xref: /dflybsd-src/sys/kern/kern_time.c (revision b721b9ee4205a28b1f2d01b5c95851add56bf82b)
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  * $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 1.18 2005/03/15 01:12:23 dillon Exp $
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/buf.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>
48 #include <sys/proc.h>
49 #include <sys/time.h>
50 #include <sys/vnode.h>
51 #include <sys/sysctl.h>
52 #include <vm/vm.h>
53 #include <vm/vm_extern.h>
54 #include <sys/msgport2.h>
55 #include <sys/thread2.h>
56 
57 struct timezone tz;
58 
59 /*
60  * Time of day and interval timer support.
61  *
62  * These routines provide the kernel entry points to get and set
63  * the time-of-day and per-process interval timers.  Subroutines
64  * here provide support for adding and subtracting timeval structures
65  * and decrementing interval timers, optionally reloading the interval
66  * timers when they expire.
67  */
68 
69 static int	nanosleep1 (struct timespec *rqt,
70 		    struct timespec *rmt);
71 static int	settime (struct timeval *);
72 static void	timevalfix (struct timeval *);
73 static void	no_lease_updatetime (int);
74 
75 static int     sleep_hard_us = 100;
76 SYSCTL_INT(_kern, OID_AUTO, sleep_hard_us, CTLFLAG_RW, &sleep_hard_us, 0, "")
77 
78 static void
79 no_lease_updatetime(deltat)
80 	int deltat;
81 {
82 }
83 
84 void (*lease_updatetime) (int)  = no_lease_updatetime;
85 
86 static int
87 settime(tv)
88 	struct timeval *tv;
89 {
90 	struct timeval delta, tv1, tv2;
91 	static struct timeval maxtime, laststep;
92 	struct timespec ts;
93 
94 	crit_enter();
95 	microtime(&tv1);
96 	delta = *tv;
97 	timevalsub(&delta, &tv1);
98 
99 	/*
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
104 	 * back to the past.
105 	 *
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.
109 	 */
110 	if (securelevel > 1) {
111 		if (delta.tv_sec < 0 || delta.tv_usec < 0) {
112 			/*
113 			 * Update maxtime to latest time we've seen.
114 			 */
115 			if (tv1.tv_sec > maxtime.tv_sec)
116 				maxtime = tv1;
117 			tv2 = *tv;
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");
122 			}
123 		} else {
124 			if (tv1.tv_sec == laststep.tv_sec) {
125 				crit_exit();
126 				return (EPERM);
127 			}
128 			if (delta.tv_sec > 1) {
129 				tv->tv_sec = tv1.tv_sec + 1;
130 				printf("Time adjustment clamped to +1 second\n");
131 			}
132 			laststep = *tv;
133 		}
134 	}
135 
136 	ts.tv_sec = tv->tv_sec;
137 	ts.tv_nsec = tv->tv_usec * 1000;
138 	set_timeofday(&ts);
139 	lease_updatetime(delta.tv_sec);
140 	crit_exit();
141 	resettodr();
142 	return (0);
143 }
144 
145 /* ARGSUSED */
146 int
147 clock_gettime(struct clock_gettime_args *uap)
148 {
149 	struct timespec ats;
150 
151 	switch(uap->clock_id) {
152 	case CLOCK_REALTIME:
153 		nanotime(&ats);
154 		return (copyout(&ats, uap->tp, sizeof(ats)));
155 	case CLOCK_MONOTONIC:
156 		nanouptime(&ats);
157 		return (copyout(&ats, uap->tp, sizeof(ats)));
158 	default:
159 		return (EINVAL);
160 	}
161 }
162 
163 /* ARGSUSED */
164 int
165 clock_settime(struct clock_settime_args *uap)
166 {
167 	struct thread *td = curthread;
168 	struct timeval atv;
169 	struct timespec ats;
170 	int error;
171 
172 	if ((error = suser(td)) != 0)
173 		return (error);
174 	switch(uap->clock_id) {
175 	case CLOCK_REALTIME:
176 		if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
177 			return (error);
178 		if (ats.tv_nsec < 0 || ats.tv_nsec >= 1000000000)
179 			return (EINVAL);
180 		/* XXX Don't convert nsec->usec and back */
181 		TIMESPEC_TO_TIMEVAL(&atv, &ats);
182 		error = settime(&atv);
183 		return (error);
184 	default:
185 		return (EINVAL);
186 	}
187 }
188 
189 int
190 clock_getres(struct clock_getres_args *uap)
191 {
192 	struct timespec ts;
193 
194 	switch(uap->clock_id) {
195 	case CLOCK_REALTIME:
196 	case CLOCK_MONOTONIC:
197 		/*
198 		 * Round up the result of the division cheaply
199 		 * by adding 1.  Rounding up is especially important
200 		 * if rounding down would give 0.  Perfect rounding
201 		 * is unimportant.
