Home | History | Annotate | Line # | Download | only in kern
kern_time.c revision 1.61.10.2
      1 /*	$NetBSD: kern_time.c,v 1.61.10.2 2005/12/07 10:43:07 tron Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2000 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.
      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  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the NetBSD
     21  *	Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 
     39 /*
     40  * Copyright (c) 1982, 1986, 1989, 1993
     41  *	The Regents of the University of California.  All rights reserved.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  * 3. All advertising materials mentioning features or use of this software
     52  *    must display the following acknowledgement:
     53  *	This product includes software developed by the University of
     54  *	California, Berkeley and its contributors.
     55  * 4. Neither the name of the University nor the names of its contributors
     56  *    may be used to endorse or promote products derived from this software
     57  *    without specific prior written permission.
     58  *
     59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     69  * SUCH DAMAGE.
     70  *
     71  *	@(#)kern_time.c	8.4 (Berkeley) 5/26/95
     72  */
     73 
     74 #include <sys/cdefs.h>
     75 __KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.61.10.2 2005/12/07 10:43:07 tron Exp $");
     76 
     77 #include "fs_nfs.h"
     78 #include "opt_nfs.h"
     79 #include "opt_nfsserver.h"
     80 
     81 #include <sys/param.h>
     82 #include <sys/resourcevar.h>
     83 #include <sys/kernel.h>
     84 #include <sys/systm.h>
     85 #include <sys/proc.h>
     86 #include <sys/vnode.h>
     87 #include <sys/signalvar.h>
     88 #include <sys/syslog.h>
     89 
     90 #include <sys/mount.h>
     91 #include <sys/syscallargs.h>
     92 
     93 #include <uvm/uvm_extern.h>
     94 
     95 #if defined(NFS) || defined(NFSSERVER)
     96 #include <nfs/rpcv2.h>
     97 #include <nfs/nfsproto.h>
     98 #include <nfs/nfs_var.h>
     99 #endif
    100 
    101 #include <machine/cpu.h>
    102 
    103 /*
    104  * Time of day and interval timer support.
    105  *
    106  * These routines provide the kernel entry points to get and set
    107  * the time-of-day and per-process interval timers.  Subroutines
    108  * here provide support for adding and subtracting timeval structures
    109  * and decrementing interval timers, optionally reloading the interval
    110  * timers when they expire.
    111  */
    112 
    113 /* This function is used by clock_settime and settimeofday */
    114 int
    115 settime(tv)
    116 	struct timeval *tv;
    117 {
    118 	struct timeval delta;
    119 	struct cpu_info *ci;
    120 	int s;
    121 
    122 	/*
    123 	 * Don't allow the time to be set forward so far it will wrap
    124 	 * and become negative, thus allowing an attacker to bypass
    125 	 * the next check below.  The cutoff is 1 year before rollover
    126 	 * occurs, so even if the attacker uses adjtime(2) to move
    127 	 * the time past the cutoff, it will take a very long time
    128 	 * to get to the wrap point.
    129 	 *
    130 	 * XXX: we check against INT_MAX since on 64-bit
    131 	 *	platforms, sizeof(int) != sizeof(long) and
    132 	 *	time_t is 32 bits even when atv.tv_sec is 64 bits.
    133 	 */
    134 	if (tv->tv_sec > INT_MAX - 365*24*60*60) {
    135 		struct proc *p = curproc;
    136 		struct proc *pp = p->p_pptr;
    137 		log(LOG_WARNING, "pid %d (%s) "
    138 		    "invoked by uid %d ppid %d (%s) "
    139 		    "tried to set clock forward to %ld\n",
    140 		    p->p_pid, p->p_comm, pp->p_ucred->cr_uid,
    141 		    pp->p_pid, pp->p_comm, (long)tv->tv_sec);
    142 		return (EPERM);
    143 	}
    144 	/* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */
    145 	s = splclock();
    146 	timersub(tv, &time, &delta);
    147 	if ((delta.tv_sec < 0 || delta.tv_usec < 0) && securelevel > 1) {
    148 		splx(s);
    149 		return (EPERM);
    150 	}
    151 #ifdef notyet
    152 	if ((delta.tv_sec < 86400) && securelevel > 0) {
    153 		splx(s);
    154 		return (EPERM);
    155 	}
    156 #endif
    157 	time = *tv;
    158 	(void) spllowersoftclock();
    159 	timeradd(&boottime, &delta, &boottime);
    160 	/*
    161 	 * XXXSMP
    162 	 * This is wrong.  We should traverse a list of all
    163 	 * CPUs and add the delta to the runtime of those
    164 	 * CPUs which have a process on them.
