2 * IRC - Internet Relay Chat, ircd/engine_devpoll.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"
33 #include <sys/devpoll.h>
35 #include <sys/socket.h>
37 #include <sys/types.h>
40 #define DEVPOLL_ERROR_THRESHOLD 20 /* after 20 devpoll errors, restart */
41 #define ERROR_EXPIRE_TIME 3600 /* expire errors after an hour */
43 #define POLLS_PER_DEVPOLL 20 /* get 20 pollfd's per turn */
45 /* Figure out what bits to set for read */
46 #if defined(POLLMSG) && defined(POLLIN) && defined(POLLRDNORM)
47 # define POLLREADFLAGS (POLLMSG|POLLIN|POLLRDNORM)
48 #elif defined(POLLIN) && defined(POLLRDNORM)
49 # define POLLREADFLAGS (POLLIN|POLLRDNORM)
51 # define POLLREADFLAGS POLLIN
52 #elif defined(POLLRDNORM)
53 # define POLLREADFLAGS POLLRDNORM
56 /* Figure out what bits to set for write */
57 #if defined(POLLOUT) && defined(POLLWRNORM)
58 # define POLLWRITEFLAGS (POLLOUT|POLLWRNORM)
59 #elif defined(POLLOUT)
60 # define POLLWRITEFLAGS POLLOUT
61 #elif defined(POLLWRNORM)
62 # define POLLWRITEFLAGS POLLWRNORM
65 static struct Socket** sockList;
66 static int devpoll_max;
67 static int devpoll_fd;
69 static int errors = 0;
70 static struct Timer clear_error;
72 /* decrements the error count once per hour */
74 error_clear(struct Event* ev)
76 if (!--errors) /* remove timer when error count reaches 0 */
77 timer_del(ev_timer(ev));
80 /* initialize the devpoll engine */
82 engine_init(int max_sockets)
86 if ((devpoll_fd = open("/dev/poll", O_RDWR)) < 0) {
87 log_write(LS_SYSTEM, L_WARNING, 0,
88 "/dev/poll engine cannot open device: %m");
89 return 0; /* engine cannot be initialized; defer */
92 /* allocate necessary memory */
93 sockList = (struct Socket**) MyMalloc(sizeof(struct Socket*) * max_sockets);
95 /* initialize the data */
96 for (i = 0; i < max_sockets; i++)
99 devpoll_max = max_sockets; /* number of sockets allocated */
104 /* Figure out what events go with a given state */
106 state_to_events(enum SocketState state, unsigned int events)
109 case SS_CONNECTING: /* connecting socket */
110 return SOCK_EVENT_WRITABLE;
113 case SS_LISTENING: /* listening socket */
114 case SS_NOTSOCK: /* our signal socket */
115 return SOCK_EVENT_READABLE;
118 case SS_CONNECTED: case SS_DATAGRAM: case SS_CONNECTDG:
119 return events; /* ordinary socket */
127 /* Reset the desired events */
129 set_events(struct Socket* sock, unsigned int events)
135 if (s_ed_int(sock)) { /* is one in /dev/poll already? */
136 pfd.events = POLLREMOVE; /* First, remove old pollfd */
138 Debug((DEBUG_ENGINE, "devpoll: Removing old entry for socket %d [%p]",
141 if (write(devpoll_fd, &pfd, sizeof(pfd)) != sizeof(pfd)) {
142 event_generate(ET_ERROR, sock, errno); /* report error */
146 s_ed_int(sock) = 0; /* mark that it's gone */
149 if (!(events & SOCK_EVENT_MASK)) /* no events, so stop here */
152 pfd.events = 0; /* Now, set up new pollfd... */
153 if (events & SOCK_EVENT_READABLE)
154 pfd.events |= POLLREADFLAGS; /* look for readable conditions */
155 if (events & SOCK_EVENT_WRITABLE)
156 pfd.events |= POLLWRITEFLAGS; /* look for writable conditions */
158 Debug((DEBUG_ENGINE, "devpoll: Registering interest on %d [%p] (state %s, "
159 "mask [%s])", s_fd(sock), sock, state_to_name(s_state(sock)),
