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  * 3. 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  * $FreeBSD: src/sys/kern/kern_time.c,v 1.68.2.1 2002/10/01 08:00:41 bde Exp $
  31  */
  32
  33 #include <sys/param.h>
  34 #include <sys/systm.h>
  35 #include <sys/buf.h>
  36 #include <sys/sysproto.h>
  37 #include <sys/resourcevar.h>
  38 #include <sys/signalvar.h>
  39 #include <sys/kernel.h>
  40 #include <sys/sysent.h>
  41 #include <sys/sysunion.h>
  42 #include <sys/proc.h>
  43 #include <sys/priv.h>
  44 #include <sys/time.h>
  45 #include <sys/vnode.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/kern_syscall.h>
  48 #include <vm/vm.h>
  49 #include <vm/vm_extern.h>
  50
  51 #include <sys/msgport2.h>
  52 #include <sys/spinlock2.h>
  53 #include <sys/thread2.h>
  54
  55 extern struct spinlock ntp_spin;
  56
  57 struct timezone tz;
  58
  59 /*
  60  * Time of day and interval timer support.
  61  *
  62  * These routines provide the kernel entry points to get and set
  63  * the time-of-day and per-process interval timers.  Subroutines
  64  * here provide support for adding and subtracting timeval structures
  65  * and decrementing interval timers, optionally reloading the interval
  66  * timers when they expire.
  67  */
  68
  69 static int      settime(struct timeval *);
  70 static void     timevalfix(struct timeval *);
  71 static void     realitexpire(void *arg);
  72
  73 /*
  74  * Nanosleep tries very hard to sleep for a precisely requested time
  75  * interval, down to 1uS.  The administrator can impose a minimum delay
  76  * and a delay below which we hard-loop instead of initiate a timer
  77  * interrupt and sleep.
  78  *
  79  * For machines under high loads it might be beneficial to increase min_us
  80  * to e.g. 1000uS (1ms) so spining processes sleep meaningfully.
  81  */
  82 static int     nanosleep_min_us = 10;
  83 static int     nanosleep_hard_us = 100;
  84 static int     gettimeofday_quick = 0;
  85 SYSCTL_INT(_kern, OID_AUTO, nanosleep_min_us, CTLFLAG_RW,
  86            &nanosleep_min_us, 0, "");
  87 SYSCTL_INT(_kern, OID_AUTO, nanosleep_hard_us, CTLFLAG_RW,
  88            &nanosleep_hard_us, 0, "");
  89 SYSCTL_INT(_kern, OID_AUTO, gettimeofday_quick, CTLFLAG_RW,
  90            &gettimeofday_quick, 0, "");
  91
  92 static struct lock masterclock_lock = LOCK_INITIALIZER("mstrclk", 0, 0);
  93
  94 static int
  95 settime(struct timeval *tv)
  96 {
  97         struct timeval delta, tv1, tv2;
  98         static struct timeval maxtime, laststep;
  99         struct timespec ts;
 100         int origcpu;
 101
 102         if ((origcpu = mycpu->gd_cpuid) != 0)
 103                 lwkt_setcpu_self(globaldata_find(0));
 104
 105         crit_enter();
 106         microtime(&tv1);
 107         delta = *tv;
 108         timevalsub(&delta, &tv1);
 109
 110         /*
 111          * If the system is secure, we do not allow the time to be
 112          * set to a value earlier than 1 second less than the highest
 113          * time we have yet seen. The worst a miscreant can do in
 114          * this circumstance is "freeze" time. He couldn't go
 115          * back to the past.
 116          *
 117          * We similarly do not allow the clock to be stepped more
 118          * than one second, nor more than once per second. This allows
 119          * a miscreant to make the clock march double-time, but no worse.
 120          */
 121         if (securelevel > 1) {
 122                 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
 123                         /*
 124                          * Update maxtime to latest time we've seen.
