1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 #include "opt_ktrace.h" 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/limits.h> 38 #include <sys/clock.h> 39 #include <sys/lock.h> 40 #include <sys/mutex.h> 41 #include <sys/sysproto.h> 42 #include <sys/resourcevar.h> 43 #include <sys/signalvar.h> 44 #include <sys/kernel.h> 45 #include <sys/sleepqueue.h> 46 #include <sys/syscallsubr.h> 47 #include <sys/sysctl.h> 48 #include <sys/priv.h> 49 #include <sys/proc.h> 50 #include <sys/posix4.h> 51 #include <sys/time.h> 52 #include <sys/timers.h> 53 #include <sys/timetc.h> 54 #include <sys/vnode.h> 55 #ifdef KTRACE 56 #include <sys/ktrace.h> 57 #endif 58 59 #include <vm/vm.h> 60 #include <vm/vm_extern.h> 61 #include <vm/uma.h> 62 63 #define MAX_CLOCKS (CLOCK_MONOTONIC+1) 64 #define CPUCLOCK_BIT 0x80000000 65 #define CPUCLOCK_PROCESS_BIT 0x40000000 66 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT)) 67 #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid)) 68 #define MAKE_PROCESS_CPUCLOCK(pid) \ 69 (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid)) 70 71 #define NS_PER_SEC 1000000000 72 73 static struct kclock posix_clocks[MAX_CLOCKS]; 74 static uma_zone_t itimer_zone = NULL; 75 76 /* 77 * Time of day and interval timer support. 78 * 79 * These routines provide the kernel entry points to get and set 80 * the time-of-day and per-process interval timers. Subroutines 81 * here provide support for adding and subtracting timeval structures 82 * and decrementing interval timers, optionally reloading the interval 83 * timers when they expire. 84 */ 85 86 static int settime(struct thread *, struct timeval *); 87 static void timevalfix(struct timeval *); 88 static int user_clock_nanosleep(struct thread *td, clockid_t clock_id, 89 int flags, const struct timespec *ua_rqtp, 90 struct timespec *ua_rmtp); 91 92 static void itimer_start(void); 93 static int itimer_init(void *, int, int); 94 static void itimer_fini(void *, int); 95 static void itimer_enter(struct itimer *); 96 static void itimer_leave(struct itimer *); 97 static struct itimer *itimer_find(struct proc *, int); 98 static void itimers_alloc(struct proc *); 99 static int realtimer_create(struct itimer *); 100 static int realtimer_gettime(struct itimer *, struct itimerspec *); 101 static int realtimer_settime(struct itimer *, int, 102 struct itimerspec *, struct itimerspec *); 103 static int realtimer_delete(struct itimer *); 104 static void realtimer_clocktime(clockid_t, struct timespec *); 105 static void realtimer_expire(void *); 106 static void realtimer_expire_l(struct itimer *it, bool proc_locked); 107 108 static void realitexpire(void *arg); 109 110 static int register_posix_clock(int, const struct kclock *); 111 static void itimer_fire(struct itimer *it); 112 static 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 static int 120 settime(struct thread *td, struct timeval *tv) 121 { 122 struct timeval delta, tv1, tv2; 123 static struct timeval maxtime, laststep; 124 struct timespec ts; 125 126 microtime(&tv1); 127 delta = *tv; 128 timevalsub(&delta, &tv1); 129 130 /* 131 * If the system is secure, we do not allow the time to be 132 * set to a value earlier than 1 second less than the highest 133 * time we have yet seen. The worst a miscreant can do in 134 * this circumstance is "freeze" time. He couldn't go 135 * back to the past. 136 * 137 * We similarly do not allow the clock to be stepped more 138 * than one second, nor more than once per second. This allows 139 * a miscreant to make the clock march double-time, but no worse. 140 */ 141 if (securelevel_gt(td->td_ucred, 1) != 0) { 142 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 return (EPERM); 157 if (delta.tv_sec > 1) { 158 tv->tv_sec = tv1.tv_sec + 1; 159 printf("Time adjustment clamped to +1 second\n"); 160 } 161 laststep = *tv; 162 } 163 } 164 165 ts.tv_sec = tv->tv_sec; 166 ts.tv_nsec = tv->tv_usec * 1000; 167 tc_setclock(&ts); 168 resettodr(); 169 return (0); 170 } 171 172 #ifndef _SYS_SYSPROTO_H_ 173 struct clock_getcpuclockid2_args { 174 id_t id; 175 int which, 176 clockid_t *clock_id; 177 }; 178 #endif 179 /* ARGSUSED */ 180 int 181 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap) 182 { 183 clockid_t clk_id; 184 int error; 185 186 error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id); 187 if (error == 0) 188 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); 189 return (error); 190 } 191 192 int 193 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which, 194 clockid_t *clk_id) 195 { 196 struct proc *p; 197 pid_t pid; 198 lwpid_t tid; 199 int error; 200 201 switch (which) { 202 case CPUCLOCK_WHICH_PID: 203 if (id != 0) { 204 error = pget(id, PGET_CANSEE | PGET_NOTID, &p); 205 if (error != 0) 206 return (error); 207 PROC_UNLOCK(p); 208 pid = id; 209 } else { 210 pid = td->td_proc->p_pid; 211 } 212 *clk_id = MAKE_PROCESS_CPUCLOCK(pid); 213 return (0); 214 case CPUCLOCK_WHICH_TID: 215 tid = id == 0 ? td->td_tid : id; 216 *clk_id = MAKE_THREAD_CPUCLOCK(tid); 217 return (0); 218 default: 219 return (EINVAL); 220 } 221 } 222 223 #ifndef _SYS_SYSPROTO_H_ 224 struct clock_gettime_args { 225 clockid_t clock_id; 226 struct timespec *tp; 227 }; 228 #endif 229 /* ARGSUSED */ 230 int 231 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap) 232 { 233 struct timespec ats; 234 int error; 235 236 error = kern_clock_gettime(td, uap->clock_id, &ats); 237 if (error == 0) 238 error = copyout(&ats, uap->tp, sizeof(ats)); 239 240 return (error); 241 } 242 243 static inline void 244 cputick2timespec(uint64_t runtime, struct timespec *ats) 245 { 246 uint64_t tr; 247 tr = cpu_tickrate(); 248 ats->tv_sec = runtime / tr; 249 ats->tv_nsec = ((runtime % tr) * 1000000000ULL) / tr; 250 } 251 252 void 253 kern_thread_cputime(struct thread *targettd, struct timespec *ats) 254 { 255 uint64_t runtime, curtime, switchtime; 