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 <sys/upmap.h> 49 #include <vm/vm.h> 50 #include <vm/vm_extern.h> 51 52 #include <sys/msgport2.h> 53 #include <sys/spinlock2.h> 54 #include <sys/thread2.h> 55 56 extern struct spinlock ntp_spin; 57 58 #define CPUCLOCK_BIT 0x80000000 59 #define CPUCLOCK_ID_MASK ~CPUCLOCK_BIT 60 #define CPUCLOCK2LWPID(clock_id) (((clockid_t)(clock_id) >> 32) & CPUCLOCK_ID_MASK) 61 #define CPUCLOCK2PID(clock_id) ((clock_id) & CPUCLOCK_ID_MASK) 62 #define MAKE_CPUCLOCK(pid, lwp_id) ((clockid_t)(lwp_id) << 32 | (pid) | CPUCLOCK_BIT) 63 64 struct timezone tz; 65 66 /* 67 * Time of day and interval timer support. 68 * 69 * These routines provide the kernel entry points to get and set 70 * the time-of-day and per-process interval timers. Subroutines 71 * here provide support for adding and subtracting timeval structures 72 * and decrementing interval timers, optionally reloading the interval 73 * timers when they expire. 74 */ 75 76 static int settime(struct timeval *); 77 static void timevalfix(struct timeval *); 78 static void realitexpire(void *arg); 79 80 static int sysctl_gettimeofday_quick(SYSCTL_HANDLER_ARGS); 81 82 83 /* 84 * Nanosleep tries very hard to sleep for a precisely requested time 85 * interval, down to 1uS. The administrator can impose a minimum delay 86 * and a delay below which we hard-loop instead of initiate a timer 87 * interrupt and sleep. 88 * 89 * For machines under high loads it might be beneficial to increase min_us 90 * to e.g. 1000uS (1ms) so spining processes sleep meaningfully. 91 */ 92 static int nanosleep_min_us = 10; 93 static int nanosleep_hard_us = 100; 94 static int gettimeofday_quick = 0; 95 SYSCTL_INT(_kern, OID_AUTO, nanosleep_min_us, CTLFLAG_RW, 96 &nanosleep_min_us, 0, ""); 97 SYSCTL_INT(_kern, OID_AUTO, nanosleep_hard_us, CTLFLAG_RW, 98 &nanosleep_hard_us, 0, ""); 99 SYSCTL_PROC(_kern, OID_AUTO, gettimeofday_quick, CTLTYPE_INT | CTLFLAG_RW, 100 0, 0, sysctl_gettimeofday_quick, "I", "Quick mode gettimeofday"); 101 102 static struct lock masterclock_lock = LOCK_INITIALIZER("mstrclk", 0, 0); 103 104 static int 105 settime(struct timeval *tv) 106 { 107 struct timeval delta, tv1, tv2; 108 static struct timeval maxtime, laststep; 109 struct timespec ts; 110 int origcpu; 111 112 if ((origcpu = mycpu->gd_cpuid) != 0) 113 lwkt_setcpu_self(globaldata_find(0)); 114 115 crit_enter(); 116 microtime(&tv1); 117 delta = *tv; 118 timevalsub(&delta, &tv1); 119 120 /* 121 * If the system is secure, we do not allow the time to be 122 * set to a value earlier than 1 second less than the highest 123 * time we have yet seen. The worst a miscreant can do in 124 * this circumstance is "freeze" time. He couldn't go 125 * back to the past. 126 * 127 * We similarly do not allow the clock to be stepped more 128 * than one second, nor more than once per second. This allows 129 * a miscreant to make the clock march double-time, but no worse. 130 */ 131 if (securelevel > 1) { 132 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 133 /* 134 * Update maxtime to latest time we've seen. 135 */ 136 if (tv1.tv_sec > maxtime.tv_sec) 137 maxtime = tv1; 138 tv2 = *tv; 139 timevalsub(&tv2, &maxtime); 140 if (tv2.tv_sec < -1) { 141 tv->tv_sec = maxtime.tv_sec - 1; 142 kprintf("Time adjustment clamped to -1 second\n"); 143 } 144 } else { 145 if (tv1.tv_sec == laststep.tv_sec) { 146 crit_exit(); 147 return (EPERM); 148 } 149 if (delta.tv_sec > 1) { 150 tv->tv_sec = tv1.tv_sec + 1; 151 kprintf("Time adjustment clamped to +1 second\n"); 152 } 153 laststep = *tv; 154 } 155 } 156 157 ts.tv_sec = tv->tv_sec; 158 ts.tv_nsec = tv->tv_usec * 1000; 159 set_timeofday(&ts); 160 crit_exit(); 161 162 if (origcpu != 0) 163 lwkt_setcpu_self(globaldata_find(origcpu)); 