blob: dba15fecf23e376f12ad2cb367424cf202d3a819 [file] [log] [blame]
Davide Libenzie1ad7462007-05-10 22:23:19 -07001/*
2 * fs/eventfd.c
3 *
4 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
5 *
6 */
7
8#include <linux/file.h>
9#include <linux/poll.h>
10#include <linux/init.h>
11#include <linux/fs.h>
12#include <linux/sched.h>
13#include <linux/kernel.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090014#include <linux/slab.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070015#include <linux/list.h>
16#include <linux/spinlock.h>
17#include <linux/anon_inodes.h>
Adrian Bunk7747cdb2008-02-06 01:36:49 -080018#include <linux/syscalls.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050019#include <linux/export.h>
Davide Libenzi13389012009-06-30 11:41:11 -070020#include <linux/kref.h>
21#include <linux/eventfd.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070022
23struct eventfd_ctx {
Davide Libenzi13389012009-06-30 11:41:11 -070024 struct kref kref;
Davide Libenzie1ad7462007-05-10 22:23:19 -070025 wait_queue_head_t wqh;
26 /*
27 * Every time that a write(2) is performed on an eventfd, the
28 * value of the __u64 being written is added to "count" and a
29 * wakeup is performed on "wqh". A read(2) will return the "count"
30 * value to userspace, and will reset "count" to zero. The kernel
Davide Libenzi13389012009-06-30 11:41:11 -070031 * side eventfd_signal() also, adds to the "count" counter and
Davide Libenzie1ad7462007-05-10 22:23:19 -070032 * issue a wakeup.
33 */
34 __u64 count;
Davide Libenzibcd0b232009-03-31 15:24:18 -070035 unsigned int flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -070036};
37
Davide Libenzi13389012009-06-30 11:41:11 -070038/**
39 * eventfd_signal - Adds @n to the eventfd counter.
40 * @ctx: [in] Pointer to the eventfd context.
41 * @n: [in] Value of the counter to be added to the eventfd internal counter.
42 * The value cannot be negative.
43 *
44 * This function is supposed to be called by the kernel in paths that do not
45 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
46 * value, and we signal this as overflow condition by returining a POLLERR
47 * to poll(2).
48 *
49 * Returns @n in case of success, a non-negative number lower than @n in case
50 * of overflow, or the following error codes:
51 *
52 * -EINVAL : The value of @n is negative.
Davide Libenzie1ad7462007-05-10 22:23:19 -070053 */
Davide Libenzi13389012009-06-30 11:41:11 -070054int eventfd_signal(struct eventfd_ctx *ctx, int n)
Davide Libenzie1ad7462007-05-10 22:23:19 -070055{
Davide Libenzie1ad7462007-05-10 22:23:19 -070056 unsigned long flags;
57
58 if (n < 0)
59 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -070060 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070061 if (ULLONG_MAX - ctx->count < n)
62 n = (int) (ULLONG_MAX - ctx->count);
63 ctx->count += n;
64 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -070065 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzid48eb232007-05-18 12:02:33 -070066 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070067
68 return n;
69}
Rusty Russell57186072009-06-12 22:27:09 -060070EXPORT_SYMBOL_GPL(eventfd_signal);
Davide Libenzie1ad7462007-05-10 22:23:19 -070071
Davide Libenzi562787a2009-09-22 16:43:57 -070072static void eventfd_free_ctx(struct eventfd_ctx *ctx)
73{
74 kfree(ctx);
75}
76
Davide Libenzi13389012009-06-30 11:41:11 -070077static void eventfd_free(struct kref *kref)
78{
79 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
80
Davide Libenzi562787a2009-09-22 16:43:57 -070081 eventfd_free_ctx(ctx);
Davide Libenzi13389012009-06-30 11:41:11 -070082}
83
84/**
85 * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
86 * @ctx: [in] Pointer to the eventfd context.
87 *
88 * Returns: In case of success, returns a pointer to the eventfd context.
89 */
90struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
91{
92 kref_get(&ctx->kref);
93 return ctx;
94}
95EXPORT_SYMBOL_GPL(eventfd_ctx_get);
96
97/**
98 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
99 * @ctx: [in] Pointer to eventfd context.
100 *
101 * The eventfd context reference must have been previously acquired either
Randy Dunlap36182182011-02-20 20:08:35 -0800102 * with eventfd_ctx_get() or eventfd_ctx_fdget().
