blob: 35470d9b96e68951de6f1a63fe3657f09a8296c4 [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>
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -080022#include <linux/proc_fs.h>
23#include <linux/seq_file.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070024
25struct eventfd_ctx {
Davide Libenzi13389012009-06-30 11:41:11 -070026 struct kref kref;
Davide Libenzie1ad7462007-05-10 22:23:19 -070027 wait_queue_head_t wqh;
28 /*
29 * Every time that a write(2) is performed on an eventfd, the
30 * value of the __u64 being written is added to "count" and a
31 * wakeup is performed on "wqh". A read(2) will return the "count"
32 * value to userspace, and will reset "count" to zero. The kernel
Davide Libenzi13389012009-06-30 11:41:11 -070033 * side eventfd_signal() also, adds to the "count" counter and
Davide Libenzie1ad7462007-05-10 22:23:19 -070034 * issue a wakeup.
35 */
36 __u64 count;
Davide Libenzibcd0b232009-03-31 15:24:18 -070037 unsigned int flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -070038};
39
Davide Libenzi13389012009-06-30 11:41:11 -070040/**
41 * eventfd_signal - Adds @n to the eventfd counter.
42 * @ctx: [in] Pointer to the eventfd context.
43 * @n: [in] Value of the counter to be added to the eventfd internal counter.
44 * The value cannot be negative.
45 *
46 * This function is supposed to be called by the kernel in paths that do not
47 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
48 * value, and we signal this as overflow condition by returining a POLLERR
49 * to poll(2).
50 *
Sha Zhengjuee62c6b2012-05-31 16:26:41 -070051 * Returns the amount by which the counter was incrememnted. This will be less
52 * than @n if the counter has overflowed.
Davide Libenzie1ad7462007-05-10 22:23:19 -070053 */
Sha Zhengjuee62c6b2012-05-31 16:26:41 -070054__u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
Davide Libenzie1ad7462007-05-10 22:23:19 -070055{
Davide Libenzie1ad7462007-05-10 22:23:19 -070056 unsigned long flags;
57
Davide Libenzid48eb232007-05-18 12:02:33 -070058 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070059 if (ULLONG_MAX - ctx->count < n)
Sha Zhengjuee62c6b2012-05-31 16:26:41 -070060 n = ULLONG_MAX - ctx->count;
Davide Libenzie1ad7462007-05-10 22:23:19 -070061 ctx->count += n;
62 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -070063 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzid48eb232007-05-18 12:02:33 -070064 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070065
66 return n;
67}
Rusty Russell57186072009-06-12 22:27:09 -060068EXPORT_SYMBOL_GPL(eventfd_signal);
Davide Libenzie1ad7462007-05-10 22:23:19 -070069
Davide Libenzi562787a2009-09-22 16:43:57 -070070static void eventfd_free_ctx(struct eventfd_ctx *ctx)
71{
72 kfree(ctx);
73}
74
Davide Libenzi13389012009-06-30 11:41:11 -070075static void eventfd_free(struct kref *kref)
76{
77 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
78
Davide Libenzi562787a2009-09-22 16:43:57 -070079 eventfd_free_ctx(ctx);
Davide Libenzi13389012009-06-30 11:41:11 -070080}
81
82/**
83 * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
84 * @ctx: [in] Pointer to the eventfd context.
85 *
86 * Returns: In case of success, returns a pointer to the eventfd context.
87 */
88struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
89{
90 kref_get(&ctx->kref);
91 return ctx;
92}
93EXPORT_SYMBOL_GPL(eventfd_ctx_get);
94
95/**
96 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
97 * @ctx: [in] Pointer to eventfd context.
98 *
99 * The eventfd context reference must have been previously acquired either
Randy Dunlap36182182011-02-20 20:08:35 -0800100 * with eventfd_ctx_get() or eventfd_ctx_fdget().
