blob: 98158de91d24a89146b4a0cbe45c0120e511a914 [file] [log] [blame]
Davide Libenzib215e282007-05-10 22:23:16 -07001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090017#include <linux/slab.h>
Davide Libenzib215e282007-05-10 22:23:16 -070018#include <linux/list.h>
19#include <linux/spinlock.h>
20#include <linux/time.h>
21#include <linux/hrtimer.h>
22#include <linux/anon_inodes.h>
23#include <linux/timerfd.h>
Adrian Bunk45cc2b92008-04-29 00:58:59 -070024#include <linux/syscalls.h>
Davide Libenzib215e282007-05-10 22:23:16 -070025
26struct timerfd_ctx {
27 struct hrtimer tmr;
28 ktime_t tintv;
Davide Libenzib215e282007-05-10 22:23:16 -070029 wait_queue_head_t wqh;
Davide Libenzi4d672e72008-02-04 22:27:26 -080030 u64 ticks;
Davide Libenzib215e282007-05-10 22:23:16 -070031 int expired;
Davide Libenzi4d672e72008-02-04 22:27:26 -080032 int clockid;
Davide Libenzib215e282007-05-10 22:23:16 -070033};
34
35/*
36 * This gets called when the timer event triggers. We set the "expired"
37 * flag, but we do not re-arm the timer (in case it's necessary,
Davide Libenzi4d672e72008-02-04 22:27:26 -080038 * tintv.tv64 != 0) until the timer is accessed.
Davide Libenzib215e282007-05-10 22:23:16 -070039 */
40static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
41{
42 struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx, tmr);
43 unsigned long flags;
44
Davide Libenzi18963c02007-05-18 12:02:33 -070045 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzib215e282007-05-10 22:23:16 -070046 ctx->expired = 1;
Davide Libenzi4d672e72008-02-04 22:27:26 -080047 ctx->ticks++;
Davide Libenzib215e282007-05-10 22:23:16 -070048 wake_up_locked(&ctx->wqh);
Davide Libenzi18963c02007-05-18 12:02:33 -070049 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzib215e282007-05-10 22:23:16 -070050
51 return HRTIMER_NORESTART;
52}
53
Davide Libenzi4d672e72008-02-04 22:27:26 -080054static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
55{
Arjan van de Ven76369472008-09-01 15:00:14 -070056 ktime_t remaining;
Davide Libenzi4d672e72008-02-04 22:27:26 -080057
Arjan van de Ven76369472008-09-01 15:00:14 -070058 remaining = hrtimer_expires_remaining(&ctx->tmr);
Davide Libenzi4d672e72008-02-04 22:27:26 -080059 return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
60}
61
62static void timerfd_setup(struct timerfd_ctx *ctx, int flags,
Davide Libenzib215e282007-05-10 22:23:16 -070063 const struct itimerspec *ktmr)
64{
65 enum hrtimer_mode htmode;
66 ktime_t texp;
67
68 htmode = (flags & TFD_TIMER_ABSTIME) ?
69 HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
70
71 texp = timespec_to_ktime(ktmr->it_value);
72 ctx->expired = 0;
Davide Libenzi4d672e72008-02-04 22:27:26 -080073 ctx->ticks = 0;
Davide Libenzib215e282007-05-10 22:23:16 -070074 ctx->tintv = timespec_to_ktime(ktmr->it_interval);
Davide Libenzi4d672e72008-02-04 22:27:26 -080075 hrtimer_init(&ctx->tmr, ctx->clockid, htmode);
Arjan van de Ven76369472008-09-01 15:00:14 -070076 hrtimer_set_expires(&ctx->tmr, texp);
Davide Libenzib215e282007-05-10 22:23:16 -070077 ctx->tmr.function = timerfd_tmrproc;
78 if (texp.tv64 != 0)
79 hrtimer_start(&ctx->tmr, texp, htmode);
80}
81
82static int timerfd_release(struct inode *inode, struct file *file)
83{
84 struct timerfd_ctx *ctx = file->private_data;
85
86 hrtimer_cancel(&ctx->tmr);
87 kfree(ctx);
88 return 0;
89}
90
91static unsigned int timerfd_poll(struct file *file, poll_table *wait)
92{
93 struct timerfd_ctx *ctx = file->private_data;
94 unsigned int events = 0;
95 unsigned long flags;
96
97 poll_wait(file, &ctx->wqh, wait);
98
Davide Libenzi18963c02007-05-18 12:02:33 -070099 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800100 if (ctx->ticks)
Davide Libenzib215e282007-05-10 22:23:16 -0700101 events |= POLLIN;
