sys/kern/kern_time.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 4. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)kern_time.c 8.1 (Berkeley) 6/10/93
  30  * $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/thread2.h>
  53 #include <sys/mplock2.h>
  54
  55 struct timezone tz;
  56
  57 /*
  58  * Time of day and interval timer support.
  59  *
  60  * These routines provide the kernel entry points to get and set
  61  * the time-of-day and per-process interval timers.  Subroutines
  62  * here provide support for adding and subtracting timeval structures
  63  * and decrementing interval timers, optionally reloading the interval
  64  * timers when they expire.
  65  */
  66
  67 static int      settime(struct timeval *);
  68 static void     timevalfix(struct timeval *);
  69
  70 /*
  71  * Nanosleep tries very hard to sleep for a precisely requested time
  72  * interval, down to 1uS.  The administrator can impose a minimum delay
  73  * and a delay below which we hard-loop instead of initiate a timer
  74  * interrupt and sleep.
  75  *
  76  * For machines under high loads it might be beneficial to increase min_us
  77  * to e.g. 1000uS (1ms) so spining processes sleep meaningfully.
  78  */
  79 static int     nanosleep_min_us = 10;
  80 static int     nanosleep_hard_us = 100;
  81 static int     gettimeofday_quick = 0;
  82 SYSCTL_INT(_kern, OID_AUTO, nanosleep_min_us, CTLFLAG_RW,
  83            &nanosleep_min_us, 0, "")
  84 SYSCTL_INT(_kern, OID_AUTO, nanosleep_hard_us, CTLFLAG_RW,
  85            &nanosleep_hard_us, 0, "")
  86 SYSCTL_INT(_kern, OID_AUTO, gettimeofday_quick, CTLFLAG_RW,
  87            &gettimeofday_quick, 0, "")
  88
  89 static int
  90 settime(struct timeval *tv)
  91 {
  92         struct timeval delta, tv1, tv2;
  93         static struct timeval maxtime, laststep;
  94         struct timespec ts;
  95         int origcpu;
  96
  97         if ((origcpu = mycpu->gd_cpuid) != 0)
  98                 lwkt_setcpu_self(globaldata_find(0));
  99
 100         crit_enter();
 101         microtime(&tv1);
 102         delta = *tv;
 103         timevalsub(&delta, &tv1);
 104
 105         /*
 106          * If the system is secure, we do not allow the time to be
 107          * set to a value earlier than 1 second less than the highest
 108          * time we have yet seen. The worst a miscreant can do in
 109          * this circumstance is "freeze" time. He couldn't go
 110          * back to the past.
 111          *
 112          * We similarly do not allow the clock to be stepped more
 113          * than one second, nor more than once per second. This allows
 114          * a miscreant to make the clock march double-time, but no worse.
 115          */
 116         if (securelevel > 1) {
 117                 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
 118                         /*
 119                          * Update maxtime to latest time we've seen.
 120                          */
 121                         if (tv1.tv_sec > maxtime.tv_sec)
 122                                 maxtime = tv1;
 123                         tv2 = *tv;
 124                         timevalsub(&tv2, &maxtime);
 125                         if (tv2.tv_sec < -1) {
 126                                 tv->tv_sec = maxtime.tv_sec - 1;
 127                                 kprintf("Time adjustment clamped to -1 second\n");
 128                         }
 129                 } else {
 130                         if (tv1.tv_sec == laststep.tv_sec) {
 131                                 crit_exit();
 132                                 return (EPERM);
 133                         }
 134                         if (delta.tv_sec > 1) {
 135                                 tv->tv_sec = tv1.tv_sec + 1;
 136                                 kprintf("Time adjustment clamped to +1 second\n");
 137                         }
 138                         laststep = *tv;
 139                 }
 140         }
 141
 142         ts.tv_sec = tv->tv_sec;
 143         ts.tv_nsec = tv->tv_usec * 1000;
 144         set_timeofday(&ts);
 145         crit_exit();
 146
 147         if (origcpu != 0)
 148                 lwkt_setcpu_self(globaldata_find(origcpu));
 149
 150         resettodr();
 151         return (0);
 152 }
 153
 154 static void
 155 get_curthread_cputime(struct timespec *ats)
 156 {
 157         struct thread *td = curthread;
 158
 159         crit_enter();
 160         /*
 161          * These are 64-bit fields but the actual values should never reach
 162          * the limit. We don't care about overflows.
