xref: /freebsd/sys/kern/kern_time.c (revision 15c433351f54e7cd5bec8d36c8e89e6a7fa55b26)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)kern_time.c	8.1 (Berkeley) 6/10/93
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/limits.h>
38 #include <sys/clock.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/eventhandler.h>
43 #include <sys/resourcevar.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/sleepqueue.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/posix4.h>
53 #include <sys/time.h>
54 #include <sys/timers.h>
55 #include <sys/timetc.h>
56 #include <sys/vnode.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_extern.h>
60 
61 #define MAX_CLOCKS 	(CLOCK_MONOTONIC+1)
62 #define CPUCLOCK_BIT		0x80000000
63 #define CPUCLOCK_PROCESS_BIT	0x40000000
64 #define CPUCLOCK_ID_MASK	(~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
65 #define MAKE_THREAD_CPUCLOCK(tid)	(CPUCLOCK_BIT|(tid))
66 #define MAKE_PROCESS_CPUCLOCK(pid)	\
67 	(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
68 
69 static struct kclock	posix_clocks[MAX_CLOCKS];
70 static uma_zone_t	itimer_zone = NULL;
71 
72 /*
73  * Time of day and interval timer support.
74  *
75  * These routines provide the kernel entry points to get and set
76  * the time-of-day and per-process interval timers.  Subroutines
77  * here provide support for adding and subtracting timeval structures
78  * and decrementing interval timers, optionally reloading the interval
79  * timers when they expire.
80  */
81 
82 static int	settime(struct thread *, struct timeval *);
83 static void	timevalfix(struct timeval *);
84 
85 static void	itimer_start(void);
86 static int	itimer_init(void *, int, int);
87 static void	itimer_fini(void *, int);
88 static void	itimer_enter(struct itimer *);
89 static void	itimer_leave(struct itimer *);
90 static struct itimer *itimer_find(struct proc *, int);
91 static void	itimers_alloc(struct proc *);
92 static void	itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
93 static void	itimers_event_hook_exit(void *arg, struct proc *p);
94 static int	realtimer_create(struct itimer *);
95 static int	realtimer_gettime(struct itimer *, struct itimerspec *);
96 static int	realtimer_settime(struct itimer *, int,
97 			struct itimerspec *, struct itimerspec *);
98 static int	realtimer_delete(struct itimer *);
99 static void	realtimer_clocktime(clockid_t, struct timespec *);
100 static void	realtimer_expire(void *);
101 
102 int		register_posix_clock(int, struct kclock *);
103 void		itimer_fire(struct itimer *it);
104 int		itimespecfix(struct timespec *ts);
105 
106 #define CLOCK_CALL(clock, call, arglist)		\
107 	((*posix_clocks[clock].call) arglist)
108 
109 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
110 
111 
112 static int
113 settime(struct thread *td, struct timeval *tv)
114 {
115 	struct timeval delta, tv1, tv2;
116 	static struct timeval maxtime, laststep;
117 	struct timespec ts;
118 	int s;
119 
120 	s = splclock();
121 	microtime(&tv1);
122 	delta = *tv;
123 	timevalsub(&delta, &tv1);
124 
125 	/*
126 	 * If the system is secure, we do not allow the time to be
127 	 * set to a value earlier than 1 second less than the highest
128 	 * time we have yet seen. The worst a miscreant can do in
129 	 * this circumstance is "freeze" time. He couldn't go
130 	 * back to the past.
131 	 *
132 	 * We similarly do not allow the clock to be stepped more
133 	 * than one second, nor more than once per second. This allows
134 	 * a miscreant to make the clock march double-time, but no worse.
135 	 */
136 	if (securelevel_gt(td->td_ucred, 1) != 0) {
137 		if (delta.tv_sec < 0 || delta.tv_usec < 0) {
138 			/*
139 			 * Update maxtime to latest time we've seen.
140 			 */
141 			if (tv1.tv_sec > maxtime.tv_sec)
142 				maxtime = tv1;
143 			tv2 = *tv;
144 			timevalsub(&tv2, &maxtime);
145 			if (tv2.tv_sec < -1) {
146 				tv->tv_sec = maxtime.tv_sec - 1;
147 				printf("Time adjustment clamped to -1 second\n");
148 			}
149 		} else {
150 			if (tv1.tv_sec == laststep.tv_sec) {
151 				splx(s);
152 				return (EPERM);
153 			}
154 			if (delta.tv_sec > 1) {
155 				tv->tv_sec = tv1.tv_sec + 1;
156 				printf("Time adjustment clamped to +1 second\n");
157 			}
158 			laststep = *tv;
159 		}
160 	}
161 
162 	ts.tv_sec = tv->tv_sec;
163 	ts.tv_nsec = tv->tv_usec * 1000;
164 	mtx_lock(&Giant);
165 	tc_setclock(&ts);
166 	resettodr();
167 	mtx_unlock(&Giant);
168 	return (0);
169 }
170 
171 #ifndef _SYS_SYSPROTO_H_
172 struct clock_getcpuclockid2_args {
173 	id_t id;
174 	int which,
175 	clockid_t *clock_id;
176 };
177 #endif
178 /* ARGSUSED */
179 int
180 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
181 {
182 	clockid_t clk_id;
183 	int error;
184 
185 	error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
186 	if (error == 0)
187 		error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
188 	return (error);
189 }
190 
191 int
192 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
193     clockid_t *clk_id)
194 {
195 	struct proc *p;
196 	pid_t pid;
197 	lwpid_t tid;
198 	int error;
199 
200 	switch (which) {
201 	case CPUCLOCK_WHICH_PID:
202 		if (id != 0) {
203 			error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
204 			if (error != 0)
205 				return (error);
206 			PROC_UNLOCK(p);
207 			pid = id;
208 		} else {
209 			pid = td->td_proc->p_pid;
210 		}
211 		*clk_id = MAKE_PROCESS_CPUCLOCK(pid);
212 		return (0);
213 	case CPUCLOCK_WHICH_TID:
214 		tid = id == 0 ? td->td_tid : id;
215 		*clk_id = MAKE_THREAD_CPUCLOCK(tid);
216 		return (0);
217 	default:
218 		return (EINVAL);
219 	}
220 }
221 
