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/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 case CLOCK_PROCESS_CPUTIME_ID: 208 p = curproc; 209 ats->tv_sec = p->p_timer[ITIMER_PROF].it_value.tv_sec; 210 ats->tv_nsec = p->p_timer[ITIMER_PROF].it_value.tv_usec * 211 1000; 212 break; 213 case CLOCK_SECOND: 214 ats->tv_sec = time_second; 215 ats->tv_nsec = 0; 216 break; 217 case CLOCK_THREAD_CPUTIME_ID: 218 get_curthread_cputime(ats); 219 break; 220 default: 221 error = EINVAL; 222 break; 223 } 224 return (error); 225 } 226 227 /* 228 * MPSAFE 229 */ 230 int 231 sys_clock_gettime(struct clock_gettime_args *uap) 232 { 233 struct timespec ats; 234 int error; 235 236 error = kern_clock_gettime(uap->clock_id, &ats); 237 if (error == 0) 238 error = copyout(&ats, uap->tp, sizeof(ats)); 239 240 return (error); 241 } 242 243 int 244 kern_clock_settime(clockid_t clock_id, struct timespec *ats) 245 { 246 struct thread *td = curthread; 247 struct timeval atv; 248 int error; 249 250 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 251 return (error); 252 if (clock_id != CLOCK_REALTIME) 253 return (EINVAL); 254 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000) 255 return (EINVAL); 256 257 TIMESPEC_TO_TIMEVAL(&atv, ats); 258 error = settime(&atv); 259 return (error); 260 } 261 262 /* 263 * MPALMOSTSAFE 264 */ 265 int 266 sys_clock_settime(struct clock_settime_args *uap) 267 { 268 struct timespec ats; 269 int error; 270 271 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 272 return (error); 273 274 get_mplock(); 275 error = kern_clock_settime(uap->clock_id, &ats); 276 rel_mplock(); 277 return (error); 278 } 279 280 /* 281 * MPSAFE 282 */ 283 int 284 kern_clock_getres(clockid_t clock_id, struct timespec *ts) 285 { 286 int error; 287 288 switch(clock_id) { 289 case CLOCK_REALTIME: 290 case CLOCK_REALTIME_FAST: 291 case CLOCK_REALTIME_PRECISE: 292 case CLOCK_MONOTONIC: 293 case CLOCK_MONOTONIC_FAST: 294 case CLOCK_MONOTONIC_PRECISE: 295 case CLOCK_UPTIME: 296 case CLOCK_UPTIME_FAST: 297 case CLOCK_UPTIME_PRECISE: 298 case CLOCK_THREAD_CPUTIME_ID: 299 case CLOCK_PROCESS_CPUTIME_ID: 300 /* 301 * Round up the result of the division cheaply 302 * by adding 1. Rounding up is especially important 303 * if rounding down would give 0. Perfect rounding 304 * is unimportant. 305 */ 306 ts->tv_sec = 0; 307 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1; 308 error = 0; 309 break; 310 case CLOCK_VIRTUAL: 311 case CLOCK_PROF: 312 /* Accurately round up here because we can do so cheaply. */ 313 ts->tv_sec = 0; 314 ts->tv_nsec = (1000000000 + hz - 1) / hz; 315 error = 0; 316 break; 317 case CLOCK_SECOND: 318 ts->tv_sec = 1; 319 ts->tv_nsec = 0; 320 error = 0; 321 break; 322 default: 323 error = EINVAL; 324 break; 325 } 326 327 return(error); 328 } 329 330 /* 331 * MPSAFE 332 */ 333 int 334 sys_clock_getres(struct clock_getres_args *uap) 335 { 336 int error; 337 struct timespec ts; 338 339 error = kern_clock_getres(uap->clock_id, &ts); 340 if (error == 0) 341 error = copyout(&ts, uap->tp, sizeof(ts)); 342 343 return (error); 344 } 345 346 /* 347 * nanosleep1() 348 * 349 * This is a general helper function for nanosleep() (aka sleep() aka 350 * usleep()). 351 * 352 * If there is less then one tick's worth of time left and 353 * we haven't done a yield, or the remaining microseconds is 354 * ridiculously low, do a yield. This avoids having 355 * to deal with systimer overheads when the system is under 356 * heavy loads. If we have done a yield already then use 357 * a systimer and an uninterruptable thread wait. 358 * 359 * If there is more then a tick's worth of time left, 360 * calculate the baseline ticks and use an interruptable 361 * tsleep, then handle the fine-grained delay on the next 362 * loop. This usually results in two sleeps occuring, a long one 363 * and a short one. 364 * 365 * MPSAFE 366 */ 367 static void 368 