Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0+
   2/*
   3 * 2002-10-15  Posix Clocks & timers
   4 *                           by George Anzinger george@mvista.com
   5 *			     Copyright (C) 2002 2003 by MontaVista Software.
   6 *
   7 * 2004-06-01  Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug.
   8 *			     Copyright (C) 2004 Boris Hu
   9 *
  10 * These are all the functions necessary to implement POSIX clocks & timers
  11 */
  12#include <linux/mm.h>
  13#include <linux/interrupt.h>
  14#include <linux/slab.h>
  15#include <linux/time.h>
  16#include <linux/mutex.h>
  17#include <linux/sched/task.h>
  18
  19#include <linux/uaccess.h>
  20#include <linux/list.h>
  21#include <linux/init.h>
  22#include <linux/compiler.h>
  23#include <linux/hash.h>
  24#include <linux/posix-clock.h>
  25#include <linux/posix-timers.h>
  26#include <linux/syscalls.h>
  27#include <linux/wait.h>
  28#include <linux/workqueue.h>
  29#include <linux/export.h>
  30#include <linux/hashtable.h>
  31#include <linux/compat.h>
  32#include <linux/nospec.h>
 
  33
  34#include "timekeeping.h"
  35#include "posix-timers.h"
  36
  37/*
  38 * Management arrays for POSIX timers. Timers are now kept in static hash table
  39 * with 512 entries.
  40 * Timer ids are allocated by local routine, which selects proper hash head by
  41 * key, constructed from current->signal address and per signal struct counter.
  42 * This keeps timer ids unique per process, but now they can intersect between
  43 * processes.
  44 */
  45
  46/*
  47 * Lets keep our timers in a slab cache :-)
 
 
 
 
 
 
  48 */
  49static struct kmem_cache *posix_timers_cache;
  50
  51static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
  52static DEFINE_SPINLOCK(hash_lock);
  53
  54static const struct k_clock * const posix_clocks[];
  55static const struct k_clock *clockid_to_kclock(const clockid_t id);
  56static const struct k_clock clock_realtime, clock_monotonic;
  57
  58/*
  59 * we assume that the new SIGEV_THREAD_ID shares no bits with the other
  60 * SIGEV values.  Here we put out an error if this assumption fails.
  61 */
  62#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
  63                       ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
  64#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
  65#endif
  66
  67/*
  68 * The timer ID is turned into a timer address by idr_find().
  69 * Verifying a valid ID consists of:
  70 *
  71 * a) checking that idr_find() returns other than -1.
  72 * b) checking that the timer id matches the one in the timer itself.
  73 * c) that the timer owner is in the callers thread group.
  74 */
  75
  76/*
  77 * CLOCKs: The POSIX standard calls for a couple of clocks and allows us
  78 *	    to implement others.  This structure defines the various
  79 *	    clocks.
  80 *
  81 * RESOLUTION: Clock resolution is used to round up timer and interval
  82 *	    times, NOT to report clock times, which are reported with as
  83 *	    much resolution as the system can muster.  In some cases this
  84 *	    resolution may depend on the underlying clock hardware and
  85 *	    may not be quantifiable until run time, and only then is the
  86 *	    necessary code is written.	The standard says we should say
  87 *	    something about this issue in the documentation...
  88 *
  89 * FUNCTIONS: The CLOCKs structure defines possible functions to
  90 *	    handle various clock functions.
  91 *
  92 *	    The standard POSIX timer management code assumes the
  93 *	    following: 1.) The k_itimer struct (sched.h) is used for
  94 *	    the timer.  2.) The list, it_lock, it_clock, it_id and
  95 *	    it_pid fields are not modified by timer code.
  96 *
  97 * Permissions: It is assumed that the clock_settime() function defined
  98 *	    for each clock will take care of permission checks.	 Some
  99 *	    clocks may be set able by any user (i.e. local process
 100 *	    clocks) others not.	 Currently the only set able clock we
 101 *	    have is CLOCK_REALTIME and its high res counter part, both of
 102 *	    which we beg off on and pass to do_sys_settimeofday().
 103 */
 104static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
 105
 106#define lock_timer(tid, flags)						   \
 107({	struct k_itimer *__timr;					   \
 108	__cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags));  \
 109	__timr;								   \
 110})
 111
 112static int hash(struct signal_struct *sig, unsigned int nr)
 113{
 114	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
 115}
 116
 117static struct k_itimer *__posix_timers_find(struct hlist_head *head,
 118					    struct signal_struct *sig,
 119					    timer_t id)
 120{
 121	struct k_itimer *timer;
 122
 123	hlist_for_each_entry_rcu(timer, head, t_hash) {
 124		if ((timer->it_signal == sig) && (timer->it_id == id))
 
 125			return timer;
 126	}
 127	return NULL;
 128}
 129
 130static struct k_itimer *posix_timer_by_id(timer_t id)
 131{
 132	struct signal_struct *sig = current->signal;
 133	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
 134
 135	return __posix_timers_find(head, sig, id);
 136}
 137
 138static int posix_timer_add(struct k_itimer *timer)
 139{
 140	struct signal_struct *sig = current->signal;
 141	int first_free_id = sig->posix_timer_id;
 142	struct hlist_head *head;
 143	int ret = -ENOENT;
 144
 145	do {
 
 
 
 
 146		spin_lock(&hash_lock);
 147		head = &posix_timers_hashtable[hash(sig, sig->posix_timer_id)];
 148		if (!__posix_timers_find(head, sig, sig->posix_timer_id)) {
 
 
 
 
 
 149			hlist_add_head_rcu(&timer->t_hash, head);
 150			ret = sig->posix_timer_id;
 
 151		}
 152		if (++sig->posix_timer_id < 0)
 153			sig->posix_timer_id = 0;
 154		if ((sig->posix_timer_id == first_free_id) && (ret == -ENOENT))
 155			/* Loop over all possible ids completed */
 156			ret = -EAGAIN;
 157		spin_unlock(&hash_lock);
 158	} while (ret == -ENOENT);
 159	return ret;
 
 160}
 161
 162static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
 163{
 164	spin_unlock_irqrestore(&timr->it_lock, flags);
 165}
 166
 167/* Get clock_realtime */
 168static int posix_clock_realtime_get(clockid_t which_clock, struct timespec64 *tp)
 169{
 170	ktime_get_real_ts64(tp);
 171	return 0;
 172}
 173
 174/* Set clock_realtime */
 
 
 
 
 175static int posix_clock_realtime_set(const clockid_t which_clock,
 176				    const struct timespec64 *tp)
 177{
 178	return do_sys_settimeofday64(tp, NULL);
 179}
 180
 181static int posix_clock_realtime_adj(const clockid_t which_clock,
 182				    struct __kernel_timex *t)
 183{
 184	return do_adjtimex(t);
 185}
 186
 187/*
 188 * Get monotonic time for posix timers
 189 */
 190static int posix_ktime_get_ts(clockid_t which_clock, struct timespec64 *tp)
 191{
 192	ktime_get_ts64(tp);
 
 193	return 0;
 194}
 195
 196/*
 197 * Get monotonic-raw time for posix timers
 198 */
 
 
 199static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
 200{
 201	ktime_get_raw_ts64(tp);
 
 202	return 0;
 203}
 204
 205
 206static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
 207{
 208	ktime_get_coarse_real_ts64(tp);
 209	return 0;
 210}
 211
 212static int posix_get_monotonic_coarse(clockid_t which_clock,
 213						struct timespec64 *tp)
 214{
 215	ktime_get_coarse_ts64(tp);
 
 216	return 0;
 217}
 218
 219static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
 220{
 221	*tp = ktime_to_timespec64(KTIME_LOW_RES);
 222	return 0;
 223}
 224
 225static int posix_get_boottime(const clockid_t which_clock, struct timespec64 *tp)
 226{
 227	ktime_get_boottime_ts64(tp);
 
 228	return 0;
 229}
 230
 231static int posix_get_tai(clockid_t which_clock, struct timespec64 *tp)
 
 
 
 
 
 232{
 233	ktime_get_clocktai_ts64(tp);
 234	return 0;
 235}
 236
 
 
 
 
 
 237static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
 238{
 239	tp->tv_sec = 0;
 240	tp->tv_nsec = hrtimer_resolution;
 241	return 0;
 242}
 243
 244/*
 245 * Initialize everything, well, just everything in Posix clocks/timers ;)
 246 */
 247static __init int init_posix_timers(void)
 248{
 249	posix_timers_cache = kmem_cache_create("posix_timers_cache",
 250					sizeof (struct k_itimer), 0, SLAB_PANIC,
 251					NULL);
 252	return 0;
 253}
 254__initcall(init_posix_timers);
 255
 256/*
 257 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
 258 * are of type int. Clamp the overrun value to INT_MAX
 259 */
 260static inline int timer_overrun_to_int(struct k_itimer *timr, int baseval)
 261{
 262	s64 sum = timr->it_overrun_last + (s64)baseval;
 263
 264	return sum > (s64)INT_MAX ? INT_MAX : (int)sum;
 265}
 266
 267static void common_hrtimer_rearm(struct k_itimer *timr)
 268{
 269	struct hrtimer *timer = &timr->it.real.timer;
 270
 271	timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
 272					    timr->it_interval);
 273	hrtimer_restart(timer);
 274}
 275
 276/*
 277 * This function is exported for use by the signal deliver code.  It is
 278 * called just prior to the info block being released and passes that
 279 * block to us.  It's function is to update the overrun entry AND to
 280 * restart the timer.  It should only be called if the timer is to be
 281 * restarted (i.e. we have flagged this in the sys_private entry of the
 282 * info block).
 283 *
 284 * To protect against the timer going away while the interrupt is queued,
 285 * we require that the it_requeue_pending flag be set.
 286 */
 287void posixtimer_rearm(struct kernel_siginfo *info)
 288{
 289	struct k_itimer *timr;
 290	unsigned long flags;
 291
 292	timr = lock_timer(info->si_tid, &flags);
 293	if (!timr)
 294		return;
 295
 296	if (timr->it_interval && timr->it_requeue_pending == info->si_sys_private) {
 297		timr->kclock->timer_rearm(timr);
 298
 299		timr->it_active = 1;
 300		timr->it_overrun_last = timr->it_overrun;
 301		timr->it_overrun = -1LL;
 302		++timr->it_requeue_pending;
 303
 304		info->si_overrun = timer_overrun_to_int(timr, info->si_overrun);
 305	}
 306
 307	unlock_timer(timr, flags);
 308}
 309
 310int posix_timer_event(struct k_itimer *timr, int si_private)
 311{
 312	enum pid_type type;
 313	int ret = -1;
 314	/*
 315	 * FIXME: if ->sigq is queued we can race with
 316	 * dequeue_signal()->posixtimer_rearm().
 317	 *
 318	 * If dequeue_signal() sees the "right" value of
 319	 * si_sys_private it calls posixtimer_rearm().
 320	 * We re-queue ->sigq and drop ->it_lock().
 321	 * posixtimer_rearm() locks the timer
 322	 * and re-schedules it while ->sigq is pending.
 323	 * Not really bad, but not that we want.
 324	 */
 325	timr->sigq->info.si_sys_private = si_private;
 326
 327	type = !(timr->it_sigev_notify & SIGEV_THREAD_ID) ? PIDTYPE_TGID : PIDTYPE_PID;
 328	ret = send_sigqueue(timr->sigq, timr->it_pid, type);
 329	/* If we failed to send the signal the timer stops. */
 330	return ret > 0;
 331}
 332
 333/*
 334 * This function gets called when a POSIX.1b interval timer expires.  It
 335 * is used as a callback from the kernel internal timer.  The
 336 * run_timer_list code ALWAYS calls with interrupts on.
 337
 338 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
 339 */
 340static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
 341{
 
 342	struct k_itimer *timr;
 343	unsigned long flags;
 344	int si_private = 0;
 345	enum hrtimer_restart ret = HRTIMER_NORESTART;
 346
 347	timr = container_of(timer, struct k_itimer, it.real.timer);
 348	spin_lock_irqsave(&timr->it_lock, flags);
 349
 350	timr->it_active = 0;
 351	if (timr->it_interval != 0)
 352		si_private = ++timr->it_requeue_pending;
 353
 354	if (posix_timer_event(timr, si_private)) {
 355		/*
 356		 * signal was not sent because of sig_ignor
 357		 * we will not get a call back to restart it AND
 358		 * it should be restarted.
 