 125                          */
 126                         if (tv1.tv_sec > maxtime.tv_sec)
 127                                 maxtime = tv1;
 128                         tv2 = *tv;
 129                         timevalsub(&tv2, &maxtime);
 130                         if (tv2.tv_sec < -1) {
 131                                 tv->tv_sec = maxtime.tv_sec - 1;
 132                                 kprintf("Time adjustment clamped to -1 second\n");
 133                         }
 134                 } else {
 135                         if (tv1.tv_sec == laststep.tv_sec) {
 136                                 crit_exit();
 137                                 return (EPERM);
 138                         }
 139                         if (delta.tv_sec > 1) {
 140                                 tv->tv_sec = tv1.tv_sec + 1;
 141                                 kprintf("Time adjustment clamped to +1 second\n");
 142                         }
 143                         laststep = *tv;
 144                 }
 145         }
 146
 147         ts.tv_sec = tv->tv_sec;
 148         ts.tv_nsec = tv->tv_usec * 1000;
 149         set_timeofday(&ts);
 150         crit_exit();
 151
 152         if (origcpu != 0)
 153                 lwkt_setcpu_self(globaldata_find(origcpu));
 154
 155         resettodr();
 156         return (0);
 157 }
 158
 159 static void
 160 get_process_cputime(struct proc *p, struct timespec *ats)
 161 {
 162         struct rusage ru;
 163
 164         lwkt_gettoken(&p->p_token);
 165         calcru_proc(p, &ru);
 166         lwkt_reltoken(&p->p_token);
 167         timevaladd(&ru.ru_utime, &ru.ru_stime);
 168         TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats);
 169 }
 170
 171 static void
 172 get_process_usertime(struct proc *p, struct timespec *ats)
 173 {
 174         struct rusage ru;
 175
 176         lwkt_gettoken(&p->p_token);
 177         calcru_proc(p, &ru);
 178         lwkt_reltoken(&p->p_token);
 179         TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats);
 180 }
 181
 182 static void
 183 get_thread_cputime(struct thread *td, struct timespec *ats)
 184 {
 185         struct timeval sys, user;
 186
 187         calcru(td->td_lwp, &user, &sys);
 188         timevaladd(&user, &sys);
 189         TIMEVAL_TO_TIMESPEC(&user, ats);
 190 }
 191
 192 /*
 193  * MPSAFE
 194  */
 195 int
 196 kern_clock_gettime(clockid_t clock_id, struct timespec *ats)
 197 {
 198         struct proc *p;
 199
 200         p = curproc;
 201         switch(clock_id) {
 202         case CLOCK_REALTIME:
 203         case CLOCK_REALTIME_PRECISE:
 204                 nanotime(ats);
 205                 break;
 206         case CLOCK_REALTIME_FAST:
 207                 getnanotime(ats);
 208                 break;
 209         case CLOCK_MONOTONIC:
 210         case CLOCK_MONOTONIC_PRECISE:
 211         case CLOCK_UPTIME:
 212         case CLOCK_UPTIME_PRECISE:
 213                 nanouptime(ats);
 214                 break;
 215         case CLOCK_MONOTONIC_FAST:
 216         case CLOCK_UPTIME_FAST:
 217                 getnanouptime(ats);
 218                 break;
 219         case CLOCK_VIRTUAL:
 220                 get_process_usertime(p, ats);
 221                 break;
 222         case CLOCK_PROF:
 223         case CLOCK_PROCESS_CPUTIME_ID:
 224                 get_process_cputime(p, ats);
 225                 break;
 226         case CLOCK_SECOND:
 227                 ats->tv_sec = time_second;
 228                 ats->tv_nsec = 0;
 229                 break;
 230         case CLOCK_THREAD_CPUTIME_ID:
 231                 get_thread_cputime(curthread, ats);
 232                 break;
 233         default:
 234                 return (EINVAL);
 235         }
 236         return (0);
 237 }
 238
 239 /*
 240  * MPSAFE
 241  */
 242 int
 243 sys_clock_gettime(struct clock_gettime_args *uap)
 244 {
 245         struct timespec ats;
 246         int error;
 247
 248         error = kern_clock_gettime(uap->clock_id, &ats);
