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/sysmsg.h> 37 #include <sys/resourcevar.h> 38 #include <sys/signalvar.h> 39 #include <sys/kernel.h> 40 #include <sys/sysent.h> 41 #include <sys/proc.h> 42 #include <sys/priv.h> 43 #include <sys/time.h> 44 #include <sys/vnode.h> 45 #include <sys/sysctl.h> 46 #include <sys/kern_syscall.h> 47 #include <sys/upmap.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 static int sysctl_gettimeofday_quick(SYSCTL_HANDLER_ARGS); 80 81 82 /* 83 * Nanosleep tries very hard to sleep for a precisely requested time 84 * interval, down to 1uS. The administrator can impose a minimum delay 85 * and a delay below which we hard-loop instead of initiate a timer 86 * interrupt and sleep. 87 * 88 * For machines under high loads it might be beneficial to increase min_us 89 * to e.g. 1000uS (1ms) so spining processes sleep meaningfully. 90 */ 91 static int nanosleep_min_us = 10; 92 static int nanosleep_hard_us = 100; 93 static int gettimeofday_quick = 0; 94 SYSCTL_INT(_kern, OID_AUTO, nanosleep_min_us, CTLFLAG_RW, 95 &nanosleep_min_us, 0, ""); 96 SYSCTL_INT(_kern, OID_AUTO, nanosleep_hard_us, CTLFLAG_RW, 97 &nanosleep_hard_us, 0, ""); 98 SYSCTL_PROC(_kern, OID_AUTO, gettimeofday_quick, CTLTYPE_INT | CTLFLAG_RW, 99 0, 0, sysctl_gettimeofday_quick, "I", "Quick mode gettimeofday"); 100 101 static struct lock masterclock_lock = LOCK_INITIALIZER("mstrclk", 0, 0); 102 103 static int 104 settime(struct timeval *tv) 105 { 106 struct timeval delta, tv1, tv2; 107 static struct timeval maxtime, laststep; 108 struct timespec ts; 109 int origcpu; 110 111 if ((origcpu = mycpu->gd_cpuid) != 0) 112 lwkt_setcpu_self(globaldata_find(0)); 113 114 crit_enter(); 115 microtime(&tv1); 116 delta = *tv; 117 timevalsub(&delta, &tv1); 118 119 /* 120 * If the system is secure, we do not allow the time to be 121 * set to a value earlier than 1 second less than the highest 122 * time we have yet seen. The worst a miscreant can do in 123 * this circumstance is "freeze" time. He couldn't go 124 * back to the past. 125 * 126 * We similarly do not allow the clock to be stepped more 127 * than one second, nor more than once per second. This allows 128 * a miscreant to make the clock march double-time, but no worse. 129 */ 130 if (securelevel > 1) { 131 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 132 /* 133 * Update maxtime to latest time we've seen. 134 */ 135 if (tv1.tv_sec > maxtime.tv_sec) 136 maxtime = tv1; 137 tv2 = *tv; 138 timevalsub(&tv2, &maxtime); 139 if (tv2.tv_sec < -1) { 140 tv->tv_sec = maxtime.tv_sec - 1; 141 kprintf("Time adjustment clamped to -1 second\n"); 142 } 143 } else { 144 if (tv1.tv_sec == laststep.tv_sec) { 145 crit_exit(); 146 return (EPERM); 147 } 148 if (delta.tv_sec > 1) { 149 tv->tv_sec = tv1.tv_sec + 1; 150 kprintf("Time adjustment clamped to +1 second\n"); 151 } 152 laststep = *tv; 153 } 154 } 155 156 ts.tv_sec = tv->tv_sec; 157 ts.tv_nsec = tv->tv_usec * 1000; 158 set_timeofday(&ts); 159 crit_exit(); 160 161 if (origcpu != 0) 162 lwkt_setcpu_self(globaldata_find(origcpu)); 163 164 resettodr(); 165 return (0); 166 } 167 168 static void 169 get_process_cputime(struct proc *p, struct timespec *ats) 170 { 171 struct rusage ru; 172 173 lwkt_gettoken(&p->p_token); 174 calcru_proc(p, &ru); 175 lwkt_reltoken(&p->p_token); 176 timevaladd(&ru.ru_utime, &ru.ru_stime); 177 TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats); 178 } 179 180 static void 181 get_process_usertime(struct proc *p, struct timespec *ats) 182 { 183 struct rusage ru; 184 185 lwkt_gettoken(&p->p_token); 186 calcru_proc(p, &ru); 187 lwkt_reltoken(&p->p_token); 188 