xref: /netbsd-src/sys/kern/kern_time.c (revision 46f5119e40af2e51998f686b2fdcc76b5488f7f3)
1 /*	$NetBSD: kern_time.c,v 1.168 2011/04/08 10:35:37 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 2000, 2004, 2005, 2007, 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Christopher G. Demetriou, and by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1982, 1986, 1989, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the University nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  *	@(#)kern_time.c	8.4 (Berkeley) 5/26/95
61  */
62 
63 #include <sys/cdefs.h>
64 __KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.168 2011/04/08 10:35:37 yamt Exp $");
65 
66 #include <sys/param.h>
67 #include <sys/resourcevar.h>
68 #include <sys/kernel.h>
69 #include <sys/systm.h>
70 #include <sys/proc.h>
71 #include <sys/vnode.h>
72 #include <sys/signalvar.h>
73 #include <sys/syslog.h>
74 #include <sys/timetc.h>
75 #include <sys/timex.h>
76 #include <sys/kauth.h>
77 #include <sys/mount.h>
78 #include <sys/sa.h>
79 #include <sys/savar.h>
80 #include <sys/syscallargs.h>
81 #include <sys/cpu.h>
82 
83 #include <uvm/uvm_extern.h>
84 
85 #include "opt_sa.h"
86 
87 static void	timer_intr(void *);
88 static void	itimerfire(struct ptimer *);
89 static void	itimerfree(struct ptimers *, int);
90 
91 kmutex_t	timer_lock;
92 
93 static void	*timer_sih;
94 static TAILQ_HEAD(, ptimer) timer_queue;
95 
96 struct pool ptimer_pool, ptimers_pool;
97 
98 #define	CLOCK_VIRTUAL_P(clockid)	\
99 	((clockid) == CLOCK_VIRTUAL || (clockid) == CLOCK_PROF)
100 
101 CTASSERT(ITIMER_REAL == CLOCK_REALTIME);
102 CTASSERT(ITIMER_VIRTUAL == CLOCK_VIRTUAL);
103 CTASSERT(ITIMER_PROF == CLOCK_PROF);
104 
105 /*
106  * Initialize timekeeping.
107  */
108 void
109 time_init(void)
110 {
111 
112 	pool_init(&ptimer_pool, sizeof(struct ptimer), 0, 0, 0, "ptimerpl",
113 	    &pool_allocator_nointr, IPL_NONE);
114 	pool_init(&ptimers_pool, sizeof(struct ptimers), 0, 0, 0, "ptimerspl",
115 	    &pool_allocator_nointr, IPL_NONE);
116 }
117 
118 void
119 time_init2(void)
120 {
121 
122 	TAILQ_INIT(&timer_queue);
123 	mutex_init(&timer_lock, MUTEX_DEFAULT, IPL_SCHED);
124 	timer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
125 	    timer_intr, NULL);
126 }
127 
128 /* Time of day and interval timer support.
129  *
130  * These routines provide the kernel entry points to get and set
131  * the time-of-day and per-process interval timers.  Subroutines
132  * here provide support for adding and subtracting timeval structures
133  * and decrementing interval timers, optionally reloading the interval
134  * timers when they expire.
135  */
136 
137 /* This function is used by clock_settime and settimeofday */
138 static int
139 settime1(struct proc *p, const struct timespec *ts, bool check_kauth)
140 {
141 	struct timespec delta, now;
142 	int s;
143 
144 	/* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */
145 	s = splclock();
146 	nanotime(&now);
147 	timespecsub(ts, &now, &delta);
148 
149 	if (check_kauth && kauth_authorize_system(kauth_cred_get(),
150 	    KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, __UNCONST(ts),
151 	    &delta, KAUTH_ARG(check_kauth ? false : true)) != 0) {
152 		splx(s);
153 		return (EPERM);
154 	}
155 
156 #ifdef notyet
157 	if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */
158 		splx(s);
159 		return (EPERM);
160 	}
161 #endif
162 
163 	tc_setclock(ts);
164 
165 	timespecadd(&boottime, &delta, &boottime);
166 
167 	resettodr();
168 	splx(s);
169 
170 	return (0);
171 }
172 
173 int
174 settime(struct proc *p, struct timespec *ts)
175 {
176 	return (settime1(p, ts, true));
177 }
178 
179 /* ARGSUSED */
180 int
181 sys___clock_gettime50(struct lwp *l,
182     const struct sys___clock_gettime50_args *uap, register_t *retval)
183 {
184 	/* {
185 		syscallarg(clockid_t) clock_id;
186 		syscallarg(struct timespec *) tp;
187 	} */
188 	int error;
189 	struct timespec ats;
190 
191 	error = clock_gettime1(SCARG(uap, clock_id), &ats);
192 	if (error != 0)
193 		return error;
194 
195 	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
196 }
197 
198 int
199 clock_gettime1(clockid_t clock_id, struct timespec *ts)
200 {
201 
202 	switch (clock_id) {
203 	case CLOCK_REALTIME:
204 		nanotime(ts);
205 		break;
206 	case CLOCK_MONOTONIC:
207 		nanouptime(ts);
208 		break;
209 	default:
210 		return EINVAL;
211 	}
212 
213 	return 0;
214 }
215 
216 /* ARGSUSED */
217 int
218 sys___clock_settime50(struct lwp *l,
219     const struct sys___clock_settime50_args *uap, register_t *retval)
220 {
221 	/* {
222 		syscallarg(clockid_t) clock_id;
223 		syscallarg(const struct timespec *) tp;
224 	} */
225 	int error;
226 	struct timespec ats;
227 
228 	if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
229 		return error;
230 
231 	return clock_settime1(l->l_proc, SCARG(uap, clock_id), &ats, true);
232 }
233 
234 
235 int
236 clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp,
237     bool check_kauth)
238 {
239 	int error;
240 
241 	switch (clock_id) {
242 	case CLOCK_REALTIME:
243 		if ((error = settime1(p, tp, check_kauth)) != 0)
244 			return (error);
245 		break;
246 	case CLOCK_MONOTONIC:
247 		return (EINVAL);	/* read-only clock */
248 	default:
249 		return (EINVAL);
250 	}
251 
252 	return 0;
253 }
254 
255 int
256 sys___clock_getres50(struct lwp *l, const struct sys___clock_getres50_args *uap,
257     register_t *retval)
258 {
259 	/* {
260 		syscallarg(clockid_t) clock_id;
261 		syscallarg(struct timespec *) tp;
262 	} */
263 	struct timespec ts;
264 	int error = 0;
265 
266 	if ((error = clock_getres1(SCARG(uap, clock_id), &ts)) != 0)
267 		return error;
268 
269 	if (SCARG(uap, tp))
270 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
271 
272 	return error;
273 }
274 
275 int
276 clock_getres1(clockid_t clock_id, struct timespec *ts)