Davide Libenzi13389012009-06-30 11:41:11 -0700101 */
102void eventfd_ctx_put(struct eventfd_ctx *ctx)
103{
104 kref_put(&ctx->kref, eventfd_free);
105}
106EXPORT_SYMBOL_GPL(eventfd_ctx_put);
107
Davide Libenzie1ad7462007-05-10 22:23:19 -0700108static int eventfd_release(struct inode *inode, struct file *file)
109{
Davide Libenzi13389012009-06-30 11:41:11 -0700110 struct eventfd_ctx *ctx = file->private_data;
111
112 wake_up_poll(&ctx->wqh, POLLHUP);
113 eventfd_ctx_put(ctx);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700114 return 0;
115}
116
117static unsigned int eventfd_poll(struct file *file, poll_table *wait)
118{
119 struct eventfd_ctx *ctx = file->private_data;
120 unsigned int events = 0;
121 unsigned long flags;
122
123 poll_wait(file, &ctx->wqh, wait);
124
Davide Libenzid48eb232007-05-18 12:02:33 -0700125 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700126 if (ctx->count > 0)
127 events |= POLLIN;
128 if (ctx->count == ULLONG_MAX)
129 events |= POLLERR;
130 if (ULLONG_MAX - 1 > ctx->count)
131 events |= POLLOUT;
Davide Libenzid48eb232007-05-18 12:02:33 -0700132 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700133
134 return events;
135}
136
Davide Libenzicb289d62010-01-13 09:34:36 -0800137static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700138{
Davide Libenzicb289d62010-01-13 09:34:36 -0800139 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
140 ctx->count -= *cnt;
141}
142
143/**
144 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
145 * @ctx: [in] Pointer to eventfd context.
146 * @wait: [in] Wait queue to be removed.
Randy Dunlap36182182011-02-20 20:08:35 -0800147 * @cnt: [out] Pointer to the 64-bit counter value.
Davide Libenzicb289d62010-01-13 09:34:36 -0800148 *
Randy Dunlap36182182011-02-20 20:08:35 -0800149 * Returns %0 if successful, or the following error codes:
Davide Libenzicb289d62010-01-13 09:34:36 -0800150 *
151 * -EAGAIN : The operation would have blocked.
152 *
153 * This is used to atomically remove a wait queue entry from the eventfd wait
154 * queue head, and read/reset the counter value.
155 */
156int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_t *wait,
157 __u64 *cnt)
158{
159 unsigned long flags;
160
161 spin_lock_irqsave(&ctx->wqh.lock, flags);
162 eventfd_ctx_do_read(ctx, cnt);
163 __remove_wait_queue(&ctx->wqh, wait);
164 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
165 wake_up_locked_poll(&ctx->wqh, POLLOUT);
166 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
167
168 return *cnt != 0 ? 0 : -EAGAIN;
169}
170EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
171
172/**
173 * eventfd_ctx_read - Reads the eventfd counter or wait if it is zero.
174 * @ctx: [in] Pointer to eventfd context.
175 * @no_wait: [in] Different from zero if the operation should not block.
Randy Dunlap36182182011-02-20 20:08:35 -0800176 * @cnt: [out] Pointer to the 64-bit counter value.
Davide Libenzicb289d62010-01-13 09:34:36 -0800177 *
Randy Dunlap36182182011-02-20 20:08:35 -0800178 * Returns %0 if successful, or the following error codes:
Davide Libenzicb289d62010-01-13 09:34:36 -0800179 *
Randy Dunlap36182182011-02-20 20:08:35 -0800180 * -EAGAIN : The operation would have blocked but @no_wait was non-zero.
Davide Libenzicb289d62010-01-13 09:34:36 -0800181 * -ERESTARTSYS : A signal interrupted the wait operation.
182 *
183 * If @no_wait is zero, the function might sleep until the eventfd internal
184 * counter becomes greater than zero.
