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. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93 34 * $FreeBSD: src/sys/kern/kern_time.c,v 1.68.2.1 2002/10/01 08:00:41 bde Exp $ 35 */ 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/buf.h> 40 #include <sys/sysproto.h> 41 #include <sys/resourcevar.h> 42 #include <sys/signalvar.h> 43 #include <sys/kernel.h> 44 #include <sys/sysent.h> 45 #include <sys/sysunion.h> 46 #include <sys/proc.h> 47 #include <sys/priv.h> 48 #include <sys/time.h> 49 #include <sys/vnode.h> 50 #include <sys/sysctl.h> 51 #include <sys/kern_syscall.h> 52 #include <vm/vm.h> 53 #include <vm/vm_extern.h> 54 55 #include <sys/msgport2.h> 56 #include <sys/thread2.h> 57 #include <sys/mplock2.h> 58 59 struct timezone tz; 60 61 /* 62 * Time of day and interval timer support. 63 * 64 * These routines provide the kernel entry points to get and set 65 * the time-of-day and per-process interval timers. Subroutines 66 * here provide support for adding and subtracting timeval structures 67 * and decrementing interval timers, optionally reloading the interval 68 * timers when they expire. 69 */ 70 71 static int settime(struct timeval *); 72 static void timevalfix(struct timeval *); 73 74 /* 75 * Nanosleep tries very hard to sleep for a precisely requested time 76 * interval, down to 1uS. The administrator can impose a minimum delay 77 * and a delay below which we hard-loop instead of initiate a timer 78 * interrupt and sleep. 79 * 80 * For machines under high loads it might be beneficial to increase min_us 81 * to e.g. 1000uS (1ms) so spining processes sleep meaningfully. 82 */ 83 static int nanosleep_min_us = 10; 84 static int nanosleep_hard_us = 100; 85 static int gettimeofday_quick = 0; 86 SYSCTL_INT(_kern, OID_AUTO, nanosleep_min_us, CTLFLAG_RW, 87 &nanosleep_min_us, 0, "") 88 SYSCTL_INT(_kern, OID_AUTO, nanosleep_hard_us, CTLFLAG_RW, 89 &nanosleep_hard_us, 0, "") 90 SYSCTL_INT(_kern, OID_AUTO, gettimeofday_quick, CTLFLAG_RW, 91 &gettimeofday_quick, 0, "") 92 93 static int 94 settime(struct timeval *tv) 95 { 96 struct timeval delta, tv1, tv2; 97 static struct timeval maxtime, laststep; 98 struct timespec ts; 99 int origcpu; 100 101 if ((origcpu = mycpu->gd_cpuid) != 0) 102 lwkt_setcpu_self(globaldata_find(0)); 103 104 crit_enter(); 105 microtime(&tv1); 106 delta = *tv; 107 timevalsub(&delta, &tv1); 108 109 /* 110 * If the system is secure, we do not allow the time to be 111 * set to a value earlier than 1 second less than the highest 112 * time we have yet seen. The worst a miscreant can do in 113 * this circumstance is "freeze" time. He couldn't go 114 * back to the past. 115 * 116 * We similarly do not allow the clock to be stepped more 117 * than one second, nor more than once per second. This allows 118 * a miscreant to make the clock march double-time, but no worse. 119 */ 120 if (securelevel > 1) { 121 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 122 /* 123 * Update maxtime to latest time we've seen. 124 */ 125 if (tv1.tv_sec > maxtime.tv_sec) 126 maxtime = tv1; 127 tv2 = *tv; 128 timevalsub(&tv2, &maxtime); 129 if (tv2.tv_sec < -1) { 130 tv->tv_sec = maxtime.tv_sec - 1; 131 kprintf("Time adjustment clamped to -1 second\n"); 132 } 133 } else { 134 if (tv1.tv_sec == laststep.tv_sec) { 135 crit_exit(); 136 return (EPERM); 137 } 138 if (delta.tv_sec > 1) { 139 tv->tv_sec = tv1.tv_sec + 1; 140 kprintf("Time adjustment clamped to +1 second\n"); 141 } 142 laststep = *tv; 143 } 144 } 145 146 ts.tv_sec = tv->tv_sec; 147 ts.tv_nsec = tv->tv_usec * 1000; 148 set_timeofday(&ts); 149 crit_exit(); 150 151 if (origcpu != 0) 152 lwkt_setcpu_self(globaldata_find(origcpu)); 153 154 resettodr(); 155 return (0); 156 } 157 158 static void 159 get_curthread_cputime(struct timespec *ats) 160 { 161 struct thread *td = curthread; 162 163 crit_enter(); 164 /* 165 * These are 64-bit fields but the actual values should never reach 166 * the limit. We don't care about overflows. 