kern_time.c revision 303093
1254721Semaste/*- 2254721Semaste * Copyright (c) 1982, 1986, 1989, 1993 3254721Semaste * The Regents of the University of California. All rights reserved. 4254721Semaste * 5254721Semaste * Redistribution and use in source and binary forms, with or without 6254721Semaste * modification, are permitted provided that the following conditions 7254721Semaste * are met: 8254721Semaste * 1. Redistributions of source code must retain the above copyright 9254721Semaste * notice, this list of conditions and the following disclaimer. 10254721Semaste * 2. Redistributions in binary form must reproduce the above copyright 11254721Semaste * notice, this list of conditions and the following disclaimer in the 12254721Semaste * documentation and/or other materials provided with the distribution. 13254721Semaste * 4. Neither the name of the University nor the names of its contributors 14254721Semaste * may be used to endorse or promote products derived from this software 15254721Semaste * without specific prior written permission. 16254721Semaste * 17254721Semaste * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18254721Semaste * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19254721Semaste * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20254721Semaste * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21254721Semaste * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22254721Semaste * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23254721Semaste * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24254721Semaste * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25254721Semaste * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26254721Semaste * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27254721Semaste * SUCH DAMAGE. 28254721Semaste * 29254721Semaste * @(#)kern_time.c 8.1 (Berkeley) 6/10/93 30254721Semaste */ 31254721Semaste 32254721Semaste#include <sys/cdefs.h> 33254721Semaste__FBSDID("$FreeBSD: stable/10/sys/kern/kern_time.c 303093 2016-07-20 15:07:52Z kib $"); 34254721Semaste 35254721Semaste#include "opt_ktrace.h" 36254721Semaste 37254721Semaste#include <sys/param.h> 38254721Semaste#include <sys/systm.h> 39254721Semaste#include <sys/limits.h> 40254721Semaste#include <sys/clock.h> 41254721Semaste#include <sys/lock.h> 42254721Semaste#include <sys/mutex.h> 43254721Semaste#include <sys/sysproto.h> 44254721Semaste#include <sys/eventhandler.h> 45254721Semaste#include <sys/resourcevar.h> 46254721Semaste#include <sys/signalvar.h> 47254721Semaste#include <sys/kernel.h> 48254721Semaste#include <sys/sleepqueue.h> 49254721Semaste#include <sys/syscallsubr.h> 50254721Semaste#include <sys/sysctl.h> 51254721Semaste#include <sys/sysent.h> 52254721Semaste#include <sys/priv.h> 53254721Semaste#include <sys/proc.h> 54254721Semaste#include <sys/posix4.h> 55254721Semaste#include <sys/time.h> 56254721Semaste#include <sys/timers.h> 57254721Semaste#include <sys/timetc.h> 58254721Semaste#include <sys/vnode.h> 59254721Semaste#ifdef KTRACE 60254721Semaste#include <sys/ktrace.h> 61254721Semaste#endif 62254721Semaste 63254721Semaste#include <vm/vm.h> 64254721Semaste#include <vm/vm_extern.h> 65254721Semaste 66254721Semaste#define MAX_CLOCKS (CLOCK_MONOTONIC+1) 67254721Semaste#define CPUCLOCK_BIT 0x80000000 68254721Semaste#define CPUCLOCK_PROCESS_BIT 0x40000000 69254721Semaste#define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT)) 70254721Semaste#define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid)) 71254721Semaste#define MAKE_PROCESS_CPUCLOCK(pid) \ 72254721Semaste (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid)) 73254721Semaste 74254721Semastestatic struct kclock posix_clocks[MAX_CLOCKS]; 75254721Semastestatic uma_zone_t itimer_zone = NULL; 76254721Semaste 77254721Semaste/* 78254721Semaste * Time of day and interval timer support. 79254721Semaste * 80254721Semaste * These routines provide the kernel entry points to get and set 81254721Semaste * the time-of-day and per-process interval timers. Subroutines 82254721Semaste * here provide support for adding and subtracting timeval structures 83254721Semaste * and decrementing interval timers, optionally reloading the interval 84254721Semaste * timers when they expire. 85254721Semaste */ 86254721Semaste 87254721Semastestatic int settime(struct thread *, struct timeval *); 88254721Semastestatic void timevalfix(struct timeval *); 89254721Semaste 90254721Semastestatic void itimer_start(void); 91254721Semastestatic int itimer_init(void *, int, int); 92254721Semastestatic void itimer_fini(void *, int); 93254721Semastestatic void itimer_enter(struct itimer *); 94254721Semastestatic void itimer_leave(struct itimer *); 95254721Semastestatic struct itimer *itimer_find(struct proc *, int); 96254721Semastestatic void itimers_alloc(struct proc *); 97254721Semastestatic void itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp); 98254721Semastestatic void itimers_event_hook_exit(void *arg, struct proc *p); 99254721Semastestatic int realtimer_create(struct itimer *); 100254721Semastestatic int realtimer_gettime(struct itimer *, struct itimerspec *); 101254721Semastestatic int realtimer_settime(struct itimer *, int, 102254721Semaste struct itimerspec *, struct itimerspec *); 103254721Semastestatic int realtimer_delete(struct itimer *); 104254721Semastestatic void realtimer_clocktime(clockid_t, struct timespec *); 105254721Semastestatic void realtimer_expire(void *); 106254721Semaste 107254721Semasteint register_posix_clock(int, struct kclock *); 108254721Semastevoid itimer_fire(struct itimer *it); 109254721Semasteint itimespecfix(struct timespec *ts); 110254721Semaste 111254721Semaste#define CLOCK_CALL(clock, call, arglist) \ 112254721Semaste ((*posix_clocks[clock].call) arglist) 