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 #define CPUCLOCK_BIT 0x80000000 58 #define CPUCLOCK_ID_MASK ~CPUCLOCK_BIT 59 #define CPUCLOCK2LWPID(clock_id) (((clockid_t)(clock_id) >> 32) & CPUCLOCK_ID_MASK) 60 #define CPUCLOCK2PID(clock_id) ((clock_id) & CPUCLOCK_ID_MASK) 61 #define MAKE_CPUCLOCK(pid, lwp_id) ((clockid_t)(lwp_id) << 32 | (pid) | CPUCLOCK_BIT) 62 63 struct timezone tz; 64 65 /* 66 * Time of day and interval timer support. 67 * 68 * These routines provide the kernel entry points to get and set 69 * the time-of-day and per-process interval timers. Subroutines 70 * here provide support for adding and subtracting timeval structures 71 * and decrementing interval timers, optionally reloading the interval 72 * timers when they expire. 73 */ 74 75 static int settime(struct timeval *); 76 static void timevalfix(struct timeval *); 77 static void realitexpire(void *arg); 78 79 /* 80 * Nanosleep tries very hard to sleep for a precisely requested time 81 * interval, down to 1uS. The administrator can impose a minimum delay 82 * and a delay below which we hard-loop instead of initiate a timer 83 * interrupt and sleep. 84 * 85 * For machines under high loads it might be beneficial to increase min_us 86 * to e.g. 1000uS (1ms) so spining processes sleep meaningfully. 87 */ 88 static int nanosleep_min_us = 10; 89 static int nanosleep_hard_us = 100; 90 static int gettimeofday_quick = 0; 91 SYSCTL_INT(_kern, OID_AUTO, nanosleep_min_us, CTLFLAG_RW, 92 &nanosleep_min_us, 0, ""); 93 SYSCTL_INT(_kern, OID_AUTO, nanosleep_hard_us, CTLFLAG_RW, 94 &nanosleep_hard_us, 0, ""); 95 SYSCTL_INT(_kern, OID_AUTO, gettimeofday_quick, CTLFLAG_RW, 96 &gettimeofday_quick, 0, ""); 97 98 static struct lock masterclock_lock = LOCK_INITIALIZER("mstrclk", 0, 0); 99 100 static int 101 settime(struct timeval *tv) 102 { 103 struct timeval delta, tv1, tv2; 104 static struct timeval maxtime, laststep; 105 struct timespec ts; 106 int origcpu; 107 108 if ((origcpu = mycpu->gd_cpuid) != 0) 109 lwkt_setcpu_self(globaldata_find(0)); 110 111 crit_enter(); 112 microtime(&tv1); 113 delta = *tv; 114 timevalsub(&delta, &tv1); 115 116 /* 117 * If the system is secure, we do not allow the time to be 118 * set to a value earlier than 1 second less than the highest 119 * time we have yet seen. The worst a miscreant can do in 120 * this circumstance is "freeze" time. He couldn't go 121 * back to the past. 122 * 123 * We similarly do not allow the clock to be stepped more 124 * than one second, nor more than once per second. This allows 125 * a miscreant to make the clock march double-time, but no worse. 126 */ 127 if (securelevel > 1) { 128 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 129 /* 130 * Update maxtime to latest time we've seen. 131 */ 132 if (tv1.tv_sec > maxtime.tv_sec) 133 maxtime = tv1; 134 tv2 = *tv; 135 timevalsub(&tv2, &maxtime); 136 if (tv2.tv_sec < -1) { 137 tv->tv_sec = maxtime.tv_sec - 1; 138 kprintf("Time adjustment clamped to -1 second\n"); 139 } 140 } else { 141 if (tv1.tv_sec == laststep.tv_sec) { 142 crit_exit(); 143 return (EPERM); 144 } 145 if (delta.tv_sec > 1) { 146 tv->tv_sec = tv1.tv_sec + 1; 147 kprintf("Time adjustment clamped to +1 second\n"); 148 } 149 laststep = *tv; 150 } 151 } 152 153 ts.tv_sec = tv->tv_sec; 154 ts.tv_nsec = tv->tv_usec * 1000; 155 set_timeofday(&ts); 156 crit_exit(); 157 158 if (origcpu != 0) 159 lwkt_setcpu_self(globaldata_find(origcpu)); 160 161 resettodr(); 162 return (0); 163 } 164 165 static void 166 get_process_cputime(struct proc *p, struct timespec *ats) 167 { 168 struct rusage ru; 169 170 lwkt_gettoken(&p->p_token); 171 calcru_proc(p, &ru); 172 