xref: /netbsd-src/sys/kern/kern_time.c (revision 8feb0f0b7eaff0608f8350bbfa3098827b4bb91b)
1 /*	$NetBSD: kern_time.c,v 1.219 2023/02/18 14:04:17 thorpej Exp $	*/
2 
3 /*-
4  * Copyright (c) 2000, 2004, 2005, 2007, 2008, 2009, 2020
5  *     The NetBSD Foundation, Inc.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to The NetBSD Foundation
9  * by Christopher G. Demetriou, by Andrew Doran, and by Jason R. Thorpe.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)kern_time.c	8.4 (Berkeley) 5/26/95
62  */
63 
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.219 2023/02/18 14:04:17 thorpej Exp $");
66 
67 #include <sys/param.h>
68 #include <sys/resourcevar.h>
69 #include <sys/kernel.h>
70 #include <sys/systm.h>
71 #include <sys/proc.h>
72 #include <sys/vnode.h>
73 #include <sys/signalvar.h>
74 #include <sys/syslog.h>
75 #include <sys/timetc.h>
76 #include <sys/timevar.h>
77 #include <sys/timex.h>
78 #include <sys/kauth.h>
79 #include <sys/mount.h>
80 #include <sys/syscallargs.h>
81 #include <sys/cpu.h>
82 
83 kmutex_t	itimer_mutex __cacheline_aligned;	/* XXX static */
84 static struct itlist itimer_realtime_changed_notify;
85 
86 static void	ptimer_intr(void *);
87 static void	*ptimer_sih __read_mostly;
88 static TAILQ_HEAD(, ptimer) ptimer_queue;
89 
90 #define	CLOCK_VIRTUAL_P(clockid)	\
91 	((clockid) == CLOCK_VIRTUAL || (clockid) == CLOCK_PROF)
92 
93 CTASSERT(ITIMER_REAL == CLOCK_REALTIME);
94 CTASSERT(ITIMER_VIRTUAL == CLOCK_VIRTUAL);
95 CTASSERT(ITIMER_PROF == CLOCK_PROF);
96 CTASSERT(ITIMER_MONOTONIC == CLOCK_MONOTONIC);
97 
98 #define	DELAYTIMER_MAX	32
99 
100 /*
101  * Initialize timekeeping.
102  */
103 void
104 time_init(void)
105 {
106 
107 	mutex_init(&itimer_mutex, MUTEX_DEFAULT, IPL_SCHED);
108 	LIST_INIT(&itimer_realtime_changed_notify);
109 
110 	TAILQ_INIT(&ptimer_queue);
111 	ptimer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
112 	    ptimer_intr, NULL);
113 }
114 
115 /*
116  * Check if the time will wrap if set to ts.
117  *
118  * ts - timespec describing the new time
119  * delta - the delta between the current time and ts
120  */
121 bool
122 time_wraps(struct timespec *ts, struct timespec *delta)
123 {
124 
125 	/*
126 	 * Don't allow the time to be set forward so far it
127 	 * will wrap and become negative, thus allowing an
128 	 * attacker to bypass the next check below.  The
129 	 * cutoff is 1 year before rollover occurs, so even
130 	 * if the attacker uses adjtime(2) to move the time
131 	 * past the cutoff, it will take a very long time
132 	 * to get to the wrap point.
133 	 */
134 	if ((ts->tv_sec > LLONG_MAX - 365*24*60*60) ||
135 	    (delta->tv_sec < 0 || delta->tv_nsec < 0))
136 		return true;
137 
138 	return false;
139 }
140 
141 /*
142  * itimer_lock:
143  *
144  *	Acquire the interval timer data lock.
145  */
146 void
147 itimer_lock(void)
148 {
149 	mutex_spin_enter(&itimer_mutex);
150 }
151 
152 /*
153  * itimer_unlock:
154  *
155  *	Release the interval timer data lock.
156  */
157 void
158 itimer_unlock(void)
159 {
160 	mutex_spin_exit(&itimer_mutex);
161 }
162 
163 /*
164  * itimer_lock_held:
165  *
166  *	Check that the interval timer lock is held for diagnostic
167  *	assertions.
168  */
169 inline bool __diagused
170 itimer_lock_held(void)
171 {
172 	return mutex_owned(&itimer_mutex);
173 }
174 
175 /*
176  * Time of day and interval timer support.
177  *
178  * These routines provide the kernel entry points to get and set
179  * the time-of-day and per-process interval timers.  Subroutines
180  * here provide support for adding and subtracting timeval structures
181  * and decrementing interval timers, optionally reloading the interval
182  * timers when they expire.
183  */
184 
185 /* This function is used by clock_settime and settimeofday */
186 static int
187 settime1(struct proc *p, const struct timespec *ts, bool check_kauth)
188 {
189 	struct timespec delta, now;
190 
191 	/*
192 	 * The time being set to an unreasonable value will cause
193 	 * unreasonable system behaviour.
194 	 */
195 	if (ts->tv_sec < 0 || ts->tv_sec > (1LL << 36))
196 		return EINVAL;
197 
198 	nanotime(&now);
199 	timespecsub(ts, &now, &delta);
200 
201 	if (check_kauth && kauth_authorize_system(kauth_cred_get(),
202 	    KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, __UNCONST(ts),
203 	    &delta, KAUTH_ARG(check_kauth ? false : true)) != 0) {
204 		return EPERM;
205 	}
206 
207 #ifdef notyet
208 	if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */
209 		return EPERM;
210 	}
211 #endif
212 
213 	tc_setclock(ts);
214 
215 	resettodr();
216 
217 	/*
218 	 * Notify pending CLOCK_REALTIME timers about the real time change.
219 	 * There may be inactive timers on this list, but this happens
220 	 * comparatively less often than timers firing, and so it's better
221 	 * to put the extra checks here than to complicate the other code
222 	 * path.
