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_curthread_cputime(struct timespec *ats)
 184 {
 185         struct thread *td = curthread;
 186         struct timeval sys, user;
 187
 188         calcru(td->td_lwp, &user, &sys);
 189         timevaladd(&user, &sys);
 190         TIMEVAL_TO_TIMESPEC(&user, ats);
 191 }
 192
 193 /*
 194  * MPSAFE
 195  */
 196 int
 197 kern_clock_gettime(clockid_t clock_id, struct timespec *ats)
 198 {
 199         struct proc *p;
 200
 201         p = curproc;
 202         switch(clock_id) {
 203         case CLOCK_REALTIME:
 204         case CLOCK_REALTIME_PRECISE:
 205                 nanotime(ats);
 206                 break;
 207         case CLOCK_REALTIME_FAST:
 208                 getnanotime(ats);
 209                 break;
 210         case CLOCK_MONOTONIC:
 211         case CLOCK_MONOTONIC_PRECISE:
 212         case CLOCK_UPTIME:
 213         case CLOCK_UPTIME_PRECISE:
 214                 nanouptime(ats);
 215                 break;
 216         case CLOCK_MONOTONIC_FAST:
 217         case CLOCK_UPTIME_FAST:
 218                 getnanouptime(ats);
 219                 break;
 220         case CLOCK_VIRTUAL:
 221                 get_process_usertime(p, ats);
 222                 break;
 223         case CLOCK_PROF:
 224         case CLOCK_PROCESS_CPUTIME_ID:
 225                 get_process_cputime(p, ats);
 226                 break;
 227         case CLOCK_SECOND:
 228                 ats->tv_sec = time_second;
 229                 ats->tv_nsec = 0;
 230                 break;
 231         case CLOCK_THREAD_CPUTIME_ID:
 232                 get_curthread_cputime(ats);
 233                 break;
 234         default:
 235                 return (EINVAL);
 236         }
 237         return (0);
 238 }
 239
 240 /*
 241  * MPSAFE
 242  */
 243 int
 244 sys_clock_gettime(struct clock_gettime_args *uap)
 245 {
 246         struct timespec ats;
 247         int error;
 248
 249         error = kern_clock_gettime(uap->clock_id, &ats);
 250         if (error == 0)
 251                 error = copyout(&ats, uap->tp, sizeof(ats));
 252
 253         return (error);
 254 }
 255
 256 int
 257 kern_clock_settime(clockid_t clock_id, struct timespec *ats)
 258 {
 259         struct thread *td = curthread;
 260         struct timeval atv;
 261         int error;
 262
 263         if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
 264                 return (error);
 265         if (clock_id != CLOCK_REALTIME)
 266                 return (EINVAL);
 267         if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
 268                 return (EINVAL);
 269
 270         lockmgr(&masterclock_lock, LK_EXCLUSIVE);
 271         TIMESPEC_TO_TIMEVAL(&atv, ats);
 272         error = settime(&atv);
 273         lockmgr(&masterclock_lock, LK_RELEASE);
 274
 275         return (error);
 276 }
 277
 278 /*
 279  * MPALMOSTSAFE
 280  */
 281 int
 282 sys_clock_settime(struct clock_settime_args *uap)
 283 {
 284         struct timespec ats;
 285         int error;
 286
 287         if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
 288                 return (error);
 289
 290         error = kern_clock_settime(uap->clock_id, &ats);
 291
 292         return (error);
 293 }
 294
 295 /*
 296  * MPSAFE
 297  */
 298 int
 299 kern_clock_getres(clockid_t clock_id, struct timespec *ts)
 300 {
 301         ts->tv_sec = 0;
 302         switch(clock_id) {
 303         case CLOCK_REALTIME:
 304         case CLOCK_REALTIME_FAST:
 305         case CLOCK_REALTIME_PRECISE:
 306         case CLOCK_MONOTONIC:
 307         case CLOCK_MONOTONIC_FAST:
 308         case CLOCK_MONOTONIC_PRECISE:
 309         case CLOCK_UPTIME:
 310         case CLOCK_UPTIME_FAST:
 311         case CLOCK_UPTIME_PRECISE:
 312                 /*
 313                  * Round up the result of the division cheaply
 314                  * by adding 1.  Rounding up is especially important
 315                  * if rounding down would give 0.  Perfect rounding
 316                  * is unimportant.
