blob: 249ca6c0b7843ff1fa9d102ad14c15e54c0ad578 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Davide Libenzie1ad7462007-05-10 22:23:19 -07002/*
3 * fs/eventfd.c
4 *
5 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
6 *
7 */
8
9#include <linux/file.h>
10#include <linux/poll.h>
11#include <linux/init.h>
12#include <linux/fs.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010013#include <linux/sched/signal.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070014#include <linux/kernel.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090015#include <linux/slab.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070016#include <linux/list.h>
17#include <linux/spinlock.h>
18#include <linux/anon_inodes.h>
Adrian Bunk7747cdb2008-02-06 01:36:49 -080019#include <linux/syscalls.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050020#include <linux/export.h>
Davide Libenzi13389012009-06-30 11:41:11 -070021#include <linux/kref.h>
22#include <linux/eventfd.h>
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -080023#include <linux/proc_fs.h>
24#include <linux/seq_file.h>
Masatake YAMATOb556db12019-05-14 15:45:19 -070025#include <linux/idr.h>
Jens Axboe12aceb82020-05-01 13:11:09 -060026#include <linux/uio.h>
Masatake YAMATOb556db12019-05-14 15:45:19 -070027
YueHaibingce528c42019-05-14 15:45:22 -070028static DEFINE_IDA(eventfd_ida);
Davide Libenzie1ad7462007-05-10 22:23:19 -070029
30struct eventfd_ctx {
Davide Libenzi13389012009-06-30 11:41:11 -070031 struct kref kref;
Davide Libenzie1ad7462007-05-10 22:23:19 -070032 wait_queue_head_t wqh;
33 /*
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
Davide Libenzi13389012009-06-30 11:41:11 -070038 * side eventfd_signal() also, adds to the "count" counter and
Davide Libenzie1ad7462007-05-10 22:23:19 -070039 * issue a wakeup.
40 */
41 __u64 count;
Davide Libenzibcd0b232009-03-31 15:24:18 -070042 unsigned int flags;
Masatake YAMATOb556db12019-05-14 15:45:19 -070043 int id;
Davide Libenzie1ad7462007-05-10 22:23:19 -070044};
45
Jens Axboe03e02ac2022-11-20 10:13:44 -070046__u64 eventfd_signal_mask(struct eventfd_ctx *ctx, __u64 n, unsigned mask)
Davide Libenzie1ad7462007-05-10 22:23:19 -070047{
Davide Libenzie1ad7462007-05-10 22:23:19 -070048 unsigned long flags;
49
Jens Axboeb5e683d2020-02-02 08:23:03 -070050 /*
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
Thomas Gleixnerb542e382021-07-29 13:01:59 +020054 * check eventfd_signal_allowed() before calling this function. If
55 * it returns false, the eventfd_signal() call should be deferred to a
Jens Axboeb5e683d2020-02-02 08:23:03 -070056 * safe context.
57 */
Dylan Yudaken9f0deaa2022-08-16 06:59:59 -070058 if (WARN_ON_ONCE(current->in_eventfd))
Jens Axboeb5e683d2020-02-02 08:23:03 -070059 return 0;
60
Davide Libenzid48eb232007-05-18 12:02:33 -070061 spin_lock_irqsave(&ctx->wqh.lock, flags);
Dylan Yudaken9f0deaa2022-08-16 06:59:59 -070062 current->in_eventfd = 1;
Davide Libenzie1ad7462007-05-10 22:23:19 -070063 if (ULLONG_MAX - ctx->count < n)
Sha Zhengjuee62c6b2012-05-31 16:26:41 -070064 n = ULLONG_MAX - ctx->count;
Davide Libenzie1ad7462007-05-10 22:23:19 -070065 ctx->count += n;
66 if (waitqueue_active(&ctx->wqh))
Jens Axboe03e02ac2022-11-20 10:13:44 -070067 wake_up_locked_poll(&ctx->wqh, EPOLLIN | mask);
Dylan Yudaken9f0deaa2022-08-16 06:59:59 -070068 current->in_eventfd = 0;
Davide Libenzid48eb232007-05-18 12:02:33 -070069 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070070
71 return n;
72}
Jens Axboe03e02ac2022-11-20 10:13:44 -070073
74/**
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.
79 *
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
83 * to poll(2).
84 *
85 * Returns the amount by which the counter was incremented. This will be less
86 * than @n if the counter has overflowed.