223 	 */
224 	struct itimer *it;
225 	itimer_lock();
226 	LIST_FOREACH(it, &itimer_realtime_changed_notify, it_rtchgq) {
227 		KASSERT(it->it_ops->ito_realtime_changed != NULL);
228 		if (timespecisset(&it->it_time.it_value)) {
229 			(*it->it_ops->ito_realtime_changed)(it);
230 		}
231 	}
232 	itimer_unlock();
233 
234 	return 0;
235 }
236 
237 int
238 settime(struct proc *p, struct timespec *ts)
239 {
240 	return settime1(p, ts, true);
241 }
242 
243 /* ARGSUSED */
244 int
245 sys___clock_gettime50(struct lwp *l,
246     const struct sys___clock_gettime50_args *uap, register_t *retval)
247 {
248 	/* {
249 		syscallarg(clockid_t) clock_id;
250 		syscallarg(struct timespec *) tp;
251 	} */
252 	int error;
253 	struct timespec ats;
254 
255 	error = clock_gettime1(SCARG(uap, clock_id), &ats);
256 	if (error != 0)
257 		return error;
258 
259 	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
260 }
261 
262 /* ARGSUSED */
263 int
264 sys___clock_settime50(struct lwp *l,
265     const struct sys___clock_settime50_args *uap, register_t *retval)
266 {
267 	/* {
268 		syscallarg(clockid_t) clock_id;
269 		syscallarg(const struct timespec *) tp;
270 	} */
271 	int error;
272 	struct timespec ats;
273 
274 	if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
275 		return error;
276 
277 	return clock_settime1(l->l_proc, SCARG(uap, clock_id), &ats, true);
278 }
279 
280 
281 int
282 clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp,
283     bool check_kauth)
284 {
285 	int error;
286 
287 	if (tp->tv_nsec < 0 || tp->tv_nsec >= 1000000000L)
288 		return EINVAL;
289 
290 	switch (clock_id) {
291 	case CLOCK_REALTIME:
292 		if ((error = settime1(p, tp, check_kauth)) != 0)
293 			return error;
294 		break;
295 	case CLOCK_MONOTONIC:
296 		return EINVAL;	/* read-only clock */
297 	default:
298 		return EINVAL;
299 	}
300 
301 	return 0;
302 }
303 
304 int
305 sys___clock_getres50(struct lwp *l, const struct sys___clock_getres50_args *uap,
306     register_t *retval)
307 {
308 	/* {
309 		syscallarg(clockid_t) clock_id;
310 		syscallarg(struct timespec *) tp;
311 	} */
312 	struct timespec ts;
313 	int error;
314 
315 	if ((error = clock_getres1(SCARG(uap, clock_id), &ts)) != 0)
316 		return error;
317 
318 	if (SCARG(uap, tp))
319 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
320 
321 	return error;
322 }
323 
324 int
325 clock_getres1(clockid_t clock_id, struct timespec *ts)
326 {
327 
328 	switch (clock_id) {
329 	case CLOCK_REALTIME:
330 	case CLOCK_MONOTONIC:
331 		ts->tv_sec = 0;
332 		if (tc_getfrequency() > 1000000000)
333 			ts->tv_nsec = 1;
334 		else
335 			ts->tv_nsec = 1000000000 / tc_getfrequency();
336 		break;
337 	default:
338 		return EINVAL;
339 	}
340 
341 	return 0;
342 }
343 
344 /* ARGSUSED */
345 int
346 sys___nanosleep50(struct lwp *l, const struct sys___nanosleep50_args *uap,
347     register_t *retval)
348 {
349 	/* {
350 		syscallarg(struct timespec *) rqtp;
351 		syscallarg(struct timespec *) rmtp;
352 	} */
353 	struct timespec rmt, rqt;
354 	int error, error1;
355 
356 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
357 	if (error)
358 		return error;
359 
360 	error = nanosleep1(l, CLOCK_MONOTONIC, 0, &rqt,
361 	    SCARG(uap, rmtp) ? &rmt : NULL);
362 	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
363 		return error;
364 
365 	error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt));
366 	return error1 ? error1 : error;
367 }
368 
369 /* ARGSUSED */
370 int
371 sys_clock_nanosleep(struct lwp *l, const struct sys_clock_nanosleep_args *uap,
372     register_t *retval)
373 {
374 	/* {
375 		syscallarg(clockid_t) clock_id;
376 		syscallarg(int) flags;
377 		syscallarg(struct timespec *) rqtp;
378 		syscallarg(struct timespec *) rmtp;
379 	} */
380 	struct timespec rmt, rqt;
381 	int error, error1;
382 
383 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
384 	if (error)
385 		goto out;
386 
387 	error = nanosleep1(l, SCARG(uap, clock_id), SCARG(uap, flags), &rqt,
388 	    SCARG(uap, rmtp) ? &rmt : NULL);
389 	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
390 		goto out;
391 
392 	if ((SCARG(uap, flags) & TIMER_ABSTIME) == 0 &&
393 	    (error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt))) != 0)
394 		error = error1;
395 out:
396 	*retval = error;
397 	return 0;
398 }
399 
400 int
401 nanosleep1(struct lwp *l, clockid_t clock_id, int flags, struct timespec *rqt,
402     struct timespec *rmt)
403 {
404 	struct timespec rmtstart;
405 	int error, timo;
406 
407 	if ((error = ts2timo(clock_id, flags, rqt, &timo, &rmtstart)) != 0) {
408 		if (error == ETIMEDOUT) {
409 			error = 0;
410 			if (rmt != NULL)
411 				rmt->tv_sec = rmt->tv_nsec = 0;
412 		}
413 		return error;
414 	}
415 
416 	/*
417 	 * Avoid inadvertently sleeping forever
418 	 */
419 	if (timo == 0)
420 		timo = 1;
421 again:
422 	error = kpause("nanoslp", true, timo, NULL);
423 	if (error == EWOULDBLOCK)
424 		error = 0;
425 	if (rmt != NULL || error == 0) {
426 		struct timespec rmtend;
427 		struct timespec t0;
428 		struct timespec *t;
429 		int err;
430 
431 		err = clock_gettime1(clock_id, &rmtend);
432 		if (err != 0)
433 			return err;
434 
435 		t = (rmt != NULL) ? rmt : &t0;
436 		if (flags & TIMER_ABSTIME) {
437 			timespecsub(rqt, &rmtend, t);
438 		} else {
439 			if (timespeccmp(&rmtend, &rmtstart, <))
440 				timespecclear(t); /* clock wound back */
441 			else
442 				timespecsub(&rmtend, &rmtstart, t);
443 			if (timespeccmp(rqt, t, <))
444 				timespecclear(t);
445 			else
446 				timespecsub(rqt, t, t);
447 		}
448 		if (t->tv_sec < 0)
449 			timespecclear(t);
450 		if (error == 0) {
451 			timo = tstohz(t);
452 			if (timo > 0)
453 				goto again;
454 		}
455 	}
456 
457 	if (error == ERESTART)
458 		error = EINTR;
459 
460 	return error;
461 }
462 
463 int
464 sys_clock_getcpuclockid2(struct lwp *l,
465     const struct sys_clock_getcpuclockid2_args *uap,
466     register_t *retval)
467 {
468 	/* {
469 		syscallarg(idtype_t idtype;
470 		syscallarg(id_t id);
471 		syscallarg(clockid_t *)clock_id;
472 	} */
473 	pid_t pid;
474 	lwpid_t lid;
475 	clockid_t clock_id;
476 	id_t id = SCARG(uap, id);
477 
478 	switch (SCARG(uap, idtype)) {
479 	case P_PID:
480 		pid = id == 0 ? l->l_proc->p_pid : id;
481 		clock_id = CLOCK_PROCESS_CPUTIME_ID | pid;
482 		break;
483 	case P_LWPID:
484 		lid = id == 0 ? l->l_lid : id;
485 		clock_id = CLOCK_THREAD_CPUTIME_ID | lid;
486 		break;
487 	default:
488 		return EINVAL;
489 	}
490 	return copyout(&clock_id, SCARG(uap, clock_id), sizeof(clock_id));
491 }
492 
493 /* ARGSUSED */
494 int
495 sys___gettimeofday50(struct lwp *l, const struct sys___gettimeofday50_args *uap,
496     register_t *retval)
497 {
498 	/* {
499 		syscallarg(struct timeval *) tp;
500 		syscallarg(void *) tzp;		really "struct timezone *";
501 	} */
502 	struct timeval atv;
503 	int error = 0;
504 	struct timezone tzfake;
505 
506 	if (SCARG(uap, tp)) {
507 		memset(&atv, 0, sizeof(atv));
508 		microtime(&atv);
509 		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
510 		if (error)
511 			return error;
512 	}
513 	if (SCARG(uap, tzp)) {
514 		/*
515 		 * NetBSD has no kernel notion of time zone, so we just
516 		 * fake up a timezone struct and return it if demanded.
