posix-timers.c

   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/compat.h>
  13#include <linux/compiler.h>
  14#include <linux/init.h>
  15#include <linux/jhash.h>
  16#include <linux/interrupt.h>
  17#include <linux/list.h>
  18#include <linux/memblock.h>
  19#include <linux/nospec.h>
  20#include <linux/posix-clock.h>
  21#include <linux/posix-timers.h>
  22#include <linux/prctl.h>
  23#include <linux/sched/task.h>
  24#include <linux/slab.h>
  25#include <linux/syscalls.h>
  26#include <linux/time.h>
  27#include <linux/time_namespace.h>
  28#include <linux/uaccess.h>
  29
  30#include "timekeeping.h"
  31#include "posix-timers.h"
  32
  33/*
  34 * Timers are managed in a hash table for lockless lookup. The hash key is
  35 * constructed from current::signal and the timer ID and the timer is
  36 * matched against current::signal and the timer ID when walking the hash
  37 * bucket list.
  38 *
  39 * This allows checkpoint/restore to reconstruct the exact timer IDs for
  40 * a process.
  41 */
  42struct timer_hash_bucket {
  43	spinlock_t		lock;
  44	struct hlist_head	head;
  45};
  46
  47static struct {
  48	struct timer_hash_bucket	*buckets;
  49	unsigned long			mask;
  50	struct kmem_cache		*cache;
  51} __timer_data __ro_after_init __aligned(4*sizeof(long));
  52
  53#define timer_buckets		(__timer_data.buckets)
  54#define timer_hashmask		(__timer_data.mask)
  55#define posix_timers_cache	(__timer_data.cache)
  56
  57static const struct k_clock * const posix_clocks[];
  58static const struct k_clock *clockid_to_kclock(const clockid_t id);
  59static const struct k_clock clock_realtime, clock_monotonic;
  60
  61#define TIMER_ANY_ID		INT_MIN
  62
  63/* SIGEV_THREAD_ID cannot share a bit with the other SIGEV values. */
  64#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
  65			~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
  66#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
  67#endif
  68
  69static struct k_itimer *__lock_timer(timer_t timer_id);
  70
  71#define lock_timer(tid)							\
  72({	struct k_itimer *__timr;					\
  73	__cond_lock(&__timr->it_lock, __timr = __lock_timer(tid));	\
  74	__timr;								\
  75})
  76
  77static inline void unlock_timer(struct k_itimer *timr)
  78{
  79	if (likely((timr)))
  80		spin_unlock_irq(&timr->it_lock);
  81}
  82
  83#define scoped_timer_get_or_fail(_id)					\
  84	scoped_cond_guard(lock_timer, return -EINVAL, _id)
  85
  86#define scoped_timer				(scope)
  87
  88DEFINE_CLASS(lock_timer, struct k_itimer *, unlock_timer(_T), __lock_timer(id), timer_t id);
  89DEFINE_CLASS_IS_COND_GUARD(lock_timer);
  90
  91static struct timer_hash_bucket *hash_bucket(struct signal_struct *sig, unsigned int nr)
  92{
  93	return &timer_buckets[jhash2((u32 *)&sig, sizeof(sig) / sizeof(u32), nr) & timer_hashmask];
  94}
  95
  96static struct k_itimer *posix_timer_by_id(timer_t id)
  97{
  98	struct signal_struct *sig = current->signal;
  99	struct timer_hash_bucket *bucket = hash_bucket(sig, id);
 100	struct k_itimer *timer;
 101
 102	hlist_for_each_entry_rcu(timer, &bucket->head, t_hash) {
 103		/* timer->it_signal can be set concurrently */
 104		if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id))
 105			return timer;
 106	}
 107	return NULL;
 108}
 109
 110static inline struct signal_struct *posix_sig_owner(const struct k_itimer *timer)
 111{
 112	unsigned long val = (unsigned long)timer->it_signal;
 113
 114	/*
 115	 * Mask out bit 0, which acts as invalid marker to prevent
 116	 * posix_timer_by_id() detecting it as valid.
 117	 */
 118	return (struct signal_struct *)(val & ~1UL);
 119}
 120
 121static bool posix_timer_hashed(struct timer_hash_bucket *bucket, struct signal_struct *sig,
 122			       timer_t id)
 123{
 124	struct hlist_head *head = &bucket->head;
 125	struct k_itimer *timer;
 126
 127	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&bucket->lock)) {
 128		if ((posix_sig_owner(timer) == sig) && (timer->it_id == id))
 129			return true;
 130	}
 131	return false;
 132}
 133
 134static bool posix_timer_add_at(struct k_itimer *timer, struct signal_struct *sig, unsigned int id)
 135{
 136	struct timer_hash_bucket *bucket = hash_bucket(sig, id);
 137
 138	scoped_guard (spinlock, &bucket->lock) {
 139		/*
 140		 * Validate under the lock as this could have raced against
 141		 * another thread ending up with the same ID, which is
 142		 * highly unlikely, but possible.
 143		 */
 144		if (!posix_timer_hashed(bucket, sig, id)) {
 145			/*
 146			 * Set the timer ID and the signal pointer to make
 147			 * it identifiable in the hash table. The signal
 148			 * pointer has bit 0 set to indicate that it is not
 149			 * yet fully initialized. posix_timer_hashed()
 150			 * masks this bit out, but the syscall lookup fails
 151			 * to match due to it being set. This guarantees
 152			 * that there can't be duplicate timer IDs handed
 153			 * out.
 154			 */
 155			timer->it_id = (timer_t)id;
 156			timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
 157			hlist_add_head_rcu(&timer->t_hash, &bucket->head);
 158			return true;
 159		}
 160	}
 161	return false;
 162}
 163
 164static int posix_timer_add(struct k_itimer *timer, int req_id)
 165{
 166	struct signal_struct *sig = current->signal;
 167
 168	if (unlikely(req_id != TIMER_ANY_ID)) {
 169		if (!posix_timer_add_at(timer, sig, req_id))
 170			return -EBUSY;
 171
 172		/*
 173		 * Move the ID counter past the requested ID, so that after
 174		 * switching back to normal mode the IDs are outside of the
 175		 * exact allocated region. That avoids ID collisions on the
 176		 * next regular timer_create() invocations.