 249         if (error == 0)
 250                 error = copyout(&ats, uap->tp, sizeof(ats));
 251
 252         return (error);
 253 }
 254
 255 int
 256 kern_clock_settime(clockid_t clock_id, struct timespec *ats)
 257 {
 258         struct thread *td = curthread;
 259         struct timeval atv;
 260         int error;
 261
 262         if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
 263                 return (error);
 264         if (clock_id != CLOCK_REALTIME)
 265                 return (EINVAL);
 266         if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
 267                 return (EINVAL);
 268
 269         lockmgr(&masterclock_lock, LK_EXCLUSIVE);
 270         TIMESPEC_TO_TIMEVAL(&atv, ats);
 271         error = settime(&atv);
 272         lockmgr(&masterclock_lock, LK_RELEASE);
 273
 274         return (error);
 275 }
 276
 277 /*
 278  * MPALMOSTSAFE
 279  */
 280 int
 281 sys_clock_settime(struct clock_settime_args *uap)
 282 {
 283         struct timespec ats;
 284         int error;
 285
 286         if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
 287                 return (error);
 288
 289         error = kern_clock_settime(uap->clock_id, &ats);
 290
 291         return (error);
 292 }
 293
 294 /*
 295  * MPSAFE
 296  */
 297 int
 298 kern_clock_getres(clockid_t clock_id, struct timespec *ts)
 299 {
 300         ts->tv_sec = 0;
 301         switch(clock_id) {
 302         case CLOCK_REALTIME:
 303         case CLOCK_REALTIME_FAST:
 304         case CLOCK_REALTIME_PRECISE:
 305         case CLOCK_MONOTONIC:
 306         case CLOCK_MONOTONIC_FAST:
 307         case CLOCK_MONOTONIC_PRECISE:
 308         case CLOCK_UPTIME:
 309         case CLOCK_UPTIME_FAST:
 310         case CLOCK_UPTIME_PRECISE:
 311                 /*
 312                  * Round up the result of the division cheaply
 313                  * by adding 1.  Rounding up is especially important
 314                  * if rounding down would give 0.  Perfect rounding
 315                  * is unimportant.
 316                  */
 317                 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1;
 318                 break;
 319         case CLOCK_VIRTUAL:
 320         case CLOCK_PROF:
 321                 /* Accurately round up here because we can do so cheaply. */
 322                 ts->tv_nsec = (1000000000 + hz - 1) / hz;
 323                 break;
 324         case CLOCK_SECOND:
 325                 ts->tv_sec = 1;
 326                 ts->tv_nsec = 0;
 327                 break;
 328         case CLOCK_THREAD_CPUTIME_ID:
 329         case CLOCK_PROCESS_CPUTIME_ID:
 330                 ts->tv_nsec = 1000;
 331                 break;
 332         default:
 333                 return (EINVAL);
 334         }
 335
 336         return (0);
 337 }
 338
 339 /*
 340  * MPSAFE
 341  */
 342 int
 343 sys_clock_getres(struct clock_getres_args *uap)
 344 {
 345         int error;
 346         struct timespec ts;
 347
 348         error = kern_clock_getres(uap->clock_id, &ts);
 349         if (error == 0)
 350                 error = copyout(&ts, uap->tp, sizeof(ts));
 351
 352         return (error);
 353 }
 354
 355 /*
 356  * nanosleep1()
 357  *
 358  *      This is a general helper function for nanosleep() (aka sleep() aka
 359  *      usleep()).
 360  *
 361  *      If there is less then one tick's worth of time left and
 362  *      we haven't done a yield, or the remaining microseconds is
 363  *      ridiculously low, do a yield.  This avoids having
 364  *      to deal with systimer overheads when the system is under
 365  *      heavy loads.  If we have done a yield already then use
 366  *      a systimer and an uninterruptable thread wait.
 367  *
 368  *      If there is more then a tick's worth of time left,
 369  *      calculate the baseline ticks and use an interruptable
 370  *      tsleep, then handle the fine-grained delay on the next
 371  *      loop.  This usually results in two sleeps occuring, a long one
 372  *      and a short one.
