From: Jonathan McDowell <noodles@earth.li>
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / timerfd.c
blob43feb6d39d9ae815bfc5d3dd90990d5766470ecd
1 /*
2 * fs/timerfd.c
4 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
7 * Thanks to Thomas Gleixner for code reviews and useful comments.
9 */
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 #include <linux/syscalls.h>
25 struct timerfd_ctx {
26 struct hrtimer tmr;
27 ktime_t tintv;
28 wait_queue_head_t wqh;
29 u64 ticks;
30 int expired;
31 int clockid;
35 * This gets called when the timer event triggers. We set the "expired"
36 * flag, but we do not re-arm the timer (in case it's necessary,
37 * tintv.tv64 != 0) until the timer is accessed.
39 static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
41 struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx, tmr);
42 unsigned long flags;
44 spin_lock_irqsave(&ctx->wqh.lock, flags);
45 ctx->expired = 1;
46 ctx->ticks++;
47 wake_up_locked(&ctx->wqh);
48 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
50 return HRTIMER_NORESTART;
53 static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
55 ktime_t now, remaining;
57 now = ctx->tmr.base->get_time();
58 remaining = ktime_sub(ctx->tmr.expires, now);
60 return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
63 static void timerfd_setup(struct timerfd_ctx *ctx, int flags,
64 const struct itimerspec *ktmr)
66 enum hrtimer_mode htmode;
67 ktime_t texp;
69 htmode = (flags & TFD_TIMER_ABSTIME) ?
70 HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
72 texp = timespec_to_ktime(ktmr->it_value);
73 ctx->expired = 0;
74 ctx->ticks = 0;
75 ctx->tintv = timespec_to_ktime(ktmr->it_interval);
76 hrtimer_init(&ctx->tmr, ctx->clockid, htmode);
77 ctx->tmr.expires = texp;
78 ctx->tmr.function = timerfd_tmrproc;
79 if (texp.tv64 != 0)
80 hrtimer_start(&ctx->tmr, texp, htmode);
83 static int timerfd_release(struct inode *inode, struct file *file)
85 struct timerfd_ctx *ctx = file->private_data;
87 hrtimer_cancel(&ctx->tmr);
88 kfree(ctx);
89 return 0;
92 static unsigned int timerfd_poll(struct file *file, poll_table *wait)
94 struct timerfd_ctx *ctx = file->private_data;
95 unsigned int events = 0;
96 unsigned long flags;
98 poll_wait(file, &ctx->wqh, wait);
100 spin_lock_irqsave(&ctx->wqh.lock, flags);
101 if (ctx->ticks)
102 events |= POLLIN;
103 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
105 return events;
108 static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
109 loff_t *ppos)
111 struct timerfd_ctx *ctx = file->private_data;
112 ssize_t res;
113 u64 ticks = 0;
114 DECLARE_WAITQUEUE(wait, current);
116 if (count < sizeof(ticks))
117 return -EINVAL;
118 spin_lock_irq(&ctx->wqh.lock);
119 res = -EAGAIN;
120 if (!ctx->ticks && !(file->f_flags & O_NONBLOCK)) {
121 __add_wait_queue(&ctx->wqh, &wait);
122 for (res = 0;;) {
123 set_current_state(TASK_INTERRUPTIBLE);
124 if (ctx->ticks) {
125 res = 0;
126 break;
128 if (signal_pending(current)) {
129 res = -ERESTARTSYS;
130 break;
132 spin_unlock_irq(&ctx->wqh.lock);
133 schedule();
134 spin_lock_irq(&ctx->wqh.lock);
136 __remove_wait_queue(&ctx->wqh, &wait);
137 __set_current_state(TASK_RUNNING);
139 if (ctx->ticks) {
140 ticks = ctx->ticks;
141 if (ctx->expired && ctx->tintv.tv64) {
143 * If tintv.tv64 != 0, this is a periodic timer that
144 * needs to be re-armed. We avoid doing it in the timer
145 * callback to avoid DoS attacks specifying a very
146 * short timer period.
