projects/linux·blob master · refs

1582 lines · 43.8 KiB

1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * 2002-10-15 Posix Clocks & timers
4 * by George Anzinger [email protected]
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/compat.h>
13#include <linux/compiler.h>
14#include <linux/init.h>
15#include <linux/jhash.h>
16#include <linux/interrupt.h>
17#include <linux/list.h>
18#include <linux/memblock.h>
19#include <linux/nospec.h>
20#include <linux/posix-clock.h>
21#include <linux/posix-timers.h>
22#include <linux/prctl.h>
23#include <linux/sched/task.h>
24#include <linux/slab.h>
25#include <linux/syscalls.h>
26#include <linux/time.h>
27#include <linux/time_namespace.h>
28#include <linux/uaccess.h>
29
30#include "timekeeping.h"
31#include "posix-timers.h"
32
33/*
34 * Timers are managed in a hash table for lockless lookup. The hash key is
35 * constructed from current::signal and the timer ID and the timer is
36 * matched against current::signal and the timer ID when walking the hash
37 * bucket list.
38 *
39 * This allows checkpoint/restore to reconstruct the exact timer IDs for
40 * a process.
41 */
42struct timer_hash_bucket {
43 spinlock_t lock;
44 struct hlist_head head;
45};
46
47static struct {
48 struct timer_hash_bucket *buckets;
49 unsigned long mask;
50 struct kmem_cache *cache;
51} __timer_data __ro_after_init __aligned(4*sizeof(long));
52
53#define timer_buckets (__timer_data.buckets)
54#define timer_hashmask (__timer_data.mask)
55#define posix_timers_cache (__timer_data.cache)
56
57static const struct k_clock * const posix_clocks[];
58static const struct k_clock *clockid_to_kclock(const clockid_t id);
59static const struct k_clock clock_realtime, clock_monotonic;
60
61#define TIMER_ANY_ID INT_MIN
62
63/* SIGEV_THREAD_ID cannot share a bit with the other SIGEV values. */
64#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
65 ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
66#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
67#endif
68
69static struct k_itimer *lock_timer(timer_t timer_id);
70static inline void unlock_timer(struct k_itimer *timr)
71{
72 if (likely((timr)))
73 spin_unlock_irq(&timr->it_lock);
74}
75
76#define scoped_timer_get_or_fail(_id) \
77 scoped_cond_guard(lock_timer, return -EINVAL, _id)
78
79#define scoped_timer (scope)
80
81DEFINE_CLASS(lock_timer, struct k_itimer *, unlock_timer(_T), lock_timer(id), timer_t id);
82DEFINE_CLASS_IS_COND_GUARD(lock_timer);
83
84static struct timer_hash_bucket *hash_bucket(struct signal_struct *sig, unsigned int nr)
85{
86 return &timer_buckets[jhash2((u32 *)&sig, sizeof(sig) / sizeof(u32), nr) & timer_hashmask];
87}
88
89static struct k_itimer *posix_timer_by_id(timer_t id)
90{
91 struct signal_struct *sig = current->signal;
92 struct timer_hash_bucket *bucket = hash_bucket(sig, id);
93 struct k_itimer *timer;
94
95 hlist_for_each_entry_rcu(timer, &bucket->head, t_hash) {
96 /* timer->it_signal can be set concurrently */
97 if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id))
98 return timer;
99 }
100 return NULL;
101}
102
103static inline struct signal_struct *posix_sig_owner(const struct k_itimer *timer)
104{
105 unsigned long val = (unsigned long)timer->it_signal;
106
107 /*
108 * Mask out bit 0, which acts as invalid marker to prevent
109 * posix_timer_by_id() detecting it as valid.
110 */
111 return (struct signal_struct *)(val & ~1UL);
112}
113
114static bool posix_timer_hashed(struct timer_hash_bucket *bucket, struct signal_struct *sig,
115 timer_t id)
116{
117 struct hlist_head *head = &bucket->head;
118 struct k_itimer *timer;
119
120 hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&bucket->lock)) {
121 if ((posix_sig_owner(timer) == sig) && (timer->it_id == id))
122 return true;
123 }
124 return false;
125}
126
127static bool posix_timer_add_at(struct k_itimer *timer, struct signal_struct *sig, unsigned int id)
128{
129 struct timer_hash_bucket *bucket = hash_bucket(sig, id);
130
131 scoped_guard (spinlock, &bucket->lock) {
132 /*
133 * Validate under the lock as this could have raced against
134 * another thread ending up with the same ID, which is
135 * highly unlikely, but possible.
136 */
137 if (!posix_timer_hashed(bucket, sig, id)) {
138 /*
139 * Set the timer ID and the signal pointer to make
140 * it identifiable in the hash table. The signal
141 * pointer has bit 0 set to indicate that it is not
142 * yet fully initialized. posix_timer_hashed()
143 * masks this bit out, but the syscall lookup fails
144 * to match due to it being set. This guarantees
145 * that there can't be duplicate timer IDs handed
146 * out.
147 */
148 timer->it_id = (timer_t)id;
149 timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
150 hlist_add_head_rcu(&timer->t_hash, &bucket->head);
151 return true;
152 }
153 }
154 return false;
155}
156
157static int posix_timer_add(struct k_itimer *timer, int req_id)
158{
159 struct signal_struct *sig = current->signal;
160
161 if (unlikely(req_id != TIMER_ANY_ID)) {
162 if (!posix_timer_add_at(timer, sig, req_id))
163 return -EBUSY;
164
165 /*
166 * Move the ID counter past the requested ID, so that after
167 * switching back to normal mode the IDs are outside of the
168 * exact allocated region. That avoids ID collisions on the
169 * next regular timer_create() invocations.