 163          */
 164         ats->tv_sec = td->td_uticks / 1000000;
 165         ats->tv_sec += td->td_sticks / 1000000;
 166         ats->tv_sec += td->td_iticks / 1000000;
 167         ats->tv_nsec = (td->td_uticks % 1000000) * 1000;
 168         ats->tv_nsec += (td->td_sticks % 1000000) * 1000;
 169         ats->tv_nsec += (td->td_iticks % 1000000) * 1000;
 170         crit_exit();
 171 }
 172
 173 /*
 174  * MPSAFE
 175  */
 176 int
 177 kern_clock_gettime(clockid_t clock_id, struct timespec *ats)
 178 {
 179         int error = 0;
 180         struct proc *p;
 181
 182         switch(clock_id) {
 183         case CLOCK_REALTIME:
 184         case CLOCK_REALTIME_PRECISE:
 185                 nanotime(ats);
 186                 break;
 187         case CLOCK_REALTIME_FAST:
 188                 getnanotime(ats);
 189                 break;
 190         case CLOCK_MONOTONIC:
 191         case CLOCK_MONOTONIC_PRECISE:
 192         case CLOCK_UPTIME:
 193         case CLOCK_UPTIME_PRECISE:
 194                 nanouptime(ats);
 195                 break;
 196         case CLOCK_MONOTONIC_FAST:
 197         case CLOCK_UPTIME_FAST:
 198                 getnanouptime(ats);
 199                 break;
 200         case CLOCK_VIRTUAL:
 201                 p = curproc;
 202                 ats->tv_sec = p->p_timer[ITIMER_VIRTUAL].it_value.tv_sec;
 203                 ats->tv_nsec = p->p_timer[ITIMER_VIRTUAL].it_value.tv_usec *
 204                                1000;
 205                 break;
 206         case CLOCK_PROF:
 207                 p = curproc;
 208                 ats->tv_sec = p->p_timer[ITIMER_PROF].it_value.tv_sec;
 209                 ats->tv_nsec = p->p_timer[ITIMER_PROF].it_value.tv_usec *
 210                                1000;
 211                 break;
 212         case CLOCK_SECOND:
 213                 ats->tv_sec = time_second;
 214                 ats->tv_nsec = 0;
 215                 break;
 216         case CLOCK_THREAD_CPUTIME_ID:
 217                 get_curthread_cputime(ats);
 218                 break;
 219         default:
 220                 error = EINVAL;
 221                 break;
 222         }
 223         return (error);
 224 }
 225
 226 /*
 227  * MPSAFE
 228  */
 229 int
 230 sys_clock_gettime(struct clock_gettime_args *uap)
 231 {
 232         struct timespec ats;
 233         int error;
 234
 235         error = kern_clock_gettime(uap->clock_id, &ats);
 236         if (error == 0)
 237                 error = copyout(&ats, uap->tp, sizeof(ats));
 238
 239         return (error);
 240 }
 241
 242 int
 243 kern_clock_settime(clockid_t clock_id, struct timespec *ats)
 244 {
 245         struct thread *td = curthread;
 246         struct timeval atv;
 247         int error;
 248
 249         if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
 250                 return (error);
 251         if (clock_id != CLOCK_REALTIME)
 252                 return (EINVAL);
 253         if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
 254                 return (EINVAL);
 255
 256         TIMESPEC_TO_TIMEVAL(&atv, ats);
 257         error = settime(&atv);
 258         return (error);
 259 }
 260
 261 /*
 262  * MPALMOSTSAFE
 263  */
 264 int
 265 sys_clock_settime(struct clock_settime_args *uap)
 266 {
 267         struct timespec ats;
 268         int error;
 269
 270         if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
 271                 return (error);
 272
 273         get_mplock();
 274         error = kern_clock_settime(uap->clock_id, &ats);
 275         rel_mplock();
 276         return (error);
 277 }
 278
 279 /*
 280  * MPSAFE
 281  */
 282 int
 283 kern_clock_getres(clockid_t clock_id, struct timespec *ts)
 284 {
 285         int error;
 286
 287         switch(clock_id) {
 288         case CLOCK_REALTIME:
 289         case CLOCK_REALTIME_FAST:
 290         case CLOCK_REALTIME_PRECISE:
 291         case CLOCK_MONOTONIC:
 292         case CLOCK_MONOTONIC_FAST:
 293         case CLOCK_MONOTONIC_PRECISE:
 294         case CLOCK_UPTIME:
 295         case CLOCK_UPTIME_FAST:
 296         case CLOCK_UPTIME_PRECISE:
 297         case CLOCK_THREAD_CPUTIME_ID:
 298                 /*
 299                  * Round up the result of the division cheaply
 300                  * by adding 1.  Rounding up is especially important
 301                  * if rounding down would give 0.  Perfect rounding
 302                  * is unimportant.
