Home | History | Annotate | Line # | Download | only in kern
kern_time.c revision 1.142
      1 /*	$NetBSD: kern_time.c,v 1.142 2008/04/21 00:13:46 ad Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2000, 2004, 2005, 2007, 2008 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. Neither the name of the University nor the names of its contributors
     52  *    may be used to endorse or promote products derived from this software
     53  *    without specific prior written permission.
     54  *
     55  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     56  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     65  * SUCH DAMAGE.
     66  *
     67  *	@(#)kern_time.c	8.4 (Berkeley) 5/26/95
     68  */
     69 
     70 #include <sys/cdefs.h>
     71 __KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.142 2008/04/21 00:13:46 ad Exp $");
     72 
     73 #include <sys/param.h>
     74 #include <sys/resourcevar.h>
     75 #include <sys/kernel.h>
     76 #include <sys/systm.h>
     77 #include <sys/proc.h>
     78 #include <sys/vnode.h>
     79 #include <sys/signalvar.h>
     80 #include <sys/syslog.h>
     81 #include <sys/timetc.h>
     82 #include <sys/kauth.h>
     83 
     84 #include <sys/mount.h>
     85 #include <sys/syscallargs.h>
     86 
     87 #include <uvm/uvm_extern.h>
     88 
     89 #include <sys/cpu.h>
     90 
     91 static void	timer_intr(void *);
     92 static void	itimerfire(struct ptimer *);
     93 static void	itimerfree(struct ptimers *, int);
     94 
     95 kmutex_t	time_lock;
     96 kmutex_t	timer_lock;
     97 
     98 static void	*timer_sih;
     99 static TAILQ_HEAD(, ptimer) timer_queue;
    100 
    101 POOL_INIT(ptimer_pool, sizeof(struct ptimer), 0, 0, 0, "ptimerpl",
    102     &pool_allocator_nointr, IPL_NONE);
    103 POOL_INIT(ptimers_pool, sizeof(struct ptimers), 0, 0, 0, "ptimerspl",
    104     &pool_allocator_nointr, IPL_NONE);
    105 
    106 /*
    107  * Initialize timekeeping.
    108  */
    109 void
    110 time_init(void)
    111 {
    112 
    113 	mutex_init(&time_lock, MUTEX_DEFAULT, IPL_NONE);
    114 }
    115 
    116 void
    117 time_init2(void)
    118 {
    119 
    120 	TAILQ_INIT(&timer_queue);
    121 	mutex_init(&timer_lock, MUTEX_DEFAULT, IPL_SCHED);
    122 	timer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
    123 	    timer_intr, NULL);
    124 }
    125 
    126 /* Time of day and interval timer support.
    127  *
    128  * These routines provide the kernel entry points to get and set
    129  * the time-of-day and per-process interval timers.  Subroutines
    130  * here provide support for adding and subtracting timeval structures
    131  * and decrementing interval timers, optionally reloading the interval
    132  * timers when they expire.
    133  */
    134 
    135 /* This function is used by clock_settime and settimeofday */
    136 static int
    137 settime1(struct proc *p, struct timespec *ts, bool check_kauth)
    138 {
    139 	struct timeval delta, tv;
    140 	struct timeval now;
    141 	struct timespec ts1;
    142 	struct bintime btdelta;
    143 	lwp_t *l;
    144 	int s;
    145 
    146 	TIMESPEC_TO_TIMEVAL(&tv, ts);
    147 
    148 	/* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */
    149 	s = splclock();
    150 	microtime(&now);
    151 	timersub(&tv, &now, &delta);
    152 
    153 	if (check_kauth && kauth_authorize_system(kauth_cred_get(),
    154 	    KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, ts, &delta,
    155 	    KAUTH_ARG(check_kauth ? false : true)) != 0) {
    156 		splx(s);
    157 		return (EPERM);
    158 	}
    159 
    160 #ifdef notyet
    161 	if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */
    162 		splx(s);
    163 		return (EPERM);
    164 	}
    165 #endif
    166 
    167 	TIMEVAL_TO_TIMESPEC(&tv, &ts1);
    168 	tc_setclock(&ts1);
    169 
    170 	timeradd(&boottime, &delta, &boottime);
    171 
    172 	/*
    173 	 * XXXSMP: There is a short race between setting the time above
    174 	 * and adjusting LWP's run times.  Fixing this properly means
    175 	 * pausing all CPUs while we adjust the clock.
    176 	 */
    177 	timeval2bintime(&delta, &btdelta);
    178 	mutex_enter(&proclist_lock);
    179 	LIST_FOREACH(l, &alllwp, l_list) {
    180 		lwp_lock(l);
    181 		bintime_add(&l->l_stime, &btdelta);
    182 		lwp_unlock(l);
    183 	}
    184 	mutex_exit(&proclist_lock);
    185 	resettodr();
    186 	splx(s);
    187 
    188 	return (0);
    189 }
    190 
    191 int
    192 settime(struct proc *p, struct timespec *ts)
    193 {
    194 	return (settime1(p, ts, true));
    195 }
    196 
    197 /* ARGSUSED */
    198 int
    199 sys_clock_gettime(struct lwp *l, const struct sys_clock_gettime_args *uap,
    200     register_t *retval)
    201 {
    202 	/* {
    203 		syscallarg(clockid_t) clock_id;
    204 		syscallarg(struct timespec *) tp;
    205 	} */
    206 	clockid_t clock_id;
    207 	struct timespec ats;
    208 
    209 	clock_id = SCARG(uap, clock_id);
    210 	switch (clock_id) {
    211 	case CLOCK_REALTIME:
    212 		nanotime(&ats);
    213 		break;
    214 	case CLOCK_MONOTONIC:
    215 		nanouptime(&ats);
    216 		break;
    217 	default:
    218 		return (EINVAL);
    219 	}
    220 
    221 	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
    222 }
    223 
    224 /* ARGSUSED */
    225 int
    226 sys_clock_settime(struct lwp *l, const struct sys_clock_settime_args *uap,
