xref: /freebsd/sys/kern/kern_time.c (revision 25ecdc7d52770caf1c9b44b5ec11f468f6b636f3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)kern_time.c	8.1 (Berkeley) 6/10/93
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_ktrace.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/limits.h>
42 #include <sys/clock.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/sysproto.h>
46 #include <sys/resourcevar.h>
47 #include <sys/signalvar.h>
48 #include <sys/kernel.h>
49 #include <sys/sleepqueue.h>
50 #include <sys/syscallsubr.h>
51 #include <sys/sysctl.h>
52 #include <sys/sysent.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/posix4.h>
56 #include <sys/time.h>
57 #include <sys/timers.h>
58 #include <sys/timetc.h>
59 #include <sys/vnode.h>
60 #ifdef KTRACE
61 #include <sys/ktrace.h>
62 #endif
63 
64 #include <vm/vm.h>
65 #include <vm/vm_extern.h>
66 
67 #define MAX_CLOCKS 	(CLOCK_MONOTONIC+1)
68 #define CPUCLOCK_BIT		0x80000000
69 #define CPUCLOCK_PROCESS_BIT	0x40000000
70 #define CPUCLOCK_ID_MASK	(~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
71 #define MAKE_THREAD_CPUCLOCK(tid)	(CPUCLOCK_BIT|(tid))
72 #define MAKE_PROCESS_CPUCLOCK(pid)	\
73 	(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
74 
75 #define NS_PER_SEC	1000000000
76 
77 static struct kclock	posix_clocks[MAX_CLOCKS];
78 static uma_zone_t	itimer_zone = NULL;
79 
80 /*
81  * Time of day and interval timer support.
82  *
83  * These routines provide the kernel entry points to get and set
84  * the time-of-day and per-process interval timers.  Subroutines
85  * here provide support for adding and subtracting timeval structures
86  * and decrementing interval timers, optionally reloading the interval
87  * timers when they expire.
88  */
89 
90 static int	settime(struct thread *, struct timeval *);
91 static void	timevalfix(struct timeval *);
92 static int	user_clock_nanosleep(struct thread *td, clockid_t clock_id,
93 		    int flags, const struct timespec *ua_rqtp,
94 		    struct timespec *ua_rmtp);
95 
96 static void	itimer_start(void);
97 static int	itimer_init(void *, int, int);
98 static void	itimer_fini(void *, int);
99 static void	itimer_enter(struct itimer *);
100 static void	itimer_leave(struct itimer *);
101 static struct itimer *itimer_find(struct proc *, int);
102 static void	itimers_alloc(struct proc *);
103 static int	realtimer_create(struct itimer *);
104 static int	realtimer_gettime(struct itimer *, struct itimerspec *);
105 static int	realtimer_settime(struct itimer *, int,
106 			struct itimerspec *, struct itimerspec *);
107 static int	realtimer_delete(struct itimer *);
108 static void	realtimer_clocktime(clockid_t, struct timespec *);
109 static void	realtimer_expire(void *);
110 static void	realtimer_expire_l(struct itimer *it, bool proc_locked);
111 
112 static int	register_posix_clock(int, const struct kclock *);
113 static void	itimer_fire(struct itimer *it);
114 static int	itimespecfix(struct timespec *ts);
115 
116 #define CLOCK_CALL(clock, call, arglist)		\
117 	((*posix_clocks[clock].call) arglist)
118 
119 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
120 
121 static int
122 settime(struct thread *td, struct timeval *tv)
123 {
124 	struct timeval delta, tv1, tv2;
125 	static struct timeval maxtime, laststep;
126 	struct timespec ts;
127 
128 	microtime(&tv1);
129 	delta = *tv;
130 	timevalsub(&delta, &tv1);
131 
132 	/*
133 	 * If the system is secure, we do not allow the time to be
134 	 * set to a value earlier than 1 second less than the highest
135 	 * time we have yet seen. The worst a miscreant can do in
136 	 * this circumstance is "freeze" time. He couldn't go
137 	 * back to the past.
138 	 *
139 	 * We similarly do not allow the clock to be stepped more
140 	 * than one second, nor more than once per second. This allows
141 	 * a miscreant to make the clock march double-time, but no worse.
142 	 */
143 	if (securelevel_gt(td->td_ucred, 1) != 0) {
144 		if (delta.tv_sec < 0 || delta.tv_usec < 0) {
145 			/*
146 			 * Update maxtime to latest time we've seen.
147 			 */
148 			if (tv1.tv_sec > maxtime.tv_sec)
149 				maxtime = tv1;
150 			tv2 = *tv;
151 			timevalsub(&tv2, &maxtime);
152 			if (tv2.tv_sec < -1) {
153 				tv->tv_sec = maxtime.tv_sec - 1;
154 				printf("Time adjustment clamped to -1 second\n");
155 			}
156 		} else {
157 			if (tv1.tv_sec == laststep.tv_sec)
158 				return (EPERM);
159 			if (delta.tv_sec > 1) {
160 				tv->tv_sec = tv1.tv_sec + 1;
161 				printf("Time adjustment clamped to +1 second\n");
162 			}
163 			laststep = *tv;
164 		}
165 	}
166 
167 	ts.tv_sec = tv->tv_sec;
168 	ts.tv_nsec = tv->tv_usec * 1000;
169 	tc_setclock(&ts);
170 	resettodr();
171 	return (0);
172 }
173 
174 #ifndef _SYS_SYSPROTO_H_
175 struct clock_getcpuclockid2_args {
176 	id_t id;
177 	int which,
178 	clockid_t *clock_id;
179 };
180 #endif
181 /* ARGSUSED */
182 int
183 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
184 {
185 	clockid_t clk_id;
186 	int error;
187 
188 	error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
189 	if (error == 0)
190 		error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
191 	return (error);
192 }
193 
194 int
195 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
196     clockid_t *clk_id)
197 {
198 	struct proc *p;
199 	pid_t pid;
200 	lwpid_t tid;
201 	int error;
202 
203 	switch (which) {
204 	case CPUCLOCK_WHICH_PID:
205 		if (id != 0) {
206 			error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
207 			if (error != 0)
208 				return (error);
209 			PROC_UNLOCK(p);
210 			pid = id;
211 		} else {
212 			pid = td->td_proc->p_pid;
213 		}
214 		*clk_id = MAKE_PROCESS_CPUCLOCK(pid);
215 		return (0);
216 	case CPUCLOCK_WHICH_TID:
217 		tid = id == 0 ? td->td_tid : id;
218 		*clk_id = MAKE_THREAD_CPUCLOCK(tid);
219 		return (0);
220 	default:
221 		return (EINVAL);
222 	}
223 }
224 
225 #ifndef _SYS_SYSPROTO_H_
226 struct clock_gettime_args {
227 	clockid_t clock_id;
228 	struct	timespec *tp;
229 };
230 #endif
231 /* ARGSUSED */
232 int
233 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