 303                  */
 304                 ts->tv_sec = 0;
 305                 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1;
 306                 error = 0;
 307                 break;
 308         case CLOCK_VIRTUAL:
 309         case CLOCK_PROF:
 310                 /* Accurately round up here because we can do so cheaply. */
 311                 ts->tv_sec = 0;
 312                 ts->tv_nsec = (1000000000 + hz - 1) / hz;
 313                 error = 0;
 314                 break;
 315         case CLOCK_SECOND:
 316                 ts->tv_sec = 1;
 317                 ts->tv_nsec = 0;
 318                 error = 0;
 319                 break;
 320         default:
 321                 error = EINVAL;
 322                 break;
 323         }
 324
 325         return(error);
 326 }
 327
 328 /*
 329  * MPSAFE
 330  */
 331 int
 332 sys_clock_getres(struct clock_getres_args *uap)
 333 {
 334         int error;
 335         struct timespec ts;
 336
 337         error = kern_clock_getres(uap->clock_id, &ts);
 338         if (error == 0)
 339                 error = copyout(&ts, uap->tp, sizeof(ts));
 340
 341         return (error);
 342 }
 343
 344 /*
 345  * nanosleep1()
 346  *
 347  *      This is a general helper function for nanosleep() (aka sleep() aka
 348  *      usleep()).
 349  *
 350  *      If there is less then one tick's worth of time left and
 351  *      we haven't done a yield, or the remaining microseconds is
 352  *      ridiculously low, do a yield.  This avoids having
 353  *      to deal with systimer overheads when the system is under
 354  *      heavy loads.  If we have done a yield already then use
 355  *      a systimer and an uninterruptable thread wait.
 356  *
 357  *      If there is more then a tick's worth of time left,
 358  *      calculate the baseline ticks and use an interruptable
 359  *      tsleep, then handle the fine-grained delay on the next
 360  *      loop.  This usually results in two sleeps occuring, a long one
 361  *      and a short one.
 362  *
 363  * MPSAFE
 364  */
 365 static void
 366 ns1_systimer(systimer_t info, int in_ipi __unused,
 367     struct intrframe *frame __unused)
 368 {
 369         lwkt_schedule(info->data);
 370 }
 371
 372 int
 373 nanosleep1(struct timespec *rqt, struct timespec *rmt)
 374 {
 375         static int nanowait;
 376         struct timespec ts, ts2, ts3;
 377         struct timeval tv;
 378         int error;
 379
 380         if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
 381                 return (EINVAL);
 382         /* XXX: imho this should return EINVAL at least for tv_sec < 0 */
 383         if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
 384                 return (0);
 385         nanouptime(&ts);
 386         timespecadd(&ts, rqt);          /* ts = target timestamp compare */
 387         TIMESPEC_TO_TIMEVAL(&tv, rqt);  /* tv = sleep interval */
 388
 389         for (;;) {
 390                 int ticks;
 391                 struct systimer info;
 392
 393                 ticks = tv.tv_usec / ustick;    /* approximate */
 394
 395                 if (tv.tv_sec == 0 && ticks == 0) {
 396                         thread_t td = curthread;
 397                         if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us)
 398                                 tv.tv_usec = nanosleep_min_us;
 399                         if (tv.tv_usec < nanosleep_hard_us) {
 400                                 lwkt_user_yield();
 401                                 cpu_pause();
 402                         } else {
 403                                 crit_enter_quick(td);
 404                                 systimer_init_oneshot(&info, ns1_systimer,
 405                                                 td, tv.tv_usec);
 406                                 lwkt_deschedule_self(td);
 407                                 crit_exit_quick(td);
 408                                 lwkt_switch();
 409                                 systimer_del(&info); /* make sure it's gone */
 410                         }
 411                         error = iscaught(td->td_lwp);
 412                 } else if (tv.tv_sec == 0) {
 413                         error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
 414                 } else {
 415                         ticks = tvtohz_low(&tv); /* also handles overflow */
 416                         error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
 417                 }
 418                 nanouptime(&ts2);
 419                 if (error && error != EWOULDBLOCK) {
 420                         if (error == ERESTART)
 421                                 error = EINTR;
 422                         if (rmt != NULL) {
 423                                 timespecsub(&ts, &ts2);
 424                                 if (ts.tv_sec < 0)
 425                                         timespecclear(&ts);
 426                                 *rmt = ts;
 427                         }
 428                         return (error);
 429                 }
 430                 if (timespeccmp(&ts2, &ts, >=))
 431                         return (0);
 432                 ts3 = ts;
 433                 timespecsub(&ts3, &ts2);
 434                 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
 435         }
 436 }
 437
 438 /*
 439  * MPSAFE
 440  */
 441 int
 442 sys_nanosleep(struct nanosleep_args *uap)
 443 {
 444         int error;
 445         struct timespec rqt;
 446         struct timespec rmt;
 447
 448         error = copyin(uap->rqtp, &rqt, sizeof(rqt));
 449         if (error)
 450                 return (error);
 451
 452         error = nanosleep1(&rqt, &rmt);
 453
 454         /*
 455          * copyout the residual if nanosleep was interrupted.
