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
48 __u64 eventfd_signal_mask(struct eventfd_ctx *ctx, __u64 n, unsigned mask)
53 * Deadlock or stack overflow issues can happen if we recurse here
54 * through waitqueue wakeup handlers. If the caller users potentially
55 * nested waitqueues with custom wakeup handlers, then it should
56 * check eventfd_signal_count() before calling this function. If
57 * it returns true, the eventfd_signal() call should be deferred to a
60 if (WARN_ON_ONCE(this_cpu_read(eventfd_wake_count)))
63 spin_lock_irqsave(&ctx->wqh.lock, flags);
64 this_cpu_inc(eventfd_wake_count);
65 if (ULLONG_MAX - ctx->count < n)
66 n = ULLONG_MAX - ctx->count;
68 if (waitqueue_active(&ctx->wqh))
69 wake_up_locked_poll(&ctx->wqh, EPOLLIN | mask);
70 this_cpu_dec(eventfd_wake_count);
71 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
77 * eventfd_signal - Adds @n to the eventfd counter.
78 * @ctx: [in] Pointer to the eventfd context.
79 * @n: [in] Value of the counter to be added to the eventfd internal counter.
80 * The value cannot be negative.
82 * This function is supposed to be called by the kernel in paths that do not
83 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
84 * value, and we signal this as overflow condition by returning a EPOLLERR
87 * Returns the amount by which the counter was incremented. This will be less
88 * than @n if the counter has overflowed.
90 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
92 return eventfd_signal_mask(ctx, n, 0);
94 EXPORT_SYMBOL_GPL(eventfd_signal);
96 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
99 ida_simple_remove(&eventfd_ida, ctx->id);
103 static void eventfd_free(struct kref *kref)
105 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
107 eventfd_free_ctx(ctx);
111 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
112 * @ctx: [in] Pointer to eventfd context.
114 * The eventfd context reference must have been previously acquired either
115 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
117 void eventfd_ctx_put(struct eventfd_ctx *ctx)
119 kref_put(&ctx->kref, eventfd_free);
121 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
123 static int eventfd_release(struct inode *inode, struct file *file)
125 struct eventfd_ctx *ctx = file->private_data;
127 wake_up_poll(&ctx->wqh, EPOLLHUP);
128 eventfd_ctx_put(ctx);
132 static __poll_t eventfd_poll(struct file *file, poll_table *wait)
134 struct eventfd_ctx *ctx = file->private_data;
138 poll_wait(file, &ctx->wqh, wait);
141 * All writes to ctx->count occur within ctx->wqh.lock. This read
142 * can be done outside ctx->wqh.lock because we know that poll_wait
143 * takes that lock (through add_wait_queue) if our caller will sleep.
145 * The read _can_ therefore seep into add_wait_queue's critical
146 * section, but cannot move above it! add_wait_queue's spin_lock acts
147 * as an acquire barrier and ensures that the read be ordered properly
148 * against the writes. The following CAN happen and is safe:
151 * ----------------- ------------
152 * lock ctx->wqh.lock (in poll_wait)
155 * unlock ctx->wqh.lock
158 * if (waitqueue_active)
159 * wake_up_locked_poll
160 * unlock ctx->qwh.lock
161 * eventfd_poll returns 0
163 * but the following, which would miss a wakeup, cannot happen:
166 * ----------------- ------------
167 * count = ctx->count (INVALID!)
170 * **waitqueue_active is false**
171 * **no wake_up_locked_poll!**
172 * unlock ctx->qwh.lock
173 * lock ctx->wqh.lock (in poll_wait)
175 * unlock ctx->wqh.lock
176 * eventfd_poll returns 0
178 count = READ_ONCE(ctx->count);
182 if (count == ULLONG_MAX)
184 if (ULLONG_MAX - 1 > count)
190 void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
192 lockdep_assert_held(&ctx->wqh.lock);
194 *cnt = ((ctx->flags & EFD_SEMAPHORE) && ctx->count) ? 1 : ctx->count;
197 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
200 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
201 * @ctx: [in] Pointer to eventfd context.
202 * @wait: [in] Wait queue to be removed.
203 * @cnt: [out] Pointer to the 64-bit counter value.
205 * Returns %0 if successful, or the following error codes:
207 * -EAGAIN : The operation would have blocked.
209 * This is used to atomically remove a wait queue entry from the eventfd wait
210 * queue head, and read/reset the counter value.