164 165 resettodr(); 166 return (0); 167 } 168 169 static void 170 get_process_cputime(struct proc *p, struct timespec *ats) 171 { 172 struct rusage ru; 173 174 lwkt_gettoken(&p->p_token); 175 calcru_proc(p, &ru); 176 lwkt_reltoken(&p->p_token); 177 timevaladd(&ru.ru_utime, &ru.ru_stime); 178 TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats); 179 } 180 181 static void 182 get_process_usertime(struct proc *p, struct timespec *ats) 183 { 184 struct rusage ru; 185 186 lwkt_gettoken(&p->p_token); 187 calcru_proc(p, &ru); 188 lwkt_reltoken(&p->p_token); 189 TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats); 190 } 191 192 static void 193 get_thread_cputime(struct thread *td, struct timespec *ats) 194 { 195 struct timeval sys, user; 196 197 calcru(td->td_lwp, &user, &sys); 198 timevaladd(&user, &sys); 199 TIMEVAL_TO_TIMESPEC(&user, ats); 200 } 201 202 /* 203 * MPSAFE 204 */ 205 int 206 kern_clock_gettime(clockid_t clock_id, struct timespec *ats) 207 { 208 struct proc *p; 209 struct lwp *lp; 210 lwpid_t lwp_id; 211 212 p = curproc; 213 switch(clock_id) { 214 case CLOCK_REALTIME: 215 case CLOCK_REALTIME_PRECISE: 216 nanotime(ats); 217 break; 218 case CLOCK_REALTIME_FAST: 219 getnanotime(ats); 220 break; 221 case CLOCK_MONOTONIC: 222 case CLOCK_MONOTONIC_PRECISE: 223 case CLOCK_UPTIME: 224 case CLOCK_UPTIME_PRECISE: 225 nanouptime(ats); 226 break; 227 case CLOCK_MONOTONIC_FAST: 228 case CLOCK_UPTIME_FAST: 229 getnanouptime(ats); 230 break; 231 case CLOCK_VIRTUAL: 232 get_process_usertime(p, ats); 233 break; 234 case CLOCK_PROF: 235 case CLOCK_PROCESS_CPUTIME_ID: 236 get_process_cputime(p, ats); 237 break; 238 case CLOCK_SECOND: 239 ats->tv_sec = time_second; 240 ats->tv_nsec = 0; 241 break; 242 case CLOCK_THREAD_CPUTIME_ID: 243 get_thread_cputime(curthread, ats); 244 break; 245 default: 246 if ((clock_id & CPUCLOCK_BIT) == 0) 247 return (EINVAL); 248 if ((p = pfind(CPUCLOCK2PID(clock_id))) == NULL) 249 return (EINVAL); 250 lwp_id = CPUCLOCK2LWPID(clock_id); 251 if (lwp_id == 0) { 252 get_process_cputime(p, ats); 253 } else { 254 lwkt_gettoken(&p->p_token); 255 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, lwp_id); 256 if (lp == NULL) { 257 lwkt_reltoken(&p->p_token); 258 PRELE(p); 259 return (EINVAL); 260 } 261 get_thread_cputime(lp->lwp_thread, ats); 262 lwkt_reltoken(&p->p_token); 263 } 264 PRELE(p); 265 } 266 return (0); 267 } 268 269 /* 270 * MPSAFE 271 */ 272 int 273 sys_clock_gettime(struct clock_gettime_args *uap) 274 { 275 struct timespec ats; 276 int error; 277 278 error = kern_clock_gettime(uap->clock_id, &ats); 279 if (error == 0) 280 error = copyout(&ats, uap->tp, sizeof(ats)); 281 282 return (error); 283 } 284 285 int 286 kern_clock_settime(clockid_t clock_id, struct timespec *ats) 287 { 288 struct thread *td = curthread; 289 struct timeval atv; 290 int error; 291 292 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 293 return (error); 294 if (clock_id != CLOCK_REALTIME) 295 return (EINVAL); 296 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000) 297 return (EINVAL); 298 299 lockmgr(&masterclock_lock, LK_EXCLUSIVE); 300 TIMESPEC_TO_TIMEVAL(&atv, ats); 301 error = settime(&atv); 302 lockmgr(&masterclock_lock, LK_RELEASE); 303 304 return (error); 305 } 306 307 /* 308 * MPALMOSTSAFE 309 */ 310 int 311 sys_clock_settime(struct clock_settime_args *uap) 312 { 313 struct timespec ats; 314 int error; 315 316 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 317 return (error); 318 319 error = kern_clock_settime(uap->clock_id, &ats); 320 321 return (error); 322 } 323 324 /* 325 * MPSAFE 326 */ 327 int 328 kern_clock_getres(clockid_t clock_id, struct timespec *ts) 329 { 330 ts->tv_sec = 0; 331 switch(clock_id) { 332 case CLOCK_REALTIME: 333 case CLOCK_REALTIME_FAST: 334 case CLOCK_REALTIME_PRECISE: 335 case CLOCK_MONOTONIC: 336 case CLOCK_MONOTONIC_FAST: 337 case CLOCK_MONOTONIC_PRECISE: 338 case CLOCK_UPTIME: 339 case CLOCK_UPTIME_FAST: 340 case CLOCK_UPTIME_PRECISE: 341 /* 342 * Round up the result of the division cheaply 343 * by adding 1. Rounding up is especially important 344 * if rounding down would give 0. Perfect rounding 345 * is unimportant. 