 359		 */
 360		if (timr->it_interval != 0) {
 361			ktime_t now = hrtimer_cb_get_time(timer);
 362
 363			/*
 364			 * FIXME: What we really want, is to stop this
 365			 * timer completely and restart it in case the
 366			 * SIG_IGN is removed. This is a non trivial
 367			 * change which involves sighand locking
 368			 * (sigh !), which we don't want to do late in
 369			 * the release cycle.
 
 
 
 
 
 
 370			 *
 371			 * For now we just let timers with an interval
 372			 * less than a jiffie expire every jiffie to
 373			 * avoid softirq starvation in case of SIG_IGN
 374			 * and a very small interval, which would put
 375			 * the timer right back on the softirq pending
 376			 * list. By moving now ahead of time we trick
 377			 * hrtimer_forward() to expire the timer
 378			 * later, while we still maintain the overrun
 379			 * accuracy, but have some inconsistency in
 380			 * the timer_gettime() case. This is at least
 381			 * better than a starved softirq. A more
 382			 * complex fix which solves also another related
 383			 * inconsistency is already in the pipeline.
 384			 */
 385#ifdef CONFIG_HIGH_RES_TIMERS
 386			{
 387				ktime_t kj = NSEC_PER_SEC / HZ;
 388
 389				if (timr->it_interval < kj)
 390					now = ktime_add(now, kj);
 391			}
 392#endif
 393			timr->it_overrun += hrtimer_forward(timer, now,
 394							    timr->it_interval);
 395			ret = HRTIMER_RESTART;
 396			++timr->it_requeue_pending;
 397			timr->it_active = 1;
 398		}
 399	}
 400
 401	unlock_timer(timr, flags);
 402	return ret;
 403}
 404
 405static struct pid *good_sigevent(sigevent_t * event)
 406{
 407	struct pid *pid = task_tgid(current);
 408	struct task_struct *rtn;
 409
 410	switch (event->sigev_notify) {
 411	case SIGEV_SIGNAL | SIGEV_THREAD_ID:
 412		pid = find_vpid(event->sigev_notify_thread_id);
 413		rtn = pid_task(pid, PIDTYPE_PID);
 414		if (!rtn || !same_thread_group(rtn, current))
 415			return NULL;
 416		/* FALLTHRU */
 417	case SIGEV_SIGNAL:
 418	case SIGEV_THREAD:
 419		if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
 420			return NULL;
 421		/* FALLTHRU */
 422	case SIGEV_NONE:
 423		return pid;
 424	default:
 425		return NULL;
 426	}
 427}
 428
 429static struct k_itimer * alloc_posix_timer(void)
 430{
 431	struct k_itimer *tmr;
 432	tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
 433	if (!tmr)
 434		return tmr;
 435	if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
 436		kmem_cache_free(posix_timers_cache, tmr);
 437		return NULL;
 438	}
 439	clear_siginfo(&tmr->sigq->info);
 440	return tmr;
 441}
 442
 443static void k_itimer_rcu_free(struct rcu_head *head)
 444{
 445	struct k_itimer *tmr = container_of(head, struct k_itimer, rcu);
 446
 447	kmem_cache_free(posix_timers_cache, tmr);
 448}
 449
 450#define IT_ID_SET	1
 451#define IT_ID_NOT_SET	0
 452static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
 453{
 454	if (it_id_set) {
 455		unsigned long flags;
 456		spin_lock_irqsave(&hash_lock, flags);
 457		hlist_del_rcu(&tmr->t_hash);
 458		spin_unlock_irqrestore(&hash_lock, flags);
 459	}
 460	put_pid(tmr->it_pid);
 461	sigqueue_free(tmr->sigq);
 462	call_rcu(&tmr->rcu, k_itimer_rcu_free);
 463}
 464
 
 
 
 
 
 
 
 
 465static int common_timer_create(struct k_itimer *new_timer)
 466{
 467	hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
 468	return 0;
 469}
 470
 471/* Create a POSIX.1b interval timer. */
 472static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 473			   timer_t __user *created_timer_id)
 474{
 475	const struct k_clock *kc = clockid_to_kclock(which_clock);
 476	struct k_itimer *new_timer;
 477	int error, new_timer_id;
 478	int it_id_set = IT_ID_NOT_SET;
 479
 480	if (!kc)
 481		return -EINVAL;
 482	if (!kc->timer_create)
 483		return -EOPNOTSUPP;
 484
 485	new_timer = alloc_posix_timer();
 486	if (unlikely(!new_timer))
 487		return -EAGAIN;
 488
 489	spin_lock_init(&new_timer->it_lock);
 
 
 
 
 
 
 490	new_timer_id = posix_timer_add(new_timer);
 491	if (new_timer_id < 0) {
 492		error = new_timer_id;
 493		goto out;
 494	}
 495
 496	it_id_set = IT_ID_SET;
 497	new_timer->it_id = (timer_t) new_timer_id;
 498	new_timer->it_clock = which_clock;
 499	new_timer->kclock = kc;
 500	new_timer->it_overrun = -1LL;
 501
 502	if (event) {
 503		rcu_read_lock();
 504		new_timer->it_pid = get_pid(good_sigevent(event));
 505		rcu_read_unlock();
 506		if (!new_timer->it_pid) {
 507			error = -EINVAL;
 508			goto out;
 509		}
 510		new_timer->it_sigev_notify     = event->sigev_notify;
 511		new_timer->sigq->info.si_signo = event->sigev_signo;
 512		new_timer->sigq->info.si_value = event->sigev_value;
 513	} else {
 514		new_timer->it_sigev_notify     = SIGEV_SIGNAL;
 515		new_timer->sigq->info.si_signo = SIGALRM;
 516		memset(&new_timer->sigq->info.si_value, 0, sizeof(sigval_t));
 517		new_timer->sigq->info.si_value.sival_int = new_timer->it_id;
 518		new_timer->it_pid = get_pid(task_tgid(current));
 519	}
 520
 521	new_timer->sigq->info.si_tid   = new_timer->it_id;
 522	new_timer->sigq->info.si_code  = SI_TIMER;
 523
 524	if (copy_to_user(created_timer_id,
 525			 &new_timer_id, sizeof (new_timer_id))) {
 526		error = -EFAULT;
 527		goto out;
 528	}
 529
 
 
 
 
 
 
 530	error = kc->timer_create(new_timer);
 531	if (error)
 532		goto out;
 533
 534	spin_lock_irq(&current->sighand->siglock);
 535	new_timer->it_signal = current->signal;
 
 536	list_add(&new_timer->list, &current->signal->posix_timers);
 537	spin_unlock_irq(&current->sighand->siglock);
 538
 539	return 0;
 540	/*
 541	 * In the case of the timer belonging to another task, after
 542	 * the task is unlocked, the timer is owned by the other task
 543	 * and may cease to exist at any time.  Don't use or modify
 544	 * new_timer after the unlock call.
 545	 */
 
 546out:
 547	release_posix_timer(new_timer, it_id_set);
 548	return error;
 549}
 550
 551SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
 552		struct sigevent __user *, timer_event_spec,
 553		timer_t __user *, created_timer_id)
 554{
 555	if (timer_event_spec) {
 556		sigevent_t event;
 557
 558		if (copy_from_user(&event, timer_event_spec, sizeof (event)))
 559			return -EFAULT;
 560		return do_timer_create(which_clock, &event, created_timer_id);
 561	}
 562	return do_timer_create(which_clock, NULL, created_timer_id);
 563}
 564
 565#ifdef CONFIG_COMPAT
 566COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
 567		       struct compat_sigevent __user *, timer_event_spec,
 568		       timer_t __user *, created_timer_id)
 569{
 570	if (timer_event_spec) {
 571		sigevent_t event;
 572
 573		if (get_compat_sigevent(&event, timer_event_spec))
 574			return -EFAULT;
 575		return do_timer_create(which_clock, &event, created_timer_id);
 576	}
 577	return do_timer_create(which_clock, NULL, created_timer_id);
 578}
 579#endif
 580
 581/*
 582 * Locking issues: We need to protect the result of the id look up until
 583 * we get the timer locked down so it is not deleted under us.  The
 584 * removal is done under the idr spinlock so we use that here to bridge
 585 * the find to the timer lock.  To avoid a dead lock, the timer id MUST
 586 * be release with out holding the timer lock.
 587 */
 588static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 589{
 590	struct k_itimer *timr;
 591
 592	/*
 593	 * timer_t could be any type >= int and we want to make sure any
 594	 * @timer_id outside positive int range fails lookup.
 595	 */
 596	if ((unsigned long long)timer_id > INT_MAX)
 597		return NULL;
 598
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 599	rcu_read_lock();
 600	timr = posix_timer_by_id(timer_id);
 601	if (timr) {
 602		spin_lock_irqsave(&timr->it_lock, *flags);
 
 
 
 
 603		if (timr->it_signal == current->signal) {
 604			rcu_read_unlock();
 605			return timr;
 606		}
 607		spin_unlock_irqrestore(&timr->it_lock, *flags);
 608	}
 609	rcu_read_unlock();
 610
 611	return NULL;
 612}
 613
 614static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
 615{
 616	struct hrtimer *timer = &timr->it.real.timer;
 617
 618	return __hrtimer_expires_remaining_adjusted(timer, now);
 619}
 620
 621static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
 622{
 623	struct hrtimer *timer = &timr->it.real.timer;
 624
 625	return hrtimer_forward(timer, now, timr->it_interval);
 626}
 627
 628/*
 629 * Get the time remaining on a POSIX.1b interval timer.  This function
 630 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
 631 * mess with irq.
 632 *
 633 * We have a couple of messes to clean up here.  First there is the case
 634 * of a timer that has a requeue pending.  These timers should appear to
 635 * be in the timer list with an expiry as if we were to requeue them
 636 * now.
 637 *
 638 * The second issue is the SIGEV_NONE timer which may be active but is
 639 * not really ever put in the timer list (to save system resources).
 640 * This timer may be expired, and if so, we will do it here.  Otherwise
 641 * it is the same as a requeue pending timer WRT to what we should
 642 * report.
 643 */
 644void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
 645{
 646	const struct k_clock *kc = timr->kclock;
 647	ktime_t now, remaining, iv;
 648	struct timespec64 ts64;
 649	bool sig_none;
 650
 651	sig_none = timr->it_sigev_notify == SIGEV_NONE;
 652	iv = timr->it_interval;
 653
 654	/* interval timer ? */
 655	if (iv) {
 656		cur_setting->it_interval = ktime_to_timespec64(iv);
 657	} else if (!timr->it_active) {
 658		/*
 659		 * SIGEV_NONE oneshot timers are never queued. Check them
 660		 * below.
 
 
 
 
 661		 */
 662		if (!sig_none)
 663			return;
 664	}
 665
 666	/*
 667	 * The timespec64 based conversion is suboptimal, but it's not
 668	 * worth to implement yet another callback.
 669	 */
 670	kc->clock_get(timr->it_clock, &ts64);
 671	now = timespec64_to_ktime(ts64);
 672
 673	/*
 674	 * When a requeue is pending or this is a SIGEV_NONE timer move the
 675	 * expiry time forward by intervals, so expiry is > now.
 