113254721Semaste 114254721SemasteSYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL); 115254721Semaste 116254721Semaste 117254721Semastestatic int 118254721Semastesettime(struct thread *td, struct timeval *tv) 119254721Semaste{ 120254721Semaste struct timeval delta, tv1, tv2; 121254721Semaste static struct timeval maxtime, laststep; 122254721Semaste struct timespec ts; 123254721Semaste int s; 124254721Semaste 125254721Semaste s = splclock(); 126254721Semaste microtime(&tv1); 127254721Semaste delta = *tv; 128254721Semaste timevalsub(&delta, &tv1); 129254721Semaste 130254721Semaste /* 131254721Semaste * If the system is secure, we do not allow the time to be 132254721Semaste * set to a value earlier than 1 second less than the highest 133254721Semaste * time we have yet seen. The worst a miscreant can do in 134254721Semaste * this circumstance is "freeze" time. He couldn't go 135254721Semaste * back to the past. 136254721Semaste * 137254721Semaste * We similarly do not allow the clock to be stepped more 138254721Semaste * than one second, nor more than once per second. This allows 139254721Semaste * a miscreant to make the clock march double-time, but no worse. 140254721Semaste */ 141254721Semaste if (securelevel_gt(td->td_ucred, 1) != 0) { 142254721Semaste if (delta.tv_sec < 0 || delta.tv_usec < 0) { 143 /* 144 * Update maxtime to latest time we've seen. 145 */ 146 if (tv1.tv_sec > maxtime.tv_sec) 147 maxtime = tv1; 148 tv2 = *tv; 149 timevalsub(&tv2, &maxtime); 150 if (tv2.tv_sec < -1) { 151 tv->tv_sec = maxtime.tv_sec - 1; 152 printf("Time adjustment clamped to -1 second\n"); 153 } 154 } else { 155 if (tv1.tv_sec == laststep.tv_sec) { 156 splx(s); 157 return (EPERM); 158 } 159 if (delta.tv_sec > 1) { 160 tv->tv_sec = tv1.tv_sec + 1; 161 printf("Time adjustment clamped to +1 second\n"); 162 } 163 laststep = *tv; 164 } 165 } 166 167 ts.tv_sec = tv->tv_sec; 168 ts.tv_nsec = tv->tv_usec * 1000; 169 mtx_lock(&Giant); 170 tc_setclock(&ts); 171 resettodr(); 172 mtx_unlock(&Giant); 173 return (0); 174} 175 176#ifndef _SYS_SYSPROTO_H_ 177struct clock_getcpuclockid2_args { 178 id_t id; 179 int which, 180 clockid_t *clock_id; 181}; 182#endif 183/* ARGSUSED */ 184int 185sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap) 186{ 187 clockid_t clk_id; 188 int error; 189 190 error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id); 191 if (error == 0) 192 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 193 return (error); 194} 195 196int 197kern_clock_getcpuclockid2(struct thread *td, id_t id, int which, 198 clockid_t *clk_id) 199{ 200 struct proc *p; 201 pid_t pid; 202 lwpid_t tid; 203 int error; 204 205 switch (which) { 206 case CPUCLOCK_WHICH_PID: 207 if (id != 0) { 208 error = pget(id, PGET_CANSEE | PGET_NOTID, &p); 209 if (error != 0) 210 return (error); 211 PROC_UNLOCK(p); 212 pid = id; 213 } else { 214 pid = td->td_proc->p_pid; 215 } 216 *clk_id = MAKE_PROCESS_CPUCLOCK(pid); 217 return (0); 218 case CPUCLOCK_WHICH_TID: 219 tid = id == 0 ? td->td_tid : id; 220 *clk_id = MAKE_THREAD_CPUCLOCK(tid); 221 return (0); 222 default: 223 return (EINVAL); 224 } 225} 226 227#ifndef _SYS_SYSPROTO_H_ 228struct clock_gettime_args { 229 clockid_t clock_id; 230 struct timespec *tp; 231}; 232#endif 233/* ARGSUSED */ 234int 235sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap) 236{ 237 struct timespec ats; 238 int error; 239 240 error = kern_clock_gettime(td, uap->clock_id, &ats); 241 if (error == 0) 242 error = copyout(&ats, uap->tp, sizeof(ats)); 243 244 return (error); 245} 246 247static inline void 248cputick2timespec(uint64_t runtime, struct timespec *ats) 249{ 250 runtime = cputick2usec(runtime); 251 ats->tv_sec = runtime / 1000000; 252 ats->tv_nsec = runtime % 1000000 * 1000; 253} 254 255static void 256get_thread_cputime(struct thread *targettd, struct timespec *ats) 257{ 258 uint64_t runtime, curtime, switchtime; 259 260 if (targettd == NULL) { /* current thread */ 261 critical_enter(); 262 switchtime = PCPU_GET(switchtime); 263 curtime = cpu_ticks(); 264 runtime = curthread->td_runtime; 265 critical_exit(); 266 runtime += curtime - switchtime; 267 } else { 268 thread_lock(targettd); 269 runtime = targettd->td_runtime; 270 thread_unlock(targettd); 271 } 272 cputick2timespec(runtime, ats); 273} 274 275static void 276get_process_cputime(struct proc *targetp, struct timespec *ats) 277{ 278 uint64_t runtime; 279 struct rusage ru; 280 281 PROC_STATLOCK(targetp); 282 rufetch(targetp, &ru); 283 runtime = targetp->p_rux.rux_runtime; 284 PROC_STATUNLOCK(targetp); 285 cputick2timespec(runtime, ats); 286} 287 288static int 289get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats) 290{ 291 struct proc *p, *p2; 292 struct thread *td2; 293 lwpid_t tid; 294 pid_t pid; 295 int error; 296 297 p = td->td_proc; 298 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) { 299 tid = clock_id & CPUCLOCK_ID_MASK; 300 td2 = tdfind(tid, p->p_pid); 301 if (td2 == NULL) 302 return (EINVAL); 303 get_thread_cputime(td2, ats); 304 PROC_UNLOCK(td2->td_proc); 305 } else { 306 pid = clock_id & CPUCLOCK_ID_MASK; 307 error = pget(pid, PGET_CANSEE, &p2); 308 if (error != 0) 309 return (EINVAL); 310 get_process_cputime(p2, ats); 311 PROC_UNLOCK(p2); 312 } 313 return (0); 314} 315 316int 317kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats) 318{ 319 struct timeval sys, user; 320 struct proc *p; 321 322 p = td->td_proc; 323 switch (clock_id) { 324 case CLOCK_REALTIME: /* Default to precise. */ 325 case CLOCK_REALTIME_PRECISE: 326 nanotime(ats); 327 break; 328 case CLOCK_REALTIME_FAST: 329 getnanotime(ats); 330 break; 331 case CLOCK_VIRTUAL: 332 PROC_LOCK(p); 333 PROC_STATLOCK(p); 334 calcru(p, &user, &sys); 335 PROC_STATUNLOCK(p); 336 PROC_UNLOCK(p); 337 TIMEVAL_TO_TIMESPEC(&user, ats); 