256 257 if (targettd == NULL) { /* current thread */ 258 spinlock_enter(); 259 switchtime = PCPU_GET(switchtime); 260 curtime = cpu_ticks(); 261 runtime = curthread->td_runtime; 262 spinlock_exit(); 263 runtime += curtime - switchtime; 264 } else { 265 PROC_LOCK_ASSERT(targettd->td_proc, MA_OWNED); 266 thread_lock(targettd); 267 runtime = targettd->td_runtime; 268 thread_unlock(targettd); 269 } 270 cputick2timespec(runtime, ats); 271 } 272 273 void 274 kern_process_cputime(struct proc *targetp, struct timespec *ats) 275 { 276 uint64_t runtime; 277 struct rusage ru; 278 279 PROC_LOCK_ASSERT(targetp, MA_OWNED); 280 PROC_STATLOCK(targetp); 281 rufetch(targetp, &ru); 282 runtime = targetp->p_rux.rux_runtime; 283 if (curthread->td_proc == targetp) 284 runtime += cpu_ticks() - PCPU_GET(switchtime); 285 PROC_STATUNLOCK(targetp); 286 cputick2timespec(runtime, ats); 287 } 288 289 static int 290 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats) 291 { 292 struct proc *p, *p2; 293 struct thread *td2; 294 lwpid_t tid; 295 pid_t pid; 296 int error; 297 298 p = td->td_proc; 299 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) { 300 tid = clock_id & CPUCLOCK_ID_MASK; 301 td2 = tdfind(tid, p->p_pid); 302 if (td2 == NULL) 303 return (EINVAL); 304 kern_thread_cputime(td2, ats); 305 PROC_UNLOCK(td2->td_proc); 306 } else { 307 pid = clock_id & CPUCLOCK_ID_MASK; 308 error = pget(pid, PGET_CANSEE, &p2); 309 if (error != 0) 310 return (EINVAL); 311 kern_process_cputime(p2, ats); 312 PROC_UNLOCK(p2); 313 } 314 return (0); 315 } 316 317 int 318 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats) 319 { 320 struct timeval sys, user; 321 struct proc *p; 322 323 p = td->td_proc; 324 switch (clock_id) { 325 case CLOCK_REALTIME: /* Default to precise. */ 326 case CLOCK_REALTIME_PRECISE: 327 nanotime(ats); 328 break; 329 case CLOCK_REALTIME_FAST: 330 getnanotime(ats); 331 break; 332 case CLOCK_VIRTUAL: 333 PROC_LOCK(p); 334 PROC_STATLOCK(p); 335 calcru(p, &user, &sys); 336 PROC_STATUNLOCK(p); 337 PROC_UNLOCK(p); 338 TIMEVAL_TO_TIMESPEC(&user, ats); 339 break; 340 case CLOCK_PROF: 341 PROC_LOCK(p); 342 PROC_STATLOCK(p); 343 calcru(p, &user, &sys); 344 PROC_STATUNLOCK(p); 345 PROC_UNLOCK(p); 346 timevaladd(&user, &sys); 347 TIMEVAL_TO_TIMESPEC(&user, ats); 348 break; 349 case CLOCK_MONOTONIC: /* Default to precise. */ 350 case CLOCK_MONOTONIC_PRECISE: 351 case CLOCK_UPTIME: 352 case CLOCK_UPTIME_PRECISE: 353 nanouptime(ats); 354 break; 355 case CLOCK_UPTIME_FAST: 356 case CLOCK_MONOTONIC_FAST: 357 getnanouptime(ats); 358 break; 359 case CLOCK_SECOND: 360 ats->tv_sec = time_second; 361 ats->tv_nsec = 0; 362 break; 363 case CLOCK_THREAD_CPUTIME_ID: 364 kern_thread_cputime(NULL, ats); 365 break; 366 case CLOCK_PROCESS_CPUTIME_ID: 367 PROC_LOCK(p); 368 kern_process_cputime(p, ats); 369 PROC_UNLOCK(p); 370 break; 371 default: 372 if ((int)clock_id >= 0) 373 return (EINVAL); 374 return (get_cputime(td, clock_id, ats)); 375 } 376 return (0); 377 } 378 379 #ifndef _SYS_SYSPROTO_H_ 380 struct clock_settime_args { 381 clockid_t clock_id; 382 const struct timespec *tp; 383 }; 384 #endif 385 /* ARGSUSED */ 386 int 387 sys_clock_settime(struct thread *td, struct clock_settime_args *uap) 388 { 389 struct timespec ats; 390 int error; 391 392 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) 393 return (error); 394 return (kern_clock_settime(td, uap->clock_id, &ats)); 395 } 396 397 static int allow_insane_settime = 0; 398 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN, 399 &allow_insane_settime, 0, 400 "do not perform possibly restrictive checks on settime(2) args"); 401 402 int 403 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats) 404 { 405 struct timeval atv; 406 int error; 407 408 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) 409 return (error); 410 if (clock_id != CLOCK_REALTIME) 411 return (EINVAL); 412 if (!timespecvalid_interval(ats)) 413 return (EINVAL); 414 if (!allow_insane_settime && 415 (ats->tv_sec > 8000ULL * 365 * 24 * 60 * 60 || 416 ats->tv_sec < utc_offset())) 417 return (EINVAL); 418 /* XXX Don't convert nsec->usec and back */ 419 TIMESPEC_TO_TIMEVAL(&atv, ats); 420 error = settime(td, &atv); 421 return (error); 422 } 423 424 #ifndef _SYS_SYSPROTO_H_ 425 struct clock_getres_args { 426 clockid_t clock_id; 427 struct timespec *tp; 428 }; 429 #endif 430 int 431 sys_clock_getres(struct thread *td, struct clock_getres_args *uap) 432 { 433 struct timespec ts; 434 int error; 435 436 if (uap->tp == NULL) 437 return (0); 438 439 error = kern_clock_getres(td, uap->clock_id, &ts); 440 if (error == 0) 441 error = copyout(&ts, uap->tp, sizeof(ts)); 442 return (error); 443 } 444 445 int 446 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts) 447 { 448 449 ts->tv_sec = 0; 450 switch (clock_id) { 451 case CLOCK_REALTIME: 452 case CLOCK_REALTIME_FAST: 453 case CLOCK_REALTIME_PRECISE: 454 case CLOCK_MONOTONIC: 455 case CLOCK_MONOTONIC_FAST: 456 case CLOCK_MONOTONIC_PRECISE: 457 case CLOCK_UPTIME: 458 case CLOCK_UPTIME_FAST: 459 case CLOCK_UPTIME_PRECISE: 460 /* 461 * Round up the result of the division cheaply by adding 1. 462 * Rounding up is especially important if rounding down 463 * would give 0. Perfect rounding is unimportant. 