 676	 */
 677	if (iv && (timr->it_requeue_pending & REQUEUE_PENDING || sig_none))
 678		timr->it_overrun += kc->timer_forward(timr, now);
 679
 680	remaining = kc->timer_remaining(timr, now);
 681	/* Return 0 only, when the timer is expired and not pending */
 
 
 
 
 
 
 
 
 682	if (remaining <= 0) {
 683		/*
 684		 * A single shot SIGEV_NONE timer must return 0, when
 685		 * it is expired !
 
 
 686		 */
 687		if (!sig_none)
 688			cur_setting->it_value.tv_nsec = 1;
 689	} else {
 690		cur_setting->it_value = ktime_to_timespec64(remaining);
 691	}
 692}
 693
 694/* Get the time remaining on a POSIX.1b interval timer. */
 695static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 696{
 697	struct k_itimer *timr;
 698	const struct k_clock *kc;
 
 699	unsigned long flags;
 700	int ret = 0;
 701
 702	timr = lock_timer(timer_id, &flags);
 703	if (!timr)
 704		return -EINVAL;
 705
 706	memset(setting, 0, sizeof(*setting));
 707	kc = timr->kclock;
 708	if (WARN_ON_ONCE(!kc || !kc->timer_get))
 709		ret = -EINVAL;
 710	else
 711		kc->timer_get(timr, setting);
 712
 713	unlock_timer(timr, flags);
 714	return ret;
 715}
 716
 717/* Get the time remaining on a POSIX.1b interval timer. */
 718SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
 719		struct __kernel_itimerspec __user *, setting)
 720{
 721	struct itimerspec64 cur_setting;
 722
 723	int ret = do_timer_gettime(timer_id, &cur_setting);
 724	if (!ret) {
 725		if (put_itimerspec64(&cur_setting, setting))
 726			ret = -EFAULT;
 727	}
 728	return ret;
 729}
 730
 731#ifdef CONFIG_COMPAT_32BIT_TIME
 732
 733SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
 734		struct old_itimerspec32 __user *, setting)
 735{
 736	struct itimerspec64 cur_setting;
 737
 738	int ret = do_timer_gettime(timer_id, &cur_setting);
 739	if (!ret) {
 740		if (put_old_itimerspec32(&cur_setting, setting))
 741			ret = -EFAULT;
 742	}
 743	return ret;
 744}
 745
 746#endif
 747
 748/*
 749 * Get the number of overruns of a POSIX.1b interval timer.  This is to
 750 * be the overrun of the timer last delivered.  At the same time we are
 751 * accumulating overruns on the next timer.  The overrun is frozen when
 752 * the signal is delivered, either at the notify time (if the info block
 753 * is not queued) or at the actual delivery time (as we are informed by
 754 * the call back to posixtimer_rearm().  So all we need to do is
 755 * to pick up the frozen overrun.
 
 
 
 
 
 
 
 
 
 756 */
 757SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 758{
 759	struct k_itimer *timr;
 760	int overrun;
 761	unsigned long flags;
 
 762
 763	timr = lock_timer(timer_id, &flags);
 764	if (!timr)
 765		return -EINVAL;
 766
 767	overrun = timer_overrun_to_int(timr, 0);
 768	unlock_timer(timr, flags);
 769
 770	return overrun;
 771}
 772
 773static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
 774			       bool absolute, bool sigev_none)
 775{
 776	struct hrtimer *timer = &timr->it.real.timer;
 777	enum hrtimer_mode mode;
 778
 779	mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
 780	/*
 781	 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
 782	 * clock modifications, so they become CLOCK_MONOTONIC based under the
 783	 * hood. See hrtimer_init(). Update timr->kclock, so the generic
 784	 * functions which use timr->kclock->clock_get() work.
 785	 *
 786	 * Note: it_clock stays unmodified, because the next timer_set() might
 787	 * use ABSTIME, so it needs to switch back.
 788	 */
 789	if (timr->it_clock == CLOCK_REALTIME)
 790		timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
 791
 792	hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
 793	timr->it.real.timer.function = posix_timer_fn;
 794
 795	if (!absolute)
 796		expires = ktime_add_safe(expires, timer->base->get_time());
 797	hrtimer_set_expires(timer, expires);
 798
 799	if (!sigev_none)
 800		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 801}
 802
 803static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
 804{
 805	return hrtimer_try_to_cancel(&timr->it.real.timer);
 806}
 807
 808static void common_timer_wait_running(struct k_itimer *timer)
 809{
 810	hrtimer_cancel_wait_running(&timer->it.real.timer);
 811}
 812
 813/*
 814 * On PREEMPT_RT this prevent priority inversion against softirq kthread in
 815 * case it gets preempted while executing a timer callback. See comments in
 816 * hrtimer_cancel_wait_running. For PREEMPT_RT=n this just results in a
 817 * cpu_relax().
 
 
 
 
 
 
 
 
 818 */
 819static struct k_itimer *timer_wait_running(struct k_itimer *timer,
 820					   unsigned long *flags)
 821{
 822	const struct k_clock *kc = READ_ONCE(timer->kclock);
 823	timer_t timer_id = READ_ONCE(timer->it_id);
 824
 825	/* Prevent kfree(timer) after dropping the lock */
 826	rcu_read_lock();
 827	unlock_timer(timer, *flags);
 828
 
 
 
 
 829	if (!WARN_ON_ONCE(!kc->timer_wait_running))
 830		kc->timer_wait_running(timer);
 831
 832	rcu_read_unlock();
 833	/* Relock the timer. It might be not longer hashed. */
 834	return lock_timer(timer_id, flags);
 835}
 836
 837/* Set a POSIX.1b interval timer. */
 838int common_timer_set(struct k_itimer *timr, int flags,
 839		     struct itimerspec64 *new_setting,
 840		     struct itimerspec64 *old_setting)
 841{
 842	const struct k_clock *kc = timr->kclock;
 843	bool sigev_none;
 844	ktime_t expires;
 845
 846	if (old_setting)
 847		common_timer_get(timr, old_setting);
 848
 849	/* Prevent rearming by clearing the interval */
 850	timr->it_interval = 0;
 851	/*
 852	 * Careful here. On SMP systems the timer expiry function could be
 853	 * active and spinning on timr->it_lock.
 854	 */
 855	if (kc->timer_try_to_cancel(timr) < 0)
 856		return TIMER_RETRY;
 857
 858	timr->it_active = 0;
 859	timr->it_requeue_pending = (timr->it_requeue_pending + 2) &
 860		~REQUEUE_PENDING;
 861	timr->it_overrun_last = 0;
 862
 863	/* Switch off the timer when it_value is zero */
 864	if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
 865		return 0;
 866
 867	timr->it_interval = timespec64_to_ktime(new_setting->it_interval);
 868	expires = timespec64_to_ktime(new_setting->it_value);
 
 
 869	sigev_none = timr->it_sigev_notify == SIGEV_NONE;
 870
 871	kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
 872	timr->it_active = !sigev_none;
 873	return 0;
 874}
 875
 876static int do_timer_settime(timer_t timer_id, int tmr_flags,
 877			    struct itimerspec64 *new_spec64,
 878			    struct itimerspec64 *old_spec64)
 879{
 880	const struct k_clock *kc;
 881	struct k_itimer *timr;
 882	unsigned long flags;
 883	int error = 0;
 884
 885	if (!timespec64_valid(&new_spec64->it_interval) ||
 886	    !timespec64_valid(&new_spec64->it_value))
 887		return -EINVAL;
 888
 889	if (old_spec64)
 890		memset(old_spec64, 0, sizeof(*old_spec64));
 891
 892	timr = lock_timer(timer_id, &flags);
 893retry:
 894	if (!timr)
 895		return -EINVAL;
 896
 897	kc = timr->kclock;
 898	if (WARN_ON_ONCE(!kc || !kc->timer_set))
 899		error = -EINVAL;
 900	else
 901		error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
 902
 903	if (error == TIMER_RETRY) {
 904		// We already got the old time...
 905		old_spec64 = NULL;
 906		/* Unlocks and relocks the timer if it still exists */
 907		timr = timer_wait_running(timr, &flags);
 908		goto retry;
 909	}
 910	unlock_timer(timr, flags);
 911
 912	return error;
 913}
 914
 915/* Set a POSIX.1b interval timer */
 916SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
 917		const struct __kernel_itimerspec __user *, new_setting,
 918		struct __kernel_itimerspec __user *, old_setting)
 919{
 920	struct itimerspec64 new_spec, old_spec;
 921	struct itimerspec64 *rtn = old_setting ? &old_spec : NULL;
 922	int error = 0;
 923
 924	if (!new_setting)
 925		return -EINVAL;
 926
 927	if (get_itimerspec64(&new_spec, new_setting))
 928		return -EFAULT;
 929
 
 930	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 931	if (!error && old_setting) {
 932		if (put_itimerspec64(&old_spec, old_setting))
 933			error = -EFAULT;
 934	}
 935	return error;
 936}
 937
 938#ifdef CONFIG_COMPAT_32BIT_TIME
 939SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
 940		struct old_itimerspec32 __user *, new,
 941		struct old_itimerspec32 __user *, old)
 942{
 943	struct itimerspec64 new_spec, old_spec;
 944	struct itimerspec64 *rtn = old ? &old_spec : NULL;
 945	int error = 0;
 946
 947	if (!new)
 948		return -EINVAL;
 949	if (get_old_itimerspec32(&new_spec, new))
 950		return -EFAULT;
 951
 952	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 953	if (!error && old) {
 954		if (put_old_itimerspec32(&old_spec, old))
 955			error = -EFAULT;
 956	}
 957	return error;
 958}
 959#endif
 960
 961int common_timer_del(struct k_itimer *timer)
 962{
 963	const struct k_clock *kc = timer->kclock;
 964
 965	timer->it_interval = 0;
 966	if (kc->timer_try_to_cancel(timer) < 0)
 967		return TIMER_RETRY;
 968	timer->it_active = 0;
 969	return 0;
 970}
 971
 972static inline int timer_delete_hook(struct k_itimer *timer)
 973{
 974	const struct k_clock *kc = timer->kclock;
 975
 976	if (WARN_ON_ONCE(!kc || !kc->timer_del))
 977		return -EINVAL;
 978	return kc->timer_del(timer);
 979}
 980
 981/* Delete a POSIX.1b interval timer. */
 982SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 983{
 984	struct k_itimer *timer;
 985	unsigned long flags;
 986
 987	timer = lock_timer(timer_id, &flags);
 988
 989retry_delete:
 990	if (!timer)
 991		return -EINVAL;
 992
 993	if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
 994		/* Unlocks and relocks the timer if it still exists */
 995		timer = timer_wait_running(timer, &flags);
 996		goto retry_delete;
 997	}
 998
 999	spin_lock(&current->sighand->siglock);
1000	list_del(&timer->list);
1001	spin_unlock(&current->sighand->siglock);
1002	/*
1003	 * This keeps any tasks waiting on the spin lock from thinking
1004	 * they got something (see the lock code above).
1005	 */
1006	timer->it_signal = NULL;
1007
1008	unlock_timer(timer, flags);
1009	release_posix_timer(timer, IT_ID_SET);
1010	return 0;
1011}
1012
1013/*
1014 * return timer owned by the process, used by exit_itimers
 
1015 */
1016static void itimer_delete(struct k_itimer *timer)
1017{
1018retry_delete:
1019	spin_lock_irq(&timer->it_lock);
 
 
 
 
1020
 
 
 
 
 
 
 
1021	if (timer_delete_hook(timer) == TIMER_RETRY) {
1022		spin_unlock_irq(&timer->it_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
1023		goto retry_delete;
1024	}
1025	list_del(&timer->list);
1026
1027	spin_unlock_irq(&timer->it_lock);
1028	release_posix_timer(timer, IT_ID_SET);
 
 
 
 
 
 
 
 
1029}
1030
1031/*
1032 * This is called by do_exit or de_thread, only when there are no more
1033 * references to the shared signal_struct.
 
