1/*-
2 * Copyright (c) 1982, 1986, 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 *    may be used to endorse or promote products derived from this software
15 *    without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 *	@(#)kern_time.c	8.1 (Berkeley) 6/10/93
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: stable/10/sys/kern/kern_time.c 330422 2018-03-04 23:31:25Z bdrewery $");
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
74static struct kclock	posix_clocks[MAX_CLOCKS];
75static 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
87static int	settime(struct thread *, struct timeval *);
88static void	timevalfix(struct timeval *);
89
90static void	itimer_start(void);
91static int	itimer_init(void *, int, int);
92static void	itimer_fini(void *, int);
93static void	itimer_enter(struct itimer *);
94static void	itimer_leave(struct itimer *);
95static struct itimer *itimer_find(struct proc *, int);
96static void	itimers_alloc(struct proc *);
97static void	itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
98static void	itimers_event_hook_exit(void *arg, struct proc *p);
99static int	realtimer_create(struct itimer *);
100static int	realtimer_gettime(struct itimer *, struct itimerspec *);
101static int	realtimer_settime(struct itimer *, int,
102			struct itimerspec *, struct itimerspec *);
103static int	realtimer_delete(struct itimer *);
104static void	realtimer_clocktime(clockid_t, struct timespec *);
105static void	realtimer_expire(void *);
106
107int		register_posix_clock(int, struct kclock *);
108void		itimer_fire(struct itimer *it);
109int		itimespecfix(struct timespec *ts);
110
111#define CLOCK_CALL(clock, call, arglist)		\
112	((*posix_clocks[clock].call) arglist)
113
114SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
115
116
117static int
118settime(struct thread *td, struct timeval *tv)
119{
120	struct timeval delta, tv1, tv2;
121	static struct timeval maxtime, laststep;
122	struct timespec ts;
123	int s;
124
125	s = splclock();
126	microtime(&tv1);
127	delta = *tv;
128	timevalsub(&delta, &tv1);
129
130	/*
131	 * If the system is secure, we do not allow the time to be
132	 * set to a value earlier than 1 second less than the highest
133	 * time we have yet seen. The worst a miscreant can do in
134	 * this circumstance is "freeze" time. He couldn't go
135	 * back to the past.
136	 *
137	 * We similarly do not allow the clock to be stepped more
138	 * than one second, nor more than once per second. This allows
139	 * a miscreant to make the clock march double-time, but no worse.
140	 */
141	if (securelevel_gt(td->td_ucred, 1) != 0) {
142		if (delta.tv_sec < 0 || delta.tv_usec < 0) {
143			/*
144			 * Update maxtime to latest time we've seen.
145			 */
146			if (tv1.tv_sec > maxtime.tv_sec)
147				maxtime = tv1;
148			tv2 = *tv;
149			timevalsub(&tv2, &maxtime);
150			if (tv2.tv_sec < -1) {
151				tv->tv_sec = maxtime.tv_sec - 1;
152				printf("Time adjustment clamped to -1 second\n");
153			}
154		} else {
155			if (tv1.tv_sec == laststep.tv_sec) {
156				splx(s);
157				return (EPERM);
158			}
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	mtx_lock(&Giant);
170	tc_setclock(&ts);
171	resettodr();
172	mtx_unlock(&Giant);
173	return (0);
174}
175
176#ifndef _SYS_SYSPROTO_H_
177struct clock_getcpuclockid2_args {
178	id_t id;
179	int which,
180	clockid_t *clock_id;
181};
182#endif
183/* ARGSUSED */
184int
185sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
186{
187	clockid_t clk_id;
188	int error;
189
190	error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
191	if (error == 0)
192		error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
193	return (error);
194}
195
196int
197kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
198    clockid_t *clk_id)
199{
200	struct proc *p;
201	pid_t pid;
202	lwpid_t tid;
203	int error;
204
205	switch (which) {
206	case CPUCLOCK_WHICH_PID:
207		if (id != 0) {
208			error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
209			if (error != 0)
210				return (error);
211			PROC_UNLOCK(p);
212			pid = id;
213		} else {
214			pid = td->td_proc->p_pid;
215		}
216		*clk_id = MAKE_PROCESS_CPUCLOCK(pid);
217		return (0);
218	case CPUCLOCK_WHICH_TID:
219		tid = id == 0 ? td->td_tid : id;
220		*clk_id = MAKE_THREAD_CPUCLOCK(tid);
221		return (0);
222	default:
223		return (EINVAL);
224	}
225}
226
227#ifndef _SYS_SYSPROTO_H_
228struct clock_gettime_args {
229	clockid_t clock_id;
230	struct	timespec *tp;
231};
232#endif
233/* ARGSUSED */
234int
235sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
236{
237	struct timespec ats;
238	int error;
239
240	error = kern_clock_gettime(td, uap->clock_id, &ats);
241	if (error == 0)
242		error = copyout(&ats, uap->tp, sizeof(ats));
243
244	return (error);
245}
246
247static inline void
248cputick2timespec(uint64_t runtime, struct timespec *ats)
249{
250	runtime = cputick2usec(runtime);
251	ats->tv_sec = runtime / 1000000;
