xref: /netbsd-src/sys/kern/kern_time.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: kern_time.c,v 1.108 2006/10/12 01:32:17 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2000, 2004, 2005 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 1.108 2006/10/12 01:32:17 christos Exp $");
72 
73 #include "fs_nfs.h"
74 #include "opt_nfs.h"
75 #include "opt_nfsserver.h"
76 
77 #include <sys/param.h>
78 #include <sys/resourcevar.h>
79 #include <sys/kernel.h>
80 #include <sys/systm.h>
81 #include <sys/proc.h>
82 #include <sys/sa.h>
83 #include <sys/savar.h>
84 #include <sys/vnode.h>
85 #include <sys/signalvar.h>
86 #include <sys/syslog.h>
87 #ifdef __HAVE_TIMECOUNTER
88 #include <sys/timetc.h>
89 #else /* !__HAVE_TIMECOUNTER */
90 #include <sys/timevar.h>
91 #endif /* !__HAVE_TIMECOUNTER */
92 #include <sys/kauth.h>
93 
94 #include <sys/mount.h>
95 #include <sys/syscallargs.h>
96 
97 #include <uvm/uvm_extern.h>
98 
99 #if defined(NFS) || defined(NFSSERVER)
100 #include <nfs/rpcv2.h>
101 #include <nfs/nfsproto.h>
102 #include <nfs/nfs.h>
103 #include <nfs/nfs_var.h>
104 #endif
105 
106 #include <machine/cpu.h>
107 
108 POOL_INIT(ptimer_pool, sizeof(struct ptimer), 0, 0, 0, "ptimerpl",
109     &pool_allocator_nointr);
110 POOL_INIT(ptimers_pool, sizeof(struct ptimers), 0, 0, 0, "ptimerspl",
111     &pool_allocator_nointr);
112 
113 static void timerupcall(struct lwp *, void *);
114 #ifdef __HAVE_TIMECOUNTER
115 static int itimespecfix(struct timespec *);		/* XXX move itimerfix to timespecs */
116 #endif /* __HAVE_TIMECOUNTER */
117 
118 /* Time of day and interval timer support.
119  *
120  * These routines provide the kernel entry points to get and set
121  * the time-of-day and per-process interval timers.  Subroutines
122  * here provide support for adding and subtracting timeval structures
123  * and decrementing interval timers, optionally reloading the interval
124  * timers when they expire.
125  */
126 
127 /* This function is used by clock_settime and settimeofday */
128 int
129 settime(struct proc *p, struct timespec *ts)
130 {
131 	struct timeval delta, tv;
132 #ifdef __HAVE_TIMECOUNTER
133 	struct timeval now;
134 	struct timespec ts1;
135 #endif /* !__HAVE_TIMECOUNTER */
136 	struct cpu_info *ci;
137 	int s;
138 
139 	/*
140 	 * Don't allow the time to be set forward so far it will wrap
141 	 * and become negative, thus allowing an attacker to bypass
142 	 * the next check below.  The cutoff is 1 year before rollover
143 	 * occurs, so even if the attacker uses adjtime(2) to move
144 	 * the time past the cutoff, it will take a very long time
145 	 * to get to the wrap point.
146 	 *
147 	 * XXX: we check against INT_MAX since on 64-bit
148 	 *	platforms, sizeof(int) != sizeof(long) and
149 	 *	time_t is 32 bits even when atv.tv_sec is 64 bits.
150 	 */
151 	if (ts->tv_sec > INT_MAX - 365*24*60*60) {
152 		struct proc *pp = p->p_pptr;
153 		log(LOG_WARNING, "pid %d (%s) "
154 		    "invoked by uid %d ppid %d (%s) "
155 		    "tried to set clock forward to %ld\n",
156 		    p->p_pid, p->p_comm, kauth_cred_geteuid(pp->p_cred),
157 		    pp->p_pid, pp->p_comm, (long)ts->tv_sec);
158 		return (EPERM);
159 	}
160 	TIMESPEC_TO_TIMEVAL(&tv, ts);
161 
162 	/* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */
163 	s = splclock();
164 #ifdef __HAVE_TIMECOUNTER
165 	microtime(&now);
166 	timersub(&tv, &now, &delta);
167 #else /* !__HAVE_TIMECOUNTER */
168 	timersub(&tv, &time, &delta);
169 #endif /* !__HAVE_TIMECOUNTER */
170 	if ((delta.tv_sec < 0 || delta.tv_usec < 0) &&
171 	    kauth_authorize_system(p->p_cred, KAUTH_SYSTEM_TIME,
172 	    KAUTH_REQ_SYSTEM_TIME_BACKWARDS, NULL, NULL, NULL)) {
173 		splx(s);
174 		return (EPERM);
175 	}
176 #ifdef notyet
177 	if ((delta.tv_sec < 86400) && securelevel > 0) {
178 		splx(s);
179 		return (EPERM);
180 	}
181 #endif
182 
183 #ifdef __HAVE_TIMECOUNTER
184 	TIMEVAL_TO_TIMESPEC(&tv, &ts1);
185 	tc_setclock(&ts1);
186 #else /* !__HAVE_TIMECOUNTER */
187 	time = tv;
188 #endif /* !__HAVE_TIMECOUNTER */
189 
190 	(void) spllowersoftclock();
191 
192 	timeradd(&boottime, &delta, &boottime);
193 
194 	/*
195 	 * XXXSMP
196 	 * This is wrong.  We should traverse a list of all
197 	 * CPUs and add the delta to the runtime of those
198 	 * CPUs which have a process on them.
