Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0+
   2/*
   3 * 2002-10-15  Posix Clocks & timers
   4 *                           by George Anzinger george@mvista.com
   5 *			     Copyright (C) 2002 2003 by MontaVista Software.
   6 *
   7 * 2004-06-01  Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug.
   8 *			     Copyright (C) 2004 Boris Hu
   9 *
  10 * These are all the functions necessary to implement POSIX clocks & timers
  11 */
  12#include <linux/mm.h>
  13#include <linux/interrupt.h>
  14#include <linux/slab.h>
  15#include <linux/time.h>
  16#include <linux/mutex.h>
  17#include <linux/sched/task.h>
  18
  19#include <linux/uaccess.h>
  20#include <linux/list.h>
  21#include <linux/init.h>
  22#include <linux/compiler.h>
  23#include <linux/hash.h>
  24#include <linux/posix-clock.h>
  25#include <linux/posix-timers.h>
  26#include <linux/syscalls.h>
  27#include <linux/wait.h>
  28#include <linux/workqueue.h>
  29#include <linux/export.h>
  30#include <linux/hashtable.h>
  31#include <linux/compat.h>
  32#include <linux/nospec.h>
  33#include <linux/time_namespace.h>
  34
  35#include "timekeeping.h"
  36#include "posix-timers.h"
  37
  38/*
  39 * Management arrays for POSIX timers. Timers are now kept in static hash table
  40 * with 512 entries.
  41 * Timer ids are allocated by local routine, which selects proper hash head by
  42 * key, constructed from current->signal address and per signal struct counter.
  43 * This keeps timer ids unique per process, but now they can intersect between
  44 * processes.
  45 */
  46
  47/*
  48 * Lets keep our timers in a slab cache :-)
 
 
 
 
 
 
  49 */
  50static struct kmem_cache *posix_timers_cache;
  51
  52static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
  53static DEFINE_SPINLOCK(hash_lock);
  54
  55static const struct k_clock * const posix_clocks[];
  56static const struct k_clock *clockid_to_kclock(const clockid_t id);
  57static const struct k_clock clock_realtime, clock_monotonic;
  58
  59/*
  60 * we assume that the new SIGEV_THREAD_ID shares no bits with the other
  61 * SIGEV values.  Here we put out an error if this assumption fails.
  62 */
  63#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
  64                       ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
  65#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
  66#endif
  67
  68/*
  69 * The timer ID is turned into a timer address by idr_find().
  70 * Verifying a valid ID consists of:
  71 *
  72 * a) checking that idr_find() returns other than -1.
  73 * b) checking that the timer id matches the one in the timer itself.
  74 * c) that the timer owner is in the callers thread group.
  75 */
  76
  77/*
  78 * CLOCKs: The POSIX standard calls for a couple of clocks and allows us
  79 *	    to implement others.  This structure defines the various
  80 *	    clocks.
  81 *
  82 * RESOLUTION: Clock resolution is used to round up timer and interval
  83 *	    times, NOT to report clock times, which are reported with as
  84 *	    much resolution as the system can muster.  In some cases this
  85 *	    resolution may depend on the underlying clock hardware and
  86 *	    may not be quantifiable until run time, and only then is the
  87 *	    necessary code is written.	The standard says we should say
  88 *	    something about this issue in the documentation...
  89 *
  90 * FUNCTIONS: The CLOCKs structure defines possible functions to
  91 *	    handle various clock functions.
  92 *
  93 *	    The standard POSIX timer management code assumes the
  94 *	    following: 1.) The k_itimer struct (sched.h) is used for
  95 *	    the timer.  2.) The list, it_lock, it_clock, it_id and
  96 *	    it_pid fields are not modified by timer code.
  97 *
  98 * Permissions: It is assumed that the clock_settime() function defined
  99 *	    for each clock will take care of permission checks.	 Some
 100 *	    clocks may be set able by any user (i.e. local process
 101 *	    clocks) others not.	 Currently the only set able clock we
 102 *	    have is CLOCK_REALTIME and its high res counter part, both of
 103 *	    which we beg off on and pass to do_sys_settimeofday().
 104 */
 105static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
 106
 107#define lock_timer(tid, flags)						   \
 108({	struct k_itimer *__timr;					   \
 109	__cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags));  \
 110	__timr;								   \
 111})
 112
 113static int hash(struct signal_struct *sig, unsigned int nr)
 114{
 115	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
 116}
 117
 118static struct k_itimer *__posix_timers_find(struct hlist_head *head,
 119					    struct signal_struct *sig,
 120					    timer_t id)
 121{
 122	struct k_itimer *timer;
 123
 124	hlist_for_each_entry_rcu(timer, head, t_hash,
 125				 lockdep_is_held(&hash_lock)) {
 126		if ((timer->it_signal == sig) && (timer->it_id == id))
 127			return timer;
 128	}
 129	return NULL;
 130}
 131
 132static struct k_itimer *posix_timer_by_id(timer_t id)
 133{
 134	struct signal_struct *sig = current->signal;
 135	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
 136
 137	return __posix_timers_find(head, sig, id);
 138}
 139
 140static int posix_timer_add(struct k_itimer *timer)
 141{
 142	struct signal_struct *sig = current->signal;
 143	int first_free_id = sig->posix_timer_id;
 144	struct hlist_head *head;
 145	int ret = -ENOENT;
 146
 147	do {
 
 
 
 
 148		spin_lock(&hash_lock);
 149		head = &posix_timers_hashtable[hash(sig, sig->posix_timer_id)];
 150		if (!__posix_timers_find(head, sig, sig->posix_timer_id)) {
 
 
 
 
 
 151			hlist_add_head_rcu(&timer->t_hash, head);
 152			ret = sig->posix_timer_id;
 
 153		}
 154		if (++sig->posix_timer_id < 0)
 155			sig->posix_timer_id = 0;
 156		if ((sig->posix_timer_id == first_free_id) && (ret == -ENOENT))
 157			/* Loop over all possible ids completed */
 158			ret = -EAGAIN;
 159		spin_unlock(&hash_lock);
 160	} while (ret == -ENOENT);
 161	return ret;
 
 162}
 163
 164static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
 165{
 166	spin_unlock_irqrestore(&timr->it_lock, flags);
 167}
 168
 169/* Get clock_realtime */
 170static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp)
 171{
 172	ktime_get_real_ts64(tp);
 173	return 0;
 174}
 175
 176static ktime_t posix_get_realtime_ktime(clockid_t which_clock)
 177{
 178	return ktime_get_real();
 179}
 180
 181/* Set clock_realtime */
 182static int posix_clock_realtime_set(const clockid_t which_clock,
 183				    const struct timespec64 *tp)
 184{
 185	return do_sys_settimeofday64(tp, NULL);
 186}
 187
 188static int posix_clock_realtime_adj(const clockid_t which_clock,
 189				    struct __kernel_timex *t)
 190{
 191	return do_adjtimex(t);
 192}
 193
 194/*
 195 * Get monotonic time for posix timers
 196 */
 197static int posix_get_monotonic_timespec(clockid_t which_clock, struct timespec64 *tp)
 198{
 199	ktime_get_ts64(tp);
 200	timens_add_monotonic(tp);
 201	return 0;
 202}
 203
 204static ktime_t posix_get_monotonic_ktime(clockid_t which_clock)
 205{
 206	return ktime_get();
 207}
 208
 209/*
 210 * Get monotonic-raw time for posix timers
 211 */
 212static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
 213{
 214	ktime_get_raw_ts64(tp);
 215	timens_add_monotonic(tp);
 216	return 0;
 217}
 218
 219
 220static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
 221{
 222	ktime_get_coarse_real_ts64(tp);
 223	return 0;
 224}
 225
 226static int posix_get_monotonic_coarse(clockid_t which_clock,
 227						struct timespec64 *tp)
 228{
 229	ktime_get_coarse_ts64(tp);
 230	timens_add_monotonic(tp);
 231	return 0;
 232}
 233
 234static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
 235{
 236	*tp = ktime_to_timespec64(KTIME_LOW_RES);
 237	return 0;
 238}
 239
 240static int posix_get_boottime_timespec(const clockid_t which_clock, struct timespec64 *tp)
 241{
 242	ktime_get_boottime_ts64(tp);
 243	timens_add_boottime(tp);
 244	return 0;
 245}
 246
 247static ktime_t posix_get_boottime_ktime(const clockid_t which_clock)
 248{
 249	return ktime_get_boottime();
 250}
 251
 252static int posix_get_tai_timespec(clockid_t which_clock, struct timespec64 *tp)
 253{
 254	ktime_get_clocktai_ts64(tp);
 255	return 0;
 256}
 257
 258static ktime_t posix_get_tai_ktime(clockid_t which_clock)
 259{
 260	return ktime_get_clocktai();
 261}
 262
 263static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
 264{
 265	tp->tv_sec = 0;
 266	tp->tv_nsec = hrtimer_resolution;
 267	return 0;
 268}
 269
 270/*
 271 * Initialize everything, well, just everything in Posix clocks/timers ;)
 272 */
 273static __init int init_posix_timers(void)
 274{
 275	posix_timers_cache = kmem_cache_create("posix_timers_cache",
 276					sizeof(struct k_itimer), 0,
 277					SLAB_PANIC | SLAB_ACCOUNT, NULL);
 278	return 0;
 279}
 280__initcall(init_posix_timers);
 281
 282/*
 283 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
 284 * are of type int. Clamp the overrun value to INT_MAX
 285 */
 286static inline int timer_overrun_to_int(struct k_itimer *timr, int baseval)
 287{
 288	s64 sum = timr->it_overrun_last + (s64)baseval;
 
 289
 290	return sum > (s64)INT_MAX ? INT_MAX : (int)sum;
 291}
 292
 293static void common_hrtimer_rearm(struct k_itimer *timr)
 294{
 295	struct hrtimer *timer = &timr->it.real.timer;
 296
 297	timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
 298					    timr->it_interval);
 299	hrtimer_restart(timer);
 300}
 301
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 302/*
 303 * This function is exported for use by the signal deliver code.  It is
 304 * called just prior to the info block being released and passes that
 305 * block to us.  It's function is to update the overrun entry AND to
 306 * restart the timer.  It should only be called if the timer is to be
 307 * restarted (i.e. we have flagged this in the sys_private entry of the
 308 * info block).
 309 *
 310 * To protect against the timer going away while the interrupt is queued,
 311 * we require that the it_requeue_pending flag be set.
 312 */
 313void posixtimer_rearm(struct kernel_siginfo *info)
 314{
 315	struct k_itimer *timr;
 316	unsigned long flags;
 317
 318	timr = lock_timer(info->si_tid, &flags);
 319	if (!timr)
 320		return;
 321
 322	if (timr->it_interval && timr->it_requeue_pending == info->si_sys_private) {
 323		timr->kclock->timer_rearm(timr);
 
 
 
 324
 325		timr->it_active = 1;
 326		timr->it_overrun_last = timr->it_overrun;
 327		timr->it_overrun = -1LL;
 328		++timr->it_requeue_pending;
 329
 330		info->si_overrun = timer_overrun_to_int(timr, info->si_overrun);
 331	}
 332
 333	unlock_timer(timr, flags);
 