Davide Libenzi18963c02007-05-18 12:02:33 -0700102 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzib215e282007-05-10 22:23:16 -0700103
104 return events;
105}
106
107static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
108 loff_t *ppos)
109{
110 struct timerfd_ctx *ctx = file->private_data;
111 ssize_t res;
Davide Libenzi09828402007-07-26 10:41:07 -0700112 u64 ticks = 0;
Davide Libenzib215e282007-05-10 22:23:16 -0700113 DECLARE_WAITQUEUE(wait, current);
114
115 if (count < sizeof(ticks))
116 return -EINVAL;
Davide Libenzi18963c02007-05-18 12:02:33 -0700117 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700118 res = -EAGAIN;
Davide Libenzi4d672e72008-02-04 22:27:26 -0800119 if (!ctx->ticks && !(file->f_flags & O_NONBLOCK)) {
Davide Libenzib215e282007-05-10 22:23:16 -0700120 __add_wait_queue(&ctx->wqh, &wait);
121 for (res = 0;;) {
122 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800123 if (ctx->ticks) {
Davide Libenzib215e282007-05-10 22:23:16 -0700124 res = 0;
125 break;
126 }
127 if (signal_pending(current)) {
128 res = -ERESTARTSYS;
129 break;
130 }
Davide Libenzi18963c02007-05-18 12:02:33 -0700131 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700132 schedule();
Davide Libenzi18963c02007-05-18 12:02:33 -0700133 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700134 }
135 __remove_wait_queue(&ctx->wqh, &wait);
136 __set_current_state(TASK_RUNNING);
137 }
Davide Libenzi4d672e72008-02-04 22:27:26 -0800138 if (ctx->ticks) {
139 ticks = ctx->ticks;
140 if (ctx->expired && ctx->tintv.tv64) {
Davide Libenzib215e282007-05-10 22:23:16 -0700141 /*
142 * If tintv.tv64 != 0, this is a periodic timer that
143 * needs to be re-armed. We avoid doing it in the timer
144 * callback to avoid DoS attacks specifying a very
145 * short timer period.
146 */
Davide Libenzi4d672e72008-02-04 22:27:26 -0800147 ticks += hrtimer_forward_now(&ctx->tmr,
148 ctx->tintv) - 1;
Davide Libenzib215e282007-05-10 22:23:16 -0700149 hrtimer_restart(&ctx->tmr);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800150 }
151 ctx->expired = 0;
152 ctx->ticks = 0;
Davide Libenzib215e282007-05-10 22:23:16 -0700153 }
Davide Libenzi18963c02007-05-18 12:02:33 -0700154 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700155 if (ticks)
Davide Libenzi09828402007-07-26 10:41:07 -0700156 res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
Davide Libenzib215e282007-05-10 22:23:16 -0700157 return res;
158}
159
160static const struct file_operations timerfd_fops = {
161 .release = timerfd_release,
162 .poll = timerfd_poll,
163 .read = timerfd_read,
164};
165
Davide Libenzi4d672e72008-02-04 22:27:26 -0800166static struct file *timerfd_fget(int fd)
Davide Libenzib215e282007-05-10 22:23:16 -0700167{
Davide Libenzi4d672e72008-02-04 22:27:26 -0800168 struct file *file;
169
170 file = fget(fd);
171 if (!file)
172 return ERR_PTR(-EBADF);
173 if (file->f_op != &timerfd_fops) {
174 fput(file);
175 return ERR_PTR(-EINVAL);
176 }
177
178 return file;
179}
180
Heiko Carstens836f92a2009-01-14 14:14:33 +0100181SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
Davide Libenzi4d672e72008-02-04 22:27:26 -0800182{
Al Viro2030a422008-02-23 06:46:49 -0500183 int ufd;
Davide Libenzib215e282007-05-10 22:23:16 -0700184 struct timerfd_ctx *ctx;
Davide Libenzi4d672e72008-02-04 22:27:26 -0800185
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700186 /* Check the TFD_* constants for consistency. */
187 BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
188 BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
189
Davide Libenzi610d18f2009-02-18 14:48:18 -0800190 if ((flags & ~TFD_CREATE_FLAGS) ||
191 (clockid != CLOCK_MONOTONIC &&
192 clockid != CLOCK_REALTIME))
Davide Libenzi4d672e72008-02-04 22:27:26 -0800193 return -EINVAL;
