xref: /linux/kernel/time/posix-timers.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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, int baseval)
237 {
238 	s64 sum = timr->it_overrun_last + (s64)baseval;
239 
240 	return sum > (s64)INT_MAX ? INT_MAX : (int)sum;
241 }
242 
243 static void common_hrtimer_rearm(struct k_itimer *timr)
244 {
245 	struct hrtimer *timer = &timr->it.real.timer;
246 
247 	timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
248 					    timr->it_interval);
249 	hrtimer_restart(timer);
250 }
251 
252 /*
253  * This function is called from the signal delivery code if
254  * info->si_sys_private is not zero, which indicates that the timer has to
255  * be rearmed. Restart the timer and update info::si_overrun.
256  */
257 void posixtimer_rearm(struct kernel_siginfo *info)
258 {
259 	struct k_itimer *timr;
260 	unsigned long flags;
261 
262 	timr = lock_timer(info->si_tid, &flags);
263 	if (!timr)
264 		return;
265 
266 	if (timr->it_interval && timr->it_requeue_pending == info->si_sys_private) {
267 		timr->kclock->timer_rearm(timr);
268 
269 		timr->it_active = 1;
270 		timr->it_overrun_last = timr->it_overrun;
271 		timr->it_overrun = -1LL;
272 		++timr->it_requeue_pending;
273 
274 		info->si_overrun = timer_overrun_to_int(timr, info->si_overrun);
275 	}
276 
277 	unlock_timer(timr, flags);
278 }
279 
280 int posix_timer_queue_signal(struct k_itimer *timr)
281 {
282 	int ret, si_private = 0;
283 	enum pid_type type;
284 
285 	lockdep_assert_held(&timr->it_lock);
286 
287 	timr->it_active = 0;
288 	if (timr->it_interval)
289 		si_private = ++timr->it_requeue_pending;
290 
291 	/*
292 	 * FIXME: if ->sigq is queued we can race with
293 	 * dequeue_signal()->posixtimer_rearm().
294 	 *
295 	 * If dequeue_signal() sees the "right" value of
296 	 * si_sys_private it calls posixtimer_rearm().
297 	 * We re-queue ->sigq and drop ->it_lock().
298 	 * posixtimer_rearm() locks the timer
299 	 * and re-schedules it while ->sigq is pending.
300 	 * Not really bad, but not that we want.
301 	 */
302 	timr->sigq->info.si_sys_private = si_private;
303 
304 	type = !(timr->it_sigev_notify & SIGEV_THREAD_ID) ? PIDTYPE_TGID : PIDTYPE_PID;
305 	ret = send_sigqueue(timr->sigq, timr->it_pid, type);
306 	/* If we failed to send the signal the timer stops. */
307 	return ret > 0;
308 }
309 
310 /*
311  * This function gets called when a POSIX.1b interval timer expires from
312  * the HRTIMER interrupt (soft interrupt on RT kernels).
313  *
314  * Handles CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME and CLOCK_TAI
315  * based timers.
316  */
317 static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
318 {
319 	struct k_itimer *timr = container_of(timer, struct k_itimer, it.real.timer);
320 	enum hrtimer_restart ret = HRTIMER_NORESTART;
321 	unsigned long flags;
322 
323 	spin_lock_irqsave(&timr->it_lock, flags);
324 
325 	if (posix_timer_queue_signal(timr)) {
326 		/*
327 		 * The signal was not queued due to SIG_IGN. As a
328 		 * consequence the timer is not going to be rearmed from
329 		 * the signal delivery path. But as a real signal handler
330 		 * can be installed later the timer must be rearmed here.
331 		 */
332 		if (timr->it_interval != 0) {
333 			ktime_t now = hrtimer_cb_get_time(timer);
334 
335 			/*
336 			 * FIXME: What we really want, is to stop this
337 			 * timer completely and restart it in case the
338 			 * SIG_IGN is removed. This is a non trivial
339 			 * change to the signal handling code.
340 			 *
341 			 * For now let timers with an interval less than a
342 			 * jiffy expire every jiffy and recheck for a
343 			 * valid signal handler.
344 			 *
345 			 * This avoids interrupt starvation in case of a
346 			 * very small interval, which would expire the
347 			 * timer immediately again.
348 			 *
349 			 * Moving now ahead of time by one jiffy tricks
350 			 * hrtimer_forward() to expire the timer later,
351 			 * while it still maintains the overrun accuracy
352 			 * for the price of a slight inconsistency in the
353 			 * timer_gettime() case. This is at least better
354 			 * than a timer storm.
355 			 *
356 			 * Only required when high resolution timers are
357 			 * enabled as the periodic tick based timers are
358 			 * automatically aligned to the next tick.
