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, int in_ipi __unused, 288 struct intrframe *frame __unused) 289 { 290 lwkt_schedule(info->data); 291 } 292 293 int 294 nanosleep1(struct timespec *rqt, struct timespec *rmt) 295 { 296 static int nanowait; 297 struct timespec ts, ts2, ts3; 298 struct timeval tv; 299 int error; 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 */ 309 310 for (;;) { 311 int ticks; 312 struct systimer info; 313 314 ticks = tv.tv_usec / ustick; /* approximate */ 315 316 if (tv.tv_sec == 0 && ticks == 0) { 317 thread_t td = curthread; 318 if (tv.tv_usec < sleep_hard_us) { 319 lwkt_user_yield(); 320 } else { 321 crit_enter_quick(td); 322 systimer_init_oneshot(&info, ns1_systimer, 323 td, tv.tv_usec); 324 lwkt_deschedule_self(td); 325 crit_exit_quick(td); 326 lwkt_switch(); 327 systimer_del(&info); /* make sure it's gone */ 328 } 329 error = iscaught(td->td_lwp); 330 } else if (tv.tv_sec == 0) { 331 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 332 } else { 333 ticks = tvtohz_low(&tv); /* also handles overflow */ 334 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 335 } 336 nanouptime(&ts2); 337 if (error && error != EWOULDBLOCK) { 338 if (error == ERESTART) 339 error = EINTR; 340 if (rmt != NULL) { 341 timespecsub(&ts, &ts2); 342 if (ts.tv_sec < 0) 343 timespecclear(&ts); 344 *rmt = ts; 345 } 346 return (error); 347 } 348 if (timespeccmp(&ts2, &ts, >=)) 349 return (0); 350 ts3 = ts; 351 timespecsub(&ts3, &ts2); 352 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 353 } 354 } 355 356 /* 357 * MPSAFE 358 */ 359 int 360 sys_nanosleep(struct nanosleep_args *uap) 361 { 362 int error; 363 struct timespec rqt; 364 struct timespec rmt; 365 366 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 367 if (error) 368 return (error); 369 370 error = nanosleep1(&rqt, &rmt); 371 372 /* 373 * copyout the residual if nanosleep was interrupted. 374 */ 375 if (error && uap->rmtp) { 376 int error2; 377 378 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 379 if (error2) 380 error = error2; 381 } 382 return (error); 383 } 384 385 /* 386 * MPSAFE 387 */ 388 int 389 sys_gettimeofday(struct gettimeofday_args *uap) 390 { 391 struct timeval atv; 392 int error = 0; 393 394 if (uap->tp) { 395 microtime(&atv); 396 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 397 sizeof (atv)))) 398 return (error); 399 } 400 if (uap->tzp) 401 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 402 sizeof (tz)); 403 return (error); 404 } 405 406 /* 407 * MPALMOSTSAFE 408 */ 409 int 410 sys_settimeofday(struct settimeofday_args *uap) 411 { 412 struct thread *td = curthread; 413 struct timeval atv; 414 struct timezone atz; 415 int error; 416 417 if ((error = priv_check(td, PRIV_SETTIMEOFDAY))) 418 return (error); 419 /* Verify all parameters before changing time. */ 420 if (uap->tv) { 421 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 422 sizeof(atv)))) 423 return (error); 424 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 425 return (EINVAL); 426 } 427 if (uap->tzp && 428 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 429 return (error); 430 431 get_mplock(); 432 if (uap->tv && (error = settime(&atv))) { 433 rel_mplock(); 434 return (error); 435 } 436 rel_mplock(); 437 if (uap->tzp) 438 tz = atz; 439 return (0); 440 } 441 442 static void 443 kern_adjtime_common(void) 444 { 445 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 446 (ntp_delta < 0 && ntp_delta > -ntp_default_tick_delta)) 447 ntp_tick_delta = ntp_delta; 448 else if (ntp_delta > ntp_big_delta) 449 ntp_tick_delta = 10 * ntp_default_tick_delta; 450 