TIMEVAL_TO_TIMESPEC(&ru.ru_utime, ats); 189 } 190 191 static void 192 get_thread_cputime(struct thread *td, struct timespec *ats) 193 { 194 struct timeval sys, user; 195 196 calcru(td->td_lwp, &user, &sys); 197 timevaladd(&user, &sys); 198 TIMEVAL_TO_TIMESPEC(&user, ats); 199 } 200 201 /* 202 * MPSAFE 203 */ 204 int 205 kern_clock_gettime(clockid_t clock_id, struct timespec *ats) 206 { 207 struct proc *p; 208 struct lwp *lp; 209 lwpid_t lwp_id; 210 211 p = curproc; 212 switch(clock_id) { 213 case CLOCK_REALTIME: 214 case CLOCK_REALTIME_PRECISE: 215 nanotime(ats); 216 break; 217 case CLOCK_REALTIME_FAST: 218 getnanotime(ats); 219 break; 220 case CLOCK_MONOTONIC: 221 case CLOCK_MONOTONIC_PRECISE: 222 case CLOCK_UPTIME: 223 case CLOCK_UPTIME_PRECISE: 224 nanouptime(ats); 225 break; 226 case CLOCK_MONOTONIC_FAST: 227 case CLOCK_UPTIME_FAST: 228 getnanouptime(ats); 229 break; 230 case CLOCK_VIRTUAL: 231 get_process_usertime(p, ats); 232 break; 233 case CLOCK_PROF: 234 case CLOCK_PROCESS_CPUTIME_ID: 235 get_process_cputime(p, ats); 236 break; 237 case CLOCK_SECOND: 238 ats->tv_sec = time_second; 239 ats->tv_nsec = 0; 240 break; 241 case CLOCK_THREAD_CPUTIME_ID: 242 get_thread_cputime(curthread, ats); 243 break; 244 default: 245 if ((clock_id & CPUCLOCK_BIT) == 0) 246 return (EINVAL); 247 if ((p = pfind(CPUCLOCK2PID(clock_id))) == NULL) 248 return (EINVAL); 249 lwp_id = CPUCLOCK2LWPID(clock_id); 250 if (lwp_id == 0) { 251 get_process_cputime(p, ats); 252 } else { 253 lwkt_gettoken(&p->p_token); 254 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, lwp_id); 255 if (lp == NULL) { 256 lwkt_reltoken(&p->p_token); 257 PRELE(p); 258 return (EINVAL); 259 } 260 get_thread_cputime(lp->lwp_thread, ats); 261 lwkt_reltoken(&p->p_token); 262 } 263 PRELE(p); 264 } 265 return (0); 266 } 267 268 /* 269 * MPSAFE 270 */ 271 int 272 sys_clock_gettime(struct sysmsg *sysmsg, const struct clock_gettime_args *uap) 273 { 274 struct timespec ats; 275 int error; 276 277 error = kern_clock_gettime(uap->clock_id, &ats); 278 if (error == 0) 279 error = copyout(&ats, uap->tp, sizeof(ats)); 280 281 return (error); 282 } 283 284 int 285 kern_clock_settime(clockid_t clock_id, struct timespec *ats) 286 { 287 struct thread *td = curthread; 288 struct timeval atv; 289 int error; 290 291 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 292 return (error); 293 if (clock_id != CLOCK_REALTIME) 294 return (EINVAL); 295 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000) 296 return (EINVAL); 297 298 lockmgr(&masterclock_lock, LK_EXCLUSIVE); 299 TIMESPEC_TO_TIMEVAL(&atv, ats); 300 error = settime(&atv); 301 lockmgr(&masterclock_lock, LK_RELEASE); 302 303 return (error); 304 } 305 306 /* 307 * MPALMOSTSAFE 308 */ 309 int 310 sys_clock_settime(struct sysmsg *sysmsg, const struct clock_settime_args *uap) 311 { 312 struct timespec ats; 313 int error; 314 315 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 316 return (error); 317 318 error = kern_clock_settime(uap->clock_id, &ats); 319 320 return (error); 321 } 322 323 /* 324 * MPSAFE 325 */ 326 int 327 kern_clock_getres(clockid_t clock_id, struct timespec *ts) 328 { 329 ts->tv_sec = 0; 330 331 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 * Minimum reportable resolution is 1ns. Rounding is 343 * otherwise unimportant. 