346 */ 347 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1; 348 break; 349 case CLOCK_VIRTUAL: 350 case CLOCK_PROF: 351 /* Accurately round up here because we can do so cheaply. */ 352 ts->tv_nsec = (1000000000 + hz - 1) / hz; 353 break; 354 case CLOCK_SECOND: 355 ts->tv_sec = 1; 356 ts->tv_nsec = 0; 357 break; 358 case CLOCK_THREAD_CPUTIME_ID: 359 case CLOCK_PROCESS_CPUTIME_ID: 360 ts->tv_nsec = 1000; 361 break; 362 default: 363 if ((clock_id & CPUCLOCK_BIT) != 0) 364 ts->tv_nsec = 1000; 365 else 366 return (EINVAL); 367 } 368 369 return (0); 370 } 371 372 /* 373 * MPSAFE 374 */ 375 int 376 sys_clock_getres(struct clock_getres_args *uap) 377 { 378 int error; 379 struct timespec ts; 380 381 error = kern_clock_getres(uap->clock_id, &ts); 382 if (error == 0) 383 error = copyout(&ts, uap->tp, sizeof(ts)); 384 385 return (error); 386 } 387 388 static int 389 kern_getcpuclockid(pid_t pid, lwpid_t lwp_id, clockid_t *clock_id) 390 { 391 struct proc *p; 392 int error = 0; 393 394 if (pid == 0) { 395 p = curproc; 396 pid = p->p_pid; 397 PHOLD(p); 398 } else { 399 p = pfind(pid); 400 if (p == NULL) 401 return (ESRCH); 402 } 403 /* lwp_id can be 0 when called by clock_getcpuclockid() */ 404 if (lwp_id < 0) { 405 error = EINVAL; 406 goto out; 407 } 408 lwkt_gettoken(&p->p_token); 409 if (lwp_id > 0 && 410 lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, lwp_id) == NULL) { 411 lwkt_reltoken(&p->p_token); 412 error = ESRCH; 413 goto out; 414 } 415 *clock_id = MAKE_CPUCLOCK(pid, lwp_id); 416 lwkt_reltoken(&p->p_token); 417 out: 418 PRELE(p); 419 return (error); 420 } 421 422 int 423 sys_getcpuclockid(struct getcpuclockid_args *uap) 424 { 425 clockid_t clk_id; 426 int error; 427 428 error = kern_getcpuclockid(uap->pid, uap->lwp_id, &clk_id); 429 if (error == 0) 430 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 431 432 return (error); 433 } 434 435 /* 436 * nanosleep1() 437 * 438 * This is a general helper function for nanosleep() (aka sleep() aka 439 * usleep()). 440 * 441 * If there is less then one tick's worth of time left and 442 * we haven't done a yield, or the remaining microseconds is 443 * ridiculously low, do a yield. This avoids having 444 * to deal with systimer overheads when the system is under 445 * heavy loads. If we have done a yield already then use 446 * a systimer and an uninterruptable thread wait. 447 * 448 * If there is more then a tick's worth of time left, 449 * calculate the baseline ticks and use an interruptable 450 * tsleep, then handle the fine-grained delay on the next 451 * loop. This usually results in two sleeps occuring, a long one 452 * and a short one. 453 * 454 * MPSAFE 455 */ 456 static void 457 ns1_systimer(systimer_t info, int in_ipi __unused, 458 struct intrframe *frame __unused) 459 { 460 lwkt_schedule(info->data); 461 } 462 463 int 464 nanosleep1(struct timespec *rqt, struct timespec *rmt) 465 { 466 static int nanowait; 467 struct timespec ts, ts2, ts3; 468 struct timeval tv; 469 int error; 470 471 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 472 return (EINVAL); 473 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 474 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 475 return (0); 476 nanouptime(&ts); 477 timespecadd(&ts, rqt, &ts); /* ts = target timestamp compare */ 478 