kern_time.c revision 1.187
1/*	$NetBSD: kern_time.c,v 1.187 2016/06/10 23:29:20 christos 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 1.187 2016/06/10 23:29:20 christos 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/syscallargs.h>
79#include <sys/cpu.h>
80
81static void	timer_intr(void *);
82static void	itimerfire(struct ptimer *);
83static void	itimerfree(struct ptimers *, int);
84
85kmutex_t	timer_lock;
86
87static void	*timer_sih;
88static TAILQ_HEAD(, ptimer) timer_queue;
89
90struct pool ptimer_pool, ptimers_pool;
91
92#define	CLOCK_VIRTUAL_P(clockid)	\
93	((clockid) == CLOCK_VIRTUAL || (clockid) == CLOCK_PROF)
94
95CTASSERT(ITIMER_REAL == CLOCK_REALTIME);
96CTASSERT(ITIMER_VIRTUAL == CLOCK_VIRTUAL);
97CTASSERT(ITIMER_PROF == CLOCK_PROF);
98CTASSERT(ITIMER_MONOTONIC == CLOCK_MONOTONIC);
99
100#define	DELAYTIMER_MAX	32
101
102/*
103 * Initialize timekeeping.
104 */
105void
106time_init(void)
107{
108
109	pool_init(&ptimer_pool, sizeof(struct ptimer), 0, 0, 0, "ptimerpl",
110	    &pool_allocator_nointr, IPL_NONE);
111	pool_init(&ptimers_pool, sizeof(struct ptimers), 0, 0, 0, "ptimerspl",
112	    &pool_allocator_nointr, IPL_NONE);
113}
114
115void
116time_init2(void)
117{
118
119	TAILQ_INIT(&timer_queue);
120	mutex_init(&timer_lock, MUTEX_DEFAULT, IPL_SCHED);
121	timer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
122	    timer_intr, NULL);
123}
124
125/* Time of day and interval timer support.
126 *
127 * These routines provide the kernel entry points to get and set
128 * the time-of-day and per-process interval timers.  Subroutines
129 * here provide support for adding and subtracting timeval structures
130 * and decrementing interval timers, optionally reloading the interval
131 * timers when they expire.
132 */
133
134/* This function is used by clock_settime and settimeofday */
135static int
136settime1(struct proc *p, const struct timespec *ts, bool check_kauth)
137{
138	struct timespec delta, now;
139	int s;
140
141	/* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */
142	s = splclock();
143	nanotime(&now);
144	timespecsub(ts, &now, &delta);
145
146	if (check_kauth && kauth_authorize_system(kauth_cred_get(),
147	    KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, __UNCONST(ts),
148	    &delta, KAUTH_ARG(check_kauth ? false : true)) != 0) {
149		splx(s);
150		return (EPERM);
151	}
152
153#ifdef notyet
154	if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */
155		splx(s);
156		return (EPERM);
157	}
158#endif
159
160	tc_setclock(ts);
161
162	timespecadd(&boottime, &delta, &boottime);
163
164	resettodr();
165	splx(s);
166
167	return (0);
168}
169
170int
171settime(struct proc *p, struct timespec *ts)
172{
173	return (settime1(p, ts, true));
174}
175
176/* ARGSUSED */
177int
178sys___clock_gettime50(struct lwp *l,
179    const struct sys___clock_gettime50_args *uap, register_t *retval)
180{
181	/* {
182		syscallarg(clockid_t) clock_id;
183		syscallarg(struct timespec *) tp;
184	} */
185	int error;
186	struct timespec ats;
187
188	error = clock_gettime1(SCARG(uap, clock_id), &ats);
189	if (error != 0)
190		return error;
191
192	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
193}
194
195/* ARGSUSED */
196int
197sys___clock_settime50(struct lwp *l,
198    const struct sys___clock_settime50_args *uap, register_t *retval)
199{
200	/* {
201		syscallarg(clockid_t) clock_id;
202		syscallarg(const struct timespec *) tp;
203	} */
204	int error;
205	struct timespec ats;
206
207	if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
208		return error;
209
210	return clock_settime1(l->l_proc, SCARG(uap, clock_id), &ats, true);
211}
212
213
214int
215clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp,
216    bool check_kauth)
217{
218	int error;
219
220	switch (clock_id) {
221	case CLOCK_REALTIME:
222		if ((error = settime1(p, tp, check_kauth)) != 0)
223			return (error);
224		break;
225	case CLOCK_MONOTONIC:
226		return (EINVAL);	/* read-only clock */
227	default:
228		return (EINVAL);
229	}
230
231	return 0;
232}
233
234int
235sys___clock_getres50(struct lwp *l, const struct sys___clock_getres50_args *uap,
236    register_t *retval)
237{
238	/* {
239		syscallarg(clockid_t) clock_id;
240		syscallarg(struct timespec *) tp;
241	} */
242	struct timespec ts;
243	int error;
244
245	if ((error = clock_getres1(SCARG(uap, clock_id), &ts)) != 0)
246		return error;
247
248	if (SCARG(uap, tp))
249		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
250
251	return error;
252}
253
254int
255clock_getres1(clockid_t clock_id, struct timespec *ts)
256{
257
258	switch (clock_id) {
259	case CLOCK_REALTIME:
260	case CLOCK_MONOTONIC:
261		ts->tv_sec = 0;
262		if (tc_getfrequency() > 1000000000)
263			ts->tv_nsec = 1;
264		else
265			ts->tv_nsec = 1000000000 / tc_getfrequency();
266		break;
267	default:
268		return EINVAL;
269	}
270
271	return 0;
272}
273
274/* ARGSUSED */
275int
