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/*
2 * linux/kernel/posix-timers.c
3 *
4 *
5 * 2002-10-15 Posix Clocks & timers
6 * by George Anzinger george@mvista.com
7 *
8 * Copyright (C) 2002 2003 by MontaVista Software.
9 *
10 * 2004-06-01 Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug.
11 * Copyright (C) 2004 Boris Hu
12 *
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or (at
16 * your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful, but
19 * WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * General Public License for more details.
22
23 * You should have received a copy of the GNU General Public License
24 * along with this program; if not, write to the Free Software
25 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 *
27 * MontaVista Software | 1237 East Arques Avenue | Sunnyvale | CA 94085 | USA
28 */
29
30/* These are all the functions necessary to implement
31 * POSIX clocks & timers
32 */
33#include <linux/mm.h>
34#include <linux/interrupt.h>
35#include <linux/slab.h>
36#include <linux/time.h>
37#include <linux/mutex.h>
38
39#include <asm/uaccess.h>
40#include <linux/list.h>
41#include <linux/init.h>
42#include <linux/compiler.h>
43#include <linux/hash.h>
44#include <linux/posix-clock.h>
45#include <linux/posix-timers.h>
46#include <linux/syscalls.h>
47#include <linux/wait.h>
48#include <linux/workqueue.h>
49#include <linux/export.h>
50#include <linux/hashtable.h>
51
52#include "timekeeping.h"
53
54/*
55 * Management arrays for POSIX timers. Timers are now kept in static hash table
56 * with 512 entries.
57 * Timer ids are allocated by local routine, which selects proper hash head by
58 * key, constructed from current->signal address and per signal struct counter.
59 * This keeps timer ids unique per process, but now they can intersect between
60 * processes.
61 */
62
63/*
64 * Lets keep our timers in a slab cache :-)
65 */
66static struct kmem_cache *posix_timers_cache;
67
68static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
69static DEFINE_SPINLOCK(hash_lock);
70
71/*
72 * we assume that the new SIGEV_THREAD_ID shares no bits with the other
73 * SIGEV values. Here we put out an error if this assumption fails.
74 */
75#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
76 ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
77#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
78#endif
79
80/*
81 * parisc wants ENOTSUP instead of EOPNOTSUPP
82 */
83#ifndef ENOTSUP
84# define ENANOSLEEP_NOTSUP EOPNOTSUPP
85#else
86# define ENANOSLEEP_NOTSUP ENOTSUP
87#endif
88
89/*
90 * The timer ID is turned into a timer address by idr_find().
91 * Verifying a valid ID consists of:
92 *
93 * a) checking that idr_find() returns other than -1.
94 * b) checking that the timer id matches the one in the timer itself.
95 * c) that the timer owner is in the callers thread group.
96 */
97
98/*
99 * CLOCKs: The POSIX standard calls for a couple of clocks and allows us
100 * to implement others. This structure defines the various
101 * clocks.
102 *
103 * RESOLUTION: Clock resolution is used to round up timer and interval
104 * times, NOT to report clock times, which are reported with as
105 * much resolution as the system can muster. In some cases this
106 * resolution may depend on the underlying clock hardware and
107 * may not be quantifiable until run time, and only then is the
108 * necessary code is written. The standard says we should say
109 * something about this issue in the documentation...
110 *
111 * FUNCTIONS: The CLOCKs structure defines possible functions to
112 * handle various clock functions.
113 *
114 * The standard POSIX timer management code assumes the
115 * following: 1.) The k_itimer struct (sched.h) is used for
116 * the timer. 2.) The list, it_lock, it_clock, it_id and
117 * it_pid fields are not modified by timer code.
118 *
119 * Permissions: It is assumed that the clock_settime() function defined
120 * for each clock will take care of permission checks. Some
121 * clocks may be set able by any user (i.e. local process
122 * clocks) others not. Currently the only set able clock we
123 * have is CLOCK_REALTIME and its high res counter part, both of
124 * which we beg off on and pass to do_sys_settimeofday().
125 */
126
127static struct k_clock posix_clocks[MAX_CLOCKS];
128
129/*
130 * These ones are defined below.