87 */
88__u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
89{
90 return eventfd_signal_mask(ctx, n, 0);
91}
Rusty Russell57186072009-06-12 22:27:09 -060092EXPORT_SYMBOL_GPL(eventfd_signal);
Davide Libenzie1ad7462007-05-10 22:23:19 -070093
Davide Libenzi562787a2009-09-22 16:43:57 -070094static void eventfd_free_ctx(struct eventfd_ctx *ctx)
95{
Masatake YAMATOb556db12019-05-14 15:45:19 -070096 if (ctx->id >= 0)
97 ida_simple_remove(&eventfd_ida, ctx->id);
Davide Libenzi562787a2009-09-22 16:43:57 -070098 kfree(ctx);
99}
100
Davide Libenzi13389012009-06-30 11:41:11 -0700101static void eventfd_free(struct kref *kref)
102{
103 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
104
Davide Libenzi562787a2009-09-22 16:43:57 -0700105 eventfd_free_ctx(ctx);
Davide Libenzi13389012009-06-30 11:41:11 -0700106}
107
108/**
Davide Libenzi13389012009-06-30 11:41:11 -0700109 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
110 * @ctx: [in] Pointer to eventfd context.
111 *
112 * The eventfd context reference must have been previously acquired either
Eric Biggers105f2b72018-01-06 09:45:44 -0800113 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
Davide Libenzi13389012009-06-30 11:41:11 -0700114 */
115void eventfd_ctx_put(struct eventfd_ctx *ctx)
116{
117 kref_put(&ctx->kref, eventfd_free);
118}
119EXPORT_SYMBOL_GPL(eventfd_ctx_put);
120
Davide Libenzie1ad7462007-05-10 22:23:19 -0700121static int eventfd_release(struct inode *inode, struct file *file)
122{
Davide Libenzi13389012009-06-30 11:41:11 -0700123 struct eventfd_ctx *ctx = file->private_data;
124
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800125 wake_up_poll(&ctx->wqh, EPOLLHUP);
Davide Libenzi13389012009-06-30 11:41:11 -0700126 eventfd_ctx_put(ctx);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700127 return 0;
128}
129
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700130static __poll_t eventfd_poll(struct file *file, poll_table *wait)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700131{
132 struct eventfd_ctx *ctx = file->private_data;
Al Viro076ccb72017-07-03 01:02:18 -0400133 __poll_t events = 0;
Chris Masone22553e2015-02-17 13:46:07 -0800134 u64 count;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700135
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700136 poll_wait(file, &ctx->wqh, wait);
137
Paolo Bonzinia484c3d2016-03-22 14:27:14 -0700138 /*
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.
142 *
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:
147 *
148 * poll write
149 * ----------------- ------------
150 * lock ctx->wqh.lock (in poll_wait)
151 * count = ctx->count
152 * __add_wait_queue
153 * unlock ctx->wqh.lock
154 * lock ctx->qwh.lock
155 * ctx->count += n
156 * if (waitqueue_active)
157 * wake_up_locked_poll
158 * unlock ctx->qwh.lock
159 * eventfd_poll returns 0
160 *
161 * but the following, which would miss a wakeup, cannot happen:
162 *
163 * poll write
164 * ----------------- ------------
165 * count = ctx->count (INVALID!)
166 * lock ctx->qwh.lock
167 * ctx->count += n
168 * **waitqueue_active is false**
169 * **no wake_up_locked_poll!**
170 * unlock ctx->qwh.lock
171 * lock ctx->wqh.lock (in poll_wait)
172 * __add_wait_queue
173 * unlock ctx->wqh.lock
174 * eventfd_poll returns 0
175 */
176 count = READ_ONCE(ctx->count);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700177
Chris Masone22553e2015-02-17 13:46:07 -0800178 if (count > 0)
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700179 events |= EPOLLIN;
Chris Masone22553e2015-02-17 13:46:07 -0800180 if (count == ULLONG_MAX)
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800181 events |= EPOLLERR;
Chris Masone22553e2015-02-17 13:46:07 -0800182 if (ULLONG_MAX - 1 > count)
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700183 events |= EPOLLOUT;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700184
185 return events;
186}
187
David Woodhouse28f13262020-10-27 13:55:21 +0000188void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700189{
David Woodhouse28f13262020-10-27 13:55:21 +0000190 lockdep_assert_held(&ctx->wqh.lock);
191
Davide Libenzicb289d62010-01-13 09:34:36 -0800192 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
193 ctx->count -= *cnt;
194}
David Woodhouse28f13262020-10-27 13:55:21 +0000195EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
Davide Libenzicb289d62010-01-13 09:34:36 -0800196
197/**
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.
