Loading...
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/*
39 * Management arrays for POSIX timers. Timers are now kept in static hash table
40 * with 512 entries.
41 * Timer ids are allocated by local routine, which selects proper hash head by
42 * key, constructed from current->signal address and per signal struct counter.
43 * This keeps timer ids unique per process, but now they can intersect between
44 * processes.
45 */
46
47/*
48 * Lets keep our timers in a slab cache :-)
49 */
50static struct kmem_cache *posix_timers_cache;
51
52static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
53static DEFINE_SPINLOCK(hash_lock);
54
55static const struct k_clock * const posix_clocks[];
56static const struct k_clock *clockid_to_kclock(const clockid_t id);
57static const struct k_clock clock_realtime, clock_monotonic;
58
59/*
60 * we assume that the new SIGEV_THREAD_ID shares no bits with the other
61 * SIGEV values. Here we put out an error if this assumption fails.
62 */
63#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
64 ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
65#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
66#endif
67
68/*
69 * The timer ID is turned into a timer address by idr_find().
70 * Verifying a valid ID consists of:
71 *
72 * a) checking that idr_find() returns other than -1.
73 * b) checking that the timer id matches the one in the timer itself.
74 * c) that the timer owner is in the callers thread group.
75 */
76
77/*
78 * CLOCKs: The POSIX standard calls for a couple of clocks and allows us
79 * to implement others. This structure defines the various
80 * clocks.
81 *
82 * RESOLUTION: Clock resolution is used to round up timer and interval
83 * times, NOT to report clock times, which are reported with as
84 * much resolution as the system can muster. In some cases this
85 * resolution may depend on the underlying clock hardware and
86 * may not be quantifiable until run time, and only then is the
87 * necessary code is written. The standard says we should say
88 * something about this issue in the documentation...
89 *
90 * FUNCTIONS: The CLOCKs structure defines possible functions to
91 * handle various clock functions.
92 *
93 * The standard POSIX timer management code assumes the
94 * following: 1.) The k_itimer struct (sched.h) is used for
95 * the timer. 2.) The list, it_lock, it_clock, it_id and
96 * it_pid fields are not modified by timer code.
97 *
98 * Permissions: It is assumed that the clock_settime() function defined
99 * for each clock will take care of permission checks. Some
100 * clocks may be set able by any user (i.e. local process
101 * clocks) others not. Currently the only set able clock we
102 * have is CLOCK_REALTIME and its high res counter part, both of
103 * which we beg off on and pass to do_sys_settimeofday().
104 */
105static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
106
107#define lock_timer(tid, flags) \
108({ struct k_itimer *__timr; \
109 __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags)); \
110 __timr; \
111})
112
113static int hash(struct signal_struct *sig, unsigned int nr)
114{
115 return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
116}
117
118static struct k_itimer *__posix_timers_find(struct hlist_head *head,
119 struct signal_struct *sig,
120 timer_t id)
121{
122 struct k_itimer *timer;
123
124 hlist_for_each_entry_rcu(timer, head, t_hash,
125 lockdep_is_held(&hash_lock)) {
126 if ((timer->it_signal == sig) && (timer->it_id == id))
127 return timer;
128 }
129 return NULL;
130}
131
132static struct k_itimer *posix_timer_by_id(timer_t id)
133{
134 struct signal_struct *sig = current->signal;
135 struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
136
137 return __posix_timers_find(head, sig, id);
138}
139
140static int posix_timer_add(struct k_itimer *timer)
141{
142 struct signal_struct *sig = current->signal;
143 int first_free_id = sig->posix_timer_id;
144 struct hlist_head *head;
145 int ret = -ENOENT;
146
147 do {
148 spin_lock(&hash_lock);
149 head = &posix_timers_hashtable[hash(sig, sig->posix_timer_id)];
150 if (!__posix_timers_find(head, sig, sig->posix_timer_id)) {
151 hlist_add_head_rcu(&timer->t_hash, head);
152 ret = sig->posix_timer_id;
153 }
154 if (++sig->posix_timer_id < 0)
155 sig->posix_timer_id = 0;
156 if ((sig->posix_timer_id == first_free_id) && (ret == -ENOENT))
157 /* Loop over all possible ids completed */
158 ret = -EAGAIN;
159 spin_unlock(&hash_lock);
160 } while (ret == -ENOENT);
161 return ret;
162}
163
164static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
165{
166 spin_unlock_irqrestore(&timr->it_lock, flags);
167}
168
169/* Get clock_realtime */
170static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp)
171{
172 ktime_get_real_ts64(tp);
173 return 0;
174}
175
176static ktime_t posix_get_realtime_ktime(clockid_t which_clock)
177{
178 return ktime_get_real();
179}
180
181/* Set clock_realtime */
182static int posix_clock_realtime_set(const clockid_t which_clock,
183 const struct timespec64 *tp)
184{
185 return do_sys_settimeofday64(tp, NULL);
186}
187
188static int posix_clock_realtime_adj(const clockid_t which_clock,
189 struct __kernel_timex *t)
190{
191 return do_adjtimex(t);
192}
193
194/*
195 * Get monotonic time for posix timers
196 */
197static int posix_get_monotonic_timespec(clockid_t which_clock, struct timespec64 *tp)
198{
199 ktime_get_ts64(tp);
200 timens_add_monotonic(tp);
201 return 0;
202}
203
204static ktime_t posix_get_monotonic_ktime(clockid_t which_clock)
205{
206 return ktime_get();
207}
208
209/*
210 * Get monotonic-raw time for posix timers
211 */
212static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
213{
214 ktime_get_raw_ts64(tp);
215 timens_add_monotonic(tp);
216 return 0;
217}
218
219
220static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
221{
222 ktime_get_coarse_real_ts64(tp);
223 return 0;
224}
225
226static int posix_get_monotonic_coarse(clockid_t which_clock,
227 struct timespec64 *tp)
228{
229 ktime_get_coarse_ts64(tp);
230 timens_add_monotonic(tp);
231 return 0;
232}
233
234static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
235{
236 *tp = ktime_to_timespec64(KTIME_LOW_RES);
237 return 0;
238}
239
240static int posix_get_boottime_timespec(const clockid_t which_clock, struct timespec64 *tp)
241{
242 ktime_get_boottime_ts64(tp);
243 timens_add_boottime(tp);
244 return 0;
245}
246
247static ktime_t posix_get_boottime_ktime(const clockid_t which_clock)
248{
249 return ktime_get_boottime();
250}
251
252static int posix_get_tai_timespec(clockid_t which_clock, struct timespec64 *tp)
253{
254 ktime_get_clocktai_ts64(tp);
255 return 0;
256}
257
258static ktime_t posix_get_tai_ktime(clockid_t which_clock)
259{
260 return ktime_get_clocktai();
261}
262
263static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
264{
265 tp->tv_sec = 0;
266 tp->tv_nsec = hrtimer_resolution;
267 return 0;
268}
269
270/*
271 * Initialize everything, well, just everything in Posix clocks/timers ;)
272 */
273static __init int init_posix_timers(void)
274{
275 posix_timers_cache = kmem_cache_create("posix_timers_cache",
276 sizeof(struct k_itimer), 0,
277 SLAB_PANIC | SLAB_ACCOUNT, NULL);
278 return 0;
279}
280__initcall(init_posix_timers);
281
282/*
283 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
284 * are of type int. Clamp the overrun value to INT_MAX
285 */
286static inline int timer_overrun_to_int(struct k_itimer *timr, int baseval)
287{
288 s64 sum = timr->it_overrun_last + (s64)baseval;
289
290 return sum > (s64)INT_MAX ? INT_MAX : (int)sum;
291}
292
293static void common_hrtimer_rearm(struct k_itimer *timr)
294{
295 struct hrtimer *timer = &timr->it.real.timer;
296
297 timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
298 timr->it_interval);
299 hrtimer_restart(timer);
300}
301
302/*
303 * This function is exported for use by the signal deliver code. It is
304 * called just prior to the info block being released and passes that
305 * block to us. It's function is to update the overrun entry AND to
306 * restart the timer. It should only be called if the timer is to be
307 * restarted (i.e. we have flagged this in the sys_private entry of the
308 * info block).
309 *
310 * To protect against the timer going away while the interrupt is queued,
311 * we require that the it_requeue_pending flag be set.
312 */
313void posixtimer_rearm(struct kernel_siginfo *info)
314{
315 struct k_itimer *timr;
316 unsigned long flags;
317
318 timr = lock_timer(info->si_tid, &flags);
319 if (!timr)
320 return;
321
322 if (timr->it_interval && timr->it_requeue_pending == info->si_sys_private) {
323 timr->kclock->timer_rearm(timr);
324
325 timr->it_active = 1;
326 timr->it_overrun_last = timr->it_overrun;
327 timr->it_overrun = -1LL;
328 ++timr->it_requeue_pending;
329
330 info->si_overrun = timer_overrun_to_int(timr, info->si_overrun);
331 }
332
333 unlock_timer(timr, flags);
334}
335
336int posix_timer_event(struct k_itimer *timr, int si_private)
337{
338 enum pid_type type;
339 int ret;
340 /*
341 * FIXME: if ->sigq is queued we can race with
342 * dequeue_signal()->posixtimer_rearm().
343 *
344 * If dequeue_signal() sees the "right" value of
345 * si_sys_private it calls posixtimer_rearm().
346 * We re-queue ->sigq and drop ->it_lock().
347 * posixtimer_rearm() locks the timer
348 * and re-schedules it while ->sigq is pending.
349 * Not really bad, but not that we want.
350 */
351 timr->sigq->info.si_sys_private = si_private;
352
353 type = !(timr->it_sigev_notify & SIGEV_THREAD_ID) ? PIDTYPE_TGID : PIDTYPE_PID;
354 ret = send_sigqueue(timr->sigq, timr->it_pid, type);
355 /* If we failed to send the signal the timer stops. */
356 return ret > 0;
357}
358
359/*
360 * This function gets called when a POSIX.1b interval timer expires. It
361 * is used as a callback from the kernel internal timer. The
362 * run_timer_list code ALWAYS calls with interrupts on.
363
364 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
365 */
366static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
367{
368 struct k_itimer *timr;
369 unsigned long flags;
370 int si_private = 0;
371 enum hrtimer_restart ret = HRTIMER_NORESTART;
372
373 timr = container_of(timer, struct k_itimer, it.real.timer);
374 spin_lock_irqsave(&timr->it_lock, flags);
375
376 timr->it_active = 0;
377 if (timr->it_interval != 0)
378 si_private = ++timr->it_requeue_pending;
379
380 if (posix_timer_event(timr, si_private)) {
381 /*
382 * signal was not sent because of sig_ignor
383 * we will not get a call back to restart it AND
384 * it should be restarted.