252	ats->tv_nsec = runtime % 1000000 * 1000;
253}
254
255static void
256get_thread_cputime(struct thread *targettd, struct timespec *ats)
257{
258	uint64_t runtime, curtime, switchtime;
259
260	if (targettd == NULL) { /* current thread */
261		critical_enter();
262		switchtime = PCPU_GET(switchtime);
263		curtime = cpu_ticks();
264		runtime = curthread->td_runtime;
265		critical_exit();
266		runtime += curtime - switchtime;
267	} else {
268		thread_lock(targettd);
269		runtime = targettd->td_runtime;
270		thread_unlock(targettd);
271	}
272	cputick2timespec(runtime, ats);
273}
274
275static void
276get_process_cputime(struct proc *targetp, struct timespec *ats)
277{
278	uint64_t runtime;
279	struct rusage ru;
280
281	PROC_STATLOCK(targetp);
282	rufetch(targetp, &ru);
283	runtime = targetp->p_rux.rux_runtime;
284	PROC_STATUNLOCK(targetp);
285	cputick2timespec(runtime, ats);
286}
287
288static int
289get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
290{
291	struct proc *p, *p2;
292	struct thread *td2;
293	lwpid_t tid;
294	pid_t pid;
295	int error;
296
297	p = td->td_proc;
298	if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
299		tid = clock_id & CPUCLOCK_ID_MASK;
300		td2 = tdfind(tid, p->p_pid);
301		if (td2 == NULL)
302			return (EINVAL);
303		get_thread_cputime(td2, ats);
304		PROC_UNLOCK(td2->td_proc);
305	} else {
306		pid = clock_id & CPUCLOCK_ID_MASK;
307		error = pget(pid, PGET_CANSEE, &p2);
308		if (error != 0)
309			return (EINVAL);
310		get_process_cputime(p2, ats);
311		PROC_UNLOCK(p2);
312	}
313	return (0);
314}
315
316int
317kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
318{
319	struct timeval sys, user;
320	struct proc *p;
321
322	p = td->td_proc;
323	switch (clock_id) {
324	case CLOCK_REALTIME:		/* Default to precise. */
325	case CLOCK_REALTIME_PRECISE:
326		nanotime(ats);
327		break;
328	case CLOCK_REALTIME_FAST:
329		getnanotime(ats);
330		break;
331	case CLOCK_VIRTUAL:
332		PROC_LOCK(p);
333		PROC_STATLOCK(p);
334		calcru(p, &user, &sys);
335		PROC_STATUNLOCK(p);
336		PROC_UNLOCK(p);
337		TIMEVAL_TO_TIMESPEC(&user, ats);
338		break;
339	case CLOCK_PROF:
340		PROC_LOCK(p);
341		PROC_STATLOCK(p);
342		calcru(p, &user, &sys);
343		PROC_STATUNLOCK(p);
344		PROC_UNLOCK(p);
345		timevaladd(&user, &sys);
346		TIMEVAL_TO_TIMESPEC(&user, ats);
347		break;
348	case CLOCK_MONOTONIC:		/* Default to precise. */
349	case CLOCK_MONOTONIC_PRECISE:
350	case CLOCK_UPTIME:
351	case CLOCK_UPTIME_PRECISE:
352		nanouptime(ats);
353		break;
354	case CLOCK_UPTIME_FAST:
355	case CLOCK_MONOTONIC_FAST:
356		getnanouptime(ats);
357		break;
358	case CLOCK_SECOND:
359		ats->tv_sec = time_second;
360		ats->tv_nsec = 0;
361		break;
362	case CLOCK_THREAD_CPUTIME_ID:
363		get_thread_cputime(NULL, ats);
364		break;
365	case CLOCK_PROCESS_CPUTIME_ID:
366		PROC_LOCK(p);
367		get_process_cputime(p, ats);
368		PROC_UNLOCK(p);
369		break;
370	default:
371		if ((int)clock_id >= 0)
372			return (EINVAL);
373		return (get_cputime(td, clock_id, ats));
374	}
375	return (0);
376}
377
378#ifndef _SYS_SYSPROTO_H_
379struct clock_settime_args {
380	clockid_t clock_id;
381	const struct	timespec *tp;
382};
383#endif
384/* ARGSUSED */
385int
386sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
387{
388	struct timespec ats;
389	int error;
390
391	if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
392		return (error);
393	return (kern_clock_settime(td, uap->clock_id, &ats));
394}
395
396int
397kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
398{
399	struct timeval atv;
400	int error;
401
402	if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
403		return (error);
404	if (clock_id != CLOCK_REALTIME)
405		return (EINVAL);
406	if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
407		return (EINVAL);
408	/* XXX Don't convert nsec->usec and back */
409	TIMESPEC_TO_TIMEVAL(&atv, ats);
410	error = settime(td, &atv);
411	return (error);
412}
413
414#ifndef _SYS_SYSPROTO_H_
415struct clock_getres_args {
416	clockid_t clock_id;
417	struct	timespec *tp;
418};
419#endif
420int
421sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
422{
423	struct timespec ts;
424	int error;
425
426	if (uap->tp == NULL)
427		return (0);
428
429	error = kern_clock_getres(td, uap->clock_id, &ts);
430	if (error == 0)
431		error = copyout(&ts, uap->tp, sizeof(ts));
432	return (error);
433}
434
435int
436kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
437{
438
439	ts->tv_sec = 0;
440	switch (clock_id) {
441	case CLOCK_REALTIME:
442	case CLOCK_REALTIME_FAST:
443	case CLOCK_REALTIME_PRECISE:
444	case CLOCK_MONOTONIC:
445	case CLOCK_MONOTONIC_FAST:
446	case CLOCK_MONOTONIC_PRECISE:
447	case CLOCK_UPTIME:
448	case CLOCK_UPTIME_FAST:
449	case CLOCK_UPTIME_PRECISE:
450		/*
451		 * Round up the result of the division cheaply by adding 1.
452		 * Rounding up is especially important if rounding down
453		 * would give 0.  Perfect rounding is unimportant.