222 #ifndef _SYS_SYSPROTO_H_
223 struct clock_gettime_args {
224 	clockid_t clock_id;
225 	struct	timespec *tp;
226 };
227 #endif
228 /* ARGSUSED */
229 int
230 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
231 {
232 	struct timespec ats;
233 	int error;
234 
235 	error = kern_clock_gettime(td, uap->clock_id, &ats);
236 	if (error == 0)
237 		error = copyout(&ats, uap->tp, sizeof(ats));
238 
239 	return (error);
240 }
241 
242 static inline void
243 cputick2timespec(uint64_t runtime, struct timespec *ats)
244 {
245 	runtime = cputick2usec(runtime);
246 	ats->tv_sec = runtime / 1000000;
247 	ats->tv_nsec = runtime % 1000000 * 1000;
248 }
249 
250 static void
251 get_thread_cputime(struct thread *targettd, struct timespec *ats)
252 {
253 	uint64_t runtime, curtime, switchtime;
254 
255 	if (targettd == NULL) { /* current thread */
256 		critical_enter();
257 		switchtime = PCPU_GET(switchtime);
258 		curtime = cpu_ticks();
259 		runtime = curthread->td_runtime;
260 		critical_exit();
261 		runtime += curtime - switchtime;
262 	} else {
263 		thread_lock(targettd);
264 		runtime = targettd->td_runtime;
265 		thread_unlock(targettd);
266 	}
267 	cputick2timespec(runtime, ats);
268 }
269 
270 static void
271 get_process_cputime(struct proc *targetp, struct timespec *ats)
272 {
273 	uint64_t runtime;
274 	struct rusage ru;
275 
276 	PROC_STATLOCK(targetp);
277 	rufetch(targetp, &ru);
278 	runtime = targetp->p_rux.rux_runtime;
279 	PROC_STATUNLOCK(targetp);
280 	cputick2timespec(runtime, ats);
281 }
282 
283 static int
284 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
285 {
286 	struct proc *p, *p2;
287 	struct thread *td2;
288 	lwpid_t tid;
289 	pid_t pid;
290 	int error;
291 
292 	p = td->td_proc;
293 	if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
294 		tid = clock_id & CPUCLOCK_ID_MASK;
295 		td2 = tdfind(tid, p->p_pid);
296 		if (td2 == NULL)
297 			return (EINVAL);
298 		get_thread_cputime(td2, ats);
299 		PROC_UNLOCK(td2->td_proc);
300 	} else {
301 		pid = clock_id & CPUCLOCK_ID_MASK;
302 		error = pget(pid, PGET_CANSEE, &p2);
303 		if (error != 0)
304 			return (EINVAL);
305 		get_process_cputime(p2, ats);
306 		PROC_UNLOCK(p2);
307 	}
308 	return (0);
309 }
310 
311 int
312 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
313 {
314 	struct timeval sys, user;
315 	struct proc *p;
316 
317 	p = td->td_proc;
318 	switch (clock_id) {
319 	case CLOCK_REALTIME:		/* Default to precise. */
320 	case CLOCK_REALTIME_PRECISE:
321 		nanotime(ats);
322 		break;
323 	case CLOCK_REALTIME_FAST:
324 		getnanotime(ats);
325 		break;
326 	case CLOCK_VIRTUAL:
327 		PROC_LOCK(p);
328 		PROC_STATLOCK(p);
329 		calcru(p, &user, &sys);
330 		PROC_STATUNLOCK(p);
331 		PROC_UNLOCK(p);
332 		TIMEVAL_TO_TIMESPEC(&user, ats);
333 		break;
334 	case CLOCK_PROF:
335 		PROC_LOCK(p);
336 		PROC_STATLOCK(p);
337 		calcru(p, &user, &sys);
338 		PROC_STATUNLOCK(p);
339 		PROC_UNLOCK(p);
340 		timevaladd(&user, &sys);
341 		TIMEVAL_TO_TIMESPEC(&user, ats);
342 		break;
343 	case CLOCK_MONOTONIC:		/* Default to precise. */
344 	case CLOCK_MONOTONIC_PRECISE:
345 	case CLOCK_UPTIME:
346 	case CLOCK_UPTIME_PRECISE:
347 		nanouptime(ats);
348 		break;
349 	case CLOCK_UPTIME_FAST:
350 	case CLOCK_MONOTONIC_FAST:
351 		getnanouptime(ats);
352 		break;
353 	case CLOCK_SECOND:
354 		ats->tv_sec = time_second;
355 		ats->tv_nsec = 0;
356 		break;
357 	case CLOCK_THREAD_CPUTIME_ID:
358 		get_thread_cputime(NULL, ats);
359 		break;
360 	case CLOCK_PROCESS_CPUTIME_ID:
361 		PROC_LOCK(p);
362 		get_process_cputime(p, ats);
363 		PROC_UNLOCK(p);
364 		break;
365 	default:
366 		if ((int)clock_id >= 0)
367 			return (EINVAL);
368 		return (get_cputime(td, clock_id, ats));
369 	}
370 	return (0);
371 }
372 
373 #ifndef _SYS_SYSPROTO_H_
374 struct clock_settime_args {
375 	clockid_t clock_id;
376 	const struct	timespec *tp;
377 };
378 #endif
379 /* ARGSUSED */
380 int
381 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
382 {
383 	struct timespec ats;
384 	int error;
385 
386 	if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
387 		return (error);
388 	return (kern_clock_settime(td, uap->clock_id, &ats));
389 }
390 
391 int
392 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
393 {
394 	struct timeval atv;
395 	int error;
396 
397 	if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
398 		return (error);
399 	if (clock_id != CLOCK_REALTIME)
400 		return (EINVAL);
401 	if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 ||
402 	    ats->tv_sec < 0)
403 		return (EINVAL);
404 	/* XXX Don't convert nsec->usec and back */
405 	TIMESPEC_TO_TIMEVAL(&atv, ats);
406 	error = settime(td, &atv);
407 	return (error);
408 }
409 
410 #ifndef _SYS_SYSPROTO_H_
411 struct clock_getres_args {
412 	clockid_t clock_id;
413 	struct	timespec *tp;
414 };
415 #endif
416 int
417 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
418 {
419 	struct timespec ts;
420 	int error;
421 
422 	if (uap->tp == NULL)
423 		return (0);
424 
425 	error = kern_clock_getres(td, uap->clock_id, &ts);
426 	if (error == 0)
427 		error = copyout(&ts, uap->tp, sizeof(ts));
428 	return (error);
429 }
430 
431 int
432 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
433 {
434 
435 	ts->tv_sec = 0;
436 	switch (clock_id) {
437 	case CLOCK_REALTIME:
438 	case CLOCK_REALTIME_FAST:
439 	case CLOCK_REALTIME_PRECISE:
440 	case CLOCK_MONOTONIC:
441 	case CLOCK_MONOTONIC_FAST:
442 	case CLOCK_MONOTONIC_PRECISE:
443 	case CLOCK_UPTIME:
444 	case CLOCK_UPTIME_FAST:
445 	case CLOCK_UPTIME_PRECISE:
446 		/*
447 		 * Round up the result of the division cheaply by adding 1.