234 {
235 	struct timespec ats;
236 	int error;
237 
238 	error = kern_clock_gettime(td, uap->clock_id, &ats);
239 	if (error == 0)
240 		error = copyout(&ats, uap->tp, sizeof(ats));
241 
242 	return (error);
243 }
244 
245 static inline void
246 cputick2timespec(uint64_t runtime, struct timespec *ats)
247 {
248 	runtime = cputick2usec(runtime);
249 	ats->tv_sec = runtime / 1000000;
250 	ats->tv_nsec = runtime % 1000000 * 1000;
251 }
252 
253 void
254 kern_thread_cputime(struct thread *targettd, struct timespec *ats)
255 {
256 	uint64_t runtime, curtime, switchtime;
257 
258 	if (targettd == NULL) { /* current thread */
259 		critical_enter();
260 		switchtime = PCPU_GET(switchtime);
261 		curtime = cpu_ticks();
262 		runtime = curthread->td_runtime;
263 		critical_exit();
264 		runtime += curtime - switchtime;
265 	} else {
266 		PROC_LOCK_ASSERT(targettd->td_proc, MA_OWNED);
267 		thread_lock(targettd);
268 		runtime = targettd->td_runtime;
269 		thread_unlock(targettd);
270 	}
271 	cputick2timespec(runtime, ats);
272 }
273 
274 void
275 kern_process_cputime(struct proc *targetp, struct timespec *ats)
276 {
277 	uint64_t runtime;
278 	struct rusage ru;
279 
280 	PROC_LOCK_ASSERT(targetp, MA_OWNED);
281 	PROC_STATLOCK(targetp);
282 	rufetch(targetp, &ru);
283 	runtime = targetp->p_rux.rux_runtime;
284 	if (curthread->td_proc == targetp)
285 		runtime += cpu_ticks() - PCPU_GET(switchtime);
286 	PROC_STATUNLOCK(targetp);
287 	cputick2timespec(runtime, ats);
288 }
289 
290 static int
291 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
292 {
293 	struct proc *p, *p2;
294 	struct thread *td2;
295 	lwpid_t tid;
296 	pid_t pid;
297 	int error;
298 
299 	p = td->td_proc;
300 	if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
301 		tid = clock_id & CPUCLOCK_ID_MASK;
302 		td2 = tdfind(tid, p->p_pid);
303 		if (td2 == NULL)
304 			return (EINVAL);
305 		kern_thread_cputime(td2, ats);
306 		PROC_UNLOCK(td2->td_proc);
307 	} else {
308 		pid = clock_id & CPUCLOCK_ID_MASK;
309 		error = pget(pid, PGET_CANSEE, &p2);
310 		if (error != 0)
311 			return (EINVAL);
312 		kern_process_cputime(p2, ats);
313 		PROC_UNLOCK(p2);
314 	}
315 	return (0);
316 }
317 
318 int
319 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
320 {
321 	struct timeval sys, user;
322 	struct proc *p;
323 
324 	p = td->td_proc;
325 	switch (clock_id) {
326 	case CLOCK_REALTIME:		/* Default to precise. */
327 	case CLOCK_REALTIME_PRECISE:
328 		nanotime(ats);
329 		break;
330 	case CLOCK_REALTIME_FAST:
331 		getnanotime(ats);
332 		break;
333 	case CLOCK_VIRTUAL:
334 		PROC_LOCK(p);
335 		PROC_STATLOCK(p);
336 		calcru(p, &user, &sys);
337 		PROC_STATUNLOCK(p);
338 		PROC_UNLOCK(p);
339 		TIMEVAL_TO_TIMESPEC(&user, ats);
340 		break;
341 	case CLOCK_PROF:
342 		PROC_LOCK(p);
343 		PROC_STATLOCK(p);
344 		calcru(p, &user, &sys);
345 		PROC_STATUNLOCK(p);
346 		PROC_UNLOCK(p);
347 		timevaladd(&user, &sys);
348 		TIMEVAL_TO_TIMESPEC(&user, ats);
349 		break;
350 	case CLOCK_MONOTONIC:		/* Default to precise. */
351 	case CLOCK_MONOTONIC_PRECISE:
352 	case CLOCK_UPTIME:
353 	case CLOCK_UPTIME_PRECISE:
354 		nanouptime(ats);
355 		break;
356 	case CLOCK_UPTIME_FAST:
357 	case CLOCK_MONOTONIC_FAST:
358 		getnanouptime(ats);
359 		break;
360 	case CLOCK_SECOND:
361 		ats->tv_sec = time_second;
362 		ats->tv_nsec = 0;
363 		break;
364 	case CLOCK_THREAD_CPUTIME_ID:
365 		kern_thread_cputime(NULL, ats);
366 		break;
367 	case CLOCK_PROCESS_CPUTIME_ID:
368 		PROC_LOCK(p);
369 		kern_process_cputime(p, ats);
370 		PROC_UNLOCK(p);
371 		break;
372 	default:
373 		if ((int)clock_id >= 0)
374 			return (EINVAL);
375 		return (get_cputime(td, clock_id, ats));
376 	}
377 	return (0);
378 }
379 
380 #ifndef _SYS_SYSPROTO_H_
381 struct clock_settime_args {
382 	clockid_t clock_id;
383 	const struct	timespec *tp;
384 };
385 #endif
386 /* ARGSUSED */
387 int
388 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
389 {
390 	struct timespec ats;
391 	int error;
392 
393 	if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
394 		return (error);
395 	return (kern_clock_settime(td, uap->clock_id, &ats));
396 }
397 
398 static int allow_insane_settime = 0;
399 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN,
400     &allow_insane_settime, 0,
401     "do not perform possibly restrictive checks on settime(2) args");
402 
403 int
404 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
405 {
406 	struct timeval atv;
407 	int error;
408 
409 	if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
410 		return (error);
411 	if (clock_id != CLOCK_REALTIME)
412 		return (EINVAL);
413 	if (ats->tv_nsec < 0 || ats->tv_nsec >= NS_PER_SEC || ats->tv_sec < 0)
414 		return (EINVAL);
415 	if (!allow_insane_settime &&
416 	    (ats->tv_sec > 8000ULL * 365 * 24 * 60 * 60 ||
417 	    ats->tv_sec < utc_offset()))
418 		return (EINVAL);
419 	/* XXX Don't convert nsec->usec and back */
420 	TIMESPEC_TO_TIMEVAL(&atv, ats);
421 	error = settime(td, &atv);
422 	return (error);
423 }
424 
425 #ifndef _SYS_SYSPROTO_H_
426 struct clock_getres_args {
427 	clockid_t clock_id;
428 	struct	timespec *tp;
429 };
430 #endif
431 int
432 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
433 {
434 	struct timespec ts;
435 	int error;
436 
437 	if (uap->tp == NULL)
438 		return (0);
439 
440 	error = kern_clock_getres(td, uap->clock_id, &ts);
441 	if (error == 0)
442 		error = copyout(&ts, uap->tp, sizeof(ts));
443 	return (error);
444 }
445 
446 int
447 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
448 {
449 
450 	ts->tv_sec = 0;
451 	switch (clock_id) {
452 	case CLOCK_REALTIME:
453 	case CLOCK_REALTIME_FAST:
454 	case CLOCK_REALTIME_PRECISE:
455 	case CLOCK_MONOTONIC:
456 	case CLOCK_MONOTONIC_FAST:
457 	case CLOCK_MONOTONIC_PRECISE:
458 	case CLOCK_UPTIME:
459 	case CLOCK_UPTIME_FAST:
460 	case CLOCK_UPTIME_PRECISE:
461 		/*
462 		 * Round up the result of the division cheaply by adding 1.
463 		 * Rounding up is especially important if rounding down
464 		 * would give 0.  Perfect rounding is unimportant.
