The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_time.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * SPDX-License-Identifier: BSD-3-Clause
    3  *
    4  * Copyright (c) 1982, 1986, 1989, 1993
    5  *      The Regents of the University of California.  All rights reserved.
    6  *
    7  * Redistribution and use in source and binary forms, with or without
    8  * modification, are permitted provided that the following conditions
    9  * are met:
   10  * 1. Redistributions of source code must retain the above copyright
   11  *    notice, this list of conditions and the following disclaimer.
   12  * 2. Redistributions in binary form must reproduce the above copyright
   13  *    notice, this list of conditions and the following disclaimer in the
   14  *    documentation and/or other materials provided with the distribution.
   15  * 3. Neither the name of the University nor the names of its contributors
   16  *    may be used to endorse or promote products derived from this software
   17  *    without specific prior written permission.
   18  *
   19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   29  * SUCH DAMAGE.
   30  *
   31  *      @(#)kern_time.c 8.1 (Berkeley) 6/10/93
   32  */
   33 
   34 #include <sys/cdefs.h>
   35 __FBSDID("$FreeBSD$");
   36 
   37 #include "opt_ktrace.h"
   38 
   39 #include <sys/param.h>
   40 #include <sys/systm.h>
   41 #include <sys/limits.h>
   42 #include <sys/clock.h>
   43 #include <sys/lock.h>
   44 #include <sys/mutex.h>
   45 #include <sys/sysproto.h>
   46 #include <sys/resourcevar.h>
   47 #include <sys/signalvar.h>
   48 #include <sys/kernel.h>
   49 #include <sys/sleepqueue.h>
   50 #include <sys/syscallsubr.h>
   51 #include <sys/sysctl.h>
   52 #include <sys/priv.h>
   53 #include <sys/proc.h>
   54 #include <sys/posix4.h>
   55 #include <sys/time.h>
   56 #include <sys/timers.h>
   57 #include <sys/timetc.h>
   58 #include <sys/vnode.h>
   59 #ifdef KTRACE
   60 #include <sys/ktrace.h>
   61 #endif
   62 
   63 #include <vm/vm.h>
   64 #include <vm/vm_extern.h>
   65 
   66 #define MAX_CLOCKS      (CLOCK_MONOTONIC+1)
   67 #define CPUCLOCK_BIT            0x80000000
   68 #define CPUCLOCK_PROCESS_BIT    0x40000000
   69 #define CPUCLOCK_ID_MASK        (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
   70 #define MAKE_THREAD_CPUCLOCK(tid)       (CPUCLOCK_BIT|(tid))
   71 #define MAKE_PROCESS_CPUCLOCK(pid)      \
   72         (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
   73 
   74 #define NS_PER_SEC      1000000000
   75 
   76 static struct kclock    posix_clocks[MAX_CLOCKS];
   77 static uma_zone_t       itimer_zone = NULL;
   78 
   79 /*
   80  * Time of day and interval timer support.
   81  *
   82  * These routines provide the kernel entry points to get and set
   83  * the time-of-day and per-process interval timers.  Subroutines
   84  * here provide support for adding and subtracting timeval structures
   85  * and decrementing interval timers, optionally reloading the interval
   86  * timers when they expire.
   87  */
   88 
   89 static int      settime(struct thread *, struct timeval *);
   90 static void     timevalfix(struct timeval *);
   91 static int      user_clock_nanosleep(struct thread *td, clockid_t clock_id,
   92                     int flags, const struct timespec *ua_rqtp,
   93                     struct timespec *ua_rmtp);
   94 
   95 static void     itimer_start(void);
   96 static int      itimer_init(void *, int, int);
   97 static void     itimer_fini(void *, int);
   98 static void     itimer_enter(struct itimer *);
   99 static void     itimer_leave(struct itimer *);
  100 static struct itimer *itimer_find(struct proc *, int);
  101 static void     itimers_alloc(struct proc *);
  102 static int      realtimer_create(struct itimer *);
  103 static int      realtimer_gettime(struct itimer *, struct itimerspec *);
  104 static int      realtimer_settime(struct itimer *, int,
  105                         struct itimerspec *, struct itimerspec *);
  106 static int      realtimer_delete(struct itimer *);
  107 static void     realtimer_clocktime(clockid_t, struct timespec *);
  108 static void     realtimer_expire(void *);
  109 static void     realtimer_expire_l(struct itimer *it, bool proc_locked);
  110 
  111 static void     realitexpire(void *arg);
  112 
  113 static int      register_posix_clock(int, const struct kclock *);
  114 static void     itimer_fire(struct itimer *it);
  115 static int      itimespecfix(struct timespec *ts);
  116 
  117 #define CLOCK_CALL(clock, call, arglist)                \
  118         ((*posix_clocks[clock].call) arglist)
  119 
  120 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
  121 
  122 static int
  123 settime(struct thread *td, struct timeval *tv)
  124 {
  125         struct timeval delta, tv1, tv2;
  126         static struct timeval maxtime, laststep;
  127         struct timespec ts;
  128 
  129         microtime(&tv1);
  130         delta = *tv;
  131         timevalsub(&delta, &tv1);
  132 
  133         /*
  134          * If the system is secure, we do not allow the time to be 
  135          * set to a value earlier than 1 second less than the highest
  136          * time we have yet seen. The worst a miscreant can do in
  137          * this circumstance is "freeze" time. He couldn't go
  138          * back to the past.
  139          *
  140          * We similarly do not allow the clock to be stepped more
  141          * than one second, nor more than once per second. This allows
  142          * a miscreant to make the clock march double-time, but no worse.
  143          */
  144         if (securelevel_gt(td->td_ucred, 1) != 0) {
  145                 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
  146                         /*
  147                          * Update maxtime to latest time we've seen.