    165 	 */
    166 	ci = curcpu();
    167 	timeradd(&ci->ci_schedstate.spc_runtime, &delta,
    168 	    &ci->ci_schedstate.spc_runtime);
    169 #	if (defined(NFS) && !defined (NFS_V2_ONLY)) || defined(NFSSERVER)
    170 		nqnfs_lease_updatetime(delta.tv_sec);
    171 #	endif
    172 	splx(s);
    173 	resettodr();
    174 	return (0);
    175 }
    176 
    177 /* ARGSUSED */
    178 int
    179 sys_clock_gettime(p, v, retval)
    180 	struct proc *p;
    181 	void *v;
    182 	register_t *retval;
    183 {
    184 	struct sys_clock_gettime_args /* {
    185 		syscallarg(clockid_t) clock_id;
    186 		syscallarg(struct timespec *) tp;
    187 	} */ *uap = v;
    188 	clockid_t clock_id;
    189 	struct timeval atv;
    190 	struct timespec ats;
    191 	int s;
    192 
    193 	clock_id = SCARG(uap, clock_id);
    194 	switch (clock_id) {
    195 	case CLOCK_REALTIME:
    196 		microtime(&atv);
    197 		TIMEVAL_TO_TIMESPEC(&atv,&ats);
    198 		break;
    199 	case CLOCK_MONOTONIC:
    200 		/* XXX "hz" granularity */
    201 		s = splclock();
    202 		atv = mono_time;
    203 		splx(s);
    204 		TIMEVAL_TO_TIMESPEC(&atv,&ats);
    205 		break;
    206 	default:
    207 		return (EINVAL);
    208 	}
    209 
    210 	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
    211 }
    212 
    213 /* ARGSUSED */
    214 int
    215 sys_clock_settime(p, v, retval)
    216 	struct proc *p;
    217 	void *v;
    218 	register_t *retval;
    219 {
    220 	struct sys_clock_settime_args /* {
    221 		syscallarg(clockid_t) clock_id;
    222 		syscallarg(const struct timespec *) tp;
    223 	} */ *uap = v;
    224 	int error;
    225 
    226 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
    227 		return (error);
    228 
    229 	return (clock_settime1(SCARG(uap, clock_id), SCARG(uap, tp)));
    230 }
    231 
    232 
    233 int
    234 clock_settime1(clock_id, tp)
    235 	clockid_t clock_id;
    236 	const struct timespec *tp;
    237 {
    238 	struct timespec ats;
    239 	struct timeval atv;
    240 	int error;
    241 
    242 	if ((error = copyin(tp, &ats, sizeof(ats))) != 0)
    243 		return (error);
    244 
    245 	switch (clock_id) {
    246 	case CLOCK_REALTIME:
    247 		TIMESPEC_TO_TIMEVAL(&atv, &ats);
    248 		if ((error = settime(&atv)) != 0)
    249 			return (error);
    250 		break;
    251 	case CLOCK_MONOTONIC:
    252 		return (EINVAL);	/* read-only clock */
    253 	default:
    254 		return (EINVAL);
    255 	}
    256 
    257 	return 0;
    258 }
    259 
    260 int
    261 sys_clock_getres(p, v, retval)
    262 	struct proc *p;
    263 	void *v;
    264 	register_t *retval;
    265 {
    266 	struct sys_clock_getres_args /* {
    267 		syscallarg(clockid_t) clock_id;
    268 		syscallarg(struct timespec *) tp;
    269 	} */ *uap = v;
    270 	clockid_t clock_id;
    271 	struct timespec ts;