170 */
171 atomic_set(&sig->next_posix_timer_id, req_id + 1);
172 return req_id;
173 }
174
175 for (unsigned int cnt = 0; cnt <= INT_MAX; cnt++) {
176 /* Get the next timer ID and clamp it to positive space */
177 unsigned int id = atomic_fetch_inc(&sig->next_posix_timer_id) & INT_MAX;
178
179 if (posix_timer_add_at(timer, sig, id))
180 return id;
181 cond_resched();
182 }
183 /* POSIX return code when no timer ID could be allocated */
184 return -EAGAIN;
185}
186
187static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp)
188{
189 ktime_get_real_ts64(tp);
190 return 0;
191}
192
193static ktime_t posix_get_realtime_ktime(clockid_t which_clock)
194{
195 return ktime_get_real();
196}
197
198static int posix_clock_realtime_set(const clockid_t which_clock,
199 const struct timespec64 *tp)
200{
201 return do_sys_settimeofday64(tp, NULL);
202}
203
204static int posix_clock_realtime_adj(const clockid_t which_clock,
205 struct __kernel_timex *t)
206{
207 return do_adjtimex(t);
208}
209
210static int posix_get_monotonic_timespec(clockid_t which_clock, struct timespec64 *tp)
211{
212 ktime_get_ts64(tp);
213 timens_add_monotonic(tp);
214 return 0;
215}
216
217static ktime_t posix_get_monotonic_ktime(clockid_t which_clock)
218{
219 return ktime_get();
220}
221
222static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
223{
224 ktime_get_raw_ts64(tp);
225 timens_add_monotonic(tp);
226 return 0;
227}
228
229static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
230{
231 ktime_get_coarse_real_ts64(tp);
232 return 0;
233}
234
235static int posix_get_monotonic_coarse(clockid_t which_clock,
236 struct timespec64 *tp)
237{
238 ktime_get_coarse_ts64(tp);
239 timens_add_monotonic(tp);
240 return 0;
241}
242
243static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
244{
245 *tp = ktime_to_timespec64(KTIME_LOW_RES);
246 return 0;
247}
248
249static int posix_get_boottime_timespec(const clockid_t which_clock, struct timespec64 *tp)
250{
251 ktime_get_boottime_ts64(tp);
252 timens_add_boottime(tp);
253 return 0;
254}
255
256static ktime_t posix_get_boottime_ktime(const clockid_t which_clock)
257{
258 return ktime_get_boottime();
259}
260
261static int posix_get_tai_timespec(clockid_t which_clock, struct timespec64 *tp)
262{
263 ktime_get_clocktai_ts64(tp);
264 return 0;
265}
266
267static ktime_t posix_get_tai_ktime(clockid_t which_clock)
268{
269 return ktime_get_clocktai();
270}
271
272static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
273{
274 tp->tv_sec = 0;
275 tp->tv_nsec = hrtimer_resolution;
276 return 0;
277}
278
279/*
280 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
281 * are of type int. Clamp the overrun value to INT_MAX
282 */
283static inline int timer_overrun_to_int(struct k_itimer *timr)
284{
285 if (timr->it_overrun_last > (s64)INT_MAX)
286 return INT_MAX;
287
288 return (int)timr->it_overrun_last;
289}
290
291static bool common_hrtimer_rearm(struct k_itimer *timr)
292{
293 struct hrtimer *timer = &timr->it.real.timer;
294
295 timr->it_overrun += hrtimer_forward_now(timer, timr->it_interval);
296 return hrtimer_start_expires_user(timer, HRTIMER_MODE_ABS);
297}
298
299static bool __posixtimer_deliver_signal(struct kernel_siginfo *info, struct k_itimer *timr)
300{
301 bool queued;
302
303 guard(spinlock)(&timr->it_lock);
304
305 /*
306 * Check if the timer is still alive or whether it got modified
307 * since the signal was queued. In either case, don't rearm and
308 * drop the signal.
309 */
310 if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!posixtimer_valid(timr)))
311 return false;
312
313 if (!timr->it_interval || WARN_ON_ONCE(timr->it_status != POSIX_TIMER_REQUEUE_PENDING))
314 return true;
315
316 /* timer_rearm() updates timr::it_overrun */
317 queued = timr->kclock->timer_rearm(timr);
318
319 timr->it_overrun_last = timr->it_overrun;
320 timr->it_overrun = -1LL;
321 ++timr->it_signal_seq;
322 info->si_overrun = timer_overrun_to_int(timr);
323
324 if (queued)
325 timr->it_status = POSIX_TIMER_ARMED;
326 else
327 posix_timer_queue_signal(timr);
328 return true;
329}
330
331/*
332 * This function is called from the signal delivery code. It decides
333 * whether the signal should be dropped and rearms interval timers. The
334 * timer can be unconditionally accessed as there is a reference held on
335 * it.
336 */
337bool posixtimer_deliver_signal(struct kernel_siginfo *info, struct sigqueue *timer_sigq)
338{
339 struct k_itimer *timr = container_of(timer_sigq, struct k_itimer, sigq);
340 bool ret;
341
342 /*
343 * Release siglock to ensure proper locking order versus
344 * timr::it_lock. Keep interrupts disabled.
345 */
346 spin_unlock(&current->sighand->siglock);
347
348 ret = __posixtimer_deliver_signal(info, timr);
349
350 /* Drop the reference which was acquired when the signal was queued */
351 posixtimer_putref(timr);
352
353 spin_lock(&current->sighand->siglock);
354 return ret;
355}
356
357void posix_timer_queue_signal(struct k_itimer *timr)
358{
359 lockdep_assert_held(&timr->it_lock);
360
361 if (!posixtimer_valid(timr))
362 return;
363
364 timr->it_status = timr->it_interval ? POSIX_TIMER_REQUEUE_PENDING : POSIX_TIMER_DISARMED;
365 posixtimer_send_sigqueue(timr);
366}
367
368/*
369 * This function gets called when a POSIX.1b interval timer expires from
370 * the HRTIMER interrupt (soft interrupt on RT kernels).
371 *
372 * Handles CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME and CLOCK_TAI
373 * based timers.