 177		 */
 178		atomic_set(&sig->next_posix_timer_id, req_id + 1);
 179		return req_id;
 180	}
 181
 182	for (unsigned int cnt = 0; cnt <= INT_MAX; cnt++) {
 183		/* Get the next timer ID and clamp it to positive space */
 184		unsigned int id = atomic_fetch_inc(&sig->next_posix_timer_id) & INT_MAX;
 185
 186		if (posix_timer_add_at(timer, sig, id))
 187			return id;
 188		cond_resched();
 189	}
 190	/* POSIX return code when no timer ID could be allocated */
 191	return -EAGAIN;
 192}
 193
 194static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp)
 195{
 196	ktime_get_real_ts64(tp);
 197	return 0;
 198}
 199
 200static ktime_t posix_get_realtime_ktime(clockid_t which_clock)
 201{
 202	return ktime_get_real();
 203}
 204
 205static int posix_clock_realtime_set(const clockid_t which_clock,
 206				    const struct timespec64 *tp)
 207{
 208	return do_sys_settimeofday64(tp, NULL);
 209}
 210
 211static int posix_clock_realtime_adj(const clockid_t which_clock,
 212				    struct __kernel_timex *t)
 213{
 214	return do_adjtimex(t);
 215}
 216
 217static int posix_get_monotonic_timespec(clockid_t which_clock, struct timespec64 *tp)
 218{
 219	ktime_get_ts64(tp);
 220	timens_add_monotonic(tp);
 221	return 0;
 222}
 223
 224static ktime_t posix_get_monotonic_ktime(clockid_t which_clock)
 225{
 226	return ktime_get();
 227}
 228
 229static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec64 *tp)
 230{
 231	ktime_get_raw_ts64(tp);
 232	timens_add_monotonic(tp);
 233	return 0;
 234}
 235
 236static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec64 *tp)
 237{
 238	ktime_get_coarse_real_ts64(tp);
 239	return 0;
 240}
 241
 242static int posix_get_monotonic_coarse(clockid_t which_clock,
 243						struct timespec64 *tp)
 244{
 245	ktime_get_coarse_ts64(tp);
 246	timens_add_monotonic(tp);
 247	return 0;
 248}
 249
 250static int posix_get_coarse_res(const clockid_t which_clock, struct timespec64 *tp)
 251{
 252	*tp = ktime_to_timespec64(KTIME_LOW_RES);
 253	return 0;
 254}
 255
 256static int posix_get_boottime_timespec(const clockid_t which_clock, struct timespec64 *tp)
 257{
 258	ktime_get_boottime_ts64(tp);
 259	timens_add_boottime(tp);
 260	return 0;
 261}
 262
 263static ktime_t posix_get_boottime_ktime(const clockid_t which_clock)
 264{
 265	return ktime_get_boottime();
 266}
 267
 268static int posix_get_tai_timespec(clockid_t which_clock, struct timespec64 *tp)
 269{
 270	ktime_get_clocktai_ts64(tp);
 271	return 0;
 272}
 273
 274static ktime_t posix_get_tai_ktime(clockid_t which_clock)
 275{
 276	return ktime_get_clocktai();
 277}
 278
 279static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec64 *tp)
 280{
 281	tp->tv_sec = 0;
 282	tp->tv_nsec = hrtimer_resolution;
 283	return 0;
 284}
 285
 286/*
 287 * The siginfo si_overrun field and the return value of timer_getoverrun(2)
 288 * are of type int. Clamp the overrun value to INT_MAX
 289 */
 290static inline int timer_overrun_to_int(struct k_itimer *timr)
 291{
 292	if (timr->it_overrun_last > (s64)INT_MAX)
 293		return INT_MAX;
 294
 295	return (int)timr->it_overrun_last;
 296}
 297
 298static void common_hrtimer_rearm(struct k_itimer *timr)
 299{
 300	struct hrtimer *timer = &timr->it.real.timer;
 301
 302	timr->it_overrun += hrtimer_forward_now(timer, timr->it_interval);
 303	hrtimer_restart(timer);
 304}
 305
 306static bool __posixtimer_deliver_signal(struct kernel_siginfo *info, struct k_itimer *timr)
 307{
 308	guard(spinlock)(&timr->it_lock);
 309
 310	/*
 311	 * Check if the timer is still alive or whether it got modified
 312	 * since the signal was queued. In either case, don't rearm and
 313	 * drop the signal.
 314	 */
 315	if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!posixtimer_valid(timr)))
 316		return false;
 317
 318	if (!timr->it_interval || WARN_ON_ONCE(timr->it_status != POSIX_TIMER_REQUEUE_PENDING))
 319		return true;
 320
 321	timr->kclock->timer_rearm(timr);
 322	timr->it_status = POSIX_TIMER_ARMED;
 323	timr->it_overrun_last = timr->it_overrun;
 324	timr->it_overrun = -1LL;
 325	++timr->it_signal_seq;
 326	info->si_overrun = timer_overrun_to_int(timr);
 327	return true;
 328}
 329
 330/*
 331 * This function is called from the signal delivery code. It decides
 332 * whether the signal should be dropped and rearms interval timers.  The
 333 * timer can be unconditionally accessed as there is a reference held on
 334 * it.
 335 */
 336bool posixtimer_deliver_signal(struct kernel_siginfo *info, struct sigqueue *timer_sigq)
 337{
 338	struct k_itimer *timr = container_of(timer_sigq, struct k_itimer, sigq);
 339	bool ret;
 340
 341	/*
 342	 * Release siglock to ensure proper locking order versus
 343	 * timr::it_lock. Keep interrupts disabled.
 344	 */
 345	spin_unlock(&current->sighand->siglock);
 346
 347	ret = __posixtimer_deliver_signal(info, timr);
 348
 349	/* Drop the reference which was acquired when the signal was queued */
 350	posixtimer_putref(timr);
 351
 352	spin_lock(&current->sighand->siglock);
 353	return ret;
 354}
 355
 356void posix_timer_queue_signal(struct k_itimer *timr)
 357{
 358	lockdep_assert_held(&timr->it_lock);
 359
 360	if (!posixtimer_valid(timr))
 361		return;
 362
 363	timr->it_status = timr->it_interval ? POSIX_TIMER_REQUEUE_PENDING : POSIX_TIMER_DISARMED;
 364	posixtimer_send_sigqueue(timr);
 365}
 366
 367/*
 368 * This function gets called when a POSIX.1b interval timer expires from
 369 * the HRTIMER interrupt (soft interrupt on RT kernels).
 370 *
 371 * Handles CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME and CLOCK_TAI
 372 * based timers.
