Commit | Line | Data |
---|---|---|
e1ad7468 DL |
1 | /* |
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> | |
174cd4b1 | 12 | #include <linux/sched/signal.h> |
e1ad7468 | 13 | #include <linux/kernel.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
e1ad7468 DL |
15 | #include <linux/list.h> |
16 | #include <linux/spinlock.h> | |
17 | #include <linux/anon_inodes.h> | |
7747cdb2 | 18 | #include <linux/syscalls.h> |
630d9c47 | 19 | #include <linux/export.h> |
13389010 DL |
20 | #include <linux/kref.h> |
21 | #include <linux/eventfd.h> | |
cbac5542 CG |
22 | #include <linux/proc_fs.h> |
23 | #include <linux/seq_file.h> | |
e1ad7468 DL |
24 | |
25 | struct eventfd_ctx { | |
13389010 | 26 | struct kref kref; |
e1ad7468 DL |
27 | 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 | |
13389010 | 33 | * side eventfd_signal() also, adds to the "count" counter and |
e1ad7468 DL |
34 | * issue a wakeup. |
35 | */ | |
36 | __u64 count; | |
bcd0b235 | 37 | unsigned int flags; |
e1ad7468 DL |
38 | }; |
39 | ||
13389010 DL |
40 | /** |
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 | |
a9a08845 | 48 | * value, and we signal this as overflow condition by returning a EPOLLERR |
13389010 DL |
49 | * to poll(2). |
50 | * | |
20d5a865 | 51 | * Returns the amount by which the counter was incremented. This will be less |
ee62c6b2 | 52 | * than @n if the counter has overflowed. |
e1ad7468 | 53 | */ |
ee62c6b2 | 54 | __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n) |
e1ad7468 | 55 | { |
e1ad7468 DL |
56 | unsigned long flags; |
57 | ||
d48eb233 | 58 | spin_lock_irqsave(&ctx->wqh.lock, flags); |
e1ad7468 | 59 | if (ULLONG_MAX - ctx->count < n) |
ee62c6b2 | 60 | n = ULLONG_MAX - ctx->count; |
e1ad7468 DL |
61 | ctx->count += n; |
62 | if (waitqueue_active(&ctx->wqh)) | |
a9a08845 | 63 | wake_up_locked_poll(&ctx->wqh, EPOLLIN); |
d48eb233 | 64 | spin_unlock_irqrestore(&ctx->wqh.lock, flags); |
e1ad7468 DL |
65 | |
66 | return n; | |
67 | } | |
5718607b | 68 | EXPORT_SYMBOL_GPL(eventfd_signal); |
e1ad7468 | 69 | |
562787a5 DL |
70 | static void eventfd_free_ctx(struct eventfd_ctx *ctx) |
71 | { | |
72 | kfree(ctx); | |
73 | } | |
74 | ||
13389010 DL |
75 | static void eventfd_free(struct kref *kref) |
76 | { | |
77 | struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref); | |
78 | ||
562787a5 | 79 | eventfd_free_ctx(ctx); |
13389010 DL |
80 | } |
81 | ||
13389010 DL |
82 | /** |
83 | * eventfd_ctx_put - Releases a reference to the internal eventfd context. | |
84 | * @ctx: [in] Pointer to eventfd context. | |
85 | * | |
86 | * The eventfd context reference must have been previously acquired either | |
105f2b70 | 87 | * with eventfd_ctx_fdget() or eventfd_ctx_fileget(). |
13389010 DL |
88 | */ |
89 | void eventfd_ctx_put(struct eventfd_ctx *ctx) | |
90 | { | |
91 | kref_put(&ctx->kref, eventfd_free); | |
92 | } | |
93 | EXPORT_SYMBOL_GPL(eventfd_ctx_put); | |
94 | ||
e1ad7468 DL |
95 | static int eventfd_release(struct inode *inode, struct file *file) |
96 | { | |
13389010 DL |
97 | struct eventfd_ctx *ctx = file->private_data; |
98 | ||
a9a08845 | 99 | wake_up_poll(&ctx->wqh, EPOLLHUP); |
13389010 | 100 | eventfd_ctx_put(ctx); |
e1ad7468 DL |
101 | return 0; |
102 | } | |
103 | ||
9e42f195 CH |
104 | static struct wait_queue_head * |
105 | eventfd_get_poll_head(struct file *file, __poll_t events) | |
106 | { | |