131 */
132static int common_nsleep(const clockid_t, int flags, struct timespec *t,
133 struct timespec __user *rmtp);
134static int common_timer_create(struct k_itimer *new_timer);
135static void common_timer_get(struct k_itimer *, struct itimerspec *);
136static int common_timer_set(struct k_itimer *, int,
137 struct itimerspec *, struct itimerspec *);
138static int common_timer_del(struct k_itimer *timer);
139
140static enum hrtimer_restart posix_timer_fn(struct hrtimer *data);
141
142static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
143
144#define lock_timer(tid, flags) \
145({ struct k_itimer *__timr; \
146 __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags)); \
147 __timr; \
148})
149
150static int hash(struct signal_struct *sig, unsigned int nr)
151{
152 return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
153}
154
155static struct k_itimer *__posix_timers_find(struct hlist_head *head,
156 struct signal_struct *sig,
157 timer_t id)
158{
159 struct k_itimer *timer;
160
161 hlist_for_each_entry_rcu(timer, head, t_hash) {
162 if ((timer->it_signal == sig) && (timer->it_id == id))
163 return timer;
164 }
165 return NULL;
166}
167
168static struct k_itimer *posix_timer_by_id(timer_t id)
169{
170 struct signal_struct *sig = current->signal;
171 struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
172
173 return __posix_timers_find(head, sig, id);
174}
175
176static int posix_timer_add(struct k_itimer *timer)
177{
178 struct signal_struct *sig = current->signal;
179 int first_free_id = sig->posix_timer_id;
180 struct hlist_head *head;
181 int ret = -ENOENT;
182
183 do {
184 spin_lock(&hash_lock);
185 head = &posix_timers_hashtable[hash(sig, sig->posix_timer_id)];
186 if (!__posix_timers_find(head, sig, sig->posix_timer_id)) {
187 hlist_add_head_rcu(&timer->t_hash, head);
188 ret = sig->posix_timer_id;
189 }
190 if (++sig->posix_timer_id < 0)
191 sig->posix_timer_id = 0;
192 if ((sig->posix_timer_id == first_free_id) && (ret == -ENOENT))
193 /* Loop over all possible ids completed */
194 ret = -EAGAIN;
195 spin_unlock(&hash_lock);
196 } while (ret == -ENOENT);
197 return ret;
198}
199
200static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
201{
202 spin_unlock_irqrestore(&timr->it_lock, flags);
203}
204
205/* Get clock_realtime */
206static int posix_clock_realtime_get(clockid_t which_clock, struct timespec *tp)
207{
208 ktime_get_real_ts(tp);
209 return 0;
210}
211
212/* Set clock_realtime */
213static int posix_clock_realtime_set(const clockid_t which_clock,
214 const struct timespec *tp)
215{
216 return do_sys_settimeofday(tp, NULL);
217}
218
219static int posix_clock_realtime_adj(const clockid_t which_clock,
220 struct timex *t)
221{
222 return do_adjtimex(t);
223}
224
225/*
226 * Get monotonic time for posix timers
227 */
228static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp)
229{
230 ktime_get_ts(tp);
231 return 0;
232}
233
234/*
235 * Get monotonic-raw time for posix timers
236 */
237static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp)
238{
239 getrawmonotonic(tp);
240 return 0;
241}
242
243
244static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec *tp)
245{
246 *tp = current_kernel_time();
247 return 0;
248}
249
250static int posix_get_monotonic_coarse(clockid_t which_clock,
251 struct timespec *tp)
252{
253 *tp = get_monotonic_coarse();
254 return 0;
255}
256
257static int posix_get_coarse_res(const clockid_t which_clock, struct timespec *tp)
258{
259 *tp = ktime_to_timespec(KTIME_LOW_RES);
260 return 0;
261}
262
263static int posix_get_boottime(const clockid_t which_clock, struct timespec *tp)
264{
265 get_monotonic_boottime(tp);
266 return 0;
267}
268
269static int posix_get_tai(clockid_t which_clock, struct timespec *tp)
270{
271 timekeeping_clocktai(tp);
272 return 0;
273}
274
275static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec *tp)
276{
277 tp->tv_sec = 0;
278 tp->tv_nsec = hrtimer_resolution;
279 return 0;
280}
281
282/*
283 * Initialize everything, well, just everything in Posix clocks/timers ;)
284 */
285static __init int init_posix_timers(void)
286{
287 struct k_clock clock_realtime = {
288 .clock_getres = posix_get_hrtimer_res,