    272 	int error = 0;
    273 
    274 	clock_id = SCARG(uap, clock_id);
    275 	switch (clock_id) {
    276 	case CLOCK_REALTIME:
    277 	case CLOCK_MONOTONIC:
    278 		ts.tv_sec = 0;
    279 		ts.tv_nsec = 1000000000 / hz;
    280 		break;
    281 	default:
    282 		return (EINVAL);
    283 	}
    284 
    285 	if (SCARG(uap, tp))
    286 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
    287 
    288 	return error;
    289 }
    290 
    291 /* ARGSUSED */
    292 int
    293 sys_nanosleep(p, v, retval)
    294 	struct proc *p;
    295 	void *v;
    296 	register_t *retval;
    297 {
    298 	static int nanowait;
    299 	struct sys_nanosleep_args/* {
    300 		syscallarg(struct timespec *) rqtp;
    301 		syscallarg(struct timespec *) rmtp;
    302 	} */ *uap = v;
    303 	struct timespec rqt;
    304 	struct timespec rmt;
    305 	struct timeval atv, utv;
    306 	int error, s, timo;
    307 
    308 	error = copyin((caddr_t)SCARG(uap, rqtp), (caddr_t)&rqt,
    309 		       sizeof(struct timespec));
    310 	if (error)
    311 		return (error);
    312 
    313 	TIMESPEC_TO_TIMEVAL(&atv,&rqt)
    314 	if (itimerfix(&atv) || atv.tv_sec > 1000000000)
    315 		return (EINVAL);
    316 
    317 	s = splclock();
    318 	timeradd(&atv,&time,&atv);
    319 	timo = hzto(&atv);
    320 	/*
    321 	 * Avoid inadvertantly sleeping forever
    322 	 */
    323 	if (timo == 0)
    324 		timo = 1;
    325 	splx(s);
    326 
    327 	error = tsleep(&nanowait, PWAIT | PCATCH, "nanosleep", timo);
    328 	if (error == ERESTART)
    329 		error = EINTR;
    330 	if (error == EWOULDBLOCK)
    331 		error = 0;
    332 
    333 	if (SCARG(uap, rmtp)) {
    334 		int error;
    335 
    336 		s = splclock();
    337 		utv = time;
    338 		splx(s);
    339 
    340 		timersub(&atv, &utv, &utv);
    341 		if (utv.tv_sec < 0)
    342 			timerclear(&utv);
    343 
    344 		TIMEVAL_TO_TIMESPEC(&utv,&rmt);
    345 		error = copyout((caddr_t)&rmt, (caddr_t)SCARG(uap,rmtp),
    346 			sizeof(rmt));
    347 		if (error)
    348 			return (error);
    349 	}
    350 
    351 	return error;
    352 }
    353 
    354 /* ARGSUSED */
    355 int
    356 sys_gettimeofday(p, v, retval)
    357 	struct proc *p;
    358 	void *v;
    359 	register_t *retval;
    360 {
    361 	struct sys_gettimeofday_args /* {
    362 		syscallarg(struct timeval *) tp;
    363 		syscallarg(struct timezone *) tzp;
    364 	} */ *uap = v;
    365 	struct timeval atv;
    366 	int error = 0;
    367 	struct timezone tzfake;
    368 
    369 	if (SCARG(uap, tp)) {
    370 		microtime(&atv);
    371 		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
    372 		if (error)
    373 			return (error);
    374 	}
    375 	if (SCARG(uap, tzp)) {
    376 		/*
    377 		 * NetBSD has no kernel notion of time zone, so we just
    378 		 * fake up a timezone struct and return it if demanded.
