[linux.git] / fs / timerfd.c

/*
 *  fs/timerfd.c
 *
 *  Copyright (C) 2007  Davide Libenzi <[email protected]>
 *
 *
 *  Thanks to Thomas Gleixner for code reviews and useful comments.
 *
 */

#include <linux/alarmtimer.h>
#include <linux/file.h>
#include <linux/poll.h>
#include <linux/init.h>
#include <linux/fs.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/list.h>
#include <linux/spinlock.h>
#include <linux/time.h>
#include <linux/hrtimer.h>
#include <linux/anon_inodes.h>
#include <linux/timerfd.h>
#include <linux/syscalls.h>
#include <linux/compat.h>
#include <linux/rcupdate.h>

struct timerfd_ctx {
	union {
		struct hrtimer tmr;
		struct alarm alarm;
	} t;
	ktime_t tintv;
	ktime_t moffs;
	wait_queue_head_t wqh;
	u64 ticks;
	int expired;
	int clockid;
	struct rcu_head rcu;
	struct list_head clist;
	bool might_cancel;
};

static LIST_HEAD(cancel_list);
static DEFINE_SPINLOCK(cancel_lock);

static inline bool isalarm(struct timerfd_ctx *ctx)
{
	return ctx->clockid == CLOCK_REALTIME_ALARM ||
		ctx->clockid == CLOCK_BOOTTIME_ALARM;
}

/*
 * This gets called when the timer event triggers. We set the "expired"
 * flag, but we do not re-arm the timer (in case it's necessary,
 * tintv.tv64 != 0) until the timer is accessed.
 */
static void timerfd_triggered(struct timerfd_ctx *ctx)
{
	unsigned long flags;

	spin_lock_irqsave(&ctx->wqh.lock, flags);
	ctx->expired = 1;
	ctx->ticks++;
	wake_up_locked(&ctx->wqh);
	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
}

static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
{
	struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx,
					       t.tmr);
	timerfd_triggered(ctx);
	return HRTIMER_NORESTART;
}

static enum alarmtimer_restart timerfd_alarmproc(struct alarm *alarm,
	ktime_t now)
{
	struct timerfd_ctx *ctx = container_of(alarm, struct timerfd_ctx,
					       t.alarm);
	timerfd_triggered(ctx);
	return ALARMTIMER_NORESTART;
}

/*
 * Called when the clock was set to cancel the timers in the cancel
 * list. This will wake up processes waiting on these timers. The
 * wake-up requires ctx->ticks to be non zero, therefore we increment
 * it before calling wake_up_locked().
 */
void timerfd_clock_was_set(void)
{
	ktime_t moffs = ktime_get_monotonic_offset();
	struct timerfd_ctx *ctx;
	unsigned long flags;

	rcu_read_lock();
	list_for_each_entry_rcu(ctx, &cancel_list, clist) {
		if (!ctx->might_cancel)
			continue;
		spin_lock_irqsave(&ctx->wqh.lock, flags);
		if (ctx->moffs.tv64 != moffs.tv64) {
			ctx->moffs.tv64 = KTIME_MAX;
			ctx->ticks++;
			wake_up_locked(&ctx->wqh);
		}
		spin_unlock_irqrestore(&ctx->wqh.lock, flags);
	}
	rcu_read_unlock();
}

static void timerfd_remove_cancel(struct timerfd_ctx *ctx)
{
	if (ctx->might_cancel) {
		ctx->might_cancel = false;
		spin_lock(&cancel_lock);
		list_del_rcu(&ctx->clist);
		spin_unlock(&cancel_lock);
	}
}

static bool timerfd_canceled(struct timerfd_ctx *ctx)
{
	if (!ctx->might_cancel || ctx->moffs.tv64 != KTIME_MAX)
		return false;
	ctx->moffs = ktime_get_monotonic_offset();
	return true;
}

static void timerfd_setup_cancel(struct timerfd_ctx *ctx, int flags)
{
	if ((ctx->clockid == CLOCK_REALTIME ||
	     ctx->clockid == CLOCK_REALTIME_ALARM) &&
	    (flags & TFD_TIMER_ABSTIME) && (flags & TFD_TIMER_CANCEL_ON_SET)) {
		if (!ctx->might_cancel) {
			ctx->might_cancel = true;
			spin_lock(&cancel_lock);
			list_add_rcu(&ctx->clist, &cancel_list);
			spin_unlock(&cancel_lock);
		}
	} else if (ctx->might_cancel) {
		timerfd_remove_cancel(ctx);
	}
}