385 */
386 if (timr->it_interval != 0) {
387 ktime_t now = hrtimer_cb_get_time(timer);
388
389 /*
390 * FIXME: What we really want, is to stop this
391 * timer completely and restart it in case the
392 * SIG_IGN is removed. This is a non trivial
393 * change which involves sighand locking
394 * (sigh !), which we don't want to do late in
395 * the release cycle.
396 *
397 * For now we just let timers with an interval
398 * less than a jiffie expire every jiffie to
399 * avoid softirq starvation in case of SIG_IGN
400 * and a very small interval, which would put
401 * the timer right back on the softirq pending
402 * list. By moving now ahead of time we trick
403 * hrtimer_forward() to expire the timer
404 * later, while we still maintain the overrun
405 * accuracy, but have some inconsistency in
406 * the timer_gettime() case. This is at least
407 * better than a starved softirq. A more
408 * complex fix which solves also another related
409 * inconsistency is already in the pipeline.
410 */
411#ifdef CONFIG_HIGH_RES_TIMERS
412 {
413 ktime_t kj = NSEC_PER_SEC / HZ;
414
415 if (timr->it_interval < kj)
416 now = ktime_add(now, kj);
417 }
418#endif
419 timr->it_overrun += hrtimer_forward(timer, now,
420 timr->it_interval);
421 ret = HRTIMER_RESTART;
422 ++timr->it_requeue_pending;
423 timr->it_active = 1;
424 }
425 }
426
427 unlock_timer(timr, flags);
428 return ret;
429}
430
431static struct pid *good_sigevent(sigevent_t * event)
432{
433 struct pid *pid = task_tgid(current);
434 struct task_struct *rtn;
435
436 switch (event->sigev_notify) {
437 case SIGEV_SIGNAL | SIGEV_THREAD_ID:
438 pid = find_vpid(event->sigev_notify_thread_id);
439 rtn = pid_task(pid, PIDTYPE_PID);
440 if (!rtn || !same_thread_group(rtn, current))
441 return NULL;
442 fallthrough;
443 case SIGEV_SIGNAL:
444 case SIGEV_THREAD:
445 if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
446 return NULL;
447 fallthrough;
448 case SIGEV_NONE:
449 return pid;
450 default:
451 return NULL;
452 }
453}
454
455static struct k_itimer * alloc_posix_timer(void)
456{
457 struct k_itimer *tmr;
458 tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
459 if (!tmr)
460 return tmr;
461 if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
462 kmem_cache_free(posix_timers_cache, tmr);
463 return NULL;
464 }
465 clear_siginfo(&tmr->sigq->info);
466 return tmr;
467}
468
469static void k_itimer_rcu_free(struct rcu_head *head)
470{
471 struct k_itimer *tmr = container_of(head, struct k_itimer, rcu);
472
473 kmem_cache_free(posix_timers_cache, tmr);
474}
475
476#define IT_ID_SET 1
477#define IT_ID_NOT_SET 0
478static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
479{
480 if (it_id_set) {
481 unsigned long flags;
482 spin_lock_irqsave(&hash_lock, flags);
483 hlist_del_rcu(&tmr->t_hash);
484 spin_unlock_irqrestore(&hash_lock, flags);
485 }
486 put_pid(tmr->it_pid);
487 sigqueue_free(tmr->sigq);
488 call_rcu(&tmr->rcu, k_itimer_rcu_free);
489}
490
491static int common_timer_create(struct k_itimer *new_timer)
492{
493 hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
494 return 0;
495}
496
497/* Create a POSIX.1b interval timer. */
498static int do_timer_create(clockid_t which_clock, struct sigevent *event,
499 timer_t __user *created_timer_id)
500{
501 const struct k_clock *kc = clockid_to_kclock(which_clock);
502 struct k_itimer *new_timer;
503 int error, new_timer_id;
504 int it_id_set = IT_ID_NOT_SET;
505
506 if (!kc)
507 return -EINVAL;
508 if (!kc->timer_create)
509 return -EOPNOTSUPP;
510
511 new_timer = alloc_posix_timer();
512 if (unlikely(!new_timer))
513 return -EAGAIN;
514
515 spin_lock_init(&new_timer->it_lock);
516 new_timer_id = posix_timer_add(new_timer);
517 if (new_timer_id < 0) {
518 error = new_timer_id;
519 goto out;
520 }
521
522 it_id_set = IT_ID_SET;
523 new_timer->it_id = (timer_t) new_timer_id;
524 new_timer->it_clock = which_clock;
525 new_timer->kclock = kc;
526 new_timer->it_overrun = -1LL;
527
528 if (event) {
529 rcu_read_lock();
530 new_timer->it_pid = get_pid(good_sigevent(event));
531 rcu_read_unlock();
532 if (!new_timer->it_pid) {
533 error = -EINVAL;
534 goto out;
535 }
536 new_timer->it_sigev_notify = event->sigev_notify;
537 new_timer->sigq->info.si_signo = event->sigev_signo;
538 new_timer->sigq->info.si_value = event->sigev_value;
539 } else {
540 new_timer->it_sigev_notify = SIGEV_SIGNAL;
541 new_timer->sigq->info.si_signo = SIGALRM;
542 memset(&new_timer->sigq->info.si_value, 0, sizeof(sigval_t));
543 new_timer->sigq->info.si_value.sival_int = new_timer->it_id;
544 new_timer->it_pid = get_pid(task_tgid(current));
545 }
546
547 new_timer->sigq->info.si_tid = new_timer->it_id;
548 new_timer->sigq->info.si_code = SI_TIMER;
549
550 if (copy_to_user(created_timer_id,
551 &new_timer_id, sizeof (new_timer_id))) {
552 error = -EFAULT;
553 goto out;
554 }
555
556 error = kc->timer_create(new_timer);
557 if (error)
558 goto out;
559
560 spin_lock_irq(¤t->sighand->siglock);
561 new_timer->it_signal = current->signal;
562 list_add(&new_timer->list, ¤t->signal->posix_timers);
563 spin_unlock_irq(¤t->sighand->siglock);
564
565 return 0;
566 /*
567 * In the case of the timer belonging to another task, after
568 * the task is unlocked, the timer is owned by the other task
569 * and may cease to exist at any time. Don't use or modify
570 * new_timer after the unlock call.
571 */
572out:
573 release_posix_timer(new_timer, it_id_set);
574 return error;
575}
576
577SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
578 struct sigevent __user *, timer_event_spec,
579 timer_t __user *, created_timer_id)
580{
581 if (timer_event_spec) {
582 sigevent_t event;
583
584 if (copy_from_user(&event, timer_event_spec, sizeof (event)))
585 return -EFAULT;
586 return do_timer_create(which_clock, &event, created_timer_id);
587 }
588 return do_timer_create(which_clock, NULL, created_timer_id);
589}
590
591#ifdef CONFIG_COMPAT
592COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
593 struct compat_sigevent __user *, timer_event_spec,
594 timer_t __user *, created_timer_id)
595{
596 if (timer_event_spec) {
597 sigevent_t event;
598
599 if (get_compat_sigevent(&event, timer_event_spec))
600 return -EFAULT;
601 return do_timer_create(which_clock, &event, created_timer_id);
602 }
603 return do_timer_create(which_clock, NULL, created_timer_id);
604}
605#endif
606
607/*
608 * Locking issues: We need to protect the result of the id look up until
609 * we get the timer locked down so it is not deleted under us. The
610 * removal is done under the idr spinlock so we use that here to bridge
611 * the find to the timer lock. To avoid a dead lock, the timer id MUST
612 * be release with out holding the timer lock.
613 */
614static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
615{
616 struct k_itimer *timr;
617
618 /*
619 * timer_t could be any type >= int and we want to make sure any
620 * @timer_id outside positive int range fails lookup.
621 */
622 if ((unsigned long long)timer_id > INT_MAX)
623 return NULL;
624
625 rcu_read_lock();
626 timr = posix_timer_by_id(timer_id);
627 if (timr) {
628 spin_lock_irqsave(&timr->it_lock, *flags);
629 if (timr->it_signal == current->signal) {
630 rcu_read_unlock();
631 return timr;
632 }
633 spin_unlock_irqrestore(&timr->it_lock, *flags);
634 }
635 rcu_read_unlock();
636
637 return NULL;
638}
639
640static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
641{
642 struct hrtimer *timer = &timr->it.real.timer;
643
644 return __hrtimer_expires_remaining_adjusted(timer, now);
645}
646
647static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
648{
649 struct hrtimer *timer = &timr->it.real.timer;
650
651 return hrtimer_forward(timer, now, timr->it_interval);
652}
653
654/*
655 * Get the time remaining on a POSIX.1b interval timer. This function
656 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
657 * mess with irq.
658 *
659 * We have a couple of messes to clean up here. First there is the case
660 * of a timer that has a requeue pending. These timers should appear to
661 * be in the timer list with an expiry as if we were to requeue them
662 * now.
663 *
664 * The second issue is the SIGEV_NONE timer which may be active but is
665 * not really ever put in the timer list (to save system resources).
666 * This timer may be expired, and if so, we will do it here. Otherwise
667 * it is the same as a requeue pending timer WRT to what we should
668 * report.
669 */
670void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
671{
672 const struct k_clock *kc = timr->kclock;
673 ktime_t now, remaining, iv;
674 bool sig_none;
675
676 sig_none = timr->it_sigev_notify == SIGEV_NONE;
677 iv = timr->it_interval;
678
679 /* interval timer ? */
680 if (iv) {
681 cur_setting->it_interval = ktime_to_timespec64(iv);
682 } else if (!timr->it_active) {
683 /*
684 * SIGEV_NONE oneshot timers are never queued. Check them
685 * below.
686 */
687 if (!sig_none)
688 return;
689 }
690
691 now = kc->clock_get_ktime(timr->it_clock);
692
693 /*
694 * When a requeue is pending or this is a SIGEV_NONE timer move the
695 * expiry time forward by intervals, so expiry is > now.
696 */
697 if (iv && (timr->it_requeue_pending & REQUEUE_PENDING || sig_none))
698 timr->it_overrun += kc->timer_forward(timr, now);
699
700 remaining = kc->timer_remaining(timr, now);
701 /* Return 0 only, when the timer is expired and not pending */
702 if (remaining <= 0) {
703 /*
704 * A single shot SIGEV_NONE timer must return 0, when
705 * it is expired !