517 		 */
518 		tzfake.tz_minuteswest = 0;
519 		tzfake.tz_dsttime = 0;
520 		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
521 	}
522 	return error;
523 }
524 
525 /* ARGSUSED */
526 int
527 sys___settimeofday50(struct lwp *l, const struct sys___settimeofday50_args *uap,
528     register_t *retval)
529 {
530 	/* {
531 		syscallarg(const struct timeval *) tv;
532 		syscallarg(const void *) tzp; really "const struct timezone *";
533 	} */
534 
535 	return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true);
536 }
537 
538 int
539 settimeofday1(const struct timeval *utv, bool userspace,
540     const void *utzp, struct lwp *l, bool check_kauth)
541 {
542 	struct timeval atv;
543 	struct timespec ts;
544 	int error;
545 
546 	/* Verify all parameters before changing time. */
547 
548 	/*
549 	 * NetBSD has no kernel notion of time zone, and only an
550 	 * obsolete program would try to set it, so we log a warning.
551 	 */
552 	if (utzp)
553 		log(LOG_WARNING, "pid %d attempted to set the "
554 		    "(obsolete) kernel time zone\n", l->l_proc->p_pid);
555 
556 	if (utv == NULL)
557 		return 0;
558 
559 	if (userspace) {
560 		if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
561 			return error;
562 		utv = &atv;
563 	}
564 
565 	if (utv->tv_usec < 0 || utv->tv_usec >= 1000000)
566 		return EINVAL;
567 
568 	TIMEVAL_TO_TIMESPEC(utv, &ts);
569 	return settime1(l->l_proc, &ts, check_kauth);
570 }
571 
572 int	time_adjusted;			/* set if an adjustment is made */
573 
574 /* ARGSUSED */
575 int
576 sys___adjtime50(struct lwp *l, const struct sys___adjtime50_args *uap,
577     register_t *retval)
578 {
579 	/* {
580 		syscallarg(const struct timeval *) delta;
581 		syscallarg(struct timeval *) olddelta;
582 	} */
583 	int error;
584 	struct timeval atv, oldatv;
585 
586 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
587 	    KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0)
588 		return error;
589 
590 	if (SCARG(uap, delta)) {
591 		error = copyin(SCARG(uap, delta), &atv,
592 		    sizeof(*SCARG(uap, delta)));
593 		if (error)
594 			return error;
595 	}
596 	adjtime1(SCARG(uap, delta) ? &atv : NULL,
597 	    SCARG(uap, olddelta) ? &oldatv : NULL, l->l_proc);
598 	if (SCARG(uap, olddelta))
599 		error = copyout(&oldatv, SCARG(uap, olddelta),
600 		    sizeof(*SCARG(uap, olddelta)));
601 	return error;
602 }
603 
604 void
605 adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p)
606 {
607 
608 	if (olddelta) {
609 		memset(olddelta, 0, sizeof(*olddelta));
610 		mutex_spin_enter(&timecounter_lock);
611 		olddelta->tv_sec = time_adjtime / 1000000;
612 		olddelta->tv_usec = time_adjtime % 1000000;
613 		if (olddelta->tv_usec < 0) {
614 			olddelta->tv_usec += 1000000;
615 			olddelta->tv_sec--;
616 		}
617 		mutex_spin_exit(&timecounter_lock);
618 	}
619 
620 	if (delta) {
621 		mutex_spin_enter(&timecounter_lock);
622 		/*
623 		 * XXX This should maybe just report failure to
624 		 * userland for nonsense deltas.
625 		 */
626 		if (delta->tv_sec > INT64_MAX/1000000 - 1) {
627 			time_adjtime = INT64_MAX;
628 		} else if (delta->tv_sec < INT64_MIN/1000000 + 1) {
629 			time_adjtime = INT64_MIN;
630 		} else {
631 			time_adjtime = delta->tv_sec * 1000000
632 			    + MAX(-999999, MIN(999999, delta->tv_usec));
633 		}
634 
635 		if (time_adjtime) {
636 			/* We need to save the system time during shutdown */
637 			time_adjusted |= 1;
638 		}
639 		mutex_spin_exit(&timecounter_lock);
640 	}
641 }
642 
643 /*
644  * Interval timer support.
645  *
646  * The itimer_*() routines provide generic support for interval timers,
647  * both real (CLOCK_REALTIME, CLOCK_MONOTIME), and virtual (CLOCK_VIRTUAL,
648  * CLOCK_PROF).
649  *
650  * Real timers keep their deadline as an absolute time, and are fired
651  * by a callout.  Virtual timers are kept as a linked-list of deltas,
652  * and are processed by hardclock().
653  *
654  * Because the real time timer callout may be delayed in real time due
655  * to interrupt processing on the system, it is possible for the real
656  * time timeout routine (itimer_callout()) run past after its deadline.
657  * It does not suffice, therefore, to reload the real timer .it_value
658  * from the timer's .it_interval.  Rather, we compute the next deadline
659  * in absolute time based on the current time and the .it_interval value,
660  * and report any overruns.
661  *
662  * Note that while the virtual timers are supported in a generic fashion
663  * here, they only (currently) make sense as per-process timers, and thus
664  * only really work for that case.
665  */
666 
667 /*
668  * itimer_init:
669  *
670  *	Initialize the common data for an interval timer.
671  */
672 void
673 itimer_init(struct itimer * const it, const struct itimer_ops * const ops,
674     clockid_t const id, struct itlist * const itl)
675 {
676 
677 	KASSERT(itimer_lock_held());
678 	KASSERT(ops != NULL);
679 
680 	timespecclear(&it->it_time.it_value);
681 	it->it_ops = ops;
682 	it->it_clockid = id;
683 	it->it_overruns = 0;
684 	it->it_dying = false;
685 	if (!CLOCK_VIRTUAL_P(id)) {
686 		KASSERT(itl == NULL);
687 		callout_init(&it->it_ch, CALLOUT_MPSAFE);
688 		if (id == CLOCK_REALTIME && ops->ito_realtime_changed != NULL) {
689 			LIST_INSERT_HEAD(&itimer_realtime_changed_notify,
690 			    it, it_rtchgq);
691 		}
692 	} else {
693 		KASSERT(itl != NULL);
694 		it->it_vlist = itl;
695 		it->it_active = false;
696 	}
697 }
698 
699 /*
700  * itimer_poison:
701  *
702  *	Poison an interval timer, preventing it from being scheduled
703  *	or processed, in preparation for freeing the timer.
704  */
705 void
706 itimer_poison(struct itimer * const it)
707 {
708 
709 	KASSERT(itimer_lock_held());
710 
711 	it->it_dying = true;
712 
713 	/*
714 	 * For non-virtual timers, stop the callout, or wait for it to
715 	 * run if it has already fired.  It cannot restart again after
716 	 * this point: the callout won't restart itself when dying, no
717 	 * other users holding the lock can restart it, and any other
718 	 * users waiting for callout_halt concurrently (itimer_settime)
719 	 * will restart from the top.
720 	 */
721 	if (!CLOCK_VIRTUAL_P(it->it_clockid)) {
722 		callout_halt(&it->it_ch, &itimer_mutex);
723 		if (it->it_clockid == CLOCK_REALTIME &&
724 		    it->it_ops->ito_realtime_changed != NULL) {
725 			LIST_REMOVE(it, it_rtchgq);
726 		}
727 	}
728 }
729 
730 /*
731  * itimer_fini:
732  *
733  *	Release resources used by an interval timer.
734  *
735  *	N.B. itimer_lock must be held on entry, and is released on exit.
736  */
737 void
738 itimer_fini(struct itimer * const it)
739 {
740 
741 	KASSERT(itimer_lock_held());
742 
743 	/* All done with the global state. */
744 	itimer_unlock();
745 
746 	/* Destroy the callout, if needed. */
747 	if (!CLOCK_VIRTUAL_P(it->it_clockid))
748 		callout_destroy(&it->it_ch);
749 }
750 
751 /*
752  * itimer_decr:
753  *
754  *	Decrement an interval timer by a specified number of nanoseconds,
755  *	which must be less than a second, i.e. < 1000000000.  If the timer
756  *	expires, then reload it.  In this case, carry over (nsec - old value)
757  *	to reduce the value reloaded into the timer so that the timer does
758  *	not drift.  This routine assumes that it is called in a context where
759  *	the timers on which it is operating cannot change in value.
