1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
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>
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
27 DEFINE_PER_CPU(int, eventfd_wake_count);
28
29 static DEFINE_IDA(eventfd_ida);
30
31 struct eventfd_ctx {
32 struct kref kref;
33 wait_queue_head_t wqh;
34 /*
35 * Every time that a write(2) is performed on an eventfd, the
36 * value of the __u64 being written is added to "count" and a
37 * wakeup is performed on "wqh". A read(2) will return the "count"
38 * value to userspace, and will reset "count" to zero. The kernel
39 * side eventfd_signal() also, adds to the "count" counter and
40 * issue a wakeup.
41 */
42 __u64 count;
43 unsigned int flags;
44 int id;
45 };
46
47 /**
48 * eventfd_signal - Adds @n to the eventfd counter.
49 * @ctx: [in] Pointer to the eventfd context.
50 * @n: [in] Value of the counter to be added to the eventfd internal counter.
51 * The value cannot be negative.
52 *
53 * This function is supposed to be called by the kernel in paths that do not
54 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
55 * value, and we signal this as overflow condition by returning a EPOLLERR
56 * to poll(2).
57 *
58 * Returns the amount by which the counter was incremented. This will be less
59 * than @n if the counter has overflowed.
60 */
eventfd_signal(struct eventfd_ctx * ctx,__u64 n)61 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
62 {
63 unsigned long flags;
64
65 /*
66 * Deadlock or stack overflow issues can happen if we recurse here
67 * through waitqueue wakeup handlers. If the caller users potentially
68 * nested waitqueues with custom wakeup handlers, then it should
69 * check eventfd_signal_count() before calling this function. If
70 * it returns true, the eventfd_signal() call should be deferred to a
71 * safe context.
72 */
73 if (WARN_ON_ONCE(this_cpu_read(eventfd_wake_count)))
74 return 0;
75
76 spin_lock_irqsave(&ctx->wqh.lock, flags);
77 this_cpu_inc(eventfd_wake_count);
78 if (ULLONG_MAX - ctx->count < n)
79 n = ULLONG_MAX - ctx->count;
80 ctx->count += n;
81 if (waitqueue_active(&ctx->wqh))
82 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
83 this_cpu_dec(eventfd_wake_count);
84 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
85
86 return n;
87 }
88 EXPORT_SYMBOL_GPL(eventfd_signal);
89
eventfd_free_ctx(struct eventfd_ctx * ctx)90 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
91 {
92 if (ctx->id >= 0)
93 ida_simple_remove(&eventfd_ida, ctx->id);
94 kfree(ctx);
95 }
96
eventfd_free(struct kref * kref)97 static void eventfd_free(struct kref *kref)
98 {
99 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
100
101 eventfd_free_ctx(ctx);
102 }
103
104 /**
105 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
106 * @ctx: [in] Pointer to eventfd context.
107 *
108 * The eventfd context reference must have been previously acquired either
109 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
110 */
eventfd_ctx_put(struct eventfd_ctx * ctx)111 void eventfd_ctx_put(struct eventfd_ctx *ctx)
112 {
113 kref_put(&ctx->kref, eventfd_free);
114 }
115 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
116
eventfd_release(struct inode * inode,struct file * file)117 static int eventfd_release(struct inode *inode, struct file *file)
118 {
119 struct eventfd_ctx *ctx = file->private_data;
120
121 wake_up_poll(&ctx->wqh, EPOLLHUP);
122 eventfd_ctx_put(ctx);
123 return 0;
124 }
125
eventfd_poll(struct file * file,poll_table * wait)126 static __poll_t eventfd_poll(struct file *file, poll_table *wait)
127 {
128 struct eventfd_ctx *ctx = file->private_data;
129 __poll_t events = 0;
130 u64 count;
131
132 poll_wait(file, &ctx->wqh, wait);
133
134 /*
135 * All writes to ctx->count occur within ctx->wqh.lock. This read
136 * can be done outside ctx->wqh.lock because we know that poll_wait
137 * takes that lock (through add_wait_queue) if our caller will sleep.
138 *
139 * The read _can_ therefore seep into add_wait_queue's critical
140 * section, but cannot move above it! add_wait_queue's spin_lock acts
141 * as an acquire barrier and ensures that the read be ordered properly
142 * against the writes. The following CAN happen and is safe:
143 *
144 * poll write
145 * ----------------- ------------
146 * lock ctx->wqh.lock (in poll_wait)
147 * count = ctx->count
148 * __add_wait_queue
149 * unlock ctx->wqh.lock
150 * lock ctx->qwh.lock
151 * ctx->count += n
152 * if (waitqueue_active)
153 * wake_up_locked_poll
154 * unlock ctx->qwh.lock
155 * eventfd_poll returns 0
156 *
157 * but the following, which would miss a wakeup, cannot happen:
158 *
159 * poll write
160 * ----------------- ------------
161 * count = ctx->count (INVALID!)
