Loading...
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}
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
38static struct kmem_cache *posix_timers_cache;
39
40/*
41 * Timers are managed in a hash table for lockless lookup. The hash key is
42 * constructed from current::signal and the timer ID and the timer is
43 * matched against current::signal and the timer ID when walking the hash
44 * bucket list.
45 *
46 * This allows checkpoint/restore to reconstruct the exact timer IDs for
47 * a process.
48 */
49static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
50static DEFINE_SPINLOCK(hash_lock);
51
52static const struct k_clock * const posix_clocks[];
53static const struct k_clock *clockid_to_kclock(const clockid_t id);
54static const struct k_clock clock_realtime, clock_monotonic;
55
56/* SIGEV_THREAD_ID cannot share a bit with the other SIGEV values. */
57#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
58 ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
59#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
60#endif
61
62static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
63
64#define lock_timer(tid, flags) \
65({ struct k_itimer *__timr; \
66 __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags)); \
67 __timr; \
68})
69
70static int hash(struct signal_struct *sig, unsigned int nr)
71{
72 return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
73}
74
75static struct k_itimer *__posix_timers_find(struct hlist_head *head,
76 struct signal_struct *sig,
77 timer_t id)
78{
79 struct k_itimer *timer;
80
81 hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) {
82 /* timer->it_signal can be set concurrently */
83 if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id))
84 return timer;
85 }
86 return NULL;
87}
88
89static struct k_itimer *posix_timer_by_id(timer_t id)
90{
91 struct signal_struct *sig = current->signal;
92 struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
93
94 return __posix_timers_find(head, sig, id);
95}
96
97static int posix_timer_add(struct k_itimer *timer)
98{
99 struct signal_struct *sig = current->signal;
100 struct hlist_head *head;
101 unsigned int cnt, id;
102
103 /*
104 * FIXME: Replace this by a per signal struct xarray once there is
105 * a plan to handle the resulting CRIU regression gracefully.
106 */
107 for (cnt = 0; cnt <= INT_MAX; cnt++) {
108 spin_lock(&hash_lock);
109 id = sig->next_posix_timer_id;
110
111 /* Write the next ID back. Clamp it to the positive space */
112 sig->next_posix_timer_id = (id + 1) & INT_MAX;
113
114 head = &posix_timers_hashtable[hash(sig, id)];
115 if (!__posix_timers_find(head, sig, id)) {
116 hlist_add_head_rcu(&timer->t_hash, head);
117 spin_unlock(&hash_lock);
118 return id;
119 }
120 spin_unlock(&hash_lock);
121 }
122 /* POSIX return code when no timer ID could be allocated */
123 return -EAGAIN;
124}
125
126static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
127{
128 spin_unlock_irqrestore(&timr->it_lock, flags);
129}
130
131static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp)
132{
133 ktime_get_real_ts64(tp);
134 return 0;
135}
136
137static ktime_t posix_get_realtime_ktime(clockid_t which_clock)
138{
139 return ktime_get_real();
140}
141
142static int posix_clock_realtime_set(const clockid_t which_clock,
143 const struct timespec64 *tp)
144{
145 return do_sys_settimeofday64(tp, NULL);
146}
147
148static int posix_clock_realtime_adj(const clockid_t which_clock,
149 struct __kernel_timex *t)
150{
151 return do_adjtimex(t);
152}
153
154static int posix_get_monotonic_timespec(clockid_t which_clock, struct timespec64 *tp)
155{
156 ktime_get_ts64(tp);
157 timens_add_monotonic(tp);
158 return 0;
159}
160
161static ktime_t posix_get_monotonic_ktime(clockid_t which_clock)
162{
163 return ktime_get();
164}
165
166static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
167{
168 ktime_get_raw_ts64(tp);
169 timens_add_monotonic(tp);
170 return 0;
171}
172
173static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
174{
175 ktime_get_coarse_real_ts64(tp);
176 return 0;
177}
178
179static int posix_get_monotonic_coarse(clockid_t which_clock,
180 struct timespec64 *tp)
181{
182 ktime_get_coarse_ts64(tp);
183 timens_add_monotonic(tp);
184 return 0;
185}
186
187static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
188{
189 *tp = ktime_to_timespec64(KTIME_LOW_RES);
190 return 0;
191}
192
193static int posix_get_boottime_timespec(const clockid_t which_clock, struct timespec64 *tp)
194{
195 ktime_get_boottime_ts64(tp);
196 timens_add_boottime(tp);
197 return 0;
198}
199
200static ktime_t posix_get_boottime_ktime(const clockid_t which_clock)
201{
202 return ktime_get_boottime();
203}
204
205static int posix_get_tai_timespec(clockid_t which_clock, struct timespec64 *tp)
206{
207 ktime_get_clocktai_ts64(tp);
208 return 0;
209}
210
211static ktime_t posix_get_tai_ktime(clockid_t which_clock)
212{
213 return ktime_get_clocktai();
214}
215
216static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
217{
218 tp->tv_sec = 0;
219 tp->tv_nsec = hrtimer_resolution;
220 return 0;
221}
222
223static __init int init_posix_timers(void)
224{
225 posix_timers_cache = kmem_cache_create("posix_timers_cache",
226 sizeof(struct k_itimer), 0,
227 SLAB_PANIC | SLAB_ACCOUNT, NULL);
228 return 0;
229}
230__initcall(init_posix_timers);
231
232/*
233 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
234 * are of type int. Clamp the overrun value to INT_MAX
235 */
236static inline int timer_overrun_to_int(struct k_itimer *timr)
237{
238 if (timr->it_overrun_last > (s64)INT_MAX)
239 return INT_MAX;
240
241 return (int)timr->it_overrun_last;
242}
243
244static void common_hrtimer_rearm(struct k_itimer *timr)
245{
246 struct hrtimer *timer = &timr->it.real.timer;
247
248 timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
249 timr->it_interval);
250 hrtimer_restart(timer);
251}
252
253static bool __posixtimer_deliver_signal(struct kernel_siginfo *info, struct k_itimer *timr)
254{
255 guard(spinlock)(&timr->it_lock);
256
257 /*
258 * Check if the timer is still alive or whether it got modified
259 * since the signal was queued. In either case, don't rearm and
260 * drop the signal.
