xref: /freebsd/sys/kern/kern_time.c (revision 6ae1554a5d9b318f8ad53ccc39fa5a961403da73)
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/sleepqueue.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/posix4.h>
53 #include <sys/time.h>
54 #include <sys/timers.h>
55 #include <sys/timetc.h>
56 #include <sys/vnode.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_extern.h>
60 
61 #define MAX_CLOCKS 	(CLOCK_MONOTONIC+1)
62 #define CPUCLOCK_BIT		0x80000000
63 #define CPUCLOCK_PROCESS_BIT	0x40000000
64 #define CPUCLOCK_ID_MASK	(~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
65 #define MAKE_THREAD_CPUCLOCK(tid)	(CPUCLOCK_BIT|(tid))
66 #define MAKE_PROCESS_CPUCLOCK(pid)	\
67 	(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
68 
69 static struct kclock	posix_clocks[MAX_CLOCKS];
70 static uma_zone_t	itimer_zone = NULL;
71 
72 /*
73  * Time of day and interval timer support.
74  *
75  * These routines provide the kernel entry points to get and set
76  * the time-of-day and per-process interval timers.  Subroutines
77  * here provide support for adding and subtracting timeval structures
78  * and decrementing interval timers, optionally reloading the interval
79  * timers when they expire.
80  */
81 
82 static int	settime(struct thread *, struct timeval *);
83 static void	timevalfix(struct timeval *);
84 
85 static void	itimer_start(void);
86 static int	itimer_init(void *, int, int);
87 static void	itimer_fini(void *, int);
88 static void	itimer_enter(struct itimer *);
89 static void	itimer_leave(struct itimer *);
90 static struct itimer *itimer_find(struct proc *, int);
91 static void	itimers_alloc(struct proc *);
92 static void	itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
93 static void	itimers_event_hook_exit(void *arg, struct proc *p);
94 static int	realtimer_create(struct itimer *);
95 static int	realtimer_gettime(struct itimer *, struct itimerspec *);
96 static int	realtimer_settime(struct itimer *, int,
97 			struct itimerspec *, struct itimerspec *);
98 static int	realtimer_delete(struct itimer *);
99 static void	realtimer_clocktime(clockid_t, struct timespec *);
100 static void	realtimer_expire(void *);
101 
102 int		register_posix_clock(int, struct kclock *);
103 void		itimer_fire(struct itimer *it);
104 int		itimespecfix(struct timespec *ts);
105 
106 #define CLOCK_CALL(clock, call, arglist)		\
107 	((*posix_clocks[clock].call) arglist)
108 
109 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
110 
111 
112 static int
113 settime(struct thread *td, struct timeval *tv)
114 {
115 	struct timeval delta, tv1, tv2;
116 	static struct timeval maxtime, laststep;
117 	struct timespec ts;
118 	int s;
119 
120 	s = splclock();
121 	microtime(&tv1);
122 	delta = *tv;
123 	timevalsub(&delta, &tv1);
124 
125 	/*
126 	 * If the system is secure, we do not allow the time to be
127 	 * set to a value earlier than 1 second less than the highest
128 	 * time we have yet seen. The worst a miscreant can do in
129 	 * this circumstance is "freeze" time. He couldn't go
130 	 * back to the past.
131 	 *
132 	 * We similarly do not allow the clock to be stepped more
133 	 * than one second, nor more than once per second. This allows
134 	 * a miscreant to make the clock march double-time, but no worse.
135 	 */
136 	if (securelevel_gt(td->td_ucred, 1) != 0) {
137 		if (delta.tv_sec < 0 || delta.tv_usec < 0) {
138 			/*
139 			 * Update maxtime to latest time we've seen.
140 			 */
141 			if (tv1.tv_sec > maxtime.tv_sec)
142 				maxtime = tv1;
143 			tv2 = *tv;
144 			timevalsub(&tv2, &maxtime);
145 			if (tv2.tv_sec < -1) {
146 				tv->tv_sec = maxtime.tv_sec - 1;
147 				printf("Time adjustment clamped to -1 second\n");
148 			}
149 		} else {
150 			if (tv1.tv_sec == laststep.tv_sec) {
151 				splx(s);
152 				return (EPERM);
153 			}
154 			if (delta.tv_sec > 1) {
155 				tv->tv_sec = tv1.tv_sec + 1;
156 				printf("Time adjustment clamped to +1 second\n");
157 			}
158 			laststep = *tv;
159 		}
160 	}
161 
162 	ts.tv_sec = tv->tv_sec;
163 	ts.tv_nsec = tv->tv_usec * 1000;
164 	mtx_lock(&Giant);
165 	tc_setclock(&ts);
166 	resettodr();
167 	mtx_unlock(&Giant);
168 	return (0);
169 }
170 
171 #ifndef _SYS_SYSPROTO_H_
172 struct clock_getcpuclockid2_args {
173 	id_t id;
174 	int which,
175 	clockid_t *clock_id;
176 };
177 #endif
178 /* ARGSUSED */
179 int
180 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
181 {
182 	clockid_t clk_id;
183 	int error;
184 
185 	error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
186 	if (error == 0)
187 		error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
188 	return (error);
189 }
190 
191 int
192 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
193     clockid_t *clk_id)
194 {
195 	struct proc *p;
196 	pid_t pid;
197 	lwpid_t tid;
198 	int error;
199 
200 	switch (which) {
201 	case CPUCLOCK_WHICH_PID:
202 		if (id != 0) {
203 			error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
204 			if (error != 0)
205 				return (error);
206 			PROC_UNLOCK(p);
207 			pid = id;
208 		} else {
209 			pid = td->td_proc->p_pid;
210 		}
211 		*clk_id = MAKE_PROCESS_CPUCLOCK(pid);
212 		return (0);
213 	case CPUCLOCK_WHICH_TID:
214 		tid = id == 0 ? td->td_tid : id;
215 		*clk_id = MAKE_THREAD_CPUCLOCK(tid);
216 		return (0);
217 	default:
218 		return (EINVAL);
219 	}
220 }
221 
222 #ifndef _SYS_SYSPROTO_H_
223 struct clock_gettime_args {
224 	clockid_t clock_id;
225 	struct	timespec *tp;
226 };
227 #endif
