xref: /freebsd/sys/kern/kern_time.c (revision af6a5351a1fdb1130f18be6c782c4d48916eb971)
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  * 3. 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 "opt_ktrace.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/limits.h>
40 #include <sys/clock.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/eventhandler.h>
45 #include <sys/resourcevar.h>
46 #include <sys/signalvar.h>
47 #include <sys/kernel.h>
48 #include <sys/sleepqueue.h>
49 #include <sys/syscallsubr.h>
50 #include <sys/sysctl.h>
51 #include <sys/sysent.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/posix4.h>
55 #include <sys/time.h>
56 #include <sys/timers.h>
57 #include <sys/timetc.h>
58 #include <sys/vnode.h>
59 #ifdef KTRACE
60 #include <sys/ktrace.h>
61 #endif
62 
63 #include <vm/vm.h>
64 #include <vm/vm_extern.h>
65 
66 #define MAX_CLOCKS 	(CLOCK_MONOTONIC+1)
67 #define CPUCLOCK_BIT		0x80000000
68 #define CPUCLOCK_PROCESS_BIT	0x40000000
69 #define CPUCLOCK_ID_MASK	(~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
70 #define MAKE_THREAD_CPUCLOCK(tid)	(CPUCLOCK_BIT|(tid))
71 #define MAKE_PROCESS_CPUCLOCK(pid)	\
72 	(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
73 
74 static struct kclock	posix_clocks[MAX_CLOCKS];
75 static uma_zone_t	itimer_zone = NULL;
76 
77 /*
78  * Time of day and interval timer support.
79  *
80  * These routines provide the kernel entry points to get and set
81  * the time-of-day and per-process interval timers.  Subroutines
82  * here provide support for adding and subtracting timeval structures
83  * and decrementing interval timers, optionally reloading the interval
84  * timers when they expire.
85  */
86 
87 static int	settime(struct thread *, struct timeval *);
88 static void	timevalfix(struct timeval *);
89 
90 static void	itimer_start(void);
91 static int	itimer_init(void *, int, int);
92 static void	itimer_fini(void *, int);
93 static void	itimer_enter(struct itimer *);
94 static void	itimer_leave(struct itimer *);
95 static struct itimer *itimer_find(struct proc *, int);
96 static void	itimers_alloc(struct proc *);
97 static void	itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
98 static void	itimers_event_hook_exit(void *arg, struct proc *p);
99 static int	realtimer_create(struct itimer *);
100 static int	realtimer_gettime(struct itimer *, struct itimerspec *);
101 static int	realtimer_settime(struct itimer *, int,
102 			struct itimerspec *, struct itimerspec *);
103 static int	realtimer_delete(struct itimer *);
104 static void	realtimer_clocktime(clockid_t, struct timespec *);
105 static void	realtimer_expire(void *);
106 
107 int		register_posix_clock(int, struct kclock *);
108 void		itimer_fire(struct itimer *it);
109 int		itimespecfix(struct timespec *ts);
110 
111 #define CLOCK_CALL(clock, call, arglist)		\
112 	((*posix_clocks[clock].call) arglist)
113 
114 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
115 
116 
117 static int
118 settime(struct thread *td, struct timeval *tv)
119 {
120 	struct timeval delta, tv1, tv2;
121 	static struct timeval maxtime, laststep;
122 	struct timespec ts;
123 
124 	microtime(&tv1);
125 	delta = *tv;
126 	timevalsub(&delta, &tv1);
127 
128 	/*
129 	 * If the system is secure, we do not allow the time to be
130 	 * set to a value earlier than 1 second less than the highest
131 	 * time we have yet seen. The worst a miscreant can do in
132 	 * this circumstance is "freeze" time. He couldn't go
133 	 * back to the past.
134 	 *
135 	 * We similarly do not allow the clock to be stepped more
136 	 * than one second, nor more than once per second. This allows
137 	 * a miscreant to make the clock march double-time, but no worse.
138 	 */
139 	if (securelevel_gt(td->td_ucred, 1) != 0) {
140 		if (delta.tv_sec < 0 || delta.tv_usec < 0) {
141 			/*
142 			 * Update maxtime to latest time we've seen.
143 			 */
144 			if (tv1.tv_sec > maxtime.tv_sec)
145 				maxtime = tv1;
146 			tv2 = *tv;
147 			timevalsub(&tv2, &maxtime);
148 			if (tv2.tv_sec < -1) {
149 				tv->tv_sec = maxtime.tv_sec - 1;
150 				printf("Time adjustment clamped to -1 second\n");
151 			}
152 		} else {
153 			if (tv1.tv_sec == laststep.tv_sec)
154 				return (EPERM);
155 			if (delta.tv_sec > 1) {
156 				tv->tv_sec = tv1.tv_sec + 1;
157 				printf("Time adjustment clamped to +1 second\n");
158 			}
159 			laststep = *tv;
160 		}
161 	}
162 
163 	ts.tv_sec = tv->tv_sec;
164 	ts.tv_nsec = tv->tv_usec * 1000;
165 	tc_setclock(&ts);
166 	resettodr();
167 	return (0);
168 }
169 
170 #ifndef _SYS_SYSPROTO_H_
171 struct clock_getcpuclockid2_args {
172 	id_t id;
173 	int which,
174 	clockid_t *clock_id;
175 };
176 #endif
177 /* ARGSUSED */
178 int
179 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
180 {
181 	clockid_t clk_id;
182 	int error;
183 
184 	error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
185 	if (error == 0)
186 		error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
187 	return (error);
188 }
189 
190 int
191 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
192     clockid_t *clk_id)
193 {
194 	struct proc *p;
195 	pid_t pid;
196 	lwpid_t tid;
197 	int error;
198 
199 	switch (which) {
200 	case CPUCLOCK_WHICH_PID:
201 		if (id != 0) {
202 			error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
203 			if (error != 0)
204 				return (error);
205 			PROC_UNLOCK(p);
206 			pid = id;
207 		} else {
208 			pid = td->td_proc->p_pid;
209 		}
210 		*clk_id = MAKE_PROCESS_CPUCLOCK(pid);
211 		return (0);
212 	case CPUCLOCK_WHICH_TID:
213 		tid = id == 0 ? td->td_tid : id;
214 		*clk_id = MAKE_THREAD_CPUCLOCK(tid);
215 		return (0);
216 	default:
217 		return (EINVAL);
218 	}
219 }
220 
221 #ifndef _SYS_SYSPROTO_H_
222 struct clock_gettime_args {
223 	clockid_t clock_id;
