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 static DEFINE_IDA(eventfd_ida);
32 wait_queue_head_t wqh;
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". A read(2) will return the "count"
37 * value to userspace, and will reset "count" to zero. The kernel
38 * side eventfd_signal() also, adds to the "count" counter and
47 * eventfd_signal - Adds @n to the eventfd counter.
48 * @ctx: [in] Pointer to the eventfd context.
49 * @n: [in] Value of the counter to be added to the eventfd internal counter.
50 * The value cannot be negative.
52 * This function is supposed to be called by the kernel in paths that do not
53 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
54 * value, and we signal this as overflow condition by returning a EPOLLERR
57 * Returns the amount by which the counter was incremented. This will be less
58 * than @n if the counter has overflowed.
60 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
65 * Deadlock or stack overflow issues can happen if we recurse here
66 * through waitqueue wakeup handlers. If the caller users potentially
67 * nested waitqueues with custom wakeup handlers, then it should
68 * check eventfd_signal_allowed() before calling this function. If
69 * it returns false, the eventfd_signal() call should be deferred to a
72 if (WARN_ON_ONCE(current->in_eventfd_signal))
75 spin_lock_irqsave(&ctx->wqh.lock, flags);
76 current->in_eventfd_signal = 1;
77 if (ULLONG_MAX - ctx->count < n)
78 n = ULLONG_MAX - ctx->count;
80 if (waitqueue_active(&ctx->wqh))
81 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
82 current->in_eventfd_signal = 0;
83 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
87 EXPORT_SYMBOL_GPL(eventfd_signal);
89 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
92 ida_simple_remove(&eventfd_ida, ctx->id);
96 static void eventfd_free(struct kref *kref)
98 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
100 eventfd_free_ctx(ctx);
104 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
105 * @ctx: [in] Pointer to eventfd context.
107 * The eventfd context reference must have been previously acquired either
108 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
110 void eventfd_ctx_put(struct eventfd_ctx *ctx)
112 kref_put(&ctx->kref, eventfd_free);
114 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
116 static int eventfd_release(struct inode *inode, struct file *file)
118 struct eventfd_ctx *ctx = file->private_data;
120 wake_up_poll(&ctx->wqh, EPOLLHUP);
121 eventfd_ctx_put(ctx);
125 static __poll_t eventfd_poll(struct file *file, poll_table *wait)
127 struct eventfd_ctx *ctx = file->private_data;
131 poll_wait(file, &ctx->wqh, wait);
134 * All writes to ctx->count occur within ctx->wqh.lock. This read
135 * can be done outside ctx->wqh.lock because we know that poll_wait
136 * takes that lock (through add_wait_queue) if our caller will sleep.
138 * The read _can_ therefore seep into add_wait_queue's critical
139 * section, but cannot move above it! add_wait_queue's spin_lock acts
140 * as an acquire barrier and ensures that the read be ordered properly
141 * against the writes. The following CAN happen and is safe:
144 * ----------------- ------------
145 * lock ctx->wqh.lock (in poll_wait)
148 * unlock ctx->wqh.lock
151 * if (waitqueue_active)
152 * wake_up_locked_poll
153 * unlock ctx->qwh.lock
154 * eventfd_poll returns 0
156 * but the following, which would miss a wakeup, cannot happen:
159 * ----------------- ------------
160 * count = ctx->count (INVALID!)
163 * **waitqueue_active is false**
164 * **no wake_up_locked_poll!**
165 * unlock ctx->qwh.lock
166 * lock ctx->wqh.lock (in poll_wait)
168 * unlock ctx->wqh.lock
169 * eventfd_poll returns 0
171 count = READ_ONCE(ctx->count);
175 if (count == ULLONG_MAX)
177 if (ULLONG_MAX - 1 > count)
183 void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
185 lockdep_assert_held(&ctx->wqh.lock);
187 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
190 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
193 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
194 * @ctx: [in] Pointer to eventfd context.
195 * @wait: [in] Wait queue to be removed.
196 * @cnt: [out] Pointer to the 64-bit counter value.
198 * Returns %0 if successful, or the following error codes:
200 * -EAGAIN : The operation would have blocked.
202 * This is used to atomically remove a wait queue entry from the eventfd wait
203 * queue head, and read/reset the counter value.