359 			 */
360 			if (IS_ENABLED(CONFIG_HIGH_RES_TIMERS)) {
361 				ktime_t kj = TICK_NSEC;
362 
363 				if (timr->it_interval < kj)
364 					now = ktime_add(now, kj);
365 			}
366 
367 			timr->it_overrun += hrtimer_forward(timer, now, timr->it_interval);
368 			ret = HRTIMER_RESTART;
369 			++timr->it_requeue_pending;
370 			timr->it_active = 1;
371 		}
372 	}
373 
374 	unlock_timer(timr, flags);
375 	return ret;
376 }
377 
378 static struct pid *good_sigevent(sigevent_t * event)
379 {
380 	struct pid *pid = task_tgid(current);
381 	struct task_struct *rtn;
382 
383 	switch (event->sigev_notify) {
384 	case SIGEV_SIGNAL | SIGEV_THREAD_ID:
385 		pid = find_vpid(event->sigev_notify_thread_id);
386 		rtn = pid_task(pid, PIDTYPE_PID);
387 		if (!rtn || !same_thread_group(rtn, current))
388 			return NULL;
389 		fallthrough;
390 	case SIGEV_SIGNAL:
391 	case SIGEV_THREAD:
392 		if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
393 			return NULL;
394 		fallthrough;
395 	case SIGEV_NONE:
396 		return pid;
397 	default:
398 		return NULL;
399 	}
400 }
401 
402 static struct k_itimer * alloc_posix_timer(void)
403 {
404 	struct k_itimer *tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
405 
406 	if (!tmr)
407 		return tmr;
408 	if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
409 		kmem_cache_free(posix_timers_cache, tmr);
410 		return NULL;
411 	}
412 	clear_siginfo(&tmr->sigq->info);
413 	return tmr;
414 }
415 
416 static void k_itimer_rcu_free(struct rcu_head *head)
417 {
418 	struct k_itimer *tmr = container_of(head, struct k_itimer, rcu);
419 
420 	kmem_cache_free(posix_timers_cache, tmr);
421 }
422 
423 static void posix_timer_free(struct k_itimer *tmr)
424 {
425 	put_pid(tmr->it_pid);
426 	sigqueue_free(tmr->sigq);
427 	call_rcu(&tmr->rcu, k_itimer_rcu_free);
428 }
429 
430 static void posix_timer_unhash_and_free(struct k_itimer *tmr)
431 {
432 	spin_lock(&hash_lock);
433 	hlist_del_rcu(&tmr->t_hash);
434 	spin_unlock(&hash_lock);
435 	posix_timer_free(tmr);
436 }
437 
438 static int common_timer_create(struct k_itimer *new_timer)
439 {
440 	hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
441 	return 0;
442 }
443 
444 /* Create a POSIX.1b interval timer. */
445 static int do_timer_create(clockid_t which_clock, struct sigevent *event,
446 			   timer_t __user *created_timer_id)
447 {
448 	const struct k_clock *kc = clockid_to_kclock(which_clock);
449 	struct k_itimer *new_timer;
450 	int error, new_timer_id;
451 
452 	if (!kc)
453 		return -EINVAL;
454 	if (!kc->timer_create)
455 		return -EOPNOTSUPP;
456 
457 	new_timer = alloc_posix_timer();
458 	if (unlikely(!new_timer))
459 		return -EAGAIN;
460 
461 	spin_lock_init(&new_timer->it_lock);
462 
463 	/*
464 	 * Add the timer to the hash table. The timer is not yet valid
465 	 * because new_timer::it_signal is still NULL. The timer id is also
466 	 * not yet visible to user space.
467 	 */
468 	new_timer_id = posix_timer_add(new_timer);
469 	if (new_timer_id < 0) {
470 		posix_timer_free(new_timer);
471 		return new_timer_id;
472 	}
473 
474 	new_timer->it_id = (timer_t) new_timer_id;
475 	new_timer->it_clock = which_clock;
476 	new_timer->kclock = kc;
477 	new_timer->it_overrun = -1LL;
478 
479 	if (event) {
480 		rcu_read_lock();
481 		new_timer->it_pid = get_pid(good_sigevent(event));
482 		rcu_read_unlock();
483 		if (!new_timer->it_pid) {
484 			error = -EINVAL;
485 			goto out;
486 		}
487 		new_timer->it_sigev_notify     = event->sigev_notify;
488 		new_timer->sigq->info.si_signo = event->sigev_signo;
489 		new_timer->sigq->info.si_value = event->sigev_value;
490 	} else {
491 		new_timer->it_sigev_notify     = SIGEV_SIGNAL;
492 		new_timer->sigq->info.si_signo = SIGALRM;
493 		memset(&new_timer->sigq->info.si_value, 0, sizeof(sigval_t));
494 		new_timer->sigq->info.si_value.sival_int = new_timer->it_id;
495 		new_timer->it_pid = get_pid(task_tgid(current));
496 	}
497 
498 	new_timer->sigq->info.si_tid   = new_timer->it_id;
499 	new_timer->sigq->info.si_code  = SI_TIMER;
500 
501 	if (copy_to_user(created_timer_id, &new_timer_id, sizeof (new_timer_id))) {
502 		error = -EFAULT;
503 		goto out;
504 	}
505 	/*
506 	 * After succesful copy out, the timer ID is visible to user space
507 	 * now but not yet valid because new_timer::signal is still NULL.
508 	 *
509 	 * Complete the initialization with the clock specific create
510 	 * callback.
511 	 */
512 	error = kc->timer_create(new_timer);
513 	if (error)
514 		goto out;
515 
516 	spin_lock_irq(&current->sighand->siglock);
517 	/* This makes the timer valid in the hash table */
518 	WRITE_ONCE(new_timer->it_signal, current->signal);
519 	hlist_add_head(&new_timer->list, &current->signal->posix_timers);
520 	spin_unlock_irq(&current->sighand->siglock);
521 	/*
522 	 * After unlocking sighand::siglock @new_timer is subject to
523 	 * concurrent removal and cannot be touched anymore
524 	 */
525 	return 0;
526 out:
527 	posix_timer_unhash_and_free(new_timer);
528 	return error;
529 }
530 
531 SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
532 		struct sigevent __user *, timer_event_spec,
533 		timer_t __user *, created_timer_id)
534 {
535 	if (timer_event_spec) {
536 		sigevent_t event;
537 
538 		if (copy_from_user(&event, timer_event_spec, sizeof (event)))
539 			return -EFAULT;
540 		return do_timer_create(which_clock, &event, created_timer_id);
541 	}
542 	return do_timer_create(which_clock, NULL, created_timer_id);
543 }
544 
545 #ifdef CONFIG_COMPAT
546 COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
547 		       struct compat_sigevent __user *, timer_event_spec,
548 		       timer_t __user *, created_timer_id)
549 {
550 	if (timer_event_spec) {
551 		sigevent_t event;
552 
553 		if (get_compat_sigevent(&event, timer_event_spec))
554 			return -EFAULT;
555 		return do_timer_create(which_clock, &event, created_timer_id);
556 	}
557 	return do_timer_create(which_clock, NULL, created_timer_id);
558 }
559 #endif
560 
561 static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
562 {
563 	struct k_itimer *timr;
564 
565 	/*
566 	 * timer_t could be any type >= int and we want to make sure any
567 	 * @timer_id outside positive int range fails lookup.
