1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_ktrace.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/limits.h> 42 #include <sys/clock.h> 43 #include <sys/lock.h> 44 #include <sys/mutex.h> 45 #include <sys/sysproto.h> 46 #include <sys/eventhandler.h> 47 #include <sys/resourcevar.h> 48 #include <sys/signalvar.h> 49 #include <sys/kernel.h> 50 #include <sys/sleepqueue.h> 51 #include <sys/syscallsubr.h> 52 #include <sys/sysctl.h> 53 #include <sys/sysent.h> 54 #include <sys/priv.h> 55 #include <sys/proc.h> 56 #include <sys/posix4.h> 57 #include <sys/time.h> 58 #include <sys/timers.h> 59 #include <sys/timetc.h> 60 #include <sys/vnode.h> 61 #ifdef KTRACE 62 #include <sys/ktrace.h> 63 #endif 64 65 #include <vm/vm.h> 66 #include <vm/vm_extern.h> 67 68 #define MAX_CLOCKS (CLOCK_MONOTONIC+1) 69 #define CPUCLOCK_BIT 0x80000000 70 #define CPUCLOCK_PROCESS_BIT 0x40000000 71 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT)) 72 #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid)) 73 #define MAKE_PROCESS_CPUCLOCK(pid) \ 74 (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid)) 75 76 static struct kclock posix_clocks[MAX_CLOCKS]; 77 static uma_zone_t itimer_zone = NULL; 78 79 /* 80 * Time of day and interval timer support. 81 * 82 * These routines provide the kernel entry points to get and set 83 * the time-of-day and per-process interval timers. Subroutines 84 * here provide support for adding and subtracting timeval structures 85 * and decrementing interval timers, optionally reloading the interval 86 * timers when they expire. 87 */ 88 89 static int settime(struct thread *, struct timeval *); 90 static void timevalfix(struct timeval *); 91 static int user_clock_nanosleep(struct thread *td, clockid_t clock_id, 92 int flags, const struct timespec *ua_rqtp, 93 struct timespec *ua_rmtp); 94 95 static void itimer_start(void); 96 static int itimer_init(void *, int, int); 97 static void itimer_fini(void *, int); 98 static void itimer_enter(struct itimer *); 99 static void itimer_leave(struct itimer *); 100 static struct itimer *itimer_find(struct proc *, int); 101 static void itimers_alloc(struct proc *); 102 static void itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp); 103 static void itimers_event_hook_exit(void *arg, struct proc *p); 104 static int realtimer_create(struct itimer *); 105 static int realtimer_gettime(struct itimer *, struct itimerspec *); 106 static int realtimer_settime(struct itimer *, int, 107 struct itimerspec *, struct itimerspec *); 108 static int realtimer_delete(struct itimer *); 109 static void realtimer_clocktime(clockid_t, struct timespec *); 110 static void realtimer_expire(void *); 111 112 int register_posix_clock(int, struct kclock *); 113 void itimer_fire(struct itimer *it); 114 int itimespecfix(struct timespec *ts); 115 116 #define CLOCK_CALL(clock, call, arglist) \ 117 ((*posix_clocks[clock].call) arglist) 118 119 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL); 120 121 122 static int 123 settime(struct thread *td, struct timeval *tv) 124 { 125 struct timeval delta, tv1, tv2; 126 static struct timeval maxtime, laststep; 127 struct timespec ts; 128 129 microtime(&tv1); 130 delta = *tv; 131 timevalsub(&delta, &tv1); 132 133 /* 134 * If the system is secure, we do not allow the time to be 135 * set to a value earlier than 1 second less than the highest 136 * time we have yet seen. The worst a miscreant can do in 137 * this circumstance is "freeze" time. He couldn't go 138 * back to the past. 139 * 140 * We similarly do not allow the clock to be stepped more 141 * than one second, nor more than once per second. This allows 142 * a miscreant to make the clock march double-time, but no worse. 143 */ 144 if (securelevel_gt(td->td_ucred, 1) != 0) { 145 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 146 /* 147 * Update maxtime to latest time we've seen. 148 */ 149 if (tv1.tv_sec > maxtime.tv_sec) 150 maxtime = tv1; 151 tv2 = *tv; 152 timevalsub(&tv2, &maxtime); 153 if (tv2.tv_sec < -1) { 154 tv->tv_sec = maxtime.tv_sec - 1; 155 printf("Time adjustment clamped to -1 second\n"); 156 } 157 } else { 158 if (tv1.tv_sec == laststep.tv_sec) 159 return (EPERM); 160 if (delta.tv_sec > 1) { 161 tv->tv_sec = tv1.tv_sec + 1; 162 printf("Time adjustment clamped to +1 second\n"); 163 } 164 laststep = *tv; 165 } 166 } 167 168 ts.tv_sec = tv->tv_sec; 169 ts.tv_nsec = tv->tv_usec * 1000; 170 tc_setclock(&ts); 171 resettodr(); 172 return (0); 173 } 174 175 #ifndef _SYS_SYSPROTO_H_ 176 struct clock_getcpuclockid2_args { 177 id_t id; 178 int which, 179 clockid_t *clock_id; 180 }; 181 #endif 182 /* ARGSUSED */ 183 int 184 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap) 185 { 186 clockid_t clk_id; 187 int error; 188 189 error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id); 190 if (error == 0) 191 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 192 return (error); 193 } 194 195 int 196 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which, 197 clockid_t *clk_id) 198 { 199 struct proc *p; 200 pid_t pid; 201 lwpid_t tid; 202 int error; 203 204 switch (which) { 205 case CPUCLOCK_WHICH_PID: 206 if (id != 0) { 207 error = pget(id, PGET_CANSEE | PGET_NOTID, &p); 208 if (error != 0) 209 return (error); 210 PROC_UNLOCK(p); 211 pid = id; 212 } else { 213 pid = td->td_proc->p_pid; 214 } 215 *clk_id = MAKE_PROCESS_CPUCLOCK(pid); 216 return (0); 217 case CPUCLOCK_WHICH_TID: 218 tid = id == 0 ? td->td_tid : id; 219 *clk_id = MAKE_THREAD_CPUCLOCK(tid); 220 return (0); 221 default: 222 return (EINVAL); 223 } 224 } 225 226 #ifndef _SYS_SYSPROTO_H_ 227 struct clock_gettime_args { 228 clockid_t clock_id; 229 struct timespec *tp; 230 }; 231 #endif 232 /* ARGSUSED */ 233 int 234 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap) 235 { 236 struct timespec ats; 237 int error; 238 239 error = kern_clock_gettime(td, uap->clock_id, &ats); 240 if (error == 0) 241 error = copyout(&ats, uap->tp, sizeof(ats)); 242 243 return (error); 244 } 245 246 static inline void 247 cputick2timespec(uint64_t runtime, struct timespec *ats) 248 { 249 runtime = cputick2usec(runtime); 250 ats->tv_sec = runtime / 1000000; 251 ats->tv_nsec = runtime % 1000000 * 1000; 252 } 253 254 static void 255 get_thread_cputime(struct thread *targettd, struct timespec *ats) 256 { 257 uint64_t runtime, curtime, switchtime; 258 259 if (targettd == NULL) { /* current thread */ 260 critical_enter(); 261 switchtime = PCPU_GET(switchtime); 262 curtime = cpu_ticks(); 263 runtime = curthread->td_runtime; 264 critical_exit(); 265 runtime += curtime - switchtime; 266 } else { 267 thread_lock(targettd); 268 runtime = targettd->td_runtime; 269 thread_unlock(targettd); 270 } 271 cputick2timespec(runtime, ats); 272 } 273 274 static void 275 get_process_cputime(struct proc *targetp, struct timespec *ats) 276 { 277 uint64_t runtime; 278 struct rusage ru; 279 280 PROC_STATLOCK(targetp); 281 rufetch(targetp, &ru); 282 runtime = targetp->p_rux.rux_runtime; 283 PROC_STATUNLOCK(targetp); 284 cputick2timespec(runtime, ats); 285 } 286 287 static int 288 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats) 289 { 290 struct proc *p, *p2; 291 struct thread *td2; 292 lwpid_t tid; 293 pid_t pid; 294 int error; 295 296 p = td->td_proc; 297 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) { 298 tid = clock_id & CPUCLOCK_ID_MASK; 299 td2 = tdfind(tid, p->p_pid); 300 if (td2 == NULL) 301 return (EINVAL); 302 get_thread_cputime(td2, ats); 303 PROC_UNLOCK(td2->td_proc); 304 } else { 305 pid = clock_id & CPUCLOCK_ID_MASK; 306 error = pget(pid, PGET_CANSEE, &p2); 307 if (error != 0) 308 return (EINVAL); 309 get_process_cputime(p2, ats); 310 PROC_UNLOCK(p2); 311 } 312 return (0); 313 } 314 315 int 316 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats) 317 { 318 struct timeval sys, user; 319 struct proc *p; 320 321 p = td->td_proc; 322 switch (clock_id) { 323 case CLOCK_REALTIME: /* Default to precise. */ 324 case CLOCK_REALTIME_PRECISE: 325 nanotime(ats); 326 break; 327 case CLOCK_REALTIME_FAST: 328 getnanotime(ats); 329 break; 330 case CLOCK_VIRTUAL: 331 PROC_LOCK(p); 332 PROC_STATLOCK(p); 333 calcru(p, &user, &sys); 334 PROC_STATUNLOCK(p); 335 PROC_UNLOCK(p); 336 TIMEVAL_TO_TIMESPEC(&user, ats); 337 break; 338 case CLOCK_PROF: 339 PROC_LOCK(p); 340 PROC_STATLOCK(p); 341 calcru(p, &user, &sys); 342 PROC_STATUNLOCK(p); 343 PROC_UNLOCK(p); 344 timevaladd(&user, &sys); 345 TIMEVAL_TO_TIMESPEC(&user, ats); 346 break; 347 case CLOCK_MONOTONIC: /* Default to precise. */ 348 case CLOCK_MONOTONIC_PRECISE: 349 case CLOCK_UPTIME: 350 case CLOCK_UPTIME_PRECISE: 351 nanouptime(ats); 352 break; 353 case CLOCK_UPTIME_FAST: 354 case CLOCK_MONOTONIC_FAST: 355 getnanouptime(ats); 356 break; 357 case CLOCK_SECOND: 358 ats->tv_sec = time_second; 359 ats->tv_nsec = 0; 360 break; 361 case CLOCK_THREAD_CPUTIME_ID: 362 get_thread_cputime(NULL, ats); 363 break; 364 case CLOCK_PROCESS_CPUTIME_ID: 365 PROC_LOCK(p); 366 get_process_cputime(p, ats); 367 PROC_UNLOCK(p); 368 break; 369 default: 370 if ((int)clock_id >= 0) 371 return (EINVAL); 372 return (get_cputime(td, clock_id, ats)); 373 } 374 return (0); 375 } 376 377 #ifndef _SYS_SYSPROTO_H_ 378 struct clock_settime_args { 379 clockid_t clock_id; 380 const struct timespec *tp; 381 }; 382 #endif 383 /* ARGSUSED */ 384 int 385 sys_clock_settime(struct thread *td, struct clock_settime_args *uap) 386 { 387 struct timespec ats; 388 int error; 389 390 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 391 return (error); 392 return (kern_clock_settime(td, uap->clock_id, &ats)); 393 } 394 395 static int allow_insane_settime = 0; 396 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN, 397 &allow_insane_settime, 0, 398 "do not perform possibly restrictive checks on settime(2) args"); 399 400 int 401 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats) 402 { 403 struct timeval atv; 404 int error; 405 406 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 407 return (error); 408 if (clock_id != CLOCK_REALTIME) 409 return (EINVAL); 410 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 || 411 ats->tv_sec < 0) 412 return (EINVAL); 413 if (!allow_insane_settime && ats->tv_sec > 8000ULL * 365 * 24 * 60 * 60) 414 return (EINVAL); 415 /* XXX Don't convert nsec->usec and back */ 416 TIMESPEC_TO_TIMEVAL(&atv, ats); 417 error = settime(td, &atv); 418 return (error); 419 } 420 421 #ifndef _SYS_SYSPROTO_H_ 422 struct clock_getres_args { 423 clockid_t clock_id; 424 struct timespec *tp; 425 }; 426 #endif 427 int 428 sys_clock_getres(struct thread *td, struct clock_getres_args *uap) 429 { 430 struct timespec ts; 431 int error; 432 433 if (uap->tp == NULL) 434 return (0); 435 436 error = kern_clock_getres(td, uap->clock_id, &ts); 437 if (error == 0) 438 error = copyout(&ts, uap->tp, sizeof(ts)); 439 return (error); 440 } 441 442 int 443 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts) 444 { 445 446 ts->tv_sec = 0; 447 switch (clock_id) { 448 case CLOCK_REALTIME: 449 case CLOCK_REALTIME_FAST: 450 case CLOCK_REALTIME_PRECISE: 451 case CLOCK_MONOTONIC: 452 case CLOCK_MONOTONIC_FAST: 453 case CLOCK_MONOTONIC_PRECISE: 454 case CLOCK_UPTIME: 455 case CLOCK_UPTIME_FAST: 456 case CLOCK_UPTIME_PRECISE: 457 /* 458 * Round up the result of the division cheaply by adding 1. 459 * Rounding up is especially important if rounding down 460 * would give 0. Perfect rounding is unimportant. 