277 {
278 
279 	switch (clock_id) {
280 	case CLOCK_REALTIME:
281 	case CLOCK_MONOTONIC:
282 		ts->tv_sec = 0;
283 		if (tc_getfrequency() > 1000000000)
284 			ts->tv_nsec = 1;
285 		else
286 			ts->tv_nsec = 1000000000 / tc_getfrequency();
287 		break;
288 	default:
289 		return EINVAL;
290 	}
291 
292 	return 0;
293 }
294 
295 /* ARGSUSED */
296 int
297 sys___nanosleep50(struct lwp *l, const struct sys___nanosleep50_args *uap,
298     register_t *retval)
299 {
300 	/* {
301 		syscallarg(struct timespec *) rqtp;
302 		syscallarg(struct timespec *) rmtp;
303 	} */
304 	struct timespec rmt, rqt;
305 	int error, error1;
306 
307 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
308 	if (error)
309 		return (error);
310 
311 	error = nanosleep1(l, &rqt, SCARG(uap, rmtp) ? &rmt : NULL);
312 	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
313 		return error;
314 
315 	error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt));
316 	return error1 ? error1 : error;
317 }
318 
319 int
320 nanosleep1(struct lwp *l, struct timespec *rqt, struct timespec *rmt)
321 {
322 	struct timespec rmtstart;
323 	int error, timo;
324 
325 	if ((error = itimespecfix(rqt)) != 0)
326 		return error;
327 
328 	timo = tstohz(rqt);
329 	/*
330 	 * Avoid inadvertantly sleeping forever
331 	 */
332 	if (timo == 0)
333 		timo = 1;
334 	getnanouptime(&rmtstart);
335 again:
336 	error = kpause("nanoslp", true, timo, NULL);
337 	if (rmt != NULL || error == 0) {
338 		struct timespec rmtend;
339 		struct timespec t0;
340 		struct timespec *t;
341 
342 		getnanouptime(&rmtend);
343 		t = (rmt != NULL) ? rmt : &t0;
344 		timespecsub(&rmtend, &rmtstart, t);
345 		timespecsub(rqt, t, t);
346 		if (t->tv_sec < 0)
347 			timespecclear(t);
348 		if (error == 0) {
349 			timo = tstohz(t);
350 			if (timo > 0)
351 				goto again;
352 		}
353 	}
354 
355 	if (error == ERESTART)
356 		error = EINTR;
357 	if (error == EWOULDBLOCK)
358 		error = 0;
359 
360 	return error;
361 }
362 
363 /* ARGSUSED */
364 int
365 sys___gettimeofday50(struct lwp *l, const struct sys___gettimeofday50_args *uap,
366     register_t *retval)
367 {
368 	/* {
369 		syscallarg(struct timeval *) tp;
370 		syscallarg(void *) tzp;		really "struct timezone *";
371 	} */
372 	struct timeval atv;
373 	int error = 0;
374 	struct timezone tzfake;
375 
376 	if (SCARG(uap, tp)) {
377 		microtime(&atv);
378 		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
379 		if (error)
380 			return (error);
381 	}
382 	if (SCARG(uap, tzp)) {
383 		/*
384 		 * NetBSD has no kernel notion of time zone, so we just
385 		 * fake up a timezone struct and return it if demanded.
386 		 */
387 		tzfake.tz_minuteswest = 0;
388 		tzfake.tz_dsttime = 0;
389 		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
390 	}
391 	return (error);
392 }
393 
394 /* ARGSUSED */
395 int
396 sys___settimeofday50(struct lwp *l, const struct sys___settimeofday50_args *uap,
397     register_t *retval)
398 {
399 	/* {
400 		syscallarg(const struct timeval *) tv;
401 		syscallarg(const void *) tzp; really "const struct timezone *";
402 	} */
403 
404 	return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true);
405 }
406 
407 int
408 settimeofday1(const struct timeval *utv, bool userspace,
409     const void *utzp, struct lwp *l, bool check_kauth)
410 {
411 	struct timeval atv;
412 	struct timespec ts;
413 	int error;
414 
415 	/* Verify all parameters before changing time. */
416 
417 	/*
418 	 * NetBSD has no kernel notion of time zone, and only an
419 	 * obsolete program would try to set it, so we log a warning.
420 	 */
421 	if (utzp)
422 		log(LOG_WARNING, "pid %d attempted to set the "
423 		    "(obsolete) kernel time zone\n", l->l_proc->p_pid);
424 
425 	if (utv == NULL)
426 		return 0;
427 
428 	if (userspace) {
429 		if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
430 			return error;
431 		utv = &atv;
432 	}
433 
434 	TIMEVAL_TO_TIMESPEC(utv, &ts);
435 	return settime1(l->l_proc, &ts, check_kauth);
436 }
437 
438 int	time_adjusted;			/* set if an adjustment is made */
439 
440 /* ARGSUSED */
441 int
442 sys___adjtime50(struct lwp *l, const struct sys___adjtime50_args *uap,
443     register_t *retval)
444 {
445 	/* {
446 		syscallarg(const struct timeval *) delta;
447 		syscallarg(struct timeval *) olddelta;
448 	} */
449 	int error = 0;
450 	struct timeval atv, oldatv;
451 
452 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
453 	    KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0)
454 		return error;
455 
456 	if (SCARG(uap, delta)) {
457 		error = copyin(SCARG(uap, delta), &atv,
458 		    sizeof(*SCARG(uap, delta)));
459 		if (error)
460 			return (error);
461 	}
462 	adjtime1(SCARG(uap, delta) ? &atv : NULL,
463 	    SCARG(uap, olddelta) ? &oldatv : NULL, l->l_proc);
464 	if (SCARG(uap, olddelta))
465 		error = copyout(&oldatv, SCARG(uap, olddelta),
466 		    sizeof(*SCARG(uap, olddelta)));
467 	return error;
468 }
469 
470 void
471 adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p)
472 {
473 	extern int64_t time_adjtime;  /* in kern_ntptime.c */
474 
475 	if (olddelta) {
476 		mutex_spin_enter(&timecounter_lock);
477 		olddelta->tv_sec = time_adjtime / 1000000;
478 		olddelta->tv_usec = time_adjtime % 1000000;
479 		if (olddelta->tv_usec < 0) {
480 			olddelta->tv_usec += 1000000;
481 			olddelta->tv_sec--;
482 		}
483 		mutex_spin_exit(&timecounter_lock);
484 	}
485 
486 	if (delta) {
487 		mutex_spin_enter(&timecounter_lock);
488 		time_adjtime = delta->tv_sec * 1000000 + delta->tv_usec;
489 
490 		if (time_adjtime) {
491 			/* We need to save the system time during shutdown */
492 			time_adjusted |= 1;
493 		}
494 		mutex_spin_exit(&timecounter_lock);
495 	}
496 }
497 
498 /*
499  * Interval timer support. Both the BSD getitimer() family and the POSIX
500  * timer_*() family of routines are supported.
