2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
33 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93
34 * $FreeBSD: src/sys/kern/kern_time.c,v 1.68.2.1 2002/10/01 08:00:41 bde Exp $
35 * $DragonFly: src/sys/kern/kern_time.c,v 1.10 2003/08/26 21:09:02 rob Exp $
38 #include <sys/param.h>
39 #include <sys/systm.h>
41 #include <sys/sysproto.h>
42 #include <sys/resourcevar.h>
43 #include <sys/signalvar.h>
44 #include <sys/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/sysent.h>
49 #include <sys/vnode.h>
51 #include <vm/vm_extern.h>
52 #include <sys/msgport2.h>
57 * Time of day and interval timer support.
59 * These routines provide the kernel entry points to get and set
60 * the time-of-day and per-process interval timers. Subroutines
61 * here provide support for adding and subtracting timeval structures
62 * and decrementing interval timers, optionally reloading the interval
63 * timers when they expire.
66 static int nanosleep1 (struct timespec *rqt,
67 struct timespec *rmt);
68 static int settime (struct timeval *);
69 static void timevalfix (struct timeval *);
70 static void no_lease_updatetime (int);
73 no_lease_updatetime(deltat)
78 void (*lease_updatetime) (int) = no_lease_updatetime;
84 struct timeval delta, tv1, tv2;
85 static struct timeval maxtime, laststep;
92 timevalsub(&delta, &tv1);
95 * If the system is secure, we do not allow the time to be
96 * set to a value earlier than 1 second less than the highest
97 * time we have yet seen. The worst a miscreant can do in
98 * this circumstance is "freeze" time. He couldn't go
101 * We similarly do not allow the clock to be stepped more
102 * than one second, nor more than once per second. This allows
103 * a miscreant to make the clock march double-time, but no worse.
105 if (securelevel > 1) {
106 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
108 * Update maxtime to latest time we've seen.
110 if (tv1.tv_sec > maxtime.tv_sec)
113 timevalsub(&tv2, &maxtime);
114 if (tv2.tv_sec < -1) {
115 tv->tv_sec = maxtime.tv_sec - 1;
116 printf("Time adjustment clamped to -1 second\n");
119 if (tv1.tv_sec == laststep.tv_sec) {
123 if (delta.tv_sec > 1) {
124 tv->tv_sec = tv1.tv_sec + 1;
125 printf("Time adjustment clamped to +1 second\n");
131 ts.tv_sec = tv->tv_sec;
132 ts.tv_nsec = tv->tv_usec * 1000;
133 set_timecounter(&ts);
134 (void) splsoftclock();
135 lease_updatetime(delta.tv_sec);
143 clock_gettime(struct clock_gettime_args *uap)
147 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
150 return (copyout(&ats, SCARG(uap, tp), sizeof(ats)));
155 clock_settime(struct clock_settime_args *uap)
157 struct thread *td = curthread;
162 if ((error = suser(td)) != 0)
164 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
166 if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
168 if (ats.tv_nsec < 0 || ats.tv_nsec >= 1000000000)
170 /* XXX Don't convert nsec->usec and back */
171 TIMESPEC_TO_TIMEVAL(&atv, &ats);
172 if ((error = settime(&atv)))
178 clock_getres(struct clock_getres_args *uap)
183 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
186 if (SCARG(uap, tp)) {
189 * Round up the result of the division cheaply by adding 1.
190 * Rounding up is especially important if rounding down
191 * would give 0. Perfect rounding is unimportant.