167 */ 168 ats->tv_sec = td->td_uticks / 1000000; 169 ats->tv_sec += td->td_sticks / 1000000; 170 ats->tv_sec += td->td_iticks / 1000000; 171 ats->tv_nsec = (td->td_uticks % 1000000) * 1000; 172 ats->tv_nsec += (td->td_sticks % 1000000) * 1000; 173 ats->tv_nsec += (td->td_iticks % 1000000) * 1000; 174 crit_exit(); 175 } 176 177 /* 178 * MPSAFE 179 */ 180 int 181 kern_clock_gettime(clockid_t clock_id, struct timespec *ats) 182 { 183 int error = 0; 184 struct proc *p; 185 186 switch(clock_id) { 187 case CLOCK_REALTIME: 188 case CLOCK_REALTIME_PRECISE: 189 nanotime(ats); 190 break; 191 case CLOCK_REALTIME_FAST: 192 getnanotime(ats); 193 break; 194 case CLOCK_MONOTONIC: 195 case CLOCK_MONOTONIC_PRECISE: 196 case CLOCK_UPTIME: 197 case CLOCK_UPTIME_PRECISE: 198 nanouptime(ats); 199 break; 200 case CLOCK_MONOTONIC_FAST: 201 case CLOCK_UPTIME_FAST: 202 getnanouptime(ats); 203 break; 204 case CLOCK_VIRTUAL: 205 p = curproc; 206 ats->tv_sec = p->p_timer[ITIMER_VIRTUAL].it_value.tv_sec; 207 ats->tv_nsec = p->p_timer[ITIMER_VIRTUAL].it_value.tv_usec * 208 1000; 209 break; 210 case CLOCK_PROF: 211 p = curproc; 212 ats->tv_sec = p->p_timer[ITIMER_PROF].it_value.tv_sec; 213 ats->tv_nsec = p->p_timer[ITIMER_PROF].it_value.tv_usec * 214 1000; 215 break; 216 case CLOCK_SECOND: 217 ats->tv_sec = time_second; 218 ats->tv_nsec = 0; 219 break; 220 case CLOCK_THREAD_CPUTIME_ID: 221 get_curthread_cputime(ats); 222 break; 223 default: 224 error = EINVAL; 225 break; 226 } 227 return (error); 228 } 229 230 /* 231 * MPSAFE 232 */ 233 int 234 sys_clock_gettime(struct clock_gettime_args *uap) 235 { 236 struct timespec ats; 237 int error; 238 239 error = kern_clock_gettime(uap->clock_id, &ats); 240 if (error == 0) 241 error = copyout(&ats, uap->tp, sizeof(ats)); 242 243 return (error); 244 } 245 246 int 247 kern_clock_settime(clockid_t clock_id, struct timespec *ats) 248 { 249 struct thread *td = curthread; 250 struct timeval atv; 251 int error; 252 253 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 254 return (error); 255 if (clock_id != CLOCK_REALTIME) 256 return (EINVAL); 257 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000) 258 return (EINVAL); 259 260 TIMESPEC_TO_TIMEVAL(&atv, ats); 261 error = settime(&atv); 262 return (error); 263 } 264 265 /* 266 * MPALMOSTSAFE 267 */ 268 int 269 sys_clock_settime(struct clock_settime_args *uap) 270 { 271 struct timespec ats; 272 int error; 273 274 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 275 return (error); 276 277 get_mplock(); 278 error = kern_clock_settime(uap->clock_id, &ats); 279 rel_mplock(); 280 return (error); 281 } 282 283 /* 284 * MPSAFE 285 */ 286 int 287 kern_clock_getres(clockid_t clock_id, struct timespec *ts) 288 { 289 int error; 290 291 switch(clock_id) { 292 case CLOCK_REALTIME: 293 case CLOCK_REALTIME_FAST: 294 case CLOCK_REALTIME_PRECISE: 295 case CLOCK_MONOTONIC: 296 case CLOCK_MONOTONIC_FAST: 297 case CLOCK_MONOTONIC_PRECISE: 298 case CLOCK_UPTIME: 299 case CLOCK_UPTIME_FAST: 300 case CLOCK_UPTIME_PRECISE: 301 case CLOCK_THREAD_CPUTIME_ID: 302 /* 303 * Round up the result of the division cheaply 304 * by adding 1. Rounding up is especially important 305 * if rounding down would give 0. Perfect rounding 306 * is unimportant. 