xref: /freebsd/sys/kern/kern_time.c (revision a8445737e740901f5f2c8d24c12ef7fc8b00134e)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)kern_time.c	8.1 (Berkeley) 6/10/93
34  * $Id: kern_time.c,v 1.57 1998/05/17 20:13:01 bde Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/sysproto.h>
39 #include <sys/resourcevar.h>
40 #include <sys/signalvar.h>
41 #include <sys/kernel.h>
42 #include <sys/systm.h>
43 #include <sys/sysent.h>
44 #include <sys/proc.h>
45 #include <sys/time.h>
46 #include <sys/vnode.h>
47 #include <vm/vm.h>
48 #include <vm/vm_extern.h>
49 
50 struct timezone tz;
51 
52 /*
53  * Time of day and interval timer support.
54  *
55  * These routines provide the kernel entry points to get and set
56  * the time-of-day and per-process interval timers.  Subroutines
57  * here provide support for adding and subtracting timeval structures
58  * and decrementing interval timers, optionally reloading the interval
59  * timers when they expire.
60  */
61 
62 static int	nanosleep1 __P((struct proc *p, struct timespec *rqt,
63 		    struct timespec *rmt));
64 static int	settime __P((struct timeval *));
65 static void	timevalfix __P((struct timeval *));
66 static void	no_lease_updatetime __P((int));
67 
68 static void
69 no_lease_updatetime(deltat)
70 	int deltat;
71 {
72 }
73 
74 void (*lease_updatetime) __P((int))  = no_lease_updatetime;
75 
76 static int
77 settime(tv)
78 	struct timeval *tv;
79 {
80 	struct timeval delta, tv1;
81 	struct timespec ts;
82 	struct proc *p;
83 	int s;
84 
85 	s = splclock();
86 	microtime(&tv1);
87 	delta = *tv;
88 	timevalsub(&delta, &tv1);
89 
90 	/*
91 	 * If the system is secure, we do not allow the time to be
92 	 * set to an earlier value (it may be slowed using adjtime,
93 	 * but not set back). This feature prevent interlopers from
94 	 * setting arbitrary time stamps on files.
95 	 */
96 	if (delta.tv_sec < 0 && securelevel > 1) {
97 		splx(s);
98 		return (EPERM);
99 	}
100 
101 	ts.tv_sec = tv->tv_sec;
102 	ts.tv_nsec = tv->tv_usec * 1000;
103 	set_timecounter(&ts);
104 	(void) splsoftclock();
105 	lease_updatetime(delta.tv_sec);
106 	splx(s);
107 	resettodr();
108 	return (0);
109 }
110 
111 #ifndef _SYS_SYSPROTO_H_
112 struct clock_gettime_args {
113 	clockid_t clock_id;
114 	struct	timespec *tp;
115 };
116 #endif
117 
118 /* ARGSUSED */
119 int
120 clock_gettime(p, uap)
121 	struct proc *p;
122 	struct clock_gettime_args *uap;
123 {
124 	struct timespec ats;
125 
126 	if (SCARG(uap, clock_id) != CLOCK_REALTIME)
127 		return (EINVAL);
128 	nanotime(&ats);
129 	return (copyout(&ats, SCARG(uap, tp), sizeof(ats)));
130 }
131 
132 #ifndef _SYS_SYSPROTO_H_
133 struct clock_settime_args {
134 	clockid_t clock_id;
135 	const struct	timespec *tp;
136 };
137 #endif
138 
139 /* ARGSUSED */
140 int
141 clock_settime(p, uap)
142 	struct proc *p;
143 	struct clock_settime_args *uap;
144 {
145 	struct timeval atv;
146 	struct timespec ats;
147 	int error;
148 
149 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
150 		return (error);
151 	if (SCARG(uap, clock_id) != CLOCK_REALTIME)
152 		return (EINVAL);
153 	if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
154 		return (error);
155 	if (ats.tv_nsec < 0 || ats.tv_nsec >= 1000000000)
156 		return (EINVAL);
157 	/* XXX Don't convert nsec->usec and back */
158 	TIMESPEC_TO_TIMEVAL(&atv, &ats);
159 	if ((error = settime(&atv)))
160 		return (error);
161 	return (0);
162 }
163 
164 #ifndef _SYS_SYSPROTO_H_
165 struct clock_getres_args {
