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; 331 332 switch(clock_id) { 333 case CLOCK_REALTIME: 334 case CLOCK_REALTIME_FAST: 335 case CLOCK_REALTIME_PRECISE: 336 case CLOCK_MONOTONIC: 337 case CLOCK_MONOTONIC_FAST: 338 case CLOCK_MONOTONIC_PRECISE: 339 case CLOCK_UPTIME: 340 case CLOCK_UPTIME_FAST: 341 case CLOCK_UPTIME_PRECISE: 342 /* 343 * Minimum reportable resolution is 1ns. Rounding is 344 * otherwise unimportant. 345 */ 346 ts->tv_nsec = 999999999 / sys_cputimer->freq + 1; 347 break; 348 case CLOCK_VIRTUAL: 349 case CLOCK_PROF: 350 /* Accurately round up here because we can do so cheaply. */ 351 ts->tv_nsec = (1000000000 + hz - 1) / hz; 352 break; 353 case CLOCK_SECOND: 354 ts->tv_sec = 1; 355 ts->tv_nsec = 0; 356 break; 357 case CLOCK_THREAD_CPUTIME_ID: 358 case CLOCK_PROCESS_CPUTIME_ID: 359 ts->tv_nsec = 1000; 360 break; 361 default: 362 if ((clock_id & CPUCLOCK_BIT) != 0) 363 ts->tv_nsec = 1000; 364 else 365 return (EINVAL); 366 } 367 368 return (0); 369 } 370 371 /* 372 * MPSAFE 373 */ 374 int 375 sys_clock_getres(struct clock_getres_args *uap) 376 { 377 int error; 378 struct timespec ts; 379 380 error = kern_clock_getres(uap->clock_id, &ts); 381 if (error == 0) 382 error = copyout(&ts, uap->tp, sizeof(ts)); 383 384 return (error); 385 } 386 387 static int 388 kern_getcpuclockid(pid_t pid, lwpid_t lwp_id, clockid_t *clock_id) 389 { 390 struct proc *p; 391 int error = 0; 392 393 if (pid == 0) { 394 p = curproc; 395 pid = p->p_pid; 396 PHOLD(p); 397 } else { 398 p = pfind(pid); 399 if (p == NULL) 400 return (ESRCH); 401 } 402 /* lwp_id can be 0 when called by clock_getcpuclockid() */ 403 if (lwp_id < 0) { 404 error = EINVAL; 405 goto out; 406 } 407 lwkt_gettoken(&p->p_token); 408 if (lwp_id > 0 && 409 lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, lwp_id) == NULL) { 410 lwkt_reltoken(&p->p_token); 411 error = ESRCH; 412 goto out; 413 } 414 *clock_id = MAKE_CPUCLOCK(pid, lwp_id); 415 lwkt_reltoken(&p->p_token); 416 out: 417 PRELE(p); 418 return (error); 419 } 420 421 int 422 sys_getcpuclockid(struct getcpuclockid_args *uap) 423 { 424 clockid_t clk_id; 425 int error; 426 427 error = kern_getcpuclockid(uap->pid, uap->lwp_id, &clk_id); 428 if (error == 0) 429 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 430 431 return (error); 432 } 433 434 /* 435 * nanosleep1() 436 * 437 * This is a general helper function for nanosleep() (aka sleep() aka 438 * usleep()). 439 * 440 * If there is less then one tick's worth of time left and 441 * we haven't done a yield, or the remaining microseconds is 442 * ridiculously low, do a yield. This avoids having 443 * to deal with systimer overheads when the system is under 444 * heavy loads. If we have done a yield already then use 445 * a systimer and an uninterruptable thread wait. 446 * 447 * If there is more then a tick's worth of time left, 448 * calculate the baseline ticks and use an interruptable 449 * tsleep, then handle the fine-grained delay on the next 450 * loop. This usually results in two sleeps occuring, a long one 451 * and a short one. 452 * 453 * MPSAFE 454 */ 455 static void 456 ns1_systimer(systimer_t info, int in_ipi __unused, 457 struct intrframe *frame __unused) 458 { 459 lwkt_schedule(info->data); 460 } 461 462 int 463 nanosleep1(struct timespec *rqt, struct timespec *rmt) 464 { 465 static int nanowait; 466 struct timespec ts, ts2, ts3; 467 struct timeval tv; 468 int error; 469 470 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 471 return (EINVAL); 472 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 473 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 474 return (0); 475 nanouptime(&ts); 476 timespecadd(&ts, rqt, &ts); /* ts = target timestamp compare */ 477 