224 	struct	timespec *tp;
225 };
226 #endif
227 /* ARGSUSED */
228 int
229 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
230 {
231 	struct timespec ats;
232 	int error;
233 
234 	error = kern_clock_gettime(td, uap->clock_id, &ats);
235 	if (error == 0)
236 		error = copyout(&ats, uap->tp, sizeof(ats));
237 
238 	return (error);
239 }
240 
241 static inline void
242 cputick2timespec(uint64_t runtime, struct timespec *ats)
243 {
244 	runtime = cputick2usec(runtime);
245 	ats->tv_sec = runtime / 1000000;
246 	ats->tv_nsec = runtime % 1000000 * 1000;
247 }
248 
249 static void
250 get_thread_cputime(struct thread *targettd, struct timespec *ats)
251 {
252 	uint64_t runtime, curtime, switchtime;
253 
254 	if (targettd == NULL) { /* current thread */
255 		critical_enter();
256 		switchtime = PCPU_GET(switchtime);
257 		curtime = cpu_ticks();
258 		runtime = curthread->td_runtime;
259 		critical_exit();
260 		runtime += curtime - switchtime;
261 	} else {
262 		thread_lock(targettd);
263 		runtime = targettd->td_runtime;
264 		thread_unlock(targettd);
265 	}
266 	cputick2timespec(runtime, ats);
267 }
268 
269 static void
270 get_process_cputime(struct proc *targetp, struct timespec *ats)
271 {
272 	uint64_t runtime;
273 	struct rusage ru;
274 
275 	PROC_STATLOCK(targetp);
276 	rufetch(targetp, &ru);
277 	runtime = targetp->p_rux.rux_runtime;
278 	PROC_STATUNLOCK(targetp);
279 	cputick2timespec(runtime, ats);
280 }
281 
282 static int
283 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
284 {
285 	struct proc *p, *p2;
286 	struct thread *td2;
287 	lwpid_t tid;
288 	pid_t pid;
289 	int error;
290 
291 	p = td->td_proc;
292 	if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
293 		tid = clock_id & CPUCLOCK_ID_MASK;
294 		td2 = tdfind(tid, p->p_pid);
295 		if (td2 == NULL)
296 			return (EINVAL);
297 		get_thread_cputime(td2, ats);
298 		PROC_UNLOCK(td2->td_proc);
299 	} else {
300 		pid = clock_id & CPUCLOCK_ID_MASK;
301 		error = pget(pid, PGET_CANSEE, &p2);
302 		if (error != 0)
303 			return (EINVAL);
304 		get_process_cputime(p2, ats);
305 		PROC_UNLOCK(p2);
306 	}
307 	return (0);
308 }
309 
310 int
311 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
312 {
313 	struct timeval sys, user;
314 	struct proc *p;
315 
316 	p = td->td_proc;
317 	switch (clock_id) {
318 	case CLOCK_REALTIME:		/* Default to precise. */
319 	case CLOCK_REALTIME_PRECISE:
320 		nanotime(ats);
321 		break;
322 	case CLOCK_REALTIME_FAST:
323 		getnanotime(ats);
324 		break;
325 	case CLOCK_VIRTUAL:
326 		PROC_LOCK(p);
327 		PROC_STATLOCK(p);
328 		calcru(p, &user, &sys);
329 		PROC_STATUNLOCK(p);
330 		PROC_UNLOCK(p);
331 		TIMEVAL_TO_TIMESPEC(&user, ats);
332 		break;
333 	case CLOCK_PROF:
334 		PROC_LOCK(p);
335 		PROC_STATLOCK(p);
336 		calcru(p, &user, &sys);
337 		PROC_STATUNLOCK(p);
338 		PROC_UNLOCK(p);
339 		timevaladd(&user, &sys);
340 		TIMEVAL_TO_TIMESPEC(&user, ats);
341 		break;
342 	case CLOCK_MONOTONIC:		/* Default to precise. */
343 	case CLOCK_MONOTONIC_PRECISE:
344 	case CLOCK_UPTIME:
345 	case CLOCK_UPTIME_PRECISE:
346 		nanouptime(ats);
347 		break;
348 	case CLOCK_UPTIME_FAST:
349 	case CLOCK_MONOTONIC_FAST:
350 		getnanouptime(ats);
351 		break;
352 	case CLOCK_SECOND:
353 		ats->tv_sec = time_second;
354 		ats->tv_nsec = 0;
355 		break;
356 	case CLOCK_THREAD_CPUTIME_ID:
357 		get_thread_cputime(NULL, ats);
358 		break;
359 	case CLOCK_PROCESS_CPUTIME_ID:
360 		PROC_LOCK(p);
361 		get_process_cputime(p, ats);
362 		PROC_UNLOCK(p);
363 		break;
364 	default:
365 		if ((int)clock_id >= 0)
366 			return (EINVAL);
367 		return (get_cputime(td, clock_id, ats));
368 	}
369 	return (0);
370 }
371 
372 #ifndef _SYS_SYSPROTO_H_
373 struct clock_settime_args {
374 	clockid_t clock_id;
375 	const struct	timespec *tp;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
381 {
382 	struct timespec ats;
383 	int error;
384 
385 	if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
386 		return (error);
387 	return (kern_clock_settime(td, uap->clock_id, &ats));
388 }
389 
390 static int allow_insane_settime = 0;
391 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN,
392     &allow_insane_settime, 0,
393     "do not perform possibly restrictive checks on settime(2) args");
394 
395 int
396 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
397 {
398 	struct timeval atv;
399 	int error;
400 
401 	if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
402 		return (error);
403 	if (clock_id != CLOCK_REALTIME)
404 		return (EINVAL);
405 	if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 ||
406 	    ats->tv_sec < 0)
407 		return (EINVAL);
408 	if (!allow_insane_settime && ats->tv_sec > 9999ULL * 366 * 24 * 60 * 60)
409 		return (EINVAL);
410 	/* XXX Don't convert nsec->usec and back */
411 	TIMESPEC_TO_TIMEVAL(&atv, ats);
412 	error = settime(td, &atv);
413 	return (error);
414 }
415 
416 #ifndef _SYS_SYSPROTO_H_
417 struct clock_getres_args {
418 	clockid_t clock_id;
419 	struct	timespec *tp;
420 };
421 #endif
422 int
423 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
424 {
425 	struct timespec ts;
426 	int error;
427 
428 	if (uap->tp == NULL)
429 		return (0);
430 
431 	error = kern_clock_getres(td, uap->clock_id, &ts);
432 	if (error == 0)
433 		error = copyout(&ts, uap->tp, sizeof(ts));
434 	return (error);
435 }
436 
437 int
438 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
439 {
440 
441 	ts->tv_sec = 0;
442 	switch (clock_id) {
443 	case CLOCK_REALTIME:
444 	case CLOCK_REALTIME_FAST:
445 	case CLOCK_REALTIME_PRECISE:
446 	case CLOCK_MONOTONIC:
447 	case CLOCK_MONOTONIC_FAST:
448 	case CLOCK_MONOTONIC_PRECISE:
449 	case CLOCK_UPTIME:
450 	case CLOCK_UPTIME_FAST:
451 	case CLOCK_UPTIME_PRECISE:
452 		/*
453 		 * Round up the result of the division cheaply by adding 1.