Randy Dunlap36182182011-02-20 20:08:35 -0800201 * @cnt: [out] Pointer to the 64-bit counter value.
Davide Libenzicb289d62010-01-13 09:34:36 -0800202 *
Randy Dunlap36182182011-02-20 20:08:35 -0800203 * Returns %0 if successful, or the following error codes:
Davide Libenzicb289d62010-01-13 09:34:36 -0800204 *
205 * -EAGAIN : The operation would have blocked.
206 *
207 * This is used to atomically remove a wait queue entry from the eventfd wait
208 * queue head, and read/reset the counter value.
209 */
Ingo Molnarac6424b2017-06-20 12:06:13 +0200210int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
Davide Libenzicb289d62010-01-13 09:34:36 -0800211 __u64 *cnt)
212{
213 unsigned long flags;
214
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))
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800219 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
Davide Libenzicb289d62010-01-13 09:34:36 -0800220 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
221
222 return *cnt != 0 ? 0 : -EAGAIN;
223}
224EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
225
Jens Axboe12aceb82020-05-01 13:11:09 -0600226static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to)
Davide Libenzicb289d62010-01-13 09:34:36 -0800227{
Jens Axboe12aceb82020-05-01 13:11:09 -0600228 struct file *file = iocb->ki_filp;
Eric Biggersb6364572018-01-06 09:45:43 -0800229 struct eventfd_ctx *ctx = file->private_data;
Eric Biggersb6364572018-01-06 09:45:43 -0800230 __u64 ucnt = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700231 DECLARE_WAITQUEUE(wait, current);
232
Jens Axboe12aceb82020-05-01 13:11:09 -0600233 if (iov_iter_count(to) < sizeof(ucnt))
Eric Biggersb6364572018-01-06 09:45:43 -0800234 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -0700235 spin_lock_irq(&ctx->wqh.lock);
Jens Axboe12aceb82020-05-01 13:11:09 -0600236 if (!ctx->count) {
237 if ((file->f_flags & O_NONBLOCK) ||
238 (iocb->ki_flags & IOCB_NOWAIT)) {
239 spin_unlock_irq(&ctx->wqh.lock);
240 return -EAGAIN;
241 }
Davide Libenzie1ad7462007-05-10 22:23:19 -0700242 __add_wait_queue(&ctx->wqh, &wait);
Davide Libenzicb289d62010-01-13 09:34:36 -0800243 for (;;) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700244 set_current_state(TASK_INTERRUPTIBLE);
Jens Axboe12aceb82020-05-01 13:11:09 -0600245 if (ctx->count)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700246 break;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700247 if (signal_pending(current)) {
Jens Axboe12aceb82020-05-01 13:11:09 -0600248 __remove_wait_queue(&ctx->wqh, &wait);
249 __set_current_state(TASK_RUNNING);
250 spin_unlock_irq(&ctx->wqh.lock);
251 return -ERESTARTSYS;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700252 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700253 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700254 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700255 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700256 }
257 __remove_wait_queue(&ctx->wqh, &wait);
258 __set_current_state(TASK_RUNNING);
259 }
Jens Axboe12aceb82020-05-01 13:11:09 -0600260 eventfd_ctx_do_read(ctx, &ucnt);
Dylan Yudaken9f0deaa2022-08-16 06:59:59 -0700261 current->in_eventfd = 1;
Jens Axboe12aceb82020-05-01 13:11:09 -0600262 if (waitqueue_active(&ctx->wqh))
263 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
Dylan Yudaken9f0deaa2022-08-16 06:59:59 -0700264 current->in_eventfd = 0;
Davide Libenzid48eb232007-05-18 12:02:33 -0700265 spin_unlock_irq(&ctx->wqh.lock);
Jens Axboe12aceb82020-05-01 13:11:09 -0600266 if (unlikely(copy_to_iter(&ucnt, sizeof(ucnt), to) != sizeof(ucnt)))
Eric Biggersb6364572018-01-06 09:45:43 -0800267 return -EFAULT;
268
Jens Axboe12aceb82020-05-01 13:11:09 -0600269 return sizeof(ucnt);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700270}
271
272static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
273 loff_t *ppos)
274{
275 struct eventfd_ctx *ctx = file->private_data;
276 ssize_t res;
277 __u64 ucnt;
278 DECLARE_WAITQUEUE(wait, current);
279
280 if (count < sizeof(ucnt))
