sys/kern/kern_time.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 4. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)kern_time.c 8.1 (Berkeley) 6/10/93
  30  */
  31
  32 #include <sys/cdefs.h>
  33 __FBSDID("$FreeBSD$");
  34
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/limits.h>
  38 #include <sys/clock.h>
  39 #include <sys/lock.h>
  40 #include <sys/mutex.h>
  41 #include <sys/sysproto.h>
  42 #include <sys/eventhandler.h>
  43 #include <sys/resourcevar.h>
  44 #include <sys/signalvar.h>
  45 #include <sys/kernel.h>
  46 #include <sys/syscallsubr.h>
  47 #include <sys/sysctl.h>
  48 #include <sys/sysent.h>
  49 #include <sys/priv.h>
  50 #include <sys/proc.h>
  51 #include <sys/posix4.h>
  52 #include <sys/time.h>
  53 #include <sys/timers.h>
  54 #include <sys/timetc.h>
  55 #include <sys/vnode.h>
  56
  57 #include <vm/vm.h>
  58 #include <vm/vm_extern.h>
  59
  60 #define MAX_CLOCKS      (CLOCK_MONOTONIC+1)
  61
  62 static struct kclock    posix_clocks[MAX_CLOCKS];
  63 static uma_zone_t       itimer_zone = NULL;
  64
  65 /*
  66  * Time of day and interval timer support.
  67  *
  68  * These routines provide the kernel entry points to get and set
  69  * the time-of-day and per-process interval timers.  Subroutines
  70  * here provide support for adding and subtracting timeval structures
  71  * and decrementing interval timers, optionally reloading the interval
  72  * timers when they expire.
  73  */
  74
  75 static int      settime(struct thread *, struct timeval *);
  76 static void     timevalfix(struct timeval *);
  77 static void     no_lease_updatetime(int);
  78
  79 static void     itimer_start(void);
  80 static int      itimer_init(void *, int, int);
  81 static void     itimer_fini(void *, int);
  82 static void     itimer_enter(struct itimer *);
  83 static void     itimer_leave(struct itimer *);
  84 static struct itimer *itimer_find(struct proc *, int);
  85 static void     itimers_alloc(struct proc *);
  86 static void     itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
  87 static void     itimers_event_hook_exit(void *arg, struct proc *p);
  88 static int      realtimer_create(struct itimer *);
  89 static int      realtimer_gettime(struct itimer *, struct itimerspec *);
  90 static int      realtimer_settime(struct itimer *, int,
  91                         struct itimerspec *, struct itimerspec *);
  92 static int      realtimer_delete(struct itimer *);
  93 static void     realtimer_clocktime(clockid_t, struct timespec *);
  94 static void     realtimer_expire(void *);
  95 static int      kern_timer_create(struct thread *, clockid_t,
  96                         struct sigevent *, int *, int);
  97 static int      kern_timer_delete(struct thread *, int);
  98
  99 int             register_posix_clock(int, struct kclock *);
 100 void            itimer_fire(struct itimer *it);
 101 int             itimespecfix(struct timespec *ts);
 102
 103 #define CLOCK_CALL(clock, call, arglist)                \
 104         ((*posix_clocks[clock].call) arglist)
 105
 106 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
 107
 108
 109 static void
 110 no_lease_updatetime(deltat)
 111         int deltat;
 112 {
 113 }
 114
 115 void (*lease_updatetime)(int)  = no_lease_updatetime;
 116
 117 static int
 118 settime(struct thread *td, struct timeval *tv)
 119 {
 120         struct timeval delta, tv1, tv2;
 121         static struct timeval maxtime, laststep;
 122         struct timespec ts;
 123         int s;
 124
 125         s = splclock();
 126         microtime(&tv1);
 127         delta = *tv;
 128         timevalsub(&delta, &tv1);
 129
 130         /*
 131          * If the system is secure, we do not allow the time to be
 132          * set to a value earlier than 1 second less than the highest
 133          * time we have yet seen. The worst a miscreant can do in
 134          * this circumstance is "freeze" time. He couldn't go
 135          * back to the past.
 136          *
 137          * We similarly do not allow the clock to be stepped more
 138          * than one second, nor more than once per second. This allows
 139          * a miscreant to make the clock march double-time, but no worse.
 140          */
 141         if (securelevel_gt(td->td_ucred, 1) != 0) {
 142                 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
 143                         /*
 144                          * Update maxtime to latest time we've seen.