 373 */
 374static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
 375{
 376	struct k_itimer *timr = container_of(timer, struct k_itimer, it.real.timer);
 377
 378	guard(spinlock_irqsave)(&timr->it_lock);
 379	posix_timer_queue_signal(timr);
 380	return HRTIMER_NORESTART;
 381}
 382
 383long posixtimer_create_prctl(unsigned long ctrl)
 384{
 385	switch (ctrl) {
 386	case PR_TIMER_CREATE_RESTORE_IDS_OFF:
 387		current->signal->timer_create_restore_ids = 0;
 388		return 0;
 389	case PR_TIMER_CREATE_RESTORE_IDS_ON:
 390		current->signal->timer_create_restore_ids = 1;
 391		return 0;
 392	case PR_TIMER_CREATE_RESTORE_IDS_GET:
 393		return current->signal->timer_create_restore_ids;
 394	}
 395	return -EINVAL;
 396}
 397
 398static struct pid *good_sigevent(sigevent_t * event)
 399{
 400	struct pid *pid = task_tgid(current);
 401	struct task_struct *rtn;
 402
 403	switch (event->sigev_notify) {
 404	case SIGEV_SIGNAL | SIGEV_THREAD_ID:
 405		pid = find_vpid(event->sigev_notify_thread_id);
 406		rtn = pid_task(pid, PIDTYPE_PID);
 407		if (!rtn || !same_thread_group(rtn, current))
 408			return NULL;
 409		fallthrough;
 410	case SIGEV_SIGNAL:
 411	case SIGEV_THREAD:
 412		if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
 413			return NULL;
 414		fallthrough;
 415	case SIGEV_NONE:
 416		return pid;
 417	default:
 418		return NULL;
 419	}
 420}
 421
 422static struct k_itimer *alloc_posix_timer(void)
 423{
 424	struct k_itimer *tmr;
 425
 426	if (unlikely(!posix_timers_cache))
 427		return NULL;
 428
 429	tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
 430	if (!tmr)
 431		return tmr;
 432
 433	if (unlikely(!posixtimer_init_sigqueue(&tmr->sigq))) {
 434		kmem_cache_free(posix_timers_cache, tmr);
 435		return NULL;
 436	}
 437	rcuref_init(&tmr->rcuref, 1);
 438	return tmr;
 439}
 440
 441void posixtimer_free_timer(struct k_itimer *tmr)
 442{
 443	put_pid(tmr->it_pid);
 444	if (tmr->sigq.ucounts)
 445		dec_rlimit_put_ucounts(tmr->sigq.ucounts, UCOUNT_RLIMIT_SIGPENDING);
 446	kfree_rcu(tmr, rcu);
 447}
 448
 449static void posix_timer_unhash_and_free(struct k_itimer *tmr)
 450{
 451	struct timer_hash_bucket *bucket = hash_bucket(posix_sig_owner(tmr), tmr->it_id);
 452
 453	scoped_guard (spinlock, &bucket->lock)
 454		hlist_del_rcu(&tmr->t_hash);
 455	posixtimer_putref(tmr);
 456}
 457
 458static int common_timer_create(struct k_itimer *new_timer)
 459{
 460	hrtimer_setup(&new_timer->it.real.timer, posix_timer_fn, new_timer->it_clock, 0);
 461	return 0;
 462}
 463
 464/* Create a POSIX.1b interval timer. */
 465static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 466			   timer_t __user *created_timer_id)
 467{
 468	const struct k_clock *kc = clockid_to_kclock(which_clock);
 469	timer_t req_id = TIMER_ANY_ID;
 470	struct k_itimer *new_timer;
 471	int error, new_timer_id;
 472
 473	if (!kc)
 474		return -EINVAL;
 475	if (!kc->timer_create)
 476		return -EOPNOTSUPP;
 477
 478	/* Special case for CRIU to restore timers with a given timer ID. */
 479	if (unlikely(current->signal->timer_create_restore_ids)) {
 480		if (copy_from_user(&req_id, created_timer_id, sizeof(req_id)))
 481			return -EFAULT;
 482		/* Valid IDs are 0..INT_MAX */
 483		if ((unsigned int)req_id > INT_MAX)
 484			return -EINVAL;
 485	}
 486
 487	new_timer = alloc_posix_timer();
 488	if (unlikely(!new_timer))
 489		return -EAGAIN;
 490
 491	spin_lock_init(&new_timer->it_lock);
 492
 493	/*
 494	 * Add the timer to the hash table. The timer is not yet valid
 495	 * after insertion, but has a unique ID allocated.
 496	 */
 497	new_timer_id = posix_timer_add(new_timer, req_id);
 498	if (new_timer_id < 0) {
 499		posixtimer_free_timer(new_timer);
 500		return new_timer_id;
 501	}
 502
 503	new_timer->it_clock = which_clock;
 504	new_timer->kclock = kc;
 505	new_timer->it_overrun = -1LL;
 506
 507	if (event) {
 508		scoped_guard (rcu)
 509			new_timer->it_pid = get_pid(good_sigevent(event));
 510		if (!new_timer->it_pid) {
 511			error = -EINVAL;
 512			goto out;
 513		}
 514		new_timer->it_sigev_notify     = event->sigev_notify;
 515		new_timer->sigq.info.si_signo = event->sigev_signo;
 516		new_timer->sigq.info.si_value = event->sigev_value;
 517	} else {
 518		new_timer->it_sigev_notify     = SIGEV_SIGNAL;
 519		new_timer->sigq.info.si_signo = SIGALRM;
 520		new_timer->sigq.info.si_value.sival_int = new_timer->it_id;
 521		new_timer->it_pid = get_pid(task_tgid(current));
 522	}
 523
 524	if (new_timer->it_sigev_notify & SIGEV_THREAD_ID)
 525		new_timer->it_pid_type = PIDTYPE_PID;
 526	else
 527		new_timer->it_pid_type = PIDTYPE_TGID;
 528
 529	new_timer->sigq.info.si_tid = new_timer->it_id;
 530	new_timer->sigq.info.si_code = SI_TIMER;
 531
 532	if (copy_to_user(created_timer_id, &new_timer_id, sizeof (new_timer_id))) {
 533		error = -EFAULT;
 534		goto out;
 535	}
 536	/*
 537	 * After successful copy out, the timer ID is visible to user space
 538	 * now but not yet valid because new_timer::signal low order bit is 1.
 539	 *
 540	 * Complete the initialization with the clock specific create
 541	 * callback.
 542	 */
 543	error = kc->timer_create(new_timer);
 544	if (error)
 545		goto out;
 546
 547	/*
 548	 * timer::it_lock ensures that __lock_timer() observes a fully
 549	 * initialized timer when it observes a valid timer::it_signal.
 550	 *
 551	 * sighand::siglock is required to protect signal::posix_timers.
 552	 */
 553	scoped_guard (spinlock_irq, &new_timer->it_lock) {
 554		guard(spinlock)(&current->sighand->siglock);
 555		/*
 556		 * new_timer::it_signal contains the signal pointer with
 557		 * bit 0 set, which makes it invalid for syscall operations.
 558		 * Store the unmodified signal pointer to make it valid.
 559		 */
 560		WRITE_ONCE(new_timer->it_signal, current->signal);
 561		hlist_add_head_rcu(&new_timer->list, &current->signal->posix_timers);
 562	}
 563	/*
 564	 * After unlocking @new_timer is subject to concurrent removal and
 565	 * cannot be touched anymore
 566	 */
 567	return 0;
 568out:
 569	posix_timer_unhash_and_free(new_timer);
 570	return error;
 571}
 572
 573SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
 574		struct sigevent __user *, timer_event_spec,
 575		timer_t __user *, created_timer_id)
 576{
 577	if (timer_event_spec) {
 578		sigevent_t event;
 579
 580		if (copy_from_user(&event, timer_event_spec, sizeof (event)))
 581			return -EFAULT;
 582		return do_timer_create(which_clock, &event, created_timer_id);
 583	}
 584	return do_timer_create(which_clock, NULL, created_timer_id);
 585}
 586
 587#ifdef CONFIG_COMPAT
 588COMPAT_SYSCALL_DEFINE3(timer_create, clockid_t, which_clock,
 589		       struct compat_sigevent __user *, timer_event_spec,
 590		       timer_t __user *, created_timer_id)
 591{
 592	if (timer_event_spec) {
 593		sigevent_t event;
 594
 595		if (get_compat_sigevent(&event, timer_event_spec))
 596			return -EFAULT;
 597		return do_timer_create(which_clock, &event, created_timer_id);
 598	}
 599	return do_timer_create(which_clock, NULL, created_timer_id);
 600}
 601#endif
 602
 603static struct k_itimer *__lock_timer(timer_t timer_id)
 604{
 605	struct k_itimer *timr;
 606
 607	/*
 608	 * timer_t could be any type >= int and we want to make sure any
 609	 * @timer_id outside positive int range fails lookup.
 610	 */
 611	if ((unsigned long long)timer_id > INT_MAX)
 612		return NULL;
 613
 614	/*
 615	 * The hash lookup and the timers are RCU protected.
