1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_ktrace.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/limits.h> 40 #include <sys/clock.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/sysproto.h> 44 #include <sys/eventhandler.h> 45 #include <sys/resourcevar.h> 46 #include <sys/signalvar.h> 47 #include <sys/kernel.h> 48 #include <sys/sleepqueue.h> 49 #include <sys/syscallsubr.h> 50 #include <sys/sysctl.h> 51 #include <sys/sysent.h> 52 #include <sys/priv.h> 53 #include <sys/proc.h> 54 #include <sys/posix4.h> 55 #include <sys/time.h> 56 #include <sys/timers.h> 57 #include <sys/timetc.h> 58 #include <sys/vnode.h> 59 #ifdef KTRACE 60 #include <sys/ktrace.h> 61 #endif 62 63 #include <vm/vm.h> 64 #include <vm/vm_extern.h> 65 66 #define MAX_CLOCKS (CLOCK_MONOTONIC+1) 67 #define CPUCLOCK_BIT 0x80000000 68 #define CPUCLOCK_PROCESS_BIT 0x40000000 69 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT)) 70 #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid)) 71 #define MAKE_PROCESS_CPUCLOCK(pid) \ 72 (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid)) 73 74 static struct kclock posix_clocks[MAX_CLOCKS]; 75 static uma_zone_t itimer_zone = NULL; 76 77 /* 78 * Time of day and interval timer support. 79 * 80 * These routines provide the kernel entry points to get and set 81 * the time-of-day and per-process interval timers. Subroutines 82 * here provide support for adding and subtracting timeval structures 83 * and decrementing interval timers, optionally reloading the interval 84 * timers when they expire. 85 */ 86 87 static int settime(struct thread *, struct timeval *); 88 static void timevalfix(struct timeval *); 89 static int user_clock_nanosleep(struct thread *td, clockid_t clock_id, 90 int flags, const struct timespec *ua_rqtp, 91 struct timespec *ua_rmtp); 92 93 static void itimer_start(void); 94 static int itimer_init(void *, int, int); 95 static void itimer_fini(void *, int); 96 static void itimer_enter(struct itimer *); 97 static void itimer_leave(struct itimer *); 98 static struct itimer *itimer_find(struct proc *, int); 99 static void itimers_alloc(struct proc *); 100 static void itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp); 101 static void itimers_event_hook_exit(void *arg, struct proc *p); 102 static int realtimer_create(struct itimer *); 103 static int realtimer_gettime(struct itimer *, struct itimerspec *); 104 static int realtimer_settime(struct itimer *, int, 105 struct itimerspec *, struct itimerspec *); 106 static int realtimer_delete(struct itimer *); 107 static void realtimer_clocktime(clockid_t, struct timespec *); 108 static void realtimer_expire(void *); 109 110 int register_posix_clock(int, struct kclock *); 111 void itimer_fire(struct itimer *it); 112 int itimespecfix(struct timespec *ts); 113 114 #define CLOCK_CALL(clock, call, arglist) \ 115 ((*posix_clocks[clock].call) arglist) 116 117 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL); 118 119 120 static int 121 settime(struct thread *td, struct timeval *tv) 122 { 123 struct timeval delta, tv1, tv2; 124 static struct timeval maxtime, laststep; 125 struct timespec ts; 126 127 microtime(&tv1); 128 delta = *tv; 129 timevalsub(&delta, &tv1); 130 131 /* 132 * If the system is secure, we do not allow the time to be 133 * set to a value earlier than 1 second less than the highest 134 * time we have yet seen. The worst a miscreant can do in 135 * this circumstance is "freeze" time. He couldn't go 136 * back to the past. 137 * 138 * We similarly do not allow the clock to be stepped more 139 * than one second, nor more than once per second. This allows 140 * a miscreant to make the clock march double-time, but no worse. 141 */ 142 if (securelevel_gt(td->td_ucred, 1) != 0) { 143 if (delta.tv_sec < 0 || delta.tv_usec < 0) { 144 /* 145 * Update maxtime to latest time we've seen. 146 */ 147 if (tv1.tv_sec > maxtime.tv_sec) 148 maxtime = tv1; 149 tv2 = *tv; 150 timevalsub(&tv2, &maxtime); 151 if (tv2.tv_sec < -1) { 152 tv->tv_sec = maxtime.tv_sec - 1; 153 printf("Time adjustment clamped to -1 second\n"); 154 } 155 } else { 156 if (tv1.tv_sec == laststep.tv_sec) 157 return (EPERM); 158 if (delta.tv_sec > 1) { 159 tv->tv_sec = tv1.tv_sec + 1; 160 printf("Time adjustment clamped to +1 second\n"); 161 } 162 laststep = *tv; 163 } 164 } 165 166 ts.tv_sec = tv->tv_sec; 167 ts.tv_nsec = tv->tv_usec * 1000; 168 tc_setclock(&ts); 169 resettodr(); 170 return (0); 171 } 172 173 #ifndef _SYS_SYSPROTO_H_ 174 struct clock_getcpuclockid2_args { 175 id_t id; 176 int which, 177 clockid_t *clock_id; 178 }; 179 #endif 180 /* ARGSUSED */ 181 int 182 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap) 183 { 184 clockid_t clk_id; 185 int error; 186 187 error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id); 188 if (error == 0) 189 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 190 return (error); 191 } 192 193 int 194 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which, 195 clockid_t *clk_id) 196 { 197 struct proc *p; 198 pid_t pid; 199 lwpid_t tid; 200 int error; 201 202 switch (which) { 203 case CPUCLOCK_WHICH_PID: 204 if (id != 0) { 205 error = pget(id, PGET_CANSEE | PGET_NOTID, &p); 206 if (error != 0) 207 return (error); 208 PROC_UNLOCK(p); 209 pid = id; 210 } else { 211 pid = td->td_proc->p_pid; 212 } 213 *clk_id = MAKE_PROCESS_CPUCLOCK(pid); 214 return (0); 215 case CPUCLOCK_WHICH_TID: 216 tid = id == 0 ? td->td_tid : id; 217 *clk_id = MAKE_THREAD_CPUCLOCK(tid); 218 return (0); 219 default: 220 return (EINVAL); 221 } 222 } 223 224 #ifndef _SYS_SYSPROTO_H_ 225 struct clock_gettime_args { 226 clockid_t clock_id; 227 struct timespec *tp; 228 }; 229 #endif 230 /* ARGSUSED */ 231 int 232 