212 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
217 spin_lock_irqsave(&ctx->wqh.lock, flags);
218 eventfd_ctx_do_read(ctx, cnt);
219 __remove_wait_queue(&ctx->wqh, wait);
220 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
221 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
222 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
224 return *cnt != 0 ? 0 : -EAGAIN;
226 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
228 static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to)
230 struct file *file = iocb->ki_filp;
231 struct eventfd_ctx *ctx = file->private_data;
233 DECLARE_WAITQUEUE(wait, current);
235 if (iov_iter_count(to) < sizeof(ucnt))
237 spin_lock_irq(&ctx->wqh.lock);
239 if ((file->f_flags & O_NONBLOCK) ||
240 (iocb->ki_flags & IOCB_NOWAIT)) {
241 spin_unlock_irq(&ctx->wqh.lock);
244 __add_wait_queue(&ctx->wqh, &wait);
246 set_current_state(TASK_INTERRUPTIBLE);
249 if (signal_pending(current)) {
250 __remove_wait_queue(&ctx->wqh, &wait);
251 __set_current_state(TASK_RUNNING);
252 spin_unlock_irq(&ctx->wqh.lock);
255 spin_unlock_irq(&ctx->wqh.lock);
257 spin_lock_irq(&ctx->wqh.lock);
259 __remove_wait_queue(&ctx->wqh, &wait);
260 __set_current_state(TASK_RUNNING);
262 eventfd_ctx_do_read(ctx, &ucnt);
263 if (waitqueue_active(&ctx->wqh))
264 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
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 if (waitqueue_active(&ctx->wqh))
312 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
314 spin_unlock_irq(&ctx->wqh.lock);
319 #ifdef CONFIG_PROC_FS
320 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
322 struct eventfd_ctx *ctx = f->private_data;
324 spin_lock_irq(&ctx->wqh.lock);
325 seq_printf(m, "eventfd-count: %16llx\n",
326 (unsigned long long)ctx->count);
327 spin_unlock_irq(&ctx->wqh.lock);
328 seq_printf(m, "eventfd-id: %d\n", ctx->id);
332 static const struct file_operations eventfd_fops = {
333 #ifdef CONFIG_PROC_FS
334 .show_fdinfo = eventfd_show_fdinfo,
336 .release = eventfd_release,
337 .poll = eventfd_poll,
338 .read_iter = eventfd_read,
339 .write = eventfd_write,
340 .llseek = noop_llseek,
344 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
345 * @fd: [in] Eventfd file descriptor.
347 * Returns a pointer to the eventfd file structure in case of success, or the
348 * following error pointer:
350 * -EBADF : Invalid @fd file descriptor.
351 * -EINVAL : The @fd file descriptor is not an eventfd file.
353 struct file *eventfd_fget(int fd)
359 return ERR_PTR(-EBADF);
360 if (file->f_op != &eventfd_fops) {
362 return ERR_PTR(-EINVAL);
367 EXPORT_SYMBOL_GPL(eventfd_fget);
370 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
371 * @fd: [in] Eventfd file descriptor.
373 * Returns a pointer to the internal eventfd context, otherwise the error
374 * pointers returned by the following functions:
378 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
380 struct eventfd_ctx *ctx;
381 struct fd f = fdget(fd);
383 return ERR_PTR(-EBADF);
384 ctx = eventfd_ctx_fileget(f.file);
388 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
391 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
392 * @file: [in] Eventfd file pointer.
394 * Returns a pointer to the internal eventfd context, otherwise the error
397 * -EINVAL : The @fd file descriptor is not an eventfd file.
399 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
401 struct eventfd_ctx *ctx;
403 if (file->f_op != &eventfd_fops)
404 return ERR_PTR(-EINVAL);
406 ctx = file->private_data;
407 kref_get(&ctx->kref);
410 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
412 static int do_eventfd(unsigned int count, int flags)
414 struct eventfd_ctx *ctx;
418 /* Check the EFD_* constants for consistency. */
419 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
420 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
422 if (flags & ~EFD_FLAGS_SET)
425 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
429 kref_init(&ctx->kref);
430 init_waitqueue_head(&ctx->wqh);
433 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
435 flags &= EFD_SHARED_FCNTL_FLAGS;
437 fd = get_unused_fd_flags(flags);
441 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags);
448 file->f_mode |= FMODE_NOWAIT;
449 fd_install(fd, file);
452 eventfd_free_ctx(ctx);
456 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
458 return do_eventfd(count, flags);
461 SYSCALL_DEFINE1(eventfd, unsigned int, count)
463 return do_eventfd(count, 0);