454 		 * Rounding up is especially important if rounding down
455 		 * would give 0.  Perfect rounding is unimportant.
456 		 */
457 		ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
458 		break;
459 	case CLOCK_VIRTUAL:
460 	case CLOCK_PROF:
461 		/* Accurately round up here because we can do so cheaply. */
462 		ts->tv_nsec = howmany(1000000000, hz);
463 		break;
464 	case CLOCK_SECOND:
465 		ts->tv_sec = 1;
466 		ts->tv_nsec = 0;
467 		break;
468 	case CLOCK_THREAD_CPUTIME_ID:
469 	case CLOCK_PROCESS_CPUTIME_ID:
470 	cputime:
471 		/* sync with cputick2usec */
472 		ts->tv_nsec = 1000000 / cpu_tickrate();
473 		if (ts->tv_nsec == 0)
474 			ts->tv_nsec = 1000;
475 		break;
476 	default:
477 		if ((int)clock_id < 0)
478 			goto cputime;
479 		return (EINVAL);
480 	}
481 	return (0);
482 }
483 
484 static uint8_t nanowait[MAXCPU];
485 
486 int
487 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
488 {
489 	struct timespec ts;
490 	sbintime_t sbt, sbtt, prec, tmp;
491 	time_t over;
492 	int error;
493 
494 	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
495 		return (EINVAL);
496 	if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
497 		return (0);
498 	ts = *rqt;
499 	if (ts.tv_sec > INT32_MAX / 2) {
500 		over = ts.tv_sec - INT32_MAX / 2;
501 		ts.tv_sec -= over;
502 	} else
503 		over = 0;
504 	tmp = tstosbt(ts);
505 	prec = tmp;
506 	prec >>= tc_precexp;
507 	if (TIMESEL(&sbt, tmp))
508 		sbt += tc_tick_sbt;
509 	sbt += tmp;
510 	error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
511 	    sbt, prec, C_ABSOLUTE);
512 	if (error != EWOULDBLOCK) {
513 		if (error == ERESTART)
514 			error = EINTR;
515 		TIMESEL(&sbtt, tmp);
516 		if (rmt != NULL) {
517 			ts = sbttots(sbt - sbtt);
518 			ts.tv_sec += over;
519 			if (ts.tv_sec < 0)
520 				timespecclear(&ts);
521 			*rmt = ts;
522 		}
523 		if (sbtt >= sbt)
524 			return (0);
525 		return (error);
526 	}
527 	return (0);
528 }
529 
530 #ifndef _SYS_SYSPROTO_H_
531 struct nanosleep_args {
532 	struct	timespec *rqtp;
533 	struct	timespec *rmtp;
534 };
535 #endif
536 /* ARGSUSED */
537 int
538 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
539 {
540 	struct timespec rmt, rqt;
541 	int error;
542 
543 	error = copyin(uap->rqtp, &rqt, sizeof(rqt));
544 	if (error)
545 		return (error);
546 
547 	if (uap->rmtp &&
548 	    !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE))
549 			return (EFAULT);
550 	error = kern_nanosleep(td, &rqt, &rmt);
551 	if (error && uap->rmtp) {
552 		int error2;
553 
554 		error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
555 		if (error2)
556 			error = error2;
557 	}
558 	return (error);
559 }
560 
561 #ifndef _SYS_SYSPROTO_H_
562 struct gettimeofday_args {
563 	struct	timeval *tp;
564 	struct	timezone *tzp;
565 };
566 #endif
567 /* ARGSUSED */
568 int
569 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
570 {
571 	struct timeval atv;
572 	struct timezone rtz;
573 	int error = 0;
574 
575 	if (uap->tp) {
576 		microtime(&atv);
577 		error = copyout(&atv, uap->tp, sizeof (atv));
578 	}
579 	if (error == 0 && uap->tzp != NULL) {
580 		rtz.tz_minuteswest = tz_minuteswest;
581 		rtz.tz_dsttime = tz_dsttime;
582 		error = copyout(&rtz, uap->tzp, sizeof (rtz));
583 	}
584 	return (error);
585 }
586 
587 #ifndef _SYS_SYSPROTO_H_
588 struct settimeofday_args {
589 	struct	timeval *tv;
590 	struct	timezone *tzp;
591 };
592 #endif
593 /* ARGSUSED */
594 int
595 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
596 {
597 	struct timeval atv, *tvp;
598 	struct timezone atz, *tzp;
599 	int error;
600 
601 	if (uap->tv) {
602 		error = copyin(uap->tv, &atv, sizeof(atv));
603 		if (error)
604 			return (error);
605 		tvp = &atv;
606 	} else
607 		tvp = NULL;
608 	if (uap->tzp) {
609 		error = copyin(uap->tzp, &atz, sizeof(atz));
610 		if (error)
611 			return (error);
612 		tzp = &atz;
613 	} else
614 		tzp = NULL;
615 	return (kern_settimeofday(td, tvp, tzp));
616 }
617 
618 int
619 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
620 {
621 	int error;
622 
623 	error = priv_check(td, PRIV_SETTIMEOFDAY);
624 	if (error)
625 		return (error);
626 	/* Verify all parameters before changing time. */
627 	if (tv) {
628 		if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
629 		    tv->tv_sec < 0)
630 			return (EINVAL);
631 		error = settime(td, tv);
632 	}
633 	if (tzp && error == 0) {
634 		tz_minuteswest = tzp->tz_minuteswest;
635 		tz_dsttime = tzp->tz_dsttime;
636 	}
637 	return (error);
638 }
639 
640 /*
641  * Get value of an interval timer.  The process virtual and profiling virtual
642  * time timers are kept in the p_stats area, since they can be swapped out.
643  * These are kept internally in the way they are specified externally: in
644  * time until they expire.
645  *
646  * The real time interval timer is kept in the process table slot for the
647  * process, and its value (it_value) is kept as an absolute time rather than
648  * as a delta, so that it is easy to keep periodic real-time signals from
649  * drifting.