465 		 */
466 		ts->tv_nsec = NS_PER_SEC / tc_getfrequency() + 1;
467 		break;
468 	case CLOCK_VIRTUAL:
469 	case CLOCK_PROF:
470 		/* Accurately round up here because we can do so cheaply. */
471 		ts->tv_nsec = howmany(NS_PER_SEC, hz);
472 		break;
473 	case CLOCK_SECOND:
474 		ts->tv_sec = 1;
475 		ts->tv_nsec = 0;
476 		break;
477 	case CLOCK_THREAD_CPUTIME_ID:
478 	case CLOCK_PROCESS_CPUTIME_ID:
479 	cputime:
480 		/* sync with cputick2usec */
481 		ts->tv_nsec = 1000000 / cpu_tickrate();
482 		if (ts->tv_nsec == 0)
483 			ts->tv_nsec = 1000;
484 		break;
485 	default:
486 		if ((int)clock_id < 0)
487 			goto cputime;
488 		return (EINVAL);
489 	}
490 	return (0);
491 }
492 
493 int
494 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
495 {
496 
497 	return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt,
498 	    rmt));
499 }
500 
501 static uint8_t nanowait[MAXCPU];
502 
503 int
504 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
505     const struct timespec *rqt, struct timespec *rmt)
506 {
507 	struct timespec ts, now;
508 	sbintime_t sbt, sbtt, prec, tmp;
509 	time_t over;
510 	int error;
511 	bool is_abs_real;
512 
513 	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= NS_PER_SEC)
514 		return (EINVAL);
515 	if ((flags & ~TIMER_ABSTIME) != 0)
516 		return (EINVAL);
517 	switch (clock_id) {
518 	case CLOCK_REALTIME:
519 	case CLOCK_REALTIME_PRECISE:
520 	case CLOCK_REALTIME_FAST:
521 	case CLOCK_SECOND:
522 		is_abs_real = (flags & TIMER_ABSTIME) != 0;
523 		break;
524 	case CLOCK_MONOTONIC:
525 	case CLOCK_MONOTONIC_PRECISE:
526 	case CLOCK_MONOTONIC_FAST:
527 	case CLOCK_UPTIME:
528 	case CLOCK_UPTIME_PRECISE:
529 	case CLOCK_UPTIME_FAST:
530 		is_abs_real = false;
531 		break;
532 	case CLOCK_VIRTUAL:
533 	case CLOCK_PROF:
534 	case CLOCK_PROCESS_CPUTIME_ID:
535 		return (ENOTSUP);
536 	case CLOCK_THREAD_CPUTIME_ID:
537 	default:
538 		return (EINVAL);
539 	}
540 	do {
541 		ts = *rqt;
542 		if ((flags & TIMER_ABSTIME) != 0) {
543 			if (is_abs_real)
544 				td->td_rtcgen =
545 				    atomic_load_acq_int(&rtc_generation);
546 			error = kern_clock_gettime(td, clock_id, &now);
547 			KASSERT(error == 0, ("kern_clock_gettime: %d", error));
548 			timespecsub(&ts, &now, &ts);
549 		}
550 		if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) {
551 			error = EWOULDBLOCK;
552 			break;
553 		}
554 		if (ts.tv_sec > INT32_MAX / 2) {
555 			over = ts.tv_sec - INT32_MAX / 2;
556 			ts.tv_sec -= over;
557 		} else
558 			over = 0;
559 		tmp = tstosbt(ts);
560 		prec = tmp;
561 		prec >>= tc_precexp;
562 		if (TIMESEL(&sbt, tmp))
563 			sbt += tc_tick_sbt;
564 		sbt += tmp;
565 		error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
566 		    sbt, prec, C_ABSOLUTE);
567 	} while (error == 0 && is_abs_real && td->td_rtcgen == 0);
568 	td->td_rtcgen = 0;
569 	if (error != EWOULDBLOCK) {
570 		if (TIMESEL(&sbtt, tmp))
571 			sbtt += tc_tick_sbt;
572 		if (sbtt >= sbt)
573 			return (0);
574 		if (error == ERESTART)
575 			error = EINTR;
576 		if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) {
577 			ts = sbttots(sbt - sbtt);
578 			ts.tv_sec += over;
579 			if (ts.tv_sec < 0)
580 				timespecclear(&ts);
581 			*rmt = ts;
582 		}
583 		return (error);
584 	}
585 	return (0);
586 }
587 
588 #ifndef _SYS_SYSPROTO_H_
589 struct nanosleep_args {
590 	struct	timespec *rqtp;
591 	struct	timespec *rmtp;
592 };
593 #endif
594 /* ARGSUSED */
595 int
596 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
597 {
598 
599 	return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME,
600 	    uap->rqtp, uap->rmtp));
601 }
602 
603 #ifndef _SYS_SYSPROTO_H_
604 struct clock_nanosleep_args {
605 	clockid_t clock_id;
606 	int 	  flags;
607 	struct	timespec *rqtp;
608 	struct	timespec *rmtp;
609 };
610 #endif
611 /* ARGSUSED */
612 int
613 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap)
614 {
615 	int error;
616 
617 	error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp,
618 	    uap->rmtp);
619 	return (kern_posix_error(td, error));
620 }
621 
622 static int
623 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
624     const struct timespec *ua_rqtp, struct timespec *ua_rmtp)
625 {
626 	struct timespec rmt, rqt;
627 	int error, error2;
628 
629 	error = copyin(ua_rqtp, &rqt, sizeof(rqt));
630 	if (error)
631 		return (error);
632 	error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt);
633 	if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) {
634 		error2 = copyout(&rmt, ua_rmtp, sizeof(rmt));
635 		if (error2 != 0)
636 			error = error2;
637 	}
638 	return (error);
639 }
640 
641 #ifndef _SYS_SYSPROTO_H_
642 struct gettimeofday_args {
643 	struct	timeval *tp;
644 	struct	timezone *tzp;
645 };
646 #endif
647 /* ARGSUSED */
648 int
649 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
650 {
651 	struct timeval atv;
652 	struct timezone rtz;
653 	int error = 0;
654 
655 	if (uap->tp) {
656 		microtime(&atv);
657 		error = copyout(&atv, uap->tp, sizeof (atv));
658 	}
659 	if (error == 0 && uap->tzp != NULL) {
660 		rtz.tz_minuteswest = 0;
661 		rtz.tz_dsttime = 0;
662 		error = copyout(&rtz, uap->tzp, sizeof (rtz));
663 	}
664 	return (error);
665 }
666 
667 #ifndef _SYS_SYSPROTO_H_
668 struct settimeofday_args {
669 	struct	timeval *tv;
670 	struct	timezone *tzp;
671 };
672 #endif
673 /* ARGSUSED */
674 int
675 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
676 {
677 	struct timeval atv, *tvp;
678 	struct timezone atz, *tzp;
679 	int error;
680 
681 	if (uap->tv) {
682 		error = copyin(uap->tv, &atv, sizeof(atv));
683 		if (error)
684 			return (error);
685 		tvp = &atv;
686 	} else
687 		tvp = NULL;
688 	if (uap->tzp) {
689 		error = copyin(uap->tzp, &atz, sizeof(atz));
690 		if (error)
691 			return (error);
692 		tzp = &atz;
693 	} else
694 		tzp = NULL;
695 	return (kern_settimeofday(td, tvp, tzp));
696 }
697 
698 int
699 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
700 {
701 	int error;
702 
703 	error = priv_check(td, PRIV_SETTIMEOFDAY);
704 	if (error)
705 		return (error);
706 	/* Verify all parameters before changing time. */
707 	if (tv) {
708 		if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
709 		    tv->tv_sec < 0)
710 			return (EINVAL);
711 		error = settime(td, tv);
712 	}
713 	return (error);
714 }
715 
716 /*
717  * Get value of an interval timer.  The process virtual and profiling virtual
718  * time timers are kept in the p_stats area, since they can be swapped out.