    379 		 */
    380 		tzfake.tz_minuteswest = 0;
    381 		tzfake.tz_dsttime = 0;
    382 		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
    383 	}
    384 	return (error);
    385 }
    386 
    387 /* ARGSUSED */
    388 int
    389 sys_settimeofday(p, v, retval)
    390 	struct proc *p;
    391 	void *v;
    392 	register_t *retval;
    393 {
    394 	struct sys_settimeofday_args /* {
    395 		syscallarg(const struct timeval *) tv;
    396 		syscallarg(const struct timezone *) tzp;
    397 	} */ *uap = v;
    398 	int error;
    399 
    400 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
    401 		return (error);
    402 
    403 	return settimeofday1(SCARG(uap, tv), SCARG(uap, tzp), p);
    404 }
    405 
    406 int
    407 settimeofday1(utv, utzp, p)
    408 	const struct timeval *utv;
    409 	const struct timezone *utzp;
    410 	struct proc *p;
    411 {
    412 	struct timeval atv;
    413 	struct timezone atz;
    414 	struct timeval *tv = NULL;
    415 	struct timezone *tzp = NULL;
    416 	int error;
    417 
    418 	/* Verify all parameters before changing time. */
    419 	if (utv) {
    420 		if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
    421 			return (error);
    422 		tv = &atv;
    423 	}
    424 	/* XXX since we don't use tz, probably no point in doing copyin. */
    425 	if (utzp) {
    426 		if ((error = copyin(utzp, &atz, sizeof(atz))) != 0)
    427 			return (error);
    428 		tzp = &atz;
    429 	}
    430 
    431 	if (tv)
    432 		if ((error = settime(tv)) != 0)
    433 			return (error);
    434 	/*
    435 	 * NetBSD has no kernel notion of time zone, and only an
    436 	 * obsolete program would try to set it, so we log a warning.
    437 	 */
    438 	if (tzp)
    439 		log(LOG_WARNING, "pid %d attempted to set the "
    440 		    "(obsolete) kernel time zone\n", p->p_pid);
    441 	return (0);
    442 }
    443 
    444 int	tickdelta;			/* current clock skew, us. per tick */
    445 long	timedelta;			/* unapplied time correction, us. */
    446 long	bigadj = 1000000;		/* use 10x skew above bigadj us. */
    447 
    448 /* ARGSUSED */
    449 int
    450 sys_adjtime(p, v, retval)
    451 	struct proc *p;
    452 	void *v;
    453 	register_t *retval;
    454 {
    455 	struct sys_adjtime_args /* {
    456 		syscallarg(const struct timeval *) delta;
    457 		syscallarg(struct timeval *) olddelta;
    458 	} */ *uap = v;
    459 	int error;
    460 
    461 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
    462 		return (error);
    463 
    464 	return adjtime1(SCARG(uap, delta), SCARG(uap, olddelta), p);
    465 }
    466 
    467 int
    468 adjtime1(delta, olddelta, p)
    469 	const struct timeval *delta;
    470 	struct timeval *olddelta;
    471 	struct proc *p;
    472 {
    473 	struct timeval atv;
    474 	struct timeval *oatv = NULL;
    475 	long ndelta, ntickdelta, odelta;
    476 	int error;
    477 	int s;
    478 
    479 	error = copyin(delta, &atv, sizeof(struct timeval));
    480 	if (error)
    481 		return (error);
    482 
    483 	if (olddelta != NULL) {
    484 		if (uvm_useracc((caddr_t)olddelta,
    485 		    sizeof(struct timeval), B_WRITE) == FALSE)
    486 			return (EFAULT);
    487 		oatv = olddelta;
    488 	}
    489 
    490 	/*
    491 	 * Compute the total correction and the rate at which to apply it.
    492 	 * Round the adjustment down to a whole multiple of the per-tick
    493 	 * delta, so that after some number of incremental changes in
    494 	 * hardclock(), tickdelta will become zero, lest the correction
    495 	 * overshoot and start taking us away from the desired final time.