261 */
262 if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal))
263 return false;
264
265 if (!timr->it_interval || WARN_ON_ONCE(timr->it_status != POSIX_TIMER_REQUEUE_PENDING))
266 return true;
267
268 timr->kclock->timer_rearm(timr);
269 timr->it_status = POSIX_TIMER_ARMED;
270 timr->it_overrun_last = timr->it_overrun;
271 timr->it_overrun = -1LL;
272 ++timr->it_signal_seq;
273 info->si_overrun = timer_overrun_to_int(timr);
274 return true;
275}
276
277/*
278 * This function is called from the signal delivery code. It decides
279 * whether the signal should be dropped and rearms interval timers. The
280 * timer can be unconditionally accessed as there is a reference held on
281 * it.
282 */
283bool posixtimer_deliver_signal(struct kernel_siginfo *info, struct sigqueue *timer_sigq)
284{
285 struct k_itimer *timr = container_of(timer_sigq, struct k_itimer, sigq);
286 bool ret;
287
288 /*
289 * Release siglock to ensure proper locking order versus
290 * timr::it_lock. Keep interrupts disabled.
291 */
292 spin_unlock(¤t->sighand->siglock);
293
294 ret = __posixtimer_deliver_signal(info, timr);
295
296 /* Drop the reference which was acquired when the signal was queued */
297 posixtimer_putref(timr);
298
299 spin_lock(¤t->sighand->siglock);
300 return ret;
301}
302
303void posix_timer_queue_signal(struct k_itimer *timr)
304{
305 lockdep_assert_held(&timr->it_lock);
306
307 timr->it_status = timr->it_interval ? POSIX_TIMER_REQUEUE_PENDING : POSIX_TIMER_DISARMED;
308 posixtimer_send_sigqueue(timr);
309}
310
311/*
312 * This function gets called when a POSIX.1b interval timer expires from
313 * the HRTIMER interrupt (soft interrupt on RT kernels).
314 *
315 * Handles CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME and CLOCK_TAI
316 * based timers.
317 */
318static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
319{
320 struct k_itimer *timr = container_of(timer, struct k_itimer, it.real.timer);
321
322 guard(spinlock_irqsave)(&timr->it_lock);
323 posix_timer_queue_signal(timr);
324 return HRTIMER_NORESTART;
325}
326
327static struct pid *good_sigevent(sigevent_t * event)
328{
329 struct pid *pid = task_tgid(current);
330 struct task_struct *rtn;
331
332 switch (event->sigev_notify) {
333 case SIGEV_SIGNAL | SIGEV_THREAD_ID:
334 pid = find_vpid(event->sigev_notify_thread_id);
335 rtn = pid_task(pid, PIDTYPE_PID);
336 if (!rtn || !same_thread_group(rtn, current))
337 return NULL;
338 fallthrough;
339 case SIGEV_SIGNAL:
340 case SIGEV_THREAD:
341 if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
342 return NULL;
343 fallthrough;
344 case SIGEV_NONE:
345 return pid;
346 default:
347 return NULL;
348 }
349}
350
351static struct k_itimer *alloc_posix_timer(void)
352{
353 struct k_itimer *tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
354
355 if (!tmr)
356 return tmr;
357
358 if (unlikely(!posixtimer_init_sigqueue(&tmr->sigq))) {
359 kmem_cache_free(posix_timers_cache, tmr);
360 return NULL;
361 }
362 rcuref_init(&tmr->rcuref, 1);
363 return tmr;
364}
365
366void posixtimer_free_timer(struct k_itimer *tmr)
367{
368 put_pid(tmr->it_pid);
369 if (tmr->sigq.ucounts)
370 dec_rlimit_put_ucounts(tmr->sigq.ucounts, UCOUNT_RLIMIT_SIGPENDING);
371 kfree_rcu(tmr, rcu);
372}
373
374static void posix_timer_unhash_and_free(struct k_itimer *tmr)
375{
376 spin_lock(&hash_lock);
377 hlist_del_rcu(&tmr->t_hash);
378 spin_unlock(&hash_lock);
379 posixtimer_putref(tmr);
380}
381
382static int common_timer_create(struct k_itimer *new_timer)
383{
384 hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
385 return 0;
386}
387
388/* Create a POSIX.1b interval timer. */
389static int do_timer_create(clockid_t which_clock, struct sigevent *event,
390 timer_t __user *created_timer_id)
391{
392 const struct k_clock *kc = clockid_to_kclock(which_clock);
393 struct k_itimer *new_timer;
394 int error, new_timer_id;
395
396 if (!kc)
397 return -EINVAL;
398 if (!kc->timer_create)
399 return -EOPNOTSUPP;
400
401 new_timer = alloc_posix_timer();
402 if (unlikely(!new_timer))
403 return -EAGAIN;
404
405 spin_lock_init(&new_timer->it_lock);
406
407 /*
408 * Add the timer to the hash table. The timer is not yet valid
409 * because new_timer::it_signal is still NULL. The timer id is also
410 * not yet visible to user space.
411 */
412 new_timer_id = posix_timer_add(new_timer);
413 if (new_timer_id < 0) {
414 posixtimer_free_timer(new_timer);
415 return new_timer_id;
416 }
417
418 new_timer->it_id = (timer_t) new_timer_id;
419 new_timer->it_clock = which_clock;
420 new_timer->kclock = kc;
421 new_timer->it_overrun = -1LL;
422
423 if (event) {
424 rcu_read_lock();
425 new_timer->it_pid = get_pid(good_sigevent(event));
426 rcu_read_unlock();
427 if (!new_timer->it_pid) {
428 error = -EINVAL;
429 goto out;
430 }
431 new_timer->it_sigev_notify = event->sigev_notify;
432 new_timer->sigq.info.si_signo = event->sigev_signo;
433 new_timer->sigq.info.si_value = event->sigev_value;
434 } else {
435 new_timer->it_sigev_notify = SIGEV_SIGNAL;
436 new_timer->sigq.info.si_signo = SIGALRM;
437 memset(&new_timer->sigq.info.si_value, 0, sizeof(sigval_t));
438 new_timer->sigq.info.si_value.sival_int = new_timer->it_id;
439 new_timer->it_pid = get_pid(task_tgid(current));
440 }
441
442 if (new_timer->it_sigev_notify & SIGEV_THREAD_ID)
443 new_timer->it_pid_type = PIDTYPE_PID;
444 else
445 new_timer->it_pid_type = PIDTYPE_TGID;
446
447 new_timer->sigq.info.si_tid = new_timer->it_id;
448 new_timer->sigq.info.si_code = SI_TIMER;
449
450 if (copy_to_user(created_timer_id, &new_timer_id, sizeof (new_timer_id))) {
451 error = -EFAULT;
452 goto out;
453 }
454 /*
455 * After succesful copy out, the timer ID is visible to user space
456 * now but not yet valid because new_timer::signal is still NULL.