lwkt_reltoken(&p->p_token); 173 timevaladd(&ru.ru_utime, &ru.ru_stime); 174 TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats); 175 } 176 177 static void 178 get_process_usertime(struct proc *p, struct timespec *ats) 179 { 180 struct rusage ru; 181 182 lwkt_gettoken(&p->p_token); 183 calcru_proc(p, &ru); 184 lwkt_reltoken(&p->p_token); 185 TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats); 186 } 187 188 static void 189 get_thread_cputime(struct thread *td, struct timespec *ats) 190 { 191 struct timeval sys, user; 192 193 calcru(td->td_lwp, &user, &sys); 194 timevaladd(&user, &sys); 195 TIMEVAL_TO_TIMESPEC(&user, ats); 196 } 197 198 /* 199 * MPSAFE 200 */ 201 int 202 kern_clock_gettime(clockid_t clock_id, struct timespec *ats) 203 { 204 struct proc *p; 205 struct lwp *lp; 206 lwpid_t lwp_id; 207 208 p = curproc; 209 switch(clock_id) { 210 case CLOCK_REALTIME: 211 case CLOCK_REALTIME_PRECISE: 212 nanotime(ats); 213 break; 214 case CLOCK_REALTIME_FAST: 215 getnanotime(ats); 216 break; 217 case CLOCK_MONOTONIC: 218 case CLOCK_MONOTONIC_PRECISE: 219 case CLOCK_UPTIME: 220 case CLOCK_UPTIME_PRECISE: 221 nanouptime(ats); 222 break; 223 case CLOCK_MONOTONIC_FAST: 224 case CLOCK_UPTIME_FAST: 225 getnanouptime(ats); 226 break; 227 case CLOCK_VIRTUAL: 228 get_process_usertime(p, ats); 229 break; 230 case CLOCK_PROF: 231 case CLOCK_PROCESS_CPUTIME_ID: 232 get_process_cputime(p, ats); 233 break; 234 case CLOCK_SECOND: 235 ats->tv_sec = time_second; 236 ats->tv_nsec = 0; 237 break; 238 case CLOCK_THREAD_CPUTIME_ID: 239 get_thread_cputime(curthread, ats); 240 break; 241 default: 242 if ((clock_id & CPUCLOCK_BIT) == 0) 243 return (EINVAL); 244 if ((p = pfind(CPUCLOCK2PID(clock_id))) == NULL) 245 return (EINVAL); 246 lwp_id = CPUCLOCK2LWPID(clock_id); 247 if (lwp_id == 0) { 248 get_process_cputime(p, ats); 249 } else { 250 lwkt_gettoken(&p->p_token); 251 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, lwp_id); 252 if (lp == NULL) { 253 lwkt_reltoken(&p->p_token); 254 PRELE(p); 255 return (EINVAL); 256 } 257 get_thread_cputime(lp->lwp_thread, ats); 258 lwkt_reltoken(&p->p_token); 259 } 260 PRELE(p); 261 } 262 return (0); 263 } 264 265 /* 266 * MPSAFE 267 */ 268 int 269 sys_clock_gettime(struct clock_gettime_args *uap) 270 { 271 struct timespec ats; 272 int error; 273 274 error = kern_clock_gettime(uap->clock_id, &ats); 275 if (error == 0) 276 error = copyout(&ats, uap->tp, sizeof(ats)); 277 278 return (error); 279 } 280 281 int 282 kern_clock_settime(clockid_t clock_id, struct timespec *ats) 283 { 284 struct thread *td = curthread; 285 struct timeval atv; 286 int error; 287 288 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 289 return (error); 290 if (clock_id != CLOCK_REALTIME) 291 return (EINVAL); 292 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000) 293 return (EINVAL); 294 295 lockmgr(&masterclock_lock, LK_EXCLUSIVE); 296 TIMESPEC_TO_TIMEVAL(&atv, ats); 297 error = settime(&atv); 298 lockmgr(&masterclock_lock, LK_RELEASE); 299 300 return (error); 301 } 302 303 /* 304 * MPALMOSTSAFE 305 */ 306 int 307 sys_clock_settime(struct clock_settime_args *uap) 308 { 309 struct timespec ats; 310 int error; 311 312 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 313 return (error); 314 315 error = kern_clock_settime(uap->clock_id, &ats); 316 317 return (error); 318 } 319 320 /* 321 * MPSAFE 322 */ 323 int 324 kern_clock_getres(clockid_t clock_id, struct timespec *ts) 325 { 326 ts->tv_sec = 0; 327 switch(clock_id) { 328 case CLOCK_REALTIME: 329 case CLOCK_REALTIME_FAST: 330 case CLOCK_REALTIME_PRECISE: 331 case CLOCK_MONOTONIC: 332 case CLOCK_MONOTONIC_FAST: 333 case CLOCK_MONOTONIC_PRECISE: 334 case CLOCK_UPTIME: 335 case CLOCK_UPTIME_FAST: 336 case CLOCK_UPTIME_PRECISE: 337 /* 338 * Round up the result of the division cheaply 339 * by adding 1. Rounding up is especially important 340 * if rounding down would give 0. Perfect rounding 341 * is unimportant. 