719  * These are kept internally in the way they are specified externally: in
720  * time until they expire.
721  *
722  * The real time interval timer is kept in the process table slot for the
723  * process, and its value (it_value) is kept as an absolute time rather than
724  * as a delta, so that it is easy to keep periodic real-time signals from
725  * drifting.
726  *
727  * Virtual time timers are processed in the hardclock() routine of
728  * kern_clock.c.  The real time timer is processed by a timeout routine,
729  * called from the softclock() routine.  Since a callout may be delayed in
730  * real time due to interrupt processing in the system, it is possible for
731  * the real time timeout routine (realitexpire, given below), to be delayed
732  * in real time past when it is supposed to occur.  It does not suffice,
733  * therefore, to reload the real timer .it_value from the real time timers
734  * .it_interval.  Rather, we compute the next time in absolute time the timer
735  * should go off.
736  */
737 #ifndef _SYS_SYSPROTO_H_
738 struct getitimer_args {
739 	u_int	which;
740 	struct	itimerval *itv;
741 };
742 #endif
743 int
744 sys_getitimer(struct thread *td, struct getitimer_args *uap)
745 {
746 	struct itimerval aitv;
747 	int error;
748 
749 	error = kern_getitimer(td, uap->which, &aitv);
750 	if (error != 0)
751 		return (error);
752 	return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
753 }
754 
755 int
756 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
757 {
758 	struct proc *p = td->td_proc;
759 	struct timeval ctv;
760 
761 	if (which > ITIMER_PROF)
762 		return (EINVAL);
763 
764 	if (which == ITIMER_REAL) {
765 		/*
766 		 * Convert from absolute to relative time in .it_value
767 		 * part of real time timer.  If time for real time timer
768 		 * has passed return 0, else return difference between
769 		 * current time and time for the timer to go off.
770 		 */
771 		PROC_LOCK(p);
772 		*aitv = p->p_realtimer;
773 		PROC_UNLOCK(p);
774 		if (timevalisset(&aitv->it_value)) {
775 			microuptime(&ctv);
776 			if (timevalcmp(&aitv->it_value, &ctv, <))
777 				timevalclear(&aitv->it_value);
778 			else
779 				timevalsub(&aitv->it_value, &ctv);
780 		}
781 	} else {
782 		PROC_ITIMLOCK(p);
783 		*aitv = p->p_stats->p_timer[which];
784 		PROC_ITIMUNLOCK(p);
785 	}
786 #ifdef KTRACE
787 	if (KTRPOINT(td, KTR_STRUCT))
788 		ktritimerval(aitv);
789 #endif
790 	return (0);
791 }
792 
793 #ifndef _SYS_SYSPROTO_H_
794 struct setitimer_args {
795 	u_int	which;
796 	struct	itimerval *itv, *oitv;
797 };
798 #endif
799 int
800 sys_setitimer(struct thread *td, struct setitimer_args *uap)
801 {
802 	struct itimerval aitv, oitv;
803 	int error;
804 
805 	if (uap->itv == NULL) {
806 		uap->itv = uap->oitv;
807 		return (sys_getitimer(td, (struct getitimer_args *)uap));
808 	}
809 
810 	if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
811 		return (error);
812 	error = kern_setitimer(td, uap->which, &aitv, &oitv);
813 	if (error != 0 || uap->oitv == NULL)
814 		return (error);
815 	return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
816 }
817 
818 int
819 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
820     struct itimerval *oitv)
821 {
822 	struct proc *p = td->td_proc;
823 	struct timeval ctv;
824 	sbintime_t sbt, pr;
825 
826 	if (aitv == NULL)
827 		return (kern_getitimer(td, which, oitv));
828 
829 	if (which > ITIMER_PROF)
830 		return (EINVAL);
831 #ifdef KTRACE
832 	if (KTRPOINT(td, KTR_STRUCT))
833 		ktritimerval(aitv);
834 #endif
835 	if (itimerfix(&aitv->it_value) ||
836 	    aitv->it_value.tv_sec > INT32_MAX / 2)
837 		return (EINVAL);
838 	if (!timevalisset(&aitv->it_value))
839 		timevalclear(&aitv->it_interval);
840 	else if (itimerfix(&aitv->it_interval) ||
841 	    aitv->it_interval.tv_sec > INT32_MAX / 2)
842 		return (EINVAL);
843 
844 	if (which == ITIMER_REAL) {
845 		PROC_LOCK(p);
846 		if (timevalisset(&p->p_realtimer.it_value))
847 			callout_stop(&p->p_itcallout);
848 		microuptime(&ctv);
849 		if (timevalisset(&aitv->it_value)) {
850 			pr = tvtosbt(aitv->it_value) >> tc_precexp;
851 			timevaladd(&aitv->it_value, &ctv);
852 			sbt = tvtosbt(aitv->it_value);
853 			callout_reset_sbt(&p->p_itcallout, sbt, pr,
854 			    realitexpire, p, C_ABSOLUTE);
855 		}
856 		*oitv = p->p_realtimer;
857 		p->p_realtimer = *aitv;
858 		PROC_UNLOCK(p);
859 		if (timevalisset(&oitv->it_value)) {
860 			if (timevalcmp(&oitv->it_value, &ctv, <))
861 				timevalclear(&oitv->it_value);
862 			else
863 				timevalsub(&oitv->it_value, &ctv);
864 		}
865 	} else {
866 		if (aitv->it_interval.tv_sec == 0 &&
867 		    aitv->it_interval.tv_usec != 0 &&
868 		    aitv->it_interval.tv_usec < tick)
869 			aitv->it_interval.tv_usec = tick;
870 		if (aitv->it_value.tv_sec == 0 &&
871 		    aitv->it_value.tv_usec != 0 &&
872 		    aitv->it_value.tv_usec < tick)
873 			aitv->it_value.tv_usec = tick;
874 		PROC_ITIMLOCK(p);
875 		*oitv = p->p_stats->p_timer[which];
876 		p->p_stats->p_timer[which] = *aitv;
877 		PROC_ITIMUNLOCK(p);
878 	}
879 #ifdef KTRACE
880 	if (KTRPOINT(td, KTR_STRUCT))
881 		ktritimerval(oitv);
882 #endif
883 	return (0);
884 }
885 
886 static void
887 realitexpire_reset_callout(struct proc *p, sbintime_t *isbtp)
888 {
889 	sbintime_t prec;
890 
891 	prec = isbtp == NULL ? tvtosbt(p->p_realtimer.it_interval) : *isbtp;
892 	callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
893 	    prec >> tc_precexp, realitexpire, p, C_ABSOLUTE);
894 }
895 
896 void
897 itimer_proc_continue(struct proc *p)
898 {
899 	struct timeval ctv;
900 	struct itimer *it;
901 	int id;
902 
903 	PROC_LOCK_ASSERT(p, MA_OWNED);
904 
905 	if ((p->p_flag2 & P2_ITSTOPPED) != 0) {
906 		p->p_flag2 &= ~P2_ITSTOPPED;
907 		microuptime(&ctv);
908 		if (timevalcmp(&p->p_realtimer.it_value, &ctv, >=))
909 			realitexpire(p);
910 		else
911 			realitexpire_reset_callout(p, NULL);
912 	}
913 
914 	if (p->p_itimers != NULL) {
915 		for (id = 3; id < TIMER_MAX; id++) {
916 			it = p->p_itimers->its_timers[id];
917 			if (it == NULL)
918 				continue;
919 			if ((it->it_flags & ITF_PSTOPPED) != 0) {
920 				ITIMER_LOCK(it);
921 				if ((it->it_flags & ITF_PSTOPPED) != 0) {
922 					it->it_flags &= ~ITF_PSTOPPED;
923 					if ((it->it_flags & ITF_DELETING) == 0)
924 						realtimer_expire_l(it, true);
925 				}
926 				ITIMER_UNLOCK(it);
927 			}
928 		}
929 	}
930 }
931 
932 /*
933  * Real interval timer expired:
934  * send process whose timer expired an alarm signal.
