1/*
2 *  fs/timerfd.c
3 *
4 *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
5 *
6 *
7 *  Thanks to Thomas Gleixner for code reviews and useful comments.
8 *
9 */
10
11#include <linux/file.h>
12#include <linux/poll.h>
13#include <linux/init.h>
14#include <linux/fs.h>
15#include <linux/sched.h>
16#include <linux/kernel.h>
17#include <linux/list.h>
18#include <linux/spinlock.h>
19#include <linux/time.h>
20#include <linux/hrtimer.h>
21#include <linux/anon_inodes.h>
22#include <linux/timerfd.h>
23
24struct timerfd_ctx {
25	struct hrtimer tmr;
26	ktime_t tintv;
27	wait_queue_head_t wqh;
28	int expired;
29};
30
31/*
32 * This gets called when the timer event triggers. We set the "expired"
33 * flag, but we do not re-arm the timer (in case it's necessary,
34 * tintv.tv64 != 0) until the timer is read.
35 */
36static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
37{
38	struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx, tmr);
39	unsigned long flags;
40
41	spin_lock_irqsave(&ctx->wqh.lock, flags);
42	ctx->expired = 1;
43	wake_up_locked(&ctx->wqh);
44	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
45
46	return HRTIMER_NORESTART;
47}
48
49static void timerfd_setup(struct timerfd_ctx *ctx, int clockid, int flags,
50			  const struct itimerspec *ktmr)
51{
52	enum hrtimer_mode htmode;
53	ktime_t texp;
54
55	htmode = (flags & TFD_TIMER_ABSTIME) ?
56		HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
57
58	texp = timespec_to_ktime(ktmr->it_value);
59	ctx->expired = 0;
60	ctx->tintv = timespec_to_ktime(ktmr->it_interval);
61	hrtimer_init(&ctx->tmr, clockid, htmode);
62	ctx->tmr.expires = texp;
63	ctx->tmr.function = timerfd_tmrproc;
64	if (texp.tv64 != 0)
65		hrtimer_start(&ctx->tmr, texp, htmode);
66}
67
68static int timerfd_release(struct inode *inode, struct file *file)
69{
70	struct timerfd_ctx *ctx = file->private_data;
71
72	hrtimer_cancel(&ctx->tmr);
73	kfree(ctx);
74	return 0;
75}
76
77static unsigned int timerfd_poll(struct file *file, poll_table *wait)
78{
79	struct timerfd_ctx *ctx = file->private_data;
80	unsigned int events = 0;
81	unsigned long flags;
82
83	poll_wait(file, &ctx->wqh, wait);
84
85	spin_lock_irqsave(&ctx->wqh.lock, flags);
86	if (ctx->expired)
87		events |= POLLIN;
88	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
89
90	return events;
91}
92
93static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
94			    loff_t *ppos)
95{
96	struct timerfd_ctx *ctx = file->private_data;
97	ssize_t res;
98	u32 ticks = 0;
99	DECLARE_WAITQUEUE(wait, current);
100
101	if (count < sizeof(ticks))
102		return -EINVAL;
103	spin_lock_irq(&ctx->wqh.lock);
104	res = -EAGAIN;
105	if (!ctx->expired && !(file->f_flags & O_NONBLOCK)) {
106		__add_wait_queue(&ctx->wqh, &wait);
107		for (res = 0;;) {
108			set_current_state(TASK_INTERRUPTIBLE);
109			if (ctx->expired) {
110				res = 0;
111				break;
112			}
113			if (signal_pending(current)) {
114				res = -ERESTARTSYS;
115				break;
116			}
117			spin_unlock_irq(&ctx->wqh.lock);
118			schedule();
119			spin_lock_irq(&ctx->wqh.lock);
120		}
121		__remove_wait_queue(&ctx->wqh, &wait);
122		__set_current_state(TASK_RUNNING);
123	}
124	if (ctx->expired) {
125		ctx->expired = 0;
126		if (ctx->tintv.tv64 != 0) {
127			/*
128			 * If tintv.tv64 != 0, this is a periodic timer that
129			 * needs to be re-armed. We avoid doing it in the timer
130			 * callback to avoid DoS attacks specifying a very
131			 * short timer period.
132			 */
133			ticks = (u32)
134				hrtimer_forward(&ctx->tmr,
135						hrtimer_cb_get_time(&ctx->tmr),
136						ctx->tintv);
137			hrtimer_restart(&ctx->tmr);
138		} else
139			ticks = 1;
140	}
141	spin_unlock_irq(&ctx->wqh.lock);
142	if (ticks)
143		res = put_user(ticks, buf) ? -EFAULT: sizeof(ticks);
144	return res;
145}
146
147static const struct file_operations timerfd_fops = {
148	.release	= timerfd_release,
149	.poll		= timerfd_poll,
150	.read		= timerfd_read,
151};
152
153asmlinkage long sys_timerfd(int ufd, int clockid, int flags,
154			    const struct itimerspec __user *utmr)
155{
156	int error;
157	struct timerfd_ctx *ctx;
158	struct file *file;
159	struct inode *inode;
160	struct itimerspec ktmr;
161
162	if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
163		return -EFAULT;
164
165	if (clockid != CLOCK_MONOTONIC &&
166	    clockid != CLOCK_REALTIME)
167		return -EINVAL;
168	if (!timespec_valid(&ktmr.it_value) ||
169	    !timespec_valid(&ktmr.it_interval))
170		return -EINVAL;
171
172	if (ufd == -1) {
173		ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
174		if (!ctx)
175			return -ENOMEM;
176
177		init_waitqueue_head(&ctx->wqh);
178
179		timerfd_setup(ctx, clockid, flags, &ktmr);
180
181		/*
182		 * When we call this, the initialization must be complete, since
183		 * anon_inode_getfd() will install the fd.
184		 */
185		error = anon_inode_getfd(&ufd, &inode, &file, "[timerfd]",
186					 &timerfd_fops, ctx);
187		if (error)
188			goto err_tmrcancel;
189	} else {
190		file = fget(ufd);
191		if (!file)
192			return -EBADF;
193		ctx = file->private_data;
194		if (file->f_op != &timerfd_fops) {
195			fput(file);
196			return -EINVAL;
197		}
198		/*
199		 * We need to stop the existing timer before reprogramming
200		 * it to the new values.
201		 */
202		for (;;) {
203			spin_lock_irq(&ctx->wqh.lock);
204			if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
205				break;
206			spin_unlock_irq(&ctx->wqh.lock);
207			cpu_relax();
208		}
209		/*
210		 * Re-program the timer to the new value ...
211		 */
212		timerfd_setup(ctx, clockid, flags, &ktmr);
213
214		spin_unlock_irq(&ctx->wqh.lock);
215		fput(file);
216	}
217
218	return ufd;
219
220err_tmrcancel:
221	hrtimer_cancel(&ctx->tmr);
222	kfree(ctx);
223	return error;
224}
225