166 	clockid_t clock_id;
167 	struct	timespec *tp;
168 };
169 #endif
170 
171 int
172 clock_getres(p, uap)
173 	struct proc *p;
174 	struct clock_getres_args *uap;
175 {
176 	struct timespec ts;
177 	int error;
178 
179 	if (SCARG(uap, clock_id) != CLOCK_REALTIME)
180 		return (EINVAL);
181 	error = 0;
182 	if (SCARG(uap, tp)) {
183 		ts.tv_sec = 0;
184 		ts.tv_nsec = 1000000000 / timecounter->tc_frequency;
185 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
186 	}
187 	return (error);
188 }
189 
190 static int nanowait;
191 
192 static int
193 nanosleep1(p, rqt, rmt)
194 	struct proc *p;
195 	struct timespec *rqt, *rmt;
196 {
197 	struct timespec ts, ts2, ts3;
198 	struct timeval tv;
199 	int error;
200 
201 	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
202 		return (EINVAL);
203 	if (rqt->tv_sec < 0 || rqt->tv_sec == 0 && rqt->tv_nsec == 0)
204 		return (0);
205 	getnanouptime(&ts);
206 	timespecadd(&ts, rqt);
207 	TIMESPEC_TO_TIMEVAL(&tv, rqt);
208 	for (;;) {
209 		error = tsleep(&nanowait, PWAIT | PCATCH, "nanslp",
210 		    tvtohz(&tv));
211 		getnanouptime(&ts2);
212 		if (error != EWOULDBLOCK) {
213 			if (error == ERESTART)
214 				error = EINTR;
215 			if (rmt != NULL) {
216 				timespecsub(&ts, &ts2);
217 				if (ts.tv_sec < 0)
218 					timespecclear(&ts);
219 				*rmt = ts;
220 			}
221 			return (error);
222 		}
223 		if (timespeccmp(&ts2, &ts, >=))
224 			return (0);
225 		ts3 = ts;
226 		timespecsub(&ts3, &ts2);
227 		TIMESPEC_TO_TIMEVAL(&tv, &ts3);
228 	}
229 }
230 
231 #ifndef _SYS_SYSPROTO_H_
232 struct nanosleep_args {
233 	struct	timespec *rqtp;
234 	struct	timespec *rmtp;
235 };
236 #endif
237 
238 /* ARGSUSED */
239 int
240 nanosleep(p, uap)
241 	struct proc *p;
242 	struct nanosleep_args *uap;
243 {
244 	struct timespec rmt, rqt;
245 	int error, error2;
246 
247 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(rqt));
248 	if (error)
249 		return (error);
250 	if (SCARG(uap, rmtp))
251 		if (!useracc((caddr_t)SCARG(uap, rmtp), sizeof(rmt), B_WRITE))
252 			return (EFAULT);
253 	error = nanosleep1(p, &rqt, &rmt);
254 	if (error && SCARG(uap, rmtp)) {
255 		error2 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt));
256 		if (error2)	/* XXX shouldn't happen, did useracc() above */
257 			return (error2);
258 	}
259 	return (error);
260 }
261 
262 #ifndef _SYS_SYSPROTO_H_
263 struct gettimeofday_args {
264 	struct	timeval *tp;
265 	struct	timezone *tzp;
266 };
267 #endif
268 /* ARGSUSED */
269 int
270 gettimeofday(p, uap)
271 	struct proc *p;
272 	register struct gettimeofday_args *uap;
273 {
274 	struct timeval atv;
275 	int error = 0;
276 
277 	if (uap->tp) {
278 		microtime(&atv);
279 		if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
280 		    sizeof (atv))))
281 			return (error);
282 	}
283 	if (uap->tzp)
284 		error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
285 		    sizeof (tz));
286 	return (error);
287 }
288 
289 #ifndef _SYS_SYSPROTO_H_
290 struct settimeofday_args {
291 	struct	timeval *tv;
292 	struct	timezone *tzp;
293 };
294 #endif
295 /* ARGSUSED */
296 int
297 settimeofday(p, uap)
298 	struct proc *p;
299 	struct settimeofday_args *uap;
300 {
301 	struct timeval atv;
302 	struct timezone atz;
303 	int error;
304 
305 	if ((error = suser(p->p_ucred, &p->p_acflag)))
306 		return (error);
307 	/* Verify all parameters before changing time. */
308 	if (uap->tv) {
309 		if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
310 		    sizeof(atv))))
311 			return (error);