457 *
458 * Complete the initialization with the clock specific create
459 * callback.
460 */
461 error = kc->timer_create(new_timer);
462 if (error)
463 goto out;
464
465 spin_lock_irq(¤t->sighand->siglock);
466 /* This makes the timer valid in the hash table */
467 WRITE_ONCE(new_timer->it_signal, current->signal);
468 hlist_add_head(&new_timer->list, ¤t->signal->posix_timers);
469 spin_unlock_irq(¤t->sighand->siglock);
470 /*
471 * After unlocking sighand::siglock @new_timer is subject to
472 * concurrent removal and cannot be touched anymore
473 */
474 return 0;
475out:
476 posix_timer_unhash_and_free(new_timer);
477 return error;
478}
479
480SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
481 struct sigevent __user *, timer_event_spec,
482 timer_t __user *, created_timer_id)
483{
484 if (timer_event_spec) {
485 sigevent_t event;
486
487 if (copy_from_user(&event, timer_event_spec, sizeof (event)))
488 return -EFAULT;
489 return do_timer_create(which_clock, &event, created_timer_id);
490 }
491 return do_timer_create(which_clock, NULL, created_timer_id);
492}
493
494#ifdef CONFIG_COMPAT
495COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
496 struct compat_sigevent __user *, timer_event_spec,
497 timer_t __user *, created_timer_id)
498{
499 if (timer_event_spec) {
500 sigevent_t event;
501
502 if (get_compat_sigevent(&event, timer_event_spec))
503 return -EFAULT;
504 return do_timer_create(which_clock, &event, created_timer_id);
505 }
506 return do_timer_create(which_clock, NULL, created_timer_id);
507}
508#endif
509
510static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
511{
512 struct k_itimer *timr;
513
514 /*
515 * timer_t could be any type >= int and we want to make sure any
516 * @timer_id outside positive int range fails lookup.
517 */
518 if ((unsigned long long)timer_id > INT_MAX)
519 return NULL;
520
521 /*
522 * The hash lookup and the timers are RCU protected.
523 *
524 * Timers are added to the hash in invalid state where
525 * timr::it_signal == NULL. timer::it_signal is only set after the
526 * rest of the initialization succeeded.
527 *
528 * Timer destruction happens in steps:
529 * 1) Set timr::it_signal to NULL with timr::it_lock held
530 * 2) Release timr::it_lock
531 * 3) Remove from the hash under hash_lock
532 * 4) Put the reference count.
533 *
534 * The reference count might not drop to zero if timr::sigq is
535 * queued. In that case the signal delivery or flush will put the
536 * last reference count.
537 *
538 * When the reference count reaches zero, the timer is scheduled
539 * for RCU removal after the grace period.
540 *
541 * Holding rcu_read_lock() accross the lookup ensures that
542 * the timer cannot be freed.
543 *
544 * The lookup validates locklessly that timr::it_signal ==
545 * current::it_signal and timr::it_id == @timer_id. timr::it_id
546 * can't change, but timr::it_signal becomes NULL during
547 * destruction.
548 */
549 rcu_read_lock();
550 timr = posix_timer_by_id(timer_id);
551 if (timr) {
552 spin_lock_irqsave(&timr->it_lock, *flags);
553 /*
554 * Validate under timr::it_lock that timr::it_signal is
555 * still valid. Pairs with #1 above.
556 */
557 if (timr->it_signal == current->signal) {
558 rcu_read_unlock();
559 return timr;
560 }
561 spin_unlock_irqrestore(&timr->it_lock, *flags);
562 }
563 rcu_read_unlock();
564
565 return NULL;
566}
567
568static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
569{
570 struct hrtimer *timer = &timr->it.real.timer;
571
572 return __hrtimer_expires_remaining_adjusted(timer, now);
573}
574
575static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
576{
577 struct hrtimer *timer = &timr->it.real.timer;
578
579 return hrtimer_forward(timer, now, timr->it_interval);
580}
581
582/*
583 * Get the time remaining on a POSIX.1b interval timer.
584 *
585 * Two issues to handle here:
586 *
587 * 1) The timer has a requeue pending. The return value must appear as
588 * if the timer has been requeued right now.
589 *
590 * 2) The timer is a SIGEV_NONE timer. These timers are never enqueued
591 * into the hrtimer queue and therefore never expired. Emulate expiry
592 * here taking #1 into account.
593 */
594void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
595{
596 const struct k_clock *kc = timr->kclock;
597 ktime_t now, remaining, iv;
598 bool sig_none;
599
600 sig_none = timr->it_sigev_notify == SIGEV_NONE;
601 iv = timr->it_interval;
602
603 /* interval timer ? */
604 if (iv) {
605 cur_setting->it_interval = ktime_to_timespec64(iv);
606 } else if (timr->it_status == POSIX_TIMER_DISARMED) {
607 /*
608 * SIGEV_NONE oneshot timers are never queued and therefore
609 * timr->it_status is always DISARMED. The check below
610 * vs. remaining time will handle this case.
611 *
612 * For all other timers there is nothing to update here, so
613 * return.
614 */
615 if (!sig_none)
616 return;
617 }
618
619 now = kc->clock_get_ktime(timr->it_clock);
620
621 /*
622 * If this is an interval timer and either has requeue pending or
623 * is a SIGEV_NONE timer move the expiry time forward by intervals,
624 * so expiry is > now.
625 */
626 if (iv && timr->it_status != POSIX_TIMER_ARMED)
627 timr->it_overrun += kc->timer_forward(timr, now);
628
629 remaining = kc->timer_remaining(timr, now);
630 /*
631 * As @now is retrieved before a possible timer_forward() and
632 * cannot be reevaluated by the compiler @remaining is based on the
633 * same @now value. Therefore @remaining is consistent vs. @now.
