xref: /freebsd/sys/kern/kern_time.c (revision 7aa383846770374466b1dcb2cefd71bde9acf463)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)kern_time.c	8.1 (Berkeley) 6/10/93
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/limits.h>
38 #include <sys/clock.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/eventhandler.h>
43 #include <sys/resourcevar.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysctl.h>
48 #include <sys/sysent.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/posix4.h>
52 #include <sys/time.h>
53 #include <sys/timers.h>
54 #include <sys/timetc.h>
55 #include <sys/vnode.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_extern.h>
59 
60 #define MAX_CLOCKS 	(CLOCK_MONOTONIC+1)
61 
62 static struct kclock	posix_clocks[MAX_CLOCKS];
63 static uma_zone_t	itimer_zone = NULL;
64 
65 /*
66  * Time of day and interval timer support.
67  *
68  * These routines provide the kernel entry points to get and set
69  * the time-of-day and per-process interval timers.  Subroutines
70  * here provide support for adding and subtracting timeval structures
71  * and decrementing interval timers, optionally reloading the interval
72  * timers when they expire.
73  */
74 
75 static int	settime(struct thread *, struct timeval *);
76 static void	timevalfix(struct timeval *);
77 
78 static void	itimer_start(void);
79 static int	itimer_init(void *, int, int);
80 static void	itimer_fini(void *, int);
81 static void	itimer_enter(struct itimer *);
82 static void	itimer_leave(struct itimer *);
83 static struct itimer *itimer_find(struct proc *, int);
84 static void	itimers_alloc(struct proc *);
85 static void	itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
86 static void	itimers_event_hook_exit(void *arg, struct proc *p);
87 static int	realtimer_create(struct itimer *);
88 static int	realtimer_gettime(struct itimer *, struct itimerspec *);
89 static int	realtimer_settime(struct itimer *, int,
90 			struct itimerspec *, struct itimerspec *);
91 static int	realtimer_delete(struct itimer *);
92 static void	realtimer_clocktime(clockid_t, struct timespec *);
93 static void	realtimer_expire(void *);
94 static int	kern_timer_create(struct thread *, clockid_t,
95 			struct sigevent *, int *, int);
96 static int	kern_timer_delete(struct thread *, int);
97 
98 int		register_posix_clock(int, struct kclock *);
99 void		itimer_fire(struct itimer *it);
100 int		itimespecfix(struct timespec *ts);
101 
102 #define CLOCK_CALL(clock, call, arglist)		\
103 	((*posix_clocks[clock].call) arglist)
104 
105 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
106 
107 
108 static int
109 settime(struct thread *td, struct timeval *tv)
110 {
111 	struct timeval delta, tv1, tv2;
112 	static struct timeval maxtime, laststep;
113 	struct timespec ts;
114 	int s;
115 
116 	s = splclock();
117 	microtime(&tv1);
118 	delta = *tv;
119 	timevalsub(&delta, &tv1);
120 
121 	/*
122 	 * If the system is secure, we do not allow the time to be
123 	 * set to a value earlier than 1 second less than the highest
124 	 * time we have yet seen. The worst a miscreant can do in
125 	 * this circumstance is "freeze" time. He couldn't go
126 	 * back to the past.
127 	 *
128 	 * We similarly do not allow the clock to be stepped more
129 	 * than one second, nor more than once per second. This allows
130 	 * a miscreant to make the clock march double-time, but no worse.
131 	 */
132 	if (securelevel_gt(td->td_ucred, 1) != 0) {
133 		if (delta.tv_sec < 0 || delta.tv_usec < 0) {
134 			/*
135 			 * Update maxtime to latest time we've seen.
136 			 */
137 			if (tv1.tv_sec > maxtime.tv_sec)
138 				maxtime = tv1;
139 			tv2 = *tv;
140 			timevalsub(&tv2, &maxtime);
141 			if (tv2.tv_sec < -1) {
142 				tv->tv_sec = maxtime.tv_sec - 1;
143 				printf("Time adjustment clamped to -1 second\n");
144 			}
145 		} else {
146 			if (tv1.tv_sec == laststep.tv_sec) {
147 				splx(s);
148 				return (EPERM);
149 			}
150 			if (delta.tv_sec > 1) {
151 				tv->tv_sec = tv1.tv_sec + 1;
152 				printf("Time adjustment clamped to +1 second\n");
153 			}
154 			laststep = *tv;
155 		}
156 	}
157 
158 	ts.tv_sec = tv->tv_sec;
159 	ts.tv_nsec = tv->tv_usec * 1000;
160 	mtx_lock(&Giant);
161 	tc_setclock(&ts);
162 	resettodr();
163 	mtx_unlock(&Giant);
164 	return (0);
165 }
166 
167 #ifndef _SYS_SYSPROTO_H_
168 struct clock_gettime_args {
169 	clockid_t clock_id;
170 	struct	timespec *tp;
171 };
172 #endif
173 /* ARGSUSED */
174 int
175 clock_gettime(struct thread *td, struct clock_gettime_args *uap)
176 {
177 	struct timespec ats;
178 	int error;
179 
180 	error = kern_clock_gettime(td, uap->clock_id, &ats);
181 	if (error == 0)
182 		error = copyout(&ats, uap->tp, sizeof(ats));
183 
184 	return (error);
185 }
186 
187 int
188 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
189 {
190 	struct timeval sys, user;