289 .clock_get = posix_clock_realtime_get,
290 .clock_set = posix_clock_realtime_set,
291 .clock_adj = posix_clock_realtime_adj,
292 .nsleep = common_nsleep,
293 .nsleep_restart = hrtimer_nanosleep_restart,
294 .timer_create = common_timer_create,
295 .timer_set = common_timer_set,
296 .timer_get = common_timer_get,
297 .timer_del = common_timer_del,
298 };
299 struct k_clock clock_monotonic = {
300 .clock_getres = posix_get_hrtimer_res,
301 .clock_get = posix_ktime_get_ts,
302 .nsleep = common_nsleep,
303 .nsleep_restart = hrtimer_nanosleep_restart,
304 .timer_create = common_timer_create,
305 .timer_set = common_timer_set,
306 .timer_get = common_timer_get,
307 .timer_del = common_timer_del,
308 };
309 struct k_clock clock_monotonic_raw = {
310 .clock_getres = posix_get_hrtimer_res,
311 .clock_get = posix_get_monotonic_raw,
312 };
313 struct k_clock clock_realtime_coarse = {
314 .clock_getres = posix_get_coarse_res,
315 .clock_get = posix_get_realtime_coarse,
316 };
317 struct k_clock clock_monotonic_coarse = {
318 .clock_getres = posix_get_coarse_res,
319 .clock_get = posix_get_monotonic_coarse,
320 };
321 struct k_clock clock_tai = {
322 .clock_getres = posix_get_hrtimer_res,
323 .clock_get = posix_get_tai,
324 .nsleep = common_nsleep,
325 .nsleep_restart = hrtimer_nanosleep_restart,
326 .timer_create = common_timer_create,
327 .timer_set = common_timer_set,
328 .timer_get = common_timer_get,
329 .timer_del = common_timer_del,
330 };
331 struct k_clock clock_boottime = {
332 .clock_getres = posix_get_hrtimer_res,
333 .clock_get = posix_get_boottime,
334 .nsleep = common_nsleep,
335 .nsleep_restart = hrtimer_nanosleep_restart,
336 .timer_create = common_timer_create,
337 .timer_set = common_timer_set,
338 .timer_get = common_timer_get,
339 .timer_del = common_timer_del,
340 };
341
342 posix_timers_register_clock(CLOCK_REALTIME, &clock_realtime);
343 posix_timers_register_clock(CLOCK_MONOTONIC, &clock_monotonic);
344 posix_timers_register_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw);
345 posix_timers_register_clock(CLOCK_REALTIME_COARSE, &clock_realtime_coarse);
346 posix_timers_register_clock(CLOCK_MONOTONIC_COARSE, &clock_monotonic_coarse);
347 posix_timers_register_clock(CLOCK_BOOTTIME, &clock_boottime);
348 posix_timers_register_clock(CLOCK_TAI, &clock_tai);
349
350 posix_timers_cache = kmem_cache_create("posix_timers_cache",
351 sizeof (struct k_itimer), 0, SLAB_PANIC,
352 NULL);
353 return 0;
354}
355
356__initcall(init_posix_timers);
357
358static void schedule_next_timer(struct k_itimer *timr)
359{
360 struct hrtimer *timer = &timr->it.real.timer;
361
362 if (timr->it.real.interval.tv64 == 0)
363 return;
364
365 timr->it_overrun += (unsigned int) hrtimer_forward(timer,
366 timer->base->get_time(),
367 timr->it.real.interval);
368
369 timr->it_overrun_last = timr->it_overrun;
370 timr->it_overrun = -1;
371 ++timr->it_requeue_pending;
372 hrtimer_restart(timer);
373}
374
375/*
376 * This function is exported for use by the signal deliver code. It is
377 * called just prior to the info block being released and passes that
378 * block to us. It's function is to update the overrun entry AND to
379 * restart the timer. It should only be called if the timer is to be
380 * restarted (i.e. we have flagged this in the sys_private entry of the
381 * info block).
382 *
383 * To protect against the timer going away while the interrupt is queued,
384 * we require that the it_requeue_pending flag be set.
385 */
386void do_schedule_next_timer(struct siginfo *info)
387{
388 struct k_itimer *timr;
389 unsigned long flags;
390
391 timr = lock_timer(info->si_tid, &flags);
392
393 if (timr && timr->it_requeue_pending == info->si_sys_private) {
394 if (timr->it_clock < 0)
395 posix_cpu_timer_schedule(timr);
396 else
397 schedule_next_timer(timr);
398
399 info->si_overrun += timr->it_overrun_last;
400 }
401
402 if (timr)
403 unlock_timer(timr, flags);
404}
405
406int posix_timer_event(struct k_itimer *timr, int si_private)
407{
408 struct task_struct *task;
409 int shared, ret = -1;
410 /*
411 * FIXME: if ->sigq is queued we can race with
412 * dequeue_signal()->do_schedule_next_timer().