374 */
375static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
376{
377 struct k_itimer *timr = container_of(timer, struct k_itimer, it.real.timer);
378
379 guard(spinlock_irqsave)(&timr->it_lock);
380 posix_timer_queue_signal(timr);
381 return HRTIMER_NORESTART;
382}
383
384long posixtimer_create_prctl(unsigned long ctrl)
385{
386 switch (ctrl) {
387 case PR_TIMER_CREATE_RESTORE_IDS_OFF:
388 current->signal->timer_create_restore_ids = 0;
389 return 0;
390 case PR_TIMER_CREATE_RESTORE_IDS_ON:
391 current->signal->timer_create_restore_ids = 1;
392 return 0;
393 case PR_TIMER_CREATE_RESTORE_IDS_GET:
394 return current->signal->timer_create_restore_ids;
395 }
396 return -EINVAL;
397}
398
399static struct pid *good_sigevent(sigevent_t * event)
400{
401 struct pid *pid = task_tgid(current);
402 struct task_struct *rtn;
403
404 switch (event->sigev_notify) {
405 case SIGEV_SIGNAL | SIGEV_THREAD_ID:
406 pid = find_vpid(event->sigev_notify_thread_id);
407 rtn = pid_task(pid, PIDTYPE_PID);
408 if (!rtn || !same_thread_group(rtn, current))
409 return NULL;
410 fallthrough;
411 case SIGEV_SIGNAL:
412 case SIGEV_THREAD:
413 if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
414 return NULL;
415 fallthrough;
416 case SIGEV_NONE:
417 return pid;
418 default:
419 return NULL;
420 }
421}
422
423static struct k_itimer *alloc_posix_timer(void)
424{
425 struct k_itimer *tmr;
426
427 if (unlikely(!posix_timers_cache))
428 return NULL;
429
430 tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
431 if (!tmr)
432 return tmr;
433
434 if (unlikely(!posixtimer_init_sigqueue(&tmr->sigq))) {
435 kmem_cache_free(posix_timers_cache, tmr);
436 return NULL;
437 }
438 rcuref_init(&tmr->rcuref, 1);
439 return tmr;
440}
441
442void posixtimer_free_timer(struct k_itimer *tmr)
443{
444 put_pid(tmr->it_pid);
445 if (tmr->sigq.ucounts)
446 dec_rlimit_put_ucounts(tmr->sigq.ucounts, UCOUNT_RLIMIT_SIGPENDING);
447 kfree_rcu(tmr, rcu);
448}
449
450static void posix_timer_unhash_and_free(struct k_itimer *tmr)
451{
452 struct timer_hash_bucket *bucket = hash_bucket(posix_sig_owner(tmr), tmr->it_id);
453
454 scoped_guard (spinlock, &bucket->lock)
455 hlist_del_rcu(&tmr->t_hash);
456 posixtimer_putref(tmr);
457}
458
459static int common_timer_create(struct k_itimer *new_timer)
460{
461 hrtimer_setup(&new_timer->it.real.timer, posix_timer_fn, new_timer->it_clock, 0);
462 return 0;
463}
464
465/* Create a POSIX.1b interval timer. */
466static int do_timer_create(clockid_t which_clock, struct sigevent *event,
467 timer_t __user *created_timer_id)
468{
469 const struct k_clock *kc = clockid_to_kclock(which_clock);
470 timer_t req_id = TIMER_ANY_ID;
471 struct k_itimer *new_timer;
472 int error, new_timer_id;
473
474 if (!kc)
475 return -EINVAL;
476 if (!kc->timer_create)
477 return -EOPNOTSUPP;
478
479 /* Special case for CRIU to restore timers with a given timer ID. */
480 if (unlikely(current->signal->timer_create_restore_ids)) {
481 if (copy_from_user(&req_id, created_timer_id, sizeof(req_id)))
482 return -EFAULT;
483 /* Valid IDs are 0..INT_MAX */
484 if ((unsigned int)req_id > INT_MAX)
485 return -EINVAL;
486 }
487
488 new_timer = alloc_posix_timer();
489 if (unlikely(!new_timer))
490 return -EAGAIN;
491
492 spin_lock_init(&new_timer->it_lock);
493
494 /*
495 * Add the timer to the hash table. The timer is not yet valid
496 * after insertion, but has a unique ID allocated.
497 */
498 new_timer_id = posix_timer_add(new_timer, req_id);
499 if (new_timer_id < 0) {
500 posixtimer_free_timer(new_timer);
501 return new_timer_id;
502 }
503
504 new_timer->it_clock = which_clock;
505 new_timer->kclock = kc;
506 new_timer->it_overrun = -1LL;
507
508 if (event) {
509 scoped_guard (rcu)
510 new_timer->it_pid = get_pid(good_sigevent(event));
511 if (!new_timer->it_pid) {
512 error = -EINVAL;
513 goto out;
514 }
515 new_timer->it_sigev_notify = event->sigev_notify;
516 new_timer->sigq.info.si_signo = event->sigev_signo;
517 new_timer->sigq.info.si_value = event->sigev_value;
518 } else {
519 new_timer->it_sigev_notify = SIGEV_SIGNAL;
520 new_timer->sigq.info.si_signo = SIGALRM;
521 new_timer->sigq.info.si_value.sival_int = new_timer->it_id;
522 new_timer->it_pid = get_pid(task_tgid(current));
523 }
524
525 if (new_timer->it_sigev_notify & SIGEV_THREAD_ID)
526 new_timer->it_pid_type = PIDTYPE_PID;
527 else
528 new_timer->it_pid_type = PIDTYPE_TGID;
529
530 new_timer->sigq.info.si_tid = new_timer->it_id;
531 new_timer->sigq.info.si_code = SI_TIMER;
532
533 if (copy_to_user(created_timer_id, &new_timer_id, sizeof (new_timer_id))) {
534 error = -EFAULT;
535 goto out;
536 }
537 /*
538 * After successful copy out, the timer ID is visible to user space
539 * now but not yet valid because new_timer::signal low order bit is 1.
540 *
541 * Complete the initialization with the clock specific create
542 * callback.
543 */
544 error = kc->timer_create(new_timer);
545 if (error)
546 goto out;
547
548 /*
549 * timer::it_lock ensures that __lock_timer() observes a fully
550 * initialized timer when it observes a valid timer::it_signal.
551 *
552 * sighand::siglock is required to protect signal::posix_timers.
553 */
554 scoped_guard (spinlock_irq, &new_timer->it_lock) {
555 guard(spinlock)(&current->sighand->siglock);
556 /*
557 * new_timer::it_signal contains the signal pointer with
558 * bit 0 set, which makes it invalid for syscall operations.
559 * Store the unmodified signal pointer to make it valid.
560 */
561 WRITE_ONCE(new_timer->it_signal, current->signal);
562 hlist_add_head_rcu(&new_timer->list, &current->signal->posix_timers);
563 }
564 /*
565 * After unlocking @new_timer is subject to concurrent removal and
566 * cannot be touched anymore
567 */
568 return 0;
569out:
570 posix_timer_unhash_and_free(new_timer);
571 return error;
572}
573
574SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
575 struct sigevent __user *, timer_event_spec,
576 timer_t __user *, created_timer_id)
577{
578 if (timer_event_spec) {
579 sigevent_t event;
580
581 if (copy_from_user(&event, timer_event_spec, sizeof (event)))
582 return -EFAULT;
583 return do_timer_create(which_clock, &event, created_timer_id);
584 }
585 return do_timer_create(which_clock, NULL, created_timer_id);
586}
587
588#ifdef CONFIG_COMPAT
589COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
590 struct compat_sigevent __user *, timer_event_spec,
591 timer_t __user *, created_timer_id)
592{
593 if (timer_event_spec) {
594 sigevent_t event;
595
596 if (get_compat_sigevent(&event, timer_event_spec))
597 return -EFAULT;
598 return do_timer_create(which_clock, &event, created_timer_id);
599 }
600 return do_timer_create(which_clock, NULL, created_timer_id);
601}
602#endif
603
604static struct k_itimer *lock_timer(timer_t timer_id)
605{
606 struct k_itimer *timr;
607
608 /*
609 * timer_t could be any type >= int and we want to make sure any
610 * @timer_id outside positive int range fails lookup.
611 */
612 if ((unsigned long long)timer_id > INT_MAX)
613 return NULL;
614
615 /*
616 * The hash lookup and the timers are RCU protected.
617 *
618 * Timers are added to the hash in invalid state where
619 * timr::it_signal is marked invalid. timer::it_signal is only set
620 * after the rest of the initialization succeeded.