760  *
761  *	Returns true if the timer has expired.
762  */
763 static bool
764 itimer_decr(struct itimer *it, int nsec)
765 {
766 	struct itimerspec *itp;
767 	int error __diagused;
768 
769 	KASSERT(itimer_lock_held());
770 	KASSERT(CLOCK_VIRTUAL_P(it->it_clockid));
771 
772 	itp = &it->it_time;
773 	if (itp->it_value.tv_nsec < nsec) {
774 		if (itp->it_value.tv_sec == 0) {
775 			/* expired, and already in next interval */
776 			nsec -= itp->it_value.tv_nsec;
777 			goto expire;
778 		}
779 		itp->it_value.tv_nsec += 1000000000;
780 		itp->it_value.tv_sec--;
781 	}
782 	itp->it_value.tv_nsec -= nsec;
783 	nsec = 0;
784 	if (timespecisset(&itp->it_value))
785 		return false;
786 	/* expired, exactly at end of interval */
787  expire:
788 	if (timespecisset(&itp->it_interval)) {
789 		itp->it_value = itp->it_interval;
790 		itp->it_value.tv_nsec -= nsec;
791 		if (itp->it_value.tv_nsec < 0) {
792 			itp->it_value.tv_nsec += 1000000000;
793 			itp->it_value.tv_sec--;
794 		}
795 		error = itimer_settime(it);
796 		KASSERT(error == 0); /* virtual, never fails */
797 	} else
798 		itp->it_value.tv_nsec = 0;		/* sec is already 0 */
799 	return true;
800 }
801 
802 static void itimer_callout(void *);
803 
804 /*
805  * itimer_arm_real:
806  *
807  *	Arm a non-virtual timer.
808  */
809 static void
810 itimer_arm_real(struct itimer * const it)
811 {
812 	KASSERT(!it->it_dying);
813 
814 	/*
815 	 * Don't need to check tshzto() return value, here.
816 	 * callout_reset() does it for us.
817 	 */
818 	callout_reset(&it->it_ch,
819 	    (it->it_clockid == CLOCK_MONOTONIC
820 		? tshztoup(&it->it_time.it_value)
821 		: tshzto(&it->it_time.it_value)),
822 	    itimer_callout, it);
823 }
824 
825 /*
826  * itimer_callout:
827  *
828  *	Callout to expire a non-virtual timer.  Queue it up for processing,
829  *	and then reload, if it is configured to do so.
830  *
831  *	N.B. A delay in processing this callout causes multiple
832  *	SIGALRM calls to be compressed into one.
833  */
834 static void
835 itimer_callout(void *arg)
836 {
837 	uint64_t last_val, next_val, interval, now_ns;
838 	struct timespec now, next;
839 	struct itimer * const it = arg;
840 	int backwards;
841 
842 	itimer_lock();
843 	(*it->it_ops->ito_fire)(it);
844 
845 	if (!timespecisset(&it->it_time.it_interval)) {
846 		timespecclear(&it->it_time.it_value);
847 		itimer_unlock();
848 		return;
849 	}
850 
851 	if (it->it_clockid == CLOCK_MONOTONIC) {
852 		getnanouptime(&now);
853 	} else {
854 		getnanotime(&now);
855 	}
856 
857 	backwards = (timespeccmp(&it->it_time.it_value, &now, >));
858 
859 	/* Nonnegative interval guaranteed by itimerfix.  */
860 	KASSERT(it->it_time.it_interval.tv_sec >= 0);
861 	KASSERT(it->it_time.it_interval.tv_nsec >= 0);
862 
863 	/* Handle the easy case of non-overflown timers first. */
864 	if (!backwards &&
865 	    timespecaddok(&it->it_time.it_value, &it->it_time.it_interval)) {
866 		timespecadd(&it->it_time.it_value, &it->it_time.it_interval,
867 		    &next);
868 		it->it_time.it_value = next;
869 	} else {
870 		now_ns = timespec2ns(&now);
871 		last_val = timespec2ns(&it->it_time.it_value);
872 		interval = timespec2ns(&it->it_time.it_interval);
873 
874 		next_val = now_ns +
875 		    (now_ns - last_val + interval - 1) % interval;
876 
877 		if (backwards)
878 			next_val += interval;
879 		else
880 			it->it_overruns += (now_ns - last_val) / interval;
881 
882 		it->it_time.it_value.tv_sec = next_val / 1000000000;
883 		it->it_time.it_value.tv_nsec = next_val % 1000000000;
884 	}
885 
886 	/*
887 	 * Reset the callout, if it's not going away.
888 	 */
889 	if (!it->it_dying)
890 		itimer_arm_real(it);
891 	itimer_unlock();
892 }
893 
894 /*
895  * itimer_settime:
896  *
897  *	Set up the given interval timer. The value in it->it_time.it_value
898  *	is taken to be an absolute time for CLOCK_REALTIME/CLOCK_MONOTONIC
899  *	timers and a relative time for CLOCK_VIRTUAL/CLOCK_PROF timers.
900  *
901  *	If the callout had already fired but not yet run, fails with
902  *	ERESTART -- caller must restart from the top to look up a timer.
903  */
904 int
905 itimer_settime(struct itimer *it)
906 {
907 	struct itimer *itn, *pitn;
908 	struct itlist *itl;
909 
910 	KASSERT(itimer_lock_held());
911 
912 	if (!CLOCK_VIRTUAL_P(it->it_clockid)) {
913 		/*
914 		 * Try to stop the callout.  However, if it had already
915 		 * fired, we have to drop the lock to wait for it, so
916 		 * the world may have changed and pt may not be there
917 		 * any more.  In that case, tell the caller to start
918 		 * over from the top.
919 		 */
920 		if (callout_halt(&it->it_ch, &itimer_mutex))
921 			return ERESTART;
922 
923 		/* Now we can touch it and start it up again. */
924 		if (timespecisset(&it->it_time.it_value))
925 			itimer_arm_real(it);
926 	} else {
927 		if (it->it_active) {
928 			itn = LIST_NEXT(it, it_list);
929 			LIST_REMOVE(it, it_list);
930 			for ( ; itn; itn = LIST_NEXT(itn, it_list))
931 				timespecadd(&it->it_time.it_value,
932 				    &itn->it_time.it_value,
933 				    &itn->it_time.it_value);
934 		}
935 		if (timespecisset(&it->it_time.it_value)) {
936 			itl = it->it_vlist;
937 			for (itn = LIST_FIRST(itl), pitn = NULL;
938 			     itn && timespeccmp(&it->it_time.it_value,
939 				 &itn->it_time.it_value, >);
940 			     pitn = itn, itn = LIST_NEXT(itn, it_list))
941 				timespecsub(&it->it_time.it_value,
942 				    &itn->it_time.it_value,
943 				    &it->it_time.it_value);
944 
945 			if (pitn)
946 				LIST_INSERT_AFTER(pitn, it, it_list);
947 			else
948 				LIST_INSERT_HEAD(itl, it, it_list);
949 
950 			for ( ; itn ; itn = LIST_NEXT(itn, it_list))
951 				timespecsub(&itn->it_time.it_value,
952 				    &it->it_time.it_value,
953 				    &itn->it_time.it_value);
954 
955 			it->it_active = true;
956 		} else {
957 			it->it_active = false;
958 		}
959 	}
960 
961 	/* Success!  */
962 	return 0;
963 }
964 
965 /*
966  * itimer_gettime:
967  *
968  *	Return the remaining time of an interval timer.