 317                  */
 318                 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1;
 319                 break;
 320         case CLOCK_VIRTUAL:
 321         case CLOCK_PROF:
 322                 /* Accurately round up here because we can do so cheaply. */
 323                 ts->tv_nsec = (1000000000 + hz - 1) / hz;
 324                 break;
 325         case CLOCK_SECOND:
 326                 ts->tv_sec = 1;
 327                 ts->tv_nsec = 0;
 328                 break;
 329         case CLOCK_THREAD_CPUTIME_ID:
 330         case CLOCK_PROCESS_CPUTIME_ID:
 331                 ts->tv_nsec = 1000;
 332                 break;
 333         default:
 334                 return (EINVAL);
 335         }
 336
 337         return (0);
 338 }
 339
 340 /*
 341  * MPSAFE
 342  */
 343 int
 344 sys_clock_getres(struct clock_getres_args *uap)
 345 {
 346         int error;
 347         struct timespec ts;
 348
 349         error = kern_clock_getres(uap->clock_id, &ts);
 350         if (error == 0)
 351                 error = copyout(&ts, uap->tp, sizeof(ts));
 352
 353         return (error);
 354 }
 355
 356 /*
 357  * nanosleep1()
 358  *
 359  *      This is a general helper function for nanosleep() (aka sleep() aka
 360  *      usleep()).
 361  *
 362  *      If there is less then one tick's worth of time left and
 363  *      we haven't done a yield, or the remaining microseconds is
 364  *      ridiculously low, do a yield.  This avoids having
 365  *      to deal with systimer overheads when the system is under
 366  *      heavy loads.  If we have done a yield already then use
 367  *      a systimer and an uninterruptable thread wait.
 368  *
 369  *      If there is more then a tick's worth of time left,
 370  *      calculate the baseline ticks and use an interruptable
 371  *      tsleep, then handle the fine-grained delay on the next
 372  *      loop.  This usually results in two sleeps occuring, a long one
 373  *      and a short one.
 374  *
 375  * MPSAFE
 376  */
 377 static void
 378 ns1_systimer(systimer_t info, int in_ipi __unused,
 379     struct intrframe *frame __unused)
 380 {
 381         lwkt_schedule(info->data);
 382 }
 383
 384 int
 385 nanosleep1(struct timespec *rqt, struct timespec *rmt)
 386 {
 387         static int nanowait;
 388         struct timespec ts, ts2, ts3;
 389         struct timeval tv;
 390         int error;
 391
 392         if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
 393                 return (EINVAL);
 394         /* XXX: imho this should return EINVAL at least for tv_sec < 0 */
 395         if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
 396                 return (0);
 397         nanouptime(&ts);
 398         timespecadd(&ts, rqt);          /* ts = target timestamp compare */
 399         TIMESPEC_TO_TIMEVAL(&tv, rqt);  /* tv = sleep interval */
 400
 401         for (;;) {
 402                 int ticks;
 403                 struct systimer info;
 404
 405                 ticks = tv.tv_usec / ustick;    /* approximate */
 406
 407                 if (tv.tv_sec == 0 && ticks == 0) {
 408                         thread_t td = curthread;
 409                         if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us)
 410                                 tv.tv_usec = nanosleep_min_us;
 411                         if (tv.tv_usec < nanosleep_hard_us) {
 412                                 lwkt_user_yield();
 413                                 cpu_pause();
 414                         } else {
 415                                 crit_enter_quick(td);
 416                                 systimer_init_oneshot(&info, ns1_systimer,
 417                                                 td, tv.tv_usec);
 418                                 lwkt_deschedule_self(td);
 419                                 crit_exit_quick(td);
 420                                 lwkt_switch();
 421                                 systimer_del(&info); /* make sure it's gone */
 422                         }
 423                         error = iscaught(td->td_lwp);
 424                 } else if (tv.tv_sec == 0) {
 425                         error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
 426                 } else {
 427                         ticks = tvtohz_low(&tv); /* also handles overflow */
 428                         error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
 429                 }
 430                 nanouptime(&ts2);
 431                 if (error && error != EWOULDBLOCK) {
 432                         if (error == ERESTART)
 433                                 error = EINTR;
 434                         if (rmt != NULL) {
 435                                 timespecsub(&ts, &ts2);
 436                                 if (ts.tv_sec < 0)
 437                                         timespecclear(&ts);
 438                                 *rmt = ts;
 439                         }
 440                         return (error);
 441                 }
 442                 if (timespeccmp(&ts2, &ts, >=))
 443                         return (0);
 444                 ts3 = ts;
 445                 timespecsub(&ts3, &ts2);
 446                 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
 447         }
 448 }
 449
 450 /*
 451  * MPSAFE
 452  */
 453 int
 454 sys_nanosleep(struct nanosleep_args *uap)
 455 {
 456         int error;
 457         struct timespec rqt;
 458         struct timespec rmt;
 459
 460         error = copyin(uap->rqtp, &rqt, sizeof(rqt));
 461         if (error)
 462                 return (error);
 463
 464         error = nanosleep1(&rqt, &rmt);
 465
 466         /*
 467          * copyout the residual if nanosleep was interrupted.