461 */ 462 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1; 463 break; 464 case CLOCK_VIRTUAL: 465 case CLOCK_PROF: 466 /* Accurately round up here because we can do so cheaply. */ 467 ts->tv_nsec = howmany(1000000000, hz); 468 break; 469 case CLOCK_SECOND: 470 ts->tv_sec = 1; 471 ts->tv_nsec = 0; 472 break; 473 case CLOCK_THREAD_CPUTIME_ID: 474 case CLOCK_PROCESS_CPUTIME_ID: 475 cputime: 476 /* sync with cputick2usec */ 477 ts->tv_nsec = 1000000 / cpu_tickrate(); 478 if (ts->tv_nsec == 0) 479 ts->tv_nsec = 1000; 480 break; 481 default: 482 if ((int)clock_id < 0) 483 goto cputime; 484 return (EINVAL); 485 } 486 return (0); 487 } 488 489 int 490 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt) 491 { 492 493 return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt, 494 rmt)); 495 } 496 497 static uint8_t nanowait[MAXCPU]; 498 499 int 500 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, 501 const struct timespec *rqt, struct timespec *rmt) 502 { 503 struct timespec ts, now; 504 sbintime_t sbt, sbtt, prec, tmp; 505 time_t over; 506 int error; 507 bool is_abs_real; 508 509 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 510 return (EINVAL); 511 if ((flags & ~TIMER_ABSTIME) != 0) 512 return (EINVAL); 513 switch (clock_id) { 514 case CLOCK_REALTIME: 515 case CLOCK_REALTIME_PRECISE: 516 case CLOCK_REALTIME_FAST: 517 case CLOCK_SECOND: 518 is_abs_real = (flags & TIMER_ABSTIME) != 0; 519 break; 520 case CLOCK_MONOTONIC: 521 case CLOCK_MONOTONIC_PRECISE: 522 case CLOCK_MONOTONIC_FAST: 523 case CLOCK_UPTIME: 524 case CLOCK_UPTIME_PRECISE: 525 case CLOCK_UPTIME_FAST: 526 is_abs_real = false; 527 break; 528 case CLOCK_VIRTUAL: 529 case CLOCK_PROF: 530 case CLOCK_PROCESS_CPUTIME_ID: 531 return (ENOTSUP); 532 case CLOCK_THREAD_CPUTIME_ID: 533 default: 534 return (EINVAL); 535 } 536 do { 537 ts = *rqt; 538 if ((flags & TIMER_ABSTIME) != 0) { 539 if (is_abs_real) 540 td->td_rtcgen = 541 atomic_load_acq_int(&rtc_generation); 542 error = kern_clock_gettime(td, clock_id, &now); 543 KASSERT(error == 0, ("kern_clock_gettime: %d", error)); 544 timespecsub(&ts, &now); 545 } 546 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) { 547 error = EWOULDBLOCK; 548 break; 549 } 550 if (ts.tv_sec > INT32_MAX / 2) { 551 over = ts.tv_sec - INT32_MAX / 2; 552 ts.tv_sec -= over; 553 } else 554 over = 0; 555 tmp = tstosbt(ts); 556 prec = tmp; 557 prec >>= tc_precexp; 558 if (TIMESEL(&sbt, tmp)) 559 sbt += tc_tick_sbt; 560 sbt += tmp; 561 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp", 562 sbt, prec, C_ABSOLUTE); 563 } while (error == 0 && is_abs_real && td->td_rtcgen == 0); 564 td->td_rtcgen = 0; 565 if (error != EWOULDBLOCK) { 566 TIMESEL(&sbtt, tmp); 567 if (sbtt >= sbt) 568 return (0); 569 if (error == ERESTART) 570 error = EINTR; 571 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) { 572 ts = sbttots(sbt - sbtt); 573 ts.tv_sec += over; 574 if (ts.tv_sec < 0) 575 timespecclear(&ts); 576 *rmt = ts; 577 } 578 return (error); 579 } 580 return (0); 581 } 582 583 #ifndef _SYS_SYSPROTO_H_ 584 struct nanosleep_args { 585 struct timespec *rqtp; 586 struct timespec *rmtp; 587 }; 588 #endif 589 /* ARGSUSED */ 590 int 591 sys_nanosleep(struct thread *td, struct nanosleep_args *uap) 592 { 593 594 return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, 595 uap->rqtp, uap->rmtp)); 596 } 597 598 #ifndef _SYS_SYSPROTO_H_ 599 struct clock_nanosleep_args { 600 clockid_t clock_id; 601 int flags; 602 struct timespec *rqtp; 603 struct timespec *rmtp; 604 }; 605 #endif 606 /* ARGSUSED */ 607 int 608 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap) 609 { 610 int error; 611 612 error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp, 613 uap->rmtp); 614 return (kern_posix_error(td, error)); 615 } 616 617 static int 618 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, 619 const struct timespec *ua_rqtp, struct timespec *ua_rmtp) 620 { 621 struct timespec rmt, rqt; 622 int error; 623 624 error = copyin(ua_rqtp, &rqt, sizeof(rqt)); 625 if (error) 626 return (error); 627 if (ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0 && 628 !useracc(ua_rmtp, sizeof(rmt), VM_PROT_WRITE)) 629 return (EFAULT); 630 error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt); 631 if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) { 632 int error2; 633 634 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt)); 635 if (error2) 636 error = error2; 637 } 638 return (error); 639 } 640 641 #ifndef _SYS_SYSPROTO_H_ 642 struct gettimeofday_args { 643 struct timeval *tp; 644 struct timezone *tzp; 645 }; 646 #endif 647 /* ARGSUSED */ 648 int 649 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap) 650 { 651 struct timeval atv; 652 struct timezone rtz; 653 int error = 0; 654 655 if (uap->tp) { 656 microtime(&atv); 657 error = copyout(&atv, uap->tp, sizeof (atv)); 658 } 659 if (error == 0 && uap->tzp != NULL) { 660 rtz.tz_minuteswest = tz_minuteswest; 661 rtz.tz_dsttime = tz_dsttime; 662 error = copyout(&rtz, uap->tzp, sizeof (rtz)); 663 } 664 return (error); 665 } 666 667 #ifndef _SYS_SYSPROTO_H_ 668 struct settimeofday_args { 669 struct timeval *tv; 670 struct timezone *tzp; 671 }; 672 #endif 673 /* ARGSUSED */ 674 int 675 sys_settimeofday(struct thread *td, struct settimeofday_args *uap) 676 { 677 struct timeval atv, *tvp; 678 struct timezone atz, *tzp; 679 int error; 680 681 if (uap->tv) { 682 error = copyin(uap->tv, &atv, sizeof(atv)); 683 if (error) 684 return (error); 685 tvp = &atv; 686 } else 687 tvp = NULL; 688 if (uap->tzp) { 689 error = copyin(uap->tzp, &atz, sizeof(atz)); 690 if (error) 691 return (error); 692 tzp = &atz; 693 } else 694 tzp = NULL; 695 return (kern_settimeofday(td, tvp, tzp)); 696 } 697 698 int 699 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp) 700 { 701 int error; 702 703 error = priv_check(td, PRIV_SETTIMEOFDAY); 704 if (error) 705 return (error); 706 /* Verify all parameters before changing time. */ 707 if (tv) { 708 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 || 709 tv->tv_sec < 0) 710 return (EINVAL); 711 error = settime(td, tv); 712 } 713 if (tzp && error == 0) { 714 tz_minuteswest = tzp->tz_minuteswest; 715 tz_dsttime = tzp->tz_dsttime; 716 } 717 return (error); 718 } 719 720 /* 721 * Get value of an interval timer. The process virtual and profiling virtual 722 * time timers are kept in the p_stats area, since they can be swapped out. 723 * These are kept internally in the way they are specified externally: in 724 * time until they expire. 