202 		 */
203 		ts.tv_sec = 0;
204 		ts.tv_nsec = 1000000000 / cputimer_freq + 1;
205 		return(copyout(&ts, uap->tp, sizeof(ts)));
206 	default:
207 		return(EINVAL);
208 	}
209 }
210 
211 /*
212  * nanosleep1()
213  *
214  *	This is a general helper function for nanosleep() (aka sleep() aka
215  *	usleep()).
216  *
217  *	If there is less then one tick's worth of time left and
218  *	we haven't done a yield, or the remaining microseconds is
219  *	ridiculously low, do a yield.  This avoids having
220  *	to deal with systimer overheads when the system is under
221  *	heavy loads.  If we have done a yield already then use
222  *	a systimer and an uninterruptable thread wait.
223  *
224  *	If there is more then a tick's worth of time left,
225  *	calculate the baseline ticks and use an interruptable
226  *	tsleep, then handle the fine-grained delay on the next
227  *	loop.  This usually results in two sleeps occuring, a long one
228  *	and a short one.
229  */
230 static void
231 ns1_systimer(systimer_t info)
232 {
233 	lwkt_schedule(info->data);
234 }
235 
236 static int
237 nanosleep1(struct timespec *rqt, struct timespec *rmt)
238 {
239 	static int nanowait;
240 	struct timespec ts, ts2, ts3;
241 	struct timeval tv;
242 	int error;
243 	int tried_yield;
244 
245 	if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
246 		return (EINVAL);
247 	if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
248 		return (0);
249 	nanouptime(&ts);
250 	timespecadd(&ts, rqt);		/* ts = target timestamp compare */
251 	TIMESPEC_TO_TIMEVAL(&tv, rqt);	/* tv = sleep interval */
252 	tried_yield = 0;
253 
254 	for (;;) {
255 		int ticks;
256 		struct systimer info;
257 
258 		ticks = tv.tv_usec / tick;	/* approximate */
259 
260 		if (tv.tv_sec == 0 && ticks == 0) {
261 			thread_t td = curthread;
262 			if (tried_yield || tv.tv_usec < sleep_hard_us) {
263 				tried_yield = 0;
264 				uio_yield();
265 			} else {
266 				crit_enter_quick(td);
267 				systimer_init_oneshot(&info, ns1_systimer,
268 						td, tv.tv_usec);
269 				lwkt_deschedule_self(td);
270 				crit_exit_quick(td);
271 				lwkt_switch();
272 				systimer_del(&info); /* make sure it's gone */
273 			}
274 			error = iscaught(td->td_proc);
275 		} else if (tv.tv_sec == 0) {
276 			error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
277 		} else {
278 			ticks = tvtohz_low(&tv); /* also handles overflow */
279 			error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
280 		}
281 		nanouptime(&ts2);
282 		if (error && error != EWOULDBLOCK) {
283 			if (error == ERESTART)
284 				error = EINTR;
285 			if (rmt != NULL) {
286 				timespecsub(&ts, &ts2);
287 				if (ts.tv_sec < 0)
288 					timespecclear(&ts);
289 				*rmt = ts;
290 			}
291 			return (error);
292 		}
293 		if (timespeccmp(&ts2, &ts, >=))
294 			return (0);
295 		ts3 = ts;
296 		timespecsub(&ts3, &ts2);
297 		TIMESPEC_TO_TIMEVAL(&tv, &ts3);
298 	}
299 }
300 
301 static void nanosleep_done(void *arg);
302 static void nanosleep_copyout(union sysunion *sysun);
303 
304 /* ARGSUSED */
305 int
306 nanosleep(struct nanosleep_args *uap)
307 {
308 	int error;
309 	struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep;
310 
311 	error = copyin(uap->rqtp, &smsleep->rqt, sizeof(smsleep->rqt));
312 	if (error)
313 		return (error);
314 	/*
315 	 * YYY clean this up to always use the callout, note that an abort
316 	 * implementation should record the residual in the async case.