191 	struct proc *p;
192 	uint64_t runtime, curtime, switchtime;
193 
194 	p = td->td_proc;
195 	switch (clock_id) {
196 	case CLOCK_REALTIME:		/* Default to precise. */
197 	case CLOCK_REALTIME_PRECISE:
198 		nanotime(ats);
199 		break;
200 	case CLOCK_REALTIME_FAST:
201 		getnanotime(ats);
202 		break;
203 	case CLOCK_VIRTUAL:
204 		PROC_LOCK(p);
205 		PROC_SLOCK(p);
206 		calcru(p, &user, &sys);
207 		PROC_SUNLOCK(p);
208 		PROC_UNLOCK(p);
209 		TIMEVAL_TO_TIMESPEC(&user, ats);
210 		break;
211 	case CLOCK_PROF:
212 		PROC_LOCK(p);
213 		PROC_SLOCK(p);
214 		calcru(p, &user, &sys);
215 		PROC_SUNLOCK(p);
216 		PROC_UNLOCK(p);
217 		timevaladd(&user, &sys);
218 		TIMEVAL_TO_TIMESPEC(&user, ats);
219 		break;
220 	case CLOCK_MONOTONIC:		/* Default to precise. */
221 	case CLOCK_MONOTONIC_PRECISE:
222 	case CLOCK_UPTIME:
223 	case CLOCK_UPTIME_PRECISE:
224 		nanouptime(ats);
225 		break;
226 	case CLOCK_UPTIME_FAST:
227 	case CLOCK_MONOTONIC_FAST:
228 		getnanouptime(ats);
229 		break;
230 	case CLOCK_SECOND:
231 		ats->tv_sec = time_second;
232 		ats->tv_nsec = 0;
233 		break;
234 	case CLOCK_THREAD_CPUTIME_ID:
235 		critical_enter();
236 		switchtime = PCPU_GET(switchtime);
237 		curtime = cpu_ticks();
238 		runtime = td->td_runtime;
239 		critical_exit();
240 		runtime = cputick2usec(runtime + curtime - switchtime);
241 		ats->tv_sec = runtime / 1000000;
242 		ats->tv_nsec = runtime % 1000000 * 1000;
243 		break;
244 	default:
245 		return (EINVAL);
246 	}
247 	return (0);
248 }
249 
250 #ifndef _SYS_SYSPROTO_H_
251 struct clock_settime_args {
252 	clockid_t clock_id;
253 	const struct	timespec *tp;
254 };
255 #endif
256 /* ARGSUSED */
257 int
258 clock_settime(struct thread *td, struct clock_settime_args *uap)
259 {
260 	struct timespec ats;
261 	int error;
262 
263 	if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
264 		return (error);
265 	return (kern_clock_settime(td, uap->clock_id, &ats));
266 }
267 
268 int
269 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
270 {
271 	struct timeval atv;
272 	int error;
273 
274 	if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
275 		return (error);
276 	if (clock_id != CLOCK_REALTIME)
277 		return (EINVAL);
278 	if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
279 		return (EINVAL);
280 	/* XXX Don't convert nsec->usec and back */
281 	TIMESPEC_TO_TIMEVAL(&atv, ats);
282 	error = settime(td, &atv);
283 	return (error);
284 }
285 
286 #ifndef _SYS_SYSPROTO_H_
287 struct clock_getres_args {
288 	clockid_t clock_id;
289 	struct	timespec *tp;
290 };
291 #endif
292 int
293 clock_getres(struct thread *td, struct clock_getres_args *uap)
294 {
295 	struct timespec ts;
296 	int error;
297 
298 	if (uap->tp == NULL)
299 		return (0);
300 
301 	error = kern_clock_getres(td, uap->clock_id, &ts);
302 	if (error == 0)
303 		error = copyout(&ts, uap->tp, sizeof(ts));
304 	return (error);
305 }
306 
307 int
308 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
309 {
310 
311 	ts->tv_sec = 0;
312 	switch (clock_id) {
313 	case CLOCK_REALTIME:
314 	case CLOCK_REALTIME_FAST:
315 	case CLOCK_REALTIME_PRECISE:
316 	case CLOCK_MONOTONIC:
317 	case CLOCK_MONOTONIC_FAST:
318 	case CLOCK_MONOTONIC_PRECISE:
319 	case CLOCK_UPTIME:
320 	case CLOCK_UPTIME_FAST:
321 	case CLOCK_UPTIME_PRECISE:
322 		/*
323 		 * Round up the result of the division cheaply by adding 1.
324 		 * Rounding up is especially important if rounding down
325 		 * would give 0.  Perfect rounding is unimportant.
326 		 */
327 		ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
328 		break;
329 	case CLOCK_VIRTUAL:
330 	case CLOCK_PROF:
331 		/* Accurately round up here because we can do so cheaply. */
332 		ts->tv_nsec = (1000000000 + hz - 1) / hz;
333 		break;
334 	case CLOCK_SECOND:
335 		ts->tv_sec = 1;
336 		ts->tv_nsec = 0;
337 		break;
338 	case CLOCK_THREAD_CPUTIME_ID:
339 		/* sync with cputick2usec */
340 		ts->tv_nsec = 1000000 / cpu_tickrate();
341 		if (ts->tv_nsec == 0)
342 			ts->tv_nsec = 1000;
343 		break;
344 	default:
345 		return (EINVAL);
346 	}
347 	return (0);
348 }
349 
350 static int nanowait;
351 
352 int
353 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
354 {
355 	struct timespec ts, ts2, ts3;
356 	struct timeval tv;
357 	int error;
358 
359 	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
360 		return (EINVAL);
361 	if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
362 		return (0);
363 	getnanouptime(&ts);
364 	timespecadd(&ts, rqt);
365 	TIMESPEC_TO_TIMEVAL(&tv, rqt);
366 	for (;;) {
367 		error = tsleep(&nanowait, PWAIT | PCATCH, "nanslp",
368 		    tvtohz(&tv));
369 		getnanouptime(&ts2);
370 		if (error != EWOULDBLOCK) {
371 			if (error == ERESTART)
372 				error = EINTR;
373 			if (rmt != NULL) {
374 				timespecsub(&ts, &ts2);
375 				if (ts.tv_sec < 0)
376 					timespecclear(&ts);
