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 int tried_yield; 300 301 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 302 return (EINVAL); 303 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */ 304 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 305 return (0); 306 nanouptime(&ts); 307 timespecadd(&ts, rqt); /* ts = target timestamp compare */ 308 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 309 tried_yield = 0; 310 311 for (;;) { 312 int ticks; 313 struct systimer info; 314 315 ticks = tv.tv_usec / ustick; /* approximate */ 316 317 if (tv.tv_sec == 0 && ticks == 0) { 318 thread_t td = curthread; 319 if (tried_yield || tv.tv_usec < sleep_hard_us) { 320 tried_yield = 0; 321 uio_yield(); 322 } else { 323 crit_enter_quick(td); 324 systimer_init_oneshot(&info, ns1_systimer, 325 td, tv.tv_usec); 326 lwkt_deschedule_self(td); 327 crit_exit_quick(td); 328 lwkt_switch(); 329 systimer_del(&info); /* make sure it's gone */ 330 } 331 error = iscaught(td->td_lwp); 332 } else if (tv.tv_sec == 0) { 333 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 334 } else { 335 ticks = tvtohz_low(&tv); /* also handles overflow */ 336 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 337 } 338 nanouptime(&ts2); 339 if (error && error != EWOULDBLOCK) { 340 if (error == ERESTART) 341 error = EINTR; 342 if (rmt != NULL) { 343 timespecsub(&ts, &ts2); 344 if (ts.tv_sec < 0) 345 timespecclear(&ts); 346 *rmt = ts; 347 } 348 return (error); 349 } 350 if (timespeccmp(&ts2, &ts, >=)) 351 return (0); 352 ts3 = ts; 353 timespecsub(&ts3, &ts2); 354 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 355 } 356 } 357 358 /* 359 * MPSAFE 360 */ 361 int 362 sys_nanosleep(struct nanosleep_args *uap) 363 { 364 int error; 365 struct timespec rqt; 366 struct timespec rmt; 367 368 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 369 if (error) 370 return (error); 371 372 error = nanosleep1(&rqt, &rmt); 373 374 /* 375 * copyout the residual if nanosleep was interrupted. 376 */ 377 if (error && uap->rmtp) { 378 int error2; 379 380 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 381 if (error2) 382 error = error2; 383 } 384 return (error); 385 } 386 387 /* 388 * MPSAFE 389 */ 390 int 391 sys_gettimeofday(struct gettimeofday_args *uap) 392 { 393 struct timeval atv; 394 int error = 0; 395 396 if (uap->tp) { 397 microtime(&atv); 398 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 399 sizeof (atv)))) 400 return (error); 401 } 402 if (uap->tzp) 403 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 404 sizeof (tz)); 405 return (error); 406 } 407 408 /* 409 * MPALMOSTSAFE 410 */ 411 int 412 sys_settimeofday(struct settimeofday_args *uap) 413 { 414 struct thread *td = curthread; 415 struct timeval atv; 416 struct timezone atz; 417 int error; 418 419 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 420 return (error); 421 /* Verify all parameters before changing time. */ 422 if (uap->tv) { 423 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 424 sizeof(atv)))) 425 return (error); 426 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 427 return (EINVAL); 428 } 429 if (uap->tzp && 430 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 431 return (error); 432 433 get_mplock(); 434 if (uap->tv && (error = settime(&atv))) { 435 rel_mplock(); 436 return (error); 437 } 438 rel_mplock(); 439 if (uap->tzp) 440 tz = atz; 441 return (0); 442 } 443 444 static void 445 kern_adjtime_common(void) 446 { 447 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 448 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 449 ntp_tick_delta = ntp_delta; 450 else if (ntp_delta > ntp_big_delta) 451 ntp_tick_delta = 10 * ntp_default_tick_delta; 452 else if (ntp_delta < -ntp_big_delta) 453 ntp_tick_delta = -10 * ntp_default_tick_delta; 454 else if (ntp_delta > 0) 455 ntp_tick_delta = ntp_default_tick_delta; 456 else 457 ntp_tick_delta = -ntp_default_tick_delta; 458 } 459 460 void 461 kern_adjtime(int64_t delta, int64_t *odelta) 462 { 463 int origcpu; 464 465 if ((origcpu = mycpu->gd_cpuid) != 0) 466 lwkt_setcpu_self(globaldata_find(0)); 467 468 crit_enter(); 469 *odelta = ntp_delta; 470 ntp_delta = delta; 471 kern_adjtime_common(); 472 crit_exit(); 473 474 if (origcpu != 0) 475 lwkt_setcpu_self(globaldata_find(origcpu)); 476 } 477 478 static void 479 kern_get_ntp_delta(int64_t *delta) 480 { 481 int origcpu; 482 483 if ((origcpu = mycpu->gd_cpuid) != 0) 484 lwkt_setcpu_self(globaldata_find(0)); 485 486 crit_enter(); 487 *delta = ntp_delta; 488 crit_exit(); 489 490 