blob: 0862f0e49d0cf8b5af985a2ea46991d25b6b9eaf [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>
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>
Adrian Bunk45cc2b92008-04-29 00:58:59 -070023#include <linux/syscalls.h>
Davide Libenzib215e282007-05-10 22:23:16 -070024
25struct timerfd_ctx {
26 struct hrtimer tmr;
27 ktime_t tintv;
Davide Libenzib215e282007-05-10 22:23:16 -070028 wait_queue_head_t wqh;
Davide Libenzi4d672e72008-02-04 22:27:26 -080029 u64 ticks;
Davide Libenzib215e282007-05-10 22:23:16 -070030 int expired;
Davide Libenzi4d672e72008-02-04 22:27:26 -080031 int clockid;
Davide Libenzib215e282007-05-10 22:23:16 -070032};
33
34/*
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,
Davide Libenzi4d672e72008-02-04 22:27:26 -080037 * tintv.tv64 != 0) until the timer is accessed.
Davide Libenzib215e282007-05-10 22:23:16 -070038 */
39static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
40{
41 struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx, tmr);
42 unsigned long flags;
43
Davide Libenzi18963c02007-05-18 12:02:33 -070044 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzib215e282007-05-10 22:23:16 -070045 ctx->expired = 1;
Davide Libenzi4d672e72008-02-04 22:27:26 -080046 ctx->ticks++;
Davide Libenzib215e282007-05-10 22:23:16 -070047 wake_up_locked(&ctx->wqh);
Davide Libenzi18963c02007-05-18 12:02:33 -070048 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzib215e282007-05-10 22:23:16 -070049
50 return HRTIMER_NORESTART;
51}
52
Davide Libenzi4d672e72008-02-04 22:27:26 -080053static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
54{
Arjan van de Ven76369472008-09-01 15:00:14 -070055 ktime_t remaining;
Davide Libenzi4d672e72008-02-04 22:27:26 -080056
Arjan van de Ven76369472008-09-01 15:00:14 -070057 remaining = hrtimer_expires_remaining(&ctx->tmr);
Davide Libenzi4d672e72008-02-04 22:27:26 -080058 return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
59}
60
61static void timerfd_setup(struct timerfd_ctx *ctx, int flags,
Davide Libenzib215e282007-05-10 22:23:16 -070062 const struct itimerspec *ktmr)
63{
64 enum hrtimer_mode htmode;
65 ktime_t texp;
66
67 htmode = (flags & TFD_TIMER_ABSTIME) ?
68 HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
69
70 texp = timespec_to_ktime(ktmr->it_value);
71 ctx->expired = 0;
Davide Libenzi4d672e72008-02-04 22:27:26 -080072 ctx->ticks = 0;
Davide Libenzib215e282007-05-10 22:23:16 -070073 ctx->tintv = timespec_to_ktime(ktmr->it_interval);
Davide Libenzi4d672e72008-02-04 22:27:26 -080074 hrtimer_init(&ctx->tmr, ctx->clockid, htmode);
Arjan van de Ven76369472008-09-01 15:00:14 -070075 hrtimer_set_expires(&ctx->tmr, texp);
Davide Libenzib215e282007-05-10 22:23:16 -070076 ctx->tmr.function = timerfd_tmrproc;
77 if (texp.tv64 != 0)
78 hrtimer_start(&ctx->tmr, texp, htmode);
79}
80
81static int timerfd_release(struct inode *inode, struct file *file)
82{
83 struct timerfd_ctx *ctx = file->private_data;
84
85 hrtimer_cancel(&ctx->tmr);
86 kfree(ctx);
87 return 0;
88}
89
90static unsigned int timerfd_poll(struct file *file, poll_table *wait)
91{
92 struct timerfd_ctx *ctx = file->private_data;
93 unsigned int events = 0;
94 unsigned long flags;
95
96 poll_wait(file, &ctx->wqh, wait);
97
Davide Libenzi18963c02007-05-18 12:02:33 -070098 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzi4d672e72008-02-04 22:27:26 -080099 if (ctx->ticks)
