]> jfr.im git - irc/rqf/shadowircd.git/blob - libratbox/src/epoll.c
Copied libratbox and related stuff from shadowircd upstream.
[irc/rqf/shadowircd.git] / libratbox / src / epoll.c
1 /*
2 * ircd-ratbox: A slightly useful ircd.
3 * epoll.c: Linux epoll compatible network routines.
4 *
5 * Copyright (C) 1990 Jarkko Oikarinen and University of Oulu, Co Center
6 * Copyright (C) 1996-2002 Hybrid Development Team
7 * Copyright (C) 2001 Adrian Chadd <adrian@creative.net.au>
8 * Copyright (C) 2002-2005 ircd-ratbox development team
9 * Copyright (C) 2002 Aaron Sethman <androsyn@ratbox.org>
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2 of the License, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301
24 * USA
25 *
26 * $Id: epoll.c 26092 2008-09-19 15:13:52Z androsyn $
27 */
28 #define _GNU_SOURCE 1
29
30 #include <libratbox_config.h>
31 #include <ratbox_lib.h>
32 #include <commio-int.h>
33 #include <event-int.h>
34 #if defined(HAVE_EPOLL_CTL) && (HAVE_SYS_EPOLL_H)
35 #define USING_EPOLL
36 #include <fcntl.h>
37 #include <sys/epoll.h>
38
39 #if defined(HAVE_SIGNALFD) && (HAVE_SYS_SIGNALFD_H) && (USE_TIMER_CREATE) && (HAVE_SYS_UIO_H)
40 #include <signal.h>
41 #include <sys/signalfd.h>
42 #include <sys/uio.h>
43 #define EPOLL_SCHED_EVENT 1
44 #endif
45
46 #if defined(USE_TIMERFD_CREATE)
47 #include <sys/timerfd.h>
48 #endif
49
50 #define RTSIGNAL SIGRTMIN
51 struct epoll_info
52 {
53 int ep;
54 struct epoll_event *pfd;
55 int pfd_size;
56 };
57
58 static struct epoll_info *ep_info;
59 static int can_do_event;
60 static int can_do_timerfd;
61
62 /*
63 * rb_init_netio
64 *
65 * This is a needed exported function which will be called to initialise
66 * the network loop code.
67 */
68 int
69 rb_init_netio_epoll(void)
70 {
71 can_do_event = 0; /* shut up gcc */
72 ep_info = rb_malloc(sizeof(struct epoll_info));
73 ep_info->pfd_size = getdtablesize();
74 ep_info->ep = epoll_create(ep_info->pfd_size);
75 if(ep_info->ep < 0)
76 {
77 return -1;
78 }
79 rb_open(ep_info->ep, RB_FD_UNKNOWN, "epoll file descriptor");
80 ep_info->pfd = rb_malloc(sizeof(struct epoll_event) * ep_info->pfd_size);
81
82 return 0;
83 }
84
85 int
86 rb_setup_fd_epoll(rb_fde_t *F)
87 {
88 return 0;
89 }
90
91
92 /*
93 * rb_setselect
94 *
95 * This is a needed exported function which will be called to register
96 * and deregister interest in a pending IO state for a given FD.
97 */
98 void
99 rb_setselect_epoll(rb_fde_t *F, unsigned int type, PF * handler, void *client_data)
100 {
101 struct epoll_event ep_event;
102 int old_flags = F->pflags;
103 int op = -1;
104
105 lrb_assert(IsFDOpen(F));
106
107 /* Update the list, even though we're not using it .. */
108 if(type & RB_SELECT_READ)
109 {
110 if(handler != NULL)
111 F->pflags |= EPOLLIN;
112 else
113 F->pflags &= ~EPOLLIN;
114 F->read_handler = handler;
115 F->read_data = client_data;
116 }
117
118 if(type & RB_SELECT_WRITE)
119 {
120 if(handler != NULL)
121 F->pflags |= EPOLLOUT;
122 else
123 F->pflags &= ~EPOLLOUT;
124 F->write_handler = handler;
125 F->write_data = client_data;
126 }
127
128 if(old_flags == 0 && F->pflags == 0)
129 return;
130 else if(F->pflags <= 0)
131 op = EPOLL_CTL_DEL;
132 else if(old_flags == 0 && F->pflags > 0)
133 op = EPOLL_CTL_ADD;
134 else if(F->pflags != old_flags)
135 op = EPOLL_CTL_MOD;
136
137 if(op == -1)
138 return;
139
140 ep_event.events = F->pflags;
141 ep_event.data.ptr = F;
142
143 if(op == EPOLL_CTL_ADD || op == EPOLL_CTL_MOD)
144 ep_event.events |= EPOLLET;
145
146 if(epoll_ctl(ep_info->ep, op, F->fd, &ep_event) != 0)
147 {
148 rb_lib_log("rb_setselect_epoll(): epoll_ctl failed: %s", strerror(errno));
149 abort();
150 }
151
152
153 }
154
155 /*
156 * rb_select
157 *
158 * Called to do the new-style IO, courtesy of squid (like most of this
159 * new IO code). This routine handles the stuff we've hidden in
160 * rb_setselect and fd_table[] and calls callbacks for IO ready
161 * events.