1034 */
1035void exit_itimers(struct signal_struct *sig)
1036{
 
1037	struct k_itimer *tmr;
1038
1039	while (!list_empty(&sig->posix_timers)) {
1040		tmr = list_entry(sig->posix_timers.next, struct k_itimer, list);
 
 
 
 
 
 
 
 
 
1041		itimer_delete(tmr);
1042	}
1043}
1044
1045SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1046		const struct __kernel_timespec __user *, tp)
1047{
1048	const struct k_clock *kc = clockid_to_kclock(which_clock);
1049	struct timespec64 new_tp;
1050
1051	if (!kc || !kc->clock_set)
1052		return -EINVAL;
1053
1054	if (get_timespec64(&new_tp, tp))
1055		return -EFAULT;
1056
 
 
 
 
1057	return kc->clock_set(which_clock, &new_tp);
1058}
1059
1060SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1061		struct __kernel_timespec __user *, tp)
1062{
1063	const struct k_clock *kc = clockid_to_kclock(which_clock);
1064	struct timespec64 kernel_tp;
1065	int error;
1066
1067	if (!kc)
1068		return -EINVAL;
1069
1070	error = kc->clock_get(which_clock, &kernel_tp);
1071
1072	if (!error && put_timespec64(&kernel_tp, tp))
1073		error = -EFAULT;
1074
1075	return error;
1076}
1077
1078int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1079{
1080	const struct k_clock *kc = clockid_to_kclock(which_clock);
1081
1082	if (!kc)
1083		return -EINVAL;
1084	if (!kc->clock_adj)
1085		return -EOPNOTSUPP;
1086
1087	return kc->clock_adj(which_clock, ktx);
1088}
1089
1090SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1091		struct __kernel_timex __user *, utx)
1092{
1093	struct __kernel_timex ktx;
1094	int err;
1095
1096	if (copy_from_user(&ktx, utx, sizeof(ktx)))
1097		return -EFAULT;
1098
1099	err = do_clock_adjtime(which_clock, &ktx);
1100
1101	if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1102		return -EFAULT;
1103
1104	return err;
1105}
1106
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1107SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1108		struct __kernel_timespec __user *, tp)
1109{
1110	const struct k_clock *kc = clockid_to_kclock(which_clock);
1111	struct timespec64 rtn_tp;
1112	int error;
1113
1114	if (!kc)
1115		return -EINVAL;
1116
1117	error = kc->clock_getres(which_clock, &rtn_tp);
1118
1119	if (!error && tp && put_timespec64(&rtn_tp, tp))
1120		error = -EFAULT;
1121
1122	return error;
1123}
1124
1125#ifdef CONFIG_COMPAT_32BIT_TIME
1126
1127SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1128		struct old_timespec32 __user *, tp)
1129{
1130	const struct k_clock *kc = clockid_to_kclock(which_clock);
1131	struct timespec64 ts;
1132
1133	if (!kc || !kc->clock_set)
1134		return -EINVAL;
1135
1136	if (get_old_timespec32(&ts, tp))
1137		return -EFAULT;
1138
1139	return kc->clock_set(which_clock, &ts);
1140}
1141
1142SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1143		struct old_timespec32 __user *, tp)
1144{
1145	const struct k_clock *kc = clockid_to_kclock(which_clock);
1146	struct timespec64 ts;
1147	int err;
1148
1149	if (!kc)
1150		return -EINVAL;
1151
1152	err = kc->clock_get(which_clock, &ts);
1153
1154	if (!err && put_old_timespec32(&ts, tp))
1155		err = -EFAULT;
1156
1157	return err;
1158}
1159
1160SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1161		struct old_timex32 __user *, utp)
1162{
1163	struct __kernel_timex ktx;
1164	int err;
1165
1166	err = get_old_timex32(&ktx, utp);
1167	if (err)
1168		return err;
1169
1170	err = do_clock_adjtime(which_clock, &ktx);
1171
1172	if (err >= 0)
1173		err = put_old_timex32(utp, &ktx);
1174
1175	return err;
1176}
1177
1178SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1179		struct old_timespec32 __user *, tp)
1180{
1181	const struct k_clock *kc = clockid_to_kclock(which_clock);
1182	struct timespec64 ts;
1183	int err;
1184
1185	if (!kc)
1186		return -EINVAL;
1187
1188	err = kc->clock_getres(which_clock, &ts);
1189	if (!err && tp && put_old_timespec32(&ts, tp))
1190		return -EFAULT;
1191
1192	return err;
1193}
1194
1195#endif
1196
1197/*
1198 * nanosleep for monotonic and realtime clocks
1199 */
1200static int common_nsleep(const clockid_t which_clock, int flags,
1201			 const struct timespec64 *rqtp)
1202{
1203	return hrtimer_nanosleep(rqtp, flags & TIMER_ABSTIME ?
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1204				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1205				 which_clock);
1206}
1207
1208SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1209		const struct __kernel_timespec __user *, rqtp,
1210		struct __kernel_timespec __user *, rmtp)
1211{
1212	const struct k_clock *kc = clockid_to_kclock(which_clock);
1213	struct timespec64 t;
1214
1215	if (!kc)
1216		return -EINVAL;
1217	if (!kc->nsleep)
1218		return -EOPNOTSUPP;
1219
1220	if (get_timespec64(&t, rqtp))
1221		return -EFAULT;
1222
1223	if (!timespec64_valid(&t))
1224		return -EINVAL;
1225	if (flags & TIMER_ABSTIME)
1226		rmtp = NULL;
 
1227	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1228	current->restart_block.nanosleep.rmtp = rmtp;
1229
1230	return kc->nsleep(which_clock, flags, &t);
1231}
1232
1233#ifdef CONFIG_COMPAT_32BIT_TIME
1234
1235SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1236		struct old_timespec32 __user *, rqtp,
1237		struct old_timespec32 __user *, rmtp)
1238{
1239	const struct k_clock *kc = clockid_to_kclock(which_clock);
1240	struct timespec64 t;
1241
1242	if (!kc)
1243		return -EINVAL;
1244	if (!kc->nsleep)
1245		return -EOPNOTSUPP;
1246
1247	if (get_old_timespec32(&t, rqtp))
1248		return -EFAULT;
1249
1250	if (!timespec64_valid(&t))
1251		return -EINVAL;
1252	if (flags & TIMER_ABSTIME)
1253		rmtp = NULL;
 
1254	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1255	current->restart_block.nanosleep.compat_rmtp = rmtp;
1256
1257	return kc->nsleep(which_clock, flags, &t);
1258}
1259
1260#endif
1261
1262static const struct k_clock clock_realtime = {
1263	.clock_getres		= posix_get_hrtimer_res,
1264	.clock_get		= posix_clock_realtime_get,
 
1265	.clock_set		= posix_clock_realtime_set,
1266	.clock_adj		= posix_clock_realtime_adj,
1267	.nsleep			= common_nsleep,
1268	.timer_create		= common_timer_create,
1269	.timer_set		= common_timer_set,
1270	.timer_get		= common_timer_get,
1271	.timer_del		= common_timer_del,
1272	.timer_rearm		= common_hrtimer_rearm,
1273	.timer_forward		= common_hrtimer_forward,
1274	.timer_remaining	= common_hrtimer_remaining,
1275	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1276	.timer_wait_running	= common_timer_wait_running,
1277	.timer_arm		= common_hrtimer_arm,
1278};
1279
1280static const struct k_clock clock_monotonic = {
1281	.clock_getres		= posix_get_hrtimer_res,
1282	.clock_get		= posix_ktime_get_ts,
1283	.nsleep			= common_nsleep,
 
1284	.timer_create		= common_timer_create,
1285	.timer_set		= common_timer_set,
1286	.timer_get		= common_timer_get,
1287	.timer_del		= common_timer_del,
1288	.timer_rearm		= common_hrtimer_rearm,
1289	.timer_forward		= common_hrtimer_forward,
1290	.timer_remaining	= common_hrtimer_remaining,
1291	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1292	.timer_wait_running	= common_timer_wait_running,
1293	.timer_arm		= common_hrtimer_arm,
1294};
1295
1296static const struct k_clock clock_monotonic_raw = {
1297	.clock_getres		= posix_get_hrtimer_res,
1298	.clock_get		= posix_get_monotonic_raw,
1299};
1300
1301static const struct k_clock clock_realtime_coarse = {
1302	.clock_getres		= posix_get_coarse_res,
1303	.clock_get		= posix_get_realtime_coarse,
1304};
1305
1306static const struct k_clock clock_monotonic_coarse = {
1307	.clock_getres		= posix_get_coarse_res,
1308	.clock_get		= posix_get_monotonic_coarse,
1309};
1310
1311static const struct k_clock clock_tai = {
1312	.clock_getres		= posix_get_hrtimer_res,
1313	.clock_get		= posix_get_tai,
 
1314	.nsleep			= common_nsleep,
1315	.timer_create		= common_timer_create,
1316	.timer_set		= common_timer_set,
1317	.timer_get		= common_timer_get,
1318	.timer_del		= common_timer_del,
1319	.timer_rearm		= common_hrtimer_rearm,
1320	.timer_forward		= common_hrtimer_forward,
1321	.timer_remaining	= common_hrtimer_remaining,
1322	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1323	.timer_wait_running	= common_timer_wait_running,
1324	.timer_arm		= common_hrtimer_arm,
1325};
1326
1327static const struct k_clock clock_boottime = {
1328	.clock_getres		= posix_get_hrtimer_res,
1329	.clock_get		= posix_get_boottime,
1330	.nsleep			= common_nsleep,
 
1331	.timer_create		= common_timer_create,
1332	.timer_set		= common_timer_set,
1333	.timer_get		= common_timer_get,
1334	.timer_del		= common_timer_del,
1335	.timer_rearm		= common_hrtimer_rearm,
1336	.timer_forward		= common_hrtimer_forward,
1337	.timer_remaining	= common_hrtimer_remaining,
1338	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1339	.timer_wait_running	= common_timer_wait_running,
1340	.timer_arm		= common_hrtimer_arm,
1341};
1342
1343static const struct k_clock * const posix_clocks[] = {
1344	[CLOCK_REALTIME]		= &clock_realtime,
1345	[CLOCK_MONOTONIC]		= &clock_monotonic,
1346	[CLOCK_PROCESS_CPUTIME_ID]	= &clock_process,
1347	[CLOCK_THREAD_CPUTIME_ID]	= &clock_thread,
1348	[CLOCK_MONOTONIC_RAW]		= &clock_monotonic_raw,
1349	[CLOCK_REALTIME_COARSE]		= &clock_realtime_coarse,
1350	[CLOCK_MONOTONIC_COARSE]	= &clock_monotonic_coarse,
1351	[CLOCK_BOOTTIME]		= &clock_boottime,
1352	[CLOCK_REALTIME_ALARM]		= &alarm_clock,
1353	[CLOCK_BOOTTIME_ALARM]		= &alarm_clock,
1354	[CLOCK_TAI]			= &clock_tai,
1355};
1356
1357static const struct k_clock *clockid_to_kclock(const clockid_t id)
1358{
1359	clockid_t idx = id;
1360
1361	if (id < 0) {
1362		return (id & CLOCKFD_MASK) == CLOCKFD ?
1363			&clock_posix_dynamic : &clock_posix_cpu;
1364	}
1365
1366	if (id >= ARRAY_SIZE(posix_clocks))
1367		return NULL;
1368
1369	return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1370}
v6.8
   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, int baseval)
 237{
 238	s64 sum = timr->it_overrun_last + (s64)baseval;
 239
 240	return sum > (s64)INT_MAX ? INT_MAX : (int)sum;
 241}
 242
 243static void common_hrtimer_rearm(struct k_itimer *timr)
 244{
 245	struct hrtimer *timer = &timr->it.real.timer;
 246
 247	timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
 248					    timr->it_interval);
 249	hrtimer_restart(timer);
 250}
 251
 252/*
 253 * This function is called from the signal delivery code if
 254 * info->si_sys_private is not zero, which indicates that the timer has to
 255 * be rearmed. Restart the timer and update info::si_overrun.
 