 373  *
 374  * MPSAFE
 375  */
 376 static void
 377 ns1_systimer(systimer_t info, int in_ipi __unused,
 378     struct intrframe *frame __unused)
 379 {
 380         lwkt_schedule(info->data);
 381 }
 382
 383 int
 384 nanosleep1(struct timespec *rqt, struct timespec *rmt)
 385 {
 386         static int nanowait;
 387         struct timespec ts, ts2, ts3;
 388         struct timeval tv;
 389         int error;
 390
 391         if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
 392                 return (EINVAL);
 393         /* XXX: imho this should return EINVAL at least for tv_sec < 0 */
 394         if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
 395                 return (0);
 396         nanouptime(&ts);
 397         timespecadd(&ts, rqt);          /* ts = target timestamp compare */
 398         TIMESPEC_TO_TIMEVAL(&tv, rqt);  /* tv = sleep interval */
 399
 400         for (;;) {
 401                 int ticks;
 402                 struct systimer info;
 403
 404                 ticks = tv.tv_usec / ustick;    /* approximate */
 405
 406                 if (tv.tv_sec == 0 && ticks == 0) {
 407                         thread_t td = curthread;
 408                         if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us)
 409                                 tv.tv_usec = nanosleep_min_us;
 410                         if (tv.tv_usec < nanosleep_hard_us) {
 411                                 lwkt_user_yield();
 412                                 cpu_pause();
 413                         } else {
 414                                 crit_enter_quick(td);
 415                                 systimer_init_oneshot(&info, ns1_systimer,
 416                                                 td, tv.tv_usec);
 417                                 lwkt_deschedule_self(td);
 418                                 crit_exit_quick(td);
 419                                 lwkt_switch();
 420                                 systimer_del(&info); /* make sure it's gone */
 421                         }
 422                         error = iscaught(td->td_lwp);
 423                 } else if (tv.tv_sec == 0) {
 424                         error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
 425                 } else {
 426                         ticks = tvtohz_low(&tv); /* also handles overflow */
 427                         error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
 428                 }
 429                 nanouptime(&ts2);
 430                 if (error && error != EWOULDBLOCK) {
 431                         if (error == ERESTART)
 432                                 error = EINTR;
 433                         if (rmt != NULL) {
 434                                 timespecsub(&ts, &ts2);
 435                                 if (ts.tv_sec < 0)
 436                                         timespecclear(&ts);
 437                                 *rmt = ts;
 438                         }
 439                         return (error);
 440                 }
 441                 if (timespeccmp(&ts2, &ts, >=))
 442                         return (0);
 443                 ts3 = ts;
 444                 timespecsub(&ts3, &ts2);
 445                 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
 446         }
 447 }
 448
 449 /*
 450  * MPSAFE
 451  */
 452 int
 453 sys_nanosleep(struct nanosleep_args *uap)
 454 {
 455         int error;
 456         struct timespec rqt;
 457         struct timespec rmt;
 458
 459         error = copyin(uap->rqtp, &rqt, sizeof(rqt));
 460         if (error)
 461                 return (error);
 462
 463         error = nanosleep1(&rqt, &rmt);
 464
 465         /*
 466          * copyout the residual if nanosleep was interrupted.
 467          */
 468         if (error && uap->rmtp) {
 469                 int error2;
 470
 471                 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
 472                 if (error2)
 473                         error = error2;
 474         }
 475         return (error);
 476 }
 477
 478 /*
 479  * The gettimeofday() system call is supposed to return a fine-grained
 480  * realtime stamp.  However, acquiring a fine-grained stamp can create a
 481  * bottleneck when multiple cpu cores are trying to accessing e.g. the
 482  * HPET hardware timer all at the same time, so we have a sysctl that
 483  * allows its behavior to be changed to a more coarse-grained timestamp
 484  * which does not have to access a hardware timer.
 485  */
 486 int
 487 sys_gettimeofday(struct gettimeofday_args *uap)
 488 {
 489         struct timeval atv;
 490         int error = 0;
 491
 492         if (uap->tp) {
 493                 if (gettimeofday_quick)
 494                         getmicrotime(&atv);
 495                 else
 496                         microtime(&atv);
 497                 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
 498                     sizeof (atv))))
 499                         return (error);
 500         }
 501         if (uap->tzp)
 502                 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
 503                     sizeof (tz));
 504         return (error);
 505 }
 506
 507 /*
 508  * MPALMOSTSAFE
 509  */
 510 int
 511 sys_settimeofday(struct settimeofday_args *uap)
 512 {
 513         struct thread *td = curthread;
 514         struct timeval atv;
 515         struct timezone atz;
 516         int error;
 517
 518         if ((error = priv_check(td, PRIV_SETTIMEOFDAY)))
 519                 return (error);
 520         /*
 521          * Verify all parameters before changing time.
 522          *
 523          * XXX: We do not allow the time to be set to 0.0, which also by
 524          *      happy coincidence works around a pkgsrc bulk build bug.