 334}
 335
 336int posix_timer_event(struct k_itimer *timr, int si_private)
 337{
 338	enum pid_type type;
 339	int ret;
 340	/*
 341	 * FIXME: if ->sigq is queued we can race with
 342	 * dequeue_signal()->posixtimer_rearm().
 343	 *
 344	 * If dequeue_signal() sees the "right" value of
 345	 * si_sys_private it calls posixtimer_rearm().
 346	 * We re-queue ->sigq and drop ->it_lock().
 347	 * posixtimer_rearm() locks the timer
 348	 * and re-schedules it while ->sigq is pending.
 349	 * Not really bad, but not that we want.
 350	 */
 351	timr->sigq->info.si_sys_private = si_private;
 352
 353	type = !(timr->it_sigev_notify & SIGEV_THREAD_ID) ? PIDTYPE_TGID : PIDTYPE_PID;
 354	ret = send_sigqueue(timr->sigq, timr->it_pid, type);
 355	/* If we failed to send the signal the timer stops. */
 356	return ret > 0;
 357}
 358
 359/*
 360 * This function gets called when a POSIX.1b interval timer expires.  It
 361 * is used as a callback from the kernel internal timer.  The
 362 * run_timer_list code ALWAYS calls with interrupts on.
 363
 364 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
 365 */
 366static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
 367{
 368	struct k_itimer *timr;
 369	unsigned long flags;
 370	int si_private = 0;
 371	enum hrtimer_restart ret = HRTIMER_NORESTART;
 372
 373	timr = container_of(timer, struct k_itimer, it.real.timer);
 374	spin_lock_irqsave(&timr->it_lock, flags);
 375
 376	timr->it_active = 0;
 377	if (timr->it_interval != 0)
 378		si_private = ++timr->it_requeue_pending;
 379
 380	if (posix_timer_event(timr, si_private)) {
 381		/*
 382		 * signal was not sent because of sig_ignor
 383		 * we will not get a call back to restart it AND
 384		 * it should be restarted.
 385		 */
 386		if (timr->it_interval != 0) {
 387			ktime_t now = hrtimer_cb_get_time(timer);
 388
 389			/*
 390			 * FIXME: What we really want, is to stop this
 391			 * timer completely and restart it in case the
 392			 * SIG_IGN is removed. This is a non trivial
 393			 * change which involves sighand locking
 394			 * (sigh !), which we don't want to do late in
 395			 * the release cycle.
 396			 *
 397			 * For now we just let timers with an interval
 398			 * less than a jiffie expire every jiffie to
 399			 * avoid softirq starvation in case of SIG_IGN
 400			 * and a very small interval, which would put
 401			 * the timer right back on the softirq pending
 402			 * list. By moving now ahead of time we trick
 403			 * hrtimer_forward() to expire the timer
 404			 * later, while we still maintain the overrun
 405			 * accuracy, but have some inconsistency in
 406			 * the timer_gettime() case. This is at least
 407			 * better than a starved softirq. A more
 408			 * complex fix which solves also another related
 409			 * inconsistency is already in the pipeline.
 410			 */
 411#ifdef CONFIG_HIGH_RES_TIMERS
 412			{
 413				ktime_t kj = NSEC_PER_SEC / HZ;
 414
 415				if (timr->it_interval < kj)
 416					now = ktime_add(now, kj);
 417			}
 418#endif
 419			timr->it_overrun += hrtimer_forward(timer, now,
 420							    timr->it_interval);
 421			ret = HRTIMER_RESTART;
 422			++timr->it_requeue_pending;
 423			timr->it_active = 1;
 424		}
 425	}
 426
 427	unlock_timer(timr, flags);
 428	return ret;
 429}
 430
 431static struct pid *good_sigevent(sigevent_t * event)
 432{
 433	struct pid *pid = task_tgid(current);
 434	struct task_struct *rtn;
 435
 436	switch (event->sigev_notify) {
 437	case SIGEV_SIGNAL | SIGEV_THREAD_ID:
 438		pid = find_vpid(event->sigev_notify_thread_id);
 439		rtn = pid_task(pid, PIDTYPE_PID);
 440		if (!rtn || !same_thread_group(rtn, current))
 441			return NULL;
 442		fallthrough;
 443	case SIGEV_SIGNAL:
 444	case SIGEV_THREAD:
 445		if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
 446			return NULL;
 447		fallthrough;
 448	case SIGEV_NONE:
 449		return pid;
 450	default:
 451		return NULL;
 452	}
 453}
 454
 455static struct k_itimer * alloc_posix_timer(void)
 456{
 457	struct k_itimer *tmr;
 458	tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
 459	if (!tmr)
 460		return tmr;
 461	if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
 
 462		kmem_cache_free(posix_timers_cache, tmr);
 463		return NULL;
 464	}
 465	clear_siginfo(&tmr->sigq->info);
 466	return tmr;
 467}
 468
 469static void k_itimer_rcu_free(struct rcu_head *head)
 470{
 471	struct k_itimer *tmr = container_of(head, struct k_itimer, rcu);
 472
 473	kmem_cache_free(posix_timers_cache, tmr);
 
 474}
 475
 476#define IT_ID_SET	1
 477#define IT_ID_NOT_SET	0
 478static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
 479{
 480	if (it_id_set) {
 481		unsigned long flags;
 482		spin_lock_irqsave(&hash_lock, flags);
 483		hlist_del_rcu(&tmr->t_hash);
 484		spin_unlock_irqrestore(&hash_lock, flags);
 485	}
 486	put_pid(tmr->it_pid);
 487	sigqueue_free(tmr->sigq);
 488	call_rcu(&tmr->rcu, k_itimer_rcu_free);
 489}
 490
 491static int common_timer_create(struct k_itimer *new_timer)
 492{
 493	hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
 494	return 0;
 495}
 496
 497/* Create a POSIX.1b interval timer. */
 498static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 499			   timer_t __user *created_timer_id)
 500{
 501	const struct k_clock *kc = clockid_to_kclock(which_clock);
 502	struct k_itimer *new_timer;
 503	int error, new_timer_id;
 504	int it_id_set = IT_ID_NOT_SET;
 505
 506	if (!kc)
 507		return -EINVAL;
 508	if (!kc->timer_create)
 509		return -EOPNOTSUPP;
 510
 511	new_timer = alloc_posix_timer();
 512	if (unlikely(!new_timer))
 513		return -EAGAIN;
 514
 515	spin_lock_init(&new_timer->it_lock);
 
 
 
 
 
 
 516	new_timer_id = posix_timer_add(new_timer);
 517	if (new_timer_id < 0) {
 518		error = new_timer_id;
 519		goto out;
 520	}
 521
 522	it_id_set = IT_ID_SET;
 523	new_timer->it_id = (timer_t) new_timer_id;
 524	new_timer->it_clock = which_clock;
 525	new_timer->kclock = kc;
 526	new_timer->it_overrun = -1LL;
 527
 528	if (event) {
 529		rcu_read_lock();
 530		new_timer->it_pid = get_pid(good_sigevent(event));
 531		rcu_read_unlock();
 532		if (!new_timer->it_pid) {
 533			error = -EINVAL;
 534			goto out;
 535		}
 536		new_timer->it_sigev_notify     = event->sigev_notify;
 537		new_timer->sigq->info.si_signo = event->sigev_signo;
 538		new_timer->sigq->info.si_value = event->sigev_value;
 539	} else {
 540		new_timer->it_sigev_notify     = SIGEV_SIGNAL;
 541		new_timer->sigq->info.si_signo = SIGALRM;
 542		memset(&new_timer->sigq->info.si_value, 0, sizeof(sigval_t));
 543		new_timer->sigq->info.si_value.sival_int = new_timer->it_id;
 544		new_timer->it_pid = get_pid(task_tgid(current));
 545	}
 546
 547	new_timer->sigq->info.si_tid   = new_timer->it_id;
 548	new_timer->sigq->info.si_code  = SI_TIMER;
 
 
 
 
 
 549
 550	if (copy_to_user(created_timer_id,
 551			 &new_timer_id, sizeof (new_timer_id))) {
 552		error = -EFAULT;
 553		goto out;
 554	}
 555
 
 
 
 
 
 
 556	error = kc->timer_create(new_timer);
 557	if (error)
 558		goto out;
 559
 560	spin_lock_irq(&current->sighand->siglock);
 561	new_timer->it_signal = current->signal;
 562	list_add(&new_timer->list, &current->signal->posix_timers);
 
 563	spin_unlock_irq(&current->sighand->siglock);
 564
 565	return 0;
 566	/*
 567	 * In the case of the timer belonging to another task, after
 568	 * the task is unlocked, the timer is owned by the other task
 569	 * and may cease to exist at any time.  Don't use or modify
 570	 * new_timer after the unlock call.
 571	 */
 
 572out:
 573	release_posix_timer(new_timer, it_id_set);
 574	return error;
 575}
 576
 577SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
 578		struct sigevent __user *, timer_event_spec,
 579		timer_t __user *, created_timer_id)
 580{
 581	if (timer_event_spec) {
 582		sigevent_t event;
 583
 584		if (copy_from_user(&event, timer_event_spec, sizeof (event)))
 585			return -EFAULT;
 586		return do_timer_create(which_clock, &event, created_timer_id);
 587	}
 588	return do_timer_create(which_clock, NULL, created_timer_id);
 589}
 590
 591#ifdef CONFIG_COMPAT
 592COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
 593		       struct compat_sigevent __user *, timer_event_spec,
 594		       timer_t __user *, created_timer_id)
 595{
 596	if (timer_event_spec) {
 597		sigevent_t event;
 598
 599		if (get_compat_sigevent(&event, timer_event_spec))
 600			return -EFAULT;
 601		return do_timer_create(which_clock, &event, created_timer_id);
 602	}
 603	return do_timer_create(which_clock, NULL, created_timer_id);
 604}
 605#endif
 606
 607/*
 608 * Locking issues: We need to protect the result of the id look up until
 609 * we get the timer locked down so it is not deleted under us.  The
 610 * removal is done under the idr spinlock so we use that here to bridge
 611 * the find to the timer lock.  To avoid a dead lock, the timer id MUST
 612 * be release with out holding the timer lock.
 613 */
 614static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 615{
 616	struct k_itimer *timr;
 617
 618	/*
 619	 * timer_t could be any type >= int and we want to make sure any
 620	 * @timer_id outside positive int range fails lookup.
 621	 */
 622	if ((unsigned long long)timer_id > INT_MAX)
 623		return NULL;
 624
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 625	rcu_read_lock();
 626	timr = posix_timer_by_id(timer_id);
 627	if (timr) {
 628		spin_lock_irqsave(&timr->it_lock, *flags);
 
 
 
 
 629		if (timr->it_signal == current->signal) {
 630			rcu_read_unlock();
 631			return timr;
 632		}
 633		spin_unlock_irqrestore(&timr->it_lock, *flags);
 634	}
 635	rcu_read_unlock();
 636
 637	return NULL;
 638}
 639
 640static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
 641{
 642	struct hrtimer *timer = &timr->it.real.timer;
 643
 644	return __hrtimer_expires_remaining_adjusted(timer, now);
 645}
 646
 647static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
 648{
 649	struct hrtimer *timer = &timr->it.real.timer;
 650
 651	return hrtimer_forward(timer, now, timr->it_interval);
 652}
 653
 654/*
 655 * Get the time remaining on a POSIX.1b interval timer.  This function
 656 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
 657 * mess with irq.
 658 *
 659 * We have a couple of messes to clean up here.  First there is the case
 660 * of a timer that has a requeue pending.  These timers should appear to
 661 * be in the timer list with an expiry as if we were to requeue them
 662 * now.
 663 *
 664 * The second issue is the SIGEV_NONE timer which may be active but is
 665 * not really ever put in the timer list (to save system resources).
 666 * This timer may be expired, and if so, we will do it here.  Otherwise
 667 * it is the same as a requeue pending timer WRT to what we should
 668 * report.
 669 */
 670void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
 671{
 672	const struct k_clock *kc = timr->kclock;
 673	ktime_t now, remaining, iv;
 674	bool sig_none;
 675
 676	sig_none = timr->it_sigev_notify == SIGEV_NONE;
 677	iv = timr->it_interval;
 678
 679	/* interval timer ? */
 680	if (iv) {
 681		cur_setting->it_interval = ktime_to_timespec64(iv);
 682	} else if (!timr->it_active) {
 683		/*
 684		 * SIGEV_NONE oneshot timers are never queued. Check them
 685		 * below.
 
 
 
 
 686		 */
 687		if (!sig_none)
 688			return;
 689	}
 690
 691	now = kc->clock_get_ktime(timr->it_clock);
 692
 693	/*
 694	 * When a requeue is pending or this is a SIGEV_NONE timer move the
 695	 * expiry time forward by intervals, so expiry is > now.
 
 696	 */
 697	if (iv && (timr->it_requeue_pending & REQUEUE_PENDING || sig_none))
 698		timr->it_overrun += kc->timer_forward(timr, now);
 699
 700	remaining = kc->timer_remaining(timr, now);
 701	/* Return 0 only, when the timer is expired and not pending */
 
 
 
 
 
 
 
 
 702	if (remaining <= 0) {
 703		/*
 704		 * A single shot SIGEV_NONE timer must return 0, when
 705		 * it is expired !
 
 
 706		 */
 707		if (!sig_none)
 708			cur_setting->it_value.tv_nsec = 1;
 709	} else {
 710		cur_setting->it_value = ktime_to_timespec64(remaining);
 711	}
 712}
 713
 714/* Get the time remaining on a POSIX.1b interval timer. */
 715static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 716{
 717	struct k_itimer *timr;
 718	const struct k_clock *kc;
 
 719	unsigned long flags;
 720	int ret = 0;
 721
 722	timr = lock_timer(timer_id, &flags);
 723	if (!timr)
 724		return -EINVAL;
 725
 726	memset(setting, 0, sizeof(*setting));
 727	kc = timr->kclock;
 728	if (WARN_ON_ONCE(!kc || !kc->timer_get))
 729		ret = -EINVAL;
 730	else
 731		kc->timer_get(timr, setting);
 732
 733	unlock_timer(timr, flags);
 734	return ret;
 735}
 736
 737/* Get the time remaining on a POSIX.1b interval timer. */
 738SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
 739		struct __kernel_itimerspec __user *, setting)
 740{
 741	struct itimerspec64 cur_setting;
 742
 743	int ret = do_timer_gettime(timer_id, &cur_setting);
 744	if (!ret) {
 745		if (put_itimerspec64(&cur_setting, setting))
 746			ret = -EFAULT;
 747	}
 748	return ret;
 749}
 750
 751#ifdef CONFIG_COMPAT_32BIT_TIME
 752
 753SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
 754		struct old_itimerspec32 __user *, setting)
 755{
 756	struct itimerspec64 cur_setting;
 757
 758	int ret = do_timer_gettime(timer_id, &cur_setting);
 759	if (!ret) {
 760		if (put_old_itimerspec32(&cur_setting, setting))
 761			ret = -EFAULT;
 762	}
 763	return ret;
 764}
 765
 766#endif
 767
 768/*
 769 * Get the number of overruns of a POSIX.1b interval timer.  This is to
 770 * be the overrun of the timer last delivered.  At the same time we are
 771 * accumulating overruns on the next timer.  The overrun is frozen when
 772 * the signal is delivered, either at the notify time (if the info block
 773 * is not queued) or at the actual delivery time (as we are informed by
 774 * the call back to posixtimer_rearm().  So all we need to do is
 775 * to pick up the frozen overrun.
 
