treewide: Use array_size() in f2fs_kzalloc()
[linux-2.6-block.git] / fs / timerfd.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
b215e283
DL
2/*
3 * fs/timerfd.c
4 *
5 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
6 *
7 *
8 * Thanks to Thomas Gleixner for code reviews and useful comments.
9 *
10 */
11
11ffa9d6 12#include <linux/alarmtimer.h>
b215e283
DL
13#include <linux/file.h>
14#include <linux/poll.h>
15#include <linux/init.h>
16#include <linux/fs.h>
17#include <linux/sched.h>
18#include <linux/kernel.h>
5a0e3ad6 19#include <linux/slab.h>
b215e283
DL
20#include <linux/list.h>
21#include <linux/spinlock.h>
22#include <linux/time.h>
23#include <linux/hrtimer.h>
24#include <linux/anon_inodes.h>
25#include <linux/timerfd.h>
45cc2b96 26#include <linux/syscalls.h>
9d94b9e2 27#include <linux/compat.h>
9ec26907 28#include <linux/rcupdate.h>
b215e283
DL
29
30struct timerfd_ctx {
11ffa9d6
TP
31 union {
32 struct hrtimer tmr;
33 struct alarm alarm;
34 } t;
b215e283 35 ktime_t tintv;
99ee5315 36 ktime_t moffs;
b215e283 37 wait_queue_head_t wqh;
4d672e7a 38 u64 ticks;
4d672e7a 39 int clockid;
af9c4957
CG
40 short unsigned expired;
41 short unsigned settime_flags; /* to show in fdinfo */
9ec26907
TG
42 struct rcu_head rcu;
43 struct list_head clist;
1e38da30 44 spinlock_t cancel_lock;
99ee5315 45 bool might_cancel;
b215e283
DL
46};
47
9ec26907
TG
48static LIST_HEAD(cancel_list);
49static DEFINE_SPINLOCK(cancel_lock);
50
11ffa9d6
TP
51static inline bool isalarm(struct timerfd_ctx *ctx)
52{
53 return ctx->clockid == CLOCK_REALTIME_ALARM ||
54 ctx->clockid == CLOCK_BOOTTIME_ALARM;
55}
56
b215e283
DL
57/*
58 * This gets called when the timer event triggers. We set the "expired"
59 * flag, but we do not re-arm the timer (in case it's necessary,
2456e855 60 * tintv != 0) until the timer is accessed.
b215e283 61 */
11ffa9d6 62static void timerfd_triggered(struct timerfd_ctx *ctx)
b215e283 63{
b215e283
DL
64 unsigned long flags;
65
18963c01 66 spin_lock_irqsave(&ctx->wqh.lock, flags);
b215e283 67 ctx->expired = 1;
4d672e7a 68 ctx->ticks++;
b215e283 69 wake_up_locked(&ctx->wqh);
18963c01 70 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
11ffa9d6 71}
b215e283 72
11ffa9d6
TP
73static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
74{
75 struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx,
76 t.tmr);
77 timerfd_triggered(ctx);
b215e283
DL
78 return HRTIMER_NORESTART;
79}
80
11ffa9d6
TP
81static enum alarmtimer_restart timerfd_alarmproc(struct alarm *alarm,
82 ktime_t now)
83{
84 struct timerfd_ctx *ctx = container_of(alarm, struct timerfd_ctx,
85 t.alarm);
86 timerfd_triggered(ctx);
87 return ALARMTIMER_NORESTART;
88}
89
9ec26907
TG
90/*
91 * Called when the clock was set to cancel the timers in the cancel
1123d939
MA
92 * list. This will wake up processes waiting on these timers. The
93 * wake-up requires ctx->ticks to be non zero, therefore we increment
94 * it before calling wake_up_locked().