276sys___nanosleep50(struct lwp *l, const struct sys___nanosleep50_args *uap,
277    register_t *retval)
278{
279	/* {
280		syscallarg(struct timespec *) rqtp;
281		syscallarg(struct timespec *) rmtp;
282	} */
283	struct timespec rmt, rqt;
284	int error, error1;
285
286	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
287	if (error)
288		return (error);
289
290	error = nanosleep1(l, CLOCK_MONOTONIC, 0, &rqt,
291	    SCARG(uap, rmtp) ? &rmt : NULL);
292	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
293		return error;
294
295	error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt));
296	return error1 ? error1 : error;
297}
298
299/* ARGSUSED */
300int
301sys_clock_nanosleep(struct lwp *l, const struct sys_clock_nanosleep_args *uap,
302    register_t *retval)
303{
304	/* {
305		syscallarg(clockid_t) clock_id;
306		syscallarg(int) flags;
307		syscallarg(struct timespec *) rqtp;
308		syscallarg(struct timespec *) rmtp;
309	} */
310	struct timespec rmt, rqt;
311	int error, error1;
312
313	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
314	if (error)
315		goto out;
316
317	error = nanosleep1(l, SCARG(uap, clock_id), SCARG(uap, flags), &rqt,
318	    SCARG(uap, rmtp) ? &rmt : NULL);
319	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
320		goto out;
321
322	if ((error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt))) != 0)
323		error = error1;
324out:
325	*retval = error;
326	return 0;
327}
328
329int
330nanosleep1(struct lwp *l, clockid_t clock_id, int flags, struct timespec *rqt,
331    struct timespec *rmt)
332{
333	struct timespec rmtstart;
334	int error, timo;
335
336	if ((error = ts2timo(clock_id, flags, rqt, &timo, &rmtstart)) != 0) {
337		if (error == ETIMEDOUT) {
338			error = 0;
339			if (rmt != NULL)
340				rmt->tv_sec = rmt->tv_nsec = 0;
341		}
342		return error;
343	}
344
345	/*
346	 * Avoid inadvertently sleeping forever
347	 */
348	if (timo == 0)
349		timo = 1;
350again:
351	error = kpause("nanoslp", true, timo, NULL);
352	if (rmt != NULL || error == 0) {
353		struct timespec rmtend;
354		struct timespec t0;
355		struct timespec *t;
356
357		(void)clock_gettime1(clock_id, &rmtend);
358		t = (rmt != NULL) ? rmt : &t0;
359		if (flags & TIMER_ABSTIME) {
360			timespecsub(rqt, &rmtend, t);
361		} else {
362			timespecsub(&rmtend, &rmtstart, t);
363			timespecsub(rqt, t, t);
364		}
365		if (t->tv_sec < 0)
366			timespecclear(t);
367		if (error == 0) {
368			timo = tstohz(t);
369			if (timo > 0)
370				goto again;
371		}
372	}
373
374	if (error == ERESTART)
375		error = EINTR;
376	if (error == EWOULDBLOCK)
377		error = 0;
378
379	return error;
380}
381
382int
383sys_clock_getcpuclockid2(struct lwp *l,
384    const struct sys_clock_getcpuclockid2_args *uap,
385    register_t *retval)
386{
387	/* {
388		syscallarg(idtype_t idtype;
389		syscallarg(id_t id);
390		syscallarg(clockid_t *)clock_id;
391	} */
392	pid_t pid;
393	lwpid_t lid;
394	clockid_t clock_id;
395	id_t id = SCARG(uap, id);
396
397	switch (SCARG(uap, idtype)) {
398	case P_PID:
399		pid = id == 0 ? l->l_proc->p_pid : id;
400		clock_id = CLOCK_PROCESS_CPUTIME_ID | pid;
401		break;
402	case P_LWPID:
403		lid = id == 0 ? l->l_lid : id;
404		clock_id = CLOCK_THREAD_CPUTIME_ID | lid;
405		break;
406	default:
407		return EINVAL;
408	}
409	return copyout(&clock_id, SCARG(uap, clock_id), sizeof(clock_id));
410}
411
412/* ARGSUSED */
413int
414sys___gettimeofday50(struct lwp *l, const struct sys___gettimeofday50_args *uap,
415    register_t *retval)
416{
417	/* {
418		syscallarg(struct timeval *) tp;
419		syscallarg(void *) tzp;		really "struct timezone *";
420	} */
421	struct timeval atv;
422	int error = 0;
423	struct timezone tzfake;
424
425	if (SCARG(uap, tp)) {
426		microtime(&atv);
427		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
428		if (error)
429			return (error);
430	}
431	if (SCARG(uap, tzp)) {
432		/*
433		 * NetBSD has no kernel notion of time zone, so we just
434		 * fake up a timezone struct and return it if demanded.
435		 */
436		tzfake.tz_minuteswest = 0;
437		tzfake.tz_dsttime = 0;
438		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
439	}
440	return (error);
441}
442
443/* ARGSUSED */
444int
445sys___settimeofday50(struct lwp *l, const struct sys___settimeofday50_args *uap,
446    register_t *retval)
447{
448	/* {
449		syscallarg(const struct timeval *) tv;
450		syscallarg(const void *) tzp; really "const struct timezone *";
451	} */
452
453	return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true);
454}
455
456int
457settimeofday1(const struct timeval *utv, bool userspace,
458    const void *utzp, struct lwp *l, bool check_kauth)
459{
460	struct timeval atv;
461	struct timespec ts;
462	int error;
463
464	/* Verify all parameters before changing time. */
465
466	/*
467	 * NetBSD has no kernel notion of time zone, and only an
468	 * obsolete program would try to set it, so we log a warning.