  148                          */
  149                         if (tv1.tv_sec > maxtime.tv_sec)
  150                                 maxtime = tv1;
  151                         tv2 = *tv;
  152                         timevalsub(&tv2, &maxtime);
  153                         if (tv2.tv_sec < -1) {
  154                                 tv->tv_sec = maxtime.tv_sec - 1;
  155                                 printf("Time adjustment clamped to -1 second\n");
  156                         }
  157                 } else {
  158                         if (tv1.tv_sec == laststep.tv_sec)
  159                                 return (EPERM);
  160                         if (delta.tv_sec > 1) {
  161                                 tv->tv_sec = tv1.tv_sec + 1;
  162                                 printf("Time adjustment clamped to +1 second\n");
  163                         }
  164                         laststep = *tv;
  165                 }
  166         }
  167 
  168         ts.tv_sec = tv->tv_sec;
  169         ts.tv_nsec = tv->tv_usec * 1000;
  170         tc_setclock(&ts);
  171         resettodr();
  172         return (0);
  173 }
  174 
  175 #ifndef _SYS_SYSPROTO_H_
  176 struct clock_getcpuclockid2_args {
  177         id_t id;
  178         int which,
  179         clockid_t *clock_id;
  180 };
  181 #endif
  182 /* ARGSUSED */
  183 int
  184 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
  185 {
  186         clockid_t clk_id;
  187         int error;
  188 
  189         error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
  190         if (error == 0)
  191                 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
  192         return (error);
  193 }
  194 
  195 int
  196 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
  197     clockid_t *clk_id)
  198 {
  199         struct proc *p;
  200         pid_t pid;
  201         lwpid_t tid;
  202         int error;
  203 
  204         switch (which) {
  205         case CPUCLOCK_WHICH_PID:
  206                 if (id != 0) {
  207                         error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
  208                         if (error != 0)
  209                                 return (error);
  210                         PROC_UNLOCK(p);
  211                         pid = id;
  212                 } else {
  213                         pid = td->td_proc->p_pid;
  214                 }
  215                 *clk_id = MAKE_PROCESS_CPUCLOCK(pid);
  216                 return (0);
  217         case CPUCLOCK_WHICH_TID:
  218                 tid = id == 0 ? td->td_tid : id;
  219                 *clk_id = MAKE_THREAD_CPUCLOCK(tid);
  220                 return (0);
  221         default:
  222                 return (EINVAL);
  223         }
  224 }
  225 
  226 #ifndef _SYS_SYSPROTO_H_
  227 struct clock_gettime_args {
  228         clockid_t clock_id;
  229         struct  timespec *tp;
  230 };
  231 #endif
  232 /* ARGSUSED */
  233 int
  234 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
  235 {
  236         struct timespec ats;
  237         int error;
  238 
  239         error = kern_clock_gettime(td, uap->clock_id, &ats);
  240         if (error == 0)
  241                 error = copyout(&ats, uap->tp, sizeof(ats));
  242 
  243         return (error);
  244 }
  245 
  246 static inline void
  247 cputick2timespec(uint64_t runtime, struct timespec *ats)
  248 {
  249         runtime = cputick2usec(runtime);
  250         ats->tv_sec = runtime / 1000000;
  251         ats->tv_nsec = runtime % 1000000 * 1000;
  252 }
  253 
  254 void
  255 kern_thread_cputime(struct thread *targettd, struct timespec *ats)
  256 {
  257         uint64_t runtime, curtime, switchtime;
  258 
  259         if (targettd == NULL) { /* current thread */
  260                 spinlock_enter();
  261                 switchtime = PCPU_GET(switchtime);
  262                 curtime = cpu_ticks();
  263                 runtime = curthread->td_runtime;
  264                 spinlock_exit();
  265                 runtime += curtime - switchtime;
  266         } else {
  267                 PROC_LOCK_ASSERT(targettd->td_proc, MA_OWNED);
  268                 thread_lock(targettd);
  269                 runtime = targettd->td_runtime;
  270                 thread_unlock(targettd);
  271         }
  272         cputick2timespec(runtime, ats);
  273 }
  274 
  275 void
  276 kern_process_cputime(struct proc *targetp, struct timespec *ats)
  277 {
  278         uint64_t runtime;
  279         struct rusage ru;
  280 
  281         PROC_LOCK_ASSERT(targetp, MA_OWNED);
  282         PROC_STATLOCK(targetp);
  283         rufetch(targetp, &ru);
  284         runtime = targetp->p_rux.rux_runtime;
  285         if (curthread->td_proc == targetp)
  286                 runtime += cpu_ticks() - PCPU_GET(switchtime);
  287         PROC_STATUNLOCK(targetp);
  288         cputick2timespec(runtime, ats);
  289 }
  290 
  291 static int
  292 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
  293 {
  294         struct proc *p, *p2;
  295         struct thread *td2;
  296         lwpid_t tid;
  297         pid_t pid;
  298         int error;
  299 
  300         p = td->td_proc;
  301         if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
  302                 tid = clock_id & CPUCLOCK_ID_MASK;
  303                 td2 = tdfind(tid, p->p_pid);
  304                 if (td2 == NULL)
  305                         return (EINVAL);
  306                 kern_thread_cputime(td2, ats);
  307                 PROC_UNLOCK(td2->td_proc);
  308         } else {
  309                 pid = clock_id & CPUCLOCK_ID_MASK;
  310                 error = pget(pid, PGET_CANSEE, &p2);
  311                 if (error != 0)
  312                         return (EINVAL);
  313                 kern_process_cputime(p2, ats);
  314                 PROC_UNLOCK(p2);
  315         }
  316         return (0);
  317 }
  318 
  319 int
  320 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
  321 {
  322         struct timeval sys, user;
  323         struct proc *p;
  324 
  325         p = td->td_proc;
  326         switch (clock_id) {
  327         case CLOCK_REALTIME:            /* Default to precise. */
  328         case CLOCK_REALTIME_PRECISE:
  329                 nanotime(ats);
  330                 break;
  331         case CLOCK_REALTIME_FAST:
  332                 getnanotime(ats);
  333                 break;
  334         case CLOCK_VIRTUAL:
  335                 PROC_LOCK(p);
  336                 PROC_STATLOCK(p);
  337                 calcru(p, &user, &sys);
  338                 PROC_STATUNLOCK(p);
  339                 PROC_UNLOCK(p);
  340                 TIMEVAL_TO_TIMESPEC(&user, ats);
  341                 break;
  342         case CLOCK_PROF:
  343                 PROC_LOCK(p);
  344                 PROC_STATLOCK(p);
  345                 calcru(p, &user, &sys);
  346                 PROC_STATUNLOCK(p);
  347                 PROC_UNLOCK(p);
  348                 timevaladd(&user, &sys);
  349                 TIMEVAL_TO_TIMESPEC(&user, ats);
  350                 break;
  351         case CLOCK_MONOTONIC:           /* Default to precise. */
  352         case CLOCK_MONOTONIC_PRECISE:
  353         case CLOCK_UPTIME:
  354         case CLOCK_UPTIME_PRECISE:
  355                 nanouptime(ats);
  356                 break;
  357         case CLOCK_UPTIME_FAST:
  358         case CLOCK_MONOTONIC_FAST:
  359                 getnanouptime(ats);
  360                 break;
  361         case CLOCK_SECOND:
  362                 ats->tv_sec = time_second;
  363                 ats->tv_nsec = 0;
  364                 break;
  365         case CLOCK_THREAD_CPUTIME_ID:
  366                 kern_thread_cputime(NULL, ats);
  367                 break;
  368         case CLOCK_PROCESS_CPUTIME_ID:
  369                 PROC_LOCK(p);
  370                 kern_process_cputime(p, ats);
  371                 PROC_UNLOCK(p);
  372                 break;
  373         default:
  374                 if ((int)clock_id >= 0)
  375                         return (EINVAL);
  376                 return (get_cputime(td, clock_id, ats));
  377         }
  378         return (0);
  379 }
  380 
  381 #ifndef _SYS_SYSPROTO_H_
  382 struct clock_settime_args {
  383         clockid_t clock_id;
  384         const struct    timespec *tp;
  385 };
  386 #endif
  387 /* ARGSUSED */
  388 int
  389 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
  390 {
  391         struct timespec ats;
  392         int error;
  393 
  394         if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
  395                 return (error);
  396         return (kern_clock_settime(td, uap->clock_id, &ats));
  397 }
  398 
  399 static int allow_insane_settime = 0;
  400 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN,
  401     &allow_insane_settime, 0,
  402     "do not perform possibly restrictive checks on settime(2) args");
  403 
  404 int
  405 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
  406 {
  407         struct timeval atv;
  408         int error;
  409 
  410         if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
  411                 return (error);
  412         if (clock_id != CLOCK_REALTIME)
  413                 return (EINVAL);
  414         if (!timespecvalid_interval(ats))
  415                 return (EINVAL);
  416         if (!allow_insane_settime &&
  417             (ats->tv_sec > 8000ULL * 365 * 24 * 60 * 60 ||
  418             ats->tv_sec < utc_offset()))
  419                 return (EINVAL);
  420         /* XXX Don't convert nsec->usec and back */
  421         TIMESPEC_TO_TIMEVAL(&atv, ats);
  422         error = settime(td, &atv);
  423         return (error);
  424 }
  425 
  426 #ifndef _SYS_SYSPROTO_H_
  427 struct clock_getres_args {
  428         clockid_t clock_id;
  429         struct  timespec *tp;
  430 };
  431 #endif
  432 int
  433 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
  434 {
  435         struct timespec ts;
  436         int error;
  437 
  438         if (uap->tp == NULL)
  439                 return (0);
  440 
  441         error = kern_clock_getres(td, uap->clock_id, &ts);
  442         if (error == 0)
  443                 error = copyout(&ts, uap->tp, sizeof(ts));
  444         return (error);
  445 }
  446 
  447 int
  448 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
  449 {
  450 
  451         ts->tv_sec = 0;
  452         switch (clock_id) {
  453         case CLOCK_REALTIME:
  454         case CLOCK_REALTIME_FAST:
  455         case CLOCK_REALTIME_PRECISE:
  456         case CLOCK_MONOTONIC:
  457         case CLOCK_MONOTONIC_FAST:
  458         case CLOCK_MONOTONIC_PRECISE:
  459         case CLOCK_UPTIME:
  460         case CLOCK_UPTIME_FAST:
  461         case CLOCK_UPTIME_PRECISE:
  462                 /*
  463                  * Round up the result of the division cheaply by adding 1.