    227     register_t *retval)
    228 {
    229 	/* {
    230 		syscallarg(clockid_t) clock_id;
    231 		syscallarg(const struct timespec *) tp;
    232 	} */
    233 
    234 	return clock_settime1(l->l_proc, SCARG(uap, clock_id), SCARG(uap, tp),
    235 	    true);
    236 }
    237 
    238 
    239 int
    240 clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp,
    241     bool check_kauth)
    242 {
    243 	struct timespec ats;
    244 	int error;
    245 
    246 	if ((error = copyin(tp, &ats, sizeof(ats))) != 0)
    247 		return (error);
    248 
    249 	switch (clock_id) {
    250 	case CLOCK_REALTIME:
    251 		if ((error = settime1(p, &ats, check_kauth)) != 0)
    252 			return (error);
    253 		break;
    254 	case CLOCK_MONOTONIC:
    255 		return (EINVAL);	/* read-only clock */
    256 	default:
    257 		return (EINVAL);
    258 	}
    259 
    260 	return 0;
    261 }
    262 
    263 int
    264 sys_clock_getres(struct lwp *l, const struct sys_clock_getres_args *uap,
    265     register_t *retval)
    266 {
    267 	/* {
    268 		syscallarg(clockid_t) clock_id;
    269 		syscallarg(struct timespec *) tp;
    270 	} */
    271 	clockid_t clock_id;
    272 	struct timespec ts;
    273 	int error = 0;
    274 
    275 	clock_id = SCARG(uap, clock_id);
    276 	switch (clock_id) {
    277 	case CLOCK_REALTIME:
    278 	case CLOCK_MONOTONIC:
    279 		ts.tv_sec = 0;
    280 		if (tc_getfrequency() > 1000000000)
    281 			ts.tv_nsec = 1;
    282 		else
    283 			ts.tv_nsec = 1000000000 / tc_getfrequency();
    284 		break;
    285 	default:
    286 		return (EINVAL);
    287 	}
    288 
    289 	if (SCARG(uap, tp))
    290 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
    291 
    292 	return error;
    293 }
    294 
    295 /* ARGSUSED */
    296 int
    297 sys_nanosleep(struct lwp *l, const struct sys_nanosleep_args *uap,
    298     register_t *retval)
    299 {
    300 	/* {
    301 		syscallarg(struct timespec *) rqtp;
    302 		syscallarg(struct timespec *) rmtp;
    303 	} */
    304 	struct timespec rmt, rqt;
    305 	int error, error1;
    306 
    307 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
    308 	if (error)
    309 		return (error);
    310 
    311 	error = nanosleep1(l, &rqt, SCARG(uap, rmtp) ? &rmt : NULL);
    312 	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
    313 		return error;
    314 
    315 	error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt));
    316 	return error1 ? error1 : error;
    317 }
    318 
    319 int
    320 nanosleep1(struct lwp *l, struct timespec *rqt, struct timespec *rmt)
    321 {
    322 	struct timespec rmtstart;
    323 	int error, timo;
    324 
    325 	if (itimespecfix(rqt))
    326 		return (EINVAL);
    327 
    328 	timo = tstohz(rqt);
    329 	/*
    330 	 * Avoid inadvertantly sleeping forever
    331 	 */
    332 	if (timo == 0)
    333 		timo = 1;
    334 	getnanouptime(&rmtstart);
    335 again:
    336 	error = kpause("nanoslp", true, timo, NULL);
    337 	if (rmt != NULL || error == 0) {
    338 		struct timespec rmtend;
    339 		struct timespec t0;
    340 		struct timespec *t;
    341 
    342 		getnanouptime(&rmtend);
    343 		t = (rmt != NULL) ? rmt : &t0;
    344 		timespecsub(&rmtend, &rmtstart, t);
    345 		timespecsub(rqt, t, t);
    346 		if (t->tv_sec < 0)
    347 			timespecclear(t);
    348 		if (error == 0) {
    349 			timo = tstohz(t);
    350 			if (timo > 0)
    351 				goto again;
    352 		}
    353 	}
    354 
    355 	if (error == ERESTART)
    356 		error = EINTR;
    357 	if (error == EWOULDBLOCK)
    358 		error = 0;
    359 
    360 	return error;
    361 }
    362 
    363 /* ARGSUSED */
    364 int
    365 sys_gettimeofday(struct lwp *l, const struct sys_gettimeofday_args *uap,
    366     register_t *retval)
    367 {
    368 	/* {
    369 		syscallarg(struct timeval *) tp;
    370 		syscallarg(void *) tzp;		really "struct timezone *";
    371 	} */
    372 	struct timeval atv;
    373 	int error = 0;
    374 	struct timezone tzfake;
    375 
    376 	if (SCARG(uap, tp)) {
    377 		microtime(&atv);
    378 		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
    379 		if (error)
    380 			return (error);
    381 	}
    382 	if (SCARG(uap, tzp)) {
    383 		/*
    384 		 * NetBSD has no kernel notion of time zone, so we just
    385 		 * fake up a timezone struct and return it if demanded.
    386 		 */
    387 		tzfake.tz_minuteswest = 0;
    388 		tzfake.tz_dsttime = 0;
    389 		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
    390 	}
    391 	return (error);
    392 }
    393 
    394 /* ARGSUSED */
    395 int
    396 sys_settimeofday(struct lwp *l, const struct sys_settimeofday_args *uap,
    397     register_t *retval)
    398 {
    399 	/* {
    400 		syscallarg(const struct timeval *) tv;
    401 		syscallarg(const void *) tzp; really "const struct timezone *";
    402 	} */
    403 
    404 	return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true);
    405 }
    406 
    407 int
    408 settimeofday1(const struct timeval *utv, bool userspace,
    409     const void *utzp, struct lwp *l, bool check_kauth)
    410 {
    411 	struct timeval atv;
    412 	struct timespec ts;
    413 	int error;
    414 
    415 	/* Verify all parameters before changing time. */
    416 
    417 	/*
    418 	 * NetBSD has no kernel notion of time zone, and only an
    419 	 * obsolete program would try to set it, so we log a warning.