725 * 726 * The real time interval timer is kept in the process table slot for the 727 * process, and its value (it_value) is kept as an absolute time rather than 728 * as a delta, so that it is easy to keep periodic real-time signals from 729 * drifting. 730 * 731 * Virtual time timers are processed in the hardclock() routine of 732 * kern_clock.c. The real time timer is processed by a timeout routine, 733 * called from the softclock() routine. Since a callout may be delayed in 734 * real time due to interrupt processing in the system, it is possible for 735 * the real time timeout routine (realitexpire, given below), to be delayed 736 * in real time past when it is supposed to occur. It does not suffice, 737 * therefore, to reload the real timer .it_value from the real time timers 738 * .it_interval. Rather, we compute the next time in absolute time the timer 739 * should go off. 740 */ 741 #ifndef _SYS_SYSPROTO_H_ 742 struct getitimer_args { 743 u_int which; 744 struct itimerval *itv; 745 }; 746 #endif 747 int 748 sys_getitimer(struct thread *td, struct getitimer_args *uap) 749 { 750 struct itimerval aitv; 751 int error; 752 753 error = kern_getitimer(td, uap->which, &aitv); 754 if (error != 0) 755 return (error); 756 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 757 } 758 759 int 760 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv) 761 { 762 struct proc *p = td->td_proc; 763 struct timeval ctv; 764 765 if (which > ITIMER_PROF) 766 return (EINVAL); 767 768 if (which == ITIMER_REAL) { 769 /* 770 * Convert from absolute to relative time in .it_value 771 * part of real time timer. If time for real time timer 772 * has passed return 0, else return difference between 773 * current time and time for the timer to go off. 774 */ 775 PROC_LOCK(p); 776 *aitv = p->p_realtimer; 777 PROC_UNLOCK(p); 778 if (timevalisset(&aitv->it_value)) { 779 microuptime(&ctv); 780 if (timevalcmp(&aitv->it_value, &ctv, <)) 781 timevalclear(&aitv->it_value); 782 else 783 timevalsub(&aitv->it_value, &ctv); 784 } 785 } else { 786 PROC_ITIMLOCK(p); 787 *aitv = p->p_stats->p_timer[which]; 788 PROC_ITIMUNLOCK(p); 789 } 790 #ifdef KTRACE 791 if (KTRPOINT(td, KTR_STRUCT)) 792 ktritimerval(aitv); 793 #endif 794 return (0); 795 } 796 797 #ifndef _SYS_SYSPROTO_H_ 798 struct setitimer_args { 799 u_int which; 800 struct itimerval *itv, *oitv; 801 }; 802 #endif 803 int 804 sys_setitimer(struct thread *td, struct setitimer_args *uap) 805 { 806 struct itimerval aitv, oitv; 807 int error; 808 809 if (uap->itv == NULL) { 810 uap->itv = uap->oitv; 811 return (sys_getitimer(td, (struct getitimer_args *)uap)); 812 } 813 814 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval)))) 815 return (error); 816 error = kern_setitimer(td, uap->which, &aitv, &oitv); 817 if (error != 0 || uap->oitv == NULL) 818 return (error); 819 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval))); 820 } 821 822 int 823 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv, 824 struct itimerval *oitv) 825 { 826 struct proc *p = td->td_proc; 827 struct timeval ctv; 828 sbintime_t sbt, pr; 829 830 if (aitv == NULL) 831 return (kern_getitimer(td, which, oitv)); 832 833 if (which > ITIMER_PROF) 834 return (EINVAL); 835 #ifdef KTRACE 836 if (KTRPOINT(td, KTR_STRUCT)) 837 ktritimerval(aitv); 838 #endif 839 if (itimerfix(&aitv->it_value) || 840 aitv->it_value.tv_sec > INT32_MAX / 2) 841 return (EINVAL); 842 if (!timevalisset(&aitv->it_value)) 843 timevalclear(&aitv->it_interval); 844 else if (itimerfix(&aitv->it_interval) || 845 aitv->it_interval.tv_sec > INT32_MAX / 2) 846 return (EINVAL); 847 848 if (which == ITIMER_REAL) { 849 PROC_LOCK(p); 850 if (timevalisset(&p->p_realtimer.it_value)) 851 callout_stop(&p->p_itcallout); 852 microuptime(&ctv); 853 if (timevalisset(&aitv->it_value)) { 854 pr = tvtosbt(aitv->it_value) >> tc_precexp; 855 timevaladd(&aitv->it_value, &ctv); 856 sbt = tvtosbt(aitv->it_value); 857 callout_reset_sbt(&p->p_itcallout, sbt, pr, 858 realitexpire, p, C_ABSOLUTE); 859 } 860 *oitv = p->p_realtimer; 861 p->p_realtimer = *aitv; 862 PROC_UNLOCK(p); 863 if (timevalisset(&oitv->it_value)) { 864 if (timevalcmp(&oitv->it_value, &ctv, <)) 865 timevalclear(&oitv->it_value); 866 else 867 timevalsub(&oitv->it_value, &ctv); 868 } 869 } else { 870 if (aitv->it_interval.tv_sec == 0 && 871 aitv->it_interval.tv_usec != 0 && 872 aitv->it_interval.tv_usec < tick) 873 aitv->it_interval.tv_usec = tick; 874 if (aitv->it_value.tv_sec == 0 && 875 aitv->it_value.tv_usec != 0 && 876 aitv->it_value.tv_usec < tick) 877 aitv->it_value.tv_usec = tick; 878 PROC_ITIMLOCK(p); 879 *oitv = p->p_stats->p_timer[which]; 880 p->p_stats->p_timer[which] = *aitv; 881 PROC_ITIMUNLOCK(p); 882 } 883 #ifdef KTRACE 884 if (KTRPOINT(td, KTR_STRUCT)) 885 ktritimerval(oitv); 886 #endif 887 return (0); 888 } 889 890 /* 891 * Real interval timer expired: 892 * send process whose timer expired an alarm signal. 