  464                  * Rounding up is especially important if rounding down
  465                  * would give 0.  Perfect rounding is unimportant.
  466                  */
  467                 ts->tv_nsec = NS_PER_SEC / tc_getfrequency() + 1;
  468                 break;
  469         case CLOCK_VIRTUAL:
  470         case CLOCK_PROF:
  471                 /* Accurately round up here because we can do so cheaply. */
  472                 ts->tv_nsec = howmany(NS_PER_SEC, hz);
  473                 break;
  474         case CLOCK_SECOND:
  475                 ts->tv_sec = 1;
  476                 ts->tv_nsec = 0;
  477                 break;
  478         case CLOCK_THREAD_CPUTIME_ID:
  479         case CLOCK_PROCESS_CPUTIME_ID:
  480         cputime:
  481                 /* sync with cputick2usec */
  482                 ts->tv_nsec = 1000000 / cpu_tickrate();
  483                 if (ts->tv_nsec == 0)
  484                         ts->tv_nsec = 1000;
  485                 break;
  486         default:
  487                 if ((int)clock_id < 0)
  488                         goto cputime;
  489                 return (EINVAL);
  490         }
  491         return (0);
  492 }
  493 
  494 int
  495 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
  496 {
  497 
  498         return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt,
  499             rmt));
  500 }
  501 
  502 static uint8_t nanowait[MAXCPU];
  503 
  504 int
  505 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
  506     const struct timespec *rqt, struct timespec *rmt)
  507 {
  508         struct timespec ts, now;
  509         sbintime_t sbt, sbtt, prec, tmp;
  510         time_t over;
  511         int error;
  512         bool is_abs_real;
  513 
  514         if (rqt->tv_nsec < 0 || rqt->tv_nsec >= NS_PER_SEC)
  515                 return (EINVAL);
  516         if ((flags & ~TIMER_ABSTIME) != 0)
  517                 return (EINVAL);
  518         switch (clock_id) {
  519         case CLOCK_REALTIME:
  520         case CLOCK_REALTIME_PRECISE:
  521         case CLOCK_REALTIME_FAST:
  522         case CLOCK_SECOND:
  523                 is_abs_real = (flags & TIMER_ABSTIME) != 0;
  524                 break;
  525         case CLOCK_MONOTONIC:
  526         case CLOCK_MONOTONIC_PRECISE:
  527         case CLOCK_MONOTONIC_FAST:
  528         case CLOCK_UPTIME:
  529         case CLOCK_UPTIME_PRECISE:
  530         case CLOCK_UPTIME_FAST:
  531                 is_abs_real = false;
  532                 break;
  533         case CLOCK_VIRTUAL:
  534         case CLOCK_PROF:
  535         case CLOCK_PROCESS_CPUTIME_ID:
  536                 return (ENOTSUP);
  537         case CLOCK_THREAD_CPUTIME_ID:
  538         default:
  539                 return (EINVAL);
  540         }
  541         do {
  542                 ts = *rqt;
  543                 if ((flags & TIMER_ABSTIME) != 0) {
  544                         if (is_abs_real)
  545                                 td->td_rtcgen =
  546                                     atomic_load_acq_int(&rtc_generation);
  547                         error = kern_clock_gettime(td, clock_id, &now);
  548                         KASSERT(error == 0, ("kern_clock_gettime: %d", error));
  549                         timespecsub(&ts, &now, &ts);
  550                 }
  551                 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) {
  552                         error = EWOULDBLOCK;
  553                         break;
  554                 }
  555                 if (ts.tv_sec > INT32_MAX / 2) {
  556                         over = ts.tv_sec - INT32_MAX / 2;
  557                         ts.tv_sec -= over;
  558                 } else
  559                         over = 0;
  560                 tmp = tstosbt(ts);
  561                 prec = tmp;
  562                 prec >>= tc_precexp;
  563                 if (TIMESEL(&sbt, tmp))
  564                         sbt += tc_tick_sbt;
  565                 sbt += tmp;
  566                 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
  567                     sbt, prec, C_ABSOLUTE);
  568         } while (error == 0 && is_abs_real && td->td_rtcgen == 0);
  569         td->td_rtcgen = 0;
  570         if (error != EWOULDBLOCK) {
  571                 if (TIMESEL(&sbtt, tmp))
  572                         sbtt += tc_tick_sbt;
  573                 if (sbtt >= sbt)
  574                         return (0);
  575                 if (error == ERESTART)
  576                         error = EINTR;
  577                 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) {
  578                         ts = sbttots(sbt - sbtt);
  579                         ts.tv_sec += over;
  580                         if (ts.tv_sec < 0)
  581                                 timespecclear(&ts);
  582                         *rmt = ts;
  583                 }
  584                 return (error);
  585         }
  586         return (0);
  587 }
  588 
  589 #ifndef _SYS_SYSPROTO_H_
  590 struct nanosleep_args {
  591         struct  timespec *rqtp;
  592         struct  timespec *rmtp;
  593 };
  594 #endif
  595 /* ARGSUSED */
  596 int
  597 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
  598 {
  599 
  600         return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME,
  601             uap->rqtp, uap->rmtp));
  602 }
  603 
  604 #ifndef _SYS_SYSPROTO_H_
  605 struct clock_nanosleep_args {
  606         clockid_t clock_id;
  607         int       flags;
  608         struct  timespec *rqtp;
  609         struct  timespec *rmtp;
  610 };
  611 #endif
  612 /* ARGSUSED */
  613 int
  614 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap)
  615 {
  616         int error;
  617 
  618         error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp,
  619             uap->rmtp);
  620         return (kern_posix_error(td, error));
  621 }
  622 
  623 static int
  624 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
  625     const struct timespec *ua_rqtp, struct timespec *ua_rmtp)
  626 {
  627         struct timespec rmt, rqt;
  628         int error, error2;
  629 
  630         error = copyin(ua_rqtp, &rqt, sizeof(rqt));
  631         if (error)
  632                 return (error);
  633         error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt);
  634         if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) {
  635                 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt));
  636                 if (error2 != 0)
  637                         error = error2;
  638         }
  639         return (error);
  640 }
  641 
  642 #ifndef _SYS_SYSPROTO_H_
  643 struct gettimeofday_args {
  644         struct  timeval *tp;
  645         struct  timezone *tzp;
  646 };
  647 #endif
  648 /* ARGSUSED */
  649 int
  650 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
  651 {
  652         struct timeval atv;
  653         struct timezone rtz;
  654         int error = 0;
  655 
  656         if (uap->tp) {
  657                 microtime(&atv);
  658                 error = copyout(&atv, uap->tp, sizeof (atv));
  659         }
  660         if (error == 0 && uap->tzp != NULL) {
  661                 rtz.tz_minuteswest = 0;
  662                 rtz.tz_dsttime = 0;
  663                 error = copyout(&rtz, uap->tzp, sizeof (rtz));
  664         }
  665         return (error);
  666 }
  667 
  668 #ifndef _SYS_SYSPROTO_H_
  669 struct settimeofday_args {
  670         struct  timeval *tv;
  671         struct  timezone *tzp;
  672 };
  673 #endif
  674 /* ARGSUSED */
  675 int
  676 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
  677 {
  678         struct timeval atv, *tvp;
  679         struct timezone atz, *tzp;
  680         int error;
  681 
  682         if (uap->tv) {
  683                 error = copyin(uap->tv, &atv, sizeof(atv));
  684                 if (error)
  685                         return (error);
  686                 tvp = &atv;
  687         } else
  688                 tvp = NULL;
  689         if (uap->tzp) {
  690                 error = copyin(uap->tzp, &atz, sizeof(atz));
  691                 if (error)
  692                         return (error);
  693                 tzp = &atz;
  694         } else
  695                 tzp = NULL;
  696         return (kern_settimeofday(td, tvp, tzp));
  697 }
  698 
  699 int
  700 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
  701 {
  702         int error;
  703 
  704         error = priv_check(td, PRIV_SETTIMEOFDAY);
  705         if (error)
  706                 return (error);
  707         /* Verify all parameters before changing time. */
  708         if (tv) {
  709                 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
  710                     tv->tv_sec < 0)
  711                         return (EINVAL);
  712                 error = settime(td, tv);
  713         }
  714         return (error);
  715 }
  716 
  717 /*
  718  * Get value of an interval timer.  The process virtual and profiling virtual
  719  * time timers are kept in the p_stats area, since they can be swapped out.