9ec26907
TG
95 */
96void timerfd_clock_was_set(void)
4d672e7a 97{
2456e855 98 ktime_t moffs = ktime_mono_to_real(0);
9ec26907
TG
99 struct timerfd_ctx *ctx;
100 unsigned long flags;
4d672e7a 101
9ec26907
TG
102 rcu_read_lock();
103 list_for_each_entry_rcu(ctx, &cancel_list, clist) {
104 if (!ctx->might_cancel)
105 continue;
106 spin_lock_irqsave(&ctx->wqh.lock, flags);
2456e855
TG
107 if (ctx->moffs != moffs) {
108 ctx->moffs = KTIME_MAX;
1123d939 109 ctx->ticks++;
9ec26907
TG
110 wake_up_locked(&ctx->wqh);
111 }
112 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
113 }
114 rcu_read_unlock();
4d672e7a
DL
115}
116
1e38da30 117static void __timerfd_remove_cancel(struct timerfd_ctx *ctx)
99ee5315 118{
9ec26907
TG
119 if (ctx->might_cancel) {
120 ctx->might_cancel = false;
121 spin_lock(&cancel_lock);
122 list_del_rcu(&ctx->clist);
123 spin_unlock(&cancel_lock);
124 }
125}
99ee5315 126
1e38da30
TG
127static void timerfd_remove_cancel(struct timerfd_ctx *ctx)
128{
129 spin_lock(&ctx->cancel_lock);
130 __timerfd_remove_cancel(ctx);
131 spin_unlock(&ctx->cancel_lock);
132}
133
9ec26907
TG
134static bool timerfd_canceled(struct timerfd_ctx *ctx)
135{
2456e855 136 if (!ctx->might_cancel || ctx->moffs != KTIME_MAX)
99ee5315 137 return false;
2456e855 138 ctx->moffs = ktime_mono_to_real(0);
9ec26907
TG
139 return true;
140}
99ee5315 141
9ec26907
TG
142static void timerfd_setup_cancel(struct timerfd_ctx *ctx, int flags)
143{
1e38da30 144 spin_lock(&ctx->cancel_lock);
11ffa9d6
TP
145 if ((ctx->clockid == CLOCK_REALTIME ||
146 ctx->clockid == CLOCK_REALTIME_ALARM) &&
147 (flags & TFD_TIMER_ABSTIME) && (flags & TFD_TIMER_CANCEL_ON_SET)) {
9ec26907
TG
148 if (!ctx->might_cancel) {
149 ctx->might_cancel = true;
150 spin_lock(&cancel_lock);
151 list_add_rcu(&ctx->clist, &cancel_list);
152 spin_unlock(&cancel_lock);
153 }
1e38da30
TG
154 } else {
155 __timerfd_remove_cancel(ctx);
9ec26907 156 }
1e38da30 157 spin_unlock(&ctx->cancel_lock);
9ec26907 158}
99ee5315 159
9ec26907
TG
160static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
161{
162 ktime_t remaining;
99ee5315 163
11ffa9d6
TP
164 if (isalarm(ctx))
165 remaining = alarm_expires_remaining(&ctx->t.alarm);
166 else
b62526ed 167 remaining = hrtimer_expires_remaining_adjusted(&ctx->t.tmr);
11ffa9d6 168
8b0e1953 169 return remaining < 0 ? 0: remaining;
99ee5315
TG
170}
171
172static int timerfd_setup(struct timerfd_ctx *ctx, int flags,
bff41203 173 const struct itimerspec64 *ktmr)
b215e283
DL
174{
175 enum hrtimer_mode htmode;
176 ktime_t texp;
99ee5315 177 int clockid = ctx->clockid;
b215e283
DL
178
179 htmode = (flags & TFD_TIMER_ABSTIME) ?
180 HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
181
bff41203 182 texp = timespec64_to_ktime(ktmr->it_value);
b215e283 183 ctx->expired = 0;
4d672e7a 184 ctx->ticks = 0;
bff41203 185 ctx->tintv = timespec64_to_ktime(ktmr->it_interval);
11ffa9d6
TP
186
187 if (isalarm(ctx)) {
188 alarm_init(&ctx->t.alarm,
189 ctx->clockid == CLOCK_REALTIME_ALARM ?