    420 	 */
    421 	if (utzp)
    422 		log(LOG_WARNING, "pid %d attempted to set the "
    423 		    "(obsolete) kernel time zone\n", l->l_proc->p_pid);
    424 
    425 	if (utv == NULL)
    426 		return 0;
    427 
    428 	if (userspace) {
    429 		if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
    430 			return error;
    431 		utv = &atv;
    432 	}
    433 
    434 	TIMEVAL_TO_TIMESPEC(utv, &ts);
    435 	return settime1(l->l_proc, &ts, check_kauth);
    436 }
    437 
    438 int	time_adjusted;			/* set if an adjustment is made */
    439 
    440 /* ARGSUSED */
    441 int
    442 sys_adjtime(struct lwp *l, const struct sys_adjtime_args *uap,
    443     register_t *retval)
    444 {
    445 	/* {
    446 		syscallarg(const struct timeval *) delta;
    447 		syscallarg(struct timeval *) olddelta;
    448 	} */
    449 	int error;
    450 
    451 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
    452 	    KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0)
    453 		return (error);
    454 
    455 	return adjtime1(SCARG(uap, delta), SCARG(uap, olddelta), l->l_proc);
    456 }
    457 
    458 int
    459 adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p)
    460 {
    461 	struct timeval atv;
    462 	int error = 0;
    463 
    464 	extern int64_t time_adjtime;  /* in kern_ntptime.c */
    465 
    466 	if (olddelta) {
    467 		atv.tv_sec = time_adjtime / 1000000;
    468 		atv.tv_usec = time_adjtime % 1000000;
    469 		if (atv.tv_usec < 0) {
    470 			atv.tv_usec += 1000000;
    471 			atv.tv_sec--;
    472 		}
    473 		error = copyout(&atv, olddelta, sizeof(struct timeval));
    474 		if (error)
    475 			return (error);
    476 	}
    477 
    478 	if (delta) {
    479 		error = copyin(delta, &atv, sizeof(struct timeval));
    480 		if (error)
    481 			return (error);
    482 
    483 		time_adjtime = (int64_t)atv.tv_sec * 1000000 +
    484 			atv.tv_usec;
    485 
    486 		if (time_adjtime)
    487 			/* We need to save the system time during shutdown */
    488 			time_adjusted |= 1;
    489 	}
    490 
    491 	return error;
    492 }
    493 
    494 /*
    495  * Interval timer support. Both the BSD getitimer() family and the POSIX
    496  * timer_*() family of routines are supported.
    497  *
    498  * All timers are kept in an array pointed to by p_timers, which is
    499  * allocated on demand - many processes don't use timers at all. The
    500  * first three elements in this array are reserved for the BSD timers:
    501  * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, and element
    502  * 2 is ITIMER_PROF. The rest may be allocated by the timer_create()
    503  * syscall.
    504  *
    505  * Realtime timers are kept in the ptimer structure as an absolute
    506  * time; virtual time timers are kept as a linked list of deltas.
    507  * Virtual time timers are processed in the hardclock() routine of
    508  * kern_clock.c.  The real time timer is processed by a callout
    509  * routine, called from the softclock() routine.  Since a callout may
    510  * be delayed in real time due to interrupt processing in the system,
    511  * it is possible for the real time timeout routine (realtimeexpire,
    512  * given below), to be delayed in real time past when it is supposed
    513  * to occur.  It does not suffice, therefore, to reload the real timer
    514  * .it_value from the real time timers .it_interval.  Rather, we
    515  * compute the next time in absolute time the timer should go off.  */
    516 
    517 /* Allocate a POSIX realtime timer. */
    518 int
    519 sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap,
    520     register_t *retval)
    521 {
    522 	/* {
    523 		syscallarg(clockid_t) clock_id;
    524 		syscallarg(struct sigevent *) evp;
    525 		syscallarg(timer_t *) timerid;
    526 	} */
    527 
    528 	return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id),
    529 	    SCARG(uap, evp), copyin, l);
    530 }
    531 
    532 int
    533 timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp,
    534     copyin_t fetch_event, struct lwp *l)
    535 {
    536 	int error;
    537 	timer_t timerid;
    538 	struct ptimers *pts;
    539 	struct ptimer *pt;
    540 	struct proc *p;
    541 
    542 	p = l->l_proc;
    543 
    544 	if (id < CLOCK_REALTIME || id > CLOCK_PROF)
    545 		return (EINVAL);
    546 
    547 	if ((pts = p->p_timers) == NULL)
    548 		pts = timers_alloc(p);
    549 
    550 	pt = pool_get(&ptimer_pool, PR_WAITOK);
    551 	if (evp != NULL) {
    552 		if (((error =
    553 		    (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) ||
    554 		    ((pt->pt_ev.sigev_notify < SIGEV_NONE) ||
    555 			(pt->pt_ev.sigev_notify > SIGEV_SA))) {
    556 			pool_put(&ptimer_pool, pt);
    557 			return (error ? error : EINVAL);
    558 		}
    559 	}
    560 
    561 	/* Find a free timer slot, skipping those reserved for setitimer(). */
    562 	mutex_spin_enter(&timer_lock);
    563 	for (timerid = 3; timerid < TIMER_MAX; timerid++)
    564 		if (pts->pts_timers[timerid] == NULL)
    565 			break;
    566 	if (timerid == TIMER_MAX) {
    567 		mutex_spin_exit(&timer_lock);
    568 		pool_put(&ptimer_pool, pt);
    569 		return EAGAIN;
    570 	}
    571 	if (evp == NULL) {
    572 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
    573 		switch (id) {
    574 		case CLOCK_REALTIME:
    575 			pt->pt_ev.sigev_signo = SIGALRM;
    576 			break;
    577 		case CLOCK_VIRTUAL:
    578 			pt->pt_ev.sigev_signo = SIGVTALRM;
    579 			break;
    580 		case CLOCK_PROF:
    581 			pt->pt_ev.sigev_signo = SIGPROF;
    582 			break;
    583 		}
    584 		pt->pt_ev.sigev_value.sival_int = timerid;
    585 	}
    586 	pt->pt_info.ksi_signo = pt->pt_ev.sigev_signo;
    587 	pt->pt_info.ksi_errno = 0;
    588 	pt->pt_info.ksi_code = 0;
    589 	pt->pt_info.ksi_pid = p->p_pid;
    590 	pt->pt_info.ksi_uid = kauth_cred_getuid(l->l_cred);
    591 	pt->pt_info.ksi_value = pt->pt_ev.sigev_value;
    592 	pt->pt_type = id;
    593 	pt->pt_proc = p;
    594 	pt->pt_overruns = 0;
    595 	pt->pt_poverruns = 0;
    596 	pt->pt_entry = timerid;
    597 	pt->pt_queued = false;
    598 	pt->pt_active = 0;
    599 	timerclear(&pt->pt_time.it_value);
    600 	callout_init(&pt->pt_ch, 0);
    601 	pts->pts_timers[timerid] = pt;
    602 	mutex_spin_exit(&timer_lock);
    603 
    604 	return copyout(&timerid, tid, sizeof(timerid));
    605 }
    606 
    607 /* Delete a POSIX realtime timer */
    608 int
    609 sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap,
    610     register_t *retval)
    611 {
    612 	/* {
    613 		syscallarg(timer_t) timerid;
    614 	} */
    615 	struct proc *p = l->l_proc;
    616 	timer_t timerid;
    617 	struct ptimers *pts;
    618 	struct ptimer *pt, *ptn;
    619 
    620 	timerid = SCARG(uap, timerid);
    621 	pts = p->p_timers;
    622 
    623 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    624 		return (EINVAL);
    625 
    626 	mutex_spin_enter(&timer_lock);
    627 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    628 		mutex_spin_exit(&timer_lock);
    629 		return (EINVAL);
    630 	}
    631 	if (pt->pt_active) {
    632 		ptn = LIST_NEXT(pt, pt_list);
    633 		LIST_REMOVE(pt, pt_list);
    634 		for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
    635 			timeradd(&pt->pt_time.it_value, &ptn->pt_time.it_value,
    636 			    &ptn->pt_time.it_value);
    637 		pt->pt_active = 0;
    638 	}
    639 	itimerfree(pts, timerid);
    640 
    641 	return (0);
    642 }
    643 
    644 /*
    645  * Set up the given timer. The value in pt->pt_time.it_value is taken
    646  * to be an absolute time for CLOCK_REALTIME timers and a relative
    647  * time for virtual timers.