469	 */
470	if (utzp)
471		log(LOG_WARNING, "pid %d attempted to set the "
472		    "(obsolete) kernel time zone\n", l->l_proc->p_pid);
473
474	if (utv == NULL)
475		return 0;
476
477	if (userspace) {
478		if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
479			return error;
480		utv = &atv;
481	}
482
483	TIMEVAL_TO_TIMESPEC(utv, &ts);
484	return settime1(l->l_proc, &ts, check_kauth);
485}
486
487int	time_adjusted;			/* set if an adjustment is made */
488
489/* ARGSUSED */
490int
491sys___adjtime50(struct lwp *l, const struct sys___adjtime50_args *uap,
492    register_t *retval)
493{
494	/* {
495		syscallarg(const struct timeval *) delta;
496		syscallarg(struct timeval *) olddelta;
497	} */
498	int error;
499	struct timeval atv, oldatv;
500
501	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
502	    KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0)
503		return error;
504
505	if (SCARG(uap, delta)) {
506		error = copyin(SCARG(uap, delta), &atv,
507		    sizeof(*SCARG(uap, delta)));
508		if (error)
509			return (error);
510	}
511	adjtime1(SCARG(uap, delta) ? &atv : NULL,
512	    SCARG(uap, olddelta) ? &oldatv : NULL, l->l_proc);
513	if (SCARG(uap, olddelta))
514		error = copyout(&oldatv, SCARG(uap, olddelta),
515		    sizeof(*SCARG(uap, olddelta)));
516	return error;
517}
518
519void
520adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p)
521{
522	extern int64_t time_adjtime;  /* in kern_ntptime.c */
523
524	if (olddelta) {
525		mutex_spin_enter(&timecounter_lock);
526		olddelta->tv_sec = time_adjtime / 1000000;
527		olddelta->tv_usec = time_adjtime % 1000000;
528		if (olddelta->tv_usec < 0) {
529			olddelta->tv_usec += 1000000;
530			olddelta->tv_sec--;
531		}
532		mutex_spin_exit(&timecounter_lock);
533	}
534
535	if (delta) {
536		mutex_spin_enter(&timecounter_lock);
537		time_adjtime = delta->tv_sec * 1000000 + delta->tv_usec;
538
539		if (time_adjtime) {
540			/* We need to save the system time during shutdown */
541			time_adjusted |= 1;
542		}
543		mutex_spin_exit(&timecounter_lock);
544	}
545}
546
547/*
548 * Interval timer support. Both the BSD getitimer() family and the POSIX
549 * timer_*() family of routines are supported.
550 *
551 * All timers are kept in an array pointed to by p_timers, which is
552 * allocated on demand - many processes don't use timers at all. The
553 * first four elements in this array are reserved for the BSD timers:
554 * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, element
555 * 2 is ITIMER_PROF, and element 3 is ITIMER_MONOTONIC. The rest may be
556 * allocated by the timer_create() syscall.
557 *
558 * Realtime timers are kept in the ptimer structure as an absolute
559 * time; virtual time timers are kept as a linked list of deltas.
560 * Virtual time timers are processed in the hardclock() routine of
561 * kern_clock.c.  The real time timer is processed by a callout
562 * routine, called from the softclock() routine.  Since a callout may
563 * be delayed in real time due to interrupt processing in the system,
564 * it is possible for the real time timeout routine (realtimeexpire,
565 * given below), to be delayed in real time past when it is supposed
566 * to occur.  It does not suffice, therefore, to reload the real timer
567 * .it_value from the real time timers .it_interval.  Rather, we
568 * compute the next time in absolute time the timer should go off.  */
569
570/* Allocate a POSIX realtime timer. */
571int
572sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap,
573    register_t *retval)
574{
575	/* {
576		syscallarg(clockid_t) clock_id;
577		syscallarg(struct sigevent *) evp;
578		syscallarg(timer_t *) timerid;
579	} */
580
581	return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id),
582	    SCARG(uap, evp), copyin, l);
583}
584
585int
586timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp,
587    copyin_t fetch_event, struct lwp *l)
588{
589	int error;
590	timer_t timerid;
591	struct ptimers *pts;
592	struct ptimer *pt;
593	struct proc *p;
594
595	p = l->l_proc;
596
597	if ((u_int)id > CLOCK_MONOTONIC)
598		return (EINVAL);
599
600	if ((pts = p->p_timers) == NULL)
601		pts = timers_alloc(p);
602
603	pt = pool_get(&ptimer_pool, PR_WAITOK);
604	if (evp != NULL) {
605		if (((error =
606		    (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) ||
607		    ((pt->pt_ev.sigev_notify < SIGEV_NONE) ||
608			(pt->pt_ev.sigev_notify > SIGEV_SA)) ||
609			(pt->pt_ev.sigev_notify == SIGEV_SIGNAL &&
610			 (pt->pt_ev.sigev_signo <= 0 ||
611			  pt->pt_ev.sigev_signo >= NSIG))) {
612			pool_put(&ptimer_pool, pt);
613			return (error ? error : EINVAL);
614		}
615	}
616
617	/* Find a free timer slot, skipping those reserved for setitimer(). */
618	mutex_spin_enter(&timer_lock);
619	for (timerid = TIMER_MIN; timerid < TIMER_MAX; timerid++)