621 *
622 * Timer destruction happens in steps:
623 * 1) Set timr::it_signal marked invalid with timr::it_lock held
624 * 2) Release timr::it_lock
625 * 3) Remove from the hash under hash_lock
626 * 4) Put the reference count.
627 *
628 * The reference count might not drop to zero if timr::sigq is
629 * queued. In that case the signal delivery or flush will put the
630 * last reference count.
631 *
632 * When the reference count reaches zero, the timer is scheduled
633 * for RCU removal after the grace period.
634 *
635 * Holding rcu_read_lock() across the lookup ensures that
636 * the timer cannot be freed.
637 *
638 * The lookup validates locklessly that timr::it_signal ==
639 * current::it_signal and timr::it_id == @timer_id. timr::it_id
640 * can't change, but timr::it_signal can become invalid during
641 * destruction, which makes the locked check fail.
642 */
643 guard(rcu)();
644 timr = posix_timer_by_id(timer_id);
645 if (timr) {
646 spin_lock_irq(&timr->it_lock);
647 /*
648 * Validate under timr::it_lock that timr::it_signal is
649 * still valid. Pairs with #1 above.
650 */
651 if (timr->it_signal == current->signal)
652 return timr;
653 spin_unlock_irq(&timr->it_lock);
654 }
655 return NULL;
656}
657
658static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
659{
660 struct hrtimer *timer = &timr->it.real.timer;
661
662 return __hrtimer_expires_remaining_adjusted(timer, now);
663}
664
665static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
666{
667 struct hrtimer *timer = &timr->it.real.timer;
668
669 return hrtimer_forward(timer, now, timr->it_interval);
670}
671
672/*
673 * Get the time remaining on a POSIX.1b interval timer.
674 *
675 * Two issues to handle here:
676 *
677 * 1) The timer has a requeue pending. The return value must appear as
678 * if the timer has been requeued right now.
679 *
680 * 2) The timer is a SIGEV_NONE timer. These timers are never enqueued
681 * into the hrtimer queue and therefore never expired. Emulate expiry
682 * here taking #1 into account.
683 */
684void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
685{
686 const struct k_clock *kc = timr->kclock;
687 ktime_t now, remaining, iv;
688 bool sig_none;
689
690 sig_none = timr->it_sigev_notify == SIGEV_NONE;
691 iv = timr->it_interval;
692
693 /* interval timer ? */
694 if (iv) {
695 cur_setting->it_interval = ktime_to_timespec64(iv);
696 } else if (timr->it_status == POSIX_TIMER_DISARMED) {
697 /*
698 * SIGEV_NONE oneshot timers are never queued and therefore
699 * timr->it_status is always DISARMED. The check below
700 * vs. remaining time will handle this case.
701 *
702 * For all other timers there is nothing to update here, so
703 * return.
704 */
705 if (!sig_none)
706 return;
707 }
708
709 now = kc->clock_get_ktime(timr->it_clock);
710
711 /*
712 * If this is an interval timer and either has requeue pending or
713 * is a SIGEV_NONE timer move the expiry time forward by intervals,
714 * so expiry is > now.
715 */
716 if (iv && timr->it_status != POSIX_TIMER_ARMED)
717 timr->it_overrun += kc->timer_forward(timr, now);
718
719 remaining = kc->timer_remaining(timr, now);
720 /*
721 * As @now is retrieved before a possible timer_forward() and
722 * cannot be reevaluated by the compiler @remaining is based on the
723 * same @now value. Therefore @remaining is consistent vs. @now.
724 *
725 * Consequently all interval timers, i.e. @iv > 0, cannot have a
726 * remaining time <= 0 because timer_forward() guarantees to move
727 * them forward so that the next timer expiry is > @now.
728 */
729 if (remaining <= 0) {
730 /*
731 * A single shot SIGEV_NONE timer must return 0, when it is
732 * expired! Timers which have a real signal delivery mode
733 * must return a remaining time greater than 0 because the
734 * signal has not yet been delivered.
735 */
736 if (!sig_none)
737 cur_setting->it_value.tv_nsec = 1;
738 } else {
739 cur_setting->it_value = ktime_to_timespec64(remaining);
740 }
741}
742
743static int do_timer_gettime(timer_t timer_id, struct itimerspec64 *setting)
744{
745 memset(setting, 0, sizeof(*setting));
746 scoped_timer_get_or_fail(timer_id)
747 scoped_timer->kclock->timer_get(scoped_timer, setting);
748 return 0;
749}
750
751/* Get the time remaining on a POSIX.1b interval timer. */
752SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
753 struct __kernel_itimerspec __user *, setting)
754{
755 struct itimerspec64 cur_setting;
756
757 int ret = do_timer_gettime(timer_id, &cur_setting);
758 if (!ret) {
759 if (put_itimerspec64(&cur_setting, setting))
760 ret = -EFAULT;
761 }
762 return ret;
763}
764
765#ifdef CONFIG_COMPAT_32BIT_TIME
766
767SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
768 struct old_itimerspec32 __user *, setting)
769{
770 struct itimerspec64 cur_setting;
771
772 int ret = do_timer_gettime(timer_id, &cur_setting);
773 if (!ret) {
774 if (put_old_itimerspec32(&cur_setting, setting))
775 ret = -EFAULT;
776 }
777 return ret;
778}
779
780#endif
781
782/**
783 * sys_timer_getoverrun - Get the number of overruns of a POSIX.1b interval timer
784 * @timer_id: The timer ID which identifies the timer
785 *
786 * The "overrun count" of a timer is one plus the number of expiration
787 * intervals which have elapsed between the first expiry, which queues the
788 * signal and the actual signal delivery. On signal delivery the "overrun
789 * count" is calculated and cached, so it can be returned directly here.
790 *
791 * As this is relative to the last queued signal the returned overrun count
792 * is meaningless outside of the signal delivery path and even there it
793 * does not accurately reflect the current state when user space evaluates
794 * it.
795 *
796 * Returns:
797 * -EINVAL @timer_id is invalid
798 * 1..INT_MAX The number of overruns related to the last delivered signal
799 */
800SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
801{
802 scoped_timer_get_or_fail(timer_id)
803 return timer_overrun_to_int(scoped_timer);
804}
805
806static bool common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
807 bool absolute, bool sigev_none)
808{
809 struct hrtimer *timer = &timr->it.real.timer;
810 enum hrtimer_mode mode;
811
812 mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
813 /*
814 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
815 * clock modifications, so they become CLOCK_MONOTONIC based under the
816 * hood. See hrtimer_setup(). Update timr->kclock, so the generic
817 * functions which use timr->kclock->clock_get_*() work.
818 *
819 * Note: it_clock stays unmodified, because the next timer_set() might
820 * use ABSTIME, so it needs to switch back.