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 478 479 for (;;) { 480 int ticks; 481 struct systimer info; 482 483 ticks = tv.tv_usec / ustick; /* approximate */ 484 485 if (tv.tv_sec == 0 && ticks == 0) { 486 thread_t td = curthread; 487 if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us) 488 tv.tv_usec = nanosleep_min_us; 489 if (tv.tv_usec < nanosleep_hard_us) { 490 lwkt_user_yield(); 491 cpu_pause(); 492 } else { 493 crit_enter_quick(td); 494 systimer_init_oneshot(&info, ns1_systimer, 495 td, tv.tv_usec); 496 lwkt_deschedule_self(td); 497 crit_exit_quick(td); 498 lwkt_switch(); 499 systimer_del(&info); /* make sure it's gone */ 500 } 501 error = iscaught(td->td_lwp); 502 } else if (tv.tv_sec == 0) { 503 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 504 } else { 505 ticks = tvtohz_low(&tv); /* also handles overflow */ 506 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 507 } 508 nanouptime(&ts2); 509 if (error && error != EWOULDBLOCK) { 510 if (error == ERESTART) 511 error = EINTR; 512 if (rmt != NULL) { 513 timespecsub(&ts, &ts2, &ts); 514 if (ts.tv_sec < 0) 515 timespecclear(&ts); 516 *rmt = ts; 517 } 518 return (error); 519 } 520 if (timespeccmp(&ts2, &ts, >=)) 521 return (0); 522 timespecsub(&ts, &ts2, &ts3); 523 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 524 } 525 } 526 527 /* 528 * MPSAFE 529 */ 530 int 531 sys_nanosleep(struct nanosleep_args *uap) 532 { 533 int error; 534 struct timespec rqt; 535 struct timespec rmt; 536 537 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 538 if (error) 539 return (error); 540 541 error = nanosleep1(&rqt, &rmt); 542 543 /* 544 * copyout the residual if nanosleep was interrupted. 545 */ 546 if (error && uap->rmtp) { 547 int error2; 548 549 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 550 if (error2) 551 error = error2; 552 } 553 return (error); 554 } 555 556 /* 557 * The gettimeofday() system call is supposed to return a fine-grained 558 * realtime stamp. However, acquiring a fine-grained stamp can create a 559 * bottleneck when multiple cpu cores are trying to accessing e.g. the 560 * HPET hardware timer all at the same time, so we have a sysctl that 561 * allows its behavior to be changed to a more coarse-grained timestamp 562 * which does not have to access a hardware timer. 563 */ 564 int 565 sys_gettimeofday(struct gettimeofday_args *uap) 566 { 567 struct timeval atv; 568 int error = 0; 569 570 if (uap->tp) { 571 if (gettimeofday_quick) 572 getmicrotime(&atv); 573 else 574 microtime(&atv); 575 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 576 sizeof (atv)))) 577 return (error); 578 } 579 if (uap->tzp) 580 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 581 sizeof (tz)); 582 return (error); 583 } 584 585 /* 586 * MPALMOSTSAFE 587 */ 588 int 589 sys_settimeofday(struct settimeofday_args *uap) 590 { 591 struct thread *td = curthread; 592 struct timeval atv; 593 struct timezone atz; 594 int error; 595 596 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 597 return (error); 598 /* 599 * Verify all parameters before changing time. 600 * 601 * XXX: We do not allow the time to be set to 0.0, which also by 602 * happy coincidence works around a pkgsrc bulk build bug. 603 */ 604 if (uap->tv) { 605 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 606 sizeof(atv)))) 607 return (error); 608 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 609 return (EINVAL); 610 if (atv.tv_sec == 0 && atv.tv_usec == 0) 611 return (EINVAL); 612 } 613 if (uap->tzp && 614 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 615 return (error); 616 617 lockmgr(&masterclock_lock, LK_EXCLUSIVE); 618 if (uap->tv && (error = settime(&atv))) { 619 lockmgr(&masterclock_lock, LK_RELEASE); 620 