  720  * These are kept internally in the way they are specified externally: in
  721  * time until they expire.
  722  *
  723  * The real time interval timer is kept in the process table slot for the
  724  * process, and its value (it_value) is kept as an absolute time rather than
  725  * as a delta, so that it is easy to keep periodic real-time signals from
  726  * drifting.
  727  *
  728  * Virtual time timers are processed in the hardclock() routine of
  729  * kern_clock.c.  The real time timer is processed by a timeout routine,
  730  * called from the softclock() routine.  Since a callout may be delayed in
  731  * real time due to interrupt processing in the system, it is possible for
  732  * the real time timeout routine (realitexpire, given below), to be delayed
  733  * in real time past when it is supposed to occur.  It does not suffice,
  734  * therefore, to reload the real timer .it_value from the real time timers
  735  * .it_interval.  Rather, we compute the next time in absolute time the timer
  736  * should go off.
  737  */
  738 #ifndef _SYS_SYSPROTO_H_
  739 struct getitimer_args {
  740         u_int   which;
  741         struct  itimerval *itv;
  742 };
  743 #endif
  744 int
  745 sys_getitimer(struct thread *td, struct getitimer_args *uap)
  746 {
  747         struct itimerval aitv;
  748         int error;
  749 
  750         error = kern_getitimer(td, uap->which, &aitv);
  751         if (error != 0)
  752                 return (error);
  753         return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
  754 }
  755 
  756 int
  757 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
  758 {
  759         struct proc *p = td->td_proc;
  760         struct timeval ctv;
  761 
  762         if (which > ITIMER_PROF)
  763                 return (EINVAL);
  764 
  765         if (which == ITIMER_REAL) {
  766                 /*
  767                  * Convert from absolute to relative time in .it_value
  768                  * part of real time timer.  If time for real time timer
  769                  * has passed return 0, else return difference between
  770                  * current time and time for the timer to go off.
  771                  */
  772                 PROC_LOCK(p);
  773                 *aitv = p->p_realtimer;
  774                 PROC_UNLOCK(p);
  775                 if (timevalisset(&aitv->it_value)) {
  776                         microuptime(&ctv);
  777                         if (timevalcmp(&aitv->it_value, &ctv, <))
  778                                 timevalclear(&aitv->it_value);
  779                         else
  780                                 timevalsub(&aitv->it_value, &ctv);
  781                 }
  782         } else {
  783                 PROC_ITIMLOCK(p);
  784                 *aitv = p->p_stats->p_timer[which];
  785                 PROC_ITIMUNLOCK(p);
  786         }
  787 #ifdef KTRACE
  788         if (KTRPOINT(td, KTR_STRUCT))
  789                 ktritimerval(aitv);
  790 #endif
  791         return (0);
  792 }
  793 
  794 #ifndef _SYS_SYSPROTO_H_
  795 struct setitimer_args {
  796         u_int   which;
  797         struct  itimerval *itv, *oitv;
  798 };
  799 #endif
  800 int
  801 sys_setitimer(struct thread *td, struct setitimer_args *uap)
  802 {
  803         struct itimerval aitv, oitv;
  804         int error;
  805 
  806         if (uap->itv == NULL) {
  807                 uap->itv = uap->oitv;
  808                 return (sys_getitimer(td, (struct getitimer_args *)uap));
  809         }
  810 
  811         if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
  812                 return (error);
  813         error = kern_setitimer(td, uap->which, &aitv, &oitv);
  814         if (error != 0 || uap->oitv == NULL)
  815                 return (error);
  816         return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
  817 }
  818 
  819 int
  820 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
  821     struct itimerval *oitv)
  822 {
  823         struct proc *p = td->td_proc;
  824         struct timeval ctv;
  825         sbintime_t sbt, pr;
  826 
  827         if (aitv == NULL)
  828                 return (kern_getitimer(td, which, oitv));
  829 
  830         if (which > ITIMER_PROF)
  831                 return (EINVAL);
  832 #ifdef KTRACE
  833         if (KTRPOINT(td, KTR_STRUCT))
  834                 ktritimerval(aitv);
  835 #endif
  836         if (itimerfix(&aitv->it_value) ||
  837             aitv->it_value.tv_sec > INT32_MAX / 2)
  838                 return (EINVAL);
  839         if (!timevalisset(&aitv->it_value))
  840                 timevalclear(&aitv->it_interval);
  841         else if (itimerfix(&aitv->it_interval) ||
  842             aitv->it_interval.tv_sec > INT32_MAX / 2)
  843                 return (EINVAL);
  844 
  845         if (which == ITIMER_REAL) {
  846                 PROC_LOCK(p);
  847                 if (timevalisset(&p->p_realtimer.it_value))
  848                         callout_stop(&p->p_itcallout);
  849                 microuptime(&ctv);
  850                 if (timevalisset(&aitv->it_value)) {
  851                         pr = tvtosbt(aitv->it_value) >> tc_precexp;
  852                         timevaladd(&aitv->it_value, &ctv);
  853                         sbt = tvtosbt(aitv->it_value);
  854                         callout_reset_sbt(&p->p_itcallout, sbt, pr,
  855                             realitexpire, p, C_ABSOLUTE);
  856                 }
  857                 *oitv = p->p_realtimer;
  858                 p->p_realtimer = *aitv;
  859                 PROC_UNLOCK(p);
  860                 if (timevalisset(&oitv->it_value)) {
  861                         if (timevalcmp(&oitv->it_value, &ctv, <))
  862                                 timevalclear(&oitv->it_value);
  863                         else
  864                                 timevalsub(&oitv->it_value, &ctv);
  865                 }
  866         } else {
  867                 if (aitv->it_interval.tv_sec == 0 &&
  868                     aitv->it_interval.tv_usec != 0 &&
  869                     aitv->it_interval.tv_usec < tick)
  870                         aitv->it_interval.tv_usec = tick;
  871                 if (aitv->it_value.tv_sec == 0 &&
  872                     aitv->it_value.tv_usec != 0 &&
  873                     aitv->it_value.tv_usec < tick)
  874                         aitv->it_value.tv_usec = tick;
  875                 PROC_ITIMLOCK(p);
  876                 *oitv = p->p_stats->p_timer[which];
  877                 p->p_stats->p_timer[which] = *aitv;
  878                 PROC_ITIMUNLOCK(p);
  879         }
  880 #ifdef KTRACE
  881         if (KTRPOINT(td, KTR_STRUCT))
  882                 ktritimerval(oitv);
  883 #endif
  884         return (0);
  885 }
  886 
  887 static void
  888 realitexpire_reset_callout(struct proc *p, sbintime_t *isbtp)
  889 {
  890         sbintime_t prec;
  891 
  892         prec = isbtp == NULL ? tvtosbt(p->p_realtimer.it_interval) : *isbtp;
  893         callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
  894             prec >> tc_precexp, realitexpire, p, C_ABSOLUTE);
  895 }
  896 
  897 void
  898 itimer_proc_continue(struct proc *p)
  899 {
  900         struct timeval ctv;
  901         struct itimer *it;
  902         int id;
  903 
  904         PROC_LOCK_ASSERT(p, MA_OWNED);
  905 
  906         if ((p->p_flag2 & P2_ITSTOPPED) != 0) {
  907                 p->p_flag2 &= ~P2_ITSTOPPED;
  908                 microuptime(&ctv);
  909                 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >=))
  910                         realitexpire(p);
  911                 else
  912                         realitexpire_reset_callout(p, NULL);
  913         }
  914 
  915         if (p->p_itimers != NULL) {
  916                 for (id = 3; id < TIMER_MAX; id++) {
  917                         it = p->p_itimers->its_timers[id];
  918                         if (it == NULL)
  919                                 continue;
  920                         if ((it->it_flags & ITF_PSTOPPED) != 0) {
  921                                 ITIMER_LOCK(it);
  922                                 if ((it->it_flags & ITF_PSTOPPED) != 0) {
  923                                         it->it_flags &= ~ITF_PSTOPPED;
  924                                         if ((it->it_flags & ITF_DELETING) == 0)
  925                                                 realtimer_expire_l(it, true);
  926                                 }
  927                                 ITIMER_UNLOCK(it);
  928                         }
  929                 }
  930         }
  931 }
  932 
  933 /*
  934  * Real interval timer expired:
  935  * send process whose timer expired an alarm signal.