935  * If time is not set up to reload, then just return.
936  * Else compute next time timer should go off which is > current time.
937  * This is where delay in processing this timeout causes multiple
938  * SIGALRM calls to be compressed into one.
939  * tvtohz() always adds 1 to allow for the time until the next clock
940  * interrupt being strictly less than 1 clock tick, but we don't want
941  * that here since we want to appear to be in sync with the clock
942  * interrupt even when we're delayed.
943  */
944 void
945 realitexpire(void *arg)
946 {
947 	struct proc *p;
948 	struct timeval ctv;
949 	sbintime_t isbt;
950 
951 	p = (struct proc *)arg;
952 	kern_psignal(p, SIGALRM);
953 	if (!timevalisset(&p->p_realtimer.it_interval)) {
954 		timevalclear(&p->p_realtimer.it_value);
955 		if (p->p_flag & P_WEXIT)
956 			wakeup(&p->p_itcallout);
957 		return;
958 	}
959 
960 	isbt = tvtosbt(p->p_realtimer.it_interval);
961 	if (isbt >= sbt_timethreshold)
962 		getmicrouptime(&ctv);
963 	else
964 		microuptime(&ctv);
965 	do {
966 		timevaladd(&p->p_realtimer.it_value,
967 		    &p->p_realtimer.it_interval);
968 	} while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
969 
970 	if (P_SHOULDSTOP(p) || P_KILLED(p)) {
971 		p->p_flag2 |= P2_ITSTOPPED;
972 		return;
973 	}
974 
975 	p->p_flag2 &= ~P2_ITSTOPPED;
976 	realitexpire_reset_callout(p, &isbt);
977 }
978 
979 /*
980  * Check that a proposed value to load into the .it_value or
981  * .it_interval part of an interval timer is acceptable, and
982  * fix it to have at least minimal value (i.e. if it is less
983  * than the resolution of the clock, round it up.)
984  */
985 int
986 itimerfix(struct timeval *tv)
987 {
988 
989 	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
990 		return (EINVAL);
991 	if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
992 	    tv->tv_usec < (u_int)tick / 16)
993 		tv->tv_usec = (u_int)tick / 16;
994 	return (0);
995 }
996 
997 /*
998  * Decrement an interval timer by a specified number
999  * of microseconds, which must be less than a second,
1000  * i.e. < 1000000.  If the timer expires, then reload
1001  * it.  In this case, carry over (usec - old value) to
1002  * reduce the value reloaded into the timer so that
1003  * the timer does not drift.  This routine assumes
1004  * that it is called in a context where the timers
1005  * on which it is operating cannot change in value.
1006  */
1007 int
1008 itimerdecr(struct itimerval *itp, int usec)
1009 {
1010 
1011 	if (itp->it_value.tv_usec < usec) {
1012 		if (itp->it_value.tv_sec == 0) {
1013 			/* expired, and already in next interval */
1014 			usec -= itp->it_value.tv_usec;
1015 			goto expire;
1016 		}
1017 		itp->it_value.tv_usec += 1000000;
1018 		itp->it_value.tv_sec--;
1019 	}
1020 	itp->it_value.tv_usec -= usec;
1021 	usec = 0;
1022 	if (timevalisset(&itp->it_value))
1023 		return (1);
1024 	/* expired, exactly at end of interval */
1025 expire:
1026 	if (timevalisset(&itp->it_interval)) {
1027 		itp->it_value = itp->it_interval;
1028 		itp->it_value.tv_usec -= usec;
1029 		if (itp->it_value.tv_usec < 0) {
1030 			itp->it_value.tv_usec += 1000000;
1031 			itp->it_value.tv_sec--;
1032 		}
1033 	} else
1034 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
1035 	return (0);
1036 }
1037 
1038 /*
1039  * Add and subtract routines for timevals.
1040  * N.B.: subtract routine doesn't deal with
1041  * results which are before the beginning,
1042  * it just gets very confused in this case.
1043  * Caveat emptor.
1044  */
1045 void
1046 timevaladd(struct timeval *t1, const struct timeval *t2)
1047 {
1048 
1049 	t1->tv_sec += t2->tv_sec;
1050 	t1->tv_usec += t2->tv_usec;
1051 	timevalfix(t1);
1052 }
1053 
1054 void
1055 timevalsub(struct timeval *t1, const struct timeval *t2)
1056 {
1057 
1058 	t1->tv_sec -= t2->tv_sec;
1059 	t1->tv_usec -= t2->tv_usec;
1060 	timevalfix(t1);
1061 }
1062 
1063 static void
1064 timevalfix(struct timeval *t1)
1065 {
1066 
1067 	if (t1->tv_usec < 0) {
1068 		t1->tv_sec--;
1069 		t1->tv_usec += 1000000;
1070 	}
1071 	if (t1->tv_usec >= 1000000) {
1072 		t1->tv_sec++;
1073 		t1->tv_usec -= 1000000;
1074 	}
1075 }
1076 
1077 /*
1078  * ratecheck(): simple time-based rate-limit checking.
1079  */
1080 int
1081 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1082 {
1083 	struct timeval tv, delta;
1084 	int rv = 0;
1085 
1086 	getmicrouptime(&tv);		/* NB: 10ms precision */
1087 	delta = tv;
1088 	timevalsub(&delta, lasttime);
1089 
1090 	/*
1091 	 * check for 0,0 is so that the message will be seen at least once,
1092 	 * even if interval is huge.
1093 	 */
1094 	if (timevalcmp(&delta, mininterval, >=) ||
1095 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1096 		*lasttime = tv;
1097 		rv = 1;
1098 	}
1099 
1100 	return (rv);
1101 }
1102 
1103 /*
1104  * ppsratecheck(): packets (or events) per second limitation.
1105  *
1106  * Return 0 if the limit is to be enforced (e.g. the caller
1107  * should drop a packet because of the rate limitation).
1108  *
1109  * maxpps of 0 always causes zero to be returned.  maxpps of -1
1110  * always causes 1 to be returned; this effectively defeats rate
1111  * limiting.
1112  *
1113  * Note that we maintain the struct timeval for compatibility
1114  * with other bsd systems.  We reuse the storage and just monitor
1115  * clock ticks for minimal overhead.
1116  */
1117 int
1118 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
1119 {
1120 	int now;
1121 
1122 	/*
1123 	 * Reset the last time and counter if this is the first call
1124 	 * or more than a second has passed since the last update of
1125 	 * lasttime.