    648  * Must be called at splclock().
    649  */
    650 void
    651 timer_settime(struct ptimer *pt)
    652 {
    653 	struct ptimer *ptn, *pptn;
    654 	struct ptlist *ptl;
    655 
    656 	KASSERT(mutex_owned(&timer_lock));
    657 
    658 	if (pt->pt_type == CLOCK_REALTIME) {
    659 		callout_stop(&pt->pt_ch);
    660 		if (timerisset(&pt->pt_time.it_value)) {
    661 			/*
    662 			 * Don't need to check hzto() return value, here.
    663 			 * callout_reset() does it for us.
    664 			 */
    665 			callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value),
    666 			    realtimerexpire, pt);
    667 		}
    668 	} else {
    669 		if (pt->pt_active) {
    670 			ptn = LIST_NEXT(pt, pt_list);
    671 			LIST_REMOVE(pt, pt_list);
    672 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
    673 				timeradd(&pt->pt_time.it_value,
    674 				    &ptn->pt_time.it_value,
    675 				    &ptn->pt_time.it_value);
    676 		}
    677 		if (timerisset(&pt->pt_time.it_value)) {
    678 			if (pt->pt_type == CLOCK_VIRTUAL)
    679 				ptl = &pt->pt_proc->p_timers->pts_virtual;
    680 			else
    681 				ptl = &pt->pt_proc->p_timers->pts_prof;
    682 
    683 			for (ptn = LIST_FIRST(ptl), pptn = NULL;
    684 			     ptn && timercmp(&pt->pt_time.it_value,
    685 				 &ptn->pt_time.it_value, >);
    686 			     pptn = ptn, ptn = LIST_NEXT(ptn, pt_list))
    687 				timersub(&pt->pt_time.it_value,
    688 				    &ptn->pt_time.it_value,
    689 				    &pt->pt_time.it_value);
    690 
    691 			if (pptn)
    692 				LIST_INSERT_AFTER(pptn, pt, pt_list);
    693 			else
    694 				LIST_INSERT_HEAD(ptl, pt, pt_list);
    695 
    696 			for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list))
    697 				timersub(&ptn->pt_time.it_value,
    698 				    &pt->pt_time.it_value,
    699 				    &ptn->pt_time.it_value);
    700 
    701 			pt->pt_active = 1;
    702 		} else
    703 			pt->pt_active = 0;
    704 	}
    705 }
    706 
    707 void
    708 timer_gettime(struct ptimer *pt, struct itimerval *aitv)
    709 {
    710 	struct timeval now;
    711 	struct ptimer *ptn;
    712 
    713 	KASSERT(mutex_owned(&timer_lock));
    714 
    715 	*aitv = pt->pt_time;
    716 	if (pt->pt_type == CLOCK_REALTIME) {
    717 		/*
    718 		 * Convert from absolute to relative time in .it_value
    719 		 * part of real time timer.  If time for real time
    720 		 * timer has passed return 0, else return difference
    721 		 * between current time and time for the timer to go
    722 		 * off.
    723 		 */
    724 		if (timerisset(&aitv->it_value)) {
    725 			getmicrotime(&now);
    726 			if (timercmp(&aitv->it_value, &now, <))
    727 				timerclear(&aitv->it_value);
    728 			else
    729 				timersub(&aitv->it_value, &now,
    730 				    &aitv->it_value);
    731 		}
    732 	} else if (pt->pt_active) {
    733 		if (pt->pt_type == CLOCK_VIRTUAL)
    734 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual);
    735 		else
    736 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof);
    737 		for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list))
    738 			timeradd(&aitv->it_value,
    739 			    &ptn->pt_time.it_value, &aitv->it_value);
    740 		KASSERT(ptn != NULL); /* pt should be findable on the list */
    741 	} else
    742 		timerclear(&aitv->it_value);
    743 }
    744 
    745 
    746 
    747 /* Set and arm a POSIX realtime timer */
    748 int
    749 sys_timer_settime(struct lwp *l, const struct sys_timer_settime_args *uap,
    750     register_t *retval)
    751 {
    752 	/* {
    753 		syscallarg(timer_t) timerid;
    754 		syscallarg(int) flags;
    755 		syscallarg(const struct itimerspec *) value;
    756 		syscallarg(struct itimerspec *) ovalue;
    757 	} */
    758 	int error;
    759 	struct itimerspec value, ovalue, *ovp = NULL;
    760 
    761 	if ((error = copyin(SCARG(uap, value), &value,
    762 	    sizeof(struct itimerspec))) != 0)
    763 		return (error);
    764 
    765 	if (SCARG(uap, ovalue))
    766 		ovp = &ovalue;
    767 
    768 	if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp,
    769 	    SCARG(uap, flags), l->l_proc)) != 0)
    770 		return error;
    771 
    772 	if (ovp)
    773 		return copyout(&ovalue, SCARG(uap, ovalue),
    774 		    sizeof(struct itimerspec));
    775 	return 0;
    776 }
    777 
    778 int
    779 dotimer_settime(int timerid, struct itimerspec *value,
    780     struct itimerspec *ovalue, int flags, struct proc *p)
    781 {
    782 	struct timeval now;
    783 	struct itimerval val, oval;
    784 	struct ptimers *pts;
    785 	struct ptimer *pt;
    786 
    787 	pts = p->p_timers;
    788 
    789 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    790 		return EINVAL;
    791 	TIMESPEC_TO_TIMEVAL(&val.it_value, &value->it_value);
    792 	TIMESPEC_TO_TIMEVAL(&val.it_interval, &value->it_interval);
    793 	if (itimerfix(&val.it_value) || itimerfix(&val.it_interval))
    794 		return (EINVAL);
    795 
    796 	mutex_spin_enter(&timer_lock);
    797 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    798 		mutex_spin_exit(&timer_lock);
    799 		return (EINVAL);
    800 	}
    801 
    802 	oval = pt->pt_time;
    803 	pt->pt_time = val;
    804 
    805 	/*
    806 	 * If we've been passed a relative time for a realtime timer,
    807 	 * convert it to absolute; if an absolute time for a virtual
    808 	 * timer, convert it to relative and make sure we don't set it
    809 	 * to zero, which would cancel the timer, or let it go
    810 	 * negative, which would confuse the comparison tests.
