2 * libev event processing core, watcher management
4 * Copyright (c) 2007 Marc Alexander Lehmann <libev@schmorp.de>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
14 * * Redistributions in binary form must reproduce the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer in the documentation and/or other materials provided
17 * with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
46 #include <sys/types.h>
48 # include <sys/wait.h>
55 #ifndef EV_USE_MONOTONIC
56 # define EV_USE_MONOTONIC 1
60 # define EV_USE_SELECT 1
64 # define EV_USE_POLL 0 /* poll is usually slower than select, and not as well tested */
68 # define EV_USE_EPOLL 0
72 # define EV_USE_KQUEUE 0
75 #ifndef EV_USE_REALTIME
76 # define EV_USE_REALTIME 1
81 #ifndef CLOCK_MONOTONIC
82 # undef EV_USE_MONOTONIC
83 # define EV_USE_MONOTONIC 0
86 #ifndef CLOCK_REALTIME
87 # undef EV_USE_REALTIME
88 # define EV_USE_REALTIME 0
93 #define MIN_TIMEJUMP 1. /* minimum timejump that gets detected (if monotonic clock available) */
94 #define MAX_BLOCKTIME 59.731 /* never wait longer than this time (to detect time jumps) */
95 #define PID_HASHSIZE 16 /* size of pid hash table, must be power of two */
96 /*#define CLEANUP_INTERVAL 300. /* how often to try to free memory and re-check fds */
101 # define expect(expr,value) __builtin_expect ((expr),(value))
102 # define inline inline
104 # define expect(expr,value) (expr)
105 # define inline static
108 #define expect_false(expr) expect ((expr) != 0, 0)
109 #define expect_true(expr) expect ((expr) != 0, 1)
111 #define NUMPRI (EV_MAXPRI - EV_MINPRI + 1)
112 #define ABSPRI(w) ((w)->priority - EV_MINPRI)
114 typedef struct ev_watcher *W;
115 typedef struct ev_watcher_list *WL;
116 typedef struct ev_watcher_time *WT;
118 static ev_tstamp now_floor, now, diff; /* monotonic clock */
122 static int have_monotonic; /* runtime */
124 static ev_tstamp method_fudge; /* stupid epoll-returns-early bug */
125 static void (*method_modify)(int fd, int oev, int nev);
126 static void (*method_poll)(ev_tstamp timeout);
128 /*****************************************************************************/
135 clock_gettime (CLOCK_REALTIME, &ts);
136 return ts.tv_sec + ts.tv_nsec * 1e-9;
139 gettimeofday (&tv, 0);
140 return tv.tv_sec + tv.tv_usec * 1e-6;
148 if (expect_true (have_monotonic))
151 clock_gettime (CLOCK_MONOTONIC, &ts);
152 return ts.tv_sec + ts.tv_nsec * 1e-9;
159 #define array_roundsize(base,n) ((n) | 4 & ~3)
161 #define array_needsize(base,cur,cnt,init) \
162 if (expect_false ((cnt) > cur)) \
167 newcnt = array_roundsize (base, newcnt << 1); \
169 while ((cnt) > newcnt); \
171 base = realloc (base, sizeof (*base) * (newcnt)); \
172 init (base + cur, newcnt - cur); \
176 /*****************************************************************************/
180 struct ev_watcher_list *head;
181 unsigned char events;
189 anfds_init (ANFD *base, int count)
194 base->events = EV_NONE;
207 static ANPENDING *pendings [NUMPRI];
208 static int pendingmax [NUMPRI], pendingcnt [NUMPRI];
211 event (W w, int events)
215 pendings [ABSPRI (w)][w->pending - 1].events |= events;
219 w->pending = ++pendingcnt [ABSPRI (w)];
220 array_needsize (pendings [ABSPRI (w)], pendingmax [ABSPRI (w)], pendingcnt [ABSPRI (w)], );
221 pendings [ABSPRI (w)][w->pending - 1].w = w;
222 pendings [ABSPRI (w)][w->pending - 1].events = events;
226 queue_events (W *events, int eventcnt, int type)
230 for (i = 0; i < eventcnt; ++i)
231 event (events [i], type);
235 fd_event (int fd, int events)
237 ANFD *anfd = anfds + fd;
240 for (w = (struct ev_io *)anfd->head; w; w = (struct ev_io *)((WL)w)->next)
242 int ev = w->events & events;
249 /*****************************************************************************/
251 static int *fdchanges;
