kern_time.c revision 1.232
1 /* $NetBSD: kern_time.c,v 1.232 2026/03/15 12:00:58 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 2000, 2004, 2005, 2007, 2008, 2009, 2020 5 * The NetBSD Foundation, Inc. 6 * All rights reserved. 7 * 8 * This code is derived from software contributed to The NetBSD Foundation 9 * by Christopher G. Demetriou, by Andrew Doran, and by Jason R. Thorpe. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1989, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)kern_time.c 8.4 (Berkeley) 5/26/95 62 */ 63 64 #include <sys/cdefs.h> 65 __KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.232 2026/03/15 12:00:58 yamt Exp $"); 66 67 #include <sys/param.h> 68 #include <sys/types.h> 69 70 #include <sys/callout.h> 71 #include <sys/cpu.h> 72 #include <sys/errno.h> 73 #include <sys/intr.h> 74 #include <sys/kauth.h> 75 #include <sys/kernel.h> 76 #include <sys/kmem.h> 77 #include <sys/lwp.h> 78 #include <sys/mount.h> 79 #include <sys/mutex.h> 80 #include <sys/proc.h> 81 #include <sys/queue.h> 82 #include <sys/resourcevar.h> 83 #include <sys/sdt.h> 84 #include <sys/signal.h> 85 #include <sys/signalvar.h> 86 #include <sys/syscallargs.h> 87 #include <sys/syslog.h> 88 #include <sys/systm.h> 89 #include <sys/timetc.h> 90 #include <sys/timevar.h> 91 #include <sys/timex.h> 92 #include <sys/vnode.h> 93 94 #include <machine/limits.h> 95 96 kmutex_t itimer_mutex __cacheline_aligned; /* XXX static */ 97 static struct itlist itimer_realtime_changed_notify; 98 99 static void itimer_callout(void *); 100 static void ptimer_intr(void *); 101 static void *ptimer_sih __read_mostly; 102 static TAILQ_HEAD(, ptimer) ptimer_queue; 103 104 #define CLOCK_VIRTUAL_P(clockid) \ 105 ((clockid) == CLOCK_VIRTUAL || (clockid) == CLOCK_PROF) 106 107 #define IS_ITIMER(id) (0 <= (id) && (id) < TIMER_MIN) 108 #define IS_POSIX_TIMER(id) (TIMER_MIN <= (id) && (id) < TIMER_MAX) 109 110 CTASSERT(ITIMER_REAL == CLOCK_REALTIME); 111 CTASSERT(ITIMER_VIRTUAL == CLOCK_VIRTUAL); 112 CTASSERT(ITIMER_PROF == CLOCK_PROF); 113 CTASSERT(ITIMER_MONOTONIC == CLOCK_MONOTONIC); 114 115 CTASSERT(IS_ITIMER(ITIMER_REAL)); 116 CTASSERT(IS_ITIMER(ITIMER_VIRTUAL)); 117 CTASSERT(IS_ITIMER(ITIMER_PROF)); 118 CTASSERT(IS_ITIMER(ITIMER_MONOTONIC)); 119 120 CTASSERT(!IS_POSIX_TIMER(ITIMER_REAL)); 121 CTASSERT(!IS_POSIX_TIMER(ITIMER_VIRTUAL)); 122 CTASSERT(!IS_POSIX_TIMER(ITIMER_PROF)); 123 CTASSERT(!IS_POSIX_TIMER(ITIMER_MONOTONIC)); 124 125 /* 126 * Initialize timekeeping. 127 */ 128 void 129 time_init(void) 130 { 131 132 mutex_init(&itimer_mutex, MUTEX_DEFAULT, IPL_SCHED); 133 LIST_INIT(&itimer_realtime_changed_notify); 134 135 TAILQ_INIT(&ptimer_queue); 136 ptimer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE, 137 ptimer_intr, NULL); 138 } 139 140 /* 141 * Check if the time will wrap if set to ts. 142 * 143 * ts - timespec describing the new time 144 * delta - the delta between the current time and ts 145 */ 146 bool 147 time_wraps(struct timespec *ts, struct timespec *delta) 148 { 149 150 /* 151 * Don't allow the time to be set forward so far it 152 * will wrap and become negative, thus allowing an 153 * attacker to bypass the next check below. The 154 * cutoff is 1 year before rollover occurs, so even 155 * if the attacker uses adjtime(2) to move the time 156 * past the cutoff, it will take a very long time 157 * to get to the wrap point. 158 */ 159 if ((ts->tv_sec > LLONG_MAX - 365*24*60*60) || 160 (delta->tv_sec < 0 || delta->tv_nsec < 0)) 161 return true; 162 163 return false; 164 } 165 166 /* 167 * itimer_lock: 168 * 169 * Acquire the interval timer data lock. 170 */ 171 void 172 itimer_lock(void) 173 { 174 mutex_spin_enter(&itimer_mutex); 175 } 176 177 /* 178 * itimer_unlock: 179 * 180 * Release the interval timer data lock. 181 */ 182 void 183 itimer_unlock(void) 184 { 185 mutex_spin_exit(&itimer_mutex); 186 } 187 188 /* 189 * itimer_lock_held: 190 * 191 * Check that the interval timer lock is held for diagnostic 192 * assertions. 193 */ 194 inline bool __diagused 195 itimer_lock_held(void) 196 { 197 return mutex_owned(&itimer_mutex); 198 } 199 200 /* 201 * Time of day and interval timer support. 202 * 203 * These routines provide the kernel entry points to get and set 204 * the time-of-day and per-process interval timers. Subroutines 205 * here provide support for adding and subtracting timeval structures 206 * and decrementing interval timers, optionally reloading the interval 207 * timers when they expire. 208 */ 209 210 /* This function is used by clock_settime and settimeofday */ 211 static int 212 settime1(struct proc *p, const struct timespec *ts, bool check_kauth) 213 { 214 struct timespec delta, now; 215 216 /* 217 * The time being set to an unreasonable value will cause 218 * unreasonable system behaviour. 219 */ 220 if (ts->tv_sec < 0 || ts->tv_sec > (1LL << 36)) 221 return SET_ERROR(EINVAL); 222 223 nanotime(&now); 224 timespecsub(ts, &now, &delta); 225 226 if (check_kauth && kauth_authorize_system(kauth_cred_get(), 227 KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, __UNCONST(ts), 228 &delta, KAUTH_ARG(check_kauth ? false : true)) != 0) { 229 return SET_ERROR(EPERM); 230 } 231 232 #ifdef notyet 233 if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */ 234 return SET_ERROR(EPERM); 235 } 236 #endif 237 238 tc_setclock(ts); 239 240 resettodr(); 241 242 /* 243 * Notify pending CLOCK_REALTIME timers about the real time change. 