501  *
502  * All timers are kept in an array pointed to by p_timers, which is
503  * allocated on demand - many processes don't use timers at all. The
504  * first three elements in this array are reserved for the BSD timers:
505  * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, and element
506  * 2 is ITIMER_PROF. The rest may be allocated by the timer_create()
507  * syscall.
508  *
509  * Realtime timers are kept in the ptimer structure as an absolute
510  * time; virtual time timers are kept as a linked list of deltas.
511  * Virtual time timers are processed in the hardclock() routine of
512  * kern_clock.c.  The real time timer is processed by a callout
513  * routine, called from the softclock() routine.  Since a callout may
514  * be delayed in real time due to interrupt processing in the system,
515  * it is possible for the real time timeout routine (realtimeexpire,
516  * given below), to be delayed in real time past when it is supposed
517  * to occur.  It does not suffice, therefore, to reload the real timer
518  * .it_value from the real time timers .it_interval.  Rather, we
519  * compute the next time in absolute time the timer should go off.  */
520 
521 /* Allocate a POSIX realtime timer. */
522 int
523 sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap,
524     register_t *retval)
525 {
526 	/* {
527 		syscallarg(clockid_t) clock_id;
528 		syscallarg(struct sigevent *) evp;
529 		syscallarg(timer_t *) timerid;
530 	} */
531 
532 	return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id),
533 	    SCARG(uap, evp), copyin, l);
534 }
535 
536 int
537 timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp,
538     copyin_t fetch_event, struct lwp *l)
539 {
540 	int error;
541 	timer_t timerid;
542 	struct ptimers *pts;
543 	struct ptimer *pt;
544 	struct proc *p;
545 
546 	p = l->l_proc;
547 
548 	if (id != CLOCK_REALTIME && id != CLOCK_VIRTUAL &&
549 	    id != CLOCK_PROF && id != CLOCK_MONOTONIC)
550 		return (EINVAL);
551 
552 	if ((pts = p->p_timers) == NULL)
553 		pts = timers_alloc(p);
554 
555 	pt = pool_get(&ptimer_pool, PR_WAITOK);
556 	if (evp != NULL) {
557 		if (((error =
558 		    (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) ||
559 		    ((pt->pt_ev.sigev_notify < SIGEV_NONE) ||
560 			(pt->pt_ev.sigev_notify > SIGEV_SA)) ||
561 			(pt->pt_ev.sigev_notify == SIGEV_SIGNAL &&
562 			 (pt->pt_ev.sigev_signo <= 0 ||
563 			  pt->pt_ev.sigev_signo >= NSIG))) {
564 			pool_put(&ptimer_pool, pt);
565 			return (error ? error : EINVAL);
566 		}
567 	}
568 
569 	/* Find a free timer slot, skipping those reserved for setitimer(). */
570 	mutex_spin_enter(&timer_lock);
571 	for (timerid = 3; timerid < TIMER_MAX; timerid++)
572 		if (pts->pts_timers[timerid] == NULL)
573 			break;
574 	if (timerid == TIMER_MAX) {
575 		mutex_spin_exit(&timer_lock);
576 		pool_put(&ptimer_pool, pt);
577 		return EAGAIN;
578 	}
579 	if (evp == NULL) {
580 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
581 		switch (id) {
582 		case CLOCK_REALTIME:
583 		case CLOCK_MONOTONIC:
584 			pt->pt_ev.sigev_signo = SIGALRM;
585 			break;
586 		case CLOCK_VIRTUAL:
587 			pt->pt_ev.sigev_signo = SIGVTALRM;
588 			break;
589 		case CLOCK_PROF:
590 			pt->pt_ev.sigev_signo = SIGPROF;
591 			break;
592 		}
593 		pt->pt_ev.sigev_value.sival_int = timerid;
594 	}
595 	pt->pt_info.ksi_signo = pt->pt_ev.sigev_signo;
596 	pt->pt_info.ksi_errno = 0;
597 	pt->pt_info.ksi_code = 0;
598 	pt->pt_info.ksi_pid = p->p_pid;
599 	pt->pt_info.ksi_uid = kauth_cred_getuid(l->l_cred);
600 	pt->pt_info.ksi_value = pt->pt_ev.sigev_value;
601 	pt->pt_type = id;
602 	pt->pt_proc = p;
603 	pt->pt_overruns = 0;
604 	pt->pt_poverruns = 0;
605 	pt->pt_entry = timerid;
606 	pt->pt_queued = false;
607 	timespecclear(&pt->pt_time.it_value);
608 	if (!CLOCK_VIRTUAL_P(id))
609 		callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
610 	else
611 		pt->pt_active = 0;
612 
613 	pts->pts_timers[timerid] = pt;
614 	mutex_spin_exit(&timer_lock);
615 
616 	return copyout(&timerid, tid, sizeof(timerid));
617 }
618 
619 /* Delete a POSIX realtime timer */
620 int
621 sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap,
622     register_t *retval)
623 {
624 	/* {
625 		syscallarg(timer_t) timerid;
626 	} */
627 	struct proc *p = l->l_proc;
628 	timer_t timerid;
629 	struct ptimers *pts;
630 	struct ptimer *pt, *ptn;
631 
632 	timerid = SCARG(uap, timerid);
633 	pts = p->p_timers;
634 
635 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
636 		return (EINVAL);
637 
638 	mutex_spin_enter(&timer_lock);
639 	if ((pt = pts->pts_timers[timerid]) == NULL) {
640 		mutex_spin_exit(&timer_lock);
641 		return (EINVAL);
642 	}
643 	if (CLOCK_VIRTUAL_P(pt->pt_type)) {
644 		if (pt->pt_active) {
645 			ptn = LIST_NEXT(pt, pt_list);
646 			LIST_REMOVE(pt, pt_list);
647 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
648 				timespecadd(&pt->pt_time.it_value,
649 				    &ptn->pt_time.it_value,
650 				    &ptn->pt_time.it_value);
651 			pt->pt_active = 0;
652 		}
653 	}
654 	itimerfree(pts, timerid);
655 
656 	return (0);
657 }
658 
659 /*
660  * Set up the given timer. The value in pt->pt_time.it_value is taken
661  * to be an absolute time for CLOCK_REALTIME/CLOCK_MONOTONIC timers and
662  * a relative time for CLOCK_VIRTUAL/CLOCK_PROF timers.