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap) 233 { 234 struct timespec ats; 235 int error; 236 237 error = kern_clock_gettime(td, uap->clock_id, &ats); 238 if (error == 0) 239 error = copyout(&ats, uap->tp, sizeof(ats)); 240 241 return (error); 242 } 243 244 static inline void 245 cputick2timespec(uint64_t runtime, struct timespec *ats) 246 { 247 runtime = cputick2usec(runtime); 248 ats->tv_sec = runtime / 1000000; 249 ats->tv_nsec = runtime % 1000000 * 1000; 250 } 251 252 static void 253 get_thread_cputime(struct thread *targettd, struct timespec *ats) 254 { 255 uint64_t runtime, curtime, switchtime; 256 257 if (targettd == NULL) { /* current thread */ 258 critical_enter(); 259 switchtime = PCPU_GET(switchtime); 260 curtime = cpu_ticks(); 261 runtime = curthread->td_runtime; 262 critical_exit(); 263 runtime += curtime - switchtime; 264 } else { 265 thread_lock(targettd); 266 runtime = targettd->td_runtime; 267 thread_unlock(targettd); 268 } 269 cputick2timespec(runtime, ats); 270 } 271 272 static void 273 get_process_cputime(struct proc *targetp, struct timespec *ats) 274 { 275 uint64_t runtime; 276 struct rusage ru; 277 278 PROC_STATLOCK(targetp); 279 rufetch(targetp, &ru); 280 runtime = targetp->p_rux.rux_runtime; 281 PROC_STATUNLOCK(targetp); 282 cputick2timespec(runtime, ats); 283 } 284 285 static int 286 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats) 287 { 288 struct proc *p, *p2; 289 struct thread *td2; 290 lwpid_t tid; 291 pid_t pid; 292 int error; 293 294 p = td->td_proc; 295 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) { 296 tid = clock_id & CPUCLOCK_ID_MASK; 297 td2 = tdfind(tid, p->p_pid); 298 if (td2 == NULL) 299 return (EINVAL); 300 get_thread_cputime(td2, ats); 301 PROC_UNLOCK(td2->td_proc); 302 } else { 303 pid = clock_id & CPUCLOCK_ID_MASK; 304 error = pget(pid, PGET_CANSEE, &p2); 305 if (error != 0) 306 return (EINVAL); 307 get_process_cputime(p2, ats); 308 PROC_UNLOCK(p2); 309 } 310 return (0); 311 } 312 313 int 314 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats) 315 { 316 struct timeval sys, user; 317 struct proc *p; 318 319 p = td->td_proc; 320 switch (clock_id) { 321 case CLOCK_REALTIME: /* Default to precise. */ 322 case CLOCK_REALTIME_PRECISE: 323 nanotime(ats); 324 break; 325 case CLOCK_REALTIME_FAST: 326 getnanotime(ats); 327 break; 328 case CLOCK_VIRTUAL: 329 PROC_LOCK(p); 330 PROC_STATLOCK(p); 331 calcru(p, &user, &sys); 332 PROC_STATUNLOCK(p); 333 PROC_UNLOCK(p); 334 TIMEVAL_TO_TIMESPEC(&user, ats); 335 break; 336 case CLOCK_PROF: 337 PROC_LOCK(p); 338 PROC_STATLOCK(p); 339 calcru(p, &user, &sys); 340 PROC_STATUNLOCK(p); 341 PROC_UNLOCK(p); 342 timevaladd(&user, &sys); 343 TIMEVAL_TO_TIMESPEC(&user, ats); 344 break; 345 case CLOCK_MONOTONIC: /* Default to precise. */ 346 case CLOCK_MONOTONIC_PRECISE: 347 case CLOCK_UPTIME: 348 case CLOCK_UPTIME_PRECISE: 349 nanouptime(ats); 350 break; 351 case CLOCK_UPTIME_FAST: 352 case CLOCK_MONOTONIC_FAST: 353 getnanouptime(ats); 354 break; 355 case CLOCK_SECOND: 356 ats->tv_sec = time_second; 357 ats->tv_nsec = 0; 358 break; 359 case CLOCK_THREAD_CPUTIME_ID: 360 get_thread_cputime(NULL, ats); 361 break; 362 case CLOCK_PROCESS_CPUTIME_ID: 363 PROC_LOCK(p); 364 get_process_cputime(p, ats); 365 PROC_UNLOCK(p); 366 break; 367 default: 368 if ((int)clock_id >= 0) 369 return (EINVAL); 370 return (get_cputime(td, clock_id, ats)); 371 } 372 return (0); 373 } 374 375 #ifndef _SYS_SYSPROTO_H_ 376 struct clock_settime_args { 377 clockid_t clock_id; 378 const struct timespec *tp; 379 }; 380 #endif 381 /* ARGSUSED */ 382 int 383 sys_clock_settime(struct thread *td, struct clock_settime_args *uap) 384 { 385 struct timespec ats; 386 int error; 387 388 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 389 return (error); 390 return (kern_clock_settime(td, uap->clock_id, &ats)); 391 } 392 393 static int allow_insane_settime = 0; 394 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN, 395 &allow_insane_settime, 0, 396 "do not perform possibly restrictive checks on settime(2) args"); 397 398 int 399 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats) 400 { 401 struct timeval atv; 402 int error; 403 404 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 405 return (error); 406 if (clock_id != CLOCK_REALTIME) 407 return (EINVAL); 408 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 || 409 ats->tv_sec < 0) 410 return (EINVAL); 411 if (!allow_insane_settime && ats->tv_sec > 9999ULL * 366 * 24 * 60 * 60) 412 return (EINVAL); 413 /* XXX Don't convert nsec->usec and back */ 414 TIMESPEC_TO_TIMEVAL(&atv, ats); 415 error = settime(td, &atv); 416 return (error); 417 } 418 419 #ifndef _SYS_SYSPROTO_H_ 420 struct clock_getres_args { 421 clockid_t clock_id; 422 struct timespec *tp; 423 }; 424 #endif 425 int 426 sys_clock_getres(struct thread *td, struct clock_getres_args *uap) 427 { 428 struct timespec ts; 429 int error; 430 431 if (uap->tp == NULL) 432 return (0); 433 434 error = kern_clock_getres(td, uap->clock_id, &ts); 435 if (error == 0) 436 error = copyout(&ts, uap->tp, sizeof(ts)); 437 return (error); 438 } 439 440 int 441 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts) 442 { 443 444 ts->tv_sec = 0; 445 switch (clock_id) { 446 case CLOCK_REALTIME: 447 case CLOCK_REALTIME_FAST: 448 case CLOCK_REALTIME_PRECISE: 449 case CLOCK_MONOTONIC: 450 case CLOCK_MONOTONIC_FAST: 451 case CLOCK_MONOTONIC_PRECISE: 452 case CLOCK_UPTIME: 453 case CLOCK_UPTIME_FAST: 454 case CLOCK_UPTIME_PRECISE: 455 /* 456 * Round up the result of the division cheaply by adding 1. 457 * Rounding up is especially important if rounding down 458 * would give 0. Perfect rounding is unimportant. 