706 */
707 if (!sig_none)
708 cur_setting->it_value.tv_nsec = 1;
709 } else {
710 cur_setting->it_value = ktime_to_timespec64(remaining);
711 }
712}
713
714/* Get the time remaining on a POSIX.1b interval timer. */
715static int do_timer_gettime(timer_t timer_id, struct itimerspec64 *setting)
716{
717 struct k_itimer *timr;
718 const struct k_clock *kc;
719 unsigned long flags;
720 int ret = 0;
721
722 timr = lock_timer(timer_id, &flags);
723 if (!timr)
724 return -EINVAL;
725
726 memset(setting, 0, sizeof(*setting));
727 kc = timr->kclock;
728 if (WARN_ON_ONCE(!kc || !kc->timer_get))
729 ret = -EINVAL;
730 else
731 kc->timer_get(timr, setting);
732
733 unlock_timer(timr, flags);
734 return ret;
735}
736
737/* Get the time remaining on a POSIX.1b interval timer. */
738SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
739 struct __kernel_itimerspec __user *, setting)
740{
741 struct itimerspec64 cur_setting;
742
743 int ret = do_timer_gettime(timer_id, &cur_setting);
744 if (!ret) {
745 if (put_itimerspec64(&cur_setting, setting))
746 ret = -EFAULT;
747 }
748 return ret;
749}
750
751#ifdef CONFIG_COMPAT_32BIT_TIME
752
753SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
754 struct old_itimerspec32 __user *, setting)
755{
756 struct itimerspec64 cur_setting;
757
758 int ret = do_timer_gettime(timer_id, &cur_setting);
759 if (!ret) {
760 if (put_old_itimerspec32(&cur_setting, setting))
761 ret = -EFAULT;
762 }
763 return ret;
764}
765
766#endif
767
768/*
769 * Get the number of overruns of a POSIX.1b interval timer. This is to
770 * be the overrun of the timer last delivered. At the same time we are
771 * accumulating overruns on the next timer. The overrun is frozen when
772 * the signal is delivered, either at the notify time (if the info block
773 * is not queued) or at the actual delivery time (as we are informed by
774 * the call back to posixtimer_rearm(). So all we need to do is
775 * to pick up the frozen overrun.
776 */
777SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
778{
779 struct k_itimer *timr;
780 int overrun;
781 unsigned long flags;
782
783 timr = lock_timer(timer_id, &flags);
784 if (!timr)
785 return -EINVAL;
786
787 overrun = timer_overrun_to_int(timr, 0);
788 unlock_timer(timr, flags);
789
790 return overrun;
791}
792
793static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
794 bool absolute, bool sigev_none)
795{
796 struct hrtimer *timer = &timr->it.real.timer;
797 enum hrtimer_mode mode;
798
799 mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
800 /*
801 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
802 * clock modifications, so they become CLOCK_MONOTONIC based under the
803 * hood. See hrtimer_init(). Update timr->kclock, so the generic
804 * functions which use timr->kclock->clock_get_*() work.
805 *
806 * Note: it_clock stays unmodified, because the next timer_set() might
807 * use ABSTIME, so it needs to switch back.
808 */
809 if (timr->it_clock == CLOCK_REALTIME)
810 timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
811
812 hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
813 timr->it.real.timer.function = posix_timer_fn;
814
815 if (!absolute)
816 expires = ktime_add_safe(expires, timer->base->get_time());
817 hrtimer_set_expires(timer, expires);
818
819 if (!sigev_none)
820 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
821}
822
823static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
824{
825 return hrtimer_try_to_cancel(&timr->it.real.timer);
826}
827
828static void common_timer_wait_running(struct k_itimer *timer)
829{
830 hrtimer_cancel_wait_running(&timer->it.real.timer);
831}
832
833/*
834 * On PREEMPT_RT this prevent priority inversion against softirq kthread in
835 * case it gets preempted while executing a timer callback. See comments in
836 * hrtimer_cancel_wait_running. For PREEMPT_RT=n this just results in a
837 * cpu_relax().
838 */
839static struct k_itimer *timer_wait_running(struct k_itimer *timer,
840 unsigned long *flags)
841{
842 const struct k_clock *kc = READ_ONCE(timer->kclock);
843 timer_t timer_id = READ_ONCE(timer->it_id);
844
845 /* Prevent kfree(timer) after dropping the lock */
846 rcu_read_lock();
847 unlock_timer(timer, *flags);
848
849 if (!WARN_ON_ONCE(!kc->timer_wait_running))
850 kc->timer_wait_running(timer);
851
852 rcu_read_unlock();
853 /* Relock the timer. It might be not longer hashed. */
854 return lock_timer(timer_id, flags);
855}
856
857/* Set a POSIX.1b interval timer. */
858int common_timer_set(struct k_itimer *timr, int flags,
859 struct itimerspec64 *new_setting,
860 struct itimerspec64 *old_setting)
861{
862 const struct k_clock *kc = timr->kclock;
863 bool sigev_none;
864 ktime_t expires;
865
866 if (old_setting)
867 common_timer_get(timr, old_setting);
868
869 /* Prevent rearming by clearing the interval */
870 timr->it_interval = 0;
871 /*
872 * Careful here. On SMP systems the timer expiry function could be
873 * active and spinning on timr->it_lock.
874 */
875 if (kc->timer_try_to_cancel(timr) < 0)
876 return TIMER_RETRY;
877
878 timr->it_active = 0;
879 timr->it_requeue_pending = (timr->it_requeue_pending + 2) &
880 ~REQUEUE_PENDING;
881 timr->it_overrun_last = 0;
882
883 /* Switch off the timer when it_value is zero */
884 if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
885 return 0;
886
887 timr->it_interval = timespec64_to_ktime(new_setting->it_interval);
888 expires = timespec64_to_ktime(new_setting->it_value);
889 if (flags & TIMER_ABSTIME)
890 expires = timens_ktime_to_host(timr->it_clock, expires);
891 sigev_none = timr->it_sigev_notify == SIGEV_NONE;
892
893 kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
894 timr->it_active = !sigev_none;
895 return 0;
896}
897
898static int do_timer_settime(timer_t timer_id, int tmr_flags,
899 struct itimerspec64 *new_spec64,
900 struct itimerspec64 *old_spec64)
901{
902 const struct k_clock *kc;
903 struct k_itimer *timr;
904 unsigned long flags;
905 int error = 0;
906
907 if (!timespec64_valid(&new_spec64->it_interval) ||
908 !timespec64_valid(&new_spec64->it_value))
909 return -EINVAL;
910
911 if (old_spec64)
912 memset(old_spec64, 0, sizeof(*old_spec64));
913
914 timr = lock_timer(timer_id, &flags);
915retry:
916 if (!timr)
917 return -EINVAL;
918
919 kc = timr->kclock;
920 if (WARN_ON_ONCE(!kc || !kc->timer_set))
921 error = -EINVAL;
922 else
923 error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
924
925 if (error == TIMER_RETRY) {
926 // We already got the old time...
927 old_spec64 = NULL;
928 /* Unlocks and relocks the timer if it still exists */
929 timr = timer_wait_running(timr, &flags);
930 goto retry;
931 }
932 unlock_timer(timr, flags);
933
934 return error;
935}
936
937/* Set a POSIX.1b interval timer */
938SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
939 const struct __kernel_itimerspec __user *, new_setting,
940 struct __kernel_itimerspec __user *, old_setting)
941{
942 struct itimerspec64 new_spec, old_spec;
943 struct itimerspec64 *rtn = old_setting ? &old_spec : NULL;
944 int error = 0;
945
946 if (!new_setting)
947 return -EINVAL;
948
949 if (get_itimerspec64(&new_spec, new_setting))
950 return -EFAULT;
951
952 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
953 if (!error && old_setting) {
954 if (put_itimerspec64(&old_spec, old_setting))
955 error = -EFAULT;
956 }
957 return error;
958}
959
960#ifdef CONFIG_COMPAT_32BIT_TIME
961SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
962 struct old_itimerspec32 __user *, new,
963 struct old_itimerspec32 __user *, old)
964{
965 struct itimerspec64 new_spec, old_spec;
966 struct itimerspec64 *rtn = old ? &old_spec : NULL;
967 int error = 0;
968
969 if (!new)
970 return -EINVAL;
971 if (get_old_itimerspec32(&new_spec, new))
972 return -EFAULT;
973
974 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
975 if (!error && old) {
976 if (put_old_itimerspec32(&old_spec, old))
977 error = -EFAULT;
978 }
979 return error;
980}
981#endif
982
983int common_timer_del(struct k_itimer *timer)
984{
985 const struct k_clock *kc = timer->kclock;
986
987 timer->it_interval = 0;
988 if (kc->timer_try_to_cancel(timer) < 0)
989 return TIMER_RETRY;
990 timer->it_active = 0;
991 return 0;
992}
993
994static inline int timer_delete_hook(struct k_itimer *timer)
995{
996 const struct k_clock *kc = timer->kclock;
997
998 if (WARN_ON_ONCE(!kc || !kc->timer_del))
999 return -EINVAL;
1000 return kc->timer_del(timer);
1001}
1002
1003/* Delete a POSIX.1b interval timer. */
1004SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
1005{
1006 struct k_itimer *timer;
1007 unsigned long flags;
1008
1009 timer = lock_timer(timer_id, &flags);
1010
1011retry_delete:
1012 if (!timer)
1013 return -EINVAL;
1014
1015 if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
1016 /* Unlocks and relocks the timer if it still exists */
1017 timer = timer_wait_running(timer, &flags);
1018 goto retry_delete;
1019 }
1020
1021 spin_lock(¤t->sighand->siglock);
1022 list_del(&timer->list);
1023 spin_unlock(¤t->sighand->siglock);
1024 /*
1025 * This keeps any tasks waiting on the spin lock from thinking
1026 * they got something (see the lock code above).
1027 */
1028 timer->it_signal = NULL;
1029
1030 unlock_timer(timer, flags);
1031 release_posix_timer(timer, IT_ID_SET);
1032 return 0;
1033}
1034
1035/*
1036 * return timer owned by the process, used by exit_itimers
1037 */
1038static void itimer_delete(struct k_itimer *timer)
1039{
1040retry_delete:
1041 spin_lock_irq(&timer->it_lock);
1042
1043 if (timer_delete_hook(timer) == TIMER_RETRY) {
1044 spin_unlock_irq(&timer->it_lock);
1045 goto retry_delete;
1046 }
1047 list_del(&timer->list);
1048
1049 spin_unlock_irq(&timer->it_lock);
1050 release_posix_timer(timer, IT_ID_SET);
1051}
1052
1053/*
1054 * This is called by do_exit or de_thread, only when nobody else can
1055 * modify the signal->posix_timers list. Yet we need sighand->siglock
1056 * to prevent the race with /proc/pid/timers.