413 *
414 * If dequeue_signal() sees the "right" value of
415 * si_sys_private it calls do_schedule_next_timer().
416 * We re-queue ->sigq and drop ->it_lock().
417 * do_schedule_next_timer() locks the timer
418 * and re-schedules it while ->sigq is pending.
419 * Not really bad, but not that we want.
420 */
421 timr->sigq->info.si_sys_private = si_private;
422
423 rcu_read_lock();
424 task = pid_task(timr->it_pid, PIDTYPE_PID);
425 if (task) {
426 shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID);
427 ret = send_sigqueue(timr->sigq, task, shared);
428 }
429 rcu_read_unlock();
430 /* If we failed to send the signal the timer stops. */
431 return ret > 0;
432}
433EXPORT_SYMBOL_GPL(posix_timer_event);
434
435/*
436 * This function gets called when a POSIX.1b interval timer expires. It
437 * is used as a callback from the kernel internal timer. The
438 * run_timer_list code ALWAYS calls with interrupts on.
439
440 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
441 */
442static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
443{
444 struct k_itimer *timr;
445 unsigned long flags;
446 int si_private = 0;
447 enum hrtimer_restart ret = HRTIMER_NORESTART;
448
449 timr = container_of(timer, struct k_itimer, it.real.timer);
450 spin_lock_irqsave(&timr->it_lock, flags);
451
452 if (timr->it.real.interval.tv64 != 0)
453 si_private = ++timr->it_requeue_pending;
454
455 if (posix_timer_event(timr, si_private)) {
456 /*
457 * signal was not sent because of sig_ignor
458 * we will not get a call back to restart it AND
459 * it should be restarted.
460 */
461 if (timr->it.real.interval.tv64 != 0) {
462 ktime_t now = hrtimer_cb_get_time(timer);
463
464 /*
465 * FIXME: What we really want, is to stop this
466 * timer completely and restart it in case the
467 * SIG_IGN is removed. This is a non trivial
468 * change which involves sighand locking
469 * (sigh !), which we don't want to do late in
470 * the release cycle.
471 *
472 * For now we just let timers with an interval
473 * less than a jiffie expire every jiffie to
474 * avoid softirq starvation in case of SIG_IGN
475 * and a very small interval, which would put
476 * the timer right back on the softirq pending
477 * list. By moving now ahead of time we trick
478 * hrtimer_forward() to expire the timer
479 * later, while we still maintain the overrun
480 * accuracy, but have some inconsistency in
481 * the timer_gettime() case. This is at least
482 * better than a starved softirq. A more
483 * complex fix which solves also another related
484 * inconsistency is already in the pipeline.
485 */
486#ifdef CONFIG_HIGH_RES_TIMERS
487 {
488 ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ);
489
490 if (timr->it.real.interval.tv64 < kj.tv64)
491 now = ktime_add(now, kj);
492 }
493#endif
494 timr->it_overrun += (unsigned int)
495 hrtimer_forward(timer, now,
496 timr->it.real.interval);
497 ret = HRTIMER_RESTART;
498 ++timr->it_requeue_pending;
499 }
500 }
501
502 unlock_timer(timr, flags);
503 return ret;
504}
505
506static struct pid *good_sigevent(sigevent_t * event)
507{
508 struct task_struct *rtn = current->group_leader;
509
510 if ((event->sigev_notify & SIGEV_THREAD_ID ) &&
511 (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) ||
512 !same_thread_group(rtn, current) ||
513 (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL))
514 return NULL;
515
516 if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) &&
517 ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX)))
518 return NULL;
519
520 return task_pid(rtn);
521}
522
523void posix_timers_register_clock(const clockid_t clock_id,
524 struct k_clock *new_clock)
525{
526 if ((unsigned) clock_id >= MAX_CLOCKS) {
527 printk(KERN_WARNING "POSIX clock register failed for clock_id %d\n",
528 clock_id);
529 return;
530 }
531
532 if (!new_clock->clock_get) {
533 printk(KERN_WARNING "POSIX clock id %d lacks clock_get()\n",
534 clock_id);
535 return;
536 }
537 if (!new_clock->clock_getres) {
538 printk(KERN_WARNING "POSIX clock id %d lacks clock_getres()\n",
539 clock_id);
540 return;
541 }
542
543 posix_clocks[clock_id] = *new_clock;
544}
545EXPORT_SYMBOL_GPL(posix_timers_register_clock);
546
547static struct k_itimer * alloc_posix_timer(void)
548{
549 struct k_itimer *tmr;
550 tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
551 if (!tmr)
552 return tmr;
553 if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
554 kmem_cache_free(posix_timers_cache, tmr);
555 return NULL;
556 }
557 memset(&tmr->sigq->info, 0, sizeof(siginfo_t));
558 return tmr;
559}
560
561static void k_itimer_rcu_free(struct rcu_head *head)
562{
563 struct k_itimer *tmr = container_of(head, struct k_itimer, it.rcu);
564
565 kmem_cache_free(posix_timers_cache, tmr);
566}
567
568#define IT_ID_SET 1
569#define IT_ID_NOT_SET 0
570static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
571{
572 if (it_id_set) {
573 unsigned long flags;
574 spin_lock_irqsave(&hash_lock, flags);
575 hlist_del_rcu(&tmr->t_hash);
576 spin_unlock_irqrestore(&hash_lock, flags);
577 }
578 put_pid(tmr->it_pid);
579 sigqueue_free(tmr->sigq);
580 call_rcu(&tmr->it.rcu, k_itimer_rcu_free);
581}
582
583static struct k_clock *clockid_to_kclock(const clockid_t id)
584{
585 if (id < 0)
586 return (id & CLOCKFD_MASK) == CLOCKFD ?
