blob: 9736df2ce89d2db41e6d19cba88a808e034319e9 [file] [log] [blame]
Davide Libenzie1ad7462007-05-10 22:23:19 -07001/*
2 * fs/eventfd.c
3 *
4 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
5 *
6 */
7
8#include <linux/file.h>
9#include <linux/poll.h>
10#include <linux/init.h>
11#include <linux/fs.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010012#include <linux/sched/signal.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070013#include <linux/kernel.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090014#include <linux/slab.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070015#include <linux/list.h>
16#include <linux/spinlock.h>
17#include <linux/anon_inodes.h>
Adrian Bunk7747cdb2008-02-06 01:36:49 -080018#include <linux/syscalls.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050019#include <linux/export.h>
Davide Libenzi13389012009-06-30 11:41:11 -070020#include <linux/kref.h>
21#include <linux/eventfd.h>
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -080022#include <linux/proc_fs.h>
23#include <linux/seq_file.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070024
25struct eventfd_ctx {
Davide Libenzi13389012009-06-30 11:41:11 -070026 struct kref kref;
Davide Libenzie1ad7462007-05-10 22:23:19 -070027 wait_queue_head_t wqh;
28 /*
29 * Every time that a write(2) is performed on an eventfd, the
30 * value of the __u64 being written is added to "count" and a
31 * wakeup is performed on "wqh". A read(2) will return the "count"
32 * value to userspace, and will reset "count" to zero. The kernel
Davide Libenzi13389012009-06-30 11:41:11 -070033 * side eventfd_signal() also, adds to the "count" counter and
Davide Libenzie1ad7462007-05-10 22:23:19 -070034 * issue a wakeup.
35 */
36 __u64 count;
Davide Libenzibcd0b232009-03-31 15:24:18 -070037 unsigned int flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -070038};
39
Davide Libenzi13389012009-06-30 11:41:11 -070040/**
41 * eventfd_signal - Adds @n to the eventfd counter.
42 * @ctx: [in] Pointer to the eventfd context.
43 * @n: [in] Value of the counter to be added to the eventfd internal counter.
44 * The value cannot be negative.
45 *
46 * This function is supposed to be called by the kernel in paths that do not
47 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
Masanari Iida20d5a862015-09-22 12:04:17 +090048 * value, and we signal this as overflow condition by returning a POLLERR
Davide Libenzi13389012009-06-30 11:41:11 -070049 * to poll(2).
50 *
Masanari Iida20d5a862015-09-22 12:04:17 +090051 * Returns the amount by which the counter was incremented. This will be less
Sha Zhengjuee62c6b2012-05-31 16:26:41 -070052 * than @n if the counter has overflowed.
Davide Libenzie1ad7462007-05-10 22:23:19 -070053 */
Sha Zhengjuee62c6b2012-05-31 16:26:41 -070054__u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
Davide Libenzie1ad7462007-05-10 22:23:19 -070055{
Davide Libenzie1ad7462007-05-10 22:23:19 -070056 unsigned long flags;
57
Davide Libenzid48eb232007-05-18 12:02:33 -070058 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070059 if (ULLONG_MAX - ctx->count < n)
Sha Zhengjuee62c6b2012-05-31 16:26:41 -070060 n = ULLONG_MAX - ctx->count;
Davide Libenzie1ad7462007-05-10 22:23:19 -070061 ctx->count += n;
62 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -070063 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzid48eb232007-05-18 12:02:33 -070064 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070065
66 return n;
67}
Rusty Russell57186072009-06-12 22:27:09 -060068EXPORT_SYMBOL_GPL(eventfd_signal);
Davide Libenzie1ad7462007-05-10 22:23:19 -070069
Davide Libenzi562787a2009-09-22 16:43:57 -070070static void eventfd_free_ctx(struct eventfd_ctx *ctx)
71{
72 kfree(ctx);
73}
74
Davide Libenzi13389012009-06-30 11:41:11 -070075static void eventfd_free(struct kref *kref)
76{
77 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
78
Davide Libenzi562787a2009-09-22 16:43:57 -070079 eventfd_free_ctx(ctx);
Davide Libenzi13389012009-06-30 11:41:11 -070080}
81
82/**
83 * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
84 * @ctx: [in] Pointer to the eventfd context.