199 	 */
200 	ci = curcpu();
201 	timeradd(&ci->ci_schedstate.spc_runtime, &delta,
202 	    &ci->ci_schedstate.spc_runtime);
203 #if (defined(NFS) && !defined (NFS_V2_ONLY)) || defined(NFSSERVER)
204 	nqnfs_lease_updatetime(delta.tv_sec);
205 #endif
206 	splx(s);
207 	resettodr();
208 	return (0);
209 }
210 
211 /* ARGSUSED */
212 int
213 sys_clock_gettime(struct lwp *l __unused, void *v, register_t *retval __unused)
214 {
215 	struct sys_clock_gettime_args /* {
216 		syscallarg(clockid_t) clock_id;
217 		syscallarg(struct timespec *) tp;
218 	} */ *uap = v;
219 	clockid_t clock_id;
220 	struct timespec ats;
221 
222 	clock_id = SCARG(uap, clock_id);
223 	switch (clock_id) {
224 	case CLOCK_REALTIME:
225 		nanotime(&ats);
226 		break;
227 	case CLOCK_MONOTONIC:
228 #ifdef __HAVE_TIMECOUNTER
229 		nanouptime(&ats);
230 #else /* !__HAVE_TIMECOUNTER */
231 		{
232 		int s;
233 
234 		/* XXX "hz" granularity */
235 		s = splclock();
236 		TIMEVAL_TO_TIMESPEC(&mono_time,&ats);
237 		splx(s);
238 		}
239 #endif /* !__HAVE_TIMECOUNTER */
240 		break;
241 	default:
242 		return (EINVAL);
243 	}
244 
245 	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
246 }
247 
248 /* ARGSUSED */
249 int
250 sys_clock_settime(struct lwp *l, void *v, register_t *retval __unused)
251 {
252 	struct sys_clock_settime_args /* {
253 		syscallarg(clockid_t) clock_id;
254 		syscallarg(const struct timespec *) tp;
255 	} */ *uap = v;
256 	int error;
257 
258 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
259 	    KAUTH_REQ_SYSTEM_TIME_SYSTEM, NULL, NULL, NULL)) != 0)
260 		return (error);
261 
262 	return clock_settime1(l->l_proc, SCARG(uap, clock_id), SCARG(uap, tp));
263 }
264 
265 
266 int
267 clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp)
268 {
269 	struct timespec ats;
270 	int error;
271 
272 	if ((error = copyin(tp, &ats, sizeof(ats))) != 0)
273 		return (error);
274 
275 	switch (clock_id) {
276 	case CLOCK_REALTIME:
277 		if ((error = settime(p, &ats)) != 0)
278 			return (error);
279 		break;
280 	case CLOCK_MONOTONIC:
281 		return (EINVAL);	/* read-only clock */
282 	default:
283 		return (EINVAL);
284 	}
285 
286 	return 0;
287 }
288 
289 int
290 sys_clock_getres(struct lwp *l __unused, void *v, register_t *retval __unused)
291 {
292 	struct sys_clock_getres_args /* {
293 		syscallarg(clockid_t) clock_id;
294 		syscallarg(struct timespec *) tp;
295 	} */ *uap = v;
296 	clockid_t clock_id;
297 	struct timespec ts;
298 	int error = 0;
299 
300 	clock_id = SCARG(uap, clock_id);
301 	switch (clock_id) {
302 	case CLOCK_REALTIME:
303 	case CLOCK_MONOTONIC:
304 		ts.tv_sec = 0;
305 #ifdef __HAVE_TIMECOUNTER
306 		if (tc_getfrequency() > 1000000000)
307 			ts.tv_nsec = 1;
308 		else
309 			ts.tv_nsec = 1000000000 / tc_getfrequency();
310 #else /* !__HAVE_TIMECOUNTER */
311 		ts.tv_nsec = 1000000000 / hz;
312 #endif /* !__HAVE_TIMECOUNTER */
313 		break;
314 	default:
315 		return (EINVAL);
316 	}
317 
318 	if (SCARG(uap, tp))
319 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
320 
321 	return error;
322 }
323 
324 /* ARGSUSED */
325 int
326 sys_nanosleep(struct lwp *l __unused, void *v, register_t *retval __unused)
327 {
328 #ifdef __HAVE_TIMECOUNTER
329 	static int nanowait;
330 	struct sys_nanosleep_args/* {
331 		syscallarg(struct timespec *) rqtp;
332 		syscallarg(struct timespec *) rmtp;
333 	} */ *uap = v;
334 	struct timespec rmt, rqt;
335 	int error, timo;
336 
337 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
338 	if (error)
339 		return (error);
340 
341 	if (itimespecfix(&rqt))
342 		return (EINVAL);
343 
344 	timo = tstohz(&rqt);
345 	/*
346 	 * Avoid inadvertantly sleeping forever
347 	 */
348 	if (timo == 0)
349 		timo = 1;
350 
351 	getnanouptime(&rmt);
352 
353 	error = tsleep(&nanowait, PWAIT | PCATCH, "nanosleep", timo);
354 	if (error == ERESTART)
355 		error = EINTR;
356 	if (error == EWOULDBLOCK)
357 		error = 0;
358 
359 	if (SCARG(uap, rmtp)) {
360 		int error1;
361 		struct timespec rmtend;
362 
363 		getnanouptime(&rmtend);
364 
365 		timespecsub(&rmtend, &rmt, &rmt);
366 		timespecsub(&rqt, &rmt, &rmt);
367 		if (rmt.tv_sec < 0)
368 			timespecclear(&rmt);
369 
370 		error1 = copyout((caddr_t)&rmt, (caddr_t)SCARG(uap,rmtp),
371 			sizeof(rmt));
372 		if (error1)
373 			return (error1);
374 	}
375 
376 	return error;
377 #else /* !__HAVE_TIMECOUNTER */
378 	static int nanowait;
379 	struct sys_nanosleep_args/* {
380 		syscallarg(struct timespec *) rqtp;
381 		syscallarg(struct timespec *) rmtp;
382 	} */ *uap = v;
383 	struct timespec rqt;
384 	struct timespec rmt;
385 	struct timeval atv, utv;
386 	int error, s, timo;
387 
388 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
389 	if (error)
390 		return (error);
391 
392 	TIMESPEC_TO_TIMEVAL(&atv,&rqt);
393 	if (itimerfix(&atv))
394 		return (EINVAL);
395 
396 	s = splclock();
397 	timeradd(&atv,&time,&atv);
398 	timo = hzto(&atv);
399 	/*
400 	 * Avoid inadvertantly sleeping forever
401 	 */
402 	if (timo == 0)
403 		timo = 1;
404 	splx(s);
405 
406 	error = tsleep(&nanowait, PWAIT | PCATCH, "nanosleep", timo);
407 	if (error == ERESTART)
408 		error = EINTR;
409 	if (error == EWOULDBLOCK)
410 		error = 0;
411 
412 	if (SCARG(uap, rmtp)) {
413 		int error1;
414 
415 		s = splclock();
416 		utv = time;
417 		splx(s);
418 
419 		timersub(&atv, &utv, &utv);
420 		if (utv.tv_sec < 0)
421 			timerclear(&utv);
422 
423 		TIMEVAL_TO_TIMESPEC(&utv,&rmt);
424 		error1 = copyout((caddr_t)&rmt, (caddr_t)SCARG(uap,rmtp),
425 			sizeof(rmt));
426 		if (error1)
427 			return (error1);
428 	}
429 
430 	return error;
431 #endif /* !__HAVE_TIMECOUNTER */
432 }
433 
434 /* ARGSUSED */
435 int
436 sys_gettimeofday(struct lwp *l __unused, void *v, register_t *retval __unused)
437 {
438 	struct sys_gettimeofday_args /* {
439 		syscallarg(struct timeval *) tp;
440 		syscallarg(void *) tzp;		really "struct timezone *"
441 	} */ *uap = v;
442 	struct timeval atv;
443 	int error = 0;
444 	struct timezone tzfake;
445 
446 	if (SCARG(uap, tp)) {
447 		microtime(&atv);
448 		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
449 		if (error)
450 			return (error);
451 	}
452 	if (SCARG(uap, tzp)) {
453 		/*
454 		 * NetBSD has no kernel notion of time zone, so we just
455 		 * fake up a timezone struct and return it if demanded.
456 		 */
457 		tzfake.tz_minuteswest = 0;
458 		tzfake.tz_dsttime = 0;
459 		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
460 	}
461 	return (error);
462 }
463 
464 /* ARGSUSED */
465 int
466 sys_settimeofday(struct lwp *l, void *v, register_t *retval __unused)
467 {
468 	struct sys_settimeofday_args /* {
469 		syscallarg(const struct timeval *) tv;
470 		syscallarg(const void *) tzp;	really "const struct timezone *"
471 	} */ *uap = v;
472 	int error;
473 
474 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
475 	    KAUTH_REQ_SYSTEM_TIME_SYSTEM, NULL, NULL, NULL)) != 0)
476 		return (error);
477 
478 	return settimeofday1(SCARG(uap, tv), SCARG(uap, tzp), l->l_proc);
479 }
480 
481 int
482 settimeofday1(const struct timeval *utv, const struct timezone *utzp,
483     struct proc *p)
484 {
485 	struct timeval atv;
486 	struct timespec ts;
487 	int error;
488 
489 	/* Verify all parameters before changing time. */
490 	/*
491 	 * NetBSD has no kernel notion of time zone, and only an
492 	 * obsolete program would try to set it, so we log a warning.