194
195 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
196 if (!ctx)
197 return -ENOMEM;
198
199 init_waitqueue_head(&ctx->wqh);
200 ctx->clockid = clockid;
201 hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
202
Ulrich Drepper11fcb6c2008-07-23 21:29:26 -0700203 ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
Roland Dreier628ff7c2009-12-18 09:41:24 -0800204 O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
Al Viro2030a422008-02-23 06:46:49 -0500205 if (ufd < 0)
Davide Libenzi4d672e72008-02-04 22:27:26 -0800206 kfree(ctx);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800207
208 return ufd;
209}
210
Heiko Carstens836f92a2009-01-14 14:14:33 +0100211SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
212 const struct itimerspec __user *, utmr,
213 struct itimerspec __user *, otmr)
Davide Libenzi4d672e72008-02-04 22:27:26 -0800214{
215 struct file *file;
216 struct timerfd_ctx *ctx;
217 struct itimerspec ktmr, kotmr;
Davide Libenzib215e282007-05-10 22:23:16 -0700218
219 if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
220 return -EFAULT;
221
Davide Libenzi610d18f2009-02-18 14:48:18 -0800222 if ((flags & ~TFD_SETTIME_FLAGS) ||
223 !timespec_valid(&ktmr.it_value) ||
Davide Libenzib215e282007-05-10 22:23:16 -0700224 !timespec_valid(&ktmr.it_interval))
225 return -EINVAL;
226
Davide Libenzi4d672e72008-02-04 22:27:26 -0800227 file = timerfd_fget(ufd);
228 if (IS_ERR(file))
229 return PTR_ERR(file);
230 ctx = file->private_data;
Davide Libenzib215e282007-05-10 22:23:16 -0700231
Davide Libenzi4d672e72008-02-04 22:27:26 -0800232 /*
233 * We need to stop the existing timer before reprogramming
234 * it to the new values.
235 */
236 for (;;) {
237 spin_lock_irq(&ctx->wqh.lock);
238 if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
239 break;
Davide Libenzi18963c02007-05-18 12:02:33 -0700240 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800241 cpu_relax();
Davide Libenzib215e282007-05-10 22:23:16 -0700242 }
243
Davide Libenzi4d672e72008-02-04 22:27:26 -0800244 /*
245 * If the timer is expired and it's periodic, we need to advance it
246 * because the caller may want to know the previous expiration time.
247 * We do not update "ticks" and "expired" since the timer will be
248 * re-programmed again in the following timerfd_setup() call.
249 */
250 if (ctx->expired && ctx->tintv.tv64)
251 hrtimer_forward_now(&ctx->tmr, ctx->tintv);
Davide Libenzib215e282007-05-10 22:23:16 -0700252
Davide Libenzi4d672e72008-02-04 22:27:26 -0800253 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
254 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
255
256 /*
257 * Re-program the timer to the new value ...
258 */
259 timerfd_setup(ctx, flags, &ktmr);
260
261 spin_unlock_irq(&ctx->wqh.lock);
262 fput(file);
263 if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
264 return -EFAULT;
265
266 return 0;
267}
268
Heiko Carstensd4e82042009-01-14 14:14:34 +0100269SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
Davide Libenzi4d672e72008-02-04 22:27:26 -0800270{
271 struct file *file;
272 struct timerfd_ctx *ctx;
273 struct itimerspec kotmr;
274
275 file = timerfd_fget(ufd);
276 if (IS_ERR(file))
277 return PTR_ERR(file);
278 ctx = file->private_data;
279
280 spin_lock_irq(&ctx->wqh.lock);
281 if (ctx->expired && ctx->tintv.tv64) {
282 ctx->expired = 0;
283 ctx->ticks +=
284 hrtimer_forward_now(&ctx->tmr, ctx->tintv) - 1;
285 hrtimer_restart(&ctx->tmr);
286 }
287 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
288 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
289 spin_unlock_irq(&ctx->wqh.lock);
290 fput(file);
291
292 return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
Davide Libenzib215e282007-05-10 22:23:16 -0700293}
294