634 *
635 * Consequently all interval timers, i.e. @iv > 0, cannot have a
636 * remaining time <= 0 because timer_forward() guarantees to move
637 * them forward so that the next timer expiry is > @now.
638 */
639 if (remaining <= 0) {
640 /*
641 * A single shot SIGEV_NONE timer must return 0, when it is
642 * expired! Timers which have a real signal delivery mode
643 * must return a remaining time greater than 0 because the
644 * signal has not yet been delivered.
645 */
646 if (!sig_none)
647 cur_setting->it_value.tv_nsec = 1;
648 } else {
649 cur_setting->it_value = ktime_to_timespec64(remaining);
650 }
651}
652
653static int do_timer_gettime(timer_t timer_id, struct itimerspec64 *setting)
654{
655 const struct k_clock *kc;
656 struct k_itimer *timr;
657 unsigned long flags;
658 int ret = 0;
659
660 timr = lock_timer(timer_id, &flags);
661 if (!timr)
662 return -EINVAL;
663
664 memset(setting, 0, sizeof(*setting));
665 kc = timr->kclock;
666 if (WARN_ON_ONCE(!kc || !kc->timer_get))
667 ret = -EINVAL;
668 else
669 kc->timer_get(timr, setting);
670
671 unlock_timer(timr, flags);
672 return ret;
673}
674
675/* Get the time remaining on a POSIX.1b interval timer. */
676SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
677 struct __kernel_itimerspec __user *, setting)
678{
679 struct itimerspec64 cur_setting;
680
681 int ret = do_timer_gettime(timer_id, &cur_setting);
682 if (!ret) {
683 if (put_itimerspec64(&cur_setting, setting))
684 ret = -EFAULT;
685 }
686 return ret;
687}
688
689#ifdef CONFIG_COMPAT_32BIT_TIME
690
691SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
692 struct old_itimerspec32 __user *, setting)
693{
694 struct itimerspec64 cur_setting;
695
696 int ret = do_timer_gettime(timer_id, &cur_setting);
697 if (!ret) {
698 if (put_old_itimerspec32(&cur_setting, setting))
699 ret = -EFAULT;
700 }
701 return ret;
702}
703
704#endif
705
706/**
707 * sys_timer_getoverrun - Get the number of overruns of a POSIX.1b interval timer
708 * @timer_id: The timer ID which identifies the timer
709 *
710 * The "overrun count" of a timer is one plus the number of expiration
711 * intervals which have elapsed between the first expiry, which queues the
712 * signal and the actual signal delivery. On signal delivery the "overrun
713 * count" is calculated and cached, so it can be returned directly here.
714 *
715 * As this is relative to the last queued signal the returned overrun count
716 * is meaningless outside of the signal delivery path and even there it
717 * does not accurately reflect the current state when user space evaluates
718 * it.
719 *
720 * Returns:
721 * -EINVAL @timer_id is invalid
722 * 1..INT_MAX The number of overruns related to the last delivered signal
723 */
724SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
725{
726 struct k_itimer *timr;
727 unsigned long flags;
728 int overrun;
729
730 timr = lock_timer(timer_id, &flags);
731 if (!timr)
732 return -EINVAL;
733
734 overrun = timer_overrun_to_int(timr);
735 unlock_timer(timr, flags);
736
737 return overrun;
738}
739
740static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
741 bool absolute, bool sigev_none)
742{
743 struct hrtimer *timer = &timr->it.real.timer;
744 enum hrtimer_mode mode;
745
746 mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
747 /*
748 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
749 * clock modifications, so they become CLOCK_MONOTONIC based under the
750 * hood. See hrtimer_init(). Update timr->kclock, so the generic
751 * functions which use timr->kclock->clock_get_*() work.
752 *
753 * Note: it_clock stays unmodified, because the next timer_set() might
754 * use ABSTIME, so it needs to switch back.
755 */
756 if (timr->it_clock == CLOCK_REALTIME)
757 timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
758
759 hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
760 timr->it.real.timer.function = posix_timer_fn;
761
762 if (!absolute)
763 expires = ktime_add_safe(expires, timer->base->get_time());
764 hrtimer_set_expires(timer, expires);
765
766 if (!sigev_none)
767 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
768}
769
770static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
771{
772 return hrtimer_try_to_cancel(&timr->it.real.timer);
773}
774
775static void common_timer_wait_running(struct k_itimer *timer)
776{
777 hrtimer_cancel_wait_running(&timer->it.real.timer);
778}
779
780/*
781 * On PREEMPT_RT this prevents priority inversion and a potential livelock
782 * against the ksoftirqd thread in case that ksoftirqd gets preempted while
783 * executing a hrtimer callback.
784 *
785 * See the comments in hrtimer_cancel_wait_running(). For PREEMPT_RT=n this
786 * just results in a cpu_relax().
787 *
788 * For POSIX CPU timers with CONFIG_POSIX_CPU_TIMERS_TASK_WORK=n this is
789 * just a cpu_relax(). With CONFIG_POSIX_CPU_TIMERS_TASK_WORK=y this
790 * prevents spinning on an eventually scheduled out task and a livelock
791 * when the task which tries to delete or disarm the timer has preempted
792 * the task which runs the expiry in task work context.
793 */
794static struct k_itimer *timer_wait_running(struct k_itimer *timer,
795 unsigned long *flags)
796{
797 const struct k_clock *kc = READ_ONCE(timer->kclock);
798 timer_t timer_id = READ_ONCE(timer->it_id);
799
800 /* Prevent kfree(timer) after dropping the lock */
801 rcu_read_lock();
802 unlock_timer(timer, *flags);
803
804 /*
805 * kc->timer_wait_running() might drop RCU lock. So @timer
806 * cannot be touched anymore after the function returns!