464 */ 465 ts->tv_nsec = NS_PER_SEC / tc_getfrequency() + 1; 466 break; 467 case CLOCK_VIRTUAL: 468 case CLOCK_PROF: 469 /* Accurately round up here because we can do so cheaply. */ 470 ts->tv_nsec = howmany(NS_PER_SEC, hz); 471 break; 472 case CLOCK_SECOND: 473 ts->tv_sec = 1; 474 ts->tv_nsec = 0; 475 break; 476 case CLOCK_THREAD_CPUTIME_ID: 477 case CLOCK_PROCESS_CPUTIME_ID: 478 cputime: 479 ts->tv_nsec = 1000000000 / cpu_tickrate() + 1; 480 break; 481 default: 482 if ((int)clock_id < 0) 483 goto cputime; 484 return (EINVAL); 485 } 486 return (0); 487 } 488 489 int 490 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt) 491 { 492 493 return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt, 494 rmt)); 495 } 496 497 static uint8_t nanowait[MAXCPU]; 498 499 int 500 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, 501 const struct timespec *rqt, struct timespec *rmt) 502 { 503 struct timespec ts, now; 504 sbintime_t sbt, sbtt, prec, tmp; 505 time_t over; 506 int error; 507 bool is_abs_real; 508 509 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= NS_PER_SEC) 510 return (EINVAL); 511 if ((flags & ~TIMER_ABSTIME) != 0) 512 return (EINVAL); 513 switch (clock_id) { 514 case CLOCK_REALTIME: 515 case CLOCK_REALTIME_PRECISE: 516 case CLOCK_REALTIME_FAST: 517 case CLOCK_SECOND: 518 is_abs_real = (flags & TIMER_ABSTIME) != 0; 519 break; 520 case CLOCK_MONOTONIC: 521 case CLOCK_MONOTONIC_PRECISE: 522 case CLOCK_MONOTONIC_FAST: 523 case CLOCK_UPTIME: 524 case CLOCK_UPTIME_PRECISE: 525 case CLOCK_UPTIME_FAST: 526 is_abs_real = false; 527 break; 528 case CLOCK_VIRTUAL: 529 case CLOCK_PROF: 530 case CLOCK_PROCESS_CPUTIME_ID: 531 return (ENOTSUP); 532 case CLOCK_THREAD_CPUTIME_ID: 533 default: 534 return (EINVAL); 535 } 536 do { 537 ts = *rqt; 538 if ((flags & TIMER_ABSTIME) != 0) { 539 if (is_abs_real) 540 td->td_rtcgen = 541 atomic_load_acq_int(&rtc_generation); 542 error = kern_clock_gettime(td, clock_id, &now); 543 KASSERT(error == 0, ("kern_clock_gettime: %d", error)); 544 timespecsub(&ts, &now, &ts); 545 } 546 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) { 547 error = EWOULDBLOCK; 548 break; 549 } 550 if (ts.tv_sec > INT32_MAX / 2) { 551 over = ts.tv_sec - INT32_MAX / 2; 552 ts.tv_sec -= over; 553 } else 554 over = 0; 555 tmp = tstosbt(ts); 556 prec = tmp; 557 prec >>= tc_precexp; 558 if (TIMESEL(&sbt, tmp)) 559 sbt += tc_tick_sbt; 560 sbt += tmp; 561 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp", 562 sbt, prec, C_ABSOLUTE); 563 } while (error == 0 && is_abs_real && td->td_rtcgen == 0); 564 td->td_rtcgen = 0; 565 if (error != EWOULDBLOCK) { 566 if (TIMESEL(&sbtt, tmp)) 567 sbtt += tc_tick_sbt; 568 if (sbtt >= sbt) 569 return (0); 570 if (error == ERESTART) 571 error = EINTR; 572 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) { 573 ts = sbttots(sbt - sbtt); 574 ts.tv_sec += over; 575 if (ts.tv_sec < 0) 576 timespecclear(&ts); 577 *rmt = ts; 578 } 579 return (error); 580 } 581 return (0); 582 } 583 584 #ifndef _SYS_SYSPROTO_H_ 585 struct nanosleep_args { 586 struct timespec *rqtp; 587 struct timespec *rmtp; 588 }; 589 #endif 590 /* ARGSUSED */ 591 int 592 sys_nanosleep(struct thread *td, struct nanosleep_args *uap) 593 { 594 595 return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, 596 uap->rqtp, uap->rmtp)); 597 } 598 599 #ifndef _SYS_SYSPROTO_H_ 600 struct clock_nanosleep_args { 601 clockid_t clock_id; 602 int flags; 603 struct timespec *rqtp; 604 struct timespec *rmtp; 605 }; 606 #endif 607 /* ARGSUSED */ 608 int 609 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap) 610 { 611 int error; 612 613 error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp, 614 uap->rmtp); 615 return (kern_posix_error(td, error)); 616 } 617 618 static int 619 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, 620 const struct timespec *ua_rqtp, struct timespec *ua_rmtp) 621 { 622 struct timespec rmt, rqt; 623 int error, error2; 624 625 error = copyin(ua_rqtp, &rqt, sizeof(rqt)); 626 if (error) 627 return (error); 628 error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt); 629 if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) { 630 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt)); 631 if (error2 != 0) 632 error = error2; 633 } 634 return (error); 635 } 636 637 #ifndef _SYS_SYSPROTO_H_ 638 struct gettimeofday_args { 639 struct timeval *tp; 640 struct timezone *tzp; 641 }; 642 #endif 643 /* ARGSUSED */ 644 int 645 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap) 646 { 647 struct timeval atv; 648 struct timezone rtz; 649 int error = 0; 650 651 if (uap->tp) { 652 microtime(&atv); 653 error = copyout(&atv, uap->tp, sizeof (atv)); 654 } 655 if (error == 0 && uap->tzp != NULL) { 656 rtz.tz_minuteswest = 0; 657 rtz.tz_dsttime = 0; 658 error = copyout(&rtz, uap->tzp, sizeof (rtz)); 659 } 660 return (error); 661 } 662 663 #ifndef _SYS_SYSPROTO_H_ 664 struct settimeofday_args { 665 struct timeval *tv; 666 struct timezone *tzp; 667 }; 668 #endif 669 /* ARGSUSED */ 670 int 671 sys_settimeofday(struct thread *td, struct settimeofday_args *uap) 672 { 673 struct timeval atv, *tvp; 674 struct timezone atz, *tzp; 675 int error; 676 677 if (uap->tv) { 678 error = copyin(uap->tv, &atv, sizeof(atv)); 679 if (error) 680 return (error); 681 tvp = &atv; 682 } else 683 tvp = NULL; 684 if (uap->tzp) { 685 error = copyin(uap->tzp, &atz, sizeof(atz)); 686 if (error) 687 return (error); 688 tzp = &atz; 689 } else 690 tzp = NULL; 691 return (kern_settimeofday(td, tvp, tzp)); 692 } 693 694 int 695 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp) 696 { 697 int error; 698 699 error = priv_check(td, PRIV_SETTIMEOFDAY); 700 if (error) 701 return (error); 702 /* Verify all parameters before changing time. */ 703 if (tv) { 704 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 || 705 tv->tv_sec < 0) 706 return (EINVAL); 707 error = settime(td, tv); 708 } 709 return (error); 710 } 711 712 /* 713 * Get value of an interval timer. The process virtual and profiling virtual 714 * time timers are kept in the p_stats area, since they can be swapped out. 715 * These are kept internally in the way they are specified externally: in 716 * time until they expire. 717 * 718 * The real time interval timer is kept in the process table slot for the 719 * process, and its value (it_value) is kept as an absolute time rather than 720 * as a delta, so that it is easy to keep periodic real-time signals from 721 * drifting. 