 
 
 
 
 
 256 */
 257void posixtimer_rearm(struct kernel_siginfo *info)
 258{
 259	struct k_itimer *timr;
 260	unsigned long flags;
 261
 262	timr = lock_timer(info->si_tid, &flags);
 263	if (!timr)
 264		return;
 265
 266	if (timr->it_interval && timr->it_requeue_pending == info->si_sys_private) {
 267		timr->kclock->timer_rearm(timr);
 268
 269		timr->it_active = 1;
 270		timr->it_overrun_last = timr->it_overrun;
 271		timr->it_overrun = -1LL;
 272		++timr->it_requeue_pending;
 273
 274		info->si_overrun = timer_overrun_to_int(timr, info->si_overrun);
 275	}
 276
 277	unlock_timer(timr, flags);
 278}
 279
 280int posix_timer_event(struct k_itimer *timr, int si_private)
 281{
 282	enum pid_type type;
 283	int ret;
 284	/*
 285	 * FIXME: if ->sigq is queued we can race with
 286	 * dequeue_signal()->posixtimer_rearm().
 287	 *
 288	 * If dequeue_signal() sees the "right" value of
 289	 * si_sys_private it calls posixtimer_rearm().
 290	 * We re-queue ->sigq and drop ->it_lock().
 291	 * posixtimer_rearm() locks the timer
 292	 * and re-schedules it while ->sigq is pending.
 293	 * Not really bad, but not that we want.
 294	 */
 295	timr->sigq->info.si_sys_private = si_private;
 296
 297	type = !(timr->it_sigev_notify & SIGEV_THREAD_ID) ? PIDTYPE_TGID : PIDTYPE_PID;
 298	ret = send_sigqueue(timr->sigq, timr->it_pid, type);
 299	/* If we failed to send the signal the timer stops. */
 300	return ret > 0;
 301}
 302
 303/*
 304 * This function gets called when a POSIX.1b interval timer expires from
 305 * the HRTIMER interrupt (soft interrupt on RT kernels).
 306 *
 307 * Handles CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME and CLOCK_TAI
 308 * based timers.
 309 */
 310static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
 311{
 312	enum hrtimer_restart ret = HRTIMER_NORESTART;
 313	struct k_itimer *timr;
 314	unsigned long flags;
 315	int si_private = 0;
 
 316
 317	timr = container_of(timer, struct k_itimer, it.real.timer);
 318	spin_lock_irqsave(&timr->it_lock, flags);
 319
 320	timr->it_active = 0;
 321	if (timr->it_interval != 0)
 322		si_private = ++timr->it_requeue_pending;
 323
 324	if (posix_timer_event(timr, si_private)) {
 325		/*
 326		 * The signal was not queued due to SIG_IGN. As a
 327		 * consequence the timer is not going to be rearmed from
 328		 * the signal delivery path. But as a real signal handler
 329		 * can be installed later the timer must be rearmed here.
 330		 */
 331		if (timr->it_interval != 0) {
 332			ktime_t now = hrtimer_cb_get_time(timer);
 333
 334			/*
 335			 * FIXME: What we really want, is to stop this
 336			 * timer completely and restart it in case the
 337			 * SIG_IGN is removed. This is a non trivial
 338			 * change to the signal handling code.
 339			 *
 340			 * For now let timers with an interval less than a
 341			 * jiffie expire every jiffie and recheck for a
 342			 * valid signal handler.
 343			 *
 344			 * This avoids interrupt starvation in case of a
 345			 * very small interval, which would expire the
 346			 * timer immediately again.
 347			 *
 348			 * Moving now ahead of time by one jiffie tricks
 349			 * hrtimer_forward() to expire the timer later,
 350			 * while it still maintains the overrun accuracy
 351			 * for the price of a slight inconsistency in the
 352			 * timer_gettime() case. This is at least better
 353			 * than a timer storm.
 354			 *
 355			 * Only required when high resolution timers are
 356			 * enabled as the periodic tick based timers are
 357			 * automatically aligned to the next tick.
 
 
 
 358			 */
 359			if (IS_ENABLED(CONFIG_HIGH_RES_TIMERS)) {
 360				ktime_t kj = TICK_NSEC;
 
 361
 362				if (timr->it_interval < kj)
 363					now = ktime_add(now, kj);
 364			}
 365
 366			timr->it_overrun += hrtimer_forward(timer, now, timr->it_interval);
 
 367			ret = HRTIMER_RESTART;
 368			++timr->it_requeue_pending;
 369			timr->it_active = 1;
 370		}
 371	}
 372
 373	unlock_timer(timr, flags);
 374	return ret;
 375}
 376
 377static struct pid *good_sigevent(sigevent_t * event)
 378{
 379	struct pid *pid = task_tgid(current);
 380	struct task_struct *rtn;
 381
 382	switch (event->sigev_notify) {
 383	case SIGEV_SIGNAL | SIGEV_THREAD_ID:
 384		pid = find_vpid(event->sigev_notify_thread_id);
 385		rtn = pid_task(pid, PIDTYPE_PID);
 386		if (!rtn || !same_thread_group(rtn, current))
 387			return NULL;
 388		fallthrough;
 389	case SIGEV_SIGNAL:
 390	case SIGEV_THREAD:
 391		if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
 392			return NULL;
 393		fallthrough;
 394	case SIGEV_NONE:
 395		return pid;
 396	default:
 397		return NULL;
 398	}
 399}
 400
 401static struct k_itimer * alloc_posix_timer(void)
 402{
 403	struct k_itimer *tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
 404
 405	if (!tmr)
 406		return tmr;
 407	if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
 408		kmem_cache_free(posix_timers_cache, tmr);
 409		return NULL;
 410	}
 411	clear_siginfo(&tmr->sigq->info);
 412	return tmr;
 413}
 414
 415static void k_itimer_rcu_free(struct rcu_head *head)
 416{
 417	struct k_itimer *tmr = container_of(head, struct k_itimer, rcu);
 418
 419	kmem_cache_free(posix_timers_cache, tmr);
 420}
 421
 422static void posix_timer_free(struct k_itimer *tmr)
 423{
 
 
 
 
 
 
 
 
 424	put_pid(tmr->it_pid);
 425	sigqueue_free(tmr->sigq);
 426	call_rcu(&tmr->rcu, k_itimer_rcu_free);
 427}
 428
 429static void posix_timer_unhash_and_free(struct k_itimer *tmr)
 430{
 431	spin_lock(&hash_lock);
 432	hlist_del_rcu(&tmr->t_hash);
 433	spin_unlock(&hash_lock);
 434	posix_timer_free(tmr);
 435}
 436
 437static int common_timer_create(struct k_itimer *new_timer)
 438{
 439	hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
 440	return 0;
 441}
 442
 443/* Create a POSIX.1b interval timer. */
 444static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 445			   timer_t __user *created_timer_id)
 446{
 447	const struct k_clock *kc = clockid_to_kclock(which_clock);
 448	struct k_itimer *new_timer;
 449	int error, new_timer_id;
 
 450
 451	if (!kc)
 452		return -EINVAL;
 453	if (!kc->timer_create)
 454		return -EOPNOTSUPP;
 455
 456	new_timer = alloc_posix_timer();
 457	if (unlikely(!new_timer))
 458		return -EAGAIN;
 459
 460	spin_lock_init(&new_timer->it_lock);
 461
 462	/*
 463	 * Add the timer to the hash table. The timer is not yet valid
 464	 * because new_timer::it_signal is still NULL. The timer id is also
 465	 * not yet visible to user space.
 466	 */
 467	new_timer_id = posix_timer_add(new_timer);
 468	if (new_timer_id < 0) {
 469		posix_timer_free(new_timer);
 470		return new_timer_id;
 471	}
 472
 
 473	new_timer->it_id = (timer_t) new_timer_id;
 474	new_timer->it_clock = which_clock;
 475	new_timer->kclock = kc;
 476	new_timer->it_overrun = -1LL;
 477
 478	if (event) {
 479		rcu_read_lock();
 480		new_timer->it_pid = get_pid(good_sigevent(event));
 481		rcu_read_unlock();
 482		if (!new_timer->it_pid) {
 483			error = -EINVAL;
 484			goto out;
 485		}
 486		new_timer->it_sigev_notify     = event->sigev_notify;
 487		new_timer->sigq->info.si_signo = event->sigev_signo;
 488		new_timer->sigq->info.si_value = event->sigev_value;
 489	} else {
 490		new_timer->it_sigev_notify     = SIGEV_SIGNAL;
 491		new_timer->sigq->info.si_signo = SIGALRM;
 492		memset(&new_timer->sigq->info.si_value, 0, sizeof(sigval_t));
 493		new_timer->sigq->info.si_value.sival_int = new_timer->it_id;
 494		new_timer->it_pid = get_pid(task_tgid(current));
 495	}
 496
 497	new_timer->sigq->info.si_tid   = new_timer->it_id;
 498	new_timer->sigq->info.si_code  = SI_TIMER;
 499
 500	if (copy_to_user(created_timer_id, &new_timer_id, sizeof (new_timer_id))) {
 
 501		error = -EFAULT;
 502		goto out;
 503	}
 504	/*
 505	 * After succesful copy out, the timer ID is visible to user space
 506	 * now but not yet valid because new_timer::signal is still NULL.
 507	 *
 508	 * Complete the initialization with the clock specific create
 509	 * callback.
 510	 */
 511	error = kc->timer_create(new_timer);
 512	if (error)
 513		goto out;
 514
 515	spin_lock_irq(&current->sighand->siglock);
 516	/* This makes the timer valid in the hash table */
 517	WRITE_ONCE(new_timer->it_signal, current->signal);
 518	list_add(&new_timer->list, &current->signal->posix_timers);
 519	spin_unlock_irq(&current->sighand->siglock);
 
 
 520	/*
 521	 * After unlocking sighand::siglock @new_timer is subject to
 522	 * concurrent removal and cannot be touched anymore
 
 
 523	 */
 524	return 0;
 525out:
 526	posix_timer_unhash_and_free(new_timer);
 527	return error;
 528}
 529
 530SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
 531		struct sigevent __user *, timer_event_spec,
 532		timer_t __user *, created_timer_id)
 533{
 534	if (timer_event_spec) {
 535		sigevent_t event;
 536
 537		if (copy_from_user(&event, timer_event_spec, sizeof (event)))
 538			return -EFAULT;
 539		return do_timer_create(which_clock, &event, created_timer_id);
 540	}
 541	return do_timer_create(which_clock, NULL, created_timer_id);
 542}
 543
 544#ifdef CONFIG_COMPAT
 545COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
 546		       struct compat_sigevent __user *, timer_event_spec,
 547		       timer_t __user *, created_timer_id)
 548{
 549	if (timer_event_spec) {
 550		sigevent_t event;
 551
 552		if (get_compat_sigevent(&event, timer_event_spec))
 553			return -EFAULT;
 554		return do_timer_create(which_clock, &event, created_timer_id);
 555	}
 556	return do_timer_create(which_clock, NULL, created_timer_id);
 557}
 558#endif
 559
 