454		 */
455		ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
456		break;
457	case CLOCK_VIRTUAL:
458	case CLOCK_PROF:
459		/* Accurately round up here because we can do so cheaply. */
460		ts->tv_nsec = (1000000000 + hz - 1) / hz;
461		break;
462	case CLOCK_SECOND:
463		ts->tv_sec = 1;
464		ts->tv_nsec = 0;
465		break;
466	case CLOCK_THREAD_CPUTIME_ID:
467	case CLOCK_PROCESS_CPUTIME_ID:
468	cputime:
469		/* sync with cputick2usec */
470		ts->tv_nsec = 1000000 / cpu_tickrate();
471		if (ts->tv_nsec == 0)
472			ts->tv_nsec = 1000;
473		break;
474	default:
475		if ((int)clock_id < 0)
476			goto cputime;
477		return (EINVAL);
478	}
479	return (0);
480}
481
482static uint8_t nanowait[MAXCPU];
483
484int
485kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
486{
487	struct timespec ts;
488	sbintime_t sbt, sbtt, prec, tmp;
489	time_t over;
490	int error;
491
492	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
493		return (EINVAL);
494	if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
495		return (0);
496	ts = *rqt;
497	if (ts.tv_sec > INT32_MAX / 2) {
498		over = ts.tv_sec - INT32_MAX / 2;
499		ts.tv_sec -= over;
500	} else
501		over = 0;
502	tmp = tstosbt(ts);
503	prec = tmp;
504	prec >>= tc_precexp;
505	if (TIMESEL(&sbt, tmp))
506		sbt += tc_tick_sbt;
507	sbt += tmp;
508	error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
509	    sbt, prec, C_ABSOLUTE);
510	if (error != EWOULDBLOCK) {
511		if (error == ERESTART)
512			error = EINTR;
513		if (TIMESEL(&sbtt, tmp))
514			sbtt += tc_tick_sbt;
515		if (rmt != NULL) {
516			ts = sbttots(sbt - sbtt);
517			ts.tv_sec += over;
518			if (ts.tv_sec < 0)
519				timespecclear(&ts);
520			*rmt = ts;
521		}
522		if (sbtt >= sbt)
523			return (0);
524		return (error);
525	}
526	return (0);
527}
528
529#ifndef _SYS_SYSPROTO_H_
530struct nanosleep_args {
531	struct	timespec *rqtp;
532	struct	timespec *rmtp;
533};
534#endif
535/* ARGSUSED */
536int
537sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
538{
539	struct timespec rmt, rqt;
540	int error;
541
542	error = copyin(uap->rqtp, &rqt, sizeof(rqt));
543	if (error)
544		return (error);
545
546	if (uap->rmtp &&
547	    !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE))
548			return (EFAULT);
549	error = kern_nanosleep(td, &rqt, &rmt);
550	if (error == EINTR && uap->rmtp) {
551		int error2;
552
553		error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
554		if (error2)
555			error = error2;
556	}
557	return (error);
558}
559
560#ifndef _SYS_SYSPROTO_H_
561struct gettimeofday_args {
562	struct	timeval *tp;
563	struct	timezone *tzp;
564};
565#endif
566/* ARGSUSED */
567int
568sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
569{
570	struct timeval atv;
571	struct timezone rtz;
572	int error = 0;
573
574	if (uap->tp) {
575		microtime(&atv);
576		error = copyout(&atv, uap->tp, sizeof (atv));
577	}
578	if (error == 0 && uap->tzp != NULL) {
579		rtz.tz_minuteswest = tz_minuteswest;
580		rtz.tz_dsttime = tz_dsttime;
581		error = copyout(&rtz, uap->tzp, sizeof (rtz));
582	}
583	return (error);
584}
585
586#ifndef _SYS_SYSPROTO_H_
587struct settimeofday_args {
588	struct	timeval *tv;
589	struct	timezone *tzp;
590};
591#endif
592/* ARGSUSED */
593int
594sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
595{
596	struct timeval atv, *tvp;
597	struct timezone atz, *tzp;
598	int error;
599
600	if (uap->tv) {
601		error = copyin(uap->tv, &atv, sizeof(atv));
602		if (error)
603			return (error);
604		tvp = &atv;
605	} else
606		tvp = NULL;
607	if (uap->tzp) {
608		error = copyin(uap->tzp, &atz, sizeof(atz));
609		if (error)
610			return (error);
611		tzp = &atz;
612	} else
613		tzp = NULL;
614	return (kern_settimeofday(td, tvp, tzp));
615}
616
617int
618kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
619{
620	int error;
621
622	error = priv_check(td, PRIV_SETTIMEOFDAY);
623	if (error)
624		return (error);
625	/* Verify all parameters before changing time. */
626	if (tv) {
627		if (tv->tv_usec < 0 || tv->tv_usec >= 1000000)
628			return (EINVAL);
629		error = settime(td, tv);
630	}
631	if (tzp && error == 0) {
632		tz_minuteswest = tzp->tz_minuteswest;
633		tz_dsttime = tzp->tz_dsttime;
634	}
635	return (error);
636}
637
638/*
639 * Get value of an interval timer.  The process virtual and profiling virtual
640 * time timers are kept in the p_stats area, since they can be swapped out.
641 * These are kept internally in the way they are specified externally: in
642 * time until they expire.
643 *
644 * The real time interval timer is kept in the process table slot for the
645 * process, and its value (it_value) is kept as an absolute time rather than
646 * as a delta, so that it is easy to keep periodic real-time signals from
647 * drifting.
648 *
649 * Virtual time timers are processed in the hardclock() routine of
650 * kern_clock.c.  The real time timer is processed by a timeout routine,
651 * called from the softclock() routine.  Since a callout may be delayed in
652 * real time due to interrupt processing in the system, it is possible for
653 * the real time timeout routine (realitexpire, given below), to be delayed
654 * in real time past when it is supposed to occur.  It does not suffice,
655 * therefore, to reload the real timer .it_value from the real time timers
656 * .it_interval.  Rather, we compute the next time in absolute time the timer
657 * should go off.
658 */
659#ifndef _SYS_SYSPROTO_H_
660struct getitimer_args {
661	u_int	which;
662	struct	itimerval *itv;
663};
664#endif
665int
666sys_getitimer(struct thread *td, struct getitimer_args *uap)
667{
668	struct itimerval aitv;
669	int error;
670
671	error = kern_getitimer(td, uap->which, &aitv);
672	if (error != 0)
673		return (error);
674	return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
675}
676
677int
678kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
679{
680	struct proc *p = td->td_proc;
681	struct timeval ctv;
682
683	if (which > ITIMER_PROF)
684		return (EINVAL);
685
686	if (which == ITIMER_REAL) {
687		/*
688		 * Convert from absolute to relative time in .it_value
689		 * part of real time timer.  If time for real time timer
690		 * has passed return 0, else return difference between
691		 * current time and time for the timer to go off.