252 static int fdchangemax, fdchangecnt;
259 for (i = 0; i < fdchangecnt; ++i)
261 int fd = fdchanges [i];
262 ANFD *anfd = anfds + fd;
267 for (w = (struct ev_io *)anfd->head; w; w = (struct ev_io *)((WL)w)->next)
272 if (anfd->events != events)
274 method_modify (fd, anfd->events, events);
275 anfd->events = events;
285 if (anfds [fd].reify || fdchangecnt < 0)
288 anfds [fd].reify = 1;
291 array_needsize (fdchanges, fdchangemax, fdchangecnt, );
292 fdchanges [fdchangecnt - 1] = fd;
300 while ((w = (struct ev_io *)anfds [fd].head))
303 event ((W)w, EV_ERROR | EV_READ | EV_WRITE);
307 /* called on EBADF to verify fds */
313 for (fd = 0; fd < anfdmax; ++fd)
314 if (anfds [fd].events)
315 if (fcntl (fd, F_GETFD) == -1 && errno == EBADF)
319 /* called on ENOMEM in select/poll to kill some fds and retry */
326 if (anfds [fd].events)
334 /*****************************************************************************/
336 static struct ev_timer **timers;
337 static int timermax, timercnt;
339 static struct ev_periodic **periodics;
340 static int periodicmax, periodiccnt;
343 upheap (WT *timers, int k)
347 while (k && timers [k >> 1]->at > w->at)
349 timers [k] = timers [k >> 1];
350 timers [k]->active = k + 1;
355 timers [k]->active = k + 1;
360 downheap (WT *timers, int N, int k)
368 if (j + 1 < N && timers [j]->at > timers [j + 1]->at)
371 if (w->at <= timers [j]->at)
374 timers [k] = timers [j];
375 timers [k]->active = k + 1;
380 timers [k]->active = k + 1;
383 /*****************************************************************************/
387 struct ev_watcher_list *head;
388 sig_atomic_t volatile gotsig;
391 static ANSIG *signals;
392 static int signalmax;
394 static int sigpipe [2];
395 static sig_atomic_t volatile gotsig;
396 static struct ev_io sigev;
399 signals_init (ANSIG *base, int count)
411 sighandler (int signum)
413 signals [signum - 1].gotsig = 1;
417 int old_errno = errno;
419 write (sigpipe [1], &signum, 1);
425 sigcb (struct ev_io *iow, int revents)
427 struct ev_watcher_list *w;
430 read (sigpipe [0], &revents, 1);
433 for (signum = signalmax; signum--; )
434 if (signals [signum].gotsig)
436 signals [signum].gotsig = 0;
438 for (w = signals [signum].head; w; w = w->next)
439 event ((W)w, EV_SIGNAL);
447 fcntl (sigpipe [0], F_SETFD, FD_CLOEXEC);
448 fcntl (sigpipe [1], F_SETFD, FD_CLOEXEC);
450 /* rather than sort out wether we really need nb, set it */
451 fcntl (sigpipe [0], F_SETFL, O_NONBLOCK);
452 fcntl (sigpipe [1], F_SETFL, O_NONBLOCK);
455 ev_io_set (&sigev, sigpipe [0], EV_READ);
456 ev_io_start (&sigev);
459 /*****************************************************************************/
461 static struct ev_idle **idles;
462 static int idlemax, idlecnt;
464 static struct ev_prepare **prepares;
465 static int preparemax, preparecnt;
467 static struct ev_check **checks;
468 static int checkmax, checkcnt;
470 /*****************************************************************************/
472 static struct ev_child *childs [PID_HASHSIZE];
473 static struct ev_signal childev;
478 # define WCONTINUED 0
482 child_reap (struct ev_signal *sw, int chain, int pid, int status)
486 for (w = (struct ev_child *)childs [chain & (PID_HASHSIZE - 1)]; w; w = (struct ev_child *)((WL)w)->next)
487 if (w->pid == pid || !w->pid)
489 w->priority = sw->priority; /* need to do it *now* */
492 event ((W)w, EV_CHILD);
497 childcb (struct ev_signal *sw, int revents)
501 if (0 < (pid = waitpid (-1, &status, WNOHANG | WUNTRACED | WCONTINUED)))
503 /* make sure we are called again until all childs have been reaped */
504 event ((W)sw, EV_SIGNAL);
506 child_reap (sw, pid, pid, status);
507 child_reap (sw, 0, pid, status); /* this might trigger a watcher twice, but event catches that */
513 /*****************************************************************************/