 468          */
 469         if (error && uap->rmtp) {
 470                 int error2;
 471
 472                 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
 473                 if (error2)
 474                         error = error2;
 475         }
 476         return (error);
 477 }
 478
 479 /*
 480  * The gettimeofday() system call is supposed to return a fine-grained
 481  * realtime stamp.  However, acquiring a fine-grained stamp can create a
 482  * bottleneck when multiple cpu cores are trying to accessing e.g. the
 483  * HPET hardware timer all at the same time, so we have a sysctl that
 484  * allows its behavior to be changed to a more coarse-grained timestamp
 485  * which does not have to access a hardware timer.
 486  */
 487 int
 488 sys_gettimeofday(struct gettimeofday_args *uap)
 489 {
 490         struct timeval atv;
 491         int error = 0;
 492
 493         if (uap->tp) {
 494                 if (gettimeofday_quick)
 495                         getmicrotime(&atv);
 496                 else
 497                         microtime(&atv);
 498                 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
 499                     sizeof (atv))))
 500                         return (error);
 501         }
 502         if (uap->tzp)
 503                 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
 504                     sizeof (tz));
 505         return (error);
 506 }
 507
 508 /*
 509  * MPALMOSTSAFE
 510  */
 511 int
 512 sys_settimeofday(struct settimeofday_args *uap)
 513 {
 514         struct thread *td = curthread;
 515         struct timeval atv;
 516         struct timezone atz;
 517         int error;
 518
 519         if ((error = priv_check(td, PRIV_SETTIMEOFDAY)))
 520                 return (error);
 521         /*
 522          * Verify all parameters before changing time.
 523          *
 524          * XXX: We do not allow the time to be set to 0.0, which also by
 525          *      happy coincidence works around a pkgsrc bulk build bug.
 526          */
 527         if (uap->tv) {
 528                 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
 529                     sizeof(atv))))
 530                         return (error);
 531                 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
 532                         return (EINVAL);
 533                 if (atv.tv_sec == 0 && atv.tv_usec == 0)
 534                         return (EINVAL);
 535         }
 536         if (uap->tzp &&
 537             (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
 538                 return (error);
 539
 540         lockmgr(&masterclock_lock, LK_EXCLUSIVE);
 541         if (uap->tv && (error = settime(&atv))) {
 542                 lockmgr(&masterclock_lock, LK_RELEASE);
 543                 return (error);
 544         }
 545         lockmgr(&masterclock_lock, LK_RELEASE);
 546
 547         if (uap->tzp)
 548                 tz = atz;
 549         return (0);
 550 }
 551
 552 /*
 553  * WARNING! Run with ntp_spin held
 554  */
 555 static void
 556 kern_adjtime_common(void)
 557 {
 558         if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) ||
 559             (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta))
 560                 ntp_tick_delta = ntp_delta;
 561         else if (ntp_delta > ntp_big_delta)
 562                 ntp_tick_delta = 10 * ntp_default_tick_delta;
 563         else if (ntp_delta < -ntp_big_delta)
 564                 ntp_tick_delta = -10 * ntp_default_tick_delta;
 565         else if (ntp_delta > 0)
 566                 ntp_tick_delta = ntp_default_tick_delta;
 567         else