    811 	 */
    812 	if (timerisset(&pt->pt_time.it_value)) {
    813 		if (pt->pt_type == CLOCK_REALTIME) {
    814 			if ((flags & TIMER_ABSTIME) == 0) {
    815 				getmicrotime(&now);
    816 				timeradd(&pt->pt_time.it_value, &now,
    817 				    &pt->pt_time.it_value);
    818 			}
    819 		} else {
    820 			if ((flags & TIMER_ABSTIME) != 0) {
    821 				getmicrotime(&now);
    822 				timersub(&pt->pt_time.it_value, &now,
    823 				    &pt->pt_time.it_value);
    824 				if (!timerisset(&pt->pt_time.it_value) ||
    825 				    pt->pt_time.it_value.tv_sec < 0) {
    826 					pt->pt_time.it_value.tv_sec = 0;
    827 					pt->pt_time.it_value.tv_usec = 1;
    828 				}
    829 			}
    830 		}
    831 	}
    832 
    833 	timer_settime(pt);
    834 	mutex_spin_exit(&timer_lock);
    835 
    836 	if (ovalue) {
    837 		TIMEVAL_TO_TIMESPEC(&oval.it_value, &ovalue->it_value);
    838 		TIMEVAL_TO_TIMESPEC(&oval.it_interval, &ovalue->it_interval);
    839 	}
    840 
    841 	return (0);
    842 }
    843 
    844 /* Return the time remaining until a POSIX timer fires. */
    845 int
    846 sys_timer_gettime(struct lwp *l, const struct sys_timer_gettime_args *uap,
    847     register_t *retval)
    848 {
    849 	/* {
    850 		syscallarg(timer_t) timerid;
    851 		syscallarg(struct itimerspec *) value;
    852 	} */
    853 	struct itimerspec its;
    854 	int error;
    855 
    856 	if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc,
    857 	    &its)) != 0)
    858 		return error;
    859 
    860 	return copyout(&its, SCARG(uap, value), sizeof(its));
    861 }
    862 
    863 int
    864 dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its)
    865 {
    866 	struct ptimer *pt;
    867 	struct ptimers *pts;
    868 	struct itimerval aitv;
    869 
    870 	pts = p->p_timers;
    871 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    872 		return (EINVAL);
    873 	mutex_spin_enter(&timer_lock);
    874 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    875 		mutex_spin_exit(&timer_lock);
    876 		return (EINVAL);
    877 	}
    878 	timer_gettime(pt, &aitv);
    879 	mutex_spin_exit(&timer_lock);
    880 
    881 	TIMEVAL_TO_TIMESPEC(&aitv.it_interval, &its->it_interval);
    882 	TIMEVAL_TO_TIMESPEC(&aitv.it_value, &its->it_value);
    883 
    884 	return 0;
    885 }
    886 
    887 /*
    888  * Return the count of the number of times a periodic timer expired
    889  * while a notification was already pending. The counter is reset when
    890  * a timer expires and a notification can be posted.
    891  */
    892 int
    893 sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap,
    894     register_t *retval)
    895 {
    896 	/* {
    897 		syscallarg(timer_t) timerid;
    898 	} */
    899 	struct proc *p = l->l_proc;
    900 	struct ptimers *pts;
    901 	int timerid;
    902 	struct ptimer *pt;
    903 
    904 	timerid = SCARG(uap, timerid);
    905 
    906 	pts = p->p_timers;
    907 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    908 		return (EINVAL);
    909 	mutex_spin_enter(&timer_lock);
    910 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    911 		mutex_spin_exit(&timer_lock);
    912 		return (EINVAL);
    913 	}
    914 	*retval = pt->pt_poverruns;
    915 	mutex_spin_exit(&timer_lock);
    916 
    917 	return (0);
    918 }
    919 
    920 /*
    921  * Real interval timer expired:
    922  * send process whose timer expired an alarm signal.
    923  * If time is not set up to reload, then just return.
    924  * Else compute next time timer should go off which is > current time.
    925  * This is where delay in processing this timeout causes multiple
    926  * SIGALRM calls to be compressed into one.
    927  */
    928 void
    929 realtimerexpire(void *arg)
    930 {
    931 	struct timeval now;
    932 	struct ptimer *pt;
    933 
    934 	pt = arg;
    935 
    936 	mutex_spin_enter(&timer_lock);
    937 	itimerfire(pt);
    938 
    939 	if (!timerisset(&pt->pt_time.it_interval)) {
    940 		timerclear(&pt->pt_time.it_value);
    941 		mutex_spin_exit(&timer_lock);
    942 		return;
    943 	}
    944 	for (;;) {
    945 		timeradd(&pt->pt_time.it_value,
    946 		    &pt->pt_time.it_interval, &pt->pt_time.it_value);
    947 		getmicrotime(&now);
    948 		if (timercmp(&pt->pt_time.it_value, &now, >)) {
    949 			/*
    950 			 * Don't need to check hzto() return value, here.