377 				*rmt = ts;
378 			}
379 			return (error);
380 		}
381 		if (timespeccmp(&ts2, &ts, >=))
382 			return (0);
383 		ts3 = ts;
384 		timespecsub(&ts3, &ts2);
385 		TIMESPEC_TO_TIMEVAL(&tv, &ts3);
386 	}
387 }
388 
389 #ifndef _SYS_SYSPROTO_H_
390 struct nanosleep_args {
391 	struct	timespec *rqtp;
392 	struct	timespec *rmtp;
393 };
394 #endif
395 /* ARGSUSED */
396 int
397 nanosleep(struct thread *td, struct nanosleep_args *uap)
398 {
399 	struct timespec rmt, rqt;
400 	int error;
401 
402 	error = copyin(uap->rqtp, &rqt, sizeof(rqt));
403 	if (error)
404 		return (error);
405 
406 	if (uap->rmtp &&
407 	    !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE))
408 			return (EFAULT);
409 	error = kern_nanosleep(td, &rqt, &rmt);
410 	if (error && uap->rmtp) {
411 		int error2;
412 
413 		error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
414 		if (error2)
415 			error = error2;
416 	}
417 	return (error);
418 }
419 
420 #ifndef _SYS_SYSPROTO_H_
421 struct gettimeofday_args {
422 	struct	timeval *tp;
423 	struct	timezone *tzp;
424 };
425 #endif
426 /* ARGSUSED */
427 int
428 gettimeofday(struct thread *td, struct gettimeofday_args *uap)
429 {
430 	struct timeval atv;
431 	struct timezone rtz;
432 	int error = 0;
433 
434 	if (uap->tp) {
435 		microtime(&atv);
436 		error = copyout(&atv, uap->tp, sizeof (atv));
437 	}
438 	if (error == 0 && uap->tzp != NULL) {
439 		rtz.tz_minuteswest = tz_minuteswest;
440 		rtz.tz_dsttime = tz_dsttime;
441 		error = copyout(&rtz, uap->tzp, sizeof (rtz));
442 	}
443 	return (error);
444 }
445 
446 #ifndef _SYS_SYSPROTO_H_
447 struct settimeofday_args {
448 	struct	timeval *tv;
449 	struct	timezone *tzp;
450 };
451 #endif
452 /* ARGSUSED */
453 int
454 settimeofday(struct thread *td, struct settimeofday_args *uap)
455 {
456 	struct timeval atv, *tvp;
457 	struct timezone atz, *tzp;
458 	int error;
459 
460 	if (uap->tv) {
461 		error = copyin(uap->tv, &atv, sizeof(atv));
462 		if (error)
463 			return (error);
464 		tvp = &atv;
465 	} else
466 		tvp = NULL;
467 	if (uap->tzp) {
468 		error = copyin(uap->tzp, &atz, sizeof(atz));
469 		if (error)
470 			return (error);
471 		tzp = &atz;
472 	} else
473 		tzp = NULL;
474 	return (kern_settimeofday(td, tvp, tzp));
475 }
476 
477 int
478 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
479 {
480 	int error;
481 
482 	error = priv_check(td, PRIV_SETTIMEOFDAY);
483 	if (error)
484 		return (error);
485 	/* Verify all parameters before changing time. */
486 	if (tv) {
487 		if (tv->tv_usec < 0 || tv->tv_usec >= 1000000)
488 			return (EINVAL);
489 		error = settime(td, tv);
490 	}
491 	if (tzp && error == 0) {
492 		tz_minuteswest = tzp->tz_minuteswest;
493 		tz_dsttime = tzp->tz_dsttime;
494 	}
495 	return (error);
496 }
497 
498 /*
499  * Get value of an interval timer.  The process virtual and profiling virtual
500  * time timers are kept in the p_stats area, since they can be swapped out.
501  * These are kept internally in the way they are specified externally: in
502  * time until they expire.
503  *
504  * The real time interval timer is kept in the process table slot for the
505  * process, and its value (it_value) is kept as an absolute time rather than
506  * as a delta, so that it is easy to keep periodic real-time signals from
507  * drifting.
508  *
509  * Virtual time timers are processed in the hardclock() routine of
510  * kern_clock.c.  The real time timer is processed by a timeout routine,
511  * called from the softclock() routine.  Since a callout may be delayed in
512  * real time due to interrupt processing in the system, it is possible for
513  * the real time timeout routine (realitexpire, given below), to be delayed
514  * in real time past when it is supposed to occur.  It does not suffice,
515  * therefore, to reload the real timer .it_value from the real time timers
516  * .it_interval.  Rather, we compute the next time in absolute time the timer
517  * should go off.
518  */
519 #ifndef _SYS_SYSPROTO_H_
520 struct getitimer_args {
521 	u_int	which;
522 	struct	itimerval *itv;
523 };
524 #endif
525 int
526 getitimer(struct thread *td, struct getitimer_args *uap)
527 {
528 	struct itimerval aitv;
529 	int error;
530 
531 	error = kern_getitimer(td, uap->which, &aitv);
532 	if (error != 0)
533 		return (error);
534 	return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
535 }
536 
537 int
538 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
539 {
540 	struct proc *p = td->td_proc;
541 	struct timeval ctv;
542 
543 	if (which > ITIMER_PROF)
544 		return (EINVAL);
545 
546 	if (which == ITIMER_REAL) {
547 		/*
548 		 * Convert from absolute to relative time in .it_value
549 		 * part of real time timer.  If time for real time timer
550 		 * has passed return 0, else return difference between
551 		 * current time and time for the timer to go off.