85 *
86 * Returns: In case of success, returns a pointer to the eventfd context.
87 */
88struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
89{
90 kref_get(&ctx->kref);
91 return ctx;
92}
93EXPORT_SYMBOL_GPL(eventfd_ctx_get);
94
95/**
96 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
97 * @ctx: [in] Pointer to eventfd context.
98 *
99 * The eventfd context reference must have been previously acquired either
Randy Dunlap36182182011-02-20 20:08:35 -0800100 * with eventfd_ctx_get() or eventfd_ctx_fdget().
Davide Libenzi13389012009-06-30 11:41:11 -0700101 */
102void eventfd_ctx_put(struct eventfd_ctx *ctx)
103{
104 kref_put(&ctx->kref, eventfd_free);
105}
106EXPORT_SYMBOL_GPL(eventfd_ctx_put);
107
Davide Libenzie1ad7462007-05-10 22:23:19 -0700108static int eventfd_release(struct inode *inode, struct file *file)
109{
Davide Libenzi13389012009-06-30 11:41:11 -0700110 struct eventfd_ctx *ctx = file->private_data;
111
112 wake_up_poll(&ctx->wqh, POLLHUP);
113 eventfd_ctx_put(ctx);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700114 return 0;
115}
116
117static unsigned int eventfd_poll(struct file *file, poll_table *wait)
118{
119 struct eventfd_ctx *ctx = file->private_data;
120 unsigned int events = 0;
Chris Masone22553e2015-02-17 13:46:07 -0800121 u64 count;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700122
123 poll_wait(file, &ctx->wqh, wait);
Paolo Bonzinia484c3d2016-03-22 14:27:14 -0700124
125 /*
126 * All writes to ctx->count occur within ctx->wqh.lock. This read
127 * can be done outside ctx->wqh.lock because we know that poll_wait
128 * takes that lock (through add_wait_queue) if our caller will sleep.
129 *
130 * The read _can_ therefore seep into add_wait_queue's critical
131 * section, but cannot move above it! add_wait_queue's spin_lock acts
132 * as an acquire barrier and ensures that the read be ordered properly
133 * against the writes. The following CAN happen and is safe:
134 *
135 * poll write
136 * ----------------- ------------
137 * lock ctx->wqh.lock (in poll_wait)
138 * count = ctx->count
139 * __add_wait_queue
140 * unlock ctx->wqh.lock
141 * lock ctx->qwh.lock
142 * ctx->count += n
143 * if (waitqueue_active)
144 * wake_up_locked_poll
145 * unlock ctx->qwh.lock
146 * eventfd_poll returns 0
147 *
148 * but the following, which would miss a wakeup, cannot happen:
149 *
150 * poll write
151 * ----------------- ------------
152 * count = ctx->count (INVALID!)
153 * lock ctx->qwh.lock
154 * ctx->count += n
155 * **waitqueue_active is false**
156 * **no wake_up_locked_poll!**
157 * unlock ctx->qwh.lock
158 * lock ctx->wqh.lock (in poll_wait)
159 * __add_wait_queue
160 * unlock ctx->wqh.lock
161 * eventfd_poll returns 0
162 */
163 count = READ_ONCE(ctx->count);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700164
Chris Masone22553e2015-02-17 13:46:07 -0800165 if (count > 0)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700166 events |= POLLIN;
Chris Masone22553e2015-02-17 13:46:07 -0800167 if (count == ULLONG_MAX)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700168 events |= POLLERR;
Chris Masone22553e2015-02-17 13:46:07 -0800169 if (ULLONG_MAX - 1 > count)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700170 events |= POLLOUT;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700171
172 return events;
173}
174
Davide Libenzicb289d62010-01-13 09:34:36 -0800175static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700176{
Davide Libenzicb289d62010-01-13 09:34:36 -0800177 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
178 ctx->count -= *cnt;
179}
180
181/**
182 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
183 * @ctx: [in] Pointer to eventfd context.