else if (ntp_delta < -ntp_big_delta) 451 ntp_tick_delta = -10 * ntp_default_tick_delta; 452 else if (ntp_delta > 0) 453 ntp_tick_delta = ntp_default_tick_delta; 454 else 455 ntp_tick_delta = -ntp_default_tick_delta; 456 } 457 458 void 459 kern_adjtime(int64_t delta, int64_t *odelta) 460 { 461 int origcpu; 462 463 if ((origcpu = mycpu->gd_cpuid) != 0) 464 lwkt_setcpu_self(globaldata_find(0)); 465 466 crit_enter(); 467 *odelta = ntp_delta; 468 ntp_delta = delta; 469 kern_adjtime_common(); 470 crit_exit(); 471 472 if (origcpu != 0) 473 lwkt_setcpu_self(globaldata_find(origcpu)); 474 } 475 476 static void 477 kern_get_ntp_delta(int64_t *delta) 478 { 479 int origcpu; 480 481 if ((origcpu = mycpu->gd_cpuid) != 0) 482 lwkt_setcpu_self(globaldata_find(0)); 483 484 crit_enter(); 485 *delta = ntp_delta; 486 crit_exit(); 487 488 if (origcpu != 0) 489 lwkt_setcpu_self(globaldata_find(origcpu)); 490 } 491 492 void 493 kern_reladjtime(int64_t delta) 494 { 495 int origcpu; 496 497 if ((origcpu = mycpu->gd_cpuid) != 0) 498 lwkt_setcpu_self(globaldata_find(0)); 499 500 crit_enter(); 501 ntp_delta += delta; 502 kern_adjtime_common(); 503 crit_exit(); 504 505 if (origcpu != 0) 506 lwkt_setcpu_self(globaldata_find(origcpu)); 507 } 508 509 static void 510 kern_adjfreq(int64_t rate) 511 { 512 int origcpu; 513 514 if ((origcpu = mycpu->gd_cpuid) != 0) 515 lwkt_setcpu_self(globaldata_find(0)); 516 517 crit_enter(); 518 ntp_tick_permanent = rate; 519 crit_exit(); 520 521 if (origcpu != 0) 522 lwkt_setcpu_self(globaldata_find(origcpu)); 523 } 524 525 /* 526 * MPALMOSTSAFE 527 */ 528 int 529 sys_adjtime(struct adjtime_args *uap) 530 { 531 struct thread *td = curthread; 532 struct timeval atv; 533 int64_t ndelta, odelta; 534 int error; 535 536 if ((error = priv_check(td, PRIV_ADJTIME))) 537 return (error); 538 error = copyin(uap->delta, &atv, sizeof(struct timeval)); 539 if (error) 540 return (error); 541 542 /* 543 * Compute the total correction and the rate at which to apply it. 544 * Round the adjustment down to a whole multiple of the per-tick 545 * delta, so that after some number of incremental changes in 546 * hardclock(), tickdelta will become zero, lest the correction 547 * overshoot and start taking us away from the desired final time. 548 */ 549 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 550 get_mplock(); 551 kern_adjtime(ndelta, &odelta); 552 rel_mplock(); 553 554 if (uap->olddelta) { 555 atv.tv_sec = odelta / 1000000000; 556 atv.tv_usec = odelta % 1000000000 / 1000; 557 copyout(&atv, uap->olddelta, sizeof(struct timeval)); 558 } 559 return (0); 560 } 561 562 static int 563 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 564 { 565 int64_t delta; 566 int error; 567 568 if (req->newptr != NULL) { 569 if (priv_check(curthread, PRIV_ROOT)) 570 return (EPERM); 571 error = SYSCTL_IN(req, &delta, sizeof(delta)); 572 if (error) 573 return (error); 574 kern_reladjtime(delta); 575 } 576 577 if (req->oldptr) 578 kern_get_ntp_delta(&delta); 579 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 580 return (error); 581 } 582 583 /* 584 * delta is in nanoseconds. 585 */ 586 static int 587 sysctl_delta(SYSCTL_HANDLER_ARGS) 588 { 589 int64_t delta, old_delta; 590 int error; 591 592 if (req->newptr != NULL) { 593 if (priv_check(curthread, PRIV_ROOT)) 594 return (EPERM); 595 error = SYSCTL_IN(req, &delta, sizeof(delta)); 596 if (error) 597 return (error); 598 kern_adjtime(delta, &old_delta); 599 } 600 601 if (req->oldptr != NULL) 602 kern_get_ntp_delta(&old_delta); 603 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 604 return (error); 605 } 606 607 /* 608 * frequency is in nanoseconds per second shifted left 32. 