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
275/*
276 * Initialize everything, well, just everything in Posix clocks/timers ;)
277 */
278static __init int init_posix_timers(void)
279{
280	struct k_clock clock_realtime = {
281		.clock_getres	= hrtimer_get_res,
282		.clock_get	= posix_clock_realtime_get,
283		.clock_set	= posix_clock_realtime_set,
284		.clock_adj	= posix_clock_realtime_adj,
285		.nsleep		= common_nsleep,
286		.nsleep_restart	= hrtimer_nanosleep_restart,
287		.timer_create	= common_timer_create,
288		.timer_set	= common_timer_set,
289		.timer_get	= common_timer_get,
290		.timer_del	= common_timer_del,
291	};
292	struct k_clock clock_monotonic = {
293		.clock_getres	= hrtimer_get_res,
294		.clock_get	= posix_ktime_get_ts,
295		.nsleep		= common_nsleep,
296		.nsleep_restart	= hrtimer_nanosleep_restart,
297		.timer_create	= common_timer_create,
298		.timer_set	= common_timer_set,
299		.timer_get	= common_timer_get,
300		.timer_del	= common_timer_del,
301	};
302	struct k_clock clock_monotonic_raw = {
303		.clock_getres	= hrtimer_get_res,
304		.clock_get	= posix_get_monotonic_raw,
305	};
306	struct k_clock clock_realtime_coarse = {
307		.clock_getres	= posix_get_coarse_res,
308		.clock_get	= posix_get_realtime_coarse,
309	};
310	struct k_clock clock_monotonic_coarse = {
311		.clock_getres	= posix_get_coarse_res,
312		.clock_get	= posix_get_monotonic_coarse,
313	};
314	struct k_clock clock_tai = {
315		.clock_getres	= hrtimer_get_res,
316		.clock_get	= posix_get_tai,
317		.nsleep		= common_nsleep,
318		.nsleep_restart	= hrtimer_nanosleep_restart,
319		.timer_create	= common_timer_create,
320		.timer_set	= common_timer_set,
321		.timer_get	= common_timer_get,
322		.timer_del	= common_timer_del,
323	};
324	struct k_clock clock_boottime = {
325		.clock_getres	= hrtimer_get_res,
326		.clock_get	= posix_get_boottime,
327		.nsleep		= common_nsleep,
328		.nsleep_restart	= hrtimer_nanosleep_restart,
329		.timer_create	= common_timer_create,
330		.timer_set	= common_timer_set,
331		.timer_get	= common_timer_get,
332		.timer_del	= common_timer_del,
333	};
334
335	posix_timers_register_clock(CLOCK_REALTIME, &clock_realtime);
336	posix_timers_register_clock(CLOCK_MONOTONIC, &clock_monotonic);
337	posix_timers_register_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw);
338	posix_timers_register_clock(CLOCK_REALTIME_COARSE, &clock_realtime_coarse);
339	posix_timers_register_clock(CLOCK_MONOTONIC_COARSE, &clock_monotonic_coarse);
340	posix_timers_register_clock(CLOCK_BOOTTIME, &clock_boottime);
341	posix_timers_register_clock(CLOCK_TAI, &clock_tai);
342
343	posix_timers_cache = kmem_cache_create("posix_timers_cache",
344					sizeof (struct k_itimer), 0, SLAB_PANIC,
345					NULL);
346	return 0;
347}
348
349__initcall(init_posix_timers);
350
351static void schedule_next_timer(struct k_itimer *timr)
352{
353	struct hrtimer *timer = &timr->it.real.timer;
354
355	if (timr->it.real.interval.tv64 == 0)
356		return;
357
358	timr->it_overrun += (unsigned int) hrtimer_forward(timer,
359						timer->base->get_time(),
360						timr->it.real.interval);
361
362	timr->it_overrun_last = timr->it_overrun;
363	timr->it_overrun = -1;
364	++timr->it_requeue_pending;
365	hrtimer_restart(timer);
366}
367
368/*
369 * This function is exported for use by the signal deliver code.  It is
370 * called just prior to the info block being released and passes that
371 * block to us.  It's function is to update the overrun entry AND to
372 * restart the timer.  It should only be called if the timer is to be
373 * restarted (i.e. we have flagged this in the sys_private entry of the
374 * info block).