722 * 723 * Virtual time timers are processed in the hardclock() routine of 724 * kern_clock.c. The real time timer is processed by a timeout routine, 725 * called from the softclock() routine. Since a callout may be delayed in 726 * real time due to interrupt processing in the system, it is possible for 727 * the real time timeout routine (realitexpire, given below), to be delayed 728 * in real time past when it is supposed to occur. It does not suffice, 729 * therefore, to reload the real timer .it_value from the real time timers 730 * .it_interval. Rather, we compute the next time in absolute time the timer 731 * should go off. 732 */ 733 #ifndef _SYS_SYSPROTO_H_ 734 struct getitimer_args { 735 u_int which; 736 struct itimerval *itv; 737 }; 738 #endif 739 int 740 sys_getitimer(struct thread *td, struct getitimer_args *uap) 741 { 742 struct itimerval aitv; 743 int error; 744 745 error = kern_getitimer(td, uap->which, &aitv); 746 if (error != 0) 747 return (error); 748 return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); 749 } 750 751 int 752 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv) 753 { 754 struct proc *p = td->td_proc; 755 struct timeval ctv; 756 757 if (which > ITIMER_PROF) 758 return (EINVAL); 759 760 if (which == ITIMER_REAL) { 761 /* 762 * Convert from absolute to relative time in .it_value 763 * part of real time timer. If time for real time timer 764 * has passed return 0, else return difference between 765 * current time and time for the timer to go off. 766 */ 767 PROC_LOCK(p); 768 *aitv = p->p_realtimer; 769 PROC_UNLOCK(p); 770 if (timevalisset(&aitv->it_value)) { 771 microuptime(&ctv); 772 if (timevalcmp(&aitv->it_value, &ctv, <)) 773 timevalclear(&aitv->it_value); 774 else 775 timevalsub(&aitv->it_value, &ctv); 776 } 777 } else { 778 PROC_ITIMLOCK(p); 779 *aitv = p->p_stats->p_timer[which]; 780 PROC_ITIMUNLOCK(p); 781 } 782 #ifdef KTRACE 783 if (KTRPOINT(td, KTR_STRUCT)) 784 ktritimerval(aitv); 785 #endif 786 return (0); 787 } 788 789 #ifndef _SYS_SYSPROTO_H_ 790 struct setitimer_args { 791 u_int which; 792 struct itimerval *itv, *oitv; 793 }; 794 #endif 795 int 796 sys_setitimer(struct thread *td, struct setitimer_args *uap) 797 { 798 struct itimerval aitv, oitv; 799 int error; 800 801 if (uap->itv == NULL) { 802 uap->itv = uap->oitv; 803 return (sys_getitimer(td, (struct getitimer_args *)uap)); 804 } 805 806 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval)))) 807 return (error); 808 error = kern_setitimer(td, uap->which, &aitv, &oitv); 809 if (error != 0 || uap->oitv == NULL) 810 return (error); 811 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval))); 812 } 813 814 int 815 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv, 816 struct itimerval *oitv) 817 { 818 struct proc *p = td->td_proc; 819 struct timeval ctv; 820 sbintime_t sbt, pr; 821 822 if (aitv == NULL) 823 return (kern_getitimer(td, which, oitv)); 824 825 if (which > ITIMER_PROF) 826 return (EINVAL); 827 #ifdef KTRACE 828 if (KTRPOINT(td, KTR_STRUCT)) 829 ktritimerval(aitv); 830 #endif 831 if (itimerfix(&aitv->it_value) || 832 aitv->it_value.tv_sec > INT32_MAX / 2) 833 return (EINVAL); 834 if (!timevalisset(&aitv->it_value)) 835 timevalclear(&aitv->it_interval); 836 else if (itimerfix(&aitv->it_interval) || 837 aitv->it_interval.tv_sec > INT32_MAX / 2) 838 return (EINVAL); 839 840 if (which == ITIMER_REAL) { 841 PROC_LOCK(p); 842 if (timevalisset(&p->p_realtimer.it_value)) 843 callout_stop(&p->p_itcallout); 844 microuptime(&ctv); 845 if (timevalisset(&aitv->it_value)) { 846 pr = tvtosbt(aitv->it_value) >> tc_precexp; 847 timevaladd(&aitv->it_value, &ctv); 848 sbt = tvtosbt(aitv->it_value); 849 callout_reset_sbt(&p->p_itcallout, sbt, pr, 850 realitexpire, p, C_ABSOLUTE); 851 } 852 *oitv = p->p_realtimer; 853 p->p_realtimer = *aitv; 854 PROC_UNLOCK(p); 855 if (timevalisset(&oitv->it_value)) { 856 if (timevalcmp(&oitv->it_value, &ctv, <)) 857 timevalclear(&oitv->it_value); 858 else 859 timevalsub(&oitv->it_value, &ctv); 860 } 861 } else { 862 if (aitv->it_interval.tv_sec == 0 && 863 aitv->it_interval.tv_usec != 0 && 864 aitv->it_interval.tv_usec < tick) 865 aitv->it_interval.tv_usec = tick; 866 if (aitv->it_value.tv_sec == 0 && 867 aitv->it_value.tv_usec != 0 && 868 aitv->it_value.tv_usec < tick) 869 aitv->it_value.tv_usec = tick; 870 PROC_ITIMLOCK(p); 871 *oitv = p->p_stats->p_timer[which]; 872 p->p_stats->p_timer[which] = *aitv; 873 PROC_ITIMUNLOCK(p); 874 } 875 #ifdef KTRACE 876 if (KTRPOINT(td, KTR_STRUCT)) 877 ktritimerval(oitv); 878 #endif 879 return (0); 880 } 881 882 static void 883 realitexpire_reset_callout(struct proc *p, sbintime_t *isbtp) 884 { 885 sbintime_t prec; 886 887 if ((p->p_flag & P_WEXIT) != 0) 888 return; 889 prec = isbtp == NULL ? tvtosbt(p->p_realtimer.it_interval) : *isbtp; 890 callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value), 891 prec >> tc_precexp, realitexpire, p, C_ABSOLUTE); 892 } 893 894 void 895 itimer_proc_continue(struct proc *p) 896 { 897 struct timeval ctv; 898 struct itimer *it; 899 int id; 900 901 PROC_LOCK_ASSERT(p, MA_OWNED); 902 903 if ((p->p_flag2 & P2_ITSTOPPED) != 0) { 904 p->p_flag2 &= ~P2_ITSTOPPED; 905 microuptime(&ctv); 906 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >=)) 907 realitexpire(p); 908 else 909 realitexpire_reset_callout(p, NULL); 910 } 911 912 if (p->p_itimers != NULL) { 913 for (id = 3; id < TIMER_MAX; id++) { 914 it = p->p_itimers->its_timers[id]; 915 if (it == NULL) 916 continue; 917 if ((it->it_flags & ITF_PSTOPPED) != 0) { 918 ITIMER_LOCK(it); 919 if ((it->it_flags & ITF_PSTOPPED) != 0) { 920 it->it_flags &= ~ITF_PSTOPPED; 921 if ((it->it_flags & ITF_DELETING) == 0) 922 realtimer_expire_l(it, true); 923 } 924 ITIMER_UNLOCK(it); 925 } 926 } 927 } 928 } 929 930 /* 931 * Real interval timer expired: 932 * send process whose timer expired an alarm signal. 