342 */ 343 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1; 344 break; 345 case CLOCK_VIRTUAL: 346 case CLOCK_PROF: 347 /* Accurately round up here because we can do so cheaply. */ 348 ts->tv_nsec = (1000000000 + hz - 1) / hz; 349 break; 350 case CLOCK_SECOND: 351 ts->tv_sec = 1; 352 ts->tv_nsec = 0; 353 break; 354 case CLOCK_THREAD_CPUTIME_ID: 355 case CLOCK_PROCESS_CPUTIME_ID: 356 ts->tv_nsec = 1000; 357 break; 358 default: 359 if ((clock_id & CPUCLOCK_BIT) != 0) 360 ts->tv_nsec = 1000; 361 else 362 return (EINVAL); 363 } 364 365 return (0); 366 } 367 368 /* 369 * MPSAFE 370 */ 371 int 372 sys_clock_getres(struct clock_getres_args *uap) 373 { 374 int error; 375 struct timespec ts; 376 377 error = kern_clock_getres(uap->clock_id, &ts); 378 if (error == 0) 379 error = copyout(&ts, uap->tp, sizeof(ts)); 380 381 return (error); 382 } 383 384 static int 385 kern_getcpuclockid(pid_t pid, lwpid_t lwp_id, clockid_t *clock_id) 386 { 387 struct proc *p; 388 int error = 0; 389 390 if (pid == 0) { 391 p = curproc; 392 pid = p->p_pid; 393 PHOLD(p); 394 } else { 395 p = pfind(pid); 396 if (p == NULL) 397 return (ESRCH); 398 } 399 /* lwp_id can be 0 when called by clock_getcpuclockid() */ 400 if (lwp_id < 0) { 401 error = EINVAL; 402 goto out; 403 } 404 lwkt_gettoken(&p->p_token); 405 if (lwp_id > 0 && 406 lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, lwp_id) == NULL) { 407 lwkt_reltoken(&p->p_token); 408 error = ESRCH; 409 goto out; 410 } 411 *clock_id = MAKE_CPUCLOCK(pid, lwp_id); 412 lwkt_reltoken(&p->p_token); 413 out: 414 PRELE(p); 415 return (error); 416 } 417 418 int 419 sys_getcpuclockid(struct getcpuclockid_args *uap) 420 { 421 clockid_t clk_id; 422 int error; 423 424 error = kern_getcpuclockid(uap->pid, uap->lwp_id, &clk_id); 425 if (error == 0) 426 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 427 428 return (error); 429 } 430 431 /* 432 * nanosleep1() 433 * 434 * This is a general helper function for nanosleep() (aka sleep() aka 435 * usleep()). 436 * 437 * If there is less then one tick's worth of time left and 438 * we haven't done a yield, or the remaining microseconds is 439 * ridiculously low, do a yield. This avoids having 440 * to deal with systimer overheads when the system is under 441 * heavy loads. If we have done a yield already then use 442 * a systimer and an uninterruptable thread wait. 443 * 444 * If there is more then a tick's worth of time left, 445 * calculate the baseline ticks and use an interruptable 446 * tsleep, then handle the fine-grained delay on the next 447 * loop. This usually results in two sleeps occuring, a long one 448 * and a short one. 449 * 450 * MPSAFE 451 */ 452 static void 453 ns1_systimer(systimer_t info, int in_ipi __unused, 454 struct intrframe *frame __unused) 455 { 456 lwkt_schedule(info->data); 457 } 458 459 int 460 nanosleep1(struct timespec *rqt, struct timespec *rmt) 461 { 462 static int nanowait; 463 struct timespec ts, ts2, ts3; 464 struct timeval tv; 465 int error; 466 467 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 468 return (EINVAL); 469 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 470 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 471 return (0); 472 nanouptime(&ts); 473 timespecadd(&ts, rqt); /* ts = target timestamp compare */ 474 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 475 476 for (;;) { 477 int