663  */
664 void
665 timer_settime(struct ptimer *pt)
666 {
667 	struct ptimer *ptn, *pptn;
668 	struct ptlist *ptl;
669 
670 	KASSERT(mutex_owned(&timer_lock));
671 
672 	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
673 		callout_halt(&pt->pt_ch, &timer_lock);
674 		if (timespecisset(&pt->pt_time.it_value)) {
675 			/*
676 			 * Don't need to check tshzto() return value, here.
677 			 * callout_reset() does it for us.
678 			 */
679 			callout_reset(&pt->pt_ch, tshzto(&pt->pt_time.it_value),
680 			    realtimerexpire, pt);
681 		}
682 	} else {
683 		if (pt->pt_active) {
684 			ptn = LIST_NEXT(pt, pt_list);
685 			LIST_REMOVE(pt, pt_list);
686 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
687 				timespecadd(&pt->pt_time.it_value,
688 				    &ptn->pt_time.it_value,
689 				    &ptn->pt_time.it_value);
690 		}
691 		if (timespecisset(&pt->pt_time.it_value)) {
692 			if (pt->pt_type == CLOCK_VIRTUAL)
693 				ptl = &pt->pt_proc->p_timers->pts_virtual;
694 			else
695 				ptl = &pt->pt_proc->p_timers->pts_prof;
696 
697 			for (ptn = LIST_FIRST(ptl), pptn = NULL;
698 			     ptn && timespeccmp(&pt->pt_time.it_value,
699 				 &ptn->pt_time.it_value, >);
700 			     pptn = ptn, ptn = LIST_NEXT(ptn, pt_list))
701 				timespecsub(&pt->pt_time.it_value,
702 				    &ptn->pt_time.it_value,
703 				    &pt->pt_time.it_value);
704 
705 			if (pptn)
706 				LIST_INSERT_AFTER(pptn, pt, pt_list);
707 			else
708 				LIST_INSERT_HEAD(ptl, pt, pt_list);
709 
710 			for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list))
711 				timespecsub(&ptn->pt_time.it_value,
712 				    &pt->pt_time.it_value,
713 				    &ptn->pt_time.it_value);
714 
715 			pt->pt_active = 1;
716 		} else
717 			pt->pt_active = 0;
718 	}
719 }
720 
721 void
722 timer_gettime(struct ptimer *pt, struct itimerspec *aits)
723 {
724 	struct timespec now;
725 	struct ptimer *ptn;
726 
727 	KASSERT(mutex_owned(&timer_lock));
728 
729 	*aits = pt->pt_time;
730 	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
731 		/*
732 		 * Convert from absolute to relative time in .it_value
733 		 * part of real time timer.  If time for real time
734 		 * timer has passed return 0, else return difference
735 		 * between current time and time for the timer to go
736 		 * off.
737 		 */
738 		if (timespecisset(&aits->it_value)) {
739 			if (pt->pt_type == CLOCK_REALTIME) {
740 				getnanotime(&now);
741 			} else { /* CLOCK_MONOTONIC */
742 				getnanouptime(&now);
743 			}
744 			if (timespeccmp(&aits->it_value, &now, <))
745 				timespecclear(&aits->it_value);
746 			else
747 				timespecsub(&aits->it_value, &now,
748 				    &aits->it_value);
749 		}
750 	} else if (pt->pt_active) {
751 		if (pt->pt_type == CLOCK_VIRTUAL)
752 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual);
753 		else
754 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof);
755 		for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list))
756 			timespecadd(&aits->it_value,
757 			    &ptn->pt_time.it_value, &aits->it_value);
758 		KASSERT(ptn != NULL); /* pt should be findable on the list */
759 	} else
760 		timespecclear(&aits->it_value);
761 }
762 
763 
764 
765 /* Set and arm a POSIX realtime timer */
766 int
767 sys___timer_settime50(struct lwp *l,
768     const struct sys___timer_settime50_args *uap,
769     register_t *retval)
770 {
771 	/* {
772 		syscallarg(timer_t) timerid;
773 		syscallarg(int) flags;
774 		syscallarg(const struct itimerspec *) value;
775 		syscallarg(struct itimerspec *) ovalue;
776 	} */
777 	int error;
778 	struct itimerspec value, ovalue, *ovp = NULL;
779 
780 	if ((error = copyin(SCARG(uap, value), &value,
781 	    sizeof(struct itimerspec))) != 0)
782 		return (error);
783 
784 	if (SCARG(uap, ovalue))
785 		ovp = &ovalue;
786 
787 	if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp,
788 	    SCARG(uap, flags), l->l_proc)) != 0)
789 		return error;
790 
791 	if (ovp)
792 		return copyout(&ovalue, SCARG(uap, ovalue),
793 		    sizeof(struct itimerspec));
794 	return 0;
795 }
796 
797 int
798 dotimer_settime(int timerid, struct itimerspec *value,
799     struct itimerspec *ovalue, int flags, struct proc *p)
800 {
801 	struct timespec now;
802 	struct itimerspec val, oval;
803 	struct ptimers *pts;
804 	struct ptimer *pt;
805 	int error;
806 
807 	pts = p->p_timers;
808 
809 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
810 		return EINVAL;
811 	val = *value;
812 	if ((error = itimespecfix(&val.it_value)) != 0 ||
813 	    (error = itimespecfix(&val.it_interval)) != 0)
814 		return error;
815 
816 	mutex_spin_enter(&timer_lock);
817 	if ((pt = pts->pts_timers[timerid]) == NULL) {
818 		mutex_spin_exit(&timer_lock);
819 		return EINVAL;
820 	}
821 
822 	oval = pt->pt_time;
823 	pt->pt_time = val;
824 
825 	/*
826 	 * If we've been passed a relative time for a realtime timer,
827 	 * convert it to absolute; if an absolute time for a virtual
828 	 * timer, convert it to relative and make sure we don't set it
829 	 * to zero, which would cancel the timer, or let it go
830 	 * negative, which would confuse the comparison tests.
