Home | History | Annotate | Line # | Download | only in kern
kern_time.c revision 1.170
      1 /*	$NetBSD: kern_time.c,v 1.170 2011/10/27 16:12:52 christos Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2000, 2004, 2005, 2007, 2008, 2009 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, and by Andrew Doran.
      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  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 1982, 1986, 1989, 1993
     34  *	The Regents of the University of California.  All rights reserved.
     35  *
     36  * Redistribution and use in source and binary forms, with or without
     37  * modification, are permitted provided that the following conditions
     38  * are met:
     39  * 1. Redistributions of source code must retain the above copyright
     40  *    notice, this list of conditions and the following disclaimer.
     41  * 2. Redistributions in binary form must reproduce the above copyright
     42  *    notice, this list of conditions and the following disclaimer in the
     43  *    documentation and/or other materials provided with the distribution.
     44  * 3. Neither the name of the University nor the names of its contributors
     45  *    may be used to endorse or promote products derived from this software
     46  *    without specific prior written permission.
     47  *
     48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     58  * SUCH DAMAGE.
     59  *
     60  *	@(#)kern_time.c	8.4 (Berkeley) 5/26/95
     61  */
     62 
     63 #include <sys/cdefs.h>
     64 __KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.170 2011/10/27 16:12:52 christos Exp $");
     65 
     66 #include <sys/param.h>
     67 #include <sys/resourcevar.h>
     68 #include <sys/kernel.h>
     69 #include <sys/systm.h>
     70 #include <sys/proc.h>
     71 #include <sys/vnode.h>
     72 #include <sys/signalvar.h>
     73 #include <sys/syslog.h>
     74 #include <sys/timetc.h>
     75 #include <sys/timex.h>
     76 #include <sys/kauth.h>
     77 #include <sys/mount.h>
     78 #include <sys/sa.h>
     79 #include <sys/savar.h>
     80 #include <sys/syscallargs.h>
     81 #include <sys/cpu.h>
     82 
     83 #include "opt_sa.h"
     84 
     85 static void	timer_intr(void *);
     86 static void	itimerfire(struct ptimer *);
     87 static void	itimerfree(struct ptimers *, int);
     88 
     89 kmutex_t	timer_lock;
     90 
     91 static void	*timer_sih;
     92 static TAILQ_HEAD(, ptimer) timer_queue;
     93 
     94 struct pool ptimer_pool, ptimers_pool;
     95 
     96 #define	CLOCK_VIRTUAL_P(clockid)	\
     97 	((clockid) == CLOCK_VIRTUAL || (clockid) == CLOCK_PROF)
     98 
     99 CTASSERT(ITIMER_REAL == CLOCK_REALTIME);
    100 CTASSERT(ITIMER_VIRTUAL == CLOCK_VIRTUAL);
    101 CTASSERT(ITIMER_PROF == CLOCK_PROF);
    102 CTASSERT(ITIMER_MONOTONIC == CLOCK_MONOTONIC);
    103 
    104 /*
    105  * Initialize timekeeping.
    106  */
    107 void
    108 time_init(void)
    109 {
    110 
    111 	pool_init(&ptimer_pool, sizeof(struct ptimer), 0, 0, 0, "ptimerpl",
    112 	    &pool_allocator_nointr, IPL_NONE);
    113 	pool_init(&ptimers_pool, sizeof(struct ptimers), 0, 0, 0, "ptimerspl",
    114 	    &pool_allocator_nointr, IPL_NONE);
    115 }
    116 
    117 void
    118 time_init2(void)
    119 {
    120 
    121 	TAILQ_INIT(&timer_queue);
    122 	mutex_init(&timer_lock, MUTEX_DEFAULT, IPL_SCHED);
    123 	timer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
    124 	    timer_intr, NULL);
    125 }
    126 
    127 /* Time of day and interval timer support.
    128  *
    129  * These routines provide the kernel entry points to get and set
    130  * the time-of-day and per-process interval timers.  Subroutines
    131  * here provide support for adding and subtracting timeval structures
    132  * and decrementing interval timers, optionally reloading the interval
    133  * timers when they expire.
    134  */
    135 
    136 /* This function is used by clock_settime and settimeofday */
    137 static int
    138 settime1(struct proc *p, const struct timespec *ts, bool check_kauth)
    139 {
    140 	struct timespec delta, now;
    141 	int s;
    142 
    143 	/* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */
    144 	s = splclock();
    145 	nanotime(&now);
    146 	timespecsub(ts, &now, &delta);
    147 
    148 	if (check_kauth && kauth_authorize_system(kauth_cred_get(),
    149 	    KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, __UNCONST(ts),
    150 	    &delta, KAUTH_ARG(check_kauth ? false : true)) != 0) {
    151 		splx(s);
    152 		return (EPERM);
    153 	}
    154 
    155 #ifdef notyet
    156 	if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */
    157 		splx(s);
    158 		return (EPERM);
    159 	}
    160 #endif
    161 
    162 	tc_setclock(ts);
    163 
    164 	timespecadd(&boottime, &delta, &boottime);
    165 
    166 	resettodr();
    167 	splx(s);
    168 
    169 	return (0);
    170 }
    171 
    172 int
    173 settime(struct proc *p, struct timespec *ts)
    174 {
    175 	return (settime1(p, ts, true));
    176 }
    177 
    178 /* ARGSUSED */
    179 int
    180 sys___clock_gettime50(struct lwp *l,
    181     const struct sys___clock_gettime50_args *uap, register_t *retval)
    182 {
    183 	/* {
    184 		syscallarg(clockid_t) clock_id;
    185 		syscallarg(struct timespec *) tp;
    186 	} */
    187 	int error;
    188 	struct timespec ats;
    189 
    190 	error = clock_gettime1(SCARG(uap, clock_id), &ats);
    191 	if (error != 0)
    192 		return error;
    193 
    194 	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
    195 }
    196 
    197 int
    198 clock_gettime1(clockid_t clock_id, struct timespec *ts)
    199 {
    200 
    201 	switch (clock_id) {
    202 	case CLOCK_REALTIME:
    203 		nanotime(ts);
    204 		break;
    205 	case CLOCK_MONOTONIC:
    206 		nanouptime(ts);
    207 		break;
    208 	default:
    209 		return EINVAL;
    210 	}
    211 
    212 	return 0;
    213 }
    214 
    215 /* ARGSUSED */
    216 int
    217 sys___clock_settime50(struct lwp *l,
    218     const struct sys___clock_settime50_args *uap, register_t *retval)
    219 {
    220 	/* {
    221 		syscallarg(clockid_t) clock_id;
    222 		syscallarg(const struct timespec *) tp;
    223 	} */
    224 	int error;
    225 	struct timespec ats;
    226 
    227 	if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
    228 		return error;
    229 
    230 	return clock_settime1(l->l_proc, SCARG(uap, clock_id), &ats, true);
    231 }
    232 
    233 
    234 int
    235 clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp,
    236     bool check_kauth)
    237 {
    238 	int error;
    239 
    240 	switch (clock_id) {
    241 	case CLOCK_REALTIME:
    242 		if ((error = settime1(p, tp, check_kauth)) != 0)
    243 			return (error);
    244 		break;
    245 	case CLOCK_MONOTONIC:
    246 		return (EINVAL);	/* read-only clock */
    247 	default:
    248 		return (EINVAL);
    249 	}
    250 
    251 	return 0;
    252 }
    253 
    254 int
    255 sys___clock_getres50(struct lwp *l, const struct sys___clock_getres50_args *uap,