 568                 ntp_tick_delta = -ntp_default_tick_delta;
 569 }
 570
 571 void
 572 kern_adjtime(int64_t delta, int64_t *odelta)
 573 {
 574         spin_lock(&ntp_spin);
 575         *odelta = ntp_delta;
 576         ntp_delta = delta;
 577         kern_adjtime_common();
 578         spin_unlock(&ntp_spin);
 579 }
 580
 581 static void
 582 kern_get_ntp_delta(int64_t *delta)
 583 {
 584         *delta = ntp_delta;
 585 }
 586
 587 void
 588 kern_reladjtime(int64_t delta)
 589 {
 590         spin_lock(&ntp_spin);
 591         ntp_delta += delta;
 592         kern_adjtime_common();
 593         spin_unlock(&ntp_spin);
 594 }
 595
 596 static void
 597 kern_adjfreq(int64_t rate)
 598 {
 599         spin_lock(&ntp_spin);
 600         ntp_tick_permanent = rate;
 601         spin_unlock(&ntp_spin);
 602 }
 603
 604 /*
 605  * MPALMOSTSAFE
 606  */
 607 int
 608 sys_adjtime(struct adjtime_args *uap)
 609 {
 610         struct thread *td = curthread;
 611         struct timeval atv;
 612         int64_t ndelta, odelta;
 613         int error;
 614
 615         if ((error = priv_check(td, PRIV_ADJTIME)))
 616                 return (error);
 617         error = copyin(uap->delta, &atv, sizeof(struct timeval));
 618         if (error)
 619                 return (error);
 620
 621         /*
 622          * Compute the total correction and the rate at which to apply it.
 623          * Round the adjustment down to a whole multiple of the per-tick
 624          * delta, so that after some number of incremental changes in
 625          * hardclock(), tickdelta will become zero, lest the correction
 626          * overshoot and start taking us away from the desired final time.
 627          */
 628         ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000;
 629         kern_adjtime(ndelta, &odelta);
 630
 631         if (uap->olddelta) {
 632                 atv.tv_sec = odelta / 1000000000;
 633                 atv.tv_usec = odelta % 1000000000 / 1000;
 634                 copyout(&atv, uap->olddelta, sizeof(struct timeval));
 635         }
 636         return (0);
 637 }
 638
 639 static int
 640 sysctl_adjtime(SYSCTL_HANDLER_ARGS)
 641 {
 642         int64_t delta;
 643         int error;
 644
 645         if (req->newptr != NULL) {
 646                 if (priv_check(curthread, PRIV_ROOT))
 647                         return (EPERM);
 648                 error = SYSCTL_IN(req, &delta, sizeof(delta));
 649                 if (error)
 650                         return (error);
 651                 kern_reladjtime(delta);
 652         }
 653
 654         if (req->oldptr)
 655                 kern_get_ntp_delta(&delta);
 656         error = SYSCTL_OUT(req, &delta, sizeof(delta));
 657         return (error);
 658 }
 659
 660 /*
 661  * delta is in nanoseconds.
 662  */
 663 static int
 664 sysctl_delta(SYSCTL_HANDLER_ARGS)
 665 {
 666         int64_t delta, old_delta;
 667         int error;
 668
 669         if (req->newptr != NULL) {
 670                 if (priv_check(curthread, PRIV_ROOT))
 671                         return (EPERM);
 672                 error = SYSCTL_IN(req, &delta, sizeof(delta));
 673                 if (error)
 674                         return (error);
 675                 kern_adjtime(delta, &old_delta);
 676         }
 677
 678         if (req->oldptr != NULL)
 679                 kern_get_ntp_delta(&old_delta);
 680         error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta));
 681         return (error);
 682 }
 683
 684 /*
 685  * frequency is in nanoseconds per second shifted left 32.
 686  * kern_adjfreq() needs it in nanoseconds per tick shifted left 32.