    951 			 * callout_reset() does it for us.
    952 			 */
    953 			callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value),
    954 			    realtimerexpire, pt);
    955 			mutex_spin_exit(&timer_lock);
    956 			return;
    957 		}
    958 		mutex_spin_exit(&timer_lock);
    959 		pt->pt_overruns++;
    960 		mutex_spin_enter(&timer_lock);
    961 	}
    962 }
    963 
    964 /* BSD routine to get the value of an interval timer. */
    965 /* ARGSUSED */
    966 int
    967 sys_getitimer(struct lwp *l, const struct sys_getitimer_args *uap,
    968     register_t *retval)
    969 {
    970 	/* {
    971 		syscallarg(int) which;
    972 		syscallarg(struct itimerval *) itv;
    973 	} */
    974 	struct proc *p = l->l_proc;
    975 	struct itimerval aitv;
    976 	int error;
    977 
    978 	error = dogetitimer(p, SCARG(uap, which), &aitv);
    979 	if (error)
    980 		return error;
    981 	return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)));
    982 }
    983 
    984 int
    985 dogetitimer(struct proc *p, int which, struct itimerval *itvp)
    986 {
    987 	struct ptimers *pts;
    988 	struct ptimer *pt;
    989 
    990 	if ((u_int)which > ITIMER_PROF)
    991 		return (EINVAL);
    992 
    993 	mutex_spin_enter(&timer_lock);
    994 	pts = p->p_timers;
    995 	if (pts == NULL || (pt = pts->pts_timers[which]) == NULL) {
    996 		timerclear(&itvp->it_value);
    997 		timerclear(&itvp->it_interval);
    998 	} else
    999 		timer_gettime(pt, itvp);
   1000 	mutex_spin_exit(&timer_lock);
   1001 
   1002 	return 0;
   1003 }
   1004 
   1005 /* BSD routine to set/arm an interval timer. */
   1006 /* ARGSUSED */
   1007 int
   1008 sys_setitimer(struct lwp *l, const struct sys_setitimer_args *uap,
   1009     register_t *retval)
   1010 {
   1011 	/* {
   1012 		syscallarg(int) which;
   1013 		syscallarg(const struct itimerval *) itv;
   1014 		syscallarg(struct itimerval *) oitv;
   1015 	} */
   1016 	struct proc *p = l->l_proc;
   1017 	int which = SCARG(uap, which);
   1018 	struct sys_getitimer_args getargs;
   1019 	const struct itimerval *itvp;
   1020 	struct itimerval aitv;
   1021 	int error;
   1022 
   1023 	if ((u_int)which > ITIMER_PROF)
   1024 		return (EINVAL);
   1025 	itvp = SCARG(uap, itv);
   1026 	if (itvp &&
   1027 	    (error = copyin(itvp, &aitv, sizeof(struct itimerval)) != 0))
   1028 		return (error);
   1029 	if (SCARG(uap, oitv) != NULL) {
   1030 		SCARG(&getargs, which) = which;
   1031 		SCARG(&getargs, itv) = SCARG(uap, oitv);
   1032 		if ((error = sys_getitimer(l, &getargs, retval)) != 0)
   1033 			return (error);
   1034 	}
   1035 	if (itvp == 0)
   1036 		return (0);
   1037 
   1038 	return dosetitimer(p, which, &aitv);
   1039 }
   1040 
   1041 int
   1042 dosetitimer(struct proc *p, int which, struct itimerval *itvp)
   1043 {
   1044 	struct timeval now;
   1045 	struct ptimers *pts;
   1046 	struct ptimer *pt, *spare;
   1047 
   1048 	if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval))
   1049 		return (EINVAL);
   1050 
   1051 	/*
   1052 	 * Don't bother allocating data structures if the process just
   1053 	 * wants to clear the timer.
   1054 	 */
   1055 	spare = NULL;
   1056 	pts = p->p_timers;
   1057  retry:
   1058 	if (!timerisset(&itvp->it_value) && (pts == NULL ||
   1059 	    pts->pts_timers[which] == NULL))
   1060 		return (0);
   1061 	if (pts == NULL)
   1062 		pts = timers_alloc(p);
   1063 	mutex_spin_enter(&timer_lock);
   1064 	pt = pts->pts_timers[which];
   1065 	if (pt == NULL) {
   1066 		if (spare == NULL) {
   1067 			mutex_spin_exit(&timer_lock);
   1068 			spare = pool_get(&ptimer_pool, PR_WAITOK);
   1069 			goto retry;
   1070 		}
   1071 		pt = spare;
   1072 		spare = NULL;
   1073 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
   1074 		pt->pt_ev.sigev_value.sival_int = which;
   1075 		pt->pt_overruns = 0;
   1076 		pt->pt_proc = p;
   1077 		pt->pt_type = which;
   1078 		pt->pt_entry = which;
   1079 		pt->pt_active = 0;
   1080 		pt->pt_queued = false;
   1081 		callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
   1082 		switch (which) {
   1083 		case ITIMER_REAL:
   1084 			pt->pt_ev.sigev_signo = SIGALRM;
   1085 			break;
   1086 		case ITIMER_VIRTUAL:
   1087 			pt->pt_ev.sigev_signo = SIGVTALRM;
   1088 			break;
   1089 		case ITIMER_PROF:
   1090 			pt->pt_ev.sigev_signo = SIGPROF;
   1091 			break;
   1092 		}
   1093 		pts->pts_timers[which] = pt;
   1094 	}
   1095 	pt->pt_time = *itvp;
   1096 
   1097 	if ((which == ITIMER_REAL) && timerisset(&pt->pt_time.it_value)) {
   1098 		/* Convert to absolute time */
   1099 		/* XXX need to wrap in splclock for timecounters case? */
   1100 		getmicrotime(&now);
   1101 		timeradd(&pt->pt_time.it_value, &now, &pt->pt_time.it_value);
   1102 	}
   1103 	timer_settime(pt);
   1104 	mutex_spin_exit(&timer_lock);
   1105 	if (spare != NULL)
   1106 		pool_put(&ptimer_pool, spare);
   1107 
   1108 	return (0);
   1109 }
   1110 
   1111 /* Utility routines to manage the array of pointers to timers. */
   1112 struct ptimers *
   1113 timers_alloc(struct proc *p)
   1114 {
   1115 	struct ptimers *pts;
   1116 	int i;
   1117 
   1118 	pts = pool_get(&ptimers_pool, PR_WAITOK);
   1119 	LIST_INIT(&pts->pts_virtual);
   1120 	LIST_INIT(&pts->pts_prof);
   1121 	for (i = 0; i < TIMER_MAX; i++)
   1122 		pts->pts_timers[i] = NULL;
   1123 	pts->pts_fired = 0;
   1124 	mutex_spin_enter(&timer_lock);
   1125 	if (p->p_timers == NULL) {
   1126 		p->p_timers = pts;
   1127 		mutex_spin_exit(&timer_lock);
   1128 		return pts;
   1129 	}
   1130 	mutex_spin_exit(&timer_lock);
   1131 	pool_put(&ptimers_pool, pts);
   1132 	return p->p_timers;
   1133 }
   1134 
   1135 /*
   1136  * Clean up the per-process timers. If "which" is set to TIMERS_ALL,
   1137  * then clean up all timers and free all the data structures. If
   1138  * "which" is set to TIMERS_POSIX, only clean up the timers allocated
   1139  * by timer_create(), not the BSD setitimer() timers, and only free the
   1140  * structure if none of those remain.