 145                          */
 146                         if (tv1.tv_sec > maxtime.tv_sec)
 147                                 maxtime = tv1;
 148                         tv2 = *tv;
 149                         timevalsub(&tv2, &maxtime);
 150                         if (tv2.tv_sec < -1) {
 151                                 tv->tv_sec = maxtime.tv_sec - 1;
 152                                 printf("Time adjustment clamped to -1 second\n");
 153                         }
 154                 } else {
 155                         if (tv1.tv_sec == laststep.tv_sec) {
 156                                 splx(s);
 157                                 return (EPERM);
 158                         }
 159                         if (delta.tv_sec > 1) {
 160                                 tv->tv_sec = tv1.tv_sec + 1;
 161                                 printf("Time adjustment clamped to +1 second\n");
 162                         }
 163                         laststep = *tv;
 164                 }
 165         }
 166
 167         ts.tv_sec = tv->tv_sec;
 168         ts.tv_nsec = tv->tv_usec * 1000;
 169         mtx_lock(&Giant);
 170         tc_setclock(&ts);
 171         (void) splsoftclock();
 172         lease_updatetime(delta.tv_sec);
 173         splx(s);
 174         resettodr();
 175         mtx_unlock(&Giant);
 176         return (0);
 177 }
 178
 179 #ifndef _SYS_SYSPROTO_H_
 180 struct clock_gettime_args {
 181         clockid_t clock_id;
 182         struct  timespec *tp;
 183 };
 184 #endif
 185 /* ARGSUSED */
 186 int
 187 clock_gettime(struct thread *td, struct clock_gettime_args *uap)
 188 {
 189         struct timespec ats;
 190         int error;
 191
 192         error = kern_clock_gettime(td, uap->clock_id, &ats);
 193         if (error == 0)
 194                 error = copyout(&ats, uap->tp, sizeof(ats));
 195
 196         return (error);
 197 }
 198
 199 int
 200 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
 201 {
 202         struct timeval sys, user;
 203         struct proc *p;
 204         uint64_t runtime, curtime, switchtime;
 205
 206         p = td->td_proc;
 207         switch (clock_id) {
 208         case CLOCK_REALTIME:            /* Default to precise. */
 209         case CLOCK_REALTIME_PRECISE:
 210                 nanotime(ats);
 211                 break;
 212         case CLOCK_REALTIME_FAST:
 213                 getnanotime(ats);
 214                 break;
 215         case CLOCK_VIRTUAL:
 216                 PROC_LOCK(p);
 217                 PROC_SLOCK(p);
 218                 calcru(p, &user, &sys);
 219                 PROC_SUNLOCK(p);
 220                 PROC_UNLOCK(p);
 221                 TIMEVAL_TO_TIMESPEC(&user, ats);
 222                 break;
 223         case CLOCK_PROF:
 224                 PROC_LOCK(p);
 225                 PROC_SLOCK(p);
 226                 calcru(p, &user, &sys);
 227                 PROC_SUNLOCK(p);
 228                 PROC_UNLOCK(p);
 229                 timevaladd(&user, &sys);
 230                 TIMEVAL_TO_TIMESPEC(&user, ats);
 231                 break;
 232         case CLOCK_MONOTONIC:           /* Default to precise. */
 233         case CLOCK_MONOTONIC_PRECISE:
 234         case CLOCK_UPTIME:
 235         case CLOCK_UPTIME_PRECISE:
 236                 nanouptime(ats);
 237                 break;
 238         case CLOCK_UPTIME_FAST:
 239         case CLOCK_MONOTONIC_FAST:
 240                 getnanouptime(ats);
 241                 break;
 242         case CLOCK_SECOND:
 243                 ats->tv_sec = time_second;
 244                 ats->tv_nsec = 0;
 245                 break;
 246         case CLOCK_THREAD_CPUTIME_ID:
 247                 critical_enter();
 248                 switchtime = PCPU_GET(switchtime);
 249                 curtime = cpu_ticks();
 250                 runtime = td->td_runtime;
 251                 critical_exit();
 252                 runtime = cputick2usec(runtime + curtime - switchtime);
 253                 ats->tv_sec = runtime / 1000000;
 254                 ats->tv_nsec = runtime % 1000000 * 1000;
 255                 break;
 256         default:
 257                 return (EINVAL);
 258         }
 259         return (0);
 260 }
 261
 262 #ifndef _SYS_SYSPROTO_H_
 263 struct clock_settime_args {
 264         clockid_t clock_id;
 265         const struct    timespec *tp;
 266 };
 267 #endif
 268 /* ARGSUSED */
 269 int
 270 clock_settime(struct thread *td, struct clock_settime_args *uap)
 271 {
 272         struct timespec ats;
 273         int error;
 274
 275         if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
 276                 return (error);
 277         return (kern_clock_settime(td, uap->clock_id, &ats));
 278 }
 279
 280 int
 281 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
 282 {
 283         struct timeval atv;
 284         int error;
 285
 286         if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
 287                 return (error);
 288         if (clock_id != CLOCK_REALTIME)
 289                 return (EINVAL);
 290         if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
 291                 return (EINVAL);
 292         /* XXX Don't convert nsec->usec and back */
 293         TIMESPEC_TO_TIMEVAL(&atv, ats);
 294         error = settime(td, &atv);
 295         return (error);
 296 }
 297
 298 #ifndef _SYS_SYSPROTO_H_
 299 struct clock_getres_args {
 300         clockid_t clock_id;
 301         struct  timespec *tp;
 302 };
 303 #endif
 304 int
 305 clock_getres(struct thread *td, struct clock_getres_args *uap)
 306 {
 307         struct timespec ts;
 308         int error;
 309
 310         if (uap->tp == NULL)
 311                 return (0);
 312
 313         error = kern_clock_getres(td, uap->clock_id, &ts);
 314         if (error == 0)
 315                 error = copyout(&ts, uap->tp, sizeof(ts));
 316         return (error);
 317 }
 318
 319 int
 320 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
 321 {
 322
 323         ts->tv_sec = 0;
 324         switch (clock_id) {
 325         case CLOCK_REALTIME:
 326         case CLOCK_REALTIME_FAST:
 327         case CLOCK_REALTIME_PRECISE:
 328         case CLOCK_MONOTONIC:
 329         case CLOCK_MONOTONIC_FAST:
 330         case CLOCK_MONOTONIC_PRECISE:
 331         case CLOCK_UPTIME:
 332         case CLOCK_UPTIME_FAST:
 333         case CLOCK_UPTIME_PRECISE:
 334                 /*
 335                  * Round up the result of the division cheaply by adding 1.
 336                  * Rounding up is especially important if rounding down
 337                  * would give 0.  Perfect rounding is unimportant.