650  *
651  * Virtual time timers are processed in the hardclock() routine of
652  * kern_clock.c.  The real time timer is processed by a timeout routine,
653  * called from the softclock() routine.  Since a callout may be delayed in
654  * real time due to interrupt processing in the system, it is possible for
655  * the real time timeout routine (realitexpire, given below), to be delayed
656  * in real time past when it is supposed to occur.  It does not suffice,
657  * therefore, to reload the real timer .it_value from the real time timers
658  * .it_interval.  Rather, we compute the next time in absolute time the timer
659  * should go off.
660  */
661 #ifndef _SYS_SYSPROTO_H_
662 struct getitimer_args {
663 	u_int	which;
664 	struct	itimerval *itv;
665 };
666 #endif
667 int
668 sys_getitimer(struct thread *td, struct getitimer_args *uap)
669 {
670 	struct itimerval aitv;
671 	int error;
672 
673 	error = kern_getitimer(td, uap->which, &aitv);
674 	if (error != 0)
675 		return (error);
676 	return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
677 }
678 
679 int
680 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
681 {
682 	struct proc *p = td->td_proc;
683 	struct timeval ctv;
684 
685 	if (which > ITIMER_PROF)
686 		return (EINVAL);
687 
688 	if (which == ITIMER_REAL) {
689 		/*
690 		 * Convert from absolute to relative time in .it_value
691 		 * part of real time timer.  If time for real time timer
692 		 * has passed return 0, else return difference between
693 		 * current time and time for the timer to go off.
694 		 */
695 		PROC_LOCK(p);
696 		*aitv = p->p_realtimer;
697 		PROC_UNLOCK(p);
698 		if (timevalisset(&aitv->it_value)) {
699 			microuptime(&ctv);
700 			if (timevalcmp(&aitv->it_value, &ctv, <))
701 				timevalclear(&aitv->it_value);
702 			else
703 				timevalsub(&aitv->it_value, &ctv);
704 		}
705 	} else {
706 		PROC_ITIMLOCK(p);
707 		*aitv = p->p_stats->p_timer[which];
708 		PROC_ITIMUNLOCK(p);
709 	}
710 #ifdef KTRACE
711 	if (KTRPOINT(td, KTR_STRUCT))
712 		ktritimerval(aitv);
713 #endif
714 	return (0);
715 }
716 
717 #ifndef _SYS_SYSPROTO_H_
718 struct setitimer_args {
719 	u_int	which;
720 	struct	itimerval *itv, *oitv;
721 };
722 #endif
723 int
724 sys_setitimer(struct thread *td, struct setitimer_args *uap)
725 {
726 	struct itimerval aitv, oitv;
727 	int error;
728 
729 	if (uap->itv == NULL) {
730 		uap->itv = uap->oitv;
731 		return (sys_getitimer(td, (struct getitimer_args *)uap));
732 	}
733 
734 	if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
735 		return (error);
736 	error = kern_setitimer(td, uap->which, &aitv, &oitv);
737 	if (error != 0 || uap->oitv == NULL)
738 		return (error);
739 	return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
740 }
741 
742 int
743 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
744     struct itimerval *oitv)
745 {
746 	struct proc *p = td->td_proc;
747 	struct timeval ctv;
748 	sbintime_t sbt, pr;
749 
750 	if (aitv == NULL)
751 		return (kern_getitimer(td, which, oitv));
752 
753 	if (which > ITIMER_PROF)
754 		return (EINVAL);
755 #ifdef KTRACE
756 	if (KTRPOINT(td, KTR_STRUCT))
757 		ktritimerval(aitv);
758 #endif
759 	if (itimerfix(&aitv->it_value) ||
760 	    aitv->it_value.tv_sec > INT32_MAX / 2)
761 		return (EINVAL);
762 	if (!timevalisset(&aitv->it_value))
763 		timevalclear(&aitv->it_interval);
764 	else if (itimerfix(&aitv->it_interval) ||
765 	    aitv->it_interval.tv_sec > INT32_MAX / 2)
766 		return (EINVAL);
767 
768 	if (which == ITIMER_REAL) {
769 		PROC_LOCK(p);
770 		if (timevalisset(&p->p_realtimer.it_value))
771 			callout_stop(&p->p_itcallout);
772 		microuptime(&ctv);
773 		if (timevalisset(&aitv->it_value)) {
774 			pr = tvtosbt(aitv->it_value) >> tc_precexp;
775 			timevaladd(&aitv->it_value, &ctv);
776 			sbt = tvtosbt(aitv->it_value);
777 			callout_reset_sbt(&p->p_itcallout, sbt, pr,
778 			    realitexpire, p, C_ABSOLUTE);
779 		}
780 		*oitv = p->p_realtimer;
781 		p->p_realtimer = *aitv;
782 		PROC_UNLOCK(p);
783 		if (timevalisset(&oitv->it_value)) {
784 			if (timevalcmp(&oitv->it_value, &ctv, <))
785 				timevalclear(&oitv->it_value);
786 			else
787 				timevalsub(&oitv->it_value, &ctv);
788 		}
789 	} else {
790 		if (aitv->it_interval.tv_sec == 0 &&
791 		    aitv->it_interval.tv_usec != 0 &&
792 		    aitv->it_interval.tv_usec < tick)
793 			aitv->it_interval.tv_usec = tick;
794 		if (aitv->it_value.tv_sec == 0 &&
795 		    aitv->it_value.tv_usec != 0 &&
796 		    aitv->it_value.tv_usec < tick)
797 			aitv->it_value.tv_usec = tick;
798 		PROC_ITIMLOCK(p);
799 		*oitv = p->p_stats->p_timer[which];
800 		p->p_stats->p_timer[which] = *aitv;
801 		PROC_ITIMUNLOCK(p);
802 	}
803 #ifdef KTRACE
804 	if (KTRPOINT(td, KTR_STRUCT))
805 		ktritimerval(oitv);
806 #endif
807 	return (0);
808 }
809 
810 /*
811  * Real interval timer expired:
812  * send process whose timer expired an alarm signal.
813  * If time is not set up to reload, then just return.
814  * Else compute next time timer should go off which is > current time.
815  * This is where delay in processing this timeout causes multiple
816  * SIGALRM calls to be compressed into one.
817  * tvtohz() always adds 1 to allow for the time until the next clock
818  * interrupt being strictly less than 1 clock tick, but we don't want
819  * that here since we want to appear to be in sync with the clock
820  * interrupt even when we're delayed.