162 */
163
164 int
165 rb_select_epoll(long delay)
166 {
167 int num, i, flags, old_flags, op;
168 struct epoll_event ep_event;
169 int o_errno;
170 void *data;
171
172 num = epoll_wait(ep_info->ep, ep_info->pfd, ep_info->pfd_size, delay);
173
174 /* save errno as rb_set_time() will likely clobber it */
175 o_errno = errno;
176 rb_set_time();
177 errno = o_errno;
178
179 if(num < 0 && !rb_ignore_errno(o_errno))
180 return RB_ERROR;
181
182 if(num <= 0)
183 return RB_OK;
184
185 for(i = 0; i < num; i++)
186 {
187 PF *hdl;
188 rb_fde_t *F = ep_info->pfd[i].data.ptr;
189 old_flags = F->pflags;
190 if(ep_info->pfd[i].events & (EPOLLIN | EPOLLHUP | EPOLLERR))
191 {
192 hdl = F->read_handler;
193 data = F->read_data;
194 F->read_handler = NULL;
195 F->read_data = NULL;
196 if(hdl)
197 {
198 hdl(F, data);
199 }
200 }
201
202 if(!IsFDOpen(F))
203 continue;
204 if(ep_info->pfd[i].events & (EPOLLOUT | EPOLLHUP | EPOLLERR))
205 {
206 hdl = F->write_handler;
207 data = F->write_data;
208 F->write_handler = NULL;
209 F->write_data = NULL;
210
211 if(hdl)
212 {
213 hdl(F, data);
214 }
215 }
216
217 if(!IsFDOpen(F))
218 continue;
219
220 flags = 0;
221
222 if(F->read_handler != NULL)
223 flags |= EPOLLIN;
224 if(F->write_handler != NULL)
225 flags |= EPOLLOUT;
226
227 if(old_flags != flags)
228 {
229 if(flags == 0)
230 op = EPOLL_CTL_DEL;
231 else
232 op = EPOLL_CTL_MOD;
233 F->pflags = ep_event.events = flags;
234 ep_event.data.ptr = F;
235 if(op == EPOLL_CTL_MOD || op == EPOLL_CTL_ADD)
236 ep_event.events |= EPOLLET;
237
238 if(epoll_ctl(ep_info->ep, op, F->fd, &ep_event) != 0)
239 {
240 rb_lib_log("rb_select_epoll(): epoll_ctl failed: %s",
241 strerror(errno));
242 }
243 }
244
245 }
246 return RB_OK;
247 }
248
249 #ifdef EPOLL_SCHED_EVENT
250 int
251 rb_epoll_supports_event(void)
252 {
253 /* try to detect at runtime if everything we need actually works */
254 timer_t timer;
255 struct sigevent ev;
256 int fd;
257 sigset_t set;
258
259 if(can_do_event == 1)
260 return 1;
261 if(can_do_event == -1)
262 return 0;
263
264 #ifdef USE_TIMERFD_CREATE
265 if((fd = timerfd_create(CLOCK_REALTIME, 0)) >= 0)
266 {
267 close(fd);
268 can_do_event = 1;
269 can_do_timerfd = 1;
270 return 0;
271 }
272 #endif
273
274 ev.sigev_signo = SIGVTALRM;
275 ev.sigev_notify = SIGEV_SIGNAL;
276 if(timer_create(CLOCK_REALTIME, &ev, &timer) != 0)
277 {
278 can_do_event = -1;
279 return 0;
280 }
281 timer_delete(timer);
282 sigemptyset(&set);
283 fd = signalfd(-1, &set, 0);
284 if(fd < 0)
285 {
286 can_do_event = -1;
287 return 0;
288 }
289 close(fd);
290 can_do_event = 1;
291 return 1;
292 }
293
294
295 /* bleh..work around a glibc header bug on 32bit systems */
296 struct our_signalfd_siginfo
297 {
298 uint32_t signo;
299 int32_t err;
300 int32_t code;
301 uint32_t pid;