193 ts.tv_nsec = 1000000000 / timecounter->tc_frequency + 1;
194 error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
202 nanosleep1(struct timespec *rqt, struct timespec *rmt)
204 struct timespec ts, ts2, ts3;
208 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
210 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
213 timespecadd(&ts, rqt);
214 TIMESPEC_TO_TIMEVAL(&tv, rqt);
216 error = tsleep(&nanowait, PCATCH, "nanslp",
219 if (error != EWOULDBLOCK) {
220 if (error == ERESTART)
223 timespecsub(&ts, &ts2);
230 if (timespeccmp(&ts2, &ts, >=))
233 timespecsub(&ts3, &ts2);
234 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
238 static void nanosleep_done(void *arg);
239 static void nanosleep_return(lwkt_port_t port, lwkt_msg_t msg);
243 nanosleep(struct nanosleep_args *uap)
246 struct sysmsg_sleep *sysmsg = &uap->sysmsg.sm_sleep;
248 error = copyin(uap->rqtp, &sysmsg->rqt, sizeof(sysmsg->rqt));
252 * YYY clean this up to always use the callout, note that an abort
253 * implementation should record the residual in the async case.
255 if (sysmsg->lmsg.ms_flags & MSGF_ASYNC) {
258 ticks = (quad_t)sysmsg->rqt.tv_nsec * hz / 1000000000LL;
259 if (sysmsg->rqt.tv_sec)
260 ticks += (quad_t)sysmsg->rqt.tv_sec * hz;
267 sysmsg->lmsg.ms_cleanupmsg = nanosleep_return;
268 callout_init(&sysmsg->timer);
269 callout_reset(&sysmsg->timer, ticks, nanosleep_done, uap);
274 * Old synchronous sleep code, copyout the residual if
275 * nanosleep was interrupted.
277 error = nanosleep1(&sysmsg->rqt, &sysmsg->rmt);
278 if (error && SCARG(uap, rmtp))
279 error = copyout(&sysmsg->rmt, SCARG(uap, rmtp), sizeof(sysmsg->rmt));
285 * Asynch completion for the nanosleep() syscall. This function may be
286 * called from any context and cannot legally access the originating
287 * thread, proc, or its user space.
289 * YYY change the callout interface API so we can simply assign the replymsg
290 * function to it directly.
293 nanosleep_done(void *arg)
295 struct nanosleep_args *uap = arg;
297 lwkt_replymsg(&uap->sysmsg.lmsg, 0);
301 * Asynch return for the nanosleep() syscall, called in the context of the
302 * originating thread when it pulls the message off the reply port. This
303 * function is responsible for any copyouts to userland. Kernel threads
304 * which do their own internal system calls will not usually call the return
308 nanosleep_return(lwkt_port_t port, lwkt_msg_t msg)
310 struct nanosleep_args *uap = (void *)msg;
311 struct sysmsg_sleep *sysmsg = &uap->sysmsg.sm_sleep;
313 if (sysmsg->lmsg.ms_error && uap->rmtp) {
314 sysmsg->lmsg.ms_error =
315 copyout(&sysmsg->rmt, uap->rmtp, sizeof(sysmsg->rmt));
321 gettimeofday(struct gettimeofday_args *uap)
328 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
333 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
340 settimeofday(struct settimeofday_args *uap)
342 struct thread *td = curthread;
347 if ((error = suser(td)))
349 /* Verify all parameters before changing time. */
351 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
354 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
358 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
360 if (uap->tv && (error = settime(&atv)))
367 int tickdelta; /* current clock skew, us. per tick */
368 long timedelta; /* unapplied time correction, us. */
369 static long bigadj = 1000000; /* use 10x skew above bigadj us. */
373 adjtime(struct adjtime_args *uap)
375 struct thread *td = curthread;
377 long ndelta, ntickdelta, odelta;
380 if ((error = suser(td)))
383 copyin((caddr_t)uap->delta, (caddr_t)&atv, sizeof(struct timeval))))
387 * Compute the total correction and the rate at which to apply it.
388 * Round the adjustment down to a whole multiple of the per-tick
389 * delta, so that after some number of incremental changes in
390 * hardclock(), tickdelta will become zero, lest the correction
391 * overshoot and start taking us away from the desired final time.
393 ndelta = atv.tv_sec * 1000000 + atv.tv_usec;
394 if (ndelta > bigadj || ndelta < -bigadj)
395 ntickdelta = 10 * tickadj;
397 ntickdelta = tickadj;
398 if (ndelta % ntickdelta)
399 ndelta = ndelta / ntickdelta * ntickdelta;
402 * To make hardclock()'s job easier, make the per-tick delta negative
403 * if we want time to run slower; then hardclock can simply compute
404 * tick + tickdelta, and subtract tickdelta from timedelta.
