Linux 3.0-rc3
[linux-block.git] / fs / timerfd.c
CommitLineData
b215e283
DL
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>
5a0e3ad6 17#include <linux/slab.h>
b215e283
DL
18#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>
45cc2b96 24#include <linux/syscalls.h>
9ec26907 25#include <linux/rcupdate.h>
b215e283
DL
26
27struct timerfd_ctx {
28 struct hrtimer tmr;
29 ktime_t tintv;
99ee5315 30 ktime_t moffs;
b215e283 31 wait_queue_head_t wqh;
4d672e7a 32 u64 ticks;
b215e283 33 int expired;
4d672e7a 34 int clockid;
9ec26907
TG
35 struct rcu_head rcu;
36 struct list_head clist;
99ee5315 37 bool might_cancel;
b215e283
DL
38};
39
9ec26907
TG
40static LIST_HEAD(cancel_list);
41static DEFINE_SPINLOCK(cancel_lock);
42
b215e283
DL
43/*
44 * This gets called when the timer event triggers. We set the "expired"
45 * flag, but we do not re-arm the timer (in case it's necessary,
4d672e7a 46 * tintv.tv64 != 0) until the timer is accessed.
b215e283
DL
47 */
48static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
49{
50 struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx, tmr);
51 unsigned long flags;
52
18963c01 53 spin_lock_irqsave(&ctx->wqh.lock, flags);
b215e283 54 ctx->expired = 1;
4d672e7a 55 ctx->ticks++;
b215e283 56 wake_up_locked(&ctx->wqh);
18963c01 57 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
b215e283
DL
58
59 return HRTIMER_NORESTART;
60}
61
9ec26907
TG
62/*
63 * Called when the clock was set to cancel the timers in the cancel
64 * list.
65 */
66void timerfd_clock_was_set(void)
4d672e7a 67{
9ec26907
TG
68 ktime_t moffs = ktime_get_monotonic_offset();
69 struct timerfd_ctx *ctx;
70 unsigned long flags;
4d672e7a 71
9ec26907
TG
72 rcu_read_lock();
73 list_for_each_entry_rcu(ctx, &cancel_list, clist) {
74 if (!ctx->might_cancel)
75 continue;
76 spin_lock_irqsave(&ctx->wqh.lock, flags);
77 if (ctx->moffs.tv64 != moffs.tv64) {
78 ctx->moffs.tv64 = KTIME_MAX;
79 wake_up_locked(&ctx->wqh);
80 }
81 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
82 }
83 rcu_read_unlock();
4d672e7a
DL
84}
85
9ec26907 86static void timerfd_remove_cancel(struct timerfd_ctx *ctx)
99ee5315 87{
9ec26907
TG
88 if (ctx->might_cancel) {
89 ctx->might_cancel = false;
90 spin_lock(&cancel_lock);
91 list_del_rcu(&ctx->clist);
92 spin_unlock(&cancel_lock);
93 }
94}
99ee5315 95
9ec26907
TG
96static bool timerfd_canceled(struct timerfd_ctx *ctx)
97{
98 if (!ctx->might_cancel || ctx->moffs.tv64 != KTIME_MAX)
99ee5315 99 return false;
9ec26907
TG
100 ctx->moffs = ktime_get_monotonic_offset();
101 return true;
102}
99ee5315 103
9ec26907
TG
104static void timerfd_setup_cancel(struct timerfd_ctx *ctx, int flags)
105{
106 if (ctx->clockid == CLOCK_REALTIME && (flags & TFD_TIMER_ABSTIME) &&
107 (flags & TFD_TIMER_CANCEL_ON_SET)) {
108 if (!ctx->might_cancel) {
109 ctx->might_cancel = true;
110 spin_lock(&cancel_lock);
111 list_add_rcu(&ctx->clist, &cancel_list);
112 spin_unlock(&cancel_lock);
113 }
114 } else if (ctx->might_cancel) {
115 timerfd_remove_cancel(ctx);
116 }
117}
99ee5315 118
9ec26907
TG
119static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
120{
121 ktime_t remaining;
99ee5315 122
9ec26907
TG
123 remaining = hrtimer_expires_remaining(&ctx->tmr);
124 return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
99ee5315
TG
125}
126
127static int timerfd_setup(struct timerfd_ctx *ctx, int flags,
128 const struct itimerspec *ktmr)
b215e283
DL
129{
130 enum hrtimer_mode htmode;
131 ktime_t texp;
99ee5315 132 int clockid = ctx->clockid;
b215e283
DL
133
134 htmode = (flags & TFD_TIMER_ABSTIME) ?