1126 	 */
1127 	now = ticks;
1128 	if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1129 		lasttime->tv_sec = now;
1130 		*curpps = 1;
1131 		return (maxpps != 0);
1132 	} else {
1133 		(*curpps)++;		/* NB: ignore potential overflow */
1134 		return (maxpps < 0 || *curpps <= maxpps);
1135 	}
1136 }
1137 
1138 static void
1139 itimer_start(void)
1140 {
1141 	static const struct kclock rt_clock = {
1142 		.timer_create  = realtimer_create,
1143 		.timer_delete  = realtimer_delete,
1144 		.timer_settime = realtimer_settime,
1145 		.timer_gettime = realtimer_gettime,
1146 	};
1147 
1148 	itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1149 		NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1150 	register_posix_clock(CLOCK_REALTIME,  &rt_clock);
1151 	register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1152 	p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1153 	p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1154 	p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1155 }
1156 
1157 static int
1158 register_posix_clock(int clockid, const struct kclock *clk)
1159 {
1160 	if ((unsigned)clockid >= MAX_CLOCKS) {
1161 		printf("%s: invalid clockid\n", __func__);
1162 		return (0);
1163 	}
1164 	posix_clocks[clockid] = *clk;
1165 	return (1);
1166 }
1167 
1168 static int
1169 itimer_init(void *mem, int size, int flags)
1170 {
1171 	struct itimer *it;
1172 
1173 	it = (struct itimer *)mem;
1174 	mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1175 	return (0);
1176 }
1177 
1178 static void
1179 itimer_fini(void *mem, int size)
1180 {
1181 	struct itimer *it;
1182 
1183 	it = (struct itimer *)mem;
1184 	mtx_destroy(&it->it_mtx);
1185 }
1186 
1187 static void
1188 itimer_enter(struct itimer *it)
1189 {
1190 
1191 	mtx_assert(&it->it_mtx, MA_OWNED);
1192 	it->it_usecount++;
1193 }
1194 
1195 static void
1196 itimer_leave(struct itimer *it)
1197 {
1198 
1199 	mtx_assert(&it->it_mtx, MA_OWNED);
1200 	KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1201 
1202 	if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1203 		wakeup(it);
1204 }
1205 
1206 #ifndef _SYS_SYSPROTO_H_
1207 struct ktimer_create_args {
1208 	clockid_t clock_id;
1209 	struct sigevent * evp;
1210 	int * timerid;
1211 };
1212 #endif
1213 int
1214 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1215 {
1216 	struct sigevent *evp, ev;
1217 	int id;
1218 	int error;
1219 
1220 	if (uap->evp == NULL) {
1221 		evp = NULL;
1222 	} else {
1223 		error = copyin(uap->evp, &ev, sizeof(ev));
1224 		if (error != 0)
1225 			return (error);
1226 		evp = &ev;
1227 	}
1228 	error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1229 	if (error == 0) {
1230 		error = copyout(&id, uap->timerid, sizeof(int));
1231 		if (error != 0)
1232 			kern_ktimer_delete(td, id);
1233 	}
1234 	return (error);
1235 }
1236 
1237 int
1238 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1239     int *timerid, int preset_id)
1240 {
1241 	struct proc *p = td->td_proc;
1242 	struct itimer *it;
1243 	int id;
1244 	int error;
1245 
1246 	if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1247 		return (EINVAL);
1248 
1249 	if (posix_clocks[clock_id].timer_create == NULL)
1250 		return (EINVAL);
1251 
1252 	if (evp != NULL) {
1253 		if (evp->sigev_notify != SIGEV_NONE &&
1254 		    evp->sigev_notify != SIGEV_SIGNAL &&
1255 		    evp->sigev_notify != SIGEV_THREAD_ID)
1256 			return (EINVAL);
1257 		if ((evp->sigev_notify == SIGEV_SIGNAL ||
1258 		     evp->sigev_notify == SIGEV_THREAD_ID) &&
1259 			!_SIG_VALID(evp->sigev_signo))
1260 			return (EINVAL);
1261 	}
1262 
1263 	if (p->p_itimers == NULL)
1264 		itimers_alloc(p);
1265 
1266 	it = uma_zalloc(itimer_zone, M_WAITOK);
1267 	it->it_flags = 0;
1268 	it->it_usecount = 0;
1269 	it->it_active = 0;
1270 	timespecclear(&it->it_time.it_value);
1271 	timespecclear(&it->it_time.it_interval);
1272 	it->it_overrun = 0;
1273 	it->it_overrun_last = 0;
1274 	it->it_clockid = clock_id;
1275 	it->it_timerid = -1;
1276 	it->it_proc = p;
1277 	ksiginfo_init(&it->it_ksi);
1278 	it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1279 	error = CLOCK_CALL(clock_id, timer_create, (it));
1280 	if (error != 0)
1281 		goto out;
1282 
1283 	PROC_LOCK(p);
1284 	if (preset_id != -1) {
1285 		KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1286 		id = preset_id;
1287 		if (p->p_itimers->its_timers[id] != NULL) {
1288 			PROC_UNLOCK(p);
1289 			error = 0;
1290 			goto out;
1291 		}
1292 	} else {
1293 		/*
1294 		 * Find a free timer slot, skipping those reserved
1295 		 * for setitimer().
1296 		 */
1297 		for (id = 3; id < TIMER_MAX; id++)
1298 			if (p->p_itimers->its_timers[id] == NULL)
1299 				break;
1300 		if (id == TIMER_MAX) {
1301 			PROC_UNLOCK(p);
1302 			error = EAGAIN;
1303 			goto out;
1304 		}
1305 	}
1306 	it->it_timerid = id;
1307 	p->p_itimers->its_timers[id] = it;
1308 	if (evp != NULL)
1309 		it->it_sigev = *evp;
1310 	else {
1311 		it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1312 		switch (clock_id) {
1313 		default:
1314 		case CLOCK_REALTIME:
1315 			it->it_sigev.sigev_signo = SIGALRM;
1316 			break;
1317 		case CLOCK_VIRTUAL:
1318  			it->it_sigev.sigev_signo = SIGVTALRM;
1319 			break;
1320 		case CLOCK_PROF:
1321 			it->it_sigev.sigev_signo = SIGPROF;
1322 			break;
1323 		}
1324 		it->it_sigev.sigev_value.sival_int = id;
1325 	}
1326 
1327 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1328 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1329 		it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1330 		it->it_ksi.ksi_code = SI_TIMER;
1331 		it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1332 		it->it_ksi.ksi_timerid = id;
1333 	}
1334 	PROC_UNLOCK(p);
1335 	*timerid = id;
1336 	return (0);
1337 
1338 out:
1339 	ITIMER_LOCK(it);
1340 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1341 	ITIMER_UNLOCK(it);
1342 	uma_zfree(itimer_zone, it);
1343 	return (error);
1344 }
1345 
1346 #ifndef _SYS_SYSPROTO_H_
1347 struct ktimer_delete_args {
1348 	int timerid;
1349 };
1350 #endif
1351 int
1352 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1353 {
1354 
1355 	return (kern_ktimer_delete(td, uap->timerid));
1356 }
1357 
1358 static struct itimer *
1359 itimer_find(struct proc *p, int timerid)
1360 {
1361 	struct itimer *it;
1362 
1363 	PROC_LOCK_ASSERT(p, MA_OWNED);
1364 	if ((p->p_itimers == NULL) ||