1057 */
1058void exit_itimers(struct task_struct *tsk)
1059{
1060 struct list_head timers;
1061 struct k_itimer *tmr;
1062
1063 if (list_empty(&tsk->signal->posix_timers))
1064 return;
1065
1066 spin_lock_irq(&tsk->sighand->siglock);
1067 list_replace_init(&tsk->signal->posix_timers, &timers);
1068 spin_unlock_irq(&tsk->sighand->siglock);
1069
1070 while (!list_empty(&timers)) {
1071 tmr = list_first_entry(&timers, struct k_itimer, list);
1072 itimer_delete(tmr);
1073 }
1074}
1075
1076SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1077 const struct __kernel_timespec __user *, tp)
1078{
1079 const struct k_clock *kc = clockid_to_kclock(which_clock);
1080 struct timespec64 new_tp;
1081
1082 if (!kc || !kc->clock_set)
1083 return -EINVAL;
1084
1085 if (get_timespec64(&new_tp, tp))
1086 return -EFAULT;
1087
1088 return kc->clock_set(which_clock, &new_tp);
1089}
1090
1091SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1092 struct __kernel_timespec __user *, tp)
1093{
1094 const struct k_clock *kc = clockid_to_kclock(which_clock);
1095 struct timespec64 kernel_tp;
1096 int error;
1097
1098 if (!kc)
1099 return -EINVAL;
1100
1101 error = kc->clock_get_timespec(which_clock, &kernel_tp);
1102
1103 if (!error && put_timespec64(&kernel_tp, tp))
1104 error = -EFAULT;
1105
1106 return error;
1107}
1108
1109int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1110{
1111 const struct k_clock *kc = clockid_to_kclock(which_clock);
1112
1113 if (!kc)
1114 return -EINVAL;
1115 if (!kc->clock_adj)
1116 return -EOPNOTSUPP;
1117
1118 return kc->clock_adj(which_clock, ktx);
1119}
1120
1121SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1122 struct __kernel_timex __user *, utx)
1123{
1124 struct __kernel_timex ktx;
1125 int err;
1126
1127 if (copy_from_user(&ktx, utx, sizeof(ktx)))
1128 return -EFAULT;
1129
1130 err = do_clock_adjtime(which_clock, &ktx);
1131
1132 if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1133 return -EFAULT;
1134
1135 return err;
1136}
1137
1138SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1139 struct __kernel_timespec __user *, tp)
1140{
1141 const struct k_clock *kc = clockid_to_kclock(which_clock);
1142 struct timespec64 rtn_tp;
1143 int error;
1144
1145 if (!kc)
1146 return -EINVAL;
1147
1148 error = kc->clock_getres(which_clock, &rtn_tp);
1149
1150 if (!error && tp && put_timespec64(&rtn_tp, tp))
1151 error = -EFAULT;
1152
1153 return error;
1154}
1155
1156#ifdef CONFIG_COMPAT_32BIT_TIME
1157
1158SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1159 struct old_timespec32 __user *, tp)
1160{
1161 const struct k_clock *kc = clockid_to_kclock(which_clock);
1162 struct timespec64 ts;
1163
1164 if (!kc || !kc->clock_set)
1165 return -EINVAL;
1166
1167 if (get_old_timespec32(&ts, tp))
1168 return -EFAULT;
1169
1170 return kc->clock_set(which_clock, &ts);
1171}
1172
1173SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1174 struct old_timespec32 __user *, tp)
1175{
1176 const struct k_clock *kc = clockid_to_kclock(which_clock);
1177 struct timespec64 ts;
1178 int err;
1179
1180 if (!kc)
1181 return -EINVAL;
1182
1183 err = kc->clock_get_timespec(which_clock, &ts);
1184
1185 if (!err && put_old_timespec32(&ts, tp))
1186 err = -EFAULT;
1187
1188 return err;
1189}
1190
1191SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1192 struct old_timex32 __user *, utp)
1193{
1194 struct __kernel_timex ktx;
1195 int err;
1196
1197 err = get_old_timex32(&ktx, utp);
1198 if (err)
1199 return err;
1200
1201 err = do_clock_adjtime(which_clock, &ktx);
1202
1203 if (err >= 0 && put_old_timex32(utp, &ktx))
1204 return -EFAULT;
1205
1206 return err;
1207}
1208
1209SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1210 struct old_timespec32 __user *, tp)
1211{
1212 const struct k_clock *kc = clockid_to_kclock(which_clock);
1213 struct timespec64 ts;
1214 int err;
1215
1216 if (!kc)
1217 return -EINVAL;
1218
1219 err = kc->clock_getres(which_clock, &ts);
1220 if (!err && tp && put_old_timespec32(&ts, tp))
1221 return -EFAULT;
1222
1223 return err;
1224}
1225
1226#endif
1227
1228/*
1229 * nanosleep for monotonic and realtime clocks
1230 */
1231static int common_nsleep(const clockid_t which_clock, int flags,
1232 const struct timespec64 *rqtp)
1233{
1234 ktime_t texp = timespec64_to_ktime(*rqtp);
1235
1236 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1237 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1238 which_clock);
1239}
1240
1241static int common_nsleep_timens(const clockid_t which_clock, int flags,
1242 const struct timespec64 *rqtp)
1243{
1244 ktime_t texp = timespec64_to_ktime(*rqtp);
1245
1246 if (flags & TIMER_ABSTIME)
1247 texp = timens_ktime_to_host(which_clock, texp);
1248
1249 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1250 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1251 which_clock);
1252}
1253
1254SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1255 const struct __kernel_timespec __user *, rqtp,
1256 struct __kernel_timespec __user *, rmtp)
1257{
1258 const struct k_clock *kc = clockid_to_kclock(which_clock);
1259 struct timespec64 t;
1260
1261 if (!kc)
1262 return -EINVAL;
1263 if (!kc->nsleep)
1264 return -EOPNOTSUPP;
1265
1266 if (get_timespec64(&t, rqtp))
1267 return -EFAULT;
1268
1269 if (!timespec64_valid(&t))
1270 return -EINVAL;
1271 if (flags & TIMER_ABSTIME)
1272 rmtp = NULL;
1273 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1274 current->restart_block.nanosleep.rmtp = rmtp;
1275
1276 return kc->nsleep(which_clock, flags, &t);
1277}
1278
1279#ifdef CONFIG_COMPAT_32BIT_TIME
1280
1281SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1282 struct old_timespec32 __user *, rqtp,
1283 struct old_timespec32 __user *, rmtp)
1284{
1285 const struct k_clock *kc = clockid_to_kclock(which_clock);
1286 struct timespec64 t;
1287
1288 if (!kc)
1289 return -EINVAL;
1290 if (!kc->nsleep)
1291 return -EOPNOTSUPP;
1292
1293 if (get_old_timespec32(&t, rqtp))
1294 return -EFAULT;
1295
1296 if (!timespec64_valid(&t))
1297 return -EINVAL;
1298 if (flags & TIMER_ABSTIME)
1299 rmtp = NULL;
1300 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1301 current->restart_block.nanosleep.compat_rmtp = rmtp;
1302
1303 return kc->nsleep(which_clock, flags, &t);
1304}
1305
1306#endif
1307
1308static const struct k_clock clock_realtime = {
1309 .clock_getres = posix_get_hrtimer_res,
1310 .clock_get_timespec = posix_get_realtime_timespec,
1311 .clock_get_ktime = posix_get_realtime_ktime,
1312 .clock_set = posix_clock_realtime_set,
1313 .clock_adj = posix_clock_realtime_adj,
1314 .nsleep = common_nsleep,
1315 .timer_create = common_timer_create,
1316 .timer_set = common_timer_set,
1317 .timer_get = common_timer_get,
1318 .timer_del = common_timer_del,
1319 .timer_rearm = common_hrtimer_rearm,
1320 .timer_forward = common_hrtimer_forward,
1321 .timer_remaining = common_hrtimer_remaining,
1322 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1323 .timer_wait_running = common_timer_wait_running,
1324 .timer_arm = common_hrtimer_arm,
1325};
1326
1327static const struct k_clock clock_monotonic = {
1328 .clock_getres = posix_get_hrtimer_res,
1329 .clock_get_timespec = posix_get_monotonic_timespec,
1330 .clock_get_ktime = posix_get_monotonic_ktime,
1331 .nsleep = common_nsleep_timens,
1332 .timer_create = common_timer_create,
1333 .timer_set = common_timer_set,
1334 .timer_get = common_timer_get,
1335 .timer_del = common_timer_del,
1336 .timer_rearm = common_hrtimer_rearm,
1337 .timer_forward = common_hrtimer_forward,
1338 .timer_remaining = common_hrtimer_remaining,
1339 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1340 .timer_wait_running = common_timer_wait_running,
1341 .timer_arm = common_hrtimer_arm,
1342};
1343
1344static const struct k_clock clock_monotonic_raw = {
1345 .clock_getres = posix_get_hrtimer_res,
1346 .clock_get_timespec = posix_get_monotonic_raw,
1347};
1348
1349static const struct k_clock clock_realtime_coarse = {
1350 .clock_getres = posix_get_coarse_res,
1351 .clock_get_timespec = posix_get_realtime_coarse,
1352};
1353
1354static const struct k_clock clock_monotonic_coarse = {
1355 .clock_getres = posix_get_coarse_res,
1356 .clock_get_timespec = posix_get_monotonic_coarse,
1357};
1358
1359static const struct k_clock clock_tai = {
1360 .clock_getres = posix_get_hrtimer_res,
1361 .clock_get_ktime = posix_get_tai_ktime,
1362 .clock_get_timespec = posix_get_tai_timespec,
1363 .nsleep = common_nsleep,
1364 .timer_create = common_timer_create,
1365 .timer_set = common_timer_set,
1366 .timer_get = common_timer_get,
1367 .timer_del = common_timer_del,
1368 .timer_rearm = common_hrtimer_rearm,
1369 .timer_forward = common_hrtimer_forward,
1370 .timer_remaining = common_hrtimer_remaining,
1371 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1372 .timer_wait_running = common_timer_wait_running,
1373 .timer_arm = common_hrtimer_arm,
1374};
1375
1376static const struct k_clock clock_boottime = {
1377 .clock_getres = posix_get_hrtimer_res,
1378 .clock_get_ktime = posix_get_boottime_ktime,
1379 .clock_get_timespec = posix_get_boottime_timespec,
1380 .nsleep = common_nsleep_timens,
1381 .timer_create = common_timer_create,
1382 .timer_set = common_timer_set,
1383 .timer_get = common_timer_get,
1384 .timer_del = common_timer_del,
1385 .timer_rearm = common_hrtimer_rearm,
1386 .timer_forward = common_hrtimer_forward,
1387 .timer_remaining = common_hrtimer_remaining,
1388 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1389 .timer_wait_running = common_timer_wait_running,
1390 .timer_arm = common_hrtimer_arm,
1391};
1392
1393static const struct k_clock * const posix_clocks[] = {
1394 [CLOCK_REALTIME] = &clock_realtime,
1395 [CLOCK_MONOTONIC] = &clock_monotonic,
1396 [CLOCK_PROCESS_CPUTIME_ID] = &clock_process,
1397 [CLOCK_THREAD_CPUTIME_ID] = &clock_thread,
1398 [CLOCK_MONOTONIC_RAW] = &clock_monotonic_raw,
1399 [CLOCK_REALTIME_COARSE] = &clock_realtime_coarse,
1400 [CLOCK_MONOTONIC_COARSE] = &clock_monotonic_coarse,
1401 [CLOCK_BOOTTIME] = &clock_boottime,
1402 [CLOCK_REALTIME_ALARM] = &alarm_clock,
1403 [CLOCK_BOOTTIME_ALARM] = &alarm_clock,
1404 [CLOCK_TAI] = &clock_tai,
1405};
1406
1407static const struct k_clock *clockid_to_kclock(const clockid_t id)
1408{
1409 clockid_t idx = id;
1410
1411 if (id < 0) {
1412 return (id & CLOCKFD_MASK) == CLOCKFD ?
