kern_time.c revision 1.216
1/* $NetBSD: kern_time.c,v 1.216 2022/06/27 00:34:24 riastradh 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 1.216 2022/06/27 00:34:24 riastradh Exp $"); 66 67#include <sys/param.h> 68#include <sys/resourcevar.h> 69#include <sys/kernel.h> 70#include <sys/systm.h> 71#include <sys/proc.h> 72#include <sys/vnode.h> 73#include <sys/signalvar.h> 74#include <sys/syslog.h> 75#include <sys/timetc.h> 76#include <sys/timex.h> 77#include <sys/kauth.h> 78#include <sys/mount.h> 79#include <sys/syscallargs.h> 80#include <sys/cpu.h> 81 82kmutex_t itimer_mutex __cacheline_aligned; /* XXX static */ 83static struct itlist itimer_realtime_changed_notify; 84 85static void ptimer_intr(void *); 86static void *ptimer_sih __read_mostly; 87static TAILQ_HEAD(, ptimer) ptimer_queue; 88 89#define CLOCK_VIRTUAL_P(clockid) \ 90 ((clockid) == CLOCK_VIRTUAL || (clockid) == CLOCK_PROF) 91 92CTASSERT(ITIMER_REAL == CLOCK_REALTIME); 93CTASSERT(ITIMER_VIRTUAL == CLOCK_VIRTUAL); 94CTASSERT(ITIMER_PROF == CLOCK_PROF); 95CTASSERT(ITIMER_MONOTONIC == CLOCK_MONOTONIC); 96 97#define DELAYTIMER_MAX 32 98 99/* 100 * Initialize timekeeping. 101 */ 102void 103time_init(void) 104{ 105 106 mutex_init(&itimer_mutex, MUTEX_DEFAULT, IPL_SCHED); 107 LIST_INIT(&itimer_realtime_changed_notify); 108 109 TAILQ_INIT(&ptimer_queue); 110 ptimer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE, 111 ptimer_intr, NULL); 112} 113 114/* 115 * Check if the time will wrap if set to ts. 116 * 117 * ts - timespec describing the new time 118 * delta - the delta between the current time and ts 119 */ 120bool 121time_wraps(struct timespec *ts, struct timespec *delta) 122{ 123 124 /* 125 * Don't allow the time to be set forward so far it 126 * will wrap and become negative, thus allowing an 127 * attacker to bypass the next check below. The 128 * cutoff is 1 year before rollover occurs, so even 129 * if the attacker uses adjtime(2) to move the time 130 * past the cutoff, it will take a very long time 131 * to get to the wrap point. 132 */ 133 if ((ts->tv_sec > LLONG_MAX - 365*24*60*60) || 134 (delta->tv_sec < 0 || delta->tv_nsec < 0)) 135 return true; 136 137 return false; 138} 139 140/* 141 * itimer_lock: 142 * 143 * Acquire the interval timer data lock. 144 */ 145void 146itimer_lock(void) 147{ 148 mutex_spin_enter(&itimer_mutex); 149} 150 151/* 152 * itimer_unlock: 153 * 154 * Release the interval timer data lock. 155 */ 156void 157itimer_unlock(void) 158{ 159 mutex_spin_exit(&itimer_mutex); 160} 161 162/* 163 * itimer_lock_held: 164 * 165 * Check that the interval timer lock is held for diagnostic 166 * assertions. 167 */ 168inline bool __diagused 169itimer_lock_held(void) 170{ 171 return mutex_owned(&itimer_mutex); 172} 173 174/* 175 * Time of day and interval timer support. 176 * 177 * These routines provide the kernel entry points to get and set 178 * the time-of-day and per-process interval timers. Subroutines 179 * here provide support for adding and subtracting timeval structures 180 * and decrementing interval timers, optionally reloading the interval 181 * timers when they expire. 182 */ 183 184/* This function is used by clock_settime and settimeofday */ 185static int 186settime1(struct proc *p, const struct timespec *ts, bool check_kauth) 187{ 188 struct timespec delta, now; 189 190 /* 191 * The time being set to an unreasonable value will cause 192 * unreasonable system behaviour. 193 */ 194 if (ts->tv_sec < 0 || ts->tv_sec > (1LL << 36)) 195 return (EINVAL); 196 197 nanotime(&now); 198 timespecsub(ts, &now, &delta); 199 200 if (check_kauth && kauth_authorize_system(kauth_cred_get(), 201 KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, __UNCONST(ts), 202 &delta, KAUTH_ARG(check_kauth ? false : true)) != 0) { 203 return (EPERM); 204 } 205 206#ifdef notyet 207 if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */ 208 return (EPERM); 209 } 210#endif 211 212 tc_setclock(ts); 213 214 resettodr(); 215 216 /* 217 * Notify pending CLOCK_REALTIME timers about the real time change. 218 * There may be inactive timers on this list, but this happens 219 * comparatively less often than timers firing, and so it's better 220 * to put the extra checks here than to complicate the other code 221 * path. 222 */ 223 struct itimer *it; 224 itimer_lock(); 225 LIST_FOREACH(it, &itimer_realtime_changed_notify, it_rtchgq) { 226 KASSERT(it->it_ops->ito_realtime_changed != NULL); 227 if (timespecisset(&it->it_time.it_value)) { 228 (*it->it_ops->ito_realtime_changed)(it); 229 } 230 } 231 itimer_unlock(); 232 233 return (0); 234} 235 236int 237settime(struct proc *p, struct timespec *ts) 238{ 239 return (settime1(p, ts, true)); 240} 241 242/* ARGSUSED */ 243int 244sys___clock_gettime50(struct lwp *l, 245 const struct sys___clock_gettime50_args *uap, register_t *retval) 246{ 247 /* { 248 syscallarg(clockid_t) clock_id; 249 syscallarg(struct timespec *) tp; 250 } */ 251 int error; 252 struct timespec ats; 253 254 error = clock_gettime1(SCARG(uap, clock_id), &ats); 255 if (error != 0) 256 return error; 257 258 return copyout(&ats, SCARG(uap, tp), sizeof(ats)); 259} 260 261/* ARGSUSED */ 262int 263sys___clock_settime50(struct lwp *l, 264 const struct sys___clock_settime50_args *uap, register_t *retval) 265{ 266 /* { 267 syscallarg(clockid_t) clock_id; 268 syscallarg(const struct timespec *) tp; 269 } */ 270 int error; 271 struct timespec ats; 272 273 if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0) 274 return error; 275 276 return clock_settime1(l->l_proc, SCARG(uap, clock_id), &ats, true); 277} 278 279 280int 281clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp, 282 bool check_kauth) 283{ 284 int error; 285 286 if (tp->tv_nsec < 0 || tp->tv_nsec >= 1000000000L) 287 return EINVAL; 288 289 switch (clock_id) { 290 case CLOCK_REALTIME: 291 if ((error = settime1(p, tp, check_kauth)) != 0) 292 return (error); 293 break; 294 case CLOCK_MONOTONIC: 295 return (EINVAL); /* read-only clock */ 296 default: 297 return (EINVAL); 298 } 299 300 return 0; 301} 302 303int 304sys___clock_getres50(struct lwp *l, const struct sys___clock_getres50_args *uap, 305 register_t *retval) 306{ 307 /* { 308 syscallarg(clockid_t) clock_id; 309 syscallarg(struct timespec *) tp; 310 } */ 311 struct timespec ts; 312 int error; 313 314 if ((error = clock_getres1(SCARG(uap, clock_id), &ts)) != 0) 315 return error; 316 317 if (SCARG(uap, tp)) 318 error = copyout(&ts, SCARG(uap, tp), sizeof(ts)); 319 320 return error; 321} 322 323int 324clock_getres1(clockid_t clock_id, struct timespec *ts) 325{ 326 327 switch (clock_id) { 328 case CLOCK_REALTIME: 329 case CLOCK_MONOTONIC: 330 ts->tv_sec = 0; 331 if (tc_getfrequency() > 1000000000) 332 ts->tv_nsec = 1; 333 else 334 ts->tv_nsec = 1000000000 / tc_getfrequency(); 335 break; 336 default: 337 return EINVAL; 338 } 339 340 return 0; 341} 342 343/* ARGSUSED */ 344int 345sys___nanosleep50(struct lwp *l, const struct sys___nanosleep50_args *uap, 346 register_t *retval) 347{ 348 /* { 349 syscallarg(struct timespec *) rqtp; 350 syscallarg(struct timespec *) rmtp; 351 } */ 352 struct timespec rmt, rqt; 353 int error, error1; 354 355 error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec)); 356 if (error) 357 return (error); 358 359 error = nanosleep1(l, CLOCK_MONOTONIC, 0, &rqt, 360 SCARG(uap, rmtp) ? &rmt : NULL); 361 if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR)) 362 return error; 363 364 error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt)); 365 return error1 ? error1 : error; 366} 367 368/* ARGSUSED */ 369int 370sys_clock_nanosleep(struct lwp *l, const struct sys_clock_nanosleep_args *uap, 371 register_t *retval) 372{ 373 /* { 374 syscallarg(clockid_t) clock_id; 375 syscallarg(int) flags; 376 syscallarg(struct timespec *) rqtp; 377 syscallarg(struct timespec *) rmtp; 378 } */ 379 struct timespec rmt, rqt; 380 int error, error1; 381 382 error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec)); 383 if (error) 384 goto out; 385 386 error = nanosleep1(l, SCARG(uap, clock_id), SCARG(uap, flags), &rqt, 387 SCARG(uap, rmtp) ? &rmt : NULL); 388 if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR)) 389 goto out; 390 391 if ((SCARG(uap, flags) & TIMER_ABSTIME) == 0 && 392 (error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt))) != 0) 393 error = error1; 394out: 395 *retval = error; 396 return 0; 397} 398 399int 400nanosleep1(struct lwp *l, clockid_t clock_id, int flags, struct timespec *rqt, 401 struct timespec *rmt) 402{ 403 struct timespec rmtstart; 404 int error, timo; 405 406 if ((error = ts2timo(clock_id, flags, rqt, &timo, &rmtstart)) != 0) { 407 if (error == ETIMEDOUT) { 408 error = 0; 409 if (rmt != NULL) 410 rmt->tv_sec = rmt->tv_nsec = 0; 411 } 412 return error; 413 } 414 415 /* 416 * Avoid inadvertently sleeping forever 417 */ 418 if (timo == 0) 419 timo = 1; 420again: 421 error = kpause("nanoslp", true, timo, NULL); 422 if (error == EWOULDBLOCK) 423 error = 0; 424 if (rmt != NULL || error == 0) { 425 struct timespec rmtend; 426 struct timespec t0; 427 struct timespec *t; 428 int err; 429 430 err = clock_gettime1(clock_id, &rmtend); 431 if (err != 0) 432 return err; 433 434 t = (rmt != NULL) ? rmt : &t0; 435 if (flags & TIMER_ABSTIME) { 436 timespecsub(rqt, &rmtend, t); 437 } else { 438 if (timespeccmp(&rmtend, &rmtstart, <)) 439 timespecclear(t); /* clock wound back */ 440 else 441 timespecsub(&rmtend, &rmtstart, t); 442 if (timespeccmp(rqt, t, <)) 443 timespecclear(t); 444 else 445 timespecsub(rqt, t, t); 446 } 447 if (t->tv_sec < 0) 448 timespecclear(t); 449 if (error == 0) { 450 timo = tstohz(t); 451 if (timo > 0) 452 goto again; 453 } 454 } 455 456 if (error == ERESTART) 457 error = EINTR; 458 459 return error; 460} 461 462int 463sys_clock_getcpuclockid2(struct lwp *l, 464 const struct sys_clock_getcpuclockid2_args *uap, 465 register_t *retval) 466{ 467 /* { 468 syscallarg(idtype_t idtype; 469 syscallarg(id_t id); 470 syscallarg(clockid_t *)clock_id; 471 } */ 472 pid_t pid; 473 lwpid_t lid; 474 clockid_t clock_id; 475 id_t id = SCARG(uap, id); 476 477 switch (SCARG(uap, idtype)) { 478 case P_PID: 479 pid = id == 0 ? l->l_proc->p_pid : id; 480 clock_id = CLOCK_PROCESS_CPUTIME_ID | pid; 481 break; 482 case P_LWPID: 483 lid = id == 0 ? l->l_lid : id; 484 clock_id = CLOCK_THREAD_CPUTIME_ID | lid; 485 break; 486 default: 487 return EINVAL; 488 } 489 return copyout(&clock_id, SCARG(uap, clock_id), sizeof(clock_id)); 490} 491 492/* ARGSUSED */ 493int 494sys___gettimeofday50(struct lwp *l, const struct sys___gettimeofday50_args *uap, 495 register_t *retval) 496{ 497 /* { 498 syscallarg(struct timeval *) tp; 499 syscallarg(void *) tzp; really "struct timezone *"; 500 } */ 501 struct timeval atv; 502 int error = 0; 503 struct timezone tzfake; 504 505 if (SCARG(uap, tp)) { 506 memset(&atv, 0, sizeof(atv)); 507 microtime(&atv); 508 error = copyout(&atv, SCARG(uap, tp), sizeof(atv)); 509 if (error) 510 return (error); 511 } 512 if (SCARG(uap, tzp)) { 513 /* 514 * NetBSD has no kernel notion of time zone, so we just 515 * fake up a timezone struct and return it if demanded. 