620		if (pts->pts_timers[timerid] == NULL)
621			break;
622	if (timerid == TIMER_MAX) {
623		mutex_spin_exit(&timer_lock);
624		pool_put(&ptimer_pool, pt);
625		return EAGAIN;
626	}
627	if (evp == NULL) {
628		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
629		switch (id) {
630		case CLOCK_REALTIME:
631		case CLOCK_MONOTONIC:
632			pt->pt_ev.sigev_signo = SIGALRM;
633			break;
634		case CLOCK_VIRTUAL:
635			pt->pt_ev.sigev_signo = SIGVTALRM;
636			break;
637		case CLOCK_PROF:
638			pt->pt_ev.sigev_signo = SIGPROF;
639			break;
640		}
641		pt->pt_ev.sigev_value.sival_int = timerid;
642	}
643	pt->pt_info.ksi_signo = pt->pt_ev.sigev_signo;
644	pt->pt_info.ksi_errno = 0;
645	pt->pt_info.ksi_code = 0;
646	pt->pt_info.ksi_pid = p->p_pid;
647	pt->pt_info.ksi_uid = kauth_cred_getuid(l->l_cred);
648	pt->pt_info.ksi_value = pt->pt_ev.sigev_value;
649	pt->pt_type = id;
650	pt->pt_proc = p;
651	pt->pt_overruns = 0;
652	pt->pt_poverruns = 0;
653	pt->pt_entry = timerid;
654	pt->pt_queued = false;
655	timespecclear(&pt->pt_time.it_value);
656	if (!CLOCK_VIRTUAL_P(id))
657		callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
658	else
659		pt->pt_active = 0;
660
661	pts->pts_timers[timerid] = pt;
662	mutex_spin_exit(&timer_lock);
663
664	return copyout(&timerid, tid, sizeof(timerid));
665}
666
667/* Delete a POSIX realtime timer */
668int
669sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap,
670    register_t *retval)
671{
672	/* {
673		syscallarg(timer_t) timerid;
674	} */
675	struct proc *p = l->l_proc;
676	timer_t timerid;
677	struct ptimers *pts;
678	struct ptimer *pt, *ptn;
679
680	timerid = SCARG(uap, timerid);
681	pts = p->p_timers;
682
683	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
684		return (EINVAL);
685
686	mutex_spin_enter(&timer_lock);
687	if ((pt = pts->pts_timers[timerid]) == NULL) {
688		mutex_spin_exit(&timer_lock);
689		return (EINVAL);
690	}
691	if (CLOCK_VIRTUAL_P(pt->pt_type)) {
692		if (pt->pt_active) {
693			ptn = LIST_NEXT(pt, pt_list);
694			LIST_REMOVE(pt, pt_list);
695			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
696				timespecadd(&pt->pt_time.it_value,
697				    &ptn->pt_time.it_value,
698				    &ptn->pt_time.it_value);
699			pt->pt_active = 0;
700		}
701	}
702	itimerfree(pts, timerid);
703
704	return (0);
705}
706
707/*
708 * Set up the given timer. The value in pt->pt_time.it_value is taken
709 * to be an absolute time for CLOCK_REALTIME/CLOCK_MONOTONIC timers and
710 * a relative time for CLOCK_VIRTUAL/CLOCK_PROF timers.
711 */
712void
713timer_settime(struct ptimer *pt)
714{
715	struct ptimer *ptn, *pptn;
716	struct ptlist *ptl;
717
718	KASSERT(mutex_owned(&timer_lock));
719
720	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
721		callout_halt(&pt->pt_ch, &timer_lock);
722		if (timespecisset(&pt->pt_time.it_value)) {
723			/*
724			 * Don't need to check tshzto() return value, here.
725			 * callout_reset() does it for us.
726			 */
727			callout_reset(&pt->pt_ch,
728			    pt->pt_type == CLOCK_MONOTONIC ?
729			    tshztoup(&pt->pt_time.it_value) :
730			    tshzto(&pt->pt_time.it_value),
731			    realtimerexpire, pt);
732		}
733	} else {
734		if (pt->pt_active) {
735			ptn = LIST_NEXT(pt, pt_list);
736			LIST_REMOVE(pt, pt_list);
737			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
738				timespecadd(&pt->pt_time.it_value,
739				    &ptn->pt_time.it_value,
740				    &ptn->pt_time.it_value);
741		}
742		if (timespecisset(&pt->pt_time.it_value)) {
743			if (pt->pt_type == CLOCK_VIRTUAL)
744				ptl = &pt->pt_proc->p_timers->pts_virtual;
745			else
746				ptl = &pt->pt_proc->p_timers->pts_prof;
747
748			for (ptn = LIST_FIRST(ptl), pptn = NULL;
749			     ptn && timespeccmp(&pt->pt_time.it_value,
750				 &ptn->pt_time.it_value, >);
751			     pptn = ptn, ptn = LIST_NEXT(ptn, pt_list))
752				timespecsub(&pt->pt_time.it_value,
753				    &ptn->pt_time.it_value,
754				    &pt->pt_time.it_value);
755
756			if (pptn)
757				LIST_INSERT_AFTER(pptn, pt, pt_list);
758			else
759				LIST_INSERT_HEAD(ptl, pt, pt_list);
760
761			for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list))
762				timespecsub(&ptn->pt_time.it_value,
763				    &pt->pt_time.it_value,
764				    &ptn->pt_time.it_value);
765
766			pt->pt_active = 1;
767		} else
768			pt->pt_active = 0;
769	}
770}
771
772void
773timer_gettime(struct ptimer *pt, struct itimerspec *aits)
774{
775	struct timespec now;
776	struct ptimer *ptn;
777
778	KASSERT(mutex_owned(&timer_lock));
779
780	*aits = pt->pt_time;
781	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
782		/*
783		 * Convert from absolute to relative time in .it_value
784		 * part of real time timer.  If time for real time
785		 * timer has passed return 0, else return difference
786		 * between current time and time for the timer to go
787		 * off.