    496 	 */
    497 	ndelta = atv.tv_sec * 1000000 + atv.tv_usec;
    498 	if (ndelta > bigadj || ndelta < -bigadj)
    499 		ntickdelta = 10 * tickadj;
    500 	else
    501 		ntickdelta = tickadj;
    502 	if (ndelta % ntickdelta)
    503 		ndelta = ndelta / ntickdelta * ntickdelta;
    504 
    505 	/*
    506 	 * To make hardclock()'s job easier, make the per-tick delta negative
    507 	 * if we want time to run slower; then hardclock can simply compute
    508 	 * tick + tickdelta, and subtract tickdelta from timedelta.
    509 	 */
    510 	if (ndelta < 0)
    511 		ntickdelta = -ntickdelta;
    512 	s = splclock();
    513 	odelta = timedelta;
    514 	timedelta = ndelta;
    515 	tickdelta = ntickdelta;
    516 	splx(s);
    517 
    518 	if (olddelta) {
    519 		atv.tv_sec = odelta / 1000000;
    520 		atv.tv_usec = odelta % 1000000;
    521 		(void) copyout(&atv, olddelta, sizeof(struct timeval));
    522 	}
    523 	return (0);
    524 }
    525 
    526 /*
    527  * Get value of an interval timer.  The process virtual and
    528  * profiling virtual time timers are kept in the p_stats area, since
    529  * they can be swapped out.  These are kept internally in the
    530  * way they are specified externally: in time until they expire.
    531  *
    532  * The real time interval timer is kept in the process table slot
    533  * for the process, and its value (it_value) is kept as an
    534  * absolute time rather than as a delta, so that it is easy to keep
    535  * periodic real-time signals from drifting.
    536  *
    537  * Virtual time timers are processed in the hardclock() routine of
    538  * kern_clock.c.  The real time timer is processed by a timeout
    539  * routine, called from the softclock() routine.  Since a callout
    540  * may be delayed in real time due to interrupt processing in the system,
    541  * it is possible for the real time timeout routine (realitexpire, given below),
    542  * to be delayed in real time past when it is supposed to occur.  It
    543  * does not suffice, therefore, to reload the real timer .it_value from the
    544  * real time timers .it_interval.  Rather, we compute the next time in
    545  * absolute time the timer should go off.
    546  */
    547 /* ARGSUSED */
    548 int
    549 sys_getitimer(p, v, retval)
    550 	struct proc *p;
    551 	void *v;
    552 	register_t *retval;
    553 {
    554 	struct sys_getitimer_args /* {
    555 		syscallarg(int) which;
    556 		syscallarg(struct itimerval *) itv;
    557 	} */ *uap = v;
    558 	int which = SCARG(uap, which);
    559 	struct itimerval aitv;
    560 	int s;
    561 
    562 	if ((u_int)which > ITIMER_PROF)
    563 		return (EINVAL);
    564 	s = splclock();
    565 	if (which == ITIMER_REAL) {
    566 		/*
    567 		 * Convert from absolute to relative time in .it_value
    568 		 * part of real time timer.  If time for real time timer
    569 		 * has passed return 0, else return difference between
    570 		 * current time and time for the timer to go off.
    571 		 */
    572 		aitv = p->p_realtimer;
    573 		if (timerisset(&aitv.it_value)) {
    574 			if (timercmp(&aitv.it_value, &time, <))
    575 				timerclear(&aitv.it_value);
    576 			else
    577 				timersub(&aitv.it_value, &time, &aitv.it_value);
    578 		}
    579 	} else
    580 		aitv = p->p_stats->p_timer[which];
    581 	splx(s);
    582 	return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)));
    583 }
    584 
    585 /* ARGSUSED */
    586 int
    587 sys_setitimer(p, v, retval)
    588 	struct proc *p;
    589 	void *v;
    590 	register_t *retval;
    591 {
    592 	struct sys_setitimer_args /* {
    593 		syscallarg(int) which;
    594 		syscallarg(const struct itimerval *) itv;
    595 		syscallarg(struct itimerval *) oitv;
    596 	} */ *uap = v;
    597 	int which = SCARG(uap, which);
    598 	struct sys_getitimer_args getargs;
    599 	struct itimerval aitv;
    600 	const struct itimerval *itvp;
    601 	int s, error;
    602 
    603 	if ((u_int)which > ITIMER_PROF)
    604 		return (EINVAL);
    605 	itvp = SCARG(uap, itv);
    606 	if (itvp &&
    607 	    (error = copyin(itvp, &aitv, sizeof(struct itimerval)) != 0))
    608 		return (error);
    609 	if (SCARG(uap, oitv) != NULL) {
    610 		SCARG(&getargs, which) = which;
    611 		SCARG(&getargs, itv) = SCARG(uap, oitv);
    612 		if ((error = sys_getitimer(p, &getargs, retval)) != 0)
    613 			return (error);
    614 	}
    615 	if (itvp == 0)
    616 		return (0);
    617 	if (itimerfix(&aitv.it_value) || itimerfix(&aitv.it_interval))
    618 		return (EINVAL);
    619 	s = splclock();
    620 	if (which == ITIMER_REAL) {
    621 		callout_stop(&p->p_realit_ch);
    622 		if (timerisset(&aitv.it_value)) {
    623 			/*
    624 			 * Don't need to check hzto() return value, here.