338 break; 339 case CLOCK_PROF: 340 PROC_LOCK(p); 341 PROC_STATLOCK(p); 342 calcru(p, &user, &sys); 343 PROC_STATUNLOCK(p); 344 PROC_UNLOCK(p); 345 timevaladd(&user, &sys); 346 TIMEVAL_TO_TIMESPEC(&user, ats); 347 break; 348 case CLOCK_MONOTONIC: /* Default to precise. */ 349 case CLOCK_MONOTONIC_PRECISE: 350 case CLOCK_UPTIME: 351 case CLOCK_UPTIME_PRECISE: 352 nanouptime(ats); 353 break; 354 case CLOCK_UPTIME_FAST: 355 case CLOCK_MONOTONIC_FAST: 356 getnanouptime(ats); 357 break; 358 case CLOCK_SECOND: 359 ats->tv_sec = time_second; 360 ats->tv_nsec = 0; 361 break; 362 case CLOCK_THREAD_CPUTIME_ID: 363 get_thread_cputime(NULL, ats); 364 break; 365 case CLOCK_PROCESS_CPUTIME_ID: 366 PROC_LOCK(p); 367 get_process_cputime(p, ats); 368 PROC_UNLOCK(p); 369 break; 370 default: 371 if ((int)clock_id >= 0) 372 return (EINVAL); 373 return (get_cputime(td, clock_id, ats)); 374 } 375 return (0); 376} 377 378#ifndef _SYS_SYSPROTO_H_ 379struct clock_settime_args { 380 clockid_t clock_id; 381 const struct timespec *tp; 382}; 383#endif 384/* ARGSUSED */ 385int 386sys_clock_settime(struct thread *td, struct clock_settime_args *uap) 387{ 388 struct timespec ats; 389 int error; 390 391 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 392 return (error); 393 return (kern_clock_settime(td, uap->clock_id, &ats)); 394} 395 396int 397kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats) 398{ 399 struct timeval atv; 400 int error; 401 402 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 403 return (error); 404 if (clock_id != CLOCK_REALTIME) 405 return (EINVAL); 406 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000) 407 return (EINVAL); 408 /* XXX Don't convert nsec->usec and back */ 409 TIMESPEC_TO_TIMEVAL(&atv, ats); 410 error = settime(td, &atv); 411 return (error); 412} 413 414#ifndef _SYS_SYSPROTO_H_ 415struct clock_getres_args { 416 clockid_t clock_id; 417 struct timespec *tp; 418}; 419#endif 420int 421sys_clock_getres(struct thread *td, struct clock_getres_args *uap) 422{ 423 struct timespec ts; 424 int error; 425 426 if (uap->tp == NULL) 427 return (0); 428 429 error = kern_clock_getres(td, uap->clock_id, &ts); 430 if (error == 0) 431 error = copyout(&ts, uap->tp, sizeof(ts)); 432 return (error); 433} 434 435int 436kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts) 437{ 438 439 ts->tv_sec = 0; 440 switch (clock_id) { 441 case CLOCK_REALTIME: 442 case CLOCK_REALTIME_FAST: 443 case CLOCK_REALTIME_PRECISE: 444 case CLOCK_MONOTONIC: 445 case CLOCK_MONOTONIC_FAST: 446 case CLOCK_MONOTONIC_PRECISE: 447 case CLOCK_UPTIME: 448 case CLOCK_UPTIME_FAST: 449 case CLOCK_UPTIME_PRECISE: 450 /* 451 * Round up the result of the division cheaply by adding 1. 452 * Rounding up is especially important if rounding down 453 * would give 0. Perfect rounding is unimportant. 454 */ 455 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1; 456 break; 457 case CLOCK_VIRTUAL: 458 case CLOCK_PROF: 459 /* Accurately round up here because we can do so cheaply. */ 460 ts->tv_nsec = (1000000000 + hz - 1) / hz; 461 break; 462 case CLOCK_SECOND: 463 ts->tv_sec = 1; 464 ts->tv_nsec = 0; 465 break; 466 case CLOCK_THREAD_CPUTIME_ID: 467 case CLOCK_PROCESS_CPUTIME_ID: 468 cputime: 469 /* sync with cputick2usec */ 470 ts->tv_nsec = 1000000 / cpu_tickrate(); 471 if (ts->tv_nsec == 0) 472 ts->tv_nsec = 1000; 473 break; 474 default: 475 if ((int)clock_id < 0) 476 goto cputime; 477 return (EINVAL); 478 } 479 return (0); 480} 481 482static uint8_t nanowait[MAXCPU]; 483 484int 485kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt) 486{ 487 struct timespec ts; 488 sbintime_t sbt, sbtt, prec, tmp; 489 time_t over; 490 int error; 491 492 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) 493 return (EINVAL); 494 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) 495 return (0); 496 ts = *rqt; 497 if (ts.tv_sec > INT32_MAX / 2) { 498 over = ts.tv_sec - INT32_MAX / 2; 499 ts.tv_sec -= over; 500 } else 501 over = 0; 502 tmp = tstosbt(ts); 503 prec = tmp; 504 prec >>= tc_precexp; 505 if (TIMESEL(&sbt, tmp)) 506 sbt += tc_tick_sbt; 507 sbt += tmp; 508 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp", 509 sbt, prec, C_ABSOLUTE); 510 if (error != EWOULDBLOCK) { 511 if (error == ERESTART) 512 error = EINTR; 513 TIMESEL(&sbtt, tmp); 514 if (rmt != NULL) { 515 ts = sbttots(sbt - sbtt); 516 ts.tv_sec += over; 517 if (ts.tv_sec < 0) 518 timespecclear(&ts); 519 *rmt = ts; 520 } 521 if (sbtt >= sbt) 522 return (0); 523 return (error); 524 } 525 return (0); 526} 527 528#ifndef _SYS_SYSPROTO_H_ 529struct nanosleep_args { 530 struct timespec *rqtp; 531 struct timespec *rmtp; 532}; 533#endif 534/* ARGSUSED */ 535int 536sys_nanosleep(struct thread *td, struct nanosleep_args *uap) 537{ 538 struct timespec rmt, rqt; 539 int error; 540 541 error = copyin(uap->rqtp, &rqt, sizeof(rqt)); 542 if (error) 543 return (error); 544 545 if (uap->rmtp && 546 !