1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * 2002-10-15 Posix Clocks & timers 4 * by George Anzinger george@mvista.com 5 * Copyright (C) 2002 2003 by MontaVista Software. 6 * 7 * 2004-06-01 Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug. 8 * Copyright (C) 2004 Boris Hu 9 * 10 * These are all the functions necessary to implement POSIX clocks & timers 11 */ 12 #include <linux/mm.h> 13 #include <linux/interrupt.h> 14 #include <linux/slab.h> 15 #include <linux/time.h> 16 #include <linux/mutex.h> 17 #include <linux/sched/task.h> 18 19 #include <linux/uaccess.h> 20 #include <linux/list.h> 21 #include <linux/init.h> 22 #include <linux/compiler.h> 23 #include <linux/hash.h> 24 #include <linux/posix-clock.h> 25 #include <linux/posix-timers.h> 26 #include <linux/syscalls.h> 27 #include <linux/wait.h> 28 #include <linux/workqueue.h> 29 #include <linux/export.h> 30 #include <linux/hashtable.h> 31 #include <linux/compat.h> 32 #include <linux/nospec.h> 33 #include <linux/time_namespace.h> 34 35 #include "timekeeping.h" 36 #include "posix-timers.h" 37 38 static struct kmem_cache *posix_timers_cache; 39 40 /* 41 * Timers are managed in a hash table for lockless lookup. The hash key is 42 * constructed from current::signal and the timer ID and the timer is 43 * matched against current::signal and the timer ID when walking the hash 44 * bucket list. 45 * 46 * This allows checkpoint/restore to reconstruct the exact timer IDs for 47 * a process. 48 */ 49 static DEFINE_HASHTABLE(posix_timers_hashtable, 9); 50 static DEFINE_SPINLOCK(hash_lock); 51 52 static const struct k_clock * const posix_clocks[]; 53 static const struct k_clock *clockid_to_kclock(const clockid_t id); 54 static const struct k_clock clock_realtime, clock_monotonic; 55 56 /* SIGEV_THREAD_ID cannot share a bit with the other SIGEV values. */ 57 #if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \ 58 ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD)) 59 #error "SIGEV_THREAD_ID must not share bit with other SIGEV values!" 60 #endif 61 62 static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags); 63 64 #define lock_timer(tid, flags) \ 65 ({ struct k_itimer *__timr; \ 66 __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags)); \ 67 __timr; \ 68 }) 69 70 static int hash(struct signal_struct *sig, unsigned int nr) 71 { 72 return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable)); 73 } 74 75 static struct k_itimer *__posix_timers_find(struct hlist_head *head, 76 struct signal_struct *sig, 77 timer_t id) 78 { 79 struct k_itimer *timer; 80 81 hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) { 82 /* timer->it_signal can be set concurrently */ 83 if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id)) 84 return timer; 85 } 86 return NULL; 87 } 88 89 static struct k_itimer *posix_timer_by_id(timer_t id) 90 { 91 struct signal_struct *sig = current->signal; 92 struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)]; 93 94 return __posix_timers_find(head, sig, id); 95 } 96 97 static int posix_timer_add(struct k_itimer *timer) 98 { 99 struct signal_struct *sig = current->signal; 100 struct hlist_head *head; 101 unsigned int cnt, id; 102 103 /* 104 * FIXME: Replace this by a per signal struct xarray once there is 105 * a plan to handle the resulting CRIU regression gracefully. 106 */ 107 for (cnt = 0; cnt <= INT_MAX; cnt++) { 108 spin_lock(&hash_lock); 109 id = sig->next_posix_timer_id; 110 111 /* Write the next ID back. Clamp it to the positive space */ 112 sig->next_posix_timer_id = (id + 1) & INT_MAX; 113 114 head = &posix_timers_hashtable[hash(sig, id)]; 115 if (!__posix_timers_find(head, sig, id)) { 116 hlist_add_head_rcu(&timer->t_hash, head); 117 spin_unlock(&hash_lock); 118 return id; 119 } 120 spin_unlock(&hash_lock); 121 } 122 /* POSIX return code when no timer ID could be allocated */ 123 return -EAGAIN; 124 } 125 126 static inline void unlock_timer(struct k_itimer *timr, unsigned long flags) 127 { 128 spin_unlock_irqrestore(&timr->it_lock, flags); 129 } 130 131 static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp) 132 { 133 ktime_get_real_ts64(tp); 134 return 0; 135 } 136 137 static ktime_t posix_get_realtime_ktime(clockid_t which_clock) 138 { 139 return ktime_get_real(); 140 } 141 142 static int posix_clock_realtime_set(const clockid_t which_clock, 143 const struct timespec64 *tp) 144 { 145 return do_sys_settimeofday64(tp, NULL); 146 } 147 148 static int posix_clock_realtime_adj(const clockid_t which_clock, 149 struct __kernel_timex *t) 150 { 151 return do_adjtimex(t); 152 } 153 154 static int posix_get_monotonic_timespec(clockid_t which_clock, struct timespec64 *tp) 155 { 156 ktime_get_ts64(tp); 157 timens_add_monotonic(tp); 158 return 0; 159 } 160 161 static ktime_t posix_get_monotonic_ktime(clockid_t which_clock) 162 { 163 return ktime_get(); 164 } 165 166 static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp) 167 { 168 ktime_get_raw_ts64(tp); 169 timens_add_monotonic(tp); 170 return 0; 171 } 172 173 static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp) 174 { 175 ktime_get_coarse_real_ts64(tp); 176 return 0; 177 } 178 179 static int posix_get_monotonic_coarse(clockid_t which_clock, 180 struct timespec64 *tp) 181 { 182 ktime_get_coarse_ts64(tp); 183 timens_add_monotonic(tp); 184 return 0; 185 } 186 187 static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp) 188 { 189 *tp = ktime_to_timespec64(KTIME_LOW_RES); 190 return 0; 191 } 192 193 static int posix_get_boottime_timespec(const clockid_t which_clock, struct timespec64 *tp) 194 { 195 ktime_get_boottime_ts64(tp); 196 timens_add_boottime(tp); 197 return 0; 198 } 199 200 static ktime_t posix_get_boottime_ktime(const clockid_t which_clock) 201 { 202 return ktime_get_boottime(); 203 } 204 205 static int posix_get_tai_timespec(clockid_t which_clock, struct timespec64 *tp) 206 { 207 ktime_get_clocktai_ts64(tp); 208 return 0; 209 } 