 616	 *
 617	 * Timers are added to the hash in invalid state where
 618	 * timr::it_signal is marked invalid. timer::it_signal is only set
 619	 * after the rest of the initialization succeeded.
 620	 *
 621	 * Timer destruction happens in steps:
 622	 *  1) Set timr::it_signal marked invalid with timr::it_lock held
 623	 *  2) Release timr::it_lock
 624	 *  3) Remove from the hash under hash_lock
 625	 *  4) Put the reference count.
 626	 *
 627	 * The reference count might not drop to zero if timr::sigq is
 628	 * queued. In that case the signal delivery or flush will put the
 629	 * last reference count.
 630	 *
 631	 * When the reference count reaches zero, the timer is scheduled
 632	 * for RCU removal after the grace period.
 633	 *
 634	 * Holding rcu_read_lock() across the lookup ensures that
 635	 * the timer cannot be freed.
 636	 *
 637	 * The lookup validates locklessly that timr::it_signal ==
 638	 * current::it_signal and timr::it_id == @timer_id. timr::it_id
 639	 * can't change, but timr::it_signal can become invalid during
 640	 * destruction, which makes the locked check fail.
 641	 */
 642	guard(rcu)();
 643	timr = posix_timer_by_id(timer_id);
 644	if (timr) {
 645		spin_lock_irq(&timr->it_lock);
 646		/*
 647		 * Validate under timr::it_lock that timr::it_signal is
 648		 * still valid. Pairs with #1 above.
 649		 */
 650		if (timr->it_signal == current->signal)
 651			return timr;
 652		spin_unlock_irq(&timr->it_lock);
 653	}
 654	return NULL;
 655}
 656
 657static ktime_t common_hrtimer_remaining(struct k_itimer *timr, ktime_t now)
 658{
 659	struct hrtimer *timer = &timr->it.real.timer;
 660
 661	return __hrtimer_expires_remaining_adjusted(timer, now);
 662}
 663
 664static s64 common_hrtimer_forward(struct k_itimer *timr, ktime_t now)
 665{
 666	struct hrtimer *timer = &timr->it.real.timer;
 667
 668	return hrtimer_forward(timer, now, timr->it_interval);
 669}
 670
 671/*
 672 * Get the time remaining on a POSIX.1b interval timer.
 673 *
 674 * Two issues to handle here:
 675 *
 676 *  1) The timer has a requeue pending. The return value must appear as
 677 *     if the timer has been requeued right now.
 678 *
 679 *  2) The timer is a SIGEV_NONE timer. These timers are never enqueued
 680 *     into the hrtimer queue and therefore never expired. Emulate expiry
 681 *     here taking #1 into account.
 682 */
 683void common_timer_get(struct k_itimer *timr, struct itimerspec64 *cur_setting)
 684{
 685	const struct k_clock *kc = timr->kclock;
 686	ktime_t now, remaining, iv;
 687	bool sig_none;
 688
 689	sig_none = timr->it_sigev_notify == SIGEV_NONE;
 690	iv = timr->it_interval;
 691
 692	/* interval timer ? */
 693	if (iv) {
 694		cur_setting->it_interval = ktime_to_timespec64(iv);
 695	} else if (timr->it_status == POSIX_TIMER_DISARMED) {
 696		/*
 697		 * SIGEV_NONE oneshot timers are never queued and therefore
 698		 * timr->it_status is always DISARMED. The check below
 699		 * vs. remaining time will handle this case.
 700		 *
 701		 * For all other timers there is nothing to update here, so
 702		 * return.
 703		 */
 704		if (!sig_none)
 705			return;
 706	}
 707
 708	now = kc->clock_get_ktime(timr->it_clock);
 709
 710	/*
 711	 * If this is an interval timer and either has requeue pending or
 712	 * is a SIGEV_NONE timer move the expiry time forward by intervals,
 713	 * so expiry is > now.
 714	 */
 715	if (iv && timr->it_status != POSIX_TIMER_ARMED)
 716		timr->it_overrun += kc->timer_forward(timr, now);
 717
 718	remaining = kc->timer_remaining(timr, now);
 719	/*
 720	 * As @now is retrieved before a possible timer_forward() and
 721	 * cannot be reevaluated by the compiler @remaining is based on the
 722	 * same @now value. Therefore @remaining is consistent vs. @now.
 723	 *
 724	 * Consequently all interval timers, i.e. @iv > 0, cannot have a
 725	 * remaining time <= 0 because timer_forward() guarantees to move
 726	 * them forward so that the next timer expiry is > @now.
 727	 */
 728	if (remaining <= 0) {
 729		/*
 730		 * A single shot SIGEV_NONE timer must return 0, when it is
 731		 * expired! Timers which have a real signal delivery mode
 732		 * must return a remaining time greater than 0 because the
 733		 * signal has not yet been delivered.
 734		 */
 735		if (!sig_none)
 736			cur_setting->it_value.tv_nsec = 1;
 737	} else {
 738		cur_setting->it_value = ktime_to_timespec64(remaining);
 739	}
 740}
 741
 742static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 743{
 744	memset(setting, 0, sizeof(*setting));
 745	scoped_timer_get_or_fail(timer_id)
 746		scoped_timer->kclock->timer_get(scoped_timer, setting);
 747	return 0;
 748}
 749
 750/* Get the time remaining on a POSIX.1b interval timer. */
 751SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
 752		struct __kernel_itimerspec __user *, setting)
 753{
 754	struct itimerspec64 cur_setting;
 755
 756	int ret = do_timer_gettime(timer_id, &cur_setting);
 757	if (!ret) {
 758		if (put_itimerspec64(&cur_setting, setting))
 759			ret = -EFAULT;
 760	}
 761	return ret;
 762}
 763
 764#ifdef CONFIG_COMPAT_32BIT_TIME
 765
 766SYSCALL_DEFINE2(timer_gettime32, timer_t, timer_id,
 767		struct old_itimerspec32 __user *, setting)
 768{
 769	struct itimerspec64 cur_setting;
 770
 771	int ret = do_timer_gettime(timer_id, &cur_setting);
 772	if (!ret) {
 773		if (put_old_itimerspec32(&cur_setting, setting))
 774			ret = -EFAULT;
 775	}
 776	return ret;
 777}
 778
 779#endif
 780
 781/**
 782 * sys_timer_getoverrun - Get the number of overruns of a POSIX.1b interval timer
 783 * @timer_id:	The timer ID which identifies the timer
 784 *
 785 * The "overrun count" of a timer is one plus the number of expiration
 786 * intervals which have elapsed between the first expiry, which queues the
 787 * signal and the actual signal delivery. On signal delivery the "overrun
 788 * count" is calculated and cached, so it can be returned directly here.
 789 *
 790 * As this is relative to the last queued signal the returned overrun count
 791 * is meaningless outside of the signal delivery path and even there it
 792 * does not accurately reflect the current state when user space evaluates
 793 * it.
 794 *
 795 * Returns:
 796 *	-EINVAL		@timer_id is invalid
 797 *	1..INT_MAX	The number of overruns related to the last delivered signal
 798 */
 799SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 800{
 801	scoped_timer_get_or_fail(timer_id)
 802		return timer_overrun_to_int(scoped_timer);
 803}
 804
 805static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
 806			       bool absolute, bool sigev_none)
 807{
 808	struct hrtimer *timer = &timr->it.real.timer;
 809	enum hrtimer_mode mode;
 810
 811	mode = absolute ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
 812	/*
 813	 * Posix magic: Relative CLOCK_REALTIME timers are not affected by
 814	 * clock modifications, so they become CLOCK_MONOTONIC based under the
 815	 * hood. See hrtimer_setup(). Update timr->kclock, so the generic
 816	 * functions which use timr->kclock->clock_get_*() work.