459 */ 460 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1; 461 break; 462 case CLOCK_VIRTUAL: 463 case CLOCK_PROF: 464 /* Accurately round up here because we can do so cheaply. */ 465 ts->tv_nsec = howmany(1000000000, hz); 466 break; 467 case CLOCK_SECOND: 468 ts->tv_sec = 1; 469 ts->tv_nsec = 0; 470 break; 471 case CLOCK_THREAD_CPUTIME_ID: 472 case CLOCK_PROCESS_CPUTIME_ID: 473 cputime: 474 /* sync with cputick2usec */ 475 ts->tv_nsec = 1000000 / cpu_tickrate(); 476 if (ts->tv_nsec == 0) 477 ts->tv_nsec = 1000; 478 break; 479 default: 480 if ((int)clock_id < 0) 481 goto cputime; 482 return (EINVAL); 483 } 484 return (0); 485 } 486 487 int 488 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt) 489 { 490 491 return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt, 492 rmt)); 493 } 494 495 static uint8_t nanowait[MAXCPU]; 496 497 int 498 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, 499 const struct timespec *rqt, struct timespec *rmt) 500 { 501 struct timespec ts, now; 502 sbintime_t sbt, sbtt, prec, tmp; 503 time_t over; 504 int error; 505 bool is_abs_real; 506 507 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 508 return (EINVAL); 509 if ((flags & ~TIMER_ABSTIME) != 0) 510 return (EINVAL); 511 switch (clock_id) { 512 case CLOCK_REALTIME: 513 case CLOCK_REALTIME_PRECISE: 514 case CLOCK_REALTIME_FAST: 515 case CLOCK_SECOND: 516 is_abs_real = (flags & TIMER_ABSTIME) != 0; 517 break; 518 case CLOCK_MONOTONIC: 519 case CLOCK_MONOTONIC_PRECISE: 520 case CLOCK_MONOTONIC_FAST: 521 case CLOCK_UPTIME: 522 case CLOCK_UPTIME_PRECISE: 523 case CLOCK_UPTIME_FAST: 524 is_abs_real = false; 525 break; 526 case CLOCK_VIRTUAL: 527 case CLOCK_PROF: 528 case CLOCK_PROCESS_CPUTIME_ID: 529 return (ENOTSUP); 530 case CLOCK_THREAD_CPUTIME_ID: 531 default: 532 return (EINVAL); 533 } 534 do { 535 ts = *rqt; 536 if ((flags & TIMER_ABSTIME) != 0) { 537 if (is_abs_real) 538 td->td_rtcgen = 539 atomic_load_acq_int(&rtc_generation); 540 error = kern_clock_gettime(td, clock_id, &now); 541 KASSERT(error == 0, ("kern_clock_gettime: %d", error)); 542 timespecsub(&ts, &now); 543 } 544 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) { 545 error = EWOULDBLOCK; 546 break; 547 } 548 if (ts.tv_sec > INT32_MAX / 2) { 549 over = ts.tv_sec - INT32_MAX / 2; 550 ts.tv_sec -= over; 551 } else 552 over = 0; 553 tmp = tstosbt(ts); 554 prec = tmp; 555 prec >>= tc_precexp; 556 if (TIMESEL(&sbt, tmp)) 557 sbt += tc_tick_sbt; 558 sbt += tmp; 559 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp", 560 sbt, prec, C_ABSOLUTE); 561 } while (error == 0 && is_abs_real && td->td_rtcgen == 0); 562 td->td_rtcgen = 0; 563 if (error != EWOULDBLOCK) { 564 TIMESEL(&sbtt, tmp); 565 if (sbtt >= sbt) 566 return (0); 567 if (error == ERESTART) 568 error = EINTR; 569 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) { 570 ts = sbttots(sbt - sbtt); 571 ts.tv_sec += over; 572 if (ts.tv_sec < 0) 573 timespecclear(&ts); 574 *rmt = ts; 575 } 576 return (error); 577 } 578 return (0); 579 } 580 581 #ifndef _SYS_SYSPROTO_H_ 582 struct nanosleep_args { 583 struct timespec *rqtp; 584 struct timespec *rmtp; 585 }; 586 #endif 587 /* ARGSUSED */ 588 int 589 sys_nanosleep(struct thread *td, struct nanosleep_args *uap) 590 { 591 592 return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, 593 uap->rqtp, uap->rmtp)); 594 } 595 596 #ifndef _SYS_SYSPROTO_H_ 597 struct clock_nanosleep_args { 598 clockid_t clock_id; 599 int flags; 600 struct timespec *rqtp; 601 struct timespec *rmtp; 602 }; 603 #endif 604 /* ARGSUSED */ 605 int 606 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap) 607 { 608 int error; 609 610 error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp, 611 uap->rmtp); 612 return (kern_posix_error(td, error)); 613 } 614 615 static int 616 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, 617 const struct timespec *ua_rqtp, struct timespec *ua_rmtp) 618 { 619 struct timespec rmt, rqt; 620 int error; 621 622 error = copyin(ua_rqtp, &rqt, sizeof(rqt)); 623 if (error) 624 return (error); 625 if (ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0 && 626 !useracc(ua_rmtp, sizeof(rmt), VM_PROT_WRITE)) 627 return (EFAULT); 628 error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt); 629 if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) { 630 int error2; 631 632 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt)); 633 if (error2) 634 error = error2; 635 } 636 return (error); 637 } 638 639 #ifndef _SYS_SYSPROTO_H_ 640 struct gettimeofday_args { 641 struct timeval *tp; 642 struct timezone *tzp; 643 }; 644 #endif 645 /* ARGSUSED */ 646 int 647 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap) 648 { 649 struct timeval atv; 650 struct timezone rtz; 651 int error = 0; 652 653 if (uap->tp) { 654 microtime(&atv); 655 error = copyout(&atv, uap->tp, sizeof (atv)); 656 } 657 if (error == 0 && uap->tzp != NULL) { 658 rtz.tz_minuteswest = tz_minuteswest; 659 rtz.tz_dsttime = tz_dsttime; 660 error = copyout(&rtz, uap->tzp, sizeof (rtz)); 661 } 662 return (error); 663 } 664 665 #ifndef _SYS_SYSPROTO_H_ 666 struct settimeofday_args { 667 struct timeval *tv; 668 struct timezone *tzp; 669 }; 670 #endif 671 /* ARGSUSED */ 672 int 673 sys_settimeofday(struct thread *td, struct settimeofday_args *uap) 674 { 675 struct timeval atv, *tvp; 676 struct timezone atz, *tzp; 677 int error; 678 679 if (uap->tv) { 680 error = copyin(uap->tv, &atv, sizeof(atv)); 681 if (error) 682 return (error); 683 tvp = &atv; 684 } else 685 tvp = NULL; 686 if (uap->tzp) { 687 error = copyin(uap->tzp, &atz, sizeof(atz)); 688 if (error) 689 return (error); 690 tzp = &atz; 691 } else 692 tzp = NULL; 693 return (kern_settimeofday(td, tvp, tzp)); 694 } 695 696 int 697 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp) 698 { 699 int error; 700 701 error = priv_check(td, PRIV_SETTIMEOFDAY); 702 if (error) 703 return (error); 704 /* Verify all parameters before changing time. */ 705 if (tv) { 706 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 || 707 tv->tv_sec < 0) 708 return (EINVAL); 709 error = settime(td, tv); 710 } 711 if (tzp && error == 0) { 712 tz_minuteswest = tzp->tz_minuteswest; 713 tz_dsttime = tzp->tz_dsttime; 714 } 715 return (error); 716 } 717 718 /* 719 * Get value of an interval timer. The process virtual and profiling virtual 720 * time timers are kept in the p_stats area, since they can be swapped out. 721 * These are kept internally in the way they are specified externally: in 722 * time until they expire. 