448 		 * Rounding up is especially important if rounding down
449 		 * would give 0.  Perfect rounding is unimportant.
450 		 */
451 		ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
452 		break;
453 	case CLOCK_VIRTUAL:
454 	case CLOCK_PROF:
455 		/* Accurately round up here because we can do so cheaply. */
456 		ts->tv_nsec = (1000000000 + hz - 1) / hz;
457 		break;
458 	case CLOCK_SECOND:
459 		ts->tv_sec = 1;
460 		ts->tv_nsec = 0;
461 		break;
462 	case CLOCK_THREAD_CPUTIME_ID:
463 	case CLOCK_PROCESS_CPUTIME_ID:
464 	cputime:
465 		/* sync with cputick2usec */
466 		ts->tv_nsec = 1000000 / cpu_tickrate();
467 		if (ts->tv_nsec == 0)
468 			ts->tv_nsec = 1000;
469 		break;
470 	default:
471 		if ((int)clock_id < 0)
472 			goto cputime;
473 		return (EINVAL);
474 	}
475 	return (0);
476 }
477 
478 static uint8_t nanowait[MAXCPU];
479 
480 int
481 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
482 {
483 	struct timespec ts;
484 	sbintime_t sbt, sbtt, prec, tmp;
485 	time_t over;
486 	int error;
487 
488 	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
489 		return (EINVAL);
490 	if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
491 		return (0);
492 	ts = *rqt;
493 	if (ts.tv_sec > INT32_MAX / 2) {
494 		over = ts.tv_sec - INT32_MAX / 2;
495 		ts.tv_sec -= over;
496 	} else
497 		over = 0;
498 	tmp = tstosbt(ts);
499 	prec = tmp;
500 	prec >>= tc_precexp;
501 	if (TIMESEL(&sbt, tmp))
502 		sbt += tc_tick_sbt;
503 	sbt += tmp;
504 	error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
505 	    sbt, prec, C_ABSOLUTE);
506 	if (error != EWOULDBLOCK) {
507 		if (error == ERESTART)
508 			error = EINTR;
509 		TIMESEL(&sbtt, tmp);
510 		if (rmt != NULL) {
511 			ts = sbttots(sbt - sbtt);
512 			ts.tv_sec += over;
513 			if (ts.tv_sec < 0)
514 				timespecclear(&ts);
515 			*rmt = ts;
516 		}
517 		if (sbtt >= sbt)
518 			return (0);
519 		return (error);
520 	}
521 	return (0);
522 }
523 
524 #ifndef _SYS_SYSPROTO_H_
525 struct nanosleep_args {
526 	struct	timespec *rqtp;
527 	struct	timespec *rmtp;
528 };
529 #endif
530 /* ARGSUSED */
531 int
532 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
533 {
534 	struct timespec rmt, rqt;
535 	int error;
536 
537 	error = copyin(uap->rqtp, &rqt, sizeof(rqt));
538 	if (error)
539 		return (error);
540 
541 	if (uap->rmtp &&
542 	    !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE))
543 			return (EFAULT);
544 	error = kern_nanosleep(td, &rqt, &rmt);
545 	if (error && uap->rmtp) {
546 		int error2;
547 
548 		error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
549 		if (error2)
550 			error = error2;
551 	}
552 	return (error);
553 }
554 
555 #ifndef _SYS_SYSPROTO_H_
556 struct gettimeofday_args {
557 	struct	timeval *tp;
558 	struct	timezone *tzp;
559 };
560 #endif
561 /* ARGSUSED */
562 int
563 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
564 {
565 	struct timeval atv;
566 	struct timezone rtz;
567 	int error = 0;
568 
569 	if (uap->tp) {
570 		microtime(&atv);
571 		error = copyout(&atv, uap->tp, sizeof (atv));
572 	}
573 	if (error == 0 && uap->tzp != NULL) {
574 		rtz.tz_minuteswest = tz_minuteswest;
575 		rtz.tz_dsttime = tz_dsttime;
576 		error = copyout(&rtz, uap->tzp, sizeof (rtz));
577 	}
578 	return (error);
579 }
580 
581 #ifndef _SYS_SYSPROTO_H_
582 struct settimeofday_args {
583 	struct	timeval *tv;
584 	struct	timezone *tzp;
585 };
586 #endif
587 /* ARGSUSED */
588 int
589 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
590 {
591 	struct timeval atv, *tvp;
592 	struct timezone atz, *tzp;
593 	int error;
594 
595 	if (uap->tv) {
596 		error = copyin(uap->tv, &atv, sizeof(atv));
597 		if (error)
598 			return (error);
599 		tvp = &atv;
600 	} else
601 		tvp = NULL;
602 	if (uap->tzp) {
603 		error = copyin(uap->tzp, &atz, sizeof(atz));
604 		if (error)
605 			return (error);
606 		tzp = &atz;
607 	} else
608 		tzp = NULL;
609 	return (kern_settimeofday(td, tvp, tzp));
610 }
611 
612 int
613 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
614 {
615 	int error;
616 
617 	error = priv_check(td, PRIV_SETTIMEOFDAY);
618 	if (error)
619 		return (error);
620 	/* Verify all parameters before changing time. */
621 	if (tv) {
622 		if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
623 		    tv->tv_sec < 0)
624 			return (EINVAL);
625 		error = settime(td, tv);
626 	}
627 	if (tzp && error == 0) {
628 		tz_minuteswest = tzp->tz_minuteswest;
629 		tz_dsttime = tzp->tz_dsttime;
630 	}
631 	return (error);
632 }
633 
634 /*
635  * Get value of an interval timer.  The process virtual and profiling virtual
636  * time timers are kept in the p_stats area, since they can be swapped out.
637  * These are kept internally in the way they are specified externally: in
638  * time until they expire.
639  *
640  * The real time interval timer is kept in the process table slot for the
641  * process, and its value (it_value) is kept as an absolute time rather than
642  * as a delta, so that it is easy to keep periodic real-time signals from
643  * drifting.
644  *
645  * Virtual time timers are processed in the hardclock() routine of
646  * kern_clock.c.  The real time timer is processed by a timeout routine,
647  * called from the softclock() routine.  Since a callout may be delayed in
648  * real time due to interrupt processing in the system, it is possible for
649  * the real time timeout routine (realitexpire, given below), to be delayed
650  * in real time past when it is supposed to occur.  It does not suffice,
651  * therefore, to reload the real timer .it_value from the real time timers
652  * .it_interval.  Rather, we compute the next time in absolute time the timer
653  * should go off.