 817	 *
 818	 * Note: it_clock stays unmodified, because the next timer_set() might
 819	 * use ABSTIME, so it needs to switch back.
 820	 */
 821	if (timr->it_clock == CLOCK_REALTIME)
 822		timr->kclock = absolute ? &clock_realtime : &clock_monotonic;
 823
 824	hrtimer_setup(&timr->it.real.timer, posix_timer_fn, timr->it_clock, mode);
 825
 826	if (!absolute)
 827		expires = ktime_add_safe(expires, hrtimer_cb_get_time(timer));
 828	hrtimer_set_expires(timer, expires);
 829
 830	if (!sigev_none)
 831		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 832}
 833
 834static int common_hrtimer_try_to_cancel(struct k_itimer *timr)
 835{
 836	return hrtimer_try_to_cancel(&timr->it.real.timer);
 837}
 838
 839static void common_timer_wait_running(struct k_itimer *timer)
 840{
 841	hrtimer_cancel_wait_running(&timer->it.real.timer);
 842}
 843
 844/*
 845 * On PREEMPT_RT this prevents priority inversion and a potential livelock
 846 * against the ksoftirqd thread in case that ksoftirqd gets preempted while
 847 * executing a hrtimer callback.
 848 *
 849 * See the comments in hrtimer_cancel_wait_running(). For PREEMPT_RT=n this
 850 * just results in a cpu_relax().
 851 *
 852 * For POSIX CPU timers with CONFIG_POSIX_CPU_TIMERS_TASK_WORK=n this is
 853 * just a cpu_relax(). With CONFIG_POSIX_CPU_TIMERS_TASK_WORK=y this
 854 * prevents spinning on an eventually scheduled out task and a livelock
 855 * when the task which tries to delete or disarm the timer has preempted
 856 * the task which runs the expiry in task work context.
 857 */
 858static void timer_wait_running(struct k_itimer *timer)
 859{
 860	/*
 861	 * kc->timer_wait_running() might drop RCU lock. So @timer
 862	 * cannot be touched anymore after the function returns!
 863	 */
 864	timer->kclock->timer_wait_running(timer);
 865}
 866
 867/*
 868 * Set up the new interval and reset the signal delivery data
 869 */
 870void posix_timer_set_common(struct k_itimer *timer, struct itimerspec64 *new_setting)
 871{
 872	if (new_setting->it_value.tv_sec || new_setting->it_value.tv_nsec)
 873		timer->it_interval = timespec64_to_ktime(new_setting->it_interval);
 874	else
 875		timer->it_interval = 0;
 876
 877	/* Reset overrun accounting */
 878	timer->it_overrun_last = 0;
 879	timer->it_overrun = -1LL;
 880}
 881
 882/* Set a POSIX.1b interval timer. */
 883int common_timer_set(struct k_itimer *timr, int flags,
 884		     struct itimerspec64 *new_setting,
 885		     struct itimerspec64 *old_setting)
 886{
 887	const struct k_clock *kc = timr->kclock;
 888	bool sigev_none;
 889	ktime_t expires;
 890
 891	if (old_setting)
 892		common_timer_get(timr, old_setting);
 893
 894	/*
 895	 * Careful here. On SMP systems the timer expiry function could be
 896	 * active and spinning on timr->it_lock.
 897	 */
 898	if (kc->timer_try_to_cancel(timr) < 0)
 899		return TIMER_RETRY;
 900
 901	timr->it_status = POSIX_TIMER_DISARMED;
 902	posix_timer_set_common(timr, new_setting);
 903
 904	/* Keep timer disarmed when it_value is zero */
 905	if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
 906		return 0;
 907
 908	expires = timespec64_to_ktime(new_setting->it_value);
 909	if (flags & TIMER_ABSTIME)
 910		expires = timens_ktime_to_host(timr->it_clock, expires);
 911	sigev_none = timr->it_sigev_notify == SIGEV_NONE;
 912
 913	kc->timer_arm(timr, expires, flags & TIMER_ABSTIME, sigev_none);
 914	if (!sigev_none)
 915		timr->it_status = POSIX_TIMER_ARMED;
 916	return 0;
 917}
 918
 919static int do_timer_settime(timer_t timer_id, int tmr_flags, struct itimerspec64 *new_spec64,
 920			    struct itimerspec64 *old_spec64)
 921{
 922	if (!timespec64_valid(&new_spec64->it_interval) ||
 923	    !timespec64_valid(&new_spec64->it_value))
 924		return -EINVAL;
 925
 926	if (old_spec64)
 927		memset(old_spec64, 0, sizeof(*old_spec64));
 928
 929	for (; ; old_spec64 = NULL) {
 930		struct k_itimer *timr;
 931
 932		scoped_timer_get_or_fail(timer_id) {
 933			timr = scoped_timer;
 934
 935			if (old_spec64)
 936				old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
 937
 938			/* Prevent signal delivery and rearming. */
 939			timr->it_signal_seq++;
 940
 941			int ret = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
 942			if (ret != TIMER_RETRY)
 943				return ret;
 944
 945			/* Protect the timer from being freed when leaving the lock scope */
 946			rcu_read_lock();
 947		}
 948		timer_wait_running(timr);
 949		rcu_read_unlock();
 950	}
 951}
 952
 953/* Set a POSIX.1b interval timer */
 954SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
 955		const struct __kernel_itimerspec __user *, new_setting,
 956		struct __kernel_itimerspec __user *, old_setting)
 957{
 958	struct itimerspec64 new_spec, old_spec, *rtn;
 959	int error = 0;
 960
 961	if (!new_setting)
 962		return -EINVAL;
 963
 964	if (get_itimerspec64(&new_spec, new_setting))
 965		return -EFAULT;
 966
 967	rtn = old_setting ? &old_spec : NULL;
 968	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 969	if (!error && old_setting) {
 970		if (put_itimerspec64(&old_spec, old_setting))
 971			error = -EFAULT;
 972	}
 973	return error;
 974}
 975
 976#ifdef CONFIG_COMPAT_32BIT_TIME
 977SYSCALL_DEFINE4(timer_settime32, timer_t, timer_id, int, flags,
 978		struct old_itimerspec32 __user *, new,
 979		struct old_itimerspec32 __user *, old)
 980{
 981	struct itimerspec64 new_spec, old_spec;
 982	struct itimerspec64 *rtn = old ? &old_spec : NULL;
 983	int error = 0;
 984
 985	if (!new)
 986		return -EINVAL;
 987	if (get_old_itimerspec32(&new_spec, new))
 988		return -EFAULT;
 989
 990	error = do_timer_settime(timer_id, flags, &new_spec, rtn);
 991	if (!error && old) {
 992		if (put_old_itimerspec32(&old_spec, old))
 993			error = -EFAULT;
 994	}
 995	return error;
 996}
 997#endif
 998
 999int common_timer_del(struct k_itimer *timer)
1000{
1001	const struct k_clock *kc = timer->kclock;
1002
1003	if (kc->timer_try_to_cancel(timer) < 0)
1004		return TIMER_RETRY;
1005	timer->it_status = POSIX_TIMER_DISARMED;
1006	return 0;
1007}
1008
1009/*
1010 * If the deleted timer is on the ignored list, remove it and
1011 * drop the associated reference.