568 	 */
569 	if ((unsigned long long)timer_id > INT_MAX)
570 		return NULL;
571 
572 	/*
573 	 * The hash lookup and the timers are RCU protected.
574 	 *
575 	 * Timers are added to the hash in invalid state where
576 	 * timr::it_signal == NULL. timer::it_signal is only set after the
577 	 * rest of the initialization succeeded.
578 	 *
579 	 * Timer destruction happens in steps:
580 	 *  1) Set timr::it_signal to NULL with timr::it_lock held
581 	 *  2) Release timr::it_lock
582 	 *  3) Remove from the hash under hash_lock
583 	 *  4) Call RCU for removal after the grace period
584 	 *
585 	 * Holding rcu_read_lock() accross the lookup ensures that
586 	 * the timer cannot be freed.
587 	 *
588 	 * The lookup validates locklessly that timr::it_signal ==
589 	 * current::it_signal and timr::it_id == @timer_id. timr::it_id
590 	 * can't change, but timr::it_signal becomes NULL during
591 	 * destruction.
592 	 */
593 	rcu_read_lock();
594 	timr = posix_timer_by_id(timer_id);
595 	if (timr) {
596 		spin_lock_irqsave(&timr->it_lock, *flags);
597 		/*
598 		 * Validate under timr::it_lock that timr::it_signal is
599 		 * still valid. Pairs with #1 above.
600 		 */
601 		if (timr->it_signal == current->signal) {
602 			rcu_read_unlock();
603 			return timr;
604 		}
605 		spin_unlock_irqrestore(&timr->it_lock, *flags);
606 	}
607 	rcu_read_unlock();
608 
609 	return NULL;
610 }
611 
612 static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
613 {
614 	struct hrtimer *timer = &timr->it.real.timer;
615 
616 	return __hrtimer_expires_remaining_adjusted(timer, now);
617 }
618 
619 static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
620 {
621 	struct hrtimer *timer = &timr->it.real.timer;
622 
623 	return hrtimer_forward(timer, now, timr->it_interval);
624 }
625 
626 /*
627  * Get the time remaining on a POSIX.1b interval timer.
628  *
629  * Two issues to handle here:
630  *
631  *  1) The timer has a requeue pending. The return value must appear as
632  *     if the timer has been requeued right now.
633  *
634  *  2) The timer is a SIGEV_NONE timer. These timers are never enqueued
635  *     into the hrtimer queue and therefore never expired. Emulate expiry
636  *     here taking #1 into account.
637  */
638 void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
639 {
640 	const struct k_clock *kc = timr->kclock;
641 	ktime_t now, remaining, iv;
642 	bool sig_none;
643 
644 	sig_none = timr->it_sigev_notify == SIGEV_NONE;
645 	iv = timr->it_interval;
646 
647 	/* interval timer ? */
648 	if (iv) {
649 		cur_setting->it_interval = ktime_to_timespec64(iv);
650 	} else if (!timr->it_active) {
651 		/*
652 		 * SIGEV_NONE oneshot timers are never queued and therefore
653 		 * timr->it_active is always false. The check below
654 		 * vs. remaining time will handle this case.
655 		 *
656 		 * For all other timers there is nothing to update here, so
657 		 * return.
658 		 */
659 		if (!sig_none)
660 			return;
661 	}
662 
663 	now = kc->clock_get_ktime(timr->it_clock);
664 
665 	/*
666 	 * If this is an interval timer and either has requeue pending or
667 	 * is a SIGEV_NONE timer move the expiry time forward by intervals,
668 	 * so expiry is > now.
669 	 */
670 	if (iv && (timr->it_requeue_pending & REQUEUE_PENDING || sig_none))
671 		timr->it_overrun += kc->timer_forward(timr, now);
672 
673 	remaining = kc->timer_remaining(timr, now);
674 	/*
675 	 * As @now is retrieved before a possible timer_forward() and
676 	 * cannot be reevaluated by the compiler @remaining is based on the
677 	 * same @now value. Therefore @remaining is consistent vs. @now.
678 	 *
679 	 * Consequently all interval timers, i.e. @iv > 0, cannot have a
680 	 * remaining time <= 0 because timer_forward() guarantees to move
681 	 * them forward so that the next timer expiry is > @now.
682 	 */
683 	if (remaining <= 0) {
684 		/*
685 		 * A single shot SIGEV_NONE timer must return 0, when it is
686 		 * expired! Timers which have a real signal delivery mode
687 		 * must return a remaining time greater than 0 because the
688 		 * signal has not yet been delivered.
689 		 */
690 		if (!sig_none)
691 			cur_setting->it_value.tv_nsec = 1;
692 	} else {
693 		cur_setting->it_value = ktime_to_timespec64(remaining);
694 	}
695 }
696 
697 static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
698 {
699 	const struct k_clock *kc;
700 	struct k_itimer *timr;
701 	unsigned long flags;
702 	int ret = 0;
703 
704 	timr = lock_timer(timer_id, &flags);
705 	if (!timr)
706 		return -EINVAL;
707 
708 	memset(setting, 0, sizeof(*setting));
709 	kc = timr->kclock;
710 	if (WARN_ON_ONCE(!kc || !kc->timer_get))
711 		ret = -EINVAL;
712 	else
713 		kc->timer_get(timr, setting);
714 
715 	unlock_timer(timr, flags);
716 	return ret;
717 }
718 
719 /* Get the time remaining on a POSIX.1b interval timer. */
720 SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
721 		struct __kernel_itimerspec __user *, setting)
722 {
723 	struct itimerspec64 cur_setting;
724 
725 	int ret = do_timer_gettime(timer_id, &cur_setting);
726 	if (!ret) {
727 		if (put_itimerspec64(&cur_setting, setting))
728 			ret = -EFAULT;
729 	}
730 	return ret;
731 }
732 
733 #ifdef CONFIG_COMPAT_32BIT_TIME
734 
735 SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
736 		struct old_itimerspec32 __user *, setting)
737 {
738 	struct itimerspec64 cur_setting;
739 
740 	int ret = do_timer_gettime(timer_id, &cur_setting);
741 	if (!ret) {
742 		if (put_old_itimerspec32(&cur_setting, setting))
743 			ret = -EFAULT;
744 	}
745 	return ret;
746 }
747 
748 #endif
749 
750 /**
751  * sys_timer_getoverrun - Get the number of overruns of a POSIX.1b interval timer
752  * @timer_id:	The timer ID which identifies the timer
753  *
754  * The "overrun count" of a timer is one plus the number of expiration
755  * intervals which have elapsed between the first expiry, which queues the
756  * signal and the actual signal delivery. On signal delivery the "overrun
757  * count" is calculated and cached, so it can be returned directly here.