516 */ 517 tzfake.tz_minuteswest = 0; 518 tzfake.tz_dsttime = 0; 519 error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake)); 520 } 521 return (error); 522} 523 524/* ARGSUSED */ 525int 526sys___settimeofday50(struct lwp *l, const struct sys___settimeofday50_args *uap, 527 register_t *retval) 528{ 529 /* { 530 syscallarg(const struct timeval *) tv; 531 syscallarg(const void *) tzp; really "const struct timezone *"; 532 } */ 533 534 return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true); 535} 536 537int 538settimeofday1(const struct timeval *utv, bool userspace, 539 const void *utzp, struct lwp *l, bool check_kauth) 540{ 541 struct timeval atv; 542 struct timespec ts; 543 int error; 544 545 /* Verify all parameters before changing time. */ 546 547 /* 548 * NetBSD has no kernel notion of time zone, and only an 549 * obsolete program would try to set it, so we log a warning. 550 */ 551 if (utzp) 552 log(LOG_WARNING, "pid %d attempted to set the " 553 "(obsolete) kernel time zone\n", l->l_proc->p_pid); 554 555 if (utv == NULL) 556 return 0; 557 558 if (userspace) { 559 if ((error = copyin(utv, &atv, sizeof(atv))) != 0) 560 return error; 561 utv = &atv; 562 } 563 564 if (utv->tv_usec < 0 || utv->tv_usec >= 1000000) 565 return EINVAL; 566 567 TIMEVAL_TO_TIMESPEC(utv, &ts); 568 return settime1(l->l_proc, &ts, check_kauth); 569} 570 571int time_adjusted; /* set if an adjustment is made */ 572 573/* ARGSUSED */ 574int 575sys___adjtime50(struct lwp *l, const struct sys___adjtime50_args *uap, 576 register_t *retval) 577{ 578 /* { 579 syscallarg(const struct timeval *) delta; 580 syscallarg(struct timeval *) olddelta; 581 } */ 582 int error; 583 struct timeval atv, oldatv; 584 585 if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME, 586 KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0) 587 return error; 588 589 if (SCARG(uap, delta)) { 590 error = copyin(SCARG(uap, delta), &atv, 591 sizeof(*SCARG(uap, delta))); 592 if (error) 593 return (error); 594 } 595 adjtime1(SCARG(uap, delta) ? &atv : NULL, 596 SCARG(uap, olddelta) ? &oldatv : NULL, l->l_proc); 597 if (SCARG(uap, olddelta)) 598 error = copyout(&oldatv, SCARG(uap, olddelta), 599 sizeof(*SCARG(uap, olddelta))); 600 return error; 601} 602 603void 604adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p) 605{ 606 extern int64_t time_adjtime; /* in kern_ntptime.c */ 607 608 if (olddelta) { 609 memset(olddelta, 0, sizeof(*olddelta)); 610 mutex_spin_enter(&timecounter_lock); 611 olddelta->tv_sec = time_adjtime / 1000000; 612 olddelta->tv_usec = time_adjtime % 1000000; 613 if (olddelta->tv_usec < 0) { 614 olddelta->tv_usec += 1000000; 615 olddelta->tv_sec--; 616 } 617 mutex_spin_exit(&timecounter_lock); 618 } 619 620 if (delta) { 621 mutex_spin_enter(&timecounter_lock); 622 /* 623 * XXX This should maybe just report failure to 624 * userland for nonsense deltas. 625 */ 626 if (delta->tv_sec > INT64_MAX/1000000 - 1) { 627 time_adjtime = INT64_MAX; 628 } else if (delta->tv_sec < INT64_MIN/1000000 + 1) { 629 time_adjtime = INT64_MIN; 630 } else { 631 time_adjtime = delta->tv_sec * 1000000 632 + MAX(-999999, MIN(999999, delta->tv_usec)); 633 } 634 635 if (time_adjtime) { 636 /* We need to save the system time during shutdown */ 637 time_adjusted |= 1; 638 } 639 mutex_spin_exit(&timecounter_lock); 640 } 641} 642 643/* 644 * Interval timer support. 645 * 646 * The itimer_*() routines provide generic support for interval timers, 647 * both real (CLOCK_REALTIME, CLOCK_MONOTIME), and virtual (CLOCK_VIRTUAL, 648 * CLOCK_PROF). 649 * 650 * Real timers keep their deadline as an absolute time, and are fired 651 * by a callout. Virtual timers are kept as a linked-list of deltas, 652 * and are processed by hardclock(). 653 * 654 * Because the real time timer callout may be delayed in real time due 655 * to interrupt processing on the system, it is possible for the real 656 * time timeout routine (itimer_callout()) run past after its deadline. 657 * It does not suffice, therefore, to reload the real timer .it_value 658 * from the timer's .it_interval. Rather, we compute the next deadline 659 * in absolute time based on the current time and the .it_interval value, 660 * and report any overruns. 661 * 662 * Note that while the virtual timers are supported in a generic fashion 663 * here, they only (currently) make sense as per-process timers, and thus 664 * only really work for that case. 665 */ 666 667/* 668 * itimer_init: 669 * 670 * Initialize the common data for an interval timer. 671 */ 672void 673itimer_init(struct itimer * const it, const struct itimer_ops * const ops, 674 clockid_t const id, struct itlist * const itl) 675{ 676 677 KASSERT(itimer_lock_held()); 678 KASSERT(ops != NULL); 679 680 timespecclear(&it->it_time.it_value); 681 it->it_ops = ops; 682 it->it_clockid = id; 683 it->it_overruns = 0; 684 it->it_dying = false; 685 if (!CLOCK_VIRTUAL_P(id)) { 686 KASSERT(itl == NULL); 687 callout_init(&it->it_ch, CALLOUT_MPSAFE); 688 if (id == CLOCK_REALTIME && ops->ito_realtime_changed != NULL) { 689 LIST_INSERT_HEAD(&itimer_realtime_changed_notify, 690 it, it_rtchgq); 691 } 692 } else { 693 KASSERT(itl != NULL); 694 it->it_vlist = itl; 695 it->it_active = false; 696 } 697} 698 699/* 700 * itimer_poison: 701 * 702 * Poison an interval timer, preventing it from being scheduled 703 * or processed, in preparation for freeing the timer. 704 */ 705void 706itimer_poison(struct itimer * const it) 707{ 708 709 KASSERT(itimer_lock_held()); 710 711 it->it_dying = true; 712 713 /* 714 * For non-virtual timers, stop the callout, or wait for it to 715 * run if it has already fired. It cannot restart again after 716 * this point: the callout won't restart itself when dying, no 717 * other users holding the lock can restart it, and any other 718 * users waiting for callout_halt concurrently (itimer_settime) 719 * will restart from the top. 720 */ 721 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 722 callout_halt(&it->it_ch, &itimer_mutex); 723 if (it->it_clockid == CLOCK_REALTIME && 724 it->it_ops->ito_realtime_changed != NULL) { 725 LIST_REMOVE(it, it_rtchgq); 726 } 727 } 728} 729 730/* 731 * itimer_fini: 732 * 733 * Release resources used by an interval timer. 