1413 &clock_posix_dynamic : &clock_posix_cpu;
1414 }
1415
1416 if (id >= ARRAY_SIZE(posix_clocks))
1417 return NULL;
1418
1419 return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1420}
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
34#include "timekeeping.h"
35#include "posix-timers.h"
36
37/*
38 * Management arrays for POSIX timers. Timers are now kept in static hash table
39 * with 512 entries.
40 * Timer ids are allocated by local routine, which selects proper hash head by
41 * key, constructed from current->signal address and per signal struct counter.
42 * This keeps timer ids unique per process, but now they can intersect between
43 * processes.
44 */
45
46/*
47 * Lets keep our timers in a slab cache :-)
48 */
49static struct kmem_cache *posix_timers_cache;
50
51static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
52static DEFINE_SPINLOCK(hash_lock);
53
54static const struct k_clock * const posix_clocks[];
55static const struct k_clock *clockid_to_kclock(const clockid_t id);
56static const struct k_clock clock_realtime, clock_monotonic;
57
58/*
59 * we assume that the new SIGEV_THREAD_ID shares no bits with the other
60 * SIGEV values. Here we put out an error if this assumption fails.
61 */
62#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
63 ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
64#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
65#endif
66
67/*
68 * The timer ID is turned into a timer address by idr_find().
69 * Verifying a valid ID consists of:
70 *
71 * a) checking that idr_find() returns other than -1.
72 * b) checking that the timer id matches the one in the timer itself.
73 * c) that the timer owner is in the callers thread group.
74 */
75
76/*
77 * CLOCKs: The POSIX standard calls for a couple of clocks and allows us
78 * to implement others. This structure defines the various
79 * clocks.
80 *
81 * RESOLUTION: Clock resolution is used to round up timer and interval
82 * times, NOT to report clock times, which are reported with as
83 * much resolution as the system can muster. In some cases this
84 * resolution may depend on the underlying clock hardware and
85 * may not be quantifiable until run time, and only then is the
86 * necessary code is written. The standard says we should say
87 * something about this issue in the documentation...
88 *
89 * FUNCTIONS: The CLOCKs structure defines possible functions to
90 * handle various clock functions.
91 *
92 * The standard POSIX timer management code assumes the
93 * following: 1.) The k_itimer struct (sched.h) is used for
94 * the timer. 2.) The list, it_lock, it_clock, it_id and
95 * it_pid fields are not modified by timer code.
96 *
97 * Permissions: It is assumed that the clock_settime() function defined
98 * for each clock will take care of permission checks. Some
99 * clocks may be set able by any user (i.e. local process
100 * clocks) others not. Currently the only set able clock we
101 * have is CLOCK_REALTIME and its high res counter part, both of
102 * which we beg off on and pass to do_sys_settimeofday().
103 */
104static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
105
106#define lock_timer(tid, flags) \
107({ struct k_itimer *__timr; \
108 __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags)); \
109 __timr; \
110})
111
112static int hash(struct signal_struct *sig, unsigned int nr)
113{
114 return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
115}
116
117static struct k_itimer *__posix_timers_find(struct hlist_head *head,
118 struct signal_struct *sig,
119 timer_t id)
120{
121 struct k_itimer *timer;
122
123 hlist_for_each_entry_rcu(timer, head, t_hash) {
124 if ((timer->it_signal == sig) && (timer->it_id == id))
125 return timer;
126 }
127 return NULL;
128}
129
130static struct k_itimer *posix_timer_by_id(timer_t id)
131{
132 struct signal_struct *sig = current->signal;
133 struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
134
135 return __posix_timers_find(head, sig, id);
136}
137
138static int posix_timer_add(struct k_itimer *timer)
139{
140 struct signal_struct *sig = current->signal;
141 int first_free_id = sig->posix_timer_id;
142 struct hlist_head *head;
143 int ret = -ENOENT;
144
145 do {
146 spin_lock(&hash_lock);
147 head = &posix_timers_hashtable[hash(sig, sig->posix_timer_id)];
148 if (!__posix_timers_find(head, sig, sig->posix_timer_id)) {
149 hlist_add_head_rcu(&timer->t_hash, head);
150 ret = sig->posix_timer_id;
151 }
152 if (++sig->posix_timer_id < 0)
153 sig->posix_timer_id = 0;
154 if ((sig->posix_timer_id == first_free_id) && (ret == -ENOENT))
155 /* Loop over all possible ids completed */
156 ret = -EAGAIN;
157 spin_unlock(&hash_lock);
158 } while (ret == -ENOENT);
159 return ret;
160}
161
162static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
163{
164 spin_unlock_irqrestore(&timr->it_lock, flags);
165}
166
167/* Get clock_realtime */
168static int posix_clock_realtime_get(clockid_t which_clock, struct timespec64 *tp)
169{
170 ktime_get_real_ts64(tp);
171 return 0;
172}
173
174/* Set clock_realtime */
175static int posix_clock_realtime_set(const clockid_t which_clock,
176 const struct timespec64 *tp)
177{
178 return do_sys_settimeofday64(tp, NULL);
179}
180
181static int posix_clock_realtime_adj(const clockid_t which_clock,
182 struct __kernel_timex *t)
183{
184 return do_adjtimex(t);
185}
186
187/*
188 * Get monotonic time for posix timers
189 */
190static int posix_ktime_get_ts(clockid_t which_clock, struct timespec64 *tp)
191{
192 ktime_get_ts64(tp);
193 return 0;
194}
195
196/*
197 * Get monotonic-raw time for posix timers
198 */
199static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
200{
201 ktime_get_raw_ts64(tp);
202 return 0;
203}
204
205
206static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
207{
208 ktime_get_coarse_real_ts64(tp);
209 return 0;
210}
211
212static int posix_get_monotonic_coarse(clockid_t which_clock,
213 struct timespec64 *tp)
214{
215 ktime_get_coarse_ts64(tp);
216 return 0;
217}
218
219static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
220{
221 *tp = ktime_to_timespec64(KTIME_LOW_RES);
222 return 0;
223}
224
225static int posix_get_boottime(const clockid_t which_clock, struct timespec64 *tp)
226{
227 ktime_get_boottime_ts64(tp);
228 return 0;
229}
230
231static int posix_get_tai(clockid_t which_clock, struct timespec64 *tp)
232{
233 ktime_get_clocktai_ts64(tp);
234 return 0;
235}
236
237static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
238{
239 tp->tv_sec = 0;
240 tp->tv_nsec = hrtimer_resolution;
241 return 0;
242}
243
244/*
245 * Initialize everything, well, just everything in Posix clocks/timers ;)
246 */
247static __init int init_posix_timers(void)
248{
249 posix_timers_cache = kmem_cache_create("posix_timers_cache",
250 sizeof (struct k_itimer), 0, SLAB_PANIC,
251 NULL);
252 return 0;
253}
254__initcall(init_posix_timers);
255
256/*
257 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
258 * are of type int. Clamp the overrun value to INT_MAX
259 */
260static inline int timer_overrun_to_int(struct k_itimer *timr, int baseval)
261{
262 s64 sum = timr->it_overrun_last + (s64)baseval;
263
264 return sum > (s64)INT_MAX ? INT_MAX : (int)sum;
265}
266
267static void common_hrtimer_rearm(struct k_itimer *timr)
268{
269 struct hrtimer *timer = &timr->it.real.timer;
270
271 timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
272 timr->it_interval);
273 hrtimer_restart(timer);
274}
275
276/*
277 * This function is exported for use by the signal deliver code. It is
278 * called just prior to the info block being released and passes that
279 * block to us. It's function is to update the overrun entry AND to
280 * restart the timer. It should only be called if the timer is to be
281 * restarted (i.e. we have flagged this in the sys_private entry of the
282 * info block).
283 *
284 * To protect against the timer going away while the interrupt is queued,
285 * we require that the it_requeue_pending flag be set.
286 */
287void posixtimer_rearm(struct kernel_siginfo *info)
288{
289 struct k_itimer *timr;
290 unsigned long flags;
291
292 timr = lock_timer(info->si_tid, &flags);
293 if (!timr)
294 return;
295
296 if (timr->it_interval && timr->it_requeue_pending == info->si_sys_private) {
297 timr->kclock->timer_rearm(timr);
298
299 timr->it_active = 1;
300 timr->it_overrun_last = timr->it_overrun;
301 timr->it_overrun = -1LL;
302 ++timr->it_requeue_pending;
303
304 info->si_overrun = timer_overrun_to_int(timr, info->si_overrun);
305 }
306
307 unlock_timer(timr, flags);
308}
309
310int posix_timer_event(struct k_itimer *timr, int si_private)
311{
312 enum pid_type type;
313 int ret = -1;
314 /*
315 * FIXME: if ->sigq is queued we can race with
316 * dequeue_signal()->posixtimer_rearm().
317 *
318 * If dequeue_signal() sees the "right" value of
319 * si_sys_private it calls posixtimer_rearm().
320 * We re-queue ->sigq and drop ->it_lock().
321 * posixtimer_rearm() locks the timer
322 * and re-schedules it while ->sigq is pending.
323 * Not really bad, but not that we want.
324 */
325 timr->sigq->info.si_sys_private = si_private;
326
327 type = !(timr->it_sigev_notify & SIGEV_THREAD_ID) ? PIDTYPE_TGID : PIDTYPE_PID;
328 ret = send_sigqueue(timr->sigq, timr->it_pid, type);
329 /* If we failed to send the signal the timer stops. */
330 return ret > 0;
331}
332
333/*
334 * This function gets called when a POSIX.1b interval timer expires. It
335 * is used as a callback from the kernel internal timer. The
336 * run_timer_list code ALWAYS calls with interrupts on.
337
338 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
339 */
340static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
341{
342 struct k_itimer *timr;
343 unsigned long flags;
344 int si_private = 0;
345 enum hrtimer_restart ret = HRTIMER_NORESTART;
346
347 timr = container_of(timer, struct k_itimer, it.real.timer);
348 spin_lock_irqsave(&timr->it_lock, flags);
349
350 timr->it_active = 0;
351 if (timr->it_interval != 0)
352 si_private = ++timr->it_requeue_pending;
353
354 if (posix_timer_event(timr, si_private)) {
355 /*
356 * signal was not sent because of sig_ignor
357 * we will not get a call back to restart it AND
358 * it should be restarted.