758  *
759  * As this is relative to the last queued signal the returned overrun count
760  * is meaningless outside of the signal delivery path and even there it
761  * does not accurately reflect the current state when user space evaluates
762  * it.
763  *
764  * Returns:
765  *	-EINVAL		@timer_id is invalid
766  *	1..INT_MAX	The number of overruns related to the last delivered signal
767  */
768 SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
769 {
770 	struct k_itimer *timr;
771 	unsigned long flags;
772 	int overrun;
773 
774 	timr = lock_timer(timer_id, &flags);
775 	if (!timr)
776 		return -EINVAL;
777 
778 	overrun = timer_overrun_to_int(timr, 0);
779 	unlock_timer(timr, flags);
780 
781 	return overrun;
782 }
783 
784 static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
785 			       bool absolute, bool sigev_none)
786 {
787 	struct hrtimer *timer = &timr->it.real.timer;
788 	enum hrtimer_mode mode;
789 
790 	mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
791 	/*
792 	 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
793 	 * clock modifications, so they become CLOCK_MONOTONIC based under the
794 	 * hood. See hrtimer_init(). Update timr->kclock, so the generic
795 	 * functions which use timr->kclock->clock_get_*() work.
796 	 *
797 	 * Note: it_clock stays unmodified, because the next timer_set() might
798 	 * use ABSTIME, so it needs to switch back.
799 	 */
800 	if (timr->it_clock == CLOCK_REALTIME)
801 		timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
802 
803 	hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
804 	timr->it.real.timer.function = posix_timer_fn;
805 
806 	if (!absolute)
807 		expires = ktime_add_safe(expires, timer->base->get_time());
808 	hrtimer_set_expires(timer, expires);
809 
810 	if (!sigev_none)
811 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
812 }
813 
814 static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
815 {
816 	return hrtimer_try_to_cancel(&timr->it.real.timer);
817 }
818 
819 static void common_timer_wait_running(struct k_itimer *timer)
820 {
821 	hrtimer_cancel_wait_running(&timer->it.real.timer);
822 }
823 
824 /*
825  * On PREEMPT_RT this prevents priority inversion and a potential livelock
826  * against the ksoftirqd thread in case that ksoftirqd gets preempted while
827  * executing a hrtimer callback.
828  *
829  * See the comments in hrtimer_cancel_wait_running(). For PREEMPT_RT=n this
830  * just results in a cpu_relax().
831  *
832  * For POSIX CPU timers with CONFIG_POSIX_CPU_TIMERS_TASK_WORK=n this is
833  * just a cpu_relax(). With CONFIG_POSIX_CPU_TIMERS_TASK_WORK=y this
834  * prevents spinning on an eventually scheduled out task and a livelock
835  * when the task which tries to delete or disarm the timer has preempted
836  * the task which runs the expiry in task work context.
837  */
838 static struct k_itimer *timer_wait_running(struct k_itimer *timer,
839 					   unsigned long *flags)
840 {
841 	const struct k_clock *kc = READ_ONCE(timer->kclock);
842 	timer_t timer_id = READ_ONCE(timer->it_id);
843 
844 	/* Prevent kfree(timer) after dropping the lock */
845 	rcu_read_lock();
846 	unlock_timer(timer, *flags);
847 
848 	/*
849 	 * kc->timer_wait_running() might drop RCU lock. So @timer
850 	 * cannot be touched anymore after the function returns!
851 	 */
852 	if (!WARN_ON_ONCE(!kc->timer_wait_running))
853 		kc->timer_wait_running(timer);
854 
855 	rcu_read_unlock();
856 	/* Relock the timer. It might be not longer hashed. */
857 	return lock_timer(timer_id, flags);
858 }
859 
860 /*
861  * Set up the new interval and reset the signal delivery data
862  */
863 void posix_timer_set_common(struct k_itimer *timer, struct itimerspec64 *new_setting)
864 {
865 	if (new_setting->it_value.tv_sec || new_setting->it_value.tv_nsec)
866 		timer->it_interval = timespec64_to_ktime(new_setting->it_interval);
867 	else
868 		timer->it_interval = 0;
869 
870 	/* Prevent reloading in case there is a signal pending */
871 	timer->it_requeue_pending = (timer->it_requeue_pending + 2) & ~REQUEUE_PENDING;
872 	/* Reset overrun accounting */
873 	timer->it_overrun_last = 0;
874 	timer->it_overrun = -1LL;
875 }
876 
877 /* Set a POSIX.1b interval timer. */
878 int common_timer_set(struct k_itimer *timr, int flags,
879 		     struct itimerspec64 *new_setting,
880 		     struct itimerspec64 *old_setting)
881 {
882 	const struct k_clock *kc = timr->kclock;
883 	bool sigev_none;
884 	ktime_t expires;
885 
886 	if (old_setting)
887 		common_timer_get(timr, old_setting);
888 
889 	/* Prevent rearming by clearing the interval */
890 	timr->it_interval = 0;
891 	/*
892 	 * Careful here. On SMP systems the timer expiry function could be
893 	 * active and spinning on timr->it_lock.