609 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 610 */ 611 static int 612 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 613 { 614 int64_t freqdelta; 615 int error; 616 617 if (req->newptr != NULL) { 618 if (priv_check(curthread, PRIV_ROOT)) 619 return (EPERM); 620 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 621 if (error) 622 return (error); 623 624 freqdelta /= hz; 625 kern_adjfreq(freqdelta); 626 } 627 628 if (req->oldptr != NULL) 629 freqdelta = ntp_tick_permanent * hz; 630 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 631 if (error) 632 return (error); 633 634 return (0); 635 } 636 637 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 638 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 639 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 640 sysctl_adjfreq, "Q", "permanent correction per second"); 641 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 642 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 643 sysctl_delta, "Q", "one-time delta"); 644 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 645 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 646 "threshold for fast adjustment"); 647 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 648 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 649 "per-tick adjustment"); 650 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 651 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 652 "default per-tick adjustment"); 653 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 654 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 655 "next leap second"); 656 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 657 &ntp_leap_insert, 0, "insert or remove leap second"); 658 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 659 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 660 sysctl_adjtime, "Q", "relative adjust for delta"); 661 662 /* 663 * Get value of an interval timer. The process virtual and 664 * profiling virtual time timers are kept in the p_stats area, since 665 * they can be swapped out. These are kept internally in the 666 * way they are specified externally: in time until they expire. 667 * 668 * The real time interval timer is kept in the process table slot 669 * for the process, and its value (it_value) is kept as an 670 * absolute time rather than as a delta, so that it is easy to keep 671 * periodic real-time signals from drifting. 672 * 673 * Virtual time timers are processed in the hardclock() routine of 674 * kern_clock.c. The real time timer is processed by a timeout 675 * routine, called from the softclock() routine. Since a callout 676 * may be delayed in real time due to interrupt processing in the system, 677 * it is possible for the real time timeout routine (realitexpire, given below), 678 * to be delayed in real time past when it is supposed to occur. It 679 * does not suffice, therefore, to reload the real timer .it_value from the 680 * real time timers .it_interval. Rather, we compute the next time in 681 * absolute time the timer should go off. 682 * 683 * MPALMOSTSAFE 684 */ 685 int 686 sys_getitimer(struct getitimer_args *uap) 687 { 688 struct proc *p = curproc; 689 struct timeval ctv; 690 struct itimerval aitv; 691 692 if (uap->which > ITIMER_PROF) 693 return (EINVAL); 694 get_mplock(); 695 crit_enter(); 696 if (uap->which == ITIMER_REAL) { 697 /* 698 * Convert from absolute to relative time in .it_value 699 * part of real time timer. If time for real time timer 700 * has passed return 0, else return difference between 701 * current time and time for the timer to go off. 702 */ 703 aitv = p->p_realtimer; 704 if (timevalisset(&aitv.it_value)) { 705 getmicrouptime(&ctv); 706 if (timevalcmp(&aitv.it_value, &ctv, <)) 707 timevalclear(&aitv.it_value); 708 else 709 timevalsub(&aitv.it_value, &ctv); 710 } 711 } else { 712 aitv = p->p_timer[uap->which]; 713 } 714 crit_exit(); 715 rel_mplock(); 716 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 717 } 718 719 /* 720 * MPALMOSTSAFE 721 */ 722 int 723 sys_setitimer(struct setitimer_args *uap) 724 { 725 struct itimerval aitv; 726 struct timeval ctv; 727 struct itimerval *itvp; 728 struct proc *p = curproc; 729 int error; 730 731 if (uap->which > ITIMER_PROF) 732 return (EINVAL); 733 itvp = uap->itv; 734 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 735 sizeof(struct itimerval)))) 736 return (error); 737 if ((uap->itv = uap->oitv) && 738 (error = sys_getitimer((struct getitimer_args *)uap))) 739 return (error); 740 if (itvp == 0) 741 return (0); 742 if (itimerfix(&aitv.it_value)) 743 return (EINVAL); 744 if (!timevalisset(&aitv.it_value)) 745 timevalclear(&aitv.it_interval); 746 else if (itimerfix(&aitv.it_interval)) 747 return (EINVAL); 748 get_mplock(); 749 crit_enter(); 750 if (uap->which == ITIMER_REAL) { 751 if (timevalisset(&p->p_realtimer.it_value)) 752 callout_stop(&p->p_ithandle); 753 if (timevalisset(&aitv.it_value)) 754 callout_reset(&p->p_ithandle, 755 tvtohz_high(&aitv.it_value), realitexpire, p); 756 getmicrouptime(&ctv); 757 timevaladd(&aitv.it_value, &ctv); 758 p->p_realtimer = aitv; 759 } else { 760 p->p_timer[uap->which] = aitv; 761 } 762 crit_exit(); 763 rel_mplock(); 764 return (0); 765 } 766 767 /* 768 * Real interval timer expired: 769 * send process whose timer expired an alarm signal. 770 * If time is not set up to reload, then just return. 771 * Else compute next time timer should go off which is > current time. 772 * This is where delay in processing this timeout causes multiple 773 * SIGALRM calls to be compressed into one. 774 * tvtohz_high() always adds 1 to allow for the time until the next clock 775 * interrupt being strictly less than 1 clock tick, but we don't want 776 * that here since we want to appear to be in sync with the clock 777 * interrupt even when we're delayed. 778 */ 779 void 780 realitexpire(void *arg) 781 { 782 struct proc *p; 783 struct timeval ctv, ntv; 784 785 p = (struct proc *)arg; 786 ksignal(p, SIGALRM); 787 if (!timevalisset(&p->p_realtimer.it_interval)) { 788 timevalclear(&p->p_realtimer.it_value); 789 return; 790 } 791 for (;;) { 792 crit_enter(); 793 timevaladd(&p->p_realtimer.it_value, 794 &p->p_realtimer.it_interval); 795 getmicrouptime(&ctv); 796 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 797 ntv = p->p_realtimer.it_value; 798 timevalsub(&ntv, &ctv); 799 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 800 realitexpire, p); 801 crit_exit(); 802 return; 803 } 804 crit_exit(); 805 } 806 } 807 808 /* 809 * Check that a proposed value to load into the .it_value or 810 * .it_interval part of an interval timer is acceptable, and 811 * fix it to have at least minimal value (i.e. if it is less 812 * than the resolution of the clock, round it up.) 813 * 814 * MPSAFE 815 */ 816 int 817 itimerfix(struct timeval *tv) 818 { 819 820 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 || 821 tv->tv_usec < 0 || tv->tv_usec >= 1000000) 822 return (EINVAL); 823 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < ustick) 824 tv->tv_usec = ustick; 825 return (0); 826 } 827 828 /* 829 * Decrement an interval timer by a specified number 830 * of microseconds, which must be less than a second, 831 * i.e. < 1000000. If the timer expires, then reload 832 * it. In this case, carry over (usec - old value) to 833 * reduce the value reloaded into the timer so that 834 * the timer does not drift. This routine assumes 835 * that it is called in a context where the timers 836 * on which it is operating cannot change in value. 