    625 			 * callout_reset() does it for us.
    626 			 */
    627 			timeradd(&aitv.it_value, &time, &aitv.it_value);
    628 			callout_reset(&p->p_realit_ch, hzto(&aitv.it_value),
    629 			    realitexpire, p);
    630 		}
    631 		p->p_realtimer = aitv;
    632 	} else
    633 		p->p_stats->p_timer[which] = aitv;
    634 	splx(s);
    635 	return (0);
    636 }
    637 
    638 /*
    639  * Real interval timer expired:
    640  * send process whose timer expired an alarm signal.
    641  * If time is not set up to reload, then just return.
    642  * Else compute next time timer should go off which is > current time.
    643  * This is where delay in processing this timeout causes multiple
    644  * SIGALRM calls to be compressed into one.
    645  */
    646 void
    647 realitexpire(arg)
    648 	void *arg;
    649 {
    650 	struct proc *p;
    651 	int s;
    652 
    653 	p = (struct proc *)arg;
    654 	psignal(p, SIGALRM);
    655 	if (!timerisset(&p->p_realtimer.it_interval)) {
    656 		timerclear(&p->p_realtimer.it_value);
    657 		return;
    658 	}
    659 	for (;;) {
    660 		s = splclock();
    661 		timeradd(&p->p_realtimer.it_value,
    662 		    &p->p_realtimer.it_interval, &p->p_realtimer.it_value);
    663 		if (timercmp(&p->p_realtimer.it_value, &time, >)) {
    664 			/*
    665 			 * Don't need to check hzto() return value, here.
    666 			 * callout_reset() does it for us.
    667 			 */
    668 			callout_reset(&p->p_realit_ch,
    669 			    hzto(&p->p_realtimer.it_value), realitexpire, p);
    670 			splx(s);
    671 			return;
    672 		}
    673 		splx(s);
    674 	}
    675 }
    676 
    677 /*
    678  * Check that a proposed value to load into the .it_value or
    679  * .it_interval part of an interval timer is acceptable, and
    680  * fix it to have at least minimal value (i.e. if it is less
    681  * than the resolution of the clock, round it up.)
    682  */
    683 int
    684 itimerfix(tv)
    685 	struct timeval *tv;
    686 {
    687 
    688 	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
    689 		return (EINVAL);
    690 	if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
    691 		tv->tv_usec = tick;
    692 	return (0);
    693 }
    694 
    695 /*
    696  * Decrement an interval timer by a specified number
    697  * of microseconds, which must be less than a second,
    698  * i.e. < 1000000.  If the timer expires, then reload
    699  * it.  In this case, carry over (usec - old value) to
    700  * reduce the value reloaded into the timer so that
    701  * the timer does not drift.  This routine assumes
    702  * that it is called in a context where the timers
    703  * on which it is operating cannot change in value.