if (origcpu != 0) 491 lwkt_setcpu_self(globaldata_find(origcpu)); 492 } 493 494 void 495 kern_reladjtime(int64_t delta) 496 { 497 int origcpu; 498 499 if ((origcpu = mycpu->gd_cpuid) != 0) 500 lwkt_setcpu_self(globaldata_find(0)); 501 502 crit_enter(); 503 ntp_delta += delta; 504 kern_adjtime_common(); 505 crit_exit(); 506 507 if (origcpu != 0) 508 lwkt_setcpu_self(globaldata_find(origcpu)); 509 } 510 511 static void 512 kern_adjfreq(int64_t rate) 513 { 514 int origcpu; 515 516 if ((origcpu = mycpu->gd_cpuid) != 0) 517 lwkt_setcpu_self(globaldata_find(0)); 518 519 crit_enter(); 520 ntp_tick_permanent = rate; 521 crit_exit(); 522 523 if (origcpu != 0) 524 lwkt_setcpu_self(globaldata_find(origcpu)); 525 } 526 527 /* 528 * MPALMOSTSAFE 529 */ 530 int 531 sys_adjtime(struct adjtime_args *uap) 532 { 533 struct thread *td = curthread; 534 struct timeval atv; 535 int64_t ndelta, odelta; 536 int error; 537 538 if ((error = priv_check(td, PRIV_ADJTIME))) 539 return (error); 540 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 541 if (error) 542 return (error); 543 544 /* 545 * Compute the total correction and the rate at which to apply it. 546 * Round the adjustment down to a whole multiple of the per-tick 547 * delta, so that after some number of incremental changes in 548 * hardclock(), tickdelta will become zero, lest the correction 549 * overshoot and start taking us away from the desired final time. 550 */ 551 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 552 get_mplock(); 553 kern_adjtime(ndelta, &odelta); 554 rel_mplock(); 555 556 if (uap->olddelta) { 557 atv.tv_sec = odelta / 1000000000; 558 atv.tv_usec = odelta % 1000000000 / 1000; 559 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 560 } 561 return (0); 562 } 563 564 static int 565 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 566 { 567 int64_t delta; 568 int error; 569 570 if (req->newptr != NULL) { 571 if (priv_check(curthread, PRIV_ROOT)) 572 return (EPERM); 573 error = SYSCTL_IN(req, &delta, sizeof(delta)); 574 if (error) 575 return (error); 576 kern_reladjtime(delta); 577 } 578 579 if (req->oldptr) 580 kern_get_ntp_delta(&delta); 581 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 582 return (error); 583 } 584 585 /* 586 * delta is in nanoseconds. 587 */ 588 static int 589 sysctl_delta(SYSCTL_HANDLER_ARGS) 590 { 591 int64_t delta, old_delta; 592 int error; 593 594 if (req->newptr != NULL) { 595 if (priv_check(curthread, PRIV_ROOT)) 596 return (EPERM); 597 error = SYSCTL_IN(req, &delta, sizeof(delta)); 598 if (error) 599 return (error); 600 kern_adjtime(delta, &old_delta); 601 } 602 603 if (req->oldptr != NULL) 604 kern_get_ntp_delta(&old_delta); 605 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 606 return (error); 607 } 608 609 /* 610 * frequency is in nanoseconds per second shifted left 32. 611 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 612 */ 613 static int 614 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 615 { 616 int64_t freqdelta; 617 int error; 618 619 if (req->newptr != NULL) { 620 if (priv_check(curthread, PRIV_ROOT)) 621 return (EPERM); 622 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 623 if (error) 624 return (error); 625 626 freqdelta /= hz; 627 kern_adjfreq(freqdelta); 628 } 629 630 if (req->oldptr != NULL) 631 freqdelta = ntp_tick_permanent * hz; 632 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 633 if (error) 634 return (error); 635 636 return (0); 637 } 638 639 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 640 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 641 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 642 sysctl_adjfreq, "Q", "permanent correction per second"); 643 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 644 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 645 sysctl_delta, "Q", "one-time delta"); 646 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 647 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 648 "threshold for fast adjustment"); 649 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 650 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 651 "per-tick adjustment"); 652 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 653 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 654 "default per-tick adjustment"); 655 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 656 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 657 "next leap second"); 658 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 659 &ntp_leap_insert, 0, "insert or remove leap second"); 660 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 661 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 662 sysctl_adjtime, "Q", "relative adjust for delta"); 663 664 /* 665 * Get value of an interval timer. The process virtual and 666 * profiling virtual time timers are kept in the p_stats area, since 667 * they can be swapped out. These are kept internally in the 668 * way they are specified externally: in time until they expire. 