ns1_systimer(systimer_t info, int in_ipi __unused, 369 struct intrframe *frame __unused) 370 { 371 lwkt_schedule(info->data); 372 } 373 374 int 375 nanosleep1(struct timespec *rqt, struct timespec *rmt) 376 { 377 static int nanowait; 378 struct timespec ts, ts2, ts3; 379 struct timeval tv; 380 int error; 381 382 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 383 return (EINVAL); 384 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 385 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 386 return (0); 387 nanouptime(&ts); 388 timespecadd(&ts, rqt); /* ts = target timestamp compare */ 389 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 390 391 for (;;) { 392 int ticks; 393 struct systimer info; 394 395 ticks = tv.tv_usec / ustick; /* approximate */ 396 397 if (tv.tv_sec == 0 && ticks == 0) { 398 thread_t td = curthread; 399 if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us) 400 tv.tv_usec = nanosleep_min_us; 401 if (tv.tv_usec < nanosleep_hard_us) { 402 lwkt_user_yield(); 403 cpu_pause(); 404 } else { 405 crit_enter_quick(td); 406 systimer_init_oneshot(&info, ns1_systimer, 407 td, tv.tv_usec); 408 lwkt_deschedule_self(td); 409 crit_exit_quick(td); 410 lwkt_switch(); 411 systimer_del(&info); /* make sure it's gone */ 412 } 413 error = iscaught(td->td_lwp); 414 } else if (tv.tv_sec == 0) { 415 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 416 } else { 417 ticks = tvtohz_low(&tv); /* also handles overflow */ 418 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 419 } 420 nanouptime(&ts2); 421 if (error && error != EWOULDBLOCK) { 422 if (error == ERESTART) 423 error = EINTR; 424 if (rmt != NULL) { 425 timespecsub(&ts, &ts2); 426 if (ts.tv_sec < 0) 427 timespecclear(&ts); 428 *rmt = ts; 429 } 430 return (error); 431 } 432 if (timespeccmp(&ts2, &ts, >=)) 433 return (0); 434 ts3 = ts; 435 timespecsub(&ts3, &ts2); 436 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 437 } 438 } 439 440 /* 441 * MPSAFE 442 */ 443 int 444 sys_nanosleep(struct nanosleep_args *uap) 445 { 446 int error; 447 struct timespec rqt; 448 struct timespec rmt; 449 450 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 451 if (error) 452 return (error); 453 454 error = nanosleep1(&rqt, &rmt); 455 456 /* 457 * copyout the residual if nanosleep was interrupted. 458 */ 459 if (error && uap->rmtp) { 460 int error2; 461 462 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 463 if (error2) 464 error = error2; 465 } 466 return (error); 467 } 468 469 /* 470 * The gettimeofday() system call is supposed to return a fine-grained 471 * realtime stamp. However, acquiring a fine-grained stamp can create a 472 * bottleneck when multiple cpu cores are trying to accessing e.g. the 473 * HPET hardware timer all at the same time, so we have a sysctl that 474 * allows its behavior to be changed to a more coarse-grained timestamp 475 * which does not have to access a hardware timer. 476 */ 477 int 478 sys_gettimeofday(struct gettimeofday_args *uap) 479 { 480 struct timeval atv; 481 int error = 0; 482 483 if (uap->tp) { 484 if (gettimeofday_quick) 485 getmicrotime(&atv); 486 else 487 microtime(&atv); 488 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 489 sizeof (atv)))) 490 return (error); 491 } 492 if (uap->tzp) 493 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 494 sizeof (tz)); 495 return (error); 496 } 497 498 /* 499 * MPALMOSTSAFE 500 */ 501 int 502 sys_settimeofday(struct settimeofday_args *uap) 503 { 504 struct thread *td = curthread; 505 struct timeval atv; 506 struct timezone atz; 507 int error; 508 509 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 510 return (error); 511 /* 512 * Verify all parameters before changing time. 513 * 514 * XXX: We do not allow the time to be set to 0.0, which also by 515 * happy coincidence works around a pkgsrc bulk build bug. 516 */ 517 if (uap->tv) { 518 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 