375 *
376 * To protect against the timer going away while the interrupt is queued,
377 * we require that the it_requeue_pending flag be set.
378 */
379void do_schedule_next_timer(struct siginfo *info)
380{
381	struct k_itimer *timr;
382	unsigned long flags;
383
384	timr = lock_timer(info->si_tid, &flags);
385
386	if (timr && timr->it_requeue_pending == info->si_sys_private) {
387		if (timr->it_clock < 0)
388			posix_cpu_timer_schedule(timr);
389		else
390			schedule_next_timer(timr);
391
392		info->si_overrun += timr->it_overrun_last;
393	}
394
395	if (timr)
396		unlock_timer(timr, flags);
397}
398
399int posix_timer_event(struct k_itimer *timr, int si_private)
400{
401	struct task_struct *task;
402	int shared, ret = -1;
403	/*
404	 * FIXME: if ->sigq is queued we can race with
405	 * dequeue_signal()->do_schedule_next_timer().
406	 *
407	 * If dequeue_signal() sees the "right" value of
408	 * si_sys_private it calls do_schedule_next_timer().
409	 * We re-queue ->sigq and drop ->it_lock().
410	 * do_schedule_next_timer() locks the timer
411	 * and re-schedules it while ->sigq is pending.
412	 * Not really bad, but not that we want.
413	 */
414	timr->sigq->info.si_sys_private = si_private;
415
416	rcu_read_lock();
417	task = pid_task(timr->it_pid, PIDTYPE_PID);
418	if (task) {
419		shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID);
420		ret = send_sigqueue(timr->sigq, task, shared);
421	}
422	rcu_read_unlock();
423	/* If we failed to send the signal the timer stops. */
424	return ret > 0;
425}
426EXPORT_SYMBOL_GPL(posix_timer_event);
427
428/*
429 * This function gets called when a POSIX.1b interval timer expires.  It
430 * is used as a callback from the kernel internal timer.  The
431 * run_timer_list code ALWAYS calls with interrupts on.
432
433 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
434 */
435static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
436{
437	struct k_itimer *timr;
438	unsigned long flags;
439	int si_private = 0;
440	enum hrtimer_restart ret = HRTIMER_NORESTART;
441
442	timr = container_of(timer, struct k_itimer, it.real.timer);
443	spin_lock_irqsave(&timr->it_lock, flags);
444
445	if (timr->it.real.interval.tv64 != 0)
446		si_private = ++timr->it_requeue_pending;
447
448	if (posix_timer_event(timr, si_private)) {
449		/*
450		 * signal was not sent because of sig_ignor
451		 * we will not get a call back to restart it AND
452		 * it should be restarted.
453		 */
454		if (timr->it.real.interval.tv64 != 0) {
455			ktime_t now = hrtimer_cb_get_time(timer);
456
457			/*
458			 * FIXME: What we really want, is to stop this
459			 * timer completely and restart it in case the
460			 * SIG_IGN is removed. This is a non trivial
461			 * change which involves sighand locking
462			 * (sigh !), which we don't want to do late in
463			 * the release cycle.
464			 *
465			 * For now we just let timers with an interval
466			 * less than a jiffie expire every jiffie to
467			 * avoid softirq starvation in case of SIG_IGN
468			 * and a very small interval, which would put
469			 * the timer right back on the softirq pending
470			 * list. By moving now ahead of time we trick
471			 * hrtimer_forward() to expire the timer
472			 * later, while we still maintain the overrun
473			 * accuracy, but have some inconsistency in
474			 * the timer_gettime() case. This is at least
475			 * better than a starved softirq. A more
476			 * complex fix which solves also another related
477			 * inconsistency is already in the pipeline.