654  */
655 #ifndef _SYS_SYSPROTO_H_
656 struct getitimer_args {
657 	u_int	which;
658 	struct	itimerval *itv;
659 };
660 #endif
661 int
662 sys_getitimer(struct thread *td, struct getitimer_args *uap)
663 {
664 	struct itimerval aitv;
665 	int error;
666 
667 	error = kern_getitimer(td, uap->which, &aitv);
668 	if (error != 0)
669 		return (error);
670 	return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
671 }
672 
673 int
674 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
675 {
676 	struct proc *p = td->td_proc;
677 	struct timeval ctv;
678 
679 	if (which > ITIMER_PROF)
680 		return (EINVAL);
681 
682 	if (which == ITIMER_REAL) {
683 		/*
684 		 * Convert from absolute to relative time in .it_value
685 		 * part of real time timer.  If time for real time timer
686 		 * has passed return 0, else return difference between
687 		 * current time and time for the timer to go off.
688 		 */
689 		PROC_LOCK(p);
690 		*aitv = p->p_realtimer;
691 		PROC_UNLOCK(p);
692 		if (timevalisset(&aitv->it_value)) {
693 			microuptime(&ctv);
694 			if (timevalcmp(&aitv->it_value, &ctv, <))
695 				timevalclear(&aitv->it_value);
696 			else
697 				timevalsub(&aitv->it_value, &ctv);
698 		}
699 	} else {
700 		PROC_ITIMLOCK(p);
701 		*aitv = p->p_stats->p_timer[which];
702 		PROC_ITIMUNLOCK(p);
703 	}
704 	return (0);
705 }
706 
707 #ifndef _SYS_SYSPROTO_H_
708 struct setitimer_args {
709 	u_int	which;
710 	struct	itimerval *itv, *oitv;
711 };
712 #endif
713 int
714 sys_setitimer(struct thread *td, struct setitimer_args *uap)
715 {
716 	struct itimerval aitv, oitv;
717 	int error;
718 
719 	if (uap->itv == NULL) {
720 		uap->itv = uap->oitv;
721 		return (sys_getitimer(td, (struct getitimer_args *)uap));
722 	}
723 
724 	if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
725 		return (error);
726 	error = kern_setitimer(td, uap->which, &aitv, &oitv);
727 	if (error != 0 || uap->oitv == NULL)
728 		return (error);
729 	return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
730 }
731 
732 int
733 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
734     struct itimerval *oitv)
735 {
736 	struct proc *p = td->td_proc;
737 	struct timeval ctv;
738 	sbintime_t sbt, pr;
739 
740 	if (aitv == NULL)
741 		return (kern_getitimer(td, which, oitv));
742 
743 	if (which > ITIMER_PROF)
744 		return (EINVAL);
745 	if (itimerfix(&aitv->it_value) ||
746 	    aitv->it_value.tv_sec > INT32_MAX / 2)
747 		return (EINVAL);
748 	if (!timevalisset(&aitv->it_value))
749 		timevalclear(&aitv->it_interval);
750 	else if (itimerfix(&aitv->it_interval) ||
751 	    aitv->it_interval.tv_sec > INT32_MAX / 2)
752 		return (EINVAL);
753 
754 	if (which == ITIMER_REAL) {
755 		PROC_LOCK(p);
756 		if (timevalisset(&p->p_realtimer.it_value))
757 			callout_stop(&p->p_itcallout);
758 		microuptime(&ctv);
759 		if (timevalisset(&aitv->it_value)) {
760 			pr = tvtosbt(aitv->it_value) >> tc_precexp;
761 			timevaladd(&aitv->it_value, &ctv);
762 			sbt = tvtosbt(aitv->it_value);
763 			callout_reset_sbt(&p->p_itcallout, sbt, pr,
764 			    realitexpire, p, C_ABSOLUTE);
765 		}
766 		*oitv = p->p_realtimer;
767 		p->p_realtimer = *aitv;
768 		PROC_UNLOCK(p);
769 		if (timevalisset(&oitv->it_value)) {
770 			if (timevalcmp(&oitv->it_value, &ctv, <))
771 				timevalclear(&oitv->it_value);
772 			else
773 				timevalsub(&oitv->it_value, &ctv);
774 		}
775 	} else {
776 		if (aitv->it_interval.tv_sec == 0 &&
777 		    aitv->it_interval.tv_usec != 0 &&
778 		    aitv->it_interval.tv_usec < tick)
779 			aitv->it_interval.tv_usec = tick;
780 		if (aitv->it_value.tv_sec == 0 &&
781 		    aitv->it_value.tv_usec != 0 &&
782 		    aitv->it_value.tv_usec < tick)
783 			aitv->it_value.tv_usec = tick;
784 		PROC_ITIMLOCK(p);
785 		*oitv = p->p_stats->p_timer[which];
786 		p->p_stats->p_timer[which] = *aitv;
787 		PROC_ITIMUNLOCK(p);
788 	}
789 	return (0);
790 }
791 
792 /*
793  * Real interval timer expired:
794  * send process whose timer expired an alarm signal.
795  * If time is not set up to reload, then just return.
796  * Else compute next time timer should go off which is > current time.
797  * This is where delay in processing this timeout causes multiple
798  * SIGALRM calls to be compressed into one.
799  * tvtohz() always adds 1 to allow for the time until the next clock
800  * interrupt being strictly less than 1 clock tick, but we don't want
801  * that here since we want to appear to be in sync with the clock
802  * interrupt even when we're delayed.
803  */
804 void
805 realitexpire(void *arg)
806 {
807 	struct proc *p;
808 	struct timeval ctv;
809 	sbintime_t isbt;
810 
811 	p = (struct proc *)arg;
812 	kern_psignal(p, SIGALRM);
813 	if (!timevalisset(&p->p_realtimer.it_interval)) {
814 		timevalclear(&p->p_realtimer.it_value);
815 		if (p->p_flag & P_WEXIT)
816 			wakeup(&p->p_itcallout);
817 		return;
818 	}
819 	isbt = tvtosbt(p->p_realtimer.it_interval);
820 	if (isbt >= sbt_timethreshold)
821 		getmicrouptime(&ctv);
822 	else
823 		microuptime(&ctv);
824 	do {
825 		timevaladd(&p->p_realtimer.it_value,
826 		    &p->p_realtimer.it_interval);
827 	} while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
828 	callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
829 	    isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE);
830 }
831 
832 /*
833  * Check that a proposed value to load into the .it_value or
834  * .it_interval part of an interval timer is acceptable, and
835  * fix it to have at least minimal value (i.e. if it is less
836  * than the resolution of the clock, round it up.)