228 /* ARGSUSED */
229 int
230 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
231 {
232 	struct timespec ats;
233 	int error;
234 
235 	error = kern_clock_gettime(td, uap->clock_id, &ats);
236 	if (error == 0)
237 		error = copyout(&ats, uap->tp, sizeof(ats));
238 
239 	return (error);
240 }
241 
242 static inline void
243 cputick2timespec(uint64_t runtime, struct timespec *ats)
244 {
245 	runtime = cputick2usec(runtime);
246 	ats->tv_sec = runtime / 1000000;
247 	ats->tv_nsec = runtime % 1000000 * 1000;
248 }
249 
250 static void
251 get_thread_cputime(struct thread *targettd, struct timespec *ats)
252 {
253 	uint64_t runtime, curtime, switchtime;
254 
255 	if (targettd == NULL) { /* current thread */
256 		critical_enter();
257 		switchtime = PCPU_GET(switchtime);
258 		curtime = cpu_ticks();
259 		runtime = curthread->td_runtime;
260 		critical_exit();
261 		runtime += curtime - switchtime;
262 	} else {
263 		thread_lock(targettd);
264 		runtime = targettd->td_runtime;
265 		thread_unlock(targettd);
266 	}
267 	cputick2timespec(runtime, ats);
268 }
269 
270 static void
271 get_process_cputime(struct proc *targetp, struct timespec *ats)
272 {
273 	uint64_t runtime;
274 	struct rusage ru;
275 
276 	PROC_STATLOCK(targetp);
277 	rufetch(targetp, &ru);
278 	runtime = targetp->p_rux.rux_runtime;
279 	PROC_STATUNLOCK(targetp);
280 	cputick2timespec(runtime, ats);
281 }
282 
283 static int
284 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
285 {
286 	struct proc *p, *p2;
287 	struct thread *td2;
288 	lwpid_t tid;
289 	pid_t pid;
290 	int error;
291 
292 	p = td->td_proc;
293 	if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
294 		tid = clock_id & CPUCLOCK_ID_MASK;
295 		td2 = tdfind(tid, p->p_pid);
296 		if (td2 == NULL)
297 			return (EINVAL);
298 		get_thread_cputime(td2, ats);
299 		PROC_UNLOCK(td2->td_proc);
300 	} else {
301 		pid = clock_id & CPUCLOCK_ID_MASK;
302 		error = pget(pid, PGET_CANSEE, &p2);
303 		if (error != 0)
304 			return (EINVAL);
305 		get_process_cputime(p2, ats);
306 		PROC_UNLOCK(p2);
307 	}
308 	return (0);
309 }
310 
311 int
312 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
313 {
314 	struct timeval sys, user;
315 	struct proc *p;
316 
317 	p = td->td_proc;
318 	switch (clock_id) {
319 	case CLOCK_REALTIME:		/* Default to precise. */
320 	case CLOCK_REALTIME_PRECISE:
321 		nanotime(ats);
322 		break;
323 	case CLOCK_REALTIME_FAST:
324 		getnanotime(ats);
325 		break;
326 	case CLOCK_VIRTUAL:
327 		PROC_LOCK(p);
328 		PROC_STATLOCK(p);
329 		calcru(p, &user, &sys);
330 		PROC_STATUNLOCK(p);
331 		PROC_UNLOCK(p);
332 		TIMEVAL_TO_TIMESPEC(&user, ats);
333 		break;
334 	case CLOCK_PROF:
335 		PROC_LOCK(p);
336 		PROC_STATLOCK(p);
337 		calcru(p, &user, &sys);
338 		PROC_STATUNLOCK(p);
339 		PROC_UNLOCK(p);
340 		timevaladd(&user, &sys);
341 		TIMEVAL_TO_TIMESPEC(&user, ats);
342 		break;
343 	case CLOCK_MONOTONIC:		/* Default to precise. */
344 	case CLOCK_MONOTONIC_PRECISE:
345 	case CLOCK_UPTIME:
346 	case CLOCK_UPTIME_PRECISE:
347 		nanouptime(ats);
348 		break;
349 	case CLOCK_UPTIME_FAST:
350 	case CLOCK_MONOTONIC_FAST:
351 		getnanouptime(ats);
352 		break;
353 	case CLOCK_SECOND:
354 		ats->tv_sec = time_second;
355 		ats->tv_nsec = 0;
356 		break;
357 	case CLOCK_THREAD_CPUTIME_ID:
358 		get_thread_cputime(NULL, ats);
359 		break;
360 	case CLOCK_PROCESS_CPUTIME_ID:
361 		PROC_LOCK(p);
362 		get_process_cputime(p, ats);
363 		PROC_UNLOCK(p);
364 		break;
365 	default:
366 		if ((int)clock_id >= 0)
367 			return (EINVAL);
368 		return (get_cputime(td, clock_id, ats));
369 	}
370 	return (0);
371 }
372 
373 #ifndef _SYS_SYSPROTO_H_
374 struct clock_settime_args {
375 	clockid_t clock_id;
376 	const struct	timespec *tp;
377 };
378 #endif
379 /* ARGSUSED */
380 int
381 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
382 {
383 	struct timespec ats;
384 	int error;
385 
386 	if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
387 		return (error);
388 	return (kern_clock_settime(td, uap->clock_id, &ats));
389 }
390 
391 int
392 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
393 {
394 	struct timeval atv;
395 	int error;
396 
397 	if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
398 		return (error);
399 	if (clock_id != CLOCK_REALTIME)
400 		return (EINVAL);
401 	if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
402 		return (EINVAL);
403 	/* XXX Don't convert nsec->usec and back */
404 	TIMESPEC_TO_TIMEVAL(&atv, ats);
405 	error = settime(td, &atv);
406 	return (error);
407 }
408 
409 #ifndef _SYS_SYSPROTO_H_
410 struct clock_getres_args {
411 	clockid_t clock_id;
412 	struct	timespec *tp;
413 };
414 #endif
415 int
416 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
417 {
418 	struct timespec ts;
419 	int error;
420 
421 	if (uap->tp == NULL)
422 		return (0);
423 
424 	error = kern_clock_getres(td, uap->clock_id, &ts);
425 	if (error == 0)
426 		error = copyout(&ts, uap->tp, sizeof(ts));
427 	return (error);
428 }
429 
430 int
431 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
432 {
433 
434 	ts->tv_sec = 0;
435 	switch (clock_id) {
436 	case CLOCK_REALTIME:
437 	case CLOCK_REALTIME_FAST:
438 	case CLOCK_REALTIME_PRECISE:
439 	case CLOCK_MONOTONIC:
440 	case CLOCK_MONOTONIC_FAST:
441 	case CLOCK_MONOTONIC_PRECISE:
442 	case CLOCK_UPTIME:
443 	case CLOCK_UPTIME_FAST:
444 	case CLOCK_UPTIME_PRECISE:
445 		/*
446 		 * Round up the result of the division cheaply by adding 1.