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE)) 547 return (EFAULT); 548 error = kern_nanosleep(td, &rqt, &rmt); 549 if (error && uap->rmtp) { 550 int error2; 551 552 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); 553 if (error2) 554 error = error2; 555 } 556 return (error); 557} 558 559#ifndef _SYS_SYSPROTO_H_ 560struct gettimeofday_args { 561 struct timeval *tp; 562 struct timezone *tzp; 563}; 564#endif 565/* ARGSUSED */ 566int 567sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap) 568{ 569 struct timeval atv; 570 struct timezone rtz; 571 int error = 0; 572 573 if (uap->tp) { 574 microtime(&atv); 575 error = copyout(&atv, uap->tp, sizeof (atv)); 576 } 577 if (error == 0 && uap->tzp != NULL) { 578 rtz.tz_minuteswest = tz_minuteswest; 579 rtz.tz_dsttime = tz_dsttime; 580 error = copyout(&rtz, uap->tzp, sizeof (rtz)); 581 } 582 return (error); 583} 584 585#ifndef _SYS_SYSPROTO_H_ 586struct settimeofday_args { 587 struct timeval *tv; 588 struct timezone *tzp; 589}; 590#endif 591/* ARGSUSED */ 592int 593sys_settimeofday(struct thread *td, struct settimeofday_args *uap) 594{ 595 struct timeval atv, *tvp; 596 struct timezone atz, *tzp; 597 int error; 598 599 if (uap->tv) { 600 error = copyin(uap->tv, &atv, sizeof(atv)); 601 if (error) 602 return (error); 603 tvp = &atv; 604 } else 605 tvp = NULL; 606 if (uap->tzp) { 607 error = copyin(uap->tzp, &atz, sizeof(atz)); 608 if (error) 609 return (error); 610 tzp = &atz; 611 } else 612 tzp = NULL; 613 return (kern_settimeofday(td, tvp, tzp)); 614} 615 616int 617kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp) 618{ 619 int error; 620 621 error = priv_check(td, PRIV_SETTIMEOFDAY); 622 if (error) 623 return (error); 624 /* Verify all parameters before changing time. */ 625 if (tv) { 626 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000) 627 return (EINVAL); 628 error = settime(td, tv); 629 } 630 if (tzp && error == 0) { 631 tz_minuteswest = tzp->tz_minuteswest; 632 tz_dsttime = tzp->tz_dsttime; 633 } 634 return (error); 635} 636 637/* 638 * Get value of an interval timer. The process virtual and profiling virtual 639 * time timers are kept in the p_stats area, since they can be swapped out. 640 * These are kept internally in the way they are specified externally: in 641 * time until they expire. 642 * 643 * The real time interval timer is kept in the process table slot for the 644 * process, and its value (it_value) is kept as an absolute time rather than 645 * as a delta, so that it is easy to keep periodic real-time signals from 646 * drifting. 647 * 648 * Virtual time timers are processed in the hardclock() routine of 649 * kern_clock.c. The real time timer is processed by a timeout routine, 650 * called from the softclock() routine. Since a callout may be delayed in 651 * real time due to interrupt processing in the system, it is possible for 652 * the real time timeout routine (realitexpire, given below), to be delayed 653 * in real time past when it is supposed to occur. It does not suffice, 654 * therefore, to reload the real timer .it_value from the real time timers 655 * .it_interval. Rather, we compute the next time in absolute time the timer 656 * should go off. 657 */ 658#ifndef _SYS_SYSPROTO_H_ 659struct getitimer_args { 660 u_int which; 661 struct itimerval *itv; 662}; 663#endif 664int 665sys_getitimer(struct thread *td, struct getitimer_args *uap) 666{ 667 struct itimerval aitv; 668 int error; 669 670 error = kern_getitimer(td, uap->which, &aitv); 671 if (error != 0) 672 return (error); 673 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 674} 675 676int 677kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv) 678{ 679 struct proc *p = td->td_proc; 680 struct timeval ctv; 681 682 if (which > ITIMER_PROF) 683 return (EINVAL); 684 685 if (which == ITIMER_REAL) { 686 /* 687 * Convert from absolute to relative time in .it_value 688 * part of real time timer. If time for real time timer 689 * has passed return 0, else return difference between 690 * current time and time for the timer to go off. 691 */ 692 PROC_LOCK(p); 693 *aitv = p->p_realtimer; 694 PROC_UNLOCK(p); 695 if (timevalisset(&aitv->it_value)) { 696 microuptime(&ctv); 697 if (timevalcmp(&aitv->it_value, &ctv, <)) 698 timevalclear(&aitv->it_value); 699 else 700 timevalsub(&aitv->it_value, &ctv); 701 } 702 } else { 703 PROC_ITIMLOCK(p); 704 *aitv = p->p_stats->p_timer[which]; 705 PROC_ITIMUNLOCK(p); 706 } 707#ifdef KTRACE 708 if (KTRPOINT(td, KTR_STRUCT)) 709 ktritimerval(aitv); 710#endif 711 return (0); 712} 713 714#ifndef _SYS_SYSPROTO_H_ 715struct setitimer_args { 716 u_int which; 717 struct itimerval *itv, *oitv; 718}; 719#endif 720int 721sys_setitimer(struct thread *td, struct setitimer_args *uap) 722{ 723 struct itimerval aitv, oitv; 724 int error; 725 726 if (uap->itv == NULL) { 727 uap->itv = uap->oitv; 728 return (sys_getitimer(td, (struct getitimer_args *)uap)); 729 } 730 731 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval)))) 732 return (error); 733 error = kern_setitimer(td, uap->which, &aitv, &oitv); 734 if (error != 0 || uap->oitv == NULL) 735 return (error); 736 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval))); 737} 738 739int 740kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv, 741 struct itimerval *oitv) 742{ 743 struct proc *p = td->td_proc; 744 struct timeval ctv; 745 sbintime_t sbt, pr; 746 747 if (aitv == NULL) 748 return (kern_getitimer(td, which, oitv)); 749 750 if (which > ITIMER_PROF) 751 return (EINVAL); 752#ifdef KTRACE 753 if (KTRPOINT(td, KTR_STRUCT)) 754 ktritimerval(aitv); 755#endif 756 if (itimerfix(&aitv->it_value) || 757 aitv->it_value.tv_sec > INT32_MAX / 2) 758 return (EINVAL); 759 if (!timevalisset(&aitv->it_value)) 760 timevalclear(&aitv->it_interval); 761 else if (itimerfix(&aitv->it_interval) || 762 aitv->it_interval.tv_sec > INT32_MAX / 2) 763 return (EINVAL); 764 765 if (which == ITIMER_REAL) { 766 PROC_LOCK(p); 767 if (timevalisset(&p->p_realtimer.it_value)) 768 callout_stop(&p->p_itcallout); 769 microuptime(&ctv); 770 if (timevalisset(&aitv->it_value)) { 771 pr = tvtosbt(aitv->it_value) >> tc_precexp; 772 timevaladd(&aitv->it_value, &ctv); 773 sbt = tvtosbt(aitv->it_value); 774 callout_reset_sbt(&p->p_itcallout, sbt, pr, 775 realitexpire, p, C_ABSOLUTE); 776 } 777 *oitv = p->p_realtimer; 778 p->p_realtimer = *aitv; 779 PROC_UNLOCK(p); 780 if (timevalisset(&oitv->it_value)) { 781 if (timevalcmp(&oitv->it_value, &ctv, <)) 782 timevalclear(&oitv->it_value); 783 else 784 timevalsub(&oitv->it_value, &ctv); 785 } 786 } else { 787 if (aitv->it_interval.tv_sec == 0 && 788 aitv->it_interval.tv_usec != 0 && 789 aitv->it_interval.tv_usec < tick) 790 aitv->it_interval.tv_usec = tick; 791 if (aitv->it_value.tv_sec == 0 && 792 aitv->it_value.tv_usec != 0 && 793 aitv->it_value.tv_usec < tick) 794 aitv->it_value.tv_usec = tick; 795 PROC_ITIMLOCK(p); 796 *oitv = p->p_stats->p_timer[which]; 797 p->p_stats->p_timer[which] = *aitv; 798 PROC_ITIMUNLOCK(p); 799 } 800#ifdef KTRACE 801 if (KTRPOINT(td, KTR_STRUCT)) 802 ktritimerval(oitv); 803#endif 804 return (0); 805} 806 807/* 808 * Real interval timer expired: 809 * send process whose timer expired an alarm signal. 