 687  */
 688 static int
 689 sysctl_adjfreq(SYSCTL_HANDLER_ARGS)
 690 {
 691         int64_t freqdelta;
 692         int error;
 693
 694         if (req->newptr != NULL) {
 695                 if (priv_check(curthread, PRIV_ROOT))
 696                         return (EPERM);
 697                 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta));
 698                 if (error)
 699                         return (error);
 700
 701                 freqdelta /= hz;
 702                 kern_adjfreq(freqdelta);
 703         }
 704
 705         if (req->oldptr != NULL)
 706                 freqdelta = ntp_tick_permanent * hz;
 707         error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta));
 708         if (error)
 709                 return (error);
 710
 711         return (0);
 712 }
 713
 714 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls");
 715 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent,
 716     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 717     sysctl_adjfreq, "Q", "permanent correction per second");
 718 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta,
 719     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 720     sysctl_delta, "Q", "one-time delta");
 721 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD,
 722     &ntp_big_delta, sizeof(ntp_big_delta), "Q",
 723     "threshold for fast adjustment");
 724 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD,
 725     &ntp_tick_delta, sizeof(ntp_tick_delta), "LU",
 726     "per-tick adjustment");
 727 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD,
 728     &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU",
 729     "default per-tick adjustment");
 730 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW,
 731     &ntp_leap_second, sizeof(ntp_leap_second), "LU",
 732     "next leap second");
 733 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW,
 734     &ntp_leap_insert, 0, "insert or remove leap second");
 735 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust,
 736     CTLTYPE_QUAD|CTLFLAG_RW, 0, 0,
 737     sysctl_adjtime, "Q", "relative adjust for delta");
 738
 739 /*
 740  * Get value of an interval timer.  The process virtual and
 741  * profiling virtual time timers are kept in the p_stats area, since
 742  * they can be swapped out.  These are kept internally in the
 743  * way they are specified externally: in time until they expire.
 744  *
 745  * The real time interval timer is kept in the process table slot
 746  * for the process, and its value (it_value) is kept as an
 747  * absolute time rather than as a delta, so that it is easy to keep
 748  * periodic real-time signals from drifting.
 749  *
 750  * Virtual time timers are processed in the hardclock() routine of
 751  * kern_clock.c.  The real time timer is processed by a timeout
 752  * routine, called from the softclock() routine.  Since a callout
 753  * may be delayed in real time due to interrupt processing in the system,
 754  * it is possible for the real time timeout routine (realitexpire, given below),
 755  * to be delayed in real time past when it is supposed to occur.  It
 756  * does not suffice, therefore, to reload the real timer .it_value from the
 757  * real time timers .it_interval.  Rather, we compute the next time in
 758  * absolute time the timer should go off.
 759  *
 760  * MPALMOSTSAFE
 761  */
 762 int
 763 sys_getitimer(struct getitimer_args *uap)
 764 {
 765         struct proc *p = curproc;
 766         struct timeval ctv;
 767         struct itimerval aitv;
 768
 769         if (uap->which > ITIMER_PROF)
 770                 return (EINVAL);
 771         lwkt_gettoken(&p->p_token);
 772         if (uap->which == ITIMER_REAL) {
 773                 /*
 774                  * Convert from absolute to relative time in .it_value
 775                  * part of real time timer.  If time for real time timer
 776                  * has passed return 0, else return difference between
 777                  * current time and time for the timer to go off.
