2 * IRC - Internet Relay Chat, ircd/engine_kqueue.c
3 * Copyright (C) 2001 Kevin L. Mitchell <klmitch@mit.edu>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 1, or (at your option)
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include "ircd_events.h"
26 #include "ircd_alloc.h"
27 #include "ircd_features.h"
34 #include <sys/event.h>
35 #include <sys/socket.h>
37 #include <sys/types.h>
41 #define KQUEUE_ERROR_THRESHOLD 20 /* after 20 kqueue errors, restart */
42 #define ERROR_EXPIRE_TIME 3600 /* expire errors after an hour */
44 static struct Socket** sockList;
45 static int kqueue_max;
48 static int errors = 0;
49 static struct Timer clear_error;
51 /* decrements the error count once per hour */
53 error_clear(struct Event* ev)
55 if (!--errors) /* remove timer when error count reaches 0 */
56 timer_del(ev_timer(ev));
59 /* initialize the kqueue engine */
61 engine_init(int max_sockets)
65 if ((kqueue_id = kqueue()) < 0) { /* initialize... */
66 log_write(LS_SYSTEM, L_WARNING, 0,
67 "kqueue() engine cannot initialize: %m");
71 /* allocate necessary memory */
72 sockList = (struct Socket**) MyMalloc(sizeof(struct Socket*) * max_sockets);
74 /* initialize the data */
75 for (i = 0; i < max_sockets; i++)
78 kqueue_max = max_sockets; /* number of sockets allocated */
80 return 1; /* success! */
83 /* add a signel to be watched for */
85 engine_signal(struct Signal* sig)
87 struct kevent sigevent;
92 Debug((DEBUG_ENGINE, "kqueue: Adding filter for signal %d [%p]",
93 sig_signal(sig), sig));
95 sigevent.ident = sig_signal(sig); /* set up the kqueue event */
96 sigevent.filter = EVFILT_SIGNAL; /* looking for signals... */
97 sigevent.flags = EV_ADD | EV_ENABLE; /* add and enable it */
100 sigevent.udata = sig; /* store our user data */
102 if (kevent(kqueue_id, &sigevent, 1, 0, 0, 0) < 0) { /* add event */
103 log_write(LS_SYSTEM, L_WARNING, 0, "Unable to trap signal %d",
108 act.sa_handler = SIG_IGN; /* ignore the signal */
110 sigemptyset(&act.sa_mask);
111 sigaction(sig_signal(sig), &act, 0);
114 /* Figure out what events go with a given state */
116 state_to_events(enum SocketState state, unsigned int events)
119 case SS_CONNECTING: /* connecting socket */
120 return SOCK_EVENT_WRITABLE;
123 case SS_LISTENING: /* listening socket */
124 case SS_NOTSOCK: /* our signal socket--just in case */
125 return SOCK_EVENT_READABLE;
128 case SS_CONNECTED: case SS_DATAGRAM: case SS_CONNECTDG:
129 return events; /* ordinary socket */
137 /* Activate kqueue filters as appropriate */
139 set_or_clear(struct Socket* sock, unsigned int clear, unsigned int set)
142 struct kevent chglist[2];
145 assert(-1 < s_fd(sock));
147 if ((clear ^ set) & SOCK_EVENT_READABLE) { /* readable has changed */
148 chglist[i].ident = s_fd(sock); /* set up the change list */
149 chglist[i].filter = EVFILT_READ; /* readable filter */
150 chglist[i].flags = EV_ADD; /* adding it */
151 chglist[i].fflags = 0;
153 chglist[i].udata = 0; /* I love udata, but it can't really be used here */
155 if (set & SOCK_EVENT_READABLE) /* it's set */
156 chglist[i].flags |= EV_ENABLE;
158 chglist[i].flags |= EV_DISABLE;
160 i++; /* advance to next element */
163 if ((clear ^ set) & SOCK_EVENT_WRITABLE) { /* writable has changed */
164 chglist[i].ident = s_fd(sock); /* set up the change list */
165 chglist[i].filter = EVFILT_WRITE; /* writable filter */
166 chglist[i].flags = EV_ADD; /* adding it */
167 chglist[i].fflags = 0;