734 * 735 * N.B. itimer_lock must be held on entry, and is released on exit. 736 */ 737void 738itimer_fini(struct itimer * const it) 739{ 740 741 KASSERT(itimer_lock_held()); 742 743 /* All done with the global state. */ 744 itimer_unlock(); 745 746 /* Destroy the callout, if needed. */ 747 if (!CLOCK_VIRTUAL_P(it->it_clockid)) 748 callout_destroy(&it->it_ch); 749} 750 751/* 752 * itimer_decr: 753 * 754 * Decrement an interval timer by a specified number of nanoseconds, 755 * which must be less than a second, i.e. < 1000000000. If the timer 756 * expires, then reload it. In this case, carry over (nsec - old value) 757 * to reduce the value reloaded into the timer so that the timer does 758 * not drift. This routine assumes that it is called in a context where 759 * the timers on which it is operating cannot change in value. 760 * 761 * Returns true if the timer has expired. 762 */ 763static bool 764itimer_decr(struct itimer *it, int nsec) 765{ 766 struct itimerspec *itp; 767 int error __diagused; 768 769 KASSERT(itimer_lock_held()); 770 KASSERT(CLOCK_VIRTUAL_P(it->it_clockid)); 771 772 itp = &it->it_time; 773 if (itp->it_value.tv_nsec < nsec) { 774 if (itp->it_value.tv_sec == 0) { 775 /* expired, and already in next interval */ 776 nsec -= itp->it_value.tv_nsec; 777 goto expire; 778 } 779 itp->it_value.tv_nsec += 1000000000; 780 itp->it_value.tv_sec--; 781 } 782 itp->it_value.tv_nsec -= nsec; 783 nsec = 0; 784 if (timespecisset(&itp->it_value)) 785 return false; 786 /* expired, exactly at end of interval */ 787 expire: 788 if (timespecisset(&itp->it_interval)) { 789 itp->it_value = itp->it_interval; 790 itp->it_value.tv_nsec -= nsec; 791 if (itp->it_value.tv_nsec < 0) { 792 itp->it_value.tv_nsec += 1000000000; 793 itp->it_value.tv_sec--; 794 } 795 error = itimer_settime(it); 796 KASSERT(error == 0); /* virtual, never fails */ 797 } else 798 itp->it_value.tv_nsec = 0; /* sec is already 0 */ 799 return true; 800} 801 802static void itimer_callout(void *); 803 804/* 805 * itimer_arm_real: 806 * 807 * Arm a non-virtual timer. 808 */ 809static void 810itimer_arm_real(struct itimer * const it) 811{ 812 /* 813 * Don't need to check tshzto() return value, here. 814 * callout_reset() does it for us. 815 */ 816 callout_reset(&it->it_ch, 817 (it->it_clockid == CLOCK_MONOTONIC 818 ? tshztoup(&it->it_time.it_value) 819 : tshzto(&it->it_time.it_value)), 820 itimer_callout, it); 821} 822 823/* 824 * itimer_callout: 825 * 826 * Callout to expire a non-virtual timer. Queue it up for processing, 827 * and then reload, if it is configured to do so. 828 * 829 * N.B. A delay in processing this callout causes multiple 830 * SIGALRM calls to be compressed into one. 831 */ 832static void 833itimer_callout(void *arg) 834{ 835 uint64_t last_val, next_val, interval, now_ns; 836 struct timespec now, next; 837 struct itimer * const it = arg; 838 int backwards; 839 840 itimer_lock(); 841 (*it->it_ops->ito_fire)(it); 842 843 if (!timespecisset(&it->it_time.it_interval)) { 844 timespecclear(&it->it_time.it_value); 845 itimer_unlock(); 846 return; 847 } 848 849 if (it->it_clockid == CLOCK_MONOTONIC) { 850 getnanouptime(&now); 851 } else { 852 getnanotime(&now); 853 } 854 855 backwards = (timespeccmp(&it->it_time.it_value, &now, >)); 856 857 /* Nonnegative interval guaranteed by itimerfix. */ 858 KASSERT(it->it_time.it_interval.tv_sec >= 0); 859 KASSERT(it->it_time.it_interval.tv_nsec >= 0); 860 861 /* Handle the easy case of non-overflown timers first. */ 862 if (!backwards && 863 timespecaddok(&it->it_time.it_value, &it->it_time.it_interval)) { 864 timespecadd(&it->it_time.it_value, &it->it_time.it_interval, 865 &next); 866 it->it_time.it_value = next; 867 } else { 868 now_ns = timespec2ns(&now); 869 last_val = timespec2ns(&it->it_time.it_value); 870 interval = timespec2ns(&it->it_time.it_interval); 871 872 next_val = now_ns + 873 (now_ns - last_val + interval - 1) % interval; 874 875 if (backwards) 876 next_val += interval; 877 else 878 it->it_overruns += (now_ns - last_val) / interval; 879 880 it->it_time.it_value.tv_sec = next_val / 1000000000; 881 it->it_time.it_value.tv_nsec = next_val % 1000000000; 882 } 883 884 /* 885 * Reset the callout, if it's not going away. 886 */ 887 if (!it->it_dying) 888 itimer_arm_real(it); 889 itimer_unlock(); 890} 891 892/* 893 * itimer_settime: 894 * 895 * Set up the given interval timer. The value in it->it_time.it_value 896 * is taken to be an absolute time for CLOCK_REALTIME/CLOCK_MONOTONIC 897 * timers and a relative time for CLOCK_VIRTUAL/CLOCK_PROF timers. 898 * 899 * If the callout had already fired but not yet run, fails with 900 * ERESTART -- caller must restart from the top to look up a timer. 