788		 */
789		if (timespecisset(&aits->it_value)) {
790			if (pt->pt_type == CLOCK_REALTIME) {
791				getnanotime(&now);
792			} else { /* CLOCK_MONOTONIC */
793				getnanouptime(&now);
794			}
795			if (timespeccmp(&aits->it_value, &now, <))
796				timespecclear(&aits->it_value);
797			else
798				timespecsub(&aits->it_value, &now,
799				    &aits->it_value);
800		}
801	} else if (pt->pt_active) {
802		if (pt->pt_type == CLOCK_VIRTUAL)
803			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual);
804		else
805			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof);
806		for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list))
807			timespecadd(&aits->it_value,
808			    &ptn->pt_time.it_value, &aits->it_value);
809		KASSERT(ptn != NULL); /* pt should be findable on the list */
810	} else
811		timespecclear(&aits->it_value);
812}
813
814
815
816/* Set and arm a POSIX realtime timer */
817int
818sys___timer_settime50(struct lwp *l,
819    const struct sys___timer_settime50_args *uap,
820    register_t *retval)
821{
822	/* {
823		syscallarg(timer_t) timerid;
824		syscallarg(int) flags;
825		syscallarg(const struct itimerspec *) value;
826		syscallarg(struct itimerspec *) ovalue;
827	} */
828	int error;
829	struct itimerspec value, ovalue, *ovp = NULL;
830
831	if ((error = copyin(SCARG(uap, value), &value,
832	    sizeof(struct itimerspec))) != 0)
833		return (error);
834
835	if (SCARG(uap, ovalue))
836		ovp = &ovalue;
837
838	if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp,
839	    SCARG(uap, flags), l->l_proc)) != 0)
840		return error;
841
842	if (ovp)
843		return copyout(&ovalue, SCARG(uap, ovalue),
844		    sizeof(struct itimerspec));
845	return 0;
846}
847
848int
849dotimer_settime(int timerid, struct itimerspec *value,
850    struct itimerspec *ovalue, int flags, struct proc *p)
851{
852	struct timespec now;
853	struct itimerspec val, oval;
854	struct ptimers *pts;
855	struct ptimer *pt;
856	int error;
857
858	pts = p->p_timers;
859
860	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
861		return EINVAL;
862	val = *value;
863	if ((error = itimespecfix(&val.it_value)) != 0 ||
864	    (error = itimespecfix(&val.it_interval)) != 0)
865		return error;
866
867	mutex_spin_enter(&timer_lock);
868	if ((pt = pts->pts_timers[timerid]) == NULL) {
869		mutex_spin_exit(&timer_lock);
870		return EINVAL;
871	}
872
873	oval = pt->pt_time;
874	pt->pt_time = val;
875
876	/*
877	 * If we've been passed a relative time for a realtime timer,
878	 * convert it to absolute; if an absolute time for a virtual
879	 * timer, convert it to relative and make sure we don't set it
880	 * to zero, which would cancel the timer, or let it go
881	 * negative, which would confuse the comparison tests.
882	 */
883	if (timespecisset(&pt->pt_time.it_value)) {
884		if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
885			if ((flags & TIMER_ABSTIME) == 0) {
886				if (pt->pt_type == CLOCK_REALTIME) {
887					getnanotime(&now);
888				} else { /* CLOCK_MONOTONIC */
889					getnanouptime(&now);
890				}
891				timespecadd(&pt->pt_time.it_value, &now,
892				    &pt->pt_time.it_value);
893			}
894		} else {
895			if ((flags & TIMER_ABSTIME) != 0) {
896				getnanotime(&now);
897				timespecsub(&pt->pt_time.it_value, &now,
898				    &pt->pt_time.it_value);
899				if (!timespecisset(&pt->pt_time.it_value) ||
900				    pt->pt_time.it_value.tv_sec < 0) {
901					pt->pt_time.it_value.tv_sec = 0;
902					pt->pt_time.it_value.tv_nsec = 1;
903				}
904			}
905		}
906	}
907
908	timer_settime(pt);
909	mutex_spin_exit(&timer_lock);
910
911	if (ovalue)
912		*ovalue = oval;
913
914	return (0);
915}
916
917/* Return the time remaining until a POSIX timer fires. */
918int
919sys___timer_gettime50(struct lwp *l,
920    const struct sys___timer_gettime50_args *uap, register_t *retval)
921{
922	/* {
923		syscallarg(timer_t) timerid;
924		syscallarg(struct itimerspec *) value;
925	} */
926	struct itimerspec its;
927	int error;
928
929	if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc,
930	    &its)) != 0)
931		return error;
932
933	return copyout(&its, SCARG(uap, value), sizeof(its));
934}
935
936int
937dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its)
938{
939	struct ptimer *pt;
940	struct ptimers *pts;
941
942	pts = p->p_timers;
943	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
944		return (EINVAL);
945	mutex_spin_enter(&timer_lock);
946	if ((pt = pts->pts_timers[timerid]) == NULL) {
947		mutex_spin_exit(&timer_lock);
948		return (EINVAL);
949	}
950	timer_gettime(pt, its);
951	mutex_spin_exit(&timer_lock);
952
953	return 0;
954}
955
956/*
957 * Return the count of the number of times a periodic timer expired
958 * while a notification was already pending. The counter is reset when
959 * a timer expires and a notification can be posted.
960 */
961int
962sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap,
963    register_t *retval)
964{
965	/* {
966		syscallarg(timer_t) timerid;
967	} */
968	struct proc *p = l->l_proc;
969	struct ptimers *pts;
970	int timerid;
971	struct ptimer *pt;
972
973	timerid = SCARG(uap, timerid);
974
975	pts = p->p_timers;
976	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
977		return (EINVAL);
978	mutex_spin_enter(&timer_lock);
979	if ((pt = pts->pts_timers[timerid]) == NULL) {
980		mutex_spin_exit(&timer_lock);
981		return (EINVAL);
982	}
983	*retval = pt->pt_poverruns;
984	if (*retval >= DELAYTIMER_MAX)
985		*retval = DELAYTIMER_MAX;
986	mutex_spin_exit(&timer_lock);
987
988	return (0);
989}
990
991/*
992 * Real interval timer expired:
993 * send process whose timer expired an alarm signal.
994 * If time is not set up to reload, then just return.
995 * Else compute next time timer should go off which is > current time.
996 * This is where delay in processing this timeout causes multiple
997 * SIGALRM calls to be compressed into one.