552 		 */
553 		PROC_LOCK(p);
554 		*aitv = p->p_realtimer;
555 		PROC_UNLOCK(p);
556 		if (timevalisset(&aitv->it_value)) {
557 			getmicrouptime(&ctv);
558 			if (timevalcmp(&aitv->it_value, &ctv, <))
559 				timevalclear(&aitv->it_value);
560 			else
561 				timevalsub(&aitv->it_value, &ctv);
562 		}
563 	} else {
564 		PROC_SLOCK(p);
565 		*aitv = p->p_stats->p_timer[which];
566 		PROC_SUNLOCK(p);
567 	}
568 	return (0);
569 }
570 
571 #ifndef _SYS_SYSPROTO_H_
572 struct setitimer_args {
573 	u_int	which;
574 	struct	itimerval *itv, *oitv;
575 };
576 #endif
577 int
578 setitimer(struct thread *td, struct setitimer_args *uap)
579 {
580 	struct itimerval aitv, oitv;
581 	int error;
582 
583 	if (uap->itv == NULL) {
584 		uap->itv = uap->oitv;
585 		return (getitimer(td, (struct getitimer_args *)uap));
586 	}
587 
588 	if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
589 		return (error);
590 	error = kern_setitimer(td, uap->which, &aitv, &oitv);
591 	if (error != 0 || uap->oitv == NULL)
592 		return (error);
593 	return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
594 }
595 
596 int
597 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
598     struct itimerval *oitv)
599 {
600 	struct proc *p = td->td_proc;
601 	struct timeval ctv;
602 
603 	if (aitv == NULL)
604 		return (kern_getitimer(td, which, oitv));
605 
606 	if (which > ITIMER_PROF)
607 		return (EINVAL);
608 	if (itimerfix(&aitv->it_value))
609 		return (EINVAL);
610 	if (!timevalisset(&aitv->it_value))
611 		timevalclear(&aitv->it_interval);
612 	else if (itimerfix(&aitv->it_interval))
613 		return (EINVAL);
614 
615 	if (which == ITIMER_REAL) {
616 		PROC_LOCK(p);
617 		if (timevalisset(&p->p_realtimer.it_value))
618 			callout_stop(&p->p_itcallout);
619 		getmicrouptime(&ctv);
620 		if (timevalisset(&aitv->it_value)) {
621 			callout_reset(&p->p_itcallout, tvtohz(&aitv->it_value),
622 			    realitexpire, p);
623 			timevaladd(&aitv->it_value, &ctv);
624 		}
625 		*oitv = p->p_realtimer;
626 		p->p_realtimer = *aitv;
627 		PROC_UNLOCK(p);
628 		if (timevalisset(&oitv->it_value)) {
629 			if (timevalcmp(&oitv->it_value, &ctv, <))
630 				timevalclear(&oitv->it_value);
631 			else
632 				timevalsub(&oitv->it_value, &ctv);
633 		}
634 	} else {
635 		PROC_SLOCK(p);
636 		*oitv = p->p_stats->p_timer[which];
637 		p->p_stats->p_timer[which] = *aitv;
638 		PROC_SUNLOCK(p);
639 	}
640 	return (0);
641 }
642 
643 /*
644  * Real interval timer expired:
645  * send process whose timer expired an alarm signal.
646  * If time is not set up to reload, then just return.
647  * Else compute next time timer should go off which is > current time.
648  * This is where delay in processing this timeout causes multiple
649  * SIGALRM calls to be compressed into one.
650  * tvtohz() always adds 1 to allow for the time until the next clock
651  * interrupt being strictly less than 1 clock tick, but we don't want
652  * that here since we want to appear to be in sync with the clock
653  * interrupt even when we're delayed.
654  */
655 void
656 realitexpire(void *arg)
657 {
658 	struct proc *p;
659 	struct timeval ctv, ntv;
660 
661 	p = (struct proc *)arg;
662 	PROC_LOCK(p);
663 	psignal(p, SIGALRM);
664 	if (!timevalisset(&p->p_realtimer.it_interval)) {
665 		timevalclear(&p->p_realtimer.it_value);
666 		if (p->p_flag & P_WEXIT)
667 			wakeup(&p->p_itcallout);
668 		PROC_UNLOCK(p);
669 		return;
670 	}
671 	for (;;) {
672 		timevaladd(&p->p_realtimer.it_value,
673 		    &p->p_realtimer.it_interval);
674 		getmicrouptime(&ctv);
675 		if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
676 			ntv = p->p_realtimer.it_value;
677 			timevalsub(&ntv, &ctv);
678 			callout_reset(&p->p_itcallout, tvtohz(&ntv) - 1,
679 			    realitexpire, p);
680 			PROC_UNLOCK(p);
681 			return;
682 		}
683 	}
684 	/*NOTREACHED*/
685 }
686 
687 /*
688  * Check that a proposed value to load into the .it_value or
689  * .it_interval part of an interval timer is acceptable, and
690  * fix it to have at least minimal value (i.e. if it is less
691  * than the resolution of the clock, round it up.)
692  */
693 int
694 itimerfix(struct timeval *tv)
695 {
696 
697 	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
698 		return (EINVAL);
699 	if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
700 		tv->tv_usec = tick;
701 	return (0);
702 }
703 
704 /*
705  * Decrement an interval timer by a specified number
706  * of microseconds, which must be less than a second,
707  * i.e. < 1000000.  If the timer expires, then reload
708  * it.  In this case, carry over (usec - old value) to
709  * reduce the value reloaded into the timer so that
710  * the timer does not drift.  This routine assumes
711  * that it is called in a context where the timers
712  * on which it is operating cannot change in value.
713  */
714 int
715 itimerdecr(struct itimerval *itp, int usec)
716 {
717 
718 	if (itp->it_value.tv_usec < usec) {
719 		if (itp->it_value.tv_sec == 0) {
720 			/* expired, and already in next interval */
721 			usec -= itp->it_value.tv_usec;
722 			goto expire;
723 		}
724 		itp->it_value.tv_usec += 1000000;
725 		itp->it_value.tv_sec--;
726 	}
727 	itp->it_value.tv_usec -= usec;
728 	usec = 0;
729 	if (timevalisset(&itp->it_value))
730 		return (1);
731 	/* expired, exactly at end of interval */
732 expire:
733 	if (timevalisset(&itp->it_interval)) {
734 		itp->it_value = itp->it_interval;
735 		itp->it_value.tv_usec -= usec;
736 		if (itp->it_value.tv_usec < 0) {
737 			itp->it_value.tv_usec += 1000000;
738 			itp->it_value.tv_sec--;
739 		}
740 	} else
741 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
742 	return (0);
743 }
744 
745 /*
746  * Add and subtract routines for timevals.
747  * N.B.: subtract routine doesn't deal with
748  * results which are before the beginning,
749  * it just gets very confused in this case.
750  * Caveat emptor.
