Linux Audio

Check our new training course

Loading...
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *  fs/eventfd.c
  4 *
  5 *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
  6 *
  7 */
  8
  9#include <linux/file.h>
 10#include <linux/poll.h>
 11#include <linux/init.h>
 12#include <linux/fs.h>
 13#include <linux/sched/signal.h>
 14#include <linux/kernel.h>
 15#include <linux/slab.h>
 16#include <linux/list.h>
 17#include <linux/spinlock.h>
 18#include <linux/anon_inodes.h>
 19#include <linux/syscalls.h>
 20#include <linux/export.h>
 21#include <linux/kref.h>
 22#include <linux/eventfd.h>
 23#include <linux/proc_fs.h>
 24#include <linux/seq_file.h>
 25#include <linux/idr.h>
 26#include <linux/uio.h>
 27
 28static DEFINE_IDA(eventfd_ida);
 29
 30struct eventfd_ctx {
 31	struct kref kref;
 32	wait_queue_head_t wqh;
 33	/*
 34	 * Every time that a write(2) is performed on an eventfd, the
 35	 * value of the __u64 being written is added to "count" and a
 36	 * wakeup is performed on "wqh". If EFD_SEMAPHORE flag was not
 37	 * specified, a read(2) will return the "count" value to userspace,
 38	 * and will reset "count" to zero. The kernel side eventfd_signal()
 39	 * also, adds to the "count" counter and issue a wakeup.
 40	 */
 41	__u64 count;
 42	unsigned int flags;
 43	int id;
 44};
 45
 46/**
 47 * eventfd_signal_mask - Increment the event counter
 48 * @ctx: [in] Pointer to the eventfd context.
 49 * @mask: [in] poll mask
 
 50 *
 51 * This function is supposed to be called by the kernel in paths that do not
 52 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
 53 * value, and we signal this as overflow condition by returning a EPOLLERR
 54 * to poll(2).
 
 
 
 55 */
 56void eventfd_signal_mask(struct eventfd_ctx *ctx, __poll_t mask)
 57{
 58	unsigned long flags;
 59
 60	/*
 61	 * Deadlock or stack overflow issues can happen if we recurse here
 62	 * through waitqueue wakeup handlers. If the caller users potentially
 63	 * nested waitqueues with custom wakeup handlers, then it should
 64	 * check eventfd_signal_allowed() before calling this function. If
 65	 * it returns false, the eventfd_signal() call should be deferred to a
 66	 * safe context.
 67	 */
 68	if (WARN_ON_ONCE(current->in_eventfd))
 69		return;
 70
 71	spin_lock_irqsave(&ctx->wqh.lock, flags);
 72	current->in_eventfd = 1;
 73	if (ctx->count < ULLONG_MAX)
 74		ctx->count++;
 75	if (waitqueue_active(&ctx->wqh))
 76		wake_up_locked_poll(&ctx->wqh, EPOLLIN | mask);
 77	current->in_eventfd = 0;
 78	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 
 