 778                  */
 779                 aitv = p->p_realtimer;
 780                 if (timevalisset(&aitv.it_value)) {
 781                         getmicrouptime(&ctv);
 782                         if (timevalcmp(&aitv.it_value, &ctv, <))
 783                                 timevalclear(&aitv.it_value);
 784                         else
 785                                 timevalsub(&aitv.it_value, &ctv);
 786                 }
 787         } else {
 788                 aitv = p->p_timer[uap->which];
 789         }
 790         lwkt_reltoken(&p->p_token);
 791         return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
 792 }
 793
 794 /*
 795  * MPALMOSTSAFE
 796  */
 797 int
 798 sys_setitimer(struct setitimer_args *uap)
 799 {
 800         struct itimerval aitv;
 801         struct timeval ctv;
 802         struct itimerval *itvp;
 803         struct proc *p = curproc;
 804         int error;
 805
 806         if (uap->which > ITIMER_PROF)
 807                 return (EINVAL);
 808         itvp = uap->itv;
 809         if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
 810             sizeof(struct itimerval))))
 811                 return (error);
 812         if ((uap->itv = uap->oitv) &&
 813             (error = sys_getitimer((struct getitimer_args *)uap)))
 814                 return (error);
 815         if (itvp == NULL)
 816                 return (0);
 817         if (itimerfix(&aitv.it_value))
 818                 return (EINVAL);
 819         if (!timevalisset(&aitv.it_value))
 820                 timevalclear(&aitv.it_interval);
 821         else if (itimerfix(&aitv.it_interval))
 822                 return (EINVAL);
 823         lwkt_gettoken(&p->p_token);
 824         if (uap->which == ITIMER_REAL) {
 825                 if (timevalisset(&p->p_realtimer.it_value))
 826                         callout_stop_sync(&p->p_ithandle);
 827                 if (timevalisset(&aitv.it_value))
 828                         callout_reset(&p->p_ithandle,
 829                             tvtohz_high(&aitv.it_value), realitexpire, p);
 830                 getmicrouptime(&ctv);
 831                 timevaladd(&aitv.it_value, &ctv);
 832                 p->p_realtimer = aitv;
 833         } else {
 834                 p->p_timer[uap->which] = aitv;
 835                 switch(uap->which) {
 836                 case ITIMER_VIRTUAL:
 837                         p->p_flags &= ~P_SIGVTALRM;
 838                         break;
 839                 case ITIMER_PROF:
 840                         p->p_flags &= ~P_SIGPROF;
 841                         break;
 842                 }
 843         }
 844         lwkt_reltoken(&p->p_token);
 845         return (0);
 846 }
 847
 848 /*
 849  * Real interval timer expired:
 850  * send process whose timer expired an alarm signal.
 851  * If time is not set up to reload, then just return.
 852  * Else compute next time timer should go off which is > current time.
 853  * This is where delay in processing this timeout causes multiple
 854  * SIGALRM calls to be compressed into one.
 855  * tvtohz_high() always adds 1 to allow for the time until the next clock
 856  * interrupt being strictly less than 1 clock tick, but we don't want
 857  * that here since we want to appear to be in sync with the clock
 858  * interrupt even when we're delayed.
 859  */
 860 static
 861 void
 862 realitexpire(void *arg)
 863 {
 864         struct proc *p;
 865         struct timeval ctv, ntv;
 866
 867         p = (struct proc *)arg;
 868         PHOLD(p);
 869         lwkt_gettoken(&p->p_token);
 870         ksignal(p, SIGALRM);
 871         if (!timevalisset(&p->p_realtimer.it_interval)) {
 872                 timevalclear(&p->p_realtimer.it_value);
 873                 goto done;
 874         }
 875         for (;;) {
 876                 timevaladd(&p->p_realtimer.it_value,
 877                            &p->p_realtimer.it_interval);
 878                 getmicrouptime(&ctv);
 879                 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
 880                         ntv = p->p_realtimer.it_value;
 881                         timevalsub(&ntv, &ctv);
 882                         callout_reset(&p->p_ithandle, tvtohz_low(&ntv),
 883                                       realitexpire, p);
 884                         goto done;
 885                 }
 886         }
 887 done:
 888         lwkt_reltoken(&p->p_token);
 889         PRELE(p);
 890 }
 891
 892 /*
 893  * Used to validate itimer timeouts and utimes*() timespecs.
 894  */
 895 int
 896 itimerfix(struct timeval *tv)
 897 {
 898         if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
 899                 return (EINVAL);
 900         if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick)
 901                 tv->tv_usec = ustick;
 902         return (0);
 903 }
 904
 905 /*
 906  * Used to validate timeouts and utimes*() timespecs.
 907  */
 908 int
 909 itimespecfix(struct timespec *ts)
 910 {
 911         if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000ULL)
 912                 return (EINVAL);
 913         if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < nstick)
 914                 ts->tv_nsec = nstick;
 915         return (0);
 916 }
 917
 918 /*
 919  * Decrement an interval timer by a specified number
 920  * of microseconds, which must be less than a second,
 921  * i.e. < 1000000.  If the timer expires, then reload
 922  * it.  In this case, carry over (usec - old value) to
 923  * reduce the value reloaded into the timer so that
 924  * the timer does not drift.  This routine assumes
 925  * that it is called in a context where the timers
 926  * on which it is operating cannot change in value.