   1141  */
   1142 void
   1143 timers_free(struct proc *p, int which)
   1144 {
   1145 	struct ptimers *pts;
   1146 	struct ptimer *ptn;
   1147 	struct timeval tv;
   1148 	int i;
   1149 
   1150 	if (p->p_timers == NULL)
   1151 		return;
   1152 
   1153 	pts = p->p_timers;
   1154 	mutex_spin_enter(&timer_lock);
   1155 	if (which == TIMERS_ALL) {
   1156 		p->p_timers = NULL;
   1157 		i = 0;
   1158 	} else {
   1159 		timerclear(&tv);
   1160 		for (ptn = LIST_FIRST(&pts->pts_virtual);
   1161 		     ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL];
   1162 		     ptn = LIST_NEXT(ptn, pt_list))
   1163 			timeradd(&tv, &ptn->pt_time.it_value, &tv);
   1164 		LIST_FIRST(&pts->pts_virtual) = NULL;
   1165 		if (ptn) {
   1166 			timeradd(&tv, &ptn->pt_time.it_value,
   1167 			    &ptn->pt_time.it_value);
   1168 			LIST_INSERT_HEAD(&pts->pts_virtual, ptn, pt_list);
   1169 		}
   1170 		timerclear(&tv);
   1171 		for (ptn = LIST_FIRST(&pts->pts_prof);
   1172 		     ptn && ptn != pts->pts_timers[ITIMER_PROF];
   1173 		     ptn = LIST_NEXT(ptn, pt_list))
   1174 			timeradd(&tv, &ptn->pt_time.it_value, &tv);
   1175 		LIST_FIRST(&pts->pts_prof) = NULL;
   1176 		if (ptn) {
   1177 			timeradd(&tv, &ptn->pt_time.it_value,
   1178 			    &ptn->pt_time.it_value);
   1179 			LIST_INSERT_HEAD(&pts->pts_prof, ptn, pt_list);
   1180 		}
   1181 		i = 3;
   1182 	}
   1183 	for ( ; i < TIMER_MAX; i++) {
   1184 		if (pts->pts_timers[i] != NULL) {
   1185 			itimerfree(pts, i);
   1186 			mutex_spin_enter(&timer_lock);
   1187 		}
   1188 	}
   1189 	if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL &&
   1190 	    pts->pts_timers[2] == NULL) {
   1191 		p->p_timers = NULL;
   1192 		mutex_spin_exit(&timer_lock);
   1193 		pool_put(&ptimers_pool, pts);
   1194 	} else
   1195 		mutex_spin_exit(&timer_lock);
   1196 }
   1197 
   1198 static void
   1199 itimerfree(struct ptimers *pts, int index)
   1200 {
   1201 	struct ptimer *pt;
   1202 
   1203 	KASSERT(mutex_owned(&timer_lock));
   1204 
   1205 	pt = pts->pts_timers[index];
   1206 	pts->pts_timers[index] = NULL;
   1207 	if (pt->pt_type == CLOCK_REALTIME) {
   1208 		mutex_spin_exit(&timer_lock);
   1209 		callout_halt(&pt->pt_ch);
   1210 	} else if (pt->pt_queued) {
   1211 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
   1212 		mutex_spin_exit(&timer_lock);
   1213 	} else
   1214 		mutex_spin_exit(&timer_lock);
   1215 	callout_destroy(&pt->pt_ch);
   1216 	pool_put(&ptimer_pool, pt);
   1217 }
   1218 
   1219 /*
   1220  * Decrement an interval timer by a specified number
   1221  * of microseconds, which must be less than a second,
   1222  * i.e. < 1000000.  If the timer expires, then reload
   1223  * it.  In this case, carry over (usec - old value) to
   1224  * reduce the value reloaded into the timer so that
   1225  * the timer does not drift.  This routine assumes
   1226  * that it is called in a context where the timers
   1227  * on which it is operating cannot change in value.
   1228  */
   1229 static int
   1230 itimerdecr(struct ptimer *pt, int usec)
   1231 {
   1232 	struct itimerval *itp;
   1233 
   1234 	KASSERT(mutex_owned(&timer_lock));
   1235 
   1236 	itp = &pt->pt_time;
   1237 	if (itp->it_value.tv_usec < usec) {
   1238 		if (itp->it_value.tv_sec == 0) {
   1239 			/* expired, and already in next interval */
   1240 			usec -= itp->it_value.tv_usec;
   1241 			goto expire;
   1242 		}
   1243 		itp->it_value.tv_usec += 1000000;
   1244 		itp->it_value.tv_sec--;
   1245 	}
   1246 	itp->it_value.tv_usec -= usec;
   1247 	usec = 0;
   1248 	if (timerisset(&itp->it_value))
   1249 		return (1);
   1250 	/* expired, exactly at end of interval */
   1251 expire:
   1252 	if (timerisset(&itp->it_interval)) {
   1253 		itp->it_value = itp->it_interval;
   1254 		itp->it_value.tv_usec -= usec;
   1255 		if (itp->it_value.tv_usec < 0) {
   1256 			itp->it_value.tv_usec += 1000000;
   1257 			itp->it_value.tv_sec--;
   1258 		}
   1259 		timer_settime(pt);
   1260 	} else
   1261 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
   1262 	return (0);
   1263 }
   1264 
   1265 static void
   1266 itimerfire(struct ptimer *pt)
   1267 {
   1268 
   1269 	KASSERT(mutex_owned(&timer_lock));
   1270 
   1271 	/*
   1272 	 * XXX Can overrun, but we don't do signal queueing yet, anyway.