244 * There may be inactive timers on this list, but this happens 245 * comparatively less often than timers firing, and so it's better 246 * to put the extra checks here than to complicate the other code 247 * path. 248 */ 249 struct itimer *it; 250 itimer_lock(); 251 LIST_FOREACH(it, &itimer_realtime_changed_notify, it_rtchgq) { 252 KASSERT(it->it_ops->ito_realtime_changed != NULL); 253 if (timespecisset(&it->it_time.it_value)) { 254 (*it->it_ops->ito_realtime_changed)(it); 255 } 256 } 257 itimer_unlock(); 258 259 return 0; 260 } 261 262 int 263 settime(struct proc *p, struct timespec *ts) 264 { 265 return settime1(p, ts, true); 266 } 267 268 /* ARGSUSED */ 269 int 270 sys___clock_gettime50(struct lwp *l, 271 const struct sys___clock_gettime50_args *uap, register_t *retval) 272 { 273 /* { 274 syscallarg(clockid_t) clock_id; 275 syscallarg(struct timespec *) tp; 276 } */ 277 int error; 278 struct timespec ats; 279 280 error = clock_gettime1(SCARG(uap, clock_id), &ats); 281 if (error != 0) 282 return error; 283 284 return copyout(&ats, SCARG(uap, tp), sizeof(ats)); 285 } 286 287 /* ARGSUSED */ 288 int 289 sys___clock_settime50(struct lwp *l, 290 const struct sys___clock_settime50_args *uap, register_t *retval) 291 { 292 /* { 293 syscallarg(clockid_t) clock_id; 294 syscallarg(const struct timespec *) tp; 295 } */ 296 int error; 297 struct timespec ats; 298 299 if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0) 300 return error; 301 302 return clock_settime1(l->l_proc, SCARG(uap, clock_id), &ats, true); 303 } 304 305 306 int 307 clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp, 308 bool check_kauth) 309 { 310 int error; 311 312 if (tp->tv_nsec < 0 || tp->tv_nsec >= 1000000000L) 313 return SET_ERROR(EINVAL); 314 315 switch (clock_id) { 316 case CLOCK_REALTIME: 317 if ((error = settime1(p, tp, check_kauth)) != 0) 318 return error; 319 break; 320 case CLOCK_MONOTONIC: 321 return SET_ERROR(EINVAL); /* read-only clock */ 322 default: 323 return SET_ERROR(EINVAL); 324 } 325 326 return 0; 327 } 328 329 int 330 sys___clock_getres50(struct lwp *l, const struct sys___clock_getres50_args *uap, 331 register_t *retval) 332 { 333 /* { 334 syscallarg(clockid_t) clock_id; 335 syscallarg(struct timespec *) tp; 336 } */ 337 struct timespec ts; 338 int error; 339 340 if ((error = clock_getres1(SCARG(uap, clock_id), &ts)) != 0) 341 return error; 342 343 if (SCARG(uap, tp)) 344 error = copyout(&ts, SCARG(uap, tp), sizeof(ts)); 345 346 return error; 347 } 348 349 int 350 clock_getres1(clockid_t clock_id, struct timespec *ts) 351 { 352 353 switch (clock_id) { 354 case CLOCK_REALTIME: 355 case CLOCK_MONOTONIC: 356 case CLOCK_PROCESS_CPUTIME_ID: 357 case CLOCK_THREAD_CPUTIME_ID: 358 ts->tv_sec = 0; 359 if (tc_getfrequency() > 1000000000) 360 ts->tv_nsec = 1; 361 else 362 ts->tv_nsec = 1000000000 / tc_getfrequency(); 363 break; 364 default: 365 return SET_ERROR(EINVAL); 366 } 367 368 return 0; 369 } 370 371 /* ARGSUSED */ 372 int 373 sys___nanosleep50(struct lwp *l, const struct sys___nanosleep50_args *uap, 374 register_t *retval) 375 { 376 /* { 377 syscallarg(struct timespec *) rqtp; 378 syscallarg(struct timespec *) rmtp; 379 } */ 380 struct timespec rmt, rqt; 381 int error, error1; 382 383 error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec)); 384 if (error) 385 return error; 386 387 error = nanosleep1(l, CLOCK_MONOTONIC, 0, &rqt, 388 SCARG(uap, rmtp) ? &rmt : NULL); 389 if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR)) 390 return error; 391 392 error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt)); 393 return error1 ? error1 : error; 394 } 395 396 /* ARGSUSED */ 397 int 398 sys_clock_nanosleep(struct lwp *l, const struct sys_clock_nanosleep_args *uap, 399 register_t *retval) 400 { 401 /* { 402 syscallarg(clockid_t) clock_id; 403 syscallarg(int) flags; 404 syscallarg(struct timespec *) rqtp; 405 syscallarg(struct timespec *) rmtp; 406 } */ 407 struct timespec rmt, rqt; 408 int error, error1; 409 410 error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec)); 411 if (error) 412 goto out; 413 414 error = nanosleep1(l, SCARG(uap, clock_id), SCARG(uap, flags), &rqt, 415 SCARG(uap, rmtp) ? &rmt : NULL); 416 if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR)) 417 goto out; 418 419 if ((SCARG(uap, flags) & TIMER_ABSTIME) == 0 && 420 (error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt))) != 0) 421 error = error1; 422 out: 423 *retval = error; 424 return 0; 425 } 426 427 int 428 nanosleep1(struct lwp *l, clockid_t clock_id, int flags, struct timespec *rqt, 429 struct timespec *rmt) 430 { 431 struct timespec rmtstart; 432 int error, timo; 433 434 if ((error = ts2timo(clock_id, flags, rqt, &timo, &rmtstart)) != 0) { 435 if (error == ETIMEDOUT) { 436 error = 0; 437 if (rmt != NULL) 438 rmt->tv_sec = rmt->tv_nsec = 0; 439 } 440 return error; 441 } 442 443 /* 444 * Avoid inadvertently sleeping forever 445 */ 446 if (timo == 0) 447 timo = 1; 448 again: 449 error = kpause("nanoslp", true, timo, NULL); 450 if (error == EWOULDBLOCK) 451 error = 0; 452 if (rmt != NULL || error == 0) { 453 struct timespec rmtend; 454 struct timespec t0; 455 struct timespec *t; 456 int err; 457 458 err = clock_gettime1(clock_id, &rmtend); 459 if (err != 0) 460 return err; 461 462 t = (rmt != NULL) ? rmt : &t0; 463 if (flags & TIMER_ABSTIME) { 464 timespecsub(rqt, &rmtend, t); 465 } else { 466 if (timespeccmp(&rmtend, &rmtstart, <)) 467 timespecclear(t); /* clock wound back */ 468 else 469 timespecsub(&rmtend, &rmtstart, t); 470 if (timespeccmp(rqt, t, <)) 471 timespecclear(t); 472 else 473 timespecsub(rqt, t, t); 474 } 475 if (t->tv_sec < 0) 476 timespecclear(t); 477 if (error == 0) { 478 timo = tstohz(t); 479 if (timo > 0) 480 goto again; 481 } 482 } 483 484 if (error == ERESTART) 485 error = SET_ERROR(EINTR); 486 487 return error; 488 } 489 490 int 491 sys_clock_getcpuclockid2(struct lwp *l, 492 const struct sys_clock_getcpuclockid2_args *uap, 493 register_t *retval) 494 { 495 /* { 496 syscallarg(idtype_t idtype; 497 syscallarg(id_t id); 498 syscallarg(clockid_t *)clock_id; 499 } */ 500 pid_t pid; 501 lwpid_t lid; 502 clockid_t clock_id; 503 id_t id = SCARG(uap, id); 504 505 switch (SCARG(uap, idtype)) { 506 case P_PID: 507 pid = id == 0 ? l->l_proc->p_pid : id; 508 clock_id = CLOCK_PROCESS_CPUTIME_ID | pid; 509 break; 510 case P_LWPID: 511 lid = id == 0 ? l->l_lid : id; 512 clock_id = CLOCK_THREAD_CPUTIME_ID | lid; 513 break; 514 default: 515 return SET_ERROR(EINVAL); 516 } 517 return copyout(&clock_id, SCARG(uap, clock_id), sizeof(clock_id)); 518 } 519 520 /* ARGSUSED */ 521 int 522 sys___gettimeofday50(struct lwp *l, const struct sys___gettimeofday50_args *uap, 523 register_t *retval) 524 { 525 /* { 526 syscallarg(struct timeval *) tp; 527 syscallarg(void *) tzp; really "struct timezone *"; 528 } */ 529 struct timeval atv; 530 int error = 0; 531 struct timezone tzfake; 532 533 if (SCARG(uap, tp)) { 534 memset(&atv, 0, sizeof(atv)); 535 microtime(&atv); 536 error = copyout(&atv, SCARG(uap, tp), sizeof(atv)); 537 if (error) 538 return error; 539 } 540 if (SCARG(uap, tzp)) { 541 /* 542 * NetBSD has no kernel notion of time zone, so we just 543 * fake up a timezone struct and return it if demanded. 