807 */
808 if (!WARN_ON_ONCE(!kc->timer_wait_running))
809 kc->timer_wait_running(timer);
810
811 rcu_read_unlock();
812 /* Relock the timer. It might be not longer hashed. */
813 return lock_timer(timer_id, flags);
814}
815
816/*
817 * Set up the new interval and reset the signal delivery data
818 */
819void posix_timer_set_common(struct k_itimer *timer, struct itimerspec64 *new_setting)
820{
821 if (new_setting->it_value.tv_sec || new_setting->it_value.tv_nsec)
822 timer->it_interval = timespec64_to_ktime(new_setting->it_interval);
823 else
824 timer->it_interval = 0;
825
826 /* Reset overrun accounting */
827 timer->it_overrun_last = 0;
828 timer->it_overrun = -1LL;
829}
830
831/* Set a POSIX.1b interval timer. */
832int common_timer_set(struct k_itimer *timr, int flags,
833 struct itimerspec64 *new_setting,
834 struct itimerspec64 *old_setting)
835{
836 const struct k_clock *kc = timr->kclock;
837 bool sigev_none;
838 ktime_t expires;
839
840 if (old_setting)
841 common_timer_get(timr, old_setting);
842
843 /*
844 * Careful here. On SMP systems the timer expiry function could be
845 * active and spinning on timr->it_lock.
846 */
847 if (kc->timer_try_to_cancel(timr) < 0)
848 return TIMER_RETRY;
849
850 timr->it_status = POSIX_TIMER_DISARMED;
851 posix_timer_set_common(timr, new_setting);
852
853 /* Keep timer disarmed when it_value is zero */
854 if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
855 return 0;
856
857 expires = timespec64_to_ktime(new_setting->it_value);
858 if (flags & TIMER_ABSTIME)
859 expires = timens_ktime_to_host(timr->it_clock, expires);
860 sigev_none = timr->it_sigev_notify == SIGEV_NONE;
861
862 kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
863 if (!sigev_none)
864 timr->it_status = POSIX_TIMER_ARMED;
865 return 0;
866}
867
868static int do_timer_settime(timer_t timer_id, int tmr_flags,
869 struct itimerspec64 *new_spec64,
870 struct itimerspec64 *old_spec64)
871{
872 const struct k_clock *kc;
873 struct k_itimer *timr;
874 unsigned long flags;
875 int error;
876
877 if (!timespec64_valid(&new_spec64->it_interval) ||
878 !timespec64_valid(&new_spec64->it_value))
879 return -EINVAL;
880
881 if (old_spec64)
882 memset(old_spec64, 0, sizeof(*old_spec64));
883
884 timr = lock_timer(timer_id, &flags);
885retry:
886 if (!timr)
887 return -EINVAL;
888
889 if (old_spec64)
890 old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
891
892 /* Prevent signal delivery and rearming. */
893 timr->it_signal_seq++;
894
895 kc = timr->kclock;
896 if (WARN_ON_ONCE(!kc || !kc->timer_set))
897 error = -EINVAL;
898 else
899 error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
900
901 if (error == TIMER_RETRY) {
902 // We already got the old time...
903 old_spec64 = NULL;
904 /* Unlocks and relocks the timer if it still exists */
905 timr = timer_wait_running(timr, &flags);
906 goto retry;
907 }
908 unlock_timer(timr, flags);
909
910 return error;
911}
912
913/* Set a POSIX.1b interval timer */
914SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
915 const struct __kernel_itimerspec __user *, new_setting,
916 struct __kernel_itimerspec __user *, old_setting)
917{
918 struct itimerspec64 new_spec, old_spec, *rtn;
919 int error = 0;
920
921 if (!new_setting)
922 return -EINVAL;
923
924 if (get_itimerspec64(&new_spec, new_setting))
925 return -EFAULT;
926
927 rtn = old_setting ? &old_spec : NULL;
928 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
929 if (!error && old_setting) {
930 if (put_itimerspec64(&old_spec, old_setting))
931 error = -EFAULT;
932 }
933 return error;
934}
935
936#ifdef CONFIG_COMPAT_32BIT_TIME
937SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
938 struct old_itimerspec32 __user *, new,
939 struct old_itimerspec32 __user *, old)
940{
941 struct itimerspec64 new_spec, old_spec;
942 struct itimerspec64 *rtn = old ? &old_spec : NULL;
943 int error = 0;
944
945 if (!new)
946 return -EINVAL;
947 if (get_old_itimerspec32(&new_spec, new))
948 return -EFAULT;
949
950 error = do_timer_settime(timer_id, flags, &new_spec, rtn);
951 if (!error && old) {
952 if (put_old_itimerspec32(&old_spec, old))
953 error = -EFAULT;
954 }
955 return error;
956}
957#endif
958
959int common_timer_del(struct k_itimer *timer)
960{
961 const struct k_clock *kc = timer->kclock;
962
963 if (kc->timer_try_to_cancel(timer) < 0)
964 return TIMER_RETRY;
965 timer->it_status = POSIX_TIMER_DISARMED;
966 return 0;
967}
968
969/*
970 * If the deleted timer is on the ignored list, remove it and
971 * drop the associated reference.
972 */
973static inline void posix_timer_cleanup_ignored(struct k_itimer *tmr)
974{
975 if (!hlist_unhashed(&tmr->ignored_list)) {
976 hlist_del_init(&tmr->ignored_list);
977 posixtimer_putref(tmr);
978 }
979}
980
981static inline int timer_delete_hook(struct k_itimer *timer)
982{
983 const struct k_clock *kc = timer->kclock;
984
985 /* Prevent signal delivery and rearming. */
986 timer->it_signal_seq++;
987
988 if (WARN_ON_ONCE(!kc || !kc->timer_del))
989 return -EINVAL;
990 return kc->timer_del(timer);
991}
992
993/* Delete a POSIX.1b interval timer. */
994SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
995{
996 struct k_itimer *timer;
997 unsigned long flags;
998
999 timer = lock_timer(timer_id, &flags);
1000
1001retry_delete:
1002 if (!timer)
1003 return -EINVAL;
1004
1005 if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
1006 /* Unlocks and relocks the timer if it still exists */
1007 timer = timer_wait_running(timer, &flags);
1008 goto retry_delete;
1009 }
1010
1011 spin_lock(¤t->sighand->siglock);
1012 hlist_del(&timer->list);
1013 posix_timer_cleanup_ignored(timer);
1014 /*
1015 * A concurrent lookup could check timer::it_signal lockless. It
1016 * will reevaluate with timer::it_lock held and observe the NULL.
1017 *
1018 * It must be written with siglock held so that the signal code
1019 * observes timer->it_signal == NULL in do_sigaction(SIG_IGN),
1020 * which prevents it from moving a pending signal of a deleted
1021 * timer to the ignore list.