519 sizeof(atv)))) 520 return (error); 521 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 522 return (EINVAL); 523 if (atv.tv_sec == 0 && atv.tv_usec == 0) 524 return (EINVAL); 525 } 526 if (uap->tzp && 527 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 528 return (error); 529 530 get_mplock(); 531 if (uap->tv && (error = settime(&atv))) { 532 rel_mplock(); 533 return (error); 534 } 535 rel_mplock(); 536 if (uap->tzp) 537 tz = atz; 538 return (0); 539 } 540 541 static void 542 kern_adjtime_common(void) 543 { 544 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 545 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 546 ntp_tick_delta = ntp_delta; 547 else if (ntp_delta > ntp_big_delta) 548 ntp_tick_delta = 10 * ntp_default_tick_delta; 549 else if (ntp_delta < -ntp_big_delta) 550 ntp_tick_delta = -10 * ntp_default_tick_delta; 551 else if (ntp_delta > 0) 552 ntp_tick_delta = ntp_default_tick_delta; 553 else 554 ntp_tick_delta = -ntp_default_tick_delta; 555 } 556 557 void 558 kern_adjtime(int64_t delta, int64_t *odelta) 559 { 560 int origcpu; 561 562 if ((origcpu = mycpu->gd_cpuid) != 0) 563 lwkt_setcpu_self(globaldata_find(0)); 564 565 crit_enter(); 566 *odelta = ntp_delta; 567 ntp_delta = delta; 568 kern_adjtime_common(); 569 crit_exit(); 570 571 if (origcpu != 0) 572 lwkt_setcpu_self(globaldata_find(origcpu)); 573 } 574 575 static void 576 kern_get_ntp_delta(int64_t *delta) 577 { 578 int origcpu; 579 580 if ((origcpu = mycpu->gd_cpuid) != 0) 581 lwkt_setcpu_self(globaldata_find(0)); 582 583 crit_enter(); 584 *delta = ntp_delta; 585 crit_exit(); 586 587 if (origcpu != 0) 588 lwkt_setcpu_self(globaldata_find(origcpu)); 589 } 590 591 void 592 kern_reladjtime(int64_t delta) 593 { 594 int origcpu; 595 596 if ((origcpu = mycpu->gd_cpuid) != 0) 597 lwkt_setcpu_self(globaldata_find(0)); 598 599 crit_enter(); 600 ntp_delta += delta; 601 kern_adjtime_common(); 602 crit_exit(); 603 604 if (origcpu != 0) 605 lwkt_setcpu_self(globaldata_find(origcpu)); 606 } 607 608 static void 609 kern_adjfreq(int64_t rate) 610 { 611 int origcpu; 612 613 if ((origcpu = mycpu->gd_cpuid) != 0) 614 lwkt_setcpu_self(globaldata_find(0)); 615 616 crit_enter(); 617 ntp_tick_permanent = rate; 618 crit_exit(); 619 620 if (origcpu != 0) 621 lwkt_setcpu_self(globaldata_find(origcpu)); 622 } 623 624 /* 625 * MPALMOSTSAFE 626 */ 627 int 628 sys_adjtime(struct adjtime_args *uap) 629 { 630 struct thread *td = curthread; 631 struct timeval atv; 632 int64_t ndelta, odelta; 633 int error; 634 635 if ((error = priv_check(td, PRIV_ADJTIME))) 636 return (error); 637 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 638 if (error) 639 return (error); 640 641 /* 642 * Compute the total correction and the rate at which to apply it. 643 * Round the adjustment down to a whole multiple of the per-tick 644 * delta, so that after some number of incremental changes in 645 * hardclock(), tickdelta will become zero, lest the correction 646 * overshoot and start taking us away from the desired final time. 647 */ 648 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 649 get_mplock(); 650 kern_adjtime(ndelta, &odelta); 651 rel_mplock(); 652 653 if (uap->olddelta) { 654 atv.tv_sec = odelta / 1000000000; 655 atv.tv_usec = odelta % 1000000000 / 1000; 656 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 657 } 658 return (0); 659 } 660 661 static int 662 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 663 { 664 int64_t delta; 665 int error; 666 667 if (req->newptr != NULL) { 668 if (priv_check(curthread, PRIV_ROOT)) 669 return (EPERM); 670 error = SYSCTL_IN(req, &delta, sizeof(delta)); 671 if (error) 672 return (error); 673 kern_reladjtime(delta); 674 } 675 676 if (req->oldptr) 677 kern_get_ntp_delta(&delta); 678 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 679 return (error); 680 } 681 682 /* 683 * delta is in nanoseconds. 