190 ALARM_REALTIME : ALARM_BOOTTIME,
191 timerfd_alarmproc);
192 } else {
193 hrtimer_init(&ctx->t.tmr, clockid, htmode);
194 hrtimer_set_expires(&ctx->t.tmr, texp);
195 ctx->t.tmr.function = timerfd_tmrproc;
196 }
197
2456e855 198 if (texp != 0) {
11ffa9d6
TP
199 if (isalarm(ctx)) {
200 if (flags & TFD_TIMER_ABSTIME)
201 alarm_start(&ctx->t.alarm, texp);
202 else
203 alarm_start_relative(&ctx->t.alarm, texp);
204 } else {
205 hrtimer_start(&ctx->t.tmr, texp, htmode);
206 }
207
99ee5315
TG
208 if (timerfd_canceled(ctx))
209 return -ECANCELED;
210 }
af9c4957
CG
211
212 ctx->settime_flags = flags & TFD_SETTIME_FLAGS;
99ee5315 213 return 0;
b215e283
DL
214}
215
216static int timerfd_release(struct inode *inode, struct file *file)
217{
218 struct timerfd_ctx *ctx = file->private_data;
219
9ec26907 220 timerfd_remove_cancel(ctx);
11ffa9d6
TP
221
222 if (isalarm(ctx))
223 alarm_cancel(&ctx->t.alarm);
224 else
225 hrtimer_cancel(&ctx->t.tmr);
9ec26907 226 kfree_rcu(ctx, rcu);
b215e283
DL
227 return 0;
228}
652fe8e8
CH
229
230static struct wait_queue_head *timerfd_get_poll_head(struct file *file,
231 __poll_t eventmask)
b215e283
DL
232{
233 struct timerfd_ctx *ctx = file->private_data;
b215e283 234
652fe8e8
CH
235 return &ctx->wqh;
236}
b215e283 237
652fe8e8
CH
238static __poll_t timerfd_poll_mask(struct file *file, __poll_t eventmask)
239{
240 struct timerfd_ctx *ctx = file->private_data;
b215e283 241
652fe8e8 242 return ctx->ticks ? EPOLLIN : 0;
b215e283
DL
243}
244
245static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
246 loff_t *ppos)
247{
248 struct timerfd_ctx *ctx = file->private_data;
249 ssize_t res;
09828402 250 u64 ticks = 0;
b215e283
DL
251
252 if (count < sizeof(ticks))
253 return -EINVAL;
18963c01 254 spin_lock_irq(&ctx->wqh.lock);
8120a8aa
MN
255 if (file->f_flags & O_NONBLOCK)
256 res = -EAGAIN;
257 else
258 res = wait_event_interruptible_locked_irq(ctx->wqh, ctx->ticks);
99ee5315 259
9ec26907
TG
260 /*
261 * If clock has changed, we do not care about the
262 * ticks and we do not rearm the timer. Userspace must
263 * reevaluate anyway.
264 */
265 if (timerfd_canceled(ctx)) {
266 ctx->ticks = 0;
267 ctx->expired = 0;
268 res = -ECANCELED;
269 }
270
4d672e7a
DL
271 if (ctx->ticks) {
272 ticks = ctx->ticks;
99ee5315 273
2456e855 274 if (ctx->expired && ctx->tintv) {
b215e283 275 /*
2456e855 276 * If tintv != 0, this is a periodic timer that
b215e283
DL
277 * needs to be re-armed. We avoid doing it in the timer
278 * callback to avoid DoS attacks specifying a very
279 * short timer period.