    256     register_t *retval)
    257 {
    258 	/* {
    259 		syscallarg(clockid_t) clock_id;
    260 		syscallarg(struct timespec *) tp;
    261 	} */
    262 	struct timespec ts;
    263 	int error = 0;
    264 
    265 	if ((error = clock_getres1(SCARG(uap, clock_id), &ts)) != 0)
    266 		return error;
    267 
    268 	if (SCARG(uap, tp))
    269 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
    270 
    271 	return error;
    272 }
    273 
    274 int
    275 clock_getres1(clockid_t clock_id, struct timespec *ts)
    276 {
    277 
    278 	switch (clock_id) {
    279 	case CLOCK_REALTIME:
    280 	case CLOCK_MONOTONIC:
    281 		ts->tv_sec = 0;
    282 		if (tc_getfrequency() > 1000000000)
    283 			ts->tv_nsec = 1;
    284 		else
    285 			ts->tv_nsec = 1000000000 / tc_getfrequency();
    286 		break;
    287 	default:
    288 		return EINVAL;
    289 	}
    290 
    291 	return 0;
    292 }
    293 
    294 /* ARGSUSED */
    295 int
    296 sys___nanosleep50(struct lwp *l, const struct sys___nanosleep50_args *uap,
    297     register_t *retval)
    298 {
    299 	/* {
    300 		syscallarg(struct timespec *) rqtp;
    301 		syscallarg(struct timespec *) rmtp;
    302 	} */
    303 	struct timespec rmt, rqt;
    304 	int error, error1;
    305 
    306 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
    307 	if (error)
    308 		return (error);
    309 
    310 	error = nanosleep1(l, &rqt, SCARG(uap, rmtp) ? &rmt : NULL);
    311 	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
    312 		return error;
    313 
    314 	error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt));
    315 	return error1 ? error1 : error;
    316 }
    317 
    318 int
    319 nanosleep1(struct lwp *l, struct timespec *rqt, struct timespec *rmt)
    320 {
    321 	struct timespec rmtstart;
    322 	int error, timo;
    323 
    324 	if ((error = itimespecfix(rqt)) != 0)
    325 		return error;
    326 
    327 	timo = tstohz(rqt);
    328 	/*
    329 	 * Avoid inadvertantly sleeping forever
    330 	 */
    331 	if (timo == 0)
    332 		timo = 1;
    333 	getnanouptime(&rmtstart);
    334 again:
    335 	error = kpause("nanoslp", true, timo, NULL);
    336 	if (rmt != NULL || error == 0) {
    337 		struct timespec rmtend;
    338 		struct timespec t0;
    339 		struct timespec *t;
    340 
    341 		getnanouptime(&rmtend);
    342 		t = (rmt != NULL) ? rmt : &t0;
    343 		timespecsub(&rmtend, &rmtstart, t);
    344 		timespecsub(rqt, t, t);
    345 		if (t->tv_sec < 0)
    346 			timespecclear(t);
    347 		if (error == 0) {
    348 			timo = tstohz(t);
    349 			if (timo > 0)
    350 				goto again;
    351 		}
    352 	}
    353 
    354 	if (error == ERESTART)
    355 		error = EINTR;
    356 	if (error == EWOULDBLOCK)
    357 		error = 0;
    358 
    359 	return error;
    360 }
    361 
    362 /* ARGSUSED */
    363 int
    364 sys___gettimeofday50(struct lwp *l, const struct sys___gettimeofday50_args *uap,
    365     register_t *retval)
    366 {
    367 	/* {
    368 		syscallarg(struct timeval *) tp;
    369 		syscallarg(void *) tzp;		really "struct timezone *";
    370 	} */
    371 	struct timeval atv;
    372 	int error = 0;
    373 	struct timezone tzfake;
    374 
    375 	if (SCARG(uap, tp)) {
    376 		microtime(&atv);
    377 		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
    378 		if (error)
    379 			return (error);
    380 	}
    381 	if (SCARG(uap, tzp)) {
    382 		/*
    383 		 * NetBSD has no kernel notion of time zone, so we just
    384 		 * fake up a timezone struct and return it if demanded.
    385 		 */
    386 		tzfake.tz_minuteswest = 0;
    387 		tzfake.tz_dsttime = 0;
    388 		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
    389 	}
    390 	return (error);
    391 }
    392 
    393 /* ARGSUSED */
    394 int
    395 sys___settimeofday50(struct lwp *l, const struct sys___settimeofday50_args *uap,
    396     register_t *retval)
    397 {
    398 	/* {
    399 		syscallarg(const struct timeval *) tv;
    400 		syscallarg(const void *) tzp; really "const struct timezone *";
    401 	} */
    402 
    403 	return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true);
    404 }
    405 
    406 int
    407 settimeofday1(const struct timeval *utv, bool userspace,
    408     const void *utzp, struct lwp *l, bool check_kauth)
    409 {
    410 	struct timeval atv;
    411 	struct timespec ts;
    412 	int error;
    413 
    414 	/* Verify all parameters before changing time. */
    415 
    416 	/*
    417 	 * NetBSD has no kernel notion of time zone, and only an
    418 	 * obsolete program would try to set it, so we log a warning.
    419 	 */
    420 	if (utzp)
    421 		log(LOG_WARNING, "pid %d attempted to set the "
    422 		    "(obsolete) kernel time zone\n", l->l_proc->p_pid);
    423 
    424 	if (utv == NULL)
    425 		return 0;
    426 
    427 	if (userspace) {
    428 		if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
    429 			return error;
    430 		utv = &atv;
    431 	}
    432 
    433 	TIMEVAL_TO_TIMESPEC(utv, &ts);
    434 	return settime1(l->l_proc, &ts, check_kauth);
    435 }
    436 
    437 int	time_adjusted;			/* set if an adjustment is made */
    438 
    439 /* ARGSUSED */
    440 int
    441 sys___adjtime50(struct lwp *l, const struct sys___adjtime50_args *uap,
    442     register_t *retval)
    443 {
    444 	/* {
    445 		syscallarg(const struct timeval *) delta;
    446 		syscallarg(struct timeval *) olddelta;
    447 	} */
    448 	int error = 0;
    449 	struct timeval atv, oldatv;
    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 	if (SCARG(uap, delta)) {
    456 		error = copyin(SCARG(uap, delta), &atv,
    457 		    sizeof(*SCARG(uap, delta)));
    458 		if (error)
    459 			return (error);
    460 	}
    461 	adjtime1(SCARG(uap, delta) ? &atv : NULL,
    462 	    SCARG(uap, olddelta) ? &oldatv : NULL, l->l_proc);
    463 	if (SCARG(uap, olddelta))
    464 		error = copyout(&oldatv, SCARG(uap, olddelta),
    465 		    sizeof(*SCARG(uap, olddelta)));
    466 	return error;
    467 }
    468 
    469 void
    470 adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p)
    471 {
    472 	extern int64_t time_adjtime;  /* in kern_ntptime.c */
    473 
    474 	if (olddelta) {
    475 		mutex_spin_enter(&timecounter_lock);
    476 		olddelta->tv_sec = time_adjtime / 1000000;
    477 		olddelta->tv_usec = time_adjtime % 1000000;
    478 		if (olddelta->tv_usec < 0) {
    479 			olddelta->tv_usec += 1000000;
    480 			olddelta->tv_sec--;
    481 		}
    482 		mutex_spin_exit(&timecounter_lock);
    483 	}
    484 
    485 	if (delta) {
    486 		mutex_spin_enter(&timecounter_lock);
    487 		time_adjtime = delta->tv_sec * 1000000 + delta->tv_usec;
    488 
    489 		if (time_adjtime) {
    490 			/* We need to save the system time during shutdown */
    491 			time_adjusted |= 1;
    492 		}
    493 		mutex_spin_exit(&timecounter_lock);
    494 	}
    495 }
    496 
    497 /*
    498  * Interval timer support. Both the BSD getitimer() family and the POSIX
    499  * timer_*() family of routines are supported.