static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
{
	ktime_t remaining;

	if (isalarm(ctx))
		remaining = alarm_expires_remaining(&ctx->t.alarm);
	else
		remaining = hrtimer_expires_remaining(&ctx->t.tmr);

	return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
}

static int timerfd_setup(struct timerfd_ctx *ctx, int flags,
			 const struct itimerspec *ktmr)
{
	enum hrtimer_mode htmode;
	ktime_t texp;
	int clockid = ctx->clockid;

	htmode = (flags & TFD_TIMER_ABSTIME) ?
		HRTIMER_MODE_ABS: HRTIMER_MODE_REL;

	texp = timespec_to_ktime(ktmr->it_value);
	ctx->expired = 0;
	ctx->ticks = 0;
	ctx->tintv = timespec_to_ktime(ktmr->it_interval);

	if (isalarm(ctx)) {
		alarm_init(&ctx->t.alarm,
			   ctx->clockid == CLOCK_REALTIME_ALARM ?
			   ALARM_REALTIME : ALARM_BOOTTIME,
			   timerfd_alarmproc);
	} else {
		hrtimer_init(&ctx->t.tmr, clockid, htmode);
		hrtimer_set_expires(&ctx->t.tmr, texp);
		ctx->t.tmr.function = timerfd_tmrproc;
	}

	if (texp.tv64 != 0) {
		if (isalarm(ctx)) {
			if (flags & TFD_TIMER_ABSTIME)
				alarm_start(&ctx->t.alarm, texp);
			else
				alarm_start_relative(&ctx->t.alarm, texp);
		} else {
			hrtimer_start(&ctx->t.tmr, texp, htmode);
		}

		if (timerfd_canceled(ctx))
			return -ECANCELED;
	}
	return 0;
}

static int timerfd_release(struct inode *inode, struct file *file)
{
	struct timerfd_ctx *ctx = file->private_data;

	timerfd_remove_cancel(ctx);

	if (isalarm(ctx))
		alarm_cancel(&ctx->t.alarm);
	else
		hrtimer_cancel(&ctx->t.tmr);
	kfree_rcu(ctx, rcu);
	return 0;
}

static unsigned int timerfd_poll(struct file *file, poll_table *wait)
{
	struct timerfd_ctx *ctx = file->private_data;
	unsigned int events = 0;
	unsigned long flags;

	poll_wait(file, &ctx->wqh, wait);

	spin_lock_irqsave(&ctx->wqh.lock, flags);
	if (ctx->ticks)
		events |= POLLIN;
	spin_unlock_irqrestore(&ctx->wqh.lock, flags);

	return events;
}

static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
			    loff_t *ppos)
{
	struct timerfd_ctx *ctx = file->private_data;
	ssize_t res;
	u64 ticks = 0;

	if (count < sizeof(ticks))
		return -EINVAL;
	spin_lock_irq(&ctx->wqh.lock);
	if (file->f_flags & O_NONBLOCK)
		res = -EAGAIN;
	else
		res = wait_event_interruptible_locked_irq(ctx->wqh, ctx->ticks);

	/*
	 * If clock has changed, we do not care about the
	 * ticks and we do not rearm the timer. Userspace must
	 * reevaluate anyway.
	 */
	if (timerfd_canceled(ctx)) {
		ctx->ticks = 0;
		ctx->expired = 0;
		res = -ECANCELED;
	}

	if (ctx->ticks) {
		ticks = ctx->ticks;

		if (ctx->expired && ctx->tintv.tv64) {
			/*
			 * If tintv.tv64 != 0, this is a periodic timer that
			 * needs to be re-armed. We avoid doing it in the timer
			 * callback to avoid DoS attacks specifying a very
			 * short timer period.
			 */
			if (isalarm(ctx)) {
				ticks += alarm_forward_now(
					&ctx->t.alarm, ctx->tintv) - 1;
				alarm_restart(&ctx->t.alarm);
			} else {
				ticks += hrtimer_forward_now(&ctx->t.tmr,
							     ctx->tintv) - 1;
				hrtimer_restart(&ctx->t.tmr);
			}
		}
		ctx->expired = 0;
		ctx->ticks = 0;
	}
	spin_unlock_irq(&ctx->wqh.lock);
	if (ticks)
		res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
	return res;
}

static const struct file_operations timerfd_fops = {
	.release	= timerfd_release,
	.poll		= timerfd_poll,
	.read		= timerfd_read,
	.llseek		= noop_llseek,
};

static int timerfd_fget(int fd, struct fd *p)
{
	struct fd f = fdget(fd);
	if (!f.file)
		return -EBADF;
	if (f.file->f_op != &timerfd_fops) {
		fdput(f);
		return -EINVAL;
	}
	*p = f;
	return 0;
}

SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
{
	int ufd;
	struct timerfd_ctx *ctx;

	/* Check the TFD_* constants for consistency.  */
	BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
	BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);

	if ((flags & ~TFD_CREATE_FLAGS) ||
	    (clockid != CLOCK_MONOTONIC &&
	     clockid != CLOCK_REALTIME &&
	     clockid != CLOCK_REALTIME_ALARM &&
	     clockid != CLOCK_BOOTTIME_ALARM))
		return -EINVAL;

	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
	if (!ctx)
		return -ENOMEM;

	init_waitqueue_head(&ctx->wqh);
	ctx->clockid = clockid;

	if (isalarm(ctx))
		alarm_init(&ctx->t.alarm,
			   ctx->clockid == CLOCK_REALTIME_ALARM ?
			   ALARM_REALTIME : ALARM_BOOTTIME,
			   timerfd_alarmproc);
	else
		hrtimer_init(&ctx->t.tmr, clockid, HRTIMER_MODE_ABS);

	ctx->moffs = ktime_get_monotonic_offset();

	ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
			       O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
	if (ufd < 0)
		kfree(ctx);

	return ufd;
}

static int do_timerfd_settime(int ufd, int flags, 
		const struct itimerspec *new,
		struct itimerspec *old)
{
	struct fd f;
	struct timerfd_ctx *ctx;
	int ret;

	if ((flags & ~TFD_SETTIME_FLAGS) ||
	    !timespec_valid(&new->it_value) ||
	    !timespec_valid(&new->it_interval))
		return -EINVAL;

	ret = timerfd_fget(ufd, &f);
	if (ret)
		return ret;
	ctx = f.file->private_data;

	timerfd_setup_cancel(ctx, flags);

	/*
	 * We need to stop the existing timer before reprogramming
	 * it to the new values.
	 */
	for (;;) {
		spin_lock_irq(&ctx->wqh.lock);

		if (isalarm(ctx)) {
			if (alarm_try_to_cancel(&ctx->t.alarm) >= 0)
				break;
		} else {
			if (hrtimer_try_to_cancel(&ctx->t.tmr) >= 0)
				break;
		}
		spin_unlock_irq(&ctx->wqh.lock);
		cpu_relax();
	}

	/*
	 * If the timer is expired and it's periodic, we need to advance it
	 * because the caller may want to know the previous expiration time.
	 * We do not update "ticks" and "expired" since the timer will be
	 * re-programmed again in the following timerfd_setup() call.
	 */
	if (ctx->expired && ctx->tintv.tv64) {
		if (isalarm(ctx))
			alarm_forward_now(&ctx->t.alarm, ctx->tintv);
		else
			hrtimer_forward_now(&ctx->t.tmr, ctx->tintv);
	}

	old->it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
	old->it_interval = ktime_to_timespec(ctx->tintv);

	/*
	 * Re-program the timer to the new value ...
	 */
	ret = timerfd_setup(ctx, flags, new);

	spin_unlock_irq(&ctx->wqh.lock);
	fdput(f);
	return ret;
}

static int do_timerfd_gettime(int ufd, struct itimerspec *t)
{
	struct fd f;
	struct timerfd_ctx *ctx;
	int ret = timerfd_fget(ufd, &f);
	if (ret)
		return ret;
	ctx = f.file->private_data;

	spin_lock_irq(&ctx->wqh.lock);
	if (ctx->expired && ctx->tintv.tv64) {
		ctx->expired = 0;

		if (isalarm(ctx)) {
			ctx->ticks +=
				alarm_forward_now(
					&ctx->t.alarm, ctx->tintv) - 1;
			alarm_restart(&ctx->t.alarm);
		} else {
			ctx->ticks +=
				hrtimer_forward_now(&ctx->t.tmr, ctx->tintv)
				- 1;
			hrtimer_restart(&ctx->t.tmr);
		}
	}
	t->it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
	t->it_interval = ktime_to_timespec(ctx->tintv);
	spin_unlock_irq(&ctx->wqh.lock);
	fdput(f);
	return 0;
}

SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
		const struct itimerspec __user *, utmr,
		struct itimerspec __user *, otmr)
{
	struct itimerspec new, old;
	int ret;

	if (copy_from_user(&new, utmr, sizeof(new)))
		return -EFAULT;
	ret = do_timerfd_settime(ufd, flags, &new, &old);
	if (ret)
		return ret;
	if (otmr && copy_to_user(otmr, &old, sizeof(old)))
		return -EFAULT;

	return ret;
}

SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
{
	struct itimerspec kotmr;
	int ret = do_timerfd_gettime(ufd, &kotmr);
	if (ret)
		return ret;
	return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
}

#ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
		const struct compat_itimerspec __user *, utmr,
		struct compat_itimerspec __user *, otmr)
{
	struct itimerspec new, old;
	int ret;

	if (get_compat_itimerspec(&new, utmr))
		return -EFAULT;
	ret = do_timerfd_settime(ufd, flags, &new, &old);
	if (ret)
		return ret;
	if (otmr && put_compat_itimerspec(otmr, &old))
		return -EFAULT;
	return ret;
}

COMPAT_SYSCALL_DEFINE2(timerfd_gettime, int, ufd,
		struct compat_itimerspec __user *, otmr)
{
	struct itimerspec kotmr;
	int ret = do_timerfd_gettime(ufd, &kotmr);
	if (ret)
		return ret;
	return put_compat_itimerspec(otmr, &kotmr) ? -EFAULT: 0;
}
#endif
Commit	Line	Data
b215e283 DL	1	/*
	2	* fs/timerfd.c
	3	*
	4	* Copyright (C) 2007 Davide Libenzi <[email protected]>
	5	*
	6	*
	7	* Thanks to Thomas Gleixner for code reviews and useful comments.
	8	*
	9	*/
	10
11ffa9d6	11	#include <linux/alarmtimer.h>
b215e283 DL	12	#include <linux/file.h>
	13	#include <linux/poll.h>
	14	#include <linux/init.h>
	15	#include <linux/fs.h>
	16	#include <linux/sched.h>
	17	#include <linux/kernel.h>
5a0e3ad6	18	#include <linux/slab.h>
b215e283 DL	19	#include <linux/list.h>
	20	#include <linux/spinlock.h>
	21	#include <linux/time.h>
	22	#include <linux/hrtimer.h>
	23	#include <linux/anon_inodes.h>
	24	#include <linux/timerfd.h>
45cc2b96	25	#include <linux/syscalls.h>
9d94b9e2	26	#include <linux/compat.h>
9ec26907	27	#include <linux/rcupdate.h>
b215e283 DL	28
b215e283 DL	29	struct timerfd_ctx {
11ffa9d6 TP	30	union {
	31	struct hrtimer tmr;
	32	struct alarm alarm;
	33	} t;
b215e283	34	ktime_t tintv;
99ee5315	35	ktime_t moffs;
b215e283	36	wait_queue_head_t wqh;
4d672e7a	37	u64 ticks;
b215e283	38	int expired;
4d672e7a	39	int clockid;
9ec26907 TG	40	struct rcu_head rcu;
9ec26907 TG	41	struct list_head clist;
99ee5315	42	bool might_cancel;
b215e283 DL	43	};
b215e283 DL	44
9ec26907 TG	45	static LIST_HEAD(cancel_list);
	46	static DEFINE_SPINLOCK(cancel_lock);
	47
11ffa9d6 TP	48	static inline bool isalarm(struct timerfd_ctx *ctx)
	49	{
	50	return ctx->clockid == CLOCK_REALTIME_ALARM \|\|
	51	ctx->clockid == CLOCK_BOOTTIME_ALARM;
	52	}
	53
b215e283 DL	54	/*
	55	* This gets called when the timer event triggers. We set the "expired"
	56	* flag, but we do not re-arm the timer (in case it's necessary,
4d672e7a	57	* tintv.tv64 != 0) until the timer is accessed.
b215e283	58	*/
11ffa9d6	59	static void timerfd_triggered(struct timerfd_ctx *ctx)
b215e283	60	{
b215e283 DL	61	unsigned long flags;
b215e283 DL	62
18963c01	63	spin_lock_irqsave(&ctx->wqh.lock, flags);
b215e283	64	ctx->expired = 1;
4d672e7a	65	ctx->ticks++;
b215e283	66	wake_up_locked(&ctx->wqh);