280 */
11ffa9d6
TP
281 if (isalarm(ctx)) {
282 ticks += alarm_forward_now(
283 &ctx->t.alarm, ctx->tintv) - 1;
284 alarm_restart(&ctx->t.alarm);
285 } else {
286 ticks += hrtimer_forward_now(&ctx->t.tmr,
287 ctx->tintv) - 1;
288 hrtimer_restart(&ctx->t.tmr);
289 }
4d672e7a
DL
290 }
291 ctx->expired = 0;
292 ctx->ticks = 0;
b215e283 293 }
18963c01 294 spin_unlock_irq(&ctx->wqh.lock);
b215e283 295 if (ticks)
09828402 296 res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
b215e283
DL
297 return res;
298}
299
af9c4957 300#ifdef CONFIG_PROC_FS
a3816ab0 301static void timerfd_show(struct seq_file *m, struct file *file)
af9c4957
CG
302{
303 struct timerfd_ctx *ctx = file->private_data;
304 struct itimerspec t;
305
306 spin_lock_irq(&ctx->wqh.lock);
307 t.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
308 t.it_interval = ktime_to_timespec(ctx->tintv);
309 spin_unlock_irq(&ctx->wqh.lock);
310
a3816ab0
JP
311 seq_printf(m,
312 "clockid: %d\n"
313 "ticks: %llu\n"
314 "settime flags: 0%o\n"
315 "it_value: (%llu, %llu)\n"
316 "it_interval: (%llu, %llu)\n",
317 ctx->clockid,
318 (unsigned long long)ctx->ticks,
319 ctx->settime_flags,
320 (unsigned long long)t.it_value.tv_sec,
321 (unsigned long long)t.it_value.tv_nsec,
322 (unsigned long long)t.it_interval.tv_sec,
323 (unsigned long long)t.it_interval.tv_nsec);
af9c4957
CG
324}
325#else
326#define timerfd_show NULL
327#endif
328
5442e9fb
CG
329#ifdef CONFIG_CHECKPOINT_RESTORE
330static long timerfd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
331{
332 struct timerfd_ctx *ctx = file->private_data;
333 int ret = 0;
334
335 switch (cmd) {
336 case TFD_IOC_SET_TICKS: {
337 u64 ticks;
338
339 if (copy_from_user(&ticks, (u64 __user *)arg, sizeof(ticks)))
340 return -EFAULT;
341 if (!ticks)
342 return -EINVAL;
343
344 spin_lock_irq(&ctx->wqh.lock);
345 if (!timerfd_canceled(ctx)) {
346 ctx->ticks = ticks;
88299c9b 347 wake_up_locked(&ctx->wqh);
5442e9fb
CG
348 } else
349 ret = -ECANCELED;
350 spin_unlock_irq(&ctx->wqh.lock);
351 break;
352 }
353 default:
354 ret = -ENOTTY;
355 break;
356 }
357
358 return ret;
359}
360#else
361#define timerfd_ioctl NULL
362#endif
363
b215e283
DL
364static const struct file_operations timerfd_fops = {
365 .release = timerfd_release,
652fe8e8
CH
366 .get_poll_head = timerfd_get_poll_head,
367 .poll_mask = timerfd_poll_mask,
b215e283 368 .read = timerfd_read,
6038f373 369 .llseek = noop_llseek,
af9c4957 370 .show_fdinfo = timerfd_show,
5442e9fb 371 .unlocked_ioctl = timerfd_ioctl,
b215e283
DL
372};
373
2903ff01 374static int timerfd_fget(int fd, struct fd *p)
4d672e7a 375{
2903ff01
AV
376 struct fd f = fdget(fd);
377 if (!f.file)
378 return -EBADF;
379 if (f.file->f_op != &timerfd_fops) {
380 fdput(f);
381 return -EINVAL;
4d672e7a 382 }
2903ff01
AV
383 *p = f;
384 return 0;
4d672e7a
DL
385}
386
836f92ad 387SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
b215e283 388{
2030a42c 389 int ufd;
b215e283 390 struct timerfd_ctx *ctx;
b215e283 391
e38b36f3
UD
392 /* Check the TFD_* constants for consistency. */
393 BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
394 BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
395
610d18f4
DL
396 if ((flags & ~TFD_CREATE_FLAGS) ||
397 (clockid != CLOCK_MONOTONIC &&
11ffa9d6
TP
398 clockid != CLOCK_REALTIME &&
399 clockid != CLOCK_REALTIME_ALARM &&
4a2378a9 400 clockid != CLOCK_BOOTTIME &&
11ffa9d6 401 clockid != CLOCK_BOOTTIME_ALARM))
b215e283 402 return -EINVAL;
4d672e7a 403
25b68a8f
SS
404 if ((clockid == CLOCK_REALTIME_ALARM ||
405 clockid == CLOCK_BOOTTIME_ALARM) &&
406 !capable(CAP_WAKE_ALARM))
2895a5e5
EC
407 return -EPERM;
408
4d672e7a
DL
409 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
410 if (!ctx)
411 return -ENOMEM;
412
413 init_waitqueue_head(&ctx->wqh);
1e38da30 414 spin_lock_init(&ctx->cancel_lock);
4d672e7a 415 ctx->clockid = clockid;
11ffa9d6
TP
416
417 if (isalarm(ctx))
418 alarm_init(&ctx->t.alarm,
419 ctx->clockid == CLOCK_REALTIME_ALARM ?