893 * If time is not set up to reload, then just return. 894 * Else compute next time timer should go off which is > current time. 895 * This is where delay in processing this timeout causes multiple 896 * SIGALRM calls to be compressed into one. 897 * tvtohz() always adds 1 to allow for the time until the next clock 898 * interrupt being strictly less than 1 clock tick, but we don't want 899 * that here since we want to appear to be in sync with the clock 900 * interrupt even when we're delayed. 901 */ 902 void 903 realitexpire(void *arg) 904 { 905 struct proc *p; 906 struct timeval ctv; 907 sbintime_t isbt; 908 909 p = (struct proc *)arg; 910 kern_psignal(p, SIGALRM); 911 if (!timevalisset(&p->p_realtimer.it_interval)) { 912 timevalclear(&p->p_realtimer.it_value); 913 if (p->p_flag & P_WEXIT) 914 wakeup(&p->p_itcallout); 915 return; 916 } 917 isbt = tvtosbt(p->p_realtimer.it_interval); 918 if (isbt >= sbt_timethreshold) 919 getmicrouptime(&ctv); 920 else 921 microuptime(&ctv); 922 do { 923 timevaladd(&p->p_realtimer.it_value, 924 &p->p_realtimer.it_interval); 925 } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=)); 926 callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value), 927 isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE); 928 } 929 930 /* 931 * Check that a proposed value to load into the .it_value or 932 * .it_interval part of an interval timer is acceptable, and 933 * fix it to have at least minimal value (i.e. if it is less 934 * than the resolution of the clock, round it up.) 935 */ 936 int 937 itimerfix(struct timeval *tv) 938 { 939 940 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 941 return (EINVAL); 942 if (tv->tv_sec == 0 && tv->tv_usec != 0 && 943 tv->tv_usec < (u_int)tick / 16) 944 tv->tv_usec = (u_int)tick / 16; 945 return (0); 946 } 947 948 /* 949 * Decrement an interval timer by a specified number 950 * of microseconds, which must be less than a second, 951 * i.e. < 1000000. If the timer expires, then reload 952 * it. In this case, carry over (usec - old value) to 953 * reduce the value reloaded into the timer so that 954 * the timer does not drift. This routine assumes 955 * that it is called in a context where the timers 956 * on which it is operating cannot change in value. 957 */ 958 int 959 itimerdecr(struct itimerval *itp, int usec) 960 { 961 962 if (itp->it_value.tv_usec < usec) { 963 if (itp->it_value.tv_sec == 0) { 964 /* expired, and already in next interval */ 965 usec -= itp->it_value.tv_usec; 966 goto expire; 967 } 968 itp->it_value.tv_usec += 1000000; 969 itp->it_value.tv_sec--; 970 } 971 itp->it_value.tv_usec -= usec; 972 usec = 0; 973 if (timevalisset(&itp->it_value)) 974 return (1); 975 /* expired, exactly at end of interval */ 976 expire: 977 if (timevalisset(&itp->it_interval)) { 978 itp->it_value = itp->it_interval; 979 itp->it_value.tv_usec -= usec; 980 if (itp->it_value.tv_usec < 0) { 981 itp->it_value.tv_usec += 1000000; 982 itp->it_value.tv_sec--; 983 } 984 } else 985 itp->it_value.tv_usec = 0; /* sec is already 0 */ 986 return (0); 987 } 988 989 /* 990 * Add and subtract routines for timevals. 991 * N.B.: subtract routine doesn't deal with 992 * results which are before the beginning, 993 * it just gets very confused in this case. 994 * Caveat emptor. 995 */ 996 void 997 timevaladd(struct timeval *t1, const struct timeval *t2) 998 { 999 1000 t1->tv_sec += t2->tv_sec; 1001 t1->tv_usec += t2->tv_usec; 1002 timevalfix(t1); 1003 } 1004 1005 void 1006 timevalsub(struct timeval *t1, const struct timeval *t2) 1007 { 1008 1009 t1->tv_sec -= t2->tv_sec; 1010 t1->tv_usec -= t2->tv_usec; 1011 timevalfix(t1); 1012 } 1013 1014 static void 1015 timevalfix(struct timeval *t1) 1016 { 1017 1018 if (t1->tv_usec < 0) { 1019 t1->tv_sec--; 1020 t1->tv_usec += 1000000; 1021 } 1022 if (t1->tv_usec >= 1000000) { 1023 t1->tv_sec++; 1024 t1->tv_usec -= 1000000; 1025 } 1026 } 1027 1028 /* 1029 * ratecheck(): simple time-based rate-limit checking. 1030 */ 1031 int 1032 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1033 { 1034 struct timeval tv, delta; 1035 int rv = 0; 1036 1037 getmicrouptime(&tv); /* NB: 10ms precision */ 1038 delta = tv; 1039 timevalsub(&delta, lasttime); 1040 1041 /* 1042 * check for 0,0 is so that the message will be seen at least once, 1043 * even if interval is huge. 1044 */ 1045 if (timevalcmp(&delta, mininterval, >=) || 1046 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1047 *lasttime = tv; 1048 rv = 1; 1049 } 1050 1051 return (rv); 1052 } 1053 1054 /* 1055 * ppsratecheck(): packets (or events) per second limitation. 1056 * 1057 * Return 0 if the limit is to be enforced (e.g. the caller 1058 * should drop a packet because of the rate limitation). 1059 * 1060 * maxpps of 0 always causes zero to be returned. maxpps of -1 1061 * always causes 1 to be returned; this effectively defeats rate 1062 * limiting. 1063 * 1064 * Note that we maintain the struct timeval for compatibility 1065 * with other bsd systems. We reuse the storage and just monitor 1066 * clock ticks for minimal overhead. 1067 */ 1068 int 1069 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1070 { 1071 int now; 1072 1073 /* 1074 * Reset the last time and counter if this is the first call 1075 * or more than a second has passed since the last update of 1076 * lasttime. 