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 1.13 2004/01/07 11:08:06 dillon 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>
47 #include <sys/sysunion.h>
50 #include <sys/vnode.h>
51 #include <sys/sysctl.h>
53 #include <vm/vm_extern.h>
54 #include <sys/msgport2.h>
59 * Time of day and interval timer support.
61 * These routines provide the kernel entry points to get and set
62 * the time-of-day and per-process interval timers. Subroutines
63 * here provide support for adding and subtracting timeval structures
64 * and decrementing interval timers, optionally reloading the interval
65 * timers when they expire.
68 static int nanosleep1 (struct timespec *rqt,
69 struct timespec *rmt);
70 static int settime (struct timeval *);
71 static void timevalfix (struct timeval *);
72 static void no_lease_updatetime (int);
74 static int sleep_hardloop = 0;
75 SYSCTL_INT(_kern, OID_AUTO, sleep_hardloop, CTLFLAG_RW, &sleep_hardloop, 0, "");
78 no_lease_updatetime(deltat)
83 void (*lease_updatetime) (int) = no_lease_updatetime;
89 struct timeval delta, tv1, tv2;
90 static struct timeval maxtime, laststep;
97 timevalsub(&delta, &tv1);
100 * If the system is secure, we do not allow the time to be
101 * set to a value earlier than 1 second less than the highest
102 * time we have yet seen. The worst a miscreant can do in
103 * this circumstance is "freeze" time. He couldn't go
106 * We similarly do not allow the clock to be stepped more
107 * than one second, nor more than once per second. This allows
108 * a miscreant to make the clock march double-time, but no worse.
110 if (securelevel > 1) {
111 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
113 * Update maxtime to latest time we've seen.
115 if (tv1.tv_sec > maxtime.tv_sec)
118 timevalsub(&tv2, &maxtime);
119 if (tv2.tv_sec < -1) {
120 tv->tv_sec = maxtime.tv_sec - 1;
121 printf("Time adjustment clamped to -1 second\n");
124 if (tv1.tv_sec == laststep.tv_sec) {
128 if (delta.tv_sec > 1) {
129 tv->tv_sec = tv1.tv_sec + 1;
130 printf("Time adjustment clamped to +1 second\n");
136 ts.tv_sec = tv->tv_sec;
137 ts.tv_nsec = tv->tv_usec * 1000;
138 set_timecounter(&ts);
139 (void) splsoftclock();
140 lease_updatetime(delta.tv_sec);
148 clock_gettime(struct clock_gettime_args *uap)
152 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
155 return (copyout(&ats, SCARG(uap, tp), sizeof(ats)));
160 clock_settime(struct clock_settime_args *uap)
162 struct thread *td = curthread;
167 if ((error = suser(td)) != 0)
169 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
171 if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
173 if (ats.tv_nsec < 0 || ats.tv_nsec >= 1000000000)
175 /* XXX Don't convert nsec->usec and back */
176 TIMESPEC_TO_TIMEVAL(&atv, &ats);
177 if ((error = settime(&atv)))
183 clock_getres(struct clock_getres_args *uap)
188 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
191 if (SCARG(uap, tp)) {
194 * Round up the result of the division cheaply by adding 1.
195 * Rounding up is especially important if rounding down
196 * would give 0. Perfect rounding is unimportant.