 338                  */
 339                 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
 340                 break;
 341         case CLOCK_VIRTUAL:
 342         case CLOCK_PROF:
 343                 /* Accurately round up here because we can do so cheaply. */
 344                 ts->tv_nsec = (1000000000 + hz - 1) / hz;
 345                 break;
 346         case CLOCK_SECOND:
 347                 ts->tv_sec = 1;
 348                 ts->tv_nsec = 0;
 349                 break;
 350         case CLOCK_THREAD_CPUTIME_ID:
 351                 /* sync with cputick2usec */
 352                 ts->tv_nsec = 1000000 / cpu_tickrate();
 353                 if (ts->tv_nsec == 0)
 354                         ts->tv_nsec = 1000;
 355                 break;
 356         default:
 357                 return (EINVAL);
 358         }
 359         return (0);
 360 }
 361
 362 static int nanowait;
 363
 364 int
 365 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
 366 {
 367         struct timespec ts, ts2, ts3;
 368         struct timeval tv;
 369         int error;
 370
 371         if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
 372                 return (EINVAL);
 373         if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
 374                 return (0);
 375         getnanouptime(&ts);
 376         timespecadd(&ts, rqt);
 377         TIMESPEC_TO_TIMEVAL(&tv, rqt);
 378         for (;;) {
 379                 error = tsleep(&nanowait, PWAIT | PCATCH, "nanslp",
 380                     tvtohz(&tv));
 381                 getnanouptime(&ts2);
 382                 if (error != EWOULDBLOCK) {
 383                         if (error == ERESTART)
 384                                 error = EINTR;
 385                         if (rmt != NULL) {
 386                                 timespecsub(&ts, &ts2);
 387                                 if (ts.tv_sec < 0)
 388                                         timespecclear(&ts);
 389                                 *rmt = ts;
 390                         }
 391                         return (error);
 392                 }
 393                 if (timespeccmp(&ts2, &ts, >=))
 394                         return (0);
 395                 ts3 = ts;
 396                 timespecsub(&ts3, &ts2);
 397                 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
 398         }
 399 }
 400
 401 #ifndef _SYS_SYSPROTO_H_
 402 struct nanosleep_args {
 403         struct  timespec *rqtp;
 404         struct  timespec *rmtp;
 405 };
 406 #endif
 407 /* ARGSUSED */
 408 int
 409 nanosleep(struct thread *td, struct nanosleep_args *uap)
 410 {
 411         struct timespec rmt, rqt;
 412         int error;
 413
 414         error = copyin(uap->rqtp, &rqt, sizeof(rqt));
 415         if (error)
 416                 return (error);
 417
 418         if (uap->rmtp &&
 419             !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE))
 420                         return (EFAULT);
 421         error = kern_nanosleep(td, &rqt, &rmt);
 422         if (error && uap->rmtp) {
 423                 int error2;
 424
 425                 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
 426                 if (error2)
 427                         error = error2;
 428         }
 429         return (error);
 430 }
 431
 432 #ifndef _SYS_SYSPROTO_H_
 433 struct gettimeofday_args {
 434         struct  timeval *tp;
 435         struct  timezone *tzp;
 436 };
 437 #endif
 438 /* ARGSUSED */
 439 int
 440 gettimeofday(struct thread *td, struct gettimeofday_args *uap)
 441 {
 442         struct timeval atv;
 443         struct timezone rtz;
 444         int error = 0;
 445
 446         if (uap->tp) {
 447                 microtime(&atv);
 448                 error = copyout(&atv, uap->tp, sizeof (atv));
 449         }
 450         if (error == 0 && uap->tzp != NULL) {
 451                 rtz.tz_minuteswest = tz_minuteswest;
 452                 rtz.tz_dsttime = tz_dsttime;
 453                 error = copyout(&rtz, uap->tzp, sizeof (rtz));
 454         }
 455         return (error);
 456 }
 457
 458 #ifndef _SYS_SYSPROTO_H_
 459 struct settimeofday_args {
 460         struct  timeval *tv;
 461         struct  timezone *tzp;
 462 };
 463 #endif
 464 /* ARGSUSED */
 465 int
 466 settimeofday(struct thread *td, struct settimeofday_args *uap)
 467 {
 468         struct timeval atv, *tvp;
 469         struct timezone atz, *tzp;
 470         int error;
 471
 472         if (uap->tv) {
 473                 error = copyin(uap->tv, &atv, sizeof(atv));
 474                 if (error)
 475                         return (error);
 476                 tvp = &atv;
 477         } else
 478                 tvp = NULL;
 479         if (uap->tzp) {
 480                 error = copyin(uap->tzp, &atz, sizeof(atz));
 481                 if (error)
 482                         return (error);
 483                 tzp = &atz;
 484         } else
 485                 tzp = NULL;
 486         return (kern_settimeofday(td, tvp, tzp));
 487 }
 488
 489 int
 490 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
 491 {
 492         int error;
 493
 494         error = priv_check(td, PRIV_SETTIMEOFDAY);
 495         if (error)
 496                 return (error);
 497         /* Verify all parameters before changing time. */
 498         if (tv) {
 499                 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000)
 500                         return (EINVAL);
 501                 error = settime(td, tv);
 502         }
 503         if (tzp && error == 0) {
 504                 tz_minuteswest = tzp->tz_minuteswest;
 505                 tz_dsttime = tzp->tz_dsttime;
 506         }
 507         return (error);
 508 }
 509
 510 /*
 511  * Get value of an interval timer.  The process virtual and profiling virtual
 512  * time timers are kept in the p_stats area, since they can be swapped out.
 513  * These are kept internally in the way they are specified externally: in
 514  * time until they expire.
 515  *
 516  * The real time interval timer is kept in the process table slot for the
 517  * process, and its value (it_value) is kept as an absolute time rather than
 518  * as a delta, so that it is easy to keep periodic real-time signals from
 519  * drifting.
 520  *
 521  * Virtual time timers are processed in the hardclock() routine of
 522  * kern_clock.c.  The real time timer is processed by a timeout routine,
 523  * called from the softclock() routine.  Since a callout may be delayed in
 524  * real time due to interrupt processing in the system, it is possible for
 525  * the real time timeout routine (realitexpire, given below), to be delayed
 526  * in real time past when it is supposed to occur.  It does not suffice,
 527  * therefore, to reload the real timer .it_value from the real time timers
 528  * .it_interval.  Rather, we compute the next time in absolute time the timer
 529  * should go off.
 530  */
 531 #ifndef _SYS_SYSPROTO_H_
 532 struct getitimer_args {
 533         u_int   which;
 534         struct  itimerval *itv;
 535 };
 536 #endif
 537 int
 538 getitimer(struct thread *td, struct getitimer_args *uap)
 539 {
 540         struct itimerval aitv;
 541         int error;
 542
 543         error = kern_getitimer(td, uap->which, &aitv);
 544         if (error != 0)
 545                 return (error);
 546         return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
 547 }
 548
 549 int
 550 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
 551 {
 552         struct proc *p = td->td_proc;
 553         struct timeval ctv;
 554
 555         if (which > ITIMER_PROF)
 556                 return (EINVAL);
 557
 558         if (which == ITIMER_REAL) {
 559                 /*
 560                  * Convert from absolute to relative time in .it_value
 561                  * part of real time timer.  If time for real time timer
 562                  * has passed return 0, else return difference between
 563                  * current time and time for the timer to go off.