 79}
 80EXPORT_SYMBOL_GPL(eventfd_signal_mask);
 81
 82static void eventfd_free_ctx(struct eventfd_ctx *ctx)
 83{
 84	if (ctx->id >= 0)
 85		ida_free(&eventfd_ida, ctx->id);
 86	kfree(ctx);
 87}
 88
 89static void eventfd_free(struct kref *kref)
 90{
 91	struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
 92
 93	eventfd_free_ctx(ctx);
 94}
 95
 96/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 97 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
 98 * @ctx: [in] Pointer to eventfd context.
 99 *
100 * The eventfd context reference must have been previously acquired either
101 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
102 */
103void eventfd_ctx_put(struct eventfd_ctx *ctx)
104{
105	kref_put(&ctx->kref, eventfd_free);
106}
107EXPORT_SYMBOL_GPL(eventfd_ctx_put);
108
109static int eventfd_release(struct inode *inode, struct file *file)
110{
111	struct eventfd_ctx *ctx = file->private_data;
112
113	wake_up_poll(&ctx->wqh, EPOLLHUP);
114	eventfd_ctx_put(ctx);
115	return 0;
116}
117
118static __poll_t eventfd_poll(struct file *file, poll_table *wait)
119{
120	struct eventfd_ctx *ctx = file->private_data;
121	__poll_t events = 0;
122	u64 count;
123
124	poll_wait(file, &ctx->wqh, wait);
125
126	/*
127	 * All writes to ctx->count occur within ctx->wqh.lock.  This read
128	 * can be done outside ctx->wqh.lock because we know that poll_wait
129	 * takes that lock (through add_wait_queue) if our caller will sleep.
130	 *
131	 * The read _can_ therefore seep into add_wait_queue's critical
132	 * section, but cannot move above it!  add_wait_queue's spin_lock acts
133	 * as an acquire barrier and ensures that the read be ordered properly
134	 * against the writes.  The following CAN happen and is safe:
135	 *
136	 *     poll                               write
137	 *     -----------------                  ------------
138	 *     lock ctx->wqh.lock (in poll_wait)
139	 *     count = ctx->count
140	 *     __add_wait_queue
141	 *     unlock ctx->wqh.lock
142	 *                                        lock ctx->qwh.lock
143	 *                                        ctx->count += n
144	 *                                        if (waitqueue_active)
145	 *                                          wake_up_locked_poll
146	 *                                        unlock ctx->qwh.lock
147	 *     eventfd_poll returns 0
148	 *
149	 * but the following, which would miss a wakeup, cannot happen:
150	 *
151	 *     poll                               write
152	 *     -----------------                  ------------
153	 *     count = ctx->count (INVALID!)
154	 *                                        lock ctx->qwh.lock
155	 *                                        ctx->count += n
156	 *                                        **waitqueue_active is false**
157	 *                                        **no wake_up_locked_poll!**
158	 *                                        unlock ctx->qwh.lock
159	 *     lock ctx->wqh.lock (in poll_wait)
160	 *     __add_wait_queue
161	 *     unlock ctx->wqh.lock
162	 *     eventfd_poll returns 0
163	 */
164	count = READ_ONCE(ctx->count);
165
166	if (count > 0)
167		events |= EPOLLIN;
168	if (count == ULLONG_MAX)
169		events |= EPOLLERR;
170	if (ULLONG_MAX - 1 > count)
171		events |= EPOLLOUT;
172
173	return events;
174}
175
176void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
177{
178	lockdep_assert_held(&ctx->wqh.lock);
179
180	*cnt = ((ctx->flags & EFD_SEMAPHORE) && ctx->count) ? 1 : ctx->count;
181	ctx->count -= *cnt;
182}
183EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
184
185/**
186 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
187 * @ctx: [in] Pointer to eventfd context.
188 * @wait: [in] Wait queue to be removed.
189 * @cnt: [out] Pointer to the 64-bit counter value.
190 *
191 * Returns %0 if successful, or the following error codes:
192 *
193 * -EAGAIN      : The operation would have blocked.
194 *
195 * This is used to atomically remove a wait queue entry from the eventfd wait
196 * queue head, and read/reset the counter value.
197 */
198int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
199				  __u64 *cnt)
200{
201	unsigned long flags;
202
203	spin_lock_irqsave(&ctx->wqh.lock, flags);
204	eventfd_ctx_do_read(ctx, cnt);
205	__remove_wait_queue(&ctx->wqh, wait);
206	if (*cnt != 0 && waitqueue_active(&ctx->wqh))
207		wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
208	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
209
210	return *cnt != 0 ? 0 : -EAGAIN;
211}
212EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
213
214static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
215{
216	struct file *file = iocb->ki_filp;
217	struct eventfd_ctx *ctx = file->private_data;
218	__u64 ucnt = 0;
219
220	if (iov_iter_count(to) < sizeof(ucnt))
221		return -EINVAL;
222	spin_lock_irq(&ctx->wqh.lock);
223	if (!ctx->count) {
224		if ((file->f_flags & O_NONBLOCK) ||
225		    (iocb->ki_flags & IOCB_NOWAIT)) {
226			spin_unlock_irq(&ctx->wqh.lock);
227			return -EAGAIN;
228		}
229
230		if (wait_event_interruptible_locked_irq(ctx->wqh, ctx->count)) {
 