478			 */
479#ifdef CONFIG_HIGH_RES_TIMERS
480			{
481				ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ);
482
483				if (timr->it.real.interval.tv64 < kj.tv64)
484					now = ktime_add(now, kj);
485			}
486#endif
487			timr->it_overrun += (unsigned int)
488				hrtimer_forward(timer, now,
489						timr->it.real.interval);
490			ret = HRTIMER_RESTART;
491			++timr->it_requeue_pending;
492		}
493	}
494
495	unlock_timer(timr, flags);
496	return ret;
497}
498
499static struct pid *good_sigevent(sigevent_t * event)
500{
501	struct task_struct *rtn = current->group_leader;
502
503	if ((event->sigev_notify & SIGEV_THREAD_ID ) &&
504		(!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) ||
505		 !same_thread_group(rtn, current) ||
506		 (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL))
507		return NULL;
508
509	if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) &&
510	    ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX)))
511		return NULL;
512
513	return task_pid(rtn);
514}
515
516void posix_timers_register_clock(const clockid_t clock_id,
517				 struct k_clock *new_clock)
518{
519	if ((unsigned) clock_id >= MAX_CLOCKS) {
520		printk(KERN_WARNING "POSIX clock register failed for clock_id %d\n",
521		       clock_id);
522		return;
523	}
524
525	if (!new_clock->clock_get) {
526		printk(KERN_WARNING "POSIX clock id %d lacks clock_get()\n",
527		       clock_id);
528		return;
529	}
530	if (!new_clock->clock_getres) {
531		printk(KERN_WARNING "POSIX clock id %d lacks clock_getres()\n",
532		       clock_id);
533		return;
534	}
535
536	posix_clocks[clock_id] = *new_clock;
537}
538EXPORT_SYMBOL_GPL(posix_timers_register_clock);
539
540static struct k_itimer * alloc_posix_timer(void)
541{
542	struct k_itimer *tmr;
543	tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
544	if (!tmr)
545		return tmr;
546	if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
547		kmem_cache_free(posix_timers_cache, tmr);
548		return NULL;
549	}
550	memset(&tmr->sigq->info, 0, sizeof(siginfo_t));
551	return tmr;
552}
553
554static void k_itimer_rcu_free(struct rcu_head *head)
555{
556	struct k_itimer *tmr = container_of(head, struct k_itimer, it.rcu);
557
558	kmem_cache_free(posix_timers_cache, tmr);
559}
560
561#define IT_ID_SET	1
562#define IT_ID_NOT_SET	0
563static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
564{
565	if (it_id_set) {
566		unsigned long flags;
567		spin_lock_irqsave(&hash_lock, flags);
568		hlist_del_rcu(&tmr->t_hash);
569		spin_unlock_irqrestore(&hash_lock, flags);
570	}
571	put_pid(tmr->it_pid);
572	sigqueue_free(tmr->sigq);
573	call_rcu(&tmr->it.rcu, k_itimer_rcu_free);
574}
575
576static struct k_clock *clockid_to_kclock(const clockid_t id)
577{
578	if (id < 0)
579		return (id & CLOCKFD_MASK) == CLOCKFD ?