317 	 */
318 	if (uap->sysmsg.lmsg.ms_flags & MSGF_ASYNC) {
319 		quad_t ticks;
320 
321 		ticks = (quad_t)smsleep->rqt.tv_nsec * hz / 1000000000LL;
322 		if (smsleep->rqt.tv_sec)
323 			ticks += (quad_t)smsleep->rqt.tv_sec * hz;
324 		if (ticks <= 0) {
325 			if (ticks == 0)
326 				error = 0;
327 			else
328 				error = EINVAL;
329 		} else {
330 			uap->sysmsg.copyout = nanosleep_copyout;
331 			uap->sysmsg.lmsg.ms_flags &= ~MSGF_DONE;
332 			callout_init(&smsleep->timer);
333 			callout_reset(&smsleep->timer, ticks, nanosleep_done, uap);
334 			error = EASYNC;
335 		}
336 	} else {
337 		/*
338 		 * Old synchronous sleep code, copyout the residual if
339 		 * nanosleep was interrupted.
340 		 */
341 		error = nanosleep1(&smsleep->rqt, &smsleep->rmt);
342 		if (error && SCARG(uap, rmtp))
343 			error = copyout(&smsleep->rmt, SCARG(uap, rmtp), sizeof(smsleep->rmt));
344 	}
345 	return (error);
346 }
347 
348 /*
349  * Asynch completion for the nanosleep() syscall.  This function may be
350  * called from any context and cannot legally access the originating
351  * thread, proc, or its user space.
352  *
353  * YYY change the callout interface API so we can simply assign the replymsg
354  * function to it directly.
355  */
356 static void
357 nanosleep_done(void *arg)
358 {
359 	struct nanosleep_args *uap = arg;
360 	lwkt_msg_t msg = &uap->sysmsg.lmsg;
361 
362 	lwkt_replymsg(msg, 0);
363 }
364 
365 /*
366  * Asynch return for the nanosleep() syscall, called in the context of the
367  * originating thread when it pulls the message off the reply port.  This
368  * function is responsible for any copyouts to userland.  Kernel threads
369  * which do their own internal system calls will not usually call the return
370  * function.
371  */
372 static void
373 nanosleep_copyout(union sysunion *sysun)
374 {
375 	struct nanosleep_args *uap = &sysun->nanosleep;
376 	struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep;
377 
378 	if (sysun->lmsg.ms_error && uap->rmtp) {
379 		sysun->lmsg.ms_error =
380 		    copyout(&smsleep->rmt, uap->rmtp, sizeof(smsleep->rmt));
381 	}
382 }
383 
384 /* ARGSUSED */
385 int
386 gettimeofday(struct gettimeofday_args *uap)
387 {
388 	struct timeval atv;
389 	int error = 0;
390 
391 	if (uap->tp) {
392 		microtime(&atv);
393 		if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
394 		    sizeof (atv))))
395 			return (error);
396 	}
397 	if (uap->tzp)
398 		error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
399 		    sizeof (tz));
400 	return (error);
401 }
402 
403 /* ARGSUSED */
404 int
405 settimeofday(struct settimeofday_args *uap)
406 {
407 	struct thread *td = curthread;
408 	struct timeval atv;
409 	struct timezone atz;
410 	int error;
411 
412 	if ((error = suser(td)))
413 		return (error);
414 	/* Verify all parameters before changing time. */
415 	if (uap->tv) {
416 		if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
417 		    sizeof(atv))))
418 			return (error);
419 		if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
420 			return (EINVAL);
421 	}
422 	if (uap->tzp &&
423 	    (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
424 		return (error);
425 	if (uap->tv && (error = settime(&atv)))
426 		return (error);
427 	if (uap->tzp)
428 		tz = atz;
429 	return (0);
430 }
431 
432 int	tickdelta;			/* current clock skew, us. per tick */
433 long	timedelta;			/* unapplied time correction, us. */
434 static long	bigadj = 1000000;	/* use 10x skew above bigadj us. */
435 
436 /* ARGSUSED */
437 int
438 adjtime(struct adjtime_args *uap)
439 {
440 	struct thread *td = curthread;
441 	struct timeval atv;
442 	long ndelta, ntickdelta, odelta;
443 	int error;
444 
445 	if ((error = suser(td)))
446 		return (error);
447 	if ((error =
448 	    copyin((caddr_t)uap->delta, (caddr_t)&atv, sizeof(struct timeval))))
449 		return (error);
450 
451 	/*
452 	 * Compute the total correction and the rate at which to apply it.