 564                  */
 565                 PROC_LOCK(p);
 566                 *aitv = p->p_realtimer;
 567                 PROC_UNLOCK(p);
 568                 if (timevalisset(&aitv->it_value)) {
 569                         getmicrouptime(&ctv);
 570                         if (timevalcmp(&aitv->it_value, &ctv, <))
 571                                 timevalclear(&aitv->it_value);
 572                         else
 573                                 timevalsub(&aitv->it_value, &ctv);
 574                 }
 575         } else {
 576                 PROC_SLOCK(p);
 577                 *aitv = p->p_stats->p_timer[which];
 578                 PROC_SUNLOCK(p);
 579         }
 580         return (0);
 581 }
 582
 583 #ifndef _SYS_SYSPROTO_H_
 584 struct setitimer_args {
 585         u_int   which;
 586         struct  itimerval *itv, *oitv;
 587 };
 588 #endif
 589 int
 590 setitimer(struct thread *td, struct setitimer_args *uap)
 591 {
 592         struct itimerval aitv, oitv;
 593         int error;
 594
 595         if (uap->itv == NULL) {
 596                 uap->itv = uap->oitv;
 597                 return (getitimer(td, (struct getitimer_args *)uap));
 598         }
 599
 600         if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
 601                 return (error);
 602         error = kern_setitimer(td, uap->which, &aitv, &oitv);
 603         if (error != 0 || uap->oitv == NULL)
 604                 return (error);
 605         return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
 606 }
 607
 608 int
 609 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
 610     struct itimerval *oitv)
 611 {
 612         struct proc *p = td->td_proc;
 613         struct timeval ctv;
 614
 615         if (aitv == NULL)
 616                 return (kern_getitimer(td, which, oitv));
 617
 618         if (which > ITIMER_PROF)
 619                 return (EINVAL);
 620         if (itimerfix(&aitv->it_value))
 621                 return (EINVAL);
 622         if (!timevalisset(&aitv->it_value))
 623                 timevalclear(&aitv->it_interval);
 624         else if (itimerfix(&aitv->it_interval))
 625                 return (EINVAL);
 626
 627         if (which == ITIMER_REAL) {
 628                 PROC_LOCK(p);
 629                 if (timevalisset(&p->p_realtimer.it_value))
 630                         callout_stop(&p->p_itcallout);
 631                 getmicrouptime(&ctv);
 632                 if (timevalisset(&aitv->it_value)) {
 633                         callout_reset(&p->p_itcallout, tvtohz(&aitv->it_value),
 634                             realitexpire, p);
 635                         timevaladd(&aitv->it_value, &ctv);
 636                 }
 637                 *oitv = p->p_realtimer;
 638                 p->p_realtimer = *aitv;
 639                 PROC_UNLOCK(p);
 640                 if (timevalisset(&oitv->it_value)) {
 641                         if (timevalcmp(&oitv->it_value, &ctv, <))
 642                                 timevalclear(&oitv->it_value);
 643                         else
 644                                 timevalsub(&oitv->it_value, &ctv);
 645                 }
 646         } else {
 647                 PROC_SLOCK(p);
 648                 *oitv = p->p_stats->p_timer[which];
 649                 p->p_stats->p_timer[which] = *aitv;
 650                 PROC_SUNLOCK(p);
 651         }
 652         return (0);
 653 }
 654
 655 /*
 656  * Real interval timer expired:
 657  * send process whose timer expired an alarm signal.
 658  * If time is not set up to reload, then just return.
 659  * Else compute next time timer should go off which is > current time.
 660  * This is where delay in processing this timeout causes multiple
 661  * SIGALRM calls to be compressed into one.
 662  * tvtohz() always adds 1 to allow for the time until the next clock
 663  * interrupt being strictly less than 1 clock tick, but we don't want
 664  * that here since we want to appear to be in sync with the clock
 665  * interrupt even when we're delayed.
 666  */
 667 void
 668 realitexpire(void *arg)
 669 {
 670         struct proc *p;
 671         struct timeval ctv, ntv;
 672
 673         p = (struct proc *)arg;
 674         PROC_LOCK(p);
 675         psignal(p, SIGALRM);
 676         if (!timevalisset(&p->p_realtimer.it_interval)) {
 677                 timevalclear(&p->p_realtimer.it_value);
 678                 if (p->p_flag & P_WEXIT)
 679                         wakeup(&p->p_itcallout);
 680                 PROC_UNLOCK(p);
 681                 return;
 682         }
 683         for (;;) {
 684                 timevaladd(&p->p_realtimer.it_value,
 685                     &p->p_realtimer.it_interval);
 686                 getmicrouptime(&ctv);
 687                 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
 688                         ntv = p->p_realtimer.it_value;
 689                         timevalsub(&ntv, &ctv);
 690                         callout_reset(&p->p_itcallout, tvtohz(&ntv) - 1,
 691                             realitexpire, p);
 692                         PROC_UNLOCK(p);
 693                         return;
 694                 }
 695         }
 696         /*NOTREACHED*/
 697 }
 698
 699 /*
 700  * Check that a proposed value to load into the .it_value or
 701  * .it_interval part of an interval timer is acceptable, and
 702  * fix it to have at least minimal value (i.e. if it is less
 703  * than the resolution of the clock, round it up.)
 704  */
 705 int
 706 itimerfix(struct timeval *tv)
 707 {
 708
 709         if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
 710                 return (EINVAL);
 711         if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
 712                 tv->tv_usec = tick;
 713         return (0);
 714 }
 715
 716 /*
 717  * Decrement an interval timer by a specified number
 718  * of microseconds, which must be less than a second,
 719  * i.e. < 1000000.  If the timer expires, then reload
 720  * it.  In this case, carry over (usec - old value) to
 721  * reduce the value reloaded into the timer so that
 722  * the timer does not drift.  This routine assumes
 723  * that it is called in a context where the timers
 724  * on which it is operating cannot change in value.