205 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
210 spin_lock_irqsave(&ctx->wqh.lock, flags);
211 eventfd_ctx_do_read(ctx, cnt);
212 __remove_wait_queue(&ctx->wqh, wait);
213 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
214 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
215 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
217 return *cnt != 0 ? 0 : -EAGAIN;
219 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
221 static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to)
223 struct file *file = iocb->ki_filp;
224 struct eventfd_ctx *ctx = file->private_data;
226 DECLARE_WAITQUEUE(wait, current);
228 if (iov_iter_count(to) < sizeof(ucnt))
230 spin_lock_irq(&ctx->wqh.lock);
232 if ((file->f_flags & O_NONBLOCK) ||
233 (iocb->ki_flags & IOCB_NOWAIT)) {
234 spin_unlock_irq(&ctx->wqh.lock);
237 __add_wait_queue(&ctx->wqh, &wait);
239 set_current_state(TASK_INTERRUPTIBLE);
242 if (signal_pending(current)) {
243 __remove_wait_queue(&ctx->wqh, &wait);
244 __set_current_state(TASK_RUNNING);
245 spin_unlock_irq(&ctx->wqh.lock);
248 spin_unlock_irq(&ctx->wqh.lock);
250 spin_lock_irq(&ctx->wqh.lock);
252 __remove_wait_queue(&ctx->wqh, &wait);
253 __set_current_state(TASK_RUNNING);
255 eventfd_ctx_do_read(ctx, &ucnt);
256 if (waitqueue_active(&ctx->wqh))
257 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
258 spin_unlock_irq(&ctx->wqh.lock);
259 if (unlikely(copy_to_iter(&ucnt, sizeof(ucnt), to) != sizeof(ucnt)))
265 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
268 struct eventfd_ctx *ctx = file->private_data;
271 DECLARE_WAITQUEUE(wait, current);
273 if (count < sizeof(ucnt))
275 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
277 if (ucnt == ULLONG_MAX)
279 spin_lock_irq(&ctx->wqh.lock);
281 if (ULLONG_MAX - ctx->count > ucnt)
283 else if (!(file->f_flags & O_NONBLOCK)) {
284 __add_wait_queue(&ctx->wqh, &wait);
286 set_current_state(TASK_INTERRUPTIBLE);
287 if (ULLONG_MAX - ctx->count > ucnt) {
291 if (signal_pending(current)) {
295 spin_unlock_irq(&ctx->wqh.lock);
297 spin_lock_irq(&ctx->wqh.lock);
299 __remove_wait_queue(&ctx->wqh, &wait);
300 __set_current_state(TASK_RUNNING);
302 if (likely(res > 0)) {
304 if (waitqueue_active(&ctx->wqh))
305 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
307 spin_unlock_irq(&ctx->wqh.lock);
312 #ifdef CONFIG_PROC_FS
313 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
315 struct eventfd_ctx *ctx = f->private_data;
317 spin_lock_irq(&ctx->wqh.lock);
318 seq_printf(m, "eventfd-count: %16llx\n",
319 (unsigned long long)ctx->count);
320 spin_unlock_irq(&ctx->wqh.lock);
321 seq_printf(m, "eventfd-id: %d\n", ctx->id);
325 static const struct file_operations eventfd_fops = {
326 #ifdef CONFIG_PROC_FS
327 .show_fdinfo = eventfd_show_fdinfo,
329 .release = eventfd_release,
330 .poll = eventfd_poll,
331 .read_iter = eventfd_read,
332 .write = eventfd_write,
333 .llseek = noop_llseek,
337 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
338 * @fd: [in] Eventfd file descriptor.
340 * Returns a pointer to the eventfd file structure in case of success, or the
341 * following error pointer:
343 * -EBADF : Invalid @fd file descriptor.
344 * -EINVAL : The @fd file descriptor is not an eventfd file.
346 struct file *eventfd_fget(int fd)
352 return ERR_PTR(-EBADF);
353 if (file->f_op != &eventfd_fops) {
355 return ERR_PTR(-EINVAL);
360 EXPORT_SYMBOL_GPL(eventfd_fget);
363 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
364 * @fd: [in] Eventfd file descriptor.
366 * Returns a pointer to the internal eventfd context, otherwise the error
367 * pointers returned by the following functions:
371 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
373 struct eventfd_ctx *ctx;
374 struct fd f = fdget(fd);
376 return ERR_PTR(-EBADF);
377 ctx = eventfd_ctx_fileget(f.file);
381 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
384 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
385 * @file: [in] Eventfd file pointer.
387 * Returns a pointer to the internal eventfd context, otherwise the error
390 * -EINVAL : The @fd file descriptor is not an eventfd file.
392 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
394 struct eventfd_ctx *ctx;
396 if (file->f_op != &eventfd_fops)
397 return ERR_PTR(-EINVAL);
399 ctx = file->private_data;
400 kref_get(&ctx->kref);
403 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
405 static int do_eventfd(unsigned int count, int flags)
407 struct eventfd_ctx *ctx;
411 /* Check the EFD_* constants for consistency. */
412 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
413 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
415 if (flags & ~EFD_FLAGS_SET)
418 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
422 kref_init(&ctx->kref);
423 init_waitqueue_head(&ctx->wqh);
426 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
428 flags &= EFD_SHARED_FCNTL_FLAGS;
430 fd = get_unused_fd_flags(flags);
434 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags);
441 file->f_mode |= FMODE_NOWAIT;
442 fd_install(fd, file);
445 eventfd_free_ctx(ctx);
449 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
451 return do_eventfd(count, flags);
454 SYSCALL_DEFINE1(eventfd, unsigned int, count)
456 return do_eventfd(count, 0);