692		 */
693		PROC_LOCK(p);
694		*aitv = p->p_realtimer;
695		PROC_UNLOCK(p);
696		if (timevalisset(&aitv->it_value)) {
697			microuptime(&ctv);
698			if (timevalcmp(&aitv->it_value, &ctv, <))
699				timevalclear(&aitv->it_value);
700			else
701				timevalsub(&aitv->it_value, &ctv);
702		}
703	} else {
704		PROC_ITIMLOCK(p);
705		*aitv = p->p_stats->p_timer[which];
706		PROC_ITIMUNLOCK(p);
707	}
708#ifdef KTRACE
709	if (KTRPOINT(td, KTR_STRUCT))
710		ktritimerval(aitv);
711#endif
712	return (0);
713}
714
715#ifndef _SYS_SYSPROTO_H_
716struct setitimer_args {
717	u_int	which;
718	struct	itimerval *itv, *oitv;
719};
720#endif
721int
722sys_setitimer(struct thread *td, struct setitimer_args *uap)
723{
724	struct itimerval aitv, oitv;
725	int error;
726
727	if (uap->itv == NULL) {
728		uap->itv = uap->oitv;
729		return (sys_getitimer(td, (struct getitimer_args *)uap));
730	}
731
732	if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
733		return (error);
734	error = kern_setitimer(td, uap->which, &aitv, &oitv);
735	if (error != 0 || uap->oitv == NULL)
736		return (error);
737	return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
738}
739
740int
741kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
742    struct itimerval *oitv)
743{
744	struct proc *p = td->td_proc;
745	struct timeval ctv;
746	sbintime_t sbt, pr;
747
748	if (aitv == NULL)
749		return (kern_getitimer(td, which, oitv));
750
751	if (which > ITIMER_PROF)
752		return (EINVAL);
753#ifdef KTRACE
754	if (KTRPOINT(td, KTR_STRUCT))
755		ktritimerval(aitv);
756#endif
757	if (itimerfix(&aitv->it_value) ||
758	    aitv->it_value.tv_sec > INT32_MAX / 2)
759		return (EINVAL);
760	if (!timevalisset(&aitv->it_value))
761		timevalclear(&aitv->it_interval);
762	else if (itimerfix(&aitv->it_interval) ||
763	    aitv->it_interval.tv_sec > INT32_MAX / 2)
764		return (EINVAL);
765
766	if (which == ITIMER_REAL) {
767		PROC_LOCK(p);
768		if (timevalisset(&p->p_realtimer.it_value))
769			callout_stop(&p->p_itcallout);
770		microuptime(&ctv);
771		if (timevalisset(&aitv->it_value)) {
772			pr = tvtosbt(aitv->it_value) >> tc_precexp;
773			timevaladd(&aitv->it_value, &ctv);
774			sbt = tvtosbt(aitv->it_value);
775			callout_reset_sbt(&p->p_itcallout, sbt, pr,
776			    realitexpire, p, C_ABSOLUTE);
777		}
778		*oitv = p->p_realtimer;
779		p->p_realtimer = *aitv;
780		PROC_UNLOCK(p);
781		if (timevalisset(&oitv->it_value)) {
782			if (timevalcmp(&oitv->it_value, &ctv, <))
783				timevalclear(&oitv->it_value);
784			else
785				timevalsub(&oitv->it_value, &ctv);
786		}
787	} else {
788		if (aitv->it_interval.tv_sec == 0 &&
789		    aitv->it_interval.tv_usec != 0 &&
790		    aitv->it_interval.tv_usec < tick)
791			aitv->it_interval.tv_usec = tick;
792		if (aitv->it_value.tv_sec == 0 &&
793		    aitv->it_value.tv_usec != 0 &&
794		    aitv->it_value.tv_usec < tick)
795			aitv->it_value.tv_usec = tick;
796		PROC_ITIMLOCK(p);
797		*oitv = p->p_stats->p_timer[which];
798		p->p_stats->p_timer[which] = *aitv;
799		PROC_ITIMUNLOCK(p);
800	}
801#ifdef KTRACE
802	if (KTRPOINT(td, KTR_STRUCT))
803		ktritimerval(oitv);
804#endif
805	return (0);
806}
807
808/*
809 * Real interval timer expired:
810 * send process whose timer expired an alarm signal.
811 * If time is not set up to reload, then just return.
812 * Else compute next time timer should go off which is > current time.
813 * This is where delay in processing this timeout causes multiple
814 * SIGALRM calls to be compressed into one.
815 * tvtohz() always adds 1 to allow for the time until the next clock
816 * interrupt being strictly less than 1 clock tick, but we don't want
817 * that here since we want to appear to be in sync with the clock
818 * interrupt even when we're delayed.
819 */
820void
821realitexpire(void *arg)
822{
823	struct proc *p;
824	struct timeval ctv;
825	sbintime_t isbt;
826
827	p = (struct proc *)arg;
828	kern_psignal(p, SIGALRM);
829	if (!timevalisset(&p->p_realtimer.it_interval)) {
830		timevalclear(&p->p_realtimer.it_value);
831		if (p->p_flag & P_WEXIT)
832			wakeup(&p->p_itcallout);
833		return;
834	}
835	isbt = tvtosbt(p->p_realtimer.it_interval);
836	if (isbt >= sbt_timethreshold)
837		getmicrouptime(&ctv);
838	else
839		microuptime(&ctv);
840	do {
841		timevaladd(&p->p_realtimer.it_value,
842		    &p->p_realtimer.it_interval);
843	} while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
844	callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
845	    isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE);
846}
847
848/*
849 * Check that a proposed value to load into the .it_value or
850 * .it_interval part of an interval timer is acceptable, and
851 * fix it to have at least minimal value (i.e. if it is less
852 * than the resolution of the clock, round it up.)