 
 
 
 
 
 
 
 
 776 */
 777SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 778{
 779	struct k_itimer *timr;
 780	int overrun;
 781	unsigned long flags;
 
 782
 783	timr = lock_timer(timer_id, &flags);
 784	if (!timr)
 785		return -EINVAL;
 786
 787	overrun = timer_overrun_to_int(timr, 0);
 788	unlock_timer(timr, flags);
 789
 790	return overrun;
 791}
 792
 793static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
 794			       bool absolute, bool sigev_none)
 795{
 796	struct hrtimer *timer = &timr->it.real.timer;
 797	enum hrtimer_mode mode;
 798
 799	mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
 800	/*
 801	 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
 802	 * clock modifications, so they become CLOCK_MONOTONIC based under the
 803	 * hood. See hrtimer_init(). Update timr->kclock, so the generic
 804	 * functions which use timr->kclock->clock_get_*() work.
 805	 *
 806	 * Note: it_clock stays unmodified, because the next timer_set() might
 807	 * use ABSTIME, so it needs to switch back.
 808	 */
 809	if (timr->it_clock == CLOCK_REALTIME)
 810		timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
 811
 812	hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
 813	timr->it.real.timer.function = posix_timer_fn;
 814
 815	if (!absolute)
 816		expires = ktime_add_safe(expires, timer->base->get_time());
 817	hrtimer_set_expires(timer, expires);
 818
 819	if (!sigev_none)
 820		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 821}
 822
 823static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
 824{
 825	return hrtimer_try_to_cancel(&timr->it.real.timer);
 826}
 827
 828static void common_timer_wait_running(struct k_itimer *timer)
 829{
 830	hrtimer_cancel_wait_running(&timer->it.real.timer);
 831}
 832
 833/*
 834 * On PREEMPT_RT this prevent priority inversion against softirq kthread in
 835 * case it gets preempted while executing a timer callback. See comments in
 836 * hrtimer_cancel_wait_running. For PREEMPT_RT=n this just results in a
 837 * cpu_relax().
 
 
 
 
 
 
 
 
 838 */
 839static struct k_itimer *timer_wait_running(struct k_itimer *timer,
 840					   unsigned long *flags)
 841{
 842	const struct k_clock *kc = READ_ONCE(timer->kclock);
 843	timer_t timer_id = READ_ONCE(timer->it_id);
 844
 845	/* Prevent kfree(timer) after dropping the lock */
 846	rcu_read_lock();
 847	unlock_timer(timer, *flags);
 848
 
 
 
 
 849	if (!WARN_ON_ONCE(!kc->timer_wait_running))
 850		kc->timer_wait_running(timer);
 851
 852	rcu_read_unlock();
 853	/* Relock the timer. It might be not longer hashed. */
 854	return lock_timer(timer_id, flags);
 855}
 856
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 857/* Set a POSIX.1b interval timer. */
 858int common_timer_set(struct k_itimer *timr, int flags,
 859		     struct itimerspec64 *new_setting,
 860		     struct itimerspec64 *old_setting)
 861{
 862	const struct k_clock *kc = timr->kclock;
 863	bool sigev_none;
 864	ktime_t expires;
 865
 866	if (old_setting)
 867		common_timer_get(timr, old_setting);
 868
 869	/* Prevent rearming by clearing the interval */
 870	timr->it_interval = 0;
 871	/*
 872	 * Careful here. On SMP systems the timer expiry function could be
 873	 * active and spinning on timr->it_lock.
 874	 */
 875	if (kc->timer_try_to_cancel(timr) < 0)
 876		return TIMER_RETRY;
 877
 878	timr->it_active = 0;
 879	timr->it_requeue_pending = (timr->it_requeue_pending + 2) &
 880		~REQUEUE_PENDING;
 881	timr->it_overrun_last = 0;
 882
 883	/* Switch off the timer when it_value is zero */
 884	if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
 885		return 0;
 886
 887	timr->it_interval = timespec64_to_ktime(new_setting->it_interval);
 888	expires = timespec64_to_ktime(new_setting->it_value);
 889	if (flags & TIMER_ABSTIME)
 890		expires = timens_ktime_to_host(timr->it_clock, expires);
 891	sigev_none = timr->it_sigev_notify == SIGEV_NONE;
 892
 893	kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
 894	timr->it_active = !sigev_none;
 
 895	return 0;
 896}
 897
 898static int do_timer_settime(timer_t timer_id, int tmr_flags,
 899			    struct itimerspec64 *new_spec64,
 900			    struct itimerspec64 *old_spec64)
 901{
 902	const struct k_clock *kc;
 903	struct k_itimer *timr;
 904	unsigned long flags;
 905	int error = 0;
 906
 907	if (!timespec64_valid(&new_spec64->it_interval) ||
 908	    !timespec64_valid(&new_spec64->it_value))
 909		return -EINVAL;
 910
 911	if (old_spec64)
 912		memset(old_spec64, 0, sizeof(*old_spec64));
 913
 914	timr = lock_timer(timer_id, &flags);
 915retry:
 916	if (!timr)
 917		return -EINVAL;
 918
 
 
 
 
 
 
 919	kc = timr->kclock;
 920	if (WARN_ON_ONCE(!kc || !kc->timer_set))
 921		error = -EINVAL;
 922	else
 923		error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
 924
 925	if (error == TIMER_RETRY) {
 926		// We already got the old time...
 927		old_spec64 = NULL;
 928		/* Unlocks and relocks the timer if it still exists */
 929		timr = timer_wait_running(timr, &flags);
 930		goto retry;
 931	}
 932	unlock_timer(timr, flags);
 933
 934	return error;
 935}
 936
 937/* Set a POSIX.1b interval timer */
 938SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
 939		const struct __kernel_itimerspec __user *, new_setting,
 940		struct __kernel_itimerspec __user *, old_setting)
 941{
 942	struct itimerspec64 new_spec, old_spec;
 943	struct itimerspec64 *rtn = old_setting ? &old_spec : NULL;
 944	int error = 0;
 945
 946	if (!new_setting)
 947		return -EINVAL;
 948
 949	if (get_itimerspec64(&new_spec, new_setting))
 950		return -EFAULT;
 951
 
 952	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 953	if (!error && old_setting) {
 954		if (put_itimerspec64(&old_spec, old_setting))
 955			error = -EFAULT;
 956	}
 957	return error;
 958}
 959
 960#ifdef CONFIG_COMPAT_32BIT_TIME
 961SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
 962		struct old_itimerspec32 __user *, new,
 963		struct old_itimerspec32 __user *, old)
 964{
 965	struct itimerspec64 new_spec, old_spec;
 966	struct itimerspec64 *rtn = old ? &old_spec : NULL;
 967	int error = 0;
 968
 969	if (!new)
 970		return -EINVAL;
 971	if (get_old_itimerspec32(&new_spec, new))
 972		return -EFAULT;
 973
 974	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 975	if (!error && old) {
 976		if (put_old_itimerspec32(&old_spec, old))
 977			error = -EFAULT;
 978	}
 979	return error;
 980}
 981#endif
 982
 983int common_timer_del(struct k_itimer *timer)
 984{
 985	const struct k_clock *kc = timer->kclock;
 986
 987	timer->it_interval = 0;
 988	if (kc->timer_try_to_cancel(timer) < 0)
 989		return TIMER_RETRY;
 990	timer->it_active = 0;
 991	return 0;
 992}
 993
 
 
 
 
 
 
 
 
 
 
 
 
 994static inline int timer_delete_hook(struct k_itimer *timer)
 995{
 996	const struct k_clock *kc = timer->kclock;
 997
 
 
 
 998	if (WARN_ON_ONCE(!kc || !kc->timer_del))
 999		return -EINVAL;
1000	return kc->timer_del(timer);
1001}
1002
1003/* Delete a POSIX.1b interval timer. */
1004SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
1005{
1006	struct k_itimer *timer;
1007	unsigned long flags;
1008
1009	timer = lock_timer(timer_id, &flags);
1010
1011retry_delete:
1012	if (!timer)
1013		return -EINVAL;
1014
1015	if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
1016		/* Unlocks and relocks the timer if it still exists */
1017		timer = timer_wait_running(timer, &flags);
1018		goto retry_delete;
1019	}
1020
1021	spin_lock(&current->sighand->siglock);
1022	list_del(&timer->list);
1023	spin_unlock(&current->sighand->siglock);
1024	/*
1025	 * This keeps any tasks waiting on the spin lock from thinking
1026	 * they got something (see the lock code above).
 
 
 
 
 
1027	 */
1028	timer->it_signal = NULL;
 
1029
1030	unlock_timer(timer, flags);
1031	release_posix_timer(timer, IT_ID_SET);
1032	return 0;
1033}
1034
1035/*
1036 * return timer owned by the process, used by exit_itimers
 
1037 */
1038static void itimer_delete(struct k_itimer *timer)
1039{
1040retry_delete:
1041	spin_lock_irq(&timer->it_lock);
 
 
 
 
1042
 
 
 
 
 
 
 
1043	if (timer_delete_hook(timer) == TIMER_RETRY) {
1044		spin_unlock_irq(&timer->it_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
1045		goto retry_delete;
1046	}
1047	list_del(&timer->list);
 
 
 
 
 
 
 
 
 
 
1048
1049	spin_unlock_irq(&timer->it_lock);
1050	release_posix_timer(timer, IT_ID_SET);
1051}
1052
1053/*
1054 * This is called by do_exit or de_thread, only when nobody else can
1055 * modify the signal->posix_timers list. Yet we need sighand->siglock
1056 * to prevent the race with /proc/pid/timers.
1057 */
1058void exit_itimers(struct task_struct *tsk)
1059{
1060	struct list_head timers;
1061	struct k_itimer *tmr;
1062
1063	if (list_empty(&tsk->signal->posix_timers))
1064		return;
1065
 
1066	spin_lock_irq(&tsk->sighand->siglock);
1067	list_replace_init(&tsk->signal->posix_timers, &timers);
1068	spin_unlock_irq(&tsk->sighand->siglock);
1069
1070	while (!list_empty(&timers)) {
1071		tmr = list_first_entry(&timers, struct k_itimer, list);
1072		itimer_delete(tmr);
 
 
 
 
 
 
 
 
 
 
 
 
1073	}
1074}
1075
1076SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1077		const struct __kernel_timespec __user *, tp)
1078{
1079	const struct k_clock *kc = clockid_to_kclock(which_clock);
1080	struct timespec64 new_tp;
1081
1082	if (!kc || !kc->clock_set)
1083		return -EINVAL;
1084
1085	if (get_timespec64(&new_tp, tp))
1086		return -EFAULT;
1087
 
 
 