 525          */
 526         if (uap->tv) {
 527                 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
 528                     sizeof(atv))))
 529                         return (error);
 530                 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
 531                         return (EINVAL);
 532                 if (atv.tv_sec == 0 && atv.tv_usec == 0)
 533                         return (EINVAL);
 534         }
 535         if (uap->tzp &&
 536             (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
 537                 return (error);
 538
 539         lockmgr(&masterclock_lock, LK_EXCLUSIVE);
 540         if (uap->tv && (error = settime(&atv))) {
 541                 lockmgr(&masterclock_lock, LK_RELEASE);
 542                 return (error);
 543         }
 544         lockmgr(&masterclock_lock, LK_RELEASE);
 545
 546         if (uap->tzp)
 547                 tz = atz;
 548         return (0);
 549 }
 550
 551 /*
 552  * WARNING! Run with ntp_spin held
 553  */
 554 static void
 555 kern_adjtime_common(void)
 556 {
 557         if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) ||
 558             (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta))
 559                 ntp_tick_delta = ntp_delta;
 560         else if (ntp_delta > ntp_big_delta)
 561                 ntp_tick_delta = 10 * ntp_default_tick_delta;
 562         else if (ntp_delta < -ntp_big_delta)
 563                 ntp_tick_delta = -10 * ntp_default_tick_delta;
 564         else if (ntp_delta > 0)
 565                 ntp_tick_delta = ntp_default_tick_delta;
 566         else
 567                 ntp_tick_delta = -ntp_default_tick_delta;
 568 }
 569
 570 void
 571 kern_adjtime(int64_t delta, int64_t *odelta)
 572 {
 573         spin_lock(&ntp_spin);
 574         *odelta = ntp_delta;
 575         ntp_delta = delta;
 576         kern_adjtime_common();
 577         spin_unlock(&ntp_spin);
 578 }
 579
 580 static void
 581 kern_get_ntp_delta(int64_t *delta)
 582 {
 583         *delta = ntp_delta;
 584 }
 585
 586 void
 587 kern_reladjtime(int64_t delta)
 588 {
 589         spin_lock(&ntp_spin);
 590         ntp_delta += delta;
 591         kern_adjtime_common();
 592         spin_unlock(&ntp_spin);
 593 }
 594
 595 static void
 596 kern_adjfreq(int64_t rate)
 597 {
 598         spin_lock(&ntp_spin);
 599         ntp_tick_permanent = rate;
 600         spin_unlock(&ntp_spin);
 601 }
 602
 603 /*
 604  * MPALMOSTSAFE
 605  */
 606 int
 607 sys_adjtime(struct adjtime_args *uap)
 608 {
 609         struct thread *td = curthread;
 610         struct timeval atv;
 611         int64_t ndelta, odelta;
 612         int error;
 613
 614         if ((error = priv_check(td, PRIV_ADJTIME)))
 615                 return (error);
 616         error = copyin(uap->delta, &atv, sizeof(struct timeval));
 617         if (error)
 618                 return (error);
 619
 620         /*
 621          * Compute the total correction and the rate at which to apply it.
 622          * Round the adjustment down to a whole multiple of the per-tick
 623          * delta, so that after some number of incremental changes in
 624          * hardclock(), tickdelta will become zero, lest the correction
 625          * overshoot and start taking us away from the desired final time.
 626          */
 627         ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000;
 628         kern_adjtime(ndelta, &odelta);
 629
 630         if (uap->olddelta) {
 631                 atv.tv_sec = odelta / 1000000000;
 632                 atv.tv_usec = odelta % 1000000000 / 1000;
 633                 copyout(&atv, uap->olddelta, sizeof(struct timeval));
 634         }
 635         return (0);
 636 }
 637
 638 static int
 639 sysctl_adjtime(SYSCTL_HANDLER_ARGS)
 640 {
 641         int64_t delta;
 642         int error;
 643
 644         if (req->newptr != NULL) {
 645                 if (priv_check(curthread, PRIV_ROOT))
 646                         return (EPERM);
 647                 error = SYSCTL_IN(req, &delta, sizeof(delta));
 648                 if (error)
 649                         return (error);
 650                 kern_reladjtime(delta);
 651         }
 652
 653         if (req->oldptr)
 654                 kern_get_ntp_delta(&delta);
 655         error = SYSCTL_OUT(req, &delta, sizeof(delta));
 656         return (error);
 657 }
 658
 659 /*
 660  * delta is in nanoseconds.
 661  */
 662 static int
 663 sysctl_delta(SYSCTL_HANDLER_ARGS)
 664 {
 665         int64_t delta, old_delta;
 666         int error;
 667
 668         if (req->newptr != NULL) {
 669                 if (priv_check(curthread, PRIV_ROOT))
 670                         return (EPERM);
 671                 error = SYSCTL_IN(req, &delta, sizeof(delta));
 672                 if (error)
 673                         return (error);
 674                 kern_adjtime(delta, &old_delta);
 675         }
 676
 677         if (req->oldptr != NULL)
 678                 kern_get_ntp_delta(&old_delta);
 679         error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta));
 680         return (error);
 681 }
 682
 683 /*
 684  * frequency is in nanoseconds per second shifted left 32.