 456          */
 457         if (error && uap->rmtp) {
 458                 int error2;
 459
 460                 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
 461                 if (error2)
 462                         error = error2;
 463         }
 464         return (error);
 465 }
 466
 467 /*
 468  * The gettimeofday() system call is supposed to return a fine-grained
 469  * realtime stamp.  However, acquiring a fine-grained stamp can create a
 470  * bottleneck when multiple cpu cores are trying to accessing e.g. the
 471  * HPET hardware timer all at the same time, so we have a sysctl that
 472  * allows its behavior to be changed to a more coarse-grained timestamp
 473  * which does not have to access a hardware timer.
 474  */
 475 int
 476 sys_gettimeofday(struct gettimeofday_args *uap)
 477 {
 478         struct timeval atv;
 479         int error = 0;
 480
 481         if (uap->tp) {
 482                 if (gettimeofday_quick)
 483                         getmicrotime(&atv);
 484                 else
 485                         microtime(&atv);
 486                 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
 487                     sizeof (atv))))
 488                         return (error);
 489         }
 490         if (uap->tzp)
 491                 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
 492                     sizeof (tz));
 493         return (error);
 494 }
 495
 496 /*
 497  * MPALMOSTSAFE
 498  */
 499 int
 500 sys_settimeofday(struct settimeofday_args *uap)
 501 {
 502         struct thread *td = curthread;
 503         struct timeval atv;
 504         struct timezone atz;
 505         int error;
 506
 507         if ((error = priv_check(td, PRIV_SETTIMEOFDAY)))
 508                 return (error);
 509         /*
 510          * Verify all parameters before changing time.
 511          *
 512          * NOTE: We do not allow the time to be set to 0.0, which also by
 513          *       happy coincidence works around a pkgsrc bulk build bug.
 514          */
 515         if (uap->tv) {
 516                 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
 517                     sizeof(atv))))
 518                         return (error);
 519                 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
 520                         return (EINVAL);
 521                 if (atv.tv_sec == 0 && atv.tv_usec == 0)
 522                         return (EINVAL);
 523         }
 524         if (uap->tzp &&
 525             (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
 526                 return (error);
 527
 528         get_mplock();
 529         if (uap->tv && (error = settime(&atv))) {
 530                 rel_mplock();
 531                 return (error);
 532         }
 533         rel_mplock();
 534         if (uap->tzp)
 535                 tz = atz;
 536         return (0);
 537 }
 538
 539 static void
 540 kern_adjtime_common(void)
 541 {
 542         if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) ||
 543             (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta))
 544                 ntp_tick_delta = ntp_delta;
 545         else if (ntp_delta > ntp_big_delta)
 546                 ntp_tick_delta = 10 * ntp_default_tick_delta;
 547         else if (ntp_delta < -ntp_big_delta)
 548                 ntp_tick_delta = -10 * ntp_default_tick_delta;
 549         else if (ntp_delta > 0)
 550                 ntp_tick_delta = ntp_default_tick_delta;
 551         else
 552                 ntp_tick_delta = -ntp_default_tick_delta;
 553 }
 554
 555 void
 556 kern_adjtime(int64_t delta, int64_t *odelta)
 557 {
 558         int origcpu;
 559
 560         if ((origcpu = mycpu->gd_cpuid) != 0)
 561                 lwkt_setcpu_self(globaldata_find(0));
 562
 563         crit_enter();
 564         *odelta = ntp_delta;
 565         ntp_delta = delta;
 566         kern_adjtime_common();
 567         crit_exit();
 568
 569         if (origcpu != 0)
 570                 lwkt_setcpu_self(globaldata_find(origcpu));
 571 }