 
1088	return kc->clock_set(which_clock, &new_tp);
1089}
1090
1091SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1092		struct __kernel_timespec __user *, tp)
1093{
1094	const struct k_clock *kc = clockid_to_kclock(which_clock);
1095	struct timespec64 kernel_tp;
1096	int error;
1097
1098	if (!kc)
1099		return -EINVAL;
1100
1101	error = kc->clock_get_timespec(which_clock, &kernel_tp);
1102
1103	if (!error && put_timespec64(&kernel_tp, tp))
1104		error = -EFAULT;
1105
1106	return error;
1107}
1108
1109int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1110{
1111	const struct k_clock *kc = clockid_to_kclock(which_clock);
1112
1113	if (!kc)
1114		return -EINVAL;
1115	if (!kc->clock_adj)
1116		return -EOPNOTSUPP;
1117
1118	return kc->clock_adj(which_clock, ktx);
1119}
1120
1121SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1122		struct __kernel_timex __user *, utx)
1123{
1124	struct __kernel_timex ktx;
1125	int err;
1126
1127	if (copy_from_user(&ktx, utx, sizeof(ktx)))
1128		return -EFAULT;
1129
1130	err = do_clock_adjtime(which_clock, &ktx);
1131
1132	if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1133		return -EFAULT;
1134
1135	return err;
1136}
1137
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1138SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1139		struct __kernel_timespec __user *, tp)
1140{
1141	const struct k_clock *kc = clockid_to_kclock(which_clock);
1142	struct timespec64 rtn_tp;
1143	int error;
1144
1145	if (!kc)
1146		return -EINVAL;
1147
1148	error = kc->clock_getres(which_clock, &rtn_tp);
1149
1150	if (!error && tp && put_timespec64(&rtn_tp, tp))
1151		error = -EFAULT;
1152
1153	return error;
1154}
1155
1156#ifdef CONFIG_COMPAT_32BIT_TIME
1157
1158SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1159		struct old_timespec32 __user *, tp)
1160{
1161	const struct k_clock *kc = clockid_to_kclock(which_clock);
1162	struct timespec64 ts;
1163
1164	if (!kc || !kc->clock_set)
1165		return -EINVAL;
1166
1167	if (get_old_timespec32(&ts, tp))
1168		return -EFAULT;
1169
1170	return kc->clock_set(which_clock, &ts);
1171}
1172
1173SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1174		struct old_timespec32 __user *, tp)
1175{
1176	const struct k_clock *kc = clockid_to_kclock(which_clock);
1177	struct timespec64 ts;
1178	int err;
1179
1180	if (!kc)
1181		return -EINVAL;
1182
1183	err = kc->clock_get_timespec(which_clock, &ts);
1184
1185	if (!err && put_old_timespec32(&ts, tp))
1186		err = -EFAULT;
1187
1188	return err;
1189}
1190
1191SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1192		struct old_timex32 __user *, utp)
1193{
1194	struct __kernel_timex ktx;
1195	int err;
1196
1197	err = get_old_timex32(&ktx, utp);
1198	if (err)
1199		return err;
1200
1201	err = do_clock_adjtime(which_clock, &ktx);
1202
1203	if (err >= 0 && put_old_timex32(utp, &ktx))
1204		return -EFAULT;
1205
1206	return err;
1207}
1208
1209SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1210		struct old_timespec32 __user *, tp)
1211{
1212	const struct k_clock *kc = clockid_to_kclock(which_clock);
1213	struct timespec64 ts;
1214	int err;
1215
1216	if (!kc)
1217		return -EINVAL;
1218
1219	err = kc->clock_getres(which_clock, &ts);
1220	if (!err && tp && put_old_timespec32(&ts, tp))
1221		return -EFAULT;
1222
1223	return err;
1224}
1225
1226#endif
1227
1228/*
1229 * nanosleep for monotonic and realtime clocks
1230 */
1231static int common_nsleep(const clockid_t which_clock, int flags,
1232			 const struct timespec64 *rqtp)
1233{
1234	ktime_t texp = timespec64_to_ktime(*rqtp);
1235
1236	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1237				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1238				 which_clock);
1239}
1240
 
 
 
 
 
1241static int common_nsleep_timens(const clockid_t which_clock, int flags,
1242			 const struct timespec64 *rqtp)
1243{
1244	ktime_t texp = timespec64_to_ktime(*rqtp);
1245
1246	if (flags & TIMER_ABSTIME)
1247		texp = timens_ktime_to_host(which_clock, texp);
1248
1249	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1250				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1251				 which_clock);
1252}
1253
1254SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1255		const struct __kernel_timespec __user *, rqtp,
1256		struct __kernel_timespec __user *, rmtp)
1257{
1258	const struct k_clock *kc = clockid_to_kclock(which_clock);
1259	struct timespec64 t;
1260
1261	if (!kc)
1262		return -EINVAL;
1263	if (!kc->nsleep)
1264		return -EOPNOTSUPP;
1265
1266	if (get_timespec64(&t, rqtp))
1267		return -EFAULT;
1268
1269	if (!timespec64_valid(&t))
1270		return -EINVAL;
1271	if (flags & TIMER_ABSTIME)
1272		rmtp = NULL;
 
1273	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1274	current->restart_block.nanosleep.rmtp = rmtp;
1275
1276	return kc->nsleep(which_clock, flags, &t);
1277}
1278
1279#ifdef CONFIG_COMPAT_32BIT_TIME
1280
1281SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1282		struct old_timespec32 __user *, rqtp,
1283		struct old_timespec32 __user *, rmtp)
1284{
1285	const struct k_clock *kc = clockid_to_kclock(which_clock);
1286	struct timespec64 t;
1287
1288	if (!kc)
1289		return -EINVAL;
1290	if (!kc->nsleep)
1291		return -EOPNOTSUPP;
1292
1293	if (get_old_timespec32(&t, rqtp))
1294		return -EFAULT;
1295
1296	if (!timespec64_valid(&t))
1297		return -EINVAL;
1298	if (flags & TIMER_ABSTIME)
1299		rmtp = NULL;
 
1300	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1301	current->restart_block.nanosleep.compat_rmtp = rmtp;
1302
1303	return kc->nsleep(which_clock, flags, &t);
1304}
1305
1306#endif
1307
1308static const struct k_clock clock_realtime = {
1309	.clock_getres		= posix_get_hrtimer_res,
1310	.clock_get_timespec	= posix_get_realtime_timespec,
1311	.clock_get_ktime	= posix_get_realtime_ktime,
1312	.clock_set		= posix_clock_realtime_set,
1313	.clock_adj		= posix_clock_realtime_adj,
1314	.nsleep			= common_nsleep,
1315	.timer_create		= common_timer_create,
1316	.timer_set		= common_timer_set,
1317	.timer_get		= common_timer_get,
1318	.timer_del		= common_timer_del,
1319	.timer_rearm		= common_hrtimer_rearm,
1320	.timer_forward		= common_hrtimer_forward,
1321	.timer_remaining	= common_hrtimer_remaining,
1322	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1323	.timer_wait_running	= common_timer_wait_running,
1324	.timer_arm		= common_hrtimer_arm,
1325};
1326
1327static const struct k_clock clock_monotonic = {
1328	.clock_getres		= posix_get_hrtimer_res,
1329	.clock_get_timespec	= posix_get_monotonic_timespec,
1330	.clock_get_ktime	= posix_get_monotonic_ktime,
1331	.nsleep			= common_nsleep_timens,
1332	.timer_create		= common_timer_create,
1333	.timer_set		= common_timer_set,
1334	.timer_get		= common_timer_get,
1335	.timer_del		= common_timer_del,
1336	.timer_rearm		= common_hrtimer_rearm,
1337	.timer_forward		= common_hrtimer_forward,
1338	.timer_remaining	= common_hrtimer_remaining,
1339	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1340	.timer_wait_running	= common_timer_wait_running,
1341	.timer_arm		= common_hrtimer_arm,
1342};
1343
1344static const struct k_clock clock_monotonic_raw = {
1345	.clock_getres		= posix_get_hrtimer_res,
1346	.clock_get_timespec	= posix_get_monotonic_raw,
1347};
1348
1349static const struct k_clock clock_realtime_coarse = {
1350	.clock_getres		= posix_get_coarse_res,
1351	.clock_get_timespec	= posix_get_realtime_coarse,
1352};
1353
1354static const struct k_clock clock_monotonic_coarse = {
1355	.clock_getres		= posix_get_coarse_res,
1356	.clock_get_timespec	= posix_get_monotonic_coarse,
1357};
1358
1359static const struct k_clock clock_tai = {
1360	.clock_getres		= posix_get_hrtimer_res,
1361	.clock_get_ktime	= posix_get_tai_ktime,
1362	.clock_get_timespec	= posix_get_tai_timespec,
1363	.nsleep			= common_nsleep,
1364	.timer_create		= common_timer_create,
1365	.timer_set		= common_timer_set,
1366	.timer_get		= common_timer_get,
1367	.timer_del		= common_timer_del,
1368	.timer_rearm		= common_hrtimer_rearm,
1369	.timer_forward		= common_hrtimer_forward,
1370	.timer_remaining	= common_hrtimer_remaining,
1371	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1372	.timer_wait_running	= common_timer_wait_running,
1373	.timer_arm		= common_hrtimer_arm,
1374};
1375
1376static const struct k_clock clock_boottime = {
1377	.clock_getres		= posix_get_hrtimer_res,
1378	.clock_get_ktime	= posix_get_boottime_ktime,
1379	.clock_get_timespec	= posix_get_boottime_timespec,
1380	.nsleep			= common_nsleep_timens,
1381	.timer_create		= common_timer_create,
1382	.timer_set		= common_timer_set,
1383	.timer_get		= common_timer_get,
1384	.timer_del		= common_timer_del,
1385	.timer_rearm		= common_hrtimer_rearm,
1386	.timer_forward		= common_hrtimer_forward,
1387	.timer_remaining	= common_hrtimer_remaining,
1388	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1389	.timer_wait_running	= common_timer_wait_running,
1390	.timer_arm		= common_hrtimer_arm,
1391};
1392
1393static const struct k_clock * const posix_clocks[] = {
1394	[CLOCK_REALTIME]		= &clock_realtime,
1395	[CLOCK_MONOTONIC]		= &clock_monotonic,
1396	[CLOCK_PROCESS_CPUTIME_ID]	= &clock_process,
1397	[CLOCK_THREAD_CPUTIME_ID]	= &clock_thread,
1398	[CLOCK_MONOTONIC_RAW]		= &clock_monotonic_raw,
1399	[CLOCK_REALTIME_COARSE]		= &clock_realtime_coarse,
1400	[CLOCK_MONOTONIC_COARSE]	= &clock_monotonic_coarse,
1401	[CLOCK_BOOTTIME]		= &clock_boottime,
1402	[CLOCK_REALTIME_ALARM]		= &alarm_clock,
1403	[CLOCK_BOOTTIME_ALARM]		= &alarm_clock,
1404	[CLOCK_TAI]			= &clock_tai,
1405};
1406
1407static const struct k_clock *clockid_to_kclock(const clockid_t id)
1408{
1409	clockid_t idx = id;
1410
1411	if (id < 0) {
1412		return (id & CLOCKFD_MASK) == CLOCKFD ?
1413			&clock_posix_dynamic : &clock_posix_cpu;
1414	}
1415
1416	if (id >= ARRAY_SIZE(posix_clocks))
1417		return NULL;
1418
1419	return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1420}
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0+
   2/*
   3 * 2002-10-15  Posix Clocks & timers
   4 *                           by George Anzinger george@mvista.com
   5 *			     Copyright (C) 2002 2003 by MontaVista Software.
   6 *
   7 * 2004-06-01  Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug.
   8 *			     Copyright (C) 2004 Boris Hu
   9 *
  10 * These are all the functions necessary to implement POSIX clocks & timers
  11 */
  12#include <linux/mm.h>
  13#include <linux/interrupt.h>
  14#include <linux/slab.h>
  15#include <linux/time.h>
  16#include <linux/mutex.h>
  17#include <linux/sched/task.h>
  18
  19#include <linux/uaccess.h>
  20#include <linux/list.h>
  21#include <linux/init.h>
  22#include <linux/compiler.h>
  23#include <linux/hash.h>
  24#include <linux/posix-clock.h>
  25#include <linux/posix-timers.h>
  26#include <linux/syscalls.h>
  27#include <linux/wait.h>
  28#include <linux/workqueue.h>
  29#include <linux/export.h>
  30#include <linux/hashtable.h>
  31#include <linux/compat.h>
  32#include <linux/nospec.h>
  33#include <linux/time_namespace.h>
  34
  35#include "timekeeping.h"
  36#include "posix-timers.h"
  37
  38static struct kmem_cache *posix_timers_cache;
 
 
 
 
 
 
 
  39
  40/*
  41 * Timers are managed in a hash table for lockless lookup. The hash key is
  42 * constructed from current::signal and the timer ID and the timer is
  43 * matched against current::signal and the timer ID when walking the hash
  44 * bucket list.
  45 *
  46 * This allows checkpoint/restore to reconstruct the exact timer IDs for
  47 * a process.
  48 */
 
 
  49static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
  50static DEFINE_SPINLOCK(hash_lock);
  51
  52static const struct k_clock * const posix_clocks[];
  53static const struct k_clock *clockid_to_kclock(const clockid_t id);
  54static const struct k_clock clock_realtime, clock_monotonic;
  55
  56/* SIGEV_THREAD_ID cannot share a bit with the other SIGEV values. */
 
 
 
  57#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
  58			~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
  59#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
  60#endif
  61
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  62static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
  63
  64#define lock_timer(tid, flags)						   \
  65({	struct k_itimer *__timr;					   \
  66	__cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags));  \
  67	__timr;								   \
  68})
  69
  70static int hash(struct signal_struct *sig, unsigned int nr)
  71{
  72	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
  73}
  74
  75static struct k_itimer *__posix_timers_find(struct hlist_head *head,
  76					    struct signal_struct *sig,
  77					    timer_t id)
  78{
  79	struct k_itimer *timer;
  80
  81	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) {
  82		/* timer->it_signal can be set concurrently */
  83		if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id))
  84			return timer;
  85	}
  86	return NULL;
  87}
  88
  89static struct k_itimer *posix_timer_by_id(timer_t id)
  90{
  91	struct signal_struct *sig = current->signal;
  92	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
  93
  94	return __posix_timers_find(head, sig, id);
  95}
  96
  97static int posix_timer_add(struct k_itimer *timer)
  98{
  99	struct signal_struct *sig = current->signal;
 