516 # include "ev_kqueue.c"
519 # include "ev_epoll.c"
522 # include "ev_poll.c"
525 # include "ev_select.c"
529 ev_version_major (void)
531 return EV_VERSION_MAJOR;
535 ev_version_minor (void)
537 return EV_VERSION_MINOR;
540 /* return true if we are running with elevated privileges and should ignore env variables */
547 return getuid () != geteuid ()
548 || getgid () != getegid ();
552 int ev_init (int methods)
559 if (!clock_gettime (CLOCK_MONOTONIC, &ts))
572 if (methods == EVMETHOD_AUTO)
573 if (!enable_secure () && getenv ("LIBEV_METHODS"))
574 methods = atoi (getenv ("LIBEV_METHODS"));
576 methods = EVMETHOD_ANY;
580 if (!ev_method && (methods & EVMETHOD_KQUEUE)) kqueue_init (methods);
583 if (!ev_method && (methods & EVMETHOD_EPOLL )) epoll_init (methods);
586 if (!ev_method && (methods & EVMETHOD_POLL )) poll_init (methods);
589 if (!ev_method && (methods & EVMETHOD_SELECT)) select_init (methods);
594 ev_watcher_init (&sigev, sigcb);
595 ev_set_priority (&sigev, EV_MAXPRI);
599 ev_signal_init (&childev, childcb, SIGCHLD);
600 ev_set_priority (&childev, EV_MAXPRI);
601 ev_signal_start (&childev);
609 /*****************************************************************************/
612 ev_fork_prepare (void)
618 ev_fork_parent (void)
627 if (ev_method == EVMETHOD_EPOLL)
628 epoll_postfork_child ();
638 /*****************************************************************************/
645 for (pri = NUMPRI; pri--; )
646 while (pendingcnt [pri])
648 ANPENDING *p = pendings [pri] + --pendingcnt [pri];
653 p->w->cb (p->w, p->events);
661 while (timercnt && timers [0]->at <= now)
663 struct ev_timer *w = timers [0];
665 /* first reschedule or stop timer */
668 assert (("negative ev_timer repeat value found while processing timers", w->repeat > 0.));
669 w->at = now + w->repeat;
670 downheap ((WT *)timers, timercnt, 0);
673 ev_timer_stop (w); /* nonrepeating: stop timer */
675 event ((W)w, EV_TIMEOUT);
680 periodics_reify (void)
682 while (periodiccnt && periodics [0]->at <= ev_now)
684 struct ev_periodic *w = periodics [0];
686 /* first reschedule or stop timer */
689 w->at += floor ((ev_now - w->at) / w->interval + 1.) * w->interval;
690 assert (("ev_periodic timeout in the past detected while processing timers, negative interval?", w->at > ev_now));
691 downheap ((WT *)periodics, periodiccnt, 0);
694 ev_periodic_stop (w); /* nonrepeating: stop timer */
696 event ((W)w, EV_PERIODIC);
701 periodics_reschedule (ev_tstamp diff)
705 /* adjust periodics after time jump */
706 for (i = 0; i < periodiccnt; ++i)
708 struct ev_periodic *w = periodics [i];
712 ev_tstamp diff = ceil ((ev_now - w->at) / w->interval) * w->interval;
714 if (fabs (diff) >= 1e-4)
716 ev_periodic_stop (w);
717 ev_periodic_start (w);
719 i = 0; /* restart loop, inefficient, but time jumps should be rare */
726 time_update_monotonic (void)
730 if (expect_true (now - now_floor < MIN_TIMEJUMP * .5))
749 if (expect_true (have_monotonic))
751 if (time_update_monotonic ())
753 ev_tstamp odiff = diff;
755 for (i = 4; --i; ) /* loop a few times, before making important decisions */
759 if (fabs (odiff - diff) < MIN_TIMEJUMP)
760 return; /* all is well */
767 periodics_reschedule (diff - odiff);
768 /* no timer adjustment, as the monotonic clock doesn't jump */
776 if (expect_false (now > ev_now || now < ev_now - MAX_BLOCKTIME - MIN_TIMEJUMP))
778 periodics_reschedule (ev_now - now);
780 /* adjust timers. this is easy, as the offset is the same for all */
781 for (i = 0; i < timercnt; ++i)
782 timers [i]->at += diff;
791 void ev_loop (int flags)
794 ev_loop_done = flags & (EVLOOP_ONESHOT | EVLOOP_NONBLOCK) ? 1 : 0;
798 /* queue check watchers (and execute them) */
799 if (expect_false (preparecnt))
801 queue_events ((W *)prepares, preparecnt, EV_PREPARE);
