1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93 34 * $FreeBSD: src/sys/kern/kern_time.c,v 1.68.2.1 2002/10/01 08:00:41 bde Exp $ 35 * $DragonFly: src/sys/kern/kern_time.c,v 1.40 2008/04/02 14:16:16 sephe Exp $ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/buf.h> 41 #include <sys/sysproto.h> 42 #include <sys/resourcevar.h> 43 #include <sys/signalvar.h> 44 #include <sys/kernel.h> 45 #include <sys/systm.h> 46 #include <sys/sysent.h> 47 #include <sys/sysunion.h> 48 #include <sys/proc.h> 49 #include <sys/priv.h> 50 #include <sys/time.h> 51 #include <sys/vnode.h> 52 #include <sys/sysctl.h> 53 #include <sys/kern_syscall.h> 54 #include <vm/vm.h> 55 #include <vm/vm_extern.h> 56 57 #include <sys/msgport2.h> 58 #include <sys/thread2.h> 59 #include <sys/mplock2.h> 60 61 struct timezone tz; 62 63 /* 64 * Time of day and interval timer support. 65 * 66 * These routines provide the kernel entry points to get and set 67 * the time-of-day and per-process interval timers. Subroutines 68 * here provide support for adding and subtracting timeval structures 69 * and decrementing interval timers, optionally reloading the interval 70 * timers when they expire. 71 */ 72 73 int nanosleep1(struct timespec *rqt, struct timespec *rmt); 74 static int settime(struct timeval *); 75 static void timevalfix(struct timeval *); 76 77 static int sleep_hard_us = 100; 78 SYSCTL_INT(_kern, OID_AUTO, sleep_hard_us, CTLFLAG_RW, &sleep_hard_us, 0, "") 79 80 static int 81 settime(struct timeval *tv) 82 { 83 struct timeval delta, tv1, tv2; 84 static struct timeval maxtime, laststep; 85 struct timespec ts; 86 int origcpu; 87 88 if ((origcpu = mycpu->gd_cpuid) != 0) 89 lwkt_setcpu_self(globaldata_find(0)); 90 91 crit_enter(); 92 microtime(&tv1); 93 delta = *tv; 94 timevalsub(&delta, &tv1); 95 96 /* 97 * If the system is secure, we do not allow the time to be 98 * set to a value earlier than 1 second less than the highest 99 * time we have yet seen. The worst a miscreant can do in 100 * this circumstance is "freeze" time. He couldn't go 101 * back to the past. 102 * 103 * We similarly do not allow the clock to be stepped more 104 * than one second, nor more than once per second. This allows 105 * a miscreant to make the clock march double-time, but no worse. 106 */ 107 if (securelevel > 1) { 108 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 109 /* 110 * Update maxtime to latest time we've seen. 111 */ 112 if (tv1.tv_sec > maxtime.tv_sec) 113 maxtime = tv1; 114 tv2 = *tv; 115 timevalsub(&tv2, &maxtime); 116 if (tv2.tv_sec < -1) { 117 tv->tv_sec = maxtime.tv_sec - 1; 118 kprintf("Time adjustment clamped to -1 second\n"); 119 } 120 } else { 121 if (tv1.tv_sec == laststep.tv_sec) { 122 crit_exit(); 123 return (EPERM); 124 } 125 if (delta.tv_sec > 1) { 126 tv->tv_sec = tv1.tv_sec + 1; 127 kprintf("Time adjustment clamped to +1 second\n"); 128 } 129 laststep = *tv; 130 } 131 } 132 133 ts.tv_sec = tv->tv_sec; 134 ts.tv_nsec = tv->tv_usec * 1000; 135 set_timeofday(&ts); 136 crit_exit(); 137 138 if (origcpu != 0) 139 lwkt_setcpu_self(globaldata_find(origcpu)); 140 141 resettodr(); 142 return (0); 143 } 144 145 /* 146 * MPSAFE 147 */ 148 int 149 kern_clock_gettime(clockid_t clock_id, struct timespec *ats) 150 { 151 int error = 0; 152 153 switch(clock_id) { 154 case CLOCK_REALTIME: 155 nanotime(ats); 156 break; 157 case CLOCK_MONOTONIC: 158 nanouptime(ats); 159 break; 160 default: 161 error = EINVAL; 162 break; 163 } 164 return (error); 165 } 166 167 /* 168 * MPSAFE 169 */ 170 int 171 sys_clock_gettime(struct clock_gettime_args *uap) 172 { 173 struct timespec ats; 174 int error; 175 176 error = kern_clock_gettime(uap->clock_id, &ats); 177 if (error == 0) 178 error = copyout(&ats, uap->tp, sizeof(ats)); 179 180 return (error); 181 } 182 183 int 184 kern_clock_settime(clockid_t clock_id, struct timespec *ats) 185 { 186 struct thread *td = curthread; 187 struct timeval atv; 188 int error; 189 190 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 191 return (error); 192 if (clock_id != CLOCK_REALTIME) 193 return (EINVAL); 194 