 725  */
 726 int
 727 itimerdecr(struct itimerval *itp, int usec)
 728 {
 729
 730         if (itp->it_value.tv_usec < usec) {
 731                 if (itp->it_value.tv_sec == 0) {
 732                         /* expired, and already in next interval */
 733                         usec -= itp->it_value.tv_usec;
 734                         goto expire;
 735                 }
 736                 itp->it_value.tv_usec += 1000000;
 737                 itp->it_value.tv_sec--;
 738         }
 739         itp->it_value.tv_usec -= usec;
 740         usec = 0;
 741         if (timevalisset(&itp->it_value))
 742                 return (1);
 743         /* expired, exactly at end of interval */
 744 expire:
 745         if (timevalisset(&itp->it_interval)) {
 746                 itp->it_value = itp->it_interval;
 747                 itp->it_value.tv_usec -= usec;
 748                 if (itp->it_value.tv_usec < 0) {
 749                         itp->it_value.tv_usec += 1000000;
 750                         itp->it_value.tv_sec--;
 751                 }
 752         } else
 753                 itp->it_value.tv_usec = 0;              /* sec is already 0 */
 754         return (0);
 755 }
 756
 757 /*
 758  * Add and subtract routines for timevals.
 759  * N.B.: subtract routine doesn't deal with
 760  * results which are before the beginning,
 761  * it just gets very confused in this case.
 762  * Caveat emptor.
 763  */
 764 void
 765 timevaladd(struct timeval *t1, const struct timeval *t2)
 766 {
 767
 768         t1->tv_sec += t2->tv_sec;
 769         t1->tv_usec += t2->tv_usec;
 770         timevalfix(t1);
 771 }
 772
 773 void
 774 timevalsub(struct timeval *t1, const struct timeval *t2)
 775 {
 776
 777         t1->tv_sec -= t2->tv_sec;
 778         t1->tv_usec -= t2->tv_usec;
 779         timevalfix(t1);
 780 }
 781
 782 static void
 783 timevalfix(struct timeval *t1)
 784 {
 785
 786         if (t1->tv_usec < 0) {
 787                 t1->tv_sec--;
 788                 t1->tv_usec += 1000000;
 789         }
 790         if (t1->tv_usec >= 1000000) {
 791                 t1->tv_sec++;
 792                 t1->tv_usec -= 1000000;
 793         }
 794 }
 795
 796 /*
 797  * ratecheck(): simple time-based rate-limit checking.
 798  */
 799 int
 800 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
 801 {
 802         struct timeval tv, delta;
 803         int rv = 0;
 804
 805         getmicrouptime(&tv);            /* NB: 10ms precision */
 806         delta = tv;
 807         timevalsub(&delta, lasttime);
 808
 809         /*
 810          * check for 0,0 is so that the message will be seen at least once,
 811          * even if interval is huge.
 812          */
 813         if (timevalcmp(&delta, mininterval, >=) ||
 814             (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
 815                 *lasttime = tv;
 816                 rv = 1;
 817         }
 818
 819         return (rv);
 820 }
 821
 822 /*
 823  * ppsratecheck(): packets (or events) per second limitation.
 824  *
 825  * Return 0 if the limit is to be enforced (e.g. the caller
 826  * should drop a packet because of the rate limitation).
 827  *
 828  * maxpps of 0 always causes zero to be returned.  maxpps of -1
 829  * always causes 1 to be returned; this effectively defeats rate
 830  * limiting.
 831  *
 832  * Note that we maintain the struct timeval for compatibility
 833  * with other bsd systems.  We reuse the storage and just monitor
 834  * clock ticks for minimal overhead.
 835  */
 836 int
 837 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
 838 {
 839         int now;
 840
 841         /*
 842          * Reset the last time and counter if this is the first call
 843          * or more than a second has passed since the last update of
 844          * lasttime.
 845          */
 846         now = ticks;
 847         if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
 848                 lasttime->tv_sec = now;
 849                 *curpps = 1;
 850                 return (maxpps != 0);
 851         } else {
 852                 (*curpps)++;            /* NB: ignore potential overflow */
 853                 return (maxpps < 0 || *curpps < maxpps);
 854         }
 855 }
 856
 857 static void
 858 itimer_start(void)
 859 {
 860         struct kclock rt_clock = {
 861                 .timer_create  = realtimer_create,
 862                 .timer_delete  = realtimer_delete,
 863                 .timer_settime = realtimer_settime,
 864                 .timer_gettime = realtimer_gettime,
 865                 .event_hook    = NULL
 866         };
 867
 868         itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
 869                 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
 870         register_posix_clock(CLOCK_REALTIME,  &rt_clock);
 871         register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
 872         p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
 873         p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
 874         p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
 875         EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
 876                 (void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
 877         EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
 878                 (void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
 879 }
 880
 881 int
 882 register_posix_clock(int clockid, struct kclock *clk)
 883 {
 884         if ((unsigned)clockid >= MAX_CLOCKS) {
 885                 printf("%s: invalid clockid\n", __func__);
 886                 return (0);
 887         }
 888         posix_clocks[clockid] = *clk;
 889         return (1);
 890 }
 891
 892 static int
 893 itimer_init(void *mem, int size, int flags)
 894 {
 895         struct itimer *it;
 896
 897         it = (struct itimer *)mem;
 898         mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
 899         return (0);
 900 }
 901
 902 static void
 903 itimer_fini(void *mem, int size)
 904 {
 905         struct itimer *it;
 906
 907         it = (struct itimer *)mem;
 908         mtx_destroy(&it->it_mtx);
 909 }
 910
 911 static void
 912 itimer_enter(struct itimer *it)
 913 {
 914
 915         mtx_assert(&it->it_mtx, MA_OWNED);
 916         it->it_usecount++;
 917 }
 918
 919 static void
 920 itimer_leave(struct itimer *it)
 921 {
 922
 923         mtx_assert(&it->it_mtx, MA_OWNED);