1012 */
1013static inline void posix_timer_cleanup_ignored(struct k_itimer *tmr)
1014{
1015	if (!hlist_unhashed(&tmr->ignored_list)) {
1016		hlist_del_init(&tmr->ignored_list);
1017		posixtimer_putref(tmr);
1018	}
1019}
1020
1021static void posix_timer_delete(struct k_itimer *timer)
1022{
1023	/*
1024	 * Invalidate the timer, remove it from the linked list and remove
1025	 * it from the ignored list if pending.
1026	 *
1027	 * The invalidation must be written with siglock held so that the
1028	 * signal code observes the invalidated timer::it_signal in
1029	 * do_sigaction(), which prevents it from moving a pending signal
1030	 * of a deleted timer to the ignore list.
1031	 *
1032	 * The invalidation also prevents signal queueing, signal delivery
1033	 * and therefore rearming from the signal delivery path.
1034	 *
1035	 * A concurrent lookup can still find the timer in the hash, but it
1036	 * will check timer::it_signal with timer::it_lock held and observe
1037	 * bit 0 set, which invalidates it. That also prevents the timer ID
1038	 * from being handed out before this timer is completely gone.
1039	 */
1040	timer->it_signal_seq++;
1041
1042	scoped_guard (spinlock, &current->sighand->siglock) {
1043		unsigned long sig = (unsigned long)timer->it_signal | 1UL;
1044
1045		WRITE_ONCE(timer->it_signal, (struct signal_struct *)sig);
1046		hlist_del_rcu(&timer->list);
1047		posix_timer_cleanup_ignored(timer);
1048	}
1049
1050	while (timer->kclock->timer_del(timer) == TIMER_RETRY) {
1051		guard(rcu)();
1052		spin_unlock_irq(&timer->it_lock);
1053		timer_wait_running(timer);
1054		spin_lock_irq(&timer->it_lock);
1055	}
1056}
1057
1058/* Delete a POSIX.1b interval timer. */
1059SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
1060{
1061	struct k_itimer *timer;
1062
1063	scoped_timer_get_or_fail(timer_id) {
1064		timer = scoped_timer;
1065		posix_timer_delete(timer);
1066	}
1067	/* Remove it from the hash, which frees up the timer ID */
1068	posix_timer_unhash_and_free(timer);
1069	return 0;
1070}
1071
1072/*
1073 * Invoked from do_exit() when the last thread of a thread group exits.
1074 * At that point no other task can access the timers of the dying
1075 * task anymore.
1076 */
1077void exit_itimers(struct task_struct *tsk)
1078{
1079	struct hlist_head timers;
1080	struct hlist_node *next;
1081	struct k_itimer *timer;
1082
1083	/* Clear restore mode for exec() */
1084	tsk->signal->timer_create_restore_ids = 0;
1085
1086	if (hlist_empty(&tsk->signal->posix_timers))
1087		return;
1088
1089	/* Protect against concurrent read via /proc/$PID/timers */
1090	scoped_guard (spinlock_irq, &tsk->sighand->siglock)
1091		hlist_move_list(&tsk->signal->posix_timers, &timers);
1092
1093	/* The timers are not longer accessible via tsk::signal */
1094	hlist_for_each_entry_safe(timer, next, &timers, list) {
1095		scoped_guard (spinlock_irq, &timer->it_lock)
1096			posix_timer_delete(timer);
1097		posix_timer_unhash_and_free(timer);
1098		cond_resched();
1099	}
1100
1101	/*
1102	 * There should be no timers on the ignored list. itimer_delete() has
1103	 * mopped them up.
1104	 */
1105	if (!WARN_ON_ONCE(!hlist_empty(&tsk->signal->ignored_posix_timers)))
1106		return;
1107
1108	hlist_move_list(&tsk->signal->ignored_posix_timers, &timers);
1109	while (!hlist_empty(&timers)) {
1110		posix_timer_cleanup_ignored(hlist_entry(timers.first, struct k_itimer,
1111							ignored_list));
1112	}
1113}
1114
1115SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1116		const struct __kernel_timespec __user *, tp)
1117{
1118	const struct k_clock *kc = clockid_to_kclock(which_clock);
1119	struct timespec64 new_tp;
1120
1121	if (!kc || !kc->clock_set)
1122		return -EINVAL;
1123
1124	if (get_timespec64(&new_tp, tp))
1125		return -EFAULT;
1126
1127	/*
1128	 * Permission checks have to be done inside the clock specific
1129	 * setter callback.
1130	 */
1131	return kc->clock_set(which_clock, &new_tp);
1132}
1133
1134SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1135		struct __kernel_timespec __user *, tp)
1136{
1137	const struct k_clock *kc = clockid_to_kclock(which_clock);
1138	struct timespec64 kernel_tp;
1139	int error;
1140
1141	if (!kc)
1142		return -EINVAL;
1143
1144	error = kc->clock_get_timespec(which_clock, &kernel_tp);
1145
1146	if (!error && put_timespec64(&kernel_tp, tp))
1147		error = -EFAULT;
1148
1149	return error;
1150}
1151
1152int do_clock_adjtime(const clockid_t which_clock, struct __kernel_timex * ktx)
1153{
1154	const struct k_clock *kc = clockid_to_kclock(which_clock);
1155
1156	if (!kc)
1157		return -EINVAL;
1158	if (!kc->clock_adj)
1159		return -EOPNOTSUPP;
1160
1161	return kc->clock_adj(which_clock, ktx);
1162}
1163
1164SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1165		struct __kernel_timex __user *, utx)
1166{
1167	struct __kernel_timex ktx;
1168	int err;
1169
1170	if (copy_from_user(&ktx, utx, sizeof(ktx)))
1171		return -EFAULT;
1172
1173	err = do_clock_adjtime(which_clock, &ktx);
1174
1175	if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1176		return -EFAULT;
1177
1178	return err;
1179}
1180
1181/**
1182 * sys_clock_getres - Get the resolution of a clock
1183 * @which_clock:	The clock to get the resolution for
1184 * @tp:			Pointer to a a user space timespec64 for storage
1185 *
1186 * POSIX defines:
1187 *
1188 * "The clock_getres() function shall return the resolution of any
1189 * clock. Clock resolutions are implementation-defined and cannot be set by
1190 * a process. If the argument res is not NULL, the resolution of the
1191 * specified clock shall be stored in the location pointed to by res. If
1192 * res is NULL, the clock resolution is not returned. If the time argument
1193 * of clock_settime() is not a multiple of res, then the value is truncated
1194 * to a multiple of res."
1195 *
1196 * Due to the various hardware constraints the real resolution can vary
1197 * wildly and even change during runtime when the underlying devices are
1198 * replaced. The kernel also can use hardware devices with different
1199 * resolutions for reading the time and for arming timers.
1200 *
1201 * The kernel therefore deviates from the POSIX spec in various aspects:
1202 *
1203 * 1) The resolution returned to user space
1204 *
1205 *    For CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, CLOCK_TAI,
1206 *    CLOCK_REALTIME_ALARM, CLOCK_BOOTTIME_ALAREM and CLOCK_MONOTONIC_RAW
1207 *    the kernel differentiates only two cases:
1208 *
1209 *    I)  Low resolution mode:
1210 *
1211 *	  When high resolution timers are disabled at compile or runtime
1212 *	  the resolution returned is nanoseconds per tick, which represents
1213 *	  the precision at which timers expire.
