll */
103 events = malloc (sizeof (struct epoll_event) * eventmax); 151 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
152
153 return EVBACKEND_EPOLL;
104} 154}
155
156void inline_size
157epoll_destroy (EV_P)
158{
159 ev_free (epoll_events);
160}
161
162void inline_size
163epoll_fork (EV_P)
164{
165 close (backend_fd);
166
167 while ((backend_fd = epoll_create (256)) < 0)
168 syserr ("(libev) epoll_create");
169
170 fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
171
172 fd_rearm_all (EV_A);
173}
174

Diff Legend

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