810 * If time is not set up to reload, then just return. 811 * Else compute next time timer should go off which is > current time. 812 * This is where delay in processing this timeout causes multiple 813 * SIGALRM calls to be compressed into one. 814 * tvtohz() always adds 1 to allow for the time until the next clock 815 * interrupt being strictly less than 1 clock tick, but we don't want 816 * that here since we want to appear to be in sync with the clock 817 * interrupt even when we're delayed. 818 */ 819void 820realitexpire(void *arg) 821{ 822 struct proc *p; 823 struct timeval ctv; 824 sbintime_t isbt; 825 826 p = (struct proc *)arg; 827 kern_psignal(p, SIGALRM); 828 if (!timevalisset(&p->p_realtimer.it_interval)) { 829 timevalclear(&p->p_realtimer.it_value); 830 if (p->p_flag & P_WEXIT) 831 wakeup(&p->p_itcallout); 832 return; 833 } 834 isbt = tvtosbt(p->p_realtimer.it_interval); 835 if (isbt >= sbt_timethreshold) 836 getmicrouptime(&ctv); 837 else 838 microuptime(&ctv); 839 do { 840 timevaladd(&p->p_realtimer.it_value, 841 &p->p_realtimer.it_interval); 842 } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=)); 843 callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value), 844 isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE); 845} 846 847/* 848 * Check that a proposed value to load into the .it_value or 849 * .it_interval part of an interval timer is acceptable, and 850 * fix it to have at least minimal value (i.e. if it is less 851 * than the resolution of the clock, round it up.) 852 */ 853int 854itimerfix(struct timeval *tv) 855{ 856 857 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 858 return (EINVAL); 859 if (tv->tv_sec == 0 && tv->tv_usec != 0 && 860 tv->tv_usec < (u_int)tick / 16) 861 tv->tv_usec = (u_int)tick / 16; 862 return (0); 863} 864 865/* 866 * Decrement an interval timer by a specified number 867 * of microseconds, which must be less than a second, 868 * i.e. < 1000000. If the timer expires, then reload 869 * it. In this case, carry over (usec - old value) to 870 * reduce the value reloaded into the timer so that 871 * the timer does not drift. This routine assumes 872 * that it is called in a context where the timers 873 * on which it is operating cannot change in value. 874 */ 875int 876itimerdecr(struct itimerval *itp, int usec) 877{ 878 879 if (itp->it_value.tv_usec < usec) { 880 if (itp->it_value.tv_sec == 0) { 881 /* expired, and already in next interval */ 882 usec -= itp->it_value.tv_usec; 883 goto expire; 884 } 885 itp->it_value.tv_usec += 1000000; 886 itp->it_value.tv_sec--; 887 } 888 itp->it_value.tv_usec -= usec; 889 usec = 0; 890 if (timevalisset(&itp->it_value)) 891 return (1); 892 /* expired, exactly at end of interval */ 893expire: 894 if (timevalisset(&itp->it_interval)) { 895 itp->it_value = itp->it_interval; 896 itp->it_value.tv_usec -= usec; 897 if (itp->it_value.tv_usec < 0) { 898 itp->it_value.tv_usec += 1000000; 899 itp->it_value.tv_sec--; 900 } 901 } else 902 itp->it_value.tv_usec = 0; /* sec is already 0 */ 903 return (0); 904} 905 906/* 907 * Add and subtract routines for timevals. 908 * N.B.: subtract routine doesn't deal with 909 * results which are before the beginning, 910 * it just gets very confused in this case. 911 * Caveat emptor. 912 */ 913void 914timevaladd(struct timeval *t1, const struct timeval *t2) 915{ 916 917 t1->tv_sec += t2->tv_sec; 918 t1->tv_usec += t2->tv_usec; 919 timevalfix(t1); 920} 921 922void 923timevalsub(struct timeval *t1, const struct timeval *t2) 924{ 925 926 t1->tv_sec -= t2->tv_sec; 927 t1->tv_usec -= t2->tv_usec; 928 timevalfix(t1); 929} 930 931static void 932timevalfix(struct timeval *t1) 933{ 934 935 if (t1->tv_usec < 0) { 936 t1->tv_sec--; 937 t1->tv_usec += 1000000; 938 } 939 if (t1->tv_usec >= 1000000) { 940 t1->tv_sec++; 941 t1->tv_usec -= 1000000; 942 } 943} 944 945/* 946 * ratecheck(): simple time-based rate-limit checking. 947 */ 948int 949ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 950{ 951 struct timeval tv, delta; 952 int rv = 0; 953 954 getmicrouptime(&tv); /* NB: 10ms precision */ 955 delta = tv; 956 timevalsub(&delta, lasttime); 957 958 /* 959 * check for 0,0 is so that the message will be seen at least once, 960 * even if interval is huge. 961 */ 962 if (timevalcmp(&delta, mininterval, >=) || 963 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 964 *lasttime = tv; 965 rv = 1; 966 } 967 968 return (rv); 969} 970 971/* 972 * ppsratecheck(): packets (or events) per second limitation. 973 * 974 * Return 0 if the limit is to be enforced (e.g. the caller 975 * should drop a packet because of the rate limitation). 976 * 977 * maxpps of 0 always causes zero to be returned. maxpps of -1 978 * always causes 1 to be returned; this effectively defeats rate 979 * limiting. 980 * 981 * Note that we maintain the struct timeval for compatibility 982 * with other bsd systems. We reuse the storage and just monitor 983 * clock ticks for minimal overhead. 984 */ 985int 986ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 987{ 988 int now; 989 990 /* 991 * Reset the last time and counter if this is the first call 992 * or more than a second has passed since the last update of 993 * lasttime. 