162 * lock ctx->qwh.lock
163 * ctx->count += n
164 * **waitqueue_active is false**
165 * **no wake_up_locked_poll!**
166 * unlock ctx->qwh.lock
167 * lock ctx->wqh.lock (in poll_wait)
168 * __add_wait_queue
169 * unlock ctx->wqh.lock
170 * eventfd_poll returns 0
171 */
172 count = READ_ONCE(ctx->count);
173
174 if (count > 0)
175 events |= EPOLLIN;
176 if (count == ULLONG_MAX)
177 events |= EPOLLERR;
178 if (ULLONG_MAX - 1 > count)
179 events |= EPOLLOUT;
180
181 return events;
182 }
183
eventfd_ctx_do_read(struct eventfd_ctx * ctx,__u64 * cnt)184 void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
185 {
186 lockdep_assert_held(&ctx->wqh.lock);
187
188 *cnt = ((ctx->flags & EFD_SEMAPHORE) && ctx->count) ? 1 : ctx->count;
189 ctx->count -= *cnt;
190 }
191 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
192
193 /**
194 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
195 * @ctx: [in] Pointer to eventfd context.
196 * @wait: [in] Wait queue to be removed.
197 * @cnt: [out] Pointer to the 64-bit counter value.
198 *
199 * Returns %0 if successful, or the following error codes:
200 *
201 * -EAGAIN : The operation would have blocked.
202 *
203 * This is used to atomically remove a wait queue entry from the eventfd wait
204 * queue head, and read/reset the counter value.
205 */
eventfd_ctx_remove_wait_queue(struct eventfd_ctx * ctx,wait_queue_entry_t * wait,__u64 * cnt)206 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
207 __u64 *cnt)
208 {
209 unsigned long flags;
210
211 spin_lock_irqsave(&ctx->wqh.lock, flags);
212 eventfd_ctx_do_read(ctx, cnt);
213 __remove_wait_queue(&ctx->wqh, wait);
214 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
215 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
216 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
217
218 return *cnt != 0 ? 0 : -EAGAIN;
219 }
220 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
221
eventfd_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)222 static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
223 loff_t *ppos)
224 {
225 struct eventfd_ctx *ctx = file->private_data;
226 ssize_t res;
227 __u64 ucnt = 0;
228 DECLARE_WAITQUEUE(wait, current);
229
230 if (count < sizeof(ucnt))
231 return -EINVAL;
232
233 spin_lock_irq(&ctx->wqh.lock);
234 res = -EAGAIN;
235 if (ctx->count > 0)
236 res = sizeof(ucnt);
237 else if (!(file->f_flags & O_NONBLOCK)) {
238 __add_wait_queue(&ctx->wqh, &wait);
239 for (;;) {
240 set_current_state(TASK_INTERRUPTIBLE);
241 if (ctx->count > 0) {
242 res = sizeof(ucnt);
243 break;
244 }
245 if (signal_pending(current)) {
246 res = -ERESTARTSYS;
247 break;
248 }
249 spin_unlock_irq(&ctx->wqh.lock);
250 schedule();
251 spin_lock_irq(&ctx->wqh.lock);
252 }
253 __remove_wait_queue(&ctx->wqh, &wait);
254 __set_current_state(TASK_RUNNING);
255 }
256 if (likely(res > 0)) {
257 eventfd_ctx_do_read(ctx, &ucnt);
258 if (waitqueue_active(&ctx->wqh))
259 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
260 }
261 spin_unlock_irq(&ctx->wqh.lock);
262
263 if (res > 0 && put_user(ucnt, (__u64 __user *)buf))
264 return -EFAULT;
265
266 return res;
267 }
268
eventfd_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)269 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
270 loff_t *ppos)
271 {
272 struct eventfd_ctx *ctx = file->private_data;
273 ssize_t res;
274 __u64 ucnt;
275 DECLARE_WAITQUEUE(wait, current);
276
277 if (count < sizeof(ucnt))
278 return -EINVAL;
279 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
280 return -EFAULT;
281 if (ucnt == ULLONG_MAX)
282 return -EINVAL;
283 spin_lock_irq(&ctx->wqh.lock);
284 res = -EAGAIN;
285 if (ULLONG_MAX - ctx->count > ucnt)
286 res = sizeof(ucnt);
287 else if (!(file->f_flags & O_NONBLOCK)) {
288 __add_wait_queue(&ctx->wqh, &wait);
289 for (res = 0;;) {
290 set_current_state(TASK_INTERRUPTIBLE);
291 if (ULLONG_MAX - ctx->count > ucnt) {
292 res = sizeof(ucnt);
293 break;
294 }
295 if (signal_pending(current)) {
296 res = -ERESTARTSYS;
297 break;
298 }
299 spin_unlock_irq(&ctx->wqh.lock);
300 schedule();