447 		 * Rounding up is especially important if rounding down
448 		 * would give 0.  Perfect rounding is unimportant.
449 		 */
450 		ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
451 		break;
452 	case CLOCK_VIRTUAL:
453 	case CLOCK_PROF:
454 		/* Accurately round up here because we can do so cheaply. */
455 		ts->tv_nsec = (1000000000 + hz - 1) / hz;
456 		break;
457 	case CLOCK_SECOND:
458 		ts->tv_sec = 1;
459 		ts->tv_nsec = 0;
460 		break;
461 	case CLOCK_THREAD_CPUTIME_ID:
462 	case CLOCK_PROCESS_CPUTIME_ID:
463 	cputime:
464 		/* sync with cputick2usec */
465 		ts->tv_nsec = 1000000 / cpu_tickrate();
466 		if (ts->tv_nsec == 0)
467 			ts->tv_nsec = 1000;
468 		break;
469 	default:
470 		if ((int)clock_id < 0)
471 			goto cputime;
472 		return (EINVAL);
473 	}
474 	return (0);
475 }
476 
477 static uint8_t nanowait[MAXCPU];
478 
479 int
480 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
481 {
482 	struct timespec ts;
483 	sbintime_t sbt, sbtt, prec, tmp;
484 	time_t over;
485 	int error;
486 
487 	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
488 		return (EINVAL);
489 	if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
490 		return (0);
491 	ts = *rqt;
492 	if (ts.tv_sec > INT32_MAX / 2) {
493 		over = ts.tv_sec - INT32_MAX / 2;
494 		ts.tv_sec -= over;
495 	} else
496 		over = 0;
497 	tmp = tstosbt(ts);
498 	prec = tmp;
499 	prec >>= tc_precexp;
500 	if (TIMESEL(&sbt, tmp))
501 		sbt += tc_tick_sbt;
502 	sbt += tmp;
503 	error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
504 	    sbt, prec, C_ABSOLUTE);
505 	if (error != EWOULDBLOCK) {
506 		if (error == ERESTART)
507 			error = EINTR;
508 		TIMESEL(&sbtt, tmp);
509 		if (rmt != NULL) {
510 			ts = sbttots(sbt - sbtt);
511 			ts.tv_sec += over;
512 			if (ts.tv_sec < 0)
513 				timespecclear(&ts);
514 			*rmt = ts;
515 		}
516 		if (sbtt >= sbt)
517 			return (0);
518 		return (error);
519 	}
520 	return (0);
521 }
522 
523 #ifndef _SYS_SYSPROTO_H_
524 struct nanosleep_args {
525 	struct	timespec *rqtp;
526 	struct	timespec *rmtp;
527 };
528 #endif
529 /* ARGSUSED */
530 int
531 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
532 {
533 	struct timespec rmt, rqt;
534 	int error;
535 
536 	error = copyin(uap->rqtp, &rqt, sizeof(rqt));
537 	if (error)
538 		return (error);
539 
540 	if (uap->rmtp &&
541 	    !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE))
542 			return (EFAULT);
543 	error = kern_nanosleep(td, &rqt, &rmt);
544 	if (error && uap->rmtp) {
545 		int error2;
546 
547 		error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
548 		if (error2)
549 			error = error2;
550 	}
551 	return (error);
552 }
553 
554 #ifndef _SYS_SYSPROTO_H_
555 struct gettimeofday_args {
556 	struct	timeval *tp;
557 	struct	timezone *tzp;
558 };
559 #endif
560 /* ARGSUSED */
561 int
562 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
563 {
564 	struct timeval atv;
565 	struct timezone rtz;
566 	int error = 0;
567 
568 	if (uap->tp) {
569 		microtime(&atv);
570 		error = copyout(&atv, uap->tp, sizeof (atv));
571 	}
572 	if (error == 0 && uap->tzp != NULL) {
573 		rtz.tz_minuteswest = tz_minuteswest;
574 		rtz.tz_dsttime = tz_dsttime;
575 		error = copyout(&rtz, uap->tzp, sizeof (rtz));
576 	}
577 	return (error);
578 }
579 
580 #ifndef _SYS_SYSPROTO_H_
581 struct settimeofday_args {
582 	struct	timeval *tv;
583 	struct	timezone *tzp;
584 };
585 #endif
586 /* ARGSUSED */
587 int
588 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
589 {
590 	struct timeval atv, *tvp;
591 	struct timezone atz, *tzp;
592 	int error;
593 
594 	if (uap->tv) {
595 		error = copyin(uap->tv, &atv, sizeof(atv));
596 		if (error)
597 			return (error);
598 		tvp = &atv;
599 	} else
600 		tvp = NULL;
601 	if (uap->tzp) {
602 		error = copyin(uap->tzp, &atz, sizeof(atz));
603 		if (error)
604 			return (error);
605 		tzp = &atz;
606 	} else
607 		tzp = NULL;
608 	return (kern_settimeofday(td, tvp, tzp));
609 }
610 
611 int
612 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
613 {
614 	int error;
615 
616 	error = priv_check(td, PRIV_SETTIMEOFDAY);
617 	if (error)
618 		return (error);
619 	/* Verify all parameters before changing time. */
620 	if (tv) {
621 		if (tv->tv_usec < 0 || tv->tv_usec >= 1000000)
622 			return (EINVAL);
623 		error = settime(td, tv);
624 	}
625 	if (tzp && error == 0) {
626 		tz_minuteswest = tzp->tz_minuteswest;
627 		tz_dsttime = tzp->tz_dsttime;
628 	}
629 	return (error);
630 }
631 
632 /*
633  * Get value of an interval timer.  The process virtual and profiling virtual
634  * time timers are kept in the p_stats area, since they can be swapped out.
635  * These are kept internally in the way they are specified externally: in
636  * time until they expire.
637  *
638  * The real time interval timer is kept in the process table slot for the
639  * process, and its value (it_value) is kept as an absolute time rather than
640  * as a delta, so that it is easy to keep periodic real-time signals from
641  * drifting.
642  *
643  * Virtual time timers are processed in the hardclock() routine of
644  * kern_clock.c.  The real time timer is processed by a timeout routine,
645  * called from the softclock() routine.  Since a callout may be delayed in
646  * real time due to interrupt processing in the system, it is possible for
647  * the real time timeout routine (realitexpire, given below), to be delayed
648  * in real time past when it is supposed to occur.  It does not suffice,
649  * therefore, to reload the real timer .it_value from the real time timers
650  * .it_interval.  Rather, we compute the next time in absolute time the timer
651  * should go off.