420 ALARM_REALTIME : ALARM_BOOTTIME,
421 timerfd_alarmproc);
422 else
423 hrtimer_init(&ctx->t.tmr, clockid, HRTIMER_MODE_ABS);
424
2456e855 425 ctx->moffs = ktime_mono_to_real(0);
4d672e7a 426
11fcb6c1 427 ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
628ff7c1 428 O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
2030a42c 429 if (ufd < 0)
4d672e7a 430 kfree(ctx);
4d672e7a
DL
431
432 return ufd;
433}
434
9d94b9e2 435static int do_timerfd_settime(int ufd, int flags,
bff41203
DD
436 const struct itimerspec64 *new,
437 struct itimerspec64 *old)
4d672e7a 438{
2903ff01 439 struct fd f;
4d672e7a 440 struct timerfd_ctx *ctx;
2903ff01 441 int ret;
4d672e7a 442
610d18f4 443 if ((flags & ~TFD_SETTIME_FLAGS) ||
bff41203 444 !itimerspec64_valid(new))
b215e283
DL
445 return -EINVAL;
446
2903ff01
AV
447 ret = timerfd_fget(ufd, &f);
448 if (ret)
449 return ret;
450 ctx = f.file->private_data;
b215e283 451
25b68a8f 452 if (isalarm(ctx) && !capable(CAP_WAKE_ALARM)) {
2895a5e5
EC
453 fdput(f);
454 return -EPERM;
455 }
456
9ec26907
TG
457 timerfd_setup_cancel(ctx, flags);
458
4d672e7a
DL
459 /*
460 * We need to stop the existing timer before reprogramming
461 * it to the new values.
462 */
463 for (;;) {
464 spin_lock_irq(&ctx->wqh.lock);
11ffa9d6
TP
465
466 if (isalarm(ctx)) {
467 if (alarm_try_to_cancel(&ctx->t.alarm) >= 0)
468 break;
469 } else {
470 if (hrtimer_try_to_cancel(&ctx->t.tmr) >= 0)
471 break;
472 }
18963c01 473 spin_unlock_irq(&ctx->wqh.lock);
4d672e7a 474 cpu_relax();
b215e283
DL
475 }
476
4d672e7a
DL
477 /*
478 * If the timer is expired and it's periodic, we need to advance it
479 * because the caller may want to know the previous expiration time.
480 * We do not update "ticks" and "expired" since the timer will be
481 * re-programmed again in the following timerfd_setup() call.
482 */
2456e855 483 if (ctx->expired && ctx->tintv) {
11ffa9d6
TP
484 if (isalarm(ctx))
485 alarm_forward_now(&ctx->t.alarm, ctx->tintv);
486 else
487 hrtimer_forward_now(&ctx->t.tmr, ctx->tintv);
488 }
b215e283 489
bff41203
DD
490 old->it_value = ktime_to_timespec64(timerfd_get_remaining(ctx));
491 old->it_interval = ktime_to_timespec64(ctx->tintv);
4d672e7a
DL
492
493 /*
494 * Re-program the timer to the new value ...