312 		if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
313 			return (EINVAL);
314 	}
315 	if (uap->tzp &&
316 	    (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
317 		return (error);
318 	if (uap->tv && (error = settime(&atv)))
319 		return (error);
320 	if (uap->tzp)
321 		tz = atz;
322 	return (0);
323 }
324 
325 int	tickdelta;			/* current clock skew, us. per tick */
326 long	timedelta;			/* unapplied time correction, us. */
327 static long	bigadj = 1000000;	/* use 10x skew above bigadj us. */
328 
329 #ifndef _SYS_SYSPROTO_H_
330 struct adjtime_args {
331 	struct timeval *delta;
332 	struct timeval *olddelta;
333 };
334 #endif
335 /* ARGSUSED */
336 int
337 adjtime(p, uap)
338 	struct proc *p;
339 	register struct adjtime_args *uap;
340 {
341 	struct timeval atv;
342 	register long ndelta, ntickdelta, odelta;
343 	int s, error;
344 
345 	if ((error = suser(p->p_ucred, &p->p_acflag)))
346 		return (error);
347 	if ((error =
348 	    copyin((caddr_t)uap->delta, (caddr_t)&atv, sizeof(struct timeval))))
349 		return (error);
350 
351 	/*
352 	 * Compute the total correction and the rate at which to apply it.
353 	 * Round the adjustment down to a whole multiple of the per-tick
354 	 * delta, so that after some number of incremental changes in
355 	 * hardclock(), tickdelta will become zero, lest the correction
356 	 * overshoot and start taking us away from the desired final time.
357 	 */
358 	ndelta = atv.tv_sec * 1000000 + atv.tv_usec;
359 	if (ndelta > bigadj || ndelta < -bigadj)
360 		ntickdelta = 10 * tickadj;
361 	else
362 		ntickdelta = tickadj;
363 	if (ndelta % ntickdelta)
364 		ndelta = ndelta / ntickdelta * ntickdelta;
365 
366 	/*
367 	 * To make hardclock()'s job easier, make the per-tick delta negative
368 	 * if we want time to run slower; then hardclock can simply compute
369 	 * tick + tickdelta, and subtract tickdelta from timedelta.
370 	 */
371 	if (ndelta < 0)
372 		ntickdelta = -ntickdelta;
373 	s = splclock();
374 	odelta = timedelta;
375 	timedelta = ndelta;
376 	tickdelta = ntickdelta;
377 	splx(s);
378 
379 	if (uap->olddelta) {
380 		atv.tv_sec = odelta / 1000000;
381 		atv.tv_usec = odelta % 1000000;
382 		(void) copyout((caddr_t)&atv, (caddr_t)uap->olddelta,
383 		    sizeof(struct timeval));
384 	}
385 	return (0);
386 }
387 
388 /*
389  * Get value of an interval timer.  The process virtual and
390  * profiling virtual time timers are kept in the p_stats area, since
391  * they can be swapped out.  These are kept internally in the
392  * way they are specified externally: in time until they expire.
393  *
394  * The real time interval timer is kept in the process table slot
395  * for the process, and its value (it_value) is kept as an
396  * absolute time rather than as a delta, so that it is easy to keep
397  * periodic real-time signals from drifting.
398  *
399  * Virtual time timers are processed in the hardclock() routine of
400  * kern_clock.c.  The real time timer is processed by a timeout
401  * routine, called from the softclock() routine.  Since a callout
402  * may be delayed in real time due to interrupt processing in the system,
403  * it is possible for the real time timeout routine (realitexpire, given below),
404  * to be delayed in real time past when it is supposed to occur.  It
405  * does not suffice, therefore, to reload the real timer .it_value from the
406  * real time timers .it_interval.  Rather, we compute the next time in
407  * absolute time the timer should go off.