ticks; 478 struct systimer info; 479 480 ticks = tv.tv_usec / ustick; /* approximate */ 481 482 if (tv.tv_sec == 0 && ticks == 0) { 483 thread_t td = curthread; 484 if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us) 485 tv.tv_usec = nanosleep_min_us; 486 if (tv.tv_usec < nanosleep_hard_us) { 487 lwkt_user_yield(); 488 cpu_pause(); 489 } else { 490 crit_enter_quick(td); 491 systimer_init_oneshot(&info, ns1_systimer, 492 td, tv.tv_usec); 493 lwkt_deschedule_self(td); 494 crit_exit_quick(td); 495 lwkt_switch(); 496 systimer_del(&info); /* make sure it's gone */ 497 } 498 error = iscaught(td->td_lwp); 499 } else if (tv.tv_sec == 0) { 500 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 501 } else { 502 ticks = tvtohz_low(&tv); /* also handles overflow */ 503 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 504 } 505 nanouptime(&ts2); 506 if (error && error != EWOULDBLOCK) { 507 if (error == ERESTART) 508 error = EINTR; 509 if (rmt != NULL) { 510 timespecsub(&ts, &ts2); 511 if (ts.tv_sec < 0) 512 timespecclear(&ts); 513 *rmt = ts; 514 } 515 return (error); 516 } 517 if (timespeccmp(&ts2, &ts, >=)) 518 return (0); 519 ts3 = ts; 520 timespecsub(&ts3, &ts2); 521 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 522 } 523 } 524 525 /* 526 * MPSAFE 527 */ 528 int 529 sys_nanosleep(struct nanosleep_args *uap) 530 { 531 int error; 532 struct timespec rqt; 533 struct timespec rmt; 534 535 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 536 if (error) 537 return (error); 538 539 error = nanosleep1(&rqt, &rmt); 540 541 /* 542 * copyout the residual if nanosleep was interrupted. 543 */ 544 if (error && uap->rmtp) { 545 int error2; 546 547 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 548 if (error2) 549 error = error2; 550 } 551 return (error); 552 } 553 554 /* 555 * The gettimeofday() system call is supposed to return a fine-grained 556 * realtime stamp. However, acquiring a fine-grained stamp can create a 557 * bottleneck when multiple cpu cores are trying to accessing e.g. the 558 * HPET hardware timer all at the same time, so we have a sysctl that 559 * allows its behavior to be changed to a more coarse-grained timestamp 560 * which does not have to access a hardware timer. 561 */ 562 int 563 sys_gettimeofday(struct gettimeofday_args *uap) 564 { 565 struct timeval atv; 566 int error = 0; 567 568 if (uap->tp) { 569 if (gettimeofday_quick) 570 getmicrotime(&atv); 571 else 572 microtime(&atv); 573 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 574 sizeof (atv)))) 575 return (error); 576 } 577 if (uap->tzp) 578 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 579 sizeof (tz)); 580 return (error); 581 } 582 583 /* 584 * MPALMOSTSAFE 585 */ 586 int 587 sys_settimeofday(struct settimeofday_args *uap) 588 { 589 struct thread *td = curthread; 590 struct timeval atv; 591 struct timezone atz; 592 int error; 593 594 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 595 return (error); 596 /* 597 * Verify all parameters before changing time. 598 * 599 * XXX: We do not allow the time to be set to 0.0, which also by 600 * happy coincidence works around a pkgsrc bulk build bug. 601 */ 602 if (uap->tv) { 603 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 604 sizeof(atv)))) 605 return (error); 606 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 607 return (EINVAL); 608 if (atv.tv_sec == 0 && atv.tv_usec == 0) 609 return (EINVAL); 610 } 611 if (uap->tzp && 612 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 613 return (error); 614 615 lockmgr(&masterclock_lock, LK_EXCLUSIVE); 616 if (uap->tv && (error = settime(&atv))) { 617 lockmgr(&masterclock_lock, LK_RELEASE); 