652  */
653 #ifndef _SYS_SYSPROTO_H_
654 struct getitimer_args {
655 	u_int	which;
656 	struct	itimerval *itv;
657 };
658 #endif
659 int
660 sys_getitimer(struct thread *td, struct getitimer_args *uap)
661 {
662 	struct itimerval aitv;
663 	int error;
664 
665 	error = kern_getitimer(td, uap->which, &aitv);
666 	if (error != 0)
667 		return (error);
668 	return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
669 }
670 
671 int
672 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
673 {
674 	struct proc *p = td->td_proc;
675 	struct timeval ctv;
676 
677 	if (which > ITIMER_PROF)
678 		return (EINVAL);
679 
680 	if (which == ITIMER_REAL) {
681 		/*
682 		 * Convert from absolute to relative time in .it_value
683 		 * part of real time timer.  If time for real time timer
684 		 * has passed return 0, else return difference between
685 		 * current time and time for the timer to go off.
686 		 */
687 		PROC_LOCK(p);
688 		*aitv = p->p_realtimer;
689 		PROC_UNLOCK(p);
690 		if (timevalisset(&aitv->it_value)) {
691 			microuptime(&ctv);
692 			if (timevalcmp(&aitv->it_value, &ctv, <))
693 				timevalclear(&aitv->it_value);
694 			else
695 				timevalsub(&aitv->it_value, &ctv);
696 		}
697 	} else {
698 		PROC_ITIMLOCK(p);
699 		*aitv = p->p_stats->p_timer[which];
700 		PROC_ITIMUNLOCK(p);
701 	}
702 	return (0);
703 }
704 
705 #ifndef _SYS_SYSPROTO_H_
706 struct setitimer_args {
707 	u_int	which;
708 	struct	itimerval *itv, *oitv;
709 };
710 #endif
711 int
712 sys_setitimer(struct thread *td, struct setitimer_args *uap)
713 {
714 	struct itimerval aitv, oitv;
715 	int error;
716 
717 	if (uap->itv == NULL) {
718 		uap->itv = uap->oitv;
719 		return (sys_getitimer(td, (struct getitimer_args *)uap));
720 	}
721 
722 	if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
723 		return (error);
724 	error = kern_setitimer(td, uap->which, &aitv, &oitv);
725 	if (error != 0 || uap->oitv == NULL)
726 		return (error);
727 	return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
728 }
729 
730 int
731 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
732     struct itimerval *oitv)
733 {
734 	struct proc *p = td->td_proc;
735 	struct timeval ctv;
736 	sbintime_t sbt, pr;
737 
738 	if (aitv == NULL)
739 		return (kern_getitimer(td, which, oitv));
740 
741 	if (which > ITIMER_PROF)
742 		return (EINVAL);
743 	if (itimerfix(&aitv->it_value) ||
744 	    aitv->it_value.tv_sec > INT32_MAX / 2)
745 		return (EINVAL);
746 	if (!timevalisset(&aitv->it_value))
747 		timevalclear(&aitv->it_interval);
748 	else if (itimerfix(&aitv->it_interval) ||
749 	    aitv->it_interval.tv_sec > INT32_MAX / 2)
750 		return (EINVAL);
751 
752 	if (which == ITIMER_REAL) {
753 		PROC_LOCK(p);
754 		if (timevalisset(&p->p_realtimer.it_value))
755 			callout_stop(&p->p_itcallout);
756 		microuptime(&ctv);
757 		if (timevalisset(&aitv->it_value)) {
758 			pr = tvtosbt(aitv->it_value) >> tc_precexp;
759 			timevaladd(&aitv->it_value, &ctv);
760 			sbt = tvtosbt(aitv->it_value);
761 			callout_reset_sbt(&p->p_itcallout, sbt, pr,
762 			    realitexpire, p, C_ABSOLUTE);
763 		}
764 		*oitv = p->p_realtimer;
765 		p->p_realtimer = *aitv;
766 		PROC_UNLOCK(p);
767 		if (timevalisset(&oitv->it_value)) {
768 			if (timevalcmp(&oitv->it_value, &ctv, <))
769 				timevalclear(&oitv->it_value);
770 			else
771 				timevalsub(&oitv->it_value, &ctv);
772 		}
773 	} else {
774 		if (aitv->it_interval.tv_sec == 0 &&
775 		    aitv->it_interval.tv_usec != 0 &&
776 		    aitv->it_interval.tv_usec < tick)
777 			aitv->it_interval.tv_usec = tick;
778 		if (aitv->it_value.tv_sec == 0 &&
779 		    aitv->it_value.tv_usec != 0 &&
780 		    aitv->it_value.tv_usec < tick)
781 			aitv->it_value.tv_usec = tick;
782 		PROC_ITIMLOCK(p);
783 		*oitv = p->p_stats->p_timer[which];
784 		p->p_stats->p_timer[which] = *aitv;
785 		PROC_ITIMUNLOCK(p);
786 	}
787 	return (0);
788 }
789 
790 /*
791  * Real interval timer expired:
792  * send process whose timer expired an alarm signal.
793  * If time is not set up to reload, then just return.
794  * Else compute next time timer should go off which is > current time.
795  * This is where delay in processing this timeout causes multiple
796  * SIGALRM calls to be compressed into one.
797  * tvtohz() always adds 1 to allow for the time until the next clock
798  * interrupt being strictly less than 1 clock tick, but we don't want
799  * that here since we want to appear to be in sync with the clock
800  * interrupt even when we're delayed.
801  */
802 void
803 realitexpire(void *arg)
804 {
805 	struct proc *p;
806 	struct timeval ctv;
807 	sbintime_t isbt;
808 
809 	p = (struct proc *)arg;
810 	kern_psignal(p, SIGALRM);
811 	if (!timevalisset(&p->p_realtimer.it_interval)) {
812 		timevalclear(&p->p_realtimer.it_value);
813 		if (p->p_flag & P_WEXIT)
814 			wakeup(&p->p_itcallout);
815 		return;
816 	}
817 	isbt = tvtosbt(p->p_realtimer.it_interval);
818 	if (isbt >= sbt_timethreshold)
819 		getmicrouptime(&ctv);
820 	else
821 		microuptime(&ctv);
822 	do {
823 		timevaladd(&p->p_realtimer.it_value,
824 		    &p->p_realtimer.it_interval);
825 	} while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
826 	callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
827 	    isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE);
828 }
829 
830 /*
831  * Check that a proposed value to load into the .it_value or
832  * .it_interval part of an interval timer is acceptable, and
833  * fix it to have at least minimal value (i.e. if it is less
834  * than the resolution of the clock, round it up.)