587 &clock_posix_dynamic : &clock_posix_cpu;
588
589 if (id >= MAX_CLOCKS || !posix_clocks[id].clock_getres)
590 return NULL;
591 return &posix_clocks[id];
592}
593
594static int common_timer_create(struct k_itimer *new_timer)
595{
596 hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
597 return 0;
598}
599
600/* Create a POSIX.1b interval timer. */
601
602SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
603 struct sigevent __user *, timer_event_spec,
604 timer_t __user *, created_timer_id)
605{
606 struct k_clock *kc = clockid_to_kclock(which_clock);
607 struct k_itimer *new_timer;
608 int error, new_timer_id;
609 sigevent_t event;
610 int it_id_set = IT_ID_NOT_SET;
611
612 if (!kc)
613 return -EINVAL;
614 if (!kc->timer_create)
615 return -EOPNOTSUPP;
616
617 new_timer = alloc_posix_timer();
618 if (unlikely(!new_timer))
619 return -EAGAIN;
620
621 spin_lock_init(&new_timer->it_lock);
622 new_timer_id = posix_timer_add(new_timer);
623 if (new_timer_id < 0) {
624 error = new_timer_id;
625 goto out;
626 }
627
628 it_id_set = IT_ID_SET;
629 new_timer->it_id = (timer_t) new_timer_id;
630 new_timer->it_clock = which_clock;
631 new_timer->it_overrun = -1;
632
633 if (timer_event_spec) {
634 if (copy_from_user(&event, timer_event_spec, sizeof (event))) {
635 error = -EFAULT;
636 goto out;
637 }
638 rcu_read_lock();
639 new_timer->it_pid = get_pid(good_sigevent(&event));
640 rcu_read_unlock();
641 if (!new_timer->it_pid) {
642 error = -EINVAL;
643 goto out;
644 }
645 } else {
646 memset(&event.sigev_value, 0, sizeof(event.sigev_value));
647 event.sigev_notify = SIGEV_SIGNAL;
648 event.sigev_signo = SIGALRM;
649 event.sigev_value.sival_int = new_timer->it_id;
650 new_timer->it_pid = get_pid(task_tgid(current));
651 }
652
653 new_timer->it_sigev_notify = event.sigev_notify;
654 new_timer->sigq->info.si_signo = event.sigev_signo;
655 new_timer->sigq->info.si_value = event.sigev_value;
656 new_timer->sigq->info.si_tid = new_timer->it_id;
657 new_timer->sigq->info.si_code = SI_TIMER;
658
659 if (copy_to_user(created_timer_id,
660 &new_timer_id, sizeof (new_timer_id))) {
661 error = -EFAULT;
662 goto out;
663 }
664
665 error = kc->timer_create(new_timer);
666 if (error)
667 goto out;
668
669 spin_lock_irq(¤t->sighand->siglock);
670 new_timer->it_signal = current->signal;
671 list_add(&new_timer->list, ¤t->signal->posix_timers);
672 spin_unlock_irq(¤t->sighand->siglock);
673
674 return 0;
675 /*
676 * In the case of the timer belonging to another task, after
677 * the task is unlocked, the timer is owned by the other task
678 * and may cease to exist at any time. Don't use or modify
679 * new_timer after the unlock call.