894 	 */
895 	if (kc->timer_try_to_cancel(timr) < 0)
896 		return TIMER_RETRY;
897 
898 	timr->it_active = 0;
899 	posix_timer_set_common(timr, new_setting);
900 
901 	/* Keep timer disarmed when it_value is zero */
902 	if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
903 		return 0;
904 
905 	expires = timespec64_to_ktime(new_setting->it_value);
906 	if (flags & TIMER_ABSTIME)
907 		expires = timens_ktime_to_host(timr->it_clock, expires);
908 	sigev_none = timr->it_sigev_notify == SIGEV_NONE;
909 
910 	kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
911 	timr->it_active = !sigev_none;
912 	return 0;
913 }
914 
915 static int do_timer_settime(timer_t timer_id, int tmr_flags,
916 			    struct itimerspec64 *new_spec64,
917 			    struct itimerspec64 *old_spec64)
918 {
919 	const struct k_clock *kc;
920 	struct k_itimer *timr;
921 	unsigned long flags;
922 	int error;
923 
924 	if (!timespec64_valid(&new_spec64->it_interval) ||
925 	    !timespec64_valid(&new_spec64->it_value))
926 		return -EINVAL;
927 
928 	if (old_spec64)
929 		memset(old_spec64, 0, sizeof(*old_spec64));
930 
931 	timr = lock_timer(timer_id, &flags);
932 retry:
933 	if (!timr)
934 		return -EINVAL;
935 
936 	if (old_spec64)
937 		old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
938 
939 	kc = timr->kclock;
940 	if (WARN_ON_ONCE(!kc || !kc->timer_set))
941 		error = -EINVAL;
942 	else
943 		error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
944 
945 	if (error == TIMER_RETRY) {
946 		// We already got the old time...
947 		old_spec64 = NULL;
948 		/* Unlocks and relocks the timer if it still exists */
949 		timr = timer_wait_running(timr, &flags);
950 		goto retry;
951 	}
952 	unlock_timer(timr, flags);
953 
954 	return error;
955 }
956 
957 /* Set a POSIX.1b interval timer */
958 SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
959 		const struct __kernel_itimerspec __user *, new_setting,
960 		struct __kernel_itimerspec __user *, old_setting)
961 {
962 	struct itimerspec64 new_spec, old_spec, *rtn;
963 	int error = 0;
964 
965 	if (!new_setting)
966 		return -EINVAL;
967 
968 	if (get_itimerspec64(&new_spec, new_setting))
969 		return -EFAULT;
970 
971 	rtn = old_setting ? &old_spec : NULL;
972 	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
973 	if (!error && old_setting) {
974 		if (put_itimerspec64(&old_spec, old_setting))
975 			error = -EFAULT;
976 	}
977 	return error;
978 }
979 
980 #ifdef CONFIG_COMPAT_32BIT_TIME
981 SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
982 		struct old_itimerspec32 __user *, new,
983 		struct old_itimerspec32 __user *, old)
984 {
985 	struct itimerspec64 new_spec, old_spec;
986 	struct itimerspec64 *rtn = old ? &old_spec : NULL;
987 	int error = 0;
988 
989 	if (!new)
990 		return -EINVAL;
991 	if (get_old_itimerspec32(&new_spec, new))
992 		return -EFAULT;
993 
994 	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
995 	if (!error && old) {
996 		if (put_old_itimerspec32(&old_spec, old))
997 			error = -EFAULT;
998 	}
999 	return error;
1000 }
1001 #endif
1002 
1003 int common_timer_del(struct k_itimer *timer)
1004 {
1005 	const struct k_clock *kc = timer->kclock;
1006 
1007 	timer->it_interval = 0;
1008 	if (kc->timer_try_to_cancel(timer) < 0)
1009 		return TIMER_RETRY;
1010 	timer->it_active = 0;
1011 	return 0;
1012 }
1013 
1014 static inline int timer_delete_hook(struct k_itimer *timer)
1015 {
1016 	const struct k_clock *kc = timer->kclock;
1017 
1018 	if (WARN_ON_ONCE(!kc || !kc->timer_del))
1019 		return -EINVAL;
1020 	return kc->timer_del(timer);
1021 }
1022 
1023 /* Delete a POSIX.1b interval timer. */
1024 SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
1025 {
1026 	struct k_itimer *timer;
1027 	unsigned long flags;
1028 
1029 	timer = lock_timer(timer_id, &flags);
1030 
1031 retry_delete:
1032 	if (!timer)
1033 		return -EINVAL;
1034 
1035 	if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
1036 		/* Unlocks and relocks the timer if it still exists */
1037 		timer = timer_wait_running(timer, &flags);
1038 		goto retry_delete;
1039 	}
1040 
1041 	spin_lock(&current->sighand->siglock);
1042 	hlist_del(&timer->list);
1043 	spin_unlock(&current->sighand->siglock);
1044 	/*
1045 	 * A concurrent lookup could check timer::it_signal lockless. It
1046 	 * will reevaluate with timer::it_lock held and observe the NULL.
1047 	 */
1048 	WRITE_ONCE(timer->it_signal, NULL);
1049 
1050 	unlock_timer(timer, flags);
1051 	posix_timer_unhash_and_free(timer);
1052 	return 0;
1053 }
1054 
1055 /*
1056  * Delete a timer if it is armed, remove it from the hash and schedule it
1057  * for RCU freeing.
1058  */
1059 static void itimer_delete(struct k_itimer *timer)
1060 {
1061 	unsigned long flags;
1062 
1063 	/*
1064 	 * irqsave is required to make timer_wait_running() work.
1065 	 */
1066 	spin_lock_irqsave(&timer->it_lock, flags);
1067 
1068 retry_delete:
1069 	/*
1070 	 * Even if the timer is not longer accessible from other tasks
1071 	 * it still might be armed and queued in the underlying timer
1072 	 * mechanism. Worse, that timer mechanism might run the expiry
1073 	 * function concurrently.
1074 	 */
1075 	if (timer_delete_hook(timer) == TIMER_RETRY) {
1076 		/*
1077 		 * Timer is expired concurrently, prevent livelocks
1078 		 * and pointless spinning on RT.
1079 		 *
1080 		 * timer_wait_running() drops timer::it_lock, which opens
1081 		 * the possibility for another task to delete the timer.