return (error); 621 } 622 lockmgr(&masterclock_lock, LK_RELEASE); 623 624 if (uap->tzp) 625 tz = atz; 626 return (0); 627 } 628 629 /* 630 * WARNING! Run with ntp_spin held 631 */ 632 static void 633 kern_adjtime_common(void) 634 { 635 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 636 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 637 ntp_tick_delta = ntp_delta; 638 else if (ntp_delta > ntp_big_delta) 639 ntp_tick_delta = 10 * ntp_default_tick_delta; 640 else if (ntp_delta < -ntp_big_delta) 641 ntp_tick_delta = -10 * ntp_default_tick_delta; 642 else if (ntp_delta > 0) 643 ntp_tick_delta = ntp_default_tick_delta; 644 else 645 ntp_tick_delta = -ntp_default_tick_delta; 646 } 647 648 void 649 kern_adjtime(int64_t delta, int64_t *odelta) 650 { 651 spin_lock(&ntp_spin); 652 *odelta = ntp_delta; 653 ntp_delta = delta; 654 kern_adjtime_common(); 655 spin_unlock(&ntp_spin); 656 } 657 658 static void 659 kern_get_ntp_delta(int64_t *delta) 660 { 661 *delta = ntp_delta; 662 } 663 664 void 665 kern_reladjtime(int64_t delta) 666 { 667 spin_lock(&ntp_spin); 668 ntp_delta += delta; 669 kern_adjtime_common(); 670 spin_unlock(&ntp_spin); 671 } 672 673 static void 674 kern_adjfreq(int64_t rate) 675 { 676 spin_lock(&ntp_spin); 677 ntp_tick_permanent = rate; 678 spin_unlock(&ntp_spin); 679 } 680 681 /* 682 * MPALMOSTSAFE 683 */ 684 int 685 sys_adjtime(struct adjtime_args *uap) 686 { 687 struct thread *td = curthread; 688 struct timeval atv; 689 int64_t ndelta, odelta; 690 int error; 691 692 if ((error = priv_check(td, PRIV_ADJTIME))) 693 return (error); 694 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 695 if (error) 696 return (error); 697 698 /* 699 * Compute the total correction and the rate at which to apply it. 700 * Round the adjustment down to a whole multiple of the per-tick 701 * delta, so that after some number of incremental changes in 702 * hardclock(), tickdelta will become zero, lest the correction 703 * overshoot and start taking us away from the desired final time. 704 */ 705 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 706 kern_adjtime(ndelta, &odelta); 707 708 if (uap->olddelta) { 709 atv.tv_sec = odelta / 1000000000; 710 atv.tv_usec = odelta % 1000000000 / 1000; 711 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 712 } 713 return (0); 714 } 715 716 static int 717 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 718 { 719 int64_t delta; 720 int error; 721 722 if (req->newptr != NULL) { 723 if (priv_check(curthread, PRIV_ROOT)) 724 return (EPERM); 725 error = SYSCTL_IN(req, &delta, sizeof(delta)); 726 if (error) 727 return (error); 728 kern_reladjtime(delta); 729 } 730 731 if (req->oldptr) 732 kern_get_ntp_delta(&delta); 733 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 734 return (error); 735 } 736 737 /* 738 * delta is in nanoseconds. 739 */ 740 static int 741 sysctl_delta(SYSCTL_HANDLER_ARGS) 742 { 743 int64_t delta, old_delta; 744 int error; 745 746 if (req->newptr != NULL) { 747 if (priv_check(curthread, PRIV_ROOT)) 748 return (EPERM); 749 error = SYSCTL_IN(req, &delta, sizeof(delta)); 750 if (error) 751 return (error); 752 kern_adjtime(delta, &old_delta); 753 } 754 755 if (req->oldptr != NULL) 756 kern_get_ntp_delta(&old_delta); 757 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 758 return (error); 759 } 760 761 /* 762 * frequency is in nanoseconds per second shifted left 32. 