821  */
822 void
823 realitexpire(void *arg)
824 {
825 	struct proc *p;
826 	struct timeval ctv;
827 	sbintime_t isbt;
828 
829 	p = (struct proc *)arg;
830 	kern_psignal(p, SIGALRM);
831 	if (!timevalisset(&p->p_realtimer.it_interval)) {
832 		timevalclear(&p->p_realtimer.it_value);
833 		if (p->p_flag & P_WEXIT)
834 			wakeup(&p->p_itcallout);
835 		return;
836 	}
837 	isbt = tvtosbt(p->p_realtimer.it_interval);
838 	if (isbt >= sbt_timethreshold)
839 		getmicrouptime(&ctv);
840 	else
841 		microuptime(&ctv);
842 	do {
843 		timevaladd(&p->p_realtimer.it_value,
844 		    &p->p_realtimer.it_interval);
845 	} while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
846 	callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
847 	    isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE);
848 }
849 
850 /*
851  * Check that a proposed value to load into the .it_value or
852  * .it_interval part of an interval timer is acceptable, and
853  * fix it to have at least minimal value (i.e. if it is less
854  * than the resolution of the clock, round it up.)
855  */
856 int
857 itimerfix(struct timeval *tv)
858 {
859 
860 	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
861 		return (EINVAL);
862 	if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
863 	    tv->tv_usec < (u_int)tick / 16)
864 		tv->tv_usec = (u_int)tick / 16;
865 	return (0);
866 }
867 
868 /*
869  * Decrement an interval timer by a specified number
870  * of microseconds, which must be less than a second,
871  * i.e. < 1000000.  If the timer expires, then reload
872  * it.  In this case, carry over (usec - old value) to
873  * reduce the value reloaded into the timer so that
874  * the timer does not drift.  This routine assumes
875  * that it is called in a context where the timers
876  * on which it is operating cannot change in value.
877  */
878 int
879 itimerdecr(struct itimerval *itp, int usec)
880 {
881 
882 	if (itp->it_value.tv_usec < usec) {
883 		if (itp->it_value.tv_sec == 0) {
884 			/* expired, and already in next interval */
885 			usec -= itp->it_value.tv_usec;
886 			goto expire;
887 		}
888 		itp->it_value.tv_usec += 1000000;
889 		itp->it_value.tv_sec--;
890 	}
891 	itp->it_value.tv_usec -= usec;
892 	usec = 0;
893 	if (timevalisset(&itp->it_value))
894 		return (1);
895 	/* expired, exactly at end of interval */
896 expire:
897 	if (timevalisset(&itp->it_interval)) {
898 		itp->it_value = itp->it_interval;
899 		itp->it_value.tv_usec -= usec;
900 		if (itp->it_value.tv_usec < 0) {
901 			itp->it_value.tv_usec += 1000000;
902 			itp->it_value.tv_sec--;
903 		}
904 	} else
905 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
906 	return (0);
907 }
908 
909 /*
910  * Add and subtract routines for timevals.
911  * N.B.: subtract routine doesn't deal with
912  * results which are before the beginning,
913  * it just gets very confused in this case.
914  * Caveat emptor.
915  */
916 void
917 timevaladd(struct timeval *t1, const struct timeval *t2)
918 {
919 
920 	t1->tv_sec += t2->tv_sec;
921 	t1->tv_usec += t2->tv_usec;
922 	timevalfix(t1);
923 }
924 
925 void
926 timevalsub(struct timeval *t1, const struct timeval *t2)
927 {
928 
929 	t1->tv_sec -= t2->tv_sec;
930 	t1->tv_usec -= t2->tv_usec;
931 	timevalfix(t1);
932 }
933 
934 static void
935 timevalfix(struct timeval *t1)
936 {
937 
938 	if (t1->tv_usec < 0) {
939 		t1->tv_sec--;
940 		t1->tv_usec += 1000000;
941 	}
942 	if (t1->tv_usec >= 1000000) {
943 		t1->tv_sec++;
944 		t1->tv_usec -= 1000000;
945 	}
946 }
947 
948 /*
949  * ratecheck(): simple time-based rate-limit checking.
950  */
951 int
952 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
953 {
954 	struct timeval tv, delta;
955 	int rv = 0;
956 
957 	getmicrouptime(&tv);		/* NB: 10ms precision */
958 	delta = tv;
959 	timevalsub(&delta, lasttime);
960 
961 	/*
962 	 * check for 0,0 is so that the message will be seen at least once,
963 	 * even if interval is huge.
964 	 */
965 	if (timevalcmp(&delta, mininterval, >=) ||
966 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
967 		*lasttime = tv;
968 		rv = 1;
969 	}
970 
971 	return (rv);
972 }
973 
974 /*
975  * ppsratecheck(): packets (or events) per second limitation.
976  *
977  * Return 0 if the limit is to be enforced (e.g. the caller
978  * should drop a packet because of the rate limitation).
979  *
980  * maxpps of 0 always causes zero to be returned.  maxpps of -1
981  * always causes 1 to be returned; this effectively defeats rate
982  * limiting.
983  *
984  * Note that we maintain the struct timeval for compatibility
985  * with other bsd systems.  We reuse the storage and just monitor
986  * clock ticks for minimal overhead.
987  */
988 int
989 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
990 {
991 	int now;
992 
993 	/*
994 	 * Reset the last time and counter if this is the first call
995 	 * or more than a second has passed since the last update of
996 	 * lasttime.