281 return -EINVAL;
282 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
283 return -EFAULT;
284 if (ucnt == ULLONG_MAX)
285 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -0700286 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700287 res = -EAGAIN;
288 if (ULLONG_MAX - ctx->count > ucnt)
289 res = sizeof(ucnt);
290 else if (!(file->f_flags & O_NONBLOCK)) {
291 __add_wait_queue(&ctx->wqh, &wait);
292 for (res = 0;;) {
293 set_current_state(TASK_INTERRUPTIBLE);
294 if (ULLONG_MAX - ctx->count > ucnt) {
295 res = sizeof(ucnt);
296 break;
297 }
298 if (signal_pending(current)) {
299 res = -ERESTARTSYS;
300 break;
301 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700302 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700303 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700304 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700305 }
306 __remove_wait_queue(&ctx->wqh, &wait);
307 __set_current_state(TASK_RUNNING);
308 }
Davide Libenzibcd0b232009-03-31 15:24:18 -0700309 if (likely(res > 0)) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700310 ctx->count += ucnt;
Dylan Yudaken9f0deaa2022-08-16 06:59:59 -0700311 current->in_eventfd = 1;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700312 if (waitqueue_active(&ctx->wqh))
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800313 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
Dylan Yudaken9f0deaa2022-08-16 06:59:59 -0700314 current->in_eventfd = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700315 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700316 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700317
318 return res;
319}
320
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800321#ifdef CONFIG_PROC_FS
Joe Perchesa3816ab2014-09-29 16:08:25 -0700322static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800323{
324 struct eventfd_ctx *ctx = f->private_data;
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800325
326 spin_lock_irq(&ctx->wqh.lock);
Joe Perchesa3816ab2014-09-29 16:08:25 -0700327 seq_printf(m, "eventfd-count: %16llx\n",
328 (unsigned long long)ctx->count);
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800329 spin_unlock_irq(&ctx->wqh.lock);
Masatake YAMATOb556db12019-05-14 15:45:19 -0700330 seq_printf(m, "eventfd-id: %d\n", ctx->id);
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800331}
332#endif
333
Davide Libenzie1ad7462007-05-10 22:23:19 -0700334static const struct file_operations eventfd_fops = {
Cyrill Gorcunovcbac5542012-12-17 16:04:57 -0800335#ifdef CONFIG_PROC_FS
336 .show_fdinfo = eventfd_show_fdinfo,
337#endif
Davide Libenzie1ad7462007-05-10 22:23:19 -0700338 .release = eventfd_release,
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700339 .poll = eventfd_poll,
Jens Axboe12aceb82020-05-01 13:11:09 -0600340 .read_iter = eventfd_read,
Davide Libenzie1ad7462007-05-10 22:23:19 -0700341 .write = eventfd_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200342 .llseek = noop_llseek,
Davide Libenzie1ad7462007-05-10 22:23:19 -0700343};
344
Davide Libenzi13389012009-06-30 11:41:11 -0700345/**
346 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
347 * @fd: [in] Eventfd file descriptor.
348 *
349 * Returns a pointer to the eventfd file structure in case of success, or the
350 * following error pointer:
351 *
352 * -EBADF : Invalid @fd file descriptor.
353 * -EINVAL : The @fd file descriptor is not an eventfd file.
354 */
Davide Libenzie1ad7462007-05-10 22:23:19 -0700355struct file *eventfd_fget(int fd)
356{
357 struct file *file;
358
359 file = fget(fd);
360 if (!file)
361 return ERR_PTR(-EBADF);
362 if (file->f_op != &eventfd_fops) {
363 fput(file);
364 return ERR_PTR(-EINVAL);
365 }
366
367 return file;
368}
Rusty Russell57186072009-06-12 22:27:09 -0600369EXPORT_SYMBOL_GPL(eventfd_fget);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700370
Davide Libenzi13389012009-06-30 11:41:11 -0700371/**
372 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
373 * @fd: [in] Eventfd file descriptor.