493 	 */
494 	if (utzp)
495 		log(LOG_WARNING, "pid %d attempted to set the "
496 		    "(obsolete) kernel time zone\n", p->p_pid);
497 
498 	if (utv == NULL)
499 		return 0;
500 
501 	if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
502 		return error;
503 	TIMEVAL_TO_TIMESPEC(&atv, &ts);
504 	return settime(p, &ts);
505 }
506 
507 #ifndef __HAVE_TIMECOUNTER
508 int	tickdelta;			/* current clock skew, us. per tick */
509 long	timedelta;			/* unapplied time correction, us. */
510 long	bigadj = 1000000;		/* use 10x skew above bigadj us. */
511 #endif
512 
513 int	time_adjusted;			/* set if an adjustment is made */
514 
515 /* ARGSUSED */
516 int
517 sys_adjtime(struct lwp *l, void *v, register_t *retval __unused)
518 {
519 	struct sys_adjtime_args /* {
520 		syscallarg(const struct timeval *) delta;
521 		syscallarg(struct timeval *) olddelta;
522 	} */ *uap = v;
523 	int error;
524 
525 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
526 	    KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0)
527 		return (error);
528 
529 	return adjtime1(SCARG(uap, delta), SCARG(uap, olddelta), l->l_proc);
530 }
531 
532 int
533 adjtime1(const struct timeval *delta, struct timeval *olddelta,
534     struct proc *p __unused)
535 {
536 	struct timeval atv;
537 	int error = 0;
538 
539 #ifdef __HAVE_TIMECOUNTER
540 	extern int64_t time_adjtime;  /* in kern_ntptime.c */
541 #else /* !__HAVE_TIMECOUNTER */
542 	long ndelta, ntickdelta, odelta;
543 	int s;
544 #endif /* !__HAVE_TIMECOUNTER */
545 
546 #ifdef __HAVE_TIMECOUNTER
547 	if (olddelta) {
548 		atv.tv_sec = time_adjtime / 1000000;
549 		atv.tv_usec = time_adjtime % 1000000;
550 		if (atv.tv_usec < 0) {
551 			atv.tv_usec += 1000000;
552 			atv.tv_sec--;
553 		}
554 		error = copyout(&atv, olddelta, sizeof(struct timeval));
555 		if (error)
556 			return (error);
557 	}
558 
559 	if (delta) {
560 		error = copyin(delta, &atv, sizeof(struct timeval));
561 		if (error)
562 			return (error);
563 
564 		time_adjtime = (int64_t)atv.tv_sec * 1000000 +
565 			atv.tv_usec;
566 
567 		if (time_adjtime)
568 			/* We need to save the system time during shutdown */
569 			time_adjusted |= 1;
570 	}
571 #else /* !__HAVE_TIMECOUNTER */
572 	error = copyin(delta, &atv, sizeof(struct timeval));
573 	if (error)
574 		return (error);
575 
576 	/*
577 	 * Compute the total correction and the rate at which to apply it.
578 	 * Round the adjustment down to a whole multiple of the per-tick
579 	 * delta, so that after some number of incremental changes in
580 	 * hardclock(), tickdelta will become zero, lest the correction
581 	 * overshoot and start taking us away from the desired final time.
582 	 */
583 	ndelta = atv.tv_sec * 1000000 + atv.tv_usec;
584 	if (ndelta > bigadj || ndelta < -bigadj)
585 		ntickdelta = 10 * tickadj;
586 	else
587 		ntickdelta = tickadj;
588 	if (ndelta % ntickdelta)
589 		ndelta = ndelta / ntickdelta * ntickdelta;
590 
591 	/*
592 	 * To make hardclock()'s job easier, make the per-tick delta negative
593 	 * if we want time to run slower; then hardclock can simply compute
594 	 * tick + tickdelta, and subtract tickdelta from timedelta.
595 	 */
596 	if (ndelta < 0)
597 		ntickdelta = -ntickdelta;
598 	if (ndelta != 0)
599 		/* We need to save the system clock time during shutdown */
600 		time_adjusted |= 1;
601 	s = splclock();
602 	odelta = timedelta;
603 	timedelta = ndelta;
604 	tickdelta = ntickdelta;
605 	splx(s);
606 
607 	if (olddelta) {
608 		atv.tv_sec = odelta / 1000000;
609 		atv.tv_usec = odelta % 1000000;
610 		error = copyout(&atv, olddelta, sizeof(struct timeval));
611 	}
612 #endif /* __HAVE_TIMECOUNTER */
613 
614 	return error;
615 }
616 
617 /*
618  * Interval timer support. Both the BSD getitimer() family and the POSIX
619  * timer_*() family of routines are supported.
620  *
621  * All timers are kept in an array pointed to by p_timers, which is
622  * allocated on demand - many processes don't use timers at all. The
623  * first three elements in this array are reserved for the BSD timers:
624  * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, and element
625  * 2 is ITIMER_PROF. The rest may be allocated by the timer_create()
626  * syscall.
627  *
628  * Realtime timers are kept in the ptimer structure as an absolute
629  * time; virtual time timers are kept as a linked list of deltas.