901 */ 902int 903itimer_settime(struct itimer *it) 904{ 905 struct itimer *itn, *pitn; 906 struct itlist *itl; 907 908 KASSERT(itimer_lock_held()); 909 910 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 911 /* 912 * Try to stop the callout. However, if it had already 913 * fired, we have to drop the lock to wait for it, so 914 * the world may have changed and pt may not be there 915 * any more. In that case, tell the caller to start 916 * over from the top. 917 */ 918 if (callout_halt(&it->it_ch, &itimer_mutex)) 919 return ERESTART; 920 921 /* Now we can touch it and start it up again. */ 922 if (timespecisset(&it->it_time.it_value)) 923 itimer_arm_real(it); 924 } else { 925 if (it->it_active) { 926 itn = LIST_NEXT(it, it_list); 927 LIST_REMOVE(it, it_list); 928 for ( ; itn; itn = LIST_NEXT(itn, it_list)) 929 timespecadd(&it->it_time.it_value, 930 &itn->it_time.it_value, 931 &itn->it_time.it_value); 932 } 933 if (timespecisset(&it->it_time.it_value)) { 934 itl = it->it_vlist; 935 for (itn = LIST_FIRST(itl), pitn = NULL; 936 itn && timespeccmp(&it->it_time.it_value, 937 &itn->it_time.it_value, >); 938 pitn = itn, itn = LIST_NEXT(itn, it_list)) 939 timespecsub(&it->it_time.it_value, 940 &itn->it_time.it_value, 941 &it->it_time.it_value); 942 943 if (pitn) 944 LIST_INSERT_AFTER(pitn, it, it_list); 945 else 946 LIST_INSERT_HEAD(itl, it, it_list); 947 948 for ( ; itn ; itn = LIST_NEXT(itn, it_list)) 949 timespecsub(&itn->it_time.it_value, 950 &it->it_time.it_value, 951 &itn->it_time.it_value); 952 953 it->it_active = true; 954 } else { 955 it->it_active = false; 956 } 957 } 958 959 /* Success! */ 960 return 0; 961} 962 963/* 964 * itimer_gettime: 965 * 966 * Return the remaining time of an interval timer. 967 */ 968void 969itimer_gettime(const struct itimer *it, struct itimerspec *aits) 970{ 971 struct timespec now; 972 struct itimer *itn; 973 974 KASSERT(itimer_lock_held()); 975 976 *aits = it->it_time; 977 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 978 /* 979 * Convert from absolute to relative time in .it_value 980 * part of real time timer. If time for real time 981 * timer has passed return 0, else return difference 982 * between current time and time for the timer to go 983 * off. 984 */ 985 if (timespecisset(&aits->it_value)) { 986 if (it->it_clockid == CLOCK_REALTIME) { 987 getnanotime(&now); 988 } else { /* CLOCK_MONOTONIC */ 989 getnanouptime(&now); 990 } 991 if (timespeccmp(&aits->it_value, &now, <)) 992 timespecclear(&aits->it_value); 993 else 994 timespecsub(&aits->it_value, &now, 995 &aits->it_value); 996 } 997 } else if (it->it_active) { 998 for (itn = LIST_FIRST(it->it_vlist); itn && itn != it; 999 itn = LIST_NEXT(itn, it_list)) 1000 timespecadd(&aits->it_value, 1001 &itn->it_time.it_value, &aits->it_value); 1002 KASSERT(itn != NULL); /* it should be findable on the list */ 1003 } else 1004 timespecclear(&aits->it_value); 1005} 1006 1007/* 1008 * Per-process timer support. 1009 * 1010 * Both the BSD getitimer() family and the POSIX timer_*() family of 1011 * routines are supported. 1012 * 1013 * All timers are kept in an array pointed to by p_timers, which is 1014 * allocated on demand - many processes don't use timers at all. The 1015 * first four elements in this array are reserved for the BSD timers: 1016 * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, element 1017 * 2 is ITIMER_PROF, and element 3 is ITIMER_MONOTONIC. The rest may be 1018 * allocated by the timer_create() syscall. 1019 * 1020 * These timers are a "sub-class" of interval timer. 1021 */ 1022 1023/* 1024 * ptimer_free: 1025 * 1026 * Free the per-process timer at the specified index. 1027 */ 1028static void 1029ptimer_free(struct ptimers *pts, int index) 1030{ 1031 struct itimer *it; 1032 struct ptimer *pt; 1033 1034 KASSERT(itimer_lock_held()); 1035 1036 it = pts->pts_timers[index]; 1037 pt = container_of(it, struct ptimer, pt_itimer); 1038 pts->pts_timers[index] = NULL; 1039 itimer_poison(it); 1040 1041 /* 1042 * Remove it from the queue to be signalled. Must be done 1043 * after itimer is poisoned, because we may have had to wait 1044 * for the callout to complete. 1045 */ 1046 if (pt->pt_queued) { 1047 TAILQ_REMOVE(&ptimer_queue, pt, pt_chain); 1048 pt->pt_queued = false; 1049 } 1050 1051 itimer_fini(it); /* releases itimer_lock */ 1052 kmem_free(pt, sizeof(*pt)); 1053} 1054 1055/* 1056 * ptimers_alloc: 1057 * 1058 * Allocate a ptimers for the specified process. 1059 */ 1060static struct ptimers * 1061ptimers_alloc(struct proc *p) 1062{ 1063 struct ptimers *pts; 1064 int i; 1065 1066 pts = kmem_alloc(sizeof(*pts), KM_SLEEP); 1067 LIST_INIT(&pts->pts_virtual); 1068 LIST_INIT(&pts->pts_prof); 1069 for (i = 0; i < TIMER_MAX; i++) 1070 pts->pts_timers[i] = NULL; 1071 itimer_lock(); 1072 if (p->p_timers == NULL) { 1073 p->p_timers = pts; 1074 itimer_unlock(); 1075 return pts; 1076 } 1077 itimer_unlock(); 1078 kmem_free(pts, sizeof(*pts)); 1079 return p->p_timers; 1080} 1081 1082/* 1083 * ptimers_free: 1084 * 1085 * Clean up the per-process timers. If "which" is set to TIMERS_ALL, 1086 * then clean up all timers and free all the data structures. If 1087 * "which" is set to TIMERS_POSIX, only clean up the timers allocated 1088 * by timer_create(), not the BSD setitimer() timers, and only free the 1089 * structure if none of those remain. 1090 * 1091 * This function is exported because it is needed in the exec and 1092 * exit code paths. 1093 */ 1094void 1095ptimers_free(struct proc *p, int which) 1096{ 1097 struct ptimers *pts; 1098 struct itimer *itn; 1099 struct timespec ts; 1100 int i; 1101 1102 if (p->p_timers == NULL) 1103 return; 1104 1105 pts = p->p_timers; 1106 itimer_lock(); 1107 if (which == TIMERS_ALL) { 1108 p->p_timers = NULL; 1109 i = 0; 1110 } else { 1111 timespecclear(&ts); 1112 for (itn = LIST_FIRST(&pts->pts_virtual); 1113 itn && itn != pts->pts_timers[ITIMER_VIRTUAL]; 1114 itn = LIST_NEXT(itn, it_list)) { 1115 KASSERT(itn->it_clockid == CLOCK_VIRTUAL); 1116 timespecadd(&ts, &itn->it_time.it_value, &ts); 1117 } 1118 LIST_FIRST(&pts->pts_virtual) = NULL; 1119 if (itn) { 1120 KASSERT(itn->it_clockid == CLOCK_VIRTUAL); 1121 timespecadd(&ts, &itn->it_time.it_value, 1122 &itn->it_time.it_value); 1123 LIST_INSERT_HEAD(&pts->pts_virtual, itn, it_list); 1124 } 1125 timespecclear(&ts); 1126 for (itn = LIST_FIRST(&pts->pts_prof); 1127 itn && itn != pts->pts_timers[ITIMER_PROF]; 1128 itn = LIST_NEXT(itn, it_list)) { 1129 KASSERT(itn->it_clockid == CLOCK_PROF); 1130 timespecadd(&ts, &itn->it_time.it_value, &ts); 1131 } 1132 LIST_FIRST(&pts->pts_prof) = NULL; 1133 if (itn) { 1134 KASSERT(itn->it_clockid == CLOCK_PROF); 1135 timespecadd(&ts, &itn->it_time.it_value, 1136 &itn->it_time.it_value); 1137 LIST_INSERT_HEAD(&pts->pts_prof, itn, it_list); 1138 } 1139 i = TIMER_MIN; 1140 } 1141 for ( ; i < TIMER_MAX; i++) { 1142 if (pts->pts_timers[i] != NULL) { 1143 /* Free the timer and release the lock. */ 1144 ptimer_free(pts, i); 1145 /* Reacquire the lock for the next one. */ 1146 itimer_lock(); 1147 } 1148 } 1149 if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL && 1150 pts->pts_timers[2] == NULL && pts->pts_timers[3] == NULL) { 1151 p->p_timers = NULL; 1152 itimer_unlock(); 1153 kmem_free(pts, sizeof(*pts)); 1154 } else 1155 itimer_unlock(); 1156} 1157 1158/* 1159 * ptimer_fire: 1160 * 1161 * Fire a per-process timer. 