853 */
854int
855itimerfix(struct timeval *tv)
856{
857
858	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
859		return (EINVAL);
860	if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
861	    tv->tv_usec < (u_int)tick / 16)
862		tv->tv_usec = (u_int)tick / 16;
863	return (0);
864}
865
866/*
867 * Decrement an interval timer by a specified number
868 * of microseconds, which must be less than a second,
869 * i.e. < 1000000.  If the timer expires, then reload
870 * it.  In this case, carry over (usec - old value) to
871 * reduce the value reloaded into the timer so that
872 * the timer does not drift.  This routine assumes
873 * that it is called in a context where the timers
874 * on which it is operating cannot change in value.
875 */
876int
877itimerdecr(struct itimerval *itp, int usec)
878{
879
880	if (itp->it_value.tv_usec < usec) {
881		if (itp->it_value.tv_sec == 0) {
882			/* expired, and already in next interval */
883			usec -= itp->it_value.tv_usec;
884			goto expire;
885		}
886		itp->it_value.tv_usec += 1000000;
887		itp->it_value.tv_sec--;
888	}
889	itp->it_value.tv_usec -= usec;
890	usec = 0;
891	if (timevalisset(&itp->it_value))
892		return (1);
893	/* expired, exactly at end of interval */
894expire:
895	if (timevalisset(&itp->it_interval)) {
896		itp->it_value = itp->it_interval;
897		itp->it_value.tv_usec -= usec;
898		if (itp->it_value.tv_usec < 0) {
899			itp->it_value.tv_usec += 1000000;
900			itp->it_value.tv_sec--;
901		}
902	} else
903		itp->it_value.tv_usec = 0;		/* sec is already 0 */
904	return (0);
905}
906
907/*
908 * Add and subtract routines for timevals.
909 * N.B.: subtract routine doesn't deal with
910 * results which are before the beginning,
911 * it just gets very confused in this case.
912 * Caveat emptor.
913 */
914void
915timevaladd(struct timeval *t1, const struct timeval *t2)
916{
917
918	t1->tv_sec += t2->tv_sec;
919	t1->tv_usec += t2->tv_usec;
920	timevalfix(t1);
921}
922
923void
924timevalsub(struct timeval *t1, const struct timeval *t2)
925{
926
927	t1->tv_sec -= t2->tv_sec;
928	t1->tv_usec -= t2->tv_usec;
929	timevalfix(t1);
930}
931
932static void
933timevalfix(struct timeval *t1)
934{
935
936	if (t1->tv_usec < 0) {
937		t1->tv_sec--;
938		t1->tv_usec += 1000000;
939	}
940	if (t1->tv_usec >= 1000000) {
941		t1->tv_sec++;
942		t1->tv_usec -= 1000000;
943	}
944}
945
946/*
947 * ratecheck(): simple time-based rate-limit checking.
948 */
949int
950ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
951{
952	struct timeval tv, delta;
953	int rv = 0;
954
955	getmicrouptime(&tv);		/* NB: 10ms precision */
956	delta = tv;
957	timevalsub(&delta, lasttime);
958
959	/*
960	 * check for 0,0 is so that the message will be seen at least once,
961	 * even if interval is huge.
962	 */
963	if (timevalcmp(&delta, mininterval, >=) ||
964	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
965		*lasttime = tv;
966		rv = 1;
967	}
968
969	return (rv);
970}
971
972/*
973 * ppsratecheck(): packets (or events) per second limitation.
974 *
975 * Return 0 if the limit is to be enforced (e.g. the caller
976 * should drop a packet because of the rate limitation).
977 *
978 * maxpps of 0 always causes zero to be returned.  maxpps of -1
979 * always causes 1 to be returned; this effectively defeats rate
980 * limiting.
981 *
982 * Note that we maintain the struct timeval for compatibility
983 * with other bsd systems.  We reuse the storage and just monitor
984 * clock ticks for minimal overhead.
985 */
986int
987ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
988{
989	int now;
990
991	/*
992	 * Reset the last time and counter if this is the first call
993	 * or more than a second has passed since the last update of
994	 * lasttime.
995	 */
996	now = ticks;
997	if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
998		lasttime->tv_sec = now;
999		*curpps = 1;
1000		return (maxpps != 0);
1001	} else {
1002		(*curpps)++;		/* NB: ignore potential overflow */
1003		return (maxpps < 0 || *curpps <= maxpps);
1004	}
1005}
1006
1007static void
1008itimer_start(void)
1009{
1010	struct kclock rt_clock = {
1011		.timer_create  = realtimer_create,
1012		.timer_delete  = realtimer_delete,
1013		.timer_settime = realtimer_settime,
1014		.timer_gettime = realtimer_gettime,
1015		.event_hook    = NULL
1016	};
1017
1018	itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1019		NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1020	register_posix_clock(CLOCK_REALTIME,  &rt_clock);
1021	register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1022	p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1023	p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1024	p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1025	EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
1026		(void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
1027	EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
1028		(void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
1029}
1030
1031int
1032register_posix_clock(int clockid, struct kclock *clk)
1033{
1034	if ((unsigned)clockid >= MAX_CLOCKS) {
1035		printf("%s: invalid clockid\n", __func__);
1036		return (0);
1037	}
1038	posix_clocks[clockid] = *clk;
1039	return (1);
1040}
1041
1042static int
1043itimer_init(void *mem, int size, int flags)
1044{
1045	struct itimer *it;
1046
1047	it = (struct itimer *)mem;