302 uint32_t uid;
303 int32_t fd;
304 uint32_t tid;
305 uint32_t band;
306 uint32_t overrun;
307 uint32_t trapno;
308 int32_t status;
309 int32_t svint;
310 uint64_t svptr;
311 uint64_t utime;
312 uint64_t stime;
313 uint64_t addr;
314 uint8_t pad[48];
315 };
316
317
318 #define SIGFDIOV_COUNT 16
319 static void
320 signalfd_handler(rb_fde_t *F, void *data)
321 {
322 static struct our_signalfd_siginfo fdsig[SIGFDIOV_COUNT];
323 static struct iovec iov[SIGFDIOV_COUNT];
324 struct ev_entry *ev;
325 int ret, x;
326
327 for(x = 0; x < SIGFDIOV_COUNT; x++)
328 {
329 iov[x].iov_base = &fdsig[x];
330 iov[x].iov_len = sizeof(struct our_signalfd_siginfo);
331 }
332
333 while(1)
334 {
335 ret = readv(rb_get_fd(F), iov, SIGFDIOV_COUNT);
336 if(ret == 0 || (ret < 0 && !rb_ignore_errno(errno)))
337 {
338 rb_close(F);
339 rb_epoll_init_event();
340 return;
341 }
342
343 if(ret < 0)
344 {
345 rb_setselect(F, RB_SELECT_READ, signalfd_handler, NULL);
346 return;
347 }
348 for(x = 0; x < ret / (int)sizeof(struct our_signalfd_siginfo); x++)
349 {
350 ev = (struct ev_entry *)((uintptr_t)fdsig[x].svptr);
351 if(ev == NULL)
352 continue;
353 rb_run_event(ev);
354 }
355 }
356 }
357
358 void
359 rb_epoll_init_event(void)
360 {
361
362 sigset_t ss;
363 rb_fde_t *F;
364 int sfd;
365 rb_epoll_supports_event();
366 if(!can_do_timerfd)
367 {
368 sigemptyset(&ss);
369 sigaddset(&ss, RTSIGNAL);
370 sigprocmask(SIG_BLOCK, &ss, 0);
371 sigemptyset(&ss);
372 sigaddset(&ss, RTSIGNAL);
373 sfd = signalfd(-1, &ss, 0);
374 if(sfd == -1)
375 {
376 can_do_event = -1;
377 return;
378 }
379 F = rb_open(sfd, RB_FD_UNKNOWN, "signalfd");
380 rb_set_nb(F);
381 signalfd_handler(F, NULL);
382 }
383 }
384
385 static int
386 rb_epoll_sched_event_signalfd(struct ev_entry *event, int when)
387 {
388 timer_t *id;
389 struct sigevent ev;
390 struct itimerspec ts;
391
392 memset(&ev, 0, sizeof(&ev));
393 event->comm_ptr = rb_malloc(sizeof(timer_t));
394 id = event->comm_ptr;
395 ev.sigev_notify = SIGEV_SIGNAL;
396 ev.sigev_signo = RTSIGNAL;
397 ev.sigev_value.sival_ptr = event;
398
399 if(timer_create(CLOCK_REALTIME, &ev, id) < 0)
400 {
401 rb_lib_log("timer_create: %s\n", strerror(errno));
402 return 0;
403 }
404 memset(&ts, 0, sizeof(ts));
405 ts.it_value.tv_sec = when;
406 ts.it_value.tv_nsec = 0;
407 if(event->frequency != 0)
408 ts.it_interval = ts.it_value;
409
410 if(timer_settime(*id, 0, &ts, NULL) < 0)
411 {
412 rb_lib_log("timer_settime: %s\n", strerror(errno));
413 return 0;
414 }
415 return 1;
416 }
417
418 #ifdef USE_TIMERFD_CREATE
419 static void
420 rb_read_timerfd(rb_fde_t *F, void *data)
421 {
422 struct ev_entry *event = (struct ev_entry *)data;
423 int retlen;
424 uint64_t count;
425
426 if(event == NULL)
427 {
428 rb_close(F);
429 return;
430 }
431
432 retlen = rb_read(F, &count, sizeof(count));