 685  * kern_adjfreq() needs it in nanoseconds per tick shifted left 32.
 686  */
 687 static int
 688 sysctl_adjfreq(SYSCTL_HANDLER_ARGS)
 689 {
 690         int64_t freqdelta;
 691         int error;
 692
 693         if (req->newptr != NULL) {
 694                 if (priv_check(curthread, PRIV_ROOT))
 695                         return (EPERM);
 696                 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta));
 697                 if (error)
 698                         return (error);
 699
 700                 freqdelta /= hz;
 701                 kern_adjfreq(freqdelta);
 702         }
 703
 704         if (req->oldptr != NULL)
 705                 freqdelta = ntp_tick_permanent * hz;
 706         error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta));
 707         if (error)
 708                 return (error);
 709
 710         return (0);
 711 }
 712
 713 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls");
 714 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent,
 715     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 716     sysctl_adjfreq, "Q", "permanent correction per second");
 717 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta,
 718     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 719     sysctl_delta, "Q", "one-time delta");
 720 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD,
 721     &ntp_big_delta, sizeof(ntp_big_delta), "Q",
 722     "threshold for fast adjustment");
 723 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD,
 724     &ntp_tick_delta, sizeof(ntp_tick_delta), "LU",
 725     "per-tick adjustment");
 726 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD,
 727     &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU",
 728     "default per-tick adjustment");
 729 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW,
 730     &ntp_leap_second, sizeof(ntp_leap_second), "LU",
 731     "next leap second");
 732 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW,
 733     &ntp_leap_insert, 0, "insert or remove leap second");
 734 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust,
 735     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 736     sysctl_adjtime, "Q", "relative adjust for delta");
 737
 738 /*
 739  * Get value of an interval timer.  The process virtual and
 740  * profiling virtual time timers are kept in the p_stats area, since
 741  * they can be swapped out.  These are kept internally in the
 742  * way they are specified externally: in time until they expire.
 743  *
 744  * The real time interval timer is kept in the process table slot
 745  * for the process, and its value (it_value) is kept as an
 746  * absolute time rather than as a delta, so that it is easy to keep
 747  * periodic real-time signals from drifting.
 748  *
 749  * Virtual time timers are processed in the hardclock() routine of
 750  * kern_clock.c.  The real time timer is processed by a timeout
 751  * routine, called from the softclock() routine.  Since a callout
 752  * may be delayed in real time due to interrupt processing in the system,
 753  * it is possible for the real time timeout routine (realitexpire, given below),
 754  * to be delayed in real time past when it is supposed to occur.  It
 755  * does not suffice, therefore, to reload the real timer .it_value from the
 756  * real time timers .it_interval.  Rather, we compute the next time in
 757  * absolute time the timer should go off.
 758  *
 759  * MPALMOSTSAFE
 760  */
 761 int
 762 sys_getitimer(struct getitimer_args *uap)
 763 {
 764         struct proc *p = curproc;
 765         struct timeval ctv;
 766         struct itimerval aitv;
 767
 768         if (uap->which > ITIMER_PROF)
 769                 return (EINVAL);
 770         lwkt_gettoken(&p->p_token);
 771         if (uap->which == ITIMER_REAL) {
 772                 /*
 773                  * Convert from absolute to relative time in .it_value
 774                  * part of real time timer.  If time for real time timer
 775                  * has passed return 0, else return difference between
 776                  * current time and time for the timer to go off.