618 return (error); 619 } 620 lockmgr(&masterclock_lock, LK_RELEASE); 621 622 if (uap->tzp) 623 tz = atz; 624 return (0); 625 } 626 627 /* 628 * WARNING! Run with ntp_spin held 629 */ 630 static void 631 kern_adjtime_common(void) 632 { 633 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 634 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 635 ntp_tick_delta = ntp_delta; 636 else if (ntp_delta > ntp_big_delta) 637 ntp_tick_delta = 10 * ntp_default_tick_delta; 638 else if (ntp_delta < -ntp_big_delta) 639 ntp_tick_delta = -10 * ntp_default_tick_delta; 640 else if (ntp_delta > 0) 641 ntp_tick_delta = ntp_default_tick_delta; 642 else 643 ntp_tick_delta = -ntp_default_tick_delta; 644 } 645 646 void 647 kern_adjtime(int64_t delta, int64_t *odelta) 648 { 649 spin_lock(&ntp_spin); 650 *odelta = ntp_delta; 651 ntp_delta = delta; 652 kern_adjtime_common(); 653 spin_unlock(&ntp_spin); 654 } 655 656 static void 657 kern_get_ntp_delta(int64_t *delta) 658 { 659 *delta = ntp_delta; 660 } 661 662 void 663 kern_reladjtime(int64_t delta) 664 { 665 spin_lock(&ntp_spin); 666 ntp_delta += delta; 667 kern_adjtime_common(); 668 spin_unlock(&ntp_spin); 669 } 670 671 static void 672 kern_adjfreq(int64_t rate) 673 { 674 spin_lock(&ntp_spin); 675 ntp_tick_permanent = rate; 676 spin_unlock(&ntp_spin); 677 } 678 679 /* 680 * MPALMOSTSAFE 681 */ 682 int 683 sys_adjtime(struct adjtime_args *uap) 684 { 685 struct thread *td = curthread; 686 struct timeval atv; 687 int64_t ndelta, odelta; 688 int error; 689 690 if ((error = priv_check(td, PRIV_ADJTIME))) 691 return (error); 692 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 693 if (error) 694 return (error); 695 696 /* 697 * Compute the total correction and the rate at which to apply it. 698 * Round the adjustment down to a whole multiple of the per-tick 699 * delta, so that after some number of incremental changes in 700 * hardclock(), tickdelta will become zero, lest the correction 701 * overshoot and start taking us away from the desired final time. 702 */ 703 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 704 kern_adjtime(ndelta, &odelta); 705 706 if (uap->olddelta) { 707 atv.tv_sec = odelta / 1000000000; 708 atv.tv_usec = odelta % 1000000000 / 1000; 709 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 710 } 711 return (0); 712 } 713 714 static int 715 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 716 { 717 int64_t delta; 718 int error; 719 720 if (req->newptr != NULL) { 721 if (priv_check(curthread, PRIV_ROOT)) 722 return (EPERM); 723 error = SYSCTL_IN(req, &delta, sizeof(delta)); 724 if (error) 725 return (error); 726 kern_reladjtime(delta); 727 } 728 729 if (req->oldptr) 730 kern_get_ntp_delta(&delta); 731 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 732 return (error); 733 } 734 735 /* 736 * delta is in nanoseconds. 737 */ 738 static int 739 sysctl_delta(SYSCTL_HANDLER_ARGS) 740 { 741 int64_t delta, old_delta; 742 int error; 743 744 if (req->newptr != NULL) { 745 if (priv_check(curthread, PRIV_ROOT)) 746 return (EPERM); 747 error = SYSCTL_IN(req, &delta, sizeof(delta)); 748 if (error) 749 return (error); 750 kern_adjtime(delta, &old_delta); 751 } 752 753 if (req->oldptr != NULL) 754 kern_get_ntp_delta(&old_delta); 755 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 756 return (error); 757 } 758 759 /* 760 * frequency is in nanoseconds per second shifted left 32. 