723 * 724 * The real time interval timer is kept in the process table slot for the 725 * process, and its value (it_value) is kept as an absolute time rather than 726 * as a delta, so that it is easy to keep periodic real-time signals from 727 * drifting. 728 * 729 * Virtual time timers are processed in the hardclock() routine of 730 * kern_clock.c. The real time timer is processed by a timeout routine, 731 * called from the softclock() routine. Since a callout may be delayed in 732 * real time due to interrupt processing in the system, it is possible for 733 * the real time timeout routine (realitexpire, given below), to be delayed 734 * in real time past when it is supposed to occur. It does not suffice, 735 * therefore, to reload the real timer .it_value from the real time timers 736 * .it_interval. Rather, we compute the next time in absolute time the timer 737 * should go off. 738 */ 739 #ifndef _SYS_SYSPROTO_H_ 740 struct getitimer_args { 741 u_int which; 742 struct itimerval *itv; 743 }; 744 #endif 745 int 746 sys_getitimer(struct thread *td, struct getitimer_args *uap) 747 { 748 struct itimerval aitv; 749 int error; 750 751 error = kern_getitimer(td, uap->which, &aitv); 752 if (error != 0) 753 return (error); 754 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 755 } 756 757 int 758 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv) 759 { 760 struct proc *p = td->td_proc; 761 struct timeval ctv; 762 763 if (which > ITIMER_PROF) 764 return (EINVAL); 765 766 if (which == ITIMER_REAL) { 767 /* 768 * Convert from absolute to relative time in .it_value 769 * part of real time timer. If time for real time timer 770 * has passed return 0, else return difference between 771 * current time and time for the timer to go off. 772 */ 773 PROC_LOCK(p); 774 *aitv = p->p_realtimer; 775 PROC_UNLOCK(p); 776 if (timevalisset(&aitv->it_value)) { 777 microuptime(&ctv); 778 if (timevalcmp(&aitv->it_value, &ctv, <)) 779 timevalclear(&aitv->it_value); 780 else 781 timevalsub(&aitv->it_value, &ctv); 782 } 783 } else { 784 PROC_ITIMLOCK(p); 785 *aitv = p->p_stats->p_timer[which]; 786 PROC_ITIMUNLOCK(p); 787 } 788 #ifdef KTRACE 789 if (KTRPOINT(td, KTR_STRUCT)) 790 ktritimerval(aitv); 791 #endif 792 return (0); 793 } 794 795 #ifndef _SYS_SYSPROTO_H_ 796 struct setitimer_args { 797 u_int which; 798 struct itimerval *itv, *oitv; 799 }; 800 #endif 801 int 802 sys_setitimer(struct thread *td, struct setitimer_args *uap) 803 { 804 struct itimerval aitv, oitv; 805 int error; 806 807 if (uap->itv == NULL) { 808 uap->itv = uap->oitv; 809 return (sys_getitimer(td, (struct getitimer_args *)uap)); 810 } 811 812 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval)))) 813 return (error); 814 error = kern_setitimer(td, uap->which, &aitv, &oitv); 815 if (error != 0 || uap->oitv == NULL) 816 return (error); 817 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval))); 818 } 819 820 int 821 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv, 822 struct itimerval *oitv) 823 { 824 struct proc *p = td->td_proc; 825 struct timeval ctv; 826 sbintime_t sbt, pr; 827 828 if (aitv == NULL) 829 return (kern_getitimer(td, which, oitv)); 830 831 if (which > ITIMER_PROF) 832 return (EINVAL); 833 #ifdef KTRACE 834 if (KTRPOINT(td, KTR_STRUCT)) 835 ktritimerval(aitv); 836 #endif 837 if (itimerfix(&aitv->it_value) || 838 aitv->it_value.tv_sec > INT32_MAX / 2) 839 return (EINVAL); 840 if (!timevalisset(&aitv->it_value)) 841 timevalclear(&aitv->it_interval); 842 else if (itimerfix(&aitv->it_interval) || 843 aitv->it_interval.tv_sec > INT32_MAX / 2) 844 return (EINVAL); 845 846 if (which == ITIMER_REAL) { 847 PROC_LOCK(p); 848 if (timevalisset(&p->p_realtimer.it_value)) 849 callout_stop(&p->p_itcallout); 850 microuptime(&ctv); 851 if (timevalisset(&aitv->it_value)) { 852 pr = tvtosbt(aitv->it_value) >> tc_precexp; 853 timevaladd(&aitv->it_value, &ctv); 854 sbt = tvtosbt(aitv->it_value); 855 callout_reset_sbt(&p->p_itcallout, sbt, pr, 856 realitexpire, p, C_ABSOLUTE); 857 } 858 *oitv = p->p_realtimer; 859 p->p_realtimer = *aitv; 860 PROC_UNLOCK(p); 861 if (timevalisset(&oitv->it_value)) { 862 if (timevalcmp(&oitv->it_value, &ctv, <)) 863 timevalclear(&oitv->it_value); 864 else 865 timevalsub(&oitv->it_value, &ctv); 866 } 867 } else { 868 if (aitv->it_interval.tv_sec == 0 && 869 aitv->it_interval.tv_usec != 0 && 870 aitv->it_interval.tv_usec < tick) 871 aitv->it_interval.tv_usec = tick; 872 if (aitv->it_value.tv_sec == 0 && 873 aitv->it_value.tv_usec != 0 && 874 aitv->it_value.tv_usec < tick) 875 aitv->it_value.tv_usec = tick; 876 PROC_ITIMLOCK(p); 877 *oitv = p->p_stats->p_timer[which]; 878 p->p_stats->p_timer[which] = *aitv; 879 PROC_ITIMUNLOCK(p); 880 } 881 #ifdef KTRACE 882 if (KTRPOINT(td, KTR_STRUCT)) 883 ktritimerval(oitv); 884 #endif 885 return (0); 886 } 887 888 /* 889 * Real interval timer expired: 890 * send process whose timer expired an alarm signal. 