344 */ 345 ts->tv_nsec = 999999999 / sys_cputimer->freq + 1; 346 break; 347 case CLOCK_VIRTUAL: 348 case CLOCK_PROF: 349 /* Accurately round up here because we can do so cheaply. */ 350 ts->tv_nsec = howmany(1000000000, hz); 351 break; 352 case CLOCK_SECOND: 353 ts->tv_sec = 1; 354 ts->tv_nsec = 0; 355 break; 356 case CLOCK_THREAD_CPUTIME_ID: 357 case CLOCK_PROCESS_CPUTIME_ID: 358 ts->tv_nsec = 1000; 359 break; 360 default: 361 if ((clock_id & CPUCLOCK_BIT) != 0) 362 ts->tv_nsec = 1000; 363 else 364 return (EINVAL); 365 } 366 367 return (0); 368 } 369 370 /* 371 * MPSAFE 372 */ 373 int 374 sys_clock_getres(struct sysmsg *sysmsg, const struct clock_getres_args *uap) 375 { 376 int error; 377 struct timespec ts; 378 379 error = kern_clock_getres(uap->clock_id, &ts); 380 if (error == 0) 381 error = copyout(&ts, uap->tp, sizeof(ts)); 382 383 return (error); 384 } 385 386 static int 387 kern_getcpuclockid(pid_t pid, lwpid_t lwp_id, clockid_t *clock_id) 388 { 389 struct proc *p; 390 int error = 0; 391 392 if (pid == 0) { 393 p = curproc; 394 pid = p->p_pid; 395 PHOLD(p); 396 } else { 397 p = pfind(pid); 398 if (p == NULL) 399 return (ESRCH); 400 } 401 /* lwp_id can be 0 when called by clock_getcpuclockid() */ 402 if (lwp_id < 0) { 403 error = EINVAL; 404 goto out; 405 } 406 lwkt_gettoken(&p->p_token); 407 if (lwp_id > 0 && 408 lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, lwp_id) == NULL) { 409 lwkt_reltoken(&p->p_token); 410 error = ESRCH; 411 goto out; 412 } 413 *clock_id = MAKE_CPUCLOCK(pid, lwp_id); 414 lwkt_reltoken(&p->p_token); 415 out: 416 PRELE(p); 417 return (error); 418 } 419 420 int 421 sys_getcpuclockid(struct sysmsg *sysmsg, const struct getcpuclockid_args *uap) 422 { 423 clockid_t clk_id; 424 int error; 425 426 error = kern_getcpuclockid(uap->pid, uap->lwp_id, &clk_id); 427 if (error == 0) 428 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 429 430 return (error); 431 } 432 433 /* 434 * nanosleep1() 435 * 436 * This is a general helper function for nanosleep() (aka sleep() aka 437 * usleep()). 438 * 439 * If there is less then one tick's worth of time left and 440 * we haven't done a yield, or the remaining microseconds is 441 * ridiculously low, do a yield. This avoids having 442 * to deal with systimer overheads when the system is under 443 * heavy loads. If we have done a yield already then use 444 * a systimer and an uninterruptable thread wait. 445 * 446 * If there is more then a tick's worth of time left, 447 * calculate the baseline ticks and use an interruptable 448 * tsleep, then handle the fine-grained delay on the next 449 * loop. This usually results in two sleeps occuring, a long one 450 * and a short one. 451 * 452 * MPSAFE 453 */ 454 static void 455 ns1_systimer(systimer_t info, int in_ipi __unused, 456 struct intrframe *frame __unused) 457 { 458 lwkt_schedule(info->data); 459 } 460 461 int 462 nanosleep1(struct timespec *rqt, struct timespec *rmt) 463 { 464 static int nanowait; 465 struct timespec ts, ts2, ts3; 466 struct timeval tv; 467 int error; 468 469 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 470 return (EINVAL); 471 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 472 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 473 return (0); 474 nanouptime(&ts); 475 timespecadd(&ts, rqt, &ts); /* ts = target timestamp compare */ 476 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 477 478 for (;;) { 479 int ticks; 480 struct systimer info; 481 482 ticks = tv.tv_usec / ustick; /* approximate */ 483 484 if (tv.tv_sec == 0 && ticks == 0) { 485 thread_t td = curthread; 486 if (tv.tv_usec > 0 && tv.tv_usec < nanosleep_min_us) 487 tv.tv_usec = nanosleep_min_us; 488 if (tv.tv_usec < nanosleep_hard_us) { 489 lwkt_user_yield(); 490 cpu_pause(); 491 } else { 492 crit_enter_quick(td); 493 systimer_init_oneshot(&info, ns1_systimer, 494 td, tv.tv_usec); 495 lwkt_deschedule_self(td); 496 crit_exit_quick(td); 497 lwkt_switch(); 498 systimer_del(&info); /* make sure