 100	struct hlist_head *head;
 101	unsigned int cnt, id;
 102
 103	/*
 104	 * FIXME: Replace this by a per signal struct xarray once there is
 105	 * a plan to handle the resulting CRIU regression gracefully.
 106	 */
 107	for (cnt = 0; cnt <= INT_MAX; cnt++) {
 108		spin_lock(&hash_lock);
 109		id = sig->next_posix_timer_id;
 110
 111		/* Write the next ID back. Clamp it to the positive space */
 112		sig->next_posix_timer_id = (id + 1) & INT_MAX;
 113
 114		head = &posix_timers_hashtable[hash(sig, id)];
 115		if (!__posix_timers_find(head, sig, id)) {
 116			hlist_add_head_rcu(&timer->t_hash, head);
 117			spin_unlock(&hash_lock);
 118			return id;
 119		}
 
 
 
 
 
 120		spin_unlock(&hash_lock);
 121	}
 122	/* POSIX return code when no timer ID could be allocated */
 123	return -EAGAIN;
 124}
 125
 126static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
 127{
 128	spin_unlock_irqrestore(&timr->it_lock, flags);
 129}
 130
 
 131static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp)
 132{
 133	ktime_get_real_ts64(tp);
 134	return 0;
 135}
 136
 137static ktime_t posix_get_realtime_ktime(clockid_t which_clock)
 138{
 139	return ktime_get_real();
 140}
 141
 
 142static int posix_clock_realtime_set(const clockid_t which_clock,
 143				    const struct timespec64 *tp)
 144{
 145	return do_sys_settimeofday64(tp, NULL);
 146}
 147
 148static int posix_clock_realtime_adj(const clockid_t which_clock,
 149				    struct __kernel_timex *t)
 150{
 151	return do_adjtimex(t);
 152}
 153
 
 
 
 154static int posix_get_monotonic_timespec(clockid_t which_clock, struct timespec64 *tp)
 155{
 156	ktime_get_ts64(tp);
 157	timens_add_monotonic(tp);
 158	return 0;
 159}
 160
 161static ktime_t posix_get_monotonic_ktime(clockid_t which_clock)
 162{
 163	return ktime_get();
 164}
 165
 
 
 
 166static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
 167{
 168	ktime_get_raw_ts64(tp);
 169	timens_add_monotonic(tp);
 170	return 0;
 171}
 172
 
 173static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
 174{
 175	ktime_get_coarse_real_ts64(tp);
 176	return 0;
 177}
 178
 179static int posix_get_monotonic_coarse(clockid_t which_clock,
 180						struct timespec64 *tp)
 181{
 182	ktime_get_coarse_ts64(tp);
 183	timens_add_monotonic(tp);
 184	return 0;
 185}
 186
 187static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
 188{
 189	*tp = ktime_to_timespec64(KTIME_LOW_RES);
 190	return 0;
 191}
 192
 193static int posix_get_boottime_timespec(const clockid_t which_clock, struct timespec64 *tp)
 194{
 195	ktime_get_boottime_ts64(tp);
 196	timens_add_boottime(tp);
 197	return 0;
 198}
 199
 200static ktime_t posix_get_boottime_ktime(const clockid_t which_clock)
 201{
 202	return ktime_get_boottime();
 203}
 204
 205static int posix_get_tai_timespec(clockid_t which_clock, struct timespec64 *tp)
 206{
 207	ktime_get_clocktai_ts64(tp);
 208	return 0;
 209}
 210
 211static ktime_t posix_get_tai_ktime(clockid_t which_clock)
 212{
 213	return ktime_get_clocktai();
 214}
 215
 216static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
 217{
 218	tp->tv_sec = 0;
 219	tp->tv_nsec = hrtimer_resolution;
 220	return 0;
 221}
 222
 
 
 
 223static __init int init_posix_timers(void)
 224{
 225	posix_timers_cache = kmem_cache_create("posix_timers_cache",
 226					sizeof(struct k_itimer), 0,
 227					SLAB_PANIC | SLAB_ACCOUNT, NULL);
 228	return 0;
 229}
 230__initcall(init_posix_timers);
 231
 232/*
 233 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
 234 * are of type int. Clamp the overrun value to INT_MAX
 235 */
 236static inline int timer_overrun_to_int(struct k_itimer *timr)
 237{
 238	if (timr->it_overrun_last > (s64)INT_MAX)
 239		return INT_MAX;
 240
 241	return (int)timr->it_overrun_last;
 242}
 243
 244static void common_hrtimer_rearm(struct k_itimer *timr)
 245{
 246	struct hrtimer *timer = &timr->it.real.timer;
 247
 248	timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
 249					    timr->it_interval);
 250	hrtimer_restart(timer);
 251}
 252
 253static bool __posixtimer_deliver_signal(struct kernel_siginfo *info, struct k_itimer *timr)
 254{
 255	guard(spinlock)(&timr->it_lock);
 256
 257	/*
 258	 * Check if the timer is still alive or whether it got modified
 259	 * since the signal was queued. In either case, don't rearm and
 260	 * drop the signal.
 261	 */
 262	if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal))
 263		return false;
 264
 265	if (!timr->it_interval || WARN_ON_ONCE(timr->it_status != POSIX_TIMER_REQUEUE_PENDING))
 266		return true;
 267
 268	timr->kclock->timer_rearm(timr);
 269	timr->it_status = POSIX_TIMER_ARMED;
 270	timr->it_overrun_last = timr->it_overrun;
 271	timr->it_overrun = -1LL;
 272	++timr->it_signal_seq;
 273	info->si_overrun = timer_overrun_to_int(timr);
 274	return true;
 275}
 276
 277/*
 278 * This function is called from the signal delivery code. It decides
 279 * whether the signal should be dropped and rearms interval timers.  The
 280 * timer can be unconditionally accessed as there is a reference held on
 281 * it.
 
 
 
 
 
 282 */
 283bool posixtimer_deliver_signal(struct kernel_siginfo *info, struct sigqueue *timer_sigq)
 284{
 285	struct k_itimer *timr = container_of(timer_sigq, struct k_itimer, sigq);
 286	bool ret;
 
 
 
 
 287
 288	/*
 289	 * Release siglock to ensure proper locking order versus
 290	 * timr::it_lock. Keep interrupts disabled.
 291	 */
 292	spin_unlock(&current->sighand->siglock);
 293
 294	ret = __posixtimer_deliver_signal(info, timr);
 
 
 
 295
 296	/* Drop the reference which was acquired when the signal was queued */
 297	posixtimer_putref(timr);
 298
 299	spin_lock(&current->sighand->siglock);
 300	return ret;
 301}
 302
 303void posix_timer_queue_signal(struct k_itimer *timr)
 304{
 305	lockdep_assert_held(&timr->it_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 306
 307	timr->it_status = timr->it_interval ? POSIX_TIMER_REQUEUE_PENDING : POSIX_TIMER_DISARMED;
 308	posixtimer_send_sigqueue(timr);
 
 
 309}
 310
 311/*
 312 * This function gets called when a POSIX.1b interval timer expires from
 313 * the HRTIMER interrupt (soft interrupt on RT kernels).
 314 *
 315 * Handles CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME and CLOCK_TAI
 316 * based timers.
 317 */
 318static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
 319{
 320	struct k_itimer *timr = container_of(timer, struct k_itimer, it.real.timer);
 
 
 
 
 
 
 321
 322	guard(spinlock_irqsave)(&timr->it_lock);
 323	posix_timer_queue_signal(timr);
 324	return HRTIMER_NORESTART;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 325}
 326
 327static struct pid *good_sigevent(sigevent_t * event)
 328{
 329	struct pid *pid = task_tgid(current);
 330	struct task_struct *rtn;
 331
 332	switch (event->sigev_notify) {
 333	case SIGEV_SIGNAL | SIGEV_THREAD_ID:
 334		pid = find_vpid(event->sigev_notify_thread_id);
 335		rtn = pid_task(pid, PIDTYPE_PID);
 336		if (!rtn || !same_thread_group(rtn, current))
 337			return NULL;
 338		fallthrough;
 339	case SIGEV_SIGNAL:
 340	case SIGEV_THREAD:
 341		if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
 342			return NULL;
 343		fallthrough;
 344	case SIGEV_NONE:
 345		return pid;
 346	default:
 347		return NULL;
 348	}
 349}
 350
 351static struct k_itimer *alloc_posix_timer(void)
 352{
 353	struct k_itimer *tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
 354
 355	if (!tmr)
 356		return tmr;
 357
 358	if (unlikely(!posixtimer_init_sigqueue(&tmr->sigq))) {
 359		kmem_cache_free(posix_timers_cache, tmr);
 360		return NULL;
 361	}
 362	rcuref_init(&tmr->rcuref, 1);
 363	return tmr;
 364}
 365
 366void posixtimer_free_timer(struct k_itimer *tmr)
 367{
 368	put_pid(tmr->it_pid);
 369	if (tmr->sigq.ucounts)
 370		dec_rlimit_put_ucounts(tmr->sigq.ucounts, UCOUNT_RLIMIT_SIGPENDING);
 371	kfree_rcu(tmr, rcu);
 372}
 373
 374static void posix_timer_unhash_and_free(struct k_itimer *tmr)
 
 
 375{
 376	spin_lock(&hash_lock);
 377	hlist_del_rcu(&tmr->t_hash);
 378	spin_unlock(&hash_lock);
 379	posixtimer_putref(tmr);
 
 
 
 
 
 380}
 381
 382static int common_timer_create(struct k_itimer *new_timer)
 383{
 384	hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
 385	return 0;
 386}
 387
 388/* Create a POSIX.1b interval timer. */
 389static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 390			   timer_t __user *created_timer_id)
 391{
 392	const struct k_clock *kc = clockid_to_kclock(which_clock);
 393	struct k_itimer *new_timer;
 394	int error, new_timer_id;
 
 395
 396	if (!kc)
 397		return -EINVAL;
 398	if (!kc->timer_create)
 399		return -EOPNOTSUPP;
 400
 401	new_timer = alloc_posix_timer();
 402	if (unlikely(!new_timer))
 403		return -EAGAIN;
 404
 405	spin_lock_init(&new_timer->it_lock);
 406
 407	/*
 408	 * Add the timer to the hash table. The timer is not yet valid
 409	 * because new_timer::it_signal is still NULL. The timer id is also
 410	 * not yet visible to user space.
 411	 */
 412	new_timer_id = posix_timer_add(new_timer);
 413	if (new_timer_id < 0) {
 414		posixtimer_free_timer(new_timer);
 415		return new_timer_id;
 416	}
 417
 
 418	new_timer->it_id = (timer_t) new_timer_id;
 419	new_timer->it_clock = which_clock;
 420	new_timer->kclock = kc;
 421	new_timer->it_overrun = -1LL;
 422
 423	if (event) {
 424		rcu_read_lock();
 425		new_timer->it_pid = get_pid(good_sigevent(event));
 426		rcu_read_unlock();
 427		if (!new_timer->it_pid) {
 428			error = -EINVAL;
 429			goto out;
 430		}
 431		new_timer->it_sigev_notify     = event->sigev_notify;
 432		new_timer->sigq.info.si_signo = event->sigev_signo;
 433		new_timer->sigq.info.si_value = event->sigev_value;
 434	} else {
 435		new_timer->it_sigev_notify     = SIGEV_SIGNAL;
 436		new_timer->sigq.info.si_signo = SIGALRM;
 437		memset(&new_timer->sigq.info.si_value, 0, sizeof(sigval_t));
 438		new_timer->sigq.info.si_value.sival_int = new_timer->it_id;
 439		new_timer->it_pid = get_pid(task_tgid(current));
 440	}
 441
 442	if (new_timer->it_sigev_notify & SIGEV_THREAD_ID)
 443		new_timer->it_pid_type = PIDTYPE_PID;
 444	else
 445		new_timer->it_pid_type = PIDTYPE_TGID;
 446
 447	new_timer->sigq.info.si_tid = new_timer->it_id;
 448	new_timer->sigq.info.si_code = SI_TIMER;
 449
 450	if (copy_to_user(created_timer_id, &new_timer_id, sizeof (new_timer_id))) {
 