407 ntickdelta = -ntickdelta;
411 tickdelta = ntickdelta;
415 atv.tv_sec = odelta / 1000000;
416 atv.tv_usec = odelta % 1000000;
417 (void) copyout((caddr_t)&atv, (caddr_t)uap->olddelta,
418 sizeof(struct timeval));
424 * Get value of an interval timer. The process virtual and
425 * profiling virtual time timers are kept in the p_stats area, since
426 * they can be swapped out. These are kept internally in the
427 * way they are specified externally: in time until they expire.
429 * The real time interval timer is kept in the process table slot
430 * for the process, and its value (it_value) is kept as an
431 * absolute time rather than as a delta, so that it is easy to keep
432 * periodic real-time signals from drifting.
434 * Virtual time timers are processed in the hardclock() routine of
435 * kern_clock.c. The real time timer is processed by a timeout
436 * routine, called from the softclock() routine. Since a callout
437 * may be delayed in real time due to interrupt processing in the system,
438 * it is possible for the real time timeout routine (realitexpire, given below),
439 * to be delayed in real time past when it is supposed to occur. It
440 * does not suffice, therefore, to reload the real timer .it_value from the
441 * real time timers .it_interval. Rather, we compute the next time in
442 * absolute time the timer should go off.
446 getitimer(struct getitimer_args *uap)
448 struct proc *p = curproc;
450 struct itimerval aitv;
453 if (uap->which > ITIMER_PROF)
455 s = splclock(); /* XXX still needed ? */
456 if (uap->which == ITIMER_REAL) {
458 * Convert from absolute to relative time in .it_value
459 * part of real time timer. If time for real time timer
460 * has passed return 0, else return difference between
461 * current time and time for the timer to go off.
463 aitv = p->p_realtimer;
464 if (timevalisset(&aitv.it_value)) {
465 getmicrouptime(&ctv);
466 if (timevalcmp(&aitv.it_value, &ctv, <))
467 timevalclear(&aitv.it_value);
469 timevalsub(&aitv.it_value, &ctv);
472 aitv = p->p_stats->p_timer[uap->which];
474 return (copyout((caddr_t)&aitv, (caddr_t)uap->itv,
475 sizeof (struct itimerval)));
480 setitimer(struct setitimer_args *uap)
482 struct itimerval aitv;
484 struct itimerval *itvp;
485 struct proc *p = curproc;
488 if (uap->which > ITIMER_PROF)
491 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
492 sizeof(struct itimerval))))
494 if ((uap->itv = uap->oitv) &&
495 (error = getitimer((struct getitimer_args *)uap)))
499 if (itimerfix(&aitv.it_value))
501 if (!timevalisset(&aitv.it_value))
502 timevalclear(&aitv.it_interval);
503 else if (itimerfix(&aitv.it_interval))
505 s = splclock(); /* XXX: still needed ? */
506 if (uap->which == ITIMER_REAL) {
507 if (timevalisset(&p->p_realtimer.it_value))
508 untimeout(realitexpire, (caddr_t)p, p->p_ithandle);
509 if (timevalisset(&aitv.it_value))
510 p->p_ithandle = timeout(realitexpire, (caddr_t)p,
511 tvtohz(&aitv.it_value));
512 getmicrouptime(&ctv);
513 timevaladd(&aitv.it_value, &ctv);
514 p->p_realtimer = aitv;
516 p->p_stats->p_timer[uap->which] = aitv;
522 * Real interval timer expired:
523 * send process whose timer expired an alarm signal.
524 * If time is not set up to reload, then just return.
525 * Else compute next time timer should go off which is > current time.
526 * This is where delay in processing this timeout causes multiple
527 * SIGALRM calls to be compressed into one.
528 * tvtohz() always adds 1 to allow for the time until the next clock
529 * interrupt being strictly less than 1 clock tick, but we don't want
530 * that here since we want to appear to be in sync with the clock
531 * interrupt even when we're delayed.