184 * @wait: [in] Wait queue to be removed.
Randy Dunlap36182182011-02-20 20:08:35 -0800185 * @cnt: [out] Pointer to the 64-bit counter value.
Davide Libenzicb289d62010-01-13 09:34:36 -0800186 *
Randy Dunlap36182182011-02-20 20:08:35 -0800187 * Returns %0 if successful, or the following error codes:
Davide Libenzicb289d62010-01-13 09:34:36 -0800188 *
189 * -EAGAIN : The operation would have blocked.
190 *
191 * This is used to atomically remove a wait queue entry from the eventfd wait
192 * queue head, and read/reset the counter value.
193 */
Ingo Molnarac6424b2017-06-20 12:06:13 +0200194int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
Davide Libenzicb289d62010-01-13 09:34:36 -0800195 __u64 *cnt)
196{
197 unsigned long flags;
198
199 spin_lock_irqsave(&ctx->wqh.lock, flags);
200 eventfd_ctx_do_read(ctx, cnt);
201 __remove_wait_queue(&ctx->wqh, wait);
202 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
203 wake_up_locked_poll(&ctx->wqh, POLLOUT);
204 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
205
206 return *cnt != 0 ? 0 : -EAGAIN;
207}
208EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
209
210/**
211 * eventfd_ctx_read - Reads the eventfd counter or wait if it is zero.
212 * @ctx: [in] Pointer to eventfd context.
213 * @no_wait: [in] Different from zero if the operation should not block.
Randy Dunlap36182182011-02-20 20:08:35 -0800214 * @cnt: [out] Pointer to the 64-bit counter value.
Davide Libenzicb289d62010-01-13 09:34:36 -0800215 *
Randy Dunlap36182182011-02-20 20:08:35 -0800216 * Returns %0 if successful, or the following error codes:
Davide Libenzicb289d62010-01-13 09:34:36 -0800217 *
Randy Dunlap36182182011-02-20 20:08:35 -0800218 * -EAGAIN : The operation would have blocked but @no_wait was non-zero.
Davide Libenzicb289d62010-01-13 09:34:36 -0800219 * -ERESTARTSYS : A signal interrupted the wait operation.
220 *
221 * If @no_wait is zero, the function might sleep until the eventfd internal
222 * counter becomes greater than zero.
223 */
224ssize_t eventfd_ctx_read(struct eventfd_ctx *ctx, int no_wait, __u64 *cnt)
225{
Davide Libenzie1ad7462007-05-10 22:23:19 -0700226 ssize_t res;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700227 DECLARE_WAITQUEUE(wait, current);
228
Davide Libenzid48eb232007-05-18 12:02:33 -0700229 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzicb289d62010-01-13 09:34:36 -0800230 *cnt = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700231 res = -EAGAIN;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700232 if (ctx->count > 0)
Davide Libenzicb289d62010-01-13 09:34:36 -0800233 res = 0;
234 else if (!no_wait) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700235 __add_wait_queue(&ctx->wqh, &wait);
Davide Libenzicb289d62010-01-13 09:34:36 -0800236 for (;;) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700237 set_current_state(TASK_INTERRUPTIBLE);
238 if (ctx->count > 0) {
Davide Libenzicb289d62010-01-13 09:34:36 -0800239 res = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700240 break;
241 }
242 if (signal_pending(current)) {
243 res = -ERESTARTSYS;
244 break;
245 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700246 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700247 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700248 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700249 }
250 __remove_wait_queue(&ctx->wqh, &wait);
251 __set_current_state(TASK_RUNNING);
252 }
Davide Libenzicb289d62010-01-13 09:34:36 -0800253 if (likely(res == 0)) {
254 eventfd_ctx_do_read(ctx, cnt);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700255 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700256 wake_up_locked_poll(&ctx->wqh, POLLOUT);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700257 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700258 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700259
260 return res;
261}
Davide Libenzicb289d62010-01-13 09:34:36 -0800262EXPORT_SYMBOL_GPL(eventfd_ctx_read);
263
264static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
265 loff_t *ppos)
266{