837  */
838 int
839 itimerfix(struct timeval *tv)
840 {
841 
842 	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
843 		return (EINVAL);
844 	if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
845 	    tv->tv_usec < (u_int)tick / 16)
846 		tv->tv_usec = (u_int)tick / 16;
847 	return (0);
848 }
849 
850 /*
851  * Decrement an interval timer by a specified number
852  * of microseconds, which must be less than a second,
853  * i.e. < 1000000.  If the timer expires, then reload
854  * it.  In this case, carry over (usec - old value) to
855  * reduce the value reloaded into the timer so that
856  * the timer does not drift.  This routine assumes
857  * that it is called in a context where the timers
858  * on which it is operating cannot change in value.
859  */
860 int
861 itimerdecr(struct itimerval *itp, int usec)
862 {
863 
864 	if (itp->it_value.tv_usec < usec) {
865 		if (itp->it_value.tv_sec == 0) {
866 			/* expired, and already in next interval */
867 			usec -= itp->it_value.tv_usec;
868 			goto expire;
869 		}
870 		itp->it_value.tv_usec += 1000000;
871 		itp->it_value.tv_sec--;
872 	}
873 	itp->it_value.tv_usec -= usec;
874 	usec = 0;
875 	if (timevalisset(&itp->it_value))
876 		return (1);
877 	/* expired, exactly at end of interval */
878 expire:
879 	if (timevalisset(&itp->it_interval)) {
880 		itp->it_value = itp->it_interval;
881 		itp->it_value.tv_usec -= usec;
882 		if (itp->it_value.tv_usec < 0) {
883 			itp->it_value.tv_usec += 1000000;
884 			itp->it_value.tv_sec--;
885 		}
886 	} else
887 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
888 	return (0);
889 }
890 
891 /*
892  * Add and subtract routines for timevals.
893  * N.B.: subtract routine doesn't deal with
894  * results which are before the beginning,
895  * it just gets very confused in this case.
896  * Caveat emptor.
897  */
898 void
899 timevaladd(struct timeval *t1, const struct timeval *t2)
900 {
901 
902 	t1->tv_sec += t2->tv_sec;
903 	t1->tv_usec += t2->tv_usec;
904 	timevalfix(t1);
905 }
906 
907 void
908 timevalsub(struct timeval *t1, const struct timeval *t2)
909 {
910 
911 	t1->tv_sec -= t2->tv_sec;
912 	t1->tv_usec -= t2->tv_usec;
913 	timevalfix(t1);
914 }
915 
916 static void
917 timevalfix(struct timeval *t1)
918 {
919 
920 	if (t1->tv_usec < 0) {
921 		t1->tv_sec--;
922 		t1->tv_usec += 1000000;
923 	}
924 	if (t1->tv_usec >= 1000000) {
925 		t1->tv_sec++;
926 		t1->tv_usec -= 1000000;
927 	}
928 }
929 
930 /*
931  * ratecheck(): simple time-based rate-limit checking.
932  */
933 int
934 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
935 {
936 	struct timeval tv, delta;
937 	int rv = 0;
938 
939 	getmicrouptime(&tv);		/* NB: 10ms precision */
940 	delta = tv;
941 	timevalsub(&delta, lasttime);
942 
943 	/*
944 	 * check for 0,0 is so that the message will be seen at least once,
945 	 * even if interval is huge.
946 	 */
947 	if (timevalcmp(&delta, mininterval, >=) ||
948 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
949 		*lasttime = tv;
950 		rv = 1;
951 	}
952 
953 	return (rv);
954 }
955 
956 /*
957  * ppsratecheck(): packets (or events) per second limitation.
958  *
959  * Return 0 if the limit is to be enforced (e.g. the caller
960  * should drop a packet because of the rate limitation).
961  *
962  * maxpps of 0 always causes zero to be returned.  maxpps of -1
963  * always causes 1 to be returned; this effectively defeats rate
964  * limiting.
965  *
966  * Note that we maintain the struct timeval for compatibility
967  * with other bsd systems.  We reuse the storage and just monitor
968  * clock ticks for minimal overhead.
969  */
970 int
971 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
972 {
973 	int now;
974 
975 	/*
976 	 * Reset the last time and counter if this is the first call
977 	 * or more than a second has passed since the last update of
978 	 * lasttime.
979 	 */
980 	now = ticks;
981 	if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
982 		lasttime->tv_sec = now;
983 		*curpps = 1;
984 		return (maxpps != 0);
985 	} else {
986 		(*curpps)++;		/* NB: ignore potential overflow */
987 		return (maxpps < 0 || *curpps <= maxpps);
988 	}
989 }
990 
991 static void
992 itimer_start(void)
993 {
994 	struct kclock rt_clock = {
995 		.timer_create  = realtimer_create,
996 		.timer_delete  = realtimer_delete,
997 		.timer_settime = realtimer_settime,
998 		.timer_gettime = realtimer_gettime,
999 		.event_hook    = NULL
1000 	};
1001 
1002 	itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1003 		NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1004 	register_posix_clock(CLOCK_REALTIME,  &rt_clock);
1005 	register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1006 	p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1007 	p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1008 	p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1009 	EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
1010 		(void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
1011 	EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
1012 		(void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
1013 }
1014 
1015 int
1016 register_posix_clock(int clockid, struct kclock *clk)
1017 {
1018 	if ((unsigned)clockid >= MAX_CLOCKS) {
1019 		printf("%s: invalid clockid\n", __func__);
1020 		return (0);
1021 	}
1022 	posix_clocks[clockid] = *clk;
1023 	return (1);
1024 }
1025 
1026 static int
1027 itimer_init(void *mem, int size, int flags)
1028 {
1029 	struct itimer *it;
1030 
1031 	it = (struct itimer *)mem;
1032 	mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1033 	return (0);
