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
46 __u64 eventfd_signal_mask(struct eventfd_ctx *ctx, __u64 n, unsigned mask)
51 * Deadlock or stack overflow issues can happen if we recurse here
52 * through waitqueue wakeup handlers. If the caller users potentially
53 * nested waitqueues with custom wakeup handlers, then it should
54 * check eventfd_signal_allowed() before calling this function. If
55 * it returns false, the eventfd_signal() call should be deferred to a
58 if (WARN_ON_ONCE(current->in_eventfd))
61 spin_lock_irqsave(&ctx->wqh.lock, flags);
62 current->in_eventfd = 1;
63 if (ULLONG_MAX - ctx->count < n)
64 n = ULLONG_MAX - ctx->count;
66 if (waitqueue_active(&ctx->wqh))
67 wake_up_locked_poll(&ctx->wqh, EPOLLIN | mask);
68 current->in_eventfd = 0;
69 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
75 * eventfd_signal - Adds @n to the eventfd counter.
76 * @ctx: [in] Pointer to the eventfd context.
77 * @n: [in] Value of the counter to be added to the eventfd internal counter.
78 * The value cannot be negative.
80 * This function is supposed to be called by the kernel in paths that do not
81 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
82 * value, and we signal this as overflow condition by returning a EPOLLERR
85 * Returns the amount by which the counter was incremented. This will be less
86 * than @n if the counter has overflowed.
88 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
90 return eventfd_signal_mask(ctx, n, 0);
92 EXPORT_SYMBOL_GPL(eventfd_signal);
94 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
97 ida_simple_remove(&eventfd_ida, ctx->id);
101 static void eventfd_free(struct kref *kref)
103 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
105 eventfd_free_ctx(ctx);
109 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
110 * @ctx: [in] Pointer to eventfd context.
112 * The eventfd context reference must have been previously acquired either
113 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
115 void eventfd_ctx_put(struct eventfd_ctx *ctx)
117 kref_put(&ctx->kref, eventfd_free);
119 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
121 static int eventfd_release(struct inode *inode, struct file *file)
123 struct eventfd_ctx *ctx = file->private_data;
125 wake_up_poll(&ctx->wqh, EPOLLHUP);
126 eventfd_ctx_put(ctx);
130 static __poll_t eventfd_poll(struct file *file, poll_table *wait)
132 struct eventfd_ctx *ctx = file->private_data;
136 poll_wait(file, &ctx->wqh, wait);
139 * All writes to ctx->count occur within ctx->wqh.lock. This read
140 * can be done outside ctx->wqh.lock because we know that poll_wait
141 * takes that lock (through add_wait_queue) if our caller will sleep.
143 * The read _can_ therefore seep into add_wait_queue's critical
144 * section, but cannot move above it! add_wait_queue's spin_lock acts
145 * as an acquire barrier and ensures that the read be ordered properly
146 * against the writes. The following CAN happen and is safe:
149 * ----------------- ------------
150 * lock ctx->wqh.lock (in poll_wait)
153 * unlock ctx->wqh.lock
156 * if (waitqueue_active)
157 * wake_up_locked_poll
158 * unlock ctx->qwh.lock
159 * eventfd_poll returns 0
161 * but the following, which would miss a wakeup, cannot happen:
164 * ----------------- ------------
165 * count = ctx->count (INVALID!)
168 * **waitqueue_active is false**
169 * **no wake_up_locked_poll!**
170 * unlock ctx->qwh.lock
171 * lock ctx->wqh.lock (in poll_wait)
173 * unlock ctx->wqh.lock
174 * eventfd_poll returns 0
176 count = READ_ONCE(ctx->count);
180 if (count == ULLONG_MAX)
182 if (ULLONG_MAX - 1 > count)
188 void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
190 lockdep_assert_held(&ctx->wqh.lock);
192 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
195 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
198 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
199 * @ctx: [in] Pointer to eventfd context.
200 * @wait: [in] Wait queue to be removed.
201 * @cnt: [out] Pointer to the 64-bit counter value.
203 * Returns %0 if successful, or the following error codes:
205 * -EAGAIN : The operation would have blocked.
207 * This is used to atomically remove a wait queue entry from the eventfd wait
208 * queue head, and read/reset the counter value.