837 */ 838 int 839 itimerdecr(struct itimerval *itp, int usec) 840 { 841 842 if (itp->it_value.tv_usec < usec) { 843 if (itp->it_value.tv_sec == 0) { 844 /* expired, and already in next interval */ 845 usec -= itp->it_value.tv_usec; 846 goto expire; 847 } 848 itp->it_value.tv_usec += 1000000; 849 itp->it_value.tv_sec--; 850 } 851 itp->it_value.tv_usec -= usec; 852 usec = 0; 853 if (timevalisset(&itp->it_value)) 854 return (1); 855 /* expired, exactly at end of interval */ 856 expire: 857 if (timevalisset(&itp->it_interval)) { 858 itp->it_value = itp->it_interval; 859 itp->it_value.tv_usec -= usec; 860 if (itp->it_value.tv_usec < 0) { 861 itp->it_value.tv_usec += 1000000; 862 itp->it_value.tv_sec--; 863 } 864 } else 865 itp->it_value.tv_usec = 0; /* sec is already 0 */ 866 return (0); 867 } 868 869 /* 870 * Add and subtract routines for timevals. 871 * N.B.: subtract routine doesn't deal with 872 * results which are before the beginning, 873 * it just gets very confused in this case. 874 * Caveat emptor. 875 */ 876 void 877 timevaladd(struct timeval *t1, const struct timeval *t2) 878 { 879 880 t1->tv_sec += t2->tv_sec; 881 t1->tv_usec += t2->tv_usec; 882 timevalfix(t1); 883 } 884 885 void 886 timevalsub(struct timeval *t1, const struct timeval *t2) 887 { 888 889 t1->tv_sec -= t2->tv_sec; 890 t1->tv_usec -= t2->tv_usec; 891 timevalfix(t1); 892 } 893 894 static void 895 timevalfix(struct timeval *t1) 896 { 897 898 if (t1->tv_usec < 0) { 899 t1->tv_sec--; 900 t1->tv_usec += 1000000; 901 } 902 if (t1->tv_usec >= 1000000) { 903 t1->tv_sec++; 904 t1->tv_usec -= 1000000; 905 } 906 } 907 908 /* 909 * ratecheck(): simple time-based rate-limit checking. 910 */ 911 int 912 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 913 { 914 struct timeval tv, delta; 915 int rv = 0; 916 917 getmicrouptime(&tv); /* NB: 10ms precision */ 918 delta = tv; 919 timevalsub(&delta, lasttime); 920 921 /* 922 * check for 0,0 is so that the message will be seen at least once, 923 * even if interval is huge. 924 */ 925 if (timevalcmp(&delta, mininterval, >=) || 926 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 927 *lasttime = tv; 928 rv = 1; 929 } 930 931 return (rv); 932 } 933 934 /* 935 * ppsratecheck(): packets (or events) per second limitation. 936 * 937 * Return 0 if the limit is to be enforced (e.g. the caller 938 * should drop a packet because of the rate limitation). 939 * 940 * maxpps of 0 always causes zero to be returned. maxpps of -1 941 * always causes 1 to be returned; this effectively defeats rate 942 * limiting. 943 * 944 * Note that we maintain the struct timeval for compatibility 945 * with other bsd systems. We reuse the storage and just monitor 946 * clock ticks for minimal overhead. 947 */ 948 int 949 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 950 { 951 int now; 952 953 /* 954 * Reset the last time and counter if this is the first call 955 * or more than a second has passed since the last update of 956 * lasttime. 957 */ 958 now = ticks; 959 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 960 lasttime->tv_sec = now; 961 *curpps = 1; 962 return (maxpps != 0); 963 } else { 964 (*curpps)++; /* NB: ignore potential overflow */ 965 return (maxpps < 0 || *curpps < maxpps); 966 } 967 } 968 969