408  */
409 #ifndef _SYS_SYSPROTO_H_
410 struct getitimer_args {
411 	u_int	which;
412 	struct	itimerval *itv;
413 };
414 #endif
415 /* ARGSUSED */
416 int
417 getitimer(p, uap)
418 	struct proc *p;
419 	register struct getitimer_args *uap;
420 {
421 	struct timeval ctv;
422 	struct itimerval aitv;
423 	int s;
424 
425 	if (uap->which > ITIMER_PROF)
426 		return (EINVAL);
427 	s = splclock(); /* XXX still needed ? */
428 	if (uap->which == ITIMER_REAL) {
429 		/*
430 		 * Convert from absolute to relative time in .it_value
431 		 * part of real time timer.  If time for real time timer
432 		 * has passed return 0, else return difference between
433 		 * current time and time for the timer to go off.
434 		 */
435 		aitv = p->p_realtimer;
436 		if (timevalisset(&aitv.it_value)) {
437 			getmicrouptime(&ctv);
438 			if (timevalcmp(&aitv.it_value, &ctv, <))
439 				timevalclear(&aitv.it_value);
440 			else
441 				timevalsub(&aitv.it_value, &ctv);
442 		}
443 	} else
444 		aitv = p->p_stats->p_timer[uap->which];
445 	splx(s);
446 	return (copyout((caddr_t)&aitv, (caddr_t)uap->itv,
447 	    sizeof (struct itimerval)));
448 }
449 
450 #ifndef _SYS_SYSPROTO_H_
451 struct setitimer_args {
452 	u_int	which;
453 	struct	itimerval *itv, *oitv;
454 };
455 #endif
456 /* ARGSUSED */
457 int
458 setitimer(p, uap)
459 	struct proc *p;
460 	register struct setitimer_args *uap;
461 {
462 	struct itimerval aitv;
463 	struct timeval ctv;
464 	register struct itimerval *itvp;
465 	int s, error;
466 
467 	if (uap->which > ITIMER_PROF)
468 		return (EINVAL);
469 	itvp = uap->itv;
470 	if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
471 	    sizeof(struct itimerval))))
472 		return (error);
473 	if ((uap->itv = uap->oitv) &&
474 	    (error = getitimer(p, (struct getitimer_args *)uap)))
475 		return (error);
476 	if (itvp == 0)
477 		return (0);
478 	if (itimerfix(&aitv.it_value))
479 		return (EINVAL);
480 	if (!timevalisset(&aitv.it_value))
481 		timevalclear(&aitv.it_interval);
482 	else if (itimerfix(&aitv.it_interval))
483 		return (EINVAL);
484 	s = splclock(); /* XXX: still needed ? */
485 	if (uap->which == ITIMER_REAL) {
486 		if (timevalisset(&p->p_realtimer.it_value))
487 			untimeout(realitexpire, (caddr_t)p, p->p_ithandle);
488 		if (timevalisset(&aitv.it_value))
489 			p->p_ithandle = timeout(realitexpire, (caddr_t)p,
490 						tvtohz(&aitv.it_value));
491 		getmicrouptime(&ctv);
492 		timevaladd(&aitv.it_value, &ctv);
493 		p->p_realtimer = aitv;
494 	} else
495 		p->p_stats->p_timer[uap->which] = aitv;
496 	splx(s);
497 	return (0);
498 }
499 
500 /*
501  * Real interval timer expired:
502  * send process whose timer expired an alarm signal.
503  * If time is not set up to reload, then just return.
504  * Else compute next time timer should go off which is > current time.
505  * This is where delay in processing this timeout causes multiple
506  * SIGALRM calls to be compressed into one.
507  * tvtohz() always adds 1 to allow for the time until the next clock
508  * interrupt being strictly less than 1 clock tick, but we don't want
509  * that here since we want to appear to be in sync with the clock
510  * interrupt even when we're delayed.