751  */
752 void
753 timevaladd(struct timeval *t1, const struct timeval *t2)
754 {
755 
756 	t1->tv_sec += t2->tv_sec;
757 	t1->tv_usec += t2->tv_usec;
758 	timevalfix(t1);
759 }
760 
761 void
762 timevalsub(struct timeval *t1, const struct timeval *t2)
763 {
764 
765 	t1->tv_sec -= t2->tv_sec;
766 	t1->tv_usec -= t2->tv_usec;
767 	timevalfix(t1);
768 }
769 
770 static void
771 timevalfix(struct timeval *t1)
772 {
773 
774 	if (t1->tv_usec < 0) {
775 		t1->tv_sec--;
776 		t1->tv_usec += 1000000;
777 	}
778 	if (t1->tv_usec >= 1000000) {
779 		t1->tv_sec++;
780 		t1->tv_usec -= 1000000;
781 	}
782 }
783 
784 /*
785  * ratecheck(): simple time-based rate-limit checking.
786  */
787 int
788 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
789 {
790 	struct timeval tv, delta;
791 	int rv = 0;
792 
793 	getmicrouptime(&tv);		/* NB: 10ms precision */
794 	delta = tv;
795 	timevalsub(&delta, lasttime);
796 
797 	/*
798 	 * check for 0,0 is so that the message will be seen at least once,
799 	 * even if interval is huge.
800 	 */
801 	if (timevalcmp(&delta, mininterval, >=) ||
802 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
803 		*lasttime = tv;
804 		rv = 1;
805 	}
806 
807 	return (rv);
808 }
809 
810 /*
811  * ppsratecheck(): packets (or events) per second limitation.
812  *
813  * Return 0 if the limit is to be enforced (e.g. the caller
814  * should drop a packet because of the rate limitation).
815  *
816  * maxpps of 0 always causes zero to be returned.  maxpps of -1
817  * always causes 1 to be returned; this effectively defeats rate
818  * limiting.
819  *
820  * Note that we maintain the struct timeval for compatibility
821  * with other bsd systems.  We reuse the storage and just monitor
822  * clock ticks for minimal overhead.
823  */
824 int
825 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
826 {
827 	int now;
828 
829 	/*
830 	 * Reset the last time and counter if this is the first call
831 	 * or more than a second has passed since the last update of
832 	 * lasttime.
833 	 */
834 	now = ticks;
835 	if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
836 		lasttime->tv_sec = now;
837 		*curpps = 1;
838 		return (maxpps != 0);
839 	} else {
840 		(*curpps)++;		/* NB: ignore potential overflow */
841 		return (maxpps < 0 || *curpps < maxpps);
842 	}
843 }
844 
845 static void
846 itimer_start(void)
847 {
848 	struct kclock rt_clock = {
849 		.timer_create  = realtimer_create,
850 		.timer_delete  = realtimer_delete,
851 		.timer_settime = realtimer_settime,
852 		.timer_gettime = realtimer_gettime,
853 		.event_hook    = NULL
854 	};
855 
856 	itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
857 		NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
858 	register_posix_clock(CLOCK_REALTIME,  &rt_clock);
859 	register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
860 	p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
861 	p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
862 	p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
863 	EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
864 		(void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
865 	EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
866 		(void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
867 }
868 
869 int
870 register_posix_clock(int clockid, struct kclock *clk)
871 {
872 	if ((unsigned)clockid >= MAX_CLOCKS) {
873 		printf("%s: invalid clockid\n", __func__);
874 		return (0);
875 	}
876 	posix_clocks[clockid] = *clk;
877 	return (1);
878 }
879 
880 static int
881 itimer_init(void *mem, int size, int flags)
882 {
883 	struct itimer *it;
884 
885 	it = (struct itimer *)mem;
886 	mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
887 	return (0);
888 }
889 
890 static void
891 itimer_fini(void *mem, int size)
892 {
893 	struct itimer *it;
894 
895 	it = (struct itimer *)mem;
896 	mtx_destroy(&it->it_mtx);
897 }
898 
899 static void
900 itimer_enter(struct itimer *it)
901 {
902 
903 	mtx_assert(&it->it_mtx, MA_OWNED);
904 	it->it_usecount++;
905 }
906 
907 static void
908 itimer_leave(struct itimer *it)
909 {
910 
911 	mtx_assert(&it->it_mtx, MA_OWNED);
912 	KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
913 
914 	if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
915 		wakeup(it);
916 }
917 
918 #ifndef _SYS_SYSPROTO_H_
919 struct ktimer_create_args {
920 	clockid_t clock_id;
921 	struct sigevent * evp;
922 	int * timerid;
923 };
924 #endif
925 int
926 ktimer_create(struct thread *td, struct ktimer_create_args *uap)
927 {
928 	struct sigevent *evp1, ev;
929 	int id;
930 	int error;
931 
932 	if (uap->evp != NULL) {
933 		error = copyin(uap->evp, &ev, sizeof(ev));
934 		if (error != 0)
935 			return (error);
936 		evp1 = &ev;
937 	} else
938 		evp1 = NULL;
939 
940 	error = kern_timer_create(td, uap->clock_id, evp1, &id, -1);
941 
942 	if (error == 0) {
943 		error = copyout(&id, uap->timerid, sizeof(int));
944 		if (error != 0)
945 			kern_timer_delete(td, id);
946 	}
947 	return (error);
948 }
949 
950 static int
951 kern_timer_create(struct thread *td, clockid_t clock_id,
952 	struct sigevent *evp, int *timerid, int preset_id)
953 {
954 	struct proc *p = td->td_proc;
955 	struct itimer *it;
956 	int id;
957 	int error;
958 
959 	if (clock_id < 0 || clock_id >= MAX_CLOCKS)
960 		return (EINVAL);
961 
962 	if (posix_clocks[clock_id].timer_create == NULL)
963 		return (EINVAL);
964 
965 	if (evp != NULL) {
966 		if (evp->sigev_notify != SIGEV_NONE &&
967 		    evp->sigev_notify != SIGEV_SIGNAL &&
968 		    evp->sigev_notify != SIGEV_THREAD_ID)
969 			return (EINVAL);
970 		if ((evp->sigev_notify == SIGEV_SIGNAL ||
971 		     evp->sigev_notify == SIGEV_THREAD_ID) &&
972 			!_SIG_VALID(evp->sigev_signo))
973 			return (EINVAL);
974 	}
975 
976 	if (p->p_itimers == NULL)
977 		itimers_alloc(p);
978 
979 	it = uma_zalloc(itimer_zone, M_WAITOK);
980 	it->it_flags = 0;
981 	it->it_usecount = 0;
982 	it->it_active = 0;
983 	timespecclear(&it->it_time.it_value);
984 	timespecclear(&it->it_time.it_interval);
985 	it->it_overrun = 0;
986 	it->it_overrun_last = 0;
987 	it->it_clockid = clock_id;
988 	it->it_timerid = -1;
989 	it->it_proc = p;
990 	ksiginfo_init(&it->it_ksi);
991 	it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
992 	error = CLOCK_CALL(clock_id, timer_create, (it));
993 	if (error != 0)
994 		goto out;
995 
996 	PROC_LOCK(p);
997 	if (preset_id != -1) {
998 		KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
999 		id = preset_id;
1000 		if (p->p_itimers->its_timers[id] != NULL) {
1001 			PROC_UNLOCK(p);
1002 			error = 0;
1003 			goto out;
1004 		}
1005 	} else {
1006 		/*
1007 		 * Find a free timer slot, skipping those reserved
1008 		 * for setitimer().