    704  */
    705 int
    706 itimerdecr(itp, usec)
    707 	struct itimerval *itp;
    708 	int usec;
    709 {
    710 
    711 	if (itp->it_value.tv_usec < usec) {
    712 		if (itp->it_value.tv_sec == 0) {
    713 			/* expired, and already in next interval */
    714 			usec -= itp->it_value.tv_usec;
    715 			goto expire;
    716 		}
    717 		itp->it_value.tv_usec += 1000000;
    718 		itp->it_value.tv_sec--;
    719 	}
    720 	itp->it_value.tv_usec -= usec;
    721 	usec = 0;
    722 	if (timerisset(&itp->it_value))
    723 		return (1);
    724 	/* expired, exactly at end of interval */
    725 expire:
    726 	if (timerisset(&itp->it_interval)) {
    727 		itp->it_value = itp->it_interval;
    728 		itp->it_value.tv_usec -= usec;
    729 		if (itp->it_value.tv_usec < 0) {
    730 			itp->it_value.tv_usec += 1000000;
    731 			itp->it_value.tv_sec--;
    732 		}
    733 	} else
    734 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
    735 	return (0);
    736 }
    737 
    738 /*
    739  * ratecheck(): simple time-based rate-limit checking.  see ratecheck(9)
    740  * for usage and rationale.
    741  */
    742 int
    743 ratecheck(lasttime, mininterval)
    744 	struct timeval *lasttime;
    745 	const struct timeval *mininterval;
    746 {
    747 	struct timeval tv, delta;
    748 	int s, rv = 0;
    749 
    750 	s = splclock();
    751 	tv = mono_time;
    752 	splx(s);
    753 
    754 	timersub(&tv, lasttime, &delta);
    755 
    756 	/*
    757 	 * check for 0,0 is so that the message will be seen at least once,
    758 	 * even if interval is huge.
    759 	 */
    760 	if (timercmp(&delta, mininterval, >=) ||
    761 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
    762 		*lasttime = tv;
    763 		rv = 1;
    764 	}
    765 
    766 	return (rv);
    767 }
    768 
    769 /*
    770  * ppsratecheck(): packets (or events) per second limitation.
    771  */
    772 int
    773 ppsratecheck(lasttime, curpps, maxpps)
    774 	struct timeval *lasttime;
    775 	int *curpps;
    776 	int maxpps;	/* maximum pps allowed */
    777 {
    778 	struct timeval tv, delta;
    779 	int s, rv;
    780 
    781 	s = splclock();
    782 	tv = mono_time;
    783 	splx(s);
    784 
    785 	timersub(&tv, lasttime, &delta);
    786 
    787 	/*
    788 	 * check for 0,0 is so that the message will be seen at least once.
    789 	 * if more than one second have passed since the last update of
    790 	 * lasttime, reset the counter.
    791 	 *
    792 	 * we do increment *curpps even in *curpps < maxpps case, as some may
    793 	 * try to use *curpps for stat purposes as well.
    794 	 */
    795 	if ((lasttime->tv_sec == 0 && lasttime->tv_usec == 0) ||
    796 	    delta.tv_sec >= 1) {
    797 		*lasttime = tv;
    798 		*curpps = 0;
    799 		rv = 1;
    800 	} else if (maxpps < 0)
    801 		rv = 1;
    802 	else if (*curpps < maxpps)
    803 		rv = 1;
    804 	else
    805 		rv = 0;
    806 
    807 #if 1 /*DIAGNOSTIC?*/
    808 	/* be careful about wrap-around */
    809 	if (*curpps + 1 > *curpps)
    810 		*curpps = *curpps + 1;
    811 #else
    812 	/*
    813 	 * assume that there's not too many calls to this function.
    814 	 * not sure if the assumption holds, as it depends on *caller's*
    815 	 * behavior, not the behavior of this function.
    816 	 * IMHO it is wrong to make assumption on the caller's behavior,
    817 	 * so the above #if is #if 1, not #ifdef DIAGNOSTIC.
    818 	 */
    819 	*curpps = *curpps + 1;
    820 #endif
    821 
    822 	return (rv);
    823 }
    824