 451		error = -EFAULT;
 452		goto out;
 453	}
 454	/*
 455	 * After succesful copy out, the timer ID is visible to user space
 456	 * now but not yet valid because new_timer::signal is still NULL.
 457	 *
 458	 * Complete the initialization with the clock specific create
 459	 * callback.
 460	 */
 461	error = kc->timer_create(new_timer);
 462	if (error)
 463		goto out;
 464
 465	spin_lock_irq(&current->sighand->siglock);
 466	/* This makes the timer valid in the hash table */
 467	WRITE_ONCE(new_timer->it_signal, current->signal);
 468	hlist_add_head(&new_timer->list, &current->signal->posix_timers);
 469	spin_unlock_irq(&current->sighand->siglock);
 
 
 470	/*
 471	 * After unlocking sighand::siglock @new_timer is subject to
 472	 * concurrent removal and cannot be touched anymore
 
 
 473	 */
 474	return 0;
 475out:
 476	posix_timer_unhash_and_free(new_timer);
 477	return error;
 478}
 479
 480SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
 481		struct sigevent __user *, timer_event_spec,
 482		timer_t __user *, created_timer_id)
 483{
 484	if (timer_event_spec) {
 485		sigevent_t event;
 486
 487		if (copy_from_user(&event, timer_event_spec, sizeof (event)))
 488			return -EFAULT;
 489		return do_timer_create(which_clock, &event, created_timer_id);
 490	}
 491	return do_timer_create(which_clock, NULL, created_timer_id);
 492}
 493
 494#ifdef CONFIG_COMPAT
 495COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
 496		       struct compat_sigevent __user *, timer_event_spec,
 497		       timer_t __user *, created_timer_id)
 498{
 499	if (timer_event_spec) {
 500		sigevent_t event;
 501
 502		if (get_compat_sigevent(&event, timer_event_spec))
 503			return -EFAULT;
 504		return do_timer_create(which_clock, &event, created_timer_id);
 505	}
 506	return do_timer_create(which_clock, NULL, created_timer_id);
 507}
 508#endif
 509
 
 
 
 
 
 
 
 510static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 511{
 512	struct k_itimer *timr;
 513
 514	/*
 515	 * timer_t could be any type >= int and we want to make sure any
 516	 * @timer_id outside positive int range fails lookup.
 517	 */
 518	if ((unsigned long long)timer_id > INT_MAX)
 519		return NULL;
 520
 521	/*
 522	 * The hash lookup and the timers are RCU protected.
 523	 *
 524	 * Timers are added to the hash in invalid state where
 525	 * timr::it_signal == NULL. timer::it_signal is only set after the
 526	 * rest of the initialization succeeded.
 527	 *
 528	 * Timer destruction happens in steps:
 529	 *  1) Set timr::it_signal to NULL with timr::it_lock held
 530	 *  2) Release timr::it_lock
 531	 *  3) Remove from the hash under hash_lock
 532	 *  4) Put the reference count.
 533	 *
 534	 * The reference count might not drop to zero if timr::sigq is
 535	 * queued. In that case the signal delivery or flush will put the
 536	 * last reference count.
 537	 *
 538	 * When the reference count reaches zero, the timer is scheduled
 539	 * for RCU removal after the grace period.
 540	 *
 541	 * Holding rcu_read_lock() accross the lookup ensures that
 542	 * the timer cannot be freed.
 543	 *
 544	 * The lookup validates locklessly that timr::it_signal ==
 545	 * current::it_signal and timr::it_id == @timer_id. timr::it_id
 546	 * can't change, but timr::it_signal becomes NULL during
 547	 * destruction.
 548	 */
 549	rcu_read_lock();
 550	timr = posix_timer_by_id(timer_id);
 551	if (timr) {
 552		spin_lock_irqsave(&timr->it_lock, *flags);
 553		/*
 554		 * Validate under timr::it_lock that timr::it_signal is
 555		 * still valid. Pairs with #1 above.
 556		 */
 557		if (timr->it_signal == current->signal) {
 558			rcu_read_unlock();
 559			return timr;
 560		}
 561		spin_unlock_irqrestore(&timr->it_lock, *flags);
 562	}
 563	rcu_read_unlock();
 564
 565	return NULL;
 566}
 567
 568static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
 569{
 570	struct hrtimer *timer = &timr->it.real.timer;
 571
 572	return __hrtimer_expires_remaining_adjusted(timer, now);
 573}
 574
 575static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
 576{
 577	struct hrtimer *timer = &timr->it.real.timer;
 578
 579	return hrtimer_forward(timer, now, timr->it_interval);
 580}
 581
 582/*
 583 * Get the time remaining on a POSIX.1b interval timer.
 584 *
 585 * Two issues to handle here:
 586 *
 587 *  1) The timer has a requeue pending. The return value must appear as
 588 *     if the timer has been requeued right now.
 589 *
 590 *  2) The timer is a SIGEV_NONE timer. These timers are never enqueued
 591 *     into the hrtimer queue and therefore never expired. Emulate expiry
 592 *     here taking #1 into account.
 
 
 
 
 593 */
 594void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
 595{
 596	const struct k_clock *kc = timr->kclock;
 597	ktime_t now, remaining, iv;
 598	bool sig_none;
 599
 600	sig_none = timr->it_sigev_notify == SIGEV_NONE;
 601	iv = timr->it_interval;
 602
 603	/* interval timer ? */
 604	if (iv) {
 605		cur_setting->it_interval = ktime_to_timespec64(iv);
 606	} else if (timr->it_status == POSIX_TIMER_DISARMED) {
 607		/*
 608		 * SIGEV_NONE oneshot timers are never queued and therefore
 609		 * timr->it_status is always DISARMED. The check below
 610		 * vs. remaining time will handle this case.
 611		 *
 612		 * For all other timers there is nothing to update here, so
 613		 * return.
 614		 */
 615		if (!sig_none)
 616			return;
 617	}
 618
 619	now = kc->clock_get_ktime(timr->it_clock);
 620
 621	/*
 622	 * If this is an interval timer and either has requeue pending or
 623	 * is a SIGEV_NONE timer move the expiry time forward by intervals,
 624	 * so expiry is > now.
 625	 */
 626	if (iv && timr->it_status != POSIX_TIMER_ARMED)
 627		timr->it_overrun += kc->timer_forward(timr, now);
 628
 629	remaining = kc->timer_remaining(timr, now);
 630	/*
 631	 * As @now is retrieved before a possible timer_forward() and
 632	 * cannot be reevaluated by the compiler @remaining is based on the
 633	 * same @now value. Therefore @remaining is consistent vs. @now.
 634	 *
 635	 * Consequently all interval timers, i.e. @iv > 0, cannot have a
 636	 * remaining time <= 0 because timer_forward() guarantees to move
 637	 * them forward so that the next timer expiry is > @now.
 638	 */
 639	if (remaining <= 0) {
 640		/*
 641		 * A single shot SIGEV_NONE timer must return 0, when it is
 642		 * expired! Timers which have a real signal delivery mode
 643		 * must return a remaining time greater than 0 because the
 644		 * signal has not yet been delivered.
 645		 */
 646		if (!sig_none)
 647			cur_setting->it_value.tv_nsec = 1;
 648	} else {
 649		cur_setting->it_value = ktime_to_timespec64(remaining);
 650	}
 651}
 652
 
 653static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 654{
 
 655	const struct k_clock *kc;
 656	struct k_itimer *timr;
 657	unsigned long flags;
 658	int ret = 0;
 659
 660	timr = lock_timer(timer_id, &flags);
 661	if (!timr)
 662		return -EINVAL;
 663
 664	memset(setting, 0, sizeof(*setting));
 665	kc = timr->kclock;
 666	if (WARN_ON_ONCE(!kc || !kc->timer_get))
 667		ret = -EINVAL;
 668	else
 669		kc->timer_get(timr, setting);
 670
 671	unlock_timer(timr, flags);
 672	return ret;
 673}
 674
 675/* Get the time remaining on a POSIX.1b interval timer. */
 676SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
 677		struct __kernel_itimerspec __user *, setting)
 678{
 679	struct itimerspec64 cur_setting;
 680
 681	int ret = do_timer_gettime(timer_id, &cur_setting);
 682	if (!ret) {
 683		if (put_itimerspec64(&cur_setting, setting))
 684			ret = -EFAULT;
 685	}
 686	return ret;
 687}
 688
 689#ifdef CONFIG_COMPAT_32BIT_TIME
 690
 691SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
 692		struct old_itimerspec32 __user *, setting)
 693{
 694	struct itimerspec64 cur_setting;
 695
 696	int ret = do_timer_gettime(timer_id, &cur_setting);
 697	if (!ret) {
 698		if (put_old_itimerspec32(&cur_setting, setting))
 699			ret = -EFAULT;
 700	}
 701	return ret;
 702}
 703
 704#endif
 705
 706/**
 707 * sys_timer_getoverrun - Get the number of overruns of a POSIX.1b interval timer
 708 * @timer_id:	The timer ID which identifies the timer
 709 *
 710 * The "overrun count" of a timer is one plus the number of expiration
 711 * intervals which have elapsed between the first expiry, which queues the
 712 * signal and the actual signal delivery. On signal delivery the "overrun
 713 * count" is calculated and cached, so it can be returned directly here.
 714 *
 715 * As this is relative to the last queued signal the returned overrun count
 716 * is meaningless outside of the signal delivery path and even there it
 717 * does not accurately reflect the current state when user space evaluates
 718 * it.
 719 *
 720 * Returns:
 721 *	-EINVAL		@timer_id is invalid
 722 *	1..INT_MAX	The number of overruns related to the last delivered signal
 723 */
 724SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 725{
 726	struct k_itimer *timr;
 
 727	unsigned long flags;
 728	int overrun;
 729
 730	timr = lock_timer(timer_id, &flags);
 731	if (!timr)
 732		return -EINVAL;
 733
 734	overrun = timer_overrun_to_int(timr);
 735	unlock_timer(timr, flags);
 736
 737	return overrun;
 738}
 739
 740static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
 741			       bool absolute, bool sigev_none)
 742{
 743	struct hrtimer *timer = &timr->it.real.timer;
 744	enum hrtimer_mode mode;
 745
 746	mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
 747	/*
 748	 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
 749	 * clock modifications, so they become CLOCK_MONOTONIC based under the
 750	 * hood. See hrtimer_init(). Update timr->kclock, so the generic
 751	 * functions which use timr->kclock->clock_get_*() work.
 752	 *
 753	 * Note: it_clock stays unmodified, because the next timer_set() might
 754	 * use ABSTIME, so it needs to switch back.
 755	 */
 756	if (timr->it_clock == CLOCK_REALTIME)
 757		timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
 758
 759	hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
 760	timr->it.real.timer.function = posix_timer_fn;
 761
 762	if (!absolute)
 763		expires = ktime_add_safe(expires, timer->base->get_time());
 764	hrtimer_set_expires(timer, expires);
 765
 766	if (!sigev_none)
 767		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 768}
 769
 770static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
 771{
 772	return hrtimer_try_to_cancel(&timr->it.real.timer);
 773}
 774
 775static void common_timer_wait_running(struct k_itimer *timer)
 776{
 777	hrtimer_cancel_wait_running(&timer->it.real.timer);
 778}
 779
 780/*
 781 * On PREEMPT_RT this prevents priority inversion and a potential livelock
 782 * against the ksoftirqd thread in case that ksoftirqd gets preempted while
 783 * executing a hrtimer callback.
 784 *
 785 * See the comments in hrtimer_cancel_wait_running(). For PREEMPT_RT=n this
 786 * just results in a cpu_relax().
 787 *
 788 * For POSIX CPU timers with CONFIG_POSIX_CPU_TIMERS_TASK_WORK=n this is
 789 * just a cpu_relax(). With CONFIG_POSIX_CPU_TIMERS_TASK_WORK=y this
 790 * prevents spinning on an eventually scheduled out task and a livelock
 791 * when the task which tries to delete or disarm the timer has preempted
 792 * the task which runs the expiry in task work context.
 793 */
 794static struct k_itimer *timer_wait_running(struct k_itimer *timer,
 795					   unsigned long *flags)
 796{
 797	const struct k_clock *kc = READ_ONCE(timer->kclock);
 798	timer_t timer_id = READ_ONCE(timer->it_id);
 799
 800	/* Prevent kfree(timer) after dropping the lock */
 801	rcu_read_lock();
 802	unlock_timer(timer, *flags);
 803
 804	/*
 805	 * kc->timer_wait_running() might drop RCU lock. So @timer
 806	 * cannot be touched anymore after the function returns!
 807	 */
 808	if (!WARN_ON_ONCE(!kc->timer_wait_running))
 809		kc->timer_wait_running(timer);
 810
 811	rcu_read_unlock();
 812	/* Relock the timer. It might be not longer hashed. */
 813	return lock_timer(timer_id, flags);
 814}
 815
 816/*
 817 * Set up the new interval and reset the signal delivery data
 818 */
 819void posix_timer_set_common(struct k_itimer *timer, struct itimerspec64 *new_setting)
 820{
 821	if (new_setting->it_value.tv_sec || new_setting->it_value.tv_nsec)
 822		timer->it_interval = timespec64_to_ktime(new_setting->it_interval);
 823	else
 824		timer->it_interval = 0;
 825
 826	/* Reset overrun accounting */
 827	timer->it_overrun_last = 0;
 828	timer->it_overrun = -1LL;
 829}
 830
 831/* Set a POSIX.1b interval timer. */
 832int common_timer_set(struct k_itimer *timr, int flags,
 833		     struct itimerspec64 *new_setting,
 834		     struct itimerspec64 *old_setting)
 835{
 836	const struct k_clock *kc = timr->kclock;
 837	bool sigev_none;
 838	ktime_t expires;
 839
 840	if (old_setting)
 841		common_timer_get(timr, old_setting);
 842
 