1082 		 *
1083 		 * That's not possible here because this is invoked from
1084 		 * do_exit() only for the last thread of the thread group.
1085 		 * So no other task can access and delete that timer.
1086 		 */
1087 		if (WARN_ON_ONCE(timer_wait_running(timer, &flags) != timer))
1088 			return;
1089 
1090 		goto retry_delete;
1091 	}
1092 	hlist_del(&timer->list);
1093 
1094 	/*
1095 	 * Setting timer::it_signal to NULL is technically not required
1096 	 * here as nothing can access the timer anymore legitimately via
1097 	 * the hash table. Set it to NULL nevertheless so that all deletion
1098 	 * paths are consistent.
1099 	 */
1100 	WRITE_ONCE(timer->it_signal, NULL);
1101 
1102 	spin_unlock_irqrestore(&timer->it_lock, flags);
1103 	posix_timer_unhash_and_free(timer);
1104 }
1105 
1106 /*
1107  * Invoked from do_exit() when the last thread of a thread group exits.
1108  * At that point no other task can access the timers of the dying
1109  * task anymore.
1110  */
1111 void exit_itimers(struct task_struct *tsk)
1112 {
1113 	struct hlist_head timers;
1114 
1115 	if (hlist_empty(&tsk->signal->posix_timers))
1116 		return;
1117 
1118 	/* Protect against concurrent read via /proc/$PID/timers */
1119 	spin_lock_irq(&tsk->sighand->siglock);
1120 	hlist_move_list(&tsk->signal->posix_timers, &timers);
1121 	spin_unlock_irq(&tsk->sighand->siglock);
1122 
1123 	/* The timers are not longer accessible via tsk::signal */
1124 	while (!hlist_empty(&timers))
1125 		itimer_delete(hlist_entry(timers.first, struct k_itimer, list));
1126 }
1127 
1128 SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1129 		const struct __kernel_timespec __user *, tp)
1130 {
1131 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1132 	struct timespec64 new_tp;
1133 
1134 	if (!kc || !kc->clock_set)
1135 		return -EINVAL;
1136 
1137 	if (get_timespec64(&new_tp, tp))
1138 		return -EFAULT;
1139 
1140 	/*
1141 	 * Permission checks have to be done inside the clock specific
1142 	 * setter callback.
1143 	 */
1144 	return kc->clock_set(which_clock, &new_tp);
1145 }
1146 
1147 SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1148 		struct __kernel_timespec __user *, tp)
1149 {
1150 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1151 	struct timespec64 kernel_tp;
1152 	int error;
1153 
1154 	if (!kc)
1155 		return -EINVAL;
1156 
1157 	error = kc->clock_get_timespec(which_clock, &kernel_tp);
1158 
1159 	if (!error && put_timespec64(&kernel_tp, tp))
1160 		error = -EFAULT;
1161 
1162 	return error;
1163 }
1164 
1165 int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1166 {
1167 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1168 
1169 	if (!kc)
1170 		return -EINVAL;
1171 	if (!kc->clock_adj)
1172 		return -EOPNOTSUPP;
1173 
1174 	return kc->clock_adj(which_clock, ktx);
1175 }
1176 
1177 SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1178 		struct __kernel_timex __user *, utx)
1179 {
1180 	struct __kernel_timex ktx;
1181 	int err;
1182 
1183 	if (copy_from_user(&ktx, utx, sizeof(ktx)))
1184 		return -EFAULT;
1185 
1186 	err = do_clock_adjtime(which_clock, &ktx);
1187 
1188 	if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1189 		return -EFAULT;
1190 
1191 	return err;
1192 }
1193 
1194 /**
1195  * sys_clock_getres - Get the resolution of a clock
1196  * @which_clock:	The clock to get the resolution for
1197  * @tp:			Pointer to a a user space timespec64 for storage
1198  *
1199  * POSIX defines:
1200  *
1201  * "The clock_getres() function shall return the resolution of any
1202  * clock. Clock resolutions are implementation-defined and cannot be set by
1203  * a process. If the argument res is not NULL, the resolution of the
1204  * specified clock shall be stored in the location pointed to by res. If
1205  * res is NULL, the clock resolution is not returned. If the time argument
1206  * of clock_settime() is not a multiple of res, then the value is truncated
1207  * to a multiple of res."
1208  *
1209  * Due to the various hardware constraints the real resolution can vary
1210  * wildly and even change during runtime when the underlying devices are
1211  * replaced. The kernel also can use hardware devices with different
1212  * resolutions for reading the time and for arming timers.
1213  *
1214  * The kernel therefore deviates from the POSIX spec in various aspects:
1215  *
1216  * 1) The resolution returned to user space
1217  *
1218  *    For CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, CLOCK_TAI,
1219  *    CLOCK_REALTIME_ALARM, CLOCK_BOOTTIME_ALAREM and CLOCK_MONOTONIC_RAW
1220  *    the kernel differentiates only two cases:
1221  *
1222  *    I)  Low resolution mode:
1223  *
1224  *	  When high resolution timers are disabled at compile or runtime
1225  *	  the resolution returned is nanoseconds per tick, which represents
1226  *	  the precision at which timers expire.
1227  *
1228  *    II) High resolution mode:
1229  *
1230  *	  When high resolution timers are enabled the resolution returned
1231  *	  is always one nanosecond independent of the actual resolution of
1232  *	  the underlying hardware devices.
1233  *
1234  *	  For CLOCK_*_ALARM the actual resolution depends on system
1235  *	  state. When system is running the resolution is the same as the
1236  *	  resolution of the other clocks. During suspend the actual
1237  *	  resolution is the resolution of the underlying RTC device which
1238  *	  might be way less precise than the clockevent device used during
1239  *	  running state.
1240  *
1241  *   For CLOCK_REALTIME_COARSE and CLOCK_MONOTONIC_COARSE the resolution
1242  *   returned is always nanoseconds per tick.
1243  *
1244  *   For CLOCK_PROCESS_CPUTIME and CLOCK_THREAD_CPUTIME the resolution
1245  *   returned is always one nanosecond under the assumption that the
1246  *   underlying scheduler clock has a better resolution than nanoseconds
1247  *   per tick.