821 */
822 if (timr->it_clock == CLOCK_REALTIME)
823 timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
824
825 hrtimer_setup(&timr->it.real.timer, posix_timer_fn, timr->it_clock, mode);
826
827 if (!absolute)
828 expires = ktime_add_safe(expires, hrtimer_cb_get_time(timer));
829 hrtimer_set_expires(timer, expires);
830
831 /* For sigev_none pretend that the timer is queued */
832 if (sigev_none)
833 return true;
834
835 return hrtimer_start_expires_user(timer, HRTIMER_MODE_ABS);
836}
837
838static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
839{
840 return hrtimer_try_to_cancel(&timr->it.real.timer);
841}
842
843static void common_timer_wait_running(struct k_itimer *timer)
844{
845 hrtimer_cancel_wait_running(&timer->it.real.timer);
846}
847
848/*
849 * On PREEMPT_RT this prevents priority inversion and a potential livelock
850 * against the ksoftirqd thread in case that ksoftirqd gets preempted while
851 * executing a hrtimer callback.
852 *
853 * See the comments in hrtimer_cancel_wait_running(). For PREEMPT_RT=n this
854 * just results in a cpu_relax().
855 *
856 * For POSIX CPU timers with CONFIG_POSIX_CPU_TIMERS_TASK_WORK=n this is
857 * just a cpu_relax(). With CONFIG_POSIX_CPU_TIMERS_TASK_WORK=y this
858 * prevents spinning on an eventually scheduled out task and a livelock
859 * when the task which tries to delete or disarm the timer has preempted
860 * the task which runs the expiry in task work context.
861 */
862static void timer_wait_running(struct k_itimer *timer)
863{
864 /*
865 * kc->timer_wait_running() might drop RCU lock. So @timer
866 * cannot be touched anymore after the function returns!
867 */
868 timer->kclock->timer_wait_running(timer);
869}
870
871/*
872 * Set up the new interval and reset the signal delivery data
873 */
874void posix_timer_set_common(struct k_itimer *timer, struct itimerspec64 *new_setting)
875{
876 if (new_setting->it_value.tv_sec || new_setting->it_value.tv_nsec)
877 timer->it_interval = timespec64_to_ktime(new_setting->it_interval);
878 else
879 timer->it_interval = 0;
880
881 /* Reset overrun accounting */
882 timer->it_overrun_last = 0;
883 timer->it_overrun = -1LL;
884}
885
886/* Set a POSIX.1b interval timer. */
887int common_timer_set(struct k_itimer *timr, int flags,
888 struct itimerspec64 *new_setting,
889 struct itimerspec64 *old_setting)
890{
891 const struct k_clock *kc = timr->kclock;
892 bool sigev_none;
893 ktime_t expires;
894
895 if (old_setting)
896 common_timer_get(timr, old_setting);
897
898 /*
899 * Careful here. On SMP systems the timer expiry function could be
900 * active and spinning on timr->it_lock.
901 */
902 if (kc->timer_try_to_cancel(timr) < 0)
903 return TIMER_RETRY;
904
905 timr->it_status = POSIX_TIMER_DISARMED;
906 posix_timer_set_common(timr, new_setting);
907
908 /* Keep timer disarmed when it_value is zero */
909 if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
910 return 0;
911
912 expires = timespec64_to_ktime(new_setting->it_value);
913 if (flags & TIMER_ABSTIME)
914 expires = timens_ktime_to_host(timr->it_clock, expires);
915 sigev_none = timr->it_sigev_notify == SIGEV_NONE;
916
917 if (kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none)) {
918 if (!sigev_none)
919 timr->it_status = POSIX_TIMER_ARMED;
920 } else {
921 /* Timer was already expired, queue the signal */
922 posix_timer_queue_signal(timr);
923 }
924 return 0;
925}
926
927static int do_timer_settime(timer_t timer_id, int tmr_flags, struct itimerspec64 *new_spec64,
928 struct itimerspec64 *old_spec64)
929{
930 if (!timespec64_valid(&new_spec64->it_interval) ||
931 !timespec64_valid(&new_spec64->it_value))
932 return -EINVAL;
933
934 if (old_spec64)
935 memset(old_spec64, 0, sizeof(*old_spec64));
936
937 for (; ; old_spec64 = NULL) {
938 struct k_itimer *timr;
939
940 scoped_timer_get_or_fail(timer_id) {
941 timr = scoped_timer;
942
943 if (old_spec64)
944 old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
945
946 /* Prevent signal delivery and rearming. */
947 timr->it_signal_seq++;
948
949 int ret = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
950 if (ret != TIMER_RETRY)
951 return ret;
952
953 /* Protect the timer from being freed when leaving the lock scope */
954 rcu_read_lock();
955 }
956 timer_wait_running(timr);
957 rcu_read_unlock();
958 }
959}
960
961/* Set a POSIX.1b interval timer */
962SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
963 const struct __kernel_itimerspec __user *, new_setting,
964 struct __kernel_itimerspec __user *, old_setting)
965{
966 struct itimerspec64 new_spec, old_spec, *rtn;
967 int error = 0;
968
969 if (!new_setting)
970 return -EINVAL;
971
972 if (get_itimerspec64(&new_spec, new_setting))
973 return -EFAULT;
974
975 rtn = old_setting ? &old_spec : NULL;
976 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
977 if (!error && old_setting) {
978 if (put_itimerspec64(&old_spec, old_setting))
979 error = -EFAULT;
980 }
981 return error;
982}
983
984#ifdef CONFIG_COMPAT_32BIT_TIME
985SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
986 struct old_itimerspec32 __user *, new,
987 struct old_itimerspec32 __user *, old)
988{
989 struct itimerspec64 new_spec, old_spec;
990 struct itimerspec64 *rtn = old ? &old_spec : NULL;
991 int error = 0;
992
993 if (!new)
994 return -EINVAL;
995 if (get_old_itimerspec32(&new_spec, new))
996 return -EFAULT;
997
998 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
999 if (!error && old) {
1000 if (put_old_itimerspec32(&old_spec, old))
1001 error = -EFAULT;
1002 }
1003 return error;
1004}
1005#endif
1006
1007int common_timer_del(struct k_itimer *timer)
1008{
1009 const struct k_clock *kc = timer->kclock;
1010
1011 if (kc->timer_try_to_cancel(timer) < 0)
1012 return TIMER_RETRY;
1013 timer->it_status = POSIX_TIMER_DISARMED;
1014 return 0;
1015}
1016
1017/*
1018 * If the deleted timer is on the ignored list, remove it and
1019 * drop the associated reference.
1020 */
1021static inline void posix_timer_cleanup_ignored(struct k_itimer *tmr)
1022{
1023 if (!hlist_unhashed(&tmr->ignored_list)) {
1024 hlist_del_init(&tmr->ignored_list);
1025 posixtimer_putref(tmr);
1026 }
1027}
1028
1029static void posix_timer_delete(struct k_itimer *timer)
1030{
1031 /*
1032 * Invalidate the timer, remove it from the linked list and remove
1033 * it from the ignored list if pending.
