1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
9 #include <linux/file.h>
10 #include <linux/poll.h>
11 #include <linux/init.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>
27 static DEFINE_IDA(eventfd_ida);
31 wait_queue_head_t wqh;
33 * Every time that a write(2) is performed on an eventfd, the
34 * value of the __u64 being written is added to "count" and a
35 * wakeup is performed on "wqh". A read(2) will return the "count"
36 * value to userspace, and will reset "count" to zero. The kernel
37 * side eventfd_signal() also, adds to the "count" counter and
46 * eventfd_signal - Adds @n to the eventfd counter.
47 * @ctx: [in] Pointer to the eventfd context.
48 * @n: [in] Value of the counter to be added to the eventfd internal counter.
49 * The value cannot be negative.
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
56 * Returns the amount by which the counter was incremented. This will be less
57 * than @n if the counter has overflowed.
59 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
63 spin_lock_irqsave(&ctx->wqh.lock, flags);
64 if (ULLONG_MAX - ctx->count < n)
65 n = ULLONG_MAX - ctx->count;
67 if (waitqueue_active(&ctx->wqh))
68 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
69 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
73 EXPORT_SYMBOL_GPL(eventfd_signal);
75 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
78 ida_simple_remove(&eventfd_ida, ctx->id);
82 static void eventfd_free(struct kref *kref)
84 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
86 eventfd_free_ctx(ctx);
90 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
91 * @ctx: [in] Pointer to eventfd context.
93 * The eventfd context reference must have been previously acquired either
94 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
96 void eventfd_ctx_put(struct eventfd_ctx *ctx)
98 kref_put(&ctx->kref, eventfd_free);
100 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
102 static int eventfd_release(struct inode *inode, struct file *file)
104 struct eventfd_ctx *ctx = file->private_data;
106 wake_up_poll(&ctx->wqh, EPOLLHUP);
107 eventfd_ctx_put(ctx);
111 static __poll_t eventfd_poll(struct file *file, poll_table *wait)
113 struct eventfd_ctx *ctx = file->private_data;
117 poll_wait(file, &ctx->wqh, wait);
120 * All writes to ctx->count occur within ctx->wqh.lock. This read
121 * can be done outside ctx->wqh.lock because we know that poll_wait
122 * takes that lock (through add_wait_queue) if our caller will sleep.
124 * The read _can_ therefore seep into add_wait_queue's critical
125 * section, but cannot move above it! add_wait_queue's spin_lock acts
126 * as an acquire barrier and ensures that the read be ordered properly
127 * against the writes. The following CAN happen and is safe:
130 * ----------------- ------------
131 * lock ctx->wqh.lock (in poll_wait)
134 * unlock ctx->wqh.lock
137 * if (waitqueue_active)
138 * wake_up_locked_poll
139 * unlock ctx->qwh.lock
140 * eventfd_poll returns 0
142 * but the following, which would miss a wakeup, cannot happen:
145 * ----------------- ------------
146 * count = ctx->count (INVALID!)
149 * **waitqueue_active is false**
150 * **no wake_up_locked_poll!**
151 * unlock ctx->qwh.lock
152 * lock ctx->wqh.lock (in poll_wait)
154 * unlock ctx->wqh.lock
155 * eventfd_poll returns 0
157 count = READ_ONCE(ctx->count);
161 if (count == ULLONG_MAX)
163 if (ULLONG_MAX - 1 > count)
169 static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
171 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
176 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
177 * @ctx: [in] Pointer to eventfd context.
178 * @wait: [in] Wait queue to be removed.
179 * @cnt: [out] Pointer to the 64-bit counter value.
181 * Returns %0 if successful, or the following error codes:
183 * -EAGAIN : The operation would have blocked.
185 * This is used to atomically remove a wait queue entry from the eventfd wait
186 * queue head, and read/reset the counter value.