 
 
 
 
 
 
 560static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 561{
 562	struct k_itimer *timr;
 563
 564	/*
 565	 * timer_t could be any type >= int and we want to make sure any
 566	 * @timer_id outside positive int range fails lookup.
 567	 */
 568	if ((unsigned long long)timer_id > INT_MAX)
 569		return NULL;
 570
 571	/*
 572	 * The hash lookup and the timers are RCU protected.
 573	 *
 574	 * Timers are added to the hash in invalid state where
 575	 * timr::it_signal == NULL. timer::it_signal is only set after the
 576	 * rest of the initialization succeeded.
 577	 *
 578	 * Timer destruction happens in steps:
 579	 *  1) Set timr::it_signal to NULL with timr::it_lock held
 580	 *  2) Release timr::it_lock
 581	 *  3) Remove from the hash under hash_lock
 582	 *  4) Call RCU for removal after the grace period
 583	 *
 584	 * Holding rcu_read_lock() accross the lookup ensures that
 585	 * the timer cannot be freed.
 586	 *
 587	 * The lookup validates locklessly that timr::it_signal ==
 588	 * current::it_signal and timr::it_id == @timer_id. timr::it_id
 589	 * can't change, but timr::it_signal becomes NULL during
 590	 * destruction.
 591	 */
 592	rcu_read_lock();
 593	timr = posix_timer_by_id(timer_id);
 594	if (timr) {
 595		spin_lock_irqsave(&timr->it_lock, *flags);
 596		/*
 597		 * Validate under timr::it_lock that timr::it_signal is
 598		 * still valid. Pairs with #1 above.
 599		 */
 600		if (timr->it_signal == current->signal) {
 601			rcu_read_unlock();
 602			return timr;
 603		}
 604		spin_unlock_irqrestore(&timr->it_lock, *flags);
 605	}
 606	rcu_read_unlock();
 607
 608	return NULL;
 609}
 610
 611static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
 612{
 613	struct hrtimer *timer = &timr->it.real.timer;
 614
 615	return __hrtimer_expires_remaining_adjusted(timer, now);
 616}
 617
 618static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
 619{
 620	struct hrtimer *timer = &timr->it.real.timer;
 621
 622	return hrtimer_forward(timer, now, timr->it_interval);
 623}
 624
 625/*
 626 * Get the time remaining on a POSIX.1b interval timer.
 627 *
 628 * Two issues to handle here:
 629 *
 630 *  1) The timer has a requeue pending. The return value must appear as
 631 *     if the timer has been requeued right now.
 632 *
 633 *  2) The timer is a SIGEV_NONE timer. These timers are never enqueued
 634 *     into the hrtimer queue and therefore never expired. Emulate expiry
 635 *     here taking #1 into account.
 
 
 
 
 636 */
 637void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
 638{
 639	const struct k_clock *kc = timr->kclock;
 640	ktime_t now, remaining, iv;
 
 641	bool sig_none;
 642
 643	sig_none = timr->it_sigev_notify == SIGEV_NONE;
 644	iv = timr->it_interval;
 645
 646	/* interval timer ? */
 647	if (iv) {
 648		cur_setting->it_interval = ktime_to_timespec64(iv);
 649	} else if (!timr->it_active) {
 650		/*
 651		 * SIGEV_NONE oneshot timers are never queued and therefore
 652		 * timr->it_active is always false. The check below
 653		 * vs. remaining time will handle this case.
 654		 *
 655		 * For all other timers there is nothing to update here, so
 656		 * return.
 657		 */
 658		if (!sig_none)
 659			return;
 660	}
 661
 662	now = kc->clock_get_ktime(timr->it_clock);
 
 
 
 
 
 663
 664	/*
 665	 * If this is an interval timer and either has requeue pending or
 666	 * is a SIGEV_NONE timer move the expiry time forward by intervals,
 667	 * so expiry is > now.
 668	 */
 669	if (iv && (timr->it_requeue_pending & REQUEUE_PENDING || sig_none))
 670		timr->it_overrun += kc->timer_forward(timr, now);
 671
 672	remaining = kc->timer_remaining(timr, now);
 673	/*
 674	 * As @now is retrieved before a possible timer_forward() and
 675	 * cannot be reevaluated by the compiler @remaining is based on the
 676	 * same @now value. Therefore @remaining is consistent vs. @now.
 677	 *
 678	 * Consequently all interval timers, i.e. @iv > 0, cannot have a
 679	 * remaining time <= 0 because timer_forward() guarantees to move
 680	 * them forward so that the next timer expiry is > @now.
 681	 */
 682	if (remaining <= 0) {
 683		/*
 684		 * A single shot SIGEV_NONE timer must return 0, when it is
 685		 * expired! Timers which have a real signal delivery mode
 686		 * must return a remaining time greater than 0 because the
 687		 * signal has not yet been delivered.
 688		 */
 689		if (!sig_none)
 690			cur_setting->it_value.tv_nsec = 1;
 691	} else {
 692		cur_setting->it_value = ktime_to_timespec64(remaining);
 693	}
 694}
 695
 
 696static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 697{
 
 698	const struct k_clock *kc;
 699	struct k_itimer *timr;
 700	unsigned long flags;
 701	int ret = 0;
 702
 703	timr = lock_timer(timer_id, &flags);
 704	if (!timr)
 705		return -EINVAL;
 706
 707	memset(setting, 0, sizeof(*setting));
 708	kc = timr->kclock;
 709	if (WARN_ON_ONCE(!kc || !kc->timer_get))
 710		ret = -EINVAL;
 711	else
 712		kc->timer_get(timr, setting);
 713
 714	unlock_timer(timr, flags);
 715	return ret;
 716}
 717
 718/* Get the time remaining on a POSIX.1b interval timer. */
 719SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
 720		struct __kernel_itimerspec __user *, setting)
 721{
 722	struct itimerspec64 cur_setting;
 723
 724	int ret = do_timer_gettime(timer_id, &cur_setting);
 725	if (!ret) {
 726		if (put_itimerspec64(&cur_setting, setting))
 727			ret = -EFAULT;
 728	}
 729	return ret;
 730}
 731
 732#ifdef CONFIG_COMPAT_32BIT_TIME
 733
 734SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
 735		struct old_itimerspec32 __user *, setting)
 736{
 737	struct itimerspec64 cur_setting;
 738
 739	int ret = do_timer_gettime(timer_id, &cur_setting);
 740	if (!ret) {
 741		if (put_old_itimerspec32(&cur_setting, setting))
 742			ret = -EFAULT;
 743	}
 744	return ret;
 745}
 746
 747#endif
 748
 749/**
 750 * sys_timer_getoverrun - Get the number of overruns of a POSIX.1b interval timer
 751 * @timer_id:	The timer ID which identifies the timer
 752 *
 753 * The "overrun count" of a timer is one plus the number of expiration
 754 * intervals which have elapsed between the first expiry, which queues the
 755 * signal and the actual signal delivery. On signal delivery the "overrun
 756 * count" is calculated and cached, so it can be returned directly here.
 757 *
 758 * As this is relative to the last queued signal the returned overrun count
 759 * is meaningless outside of the signal delivery path and even there it
 760 * does not accurately reflect the current state when user space evaluates
 761 * it.
 762 *
 763 * Returns:
 764 *	-EINVAL		@timer_id is invalid
 765 *	1..INT_MAX	The number of overruns related to the last delivered signal
 766 */
 767SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 768{
 769	struct k_itimer *timr;
 
 770	unsigned long flags;
 771	int overrun;
 772
 773	timr = lock_timer(timer_id, &flags);
 774	if (!timr)
 775		return -EINVAL;
 776
 777	overrun = timer_overrun_to_int(timr, 0);
 778	unlock_timer(timr, flags);
 779
 780	return overrun;
 781}
 782
 783static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
 784			       bool absolute, bool sigev_none)
 785{
 786	struct hrtimer *timer = &timr->it.real.timer;
 787	enum hrtimer_mode mode;
 788
 789	mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
 790	/*
 791	 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
 792	 * clock modifications, so they become CLOCK_MONOTONIC based under the
 793	 * hood. See hrtimer_init(). Update timr->kclock, so the generic
 794	 * functions which use timr->kclock->clock_get_*() work.
 795	 *
 796	 * Note: it_clock stays unmodified, because the next timer_set() might
 797	 * use ABSTIME, so it needs to switch back.
 798	 */
 799	if (timr->it_clock == CLOCK_REALTIME)
 800		timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
 801
 802	hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
 803	timr->it.real.timer.function = posix_timer_fn;
 804
 805	if (!absolute)
 806		expires = ktime_add_safe(expires, timer->base->get_time());
 807	hrtimer_set_expires(timer, expires);
 808
 809	if (!sigev_none)
 810		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 811}
 812
 813static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
 814{
 815	return hrtimer_try_to_cancel(&timr->it.real.timer);
 816}
 817
 818static void common_timer_wait_running(struct k_itimer *timer)
 819{
 820	hrtimer_cancel_wait_running(&timer->it.real.timer);
 821}
 822
 823/*
 824 * On PREEMPT_RT this prevents priority inversion and a potential livelock
 825 * against the ksoftirqd thread in case that ksoftirqd gets preempted while
 826 * executing a hrtimer callback.
 827 *
 828 * See the comments in hrtimer_cancel_wait_running(). For PREEMPT_RT=n this
 829 * just results in a cpu_relax().
 830 *
 831 * For POSIX CPU timers with CONFIG_POSIX_CPU_TIMERS_TASK_WORK=n this is
 832 * just a cpu_relax(). With CONFIG_POSIX_CPU_TIMERS_TASK_WORK=y this
 833 * prevents spinning on an eventually scheduled out task and a livelock
 834 * when the task which tries to delete or disarm the timer has preempted
 835 * the task which runs the expiry in task work context.
 836 */
 837static struct k_itimer *timer_wait_running(struct k_itimer *timer,
 838					   unsigned long *flags)
 839{
 840	const struct k_clock *kc = READ_ONCE(timer->kclock);
 841	timer_t timer_id = READ_ONCE(timer->it_id);
 842
 843	/* Prevent kfree(timer) after dropping the lock */
 844	rcu_read_lock();
 845	unlock_timer(timer, *flags);
 846
 847	/*
 848	 * kc->timer_wait_running() might drop RCU lock. So @timer
 849	 * cannot be touched anymore after the function returns!
 850	 */
 851	if (!WARN_ON_ONCE(!kc->timer_wait_running))
 852		kc->timer_wait_running(timer);
 853
 854	rcu_read_unlock();
 855	/* Relock the timer. It might be not longer hashed. */
 856	return lock_timer(timer_id, flags);
 857}
 858
 859/* Set a POSIX.1b interval timer. */
 860int common_timer_set(struct k_itimer *timr, int flags,
 861		     struct itimerspec64 *new_setting,
 862		     struct itimerspec64 *old_setting)
 863{
 864	const struct k_clock *kc = timr->kclock;
 865	bool sigev_none;
 866	ktime_t expires;
 867
 868	if (old_setting)
 869		common_timer_get(timr, old_setting);
 870
 871	/* Prevent rearming by clearing the interval */
 872	timr->it_interval = 0;
 873	/*
 874	 * Careful here. On SMP systems the timer expiry function could be
 875	 * active and spinning on timr->it_lock.
 876	 */
 877	if (kc->timer_try_to_cancel(timr) < 0)
 878		return TIMER_RETRY;
 879
 880	timr->it_active = 0;
 881	timr->it_requeue_pending = (timr->it_requeue_pending + 2) &
 882		~REQUEUE_PENDING;
 883	timr->it_overrun_last = 0;
 884
 885	/* Switch off the timer when it_value is zero */
 886	if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
 887		return 0;
 888
 889	timr->it_interval = timespec64_to_ktime(new_setting->it_interval);
 890	expires = timespec64_to_ktime(new_setting->it_value);
 891	if (flags & TIMER_ABSTIME)
 892		expires = timens_ktime_to_host(timr->it_clock, expires);
 893	sigev_none = timr->it_sigev_notify == SIGEV_NONE;
 894
 895	kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
 896	timr->it_active = !sigev_none;
 897	return 0;
 898}
 899
 900static int do_timer_settime(timer_t timer_id, int tmr_flags,
 901			    struct itimerspec64 *new_spec64,
 902			    struct itimerspec64 *old_spec64)
 903{
 904	const struct k_clock *kc;
 905	struct k_itimer *timr;
 906	unsigned long flags;
 907	int error = 0;
 908
 909	if (!timespec64_valid(&new_spec64->it_interval) ||
 910	    !timespec64_valid(&new_spec64->it_value))
 911		return -EINVAL;
 912
 913	if (old_spec64)
 914		memset(old_spec64, 0, sizeof(*old_spec64));
 915
 916	timr = lock_timer(timer_id, &flags);
 917retry:
 918	if (!timr)
 919		return -EINVAL;
 920
 921	kc = timr->kclock;
 922	if (WARN_ON_ONCE(!kc || !kc->timer_set))
 923		error = -EINVAL;
 924	else
 925		error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
 926
 927	if (error == TIMER_RETRY) {
 928		// We already got the old time...
 929		old_spec64 = NULL;
 930		/* Unlocks and relocks the timer if it still exists */
 931		timr = timer_wait_running(timr, &flags);
 932		goto retry;
 933	}
 934	unlock_timer(timr, flags);
 935
 936	return error;
 937}
 938
 939/* Set a POSIX.1b interval timer */
 940SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
 941		const struct __kernel_itimerspec __user *, new_setting,
 942		struct __kernel_itimerspec __user *, old_setting)
 943{
 944	struct itimerspec64 new_spec, old_spec, *rtn;
 