198 ts.tv_nsec = 1000000000 / timecounter->tc_frequency + 1;
199 error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
207 nanosleep1(struct timespec *rqt, struct timespec *rmt)
209 struct timespec ts, ts2, ts3;
213 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
215 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
218 timespecadd(&ts, rqt); /* ts = target timestamp compare */
219 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */
222 * If hard looping is allowed and the interval is too short,
223 * hard loop with a yield, otherwise sleep with a conservative
224 * tick count. In normal mode sleep with one extra tick count
225 * which will be sufficient for most sleep values. If it
226 * isn't sufficient in normal mode we will wind up doing an
229 * sleep_hardloop = 0 Normal mode
230 * sleep_hardloop = 1 Strict hard loop
231 * sleep_hardloop = 2 Hard loop on < 1 tick requests only
233 int ticks = tvtohz_low(&tv);
235 if (sleep_hardloop) {
238 error = iscaught(curproc);
240 error = tsleep(&nanowait, PCATCH, "nanslp",
241 ticks + sleep_hardloop - 1);
244 error = tsleep(&nanowait, PCATCH, "nanslp", ticks + 1);
247 if (error != EWOULDBLOCK) {
248 if (error == ERESTART)
251 timespecsub(&ts, &ts2);
258 if (timespeccmp(&ts2, &ts, >=))
261 timespecsub(&ts3, &ts2);
262 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
266 static void nanosleep_done(void *arg);
267 static void nanosleep_copyout(union sysunion *sysun);
271 nanosleep(struct nanosleep_args *uap)
274 struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep;
276 error = copyin(uap->rqtp, &smsleep->rqt, sizeof(smsleep->rqt));
280 * YYY clean this up to always use the callout, note that an abort
281 * implementation should record the residual in the async case.
283 if (uap->sysmsg.lmsg.ms_flags & MSGF_ASYNC) {
286 ticks = (quad_t)smsleep->rqt.tv_nsec * hz / 1000000000LL;
287 if (smsleep->rqt.tv_sec)
288 ticks += (quad_t)smsleep->rqt.tv_sec * hz;
295 uap->sysmsg.copyout = nanosleep_copyout;
296 callout_init(&smsleep->timer);
297 callout_reset(&smsleep->timer, ticks, nanosleep_done, uap);
302 * Old synchronous sleep code, copyout the residual if
303 * nanosleep was interrupted.
305 error = nanosleep1(&smsleep->rqt, &smsleep->rmt);
306 if (error && SCARG(uap, rmtp))
307 error = copyout(&smsleep->rmt, SCARG(uap, rmtp), sizeof(smsleep->rmt));
313 * Asynch completion for the nanosleep() syscall. This function may be
314 * called from any context and cannot legally access the originating
315 * thread, proc, or its user space.
317 * YYY change the callout interface API so we can simply assign the replymsg
318 * function to it directly.
321 nanosleep_done(void *arg)
323 struct nanosleep_args *uap = arg;
325 lwkt_replymsg(&uap->sysmsg.lmsg, 0);
329 * Asynch return for the nanosleep() syscall, called in the context of the
330 * originating thread when it pulls the message off the reply port. This
331 * function is responsible for any copyouts to userland. Kernel threads
332 * which do their own internal system calls will not usually call the return
336 nanosleep_copyout(union sysunion *sysun)
338 struct nanosleep_args *uap = &sysun->nanosleep;
339 struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep;
341 if (sysun->lmsg.ms_error && uap->rmtp) {
342 sysun->lmsg.ms_error =
343 copyout(&smsleep->rmt, uap->rmtp, sizeof(smsleep->rmt));
349 gettimeofday(struct gettimeofday_args *uap)
356 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
361 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
368 settimeofday(struct settimeofday_args *uap)
370 struct thread *td = curthread;
375 if ((error = suser(td)))
377 /* Verify all parameters before changing time. */
379 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
382 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
386 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
388 if (uap->tv && (error = settime(&atv)))
395 int tickdelta; /* current clock skew, us. per tick */
396 long timedelta; /* unapplied time correction, us. */
397 static long bigadj = 1000000; /* use 10x skew above bigadj us. */
401 adjtime(struct adjtime_args *uap)
403 struct thread *td = curthread;
405 long ndelta, ntickdelta, odelta;
408 if ((error = suser(td)))
411 copyin((caddr_t)uap->delta, (caddr_t)&atv, sizeof(struct timeval))))
415 * Compute the total correction and the rate at which to apply it.
416 * Round the adjustment down to a whole multiple of the per-tick
417 * delta, so that after some number of incremental changes in
418 * hardclock(), tickdelta will become zero, lest the correction
419 * overshoot and start taking us away from the desired final time.