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000) 195 return (EINVAL); 196 197 TIMESPEC_TO_TIMEVAL(&atv, ats); 198 error = settime(&atv); 199 return (error); 200 } 201 202 /* 203 * MPALMOSTSAFE 204 */ 205 int 206 sys_clock_settime(struct clock_settime_args *uap) 207 { 208 struct timespec ats; 209 int error; 210 211 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 212 return (error); 213 214 get_mplock(); 215 error = kern_clock_settime(uap->clock_id, &ats); 216 rel_mplock(); 217 return (error); 218 } 219 220 /* 221 * MPSAFE 222 */ 223 int 224 kern_clock_getres(clockid_t clock_id, struct timespec *ts) 225 { 226 int error; 227 228 switch(clock_id) { 229 case CLOCK_REALTIME: 230 case CLOCK_MONOTONIC: 231 /* 232 * Round up the result of the division cheaply 233 * by adding 1. Rounding up is especially important 234 * if rounding down would give 0. Perfect rounding 235 * is unimportant. 236 */ 237 ts->tv_sec = 0; 238 ts->tv_nsec = 1000000000 / sys_cputimer->freq + 1; 239 error = 0; 240 break; 241 default: 242 error = EINVAL; 243 break; 244 } 245 246 return(error); 247 } 248 249 /* 250 * MPSAFE 251 */ 252 int 253 sys_clock_getres(struct clock_getres_args *uap) 254 { 255 int error; 256 struct timespec ts; 257 258 error = kern_clock_getres(uap->clock_id, &ts); 259 if (error == 0) 260 error = copyout(&ts, uap->tp, sizeof(ts)); 261 262 return (error); 263 } 264 265 /* 266 * nanosleep1() 267 * 268 * This is a general helper function for nanosleep() (aka sleep() aka 269 * usleep()). 270 * 271 * If there is less then one tick's worth of time left and 272 * we haven't done a yield, or the remaining microseconds is 273 * ridiculously low, do a yield. This avoids having 274 * to deal with systimer overheads when the system is under 275 * heavy loads. If we have done a yield already then use 276 * a systimer and an uninterruptable thread wait. 277 * 278 * If there is more then a tick's worth of time left, 279 * calculate the baseline ticks and use an interruptable 280 * tsleep, then handle the fine-grained delay on the next 281 * loop. This usually results in two sleeps occuring, a long one 282 * and a short one. 283 * 284 * MPSAFE 285 */ 286 static void 287 ns1_systimer(systimer_t info) 288 { 289 lwkt_schedule(info->data); 290 } 291 292 int 293 nanosleep1(struct timespec *rqt, struct timespec *rmt) 294 { 295 static int nanowait; 296 struct timespec ts, ts2, ts3; 297 struct timeval tv; 298 int error; 299 300 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 301 return (EINVAL); 302 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 303 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 304 return (0); 305 nanouptime(&ts); 306 timespecadd(&ts, rqt); /* ts = target timestamp compare */ 307 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 308 309 for (;;) { 310 int ticks; 311 struct systimer info; 312 313 ticks = tv.tv_usec / ustick; /* approximate */ 314 315 if (tv.tv_sec == 0 && ticks == 0) { 316 thread_t td = curthread; 317 if (tv.tv_usec < sleep_hard_us) { 318 lwkt_user_yield(); 319 } else { 320 crit_enter_quick(td); 321 systimer_init_oneshot(&info, ns1_systimer, 322 td, tv.tv_usec); 323 lwkt_deschedule_self(td); 324 crit_exit_quick(td); 325 lwkt_switch(); 326 systimer_del(&info); /* make sure it's gone */ 327 } 328 error = iscaught(td->td_lwp); 329 } else if (tv.tv_sec == 0) { 330 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 331 } else { 332 ticks = tvtohz_low(&tv); /* also handles overflow */ 333 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 334 } 335 nanouptime(&ts2); 336 if (error && error != EWOULDBLOCK) { 337 if (error == ERESTART) 338 error = EINTR; 339 if (rmt != NULL) { 340 timespecsub(&ts, &ts2); 341 if (ts.tv_sec < 0) 342 timespecclear(&ts); 343 *rmt = ts; 344 } 345 return (error); 346 } 347 if (timespeccmp(&ts2, &ts, >=)) 348 return (0); 349 ts3 = ts; 350 timespecsub(&ts3, &ts2); 351 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 352 } 353 } 354 355 /* 356 * MPSAFE 357 */ 358 int 359 sys_nanosleep(struct nanosleep_args *uap) 360 { 361 int error; 362 struct timespec rqt; 363 struct timespec rmt; 364 365 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 366 if (error) 367 return (error); 368 369 error = nanosleep1(&rqt, &rmt); 370 371 /* 372 * copyout the residual if nanosleep was interrupted. 