1009 		 */
1010 		for (id = 3; id < TIMER_MAX; id++)
1011 			if (p->p_itimers->its_timers[id] == NULL)
1012 				break;
1013 		if (id == TIMER_MAX) {
1014 			PROC_UNLOCK(p);
1015 			error = EAGAIN;
1016 			goto out;
1017 		}
1018 	}
1019 	it->it_timerid = id;
1020 	p->p_itimers->its_timers[id] = it;
1021 	if (evp != NULL)
1022 		it->it_sigev = *evp;
1023 	else {
1024 		it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1025 		switch (clock_id) {
1026 		default:
1027 		case CLOCK_REALTIME:
1028 			it->it_sigev.sigev_signo = SIGALRM;
1029 			break;
1030 		case CLOCK_VIRTUAL:
1031  			it->it_sigev.sigev_signo = SIGVTALRM;
1032 			break;
1033 		case CLOCK_PROF:
1034 			it->it_sigev.sigev_signo = SIGPROF;
1035 			break;
1036 		}
1037 		it->it_sigev.sigev_value.sival_int = id;
1038 	}
1039 
1040 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1041 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1042 		it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1043 		it->it_ksi.ksi_code = SI_TIMER;
1044 		it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1045 		it->it_ksi.ksi_timerid = id;
1046 	}
1047 	PROC_UNLOCK(p);
1048 	*timerid = id;
1049 	return (0);
1050 
1051 out:
1052 	ITIMER_LOCK(it);
1053 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1054 	ITIMER_UNLOCK(it);
1055 	uma_zfree(itimer_zone, it);
1056 	return (error);
1057 }
1058 
1059 #ifndef _SYS_SYSPROTO_H_
1060 struct ktimer_delete_args {
1061 	int timerid;
1062 };
1063 #endif
1064 int
1065 ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1066 {
1067 	return (kern_timer_delete(td, uap->timerid));
1068 }
1069 
1070 static struct itimer *
1071 itimer_find(struct proc *p, int timerid)
1072 {
1073 	struct itimer *it;
1074 
1075 	PROC_LOCK_ASSERT(p, MA_OWNED);
1076 	if ((p->p_itimers == NULL) ||
1077 	    (timerid < 0) || (timerid >= TIMER_MAX) ||
1078 	    (it = p->p_itimers->its_timers[timerid]) == NULL) {
1079 		return (NULL);
1080 	}
1081 	ITIMER_LOCK(it);
1082 	if ((it->it_flags & ITF_DELETING) != 0) {
1083 		ITIMER_UNLOCK(it);
1084 		it = NULL;
1085 	}
1086 	return (it);
1087 }
1088 
1089 static int
1090 kern_timer_delete(struct thread *td, int timerid)
1091 {
1092 	struct proc *p = td->td_proc;
1093 	struct itimer *it;
1094 
1095 	PROC_LOCK(p);
1096 	it = itimer_find(p, timerid);
1097 	if (it == NULL) {
1098 		PROC_UNLOCK(p);
1099 		return (EINVAL);
1100 	}
1101 	PROC_UNLOCK(p);
1102 
1103 	it->it_flags |= ITF_DELETING;
1104 	while (it->it_usecount > 0) {
1105 		it->it_flags |= ITF_WANTED;
1106 		msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1107 	}
1108 	it->it_flags &= ~ITF_WANTED;
1109 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1110 	ITIMER_UNLOCK(it);
1111 
1112 	PROC_LOCK(p);
1113 	if (KSI_ONQ(&it->it_ksi))
1114 		sigqueue_take(&it->it_ksi);
1115 	p->p_itimers->its_timers[timerid] = NULL;
1116 	PROC_UNLOCK(p);
1117 	uma_zfree(itimer_zone, it);
1118 	return (0);
1119 }
1120 
1121 #ifndef _SYS_SYSPROTO_H_
1122 struct ktimer_settime_args {
1123 	int timerid;
1124 	int flags;
1125 	const struct itimerspec * value;
1126 	struct itimerspec * ovalue;
1127 };
1128 #endif
1129 int
1130 ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1131 {
1132 	struct proc *p = td->td_proc;
1133 	struct itimer *it;
1134 	struct itimerspec val, oval, *ovalp;
1135 	int error;
1136 
1137 	error = copyin(uap->value, &val, sizeof(val));
1138 	if (error != 0)
1139 		return (error);
1140 
1141 	if (uap->ovalue != NULL)
1142 		ovalp = &oval;
1143 	else
1144 		ovalp = NULL;
1145 
1146 	PROC_LOCK(p);
1147 	if (uap->timerid < 3 ||
1148 	    (it = itimer_find(p, uap->timerid)) == NULL) {
1149 		PROC_UNLOCK(p);
1150 		error = EINVAL;
1151 	} else {
1152 		PROC_UNLOCK(p);
1153 		itimer_enter(it);
1154 		error = CLOCK_CALL(it->it_clockid, timer_settime,
1155 				(it, uap->flags, &val, ovalp));