1077 */ 1078 now = ticks; 1079 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1080 lasttime->tv_sec = now; 1081 *curpps = 1; 1082 return (maxpps != 0); 1083 } else { 1084 (*curpps)++; /* NB: ignore potential overflow */ 1085 return (maxpps < 0 || *curpps <= maxpps); 1086 } 1087 } 1088 1089 static void 1090 itimer_start(void) 1091 { 1092 struct kclock rt_clock = { 1093 .timer_create = realtimer_create, 1094 .timer_delete = realtimer_delete, 1095 .timer_settime = realtimer_settime, 1096 .timer_gettime = realtimer_gettime, 1097 .event_hook = NULL 1098 }; 1099 1100 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer), 1101 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0); 1102 register_posix_clock(CLOCK_REALTIME, &rt_clock); 1103 register_posix_clock(CLOCK_MONOTONIC, &rt_clock); 1104 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L); 1105 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX); 1106 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX); 1107 EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit, 1108 (void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY); 1109 EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec, 1110 (void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY); 1111 } 1112 1113 int 1114 register_posix_clock(int clockid, struct kclock *clk) 1115 { 1116 if ((unsigned)clockid >= MAX_CLOCKS) { 1117 printf("%s: invalid clockid\n", __func__); 1118 return (0); 1119 } 1120 posix_clocks[clockid] = *clk; 1121 return (1); 1122 } 1123 1124 static int 1125 itimer_init(void *mem, int size, int flags) 1126 { 1127 struct itimer *it; 1128 1129 it = (struct itimer *)mem; 1130 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF); 1131 return (0); 1132 } 1133 1134 static void 1135 itimer_fini(void *mem, int size) 1136 { 1137 struct itimer *it; 1138 1139 it = (struct itimer *)mem; 1140 mtx_destroy(&it->it_mtx); 1141 } 1142 1143 static void 1144 itimer_enter(struct itimer *it) 1145 { 1146 1147 mtx_assert(&it->it_mtx, MA_OWNED); 1148 it->it_usecount++; 1149 } 1150 1151 static void 1152 itimer_leave(struct itimer *it) 1153 { 1154 1155 mtx_assert(&it->it_mtx, MA_OWNED); 1156 KASSERT(it->it_usecount > 0, ("invalid it_usecount")); 1157 1158 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0) 1159 wakeup(it); 1160 } 1161 1162 #ifndef _SYS_SYSPROTO_H_ 1163 struct ktimer_create_args { 1164 clockid_t clock_id; 1165 struct sigevent * evp; 1166 int * timerid; 1167 }; 1168 #endif 1169 int 1170 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap) 1171 { 1172 struct sigevent *evp, ev; 1173 int id; 1174 int error; 1175 1176 if (uap->evp == NULL) { 1177 evp = NULL; 1178 } else { 1179 error = copyin(uap->evp, &ev, sizeof(ev)); 1180 if (error != 0) 1181 return (error); 1182 evp = &ev; 1183 } 1184 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1); 1185 if (error == 0) { 1186 error = copyout(&id, uap->timerid, sizeof(int)); 1187 if (error != 0) 1188 kern_ktimer_delete(td, id); 1189 } 1190 return (error); 1191 } 1192 1193 int 1194 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp, 1195 int *timerid, int preset_id) 1196 { 1197 struct proc *p = td->td_proc; 1198 struct itimer *it; 1199 int id; 1200 int error; 1201 1202 if (clock_id < 0 || clock_id >= MAX_CLOCKS) 1203 return (EINVAL); 1204 1205 if (posix_clocks[clock_id].timer_create == NULL) 1206 return (EINVAL); 1207 1208 if (evp != NULL) { 1209 if (evp->sigev_notify != SIGEV_NONE && 1210 evp->sigev_notify != SIGEV_SIGNAL && 1211 evp->sigev_notify != SIGEV_THREAD_ID) 1212 return (EINVAL); 1213 if ((evp->sigev_notify == SIGEV_SIGNAL || 1214 evp->sigev_notify == SIGEV_THREAD_ID) && 1215 !_SIG_VALID(evp->sigev_signo)) 1216 return (EINVAL); 1217 } 1218 1219 if (p->p_itimers == NULL) 1220 itimers_alloc(p); 1221 1222 it = uma_zalloc(itimer_zone, M_WAITOK); 1223 it->it_flags = 0; 1224 it->it_usecount = 0; 1225 it->it_active = 0; 1226 timespecclear(&it->it_time.it_value); 1227 timespecclear(&it->it_time.it_interval); 1228 it->it_overrun = 0; 1229 it->it_overrun_last = 0; 1230 it->it_clockid = clock_id; 1231 it->it_timerid = -1; 1232 it->it_proc = p; 1233 ksiginfo_init(&it->it_ksi); 1234 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT; 1235 error = CLOCK_CALL(clock_id, timer_create, (it)); 1236 if (error != 0) 1237 goto out; 1238 1239 PROC_LOCK(p); 1240 if (preset_id != -1) { 1241 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id")); 1242 id = preset_id; 1243 if (p->p_itimers->its_timers[id] != NULL) { 1244 PROC_UNLOCK(p); 1245 error = 0; 1246 goto out; 1247 } 1248 } else { 1249 /* 1250 * Find a free timer slot, skipping those reserved 1251 * for setitimer(). 1252 */ 1253 for (id = 3; id < TIMER_MAX; id++) 1254 if (p->p_itimers->its_timers[id] == NULL) 1255 break; 1256 if (id == TIMER_MAX) { 1257 PROC_UNLOCK(p); 1258 error = EAGAIN; 1259 goto out; 1260 } 1261 } 1262 it->it_timerid = id; 1263 p->p_itimers->its_timers[id] = it; 1264 if (evp != NULL) 1265 it->it_sigev = *evp; 1266 else { 1267 it->it_sigev.sigev_notify = SIGEV_SIGNAL; 1268 switch (clock_id) { 1269 default: 1270 case CLOCK_REALTIME: 1271 it->it_sigev.sigev_signo = SIGALRM; 1272 break; 1273 case CLOCK_VIRTUAL: 1274 it->it_sigev.sigev_signo = SIGVTALRM; 1275 break; 1276 case CLOCK_PROF: 1277 it->it_sigev.sigev_signo = SIGPROF; 1278 break; 1279 } 1280 it->it_sigev.sigev_value.sival_int = id; 1281 } 1282 1283 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1284 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1285 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo; 1286 it->it_ksi.ksi_code = SI_TIMER; 1287 it->it_ksi.ksi_value = it->it_sigev.sigev_value; 1288 it->it_ksi.ksi_timerid = id; 1289 } 1290 PROC_UNLOCK(p); 1291 *timerid = id; 1292 return (0); 1293 1294 out: 1295 ITIMER_LOCK(it); 1296 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1297 ITIMER_UNLOCK(it); 1298 uma_zfree(itimer_zone, it); 1299 return (error); 1300 } 1301 1302 #ifndef _SYS_SYSPROTO_H_ 1303 struct ktimer_delete_args { 1304 int timerid; 1305 }; 1306 #endif 1307 int 1308 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap) 