Davide Libenzi13389012009-06-30 11:41:11 -0700103 */
104void eventfd_ctx_put(struct eventfd_ctx *ctx)
105{
106 kref_put(&ctx->kref, eventfd_free);
107}
108EXPORT_SYMBOL_GPL(eventfd_ctx_put);
109
Davide Libenzie1ad7462007-05-10 22:23:19 -0700110static int eventfd_release(struct inode *inode, struct file *file)
111{
Davide Libenzi13389012009-06-30 11:41:11 -0700112 struct eventfd_ctx *ctx = file->private_data;
113
114 wake_up_poll(&ctx->wqh, POLLHUP);
115 eventfd_ctx_put(ctx);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700116 return 0;
117}
118
119static unsigned int eventfd_poll(struct file *file, poll_table *wait)
120{
121 struct eventfd_ctx *ctx = file->private_data;
122 unsigned int events = 0;
123 unsigned long flags;
124
125 poll_wait(file, &ctx->wqh, wait);
126
Davide Libenzid48eb232007-05-18 12:02:33 -0700127 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700128 if (ctx->count > 0)
129 events |= POLLIN;
130 if (ctx->count == ULLONG_MAX)
131 events |= POLLERR;
132 if (ULLONG_MAX - 1 > ctx->count)
133 events |= POLLOUT;
Davide Libenzid48eb232007-05-18 12:02:33 -0700134 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700135
136 return events;
137}
138
Davide Libenzicb289d62010-01-13 09:34:36 -0800139static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700140{
Davide Libenzicb289d62010-01-13 09:34:36 -0800141 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
142 ctx->count -= *cnt;
143}
144
145/**
146 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
147 * @ctx: [in] Pointer to eventfd context.
148 * @wait: [in] Wait queue to be removed.
Randy Dunlap36182182011-02-20 20:08:35 -0800149 * @cnt: [out] Pointer to the 64-bit counter value.
Davide Libenzicb289d62010-01-13 09:34:36 -0800150 *
Randy Dunlap36182182011-02-20 20:08:35 -0800151 * Returns %0 if successful, or the following error codes:
Davide Libenzicb289d62010-01-13 09:34:36 -0800152 *
153 * -EAGAIN : The operation would have blocked.
154 *
155 * This is used to atomically remove a wait queue entry from the eventfd wait
156 * queue head, and read/reset the counter value.
157 */
158int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_t *wait,
159 __u64 *cnt)
160{
161 unsigned long flags;
162
163 spin_lock_irqsave(&ctx->wqh.lock, flags);
164 eventfd_ctx_do_read(ctx, cnt);
165 __remove_wait_queue(&ctx->wqh, wait);
166 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
167 wake_up_locked_poll(&ctx->wqh, POLLOUT);
168 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
169
170 return *cnt != 0 ? 0 : -EAGAIN;
171}
172EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
173
174/**
175 * eventfd_ctx_read - Reads the eventfd counter or wait if it is zero.
176 * @ctx: [in] Pointer to eventfd context.
177 * @no_wait: [in] Different from zero if the operation should not block.
Randy Dunlap36182182011-02-20 20:08:35 -0800178 * @cnt: [out] Pointer to the 64-bit counter value.
Davide Libenzicb289d62010-01-13 09:34:36 -0800179 *
Randy Dunlap36182182011-02-20 20:08:35 -0800180 * Returns %0 if successful, or the following error codes:
Davide Libenzicb289d62010-01-13 09:34:36 -0800181 *
Randy Dunlap36182182011-02-20 20:08:35 -0800182 * -EAGAIN : The operation would have blocked but @no_wait was non-zero.
Davide Libenzicb289d62010-01-13 09:34:36 -0800183 * -ERESTARTSYS : A signal interrupted the wait operation.
184 *
185 * If @no_wait is zero, the function might sleep until the eventfd internal
186 * counter becomes greater than zero.