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 479 480 for (;;) { 481 int ticks; 482 struct systimer info; 483 484 ticks = tv.tv_usec / ustick; /* approximate */ 485 486 if (tv.tv_sec == 0 && ticks == 0) { 487 thread_t td = curthread; 488 if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us) 489 tv.tv_usec = nanosleep_min_us; 490 if (tv.tv_usec < nanosleep_hard_us) { 491 lwkt_user_yield(); 492 cpu_pause(); 493 } else { 494 crit_enter_quick(td); 495 systimer_init_oneshot(&info, ns1_systimer, 496 td, tv.tv_usec); 497 lwkt_deschedule_self(td); 498 crit_exit_quick(td); 499 lwkt_switch(); 500 systimer_del(&info); /* make sure it's gone */ 501 } 502 error = iscaught(td->td_lwp); 503 } else if (tv.tv_sec == 0) { 504 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 505 } else { 506 ticks = tvtohz_low(&tv); /* also handles overflow */ 507 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 508 } 509 nanouptime(&ts2); 510 if (error && error != EWOULDBLOCK) { 511 if (error == ERESTART) 512 error = EINTR; 513 if (rmt != NULL) { 514 timespecsub(&ts, &ts2, &ts); 515 if (ts.tv_sec < 0) 516 timespecclear(&ts); 517 *rmt = ts; 518 } 519 return (error); 520 } 521 if (timespeccmp(&ts2, &ts, >=)) 522 return (0); 523 timespecsub(&ts, &ts2, &ts3); 524 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 525 } 526 } 527 528 /* 529 * MPSAFE 530 */ 531 int 532 sys_nanosleep(struct nanosleep_args *uap) 533 { 534 int error; 535 struct timespec rqt; 536 struct timespec rmt; 537 538 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 539 if (error) 540 return (error); 541 542 error = nanosleep1(&rqt, &rmt); 543 544 /* 545 * copyout the residual if nanosleep was interrupted. 546 */ 547 if (error && uap->rmtp) { 548 int error2; 549 550 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 551 if (error2) 552 error = error2; 553 } 554 return (error); 555 } 556 557 /* 558 * The gettimeofday() system call is supposed to return a fine-grained 559 * realtime stamp. However, acquiring a fine-grained stamp can create a 560 * bottleneck when multiple cpu cores are trying to accessing e.g. the 561 * HPET hardware timer all at the same time, so we have a sysctl that 562 * allows its behavior to be changed to a more coarse-grained timestamp 563 * which does not have to access a hardware timer. 564 */ 565 int 566 sys_gettimeofday(struct gettimeofday_args *uap) 567 { 568 struct timeval atv; 569 int error = 0; 570 571 if (uap->tp) { 572 if (gettimeofday_quick) 573 getmicrotime(&atv); 574 else 575 microtime(&atv); 576 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 577 sizeof (atv)))) 578 return (error); 579 } 580 if (uap->tzp) 581 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 582 sizeof (tz)); 583 return (error); 584 } 585 586 /* 587 * MPALMOSTSAFE 588 */ 589 int 590 sys_settimeofday(struct settimeofday_args *uap) 591 { 592 struct thread *td = curthread; 593 struct timeval atv; 594 struct timezone atz; 595 int error; 596 597 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 598 return (error); 599 /* 600 * Verify all parameters before changing time. 601 * 602 * XXX: We do not allow the time to be set to 0.0, which also by 603 * happy coincidence works around a pkgsrc bulk build bug. 604 */ 605 if (uap->tv) { 606 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 607 sizeof(atv)))) 608 return (error); 609 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 610 return (EINVAL); 611 if (atv.tv_sec == 0 && atv.tv_usec == 0) 612 return (EINVAL); 613 } 614 if (uap->tzp && 615 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 616 return (error); 617 618 lockmgr(&masterclock_lock, LK_EXCLUSIVE); 619 if (uap->tv && (error = settime(&atv))) { 620 lockmgr(&masterclock_lock, LK_RELEASE); 621 