994 */ 995 now = ticks; 996 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 997 lasttime->tv_sec = now; 998 *curpps = 1; 999 return (maxpps != 0); 1000 } else { 1001 (*curpps)++; /* NB: ignore potential overflow */ 1002 return (maxpps < 0 || *curpps <= maxpps); 1003 } 1004} 1005 1006static void 1007itimer_start(void) 1008{ 1009 struct kclock rt_clock = { 1010 .timer_create = realtimer_create, 1011 .timer_delete = realtimer_delete, 1012 .timer_settime = realtimer_settime, 1013 .timer_gettime = realtimer_gettime, 1014 .event_hook = NULL 1015 }; 1016 1017 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer), 1018 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0); 1019 register_posix_clock(CLOCK_REALTIME, &rt_clock); 1020 register_posix_clock(CLOCK_MONOTONIC, &rt_clock); 1021 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L); 1022 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX); 1023 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX); 1024 EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit, 1025 (void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY); 1026 EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec, 1027 (void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY); 1028} 1029 1030int 1031register_posix_clock(int clockid, struct kclock *clk) 1032{ 1033 if ((unsigned)clockid >= MAX_CLOCKS) { 1034 printf("%s: invalid clockid\n", __func__); 1035 return (0); 1036 } 1037 posix_clocks[clockid] = *clk; 1038 return (1); 1039} 1040 1041static int 1042itimer_init(void *mem, int size, int flags) 1043{ 1044 struct itimer *it; 1045 1046 it = (struct itimer *)mem; 1047 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF); 1048 return (0); 1049} 1050 1051static void 1052itimer_fini(void *mem, int size) 1053{ 1054 struct itimer *it; 1055 1056 it = (struct itimer *)mem; 1057 mtx_destroy(&it->it_mtx); 1058} 1059 1060static void 1061itimer_enter(struct itimer *it) 1062{ 1063 1064 mtx_assert(&it->it_mtx, MA_OWNED); 1065 it->it_usecount++; 1066} 1067 1068static void 1069itimer_leave(struct itimer *it) 1070{ 1071 1072 mtx_assert(&it->it_mtx, MA_OWNED); 1073 KASSERT(it->it_usecount > 0, ("invalid it_usecount")); 1074 1075 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0) 1076 wakeup(it); 1077} 1078 1079#ifndef _SYS_SYSPROTO_H_ 1080struct ktimer_create_args { 1081 clockid_t clock_id; 1082 struct sigevent * evp; 1083 int * timerid; 1084}; 1085#endif 1086int 1087sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap) 1088{ 1089 struct sigevent *evp, ev; 1090 int id; 1091 int error; 1092 1093 if (uap->evp == NULL) { 1094 evp = NULL; 1095 } else { 1096 error = copyin(uap->evp, &ev, sizeof(ev)); 1097 if (error != 0) 1098 return (error); 1099 evp = &ev; 1100 } 1101 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1); 1102 if (error == 0) { 1103 error = copyout(&id, uap->timerid, sizeof(int)); 1104 if (error != 0) 1105 kern_ktimer_delete(td, id); 1106 } 1107 return (error); 1108} 1109 1110int 1111kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp, 1112 int *timerid, int preset_id) 1113{ 1114 struct proc *p = td->td_proc; 1115 struct itimer *it; 1116 int id; 1117 int error; 1118 1119 if (clock_id < 0 || clock_id >= MAX_CLOCKS) 1120 return (EINVAL); 1121 1122 if (posix_clocks[clock_id].timer_create == NULL) 1123 return (EINVAL); 1124 1125 if (evp != NULL) { 1126 if (evp->sigev_notify != SIGEV_NONE && 1127 evp->sigev_notify != SIGEV_SIGNAL && 1128 evp->sigev_notify != SIGEV_THREAD_ID) 1129 return (EINVAL); 1130 if ((evp->sigev_notify == SIGEV_SIGNAL || 1131 evp->sigev_notify == SIGEV_THREAD_ID) && 1132 !_SIG_VALID(evp->sigev_signo)) 1133 return (EINVAL); 1134 } 1135 1136 if (p->p_itimers == NULL) 1137 itimers_alloc(p); 1138 1139 it = uma_zalloc(itimer_zone, M_WAITOK); 1140 it->it_flags = 0; 1141 it->it_usecount = 0; 1142 it->it_active = 0; 1143 timespecclear(&it->it_time.it_value); 1144 timespecclear(&it->it_time.it_interval); 1145 it->it_overrun = 0; 1146 it->it_overrun_last = 0; 1147 it->it_clockid = clock_id; 1148 it->it_timerid = -1; 1149 it->it_proc = p; 1150 ksiginfo_init(&it->it_ksi); 1151 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT; 1152 error = CLOCK_CALL(clock_id, timer_create, (it)); 1153 if (error != 0) 1154 goto out; 1155 1156 PROC_LOCK(p); 1157 if (preset_id != -1) { 1158 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id")); 1159 id = preset_id; 1160 if (p->p_itimers->its_timers[id] != NULL) { 1161 PROC_UNLOCK(p); 1162 error = 0; 1163 goto out; 1164 } 1165 } else { 1166 /* 1167 * Find a free timer slot, skipping those reserved 1168 * for setitimer(). 1169 */ 1170 for (id = 3; id < TIMER_MAX; id++) 1171 if (p->p_itimers->its_timers[id] == NULL) 1172 break; 1173 if (id == TIMER_MAX) { 1174 PROC_UNLOCK(p); 1175 error = EAGAIN; 1176 goto out; 1177 } 1178 } 1179 it->it_timerid = id; 1180 p->p_itimers->its_timers[id] = it; 1181 if (evp != NULL) 1182 it->it_sigev = *evp; 1183 else { 1184 it->it_sigev.sigev_notify = SIGEV_SIGNAL; 1185 switch (clock_id) { 1186 default: 1187 case CLOCK_REALTIME: 1188 it->it_sigev.sigev_signo = SIGALRM; 1189 break; 1190 case CLOCK_VIRTUAL: 1191 it->it_sigev.sigev_signo = SIGVTALRM; 1192 break; 1193 case CLOCK_PROF: 1194 it->it_sigev.sigev_signo = SIGPROF; 1195 break; 1196 } 1197 it->it_sigev.sigev_value.sival_int = id; 1198 } 1199 1200 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1201 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1202 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo; 1203 it->it_ksi.ksi_code = SI_TIMER; 1204 it->it_ksi.ksi_value = it->it_sigev.sigev_value; 1205 it->it_ksi.ksi_timerid = id; 1206 } 1207 PROC_UNLOCK(p); 1208 *timerid = id; 1209 return (0); 1210 1211out: 1212 ITIMER_LOCK(it); 1213 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1214 