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 1.29 2005/06/01 17:43:42 dillon 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/time.h> 50 #include <sys/vnode.h> 51 #include <sys/sysctl.h> 52 #include <vm/vm.h> 53 #include <vm/vm_extern.h> 54 #include <sys/msgport2.h> 55 #include <sys/thread2.h> 56 57 struct timezone tz; 58 59 /* 60 * Time of day and interval timer support. 61 * 62 * These routines provide the kernel entry points to get and set 63 * the time-of-day and per-process interval timers. Subroutines 64 * here provide support for adding and subtracting timeval structures 65 * and decrementing interval timers, optionally reloading the interval 66 * timers when they expire. 67 */ 68 69 static int nanosleep1 (struct timespec *rqt, 70 struct timespec *rmt); 71 static int settime (struct timeval *); 72 static void timevalfix (struct timeval *); 73 static void no_lease_updatetime (int); 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 void 79 no_lease_updatetime(deltat) 80 int deltat; 81 { 82 } 83 84 void (*lease_updatetime) (int) = no_lease_updatetime; 85 86 static int 87 settime(tv) 88 struct timeval *tv; 89 { 90 struct timeval delta, tv1, tv2; 91 static struct timeval maxtime, laststep; 92 struct timespec ts; 93 int origcpu; 94 95 if ((origcpu = mycpu->gd_cpuid) != 0) 96 lwkt_setcpu_self(globaldata_find(0)); 97 98 crit_enter(); 99 microtime(&tv1); 100 delta = *tv; 101 timevalsub(&delta, &tv1); 102 103 /* 104 * If the system is secure, we do not allow the time to be 105 * set to a value earlier than 1 second less than the highest 106 * time we have yet seen. The worst a miscreant can do in 107 * this circumstance is "freeze" time. He couldn't go 108 * back to the past. 109 * 110 * We similarly do not allow the clock to be stepped more 111 * than one second, nor more than once per second. This allows 112 * a miscreant to make the clock march double-time, but no worse. 113 */ 114 if (securelevel > 1) { 115 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 116 /* 117 * Update maxtime to latest time we've seen. 118 */ 119 if (tv1.tv_sec > maxtime.tv_sec) 120 maxtime = tv1; 121 tv2 = *tv; 122 timevalsub(&tv2, &maxtime); 123 if (tv2.tv_sec < -1) { 124 tv->tv_sec = maxtime.tv_sec - 1; 125 printf("Time adjustment clamped to -1 second\n"); 126 } 127 } else { 128 if (tv1.tv_sec == laststep.tv_sec) { 129 crit_exit(); 130 return (EPERM); 131 } 132 if (delta.tv_sec > 1) { 133 tv->tv_sec = tv1.tv_sec + 1; 134 printf("Time adjustment clamped to +1 second\n"); 135 } 136 laststep = *tv; 137 } 138 } 139 140 ts.tv_sec = tv->tv_sec; 141 ts.tv_nsec = tv->tv_usec * 1000; 142 set_timeofday(&ts); 143 lease_updatetime(delta.tv_sec); 144 crit_exit(); 145 146 if (origcpu != 0) 147 lwkt_setcpu_self(globaldata_find(origcpu)); 148 149 resettodr(); 150 return (0); 151 } 152 153 /* ARGSUSED */ 154 int 155 clock_gettime(struct clock_gettime_args *uap) 156 { 157 struct timespec ats; 158 159 switch(uap->clock_id) { 160 case CLOCK_REALTIME: 161 nanotime(&ats); 162 return (copyout(&ats, uap->tp, sizeof(ats))); 163 case CLOCK_MONOTONIC: 164 nanouptime(&ats); 165 return (copyout(&ats, uap->tp, sizeof(ats))); 166 default: 167 return (EINVAL); 168 } 169 } 170 171 /* ARGSUSED */ 172 int 173 clock_settime(struct clock_settime_args *uap) 174 { 175 struct thread *td = curthread; 176 struct timeval atv; 177 struct timespec ats; 178 int error; 179 180 if ((error = suser(td)) != 0) 181 return (error); 182 switch(uap->clock_id) { 183 case CLOCK_REALTIME: 184 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 185 return (error); 186 if (ats.tv_nsec < 0 || ats.tv_nsec >= 1000000000) 187 return (EINVAL); 188 /* XXX Don't convert nsec->usec and back */ 189 TIMESPEC_TO_TIMEVAL(&atv, &ats); 190 error = settime(&atv); 191 return (error); 192 default: 193 return (EINVAL); 194 } 195 } 196 197 int 198 clock_getres(struct clock_getres_args *uap) 199 { 200 struct timespec ts; 201 202 switch(uap->clock_id) { 203 case CLOCK_REALTIME: 204 case CLOCK_MONOTONIC: 205 /* 206 * Round up the result of the division cheaply 207 * by adding 1. Rounding up is especially important 208 * if rounding down would give 0. Perfect rounding 209 * is unimportant. 210 */ 211 ts.tv_sec = 0; 212 ts.tv_nsec = 1000000000 / sys_cputimer->freq + 1; 213 return(copyout(&ts, uap->tp, sizeof(ts))); 214 default: 215 return(EINVAL); 216 } 217 } 218 219 /* 220 * nanosleep1() 221 * 222 * This is a general helper function for nanosleep() (aka sleep() aka 223 * usleep()). 224 * 225 * If there is less then one tick's worth of time left and 226 * we haven't done a yield, or the remaining microseconds is 227 * ridiculously low, do a yield. This avoids having 228 * to deal with systimer overheads when the system is under 229 * heavy loads. If we have done a yield already then use 230 * a systimer and an uninterruptable thread wait. 231 * 232 * If there is more then a tick's worth of time left, 233 * calculate the baseline ticks and use an interruptable 234 * tsleep, then handle the fine-grained delay on the next 235 * loop. This usually results in two sleeps occuring, a long one 236 * and a short one. 237 */ 238 static void 239 ns1_systimer(systimer_t info) 240 { 241 lwkt_schedule(info->data); 242 } 243 244 static int 245 nanosleep1(struct timespec *rqt, struct timespec *rmt) 246 { 247 static int nanowait; 248 struct timespec ts, ts2, ts3; 249 struct timeval tv; 250 int error; 251 int tried_yield; 252 253 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 254 return (EINVAL); 255 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 256 return (0); 257 nanouptime(&ts); 258 timespecadd(&ts, rqt); /* ts = target timestamp compare */ 259 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */ 260 tried_yield = 0; 261 262 for (;;) { 263 int ticks; 264 struct systimer info; 265 266 ticks = tv.tv_usec / tick; /* approximate */ 267 268 if (tv.tv_sec == 0 && ticks == 0) { 269 thread_t td = curthread; 270 if (tried_yield || tv.tv_usec < sleep_hard_us) { 271 tried_yield = 0; 272 uio_yield(); 273 } else { 274 crit_enter_quick(td); 275 systimer_init_oneshot(&info, ns1_systimer, 276 td, tv.tv_usec); 277 lwkt_deschedule_self(td); 278 crit_exit_quick(td); 279 lwkt_switch(); 280 systimer_del(&info); /* make sure it's gone */ 281 } 282 error = iscaught(td->td_proc); 283 } else if (tv.tv_sec == 0) { 284 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 285 } else { 286 ticks = tvtohz_low(&tv); /* also handles overflow */ 287 error = tsleep(&nanowait, PCATCH, "nanslp", ticks); 288 } 289 nanouptime(&ts2); 290 if (error && error != EWOULDBLOCK) { 291 if (error == ERESTART) 292 error = EINTR; 293 if (rmt != NULL) { 294 timespecsub(&ts, &ts2); 295 if (ts.tv_sec < 0) 296 timespecclear(&ts); 297 *rmt = ts; 298 } 299 return (error); 300 } 301 if (timespeccmp(&ts2, &ts, >=)) 302 return (0); 303 ts3 = ts; 304 timespecsub(&ts3, &ts2); 305 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 306 } 307 } 308 309 static void nanosleep_done(void *arg); 310 static void nanosleep_copyout(union sysunion *sysun); 311 312 /* ARGSUSED */ 313 int 314 nanosleep(struct nanosleep_args *uap) 315 { 316 int error; 317 struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep; 