835  */
836 int
837 itimerfix(struct timeval *tv)
838 {
839 
840 	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
841 		return (EINVAL);
842 	if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
843 	    tv->tv_usec < (u_int)tick / 16)
844 		tv->tv_usec = (u_int)tick / 16;
845 	return (0);
846 }
847 
848 /*
849  * Decrement an interval timer by a specified number
850  * of microseconds, which must be less than a second,
851  * i.e. < 1000000.  If the timer expires, then reload
852  * it.  In this case, carry over (usec - old value) to
853  * reduce the value reloaded into the timer so that
854  * the timer does not drift.  This routine assumes
855  * that it is called in a context where the timers
856  * on which it is operating cannot change in value.
857  */
858 int
859 itimerdecr(struct itimerval *itp, int usec)
860 {
861 
862 	if (itp->it_value.tv_usec < usec) {
863 		if (itp->it_value.tv_sec == 0) {
864 			/* expired, and already in next interval */
865 			usec -= itp->it_value.tv_usec;
866 			goto expire;
867 		}
868 		itp->it_value.tv_usec += 1000000;
869 		itp->it_value.tv_sec--;
870 	}
871 	itp->it_value.tv_usec -= usec;
872 	usec = 0;
873 	if (timevalisset(&itp->it_value))
874 		return (1);
875 	/* expired, exactly at end of interval */
876 expire:
877 	if (timevalisset(&itp->it_interval)) {
878 		itp->it_value = itp->it_interval;
879 		itp->it_value.tv_usec -= usec;
880 		if (itp->it_value.tv_usec < 0) {
881 			itp->it_value.tv_usec += 1000000;
882 			itp->it_value.tv_sec--;
883 		}
884 	} else
885 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
886 	return (0);
887 }
888 
889 /*
890  * Add and subtract routines for timevals.
891  * N.B.: subtract routine doesn't deal with
892  * results which are before the beginning,
893  * it just gets very confused in this case.
894  * Caveat emptor.
895  */
896 void
897 timevaladd(struct timeval *t1, const struct timeval *t2)
898 {
899 
900 	t1->tv_sec += t2->tv_sec;
901 	t1->tv_usec += t2->tv_usec;
902 	timevalfix(t1);
903 }
904 
905 void
906 timevalsub(struct timeval *t1, const struct timeval *t2)
907 {
908 
909 	t1->tv_sec -= t2->tv_sec;
910 	t1->tv_usec -= t2->tv_usec;
911 	timevalfix(t1);
912 }
913 
914 static void
915 timevalfix(struct timeval *t1)
916 {
917 
918 	if (t1->tv_usec < 0) {
919 		t1->tv_sec--;
920 		t1->tv_usec += 1000000;
921 	}
922 	if (t1->tv_usec >= 1000000) {
923 		t1->tv_sec++;
924 		t1->tv_usec -= 1000000;
925 	}
926 }
927 
928 /*
929  * ratecheck(): simple time-based rate-limit checking.
930  */
931 int
932 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
933 {
934 	struct timeval tv, delta;
935 	int rv = 0;
936 
937 	getmicrouptime(&tv);		/* NB: 10ms precision */
938 	delta = tv;
939 	timevalsub(&delta, lasttime);
940 
941 	/*
942 	 * check for 0,0 is so that the message will be seen at least once,
943 	 * even if interval is huge.
944 	 */
945 	if (timevalcmp(&delta, mininterval, >=) ||
946 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
947 		*lasttime = tv;
948 		rv = 1;
949 	}
950 
951 	return (rv);
952 }
953 
954 /*
955  * ppsratecheck(): packets (or events) per second limitation.
956  *
957  * Return 0 if the limit is to be enforced (e.g. the caller
958  * should drop a packet because of the rate limitation).
959  *
960  * maxpps of 0 always causes zero to be returned.  maxpps of -1
961  * always causes 1 to be returned; this effectively defeats rate
962  * limiting.
963  *
964  * Note that we maintain the struct timeval for compatibility
965  * with other bsd systems.  We reuse the storage and just monitor
966  * clock ticks for minimal overhead.
967  */
968 int
969 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
970 {
971 	int now;
972 
973 	/*
974 	 * Reset the last time and counter if this is the first call
975 	 * or more than a second has passed since the last update of
976 	 * lasttime.
977 	 */
978 	now = ticks;
979 	if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
980 		lasttime->tv_sec = now;
981 		*curpps = 1;
982 		return (maxpps != 0);
983 	} else {
984 		(*curpps)++;		/* NB: ignore potential overflow */
985 		return (maxpps < 0 || *curpps <= maxpps);
986 	}
987 }
988 
989 static void
990 itimer_start(void)
991 {
992 	struct kclock rt_clock = {
993 		.timer_create  = realtimer_create,
994 		.timer_delete  = realtimer_delete,
995 		.timer_settime = realtimer_settime,
996 		.timer_gettime = realtimer_gettime,
997 		.event_hook    = NULL
998 	};
999 
1000 	itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1001 		NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1002 	register_posix_clock(CLOCK_REALTIME,  &rt_clock);
1003 	register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1004 	p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1005 	p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1006 	p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1007 	EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
1008 		(void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
1009 	EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
1010 		(void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
1011 }
1012 
1013 int
1014 register_posix_clock(int clockid, struct kclock *clk)
1015 {
1016 	if ((unsigned)clockid >= MAX_CLOCKS) {
1017 		printf("%s: invalid clockid\n", __func__);
1018 		return (0);
1019 	}
1020 	posix_clocks[clockid] = *clk;
1021 	return (1);
1022 }
1023 
1024 static int
1025 itimer_init(void *mem, int size, int flags)
1026 {
1027 	struct itimer *it;
1028 
1029 	it = (struct itimer *)mem;
1030 	mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1031 	return (0);