    500  *
    501  * All timers are kept in an array pointed to by p_timers, which is
    502  * allocated on demand - many processes don't use timers at all. The
    503  * first three elements in this array are reserved for the BSD timers:
    504  * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, element
    505  * 2 is ITIMER_PROF, and element 3 is ITIMER_MONOTONIC. The rest may be
    506  * allocated by the timer_create() syscall.
    507  *
    508  * Realtime timers are kept in the ptimer structure as an absolute
    509  * time; virtual time timers are kept as a linked list of deltas.
    510  * Virtual time timers are processed in the hardclock() routine of
    511  * kern_clock.c.  The real time timer is processed by a callout
    512  * routine, called from the softclock() routine.  Since a callout may
    513  * be delayed in real time due to interrupt processing in the system,
    514  * it is possible for the real time timeout routine (realtimeexpire,
    515  * given below), to be delayed in real time past when it is supposed
    516  * to occur.  It does not suffice, therefore, to reload the real timer
    517  * .it_value from the real time timers .it_interval.  Rather, we
    518  * compute the next time in absolute time the timer should go off.  */
    519 
    520 /* Allocate a POSIX realtime timer. */
    521 int
    522 sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap,
    523     register_t *retval)
    524 {
    525 	/* {
    526 		syscallarg(clockid_t) clock_id;
    527 		syscallarg(struct sigevent *) evp;
    528 		syscallarg(timer_t *) timerid;
    529 	} */
    530 
    531 	return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id),
    532 	    SCARG(uap, evp), copyin, l);
    533 }
    534 
    535 int
    536 timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp,
    537     copyin_t fetch_event, struct lwp *l)
    538 {
    539 	int error;
    540 	timer_t timerid;
    541 	struct ptimers *pts;
    542 	struct ptimer *pt;
    543 	struct proc *p;
    544 
    545 	p = l->l_proc;
    546 
    547 	if ((u_int)id > CLOCK_MONOTONIC)
    548 		return (EINVAL);
    549 
    550 	if ((pts = p->p_timers) == NULL)
    551 		pts = timers_alloc(p);
    552 
    553 	pt = pool_get(&ptimer_pool, PR_WAITOK);
    554 	if (evp != NULL) {
    555 		if (((error =
    556 		    (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) ||
    557 		    ((pt->pt_ev.sigev_notify < SIGEV_NONE) ||
    558 			(pt->pt_ev.sigev_notify > SIGEV_SA)) ||
    559 			(pt->pt_ev.sigev_notify == SIGEV_SIGNAL &&
    560 			 (pt->pt_ev.sigev_signo <= 0 ||
    561 			  pt->pt_ev.sigev_signo >= NSIG))) {
    562 			pool_put(&ptimer_pool, pt);
    563 			return (error ? error : EINVAL);
    564 		}
    565 	}
    566 
    567 	/* Find a free timer slot, skipping those reserved for setitimer(). */
    568 	mutex_spin_enter(&timer_lock);
    569 	for (timerid = 3; timerid < TIMER_MAX; timerid++)
    570 		if (pts->pts_timers[timerid] == NULL)
    571 			break;
    572 	if (timerid == TIMER_MAX) {
    573 		mutex_spin_exit(&timer_lock);
    574 		pool_put(&ptimer_pool, pt);
    575 		return EAGAIN;
    576 	}
    577 	if (evp == NULL) {
    578 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
    579 		switch (id) {
    580 		case CLOCK_REALTIME:
    581 		case CLOCK_MONOTONIC:
    582 			pt->pt_ev.sigev_signo = SIGALRM;
    583 			break;
    584 		case CLOCK_VIRTUAL:
    585 			pt->pt_ev.sigev_signo = SIGVTALRM;
    586 			break;
    587 		case CLOCK_PROF:
    588 			pt->pt_ev.sigev_signo = SIGPROF;
    589 			break;
    590 		}
    591 		pt->pt_ev.sigev_value.sival_int = timerid;
    592 	}
    593 	pt->pt_info.ksi_signo = pt->pt_ev.sigev_signo;
    594 	pt->pt_info.ksi_errno = 0;
    595 	pt->pt_info.ksi_code = 0;
    596 	pt->pt_info.ksi_pid = p->p_pid;
    597 	pt->pt_info.ksi_uid = kauth_cred_getuid(l->l_cred);
    598 	pt->pt_info.ksi_value = pt->pt_ev.sigev_value;
    599 	pt->pt_type = id;
    600 	pt->pt_proc = p;
    601 	pt->pt_overruns = 0;
    602 	pt->pt_poverruns = 0;
    603 	pt->pt_entry = timerid;
    604 	pt->pt_queued = false;
    605 	timespecclear(&pt->pt_time.it_value);
    606 	if (!CLOCK_VIRTUAL_P(id))
    607 		callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
    608 	else
    609 		pt->pt_active = 0;
    610 
    611 	pts->pts_timers[timerid] = pt;
    612 	mutex_spin_exit(&timer_lock);
    613 
    614 	return copyout(&timerid, tid, sizeof(timerid));
    615 }
    616 
    617 /* Delete a POSIX realtime timer */
    618 int
    619 sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap,
    620     register_t *retval)
    621 {
    622 	/* {
    623 		syscallarg(timer_t) timerid;
    624 	} */
    625 	struct proc *p = l->l_proc;
    626 	timer_t timerid;
    627 	struct ptimers *pts;
    628 	struct ptimer *pt, *ptn;
    629 
    630 	timerid = SCARG(uap, timerid);
    631 	pts = p->p_timers;
    632 
    633 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    634 		return (EINVAL);
    635 
    636 	mutex_spin_enter(&timer_lock);
    637 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    638 		mutex_spin_exit(&timer_lock);
    639 		return (EINVAL);
    640 	}
    641 	if (CLOCK_VIRTUAL_P(pt->pt_type)) {
    642 		if (pt->pt_active) {
    643 			ptn = LIST_NEXT(pt, pt_list);
    644 			LIST_REMOVE(pt, pt_list);
    645 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
    646 				timespecadd(&pt->pt_time.it_value,
    647 				    &ptn->pt_time.it_value,
    648 				    &ptn->pt_time.it_value);
    649 			pt->pt_active = 0;
    650 		}
    651 	}
    652 	itimerfree(pts, timerid);
    653 
    654 	return (0);
    655 }
    656 
    657 /*
    658  * Set up the given timer. The value in pt->pt_time.it_value is taken
    659  * to be an absolute time for CLOCK_REALTIME/CLOCK_MONOTONIC timers and
    660  * a relative time for CLOCK_VIRTUAL/CLOCK_PROF timers.