544 */ 545 tzfake.tz_minuteswest = 0; 546 tzfake.tz_dsttime = 0; 547 error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake)); 548 } 549 return error; 550 } 551 552 /* ARGSUSED */ 553 int 554 sys___settimeofday50(struct lwp *l, const struct sys___settimeofday50_args *uap, 555 register_t *retval) 556 { 557 /* { 558 syscallarg(const struct timeval *) tv; 559 syscallarg(const void *) tzp; really "const struct timezone *"; 560 } */ 561 562 return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true); 563 } 564 565 int 566 settimeofday1(const struct timeval *utv, bool userspace, 567 const void *utzp, struct lwp *l, bool check_kauth) 568 { 569 struct timeval atv; 570 struct timespec ts; 571 int error; 572 573 /* Verify all parameters before changing time. */ 574 575 /* 576 * NetBSD has no kernel notion of time zone, and only an 577 * obsolete program would try to set it, so we log a warning. 578 */ 579 if (utzp) 580 log(LOG_WARNING, "pid %d attempted to set the " 581 "(obsolete) kernel time zone\n", l->l_proc->p_pid); 582 583 if (utv == NULL) 584 return 0; 585 586 if (userspace) { 587 if ((error = copyin(utv, &atv, sizeof(atv))) != 0) 588 return error; 589 utv = &atv; 590 } 591 592 if (utv->tv_usec < 0 || utv->tv_usec >= 1000000) 593 return SET_ERROR(EINVAL); 594 595 TIMEVAL_TO_TIMESPEC(utv, &ts); 596 return settime1(l->l_proc, &ts, check_kauth); 597 } 598 599 int time_adjusted; /* set if an adjustment is made */ 600 601 /* ARGSUSED */ 602 int 603 sys___adjtime50(struct lwp *l, const struct sys___adjtime50_args *uap, 604 register_t *retval) 605 { 606 /* { 607 syscallarg(const struct timeval *) delta; 608 syscallarg(struct timeval *) olddelta; 609 } */ 610 int error; 611 struct timeval atv, oldatv; 612 613 if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME, 614 KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0) 615 return error; 616 617 if (SCARG(uap, delta)) { 618 error = copyin(SCARG(uap, delta), &atv, 619 sizeof(*SCARG(uap, delta))); 620 if (error) 621 return error; 622 } 623 adjtime1(SCARG(uap, delta) ? &atv : NULL, 624 SCARG(uap, olddelta) ? &oldatv : NULL, l->l_proc); 625 if (SCARG(uap, olddelta)) 626 error = copyout(&oldatv, SCARG(uap, olddelta), 627 sizeof(*SCARG(uap, olddelta))); 628 return error; 629 } 630 631 void 632 adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p) 633 { 634 635 if (olddelta) { 636 memset(olddelta, 0, sizeof(*olddelta)); 637 mutex_spin_enter(&timecounter_lock); 638 olddelta->tv_sec = time_adjtime / 1000000; 639 olddelta->tv_usec = time_adjtime % 1000000; 640 if (olddelta->tv_usec < 0) { 641 olddelta->tv_usec += 1000000; 642 olddelta->tv_sec--; 643 } 644 mutex_spin_exit(&timecounter_lock); 645 } 646 647 if (delta) { 648 mutex_spin_enter(&timecounter_lock); 649 /* 650 * XXX This should maybe just report failure to 651 * userland for nonsense deltas. 652 */ 653 if (delta->tv_sec > INT64_MAX/1000000 - 1) { 654 time_adjtime = INT64_MAX; 655 } else if (delta->tv_sec < INT64_MIN/1000000 + 1) { 656 time_adjtime = INT64_MIN; 657 } else { 658 time_adjtime = delta->tv_sec * 1000000 659 + MAX(-999999, MIN(999999, delta->tv_usec)); 660 } 661 662 if (time_adjtime) { 663 /* We need to save the system time during shutdown */ 664 time_adjusted |= 1; 665 } 666 mutex_spin_exit(&timecounter_lock); 667 } 668 } 669 670 /* 671 * Interval timer support. 672 * 673 * The itimer_*() routines provide generic support for interval timers, 674 * both real (CLOCK_REALTIME, CLOCK_MONOTIME), and virtual (CLOCK_VIRTUAL, 675 * CLOCK_PROF). 676 * 677 * Real timers keep their deadline as an absolute time, and are fired 678 * by a callout. Virtual timers are kept as a linked-list of deltas, 679 * and are processed by hardclock(). 680 * 681 * Because the real time timer callout may be delayed in real time due 682 * to interrupt processing on the system, it is possible for the real 683 * time timeout routine (itimer_callout()) run past after its deadline. 684 * It does not suffice, therefore, to reload the real timer .it_value 685 * from the timer's .it_interval. Rather, we compute the next deadline 686 * in absolute time based on the current time and the .it_interval value, 687 * and report any overruns. 688 * 689 * Note that while the virtual timers are supported in a generic fashion 690 * here, they only (currently) make sense as per-process timers, and thus 691 * only really work for that case. 692 */ 693 694 /* 695 * itimer_init: 696 * 697 * Initialize the common data for an interval timer. 698 */ 699 void 700 itimer_init(struct itimer * const it, const struct itimer_ops * const ops, 701 clockid_t const id, struct itlist * const itl) 702 { 703 704 KASSERT(itimer_lock_held()); 705 KASSERT(ops != NULL); 706 707 timespecclear(&it->it_time.it_value); 708 it->it_ops = ops; 709 it->it_clockid = id; 710 it->it_overruns = 0; 711 it->it_dying = false; 712 if (!CLOCK_VIRTUAL_P(id)) { 713 KASSERT(itl == NULL); 714 callout_init(&it->it_ch, CALLOUT_MPSAFE); 715 callout_setfunc(&it->it_ch, itimer_callout, it); 716 if (id == CLOCK_REALTIME && ops->ito_realtime_changed != NULL) { 717 LIST_INSERT_HEAD(&itimer_realtime_changed_notify, 718 it, it_rtchgq); 719 } 720 } else { 721 KASSERT(itl != NULL); 722 it->it_vlist = itl; 723 it->it_active = false; 724 } 725 } 726 727 /* 728 * itimer_poison: 729 * 730 * Poison an interval timer, preventing it from being scheduled 731 * or processed, in preparation for freeing the timer. 