453 	 * Round the adjustment down to a whole multiple of the per-tick
454 	 * delta, so that after some number of incremental changes in
455 	 * hardclock(), tickdelta will become zero, lest the correction
456 	 * overshoot and start taking us away from the desired final time.
457 	 */
458 	ndelta = atv.tv_sec * 1000000 + atv.tv_usec;
459 	if (ndelta > bigadj || ndelta < -bigadj)
460 		ntickdelta = 10 * tickadj;
461 	else
462 		ntickdelta = tickadj;
463 	if (ndelta % ntickdelta)
464 		ndelta = ndelta / ntickdelta * ntickdelta;
465 
466 	/*
467 	 * To make hardclock()'s job easier, make the per-tick delta negative
468 	 * if we want time to run slower; then hardclock can simply compute
469 	 * tick + tickdelta, and subtract tickdelta from timedelta.
470 	 */
471 	if (ndelta < 0)
472 		ntickdelta = -ntickdelta;
473 	/*
474 	 * XXX not MP safe , but will probably work anyway.
475 	 */
476 	crit_enter();
477 	odelta = timedelta;
478 	timedelta = ndelta;
479 	tickdelta = ntickdelta;
480 	crit_exit();
481 
482 	if (uap->olddelta) {
483 		atv.tv_sec = odelta / 1000000;
484 		atv.tv_usec = odelta % 1000000;
485 		(void) copyout((caddr_t)&atv, (caddr_t)uap->olddelta,
486 		    sizeof(struct timeval));
487 	}
488 	return (0);
489 }
490 
491 /*
492  * Get value of an interval timer.  The process virtual and
493  * profiling virtual time timers are kept in the p_stats area, since
494  * they can be swapped out.  These are kept internally in the
495  * way they are specified externally: in time until they expire.
496  *
497  * The real time interval timer is kept in the process table slot
498  * for the process, and its value (it_value) is kept as an
499  * absolute time rather than as a delta, so that it is easy to keep
500  * periodic real-time signals from drifting.
501  *
502  * Virtual time timers are processed in the hardclock() routine of
503  * kern_clock.c.  The real time timer is processed by a timeout
504  * routine, called from the softclock() routine.  Since a callout
505  * may be delayed in real time due to interrupt processing in the system,
506  * it is possible for the real time timeout routine (realitexpire, given below),
507  * to be delayed in real time past when it is supposed to occur.  It
508  * does not suffice, therefore, to reload the real timer .it_value from the
509  * real time timers .it_interval.  Rather, we compute the next time in
510  * absolute time the timer should go off.
511  */
512 /* ARGSUSED */
513 int
514 getitimer(struct getitimer_args *uap)
515 {
516 	struct proc *p = curproc;
517 	struct timeval ctv;
518 	struct itimerval aitv;
519 
520 	if (uap->which > ITIMER_PROF)
521 		return (EINVAL);
522 	crit_enter();
523 	if (uap->which == ITIMER_REAL) {
524 		/*
525 		 * Convert from absolute to relative time in .it_value
526 		 * part of real time timer.  If time for real time timer
527 		 * has passed return 0, else return difference between
528 		 * current time and time for the timer to go off.
