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