107 | struct eventfd_ctx *ctx = file->private_data; | |
108 | ||
109 | return &ctx->wqh; | |
110 | } | |
111 | ||
112 | static __poll_t eventfd_poll_mask(struct file *file, __poll_t eventmask) | |
e1ad7468 DL |
113 | { |
114 | struct eventfd_ctx *ctx = file->private_data; | |
076ccb76 | 115 | __poll_t events = 0; |
e22553e2 | 116 | u64 count; |
e1ad7468 | 117 | |
a484c3dd PB |
118 | /* |
119 | * All writes to ctx->count occur within ctx->wqh.lock. This read | |
120 | * can be done outside ctx->wqh.lock because we know that poll_wait | |
121 | * takes that lock (through add_wait_queue) if our caller will sleep. | |
122 | * | |
123 | * The read _can_ therefore seep into add_wait_queue's critical | |
124 | * section, but cannot move above it! add_wait_queue's spin_lock acts | |
125 | * as an acquire barrier and ensures that the read be ordered properly | |
126 | * against the writes. The following CAN happen and is safe: | |
127 | * | |
128 | * poll write | |
129 | * ----------------- ------------ | |
130 | * lock ctx->wqh.lock (in poll_wait) | |
131 | * count = ctx->count | |
132 | * __add_wait_queue | |
133 | * unlock ctx->wqh.lock | |
134 | * lock ctx->qwh.lock | |
135 | * ctx->count += n | |
136 | * if (waitqueue_active) | |
137 | * wake_up_locked_poll | |
138 | * unlock ctx->qwh.lock | |
139 | * eventfd_poll returns 0 | |
140 | * | |
141 | * but the following, which would miss a wakeup, cannot happen: | |
142 | * | |
143 | * poll write | |
144 | * ----------------- ------------ | |
145 | * count = ctx->count (INVALID!) | |
146 | * lock ctx->qwh.lock | |
147 | * ctx->count += n | |
148 | * **waitqueue_active is false** | |
149 | * **no wake_up_locked_poll!** | |
150 | * unlock ctx->qwh.lock | |
151 | * lock ctx->wqh.lock (in poll_wait) | |
152 | * __add_wait_queue | |
153 | * unlock ctx->wqh.lock | |
154 | * eventfd_poll returns 0 | |
155 | */ | |
156 | count = READ_ONCE(ctx->count); | |
e1ad7468 | 157 | |
e22553e2 | 158 | if (count > 0) |
a9a08845 | 159 | events |= EPOLLIN; |
e22553e2 | 160 | if (count == ULLONG_MAX) |
a9a08845 | 161 | events |= EPOLLERR; |
e22553e2 | 162 | if (ULLONG_MAX - 1 > count) |
a9a08845 | 163 | events |= EPOLLOUT; |
e1ad7468 DL |
164 | |
165 | return events; | |
166 | } | |
167 | ||
cb289d62 DL |
168 | static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt) |
169 | { | |
170 | *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count; | |
171 | ctx->count -= *cnt; | |
172 | } | |
173 | ||
174 | /** | |
175 | * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue. | |
176 | * @ctx: [in] Pointer to eventfd context. | |
177 | * @wait: [in] Wait queue to be removed. | |
36182185 | 178 | * @cnt: [out] Pointer to the 64-bit counter value. |
cb289d62 | 179 | * |
36182185 | 180 | * Returns %0 if successful, or the following error codes: |
cb289d62 DL |
181 | * |
182 | * -EAGAIN : The operation would have blocked. | |
183 | * | |
184 | * This is used to atomically remove a wait queue entry from the eventfd wait | |
185 | * queue head, and read/reset the counter value. | |
186 | */ | |
ac6424b9 | 187 | int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait, |
cb289d62 DL |
188 | __u64 *cnt) |
189 | { | |
190 | unsigned long flags; | |
191 | ||
192 | spin_lock_irqsave(&ctx->wqh.lock, flags); | |
193 | eventfd_ctx_do_read(ctx, cnt); | |
194 | __remove_wait_queue(&ctx->wqh, wait); | |
195 | if (*cnt != 0 && waitqueue_active(&ctx->wqh)) | |