933 * If time is not set up to reload, then just return. 934 * Else compute next time timer should go off which is > current time. 935 * This is where delay in processing this timeout causes multiple 936 * SIGALRM calls to be compressed into one. 937 * tvtohz() always adds 1 to allow for the time until the next clock 938 * interrupt being strictly less than 1 clock tick, but we don't want 939 * that here since we want to appear to be in sync with the clock 940 * interrupt even when we're delayed. 941 */ 942 static void 943 realitexpire(void *arg) 944 { 945 struct proc *p; 946 struct timeval ctv; 947 sbintime_t isbt; 948 949 p = (struct proc *)arg; 950 kern_psignal(p, SIGALRM); 951 if (!timevalisset(&p->p_realtimer.it_interval)) { 952 timevalclear(&p->p_realtimer.it_value); 953 return; 954 } 955 956 isbt = tvtosbt(p->p_realtimer.it_interval); 957 if (isbt >= sbt_timethreshold) 958 getmicrouptime(&ctv); 959 else 960 microuptime(&ctv); 961 do { 962 timevaladd(&p->p_realtimer.it_value, 963 &p->p_realtimer.it_interval); 964 } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=)); 965 966 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 967 p->p_flag2 |= P2_ITSTOPPED; 968 return; 969 } 970 971 p->p_flag2 &= ~P2_ITSTOPPED; 972 realitexpire_reset_callout(p, &isbt); 973 } 974 975 /* 976 * Check that a proposed value to load into the .it_value or 977 * .it_interval part of an interval timer is acceptable, and 978 * fix it to have at least minimal value (i.e. if it is less 979 * than the resolution of the clock, round it up.) 980 */ 981 int 982 itimerfix(struct timeval *tv) 983 { 984 985 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 986 return (EINVAL); 987 if (tv->tv_sec == 0 && tv->tv_usec != 0 && 988 tv->tv_usec < (u_int)tick / 16) 989 tv->tv_usec = (u_int)tick / 16; 990 return (0); 991 } 992 993 /* 994 * Decrement an interval timer by a specified number 995 * of microseconds, which must be less than a second, 996 * i.e. < 1000000. If the timer expires, then reload 997 * it. In this case, carry over (usec - old value) to 998 * reduce the value reloaded into the timer so that 999 * the timer does not drift. This routine assumes 1000 * that it is called in a context where the timers 1001 * on which it is operating cannot change in value. 1002 */ 1003 int 1004 itimerdecr(struct itimerval *itp, int usec) 1005 { 1006 1007 if (itp->it_value.tv_usec < usec) { 1008 if (itp->it_value.tv_sec == 0) { 1009 /* expired, and already in next interval */ 1010 usec -= itp->it_value.tv_usec; 1011 goto expire; 1012 } 1013 itp->it_value.tv_usec += 1000000; 1014 itp->it_value.tv_sec--; 1015 } 1016 itp->it_value.tv_usec -= usec; 1017 usec = 0; 1018 if (timevalisset(&itp->it_value)) 1019 return (1); 1020 /* expired, exactly at end of interval */ 1021 expire: 1022 if (timevalisset(&itp->it_interval)) { 1023 itp->it_value = itp->it_interval; 1024 itp->it_value.tv_usec -= usec; 1025 if (itp->it_value.tv_usec < 0) { 1026 itp->it_value.tv_usec += 1000000; 1027 itp->it_value.tv_sec--; 1028 } 1029 } else 1030 itp->it_value.tv_usec = 0; /* sec is already 0 */ 1031 return (0); 1032 } 1033 1034 /* 1035 * Add and subtract routines for timevals. 1036 * N.B.: subtract routine doesn't deal with 1037 * results which are before the beginning, 1038 * it just gets very confused in this case. 1039 * Caveat emptor. 1040 */ 1041 void 1042 timevaladd(struct timeval *t1, const struct timeval *t2) 1043 { 1044 1045 t1->tv_sec += t2->tv_sec; 1046 t1->tv_usec += t2->tv_usec; 1047 timevalfix(t1); 1048 } 1049 1050 void 1051 timevalsub(struct timeval *t1, const struct timeval *t2) 1052 { 1053 1054 t1->tv_sec -= t2->tv_sec; 1055 t1->tv_usec -= t2->tv_usec; 1056 timevalfix(t1); 1057 } 1058 1059 static void 1060 timevalfix(struct timeval *t1) 1061 { 1062 1063 if (t1->tv_usec < 0) { 1064 t1->tv_sec--; 1065 t1->tv_usec += 1000000; 1066 } 1067 if (t1->tv_usec >= 1000000) { 1068 t1->tv_sec++; 1069 t1->tv_usec -= 1000000; 1070 } 1071 } 1072 1073 /* 1074 * ratecheck(): simple time-based rate-limit checking. 1075 */ 1076 int 1077 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1078 { 1079 struct timeval tv, delta; 1080 int rv = 0; 1081 1082 getmicrouptime(&tv); /* NB: 10ms precision */ 1083 delta = tv; 1084 timevalsub(&delta, lasttime); 1085 1086 /* 1087 * check for 0,0 is so that the message will be seen at least once, 1088 * even if interval is huge. 1089 */ 1090 if (timevalcmp(&delta, mininterval, >=) || 1091 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1092 *lasttime = tv; 1093 rv = 1; 1094 } 1095 1096 return (rv); 1097 } 1098 1099 /* 1100 * eventratecheck(): events per second limitation. 1101 * 1102 * Return 0 if the limit is to be enforced (e.g. the caller 1103 * should ignore the event because of the rate limitation). 1104 * 1105 * maxeps of 0 always causes zero to be returned. maxeps of -1 1106 * always causes 1 to be returned; this effectively defeats rate 1107 * limiting. 1108 * 1109 * Note that we maintain the struct timeval for compatibility 1110 * with other bsd systems. We reuse the storage and just monitor 1111 * clock ticks for minimal overhead. 1112 */ 1113 int 1114 eventratecheck(struct timeval *lasttime, int *cureps, int maxeps) 1115 { 1116 int now; 1117 1118 /* 1119 * Reset the last time and counter if this is the first call 1120 * or more than a second has passed since the last update of 1121 * lasttime. 