580			&clock_posix_dynamic : &clock_posix_cpu;
581
582	if (id >= MAX_CLOCKS || !posix_clocks[id].clock_getres)
583		return NULL;
584	return &posix_clocks[id];
585}
586
587static int common_timer_create(struct k_itimer *new_timer)
588{
589	hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
590	return 0;
591}
592
593/* Create a POSIX.1b interval timer. */
594
595SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
596		struct sigevent __user *, timer_event_spec,
597		timer_t __user *, created_timer_id)
598{
599	struct k_clock *kc = clockid_to_kclock(which_clock);
600	struct k_itimer *new_timer;
601	int error, new_timer_id;
602	sigevent_t event;
603	int it_id_set = IT_ID_NOT_SET;
604
605	if (!kc)
606		return -EINVAL;
607	if (!kc->timer_create)
608		return -EOPNOTSUPP;
609
610	new_timer = alloc_posix_timer();
611	if (unlikely(!new_timer))
612		return -EAGAIN;
613
614	spin_lock_init(&new_timer->it_lock);
615	new_timer_id = posix_timer_add(new_timer);
616	if (new_timer_id < 0) {
617		error = new_timer_id;
618		goto out;
619	}
620
621	it_id_set = IT_ID_SET;
622	new_timer->it_id = (timer_t) new_timer_id;
623	new_timer->it_clock = which_clock;
624	new_timer->it_overrun = -1;
625
626	if (timer_event_spec) {
627		if (copy_from_user(&event, timer_event_spec, sizeof (event))) {
628			error = -EFAULT;
629			goto out;
630		}
631		rcu_read_lock();
632		new_timer->it_pid = get_pid(good_sigevent(&event));
633		rcu_read_unlock();
634		if (!new_timer->it_pid) {
635			error = -EINVAL;
636			goto out;
637		}
638	} else {
639		memset(&event.sigev_value, 0, sizeof(event.sigev_value));
640		event.sigev_notify = SIGEV_SIGNAL;
641		event.sigev_signo = SIGALRM;
642		event.sigev_value.sival_int = new_timer->it_id;
643		new_timer->it_pid = get_pid(task_tgid(current));
644	}
645
646	new_timer->it_sigev_notify     = event.sigev_notify;
647	new_timer->sigq->info.si_signo = event.sigev_signo;
648	new_timer->sigq->info.si_value = event.sigev_value;
649	new_timer->sigq->info.si_tid   = new_timer->it_id;
650	new_timer->sigq->info.si_code  = SI_TIMER;
651
652	if (copy_to_user(created_timer_id,
653			 &new_timer_id, sizeof (new_timer_id))) {
654		error = -EFAULT;
655		goto out;
656	}
657
658	error = kc->timer_create(new_timer);
659	if (error)
660		goto out;
661
662	spin_lock_irq(&current->sighand->siglock);
663	new_timer->it_signal = current->signal;
664	list_add(&new_timer->list, &current->signal->posix_timers);
665	spin_unlock_irq(&current->sighand->siglock);
666
667	return 0;
668	/*
669	 * In the case of the timer belonging to another task, after
670	 * the task is unlocked, the timer is owned by the other task
671	 * and may cease to exist at any time.  Don't use or modify
672	 * new_timer after the unlock call.
673	 */
674out:
675	release_posix_timer(new_timer, it_id_set);
676	return error;
677}
678
679/*
680 * Locking issues: We need to protect the result of the id look up until
681 * we get the timer locked down so it is not deleted under us.  The
682 * removal is done under the idr spinlock so we use that here to bridge
683 * the find to the timer lock.  To avoid a dead lock, the timer id MUST
684 * be release with out holding the timer lock.
685 */
686static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
687{
688	struct k_itimer *timr;
689
690	/*
691	 * timer_t could be any type >= int and we want to make sure any
692	 * @timer_id outside positive int range fails lookup.
693	 */
694	if ((unsigned long long)timer_id > INT_MAX)
695		return NULL;
696
697	rcu_read_lock();
698	timr = posix_timer_by_id(timer_id);
699	if (timr) {
700		spin_lock_irqsave(&timr->it_lock, *flags);
701		if (timr->it_signal == current->signal) {
702			rcu_read_unlock();
703			return timr;
704		}
705		spin_unlock_irqrestore(&timr->it_lock, *flags);
706	}
707	rcu_read_unlock();
708
709	return NULL;
710}
711
712/*
713 * Get the time remaining on a POSIX.1b interval timer.  This function
714 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
715 * mess with irq.
716 *
717 * We have a couple of messes to clean up here.  First there is the case
718 * of a timer that has a requeue pending.  These timers should appear to
719 * be in the timer list with an expiry as if we were to requeue them
720 * now.
721 *
722 * The second issue is the SIGEV_NONE timer which may be active but is
723 * not really ever put in the timer list (to save system resources).
724 * This timer may be expired, and if so, we will do it here.  Otherwise
725 * it is the same as a requeue pending timer WRT to what we should
726 * report.