return (error); 622 } 623 lockmgr(&masterclock_lock, LK_RELEASE); 624 625 if (uap->tzp) 626 tz = atz; 627 return (0); 628 } 629 630 /* 631 * WARNING! Run with ntp_spin held 632 */ 633 static void 634 kern_adjtime_common(void) 635 { 636 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 637 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 638 ntp_tick_delta = ntp_delta; 639 else if (ntp_delta > ntp_big_delta) 640 ntp_tick_delta = 10 * ntp_default_tick_delta; 641 else if (ntp_delta < -ntp_big_delta) 642 ntp_tick_delta = -10 * ntp_default_tick_delta; 643 else if (ntp_delta > 0) 644 ntp_tick_delta = ntp_default_tick_delta; 645 else 646 ntp_tick_delta = -ntp_default_tick_delta; 647 } 648 649 void 650 kern_adjtime(int64_t delta, int64_t *odelta) 651 { 652 spin_lock(&ntp_spin); 653 *odelta = ntp_delta; 654 ntp_delta = delta; 655 kern_adjtime_common(); 656 spin_unlock(&ntp_spin); 657 } 658 659 static void 660 kern_get_ntp_delta(int64_t *delta) 661 { 662 *delta = ntp_delta; 663 } 664 665 void 666 kern_reladjtime(int64_t delta) 667 { 668 spin_lock(&ntp_spin); 669 ntp_delta += delta; 670 kern_adjtime_common(); 671 spin_unlock(&ntp_spin); 672 } 673 674 static void 675 kern_adjfreq(int64_t rate) 676 { 677 spin_lock(&ntp_spin); 678 ntp_tick_permanent = rate; 679 spin_unlock(&ntp_spin); 680 } 681 682 /* 683 * MPALMOSTSAFE 684 */ 685 int 686 sys_adjtime(struct adjtime_args *uap) 687 { 688 struct thread *td = curthread; 689 struct timeval atv; 690 int64_t ndelta, odelta; 691 int error; 692 693 if ((error = priv_check(td, PRIV_ADJTIME))) 694 return (error); 695 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 696 if (error) 697 return (error); 698 699 /* 700 * Compute the total correction and the rate at which to apply it. 701 * Round the adjustment down to a whole multiple of the per-tick 702 * delta, so that after some number of incremental changes in 703 * hardclock(), tickdelta will become zero, lest the correction 704 * overshoot and start taking us away from the desired final time. 705 */ 706 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 707 kern_adjtime(ndelta, &odelta); 708 709 if (uap->olddelta) { 710 atv.tv_sec = odelta / 1000000000; 711 atv.tv_usec = odelta % 1000000000 / 1000; 712 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 713 } 714 return (0); 715 } 716 717 static int 718 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 719 { 720 int64_t delta; 721 int error; 722 723 if (req->newptr != NULL) { 724 if (priv_check(curthread, PRIV_ROOT)) 725 return (EPERM); 726 error = SYSCTL_IN(req, &delta, sizeof(delta)); 727 if (error) 728 return (error); 729 kern_reladjtime(delta); 730 } 731 732 if (req->oldptr) 733 kern_get_ntp_delta(&delta); 734 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 735 return (error); 736 } 737 738 /* 739 * delta is in nanoseconds. 740 */ 741 static int 742 sysctl_delta(SYSCTL_HANDLER_ARGS) 743 { 744 int64_t delta, old_delta; 745 int error; 746 747 if (req->newptr != NULL) { 748 if (priv_check(curthread, PRIV_ROOT)) 749 return (EPERM); 750 error = SYSCTL_IN(req, &delta, sizeof(delta)); 751 if (error) 752 return (error); 753 kern_adjtime(delta, &old_delta); 754 } 755 756 if (req->oldptr != NULL) 757 kern_get_ntp_delta(&old_delta); 758 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 759 return (error); 760 } 761 762 /* 763 * frequency is in nanoseconds per second shifted left 32. 