210 211 static ktime_t posix_get_tai_ktime(clockid_t which_clock) 212 { 213 return ktime_get_clocktai(); 214 } 215 216 static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp) 217 { 218 tp->tv_sec = 0; 219 tp->tv_nsec = hrtimer_resolution; 220 return 0; 221 } 222 223 static __init int init_posix_timers(void) 224 { 225 posix_timers_cache = kmem_cache_create("posix_timers_cache", 226 sizeof(struct k_itimer), 0, 227 SLAB_PANIC | SLAB_ACCOUNT, NULL); 228 return 0; 229 } 230 __initcall(init_posix_timers); 231 232 /* 233 * The siginfo si_overrun field and the return value of timer_getoverrun(2) 234 * are of type int. Clamp the overrun value to INT_MAX 235 */ 236 static inline int timer_overrun_to_int(struct k_itimer *timr) 237 { 238 if (timr->it_overrun_last > (s64)INT_MAX) 239 return INT_MAX; 240 241 return (int)timr->it_overrun_last; 242 } 243 244 static void common_hrtimer_rearm(struct k_itimer *timr) 245 { 246 struct hrtimer *timer = &timr->it.real.timer; 247 248 timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(), 249 timr->it_interval); 250 hrtimer_restart(timer); 251 } 252 253 static bool __posixtimer_deliver_signal(struct kernel_siginfo *info, struct k_itimer *timr) 254 { 255 guard(spinlock)(&timr->it_lock); 256 257 /* 258 * Check if the timer is still alive or whether it got modified 259 * since the signal was queued. In either case, don't rearm and 260 * drop the signal. 261 */ 262 if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal)) 263 return false; 264 265 if (!timr->it_interval || WARN_ON_ONCE(timr->it_status != POSIX_TIMER_REQUEUE_PENDING)) 266 return true; 267 268 timr->kclock->timer_rearm(timr); 269 timr->it_status = POSIX_TIMER_ARMED; 270 timr->it_overrun_last = timr->it_overrun; 271 timr->it_overrun = -1LL; 272 ++timr->it_signal_seq; 273 info->si_overrun = timer_overrun_to_int(timr); 274 return true; 275 } 276 277 /* 278 * This function is called from the signal delivery code. It decides 279 * whether the signal should be dropped and rearms interval timers. The 280 * timer can be unconditionally accessed as there is a reference held on 281 * it. 282 */ 283 bool posixtimer_deliver_signal(struct kernel_siginfo *info, struct sigqueue *timer_sigq) 284 { 285 struct k_itimer *timr = container_of(timer_sigq, struct k_itimer, sigq); 286 bool ret; 287 288 /* 289 * Release siglock to ensure proper locking order versus 290 * timr::it_lock. Keep interrupts disabled. 291 */ 292 spin_unlock(¤t->sighand->siglock); 293 294 ret = __posixtimer_deliver_signal(info, timr); 295 296 /* Drop the reference which was acquired when the signal was queued */ 297 posixtimer_putref(timr); 298 299 spin_lock(¤t->sighand->siglock); 300 return ret; 301 } 302 303 void posix_timer_queue_signal(struct k_itimer *timr) 304 { 305 lockdep_assert_held(&timr->it_lock); 306 307 timr->it_status = timr->it_interval ? POSIX_TIMER_REQUEUE_PENDING : POSIX_TIMER_DISARMED; 308 posixtimer_send_sigqueue(timr); 309 } 310 311 /* 312 * This function gets called when a POSIX.1b interval timer expires from 313 * the HRTIMER interrupt (soft interrupt on RT kernels). 314 * 315 * Handles CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME and CLOCK_TAI 316 * based timers. 317 */ 318 static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer) 319 { 320 struct k_itimer *timr = container_of(timer, struct k_itimer, it.real.timer); 321 322 guard(spinlock_irqsave)(&timr->it_lock); 323 posix_timer_queue_signal(timr); 324 return HRTIMER_NORESTART; 325 } 326 327 static struct pid *good_sigevent(sigevent_t * event) 328 { 329 struct pid *pid = task_tgid(current); 330 struct task_struct *rtn; 331 332 switch (event->sigev_notify) { 333 case SIGEV_SIGNAL | SIGEV_THREAD_ID: 334 pid = find_vpid(event->sigev_notify_thread_id); 335 rtn = pid_task(pid, PIDTYPE_PID); 336 if (!rtn || !same_thread_group(rtn, current)) 337 return NULL; 338 fallthrough; 339 case SIGEV_SIGNAL: 340 case SIGEV_THREAD: 341 if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX) 342 return NULL; 343 fallthrough; 344 case SIGEV_NONE: 345 return pid; 346 default: 347 return NULL; 348 } 349 } 350 351 static struct k_itimer *alloc_posix_timer(void) 352 { 353 struct k_itimer *tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL); 354 355 if (!tmr) 356 return tmr; 357 358 if (unlikely(!posixtimer_init_sigqueue(&tmr->sigq))) { 359 kmem_cache_free(posix_timers_cache, tmr); 360 return NULL; 361 } 362 rcuref_init(&tmr->rcuref, 1); 363 return tmr; 364 } 365 366 void posixtimer_free_timer(struct k_itimer *tmr) 367 { 368 put_pid(tmr->it_pid); 369 if (tmr->sigq.ucounts) 370 dec_rlimit_put_ucounts(tmr->sigq.ucounts, UCOUNT_RLIMIT_SIGPENDING); 371 kfree_rcu(tmr, rcu); 372 } 373 374 static void posix_timer_unhash_and_free(struct k_itimer *tmr) 375 { 376 spin_lock(&hash_lock); 377 hlist_del_rcu(&tmr->t_hash); 378 spin_unlock(&hash_lock); 379 posixtimer_putref(tmr); 380 } 381 382 static int common_timer_create(struct k_itimer *new_timer) 383 { 384 hrtimer_setup(&new_timer->it.real.timer, posix_timer_fn, new_timer->it_clock, 0); 385 return 0; 386 } 387 388 /* Create a POSIX.1b interval timer. */ 389 static int do_timer_create(clockid_t which_clock, struct sigevent *event, 390 timer_t __user *created_timer_id) 391 { 392 const struct k_clock *kc = clockid_to_kclock(which_clock); 393 struct k_itimer *new_timer; 394 int error, new_timer_id; 395 396 if (!kc) 397 return -EINVAL; 398 if (!kc->timer_create) 399 return -EOPNOTSUPP; 400 401 new_timer = alloc_posix_timer(); 402 if (unlikely(!new_timer)) 403 return -EAGAIN; 404 405 spin_lock_init(&new_timer->it_lock); 406 407 /* 408 * Add the timer to the hash table. The timer is not yet valid 409 * because new_timer::it_signal is still NULL. The timer id is also 410 * not yet visible to user space. 