630  * Virtual time timers are processed in the hardclock() routine of
631  * kern_clock.c.  The real time timer is processed by a callout
632  * routine, called from the softclock() routine.  Since a callout may
633  * be delayed in real time due to interrupt processing in the system,
634  * it is possible for the real time timeout routine (realtimeexpire,
635  * given below), to be delayed in real time past when it is supposed
636  * to occur.  It does not suffice, therefore, to reload the real timer
637  * .it_value from the real time timers .it_interval.  Rather, we
638  * compute the next time in absolute time the timer should go off.  */
639 
640 /* Allocate a POSIX realtime timer. */
641 int
642 sys_timer_create(struct lwp *l, void *v, register_t *retval __unused)
643 {
644 	struct sys_timer_create_args /* {
645 		syscallarg(clockid_t) clock_id;
646 		syscallarg(struct sigevent *) evp;
647 		syscallarg(timer_t *) timerid;
648 	} */ *uap = v;
649 
650 	return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id),
651 	    SCARG(uap, evp), copyin, l);
652 }
653 
654 int
655 timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp,
656     copyin_t fetch_event, struct lwp *l)
657 {
658 	int error;
659 	timer_t timerid;
660 	struct ptimer *pt;
661 	struct proc *p;
662 
663 	p = l->l_proc;
664 
665 	if (id < CLOCK_REALTIME ||
666 	    id > CLOCK_PROF)
667 		return (EINVAL);
668 
669 	if (p->p_timers == NULL)
670 		timers_alloc(p);
671 
672 	/* Find a free timer slot, skipping those reserved for setitimer(). */
673 	for (timerid = 3; timerid < TIMER_MAX; timerid++)
674 		if (p->p_timers->pts_timers[timerid] == NULL)
675 			break;
676 
677 	if (timerid == TIMER_MAX)
678 		return EAGAIN;
679 
680 	pt = pool_get(&ptimer_pool, PR_WAITOK);
681 	if (evp) {
682 		if (((error =
683 		    (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) ||
684 		    ((pt->pt_ev.sigev_notify < SIGEV_NONE) ||
685 			(pt->pt_ev.sigev_notify > SIGEV_SA))) {
686 			pool_put(&ptimer_pool, pt);
687 			return (error ? error : EINVAL);
688 		}
689 	} else {
690 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
691 		switch (id) {
692 		case CLOCK_REALTIME:
693 			pt->pt_ev.sigev_signo = SIGALRM;
694 			break;
695 		case CLOCK_VIRTUAL:
696 			pt->pt_ev.sigev_signo = SIGVTALRM;
697 			break;
698 		case CLOCK_PROF:
699 			pt->pt_ev.sigev_signo = SIGPROF;
700 			break;
701 		}
702 		pt->pt_ev.sigev_value.sival_int = timerid;
703 	}
704 	pt->pt_info.ksi_signo = pt->pt_ev.sigev_signo;
705 	pt->pt_info.ksi_errno = 0;
706 	pt->pt_info.ksi_code = 0;
707 	pt->pt_info.ksi_pid = p->p_pid;
708 	pt->pt_info.ksi_uid = kauth_cred_getuid(l->l_cred);
709 	pt->pt_info.ksi_sigval = pt->pt_ev.sigev_value;
710 
711 	pt->pt_type = id;
712 	pt->pt_proc = p;
713 	pt->pt_overruns = 0;
714 	pt->pt_poverruns = 0;
715 	pt->pt_entry = timerid;
716 	timerclear(&pt->pt_time.it_value);
717 	if (id == CLOCK_REALTIME)
718 		callout_init(&pt->pt_ch);
719 	else
720 		pt->pt_active = 0;
721 
722 	p->p_timers->pts_timers[timerid] = pt;
723 
724 	return copyout(&timerid, tid, sizeof(timerid));
725 }
726 
727 /* Delete a POSIX realtime timer */
728 int
729 sys_timer_delete(struct lwp *l, void *v, register_t *retval __unused)
730 {
731 	struct sys_timer_delete_args /*  {
732 		syscallarg(timer_t) timerid;
733 	} */ *uap = v;
734 	struct proc *p = l->l_proc;
735 	timer_t timerid;
736 	struct ptimer *pt, *ptn;
737 	int s;
738 
739 	timerid = SCARG(uap, timerid);
740 
741 	if ((p->p_timers == NULL) ||
742 	    (timerid < 2) || (timerid >= TIMER_MAX) ||
743 	    ((pt = p->p_timers->pts_timers[timerid]) == NULL))
744 		return (EINVAL);
745 
746 	if (pt->pt_type == CLOCK_REALTIME)
747 		callout_stop(&pt->pt_ch);
748 	else if (pt->pt_active) {
749 		s = splclock();
750 		ptn = LIST_NEXT(pt, pt_list);
751 		LIST_REMOVE(pt, pt_list);
752 		for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
753 			timeradd(&pt->pt_time.it_value, &ptn->pt_time.it_value,
754 			    &ptn->pt_time.it_value);
755 		splx(s);
756 	}
757 
758 	p->p_timers->pts_timers[timerid] = NULL;
759 	pool_put(&ptimer_pool, pt);
760 
761 	return (0);
762 }
763 
764 /*
765  * Set up the given timer. The value in pt->pt_time.it_value is taken
766  * to be an absolute time for CLOCK_REALTIME timers and a relative
767  * time for virtual timers.
768  * Must be called at splclock().
769  */
770 void
771 timer_settime(struct ptimer *pt)
772 {
773 	struct ptimer *ptn, *pptn;
774 	struct ptlist *ptl;
775 
776 	if (pt->pt_type == CLOCK_REALTIME) {
777 		callout_stop(&pt->pt_ch);
778 		if (timerisset(&pt->pt_time.it_value)) {
779 			/*
780 			 * Don't need to check hzto() return value, here.
781 			 * callout_reset() does it for us.
782 			 */
783 			callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value),
784 			    realtimerexpire, pt);
785 		}
786 	} else {
787 		if (pt->pt_active) {
788 			ptn = LIST_NEXT(pt, pt_list);
789 			LIST_REMOVE(pt, pt_list);
790 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
791 				timeradd(&pt->pt_time.it_value,
792 				    &ptn->pt_time.it_value,
793 				    &ptn->pt_time.it_value);
794 		}
795 		if (timerisset(&pt->pt_time.it_value)) {
796 			if (pt->pt_type == CLOCK_VIRTUAL)
797 				ptl = &pt->pt_proc->p_timers->pts_virtual;
798 			else
799 				ptl = &pt->pt_proc->p_timers->pts_prof;
800 
801 			for (ptn = LIST_FIRST(ptl), pptn = NULL;
802 			     ptn && timercmp(&pt->pt_time.it_value,
803 				 &ptn->pt_time.it_value, >);
804 			     pptn = ptn, ptn = LIST_NEXT(ptn, pt_list))
805 				timersub(&pt->pt_time.it_value,
806 				    &ptn->pt_time.it_value,
807 				    &pt->pt_time.it_value);
808 
809 			if (pptn)
810 				LIST_INSERT_AFTER(pptn, pt, pt_list);
811 			else
812 				LIST_INSERT_HEAD(ptl, pt, pt_list);
813 
814 			for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list))
815 				timersub(&ptn->pt_time.it_value,
816 				    &pt->pt_time.it_value,
817 				    &ptn->pt_time.it_value);
818 
819 			pt->pt_active = 1;
820 		} else
821 			pt->pt_active = 0;
822 	}
823 }
824 
825 void
826 timer_gettime(struct ptimer *pt, struct itimerval *aitv)
827 {
828 #ifdef __HAVE_TIMECOUNTER
829 	struct timeval now;
830 #endif
831 	struct ptimer *ptn;
832 
833 	*aitv = pt->pt_time;
834 	if (pt->pt_type == CLOCK_REALTIME) {
835 		/*
836 		 * Convert from absolute to relative time in .it_value
837 		 * part of real time timer.  If time for real time
838 		 * timer has passed return 0, else return difference
839 		 * between current time and time for the timer to go
840 		 * off.