763 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 764 */ 765 static int 766 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 767 { 768 int64_t freqdelta; 769 int error; 770 771 if (req->newptr != NULL) { 772 if (priv_check(curthread, PRIV_ROOT)) 773 return (EPERM); 774 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 775 if (error) 776 return (error); 777 778 freqdelta /= hz; 779 kern_adjfreq(freqdelta); 780 } 781 782 if (req->oldptr != NULL) 783 freqdelta = ntp_tick_permanent * hz; 784 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 785 if (error) 786 return (error); 787 788 return (0); 789 } 790 791 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 792 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 793 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 794 sysctl_adjfreq, "Q", "permanent correction per second"); 795 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 796 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 797 sysctl_delta, "Q", "one-time delta"); 798 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 799 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 800 "threshold for fast adjustment"); 801 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 802 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 803 "per-tick adjustment"); 804 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 805 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 806 "default per-tick adjustment"); 807 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 808 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 809 "next leap second"); 810 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 811 &ntp_leap_insert, 0, "insert or remove leap second"); 812 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 813 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 814 sysctl_adjtime, "Q", "relative adjust for delta"); 815 816 /* 817 * Get value of an interval timer. The process virtual and 818 * profiling virtual time timers are kept in the p_stats area, since 819 * they can be swapped out. These are kept internally in the 820 * way they are specified externally: in time until they expire. 821 * 822 * The real time interval timer is kept in the process table slot 823 * for the process, and its value (it_value) is kept as an 824 * absolute time rather than as a delta, so that it is easy to keep 825 * periodic real-time signals from drifting. 826 * 827 * Virtual time timers are processed in the hardclock() routine of 828 * kern_clock.c. The real time timer is processed by a timeout 829 * routine, called from the softclock() routine. Since a callout 830 * may be delayed in real time due to interrupt processing in the system, 831 * it is possible for the real time timeout routine (realitexpire, given below), 832 * to be delayed in real time past when it is supposed to occur. It 833 * does not suffice, therefore, to reload the real timer .it_value from the 834 * real time timers .it_interval. Rather, we compute the next time in 835 * absolute time the timer should go off. 836 * 837 * MPALMOSTSAFE 838 */ 839 int 840 sys_getitimer(struct getitimer_args *uap) 841 { 842 struct proc *p = curproc; 843 struct timeval ctv; 844 struct itimerval aitv; 845 846 if (uap->which > ITIMER_PROF) 847 return (EINVAL); 848 lwkt_gettoken(&p->p_token); 849 if (uap->which == ITIMER_REAL) { 850 /* 851 * Convert from absolute to relative time in .it_value 852 * part of real time timer. If time for real time timer 853 * has passed return 0, else return difference between 854 * current time and time for the timer to go off. 855 */ 856 aitv = p->p_realtimer; 857 if (timevalisset(&aitv.it_value)) { 858 getmicrouptime(&ctv); 859 if (timevalcmp(&aitv.it_value, &ctv, <)) 860 timevalclear(&aitv.it_value); 861 else 862 timevalsub(&aitv.it_value, &ctv); 863 } 864 } else { 865 aitv = p->p_timer[uap->which]; 866 } 867 lwkt_reltoken(&p->p_token); 868 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 869 } 870 871 /* 872 * MPALMOSTSAFE 873 */ 874 int 875 sys_setitimer(struct setitimer_args *uap) 876 { 877 