 
 
 
 
 
 
 
231			spin_unlock_irq(&ctx->wqh.lock);
232			return -ERESTARTSYS;
 
233		}
 
 
 
 
 
 
 
234	}
235	eventfd_ctx_do_read(ctx, &ucnt);
236	current->in_eventfd = 1;
237	if (waitqueue_active(&ctx->wqh))
238		wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
239	current->in_eventfd = 0;
240	spin_unlock_irq(&ctx->wqh.lock);
241	if (unlikely(copy_to_iter(&ucnt, sizeof(ucnt), to) != sizeof(ucnt)))
242		return -EFAULT;
243
244	return sizeof(ucnt);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
245}
246
247static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
248			     loff_t *ppos)
249{
250	struct eventfd_ctx *ctx = file->private_data;
251	ssize_t res;
252	__u64 ucnt;
 
253
254	if (count != sizeof(ucnt))
255		return -EINVAL;
256	if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
257		return -EFAULT;
258	if (ucnt == ULLONG_MAX)
259		return -EINVAL;
260	spin_lock_irq(&ctx->wqh.lock);
261	res = -EAGAIN;
262	if (ULLONG_MAX - ctx->count > ucnt)
263		res = sizeof(ucnt);
264	else if (!(file->f_flags & O_NONBLOCK)) {
265		res = wait_event_interruptible_locked_irq(ctx->wqh,
266				ULLONG_MAX - ctx->count > ucnt);
267		if (!res)
268			res = sizeof(ucnt);
 
 
 
 
 
 
 
 
 
 
 
 
 
269	}
270	if (likely(res > 0)) {
271		ctx->count += ucnt;
272		current->in_eventfd = 1;
273		if (waitqueue_active(&ctx->wqh))
274			wake_up_locked_poll(&ctx->wqh, EPOLLIN);
275		current->in_eventfd = 0;
276	}
277	spin_unlock_irq(&ctx->wqh.lock);
278
279	return res;
280}
281
282#ifdef CONFIG_PROC_FS
283static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
284{
285	struct eventfd_ctx *ctx = f->private_data;
286	__u64 cnt;
287
288	spin_lock_irq(&ctx->wqh.lock);
289	cnt = ctx->count;
290	spin_unlock_irq(&ctx->wqh.lock);
291
292	seq_printf(m,
293		   "eventfd-count: %16llx\n"
294		   "eventfd-id: %d\n"
295		   "eventfd-semaphore: %d\n",
296		   cnt,
297		   ctx->id,
298		   !!(ctx->flags & EFD_SEMAPHORE));
299}
300#endif
301
302static const struct file_operations eventfd_fops = {
303#ifdef CONFIG_PROC_FS
304	.show_fdinfo	= eventfd_show_fdinfo,
305#endif
306	.release	= eventfd_release,
307	.poll		= eventfd_poll,
308	.read_iter	= eventfd_read,
309	.write		= eventfd_write,
310	.llseek		= noop_llseek,
311};
312
313/**
314 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
315 * @fd: [in] Eventfd file descriptor.
316 *
317 * Returns a pointer to the eventfd file structure in case of success, or the
318 * following error pointer:
319 *
320 * -EBADF    : Invalid @fd file descriptor.
321 * -EINVAL   : The @fd file descriptor is not an eventfd file.
322 */
323struct file *eventfd_fget(int fd)
324{
325	struct file *file;
326
327	file = fget(fd);
328	if (!file)
329		return ERR_PTR(-EBADF);
330	if (file->f_op != &eventfd_fops) {
331		fput(file);
332		return ERR_PTR(-EINVAL);
333	}
334
335	return file;
336}
337EXPORT_SYMBOL_GPL(eventfd_fget);
338
339/**
340 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
341 * @fd: [in] Eventfd file descriptor.
342 *
343 * Returns a pointer to the internal eventfd context, otherwise the error
344 * pointers returned by the following functions:
345 *
346 * eventfd_fget
347 */
348struct eventfd_ctx *eventfd_ctx_fdget(int fd)
349{
350	CLASS(fd, f)(fd);
351	if (fd_empty(f))
352		return ERR_PTR(-EBADF);
353	return eventfd_ctx_fileget(fd_file(f));
 