18963c01	67	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
11ffa9d6	68	}
b215e283	69
11ffa9d6 TP	70	static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
	71	{
	72	struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx,
	73	t.tmr);
	74	timerfd_triggered(ctx);
b215e283 DL	75	return HRTIMER_NORESTART;
	76	}
	77
11ffa9d6 TP	78	static enum alarmtimer_restart timerfd_alarmproc(struct alarm *alarm,
	79	ktime_t now)
	80	{
	81	struct timerfd_ctx *ctx = container_of(alarm, struct timerfd_ctx,
	82	t.alarm);
	83	timerfd_triggered(ctx);
	84	return ALARMTIMER_NORESTART;
	85	}
	86
9ec26907 TG	87	/*
9ec26907 TG	88	* Called when the clock was set to cancel the timers in the cancel
1123d939 MA	89	* list. This will wake up processes waiting on these timers. The
	90	* wake-up requires ctx->ticks to be non zero, therefore we increment
	91	* it before calling wake_up_locked().
9ec26907 TG	92	*/
9ec26907 TG	93	void timerfd_clock_was_set(void)
4d672e7a	94	{
9ec26907 TG	95	ktime_t moffs = ktime_get_monotonic_offset();
	96	struct timerfd_ctx *ctx;
	97	unsigned long flags;
4d672e7a	98
9ec26907 TG	99	rcu_read_lock();
	100	list_for_each_entry_rcu(ctx, &cancel_list, clist) {
	101	if (!ctx->might_cancel)
	102	continue;
	103	spin_lock_irqsave(&ctx->wqh.lock, flags);
	104	if (ctx->moffs.tv64 != moffs.tv64) {
	105	ctx->moffs.tv64 = KTIME_MAX;
1123d939	106	ctx->ticks++;
9ec26907 TG	107	wake_up_locked(&ctx->wqh);
	108	}
	109	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
	110	}
	111	rcu_read_unlock();
4d672e7a DL	112	}
4d672e7a DL	113
9ec26907	114	static void timerfd_remove_cancel(struct timerfd_ctx *ctx)
99ee5315	115	{
9ec26907 TG	116	if (ctx->might_cancel) {
	117	ctx->might_cancel = false;
	118	spin_lock(&cancel_lock);
	119	list_del_rcu(&ctx->clist);
	120	spin_unlock(&cancel_lock);
	121	}
	122	}
99ee5315	123
9ec26907 TG	124	static bool timerfd_canceled(struct timerfd_ctx *ctx)
	125	{
	126	if (!ctx->might_cancel \|\| ctx->moffs.tv64 != KTIME_MAX)
99ee5315	127	return false;
9ec26907 TG	128	ctx->moffs = ktime_get_monotonic_offset();
	129	return true;
	130	}
99ee5315	131
9ec26907 TG	132	static void timerfd_setup_cancel(struct timerfd_ctx *ctx, int flags)
9ec26907 TG	133	{
11ffa9d6 TP	134	if ((ctx->clockid == CLOCK_REALTIME \|\|
	135	ctx->clockid == CLOCK_REALTIME_ALARM) &&
	136	(flags & TFD_TIMER_ABSTIME) && (flags & TFD_TIMER_CANCEL_ON_SET)) {
9ec26907 TG	137	if (!ctx->might_cancel) {
	138	ctx->might_cancel = true;
	139	spin_lock(&cancel_lock);
	140	list_add_rcu(&ctx->clist, &cancel_list);
	141	spin_unlock(&cancel_lock);
	142	}
	143	} else if (ctx->might_cancel) {
	144	timerfd_remove_cancel(ctx);
	145	}
	146	}
99ee5315	147
9ec26907 TG	148	static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
	149	{
	150	ktime_t remaining;
99ee5315	151
11ffa9d6 TP	152	if (isalarm(ctx))
	153	remaining = alarm_expires_remaining(&ctx->t.alarm);
	154	else
	155	remaining = hrtimer_expires_remaining(&ctx->t.tmr);
	156
9ec26907	157	return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
99ee5315 TG	158	}
	159
	160	static int timerfd_setup(struct timerfd_ctx *ctx, int flags,
	161	const struct itimerspec *ktmr)
b215e283 DL	162	{
	163	enum hrtimer_mode htmode;
	164	ktime_t texp;
99ee5315	165	int clockid = ctx->clockid;
b215e283 DL	166
	167	htmode = (flags & TFD_TIMER_ABSTIME) ?