187 */
188ssize_t eventfd_ctx_read(struct eventfd_ctx *ctx, int no_wait, __u64 *cnt)
189{
Davide Libenzie1ad7462007-05-10 22:23:19 -0700190 ssize_t res;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700191 DECLARE_WAITQUEUE(wait, current);
192
Davide Libenzid48eb232007-05-18 12:02:33 -0700193 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzicb289d62010-01-13 09:34:36 -0800194 *cnt = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700195 res = -EAGAIN;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700196 if (ctx->count > 0)
Davide Libenzicb289d62010-01-13 09:34:36 -0800197 res = 0;
198 else if (!no_wait) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700199 __add_wait_queue(&ctx->wqh, &wait);
Davide Libenzicb289d62010-01-13 09:34:36 -0800200 for (;;) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700201 set_current_state(TASK_INTERRUPTIBLE);
202 if (ctx->count > 0) {
Davide Libenzicb289d62010-01-13 09:34:36 -0800203 res = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700204 break;
205 }
206 if (signal_pending(current)) {
207 res = -ERESTARTSYS;
208 break;
209 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700210 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700211 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700212 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700213 }
214 __remove_wait_queue(&ctx->wqh, &wait);
215 __set_current_state(TASK_RUNNING);
216 }
Davide Libenzicb289d62010-01-13 09:34:36 -0800217 if (likely(res == 0)) {
218 eventfd_ctx_do_read(ctx, cnt);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700219 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700220 wake_up_locked_poll(&ctx->wqh, POLLOUT);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700221 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700222 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700223
224 return res;
225}
Davide Libenzicb289d62010-01-13 09:34:36 -0800226EXPORT_SYMBOL_GPL(eventfd_ctx_read);
227
228static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
229 loff_t *ppos)
230{
231 struct eventfd_ctx *ctx = file->private_data;
232 ssize_t res;
233 __u64 cnt;
234
235 if (count < sizeof(cnt))
236 return -EINVAL;
237 res = eventfd_ctx_read(ctx, file->f_flags & O_NONBLOCK, &cnt);
238 if (res < 0)
239 return res;
240
241 return put_user(cnt, (__u64 __user *) buf) ? -EFAULT : sizeof(cnt);
242}
Davide Libenzie1ad7462007-05-10 22:23:19 -0700243
244static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
245 loff_t *ppos)
246{
247 struct eventfd_ctx *ctx = file->private_data;
248 ssize_t res;
249 __u64 ucnt;
250 DECLARE_WAITQUEUE(wait, current);
251
252 if (count < sizeof(ucnt))
253 return -EINVAL;
254 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
255 return -EFAULT;
256 if (ucnt == ULLONG_MAX)
257 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -0700258 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700259 res = -EAGAIN;
260 if (ULLONG_MAX - ctx->count > ucnt)
261 res = sizeof(ucnt);
262 else if (!(file->f_flags & O_NONBLOCK)) {
263 __add_wait_queue(&ctx->wqh, &wait);
264 for (res = 0;;) {
265 set_current_state(TASK_INTERRUPTIBLE);
266 if (ULLONG_MAX - ctx->count > ucnt) {
267 res = sizeof(ucnt);
268 break;
269 }
270 if (signal_pending(current)) {
271 res = -ERESTARTSYS;
272 break;
273 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700274 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700275 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700276 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700277 }
278 __remove_wait_queue(&ctx->wqh, &wait);
279 __set_current_state(TASK_RUNNING);
280 }
Davide Libenzibcd0b232009-03-31 15:24:18 -0700281 if (likely(res > 0)) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700282 ctx->count += ucnt;
283 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700284 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700285 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700286 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700287
288 return res;
289}
290
291static const struct file_operations eventfd_fops = {
292 .release = eventfd_release,
293 .poll = eventfd_poll,
294 .read = eventfd_read,
295 .write = eventfd_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200296 .llseek = noop_llseek,
Davide Libenzie1ad7462007-05-10 22:23:19 -0700297};
298
Davide Libenzi13389012009-06-30 11:41:11 -0700299/**
300 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
301 * @fd: [in] Eventfd file descriptor.
302 *
303 * Returns a pointer to the eventfd file structure in case of success, or the
304 * following error pointer:
305 *
306 * -EBADF : Invalid @fd file descriptor.
307 * -EINVAL : The @fd file descriptor is not an eventfd file.
308 */
Davide Libenzie1ad7462007-05-10 22:23:19 -0700309struct file *eventfd_fget(int fd)
310{
311 struct file *file;
312
313 file = fget(fd);
314 if (!file)
315 return ERR_PTR(-EBADF);
316 if (file->f_op != &eventfd_fops) {
317 fput(file);
318 return ERR_PTR(-EINVAL);
319 }
320
321 return file;
322}
Rusty Russell57186072009-06-12 22:27:09 -0600323EXPORT_SYMBOL_GPL(eventfd_fget);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700324
Davide Libenzi13389012009-06-30 11:41:11 -0700325/**
326 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
327 * @fd: [in] Eventfd file descriptor.