411 */ 412 new_timer_id = posix_timer_add(new_timer); 413 if (new_timer_id < 0) { 414 posixtimer_free_timer(new_timer); 415 return new_timer_id; 416 } 417 418 new_timer->it_id = (timer_t) new_timer_id; 419 new_timer->it_clock = which_clock; 420 new_timer->kclock = kc; 421 new_timer->it_overrun = -1LL; 422 423 if (event) { 424 rcu_read_lock(); 425 new_timer->it_pid = get_pid(good_sigevent(event)); 426 rcu_read_unlock(); 427 if (!new_timer->it_pid) { 428 error = -EINVAL; 429 goto out; 430 } 431 new_timer->it_sigev_notify = event->sigev_notify; 432 new_timer->sigq.info.si_signo = event->sigev_signo; 433 new_timer->sigq.info.si_value = event->sigev_value; 434 } else { 435 new_timer->it_sigev_notify = SIGEV_SIGNAL; 436 new_timer->sigq.info.si_signo = SIGALRM; 437 memset(&new_timer->sigq.info.si_value, 0, sizeof(sigval_t)); 438 new_timer->sigq.info.si_value.sival_int = new_timer->it_id; 439 new_timer->it_pid = get_pid(task_tgid(current)); 440 } 441 442 if (new_timer->it_sigev_notify & SIGEV_THREAD_ID) 443 new_timer->it_pid_type = PIDTYPE_PID; 444 else 445 new_timer->it_pid_type = PIDTYPE_TGID; 446 447 new_timer->sigq.info.si_tid = new_timer->it_id; 448 new_timer->sigq.info.si_code = SI_TIMER; 449 450 if (copy_to_user(created_timer_id, &new_timer_id, sizeof (new_timer_id))) { 451 error = -EFAULT; 452 goto out; 453 } 454 /* 455 * After succesful copy out, the timer ID is visible to user space 456 * now but not yet valid because new_timer::signal is still NULL. 457 * 458 * Complete the initialization with the clock specific create 459 * callback. 460 */ 461 error = kc->timer_create(new_timer); 462 if (error) 463 goto out; 464 465 spin_lock_irq(¤t->sighand->siglock); 466 /* This makes the timer valid in the hash table */ 467 WRITE_ONCE(new_timer->it_signal, current->signal); 468 hlist_add_head(&new_timer->list, ¤t->signal->posix_timers); 469 spin_unlock_irq(¤t->sighand->siglock); 470 /* 471 * After unlocking sighand::siglock @new_timer is subject to 472 * concurrent removal and cannot be touched anymore 473 */ 474 return 0; 475 out: 476 posix_timer_unhash_and_free(new_timer); 477 return error; 478 } 479 480 SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock, 481 struct sigevent __user *, timer_event_spec, 482 timer_t __user *, created_timer_id) 483 { 484 if (timer_event_spec) { 485 sigevent_t event; 486 487 if (copy_from_user(&event, timer_event_spec, sizeof (event))) 488 return -EFAULT; 489 return do_timer_create(which_clock, &event, created_timer_id); 490 } 491 return do_timer_create(which_clock, NULL, created_timer_id); 492 } 493 494 #ifdef CONFIG_COMPAT 495 COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock, 496 struct compat_sigevent __user *, timer_event_spec, 497 timer_t __user *, created_timer_id) 498 { 499 if (timer_event_spec) { 500 sigevent_t event; 501 502 if (get_compat_sigevent(&event, timer_event_spec)) 503 return -EFAULT; 504 return do_timer_create(which_clock, &event, created_timer_id); 505 } 506 return do_timer_create(which_clock, NULL, created_timer_id); 507 } 508 #endif 509 510 static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags) 511 { 512 struct k_itimer *timr; 513 514 /* 515 * timer_t could be any type >= int and we want to make sure any 516 * @timer_id outside positive int range fails lookup. 517 */ 518 if ((unsigned long long)timer_id > INT_MAX) 519 return NULL; 520 521 /* 522 * The hash lookup and the timers are RCU protected. 523 * 524 * Timers are added to the hash in invalid state where 525 * timr::it_signal == NULL. timer::it_signal is only set after the 526 * rest of the initialization succeeded. 527 * 528 * Timer destruction happens in steps: 529 * 1) Set timr::it_signal to NULL with timr::it_lock held 530 * 2) Release timr::it_lock 531 * 3) Remove from the hash under hash_lock 532 * 4) Put the reference count. 533 * 534 * The reference count might not drop to zero if timr::sigq is 535 * queued. In that case the signal delivery or flush will put the 536 * last reference count. 537 * 538 * When the reference count reaches zero, the timer is scheduled 539 * for RCU removal after the grace period. 540 * 541 * Holding rcu_read_lock() across the lookup ensures that 542 * the timer cannot be freed. 543 * 544 * The lookup validates locklessly that timr::it_signal == 545 * current::it_signal and timr::it_id == @timer_id. timr::it_id 546 * can't change, but timr::it_signal becomes NULL during 547 * destruction. 548 */ 549 rcu_read_lock(); 550 timr = posix_timer_by_id(timer_id); 551 if (timr) { 552 spin_lock_irqsave(&timr->it_lock, *flags); 553 /* 554 * Validate under timr::it_lock that timr::it_signal is 555 * still valid. Pairs with #1 above. 556 */ 557 if (timr->it_signal == current->signal) { 558 rcu_read_unlock(); 559 return timr; 560 } 561 spin_unlock_irqrestore(&timr->it_lock, *flags); 562 } 563 rcu_read_unlock(); 564 565 return NULL; 566 } 567 568 static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now) 569 { 570 struct hrtimer *timer = &timr->it.real.timer; 571 572 return __hrtimer_expires_remaining_adjusted(timer, now); 573 } 574 575 static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now) 576 { 577 struct hrtimer *timer = &timr->it.real.timer; 578 579 return hrtimer_forward(timer, now, timr->it_interval); 580 } 581 582 /* 583 * Get the time remaining on a POSIX.1b interval timer. 584 * 585 * Two issues to handle here: 586 * 587 * 1) The timer has a requeue pending. The return value must appear as 588 * if the timer has been requeued right now. 589 * 590 * 2) The timer is a SIGEV_NONE timer. These timers are never enqueued 591 * into the hrtimer queue and therefore never expired. Emulate expiry 592 * here taking #1 into account. 593 */ 594 void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting) 595 { 596 const struct k_clock *kc = timr->kclock; 597 ktime_t now, remaining, iv; 598 bool sig_none; 599 600 sig_none = timr->it_sigev_notify == SIGEV_NONE; 601 iv = timr->it_interval; 602 603 /* interval timer ? */ 604 if (iv) { 605 cur_setting->it_interval = ktime_to_timespec64(iv); 606 } else if (timr->it_status == POSIX_TIMER_DISARMED) { 607 /* 608 * SIGEV_NONE oneshot timers are never queued and therefore 609 * timr->it_status is always DISARMED. The check below 610 * vs. remaining time will handle this case. 611 * 612 * For all other timers there is nothing to update here, so 613 * return. 