1214 *
1215 *    II) High resolution mode:
1216 *
1217 *	  When high resolution timers are enabled the resolution returned
1218 *	  is always one nanosecond independent of the actual resolution of
1219 *	  the underlying hardware devices.
1220 *
1221 *	  For CLOCK_*_ALARM the actual resolution depends on system
1222 *	  state. When system is running the resolution is the same as the
1223 *	  resolution of the other clocks. During suspend the actual
1224 *	  resolution is the resolution of the underlying RTC device which
1225 *	  might be way less precise than the clockevent device used during
1226 *	  running state.
1227 *
1228 *   For CLOCK_REALTIME_COARSE and CLOCK_MONOTONIC_COARSE the resolution
1229 *   returned is always nanoseconds per tick.
1230 *
1231 *   For CLOCK_PROCESS_CPUTIME and CLOCK_THREAD_CPUTIME the resolution
1232 *   returned is always one nanosecond under the assumption that the
1233 *   underlying scheduler clock has a better resolution than nanoseconds
1234 *   per tick.
1235 *
1236 *   For dynamic POSIX clocks (PTP devices) the resolution returned is
1237 *   always one nanosecond.
1238 *
1239 * 2) Affect on sys_clock_settime()
1240 *
1241 *    The kernel does not truncate the time which is handed in to
1242 *    sys_clock_settime(). The kernel internal timekeeping is always using
1243 *    nanoseconds precision independent of the clocksource device which is
1244 *    used to read the time from. The resolution of that device only
1245 *    affects the precision of the time returned by sys_clock_gettime().
1246 *
1247 * Returns:
1248 *	0		Success. @tp contains the resolution
1249 *	-EINVAL		@which_clock is not a valid clock ID
1250 *	-EFAULT		Copying the resolution to @tp faulted
1251 *	-ENODEV		Dynamic POSIX clock is not backed by a device
1252 *	-EOPNOTSUPP	Dynamic POSIX clock does not support getres()
1253 */
1254SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1255		struct __kernel_timespec __user *, tp)
1256{
1257	const struct k_clock *kc = clockid_to_kclock(which_clock);
1258	struct timespec64 rtn_tp;
1259	int error;
1260
1261	if (!kc)
1262		return -EINVAL;
1263
1264	error = kc->clock_getres(which_clock, &rtn_tp);
1265
1266	if (!error && tp && put_timespec64(&rtn_tp, tp))
1267		error = -EFAULT;
1268
1269	return error;
1270}
1271
1272#ifdef CONFIG_COMPAT_32BIT_TIME
1273
1274SYSCALL_DEFINE2(clock_settime32, clockid_t, which_clock,
1275		struct old_timespec32 __user *, tp)
1276{
1277	const struct k_clock *kc = clockid_to_kclock(which_clock);
1278	struct timespec64 ts;
1279
1280	if (!kc || !kc->clock_set)
1281		return -EINVAL;
1282
1283	if (get_old_timespec32(&ts, tp))
1284		return -EFAULT;
1285
1286	return kc->clock_set(which_clock, &ts);
1287}
1288
1289SYSCALL_DEFINE2(clock_gettime32, clockid_t, which_clock,
1290		struct old_timespec32 __user *, tp)
1291{
1292	const struct k_clock *kc = clockid_to_kclock(which_clock);
1293	struct timespec64 ts;
1294	int err;
1295
1296	if (!kc)
1297		return -EINVAL;
1298
1299	err = kc->clock_get_timespec(which_clock, &ts);
1300
1301	if (!err && put_old_timespec32(&ts, tp))
1302		err = -EFAULT;
1303
1304	return err;
1305}
1306
1307SYSCALL_DEFINE2(clock_adjtime32, clockid_t, which_clock,
1308		struct old_timex32 __user *, utp)
1309{
1310	struct __kernel_timex ktx;
1311	int err;
1312
1313	err = get_old_timex32(&ktx, utp);
1314	if (err)
1315		return err;
1316
1317	err = do_clock_adjtime(which_clock, &ktx);
1318
1319	if (err >= 0 && put_old_timex32(utp, &ktx))
1320		return -EFAULT;
1321
1322	return err;
1323}
1324
1325SYSCALL_DEFINE2(clock_getres_time32, clockid_t, which_clock,
1326		struct old_timespec32 __user *, tp)
1327{
1328	const struct k_clock *kc = clockid_to_kclock(which_clock);
1329	struct timespec64 ts;
1330	int err;
1331
1332	if (!kc)
1333		return -EINVAL;
1334
1335	err = kc->clock_getres(which_clock, &ts);
1336	if (!err && tp && put_old_timespec32(&ts, tp))
1337		return -EFAULT;
1338
1339	return err;
1340}
1341
1342#endif
1343
1344/*
1345 * sys_clock_nanosleep() for CLOCK_REALTIME and CLOCK_TAI
1346 */
1347static int common_nsleep(const clockid_t which_clock, int flags,
1348			 const struct timespec64 *rqtp)
1349{
1350	ktime_t texp = timespec64_to_ktime(*rqtp);
1351
1352	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1353				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1354				 which_clock);
1355}
1356
1357/*
1358 * sys_clock_nanosleep() for CLOCK_MONOTONIC and CLOCK_BOOTTIME
1359 *
1360 * Absolute nanosleeps for these clocks are time-namespace adjusted.
1361 */
1362static int common_nsleep_timens(const clockid_t which_clock, int flags,
1363				const struct timespec64 *rqtp)
1364{
1365	ktime_t texp = timespec64_to_ktime(*rqtp);
1366
1367	if (flags & TIMER_ABSTIME)
1368		texp = timens_ktime_to_host(which_clock, texp);
1369
1370	return hrtimer_nanosleep(texp, flags & TIMER_ABSTIME ?