680 */
681out:
682 release_posix_timer(new_timer, it_id_set);
683 return error;
684}
685
686/*
687 * Locking issues: We need to protect the result of the id look up until
688 * we get the timer locked down so it is not deleted under us. The
689 * removal is done under the idr spinlock so we use that here to bridge
690 * the find to the timer lock. To avoid a dead lock, the timer id MUST
691 * be release with out holding the timer lock.
692 */
693static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
694{
695 struct k_itimer *timr;
696
697 /*
698 * timer_t could be any type >= int and we want to make sure any
699 * @timer_id outside positive int range fails lookup.
700 */
701 if ((unsigned long long)timer_id > INT_MAX)
702 return NULL;
703
704 rcu_read_lock();
705 timr = posix_timer_by_id(timer_id);
706 if (timr) {
707 spin_lock_irqsave(&timr->it_lock, *flags);
708 if (timr->it_signal == current->signal) {
709 rcu_read_unlock();
710 return timr;
711 }
712 spin_unlock_irqrestore(&timr->it_lock, *flags);
713 }
714 rcu_read_unlock();
715
716 return NULL;
717}
718
719/*
720 * Get the time remaining on a POSIX.1b interval timer. This function
721 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
722 * mess with irq.
723 *
724 * We have a couple of messes to clean up here. First there is the case
725 * of a timer that has a requeue pending. These timers should appear to
726 * be in the timer list with an expiry as if we were to requeue them
727 * now.
728 *
729 * The second issue is the SIGEV_NONE timer which may be active but is
730 * not really ever put in the timer list (to save system resources).
731 * This timer may be expired, and if so, we will do it here. Otherwise
732 * it is the same as a requeue pending timer WRT to what we should
733 * report.
734 */
735static void
736common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting)
737{
738 ktime_t now, remaining, iv;
739 struct hrtimer *timer = &timr->it.real.timer;
740
741 memset(cur_setting, 0, sizeof(struct itimerspec));
742
743 iv = timr->it.real.interval;
744
745 /* interval timer ? */
746 if (iv.tv64)
747 cur_setting->it_interval = ktime_to_timespec(iv);
748 else if (!hrtimer_active(timer) &&
749 (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
750 return;
751
752 now = timer->base->get_time();
753
754 /*
755 * When a requeue is pending or this is a SIGEV_NONE
756 * timer move the expiry time forward by intervals, so
757 * expiry is > now.
758 */
759 if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING ||
760 (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE))
761 timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv);
762
763 remaining = __hrtimer_expires_remaining_adjusted(timer, now);
764 /* Return 0 only, when the timer is expired and not pending */
765 if (remaining.tv64 <= 0) {
766 /*
767 * A single shot SIGEV_NONE timer must return 0, when
768 * it is expired !
769 */
770 if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
771 cur_setting->it_value.tv_nsec = 1;
772 } else
773 cur_setting->it_value = ktime_to_timespec(remaining);
774}
775
776/* Get the time remaining on a POSIX.1b interval timer. */
777SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
778 struct itimerspec __user *, setting)
779{
780 struct itimerspec cur_setting;
781 struct k_itimer *timr;
782 struct k_clock *kc;
783 unsigned long flags;
784 int ret = 0;
785
786 timr = lock_timer(timer_id, &flags);
787 if (!timr)
788 return -EINVAL;
789
790 kc = clockid_to_kclock(timr->it_clock);
791 if (WARN_ON_ONCE(!kc || !kc->timer_get))
792 ret = -EINVAL;
793 else
794 kc->timer_get(timr, &cur_setting);
795
796 unlock_timer(timr, flags);
797
798 if (!ret && copy_to_user(setting, &cur_setting, sizeof (cur_setting)))
799 return -EFAULT;
800
801 return ret;
802}
803
804/*
805 * Get the number of overruns of a POSIX.1b interval timer. This is to
806 * be the overrun of the timer last delivered. At the same time we are
807 * accumulating overruns on the next timer. The overrun is frozen when
808 * the signal is delivered, either at the notify time (if the info block
809 * is not queued) or at the actual delivery time (as we are informed by
810 * the call back to do_schedule_next_timer(). So all we need to do is
811 * to pick up the frozen overrun.