727 */
728static void
729common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting)
730{
731	ktime_t now, remaining, iv;
732	struct hrtimer *timer = &timr->it.real.timer;
733
734	memset(cur_setting, 0, sizeof(struct itimerspec));
735
736	iv = timr->it.real.interval;
737
738	/* interval timer ? */
739	if (iv.tv64)
740		cur_setting->it_interval = ktime_to_timespec(iv);
741	else if (!hrtimer_active(timer) &&
742		 (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
743		return;
744
745	now = timer->base->get_time();
746
747	/*
748	 * When a requeue is pending or this is a SIGEV_NONE
749	 * timer move the expiry time forward by intervals, so
750	 * expiry is > now.
751	 */
752	if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING ||
753	    (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE))
754		timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv);
755
756	remaining = ktime_sub(hrtimer_get_expires(timer), now);
757	/* Return 0 only, when the timer is expired and not pending */
758	if (remaining.tv64 <= 0) {
759		/*
760		 * A single shot SIGEV_NONE timer must return 0, when
761		 * it is expired !
762		 */
763		if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
764			cur_setting->it_value.tv_nsec = 1;
765	} else
766		cur_setting->it_value = ktime_to_timespec(remaining);
767}
768
769/* Get the time remaining on a POSIX.1b interval timer. */
770SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
771		struct itimerspec __user *, setting)
772{
773	struct itimerspec cur_setting;
774	struct k_itimer *timr;
775	struct k_clock *kc;
776	unsigned long flags;
777	int ret = 0;
778
779	timr = lock_timer(timer_id, &flags);
780	if (!timr)
781		return -EINVAL;
782
783	kc = clockid_to_kclock(timr->it_clock);
784	if (WARN_ON_ONCE(!kc || !kc->timer_get))
785		ret = -EINVAL;
786	else
787		kc->timer_get(timr, &cur_setting);
788
789	unlock_timer(timr, flags);
790
791	if (!ret && copy_to_user(setting, &cur_setting, sizeof (cur_setting)))
792		return -EFAULT;
793
794	return ret;
795}
796
797/*
798 * Get the number of overruns of a POSIX.1b interval timer.  This is to
799 * be the overrun of the timer last delivered.  At the same time we are
800 * accumulating overruns on the next timer.  The overrun is frozen when
801 * the signal is delivered, either at the notify time (if the info block
802 * is not queued) or at the actual delivery time (as we are informed by
803 * the call back to do_schedule_next_timer().  So all we need to do is
804 * to pick up the frozen overrun.
805 */
806SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
807{
808	struct k_itimer *timr;
809	int overrun;
810	unsigned long flags;
811
812	timr = lock_timer(timer_id, &flags);
813	if (!timr)
814		return -EINVAL;
815
816	overrun = timr->it_overrun_last;
817	unlock_timer(timr, flags);
818
819	return overrun;
820}
821
822/* Set a POSIX.1b interval timer. */
823/* timr->it_lock is taken. */
824static int
825common_timer_set(struct k_itimer *timr, int flags,
826		 struct itimerspec *new_setting, struct itimerspec *old_setting)
827{
828	struct hrtimer *timer = &timr->it.real.timer;
829	enum hrtimer_mode mode;
830
831	if (old_setting)
832		common_timer_get(timr, old_setting);
833
834	/* disable the timer */
835	timr->it.real.interval.tv64 = 0;
836	/*
837	 * careful here.  If smp we could be in the "fire" routine which will
838	 * be spinning as we hold the lock.  But this is ONLY an SMP issue.