373 */ 374 if (error && uap->rmtp) { 375 int error2; 376 377 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 378 if (error2) 379 error = error2; 380 } 381 return (error); 382 } 383 384 /* 385 * MPSAFE 386 */ 387 int 388 sys_gettimeofday(struct gettimeofday_args *uap) 389 { 390 struct timeval atv; 391 int error = 0; 392 393 if (uap->tp) { 394 microtime(&atv); 395 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 396 sizeof (atv)))) 397 return (error); 398 } 399 if (uap->tzp) 400 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 401 sizeof (tz)); 402 return (error); 403 } 404 405 /* 406 * MPALMOSTSAFE 407 */ 408 int 409 sys_settimeofday(struct settimeofday_args *uap) 410 { 411 struct thread *td = curthread; 412 struct timeval atv; 413 struct timezone atz; 414 int error; 415 416 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 417 return (error); 418 /* Verify all parameters before changing time. */ 419 if (uap->tv) { 420 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 421 sizeof(atv)))) 422 return (error); 423 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 424 return (EINVAL); 425 } 426 if (uap->tzp && 427 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 428 return (error); 429 430 get_mplock(); 431 if (uap->tv && (error = settime(&atv))) { 432 rel_mplock(); 433 return (error); 434 } 435 rel_mplock(); 436 if (uap->tzp) 437 tz = atz; 438 return (0); 439 } 440 441 static void 442 kern_adjtime_common(void) 443 { 444 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 445 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 446 ntp_tick_delta = ntp_delta; 447 else if (ntp_delta > ntp_big_delta) 448 ntp_tick_delta = 10 * ntp_default_tick_delta; 449 else if (ntp_delta < -ntp_big_delta) 450 ntp_tick_delta = -10 * ntp_default_tick_delta; 451 else if (ntp_delta > 0) 452 ntp_tick_delta = ntp_default_tick_delta; 453 else 454 ntp_tick_delta = -ntp_default_tick_delta; 455 } 456 457 void 458 kern_adjtime(int64_t delta, int64_t *odelta) 459 { 460 int origcpu; 461 462 if ((origcpu = mycpu->gd_cpuid) != 0) 463 lwkt_setcpu_self(globaldata_find(0)); 464 465 crit_enter(); 466 *odelta = ntp_delta; 467 ntp_delta = delta; 468 kern_adjtime_common(); 469 crit_exit(); 470 471 if (origcpu != 0) 472 lwkt_setcpu_self(globaldata_find(origcpu)); 473 } 474 475 static void 476 kern_get_ntp_delta(int64_t *delta) 477 { 478 int origcpu; 479 480 if ((origcpu = mycpu->gd_cpuid) != 0) 481 lwkt_setcpu_self(globaldata_find(0)); 482 483 crit_enter(); 484 *delta = ntp_delta; 485 crit_exit(); 486 487 if (origcpu != 0) 488 lwkt_setcpu_self(globaldata_find(origcpu)); 489 } 490 491 void 492 kern_reladjtime(int64_t delta) 493 { 494 int origcpu; 495 496 if ((origcpu = mycpu->gd_cpuid) != 0) 497 lwkt_setcpu_self(globaldata_find(0)); 498 499 crit_enter(); 500 ntp_delta += delta; 501 kern_adjtime_common(); 502 crit_exit(); 503 504 if (origcpu != 0) 505 lwkt_setcpu_self(globaldata_find(origcpu)); 506 } 507 508 static void 509 kern_adjfreq(int64_t rate) 510 { 511 int origcpu; 512 513 if ((origcpu = mycpu->gd_cpuid) != 0) 514 lwkt_setcpu_self(globaldata_find(0)); 515 516 crit_enter(); 517 ntp_tick_permanent = rate; 518 crit_exit(); 519 520 if (origcpu != 0) 521 lwkt_setcpu_self(globaldata_find(origcpu)); 522 } 523 524 /* 525 * MPALMOSTSAFE 526 */ 527 int 528 sys_adjtime(struct adjtime_args *uap) 529 { 530 struct thread *td = curthread; 531 struct timeval atv; 532 int64_t ndelta, odelta; 533 int error; 534 535 if ((error = priv_check(td, PRIV_ADJTIME))) 536 return (error); 537 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 538 if (error) 539 return (error); 540 541 /* 542 * Compute the total correction and the rate at which to apply it. 543 * Round the adjustment down to a whole multiple of the per-tick 544 * delta, so that after some number of incremental changes in 545 * hardclock(), tickdelta will become zero, lest the correction 