1371				 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1372				 which_clock);
1373}
1374
1375SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1376		const struct __kernel_timespec __user *, rqtp,
1377		struct __kernel_timespec __user *, rmtp)
1378{
1379	const struct k_clock *kc = clockid_to_kclock(which_clock);
1380	struct timespec64 t;
1381
1382	if (!kc)
1383		return -EINVAL;
1384	if (!kc->nsleep)
1385		return -EOPNOTSUPP;
1386
1387	if (get_timespec64(&t, rqtp))
1388		return -EFAULT;
1389
1390	if (!timespec64_valid(&t))
1391		return -EINVAL;
1392	if (flags & TIMER_ABSTIME)
1393		rmtp = NULL;
1394	current->restart_block.fn = do_no_restart_syscall;
1395	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1396	current->restart_block.nanosleep.rmtp = rmtp;
1397
1398	return kc->nsleep(which_clock, flags, &t);
1399}
1400
1401#ifdef CONFIG_COMPAT_32BIT_TIME
1402
1403SYSCALL_DEFINE4(clock_nanosleep_time32, clockid_t, which_clock, int, flags,
1404		struct old_timespec32 __user *, rqtp,
1405		struct old_timespec32 __user *, rmtp)
1406{
1407	const struct k_clock *kc = clockid_to_kclock(which_clock);
1408	struct timespec64 t;
1409
1410	if (!kc)
1411		return -EINVAL;
1412	if (!kc->nsleep)
1413		return -EOPNOTSUPP;
1414
1415	if (get_old_timespec32(&t, rqtp))
1416		return -EFAULT;
1417
1418	if (!timespec64_valid(&t))
1419		return -EINVAL;
1420	if (flags & TIMER_ABSTIME)
1421		rmtp = NULL;
1422	current->restart_block.fn = do_no_restart_syscall;
1423	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1424	current->restart_block.nanosleep.compat_rmtp = rmtp;
1425
1426	return kc->nsleep(which_clock, flags, &t);
1427}
1428
1429#endif
1430
1431static const struct k_clock clock_realtime = {
1432	.clock_getres		= posix_get_hrtimer_res,
1433	.clock_get_timespec	= posix_get_realtime_timespec,
1434	.clock_get_ktime	= posix_get_realtime_ktime,
1435	.clock_set		= posix_clock_realtime_set,
1436	.clock_adj		= posix_clock_realtime_adj,
1437	.nsleep			= common_nsleep,
1438	.timer_create		= common_timer_create,
1439	.timer_set		= common_timer_set,
1440	.timer_get		= common_timer_get,
1441	.timer_del		= common_timer_del,
1442	.timer_rearm		= common_hrtimer_rearm,
1443	.timer_forward		= common_hrtimer_forward,
1444	.timer_remaining	= common_hrtimer_remaining,
1445	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1446	.timer_wait_running	= common_timer_wait_running,
1447	.timer_arm		= common_hrtimer_arm,
1448};
1449
1450static const struct k_clock clock_monotonic = {
1451	.clock_getres		= posix_get_hrtimer_res,
1452	.clock_get_timespec	= posix_get_monotonic_timespec,
1453	.clock_get_ktime	= posix_get_monotonic_ktime,
1454	.nsleep			= common_nsleep_timens,
1455	.timer_create		= common_timer_create,
1456	.timer_set		= common_timer_set,
1457	.timer_get		= common_timer_get,
1458	.timer_del		= common_timer_del,
1459	.timer_rearm		= common_hrtimer_rearm,
1460	.timer_forward		= common_hrtimer_forward,
1461	.timer_remaining	= common_hrtimer_remaining,
1462	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1463	.timer_wait_running	= common_timer_wait_running,
1464	.timer_arm		= common_hrtimer_arm,
1465};
1466
1467static const struct k_clock clock_monotonic_raw = {
1468	.clock_getres		= posix_get_hrtimer_res,
1469	.clock_get_timespec	= posix_get_monotonic_raw,
1470};
1471
1472static const struct k_clock clock_realtime_coarse = {
1473	.clock_getres		= posix_get_coarse_res,
1474	.clock_get_timespec	= posix_get_realtime_coarse,
1475};
1476
1477static const struct k_clock clock_monotonic_coarse = {
1478	.clock_getres		= posix_get_coarse_res,
1479	.clock_get_timespec	= posix_get_monotonic_coarse,
1480};
1481
1482static const struct k_clock clock_tai = {
1483	.clock_getres		= posix_get_hrtimer_res,
1484	.clock_get_ktime	= posix_get_tai_ktime,
1485	.clock_get_timespec	= posix_get_tai_timespec,
1486	.nsleep			= common_nsleep,
1487	.timer_create		= common_timer_create,
1488	.timer_set		= common_timer_set,
1489	.timer_get		= common_timer_get,
1490	.timer_del		= common_timer_del,
1491	.timer_rearm		= common_hrtimer_rearm,
1492	.timer_forward		= common_hrtimer_forward,
1493	.timer_remaining	= common_hrtimer_remaining,
1494	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1495	.timer_wait_running	= common_timer_wait_running,
1496	.timer_arm		= common_hrtimer_arm,
1497};
1498
1499static const struct k_clock clock_boottime = {
1500	.clock_getres		= posix_get_hrtimer_res,
1501	.clock_get_ktime	= posix_get_boottime_ktime,
1502	.clock_get_timespec	= posix_get_boottime_timespec,
1503	.nsleep			= common_nsleep_timens,
1504	.timer_create		= common_timer_create,
1505	.timer_set		= common_timer_set,
1506	.timer_get		= common_timer_get,
1507	.timer_del		= common_timer_del,
1508	.timer_rearm		= common_hrtimer_rearm,
1509	.timer_forward		= common_hrtimer_forward,
1510	.timer_remaining	= common_hrtimer_remaining,
1511	.timer_try_to_cancel	= common_hrtimer_try_to_cancel,
1512	.timer_wait_running	= common_timer_wait_running,
1513	.timer_arm		= common_hrtimer_arm,
1514};
1515
1516static const struct k_clock * const posix_clocks[] = {
1517	[CLOCK_REALTIME]		= &clock_realtime,
1518	[CLOCK_MONOTONIC]		= &clock_monotonic,
1519	[CLOCK_PROCESS_CPUTIME_ID]	= &clock_process,
1520	[CLOCK_THREAD_CPUTIME_ID]	= &clock_thread,
1521	[CLOCK_MONOTONIC_RAW]		= &clock_monotonic_raw,
1522	[CLOCK_REALTIME_COARSE]		= &clock_realtime_coarse,
1523	[CLOCK_MONOTONIC_COARSE]	= &clock_monotonic_coarse,
1524	[CLOCK_BOOTTIME]		= &clock_boottime,
1525	[CLOCK_REALTIME_ALARM]		= &alarm_clock,
1526	[CLOCK_BOOTTIME_ALARM]		= &alarm_clock,
1527	[CLOCK_TAI]			= &clock_tai,
1528#ifdef CONFIG_POSIX_AUX_CLOCKS
1529	[CLOCK_AUX ... CLOCK_AUX_LAST]	= &clock_aux,
1530#endif
1531};
1532
1533static const struct k_clock *clockid_to_kclock(const clockid_t id)
1534{
1535	clockid_t idx = id;
1536
1537	if (id < 0) {
1538		return (id & CLOCKFD_MASK) == CLOCKFD ?
1539			&clock_posix_dynamic : &clock_posix_cpu;
1540	}
1541
1542	if (id >= ARRAY_SIZE(posix_clocks))
1543		return NULL;
1544
1545	return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
1546}
1547
1548static int __init posixtimer_init(void)
1549{
1550	unsigned long i, size;
1551	unsigned int shift;
1552
1553	posix_timers_cache = kmem_cache_create("posix_timers_cache",
1554					       sizeof(struct k_itimer),
1555					       __alignof__(struct k_itimer),
1556					       SLAB_ACCOUNT, NULL);
1557
1558	if (IS_ENABLED(CONFIG_BASE_SMALL))
1559		size = 512;
1560	else
1561		size = roundup_pow_of_two(512 * num_possible_cpus());
1562
1563	timer_buckets = alloc_large_system_hash("posixtimers", sizeof(*timer_buckets),
1564						size, 0, 0, &shift, NULL, size, size);
1565	size = 1UL << shift;
1566	timer_hashmask = size - 1;
1567
1568	for (i = 0; i < size; i++) {
1569		spin_lock_init(&timer_buckets[i].lock);
1570		INIT_HLIST_HEAD(&timer_buckets[i].head);
1571	}
1572	return 0;
1573}
1574core_initcall(posixtimer_init);