359 */
360 if (timr->it_interval != 0) {
361 ktime_t now = hrtimer_cb_get_time(timer);
362
363 /*
364 * FIXME: What we really want, is to stop this
365 * timer completely and restart it in case the
366 * SIG_IGN is removed. This is a non trivial
367 * change which involves sighand locking
368 * (sigh !), which we don't want to do late in
369 * the release cycle.
370 *
371 * For now we just let timers with an interval
372 * less than a jiffie expire every jiffie to
373 * avoid softirq starvation in case of SIG_IGN
374 * and a very small interval, which would put
375 * the timer right back on the softirq pending
376 * list. By moving now ahead of time we trick
377 * hrtimer_forward() to expire the timer
378 * later, while we still maintain the overrun
379 * accuracy, but have some inconsistency in
380 * the timer_gettime() case. This is at least
381 * better than a starved softirq. A more
382 * complex fix which solves also another related
383 * inconsistency is already in the pipeline.
384 */
385#ifdef CONFIG_HIGH_RES_TIMERS
386 {
387 ktime_t kj = NSEC_PER_SEC / HZ;
388
389 if (timr->it_interval < kj)
390 now = ktime_add(now, kj);
391 }
392#endif
393 timr->it_overrun += hrtimer_forward(timer, now,
394 timr->it_interval);
395 ret = HRTIMER_RESTART;
396 ++timr->it_requeue_pending;
397 timr->it_active = 1;
398 }
399 }
400
401 unlock_timer(timr, flags);
402 return ret;
403}
404
405static struct pid *good_sigevent(sigevent_t * event)
406{
407 struct pid *pid = task_tgid(current);
408 struct task_struct *rtn;
409
410 switch (event->sigev_notify) {
411 case SIGEV_SIGNAL | SIGEV_THREAD_ID:
412 pid = find_vpid(event->sigev_notify_thread_id);
413 rtn = pid_task(pid, PIDTYPE_PID);
414 if (!rtn || !same_thread_group(rtn, current))
415 return NULL;
416 /* FALLTHRU */
417 case SIGEV_SIGNAL:
418 case SIGEV_THREAD:
419 if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
420 return NULL;
421 /* FALLTHRU */
422 case SIGEV_NONE:
423 return pid;
424 default:
425 return NULL;
426 }
427}
428
429static struct k_itimer * alloc_posix_timer(void)
430{
431 struct k_itimer *tmr;
432 tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
433 if (!tmr)
434 return tmr;
435 if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
436 kmem_cache_free(posix_timers_cache, tmr);
437 return NULL;
438 }
439 clear_siginfo(&tmr->sigq->info);
440 return tmr;
441}
442
443static void k_itimer_rcu_free(struct rcu_head *head)
444{
445 struct k_itimer *tmr = container_of(head, struct k_itimer, rcu);
446
447 kmem_cache_free(posix_timers_cache, tmr);
448}
449
450#define IT_ID_SET 1
451#define IT_ID_NOT_SET 0
452static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
453{
454 if (it_id_set) {
455 unsigned long flags;
456 spin_lock_irqsave(&hash_lock, flags);
457 hlist_del_rcu(&tmr->t_hash);
458 spin_unlock_irqrestore(&hash_lock, flags);
459 }
460 put_pid(tmr->it_pid);
461 sigqueue_free(tmr->sigq);
462 call_rcu(&tmr->rcu, k_itimer_rcu_free);
463}
464
465static int common_timer_create(struct k_itimer *new_timer)
466{
467 hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
468 return 0;
469}
470
471/* Create a POSIX.1b interval timer. */
472static int do_timer_create(clockid_t which_clock, struct sigevent *event,
473 timer_t __user *created_timer_id)
474{
475 const struct k_clock *kc = clockid_to_kclock(which_clock);
476 struct k_itimer *new_timer;
477 int error, new_timer_id;
478 int it_id_set = IT_ID_NOT_SET;
479
480 if (!kc)
481 return -EINVAL;
482 if (!kc->timer_create)
483 return -EOPNOTSUPP;
484
485 new_timer = alloc_posix_timer();
486 if (unlikely(!new_timer))
487 return -EAGAIN;
488
489 spin_lock_init(&new_timer->it_lock);
490 new_timer_id = posix_timer_add(new_timer);
491 if (new_timer_id < 0) {
492 error = new_timer_id;
493 goto out;
494 }
495
496 it_id_set = IT_ID_SET;
497 new_timer->it_id = (timer_t) new_timer_id;
498 new_timer->it_clock = which_clock;
499 new_timer->kclock = kc;
500 new_timer->it_overrun = -1LL;
501
502 if (event) {
503 rcu_read_lock();
504 new_timer->it_pid = get_pid(good_sigevent(event));
505 rcu_read_unlock();
506 if (!new_timer->it_pid) {
507 error = -EINVAL;
508 goto out;
509 }
510 new_timer->it_sigev_notify = event->sigev_notify;
511 new_timer->sigq->info.si_signo = event->sigev_signo;
512 new_timer->sigq->info.si_value = event->sigev_value;
513 } else {
514 new_timer->it_sigev_notify = SIGEV_SIGNAL;
515 new_timer->sigq->info.si_signo = SIGALRM;
516 memset(&new_timer->sigq->info.si_value, 0, sizeof(sigval_t));
517 new_timer->sigq->info.si_value.sival_int = new_timer->it_id;
518 new_timer->it_pid = get_pid(task_tgid(current));
519 }
520
521 new_timer->sigq->info.si_tid = new_timer->it_id;
522 new_timer->sigq->info.si_code = SI_TIMER;
523
524 if (copy_to_user(created_timer_id,
525 &new_timer_id, sizeof (new_timer_id))) {
526 error = -EFAULT;
527 goto out;
528 }
529
530 error = kc->timer_create(new_timer);
531 if (error)
532 goto out;
533
534 spin_lock_irq(¤t->sighand->siglock);
535 new_timer->it_signal = current->signal;
536 list_add(&new_timer->list, ¤t->signal->posix_timers);
537 spin_unlock_irq(¤t->sighand->siglock);
538
539 return 0;
540 /*
541 * In the case of the timer belonging to another task, after
542 * the task is unlocked, the timer is owned by the other task
543 * and may cease to exist at any time. Don't use or modify
544 * new_timer after the unlock call.
545 */
546out:
547 release_posix_timer(new_timer, it_id_set);
548 return error;
549}
550
551SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
552 struct sigevent __user *, timer_event_spec,
553 timer_t __user *, created_timer_id)
554{
555 if (timer_event_spec) {
556 sigevent_t event;
557
558 if (copy_from_user(&event, timer_event_spec, sizeof (event)))
559 return -EFAULT;
560 return do_timer_create(which_clock, &event, created_timer_id);
561 }
562 return do_timer_create(which_clock, NULL, created_timer_id);
563}
564
565#ifdef CONFIG_COMPAT
566COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
567 struct compat_sigevent __user *, timer_event_spec,
568 timer_t __user *, created_timer_id)
569{
570 if (timer_event_spec) {
571 sigevent_t event;
572
573 if (get_compat_sigevent(&event, timer_event_spec))
574 return -EFAULT;
575 return do_timer_create(which_clock, &event, created_timer_id);
576 }
577 return do_timer_create(which_clock, NULL, created_timer_id);
578}
579#endif
580
581/*
582 * Locking issues: We need to protect the result of the id look up until
583 * we get the timer locked down so it is not deleted under us. The
584 * removal is done under the idr spinlock so we use that here to bridge
585 * the find to the timer lock. To avoid a dead lock, the timer id MUST
586 * be release with out holding the timer lock.
587 */
588static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
589{
590 struct k_itimer *timr;
591
592 /*
593 * timer_t could be any type >= int and we want to make sure any
594 * @timer_id outside positive int range fails lookup.
595 */
596 if ((unsigned long long)timer_id > INT_MAX)
597 return NULL;
598
599 rcu_read_lock();
600 timr = posix_timer_by_id(timer_id);
601 if (timr) {
602 spin_lock_irqsave(&timr->it_lock, *flags);
603 if (timr->it_signal == current->signal) {
604 rcu_read_unlock();
605 return timr;
606 }
607 spin_unlock_irqrestore(&timr->it_lock, *flags);
608 }
609 rcu_read_unlock();
610
611 return NULL;
612}
613
614static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
615{
616 struct hrtimer *timer = &timr->it.real.timer;
617
618 return __hrtimer_expires_remaining_adjusted(timer, now);
619}
620
621static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
622{
623 struct hrtimer *timer = &timr->it.real.timer;
624
625 return hrtimer_forward(timer, now, timr->it_interval);
626}
627
628/*
629 * Get the time remaining on a POSIX.1b interval timer. This function
630 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
631 * mess with irq.
632 *
633 * We have a couple of messes to clean up here. First there is the case
634 * of a timer that has a requeue pending. These timers should appear to
635 * be in the timer list with an expiry as if we were to requeue them
636 * now.
637 *
638 * The second issue is the SIGEV_NONE timer which may be active but is
639 * not really ever put in the timer list (to save system resources).
640 * This timer may be expired, and if so, we will do it here. Otherwise
641 * it is the same as a requeue pending timer WRT to what we should
642 * report.
643 */
644void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
645{
646 const struct k_clock *kc = timr->kclock;
647 ktime_t now, remaining, iv;
648 struct timespec64 ts64;
649 bool sig_none;
650
651 sig_none = timr->it_sigev_notify == SIGEV_NONE;
652 iv = timr->it_interval;
653
654 /* interval timer ? */
655 if (iv) {
656 cur_setting->it_interval = ktime_to_timespec64(iv);
657 } else if (!timr->it_active) {
658 /*
659 * SIGEV_NONE oneshot timers are never queued. Check them
660 * below.
661 */
662 if (!sig_none)
663 return;
664 }
665
666 /*
667 * The timespec64 based conversion is suboptimal, but it's not
668 * worth to implement yet another callback.
669 */
670 kc->clock_get(timr->it_clock, &ts64);
671 now = timespec64_to_ktime(ts64);
672
673 /*
674 * When a requeue is pending or this is a SIGEV_NONE timer move the
675 * expiry time forward by intervals, so expiry is > now.
676 */
677 if (iv && (timr->it_requeue_pending & REQUEUE_PENDING || sig_none))
678 timr->it_overrun += kc->timer_forward(timr, now);
679
680 remaining = kc->timer_remaining(timr, now);
681 /* Return 0 only, when the timer is expired and not pending */
682 if (remaining <= 0) {
683 /*
684 * A single shot SIGEV_NONE timer must return 0, when
685 * it is expired !