433 if(retlen == 0 || (retlen < 0 && !rb_ignore_errno(errno)))
434 {
435 rb_close(F);
436 rb_lib_log("rb_read_timerfd: timerfd[%s] closed on error: %s", event->name,
437 strerror(errno));
438 return;
439 }
440 rb_setselect(F, RB_SELECT_READ, rb_read_timerfd, event);
441 rb_run_event(event);
442 }
443
444
445 static int
446 rb_epoll_sched_event_timerfd(struct ev_entry *event, int when)
447 {
448 struct itimerspec ts;
449 static char buf[FD_DESC_SZ + 8];
450 int fd;
451 rb_fde_t *F;
452
453 if((fd = timerfd_create(CLOCK_REALTIME, 0)) < 0)
454 {
455 rb_lib_log("timerfd_create: %s\n", strerror(errno));
456 return 0;
457 }
458
459 memset(&ts, 0, sizeof(ts));
460 ts.it_value.tv_sec = when;
461 ts.it_value.tv_nsec = 0;
462 if(event->frequency != 0)
463 ts.it_interval = ts.it_value;
464
465 if(timerfd_settime(fd, 0, &ts, NULL) < 0)
466 {
467 rb_lib_log("timerfd_settime: %s\n", strerror(errno));
468 close(fd);
469 return 0;
470 }
471 rb_snprintf(buf, sizeof(buf), "timerfd: %s", event->name);
472 F = rb_open(fd, RB_FD_UNKNOWN, buf);
473 rb_set_nb(F);
474 event->comm_ptr = F;
475 rb_setselect(F, RB_SELECT_READ, rb_read_timerfd, event);
476 return 1;
477 }
478 #endif
479
480
481
482 int
483 rb_epoll_sched_event(struct ev_entry *event, int when)
484 {
485 #ifdef USE_TIMERFD_CREATE
486 if(can_do_timerfd)
487 {
488 return rb_epoll_sched_event_timerfd(event, when);
489 }
490 #endif
491 return rb_epoll_sched_event_signalfd(event, when);
492 }
493
494 void
495 rb_epoll_unsched_event(struct ev_entry *event)
496 {
497 #ifdef USE_TIMERFD_CREATE
498 if(can_do_timerfd)
499 {
500 rb_close((rb_fde_t *)event->comm_ptr);
501 event->comm_ptr = NULL;
502 return;
503 }
504 #endif
505 timer_delete(*((timer_t *) event->comm_ptr));
506 rb_free(event->comm_ptr);
507 event->comm_ptr = NULL;
508 }
509 #endif /* EPOLL_SCHED_EVENT */
510
511 #else /* epoll not supported here */
512 int
513 rb_init_netio_epoll(void)
514 {
515 return ENOSYS;
516 }
517
518 void
519 rb_setselect_epoll(rb_fde_t *F, unsigned int type, PF * handler, void *client_data)
520 {
521 errno = ENOSYS;
522 return;
523 }
524
525 int
526 rb_select_epoll(long delay)
527 {
528 errno = ENOSYS;
529 return -1;
530 }
531
532 int
533 rb_setup_fd_epoll(rb_fde_t *F)
534 {
535 errno = ENOSYS;
536 return -1;
537 }
538
539
540 #endif
541
542 #if !defined(USING_EPOLL) || !defined(EPOLL_SCHED_EVENT)
543 void
544 rb_epoll_init_event(void)
545 {
546 return;
547 }
548
549 int
550 rb_epoll_sched_event(struct ev_entry *event, int when)
551 {
552 errno = ENOSYS;
553 return -1;
554 }
555
556 void
557 rb_epoll_unsched_event(struct ev_entry *event)
558 {
559 return;
560 }
561
562 int
563 rb_epoll_supports_event(void)
564 {
565 errno = ENOSYS;
566 return 0;
567 }
568 #endif /* !USING_EPOLL || !EPOLL_SCHED_EVENT */