805 /* update fd-related kernel structures */
808 /* calculate blocking time */
810 /* we only need this for !monotonic clockor timers, but as we basically
811 always have timers, we just calculate it always */
813 if (expect_true (have_monotonic))
814 time_update_monotonic ();
822 if (flags & EVLOOP_NONBLOCK || idlecnt)
826 block = MAX_BLOCKTIME;
830 ev_tstamp to = timers [0]->at - now + method_fudge;
831 if (block > to) block = to;
836 ev_tstamp to = periodics [0]->at - ev_now + method_fudge;
837 if (block > to) block = to;
840 if (block < 0.) block = 0.;
845 /* update ev_now, do magic */
848 /* queue pending timers and reschedule them */
849 timers_reify (); /* relative timers called last */
850 periodics_reify (); /* absolute timers called first */
852 /* queue idle watchers unless io or timers are pending */
854 queue_events ((W *)idles, idlecnt, EV_IDLE);
856 /* queue check watchers, to be executed first */
858 queue_events ((W *)checks, checkcnt, EV_CHECK);
862 while (!ev_loop_done);
864 if (ev_loop_done != 2)
868 /*****************************************************************************/
871 wlist_add (WL *head, WL elem)
878 wlist_del (WL *head, WL elem)
888 head = &(*head)->next;
893 ev_clear_pending (W w)
897 pendings [ABSPRI (w)][w->pending - 1].w = 0;
903 ev_start (W w, int active)
905 if (w->priority < EV_MINPRI) w->priority = EV_MINPRI;
906 if (w->priority > EV_MAXPRI) w->priority = EV_MAXPRI;
917 /*****************************************************************************/
920 ev_io_start (struct ev_io *w)
924 if (ev_is_active (w))
927 assert (("ev_io_start called with negative fd", fd >= 0));
930 array_needsize (anfds, anfdmax, fd + 1, anfds_init);
931 wlist_add ((WL *)&anfds[fd].head, (WL)w);
937 ev_io_stop (struct ev_io *w)
939 ev_clear_pending ((W)w);
940 if (!ev_is_active (w))
943 wlist_del ((WL *)&anfds[w->fd].head, (WL)w);
950 ev_timer_start (struct ev_timer *w)
952 if (ev_is_active (w))
957 assert (("ev_timer_start called with negative timer repeat value", w->repeat >= 0.));
959 ev_start ((W)w, ++timercnt);
960 array_needsize (timers, timermax, timercnt, );
961 timers [timercnt - 1] = w;
962 upheap ((WT *)timers, timercnt - 1);
966 ev_timer_stop (struct ev_timer *w)
968 ev_clear_pending ((W)w);
969 if (!ev_is_active (w))
972 if (w->active < timercnt--)
974 timers [w->active - 1] = timers [timercnt];
975 downheap ((WT *)timers, timercnt, w->active - 1);
984 ev_timer_again (struct ev_timer *w)
986 if (ev_is_active (w))
990 w->at = now + w->repeat;
991 downheap ((WT *)timers, timercnt, w->active - 1);
1001 ev_periodic_start (struct ev_periodic *w)
1003 if (ev_is_active (w))
1006 assert (("ev_periodic_start called with negative interval value", w->interval >= 0.));
1008 /* this formula differs from the one in periodic_reify because we do not always round up */
1010 w->at += ceil ((ev_now - w->at) / w->interval) * w->interval;
1012 ev_start ((W)w, ++periodiccnt);
1013 array_needsize (periodics, periodicmax, periodiccnt, );
1014 periodics [periodiccnt - 1] = w;
1015 upheap ((WT *)periodics, periodiccnt - 1);
1019 ev_periodic_stop (struct ev_periodic *w)
1021 ev_clear_pending ((W)w);
1022 if (!ev_is_active (w))
1025 if (w->active < periodiccnt--)
1027 periodics [w->active - 1] = periodics [periodiccnt];
1028 downheap ((WT *)periodics, periodiccnt, w->active - 1);
1035 # define SA_RESTART 0
1039 ev_signal_start (struct ev_signal *w)
1041 if (ev_is_active (w))
1044 assert (("ev_signal_start called with illegal signal number", w->signum > 0));
1047 array_needsize (signals, signalmax, w->signum, signals_init);
1048 wlist_add ((WL *)&signals [w->signum - 1].head, (WL)w);
1052 struct sigaction sa;
1053 sa.sa_handler = sighandler;
1054 sigfillset (&sa.sa_mask);