1162 */ 1163static void 1164ptimer_fire(struct itimer *it) 1165{ 1166 struct ptimer *pt = container_of(it, struct ptimer, pt_itimer); 1167 1168 KASSERT(itimer_lock_held()); 1169 1170 /* 1171 * XXX Can overrun, but we don't do signal queueing yet, anyway. 1172 * XXX Relying on the clock interrupt is stupid. 1173 */ 1174 if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) { 1175 return; 1176 } 1177 1178 if (!pt->pt_queued) { 1179 TAILQ_INSERT_TAIL(&ptimer_queue, pt, pt_chain); 1180 pt->pt_queued = true; 1181 softint_schedule(ptimer_sih); 1182 } 1183} 1184 1185/* 1186 * Operations vector for per-process timers (BSD and POSIX). 1187 */ 1188static const struct itimer_ops ptimer_itimer_ops = { 1189 .ito_fire = ptimer_fire, 1190}; 1191 1192/* 1193 * sys_timer_create: 1194 * 1195 * System call to create a POSIX timer. 1196 */ 1197int 1198sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap, 1199 register_t *retval) 1200{ 1201 /* { 1202 syscallarg(clockid_t) clock_id; 1203 syscallarg(struct sigevent *) evp; 1204 syscallarg(timer_t *) timerid; 1205 } */ 1206 1207 return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id), 1208 SCARG(uap, evp), copyin, l); 1209} 1210 1211int 1212timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp, 1213 copyin_t fetch_event, struct lwp *l) 1214{ 1215 int error; 1216 timer_t timerid; 1217 struct itlist *itl; 1218 struct ptimers *pts; 1219 struct ptimer *pt; 1220 struct proc *p; 1221 1222 p = l->l_proc; 1223 1224 if ((u_int)id > CLOCK_MONOTONIC) 1225 return (EINVAL); 1226 1227 if ((pts = p->p_timers) == NULL) 1228 pts = ptimers_alloc(p); 1229 1230 pt = kmem_zalloc(sizeof(*pt), KM_SLEEP); 1231 if (evp != NULL) { 1232 if (((error = 1233 (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) || 1234 ((pt->pt_ev.sigev_notify < SIGEV_NONE) || 1235 (pt->pt_ev.sigev_notify > SIGEV_SA)) || 1236 (pt->pt_ev.sigev_notify == SIGEV_SIGNAL && 1237 (pt->pt_ev.sigev_signo <= 0 || 1238 pt->pt_ev.sigev_signo >= NSIG))) { 1239 kmem_free(pt, sizeof(*pt)); 1240 return (error ? error : EINVAL); 1241 } 1242 } 1243 1244 /* Find a free timer slot, skipping those reserved for setitimer(). */ 1245 itimer_lock(); 1246 for (timerid = TIMER_MIN; timerid < TIMER_MAX; timerid++) 1247 if (pts->pts_timers[timerid] == NULL) 1248 break; 1249 if (timerid == TIMER_MAX) { 1250 itimer_unlock(); 1251 kmem_free(pt, sizeof(*pt)); 1252 return EAGAIN; 1253 } 1254 if (evp == NULL) { 1255 pt->pt_ev.sigev_notify = SIGEV_SIGNAL; 1256 switch (id) { 1257 case CLOCK_REALTIME: 1258 case CLOCK_MONOTONIC: 1259 pt->pt_ev.sigev_signo = SIGALRM; 1260 break; 1261 case CLOCK_VIRTUAL: 1262 pt->pt_ev.sigev_signo = SIGVTALRM; 1263 break; 1264 case CLOCK_PROF: 1265 pt->pt_ev.sigev_signo = SIGPROF; 1266 break; 1267 } 1268 pt->pt_ev.sigev_value.sival_int = timerid; 1269 } 1270 1271 switch (id) { 1272 case CLOCK_VIRTUAL: 1273 itl = &pts->pts_virtual; 1274 break; 1275 case CLOCK_PROF: 1276 itl = &pts->pts_prof; 1277 break; 1278 default: 1279 itl = NULL; 1280 } 1281 1282 itimer_init(&pt->pt_itimer, &ptimer_itimer_ops, id, itl); 1283 pt->pt_proc = p; 1284 pt->pt_poverruns = 0; 1285 pt->pt_entry = timerid; 1286 pt->pt_queued = false; 1287 1288 pts->pts_timers[timerid] = &pt->pt_itimer; 1289 itimer_unlock(); 1290 1291 return copyout(&timerid, tid, sizeof(timerid)); 1292} 1293 1294/* 1295 * sys_timer_delete: 1296 * 1297 * System call to delete a POSIX timer. 1298 */ 1299int 1300sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap, 1301 register_t *retval) 1302{ 1303 /* { 1304 syscallarg(timer_t) timerid; 1305 } */ 1306 struct proc *p = l->l_proc; 1307 timer_t timerid; 1308 struct ptimers *pts; 1309 struct itimer *it, *itn; 1310 1311 timerid = SCARG(uap, timerid); 1312 pts = p->p_timers; 1313 1314 if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX) 1315 return (EINVAL); 1316 1317 itimer_lock(); 1318 if ((it = pts->pts_timers[timerid]) == NULL) { 1319 itimer_unlock(); 1320 return (EINVAL); 1321 } 1322 1323 if (CLOCK_VIRTUAL_P(it->it_clockid)) { 1324 if (it->it_active) { 1325 itn = LIST_NEXT(it, it_list); 1326 LIST_REMOVE(it, it_list); 1327 for ( ; itn; itn = LIST_NEXT(itn, it_list)) 1328 timespecadd(&it->it_time.it_value, 1329 &itn->it_time.it_value, 1330 &itn->it_time.it_value); 1331 it->it_active = false; 1332 } 1333 } 1334 1335 /* Free the timer and release the lock. */ 1336 ptimer_free(pts, timerid); 1337 1338 return (0); 1339} 1340 1341/* 1342 * sys___timer_settime50: 1343 * 1344 * System call to set/arm a POSIX timer. 