188 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
193 spin_lock_irqsave(&ctx->wqh.lock, flags);
194 eventfd_ctx_do_read(ctx, cnt);
195 __remove_wait_queue(&ctx->wqh, wait);
196 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
197 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
198 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
200 return *cnt != 0 ? 0 : -EAGAIN;
202 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
204 static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
207 struct eventfd_ctx *ctx = file->private_data;
210 DECLARE_WAITQUEUE(wait, current);
212 if (count < sizeof(ucnt))
215 spin_lock_irq(&ctx->wqh.lock);
219 else if (!(file->f_flags & O_NONBLOCK)) {
220 __add_wait_queue(&ctx->wqh, &wait);
222 set_current_state(TASK_INTERRUPTIBLE);
223 if (ctx->count > 0) {
227 if (signal_pending(current)) {
231 spin_unlock_irq(&ctx->wqh.lock);
233 spin_lock_irq(&ctx->wqh.lock);
235 __remove_wait_queue(&ctx->wqh, &wait);
236 __set_current_state(TASK_RUNNING);
238 if (likely(res > 0)) {
239 eventfd_ctx_do_read(ctx, &ucnt);
240 if (waitqueue_active(&ctx->wqh))
241 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
243 spin_unlock_irq(&ctx->wqh.lock);
245 if (res > 0 && put_user(ucnt, (__u64 __user *)buf))
251 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
254 struct eventfd_ctx *ctx = file->private_data;
257 DECLARE_WAITQUEUE(wait, current);
259 if (count < sizeof(ucnt))
261 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
263 if (ucnt == ULLONG_MAX)
265 spin_lock_irq(&ctx->wqh.lock);
267 if (ULLONG_MAX - ctx->count > ucnt)
269 else if (!(file->f_flags & O_NONBLOCK)) {
270 __add_wait_queue(&ctx->wqh, &wait);
272 set_current_state(TASK_INTERRUPTIBLE);
273 if (ULLONG_MAX - ctx->count > ucnt) {
277 if (signal_pending(current)) {
281 spin_unlock_irq(&ctx->wqh.lock);
283 spin_lock_irq(&ctx->wqh.lock);
285 __remove_wait_queue(&ctx->wqh, &wait);
286 __set_current_state(TASK_RUNNING);
288 if (likely(res > 0)) {
290 if (waitqueue_active(&ctx->wqh))
291 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
293 spin_unlock_irq(&ctx->wqh.lock);
298 #ifdef CONFIG_PROC_FS
299 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
301 struct eventfd_ctx *ctx = f->private_data;
303 spin_lock_irq(&ctx->wqh.lock);
304 seq_printf(m, "eventfd-count: %16llx\n",
305 (unsigned long long)ctx->count);
306 spin_unlock_irq(&ctx->wqh.lock);
307 seq_printf(m, "eventfd-id: %d\n", ctx->id);
311 static const struct file_operations eventfd_fops = {
312 #ifdef CONFIG_PROC_FS
313 .show_fdinfo = eventfd_show_fdinfo,
315 .release = eventfd_release,
316 .poll = eventfd_poll,
317 .read = eventfd_read,
318 .write = eventfd_write,
319 .llseek = noop_llseek,
323 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
324 * @fd: [in] Eventfd file descriptor.
326 * Returns a pointer to the eventfd file structure in case of success, or the
327 * following error pointer:
329 * -EBADF : Invalid @fd file descriptor.
330 * -EINVAL : The @fd file descriptor is not an eventfd file.
332 struct file *eventfd_fget(int fd)
338 return ERR_PTR(-EBADF);
339 if (file->f_op != &eventfd_fops) {
341 return ERR_PTR(-EINVAL);
346 EXPORT_SYMBOL_GPL(eventfd_fget);
349 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
350 * @fd: [in] Eventfd file descriptor.
352 * Returns a pointer to the internal eventfd context, otherwise the error
353 * pointers returned by the following functions:
357 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
359 struct eventfd_ctx *ctx;
360 struct fd f = fdget(fd);
362 return ERR_PTR(-EBADF);
363 ctx = eventfd_ctx_fileget(f.file);
367 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
370 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
371 * @file: [in] Eventfd file pointer.
373 * Returns a pointer to the internal eventfd context, otherwise the error
376 * -EINVAL : The @fd file descriptor is not an eventfd file.
378 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
380 struct eventfd_ctx *ctx;
382 if (file->f_op != &eventfd_fops)
383 return ERR_PTR(-EINVAL);
385 ctx = file->private_data;
386 kref_get(&ctx->kref);
389 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
391 static int do_eventfd(unsigned int count, int flags)
393 struct eventfd_ctx *ctx;
396 /* Check the EFD_* constants for consistency. */
397 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
398 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
400 if (flags & ~EFD_FLAGS_SET)
403 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
407 kref_init(&ctx->kref);
408 init_waitqueue_head(&ctx->wqh);
411 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
413 fd = anon_inode_getfd("[eventfd]", &eventfd_fops, ctx,
414 O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
416 eventfd_free_ctx(ctx);
421 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
423 return do_eventfd(count, flags);
426 SYSCALL_DEFINE1(eventfd, unsigned int, count)
428 return do_eventfd(count, 0);