1032 }
1033 
1034 static void
1035 itimer_fini(void *mem, int size)
1036 {
1037 	struct itimer *it;
1038 
1039 	it = (struct itimer *)mem;
1040 	mtx_destroy(&it->it_mtx);
1041 }
1042 
1043 static void
1044 itimer_enter(struct itimer *it)
1045 {
1046 
1047 	mtx_assert(&it->it_mtx, MA_OWNED);
1048 	it->it_usecount++;
1049 }
1050 
1051 static void
1052 itimer_leave(struct itimer *it)
1053 {
1054 
1055 	mtx_assert(&it->it_mtx, MA_OWNED);
1056 	KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1057 
1058 	if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1059 		wakeup(it);
1060 }
1061 
1062 #ifndef _SYS_SYSPROTO_H_
1063 struct ktimer_create_args {
1064 	clockid_t clock_id;
1065 	struct sigevent * evp;
1066 	int * timerid;
1067 };
1068 #endif
1069 int
1070 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1071 {
1072 	struct sigevent *evp, ev;
1073 	int id;
1074 	int error;
1075 
1076 	if (uap->evp == NULL) {
1077 		evp = NULL;
1078 	} else {
1079 		error = copyin(uap->evp, &ev, sizeof(ev));
1080 		if (error != 0)
1081 			return (error);
1082 		evp = &ev;
1083 	}
1084 	error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1085 	if (error == 0) {
1086 		error = copyout(&id, uap->timerid, sizeof(int));
1087 		if (error != 0)
1088 			kern_ktimer_delete(td, id);
1089 	}
1090 	return (error);
1091 }
1092 
1093 int
1094 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1095     int *timerid, int preset_id)
1096 {
1097 	struct proc *p = td->td_proc;
1098 	struct itimer *it;
1099 	int id;
1100 	int error;
1101 
1102 	if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1103 		return (EINVAL);
1104 
1105 	if (posix_clocks[clock_id].timer_create == NULL)
1106 		return (EINVAL);
1107 
1108 	if (evp != NULL) {
1109 		if (evp->sigev_notify != SIGEV_NONE &&
1110 		    evp->sigev_notify != SIGEV_SIGNAL &&
1111 		    evp->sigev_notify != SIGEV_THREAD_ID)
1112 			return (EINVAL);
1113 		if ((evp->sigev_notify == SIGEV_SIGNAL ||
1114 		     evp->sigev_notify == SIGEV_THREAD_ID) &&
1115 			!_SIG_VALID(evp->sigev_signo))
1116 			return (EINVAL);
1117 	}
1118 
1119 	if (p->p_itimers == NULL)
1120 		itimers_alloc(p);
1121 
1122 	it = uma_zalloc(itimer_zone, M_WAITOK);
1123 	it->it_flags = 0;
1124 	it->it_usecount = 0;
1125 	it->it_active = 0;
1126 	timespecclear(&it->it_time.it_value);
1127 	timespecclear(&it->it_time.it_interval);
1128 	it->it_overrun = 0;
1129 	it->it_overrun_last = 0;
1130 	it->it_clockid = clock_id;
1131 	it->it_timerid = -1;
1132 	it->it_proc = p;
1133 	ksiginfo_init(&it->it_ksi);
1134 	it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1135 	error = CLOCK_CALL(clock_id, timer_create, (it));
1136 	if (error != 0)
1137 		goto out;
1138 
1139 	PROC_LOCK(p);
1140 	if (preset_id != -1) {
1141 		KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1142 		id = preset_id;
1143 		if (p->p_itimers->its_timers[id] != NULL) {
1144 			PROC_UNLOCK(p);
1145 			error = 0;
1146 			goto out;
1147 		}
1148 	} else {
1149 		/*
1150 		 * Find a free timer slot, skipping those reserved
1151 		 * for setitimer().
1152 		 */
1153 		for (id = 3; id < TIMER_MAX; id++)
1154 			if (p->p_itimers->its_timers[id] == NULL)
1155 				break;
1156 		if (id == TIMER_MAX) {
1157 			PROC_UNLOCK(p);
1158 			error = EAGAIN;
1159 			goto out;
1160 		}
1161 	}
1162 	it->it_timerid = id;
1163 	p->p_itimers->its_timers[id] = it;
1164 	if (evp != NULL)
1165 		it->it_sigev = *evp;
1166 	else {
1167 		it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1168 		switch (clock_id) {
1169 		default:
1170 		case CLOCK_REALTIME:
1171 			it->it_sigev.sigev_signo = SIGALRM;
1172 			break;
1173 		case CLOCK_VIRTUAL:
1174  			it->it_sigev.sigev_signo = SIGVTALRM;
1175 			break;
1176 		case CLOCK_PROF:
1177 			it->it_sigev.sigev_signo = SIGPROF;
1178 			break;
1179 		}
1180 		it->it_sigev.sigev_value.sival_int = id;
1181 	}
1182 
1183 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1184 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1185 		it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1186 		it->it_ksi.ksi_code = SI_TIMER;
1187 		it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1188 		it->it_ksi.ksi_timerid = id;
1189 	}
1190 	PROC_UNLOCK(p);
1191 	*timerid = id;
1192 	return (0);
1193 
1194 out:
1195 	ITIMER_LOCK(it);
1196 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1197 	ITIMER_UNLOCK(it);
1198 	uma_zfree(itimer_zone, it);
1199 	return (error);
1200 }
1201 
1202 #ifndef _SYS_SYSPROTO_H_
1203 struct ktimer_delete_args {
1204 	int timerid;
1205 };
1206 #endif
1207 int
1208 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1209 {
1210 
1211 	return (kern_ktimer_delete(td, uap->timerid));
1212 }
1213 
1214 static struct itimer *
1215 itimer_find(struct proc *p, int timerid)
1216 {
1217 	struct itimer *it;
1218 
1219 	PROC_LOCK_ASSERT(p, MA_OWNED);
1220 	if ((p->p_itimers == NULL) ||
1221 	    (timerid < 0) || (timerid >= TIMER_MAX) ||
1222 	    (it = p->p_itimers->its_timers[timerid]) == NULL) {
1223 		return (NULL);
1224 	}
1225 	ITIMER_LOCK(it);
1226 	if ((it->it_flags & ITF_DELETING) != 0) {
1227 		ITIMER_UNLOCK(it);
1228 		it = NULL;
1229 	}
1230 	return (it);
1231 }
1232 
1233 int
1234 kern_ktimer_delete(struct thread *td, int timerid)
1235 {
1236 	struct proc *p = td->td_proc;
1237 	struct itimer *it;
1238 
1239 	PROC_LOCK(p);
1240 	it = itimer_find(p, timerid);
1241 	if (it == NULL) {
1242 		PROC_UNLOCK(p);
1243 		return (EINVAL);
1244 	}
1245 	PROC_UNLOCK(p);
1246 
1247 	it->it_flags |= ITF_DELETING;
1248 	while (it->it_usecount > 0) {
1249 		it->it_flags |= ITF_WANTED;