 
 843	/*
 844	 * Careful here. On SMP systems the timer expiry function could be
 845	 * active and spinning on timr->it_lock.
 846	 */
 847	if (kc->timer_try_to_cancel(timr) < 0)
 848		return TIMER_RETRY;
 849
 850	timr->it_status = POSIX_TIMER_DISARMED;
 851	posix_timer_set_common(timr, new_setting);
 
 
 852
 853	/* Keep timer disarmed when it_value is zero */
 854	if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
 855		return 0;
 856
 
 857	expires = timespec64_to_ktime(new_setting->it_value);
 858	if (flags & TIMER_ABSTIME)
 859		expires = timens_ktime_to_host(timr->it_clock, expires);
 860	sigev_none = timr->it_sigev_notify == SIGEV_NONE;
 861
 862	kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
 863	if (!sigev_none)
 864		timr->it_status = POSIX_TIMER_ARMED;
 865	return 0;
 866}
 867
 868static int do_timer_settime(timer_t timer_id, int tmr_flags,
 869			    struct itimerspec64 *new_spec64,
 870			    struct itimerspec64 *old_spec64)
 871{
 872	const struct k_clock *kc;
 873	struct k_itimer *timr;
 874	unsigned long flags;
 875	int error;
 876
 877	if (!timespec64_valid(&new_spec64->it_interval) ||
 878	    !timespec64_valid(&new_spec64->it_value))
 879		return -EINVAL;
 880
 881	if (old_spec64)
 882		memset(old_spec64, 0, sizeof(*old_spec64));
 883
 884	timr = lock_timer(timer_id, &flags);
 885retry:
 886	if (!timr)
 887		return -EINVAL;
 888
 889	if (old_spec64)
 890		old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
 891
 892	/* Prevent signal delivery and rearming. */
 893	timr->it_signal_seq++;
 894
 895	kc = timr->kclock;
 896	if (WARN_ON_ONCE(!kc || !kc->timer_set))
 897		error = -EINVAL;
 898	else
 899		error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
 900
 901	if (error == TIMER_RETRY) {
 902		// We already got the old time...
 903		old_spec64 = NULL;
 904		/* Unlocks and relocks the timer if it still exists */
 905		timr = timer_wait_running(timr, &flags);
 906		goto retry;
 907	}
 908	unlock_timer(timr, flags);
 909
 910	return error;
 911}
 912
 913/* Set a POSIX.1b interval timer */
 914SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
 915		const struct __kernel_itimerspec __user *, new_setting,
 916		struct __kernel_itimerspec __user *, old_setting)
 917{
 918	struct itimerspec64 new_spec, old_spec, *rtn;
 
 919	int error = 0;
 920
 921	if (!new_setting)
 922		return -EINVAL;
 923
 924	if (get_itimerspec64(&new_spec, new_setting))
 925		return -EFAULT;
 926
 927	rtn = old_setting ? &old_spec : NULL;
 928	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 929	if (!error && old_setting) {
 930		if (put_itimerspec64(&old_spec, old_setting))
 931			error = -EFAULT;
 932	}
 933	return error;
 934}
 935
 936#ifdef CONFIG_COMPAT_32BIT_TIME
 937SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
 938		struct old_itimerspec32 __user *, new,
 939		struct old_itimerspec32 __user *, old)
 940{
 941	struct itimerspec64 new_spec, old_spec;
 942	struct itimerspec64 *rtn = old ? &old_spec : NULL;
 943	int error = 0;
 944
 945	if (!new)
 946		return -EINVAL;
 947	if (get_old_itimerspec32(&new_spec, new))
 948		return -EFAULT;
 949
 950	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 951	if (!error && old) {
 952		if (put_old_itimerspec32(&old_spec, old))
 953			error = -EFAULT;
 954	}
 955	return error;
 956}
 957#endif
 958
 959int common_timer_del(struct k_itimer *timer)
 960{
 961	const struct k_clock *kc = timer->kclock;
 962
 
 963	if (kc->timer_try_to_cancel(timer) < 0)
 964		return TIMER_RETRY;
 965	timer->it_status = POSIX_TIMER_DISARMED;
 966	return 0;
 967}
 968
 969/*
 970 * If the deleted timer is on the ignored list, remove it and
 971 * drop the associated reference.
 972 */
 973static inline void posix_timer_cleanup_ignored(struct k_itimer *tmr)
 974{
 975	if (!hlist_unhashed(&tmr->ignored_list)) {
 976		hlist_del_init(&tmr->ignored_list);
 977		posixtimer_putref(tmr);
 978	}
 979}
 980
 981static inline int timer_delete_hook(struct k_itimer *timer)
 982{
 983	const struct k_clock *kc = timer->kclock;
 984
 985	/* Prevent signal delivery and rearming. */
 986	timer->it_signal_seq++;
 987
 988	if (WARN_ON_ONCE(!kc || !kc->timer_del))
 989		return -EINVAL;
 990	return kc->timer_del(timer);
 991}
 992
 993/* Delete a POSIX.1b interval timer. */
 994SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 995{
 996	struct k_itimer *timer;
 997	unsigned long flags;
 998
 999	timer = lock_timer(timer_id, &flags);
1000
1001retry_delete:
1002	if (!timer)
1003		return -EINVAL;
1004
1005	if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
1006		/* Unlocks and relocks the timer if it still exists */
1007		timer = timer_wait_running(timer, &flags);
1008		goto retry_delete;
1009	}
1010
1011	spin_lock(&current->sighand->siglock);
1012	hlist_del(&timer->list);
1013	posix_timer_cleanup_ignored(timer);
1014	/*
1015	 * A concurrent lookup could check timer::it_signal lockless. It
1016	 * will reevaluate with timer::it_lock held and observe the NULL.
1017	 *
1018	 * It must be written with siglock held so that the signal code
1019	 * observes timer->it_signal == NULL in do_sigaction(SIG_IGN),
1020	 * which prevents it from moving a pending signal of a deleted
1021	 * timer to the ignore list.
1022	 */
1023	WRITE_ONCE(timer->it_signal, NULL);
1024	spin_unlock(&current->sighand->siglock);
1025
1026	unlock_timer(timer, flags);
1027	posix_timer_unhash_and_free(timer);
1028	return 0;
1029}
1030
1031/*
1032 * Delete a timer if it is armed, remove it from the hash and schedule it
1033 * for RCU freeing.
1034 */
1035static void itimer_delete(struct k_itimer *timer)
1036{
1037	unsigned long flags;
1038
1039	/*
1040	 * irqsave is required to make timer_wait_running() work.
1041	 */
1042	spin_lock_irqsave(&timer->it_lock, flags);
1043
1044retry_delete:
1045	/*
1046	 * Even if the timer is not longer accessible from other tasks
1047	 * it still might be armed and queued in the underlying timer
1048	 * mechanism. Worse, that timer mechanism might run the expiry
1049	 * function concurrently.
1050	 */
1051	if (timer_delete_hook(timer) == TIMER_RETRY) {
1052		/*
1053		 * Timer is expired concurrently, prevent livelocks
1054		 * and pointless spinning on RT.
1055		 *
1056		 * timer_wait_running() drops timer::it_lock, which opens
1057		 * the possibility for another task to delete the timer.
1058		 *
1059		 * That's not possible here because this is invoked from
1060		 * do_exit() only for the last thread of the thread group.
1061		 * So no other task can access and delete that timer.
1062		 */
1063		if (WARN_ON_ONCE(timer_wait_running(timer, &flags) != timer))
1064			return;
1065
1066		goto retry_delete;
1067	}
1068	hlist_del(&timer->list);
1069
1070	posix_timer_cleanup_ignored(timer);
1071
1072	/*
1073	 * Setting timer::it_signal to NULL is technically not required
1074	 * here as nothing can access the timer anymore legitimately via
1075	 * the hash table. Set it to NULL nevertheless so that all deletion
1076	 * paths are consistent.
1077	 */
1078	WRITE_ONCE(timer->it_signal, NULL);
1079
1080	spin_unlock_irqrestore(&timer->it_lock, flags);
1081	posix_timer_unhash_and_free(timer);
1082}
1083
1084/*
1085 * Invoked from do_exit() when the last thread of a thread group exits.
1086 * At that point no other task can access the timers of the dying
1087 * task anymore.
1088 */
1089void exit_itimers(struct task_struct *tsk)
1090{
1091	struct hlist_head timers;
 