318 319 error = copyin(uap->rqtp, &smsleep->rqt, sizeof(smsleep->rqt)); 320 if (error) 321 return (error); 322 /* 323 * YYY clean this up to always use the callout, note that an abort 324 * implementation should record the residual in the async case. 325 */ 326 if (uap->sysmsg.lmsg.ms_flags & MSGF_ASYNC) { 327 quad_t ticks; 328 329 ticks = (quad_t)smsleep->rqt.tv_nsec * hz / 1000000000LL; 330 if (smsleep->rqt.tv_sec) 331 ticks += (quad_t)smsleep->rqt.tv_sec * hz; 332 if (ticks <= 0) { 333 if (ticks == 0) 334 error = 0; 335 else 336 error = EINVAL; 337 } else { 338 uap->sysmsg.copyout = nanosleep_copyout; 339 uap->sysmsg.lmsg.ms_flags &= ~MSGF_DONE; 340 callout_init(&smsleep->timer); 341 callout_reset(&smsleep->timer, ticks, nanosleep_done, uap); 342 error = EASYNC; 343 } 344 } else { 345 /* 346 * Old synchronous sleep code, copyout the residual if 347 * nanosleep was interrupted. 348 */ 349 error = nanosleep1(&smsleep->rqt, &smsleep->rmt); 350 if (error && uap->rmtp) 351 error = copyout(&smsleep->rmt, uap->rmtp, sizeof(smsleep->rmt)); 352 } 353 return (error); 354 } 355 356 /* 357 * Asynch completion for the nanosleep() syscall. This function may be 358 * called from any context and cannot legally access the originating 359 * thread, proc, or its user space. 360 * 361 * YYY change the callout interface API so we can simply assign the replymsg 362 * function to it directly. 363 */ 364 static void 365 nanosleep_done(void *arg) 366 { 367 struct nanosleep_args *uap = arg; 368 lwkt_msg_t msg = &uap->sysmsg.lmsg; 369 370 lwkt_replymsg(msg, 0); 371 } 372 373 /* 374 * Asynch return for the nanosleep() syscall, called in the context of the 375 * originating thread when it pulls the message off the reply port. This 376 * function is responsible for any copyouts to userland. Kernel threads 377 * which do their own internal system calls will not usually call the return 378 * function. 379 */ 380 static void 381 nanosleep_copyout(union sysunion *sysun) 382 { 383 struct nanosleep_args *uap = &sysun->nanosleep; 384 struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep; 385 386 if (sysun->lmsg.ms_error && uap->rmtp) { 387 sysun->lmsg.ms_error = 388 copyout(&smsleep->rmt, uap->rmtp, sizeof(smsleep->rmt)); 389 } 390 } 391 392 /* ARGSUSED */ 393 int 394 gettimeofday(struct gettimeofday_args *uap) 395 { 396 struct timeval atv; 397 int error = 0; 398 399 if (uap->tp) { 400 microtime(&atv); 401 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp, 402 sizeof (atv)))) 403 return (error); 404 } 405 if (uap->tzp) 406 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp, 407 sizeof (tz)); 408 return (error); 409 } 410 411 /* ARGSUSED */ 412 int 413 settimeofday(struct settimeofday_args *uap) 414 { 415 struct thread *td = curthread; 416 struct timeval atv; 417 struct timezone atz; 418 int error; 419 420 if ((error = suser(td))) 421 return (error); 422 /* Verify all parameters before changing time. */ 423 if (uap->tv) { 424 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv, 425 sizeof(atv)))) 426 return (error); 427 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000) 428 return (EINVAL); 429 } 430 if (uap->tzp && 431 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz)))) 432 return (error); 433 if (uap->tv && (error = settime(&atv))) 434 return (error); 435 if (uap->tzp) 436 tz = atz; 437 return (0); 438 } 439 440 static void 441 kern_adjtime_common(void) 442 { 443 if ((ntp_delta >= 0 && ntp_delta < ntp_default_tick_delta) || 444 (ntp_delta < 0 && ntp_delta > ntp_default_tick_delta)) 445 ntp_tick_delta = ntp_delta; 446 else if (ntp_delta > ntp_big_delta) 447 ntp_tick_delta = 10 * ntp_default_tick_delta; 448 else if (ntp_delta < -ntp_big_delta) 449 ntp_tick_delta = -10 * ntp_default_tick_delta; 450 else if (ntp_delta > 0) 451 ntp_tick_delta = ntp_default_tick_delta; 452 else 453 ntp_tick_delta = -ntp_default_tick_delta; 454 } 455 456 void 457 kern_adjtime(int64_t delta, int64_t *odelta) 458 { 459 int origcpu; 460 461 if ((origcpu = mycpu->gd_cpuid) != 0) 462 lwkt_setcpu_self(globaldata_find(0)); 463 464 crit_enter(); 465 *odelta = ntp_delta; 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_get_ntp_delta(int64_t *delta) 476 { 477 int origcpu; 478 479 if ((origcpu = mycpu->gd_cpuid) != 0) 480 lwkt_setcpu_self(globaldata_find(0)); 481 482 crit_enter(); 483 *delta = ntp_delta; 484 crit_exit(); 485 486 if (origcpu != 0) 487 lwkt_setcpu_self(globaldata_find(origcpu)); 488 } 489 490 void 491 kern_reladjtime(int64_t delta) 492 { 493 int origcpu; 494 495 if ((origcpu = mycpu->gd_cpuid) != 0) 496 lwkt_setcpu_self(globaldata_find(0)); 497 498 crit_enter(); 499 ntp_delta += delta; 500 kern_adjtime_common(); 501 crit_exit(); 502 503 if (origcpu != 0) 504 lwkt_setcpu_self(globaldata_find(origcpu)); 505 } 506 507 static void 508 kern_adjfreq(int64_t rate) 509 { 510 int origcpu; 511 512 if ((origcpu = mycpu->gd_cpuid) != 0) 513 lwkt_setcpu_self(globaldata_find(0)); 514 515 crit_enter(); 516 ntp_tick_permanent = rate; 517 crit_exit(); 518 519 if (origcpu != 0) 520 lwkt_setcpu_self(globaldata_find(origcpu)); 521 } 522 523 /* ARGSUSED */ 524 int 525 adjtime(struct adjtime_args *uap) 526 { 527 struct thread *td = curthread; 528 struct timeval atv; 529 int64_t ndelta, odelta; 530 int error; 531 532 if ((error = suser(td))) 533 return (error); 534 if ((error = 535 copyin((caddr_t)uap->delta, (caddr_t)&atv, sizeof(struct timeval)))) 536 return (error); 537 538 /* 539 * Compute the total correction and the rate at which to apply it. 540 * Round the adjustment down to a whole multiple of the per-tick 541 * delta, so that after some number of incremental changes in 542 * hardclock(), tickdelta will become zero, lest the correction 543 * overshoot and start taking us away from the desired final time. 544 */ 545 ndelta = (int64_t)atv.tv_sec * 1000000000 + atv.tv_usec * 1000; 546 kern_adjtime(ndelta, &odelta); 547 548 if (uap->olddelta) { 549 atv.tv_sec = odelta / 1000000000; 550 atv.tv_usec = odelta % 1000000 / 1000; 551 (void) copyout((caddr_t)&atv, (caddr_t)uap->olddelta, 552 sizeof(struct timeval)); 553 } 554 return (0); 555 } 556 557 static int 558 sysctl_adjtime(SYSCTL_HANDLER_ARGS) 559 { 560 int64_t delta; 561 int error; 562 563 if (req->newptr != NULL) { 564 if (suser(curthread)) 565 return (EPERM); 566 error = SYSCTL_IN(req, &delta, sizeof(delta)); 567 if (error) 568 return (error); 569 kern_reladjtime(delta); 570 } 571 572 if (req->oldptr) 573 kern_get_ntp_delta(&delta); 574 error = SYSCTL_OUT(req, &delta, sizeof(delta)); 575 return (error); 576 } 577 578 /* 579 * delta is in nanoseconds. 580 */ 581 static int 582 sysctl_delta(SYSCTL_HANDLER_ARGS) 583 { 584 int64_t delta, old_delta; 585 int error; 586 587 if (req->newptr != NULL) { 588 if (suser(curthread)) 589 return (EPERM); 590 error = SYSCTL_IN(req, &delta, sizeof(delta)); 591 if (error) 592 return (error); 593 kern_adjtime(delta, &old_delta); 594 } 595 596 if (req->oldptr != NULL) 597 kern_get_ntp_delta(&old_delta); 598 error = SYSCTL_OUT(req, &old_delta, sizeof(old_delta)); 599 return (error); 600 } 601 602 /* 603 * frequency is in nanoseconds per second shifted left 32. 