  936  * If time is not set up to reload, then just return.
  937  * Else compute next time timer should go off which is > current time.
  938  * This is where delay in processing this timeout causes multiple
  939  * SIGALRM calls to be compressed into one.
  940  * tvtohz() always adds 1 to allow for the time until the next clock
  941  * interrupt being strictly less than 1 clock tick, but we don't want
  942  * that here since we want to appear to be in sync with the clock
  943  * interrupt even when we're delayed.
  944  */
  945 static void
  946 realitexpire(void *arg)
  947 {
  948         struct proc *p;
  949         struct timeval ctv;
  950         sbintime_t isbt;
  951 
  952         p = (struct proc *)arg;
  953         kern_psignal(p, SIGALRM);
  954         if (!timevalisset(&p->p_realtimer.it_interval)) {
  955                 timevalclear(&p->p_realtimer.it_value);
  956                 return;
  957         }
  958 
  959         isbt = tvtosbt(p->p_realtimer.it_interval);
  960         if (isbt >= sbt_timethreshold)
  961                 getmicrouptime(&ctv);
  962         else
  963                 microuptime(&ctv);
  964         do {
  965                 timevaladd(&p->p_realtimer.it_value,
  966                     &p->p_realtimer.it_interval);
  967         } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
  968 
  969         if (P_SHOULDSTOP(p) || P_KILLED(p)) {
  970                 p->p_flag2 |= P2_ITSTOPPED;
  971                 return;
  972         }
  973 
  974         p->p_flag2 &= ~P2_ITSTOPPED;
  975         realitexpire_reset_callout(p, &isbt);
  976 }
  977 
  978 /*
  979  * Check that a proposed value to load into the .it_value or
  980  * .it_interval part of an interval timer is acceptable, and
  981  * fix it to have at least minimal value (i.e. if it is less
  982  * than the resolution of the clock, round it up.)
  983  */
  984 int
  985 itimerfix(struct timeval *tv)
  986 {
  987 
  988         if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
  989                 return (EINVAL);
  990         if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
  991             tv->tv_usec < (u_int)tick / 16)
  992                 tv->tv_usec = (u_int)tick / 16;
  993         return (0);
  994 }
  995 
  996 /*
  997  * Decrement an interval timer by a specified number
  998  * of microseconds, which must be less than a second,
  999  * i.e. < 1000000.  If the timer expires, then reload
 1000  * it.  In this case, carry over (usec - old value) to
 1001  * reduce the value reloaded into the timer so that
 1002  * the timer does not drift.  This routine assumes
 1003  * that it is called in a context where the timers
 1004  * on which it is operating cannot change in value.
 1005  */
 1006 int
 1007 itimerdecr(struct itimerval *itp, int usec)
 1008 {
 1009 
 1010         if (itp->it_value.tv_usec < usec) {
 1011                 if (itp->it_value.tv_sec == 0) {
 1012                         /* expired, and already in next interval */
 1013                         usec -= itp->it_value.tv_usec;
 1014                         goto expire;
 1015                 }
 1016                 itp->it_value.tv_usec += 1000000;
 1017                 itp->it_value.tv_sec--;
 1018         }
 1019         itp->it_value.tv_usec -= usec;
 1020         usec = 0;
 1021         if (timevalisset(&itp->it_value))
 1022                 return (1);
 1023         /* expired, exactly at end of interval */
 1024 expire:
 1025         if (timevalisset(&itp->it_interval)) {
 1026                 itp->it_value = itp->it_interval;
 1027                 itp->it_value.tv_usec -= usec;
 1028                 if (itp->it_value.tv_usec < 0) {
 1029                         itp->it_value.tv_usec += 1000000;
 1030                         itp->it_value.tv_sec--;
 1031                 }
 1032         } else
 1033                 itp->it_value.tv_usec = 0;              /* sec is already 0 */
 1034         return (0);
 1035 }
 1036 
 1037 /*
 1038  * Add and subtract routines for timevals.
 1039  * N.B.: subtract routine doesn't deal with
 1040  * results which are before the beginning,
 1041  * it just gets very confused in this case.
 1042  * Caveat emptor.
 1043  */
 1044 void
 1045 timevaladd(struct timeval *t1, const struct timeval *t2)
 1046 {
 1047 
 1048         t1->tv_sec += t2->tv_sec;
 1049         t1->tv_usec += t2->tv_usec;
 1050         timevalfix(t1);
 1051 }
 1052 
 1053 void
 1054 timevalsub(struct timeval *t1, const struct timeval *t2)
 1055 {
 1056 
 1057         t1->tv_sec -= t2->tv_sec;
 1058         t1->tv_usec -= t2->tv_usec;
 1059         timevalfix(t1);
 1060 }
 1061 
 1062 static void
 1063 timevalfix(struct timeval *t1)
 1064 {
 1065 
 1066         if (t1->tv_usec < 0) {
 1067                 t1->tv_sec--;
 1068                 t1->tv_usec += 1000000;
 1069         }
 1070         if (t1->tv_usec >= 1000000) {
 1071                 t1->tv_sec++;
 1072                 t1->tv_usec -= 1000000;
 1073         }
 1074 }
 1075 
 1076 /*
 1077  * ratecheck(): simple time-based rate-limit checking.
 1078  */
 1079 int
 1080 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
 1081 {
 1082         struct timeval tv, delta;
 1083         int rv = 0;
 1084 
 1085         getmicrouptime(&tv);            /* NB: 10ms precision */
 1086         delta = tv;
 1087         timevalsub(&delta, lasttime);
 1088 
 1089         /*
 1090          * check for 0,0 is so that the message will be seen at least once,
 1091          * even if interval is huge.
 1092          */
 1093         if (timevalcmp(&delta, mininterval, >=) ||
 1094             (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
 1095                 *lasttime = tv;
 1096                 rv = 1;
 1097         }
 1098 
 1099         return (rv);
 1100 }
 1101 
 1102 /*
 1103  * ppsratecheck(): packets (or events) per second limitation.
 1104  *
 1105  * Return 0 if the limit is to be enforced (e.g. the caller
 1106  * should drop a packet because of the rate limitation).
 1107  *
 1108  * maxpps of 0 always causes zero to be returned.  maxpps of -1
 1109  * always causes 1 to be returned; this effectively defeats rate
 1110  * limiting.
 1111  *
 1112  * Note that we maintain the struct timeval for compatibility
 1113  * with other bsd systems.  We reuse the storage and just monitor
 1114  * clock ticks for minimal overhead.  
 1115  */
 1116 int
 1117 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
 1118 {
 1119         int now;
 1120 
 1121         /*
 1122          * Reset the last time and counter if this is the first call
 1123          * or more than a second has passed since the last update of
 1124          * lasttime.