135 HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
136
137 texp = timespec_to_ktime(ktmr->it_value);
138 ctx->expired = 0;
4d672e7a 139 ctx->ticks = 0;
b215e283 140 ctx->tintv = timespec_to_ktime(ktmr->it_interval);
99ee5315 141 hrtimer_init(&ctx->tmr, clockid, htmode);
76369470 142 hrtimer_set_expires(&ctx->tmr, texp);
b215e283 143 ctx->tmr.function = timerfd_tmrproc;
99ee5315 144 if (texp.tv64 != 0) {
b215e283 145 hrtimer_start(&ctx->tmr, texp, htmode);
99ee5315
TG
146 if (timerfd_canceled(ctx))
147 return -ECANCELED;
148 }
149 return 0;
b215e283
DL
150}
151
152static int timerfd_release(struct inode *inode, struct file *file)
153{
154 struct timerfd_ctx *ctx = file->private_data;
155
9ec26907 156 timerfd_remove_cancel(ctx);
b215e283 157 hrtimer_cancel(&ctx->tmr);
9ec26907 158 kfree_rcu(ctx, rcu);
b215e283
DL
159 return 0;
160}
161
162static unsigned int timerfd_poll(struct file *file, poll_table *wait)
163{
164 struct timerfd_ctx *ctx = file->private_data;
165 unsigned int events = 0;
166 unsigned long flags;
167
168 poll_wait(file, &ctx->wqh, wait);
169
18963c01 170 spin_lock_irqsave(&ctx->wqh.lock, flags);
4d672e7a 171 if (ctx->ticks)
b215e283 172 events |= POLLIN;
18963c01 173 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
b215e283
DL
174
175 return events;
176}
177
178static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
179 loff_t *ppos)
180{
181 struct timerfd_ctx *ctx = file->private_data;
182 ssize_t res;
09828402 183 u64 ticks = 0;
b215e283
DL
184
185 if (count < sizeof(ticks))
186 return -EINVAL;
18963c01 187 spin_lock_irq(&ctx->wqh.lock);
8120a8aa
MN
188 if (file->f_flags & O_NONBLOCK)
189 res = -EAGAIN;
190 else
191 res = wait_event_interruptible_locked_irq(ctx->wqh, ctx->ticks);
99ee5315 192
9ec26907
TG
193 /*
194 * If clock has changed, we do not care about the
195 * ticks and we do not rearm the timer. Userspace must
196 * reevaluate anyway.
197 */
198 if (timerfd_canceled(ctx)) {
199 ctx->ticks = 0;
200 ctx->expired = 0;
201 res = -ECANCELED;
202 }
203
4d672e7a
DL
204 if (ctx->ticks) {
205 ticks = ctx->ticks;
99ee5315 206
4d672e7a 207 if (ctx->expired && ctx->tintv.tv64) {
b215e283
DL
208 /*
209 * If tintv.tv64 != 0, this is a periodic timer that
210 * needs to be re-armed. We avoid doing it in the timer
211 * callback to avoid DoS attacks specifying a very
212 * short timer period.