998 */
999void
1000realtimerexpire(void *arg)
1001{
1002	uint64_t last_val, next_val, interval, now_ns;
1003	struct timespec now, next;
1004	struct ptimer *pt;
1005	int backwards;
1006
1007	pt = arg;
1008
1009	mutex_spin_enter(&timer_lock);
1010	itimerfire(pt);
1011
1012	if (!timespecisset(&pt->pt_time.it_interval)) {
1013		timespecclear(&pt->pt_time.it_value);
1014		mutex_spin_exit(&timer_lock);
1015		return;
1016	}
1017
1018	if (pt->pt_type == CLOCK_MONOTONIC) {
1019		getnanouptime(&now);
1020	} else {
1021		getnanotime(&now);
1022	}
1023	backwards = (timespeccmp(&pt->pt_time.it_value, &now, >));
1024	timespecadd(&pt->pt_time.it_value, &pt->pt_time.it_interval, &next);
1025	/* Handle the easy case of non-overflown timers first. */
1026	if (!backwards && timespeccmp(&next, &now, >)) {
1027		pt->pt_time.it_value = next;
1028	} else {
1029		now_ns = timespec2ns(&now);
1030		last_val = timespec2ns(&pt->pt_time.it_value);
1031		interval = timespec2ns(&pt->pt_time.it_interval);
1032
1033		next_val = now_ns +
1034		    (now_ns - last_val + interval - 1) % interval;
1035
1036		if (backwards)
1037			next_val += interval;
1038		else
1039			pt->pt_overruns += (now_ns - last_val) / interval;
1040
1041		pt->pt_time.it_value.tv_sec = next_val / 1000000000;
1042		pt->pt_time.it_value.tv_nsec = next_val % 1000000000;
1043	}
1044
1045	/*
1046	 * Don't need to check tshzto() return value, here.
1047	 * callout_reset() does it for us.
1048	 */
1049	callout_reset(&pt->pt_ch, pt->pt_type == CLOCK_MONOTONIC ?
1050	    tshztoup(&pt->pt_time.it_value) : tshzto(&pt->pt_time.it_value),
1051	    realtimerexpire, pt);
1052	mutex_spin_exit(&timer_lock);
1053}
1054
1055/* BSD routine to get the value of an interval timer. */
1056/* ARGSUSED */
1057int
1058sys___getitimer50(struct lwp *l, const struct sys___getitimer50_args *uap,
1059    register_t *retval)
1060{
1061	/* {
1062		syscallarg(int) which;
1063		syscallarg(struct itimerval *) itv;
1064	} */
1065	struct proc *p = l->l_proc;
1066	struct itimerval aitv;
1067	int error;
1068
1069	error = dogetitimer(p, SCARG(uap, which), &aitv);
1070	if (error)
1071		return error;
1072	return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)));
1073}
1074
1075int
1076dogetitimer(struct proc *p, int which, struct itimerval *itvp)
1077{
1078	struct ptimers *pts;
1079	struct ptimer *pt;
1080	struct itimerspec its;
1081
1082	if ((u_int)which > ITIMER_MONOTONIC)
1083		return (EINVAL);
1084
1085	mutex_spin_enter(&timer_lock);
1086	pts = p->p_timers;
1087	if (pts == NULL || (pt = pts->pts_timers[which]) == NULL) {
1088		timerclear(&itvp->it_value);
1089		timerclear(&itvp->it_interval);
1090	} else {
1091		timer_gettime(pt, &its);
1092		TIMESPEC_TO_TIMEVAL(&itvp->it_value, &its.it_value);
1093		TIMESPEC_TO_TIMEVAL(&itvp->it_interval, &its.it_interval);
1094	}
1095	mutex_spin_exit(&timer_lock);
1096
1097	return 0;
1098}
1099
1100/* BSD routine to set/arm an interval timer. */
1101/* ARGSUSED */
1102int
1103sys___setitimer50(struct lwp *l, const struct sys___setitimer50_args *uap,
1104    register_t *retval)
1105{
1106	/* {
1107		syscallarg(int) which;
1108		syscallarg(const struct itimerval *) itv;
1109		syscallarg(struct itimerval *) oitv;
1110	} */
1111	struct proc *p = l->l_proc;
1112	int which = SCARG(uap, which);
1113	struct sys___getitimer50_args getargs;
1114	const struct itimerval *itvp;
1115	struct itimerval aitv;
1116	int error;
1117
1118	if ((u_int)which > ITIMER_MONOTONIC)
1119		return (EINVAL);
1120	itvp = SCARG(uap, itv);
1121	if (itvp &&
1122	    (error = copyin(itvp, &aitv, sizeof(struct itimerval))) != 0)
1123		return (error);
1124	if (SCARG(uap, oitv) != NULL) {
1125		SCARG(&getargs, which) = which;
1126		SCARG(&getargs, itv) = SCARG(uap, oitv);
1127		if ((error = sys___getitimer50(l, &getargs, retval)) != 0)
1128			return (error);
1129	}
1130	if (itvp == 0)
1131		return (0);
1132
1133	return dosetitimer(p, which, &aitv);
1134}
1135
1136int
1137dosetitimer(struct proc *p, int which, struct itimerval *itvp)
1138{
1139	struct timespec now;
1140	struct ptimers *pts;
1141	struct ptimer *pt, *spare;
1142
1143	KASSERT((u_int)which <= CLOCK_MONOTONIC);
1144	if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval))
1145		return (EINVAL);
1146
1147	/*
1148	 * Don't bother allocating data structures if the process just
1149	 * wants to clear the timer.