    661  */
    662 void
    663 timer_settime(struct ptimer *pt)
    664 {
    665 	struct ptimer *ptn, *pptn;
    666 	struct ptlist *ptl;
    667 
    668 	KASSERT(mutex_owned(&timer_lock));
    669 
    670 	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
    671 		callout_halt(&pt->pt_ch, &timer_lock);
    672 		if (timespecisset(&pt->pt_time.it_value)) {
    673 			/*
    674 			 * Don't need to check tshzto() return value, here.
    675 			 * callout_reset() does it for us.
    676 			 */
    677 			callout_reset(&pt->pt_ch, tshzto(&pt->pt_time.it_value),
    678 			    realtimerexpire, pt);
    679 		}
    680 	} else {
    681 		if (pt->pt_active) {
    682 			ptn = LIST_NEXT(pt, pt_list);
    683 			LIST_REMOVE(pt, pt_list);
    684 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
    685 				timespecadd(&pt->pt_time.it_value,
    686 				    &ptn->pt_time.it_value,
    687 				    &ptn->pt_time.it_value);
    688 		}
    689 		if (timespecisset(&pt->pt_time.it_value)) {
    690 			if (pt->pt_type == CLOCK_VIRTUAL)
    691 				ptl = &pt->pt_proc->p_timers->pts_virtual;
    692 			else
    693 				ptl = &pt->pt_proc->p_timers->pts_prof;
    694 
    695 			for (ptn = LIST_FIRST(ptl), pptn = NULL;
    696 			     ptn && timespeccmp(&pt->pt_time.it_value,
    697 				 &ptn->pt_time.it_value, >);
    698 			     pptn = ptn, ptn = LIST_NEXT(ptn, pt_list))
    699 				timespecsub(&pt->pt_time.it_value,
    700 				    &ptn->pt_time.it_value,
    701 				    &pt->pt_time.it_value);
    702 
    703 			if (pptn)
    704 				LIST_INSERT_AFTER(pptn, pt, pt_list);
    705 			else
    706 				LIST_INSERT_HEAD(ptl, pt, pt_list);
    707 
    708 			for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list))
    709 				timespecsub(&ptn->pt_time.it_value,
    710 				    &pt->pt_time.it_value,
    711 				    &ptn->pt_time.it_value);
    712 
    713 			pt->pt_active = 1;
    714 		} else
    715 			pt->pt_active = 0;
    716 	}
    717 }
    718 
    719 void
    720 timer_gettime(struct ptimer *pt, struct itimerspec *aits)
    721 {
    722 	struct timespec now;
    723 	struct ptimer *ptn;
    724 
    725 	KASSERT(mutex_owned(&timer_lock));
    726 
    727 	*aits = pt->pt_time;
    728 	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
    729 		/*
    730 		 * Convert from absolute to relative time in .it_value
    731 		 * part of real time timer.  If time for real time
    732 		 * timer has passed return 0, else return difference
    733 		 * between current time and time for the timer to go
    734 		 * off.
    735 		 */
    736 		if (timespecisset(&aits->it_value)) {
    737 			if (pt->pt_type == CLOCK_REALTIME) {
    738 				getnanotime(&now);
    739 			} else { /* CLOCK_MONOTONIC */
    740 				getnanouptime(&now);
    741 			}
    742 			if (timespeccmp(&aits->it_value, &now, <))
    743 				timespecclear(&aits->it_value);
    744 			else
    745 				timespecsub(&aits->it_value, &now,
    746 				    &aits->it_value);
    747 		}
    748 	} else if (pt->pt_active) {
    749 		if (pt->pt_type == CLOCK_VIRTUAL)
    750 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual);
    751 		else
    752 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof);
    753 		for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list))
    754 			timespecadd(&aits->it_value,
    755 			    &ptn->pt_time.it_value, &aits->it_value);
    756 		KASSERT(ptn != NULL); /* pt should be findable on the list */
    757 	} else
    758 		timespecclear(&aits->it_value);
    759 }
    760 
    761 
    762 
    763 /* Set and arm a POSIX realtime timer */
    764 int
    765 sys___timer_settime50(struct lwp *l,
    766     const struct sys___timer_settime50_args *uap,
    767     register_t *retval)
    768 {
    769 	/* {
    770 		syscallarg(timer_t) timerid;
    771 		syscallarg(int) flags;
    772 		syscallarg(const struct itimerspec *) value;
    773 		syscallarg(struct itimerspec *) ovalue;
    774 	} */
    775 	int error;
    776 	struct itimerspec value, ovalue, *ovp = NULL;
    777 
    778 	if ((error = copyin(SCARG(uap, value), &value,
    779 	    sizeof(struct itimerspec))) != 0)
    780 		return (error);
    781 
    782 	if (SCARG(uap, ovalue))
    783 		ovp = &ovalue;
    784 
    785 	if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp,
    786 	    SCARG(uap, flags), l->l_proc)) != 0)
    787 		return error;
    788 
    789 	if (ovp)
    790 		return copyout(&ovalue, SCARG(uap, ovalue),
    791 		    sizeof(struct itimerspec));
    792 	return 0;
    793 }
    794 
    795 int
    796 dotimer_settime(int timerid, struct itimerspec *value,
    797     struct itimerspec *ovalue, int flags, struct proc *p)
    798 {
    799 	struct timespec now;
    800 	struct itimerspec val, oval;
    801 	struct ptimers *pts;
    802 	struct ptimer *pt;
    803 	int error;
    804 
    805 	pts = p->p_timers;
    806 
    807 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    808 		return EINVAL;
    809 	val = *value;
    810 	if ((error = itimespecfix(&val.it_value)) != 0 ||
    811 	    (error = itimespecfix(&val.it_interval)) != 0)
    812 		return error;
    813 
    814 	mutex_spin_enter(&timer_lock);
    815 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    816 		mutex_spin_exit(&timer_lock);
    817 		return EINVAL;
    818 	}
    819 
    820 	oval = pt->pt_time;
    821 	pt->pt_time = val;
    822 
    823 	/*
    824 	 * If we've been passed a relative time for a realtime timer,
    825 	 * convert it to absolute; if an absolute time for a virtual
    826 	 * timer, convert it to relative and make sure we don't set it
    827 	 * to zero, which would cancel the timer, or let it go
    828 	 * negative, which would confuse the comparison tests.