761 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 762 */ 763 static int 764 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 765 { 766 int64_t freqdelta; 767 int error; 768 769 if (req->newptr != NULL) { 770 if (priv_check(curthread, PRIV_ROOT)) 771 return (EPERM); 772 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 773 if (error) 774 return (error); 775 776 freqdelta /= hz; 777 kern_adjfreq(freqdelta); 778 } 779 780 if (req->oldptr != NULL) 781 freqdelta = ntp_tick_permanent * hz; 782 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 783 if (error) 784 return (error); 785 786 return (0); 787 } 788 789 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 790 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 791 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 792 sysctl_adjfreq, "Q", "permanent correction per second"); 793 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 794 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 795 sysctl_delta, "Q", "one-time delta"); 796 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 797 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 798 "threshold for fast adjustment"); 799 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 800 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 801 "per-tick adjustment"); 802 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 803 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 804 "default per-tick adjustment"); 805 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 806 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 807 "next leap second"); 808 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 809 &ntp_leap_insert, 0, "insert or remove leap second"); 810 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 811 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 812 sysctl_adjtime, "Q", "relative adjust for delta"); 813 814 /* 815 * Get value of an interval timer. The process virtual and 816 * profiling virtual time timers are kept in the p_stats area, since 817 * they can be swapped out. These are kept internally in the 818 * way they are specified externally: in time until they expire. 819 * 820 * The real time interval timer is kept in the process table slot 821 * for the process, and its value (it_value) is kept as an 822 * absolute time rather than as a delta, so that it is easy to keep 823 * periodic real-time signals from drifting. 824 * 825 * Virtual time timers are processed in the hardclock() routine of 826 * kern_clock.c. The real time timer is processed by a timeout 827 * routine, called from the softclock() routine. Since a callout 828 * may be delayed in real time due to interrupt processing in the system, 829 * it is possible for the real time timeout routine (realitexpire, given below), 830 * to be delayed in real time past when it is supposed to occur. It 831 * does not suffice, therefore, to reload the real timer .it_value from the 832 * real time timers .it_interval. Rather, we compute the next time in 833 * absolute time the timer should go off. 834 * 835 * MPALMOSTSAFE 836 */ 837 int 838 sys_getitimer(struct getitimer_args *uap) 839 { 840 struct proc *p = curproc; 841 struct timeval ctv; 842 struct itimerval aitv; 843 844 if (uap->which > ITIMER_PROF) 845 return (EINVAL); 846 lwkt_gettoken(&p->p_token); 847 if (uap->which == ITIMER_REAL) { 848 /* 849 * Convert from absolute to relative time in .it_value 850 * part of real time timer. If time for real time timer 851 * has passed return 0, else return difference between 852 * current time and time for the timer to go off. 853 */ 854 aitv = p->p_realtimer; 855 if (timevalisset(&aitv.it_value)) { 856 getmicrouptime(&ctv); 857 if (timevalcmp(&aitv.it_value, &ctv, <)) 858 timevalclear(&aitv.it_value); 859 else 860 timevalsub(&aitv.it_value, &ctv); 861 } 862 } else { 863 aitv = p->p_timer[uap->which]; 864 } 865 