 572
 573 static void
 574 kern_get_ntp_delta(int64_t *delta)
 575 {
 576         int origcpu;
 577
 578         if ((origcpu = mycpu->gd_cpuid) != 0)
 579                 lwkt_setcpu_self(globaldata_find(0));
 580
 581         crit_enter();
 582         *delta = ntp_delta;
 583         crit_exit();
 584
 585         if (origcpu != 0)
 586                 lwkt_setcpu_self(globaldata_find(origcpu));
 587 }
 588
 589 void
 590 kern_reladjtime(int64_t delta)
 591 {
 592         int origcpu;
 593
 594         if ((origcpu = mycpu->gd_cpuid) != 0)
 595                 lwkt_setcpu_self(globaldata_find(0));
 596
 597         crit_enter();
 598         ntp_delta += delta;
 599         kern_adjtime_common();
 600         crit_exit();
 601
 602         if (origcpu != 0)
 603                 lwkt_setcpu_self(globaldata_find(origcpu));
 604 }
 605
 606 static void
 607 kern_adjfreq(int64_t rate)
 608 {
 609         int origcpu;
 610
 611         if ((origcpu = mycpu->gd_cpuid) != 0)
 612                 lwkt_setcpu_self(globaldata_find(0));
 613
 614         crit_enter();
 615         ntp_tick_permanent = rate;
 616         crit_exit();
 617
 618         if (origcpu != 0)
 619                 lwkt_setcpu_self(globaldata_find(origcpu));
 620 }
 621
 622 /*
 623  * MPALMOSTSAFE
 624  */
 625 int
 626 sys_adjtime(struct adjtime_args *uap)
 627 {
 628         struct thread *td = curthread;
 629         struct timeval atv;
 630         int64_t ndelta, odelta;
 631         int error;
 632
 633         if ((error = priv_check(td, PRIV_ADJTIME)))
 634                 return (error);
 635         error = copyin(uap->delta, &atv, sizeof(struct timeval));
 636         if (error)
 637                 return (error);
 638
 639         /*
 640          * Compute the total correction and the rate at which to apply it.
 641          * Round the adjustment down to a whole multiple of the per-tick
 642          * delta, so that after some number of incremental changes in
 643          * hardclock(), tickdelta will become zero, lest the correction
 644          * overshoot and start taking us away from the desired final time.
 645          */
 646         ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000;
 647         get_mplock();
 648         kern_adjtime(ndelta, &odelta);
 649         rel_mplock();
 650
 651         if (uap->olddelta) {
 652                 atv.tv_sec = odelta / 1000000000;
 653                 atv.tv_usec = odelta % 1000000000 / 1000;
 654                 copyout(&atv, uap->olddelta, sizeof(struct timeval));
 655         }
 656         return (0);
 657 }
 658
 659 static int
 660 sysctl_adjtime(SYSCTL_HANDLER_ARGS)
 661 {
 662         int64_t delta;
 663         int error;
 664
 665         if (req->newptr != NULL) {
 666                 if (priv_check(curthread, PRIV_ROOT))
 667                         return (EPERM);
 668                 error = SYSCTL_IN(req, &delta, sizeof(delta));
 669                 if (error)
 670                         return (error);
 671                 kern_reladjtime(delta);
 672         }
 673
 674         if (req->oldptr)
 675                 kern_get_ntp_delta(&delta);
 676         error = SYSCTL_OUT(req, &delta, sizeof(delta));
 677         return (error);
 678 }
 679
 680 /*
 681  * delta is in nanoseconds.
 682  */
 683 static int
 684 sysctl_delta(SYSCTL_HANDLER_ARGS)
 685 {
 686         int64_t delta, old_delta;
 687         int error;
 688
 689         if (req->newptr != NULL) {
 690                 if (priv_check(curthread, PRIV_ROOT))
 691                         return (EPERM);
 692                 error = SYSCTL_IN(req, &delta, sizeof(delta));
 693                 if (error)
 694                         return (error);
 695                 kern_adjtime(delta, &old_delta);
 696         }
 697
 698         if (req->oldptr != NULL)
 699                 kern_get_ntp_delta(&old_delta);
 700         error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta));
 701         return (error);
 702 }
 703
 704 /*
 705  * frequency is in nanoseconds per second shifted left 32.