831 	 */
832 	if (timespecisset(&pt->pt_time.it_value)) {
833 		if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
834 			if ((flags & TIMER_ABSTIME) == 0) {
835 				if (pt->pt_type == CLOCK_REALTIME) {
836 					getnanotime(&now);
837 				} else { /* CLOCK_MONOTONIC */
838 					getnanouptime(&now);
839 				}
840 				timespecadd(&pt->pt_time.it_value, &now,
841 				    &pt->pt_time.it_value);
842 			}
843 		} else {
844 			if ((flags & TIMER_ABSTIME) != 0) {
845 				getnanotime(&now);
846 				timespecsub(&pt->pt_time.it_value, &now,
847 				    &pt->pt_time.it_value);
848 				if (!timespecisset(&pt->pt_time.it_value) ||
849 				    pt->pt_time.it_value.tv_sec < 0) {
850 					pt->pt_time.it_value.tv_sec = 0;
851 					pt->pt_time.it_value.tv_nsec = 1;
852 				}
853 			}
854 		}
855 	}
856 
857 	timer_settime(pt);
858 	mutex_spin_exit(&timer_lock);
859 
860 	if (ovalue)
861 		*ovalue = oval;
862 
863 	return (0);
864 }
865 
866 /* Return the time remaining until a POSIX timer fires. */
867 int
868 sys___timer_gettime50(struct lwp *l,
869     const struct sys___timer_gettime50_args *uap, register_t *retval)
870 {
871 	/* {
872 		syscallarg(timer_t) timerid;
873 		syscallarg(struct itimerspec *) value;
874 	} */
875 	struct itimerspec its;
876 	int error;
877 
878 	if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc,
879 	    &its)) != 0)
880 		return error;
881 
882 	return copyout(&its, SCARG(uap, value), sizeof(its));
883 }
884 
885 int
886 dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its)
887 {
888 	struct ptimer *pt;
889 	struct ptimers *pts;
890 
891 	pts = p->p_timers;
892 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
893 		return (EINVAL);
894 	mutex_spin_enter(&timer_lock);
895 	if ((pt = pts->pts_timers[timerid]) == NULL) {
896 		mutex_spin_exit(&timer_lock);
897 		return (EINVAL);
898 	}
899 	timer_gettime(pt, its);
900 	mutex_spin_exit(&timer_lock);
901 
902 	return 0;
903 }
904 
905 /*
906  * Return the count of the number of times a periodic timer expired
907  * while a notification was already pending. The counter is reset when
908  * a timer expires and a notification can be posted.
909  */
910 int
911 sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap,
912     register_t *retval)
913 {
914 	/* {
915 		syscallarg(timer_t) timerid;
916 	} */
917 	struct proc *p = l->l_proc;
918 	struct ptimers *pts;
919 	int timerid;
920 	struct ptimer *pt;
921 
922 	timerid = SCARG(uap, timerid);
923 
924 	pts = p->p_timers;
925 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
926 		return (EINVAL);
927 	mutex_spin_enter(&timer_lock);
928 	if ((pt = pts->pts_timers[timerid]) == NULL) {
929 		mutex_spin_exit(&timer_lock);
930 		return (EINVAL);
931 	}
932 	*retval = pt->pt_poverruns;
933 	mutex_spin_exit(&timer_lock);
934 
935 	return (0);
936 }
937 
938 #ifdef KERN_SA
939 /* Glue function that triggers an upcall; called from userret(). */
940 void
941 timerupcall(struct lwp *l)
942 {
943 	struct ptimers *pt = l->l_proc->p_timers;
944 	struct proc *p = l->l_proc;
945 	unsigned int i, fired, done;
946 
947 	KDASSERT(l->l_proc->p_sa);
948 	/* Bail out if we do not own the virtual processor */
949 	if (l->l_savp->savp_lwp != l)
950 		return ;
951 
952 	mutex_enter(p->p_lock);
953 
954 	fired = pt->pts_fired;
955 	done = 0;
956 	while ((i = ffs(fired)) != 0) {
957 		siginfo_t *si;
958 		int mask = 1 << --i;
959 		int f;
960 
961 		f = ~l->l_pflag & LP_SA_NOBLOCK;
962 		l->l_pflag |= LP_SA_NOBLOCK;
963 		si = siginfo_alloc(PR_WAITOK);
964 		si->_info = pt->pts_timers[i]->pt_info.ksi_info;
965 		if (sa_upcall(l, SA_UPCALL_SIGEV | SA_UPCALL_DEFER, NULL, l,
966 		    sizeof(*si), si, siginfo_free) != 0) {
967 			siginfo_free(si);
968 			/* XXX What do we do here?? */
969 		} else
970 			done |= mask;
971 		fired &= ~mask;
972 		l->l_pflag ^= f;
973 	}
974 	pt->pts_fired &= ~done;
975 	if (pt->pts_fired == 0)
976 		l->l_proc->p_timerpend = 0;
977 
978 	mutex_exit(p->p_lock);
979 }
980 #endif /* KERN_SA */
981 
982 /*
983  * Real interval timer expired:
984  * send process whose timer expired an alarm signal.
985  * If time is not set up to reload, then just return.
986  * Else compute next time timer should go off which is > current time.
987  * This is where delay in processing this timeout causes multiple
988  * SIGALRM calls to be compressed into one.
989  */
990 void
991 realtimerexpire(void *arg)
992 {
993 	uint64_t last_val, next_val, interval, now_ns;
994 	struct timespec now, next;
995 	struct ptimer *pt;
996 	int backwards;
997 
998 	pt = arg;
999 
1000 	mutex_spin_enter(&timer_lock);
1001 	itimerfire(pt);
1002 
1003 	if (!timespecisset(&pt->pt_time.it_interval)) {
1004 		timespecclear(&pt->pt_time.it_value);
1005 		mutex_spin_exit(&timer_lock);
1006 		return;
1007 	}
1008 
1009 	getnanotime(&now);
1010 	backwards = (timespeccmp(&pt->pt_time.it_value, &now, >));
1011 	timespecadd(&pt->pt_time.it_value, &pt->pt_time.it_interval, &next);
1012 	/* Handle the easy case of non-overflown timers first. */
1013 	if (!backwards && timespeccmp(&next, &now, >)) {
1014 		pt->pt_time.it_value = next;
1015 	} else {
1016 		now_ns = timespec2ns(&now);
1017 		last_val = timespec2ns(&pt->pt_time.it_value);
1018 		interval = timespec2ns(&pt->pt_time.it_interval);
1019 
1020 		next_val = now_ns +
1021 		    (now_ns - last_val + interval - 1) % interval;
1022 
1023 		if (backwards)
1024 			next_val += interval;
1025 		else
1026 			pt->pt_overruns += (now_ns - last_val) / interval;
1027 
1028 		pt->pt_time.it_value.tv_sec = next_val / 1000000000;
1029 		pt->pt_time.it_value.tv_nsec = next_val % 1000000000;
1030 	}
1031 
1032 	/*
1033 	 * Don't need to check tshzto() return value, here.
1034 	 * callout_reset() does it for us.