1150	 */
1151	spare = NULL;
1152	pts = p->p_timers;
1153 retry:
1154	if (!timerisset(&itvp->it_value) && (pts == NULL ||
1155	    pts->pts_timers[which] == NULL))
1156		return (0);
1157	if (pts == NULL)
1158		pts = timers_alloc(p);
1159	mutex_spin_enter(&timer_lock);
1160	pt = pts->pts_timers[which];
1161	if (pt == NULL) {
1162		if (spare == NULL) {
1163			mutex_spin_exit(&timer_lock);
1164			spare = pool_get(&ptimer_pool, PR_WAITOK);
1165			goto retry;
1166		}
1167		pt = spare;
1168		spare = NULL;
1169		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
1170		pt->pt_ev.sigev_value.sival_int = which;
1171		pt->pt_overruns = 0;
1172		pt->pt_proc = p;
1173		pt->pt_type = which;
1174		pt->pt_entry = which;
1175		pt->pt_queued = false;
1176		if (pt->pt_type == CLOCK_REALTIME)
1177			callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
1178		else
1179			pt->pt_active = 0;
1180
1181		switch (which) {
1182		case ITIMER_REAL:
1183		case ITIMER_MONOTONIC:
1184			pt->pt_ev.sigev_signo = SIGALRM;
1185			break;
1186		case ITIMER_VIRTUAL:
1187			pt->pt_ev.sigev_signo = SIGVTALRM;
1188			break;
1189		case ITIMER_PROF:
1190			pt->pt_ev.sigev_signo = SIGPROF;
1191			break;
1192		}
1193		pts->pts_timers[which] = pt;
1194	}
1195
1196	TIMEVAL_TO_TIMESPEC(&itvp->it_value, &pt->pt_time.it_value);
1197	TIMEVAL_TO_TIMESPEC(&itvp->it_interval, &pt->pt_time.it_interval);
1198
1199	if (timespecisset(&pt->pt_time.it_value)) {
1200		/* Convert to absolute time */
1201		/* XXX need to wrap in splclock for timecounters case? */
1202		switch (which) {
1203		case ITIMER_REAL:
1204			getnanotime(&now);
1205			timespecadd(&pt->pt_time.it_value, &now,
1206			    &pt->pt_time.it_value);
1207			break;
1208		case ITIMER_MONOTONIC:
1209			getnanouptime(&now);
1210			timespecadd(&pt->pt_time.it_value, &now,
1211			    &pt->pt_time.it_value);
1212			break;
1213		default:
1214			break;
1215		}
1216	}
1217	timer_settime(pt);
1218	mutex_spin_exit(&timer_lock);
1219	if (spare != NULL)
1220		pool_put(&ptimer_pool, spare);
1221
1222	return (0);
1223}
1224
1225/* Utility routines to manage the array of pointers to timers. */
1226struct ptimers *
1227timers_alloc(struct proc *p)
1228{
1229	struct ptimers *pts;
1230	int i;
1231
1232	pts = pool_get(&ptimers_pool, PR_WAITOK);
1233	LIST_INIT(&pts->pts_virtual);
1234	LIST_INIT(&pts->pts_prof);
1235	for (i = 0; i < TIMER_MAX; i++)
1236		pts->pts_timers[i] = NULL;
1237	mutex_spin_enter(&timer_lock);
1238	if (p->p_timers == NULL) {
1239		p->p_timers = pts;
1240		mutex_spin_exit(&timer_lock);
1241		return pts;
1242	}
1243	mutex_spin_exit(&timer_lock);
1244	pool_put(&ptimers_pool, pts);
1245	return p->p_timers;
1246}
1247
1248/*
1249 * Clean up the per-process timers. If "which" is set to TIMERS_ALL,
1250 * then clean up all timers and free all the data structures. If
1251 * "which" is set to TIMERS_POSIX, only clean up the timers allocated
1252 * by timer_create(), not the BSD setitimer() timers, and only free the
1253 * structure if none of those remain.
1254 */
1255void
1256timers_free(struct proc *p, int which)
1257{
1258	struct ptimers *pts;
1259	struct ptimer *ptn;
1260	struct timespec ts;
1261	int i;
1262
1263	if (p->p_timers == NULL)
1264		return;
1265
1266	pts = p->p_timers;
1267	mutex_spin_enter(&timer_lock);
1268	if (which == TIMERS_ALL) {
1269		p->p_timers = NULL;
1270		i = 0;
1271	} else {
1272		timespecclear(&ts);
1273		for (ptn = LIST_FIRST(&pts->pts_virtual);
1274		     ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL];
1275		     ptn = LIST_NEXT(ptn, pt_list)) {
1276			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
1277			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
1278		}
1279		LIST_FIRST(&pts->pts_virtual) = NULL;
1280		if (ptn) {
1281			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
1282			timespecadd(&ts, &ptn->pt_time.it_value,
1283			    &ptn->pt_time.it_value);
1284			LIST_INSERT_HEAD(&pts->pts_virtual, ptn, pt_list);
1285		}
1286		timespecclear(&ts);
1287		for (ptn = LIST_FIRST(&pts->pts_prof);
1288		     ptn && ptn != pts->pts_timers[ITIMER_PROF];
1289		     ptn = LIST_NEXT(ptn, pt_list)) {
1290			KASSERT(ptn->pt_type == CLOCK_PROF);
1291			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
1292		}
1293		LIST_FIRST(&pts->pts_prof) = NULL;
1294		if (ptn) {
1295			KASSERT(ptn->pt_type == CLOCK_PROF);
1296			timespecadd(&ts, &ptn->pt_time.it_value,
1297			    &ptn->pt_time.it_value);
1298			LIST_INSERT_HEAD(&pts->pts_prof, ptn, pt_list);
1299		}
1300		i = TIMER_MIN;
1301	}
1302	for ( ; i < TIMER_MAX; i++) {
1303		if (pts->pts_timers[i] != NULL) {
1304			itimerfree(pts, i);
1305			mutex_spin_enter(&timer_lock);
1306		}
1307	}
1308	if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL &&
1309	    pts->pts_timers[2] == NULL && pts->pts_timers[3] == NULL) {
1310		p->p_timers = NULL;
1311		mutex_spin_exit(&timer_lock);
1312		pool_put(&ptimers_pool, pts);
1313	} else
1314		mutex_spin_exit(&timer_lock);
1315}
1316
1317static void
1318itimerfree(struct ptimers *pts, int index)
1319{
1320	struct ptimer *pt;
1321
1322	KASSERT(mutex_owned(&timer_lock));
1323
1324	pt = pts->pts_timers[index];
1325	pts->pts_timers[index] = NULL;
1326	if (!CLOCK_VIRTUAL_P(pt->pt_type))
1327		callout_halt(&pt->pt_ch, &timer_lock);
1328	if (pt->pt_queued)
1329		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
1330	mutex_spin_exit(&timer_lock);
1331	if (!CLOCK_VIRTUAL_P(pt->pt_type))
1332		callout_destroy(&pt->pt_ch);
1333	pool_put(&ptimer_pool, pt);
1334}
1335
1336/*
1337 * Decrement an interval timer by a specified number
1338 * of nanoseconds, which must be less than a second,
1339 * i.e. < 1000000000.  If the timer expires, then reload
1340 * it.  In this case, carry over (nsec - old value) to
1341 * reduce the value reloaded into the timer so that
1342 * the timer does not drift.  This routine assumes
1343 * that it is called in a context where the timers
1344 * on which it is operating cannot change in value.