it's gone */ 499 } 500 error = iscaught(td->td_lwp); 501 } else if (tv.tv_sec == 0) { 502 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 503 } else { 504 ticks = tvtohz_low(&tv); /* also handles overflow */ 505 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 506 } 507 nanouptime(&ts2); 508 if (error && error != EWOULDBLOCK) { 509 if (error == ERESTART) 510 error = EINTR; 511 if (rmt != NULL) { 512 timespecsub(&ts, &ts2, &ts); 513 if (ts.tv_sec < 0) 514 timespecclear(&ts); 515 *rmt = ts; 516 } 517 return (error); 518 } 519 if (timespeccmp(&ts2, &ts, >=)) 520 return (0); 521 timespecsub(&ts, &ts2, &ts3); 522 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 523 } 524 } 525 526 /* 527 * MPSAFE 528 */ 529 int 530 sys_nanosleep(struct sysmsg *sysmsg, const struct nanosleep_args *uap) 531 { 532 int error; 533 struct timespec rqt; 534 struct timespec rmt; 535 536 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 537 if (error) 538 return (error); 539 540 error = nanosleep1(&rqt, &rmt); 541 542 /* 543 * copyout the residual if nanosleep was interrupted. 544 */ 545 if (error && uap->rmtp) { 546 int error2; 547 548 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 549 if (error2) 550 error = error2; 551 } 552 return (error); 553 } 554 555 /* 556 * The gettimeofday() system call is supposed to return a fine-grained 557 * realtime stamp. However, acquiring a fine-grained stamp can create a 558 * bottleneck when multiple cpu cores are trying to accessing e.g. the 559 * HPET hardware timer all at the same time, so we have a sysctl that 560 * allows its behavior to be changed to a more coarse-grained timestamp 561 * which does not have to access a hardware timer. 562 */ 563 int 564 sys_gettimeofday(struct sysmsg *sysmsg, const struct gettimeofday_args *uap) 565 { 566 struct timeval atv; 567 int error = 0; 568 569 if (uap->tp) { 570 if (gettimeofday_quick) 571 getmicrotime(&atv); 572 else 573 microtime(&atv); 574 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 575 sizeof (atv)))) 576 return (error); 577 } 578 if (uap->tzp) 579 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 580 sizeof (tz)); 581 return (error); 582 } 583 584 /* 585 * MPALMOSTSAFE 586 */ 587 int 588 sys_settimeofday(struct sysmsg *sysmsg, const struct settimeofday_args *uap) 589 { 590 struct thread *td = curthread; 591 struct timeval atv; 592 struct timezone atz; 593 int error; 594 595 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 596 return (error); 597 /* 598 * Verify all parameters before changing time. 599 * 600 * XXX: We do not allow the time to be set to 0.0, which also by 601 * happy coincidence works around a pkgsrc bulk build bug. 602 */ 603 if (uap->tv) { 604 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 605 sizeof(atv)))) 606 return (error); 607 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 608 return (EINVAL); 609 if (atv.tv_sec == 0 && atv.tv_usec == 0) 610 return (EINVAL); 611 } 612 if (uap->tzp && 613 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 614 return (error); 615 616 lockmgr(&masterclock_lock, LK_EXCLUSIVE); 617 if (uap->tv && (error = settime(&atv))) { 618 lockmgr(&masterclock_lock, LK_RELEASE); 619 return (error); 620 } 621 lockmgr(&masterclock_lock, LK_RELEASE); 622 623 if (uap->tzp) 624 tz = atz; 625 return (0); 626 } 627 628 /* 629 * WARNING! Run with ntp_spin held 630 */ 631 static void 632 kern_adjtime_common(void) 633 { 634 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 635 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 636 ntp_tick_delta = ntp_delta; 637 else if (ntp_delta > ntp_big_delta) 638 ntp_tick_delta = 