 777                  */
 778                 aitv = p->p_realtimer;
 779                 if (timevalisset(&aitv.it_value)) {
 780                         getmicrouptime(&ctv);
 781                         if (timevalcmp(&aitv.it_value, &ctv, <))
 782                                 timevalclear(&aitv.it_value);
 783                         else
 784                                 timevalsub(&aitv.it_value, &ctv);
 785                 }
 786         } else {
 787                 aitv = p->p_timer[uap->which];
 788         }
 789         lwkt_reltoken(&p->p_token);
 790         return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
 791 }
 792
 793 /*
 794  * MPALMOSTSAFE
 795  */
 796 int
 797 sys_setitimer(struct setitimer_args *uap)
 798 {
 799         struct itimerval aitv;
 800         struct timeval ctv;
 801         struct itimerval *itvp;
 802         struct proc *p = curproc;
 803         int error;
 804
 805         if (uap->which > ITIMER_PROF)
 806                 return (EINVAL);
 807         itvp = uap->itv;
 808         if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
 809             sizeof(struct itimerval))))
 810                 return (error);
 811         if ((uap->itv = uap->oitv) &&
 812             (error = sys_getitimer((struct getitimer_args *)uap)))
 813                 return (error);
 814         if (itvp == NULL)
 815                 return (0);
 816         if (itimerfix(&aitv.it_value))
 817                 return (EINVAL);
 818         if (!timevalisset(&aitv.it_value))
 819                 timevalclear(&aitv.it_interval);
 820         else if (itimerfix(&aitv.it_interval))
 821                 return (EINVAL);
 822         lwkt_gettoken(&p->p_token);
 823         if (uap->which == ITIMER_REAL) {
 824                 if (timevalisset(&p->p_realtimer.it_value))
 825                         callout_stop_sync(&p->p_ithandle);
 826                 if (timevalisset(&aitv.it_value))
 827                         callout_reset(&p->p_ithandle,
 828                             tvtohz_high(&aitv.it_value), realitexpire, p);
 829                 getmicrouptime(&ctv);
 830                 timevaladd(&aitv.it_value, &ctv);
 831                 p->p_realtimer = aitv;
 832         } else {
 833                 p->p_timer[uap->which] = aitv;
 834                 switch(uap->which) {
 835                 case ITIMER_VIRTUAL:
 836                         p->p_flags &= ~P_SIGVTALRM;
 837                         break;
 838                 case ITIMER_PROF:
 839                         p->p_flags &= ~P_SIGPROF;
 840                         break;
 841                 }
 842         }
 843         lwkt_reltoken(&p->p_token);
 844         return (0);
 845 }
 846
 847 /*
 848  * Real interval timer expired:
 849  * send process whose timer expired an alarm signal.
 850  * If time is not set up to reload, then just return.
 851  * Else compute next time timer should go off which is > current time.
 852  * This is where delay in processing this timeout causes multiple
 853  * SIGALRM calls to be compressed into one.
 854  * tvtohz_high() always adds 1 to allow for the time until the next clock
 855  * interrupt being strictly less than 1 clock tick, but we don't want
 856  * that here since we want to appear to be in sync with the clock
 857  * interrupt even when we're delayed.
 858  */
 859 static
 860 void
 861 realitexpire(void *arg)
 862 {
 863         struct proc *p;
 864         struct timeval ctv, ntv;
 865
 866         p = (struct proc *)arg;
 867         PHOLD(p);
 868         lwkt_gettoken(&p->p_token);
 869         ksignal(p, SIGALRM);
 870         if (!timevalisset(&p->p_realtimer.it_interval)) {
 871                 timevalclear(&p->p_realtimer.it_value);
 872                 goto done;
 873         }
 874         for (;;) {
 875                 timevaladd(&p->p_realtimer.it_value,
 876                            &p->p_realtimer.it_interval);
 877                 getmicrouptime(&ctv);
 878                 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
 879                         ntv = p->p_realtimer.it_value;
 880                         timevalsub(&ntv, &ctv);
 881                         callout_reset(&p->p_ithandle, tvtohz_low(&ntv),
 882                                       realitexpire, p);
 883                         goto done;
 884                 }
 885         }
 886 done:
 887         lwkt_reltoken(&p->p_token);
 888         PRELE(p);
 889 }
 890
 891 /*
 892  * Used to validate itimer timeouts and utimes*() timespecs.
 893  */
 894 int
 895 itimerfix(struct timeval *tv)
 896 {
 897         if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
 898                 return (EINVAL);
 899         if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick)
 900                 tv->tv_usec = ustick;
 901         return (0);
 902 }
 903
 904 /*
 905  * Used to validate timeouts and utimes*() timespecs.
 906  */
 907 int
 908 itimespecfix(struct timespec *ts)
 909 {
 910         if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000ULL)
 911                 return (EINVAL);
 912         if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < nstick)
 913                 ts->tv_nsec = nstick;
 914         return (0);
 915 }
 916
 917 /*
 918  * Decrement an interval timer by a specified number
 919  * of microseconds, which must be less than a second,
 920  * i.e. < 1000000.  If the timer expires, then reload
 921  * it.  In this case, carry over (usec - old value) to
 922  * reduce the value reloaded into the timer so that
 923  * the timer does not drift.  This routine assumes
 924  * that it is called in a context where the timers
 925  * on which it is operating cannot change in value.