328 *
329 * Returns a pointer to the internal eventfd context, otherwise the error
330 * pointers returned by the following functions:
331 *
332 * eventfd_fget
333 */
334struct eventfd_ctx *eventfd_ctx_fdget(int fd)
335{
336 struct file *file;
337 struct eventfd_ctx *ctx;
338
339 file = eventfd_fget(fd);
340 if (IS_ERR(file))
341 return (struct eventfd_ctx *) file;
342 ctx = eventfd_ctx_get(file->private_data);
343 fput(file);
344
345 return ctx;
346}
347EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
348
349/**
350 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
351 * @file: [in] Eventfd file pointer.
352 *
353 * Returns a pointer to the internal eventfd context, otherwise the error
354 * pointer:
355 *
356 * -EINVAL : The @fd file descriptor is not an eventfd file.
357 */
358struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
359{
360 if (file->f_op != &eventfd_fops)
361 return ERR_PTR(-EINVAL);
362
363 return eventfd_ctx_get(file->private_data);
364}
365EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
366
Davide Libenzi562787a2009-09-22 16:43:57 -0700367/**
368 * eventfd_file_create - Creates an eventfd file pointer.
369 * @count: Initial eventfd counter value.
370 * @flags: Flags for the eventfd file.
371 *
372 * This function creates an eventfd file pointer, w/out installing it into
373 * the fd table. This is useful when the eventfd file is used during the
374 * initialization of data structures that require extra setup after the eventfd
375 * creation. So the eventfd creation is split into the file pointer creation
376 * phase, and the file descriptor installation phase.
377 * In this way races with userspace closing the newly installed file descriptor
378 * can be avoided.
379 * Returns an eventfd file pointer, or a proper error pointer.
380 */
381struct file *eventfd_file_create(unsigned int count, int flags)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700382{
Davide Libenzi562787a2009-09-22 16:43:57 -0700383 struct file *file;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700384 struct eventfd_ctx *ctx;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700385
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700386 /* Check the EFD_* constants for consistency. */
387 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
388 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
389
Davide Libenzibcd0b232009-03-31 15:24:18 -0700390 if (flags & ~EFD_FLAGS_SET)
Davide Libenzi562787a2009-09-22 16:43:57 -0700391 return ERR_PTR(-EINVAL);
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700392
Davide Libenzie1ad7462007-05-10 22:23:19 -0700393 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
394 if (!ctx)
Davide Libenzi562787a2009-09-22 16:43:57 -0700395 return ERR_PTR(-ENOMEM);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700396
Davide Libenzi13389012009-06-30 11:41:11 -0700397 kref_init(&ctx->kref);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700398 init_waitqueue_head(&ctx->wqh);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700399 ctx->count = count;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700400 ctx->flags = flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700401
Davide Libenzi562787a2009-09-22 16:43:57 -0700402 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
Roland Dreier628ff7c2009-12-18 09:41:24 -0800403 O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
Davide Libenzi562787a2009-09-22 16:43:57 -0700404 if (IS_ERR(file))
405 eventfd_free_ctx(ctx);
406
407 return file;
408}
409
410SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
411{
412 int fd, error;
413 struct file *file;
414
415 error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
416 if (error < 0)
417 return error;
418 fd = error;
419
420 file = eventfd_file_create(count, flags);
421 if (IS_ERR(file)) {
422 error = PTR_ERR(file);
423 goto err_put_unused_fd;
424 }
425 fd_install(fd, file);
426
Al Viro2030a422008-02-23 06:46:49 -0500427 return fd;
Davide Libenzi562787a2009-09-22 16:43:57 -0700428
429err_put_unused_fd:
430 put_unused_fd(fd);
431
432 return error;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700433}
434
Heiko Carstensd4e82042009-01-14 14:14:34 +0100435SYSCALL_DEFINE1(eventfd, unsigned int, count)
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700436{
437 return sys_eventfd2(count, 0);
438}
Davide Libenzibcd0b232009-03-31 15:24:18 -0700439