307 */ 308 ts->tv_sec = 0; 309 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1; 310 error = 0; 311 break; 312 case CLOCK_VIRTUAL: 313 case CLOCK_PROF: 314 /* Accurately round up here because we can do so cheaply. */ 315 ts->tv_sec = 0; 316 ts->tv_nsec = (1000000000 + hz - 1) / hz; 317 error = 0; 318 break; 319 case CLOCK_SECOND: 320 ts->tv_sec = 1; 321 ts->tv_nsec = 0; 322 error = 0; 323 break; 324 default: 325 error = EINVAL; 326 break; 327 } 328 329 return(error); 330 } 331 332 /* 333 * MPSAFE 334 */ 335 int 336 sys_clock_getres(struct clock_getres_args *uap) 337 { 338 int error; 339 struct timespec ts; 340 341 error = kern_clock_getres(uap->clock_id, &ts); 342 if (error == 0) 343 error = copyout(&ts, uap->tp, sizeof(ts)); 344 345 return (error); 346 } 347 348 /* 349 * nanosleep1() 350 * 351 * This is a general helper function for nanosleep() (aka sleep() aka 352 * usleep()). 353 * 354 * If there is less then one tick's worth of time left and 355 * we haven't done a yield, or the remaining microseconds is 356 * ridiculously low, do a yield. This avoids having 357 * to deal with systimer overheads when the system is under 358 * heavy loads. If we have done a yield already then use 359 * a systimer and an uninterruptable thread wait. 360 * 361 * If there is more then a tick's worth of time left, 362 * calculate the baseline ticks and use an interruptable 363 * tsleep, then handle the fine-grained delay on the next 364 * loop. This usually results in two sleeps occuring, a long one 365 * and a short one. 366 * 367 * MPSAFE 368 */ 369 static void 370 ns1_systimer(systimer_t info, int in_ipi __unused, 371 struct intrframe *frame __unused) 372 { 373 lwkt_schedule(info->data); 374 } 375 376 int 377 nanosleep1(struct timespec *rqt, struct timespec *rmt) 378 { 379 static int nanowait; 380 struct timespec ts, ts2, ts3; 381 struct timeval tv; 382 int error; 383 384 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 385 return (EINVAL); 386 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 387 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 388 return (0); 389 nanouptime(&ts); 390 timespecadd(&ts, rqt); /* ts = target timestamp compare */ 391 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 392 393 for (;;) { 394 int ticks; 395 struct systimer info; 396 397 ticks = tv.tv_usec / ustick; /* approximate */ 398 399 if (tv.tv_sec == 0 && ticks == 0) { 400 thread_t td = curthread; 401 if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us) 402 tv.tv_usec = nanosleep_min_us; 403 if (tv.tv_usec < nanosleep_hard_us) { 404 lwkt_user_yield(); 405 cpu_pause(); 406 } else { 407 crit_enter_quick(td); 408 systimer_init_oneshot(&info, ns1_systimer, 409 td, tv.tv_usec); 410 lwkt_deschedule_self(td); 411 crit_exit_quick(td); 412 lwkt_switch(); 413 systimer_del(&info); /* make sure it's gone */ 414 } 415 error = iscaught(td->td_lwp); 416 } else if (tv.tv_sec == 0) { 417 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 418 } else { 419 ticks = tvtohz_low(&tv); /* also handles overflow */ 420 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 421 } 422 nanouptime(&ts2); 423 if (error && error != EWOULDBLOCK) { 424 if (error == ERESTART) 425 error = EINTR; 426 if (rmt != NULL) { 427 timespecsub(&ts, &ts2); 428 if (ts.tv_sec < 0) 429 timespecclear(&ts); 430 *rmt = ts; 431 } 432 return (error); 433 } 434 if (timespeccmp(&ts2, &ts, >=)) 435 return (0); 436 ts3 = ts; 437 timespecsub(&ts3, &ts2); 438 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 439 } 440 } 441 442 /* 443 * MPSAFE 444 */ 445 int 446 sys_nanosleep(struct nanosleep_args *uap) 447 { 448 int error; 449 struct timespec rqt; 450 struct timespec rmt; 451 452 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 453 if (error) 454 return (error); 455 456 error = nanosleep1(&rqt, &rmt); 457 458 /* 459 * copyout the residual if nanosleep was interrupted. 460 */ 461 if (error && uap->rmtp) { 462 int error2; 463 464 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 465 if (error2) 466 error = error2; 467 } 468 return (error); 469 } 470 471 /* 472 * The gettimeofday() system call is supposed to return a fine-grained 473 * realtime stamp. However, acquiring a fine-grained stamp can create a 474 * bottleneck when multiple cpu cores are trying to accessing e.g. the 475 * HPET hardware timer all at the same time, so we have a sysctl that 476 * allows its behavior to be changed to a more coarse-grained timestamp 477 * which does not have to access a hardware timer. 478 */ 479 int 480 sys_gettimeofday(struct gettimeofday_args *uap) 481 { 482 struct timeval atv; 483 int error = 0; 484 485 if (uap->tp) { 486 if (gettimeofday_quick) 487 getmicrotime(&atv); 488 else 489 microtime(&atv); 490 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 491 sizeof (atv)))) 492 return (error); 493 } 494 if (uap->tzp) 495 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 496 sizeof (tz)); 497 return (error); 498 } 499 500 /* 501 * MPALMOSTSAFE 502 */ 503 int 504 sys_settimeofday(struct settimeofday_args *uap) 505 { 506 struct thread *td = curthread; 507 struct timeval atv; 508 struct timezone atz; 509 int error; 510 511 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 512 return (error); 513 /* 514 * Verify all parameters before changing time. 515 * 516 * NOTE: We do not allow the time to be set to 0.0, which also by 517 * happy coincidence works around a pkgsrc bulk build bug. 518 */ 519 if (uap->tv) { 520 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 521 sizeof(atv)))) 522 return (error); 523 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 524 return (EINVAL); 525 if (atv.tv_sec == 0 && atv.tv_usec == 0) 526 return (EINVAL); 527 } 528 if (uap->tzp && 529 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 530 return (error); 531 532 get_mplock(); 533 if (uap->tv && (error = settime(&atv))) { 534 rel_mplock(); 535 return (error); 536 } 537 rel_mplock(); 538 if (uap->tzp) 539 tz = atz; 540 return (0); 541 } 542 543 static void 544 kern_adjtime_common(void) 545 { 546 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 547 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 548 ntp_tick_delta = ntp_delta; 549 else if (ntp_delta > ntp_big_delta) 550 ntp_tick_delta = 10 * ntp_default_tick_delta; 551 else if (ntp_delta < -ntp_big_delta) 552 ntp_tick_delta = -10 * ntp_default_tick_delta; 553 else if (ntp_delta > 0) 554 ntp_tick_delta = ntp_default_tick_delta; 555 else 556 ntp_tick_delta = -ntp_default_tick_delta; 557 } 558 559 void 560 kern_adjtime(int64_t delta, int64_t *odelta) 561 { 562 int origcpu; 563 564 if ((origcpu = mycpu->gd_cpuid) != 0) 565 lwkt_setcpu_self(globaldata_find(0)); 566 567 crit_enter(); 568 *odelta = ntp_delta; 569 ntp_delta = delta; 570 kern_adjtime_common(); 571 crit_exit(); 572 573 if (origcpu != 0) 574 