 1125          */
 1126         now = ticks;
 1127         if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
 1128                 lasttime->tv_sec = now;
 1129                 *curpps = 1;
 1130                 return (maxpps != 0);
 1131         } else {
 1132                 (*curpps)++;            /* NB: ignore potential overflow */
 1133                 return (maxpps < 0 || *curpps <= maxpps);
 1134         }
 1135 }
 1136 
 1137 static void
 1138 itimer_start(void)
 1139 {
 1140         static const struct kclock rt_clock = {
 1141                 .timer_create  = realtimer_create,
 1142                 .timer_delete  = realtimer_delete,
 1143                 .timer_settime = realtimer_settime,
 1144                 .timer_gettime = realtimer_gettime,
 1145         };
 1146 
 1147         itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
 1148                 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
 1149         register_posix_clock(CLOCK_REALTIME,  &rt_clock);
 1150         register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
 1151         p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
 1152         p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
 1153         p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
 1154 }
 1155 
 1156 static int
 1157 register_posix_clock(int clockid, const struct kclock *clk)
 1158 {
 1159         if ((unsigned)clockid >= MAX_CLOCKS) {
 1160                 printf("%s: invalid clockid\n", __func__);
 1161                 return (0);
 1162         }
 1163         posix_clocks[clockid] = *clk;
 1164         return (1);
 1165 }
 1166 
 1167 static int
 1168 itimer_init(void *mem, int size, int flags)
 1169 {
 1170         struct itimer *it;
 1171 
 1172         it = (struct itimer *)mem;
 1173         mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
 1174         return (0);
 1175 }
 1176 
 1177 static void
 1178 itimer_fini(void *mem, int size)
 1179 {
 1180         struct itimer *it;
 1181 
 1182         it = (struct itimer *)mem;
 1183         mtx_destroy(&it->it_mtx);
 1184 }
 1185 
 1186 static void
 1187 itimer_enter(struct itimer *it)
 1188 {
 1189 
 1190         mtx_assert(&it->it_mtx, MA_OWNED);
 1191         it->it_usecount++;
 1192 }
 1193 
 1194 static void
 1195 itimer_leave(struct itimer *it)
 1196 {
 1197 
 1198         mtx_assert(&it->it_mtx, MA_OWNED);
 1199         KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
 1200 
 1201         if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
 1202                 wakeup(it);
 1203 }
 1204 
 1205 #ifndef _SYS_SYSPROTO_H_
 1206 struct ktimer_create_args {
 1207         clockid_t clock_id;
 1208         struct sigevent * evp;
 1209         int * timerid;
 1210 };
 1211 #endif
 1212 int
 1213 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
 1214 {
 1215         struct sigevent *evp, ev;
 1216         int id;
 1217         int error;
 1218 
 1219         if (uap->evp == NULL) {
 1220                 evp = NULL;
 1221         } else {
 1222                 error = copyin(uap->evp, &ev, sizeof(ev));
 1223                 if (error != 0)
 1224                         return (error);
 1225                 evp = &ev;
 1226         }
 1227         error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
 1228         if (error == 0) {
 1229                 error = copyout(&id, uap->timerid, sizeof(int));
 1230                 if (error != 0)
 1231                         kern_ktimer_delete(td, id);
 1232         }
 1233         return (error);
 1234 }
 1235 
 1236 int
 1237 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
 1238     int *timerid, int preset_id)
 1239 {
 1240         struct proc *p = td->td_proc;
 1241         struct itimer *it;
 1242         int id;
 1243         int error;
 1244 
 1245         if (clock_id < 0 || clock_id >= MAX_CLOCKS)
 1246                 return (EINVAL);
 1247 
 1248         if (posix_clocks[clock_id].timer_create == NULL)
 1249                 return (EINVAL);
 1250 
 1251         if (evp != NULL) {
 1252                 if (evp->sigev_notify != SIGEV_NONE &&
 1253                     evp->sigev_notify != SIGEV_SIGNAL &&
 1254                     evp->sigev_notify != SIGEV_THREAD_ID)
 1255                         return (EINVAL);
 1256                 if ((evp->sigev_notify == SIGEV_SIGNAL ||
 1257                      evp->sigev_notify == SIGEV_THREAD_ID) &&
 1258                         !_SIG_VALID(evp->sigev_signo))
 1259                         return (EINVAL);
 1260         }
 1261 
 1262         if (p->p_itimers == NULL)
 1263                 itimers_alloc(p);
 1264 
 1265         it = uma_zalloc(itimer_zone, M_WAITOK);
 1266         it->it_flags = 0;
 1267         it->it_usecount = 0;
 1268         timespecclear(&it->it_time.it_value);
 1269         timespecclear(&it->it_time.it_interval);
 1270         it->it_overrun = 0;
 1271         it->it_overrun_last = 0;
 1272         it->it_clockid = clock_id;
 1273         it->it_proc = p;
 1274         ksiginfo_init(&it->it_ksi);
 1275         it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
 1276         error = CLOCK_CALL(clock_id, timer_create, (it));
 1277         if (error != 0)
 1278                 goto out;
 1279 
 1280         PROC_LOCK(p);
 1281         if (preset_id != -1) {
 1282                 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
 1283                 id = preset_id;
 1284                 if (p->p_itimers->its_timers[id] != NULL) {
 1285                         PROC_UNLOCK(p);
 1286                         error = 0;
 1287                         goto out;
 1288                 }
 1289         } else {
 1290                 /*
 1291                  * Find a free timer slot, skipping those reserved
 1292                  * for setitimer().
 1293                  */
 1294                 for (id = 3; id < TIMER_MAX; id++)
 1295                         if (p->p_itimers->its_timers[id] == NULL)
 1296                                 break;
 1297                 if (id == TIMER_MAX) {
 1298                         PROC_UNLOCK(p);
 1299                         error = EAGAIN;
 1300                         goto out;
 1301                 }
 1302         }
 1303         p->p_itimers->its_timers[id] = it;
 1304         if (evp != NULL)
 1305                 it->it_sigev = *evp;
 1306         else {
 1307                 it->it_sigev.sigev_notify = SIGEV_SIGNAL;
 1308                 switch (clock_id) {
 1309                 default:
 1310                 case CLOCK_REALTIME:
 1311                         it->it_sigev.sigev_signo = SIGALRM;
 1312                         break;
 1313                 case CLOCK_VIRTUAL:
 1314                         it->it_sigev.sigev_signo = SIGVTALRM;
 1315                         break;
 1316                 case CLOCK_PROF:
 1317                         it->it_sigev.sigev_signo = SIGPROF;
 1318                         break;
 1319                 }
 1320                 it->it_sigev.sigev_value.sival_int = id;
 1321         }
 1322 
 1323         if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
 1324             it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
 1325                 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
 1326                 it->it_ksi.ksi_code = SI_TIMER;
 1327                 it->it_ksi.ksi_value = it->it_sigev.sigev_value;
 1328                 it->it_ksi.ksi_timerid = id;
 1329         }
 1330         PROC_UNLOCK(p);
 1331         *timerid = id;
 1332         return (0);
 1333 
 1334 out:
 1335         ITIMER_LOCK(it);
 1336         CLOCK_CALL(it->it_clockid, timer_delete, (it));
 1337         ITIMER_UNLOCK(it);
 1338         uma_zfree(itimer_zone, it);
 1339         return (error);
 1340 }
 1341 
 1342 #ifndef _SYS_SYSPROTO_H_
 1343 struct ktimer_delete_args {
 1344         int timerid;
 1345 };
 1346 #endif
 1347 int
 1348 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
 1349 {
 1350 
 1351         return (kern_ktimer_delete(td, uap->timerid));
 1352 }
 1353 
 1354 static struct itimer *
 1355 itimer_find(struct proc *p, int timerid)
 1356 {
 1357         struct itimer *it;
 1358 
 1359         PROC_LOCK_ASSERT(p, MA_OWNED);