1035 	 */
1036 	callout_reset(&pt->pt_ch, tshzto(&pt->pt_time.it_value),
1037 	    realtimerexpire, pt);
1038 	mutex_spin_exit(&timer_lock);
1039 }
1040 
1041 /* BSD routine to get the value of an interval timer. */
1042 /* ARGSUSED */
1043 int
1044 sys___getitimer50(struct lwp *l, const struct sys___getitimer50_args *uap,
1045     register_t *retval)
1046 {
1047 	/* {
1048 		syscallarg(int) which;
1049 		syscallarg(struct itimerval *) itv;
1050 	} */
1051 	struct proc *p = l->l_proc;
1052 	struct itimerval aitv;
1053 	int error;
1054 
1055 	error = dogetitimer(p, SCARG(uap, which), &aitv);
1056 	if (error)
1057 		return error;
1058 	return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)));
1059 }
1060 
1061 int
1062 dogetitimer(struct proc *p, int which, struct itimerval *itvp)
1063 {
1064 	struct ptimers *pts;
1065 	struct ptimer *pt;
1066 	struct itimerspec its;
1067 
1068 	if ((u_int)which > ITIMER_PROF)
1069 		return (EINVAL);
1070 
1071 	mutex_spin_enter(&timer_lock);
1072 	pts = p->p_timers;
1073 	if (pts == NULL || (pt = pts->pts_timers[which]) == NULL) {
1074 		timerclear(&itvp->it_value);
1075 		timerclear(&itvp->it_interval);
1076 	} else {
1077 		timer_gettime(pt, &its);
1078 		TIMESPEC_TO_TIMEVAL(&itvp->it_value, &its.it_value);
1079 		TIMESPEC_TO_TIMEVAL(&itvp->it_interval, &its.it_interval);
1080 	}
1081 	mutex_spin_exit(&timer_lock);
1082 
1083 	return 0;
1084 }
1085 
1086 /* BSD routine to set/arm an interval timer. */
1087 /* ARGSUSED */
1088 int
1089 sys___setitimer50(struct lwp *l, const struct sys___setitimer50_args *uap,
1090     register_t *retval)
1091 {
1092 	/* {
1093 		syscallarg(int) which;
1094 		syscallarg(const struct itimerval *) itv;
1095 		syscallarg(struct itimerval *) oitv;
1096 	} */
1097 	struct proc *p = l->l_proc;
1098 	int which = SCARG(uap, which);
1099 	struct sys___getitimer50_args getargs;
1100 	const struct itimerval *itvp;
1101 	struct itimerval aitv;
1102 	int error;
1103 
1104 	if ((u_int)which > ITIMER_PROF)
1105 		return (EINVAL);
1106 	itvp = SCARG(uap, itv);
1107 	if (itvp &&
1108 	    (error = copyin(itvp, &aitv, sizeof(struct itimerval)) != 0))
1109 		return (error);
1110 	if (SCARG(uap, oitv) != NULL) {
1111 		SCARG(&getargs, which) = which;
1112 		SCARG(&getargs, itv) = SCARG(uap, oitv);
1113 		if ((error = sys___getitimer50(l, &getargs, retval)) != 0)
1114 			return (error);
1115 	}
1116 	if (itvp == 0)
1117 		return (0);
1118 
1119 	return dosetitimer(p, which, &aitv);
1120 }
1121 
1122 int
1123 dosetitimer(struct proc *p, int which, struct itimerval *itvp)
1124 {
1125 	struct timespec now;
1126 	struct ptimers *pts;
1127 	struct ptimer *pt, *spare;
1128 
1129 	KASSERT(which == CLOCK_REALTIME || which == CLOCK_VIRTUAL ||
1130 	    which == CLOCK_PROF);
1131 	if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval))
1132 		return (EINVAL);
1133 
1134 	/*
1135 	 * Don't bother allocating data structures if the process just
1136 	 * wants to clear the timer.
1137 	 */
1138 	spare = NULL;
1139 	pts = p->p_timers;
1140  retry:
1141 	if (!timerisset(&itvp->it_value) && (pts == NULL ||
1142 	    pts->pts_timers[which] == NULL))
1143 		return (0);
1144 	if (pts == NULL)
1145 		pts = timers_alloc(p);
1146 	mutex_spin_enter(&timer_lock);
1147 	pt = pts->pts_timers[which];
1148 	if (pt == NULL) {
1149 		if (spare == NULL) {
1150 			mutex_spin_exit(&timer_lock);
1151 			spare = pool_get(&ptimer_pool, PR_WAITOK);
1152 			goto retry;
1153 		}
1154 		pt = spare;
1155 		spare = NULL;
1156 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
1157 		pt->pt_ev.sigev_value.sival_int = which;
1158 		pt->pt_overruns = 0;
1159 		pt->pt_proc = p;
1160 		pt->pt_type = which;
1161 		pt->pt_entry = which;
1162 		pt->pt_queued = false;
1163 		if (pt->pt_type == CLOCK_REALTIME)
1164 			callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
1165 		else
1166 			pt->pt_active = 0;
1167 
1168 		switch (which) {
1169 		case ITIMER_REAL:
1170 			pt->pt_ev.sigev_signo = SIGALRM;
1171 			break;
1172 		case ITIMER_VIRTUAL:
1173 			pt->pt_ev.sigev_signo = SIGVTALRM;
1174 			break;
1175 		case ITIMER_PROF:
1176 			pt->pt_ev.sigev_signo = SIGPROF;
1177 			break;
1178 		}
1179 		pts->pts_timers[which] = pt;
1180 	}
1181 
1182 	TIMEVAL_TO_TIMESPEC(&itvp->it_value, &pt->pt_time.it_value);
1183 	TIMEVAL_TO_TIMESPEC(&itvp->it_interval, &pt->pt_time.it_interval);
1184 
1185 	if ((which == ITIMER_REAL) && timespecisset(&pt->pt_time.it_value)) {
1186 		/* Convert to absolute time */
1187 		/* XXX need to wrap in splclock for timecounters case? */
1188 		getnanotime(&now);
1189 		timespecadd(&pt->pt_time.it_value, &now, &pt->pt_time.it_value);
1190 	}
1191 	timer_settime(pt);
1192 	mutex_spin_exit(&timer_lock);
1193 	if (spare != NULL)
1194 		pool_put(&ptimer_pool, spare);
1195 
1196 	return (0);
1197 }
1198 
1199 /* Utility routines to manage the array of pointers to timers. */
1200 struct ptimers *
1201 timers_alloc(struct proc *p)
1202 {
1203 	struct ptimers *pts;
1204 	int i;
1205 
1206 	pts = pool_get(&ptimers_pool, PR_WAITOK);
1207 	LIST_INIT(&pts->pts_virtual);
1208 	LIST_INIT(&pts->pts_prof);
1209 	for (i = 0; i < TIMER_MAX; i++)
1210 		pts->pts_timers[i] = NULL;
1211 	pts->pts_fired = 0;
1212 	mutex_spin_enter(&timer_lock);
1213 	if (p->p_timers == NULL) {
1214 		p->p_timers = pts;
1215 		mutex_spin_exit(&timer_lock);
1216 		return pts;
1217 	}
1218 	mutex_spin_exit(&timer_lock);
1219 	pool_put(&ptimers_pool, pts);
1220 	return p->p_timers;
1221 }
1222 
1223 /*
1224  * Clean up the per-process timers. If "which" is set to TIMERS_ALL,
1225  * then clean up all timers and free all the data structures. If
1226  * "which" is set to TIMERS_POSIX, only clean up the timers allocated
1227  * by timer_create(), not the BSD setitimer() timers, and only free the
1228  * structure if none of those remain.