839	 */
840	if (hrtimer_try_to_cancel(timer) < 0)
841		return TIMER_RETRY;
842
843	timr->it_requeue_pending = (timr->it_requeue_pending + 2) &
844		~REQUEUE_PENDING;
845	timr->it_overrun_last = 0;
846
847	/* switch off the timer when it_value is zero */
848	if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
849		return 0;
850
851	mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
852	hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
853	timr->it.real.timer.function = posix_timer_fn;
854
855	hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value));
856
857	/* Convert interval */
858	timr->it.real.interval = timespec_to_ktime(new_setting->it_interval);
859
860	/* SIGEV_NONE timers are not queued ! See common_timer_get */
861	if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) {
862		/* Setup correct expiry time for relative timers */
863		if (mode == HRTIMER_MODE_REL) {
864			hrtimer_add_expires(timer, timer->base->get_time());
865		}
866		return 0;
867	}
868
869	hrtimer_start_expires(timer, mode);
870	return 0;
871}
872
873/* Set a POSIX.1b interval timer */
874SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
875		const struct itimerspec __user *, new_setting,
876		struct itimerspec __user *, old_setting)
877{
878	struct k_itimer *timr;
879	struct itimerspec new_spec, old_spec;
880	int error = 0;
881	unsigned long flag;
882	struct itimerspec *rtn = old_setting ? &old_spec : NULL;
883	struct k_clock *kc;
884
885	if (!new_setting)
886		return -EINVAL;
887
888	if (copy_from_user(&new_spec, new_setting, sizeof (new_spec)))
889		return -EFAULT;
890
891	if (!timespec_valid(&new_spec.it_interval) ||
892	    !timespec_valid(&new_spec.it_value))
893		return -EINVAL;
894retry:
895	timr = lock_timer(timer_id, &flag);
896	if (!timr)
897		return -EINVAL;
898
899	kc = clockid_to_kclock(timr->it_clock);
900	if (WARN_ON_ONCE(!kc || !kc->timer_set))
901		error = -EINVAL;
902	else
903		error = kc->timer_set(timr, flags, &new_spec, rtn);
904
905	unlock_timer(timr, flag);
906	if (error == TIMER_RETRY) {
907		rtn = NULL;	// We already got the old time...
908		goto retry;
909	}
910
911	if (old_setting && !error &&
912	    copy_to_user(old_setting, &old_spec, sizeof (old_spec)))
913		error = -EFAULT;
914
915	return error;
916}
917
918static int common_timer_del(struct k_itimer *timer)
919{
920	timer->it.real.interval.tv64 = 0;
921
922	if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0)
923		return TIMER_RETRY;
924	return 0;
925}
926
927static inline int timer_delete_hook(struct k_itimer *timer)
928{
929	struct k_clock *kc = clockid_to_kclock(timer->it_clock);
930
931	if (WARN_ON_ONCE(!kc || !kc->timer_del))
932		return -EINVAL;
933	return kc->timer_del(timer);
934}
935
936/* Delete a POSIX.1b interval timer. */
937SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
938{
939	struct k_itimer *timer;
940	unsigned long flags;
941
942retry_delete:
943	timer = lock_timer(timer_id, &flags);
944	if (!timer)
945		return -EINVAL;
946
947	if (timer_delete_hook(timer) == TIMER_RETRY) {
948		unlock_timer(timer, flags);
949		goto retry_delete;
950	}
951
952	spin_lock(&current->sighand->siglock);
953	list_del(&timer->list);
954	spin_unlock(&current->sighand->siglock);
955	/*
956	 * This keeps any tasks waiting on the spin lock from thinking
957	 * they got something (see the lock code above).
958	 */
959	timer->it_signal = NULL;
960
961	unlock_timer(timer, flags);
962	release_posix_timer(timer, IT_ID_SET);
963	return 0;
964}
965
966/*
967 * return timer owned by the process, used by exit_itimers
968 */
969static void itimer_delete(struct k_itimer *timer)
970{
971	unsigned long flags;
972
973retry_delete:
974	spin_lock_irqsave(&timer->it_lock, flags);
975
976	if (timer_delete_hook(timer) == TIMER_RETRY) {
977		unlock_timer(timer, flags);
978		goto retry_delete;
979	}
980	list_del(&timer->list);
981	/*
982	 * This keeps any tasks waiting on the spin lock from thinking
983	 * they got something (see the lock code above).