1048	mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1049	return (0);
1050}
1051
1052static void
1053itimer_fini(void *mem, int size)
1054{
1055	struct itimer *it;
1056
1057	it = (struct itimer *)mem;
1058	mtx_destroy(&it->it_mtx);
1059}
1060
1061static void
1062itimer_enter(struct itimer *it)
1063{
1064
1065	mtx_assert(&it->it_mtx, MA_OWNED);
1066	it->it_usecount++;
1067}
1068
1069static void
1070itimer_leave(struct itimer *it)
1071{
1072
1073	mtx_assert(&it->it_mtx, MA_OWNED);
1074	KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1075
1076	if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1077		wakeup(it);
1078}
1079
1080#ifndef _SYS_SYSPROTO_H_
1081struct ktimer_create_args {
1082	clockid_t clock_id;
1083	struct sigevent * evp;
1084	int * timerid;
1085};
1086#endif
1087int
1088sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1089{
1090	struct sigevent *evp, ev;
1091	int id;
1092	int error;
1093
1094	if (uap->evp == NULL) {
1095		evp = NULL;
1096	} else {
1097		error = copyin(uap->evp, &ev, sizeof(ev));
1098		if (error != 0)
1099			return (error);
1100		evp = &ev;
1101	}
1102	error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1103	if (error == 0) {
1104		error = copyout(&id, uap->timerid, sizeof(int));
1105		if (error != 0)
1106			kern_ktimer_delete(td, id);
1107	}
1108	return (error);
1109}
1110
1111int
1112kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1113    int *timerid, int preset_id)
1114{
1115	struct proc *p = td->td_proc;
1116	struct itimer *it;
1117	int id;
1118	int error;
1119
1120	if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1121		return (EINVAL);
1122
1123	if (posix_clocks[clock_id].timer_create == NULL)
1124		return (EINVAL);
1125
1126	if (evp != NULL) {
1127		if (evp->sigev_notify != SIGEV_NONE &&
1128		    evp->sigev_notify != SIGEV_SIGNAL &&
1129		    evp->sigev_notify != SIGEV_THREAD_ID)
1130			return (EINVAL);
1131		if ((evp->sigev_notify == SIGEV_SIGNAL ||
1132		     evp->sigev_notify == SIGEV_THREAD_ID) &&
1133			!_SIG_VALID(evp->sigev_signo))
1134			return (EINVAL);
1135	}
1136
1137	if (p->p_itimers == NULL)
1138		itimers_alloc(p);
1139
1140	it = uma_zalloc(itimer_zone, M_WAITOK);
1141	it->it_flags = 0;
1142	it->it_usecount = 0;
1143	it->it_active = 0;
1144	timespecclear(&it->it_time.it_value);
1145	timespecclear(&it->it_time.it_interval);
1146	it->it_overrun = 0;
1147	it->it_overrun_last = 0;
1148	it->it_clockid = clock_id;
1149	it->it_timerid = -1;
1150	it->it_proc = p;
1151	ksiginfo_init(&it->it_ksi);
1152	it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1153	error = CLOCK_CALL(clock_id, timer_create, (it));
1154	if (error != 0)
1155		goto out;
1156
1157	PROC_LOCK(p);
1158	if (preset_id != -1) {
1159		KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1160		id = preset_id;
1161		if (p->p_itimers->its_timers[id] != NULL) {
1162			PROC_UNLOCK(p);
1163			error = 0;
1164			goto out;
1165		}
1166	} else {
1167		/*
1168		 * Find a free timer slot, skipping those reserved
1169		 * for setitimer().
1170		 */
1171		for (id = 3; id < TIMER_MAX; id++)
1172			if (p->p_itimers->its_timers[id] == NULL)
1173				break;
1174		if (id == TIMER_MAX) {
1175			PROC_UNLOCK(p);
1176			error = EAGAIN;
1177			goto out;
1178		}
1179	}
1180	it->it_timerid = id;
1181	p->p_itimers->its_timers[id] = it;
1182	if (evp != NULL)
1183		it->it_sigev = *evp;
1184	else {
1185		it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1186		switch (clock_id) {
1187		default:
1188		case CLOCK_REALTIME:
1189			it->it_sigev.sigev_signo = SIGALRM;
1190			break;
1191		case CLOCK_VIRTUAL:
1192 			it->it_sigev.sigev_signo = SIGVTALRM;
1193			break;
1194		case CLOCK_PROF:
1195			it->it_sigev.sigev_signo = SIGPROF;
1196			break;
1197		}
1198		it->it_sigev.sigev_value.sival_int = id;
1199	}
1200
1201	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1202	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1203		it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1204		it->it_ksi.ksi_code = SI_TIMER;
1205		it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1206		it->it_ksi.ksi_timerid = id;
1207	}
1208	PROC_UNLOCK(p);
1209	*timerid = id;
1210	return (0);
1211
1212out:
1213	ITIMER_LOCK(it);
1214	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1215	ITIMER_UNLOCK(it);
1216	uma_zfree(itimer_zone, it);
1217	return (error);
1218}
1219
1220#ifndef _SYS_SYSPROTO_H_
1221struct ktimer_delete_args {
1222	int timerid;
1223};
1224#endif
1225int
1226sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1227{
1228
1229	return (kern_ktimer_delete(td, uap->timerid));
1230}
1231
1232static struct itimer *
1233itimer_find(struct proc *p, int timerid)
1234{
1235	struct itimer *it;
1236
1237	PROC_LOCK_ASSERT(p, MA_OWNED);
1238	if ((p->p_itimers == NULL) ||
1239	    (timerid < 0) || (timerid >= TIMER_MAX) ||
1240	    (it = p->p_itimers->its_timers[timerid]) == NULL) {
1241		return (NULL);
1242	}
1243	ITIMER_LOCK(it);
1244	if ((it->it_flags & ITF_DELETING) != 0) {
1245		ITIMER_UNLOCK(it);
1246		it = NULL;
1247	}
1248	return (it);
1249}
1250
1251int
1252kern_ktimer_delete(struct thread *td, int timerid)
1253{
1254	struct proc *p = td->td_proc;
1255	struct itimer *it;
1256
1257	PROC_LOCK(p);
1258	it = itimer_find(p, timerid);
1259	if (it == NULL) {
1260		PROC_UNLOCK(p);
1261		return (EINVAL);
1262	}
1263	PROC_UNLOCK(p);
1264
1265	it->it_flags |= ITF_DELETING;
1266	while (it->it_usecount > 0) {
1267		it->it_flags |= ITF_WANTED;