301 spin_lock_irq(&ctx->wqh.lock);
302 }
303 __remove_wait_queue(&ctx->wqh, &wait);
304 __set_current_state(TASK_RUNNING);
305 }
306 if (likely(res > 0)) {
307 ctx->count += ucnt;
308 if (waitqueue_active(&ctx->wqh))
309 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
310 }
311 spin_unlock_irq(&ctx->wqh.lock);
312
313 return res;
314 }
315
316 #ifdef CONFIG_PROC_FS
eventfd_show_fdinfo(struct seq_file * m,struct file * f)317 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
318 {
319 struct eventfd_ctx *ctx = f->private_data;
320
321 spin_lock_irq(&ctx->wqh.lock);
322 seq_printf(m, "eventfd-count: %16llx\n",
323 (unsigned long long)ctx->count);
324 spin_unlock_irq(&ctx->wqh.lock);
325 seq_printf(m, "eventfd-id: %d\n", ctx->id);
326 }
327 #endif
328
329 static const struct file_operations eventfd_fops = {
330 #ifdef CONFIG_PROC_FS
331 .show_fdinfo = eventfd_show_fdinfo,
332 #endif
333 .release = eventfd_release,
334 .poll = eventfd_poll,
335 .read = eventfd_read,
336 .write = eventfd_write,
337 .llseek = noop_llseek,
338 };
339
340 /**
341 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
342 * @fd: [in] Eventfd file descriptor.
343 *
344 * Returns a pointer to the eventfd file structure in case of success, or the
345 * following error pointer:
346 *
347 * -EBADF : Invalid @fd file descriptor.
348 * -EINVAL : The @fd file descriptor is not an eventfd file.
349 */
eventfd_fget(int fd)350 struct file *eventfd_fget(int fd)
351 {
352 struct file *file;
353
354 file = fget(fd);
355 if (!file)
356 return ERR_PTR(-EBADF);
357 if (file->f_op != &eventfd_fops) {
358 fput(file);
359 return ERR_PTR(-EINVAL);
360 }
361
362 return file;
363 }
364 EXPORT_SYMBOL_GPL(eventfd_fget);
365
366 /**
367 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
368 * @fd: [in] Eventfd file descriptor.
369 *
370 * Returns a pointer to the internal eventfd context, otherwise the error
371 * pointers returned by the following functions:
372 *
373 * eventfd_fget
374 */
eventfd_ctx_fdget(int fd)375 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
376 {
377 struct eventfd_ctx *ctx;
378 struct fd f = fdget(fd);
379 if (!f.file)
380 return ERR_PTR(-EBADF);
381 ctx = eventfd_ctx_fileget(f.file);
382 fdput(f);
383 return ctx;
384 }
385 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
386
387 /**
388 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
389 * @file: [in] Eventfd file pointer.
390 *
391 * Returns a pointer to the internal eventfd context, otherwise the error
392 * pointer:
393 *
394 * -EINVAL : The @fd file descriptor is not an eventfd file.
395 */
eventfd_ctx_fileget(struct file * file)396 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
397 {
398 struct eventfd_ctx *ctx;
399
400 if (file->f_op != &eventfd_fops)
401 return ERR_PTR(-EINVAL);
402
403 ctx = file->private_data;
404 kref_get(&ctx->kref);
405 return ctx;
406 }
407 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
408
do_eventfd(unsigned int count,int flags)409 static int do_eventfd(unsigned int count, int flags)
410 {
411 struct eventfd_ctx *ctx;
412 int fd;
413
414 /* Check the EFD_* constants for consistency. */
415 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
416 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
417
418 if (flags & ~EFD_FLAGS_SET)
419 return -EINVAL;
420
421 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
422 if (!ctx)
423 return -ENOMEM;
424
425 kref_init(&ctx->kref);
426 init_waitqueue_head(&ctx->wqh);
427 ctx->count = count;
428 ctx->flags = flags;
429 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
430
431 fd = anon_inode_getfd("[eventfd]", &eventfd_fops, ctx,
432 O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
433 if (fd < 0)
434 eventfd_free_ctx(ctx);
435
436 return fd;
437 }
438
SYSCALL_DEFINE2(eventfd2,unsigned int,count,int,flags)439 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
440 {
441 return do_eventfd(count, flags);
442 }
443
SYSCALL_DEFINE1(eventfd,unsigned int,count)444 SYSCALL_DEFINE1(eventfd, unsigned int, count)
445 {
446 return do_eventfd(count, 0);
447 }
448
449