1229  */
1230 void
1231 timers_free(struct proc *p, int which)
1232 {
1233 	struct ptimers *pts;
1234 	struct ptimer *ptn;
1235 	struct timespec ts;
1236 	int i;
1237 
1238 	if (p->p_timers == NULL)
1239 		return;
1240 
1241 	pts = p->p_timers;
1242 	mutex_spin_enter(&timer_lock);
1243 	if (which == TIMERS_ALL) {
1244 		p->p_timers = NULL;
1245 		i = 0;
1246 	} else {
1247 		timespecclear(&ts);
1248 		for (ptn = LIST_FIRST(&pts->pts_virtual);
1249 		     ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL];
1250 		     ptn = LIST_NEXT(ptn, pt_list)) {
1251 			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
1252 			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
1253 		}
1254 		LIST_FIRST(&pts->pts_virtual) = NULL;
1255 		if (ptn) {
1256 			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
1257 			timespecadd(&ts, &ptn->pt_time.it_value,
1258 			    &ptn->pt_time.it_value);
1259 			LIST_INSERT_HEAD(&pts->pts_virtual, ptn, pt_list);
1260 		}
1261 		timespecclear(&ts);
1262 		for (ptn = LIST_FIRST(&pts->pts_prof);
1263 		     ptn && ptn != pts->pts_timers[ITIMER_PROF];
1264 		     ptn = LIST_NEXT(ptn, pt_list)) {
1265 			KASSERT(ptn->pt_type == CLOCK_PROF);
1266 			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
1267 		}
1268 		LIST_FIRST(&pts->pts_prof) = NULL;
1269 		if (ptn) {
1270 			KASSERT(ptn->pt_type == CLOCK_PROF);
1271 			timespecadd(&ts, &ptn->pt_time.it_value,
1272 			    &ptn->pt_time.it_value);
1273 			LIST_INSERT_HEAD(&pts->pts_prof, ptn, pt_list);
1274 		}
1275 		i = 3;
1276 	}
1277 	for ( ; i < TIMER_MAX; i++) {
1278 		if (pts->pts_timers[i] != NULL) {
1279 			itimerfree(pts, i);
1280 			mutex_spin_enter(&timer_lock);
1281 		}
1282 	}
1283 	if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL &&
1284 	    pts->pts_timers[2] == NULL) {
1285 		p->p_timers = NULL;
1286 		mutex_spin_exit(&timer_lock);
1287 		pool_put(&ptimers_pool, pts);
1288 	} else
1289 		mutex_spin_exit(&timer_lock);
1290 }
1291 
1292 static void
1293 itimerfree(struct ptimers *pts, int index)
1294 {
1295 	struct ptimer *pt;
1296 
1297 	KASSERT(mutex_owned(&timer_lock));
1298 
1299 	pt = pts->pts_timers[index];
1300 	pts->pts_timers[index] = NULL;
1301 	if (!CLOCK_VIRTUAL_P(pt->pt_type))
1302 		callout_halt(&pt->pt_ch, &timer_lock);
1303 	if (pt->pt_queued)
1304 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
1305 	mutex_spin_exit(&timer_lock);
1306 	if (!CLOCK_VIRTUAL_P(pt->pt_type))
1307 		callout_destroy(&pt->pt_ch);
1308 	pool_put(&ptimer_pool, pt);
1309 }
1310 
1311 /*
1312  * Decrement an interval timer by a specified number
1313  * of nanoseconds, which must be less than a second,
1314  * i.e. < 1000000000.  If the timer expires, then reload
1315  * it.  In this case, carry over (nsec - old value) to
1316  * reduce the value reloaded into the timer so that
1317  * the timer does not drift.  This routine assumes
1318  * that it is called in a context where the timers
1319  * on which it is operating cannot change in value.
1320  */
1321 static int
1322 itimerdecr(struct ptimer *pt, int nsec)
1323 {
1324 	struct itimerspec *itp;
1325 
1326 	KASSERT(mutex_owned(&timer_lock));
1327 	KASSERT(CLOCK_VIRTUAL_P(pt->pt_type));
1328 
1329 	itp = &pt->pt_time;
1330 	if (itp->it_value.tv_nsec < nsec) {
1331 		if (itp->it_value.tv_sec == 0) {
1332 			/* expired, and already in next interval */
1333 			nsec -= itp->it_value.tv_nsec;
1334 			goto expire;
1335 		}
1336 		itp->it_value.tv_nsec += 1000000000;
1337 		itp->it_value.tv_sec--;
1338 	}
1339 	itp->it_value.tv_nsec -= nsec;
1340 	nsec = 0;
1341 	if (timespecisset(&itp->it_value))
1342 		return (1);
1343 	/* expired, exactly at end of interval */
1344 expire:
1345 	if (timespecisset(&itp->it_interval)) {
1346 		itp->it_value = itp->it_interval;
1347 		itp->it_value.tv_nsec -= nsec;
1348 		if (itp->it_value.tv_nsec < 0) {
1349 			itp->it_value.tv_nsec += 1000000000;
1350 			itp->it_value.tv_sec--;
1351 		}
1352 		timer_settime(pt);
1353 	} else
1354 		itp->it_value.tv_nsec = 0;		/* sec is already 0 */
1355 	return (0);
1356 }
1357 
1358 static void
1359 itimerfire(struct ptimer *pt)
1360 {
1361 
1362 	KASSERT(mutex_owned(&timer_lock));
1363 
1364 	/*
1365 	 * XXX Can overrun, but we don't do signal queueing yet, anyway.
1366 	 * XXX Relying on the clock interrupt is stupid.