374 *
375 * Returns a pointer to the internal eventfd context, otherwise the error
376 * pointers returned by the following functions:
377 *
378 * eventfd_fget
379 */
380struct eventfd_ctx *eventfd_ctx_fdget(int fd)
381{
Davide Libenzi13389012009-06-30 11:41:11 -0700382 struct eventfd_ctx *ctx;
Al Viro36a74112013-12-23 16:51:33 -0500383 struct fd f = fdget(fd);
384 if (!f.file)
385 return ERR_PTR(-EBADF);
386 ctx = eventfd_ctx_fileget(f.file);
387 fdput(f);
Davide Libenzi13389012009-06-30 11:41:11 -0700388 return ctx;
389}
390EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
391
392/**
393 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
394 * @file: [in] Eventfd file pointer.
395 *
396 * Returns a pointer to the internal eventfd context, otherwise the error
397 * pointer:
398 *
399 * -EINVAL : The @fd file descriptor is not an eventfd file.
400 */
401struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
402{
Eric Biggers105f2b72018-01-06 09:45:44 -0800403 struct eventfd_ctx *ctx;
404
Davide Libenzi13389012009-06-30 11:41:11 -0700405 if (file->f_op != &eventfd_fops)
406 return ERR_PTR(-EINVAL);
407
Eric Biggers105f2b72018-01-06 09:45:44 -0800408 ctx = file->private_data;
409 kref_get(&ctx->kref);
410 return ctx;
Davide Libenzi13389012009-06-30 11:41:11 -0700411}
412EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
413
Dominik Brodowski2fc96f82018-03-11 11:34:37 +0100414static int do_eventfd(unsigned int count, int flags)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700415{
Davide Libenzie1ad7462007-05-10 22:23:19 -0700416 struct eventfd_ctx *ctx;
Jens Axboe12aceb82020-05-01 13:11:09 -0600417 struct file *file;
Eric Biggers7d815162018-01-06 09:45:42 -0800418 int fd;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700419
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700420 /* Check the EFD_* constants for consistency. */
421 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
422 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
423
Davide Libenzibcd0b232009-03-31 15:24:18 -0700424 if (flags & ~EFD_FLAGS_SET)
Eric Biggers7d815162018-01-06 09:45:42 -0800425 return -EINVAL;
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700426
Davide Libenzie1ad7462007-05-10 22:23:19 -0700427 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
428 if (!ctx)
Eric Biggers7d815162018-01-06 09:45:42 -0800429 return -ENOMEM;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700430
Davide Libenzi13389012009-06-30 11:41:11 -0700431 kref_init(&ctx->kref);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700432 init_waitqueue_head(&ctx->wqh);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700433 ctx->count = count;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700434 ctx->flags = flags;
Masatake YAMATOb556db12019-05-14 15:45:19 -0700435 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700436
Jens Axboe12aceb82020-05-01 13:11:09 -0600437 flags &= EFD_SHARED_FCNTL_FLAGS;
438 flags |= O_RDWR;
439 fd = get_unused_fd_flags(flags);
Eric Biggers7d815162018-01-06 09:45:42 -0800440 if (fd < 0)
Jens Axboe12aceb82020-05-01 13:11:09 -0600441 goto err;
Davide Libenzi562787a2009-09-22 16:43:57 -0700442
Jens Axboe12aceb82020-05-01 13:11:09 -0600443 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags);
444 if (IS_ERR(file)) {
445 put_unused_fd(fd);
446 fd = PTR_ERR(file);
447 goto err;
448 }
449
450 file->f_mode |= FMODE_NOWAIT;
451 fd_install(fd, file);
452 return fd;
453err:
454 eventfd_free_ctx(ctx);
Al Viro2030a422008-02-23 06:46:49 -0500455 return fd;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700456}
457
Dominik Brodowski2fc96f82018-03-11 11:34:37 +0100458SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
459{
460 return do_eventfd(count, flags);
461}
462
Heiko Carstensd4e82042009-01-14 14:14:34 +0100463SYSCALL_DEFINE1(eventfd, unsigned int, count)
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700464{
Dominik Brodowski2fc96f82018-03-11 11:34:37 +0100465 return do_eventfd(count, 0);
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700466}
Davide Libenzibcd0b232009-03-31 15:24:18 -0700467