 706  * kern_adjfreq() needs it in nanoseconds per tick shifted left 32.
 707  */
 708 static int
 709 sysctl_adjfreq(SYSCTL_HANDLER_ARGS)
 710 {
 711         int64_t freqdelta;
 712         int error;
 713
 714         if (req->newptr != NULL) {
 715                 if (priv_check(curthread, PRIV_ROOT))
 716                         return (EPERM);
 717                 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta));
 718                 if (error)
 719                         return (error);
 720
 721                 freqdelta /= hz;
 722                 kern_adjfreq(freqdelta);
 723         }
 724
 725         if (req->oldptr != NULL)
 726                 freqdelta = ntp_tick_permanent * hz;
 727         error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta));
 728         if (error)
 729                 return (error);
 730
 731         return (0);
 732 }
 733
 734 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls");
 735 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent,
 736     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 737     sysctl_adjfreq, "Q", "permanent correction per second");
 738 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta,
 739     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 740     sysctl_delta, "Q", "one-time delta");
 741 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD,
 742     &ntp_big_delta, sizeof(ntp_big_delta), "Q",
 743     "threshold for fast adjustment");
 744 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD,
 745     &ntp_tick_delta, sizeof(ntp_tick_delta), "LU",
 746     "per-tick adjustment");
 747 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD,
 748     &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU",
 749     "default per-tick adjustment");
 750 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW,
 751     &ntp_leap_second, sizeof(ntp_leap_second), "LU",
 752     "next leap second");
 753 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW,
 754     &ntp_leap_insert, 0, "insert or remove leap second");
 755 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust,
 756     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 757     sysctl_adjtime, "Q", "relative adjust for delta");
 758
 759 /*
 760  * Get value of an interval timer.  The process virtual and
 761  * profiling virtual time timers are kept in the p_stats area, since
 762  * they can be swapped out.  These are kept internally in the
 763  * way they are specified externally: in time until they expire.
 764  *
 765  * The real time interval timer is kept in the process table slot
 766  * for the process, and its value (it_value) is kept as an
 767  * absolute time rather than as a delta, so that it is easy to keep
 768  * periodic real-time signals from drifting.
 769  *
 770  * Virtual time timers are processed in the hardclock() routine of
 771  * kern_clock.c.  The real time timer is processed by a timeout
 772  * routine, called from the softclock() routine.  Since a callout
 773  * may be delayed in real time due to interrupt processing in the system,
 774  * it is possible for the real time timeout routine (realitexpire, given below),
 775  * to be delayed in real time past when it is supposed to occur.  It
 776  * does not suffice, therefore, to reload the real timer .it_value from the
 777  * real time timers .it_interval.  Rather, we compute the next time in
 778  * absolute time the timer should go off.
 779  *
 780  * MPALMOSTSAFE
 781  */
 782 int
 783 sys_getitimer(struct getitimer_args *uap)
 784 {
 785         struct proc *p = curproc;
 786         struct timeval ctv;
 787         struct itimerval aitv;
 788
 789         if (uap->which > ITIMER_PROF)
 790                 return (EINVAL);
 791         lwkt_gettoken(&p->p_token);
 792         if (uap->which == ITIMER_REAL) {
 793                 /*
 794                  * Convert from absolute to relative time in .it_value
 795                  * part of real time timer.  If time for real time timer
 796                  * has passed return 0, else return difference between
 797                  * current time and time for the timer to go off.