148 ticks += hrtimer_forward_now(&ctx->tmr,
149 ctx->tintv) - 1;
150 hrtimer_restart(&ctx->tmr);
152 ctx->expired = 0;
153 ctx->ticks = 0;
155 spin_unlock_irq(&ctx->wqh.lock);
156 if (ticks)
157 res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
158 return res;
161 static const struct file_operations timerfd_fops = {
162 .release = timerfd_release,
163 .poll = timerfd_poll,
164 .read = timerfd_read,
167 static struct file *timerfd_fget(int fd)
169 struct file *file;
171 file = fget(fd);
172 if (!file)
173 return ERR_PTR(-EBADF);
174 if (file->f_op != &timerfd_fops) {
175 fput(file);
176 return ERR_PTR(-EINVAL);
179 return file;
182 SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
184 int ufd;
185 struct timerfd_ctx *ctx;
187 /* Check the TFD_* constants for consistency. */
188 BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
189 BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
191 if ((flags & ~TFD_CREATE_FLAGS) ||
192 (clockid != CLOCK_MONOTONIC &&
193 clockid != CLOCK_REALTIME))
194 return -EINVAL;
196 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
197 if (!ctx)
198 return -ENOMEM;
200 init_waitqueue_head(&ctx->wqh);
201 ctx->clockid = clockid;
202 hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
204 ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
205 flags & TFD_SHARED_FCNTL_FLAGS);
206 if (ufd < 0)
207 kfree(ctx);
209 return ufd;
212 SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
213 const struct itimerspec __user *, utmr,
214 struct itimerspec __user *, otmr)
216 struct file *file;
217 struct timerfd_ctx *ctx;
218 struct itimerspec ktmr, kotmr;
220 if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
221 return -EFAULT;
223 if ((flags & ~TFD_SETTIME_FLAGS) ||
224 !timespec_valid(&ktmr.it_value) ||
225 !timespec_valid(&ktmr.it_interval))
226 return -EINVAL;
228 file = timerfd_fget(ufd);
229 if (IS_ERR(file))
230 return PTR_ERR(file);
231 ctx = file->private_data;
234 * We need to stop the existing timer before reprogramming
235 * it to the new values.
237 for (;;) {
238 spin_lock_irq(&ctx->wqh.lock);
239 if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
240 break;
241 spin_unlock_irq(&ctx->wqh.lock);
242 cpu_relax();
246 * If the timer is expired and it's periodic, we need to advance it
247 * because the caller may want to know the previous expiration time.
248 * We do not update "ticks" and "expired" since the timer will be
249 * re-programmed again in the following timerfd_setup() call.
251 if (ctx->expired && ctx->tintv.tv64)
252 hrtimer_forward_now(&ctx->tmr, ctx->tintv);
254 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
255 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
258 * Re-program the timer to the new value ...
260 timerfd_setup(ctx, flags, &ktmr);
262 spin_unlock_irq(&ctx->wqh.lock);
263 fput(file);
264 if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
265 return -EFAULT;
267 return 0;
270 SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
272 struct file *file;
273 struct timerfd_ctx *ctx;
274 struct itimerspec kotmr;
276 file = timerfd_fget(ufd);
277 if (IS_ERR(file))
278 return PTR_ERR(file);
279 ctx = file->private_data;
281 spin_lock_irq(&ctx->wqh.lock);
282 if (ctx->expired && ctx->tintv.tv64) {
283 ctx->expired = 0;
284 ctx->ticks +=
285 hrtimer_forward_now(&ctx->tmr, ctx->tintv) - 1;
286 hrtimer_restart(&ctx->tmr);
288 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
289 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
290 spin_unlock_irq(&ctx->wqh.lock);
291 fput(file);
293 return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;