10 * ntp_default_tick_delta; 639 else if (ntp_delta < -ntp_big_delta) 640 ntp_tick_delta = -10 * ntp_default_tick_delta; 641 else if (ntp_delta > 0) 642 ntp_tick_delta = ntp_default_tick_delta; 643 else 644 ntp_tick_delta = -ntp_default_tick_delta; 645 } 646 647 void 648 kern_adjtime(int64_t delta, int64_t *odelta) 649 { 650 spin_lock(&ntp_spin); 651 *odelta = ntp_delta; 652 ntp_delta = delta; 653 kern_adjtime_common(); 654 spin_unlock(&ntp_spin); 655 } 656 657 static void 658 kern_get_ntp_delta(int64_t *delta) 659 { 660 *delta = ntp_delta; 661 } 662 663 void 664 kern_reladjtime(int64_t delta) 665 { 666 spin_lock(&ntp_spin); 667 ntp_delta += delta; 668 kern_adjtime_common(); 669 spin_unlock(&ntp_spin); 670 } 671 672 static void 673 kern_adjfreq(int64_t rate) 674 { 675 spin_lock(&ntp_spin); 676 ntp_tick_permanent = rate; 677 spin_unlock(&ntp_spin); 678 } 679 680 /* 681 * MPALMOSTSAFE 682 */ 683 int 684 sys_adjtime(struct sysmsg *sysmsg, const struct adjtime_args *uap) 685 { 686 struct thread *td = curthread; 687 struct timeval atv; 688 int64_t ndelta, odelta; 689 int error; 690 691 if ((error = priv_check(td, PRIV_ADJTIME))) 692 return (error); 693 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 694 if (error) 695 return (error); 696 697 /* 698 * Compute the total correction and the rate at which to apply it. 699 * Round the adjustment down to a whole multiple of the per-tick 700 * delta, so that after some number of incremental changes in 701 * hardclock(), tickdelta will become zero, lest the correction 702 * overshoot and start taking us away from the desired final time. 703 */ 704 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 705 kern_adjtime(ndelta, &odelta); 706 707 if (uap->olddelta) { 708 atv.tv_sec = odelta / 1000000000; 709 atv.tv_usec = odelta % 1000000000 / 1000; 710 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 711 } 712 return (0); 713 } 714 715 static int 716 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 717 { 718 int64_t delta; 719 int error; 720 721 if (req->newptr != NULL) { 722 if (priv_check(curthread, PRIV_ROOT)) 723 return (EPERM); 724 error = SYSCTL_IN(req, &delta, sizeof(delta)); 725 if (error) 726 return (error); 727 kern_reladjtime(delta); 728 } 729 730 if (req->oldptr) 731 kern_get_ntp_delta(&delta); 732 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 733 return (error); 734 } 735 736 /* 737 * delta is in nanoseconds. 738 */ 739 static int 740 sysctl_delta(SYSCTL_HANDLER_ARGS) 741 { 742 int64_t delta, old_delta; 743 int error; 744 745 if (req->newptr != NULL) { 746 if (priv_check(curthread, PRIV_ROOT)) 747 return (EPERM); 748 error = SYSCTL_IN(req, &delta, sizeof(delta)); 749 if (error) 750 return (error); 751 kern_adjtime(delta, &old_delta); 752 } 753 754 if (req->oldptr != NULL) 755 kern_get_ntp_delta(&old_delta); 756 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 757 return (error); 758 } 759 760 /* 761 * frequency is in nanoseconds per second shifted left 32. 762 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 763 */ 764 static int 765 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 766 { 767 int64_t freqdelta; 768 int error; 769 770 if (req->newptr != NULL) { 771 if (priv_check(curthread, PRIV_ROOT)) 772 return (EPERM); 773 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 774 if (error) 775 return (error); 776 777 freqdelta /= hz; 778 kern_adjfreq(freqdelta); 779 } 780 781 if (req->oldptr != NULL) 782 freqdelta = ntp_tick_permanent * hz; 783 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 784 if (error) 785 return (error); 786 787 return (0); 788 } 789 790 