	168	HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
	169
	170	texp = timespec_to_ktime(ktmr->it_value);
	171	ctx->expired = 0;
4d672e7a	172	ctx->ticks = 0;
b215e283	173	ctx->tintv = timespec_to_ktime(ktmr->it_interval);
11ffa9d6 TP	174
	175	if (isalarm(ctx)) {
	176	alarm_init(&ctx->t.alarm,
	177	ctx->clockid == CLOCK_REALTIME_ALARM ?
	178	ALARM_REALTIME : ALARM_BOOTTIME,
	179	timerfd_alarmproc);
	180	} else {
	181	hrtimer_init(&ctx->t.tmr, clockid, htmode);
	182	hrtimer_set_expires(&ctx->t.tmr, texp);
	183	ctx->t.tmr.function = timerfd_tmrproc;
	184	}
	185
99ee5315	186	if (texp.tv64 != 0) {
11ffa9d6 TP	187	if (isalarm(ctx)) {
	188	if (flags & TFD_TIMER_ABSTIME)
	189	alarm_start(&ctx->t.alarm, texp);
	190	else
	191	alarm_start_relative(&ctx->t.alarm, texp);
	192	} else {
	193	hrtimer_start(&ctx->t.tmr, texp, htmode);
	194	}
	195
99ee5315 TG	196	if (timerfd_canceled(ctx))
	197	return -ECANCELED;
	198	}
	199	return 0;
b215e283 DL	200	}
	201
	202	static int timerfd_release(struct inode inode, struct file file)
	203	{
	204	struct timerfd_ctx *ctx = file->private_data;
	205
9ec26907	206	timerfd_remove_cancel(ctx);
11ffa9d6 TP	207
	208	if (isalarm(ctx))
	209	alarm_cancel(&ctx->t.alarm);
	210	else
	211	hrtimer_cancel(&ctx->t.tmr);
9ec26907	212	kfree_rcu(ctx, rcu);
b215e283 DL	213	return 0;
	214	}
	215
	216	static unsigned int timerfd_poll(struct file file, poll_table wait)
	217	{
	218	struct timerfd_ctx *ctx = file->private_data;
	219	unsigned int events = 0;
	220	unsigned long flags;
	221
	222	poll_wait(file, &ctx->wqh, wait);
	223
18963c01	224	spin_lock_irqsave(&ctx->wqh.lock, flags);
4d672e7a	225	if (ctx->ticks)
b215e283	226	events \|= POLLIN;
18963c01	227	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
b215e283 DL	228
	229	return events;
	230	}
	231
	232	static ssize_t timerfd_read(struct file file, char __user buf, size_t count,
	233	loff_t *ppos)
	234	{
	235	struct timerfd_ctx *ctx = file->private_data;
	236	ssize_t res;
09828402	237	u64 ticks = 0;
b215e283 DL	238
	239	if (count < sizeof(ticks))
	240	return -EINVAL;
18963c01	241	spin_lock_irq(&ctx->wqh.lock);
8120a8aa MN	242	if (file->f_flags & O_NONBLOCK)
	243	res = -EAGAIN;
	244	else
	245	res = wait_event_interruptible_locked_irq(ctx->wqh, ctx->ticks);
99ee5315	246
9ec26907 TG	247	/*
	248	* If clock has changed, we do not care about the
	249	* ticks and we do not rearm the timer. Userspace must
	250	* reevaluate anyway.
	251	*/
	252	if (timerfd_canceled(ctx)) {
	253	ctx->ticks = 0;
	254	ctx->expired = 0;
	255	res = -ECANCELED;
	256	}
	257
4d672e7a DL	258	if (ctx->ticks) {
4d672e7a DL	259	ticks = ctx->ticks;
99ee5315	260
4d672e7a	261	if (ctx->expired && ctx->tintv.tv64) {
b215e283 DL	262	/*
	263	* If tintv.tv64 != 0, this is a periodic timer that
	264	* needs to be re-armed. We avoid doing it in the timer
	265	* callback to avoid DoS attacks specifying a very
	266	* short timer period.
	267	*/
11ffa9d6 TP	268	if (isalarm(ctx)) {
	269	ticks += alarm_forward_now(
	270	&ctx->t.alarm, ctx->tintv) - 1;
	271	alarm_restart(&ctx->t.alarm);
	272	} else {
	273	ticks += hrtimer_forward_now(&ctx->t.tmr,
	274	ctx->tintv) - 1;
	275	hrtimer_restart(&ctx->t.tmr);
	276	}
4d672e7a DL	277	}
	278	ctx->expired = 0;
	279	ctx->ticks = 0;
b215e283	280	}
18963c01	281	spin_unlock_irq(&ctx->wqh.lock);
b215e283	282	if (ticks)
