4 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
8 #include <linux/file.h>
9 #include <linux/poll.h>
10 #include <linux/init.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/list.h>
15 #include <linux/spinlock.h>
16 #include <linux/anon_inodes.h>
17 #include <linux/syscalls.h>
18 #include <linux/module.h>
19 #include <linux/kref.h>
20 #include <linux/eventfd.h>
24 wait_queue_head_t wqh;
26 * Every time that a write(2) is performed on an eventfd, the
27 * value of the __u64 being written is added to "count" and a
28 * wakeup is performed on "wqh". A read(2) will return the "count"
29 * value to userspace, and will reset "count" to zero. The kernel
30 * side eventfd_signal() also, adds to the "count" counter and
38 * eventfd_signal - Adds @n to the eventfd counter.
39 * @ctx: [in] Pointer to the eventfd context.
40 * @n: [in] Value of the counter to be added to the eventfd internal counter.
41 * The value cannot be negative.
43 * This function is supposed to be called by the kernel in paths that do not
44 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
45 * value, and we signal this as overflow condition by returining a POLLERR
48 * Returns @n in case of success, a non-negative number lower than @n in case
49 * of overflow, or the following error codes:
51 * -EINVAL : The value of @n is negative.
53 int eventfd_signal(struct eventfd_ctx *ctx, int n)
59 spin_lock_irqsave(&ctx->wqh.lock, flags);
60 if (ULLONG_MAX - ctx->count < n)
61 n = (int) (ULLONG_MAX - ctx->count);
63 if (waitqueue_active(&ctx->wqh))
64 wake_up_locked_poll(&ctx->wqh, POLLIN);
65 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
69 EXPORT_SYMBOL_GPL(eventfd_signal);
71 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
76 static void eventfd_free(struct kref *kref)
78 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
80 eventfd_free_ctx(ctx);
84 * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
85 * @ctx: [in] Pointer to the eventfd context.
87 * Returns: In case of success, returns a pointer to the eventfd context.
89 struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
94 EXPORT_SYMBOL_GPL(eventfd_ctx_get);
97 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
98 * @ctx: [in] Pointer to eventfd context.
100 * The eventfd context reference must have been previously acquired either
101 * with eventfd_ctx_get() or eventfd_ctx_fdget()).
103 void eventfd_ctx_put(struct eventfd_ctx *ctx)
105 kref_put(&ctx->kref, eventfd_free);
107 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
109 static int eventfd_release(struct inode *inode, struct file *file)
111 struct eventfd_ctx *ctx = file->private_data;
113 wake_up_poll(&ctx->wqh, POLLHUP);
114 eventfd_ctx_put(ctx);
118 static unsigned int eventfd_poll(struct file *file, poll_table *wait)
120 struct eventfd_ctx *ctx = file->private_data;
121 unsigned int events = 0;
124 poll_wait(file, &ctx->wqh, wait);
126 spin_lock_irqsave(&ctx->wqh.lock, flags);
129 if (ctx->count == ULLONG_MAX)
131 if (ULLONG_MAX - 1 > ctx->count)
133 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
138 static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
141 struct eventfd_ctx *ctx = file->private_data;
144 DECLARE_WAITQUEUE(wait, current);
146 if (count < sizeof(ucnt))
148 spin_lock_irq(&ctx->wqh.lock);
152 else if (!(file->f_flags & O_NONBLOCK)) {
153 __add_wait_queue(&ctx->wqh, &wait);
155 set_current_state(TASK_INTERRUPTIBLE);
156 if (ctx->count > 0) {
160 if (signal_pending(current)) {
164 spin_unlock_irq(&ctx->wqh.lock);
166 spin_lock_irq(&ctx->wqh.lock);
168 __remove_wait_queue(&ctx->wqh, &wait);
169 __set_current_state(TASK_RUNNING);
171 if (likely(res > 0)) {
172 ucnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
174 if (waitqueue_active(&ctx->wqh))
175 wake_up_locked_poll(&ctx->wqh, POLLOUT);
177 spin_unlock_irq(&ctx->wqh.lock);
178 if (res > 0 && put_user(ucnt, (__u64 __user *) buf))
184 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