Davide Libenzib215e282007-05-10 22:23:16 -0700100 events |= POLLIN;
Davide Libenzi18963c02007-05-18 12:02:33 -0700101 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzib215e282007-05-10 22:23:16 -0700102
103 return events;
104}
105
106static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
107 loff_t *ppos)
108{
109 struct timerfd_ctx *ctx = file->private_data;
110 ssize_t res;
Davide Libenzi09828402007-07-26 10:41:07 -0700111 u64 ticks = 0;
Davide Libenzib215e282007-05-10 22:23:16 -0700112 DECLARE_WAITQUEUE(wait, current);
113
114 if (count < sizeof(ticks))
115 return -EINVAL;
Davide Libenzi18963c02007-05-18 12:02:33 -0700116 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700117 res = -EAGAIN;
Davide Libenzi4d672e72008-02-04 22:27:26 -0800118 if (!ctx->ticks && !(file->f_flags & O_NONBLOCK)) {
Davide Libenzib215e282007-05-10 22:23:16 -0700119 __add_wait_queue(&ctx->wqh, &wait);
120 for (res = 0;;) {
121 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800122 if (ctx->ticks) {
Davide Libenzib215e282007-05-10 22:23:16 -0700123 res = 0;
124 break;
125 }
126 if (signal_pending(current)) {
127 res = -ERESTARTSYS;
128 break;
129 }
Davide Libenzi18963c02007-05-18 12:02:33 -0700130 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700131 schedule();
Davide Libenzi18963c02007-05-18 12:02:33 -0700132 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700133 }
134 __remove_wait_queue(&ctx->wqh, &wait);
135 __set_current_state(TASK_RUNNING);
136 }
Davide Libenzi4d672e72008-02-04 22:27:26 -0800137 if (ctx->ticks) {
138 ticks = ctx->ticks;
139 if (ctx->expired && ctx->tintv.tv64) {
Davide Libenzib215e282007-05-10 22:23:16 -0700140 /*
141 * If tintv.tv64 != 0, this is a periodic timer that
142 * needs to be re-armed. We avoid doing it in the timer
143 * callback to avoid DoS attacks specifying a very
144 * short timer period.
145 */
Davide Libenzi4d672e72008-02-04 22:27:26 -0800146 ticks += hrtimer_forward_now(&ctx->tmr,
147 ctx->tintv) - 1;
Davide Libenzib215e282007-05-10 22:23:16 -0700148 hrtimer_restart(&ctx->tmr);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800149 }
150 ctx->expired = 0;
151 ctx->ticks = 0;
Davide Libenzib215e282007-05-10 22:23:16 -0700152 }
Davide Libenzi18963c02007-05-18 12:02:33 -0700153 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzib215e282007-05-10 22:23:16 -0700154 if (ticks)
Davide Libenzi09828402007-07-26 10:41:07 -0700155 res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
Davide Libenzib215e282007-05-10 22:23:16 -0700156 return res;
157}
158
159static const struct file_operations timerfd_fops = {
160 .release = timerfd_release,
161 .poll = timerfd_poll,
162 .read = timerfd_read,
163};
164
Davide Libenzi4d672e72008-02-04 22:27:26 -0800165static struct file *timerfd_fget(int fd)
Davide Libenzib215e282007-05-10 22:23:16 -0700166{
Davide Libenzi4d672e72008-02-04 22:27:26 -0800167 struct file *file;
168
169 file = fget(fd);
170 if (!file)
171 return ERR_PTR(-EBADF);
172 if (file->f_op != &timerfd_fops) {
173 fput(file);
174 return ERR_PTR(-EINVAL);
175 }
176
177 return file;
178}
179
180asmlinkage long sys_timerfd_create(int clockid, int flags)
181{
Al Viro2030a422008-02-23 06:46:49 -0500182 int ufd;
Davide Libenzib215e282007-05-10 22:23:16 -0700183 struct timerfd_ctx *ctx;
Davide Libenzi4d672e72008-02-04 22:27:26 -0800184