1034 *
1035 * The invalidation must be written with siglock held so that the
1036 * signal code observes the invalidated timer::it_signal in
1037 * do_sigaction(), which prevents it from moving a pending signal
1038 * of a deleted timer to the ignore list.
1039 *
1040 * The invalidation also prevents signal queueing, signal delivery
1041 * and therefore rearming from the signal delivery path.
1042 *
1043 * A concurrent lookup can still find the timer in the hash, but it
1044 * will check timer::it_signal with timer::it_lock held and observe
1045 * bit 0 set, which invalidates it. That also prevents the timer ID
1046 * from being handed out before this timer is completely gone.
1047 */
1048 timer->it_signal_seq++;
1049
1050 scoped_guard (spinlock, &current->sighand->siglock) {
1051 unsigned long sig = (unsigned long)timer->it_signal | 1UL;
1052
1053 WRITE_ONCE(timer->it_signal, (struct signal_struct *)sig);
1054 hlist_del_rcu(&timer->list);
1055 posix_timer_cleanup_ignored(timer);
1056 }
1057
1058 while (timer->kclock->timer_del(timer) == TIMER_RETRY) {
1059 guard(rcu)();
1060 spin_unlock_irq(&timer->it_lock);
1061 timer_wait_running(timer);
1062 spin_lock_irq(&timer->it_lock);
1063 }
1064}
1065
1066/* Delete a POSIX.1b interval timer. */
1067SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
1068{
1069 struct k_itimer *timer;
1070
1071 scoped_timer_get_or_fail(timer_id) {
1072 timer = scoped_timer;
1073 posix_timer_delete(timer);
1074 }
1075 /* Remove it from the hash, which frees up the timer ID */
1076 posix_timer_unhash_and_free(timer);
1077 return 0;
1078}
1079
1080/*
1081 * Invoked from do_exit() when the last thread of a thread group exits.
1082 * At that point no other task can access the timers of the dying
1083 * task anymore.
1084 */
1085void exit_itimers(struct task_struct *tsk)
1086{
1087 struct hlist_head timers;
1088 struct hlist_node *next;
1089 struct k_itimer *timer;
1090
1091 /* Clear restore mode for exec() */
1092 tsk->signal->timer_create_restore_ids = 0;
1093
1094 if (hlist_empty(&tsk->signal->posix_timers))
1095 return;
1096
1097 /* Protect against concurrent read via /proc/$PID/timers */
1098 scoped_guard (spinlock_irq, &tsk->sighand->siglock)
1099 hlist_move_list(&tsk->signal->posix_timers, &timers);
1100
1101 /* The timers are not longer accessible via tsk::signal */
1102 hlist_for_each_entry_safe(timer, next, &timers, list) {
1103 scoped_guard (spinlock_irq, &timer->it_lock)
1104 posix_timer_delete(timer);
1105 posix_timer_unhash_and_free(timer);
1106 cond_resched();
1107 }
1108
1109 /*
1110 * There should be no timers on the ignored list. posix_timer_delete() has
1111 * mopped them up.
1112 */
1113 if (!WARN_ON_ONCE(!hlist_empty(&tsk->signal->ignored_posix_timers)))
1114 return;
1115
1116 hlist_move_list(&tsk->signal->ignored_posix_timers, &timers);
1117 while (!hlist_empty(&timers)) {
1118 posix_timer_cleanup_ignored(hlist_entry(timers.first, struct k_itimer,
1119 ignored_list));
1120 }
1121}
1122
1123SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1124 const struct __kernel_timespec __user *, tp)
1125{
1126 const struct k_clock *kc = clockid_to_kclock(which_clock);
1127 struct timespec64 new_tp;
1128
1129 if (!kc || !kc->clock_set)
1130 return -EINVAL;
1131
1132 if (get_timespec64(&new_tp, tp))
1133 return -EFAULT;
1134
1135 /*
1136 * Permission checks have to be done inside the clock specific
1137 * setter callback.
1138 */
1139 return kc->clock_set(which_clock, &new_tp);
1140}
1141
1142SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1143 struct __kernel_timespec __user *, tp)
1144{
1145 const struct k_clock *kc = clockid_to_kclock(which_clock);
1146 struct timespec64 kernel_tp;
1147 int error;
1148
1149 if (!kc)
1150 return -EINVAL;
1151
1152 error = kc->clock_get_timespec(which_clock, &kernel_tp);
1153
1154 if (!error && put_timespec64(&kernel_tp, tp))
1155 error = -EFAULT;
1156
1157 return error;
1158}
1159
1160int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1161{
1162 const struct k_clock *kc = clockid_to_kclock(which_clock);
1163
1164 if (!kc)
1165 return -EINVAL;
1166 if (!kc->clock_adj)
1167 return -EOPNOTSUPP;
1168
1169 return kc->clock_adj(which_clock, ktx);
1170}
1171
1172SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1173 struct __kernel_timex __user *, utx)
1174{
1175 struct __kernel_timex ktx;
1176 int err;
1177
1178 if (copy_from_user(&ktx, utx, sizeof(ktx)))
1179 return -EFAULT;
1180
1181 err = do_clock_adjtime(which_clock, &ktx);
1182
1183 if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1184 return -EFAULT;
1185
1186 return err;
1187}
1188
1189/**
1190 * sys_clock_getres - Get the resolution of a clock
1191 * @which_clock: The clock to get the resolution for
1192 * @tp: Pointer to a a user space timespec64 for storage
1193 *
1194 * POSIX defines:
1195 *
1196 * "The clock_getres() function shall return the resolution of any
1197 * clock. Clock resolutions are implementation-defined and cannot be set by
1198 * a process. If the argument res is not NULL, the resolution of the
1199 * specified clock shall be stored in the location pointed to by res. If
1200 * res is NULL, the clock resolution is not returned. If the time argument
1201 * of clock_settime() is not a multiple of res, then the value is truncated
1202 * to a multiple of res."
1203 *
1204 * Due to the various hardware constraints the real resolution can vary
1205 * wildly and even change during runtime when the underlying devices are
1206 * replaced. The kernel also can use hardware devices with different
1207 * resolutions for reading the time and for arming timers.
1208 *
1209 * The kernel therefore deviates from the POSIX spec in various aspects:
1210 *
1211 * 1) The resolution returned to user space
1212 *
1213 * For CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, CLOCK_TAI,
1214 * CLOCK_REALTIME_ALARM, CLOCK_BOOTTIME_ALAREM and CLOCK_MONOTONIC_RAW
1215 * the kernel differentiates only two cases:
1216 *
1217 * I) Low resolution mode:
1218 *
1219 * When high resolution timers are disabled at compile or runtime
1220 * the resolution returned is nanoseconds per tick, which represents
1221 * the precision at which timers expire.