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 791 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 792 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 793 sysctl_adjfreq, "Q", "permanent correction per second"); 794 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 795 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 796 sysctl_delta, "Q", "one-time delta"); 797 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 798 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 799 "threshold for fast adjustment"); 800 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 801 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 802 "per-tick adjustment"); 803 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 804 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 805 "default per-tick adjustment"); 806 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 807 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 808 "next leap second"); 809 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 810 &ntp_leap_insert, 0, "insert or remove leap second"); 811 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 812 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 813 sysctl_adjtime, "Q", "relative adjust for delta"); 814 815 /* 816 * Get value of an interval timer. The process virtual and 817 * profiling virtual time timers are kept in the p_stats area, since 818 * they can be swapped out. These are kept internally in the 819 * way they are specified externally: in time until they expire. 820 * 821 * The real time interval timer is kept in the process table slot 822 * for the process, and its value (it_value) is kept as an 823 * absolute time rather than as a delta, so that it is easy to keep 824 * periodic real-time signals from drifting. 825 * 826 * Virtual time timers are processed in the hardclock() routine of 827 * kern_clock.c. The real time timer is processed by a timeout 828 * routine, called from the softclock() routine. Since a callout 829 * may be delayed in real time due to interrupt processing in the system, 830 * it is possible for the real time timeout routine (realitexpire, given below), 831 * to be delayed in real time past when it is supposed to occur. It 832 * does not suffice, therefore, to reload the real timer .it_value from the 833 * real time timers .it_interval. Rather, we compute the next time in 834 * absolute time the timer should go off. 835 * 836 * MPALMOSTSAFE 837 */ 838 int 839 sys_getitimer(struct sysmsg *sysmsg, const struct getitimer_args *uap) 840 { 841 struct proc *p = curproc; 842 struct timeval ctv; 843 struct itimerval aitv; 844 845 if (uap->which > ITIMER_PROF) 846 return (EINVAL); 847 lwkt_gettoken(&p->p_token); 848 if (uap->which == ITIMER_REAL) { 849 /* 850 * Convert from absolute to relative time in .it_value 851 * part of real time timer. If time for real time timer 852 * has passed return 0, else return difference between 853 * current time and time for the timer to go off. 854 */ 855 aitv = p->p_realtimer; 856 if (timevalisset(&aitv.it_value)) { 857 getmicrouptime(&ctv); 858 if (timevalcmp(&aitv.it_value, &ctv, <)) 859 timevalclear(&aitv.it_value); 860 else 861 timevalsub(&aitv.it_value, &ctv); 862 } 863 } else { 864 aitv = p->p_timer[uap->which]; 865 } 866 lwkt_reltoken(&p->p_token); 867 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 868 } 869 870 /* 871 * MPALMOSTSAFE 872 */ 873 int 874 sys_setitimer(struct sysmsg *sysmsg, const struct setitimer_args *uap) 875 { 876 struct itimerval aitv; 877 struct timeval ctv; 878 struct itimerval *itvp; 879 struct proc *p = curproc; 880 struct getitimer_args gitargs; 