969  */
970 void
971 itimer_gettime(const struct itimer *it, struct itimerspec *aits)
972 {
973 	struct timespec now;
974 	struct itimer *itn;
975 
976 	KASSERT(itimer_lock_held());
977 
978 	*aits = it->it_time;
979 	if (!CLOCK_VIRTUAL_P(it->it_clockid)) {
980 		/*
981 		 * Convert from absolute to relative time in .it_value
982 		 * part of real time timer.  If time for real time
983 		 * timer has passed return 0, else return difference
984 		 * between current time and time for the timer to go
985 		 * off.
986 		 */
987 		if (timespecisset(&aits->it_value)) {
988 			if (it->it_clockid == CLOCK_REALTIME) {
989 				getnanotime(&now);
990 			} else { /* CLOCK_MONOTONIC */
991 				getnanouptime(&now);
992 			}
993 			if (timespeccmp(&aits->it_value, &now, <))
994 				timespecclear(&aits->it_value);
995 			else
996 				timespecsub(&aits->it_value, &now,
997 				    &aits->it_value);
998 		}
999 	} else if (it->it_active) {
1000 		for (itn = LIST_FIRST(it->it_vlist); itn && itn != it;
1001 		     itn = LIST_NEXT(itn, it_list))
1002 			timespecadd(&aits->it_value,
1003 			    &itn->it_time.it_value, &aits->it_value);
1004 		KASSERT(itn != NULL); /* it should be findable on the list */
1005 	} else
1006 		timespecclear(&aits->it_value);
1007 }
1008 
1009 /*
1010  * Per-process timer support.
1011  *
1012  * Both the BSD getitimer() family and the POSIX timer_*() family of
1013  * routines are supported.
1014  *
1015  * All timers are kept in an array pointed to by p_timers, which is
1016  * allocated on demand - many processes don't use timers at all. The
1017  * first four elements in this array are reserved for the BSD timers:
1018  * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, element
1019  * 2 is ITIMER_PROF, and element 3 is ITIMER_MONOTONIC. The rest may be
1020  * allocated by the timer_create() syscall.
1021  *
1022  * These timers are a "sub-class" of interval timer.
1023  */
1024 
1025 /*
1026  * ptimer_free:
1027  *
1028  *	Free the per-process timer at the specified index.
1029  */
1030 static void
1031 ptimer_free(struct ptimers *pts, int index)
1032 {
1033 	struct itimer *it;
1034 	struct ptimer *pt;
1035 
1036 	KASSERT(itimer_lock_held());
1037 
1038 	it = pts->pts_timers[index];
1039 	pt = container_of(it, struct ptimer, pt_itimer);
1040 	pts->pts_timers[index] = NULL;
1041 	itimer_poison(it);
1042 
1043 	/*
1044 	 * Remove it from the queue to be signalled.  Must be done
1045 	 * after itimer is poisoned, because we may have had to wait
1046 	 * for the callout to complete.
1047 	 */
1048 	if (pt->pt_queued) {
1049 		TAILQ_REMOVE(&ptimer_queue, pt, pt_chain);
1050 		pt->pt_queued = false;
1051 	}
1052 
1053 	itimer_fini(it);	/* releases itimer_lock */
1054 	kmem_free(pt, sizeof(*pt));
1055 }
1056 
1057 /*
1058  * ptimers_alloc:
1059  *
1060  *	Allocate a ptimers for the specified process.
1061  */
1062 static struct ptimers *
1063 ptimers_alloc(struct proc *p)
1064 {
1065 	struct ptimers *pts;
1066 	int i;
1067 
1068 	pts = kmem_alloc(sizeof(*pts), KM_SLEEP);
1069 	LIST_INIT(&pts->pts_virtual);
1070 	LIST_INIT(&pts->pts_prof);
1071 	for (i = 0; i < TIMER_MAX; i++)
1072 		pts->pts_timers[i] = NULL;
1073 	itimer_lock();
1074 	if (p->p_timers == NULL) {
1075 		p->p_timers = pts;
1076 		itimer_unlock();
1077 		return pts;
1078 	}
1079 	itimer_unlock();
1080 	kmem_free(pts, sizeof(*pts));
1081 	return p->p_timers;
1082 }
1083 
1084 /*
1085  * ptimers_free:
1086  *
1087  *	Clean up the per-process timers. If "which" is set to TIMERS_ALL,
1088  *	then clean up all timers and free all the data structures. If
1089  *	"which" is set to TIMERS_POSIX, only clean up the timers allocated
1090  *	by timer_create(), not the BSD setitimer() timers, and only free the
1091  *	structure if none of those remain.
1092  *
1093  *	This function is exported because it is needed in the exec and
1094  *	exit code paths.
1095  */
1096 void
1097 ptimers_free(struct proc *p, int which)
1098 {
1099 	struct ptimers *pts;
1100 	struct itimer *itn;
1101 	struct timespec ts;
1102 	int i;
1103 
1104 	if (p->p_timers == NULL)
1105 		return;
1106 
1107 	pts = p->p_timers;
1108 	itimer_lock();
1109 	if (which == TIMERS_ALL) {
1110 		p->p_timers = NULL;
1111 		i = 0;
1112 	} else {
1113 		timespecclear(&ts);
1114 		for (itn = LIST_FIRST(&pts->pts_virtual);
1115 		     itn && itn != pts->pts_timers[ITIMER_VIRTUAL];
1116 		     itn = LIST_NEXT(itn, it_list)) {
1117 			KASSERT(itn->it_clockid == CLOCK_VIRTUAL);
1118 			timespecadd(&ts, &itn->it_time.it_value, &ts);
1119 		}
1120 		LIST_FIRST(&pts->pts_virtual) = NULL;
1121 		if (itn) {
1122 			KASSERT(itn->it_clockid == CLOCK_VIRTUAL);
1123 			timespecadd(&ts, &itn->it_time.it_value,
1124 			    &itn->it_time.it_value);
1125 			LIST_INSERT_HEAD(&pts->pts_virtual, itn, it_list);
1126 		}
1127 		timespecclear(&ts);
1128 		for (itn = LIST_FIRST(&pts->pts_prof);
1129 		     itn && itn != pts->pts_timers[ITIMER_PROF];
1130 		     itn = LIST_NEXT(itn, it_list)) {
1131 			KASSERT(itn->it_clockid == CLOCK_PROF);
1132 			timespecadd(&ts, &itn->it_time.it_value, &ts);
1133 		}
1134 		LIST_FIRST(&pts->pts_prof) = NULL;
1135 		if (itn) {
1136 			KASSERT(itn->it_clockid == CLOCK_PROF);
1137 			timespecadd(&ts, &itn->it_time.it_value,
1138 			    &itn->it_time.it_value);
1139 			LIST_INSERT_HEAD(&pts->pts_prof, itn, it_list);
1140 		}
1141 		i = TIMER_MIN;
1142 	}
1143 	for ( ; i < TIMER_MAX; i++) {
1144 		if (pts->pts_timers[i] != NULL) {
1145 			/* Free the timer and release the lock.  */
1146 			ptimer_free(pts, i);
1147 			/* Reacquire the lock for the next one.  */
1148 			itimer_lock();
1149 		}
1150 	}
1151 	if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL &&
1152 	    pts->pts_timers[2] == NULL && pts->pts_timers[3] == NULL) {
1153 		p->p_timers = NULL;
1154 		itimer_unlock();
1155 		kmem_free(pts, sizeof(*pts));
1156 	} else
1157 		itimer_unlock();
1158 }
1159 
1160 /*
1161  * ptimer_fire:
1162  *
1163  *	Fire a per-process timer.