614 */ 615 if (!sig_none) 616 return; 617 } 618 619 now = kc->clock_get_ktime(timr->it_clock); 620 621 /* 622 * If this is an interval timer and either has requeue pending or 623 * is a SIGEV_NONE timer move the expiry time forward by intervals, 624 * so expiry is > now. 625 */ 626 if (iv && timr->it_status != POSIX_TIMER_ARMED) 627 timr->it_overrun += kc->timer_forward(timr, now); 628 629 remaining = kc->timer_remaining(timr, now); 630 /* 631 * As @now is retrieved before a possible timer_forward() and 632 * cannot be reevaluated by the compiler @remaining is based on the 633 * same @now value. Therefore @remaining is consistent vs. @now. 634 * 635 * Consequently all interval timers, i.e. @iv > 0, cannot have a 636 * remaining time <= 0 because timer_forward() guarantees to move 637 * them forward so that the next timer expiry is > @now. 638 */ 639 if (remaining <= 0) { 640 /* 641 * A single shot SIGEV_NONE timer must return 0, when it is 642 * expired! Timers which have a real signal delivery mode 643 * must return a remaining time greater than 0 because the 644 * signal has not yet been delivered. 645 */ 646 if (!sig_none) 647 cur_setting->it_value.tv_nsec = 1; 648 } else { 649 cur_setting->it_value = ktime_to_timespec64(remaining); 650 } 651 } 652 653 static int do_timer_gettime(timer_t timer_id, struct itimerspec64 *setting) 654 { 655 const struct k_clock *kc; 656 struct k_itimer *timr; 657 unsigned long flags; 658 int ret = 0; 659 660 timr = lock_timer(timer_id, &flags); 661 if (!timr) 662 return -EINVAL; 663 664 memset(setting, 0, sizeof(*setting)); 665 kc = timr->kclock; 666 if (WARN_ON_ONCE(!kc || !kc->timer_get)) 667 ret = -EINVAL; 668 else 669 kc->timer_get(timr, setting); 670 671 unlock_timer(timr, flags); 672 return ret; 673 } 674 675 /* Get the time remaining on a POSIX.1b interval timer. */ 676 SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id, 677 struct __kernel_itimerspec __user *, setting) 678 { 679 struct itimerspec64 cur_setting; 680 681 int ret = do_timer_gettime(timer_id, &cur_setting); 682 if (!ret) { 683 if (put_itimerspec64(&cur_setting, setting)) 684 ret = -EFAULT; 685 } 686 return ret; 687 } 688 689 #ifdef CONFIG_COMPAT_32BIT_TIME 690 691 SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id, 692 struct old_itimerspec32 __user *, setting) 693 { 694 struct itimerspec64 cur_setting; 695 696 int ret = do_timer_gettime(timer_id, &cur_setting); 697 if (!ret) { 698 if (put_old_itimerspec32(&cur_setting, setting)) 699 ret = -EFAULT; 700 } 701 return ret; 702 } 703 704 #endif 705 706 /** 707 * sys_timer_getoverrun - Get the number of overruns of a POSIX.1b interval timer 708 * @timer_id: The timer ID which identifies the timer 709 * 710 * The "overrun count" of a timer is one plus the number of expiration 711 * intervals which have elapsed between the first expiry, which queues the 712 * signal and the actual signal delivery. On signal delivery the "overrun 713 * count" is calculated and cached, so it can be returned directly here. 714 * 715 * As this is relative to the last queued signal the returned overrun count 716 * is meaningless outside of the signal delivery path and even there it 717 * does not accurately reflect the current state when user space evaluates 718 * it. 719 * 720 * Returns: 721 * -EINVAL @timer_id is invalid 722 * 1..INT_MAX The number of overruns related to the last delivered signal 723 */ 724 SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id) 725 { 726 struct k_itimer *timr; 727 unsigned long flags; 728 int overrun; 729 730 timr = lock_timer(timer_id, &flags); 731 if (!timr) 732 return -EINVAL; 733 734 overrun = timer_overrun_to_int(timr); 735 unlock_timer(timr, flags); 736 737 return overrun; 738 } 739 740 static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires, 741 bool absolute, bool sigev_none) 742 { 743 struct hrtimer *timer = &timr->it.real.timer; 744 enum hrtimer_mode mode; 745 746 mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; 747 /* 748 * Posix magic: Relative CLOCK_REALTIME timers are not affected by 749 * clock modifications, so they become CLOCK_MONOTONIC based under the 750 * hood. See hrtimer_setup(). Update timr->kclock, so the generic 751 * functions which use timr->kclock->clock_get_*() work. 752 * 753 * Note: it_clock stays unmodified, because the next timer_set() might 754 * use ABSTIME, so it needs to switch back. 755 */ 756 if (timr->it_clock == CLOCK_REALTIME) 757 timr->kclock = absolute ? &clock_realtime : &clock_monotonic; 758 759 hrtimer_setup(&timr->it.real.timer, posix_timer_fn, timr->it_clock, mode); 760 761 if (!absolute) 762 expires = ktime_add_safe(expires, timer->base->get_time()); 763 hrtimer_set_expires(timer, expires); 764 765 if (!sigev_none) 766 hrtimer_start_expires(timer, HRTIMER_MODE_ABS); 767 } 768 769 static int common_hrtimer_try_to_cancel(struct k_itimer *timr) 770 { 771 return hrtimer_try_to_cancel(&timr->it.real.timer); 772 } 773 774 static void common_timer_wait_running(struct k_itimer *timer) 775 { 776 hrtimer_cancel_wait_running(&timer->it.real.timer); 777 } 778 779 /* 780 * On PREEMPT_RT this prevents priority inversion and a potential livelock 781 * against the ksoftirqd thread in case that ksoftirqd gets preempted while 782 * executing a hrtimer callback. 783 * 784 * See the comments in hrtimer_cancel_wait_running(). For PREEMPT_RT=n this 785 * just results in a cpu_relax(). 786 * 787 * For POSIX CPU timers with CONFIG_POSIX_CPU_TIMERS_TASK_WORK=n this is 788 * just a cpu_relax(). With CONFIG_POSIX_CPU_TIMERS_TASK_WORK=y this 789 * prevents spinning on an eventually scheduled out task and a livelock 790 * when the task which tries to delete or disarm the timer has preempted 791 * the task which runs the expiry in task work context. 