891 * If time is not set up to reload, then just return. 892 * Else compute next time timer should go off which is > current time. 893 * This is where delay in processing this timeout causes multiple 894 * SIGALRM calls to be compressed into one. 895 * tvtohz() always adds 1 to allow for the time until the next clock 896 * interrupt being strictly less than 1 clock tick, but we don't want 897 * that here since we want to appear to be in sync with the clock 898 * interrupt even when we're delayed. 899 */ 900 void 901 realitexpire(void *arg) 902 { 903 struct proc *p; 904 struct timeval ctv; 905 sbintime_t isbt; 906 907 p = (struct proc *)arg; 908 kern_psignal(p, SIGALRM); 909 if (!timevalisset(&p->p_realtimer.it_interval)) { 910 timevalclear(&p->p_realtimer.it_value); 911 if (p->p_flag & P_WEXIT) 912 wakeup(&p->p_itcallout); 913 return; 914 } 915 isbt = tvtosbt(p->p_realtimer.it_interval); 916 if (isbt >= sbt_timethreshold) 917 getmicrouptime(&ctv); 918 else 919 microuptime(&ctv); 920 do { 921 timevaladd(&p->p_realtimer.it_value, 922 &p->p_realtimer.it_interval); 923 } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=)); 924 callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value), 925 isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE); 926 } 927 928 /* 929 * Check that a proposed value to load into the .it_value or 930 * .it_interval part of an interval timer is acceptable, and 931 * fix it to have at least minimal value (i.e. if it is less 932 * than the resolution of the clock, round it up.) 933 */ 934 int 935 itimerfix(struct timeval *tv) 936 { 937 938 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 939 return (EINVAL); 940 if (tv->tv_sec == 0 && tv->tv_usec != 0 && 941 tv->tv_usec < (u_int)tick / 16) 942 tv->tv_usec = (u_int)tick / 16; 943 return (0); 944 } 945 946 /* 947 * Decrement an interval timer by a specified number 948 * of microseconds, which must be less than a second, 949 * i.e. < 1000000. If the timer expires, then reload 950 * it. In this case, carry over (usec - old value) to 951 * reduce the value reloaded into the timer so that 952 * the timer does not drift. This routine assumes 953 * that it is called in a context where the timers 954 * on which it is operating cannot change in value. 955 */ 956 int 957 itimerdecr(struct itimerval *itp, int usec) 958 { 959 960 if (itp->it_value.tv_usec < usec) { 961 if (itp->it_value.tv_sec == 0) { 962 /* expired, and already in next interval */ 963 usec -= itp->it_value.tv_usec; 964 goto expire; 965 } 966 itp->it_value.tv_usec += 1000000; 967 itp->it_value.tv_sec--; 968 } 969 itp->it_value.tv_usec -= usec; 970 usec = 0; 971 if (timevalisset(&itp->it_value)) 972 return (1); 973 /* expired, exactly at end of interval */ 974 expire: 975 if (timevalisset(&itp->it_interval)) { 976 itp->it_value = itp->it_interval; 977 itp->it_value.tv_usec -= usec; 978 if (itp->it_value.tv_usec < 0) { 979 itp->it_value.tv_usec += 1000000; 980 itp->it_value.tv_sec--; 981 } 982 } else 983 itp->it_value.tv_usec = 0; /* sec is already 0 */ 984 return (0); 985 } 986 987 /* 988 * Add and subtract routines for timevals. 989 * N.B.: subtract routine doesn't deal with 990 * results which are before the beginning, 991 * it just gets very confused in this case. 992 * Caveat emptor. 993 */ 994 void 995 timevaladd(struct timeval *t1, const struct timeval *t2) 996 { 997 998 t1->tv_sec += t2->tv_sec; 999 t1->tv_usec += t2->tv_usec; 1000 timevalfix(t1); 1001 } 1002 1003 void 1004 timevalsub(struct timeval *t1, const struct timeval *t2) 1005 { 1006 1007 t1->tv_sec -= t2->tv_sec; 1008 t1->tv_usec -= t2->tv_usec; 1009 timevalfix(t1); 1010 } 1011 1012 static void 1013 timevalfix(struct timeval *t1) 1014 { 1015 1016 if (t1->tv_usec < 0) { 1017 t1->tv_sec--; 1018 t1->tv_usec += 1000000; 1019 } 1020 if (t1->tv_usec >= 1000000) { 1021 t1->tv_sec++; 1022 t1->tv_usec -= 1000000; 1023 } 1024 } 1025 1026 /* 1027 * ratecheck(): simple time-based rate-limit checking. 1028 */ 1029 int 1030 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1031 { 1032 struct timeval tv, delta; 1033 int rv = 0; 1034 1035 getmicrouptime(&tv); /* NB: 10ms precision */ 1036 delta = tv; 1037 timevalsub(&delta, lasttime); 1038 1039 /* 1040 * check for 0,0 is so that the message will be seen at least once, 1041 * even if interval is huge. 1042 */ 1043 if (timevalcmp(&delta, mininterval, >=) || 1044 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1045 *lasttime = tv; 1046 rv = 1; 1047 } 1048 1049 return (rv); 1050 } 1051 1052 /* 1053 * ppsratecheck(): packets (or events) per second limitation. 1054 * 1055 * Return 0 if the limit is to be enforced (e.g. the caller 1056 * should drop a packet because of the rate limitation). 1057 * 1058 * maxpps of 0 always causes zero to be returned. maxpps of -1 1059 * always causes 1 to be returned; this effectively defeats rate 1060 * limiting. 1061 * 1062 * Note that we maintain the struct timeval for compatibility 1063 * with other bsd systems. We reuse the storage and just monitor 1064 * clock ticks for minimal overhead. 1065 */ 1066 int 1067 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1068 { 1069 int now; 1070 1071 /* 1072 * Reset the last time and counter if this is the first call 1073 * or more than a second has passed since the last update of 1074 * lasttime. 