struct itimerval aitv; 878 struct timeval ctv; 879 struct itimerval *itvp; 880 struct proc *p = curproc; 881 int error; 882 883 if (uap->which > ITIMER_PROF) 884 return (EINVAL); 885 itvp = uap->itv; 886 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 887 sizeof(struct itimerval)))) 888 return (error); 889 if ((uap->itv = uap->oitv) && 890 (error = sys_getitimer((struct getitimer_args *)uap))) 891 return (error); 892 if (itvp == NULL) 893 return (0); 894 if (itimerfix(&aitv.it_value)) 895 return (EINVAL); 896 if (!timevalisset(&aitv.it_value)) 897 timevalclear(&aitv.it_interval); 898 else if (itimerfix(&aitv.it_interval)) 899 return (EINVAL); 900 lwkt_gettoken(&p->p_token); 901 if (uap->which == ITIMER_REAL) { 902 if (timevalisset(&p->p_realtimer.it_value)) 903 callout_cancel(&p->p_ithandle); 904 if (timevalisset(&aitv.it_value)) 905 callout_reset(&p->p_ithandle, 906 tvtohz_high(&aitv.it_value), realitexpire, p); 907 getmicrouptime(&ctv); 908 timevaladd(&aitv.it_value, &ctv); 909 p->p_realtimer = aitv; 910 } else { 911 p->p_timer[uap->which] = aitv; 912 switch(uap->which) { 913 case ITIMER_VIRTUAL: 914 p->p_flags &= ~P_SIGVTALRM; 915 break; 916 case ITIMER_PROF: 917 p->p_flags &= ~P_SIGPROF; 918 break; 919 } 920 } 921 lwkt_reltoken(&p->p_token); 922 return (0); 923 } 924 925 /* 926 * Real interval timer expired: 927 * send process whose timer expired an alarm signal. 928 * If time is not set up to reload, then just return. 929 * Else compute next time timer should go off which is > current time. 930 * This is where delay in processing this timeout causes multiple 931 * SIGALRM calls to be compressed into one. 932 * tvtohz_high() always adds 1 to allow for the time until the next clock 933 * interrupt being strictly less than 1 clock tick, but we don't want 934 * that here since we want to appear to be in sync with the clock 935 * interrupt even when we're delayed. 936 */ 937 static 938 void 939 realitexpire(void *arg) 940 { 941 struct proc *p; 942 struct timeval ctv, ntv; 943 944 p = (struct proc *)arg; 945 PHOLD(p); 946 lwkt_gettoken(&p->p_token); 947 ksignal(p, SIGALRM); 948 if (!timevalisset(&p->p_realtimer.it_interval)) { 949 timevalclear(&p->p_realtimer.it_value); 950 goto done; 951 } 952 for (;;) { 953 timevaladd(&p->p_realtimer.it_value, 954 &p->p_realtimer.it_interval); 955 getmicrouptime(&ctv); 956 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 957 ntv = p->p_realtimer.it_value; 958 timevalsub(&ntv, &ctv); 959 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 960 realitexpire, p); 961 goto done; 962 } 963 } 964 done: 965 lwkt_reltoken(&p->p_token); 966 PRELE(p); 967 } 968 969 /* 970 * Used to validate itimer timeouts and utimes*() timespecs. 971 */ 972 int 973 itimerfix(struct timeval *tv) 974 { 975 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 976 return (EINVAL); 977 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 978 tv->tv_usec = ustick; 979 return (0); 980 } 981 982 /* 983 * Used to validate timeouts and utimes*() timespecs. 984 */ 985 int 986 itimespecfix(struct timespec *ts) 987 { 988 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000ULL) 989 return (EINVAL); 990 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < nstick) 991 ts->tv_nsec = nstick; 992 return (0); 993 } 994 995 /* 996 * Decrement an interval timer by a specified number 997 * of microseconds, which must be less than a second, 998 * i.e. < 1000000. If the timer expires, then reload 999 * it. In this case, carry over (usec - old value) to 1000 * reduce the value reloaded into the timer so that 1001 * the timer does not drift. This routine assumes 1002 * that it is called in a context where the timers 1003 * on which it is operating cannot change in value. 