 927  */
 928 int
 929 itimerdecr(struct itimerval *itp, int usec)
 930 {
 931
 932         if (itp->it_value.tv_usec < usec) {
 933                 if (itp->it_value.tv_sec == 0) {
 934                         /* expired, and already in next interval */
 935                         usec -= itp->it_value.tv_usec;
 936                         goto expire;
 937                 }
 938                 itp->it_value.tv_usec += 1000000;
 939                 itp->it_value.tv_sec--;
 940         }
 941         itp->it_value.tv_usec -= usec;
 942         usec = 0;
 943         if (timevalisset(&itp->it_value))
 944                 return (1);
 945         /* expired, exactly at end of interval */
 946 expire:
 947         if (timevalisset(&itp->it_interval)) {
 948                 itp->it_value = itp->it_interval;
 949                 itp->it_value.tv_usec -= usec;
 950                 if (itp->it_value.tv_usec < 0) {
 951                         itp->it_value.tv_usec += 1000000;
 952                         itp->it_value.tv_sec--;
 953                 }
 954         } else
 955                 itp->it_value.tv_usec = 0;              /* sec is already 0 */
 956         return (0);
 957 }
 958
 959 /*
 960  * Add and subtract routines for timevals.
 961  * N.B.: subtract routine doesn't deal with
 962  * results which are before the beginning,
 963  * it just gets very confused in this case.
 964  * Caveat emptor.
 965  */
 966 void
 967 timevaladd(struct timeval *t1, const struct timeval *t2)
 968 {
 969
 970         t1->tv_sec += t2->tv_sec;
 971         t1->tv_usec += t2->tv_usec;
 972         timevalfix(t1);
 973 }
 974
 975 void
 976 timevalsub(struct timeval *t1, const struct timeval *t2)
 977 {
 978
 979         t1->tv_sec -= t2->tv_sec;
 980         t1->tv_usec -= t2->tv_usec;
 981         timevalfix(t1);
 982 }
 983
 984 static void
 985 timevalfix(struct timeval *t1)
 986 {
 987
 988         if (t1->tv_usec < 0) {
 989                 t1->tv_sec--;
 990                 t1->tv_usec += 1000000;
 991         }
 992         if (t1->tv_usec >= 1000000) {
 993                 t1->tv_sec++;
 994                 t1->tv_usec -= 1000000;
 995         }
 996 }
 997
 998 /*
 999  * ratecheck(): simple time-based rate-limit checking.
1000  */
1001 int
1002 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1003 {
1004         struct timeval tv, delta;
1005         int rv = 0;
1006
1007         getmicrouptime(&tv);            /* NB: 10ms precision */
1008         delta = tv;
1009         timevalsub(&delta, lasttime);
1010
1011         /*
1012          * check for 0,0 is so that the message will be seen at least once,
1013          * even if interval is huge.
1014          */
1015         if (timevalcmp(&delta, mininterval, >=) ||
1016             (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1017                 *lasttime = tv;
1018                 rv = 1;
1019         }
1020
1021         return (rv);
1022 }
1023
1024 /*
1025  * ppsratecheck(): packets (or events) per second limitation.
1026  *
1027  * Return 0 if the limit is to be enforced (e.g. the caller
1028  * should drop a packet because of the rate limitation).
1029  *
1030  * maxpps of 0 always causes zero to be returned.  maxpps of -1
1031  * always causes 1 to be returned; this effectively defeats rate
1032  * limiting.
1033  *
1034  * Note that we maintain the struct timeval for compatibility
1035  * with other bsd systems.  We reuse the storage and just monitor
1036  * clock ticks for minimal overhead.
1037  */
1038 int
1039 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
1040 {
1041         int now;
1042
1043         /*
1044          * Reset the last time and counter if this is the first call
1045          * or more than a second has passed since the last update of
1046          * lasttime.
1047          */
1048         now = ticks;
1049         if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1050                 lasttime->tv_sec = now;
1051                 *curpps = 1;
1052                 return (maxpps != 0);
1053         } else {
1054                 (*curpps)++;            /* NB: ignore potential overflow */
1055                 return (maxpps < 0 || *curpps < maxpps);
1056         }
1057 }