732 */ 733 void 734 itimer_poison(struct itimer * const it) 735 { 736 737 KASSERT(itimer_lock_held()); 738 739 it->it_dying = true; 740 741 /* 742 * For non-virtual timers, stop the callout, or wait for it to 743 * run if it has already fired. It cannot restart again after 744 * this point: the callout won't restart itself when dying, no 745 * other users holding the lock can restart it, and any other 746 * users waiting for callout_halt concurrently (itimer_settime) 747 * will restart from the top. 748 */ 749 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 750 callout_halt(&it->it_ch, &itimer_mutex); 751 if (it->it_clockid == CLOCK_REALTIME && 752 it->it_ops->ito_realtime_changed != NULL) { 753 LIST_REMOVE(it, it_rtchgq); 754 } 755 } 756 } 757 758 /* 759 * itimer_fini: 760 * 761 * Release resources used by an interval timer. 762 * 763 * N.B. itimer_lock must be held on entry, and is released on exit. 764 */ 765 void 766 itimer_fini(struct itimer * const it) 767 { 768 769 KASSERT(itimer_lock_held()); 770 771 /* All done with the global state. */ 772 itimer_unlock(); 773 774 /* Destroy the callout, if needed. */ 775 if (!CLOCK_VIRTUAL_P(it->it_clockid)) 776 callout_destroy(&it->it_ch); 777 } 778 779 /* 780 * itimer_decr: 781 * 782 * Decrement an interval timer by a specified number of nanoseconds, 783 * which must be less than a second, i.e. < 1000000000. If the timer 784 * expires, then reload it. In this case, carry over (nsec - old value) 785 * to reduce the value reloaded into the timer so that the timer does 786 * not drift. This routine assumes that it is called in a context where 787 * the timers on which it is operating cannot change in value. 788 * 789 * Returns true if the timer has expired. 790 */ 791 static bool 792 itimer_decr(struct itimer *it, int nsec) 793 { 794 struct itimerspec *itp; 795 int error __diagused; 796 797 KASSERT(itimer_lock_held()); 798 KASSERT(CLOCK_VIRTUAL_P(it->it_clockid)); 799 800 itp = &it->it_time; 801 if (itp->it_value.tv_nsec < nsec) { 802 if (itp->it_value.tv_sec == 0) { 803 /* expired, and already in next interval */ 804 nsec -= itp->it_value.tv_nsec; 805 goto expire; 806 } 807 itp->it_value.tv_nsec += 1000000000; 808 itp->it_value.tv_sec--; 809 } 810 itp->it_value.tv_nsec -= nsec; 811 nsec = 0; 812 if (timespecisset(&itp->it_value)) 813 return false; 814 /* expired, exactly at end of interval */ 815 expire: 816 if (timespecisset(&itp->it_interval)) { 817 itp->it_value = itp->it_interval; 818 itp->it_value.tv_nsec -= nsec; 819 if (itp->it_value.tv_nsec < 0) { 820 itp->it_value.tv_nsec += 1000000000; 821 itp->it_value.tv_sec--; 822 } 823 error = itimer_settime(it); 824 KASSERT(error == 0); /* virtual, never fails */ 825 } else 826 itp->it_value.tv_nsec = 0; /* sec is already 0 */ 827 return true; 828 } 829 830 /* 831 * itimer_arm_real: 832 * 833 * Arm a non-virtual timer. 834 */ 835 static void 836 itimer_arm_real(struct itimer * const it) 837 { 838 839 KASSERT(!it->it_dying); 840 KASSERT(!CLOCK_VIRTUAL_P(it->it_clockid)); 841 KASSERT(!callout_pending(&it->it_ch)); 842 843 /* 844 * Don't need to check tshzto() return value, here. 845 * callout_schedule() does it for us. 846 */ 847 callout_schedule(&it->it_ch, 848 (it->it_clockid == CLOCK_MONOTONIC 849 ? tshztoup(&it->it_time.it_value) 850 : tshzto(&it->it_time.it_value))); 851 } 852 853 /* 854 * itimer_callout: 855 * 856 * Callout to expire a non-virtual timer. Queue it up for processing, 857 * and then reload, if it is configured to do so. 858 * 859 * N.B. A delay in processing this callout causes multiple 860 * SIGALRM calls to be compressed into one. 861 */ 862 static void 863 itimer_callout(void *arg) 864 { 865 struct timespec now, next; 866 struct itimer * const it = arg; 867 int overruns; 868 869 itimer_lock(); 870 (*it->it_ops->ito_fire)(it); 871 872 if (!timespecisset(&it->it_time.it_interval)) { 873 timespecclear(&it->it_time.it_value); 874 itimer_unlock(); 875 return; 876 } 877 878 if (it->it_clockid == CLOCK_MONOTONIC) { 879 getnanouptime(&now); 880 } else { 881 getnanotime(&now); 882 } 883 884 /* 885 * Given the current itimer value and interval and the time 886 * now, compute the next itimer value and count overruns. 887 */ 888 itimer_transition(&it->it_time, &now, &next, &overruns); 889 it->it_time.it_value = next; 890 it->it_overruns += MIN(INT_MAX - overruns, overruns); 891 892 /* 893 * Reset the callout, if it's not going away. 894 */ 895 if (!it->it_dying) 896 itimer_arm_real(it); 897 itimer_unlock(); 898 } 899 900 /* 901 * itimer_settime: 902 * 903 * Set up the given interval timer. The value in it->it_time.it_value 904 * is taken to be an absolute time for CLOCK_REALTIME/CLOCK_MONOTONIC 905 * timers and a relative time for CLOCK_VIRTUAL/CLOCK_PROF timers. 906 * 907 * If the callout had already fired but not yet run, fails with 908 * ERESTART -- caller must restart from the top to look up a timer. 909 * 910 * Caller is responsible for validating it->it_value and 911 * it->it_interval, e.g. with itimerfix or itimespecfix. 