1345 */
1346static int
1347itimerdecr(struct ptimer *pt, int nsec)
1348{
1349	struct itimerspec *itp;
1350
1351	KASSERT(mutex_owned(&timer_lock));
1352	KASSERT(CLOCK_VIRTUAL_P(pt->pt_type));
1353
1354	itp = &pt->pt_time;
1355	if (itp->it_value.tv_nsec < nsec) {
1356		if (itp->it_value.tv_sec == 0) {
1357			/* expired, and already in next interval */
1358			nsec -= itp->it_value.tv_nsec;
1359			goto expire;
1360		}
1361		itp->it_value.tv_nsec += 1000000000;
1362		itp->it_value.tv_sec--;
1363	}
1364	itp->it_value.tv_nsec -= nsec;
1365	nsec = 0;
1366	if (timespecisset(&itp->it_value))
1367		return (1);
1368	/* expired, exactly at end of interval */
1369expire:
1370	if (timespecisset(&itp->it_interval)) {
1371		itp->it_value = itp->it_interval;
1372		itp->it_value.tv_nsec -= nsec;
1373		if (itp->it_value.tv_nsec < 0) {
1374			itp->it_value.tv_nsec += 1000000000;
1375			itp->it_value.tv_sec--;
1376		}
1377		timer_settime(pt);
1378	} else
1379		itp->it_value.tv_nsec = 0;		/* sec is already 0 */
1380	return (0);
1381}
1382
1383static void
1384itimerfire(struct ptimer *pt)
1385{
1386
1387	KASSERT(mutex_owned(&timer_lock));
1388
1389	/*
1390	 * XXX Can overrun, but we don't do signal queueing yet, anyway.
1391	 * XXX Relying on the clock interrupt is stupid.
1392	 */
1393	if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL || pt->pt_queued) {
1394		return;
1395	}
1396	TAILQ_INSERT_TAIL(&timer_queue, pt, pt_chain);
1397	pt->pt_queued = true;
1398	softint_schedule(timer_sih);
1399}
1400
1401void
1402timer_tick(lwp_t *l, bool user)
1403{
1404	struct ptimers *pts;
1405	struct ptimer *pt;
1406	proc_t *p;
1407
1408	p = l->l_proc;
1409	if (p->p_timers == NULL)
1410		return;
1411
1412	mutex_spin_enter(&timer_lock);
1413	if ((pts = l->l_proc->p_timers) != NULL) {
1414		/*
1415		 * Run current process's virtual and profile time, as needed.
1416		 */
1417		if (user && (pt = LIST_FIRST(&pts->pts_virtual)) != NULL)
1418			if (itimerdecr(pt, tick * 1000) == 0)
1419				itimerfire(pt);
1420		if ((pt = LIST_FIRST(&pts->pts_prof)) != NULL)
1421			if (itimerdecr(pt, tick * 1000) == 0)
1422				itimerfire(pt);
1423	}
1424	mutex_spin_exit(&timer_lock);
1425}
1426
1427static void
1428timer_intr(void *cookie)
1429{
1430	ksiginfo_t ksi;
1431	struct ptimer *pt;
1432	proc_t *p;
1433
1434	mutex_enter(proc_lock);
1435	mutex_spin_enter(&timer_lock);
1436	while ((pt = TAILQ_FIRST(&timer_queue)) != NULL) {
1437		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
1438		KASSERT(pt->pt_queued);
1439		pt->pt_queued = false;
1440
1441		if (pt->pt_proc->p_timers == NULL) {
1442			/* Process is dying. */
1443			continue;
1444		}
1445		p = pt->pt_proc;
1446		if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) {
1447			continue;
1448		}
1449		if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) {
1450			pt->pt_overruns++;
1451			continue;
1452		}
1453
1454		KSI_INIT(&ksi);
1455		ksi.ksi_signo = pt->pt_ev.sigev_signo;
1456		ksi.ksi_code = SI_TIMER;
1457		ksi.ksi_value = pt->pt_ev.sigev_value;
1458		pt->pt_poverruns = pt->pt_overruns;
1459		pt->pt_overruns = 0;
1460		mutex_spin_exit(&timer_lock);
1461		kpsignal(p, &ksi, NULL);
1462		mutex_spin_enter(&timer_lock);
1463	}
1464	mutex_spin_exit(&timer_lock);
1465	mutex_exit(proc_lock);
1466}
1467
1468/*
1469 * Check if the time will wrap if set to ts.
1470 *
1471 * ts - timespec describing the new time
1472 * delta - the delta between the current time and ts
1473 */
1474bool
1475time_wraps(struct timespec *ts, struct timespec *delta)
1476{
1477
1478	/*
1479	 * Don't allow the time to be set forward so far it
1480	 * will wrap and become negative, thus allowing an
1481	 * attacker to bypass the next check below.  The
1482	 * cutoff is 1 year before rollover occurs, so even
1483	 * if the attacker uses adjtime(2) to move the time
1484	 * past the cutoff, it will take a very long time
1485	 * to get to the wrap point.
1486	 */
1487	if ((ts->tv_sec > LLONG_MAX - 365*24*60*60) ||
1488	    (delta->tv_sec < 0 || delta->tv_nsec < 0))
1489		return true;
1490
1491	return false;
1492}
1493