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700185 /* Check the TFD_* constants for consistency. */
186 BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
187 BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
188
Ulrich Drepper6b1ef0e2008-07-23 21:29:39 -0700189 if (flags & ~(TFD_CLOEXEC | TFD_NONBLOCK))
Davide Libenzi4d672e72008-02-04 22:27:26 -0800190 return -EINVAL;
191 if (clockid != CLOCK_MONOTONIC &&
192 clockid != CLOCK_REALTIME)
193 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,
Ulrich Drepper6b1ef0e2008-07-23 21:29:39 -0700204 flags & (O_CLOEXEC | O_NONBLOCK));
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
211asmlinkage long sys_timerfd_settime(int ufd, int flags,
212 const struct itimerspec __user *utmr,
213 struct itimerspec __user *otmr)
214{
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 Libenzib215e282007-05-10 22:23:16 -0700222 if (!timespec_valid(&ktmr.it_value) ||
223 !timespec_valid(&ktmr.it_interval))
224 return -EINVAL;
225
Davide Libenzi4d672e72008-02-04 22:27:26 -0800226 file = timerfd_fget(ufd);
227 if (IS_ERR(file))
228 return PTR_ERR(file);
229 ctx = file->private_data;
Davide Libenzib215e282007-05-10 22:23:16 -0700230
Davide Libenzi4d672e72008-02-04 22:27:26 -0800231 /*
232 * We need to stop the existing timer before reprogramming
233 * it to the new values.
234 */
235 for (;;) {
236 spin_lock_irq(&ctx->wqh.lock);
237 if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
238 break;
Davide Libenzi18963c02007-05-18 12:02:33 -0700239 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzi4d672e72008-02-04 22:27:26 -0800240 cpu_relax();
Davide Libenzib215e282007-05-10 22:23:16 -0700241 }
242
Davide Libenzi4d672e72008-02-04 22:27:26 -0800243 /*
244 * If the timer is expired and it's periodic, we need to advance it
245 * because the caller may want to know the previous expiration time.
246 * We do not update "ticks" and "expired" since the timer will be
247 * re-programmed again in the following timerfd_setup() call.
248 */
249 if (ctx->expired && ctx->tintv.tv64)
250 hrtimer_forward_now(&ctx->tmr, ctx->tintv);
Davide Libenzib215e282007-05-10 22:23:16 -0700251
Davide Libenzi4d672e72008-02-04 22:27:26 -0800252 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
253 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
254
255 /*
256 * Re-program the timer to the new value ...
257 */
258 timerfd_setup(ctx, flags, &ktmr);
259
260 spin_unlock_irq(&ctx->wqh.lock);
261 fput(file);
262 if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
263 return -EFAULT;
264
265 return 0;
266}
267
268asmlinkage long sys_timerfd_gettime(int ufd, struct itimerspec __user *otmr)
269{
270 struct file *file;
271 struct timerfd_ctx *ctx;
272 struct itimerspec kotmr;
273
274 file = timerfd_fget(ufd);
275 if (IS_ERR(file))
276 return PTR_ERR(file);
277 ctx = file->private_data;
278
279 spin_lock_irq(&ctx->wqh.lock);
280 if (ctx->expired && ctx->tintv.tv64) {
281 ctx->expired = 0;
282 ctx->ticks +=
283 hrtimer_forward_now(&ctx->tmr, ctx->tintv) - 1;
284 hrtimer_restart(&ctx->tmr);
285 }
286 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
287 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
288 spin_unlock_irq(&ctx->wqh.lock);
289 fput(file);
290
291 return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
Davide Libenzib215e282007-05-10 22:23:16 -0700292}
293