912 */ 913 int 914 itimer_settime(struct itimer *it) 915 { 916 struct itimer *itn, *pitn; 917 struct itlist *itl; 918 919 KASSERT(itimer_lock_held()); 920 KASSERT(!it->it_dying); 921 KASSERT(it->it_time.it_value.tv_sec >= 0); 922 KASSERT(it->it_time.it_value.tv_nsec >= 0); 923 KASSERT(it->it_time.it_value.tv_nsec < 1000000000); 924 KASSERT(it->it_time.it_interval.tv_sec >= 0); 925 KASSERT(it->it_time.it_interval.tv_nsec >= 0); 926 KASSERT(it->it_time.it_interval.tv_nsec < 1000000000); 927 928 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 929 /* 930 * Try to stop the callout. However, if it had already 931 * fired, we have to drop the lock to wait for it, so 932 * the world may have changed and pt may not be there 933 * any more. In that case, tell the caller to start 934 * over from the top. 935 */ 936 if (callout_halt(&it->it_ch, &itimer_mutex)) 937 return SET_ERROR(ERESTART); 938 KASSERT(!it->it_dying); 939 940 /* Now we can touch it and start it up again. */ 941 if (timespecisset(&it->it_time.it_value)) 942 itimer_arm_real(it); 943 } else { 944 if (it->it_active) { 945 itn = LIST_NEXT(it, it_list); 946 LIST_REMOVE(it, it_list); 947 for ( ; itn; itn = LIST_NEXT(itn, it_list)) 948 timespecadd(&it->it_time.it_value, 949 &itn->it_time.it_value, 950 &itn->it_time.it_value); 951 } 952 if (timespecisset(&it->it_time.it_value)) { 953 itl = it->it_vlist; 954 for (itn = LIST_FIRST(itl), pitn = NULL; 955 itn && timespeccmp(&it->it_time.it_value, 956 &itn->it_time.it_value, >); 957 pitn = itn, itn = LIST_NEXT(itn, it_list)) 958 timespecsub(&it->it_time.it_value, 959 &itn->it_time.it_value, 960 &it->it_time.it_value); 961 962 if (pitn) 963 LIST_INSERT_AFTER(pitn, it, it_list); 964 else 965 LIST_INSERT_HEAD(itl, it, it_list); 966 967 for ( ; itn ; itn = LIST_NEXT(itn, it_list)) 968 timespecsub(&itn->it_time.it_value, 969 &it->it_time.it_value, 970 &itn->it_time.it_value); 971 972 it->it_active = true; 973 } else { 974 it->it_active = false; 975 } 976 } 977 978 /* Success! */ 979 return 0; 980 } 981 982 /* 983 * itimer_gettime: 984 * 985 * Return the remaining time of an interval timer. 986 */ 987 void 988 itimer_gettime(const struct itimer *it, struct itimerspec *aits) 989 { 990 struct timespec now; 991 struct itimer *itn; 992 993 KASSERT(itimer_lock_held()); 994 KASSERT(!it->it_dying); 995 996 *aits = it->it_time; 997 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 998 /* 999 * Convert from absolute to relative time in .it_value 1000 * part of real time timer. If time for real time 1001 * timer has passed return 0, else return difference 1002 * between current time and time for the timer to go 1003 * off. 1004 */ 1005 if (timespecisset(&aits->it_value)) { 1006 if (it->it_clockid == CLOCK_REALTIME) { 1007 getnanotime(&now); 1008 } else { /* CLOCK_MONOTONIC */ 1009 getnanouptime(&now); 1010 } 1011 if (timespeccmp(&aits->it_value, &now, <)) 1012 timespecclear(&aits->it_value); 1013 else 1014 timespecsub(&aits->it_value, &now, 1015 &aits->it_value); 1016 } 1017 } else if (it->it_active) { 1018 for (itn = LIST_FIRST(it->it_vlist); itn && itn != it; 1019 itn = LIST_NEXT(itn, it_list)) 1020 timespecadd(&aits->it_value, 1021 &itn->it_time.it_value, &aits->it_value); 1022 KASSERT(itn != NULL); /* it should be findable on the list */ 1023 } else 1024 timespecclear(&aits->it_value); 1025 } 1026 1027 /* 1028 * Per-process timer support. 1029 * 1030 * Both the BSD getitimer() family and the POSIX timer_*() family of 1031 * routines are supported. 1032 * 1033 * All timers are kept in an array pointed to by p_timers, which is 1034 * allocated on demand - many processes don't use timers at all. The 1035 * first four elements in this array are reserved for the BSD timers: 1036 * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, element 1037 * 2 is ITIMER_PROF, and element 3 is ITIMER_MONOTONIC. The rest may be 1038 * allocated by the timer_create() syscall. 1039 * 1040 * These timers are a "sub-class" of interval timer. 1041 */ 1042 1043 /* 1044 * ptimer_free: 1045 * 1046 * Free the per-process timer at the specified index. 1047 */ 1048 static void 1049 ptimer_free(struct ptimers *pts, int index) 1050 { 1051 struct itimer *it; 1052 struct ptimer *pt; 1053 1054 KASSERT(itimer_lock_held()); 1055 1056 it = pts->pts_timers[index]; 1057 pt = container_of(it, struct ptimer, pt_itimer); 1058 pts->pts_timers[index] = NULL; 1059 itimer_poison(it); 1060 1061 /* 1062 * Remove it from the queue to be signalled. Must be done 1063 * after itimer is poisoned, because we may have had to wait 1064 * for the callout to complete. 1065 */ 1066 if (pt->pt_queued) { 1067 TAILQ_REMOVE(&ptimer_queue, pt, pt_chain); 1068 pt->pt_queued = false; 1069 } 1070 1071 itimer_fini(it); /* releases itimer_lock */ 1072 kmem_free(pt, sizeof(*pt)); 1073 } 1074 1075 /* 1076 * ptimers_alloc: 1077 * 1078 * Allocate a ptimers for the specified process. 1079 */ 1080 static struct ptimers * 1081 ptimers_alloc(struct proc *p) 1082 { 1083 struct ptimers *pts; 1084 int i; 1085 1086 pts = kmem_alloc(sizeof(*pts), KM_SLEEP); 1087 LIST_INIT(&pts->pts_virtual); 1088 LIST_INIT(&pts->pts_prof); 1089 for (i = 0; i < TIMER_MAX; i++) 1090 pts->pts_timers[i] = NULL; 1091 itimer_lock(); 1092 if (p->p_timers == NULL) { 1093 p->p_timers = pts; 1094 itimer_unlock(); 1095 return pts; 1096 } 1097 itimer_unlock(); 1098 kmem_free(pts, sizeof(*pts)); 1099 return p->p_timers; 1100 } 1101 1102 /* 1103 * ptimers_free: 1104 * 1105 * Clean up the per-process timers. If "which" is set to TIMERS_ALL, 1106 * then clean up all timers and free all the data structures. If 1107 * "which" is set to TIMERS_POSIX, only clean up the timers allocated 1108 * by timer_create(), not the BSD setitimer() timers, and only free the 1109 * structure if none of those remain. 1110 * 1111 * This function is exported because it is needed in the exec and 1112 * exit code paths. 