a9a08845 | 196 | wake_up_locked_poll(&ctx->wqh, EPOLLOUT); |
cb289d62 DL |
197 | spin_unlock_irqrestore(&ctx->wqh.lock, flags); |
198 | ||
199 | return *cnt != 0 ? 0 : -EAGAIN; | |
200 | } | |
201 | EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue); | |
202 | ||
b6364572 EB |
203 | static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count, |
204 | loff_t *ppos) | |
e1ad7468 | 205 | { |
b6364572 | 206 | struct eventfd_ctx *ctx = file->private_data; |
e1ad7468 | 207 | ssize_t res; |
b6364572 | 208 | __u64 ucnt = 0; |
e1ad7468 DL |
209 | DECLARE_WAITQUEUE(wait, current); |
210 | ||
b6364572 EB |
211 | if (count < sizeof(ucnt)) |
212 | return -EINVAL; | |
213 | ||
d48eb233 | 214 | spin_lock_irq(&ctx->wqh.lock); |
e1ad7468 | 215 | res = -EAGAIN; |
bcd0b235 | 216 | if (ctx->count > 0) |
b6364572 EB |
217 | res = sizeof(ucnt); |
218 | else if (!(file->f_flags & O_NONBLOCK)) { | |
e1ad7468 | 219 | __add_wait_queue(&ctx->wqh, &wait); |
cb289d62 | 220 | for (;;) { |
e1ad7468 DL |
221 | set_current_state(TASK_INTERRUPTIBLE); |
222 | if (ctx->count > 0) { | |
b6364572 | 223 | res = sizeof(ucnt); |
e1ad7468 DL |
224 | break; |
225 | } | |
226 | if (signal_pending(current)) { | |
227 | res = -ERESTARTSYS; | |
228 | break; | |
229 | } | |
d48eb233 | 230 | spin_unlock_irq(&ctx->wqh.lock); |
e1ad7468 | 231 | schedule(); |
d48eb233 | 232 | spin_lock_irq(&ctx->wqh.lock); |
e1ad7468 DL |
233 | } |
234 | __remove_wait_queue(&ctx->wqh, &wait); | |
235 | __set_current_state(TASK_RUNNING); | |
236 | } | |
b6364572 EB |
237 | if (likely(res > 0)) { |
238 | eventfd_ctx_do_read(ctx, &ucnt); | |
e1ad7468 | 239 | if (waitqueue_active(&ctx->wqh)) |
a9a08845 | 240 | wake_up_locked_poll(&ctx->wqh, EPOLLOUT); |
e1ad7468 | 241 | } |
d48eb233 | 242 | spin_unlock_irq(&ctx->wqh.lock); |
e1ad7468 | 243 | |
b6364572 EB |
244 | if (res > 0 && put_user(ucnt, (__u64 __user *)buf)) |
245 | return -EFAULT; | |
cb289d62 | 246 | |
b6364572 | 247 | return res; |
cb289d62 | 248 | } |
e1ad7468 DL |
249 | |
250 | static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count, | |
251 | loff_t *ppos) | |
252 | { | |
253 | struct eventfd_ctx *ctx = file->private_data; | |
254 | ssize_t res; | |
255 | __u64 ucnt; | |
256 | DECLARE_WAITQUEUE(wait, current); | |
257 | ||
258 | if (count < sizeof(ucnt)) | |
259 | return -EINVAL; | |
260 | if (copy_from_user(&ucnt, buf, sizeof(ucnt))) | |
261 | return -EFAULT; | |
262 | if (ucnt == ULLONG_MAX) | |
263 | return -EINVAL; | |
d48eb233 | 264 | spin_lock_irq(&ctx->wqh.lock); |
e1ad7468 DL |
265 | res = -EAGAIN; |
266 | if (ULLONG_MAX - ctx->count > ucnt) | |
267 | res = sizeof(ucnt); | |
268 | else if (!(file->f_flags & O_NONBLOCK)) { | |
269 | __add_wait_queue(&ctx->wqh, &wait); | |
270 | for (res = 0;;) { | |
271 | set_current_state(TASK_INTERRUPTIBLE); | |
272 | if (ULLONG_MAX - ctx->count > ucnt) { | |
273 | res = sizeof(ucnt); | |
274 | break; | |
275 | } | |
276 | if (signal_pending(current)) { | |
277 | res = -ERESTARTSYS; | |
278 | break; | |
279 | } | |
d48eb233 | 280 | spin_unlock_irq(&ctx->wqh.lock); |
e1ad7468 | 281 | schedule(); |
d48eb233 | 282 | spin_lock_irq(&ctx->wqh.lock); |
e1ad7468 DL |
283 | } |
284 | __remove_wait_queue(&ctx->wqh, &wait); | |
285 | __set_current_state(TASK_RUNNING); | |
286 | } | |
bcd0b235 | 287 | if (likely(res > 0)) { |
e1ad7468 DL |
288 | ctx->count += ucnt; |