 
 
 
 
 
354}
355EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
356
357/**
358 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
359 * @file: [in] Eventfd file pointer.
360 *
361 * Returns a pointer to the internal eventfd context, otherwise the error
362 * pointer:
363 *
364 * -EINVAL   : The @fd file descriptor is not an eventfd file.
365 */
366struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
367{
368	struct eventfd_ctx *ctx;
369
370	if (file->f_op != &eventfd_fops)
371		return ERR_PTR(-EINVAL);
372
373	ctx = file->private_data;
374	kref_get(&ctx->kref);
375	return ctx;
376}
377EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
378
379static int do_eventfd(unsigned int count, int flags)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
380{
381	struct eventfd_ctx *ctx;
382	struct file *file;
383	int fd;
384
385	/* Check the EFD_* constants for consistency.  */
386	BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
387	BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
388	BUILD_BUG_ON(EFD_SEMAPHORE != (1 << 0));
389
390	if (flags & ~EFD_FLAGS_SET)
391		return -EINVAL;
392
393	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
394	if (!ctx)
395		return -ENOMEM;
396
397	kref_init(&ctx->kref);
398	init_waitqueue_head(&ctx->wqh);
399	ctx->count = count;
400	ctx->flags = flags;
401	ctx->id = ida_alloc(&eventfd_ida, GFP_KERNEL);
402
403	flags &= EFD_SHARED_FCNTL_FLAGS;
404	flags |= O_RDWR;
405	fd = get_unused_fd_flags(flags);
406	if (fd < 0)
407		goto err;
 
 
408
409	file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags);
 
 
 
 
 
 
 
 
 
 
410	if (IS_ERR(file)) {
411		put_unused_fd(fd);
412		fd = PTR_ERR(file);
413		goto err;
414	}
415
416	file->f_mode |= FMODE_NOWAIT;
417	fd_install(fd, file);
 
418	return fd;
419err:
420	eventfd_free_ctx(ctx);
421	return fd;
422}
423
424SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
425{
426	return do_eventfd(count, flags);
 
427}
428
429SYSCALL_DEFINE1(eventfd, unsigned int, count)
430{
431	return do_eventfd(count, 0);
432}
433
v3.5.6
 
  1/*
  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>
 12#include <linux/sched.h>
 13#include <linux/kernel.h>
 14#include <linux/slab.h>
 15#include <linux/list.h>
 16#include <linux/spinlock.h>
 17#include <linux/anon_inodes.h>
 18#include <linux/syscalls.h>
 19#include <linux/export.h>
 20#include <linux/kref.h>
 21#include <linux/eventfd.h>
 
 
 
 
 
 
 22
 23struct eventfd_ctx {
 24	struct kref kref;
 25	wait_queue_head_t wqh;
 26	/*
 27	 * Every time that a write(2) is performed on an eventfd, the
 28	 * value of the __u64 being written is added to "count" and a
 29	 * wakeup is performed on "wqh". A read(2) will return the "count"
 30	 * value to userspace, and will reset "count" to zero. The kernel
 31	 * side eventfd_signal() also, adds to the "count" counter and
 32	 * issue a wakeup.
 33	 */
 34	__u64 count;
 35	unsigned int flags;
 
 36};
 37
 38/**
 39 * eventfd_signal - Adds @n to the eventfd counter.
 40 * @ctx: [in] Pointer to the eventfd context.
 41 * @n: [in] Value of the counter to be added to the eventfd internal counter.
 42 *          The value cannot be negative.
 43 *
 44 * This function is supposed to be called by the kernel in paths that do not
 45 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
 46 * value, and we signal this as overflow condition by returining a POLLERR
 47 * to poll(2).
 48 *
 49 * Returns the amount by which the counter was incrememnted.  This will be less
 50 * than @n if the counter has overflowed.
 51 */
 52__u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
 53{
 54	unsigned long flags;
 55
 