529 		 */
530 		aitv = p->p_realtimer;
531 		if (timevalisset(&aitv.it_value)) {
532 			getmicrouptime(&ctv);
533 			if (timevalcmp(&aitv.it_value, &ctv, <))
534 				timevalclear(&aitv.it_value);
535 			else
536 				timevalsub(&aitv.it_value, &ctv);
537 		}
538 	} else {
539 		aitv = p->p_stats->p_timer[uap->which];
540 	}
541 	crit_exit();
542 	return (copyout((caddr_t)&aitv, (caddr_t)uap->itv,
543 	    sizeof (struct itimerval)));
544 }
545 
546 /* ARGSUSED */
547 int
548 setitimer(struct setitimer_args *uap)
549 {
550 	struct itimerval aitv;
551 	struct timeval ctv;
552 	struct itimerval *itvp;
553 	struct proc *p = curproc;
554 	int error;
555 
556 	if (uap->which > ITIMER_PROF)
557 		return (EINVAL);
558 	itvp = uap->itv;
559 	if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
560 	    sizeof(struct itimerval))))
561 		return (error);
562 	if ((uap->itv = uap->oitv) &&
563 	    (error = getitimer((struct getitimer_args *)uap)))
564 		return (error);
565 	if (itvp == 0)
566 		return (0);
567 	if (itimerfix(&aitv.it_value))
568 		return (EINVAL);
569 	if (!timevalisset(&aitv.it_value))
570 		timevalclear(&aitv.it_interval);
571 	else if (itimerfix(&aitv.it_interval))
572 		return (EINVAL);
573 	crit_enter();
574 	if (uap->which == ITIMER_REAL) {
575 		if (timevalisset(&p->p_realtimer.it_value))
576 			callout_stop(&p->p_ithandle);
577 		if (timevalisset(&aitv.it_value))
578 			callout_reset(&p->p_ithandle,
579 			    tvtohz_high(&aitv.it_value), realitexpire, p);
580 		getmicrouptime(&ctv);
581 		timevaladd(&aitv.it_value, &ctv);
582 		p->p_realtimer = aitv;
583 	} else {
584 		p->p_stats->p_timer[uap->which] = aitv;
585 	}
586 	crit_exit();
587 	return (0);
588 }
589 
590 /*
591  * Real interval timer expired:
592  * send process whose timer expired an alarm signal.
593  * If time is not set up to reload, then just return.
594  * Else compute next time timer should go off which is > current time.
595  * This is where delay in processing this timeout causes multiple
596  * SIGALRM calls to be compressed into one.
597  * tvtohz_high() always adds 1 to allow for the time until the next clock
598  * interrupt being strictly less than 1 clock tick, but we don't want
599  * that here since we want to appear to be in sync with the clock
600  * interrupt even when we're delayed.
601  */
602 void
603 realitexpire(arg)
604 	void *arg;
605 {
606 	struct proc *p;
607 	struct timeval ctv, ntv;
608 
609 	p = (struct proc *)arg;
610 	psignal(p, SIGALRM);
611 	if (!timevalisset(&p->p_realtimer.it_interval)) {
612 		timevalclear(&p->p_realtimer.it_value);
613 		return;
614 	}
615 	for (;;) {
616 		crit_enter();
617 		timevaladd(&p->p_realtimer.it_value,
618 		    &p->p_realtimer.it_interval);
619 		getmicrouptime(&ctv);
620 		if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
621 			ntv = p->p_realtimer.it_value;
622 			timevalsub(&ntv, &ctv);
623 			callout_reset(&p->p_ithandle, tvtohz_low(&ntv),
624 				      realitexpire, p);
625 			crit_exit();
626 			return;
627 		}
628 		crit_exit();
629 	}
630 }
631 
632 /*
633  * Check that a proposed value to load into the .it_value or
634  * .it_interval part of an interval timer is acceptable, and
635  * fix it to have at least minimal value (i.e. if it is less
636  * than the resolution of the clock, round it up.)
637  */
638 int
639 itimerfix(tv)
640 	struct timeval *tv;
641 {
642 
643 	if (tv->tv_sec < 0 || tv->tv_sec > 100000000 ||
644 	    tv->tv_usec < 0 || tv->tv_usec >= 1000000)
645 		return (EINVAL);
646 	if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
647 		tv->tv_usec = tick;
648 	return (0);
649 }
650 
651 /*
652  * Decrement an interval timer by a specified number
653  * of microseconds, which must be less than a second,
654  * i.e. < 1000000.  If the timer expires, then reload
655  * it.  In this case, carry over (usec - old value) to
656  * reduce the value reloaded into the timer so that
657  * the timer does not drift.  This routine assumes
658  * that it is called in a context where the timers
659  * on which it is operating cannot change in value.