 1360         if ((p->p_itimers == NULL) ||
 1361             (timerid < 0) || (timerid >= TIMER_MAX) ||
 1362             (it = p->p_itimers->its_timers[timerid]) == NULL) {
 1363                 return (NULL);
 1364         }
 1365         ITIMER_LOCK(it);
 1366         if ((it->it_flags & ITF_DELETING) != 0) {
 1367                 ITIMER_UNLOCK(it);
 1368                 it = NULL;
 1369         }
 1370         return (it);
 1371 }
 1372 
 1373 int
 1374 kern_ktimer_delete(struct thread *td, int timerid)
 1375 {
 1376         struct proc *p = td->td_proc;
 1377         struct itimer *it;
 1378 
 1379         PROC_LOCK(p);
 1380         it = itimer_find(p, timerid);
 1381         if (it == NULL) {
 1382                 PROC_UNLOCK(p);
 1383                 return (EINVAL);
 1384         }
 1385         PROC_UNLOCK(p);
 1386 
 1387         it->it_flags |= ITF_DELETING;
 1388         while (it->it_usecount > 0) {
 1389                 it->it_flags |= ITF_WANTED;
 1390                 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
 1391         }
 1392         it->it_flags &= ~ITF_WANTED;
 1393         CLOCK_CALL(it->it_clockid, timer_delete, (it));
 1394         ITIMER_UNLOCK(it);
 1395 
 1396         PROC_LOCK(p);
 1397         if (KSI_ONQ(&it->it_ksi))
 1398                 sigqueue_take(&it->it_ksi);
 1399         p->p_itimers->its_timers[timerid] = NULL;
 1400         PROC_UNLOCK(p);
 1401         uma_zfree(itimer_zone, it);
 1402         return (0);
 1403 }
 1404 
 1405 #ifndef _SYS_SYSPROTO_H_
 1406 struct ktimer_settime_args {
 1407         int timerid;
 1408         int flags;
 1409         const struct itimerspec * value;
 1410         struct itimerspec * ovalue;
 1411 };
 1412 #endif
 1413 int
 1414 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
 1415 {
 1416         struct itimerspec val, oval, *ovalp;
 1417         int error;
 1418 
 1419         error = copyin(uap->value, &val, sizeof(val));
 1420         if (error != 0)
 1421                 return (error);
 1422         ovalp = uap->ovalue != NULL ? &oval : NULL;
 1423         error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
 1424         if (error == 0 && uap->ovalue != NULL)
 1425                 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
 1426         return (error);
 1427 }
 1428 
 1429 int
 1430 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
 1431     struct itimerspec *val, struct itimerspec *oval)
 1432 {
 1433         struct proc *p;
 1434         struct itimer *it;
 1435         int error;
 1436 
 1437         p = td->td_proc;
 1438         PROC_LOCK(p);
 1439         if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
 1440                 PROC_UNLOCK(p);
 1441                 error = EINVAL;
 1442         } else {
 1443                 PROC_UNLOCK(p);
 1444                 itimer_enter(it);
 1445                 error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
 1446                     flags, val, oval));
 1447                 itimer_leave(it);
 1448                 ITIMER_UNLOCK(it);
 1449         }
 1450         return (error);
 1451 }
 1452 
 1453 #ifndef _SYS_SYSPROTO_H_
 1454 struct ktimer_gettime_args {
 1455         int timerid;
 1456         struct itimerspec * value;
 1457 };
 1458 #endif
 1459 int
 1460 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
 1461 {
 1462         struct itimerspec val;
 1463         int error;
 1464 
 1465         error = kern_ktimer_gettime(td, uap->timerid, &val);
 1466         if (error == 0)
 1467                 error = copyout(&val, uap->value, sizeof(val));
 1468         return (error);
 1469 }
 1470 
 1471 int
 1472 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
 1473 {
 1474         struct proc *p;
 1475         struct itimer *it;
 1476         int error;
 1477 
 1478         p = td->td_proc;
 1479         PROC_LOCK(p);
 1480         if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
 1481                 PROC_UNLOCK(p);
 1482                 error = EINVAL;
 1483         } else {
 1484                 PROC_UNLOCK(p);
 1485                 itimer_enter(it);
 1486                 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
 1487                 itimer_leave(it);
 1488                 ITIMER_UNLOCK(it);
 1489         }
 1490         return (error);
 1491 }
 1492 
 1493 #ifndef _SYS_SYSPROTO_H_
 1494 struct timer_getoverrun_args {
 1495         int timerid;
 1496 };
 1497 #endif
 1498 int
 1499 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
 1500 {
 1501 
 1502         return (kern_ktimer_getoverrun(td, uap->timerid));
 1503 }
 1504 
 1505 int
 1506 kern_ktimer_getoverrun(struct thread *td, int timer_id)
 1507 {
 1508         struct proc *p = td->td_proc;
 1509         struct itimer *it;
 1510         int error ;
 1511 
 1512         PROC_LOCK(p);
 1513         if (timer_id < 3 ||
 1514             (it = itimer_find(p, timer_id)) == NULL) {
 1515                 PROC_UNLOCK(p);
 1516                 error = EINVAL;
 1517         } else {
 1518                 td->td_retval[0] = it->it_overrun_last;
 1519                 ITIMER_UNLOCK(it);
 1520                 PROC_UNLOCK(p);
 1521                 error = 0;
 1522         }
 1523         return (error);
 1524 }
 1525 
 1526 static int
 1527 realtimer_create(struct itimer *it)
 1528 {
 1529         callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
 1530         return (0);
 1531 }
 1532 
 1533 static int
 1534 realtimer_delete(struct itimer *it)
 1535 {
 1536         mtx_assert(&it->it_mtx, MA_OWNED);
 1537 
 1538         /*
 1539          * clear timer's value and interval to tell realtimer_expire
 1540          * to not rearm the timer.
 1541          */
 1542         timespecclear(&it->it_time.it_value);
 1543         timespecclear(&it->it_time.it_interval);
 1544         ITIMER_UNLOCK(it);
 1545         callout_drain(&it->it_callout);
 1546         ITIMER_LOCK(it);
 1547         return (0);
 1548 }
 1549 
 1550 static int
 1551 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
 1552 {
 1553         struct timespec cts;
 1554 
 1555         mtx_assert(&it->it_mtx, MA_OWNED);
 1556 
 1557         realtimer_clocktime(it->it_clockid, &cts);
 1558         *ovalue = it->it_time;
 1559         if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
 1560                 timespecsub(&ovalue->it_value, &cts, &ovalue->it_value);
 1561                 if (ovalue->it_value.tv_sec < 0 ||
 1562                     (ovalue->it_value.tv_sec == 0 &&
 1563                      ovalue->it_value.tv_nsec == 0)) {
 1564                         ovalue->it_value.tv_sec  = 0;
 1565                         ovalue->it_value.tv_nsec = 1;
 1566                 }
 1567         }
 1568         return (0);
 1569 }
 1570 
 1571 static int
 1572 realtimer_settime(struct itimer *it, int flags, struct itimerspec *value,
 1573     struct itimerspec *ovalue)
 1574 {
 1575         struct timespec cts, ts;
 1576         struct timeval tv;
 1577         struct itimerspec val;
 1578 
 1579         mtx_assert(&it->it_mtx, MA_OWNED);
 1580 
 1581         val = *value;
 1582         if (itimespecfix(&val.it_value))
 1583                 return (EINVAL);
 1584 
 1585         if (timespecisset(&val.it_value)) {
 1586                 if (itimespecfix(&val.it_interval))
 1587                         return (EINVAL);
 1588         } else {
 1589                 timespecclear(&val.it_interval);
 1590         }
 1591 
 1592         if (ovalue != NULL)
 1593                 realtimer_gettime(it, ovalue);
 1594 
 1595         it->it_time = val;
 1596         if (timespecisset(&val.it_value)) {
 1597                 realtimer_clocktime(it->it_clockid, &cts);
 1598                 ts = val.it_value;
 1599                 if ((flags & TIMER_ABSTIME) == 0) {
 1600                         /* Convert to absolute time. */
 1601                         timespecadd(&it->it_time.it_value, &cts,
 1602                             &it->it_time.it_value);
 1603                 } else {
 1604                         timespecsub(&ts, &cts, &ts);
 1605                         /*
 1606                          * We don't care if ts is negative, tztohz will
 1607                          * fix it.