lwkt_setcpu_self(globaldata_find(origcpu)); 575 } 576 577 static void 578 kern_get_ntp_delta(int64_t *delta) 579 { 580 int origcpu; 581 582 if ((origcpu = mycpu->gd_cpuid) != 0) 583 lwkt_setcpu_self(globaldata_find(0)); 584 585 crit_enter(); 586 *delta = ntp_delta; 587 crit_exit(); 588 589 if (origcpu != 0) 590 lwkt_setcpu_self(globaldata_find(origcpu)); 591 } 592 593 void 594 kern_reladjtime(int64_t delta) 595 { 596 int origcpu; 597 598 if ((origcpu = mycpu->gd_cpuid) != 0) 599 lwkt_setcpu_self(globaldata_find(0)); 600 601 crit_enter(); 602 ntp_delta += delta; 603 kern_adjtime_common(); 604 crit_exit(); 605 606 if (origcpu != 0) 607 lwkt_setcpu_self(globaldata_find(origcpu)); 608 } 609 610 static void 611 kern_adjfreq(int64_t rate) 612 { 613 int origcpu; 614 615 if ((origcpu = mycpu->gd_cpuid) != 0) 616 lwkt_setcpu_self(globaldata_find(0)); 617 618 crit_enter(); 619 ntp_tick_permanent = rate; 620 crit_exit(); 621 622 if (origcpu != 0) 623 lwkt_setcpu_self(globaldata_find(origcpu)); 624 } 625 626 /* 627 * MPALMOSTSAFE 628 */ 629 int 630 sys_adjtime(struct adjtime_args *uap) 631 { 632 struct thread *td = curthread; 633 struct timeval atv; 634 int64_t ndelta, odelta; 635 int error; 636 637 if ((error = priv_check(td, PRIV_ADJTIME))) 638 return (error); 639 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 640 if (error) 641 return (error); 642 643 /* 644 * Compute the total correction and the rate at which to apply it. 645 * Round the adjustment down to a whole multiple of the per-tick 646 * delta, so that after some number of incremental changes in 647 * hardclock(), tickdelta will become zero, lest the correction 648 * overshoot and start taking us away from the desired final time. 649 */ 650 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 651 get_mplock(); 652 kern_adjtime(ndelta, &odelta); 653 rel_mplock(); 654 655 if (uap->olddelta) { 656 atv.tv_sec = odelta / 1000000000; 657 atv.tv_usec = odelta % 1000000000 / 1000; 658 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 659 } 660 return (0); 661 } 662 663 static int 664 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 665 { 666 int64_t 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_reladjtime(delta); 676 } 677 678 if (req->oldptr) 679 kern_get_ntp_delta(&delta); 680 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 681 return (error); 682 } 683 684 /* 685 * delta is in nanoseconds. 686 */ 687 static int 688 sysctl_delta(SYSCTL_HANDLER_ARGS) 689 { 690 int64_t delta, old_delta; 691 int error; 692 693 if (req->newptr != NULL) { 694 if (priv_check(curthread, PRIV_ROOT)) 695 return (EPERM); 696 error = SYSCTL_IN(req, &delta, sizeof(delta)); 697 if (error) 698 return (error); 699 kern_adjtime(delta, &old_delta); 700 } 701 702 if (req->oldptr != NULL) 703 kern_get_ntp_delta(&old_delta); 704 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 705 return (error); 706 } 707 708 /* 709 * frequency is in nanoseconds per second shifted left 32. 710 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 711 */ 712 static int 713 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 714 { 715 int64_t freqdelta; 716 int error; 717 718 if (req->newptr != NULL) { 719 if (priv_check(curthread, PRIV_ROOT)) 720 return (EPERM); 721 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 722 if (error) 723 return (error); 724 725 freqdelta /= hz; 726 kern_adjfreq(freqdelta); 727 } 728 729 if (req->oldptr != NULL) 730 freqdelta = ntp_tick_permanent * hz; 731 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 