1268		msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1269	}
1270	it->it_flags &= ~ITF_WANTED;
1271	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1272	ITIMER_UNLOCK(it);
1273
1274	PROC_LOCK(p);
1275	if (KSI_ONQ(&it->it_ksi))
1276		sigqueue_take(&it->it_ksi);
1277	p->p_itimers->its_timers[timerid] = NULL;
1278	PROC_UNLOCK(p);
1279	uma_zfree(itimer_zone, it);
1280	return (0);
1281}
1282
1283#ifndef _SYS_SYSPROTO_H_
1284struct ktimer_settime_args {
1285	int timerid;
1286	int flags;
1287	const struct itimerspec * value;
1288	struct itimerspec * ovalue;
1289};
1290#endif
1291int
1292sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1293{
1294	struct itimerspec val, oval, *ovalp;
1295	int error;
1296
1297	error = copyin(uap->value, &val, sizeof(val));
1298	if (error != 0)
1299		return (error);
1300	ovalp = uap->ovalue != NULL ? &oval : NULL;
1301	error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1302	if (error == 0 && uap->ovalue != NULL)
1303		error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1304	return (error);
1305}
1306
1307int
1308kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1309    struct itimerspec *val, struct itimerspec *oval)
1310{
1311	struct proc *p;
1312	struct itimer *it;
1313	int error;
1314
1315	p = td->td_proc;
1316	PROC_LOCK(p);
1317	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1318		PROC_UNLOCK(p);
1319		error = EINVAL;
1320	} else {
1321		PROC_UNLOCK(p);
1322		itimer_enter(it);
1323		error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1324		    flags, val, oval));
1325		itimer_leave(it);
1326		ITIMER_UNLOCK(it);
1327	}
1328	return (error);
1329}
1330
1331#ifndef _SYS_SYSPROTO_H_
1332struct ktimer_gettime_args {
1333	int timerid;
1334	struct itimerspec * value;
1335};
1336#endif
1337int
1338sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1339{
1340	struct itimerspec val;
1341	int error;
1342
1343	error = kern_ktimer_gettime(td, uap->timerid, &val);
1344	if (error == 0)
1345		error = copyout(&val, uap->value, sizeof(val));
1346	return (error);
1347}
1348
1349int
1350kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1351{
1352	struct proc *p;
1353	struct itimer *it;
1354	int error;
1355
1356	p = td->td_proc;
1357	PROC_LOCK(p);
1358	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1359		PROC_UNLOCK(p);
1360		error = EINVAL;
1361	} else {
1362		PROC_UNLOCK(p);
1363		itimer_enter(it);
1364		error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1365		itimer_leave(it);
1366		ITIMER_UNLOCK(it);
1367	}
1368	return (error);
1369}
1370
1371#ifndef _SYS_SYSPROTO_H_
1372struct timer_getoverrun_args {
1373	int timerid;
1374};
1375#endif
1376int
1377sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1378{
1379
1380	return (kern_ktimer_getoverrun(td, uap->timerid));
1381}
1382
1383int
1384kern_ktimer_getoverrun(struct thread *td, int timer_id)
1385{
1386	struct proc *p = td->td_proc;
1387	struct itimer *it;
1388	int error ;
1389
1390	PROC_LOCK(p);
1391	if (timer_id < 3 ||
1392	    (it = itimer_find(p, timer_id)) == NULL) {
1393		PROC_UNLOCK(p);
1394		error = EINVAL;
1395	} else {
1396		td->td_retval[0] = it->it_overrun_last;
1397		ITIMER_UNLOCK(it);
1398		PROC_UNLOCK(p);
1399		error = 0;
1400	}
1401	return (error);
1402}
1403
1404static int
1405realtimer_create(struct itimer *it)
1406{
1407	callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1408	return (0);
1409}
1410
1411static int
1412realtimer_delete(struct itimer *it)
1413{
1414	mtx_assert(&it->it_mtx, MA_OWNED);
1415
1416	/*
1417	 * clear timer's value and interval to tell realtimer_expire
1418	 * to not rearm the timer.
1419	 */
1420	timespecclear(&it->it_time.it_value);
1421	timespecclear(&it->it_time.it_interval);
1422	ITIMER_UNLOCK(it);
1423	callout_drain(&it->it_callout);
1424	ITIMER_LOCK(it);
1425	return (0);
1426}
1427
1428static int
1429realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1430{
1431	struct timespec cts;
1432
1433	mtx_assert(&it->it_mtx, MA_OWNED);
1434
1435	realtimer_clocktime(it->it_clockid, &cts);
1436	*ovalue = it->it_time;
1437	if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1438		timespecsub(&ovalue->it_value, &cts);
1439		if (ovalue->it_value.tv_sec < 0 ||
1440		    (ovalue->it_value.tv_sec == 0 &&
1441		     ovalue->it_value.tv_nsec == 0)) {
1442			ovalue->it_value.tv_sec  = 0;
1443			ovalue->it_value.tv_nsec = 1;
1444		}
1445	}
1446	return (0);
1447}
1448
1449static int
1450realtimer_settime(struct itimer *it, int flags,
1451	struct itimerspec *value, struct itimerspec *ovalue)
1452{
1453	struct timespec cts, ts;
1454	struct timeval tv;
1455	struct itimerspec val;
1456
1457	mtx_assert(&it->it_mtx, MA_OWNED);
1458
1459	val = *value;
1460	if (itimespecfix(&val.it_value))
1461		return (EINVAL);
1462
1463	if (timespecisset(&val.it_value)) {
1464		if (itimespecfix(&val.it_interval))
1465			return (EINVAL);
1466	} else {
1467		timespecclear(&val.it_interval);
1468	}
1469
1470	if (ovalue != NULL)
1471		realtimer_gettime(it, ovalue);
1472
1473	it->it_time = val;
1474	if (timespecisset(&val.it_value)) {
1475		realtimer_clocktime(it->it_clockid, &cts);
1476		ts = val.it_value;
1477		if ((flags & TIMER_ABSTIME) == 0) {
1478			/* Convert to absolute time. */
1479			timespecadd(&it->it_time.it_value, &cts);
1480		} else {
1481			timespecsub(&ts, &cts);
1482			/*
1483			 * We don't care if ts is negative, tztohz will
1484			 * fix it.