997 	 */
998 	now = ticks;
999 	if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1000 		lasttime->tv_sec = now;
1001 		*curpps = 1;
1002 		return (maxpps != 0);
1003 	} else {
1004 		(*curpps)++;		/* NB: ignore potential overflow */
1005 		return (maxpps < 0 || *curpps <= maxpps);
1006 	}
1007 }
1008 
1009 static void
1010 itimer_start(void)
1011 {
1012 	struct kclock rt_clock = {
1013 		.timer_create  = realtimer_create,
1014 		.timer_delete  = realtimer_delete,
1015 		.timer_settime = realtimer_settime,
1016 		.timer_gettime = realtimer_gettime,
1017 		.event_hook    = NULL
1018 	};
1019 
1020 	itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1021 		NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1022 	register_posix_clock(CLOCK_REALTIME,  &rt_clock);
1023 	register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1024 	p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1025 	p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1026 	p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1027 	EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
1028 		(void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
1029 	EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
1030 		(void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
1031 }
1032 
1033 int
1034 register_posix_clock(int clockid, struct kclock *clk)
1035 {
1036 	if ((unsigned)clockid >= MAX_CLOCKS) {
1037 		printf("%s: invalid clockid\n", __func__);
1038 		return (0);
1039 	}
1040 	posix_clocks[clockid] = *clk;
1041 	return (1);
1042 }
1043 
1044 static int
1045 itimer_init(void *mem, int size, int flags)
1046 {
1047 	struct itimer *it;
1048 
1049 	it = (struct itimer *)mem;
1050 	mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1051 	return (0);
1052 }
1053 
1054 static void
1055 itimer_fini(void *mem, int size)
1056 {
1057 	struct itimer *it;
1058 
1059 	it = (struct itimer *)mem;
1060 	mtx_destroy(&it->it_mtx);
1061 }
1062 
1063 static void
1064 itimer_enter(struct itimer *it)
1065 {
1066 
1067 	mtx_assert(&it->it_mtx, MA_OWNED);
1068 	it->it_usecount++;
1069 }
1070 
1071 static void
1072 itimer_leave(struct itimer *it)
1073 {
1074 
1075 	mtx_assert(&it->it_mtx, MA_OWNED);
1076 	KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1077 
1078 	if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1079 		wakeup(it);
1080 }
1081 
1082 #ifndef _SYS_SYSPROTO_H_
1083 struct ktimer_create_args {
1084 	clockid_t clock_id;
1085 	struct sigevent * evp;
1086 	int * timerid;
1087 };
1088 #endif
1089 int
1090 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1091 {
1092 	struct sigevent *evp, ev;
1093 	int id;
1094 	int error;
1095 
1096 	if (uap->evp == NULL) {
1097 		evp = NULL;
1098 	} else {
1099 		error = copyin(uap->evp, &ev, sizeof(ev));
1100 		if (error != 0)
1101 			return (error);
1102 		evp = &ev;
1103 	}
1104 	error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1105 	if (error == 0) {
1106 		error = copyout(&id, uap->timerid, sizeof(int));
1107 		if (error != 0)
1108 			kern_ktimer_delete(td, id);
1109 	}
1110 	return (error);
1111 }
1112 
1113 int
1114 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1115     int *timerid, int preset_id)
1116 {
1117 	struct proc *p = td->td_proc;
1118 	struct itimer *it;
1119 	int id;
1120 	int error;
1121 
1122 	if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1123 		return (EINVAL);
1124 
1125 	if (posix_clocks[clock_id].timer_create == NULL)
1126 		return (EINVAL);
1127 
1128 	if (evp != NULL) {
1129 		if (evp->sigev_notify != SIGEV_NONE &&
1130 		    evp->sigev_notify != SIGEV_SIGNAL &&
1131 		    evp->sigev_notify != SIGEV_THREAD_ID)
1132 			return (EINVAL);
1133 		if ((evp->sigev_notify == SIGEV_SIGNAL ||
1134 		     evp->sigev_notify == SIGEV_THREAD_ID) &&
1135 			!_SIG_VALID(evp->sigev_signo))
1136 			return (EINVAL);
1137 	}
1138 
1139 	if (p->p_itimers == NULL)
1140 		itimers_alloc(p);
1141 
1142 	it = uma_zalloc(itimer_zone, M_WAITOK);
1143 	it->it_flags = 0;
1144 	it->it_usecount = 0;
1145 	it->it_active = 0;
1146 	timespecclear(&it->it_time.it_value);
1147 	timespecclear(&it->it_time.it_interval);
1148 	it->it_overrun = 0;
1149 	it->it_overrun_last = 0;
1150 	it->it_clockid = clock_id;
1151 	it->it_timerid = -1;
1152 	it->it_proc = p;
1153 	ksiginfo_init(&it->it_ksi);
1154 	it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1155 	error = CLOCK_CALL(clock_id, timer_create, (it));
1156 	if (error != 0)
1157 		goto out;
1158 
1159 	PROC_LOCK(p);
1160 	if (preset_id != -1) {
1161 		KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1162 		id = preset_id;
1163 		if (p->p_itimers->its_timers[id] != NULL) {
1164 			PROC_UNLOCK(p);
1165 			error = 0;
1166 			goto out;
1167 		}
1168 	} else {
1169 		/*
1170 		 * Find a free timer slot, skipping those reserved
1171 		 * for setitimer().
1172 		 */
1173 		for (id = 3; id < TIMER_MAX; id++)
1174 			if (p->p_itimers->its_timers[id] == NULL)
1175 				break;
1176 		if (id == TIMER_MAX) {
1177 			PROC_UNLOCK(p);
1178 			error = EAGAIN;
1179 			goto out;
1180 		}
1181 	}
1182 	it->it_timerid = id;
1183 	p->p_itimers->its_timers[id] = it;
1184 	if (evp != NULL)
1185 		it->it_sigev = *evp;
1186 	else {
1187 		it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1188 		switch (clock_id) {
1189 		default:
1190 		case CLOCK_REALTIME:
1191 			it->it_sigev.sigev_signo = SIGALRM;
1192 			break;
1193 		case CLOCK_VIRTUAL:
1194  			it->it_sigev.sigev_signo = SIGVTALRM;
1195 			break;
1196 		case CLOCK_PROF:
1197 			it->it_sigev.sigev_signo = SIGPROF;
1198 			break;
1199 		}
1200 		it->it_sigev.sigev_value.sival_int = id;
1201 	}
1202 
1203 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1204 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1205 		it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1206 		it->it_ksi.ksi_code = SI_TIMER;
1207 		it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1208 		it->it_ksi.ksi_timerid = id;
1209 	}
1210 	PROC_UNLOCK(p);
1211 	*timerid = id;
1212 	return (0);
1213 
1214 out:
1215 	ITIMER_LOCK(it);
1216 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1217 	ITIMER_UNLOCK(it);
1218 	uma_zfree(itimer_zone, it);
1219 	return (error);
1220 }
1221 
1222 #ifndef _SYS_SYSPROTO_H_
1223 struct ktimer_delete_args {
1224 	int timerid;
1225 };
1226 #endif
1227 int
1228 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1229 {
1230 
1231 	return (kern_ktimer_delete(td, uap->timerid));
1232 }
1233 
1234 static struct itimer *