 945	int error = 0;
 946
 947	if (!new_setting)
 948		return -EINVAL;
 949
 950	if (get_itimerspec64(&new_spec, new_setting))
 951		return -EFAULT;
 952
 953	rtn = old_setting ? &old_spec : NULL;
 954	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 955	if (!error && old_setting) {
 956		if (put_itimerspec64(&old_spec, old_setting))
 957			error = -EFAULT;
 958	}
 959	return error;
 960}
 961
 962#ifdef CONFIG_COMPAT_32BIT_TIME
 963SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
 964		struct old_itimerspec32 __user *, new,
 965		struct old_itimerspec32 __user *, old)
 966{
 967	struct itimerspec64 new_spec, old_spec;
 968	struct itimerspec64 *rtn = old ? &old_spec : NULL;
 969	int error = 0;
 970
 971	if (!new)
 972		return -EINVAL;
 973	if (get_old_itimerspec32(&new_spec, new))
 974		return -EFAULT;
 975
 976	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 977	if (!error && old) {
 978		if (put_old_itimerspec32(&old_spec, old))
 979			error = -EFAULT;
 980	}
 981	return error;
 982}
 983#endif
 984
 985int common_timer_del(struct k_itimer *timer)
 986{
 987	const struct k_clock *kc = timer->kclock;
 988
 989	timer->it_interval = 0;
 990	if (kc->timer_try_to_cancel(timer) < 0)
 991		return TIMER_RETRY;
 992	timer->it_active = 0;
 993	return 0;
 994}
 995
 996static inline int timer_delete_hook(struct k_itimer *timer)
 997{
 998	const struct k_clock *kc = timer->kclock;
 999
1000	if (WARN_ON_ONCE(!kc || !kc->timer_del))
1001		return -EINVAL;
1002	return kc->timer_del(timer);
1003}
1004
1005/* Delete a POSIX.1b interval timer. */
1006SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
1007{
1008	struct k_itimer *timer;
1009	unsigned long flags;
1010
1011	timer = lock_timer(timer_id, &flags);
1012
1013retry_delete:
1014	if (!timer)
1015		return -EINVAL;
1016
1017	if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
1018		/* Unlocks and relocks the timer if it still exists */
1019		timer = timer_wait_running(timer, &flags);
1020		goto retry_delete;
1021	}
1022
1023	spin_lock(&current->sighand->siglock);
1024	list_del(&timer->list);
1025	spin_unlock(&current->sighand->siglock);
1026	/*
1027	 * A concurrent lookup could check timer::it_signal lockless. It
1028	 * will reevaluate with timer::it_lock held and observe the NULL.
1029	 */
1030	WRITE_ONCE(timer->it_signal, NULL);
1031
1032	unlock_timer(timer, flags);
1033	posix_timer_unhash_and_free(timer);
1034	return 0;
1035}
1036
1037/*
1038 * Delete a timer if it is armed, remove it from the hash and schedule it
1039 * for RCU freeing.
1040 */
1041static void itimer_delete(struct k_itimer *timer)
1042{
1043	unsigned long flags;
1044
1045	/*
1046	 * irqsave is required to make timer_wait_running() work.
1047	 */
1048	spin_lock_irqsave(&timer->it_lock, flags);
1049
1050retry_delete:
1051	/*
1052	 * Even if the timer is not longer accessible from other tasks
1053	 * it still might be armed and queued in the underlying timer
1054	 * mechanism. Worse, that timer mechanism might run the expiry
1055	 * function concurrently.
1056	 */
1057	if (timer_delete_hook(timer) == TIMER_RETRY) {
1058		/*
1059		 * Timer is expired concurrently, prevent livelocks
1060		 * and pointless spinning on RT.
1061		 *
1062		 * timer_wait_running() drops timer::it_lock, which opens
1063		 * the possibility for another task to delete the timer.
1064		 *
1065		 * That's not possible here because this is invoked from
1066		 * do_exit() only for the last thread of the thread group.
1067		 * So no other task can access and delete that timer.
1068		 */
1069		if (WARN_ON_ONCE(timer_wait_running(timer, &flags) != timer))
1070			return;
1071
1072		goto retry_delete;
1073	}
1074	list_del(&timer->list);
1075
1076	/*
1077	 * Setting timer::it_signal to NULL is technically not required
1078	 * here as nothing can access the timer anymore legitimately via
1079	 * the hash table. Set it to NULL nevertheless so that all deletion
1080	 * paths are consistent.
1081	 */
1082	WRITE_ONCE(timer->it_signal, NULL);
1083
1084	spin_unlock_irqrestore(&timer->it_lock, flags);
1085	posix_timer_unhash_and_free(timer);
1086}
1087
1088/*
1089 * Invoked from do_exit() when the last thread of a thread group exits.
1090 * At that point no other task can access the timers of the dying
1091 * task anymore.
1092 */
1093void exit_itimers(struct task_struct *tsk)
1094{
1095	struct list_head timers;
1096	struct k_itimer *tmr;
1097
1098	if (list_empty(&tsk->signal->posix_timers))
1099		return;
1100
1101	/* Protect against concurrent read via /proc/$PID/timers */
1102	spin_lock_irq(&tsk->sighand->siglock);
1103	list_replace_init(&tsk->signal->posix_timers, &timers);
1104	spin_unlock_irq(&tsk->sighand->siglock);
1105
1106	/* The timers are not longer accessible via tsk::signal */
1107	while (!list_empty(&timers)) {
1108		tmr = list_first_entry(&timers, struct k_itimer, list);
1109		itimer_delete(tmr);
1110	}
1111}
1112
1113SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1114		const struct __kernel_timespec __user *, tp)
1115{
1116	const struct k_clock *kc = clockid_to_kclock(which_clock);
1117	struct timespec64 new_tp;
1118
1119	if (!kc || !kc->clock_set)
1120		return -EINVAL;
1121
1122	if (get_timespec64(&new_tp, tp))
1123		return -EFAULT;
1124
1125	/*
1126	 * Permission checks have to be done inside the clock specific
1127	 * setter callback.
1128	 */
1129	return kc->clock_set(which_clock, &new_tp);
1130}
1131
1132SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1133		struct __kernel_timespec __user *, tp)
1134{
1135	const struct k_clock *kc = clockid_to_kclock(which_clock);
1136	struct timespec64 kernel_tp;
1137	int error;
1138
1139	if (!kc)
1140		return -EINVAL;
1141
1142	error = kc->clock_get_timespec(which_clock, &kernel_tp);
1143
1144	if (!error && put_timespec64(&kernel_tp, tp))
1145		error = -EFAULT;
1146
1147	return error;
1148}
1149
1150int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1151{
1152	const struct k_clock *kc = clockid_to_kclock(which_clock);
1153
1154	if (!kc)
1155		return -EINVAL;
1156	if (!kc->clock_adj)
1157		return -EOPNOTSUPP;
1158
1159	return kc->clock_adj(which_clock, ktx);
1160}
1161
1162SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1163		struct __kernel_timex __user *, utx)
1164{
1165	struct __kernel_timex ktx;
1166	int err;
1167
1168	if (copy_from_user(&ktx, utx, sizeof(ktx)))
1169		return -EFAULT;
1170
1171	err = do_clock_adjtime(which_clock, &ktx);
1172
1173	if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1174		return -EFAULT;
1175
1176	return err;
1177}
1178
1179/**
1180 * sys_clock_getres - Get the resolution of a clock
1181 * @which_clock:	The clock to get the resolution for
1182 * @tp:			Pointer to a a user space timespec64 for storage
1183 *
1184 * POSIX defines:
1185 *
1186 * "The clock_getres() function shall return the resolution of any
1187 * clock. Clock resolutions are implementation-defined and cannot be set by
1188 * a process. If the argument res is not NULL, the resolution of the
1189 * specified clock shall be stored in the location pointed to by res. If
1190 * res is NULL, the clock resolution is not returned. If the time argument
1191 * of clock_settime() is not a multiple of res, then the value is truncated
1192 * to a multiple of res."
1193 *
1194 * Due to the various hardware constraints the real resolution can vary
1195 * wildly and even change during runtime when the underlying devices are
1196 * replaced. The kernel also can use hardware devices with different
1197 * resolutions for reading the time and for arming timers.
1198 *
1199 * The kernel therefore deviates from the POSIX spec in various aspects:
1200 *
1201 * 1) The resolution returned to user space
1202 *
1203 *    For CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, CLOCK_TAI,
1204 *    CLOCK_REALTIME_ALARM, CLOCK_BOOTTIME_ALAREM and CLOCK_MONOTONIC_RAW
1205 *    the kernel differentiates only two cases:
1206 *
1207 *    I)  Low resolution mode:
1208 *
1209 *	  When high resolution timers are disabled at compile or runtime
1210 *	  the resolution returned is nanoseconds per tick, which represents
1211 *	  the precision at which timers expire.
1212 *
1213 *    II) High resolution mode:
1214 *
1215 *	  When high resolution timers are enabled the resolution returned
1216 *	  is always one nanosecond independent of the actual resolution of
1217 *	  the underlying hardware devices.
1218 *
1219 *	  For CLOCK_*_ALARM the actual resolution depends on system
1220 *	  state. When system is running the resolution is the same as the
1221 *	  resolution of the other clocks. During suspend the actual
1222 *	  resolution is the resolution of the underlying RTC device which
1223 *	  might be way less precise than the clockevent device used during
1224 *	  running state.
1225 *
1226 *   For CLOCK_REALTIME_COARSE and CLOCK_MONOTONIC_COARSE the resolution
1227 *   returned is always nanoseconds per tick.
1228 *
1229 *   For CLOCK_PROCESS_CPUTIME and CLOCK_THREAD_CPUTIME the resolution
1230 *   returned is always one nanosecond under the assumption that the
1231 *   underlying scheduler clock has a better resolution than nanoseconds
1232 *   per tick.
1233 *
1234 *   For dynamic POSIX clocks (PTP devices) the resolution returned is
1235 *   always one nanosecond.
1236 *
1237 * 2) Affect on sys_clock_settime()
1238 *
1239 *    The kernel does not truncate the time which is handed in to
1240 *    sys_clock_settime(). The kernel internal timekeeping is always using