 798                  */
 799                 aitv = p->p_realtimer;
 800                 if (timevalisset(&aitv.it_value)) {
 801                         getmicrouptime(&ctv);
 802                         if (timevalcmp(&aitv.it_value, &ctv, <))
 803                                 timevalclear(&aitv.it_value);
 804                         else
 805                                 timevalsub(&aitv.it_value, &ctv);
 806                 }
 807         } else {
 808                 aitv = p->p_timer[uap->which];
 809         }
 810         lwkt_reltoken(&p->p_token);
 811         return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
 812 }
 813
 814 /*
 815  * MPALMOSTSAFE
 816  */
 817 int
 818 sys_setitimer(struct setitimer_args *uap)
 819 {
 820         struct itimerval aitv;
 821         struct timeval ctv;
 822         struct itimerval *itvp;
 823         struct proc *p = curproc;
 824         int error;
 825
 826         if (uap->which > ITIMER_PROF)
 827                 return (EINVAL);
 828         itvp = uap->itv;
 829         if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
 830             sizeof(struct itimerval))))
 831                 return (error);
 832         if ((uap->itv = uap->oitv) &&
 833             (error = sys_getitimer((struct getitimer_args *)uap)))
 834                 return (error);
 835         if (itvp == NULL)
 836                 return (0);
 837         if (itimerfix(&aitv.it_value))
 838                 return (EINVAL);
 839         if (!timevalisset(&aitv.it_value))
 840                 timevalclear(&aitv.it_interval);
 841         else if (itimerfix(&aitv.it_interval))
 842                 return (EINVAL);
 843         lwkt_gettoken(&p->p_token);
 844         if (uap->which == ITIMER_REAL) {
 845                 if (timevalisset(&p->p_realtimer.it_value))
 846                         callout_stop_sync(&p->p_ithandle);
 847                 if (timevalisset(&aitv.it_value))
 848                         callout_reset(&p->p_ithandle,
 849                             tvtohz_high(&aitv.it_value), realitexpire, p);
 850                 getmicrouptime(&ctv);
 851                 timevaladd(&aitv.it_value, &ctv);
 852                 p->p_realtimer = aitv;
 853         } else {
 854                 p->p_timer[uap->which] = aitv;
 855                 switch(uap->which) {
 856                 case ITIMER_VIRTUAL:
 857                         p->p_flags &= ~P_SIGVTALRM;
 858                         break;
 859                 case ITIMER_PROF:
 860                         p->p_flags &= ~P_SIGPROF;
 861                         break;
 862                 }
 863         }
 864         lwkt_reltoken(&p->p_token);
 865         return (0);
 866 }
 867
 868 /*
 869  * Real interval timer expired:
 870  * send process whose timer expired an alarm signal.
 871  * If time is not set up to reload, then just return.
 872  * Else compute next time timer should go off which is > current time.
 873  * This is where delay in processing this timeout causes multiple
 874  * SIGALRM calls to be compressed into one.
 875  * tvtohz_high() always adds 1 to allow for the time until the next clock
 876  * interrupt being strictly less than 1 clock tick, but we don't want
 877  * that here since we want to appear to be in sync with the clock
 878  * interrupt even when we're delayed.
 879  */
 880 void
 881 realitexpire(void *arg)
 882 {
 883         struct proc *p;
 884         struct timeval ctv, ntv;
 885
 886         p = (struct proc *)arg;
 887         PHOLD(p);
 888         lwkt_gettoken(&p->p_token);
 889         ksignal(p, SIGALRM);
 890         if (!timevalisset(&p->p_realtimer.it_interval)) {
 891                 timevalclear(&p->p_realtimer.it_value);
 892                 goto done;
 893         }
 894         for (;;) {
 895                 timevaladd(&p->p_realtimer.it_value,
 896                            &p->p_realtimer.it_interval);
 897                 getmicrouptime(&ctv);
 898                 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
 899                         ntv = p->p_realtimer.it_value;
 900                         timevalsub(&ntv, &ctv);
 901                         callout_reset(&p->p_ithandle, tvtohz_low(&ntv),
 902                                       realitexpire, p);
 903                         goto done;
 904                 }
 905         }
 906 done:
 907         lwkt_reltoken(&p->p_token);
 908         PRELE(p);
 909 }
 910
 911 /*
 912  * Check that a proposed value to load into the .it_value or
 913  * .it_interval part of an interval timer is acceptable, and
 914  * fix it to have at least minimal value (i.e. if it is less
 915  * than the resolution of the clock, round it up.)
 916  *
 917  * MPSAFE
 918  */
 919 int
 920 itimerfix(struct timeval *tv)
 921 {
 922
 923         if (tv->tv_sec < 0 || tv->tv_sec > 100000000 ||
 924             tv->tv_usec < 0 || tv->tv_usec >= 1000000)
 925                 return (EINVAL);
 926         if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick)
 927                 tv->tv_usec = ustick;
 928         return (0);
 929 }
 930
 931 /*
 932  * Decrement an interval timer by a specified number
 933  * of microseconds, which must be less than a second,
 934  * i.e. < 1000000.  If the timer expires, then reload
 935  * it.  In this case, carry over (usec - old value) to
 936  * reduce the value reloaded into the timer so that
 937  * the timer does not drift.  This routine assumes
 938  * that it is called in a context where the timers
 939  * on which it is operating cannot change in value.