792 */ 793 static struct k_itimer *timer_wait_running(struct k_itimer *timer, 794 unsigned long *flags) 795 { 796 const struct k_clock *kc = READ_ONCE(timer->kclock); 797 timer_t timer_id = READ_ONCE(timer->it_id); 798 799 /* Prevent kfree(timer) after dropping the lock */ 800 rcu_read_lock(); 801 unlock_timer(timer, *flags); 802 803 /* 804 * kc->timer_wait_running() might drop RCU lock. So @timer 805 * cannot be touched anymore after the function returns! 806 */ 807 if (!WARN_ON_ONCE(!kc->timer_wait_running)) 808 kc->timer_wait_running(timer); 809 810 rcu_read_unlock(); 811 /* Relock the timer. It might be not longer hashed. */ 812 return lock_timer(timer_id, flags); 813 } 814 815 /* 816 * Set up the new interval and reset the signal delivery data 817 */ 818 void posix_timer_set_common(struct k_itimer *timer, struct itimerspec64 *new_setting) 819 { 820 if (new_setting->it_value.tv_sec || new_setting->it_value.tv_nsec) 821 timer->it_interval = timespec64_to_ktime(new_setting->it_interval); 822 else 823 timer->it_interval = 0; 824 825 /* Reset overrun accounting */ 826 timer->it_overrun_last = 0; 827 timer->it_overrun = -1LL; 828 } 829 830 /* Set a POSIX.1b interval timer. */ 831 int common_timer_set(struct k_itimer *timr, int flags, 832 struct itimerspec64 *new_setting, 833 struct itimerspec64 *old_setting) 834 { 835 const struct k_clock *kc = timr->kclock; 836 bool sigev_none; 837 ktime_t expires; 838 839 if (old_setting) 840 common_timer_get(timr, old_setting); 841 842 /* 843 * Careful here. On SMP systems the timer expiry function could be 844 * active and spinning on timr->it_lock. 845 */ 846 if (kc->timer_try_to_cancel(timr) < 0) 847 return TIMER_RETRY; 848 849 timr->it_status = POSIX_TIMER_DISARMED; 850 posix_timer_set_common(timr, new_setting); 851 852 /* Keep timer disarmed when it_value is zero */ 853 if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) 854 return 0; 855 856 expires = timespec64_to_ktime(new_setting->it_value); 857 if (flags & TIMER_ABSTIME) 858 expires = timens_ktime_to_host(timr->it_clock, expires); 859 sigev_none = timr->it_sigev_notify == SIGEV_NONE; 860 861 kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none); 862 if (!sigev_none) 863 timr->it_status = POSIX_TIMER_ARMED; 864 return 0; 865 } 866 867 static int do_timer_settime(timer_t timer_id, int tmr_flags, 868 struct itimerspec64 *new_spec64, 869 struct itimerspec64 *old_spec64) 870 { 871 const struct k_clock *kc; 872 struct k_itimer *timr; 873 unsigned long flags; 874 int error; 875 876 if (!timespec64_valid(&new_spec64->it_interval) || 877 !timespec64_valid(&new_spec64->it_value)) 878 return -EINVAL; 879 880 if (old_spec64) 881 memset(old_spec64, 0, sizeof(*old_spec64)); 882 883 timr = lock_timer(timer_id, &flags); 884 retry: 885 if (!timr) 886 return -EINVAL; 887 888 if (old_spec64) 889 old_spec64->it_interval = ktime_to_timespec64(timr->it_interval); 890 891 /* Prevent signal delivery and rearming. */ 892 timr->it_signal_seq++; 893 894 kc = timr->kclock; 895 if (WARN_ON_ONCE(!kc || !kc->timer_set)) 896 error = -EINVAL; 897 else 898 error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64); 899 900 if (error == TIMER_RETRY) { 901 // We already got the old time... 902 old_spec64 = NULL; 903 /* Unlocks and relocks the timer if it still exists */ 904 timr = timer_wait_running(timr, &flags); 905 goto retry; 906 } 907 unlock_timer(timr, flags); 908 909 return error; 910 } 911 912 /* Set a POSIX.1b interval timer */ 913 SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags, 914 const struct __kernel_itimerspec __user *, new_setting, 915 struct __kernel_itimerspec __user *, old_setting) 916 { 917 struct itimerspec64 new_spec, old_spec, *rtn; 918 int error = 0; 919 920 if (!new_setting) 921 return -EINVAL; 922 923 if (get_itimerspec64(&new_spec, new_setting)) 924 return -EFAULT; 925 926 rtn = old_setting ? &old_spec : NULL; 927 error = do_timer_settime(timer_id, flags, &new_spec, rtn); 928 if (!error && old_setting) { 929 if (put_itimerspec64(&old_spec, old_setting)) 930 error = -EFAULT; 931 } 932 return error; 933 } 934 935 #ifdef CONFIG_COMPAT_32BIT_TIME 936 SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags, 937 struct old_itimerspec32 __user *, new, 938 struct old_itimerspec32 __user *, old) 939 { 940 struct itimerspec64 new_spec, old_spec; 941 struct itimerspec64 *rtn = old ? &old_spec : NULL; 942 int error = 0; 943 944 if (!new) 945 return -EINVAL; 946 if (get_old_itimerspec32(&new_spec, new)) 947 return -EFAULT; 948 949 error = do_timer_settime(timer_id, flags, &new_spec, rtn); 950 if (!error && old) { 951 if (put_old_itimerspec32(&old_spec, old)) 952 error = -EFAULT; 953 } 954 return error; 955 } 956 #endif 957 958 int common_timer_del(struct k_itimer *timer) 959 { 960 const struct k_clock *kc = timer->kclock; 961 962 if (kc->timer_try_to_cancel(timer) < 0) 963 return TIMER_RETRY; 964 timer->it_status = POSIX_TIMER_DISARMED; 965 return 0; 966 } 967 968 /* 969 * If the deleted timer is on the ignored list, remove it and 970 * drop the associated reference. 