1022 */
1023 WRITE_ONCE(timer->it_signal, NULL);
1024 spin_unlock(¤t->sighand->siglock);
1025
1026 unlock_timer(timer, flags);
1027 posix_timer_unhash_and_free(timer);
1028 return 0;
1029}
1030
1031/*
1032 * Delete a timer if it is armed, remove it from the hash and schedule it
1033 * for RCU freeing.
1034 */
1035static void itimer_delete(struct k_itimer *timer)
1036{
1037 unsigned long flags;
1038
1039 /*
1040 * irqsave is required to make timer_wait_running() work.
1041 */
1042 spin_lock_irqsave(&timer->it_lock, flags);
1043
1044retry_delete:
1045 /*
1046 * Even if the timer is not longer accessible from other tasks
1047 * it still might be armed and queued in the underlying timer
1048 * mechanism. Worse, that timer mechanism might run the expiry
1049 * function concurrently.
1050 */
1051 if (timer_delete_hook(timer) == TIMER_RETRY) {
1052 /*
1053 * Timer is expired concurrently, prevent livelocks
1054 * and pointless spinning on RT.
1055 *
1056 * timer_wait_running() drops timer::it_lock, which opens
1057 * the possibility for another task to delete the timer.
1058 *
1059 * That's not possible here because this is invoked from
1060 * do_exit() only for the last thread of the thread group.
1061 * So no other task can access and delete that timer.
1062 */
1063 if (WARN_ON_ONCE(timer_wait_running(timer, &flags) != timer))
1064 return;
1065
1066 goto retry_delete;
1067 }
1068 hlist_del(&timer->list);
1069
1070 posix_timer_cleanup_ignored(timer);
1071
1072 /*
1073 * Setting timer::it_signal to NULL is technically not required
1074 * here as nothing can access the timer anymore legitimately via
1075 * the hash table. Set it to NULL nevertheless so that all deletion
1076 * paths are consistent.
1077 */
1078 WRITE_ONCE(timer->it_signal, NULL);
1079
1080 spin_unlock_irqrestore(&timer->it_lock, flags);
1081 posix_timer_unhash_and_free(timer);
1082}
1083
1084/*
1085 * Invoked from do_exit() when the last thread of a thread group exits.
1086 * At that point no other task can access the timers of the dying
1087 * task anymore.
1088 */
1089void exit_itimers(struct task_struct *tsk)
1090{
1091 struct hlist_head timers;
1092
1093 if (hlist_empty(&tsk->signal->posix_timers))
1094 return;
1095
1096 /* Protect against concurrent read via /proc/$PID/timers */
1097 spin_lock_irq(&tsk->sighand->siglock);
1098 hlist_move_list(&tsk->signal->posix_timers, &timers);
1099 spin_unlock_irq(&tsk->sighand->siglock);
1100
1101 /* The timers are not longer accessible via tsk::signal */
1102 while (!hlist_empty(&timers))
1103 itimer_delete(hlist_entry(timers.first, struct k_itimer, list));
1104
1105 /*
1106 * There should be no timers on the ignored list. itimer_delete() has
1107 * mopped them up.
1108 */
1109 if (!WARN_ON_ONCE(!hlist_empty(&tsk->signal->ignored_posix_timers)))
1110 return;
1111
1112 hlist_move_list(&tsk->signal->ignored_posix_timers, &timers);
1113 while (!hlist_empty(&timers)) {
1114 posix_timer_cleanup_ignored(hlist_entry(timers.first, struct k_itimer,
1115 ignored_list));
1116 }
1117}
1118
1119SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1120 const struct __kernel_timespec __user *, tp)
1121{
1122 const struct k_clock *kc = clockid_to_kclock(which_clock);
1123 struct timespec64 new_tp;
1124
1125 if (!kc || !kc->clock_set)
1126 return -EINVAL;
1127
1128 if (get_timespec64(&new_tp, tp))
1129 return -EFAULT;
1130
1131 /*
1132 * Permission checks have to be done inside the clock specific
1133 * setter callback.
1134 */
1135 return kc->clock_set(which_clock, &new_tp);
1136}
1137
1138SYSCALL_DEFINE2(clock_gettime, 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 kernel_tp;
1143 int error;
1144
1145 if (!kc)
1146 return -EINVAL;
1147
1148 error = kc->clock_get_timespec(which_clock, &kernel_tp);
1149
1150 if (!error && put_timespec64(&kernel_tp, tp))
1151 error = -EFAULT;
1152
1153 return error;
1154}
1155
1156int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1157{
1158 const struct k_clock *kc = clockid_to_kclock(which_clock);
1159
1160 if (!kc)
1161 return -EINVAL;
1162 if (!kc->clock_adj)
1163 return -EOPNOTSUPP;
1164
1165 return kc->clock_adj(which_clock, ktx);
1166}
1167
1168SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1169 struct __kernel_timex __user *, utx)
1170{
1171 struct __kernel_timex ktx;
1172 int err;
1173
1174 if (copy_from_user(&ktx, utx, sizeof(ktx)))
1175 return -EFAULT;
1176
1177 err = do_clock_adjtime(which_clock, &ktx);
1178
1179 if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1180 return -EFAULT;
1181
1182 return err;
1183}
1184
1185/**
1186 * sys_clock_getres - Get the resolution of a clock
1187 * @which_clock: The clock to get the resolution for
1188 * @tp: Pointer to a a user space timespec64 for storage
1189 *
1190 * POSIX defines:
1191 *
1192 * "The clock_getres() function shall return the resolution of any
1193 * clock. Clock resolutions are implementation-defined and cannot be set by
1194 * a process. If the argument res is not NULL, the resolution of the
1195 * specified clock shall be stored in the location pointed to by res. If
1196 * res is NULL, the clock resolution is not returned. If the time argument
1197 * of clock_settime() is not a multiple of res, then the value is truncated
1198 * to a multiple of res."
1199 *
1200 * Due to the various hardware constraints the real resolution can vary
1201 * wildly and even change during runtime when the underlying devices are
1202 * replaced. The kernel also can use hardware devices with different
1203 * resolutions for reading the time and for arming timers.
1204 *
1205 * The kernel therefore deviates from the POSIX spec in various aspects:
1206 *
1207 * 1) The resolution returned to user space
1208 *
1209 * For CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, CLOCK_TAI,
1210 * CLOCK_REALTIME_ALARM, CLOCK_BOOTTIME_ALAREM and CLOCK_MONOTONIC_RAW
1211 * the kernel differentiates only two cases:
1212 *
1213 * I) Low resolution mode:
1214 *
1215 * When high resolution timers are disabled at compile or runtime
1216 * the resolution returned is nanoseconds per tick, which represents
1217 * the precision at which timers expire.