1164  */
1165 static void
1166 ptimer_fire(struct itimer *it)
1167 {
1168 	struct ptimer *pt = container_of(it, struct ptimer, pt_itimer);
1169 
1170 	KASSERT(itimer_lock_held());
1171 
1172 	/*
1173 	 * XXX Can overrun, but we don't do signal queueing yet, anyway.
1174 	 * XXX Relying on the clock interrupt is stupid.
1175 	 */
1176 	if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) {
1177 		return;
1178 	}
1179 
1180 	if (!pt->pt_queued) {
1181 		TAILQ_INSERT_TAIL(&ptimer_queue, pt, pt_chain);
1182 		pt->pt_queued = true;
1183 		softint_schedule(ptimer_sih);
1184 	}
1185 }
1186 
1187 /*
1188  * Operations vector for per-process timers (BSD and POSIX).
1189  */
1190 static const struct itimer_ops ptimer_itimer_ops = {
1191 	.ito_fire = ptimer_fire,
1192 };
1193 
1194 /*
1195  * sys_timer_create:
1196  *
1197  *	System call to create a POSIX timer.
1198  */
1199 int
1200 sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap,
1201     register_t *retval)
1202 {
1203 	/* {
1204 		syscallarg(clockid_t) clock_id;
1205 		syscallarg(struct sigevent *) evp;
1206 		syscallarg(timer_t *) timerid;
1207 	} */
1208 
1209 	return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id),
1210 	    SCARG(uap, evp), copyin, l);
1211 }
1212 
1213 int
1214 timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp,
1215     copyin_t fetch_event, struct lwp *l)
1216 {
1217 	int error;
1218 	timer_t timerid;
1219 	struct itlist *itl;
1220 	struct ptimers *pts;
1221 	struct ptimer *pt;
1222 	struct proc *p;
1223 
1224 	p = l->l_proc;
1225 
1226 	if ((u_int)id > CLOCK_MONOTONIC)
1227 		return EINVAL;
1228 
1229 	if ((pts = p->p_timers) == NULL)
1230 		pts = ptimers_alloc(p);
1231 
1232 	pt = kmem_zalloc(sizeof(*pt), KM_SLEEP);
1233 	if (evp != NULL) {
1234 		if (((error =
1235 		    (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) ||
1236 		    ((pt->pt_ev.sigev_notify < SIGEV_NONE) ||
1237 			(pt->pt_ev.sigev_notify > SIGEV_SA)) ||
1238 			(pt->pt_ev.sigev_notify == SIGEV_SIGNAL &&
1239 			 (pt->pt_ev.sigev_signo <= 0 ||
1240 			  pt->pt_ev.sigev_signo >= NSIG))) {
1241 			kmem_free(pt, sizeof(*pt));
1242 			return (error ? error : EINVAL);
1243 		}
1244 	}
1245 
1246 	/* Find a free timer slot, skipping those reserved for setitimer(). */
1247 	itimer_lock();
1248 	for (timerid = TIMER_MIN; timerid < TIMER_MAX; timerid++)
1249 		if (pts->pts_timers[timerid] == NULL)
1250 			break;
1251 	if (timerid == TIMER_MAX) {
1252 		itimer_unlock();
1253 		kmem_free(pt, sizeof(*pt));
1254 		return EAGAIN;
1255 	}
1256 	if (evp == NULL) {
1257 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
1258 		switch (id) {
1259 		case CLOCK_REALTIME:
1260 		case CLOCK_MONOTONIC:
1261 			pt->pt_ev.sigev_signo = SIGALRM;
1262 			break;
1263 		case CLOCK_VIRTUAL:
1264 			pt->pt_ev.sigev_signo = SIGVTALRM;
1265 			break;
1266 		case CLOCK_PROF:
1267 			pt->pt_ev.sigev_signo = SIGPROF;
1268 			break;
1269 		}
1270 		pt->pt_ev.sigev_value.sival_int = timerid;
1271 	}
1272 
1273 	switch (id) {
1274 	case CLOCK_VIRTUAL:
1275 		itl = &pts->pts_virtual;
1276 		break;
1277 	case CLOCK_PROF:
1278 		itl = &pts->pts_prof;
1279 		break;
1280 	default:
1281 		itl = NULL;
1282 	}
1283 
1284 	itimer_init(&pt->pt_itimer, &ptimer_itimer_ops, id, itl);
1285 	pt->pt_proc = p;
1286 	pt->pt_poverruns = 0;
1287 	pt->pt_entry = timerid;
1288 	pt->pt_queued = false;
1289 
1290 	pts->pts_timers[timerid] = &pt->pt_itimer;
1291 	itimer_unlock();
1292 
1293 	return copyout(&timerid, tid, sizeof(timerid));
1294 }
1295 
1296 /*
1297  * sys_timer_delete:
1298  *
1299  *	System call to delete a POSIX timer.
1300  */
1301 int
1302 sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap,
1303     register_t *retval)
1304 {
1305 	/* {
1306 		syscallarg(timer_t) timerid;
1307 	} */
1308 	struct proc *p = l->l_proc;
1309 	timer_t timerid;
1310 	struct ptimers *pts;
1311 	struct itimer *it, *itn;
1312 
1313 	timerid = SCARG(uap, timerid);
1314 	pts = p->p_timers;
1315 
1316 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
1317 		return EINVAL;
1318 
1319 	itimer_lock();
1320 	if ((it = pts->pts_timers[timerid]) == NULL) {
1321 		itimer_unlock();
1322 		return EINVAL;
1323 	}
1324 
1325 	if (CLOCK_VIRTUAL_P(it->it_clockid)) {
1326 		if (it->it_active) {
1327 			itn = LIST_NEXT(it, it_list);
1328 			LIST_REMOVE(it, it_list);
1329 			for ( ; itn; itn = LIST_NEXT(itn, it_list))
1330 				timespecadd(&it->it_time.it_value,
1331 				    &itn->it_time.it_value,
1332 				    &itn->it_time.it_value);
1333 			it->it_active = false;
1334 		}
1335 	}
1336 
1337 	/* Free the timer and release the lock.  */
1338 	ptimer_free(pts, timerid);
1339 
1340 	return 0;
1341 }
1342 
1343 /*
1344  * sys___timer_settime50:
1345  *
1346  *	System call to set/arm a POSIX timer.