169 chglist[i].udata = 0;
171 if (set & SOCK_EVENT_WRITABLE) /* it's set */
172 chglist[i].flags |= EV_ENABLE;
174 chglist[i].flags |= EV_DISABLE;
176 i++; /* advance count... */
179 if (kevent(kqueue_id, chglist, i, 0, 0, 0) < 0 && errno != EBADF)
180 event_generate(ET_ERROR, sock, errno); /* report error */
183 /* add a socket to be listened on */
185 engine_add(struct Socket* sock)
188 assert(0 == sockList[s_fd(sock)]);
190 /* bounds-check... */
191 if (sock->s_fd >= kqueue_max) {
192 log_write(LS_SYSTEM, L_ERROR, 0,
193 "Attempt to add socket %d (> %d) to event engine", s_fd(sock),
198 sockList[s_fd(sock)] = sock; /* add to list */
200 Debug((DEBUG_ENGINE, "kqueue: Adding socket %d [%p], state %s, to engine",
201 s_fd(sock), sock, state_to_name(s_state(sock))));
203 /* Add socket to queue */
204 set_or_clear(sock, 0, state_to_events(s_state(sock), s_events(sock)));
206 return 1; /* success */
209 /* socket switching to new state */
211 engine_state(struct Socket* sock, enum SocketState new_state)
214 assert(sock == sockList[s_fd(sock)]);
216 Debug((DEBUG_ENGINE, "kqueue: Changing state for socket %p to %s", sock,
217 state_to_name(new_state)));
219 /* set the correct events */
221 state_to_events(s_state(sock), s_events(sock)), /* old state */
222 state_to_events(new_state, s_events(sock))); /* new state */
226 /* socket events changing */
228 engine_events(struct Socket* sock, unsigned int new_events)
231 assert(sock == sockList[s_fd(sock)]);
233 Debug((DEBUG_ENGINE, "kqueue: Changing event mask for socket %p to [%s]",
234 sock, sock_flags(new_events)));
236 /* set the correct events */
238 state_to_events(s_state(sock), s_events(sock)), /* old events */
239 state_to_events(s_state(sock), new_events)); /* new events */
242 /* socket going away */
244 engine_delete(struct Socket* sock)
246 struct kevent dellist[2];
249 assert(sock == sockList[s_fd(sock)]);
251 Debug((DEBUG_ENGINE, "kqueue: Deleting socket %d [%p], state %s",
252 s_fd(sock), sock, state_to_name(s_state(sock))));
254 dellist[0].ident = s_fd(sock); /* set up the delete list */
255 dellist[0].filter = EVFILT_READ; /* readable filter */
256 dellist[0].flags = EV_DELETE; /* delete it */
257 dellist[0].fflags = 0;
259 dellist[0].udata = 0;
261 dellist[1].ident = s_fd(sock);
262 dellist[1].filter = EVFILT_WRITE; /* writable filter */
263 dellist[1].flags = EV_DELETE; /* delete it */
264 dellist[1].fflags = 0;
266 dellist[1].udata = 0;
268 /* make it all go away */
269 if (kevent(kqueue_id, dellist, 2, 0, 0, 0) < 0)
270 log_write(LS_SOCKET, L_WARNING, 0,
271 "Unable to delete kevent items for socket %d", s_fd(sock));
273 sockList[s_fd(sock)] = 0;
276 /* engine event loop */
278 engine_loop(struct Generators* gen)
280 struct kevent *events;
283 struct timespec wait;
289 if ((events_count = feature_int(FEAT_POLLS_PER_LOOP)) < 20)
291 events = (struct kevent *)MyMalloc(sizeof(struct kevent) * events_count);
294 if ((i = feature_int(FEAT_POLLS_PER_LOOP)) >= 20 && i != events_count) {
295 events = (struct kevent *)MyRealloc(events, sizeof(struct kevent) * i);
299 /* set up the sleep time */
300 wait.tv_sec = timer_next(gen) ? (timer_next(gen) - CurrentTime) : -1;
303 Debug((DEBUG_INFO, "kqueue: delay: %Tu (%Tu) %Tu", timer_next(gen),
304 CurrentTime, wait.tv_sec));
306 /* check for active events */
307 nevs = kevent(kqueue_id, 0, 0, events, events_count,
308 wait.tv_sec < 0 ? 0 : &wait);