546 * overshoot and start taking us away from the desired final time. 547 */ 548 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 549 get_mplock(); 550 kern_adjtime(ndelta, &odelta); 551 rel_mplock(); 552 553 if (uap->olddelta) { 554 atv.tv_sec = odelta / 1000000000; 555 atv.tv_usec = odelta % 1000000000 / 1000; 556 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 557 } 558 return (0); 559 } 560 561 static int 562 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 563 { 564 int64_t delta; 565 int error; 566 567 if (req->newptr != NULL) { 568 if (priv_check(curthread, PRIV_ROOT)) 569 return (EPERM); 570 error = SYSCTL_IN(req, &delta, sizeof(delta)); 571 if (error) 572 return (error); 573 kern_reladjtime(delta); 574 } 575 576 if (req->oldptr) 577 kern_get_ntp_delta(&delta); 578 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 579 return (error); 580 } 581 582 /* 583 * delta is in nanoseconds. 584 */ 585 static int 586 sysctl_delta(SYSCTL_HANDLER_ARGS) 587 { 588 int64_t delta, old_delta; 589 int error; 590 591 if (req->newptr != NULL) { 592 if (priv_check(curthread, PRIV_ROOT)) 593 return (EPERM); 594 error = SYSCTL_IN(req, &delta, sizeof(delta)); 595 if (error) 596 return (error); 597 kern_adjtime(delta, &old_delta); 598 } 599 600 if (req->oldptr != NULL) 601 kern_get_ntp_delta(&old_delta); 602 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 603 return (error); 604 } 605 606 /* 607 * frequency is in nanoseconds per second shifted left 32. 608 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 609 */ 610 static int 611 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 612 { 613 int64_t freqdelta; 614 int error; 615 616 if (req->newptr != NULL) { 617 if (priv_check(curthread, PRIV_ROOT)) 618 return (EPERM); 619 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 620 if (error) 621 return (error); 622 623 freqdelta /= hz; 624 kern_adjfreq(freqdelta); 625 } 626 627 if (req->oldptr != NULL) 628 freqdelta = ntp_tick_permanent * hz; 629 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 630 if (error) 631 return (error); 632 633 return (0); 634 } 635 636 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 637 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 638 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 639 sysctl_adjfreq, "Q", "permanent correction per second"); 640 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 641 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 642 sysctl_delta, "Q", "one-time delta"); 643 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 644 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 645 "threshold for fast adjustment"); 646 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 647 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 648 "per-tick adjustment"); 649 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 650 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 651 "default per-tick adjustment"); 652 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 653 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 654 "next leap second"); 655 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 656 &ntp_leap_insert, 0, "insert or remove leap second"); 657 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 658 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 659 sysctl_adjtime, "Q", "relative adjust for delta"); 660 661 /* 662 * Get value of an interval timer. The process virtual and 663 * profiling virtual time timers are kept in the p_stats area, since 664 * they can be swapped out. These are kept internally in the 665 * way they are specified externally: in time until they expire. 666 * 667 * The real time interval timer is kept in the process table slot 668 * for the process, and its value (it_value) is kept as an 669 * absolute time rather than as a delta, so that it is easy to keep 670 * periodic real-time signals from drifting. 