1222 *
1223 * II) High resolution mode:
1224 *
1225 * When high resolution timers are enabled the resolution returned
1226 * is always one nanosecond independent of the actual resolution of
1227 * the underlying hardware devices.
1228 *
1229 * For CLOCK_*_ALARM the actual resolution depends on system
1230 * state. When system is running the resolution is the same as the
1231 * resolution of the other clocks. During suspend the actual
1232 * resolution is the resolution of the underlying RTC device which
1233 * might be way less precise than the clockevent device used during
1234 * running state.
1235 *
1236 * For CLOCK_REALTIME_COARSE and CLOCK_MONOTONIC_COARSE the resolution
1237 * returned is always nanoseconds per tick.
1238 *
1239 * For CLOCK_PROCESS_CPUTIME and CLOCK_THREAD_CPUTIME the resolution
1240 * returned is always one nanosecond under the assumption that the
1241 * underlying scheduler clock has a better resolution than nanoseconds
1242 * per tick.
1243 *
1244 * For dynamic POSIX clocks (PTP devices) the resolution returned is
1245 * always one nanosecond.
1246 *
1247 * 2) Affect on sys_clock_settime()
1248 *
1249 * The kernel does not truncate the time which is handed in to
1250 * sys_clock_settime(). The kernel internal timekeeping is always using
1251 * nanoseconds precision independent of the clocksource device which is
1252 * used to read the time from. The resolution of that device only
1253 * affects the precision of the time returned by sys_clock_gettime().
1254 *
1255 * Returns:
1256 * 0 Success. @tp contains the resolution
1257 * -EINVAL @which_clock is not a valid clock ID
1258 * -EFAULT Copying the resolution to @tp faulted
1259 * -ENODEV Dynamic POSIX clock is not backed by a device
1260 * -EOPNOTSUPP Dynamic POSIX clock does not support getres()
1261 */
1262SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1263 struct __kernel_timespec __user *, tp)
1264{
1265 const struct k_clock *kc = clockid_to_kclock(which_clock);
1266 struct timespec64 rtn_tp;
1267 int error;
1268
1269 if (!kc)
1270 return -EINVAL;
1271
1272 error = kc->clock_getres(which_clock, &rtn_tp);
1273
1274 if (!error && tp && put_timespec64(&rtn_tp, tp))
1275 error = -EFAULT;
1276
1277 return error;
1278}
1279
1280#ifdef CONFIG_COMPAT_32BIT_TIME
1281
1282SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1283 struct old_timespec32 __user *, tp)
1284{
1285 const struct k_clock *kc = clockid_to_kclock(which_clock);
1286 struct timespec64 ts;
1287
1288 if (!kc || !kc->clock_set)
1289 return -EINVAL;
1290
1291 if (get_old_timespec32(&ts, tp))
1292 return -EFAULT;
1293
1294 return kc->clock_set(which_clock, &ts);
1295}
1296
1297SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1298 struct old_timespec32 __user *, tp)
1299{
1300 const struct k_clock *kc = clockid_to_kclock(which_clock);
1301 struct timespec64 ts;
1302 int err;
1303
1304 if (!kc)
1305 return -EINVAL;
1306
1307 err = kc->clock_get_timespec(which_clock, &ts);
1308
1309 if (!err && put_old_timespec32(&ts, tp))
1310 err = -EFAULT;
1311
1312 return err;
1313}
1314
1315SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1316 struct old_timex32 __user *, utp)
1317{
1318 struct __kernel_timex ktx;
1319 int err;
1320
1321 err = get_old_timex32(&ktx, utp);
1322 if (err)
1323 return err;
1324
1325 err = do_clock_adjtime(which_clock, &ktx);
1326
1327 if (err >= 0 && put_old_timex32(utp, &ktx))
1328 return -EFAULT;
1329
1330 return err;
1331}
1332
1333SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1334 struct old_timespec32 __user *, tp)
1335{
1336 const struct k_clock *kc = clockid_to_kclock(which_clock);
1337 struct timespec64 ts;
1338 int err;
1339
1340 if (!kc)
1341 return -EINVAL;
1342
1343 err = kc->clock_getres(which_clock, &ts);
1344 if (!err && tp && put_old_timespec32(&ts, tp))
1345 return -EFAULT;
1346
1347 return err;
1348}
1349
1350#endif
1351
1352/*
1353 * sys_clock_nanosleep() for CLOCK_REALTIME and CLOCK_TAI
1354 */
1355static int common_nsleep(const clockid_t which_clock, int flags,
1356 const struct timespec64 *rqtp)
1357{
1358 ktime_t texp = timespec64_to_ktime(*rqtp);
1359
1360 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1361 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1362 which_clock);
1363}
1364
1365/*
1366 * sys_clock_nanosleep() for CLOCK_MONOTONIC and CLOCK_BOOTTIME
1367 *
1368 * Absolute nanosleeps for these clocks are time-namespace adjusted.