310 CurrentTime = time(0); /* set current time... */
313 if (errno != EINTR) { /* ignore kevent interrupts */
314 /* Log the kqueue error */
315 log_write(LS_SOCKET, L_ERROR, 0, "kevent() error: %m");
317 timer_add(timer_init(&clear_error), error_clear, 0, TT_PERIODIC,
319 else if (errors > KQUEUE_ERROR_THRESHOLD) /* too many errors... */
320 server_restart("too many kevent errors");
322 /* old code did a sleep(1) here; with usage these days,
323 * that may be too expensive
328 for (i = 0; i < nevs; i++) {
329 if (events[i].filter == EVFILT_SIGNAL) {
330 /* it's a signal; deal appropriately */
331 event_generate(ET_SIGNAL, events[i].udata, events[i].ident);
332 continue; /* skip socket processing loop */
335 assert(events[i].filter == EVFILT_READ ||
336 events[i].filter == EVFILT_WRITE);
338 sock = sockList[events[i].ident];
339 if (!sock) /* slots may become empty while processing events */
342 assert(s_fd(sock) == events[i].ident);
344 gen_ref_inc(sock); /* can't have it going away on us */
346 Debug((DEBUG_ENGINE, "kqueue: Checking socket %p (fd %d) state %s, "
347 "events %s", sock, s_fd(sock), state_to_name(s_state(sock)),
348 sock_flags(s_events(sock))));
350 if (s_state(sock) != SS_NOTSOCK) {
351 errcode = 0; /* check for errors on socket */
352 codesize = sizeof(errcode);
353 if (getsockopt(s_fd(sock), SOL_SOCKET, SO_ERROR, &errcode,
355 errcode = errno; /* work around Solaris implementation */
357 if (errcode) { /* an error occurred; generate an event */
358 Debug((DEBUG_ENGINE, "kqueue: Error %d on fd %d, socket %p", errcode,
360 event_generate(ET_ERROR, sock, errcode);
361 gen_ref_dec(sock); /* careful not to leak reference counts */
366 switch (s_state(sock)) {
368 if (events[i].filter == EVFILT_WRITE) { /* connection completed */
369 Debug((DEBUG_ENGINE, "kqueue: Connection completed"));
370 event_generate(ET_CONNECT, sock, 0);
375 if (events[i].filter == EVFILT_READ) { /* connect. to be accept. */
376 Debug((DEBUG_ENGINE, "kqueue: Ready for accept"));
377 event_generate(ET_ACCEPT, sock, 0);
381 case SS_NOTSOCK: /* doing nothing socket-specific */
383 if (events[i].filter == EVFILT_READ) { /* data on socket */
384 Debug((DEBUG_ENGINE, "kqueue: EOF or data to be read"));
385 event_generate(events[i].flags & EV_EOF ? ET_EOF : ET_READ, sock, 0);
387 if (events[i].filter == EVFILT_WRITE) { /* socket writable */
388 Debug((DEBUG_ENGINE, "kqueue: Data can be written"));
389 event_generate(ET_WRITE, sock, 0);
393 case SS_DATAGRAM: case SS_CONNECTDG:
394 if (events[i].filter == EVFILT_READ) { /* socket readable */
395 Debug((DEBUG_ENGINE, "kqueue: Datagram to be read"));
396 event_generate(ET_READ, sock, 0);
398 if (events[i].filter == EVFILT_WRITE) { /* socket writable */
399 Debug((DEBUG_ENGINE, "kqueue: Datagram can be written"));
400 event_generate(ET_WRITE, sock, 0);
405 assert(s_fd(sock) == events[i].ident);
407 gen_ref_dec(sock); /* we're done with it */
410 timer_run(); /* execute any pending timers */
414 struct Engine engine_kqueue = {
415 "kqueue()", /* Engine name */
416 engine_init, /* Engine initialization function */
417 engine_signal, /* Engine signal registration function */
418 engine_add, /* Engine socket registration function */
419 engine_state, /* Engine socket state change function */
420 engine_events, /* Engine socket events mask function */
421 engine_delete, /* Engine socket deletion function */
422 engine_loop /* Core engine event loop */