684 */ 685 static int 686 sysctl_delta(SYSCTL_HANDLER_ARGS) 687 { 688 int64_t delta, old_delta; 689 int error; 690 691 if (req->newptr != NULL) { 692 if (priv_check(curthread, PRIV_ROOT)) 693 return (EPERM); 694 error = SYSCTL_IN(req, &delta, sizeof(delta)); 695 if (error) 696 return (error); 697 kern_adjtime(delta, &old_delta); 698 } 699 700 if (req->oldptr != NULL) 701 kern_get_ntp_delta(&old_delta); 702 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 703 return (error); 704 } 705 706 /* 707 * frequency is in nanoseconds per second shifted left 32. 708 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 709 */ 710 static int 711 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 712 { 713 int64_t freqdelta; 714 int error; 715 716 if (req->newptr != NULL) { 717 if (priv_check(curthread, PRIV_ROOT)) 718 return (EPERM); 719 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 720 if (error) 721 return (error); 722 723 freqdelta /= hz; 724 kern_adjfreq(freqdelta); 725 } 726 727 if (req->oldptr != NULL) 728 freqdelta = ntp_tick_permanent * hz; 729 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 730 if (error) 731 return (error); 732 733 return (0); 734 } 735 736 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 737 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 738 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 739 sysctl_adjfreq, "Q", "permanent correction per second"); 740 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 741 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 742 sysctl_delta, "Q", "one-time delta"); 743 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 744 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 745 "threshold for fast adjustment"); 746 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 747 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 748 "per-tick adjustment"); 749 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 750 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 751 "default per-tick adjustment"); 752 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 753 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 754 "next leap second"); 755 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 756 &ntp_leap_insert, 0, "insert or remove leap second"); 757 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 758 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 759 sysctl_adjtime, "Q", "relative adjust for delta"); 760 761 /* 762 * Get value of an interval timer. The process virtual and 763 * profiling virtual time timers are kept in the p_stats area, since 764 * they can be swapped out. These are kept internally in the 765 * way they are specified externally: in time until they expire. 766 * 767 * The real time interval timer is kept in the process table slot 768 * for the process, and its value (it_value) is kept as an 769 * absolute time rather than as a delta, so that it is easy to keep 770 * periodic real-time signals from drifting. 771 * 772 * Virtual time timers are processed in the hardclock() routine of 773 * kern_clock.c. The real time timer is processed by a timeout 774 * routine, called from the softclock() routine. Since a callout 775 * may be delayed in real time due to interrupt processing in the system, 776 * it is possible for the real time timeout routine (realitexpire, given below), 777 * to be delayed in real time past when it is supposed to occur. It 778 * does not suffice, therefore, to reload the real timer .it_value from the 779 * real time timers .it_interval. Rather, we compute the next time in 780 * absolute time the timer should go off. 781 * 782 * MPALMOSTSAFE 783 */ 784 int 785 sys_getitimer(struct getitimer_args *uap) 786 { 787 struct proc *p = curproc; 788 struct timeval ctv; 789 struct itimerval aitv; 790 791 if (uap->which > ITIMER_PROF) 792 return (EINVAL); 793 lwkt_gettoken(&p->p_token); 794 if (uap->which == ITIMER_REAL) { 795 /* 796 * Convert from absolute to relative time in .it_value 797 * part of real time timer. If time for real time timer 798 * has passed return 0, else return difference between 799 * current time and time for the timer to go off. 800 */ 801 aitv = p->p_realtimer; 802 if (timevalisset(&aitv.it_value)) { 803 getmicrouptime(&ctv); 804 if (timevalcmp(&aitv.it_value, &ctv, <)) 805 timevalclear(&aitv.it_value); 806 else 807 timevalsub(&aitv.it_value, &ctv); 808 } 809 } else { 810 aitv = p->p_timer[uap->which]; 811 } 812 lwkt_reltoken(&p->p_token); 813 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 814 } 815 816 /* 817 * MPALMOSTSAFE 818 */ 819 int 820 sys_setitimer(struct setitimer_args *uap) 821 { 822 struct itimerval aitv; 823 struct timeval ctv; 824 struct itimerval *itvp; 825 struct proc *p = curproc; 826 int error; 827 828 if (uap->which > ITIMER_PROF) 829 return (EINVAL); 830 itvp = uap->itv; 831 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 832 sizeof(struct itimerval)))) 833 return (error); 834 if ((uap->itv = uap->oitv) && 835 (error = sys_getitimer((struct getitimer_args *)uap))) 836 return (error); 837 if (itvp == NULL) 838 return (0); 839 if (itimerfix(&aitv.it_value)) 840 return (EINVAL); 841 if (!timevalisset(&aitv.it_value)) 842 timevalclear(&aitv.it_interval); 843 else if (itimerfix(&aitv.it_interval)) 844 return (EINVAL); 845 lwkt_gettoken(&p->p_token); 846 if (uap->which == ITIMER_REAL) { 847 if (timevalisset(&p->p_realtimer.it_value)) 848 callout_stop_sync(&p->p_ithandle); 849 if (timevalisset(&aitv.it_value)) 850 callout_reset(&p->p_ithandle, 851 tvtohz_high(&aitv.it_value), realitexpire, p); 852 getmicrouptime(&ctv); 853 timevaladd(&aitv.it_value, &ctv); 854 p->p_realtimer = aitv; 855 } else { 856 p->p_timer[uap->which] = aitv; 857 switch(uap->which) { 858 case ITIMER_VIRTUAL: 859 p->p_flags &= ~P_SIGVTALRM; 860 break; 861 case ITIMER_PROF: 862 p->p_flags &= ~P_SIGPROF; 863 break; 864 } 865 } 866 lwkt_reltoken(&p->p_token); 867 return (0); 868 } 869 870 /* 871 * Real interval timer expired: 872 * send process whose timer expired an alarm signal. 873 * If time is not set up to reload, then just return. 874 * Else compute next time timer should go off which is > current time. 875 * This is where delay in processing this timeout causes multiple 876 * SIGALRM calls to be compressed into one. 877 * tvtohz_high() always adds 1 to allow for the time until the next clock 878 * interrupt being strictly less than 1 clock tick, but we don't want 879 * that here since we want to appear to be in sync with the clock 880 * interrupt even when we're delayed. 881 */ 882 void 883 realitexpire(void *arg) 884 { 885 struct proc *p; 886 struct timeval ctv, ntv; 887 888 p = (struct proc *)arg; 889 PHOLD(p); 890 lwkt_gettoken(&p->p_token); 891 ksignal(p, SIGALRM); 892 if (!timevalisset(&p->p_realtimer.it_interval)) { 893 timevalclear(&p->p_realtimer.it_value); 894 goto done; 895 } 896 for (;;) { 897 timevaladd(&p->p_realtimer.it_value, 898 &p->p_realtimer.it_interval); 899 getmicrouptime(&ctv); 900 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 901 ntv = p->p_realtimer.it_value; 902 timevalsub(&ntv, &ctv); 903 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 904 realitexpire, p); 905 goto done; 906 } 907 } 908 done: 909 lwkt_reltoken(&p->p_token); 910 PRELE(p); 911 } 912 913 /* 914 * Used to validate itimer timeouts and utimes*() timespecs. 915 */ 916 int 917 itimerfix(struct timeval *tv) 918 { 919 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 920 return (EINVAL); 921 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 922 tv->tv_usec = ustick; 923 return (0); 924 } 925 926 /* 927 * Used to validate timeouts and utimes*() timespecs. 