 926  */
 927 int
 928 itimerdecr(struct itimerval *itp, int usec)
 929 {
 930
 931         if (itp->it_value.tv_usec < usec) {
 932                 if (itp->it_value.tv_sec == 0) {
 933                         /* expired, and already in next interval */
 934                         usec -= itp->it_value.tv_usec;
 935                         goto expire;
 936                 }
 937                 itp->it_value.tv_usec += 1000000;
 938                 itp->it_value.tv_sec--;
 939         }
 940         itp->it_value.tv_usec -= usec;
 941         usec = 0;
 942         if (timevalisset(&itp->it_value))
 943                 return (1);
 944         /* expired, exactly at end of interval */
 945 expire:
 946         if (timevalisset(&itp->it_interval)) {
 947                 itp->it_value = itp->it_interval;
 948                 itp->it_value.tv_usec -= usec;
 949                 if (itp->it_value.tv_usec < 0) {
 950                         itp->it_value.tv_usec += 1000000;
 951                         itp->it_value.tv_sec--;
 952                 }
 953         } else
 954                 itp->it_value.tv_usec = 0;              /* sec is already 0 */
 955         return (0);
 956 }
 957
 958 /*
 959  * Add and subtract routines for timevals.
 960  * N.B.: subtract routine doesn't deal with
 961  * results which are before the beginning,
 962  * it just gets very confused in this case.
 963  * Caveat emptor.
 964  */
 965 void
 966 timevaladd(struct timeval *t1, const struct timeval *t2)
 967 {
 968
 969         t1->tv_sec += t2->tv_sec;
 970         t1->tv_usec += t2->tv_usec;
 971         timevalfix(t1);
 972 }
 973
 974 void
 975 timevalsub(struct timeval *t1, const struct timeval *t2)
 976 {
 977
 978         t1->tv_sec -= t2->tv_sec;
 979         t1->tv_usec -= t2->tv_usec;
 980         timevalfix(t1);
 981 }
 982
 983 static void
 984 timevalfix(struct timeval *t1)
 985 {
 986
 987         if (t1->tv_usec < 0) {
 988                 t1->tv_sec--;
 989                 t1->tv_usec += 1000000;
 990         }
 991         if (t1->tv_usec >= 1000000) {
 992                 t1->tv_sec++;
 993                 t1->tv_usec -= 1000000;
 994         }
 995 }
 996
 997 /*
 998  * ratecheck(): simple time-based rate-limit checking.
 999  */
1000 int
1001 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1002 {
1003         struct timeval tv, delta;
1004         int rv = 0;
1005
1006         getmicrouptime(&tv);            /* NB: 10ms precision */
1007         delta = tv;
1008         timevalsub(&delta, lasttime);
1009
1010         /*
1011          * check for 0,0 is so that the message will be seen at least once,
1012          * even if interval is huge.
1013          */
1014         if (timevalcmp(&delta, mininterval, >=) ||
1015             (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1016                 *lasttime = tv;
1017                 rv = 1;
1018         }
1019
1020         return (rv);
1021 }
1022
1023 /*
1024  * ppsratecheck(): packets (or events) per second limitation.
1025  *
1026  * Return 0 if the limit is to be enforced (e.g. the caller
1027  * should drop a packet because of the rate limitation).
1028  *
1029  * maxpps of 0 always causes zero to be returned.  maxpps of -1
1030  * always causes 1 to be returned; this effectively defeats rate
1031  * limiting.
1032  *
1033  * Note that we maintain the struct timeval for compatibility
1034  * with other bsd systems.  We reuse the storage and just monitor
1035  * clock ticks for minimal overhead.
1036  */
1037 int
1038 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
1039 {
1040         int now;
1041
1042         /*
1043          * Reset the last time and counter if this is the first call
1044          * or more than a second has passed since the last update of
1045          * lasttime.
1046          */
1047         now = ticks;
1048         if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1049                 lasttime->tv_sec = now;
1050                 *curpps = 1;
1051                 return (maxpps != 0);
1052         } else {
1053                 (*curpps)++;            /* NB: ignore potential overflow */
1054                 return (maxpps < 0 || *curpps < maxpps);
1055         }
1056 }