 924         KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
 925
 926         if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
 927                 wakeup(it);
 928 }
 929
 930 #ifndef _SYS_SYSPROTO_H_
 931 struct ktimer_create_args {
 932         clockid_t clock_id;
 933         struct sigevent * evp;
 934         int * timerid;
 935 };
 936 #endif
 937 int
 938 ktimer_create(struct thread *td, struct ktimer_create_args *uap)
 939 {
 940         struct sigevent *evp1, ev;
 941         int id;
 942         int error;
 943
 944         if (uap->evp != NULL) {
 945                 error = copyin(uap->evp, &ev, sizeof(ev));
 946                 if (error != 0)
 947                         return (error);
 948                 evp1 = &ev;
 949         } else
 950                 evp1 = NULL;
 951
 952         error = kern_timer_create(td, uap->clock_id, evp1, &id, -1);
 953
 954         if (error == 0) {
 955                 error = copyout(&id, uap->timerid, sizeof(int));
 956                 if (error != 0)
 957                         kern_timer_delete(td, id);
 958         }
 959         return (error);
 960 }
 961
 962 static int
 963 kern_timer_create(struct thread *td, clockid_t clock_id,
 964         struct sigevent *evp, int *timerid, int preset_id)
 965 {
 966         struct proc *p = td->td_proc;
 967         struct itimer *it;
 968         int id;
 969         int error;
 970
 971         if (clock_id < 0 || clock_id >= MAX_CLOCKS)
 972                 return (EINVAL);
 973
 974         if (posix_clocks[clock_id].timer_create == NULL)
 975                 return (EINVAL);
 976
 977         if (evp != NULL) {
 978                 if (evp->sigev_notify != SIGEV_NONE &&
 979                     evp->sigev_notify != SIGEV_SIGNAL &&
 980                     evp->sigev_notify != SIGEV_THREAD_ID)
 981                         return (EINVAL);
 982                 if ((evp->sigev_notify == SIGEV_SIGNAL ||
 983                      evp->sigev_notify == SIGEV_THREAD_ID) &&
 984                         !_SIG_VALID(evp->sigev_signo))
 985                         return (EINVAL);
 986         }
 987
 988         if (p->p_itimers == NULL)
 989                 itimers_alloc(p);
 990
 991         it = uma_zalloc(itimer_zone, M_WAITOK);
 992         it->it_flags = 0;
 993         it->it_usecount = 0;
 994         it->it_active = 0;
 995         timespecclear(&it->it_time.it_value);
 996         timespecclear(&it->it_time.it_interval);
 997         it->it_overrun = 0;
 998         it->it_overrun_last = 0;
 999         it->it_clockid = clock_id;
1000         it->it_timerid = -1;
1001         it->it_proc = p;
1002         ksiginfo_init(&it->it_ksi);
1003         it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1004         error = CLOCK_CALL(clock_id, timer_create, (it));
1005         if (error != 0)
1006                 goto out;
1007
1008         PROC_LOCK(p);
1009         if (preset_id != -1) {
1010                 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1011                 id = preset_id;
1012                 if (p->p_itimers->its_timers[id] != NULL) {
1013                         PROC_UNLOCK(p);
1014                         error = 0;
1015                         goto out;
1016                 }
1017         } else {
1018                 /*
1019                  * Find a free timer slot, skipping those reserved
1020                  * for setitimer().
1021                  */
1022                 for (id = 3; id < TIMER_MAX; id++)
1023                         if (p->p_itimers->its_timers[id] == NULL)
1024                                 break;
1025                 if (id == TIMER_MAX) {
1026                         PROC_UNLOCK(p);
1027                         error = EAGAIN;
1028                         goto out;
1029                 }
1030         }
1031         it->it_timerid = id;
1032         p->p_itimers->its_timers[id] = it;
1033         if (evp != NULL)
1034                 it->it_sigev = *evp;
1035         else {
1036                 it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1037                 switch (clock_id) {
1038                 default:
1039                 case CLOCK_REALTIME:
1040                         it->it_sigev.sigev_signo = SIGALRM;
1041                         break;
1042                 case CLOCK_VIRTUAL:
1043                         it->it_sigev.sigev_signo = SIGVTALRM;
1044                         break;
1045                 case CLOCK_PROF:
1046                         it->it_sigev.sigev_signo = SIGPROF;
1047                         break;
1048                 }
1049                 it->it_sigev.sigev_value.sival_int = id;
1050         }
1051
1052         if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1053             it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1054                 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1055                 it->it_ksi.ksi_code = SI_TIMER;
1056                 it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1057                 it->it_ksi.ksi_timerid = id;
1058         }
1059         PROC_UNLOCK(p);
1060         *timerid = id;
1061         return (0);
1062
1063 out:
1064         ITIMER_LOCK(it);
1065         CLOCK_CALL(it->it_clockid, timer_delete, (it));
1066         ITIMER_UNLOCK(it);
1067         uma_zfree(itimer_zone, it);
1068         return (error);
1069 }
1070
1071 #ifndef _SYS_SYSPROTO_H_
1072 struct ktimer_delete_args {
1073         int timerid;
1074 };
1075 #endif
1076 int
1077 ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1078 {
1079         return (kern_timer_delete(td, uap->timerid));
1080 }
1081
1082 static struct itimer *
1083 itimer_find(struct proc *p, int timerid)
1084 {
1085         struct itimer *it;
1086
1087         PROC_LOCK_ASSERT(p, MA_OWNED);
1088         if ((p->p_itimers == NULL) || (timerid >= TIMER_MAX) ||
1089             (it = p->p_itimers->its_timers[timerid]) == NULL) {
1090                 return (NULL);
1091         }
1092         ITIMER_LOCK(it);
1093         if ((it->it_flags & ITF_DELETING) != 0) {
1094                 ITIMER_UNLOCK(it);
1095                 it = NULL;
1096         }
1097         return (it);
1098 }
1099
1100 static int
1101 kern_timer_delete(struct thread *td, int timerid)
1102 {
1103         struct proc *p = td->td_proc;
1104         struct itimer *it;
1105
1106         PROC_LOCK(p);
1107         it = itimer_find(p, timerid);
1108         if (it == NULL) {