732 if (error) 733 return (error); 734 735 return (0); 736 } 737 738 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 739 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 740 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 741 sysctl_adjfreq, "Q", "permanent correction per second"); 742 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 743 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 744 sysctl_delta, "Q", "one-time delta"); 745 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 746 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 747 "threshold for fast adjustment"); 748 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 749 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 750 "per-tick adjustment"); 751 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 752 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 753 "default per-tick adjustment"); 754 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 755 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 756 "next leap second"); 757 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 758 &ntp_leap_insert, 0, "insert or remove leap second"); 759 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 760 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 761 sysctl_adjtime, "Q", "relative adjust for delta"); 762 763 /* 764 * Get value of an interval timer. The process virtual and 765 * profiling virtual time timers are kept in the p_stats area, since 766 * they can be swapped out. These are kept internally in the 767 * way they are specified externally: in time until they expire. 768 * 769 * The real time interval timer is kept in the process table slot 770 * for the process, and its value (it_value) is kept as an 771 * absolute time rather than as a delta, so that it is easy to keep 772 * periodic real-time signals from drifting. 773 * 774 * Virtual time timers are processed in the hardclock() routine of 775 * kern_clock.c. The real time timer is processed by a timeout 776 * routine, called from the softclock() routine. Since a callout 777 * may be delayed in real time due to interrupt processing in the system, 778 * it is possible for the real time timeout routine (realitexpire, given below), 779 * to be delayed in real time past when it is supposed to occur. It 780 * does not suffice, therefore, to reload the real timer .it_value from the 781 * real time timers .it_interval. Rather, we compute the next time in 782 * absolute time the timer should go off. 783 * 784 * MPALMOSTSAFE 785 */ 786 int 787 sys_getitimer(struct getitimer_args *uap) 788 { 789 struct proc *p = curproc; 790 struct timeval ctv; 791 struct itimerval aitv; 792 793 if (uap->which > ITIMER_PROF) 794 return (EINVAL); 795 lwkt_gettoken(&p->p_token); 796 if (uap->which == ITIMER_REAL) { 797 /* 798 * Convert from absolute to relative time in .it_value 799 * part of real time timer. If time for real time timer 800 * has passed return 0, else return difference between 801 * current time and time for the timer to go off. 802 */ 803 aitv = p->p_realtimer; 804 if (timevalisset(&aitv.it_value)) { 805 getmicrouptime(&ctv); 806 if (timevalcmp(&aitv.it_value, &ctv, <)) 807 timevalclear(&aitv.it_value); 808 else 809 timevalsub(&aitv.it_value, &ctv); 810 } 811 } else { 812 aitv = p->p_timer[uap->which]; 813 } 814 lwkt_reltoken(&p->p_token); 815 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 816 } 817 818 /* 819 * MPALMOSTSAFE 820 */ 821 int 822 sys_setitimer(struct