604 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32. 605 */ 606 static int 607 sysctl_adjfreq(SYSCTL_HANDLER_ARGS) 608 { 609 int64_t freqdelta; 610 int error; 611 612 if (req->newptr != NULL) { 613 if (suser(curthread)) 614 return (EPERM); 615 error = SYSCTL_IN(req, &freqdelta, sizeof(freqdelta)); 616 if (error) 617 return (error); 618 619 freqdelta /= hz; 620 kern_adjfreq(freqdelta); 621 } 622 623 if (req->oldptr != NULL) 624 freqdelta = ntp_tick_permanent * hz; 625 error = SYSCTL_OUT(req, &freqdelta, sizeof(freqdelta)); 626 if (error) 627 return (error); 628 629 return (0); 630 } 631 632 SYSCTL_NODE(_kern, OID_AUTO, ntp, CTLFLAG_RW, 0, "NTP related controls"); 633 SYSCTL_PROC(_kern_ntp, OID_AUTO, permanent, 634 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 635 sysctl_adjfreq, "Q", "permanent correction per second"); 636 SYSCTL_PROC(_kern_ntp, OID_AUTO, delta, 637 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 638 sysctl_delta, "Q", "one-time delta"); 639 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, big_delta, CTLFLAG_RD, 640 &ntp_big_delta, sizeof(ntp_big_delta), "Q", 641 "threshold for fast adjustment"); 642 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, tick_delta, CTLFLAG_RD, 643 &ntp_tick_delta, sizeof(ntp_tick_delta), "LU", 644 "per-tick adjustment"); 645 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, default_tick_delta, CTLFLAG_RD, 646 &ntp_default_tick_delta, sizeof(ntp_default_tick_delta), "LU", 647 "default per-tick adjustment"); 648 SYSCTL_OPAQUE(_kern_ntp, OID_AUTO, next_leap_second, CTLFLAG_RW, 649 &ntp_leap_second, sizeof(ntp_leap_second), "LU", 650 "next leap second"); 651 SYSCTL_INT(_kern_ntp, OID_AUTO, insert_leap_second, CTLFLAG_RW, 652 &ntp_leap_insert, 0, "insert or remove leap second"); 653 SYSCTL_PROC(_kern_ntp, OID_AUTO, adjust, 654 CTLTYPE_QUAD|CTLFLAG_RW, 0, 0, 655 sysctl_adjtime, "Q", "relative adjust for delta"); 656 657 /* 658 * Get value of an interval timer. The process virtual and 659 * profiling virtual time timers are kept in the p_stats area, since 660 * they can be swapped out. These are kept internally in the 661 * way they are specified externally: in time until they expire. 662 * 663 * The real time interval timer is kept in the process table slot 664 * for the process, and its value (it_value) is kept as an 665 * absolute time rather than as a delta, so that it is easy to keep 666 * periodic real-time signals from drifting. 667 * 668 * Virtual time timers are processed in the hardclock() routine of 669 * kern_clock.c. The real time timer is processed by a timeout 670 * routine, called from the softclock() routine. Since a callout 671 * may be delayed in real time due to interrupt processing in the system, 672 * it is possible for the real time timeout routine (realitexpire, given below), 673 * to be delayed in real time past when it is supposed to occur. It 674 * does not suffice, therefore, to reload the real timer .it_value from the 675 * real time timers .it_interval. Rather, we compute the next time in 676 * absolute time the timer should go off. 677 */ 678 /* ARGSUSED */ 679 int 680 getitimer(struct getitimer_args *uap) 681 { 682 struct proc *p = curproc; 683 struct timeval ctv; 684 struct itimerval aitv; 685 686 if (uap->which > ITIMER_PROF) 687 return (EINVAL); 688 crit_enter(); 689 if (uap->which == ITIMER_REAL) { 690 /* 691 * Convert from absolute to relative time in .it_value 692 * part of real time timer. If time for real time timer 693 * has passed return 0, else return difference between 694 * current time and time for the timer