    829 	 */
    830 	if (timespecisset(&pt->pt_time.it_value)) {
    831 		if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
    832 			if ((flags & TIMER_ABSTIME) == 0) {
    833 				if (pt->pt_type == CLOCK_REALTIME) {
    834 					getnanotime(&now);
    835 				} else { /* CLOCK_MONOTONIC */
    836 					getnanouptime(&now);
    837 				}
    838 				timespecadd(&pt->pt_time.it_value, &now,
    839 				    &pt->pt_time.it_value);
    840 			}
    841 		} else {
    842 			if ((flags & TIMER_ABSTIME) != 0) {
    843 				getnanotime(&now);
    844 				timespecsub(&pt->pt_time.it_value, &now,
    845 				    &pt->pt_time.it_value);
    846 				if (!timespecisset(&pt->pt_time.it_value) ||
    847 				    pt->pt_time.it_value.tv_sec < 0) {
    848 					pt->pt_time.it_value.tv_sec = 0;
    849 					pt->pt_time.it_value.tv_nsec = 1;
    850 				}
    851 			}
    852 		}
    853 	}
    854 
    855 	timer_settime(pt);
    856 	mutex_spin_exit(&timer_lock);
    857 
    858 	if (ovalue)
    859 		*ovalue = oval;
    860 
    861 	return (0);
    862 }
    863 
    864 /* Return the time remaining until a POSIX timer fires. */
    865 int
    866 sys___timer_gettime50(struct lwp *l,
    867     const struct sys___timer_gettime50_args *uap, register_t *retval)
    868 {
    869 	/* {
    870 		syscallarg(timer_t) timerid;
    871 		syscallarg(struct itimerspec *) value;
    872 	} */
    873 	struct itimerspec its;
    874 	int error;
    875 
    876 	if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc,
    877 	    &its)) != 0)
    878 		return error;
    879 
    880 	return copyout(&its, SCARG(uap, value), sizeof(its));
    881 }
    882 
    883 int
    884 dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its)
    885 {
    886 	struct ptimer *pt;
    887 	struct ptimers *pts;
    888 
    889 	pts = p->p_timers;
    890 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    891 		return (EINVAL);
    892 	mutex_spin_enter(&timer_lock);
    893 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    894 		mutex_spin_exit(&timer_lock);
    895 		return (EINVAL);
    896 	}
    897 	timer_gettime(pt, its);
    898 	mutex_spin_exit(&timer_lock);
    899 
    900 	return 0;
    901 }
    902 
    903 /*
    904  * Return the count of the number of times a periodic timer expired
    905  * while a notification was already pending. The counter is reset when
    906  * a timer expires and a notification can be posted.
    907  */
    908 int
    909 sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap,
    910     register_t *retval)
    911 {
    912 	/* {
    913 		syscallarg(timer_t) timerid;
    914 	} */
    915 	struct proc *p = l->l_proc;
    916 	struct ptimers *pts;
    917 	int timerid;
    918 	struct ptimer *pt;
    919 
    920 	timerid = SCARG(uap, timerid);
    921 
    922 	pts = p->p_timers;
    923 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    924 		return (EINVAL);
    925 	mutex_spin_enter(&timer_lock);
    926 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    927 		mutex_spin_exit(&timer_lock);
    928 		return (EINVAL);
    929 	}
    930 	*retval = pt->pt_poverruns;
    931 	mutex_spin_exit(&timer_lock);
    932 
    933 	return (0);
    934 }
    935 
    936 #ifdef KERN_SA
    937 /* Glue function that triggers an upcall; called from userret(). */
    938 void
    939 timerupcall(struct lwp *l)
    940 {
    941 	struct ptimers *pt = l->l_proc->p_timers;
    942 	struct proc *p = l->l_proc;
    943 	unsigned int i, fired, done;
    944 
    945 	KDASSERT(l->l_proc->p_sa);
    946 	/* Bail out if we do not own the virtual processor */
    947 	if (l->l_savp->savp_lwp != l)
    948 		return ;
    949 
    950 	mutex_enter(p->p_lock);
    951 
    952 	fired = pt->pts_fired;
    953 	done = 0;
    954 	while ((i = ffs(fired)) != 0) {
    955 		siginfo_t *si;
    956 		int mask = 1 << --i;
    957 		int f;
    958 
    959 		f = ~l->l_pflag & LP_SA_NOBLOCK;
    960 		l->l_pflag |= LP_SA_NOBLOCK;
    961 		si = siginfo_alloc(PR_WAITOK);
    962 		si->_info = pt->pts_timers[i]->pt_info.ksi_info;
    963 		if (sa_upcall(l, SA_UPCALL_SIGEV | SA_UPCALL_DEFER, NULL, l,
    964 		    sizeof(*si), si, siginfo_free) != 0) {
    965 			siginfo_free(si);
    966 			/* XXX What do we do here?? */
    967 		} else
    968 			done |= mask;
    969 		fired &= ~mask;
    970 		l->l_pflag ^= f;
    971 	}
    972 	pt->pts_fired &= ~done;
    973 	if (pt->pts_fired == 0)
    974 		l->l_proc->p_timerpend = 0;
    975 
    976 	mutex_exit(p->p_lock);
    977 }
    978 #endif /* KERN_SA */
    979 
    980 /*
    981  * Real interval timer expired:
    982  * send process whose timer expired an alarm signal.
    983  * If time is not set up to reload, then just return.
    984  * Else compute next time timer should go off which is > current time.
    985  * This is where delay in processing this timeout causes multiple
    986  * SIGALRM calls to be compressed into one.
    987  */
    988 void
    989 realtimerexpire(void *arg)
    990 {
    991 	uint64_t last_val, next_val, interval, now_ns;
    992 	struct timespec now, next;
    993 	struct ptimer *pt;
    994 	int backwards;
    995 
    996 	pt = arg;
    997 
    998 	mutex_spin_enter(&timer_lock);
    999 	itimerfire(pt);
   1000 
   1001 	if (!timespecisset(&pt->pt_time.it_interval)) {
   1002 		timespecclear(&pt->pt_time.it_value);
   1003 		mutex_spin_exit(&timer_lock);
   1004 		return;
   1005 	}
   1006 
   1007 	getnanotime(&now);
   1008 	backwards = (timespeccmp(&pt->pt_time.it_value, &now, >));
   1009 	timespecadd(&pt->pt_time.it_value, &pt->pt_time.it_interval, &next);
   1010 	/* Handle the easy case of non-overflown timers first. */
   1011 	if (!backwards && timespeccmp(&next, &now, >)) {
   1012 		pt->pt_time.it_value = next;
   1013 	} else {
   1014 		now_ns = timespec2ns(&now);
   1015 		last_val = timespec2ns(&pt->pt_time.it_value);
   1016 		interval = timespec2ns(&pt->pt_time.it_interval);
   1017 
   1018 		next_val = now_ns +
   1019 		    (now_ns - last_val + interval - 1) % interval;
   1020 
   1021 		if (backwards)
   1022 			next_val += interval;
   1023 		else
   1024 			pt->pt_overruns += (now_ns - last_val) / interval;
   1025 
   1026 		pt->pt_time.it_value.tv_sec = next_val / 1000000000;
   1027 		pt->pt_time.it_value.tv_nsec = next_val % 1000000000;
   1028 	}
   1029 
   1030 	/*
   1031 	 * Don't need to check tshzto() return value, here.