971 */ 972 static inline void posix_timer_cleanup_ignored(struct k_itimer *tmr) 973 { 974 if (!hlist_unhashed(&tmr->ignored_list)) { 975 hlist_del_init(&tmr->ignored_list); 976 posixtimer_putref(tmr); 977 } 978 } 979 980 static inline int timer_delete_hook(struct k_itimer *timer) 981 { 982 const struct k_clock *kc = timer->kclock; 983 984 /* Prevent signal delivery and rearming. */ 985 timer->it_signal_seq++; 986 987 if (WARN_ON_ONCE(!kc || !kc->timer_del)) 988 return -EINVAL; 989 return kc->timer_del(timer); 990 } 991 992 /* Delete a POSIX.1b interval timer. */ 993 SYSCALL_DEFINE1(timer_delete, timer_t, timer_id) 994 { 995 struct k_itimer *timer; 996 unsigned long flags; 997 998 timer = lock_timer(timer_id, &flags); 999 1000 retry_delete: 1001 if (!timer) 1002 return -EINVAL; 1003 1004 if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) { 1005 /* Unlocks and relocks the timer if it still exists */ 1006 timer = timer_wait_running(timer, &flags); 1007 goto retry_delete; 1008 } 1009 1010 spin_lock(¤t->sighand->siglock); 1011 hlist_del(&timer->list); 1012 posix_timer_cleanup_ignored(timer); 1013 /* 1014 * A concurrent lookup could check timer::it_signal lockless. It 1015 * will reevaluate with timer::it_lock held and observe the NULL. 1016 * 1017 * It must be written with siglock held so that the signal code 1018 * observes timer->it_signal == NULL in do_sigaction(SIG_IGN), 1019 * which prevents it from moving a pending signal of a deleted 1020 * timer to the ignore list. 1021 */ 1022 WRITE_ONCE(timer->it_signal, NULL); 1023 spin_unlock(¤t->sighand->siglock); 1024 1025 unlock_timer(timer, flags); 1026 posix_timer_unhash_and_free(timer); 1027 return 0; 1028 } 1029 1030 /* 1031 * Delete a timer if it is armed, remove it from the hash and schedule it 1032 * for RCU freeing. 1033 */ 1034 static void itimer_delete(struct k_itimer *timer) 1035 { 1036 unsigned long flags; 1037 1038 /* 1039 * irqsave is required to make timer_wait_running() work. 1040 */ 1041 spin_lock_irqsave(&timer->it_lock, flags); 1042 1043 retry_delete: 1044 /* 1045 * Even if the timer is not longer accessible from other tasks 1046 * it still might be armed and queued in the underlying timer 1047 * mechanism. Worse, that timer mechanism might run the expiry 1048 * function concurrently. 1049 */ 1050 if (timer_delete_hook(timer) == TIMER_RETRY) { 1051 /* 1052 * Timer is expired concurrently, prevent livelocks 1053 * and pointless spinning on RT. 1054 * 1055 * timer_wait_running() drops timer::it_lock, which opens 1056 * the possibility for another task to delete the timer. 1057 * 1058 * That's not possible here because this is invoked from 1059 * do_exit() only for the last thread of the thread group. 1060 * So no other task can access and delete that timer. 1061 */ 1062 if (WARN_ON_ONCE(timer_wait_running(timer, &flags) != timer)) 1063 return; 1064 1065 goto retry_delete; 1066 } 1067 hlist_del(&timer->list); 1068 1069 posix_timer_cleanup_ignored(timer); 1070 1071 /* 1072 * Setting timer::it_signal to NULL is technically not required 1073 * here as nothing can access the timer anymore legitimately via 1074 * the hash table. Set it to NULL nevertheless so that all deletion 1075 * paths are consistent. 1076 */ 1077 WRITE_ONCE(timer->it_signal, NULL); 1078 1079 spin_unlock_irqrestore(&timer->it_lock, flags); 1080 posix_timer_unhash_and_free(timer); 1081 } 1082 1083 /* 1084 * Invoked from do_exit() when the last thread of a thread group exits. 1085 * At that point no other task can access the timers of the dying 1086 * task anymore. 1087 */ 1088 void exit_itimers(struct task_struct *tsk) 1089 { 1090 struct hlist_head timers; 1091 1092 if (hlist_empty(&tsk->signal->posix_timers)) 1093 return; 1094 1095 /* Protect against concurrent read via /proc/$PID/timers */ 1096 spin_lock_irq(&tsk->sighand->siglock); 1097 hlist_move_list(&tsk->signal->posix_timers, &timers); 1098 spin_unlock_irq(&tsk->sighand->siglock); 1099 1100 /* The timers are not longer accessible via tsk::signal */ 1101 while (!hlist_empty(&timers)) 1102 itimer_delete(hlist_entry(timers.first, struct k_itimer, list)); 1103 1104 /* 1105 * There should be no timers on the ignored list. itimer_delete() has 1106 * mopped them up. 1107 */ 1108 if (!WARN_ON_ONCE(!hlist_empty(&tsk->signal->ignored_posix_timers))) 1109 return; 1110 1111 hlist_move_list(&tsk->signal->ignored_posix_timers, &timers); 1112 while (!hlist_empty(&timers)) { 1113 posix_timer_cleanup_ignored(hlist_entry(timers.first, struct k_itimer, 1114 ignored_list)); 1115 } 1116 } 1117 1118 SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock, 1119 const struct __kernel_timespec __user *, tp) 1120 { 1121 const struct k_clock *kc = clockid_to_kclock(which_clock); 1122 struct timespec64 new_tp; 1123 1124 if (!kc || !kc->clock_set) 1125 return -EINVAL; 1126 1127 if (get_timespec64(&new_tp, tp)) 1128 return -EFAULT; 1129 1130 /* 1131 * Permission checks have to be done inside the clock specific 1132 * setter callback. 1133 */ 1134 return kc->clock_set(which_clock, &new_tp); 1135 } 1136 1137 SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, 1138 struct __kernel_timespec __user *, tp) 1139 { 1140 const struct k_clock *kc = clockid_to_kclock(which_clock); 1141 struct timespec64 kernel_tp; 1142 int error; 1143 1144 if (!kc) 1145 return -EINVAL; 1146 1147 error = kc->clock_get_timespec(which_clock, &kernel_tp); 1148 1149 if (!error && put_timespec64(&kernel_tp, tp)) 1150 error = -EFAULT; 1151 1152 return error; 1153 } 1154 1155 int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx) 1156 { 1157 const struct k_clock *kc = clockid_to_kclock(which_clock); 1158 1159 if (!kc) 1160 return -EINVAL; 1161 if (!kc->clock_adj) 1162 return -EOPNOTSUPP; 1163 1164 return kc->clock_adj(which_clock, ktx); 1165 } 1166 1167 SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock, 1168 struct __kernel_timex __user *, utx) 1169 { 1170 struct __kernel_timex ktx; 1171 int err; 1172 1173 if (copy_from_user(&ktx, utx, sizeof(ktx))) 1174 return -EFAULT; 1175 1176 err = do_clock_adjtime(which_clock, &ktx); 1177 1178 if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx))) 1179 return -EFAULT; 1180 1181 return err; 1182 } 1183 1184 /** 1185 * sys_clock_getres - Get the resolution of a clock 1186 * @which_clock: The clock to get the resolution for 1187 * @tp: Pointer to a a user space timespec64 for storage 1188 * 1189 * POSIX defines: 1190 * 1191 * "The clock_getres() function shall return the resolution of any 1192 * clock. Clock resolutions are implementation-defined and cannot be set by 1193 * a process. If the argument res is not NULL, the resolution of the 1194 * specified clock shall be stored in the location pointed to by res. If 1195 * res is NULL, the clock resolution is not returned. If the time argument 1196 * of clock_settime() is not a multiple of res, then the value is truncated 1197 * to a multiple of res." 1198 * 1199 * Due to the various hardware constraints the real resolution can vary 1200 * wildly and even change during runtime when the underlying devices are 1201 * replaced. The kernel also can use hardware devices with different 1202 * resolutions for reading the time and for arming timers. 