1113 */ 1114 void 1115 ptimers_free(struct proc *p, int which) 1116 { 1117 struct ptimers *pts; 1118 struct itimer *itn; 1119 struct timespec ts; 1120 int i; 1121 1122 if (p->p_timers == NULL) 1123 return; 1124 1125 pts = p->p_timers; 1126 itimer_lock(); 1127 if (which == TIMERS_ALL) { 1128 p->p_timers = NULL; 1129 i = 0; 1130 } else { 1131 timespecclear(&ts); 1132 for (itn = LIST_FIRST(&pts->pts_virtual); 1133 itn && itn != pts->pts_timers[ITIMER_VIRTUAL]; 1134 itn = LIST_NEXT(itn, it_list)) { 1135 KASSERT(itn->it_clockid == CLOCK_VIRTUAL); 1136 timespecadd(&ts, &itn->it_time.it_value, &ts); 1137 } 1138 LIST_FIRST(&pts->pts_virtual) = NULL; 1139 if (itn) { 1140 KASSERT(itn->it_clockid == CLOCK_VIRTUAL); 1141 timespecadd(&ts, &itn->it_time.it_value, 1142 &itn->it_time.it_value); 1143 LIST_INSERT_HEAD(&pts->pts_virtual, itn, it_list); 1144 } 1145 timespecclear(&ts); 1146 for (itn = LIST_FIRST(&pts->pts_prof); 1147 itn && itn != pts->pts_timers[ITIMER_PROF]; 1148 itn = LIST_NEXT(itn, it_list)) { 1149 KASSERT(itn->it_clockid == CLOCK_PROF); 1150 timespecadd(&ts, &itn->it_time.it_value, &ts); 1151 } 1152 LIST_FIRST(&pts->pts_prof) = NULL; 1153 if (itn) { 1154 KASSERT(itn->it_clockid == CLOCK_PROF); 1155 timespecadd(&ts, &itn->it_time.it_value, 1156 &itn->it_time.it_value); 1157 LIST_INSERT_HEAD(&pts->pts_prof, itn, it_list); 1158 } 1159 i = TIMER_MIN; 1160 } 1161 for ( ; i < TIMER_MAX; i++) { 1162 if (pts->pts_timers[i] != NULL) { 1163 /* Free the timer and release the lock. */ 1164 ptimer_free(pts, i); 1165 /* Reacquire the lock for the next one. */ 1166 itimer_lock(); 1167 } 1168 } 1169 if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL && 1170 pts->pts_timers[2] == NULL && pts->pts_timers[3] == NULL) { 1171 p->p_timers = NULL; 1172 itimer_unlock(); 1173 kmem_free(pts, sizeof(*pts)); 1174 } else 1175 itimer_unlock(); 1176 } 1177 1178 /* 1179 * ptimer_fire: 1180 * 1181 * Fire a per-process timer. 1182 */ 1183 static void 1184 ptimer_fire(struct itimer *it) 1185 { 1186 struct ptimer *pt = container_of(it, struct ptimer, pt_itimer); 1187 1188 KASSERT(itimer_lock_held()); 1189 1190 /* 1191 * XXX Can overrun, but we don't do signal queueing yet, anyway. 1192 * XXX Relying on the clock interrupt is stupid. 1193 */ 1194 if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) { 1195 return; 1196 } 1197 1198 if (!pt->pt_queued) { 1199 TAILQ_INSERT_TAIL(&ptimer_queue, pt, pt_chain); 1200 pt->pt_queued = true; 1201 softint_schedule(ptimer_sih); 1202 } 1203 } 1204 1205 /* 1206 * Operations vector for per-process timers (BSD and POSIX). 1207 */ 1208 static const struct itimer_ops ptimer_itimer_ops = { 1209 .ito_fire = ptimer_fire, 1210 }; 1211 1212 /* 1213 * sys_timer_create: 1214 * 1215 * System call to create a POSIX timer. 1216 */ 1217 int 1218 sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap, 1219 register_t *retval) 1220 { 1221 /* { 1222 syscallarg(clockid_t) clock_id; 1223 syscallarg(struct sigevent *) evp; 1224 syscallarg(timer_t *) timerid; 1225 } */ 1226 1227 return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id), 1228 SCARG(uap, evp), copyin, l); 1229 } 1230 1231 int 1232 timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp, 1233 copyin_t fetch_event, struct lwp *l) 1234 { 1235 int error; 1236 timer_t timerid; 1237 struct itlist *itl; 1238 struct ptimers *pts; 1239 struct ptimer *pt; 1240 struct proc *p; 1241 1242 p = l->l_proc; 1243 1244 if ((u_int)id > CLOCK_MONOTONIC) 1245 return SET_ERROR(EINVAL); 1246 1247 if ((pts = p->p_timers) == NULL) 1248 pts = ptimers_alloc(p); 1249 1250 pt = kmem_zalloc(sizeof(*pt), KM_SLEEP); 1251 if (evp != NULL) { 1252 if (((error = 1253 (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) || 1254 ((pt->pt_ev.sigev_notify < SIGEV_NONE) || 1255 (pt->pt_ev.sigev_notify > SIGEV_SA)) || 1256 (pt->pt_ev.sigev_notify == SIGEV_SIGNAL && 1257 (pt->pt_ev.sigev_signo <= 0 || 1258 pt->pt_ev.sigev_signo >= NSIG))) { 1259 kmem_free(pt, sizeof(*pt)); 1260 return (error ? error : SET_ERROR(EINVAL)); 1261 } 1262 } 1263 1264 /* Find a free timer slot, skipping those reserved for setitimer(). */ 1265 itimer_lock(); 1266 for (timerid = TIMER_MIN; timerid < TIMER_MAX; timerid++) 1267 if (pts->pts_timers[timerid] == NULL) 1268 break; 1269 if (timerid == TIMER_MAX) { 1270 itimer_unlock(); 1271 kmem_free(pt, sizeof(*pt)); 1272 return SET_ERROR(EAGAIN); 1273 } 1274 if (evp == NULL) { 1275 pt->pt_ev.sigev_notify = SIGEV_SIGNAL; 1276 switch (id) { 1277 case CLOCK_REALTIME: 1278 case CLOCK_MONOTONIC: 1279 pt->pt_ev.sigev_signo = SIGALRM; 1280 break; 1281 case CLOCK_VIRTUAL: 1282 pt->pt_ev.sigev_signo = SIGVTALRM; 1283 break; 1284 case CLOCK_PROF: 1285 pt->pt_ev.sigev_signo = SIGPROF; 1286 break; 1287 } 1288 pt->pt_ev.sigev_value.sival_int = timerid; 1289 } 1290 1291 switch (id) { 1292 case CLOCK_VIRTUAL: 1293 itl = &pts->pts_virtual; 1294 break; 1295 case CLOCK_PROF: 1296 itl = &pts->pts_prof; 1297 break; 1298 default: 1299 itl = NULL; 1300 } 1301 1302 itimer_init(&pt->pt_itimer, &ptimer_itimer_ops, id, itl); 1303 pt->pt_proc = p; 1304 pt->pt_poverruns = 0; 1305 pt->pt_entry = timerid; 1306 pt->pt_queued = false; 1307 1308 pts->pts_timers[timerid] = &pt->pt_itimer; 1309 itimer_unlock(); 1310 1311 return copyout(&timerid, tid, sizeof(timerid)); 1312 } 1313 1314 /* 1315 * sys_timer_delete: 1316 * 1317 * System call to delete a POSIX timer. 1318 */ 1319 int 1320 sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap, 1321 register_t *retval) 1322 { 1323 /* { 1324 syscallarg(timer_t) timerid; 1325 } */ 1326 struct proc *p = l->l_proc; 1327 timer_t timerid; 1328 struct ptimers *pts; 1329 struct itimer *it, *itn; 1330 1331 timerid = SCARG(uap, timerid); 1332 pts = p->p_timers; 1333 1334 if (pts == NULL || !IS_POSIX_TIMER(timerid)) 1335 return SET_ERROR(EINVAL); 1336 1337 itimer_lock(); 1338 if ((it = pts->pts_timers[timerid]) == NULL) { 1339 itimer_unlock(); 1340 return SET_ERROR(EINVAL); 1341 } 1342 1343 if (CLOCK_VIRTUAL_P(it->it_clockid)) { 1344 if (it->it_active) { 1345 itn = LIST_NEXT(it, it_list); 1346 LIST_REMOVE(it, it_list); 1347 for ( ; itn; itn = LIST_NEXT(itn, it_list)) 1348 timespecadd(&it->it_time.it_value, 1349 &itn->it_time.it_value, 1350 &itn->it_time.it_value); 1351 it->it_active = false; 1352 } 1353 } 1354 1355 /* Free the timer and release the lock. */ 1356 ptimer_free(pts, timerid); 1357 1358 return 0; 1359 } 1360 1361 /* 1362 * sys___timer_settime50: 1363 * 1364 * System call to set/arm a POSIX timer. 