   1032 	 * callout_reset() does it for us.
   1033 	 */
   1034 	callout_reset(&pt->pt_ch, tshzto(&pt->pt_time.it_value),
   1035 	    realtimerexpire, pt);
   1036 	mutex_spin_exit(&timer_lock);
   1037 }
   1038 
   1039 /* BSD routine to get the value of an interval timer. */
   1040 /* ARGSUSED */
   1041 int
   1042 sys___getitimer50(struct lwp *l, const struct sys___getitimer50_args *uap,
   1043     register_t *retval)
   1044 {
   1045 	/* {
   1046 		syscallarg(int) which;
   1047 		syscallarg(struct itimerval *) itv;
   1048 	} */
   1049 	struct proc *p = l->l_proc;
   1050 	struct itimerval aitv;
   1051 	int error;
   1052 
   1053 	error = dogetitimer(p, SCARG(uap, which), &aitv);
   1054 	if (error)
   1055 		return error;
   1056 	return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)));
   1057 }
   1058 
   1059 int
   1060 dogetitimer(struct proc *p, int which, struct itimerval *itvp)
   1061 {
   1062 	struct ptimers *pts;
   1063 	struct ptimer *pt;
   1064 	struct itimerspec its;
   1065 
   1066 	if ((u_int)which > ITIMER_MONOTONIC)
   1067 		return (EINVAL);
   1068 
   1069 	mutex_spin_enter(&timer_lock);
   1070 	pts = p->p_timers;
   1071 	if (pts == NULL || (pt = pts->pts_timers[which]) == NULL) {
   1072 		timerclear(&itvp->it_value);
   1073 		timerclear(&itvp->it_interval);
   1074 	} else {
   1075 		timer_gettime(pt, &its);
   1076 		TIMESPEC_TO_TIMEVAL(&itvp->it_value, &its.it_value);
   1077 		TIMESPEC_TO_TIMEVAL(&itvp->it_interval, &its.it_interval);
   1078 	}
   1079 	mutex_spin_exit(&timer_lock);
   1080 
   1081 	return 0;
   1082 }
   1083 
   1084 /* BSD routine to set/arm an interval timer. */
   1085 /* ARGSUSED */
   1086 int
   1087 sys___setitimer50(struct lwp *l, const struct sys___setitimer50_args *uap,
   1088     register_t *retval)
   1089 {
   1090 	/* {
   1091 		syscallarg(int) which;
   1092 		syscallarg(const struct itimerval *) itv;
   1093 		syscallarg(struct itimerval *) oitv;
   1094 	} */
   1095 	struct proc *p = l->l_proc;
   1096 	int which = SCARG(uap, which);
   1097 	struct sys___getitimer50_args getargs;
   1098 	const struct itimerval *itvp;
   1099 	struct itimerval aitv;
   1100 	int error;
   1101 
   1102 	if ((u_int)which > ITIMER_MONOTONIC)
   1103 		return (EINVAL);
   1104 	itvp = SCARG(uap, itv);
   1105 	if (itvp &&
   1106 	    (error = copyin(itvp, &aitv, sizeof(struct itimerval)) != 0))
   1107 		return (error);
   1108 	if (SCARG(uap, oitv) != NULL) {
   1109 		SCARG(&getargs, which) = which;
   1110 		SCARG(&getargs, itv) = SCARG(uap, oitv);
   1111 		if ((error = sys___getitimer50(l, &getargs, retval)) != 0)
   1112 			return (error);
   1113 	}
   1114 	if (itvp == 0)
   1115 		return (0);
   1116 
   1117 	return dosetitimer(p, which, &aitv);
   1118 }
   1119 
   1120 int
   1121 dosetitimer(struct proc *p, int which, struct itimerval *itvp)
   1122 {
   1123 	struct timespec now;
   1124 	struct ptimers *pts;
   1125 	struct ptimer *pt, *spare;
   1126 
   1127 	KASSERT((u_int)which <= CLOCK_MONOTONIC);
   1128 	if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval))
   1129 		return (EINVAL);
   1130 
   1131 	/*
   1132 	 * Don't bother allocating data structures if the process just
   1133 	 * wants to clear the timer.
   1134 	 */
   1135 	spare = NULL;
   1136 	pts = p->p_timers;
   1137  retry:
   1138 	if (!timerisset(&itvp->it_value) && (pts == NULL ||
   1139 	    pts->pts_timers[which] == NULL))
   1140 		return (0);
   1141 	if (pts == NULL)
   1142 		pts = timers_alloc(p);
   1143 	mutex_spin_enter(&timer_lock);
   1144 	pt = pts->pts_timers[which];
   1145 	if (pt == NULL) {
   1146 		if (spare == NULL) {
   1147 			mutex_spin_exit(&timer_lock);
   1148 			spare = pool_get(&ptimer_pool, PR_WAITOK);
   1149 			goto retry;
   1150 		}
   1151 		pt = spare;
   1152 		spare = NULL;
   1153 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
   1154 		pt->pt_ev.sigev_value.sival_int = which;
   1155 		pt->pt_overruns = 0;
   1156 		pt->pt_proc = p;
   1157 		pt->pt_type = which;
   1158 		pt->pt_entry = which;
   1159 		pt->pt_queued = false;
   1160 		if (pt->pt_type == CLOCK_REALTIME)
   1161 			callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
   1162 		else
   1163 			pt->pt_active = 0;
   1164 
   1165 		switch (which) {
   1166 		case ITIMER_REAL:
   1167 		case ITIMER_MONOTONIC:
   1168 			pt->pt_ev.sigev_signo = SIGALRM;
   1169 			break;
   1170 		case ITIMER_VIRTUAL:
   1171 			pt->pt_ev.sigev_signo = SIGVTALRM;
   1172 			break;
   1173 		case ITIMER_PROF:
   1174 			pt->pt_ev.sigev_signo = SIGPROF;
   1175 			break;
   1176 		}
   1177 		pts->pts_timers[which] = pt;
   1178 	}
   1179 
   1180 	TIMEVAL_TO_TIMESPEC(&itvp->it_value, &pt->pt_time.it_value);
   1181 	TIMEVAL_TO_TIMESPEC(&itvp->it_interval, &pt->pt_time.it_interval);
   1182 
   1183 	if (timespecisset(&pt->pt_time.it_value)) {
   1184 		/* Convert to absolute time */
   1185 		/* XXX need to wrap in splclock for timecounters case? */
   1186 		switch (which) {
   1187 		case ITIMER_REAL:
   1188 			getnanotime(&now);
   1189 			timespecadd(&pt->pt_time.it_value, &now,
   1190 			    &pt->pt_time.it_value);
   1191 			break;
   1192 		case ITIMER_MONOTONIC:
   1193 			getnanouptime(&now);
   1194 			timespecadd(&pt->pt_time.it_value, &now,
   1195 			    &pt->pt_time.it_value);
   1196 			break;
   1197 		default:
   1198 			break;
   1199 		}
   1200 	}
   1201 	timer_settime(pt);
   1202 	mutex_spin_exit(&timer_lock);
   1203 	if (spare != NULL)
   1204 		pool_put(&ptimer_pool, spare);
   1205 
   1206 	return (0);
   1207 }
   1208 
   1209 /* Utility routines to manage the array of pointers to timers. */
   1210 struct ptimers *
   1211 timers_alloc(struct proc *p)
   1212 {
   1213 	struct ptimers *pts;
   1214 	int i;
   1215 
   1216 	pts = pool_get(&ptimers_pool, PR_WAITOK);
   1217 	LIST_INIT(&pts->pts_virtual);
   1218 	LIST_INIT(&pts->pts_prof);
   1219 	for (i = 0; i < TIMER_MAX; i++)
   1220 		pts->pts_timers[i] = NULL;
   1221 	pts->pts_fired = 0;
   1222 	mutex_spin_enter(&timer_lock);
   1223 	if (p->p_timers == NULL) {
   1224 		p->p_timers = pts;
   1225 		mutex_spin_exit(&timer_lock);
   1226 		return pts;
   1227 	}
   1228 	mutex_spin_exit(&timer_lock);
   1229 	pool_put(&ptimers_pool, pts);
   1230 	return p->p_timers;
   1231 }
   1232 
   1233 /*
   1234  * Clean up the per-process timers. If "which" is set to TIMERS_ALL,
   1235  * then clean up all timers and free all the data structures. If
   1236  * "which" is set to TIMERS_POSIX, only clean up the timers allocated
   1237  * by timer_create(), not the BSD setitimer() timers, and only free the
   1238  * structure if none of those remain.