1034 }
1035 
1036 static void
1037 itimer_fini(void *mem, int size)
1038 {
1039 	struct itimer *it;
1040 
1041 	it = (struct itimer *)mem;
1042 	mtx_destroy(&it->it_mtx);
1043 }
1044 
1045 static void
1046 itimer_enter(struct itimer *it)
1047 {
1048 
1049 	mtx_assert(&it->it_mtx, MA_OWNED);
1050 	it->it_usecount++;
1051 }
1052 
1053 static void
1054 itimer_leave(struct itimer *it)
1055 {
1056 
1057 	mtx_assert(&it->it_mtx, MA_OWNED);
1058 	KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1059 
1060 	if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1061 		wakeup(it);
1062 }
1063 
1064 #ifndef _SYS_SYSPROTO_H_
1065 struct ktimer_create_args {
1066 	clockid_t clock_id;
1067 	struct sigevent * evp;
1068 	int * timerid;
1069 };
1070 #endif
1071 int
1072 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1073 {
1074 	struct sigevent *evp, ev;
1075 	int id;
1076 	int error;
1077 
1078 	if (uap->evp == NULL) {
1079 		evp = NULL;
1080 	} else {
1081 		error = copyin(uap->evp, &ev, sizeof(ev));
1082 		if (error != 0)
1083 			return (error);
1084 		evp = &ev;
1085 	}
1086 	error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1087 	if (error == 0) {
1088 		error = copyout(&id, uap->timerid, sizeof(int));
1089 		if (error != 0)
1090 			kern_ktimer_delete(td, id);
1091 	}
1092 	return (error);
1093 }
1094 
1095 int
1096 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1097     int *timerid, int preset_id)
1098 {
1099 	struct proc *p = td->td_proc;
1100 	struct itimer *it;
1101 	int id;
1102 	int error;
1103 
1104 	if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1105 		return (EINVAL);
1106 
1107 	if (posix_clocks[clock_id].timer_create == NULL)
1108 		return (EINVAL);
1109 
1110 	if (evp != NULL) {
1111 		if (evp->sigev_notify != SIGEV_NONE &&
1112 		    evp->sigev_notify != SIGEV_SIGNAL &&
1113 		    evp->sigev_notify != SIGEV_THREAD_ID)
1114 			return (EINVAL);
1115 		if ((evp->sigev_notify == SIGEV_SIGNAL ||
1116 		     evp->sigev_notify == SIGEV_THREAD_ID) &&
1117 			!_SIG_VALID(evp->sigev_signo))
1118 			return (EINVAL);
1119 	}
1120 
1121 	if (p->p_itimers == NULL)
1122 		itimers_alloc(p);
1123 
1124 	it = uma_zalloc(itimer_zone, M_WAITOK);
1125 	it->it_flags = 0;
1126 	it->it_usecount = 0;
1127 	it->it_active = 0;
1128 	timespecclear(&it->it_time.it_value);
1129 	timespecclear(&it->it_time.it_interval);
1130 	it->it_overrun = 0;
1131 	it->it_overrun_last = 0;
1132 	it->it_clockid = clock_id;
1133 	it->it_timerid = -1;
1134 	it->it_proc = p;
1135 	ksiginfo_init(&it->it_ksi);
1136 	it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1137 	error = CLOCK_CALL(clock_id, timer_create, (it));
1138 	if (error != 0)
1139 		goto out;
1140 
1141 	PROC_LOCK(p);
1142 	if (preset_id != -1) {
1143 		KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1144 		id = preset_id;
1145 		if (p->p_itimers->its_timers[id] != NULL) {
1146 			PROC_UNLOCK(p);
1147 			error = 0;
1148 			goto out;
1149 		}
1150 	} else {
1151 		/*
1152 		 * Find a free timer slot, skipping those reserved
1153 		 * for setitimer().
1154 		 */
1155 		for (id = 3; id < TIMER_MAX; id++)
1156 			if (p->p_itimers->its_timers[id] == NULL)
1157 				break;
1158 		if (id == TIMER_MAX) {
1159 			PROC_UNLOCK(p);
1160 			error = EAGAIN;
1161 			goto out;
1162 		}
1163 	}
1164 	it->it_timerid = id;
1165 	p->p_itimers->its_timers[id] = it;
1166 	if (evp != NULL)
1167 		it->it_sigev = *evp;
1168 	else {
1169 		it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1170 		switch (clock_id) {
1171 		default:
1172 		case CLOCK_REALTIME:
1173 			it->it_sigev.sigev_signo = SIGALRM;
1174 			break;
1175 		case CLOCK_VIRTUAL:
1176  			it->it_sigev.sigev_signo = SIGVTALRM;
1177 			break;
1178 		case CLOCK_PROF:
1179 			it->it_sigev.sigev_signo = SIGPROF;
1180 			break;
1181 		}
1182 		it->it_sigev.sigev_value.sival_int = id;
1183 	}
1184 
1185 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1186 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1187 		it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1188 		it->it_ksi.ksi_code = SI_TIMER;
1189 		it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1190 		it->it_ksi.ksi_timerid = id;
1191 	}
1192 	PROC_UNLOCK(p);
1193 	*timerid = id;
1194 	return (0);
1195 
1196 out:
1197 	ITIMER_LOCK(it);
1198 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1199 	ITIMER_UNLOCK(it);
1200 	uma_zfree(itimer_zone, it);
1201 	return (error);
1202 }
1203 
1204 #ifndef _SYS_SYSPROTO_H_
1205 struct ktimer_delete_args {
1206 	int timerid;
1207 };
1208 #endif
1209 int
1210 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1211 {
1212 
1213 	return (kern_ktimer_delete(td, uap->timerid));
1214 }
1215 
1216 static struct itimer *
1217 itimer_find(struct proc *p, int timerid)
1218 {
1219 	struct itimer *it;
1220 
1221 	PROC_LOCK_ASSERT(p, MA_OWNED);
1222 	if ((p->p_itimers == NULL) ||
1223 	    (timerid < 0) || (timerid >= TIMER_MAX) ||
1224 	    (it = p->p_itimers->its_timers[timerid]) == NULL) {
1225 		return (NULL);
1226 	}
1227 	ITIMER_LOCK(it);
1228 	if ((it->it_flags & ITF_DELETING) != 0) {
1229 		ITIMER_UNLOCK(it);
1230 		it = NULL;
1231 	}
1232 	return (it);
1233 }
1234 
1235 int
1236 kern_ktimer_delete(struct thread *td, int timerid)
1237 {
1238 	struct proc *p = td->td_proc;
1239 	struct itimer *it;
1240 
1241 	PROC_LOCK(p);
1242 	it = itimer_find(p, timerid);
1243 	if (it == NULL) {
1244 		PROC_UNLOCK(p);
1245 		return (EINVAL);
1246 	}
1247 	PROC_UNLOCK(p);
1248 
1249 	it->it_flags |= ITF_DELETING;
1250 	while (it->it_usecount > 0) {
1251 		it->it_flags |= ITF_WANTED;