 
 
 
 
 
 
 
 
 
 
 56	spin_lock_irqsave(&ctx->wqh.lock, flags);
 57	if (ULLONG_MAX - ctx->count < n)
 58		n = ULLONG_MAX - ctx->count;
 59	ctx->count += n;
 60	if (waitqueue_active(&ctx->wqh))
 61		wake_up_locked_poll(&ctx->wqh, POLLIN);
 
 62	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 63
 64	return n;
 65}
 66EXPORT_SYMBOL_GPL(eventfd_signal);
 67
 68static void eventfd_free_ctx(struct eventfd_ctx *ctx)
 69{
 
 
 70	kfree(ctx);
 71}
 72
 73static void eventfd_free(struct kref *kref)
 74{
 75	struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
 76
 77	eventfd_free_ctx(ctx);
 78}
 79
 80/**
 81 * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
 82 * @ctx: [in] Pointer to the eventfd context.
 83 *
 84 * Returns: In case of success, returns a pointer to the eventfd context.
 85 */
 86struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
 87{
 88	kref_get(&ctx->kref);
 89	return ctx;
 90}
 91EXPORT_SYMBOL_GPL(eventfd_ctx_get);
 92
 93/**
 94 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
 95 * @ctx: [in] Pointer to eventfd context.
 96 *
 97 * The eventfd context reference must have been previously acquired either
 98 * with eventfd_ctx_get() or eventfd_ctx_fdget().
 99 */
100void eventfd_ctx_put(struct eventfd_ctx *ctx)
101{
102	kref_put(&ctx->kref, eventfd_free);
103}
104EXPORT_SYMBOL_GPL(eventfd_ctx_put);
105
106static int eventfd_release(struct inode *inode, struct file *file)
107{
108	struct eventfd_ctx *ctx = file->private_data;
109
110	wake_up_poll(&ctx->wqh, POLLHUP);
111	eventfd_ctx_put(ctx);
112	return 0;
113}
114
115static unsigned int eventfd_poll(struct file *file, poll_table *wait)
116{
117	struct eventfd_ctx *ctx = file->private_data;
118	unsigned int events = 0;
119	unsigned long flags;
120
121	poll_wait(file, &ctx->wqh, wait);
122
123	spin_lock_irqsave(&ctx->wqh.lock, flags);
124	if (ctx->count > 0)
125		events |= POLLIN;
126	if (ctx->count == ULLONG_MAX)
127		events |= POLLERR;
128	if (ULLONG_MAX - 1 > ctx->count)
129		events |= POLLOUT;
130	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
131
132	return events;
133}
134
135static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
136{
137	*cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
 
 
138	ctx->count -= *cnt;
139}
 
140
141/**
142 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
143 * @ctx: [in] Pointer to eventfd context.
144 * @wait: [in] Wait queue to be removed.
145 * @cnt: [out] Pointer to the 64-bit counter value.
146 *
147 * Returns %0 if successful, or the following error codes:
148 *
149 * -EAGAIN      : The operation would have blocked.
150 *
151 * This is used to atomically remove a wait queue entry from the eventfd wait
152 * queue head, and read/reset the counter value.
153 */
154int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_t *wait,
155				  __u64 *cnt)
156{
157	unsigned long flags;
158
159	spin_lock_irqsave(&ctx->wqh.lock, flags);
160	eventfd_ctx_do_read(ctx, cnt);
161	__remove_wait_queue(&ctx->wqh, wait);
162	if (*cnt != 0 && waitqueue_active(&ctx->wqh))
163		wake_up_locked_poll(&ctx->wqh, POLLOUT);
164	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
165
166	return *cnt != 0 ? 0 : -EAGAIN;
167}
168EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
169
170/**
171 * eventfd_ctx_read - Reads the eventfd counter or wait if it is zero.
172 * @ctx: [in] Pointer to eventfd context.
173 * @no_wait: [in] Different from zero if the operation should not block.
174 * @cnt: [out] Pointer to the 64-bit counter value.
175 *
176 * Returns %0 if successful, or the following error codes:
177 *
178 * -EAGAIN      : The operation would have blocked but @no_wait was non-zero.
179 * -ERESTARTSYS : A signal interrupted the wait operation.
180 *
181 * If @no_wait is zero, the function might sleep until the eventfd internal
182 * counter becomes greater than zero.
183 */
184ssize_t eventfd_ctx_read(struct eventfd_ctx *ctx, int no_wait, __u64 *cnt)
185{
186	ssize_t res;
187	DECLARE_WAITQUEUE(wait, current);
 