 1608                          */
 1609                 }
 1610                 TIMESPEC_TO_TIMEVAL(&tv, &ts);
 1611                 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
 1612                     it);
 1613         } else {
 1614                 callout_stop(&it->it_callout);
 1615         }
 1616 
 1617         return (0);
 1618 }
 1619 
 1620 static void
 1621 realtimer_clocktime(clockid_t id, struct timespec *ts)
 1622 {
 1623         if (id == CLOCK_REALTIME)
 1624                 getnanotime(ts);
 1625         else    /* CLOCK_MONOTONIC */
 1626                 getnanouptime(ts);
 1627 }
 1628 
 1629 int
 1630 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
 1631 {
 1632         struct itimer *it;
 1633 
 1634         PROC_LOCK_ASSERT(p, MA_OWNED);
 1635         it = itimer_find(p, timerid);
 1636         if (it != NULL) {
 1637                 ksi->ksi_overrun = it->it_overrun;
 1638                 it->it_overrun_last = it->it_overrun;
 1639                 it->it_overrun = 0;
 1640                 ITIMER_UNLOCK(it);
 1641                 return (0);
 1642         }
 1643         return (EINVAL);
 1644 }
 1645 
 1646 static int
 1647 itimespecfix(struct timespec *ts)
 1648 {
 1649 
 1650         if (!timespecvalid_interval(ts))
 1651                 return (EINVAL);
 1652         if ((UINT64_MAX - ts->tv_nsec) / NS_PER_SEC < ts->tv_sec)
 1653                 return (EINVAL);
 1654         if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
 1655                 ts->tv_nsec = tick * 1000;
 1656         return (0);
 1657 }
 1658 
 1659 #define timespectons(tsp)                       \
 1660         ((uint64_t)(tsp)->tv_sec * NS_PER_SEC + (tsp)->tv_nsec)
 1661 #define timespecfromns(ns) (struct timespec){   \
 1662         .tv_sec = (ns) / NS_PER_SEC,            \
 1663         .tv_nsec = (ns) % NS_PER_SEC            \
 1664 }
 1665 
 1666 static void
 1667 realtimer_expire_l(struct itimer *it, bool proc_locked)
 1668 {
 1669         struct timespec cts, ts;
 1670         struct timeval tv;
 1671         struct proc *p;
 1672         uint64_t interval, now, overruns, value;
 1673 
 1674         realtimer_clocktime(it->it_clockid, &cts);
 1675         /* Only fire if time is reached. */
 1676         if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
 1677                 if (timespecisset(&it->it_time.it_interval)) {
 1678                         timespecadd(&it->it_time.it_value,
 1679                             &it->it_time.it_interval,
 1680                             &it->it_time.it_value);
 1681 
 1682                         interval = timespectons(&it->it_time.it_interval);
 1683                         value = timespectons(&it->it_time.it_value);
 1684                         now = timespectons(&cts);
 1685 
 1686                         if (now >= value) {
 1687                                 /*
 1688                                  * We missed at least one period.
 1689                                  */
 1690                                 overruns = howmany(now - value + 1, interval);
 1691                                 if (it->it_overrun + overruns >=
 1692                                     it->it_overrun &&
 1693                                     it->it_overrun + overruns <= INT_MAX) {
 1694                                         it->it_overrun += (int)overruns;
 1695                                 } else {
 1696                                         it->it_overrun = INT_MAX;
 1697                                         it->it_ksi.ksi_errno = ERANGE;
 1698                                 }
 1699                                 value =
 1700                                     now + interval - (now - value) % interval;
 1701                                 it->it_time.it_value = timespecfromns(value);
 1702                         }
 1703                 } else {
 1704                         /* single shot timer ? */
 1705                         timespecclear(&it->it_time.it_value);
 1706                 }
 1707 
 1708                 p = it->it_proc;
 1709                 if (timespecisset(&it->it_time.it_value)) {
 1710                         if (P_SHOULDSTOP(p) || P_KILLED(p)) {
 1711                                 it->it_flags |= ITF_PSTOPPED;
 1712                         } else {
 1713                                 timespecsub(&it->it_time.it_value, &cts, &ts);
 1714                                 TIMESPEC_TO_TIMEVAL(&tv, &ts);
 1715                                 callout_reset(&it->it_callout, tvtohz(&tv),
 1716                                     realtimer_expire, it);
 1717                         }
 1718                 }
 1719 
 1720                 itimer_enter(it);
 1721                 ITIMER_UNLOCK(it);
 1722                 if (proc_locked)
 1723                         PROC_UNLOCK(p);
 1724                 itimer_fire(it);
 1725                 if (proc_locked)
 1726                         PROC_LOCK(p);
 1727                 ITIMER_LOCK(it);
 1728                 itimer_leave(it);
 1729         } else if (timespecisset(&it->it_time.it_value)) {
 1730                 p = it->it_proc;
 1731                 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
 1732                         it->it_flags |= ITF_PSTOPPED;
 1733                 } else {
 1734                         ts = it->it_time.it_value;
 1735                         timespecsub(&ts, &cts, &ts);
 1736                         TIMESPEC_TO_TIMEVAL(&tv, &ts);
 1737                         callout_reset(&it->it_callout, tvtohz(&tv),
 1738                             realtimer_expire, it);
 1739                 }
 1740         }
 1741 }
 1742 
 1743 /* Timeout callback for realtime timer */
 1744 static void
 1745 realtimer_expire(void *arg)
 1746 {
 1747         realtimer_expire_l(arg, false);
 1748 }
 1749 
 1750 static void
 1751 itimer_fire(struct itimer *it)
 1752 {
 1753         struct proc *p = it->it_proc;
 1754         struct thread *td;
 1755 
 1756         if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
 1757             it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
 1758                 if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
 1759                         ITIMER_LOCK(it);
 1760                         timespecclear(&it->it_time.it_value);
 1761                         timespecclear(&it->it_time.it_interval);
 1762                         callout_stop(&it->it_callout);
 1763                         ITIMER_UNLOCK(it);
 1764                         return;
 1765                 }
 1766                 if (!KSI_ONQ(&it->it_ksi)) {
 1767                         it->it_ksi.ksi_errno = 0;
 1768                         ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
 1769                         tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
 1770                 } else {
 1771                         if (it->it_overrun < INT_MAX)
 1772                                 it->it_overrun++;
 1773                         else
 1774                                 it->it_ksi.ksi_errno = ERANGE;
 1775                 }
 1776                 PROC_UNLOCK(p);
 1777         }
 1778 }
 1779 
 1780 static void
 1781 itimers_alloc(struct proc *p)
 1782 {
 1783         struct itimers *its;
 1784 
 1785         its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
 1786         PROC_LOCK(p);
 1787         if (p->p_itimers == NULL) {
 1788                 p->p_itimers = its;
 1789                 PROC_UNLOCK(p);
 1790         }
 1791         else {
 1792                 PROC_UNLOCK(p);
 1793                 free(its, M_SUBPROC);
 1794         }
 1795 }
 1796 
 1797 /* Clean up timers when some process events are being triggered. */
 1798 static void
 1799 itimers_event_exit_exec(int start_idx, struct proc *p)
 1800 {
 1801         struct itimers *its;
 1802         struct itimer *it;
 1803         int i;
 1804 
 1805         its = p->p_itimers;
 1806         if (its == NULL)
 1807                 return;
 1808 
 1809         for (i = start_idx; i < TIMER_MAX; ++i) {
 1810                 if ((it = its->its_timers[i]) != NULL)
 1811                         kern_ktimer_delete(curthread, i);
 1812         }
 1813         if (its->its_timers[0] == NULL && its->its_timers[1] == NULL &&
 1814             its->its_timers[2] == NULL) {
 1815                 /* Synchronize with itimer_proc_continue(). */
 1816                 PROC_LOCK(p);
 1817                 p->p_itimers = NULL;
 1818                 PROC_UNLOCK(p);
 1819                 free(its, M_SUBPROC);
 1820         }
 1821 }
 1822 
 1823 void
 1824 itimers_exec(struct proc *p)
 1825 {
 1826         /*
 1827          * According to susv3, XSI interval timers should be inherited
 1828          * by new image.
 1829          */
 1830         itimers_event_exit_exec(3, p);
 1831 }
 1832 
 1833 void
 1834 itimers_exit(struct proc *p)
 1835 {
 1836         itimers_event_exit_exec(0, p);
 1837 }

Cache object: 9baaceab38dc41da821f2e10a397f952


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.