1122 */ 1123 now = ticks; 1124 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { 1125 lasttime->tv_sec = now; 1126 *cureps = 1; 1127 return (maxeps != 0); 1128 } else { 1129 (*cureps)++; /* NB: ignore potential overflow */ 1130 return (maxeps < 0 || *cureps <= maxeps); 1131 } 1132 } 1133 1134 static void 1135 itimer_start(void) 1136 { 1137 static const struct kclock rt_clock = { 1138 .timer_create = realtimer_create, 1139 .timer_delete = realtimer_delete, 1140 .timer_settime = realtimer_settime, 1141 .timer_gettime = realtimer_gettime, 1142 }; 1143 1144 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer), 1145 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0); 1146 register_posix_clock(CLOCK_REALTIME, &rt_clock); 1147 register_posix_clock(CLOCK_MONOTONIC, &rt_clock); 1148 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L); 1149 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX); 1150 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX); 1151 } 1152 1153 static int 1154 register_posix_clock(int clockid, const struct kclock *clk) 1155 { 1156 if ((unsigned)clockid >= MAX_CLOCKS) { 1157 printf("%s: invalid clockid\n", __func__); 1158 return (0); 1159 } 1160 posix_clocks[clockid] = *clk; 1161 return (1); 1162 } 1163 1164 static int 1165 itimer_init(void *mem, int size, int flags) 1166 { 1167 struct itimer *it; 1168 1169 it = (struct itimer *)mem; 1170 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF); 1171 return (0); 1172 } 1173 1174 static void 1175 itimer_fini(void *mem, int size) 1176 { 1177 struct itimer *it; 1178 1179 it = (struct itimer *)mem; 1180 mtx_destroy(&it->it_mtx); 1181 } 1182 1183 static void 1184 itimer_enter(struct itimer *it) 1185 { 1186 1187 mtx_assert(&it->it_mtx, MA_OWNED); 1188 it->it_usecount++; 1189 } 1190 1191 static void 1192 itimer_leave(struct itimer *it) 1193 { 1194 1195 mtx_assert(&it->it_mtx, MA_OWNED); 1196 KASSERT(it->it_usecount > 0, ("invalid it_usecount")); 1197 1198 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0) 1199 wakeup(it); 1200 } 1201 1202 #ifndef _SYS_SYSPROTO_H_ 1203 struct ktimer_create_args { 1204 clockid_t clock_id; 1205 struct sigevent * evp; 1206 int * timerid; 1207 }; 1208 #endif 1209 int 1210 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap) 1211 { 1212 struct sigevent *evp, ev; 1213 int id; 1214 int error; 1215 1216 if (uap->evp == NULL) { 1217 evp = NULL; 1218 } else { 1219 error = copyin(uap->evp, &ev, sizeof(ev)); 1220 if (error != 0) 1221 return (error); 1222 evp = &ev; 1223 } 1224 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1); 1225 if (error == 0) { 1226 error = copyout(&id, uap->timerid, sizeof(int)); 1227 if (error != 0) 1228 kern_ktimer_delete(td, id); 1229 } 1230 return (error); 1231 } 1232 1233 int 1234 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp, 1235 int *timerid, int preset_id) 1236 { 1237 struct proc *p = td->td_proc; 1238 struct itimer *it; 1239 int id; 1240 int error; 1241 1242 if (clock_id < 0 || clock_id >= MAX_CLOCKS) 1243 return (EINVAL); 1244 1245 if (posix_clocks[clock_id].timer_create == NULL) 1246 return (EINVAL); 1247 1248 if (evp != NULL) { 1249 if (evp->sigev_notify != SIGEV_NONE && 1250 evp->sigev_notify != SIGEV_SIGNAL && 1251 evp->sigev_notify != SIGEV_THREAD_ID) 1252 return (EINVAL); 1253 if ((evp->sigev_notify == SIGEV_SIGNAL || 1254 evp->sigev_notify == SIGEV_THREAD_ID) && 1255 !_SIG_VALID(evp->sigev_signo)) 1256 return (EINVAL); 1257 } 1258 1259 if (p->p_itimers == NULL) 1260 itimers_alloc(p); 1261 1262 it = uma_zalloc(itimer_zone, M_WAITOK); 1263 it->it_flags = 0; 1264 it->it_usecount = 0; 1265 timespecclear(&it->it_time.it_value); 1266 timespecclear(&it->it_time.it_interval); 1267 it->it_overrun = 0; 1268 it->it_overrun_last = 0; 1269 it->it_clockid = clock_id; 1270 it->it_proc = p; 1271 ksiginfo_init(&it->it_ksi); 1272 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT; 1273 error = CLOCK_CALL(clock_id, timer_create, (it)); 1274 if (error != 0) 1275 goto out; 1276 1277 PROC_LOCK(p); 1278 if (preset_id != -1) { 1279 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id")); 1280 id = preset_id; 1281 if (p->p_itimers->its_timers[id] != NULL) { 1282 PROC_UNLOCK(p); 1283 error = 0; 1284 goto out; 1285 } 1286 } else { 1287 /* 1288 * Find a free timer slot, skipping those reserved 1289 * for setitimer(). 1290 */ 1291 for (id = 3; id < TIMER_MAX; id++) 1292 if (p->p_itimers->its_timers[id] == NULL) 1293 break; 1294 if (id == TIMER_MAX) { 1295 PROC_UNLOCK(p); 1296 error = EAGAIN; 1297 goto out; 1298 } 1299 } 1300 p->p_itimers->its_timers[id] = it; 1301 if (evp != NULL) 1302 it->it_sigev = *evp; 1303 else { 1304 it->it_sigev.sigev_notify = SIGEV_SIGNAL; 1305 switch (clock_id) { 1306 default: 1307 case CLOCK_REALTIME: 1308 it->it_sigev.sigev_signo = SIGALRM; 1309 break; 1310 case CLOCK_VIRTUAL: 1311 it->it_sigev.sigev_signo = SIGVTALRM; 1312 break; 1313 case CLOCK_PROF: 1314 it->it_sigev.sigev_signo = SIGPROF; 1315 break; 1316 } 1317 it->it_sigev.sigev_value.sival_int = id; 1318 } 1319 1320 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1321 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1322 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo; 1323 it->it_ksi.ksi_code = SI_TIMER; 1324 it->it_ksi.ksi_value = it->it_sigev.sigev_value; 1325 it->it_ksi.ksi_timerid = id; 1326 } 1327 PROC_UNLOCK(p); 1328 *timerid = id; 1329 return (0); 1330 1331 out: 1332 ITIMER_LOCK(it); 1333 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1334 ITIMER_UNLOCK(it); 1335 uma_zfree(itimer_zone, it); 1336 return (error); 1337 } 1338 1339 #ifndef _SYS_SYSPROTO_H_ 1340 struct ktimer_delete_args { 1341 int timerid; 1342 }; 1343 #endif 1344 int 1345 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap) 1346 { 1347 1348 return (kern_ktimer_delete(td, uap->timerid)); 1349 } 1350 1351 static struct itimer * 1352 itimer_find(struct proc *p, int timerid) 1353 { 1354 struct itimer *it; 1355 1356 PROC_LOCK_ASSERT(p, MA_OWNED); 1357 if ((p->p_itimers == NULL) || 1358 (timerid < 0) || (timerid >= TIMER_MAX) || 1359 (it = p->p_itimers->its_timers[timerid]) == NULL) { 1360 return (NULL); 1361 } 1362 ITIMER_LOCK(it); 1363 if ((it->it_flags & ITF_DELETING) != 0) { 1364 ITIMER_UNLOCK(it); 1365 it = NULL; 1366 } 1367 return (it); 1368 } 1369 1370 int 1371 kern_ktimer_delete(struct thread *td, int timerid) 