421 ndelta = atv.tv_sec * 1000000 + atv.tv_usec;
422 if (ndelta > bigadj || ndelta < -bigadj)
423 ntickdelta = 10 * tickadj;
425 ntickdelta = tickadj;
426 if (ndelta % ntickdelta)
427 ndelta = ndelta / ntickdelta * ntickdelta;
430 * To make hardclock()'s job easier, make the per-tick delta negative
431 * if we want time to run slower; then hardclock can simply compute
432 * tick + tickdelta, and subtract tickdelta from timedelta.
435 ntickdelta = -ntickdelta;
439 tickdelta = ntickdelta;
443 atv.tv_sec = odelta / 1000000;
444 atv.tv_usec = odelta % 1000000;
445 (void) copyout((caddr_t)&atv, (caddr_t)uap->olddelta,
446 sizeof(struct timeval));
452 * Get value of an interval timer. The process virtual and
453 * profiling virtual time timers are kept in the p_stats area, since
454 * they can be swapped out. These are kept internally in the
455 * way they are specified externally: in time until they expire.
457 * The real time interval timer is kept in the process table slot
458 * for the process, and its value (it_value) is kept as an
459 * absolute time rather than as a delta, so that it is easy to keep
460 * periodic real-time signals from drifting.
462 * Virtual time timers are processed in the hardclock() routine of
463 * kern_clock.c. The real time timer is processed by a timeout
464 * routine, called from the softclock() routine. Since a callout
465 * may be delayed in real time due to interrupt processing in the system,
466 * it is possible for the real time timeout routine (realitexpire, given below),
467 * to be delayed in real time past when it is supposed to occur. It
468 * does not suffice, therefore, to reload the real timer .it_value from the
469 * real time timers .it_interval. Rather, we compute the next time in
470 * absolute time the timer should go off.
474 getitimer(struct getitimer_args *uap)
476 struct proc *p = curproc;
478 struct itimerval aitv;
481 if (uap->which > ITIMER_PROF)
483 s = splclock(); /* XXX still needed ? */
484 if (uap->which == ITIMER_REAL) {
486 * Convert from absolute to relative time in .it_value
487 * part of real time timer. If time for real time timer
488 * has passed return 0, else return difference between
489 * current time and time for the timer to go off.
491 aitv = p->p_realtimer;
492 if (timevalisset(&aitv.it_value)) {
493 getmicrouptime(&ctv);
494 if (timevalcmp(&aitv.it_value, &ctv, <))
495 timevalclear(&aitv.it_value);
497 timevalsub(&aitv.it_value, &ctv);
500 aitv = p->p_stats->p_timer[uap->which];
502 return (copyout((caddr_t)&aitv, (caddr_t)uap->itv,
503 sizeof (struct itimerval)));
508 setitimer(struct setitimer_args *uap)
510 struct itimerval aitv;
512 struct itimerval *itvp;
513 struct proc *p = curproc;
516 if (uap->which > ITIMER_PROF)
519 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
520 sizeof(struct itimerval))))
522 if ((uap->itv = uap->oitv) &&
523 (error = getitimer((struct getitimer_args *)uap)))
527 if (itimerfix(&aitv.it_value))
529 if (!timevalisset(&aitv.it_value))
530 timevalclear(&aitv.it_interval);
531 else if (itimerfix(&aitv.it_interval))
533 s = splclock(); /* XXX: still needed ? */
534 if (uap->which == ITIMER_REAL) {
535 if (timevalisset(&p->p_realtimer.it_value))
536 untimeout(realitexpire, (caddr_t)p, p->p_ithandle);
537 if (timevalisset(&aitv.it_value))
538 p->p_ithandle = timeout(realitexpire, (caddr_t)p,
539 tvtohz_high(&aitv.it_value));
540 getmicrouptime(&ctv);
541 timevaladd(&aitv.it_value, &ctv);
542 p->p_realtimer = aitv;
544 p->p_stats->p_timer[uap->which] = aitv;
550 * Real interval timer expired:
551 * send process whose timer expired an alarm signal.
552 * If time is not set up to reload, then just return.
553 * Else compute next time timer should go off which is > current time.
554 * This is where delay in processing this timeout causes multiple
555 * SIGALRM calls to be compressed into one.