160 sock_flags(s_events(sock))));
162 if (write(devpoll_fd, &pfd, sizeof(pfd)) != sizeof(pfd)) {
163 event_generate(ET_ERROR, sock, errno); /* report error */
167 s_ed_int(sock) = 1; /* mark that we've added a pollfd */
170 /* add a socket to be listened on */
172 engine_add(struct Socket* sock)
175 assert(0 == sockList[s_fd(sock)]);
177 /* bounds-check... */
178 if (s_fd(sock) >= devpoll_max) {
179 log_write(LS_SYSTEM, L_ERROR, 0,
180 "Attempt to add socket %d (> %d) to event engine", s_fd(sock),
185 sockList[s_fd(sock)] = sock; /* add to list */
187 Debug((DEBUG_ENGINE, "devpoll: Adding socket %d [%p], state %s, to engine",
188 s_fd(sock), sock, state_to_name(s_state(sock))));
190 /* set the correct events */
191 set_events(sock, state_to_events(s_state(sock), s_events(sock)));
193 return 1; /* success */
196 /* socket switching to new state */
198 engine_state(struct Socket* sock, enum SocketState new_state)
201 assert(sock == sockList[s_fd(sock)]);
203 Debug((DEBUG_ENGINE, "devpoll: Changing state for socket %p to %s", sock,
204 state_to_name(new_state)));
206 /* set the correct events */
207 set_events(sock, state_to_events(new_state, s_events(sock)));
210 /* socket events changing */
212 engine_events(struct Socket* sock, unsigned int new_events)
215 assert(sock == sockList[s_fd(sock)]);
217 Debug((DEBUG_ENGINE, "devpoll: Changing event mask for socket %p to [%s]",
218 sock, sock_flags(new_events)));
220 /* set the correct events */
221 set_events(sock, state_to_events(s_state(sock), new_events));
224 /* socket going away */
226 engine_delete(struct Socket* sock)
229 assert(sock == sockList[s_fd(sock)]);
231 Debug((DEBUG_ENGINE, "devpoll: Deleting socket %d [%p], state %s",
232 s_fd(sock), sock, state_to_name(s_state(sock))));
234 set_events(sock, 0); /* get rid of the socket */
236 sockList[s_fd(sock)] = 0; /* zero the socket list entry */
239 /* engine event loop */
241 engine_loop(struct Generators* gen)
243 struct dvpoll dopoll;
244 struct pollfd polls[POLLS_PER_DEVPOLL];
252 dopoll.dp_fds = polls; /* set up the struct dvpoll */
253 dopoll.dp_nfds = POLLS_PER_DEVPOLL;
255 /* calculate the proper timeout */
256 dopoll.dp_timeout = timer_next(gen) ?
257 (timer_next(gen) - CurrentTime) * 1000 : -1;
259 Debug((DEBUG_INFO, "devpoll: delay: %Tu (%Tu) %d", timer_next(gen),
260 CurrentTime, dopoll.dp_timeout));
262 /* check for active files */
263 nfds = ioctl(devpoll_fd, DP_POLL, &dopoll);
265 CurrentTime = time(0); /* set current time... */
268 if (errno != EINTR) { /* ignore interrupts */
269 /* Log the poll error */
270 log_write(LS_SOCKET, L_ERROR, 0, "ioctl(DP_POLL) error: %m");
272 timer_add(&clear_error, error_clear, 0, TT_PERIODIC,
274 else if (errors > DEVPOLL_ERROR_THRESHOLD) /* too many errors... */
275 server_restart("too many /dev/poll errors");
277 /* old code did a sleep(1) here; with usage these days,
278 * that may be too expensive
283 for (i = 0; i < nfds; i++) {
284 assert(-1 < polls[i].fd);
285 assert(0 != sockList[polls[i].fd]);
286 assert(s_fd(sockList[polls[i].fd]) == polls[i].fd);
288 sock = sockList[polls[i].fd];
289 if (!sock) /* slots may become empty while processing events */
292 gen_ref_inc(sock); /* can't have it going away on us */