1345 */ 1346int 1347sys___timer_settime50(struct lwp *l, 1348 const struct sys___timer_settime50_args *uap, 1349 register_t *retval) 1350{ 1351 /* { 1352 syscallarg(timer_t) timerid; 1353 syscallarg(int) flags; 1354 syscallarg(const struct itimerspec *) value; 1355 syscallarg(struct itimerspec *) ovalue; 1356 } */ 1357 int error; 1358 struct itimerspec value, ovalue, *ovp = NULL; 1359 1360 if ((error = copyin(SCARG(uap, value), &value, 1361 sizeof(struct itimerspec))) != 0) 1362 return (error); 1363 1364 if (SCARG(uap, ovalue)) 1365 ovp = &ovalue; 1366 1367 if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp, 1368 SCARG(uap, flags), l->l_proc)) != 0) 1369 return error; 1370 1371 if (ovp) 1372 return copyout(&ovalue, SCARG(uap, ovalue), 1373 sizeof(struct itimerspec)); 1374 return 0; 1375} 1376 1377int 1378dotimer_settime(int timerid, struct itimerspec *value, 1379 struct itimerspec *ovalue, int flags, struct proc *p) 1380{ 1381 struct timespec now; 1382 struct itimerspec val, oval; 1383 struct ptimers *pts; 1384 struct itimer *it; 1385 int error; 1386 1387 pts = p->p_timers; 1388 1389 if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX) 1390 return EINVAL; 1391 val = *value; 1392 if ((error = itimespecfix(&val.it_value)) != 0 || 1393 (error = itimespecfix(&val.it_interval)) != 0) 1394 return error; 1395 1396 itimer_lock(); 1397 restart: 1398 if ((it = pts->pts_timers[timerid]) == NULL) { 1399 itimer_unlock(); 1400 return EINVAL; 1401 } 1402 1403 oval = it->it_time; 1404 it->it_time = val; 1405 1406 /* 1407 * If we've been passed a relative time for a realtime timer, 1408 * convert it to absolute; if an absolute time for a virtual 1409 * timer, convert it to relative and make sure we don't set it 1410 * to zero, which would cancel the timer, or let it go 1411 * negative, which would confuse the comparison tests. 1412 */ 1413 if (timespecisset(&it->it_time.it_value)) { 1414 if (!CLOCK_VIRTUAL_P(it->it_clockid)) { 1415 if ((flags & TIMER_ABSTIME) == 0) { 1416 if (it->it_clockid == CLOCK_REALTIME) { 1417 getnanotime(&now); 1418 } else { /* CLOCK_MONOTONIC */ 1419 getnanouptime(&now); 1420 } 1421 timespecadd(&it->it_time.it_value, &now, 1422 &it->it_time.it_value); 1423 } 1424 } else { 1425 if ((flags & TIMER_ABSTIME) != 0) { 1426 getnanotime(&now); 1427 timespecsub(&it->it_time.it_value, &now, 1428 &it->it_time.it_value); 1429 if (!timespecisset(&it->it_time.it_value) || 1430 it->it_time.it_value.tv_sec < 0) { 1431 it->it_time.it_value.tv_sec = 0; 1432 it->it_time.it_value.tv_nsec = 1; 1433 } 1434 } 1435 } 1436 } 1437 1438 error = itimer_settime(it); 1439 if (error == ERESTART) { 1440 KASSERT(!CLOCK_VIRTUAL_P(it->it_clockid)); 1441 goto restart; 1442 } 1443 KASSERT(error == 0); 1444 itimer_unlock(); 1445 1446 if (ovalue) 1447 *ovalue = oval; 1448 1449 return (0); 1450} 1451 1452/* 1453 * sys___timer_gettime50: 1454 * 1455 * System call to return the time remaining until a POSIX timer fires. 1456 */ 1457int 1458sys___timer_gettime50(struct lwp *l, 1459 const struct sys___timer_gettime50_args *uap, register_t *retval) 1460{ 1461 /* { 1462 syscallarg(timer_t) timerid; 1463 syscallarg(struct itimerspec *) value; 1464 } */ 1465 struct itimerspec its; 1466 int error; 1467 1468 if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc, 1469 &its)) != 0) 1470 return error; 1471 1472 return copyout(&its, SCARG(uap, value), sizeof(its)); 1473} 1474 1475int 1476dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its) 1477{ 1478 struct itimer *it; 1479 struct ptimers *pts; 1480 1481 pts = p->p_timers; 1482 if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX) 1483 return (EINVAL); 1484 itimer_lock(); 1485 if ((it = pts->pts_timers[timerid]) == NULL) { 1486 itimer_unlock(); 1487 return (EINVAL); 1488 } 1489 itimer_gettime(it, its); 1490 itimer_unlock(); 1491 1492 return 0; 1493} 1494 1495/* 1496 * sys_timer_getoverrun: 1497 * 1498 * System call to return the number of times a POSIX timer has 1499 * expired while a notification was already pending. The counter 1500 * is reset when a timer expires and a notification can be posted. 1501 */ 1502int 1503sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap, 1504 register_t *retval) 1505{ 1506 /* { 1507 syscallarg(timer_t) timerid; 1508 } */ 1509 struct proc *p = l->l_proc; 1510 struct ptimers *pts; 1511 int timerid; 1512 struct itimer *it; 1513 struct ptimer *pt; 1514 1515 timerid = SCARG(uap, timerid); 1516 1517 pts = p->p_timers; 1518 if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX) 1519 return (EINVAL); 1520 itimer_lock(); 1521 if ((it = pts->pts_timers[timerid]) == NULL) { 1522 itimer_unlock(); 1523 return (EINVAL); 1524 } 1525 pt = container_of(it, struct ptimer, pt_itimer); 1526 *retval = pt->pt_poverruns; 1527 if (*retval >= DELAYTIMER_MAX) 1528 *retval = DELAYTIMER_MAX; 1529 itimer_unlock(); 1530 1531 return (0); 1532} 1533 1534/* 1535 * sys___getitimer50: 1536 * 1537 * System call to get the time remaining before a BSD timer fires. 1538 */ 1539int 1540sys___getitimer50(struct lwp *l, const struct sys___getitimer50_args *uap, 1541 register_t *retval) 1542{ 1543 /* { 1544 syscallarg(int) which; 1545 syscallarg(struct itimerval *) itv; 1546 } */ 1547 struct proc *p = l->l_proc; 1548 struct itimerval aitv; 1549 int error; 1550 1551 memset(&aitv, 0, sizeof(aitv)); 1552 error = dogetitimer(p, SCARG(uap, which), &aitv); 1553 if (error) 1554 return error; 1555 return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval))); 1556} 1557 1558int 1559dogetitimer(struct proc *p, int which, struct itimerval *itvp) 1560{ 1561 struct ptimers *pts; 1562 struct itimer *it; 1563 struct itimerspec its; 1564 1565 if ((u_int)which > ITIMER_MONOTONIC) 1566 return (EINVAL); 1567 1568 itimer_lock(); 1569 pts = p->p_timers; 1570 if (pts == NULL || (it = pts->pts_timers[which]) == NULL) { 1571 timerclear(&itvp->it_value); 1572 timerclear(&itvp->it_interval); 1573 } else { 1574 itimer_gettime(it, &its); 1575 TIMESPEC_TO_TIMEVAL(&itvp->it_value, &its.it_value); 1576 TIMESPEC_TO_TIMEVAL(&itvp->it_interval, &its.it_interval); 1577 } 1578 itimer_unlock(); 1579 1580 return 0; 1581} 1582 1583/* 1584 * sys___setitimer50: 1585 * 1586 * System call to set/arm a BSD timer. 