lwkt_reltoken(&p->p_token); 866 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 867 } 868 869 /* 870 * MPALMOSTSAFE 871 */ 872 int 873 sys_setitimer(struct setitimer_args *uap) 874 { 875 struct itimerval aitv; 876 struct timeval ctv; 877 struct itimerval *itvp; 878 struct proc *p = curproc; 879 int error; 880 881 if (uap->which > ITIMER_PROF) 882 return (EINVAL); 883 itvp = uap->itv; 884 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 885 sizeof(struct itimerval)))) 886 return (error); 887 if ((uap->itv = uap->oitv) && 888 (error = sys_getitimer((struct getitimer_args *)uap))) 889 return (error); 890 if (itvp == NULL) 891 return (0); 892 if (itimerfix(&aitv.it_value)) 893 return (EINVAL); 894 if (!timevalisset(&aitv.it_value)) 895 timevalclear(&aitv.it_interval); 896 else if (itimerfix(&aitv.it_interval)) 897 return (EINVAL); 898 lwkt_gettoken(&p->p_token); 899 if (uap->which == ITIMER_REAL) { 900 if (timevalisset(&p->p_realtimer.it_value)) 901 callout_stop_sync(&p->p_ithandle); 902 if (timevalisset(&aitv.it_value)) 903 callout_reset(&p->p_ithandle, 904 tvtohz_high(&aitv.it_value), realitexpire, p); 905 getmicrouptime(&ctv); 906 timevaladd(&aitv.it_value, &ctv); 907 p->p_realtimer = aitv; 908 } else { 909 p->p_timer[uap->which] = aitv; 910 switch(uap->which) { 911 case ITIMER_VIRTUAL: 912 p->p_flags &= ~P_SIGVTALRM; 913 break; 914 case ITIMER_PROF: 915 p->p_flags &= ~P_SIGPROF; 916 break; 917 } 918 } 919 lwkt_reltoken(&p->p_token); 920 return (0); 921 } 922 923 /* 924 * Real interval timer expired: 925 * send process whose timer expired an alarm signal. 926 * If time is not set up to reload, then just return. 927 * Else compute next time timer should go off which is > current time. 928 * This is where delay in processing this timeout causes multiple 929 * SIGALRM calls to be compressed into one. 930 * tvtohz_high() always adds 1 to allow for the time until the next clock 931 * interrupt being strictly less than 1 clock tick, but we don't want 932 * that here since we want to appear to be in sync with the clock 933 * interrupt even when we're delayed. 934 */ 935 static 936 void 937 realitexpire(void *arg) 938 { 939 struct proc *p; 940 struct timeval ctv, ntv; 941 942 p = (struct proc *)arg; 943 PHOLD(p); 944 lwkt_gettoken(&p->p_token); 945 ksignal(p, SIGALRM); 946 if (!timevalisset(&p->p_realtimer.it_interval)) { 947 timevalclear(&p->p_realtimer.it_value); 948 goto done; 949 } 950 for (;;) { 951 timevaladd(&p->p_realtimer.it_value, 952 &p->p_realtimer.it_interval); 953 getmicrouptime(&ctv); 954 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 955 ntv = p->p_realtimer.it_value; 956 timevalsub(&ntv, &ctv); 957 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 958 realitexpire, p); 959 goto done; 960 } 961 } 962 done: 963 lwkt_reltoken(&p->p_token); 964 PRELE(p); 965 } 966 967 /* 968 * Used to validate itimer timeouts and utimes*() timespecs. 969 */ 970 int 971 itimerfix(struct timeval *tv) 972 { 973 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 974 return (EINVAL); 975 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 976 tv->tv_usec = ustick; 977 return (0); 978 } 979 980 /* 981 * Used to validate timeouts and utimes*() timespecs. 982 */ 983 int 984 itimespecfix(struct timespec *ts) 985 { 986 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000ULL) 987 return (EINVAL); 988 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < nstick) 989 ts->tv_nsec = nstick; 990 return (0); 991 } 992 993 /* 994 * Decrement an interval timer by a specified number 995 * of microseconds, which must be less than a second, 996 * i.e. < 1000000. If the timer expires, then reload 997 * it. In this case, carry over (usec - old value) to 998 * reduce the value reloaded into the timer so that 999 * the timer does not drift. This routine assumes 1000 * that it is called in a context where the timers 1001 * on which it is operating cannot change in value. 