187 struct eventfd_ctx *ctx = file->private_data;
190 DECLARE_WAITQUEUE(wait, current);
192 if (count < sizeof(ucnt))
194 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
196 if (ucnt == ULLONG_MAX)
198 spin_lock_irq(&ctx->wqh.lock);
200 if (ULLONG_MAX - ctx->count > ucnt)
202 else if (!(file->f_flags & O_NONBLOCK)) {
203 __add_wait_queue(&ctx->wqh, &wait);
205 set_current_state(TASK_INTERRUPTIBLE);
206 if (ULLONG_MAX - ctx->count > ucnt) {
210 if (signal_pending(current)) {
214 spin_unlock_irq(&ctx->wqh.lock);
216 spin_lock_irq(&ctx->wqh.lock);
218 __remove_wait_queue(&ctx->wqh, &wait);
219 __set_current_state(TASK_RUNNING);
221 if (likely(res > 0)) {
223 if (waitqueue_active(&ctx->wqh))
224 wake_up_locked_poll(&ctx->wqh, POLLIN);
226 spin_unlock_irq(&ctx->wqh.lock);
231 static const struct file_operations eventfd_fops = {
232 .release = eventfd_release,
233 .poll = eventfd_poll,
234 .read = eventfd_read,
235 .write = eventfd_write,
239 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
240 * @fd: [in] Eventfd file descriptor.
242 * Returns a pointer to the eventfd file structure in case of success, or the
243 * following error pointer:
245 * -EBADF : Invalid @fd file descriptor.
246 * -EINVAL : The @fd file descriptor is not an eventfd file.
248 struct file *eventfd_fget(int fd)
254 return ERR_PTR(-EBADF);
255 if (file->f_op != &eventfd_fops) {
257 return ERR_PTR(-EINVAL);
262 EXPORT_SYMBOL_GPL(eventfd_fget);
265 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
266 * @fd: [in] Eventfd file descriptor.
268 * Returns a pointer to the internal eventfd context, otherwise the error
269 * pointers returned by the following functions:
273 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
276 struct eventfd_ctx *ctx;
278 file = eventfd_fget(fd);
280 return (struct eventfd_ctx *) file;
281 ctx = eventfd_ctx_get(file->private_data);
286 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
289 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
290 * @file: [in] Eventfd file pointer.
292 * Returns a pointer to the internal eventfd context, otherwise the error
295 * -EINVAL : The @fd file descriptor is not an eventfd file.
297 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
299 if (file->f_op != &eventfd_fops)
300 return ERR_PTR(-EINVAL);
302 return eventfd_ctx_get(file->private_data);
304 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
307 * eventfd_file_create - Creates an eventfd file pointer.
308 * @count: Initial eventfd counter value.
309 * @flags: Flags for the eventfd file.
311 * This function creates an eventfd file pointer, w/out installing it into
312 * the fd table. This is useful when the eventfd file is used during the
313 * initialization of data structures that require extra setup after the eventfd
314 * creation. So the eventfd creation is split into the file pointer creation
315 * phase, and the file descriptor installation phase.
316 * In this way races with userspace closing the newly installed file descriptor
318 * Returns an eventfd file pointer, or a proper error pointer.
320 struct file *eventfd_file_create(unsigned int count, int flags)
323 struct eventfd_ctx *ctx;
325 /* Check the EFD_* constants for consistency. */
326 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
327 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
329 if (flags & ~EFD_FLAGS_SET)
330 return ERR_PTR(-EINVAL);
332 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
334 return ERR_PTR(-ENOMEM);
336 kref_init(&ctx->kref);
337 init_waitqueue_head(&ctx->wqh);
341 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
342 flags & EFD_SHARED_FCNTL_FLAGS);
344 eventfd_free_ctx(ctx);
349 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
354 error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
359 file = eventfd_file_create(count, flags);
361 error = PTR_ERR(file);
362 goto err_put_unused_fd;
364 fd_install(fd, file);
374 SYSCALL_DEFINE1(eventfd, unsigned int, count)
376 return sys_eventfd2(count, 0);