841 		 */
842 		if (timerisset(&aitv->it_value)) {
843 #ifdef __HAVE_TIMECOUNTER
844 			getmicrotime(&now);
845 			if (timercmp(&aitv->it_value, &now, <))
846 				timerclear(&aitv->it_value);
847 			else
848 				timersub(&aitv->it_value, &now,
849 				    &aitv->it_value);
850 #else /* !__HAVE_TIMECOUNTER */
851 			if (timercmp(&aitv->it_value, &time, <))
852 				timerclear(&aitv->it_value);
853 			else
854 				timersub(&aitv->it_value, &time,
855 				    &aitv->it_value);
856 #endif /* !__HAVE_TIMECOUNTER */
857 		}
858 	} else if (pt->pt_active) {
859 		if (pt->pt_type == CLOCK_VIRTUAL)
860 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual);
861 		else
862 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof);
863 		for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list))
864 			timeradd(&aitv->it_value,
865 			    &ptn->pt_time.it_value, &aitv->it_value);
866 		KASSERT(ptn != NULL); /* pt should be findable on the list */
867 	} else
868 		timerclear(&aitv->it_value);
869 }
870 
871 
872 
873 /* Set and arm a POSIX realtime timer */
874 int
875 sys_timer_settime(struct lwp *l, void *v, register_t *retval __unused)
876 {
877 	struct sys_timer_settime_args /* {
878 		syscallarg(timer_t) timerid;
879 		syscallarg(int) flags;
880 		syscallarg(const struct itimerspec *) value;
881 		syscallarg(struct itimerspec *) ovalue;
882 	} */ *uap = v;
883 	int error;
884 	struct itimerspec value, ovalue, *ovp = NULL;
885 
886 	if ((error = copyin(SCARG(uap, value), &value,
887 	    sizeof(struct itimerspec))) != 0)
888 		return (error);
889 
890 	if (SCARG(uap, ovalue))
891 		ovp = &ovalue;
892 
893 	if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp,
894 	    SCARG(uap, flags), l->l_proc)) != 0)
895 		return error;
896 
897 	if (ovp)
898 		return copyout(&ovalue, SCARG(uap, ovalue),
899 		    sizeof(struct itimerspec));
900 	return 0;
901 }
902 
903 int
904 dotimer_settime(int timerid, struct itimerspec *value,
905     struct itimerspec *ovalue, int flags, struct proc *p)
906 {
907 #ifdef __HAVE_TIMECOUNTER
908 	struct timeval now;
909 #endif
910 	struct itimerval val, oval;
911 	struct ptimer *pt;
912 	int s;
913 
914 	if ((p->p_timers == NULL) ||
915 	    (timerid < 2) || (timerid >= TIMER_MAX) ||
916 	    ((pt = p->p_timers->pts_timers[timerid]) == NULL))
917 		return (EINVAL);
918 
919 	TIMESPEC_TO_TIMEVAL(&val.it_value, &value->it_value);
920 	TIMESPEC_TO_TIMEVAL(&val.it_interval, &value->it_interval);
921 	if (itimerfix(&val.it_value) || itimerfix(&val.it_interval))
922 		return (EINVAL);
923 
924 	oval = pt->pt_time;
925 	pt->pt_time = val;
926 
927 	s = splclock();
928 	/*
929 	 * If we've been passed a relative time for a realtime timer,
930 	 * convert it to absolute; if an absolute time for a virtual
931 	 * timer, convert it to relative and make sure we don't set it
932 	 * to zero, which would cancel the timer, or let it go
933 	 * negative, which would confuse the comparison tests.
934 	 */
935 	if (timerisset(&pt->pt_time.it_value)) {
936 		if (pt->pt_type == CLOCK_REALTIME) {
937 #ifdef __HAVE_TIMECOUNTER
938 			if ((flags & TIMER_ABSTIME) == 0) {
939 				getmicrotime(&now);
940 				timeradd(&pt->pt_time.it_value, &now,
941 				    &pt->pt_time.it_value);
942 			}
943 #else /* !__HAVE_TIMECOUNTER */
944 			if ((flags & TIMER_ABSTIME) == 0)
945 				timeradd(&pt->pt_time.it_value, &time,
946 				    &pt->pt_time.it_value);
947 #endif /* !__HAVE_TIMECOUNTER */
948 		} else {
949 			if ((flags & TIMER_ABSTIME) != 0) {
950 #ifdef __HAVE_TIMECOUNTER
951 				getmicrotime(&now);
952 				timersub(&pt->pt_time.it_value, &now,
953 				    &pt->pt_time.it_value);
954 #else /* !__HAVE_TIMECOUNTER */
955 				timersub(&pt->pt_time.it_value, &time,
956 				    &pt->pt_time.it_value);
957 #endif /* !__HAVE_TIMECOUNTER */
958 				if (!timerisset(&pt->pt_time.it_value) ||
959 				    pt->pt_time.it_value.tv_sec < 0) {
960 					pt->pt_time.it_value.tv_sec = 0;
961 					pt->pt_time.it_value.tv_usec = 1;
962 				}
963 			}
964 		}
965 	}
966 
967 	timer_settime(pt);
968 	splx(s);
969 
970 	if (ovalue) {
971 		TIMEVAL_TO_TIMESPEC(&oval.it_value, &ovalue->it_value);
972 		TIMEVAL_TO_TIMESPEC(&oval.it_interval, &ovalue->it_interval);
973 	}
974 
975 	return (0);
976 }
977 
978 /* Return the time remaining until a POSIX timer fires. */
979 int
980 sys_timer_gettime(struct lwp *l, void *v, register_t *retval __unused)
981 {
982 	struct sys_timer_gettime_args /* {
983 		syscallarg(timer_t) timerid;
984 		syscallarg(struct itimerspec *) value;
985 	} */ *uap = v;
986 	struct itimerspec its;
987 	int error;
988 
989 	if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc,
990 	    &its)) != 0)
991 		return error;
992 
993 	return copyout(&its, SCARG(uap, value), sizeof(its));
994 }
995 
996 int
997 dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its)
998 {
999 	int s;
1000 	struct ptimer *pt;
1001 	struct itimerval aitv;
1002 
1003 	if ((p->p_timers == NULL) ||
1004 	    (timerid < 2) || (timerid >= TIMER_MAX) ||
1005 	    ((pt = p->p_timers->pts_timers[timerid]) == NULL))
1006 		return (EINVAL);
1007 
1008 	s = splclock();
1009 	timer_gettime(pt, &aitv);
1010 	splx(s);
1011 
1012 	TIMEVAL_TO_TIMESPEC(&aitv.it_interval, &its->it_interval);
1013 	TIMEVAL_TO_TIMESPEC(&aitv.it_value, &its->it_value);
1014 
1015 	return 0;
1016 }
1017 
1018 /*
1019  * Return the count of the number of times a periodic timer expired
1020  * while a notification was already pending. The counter is reset when
1021  * a timer expires and a notification can be posted.