812 */
813SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
814{
815 struct k_itimer *timr;
816 int overrun;
817 unsigned long flags;
818
819 timr = lock_timer(timer_id, &flags);
820 if (!timr)
821 return -EINVAL;
822
823 overrun = timr->it_overrun_last;
824 unlock_timer(timr, flags);
825
826 return overrun;
827}
828
829/* Set a POSIX.1b interval timer. */
830/* timr->it_lock is taken. */
831static int
832common_timer_set(struct k_itimer *timr, int flags,
833 struct itimerspec *new_setting, struct itimerspec *old_setting)
834{
835 struct hrtimer *timer = &timr->it.real.timer;
836 enum hrtimer_mode mode;
837
838 if (old_setting)
839 common_timer_get(timr, old_setting);
840
841 /* disable the timer */
842 timr->it.real.interval.tv64 = 0;
843 /*
844 * careful here. If smp we could be in the "fire" routine which will
845 * be spinning as we hold the lock. But this is ONLY an SMP issue.
846 */
847 if (hrtimer_try_to_cancel(timer) < 0)
848 return TIMER_RETRY;
849
850 timr->it_requeue_pending = (timr->it_requeue_pending + 2) &
851 ~REQUEUE_PENDING;
852 timr->it_overrun_last = 0;
853
854 /* switch off the timer when it_value is zero */
855 if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
856 return 0;
857
858 mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
859 hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
860 timr->it.real.timer.function = posix_timer_fn;
861
862 hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value));
863
864 /* Convert interval */
865 timr->it.real.interval = timespec_to_ktime(new_setting->it_interval);
866
867 /* SIGEV_NONE timers are not queued ! See common_timer_get */
868 if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) {
869 /* Setup correct expiry time for relative timers */
870 if (mode == HRTIMER_MODE_REL) {
871 hrtimer_add_expires(timer, timer->base->get_time());
872 }
873 return 0;
874 }
875
876 hrtimer_start_expires(timer, mode);
877 return 0;
878}
879
880/* Set a POSIX.1b interval timer */
881SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
882 const struct itimerspec __user *, new_setting,
883 struct itimerspec __user *, old_setting)
884{
885 struct k_itimer *timr;
886 struct itimerspec new_spec, old_spec;
887 int error = 0;
888 unsigned long flag;
889 struct itimerspec *rtn = old_setting ? &old_spec : NULL;
890 struct k_clock *kc;
891
892 if (!new_setting)
893 return -EINVAL;
894
895 if (copy_from_user(&new_spec, new_setting, sizeof (new_spec)))
896 return -EFAULT;
897
898 if (!timespec_valid(&new_spec.it_interval) ||
899 !timespec_valid(&new_spec.it_value))
900 return -EINVAL;
901retry:
902 timr = lock_timer(timer_id, &flag);
903 if (!timr)
904 return -EINVAL;
905
906 kc = clockid_to_kclock(timr->it_clock);
907 if (WARN_ON_ONCE(!kc || !kc->timer_set))
908 error = -EINVAL;
909 else
910 error = kc->timer_set(timr, flags, &new_spec, rtn);
911
912 unlock_timer(timr, flag);
913 if (error == TIMER_RETRY) {
914 rtn = NULL; // We already got the old time...
915 goto retry;
916 }
917
918 if (old_setting && !error &&
919 copy_to_user(old_setting, &old_spec, sizeof (old_spec)))
920 error = -EFAULT;
921
922 return error;
923}
924
925static int common_timer_del(struct k_itimer *timer)
926{
927 timer->it.real.interval.tv64 = 0;
928
929 if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0)
930 return TIMER_RETRY;
931 return 0;
932}
933
934static inline int timer_delete_hook(struct k_itimer *timer)
935{
936 struct k_clock *kc = clockid_to_kclock(timer->it_clock);
937
938 if (WARN_ON_ONCE(!kc || !kc->timer_del))
939 return -EINVAL;
940 return kc->timer_del(timer);
941}
942
943/* Delete a POSIX.1b interval timer. */
944SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
945{
946 struct k_itimer *timer;
947 unsigned long flags;
948
949retry_delete:
950 timer = lock_timer(timer_id, &flags);
951 if (!timer)
952 return -EINVAL;
953
954 if (timer_delete_hook(timer) == TIMER_RETRY) {
955 unlock_timer(timer, flags);
956 goto retry_delete;
957 }
958
959 spin_lock(¤t->sighand->siglock);
960 list_del(&timer->list);
961 spin_unlock(¤t->sighand->siglock);
962 /*
963 * This keeps any tasks waiting on the spin lock from thinking
964 * they got something (see the lock code above).