1156 		itimer_leave(it);
1157 		ITIMER_UNLOCK(it);
1158 	}
1159 	if (error == 0 && uap->ovalue != NULL)
1160 		error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1161 	return (error);
1162 }
1163 
1164 #ifndef _SYS_SYSPROTO_H_
1165 struct ktimer_gettime_args {
1166 	int timerid;
1167 	struct itimerspec * value;
1168 };
1169 #endif
1170 int
1171 ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1172 {
1173 	struct proc *p = td->td_proc;
1174 	struct itimer *it;
1175 	struct itimerspec val;
1176 	int error;
1177 
1178 	PROC_LOCK(p);
1179 	if (uap->timerid < 3 ||
1180 	   (it = itimer_find(p, uap->timerid)) == NULL) {
1181 		PROC_UNLOCK(p);
1182 		error = EINVAL;
1183 	} else {
1184 		PROC_UNLOCK(p);
1185 		itimer_enter(it);
1186 		error = CLOCK_CALL(it->it_clockid, timer_gettime,
1187 				(it, &val));
1188 		itimer_leave(it);
1189 		ITIMER_UNLOCK(it);
1190 	}
1191 	if (error == 0)
1192 		error = copyout(&val, uap->value, sizeof(val));
1193 	return (error);
1194 }
1195 
1196 #ifndef _SYS_SYSPROTO_H_
1197 struct timer_getoverrun_args {
1198 	int timerid;
1199 };
1200 #endif
1201 int
1202 ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1203 {
1204 	struct proc *p = td->td_proc;
1205 	struct itimer *it;
1206 	int error ;
1207 
1208 	PROC_LOCK(p);
1209 	if (uap->timerid < 3 ||
1210 	    (it = itimer_find(p, uap->timerid)) == NULL) {
1211 		PROC_UNLOCK(p);
1212 		error = EINVAL;
1213 	} else {
1214 		td->td_retval[0] = it->it_overrun_last;
1215 		ITIMER_UNLOCK(it);
1216 		PROC_UNLOCK(p);
1217 		error = 0;
1218 	}
1219 	return (error);
1220 }
1221 
1222 static int
1223 realtimer_create(struct itimer *it)
1224 {
1225 	callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1226 	return (0);
1227 }
1228 
1229 static int
1230 realtimer_delete(struct itimer *it)
1231 {
1232 	mtx_assert(&it->it_mtx, MA_OWNED);
1233 
1234 	/*
1235 	 * clear timer's value and interval to tell realtimer_expire
1236 	 * to not rearm the timer.
1237 	 */
1238 	timespecclear(&it->it_time.it_value);
1239 	timespecclear(&it->it_time.it_interval);
1240 	ITIMER_UNLOCK(it);
1241 	callout_drain(&it->it_callout);
1242 	ITIMER_LOCK(it);
1243 	return (0);
1244 }
1245 
1246 static int
1247 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1248 {
1249 	struct timespec cts;
1250 
1251 	mtx_assert(&it->it_mtx, MA_OWNED);
1252 
1253 	realtimer_clocktime(it->it_clockid, &cts);
1254 	*ovalue = it->it_time;
1255 	if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1256 		timespecsub(&ovalue->it_value, &cts);
1257 		if (ovalue->it_value.tv_sec < 0 ||
1258 		    (ovalue->it_value.tv_sec == 0 &&
1259 		     ovalue->it_value.tv_nsec == 0)) {
1260 			ovalue->it_value.tv_sec  = 0;
1261 			ovalue->it_value.tv_nsec = 1;
1262 		}
1263 	}
1264 	return (0);
1265 }
1266 
1267 static int
1268 realtimer_settime(struct itimer *it, int flags,
1269 	struct itimerspec *value, struct itimerspec *ovalue)
1270 {
1271 	struct timespec cts, ts;
1272 	struct timeval tv;
1273 	struct itimerspec val;
1274 
1275 	mtx_assert(&it->it_mtx, MA_OWNED);
1276 
1277 	val = *value;
1278 	if (itimespecfix(&val.it_value))
1279 		return (EINVAL);
1280 
1281 	if (timespecisset(&val.it_value)) {
1282 		if (itimespecfix(&val.it_interval))
1283 			return (EINVAL);
1284 	} else {
1285 		timespecclear(&val.it_interval);
1286 	}
1287 
1288 	if (ovalue != NULL)
1289 		realtimer_gettime(it, ovalue);
1290 
1291 	it->it_time = val;
1292 	if (timespecisset(&val.it_value)) {
1293 		realtimer_clocktime(it->it_clockid, &cts);
1294 		ts = val.it_value;
1295 		if ((flags & TIMER_ABSTIME) == 0) {
1296 			/* Convert to absolute time. */
1297 			timespecadd(&it->it_time.it_value, &cts);
1298 		} else {
1299 			timespecsub(&ts, &cts);
1300 			/*
1301 			 * We don't care if ts is negative, tztohz will
1302 			 * fix it.