660  */
661 int
662 itimerdecr(itp, usec)
663 	struct itimerval *itp;
664 	int usec;
665 {
666 
667 	if (itp->it_value.tv_usec < usec) {
668 		if (itp->it_value.tv_sec == 0) {
669 			/* expired, and already in next interval */
670 			usec -= itp->it_value.tv_usec;
671 			goto expire;
672 		}
673 		itp->it_value.tv_usec += 1000000;
674 		itp->it_value.tv_sec--;
675 	}
676 	itp->it_value.tv_usec -= usec;
677 	usec = 0;
678 	if (timevalisset(&itp->it_value))
679 		return (1);
680 	/* expired, exactly at end of interval */
681 expire:
682 	if (timevalisset(&itp->it_interval)) {
683 		itp->it_value = itp->it_interval;
684 		itp->it_value.tv_usec -= usec;
685 		if (itp->it_value.tv_usec < 0) {
686 			itp->it_value.tv_usec += 1000000;
687 			itp->it_value.tv_sec--;
688 		}
689 	} else
690 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
691 	return (0);
692 }
693 
694 /*
695  * Add and subtract routines for timevals.
696  * N.B.: subtract routine doesn't deal with
697  * results which are before the beginning,
698  * it just gets very confused in this case.
699  * Caveat emptor.
700  */
701 void
702 timevaladd(t1, t2)
703 	struct timeval *t1, *t2;
704 {
705 
706 	t1->tv_sec += t2->tv_sec;
707 	t1->tv_usec += t2->tv_usec;
708 	timevalfix(t1);
709 }
710 
711 void
712 timevalsub(t1, t2)
713 	struct timeval *t1, *t2;
714 {
715 
716 	t1->tv_sec -= t2->tv_sec;
717 	t1->tv_usec -= t2->tv_usec;
718 	timevalfix(t1);
719 }
720 
721 static void
722 timevalfix(t1)
723 	struct timeval *t1;
724 {
725 
726 	if (t1->tv_usec < 0) {
727 		t1->tv_sec--;
728 		t1->tv_usec += 1000000;
729 	}
730 	if (t1->tv_usec >= 1000000) {
731 		t1->tv_sec++;
732 		t1->tv_usec -= 1000000;
733 	}
734 }
735 
736 /*
737  * ratecheck(): simple time-based rate-limit checking.
738  */
739 int
740 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
741 {
742 	struct timeval tv, delta;
743 	int rv = 0;
744 
745 	getmicrouptime(&tv);		/* NB: 10ms precision */
746 	delta = tv;
747 	timevalsub(&delta, lasttime);
748 
749 	/*
750 	 * check for 0,0 is so that the message will be seen at least once,
751 	 * even if interval is huge.
752 	 */
753 	if (timevalcmp(&delta, mininterval, >=) ||
754 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
755 		*lasttime = tv;
756 		rv = 1;
757 	}
758 
759 	return (rv);
760 }
761 
762 /*
763  * ppsratecheck(): packets (or events) per second limitation.
764  *
765  * Return 0 if the limit is to be enforced (e.g. the caller
766  * should drop a packet because of the rate limitation).
767  *
768  * maxpps of 0 always causes zero to be returned.  maxpps of -1
769  * always causes 1 to be returned; this effectively defeats rate
770  * limiting.
771  *
772  * Note that we maintain the struct timeval for compatibility
773  * with other bsd systems.  We reuse the storage and just monitor
774  * clock ticks for minimal overhead.
775  */
776 int
777 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
778 {
779 	int now;
780 
781 	/*
782 	 * Reset the last time and counter if this is the first call
783 	 * or more than a second has passed since the last update of
784 	 * lasttime.
785 	 */
786 	now = ticks;
787 	if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
788 		lasttime->tv_sec = now;
789 		*curpps = 1;
790 		return (maxpps != 0);
791 	} else {
792 		(*curpps)++;		/* NB: ignore potential overflow */
793 		return (maxpps < 0 || *curpps < maxpps);
794 	}
795 }
796 
797