1055 sa.sa_flags = SA_RESTART; /* if restarting works we save one iteration */
1056 sigaction (w->signum, &sa, 0);
1061 ev_signal_stop (struct ev_signal *w)
1063 ev_clear_pending ((W)w);
1064 if (!ev_is_active (w))
1067 wlist_del ((WL *)&signals [w->signum - 1].head, (WL)w);
1070 if (!signals [w->signum - 1].head)
1071 signal (w->signum, SIG_DFL);
1075 ev_idle_start (struct ev_idle *w)
1077 if (ev_is_active (w))
1080 ev_start ((W)w, ++idlecnt);
1081 array_needsize (idles, idlemax, idlecnt, );
1082 idles [idlecnt - 1] = w;
1086 ev_idle_stop (struct ev_idle *w)
1088 ev_clear_pending ((W)w);
1089 if (ev_is_active (w))
1092 idles [w->active - 1] = idles [--idlecnt];
1097 ev_prepare_start (struct ev_prepare *w)
1099 if (ev_is_active (w))
1102 ev_start ((W)w, ++preparecnt);
1103 array_needsize (prepares, preparemax, preparecnt, );
1104 prepares [preparecnt - 1] = w;
1108 ev_prepare_stop (struct ev_prepare *w)
1110 ev_clear_pending ((W)w);
1111 if (ev_is_active (w))
1114 prepares [w->active - 1] = prepares [--preparecnt];
1119 ev_check_start (struct ev_check *w)
1121 if (ev_is_active (w))
1124 ev_start ((W)w, ++checkcnt);
1125 array_needsize (checks, checkmax, checkcnt, );
1126 checks [checkcnt - 1] = w;
1130 ev_check_stop (struct ev_check *w)
1132 ev_clear_pending ((W)w);
1133 if (ev_is_active (w))
1136 checks [w->active - 1] = checks [--checkcnt];
1141 ev_child_start (struct ev_child *w)
1143 if (ev_is_active (w))
1147 wlist_add ((WL *)&childs [w->pid & (PID_HASHSIZE - 1)], (WL)w);
1151 ev_child_stop (struct ev_child *w)
1153 ev_clear_pending ((W)w);
1154 if (ev_is_active (w))
1157 wlist_del ((WL *)&childs [w->pid & (PID_HASHSIZE - 1)], (WL)w);
1161 /*****************************************************************************/
1167 void (*cb)(int revents, void *arg);
1172 once_cb (struct ev_once *once, int revents)
1174 void (*cb)(int revents, void *arg) = once->cb;
1175 void *arg = once->arg;
1177 ev_io_stop (&once->io);
1178 ev_timer_stop (&once->to);
1185 once_cb_io (struct ev_io *w, int revents)
1187 once_cb ((struct ev_once *)(((char *)w) - offsetof (struct ev_once, io)), revents);
1191 once_cb_to (struct ev_timer *w, int revents)
1193 once_cb ((struct ev_once *)(((char *)w) - offsetof (struct ev_once, to)), revents);
1197 ev_once (int fd, int events, ev_tstamp timeout, void (*cb)(int revents, void *arg), void *arg)
1199 struct ev_once *once = malloc (sizeof (struct ev_once));
1202 cb (EV_ERROR | EV_READ | EV_WRITE | EV_TIMEOUT, arg);
1208 ev_watcher_init (&once->io, once_cb_io);
1211 ev_io_set (&once->io, fd, events);
1212 ev_io_start (&once->io);
1215 ev_watcher_init (&once->to, once_cb_to);
1218 ev_timer_set (&once->to, timeout, 0.);
1219 ev_timer_start (&once->to);
1224 /*****************************************************************************/
1231 sin_cb (struct ev_io *w, int revents)
1233 fprintf (stderr, "sin %d, revents %d\n", w->fd, revents);
1237 ocb (struct ev_timer *w, int revents)
1239 //fprintf (stderr, "timer %f,%f (%x) (%f) d%p\n", w->at, w->repeat, revents, w->at - ev_time (), w->data);
1245 scb (struct ev_signal *w, int revents)
1247 fprintf (stderr, "signal %x,%d\n", revents, w->signum);
1253 gcb (struct ev_signal *w, int revents)
1255 fprintf (stderr, "generic %x\n", revents);
1263 ev_io_init (&wio, sin_cb, 0, EV_READ);
1266 struct ev_timer t[10000];
1270 for (i = 0; i < 10000; ++i)
1272 struct ev_timer *w = t + i;
1273 ev_watcher_init (w, ocb, i);
1274 ev_timer_init_abs (w, ocb, drand48 (), 0.99775533);
1276 if (drand48 () < 0.5)
1282 ev_timer_init (&t1, ocb, 5, 10);
1283 ev_timer_start (&t1);
1285 struct ev_signal sig;
1286 ev_signal_init (&sig, scb, SIGQUIT);
1287 ev_signal_start (&sig);
1290 ev_check_init (&cw, gcb);
1291 ev_check_start (&cw);
1294 ev_idle_init (&iw, gcb);
1295 ev_idle_start (&iw);