965 */
966 timer->it_signal = NULL;
967
968 unlock_timer(timer, flags);
969 release_posix_timer(timer, IT_ID_SET);
970 return 0;
971}
972
973/*
974 * return timer owned by the process, used by exit_itimers
975 */
976static void itimer_delete(struct k_itimer *timer)
977{
978 unsigned long flags;
979
980retry_delete:
981 spin_lock_irqsave(&timer->it_lock, flags);
982
983 if (timer_delete_hook(timer) == TIMER_RETRY) {
984 unlock_timer(timer, flags);
985 goto retry_delete;
986 }
987 list_del(&timer->list);
988 /*
989 * This keeps any tasks waiting on the spin lock from thinking
990 * they got something (see the lock code above).
991 */
992 timer->it_signal = NULL;
993
994 unlock_timer(timer, flags);
995 release_posix_timer(timer, IT_ID_SET);
996}
997
998/*
999 * This is called by do_exit or de_thread, only when there are no more
1000 * references to the shared signal_struct.
1001 */
1002void exit_itimers(struct signal_struct *sig)
1003{
1004 struct k_itimer *tmr;
1005
1006 while (!list_empty(&sig->posix_timers)) {
1007 tmr = list_entry(sig->posix_timers.next, struct k_itimer, list);
1008 itimer_delete(tmr);
1009 }
1010}
1011
1012SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1013 const struct timespec __user *, tp)
1014{
1015 struct k_clock *kc = clockid_to_kclock(which_clock);
1016 struct timespec new_tp;
1017
1018 if (!kc || !kc->clock_set)
1019 return -EINVAL;
1020
1021 if (copy_from_user(&new_tp, tp, sizeof (*tp)))
1022 return -EFAULT;
1023
1024 return kc->clock_set(which_clock, &new_tp);
1025}
1026
1027SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1028 struct timespec __user *,tp)
1029{
1030 struct k_clock *kc = clockid_to_kclock(which_clock);
1031 struct timespec kernel_tp;
1032 int error;
1033
1034 if (!kc)
1035 return -EINVAL;
1036
1037 error = kc->clock_get(which_clock, &kernel_tp);
1038
1039 if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp)))
1040 error = -EFAULT;
1041
1042 return error;
1043}
1044
1045SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1046 struct timex __user *, utx)
1047{
1048 struct k_clock *kc = clockid_to_kclock(which_clock);
1049 struct timex ktx;
1050 int err;
1051
1052 if (!kc)
1053 return -EINVAL;
1054 if (!kc->clock_adj)
1055 return -EOPNOTSUPP;
1056
1057 if (copy_from_user(&ktx, utx, sizeof(ktx)))
1058 return -EFAULT;
1059
1060 err = kc->clock_adj(which_clock, &ktx);
1061
1062 if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1063 return -EFAULT;
1064
1065 return err;
1066}
1067
1068SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1069 struct timespec __user *, tp)
1070{
1071 struct k_clock *kc = clockid_to_kclock(which_clock);
1072 struct timespec rtn_tp;
1073 int error;
1074
1075 if (!kc)
1076 return -EINVAL;
1077
1078 error = kc->clock_getres(which_clock, &rtn_tp);
1079
1080 if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp)))
1081 error = -EFAULT;
1082
1083 return error;
1084}
1085
1086/*
1087 * nanosleep for monotonic and realtime clocks
1088 */
1089static int common_nsleep(const clockid_t which_clock, int flags,
1090 struct timespec *tsave, struct timespec __user *rmtp)
1091{
1092 return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ?
1093 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1094 which_clock);
1095}
1096
1097SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1098 const struct timespec __user *, rqtp,
1099 struct timespec __user *, rmtp)
1100{
1101 struct k_clock *kc = clockid_to_kclock(which_clock);
1102 struct timespec t;
1103
1104 if (!kc)
1105 return -EINVAL;
1106 if (!kc->nsleep)
1107 return -ENANOSLEEP_NOTSUP;
1108
1109 if (copy_from_user(&t, rqtp, sizeof (struct timespec)))
1110 return -EFAULT;
1111
1112 if (!timespec_valid(&t))
1113 return -EINVAL;
1114
1115 return kc->nsleep(which_clock, flags, &t, rmtp);
1116}
1117
1118/*
1119 * This will restart clock_nanosleep. This is required only by
1120 * compat_clock_nanosleep_restart for now.
1121 */
1122long clock_nanosleep_restart(struct restart_block *restart_block)
1123{
1124 clockid_t which_clock = restart_block->nanosleep.clockid;
1125 struct k_clock *kc = clockid_to_kclock(which_clock);
1126
1127 if (WARN_ON_ONCE(!kc || !kc->nsleep_restart))
1128 return -EINVAL;
1129
1130 return kc->nsleep_restart(restart_block);
1131}