671 * 672 * Virtual time timers are processed in the hardclock() routine of 673 * kern_clock.c. The real time timer is processed by a timeout 674 * routine, called from the softclock() routine. Since a callout 675 * may be delayed in real time due to interrupt processing in the system, 676 * it is possible for the real time timeout routine (realitexpire, given below), 677 * to be delayed in real time past when it is supposed to occur. It 678 * does not suffice, therefore, to reload the real timer .it_value from the 679 * real time timers .it_interval. Rather, we compute the next time in 680 * absolute time the timer should go off. 681 * 682 * MPALMOSTSAFE 683 */ 684 int 685 sys_getitimer(struct getitimer_args *uap) 686 { 687 struct proc *p = curproc; 688 struct timeval ctv; 689 struct itimerval aitv; 690 691 if (uap->which > ITIMER_PROF) 692 return (EINVAL); 693 get_mplock(); 694 crit_enter(); 695 if (uap->which == ITIMER_REAL) { 696 /* 697 * Convert from absolute to relative time in .it_value 698 * part of real time timer. If time for real time timer 699 * has passed return 0, else return difference between 700 * current time and time for the timer to go off. 701 */ 702 aitv = p->p_realtimer; 703 if (timevalisset(&aitv.it_value)) { 704 getmicrouptime(&ctv); 705 if (timevalcmp(&aitv.it_value, &ctv, <)) 706 timevalclear(&aitv.it_value); 707 else 708 timevalsub(&aitv.it_value, &ctv); 709 } 710 } else { 711 aitv = p->p_timer[uap->which]; 712 } 713 crit_exit(); 714 rel_mplock(); 715 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 716 } 717 718 /* 719 * MPALMOSTSAFE 720 */ 721 int 722 sys_setitimer(struct setitimer_args *uap) 723 { 724 struct itimerval aitv; 725 struct timeval ctv; 726 struct itimerval *itvp; 727 struct proc *p = curproc; 728 int error; 729 730 if (uap->which > ITIMER_PROF) 731 return (EINVAL); 732 itvp = uap->itv; 733 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 734 sizeof(struct itimerval)))) 735 return (error); 736 if ((uap->itv = uap->oitv) && 737 (error = sys_getitimer((struct getitimer_args *)uap))) 738 return (error); 739 if (itvp == 0) 740 return (0); 741 if (itimerfix(&aitv.it_value)) 742 return (EINVAL); 743 if (!timevalisset(&aitv.it_value)) 744 timevalclear(&aitv.it_interval); 745 else if (itimerfix(&aitv.it_interval)) 746 return (EINVAL); 747 get_mplock(); 748 crit_enter(); 749 if (uap->which == ITIMER_REAL) { 750 if (timevalisset(&p->p_realtimer.it_value)) 751 callout_stop(&p->p_ithandle); 752 if (timevalisset(&aitv.it_value)) 753 callout_reset(&p->p_ithandle, 754 tvtohz_high(&aitv.it_value), realitexpire, p); 755 getmicrouptime(&ctv); 756 timevaladd(&aitv.it_value, &ctv); 757 p->p_realtimer = aitv; 758 } else { 759 p->p_timer[uap->which] = aitv; 760 } 761 crit_exit(); 762 rel_mplock(); 763 return (0); 764 } 765 766 /* 767 * Real interval timer expired: 768 * send process whose timer expired an alarm signal. 769 * If time is not set up to reload, then just return. 770 * Else compute next time timer should go off which is > current time. 771 * This is where delay in processing this timeout causes multiple 772 * SIGALRM calls to be compressed into one. 773 * tvtohz_high() always adds 1 to allow for the time until the next clock 774 * interrupt being strictly less than 1 clock tick, but we don't want 775 * that here since we want to appear to be in sync with the clock 776 * interrupt even when we're delayed. 777 */ 778 void 779 realitexpire(void *arg) 780 { 781 struct proc *p; 782 struct timeval ctv, ntv; 783 784 p = (struct proc *)arg; 785 ksignal(p, SIGALRM); 786 if (!timevalisset(&p->p_realtimer.it_interval)) { 787 timevalclear(&p->p_realtimer.it_value); 788 return; 789 } 790 for (;;) { 791 crit_enter(); 792 timevaladd(&p->p_realtimer.it_value, 793 &p->p_realtimer.it_interval); 794 getmicrouptime(&ctv); 795 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 796 ntv = p->p_realtimer.it_value; 797 timevalsub(&ntv, &ctv); 798 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 799 realitexpire, p); 800 crit_exit(); 801 return; 802 } 803 crit_exit(); 804 } 805 } 806 807 /* 808 * Check that a proposed value to load into the .it_value or 809 * .it_interval part of an interval timer is acceptable, and 810 * fix it to have at least minimal value (i.e. if