1075 */ 1076 now = ticks; 1077 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1078 lasttime->tv_sec = now; 1079 *curpps = 1; 1080 return (maxpps != 0); 1081 } else { 1082 (*curpps)++; /* NB: ignore potential overflow */ 1083 return (maxpps < 0 || *curpps <= maxpps); 1084 } 1085 } 1086 1087 static void 1088 itimer_start(void) 1089 { 1090 struct kclock rt_clock = { 1091 .timer_create = realtimer_create, 1092 .timer_delete = realtimer_delete, 1093 .timer_settime = realtimer_settime, 1094 .timer_gettime = realtimer_gettime, 1095 .event_hook = NULL 1096 }; 1097 1098 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer), 1099 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0); 1100 register_posix_clock(CLOCK_REALTIME, &rt_clock); 1101 register_posix_clock(CLOCK_MONOTONIC, &rt_clock); 1102 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L); 1103 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX); 1104 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX); 1105 EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit, 1106 (void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY); 1107 EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec, 1108 (void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY); 1109 } 1110 1111 int 1112 register_posix_clock(int clockid, struct kclock *clk) 1113 { 1114 if ((unsigned)clockid >= MAX_CLOCKS) { 1115 printf("%s: invalid clockid\n", __func__); 1116 return (0); 1117 } 1118 posix_clocks[clockid] = *clk; 1119 return (1); 1120 } 1121 1122 static int 1123 itimer_init(void *mem, int size, int flags) 1124 { 1125 struct itimer *it; 1126 1127 it = (struct itimer *)mem; 1128 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF); 1129 return (0); 1130 } 1131 1132 static void 1133 itimer_fini(void *mem, int size) 1134 { 1135 struct itimer *it; 1136 1137 it = (struct itimer *)mem; 1138 mtx_destroy(&it->it_mtx); 1139 } 1140 1141 static void 1142 itimer_enter(struct itimer *it) 1143 { 1144 1145 mtx_assert(&it->it_mtx, MA_OWNED); 1146 it->it_usecount++; 1147 } 1148 1149 static void 1150 itimer_leave(struct itimer *it) 1151 { 1152 1153 mtx_assert(&it->it_mtx, MA_OWNED); 1154 KASSERT(it->it_usecount > 0, ("invalid it_usecount")); 1155 1156 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0) 1157 wakeup(it); 1158 } 1159 1160 #ifndef _SYS_SYSPROTO_H_ 1161 struct ktimer_create_args { 1162 clockid_t clock_id; 1163 struct sigevent * evp; 1164 int * timerid; 1165 }; 1166 #endif 1167 int 1168 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap) 1169 { 1170 struct sigevent *evp, ev; 1171 int id; 1172 int error; 1173 1174 if (uap->evp == NULL) { 1175 evp = NULL; 1176 } else { 1177 error = copyin(uap->evp, &ev, sizeof(ev)); 1178 if (error != 0) 1179 return (error); 1180 evp = &ev; 1181 } 1182 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1); 1183 if (error == 0) { 1184 error = copyout(&id, uap->timerid, sizeof(int)); 1185 if (error != 0) 1186 kern_ktimer_delete(td, id); 1187 } 1188 return (error); 1189 } 1190 1191 int 1192 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp, 1193 int *timerid, int preset_id) 1194 { 1195 struct proc *p = td->td_proc; 1196 struct itimer *it; 1197 int id; 1198 int error; 1199 1200 if (clock_id < 0 || clock_id >= MAX_CLOCKS) 1201 return (EINVAL); 1202 1203 if (posix_clocks[clock_id].timer_create == NULL) 1204 return (EINVAL); 1205 1206 if (evp != NULL) { 1207 if (evp->sigev_notify != SIGEV_NONE && 1208 evp->sigev_notify != SIGEV_SIGNAL && 1209 evp->sigev_notify != SIGEV_THREAD_ID) 1210 return (EINVAL); 1211 if ((evp->sigev_notify == SIGEV_SIGNAL || 1212 evp->sigev_notify == SIGEV_THREAD_ID) && 1213 !_SIG_VALID(evp->sigev_signo)) 1214 return (EINVAL); 1215 } 1216 1217 if (p->p_itimers == NULL) 1218 itimers_alloc(p); 1219 1220 it = uma_zalloc(itimer_zone, M_WAITOK); 1221 it->it_flags = 0; 1222 it->it_usecount = 0; 1223 it->it_active = 0; 1224 timespecclear(&it->it_time.it_value); 1225 timespecclear(&it->it_time.it_interval); 1226 it->it_overrun = 0; 1227 it->it_overrun_last = 0; 1228 it->it_clockid = clock_id; 1229 it->it_timerid = -1; 1230 it->it_proc = p; 1231 ksiginfo_init(&it->it_ksi); 1232 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT; 1233 error = CLOCK_CALL(clock_id, timer_create, (it)); 1234 if (error != 0) 1235 goto out; 1236 1237 PROC_LOCK(p); 1238 if (preset_id != -1) { 1239 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id")); 1240 id = preset_id; 1241 if (p->p_itimers->its_timers[id] != NULL) { 1242 PROC_UNLOCK(p); 1243 error = 0; 1244 goto out; 1245 } 1246 } else { 1247 /* 1248 * Find a free timer slot, skipping those reserved 1249 * for setitimer(). 1250 */ 1251 for (id = 3; id < TIMER_MAX; id++) 1252 if (p->p_itimers->its_timers[id] == NULL) 1253 break; 1254 if (id == TIMER_MAX) { 1255 PROC_UNLOCK(p); 1256 error = EAGAIN; 1257 goto out; 1258 } 1259 } 1260 it->it_timerid = id; 1261 p->p_itimers->its_timers[id] = it; 1262 if (evp != NULL) 1263 it->it_sigev = *evp; 1264 else { 1265 it->it_sigev.sigev_notify = SIGEV_SIGNAL; 1266 switch (clock_id) { 1267 default: 1268 case CLOCK_REALTIME: 1269 it->it_sigev.sigev_signo = SIGALRM; 1270 break; 1271 case CLOCK_VIRTUAL: 1272 it->it_sigev.sigev_signo = SIGVTALRM; 1273 break; 1274 case CLOCK_PROF: 1275 it->it_sigev.sigev_signo = SIGPROF; 1276 break; 1277 } 1278 it->it_sigev.sigev_value.sival_int = id; 1279 } 1280 1281 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1282 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1283 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo; 1284 it->it_ksi.ksi_code = SI_TIMER; 1285 it->it_ksi.ksi_value = it->it_sigev.sigev_value; 1286 it->it_ksi.ksi_timerid = id; 1287 } 1288 PROC_UNLOCK(p); 1289 *timerid = id; 1290 return (0); 1291 1292 out: 1293 ITIMER_LOCK(it); 1294 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1295 ITIMER_UNLOCK(it); 1296 uma_zfree(itimer_zone, it); 1297 return (error); 1298 } 1299 1300 #ifndef _SYS_SYSPROTO_H_ 1301 struct ktimer_delete_args { 1302 int timerid; 1303 }; 1304 #endif 1305 int 1306 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap) 