to go off. 695 */ 696 aitv = p->p_realtimer; 697 if (timevalisset(&aitv.it_value)) { 698 getmicrouptime(&ctv); 699 if (timevalcmp(&aitv.it_value, &ctv, <)) 700 timevalclear(&aitv.it_value); 701 else 702 timevalsub(&aitv.it_value, &ctv); 703 } 704 } else { 705 aitv = p->p_stats->p_timer[uap->which]; 706 } 707 crit_exit(); 708 return (copyout((caddr_t)&aitv, (caddr_t)uap->itv, 709 sizeof (struct itimerval))); 710 } 711 712 /* ARGSUSED */ 713 int 714 setitimer(struct setitimer_args *uap) 715 { 716 struct itimerval aitv; 717 struct timeval ctv; 718 struct itimerval *itvp; 719 struct proc *p = curproc; 720 int error; 721 722 if (uap->which > ITIMER_PROF) 723 return (EINVAL); 724 itvp = uap->itv; 725 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv, 726 sizeof(struct itimerval)))) 727 return (error); 728 if ((uap->itv = uap->oitv) && 729 (error = getitimer((struct getitimer_args *)uap))) 730 return (error); 731 if (itvp == 0) 732 return (0); 733 if (itimerfix(&aitv.it_value)) 734 return (EINVAL); 735 if (!timevalisset(&aitv.it_value)) 736 timevalclear(&aitv.it_interval); 737 else if (itimerfix(&aitv.it_interval)) 738 return (EINVAL); 739 crit_enter(); 740 if (uap->which == ITIMER_REAL) { 741 if (timevalisset(&p->p_realtimer.it_value)) 742 callout_stop(&p->p_ithandle); 743 if (timevalisset(&aitv.it_value)) 744 callout_reset(&p->p_ithandle, 745 tvtohz_high(&aitv.it_value), realitexpire, p); 746 getmicrouptime(&ctv); 747 timevaladd(&aitv.it_value, &ctv); 748 p->p_realtimer = aitv; 749 } else { 750 p->p_stats->p_timer[uap->which] = aitv; 751 } 752 crit_exit(); 753 return (0); 754 } 755 756 /* 757 * Real interval timer expired: 758 * send process whose timer expired an alarm signal. 759 * If time is not set up to reload, then just return. 760 * Else compute next time timer should go off which is > current time. 761 * This is where delay in processing this timeout causes multiple 762 * SIGALRM calls to be compressed into one. 763 * tvtohz_high() always adds 1 to allow for the time until the next clock 764 * interrupt being strictly less than 1 clock tick, but we don't want 765 * that here since we want to appear to be in sync with the clock 766 * interrupt even when we're delayed. 767 */ 768 void 769 realitexpire(arg) 770 void *arg; 771 { 772 struct proc *p; 773 struct timeval ctv, ntv; 774 775 p = (struct proc *)arg; 776 psignal(p, SIGALRM); 777 if (!timevalisset(&p->p_realtimer.it_interval)) { 778 timevalclear(&p->p_realtimer.it_value); 779 return; 780 } 781 for (;;) { 782 crit_enter(); 783 timevaladd(&p->p_realtimer.it_value, 784 &p->p_realtimer.it_interval); 785 getmicrouptime(&ctv); 786 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) { 787 ntv = p->p_realtimer.it_value; 788 timevalsub(&ntv, &ctv); 789 callout_reset(&p->p_ithandle, tvtohz_low(&ntv), 790 realitexpire, p); 791 crit_exit(); 792 return; 793 } 794 crit_exit(); 795 } 796 } 797 798 /* 799 * Check that a proposed value to load into the .it_value or 800 * .it_interval part of an interval timer is acceptable, and 801 * fix it to have at least minimal value (i.e. if it is less 802 * than the resolution of the clock, round it up.) 803 */ 804 int 805 itimerfix(tv) 806 struct timeval *tv; 807 { 808 809 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 || 810 tv->tv_usec < 0 || tv->tv_usec >= 1000000) 811 return (EINVAL); 812 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick) 813 tv->tv_usec = tick; 814 return (0); 815 } 816 817 /* 818 * Decrement an interval timer by a specified number 819 * of microseconds, which must be less than a second, 820 * i.e. < 1000000. If the timer expires, then reload 821 * it. In this case, carry over (usec - old value) to 822 * reduce the value reloaded into the timer so that 823 * the timer does not drift. This routine assumes 824 * that it is called in a context where the timers 825 * on which it is operating cannot change in value. 