556 * tvtohz_high() always adds 1 to allow for the time until the next clock
557 * interrupt being strictly less than 1 clock tick, but we don't want
558 * that here since we want to appear to be in sync with the clock
559 * interrupt even when we're delayed.
566 struct timeval ctv, ntv;
569 p = (struct proc *)arg;
571 if (!timevalisset(&p->p_realtimer.it_interval)) {
572 timevalclear(&p->p_realtimer.it_value);
576 s = splclock(); /* XXX: still neeeded ? */
577 timevaladd(&p->p_realtimer.it_value,
578 &p->p_realtimer.it_interval);
579 getmicrouptime(&ctv);
580 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
581 ntv = p->p_realtimer.it_value;
582 timevalsub(&ntv, &ctv);
583 p->p_ithandle = timeout(realitexpire, (caddr_t)p,
593 * Check that a proposed value to load into the .it_value or
594 * .it_interval part of an interval timer is acceptable, and
595 * fix it to have at least minimal value (i.e. if it is less
596 * than the resolution of the clock, round it up.)
603 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 ||
604 tv->tv_usec < 0 || tv->tv_usec >= 1000000)
606 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
612 * Decrement an interval timer by a specified number
613 * of microseconds, which must be less than a second,
614 * i.e. < 1000000. If the timer expires, then reload
615 * it. In this case, carry over (usec - old value) to
616 * reduce the value reloaded into the timer so that
617 * the timer does not drift. This routine assumes
618 * that it is called in a context where the timers
619 * on which it is operating cannot change in value.
622 itimerdecr(itp, usec)
623 struct itimerval *itp;
627 if (itp->it_value.tv_usec < usec) {
628 if (itp->it_value.tv_sec == 0) {
629 /* expired, and already in next interval */
630 usec -= itp->it_value.tv_usec;
633 itp->it_value.tv_usec += 1000000;
634 itp->it_value.tv_sec--;
636 itp->it_value.tv_usec -= usec;
638 if (timevalisset(&itp->it_value))
640 /* expired, exactly at end of interval */
642 if (timevalisset(&itp->it_interval)) {
643 itp->it_value = itp->it_interval;
644 itp->it_value.tv_usec -= usec;
645 if (itp->it_value.tv_usec < 0) {
646 itp->it_value.tv_usec += 1000000;
647 itp->it_value.tv_sec--;
650 itp->it_value.tv_usec = 0; /* sec is already 0 */
655 * Add and subtract routines for timevals.
656 * N.B.: subtract routine doesn't deal with
657 * results which are before the beginning,
658 * it just gets very confused in this case.
663 struct timeval *t1, *t2;
666 t1->tv_sec += t2->tv_sec;
667 t1->tv_usec += t2->tv_usec;
673 struct timeval *t1, *t2;
676 t1->tv_sec -= t2->tv_sec;
677 t1->tv_usec -= t2->tv_usec;
686 if (t1->tv_usec < 0) {
688 t1->tv_usec += 1000000;
690 if (t1->tv_usec >= 1000000) {
692 t1->tv_usec -= 1000000;
697 * ratecheck(): simple time-based rate-limit checking.
700 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
702 struct timeval tv, delta;
705 getmicrouptime(&tv); /* NB: 10ms precision */
707 timevalsub(&delta, lasttime);
710 * check for 0,0 is so that the message will be seen at least once,
711 * even if interval is huge.
713 if (timevalcmp(&delta, mininterval, >=) ||
714 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
723 * ppsratecheck(): packets (or events) per second limitation.
725 * Return 0 if the limit is to be enforced (e.g. the caller
726 * should drop a packet because of the rate limitation).
728 * maxpps of 0 always causes zero to be returned. maxpps of -1
729 * always causes 1 to be returned; this effectively defeats rate
732 * Note that we maintain the struct timeval for compatibility
733 * with other bsd systems. We reuse the storage and just monitor
734 * clock ticks for minimal overhead.
737 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
742 * Reset the last time and counter if this is the first call
743 * or more than a second has passed since the last update of
747 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
748 lasttime->tv_sec = now;
750 return (maxpps != 0);
752 (*curpps)++; /* NB: ignore potential overflow */
753 return (maxpps < 0 || *curpps < maxpps);