669 * 670 * The real time interval timer is kept in the process table slot 671 * for the process, and its value (it_value) is kept as an 672 * absolute time rather than as a delta, so that it is easy to keep 673 * periodic real-time signals from drifting. 674 * 675 * Virtual time timers are processed in the hardclock() routine of 676 * kern_clock.c. The real time timer is processed by a timeout 677 * routine, called from the softclock() routine. Since a callout 678 * may be delayed in real time due to interrupt processing in the system, 679 * it is possible for the real time timeout routine (realitexpire, given below), 680 * to be delayed in real time past when it is supposed to occur. It 681 * does not suffice, therefore, to reload the real timer .it_value from the 682 * real time timers .it_interval. Rather, we compute the next time in 683 * absolute time the timer should go off. 684 * 685 * MPALMOSTSAFE 686 */ 687 int 688 sys_getitimer(struct getitimer_args *uap) 689 { 690 struct proc *p = curproc; 691 struct timeval ctv; 692 struct itimerval aitv; 693 694 if (uap->which > ITIMER_PROF) 695 return (EINVAL); 696 get_mplock(); 697 crit_enter(); 698 if (uap->which == ITIMER_REAL) { 699 /* 700 * Convert from absolute to relative time in .it_value 701 * part of real time timer. If time for real time timer 702 * has passed return 0, else return difference between 703 * current time and time for the timer to go off. 704 */ 705 aitv = p->p_realtimer; 706 if (timevalisset(&aitv.it_value)) { 707 getmicrouptime(&ctv); 708 if (timevalcmp(&aitv.it_value, &ctv, <)) 709 timevalclear(&aitv.it_value); 710 else 711 timevalsub(&aitv.it_value, &ctv); 712 } 713 } else { 714 aitv = p->p_timer[uap->which]; 715 } 716 crit_exit(); 717 rel_mplock(); 718 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 719 } 720 721 /* 722 * MPALMOSTSAFE 723 */ 724 int 725 sys_setitimer(struct setitimer_args *uap) 726 { 727 struct itimerval aitv; 728 struct timeval ctv; 729 struct itimerval *itvp; 730 struct proc *p = curproc; 731 int error; 732 733 if (uap->which > ITIMER_PROF) 734 return (EINVAL); 735 itvp = uap->itv; 736 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 737 sizeof(struct itimerval)))) 738 return (error); 739 if ((uap->itv = uap->oitv) && 740 (error = sys_getitimer((struct getitimer_args *)uap))) 741 return (error); 742 if (itvp == 0) 743 return (0); 744 if (itimerfix(&aitv.it_value)) 745 return (EINVAL); 746 if (!timevalisset(&aitv.it_value)) 747 timevalclear(&aitv.it_interval); 748 else if (itimerfix(&aitv.it_interval)) 749 return (EINVAL); 750 get_mplock(); 751 crit_enter(); 752 if (uap->which == ITIMER_REAL) { 753 if (timevalisset(&p->p_realtimer.it_value)) 754 callout_stop(&p->p_ithandle); 755 if (timevalisset(&aitv.it_value)) 756 callout_reset(&p->p_ithandle, 757 tvtohz_high(&aitv.it_value), realitexpire, p); 758 getmicrouptime(&ctv); 759 timevaladd(&aitv.it_value, &ctv); 760 p->p_realtimer = aitv; 761 } else { 762 p->p_timer[uap->which] = aitv; 763 } 764 crit_exit(); 765 rel_mplock(); 766 return (0); 767 } 768 769 /* 770 * Real interval timer expired: 771 * send process whose timer expired an alarm signal. 772 * If time is not set up to reload, then just return. 773 * Else compute next time timer should go off which is > current time. 774 * This is where delay in processing this timeout causes multiple 775 * SIGALRM calls to be compressed into one. 776 * tvtohz_high() always adds 1 to allow for the time until the next clock 777 * interrupt being strictly less than 1 clock tick, but we don't want 778 * that here since we want to appear to be in sync with the clock 779 * interrupt even when we're delayed. 