1347  */
1348 int
1349 sys___timer_settime50(struct lwp *l,
1350     const struct sys___timer_settime50_args *uap,
1351     register_t *retval)
1352 {
1353 	/* {
1354 		syscallarg(timer_t) timerid;
1355 		syscallarg(int) flags;
1356 		syscallarg(const struct itimerspec *) value;
1357 		syscallarg(struct itimerspec *) ovalue;
1358 	} */
1359 	int error;
1360 	struct itimerspec value, ovalue, *ovp = NULL;
1361 
1362 	if ((error = copyin(SCARG(uap, value), &value,
1363 	    sizeof(struct itimerspec))) != 0)
1364 		return error;
1365 
1366 	if (SCARG(uap, ovalue))
1367 		ovp = &ovalue;
1368 
1369 	if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp,
1370 	    SCARG(uap, flags), l->l_proc)) != 0)
1371 		return error;
1372 
1373 	if (ovp)
1374 		return copyout(&ovalue, SCARG(uap, ovalue),
1375 		    sizeof(struct itimerspec));
1376 	return 0;
1377 }
1378 
1379 int
1380 dotimer_settime(int timerid, struct itimerspec *value,
1381     struct itimerspec *ovalue, int flags, struct proc *p)
1382 {
1383 	struct timespec now;
1384 	struct itimerspec val, oval;
1385 	struct ptimers *pts;
1386 	struct itimer *it;
1387 	int error;
1388 
1389 	pts = p->p_timers;
1390 
1391 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
1392 		return EINVAL;
1393 	val = *value;
1394 	if ((error = itimespecfix(&val.it_value)) != 0 ||
1395 	    (error = itimespecfix(&val.it_interval)) != 0)
1396 		return error;
1397 
1398 	itimer_lock();
1399  restart:
1400 	if ((it = pts->pts_timers[timerid]) == NULL) {
1401 		itimer_unlock();
1402 		return EINVAL;
1403 	}
1404 
1405 	oval = it->it_time;
1406 	it->it_time = val;
1407 
1408 	/*
1409 	 * If we've been passed a relative time for a realtime timer,
1410 	 * convert it to absolute; if an absolute time for a virtual
1411 	 * timer, convert it to relative and make sure we don't set it
1412 	 * to zero, which would cancel the timer, or let it go
1413 	 * negative, which would confuse the comparison tests.
1414 	 */
1415 	if (timespecisset(&it->it_time.it_value)) {
1416 		if (!CLOCK_VIRTUAL_P(it->it_clockid)) {
1417 			if ((flags & TIMER_ABSTIME) == 0) {
1418 				if (it->it_clockid == CLOCK_REALTIME) {
1419 					getnanotime(&now);
1420 				} else { /* CLOCK_MONOTONIC */
1421 					getnanouptime(&now);
1422 				}
1423 				timespecadd(&it->it_time.it_value, &now,
1424 				    &it->it_time.it_value);
1425 			}
1426 		} else {
1427 			if ((flags & TIMER_ABSTIME) != 0) {
1428 				getnanotime(&now);
1429 				timespecsub(&it->it_time.it_value, &now,
1430 				    &it->it_time.it_value);
1431 				if (!timespecisset(&it->it_time.it_value) ||
1432 				    it->it_time.it_value.tv_sec < 0) {
1433 					it->it_time.it_value.tv_sec = 0;
1434 					it->it_time.it_value.tv_nsec = 1;
1435 				}
1436 			}
1437 		}
1438 	}
1439 
1440 	error = itimer_settime(it);
1441 	if (error == ERESTART) {
1442 		KASSERT(!CLOCK_VIRTUAL_P(it->it_clockid));
1443 		goto restart;
1444 	}
1445 	KASSERT(error == 0);
1446 	itimer_unlock();
1447 
1448 	if (ovalue)
1449 		*ovalue = oval;
1450 
1451 	return 0;
1452 }
1453 
1454 /*
1455  * sys___timer_gettime50:
1456  *
1457  *	System call to return the time remaining until a POSIX timer fires.
1458  */
1459 int
1460 sys___timer_gettime50(struct lwp *l,
1461     const struct sys___timer_gettime50_args *uap, register_t *retval)
1462 {
1463 	/* {
1464 		syscallarg(timer_t) timerid;
1465 		syscallarg(struct itimerspec *) value;
1466 	} */
1467 	struct itimerspec its;
1468 	int error;
1469 
1470 	if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc,
1471 	    &its)) != 0)
1472 		return error;
1473 
1474 	return copyout(&its, SCARG(uap, value), sizeof(its));
1475 }
1476 
1477 int
1478 dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its)
1479 {
1480 	struct itimer *it;
1481 	struct ptimers *pts;
1482 
1483 	pts = p->p_timers;
1484 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
1485 		return EINVAL;
1486 	itimer_lock();
1487 	if ((it = pts->pts_timers[timerid]) == NULL) {
1488 		itimer_unlock();
1489 		return EINVAL;
1490 	}
1491 	itimer_gettime(it, its);
1492 	itimer_unlock();
1493 
1494 	return 0;
1495 }
1496 
1497 /*
1498  * sys_timer_getoverrun:
1499  *
1500  *	System call to return the number of times a POSIX timer has
1501  *	expired while a notification was already pending.  The counter
1502  *	is reset when a timer expires and a notification can be posted.
1503  */
1504 int
1505 sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap,
1506     register_t *retval)
1507 {
1508 	/* {
1509 		syscallarg(timer_t) timerid;
1510 	} */
1511 	struct proc *p = l->l_proc;
1512 	struct ptimers *pts;
1513 	int timerid;
1514 	struct itimer *it;
1515 	struct ptimer *pt;
1516 
1517 	timerid = SCARG(uap, timerid);
1518 
1519 	pts = p->p_timers;
1520 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
1521 		return EINVAL;
1522 	itimer_lock();
1523 	if ((it = pts->pts_timers[timerid]) == NULL) {
1524 		itimer_unlock();
1525 		return EINVAL;
1526 	}
1527 	pt = container_of(it, struct ptimer, pt_itimer);
1528 	*retval = pt->pt_poverruns;
1529 	if (*retval >= DELAYTIMER_MAX)
1530 		*retval = DELAYTIMER_MAX;
1531 	itimer_unlock();
1532 
1533 	return 0;
1534 }
1535 
1536 /*
1537  * sys___getitimer50:
1538  *
1539  *	System call to get the time remaining before a BSD timer fires.
1540  */
1541 int
1542 sys___getitimer50(struct lwp *l, const struct sys___getitimer50_args *uap,
1543     register_t *retval)
1544 {
1545 	/* {
1546 		syscallarg(int) which;
1547 		syscallarg(struct itimerval *) itv;
1548 	} */
1549 	struct proc *p = l->l_proc;
1550 	struct itimerval aitv;
1551 	int error;
1552 
1553 	memset(&aitv, 0, sizeof(aitv));
1554 	error = dogetitimer(p, SCARG(uap, which), &aitv);
1555 	if (error)
1556 		return error;
1557 	return copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval));
1558 }
1559 
1560 int
1561 dogetitimer(struct proc *p, int which, struct itimerval *itvp)
1562 {
1563 	struct ptimers *pts;
1564 	struct itimer *it;
1565 	struct itimerspec its;
1566 
1567 	if ((u_int)which > ITIMER_MONOTONIC)
1568 		return EINVAL;
1569 
1570 	itimer_lock();
1571 	pts = p->p_timers;
1572 	if (pts == NULL || (it = pts->pts_timers[which]) == NULL) {
1573 		timerclear(&itvp->it_value);
1574 		timerclear(&itvp->it_interval);
1575 	} else {
1576 		itimer_gettime(it, &its);
1577 		TIMESPEC_TO_TIMEVAL(&itvp->it_value, &its.it_value);
1578 		TIMESPEC_TO_TIMEVAL(&itvp->it_interval, &its.it_interval);
1579 	}
1580 	itimer_unlock();
1581 
1582 	return 0;
1583 }
1584 
1585 /*
1586  * sys___setitimer50:
1587  *
1588  *	System call to set/arm a BSD timer.