09828402	283	res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
b215e283 DL	284	return res;
	285	}
	286
	287	static const struct file_operations timerfd_fops = {
	288	.release = timerfd_release,
	289	.poll = timerfd_poll,
	290	.read = timerfd_read,
6038f373	291	.llseek = noop_llseek,
b215e283 DL	292	};
b215e283 DL	293
2903ff01	294	static int timerfd_fget(int fd, struct fd *p)
4d672e7a	295	{
2903ff01 AV	296	struct fd f = fdget(fd);
	297	if (!f.file)
	298	return -EBADF;
	299	if (f.file->f_op != &timerfd_fops) {
	300	fdput(f);
	301	return -EINVAL;
4d672e7a	302	}
2903ff01 AV	303	*p = f;
2903ff01 AV	304	return 0;
4d672e7a DL	305	}
4d672e7a DL	306
836f92ad	307	SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
b215e283	308	{
2030a42c	309	int ufd;
b215e283	310	struct timerfd_ctx *ctx;
b215e283	311
e38b36f3 UD	312	/* Check the TFD_* constants for consistency. */
	313	BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
	314	BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
	315
610d18f4 DL	316	if ((flags & ~TFD_CREATE_FLAGS) \|\|
610d18f4 DL	317	(clockid != CLOCK_MONOTONIC &&
11ffa9d6 TP	318	clockid != CLOCK_REALTIME &&
	319	clockid != CLOCK_REALTIME_ALARM &&
	320	clockid != CLOCK_BOOTTIME_ALARM))
b215e283	321	return -EINVAL;
4d672e7a DL	322
	323	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
	324	if (!ctx)
	325	return -ENOMEM;
	326
	327	init_waitqueue_head(&ctx->wqh);
	328	ctx->clockid = clockid;
11ffa9d6 TP	329
	330	if (isalarm(ctx))
	331	alarm_init(&ctx->t.alarm,
	332	ctx->clockid == CLOCK_REALTIME_ALARM ?
	333	ALARM_REALTIME : ALARM_BOOTTIME,
	334	timerfd_alarmproc);
	335	else
	336	hrtimer_init(&ctx->t.tmr, clockid, HRTIMER_MODE_ABS);
	337
99ee5315	338	ctx->moffs = ktime_get_monotonic_offset();
4d672e7a	339
11fcb6c1	340	ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
628ff7c1	341	O_RDWR \| (flags & TFD_SHARED_FCNTL_FLAGS));
2030a42c	342	if (ufd < 0)
4d672e7a	343	kfree(ctx);
4d672e7a DL	344
	345	return ufd;
	346	}
	347
9d94b9e2 AV	348	static int do_timerfd_settime(int ufd, int flags,
	349	const struct itimerspec *new,
	350	struct itimerspec *old)
4d672e7a	351	{
2903ff01	352	struct fd f;
4d672e7a	353	struct timerfd_ctx *ctx;
2903ff01	354	int ret;
4d672e7a	355
610d18f4	356	if ((flags & ~TFD_SETTIME_FLAGS) \|\|
9d94b9e2 AV	357	!timespec_valid(&new->it_value) \|\|
9d94b9e2 AV	358	!timespec_valid(&new->it_interval))
b215e283 DL	359	return -EINVAL;
b215e283 DL	360
2903ff01 AV	361	ret = timerfd_fget(ufd, &f);
	362	if (ret)
	363	return ret;
	364	ctx = f.file->private_data;
b215e283	365
9ec26907 TG	366	timerfd_setup_cancel(ctx, flags);
9ec26907 TG	367
4d672e7a DL	368	/*
	369	* We need to stop the existing timer before reprogramming
	370	* it to the new values.
	371	*/
	372	for (;;) {
	373	spin_lock_irq(&ctx->wqh.lock);
11ffa9d6 TP	374
	375	if (isalarm(ctx)) {
	376	if (alarm_try_to_cancel(&ctx->t.alarm) >= 0)
	377	break;
	378	} else {
	379	if (hrtimer_try_to_cancel(&ctx->t.tmr) >= 0)
	380	break;
	381	}
18963c01	382	spin_unlock_irq(&ctx->wqh.lock);
4d672e7a	383	cpu_relax();
b215e283 DL	384	}
b215e283 DL	385
4d672e7a DL	386	/*
	387	* If the timer is expired and it's periodic, we need to advance it
	388	* because the caller may want to know the previous expiration time.
	389	* We do not update "ticks" and "expired" since the timer will be