   1273 	 * XXX Relying on the clock interrupt is stupid.
   1274 	 */
   1275 	if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL || pt->pt_queued)
   1276 		return;
   1277 	TAILQ_INSERT_TAIL(&timer_queue, pt, pt_chain);
   1278 	pt->pt_queued = true;
   1279 	softint_schedule(timer_sih);
   1280 }
   1281 
   1282 void
   1283 timer_tick(lwp_t *l, bool user)
   1284 {
   1285 	struct ptimers *pts;
   1286 	struct ptimer *pt;
   1287 	proc_t *p;
   1288 
   1289 	p = l->l_proc;
   1290 	if (p->p_timers == NULL)
   1291 		return;
   1292 
   1293 	mutex_spin_enter(&timer_lock);
   1294 	if ((pts = l->l_proc->p_timers) != NULL) {
   1295 		/*
   1296 		 * Run current process's virtual and profile time, as needed.
   1297 		 */
   1298 		if (user && (pt = LIST_FIRST(&pts->pts_virtual)) != NULL)
   1299 			if (itimerdecr(pt, tick) == 0)
   1300 				itimerfire(pt);
   1301 		if ((pt = LIST_FIRST(&pts->pts_prof)) != NULL)
   1302 			if (itimerdecr(pt, tick) == 0)
   1303 				itimerfire(pt);
   1304 	}
   1305 	mutex_spin_exit(&timer_lock);
   1306 }
   1307 
   1308 static void
   1309 timer_intr(void *cookie)
   1310 {
   1311 	ksiginfo_t ksi;
   1312 	struct ptimer *pt;
   1313 	proc_t *p;
   1314 
   1315 	mutex_spin_enter(&timer_lock);
   1316 	while ((pt = TAILQ_FIRST(&timer_queue)) != NULL) {
   1317 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
   1318 		KASSERT(pt->pt_queued);
   1319 		pt->pt_queued = false;
   1320 
   1321 		if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL)
   1322 			continue;
   1323 		p = pt->pt_proc;
   1324 		if (pt->pt_proc->p_timers == NULL) {
   1325 			/* Process is dying. */
   1326 			continue;
   1327 		}
   1328 		if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) {
   1329 			pt->pt_overruns++;
   1330 			continue;
   1331 		}
   1332 
   1333 		KSI_INIT(&ksi);
   1334 		ksi.ksi_signo = pt->pt_ev.sigev_signo;
   1335 		ksi.ksi_code = SI_TIMER;
   1336 		ksi.ksi_value = pt->pt_ev.sigev_value;
   1337 		pt->pt_poverruns = pt->pt_overruns;
   1338 		pt->pt_overruns = 0;
   1339 		mutex_spin_exit(&timer_lock);
   1340 
   1341 		mutex_enter(&proclist_mutex);
   1342 		kpsignal(p, &ksi, NULL);
   1343 		mutex_exit(&proclist_mutex);
   1344 
   1345 		mutex_spin_enter(&timer_lock);
   1346 	}
   1347 	mutex_spin_exit(&timer_lock);
   1348 }
   1349 
   1350 /*
   1351  * ratecheck(): simple time-based rate-limit checking.  see ratecheck(9)
   1352  * for usage and rationale.
   1353  */
   1354 int
   1355 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
   1356 {
   1357 	struct timeval tv, delta;
   1358 	int rv = 0;
   1359 
   1360 	getmicrouptime(&tv);
   1361 	timersub(&tv, lasttime, &delta);
   1362 
   1363 	/*
   1364 	 * check for 0,0 is so that the message will be seen at least once,
   1365 	 * even if interval is huge.
   1366 	 */
   1367 	if (timercmp(&delta, mininterval, >=) ||
   1368 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
   1369 		*lasttime = tv;
   1370 		rv = 1;
   1371 	}
   1372 
   1373 	return (rv);
   1374 }
   1375 
   1376 /*
   1377  * ppsratecheck(): packets (or events) per second limitation.
   1378  */
   1379 int
   1380 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
   1381 {
   1382 	struct timeval tv, delta;
   1383 	int rv;
   1384 
   1385 	getmicrouptime(&tv);
   1386 	timersub(&tv, lasttime, &delta);
   1387 
   1388 	/*
   1389 	 * check for 0,0 is so that the message will be seen at least once.
   1390 	 * if more than one second have passed since the last update of
   1391 	 * lasttime, reset the counter.
   1392 	 *
   1393 	 * we do increment *curpps even in *curpps < maxpps case, as some may
   1394 	 * try to use *curpps for stat purposes as well.
   1395 	 */
   1396 	if ((lasttime->tv_sec == 0 && lasttime->tv_usec == 0) ||
   1397 	    delta.tv_sec >= 1) {
   1398 		*lasttime = tv;
   1399 		*curpps = 0;
   1400 	}
   1401 	if (maxpps < 0)
   1402 		rv = 1;
   1403 	else if (*curpps < maxpps)
   1404 		rv = 1;
   1405 	else
   1406 		rv = 0;
   1407 
   1408 #if 1 /*DIAGNOSTIC?*/
   1409 	/* be careful about wrap-around */
   1410 	if (*curpps + 1 > *curpps)
   1411 		*curpps = *curpps + 1;
   1412 #else
   1413 	/*
   1414 	 * assume that there's not too many calls to this function.
   1415 	 * not sure if the assumption holds, as it depends on *caller's*
   1416 	 * behavior, not the behavior of this function.
   1417 	 * IMHO it is wrong to make assumption on the caller's behavior,
   1418 	 * so the above #if is #if 1, not #ifdef DIAGNOSTIC.
   1419 	 */
   1420 	*curpps = *curpps + 1;
   1421 #endif
   1422 
   1423 	return (rv);
   1424 }
   1425