1252 		msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1253 	}
1254 	it->it_flags &= ~ITF_WANTED;
1255 	CLOCK_CALL(it->it_clockid, timer_delete, (it));
1256 	ITIMER_UNLOCK(it);
1257 
1258 	PROC_LOCK(p);
1259 	if (KSI_ONQ(&it->it_ksi))
1260 		sigqueue_take(&it->it_ksi);
1261 	p->p_itimers->its_timers[timerid] = NULL;
1262 	PROC_UNLOCK(p);
1263 	uma_zfree(itimer_zone, it);
1264 	return (0);
1265 }
1266 
1267 #ifndef _SYS_SYSPROTO_H_
1268 struct ktimer_settime_args {
1269 	int timerid;
1270 	int flags;
1271 	const struct itimerspec * value;
1272 	struct itimerspec * ovalue;
1273 };
1274 #endif
1275 int
1276 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1277 {
1278 	struct itimerspec val, oval, *ovalp;
1279 	int error;
1280 
1281 	error = copyin(uap->value, &val, sizeof(val));
1282 	if (error != 0)
1283 		return (error);
1284 	ovalp = uap->ovalue != NULL ? &oval : NULL;
1285 	error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1286 	if (error == 0 && uap->ovalue != NULL)
1287 		error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1288 	return (error);
1289 }
1290 
1291 int
1292 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1293     struct itimerspec *val, struct itimerspec *oval)
1294 {
1295 	struct proc *p;
1296 	struct itimer *it;
1297 	int error;
1298 
1299 	p = td->td_proc;
1300 	PROC_LOCK(p);
1301 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1302 		PROC_UNLOCK(p);
1303 		error = EINVAL;
1304 	} else {
1305 		PROC_UNLOCK(p);
1306 		itimer_enter(it);
1307 		error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1308 		    flags, val, oval));
1309 		itimer_leave(it);
1310 		ITIMER_UNLOCK(it);
1311 	}
1312 	return (error);
1313 }
1314 
1315 #ifndef _SYS_SYSPROTO_H_
1316 struct ktimer_gettime_args {
1317 	int timerid;
1318 	struct itimerspec * value;
1319 };
1320 #endif
1321 int
1322 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1323 {
1324 	struct itimerspec val;
1325 	int error;
1326 
1327 	error = kern_ktimer_gettime(td, uap->timerid, &val);
1328 	if (error == 0)
1329 		error = copyout(&val, uap->value, sizeof(val));
1330 	return (error);
1331 }
1332 
1333 int
1334 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1335 {
1336 	struct proc *p;
1337 	struct itimer *it;
1338 	int error;
1339 
1340 	p = td->td_proc;
1341 	PROC_LOCK(p);
1342 	if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1343 		PROC_UNLOCK(p);
1344 		error = EINVAL;
1345 	} else {
1346 		PROC_UNLOCK(p);
1347 		itimer_enter(it);
1348 		error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1349 		itimer_leave(it);
1350 		ITIMER_UNLOCK(it);
1351 	}
1352 	return (error);
1353 }
1354 
1355 #ifndef _SYS_SYSPROTO_H_
1356 struct timer_getoverrun_args {
1357 	int timerid;
1358 };
1359 #endif
1360 int
1361 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1362 {
1363 
1364 	return (kern_ktimer_getoverrun(td, uap->timerid));
1365 }
1366 
1367 int
1368 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1369 {
1370 	struct proc *p = td->td_proc;
1371 	struct itimer *it;
1372 	int error ;
1373 
1374 	PROC_LOCK(p);
1375 	if (timer_id < 3 ||
1376 	    (it = itimer_find(p, timer_id)) == NULL) {
1377 		PROC_UNLOCK(p);
1378 		error = EINVAL;
1379 	} else {
1380 		td->td_retval[0] = it->it_overrun_last;
1381 		ITIMER_UNLOCK(it);
1382 		PROC_UNLOCK(p);
1383 		error = 0;
1384 	}
1385 	return (error);
1386 }
1387 
1388 static int
1389 realtimer_create(struct itimer *it)
1390 {
1391 	callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1392 	return (0);
1393 }
1394 
1395 static int
1396 realtimer_delete(struct itimer *it)
1397 {
1398 	mtx_assert(&it->it_mtx, MA_OWNED);
1399 
1400 	/*
1401 	 * clear timer's value and interval to tell realtimer_expire
1402 	 * to not rearm the timer.
1403 	 */
1404 	timespecclear(&it->it_time.it_value);
1405 	timespecclear(&it->it_time.it_interval);
1406 	ITIMER_UNLOCK(it);
1407 	callout_drain(&it->it_callout);
1408 	ITIMER_LOCK(it);
1409 	return (0);
1410 }
1411 
1412 static int
1413 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1414 {
1415 	struct timespec cts;
1416 
1417 	mtx_assert(&it->it_mtx, MA_OWNED);
1418 
1419 	realtimer_clocktime(it->it_clockid, &cts);
1420 	*ovalue = it->it_time;
1421 	if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1422 		timespecsub(&ovalue->it_value, &cts);
1423 		if (ovalue->it_value.tv_sec < 0 ||
1424 		    (ovalue->it_value.tv_sec == 0 &&
1425 		     ovalue->it_value.tv_nsec == 0)) {
1426 			ovalue->it_value.tv_sec  = 0;
1427 			ovalue->it_value.tv_nsec = 1;
1428 		}
1429 	}
1430 	return (0);
1431 }
1432 
1433 static int
1434 realtimer_settime(struct itimer *it, int flags,
1435 	struct itimerspec *value, struct itimerspec *ovalue)
1436 {
1437 	struct timespec cts, ts;
1438 	struct timeval tv;
1439 	struct itimerspec val;
1440 
1441 	mtx_assert(&it->it_mtx, MA_OWNED);
1442 
1443 	val = *value;
1444 	if (itimespecfix(&val.it_value))
1445 		return (EINVAL);
1446 
1447 	if (timespecisset(&val.it_value)) {
1448 		if (itimespecfix(&val.it_interval))
1449 			return (EINVAL);
1450 	} else {
1451 		timespecclear(&val.it_interval);
1452 	}
1453 
1454 	if (ovalue != NULL)
1455 		realtimer_gettime(it, ovalue);
1456 
1457 	it->it_time = val;
1458 	if (timespecisset(&val.it_value)) {
1459 		realtimer_clocktime(it->it_clockid, &cts);
1460 		ts = val.it_value;
1461 		if ((flags & TIMER_ABSTIME) == 0) {
1462 			/* Convert to absolute time. */
1463 			timespecadd(&it->it_time.it_value, &cts);
1464 		} else {
1465 			timespecsub(&ts, &cts);
1466 			/*
1467 			 * We don't care if ts is negative, tztohz will
1468 			 * fix it.