188
 
 
189	spin_lock_irq(&ctx->wqh.lock);
190	*cnt = 0;
191	res = -EAGAIN;
192	if (ctx->count > 0)
193		res = 0;
194	else if (!no_wait) {
195		__add_wait_queue(&ctx->wqh, &wait);
196		for (;;) {
197			set_current_state(TASK_INTERRUPTIBLE);
198			if (ctx->count > 0) {
199				res = 0;
200				break;
201			}
202			if (signal_pending(current)) {
203				res = -ERESTARTSYS;
204				break;
205			}
206			spin_unlock_irq(&ctx->wqh.lock);
207			schedule();
208			spin_lock_irq(&ctx->wqh.lock);
209		}
210		__remove_wait_queue(&ctx->wqh, &wait);
211		__set_current_state(TASK_RUNNING);
212	}
213	if (likely(res == 0)) {
214		eventfd_ctx_do_read(ctx, cnt);
215		if (waitqueue_active(&ctx->wqh))
216			wake_up_locked_poll(&ctx->wqh, POLLOUT);
217	}
 
 
 
 
 
218	spin_unlock_irq(&ctx->wqh.lock);
 
 
219
220	return res;
221}
222EXPORT_SYMBOL_GPL(eventfd_ctx_read);
223
224static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
225			    loff_t *ppos)
226{
227	struct eventfd_ctx *ctx = file->private_data;
228	ssize_t res;
229	__u64 cnt;
230
231	if (count < sizeof(cnt))
232		return -EINVAL;
233	res = eventfd_ctx_read(ctx, file->f_flags & O_NONBLOCK, &cnt);
234	if (res < 0)
235		return res;
236
237	return put_user(cnt, (__u64 __user *) buf) ? -EFAULT : sizeof(cnt);
238}
239
240static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
241			     loff_t *ppos)
242{
243	struct eventfd_ctx *ctx = file->private_data;
244	ssize_t res;
245	__u64 ucnt;
246	DECLARE_WAITQUEUE(wait, current);
247
248	if (count < sizeof(ucnt))
249		return -EINVAL;
250	if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
251		return -EFAULT;
252	if (ucnt == ULLONG_MAX)
253		return -EINVAL;
254	spin_lock_irq(&ctx->wqh.lock);
255	res = -EAGAIN;
256	if (ULLONG_MAX - ctx->count > ucnt)
257		res = sizeof(ucnt);
258	else if (!(file->f_flags & O_NONBLOCK)) {
259		__add_wait_queue(&ctx->wqh, &wait);
260		for (res = 0;;) {
261			set_current_state(TASK_INTERRUPTIBLE);
262			if (ULLONG_MAX - ctx->count > ucnt) {
263				res = sizeof(ucnt);
264				break;
265			}
266			if (signal_pending(current)) {
267				res = -ERESTARTSYS;
268				break;
269			}
270			spin_unlock_irq(&ctx->wqh.lock);
271			schedule();
272			spin_lock_irq(&ctx->wqh.lock);
273		}
274		__remove_wait_queue(&ctx->wqh, &wait);
275		__set_current_state(TASK_RUNNING);
276	}
277	if (likely(res > 0)) {
278		ctx->count += ucnt;
 
279		if (waitqueue_active(&ctx->wqh))
280			wake_up_locked_poll(&ctx->wqh, POLLIN);
 
281	}
282	spin_unlock_irq(&ctx->wqh.lock);
283
284	return res;
285}
286
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
287static const struct file_operations eventfd_fops = {
 
 
 