1589  */
1590 int
1591 sys___setitimer50(struct lwp *l, const struct sys___setitimer50_args *uap,
1592     register_t *retval)
1593 {
1594 	/* {
1595 		syscallarg(int) which;
1596 		syscallarg(const struct itimerval *) itv;
1597 		syscallarg(struct itimerval *) oitv;
1598 	} */
1599 	struct proc *p = l->l_proc;
1600 	int which = SCARG(uap, which);
1601 	struct sys___getitimer50_args getargs;
1602 	const struct itimerval *itvp;
1603 	struct itimerval aitv;
1604 	int error;
1605 
1606 	itvp = SCARG(uap, itv);
1607 	if (itvp &&
1608 	    (error = copyin(itvp, &aitv, sizeof(struct itimerval))) != 0)
1609 		return error;
1610 	if (SCARG(uap, oitv) != NULL) {
1611 		SCARG(&getargs, which) = which;
1612 		SCARG(&getargs, itv) = SCARG(uap, oitv);
1613 		if ((error = sys___getitimer50(l, &getargs, retval)) != 0)
1614 			return error;
1615 	}
1616 	if (itvp == 0)
1617 		return 0;
1618 
1619 	return dosetitimer(p, which, &aitv);
1620 }
1621 
1622 int
1623 dosetitimer(struct proc *p, int which, struct itimerval *itvp)
1624 {
1625 	struct timespec now;
1626 	struct ptimers *pts;
1627 	struct ptimer *spare;
1628 	struct itimer *it;
1629 	struct itlist *itl;
1630 	int error;
1631 
1632 	if ((u_int)which > ITIMER_MONOTONIC)
1633 		return EINVAL;
1634 	if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval))
1635 		return EINVAL;
1636 
1637 	/*
1638 	 * Don't bother allocating data structures if the process just
1639 	 * wants to clear the timer.
1640 	 */
1641 	spare = NULL;
1642 	pts = p->p_timers;
1643  retry:
1644 	if (!timerisset(&itvp->it_value) && (pts == NULL ||
1645 	    pts->pts_timers[which] == NULL))
1646 		return 0;
1647 	if (pts == NULL)
1648 		pts = ptimers_alloc(p);
1649 	itimer_lock();
1650  restart:
1651 	it = pts->pts_timers[which];
1652 	if (it == NULL) {
1653 		struct ptimer *pt;
1654 
1655 		if (spare == NULL) {
1656 			itimer_unlock();
1657 			spare = kmem_zalloc(sizeof(*spare), KM_SLEEP);
1658 			goto retry;
1659 		}
1660 		pt = spare;
1661 		spare = NULL;
1662 
1663 		it = &pt->pt_itimer;
1664 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
1665 		pt->pt_ev.sigev_value.sival_int = which;
1666 
1667 		switch (which) {
1668 		case ITIMER_REAL:
1669 		case ITIMER_MONOTONIC:
1670 			itl = NULL;
1671 			pt->pt_ev.sigev_signo = SIGALRM;
1672 			break;
1673 		case ITIMER_VIRTUAL:
1674 			itl = &pts->pts_virtual;
1675 			pt->pt_ev.sigev_signo = SIGVTALRM;
1676 			break;
1677 		case ITIMER_PROF:
1678 			itl = &pts->pts_prof;
1679 			pt->pt_ev.sigev_signo = SIGPROF;
1680 			break;
1681 		default:
1682 			panic("%s: can't happen %d", __func__, which);
1683 		}
1684 		itimer_init(it, &ptimer_itimer_ops, which, itl);
1685 		pt->pt_proc = p;
1686 		pt->pt_entry = which;
1687 
1688 		pts->pts_timers[which] = it;
1689 	}
1690 
1691 	TIMEVAL_TO_TIMESPEC(&itvp->it_value, &it->it_time.it_value);
1692 	TIMEVAL_TO_TIMESPEC(&itvp->it_interval, &it->it_time.it_interval);
1693 
1694 	error = 0;
1695 	if (timespecisset(&it->it_time.it_value)) {
1696 		/* Convert to absolute time */
1697 		/* XXX need to wrap in splclock for timecounters case? */
1698 		switch (which) {
1699 		case ITIMER_REAL:
1700 			getnanotime(&now);
1701 			if (!timespecaddok(&it->it_time.it_value, &now)) {
1702 				error = EINVAL;
1703 				goto out;
1704 			}
1705 			timespecadd(&it->it_time.it_value, &now,
1706 			    &it->it_time.it_value);
1707 			break;
1708 		case ITIMER_MONOTONIC:
1709 			getnanouptime(&now);
1710 			if (!timespecaddok(&it->it_time.it_value, &now)) {
1711 				error = EINVAL;
1712 				goto out;
1713 			}
1714 			timespecadd(&it->it_time.it_value, &now,
1715 			    &it->it_time.it_value);
1716 			break;
1717 		default:
1718 			break;
1719 		}
1720 	}
1721 
1722 	error = itimer_settime(it);
1723 	if (error == ERESTART) {
1724 		KASSERT(!CLOCK_VIRTUAL_P(it->it_clockid));
1725 		goto restart;
1726 	}
1727 	KASSERT(error == 0);
1728 out:
1729 	itimer_unlock();
1730 	if (spare != NULL)
1731 		kmem_free(spare, sizeof(*spare));
1732 
1733 	return error;
1734 }
1735 
1736 /*
1737  * ptimer_tick:
1738  *
1739  *	Called from hardclock() to decrement per-process virtual timers.
1740  */
1741 void
1742 ptimer_tick(lwp_t *l, bool user)
1743 {
1744 	struct ptimers *pts;
1745 	struct itimer *it;
1746 	proc_t *p;
1747 
1748 	p = l->l_proc;
1749 	if (p->p_timers == NULL)
1750 		return;
1751 
1752 	itimer_lock();
1753 	if ((pts = l->l_proc->p_timers) != NULL) {
1754 		/*
1755 		 * Run current process's virtual and profile time, as needed.
1756 		 */
1757 		if (user && (it = LIST_FIRST(&pts->pts_virtual)) != NULL)
1758 			if (itimer_decr(it, tick * 1000))
1759 				(*it->it_ops->ito_fire)(it);
1760 		if ((it = LIST_FIRST(&pts->pts_prof)) != NULL)
1761 			if (itimer_decr(it, tick * 1000))
1762 				(*it->it_ops->ito_fire)(it);
1763 	}
1764 	itimer_unlock();
1765 }
1766 
1767 /*
1768  * ptimer_intr:
1769  *
1770  *	Software interrupt handler for processing per-process
1771  *	timer expiration.
1772  */
1773 static void
1774 ptimer_intr(void *cookie)
1775 {
1776 	ksiginfo_t ksi;
1777 	struct itimer *it;
1778 	struct ptimer *pt;
1779 	proc_t *p;
1780 
1781 	mutex_enter(&proc_lock);
1782 	itimer_lock();
1783 	while ((pt = TAILQ_FIRST(&ptimer_queue)) != NULL) {
1784 		it = &pt->pt_itimer;
1785 
1786 		TAILQ_REMOVE(&ptimer_queue, pt, pt_chain);
1787 		KASSERT(pt->pt_queued);
1788 		pt->pt_queued = false;
1789 
1790 		p = pt->pt_proc;
1791 		if (p->p_timers == NULL) {
1792 			/* Process is dying. */
1793 			continue;
1794 		}
1795 		if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) {
1796 			continue;
1797 		}
1798 		if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) {
1799 			it->it_overruns++;
1800 			continue;
1801 		}
1802 
1803 		KSI_INIT(&ksi);
1804 		ksi.ksi_signo = pt->pt_ev.sigev_signo;
1805 		ksi.ksi_code = SI_TIMER;
1806 		ksi.ksi_value = pt->pt_ev.sigev_value;
1807 		pt->pt_poverruns = it->it_overruns;
1808 		it->it_overruns = 0;
1809 		itimer_unlock();
1810 		kpsignal(p, &ksi, NULL);
1811 		itimer_lock();
1812 	}
1813 	itimer_unlock();
1814 	mutex_exit(&proc_lock);
1815 }
1816