538 struct timeval ctv, ntv;
541 p = (struct proc *)arg;
543 if (!timevalisset(&p->p_realtimer.it_interval)) {
544 timevalclear(&p->p_realtimer.it_value);
548 s = splclock(); /* XXX: still neeeded ? */
549 timevaladd(&p->p_realtimer.it_value,
550 &p->p_realtimer.it_interval);
551 getmicrouptime(&ctv);
552 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
553 ntv = p->p_realtimer.it_value;
554 timevalsub(&ntv, &ctv);
555 p->p_ithandle = timeout(realitexpire, (caddr_t)p,
565 * Check that a proposed value to load into the .it_value or
566 * .it_interval part of an interval timer is acceptable, and
567 * fix it to have at least minimal value (i.e. if it is less
568 * than the resolution of the clock, round it up.)
575 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 ||
576 tv->tv_usec < 0 || tv->tv_usec >= 1000000)
578 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
584 * Decrement an interval timer by a specified number
585 * of microseconds, which must be less than a second,
586 * i.e. < 1000000. If the timer expires, then reload
587 * it. In this case, carry over (usec - old value) to
588 * reduce the value reloaded into the timer so that
589 * the timer does not drift. This routine assumes
590 * that it is called in a context where the timers
591 * on which it is operating cannot change in value.
594 itimerdecr(itp, usec)
595 struct itimerval *itp;
599 if (itp->it_value.tv_usec < usec) {
600 if (itp->it_value.tv_sec == 0) {
601 /* expired, and already in next interval */
602 usec -= itp->it_value.tv_usec;
605 itp->it_value.tv_usec += 1000000;
606 itp->it_value.tv_sec--;
608 itp->it_value.tv_usec -= usec;
610 if (timevalisset(&itp->it_value))
612 /* expired, exactly at end of interval */
614 if (timevalisset(&itp->it_interval)) {
615 itp->it_value = itp->it_interval;
616 itp->it_value.tv_usec -= usec;
617 if (itp->it_value.tv_usec < 0) {
618 itp->it_value.tv_usec += 1000000;
619 itp->it_value.tv_sec--;
622 itp->it_value.tv_usec = 0; /* sec is already 0 */
627 * Add and subtract routines for timevals.
628 * N.B.: subtract routine doesn't deal with
629 * results which are before the beginning,
630 * it just gets very confused in this case.
635 struct timeval *t1, *t2;
638 t1->tv_sec += t2->tv_sec;
639 t1->tv_usec += t2->tv_usec;
645 struct timeval *t1, *t2;
648 t1->tv_sec -= t2->tv_sec;
649 t1->tv_usec -= t2->tv_usec;
658 if (t1->tv_usec < 0) {
660 t1->tv_usec += 1000000;
662 if (t1->tv_usec >= 1000000) {
664 t1->tv_usec -= 1000000;
669 * ratecheck(): simple time-based rate-limit checking.
672 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
674 struct timeval tv, delta;
677 getmicrouptime(&tv); /* NB: 10ms precision */
679 timevalsub(&delta, lasttime);
682 * check for 0,0 is so that the message will be seen at least once,
683 * even if interval is huge.
685 if (timevalcmp(&delta, mininterval, >=) ||
686 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
695 * ppsratecheck(): packets (or events) per second limitation.
697 * Return 0 if the limit is to be enforced (e.g. the caller
698 * should drop a packet because of the rate limitation).
700 * maxpps of 0 always causes zero to be returned. maxpps of -1
701 * always causes 1 to be returned; this effectively defeats rate
704 * Note that we maintain the struct timeval for compatibility
705 * with other bsd systems. We reuse the storage and just monitor
706 * clock ticks for minimal overhead.
709 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
714 * Reset the last time and counter if this is the first call
715 * or more than a second has passed since the last update of
719 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
720 lasttime->tv_sec = now;
722 return (maxpps != 0);
724 (*curpps)++; /* NB: ignore potential overflow */
725 return (maxpps < 0 || *curpps < maxpps);