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". If EFD_SEMAPHORE flag was not
37 * specified, a read(2) will return the "count" value to userspace,
38 * and will reset "count" to zero. The kernel side eventfd_signal()
39 * also, adds to the "count" counter and issue a wakeup.
46 __u64
eventfd_signal_mask(struct eventfd_ctx
*ctx
, __u64 n
, __poll_t 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
) && ctx
->count
) ? 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
;
232 if (iov_iter_count(to
) < sizeof(ucnt
))
234 spin_lock_irq(&ctx
->wqh
.lock
);
236 if ((file
->f_flags
& O_NONBLOCK
) ||
237 (iocb
->ki_flags
& IOCB_NOWAIT
)) {
238 spin_unlock_irq(&ctx
->wqh
.lock
);
242 if (wait_event_interruptible_locked_irq(ctx
->wqh
, ctx
->count
)) {
243 spin_unlock_irq(&ctx
->wqh
.lock
);
247 eventfd_ctx_do_read(ctx
, &ucnt
);
248 current
->in_eventfd
= 1;
249 if (waitqueue_active(&ctx
->wqh
))
250 wake_up_locked_poll(&ctx
->wqh
, EPOLLOUT
);
251 current
->in_eventfd
= 0;
252 spin_unlock_irq(&ctx
->wqh
.lock
);
253 if (unlikely(copy_to_iter(&ucnt
, sizeof(ucnt
), to
) != sizeof(ucnt
)))
259 static ssize_t
eventfd_write(struct file
*file
, const char __user
*buf
, size_t count
,
262 struct eventfd_ctx
*ctx
= file
->private_data
;
266 if (count
< sizeof(ucnt
))
268 if (copy_from_user(&ucnt
, buf
, sizeof(ucnt
)))
270 if (ucnt
== ULLONG_MAX
)
272 spin_lock_irq(&ctx
->wqh
.lock
);
274 if (ULLONG_MAX
- ctx
->count
> ucnt
)
276 else if (!(file
->f_flags
& O_NONBLOCK
)) {
277 res
= wait_event_interruptible_locked_irq(ctx
->wqh
,
278 ULLONG_MAX
- ctx
->count
> ucnt
);
282 if (likely(res
> 0)) {
284 current
->in_eventfd
= 1;
285 if (waitqueue_active(&ctx
->wqh
))
286 wake_up_locked_poll(&ctx
->wqh
, EPOLLIN
);
287 current
->in_eventfd
= 0;
289 spin_unlock_irq(&ctx
->wqh
.lock
);
294 #ifdef CONFIG_PROC_FS
295 static void eventfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
297 struct eventfd_ctx
*ctx
= f
->private_data
;
299 spin_lock_irq(&ctx
->wqh
.lock
);
300 seq_printf(m
, "eventfd-count: %16llx\n",
301 (unsigned long long)ctx
->count
);
302 spin_unlock_irq(&ctx
->wqh
.lock
);
303 seq_printf(m
, "eventfd-id: %d\n", ctx
->id
);
304 seq_printf(m
, "eventfd-semaphore: %d\n",
305 !!(ctx
->flags
& EFD_SEMAPHORE
));
309 static const struct file_operations eventfd_fops
= {
310 #ifdef CONFIG_PROC_FS
311 .show_fdinfo
= eventfd_show_fdinfo
,
313 .release
= eventfd_release
,
314 .poll
= eventfd_poll
,
315 .read_iter
= eventfd_read
,
316 .write
= eventfd_write
,
317 .llseek
= noop_llseek
,
321 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
322 * @fd: [in] Eventfd file descriptor.
324 * Returns a pointer to the eventfd file structure in case of success, or the
325 * following error pointer:
327 * -EBADF : Invalid @fd file descriptor.
328 * -EINVAL : The @fd file descriptor is not an eventfd file.
330 struct file
*eventfd_fget(int fd
)
336 return ERR_PTR(-EBADF
);
337 if (file
->f_op
!= &eventfd_fops
) {
339 return ERR_PTR(-EINVAL
);
344 EXPORT_SYMBOL_GPL(eventfd_fget
);
347 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
348 * @fd: [in] Eventfd file descriptor.
350 * Returns a pointer to the internal eventfd context, otherwise the error
351 * pointers returned by the following functions:
355 struct eventfd_ctx
*eventfd_ctx_fdget(int fd
)
357 struct eventfd_ctx
*ctx
;
358 struct fd f
= fdget(fd
);
360 return ERR_PTR(-EBADF
);
361 ctx
= eventfd_ctx_fileget(f
.file
);
365 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget
);
368 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
369 * @file: [in] Eventfd file pointer.
371 * Returns a pointer to the internal eventfd context, otherwise the error
374 * -EINVAL : The @fd file descriptor is not an eventfd file.
376 struct eventfd_ctx
*eventfd_ctx_fileget(struct file
*file
)
378 struct eventfd_ctx
*ctx
;
380 if (file
->f_op
!= &eventfd_fops
)
381 return ERR_PTR(-EINVAL
);
383 ctx
= file
->private_data
;
384 kref_get(&ctx
->kref
);
387 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget
);
389 static int do_eventfd(unsigned int count
, int flags
)
391 struct eventfd_ctx
*ctx
;
395 /* Check the EFD_* constants for consistency. */
396 BUILD_BUG_ON(EFD_CLOEXEC
!= O_CLOEXEC
);
397 BUILD_BUG_ON(EFD_NONBLOCK
!= O_NONBLOCK
);
399 if (flags
& ~EFD_FLAGS_SET
)
402 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
406 kref_init(&ctx
->kref
);
407 init_waitqueue_head(&ctx
->wqh
);
410 ctx
->id
= ida_simple_get(&eventfd_ida
, 0, 0, GFP_KERNEL
);
412 flags
&= EFD_SHARED_FCNTL_FLAGS
;
414 fd
= get_unused_fd_flags(flags
);
418 file
= anon_inode_getfile("[eventfd]", &eventfd_fops
, ctx
, flags
);
425 file
->f_mode
|= FMODE_NOWAIT
;
426 fd_install(fd
, file
);
429 eventfd_free_ctx(ctx
);
433 SYSCALL_DEFINE2(eventfd2
, unsigned int, count
, int, flags
)
435 return do_eventfd(count
, flags
);
438 SYSCALL_DEFINE1(eventfd
, unsigned int, count
)
440 return do_eventfd(count
, 0);