1022  */
1023 int
1024 sys_timer_getoverrun(struct lwp *l, void *v, register_t *retval)
1025 {
1026 	struct sys_timer_getoverrun_args /* {
1027 		syscallarg(timer_t) timerid;
1028 	} */ *uap = v;
1029 	struct proc *p = l->l_proc;
1030 	int timerid;
1031 	struct ptimer *pt;
1032 
1033 	timerid = SCARG(uap, timerid);
1034 
1035 	if ((p->p_timers == NULL) ||
1036 	    (timerid < 2) || (timerid >= TIMER_MAX) ||
1037 	    ((pt = p->p_timers->pts_timers[timerid]) == NULL))
1038 		return (EINVAL);
1039 
1040 	*retval = pt->pt_poverruns;
1041 
1042 	return (0);
1043 }
1044 
1045 /* Glue function that triggers an upcall; called from userret(). */
1046 static void
1047 timerupcall(struct lwp *l, void *arg)
1048 {
1049 	struct ptimers *pt = (struct ptimers *)arg;
1050 	unsigned int i, fired, done;
1051 
1052 	KDASSERT(l->l_proc->p_sa);
1053 	/* Bail out if we do not own the virtual processor */
1054 	if (l->l_savp->savp_lwp != l)
1055 		return ;
1056 
1057 	KERNEL_PROC_LOCK(l);
1058 
1059 	fired = pt->pts_fired;
1060 	done = 0;
1061 	while ((i = ffs(fired)) != 0) {
1062 		siginfo_t *si;
1063 		int mask = 1 << --i;
1064 		int f;
1065 
1066 		f = l->l_flag & L_SA;
1067 		l->l_flag &= ~L_SA;
1068 		si = siginfo_alloc(PR_WAITOK);
1069 		si->_info = pt->pts_timers[i]->pt_info.ksi_info;
1070 		if (sa_upcall(l, SA_UPCALL_SIGEV | SA_UPCALL_DEFER, NULL, l,
1071 		    sizeof(*si), si, siginfo_free) != 0) {
1072 			siginfo_free(si);
1073 			/* XXX What do we do here?? */
1074 		} else
1075 			done |= mask;
1076 		fired &= ~mask;
1077 		l->l_flag |= f;
1078 	}
1079 	pt->pts_fired &= ~done;
1080 	if (pt->pts_fired == 0)
1081 		l->l_proc->p_userret = NULL;
1082 
1083 	KERNEL_PROC_UNLOCK(l);
1084 }
1085 
1086 /*
1087  * Real interval timer expired:
1088  * send process whose timer expired an alarm signal.
1089  * If time is not set up to reload, then just return.
1090  * Else compute next time timer should go off which is > current time.
1091  * This is where delay in processing this timeout causes multiple
1092  * SIGALRM calls to be compressed into one.
1093  */
1094 void
1095 realtimerexpire(void *arg)
1096 {
1097 #ifdef __HAVE_TIMECOUNTER
1098 	struct timeval now;
1099 #endif
1100 	struct ptimer *pt;
1101 	int s;
1102 
1103 	pt = (struct ptimer *)arg;
1104 
1105 	itimerfire(pt);
1106 
1107 	if (!timerisset(&pt->pt_time.it_interval)) {
1108 		timerclear(&pt->pt_time.it_value);
1109 		return;
1110 	}
1111 #ifdef __HAVE_TIMECOUNTER
1112 	for (;;) {
1113 		s = splclock();	/* XXX need spl now? */
1114 		timeradd(&pt->pt_time.it_value,
1115 		    &pt->pt_time.it_interval, &pt->pt_time.it_value);
1116 		getmicrotime(&now);
1117 		if (timercmp(&pt->pt_time.it_value, &now, >)) {
1118 			/*
1119 			 * Don't need to check hzto() return value, here.
1120 			 * callout_reset() does it for us.
1121 			 */
1122 			callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value),
1123 			    realtimerexpire, pt);
1124 			splx(s);
1125 			return;
1126 		}
1127 		splx(s);
1128 		pt->pt_overruns++;
1129 	}
1130 #else /* !__HAVE_TIMECOUNTER */
1131 	for (;;) {
1132 		s = splclock();
1133 		timeradd(&pt->pt_time.it_value,
1134 		    &pt->pt_time.it_interval, &pt->pt_time.it_value);
1135 		if (timercmp(&pt->pt_time.it_value, &time, >)) {
1136 			/*
1137 			 * Don't need to check hzto() return value, here.
1138 			 * callout_reset() does it for us.
1139 			 */
1140 			callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value),
1141 			    realtimerexpire, pt);
1142 			splx(s);
1143 			return;
1144 		}
1145 		splx(s);
1146 		pt->pt_overruns++;
1147 	}
1148 #endif /* !__HAVE_TIMECOUNTER */
1149 }
1150 
1151 /* BSD routine to get the value of an interval timer. */
1152 /* ARGSUSED */
1153 int
1154 sys_getitimer(struct lwp *l, void *v, register_t *retval __unused)
1155 {
1156 	struct sys_getitimer_args /* {
1157 		syscallarg(int) which;
1158 		syscallarg(struct itimerval *) itv;
1159 	} */ *uap = v;
1160 	struct proc *p = l->l_proc;
1161 	struct itimerval aitv;
1162 	int error;
1163 
1164 	error = dogetitimer(p, SCARG(uap, which), &aitv);
1165 	if (error)
1166 		return error;
1167 	return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)));
1168 }
1169 
1170 int
1171 dogetitimer(struct proc *p, int which, struct itimerval *itvp)
1172 {
1173 	int s;
1174 
1175 	if ((u_int)which > ITIMER_PROF)
1176 		return (EINVAL);
1177 
1178 	if ((p->p_timers == NULL) || (p->p_timers->pts_timers[which] == NULL)){
1179 		timerclear(&itvp->it_value);
1180 		timerclear(&itvp->it_interval);
1181 	} else {
1182 		s = splclock();
1183 		timer_gettime(p->p_timers->pts_timers[which], itvp);
1184 		splx(s);
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 /* BSD routine to set/arm an interval timer. */
1191 /* ARGSUSED */
1192 int
1193 sys_setitimer(struct lwp *l, void *v, register_t *retval)
1194 {
1195 	struct sys_setitimer_args /* {
1196 		syscallarg(int) which;
1197 		syscallarg(const struct itimerval *) itv;
1198 		syscallarg(struct itimerval *) oitv;
1199 	} */ *uap = v;
1200 	struct proc *p = l->l_proc;
1201 	int which = SCARG(uap, which);
1202 	struct sys_getitimer_args getargs;
1203 	const struct itimerval *itvp;
1204 	struct itimerval aitv;
1205 	int error;
1206 
1207 	if ((u_int)which > ITIMER_PROF)
1208 		return (EINVAL);
1209 	itvp = SCARG(uap, itv);
1210 	if (itvp &&
1211 	    (error = copyin(itvp, &aitv, sizeof(struct itimerval)) != 0))
1212 		return (error);
1213 	if (SCARG(uap, oitv) != NULL) {
1214 		SCARG(&getargs, which) = which;
1215 		SCARG(&getargs, itv) = SCARG(uap, oitv);
1216 		if ((error = sys_getitimer(l, &getargs, retval)) != 0)
1217 			return (error);
1218 	}
1219 	if (itvp == 0)
1220 		return (0);
1221 
1222 	return dosetitimer(p, which, &aitv);
1223 }
1224 
1225 int
1226 dosetitimer(struct proc *p, int which, struct itimerval *itvp)
1227 {
1228 #ifdef __HAVE_TIMECOUNTER
1229 	struct timeval now;
1230 #endif
1231 	struct ptimer *pt;
1232 	int s;
1233 
1234 	if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval))
1235 		return (EINVAL);
1236 
1237 	/*
1238 	 * Don't bother allocating data structures if the process just
1239 	 * wants to clear the timer.