764 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 765 */ 766 static int 767 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 768 { 769 int64_t freqdelta; 770 int error; 771 772 if (req->newptr != NULL) { 773 if (priv_check(curthread, PRIV_ROOT)) 774 return (EPERM); 775 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 776 if (error) 777 return (error); 778 779 freqdelta /= hz; 780 kern_adjfreq(freqdelta); 781 } 782 783 if (req->oldptr != NULL) 784 freqdelta = ntp_tick_permanent * hz; 785 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 786 if (error) 787 return (error); 788 789 return (0); 790 } 791 792 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 793 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 794 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 795 sysctl_adjfreq, "Q", "permanent correction per second"); 796 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 797 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 798 sysctl_delta, "Q", "one-time delta"); 799 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 800 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 801 "threshold for fast adjustment"); 802 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 803 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 804 "per-tick adjustment"); 805 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 806 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 807 "default per-tick adjustment"); 808 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 809 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 810 "next leap second"); 811 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 812 &ntp_leap_insert, 0, "insert or remove leap second"); 813 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 814 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 815 sysctl_adjtime, "Q", "relative adjust for delta"); 816 817 /* 818 * Get value of an interval timer. The process virtual and 819 * profiling virtual time timers are kept in the p_stats area, since 820 * they can be swapped out. These are kept internally in the 821 * way they are specified externally: in time until they expire. 822 * 823 * The real time interval timer is kept in the process table slot 824 * for the process, and its value (it_value) is kept as an 825 * absolute time rather than as a delta, so that it is easy to keep 826 * periodic real-time signals from drifting. 827 * 828 * Virtual time timers are processed in the hardclock() routine of 829 * kern_clock.c. The real time timer is processed by a timeout 830 * routine, called from the softclock() routine. Since a callout 831 * may be delayed in real time due to interrupt processing in the system, 832 * it is possible for the real time timeout routine (realitexpire, given below), 833 * to be delayed in real time past when it is supposed to occur. It 834 * does not suffice, therefore, to reload the real timer .it_value from the 835 * real time timers .it_interval. Rather, we compute the next time in 836 * absolute time the timer should go off. 837 * 838 * MPALMOSTSAFE 839 */ 840 int 841 sys_getitimer(struct getitimer_args *uap) 842 { 843 struct proc *p = curproc; 844 struct timeval ctv; 845 struct itimerval aitv; 846 847 if (uap->which > ITIMER_PROF) 848 return (EINVAL); 849 lwkt_gettoken(&p->p_token); 850 if (uap->which == ITIMER_REAL) { 851 /* 852 * Convert from absolute to relative time in .it_value 853 * part of real time timer. If time for real time timer 854 * has passed return 0, else return difference between 855 * current time and time for the timer to go off. 856 */ 857 aitv = p->p_realtimer; 858 if (timevalisset(&aitv.it_value)) { 859 getmicrouptime(&ctv); 860 if (timevalcmp(&aitv.it_value, &ctv, <)) 861 timevalclear(&aitv.it_value); 862 else 863 timevalsub(&aitv.it_value, &ctv); 864 } 865 } else { 866 aitv = p->p_timer[uap->which]; 867 } 868 lwkt_reltoken(&p->p_token); 869 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 870 } 871 872 /* 873 * MPALMOSTSAFE 874 */ 875 int 876 sys_setitimer(struct setitimer_args *uap) 877 { 878 