1203 * 1204 * The kernel therefore deviates from the POSIX spec in various aspects: 1205 * 1206 * 1) The resolution returned to user space 1207 * 1208 * For CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, CLOCK_TAI, 1209 * CLOCK_REALTIME_ALARM, CLOCK_BOOTTIME_ALAREM and CLOCK_MONOTONIC_RAW 1210 * the kernel differentiates only two cases: 1211 * 1212 * I) Low resolution mode: 1213 * 1214 * When high resolution timers are disabled at compile or runtime 1215 * the resolution returned is nanoseconds per tick, which represents 1216 * the precision at which timers expire. 1217 * 1218 * II) High resolution mode: 1219 * 1220 * When high resolution timers are enabled the resolution returned 1221 * is always one nanosecond independent of the actual resolution of 1222 * the underlying hardware devices. 1223 * 1224 * For CLOCK_*_ALARM the actual resolution depends on system 1225 * state. When system is running the resolution is the same as the 1226 * resolution of the other clocks. During suspend the actual 1227 * resolution is the resolution of the underlying RTC device which 1228 * might be way less precise than the clockevent device used during 1229 * running state. 1230 * 1231 * For CLOCK_REALTIME_COARSE and CLOCK_MONOTONIC_COARSE the resolution 1232 * returned is always nanoseconds per tick. 1233 * 1234 * For CLOCK_PROCESS_CPUTIME and CLOCK_THREAD_CPUTIME the resolution 1235 * returned is always one nanosecond under the assumption that the 1236 * underlying scheduler clock has a better resolution than nanoseconds 1237 * per tick. 1238 * 1239 * For dynamic POSIX clocks (PTP devices) the resolution returned is 1240 * always one nanosecond. 1241 * 1242 * 2) Affect on sys_clock_settime() 1243 * 1244 * The kernel does not truncate the time which is handed in to 1245 * sys_clock_settime(). The kernel internal timekeeping is always using 1246 * nanoseconds precision independent of the clocksource device which is 1247 * used to read the time from. The resolution of that device only 1248 * affects the presicion of the time returned by sys_clock_gettime(). 1249 * 1250 * Returns: 1251 * 0 Success. @tp contains the resolution 1252 * -EINVAL @which_clock is not a valid clock ID 1253 * -EFAULT Copying the resolution to @tp faulted 1254 * -ENODEV Dynamic POSIX clock is not backed by a device 1255 * -EOPNOTSUPP Dynamic POSIX clock does not support getres() 1256 */ 1257 SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock, 1258 struct __kernel_timespec __user *, tp) 1259 { 1260 const struct k_clock *kc = clockid_to_kclock(which_clock); 1261 struct timespec64 rtn_tp; 1262 int error; 1263 1264 if (!kc) 1265 return -EINVAL; 1266 1267 error = kc->clock_getres(which_clock, &rtn_tp); 1268 1269 if (!error && tp && put_timespec64(&rtn_tp, tp)) 1270 error = -EFAULT; 1271 1272 return error; 1273 } 1274 1275 #ifdef CONFIG_COMPAT_32BIT_TIME 1276 1277 SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock, 1278 struct old_timespec32 __user *, tp) 1279 { 1280 const struct k_clock *kc = clockid_to_kclock(which_clock); 1281 struct timespec64 ts; 1282 1283 if (!kc || !kc->clock_set) 1284 return -EINVAL; 1285 1286 if (get_old_timespec32(&ts, tp)) 1287 return -EFAULT; 1288 1289 return kc->clock_set(which_clock, &ts); 1290 } 1291 1292 SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock, 1293 struct old_timespec32 __user *, tp) 1294 { 1295 const struct k_clock *kc = clockid_to_kclock(which_clock); 1296 struct timespec64 ts; 1297 int err; 1298 1299 if (!kc) 1300 return -EINVAL; 1301 1302 err = kc->clock_get_timespec(which_clock, &ts); 1303 1304 if (!err && put_old_timespec32(&ts, tp)) 1305 err = -EFAULT; 1306 1307 return err; 1308 } 1309 1310 SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock, 1311 struct old_timex32 __user *, utp) 1312 { 1313 struct __kernel_timex ktx; 1314 int err; 1315 1316 err = get_old_timex32(&ktx, utp); 1317 if (err) 1318 return err; 1319 1320 err = do_clock_adjtime(which_clock, &ktx); 1321 1322 if (err >= 0 && put_old_timex32(utp, &ktx)) 1323 return -EFAULT; 1324 1325 return err; 1326 } 1327 1328 SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock, 1329 struct old_timespec32 __user *, tp) 1330 { 1331 const struct k_clock *kc = clockid_to_kclock(which_clock); 1332 struct timespec64 ts; 1333 int err; 1334 1335 if (!kc) 1336 return -EINVAL; 1337 1338 err = kc->clock_getres(which_clock, &ts); 1339 if (!err && tp && put_old_timespec32(&ts, tp)) 1340 return -EFAULT; 1341 1342 return err; 1343 } 1344 1345 #endif 1346 1347 /* 1348 * sys_clock_nanosleep() for CLOCK_REALTIME and CLOCK_TAI 1349 */ 1350 static int common_nsleep(const clockid_t which_clock, int flags, 1351 const struct timespec64 *rqtp) 1352 { 1353 ktime_t texp = timespec64_to_ktime(*rqtp); 1354 1355 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ? 1356 HRTIMER_MODE_ABS : HRTIMER_MODE_REL, 1357 which_clock); 1358 } 1359 1360 /* 1361 * sys_clock_nanosleep() for CLOCK_MONOTONIC and CLOCK_BOOTTIME 1362 * 1363 * Absolute nanosleeps for these clocks are time-namespace adjusted. 1364 */ 1365 static int common_nsleep_timens(const clockid_t which_clock, int flags, 1366 const struct timespec64 *rqtp) 1367 { 1368 ktime_t texp = timespec64_to_ktime(*rqtp); 1369 1370 if (flags & TIMER_ABSTIME) 1371 texp = timens_ktime_to_host(which_clock, texp); 1372 1373 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ? 