495 */
9d94b9e2 496 ret = timerfd_setup(ctx, flags, new);
4d672e7a
DL
497
498 spin_unlock_irq(&ctx->wqh.lock);
2903ff01 499 fdput(f);
99ee5315 500 return ret;
4d672e7a
DL
501}
502
bff41203 503static int do_timerfd_gettime(int ufd, struct itimerspec64 *t)
4d672e7a 504{
2903ff01 505 struct fd f;
4d672e7a 506 struct timerfd_ctx *ctx;
2903ff01
AV
507 int ret = timerfd_fget(ufd, &f);
508 if (ret)
509 return ret;
510 ctx = f.file->private_data;
4d672e7a
DL
511
512 spin_lock_irq(&ctx->wqh.lock);
2456e855 513 if (ctx->expired && ctx->tintv) {
4d672e7a 514 ctx->expired = 0;
11ffa9d6
TP
515
516 if (isalarm(ctx)) {
517 ctx->ticks +=
518 alarm_forward_now(
519 &ctx->t.alarm, ctx->tintv) - 1;
520 alarm_restart(&ctx->t.alarm);
521 } else {
522 ctx->ticks +=
523 hrtimer_forward_now(&ctx->t.tmr, ctx->tintv)
524 - 1;
525 hrtimer_restart(&ctx->t.tmr);
526 }
4d672e7a 527 }
bff41203
DD
528 t->it_value = ktime_to_timespec64(timerfd_get_remaining(ctx));
529 t->it_interval = ktime_to_timespec64(ctx->tintv);
4d672e7a 530 spin_unlock_irq(&ctx->wqh.lock);
2903ff01 531 fdput(f);
9d94b9e2
AV
532 return 0;
533}
534
535SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
536 const struct itimerspec __user *, utmr,
537 struct itimerspec __user *, otmr)
538{
bff41203 539 struct itimerspec64 new, old;
9d94b9e2
AV
540 int ret;
541
bff41203 542 if (get_itimerspec64(&new, utmr))
9d94b9e2
AV
543 return -EFAULT;
544 ret = do_timerfd_settime(ufd, flags, &new, &old);
545 if (ret)
546 return ret;
bff41203 547 if (otmr && put_itimerspec64(&old, otmr))
9d94b9e2
AV
548 return -EFAULT;
549
550 return ret;
551}
4d672e7a 552
9d94b9e2
AV
553SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
554{
bff41203 555 struct itimerspec64 kotmr;
9d94b9e2
AV
556 int ret = do_timerfd_gettime(ufd, &kotmr);
557 if (ret)
558 return ret;
bff41203 559 return put_itimerspec64(&kotmr, otmr) ? -EFAULT : 0;
b215e283
DL
560}
561
0e803baf 562#ifdef CONFIG_COMPAT
9d94b9e2 563COMPAT_SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
0e803baf
HC
564 const struct compat_itimerspec __user *, utmr,
565 struct compat_itimerspec __user *, otmr)
9d94b9e2 566{
bff41203 567 struct itimerspec64 new, old;
9d94b9e2
AV
568 int ret;
569
bff41203 570 if (get_compat_itimerspec64(&new, utmr))
9d94b9e2
AV
571 return -EFAULT;
572 ret = do_timerfd_settime(ufd, flags, &new, &old);
573 if (ret)
574 return ret;
bff41203 575 if (otmr && put_compat_itimerspec64(&old, otmr))
9d94b9e2
AV
576 return -EFAULT;
577 return ret;
578}
579
580COMPAT_SYSCALL_DEFINE2(timerfd_gettime, int, ufd,
0e803baf 581 struct compat_itimerspec __user *, otmr)
9d94b9e2 582{
bff41203 583 struct itimerspec64 kotmr;
9d94b9e2
AV
584 int ret = do_timerfd_gettime(ufd, &kotmr);
585 if (ret)
586 return ret;
bff41203 587 return put_compat_itimerspec64(&kotmr, otmr) ? -EFAULT : 0;
9d94b9e2
AV
588}
589#endif