1307 { 1308 1309 return (kern_ktimer_delete(td, uap->timerid)); 1310 } 1311 1312 static struct itimer * 1313 itimer_find(struct proc *p, int timerid) 1314 { 1315 struct itimer *it; 1316 1317 PROC_LOCK_ASSERT(p, MA_OWNED); 1318 if ((p->p_itimers == NULL) || 1319 (timerid < 0) || (timerid >= TIMER_MAX) || 1320 (it = p->p_itimers->its_timers[timerid]) == NULL) { 1321 return (NULL); 1322 } 1323 ITIMER_LOCK(it); 1324 if ((it->it_flags & ITF_DELETING) != 0) { 1325 ITIMER_UNLOCK(it); 1326 it = NULL; 1327 } 1328 return (it); 1329 } 1330 1331 int 1332 kern_ktimer_delete(struct thread *td, int timerid) 1333 { 1334 struct proc *p = td->td_proc; 1335 struct itimer *it; 1336 1337 PROC_LOCK(p); 1338 it = itimer_find(p, timerid); 1339 if (it == NULL) { 1340 PROC_UNLOCK(p); 1341 return (EINVAL); 1342 } 1343 PROC_UNLOCK(p); 1344 1345 it->it_flags |= ITF_DELETING; 1346 while (it->it_usecount > 0) { 1347 it->it_flags |= ITF_WANTED; 1348 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0); 1349 } 1350 it->it_flags &= ~ITF_WANTED; 1351 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1352 ITIMER_UNLOCK(it); 1353 1354 PROC_LOCK(p); 1355 if (KSI_ONQ(&it->it_ksi)) 1356 sigqueue_take(&it->it_ksi); 1357 p->p_itimers->its_timers[timerid] = NULL; 1358 PROC_UNLOCK(p); 1359 uma_zfree(itimer_zone, it); 1360 return (0); 1361 } 1362 1363 #ifndef _SYS_SYSPROTO_H_ 1364 struct ktimer_settime_args { 1365 int timerid; 1366 int flags; 1367 const struct itimerspec * value; 1368 struct itimerspec * ovalue; 1369 }; 1370 #endif 1371 int 1372 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap) 1373 { 1374 struct itimerspec val, oval, *ovalp; 1375 int error; 1376 1377 error = copyin(uap->value, &val, sizeof(val)); 1378 if (error != 0) 1379 return (error); 1380 ovalp = uap->ovalue != NULL ? &oval : NULL; 1381 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp); 1382 if (error == 0 && uap->ovalue != NULL) 1383 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp)); 1384 return (error); 1385 } 1386 1387 int 1388 kern_ktimer_settime(struct thread *td, int timer_id, int flags, 1389 struct itimerspec *val, struct itimerspec *oval) 1390 { 1391 struct proc *p; 1392 struct itimer *it; 1393 int error; 1394 1395 p = td->td_proc; 1396 PROC_LOCK(p); 1397 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1398 PROC_UNLOCK(p); 1399 error = EINVAL; 1400 } else { 1401 PROC_UNLOCK(p); 1402 itimer_enter(it); 1403 error = CLOCK_CALL(it->it_clockid, timer_settime, (it, 1404 flags, val, oval)); 1405 itimer_leave(it); 1406 ITIMER_UNLOCK(it); 1407 } 1408 return (error); 1409 } 1410 1411 #ifndef _SYS_SYSPROTO_H_ 1412 struct ktimer_gettime_args { 1413 int timerid; 1414 struct itimerspec * value; 1415 }; 1416 #endif 1417 int 1418 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap) 1419 { 1420 struct itimerspec val; 1421 int error; 1422 1423 error = kern_ktimer_gettime(td, uap->timerid, &val); 1424 if (error == 0) 1425 error = copyout(&val, uap->value, sizeof(val)); 1426 return (error); 1427 } 1428 1429 int 1430 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val) 1431 { 1432 struct proc *p; 1433 struct itimer *it; 1434 int error; 1435 1436 p = td->td_proc; 1437 PROC_LOCK(p); 1438 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1439 PROC_UNLOCK(p); 1440 error = EINVAL; 1441 } else { 1442 PROC_UNLOCK(p); 1443 itimer_enter(it); 1444 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val)); 1445 itimer_leave(it); 1446 ITIMER_UNLOCK(it); 1447 } 1448 return (error); 1449 } 1450 1451 #ifndef _SYS_SYSPROTO_H_ 1452 struct timer_getoverrun_args { 1453 int timerid; 1454 }; 1455 #endif 1456 int 1457 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap) 1458 { 1459 1460 return (kern_ktimer_getoverrun(td, uap->timerid)); 1461 } 1462 1463 int 1464 kern_ktimer_getoverrun(struct thread *td, int timer_id) 1465 { 1466 struct proc *p = td->td_proc; 1467 struct itimer *it; 1468 int error ; 1469 1470 PROC_LOCK(p); 1471 if (timer_id < 3 || 1472 (it = itimer_find(p, timer_id)) == NULL) { 1473 PROC_UNLOCK(p); 1474 error = EINVAL; 1475 } else { 1476 td->td_retval[0] = it->it_overrun_last; 1477 ITIMER_UNLOCK(it); 1478 PROC_UNLOCK(p); 1479 error = 0; 1480 } 1481 return (error); 1482 } 1483 1484 static int 1485 realtimer_create(struct itimer *it) 1486 { 1487 callout_init_mtx(&it->it_callout, &it->it_mtx, 0); 1488 return (0); 1489 } 1490 1491 static int 1492 realtimer_delete(struct itimer *it) 1493 { 1494 mtx_assert(&it->it_mtx, MA_OWNED); 1495 1496 /* 1497 * clear timer's value and interval to tell realtimer_expire 1498 * to not rearm the timer. 1499 */ 1500 timespecclear(&it->it_time.it_value); 1501 timespecclear(&it->it_time.it_interval); 1502 ITIMER_UNLOCK(it); 1503 callout_drain(&it->it_callout); 1504 ITIMER_LOCK(it); 1505 return (0); 1506 } 1507 1508 static int 1509 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue) 1510 { 1511 struct timespec cts; 1512 1513 mtx_assert(&it->it_mtx, MA_OWNED); 1514 1515 realtimer_clocktime(it->it_clockid, &cts); 1516 *ovalue = it->it_time; 1517 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) { 1518 timespecsub(&ovalue->it_value, &cts); 1519 if (ovalue->it_value.tv_sec < 0 || 1520 (ovalue->it_value.tv_sec == 0 && 1521 ovalue->it_value.tv_nsec == 0)) { 1522 ovalue->it_value.tv_sec = 0; 1523 ovalue->it_value.tv_nsec = 1; 1524 } 1525 } 1526 return (0); 1527 } 1528 1529 static int 1530 realtimer_settime(struct itimer *it, int flags, 1531 struct itimerspec *value, struct itimerspec *ovalue) 1532 { 1533 struct timespec cts, ts; 1534 struct timeval tv; 1535 struct itimerspec val; 1536 1537 mtx_assert(&it->it_mtx, MA_OWNED); 1538 1539 val = *value; 1540 if (itimespecfix(&val.it_value)) 1541 