	390	* re-programmed again in the following timerfd_setup() call.
	391	*/
11ffa9d6 TP	392	if (ctx->expired && ctx->tintv.tv64) {
	393	if (isalarm(ctx))
	394	alarm_forward_now(&ctx->t.alarm, ctx->tintv);
	395	else
	396	hrtimer_forward_now(&ctx->t.tmr, ctx->tintv);
	397	}
b215e283	398
9d94b9e2 AV	399	old->it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
9d94b9e2 AV	400	old->it_interval = ktime_to_timespec(ctx->tintv);
4d672e7a DL	401
	402	/*
	403	* Re-program the timer to the new value ...
	404	*/
9d94b9e2	405	ret = timerfd_setup(ctx, flags, new);
4d672e7a DL	406
4d672e7a DL	407	spin_unlock_irq(&ctx->wqh.lock);
2903ff01	408	fdput(f);
99ee5315	409	return ret;
4d672e7a DL	410	}
4d672e7a DL	411
9d94b9e2	412	static int do_timerfd_gettime(int ufd, struct itimerspec *t)
4d672e7a	413	{
2903ff01	414	struct fd f;
4d672e7a	415	struct timerfd_ctx *ctx;
2903ff01 AV	416	int ret = timerfd_fget(ufd, &f);
	417	if (ret)
	418	return ret;
	419	ctx = f.file->private_data;
4d672e7a DL	420
	421	spin_lock_irq(&ctx->wqh.lock);
	422	if (ctx->expired && ctx->tintv.tv64) {
	423	ctx->expired = 0;
11ffa9d6 TP	424
	425	if (isalarm(ctx)) {
	426	ctx->ticks +=
	427	alarm_forward_now(
	428	&ctx->t.alarm, ctx->tintv) - 1;
	429	alarm_restart(&ctx->t.alarm);
	430	} else {
	431	ctx->ticks +=
	432	hrtimer_forward_now(&ctx->t.tmr, ctx->tintv)
	433	- 1;
	434	hrtimer_restart(&ctx->t.tmr);
	435	}
4d672e7a	436	}
9d94b9e2 AV	437	t->it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
9d94b9e2 AV	438	t->it_interval = ktime_to_timespec(ctx->tintv);
4d672e7a	439	spin_unlock_irq(&ctx->wqh.lock);
2903ff01	440	fdput(f);
9d94b9e2 AV	441	return 0;
	442	}
	443
	444	SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
	445	const struct itimerspec __user *, utmr,
	446	struct itimerspec __user *, otmr)
	447	{
	448	struct itimerspec new, old;
	449	int ret;
	450
	451	if (copy_from_user(&new, utmr, sizeof(new)))
	452	return -EFAULT;
	453	ret = do_timerfd_settime(ufd, flags, &new, &old);
	454	if (ret)
	455	return ret;
	456	if (otmr && copy_to_user(otmr, &old, sizeof(old)))
	457	return -EFAULT;
	458
	459	return ret;
	460	}
4d672e7a	461
9d94b9e2 AV	462	SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
	463	{
	464	struct itimerspec kotmr;
	465	int ret = do_timerfd_gettime(ufd, &kotmr);
	466	if (ret)
	467	return ret;
4d672e7a	468	return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
b215e283 DL	469	}
b215e283 DL	470
0e803baf	471	#ifdef CONFIG_COMPAT
9d94b9e2	472	COMPAT_SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
0e803baf HC	473	const struct compat_itimerspec __user *, utmr,
0e803baf HC	474	struct compat_itimerspec __user *, otmr)
9d94b9e2 AV	475	{
	476	struct itimerspec new, old;
	477	int ret;
	478
	479	if (get_compat_itimerspec(&new, utmr))
	480	return -EFAULT;
	481	ret = do_timerfd_settime(ufd, flags, &new, &old);
	482	if (ret)
	483	return ret;
	484	if (otmr && put_compat_itimerspec(otmr, &old))
	485	return -EFAULT;
	486	return ret;
	487	}
	488
	489	COMPAT_SYSCALL_DEFINE2(timerfd_gettime, int, ufd,
0e803baf	490	struct compat_itimerspec __user *, otmr)
9d94b9e2 AV	491	{
	492	struct itimerspec kotmr;
	493	int ret = do_timerfd_gettime(ufd, &kotmr);
	494	if (ret)
	495	return ret;
0e803baf	496	return put_compat_itimerspec(otmr, &kotmr) ? -EFAULT: 0;
9d94b9e2 AV	497	}
9d94b9e2 AV	498	#endif