1092
1093	if (hlist_empty(&tsk->signal->posix_timers))
1094		return;
1095
1096	/* Protect against concurrent read via /proc/$PID/timers */
1097	spin_lock_irq(&tsk->sighand->siglock);
1098	hlist_move_list(&tsk->signal->posix_timers, &timers);
1099	spin_unlock_irq(&tsk->sighand->siglock);
1100
1101	/* The timers are not longer accessible via tsk::signal */
1102	while (!hlist_empty(&timers))
1103		itimer_delete(hlist_entry(timers.first, struct k_itimer, list));
1104
1105	/*
1106	 * There should be no timers on the ignored list. itimer_delete() has
1107	 * mopped them up.
1108	 */
1109	if (!WARN_ON_ONCE(!hlist_empty(&tsk->signal->ignored_posix_timers)))
1110		return;
1111
1112	hlist_move_list(&tsk->signal->ignored_posix_timers, &timers);
1113	while (!hlist_empty(&timers)) {
1114		posix_timer_cleanup_ignored(hlist_entry(timers.first, struct k_itimer,
1115							ignored_list));
1116	}
1117}
1118
1119SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1120		const struct __kernel_timespec __user *, tp)
1121{
1122	const struct k_clock *kc = clockid_to_kclock(which_clock);
1123	struct timespec64 new_tp;
1124
1125	if (!kc || !kc->clock_set)
1126		return -EINVAL;
1127
1128	if (get_timespec64(&new_tp, tp))
1129		return -EFAULT;
1130
1131	/*
1132	 * Permission checks have to be done inside the clock specific
1133	 * setter callback.
1134	 */
1135	return kc->clock_set(which_clock, &new_tp);
1136}
1137
1138SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1139		struct __kernel_timespec __user *, tp)
1140{
1141	const struct k_clock *kc = clockid_to_kclock(which_clock);
1142	struct timespec64 kernel_tp;
1143	int error;
1144
1145	if (!kc)
1146		return -EINVAL;
1147
1148	error = kc->clock_get_timespec(which_clock, &kernel_tp);
1149
1150	if (!error && put_timespec64(&kernel_tp, tp))
1151		error = -EFAULT;
1152
1153	return error;
1154}
1155
1156int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1157{
1158	const struct k_clock *kc = clockid_to_kclock(which_clock);
1159
1160	if (!kc)
1161		return -EINVAL;
1162	if (!kc->clock_adj)
1163		return -EOPNOTSUPP;
1164
1165	return kc->clock_adj(which_clock, ktx);
1166}
1167
1168SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1169		struct __kernel_timex __user *, utx)
1170{
1171	struct __kernel_timex ktx;
1172	int err;
1173
1174	if (copy_from_user(&ktx, utx, sizeof(ktx)))
1175		return -EFAULT;
1176
1177	err = do_clock_adjtime(which_clock, &ktx);
1178
1179	if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1180		return -EFAULT;
1181
1182	return err;
1183}
1184
1185/**
1186 * sys_clock_getres - Get the resolution of a clock
1187 * @which_clock:	The clock to get the resolution for
1188 * @tp:			Pointer to a a user space timespec64 for storage
1189 *
1190 * POSIX defines:
1191 *
1192 * "The clock_getres() function shall return the resolution of any
1193 * clock. Clock resolutions are implementation-defined and cannot be set by
1194 * a process. If the argument res is not NULL, the resolution of the
1195 * specified clock shall be stored in the location pointed to by res. If
1196 * res is NULL, the clock resolution is not returned. If the time argument
1197 * of clock_settime() is not a multiple of res, then the value is truncated
1198 * to a multiple of res."
1199 *
1200 * Due to the various hardware constraints the real resolution can vary
1201 * wildly and even change during runtime when the underlying devices are
1202 * replaced. The kernel also can use hardware devices with different
1203 * resolutions for reading the time and for arming timers.
1204 *
1205 * The kernel therefore deviates from the POSIX spec in various aspects:
1206 *
1207 * 1) The resolution returned to user space
1208 *
1209 *    For CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, CLOCK_TAI,
1210 *    CLOCK_REALTIME_ALARM, CLOCK_BOOTTIME_ALAREM and CLOCK_MONOTONIC_RAW
1211 *    the kernel differentiates only two cases:
1212 *
1213 *    I)  Low resolution mode:
1214 *
1215 *	  When high resolution timers are disabled at compile or runtime
1216 *	  the resolution returned is nanoseconds per tick, which represents
1217 *	  the precision at which timers expire.
1218 *
1219 *    II) High resolution mode:
1220 *
1221 *	  When high resolution timers are enabled the resolution returned
1222 *	  is always one nanosecond independent of the actual resolution of
1223 *	  the underlying hardware devices.
1224 *
1225 *	  For CLOCK_*_ALARM the actual resolution depends on system
1226 *	  state. When system is running the resolution is the same as the
1227 *	  resolution of the other clocks. During suspend the actual
1228 *	  resolution is the resolution of the underlying RTC device which
1229 *	  might be way less precise than the clockevent device used during
1230 *	  running state.
1231 *
1232 *   For CLOCK_REALTIME_COARSE and CLOCK_MONOTONIC_COARSE the resolution
1233 *   returned is always nanoseconds per tick.
1234 *
1235 *   For CLOCK_PROCESS_CPUTIME and CLOCK_THREAD_CPUTIME the resolution
1236 *   returned is always one nanosecond under the assumption that the
1237 *   underlying scheduler clock has a better resolution than nanoseconds
1238 *   per tick.
1239 *
1240 *   For dynamic POSIX clocks (PTP devices) the resolution returned is
1241 *   always one nanosecond.
1242 *
1243 * 2) Affect on sys_clock_settime()
1244 *
1245 *    The kernel does not truncate the time which is handed in to
1246 *    sys_clock_settime(). The kernel internal timekeeping is always using
1247 *    nanoseconds precision independent of the clocksource device which is
1248 *    used to read the time from. The resolution of that device only
1249 *    affects the presicion of the time returned by sys_clock_gettime().
1250 *
1251 * Returns:
1252 *	0		Success. @tp contains the resolution
1253 *	-EINVAL		@which_clock is not a valid clock ID
1254 *	-EFAULT		Copying the resolution to @tp faulted
1255 *	-ENODEV		Dynamic POSIX clock is not backed by a device
1256 *	-EOPNOTSUPP	Dynamic POSIX clock does not support getres()
1257 */
1258SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1259		struct __kernel_timespec __user *, tp)
1260{
1261	const struct k_clock *kc = clockid_to_kclock(which_clock);
1262	struct timespec64 rtn_tp;
1263	int error;
1264
1265	if (!kc)
1266		return -EINVAL;
1267
1268	error = kc->clock_getres(which_clock, &rtn_tp);
1269
1270	if (!error && tp && put_timespec64(&rtn_tp, tp))
1271		error = -EFAULT;
1272
1273	return error;
1274}
1275
1276#ifdef CONFIG_COMPAT_32BIT_TIME
1277
1278SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1279		struct old_timespec32 __user *, tp)
1280{
1281	const struct k_clock *kc = clockid_to_kclock(which_clock);
1282	struct timespec64 ts;
1283
1284	if (!kc || !kc->clock_set)
1285		return -EINVAL;
1286
1287	if (get_old_timespec32(&ts, tp))
1288		return -EFAULT;
1289
1290	return kc->clock_set(which_clock, &ts);
1291}
1292
1293SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1294		struct old_timespec32 __user *, tp)
1295{
1296	const struct k_clock *kc = clockid_to_kclock(which_clock);
1297	struct timespec64 ts;
1298	int err;
1299
1300	if (!kc)
1301		return -EINVAL;
1302
1303	err = kc->clock_get_timespec(which_clock, &ts);
1304
1305	if (!err && put_old_timespec32(&ts, tp))
1306		err = -EFAULT;
1307
1308	return err;
1309}
1310
1311SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1312		struct old_timex32 __user *, utp)
1313{
1314	struct __kernel_timex ktx;
1315	int err;
1316
1317	err = get_old_timex32(&ktx, utp);
1318	if (err)
1319		return err;
1320
1321	err = do_clock_adjtime(which_clock, &ktx);
1322
1323	if (err >= 0 && put_old_timex32(utp, &ktx))
1324		return -EFAULT;
1325
1326	return err;
1327}
1328
1329SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1330		struct old_timespec32 __user *, tp)
1331{
1332	const struct k_clock *kc = clockid_to_kclock(which_clock);
1333	struct timespec64 ts;
1334	int err;
1335
1336	if (!kc)
1337		return -EINVAL;
1338
1339	err = kc->clock_getres(which_clock, &ts);
1340	if (!err && tp && put_old_timespec32(&ts, tp))
1341		return -EFAULT;
1342
1343	return err;
1344}
1345
1346#endif
1347
1348/*
1349 * sys_clock_nanosleep() for CLOCK_REALTIME and CLOCK_TAI
1350 */
1351static int common_nsleep(const clockid_t which_clock, int flags,
1352			 const struct timespec64 *rqtp)
1353{
1354	ktime_t texp = timespec64_to_ktime(*rqtp);
1355
1356	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1357				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1358				 which_clock);
1359}
1360
1361/*
1362 * sys_clock_nanosleep() for CLOCK_MONOTONIC and CLOCK_BOOTTIME
1363 *
1364 * Absolute nanosleeps for these clocks are time-namespace adjusted.
1365 */
1366static int common_nsleep_timens(const clockid_t which_clock, int flags,
1367				const struct timespec64 *rqtp)
1368{
1369	ktime_t texp = timespec64_to_ktime(*rqtp);
1370
1371	if (flags & TIMER_ABSTIME)
1372		texp = timens_ktime_to_host(which_clock, texp);
1373
1374	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1375				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1376				 which_clock);
1377}
1378
1379SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1380		const struct __kernel_timespec __user *, rqtp,
1381		struct __kernel_timespec __user *, rmtp)
1382{
1383	const struct k_clock *kc = clockid_to_kclock(which_clock);
1384	struct timespec64 t;
1385
1386	if (!kc)
1387		return -EINVAL;
1388	if (!kc->nsleep)
1389		return -EOPNOTSUPP;
1390
1391	if (get_timespec64(&t, rqtp))
1392		return -EFAULT;
1393
1394	if (!timespec64_valid(&t))
1395		return -EINVAL;
1396	if (flags & TIMER_ABSTIME)
1397		rmtp = NULL;
1398	current->restart_block.fn = do_no_restart_syscall;
1399	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1400	current->restart_block.nanosleep.rmtp = rmtp;
1401
1402	return kc->nsleep(which_clock, flags, &t);
1403}
1404
1405#ifdef CONFIG_COMPAT_32BIT_TIME
1406
1407SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1408		struct old_timespec32 __user *, rqtp,
1409		struct old_timespec32 __user *, rmtp)
1410{
1411	const struct k_clock *kc = clockid_to_kclock(which_clock);
1412	struct timespec64 t;
1413
1414	if (!kc)
1415		return -EINVAL;
1416	if (!kc->nsleep)
1417		return -EOPNOTSUPP;
1418
1419	if (get_old_timespec32(&t, rqtp))
1420		return -EFAULT;
1421
1422	if (!timespec64_valid(&t))
1423		return -EINVAL;
1424	if (flags & TIMER_ABSTIME)
1425		rmtp = NULL;
1426	current->restart_block.fn = do_no_restart_syscall;
1427	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1428	current->restart_block.nanosleep.compat_rmtp = rmtp;
1429
1430	return kc->nsleep(which_clock, flags, &t);
1431}
1432
1433#endif
1434
1435static const struct k_clock clock_realtime = {
1436	.clock_getres		= posix_get_hrtimer_res,
1437	.clock_get_timespec	= posix_get_realtime_timespec,
1438	.clock_get_ktime	= posix_get_realtime_ktime,
1439	.clock_set		= posix_clock_realtime_set,
1440	.clock_adj		= posix_clock_realtime_adj,
1441	.nsleep			= common_nsleep,
1442	.timer_create		= common_timer_create,
1443	.timer_set		= common_timer_set,
1444	.timer_get		= common_timer_get,
1445	.timer_del		= common_timer_del,
1446	.timer_rearm		= common_hrtimer_rearm,
1447	.timer_forward		= common_hrtimer_forward,
1448	.timer_remaining	= common_hrtimer_remaining,
1449	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1450	.timer_wait_running	= common_timer_wait_running,
1451	.timer_arm		= common_hrtimer_arm,
1452};
1453
1454static const struct k_clock clock_monotonic = {
1455	.clock_getres		= posix_get_hrtimer_res,
1456	.clock_get_timespec	= posix_get_monotonic_timespec,
1457	.clock_get_ktime	= posix_get_monotonic_ktime,
1458	.nsleep			= common_nsleep_timens,
1459	.timer_create		= common_timer_create,
1460	.timer_set		= common_timer_set,
1461	.timer_get		= common_timer_get,
1462	.timer_del		= common_timer_del,
1463	.timer_rearm		= common_hrtimer_rearm,
1464	.timer_forward		= common_hrtimer_forward,
1465	.timer_remaining	= common_hrtimer_remaining,
1466	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1467	.timer_wait_running	= common_timer_wait_running,
1468	.timer_arm		= common_hrtimer_arm,
1469};
1470
1471static const struct k_clock clock_monotonic_raw = {
1472	.clock_getres		= posix_get_hrtimer_res,
1473	.clock_get_timespec	= posix_get_monotonic_raw,
1474};
1475
1476static const struct k_clock clock_realtime_coarse = {
1477	.clock_getres		= posix_get_coarse_res,
1478	.clock_get_timespec	= posix_get_realtime_coarse,
1479};
1480
1481static const struct k_clock clock_monotonic_coarse = {
1482	.clock_getres		= posix_get_coarse_res,
1483	.clock_get_timespec	= posix_get_monotonic_coarse,
1484};
1485
1486static const struct k_clock clock_tai = {
1487	.clock_getres		= posix_get_hrtimer_res,
1488	.clock_get_ktime	= posix_get_tai_ktime,
1489	.clock_get_timespec	= posix_get_tai_timespec,
1490	.nsleep			= common_nsleep,
1491	.timer_create		= common_timer_create,
1492	.timer_set		= common_timer_set,
1493	.timer_get		= common_timer_get,
1494	.timer_del		= common_timer_del,
1495	.timer_rearm		= common_hrtimer_rearm,
1496	.timer_forward		= common_hrtimer_forward,
1497	.timer_remaining	= common_hrtimer_remaining,
1498	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1499	.timer_wait_running	= common_timer_wait_running,
1500	.timer_arm		= common_hrtimer_arm,
1501};
1502
1503static const struct k_clock clock_boottime = {
1504	.clock_getres		= posix_get_hrtimer_res,
1505	.clock_get_ktime	= posix_get_boottime_ktime,
1506	.clock_get_timespec	= posix_get_boottime_timespec,
1507	.nsleep			= common_nsleep_timens,
1508	.timer_create		= common_timer_create,
1509	.timer_set		= common_timer_set,
1510	.timer_get		= common_timer_get,
1511	.timer_del		= common_timer_del,
1512	.timer_rearm		= common_hrtimer_rearm,
1513	.timer_forward		= common_hrtimer_forward,
1514	.timer_remaining	= common_hrtimer_remaining,
1515	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1516	.timer_wait_running	= common_timer_wait_running,
1517	.timer_arm		= common_hrtimer_arm,
1518};
1519
1520static const struct k_clock * const posix_clocks[] = {
1521	[CLOCK_REALTIME]		= &clock_realtime,
1522	[CLOCK_MONOTONIC]		= &clock_monotonic,
1523	[CLOCK_PROCESS_CPUTIME_ID]	= &clock_process,
1524	[CLOCK_THREAD_CPUTIME_ID]	= &clock_thread,
1525	[CLOCK_MONOTONIC_RAW]		= &clock_monotonic_raw,
1526	[CLOCK_REALTIME_COARSE]		= &clock_realtime_coarse,
1527	[CLOCK_MONOTONIC_COARSE]	= &clock_monotonic_coarse,
1528	[CLOCK_BOOTTIME]		= &clock_boottime,
1529	[CLOCK_REALTIME_ALARM]		= &alarm_clock,
1530	[CLOCK_BOOTTIME_ALARM]		= &alarm_clock,
1531	[CLOCK_TAI]			= &clock_tai,
1532};
1533
1534static const struct k_clock *clockid_to_kclock(const clockid_t id)
1535{
1536	clockid_t idx = id;
1537
1538	if (id < 0) {
1539		return (id & CLOCKFD_MASK) == CLOCKFD ?
1540			&clock_posix_dynamic : &clock_posix_cpu;
1541	}
1542
1543	if (id >= ARRAY_SIZE(posix_clocks))
1544		return NULL;
1545
1546	return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1547}