1303 			 */
1304 		}
1305 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1306 		callout_reset(&it->it_callout, tvtohz(&tv),
1307 			realtimer_expire, it);
1308 	} else {
1309 		callout_stop(&it->it_callout);
1310 	}
1311 
1312 	return (0);
1313 }
1314 
1315 static void
1316 realtimer_clocktime(clockid_t id, struct timespec *ts)
1317 {
1318 	if (id == CLOCK_REALTIME)
1319 		getnanotime(ts);
1320 	else	/* CLOCK_MONOTONIC */
1321 		getnanouptime(ts);
1322 }
1323 
1324 int
1325 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1326 {
1327 	struct itimer *it;
1328 
1329 	PROC_LOCK_ASSERT(p, MA_OWNED);
1330 	it = itimer_find(p, timerid);
1331 	if (it != NULL) {
1332 		ksi->ksi_overrun = it->it_overrun;
1333 		it->it_overrun_last = it->it_overrun;
1334 		it->it_overrun = 0;
1335 		ITIMER_UNLOCK(it);
1336 		return (0);
1337 	}
1338 	return (EINVAL);
1339 }
1340 
1341 int
1342 itimespecfix(struct timespec *ts)
1343 {
1344 
1345 	if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1346 		return (EINVAL);
1347 	if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1348 		ts->tv_nsec = tick * 1000;
1349 	return (0);
1350 }
1351 
1352 /* Timeout callback for realtime timer */
1353 static void
1354 realtimer_expire(void *arg)
1355 {
1356 	struct timespec cts, ts;
1357 	struct timeval tv;
1358 	struct itimer *it;
1359 
1360 	it = (struct itimer *)arg;
1361 
1362 	realtimer_clocktime(it->it_clockid, &cts);
1363 	/* Only fire if time is reached. */
1364 	if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1365 		if (timespecisset(&it->it_time.it_interval)) {
1366 			timespecadd(&it->it_time.it_value,
1367 				    &it->it_time.it_interval);
1368 			while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1369 				if (it->it_overrun < INT_MAX)
1370 					it->it_overrun++;
1371 				else
1372 					it->it_ksi.ksi_errno = ERANGE;
1373 				timespecadd(&it->it_time.it_value,
1374 					    &it->it_time.it_interval);
1375 			}
1376 		} else {
1377 			/* single shot timer ? */
1378 			timespecclear(&it->it_time.it_value);
1379 		}
1380 		if (timespecisset(&it->it_time.it_value)) {
1381 			ts = it->it_time.it_value;
1382 			timespecsub(&ts, &cts);
1383 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1384 			callout_reset(&it->it_callout, tvtohz(&tv),
1385 				 realtimer_expire, it);
1386 		}
1387 		itimer_enter(it);
1388 		ITIMER_UNLOCK(it);
1389 		itimer_fire(it);
1390 		ITIMER_LOCK(it);
1391 		itimer_leave(it);
1392 	} else if (timespecisset(&it->it_time.it_value)) {
1393 		ts = it->it_time.it_value;
1394 		timespecsub(&ts, &cts);
1395 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1396 		callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1397  			it);
1398 	}
1399 }
1400 
1401 void
1402 itimer_fire(struct itimer *it)
1403 {
1404 	struct proc *p = it->it_proc;
1405 	int ret;
1406 
1407 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1408 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1409 		PROC_LOCK(p);
1410 		if (!KSI_ONQ(&it->it_ksi)) {
1411 			it->it_ksi.ksi_errno = 0;
1412 			ret = psignal_event(p, &it->it_sigev, &it->it_ksi);
1413 			if (__predict_false(ret != 0)) {
1414 				it->it_overrun++;
1415 				/*
1416 				 * Broken userland code, thread went
1417 				 * away, disarm the timer.
1418 				 */
1419 				if (ret == ESRCH) {
1420 					ITIMER_LOCK(it);
1421 					timespecclear(&it->it_time.it_value);
1422 					timespecclear(&it->it_time.it_interval);
1423 					callout_stop(&it->it_callout);
1424 					ITIMER_UNLOCK(it);
1425 				}
1426 			}
1427 		} else {
1428 			if (it->it_overrun < INT_MAX)
1429 				it->it_overrun++;
1430 			else
1431 				it->it_ksi.ksi_errno = ERANGE;
1432 		}
1433 		PROC_UNLOCK(p);
1434 	}
1435 }
1436 
1437 static void
1438 itimers_alloc(struct proc *p)
1439 {
1440 	struct itimers *its;
1441 	int i;
1442 
1443 	its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1444 	LIST_INIT(&its->its_virtual);
1445 	LIST_INIT(&its->its_prof);
1446 	TAILQ_INIT(&its->its_worklist);
1447 	for (i = 0; i < TIMER_MAX; i++)
1448 		its->its_timers[i] = NULL;
1449 	PROC_LOCK(p);
1450 	if (p->p_itimers == NULL) {
1451 		p->p_itimers = its;
1452 		PROC_UNLOCK(p);
1453 	}
1454 	else {
1455 		PROC_UNLOCK(p);
1456 		free(its, M_SUBPROC);
1457 	}
1458 }
1459 
1460 static void
1461 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1462 {
1463 	itimers_event_hook_exit(arg, p);
1464 }
1465 
1466 /* Clean up timers when some process events are being triggered. */
1467 static void
1468 itimers_event_hook_exit(void *arg, struct proc *p)
1469 {
1470 	struct itimers *its;
1471 	struct itimer *it;
1472 	int event = (int)(intptr_t)arg;
1473 	int i;
1474 
1475 	if (p->p_itimers != NULL) {
1476 		its = p->p_itimers;
1477 		for (i = 0; i < MAX_CLOCKS; ++i) {
1478 			if (posix_clocks[i].event_hook != NULL)
1479 				CLOCK_CALL(i, event_hook, (p, i, event));
1480 		}
1481 		/*
1482 		 * According to susv3, XSI interval timers should be inherited
1483 		 * by new image.
1484 		 */
1485 		if (event == ITIMER_EV_EXEC)
1486 			i = 3;
1487 		else if (event == ITIMER_EV_EXIT)
1488 			i = 0;
1489 		else
1490 			panic("unhandled event");
1491 		for (; i < TIMER_MAX; ++i) {
1492 			if ((it = its->its_timers[i]) != NULL)
1493 				kern_timer_delete(curthread, i);
1494 		}
1495 		if (its->its_timers[0] == NULL &&
1496 		    its->its_timers[1] == NULL &&
1497 		    its->its_timers[2] == NULL) {
1498 			free(its, M_SUBPROC);
1499 			p->p_itimers = NULL;
1500 		}
1501 	}
1502 }
1503