780 */ 781 void 782 realitexpire(void *arg) 783 { 784 struct proc *p; 785 struct timeval ctv, ntv; 786 787 p = (struct proc *)arg; 788 ksignal(p, SIGALRM); 789 if (!timevalisset(&p->p_realtimer.it_interval)) { 790 timevalclear(&p->p_realtimer.it_value); 791 return; 792 } 793 for (;;) { 794 crit_enter(); 795 timevaladd(&p->p_realtimer.it_value, 796 &p->p_realtimer.it_interval); 797 getmicrouptime(&ctv); 798 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 799 ntv = p->p_realtimer.it_value; 800 timevalsub(&ntv, &ctv); 801 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 802 realitexpire, p); 803 crit_exit(); 804 return; 805 } 806 crit_exit(); 807 } 808 } 809 810 /* 811 * Check that a proposed value to load into the .it_value or 812 * .it_interval part of an interval timer is acceptable, and 813 * fix it to have at least minimal value (i.e. if it is less 814 * than the resolution of the clock, round it up.) 815 * 816 * MPSAFE 817 */ 818 int 819 itimerfix(struct timeval *tv) 820 { 821 822 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 || 823 tv->tv_usec < 0 || tv->tv_usec >= 1000000) 824 return (EINVAL); 825 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 826 tv->tv_usec = ustick; 827 return (0); 828 } 829 830 /* 831 * Decrement an interval timer by a specified number 832 * of microseconds, which must be less than a second, 833 * i.e. < 1000000. If the timer expires, then reload 834 * it. In this case, carry over (usec - old value) to 835 * reduce the value reloaded into the timer so that 836 * the timer does not drift. This routine assumes 837 * that it is called in a context where the timers 838 * on which it is operating cannot change in value. 839 */ 840 int 841 itimerdecr(struct itimerval *itp, int usec) 842 { 843 844 if (itp->it_value.tv_usec < usec) { 845 if (itp->it_value.tv_sec == 0) { 846 /* expired, and already in next interval */ 847 usec -= itp->it_value.tv_usec; 848 goto expire; 849 } 850 itp->it_value.tv_usec += 1000000; 851 itp->it_value.tv_sec--; 852 } 853 itp->it_value.tv_usec -= usec; 854 usec = 0; 855 if (timevalisset(&itp->it_value)) 856 return (1); 857 /* expired, exactly at end of interval */ 858 expire: 859 if (timevalisset(&itp->it_interval)) { 860 itp->it_value = itp->it_interval; 861 itp->it_value.tv_usec -= usec; 862 if (itp->it_value.tv_usec < 0) { 863 itp->it_value.tv_usec += 1000000; 864 itp->it_value.tv_sec--; 865 } 866 } else 867 itp->it_value.tv_usec = 0; /* sec is already 0 */ 868 return (0); 869 } 870 871 /* 872 * Add and subtract routines for timevals. 873 * N.B.: subtract routine doesn't deal with 874 * results which are before the beginning, 875 * it just gets very confused in this case. 876 * Caveat emptor. 877 */ 878 void 879 timevaladd(struct timeval *t1, const struct timeval *t2) 880 { 881 882 t1->tv_sec += t2->tv_sec; 883 t1->tv_usec += t2->tv_usec; 884 timevalfix(t1); 885 } 886 887 void 888 timevalsub(struct timeval *t1, const struct timeval *t2) 889 { 890 891 t1->tv_sec -= t2->tv_sec; 892 t1->tv_usec -= t2->tv_usec; 893 timevalfix(t1); 894 } 895 896 static void 897 timevalfix(struct timeval *t1) 898 { 899 900 if (t1->tv_usec < 0) { 901 t1->tv_sec--; 902 t1->tv_usec += 1000000; 903 } 904 if (t1->tv_usec >= 1000000) { 905 t1->tv_sec++; 906 t1->tv_usec -= 1000000; 907 } 908 } 909 910 /* 911 * ratecheck(): simple time-based rate-limit checking. 912 */ 913 int 914 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 915 { 916 struct timeval tv, delta; 917 int rv = 0; 918 919 getmicrouptime(&tv); /* NB: 10ms precision */ 920 delta = tv; 921 timevalsub(&delta, lasttime); 922 923 /* 924 * check for 0,0 is so that the message will be seen at least once, 925 * even if interval is huge. 926 */ 927 if (timevalcmp(&delta, mininterval, >=) || 928 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 929 *lasttime = tv; 930 rv = 1; 931 } 932 933 return (rv); 934 } 935 936 /* 937 * ppsratecheck(): packets (or events) per second limitation. 938 * 939 * Return 0 if the limit is to be enforced (e.g. the caller 940 * should drop a packet because of the rate limitation). 941 * 942 * maxpps of 0 always causes zero to be returned. maxpps of -1 943 * always causes 1 to be returned; this effectively defeats rate 944 * limiting. 945 * 946 * Note that we maintain the struct timeval for compatibility 947 * with other bsd systems. We reuse the storage and just monitor 948 * clock ticks for minimal overhead. 949 */ 950 int 951 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 952 { 953 int now; 954 955 /* 956 * Reset the last time and counter if this is the first call 957 * or more than a second has passed since the last update of 958 * lasttime. 959 */ 960 now = ticks; 961 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 962 lasttime->tv_sec = now; 963 *curpps = 1; 964 return (maxpps != 0); 965 } else { 966 (*curpps)++; /* NB: ignore potential overflow */ 967 return (maxpps < 0 || *curpps < maxpps); 968 } 969 } 970 971