it is less 811 * than the resolution of the clock, round it up.) 812 * 813 * MPSAFE 814 */ 815 int 816 itimerfix(struct timeval *tv) 817 { 818 819 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 || 820 tv->tv_usec < 0 || tv->tv_usec >= 1000000) 821 return (EINVAL); 822 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 823 tv->tv_usec = ustick; 824 return (0); 825 } 826 827 /* 828 * Decrement an interval timer by a specified number 829 * of microseconds, which must be less than a second, 830 * i.e. < 1000000. If the timer expires, then reload 831 * it. In this case, carry over (usec - old value) to 832 * reduce the value reloaded into the timer so that 833 * the timer does not drift. This routine assumes 834 * that it is called in a context where the timers 835 * on which it is operating cannot change in value. 836 */ 837 int 838 itimerdecr(struct itimerval *itp, int usec) 839 { 840 841 if (itp->it_value.tv_usec < usec) { 842 if (itp->it_value.tv_sec == 0) { 843 /* expired, and already in next interval */ 844 usec -= itp->it_value.tv_usec; 845 goto expire; 846 } 847 itp->it_value.tv_usec += 1000000; 848 itp->it_value.tv_sec--; 849 } 850 itp->it_value.tv_usec -= usec; 851 usec = 0; 852 if (timevalisset(&itp->it_value)) 853 return (1); 854 /* expired, exactly at end of interval */ 855 expire: 856 if (timevalisset(&itp->it_interval)) { 857 itp->it_value = itp->it_interval; 858 itp->it_value.tv_usec -= usec; 859 if (itp->it_value.tv_usec < 0) { 860 itp->it_value.tv_usec += 1000000; 861 itp->it_value.tv_sec--; 862 } 863 } else 864 itp->it_value.tv_usec = 0; /* sec is already 0 */ 865 return (0); 866 } 867 868 /* 869 * Add and subtract routines for timevals. 870 * N.B.: subtract routine doesn't deal with 871 * results which are before the beginning, 872 * it just gets very confused in this case. 873 * Caveat emptor. 874 */ 875 void 876 timevaladd(struct timeval *t1, const struct timeval *t2) 877 { 878 879 t1->tv_sec += t2->tv_sec; 880 t1->tv_usec += t2->tv_usec; 881 timevalfix(t1); 882 } 883 884 void 885 timevalsub(struct timeval *t1, const struct timeval *t2) 886 { 887 888 t1->tv_sec -= t2->tv_sec; 889 t1->tv_usec -= t2->tv_usec; 890 timevalfix(t1); 891 } 892 893 static void 894 timevalfix(struct timeval *t1) 895 { 896 897 if (t1->tv_usec < 0) { 898 t1->tv_sec--; 899 t1->tv_usec += 1000000; 900 } 901 if (t1->tv_usec >= 1000000) { 902 t1->tv_sec++; 903 t1->tv_usec -= 1000000; 904 } 905 } 906 907 /* 908 * ratecheck(): simple time-based rate-limit checking. 909 */ 910 int 911 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 912 { 913 struct timeval tv, delta; 914 int rv = 0; 915 916 getmicrouptime(&tv); /* NB: 10ms precision */ 917 delta = tv; 918 timevalsub(&delta, lasttime); 919 920 /* 921 * check for 0,0 is so that the message will be seen at least once, 922 * even if interval is huge. 923 */ 924 if (timevalcmp(&delta, mininterval, >=) || 925 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 926 *lasttime = tv; 927 rv = 1; 928 } 929 930 return (rv); 931 } 932 933 /* 934 * ppsratecheck(): packets (or events) per second limitation. 935 * 936 * Return 0 if the limit is to be enforced (e.g. the caller 937 * should drop a packet because of the rate limitation). 938 * 939 * maxpps of 0 always causes zero to be returned. maxpps of -1 940 * always causes 1 to be returned; this effectively defeats rate 941 * limiting. 942 * 943 * Note that we maintain the struct timeval for compatibility 944 * with other bsd systems. We reuse the storage and just monitor 945 * clock ticks for minimal overhead. 946 */ 947 int 948 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 949 { 950 int now; 951 952 /* 953 * Reset the last time and counter if this is the first call 954 * or more than a second has passed since the last update of 955 * lasttime. 956 */ 957 now = ticks; 958 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 959 lasttime->tv_sec = now; 960 *curpps = 1; 961 return (maxpps != 0); 962 } else { 963 (*curpps)++; /* NB: ignore potential overflow */ 964 return (maxpps < 0 || *curpps < maxpps); 965 } 966 } 967 968