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); 889 890 if (uap->oitv) { 891 gitargs.which = uap->which; 892 gitargs.itv = uap->oitv; 893 error = sys_getitimer(sysmsg, &gitargs); 894 if (error) 895 return error; 896 } 897 if (itvp == NULL) 898 return (0); 899 if (itimerfix(&aitv.it_value)) 900 return (EINVAL); 901 if (!timevalisset(&aitv.it_value)) 902 timevalclear(&aitv.it_interval); 903 else if (itimerfix(&aitv.it_interval)) 904 return (EINVAL); 905 lwkt_gettoken(&p->p_token); 906 if (uap->which == ITIMER_REAL) { 907 if (timevalisset(&p->p_realtimer.it_value)) 908 callout_cancel(&p->p_ithandle); 909 if (timevalisset(&aitv.it_value)) 910 callout_reset(&p->p_ithandle, 911 tvtohz_high(&aitv.it_value), realitexpire, p); 912 getmicrouptime(&ctv); 913 timevaladd(&aitv.it_value, &ctv); 914 p->p_realtimer = aitv; 915 } else { 916 p->p_timer[uap->which] = aitv; 917 switch(uap->which) { 918 case ITIMER_VIRTUAL: 919 p->p_flags &= ~P_SIGVTALRM; 920 break; 921 case ITIMER_PROF: 922 p->p_flags &= ~P_SIGPROF; 923 break; 924 } 925 } 926 lwkt_reltoken(&p->p_token); 927 return (0); 928 } 929 930 /* 931 * Real interval timer expired: 932 * send process whose timer expired an alarm signal. 933 * If time is not set up to reload, then just return. 934 * Else compute next time timer should go off which is > current time. 935 * This is where delay in processing this timeout causes multiple 936 * SIGALRM calls to be compressed into one. 937 * tvtohz_high() always adds 1 to allow for the time until the next clock 938 * interrupt being strictly less than 1 clock tick, but we don't want 939 * that here since we want to appear to be in sync with the clock 940 * interrupt even when we're delayed. 941 */ 942 static 943 void 944 realitexpire(void *arg) 945 { 946 struct proc *p; 947 struct timeval ctv, ntv; 948 949 p = (struct proc *)arg; 950 PHOLD(p); 951 lwkt_gettoken(&p->p_token); 952 ksignal(p, SIGALRM); 953 if (!timevalisset(&p->p_realtimer.it_interval)) { 954 timevalclear(&p->p_realtimer.it_value); 955 goto done; 956 } 957 for (;;) { 958 timevaladd(&p->p_realtimer.it_value, 959 &p->p_realtimer.it_interval); 960 getmicrouptime(&ctv); 961 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 962 ntv = p->p_realtimer.it_value; 963 timevalsub(&ntv, &ctv); 964 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 965 realitexpire, p); 966 goto done; 967 } 968 } 969 done: 970 lwkt_reltoken(&p->p_token); 971 PRELE(p); 972 } 973 974 /* 975 * Used to validate itimer timeouts and utimes*() timespecs. 976 */ 977 int 978 itimerfix(struct timeval *tv) 979 { 980 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 981 return (EINVAL); 982 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 983 tv->tv_usec = ustick; 984 return (0); 985 } 986 987 /* 988 * Used to validate timeouts and utimes*() timespecs. 989 */ 990 int 991 itimespecfix(struct timespec *ts) 992 { 993 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000ULL) 994 return (EINVAL); 995 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < nstick) 996 ts->tv_nsec = nstick; 997 return (0); 998 } 999 1000 /* 1001 * Decrement an interval timer by a specified number 1002 * of microseconds, which must be less than a second, 1003 * i.e. < 1000000. If the timer expires, then reload 1004 * it. In this case, carry over (usec - old value) to 1005 * reduce the value reloaded into the timer so that 1006 * the timer does not drift. This routine assumes 1007 * that it is called in a context where the timers 1008 * on which it is operating cannot change in value. 