ITIMER_UNLOCK(it); 1215 uma_zfree(itimer_zone, it); 1216 return (error); 1217} 1218 1219#ifndef _SYS_SYSPROTO_H_ 1220struct ktimer_delete_args { 1221 int timerid; 1222}; 1223#endif 1224int 1225sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap) 1226{ 1227 1228 return (kern_ktimer_delete(td, uap->timerid)); 1229} 1230 1231static struct itimer * 1232itimer_find(struct proc *p, int timerid) 1233{ 1234 struct itimer *it; 1235 1236 PROC_LOCK_ASSERT(p, MA_OWNED); 1237 if ((p->p_itimers == NULL) || 1238 (timerid < 0) || (timerid >= TIMER_MAX) || 1239 (it = p->p_itimers->its_timers[timerid]) == NULL) { 1240 return (NULL); 1241 } 1242 ITIMER_LOCK(it); 1243 if ((it->it_flags & ITF_DELETING) != 0) { 1244 ITIMER_UNLOCK(it); 1245 it = NULL; 1246 } 1247 return (it); 1248} 1249 1250int 1251kern_ktimer_delete(struct thread *td, int timerid) 1252{ 1253 struct proc *p = td->td_proc; 1254 struct itimer *it; 1255 1256 PROC_LOCK(p); 1257 it = itimer_find(p, timerid); 1258 if (it == NULL) { 1259 PROC_UNLOCK(p); 1260 return (EINVAL); 1261 } 1262 PROC_UNLOCK(p); 1263 1264 it->it_flags |= ITF_DELETING; 1265 while (it->it_usecount > 0) { 1266 it->it_flags |= ITF_WANTED; 1267 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0); 1268 } 1269 it->it_flags &= ~ITF_WANTED; 1270 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1271 ITIMER_UNLOCK(it); 1272 1273 PROC_LOCK(p); 1274 if (KSI_ONQ(&it->it_ksi)) 1275 sigqueue_take(&it->it_ksi); 1276 p->p_itimers->its_timers[timerid] = NULL; 1277 PROC_UNLOCK(p); 1278 uma_zfree(itimer_zone, it); 1279 return (0); 1280} 1281 1282#ifndef _SYS_SYSPROTO_H_ 1283struct ktimer_settime_args { 1284 int timerid; 1285 int flags; 1286 const struct itimerspec * value; 1287 struct itimerspec * ovalue; 1288}; 1289#endif 1290int 1291sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap) 1292{ 1293 struct itimerspec val, oval, *ovalp; 1294 int error; 1295 1296 error = copyin(uap->value, &val, sizeof(val)); 1297 if (error != 0) 1298 return (error); 1299 ovalp = uap->ovalue != NULL ? &oval : NULL; 1300 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp); 1301 if (error == 0 && uap->ovalue != NULL) 1302 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp)); 1303 return (error); 1304} 1305 1306int 1307kern_ktimer_settime(struct thread *td, int timer_id, int flags, 1308 struct itimerspec *val, struct itimerspec *oval) 1309{ 1310 struct proc *p; 1311 struct itimer *it; 1312 int error; 1313 1314 p = td->td_proc; 1315 PROC_LOCK(p); 1316 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1317 PROC_UNLOCK(p); 1318 error = EINVAL; 1319 } else { 1320 PROC_UNLOCK(p); 1321 itimer_enter(it); 1322 error = CLOCK_CALL(it->it_clockid, timer_settime, (it, 1323 flags, val, oval)); 1324 itimer_leave(it); 1325 ITIMER_UNLOCK(it); 1326 } 1327 return (error); 1328} 1329 1330#ifndef _SYS_SYSPROTO_H_ 1331struct ktimer_gettime_args { 1332 int timerid; 1333 struct itimerspec * value; 1334}; 1335#endif 1336int 1337sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap) 1338{ 1339 struct itimerspec val; 1340 int error; 1341 1342 error = kern_ktimer_gettime(td, uap->timerid, &val); 1343 if (error == 0) 1344 error = copyout(&val, uap->value, sizeof(val)); 1345 return (error); 1346} 1347 1348int 1349kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val) 1350{ 1351 struct proc *p; 1352 struct itimer *it; 1353 int error; 1354 1355 p = td->td_proc; 1356 PROC_LOCK(p); 1357 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1358 PROC_UNLOCK(p); 1359 error = EINVAL; 1360 } else { 1361 PROC_UNLOCK(p); 1362 itimer_enter(it); 1363 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val)); 1364 itimer_leave(it); 1365 ITIMER_UNLOCK(it); 1366 } 1367 return (error); 1368} 1369 1370#ifndef _SYS_SYSPROTO_H_ 1371struct timer_getoverrun_args { 1372 int timerid; 1373}; 1374#endif 1375int 1376sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap) 1377{ 1378 1379 return (kern_ktimer_getoverrun(td, uap->timerid)); 1380} 1381 1382int 1383kern_ktimer_getoverrun(struct thread *td, int timer_id) 1384{ 1385 struct proc *p = td->td_proc; 1386 struct itimer *it; 1387 int error ; 1388 1389 PROC_LOCK(p); 1390 if (timer_id < 3 || 1391 (it = itimer_find(p, timer_id)) == NULL) { 1392 PROC_UNLOCK(p); 1393 error = EINVAL; 1394 } else { 1395 td->td_retval[0] = it->it_overrun_last; 1396 ITIMER_UNLOCK(it); 1397 PROC_UNLOCK(p); 1398 error = 0; 1399 } 1400 return (error); 1401} 1402 1403static int 1404realtimer_create(struct itimer *it) 1405{ 1406 callout_init_mtx(&it->it_callout, &it->it_mtx, 0); 1407 return (0); 1408} 1409 1410static int 1411realtimer_delete(struct itimer *it) 1412{ 1413 mtx_assert(&it->it_mtx, MA_OWNED); 1414 1415 /* 1416 * clear timer's value and interval to tell realtimer_expire 1417 * to not rearm the timer. 1418 */ 1419 timespecclear(&it->it_time.it_value); 1420 timespecclear(&it->it_time.it_interval); 1421 ITIMER_UNLOCK(it); 1422 callout_drain(&it->it_callout); 1423 ITIMER_LOCK(it); 1424 return (0); 1425} 1426 1427static int 1428realtimer_gettime(struct itimer *it, struct itimerspec *ovalue) 1429{ 1430 struct timespec cts; 1431 1432 mtx_assert(&it->it_mtx, MA_OWNED); 1433 1434 realtimer_clocktime(it->it_clockid, &cts); 1435 *ovalue = it->it_time; 1436 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) { 1437 timespecsub(&ovalue->it_value, &cts); 1438 if (ovalue->it_value.tv_sec < 0 || 1439 (ovalue->it_value.tv_sec == 0 && 1440 ovalue->it_value.tv_nsec == 0)) { 1441 ovalue->it_value.tv_sec = 0; 1442 ovalue->it_value.tv_nsec = 1; 1443 } 1444 } 1445 return (0); 1446} 1447 1448static int 1449realtimer_settime(struct itimer *it, int flags, 1450 struct itimerspec *value, struct itimerspec *ovalue) 1451{ 1452 struct timespec cts, ts; 1453 struct timeval tv; 1454 struct itimerspec val; 1455 