185 */
186ssize_t eventfd_ctx_read(struct eventfd_ctx *ctx, int no_wait, __u64 *cnt)
187{
Davide Libenzie1ad7462007-05-10 22:23:19 -0700188 ssize_t res;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700189 DECLARE_WAITQUEUE(wait, current);
190
Davide Libenzid48eb232007-05-18 12:02:33 -0700191 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzicb289d62010-01-13 09:34:36 -0800192 *cnt = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700193 res = -EAGAIN;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700194 if (ctx->count > 0)
Davide Libenzicb289d62010-01-13 09:34:36 -0800195 res = 0;
196 else if (!no_wait) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700197 __add_wait_queue(&ctx->wqh, &wait);
Davide Libenzicb289d62010-01-13 09:34:36 -0800198 for (;;) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700199 set_current_state(TASK_INTERRUPTIBLE);
200 if (ctx->count > 0) {
Davide Libenzicb289d62010-01-13 09:34:36 -0800201 res = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700202 break;
203 }
204 if (signal_pending(current)) {
205 res = -ERESTARTSYS;
206 break;
207 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700208 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700209 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700210 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700211 }
212 __remove_wait_queue(&ctx->wqh, &wait);
213 __set_current_state(TASK_RUNNING);
214 }
Davide Libenzicb289d62010-01-13 09:34:36 -0800215 if (likely(res == 0)) {
216 eventfd_ctx_do_read(ctx, cnt);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700217 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700218 wake_up_locked_poll(&ctx->wqh, POLLOUT);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700219 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700220 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700221
222 return res;
223}
Davide Libenzicb289d62010-01-13 09:34:36 -0800224EXPORT_SYMBOL_GPL(eventfd_ctx_read);
225
226static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
227 loff_t *ppos)
228{
229 struct eventfd_ctx *ctx = file->private_data;
230 ssize_t res;
231 __u64 cnt;
232
233 if (count < sizeof(cnt))
234 return -EINVAL;
235 res = eventfd_ctx_read(ctx, file->f_flags & O_NONBLOCK, &cnt);
236 if (res < 0)
237 return res;
238
239 return put_user(cnt, (__u64 __user *) buf) ? -EFAULT : sizeof(cnt);
240}
Davide Libenzie1ad7462007-05-10 22:23:19 -0700241
242static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
243 loff_t *ppos)
244{
245 struct eventfd_ctx *ctx = file->private_data;
246 ssize_t res;
247 __u64 ucnt;
248 DECLARE_WAITQUEUE(wait, current);
249
250 if (count < sizeof(ucnt))
251 return -EINVAL;
252 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
253 return -EFAULT;
254 if (ucnt == ULLONG_MAX)
255 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -0700256 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700257 res = -EAGAIN;
258 if (ULLONG_MAX - ctx->count > ucnt)
259 res = sizeof(ucnt);
260 else if (!(file->f_flags & O_NONBLOCK)) {
261 __add_wait_queue(&ctx->wqh, &wait);
262 for (res = 0;;) {
263 set_current_state(TASK_INTERRUPTIBLE);
264 if (ULLONG_MAX - ctx->count > ucnt) {
265 res = sizeof(ucnt);
266 break;
267 }
268 if (signal_pending(current)) {
269 res = -ERESTARTSYS;
270 break;
271 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700272 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700273 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700274 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700275 }
276 __remove_wait_queue(&ctx->wqh, &wait);
277 __set_current_state(TASK_RUNNING);
278 }
Davide Libenzibcd0b232009-03-31 15:24:18 -0700279 if (likely(res > 0)) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700280 ctx->count += ucnt;
281 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700282 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700283 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700284 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700285
286 return res;
287}
288
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800289#ifdef CONFIG_PROC_FS
290static int eventfd_show_fdinfo(struct seq_file *m, struct file *f)
291{
292 struct eventfd_ctx *ctx = f->private_data;
293 int ret;
294
295 spin_lock_irq(&ctx->wqh.lock);
296 ret = seq_printf(m, "eventfd-count: %16llx\n",
297 (unsigned long long)ctx->count);
298 spin_unlock_irq(&ctx->wqh.lock);
299
300 return ret;
301}
302#endif
303
Davide Libenzie1ad7462007-05-10 22:23:19 -0700304static const struct file_operations eventfd_fops = {
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800305#ifdef CONFIG_PROC_FS
306 .show_fdinfo = eventfd_show_fdinfo,
307#endif
Davide Libenzie1ad7462007-05-10 22:23:19 -0700308 .release = eventfd_release,
309 .poll = eventfd_poll,
310 .read = eventfd_read,
311 .write = eventfd_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200312 .llseek = noop_llseek,
Davide Libenzie1ad7462007-05-10 22:23:19 -0700313};
314
Davide Libenzi13389012009-06-30 11:41:11 -0700315/**
316 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
317 * @fd: [in] Eventfd file descriptor.
318 *
319 * Returns a pointer to the eventfd file structure in case of success, or the
320 * following error pointer:
321 *
322 * -EBADF : Invalid @fd file descriptor.
323 * -EINVAL : The @fd file descriptor is not an eventfd file.