984	 */
985	timer->it_signal = NULL;
986
987	unlock_timer(timer, flags);
988	release_posix_timer(timer, IT_ID_SET);
989}
990
991/*
992 * This is called by do_exit or de_thread, only when there are no more
993 * references to the shared signal_struct.
994 */
995void exit_itimers(struct signal_struct *sig)
996{
997	struct k_itimer *tmr;
998
999	while (!list_empty(&sig->posix_timers)) {
1000		tmr = list_entry(sig->posix_timers.next, struct k_itimer, list);
1001		itimer_delete(tmr);
1002	}
1003}
1004
1005SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1006		const struct timespec __user *, tp)
1007{
1008	struct k_clock *kc = clockid_to_kclock(which_clock);
1009	struct timespec new_tp;
1010
1011	if (!kc || !kc->clock_set)
1012		return -EINVAL;
1013
1014	if (copy_from_user(&new_tp, tp, sizeof (*tp)))
1015		return -EFAULT;
1016
1017	return kc->clock_set(which_clock, &new_tp);
1018}
1019
1020SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1021		struct timespec __user *,tp)
1022{
1023	struct k_clock *kc = clockid_to_kclock(which_clock);
1024	struct timespec kernel_tp;
1025	int error;
1026
1027	if (!kc)
1028		return -EINVAL;
1029
1030	error = kc->clock_get(which_clock, &kernel_tp);
1031
1032	if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp)))
1033		error = -EFAULT;
1034
1035	return error;
1036}
1037
1038SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1039		struct timex __user *, utx)
1040{
1041	struct k_clock *kc = clockid_to_kclock(which_clock);
1042	struct timex ktx;
1043	int err;
1044
1045	if (!kc)
1046		return -EINVAL;
1047	if (!kc->clock_adj)
1048		return -EOPNOTSUPP;
1049
1050	if (copy_from_user(&ktx, utx, sizeof(ktx)))
1051		return -EFAULT;
1052
1053	err = kc->clock_adj(which_clock, &ktx);
1054
1055	if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1056		return -EFAULT;
1057
1058	return err;
1059}
1060
1061SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1062		struct timespec __user *, tp)
1063{
1064	struct k_clock *kc = clockid_to_kclock(which_clock);
1065	struct timespec rtn_tp;
1066	int error;
1067
1068	if (!kc)
1069		return -EINVAL;
1070
1071	error = kc->clock_getres(which_clock, &rtn_tp);
1072
1073	if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp)))
1074		error = -EFAULT;
1075
1076	return error;
1077}
1078
1079/*
1080 * nanosleep for monotonic and realtime clocks
1081 */
1082static int common_nsleep(const clockid_t which_clock, int flags,
1083			 struct timespec *tsave, struct timespec __user *rmtp)
1084{
1085	return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ?
1086				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1087				 which_clock);
1088}
1089
1090SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1091		const struct timespec __user *, rqtp,
1092		struct timespec __user *, rmtp)
1093{
1094	struct k_clock *kc = clockid_to_kclock(which_clock);
1095	struct timespec t;
1096
1097	if (!kc)
1098		return -EINVAL;
1099	if (!kc->nsleep)
1100		return -ENANOSLEEP_NOTSUP;
1101
1102	if (copy_from_user(&t, rqtp, sizeof (struct timespec)))
1103		return -EFAULT;
1104
1105	if (!timespec_valid(&t))
1106		return -EINVAL;
1107
1108	return kc->nsleep(which_clock, flags, &t, rmtp);
1109}
1110
1111/*
1112 * This will restart clock_nanosleep. This is required only by
1113 * compat_clock_nanosleep_restart for now.
1114 */
1115long clock_nanosleep_restart(struct restart_block *restart_block)
1116{
1117	clockid_t which_clock = restart_block->nanosleep.clockid;
1118	struct k_clock *kc = clockid_to_kclock(which_clock);
1119
1120	if (WARN_ON_ONCE(!kc || !kc->nsleep_restart))
1121		return -EINVAL;
1122
1123	return kc->nsleep_restart(restart_block);
1124}
1125