213 */
4d672e7a
DL
214 ticks += hrtimer_forward_now(&ctx->tmr,
215 ctx->tintv) - 1;
b215e283 216 hrtimer_restart(&ctx->tmr);
4d672e7a
DL
217 }
218 ctx->expired = 0;
219 ctx->ticks = 0;
b215e283 220 }
18963c01 221 spin_unlock_irq(&ctx->wqh.lock);
b215e283 222 if (ticks)
09828402 223 res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
b215e283
DL
224 return res;
225}
226
227static const struct file_operations timerfd_fops = {
228 .release = timerfd_release,
229 .poll = timerfd_poll,
230 .read = timerfd_read,
6038f373 231 .llseek = noop_llseek,
b215e283
DL
232};
233
4d672e7a
DL
234static struct file *timerfd_fget(int fd)
235{
236 struct file *file;
237
238 file = fget(fd);
239 if (!file)
240 return ERR_PTR(-EBADF);
241 if (file->f_op != &timerfd_fops) {
242 fput(file);
243 return ERR_PTR(-EINVAL);
244 }
245
246 return file;
247}
248
836f92ad 249SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
b215e283 250{
2030a42c 251 int ufd;
b215e283 252 struct timerfd_ctx *ctx;
b215e283 253
e38b36f3
UD
254 /* Check the TFD_* constants for consistency. */
255 BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
256 BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
257
610d18f4
DL
258 if ((flags & ~TFD_CREATE_FLAGS) ||
259 (clockid != CLOCK_MONOTONIC &&
260 clockid != CLOCK_REALTIME))
b215e283 261 return -EINVAL;
4d672e7a
DL
262
263 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
264 if (!ctx)
265 return -ENOMEM;
266
267 init_waitqueue_head(&ctx->wqh);
268 ctx->clockid = clockid;
269 hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
99ee5315 270 ctx->moffs = ktime_get_monotonic_offset();
4d672e7a 271
11fcb6c1 272 ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
628ff7c1 273 O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
2030a42c 274 if (ufd < 0)
4d672e7a 275 kfree(ctx);
4d672e7a
DL
276
277 return ufd;
278}
279
836f92ad
HC
280SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
281 const struct itimerspec __user *, utmr,
282 struct itimerspec __user *, otmr)
4d672e7a
DL
283{
284 struct file *file;
285 struct timerfd_ctx *ctx;
286 struct itimerspec ktmr, kotmr;
99ee5315 287 int ret;
4d672e7a
DL
288
289 if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
290 return -EFAULT;
291
610d18f4
DL
292 if ((flags & ~TFD_SETTIME_FLAGS) ||
293 !timespec_valid(&ktmr.it_value) ||
b215e283
DL
294 !timespec_valid(&ktmr.it_interval))
295 return -EINVAL;
296
4d672e7a
DL
297 file = timerfd_fget(ufd);
298 if (IS_ERR(file))
299 return PTR_ERR(file);
300 ctx = file->private_data;
b215e283 301
9ec26907
TG
302 timerfd_setup_cancel(ctx, flags);
303
4d672e7a
DL
304 /*
305 * We need to stop the existing timer before reprogramming
306 * it to the new values.
307 */
308 for (;;) {
309 spin_lock_irq(&ctx->wqh.lock);
310 if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
311 break;
18963c01 312 spin_unlock_irq(&ctx->wqh.lock);
4d672e7a 313 cpu_relax();
b215e283
DL
314 }
315
4d672e7a
DL
316 /*
317 * If the timer is expired and it's periodic, we need to advance it
318 * because the caller may want to know the previous expiration time.
319 * We do not update "ticks" and "expired" since the timer will be
320 * re-programmed again in the following timerfd_setup() call.
321 */
322 if (ctx->expired && ctx->tintv.tv64)
323 hrtimer_forward_now(&ctx->tmr, ctx->tintv);
b215e283 324
4d672e7a
DL
325 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
326 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
327
328 /*
329 * Re-program the timer to the new value ...
330 */
99ee5315 331 ret = timerfd_setup(ctx, flags, &ktmr);
4d672e7a
DL
332
333 spin_unlock_irq(&ctx->wqh.lock);
334 fput(file);
335 if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
336 return -EFAULT;
337
99ee5315 338 return ret;
4d672e7a
DL
339}
340
d4e82042 341SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
4d672e7a
DL
342{
343 struct file *file;
344 struct timerfd_ctx *ctx;
345 struct itimerspec kotmr;
346
347 file = timerfd_fget(ufd);
348 if (IS_ERR(file))
349 return PTR_ERR(file);
350 ctx = file->private_data;
351
352 spin_lock_irq(&ctx->wqh.lock);
353 if (ctx->expired && ctx->tintv.tv64) {
354 ctx->expired = 0;
355 ctx->ticks +=
356 hrtimer_forward_now(&ctx->tmr, ctx->tintv) - 1;
357 hrtimer_restart(&ctx->tmr);
358 }
359 kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
360 kotmr.it_interval = ktime_to_timespec(ctx->tintv);
361 spin_unlock_irq(&ctx->wqh.lock);
362 fput(file);
363
364 return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
b215e283
DL
365}
366