1250 		msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1251 	}
1252 	it->it_flags &= ~ITF_WANTED;
1253 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1254 	ITIMER_UNLOCK(it);
1255 
1256 	PROC_LOCK(p);
1257 	if (KSI_ONQ(&it->it_ksi))
1258 		sigqueue_take(&it->it_ksi);
1259 	p->p_itimers->its_timers[timerid] = NULL;
1260 	PROC_UNLOCK(p);
1261 	uma_zfree(itimer_zone, it);
1262 	return (0);
1263 }
1264 
1265 #ifndef _SYS_SYSPROTO_H_
1266 struct ktimer_settime_args {
1267 	int timerid;
1268 	int flags;
1269 	const struct itimerspec * value;
1270 	struct itimerspec * ovalue;
1271 };
1272 #endif
1273 int
1274 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1275 {
1276 	struct itimerspec val, oval, *ovalp;
1277 	int error;
1278 
1279 	error = copyin(uap->value, &val, sizeof(val));
1280 	if (error != 0)
1281 		return (error);
1282 	ovalp = uap->ovalue != NULL ? &oval : NULL;
1283 	error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1284 	if (error == 0 && uap->ovalue != NULL)
1285 		error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1286 	return (error);
1287 }
1288 
1289 int
1290 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1291     struct itimerspec *val, struct itimerspec *oval)
1292 {
1293 	struct proc *p;
1294 	struct itimer *it;
1295 	int error;
1296 
1297 	p = td->td_proc;
1298 	PROC_LOCK(p);
1299 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1300 		PROC_UNLOCK(p);
1301 		error = EINVAL;
1302 	} else {
1303 		PROC_UNLOCK(p);
1304 		itimer_enter(it);
1305 		error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1306 		    flags, val, oval));
1307 		itimer_leave(it);
1308 		ITIMER_UNLOCK(it);
1309 	}
1310 	return (error);
1311 }
1312 
1313 #ifndef _SYS_SYSPROTO_H_
1314 struct ktimer_gettime_args {
1315 	int timerid;
1316 	struct itimerspec * value;
1317 };
1318 #endif
1319 int
1320 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1321 {
1322 	struct itimerspec val;
1323 	int error;
1324 
1325 	error = kern_ktimer_gettime(td, uap->timerid, &val);
1326 	if (error == 0)
1327 		error = copyout(&val, uap->value, sizeof(val));
1328 	return (error);
1329 }
1330 
1331 int
1332 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1333 {
1334 	struct proc *p;
1335 	struct itimer *it;
1336 	int error;
1337 
1338 	p = td->td_proc;
1339 	PROC_LOCK(p);
1340 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1341 		PROC_UNLOCK(p);
1342 		error = EINVAL;
1343 	} else {
1344 		PROC_UNLOCK(p);
1345 		itimer_enter(it);
1346 		error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1347 		itimer_leave(it);
1348 		ITIMER_UNLOCK(it);
1349 	}
1350 	return (error);
1351 }
1352 
1353 #ifndef _SYS_SYSPROTO_H_
1354 struct timer_getoverrun_args {
1355 	int timerid;
1356 };
1357 #endif
1358 int
1359 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1360 {
1361 
1362 	return (kern_ktimer_getoverrun(td, uap->timerid));
1363 }
1364 
1365 int
1366 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1367 {
1368 	struct proc *p = td->td_proc;
1369 	struct itimer *it;
1370 	int error ;
1371 
1372 	PROC_LOCK(p);
1373 	if (timer_id < 3 ||
1374 	    (it = itimer_find(p, timer_id)) == NULL) {
1375 		PROC_UNLOCK(p);
1376 		error = EINVAL;
1377 	} else {
1378 		td->td_retval[0] = it->it_overrun_last;
1379 		ITIMER_UNLOCK(it);
1380 		PROC_UNLOCK(p);
1381 		error = 0;
1382 	}
1383 	return (error);
1384 }
1385 
1386 static int
1387 realtimer_create(struct itimer *it)
1388 {
1389 	callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1390 	return (0);
1391 }
1392 
1393 static int
1394 realtimer_delete(struct itimer *it)
1395 {
1396 	mtx_assert(&it->it_mtx, MA_OWNED);
1397 
1398 	/*
1399 	 * clear timer's value and interval to tell realtimer_expire
1400 	 * to not rearm the timer.
1401 	 */
1402 	timespecclear(&it->it_time.it_value);
1403 	timespecclear(&it->it_time.it_interval);
1404 	ITIMER_UNLOCK(it);
1405 	callout_drain(&it->it_callout);
1406 	ITIMER_LOCK(it);
1407 	return (0);
1408 }
1409 
1410 static int
1411 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1412 {
1413 	struct timespec cts;
1414 
1415 	mtx_assert(&it->it_mtx, MA_OWNED);
1416 
1417 	realtimer_clocktime(it->it_clockid, &cts);
1418 	*ovalue = it->it_time;
1419 	if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1420 		timespecsub(&ovalue->it_value, &cts);
1421 		if (ovalue->it_value.tv_sec < 0 ||
1422 		    (ovalue->it_value.tv_sec == 0 &&
1423 		     ovalue->it_value.tv_nsec == 0)) {
1424 			ovalue->it_value.tv_sec  = 0;
1425 			ovalue->it_value.tv_nsec = 1;
1426 		}
1427 	}
1428 	return (0);
1429 }
1430 
1431 static int
1432 realtimer_settime(struct itimer *it, int flags,
1433 	struct itimerspec *value, struct itimerspec *ovalue)
1434 {
1435 	struct timespec cts, ts;
1436 	struct timeval tv;
1437 	struct itimerspec val;
1438 
1439 	mtx_assert(&it->it_mtx, MA_OWNED);
1440 
1441 	val = *value;
1442 	if (itimespecfix(&val.it_value))
1443 		return (EINVAL);
1444 
1445 	if (timespecisset(&val.it_value)) {
1446 		if (itimespecfix(&val.it_interval))
1447 			return (EINVAL);
1448 	} else {
1449 		timespecclear(&val.it_interval);
1450 	}
1451 
1452 	if (ovalue != NULL)
1453 		realtimer_gettime(it, ovalue);
1454 
1455 	it->it_time = val;
1456 	if (timespecisset(&val.it_value)) {
1457 		realtimer_clocktime(it->it_clockid, &cts);
1458 		ts = val.it_value;
1459 		if ((flags & TIMER_ABSTIME) == 0) {
1460 			/* Convert to absolute time. */
1461 			timespecadd(&it->it_time.it_value, &cts);
1462 		} else {
1463 			timespecsub(&ts, &cts);
1464 			/*
1465 			 * We don't care if ts is negative, tztohz will
1466 			 * fix it.