267 struct eventfd_ctx *ctx = file->private_data;
268 ssize_t res;
269 __u64 cnt;
270
271 if (count < sizeof(cnt))
272 return -EINVAL;
273 res = eventfd_ctx_read(ctx, file->f_flags & O_NONBLOCK, &cnt);
274 if (res < 0)
275 return res;
276
277 return put_user(cnt, (__u64 __user *) buf) ? -EFAULT : sizeof(cnt);
278}
Davide Libenzie1ad7462007-05-10 22:23:19 -0700279
280static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
281 loff_t *ppos)
282{
283 struct eventfd_ctx *ctx = file->private_data;
284 ssize_t res;
285 __u64 ucnt;
286 DECLARE_WAITQUEUE(wait, current);
287
288 if (count < sizeof(ucnt))
289 return -EINVAL;
290 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
291 return -EFAULT;
292 if (ucnt == ULLONG_MAX)
293 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -0700294 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700295 res = -EAGAIN;
296 if (ULLONG_MAX - ctx->count > ucnt)
297 res = sizeof(ucnt);
298 else if (!(file->f_flags & O_NONBLOCK)) {
299 __add_wait_queue(&ctx->wqh, &wait);
300 for (res = 0;;) {
301 set_current_state(TASK_INTERRUPTIBLE);
302 if (ULLONG_MAX - ctx->count > ucnt) {
303 res = sizeof(ucnt);
304 break;
305 }
306 if (signal_pending(current)) {
307 res = -ERESTARTSYS;
308 break;
309 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700310 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700311 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700312 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700313 }
314 __remove_wait_queue(&ctx->wqh, &wait);
315 __set_current_state(TASK_RUNNING);
316 }
Davide Libenzibcd0b232009-03-31 15:24:18 -0700317 if (likely(res > 0)) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700318 ctx->count += ucnt;
319 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700320 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700321 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700322 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700323
324 return res;
325}
326
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800327#ifdef CONFIG_PROC_FS
Joe Perchesa3816ab2014-09-29 16:08:25 -0700328static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800329{
330 struct eventfd_ctx *ctx = f->private_data;
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800331
332 spin_lock_irq(&ctx->wqh.lock);
Joe Perchesa3816ab2014-09-29 16:08:25 -0700333 seq_printf(m, "eventfd-count: %16llx\n",
334 (unsigned long long)ctx->count);
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800335 spin_unlock_irq(&ctx->wqh.lock);
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800336}
337#endif
338
Davide Libenzie1ad7462007-05-10 22:23:19 -0700339static const struct file_operations eventfd_fops = {
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800340#ifdef CONFIG_PROC_FS
341 .show_fdinfo = eventfd_show_fdinfo,
342#endif
Davide Libenzie1ad7462007-05-10 22:23:19 -0700343 .release = eventfd_release,
344 .poll = eventfd_poll,
345 .read = eventfd_read,
346 .write = eventfd_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200347 .llseek = noop_llseek,
Davide Libenzie1ad7462007-05-10 22:23:19 -0700348};
349
Davide Libenzi13389012009-06-30 11:41:11 -0700350/**
351 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
352 * @fd: [in] Eventfd file descriptor.
353 *
354 * Returns a pointer to the eventfd file structure in case of success, or the
355 * following error pointer:
356 *
357 * -EBADF : Invalid @fd file descriptor.
358 * -EINVAL : The @fd file descriptor is not an eventfd file.
359 */
Davide Libenzie1ad7462007-05-10 22:23:19 -0700360struct file *eventfd_fget(int fd)
361{
362 struct file *file;
363
364 file = fget(fd);
365 if (!file)
366 return ERR_PTR(-EBADF);
367 if (file->f_op != &eventfd_fops) {
368 fput(file);
369 return ERR_PTR(-EINVAL);
370 }
371
372 return file;
373}
Rusty Russell57186072009-06-12 22:27:09 -0600374EXPORT_SYMBOL_GPL(eventfd_fget);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700375
Davide Libenzi13389012009-06-30 11:41:11 -0700376/**
377 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
378 * @fd: [in] Eventfd file descriptor.