1218 *
1219 * II) High resolution mode:
1220 *
1221 * When high resolution timers are enabled the resolution returned
1222 * is always one nanosecond independent of the actual resolution of
1223 * the underlying hardware devices.
1224 *
1225 * For CLOCK_*_ALARM the actual resolution depends on system
1226 * state. When system is running the resolution is the same as the
1227 * resolution of the other clocks. During suspend the actual
1228 * resolution is the resolution of the underlying RTC device which
1229 * might be way less precise than the clockevent device used during
1230 * running state.
1231 *
1232 * For CLOCK_REALTIME_COARSE and CLOCK_MONOTONIC_COARSE the resolution
1233 * returned is always nanoseconds per tick.
1234 *
1235 * For CLOCK_PROCESS_CPUTIME and CLOCK_THREAD_CPUTIME the resolution
1236 * returned is always one nanosecond under the assumption that the
1237 * underlying scheduler clock has a better resolution than nanoseconds
1238 * per tick.
1239 *
1240 * For dynamic POSIX clocks (PTP devices) the resolution returned is
1241 * always one nanosecond.
1242 *
1243 * 2) Affect on sys_clock_settime()
1244 *
1245 * The kernel does not truncate the time which is handed in to
1246 * sys_clock_settime(). The kernel internal timekeeping is always using
1247 * nanoseconds precision independent of the clocksource device which is
1248 * used to read the time from. The resolution of that device only
1249 * affects the presicion of the time returned by sys_clock_gettime().
1250 *
1251 * Returns:
1252 * 0 Success. @tp contains the resolution
1253 * -EINVAL @which_clock is not a valid clock ID
1254 * -EFAULT Copying the resolution to @tp faulted
1255 * -ENODEV Dynamic POSIX clock is not backed by a device
1256 * -EOPNOTSUPP Dynamic POSIX clock does not support getres()
1257 */
1258SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1259 struct __kernel_timespec __user *, tp)
1260{
1261 const struct k_clock *kc = clockid_to_kclock(which_clock);
1262 struct timespec64 rtn_tp;
1263 int error;
1264
1265 if (!kc)
1266 return -EINVAL;
1267
1268 error = kc->clock_getres(which_clock, &rtn_tp);
1269
1270 if (!error && tp && put_timespec64(&rtn_tp, tp))
1271 error = -EFAULT;
1272
1273 return error;
1274}
1275
1276#ifdef CONFIG_COMPAT_32BIT_TIME
1277
1278SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1279 struct old_timespec32 __user *, tp)
1280{
1281 const struct k_clock *kc = clockid_to_kclock(which_clock);
1282 struct timespec64 ts;
1283
1284 if (!kc || !kc->clock_set)
1285 return -EINVAL;
1286
1287 if (get_old_timespec32(&ts, tp))
1288 return -EFAULT;
1289
1290 return kc->clock_set(which_clock, &ts);
1291}
1292
1293SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1294 struct old_timespec32 __user *, tp)
1295{
1296 const struct k_clock *kc = clockid_to_kclock(which_clock);
1297 struct timespec64 ts;
1298 int err;
1299
1300 if (!kc)
1301 return -EINVAL;
1302
1303 err = kc->clock_get_timespec(which_clock, &ts);
1304
1305 if (!err && put_old_timespec32(&ts, tp))
1306 err = -EFAULT;
1307
1308 return err;
1309}
1310
1311SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1312 struct old_timex32 __user *, utp)
1313{
1314 struct __kernel_timex ktx;
1315 int err;
1316
1317 err = get_old_timex32(&ktx, utp);
1318 if (err)
1319 return err;
1320
1321 err = do_clock_adjtime(which_clock, &ktx);
1322
1323 if (err >= 0 && put_old_timex32(utp, &ktx))
1324 return -EFAULT;
1325
1326 return err;
1327}
1328
1329SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1330 struct old_timespec32 __user *, tp)
1331{
1332 const struct k_clock *kc = clockid_to_kclock(which_clock);
1333 struct timespec64 ts;
1334 int err;
1335
1336 if (!kc)
1337 return -EINVAL;
1338
1339 err = kc->clock_getres(which_clock, &ts);
1340 if (!err && tp && put_old_timespec32(&ts, tp))
1341 return -EFAULT;
1342
1343 return err;
1344}
1345
1346#endif
1347
1348/*
1349 * sys_clock_nanosleep() for CLOCK_REALTIME and CLOCK_TAI
1350 */
1351static int common_nsleep(const clockid_t which_clock, int flags,
1352 const struct timespec64 *rqtp)
1353{
1354 ktime_t texp = timespec64_to_ktime(*rqtp);
1355
1356 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1357 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1358 which_clock);
1359}
1360
1361/*
1362 * sys_clock_nanosleep() for CLOCK_MONOTONIC and CLOCK_BOOTTIME
1363 *
1364 * Absolute nanosleeps for these clocks are time-namespace adjusted.