1235 itimer_find(struct proc *p, int timerid)
1236 {
1237 	struct itimer *it;
1238 
1239 	PROC_LOCK_ASSERT(p, MA_OWNED);
1240 	if ((p->p_itimers == NULL) ||
1241 	    (timerid < 0) || (timerid >= TIMER_MAX) ||
1242 	    (it = p->p_itimers->its_timers[timerid]) == NULL) {
1243 		return (NULL);
1244 	}
1245 	ITIMER_LOCK(it);
1246 	if ((it->it_flags & ITF_DELETING) != 0) {
1247 		ITIMER_UNLOCK(it);
1248 		it = NULL;
1249 	}
1250 	return (it);
1251 }
1252 
1253 int
1254 kern_ktimer_delete(struct thread *td, int timerid)
1255 {
1256 	struct proc *p = td->td_proc;
1257 	struct itimer *it;
1258 
1259 	PROC_LOCK(p);
1260 	it = itimer_find(p, timerid);
1261 	if (it == NULL) {
1262 		PROC_UNLOCK(p);
1263 		return (EINVAL);
1264 	}
1265 	PROC_UNLOCK(p);
1266 
1267 	it->it_flags |= ITF_DELETING;
1268 	while (it->it_usecount > 0) {
1269 		it->it_flags |= ITF_WANTED;
1270 		msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1271 	}
1272 	it->it_flags &= ~ITF_WANTED;
1273 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1274 	ITIMER_UNLOCK(it);
1275 
1276 	PROC_LOCK(p);
1277 	if (KSI_ONQ(&it->it_ksi))
1278 		sigqueue_take(&it->it_ksi);
1279 	p->p_itimers->its_timers[timerid] = NULL;
1280 	PROC_UNLOCK(p);
1281 	uma_zfree(itimer_zone, it);
1282 	return (0);
1283 }
1284 
1285 #ifndef _SYS_SYSPROTO_H_
1286 struct ktimer_settime_args {
1287 	int timerid;
1288 	int flags;
1289 	const struct itimerspec * value;
1290 	struct itimerspec * ovalue;
1291 };
1292 #endif
1293 int
1294 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1295 {
1296 	struct itimerspec val, oval, *ovalp;
1297 	int error;
1298 
1299 	error = copyin(uap->value, &val, sizeof(val));
1300 	if (error != 0)
1301 		return (error);
1302 	ovalp = uap->ovalue != NULL ? &oval : NULL;
1303 	error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1304 	if (error == 0 && uap->ovalue != NULL)
1305 		error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1306 	return (error);
1307 }
1308 
1309 int
1310 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1311     struct itimerspec *val, struct itimerspec *oval)
1312 {
1313 	struct proc *p;
1314 	struct itimer *it;
1315 	int error;
1316 
1317 	p = td->td_proc;
1318 	PROC_LOCK(p);
1319 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1320 		PROC_UNLOCK(p);
1321 		error = EINVAL;
1322 	} else {
1323 		PROC_UNLOCK(p);
1324 		itimer_enter(it);
1325 		error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1326 		    flags, val, oval));
1327 		itimer_leave(it);
1328 		ITIMER_UNLOCK(it);
1329 	}
1330 	return (error);
1331 }
1332 
1333 #ifndef _SYS_SYSPROTO_H_
1334 struct ktimer_gettime_args {
1335 	int timerid;
1336 	struct itimerspec * value;
1337 };
1338 #endif
1339 int
1340 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1341 {
1342 	struct itimerspec val;
1343 	int error;
1344 
1345 	error = kern_ktimer_gettime(td, uap->timerid, &val);
1346 	if (error == 0)
1347 		error = copyout(&val, uap->value, sizeof(val));
1348 	return (error);
1349 }
1350 
1351 int
1352 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1353 {
1354 	struct proc *p;
1355 	struct itimer *it;
1356 	int error;
1357 
1358 	p = td->td_proc;
1359 	PROC_LOCK(p);
1360 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1361 		PROC_UNLOCK(p);
1362 		error = EINVAL;
1363 	} else {
1364 		PROC_UNLOCK(p);
1365 		itimer_enter(it);
1366 		error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1367 		itimer_leave(it);
1368 		ITIMER_UNLOCK(it);
1369 	}
1370 	return (error);
1371 }
1372 
1373 #ifndef _SYS_SYSPROTO_H_
1374 struct timer_getoverrun_args {
1375 	int timerid;
1376 };
1377 #endif
1378 int
1379 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1380 {
1381 
1382 	return (kern_ktimer_getoverrun(td, uap->timerid));
1383 }
1384 
1385 int
1386 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1387 {
1388 	struct proc *p = td->td_proc;
1389 	struct itimer *it;
1390 	int error ;
1391 
1392 	PROC_LOCK(p);
1393 	if (timer_id < 3 ||
1394 	    (it = itimer_find(p, timer_id)) == NULL) {
1395 		PROC_UNLOCK(p);
1396 		error = EINVAL;
1397 	} else {
1398 		td->td_retval[0] = it->it_overrun_last;
1399 		ITIMER_UNLOCK(it);
1400 		PROC_UNLOCK(p);
1401 		error = 0;
1402 	}
1403 	return (error);
1404 }
1405 
1406 static int
1407 realtimer_create(struct itimer *it)
1408 {
1409 	callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1410 	return (0);
1411 }
1412 
1413 static int
1414 realtimer_delete(struct itimer *it)
1415 {
1416 	mtx_assert(&it->it_mtx, MA_OWNED);
1417 
1418 	/*
1419 	 * clear timer's value and interval to tell realtimer_expire
1420 	 * to not rearm the timer.
1421 	 */
1422 	timespecclear(&it->it_time.it_value);
1423 	timespecclear(&it->it_time.it_interval);
1424 	ITIMER_UNLOCK(it);
1425 	callout_drain(&it->it_callout);
1426 	ITIMER_LOCK(it);
1427 	return (0);
1428 }
1429 
1430 static int
1431 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1432 {
1433 	struct timespec cts;
1434 
1435 	mtx_assert(&it->it_mtx, MA_OWNED);
1436 
1437 	realtimer_clocktime(it->it_clockid, &cts);
1438 	*ovalue = it->it_time;
1439 	if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1440 		timespecsub(&ovalue->it_value, &cts);
1441 		if (ovalue->it_value.tv_sec < 0 ||
1442 		    (ovalue->it_value.tv_sec == 0 &&
1443 		     ovalue->it_value.tv_nsec == 0)) {
1444 			ovalue->it_value.tv_sec  = 0;
1445 			ovalue->it_value.tv_nsec = 1;
1446 		}
1447 	}
1448 	return (0);
1449 }
1450 
1451 static int
1452 realtimer_settime(struct itimer *it, int flags,
1453 	struct itimerspec *value, struct itimerspec *ovalue)
1454 {
1455 	struct timespec cts, ts;
1456 	struct timeval tv;
1457 	struct itimerspec val;
1458 
1459 	mtx_assert(&it->it_mtx, MA_OWNED);
1460 
1461 	val = *value;
1462 	if (itimespecfix(&val.it_value))
1463 		return (EINVAL);
1464 
1465 	if (timespecisset(&val.it_value)) {
1466 		if (itimespecfix(&val.it_interval))
1467 			return (EINVAL);
1468 	} else {
1469 		timespecclear(&val.it_interval);
1470 	}
1471 
1472 	if (ovalue != NULL)
1473 		realtimer_gettime(it, ovalue);
1474 
1475 	it->it_time = val;
1476 	if (timespecisset(&val.it_value)) {
1477 		realtimer_clocktime(it->it_clockid, &cts);
1478 		ts = val.it_value;
1479 		if ((flags & TIMER_ABSTIME) == 0) {
1480 			/* Convert to absolute time. */
1481 			timespecadd(&it->it_time.it_value, &cts);
1482 		} else {
1483 			timespecsub(&ts, &cts);
1484 			/*
1485 			 * We don't care if ts is negative, tztohz will
1486 			 * fix it.