1109                 PROC_UNLOCK(p);
1110                 return (EINVAL);
1111         }
1112         PROC_UNLOCK(p);
1113
1114         it->it_flags |= ITF_DELETING;
1115         while (it->it_usecount > 0) {
1116                 it->it_flags |= ITF_WANTED;
1117                 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1118         }
1119         it->it_flags &= ~ITF_WANTED;
1120         CLOCK_CALL(it->it_clockid, timer_delete, (it));
1121         ITIMER_UNLOCK(it);
1122
1123         PROC_LOCK(p);
1124         if (KSI_ONQ(&it->it_ksi))
1125                 sigqueue_take(&it->it_ksi);
1126         p->p_itimers->its_timers[timerid] = NULL;
1127         PROC_UNLOCK(p);
1128         uma_zfree(itimer_zone, it);
1129         return (0);
1130 }
1131
1132 #ifndef _SYS_SYSPROTO_H_
1133 struct ktimer_settime_args {
1134         int timerid;
1135         int flags;
1136         const struct itimerspec * value;
1137         struct itimerspec * ovalue;
1138 };
1139 #endif
1140 int
1141 ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1142 {
1143         struct proc *p = td->td_proc;
1144         struct itimer *it;
1145         struct itimerspec val, oval, *ovalp;
1146         int error;
1147
1148         error = copyin(uap->value, &val, sizeof(val));
1149         if (error != 0)
1150                 return (error);
1151
1152         if (uap->ovalue != NULL)
1153                 ovalp = &oval;
1154         else
1155                 ovalp = NULL;
1156
1157         PROC_LOCK(p);
1158         if (uap->timerid < 3 ||
1159             (it = itimer_find(p, uap->timerid)) == NULL) {
1160                 PROC_UNLOCK(p);
1161                 error = EINVAL;
1162         } else {
1163                 PROC_UNLOCK(p);
1164                 itimer_enter(it);
1165                 error = CLOCK_CALL(it->it_clockid, timer_settime,
1166                                 (it, uap->flags, &val, ovalp));
1167                 itimer_leave(it);
1168                 ITIMER_UNLOCK(it);
1169         }
1170         if (error == 0 && uap->ovalue != NULL)
1171                 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1172         return (error);
1173 }
1174
1175 #ifndef _SYS_SYSPROTO_H_
1176 struct ktimer_gettime_args {
1177         int timerid;
1178         struct itimerspec * value;
1179 };
1180 #endif
1181 int
1182 ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1183 {
1184         struct proc *p = td->td_proc;
1185         struct itimer *it;
1186         struct itimerspec val;
1187         int error;
1188
1189         PROC_LOCK(p);
1190         if (uap->timerid < 3 ||
1191            (it = itimer_find(p, uap->timerid)) == NULL) {
1192                 PROC_UNLOCK(p);
1193                 error = EINVAL;
1194         } else {
1195                 PROC_UNLOCK(p);
1196                 itimer_enter(it);
1197                 error = CLOCK_CALL(it->it_clockid, timer_gettime,
1198                                 (it, &val));
1199                 itimer_leave(it);
1200                 ITIMER_UNLOCK(it);
1201         }
1202         if (error == 0)
1203                 error = copyout(&val, uap->value, sizeof(val));
1204         return (error);
1205 }
1206
1207 #ifndef _SYS_SYSPROTO_H_
1208 struct timer_getoverrun_args {
1209         int timerid;
1210 };
1211 #endif
1212 int
1213 ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1214 {
1215         struct proc *p = td->td_proc;
1216         struct itimer *it;
1217         int error ;
1218
1219         PROC_LOCK(p);
1220         if (uap->timerid < 3 ||
1221             (it = itimer_find(p, uap->timerid)) == NULL) {
1222                 PROC_UNLOCK(p);
1223                 error = EINVAL;
1224         } else {
1225                 td->td_retval[0] = it->it_overrun_last;
1226                 ITIMER_UNLOCK(it);
1227                 PROC_UNLOCK(p);
1228                 error = 0;
1229         }
1230         return (error);
1231 }
1232
1233 static int
1234 realtimer_create(struct itimer *it)
1235 {
1236         callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1237         return (0);
1238 }
1239
1240 static int
1241 realtimer_delete(struct itimer *it)
1242 {
1243         mtx_assert(&it->it_mtx, MA_OWNED);
1244
1245         ITIMER_UNLOCK(it);
1246         callout_drain(&it->it_callout);
1247         ITIMER_LOCK(it);
1248         return (0);
1249 }
1250
1251 static int
1252 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1253 {
1254         struct timespec cts;
1255
1256         mtx_assert(&it->it_mtx, MA_OWNED);
1257
1258         realtimer_clocktime(it->it_clockid, &cts);
1259         *ovalue = it->it_time;
1260         if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1261                 timespecsub(&ovalue->it_value, &cts);
1262                 if (ovalue->it_value.tv_sec < 0 ||
1263                     (ovalue->it_value.tv_sec == 0 &&
1264                      ovalue->it_value.tv_nsec == 0)) {
1265                         ovalue->it_value.tv_sec  = 0;
1266                         ovalue->it_value.tv_nsec = 1;
1267                 }
1268         }
1269         return (0);
1270 }
1271
1272 static int
1273 realtimer_settime(struct itimer *it, int flags,
1274         struct itimerspec *value, struct itimerspec *ovalue)
1275 {
1276         struct timespec cts, ts;
1277         struct timeval tv;
1278         struct itimerspec val;
1279
1280         mtx_assert(&it->it_mtx, MA_OWNED);
1281
1282         val = *value;
1283         if (itimespecfix(&val.it_value))
1284                 return (EINVAL);
1285
1286         if (timespecisset(&val.it_value)) {
1287                 if (itimespecfix(&val.it_interval))
1288                         return (EINVAL);
1289         } else {
1290                 timespecclear(&val.it_interval);
1291         }
1292
1293         if (ovalue != NULL)
1294                 realtimer_gettime(it, ovalue);
1295
1296         it->it_time = val;
1297         if (timespecisset(&val.it_value)) {
1298                 realtimer_clocktime(it->it_clockid, &cts);
1299                 ts = val.it_value;
1300                 if ((flags & TIMER_ABSTIME) == 0) {
1301                         /* Convert to absolute time. */
1302                         timespecadd(&it->it_time.it_value, &cts);
1303                 } else {
1304                         timespecsub(&ts, &cts);
1305                         /*
1306                          * We don't care if ts is negative, tztohz will
1307                          * fix it.