   1239  */
   1240 void
   1241 timers_free(struct proc *p, int which)
   1242 {
   1243 	struct ptimers *pts;
   1244 	struct ptimer *ptn;
   1245 	struct timespec ts;
   1246 	int i;
   1247 
   1248 	if (p->p_timers == NULL)
   1249 		return;
   1250 
   1251 	pts = p->p_timers;
   1252 	mutex_spin_enter(&timer_lock);
   1253 	if (which == TIMERS_ALL) {
   1254 		p->p_timers = NULL;
   1255 		i = 0;
   1256 	} else {
   1257 		timespecclear(&ts);
   1258 		for (ptn = LIST_FIRST(&pts->pts_virtual);
   1259 		     ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL];
   1260 		     ptn = LIST_NEXT(ptn, pt_list)) {
   1261 			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
   1262 			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
   1263 		}
   1264 		LIST_FIRST(&pts->pts_virtual) = NULL;
   1265 		if (ptn) {
   1266 			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
   1267 			timespecadd(&ts, &ptn->pt_time.it_value,
   1268 			    &ptn->pt_time.it_value);
   1269 			LIST_INSERT_HEAD(&pts->pts_virtual, ptn, pt_list);
   1270 		}
   1271 		timespecclear(&ts);
   1272 		for (ptn = LIST_FIRST(&pts->pts_prof);
   1273 		     ptn && ptn != pts->pts_timers[ITIMER_PROF];
   1274 		     ptn = LIST_NEXT(ptn, pt_list)) {
   1275 			KASSERT(ptn->pt_type == CLOCK_PROF);
   1276 			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
   1277 		}
   1278 		LIST_FIRST(&pts->pts_prof) = NULL;
   1279 		if (ptn) {
   1280 			KASSERT(ptn->pt_type == CLOCK_PROF);
   1281 			timespecadd(&ts, &ptn->pt_time.it_value,
   1282 			    &ptn->pt_time.it_value);
   1283 			LIST_INSERT_HEAD(&pts->pts_prof, ptn, pt_list);
   1284 		}
   1285 		i = 3;
   1286 	}
   1287 	for ( ; i < TIMER_MAX; i++) {
   1288 		if (pts->pts_timers[i] != NULL) {
   1289 			itimerfree(pts, i);
   1290 			mutex_spin_enter(&timer_lock);
   1291 		}
   1292 	}
   1293 	if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL &&
   1294 	    pts->pts_timers[2] == NULL) {
   1295 		p->p_timers = NULL;
   1296 		mutex_spin_exit(&timer_lock);
   1297 		pool_put(&ptimers_pool, pts);
   1298 	} else
   1299 		mutex_spin_exit(&timer_lock);
   1300 }
   1301 
   1302 static void
   1303 itimerfree(struct ptimers *pts, int index)
   1304 {
   1305 	struct ptimer *pt;
   1306 
   1307 	KASSERT(mutex_owned(&timer_lock));
   1308 
   1309 	pt = pts->pts_timers[index];
   1310 	pts->pts_timers[index] = NULL;
   1311 	if (!CLOCK_VIRTUAL_P(pt->pt_type))
   1312 		callout_halt(&pt->pt_ch, &timer_lock);
   1313 	if (pt->pt_queued)
   1314 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
   1315 	mutex_spin_exit(&timer_lock);
   1316 	if (!CLOCK_VIRTUAL_P(pt->pt_type))
   1317 		callout_destroy(&pt->pt_ch);
   1318 	pool_put(&ptimer_pool, pt);
   1319 }
   1320 
   1321 /*
   1322  * Decrement an interval timer by a specified number
   1323  * of nanoseconds, which must be less than a second,
   1324  * i.e. < 1000000000.  If the timer expires, then reload
   1325  * it.  In this case, carry over (nsec - old value) to
   1326  * reduce the value reloaded into the timer so that
   1327  * the timer does not drift.  This routine assumes
   1328  * that it is called in a context where the timers
   1329  * on which it is operating cannot change in value.
   1330  */
   1331 static int
   1332 itimerdecr(struct ptimer *pt, int nsec)
   1333 {
   1334 	struct itimerspec *itp;
   1335 
   1336 	KASSERT(mutex_owned(&timer_lock));
   1337 	KASSERT(CLOCK_VIRTUAL_P(pt->pt_type));
   1338 
   1339 	itp = &pt->pt_time;
   1340 	if (itp->it_value.tv_nsec < nsec) {
   1341 		if (itp->it_value.tv_sec == 0) {
   1342 			/* expired, and already in next interval */
   1343 			nsec -= itp->it_value.tv_nsec;
   1344 			goto expire;
   1345 		}
   1346 		itp->it_value.tv_nsec += 1000000000;
   1347 		itp->it_value.tv_sec--;
   1348 	}
   1349 	itp->it_value.tv_nsec -= nsec;
   1350 	nsec = 0;
   1351 	if (timespecisset(&itp->it_value))
   1352 		return (1);
   1353 	/* expired, exactly at end of interval */
   1354 expire:
   1355 	if (timespecisset(&itp->it_interval)) {
   1356 		itp->it_value = itp->it_interval;
   1357 		itp->it_value.tv_nsec -= nsec;
   1358 		if (itp->it_value.tv_nsec < 0) {
   1359 			itp->it_value.tv_nsec += 1000000000;
   1360 			itp->it_value.tv_sec--;
   1361 		}
   1362 		timer_settime(pt);
   1363 	} else
   1364 		itp->it_value.tv_nsec = 0;		/* sec is already 0 */
   1365 	return (0);
   1366 }
   1367 
   1368 static void
   1369 itimerfire(struct ptimer *pt)
   1370 {
   1371 
   1372 	KASSERT(mutex_owned(&timer_lock));
   1373 
   1374 	/*
   1375 	 * XXX Can overrun, but we don't do signal queueing yet, anyway.