511  */
512 void
513 realitexpire(arg)
514 	void *arg;
515 {
516 	register struct proc *p;
517 	struct timeval ctv, ntv;
518 	int s;
519 
520 	p = (struct proc *)arg;
521 	psignal(p, SIGALRM);
522 	if (!timevalisset(&p->p_realtimer.it_interval)) {
523 		timevalclear(&p->p_realtimer.it_value);
524 		return;
525 	}
526 	for (;;) {
527 		s = splclock(); /* XXX: still neeeded ? */
528 		timevaladd(&p->p_realtimer.it_value,
529 		    &p->p_realtimer.it_interval);
530 		getmicrouptime(&ctv);
531 		if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
532 			ntv = p->p_realtimer.it_value;
533 			timevalsub(&ntv, &ctv);
534 			p->p_ithandle = timeout(realitexpire, (caddr_t)p,
535 			    tvtohz(&ntv) - 1);
536 			splx(s);
537 			return;
538 		}
539 		splx(s);
540 	}
541 }
542 
543 /*
544  * Check that a proposed value to load into the .it_value or
545  * .it_interval part of an interval timer is acceptable, and
546  * fix it to have at least minimal value (i.e. if it is less
547  * than the resolution of the clock, round it up.)
548  */
549 int
550 itimerfix(tv)
551 	struct timeval *tv;
552 {
553 
554 	if (tv->tv_sec < 0 || tv->tv_sec > 100000000 ||
555 	    tv->tv_usec < 0 || tv->tv_usec >= 1000000)
556 		return (EINVAL);
557 	if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
558 		tv->tv_usec = tick;
559 	return (0);
560 }
561 
562 /*
563  * Decrement an interval timer by a specified number
564  * of microseconds, which must be less than a second,
565  * i.e. < 1000000.  If the timer expires, then reload
566  * it.  In this case, carry over (usec - old value) to
567  * reduce the value reloaded into the timer so that
568  * the timer does not drift.  This routine assumes
569  * that it is called in a context where the timers
570  * on which it is operating cannot change in value.
571  */
572 int
573 itimerdecr(itp, usec)
574 	register struct itimerval *itp;
575 	int usec;
576 {
577 
578 	if (itp->it_value.tv_usec < usec) {
579 		if (itp->it_value.tv_sec == 0) {
580 			/* expired, and already in next interval */
581 			usec -= itp->it_value.tv_usec;
582 			goto expire;
583 		}
584 		itp->it_value.tv_usec += 1000000;
585 		itp->it_value.tv_sec--;
586 	}
587 	itp->it_value.tv_usec -= usec;
588 	usec = 0;
589 	if (timevalisset(&itp->it_value))
590 		return (1);
591 	/* expired, exactly at end of interval */
592 expire:
593 	if (timevalisset(&itp->it_interval)) {
594 		itp->it_value = itp->it_interval;
595 		itp->it_value.tv_usec -= usec;
596 		if (itp->it_value.tv_usec < 0) {
597 			itp->it_value.tv_usec += 1000000;
598 			itp->it_value.tv_sec--;
599 		}
600 	} else
601 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
602 	return (0);
603 }
604 
605 /*
606  * Add and subtract routines for timevals.
607  * N.B.: subtract routine doesn't deal with
608  * results which are before the beginning,
609  * it just gets very confused in this case.
610  * Caveat emptor.
611  */
612 void
613 timevaladd(t1, t2)
614 	struct timeval *t1, *t2;
615 {
616 
617 	t1->tv_sec += t2->tv_sec;
618 	t1->tv_usec += t2->tv_usec;
619 	timevalfix(t1);
620 }
621 
622 void
623 timevalsub(t1, t2)
624 	struct timeval *t1, *t2;
625 {
626 
627 	t1->tv_sec -= t2->tv_sec;
628 	t1->tv_usec -= t2->tv_usec;
629 	timevalfix(t1);
630 }
631 
632 static void
633 timevalfix(t1)
634 	struct timeval *t1;
635 {
636 
637 	if (t1->tv_usec < 0) {
638 		t1->tv_sec--;
639 		t1->tv_usec += 1000000;
640 	}
641 	if (t1->tv_usec >= 1000000) {
642 		t1->tv_sec++;
643 		t1->tv_usec -= 1000000;
644 	}
645 }
646