210 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
215 spin_lock_irqsave(&ctx->wqh.lock, flags);
216 eventfd_ctx_do_read(ctx, cnt);
217 __remove_wait_queue(&ctx->wqh, wait);
218 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
219 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
220 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
222 return *cnt != 0 ? 0 : -EAGAIN;
224 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
226 static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to)
228 struct file *file = iocb->ki_filp;
229 struct eventfd_ctx *ctx = file->private_data;
231 DECLARE_WAITQUEUE(wait, current);
233 if (iov_iter_count(to) < sizeof(ucnt))
235 spin_lock_irq(&ctx->wqh.lock);
237 if ((file->f_flags & O_NONBLOCK) ||
238 (iocb->ki_flags & IOCB_NOWAIT)) {
239 spin_unlock_irq(&ctx->wqh.lock);
242 __add_wait_queue(&ctx->wqh, &wait);
244 set_current_state(TASK_INTERRUPTIBLE);
247 if (signal_pending(current)) {
248 __remove_wait_queue(&ctx->wqh, &wait);
249 __set_current_state(TASK_RUNNING);
250 spin_unlock_irq(&ctx->wqh.lock);
253 spin_unlock_irq(&ctx->wqh.lock);
255 spin_lock_irq(&ctx->wqh.lock);
257 __remove_wait_queue(&ctx->wqh, &wait);
258 __set_current_state(TASK_RUNNING);
260 eventfd_ctx_do_read(ctx, &ucnt);
261 current->in_eventfd = 1;
262 if (waitqueue_active(&ctx->wqh))
263 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
264 current->in_eventfd = 0;
265 spin_unlock_irq(&ctx->wqh.lock);
266 if (unlikely(copy_to_iter(&ucnt, sizeof(ucnt), to) != sizeof(ucnt)))
272 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
275 struct eventfd_ctx *ctx = file->private_data;
278 DECLARE_WAITQUEUE(wait, current);
280 if (count < sizeof(ucnt))
282 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
284 if (ucnt == ULLONG_MAX)
286 spin_lock_irq(&ctx->wqh.lock);
288 if (ULLONG_MAX - ctx->count > ucnt)
290 else if (!(file->f_flags & O_NONBLOCK)) {
291 __add_wait_queue(&ctx->wqh, &wait);
293 set_current_state(TASK_INTERRUPTIBLE);
294 if (ULLONG_MAX - ctx->count > ucnt) {
298 if (signal_pending(current)) {
302 spin_unlock_irq(&ctx->wqh.lock);
304 spin_lock_irq(&ctx->wqh.lock);
306 __remove_wait_queue(&ctx->wqh, &wait);
307 __set_current_state(TASK_RUNNING);
309 if (likely(res > 0)) {
311 current->in_eventfd = 1;
312 if (waitqueue_active(&ctx->wqh))
313 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
314 current->in_eventfd = 0;
316 spin_unlock_irq(&ctx->wqh.lock);
321 #ifdef CONFIG_PROC_FS
322 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
324 struct eventfd_ctx *ctx = f->private_data;
326 spin_lock_irq(&ctx->wqh.lock);
327 seq_printf(m, "eventfd-count: %16llx\n",
328 (unsigned long long)ctx->count);
329 spin_unlock_irq(&ctx->wqh.lock);
330 seq_printf(m, "eventfd-id: %d\n", ctx->id);
334 static const struct file_operations eventfd_fops = {
335 #ifdef CONFIG_PROC_FS
336 .show_fdinfo = eventfd_show_fdinfo,
338 .release = eventfd_release,
339 .poll = eventfd_poll,
340 .read_iter = eventfd_read,
341 .write = eventfd_write,
342 .llseek = noop_llseek,
346 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
347 * @fd: [in] Eventfd file descriptor.
349 * Returns a pointer to the eventfd file structure in case of success, or the
350 * following error pointer:
352 * -EBADF : Invalid @fd file descriptor.
353 * -EINVAL : The @fd file descriptor is not an eventfd file.
355 struct file *eventfd_fget(int fd)
361 return ERR_PTR(-EBADF);
362 if (file->f_op != &eventfd_fops) {
364 return ERR_PTR(-EINVAL);
369 EXPORT_SYMBOL_GPL(eventfd_fget);
372 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
373 * @fd: [in] Eventfd file descriptor.
375 * Returns a pointer to the internal eventfd context, otherwise the error
376 * pointers returned by the following functions:
380 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
382 struct eventfd_ctx *ctx;
383 struct fd f = fdget(fd);
385 return ERR_PTR(-EBADF);
386 ctx = eventfd_ctx_fileget(f.file);
390 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
393 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
394 * @file: [in] Eventfd file pointer.
396 * Returns a pointer to the internal eventfd context, otherwise the error
399 * -EINVAL : The @fd file descriptor is not an eventfd file.
401 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
403 struct eventfd_ctx *ctx;
405 if (file->f_op != &eventfd_fops)
406 return ERR_PTR(-EINVAL);
408 ctx = file->private_data;
409 kref_get(&ctx->kref);
412 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
414 static int do_eventfd(unsigned int count, int flags)
416 struct eventfd_ctx *ctx;
420 /* Check the EFD_* constants for consistency. */
421 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
422 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
424 if (flags & ~EFD_FLAGS_SET)
427 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
431 kref_init(&ctx->kref);
432 init_waitqueue_head(&ctx->wqh);
435 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
437 flags &= EFD_SHARED_FCNTL_FLAGS;
439 fd = get_unused_fd_flags(flags);
443 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags);
450 file->f_mode |= FMODE_NOWAIT;
451 fd_install(fd, file);
454 eventfd_free_ctx(ctx);
458 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
460 return do_eventfd(count, flags);
463 SYSCALL_DEFINE1(eventfd, unsigned int, count)
465 return do_eventfd(count, 0);