1365 */ 1366 int 1367 sys___timer_settime50(struct lwp *l, 1368 const struct sys___timer_settime50_args *uap, 1369 register_t *retval) 1370 { 1371 /* { 1372 syscallarg(timer_t) timerid; 1373 syscallarg(int) flags; 1374 syscallarg(const struct itimerspec *) value; 1375 syscallarg(struct itimerspec *) ovalue; 1376 } */ 1377 int error; 1378 struct itimerspec value, ovalue, *ovp = NULL; 1379 1380 if ((error = copyin(SCARG(uap, value), &value, 1381 sizeof(struct itimerspec))) != 0) 1382 return error; 1383 1384 if (SCARG(uap, ovalue)) 1385 ovp = &ovalue; 1386 1387 if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp, 1388 SCARG(uap, flags), l->l_proc)) != 0) 1389 return error; 1390 1391 if (ovp) 1392 return copyout(&ovalue, SCARG(uap, ovalue), 1393 sizeof(struct itimerspec)); 1394 return 0; 1395 } 1396 1397 int 1398 dotimer_settime(int timerid, struct itimerspec *value, 1399 struct itimerspec *ovalue, int flags, struct proc *p) 1400 { 1401 struct timespec now; 1402 struct itimerspec val; 1403 struct ptimers *pts; 1404 struct itimer *it; 1405 int error; 1406 1407 pts = p->p_timers; 1408 1409 if (pts == NULL || !IS_POSIX_TIMER(timerid)) 1410 return SET_ERROR(EINVAL); 1411 val = *value; 1412 if (itimespecfix(&val.it_value) != 0 || 1413 itimespecfix(&val.it_interval) != 0) 1414 return SET_ERROR(EINVAL); 1415 1416 itimer_lock(); 1417 restart: 1418 if ((it = pts->pts_timers[timerid]) == NULL) { 1419 error = SET_ERROR(EINVAL); 1420 goto out; 1421 } 1422 1423 if (ovalue) 1424 itimer_gettime(it, ovalue); 1425 it->it_time = val; 1426 1427 /* 1428 * If we've been passed a relative time for a realtime timer, 1429 * convert it to absolute; if an absolute time for a virtual 1430 * timer, convert it to relative and make sure we don't set it 1431 * to zero, which would cancel the timer, or let it go 1432 * negative, which would confuse the comparison tests. 1433 */ 1434 if (timespecisset(&it->it_time.it_value)) { 1435 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 1436 if ((flags & TIMER_ABSTIME) == 0) { 1437 if (it->it_clockid == CLOCK_REALTIME) { 1438 getnanotime(&now); 1439 } else { /* CLOCK_MONOTONIC */ 1440 getnanouptime(&now); 1441 } 1442 if (!timespecaddok(&it->it_time.it_value, 1443 &now)) { 1444 error = SET_ERROR(EINVAL); 1445 goto out; 1446 } 1447 timespecadd(&it->it_time.it_value, &now, 1448 &it->it_time.it_value); 1449 } 1450 } else { 1451 if ((flags & TIMER_ABSTIME) != 0) { 1452 getnanotime(&now); 1453 if (!timespecsubok(&it->it_time.it_value, 1454 &now)) { 1455 error = SET_ERROR(EINVAL); 1456 goto out; 1457 } 1458 timespecsub(&it->it_time.it_value, &now, 1459 &it->it_time.it_value); 1460 if (!timespecisset(&it->it_time.it_value) || 1461 it->it_time.it_value.tv_sec < 0) { 1462 it->it_time.it_value.tv_sec = 0; 1463 it->it_time.it_value.tv_nsec = 1; 1464 } 1465 } 1466 } 1467 } 1468 1469 error = itimer_settime(it); 1470 if (error == ERESTART) { 1471 KASSERT(!CLOCK_VIRTUAL_P(it->it_clockid)); 1472 goto restart; 1473 } 1474 KASSERT(error == 0); 1475 out: 1476 itimer_unlock(); 1477 1478 return error; 1479 } 1480 1481 /* 1482 * sys___timer_gettime50: 1483 * 1484 * System call to return the time remaining until a POSIX timer fires. 1485 */ 1486 int 1487 sys___timer_gettime50(struct lwp *l, 1488 const struct sys___timer_gettime50_args *uap, register_t *retval) 1489 { 1490 /* { 1491 syscallarg(timer_t) timerid; 1492 syscallarg(struct itimerspec *) value; 1493 } */ 1494 struct itimerspec its; 1495 int error; 1496 1497 if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc, 1498 &its)) != 0) 1499 return error; 1500 1501 return copyout(&its, SCARG(uap, value), sizeof(its)); 1502 } 1503 1504 int 1505 dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its) 1506 { 1507 struct itimer *it; 1508 struct ptimers *pts; 1509 1510 pts = p->p_timers; 1511 if (pts == NULL || !IS_POSIX_TIMER(timerid)) 1512 return SET_ERROR(EINVAL); 1513 itimer_lock(); 1514 if ((it = pts->pts_timers[timerid]) == NULL) { 1515 itimer_unlock(); 1516 return SET_ERROR(EINVAL); 1517 } 1518 itimer_gettime(it, its); 1519 itimer_unlock(); 1520 1521 return 0; 1522 } 1523 1524 /* 1525 * sys_timer_getoverrun: 1526 * 1527 * System call to return the number of times a POSIX timer has 1528 * expired while a notification was already pending. The counter 1529 * is reset when a timer expires and a notification can be posted. 1530 */ 1531 int 1532 sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap, 1533 register_t *retval) 1534 { 1535 /* { 1536 syscallarg(timer_t) timerid; 1537 } */ 1538 struct proc *p = l->l_proc; 1539 struct ptimers *pts; 1540 int timerid; 1541 struct itimer *it; 1542 struct ptimer *pt; 1543 1544 timerid = SCARG(uap, timerid); 1545 1546 pts = p->p_timers; 1547 if (pts == NULL || !IS_POSIX_TIMER(timerid)) 1548 return SET_ERROR(EINVAL); 1549 itimer_lock(); 1550 if ((it = pts->pts_timers[timerid]) == NULL) { 1551 itimer_unlock(); 1552 return SET_ERROR(EINVAL); 1553 } 1554 pt = container_of(it, struct ptimer, pt_itimer); 1555 *retval = pt->pt_poverruns; 1556 if (*retval >= DELAYTIMER_MAX) 1557 *retval = DELAYTIMER_MAX; 1558 itimer_unlock(); 1559 1560 return 0; 1561 } 1562 1563 /* 1564 * sys___getitimer50: 1565 * 1566 * System call to get the time remaining before a BSD timer fires. 1567 */ 1568 int 1569 sys___getitimer50(struct lwp *l, const struct sys___getitimer50_args *uap, 1570 register_t *retval) 1571 { 1572 /* { 1573 syscallarg(int) which; 1574 syscallarg(struct itimerval *) itv; 1575 } */ 1576 struct proc *p = l->l_proc; 1577 struct itimerval aitv; 1578 int error; 1579 1580 memset(&aitv, 0, sizeof(aitv)); 1581 error = dogetitimer(p, SCARG(uap, which), &aitv); 1582 if (error) 1583 return error; 1584 return copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)); 1585 } 1586 1587 int 1588 dogetitimer(struct proc *p, int which, struct itimerval *itvp) 1589 { 1590 struct ptimers *pts; 1591 struct itimer *it; 1592 struct itimerspec its; 1593 1594 if (!IS_ITIMER(which)) 1595 return SET_ERROR(EINVAL); 1596 1597 itimer_lock(); 1598 pts = p->p_timers; 1599 if (pts == NULL || (it = pts->pts_timers[which]) == NULL) { 1600 timerclear(&itvp->it_value); 1601 timerclear(&itvp->it_interval); 1602 } else { 1603 itimer_gettime(it, &its); 1604 TIMESPEC_TO_TIMEVAL(&itvp->it_value, &its.it_value); 1605 TIMESPEC_TO_TIMEVAL(&itvp->it_interval, &its.it_interval); 1606 } 1607 itimer_unlock(); 1608 1609 return 0; 1610 } 1611 1612 /* 1613 * sys___setitimer50: 1614 * 1615 * System call to set/arm a BSD timer. 