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>
26 #include <linux/uio.h>
28 DEFINE_PER_CPU(int, eventfd_wake_count);
30 static DEFINE_IDA(eventfd_ida);
34 wait_queue_head_t wqh;
36 * Every time that a write(2) is performed on an eventfd, the
37 * value of the __u64 being written is added to "count" and a
38 * wakeup is performed on "wqh". A read(2) will return the "count"
39 * value to userspace, and will reset "count" to zero. The kernel
40 * side eventfd_signal() also, adds to the "count" counter and
49 * eventfd_signal - Adds @n to the eventfd counter.
50 * @ctx: [in] Pointer to the eventfd context.
51 * @n: [in] Value of the counter to be added to the eventfd internal counter.
52 * The value cannot be negative.
54 * This function is supposed to be called by the kernel in paths that do not
55 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
56 * value, and we signal this as overflow condition by returning a EPOLLERR
59 * Returns the amount by which the counter was incremented. This will be less
60 * than @n if the counter has overflowed.
62 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
67 * Deadlock or stack overflow issues can happen if we recurse here
68 * through waitqueue wakeup handlers. If the caller users potentially
69 * nested waitqueues with custom wakeup handlers, then it should
70 * check eventfd_signal_count() before calling this function. If
71 * it returns true, the eventfd_signal() call should be deferred to a
74 if (WARN_ON_ONCE(this_cpu_read(eventfd_wake_count)))
77 spin_lock_irqsave(&ctx->wqh.lock, flags);
78 this_cpu_inc(eventfd_wake_count);
79 if (ULLONG_MAX - ctx->count < n)
80 n = ULLONG_MAX - ctx->count;
82 if (waitqueue_active(&ctx->wqh))
83 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
84 this_cpu_dec(eventfd_wake_count);
85 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
89 EXPORT_SYMBOL_GPL(eventfd_signal);
91 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
94 ida_simple_remove(&eventfd_ida, ctx->id);
98 static void eventfd_free(struct kref *kref)
100 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
102 eventfd_free_ctx(ctx);
106 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
107 * @ctx: [in] Pointer to eventfd context.
109 * The eventfd context reference must have been previously acquired either
110 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
112 void eventfd_ctx_put(struct eventfd_ctx *ctx)
114 kref_put(&ctx->kref, eventfd_free);
116 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
118 static int eventfd_release(struct inode *inode, struct file *file)
120 struct eventfd_ctx *ctx = file->private_data;
122 wake_up_poll(&ctx->wqh, EPOLLHUP);
123 eventfd_ctx_put(ctx);
127 static __poll_t eventfd_poll(struct file *file, poll_table *wait)
129 struct eventfd_ctx *ctx = file->private_data;
133 poll_wait(file, &ctx->wqh, wait);
136 * All writes to ctx->count occur within ctx->wqh.lock. This read
137 * can be done outside ctx->wqh.lock because we know that poll_wait
138 * takes that lock (through add_wait_queue) if our caller will sleep.
140 * The read _can_ therefore seep into add_wait_queue's critical
141 * section, but cannot move above it! add_wait_queue's spin_lock acts
142 * as an acquire barrier and ensures that the read be ordered properly
143 * against the writes. The following CAN happen and is safe:
146 * ----------------- ------------
147 * lock ctx->wqh.lock (in poll_wait)
150 * unlock ctx->wqh.lock
153 * if (waitqueue_active)
154 * wake_up_locked_poll
155 * unlock ctx->qwh.lock
156 * eventfd_poll returns 0
158 * but the following, which would miss a wakeup, cannot happen:
161 * ----------------- ------------
162 * count = ctx->count (INVALID!)
165 * **waitqueue_active is false**
166 * **no wake_up_locked_poll!**
167 * unlock ctx->qwh.lock
168 * lock ctx->wqh.lock (in poll_wait)
170 * unlock ctx->wqh.lock
171 * eventfd_poll returns 0
173 count = READ_ONCE(ctx->count);
177 if (count == ULLONG_MAX)
179 if (ULLONG_MAX - 1 > count)
185 void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
187 lockdep_assert_held(&ctx->wqh.lock);
189 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
192 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
195 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
196 * @ctx: [in] Pointer to eventfd context.
197 * @wait: [in] Wait queue to be removed.
198 * @cnt: [out] Pointer to the 64-bit counter value.
200 * Returns %0 if successful, or the following error codes:
202 * -EAGAIN : The operation would have blocked.
204 * This is used to atomically remove a wait queue entry from the eventfd wait
205 * queue head, and read/reset the counter value.