1308                          */
1309                 }
1310                 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1311                 callout_reset(&it->it_callout, tvtohz(&tv),
1312                         realtimer_expire, it);
1313         } else {
1314                 callout_stop(&it->it_callout);
1315         }
1316
1317         return (0);
1318 }
1319
1320 static void
1321 realtimer_clocktime(clockid_t id, struct timespec *ts)
1322 {
1323         if (id == CLOCK_REALTIME)
1324                 getnanotime(ts);
1325         else    /* CLOCK_MONOTONIC */
1326                 getnanouptime(ts);
1327 }
1328
1329 int
1330 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1331 {
1332         struct itimer *it;
1333
1334         PROC_LOCK_ASSERT(p, MA_OWNED);
1335         it = itimer_find(p, timerid);
1336         if (it != NULL) {
1337                 ksi->ksi_overrun = it->it_overrun;
1338                 it->it_overrun_last = it->it_overrun;
1339                 it->it_overrun = 0;
1340                 ITIMER_UNLOCK(it);
1341                 return (0);
1342         }
1343         return (EINVAL);
1344 }
1345
1346 int
1347 itimespecfix(struct timespec *ts)
1348 {
1349
1350         if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1351                 return (EINVAL);
1352         if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1353                 ts->tv_nsec = tick * 1000;
1354         return (0);
1355 }
1356
1357 /* Timeout callback for realtime timer */
1358 static void
1359 realtimer_expire(void *arg)
1360 {
1361         struct timespec cts, ts;
1362         struct timeval tv;
1363         struct itimer *it;
1364         struct proc *p;
1365
1366         it = (struct itimer *)arg;
1367         p = it->it_proc;
1368
1369         realtimer_clocktime(it->it_clockid, &cts);
1370         /* Only fire if time is reached. */
1371         if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1372                 if (timespecisset(&it->it_time.it_interval)) {
1373                         timespecadd(&it->it_time.it_value,
1374                                     &it->it_time.it_interval);
1375                         while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1376                                 if (it->it_overrun < INT_MAX)
1377                                         it->it_overrun++;
1378                                 else
1379                                         it->it_ksi.ksi_errno = ERANGE;
1380                                 timespecadd(&it->it_time.it_value,
1381                                             &it->it_time.it_interval);
1382                         }
1383                 } else {
1384                         /* single shot timer ? */
1385                         timespecclear(&it->it_time.it_value);
1386                 }
1387                 if (timespecisset(&it->it_time.it_value)) {
1388                         ts = it->it_time.it_value;
1389                         timespecsub(&ts, &cts);
1390                         TIMESPEC_TO_TIMEVAL(&tv, &ts);
1391                         callout_reset(&it->it_callout, tvtohz(&tv),
1392                                  realtimer_expire, it);
1393                 }
1394                 ITIMER_UNLOCK(it);
1395                 itimer_fire(it);
1396                 ITIMER_LOCK(it);
1397         } else if (timespecisset(&it->it_time.it_value)) {
1398                 ts = it->it_time.it_value;
1399                 timespecsub(&ts, &cts);
1400                 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1401                 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1402                         it);
1403         }
1404 }
1405
1406 void
1407 itimer_fire(struct itimer *it)
1408 {
1409         struct proc *p = it->it_proc;
1410         int ret;
1411
1412         if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1413             it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1414                 PROC_LOCK(p);
1415                 if (!KSI_ONQ(&it->it_ksi)) {
1416                         it->it_ksi.ksi_errno = 0;
1417                         ret = psignal_event(p, &it->it_sigev, &it->it_ksi);
1418                         if (__predict_false(ret != 0)) {
1419                                 it->it_overrun++;
1420                                 /*
1421                                  * Broken userland code, thread went
1422                                  * away, disarm the timer.
1423                                  */
1424                                 if (ret == ESRCH) {
1425                                         ITIMER_LOCK(it);
1426                                         timespecclear(&it->it_time.it_value);
1427                                         timespecclear(&it->it_time.it_interval);
1428                                         callout_stop(&it->it_callout);
1429                                         ITIMER_UNLOCK(it);
1430                                 }
1431                         }
1432                 } else {
1433                         if (it->it_overrun < INT_MAX)
1434                                 it->it_overrun++;
1435                         else
1436                                 it->it_ksi.ksi_errno = ERANGE;
1437                 }
1438                 PROC_UNLOCK(p);
1439         }
1440 }
1441
1442 static void
1443 itimers_alloc(struct proc *p)
1444 {
1445         struct itimers *its;
1446         int i;
1447
1448         its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1449         LIST_INIT(&its->its_virtual);
1450         LIST_INIT(&its->its_prof);
1451         TAILQ_INIT(&its->its_worklist);
1452         for (i = 0; i < TIMER_MAX; i++)
1453                 its->its_timers[i] = NULL;
1454         PROC_LOCK(p);
1455         if (p->p_itimers == NULL) {
1456                 p->p_itimers = its;
1457                 PROC_UNLOCK(p);
1458         }
1459         else {
1460                 PROC_UNLOCK(p);
1461                 free(its, M_SUBPROC);
1462         }
1463 }
1464
1465 static void
1466 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1467 {
1468         itimers_event_hook_exit(arg, p);
1469 }
1470
1471 /* Clean up timers when some process events are being triggered. */
1472 static void
1473 itimers_event_hook_exit(void *arg, struct proc *p)
1474 {
1475         struct itimers *its;
1476         struct itimer *it;
1477         int event = (int)(intptr_t)arg;
1478         int i;
1479
1480         if (p->p_itimers != NULL) {
1481                 its = p->p_itimers;
1482                 for (i = 0; i < MAX_CLOCKS; ++i) {
1483                         if (posix_clocks[i].event_hook != NULL)
1484                                 CLOCK_CALL(i, event_hook, (p, i, event));
1485                 }
1486                 /*
1487                  * According to susv3, XSI interval timers should be inherited
1488                  * by new image.
1489                  */
1490                 if (event == ITIMER_EV_EXEC)
1491                         i = 3;
1492                 else if (event == ITIMER_EV_EXIT)
1493                         i = 0;
1494                 else
1495                         panic("unhandled event");
1496                 for (; i < TIMER_MAX; ++i) {
1497                         if ((it = its->its_timers[i]) != NULL)
1498                                 kern_timer_delete(curthread, i);
1499                 }
1500                 if (its->its_timers[0] == NULL &&
1501                     its->its_timers[1] == NULL &&
1502                     its->its_timers[2] == NULL) {
1503                         free(its, M_SUBPROC);
1504                         p->p_itimers = NULL;
1505                 }
1506         }
1507 }