288	.release	= eventfd_release,
289	.poll		= eventfd_poll,
290	.read		= eventfd_read,
291	.write		= eventfd_write,
292	.llseek		= noop_llseek,
293};
294
295/**
296 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
297 * @fd: [in] Eventfd file descriptor.
298 *
299 * Returns a pointer to the eventfd file structure in case of success, or the
300 * following error pointer:
301 *
302 * -EBADF    : Invalid @fd file descriptor.
303 * -EINVAL   : The @fd file descriptor is not an eventfd file.
304 */
305struct file *eventfd_fget(int fd)
306{
307	struct file *file;
308
309	file = fget(fd);
310	if (!file)
311		return ERR_PTR(-EBADF);
312	if (file->f_op != &eventfd_fops) {
313		fput(file);
314		return ERR_PTR(-EINVAL);
315	}
316
317	return file;
318}
319EXPORT_SYMBOL_GPL(eventfd_fget);
320
321/**
322 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
323 * @fd: [in] Eventfd file descriptor.
324 *
325 * Returns a pointer to the internal eventfd context, otherwise the error
326 * pointers returned by the following functions:
327 *
328 * eventfd_fget
329 */
330struct eventfd_ctx *eventfd_ctx_fdget(int fd)
331{
332	struct file *file;
333	struct eventfd_ctx *ctx;
334
335	file = eventfd_fget(fd);
336	if (IS_ERR(file))
337		return (struct eventfd_ctx *) file;
338	ctx = eventfd_ctx_get(file->private_data);
339	fput(file);
340
341	return ctx;
342}
343EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
344
345/**
346 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
347 * @file: [in] Eventfd file pointer.
348 *
349 * Returns a pointer to the internal eventfd context, otherwise the error
350 * pointer:
351 *
352 * -EINVAL   : The @fd file descriptor is not an eventfd file.
353 */
354struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
355{
 
 
356	if (file->f_op != &eventfd_fops)
357		return ERR_PTR(-EINVAL);
358
359	return eventfd_ctx_get(file->private_data);
 
 
360}
361EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
362
363/**
364 * eventfd_file_create - Creates an eventfd file pointer.
365 * @count: Initial eventfd counter value.
366 * @flags: Flags for the eventfd file.
367 *
368 * This function creates an eventfd file pointer, w/out installing it into
369 * the fd table. This is useful when the eventfd file is used during the
370 * initialization of data structures that require extra setup after the eventfd
371 * creation. So the eventfd creation is split into the file pointer creation
372 * phase, and the file descriptor installation phase.
373 * In this way races with userspace closing the newly installed file descriptor
374 * can be avoided.
375 * Returns an eventfd file pointer, or a proper error pointer.
376 */
377struct file *eventfd_file_create(unsigned int count, int flags)
378{
 
379	struct file *file;
380	struct eventfd_ctx *ctx;
381
382	/* Check the EFD_* constants for consistency.  */
383	BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
384	BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
 
385
386	if (flags & ~EFD_FLAGS_SET)
387		return ERR_PTR(-EINVAL);
388
389	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
390	if (!ctx)
391		return ERR_PTR(-ENOMEM);
392
393	kref_init(&ctx->kref);
394	init_waitqueue_head(&ctx->wqh);
395	ctx->count = count;
396	ctx->flags = flags;
 
397
398	file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
399				  O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
400	if (IS_ERR(file))
401		eventfd_free_ctx(ctx);
402
403	return file;
404}
405
406SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
407{
408	int fd, error;
409	struct file *file;
410
411	error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
412	if (error < 0)
413		return error;
414	fd = error;
415
416	file = eventfd_file_create(count, flags);
417	if (IS_ERR(file)) {
418		error = PTR_ERR(file);
419		goto err_put_unused_fd;
 
420	}
 
 
421	fd_install(fd, file);
422
423	return fd;
 
 
 
 
424
425err_put_unused_fd:
426	put_unused_fd(fd);
427
428	return error;
429}
430
431SYSCALL_DEFINE1(eventfd, unsigned int, count)
432{
433	return sys_eventfd2(count, 0);
434}
435