1004 */ 1005 int 1006 itimerdecr(struct itimerval *itp, int usec) 1007 { 1008 1009 if (itp->it_value.tv_usec < usec) { 1010 if (itp->it_value.tv_sec == 0) { 1011 /* expired, and already in next interval */ 1012 usec -= itp->it_value.tv_usec; 1013 goto expire; 1014 } 1015 itp->it_value.tv_usec += 1000000; 1016 itp->it_value.tv_sec--; 1017 } 1018 itp->it_value.tv_usec -= usec; 1019 usec = 0; 1020 if (timevalisset(&itp->it_value)) 1021 return (1); 1022 /* expired, exactly at end of interval */ 1023 expire: 1024 if (timevalisset(&itp->it_interval)) { 1025 itp->it_value = itp->it_interval; 1026 itp->it_value.tv_usec -= usec; 1027 if (itp->it_value.tv_usec < 0) { 1028 itp->it_value.tv_usec += 1000000; 1029 itp->it_value.tv_sec--; 1030 } 1031 } else 1032 itp->it_value.tv_usec = 0; /* sec is already 0 */ 1033 return (0); 1034 } 1035 1036 /* 1037 * Add and subtract routines for timevals. 1038 * N.B.: subtract routine doesn't deal with 1039 * results which are before the beginning, 1040 * it just gets very confused in this case. 1041 * Caveat emptor. 1042 */ 1043 void 1044 timevaladd(struct timeval *t1, const struct timeval *t2) 1045 { 1046 1047 t1->tv_sec += t2->tv_sec; 1048 t1->tv_usec += t2->tv_usec; 1049 timevalfix(t1); 1050 } 1051 1052 void 1053 timevalsub(struct timeval *t1, const struct timeval *t2) 1054 { 1055 1056 t1->tv_sec -= t2->tv_sec; 1057 t1->tv_usec -= t2->tv_usec; 1058 timevalfix(t1); 1059 } 1060 1061 static void 1062 timevalfix(struct timeval *t1) 1063 { 1064 1065 if (t1->tv_usec < 0) { 1066 t1->tv_sec--; 1067 t1->tv_usec += 1000000; 1068 } 1069 if (t1->tv_usec >= 1000000) { 1070 t1->tv_sec++; 1071 t1->tv_usec -= 1000000; 1072 } 1073 } 1074 1075 /* 1076 * ratecheck(): simple time-based rate-limit checking. 1077 */ 1078 int 1079 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1080 { 1081 struct timeval tv, delta; 1082 int rv = 0; 1083 1084 getmicrouptime(&tv); /* NB: 10ms precision */ 1085 delta = tv; 1086 timevalsub(&delta, lasttime); 1087 1088 /* 1089 * check for 0,0 is so that the message will be seen at least once, 1090 * even if interval is huge. 1091 */ 1092 if (timevalcmp(&delta, mininterval, >=) || 1093 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1094 *lasttime = tv; 1095 rv = 1; 1096 } 1097 1098 return (rv); 1099 } 1100 1101 /* 1102 * ppsratecheck(): packets (or events) per second limitation. 1103 * 1104 * Return 0 if the limit is to be enforced (e.g. the caller 1105 * should drop a packet because of the rate limitation). 1106 * 1107 * maxpps of 0 always causes zero to be returned. maxpps of -1 1108 * always causes 1 to be returned; this effectively defeats rate 1109 * limiting. 1110 * 1111 * Note that we maintain the struct timeval for compatibility 1112 * with other bsd systems. We reuse the storage and just monitor 1113 * clock ticks for minimal overhead. 1114 */ 1115 int 1116 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1117 { 1118 int now; 1119 1120 /* 1121 * Reset the last time and counter if this is the first call 1122 * or more than a second has passed since the last update of 1123 * lasttime. 1124 */ 1125 now = ticks; 1126 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1127 lasttime->tv_sec = now; 1128 *curpps = 1; 1129 return (maxpps != 0); 1130 } else { 1131 (*curpps)++; /* NB: ignore potential overflow */ 1132 return (maxpps < 0 || *curpps < maxpps); 1133 } 1134 } 1135 1136 static int 1137 sysctl_gettimeofday_quick(SYSCTL_HANDLER_ARGS) 1138 { 1139 int error; 1140 int gtod; 1141 1142 gtod = gettimeofday_quick; 1143 error = sysctl_handle_int(oidp, >od, 0, req); 1144 if (error || req->newptr == NULL) 1145 return error; 1146 gettimeofday_quick = gtod; 1147 if (kpmap) 1148 kpmap->fast_gtod = gtod; 1149 return 0; 1150 } 1151