1240 	 */
1241 	if (!timerisset(&itvp->it_value) &&
1242 	    ((p->p_timers == NULL) ||(p->p_timers->pts_timers[which] == NULL)))
1243 		return (0);
1244 
1245 	if (p->p_timers == NULL)
1246 		timers_alloc(p);
1247 	if (p->p_timers->pts_timers[which] == NULL) {
1248 		pt = pool_get(&ptimer_pool, PR_WAITOK);
1249 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
1250 		pt->pt_ev.sigev_value.sival_int = which;
1251 		pt->pt_overruns = 0;
1252 		pt->pt_proc = p;
1253 		pt->pt_type = which;
1254 		pt->pt_entry = which;
1255 		switch (which) {
1256 		case ITIMER_REAL:
1257 			callout_init(&pt->pt_ch);
1258 			pt->pt_ev.sigev_signo = SIGALRM;
1259 			break;
1260 		case ITIMER_VIRTUAL:
1261 			pt->pt_active = 0;
1262 			pt->pt_ev.sigev_signo = SIGVTALRM;
1263 			break;
1264 		case ITIMER_PROF:
1265 			pt->pt_active = 0;
1266 			pt->pt_ev.sigev_signo = SIGPROF;
1267 			break;
1268 		}
1269 	} else
1270 		pt = p->p_timers->pts_timers[which];
1271 
1272 	pt->pt_time = *itvp;
1273 	p->p_timers->pts_timers[which] = pt;
1274 
1275 	s = splclock();
1276 	if ((which == ITIMER_REAL) && timerisset(&pt->pt_time.it_value)) {
1277 		/* Convert to absolute time */
1278 #ifdef __HAVE_TIMECOUNTER
1279 		/* XXX need to wrap in splclock for timecounters case? */
1280 		getmicrotime(&now);
1281 		timeradd(&pt->pt_time.it_value, &now, &pt->pt_time.it_value);
1282 #else /* !__HAVE_TIMECOUNTER */
1283 		timeradd(&pt->pt_time.it_value, &time, &pt->pt_time.it_value);
1284 #endif /* !__HAVE_TIMECOUNTER */
1285 	}
1286 	timer_settime(pt);
1287 	splx(s);
1288 
1289 	return (0);
1290 }
1291 
1292 /* Utility routines to manage the array of pointers to timers. */
1293 void
1294 timers_alloc(struct proc *p)
1295 {
1296 	int i;
1297 	struct ptimers *pts;
1298 
1299 	pts = pool_get(&ptimers_pool, PR_WAITOK);
1300 	LIST_INIT(&pts->pts_virtual);
1301 	LIST_INIT(&pts->pts_prof);
1302 	for (i = 0; i < TIMER_MAX; i++)
1303 		pts->pts_timers[i] = NULL;
1304 	pts->pts_fired = 0;
1305 	p->p_timers = pts;
1306 }
1307 
1308 /*
1309  * Clean up the per-process timers. If "which" is set to TIMERS_ALL,
1310  * then clean up all timers and free all the data structures. If
1311  * "which" is set to TIMERS_POSIX, only clean up the timers allocated
1312  * by timer_create(), not the BSD setitimer() timers, and only free the
1313  * structure if none of those remain.
1314  */
1315 void
1316 timers_free(struct proc *p, int which)
1317 {
1318 	int i, s;
1319 	struct ptimers *pts;
1320 	struct ptimer *pt, *ptn;
1321 	struct timeval tv;
1322 
1323 	if (p->p_timers) {
1324 		pts = p->p_timers;
1325 		if (which == TIMERS_ALL)
1326 			i = 0;
1327 		else {
1328 			s = splclock();
1329 			timerclear(&tv);
1330 			for (ptn = LIST_FIRST(&p->p_timers->pts_virtual);
1331 			     ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL];
1332 			     ptn = LIST_NEXT(ptn, pt_list))
1333 				timeradd(&tv, &ptn->pt_time.it_value, &tv);
1334 			LIST_FIRST(&p->p_timers->pts_virtual) = NULL;
1335 			if (ptn) {
1336 				timeradd(&tv, &ptn->pt_time.it_value,
1337 				    &ptn->pt_time.it_value);
1338 				LIST_INSERT_HEAD(&p->p_timers->pts_virtual,
1339 				    ptn, pt_list);
1340 			}
1341 
1342 			timerclear(&tv);
1343 			for (ptn = LIST_FIRST(&p->p_timers->pts_prof);
1344 			     ptn && ptn != pts->pts_timers[ITIMER_PROF];
1345 			     ptn = LIST_NEXT(ptn, pt_list))
1346 				timeradd(&tv, &ptn->pt_time.it_value, &tv);
1347 			LIST_FIRST(&p->p_timers->pts_prof) = NULL;
1348 			if (ptn) {
1349 				timeradd(&tv, &ptn->pt_time.it_value,
1350 				    &ptn->pt_time.it_value);
1351 				LIST_INSERT_HEAD(&p->p_timers->pts_prof, ptn,
1352 				    pt_list);
1353 			}
1354 			splx(s);
1355 			i = 3;
1356 		}
1357 		for ( ; i < TIMER_MAX; i++)
1358 			if ((pt = pts->pts_timers[i]) != NULL) {
1359 				if (pt->pt_type == CLOCK_REALTIME)
1360 					callout_stop(&pt->pt_ch);
1361 				pts->pts_timers[i] = NULL;
1362 				pool_put(&ptimer_pool, pt);
1363 			}
1364 		if ((pts->pts_timers[0] == NULL) &&
1365 		    (pts->pts_timers[1] == NULL) &&
1366 		    (pts->pts_timers[2] == NULL)) {
1367 			p->p_timers = NULL;
1368 			pool_put(&ptimers_pool, pts);
1369 		}
1370 	}
1371 }
1372 
1373 /*
1374  * Check that a proposed value to load into the .it_value or
1375  * .it_interval part of an interval timer is acceptable, and
1376  * fix it to have at least minimal value (i.e. if it is less
1377  * than the resolution of the clock, round it up.)
1378  */
1379 int
1380 itimerfix(struct timeval *tv)
1381 {
1382 
1383 	if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
1384 		return (EINVAL);
1385 	if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
1386 		tv->tv_usec = tick;
1387 	return (0);
1388 }
1389 
1390 #ifdef __HAVE_TIMECOUNTER
1391 int
1392 itimespecfix(struct timespec *ts)
1393 {
1394 
1395 	if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1396 		return (EINVAL);
1397 	if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1398 		ts->tv_nsec = tick * 1000;
1399 	return (0);
1400 }
1401 #endif /* __HAVE_TIMECOUNTER */
1402 
1403 /*
1404  * Decrement an interval timer by a specified number
1405  * of microseconds, which must be less than a second,
1406  * i.e. < 1000000.  If the timer expires, then reload
1407  * it.  In this case, carry over (usec - old value) to
1408  * reduce the value reloaded into the timer so that
1409  * the timer does not drift.  This routine assumes
1410  * that it is called in a context where the timers
1411  * on which it is operating cannot change in value.