1241 *    nanoseconds precision independent of the clocksource device which is
1242 *    used to read the time from. The resolution of that device only
1243 *    affects the presicion of the time returned by sys_clock_gettime().
1244 *
1245 * Returns:
1246 *	0		Success. @tp contains the resolution
1247 *	-EINVAL		@which_clock is not a valid clock ID
1248 *	-EFAULT		Copying the resolution to @tp faulted
1249 *	-ENODEV		Dynamic POSIX clock is not backed by a device
1250 *	-EOPNOTSUPP	Dynamic POSIX clock does not support getres()
1251 */
1252SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1253		struct __kernel_timespec __user *, tp)
1254{
1255	const struct k_clock *kc = clockid_to_kclock(which_clock);
1256	struct timespec64 rtn_tp;
1257	int error;
1258
1259	if (!kc)
1260		return -EINVAL;
1261
1262	error = kc->clock_getres(which_clock, &rtn_tp);
1263
1264	if (!error && tp && put_timespec64(&rtn_tp, tp))
1265		error = -EFAULT;
1266
1267	return error;
1268}
1269
1270#ifdef CONFIG_COMPAT_32BIT_TIME
1271
1272SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1273		struct old_timespec32 __user *, tp)
1274{
1275	const struct k_clock *kc = clockid_to_kclock(which_clock);
1276	struct timespec64 ts;
1277
1278	if (!kc || !kc->clock_set)
1279		return -EINVAL;
1280
1281	if (get_old_timespec32(&ts, tp))
1282		return -EFAULT;
1283
1284	return kc->clock_set(which_clock, &ts);
1285}
1286
1287SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1288		struct old_timespec32 __user *, tp)
1289{
1290	const struct k_clock *kc = clockid_to_kclock(which_clock);
1291	struct timespec64 ts;
1292	int err;
1293
1294	if (!kc)
1295		return -EINVAL;
1296
1297	err = kc->clock_get_timespec(which_clock, &ts);
1298
1299	if (!err && put_old_timespec32(&ts, tp))
1300		err = -EFAULT;
1301
1302	return err;
1303}
1304
1305SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1306		struct old_timex32 __user *, utp)
1307{
1308	struct __kernel_timex ktx;
1309	int err;
1310
1311	err = get_old_timex32(&ktx, utp);
1312	if (err)
1313		return err;
1314
1315	err = do_clock_adjtime(which_clock, &ktx);
1316
1317	if (err >= 0 && put_old_timex32(utp, &ktx))
1318		return -EFAULT;
1319
1320	return err;
1321}
1322
1323SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1324		struct old_timespec32 __user *, tp)
1325{
1326	const struct k_clock *kc = clockid_to_kclock(which_clock);
1327	struct timespec64 ts;
1328	int err;
1329
1330	if (!kc)
1331		return -EINVAL;
1332
1333	err = kc->clock_getres(which_clock, &ts);
1334	if (!err && tp && put_old_timespec32(&ts, tp))
1335		return -EFAULT;
1336
1337	return err;
1338}
1339
1340#endif
1341
1342/*
1343 * sys_clock_nanosleep() for CLOCK_REALTIME and CLOCK_TAI
1344 */
1345static int common_nsleep(const clockid_t which_clock, int flags,
1346			 const struct timespec64 *rqtp)
1347{
1348	ktime_t texp = timespec64_to_ktime(*rqtp);
1349
1350	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1351				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1352				 which_clock);
1353}
1354
1355/*
1356 * sys_clock_nanosleep() for CLOCK_MONOTONIC and CLOCK_BOOTTIME
1357 *
1358 * Absolute nanosleeps for these clocks are time-namespace adjusted.
1359 */
1360static int common_nsleep_timens(const clockid_t which_clock, int flags,
1361				const struct timespec64 *rqtp)
1362{
1363	ktime_t texp = timespec64_to_ktime(*rqtp);
1364
1365	if (flags & TIMER_ABSTIME)
1366		texp = timens_ktime_to_host(which_clock, texp);
1367
1368	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1369				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1370				 which_clock);
1371}
1372
1373SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1374		const struct __kernel_timespec __user *, rqtp,
1375		struct __kernel_timespec __user *, rmtp)
1376{
1377	const struct k_clock *kc = clockid_to_kclock(which_clock);
1378	struct timespec64 t;
1379
1380	if (!kc)
1381		return -EINVAL;
1382	if (!kc->nsleep)
1383		return -EOPNOTSUPP;
1384
1385	if (get_timespec64(&t, rqtp))
1386		return -EFAULT;
1387
1388	if (!timespec64_valid(&t))
1389		return -EINVAL;
1390	if (flags & TIMER_ABSTIME)
1391		rmtp = NULL;
1392	current->restart_block.fn = do_no_restart_syscall;
1393	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1394	current->restart_block.nanosleep.rmtp = rmtp;
1395
1396	return kc->nsleep(which_clock, flags, &t);
1397}
1398
1399#ifdef CONFIG_COMPAT_32BIT_TIME
1400
1401SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1402		struct old_timespec32 __user *, rqtp,
1403		struct old_timespec32 __user *, rmtp)
1404{
1405	const struct k_clock *kc = clockid_to_kclock(which_clock);
1406	struct timespec64 t;
1407
1408	if (!kc)
1409		return -EINVAL;
1410	if (!kc->nsleep)
1411		return -EOPNOTSUPP;
1412
1413	if (get_old_timespec32(&t, rqtp))
1414		return -EFAULT;
1415
1416	if (!timespec64_valid(&t))
1417		return -EINVAL;
1418	if (flags & TIMER_ABSTIME)
1419		rmtp = NULL;
1420	current->restart_block.fn = do_no_restart_syscall;
1421	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1422	current->restart_block.nanosleep.compat_rmtp = rmtp;
1423
1424	return kc->nsleep(which_clock, flags, &t);
1425}
1426
1427#endif
1428
1429static const struct k_clock clock_realtime = {
1430	.clock_getres		= posix_get_hrtimer_res,
1431	.clock_get_timespec	= posix_get_realtime_timespec,
1432	.clock_get_ktime	= posix_get_realtime_ktime,
1433	.clock_set		= posix_clock_realtime_set,
1434	.clock_adj		= posix_clock_realtime_adj,
1435	.nsleep			= common_nsleep,
1436	.timer_create		= common_timer_create,
1437	.timer_set		= common_timer_set,
1438	.timer_get		= common_timer_get,
1439	.timer_del		= common_timer_del,
1440	.timer_rearm		= common_hrtimer_rearm,
1441	.timer_forward		= common_hrtimer_forward,
1442	.timer_remaining	= common_hrtimer_remaining,
1443	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1444	.timer_wait_running	= common_timer_wait_running,
1445	.timer_arm		= common_hrtimer_arm,
1446};
1447
1448static const struct k_clock clock_monotonic = {
1449	.clock_getres		= posix_get_hrtimer_res,
1450	.clock_get_timespec	= posix_get_monotonic_timespec,
1451	.clock_get_ktime	= posix_get_monotonic_ktime,
1452	.nsleep			= common_nsleep_timens,
1453	.timer_create		= common_timer_create,
1454	.timer_set		= common_timer_set,
1455	.timer_get		= common_timer_get,
1456	.timer_del		= common_timer_del,
1457	.timer_rearm		= common_hrtimer_rearm,
1458	.timer_forward		= common_hrtimer_forward,
1459	.timer_remaining	= common_hrtimer_remaining,
1460	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1461	.timer_wait_running	= common_timer_wait_running,
1462	.timer_arm		= common_hrtimer_arm,
1463};
1464
1465static const struct k_clock clock_monotonic_raw = {
1466	.clock_getres		= posix_get_hrtimer_res,
1467	.clock_get_timespec	= posix_get_monotonic_raw,
1468};
1469
1470static const struct k_clock clock_realtime_coarse = {
1471	.clock_getres		= posix_get_coarse_res,
1472	.clock_get_timespec	= posix_get_realtime_coarse,
1473};
1474
1475static const struct k_clock clock_monotonic_coarse = {
1476	.clock_getres		= posix_get_coarse_res,
1477	.clock_get_timespec	= posix_get_monotonic_coarse,
1478};
1479
1480static const struct k_clock clock_tai = {
1481	.clock_getres		= posix_get_hrtimer_res,
1482	.clock_get_ktime	= posix_get_tai_ktime,
1483	.clock_get_timespec	= posix_get_tai_timespec,
1484	.nsleep			= common_nsleep,
1485	.timer_create		= common_timer_create,
1486	.timer_set		= common_timer_set,
1487	.timer_get		= common_timer_get,
1488	.timer_del		= common_timer_del,
1489	.timer_rearm		= common_hrtimer_rearm,
1490	.timer_forward		= common_hrtimer_forward,
1491	.timer_remaining	= common_hrtimer_remaining,
1492	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1493	.timer_wait_running	= common_timer_wait_running,
1494	.timer_arm		= common_hrtimer_arm,
1495};
1496
1497static const struct k_clock clock_boottime = {
1498	.clock_getres		= posix_get_hrtimer_res,
1499	.clock_get_ktime	= posix_get_boottime_ktime,
1500	.clock_get_timespec	= posix_get_boottime_timespec,
1501	.nsleep			= common_nsleep_timens,
1502	.timer_create		= common_timer_create,
1503	.timer_set		= common_timer_set,
1504	.timer_get		= common_timer_get,
1505	.timer_del		= common_timer_del,
1506	.timer_rearm		= common_hrtimer_rearm,
1507	.timer_forward		= common_hrtimer_forward,
1508	.timer_remaining	= common_hrtimer_remaining,
1509	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1510	.timer_wait_running	= common_timer_wait_running,
1511	.timer_arm		= common_hrtimer_arm,
1512};
1513
1514static const struct k_clock * const posix_clocks[] = {
1515	[CLOCK_REALTIME]		= &clock_realtime,
1516	[CLOCK_MONOTONIC]		= &clock_monotonic,
1517	[CLOCK_PROCESS_CPUTIME_ID]	= &clock_process,
1518	[CLOCK_THREAD_CPUTIME_ID]	= &clock_thread,
1519	[CLOCK_MONOTONIC_RAW]		= &clock_monotonic_raw,
1520	[CLOCK_REALTIME_COARSE]		= &clock_realtime_coarse,
1521	[CLOCK_MONOTONIC_COARSE]	= &clock_monotonic_coarse,
1522	[CLOCK_BOOTTIME]		= &clock_boottime,
1523	[CLOCK_REALTIME_ALARM]		= &alarm_clock,
1524	[CLOCK_BOOTTIME_ALARM]		= &alarm_clock,
1525	[CLOCK_TAI]			= &clock_tai,
1526};
1527
1528static const struct k_clock *clockid_to_kclock(const clockid_t id)
1529{
1530	clockid_t idx = id;
1531
1532	if (id < 0) {
1533		return (id & CLOCKFD_MASK) == CLOCKFD ?
1534			&clock_posix_dynamic : &clock_posix_cpu;
1535	}
1536
1537	if (id >= ARRAY_SIZE(posix_clocks))
1538		return NULL;
1539
1540	return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1541}