1365 	    (timerid < 0) || (timerid >= TIMER_MAX) ||
1366 	    (it = p->p_itimers->its_timers[timerid]) == NULL) {
1367 		return (NULL);
1368 	}
1369 	ITIMER_LOCK(it);
1370 	if ((it->it_flags & ITF_DELETING) != 0) {
1371 		ITIMER_UNLOCK(it);
1372 		it = NULL;
1373 	}
1374 	return (it);
1375 }
1376 
1377 int
1378 kern_ktimer_delete(struct thread *td, int timerid)
1379 {
1380 	struct proc *p = td->td_proc;
1381 	struct itimer *it;
1382 
1383 	PROC_LOCK(p);
1384 	it = itimer_find(p, timerid);
1385 	if (it == NULL) {
1386 		PROC_UNLOCK(p);
1387 		return (EINVAL);
1388 	}
1389 	PROC_UNLOCK(p);
1390 
1391 	it->it_flags |= ITF_DELETING;
1392 	while (it->it_usecount > 0) {
1393 		it->it_flags |= ITF_WANTED;
1394 		msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1395 	}
1396 	it->it_flags &= ~ITF_WANTED;
1397 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1398 	ITIMER_UNLOCK(it);
1399 
1400 	PROC_LOCK(p);
1401 	if (KSI_ONQ(&it->it_ksi))
1402 		sigqueue_take(&it->it_ksi);
1403 	p->p_itimers->its_timers[timerid] = NULL;
1404 	PROC_UNLOCK(p);
1405 	uma_zfree(itimer_zone, it);
1406 	return (0);
1407 }
1408 
1409 #ifndef _SYS_SYSPROTO_H_
1410 struct ktimer_settime_args {
1411 	int timerid;
1412 	int flags;
1413 	const struct itimerspec * value;
1414 	struct itimerspec * ovalue;
1415 };
1416 #endif
1417 int
1418 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1419 {
1420 	struct itimerspec val, oval, *ovalp;
1421 	int error;
1422 
1423 	error = copyin(uap->value, &val, sizeof(val));
1424 	if (error != 0)
1425 		return (error);
1426 	ovalp = uap->ovalue != NULL ? &oval : NULL;
1427 	error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1428 	if (error == 0 && uap->ovalue != NULL)
1429 		error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1430 	return (error);
1431 }
1432 
1433 int
1434 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1435     struct itimerspec *val, struct itimerspec *oval)
1436 {
1437 	struct proc *p;
1438 	struct itimer *it;
1439 	int error;
1440 
1441 	p = td->td_proc;
1442 	PROC_LOCK(p);
1443 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1444 		PROC_UNLOCK(p);
1445 		error = EINVAL;
1446 	} else {
1447 		PROC_UNLOCK(p);
1448 		itimer_enter(it);
1449 		error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1450 		    flags, val, oval));
1451 		itimer_leave(it);
1452 		ITIMER_UNLOCK(it);
1453 	}
1454 	return (error);
1455 }
1456 
1457 #ifndef _SYS_SYSPROTO_H_
1458 struct ktimer_gettime_args {
1459 	int timerid;
1460 	struct itimerspec * value;
1461 };
1462 #endif
1463 int
1464 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1465 {
1466 	struct itimerspec val;
1467 	int error;
1468 
1469 	error = kern_ktimer_gettime(td, uap->timerid, &val);
1470 	if (error == 0)
1471 		error = copyout(&val, uap->value, sizeof(val));
1472 	return (error);
1473 }
1474 
1475 int
1476 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1477 {
1478 	struct proc *p;
1479 	struct itimer *it;
1480 	int error;
1481 
1482 	p = td->td_proc;
1483 	PROC_LOCK(p);
1484 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1485 		PROC_UNLOCK(p);
1486 		error = EINVAL;
1487 	} else {
1488 		PROC_UNLOCK(p);
1489 		itimer_enter(it);
1490 		error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1491 		itimer_leave(it);
1492 		ITIMER_UNLOCK(it);
1493 	}
1494 	return (error);
1495 }
1496 
1497 #ifndef _SYS_SYSPROTO_H_
1498 struct timer_getoverrun_args {
1499 	int timerid;
1500 };
1501 #endif
1502 int
1503 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1504 {
1505 
1506 	return (kern_ktimer_getoverrun(td, uap->timerid));
1507 }
1508 
1509 int
1510 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1511 {
1512 	struct proc *p = td->td_proc;
1513 	struct itimer *it;
1514 	int error ;
1515 
1516 	PROC_LOCK(p);
1517 	if (timer_id < 3 ||
1518 	    (it = itimer_find(p, timer_id)) == NULL) {
1519 		PROC_UNLOCK(p);
1520 		error = EINVAL;
1521 	} else {
1522 		td->td_retval[0] = it->it_overrun_last;
1523 		ITIMER_UNLOCK(it);
1524 		PROC_UNLOCK(p);
1525 		error = 0;
1526 	}
1527 	return (error);
1528 }
1529 
1530 static int
1531 realtimer_create(struct itimer *it)
1532 {
1533 	callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1534 	return (0);
1535 }
1536 
1537 static int
1538 realtimer_delete(struct itimer *it)
1539 {
1540 	mtx_assert(&it->it_mtx, MA_OWNED);
1541 
1542 	/*
1543 	 * clear timer's value and interval to tell realtimer_expire
1544 	 * to not rearm the timer.
1545 	 */
1546 	timespecclear(&it->it_time.it_value);
1547 	timespecclear(&it->it_time.it_interval);
1548 	ITIMER_UNLOCK(it);
1549 	callout_drain(&it->it_callout);
1550 	ITIMER_LOCK(it);
1551 	return (0);
1552 }
1553 
1554 static int
1555 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1556 {
1557 	struct timespec cts;
1558 
1559 	mtx_assert(&it->it_mtx, MA_OWNED);
1560 
1561 	realtimer_clocktime(it->it_clockid, &cts);
1562 	*ovalue = it->it_time;
1563 	if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1564 		timespecsub(&ovalue->it_value, &cts, &ovalue->it_value);
1565 		if (ovalue->it_value.tv_sec < 0 ||
1566 		    (ovalue->it_value.tv_sec == 0 &&
1567 		     ovalue->it_value.tv_nsec == 0)) {
1568 			ovalue->it_value.tv_sec  = 0;
1569 			ovalue->it_value.tv_nsec = 1;
1570 		}
1571 	}
1572 	return (0);
1573 }
1574 
1575 static int
1576 realtimer_settime(struct itimer *it, int flags, struct itimerspec *value,
1577     struct itimerspec *ovalue)
1578 {
1579 	struct timespec cts, ts;
1580 	struct timeval tv;
1581 	struct itimerspec val;
1582 
1583 	mtx_assert(&it->it_mtx, MA_OWNED);
1584 
1585 	val = *value;
1586 	if (itimespecfix(&val.it_value))
1587 		return (EINVAL);
1588 
1589 	if (timespecisset(&val.it_value)) {
1590 		if (itimespecfix(&val.it_interval))
1591 			return (EINVAL);
1592 	} else {
1593 		timespecclear(&val.it_interval);
1594 	}
1595 
1596 	if (ovalue != NULL)
1597 		realtimer_gettime(it, ovalue);
1598 
1599 	it->it_time = val;
1600 	if (timespecisset(&val.it_value)) {
1601 		realtimer_clocktime(it->it_clockid, &cts);
1602 		ts = val.it_value;
1603 		if ((flags & TIMER_ABSTIME) == 0) {
1604 			/* Convert to absolute time. */
1605 			timespecadd(&it->it_time.it_value, &cts,
1606 			    &it->it_time.it_value);
1607 		} else {
1608 			timespecsub(&ts, &cts, &ts);
1609 			/*
1610 			 * We don't care if ts is negative, tztohz will
1611 			 * fix it.