 940  */
 941 int
 942 itimerdecr(struct itimerval *itp, int usec)
 943 {
 944
 945         if (itp->it_value.tv_usec < usec) {
 946                 if (itp->it_value.tv_sec == 0) {
 947                         /* expired, and already in next interval */
 948                         usec -= itp->it_value.tv_usec;
 949                         goto expire;
 950                 }
 951                 itp->it_value.tv_usec += 1000000;
 952                 itp->it_value.tv_sec--;
 953         }
 954         itp->it_value.tv_usec -= usec;
 955         usec = 0;
 956         if (timevalisset(&itp->it_value))
 957                 return (1);
 958         /* expired, exactly at end of interval */
 959 expire:
 960         if (timevalisset(&itp->it_interval)) {
 961                 itp->it_value = itp->it_interval;
 962                 itp->it_value.tv_usec -= usec;
 963                 if (itp->it_value.tv_usec < 0) {
 964                         itp->it_value.tv_usec += 1000000;
 965                         itp->it_value.tv_sec--;
 966                 }
 967         } else
 968                 itp->it_value.tv_usec = 0;              /* sec is already 0 */
 969         return (0);
 970 }
 971
 972 /*
 973  * Add and subtract routines for timevals.
 974  * N.B.: subtract routine doesn't deal with
 975  * results which are before the beginning,
 976  * it just gets very confused in this case.
 977  * Caveat emptor.
 978  */
 979 void
 980 timevaladd(struct timeval *t1, const struct timeval *t2)
 981 {
 982
 983         t1->tv_sec += t2->tv_sec;
 984         t1->tv_usec += t2->tv_usec;
 985         timevalfix(t1);
 986 }
 987
 988 void
 989 timevalsub(struct timeval *t1, const struct timeval *t2)
 990 {
 991
 992         t1->tv_sec -= t2->tv_sec;
 993         t1->tv_usec -= t2->tv_usec;
 994         timevalfix(t1);
 995 }
 996
 997 static void
 998 timevalfix(struct timeval *t1)
 999 {
1000
1001         if (t1->tv_usec < 0) {
1002                 t1->tv_sec--;
1003                 t1->tv_usec += 1000000;
1004         }
1005         if (t1->tv_usec >= 1000000) {
1006                 t1->tv_sec++;
1007                 t1->tv_usec -= 1000000;
1008         }
1009 }
1010
1011 /*
1012  * ratecheck(): simple time-based rate-limit checking.
1013  */
1014 int
1015 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1016 {
1017         struct timeval tv, delta;
1018         int rv = 0;
1019
1020         getmicrouptime(&tv);            /* NB: 10ms precision */
1021         delta = tv;
1022         timevalsub(&delta, lasttime);
1023
1024         /*
1025          * check for 0,0 is so that the message will be seen at least once,
1026          * even if interval is huge.
1027          */
1028         if (timevalcmp(&delta, mininterval, >=) ||
1029             (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1030                 *lasttime = tv;
1031                 rv = 1;
1032         }
1033
1034         return (rv);
1035 }
1036
1037 /*
1038  * ppsratecheck(): packets (or events) per second limitation.
1039  *
1040  * Return 0 if the limit is to be enforced (e.g. the caller
1041  * should drop a packet because of the rate limitation).
1042  *
1043  * maxpps of 0 always causes zero to be returned.  maxpps of -1
1044  * always causes 1 to be returned; this effectively defeats rate
1045  * limiting.
1046  *
1047  * Note that we maintain the struct timeval for compatibility
1048  * with other bsd systems.  We reuse the storage and just monitor
1049  * clock ticks for minimal overhead.
1050  */
1051 int
1052 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
1053 {
1054         int now;
1055
1056         /*
1057          * Reset the last time and counter if this is the first call
1058          * or more than a second has passed since the last update of
1059          * lasttime.
1060          */
1061         now = ticks;
1062         if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1063                 lasttime->tv_sec = now;
1064                 *curpps = 1;
1065                 return (maxpps != 0);
1066         } else {
1067                 (*curpps)++;            /* NB: ignore potential overflow */
1068                 return (maxpps < 0 || *curpps < maxpps);
1069         }
1070 }