1248  *
1249  *   For dynamic POSIX clocks (PTP devices) the resolution returned is
1250  *   always one nanosecond.
1251  *
1252  * 2) Affect on sys_clock_settime()
1253  *
1254  *    The kernel does not truncate the time which is handed in to
1255  *    sys_clock_settime(). The kernel internal timekeeping is always using
1256  *    nanoseconds precision independent of the clocksource device which is
1257  *    used to read the time from. The resolution of that device only
1258  *    affects the presicion of the time returned by sys_clock_gettime().
1259  *
1260  * Returns:
1261  *	0		Success. @tp contains the resolution
1262  *	-EINVAL		@which_clock is not a valid clock ID
1263  *	-EFAULT		Copying the resolution to @tp faulted
1264  *	-ENODEV		Dynamic POSIX clock is not backed by a device
1265  *	-EOPNOTSUPP	Dynamic POSIX clock does not support getres()
1266  */
1267 SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1268 		struct __kernel_timespec __user *, tp)
1269 {
1270 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1271 	struct timespec64 rtn_tp;
1272 	int error;
1273 
1274 	if (!kc)
1275 		return -EINVAL;
1276 
1277 	error = kc->clock_getres(which_clock, &rtn_tp);
1278 
1279 	if (!error && tp && put_timespec64(&rtn_tp, tp))
1280 		error = -EFAULT;
1281 
1282 	return error;
1283 }
1284 
1285 #ifdef CONFIG_COMPAT_32BIT_TIME
1286 
1287 SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1288 		struct old_timespec32 __user *, tp)
1289 {
1290 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1291 	struct timespec64 ts;
1292 
1293 	if (!kc || !kc->clock_set)
1294 		return -EINVAL;
1295 
1296 	if (get_old_timespec32(&ts, tp))
1297 		return -EFAULT;
1298 
1299 	return kc->clock_set(which_clock, &ts);
1300 }
1301 
1302 SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1303 		struct old_timespec32 __user *, tp)
1304 {
1305 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1306 	struct timespec64 ts;
1307 	int err;
1308 
1309 	if (!kc)
1310 		return -EINVAL;
1311 
1312 	err = kc->clock_get_timespec(which_clock, &ts);
1313 
1314 	if (!err && put_old_timespec32(&ts, tp))
1315 		err = -EFAULT;
1316 
1317 	return err;
1318 }
1319 
1320 SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1321 		struct old_timex32 __user *, utp)
1322 {
1323 	struct __kernel_timex ktx;
1324 	int err;
1325 
1326 	err = get_old_timex32(&ktx, utp);
1327 	if (err)
1328 		return err;
1329 
1330 	err = do_clock_adjtime(which_clock, &ktx);
1331 
1332 	if (err >= 0 && put_old_timex32(utp, &ktx))
1333 		return -EFAULT;
1334 
1335 	return err;
1336 }
1337 
1338 SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1339 		struct old_timespec32 __user *, tp)
1340 {
1341 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1342 	struct timespec64 ts;
1343 	int err;
1344 
1345 	if (!kc)
1346 		return -EINVAL;
1347 
1348 	err = kc->clock_getres(which_clock, &ts);
1349 	if (!err && tp && put_old_timespec32(&ts, tp))
1350 		return -EFAULT;
1351 
1352 	return err;
1353 }
1354 
1355 #endif
1356 
1357 /*
1358  * sys_clock_nanosleep() for CLOCK_REALTIME and CLOCK_TAI
1359  */
1360 static int common_nsleep(const clockid_t which_clock, int flags,
1361 			 const struct timespec64 *rqtp)
1362 {
1363 	ktime_t texp = timespec64_to_ktime(*rqtp);
1364 
1365 	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1366 				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1367 				 which_clock);
1368 }
1369 
1370 /*
1371  * sys_clock_nanosleep() for CLOCK_MONOTONIC and CLOCK_BOOTTIME
1372  *
1373  * Absolute nanosleeps for these clocks are time-namespace adjusted.
1374  */
1375 static int common_nsleep_timens(const clockid_t which_clock, int flags,
1376 				const struct timespec64 *rqtp)
1377 {
1378 	ktime_t texp = timespec64_to_ktime(*rqtp);
1379 
1380 	if (flags & TIMER_ABSTIME)
1381 		texp = timens_ktime_to_host(which_clock, texp);
1382 
1383 	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1384 				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1385 				 which_clock);
1386 }
1387 
1388 SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1389 		const struct __kernel_timespec __user *, rqtp,
1390 		struct __kernel_timespec __user *, rmtp)
1391 {
1392 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1393 	struct timespec64 t;
1394 
1395 	if (!kc)
1396 		return -EINVAL;
1397 	if (!kc->nsleep)
1398 		return -EOPNOTSUPP;
1399 
1400 	if (get_timespec64(&t, rqtp))
1401 		return -EFAULT;
1402 
1403 	if (!timespec64_valid(&t))
1404 		return -EINVAL;
1405 	if (flags & TIMER_ABSTIME)
1406 		rmtp = NULL;
1407 	current->restart_block.fn = do_no_restart_syscall;
1408 	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1409 	current->restart_block.nanosleep.rmtp = rmtp;
1410 
1411 	return kc->nsleep(which_clock, flags, &t);
1412 }
1413 
1414 #ifdef CONFIG_COMPAT_32BIT_TIME
1415 
1416 SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1417 		struct old_timespec32 __user *, rqtp,
1418 		struct old_timespec32 __user *, rmtp)
1419 {