1309 { 1310 1311 return (kern_ktimer_delete(td, uap->timerid)); 1312 } 1313 1314 static struct itimer * 1315 itimer_find(struct proc *p, int timerid) 1316 { 1317 struct itimer *it; 1318 1319 PROC_LOCK_ASSERT(p, MA_OWNED); 1320 if ((p->p_itimers == NULL) || 1321 (timerid < 0) || (timerid >= TIMER_MAX) || 1322 (it = p->p_itimers->its_timers[timerid]) == NULL) { 1323 return (NULL); 1324 } 1325 ITIMER_LOCK(it); 1326 if ((it->it_flags & ITF_DELETING) != 0) { 1327 ITIMER_UNLOCK(it); 1328 it = NULL; 1329 } 1330 return (it); 1331 } 1332 1333 int 1334 kern_ktimer_delete(struct thread *td, int timerid) 1335 { 1336 struct proc *p = td->td_proc; 1337 struct itimer *it; 1338 1339 PROC_LOCK(p); 1340 it = itimer_find(p, timerid); 1341 if (it == NULL) { 1342 PROC_UNLOCK(p); 1343 return (EINVAL); 1344 } 1345 PROC_UNLOCK(p); 1346 1347 it->it_flags |= ITF_DELETING; 1348 while (it->it_usecount > 0) { 1349 it->it_flags |= ITF_WANTED; 1350 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0); 1351 } 1352 it->it_flags &= ~ITF_WANTED; 1353 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1354 ITIMER_UNLOCK(it); 1355 1356 PROC_LOCK(p); 1357 if (KSI_ONQ(&it->it_ksi)) 1358 sigqueue_take(&it->it_ksi); 1359 p->p_itimers->its_timers[timerid] = NULL; 1360 PROC_UNLOCK(p); 1361 uma_zfree(itimer_zone, it); 1362 return (0); 1363 } 1364 1365 #ifndef _SYS_SYSPROTO_H_ 1366 struct ktimer_settime_args { 1367 int timerid; 1368 int flags; 1369 const struct itimerspec * value; 1370 struct itimerspec * ovalue; 1371 }; 1372 #endif 1373 int 1374 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap) 1375 { 1376 struct itimerspec val, oval, *ovalp; 1377 int error; 1378 1379 error = copyin(uap->value, &val, sizeof(val)); 1380 if (error != 0) 1381 return (error); 1382 ovalp = uap->ovalue != NULL ? &oval : NULL; 1383 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp); 1384 if (error == 0 && uap->ovalue != NULL) 1385 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp)); 1386 return (error); 1387 } 1388 1389 int 1390 kern_ktimer_settime(struct thread *td, int timer_id, int flags, 1391 struct itimerspec *val, struct itimerspec *oval) 1392 { 1393 struct proc *p; 1394 struct itimer *it; 1395 int error; 1396 1397 p = td->td_proc; 1398 PROC_LOCK(p); 1399 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1400 PROC_UNLOCK(p); 1401 error = EINVAL; 1402 } else { 1403 PROC_UNLOCK(p); 1404 itimer_enter(it); 1405 error = CLOCK_CALL(it->it_clockid, timer_settime, (it, 1406 flags, val, oval)); 1407 itimer_leave(it); 1408 ITIMER_UNLOCK(it); 1409 } 1410 return (error); 1411 } 1412 1413 #ifndef _SYS_SYSPROTO_H_ 1414 struct ktimer_gettime_args { 1415 int timerid; 1416 struct itimerspec * value; 1417 }; 1418 #endif 1419 int 1420 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap) 1421 { 1422 struct itimerspec val; 1423 int error; 1424 1425 error = kern_ktimer_gettime(td, uap->timerid, &val); 1426 if (error == 0) 1427 error = copyout(&val, uap->value, sizeof(val)); 1428 return (error); 1429 } 1430 1431 int 1432 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val) 1433 { 1434 struct proc *p; 1435 struct itimer *it; 1436 int error; 1437 1438 p = td->td_proc; 1439 PROC_LOCK(p); 1440 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1441 PROC_UNLOCK(p); 1442 error = EINVAL; 1443 } else { 1444 PROC_UNLOCK(p); 1445 itimer_enter(it); 1446 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val)); 1447 itimer_leave(it); 1448 ITIMER_UNLOCK(it); 1449 } 1450 return (error); 1451 } 1452 1453 #ifndef _SYS_SYSPROTO_H_ 1454 struct timer_getoverrun_args { 1455 int timerid; 1456 }; 1457 #endif 1458 int 1459 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap) 1460 { 1461 1462 return (kern_ktimer_getoverrun(td, uap->timerid)); 1463 } 1464 1465 int 1466 kern_ktimer_getoverrun(struct thread *td, int timer_id) 1467 { 1468 struct proc *p = td->td_proc; 1469 struct itimer *it; 1470 int error ; 1471 1472 PROC_LOCK(p); 1473 if (timer_id < 3 || 1474 (it = itimer_find(p, timer_id)) == NULL) { 1475 PROC_UNLOCK(p); 1476 error = EINVAL; 1477 } else { 1478 td->td_retval[0] = it->it_overrun_last; 1479 ITIMER_UNLOCK(it); 1480 PROC_UNLOCK(p); 1481 error = 0; 1482 } 1483 return (error); 1484 } 1485 1486 static int 1487 realtimer_create(struct itimer *it) 1488 { 1489 callout_init_mtx(&it->it_callout, &it->it_mtx, 0); 1490 return (0); 1491 } 1492 1493 static int 1494 realtimer_delete(struct itimer *it) 1495 { 1496 mtx_assert(&it->it_mtx, MA_OWNED); 1497 1498 /* 1499 * clear timer's value and interval to tell realtimer_expire 1500 * to not rearm the timer. 1501 */ 1502 timespecclear(&it->it_time.it_value); 1503 timespecclear(&it->it_time.it_interval); 1504 ITIMER_UNLOCK(it); 1505 callout_drain(&it->it_callout); 1506 ITIMER_LOCK(it); 1507 return (0); 1508 } 1509 1510 static int 1511 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue) 1512 { 1513 struct timespec cts; 1514 1515 mtx_assert(&it->it_mtx, MA_OWNED); 1516 1517 realtimer_clocktime(it->it_clockid, &cts); 1518 *ovalue = it->it_time; 1519 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) { 1520 timespecsub(&ovalue->it_value, &cts); 1521 if (ovalue->it_value.tv_sec < 0 || 1522 (ovalue->it_value.tv_sec == 0 && 1523 ovalue->it_value.tv_nsec == 0)) { 1524 ovalue->it_value.tv_sec = 0; 1525 ovalue->it_value.tv_nsec = 1; 1526 } 1527 } 1528 return (0); 1529 } 1530 1531 static int 1532 realtimer_settime(struct itimer *it, int flags, 1533 struct itimerspec *value, struct itimerspec *ovalue) 1534 { 1535 struct timespec cts, ts; 1536 struct timeval tv; 1537 struct itimerspec val; 1538 1539 mtx_assert(&it->it_mtx, MA_OWNED); 1540 1541 val = *value; 1542 if (itimespecfix(&val.it_value)) 1543 