1002 */ 1003 int 1004 itimerdecr(struct itimerval *itp, int usec) 1005 { 1006 1007 if (itp->it_value.tv_usec < usec) { 1008 if (itp->it_value.tv_sec == 0) { 1009 /* expired, and already in next interval */ 1010 usec -= itp->it_value.tv_usec; 1011 goto expire; 1012 } 1013 itp->it_value.tv_usec += 1000000; 1014 itp->it_value.tv_sec--; 1015 } 1016 itp->it_value.tv_usec -= usec; 1017 usec = 0; 1018 if (timevalisset(&itp->it_value)) 1019 return (1); 1020 /* expired, exactly at end of interval */ 1021 expire: 1022 if (timevalisset(&itp->it_interval)) { 1023 itp->it_value = itp->it_interval; 1024 itp->it_value.tv_usec -= usec; 1025 if (itp->it_value.tv_usec < 0) { 1026 itp->it_value.tv_usec += 1000000; 1027 itp->it_value.tv_sec--; 1028 } 1029 } else 1030 itp->it_value.tv_usec = 0; /* sec is already 0 */ 1031 return (0); 1032 } 1033 1034 /* 1035 * Add and subtract routines for timevals. 1036 * N.B.: subtract routine doesn't deal with 1037 * results which are before the beginning, 1038 * it just gets very confused in this case. 1039 * Caveat emptor. 1040 */ 1041 void 1042 timevaladd(struct timeval *t1, const struct timeval *t2) 1043 { 1044 1045 t1->tv_sec += t2->tv_sec; 1046 t1->tv_usec += t2->tv_usec; 1047 timevalfix(t1); 1048 } 1049 1050 void 1051 timevalsub(struct timeval *t1, const struct timeval *t2) 1052 { 1053 1054 t1->tv_sec -= t2->tv_sec; 1055 t1->tv_usec -= t2->tv_usec; 1056 timevalfix(t1); 1057 } 1058 1059 static void 1060 timevalfix(struct timeval *t1) 1061 { 1062 1063 if (t1->tv_usec < 0) { 1064 t1->tv_sec--; 1065 t1->tv_usec += 1000000; 1066 } 1067 if (t1->tv_usec >= 1000000) { 1068 t1->tv_sec++; 1069 t1->tv_usec -= 1000000; 1070 } 1071 } 1072 1073 /* 1074 * ratecheck(): simple time-based rate-limit checking. 1075 */ 1076 int 1077 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1078 { 1079 struct timeval tv, delta; 1080 int rv = 0; 1081 1082 getmicrouptime(&tv); /* NB: 10ms precision */ 1083 delta = tv; 1084 timevalsub(&delta, lasttime); 1085 1086 /* 1087 * check for 0,0 is so that the message will be seen at least once, 1088 * even if interval is huge. 1089 */ 1090 if (timevalcmp(&delta, mininterval, >=) || 1091 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1092 *lasttime = tv; 1093 rv = 1; 1094 } 1095 1096 return (rv); 1097 } 1098 1099 /* 1100 * ppsratecheck(): packets (or events) per second limitation. 1101 * 1102 * Return 0 if the limit is to be enforced (e.g. the caller 1103 * should drop a packet because of the rate limitation). 1104 * 1105 * maxpps of 0 always causes zero to be returned. maxpps of -1 1106 * always causes 1 to be returned; this effectively defeats rate 1107 * limiting. 1108 * 1109 * Note that we maintain the struct timeval for compatibility 1110 * with other bsd systems. We reuse the storage and just monitor 1111 * clock ticks for minimal overhead. 1112 */ 1113 int 1114 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1115 { 1116 int now; 1117 1118 /* 1119 * Reset the last time and counter if this is the first call 1120 * or more than a second has passed since the last update of 1121 * lasttime. 1122 */ 1123 now = ticks; 1124 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1125 lasttime->tv_sec = now; 1126 *curpps = 1; 1127 return (maxpps != 0); 1128 } else { 1129 (*curpps)++; /* NB: ignore potential overflow */ 1130 return (maxpps < 0 || *curpps < maxpps); 1131 } 1132 } 1133