1456 mtx_assert(&it->it_mtx, MA_OWNED); 1457 1458 val = *value; 1459 if (itimespecfix(&val.it_value)) 1460 return (EINVAL); 1461 1462 if (timespecisset(&val.it_value)) { 1463 if (itimespecfix(&val.it_interval)) 1464 return (EINVAL); 1465 } else { 1466 timespecclear(&val.it_interval); 1467 } 1468 1469 if (ovalue != NULL) 1470 realtimer_gettime(it, ovalue); 1471 1472 it->it_time = val; 1473 if (timespecisset(&val.it_value)) { 1474 realtimer_clocktime(it->it_clockid, &cts); 1475 ts = val.it_value; 1476 if ((flags & TIMER_ABSTIME) == 0) { 1477 /* Convert to absolute time. */ 1478 timespecadd(&it->it_time.it_value, &cts); 1479 } else { 1480 timespecsub(&ts, &cts); 1481 /* 1482 * We don't care if ts is negative, tztohz will 1483 * fix it. 1484 */ 1485 } 1486 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1487 callout_reset(&it->it_callout, tvtohz(&tv), 1488 realtimer_expire, it); 1489 } else { 1490 callout_stop(&it->it_callout); 1491 } 1492 1493 return (0); 1494} 1495 1496static void 1497realtimer_clocktime(clockid_t id, struct timespec *ts) 1498{ 1499 if (id == CLOCK_REALTIME) 1500 getnanotime(ts); 1501 else /* CLOCK_MONOTONIC */ 1502 getnanouptime(ts); 1503} 1504 1505int 1506itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi) 1507{ 1508 struct itimer *it; 1509 1510 PROC_LOCK_ASSERT(p, MA_OWNED); 1511 it = itimer_find(p, timerid); 1512 if (it != NULL) { 1513 ksi->ksi_overrun = it->it_overrun; 1514 it->it_overrun_last = it->it_overrun; 1515 it->it_overrun = 0; 1516 ITIMER_UNLOCK(it); 1517 return (0); 1518 } 1519 return (EINVAL); 1520} 1521 1522int 1523itimespecfix(struct timespec *ts) 1524{ 1525 1526 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000) 1527 return (EINVAL); 1528 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000) 1529 ts->tv_nsec = tick * 1000; 1530 return (0); 1531} 1532 1533/* Timeout callback for realtime timer */ 1534static void 1535realtimer_expire(void *arg) 1536{ 1537 struct timespec cts, ts; 1538 struct timeval tv; 1539 struct itimer *it; 1540 1541 it = (struct itimer *)arg; 1542 1543 realtimer_clocktime(it->it_clockid, &cts); 1544 /* Only fire if time is reached. */ 1545 if (timespeccmp(&cts, &it->it_time.it_value, >=)) { 1546 if (timespecisset(&it->it_time.it_interval)) { 1547 timespecadd(&it->it_time.it_value, 1548 &it->it_time.it_interval); 1549 while (timespeccmp(&cts, &it->it_time.it_value, >=)) { 1550 if (it->it_overrun < INT_MAX) 1551 it->it_overrun++; 1552 else 1553 it->it_ksi.ksi_errno = ERANGE; 1554 timespecadd(&it->it_time.it_value, 1555 &it->it_time.it_interval); 1556 } 1557 } else { 1558 /* single shot timer ? */ 1559 timespecclear(&it->it_time.it_value); 1560 } 1561 if (timespecisset(&it->it_time.it_value)) { 1562 ts = it->it_time.it_value; 1563 timespecsub(&ts, &cts); 1564 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1565 callout_reset(&it->it_callout, tvtohz(&tv), 1566 realtimer_expire, it); 1567 } 1568 itimer_enter(it); 1569 ITIMER_UNLOCK(it); 1570 itimer_fire(it); 1571 ITIMER_LOCK(it); 1572 itimer_leave(it); 1573 } else if (timespecisset(&it->it_time.it_value)) { 1574 ts = it->it_time.it_value; 1575 timespecsub(&ts, &cts); 1576 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1577 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire, 1578 it); 1579 } 1580} 1581 1582void 1583itimer_fire(struct itimer *it) 1584{ 1585 struct proc *p = it->it_proc; 1586 struct thread *td; 1587 1588 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1589 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1590 if (sigev_findtd(p, &it->it_sigev, &td) != 0) { 1591 ITIMER_LOCK(it); 1592 timespecclear(&it->it_time.it_value); 1593 timespecclear(&it->it_time.it_interval); 1594 callout_stop(&it->it_callout); 1595 ITIMER_UNLOCK(it); 1596 return; 1597 } 1598 if (!KSI_ONQ(&it->it_ksi)) { 1599 it->it_ksi.ksi_errno = 0; 1600 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev); 1601 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi); 1602 } else { 1603 if (it->it_overrun < INT_MAX) 1604 it->it_overrun++; 1605 else 1606 it->it_ksi.ksi_errno = ERANGE; 1607 } 1608 PROC_UNLOCK(p); 1609 } 1610} 1611 1612static void 1613itimers_alloc(struct proc *p) 1614{ 1615 struct itimers *its; 1616 int i; 1617 1618 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO); 1619 LIST_INIT(&its->its_virtual); 1620 LIST_INIT(&its->its_prof); 1621 TAILQ_INIT(&its->its_worklist); 1622 for (i = 0; i < TIMER_MAX; i++) 1623 its->its_timers[i] = NULL; 1624 PROC_LOCK(p); 1625 if (p->p_itimers == NULL) { 1626 p->p_itimers = its; 1627 PROC_UNLOCK(p); 1628 } 1629 else { 1630 PROC_UNLOCK(p); 1631 free(its, M_SUBPROC); 1632 } 1633} 1634 1635static void 1636itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused) 1637{ 1638 itimers_event_hook_exit(arg, p); 1639} 1640 1641/* Clean up timers when some process events are being triggered. */ 1642static void 1643itimers_event_hook_exit(void *arg, struct proc *p) 1644{ 1645 struct itimers *its; 1646 struct itimer *it; 1647 int event = (int)(intptr_t)arg; 1648 int i; 1649 1650 if (p->p_itimers != NULL) { 1651 its = p->p_itimers; 1652 for (i = 0; i < MAX_CLOCKS; ++i) { 1653 if (posix_clocks[i].event_hook != NULL) 1654 CLOCK_CALL(i, event_hook, (p, i, event)); 1655 } 1656 /* 1657 * According to susv3, XSI interval timers should be inherited 1658 * by new image. 1659 */ 1660 if (event == ITIMER_EV_EXEC) 1661 i = 3; 1662 else if (event == ITIMER_EV_EXIT) 1663 i = 0; 1664 else 1665 panic("unhandled event"); 1666 for (; i < TIMER_MAX; ++i) { 1667 if ((it = its->its_timers[i]) != NULL) 1668 kern_ktimer_delete(curthread, i); 1669 } 1670 if (its->its_timers[0] == NULL && 1671 its->its_timers[1] == NULL && 1672 its->its_timers[2] == NULL) { 1673 free(its, M_SUBPROC); 1674 p->p_itimers = NULL; 1675 } 1676 } 1677} 1678