   1376 	 * XXX Relying on the clock interrupt is stupid.
   1377 	 */
   1378 	if ((pt->pt_ev.sigev_notify == SIGEV_SA && pt->pt_proc->p_sa == NULL) ||
   1379 	    (pt->pt_ev.sigev_notify != SIGEV_SIGNAL &&
   1380 	    pt->pt_ev.sigev_notify != SIGEV_SA) || pt->pt_queued)
   1381 		return;
   1382 	TAILQ_INSERT_TAIL(&timer_queue, pt, pt_chain);
   1383 	pt->pt_queued = true;
   1384 	softint_schedule(timer_sih);
   1385 }
   1386 
   1387 void
   1388 timer_tick(lwp_t *l, bool user)
   1389 {
   1390 	struct ptimers *pts;
   1391 	struct ptimer *pt;
   1392 	proc_t *p;
   1393 
   1394 	p = l->l_proc;
   1395 	if (p->p_timers == NULL)
   1396 		return;
   1397 
   1398 	mutex_spin_enter(&timer_lock);
   1399 	if ((pts = l->l_proc->p_timers) != NULL) {
   1400 		/*
   1401 		 * Run current process's virtual and profile time, as needed.
   1402 		 */
   1403 		if (user && (pt = LIST_FIRST(&pts->pts_virtual)) != NULL)
   1404 			if (itimerdecr(pt, tick * 1000) == 0)
   1405 				itimerfire(pt);
   1406 		if ((pt = LIST_FIRST(&pts->pts_prof)) != NULL)
   1407 			if (itimerdecr(pt, tick * 1000) == 0)
   1408 				itimerfire(pt);
   1409 	}
   1410 	mutex_spin_exit(&timer_lock);
   1411 }
   1412 
   1413 #ifdef KERN_SA
   1414 /*
   1415  * timer_sa_intr:
   1416  *
   1417  *	SIGEV_SA handling for timer_intr(). We are called (and return)
   1418  * with the timer lock held. We know that the process had SA enabled
   1419  * when this timer was enqueued. As timer_intr() is a soft interrupt
   1420  * handler, SA should still be enabled by the time we get here.
   1421  */
   1422 static void
   1423 timer_sa_intr(struct ptimer *pt, proc_t *p)
   1424 {
   1425 	unsigned int		i;
   1426 	struct sadata		*sa;
   1427 	struct sadata_vp	*vp;
   1428 
   1429 	/* Cause the process to generate an upcall when it returns. */
   1430 	if (!p->p_timerpend) {
   1431 		/*
   1432 		 * XXX stop signals can be processed inside tsleep,
   1433 		 * which can be inside sa_yield's inner loop, which
   1434 		 * makes testing for sa_idle alone insuffucent to
   1435 		 * determine if we really should call setrunnable.
   1436 		 */
   1437 		pt->pt_poverruns = pt->pt_overruns;
   1438 		pt->pt_overruns = 0;
   1439 		i = 1 << pt->pt_entry;
   1440 		p->p_timers->pts_fired = i;
   1441 		p->p_timerpend = 1;
   1442 
   1443 		sa = p->p_sa;
   1444 		mutex_enter(&sa->sa_mutex);
   1445 		SLIST_FOREACH(vp, &sa->sa_vps, savp_next) {
   1446 			struct lwp *vp_lwp = vp->savp_lwp;
   1447 			lwp_lock(vp_lwp);
   1448 			lwp_need_userret(vp_lwp);
   1449 			if (vp_lwp->l_flag & LW_SA_IDLE) {
   1450 				vp_lwp->l_flag &= ~LW_SA_IDLE;
   1451 				lwp_unsleep(vp_lwp, true);
   1452 				break;
   1453 			}
   1454 			lwp_unlock(vp_lwp);
   1455 		}
   1456 		mutex_exit(&sa->sa_mutex);
   1457 	} else {
   1458 		i = 1 << pt->pt_entry;
   1459 		if ((p->p_timers->pts_fired & i) == 0) {
   1460 			pt->pt_poverruns = pt->pt_overruns;
   1461 			pt->pt_overruns = 0;
   1462 			p->p_timers->pts_fired |= i;
   1463 		} else
   1464 			pt->pt_overruns++;
   1465 	}
   1466 }
   1467 #endif /* KERN_SA */
   1468 
   1469 static void
   1470 timer_intr(void *cookie)
   1471 {
   1472 	ksiginfo_t ksi;
   1473 	struct ptimer *pt;
   1474 	proc_t *p;
   1475 
   1476 	mutex_enter(proc_lock);
   1477 	mutex_spin_enter(&timer_lock);
   1478 	while ((pt = TAILQ_FIRST(&timer_queue)) != NULL) {
   1479 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
   1480 		KASSERT(pt->pt_queued);
   1481 		pt->pt_queued = false;
   1482 
   1483 		if (pt->pt_proc->p_timers == NULL) {
   1484 			/* Process is dying. */
   1485 			continue;
   1486 		}
   1487 		p = pt->pt_proc;
   1488 #ifdef KERN_SA
   1489 		if (pt->pt_ev.sigev_notify == SIGEV_SA) {
   1490 			timer_sa_intr(pt, p);
   1491 			continue;
   1492 		}
   1493 #endif /* KERN_SA */
   1494 		if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL)
   1495 			continue;
   1496 		if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) {
   1497 			pt->pt_overruns++;
   1498 			continue;
   1499 		}
   1500 
   1501 		KSI_INIT(&ksi);
   1502 		ksi.ksi_signo = pt->pt_ev.sigev_signo;
   1503 		ksi.ksi_code = SI_TIMER;
   1504 		ksi.ksi_value = pt->pt_ev.sigev_value;
   1505 		pt->pt_poverruns = pt->pt_overruns;
   1506 		pt->pt_overruns = 0;
   1507 		mutex_spin_exit(&timer_lock);
   1508 		kpsignal(p, &ksi, NULL);
   1509 		mutex_spin_enter(&timer_lock);
   1510 	}
   1511 	mutex_spin_exit(&timer_lock);
   1512 	mutex_exit(proc_lock);
   1513 }
   1514 
   1515 /*
   1516  * Check if the time will wrap if set to ts.
   1517  *
   1518  * ts - timespec describing the new time
   1519  * delta - the delta between the current time and ts
   1520  */
   1521 bool
   1522 time_wraps(struct timespec *ts, struct timespec *delta)
   1523 {
   1524 
   1525 	/*
   1526 	 * Don't allow the time to be set forward so far it
   1527 	 * will wrap and become negative, thus allowing an
   1528 	 * attacker to bypass the next check below.  The
   1529 	 * cutoff is 1 year before rollover occurs, so even
   1530 	 * if the attacker uses adjtime(2) to move the time
   1531 	 * past the cutoff, it will take a very long time
   1532 	 * to get to the wrap point.
   1533 	 */
   1534 	if ((ts->tv_sec > LLONG_MAX - 365*24*60*60) ||
   1535 	    (delta->tv_sec < 0 || delta->tv_nsec < 0))
   1536 		return true;
   1537 
   1538 	return false;
   1539 }
   1540