1365 */
1366static int common_nsleep_timens(const clockid_t which_clock, int flags,
1367 const struct timespec64 *rqtp)
1368{
1369 ktime_t texp = timespec64_to_ktime(*rqtp);
1370
1371 if (flags & TIMER_ABSTIME)
1372 texp = timens_ktime_to_host(which_clock, texp);
1373
1374 return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1375 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1376 which_clock);
1377}
1378
1379SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1380 const struct __kernel_timespec __user *, rqtp,
1381 struct __kernel_timespec __user *, rmtp)
1382{
1383 const struct k_clock *kc = clockid_to_kclock(which_clock);
1384 struct timespec64 t;
1385
1386 if (!kc)
1387 return -EINVAL;
1388 if (!kc->nsleep)
1389 return -EOPNOTSUPP;
1390
1391 if (get_timespec64(&t, rqtp))
1392 return -EFAULT;
1393
1394 if (!timespec64_valid(&t))
1395 return -EINVAL;
1396 if (flags & TIMER_ABSTIME)
1397 rmtp = NULL;
1398 current->restart_block.fn = do_no_restart_syscall;
1399 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1400 current->restart_block.nanosleep.rmtp = rmtp;
1401
1402 return kc->nsleep(which_clock, flags, &t);
1403}
1404
1405#ifdef CONFIG_COMPAT_32BIT_TIME
1406
1407SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1408 struct old_timespec32 __user *, rqtp,
1409 struct old_timespec32 __user *, rmtp)
1410{
1411 const struct k_clock *kc = clockid_to_kclock(which_clock);
1412 struct timespec64 t;
1413
1414 if (!kc)
1415 return -EINVAL;
1416 if (!kc->nsleep)
1417 return -EOPNOTSUPP;
1418
1419 if (get_old_timespec32(&t, rqtp))
1420 return -EFAULT;
1421
1422 if (!timespec64_valid(&t))
1423 return -EINVAL;
1424 if (flags & TIMER_ABSTIME)
1425 rmtp = NULL;
1426 current->restart_block.fn = do_no_restart_syscall;
1427 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1428 current->restart_block.nanosleep.compat_rmtp = rmtp;
1429
1430 return kc->nsleep(which_clock, flags, &t);
1431}
1432
1433#endif
1434
1435static const struct k_clock clock_realtime = {
1436 .clock_getres = posix_get_hrtimer_res,
1437 .clock_get_timespec = posix_get_realtime_timespec,
1438 .clock_get_ktime = posix_get_realtime_ktime,
1439 .clock_set = posix_clock_realtime_set,
1440 .clock_adj = posix_clock_realtime_adj,
1441 .nsleep = common_nsleep,
1442 .timer_create = common_timer_create,
1443 .timer_set = common_timer_set,
1444 .timer_get = common_timer_get,
1445 .timer_del = common_timer_del,
1446 .timer_rearm = common_hrtimer_rearm,
1447 .timer_forward = common_hrtimer_forward,
1448 .timer_remaining = common_hrtimer_remaining,
1449 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1450 .timer_wait_running = common_timer_wait_running,
1451 .timer_arm = common_hrtimer_arm,
1452};
1453
1454static const struct k_clock clock_monotonic = {
1455 .clock_getres = posix_get_hrtimer_res,
1456 .clock_get_timespec = posix_get_monotonic_timespec,
1457 .clock_get_ktime = posix_get_monotonic_ktime,
1458 .nsleep = common_nsleep_timens,
1459 .timer_create = common_timer_create,
1460 .timer_set = common_timer_set,
1461 .timer_get = common_timer_get,
1462 .timer_del = common_timer_del,
1463 .timer_rearm = common_hrtimer_rearm,
1464 .timer_forward = common_hrtimer_forward,
1465 .timer_remaining = common_hrtimer_remaining,
1466 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1467 .timer_wait_running = common_timer_wait_running,
1468 .timer_arm = common_hrtimer_arm,
1469};
1470
1471static const struct k_clock clock_monotonic_raw = {
1472 .clock_getres = posix_get_hrtimer_res,
1473 .clock_get_timespec = posix_get_monotonic_raw,
1474};
1475
1476static const struct k_clock clock_realtime_coarse = {
1477 .clock_getres = posix_get_coarse_res,
1478 .clock_get_timespec = posix_get_realtime_coarse,
1479};
1480
1481static const struct k_clock clock_monotonic_coarse = {
1482 .clock_getres = posix_get_coarse_res,
1483 .clock_get_timespec = posix_get_monotonic_coarse,
1484};
1485
1486static const struct k_clock clock_tai = {
1487 .clock_getres = posix_get_hrtimer_res,
1488 .clock_get_ktime = posix_get_tai_ktime,
1489 .clock_get_timespec = posix_get_tai_timespec,
1490 .nsleep = common_nsleep,
1491 .timer_create = common_timer_create,
1492 .timer_set = common_timer_set,
1493 .timer_get = common_timer_get,
1494 .timer_del = common_timer_del,
1495 .timer_rearm = common_hrtimer_rearm,
1496 .timer_forward = common_hrtimer_forward,
1497 .timer_remaining = common_hrtimer_remaining,
1498 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1499 .timer_wait_running = common_timer_wait_running,
1500 .timer_arm = common_hrtimer_arm,
1501};
1502
1503static const struct k_clock clock_boottime = {
1504 .clock_getres = posix_get_hrtimer_res,
1505 .clock_get_ktime = posix_get_boottime_ktime,
1506 .clock_get_timespec = posix_get_boottime_timespec,
1507 .nsleep = common_nsleep_timens,
1508 .timer_create = common_timer_create,
1509 .timer_set = common_timer_set,
1510 .timer_get = common_timer_get,
1511 .timer_del = common_timer_del,
1512 .timer_rearm = common_hrtimer_rearm,
1513 .timer_forward = common_hrtimer_forward,
1514 .timer_remaining = common_hrtimer_remaining,
1515 .timer_try_to_cancel = common_hrtimer_try_to_cancel,
1516 .timer_wait_running = common_timer_wait_running,
1517 .timer_arm = common_hrtimer_arm,
1518};
1519
1520static const struct k_clock * const posix_clocks[] = {
1521 [CLOCK_REALTIME] = &clock_realtime,
1522 [CLOCK_MONOTONIC] = &clock_monotonic,
1523 [CLOCK_PROCESS_CPUTIME_ID] = &clock_process,
1524 [CLOCK_THREAD_CPUTIME_ID] = &clock_thread,
1525 [CLOCK_MONOTONIC_RAW] = &clock_monotonic_raw,
1526 [CLOCK_REALTIME_COARSE] = &clock_realtime_coarse,
1527 [CLOCK_MONOTONIC_COARSE] = &clock_monotonic_coarse,
1528 [CLOCK_BOOTTIME] = &clock_boottime,
1529 [CLOCK_REALTIME_ALARM] = &alarm_clock,
1530 [CLOCK_BOOTTIME_ALARM] = &alarm_clock,
1531 [CLOCK_TAI] = &clock_tai,
1532};
1533
1534static const struct k_clock *clockid_to_kclock(const clockid_t id)
1535{
1536 clockid_t idx = id;
1537
1538 if (id < 0) {
1539 return (id & CLOCKFD_MASK) == CLOCKFD ?
1540 &clock_posix_dynamic : &clock_posix_cpu;
1541 }
1542
1543 if (id >= ARRAY_SIZE(posix_clocks))
1544 return NULL;
1545
1546 return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1547}