294 Debug((DEBUG_ENGINE, "devpoll: Checking socket %p (fd %d) state %s, "
295 "events %s", sock, s_fd(sock), state_to_name(s_state(sock)),
296 sock_flags(s_events(sock))));
298 if (s_state(sock) != SS_NOTSOCK) {
299 errcode = 0; /* check for errors on socket */
300 codesize = sizeof(errcode);
301 if (getsockopt(s_fd(sock), SOL_SOCKET, SO_ERROR, &errcode,
303 errcode = errno; /* work around Solaris implementation */
305 if (errcode) { /* an error occurred; generate an event */
306 Debug((DEBUG_ENGINE, "devpoll: Error %d on fd %d, socket %p",
307 errcode, s_fd(sock), sock));
308 event_generate(ET_ERROR, sock, errcode);
309 gen_ref_dec(sock); /* careful not to leak reference counts */
314 assert(!(polls[i].revents & POLLERR));
317 if (polls[i].revents & POLLHUP) { /* hang-up on socket */
318 Debug((DEBUG_ENGINE, "devpoll: EOF from client (POLLHUP)"));
319 event_generate(ET_EOF, sock, 0);
325 switch (s_state(sock)) {
327 if (polls[i].revents & POLLWRITEFLAGS) { /* connection completed */
328 Debug((DEBUG_ENGINE, "devpoll: Connection completed"));
329 event_generate(ET_CONNECT, sock, 0);
334 if (polls[i].revents & POLLREADFLAGS) { /* connect. to be accept. */
335 Debug((DEBUG_ENGINE, "devpoll: Ready for accept"));
336 event_generate(ET_ACCEPT, sock, 0);
341 if (polls[i].revents & POLLREADFLAGS) { /* data on socket */
342 /* can't peek; it's not a socket */
343 Debug((DEBUG_ENGINE, "devpoll: non-socket readable"));
344 event_generate(ET_READ, sock, 0);
349 if (polls[i].revents & POLLREADFLAGS) { /* data on socket */
352 switch (recv(s_fd(sock), &c, 1, MSG_PEEK)) { /* check EOF */
353 case -1: /* error occurred?!? */
354 if (errno == EAGAIN) {
355 Debug((DEBUG_ENGINE, "devpoll: Resource temporarily "
359 Debug((DEBUG_ENGINE, "devpoll: Uncaught error!"));
360 event_generate(ET_ERROR, sock, errno);
363 case 0: /* EOF from client */
364 Debug((DEBUG_ENGINE, "devpoll: EOF from client"));
365 event_generate(ET_EOF, sock, 0);
368 default: /* some data can be read */
369 Debug((DEBUG_ENGINE, "devpoll: Data to be read"));
370 event_generate(ET_READ, sock, 0);
374 if (polls[i].revents & POLLWRITEFLAGS) { /* socket writable */
375 Debug((DEBUG_ENGINE, "devpoll: Data can be written"));
376 event_generate(ET_WRITE, sock, 0);
380 case SS_DATAGRAM: case SS_CONNECTDG:
381 if (polls[i].revents & POLLREADFLAGS) { /* socket readable */
382 Debug((DEBUG_ENGINE, "devpoll: Datagram to be read"));
383 event_generate(ET_READ, sock, 0);
385 if (polls[i].revents & POLLWRITEFLAGS) { /* socket writable */
386 Debug((DEBUG_ENGINE, "devpoll: Datagram can be written"));
387 event_generate(ET_WRITE, sock, 0);
392 assert(s_fd(sock) == polls[i].fd);
394 gen_ref_dec(sock); /* we're done with it */
397 timer_run(); /* execute any pending timers */
401 struct Engine engine_devpoll = {
402 "/dev/poll", /* Engine name */
403 engine_init, /* Engine initialization function */
404 0, /* Engine signal registration function */
405 engine_add, /* Engine socket registration function */
406 engine_state, /* Engine socket state change function */
407 engine_events, /* Engine socket events mask function */
408 engine_delete, /* Engine socket deletion function */
409 engine_loop /* Core engine event loop */