826 */ 827 int 828 itimerdecr(itp, usec) 829 struct itimerval *itp; 830 int usec; 831 { 832 833 if (itp->it_value.tv_usec < usec) { 834 if (itp->it_value.tv_sec == 0) { 835 /* expired, and already in next interval */ 836 usec -= itp->it_value.tv_usec; 837 goto expire; 838 } 839 itp->it_value.tv_usec += 1000000; 840 itp->it_value.tv_sec--; 841 } 842 itp->it_value.tv_usec -= usec; 843 usec = 0; 844 if (timevalisset(&itp->it_value)) 845 return (1); 846 /* expired, exactly at end of interval */ 847 expire: 848 if (timevalisset(&itp->it_interval)) { 849 itp->it_value = itp->it_interval; 850 itp->it_value.tv_usec -= usec; 851 if (itp->it_value.tv_usec < 0) { 852 itp->it_value.tv_usec += 1000000; 853 itp->it_value.tv_sec--; 854 } 855 } else 856 itp->it_value.tv_usec = 0; /* sec is already 0 */ 857 return (0); 858 } 859 860 /* 861 * Add and subtract routines for timevals. 862 * N.B.: subtract routine doesn't deal with 863 * results which are before the beginning, 864 * it just gets very confused in this case. 865 * Caveat emptor. 866 */ 867 void 868 timevaladd(t1, t2) 869 struct timeval *t1, *t2; 870 { 871 872 t1->tv_sec += t2->tv_sec; 873 t1->tv_usec += t2->tv_usec; 874 timevalfix(t1); 875 } 876 877 void 878 timevalsub(t1, t2) 879 struct timeval *t1, *t2; 880 { 881 882 t1->tv_sec -= t2->tv_sec; 883 t1->tv_usec -= t2->tv_usec; 884 timevalfix(t1); 885 } 886 887 static void 888 timevalfix(t1) 889 struct timeval *t1; 890 { 891 892 if (t1->tv_usec < 0) { 893 t1->tv_sec--; 894 t1->tv_usec += 1000000; 895 } 896 if (t1->tv_usec >= 1000000) { 897 t1->tv_sec++; 898 t1->tv_usec -= 1000000; 899 } 900 } 901 902 /* 903 * ratecheck(): simple time-based rate-limit checking. 904 */ 905 int 906 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 907 { 908 struct timeval tv, delta; 909 int rv = 0; 910 911 getmicrouptime(&tv); /* NB: 10ms precision */ 912 delta = tv; 913 timevalsub(&delta, lasttime); 914 915 /* 916 * check for 0,0 is so that the message will be seen at least once, 917 * even if interval is huge. 918 */ 919 if (timevalcmp(&delta, mininterval, >=) || 920 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 921 *lasttime = tv; 922 rv = 1; 923 } 924 925 return (rv); 926 } 927 928 /* 929 * ppsratecheck(): packets (or events) per second limitation. 930 * 931 * Return 0 if the limit is to be enforced (e.g. the caller 932 * should drop a packet because of the rate limitation). 933 * 934 * maxpps of 0 always causes zero to be returned. maxpps of -1 935 * always causes 1 to be returned; this effectively defeats rate 936 * limiting. 937 * 938 * Note that we maintain the struct timeval for compatibility 939 * with other bsd systems. We reuse the storage and just monitor 940 * clock ticks for minimal overhead. 941 */ 942 int 943 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 944 { 945 int now; 946 947 /* 948 * Reset the last time and counter if this is the first call 949 * or more than a second has passed since the last update of 950 * lasttime. 951 */ 952 now = ticks; 953 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 954 lasttime->tv_sec = now; 955 *curpps = 1; 956 return (maxpps != 0); 957 } else { 958 (*curpps)++; /* NB: ignore potential overflow */ 959 return (maxpps < 0 || *curpps < maxpps); 960 } 961 } 962 963