struct itimerval aitv; 879 struct timeval ctv; 880 struct itimerval *itvp; 881 struct proc *p = curproc; 882 int error; 883 884 if (uap->which > ITIMER_PROF) 885 return (EINVAL); 886 itvp = uap->itv; 887 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 888 sizeof(struct itimerval)))) 889 return (error); 890 if ((uap->itv = uap->oitv) && 891 (error = sys_getitimer((struct getitimer_args *)uap))) 892 return (error); 893 if (itvp == NULL) 894 return (0); 895 if (itimerfix(&aitv.it_value)) 896 return (EINVAL); 897 if (!timevalisset(&aitv.it_value)) 898 timevalclear(&aitv.it_interval); 899 else if (itimerfix(&aitv.it_interval)) 900 return (EINVAL); 901 lwkt_gettoken(&p->p_token); 902 if (uap->which == ITIMER_REAL) { 903 if (timevalisset(&p->p_realtimer.it_value)) 904 callout_cancel(&p->p_ithandle); 905 if (timevalisset(&aitv.it_value)) 906 callout_reset(&p->p_ithandle, 907 tvtohz_high(&aitv.it_value), realitexpire, p); 908 getmicrouptime(&ctv); 909 timevaladd(&aitv.it_value, &ctv); 910 p->p_realtimer = aitv; 911 } else { 912 p->p_timer[uap->which] = aitv; 913 switch(uap->which) { 914 case ITIMER_VIRTUAL: 915 p->p_flags &= ~P_SIGVTALRM; 916 break; 917 case ITIMER_PROF: 918 p->p_flags &= ~P_SIGPROF; 919 break; 920 } 921 } 922 lwkt_reltoken(&p->p_token); 923 return (0); 924 } 925 926 /* 927 * Real interval timer expired: 928 * send process whose timer expired an alarm signal. 929 * If time is not set up to reload, then just return. 930 * Else compute next time timer should go off which is > current time. 931 * This is where delay in processing this timeout causes multiple 932 * SIGALRM calls to be compressed into one. 933 * tvtohz_high() always adds 1 to allow for the time until the next clock 934 * interrupt being strictly less than 1 clock tick, but we don't want 935 * that here since we want to appear to be in sync with the clock 936 * interrupt even when we're delayed. 937 */ 938 static 939 void 940 realitexpire(void *arg) 941 { 942 struct proc *p; 943 struct timeval ctv, ntv; 944 945 p = (struct proc *)arg; 946 PHOLD(p); 947 lwkt_gettoken(&p->p_token); 948 ksignal(p, SIGALRM); 949 if (!timevalisset(&p->p_realtimer.it_interval)) { 950 timevalclear(&p->p_realtimer.it_value); 951 goto done; 952 } 953 for (;;) { 954 timevaladd(&p->p_realtimer.it_value, 955 &p->p_realtimer.it_interval); 956 getmicrouptime(&ctv); 957 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 958 ntv = p->p_realtimer.it_value; 959 timevalsub(&ntv, &ctv); 960 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 961 realitexpire, p); 962 goto done; 963 } 964 } 965 done: 966 lwkt_reltoken(&p->p_token); 967 PRELE(p); 968 } 969 970 /* 971 * Used to validate itimer timeouts and utimes*() timespecs. 972 */ 973 int 974 itimerfix(struct timeval *tv) 975 { 976 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 977 return (EINVAL); 978 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 979 tv->tv_usec = ustick; 980 return (0); 981 } 982 983 /* 984 * Used to validate timeouts and utimes*() timespecs. 985 */ 986 int 987 itimespecfix(struct timespec *ts) 988 { 989 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000ULL) 990 return (EINVAL); 991 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < nstick) 992 ts->tv_nsec = nstick; 993 return (0); 994 } 995 996 /* 997 * Decrement an interval timer by a specified number 998 * of microseconds, which must be less than a second, 999 * i.e. < 1000000. If the timer expires, then reload 1000 * it. In this case, carry over (usec - old value) to 1001 * reduce the value reloaded into the timer so that 1002 * the timer does not drift. This routine assumes 1003 * that it is called in a context where the timers 1004 * on which it is operating cannot change in value. 