928 */ 929 int 930 itimespecfix(struct timespec *ts) 931 { 932 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000ULL) 933 return (EINVAL); 934 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < nstick) 935 ts->tv_nsec = nstick; 936 return (0); 937 } 938 939 /* 940 * Decrement an interval timer by a specified number 941 * of microseconds, which must be less than a second, 942 * i.e. < 1000000. If the timer expires, then reload 943 * it. In this case, carry over (usec - old value) to 944 * reduce the value reloaded into the timer so that 945 * the timer does not drift. This routine assumes 946 * that it is called in a context where the timers 947 * on which it is operating cannot change in value. 948 */ 949 int 950 itimerdecr(struct itimerval *itp, int usec) 951 { 952 953 if (itp->it_value.tv_usec < usec) { 954 if (itp->it_value.tv_sec == 0) { 955 /* expired, and already in next interval */ 956 usec -= itp->it_value.tv_usec; 957 goto expire; 958 } 959 itp->it_value.tv_usec += 1000000; 960 itp->it_value.tv_sec--; 961 } 962 itp->it_value.tv_usec -= usec; 963 usec = 0; 964 if (timevalisset(&itp->it_value)) 965 return (1); 966 /* expired, exactly at end of interval */ 967 expire: 968 if (timevalisset(&itp->it_interval)) { 969 itp->it_value = itp->it_interval; 970 itp->it_value.tv_usec -= usec; 971 if (itp->it_value.tv_usec < 0) { 972 itp->it_value.tv_usec += 1000000; 973 itp->it_value.tv_sec--; 974 } 975 } else 976 itp->it_value.tv_usec = 0; /* sec is already 0 */ 977 return (0); 978 } 979 980 /* 981 * Add and subtract routines for timevals. 982 * N.B.: subtract routine doesn't deal with 983 * results which are before the beginning, 984 * it just gets very confused in this case. 985 * Caveat emptor. 986 */ 987 void 988 timevaladd(struct timeval *t1, const struct timeval *t2) 989 { 990 991 t1->tv_sec += t2->tv_sec; 992 t1->tv_usec += t2->tv_usec; 993 timevalfix(t1); 994 } 995 996 void 997 timevalsub(struct timeval *t1, const struct timeval *t2) 998 { 999 1000 t1->tv_sec -= t2->tv_sec; 1001 t1->tv_usec -= t2->tv_usec; 1002 timevalfix(t1); 1003 } 1004 1005 static void 1006 timevalfix(struct timeval *t1) 1007 { 1008 1009 if (t1->tv_usec < 0) { 1010 t1->tv_sec--; 1011 t1->tv_usec += 1000000; 1012 } 1013 if (t1->tv_usec >= 1000000) { 1014 t1->tv_sec++; 1015 t1->tv_usec -= 1000000; 1016 } 1017 } 1018 1019 /* 1020 * ratecheck(): simple time-based rate-limit checking. 1021 */ 1022 int 1023 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1024 { 1025 struct timeval tv, delta; 1026 int rv = 0; 1027 1028 getmicrouptime(&tv); /* NB: 10ms precision */ 1029 delta = tv; 1030 timevalsub(&delta, lasttime); 1031 1032 /* 1033 * check for 0,0 is so that the message will be seen at least once, 1034 * even if interval is huge. 1035 */ 1036 if (timevalcmp(&delta, mininterval, >=) || 1037 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1038 *lasttime = tv; 1039 rv = 1; 1040 } 1041 1042 return (rv); 1043 } 1044 1045 /* 1046 * ppsratecheck(): packets (or events) per second limitation. 1047 * 1048 * Return 0 if the limit is to be enforced (e.g. the caller 1049 * should drop a packet because of the rate limitation). 1050 * 1051 * maxpps of 0 always causes zero to be returned. maxpps of -1 1052 * always causes 1 to be returned; this effectively defeats rate 1053 * limiting. 1054 * 1055 * Note that we maintain the struct timeval for compatibility 1056 * with other bsd systems. We reuse the storage and just monitor 1057 * clock ticks for minimal overhead. 1058 */ 1059 int 1060 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1061 { 1062 int now; 1063 1064 /* 1065 * Reset the last time and counter if this is the first call 1066 * or more than a second has passed since the last update of 1067 * lasttime. 1068 */ 1069 now = ticks; 1070 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1071 lasttime->tv_sec = now; 1072 *curpps = 1; 1073 return (maxpps != 0); 1074 } else { 1075 (*curpps)++; /* NB: ignore potential overflow */ 1076 return (maxpps < 0 || *curpps < maxpps); 1077 } 1078 } 1079