setitimer_args *uap) 823 { 824 struct itimerval aitv; 825 struct timeval ctv; 826 struct itimerval *itvp; 827 struct proc *p = curproc; 828 int error; 829 830 if (uap->which > ITIMER_PROF) 831 return (EINVAL); 832 itvp = uap->itv; 833 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 834 sizeof(struct itimerval)))) 835 return (error); 836 if ((uap->itv = uap->oitv) && 837 (error = sys_getitimer((struct getitimer_args *)uap))) 838 return (error); 839 if (itvp == NULL) 840 return (0); 841 if (itimerfix(&aitv.it_value)) 842 return (EINVAL); 843 if (!timevalisset(&aitv.it_value)) 844 timevalclear(&aitv.it_interval); 845 else if (itimerfix(&aitv.it_interval)) 846 return (EINVAL); 847 lwkt_gettoken(&p->p_token); 848 if (uap->which == ITIMER_REAL) { 849 if (timevalisset(&p->p_realtimer.it_value)) 850 callout_stop_sync(&p->p_ithandle); 851 if (timevalisset(&aitv.it_value)) 852 callout_reset(&p->p_ithandle, 853 tvtohz_high(&aitv.it_value), realitexpire, p); 854 getmicrouptime(&ctv); 855 timevaladd(&aitv.it_value, &ctv); 856 p->p_realtimer = aitv; 857 } else { 858 p->p_timer[uap->which] = aitv; 859 switch(uap->which) { 860 case ITIMER_VIRTUAL: 861 p->p_flags &= ~P_SIGVTALRM; 862 break; 863 case ITIMER_PROF: 864 p->p_flags &= ~P_SIGPROF; 865 break; 866 } 867 } 868 lwkt_reltoken(&p->p_token); 869 return (0); 870 } 871 872 /* 873 * Real interval timer expired: 874 * send process whose timer expired an alarm signal. 875 * If time is not set up to reload, then just return. 876 * Else compute next time timer should go off which is > current time. 877 * This is where delay in processing this timeout causes multiple 878 * SIGALRM calls to be compressed into one. 879 * tvtohz_high() always adds 1 to allow for the time until the next clock 880 * interrupt being strictly less than 1 clock tick, but we don't want 881 * that here since we want to appear to be in sync with the clock 882 * interrupt even when we're delayed. 883 */ 884 void 885 realitexpire(void *arg) 886 { 887 struct proc *p; 888 struct timeval ctv, ntv; 889 890 p = (struct proc *)arg; 891 PHOLD(p); 892 lwkt_gettoken(&p->p_token); 893 ksignal(p, SIGALRM); 894 if (!timevalisset(&p->p_realtimer.it_interval)) { 895 timevalclear(&p->p_realtimer.it_value); 896 goto done; 897 } 898 for (;;) { 899 timevaladd(&p->p_realtimer.it_value, 900 &p->p_realtimer.it_interval); 901 getmicrouptime(&ctv); 902 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 903 ntv = p->p_realtimer.it_value; 904 timevalsub(&ntv, &ctv); 905 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 906 realitexpire, p); 907 goto done; 908 } 909 } 910 done: 911 lwkt_reltoken(&p->p_token); 912 PRELE(p); 913 } 914 915 /* 916 * Check that a proposed value to load into the .it_value or 917 * .it_interval part of an interval timer is acceptable, and 918 * fix it to have at least minimal value (i.e. if it is less 919 * than the resolution of the clock, round it up.) 920 * 921 * MPSAFE 922 */ 923 int 924 itimerfix(struct timeval *tv) 925 { 926 927 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 || 928 tv->tv_usec < 0 || tv->tv_usec >= 1000000) 929 return (EINVAL); 930 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 931 tv->tv_usec = ustick; 932 return (0); 933 } 934 935 /* 936 * Decrement an interval timer by a specified number 937 * of microseconds, which must be less than a second, 938 * i.e. < 1000000. If the timer expires, then reload 939 * it. In this case, carry over (usec - old value) to 940 * reduce the value reloaded into the timer so that 941 * the timer does not drift. This routine assumes 942 * that it is called in a context where the timers 943 * on which it is operating cannot change in value. 