return (EINVAL); 1542 1543 if (timespecisset(&val.it_value)) { 1544 if (itimespecfix(&val.it_interval)) 1545 return (EINVAL); 1546 } else { 1547 timespecclear(&val.it_interval); 1548 } 1549 1550 if (ovalue != NULL) 1551 realtimer_gettime(it, ovalue); 1552 1553 it->it_time = val; 1554 if (timespecisset(&val.it_value)) { 1555 realtimer_clocktime(it->it_clockid, &cts); 1556 ts = val.it_value; 1557 if ((flags & TIMER_ABSTIME) == 0) { 1558 /* Convert to absolute time. */ 1559 timespecadd(&it->it_time.it_value, &cts); 1560 } else { 1561 timespecsub(&ts, &cts); 1562 /* 1563 * We don't care if ts is negative, tztohz will 1564 * fix it. 1565 */ 1566 } 1567 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1568 callout_reset(&it->it_callout, tvtohz(&tv), 1569 realtimer_expire, it); 1570 } else { 1571 callout_stop(&it->it_callout); 1572 } 1573 1574 return (0); 1575 } 1576 1577 static void 1578 realtimer_clocktime(clockid_t id, struct timespec *ts) 1579 { 1580 if (id == CLOCK_REALTIME) 1581 getnanotime(ts); 1582 else /* CLOCK_MONOTONIC */ 1583 getnanouptime(ts); 1584 } 1585 1586 int 1587 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi) 1588 { 1589 struct itimer *it; 1590 1591 PROC_LOCK_ASSERT(p, MA_OWNED); 1592 it = itimer_find(p, timerid); 1593 if (it != NULL) { 1594 ksi->ksi_overrun = it->it_overrun; 1595 it->it_overrun_last = it->it_overrun; 1596 it->it_overrun = 0; 1597 ITIMER_UNLOCK(it); 1598 return (0); 1599 } 1600 return (EINVAL); 1601 } 1602 1603 int 1604 itimespecfix(struct timespec *ts) 1605 { 1606 1607 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000) 1608 return (EINVAL); 1609 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000) 1610 ts->tv_nsec = tick * 1000; 1611 return (0); 1612 } 1613 1614 /* Timeout callback for realtime timer */ 1615 static void 1616 realtimer_expire(void *arg) 1617 { 1618 struct timespec cts, ts; 1619 struct timeval tv; 1620 struct itimer *it; 1621 1622 it = (struct itimer *)arg; 1623 1624 realtimer_clocktime(it->it_clockid, &cts); 1625 /* Only fire if time is reached. */ 1626 if (timespeccmp(&cts, &it->it_time.it_value, >=)) { 1627 if (timespecisset(&it->it_time.it_interval)) { 1628 timespecadd(&it->it_time.it_value, 1629 &it->it_time.it_interval); 1630 while (timespeccmp(&cts, &it->it_time.it_value, >=)) { 1631 if (it->it_overrun < INT_MAX) 1632 it->it_overrun++; 1633 else 1634 it->it_ksi.ksi_errno = ERANGE; 1635 timespecadd(&it->it_time.it_value, 1636 &it->it_time.it_interval); 1637 } 1638 } else { 1639 /* single shot timer ? */ 1640 timespecclear(&it->it_time.it_value); 1641 } 1642 if (timespecisset(&it->it_time.it_value)) { 1643 ts = it->it_time.it_value; 1644 timespecsub(&ts, &cts); 1645 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1646 callout_reset(&it->it_callout, tvtohz(&tv), 1647 realtimer_expire, it); 1648 } 1649 itimer_enter(it); 1650 ITIMER_UNLOCK(it); 1651 itimer_fire(it); 1652 ITIMER_LOCK(it); 1653 itimer_leave(it); 1654 } else if (timespecisset(&it->it_time.it_value)) { 1655 ts = it->it_time.it_value; 1656 timespecsub(&ts, &cts); 1657 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1658 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire, 1659 it); 1660 } 1661 } 1662 1663 void 1664 itimer_fire(struct itimer *it) 1665 { 1666 struct proc *p = it->it_proc; 1667 struct thread *td; 1668 1669 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1670 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1671 if (sigev_findtd(p, &it->it_sigev, &td) != 0) { 1672 ITIMER_LOCK(it); 1673 timespecclear(&it->it_time.it_value); 1674 timespecclear(&it->it_time.it_interval); 1675 callout_stop(&it->it_callout); 1676 ITIMER_UNLOCK(it); 1677 return; 1678 } 1679 if (!KSI_ONQ(&it->it_ksi)) { 1680 it->it_ksi.ksi_errno = 0; 1681 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev); 1682 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi); 1683 } else { 1684 if (it->it_overrun < INT_MAX) 1685 it->it_overrun++; 1686 else 1687 it->it_ksi.ksi_errno = ERANGE; 1688 } 1689 PROC_UNLOCK(p); 1690 } 1691 } 1692 1693 static void 1694 itimers_alloc(struct proc *p) 1695 { 1696 struct itimers *its; 1697 int i; 1698 1699 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO); 1700 LIST_INIT(&its->its_virtual); 1701 LIST_INIT(&its->its_prof); 1702 TAILQ_INIT(&its->its_worklist); 1703 for (i = 0; i < TIMER_MAX; i++) 1704 its->its_timers[i] = NULL; 1705 PROC_LOCK(p); 1706 if (p->p_itimers == NULL) { 1707 p->p_itimers = its; 1708 PROC_UNLOCK(p); 1709 } 1710 else { 1711 PROC_UNLOCK(p); 1712 free(its, M_SUBPROC); 1713 } 1714 } 1715 1716 static void 1717 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused) 1718 { 1719 itimers_event_hook_exit(arg, p); 1720 } 1721 1722 /* Clean up timers when some process events are being triggered. */ 1723 static void 1724 itimers_event_hook_exit(void *arg, struct proc *p) 1725 { 1726 struct itimers *its; 1727 struct itimer *it; 1728 int event = (int)(intptr_t)arg; 1729 int i; 1730 1731 if (p->p_itimers != NULL) { 1732 its = p->p_itimers; 1733 for (i = 0; i < MAX_CLOCKS; ++i) { 1734 if (posix_clocks[i].event_hook != NULL) 1735 CLOCK_CALL(i, event_hook, (p, i, event)); 1736 } 1737 /* 1738 * According to susv3, XSI interval timers should be inherited 1739 * by new image. 1740 */ 1741 if (event == ITIMER_EV_EXEC) 1742 i = 3; 1743 else if (event == ITIMER_EV_EXIT) 1744 i = 0; 1745 else 1746 panic("unhandled event"); 1747 for (; i < TIMER_MAX; ++i) { 1748 if ((it = its->its_timers[i]) != NULL) 1749 kern_ktimer_delete(curthread, i); 1750 } 1751 if (its->its_timers[0] == NULL && 1752 its->its_timers[1] == NULL && 1753 its->its_timers[2] == NULL) { 1754 free(its, M_SUBPROC); 1755 p->p_itimers = NULL; 1756 } 1757 } 1758 } 1759