1367 	 */
1368 	if ((pt->pt_ev.sigev_notify == SIGEV_SA && pt->pt_proc->p_sa == NULL) ||
1369 	    (pt->pt_ev.sigev_notify != SIGEV_SIGNAL &&
1370 	    pt->pt_ev.sigev_notify != SIGEV_SA) || pt->pt_queued)
1371 		return;
1372 	TAILQ_INSERT_TAIL(&timer_queue, pt, pt_chain);
1373 	pt->pt_queued = true;
1374 	softint_schedule(timer_sih);
1375 }
1376 
1377 void
1378 timer_tick(lwp_t *l, bool user)
1379 {
1380 	struct ptimers *pts;
1381 	struct ptimer *pt;
1382 	proc_t *p;
1383 
1384 	p = l->l_proc;
1385 	if (p->p_timers == NULL)
1386 		return;
1387 
1388 	mutex_spin_enter(&timer_lock);
1389 	if ((pts = l->l_proc->p_timers) != NULL) {
1390 		/*
1391 		 * Run current process's virtual and profile time, as needed.
1392 		 */
1393 		if (user && (pt = LIST_FIRST(&pts->pts_virtual)) != NULL)
1394 			if (itimerdecr(pt, tick * 1000) == 0)
1395 				itimerfire(pt);
1396 		if ((pt = LIST_FIRST(&pts->pts_prof)) != NULL)
1397 			if (itimerdecr(pt, tick * 1000) == 0)
1398 				itimerfire(pt);
1399 	}
1400 	mutex_spin_exit(&timer_lock);
1401 }
1402 
1403 #ifdef KERN_SA
1404 /*
1405  * timer_sa_intr:
1406  *
1407  *	SIGEV_SA handling for timer_intr(). We are called (and return)
1408  * with the timer lock held. We know that the process had SA enabled
1409  * when this timer was enqueued. As timer_intr() is a soft interrupt
1410  * handler, SA should still be enabled by the time we get here.
1411  */
1412 static void
1413 timer_sa_intr(struct ptimer *pt, proc_t *p)
1414 {
1415 	unsigned int		i;
1416 	struct sadata		*sa;
1417 	struct sadata_vp	*vp;
1418 
1419 	/* Cause the process to generate an upcall when it returns. */
1420 	if (!p->p_timerpend) {
1421 		/*
1422 		 * XXX stop signals can be processed inside tsleep,
1423 		 * which can be inside sa_yield's inner loop, which
1424 		 * makes testing for sa_idle alone insuffucent to
1425 		 * determine if we really should call setrunnable.
1426 		 */
1427 		pt->pt_poverruns = pt->pt_overruns;
1428 		pt->pt_overruns = 0;
1429 		i = 1 << pt->pt_entry;
1430 		p->p_timers->pts_fired = i;
1431 		p->p_timerpend = 1;
1432 
1433 		sa = p->p_sa;
1434 		mutex_enter(&sa->sa_mutex);
1435 		SLIST_FOREACH(vp, &sa->sa_vps, savp_next) {
1436 			struct lwp *vp_lwp = vp->savp_lwp;
1437 			lwp_lock(vp_lwp);
1438 			lwp_need_userret(vp_lwp);
1439 			if (vp_lwp->l_flag & LW_SA_IDLE) {
1440 				vp_lwp->l_flag &= ~LW_SA_IDLE;
1441 				lwp_unsleep(vp_lwp, true);
1442 				break;
1443 			}
1444 			lwp_unlock(vp_lwp);
1445 		}
1446 		mutex_exit(&sa->sa_mutex);
1447 	} else {
1448 		i = 1 << pt->pt_entry;
1449 		if ((p->p_timers->pts_fired & i) == 0) {
1450 			pt->pt_poverruns = pt->pt_overruns;
1451 			pt->pt_overruns = 0;
1452 			p->p_timers->pts_fired |= i;
1453 		} else
1454 			pt->pt_overruns++;
1455 	}
1456 }
1457 #endif /* KERN_SA */
1458 
1459 static void
1460 timer_intr(void *cookie)
1461 {
1462 	ksiginfo_t ksi;
1463 	struct ptimer *pt;
1464 	proc_t *p;
1465 
1466 	mutex_enter(proc_lock);
1467 	mutex_spin_enter(&timer_lock);
1468 	while ((pt = TAILQ_FIRST(&timer_queue)) != NULL) {
1469 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
1470 		KASSERT(pt->pt_queued);
1471 		pt->pt_queued = false;
1472 
1473 		if (pt->pt_proc->p_timers == NULL) {
1474 			/* Process is dying. */
1475 			continue;
1476 		}
1477 		p = pt->pt_proc;
1478 #ifdef KERN_SA
1479 		if (pt->pt_ev.sigev_notify == SIGEV_SA) {
1480 			timer_sa_intr(pt, p);
1481 			continue;
1482 		}
1483 #endif /* KERN_SA */
1484 		if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL)
1485 			continue;
1486 		if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) {
1487 			pt->pt_overruns++;
1488 			continue;
1489 		}
1490 
1491 		KSI_INIT(&ksi);
1492 		ksi.ksi_signo = pt->pt_ev.sigev_signo;
1493 		ksi.ksi_code = SI_TIMER;
1494 		ksi.ksi_value = pt->pt_ev.sigev_value;
1495 		pt->pt_poverruns = pt->pt_overruns;
1496 		pt->pt_overruns = 0;
1497 		mutex_spin_exit(&timer_lock);
1498 		kpsignal(p, &ksi, NULL);
1499 		mutex_spin_enter(&timer_lock);
1500 	}
1501 	mutex_spin_exit(&timer_lock);
1502 	mutex_exit(proc_lock);
1503 }
1504 
1505 /*
1506  * Check if the time will wrap if set to ts.
1507  *
1508  * ts - timespec describing the new time
1509  * delta - the delta between the current time and ts
1510  */
1511 bool
1512 time_wraps(struct timespec *ts, struct timespec *delta)
1513 {
1514 
1515 	/*
1516 	 * Don't allow the time to be set forward so far it
1517 	 * will wrap and become negative, thus allowing an
1518 	 * attacker to bypass the next check below.  The
1519 	 * cutoff is 1 year before rollover occurs, so even
1520 	 * if the attacker uses adjtime(2) to move the time
1521 	 * past the cutoff, it will take a very long time
1522 	 * to get to the wrap point.
1523 	 */
1524 	if ((ts->tv_sec > LLONG_MAX - 365*24*60*60) ||
1525 	    (delta->tv_sec < 0 || delta->tv_nsec < 0))
1526 		return true;
1527 
1528 	return false;
1529 }
1530