686 */
687 if (!sig_none)
688 cur_setting->it_value.tv_nsec = 1;
689 } else {
690 cur_setting->it_value = ktime_to_timespec64(remaining);
691 }
692}
693
694/* Get the time remaining on a POSIX.1b interval timer. */
695static int do_timer_gettime(timer_t timer_id, struct itimerspec64 *setting)
696{
697 struct k_itimer *timr;
698 const struct k_clock *kc;
699 unsigned long flags;
700 int ret = 0;
701
702 timr = lock_timer(timer_id, &flags);
703 if (!timr)
704 return -EINVAL;
705
706 memset(setting, 0, sizeof(*setting));
707 kc = timr->kclock;
708 if (WARN_ON_ONCE(!kc || !kc->timer_get))
709 ret = -EINVAL;
710 else
711 kc->timer_get(timr, setting);
712
713 unlock_timer(timr, flags);
714 return ret;
715}
716
717/* Get the time remaining on a POSIX.1b interval timer. */
718SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
719 struct __kernel_itimerspec __user *, setting)
720{
721 struct itimerspec64 cur_setting;
722
723 int ret = do_timer_gettime(timer_id, &cur_setting);
724 if (!ret) {
725 if (put_itimerspec64(&cur_setting, setting))
726 ret = -EFAULT;
727 }
728 return ret;
729}
730
731#ifdef CONFIG_COMPAT_32BIT_TIME
732
733SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
734 struct old_itimerspec32 __user *, setting)
735{
736 struct itimerspec64 cur_setting;
737
738 int ret = do_timer_gettime(timer_id, &cur_setting);
739 if (!ret) {
740 if (put_old_itimerspec32(&cur_setting, setting))
741 ret = -EFAULT;
742 }
743 return ret;
744}
745
746#endif
747
748/*
749 * Get the number of overruns of a POSIX.1b interval timer. This is to
750 * be the overrun of the timer last delivered. At the same time we are
751 * accumulating overruns on the next timer. The overrun is frozen when
752 * the signal is delivered, either at the notify time (if the info block
753 * is not queued) or at the actual delivery time (as we are informed by
754 * the call back to posixtimer_rearm(). So all we need to do is
755 * to pick up the frozen overrun.
756 */
757SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
758{
759 struct k_itimer *timr;
760 int overrun;
761 unsigned long flags;
762
763 timr = lock_timer(timer_id, &flags);
764 if (!timr)
765 return -EINVAL;
766
767 overrun = timer_overrun_to_int(timr, 0);
768 unlock_timer(timr, flags);
769
770 return overrun;
771}
772
773static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
774 bool absolute, bool sigev_none)
775{
776 struct hrtimer *timer = &timr->it.real.timer;
777 enum hrtimer_mode mode;
778
779 mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
780 /*
781 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
782 * clock modifications, so they become CLOCK_MONOTONIC based under the
783 * hood. See hrtimer_init(). Update timr->kclock, so the generic
784 * functions which use timr->kclock->clock_get() work.
785 *
786 * Note: it_clock stays unmodified, because the next timer_set() might
787 * use ABSTIME, so it needs to switch back.
788 */
789 if (timr->it_clock == CLOCK_REALTIME)
790 timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
791
792 hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
793 timr->it.real.timer.function = posix_timer_fn;
794
795 if (!absolute)
796 expires = ktime_add_safe(expires, timer->base->get_time());
797 hrtimer_set_expires(timer, expires);
798
799 if (!sigev_none)
800 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
801}
802
803static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
804{
805 return hrtimer_try_to_cancel(&timr->it.real.timer);
806}
807
808static void common_timer_wait_running(struct k_itimer *timer)
809{
810 hrtimer_cancel_wait_running(&timer->it.real.timer);
811}
812
813/*
814 * On PREEMPT_RT this prevent priority inversion against softirq kthread in
815 * case it gets preempted while executing a timer callback. See comments in
816 * hrtimer_cancel_wait_running. For PREEMPT_RT=n this just results in a
817 * cpu_relax().
818 */
819static struct k_itimer *timer_wait_running(struct k_itimer *timer,
820 unsigned long *flags)
821{
822 const struct k_clock *kc = READ_ONCE(timer->kclock);
823 timer_t timer_id = READ_ONCE(timer->it_id);
824
825 /* Prevent kfree(timer) after dropping the lock */
826 rcu_read_lock();
827 unlock_timer(timer, *flags);
828
829 if (!WARN_ON_ONCE(!kc->timer_wait_running))
830 kc->timer_wait_running(timer);
831
832 rcu_read_unlock();
833 /* Relock the timer. It might be not longer hashed. */
834 return lock_timer(timer_id, flags);
835}
836
837/* Set a POSIX.1b interval timer. */
838int common_timer_set(struct k_itimer *timr, int flags,
839 struct itimerspec64 *new_setting,
840 struct itimerspec64 *old_setting)
841{
842 const struct k_clock *kc = timr->kclock;
843 bool sigev_none;
844 ktime_t expires;
845
846 if (old_setting)
847 common_timer_get(timr, old_setting);
848
849 /* Prevent rearming by clearing the interval */
850 timr->it_interval = 0;
851 /*
852 * Careful here. On SMP systems the timer expiry function could be
853 * active and spinning on timr->it_lock.
854 */
855 if (kc->timer_try_to_cancel(timr) < 0)
856 return TIMER_RETRY;
857
858 timr->it_active = 0;
859 timr->it_requeue_pending = (timr->it_requeue_pending + 2) &
860 ~REQUEUE_PENDING;
861 timr->it_overrun_last = 0;
862
863 /* Switch off the timer when it_value is zero */
864 if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
865 return 0;
866
867 timr->it_interval = timespec64_to_ktime(new_setting->it_interval);
868 expires = timespec64_to_ktime(new_setting->it_value);
869 sigev_none = timr->it_sigev_notify == SIGEV_NONE;
870
871 kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
872 timr->it_active = !sigev_none;
873 return 0;
874}
875
876static int do_timer_settime(timer_t timer_id, int tmr_flags,
877 struct itimerspec64 *new_spec64,
878 struct itimerspec64 *old_spec64)
879{
880 const struct k_clock *kc;
881 struct k_itimer *timr;
882 unsigned long flags;
883 int error = 0;
884
885 if (!timespec64_valid(&new_spec64->it_interval) ||
886 !timespec64_valid(&new_spec64->it_value))
887 return -EINVAL;
888
889 if (old_spec64)
890 memset(old_spec64, 0, sizeof(*old_spec64));
891
892 timr = lock_timer(timer_id, &flags);
893retry:
894 if (!timr)
895 return -EINVAL;
896
897 kc = timr->kclock;
898 if (WARN_ON_ONCE(!kc || !kc->timer_set))
899 error = -EINVAL;
900 else
901 error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
902
903 if (error == TIMER_RETRY) {
904 // We already got the old time...
905 old_spec64 = NULL;
906 /* Unlocks and relocks the timer if it still exists */
907 timr = timer_wait_running(timr, &flags);
908 goto retry;
909 }
910 unlock_timer(timr, flags);
911
912 return error;
913}
914
915/* Set a POSIX.1b interval timer */
916SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
917 const struct __kernel_itimerspec __user *, new_setting,
918 struct __kernel_itimerspec __user *, old_setting)
919{
920 struct itimerspec64 new_spec, old_spec;
921 struct itimerspec64 *rtn = old_setting ? &old_spec : NULL;
922 int error = 0;
923
924 if (!new_setting)
925 return -EINVAL;
926
927 if (get_itimerspec64(&new_spec, new_setting))
928 return -EFAULT;
929
930 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
931 if (!error && old_setting) {
932 if (put_itimerspec64(&old_spec, old_setting))
933 error = -EFAULT;
934 }
935 return error;
936}
937
938#ifdef CONFIG_COMPAT_32BIT_TIME
939SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
940 struct old_itimerspec32 __user *, new,
941 struct old_itimerspec32 __user *, old)
942{
943 struct itimerspec64 new_spec, old_spec;
944 struct itimerspec64 *rtn = old ? &old_spec : NULL;
945 int error = 0;
946
947 if (!new)
948 return -EINVAL;
949 if (get_old_itimerspec32(&new_spec, new))
950 return -EFAULT;
951
952 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
953 if (!error && old) {
954 if (put_old_itimerspec32(&old_spec, old))
955 error = -EFAULT;
956 }
957 return error;
958}
959#endif
960
961int common_timer_del(struct k_itimer *timer)
962{
963 const struct k_clock *kc = timer->kclock;
964
965 timer->it_interval = 0;
966 if (kc->timer_try_to_cancel(timer) < 0)
967 return TIMER_RETRY;
968 timer->it_active = 0;
969 return 0;
970}
971
972static inline int timer_delete_hook(struct k_itimer *timer)
973{
974 const struct k_clock *kc = timer->kclock;
975
976 if (WARN_ON_ONCE(!kc || !kc->timer_del))
977 return -EINVAL;
978 return kc->timer_del(timer);
979}
980
981/* Delete a POSIX.1b interval timer. */
982SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
983{
984 struct k_itimer *timer;
985 unsigned long flags;
986
987 timer = lock_timer(timer_id, &flags);
988
989retry_delete:
990 if (!timer)
991 return -EINVAL;
992
993 if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
994 /* Unlocks and relocks the timer if it still exists */
995 timer = timer_wait_running(timer, &flags);
996 goto retry_delete;
997 }
998
999 spin_lock(¤t->sighand->siglock);
1000 list_del(&timer->list);
1001 spin_unlock(¤t->sighand->siglock);
1002 /*
1003 * This keeps any tasks waiting on the spin lock from thinking
1004 * they got something (see the lock code above).
1005 */
1006 timer->it_signal = NULL;
1007
1008 unlock_timer(timer, flags);
1009 release_posix_timer(timer, IT_ID_SET);
1010 return 0;
1011}
1012
1013/*
1014 * return timer owned by the process, used by exit_itimers
1015 */
1016static void itimer_delete(struct k_itimer *timer)
1017{
1018retry_delete:
1019 spin_lock_irq(&timer->it_lock);
1020
1021 if (timer_delete_hook(timer) == TIMER_RETRY) {
1022 spin_unlock_irq(&timer->it_lock);
1023 goto retry_delete;
1024 }
1025 list_del(&timer->list);
1026
1027 spin_unlock_irq(&timer->it_lock);
1028 release_posix_timer(timer, IT_ID_SET);
1029}
1030
1031/*
1032 * This is called by do_exit or de_thread, only when there are no more
1033 * references to the shared signal_struct.