1374 HRTIMER_MODE_ABS : HRTIMER_MODE_REL, 1375 which_clock); 1376 } 1377 1378 SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags, 1379 const struct __kernel_timespec __user *, rqtp, 1380 struct __kernel_timespec __user *, rmtp) 1381 { 1382 const struct k_clock *kc = clockid_to_kclock(which_clock); 1383 struct timespec64 t; 1384 1385 if (!kc) 1386 return -EINVAL; 1387 if (!kc->nsleep) 1388 return -EOPNOTSUPP; 1389 1390 if (get_timespec64(&t, rqtp)) 1391 return -EFAULT; 1392 1393 if (!timespec64_valid(&t)) 1394 return -EINVAL; 1395 if (flags & TIMER_ABSTIME) 1396 rmtp = NULL; 1397 current->restart_block.fn = do_no_restart_syscall; 1398 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE; 1399 current->restart_block.nanosleep.rmtp = rmtp; 1400 1401 return kc->nsleep(which_clock, flags, &t); 1402 } 1403 1404 #ifdef CONFIG_COMPAT_32BIT_TIME 1405 1406 SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags, 1407 struct old_timespec32 __user *, rqtp, 1408 struct old_timespec32 __user *, rmtp) 1409 { 1410 const struct k_clock *kc = clockid_to_kclock(which_clock); 1411 struct timespec64 t; 1412 1413 if (!kc) 1414 return -EINVAL; 1415 if (!kc->nsleep) 1416 return -EOPNOTSUPP; 1417 1418 if (get_old_timespec32(&t, rqtp)) 1419 return -EFAULT; 1420 1421 if (!timespec64_valid(&t)) 1422 return -EINVAL; 1423 if (flags & TIMER_ABSTIME) 1424 rmtp = NULL; 1425 current->restart_block.fn = do_no_restart_syscall; 1426 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE; 1427 current->restart_block.nanosleep.compat_rmtp = rmtp; 1428 1429 return kc->nsleep(which_clock, flags, &t); 1430 } 1431 1432 #endif 1433 1434 static const struct k_clock clock_realtime = { 1435 .clock_getres = posix_get_hrtimer_res, 1436 .clock_get_timespec = posix_get_realtime_timespec, 1437 .clock_get_ktime = posix_get_realtime_ktime, 1438 .clock_set = posix_clock_realtime_set, 1439 .clock_adj = posix_clock_realtime_adj, 1440 .nsleep = common_nsleep, 1441 .timer_create = common_timer_create, 1442 .timer_set = common_timer_set, 1443 .timer_get = common_timer_get, 1444 .timer_del = common_timer_del, 1445 .timer_rearm = common_hrtimer_rearm, 1446 .timer_forward = common_hrtimer_forward, 1447 .timer_remaining = common_hrtimer_remaining, 1448 .timer_try_to_cancel = common_hrtimer_try_to_cancel, 1449 .timer_wait_running = common_timer_wait_running, 1450 .timer_arm = common_hrtimer_arm, 1451 }; 1452 1453 static const struct k_clock clock_monotonic = { 1454 .clock_getres = posix_get_hrtimer_res, 1455 .clock_get_timespec = posix_get_monotonic_timespec, 1456 .clock_get_ktime = posix_get_monotonic_ktime, 1457 .nsleep = common_nsleep_timens, 1458 .timer_create = common_timer_create, 1459 .timer_set = common_timer_set, 1460 .timer_get = common_timer_get, 1461 .timer_del = common_timer_del, 1462 .timer_rearm = common_hrtimer_rearm, 1463 .timer_forward = common_hrtimer_forward, 1464 .timer_remaining = common_hrtimer_remaining, 1465 .timer_try_to_cancel = common_hrtimer_try_to_cancel, 1466 .timer_wait_running = common_timer_wait_running, 1467 .timer_arm = common_hrtimer_arm, 1468 }; 1469 1470 static const struct k_clock clock_monotonic_raw = { 1471 .clock_getres = posix_get_hrtimer_res, 1472 .clock_get_timespec = posix_get_monotonic_raw, 1473 }; 1474 1475 static const struct k_clock clock_realtime_coarse = { 1476 .clock_getres = posix_get_coarse_res, 1477 .clock_get_timespec = posix_get_realtime_coarse, 1478 }; 1479 1480 static const struct k_clock clock_monotonic_coarse = { 1481 .clock_getres = posix_get_coarse_res, 1482 .clock_get_timespec = posix_get_monotonic_coarse, 1483 }; 1484 1485 static const struct k_clock clock_tai = { 1486 .clock_getres = posix_get_hrtimer_res, 1487 .clock_get_ktime = posix_get_tai_ktime, 1488 .clock_get_timespec = posix_get_tai_timespec, 1489 .nsleep = common_nsleep, 1490 .timer_create = common_timer_create, 1491 .timer_set = common_timer_set, 1492 .timer_get = common_timer_get, 1493 .timer_del = common_timer_del, 1494 .timer_rearm = common_hrtimer_rearm, 1495 .timer_forward = common_hrtimer_forward, 1496 .timer_remaining = common_hrtimer_remaining, 1497 .timer_try_to_cancel = common_hrtimer_try_to_cancel, 1498 .timer_wait_running = common_timer_wait_running, 1499 .timer_arm = common_hrtimer_arm, 1500 }; 1501 1502 static const struct k_clock clock_boottime = { 1503 .clock_getres = posix_get_hrtimer_res, 1504 .clock_get_ktime = posix_get_boottime_ktime, 1505 .clock_get_timespec = posix_get_boottime_timespec, 1506 .nsleep = common_nsleep_timens, 1507 .timer_create = common_timer_create, 1508 .timer_set = common_timer_set, 1509 .timer_get = common_timer_get, 1510 .timer_del = common_timer_del, 1511 .timer_rearm = common_hrtimer_rearm, 1512 .timer_forward = common_hrtimer_forward, 1513 .timer_remaining = common_hrtimer_remaining, 1514 .timer_try_to_cancel = common_hrtimer_try_to_cancel, 1515 .timer_wait_running = common_timer_wait_running, 1516 .timer_arm = common_hrtimer_arm, 1517 }; 1518 1519 static const struct k_clock * const posix_clocks[] = { 1520 [CLOCK_REALTIME] = &clock_realtime, 1521 [CLOCK_MONOTONIC] = &clock_monotonic, 1522 [CLOCK_PROCESS_CPUTIME_ID] = &clock_process, 1523 [CLOCK_THREAD_CPUTIME_ID] = &clock_thread, 1524 [CLOCK_MONOTONIC_RAW] = &clock_monotonic_raw, 1525 [CLOCK_REALTIME_COARSE] = &clock_realtime_coarse, 1526 [CLOCK_MONOTONIC_COARSE] = &clock_monotonic_coarse, 1527 [CLOCK_BOOTTIME] = &clock_boottime, 1528 [CLOCK_REALTIME_ALARM] = &alarm_clock, 1529 [CLOCK_BOOTTIME_ALARM] = &alarm_clock, 1530 [CLOCK_TAI] = &clock_tai, 1531 }; 1532 1533 static const struct k_clock *clockid_to_kclock(const clockid_t id) 1534 { 1535 clockid_t idx = id; 1536 1537 if (id < 0) { 1538 return (id & CLOCKFD_MASK) == CLOCKFD ? 1539 &clock_posix_dynamic : &clock_posix_cpu; 1540 } 1541 1542 if (id >= ARRAY_SIZE(posix_clocks)) 1543 return NULL; 1544 1545 return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))]; 1546 } 1547