1009 */ 1010 int 1011 itimerdecr(struct itimerval *itp, int usec) 1012 { 1013 1014 if (itp->it_value.tv_usec < usec) { 1015 if (itp->it_value.tv_sec == 0) { 1016 /* expired, and already in next interval */ 1017 usec -= itp->it_value.tv_usec; 1018 goto expire; 1019 } 1020 itp->it_value.tv_usec += 1000000; 1021 itp->it_value.tv_sec--; 1022 } 1023 itp->it_value.tv_usec -= usec; 1024 usec = 0; 1025 if (timevalisset(&itp->it_value)) 1026 return (1); 1027 /* expired, exactly at end of interval */ 1028 expire: 1029 if (timevalisset(&itp->it_interval)) { 1030 itp->it_value = itp->it_interval; 1031 itp->it_value.tv_usec -= usec; 1032 if (itp->it_value.tv_usec < 0) { 1033 itp->it_value.tv_usec += 1000000; 1034 itp->it_value.tv_sec--; 1035 } 1036 } else 1037 itp->it_value.tv_usec = 0; /* sec is already 0 */ 1038 return (0); 1039 } 1040 1041 /* 1042 * Add and subtract routines for timevals. 1043 * N.B.: subtract routine doesn't deal with 1044 * results which are before the beginning, 1045 * it just gets very confused in this case. 1046 * Caveat emptor. 1047 */ 1048 void 1049 timevaladd(struct timeval *t1, const struct timeval *t2) 1050 { 1051 1052 t1->tv_sec += t2->tv_sec; 1053 t1->tv_usec += t2->tv_usec; 1054 timevalfix(t1); 1055 } 1056 1057 void 1058 timevalsub(struct timeval *t1, const struct timeval *t2) 1059 { 1060 1061 t1->tv_sec -= t2->tv_sec; 1062 t1->tv_usec -= t2->tv_usec; 1063 timevalfix(t1); 1064 } 1065 1066 static void 1067 timevalfix(struct timeval *t1) 1068 { 1069 1070 if (t1->tv_usec < 0) { 1071 t1->tv_sec--; 1072 t1->tv_usec += 1000000; 1073 } 1074 if (t1->tv_usec >= 1000000) { 1075 t1->tv_sec++; 1076 t1->tv_usec -= 1000000; 1077 } 1078 } 1079 1080 /* 1081 * ratecheck(): simple time-based rate-limit checking. 1082 */ 1083 int 1084 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1085 { 1086 struct timeval tv, delta; 1087 int rv = 0; 1088 1089 getmicrouptime(&tv); /* NB: 10ms precision */ 1090 delta = tv; 1091 timevalsub(&delta, lasttime); 1092 1093 /* 1094 * check for 0,0 is so that the message will be seen at least once, 1095 * even if interval is huge. 1096 */ 1097 if (timevalcmp(&delta, mininterval, >=) || 1098 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1099 *lasttime = tv; 1100 rv = 1; 1101 } 1102 1103 return (rv); 1104 } 1105 1106 /* 1107 * ppsratecheck(): packets (or events) per second limitation. 1108 * 1109 * Return 0 if the limit is to be enforced (e.g. the caller 1110 * should drop a packet because of the rate limitation). 1111 * 1112 * maxpps of 0 always causes zero to be returned. maxpps of -1 1113 * always causes 1 to be returned; this effectively defeats rate 1114 * limiting. 1115 * 1116 * Note that we maintain the struct timeval for compatibility 1117 * with other bsd systems. We reuse the storage and just monitor 1118 * clock ticks for minimal overhead. 1119 */ 1120 int 1121 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1122 { 1123 int now; 1124 1125 /* 1126 * Reset the last time and counter if this is the first call 1127 * or more than a second has passed since the last update of 1128 * lasttime. 1129 */ 1130 now = ticks; 1131 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1132 lasttime->tv_sec = now; 1133 *curpps = 1; 1134 return (maxpps != 0); 1135 } else { 1136 (*curpps)++; /* NB: ignore potential overflow */ 1137 return (maxpps < 0 || *curpps < maxpps); 1138 } 1139 } 1140 1141 static int 1142 sysctl_gettimeofday_quick(SYSCTL_HANDLER_ARGS) 1143 { 1144 int error; 1145 int gtod; 1146 1147 gtod = gettimeofday_quick; 1148 error = sysctl_handle_int(oidp, >od, 0, req); 1149 if (error || req->newptr == NULL) 1150 return error; 1151 gettimeofday_quick = gtod; 1152 if (kpmap) 1153 kpmap->fast_gtod = gtod; 1154 return 0; 1155 } 1156