1034 */
1035void exit_itimers(struct signal_struct *sig)
1036{
1037 struct k_itimer *tmr;
1038
1039 while (!list_empty(&sig->posix_timers)) {
1040 tmr = list_entry(sig->posix_timers.next, struct k_itimer, list);
1041 itimer_delete(tmr);
1042 }
1043}
1044
1045SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1046 const struct __kernel_timespec __user *, tp)
1047{
1048 const struct k_clock *kc = clockid_to_kclock(which_clock);
1049 struct timespec64 new_tp;
1050
1051 if (!kc || !kc->clock_set)
1052 return -EINVAL;
1053
1054 if (get_timespec64(&new_tp, tp))
1055 return -EFAULT;
1056
1057 return kc->clock_set(which_clock, &new_tp);
1058}
1059
1060SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1061 struct __kernel_timespec __user *, tp)
1062{
1063 const struct k_clock *kc = clockid_to_kclock(which_clock);
1064 struct timespec64 kernel_tp;
1065 int error;
1066
1067 if (!kc)
1068 return -EINVAL;
1069
1070 error = kc->clock_get(which_clock, &kernel_tp);
1071
1072 if (!error && put_timespec64(&kernel_tp, tp))
1073 error = -EFAULT;
1074
1075 return error;
1076}
1077
1078int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1079{
1080 const struct k_clock *kc = clockid_to_kclock(which_clock);
1081
1082 if (!kc)
1083 return -EINVAL;
1084 if (!kc->clock_adj)
1085 return -EOPNOTSUPP;
1086
1087 return kc->clock_adj(which_clock, ktx);
1088}
1089
1090SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1091 struct __kernel_timex __user *, utx)
1092{
1093 struct __kernel_timex ktx;
1094 int err;
1095
1096 if (copy_from_user(&ktx, utx, sizeof(ktx)))
1097 return -EFAULT;
1098
1099 err = do_clock_adjtime(which_clock, &ktx);
1100
1101 if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1102 return -EFAULT;
1103
1104 return err;
1105}
1106
1107SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1108 struct __kernel_timespec __user *, tp)
1109{
1110 const struct k_clock *kc = clockid_to_kclock(which_clock);
1111 struct timespec64 rtn_tp;
1112 int error;
1113
1114 if (!kc)
1115 return -EINVAL;
1116
1117 error = kc->clock_getres(which_clock, &rtn_tp);
1118
1119 if (!error && tp && put_timespec64(&rtn_tp, tp))
1120 error = -EFAULT;
1121
1122 return error;
1123}
1124
1125#ifdef CONFIG_COMPAT_32BIT_TIME
1126
1127SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1128 struct old_timespec32 __user *, tp)
1129{
1130 const struct k_clock *kc = clockid_to_kclock(which_clock);
1131 struct timespec64 ts;
1132
1133 if (!kc || !kc->clock_set)
1134 return -EINVAL;
1135
1136 if (get_old_timespec32(&ts, tp))
1137 return -EFAULT;
1138
1139 return kc->clock_set(which_clock, &ts);
1140}
1141
1142SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1143 struct old_timespec32 __user *, tp)
1144{
1145 const struct k_clock *kc = clockid_to_kclock(which_clock);
1146 struct timespec64 ts;
1147 int err;
1148
1149 if (!kc)
1150 return -EINVAL;
1151
1152 err = kc->clock_get(which_clock, &ts);
1153
1154 if (!err && put_old_timespec32(&ts, tp))
1155 err = -EFAULT;
1156
1157 return err;
1158}
1159
1160SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1161 struct old_timex32 __user *, utp)
1162{
1163 struct __kernel_timex ktx;
1164 int err;
1165
1166 err = get_old_timex32(&ktx, utp);
1167 if (err)
1168 return err;
1169
1170 err = do_clock_adjtime(which_clock, &ktx);
1171
1172 if (err >= 0)
1173 err = put_old_timex32(utp, &ktx);
1174
1175 return err;
1176}
1177
1178SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1179 struct old_timespec32 __user *, tp)
1180{
1181 const struct k_clock *kc = clockid_to_kclock(which_clock);
1182 struct timespec64 ts;
1183 int err;
1184
1185 if (!kc)
1186 return -EINVAL;
1187
1188 err = kc->clock_getres(which_clock, &ts);
1189 if (!err && tp && put_old_timespec32(&ts, tp))
1190 return -EFAULT;
1191
1192 return err;
1193}
1194
1195#endif
1196
1197/*
1198 * nanosleep for monotonic and realtime clocks
1199 */
1200static int common_nsleep(const clockid_t which_clock, int flags,
1201 const struct timespec64 *rqtp)
1202{
1203 return hrtimer_nanosleep(rqtp, flags & TIMER_ABSTIME ?
1204 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1205 which_clock);
1206}
1207
1208SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1209 const struct __kernel_timespec __user *, rqtp,
1210 struct __kernel_timespec __user *, rmtp)
1211{
1212 const struct k_clock *kc = clockid_to_kclock(which_clock);
1213 struct timespec64 t;
1214
1215 if (!kc)
1216 return -EINVAL;
1217 if (!kc->nsleep)
1218 return -EOPNOTSUPP;
1219
1220 if (get_timespec64(&t, rqtp))
1221 return -EFAULT;
1222
1223 if (!timespec64_valid(&t))
1224 return -EINVAL;
1225 if (flags & TIMER_ABSTIME)
1226 rmtp = NULL;
1227 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1228 current->restart_block.nanosleep.rmtp = rmtp;
1229
1230 return kc->nsleep(which_clock, flags, &t);
1231}
1232
1233#ifdef CONFIG_COMPAT_32BIT_TIME
1234
1235SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1236 struct old_timespec32 __user *, rqtp,
1237 struct old_timespec32 __user *, rmtp)
1238{
1239 const struct k_clock *kc = clockid_to_kclock(which_clock);
1240 struct timespec64 t;
1241
1242 if (!kc)
1243 return -EINVAL;
1244 if (!kc->nsleep)
1245 return -EOPNOTSUPP;
1246
1247 if (get_old_timespec32(&t, rqtp))
1248 return -EFAULT;
1249
1250 if (!timespec64_valid(&t))
1251 return -EINVAL;
1252 if (flags & TIMER_ABSTIME)
1253 rmtp = NULL;
1254 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1255 current->restart_block.nanosleep.compat_rmtp = rmtp;
1256
1257 return kc->nsleep(which_clock, flags, &t);
1258}
1259
1260#endif
1261
1262static const struct k_clock clock_realtime = {
1263 .clock_getres = posix_get_hrtimer_res,
1264 .clock_get = posix_clock_realtime_get,
1265 .clock_set = posix_clock_realtime_set,
1266 .clock_adj = posix_clock_realtime_adj,
1267 .nsleep = common_nsleep,
1268 .timer_create = common_timer_create,
1269 .timer_set = common_timer_set,
1270 .timer_get = common_timer_get,
1271 .timer_del = common_timer_del,
1272 .timer_rearm = common_hrtimer_rearm,
1273 .timer_forward = common_hrtimer_forward,
1274 .timer_remaining = common_hrtimer_remaining,
1275 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1276 .timer_wait_running = common_timer_wait_running,
1277 .timer_arm = common_hrtimer_arm,
1278};
1279
1280static const struct k_clock clock_monotonic = {
1281 .clock_getres = posix_get_hrtimer_res,
1282 .clock_get = posix_ktime_get_ts,
1283 .nsleep = common_nsleep,
1284 .timer_create = common_timer_create,
1285 .timer_set = common_timer_set,
1286 .timer_get = common_timer_get,
1287 .timer_del = common_timer_del,
1288 .timer_rearm = common_hrtimer_rearm,
1289 .timer_forward = common_hrtimer_forward,
1290 .timer_remaining = common_hrtimer_remaining,
1291 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1292 .timer_wait_running = common_timer_wait_running,
1293 .timer_arm = common_hrtimer_arm,
1294};
1295
1296static const struct k_clock clock_monotonic_raw = {
1297 .clock_getres = posix_get_hrtimer_res,
1298 .clock_get = posix_get_monotonic_raw,
1299};
1300
1301static const struct k_clock clock_realtime_coarse = {
1302 .clock_getres = posix_get_coarse_res,
1303 .clock_get = posix_get_realtime_coarse,
1304};
1305
1306static const struct k_clock clock_monotonic_coarse = {
1307 .clock_getres = posix_get_coarse_res,
1308 .clock_get = posix_get_monotonic_coarse,
1309};
1310
1311static const struct k_clock clock_tai = {
1312 .clock_getres = posix_get_hrtimer_res,
1313 .clock_get = posix_get_tai,
1314 .nsleep = common_nsleep,
1315 .timer_create = common_timer_create,
1316 .timer_set = common_timer_set,
1317 .timer_get = common_timer_get,
1318 .timer_del = common_timer_del,
1319 .timer_rearm = common_hrtimer_rearm,
1320 .timer_forward = common_hrtimer_forward,
1321 .timer_remaining = common_hrtimer_remaining,
1322 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1323 .timer_wait_running = common_timer_wait_running,
1324 .timer_arm = common_hrtimer_arm,
1325};
1326
1327static const struct k_clock clock_boottime = {
1328 .clock_getres = posix_get_hrtimer_res,
1329 .clock_get = posix_get_boottime,
1330 .nsleep = common_nsleep,
1331 .timer_create = common_timer_create,
1332 .timer_set = common_timer_set,
1333 .timer_get = common_timer_get,
1334 .timer_del = common_timer_del,
1335 .timer_rearm = common_hrtimer_rearm,
1336 .timer_forward = common_hrtimer_forward,
1337 .timer_remaining = common_hrtimer_remaining,
1338 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1339 .timer_wait_running = common_timer_wait_running,
1340 .timer_arm = common_hrtimer_arm,
1341};
1342
1343static const struct k_clock * const posix_clocks[] = {
1344 [CLOCK_REALTIME] = &clock_realtime,
1345 [CLOCK_MONOTONIC] = &clock_monotonic,
1346 [CLOCK_PROCESS_CPUTIME_ID] = &clock_process,
1347 [CLOCK_THREAD_CPUTIME_ID] = &clock_thread,
1348 [CLOCK_MONOTONIC_RAW] = &clock_monotonic_raw,
1349 [CLOCK_REALTIME_COARSE] = &clock_realtime_coarse,
1350 [CLOCK_MONOTONIC_COARSE] = &clock_monotonic_coarse,
1351 [CLOCK_BOOTTIME] = &clock_boottime,
1352 [CLOCK_REALTIME_ALARM] = &alarm_clock,
1353 [CLOCK_BOOTTIME_ALARM] = &alarm_clock,
1354 [CLOCK_TAI] = &clock_tai,
1355};
1356
1357static const struct k_clock *clockid_to_kclock(const clockid_t id)
1358{
1359 clockid_t idx = id;
1360
1361 if (id < 0) {
1362 return (id & CLOCKFD_MASK) == CLOCKFD ?
1363 &clock_posix_dynamic : &clock_posix_cpu;
1364 }
1365
1366 if (id >= ARRAY_SIZE(posix_clocks))
1367 return NULL;
1368
1369 return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1370}