324 */
Davide Libenzie1ad7462007-05-10 22:23:19 -0700325struct file *eventfd_fget(int fd)
326{
327 struct file *file;
328
329 file = fget(fd);
330 if (!file)
331 return ERR_PTR(-EBADF);
332 if (file->f_op != &eventfd_fops) {
333 fput(file);
334 return ERR_PTR(-EINVAL);
335 }
336
337 return file;
338}
Rusty Russell57186072009-06-12 22:27:09 -0600339EXPORT_SYMBOL_GPL(eventfd_fget);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700340
Davide Libenzi13389012009-06-30 11:41:11 -0700341/**
342 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
343 * @fd: [in] Eventfd file descriptor.
344 *
345 * Returns a pointer to the internal eventfd context, otherwise the error
346 * pointers returned by the following functions:
347 *
348 * eventfd_fget
349 */
350struct eventfd_ctx *eventfd_ctx_fdget(int fd)
351{
352 struct file *file;
353 struct eventfd_ctx *ctx;
354
355 file = eventfd_fget(fd);
356 if (IS_ERR(file))
357 return (struct eventfd_ctx *) file;
358 ctx = eventfd_ctx_get(file->private_data);
359 fput(file);
360
361 return ctx;
362}
363EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
364
365/**
366 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
367 * @file: [in] Eventfd file pointer.
368 *
369 * Returns a pointer to the internal eventfd context, otherwise the error
370 * pointer:
371 *
372 * -EINVAL : The @fd file descriptor is not an eventfd file.
373 */
374struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
375{
376 if (file->f_op != &eventfd_fops)
377 return ERR_PTR(-EINVAL);
378
379 return eventfd_ctx_get(file->private_data);
380}
381EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
382
Davide Libenzi562787a2009-09-22 16:43:57 -0700383/**
384 * eventfd_file_create - Creates an eventfd file pointer.
385 * @count: Initial eventfd counter value.
386 * @flags: Flags for the eventfd file.
387 *
388 * This function creates an eventfd file pointer, w/out installing it into
389 * the fd table. This is useful when the eventfd file is used during the
390 * initialization of data structures that require extra setup after the eventfd
391 * creation. So the eventfd creation is split into the file pointer creation
392 * phase, and the file descriptor installation phase.
393 * In this way races with userspace closing the newly installed file descriptor
394 * can be avoided.
395 * Returns an eventfd file pointer, or a proper error pointer.
396 */
397struct file *eventfd_file_create(unsigned int count, int flags)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700398{
Davide Libenzi562787a2009-09-22 16:43:57 -0700399 struct file *file;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700400 struct eventfd_ctx *ctx;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700401
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700402 /* Check the EFD_* constants for consistency. */
403 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
404 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
405
Davide Libenzibcd0b232009-03-31 15:24:18 -0700406 if (flags & ~EFD_FLAGS_SET)
Davide Libenzi562787a2009-09-22 16:43:57 -0700407 return ERR_PTR(-EINVAL);
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700408
Davide Libenzie1ad7462007-05-10 22:23:19 -0700409 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
410 if (!ctx)
Davide Libenzi562787a2009-09-22 16:43:57 -0700411 return ERR_PTR(-ENOMEM);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700412
Davide Libenzi13389012009-06-30 11:41:11 -0700413 kref_init(&ctx->kref);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700414 init_waitqueue_head(&ctx->wqh);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700415 ctx->count = count;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700416 ctx->flags = flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700417
Davide Libenzi562787a2009-09-22 16:43:57 -0700418 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
Roland Dreier628ff7c2009-12-18 09:41:24 -0800419 O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
Davide Libenzi562787a2009-09-22 16:43:57 -0700420 if (IS_ERR(file))
421 eventfd_free_ctx(ctx);
422
423 return file;
424}
425
426SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
427{
428 int fd, error;
429 struct file *file;
430
431 error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
432 if (error < 0)
433 return error;
434 fd = error;
435
436 file = eventfd_file_create(count, flags);
437 if (IS_ERR(file)) {
438 error = PTR_ERR(file);
439 goto err_put_unused_fd;
440 }
441 fd_install(fd, file);
442
Al Viro2030a422008-02-23 06:46:49 -0500443 return fd;
Davide Libenzi562787a2009-09-22 16:43:57 -0700444
445err_put_unused_fd:
446 put_unused_fd(fd);
447
448 return error;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700449}
450
Heiko Carstensd4e82042009-01-14 14:14:34 +0100451SYSCALL_DEFINE1(eventfd, unsigned int, count)
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700452{
453 return sys_eventfd2(count, 0);
454}
Davide Libenzibcd0b232009-03-31 15:24:18 -0700455