1372 { 1373 struct proc *p = td->td_proc; 1374 struct itimer *it; 1375 1376 PROC_LOCK(p); 1377 it = itimer_find(p, timerid); 1378 if (it == NULL) { 1379 PROC_UNLOCK(p); 1380 return (EINVAL); 1381 } 1382 PROC_UNLOCK(p); 1383 1384 it->it_flags |= ITF_DELETING; 1385 while (it->it_usecount > 0) { 1386 it->it_flags |= ITF_WANTED; 1387 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0); 1388 } 1389 it->it_flags &= ~ITF_WANTED; 1390 CLOCK_CALL(it->it_clockid, timer_delete, (it)); 1391 ITIMER_UNLOCK(it); 1392 1393 PROC_LOCK(p); 1394 if (KSI_ONQ(&it->it_ksi)) 1395 sigqueue_take(&it->it_ksi); 1396 p->p_itimers->its_timers[timerid] = NULL; 1397 PROC_UNLOCK(p); 1398 uma_zfree(itimer_zone, it); 1399 return (0); 1400 } 1401 1402 #ifndef _SYS_SYSPROTO_H_ 1403 struct ktimer_settime_args { 1404 int timerid; 1405 int flags; 1406 const struct itimerspec * value; 1407 struct itimerspec * ovalue; 1408 }; 1409 #endif 1410 int 1411 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap) 1412 { 1413 struct itimerspec val, oval, *ovalp; 1414 int error; 1415 1416 error = copyin(uap->value, &val, sizeof(val)); 1417 if (error != 0) 1418 return (error); 1419 ovalp = uap->ovalue != NULL ? &oval : NULL; 1420 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp); 1421 if (error == 0 && uap->ovalue != NULL) 1422 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp)); 1423 return (error); 1424 } 1425 1426 int 1427 kern_ktimer_settime(struct thread *td, int timer_id, int flags, 1428 struct itimerspec *val, struct itimerspec *oval) 1429 { 1430 struct proc *p; 1431 struct itimer *it; 1432 int error; 1433 1434 p = td->td_proc; 1435 PROC_LOCK(p); 1436 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1437 PROC_UNLOCK(p); 1438 error = EINVAL; 1439 } else { 1440 PROC_UNLOCK(p); 1441 itimer_enter(it); 1442 error = CLOCK_CALL(it->it_clockid, timer_settime, (it, 1443 flags, val, oval)); 1444 itimer_leave(it); 1445 ITIMER_UNLOCK(it); 1446 } 1447 return (error); 1448 } 1449 1450 #ifndef _SYS_SYSPROTO_H_ 1451 struct ktimer_gettime_args { 1452 int timerid; 1453 struct itimerspec * value; 1454 }; 1455 #endif 1456 int 1457 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap) 1458 { 1459 struct itimerspec val; 1460 int error; 1461 1462 error = kern_ktimer_gettime(td, uap->timerid, &val); 1463 if (error == 0) 1464 error = copyout(&val, uap->value, sizeof(val)); 1465 return (error); 1466 } 1467 1468 int 1469 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val) 1470 { 1471 struct proc *p; 1472 struct itimer *it; 1473 int error; 1474 1475 p = td->td_proc; 1476 PROC_LOCK(p); 1477 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { 1478 PROC_UNLOCK(p); 1479 error = EINVAL; 1480 } else { 1481 PROC_UNLOCK(p); 1482 itimer_enter(it); 1483 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val)); 1484 itimer_leave(it); 1485 ITIMER_UNLOCK(it); 1486 } 1487 return (error); 1488 } 1489 1490 #ifndef _SYS_SYSPROTO_H_ 1491 struct timer_getoverrun_args { 1492 int timerid; 1493 }; 1494 #endif 1495 int 1496 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap) 1497 { 1498 1499 return (kern_ktimer_getoverrun(td, uap->timerid)); 1500 } 1501 1502 int 1503 kern_ktimer_getoverrun(struct thread *td, int timer_id) 1504 { 1505 struct proc *p = td->td_proc; 1506 struct itimer *it; 1507 int error ; 1508 1509 PROC_LOCK(p); 1510 if (timer_id < 3 || 1511 (it = itimer_find(p, timer_id)) == NULL) { 1512 PROC_UNLOCK(p); 1513 error = EINVAL; 1514 } else { 1515 td->td_retval[0] = it->it_overrun_last; 1516 ITIMER_UNLOCK(it); 1517 PROC_UNLOCK(p); 1518 error = 0; 1519 } 1520 return (error); 1521 } 1522 1523 static int 1524 realtimer_create(struct itimer *it) 1525 { 1526 callout_init_mtx(&it->it_callout, &it->it_mtx, 0); 1527 return (0); 1528 } 1529 1530 static int 1531 realtimer_delete(struct itimer *it) 1532 { 1533 mtx_assert(&it->it_mtx, MA_OWNED); 1534 1535 /* 1536 * clear timer's value and interval to tell realtimer_expire 1537 * to not rearm the timer. 1538 */ 1539 timespecclear(&it->it_time.it_value); 1540 timespecclear(&it->it_time.it_interval); 1541 ITIMER_UNLOCK(it); 1542 callout_drain(&it->it_callout); 1543 ITIMER_LOCK(it); 1544 return (0); 1545 } 1546 1547 static int 1548 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue) 1549 { 1550 struct timespec cts; 1551 1552 mtx_assert(&it->it_mtx, MA_OWNED); 1553 1554 realtimer_clocktime(it->it_clockid, &cts); 1555 *ovalue = it->it_time; 1556 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) { 1557 timespecsub(&ovalue->it_value, &cts, &ovalue->it_value); 1558 if (ovalue->it_value.tv_sec < 0 || 1559 (ovalue->it_value.tv_sec == 0 && 1560 ovalue->it_value.tv_nsec == 0)) { 1561 ovalue->it_value.tv_sec = 0; 1562 ovalue->it_value.tv_nsec = 1; 1563 } 1564 } 1565 return (0); 1566 } 1567 1568 static int 1569 realtimer_settime(struct itimer *it, int flags, struct itimerspec *value, 1570 struct itimerspec *ovalue) 1571 { 1572 struct timespec cts, ts; 1573 struct timeval tv; 1574 struct itimerspec val; 1575 1576 mtx_assert(&it->it_mtx, MA_OWNED); 1577 1578 val = *value; 1579 if (itimespecfix(&val.it_value)) 1580 return (EINVAL); 1581 1582 if (timespecisset(&val.it_value)) { 1583 if (itimespecfix(&val.it_interval)) 1584 return (EINVAL); 1585 } else { 1586 timespecclear(&val.it_interval); 1587 } 1588 1589 if (ovalue != NULL) 1590 realtimer_gettime(it, ovalue); 1591 1592 it->it_time = val; 1593 if (timespecisset(&val.it_value)) { 1594 realtimer_clocktime(it->it_clockid, &cts); 1595 ts = val.it_value; 1596 if ((flags & TIMER_ABSTIME) == 0) { 1597 /* Convert to absolute time. */ 1598 timespecadd(&it->it_time.it_value, &cts, 1599 &it->it_time.it_value); 1600 } else { 1601 timespecsub(&ts, &cts, &ts); 1602 /* 1603 * We don't care if ts is negative, tztohz will 1604 * fix it. 1605 */ 1606 } 1607 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1608 