207 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
212 spin_lock_irqsave(&ctx->wqh.lock, flags);
213 eventfd_ctx_do_read(ctx, cnt);
214 __remove_wait_queue(&ctx->wqh, wait);
215 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
216 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
217 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
219 return *cnt != 0 ? 0 : -EAGAIN;
221 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
223 static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to)
225 struct file *file = iocb->ki_filp;
226 struct eventfd_ctx *ctx = file->private_data;
228 DECLARE_WAITQUEUE(wait, current);
230 if (iov_iter_count(to) < sizeof(ucnt))
232 spin_lock_irq(&ctx->wqh.lock);
234 if ((file->f_flags & O_NONBLOCK) ||
235 (iocb->ki_flags & IOCB_NOWAIT)) {
236 spin_unlock_irq(&ctx->wqh.lock);
239 __add_wait_queue(&ctx->wqh, &wait);
241 set_current_state(TASK_INTERRUPTIBLE);
244 if (signal_pending(current)) {
245 __remove_wait_queue(&ctx->wqh, &wait);
246 __set_current_state(TASK_RUNNING);
247 spin_unlock_irq(&ctx->wqh.lock);
250 spin_unlock_irq(&ctx->wqh.lock);
252 spin_lock_irq(&ctx->wqh.lock);
254 __remove_wait_queue(&ctx->wqh, &wait);
255 __set_current_state(TASK_RUNNING);
257 eventfd_ctx_do_read(ctx, &ucnt);
258 if (waitqueue_active(&ctx->wqh))
259 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
260 spin_unlock_irq(&ctx->wqh.lock);
261 if (unlikely(copy_to_iter(&ucnt, sizeof(ucnt), to) != sizeof(ucnt)))
267 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
270 struct eventfd_ctx *ctx = file->private_data;
273 DECLARE_WAITQUEUE(wait, current);
275 if (count < sizeof(ucnt))
277 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
279 if (ucnt == ULLONG_MAX)
281 spin_lock_irq(&ctx->wqh.lock);
283 if (ULLONG_MAX - ctx->count > ucnt)
285 else if (!(file->f_flags & O_NONBLOCK)) {
286 __add_wait_queue(&ctx->wqh, &wait);
288 set_current_state(TASK_INTERRUPTIBLE);
289 if (ULLONG_MAX - ctx->count > ucnt) {
293 if (signal_pending(current)) {
297 spin_unlock_irq(&ctx->wqh.lock);
299 spin_lock_irq(&ctx->wqh.lock);
301 __remove_wait_queue(&ctx->wqh, &wait);
302 __set_current_state(TASK_RUNNING);
304 if (likely(res > 0)) {
306 if (waitqueue_active(&ctx->wqh))
307 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
309 spin_unlock_irq(&ctx->wqh.lock);
314 #ifdef CONFIG_PROC_FS
315 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
317 struct eventfd_ctx *ctx = f->private_data;
319 spin_lock_irq(&ctx->wqh.lock);
320 seq_printf(m, "eventfd-count: %16llx\n",
321 (unsigned long long)ctx->count);
322 spin_unlock_irq(&ctx->wqh.lock);
323 seq_printf(m, "eventfd-id: %d\n", ctx->id);
327 static const struct file_operations eventfd_fops = {
328 #ifdef CONFIG_PROC_FS
329 .show_fdinfo = eventfd_show_fdinfo,
331 .release = eventfd_release,
332 .poll = eventfd_poll,
333 .read_iter = eventfd_read,
334 .write = eventfd_write,
335 .llseek = noop_llseek,
339 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
340 * @fd: [in] Eventfd file descriptor.
342 * Returns a pointer to the eventfd file structure in case of success, or the
343 * following error pointer:
345 * -EBADF : Invalid @fd file descriptor.
346 * -EINVAL : The @fd file descriptor is not an eventfd file.
348 struct file *eventfd_fget(int fd)
354 return ERR_PTR(-EBADF);
355 if (file->f_op != &eventfd_fops) {
357 return ERR_PTR(-EINVAL);
362 EXPORT_SYMBOL_GPL(eventfd_fget);
365 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
366 * @fd: [in] Eventfd file descriptor.
368 * Returns a pointer to the internal eventfd context, otherwise the error
369 * pointers returned by the following functions:
373 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
375 struct eventfd_ctx *ctx;
376 struct fd f = fdget(fd);
378 return ERR_PTR(-EBADF);
379 ctx = eventfd_ctx_fileget(f.file);
383 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
386 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
387 * @file: [in] Eventfd file pointer.
389 * Returns a pointer to the internal eventfd context, otherwise the error
392 * -EINVAL : The @fd file descriptor is not an eventfd file.
394 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
396 struct eventfd_ctx *ctx;
398 if (file->f_op != &eventfd_fops)
399 return ERR_PTR(-EINVAL);
401 ctx = file->private_data;
402 kref_get(&ctx->kref);
405 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
407 static int do_eventfd(unsigned int count, int flags)
409 struct eventfd_ctx *ctx;
413 /* Check the EFD_* constants for consistency. */
414 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
415 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
417 if (flags & ~EFD_FLAGS_SET)
420 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
424 kref_init(&ctx->kref);
425 init_waitqueue_head(&ctx->wqh);
428 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
430 flags &= EFD_SHARED_FCNTL_FLAGS;
432 fd = get_unused_fd_flags(flags);
436 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags);
443 file->f_mode |= FMODE_NOWAIT;
444 fd_install(fd, file);
447 eventfd_free_ctx(ctx);
451 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
453 return do_eventfd(count, flags);
456 SYSCALL_DEFINE1(eventfd, unsigned int, count)
458 return do_eventfd(count, 0);