1420 	const struct k_clock *kc = clockid_to_kclock(which_clock);
1421 	struct timespec64 t;
1422 
1423 	if (!kc)
1424 		return -EINVAL;
1425 	if (!kc->nsleep)
1426 		return -EOPNOTSUPP;
1427 
1428 	if (get_old_timespec32(&t, rqtp))
1429 		return -EFAULT;
1430 
1431 	if (!timespec64_valid(&t))
1432 		return -EINVAL;
1433 	if (flags & TIMER_ABSTIME)
1434 		rmtp = NULL;
1435 	current->restart_block.fn = do_no_restart_syscall;
1436 	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1437 	current->restart_block.nanosleep.compat_rmtp = rmtp;
1438 
1439 	return kc->nsleep(which_clock, flags, &t);
1440 }
1441 
1442 #endif
1443 
1444 static const struct k_clock clock_realtime = {
1445 	.clock_getres		= posix_get_hrtimer_res,
1446 	.clock_get_timespec	= posix_get_realtime_timespec,
1447 	.clock_get_ktime	= posix_get_realtime_ktime,
1448 	.clock_set		= posix_clock_realtime_set,
1449 	.clock_adj		= posix_clock_realtime_adj,
1450 	.nsleep			= common_nsleep,
1451 	.timer_create		= common_timer_create,
1452 	.timer_set		= common_timer_set,
1453 	.timer_get		= common_timer_get,
1454 	.timer_del		= common_timer_del,
1455 	.timer_rearm		= common_hrtimer_rearm,
1456 	.timer_forward		= common_hrtimer_forward,
1457 	.timer_remaining	= common_hrtimer_remaining,
1458 	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1459 	.timer_wait_running	= common_timer_wait_running,
1460 	.timer_arm		= common_hrtimer_arm,
1461 };
1462 
1463 static const struct k_clock clock_monotonic = {
1464 	.clock_getres		= posix_get_hrtimer_res,
1465 	.clock_get_timespec	= posix_get_monotonic_timespec,
1466 	.clock_get_ktime	= posix_get_monotonic_ktime,
1467 	.nsleep			= common_nsleep_timens,
1468 	.timer_create		= common_timer_create,
1469 	.timer_set		= common_timer_set,
1470 	.timer_get		= common_timer_get,
1471 	.timer_del		= common_timer_del,
1472 	.timer_rearm		= common_hrtimer_rearm,
1473 	.timer_forward		= common_hrtimer_forward,
1474 	.timer_remaining	= common_hrtimer_remaining,
1475 	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1476 	.timer_wait_running	= common_timer_wait_running,
1477 	.timer_arm		= common_hrtimer_arm,
1478 };
1479 
1480 static const struct k_clock clock_monotonic_raw = {
1481 	.clock_getres		= posix_get_hrtimer_res,
1482 	.clock_get_timespec	= posix_get_monotonic_raw,
1483 };
1484 
1485 static const struct k_clock clock_realtime_coarse = {
1486 	.clock_getres		= posix_get_coarse_res,
1487 	.clock_get_timespec	= posix_get_realtime_coarse,
1488 };
1489 
1490 static const struct k_clock clock_monotonic_coarse = {
1491 	.clock_getres		= posix_get_coarse_res,
1492 	.clock_get_timespec	= posix_get_monotonic_coarse,
1493 };
1494 
1495 static const struct k_clock clock_tai = {
1496 	.clock_getres		= posix_get_hrtimer_res,
1497 	.clock_get_ktime	= posix_get_tai_ktime,
1498 	.clock_get_timespec	= posix_get_tai_timespec,
1499 	.nsleep			= common_nsleep,
1500 	.timer_create		= common_timer_create,
1501 	.timer_set		= common_timer_set,
1502 	.timer_get		= common_timer_get,
1503 	.timer_del		= common_timer_del,
1504 	.timer_rearm		= common_hrtimer_rearm,
1505 	.timer_forward		= common_hrtimer_forward,
1506 	.timer_remaining	= common_hrtimer_remaining,
1507 	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1508 	.timer_wait_running	= common_timer_wait_running,
1509 	.timer_arm		= common_hrtimer_arm,
1510 };
1511 
1512 static const struct k_clock clock_boottime = {
1513 	.clock_getres		= posix_get_hrtimer_res,
1514 	.clock_get_ktime	= posix_get_boottime_ktime,
1515 	.clock_get_timespec	= posix_get_boottime_timespec,
1516 	.nsleep			= common_nsleep_timens,
1517 	.timer_create		= common_timer_create,
1518 	.timer_set		= common_timer_set,
1519 	.timer_get		= common_timer_get,
1520 	.timer_del		= common_timer_del,
1521 	.timer_rearm		= common_hrtimer_rearm,
1522 	.timer_forward		= common_hrtimer_forward,
1523 	.timer_remaining	= common_hrtimer_remaining,
1524 	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1525 	.timer_wait_running	= common_timer_wait_running,
1526 	.timer_arm		= common_hrtimer_arm,
1527 };
1528 
1529 static const struct k_clock * const posix_clocks[] = {
1530 	[CLOCK_REALTIME]		= &clock_realtime,
1531 	[CLOCK_MONOTONIC]		= &clock_monotonic,
1532 	[CLOCK_PROCESS_CPUTIME_ID]	= &clock_process,
1533 	[CLOCK_THREAD_CPUTIME_ID]	= &clock_thread,
1534 	[CLOCK_MONOTONIC_RAW]		= &clock_monotonic_raw,
1535 	[CLOCK_REALTIME_COARSE]		= &clock_realtime_coarse,
1536 	[CLOCK_MONOTONIC_COARSE]	= &clock_monotonic_coarse,
1537 	[CLOCK_BOOTTIME]		= &clock_boottime,
1538 	[CLOCK_REALTIME_ALARM]		= &alarm_clock,
1539 	[CLOCK_BOOTTIME_ALARM]		= &alarm_clock,
1540 	[CLOCK_TAI]			= &clock_tai,
1541 };
1542 
1543 static const struct k_clock *clockid_to_kclock(const clockid_t id)
1544 {
1545 	clockid_t idx = id;
1546 
1547 	if (id < 0) {
1548 		return (id & CLOCKFD_MASK) == CLOCKFD ?
1549 			&clock_posix_dynamic : &clock_posix_cpu;
1550 	}
1551 
1552 	if (id >= ARRAY_SIZE(posix_clocks))
1553 		return NULL;
1554 
1555 	return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1556 }
1557