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire, 1609 it); 1610 } else { 1611 callout_stop(&it->it_callout); 1612 } 1613 1614 return (0); 1615 } 1616 1617 static void 1618 realtimer_clocktime(clockid_t id, struct timespec *ts) 1619 { 1620 if (id == CLOCK_REALTIME) 1621 getnanotime(ts); 1622 else /* CLOCK_MONOTONIC */ 1623 getnanouptime(ts); 1624 } 1625 1626 int 1627 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi) 1628 { 1629 struct itimer *it; 1630 1631 PROC_LOCK_ASSERT(p, MA_OWNED); 1632 it = itimer_find(p, timerid); 1633 if (it != NULL) { 1634 ksi->ksi_overrun = it->it_overrun; 1635 it->it_overrun_last = it->it_overrun; 1636 it->it_overrun = 0; 1637 ITIMER_UNLOCK(it); 1638 return (0); 1639 } 1640 return (EINVAL); 1641 } 1642 1643 static int 1644 itimespecfix(struct timespec *ts) 1645 { 1646 1647 if (!timespecvalid_interval(ts)) 1648 return (EINVAL); 1649 if ((UINT64_MAX - ts->tv_nsec) / NS_PER_SEC < ts->tv_sec) 1650 return (EINVAL); 1651 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000) 1652 ts->tv_nsec = tick * 1000; 1653 return (0); 1654 } 1655 1656 #define timespectons(tsp) \ 1657 ((uint64_t)(tsp)->tv_sec * NS_PER_SEC + (tsp)->tv_nsec) 1658 #define timespecfromns(ns) (struct timespec){ \ 1659 .tv_sec = (ns) / NS_PER_SEC, \ 1660 .tv_nsec = (ns) % NS_PER_SEC \ 1661 } 1662 1663 static void 1664 realtimer_expire_l(struct itimer *it, bool proc_locked) 1665 { 1666 struct timespec cts, ts; 1667 struct timeval tv; 1668 struct proc *p; 1669 uint64_t interval, now, overruns, value; 1670 1671 realtimer_clocktime(it->it_clockid, &cts); 1672 /* Only fire if time is reached. */ 1673 if (timespeccmp(&cts, &it->it_time.it_value, >=)) { 1674 if (timespecisset(&it->it_time.it_interval)) { 1675 timespecadd(&it->it_time.it_value, 1676 &it->it_time.it_interval, 1677 &it->it_time.it_value); 1678 1679 interval = timespectons(&it->it_time.it_interval); 1680 value = timespectons(&it->it_time.it_value); 1681 now = timespectons(&cts); 1682 1683 if (now >= value) { 1684 /* 1685 * We missed at least one period. 1686 */ 1687 overruns = howmany(now - value + 1, interval); 1688 if (it->it_overrun + overruns >= 1689 it->it_overrun && 1690 it->it_overrun + overruns <= INT_MAX) { 1691 it->it_overrun += (int)overruns; 1692 } else { 1693 it->it_overrun = INT_MAX; 1694 it->it_ksi.ksi_errno = ERANGE; 1695 } 1696 value = 1697 now + interval - (now - value) % interval; 1698 it->it_time.it_value = timespecfromns(value); 1699 } 1700 } else { 1701 /* single shot timer ? */ 1702 timespecclear(&it->it_time.it_value); 1703 } 1704 1705 p = it->it_proc; 1706 if (timespecisset(&it->it_time.it_value)) { 1707 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 1708 it->it_flags |= ITF_PSTOPPED; 1709 } else { 1710 timespecsub(&it->it_time.it_value, &cts, &ts); 1711 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1712 callout_reset(&it->it_callout, tvtohz(&tv), 1713 realtimer_expire, it); 1714 } 1715 } 1716 1717 itimer_enter(it); 1718 ITIMER_UNLOCK(it); 1719 if (proc_locked) 1720 PROC_UNLOCK(p); 1721 itimer_fire(it); 1722 if (proc_locked) 1723 PROC_LOCK(p); 1724 ITIMER_LOCK(it); 1725 itimer_leave(it); 1726 } else if (timespecisset(&it->it_time.it_value)) { 1727 p = it->it_proc; 1728 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 1729 it->it_flags |= ITF_PSTOPPED; 1730 } else { 1731 ts = it->it_time.it_value; 1732 timespecsub(&ts, &cts, &ts); 1733 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1734 callout_reset(&it->it_callout, tvtohz(&tv), 1735 realtimer_expire, it); 1736 } 1737 } 1738 } 1739 1740 /* Timeout callback for realtime timer */ 1741 static void 1742 realtimer_expire(void *arg) 1743 { 1744 realtimer_expire_l(arg, false); 1745 } 1746 1747 static void 1748 itimer_fire(struct itimer *it) 1749 { 1750 struct proc *p = it->it_proc; 1751 struct thread *td; 1752 1753 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || 1754 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { 1755 if (sigev_findtd(p, &it->it_sigev, &td) != 0) { 1756 ITIMER_LOCK(it); 1757 timespecclear(&it->it_time.it_value); 1758 timespecclear(&it->it_time.it_interval); 1759 callout_stop(&it->it_callout); 1760 ITIMER_UNLOCK(it); 1761 return; 1762 } 1763 if (!KSI_ONQ(&it->it_ksi)) { 1764 it->it_ksi.ksi_errno = 0; 1765 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev); 1766 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi); 1767 } else { 1768 if (it->it_overrun < INT_MAX) 1769 it->it_overrun++; 1770 else 1771 it->it_ksi.ksi_errno = ERANGE; 1772 } 1773 PROC_UNLOCK(p); 1774 } 1775 } 1776 1777 static void 1778 itimers_alloc(struct proc *p) 1779 { 1780 struct itimers *its; 1781 1782 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO); 1783 PROC_LOCK(p); 1784 if (p->p_itimers == NULL) { 1785 p->p_itimers = its; 1786 PROC_UNLOCK(p); 1787 } 1788 else { 1789 PROC_UNLOCK(p); 1790 free(its, M_SUBPROC); 1791 } 1792 } 1793 1794 /* Clean up timers when some process events are being triggered. */ 1795 static void 1796 itimers_event_exit_exec(int start_idx, struct proc *p) 1797 { 1798 struct itimers *its; 1799 struct itimer *it; 1800 int i; 1801 1802 its = p->p_itimers; 1803 if (its == NULL) 1804 return; 1805 1806 for (i = start_idx; i < TIMER_MAX; ++i) { 1807 if ((it = its->its_timers[i]) != NULL) 1808 kern_ktimer_delete(curthread, i); 1809 } 1810 if (its->its_timers[0] == NULL && its->its_timers[1] == NULL && 1811 its->its_timers[2] == NULL) { 1812 /* Synchronize with itimer_proc_continue(). */ 1813 PROC_LOCK(p); 1814 p->p_itimers = NULL; 1815 PROC_UNLOCK(p); 1816 free(its, M_SUBPROC); 1817 } 1818 } 1819 1820 void 1821 itimers_exec(struct proc *p) 1822 { 1823 /* 1824 * According to susv3, XSI interval timers should be inherited 1825 * by new image. 1826 */ 1827 itimers_event_exit_exec(3, p); 1828 } 1829 1830 void 1831 itimers_exit(struct proc *p) 1832 { 1833 itimers_event_exit_exec(0, p); 1834 } 1835