1005 */ 1006 int 1007 itimerdecr(struct itimerval *itp, int usec) 1008 { 1009 1010 if (itp->it_value.tv_usec < usec) { 1011 if (itp->it_value.tv_sec == 0) { 1012 /* expired, and already in next interval */ 1013 usec -= itp->it_value.tv_usec; 1014 goto expire; 1015 } 1016 itp->it_value.tv_usec += 1000000; 1017 itp->it_value.tv_sec--; 1018 } 1019 itp->it_value.tv_usec -= usec; 1020 usec = 0; 1021 if (timevalisset(&itp->it_value)) 1022 return (1); 1023 /* expired, exactly at end of interval */ 1024 expire: 1025 if (timevalisset(&itp->it_interval)) { 1026 itp->it_value = itp->it_interval; 1027 itp->it_value.tv_usec -= usec; 1028 if (itp->it_value.tv_usec < 0) { 1029 itp->it_value.tv_usec += 1000000; 1030 itp->it_value.tv_sec--; 1031 } 1032 } else 1033 itp->it_value.tv_usec = 0; /* sec is already 0 */ 1034 return (0); 1035 } 1036 1037 /* 1038 * Add and subtract routines for timevals. 1039 * N.B.: subtract routine doesn't deal with 1040 * results which are before the beginning, 1041 * it just gets very confused in this case. 1042 * Caveat emptor. 1043 */ 1044 void 1045 timevaladd(struct timeval *t1, const struct timeval *t2) 1046 { 1047 1048 t1->tv_sec += t2->tv_sec; 1049 t1->tv_usec += t2->tv_usec; 1050 timevalfix(t1); 1051 } 1052 1053 void 1054 timevalsub(struct timeval *t1, const struct timeval *t2) 1055 { 1056 1057 t1->tv_sec -= t2->tv_sec; 1058 t1->tv_usec -= t2->tv_usec; 1059 timevalfix(t1); 1060 } 1061 1062 static void 1063 timevalfix(struct timeval *t1) 1064 { 1065 1066 if (t1->tv_usec < 0) { 1067 t1->tv_sec--; 1068 t1->tv_usec += 1000000; 1069 } 1070 if (t1->tv_usec >= 1000000) { 1071 t1->tv_sec++; 1072 t1->tv_usec -= 1000000; 1073 } 1074 } 1075 1076 /* 1077 * ratecheck(): simple time-based rate-limit checking. 1078 */ 1079 int 1080 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1081 { 1082 struct timeval tv, delta; 1083 int rv = 0; 1084 1085 getmicrouptime(&tv); /* NB: 10ms precision */ 1086 delta = tv; 1087 timevalsub(&delta, lasttime); 1088 1089 /* 1090 * check for 0,0 is so that the message will be seen at least once, 1091 * even if interval is huge. 1092 */ 1093 if (timevalcmp(&delta, mininterval, >=) || 1094 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1095 *lasttime = tv; 1096 rv = 1; 1097 } 1098 1099 return (rv); 1100 } 1101 1102 /* 1103 * ppsratecheck(): packets (or events) per second limitation. 1104 * 1105 * Return 0 if the limit is to be enforced (e.g. the caller 1106 * should drop a packet because of the rate limitation). 1107 * 1108 * maxpps of 0 always causes zero to be returned. maxpps of -1 1109 * always causes 1 to be returned; this effectively defeats rate 1110 * limiting. 1111 * 1112 * Note that we maintain the struct timeval for compatibility 1113 * with other bsd systems. We reuse the storage and just monitor 1114 * clock ticks for minimal overhead. 1115 */ 1116 int 1117 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1118 { 1119 int now; 1120 1121 /* 1122 * Reset the last time and counter if this is the first call 1123 * or more than a second has passed since the last update of 1124 * lasttime. 1125 */ 1126 now = ticks; 1127 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1128 lasttime->tv_sec = now; 1129 *curpps = 1; 1130 return (maxpps != 0); 1131 } else { 1132 (*curpps)++; /* NB: ignore potential overflow */ 1133 return (maxpps < 0 || *curpps < maxpps); 1134 } 1135 } 1136 1137 static int 1138 sysctl_gettimeofday_quick(SYSCTL_HANDLER_ARGS) 1139 { 1140 int error; 1141 int gtod; 1142 1143 gtod = gettimeofday_quick; 1144 error = sysctl_handle_int(oidp, >od, 0, req); 1145 if (error || req->newptr == NULL) 1146 return error; 1147 gettimeofday_quick = gtod; 1148 if (kpmap) 1149 kpmap->fast_gtod = gtod; 1150 return 0; 1151 } 1152