1369 */
1370static int common_nsleep_timens(const clockid_t which_clock, int flags,
1371 const struct timespec64 *rqtp)
1372{
1373 ktime_t texp = timespec64_to_ktime(*rqtp);
1374
1375 if (flags & TIMER_ABSTIME)
1376 texp = timens_ktime_to_host(which_clock, texp);
1377
1378 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1379 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1380 which_clock);
1381}
1382
1383SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1384 const struct __kernel_timespec __user *, rqtp,
1385 struct __kernel_timespec __user *, rmtp)
1386{
1387 const struct k_clock *kc = clockid_to_kclock(which_clock);
1388 struct timespec64 t;
1389
1390 if (!kc)
1391 return -EINVAL;
1392 if (!kc->nsleep)
1393 return -EOPNOTSUPP;
1394
1395 if (get_timespec64(&t, rqtp))
1396 return -EFAULT;
1397
1398 if (!timespec64_valid(&t))
1399 return -EINVAL;
1400 if (flags & TIMER_ABSTIME)
1401 rmtp = NULL;
1402 current->restart_block.fn = do_no_restart_syscall;
1403 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1404 current->restart_block.nanosleep.rmtp = rmtp;
1405
1406 return kc->nsleep(which_clock, flags, &t);
1407}
1408
1409#ifdef CONFIG_COMPAT_32BIT_TIME
1410
1411SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1412 struct old_timespec32 __user *, rqtp,
1413 struct old_timespec32 __user *, rmtp)
1414{
1415 const struct k_clock *kc = clockid_to_kclock(which_clock);
1416 struct timespec64 t;
1417
1418 if (!kc)
1419 return -EINVAL;
1420 if (!kc->nsleep)
1421 return -EOPNOTSUPP;
1422
1423 if (get_old_timespec32(&t, rqtp))
1424 return -EFAULT;
1425
1426 if (!timespec64_valid(&t))
1427 return -EINVAL;
1428 if (flags & TIMER_ABSTIME)
1429 rmtp = NULL;
1430 current->restart_block.fn = do_no_restart_syscall;
1431 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1432 current->restart_block.nanosleep.compat_rmtp = rmtp;
1433
1434 return kc->nsleep(which_clock, flags, &t);
1435}
1436
1437#endif
1438
1439static const struct k_clock clock_realtime = {
1440 .clock_getres = posix_get_hrtimer_res,
1441 .clock_get_timespec = posix_get_realtime_timespec,
1442 .clock_get_ktime = posix_get_realtime_ktime,
1443 .clock_set = posix_clock_realtime_set,
1444 .clock_adj = posix_clock_realtime_adj,
1445 .nsleep = common_nsleep,
1446 .timer_create = common_timer_create,
1447 .timer_set = common_timer_set,
1448 .timer_get = common_timer_get,
1449 .timer_del = common_timer_del,
1450 .timer_rearm = common_hrtimer_rearm,
1451 .timer_forward = common_hrtimer_forward,
1452 .timer_remaining = common_hrtimer_remaining,
1453 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1454 .timer_wait_running = common_timer_wait_running,
1455 .timer_arm = common_hrtimer_arm,
1456};
1457
1458static const struct k_clock clock_monotonic = {
1459 .clock_getres = posix_get_hrtimer_res,
1460 .clock_get_timespec = posix_get_monotonic_timespec,
1461 .clock_get_ktime = posix_get_monotonic_ktime,
1462 .nsleep = common_nsleep_timens,
1463 .timer_create = common_timer_create,
1464 .timer_set = common_timer_set,
1465 .timer_get = common_timer_get,
1466 .timer_del = common_timer_del,
1467 .timer_rearm = common_hrtimer_rearm,
1468 .timer_forward = common_hrtimer_forward,
1469 .timer_remaining = common_hrtimer_remaining,
1470 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1471 .timer_wait_running = common_timer_wait_running,
1472 .timer_arm = common_hrtimer_arm,
1473};
1474
1475static const struct k_clock clock_monotonic_raw = {
1476 .clock_getres = posix_get_hrtimer_res,
1477 .clock_get_timespec = posix_get_monotonic_raw,
1478};
1479
1480static const struct k_clock clock_realtime_coarse = {
1481 .clock_getres = posix_get_coarse_res,
1482 .clock_get_timespec = posix_get_realtime_coarse,
1483};
1484
1485static const struct k_clock clock_monotonic_coarse = {
1486 .clock_getres = posix_get_coarse_res,
1487 .clock_get_timespec = posix_get_monotonic_coarse,
1488};
1489
1490static const struct k_clock clock_tai = {
1491 .clock_getres = posix_get_hrtimer_res,
1492 .clock_get_ktime = posix_get_tai_ktime,
1493 .clock_get_timespec = posix_get_tai_timespec,
1494 .nsleep = common_nsleep,
1495 .timer_create = common_timer_create,
1496 .timer_set = common_timer_set,
1497 .timer_get = common_timer_get,
1498 .timer_del = common_timer_del,
1499 .timer_rearm = common_hrtimer_rearm,
1500 .timer_forward = common_hrtimer_forward,
1501 .timer_remaining = common_hrtimer_remaining,
1502 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1503 .timer_wait_running = common_timer_wait_running,
1504 .timer_arm = common_hrtimer_arm,
1505};
1506
1507static const struct k_clock clock_boottime = {
1508 .clock_getres = posix_get_hrtimer_res,
1509 .clock_get_ktime = posix_get_boottime_ktime,
1510 .clock_get_timespec = posix_get_boottime_timespec,
1511 .nsleep = common_nsleep_timens,
1512 .timer_create = common_timer_create,
1513 .timer_set = common_timer_set,
1514 .timer_get = common_timer_get,
1515 .timer_del = common_timer_del,
1516 .timer_rearm = common_hrtimer_rearm,
1517 .timer_forward = common_hrtimer_forward,
1518 .timer_remaining = common_hrtimer_remaining,
1519 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1520 .timer_wait_running = common_timer_wait_running,
1521 .timer_arm = common_hrtimer_arm,
1522};
1523
1524static const struct k_clock * const posix_clocks[] = {
1525 [CLOCK_REALTIME] = &clock_realtime,
1526 [CLOCK_MONOTONIC] = &clock_monotonic,
1527 [CLOCK_PROCESS_CPUTIME_ID] = &clock_process,
1528 [CLOCK_THREAD_CPUTIME_ID] = &clock_thread,
1529 [CLOCK_MONOTONIC_RAW] = &clock_monotonic_raw,
1530 [CLOCK_REALTIME_COARSE] = &clock_realtime_coarse,
1531 [CLOCK_MONOTONIC_COARSE] = &clock_monotonic_coarse,
1532 [CLOCK_BOOTTIME] = &clock_boottime,
1533 [CLOCK_REALTIME_ALARM] = &alarm_clock,
1534 [CLOCK_BOOTTIME_ALARM] = &alarm_clock,
1535 [CLOCK_TAI] = &clock_tai,
1536#ifdef CONFIG_POSIX_AUX_CLOCKS
1537 [CLOCK_AUX ... CLOCK_AUX_LAST] = &clock_aux,
1538#endif
1539};
1540
1541static const struct k_clock *clockid_to_kclock(const clockid_t id)
1542{
1543 clockid_t idx = id;
1544
1545 if (id < 0) {
1546 return (id & CLOCKFD_MASK) == CLOCKFD ?
1547 &clock_posix_dynamic : &clock_posix_cpu;
1548 }
1549
1550 if (id >= ARRAY_SIZE(posix_clocks))
1551 return NULL;
1552
1553 return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1554}
1555
1556static int __init posixtimer_init(void)
1557{
1558 unsigned long i, size;
1559 unsigned int shift;
1560
1561 posix_timers_cache = kmem_cache_create("posix_timers_cache",
1562 sizeof(struct k_itimer),
1563 __alignof__(struct k_itimer),
1564 SLAB_ACCOUNT, NULL);
1565
1566 if (IS_ENABLED(CONFIG_BASE_SMALL))
1567 size = 512;
1568 else
1569 size = roundup_pow_of_two(512 * num_possible_cpus());
1570
1571 timer_buckets = alloc_large_system_hash("posixtimers", sizeof(*timer_buckets),
1572 size, 0, 0, &shift, NULL, size, size);
1573 size = 1UL << shift;
1574 timer_hashmask = size - 1;
1575
1576 for (i = 0; i < size; i++) {
1577 spin_lock_init(&timer_buckets[i].lock);
1578 INIT_HLIST_HEAD(&timer_buckets[i].head);
1579 }
1580 return 0;
1581}
1582core_initcall(posixtimer_init);