return (EINVAL); 1544 1545 if (timespecisset(&val.it_value)) { 1546 if (itimespecfix(&val.it_interval)) 1547 return (EINVAL); 1548 } else { 1549 timespecclear(&val.it_interval); 1550 } 1551 1552 if (ovalue != NULL) 1553 realtimer_gettime(it, ovalue); 1554 1555 it->it_time = val; 1556 if (timespecisset(&val.it_value)) { 1557 realtimer_clocktime(it->it_clockid, &cts); 1558 ts = val.it_value; 1559 if ((flags & TIMER_ABSTIME) == 0) { 1560 /* Convert to absolute time. */ 1561 timespecadd(&it->it_time.it_value, &cts); 1562 } else { 1563 timespecsub(&ts, &cts); 1564 /* 1565 * We don't care if ts is negative, tztohz will 1566 * fix it. 1567 */ 1568 } 1569 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1570 callout_reset(&it->it_callout, tvtohz(&tv), 1571 realtimer_expire, it); 1572 } else { 1573 callout_stop(&it->it_callout); 1574 } 1575 1576 return (0); 1577 } 1578 1579 static void 1580 realtimer_clocktime(clockid_t id, struct timespec *ts) 1581 { 1582 if (id == CLOCK_REALTIME) 1583 getnanotime(ts); 1584 else /* CLOCK_MONOTONIC */ 1585 getnanouptime(ts); 1586 } 1587 1588 int 1589 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi) 1590 { 1591 struct itimer *it; 1592 1593 PROC_LOCK_ASSERT(p, MA_OWNED); 1594 it = itimer_find(p, timerid); 1595 if (it != NULL) { 1596 ksi->ksi_overrun = it->it_overrun; 1597 it->it_overrun_last = it->it_overrun; 1598 it->it_overrun = 0; 1599 ITIMER_UNLOCK(it); 1600 return (0); 1601 } 1602 return (EINVAL); 1603 } 1604 1605 int 1606 itimespecfix(struct timespec *ts) 1607 { 1608 1609 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000) 1610 return (EINVAL); 1611 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000) 1612 ts->tv_nsec = tick * 1000; 1613 return (0); 1614 } 1615 1616 /* Timeout callback for realtime timer */ 1617 static void 1618 realtimer_expire(void *arg) 1619 { 1620 struct timespec cts, ts; 1621 struct timeval tv; 1622 struct itimer *it; 1623 1624 it = (struct itimer *)arg; 1625 1626 realtimer_clocktime(it->it_clockid, &cts); 1627 /* Only fire if time is reached. */ 1628 if (timespeccmp(&cts, &it->it_time.it_value, >=)) { 1629 if (timespecisset(&it->it_time.it_interval)) { 1630 timespecadd(&it->it_time.it_value, 1631 &it->it_time.it_interval); 1632 while (timespeccmp(&cts, &it->it_time.it_value, >=)) { 1633 if (it->it_overrun < INT_MAX) 1634 it->it_overrun++; 1635 else 1636 it->it_ksi.ksi_errno = ERANGE; 1637 timespecadd(&it->it_time.it_value, 1638 &it->it_time.it_interval); 1639 } 1640 } else { 1641 /* single shot timer ? */ 1642 timespecclear(&it->it_time.it_value); 1643 } 1644 if (timespecisset(&it->it_time.it_value)) { 1645 ts = it->it_time.it_value; 1646 timespecsub(&ts, &cts); 1647 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1648 callout_reset(&it->it_callout, tvtohz(&tv), 1649 realtimer_expire, it); 1650 } 1651 itimer_enter(it); 1652 ITIMER_UNLOCK(it); 1653 itimer_fire(it); 1654 ITIMER_LOCK(it); 1655 itimer_leave(it); 1656 } else if (timespecisset(&it->it_time.it_value)) { 1657 ts = it->it_time.it_value; 1658 timespecsub(&ts, &cts); 1659 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1660 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire, 1661 it); 1662 } 1663 } 1664 1665 void 1666 itimer_fire(struct itimer *it) 1667 { 1668 struct proc *p = it->it_proc; 1669 struct thread *td; 1670 1671 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1672 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1673 if (sigev_findtd(p, &it->it_sigev, &td) != 0) { 1674 ITIMER_LOCK(it); 1675 timespecclear(&it->it_time.it_value); 1676 timespecclear(&it->it_time.it_interval); 1677 callout_stop(&it->it_callout); 1678 ITIMER_UNLOCK(it); 1679 return; 1680 } 1681 if (!KSI_ONQ(&it->it_ksi)) { 1682 it->it_ksi.ksi_errno = 0; 1683 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev); 1684 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi); 1685 } else { 1686 if (it->it_overrun < INT_MAX) 1687 it->it_overrun++; 1688 else 1689 it->it_ksi.ksi_errno = ERANGE; 1690 } 1691 PROC_UNLOCK(p); 1692 } 1693 } 1694 1695 static void 1696 itimers_alloc(struct proc *p) 1697 { 1698 struct itimers *its; 1699 int i; 1700 1701 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO); 1702 LIST_INIT(&its->its_virtual); 1703 LIST_INIT(&its->its_prof); 1704 TAILQ_INIT(&its->its_worklist); 1705 for (i = 0; i < TIMER_MAX; i++) 1706 its->its_timers[i] = NULL; 1707 PROC_LOCK(p); 1708 if (p->p_itimers == NULL) { 1709 p->p_itimers = its; 1710 PROC_UNLOCK(p); 1711 } 1712 else { 1713 PROC_UNLOCK(p); 1714 free(its, M_SUBPROC); 1715 } 1716 } 1717 1718 static void 1719 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused) 1720 { 1721 itimers_event_hook_exit(arg, p); 1722 } 1723 1724 /* Clean up timers when some process events are being triggered. */ 1725 static void 1726 itimers_event_hook_exit(void *arg, struct proc *p) 1727 { 1728 struct itimers *its; 1729 struct itimer *it; 1730 int event = (int)(intptr_t)arg; 1731 int i; 1732 1733 if (p->p_itimers != NULL) { 1734 its = p->p_itimers; 1735 for (i = 0; i < MAX_CLOCKS; ++i) { 1736 if (posix_clocks[i].event_hook != NULL) 1737 CLOCK_CALL(i, event_hook, (p, i, event)); 1738 } 1739 /* 1740 * According to susv3, XSI interval timers should be inherited 1741 * by new image. 1742 */ 1743 if (event == ITIMER_EV_EXEC) 1744 i = 3; 1745 else if (event == ITIMER_EV_EXIT) 1746 i = 0; 1747 else 1748 panic("unhandled event"); 1749 for (; i < TIMER_MAX; ++i) { 1750 if ((it = its->its_timers[i]) != NULL) 1751 kern_ktimer_delete(curthread, i); 1752 } 1753 if (its->its_timers[0] == NULL && 1754 its->its_timers[1] == NULL && 1755 its->its_timers[2] == NULL) { 1756 free(its, M_SUBPROC); 1757 p->p_itimers = NULL; 1758 } 1759 } 1760 } 1761