1469 			 */
1470 		}
1471 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1472 		callout_reset(&it->it_callout, tvtohz(&tv),
1473 			realtimer_expire, it);
1474 	} else {
1475 		callout_stop(&it->it_callout);
1476 	}
1477 
1478 	return (0);
1479 }
1480 
1481 static void
1482 realtimer_clocktime(clockid_t id, struct timespec *ts)
1483 {
1484 	if (id == CLOCK_REALTIME)
1485 		getnanotime(ts);
1486 	else	/* CLOCK_MONOTONIC */
1487 		getnanouptime(ts);
1488 }
1489 
1490 int
1491 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1492 {
1493 	struct itimer *it;
1494 
1495 	PROC_LOCK_ASSERT(p, MA_OWNED);
1496 	it = itimer_find(p, timerid);
1497 	if (it != NULL) {
1498 		ksi->ksi_overrun = it->it_overrun;
1499 		it->it_overrun_last = it->it_overrun;
1500 		it->it_overrun = 0;
1501 		ITIMER_UNLOCK(it);
1502 		return (0);
1503 	}
1504 	return (EINVAL);
1505 }
1506 
1507 int
1508 itimespecfix(struct timespec *ts)
1509 {
1510 
1511 	if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1512 		return (EINVAL);
1513 	if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1514 		ts->tv_nsec = tick * 1000;
1515 	return (0);
1516 }
1517 
1518 /* Timeout callback for realtime timer */
1519 static void
1520 realtimer_expire(void *arg)
1521 {
1522 	struct timespec cts, ts;
1523 	struct timeval tv;
1524 	struct itimer *it;
1525 
1526 	it = (struct itimer *)arg;
1527 
1528 	realtimer_clocktime(it->it_clockid, &cts);
1529 	/* Only fire if time is reached. */
1530 	if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1531 		if (timespecisset(&it->it_time.it_interval)) {
1532 			timespecadd(&it->it_time.it_value,
1533 				    &it->it_time.it_interval);
1534 			while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1535 				if (it->it_overrun < INT_MAX)
1536 					it->it_overrun++;
1537 				else
1538 					it->it_ksi.ksi_errno = ERANGE;
1539 				timespecadd(&it->it_time.it_value,
1540 					    &it->it_time.it_interval);
1541 			}
1542 		} else {
1543 			/* single shot timer ? */
1544 			timespecclear(&it->it_time.it_value);
1545 		}
1546 		if (timespecisset(&it->it_time.it_value)) {
1547 			ts = it->it_time.it_value;
1548 			timespecsub(&ts, &cts);
1549 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1550 			callout_reset(&it->it_callout, tvtohz(&tv),
1551 				 realtimer_expire, it);
1552 		}
1553 		itimer_enter(it);
1554 		ITIMER_UNLOCK(it);
1555 		itimer_fire(it);
1556 		ITIMER_LOCK(it);
1557 		itimer_leave(it);
1558 	} else if (timespecisset(&it->it_time.it_value)) {
1559 		ts = it->it_time.it_value;
1560 		timespecsub(&ts, &cts);
1561 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1562 		callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1563  			it);
1564 	}
1565 }
1566 
1567 void
1568 itimer_fire(struct itimer *it)
1569 {
1570 	struct proc *p = it->it_proc;
1571 	struct thread *td;
1572 
1573 	if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1574 	    it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1575 		if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1576 			ITIMER_LOCK(it);
1577 			timespecclear(&it->it_time.it_value);
1578 			timespecclear(&it->it_time.it_interval);
1579 			callout_stop(&it->it_callout);
1580 			ITIMER_UNLOCK(it);
1581 			return;
1582 		}
1583 		if (!KSI_ONQ(&it->it_ksi)) {
1584 			it->it_ksi.ksi_errno = 0;
1585 			ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1586 			tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1587 		} else {
1588 			if (it->it_overrun < INT_MAX)
1589 				it->it_overrun++;
1590 			else
1591 				it->it_ksi.ksi_errno = ERANGE;
1592 		}
1593 		PROC_UNLOCK(p);
1594 	}
1595 }
1596 
1597 static void
1598 itimers_alloc(struct proc *p)
1599 {
1600 	struct itimers *its;
1601 	int i;
1602 
1603 	its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1604 	LIST_INIT(&its->its_virtual);
1605 	LIST_INIT(&its->its_prof);
1606 	TAILQ_INIT(&its->its_worklist);
1607 	for (i = 0; i < TIMER_MAX; i++)
1608 		its->its_timers[i] = NULL;
1609 	PROC_LOCK(p);
1610 	if (p->p_itimers == NULL) {
1611 		p->p_itimers = its;
1612 		PROC_UNLOCK(p);
1613 	}
1614 	else {
1615 		PROC_UNLOCK(p);
1616 		free(its, M_SUBPROC);
1617 	}
1618 }
1619 
1620 static void
1621 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1622 {
1623 	itimers_event_hook_exit(arg, p);
1624 }
1625 
1626 /* Clean up timers when some process events are being triggered. */
1627 static void
1628 itimers_event_hook_exit(void *arg, struct proc *p)
1629 {
1630 	struct itimers *its;
1631 	struct itimer *it;
1632 	int event = (int)(intptr_t)arg;
1633 	int i;
1634 
1635 	if (p->p_itimers != NULL) {
1636 		its = p->p_itimers;
1637 		for (i = 0; i < MAX_CLOCKS; ++i) {
1638 			if (posix_clocks[i].event_hook != NULL)
1639 				CLOCK_CALL(i, event_hook, (p, i, event));
1640 		}
1641 		/*
1642 		 * According to susv3, XSI interval timers should be inherited
1643 		 * by new image.
1644 		 */
1645 		if (event == ITIMER_EV_EXEC)
1646 			i = 3;
1647 		else if (event == ITIMER_EV_EXIT)
1648 			i = 0;
1649 		else
1650 			panic("unhandled event");
1651 		for (; i < TIMER_MAX; ++i) {
1652 			if ((it = its->its_timers[i]) != NULL)
1653 				kern_ktimer_delete(curthread, i);
1654 		}
1655 		if (its->its_timers[0] == NULL &&
1656 		    its->its_timers[1] == NULL &&
1657 		    its->its_timers[2] == NULL) {
1658 			free(its, M_SUBPROC);
1659 			p->p_itimers = NULL;
1660 		}
1661 	}
1662 }
1663