379 *
380 * Returns a pointer to the internal eventfd context, otherwise the error
381 * pointers returned by the following functions:
382 *
383 * eventfd_fget
384 */
385struct eventfd_ctx *eventfd_ctx_fdget(int fd)
386{
Davide Libenzi13389012009-06-30 11:41:11 -0700387 struct eventfd_ctx *ctx;
Al Viro36a74112013-12-23 16:51:33 -0500388 struct fd f = fdget(fd);
389 if (!f.file)
390 return ERR_PTR(-EBADF);
391 ctx = eventfd_ctx_fileget(f.file);
392 fdput(f);
Davide Libenzi13389012009-06-30 11:41:11 -0700393 return ctx;
394}
395EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
396
397/**
398 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
399 * @file: [in] Eventfd file pointer.
400 *
401 * Returns a pointer to the internal eventfd context, otherwise the error
402 * pointer:
403 *
404 * -EINVAL : The @fd file descriptor is not an eventfd file.
405 */
406struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
407{
408 if (file->f_op != &eventfd_fops)
409 return ERR_PTR(-EINVAL);
410
411 return eventfd_ctx_get(file->private_data);
412}
413EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
414
Davide Libenzi562787a2009-09-22 16:43:57 -0700415/**
416 * eventfd_file_create - Creates an eventfd file pointer.
417 * @count: Initial eventfd counter value.
418 * @flags: Flags for the eventfd file.
419 *
420 * This function creates an eventfd file pointer, w/out installing it into
421 * the fd table. This is useful when the eventfd file is used during the
422 * initialization of data structures that require extra setup after the eventfd
423 * creation. So the eventfd creation is split into the file pointer creation
424 * phase, and the file descriptor installation phase.
425 * In this way races with userspace closing the newly installed file descriptor
426 * can be avoided.
427 * Returns an eventfd file pointer, or a proper error pointer.
428 */
429struct file *eventfd_file_create(unsigned int count, int flags)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700430{
Davide Libenzi562787a2009-09-22 16:43:57 -0700431 struct file *file;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700432 struct eventfd_ctx *ctx;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700433
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700434 /* Check the EFD_* constants for consistency. */
435 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
436 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
437
Davide Libenzibcd0b232009-03-31 15:24:18 -0700438 if (flags & ~EFD_FLAGS_SET)
Davide Libenzi562787a2009-09-22 16:43:57 -0700439 return ERR_PTR(-EINVAL);
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700440
Davide Libenzie1ad7462007-05-10 22:23:19 -0700441 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
442 if (!ctx)
Davide Libenzi562787a2009-09-22 16:43:57 -0700443 return ERR_PTR(-ENOMEM);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700444
Davide Libenzi13389012009-06-30 11:41:11 -0700445 kref_init(&ctx->kref);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700446 init_waitqueue_head(&ctx->wqh);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700447 ctx->count = count;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700448 ctx->flags = flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700449
Davide Libenzi562787a2009-09-22 16:43:57 -0700450 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
Roland Dreier628ff7c2009-12-18 09:41:24 -0800451 O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
Davide Libenzi562787a2009-09-22 16:43:57 -0700452 if (IS_ERR(file))
453 eventfd_free_ctx(ctx);
454
455 return file;
456}
457
458SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
459{
460 int fd, error;
461 struct file *file;
462
463 error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
464 if (error < 0)
465 return error;
466 fd = error;
467
468 file = eventfd_file_create(count, flags);
469 if (IS_ERR(file)) {
470 error = PTR_ERR(file);
471 goto err_put_unused_fd;
472 }
473 fd_install(fd, file);
474
Al Viro2030a422008-02-23 06:46:49 -0500475 return fd;
Davide Libenzi562787a2009-09-22 16:43:57 -0700476
477err_put_unused_fd:
478 put_unused_fd(fd);
479
480 return error;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700481}
482
Heiko Carstensd4e82042009-01-14 14:14:34 +0100483SYSCALL_DEFINE1(eventfd, unsigned int, count)
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700484{
485 return sys_eventfd2(count, 0);
486}
Davide Libenzibcd0b232009-03-31 15:24:18 -0700487