944 */ 945 int 946 itimerdecr(struct itimerval *itp, int usec) 947 { 948 949 if (itp->it_value.tv_usec < usec) { 950 if (itp->it_value.tv_sec == 0) { 951 /* expired, and already in next interval */ 952 usec -= itp->it_value.tv_usec; 953 goto expire; 954 } 955 itp->it_value.tv_usec += 1000000; 956 itp->it_value.tv_sec--; 957 } 958 itp->it_value.tv_usec -= usec; 959 usec = 0; 960 if (timevalisset(&itp->it_value)) 961 return (1); 962 /* expired, exactly at end of interval */ 963 expire: 964 if (timevalisset(&itp->it_interval)) { 965 itp->it_value = itp->it_interval; 966 itp->it_value.tv_usec -= usec; 967 if (itp->it_value.tv_usec < 0) { 968 itp->it_value.tv_usec += 1000000; 969 itp->it_value.tv_sec--; 970 } 971 } else 972 itp->it_value.tv_usec = 0; /* sec is already 0 */ 973 return (0); 974 } 975 976 /* 977 * Add and subtract routines for timevals. 978 * N.B.: subtract routine doesn't deal with 979 * results which are before the beginning, 980 * it just gets very confused in this case. 981 * Caveat emptor. 982 */ 983 void 984 timevaladd(struct timeval *t1, const struct timeval *t2) 985 { 986 987 t1->tv_sec += t2->tv_sec; 988 t1->tv_usec += t2->tv_usec; 989 timevalfix(t1); 990 } 991 992 void 993 timevalsub(struct timeval *t1, const struct timeval *t2) 994 { 995 996 t1->tv_sec -= t2->tv_sec; 997 t1->tv_usec -= t2->tv_usec; 998 timevalfix(t1); 999 } 1000 1001 static void 1002 timevalfix(struct timeval *t1) 1003 { 1004 1005 if (t1->tv_usec < 0) { 1006 t1->tv_sec--; 1007 t1->tv_usec += 1000000; 1008 } 1009 if (t1->tv_usec >= 1000000) { 1010 t1->tv_sec++; 1011 t1->tv_usec -= 1000000; 1012 } 1013 } 1014 1015 /* 1016 * ratecheck(): simple time-based rate-limit checking. 1017 */ 1018 int 1019 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1020 { 1021 struct timeval tv, delta; 1022 int rv = 0; 1023 1024 getmicrouptime(&tv); /* NB: 10ms precision */ 1025 delta = tv; 1026 timevalsub(&delta, lasttime); 1027 1028 /* 1029 * check for 0,0 is so that the message will be seen at least once, 1030 * even if interval is huge. 1031 */ 1032 if (timevalcmp(&delta, mininterval, >=) || 1033 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1034 *lasttime = tv; 1035 rv = 1; 1036 } 1037 1038 return (rv); 1039 } 1040 1041 /* 1042 * ppsratecheck(): packets (or events) per second limitation. 1043 * 1044 * Return 0 if the limit is to be enforced (e.g. the caller 1045 * should drop a packet because of the rate limitation). 1046 * 1047 * maxpps of 0 always causes zero to be returned. maxpps of -1 1048 * always causes 1 to be returned; this effectively defeats rate 1049 * limiting. 1050 * 1051 * Note that we maintain the struct timeval for compatibility 1052 * with other bsd systems. We reuse the storage and just monitor 1053 * clock ticks for minimal overhead. 1054 */ 1055 int 1056 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1057 { 1058 int now; 1059 1060 /* 1061 * Reset the last time and counter if this is the first call 1062 * or more than a second has passed since the last update of 1063 * lasttime. 1064 */ 1065 now = ticks; 1066 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1067 lasttime->tv_sec = now; 1068 *curpps = 1; 1069 return (maxpps != 0); 1070 } else { 1071 (*curpps)++; /* NB: ignore potential overflow */ 1072 return (maxpps < 0 || *curpps < maxpps); 1073 } 1074 } 1075