1487 			 */
1488 		}
1489 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1490 		callout_reset(&it->it_callout, tvtohz(&tv),
1491 			realtimer_expire, it);
1492 	} else {
1493 		callout_stop(&it->it_callout);
1494 	}
1495 
1496 	return (0);
1497 }
1498 
1499 static void
1500 realtimer_clocktime(clockid_t id, struct timespec *ts)
1501 {
1502 	if (id == CLOCK_REALTIME)
1503 		getnanotime(ts);
1504 	else	/* CLOCK_MONOTONIC */
1505 		getnanouptime(ts);
1506 }
1507 
1508 int
1509 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1510 {
1511 	struct itimer *it;
1512 
1513 	PROC_LOCK_ASSERT(p, MA_OWNED);
1514 	it = itimer_find(p, timerid);
1515 	if (it != NULL) {
1516 		ksi->ksi_overrun = it->it_overrun;
1517 		it->it_overrun_last = it->it_overrun;
1518 		it->it_overrun = 0;
1519 		ITIMER_UNLOCK(it);
1520 		return (0);
1521 	}
1522 	return (EINVAL);
1523 }
1524 
1525 int
1526 itimespecfix(struct timespec *ts)
1527 {
1528 
1529 	if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1530 		return (EINVAL);
1531 	if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1532 		ts->tv_nsec = tick * 1000;
1533 	return (0);
1534 }
1535 
1536 /* Timeout callback for realtime timer */
1537 static void
1538 realtimer_expire(void *arg)
1539 {
1540 	struct timespec cts, ts;
1541 	struct timeval tv;
1542 	struct itimer *it;
1543 
1544 	it = (struct itimer *)arg;
1545 
1546 	realtimer_clocktime(it->it_clockid, &cts);
1547 	/* Only fire if time is reached. */
1548 	if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1549 		if (timespecisset(&it->it_time.it_interval)) {
1550 			timespecadd(&it->it_time.it_value,
1551 				    &it->it_time.it_interval);
1552 			while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1553 				if (it->it_overrun < INT_MAX)
1554 					it->it_overrun++;
1555 				else
1556 					it->it_ksi.ksi_errno = ERANGE;
1557 				timespecadd(&it->it_time.it_value,
1558 					    &it->it_time.it_interval);
1559 			}
1560 		} else {
1561 			/* single shot timer ? */
1562 			timespecclear(&it->it_time.it_value);
1563 		}
1564 		if (timespecisset(&it->it_time.it_value)) {
1565 			ts = it->it_time.it_value;
1566 			timespecsub(&ts, &cts);
1567 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1568 			callout_reset(&it->it_callout, tvtohz(&tv),
1569 				 realtimer_expire, it);
1570 		}
1571 		itimer_enter(it);
1572 		ITIMER_UNLOCK(it);
1573 		itimer_fire(it);
1574 		ITIMER_LOCK(it);
1575 		itimer_leave(it);
1576 	} else if (timespecisset(&it->it_time.it_value)) {
1577 		ts = it->it_time.it_value;
1578 		timespecsub(&ts, &cts);
1579 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1580 		callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1581  			it);
1582 	}
1583 }
1584 
1585 void
1586 itimer_fire(struct itimer *it)
1587 {
1588 	struct proc *p = it->it_proc;
1589 	struct thread *td;
1590 
1591 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1592 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1593 		if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1594 			ITIMER_LOCK(it);
1595 			timespecclear(&it->it_time.it_value);
1596 			timespecclear(&it->it_time.it_interval);
1597 			callout_stop(&it->it_callout);
1598 			ITIMER_UNLOCK(it);
1599 			return;
1600 		}
1601 		if (!KSI_ONQ(&it->it_ksi)) {
1602 			it->it_ksi.ksi_errno = 0;
1603 			ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1604 			tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1605 		} else {
1606 			if (it->it_overrun < INT_MAX)
1607 				it->it_overrun++;
1608 			else
1609 				it->it_ksi.ksi_errno = ERANGE;
1610 		}
1611 		PROC_UNLOCK(p);
1612 	}
1613 }
1614 
1615 static void
1616 itimers_alloc(struct proc *p)
1617 {
1618 	struct itimers *its;
1619 	int i;
1620 
1621 	its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1622 	LIST_INIT(&its->its_virtual);
1623 	LIST_INIT(&its->its_prof);
1624 	TAILQ_INIT(&its->its_worklist);
1625 	for (i = 0; i < TIMER_MAX; i++)
1626 		its->its_timers[i] = NULL;
1627 	PROC_LOCK(p);
1628 	if (p->p_itimers == NULL) {
1629 		p->p_itimers = its;
1630 		PROC_UNLOCK(p);
1631 	}
1632 	else {
1633 		PROC_UNLOCK(p);
1634 		free(its, M_SUBPROC);
1635 	}
1636 }
1637 
1638 static void
1639 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1640 {
1641 	itimers_event_hook_exit(arg, p);
1642 }
1643 
1644 /* Clean up timers when some process events are being triggered. */
1645 static void
1646 itimers_event_hook_exit(void *arg, struct proc *p)
1647 {
1648 	struct itimers *its;
1649 	struct itimer *it;
1650 	int event = (int)(intptr_t)arg;
1651 	int i;
1652 
1653 	if (p->p_itimers != NULL) {
1654 		its = p->p_itimers;
1655 		for (i = 0; i < MAX_CLOCKS; ++i) {
1656 			if (posix_clocks[i].event_hook != NULL)
1657 				CLOCK_CALL(i, event_hook, (p, i, event));
1658 		}
1659 		/*
1660 		 * According to susv3, XSI interval timers should be inherited
1661 		 * by new image.
1662 		 */
1663 		if (event == ITIMER_EV_EXEC)
1664 			i = 3;
1665 		else if (event == ITIMER_EV_EXIT)
1666 			i = 0;
1667 		else
1668 			panic("unhandled event");
1669 		for (; i < TIMER_MAX; ++i) {
1670 			if ((it = its->its_timers[i]) != NULL)
1671 				kern_ktimer_delete(curthread, i);
1672 		}
1673 		if (its->its_timers[0] == NULL &&
1674 		    its->its_timers[1] == NULL &&
1675 		    its->its_timers[2] == NULL) {
1676 			free(its, M_SUBPROC);
1677 			p->p_itimers = NULL;
1678 		}
1679 	}
1680 }
1681