1467 			 */
1468 		}
1469 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1470 		callout_reset(&it->it_callout, tvtohz(&tv),
1471 			realtimer_expire, it);
1472 	} else {
1473 		callout_stop(&it->it_callout);
1474 	}
1475 
1476 	return (0);
1477 }
1478 
1479 static void
1480 realtimer_clocktime(clockid_t id, struct timespec *ts)
1481 {
1482 	if (id == CLOCK_REALTIME)
1483 		getnanotime(ts);
1484 	else	/* CLOCK_MONOTONIC */
1485 		getnanouptime(ts);
1486 }
1487 
1488 int
1489 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1490 {
1491 	struct itimer *it;
1492 
1493 	PROC_LOCK_ASSERT(p, MA_OWNED);
1494 	it = itimer_find(p, timerid);
1495 	if (it != NULL) {
1496 		ksi->ksi_overrun = it->it_overrun;
1497 		it->it_overrun_last = it->it_overrun;
1498 		it->it_overrun = 0;
1499 		ITIMER_UNLOCK(it);
1500 		return (0);
1501 	}
1502 	return (EINVAL);
1503 }
1504 
1505 int
1506 itimespecfix(struct timespec *ts)
1507 {
1508 
1509 	if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1510 		return (EINVAL);
1511 	if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1512 		ts->tv_nsec = tick * 1000;
1513 	return (0);
1514 }
1515 
1516 /* Timeout callback for realtime timer */
1517 static void
1518 realtimer_expire(void *arg)
1519 {
1520 	struct timespec cts, ts;
1521 	struct timeval tv;
1522 	struct itimer *it;
1523 
1524 	it = (struct itimer *)arg;
1525 
1526 	realtimer_clocktime(it->it_clockid, &cts);
1527 	/* Only fire if time is reached. */
1528 	if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1529 		if (timespecisset(&it->it_time.it_interval)) {
1530 			timespecadd(&it->it_time.it_value,
1531 				    &it->it_time.it_interval);
1532 			while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1533 				if (it->it_overrun < INT_MAX)
1534 					it->it_overrun++;
1535 				else
1536 					it->it_ksi.ksi_errno = ERANGE;
1537 				timespecadd(&it->it_time.it_value,
1538 					    &it->it_time.it_interval);
1539 			}
1540 		} else {
1541 			/* single shot timer ? */
1542 			timespecclear(&it->it_time.it_value);
1543 		}
1544 		if (timespecisset(&it->it_time.it_value)) {
1545 			ts = it->it_time.it_value;
1546 			timespecsub(&ts, &cts);
1547 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1548 			callout_reset(&it->it_callout, tvtohz(&tv),
1549 				 realtimer_expire, it);
1550 		}
1551 		itimer_enter(it);
1552 		ITIMER_UNLOCK(it);
1553 		itimer_fire(it);
1554 		ITIMER_LOCK(it);
1555 		itimer_leave(it);
1556 	} else if (timespecisset(&it->it_time.it_value)) {
1557 		ts = it->it_time.it_value;
1558 		timespecsub(&ts, &cts);
1559 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1560 		callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1561  			it);
1562 	}
1563 }
1564 
1565 void
1566 itimer_fire(struct itimer *it)
1567 {
1568 	struct proc *p = it->it_proc;
1569 	struct thread *td;
1570 
1571 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1572 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1573 		if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1574 			ITIMER_LOCK(it);
1575 			timespecclear(&it->it_time.it_value);
1576 			timespecclear(&it->it_time.it_interval);
1577 			callout_stop(&it->it_callout);
1578 			ITIMER_UNLOCK(it);
1579 			return;
1580 		}
1581 		if (!KSI_ONQ(&it->it_ksi)) {
1582 			it->it_ksi.ksi_errno = 0;
1583 			ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1584 			tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1585 		} else {
1586 			if (it->it_overrun < INT_MAX)
1587 				it->it_overrun++;
1588 			else
1589 				it->it_ksi.ksi_errno = ERANGE;
1590 		}
1591 		PROC_UNLOCK(p);
1592 	}
1593 }
1594 
1595 static void
1596 itimers_alloc(struct proc *p)
1597 {
1598 	struct itimers *its;
1599 	int i;
1600 
1601 	its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1602 	LIST_INIT(&its->its_virtual);
1603 	LIST_INIT(&its->its_prof);
1604 	TAILQ_INIT(&its->its_worklist);
1605 	for (i = 0; i < TIMER_MAX; i++)
1606 		its->its_timers[i] = NULL;
1607 	PROC_LOCK(p);
1608 	if (p->p_itimers == NULL) {
1609 		p->p_itimers = its;
1610 		PROC_UNLOCK(p);
1611 	}
1612 	else {
1613 		PROC_UNLOCK(p);
1614 		free(its, M_SUBPROC);
1615 	}
1616 }
1617 
1618 static void
1619 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1620 {
1621 	itimers_event_hook_exit(arg, p);
1622 }
1623 
1624 /* Clean up timers when some process events are being triggered. */
1625 static void
1626 itimers_event_hook_exit(void *arg, struct proc *p)
1627 {
1628 	struct itimers *its;
1629 	struct itimer *it;
1630 	int event = (int)(intptr_t)arg;
1631 	int i;
1632 
1633 	if (p->p_itimers != NULL) {
1634 		its = p->p_itimers;
1635 		for (i = 0; i < MAX_CLOCKS; ++i) {
1636 			if (posix_clocks[i].event_hook != NULL)
1637 				CLOCK_CALL(i, event_hook, (p, i, event));
1638 		}
1639 		/*
1640 		 * According to susv3, XSI interval timers should be inherited
1641 		 * by new image.
1642 		 */
1643 		if (event == ITIMER_EV_EXEC)
1644 			i = 3;
1645 		else if (event == ITIMER_EV_EXIT)
1646 			i = 0;
1647 		else
1648 			panic("unhandled event");
1649 		for (; i < TIMER_MAX; ++i) {
1650 			if ((it = its->its_timers[i]) != NULL)
1651 				kern_ktimer_delete(curthread, i);
1652 		}
1653 		if (its->its_timers[0] == NULL &&
1654 		    its->its_timers[1] == NULL &&
1655 		    its->its_timers[2] == NULL) {
1656 			free(its, M_SUBPROC);
1657 			p->p_itimers = NULL;
1658 		}
1659 	}
1660 }
1661