1616 */ 1617 int 1618 sys___setitimer50(struct lwp *l, const struct sys___setitimer50_args *uap, 1619 register_t *retval) 1620 { 1621 /* { 1622 syscallarg(int) which; 1623 syscallarg(const struct itimerval *) itv; 1624 syscallarg(struct itimerval *) oitv; 1625 } */ 1626 struct proc *p = l->l_proc; 1627 int which = SCARG(uap, which); 1628 struct sys___getitimer50_args getargs; 1629 const struct itimerval *itvp; 1630 struct itimerval aitv; 1631 int error; 1632 1633 itvp = SCARG(uap, itv); 1634 if (itvp && 1635 (error = copyin(itvp, &aitv, sizeof(struct itimerval))) != 0) 1636 return error; 1637 if (SCARG(uap, oitv) != NULL) { 1638 SCARG(&getargs, which) = which; 1639 SCARG(&getargs, itv) = SCARG(uap, oitv); 1640 if ((error = sys___getitimer50(l, &getargs, retval)) != 0) 1641 return error; 1642 } 1643 if (itvp == 0) 1644 return 0; 1645 1646 return dosetitimer(p, which, &aitv); 1647 } 1648 1649 int 1650 dosetitimer(struct proc *p, int which, struct itimerval *itvp) 1651 { 1652 struct timespec now; 1653 struct ptimers *pts; 1654 struct ptimer *spare; 1655 struct itimer *it; 1656 struct itlist *itl; 1657 int error; 1658 1659 if (!IS_ITIMER(which)) 1660 return SET_ERROR(EINVAL); 1661 if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval)) 1662 return SET_ERROR(EINVAL); 1663 1664 /* 1665 * Don't bother allocating data structures if the process just 1666 * wants to clear the timer. 1667 */ 1668 spare = NULL; 1669 pts = p->p_timers; 1670 retry: 1671 if (!timerisset(&itvp->it_value) && (pts == NULL || 1672 pts->pts_timers[which] == NULL)) 1673 return 0; 1674 if (pts == NULL) 1675 pts = ptimers_alloc(p); 1676 itimer_lock(); 1677 restart: 1678 it = pts->pts_timers[which]; 1679 if (it == NULL) { 1680 struct ptimer *pt; 1681 1682 if (spare == NULL) { 1683 itimer_unlock(); 1684 spare = kmem_zalloc(sizeof(*spare), KM_SLEEP); 1685 goto retry; 1686 } 1687 pt = spare; 1688 spare = NULL; 1689 1690 it = &pt->pt_itimer; 1691 pt->pt_ev.sigev_notify = SIGEV_SIGNAL; 1692 pt->pt_ev.sigev_value.sival_int = which; 1693 1694 switch (which) { 1695 case ITIMER_REAL: 1696 case ITIMER_MONOTONIC: 1697 itl = NULL; 1698 pt->pt_ev.sigev_signo = SIGALRM; 1699 break; 1700 case ITIMER_VIRTUAL: 1701 itl = &pts->pts_virtual; 1702 pt->pt_ev.sigev_signo = SIGVTALRM; 1703 break; 1704 case ITIMER_PROF: 1705 itl = &pts->pts_prof; 1706 pt->pt_ev.sigev_signo = SIGPROF; 1707 break; 1708 default: 1709 panic("%s: can't happen %d", __func__, which); 1710 } 1711 itimer_init(it, &ptimer_itimer_ops, which, itl); 1712 pt->pt_proc = p; 1713 pt->pt_entry = which; 1714 1715 pts->pts_timers[which] = it; 1716 } 1717 1718 TIMEVAL_TO_TIMESPEC(&itvp->it_value, &it->it_time.it_value); 1719 TIMEVAL_TO_TIMESPEC(&itvp->it_interval, &it->it_time.it_interval); 1720 1721 error = 0; 1722 if (timespecisset(&it->it_time.it_value)) { 1723 /* Convert to absolute time */ 1724 /* XXX need to wrap in splclock for timecounters case? */ 1725 switch (which) { 1726 case ITIMER_REAL: 1727 getnanotime(&now); 1728 if (!timespecaddok(&it->it_time.it_value, &now)) { 1729 error = SET_ERROR(EINVAL); 1730 goto out; 1731 } 1732 timespecadd(&it->it_time.it_value, &now, 1733 &it->it_time.it_value); 1734 break; 1735 case ITIMER_MONOTONIC: 1736 getnanouptime(&now); 1737 if (!timespecaddok(&it->it_time.it_value, &now)) { 1738 error = SET_ERROR(EINVAL); 1739 goto out; 1740 } 1741 timespecadd(&it->it_time.it_value, &now, 1742 &it->it_time.it_value); 1743 break; 1744 default: 1745 break; 1746 } 1747 } 1748 1749 error = itimer_settime(it); 1750 if (error == ERESTART) { 1751 KASSERT(!CLOCK_VIRTUAL_P(it->it_clockid)); 1752 goto restart; 1753 } 1754 KASSERT(error == 0); 1755 out: 1756 itimer_unlock(); 1757 if (spare != NULL) 1758 kmem_free(spare, sizeof(*spare)); 1759 1760 return error; 1761 } 1762 1763 /* 1764 * ptimer_tick: 1765 * 1766 * Called from hardclock() to decrement per-process virtual timers. 1767 */ 1768 void 1769 ptimer_tick(lwp_t *l, bool user) 1770 { 1771 struct ptimers *pts; 1772 struct itimer *it; 1773 proc_t *p; 1774 1775 p = l->l_proc; 1776 if (p->p_timers == NULL) 1777 return; 1778 1779 itimer_lock(); 1780 if ((pts = l->l_proc->p_timers) != NULL) { 1781 /* 1782 * Run current process's virtual and profile time, as needed. 1783 */ 1784 if (user && (it = LIST_FIRST(&pts->pts_virtual)) != NULL) 1785 if (itimer_decr(it, tick * 1000)) 1786 (*it->it_ops->ito_fire)(it); 1787 if ((it = LIST_FIRST(&pts->pts_prof)) != NULL) 1788 if (itimer_decr(it, tick * 1000)) 1789 (*it->it_ops->ito_fire)(it); 1790 } 1791 itimer_unlock(); 1792 } 1793 1794 /* 1795 * ptimer_intr: 1796 * 1797 * Software interrupt handler for processing per-process 1798 * timer expiration. 1799 */ 1800 static void 1801 ptimer_intr(void *cookie) 1802 { 1803 ksiginfo_t ksi; 1804 struct itimer *it; 1805 struct ptimer *pt; 1806 proc_t *p; 1807 1808 mutex_enter(&proc_lock); 1809 itimer_lock(); 1810 while ((pt = TAILQ_FIRST(&ptimer_queue)) != NULL) { 1811 it = &pt->pt_itimer; 1812 1813 TAILQ_REMOVE(&ptimer_queue, pt, pt_chain); 1814 KASSERT(pt->pt_queued); 1815 pt->pt_queued = false; 1816 1817 p = pt->pt_proc; 1818 if (p->p_timers == NULL) { 1819 /* Process is dying. */ 1820 continue; 1821 } 1822 if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) { 1823 continue; 1824 } 1825 if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) { 1826 if (it->it_overruns < INT_MAX) 1827 it->it_overruns++; 1828 continue; 1829 } 1830 1831 KSI_INIT(&ksi); 1832 ksi.ksi_signo = pt->pt_ev.sigev_signo; 1833 ksi.ksi_code = SI_TIMER; 1834 ksi.ksi_value = pt->pt_ev.sigev_value; 1835 pt->pt_poverruns = it->it_overruns; 1836 it->it_overruns = 0; 1837 itimer_unlock(); 1838 kpsignal(p, &ksi, NULL); 1839 itimer_lock(); 1840 } 1841 itimer_unlock(); 1842 mutex_exit(&proc_lock); 1843 } 1844
Indexes created Thu Sep 17 00:24:55 UTC 2026