1612 			 */
1613 		}
1614 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1615 		callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1616 		    it);
1617 	} else {
1618 		callout_stop(&it->it_callout);
1619 	}
1620 
1621 	return (0);
1622 }
1623 
1624 static void
1625 realtimer_clocktime(clockid_t id, struct timespec *ts)
1626 {
1627 	if (id == CLOCK_REALTIME)
1628 		getnanotime(ts);
1629 	else	/* CLOCK_MONOTONIC */
1630 		getnanouptime(ts);
1631 }
1632 
1633 int
1634 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1635 {
1636 	struct itimer *it;
1637 
1638 	PROC_LOCK_ASSERT(p, MA_OWNED);
1639 	it = itimer_find(p, timerid);
1640 	if (it != NULL) {
1641 		ksi->ksi_overrun = it->it_overrun;
1642 		it->it_overrun_last = it->it_overrun;
1643 		it->it_overrun = 0;
1644 		ITIMER_UNLOCK(it);
1645 		return (0);
1646 	}
1647 	return (EINVAL);
1648 }
1649 
1650 static int
1651 itimespecfix(struct timespec *ts)
1652 {
1653 
1654 	if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= NS_PER_SEC)
1655 		return (EINVAL);
1656 	if ((UINT64_MAX - ts->tv_nsec) / NS_PER_SEC < ts->tv_sec)
1657 		return (EINVAL);
1658 	if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1659 		ts->tv_nsec = tick * 1000;
1660 	return (0);
1661 }
1662 
1663 #define	timespectons(tsp)			\
1664 	((uint64_t)(tsp)->tv_sec * NS_PER_SEC + (tsp)->tv_nsec)
1665 #define	timespecfromns(ns) (struct timespec){	\
1666 	.tv_sec = (ns) / NS_PER_SEC,		\
1667 	.tv_nsec = (ns) % NS_PER_SEC		\
1668 }
1669 
1670 static void
1671 realtimer_expire_l(struct itimer *it, bool proc_locked)
1672 {
1673 	struct timespec cts, ts;
1674 	struct timeval tv;
1675 	struct proc *p;
1676 	uint64_t interval, now, overruns, value;
1677 
1678 	realtimer_clocktime(it->it_clockid, &cts);
1679 	/* Only fire if time is reached. */
1680 	if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1681 		if (timespecisset(&it->it_time.it_interval)) {
1682 			timespecadd(&it->it_time.it_value,
1683 			    &it->it_time.it_interval,
1684 			    &it->it_time.it_value);
1685 
1686 			interval = timespectons(&it->it_time.it_interval);
1687 			value = timespectons(&it->it_time.it_value);
1688 			now = timespectons(&cts);
1689 
1690 			if (now >= value) {
1691 				/*
1692 				 * We missed at least one period.
1693 				 */
1694 				overruns = howmany(now - value + 1, interval);
1695 				if (it->it_overrun + overruns >=
1696 				    it->it_overrun &&
1697 				    it->it_overrun + overruns <= INT_MAX) {
1698 					it->it_overrun += (int)overruns;
1699 				} else {
1700 					it->it_overrun = INT_MAX;
1701 					it->it_ksi.ksi_errno = ERANGE;
1702 				}
1703 				value =
1704 				    now + interval - (now - value) % interval;
1705 				it->it_time.it_value = timespecfromns(value);
1706 			}
1707 		} else {
1708 			/* single shot timer ? */
1709 			timespecclear(&it->it_time.it_value);
1710 		}
1711 
1712 		p = it->it_proc;
1713 		if (timespecisset(&it->it_time.it_value)) {
1714 			if (P_SHOULDSTOP(p) || P_KILLED(p)) {
1715 				it->it_flags |= ITF_PSTOPPED;
1716 			} else {
1717 				timespecsub(&it->it_time.it_value, &cts, &ts);
1718 				TIMESPEC_TO_TIMEVAL(&tv, &ts);
1719 				callout_reset(&it->it_callout, tvtohz(&tv),
1720 				    realtimer_expire, it);
1721 			}
1722 		}
1723 
1724 		itimer_enter(it);
1725 		ITIMER_UNLOCK(it);
1726 		if (proc_locked)
1727 			PROC_UNLOCK(p);
1728 		itimer_fire(it);
1729 		if (proc_locked)
1730 			PROC_LOCK(p);
1731 		ITIMER_LOCK(it);
1732 		itimer_leave(it);
1733 	} else if (timespecisset(&it->it_time.it_value)) {
1734 		p = it->it_proc;
1735 		if (P_SHOULDSTOP(p) || P_KILLED(p)) {
1736 			it->it_flags |= ITF_PSTOPPED;
1737 		} else {
1738 			ts = it->it_time.it_value;
1739 			timespecsub(&ts, &cts, &ts);
1740 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1741 			callout_reset(&it->it_callout, tvtohz(&tv),
1742 			    realtimer_expire, it);
1743 		}
1744 	}
1745 }
1746 
1747 /* Timeout callback for realtime timer */
1748 static void
1749 realtimer_expire(void *arg)
1750 {
1751 	realtimer_expire_l(arg, false);
1752 }
1753 
1754 static void
1755 itimer_fire(struct itimer *it)
1756 {
1757 	struct proc *p = it->it_proc;
1758 	struct thread *td;
1759 
1760 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1761 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1762 		if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1763 			ITIMER_LOCK(it);
1764 			timespecclear(&it->it_time.it_value);
1765 			timespecclear(&it->it_time.it_interval);
1766 			callout_stop(&it->it_callout);
1767 			ITIMER_UNLOCK(it);
1768 			return;
1769 		}
1770 		if (!KSI_ONQ(&it->it_ksi)) {
1771 			it->it_ksi.ksi_errno = 0;
1772 			ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1773 			tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1774 		} else {
1775 			if (it->it_overrun < INT_MAX)
1776 				it->it_overrun++;
1777 			else
1778 				it->it_ksi.ksi_errno = ERANGE;
1779 		}
1780 		PROC_UNLOCK(p);
1781 	}
1782 }
1783 
1784 static void
1785 itimers_alloc(struct proc *p)
1786 {
1787 	struct itimers *its;
1788 	int i;
1789 
1790 	its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1791 	LIST_INIT(&its->its_virtual);
1792 	LIST_INIT(&its->its_prof);
1793 	TAILQ_INIT(&its->its_worklist);
1794 	for (i = 0; i < TIMER_MAX; i++)
1795 		its->its_timers[i] = NULL;
1796 	PROC_LOCK(p);
1797 	if (p->p_itimers == NULL) {
1798 		p->p_itimers = its;
1799 		PROC_UNLOCK(p);
1800 	}
1801 	else {
1802 		PROC_UNLOCK(p);
1803 		free(its, M_SUBPROC);
1804 	}
1805 }
1806 
1807 /* Clean up timers when some process events are being triggered. */
1808 static void
1809 itimers_event_exit_exec(int start_idx, struct proc *p)
1810 {
1811 	struct itimers *its;
1812 	struct itimer *it;
1813 	int i;
1814 
1815 	its = p->p_itimers;
1816 	if (its == NULL)
1817 		return;
1818 
1819 	for (i = start_idx; i < TIMER_MAX; ++i) {
1820 		if ((it = its->its_timers[i]) != NULL)
1821 			kern_ktimer_delete(curthread, i);
1822 	}
1823 	if (its->its_timers[0] == NULL && its->its_timers[1] == NULL &&
1824 	    its->its_timers[2] == NULL) {
1825 		free(its, M_SUBPROC);
1826 		p->p_itimers = NULL;
1827 	}
1828 }
1829 
1830 void
1831 itimers_exec(struct proc *p)
1832 {
1833 	/*
1834 	 * According to susv3, XSI interval timers should be inherited
1835 	 * by new image.
1836 	 */
1837 	itimers_event_exit_exec(3, p);
1838 }
1839 
1840 void
1841 itimers_exit(struct proc *p)
1842 {
1843 	itimers_event_exit_exec(0, p);
1844 }
1845