1485			 */
1486		}
1487		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1488		callout_reset(&it->it_callout, tvtohz(&tv),
1489			realtimer_expire, it);
1490	} else {
1491		callout_stop(&it->it_callout);
1492	}
1493
1494	return (0);
1495}
1496
1497static void
1498realtimer_clocktime(clockid_t id, struct timespec *ts)
1499{
1500	if (id == CLOCK_REALTIME)
1501		getnanotime(ts);
1502	else	/* CLOCK_MONOTONIC */
1503		getnanouptime(ts);
1504}
1505
1506int
1507itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1508{
1509	struct itimer *it;
1510
1511	PROC_LOCK_ASSERT(p, MA_OWNED);
1512	it = itimer_find(p, timerid);
1513	if (it != NULL) {
1514		ksi->ksi_overrun = it->it_overrun;
1515		it->it_overrun_last = it->it_overrun;
1516		it->it_overrun = 0;
1517		ITIMER_UNLOCK(it);
1518		return (0);
1519	}
1520	return (EINVAL);
1521}
1522
1523int
1524itimespecfix(struct timespec *ts)
1525{
1526
1527	if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1528		return (EINVAL);
1529	if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1530		ts->tv_nsec = tick * 1000;
1531	return (0);
1532}
1533
1534/* Timeout callback for realtime timer */
1535static void
1536realtimer_expire(void *arg)
1537{
1538	struct timespec cts, ts;
1539	struct timeval tv;
1540	struct itimer *it;
1541
1542	it = (struct itimer *)arg;
1543
1544	realtimer_clocktime(it->it_clockid, &cts);
1545	/* Only fire if time is reached. */
1546	if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1547		if (timespecisset(&it->it_time.it_interval)) {
1548			timespecadd(&it->it_time.it_value,
1549				    &it->it_time.it_interval);
1550			while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1551				if (it->it_overrun < INT_MAX)
1552					it->it_overrun++;
1553				else
1554					it->it_ksi.ksi_errno = ERANGE;
1555				timespecadd(&it->it_time.it_value,
1556					    &it->it_time.it_interval);
1557			}
1558		} else {
1559			/* single shot timer ? */
1560			timespecclear(&it->it_time.it_value);
1561		}
1562		if (timespecisset(&it->it_time.it_value)) {
1563			ts = it->it_time.it_value;
1564			timespecsub(&ts, &cts);
1565			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1566			callout_reset(&it->it_callout, tvtohz(&tv),
1567				 realtimer_expire, it);
1568		}
1569		itimer_enter(it);
1570		ITIMER_UNLOCK(it);
1571		itimer_fire(it);
1572		ITIMER_LOCK(it);
1573		itimer_leave(it);
1574	} else if (timespecisset(&it->it_time.it_value)) {
1575		ts = it->it_time.it_value;
1576		timespecsub(&ts, &cts);
1577		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1578		callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1579 			it);
1580	}
1581}
1582
1583void
1584itimer_fire(struct itimer *it)
1585{
1586	struct proc *p = it->it_proc;
1587	struct thread *td;
1588
1589	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1590	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1591		if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1592			ITIMER_LOCK(it);
1593			timespecclear(&it->it_time.it_value);
1594			timespecclear(&it->it_time.it_interval);
1595			callout_stop(&it->it_callout);
1596			ITIMER_UNLOCK(it);
1597			return;
1598		}
1599		if (!KSI_ONQ(&it->it_ksi)) {
1600			it->it_ksi.ksi_errno = 0;
1601			ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1602			tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1603		} else {
1604			if (it->it_overrun < INT_MAX)
1605				it->it_overrun++;
1606			else
1607				it->it_ksi.ksi_errno = ERANGE;
1608		}
1609		PROC_UNLOCK(p);
1610	}
1611}
1612
1613static void
1614itimers_alloc(struct proc *p)
1615{
1616	struct itimers *its;
1617	int i;
1618
1619	its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1620	LIST_INIT(&its->its_virtual);
1621	LIST_INIT(&its->its_prof);
1622	TAILQ_INIT(&its->its_worklist);
1623	for (i = 0; i < TIMER_MAX; i++)
1624		its->its_timers[i] = NULL;
1625	PROC_LOCK(p);
1626	if (p->p_itimers == NULL) {
1627		p->p_itimers = its;
1628		PROC_UNLOCK(p);
1629	}
1630	else {
1631		PROC_UNLOCK(p);
1632		free(its, M_SUBPROC);
1633	}
1634}
1635
1636static void
1637itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1638{
1639	itimers_event_hook_exit(arg, p);
1640}
1641
1642/* Clean up timers when some process events are being triggered. */
1643static void
1644itimers_event_hook_exit(void *arg, struct proc *p)
1645{
1646	struct itimers *its;
1647	struct itimer *it;
1648	int event = (int)(intptr_t)arg;
1649	int i;
1650
1651	if (p->p_itimers != NULL) {
1652		its = p->p_itimers;
1653		for (i = 0; i < MAX_CLOCKS; ++i) {
1654			if (posix_clocks[i].event_hook != NULL)
1655				CLOCK_CALL(i, event_hook, (p, i, event));
1656		}
1657		/*
1658		 * According to susv3, XSI interval timers should be inherited
1659		 * by new image.
1660		 */
1661		if (event == ITIMER_EV_EXEC)
1662			i = 3;
1663		else if (event == ITIMER_EV_EXIT)
1664			i = 0;
1665		else
1666			panic("unhandled event");
1667		for (; i < TIMER_MAX; ++i) {
1668			if ((it = its->its_timers[i]) != NULL)
1669				kern_ktimer_delete(curthread, i);
1670		}
1671		if (its->its_timers[0] == NULL &&
1672		    its->its_timers[1] == NULL &&
1673		    its->its_timers[2] == NULL) {
1674			free(its, M_SUBPROC);
1675			p->p_itimers = NULL;
1676		}
1677	}
1678}
1679