1587 */ 1588int 1589sys___setitimer50(struct lwp *l, const struct sys___setitimer50_args *uap, 1590 register_t *retval) 1591{ 1592 /* { 1593 syscallarg(int) which; 1594 syscallarg(const struct itimerval *) itv; 1595 syscallarg(struct itimerval *) oitv; 1596 } */ 1597 struct proc *p = l->l_proc; 1598 int which = SCARG(uap, which); 1599 struct sys___getitimer50_args getargs; 1600 const struct itimerval *itvp; 1601 struct itimerval aitv; 1602 int error; 1603 1604 itvp = SCARG(uap, itv); 1605 if (itvp && 1606 (error = copyin(itvp, &aitv, sizeof(struct itimerval))) != 0) 1607 return (error); 1608 if (SCARG(uap, oitv) != NULL) { 1609 SCARG(&getargs, which) = which; 1610 SCARG(&getargs, itv) = SCARG(uap, oitv); 1611 if ((error = sys___getitimer50(l, &getargs, retval)) != 0) 1612 return (error); 1613 } 1614 if (itvp == 0) 1615 return (0); 1616 1617 return dosetitimer(p, which, &aitv); 1618} 1619 1620int 1621dosetitimer(struct proc *p, int which, struct itimerval *itvp) 1622{ 1623 struct timespec now; 1624 struct ptimers *pts; 1625 struct ptimer *spare; 1626 struct itimer *it; 1627 struct itlist *itl; 1628 int error; 1629 1630 if ((u_int)which > ITIMER_MONOTONIC) 1631 return (EINVAL); 1632 if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval)) 1633 return (EINVAL); 1634 1635 /* 1636 * Don't bother allocating data structures if the process just 1637 * wants to clear the timer. 1638 */ 1639 spare = NULL; 1640 pts = p->p_timers; 1641 retry: 1642 if (!timerisset(&itvp->it_value) && (pts == NULL || 1643 pts->pts_timers[which] == NULL)) 1644 return (0); 1645 if (pts == NULL) 1646 pts = ptimers_alloc(p); 1647 itimer_lock(); 1648 restart: 1649 it = pts->pts_timers[which]; 1650 if (it == NULL) { 1651 struct ptimer *pt; 1652 1653 if (spare == NULL) { 1654 itimer_unlock(); 1655 spare = kmem_zalloc(sizeof(*spare), KM_SLEEP); 1656 goto retry; 1657 } 1658 pt = spare; 1659 spare = NULL; 1660 1661 it = &pt->pt_itimer; 1662 pt->pt_ev.sigev_notify = SIGEV_SIGNAL; 1663 pt->pt_ev.sigev_value.sival_int = which; 1664 1665 switch (which) { 1666 case ITIMER_REAL: 1667 case ITIMER_MONOTONIC: 1668 itl = NULL; 1669 pt->pt_ev.sigev_signo = SIGALRM; 1670 break; 1671 case ITIMER_VIRTUAL: 1672 itl = &pts->pts_virtual; 1673 pt->pt_ev.sigev_signo = SIGVTALRM; 1674 break; 1675 case ITIMER_PROF: 1676 itl = &pts->pts_prof; 1677 pt->pt_ev.sigev_signo = SIGPROF; 1678 break; 1679 default: 1680 panic("%s: can't happen %d", __func__, which); 1681 } 1682 itimer_init(it, &ptimer_itimer_ops, which, itl); 1683 pt->pt_proc = p; 1684 pt->pt_entry = which; 1685 1686 pts->pts_timers[which] = it; 1687 } 1688 1689 TIMEVAL_TO_TIMESPEC(&itvp->it_value, &it->it_time.it_value); 1690 TIMEVAL_TO_TIMESPEC(&itvp->it_interval, &it->it_time.it_interval); 1691 1692 error = 0; 1693 if (timespecisset(&it->it_time.it_value)) { 1694 /* Convert to absolute time */ 1695 /* XXX need to wrap in splclock for timecounters case? */ 1696 switch (which) { 1697 case ITIMER_REAL: 1698 getnanotime(&now); 1699 if (!timespecaddok(&it->it_time.it_value, &now)) { 1700 error = EINVAL; 1701 goto out; 1702 } 1703 timespecadd(&it->it_time.it_value, &now, 1704 &it->it_time.it_value); 1705 break; 1706 case ITIMER_MONOTONIC: 1707 getnanouptime(&now); 1708 if (!timespecaddok(&it->it_time.it_value, &now)) { 1709 error = EINVAL; 1710 goto out; 1711 } 1712 timespecadd(&it->it_time.it_value, &now, 1713 &it->it_time.it_value); 1714 break; 1715 default: 1716 break; 1717 } 1718 } 1719 1720 error = itimer_settime(it); 1721 if (error == ERESTART) { 1722 KASSERT(!CLOCK_VIRTUAL_P(it->it_clockid)); 1723 goto restart; 1724 } 1725 KASSERT(error == 0); 1726out: 1727 itimer_unlock(); 1728 if (spare != NULL) 1729 kmem_free(spare, sizeof(*spare)); 1730 1731 return error; 1732} 1733 1734/* 1735 * ptimer_tick: 1736 * 1737 * Called from hardclock() to decrement per-process virtual timers. 1738 */ 1739void 1740ptimer_tick(lwp_t *l, bool user) 1741{ 1742 struct ptimers *pts; 1743 struct itimer *it; 1744 proc_t *p; 1745 1746 p = l->l_proc; 1747 if (p->p_timers == NULL) 1748 return; 1749 1750 itimer_lock(); 1751 if ((pts = l->l_proc->p_timers) != NULL) { 1752 /* 1753 * Run current process's virtual and profile time, as needed. 1754 */ 1755 if (user && (it = LIST_FIRST(&pts->pts_virtual)) != NULL) 1756 if (itimer_decr(it, tick * 1000)) 1757 (*it->it_ops->ito_fire)(it); 1758 if ((it = LIST_FIRST(&pts->pts_prof)) != NULL) 1759 if (itimer_decr(it, tick * 1000)) 1760 (*it->it_ops->ito_fire)(it); 1761 } 1762 itimer_unlock(); 1763} 1764 1765/* 1766 * ptimer_intr: 1767 * 1768 * Software interrupt handler for processing per-process 1769 * timer expiration. 1770 */ 1771static void 1772ptimer_intr(void *cookie) 1773{ 1774 ksiginfo_t ksi; 1775 struct itimer *it; 1776 struct ptimer *pt; 1777 proc_t *p; 1778 1779 mutex_enter(&proc_lock); 1780 itimer_lock(); 1781 while ((pt = TAILQ_FIRST(&ptimer_queue)) != NULL) { 1782 it = &pt->pt_itimer; 1783 1784 TAILQ_REMOVE(&ptimer_queue, pt, pt_chain); 1785 KASSERT(pt->pt_queued); 1786 pt->pt_queued = false; 1787 1788 p = pt->pt_proc; 1789 if (p->p_timers == NULL) { 1790 /* Process is dying. */ 1791 continue; 1792 } 1793 if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) { 1794 continue; 1795 } 1796 if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) { 1797 it->it_overruns++; 1798 continue; 1799 } 1800 1801 KSI_INIT(&ksi); 1802 ksi.ksi_signo = pt->pt_ev.sigev_signo; 1803 ksi.ksi_code = SI_TIMER; 1804 ksi.ksi_value = pt->pt_ev.sigev_value; 1805 pt->pt_poverruns = it->it_overruns; 1806 it->it_overruns = 0; 1807 itimer_unlock(); 1808 kpsignal(p, &ksi, NULL); 1809 itimer_lock(); 1810 } 1811 itimer_unlock(); 1812 mutex_exit(&proc_lock); 1813} 1814