1412  */
1413 int
1414 itimerdecr(struct ptimer *pt, int usec)
1415 {
1416 	struct itimerval *itp;
1417 
1418 	itp = &pt->pt_time;
1419 	if (itp->it_value.tv_usec < usec) {
1420 		if (itp->it_value.tv_sec == 0) {
1421 			/* expired, and already in next interval */
1422 			usec -= itp->it_value.tv_usec;
1423 			goto expire;
1424 		}
1425 		itp->it_value.tv_usec += 1000000;
1426 		itp->it_value.tv_sec--;
1427 	}
1428 	itp->it_value.tv_usec -= usec;
1429 	usec = 0;
1430 	if (timerisset(&itp->it_value))
1431 		return (1);
1432 	/* expired, exactly at end of interval */
1433 expire:
1434 	if (timerisset(&itp->it_interval)) {
1435 		itp->it_value = itp->it_interval;
1436 		itp->it_value.tv_usec -= usec;
1437 		if (itp->it_value.tv_usec < 0) {
1438 			itp->it_value.tv_usec += 1000000;
1439 			itp->it_value.tv_sec--;
1440 		}
1441 		timer_settime(pt);
1442 	} else
1443 		itp->it_value.tv_usec = 0;		/* sec is already 0 */
1444 	return (0);
1445 }
1446 
1447 void
1448 itimerfire(struct ptimer *pt)
1449 {
1450 	struct proc *p = pt->pt_proc;
1451 	struct sadata_vp *vp;
1452 	int s;
1453 	unsigned int i;
1454 
1455 	if (pt->pt_ev.sigev_notify == SIGEV_SIGNAL) {
1456 		/*
1457 		 * No RT signal infrastructure exists at this time;
1458 		 * just post the signal number and throw away the
1459 		 * value.
1460 		 */
1461 		if (sigismember(&p->p_sigctx.ps_siglist, pt->pt_ev.sigev_signo))
1462 			pt->pt_overruns++;
1463 		else {
1464 			ksiginfo_t ksi;
1465 			(void)memset(&ksi, 0, sizeof(ksi));
1466 			ksi.ksi_signo = pt->pt_ev.sigev_signo;
1467 			ksi.ksi_code = SI_TIMER;
1468 			ksi.ksi_sigval = pt->pt_ev.sigev_value;
1469 			pt->pt_poverruns = pt->pt_overruns;
1470 			pt->pt_overruns = 0;
1471 			kpsignal(p, &ksi, NULL);
1472 		}
1473 	} else if (pt->pt_ev.sigev_notify == SIGEV_SA && (p->p_flag & P_SA)) {
1474 		/* Cause the process to generate an upcall when it returns. */
1475 		signotify(p);
1476 		if (p->p_userret == NULL) {
1477 			/*
1478 			 * XXX stop signals can be processed inside tsleep,
1479 			 * which can be inside sa_yield's inner loop, which
1480 			 * makes testing for sa_idle alone insuffucent to
1481 			 * determine if we really should call setrunnable.
1482 			 */
1483 			pt->pt_poverruns = pt->pt_overruns;
1484 			pt->pt_overruns = 0;
1485 			i = 1 << pt->pt_entry;
1486 			p->p_timers->pts_fired = i;
1487 			p->p_userret = timerupcall;
1488 			p->p_userret_arg = p->p_timers;
1489 
1490 			SCHED_LOCK(s);
1491 			SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
1492 				if (vp->savp_lwp->l_flag & L_SA_IDLE) {
1493 					vp->savp_lwp->l_flag &= ~L_SA_IDLE;
1494 					sched_wakeup(vp->savp_lwp);
1495 					break;
1496 				}
1497 			}
1498 			SCHED_UNLOCK(s);
1499 		} else if (p->p_userret == timerupcall) {
1500 			i = 1 << pt->pt_entry;
1501 			if ((p->p_timers->pts_fired & i) == 0) {
1502 				pt->pt_poverruns = pt->pt_overruns;
1503 				pt->pt_overruns = 0;
1504 				p->p_timers->pts_fired |= i;
1505 			} else
1506 				pt->pt_overruns++;
1507 		} else {
1508 			pt->pt_overruns++;
1509 			if ((p->p_flag & P_WEXIT) == 0)
1510 				printf("itimerfire(%d): overrun %d on timer %x (userret is %p)\n",
1511 				    p->p_pid, pt->pt_overruns,
1512 				    pt->pt_ev.sigev_value.sival_int,
1513 				    p->p_userret);
1514 		}
1515 	}
1516 
1517 }
1518 
1519 /*
1520  * ratecheck(): simple time-based rate-limit checking.  see ratecheck(9)
1521  * for usage and rationale.
1522  */
1523 int
1524 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1525 {
1526 	struct timeval tv, delta;
1527 	int rv = 0;
1528 #ifndef __HAVE_TIMECOUNTER
1529 	int s;
1530 #endif
1531 
1532 #ifdef __HAVE_TIMECOUNTER
1533 	getmicrouptime(&tv);
1534 #else /* !__HAVE_TIMECOUNTER */
1535 	s = splclock();
1536 	tv = mono_time;
1537 	splx(s);
1538 #endif /* !__HAVE_TIMECOUNTER */
1539 	timersub(&tv, lasttime, &delta);
1540 
1541 	/*
1542 	 * check for 0,0 is so that the message will be seen at least once,
1543 	 * even if interval is huge.
1544 	 */
1545 	if (timercmp(&delta, mininterval, >=) ||
1546 	    (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1547 		*lasttime = tv;
1548 		rv = 1;
1549 	}
1550 
1551 	return (rv);
1552 }
1553 
1554 /*
1555  * ppsratecheck(): packets (or events) per second limitation.
1556  */
1557 int
1558 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
1559 {
1560 	struct timeval tv, delta;
1561 	int rv;
1562 #ifndef __HAVE_TIMECOUNTER
1563 	int s;
1564 #endif
1565 
1566 #ifdef __HAVE_TIMECOUNTER
1567 	getmicrouptime(&tv);
1568 #else /* !__HAVE_TIMECOUNTER */
1569 	s = splclock();
1570 	tv = mono_time;
1571 	splx(s);
1572 #endif /* !__HAVE_TIMECOUNTER */
1573 	timersub(&tv, lasttime, &delta);
1574 
1575 	/*
1576 	 * check for 0,0 is so that the message will be seen at least once.
1577 	 * if more than one second have passed since the last update of
1578 	 * lasttime, reset the counter.
1579 	 *
1580 	 * we do increment *curpps even in *curpps < maxpps case, as some may
1581 	 * try to use *curpps for stat purposes as well.
1582 	 */
1583 	if ((lasttime->tv_sec == 0 && lasttime->tv_usec == 0) ||
1584 	    delta.tv_sec >= 1) {
1585 		*lasttime = tv;
1586 		*curpps = 0;
1587 	}
1588 	if (maxpps < 0)
1589 		rv = 1;
1590 	else if (*curpps < maxpps)
1591 		rv = 1;
1592 	else
1593 		rv = 0;
1594 
1595 #if 1 /*DIAGNOSTIC?*/
1596 	/* be careful about wrap-around */
1597 	if (*curpps + 1 > *curpps)
1598 		*curpps = *curpps + 1;
1599 #else
1600 	/*
1601 	 * assume that there's not too many calls to this function.
1602 	 * not sure if the assumption holds, as it depends on *caller's*
1603 	 * behavior, not the behavior of this function.
1604 	 * IMHO it is wrong to make assumption on the caller's behavior,
1605 	 * so the above #if is #if 1, not #ifdef DIAGNOSTIC.
1606 	 */
1607 	*curpps = *curpps + 1;
1608 #endif
1609 
1610 	return (rv);
1611 }
1612