289 | if (waitqueue_active(&ctx->wqh)) | |
a9a08845 | 290 | wake_up_locked_poll(&ctx->wqh, EPOLLIN); |
e1ad7468 | 291 | } |
d48eb233 | 292 | spin_unlock_irq(&ctx->wqh.lock); |
e1ad7468 DL |
293 | |
294 | return res; | |
295 | } | |
296 | ||
cbac5542 | 297 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 298 | static void eventfd_show_fdinfo(struct seq_file *m, struct file *f) |
cbac5542 CG |
299 | { |
300 | struct eventfd_ctx *ctx = f->private_data; | |
cbac5542 CG |
301 | |
302 | spin_lock_irq(&ctx->wqh.lock); | |
a3816ab0 JP |
303 | seq_printf(m, "eventfd-count: %16llx\n", |
304 | (unsigned long long)ctx->count); | |
cbac5542 | 305 | spin_unlock_irq(&ctx->wqh.lock); |
cbac5542 CG |
306 | } |
307 | #endif | |
308 | ||
e1ad7468 | 309 | static const struct file_operations eventfd_fops = { |
cbac5542 CG |
310 | #ifdef CONFIG_PROC_FS |
311 | .show_fdinfo = eventfd_show_fdinfo, | |
312 | #endif | |
e1ad7468 | 313 | .release = eventfd_release, |
9e42f195 CH |
314 | .get_poll_head = eventfd_get_poll_head, |
315 | .poll_mask = eventfd_poll_mask, | |
e1ad7468 DL |
316 | .read = eventfd_read, |
317 | .write = eventfd_write, | |
6038f373 | 318 | .llseek = noop_llseek, |
e1ad7468 DL |
319 | }; |
320 | ||
13389010 DL |
321 | /** |
322 | * eventfd_fget - Acquire a reference of an eventfd file descriptor. | |
323 | * @fd: [in] Eventfd file descriptor. | |
324 | * | |
325 | * Returns a pointer to the eventfd file structure in case of success, or the | |
326 | * following error pointer: | |
327 | * | |
328 | * -EBADF : Invalid @fd file descriptor. | |
329 | * -EINVAL : The @fd file descriptor is not an eventfd file. | |
330 | */ | |
e1ad7468 DL |
331 | struct file *eventfd_fget(int fd) |
332 | { | |
333 | struct file *file; | |
334 | ||
335 | file = fget(fd); | |
336 | if (!file) | |
337 | return ERR_PTR(-EBADF); | |
338 | if (file->f_op != &eventfd_fops) { | |
339 | fput(file); | |
340 | return ERR_PTR(-EINVAL); | |
341 | } | |
342 | ||
343 | return file; | |
344 | } | |
5718607b | 345 | EXPORT_SYMBOL_GPL(eventfd_fget); |
e1ad7468 | 346 | |
13389010 DL |
347 | /** |
348 | * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context. | |
349 | * @fd: [in] Eventfd file descriptor. | |
350 | * | |
351 | * Returns a pointer to the internal eventfd context, otherwise the error | |
352 | * pointers returned by the following functions: | |
353 | * | |
354 | * eventfd_fget | |
355 | */ | |
356 | struct eventfd_ctx *eventfd_ctx_fdget(int fd) | |
357 | { | |
13389010 | 358 | struct eventfd_ctx *ctx; |
36a74117 AV |
359 | struct fd f = fdget(fd); |
360 | if (!f.file) | |
361 | return ERR_PTR(-EBADF); | |
362 | ctx = eventfd_ctx_fileget(f.file); | |
363 | fdput(f); | |
13389010 DL |
364 | return ctx; |
365 | } | |
366 | EXPORT_SYMBOL_GPL(eventfd_ctx_fdget); | |
367 | ||
368 | /** | |
369 | * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context. | |
370 | * @file: [in] Eventfd file pointer. | |
371 | * | |
372 | * Returns a pointer to the internal eventfd context, otherwise the error | |
373 | * pointer: | |
374 | * | |
375 | * -EINVAL : The @fd file descriptor is not an eventfd file. | |
376 | */ | |
377 | struct eventfd_ctx *eventfd_ctx_fileget(struct file *file) | |
378 | { | |
105f2b70 EB |
379 | struct eventfd_ctx *ctx; |
380 | ||
13389010 DL |
381 | if (file->f_op != &eventfd_fops) |
382 | return ERR_PTR(-EINVAL); | |
383 | ||
105f2b70 EB |
384 | ctx = file->private_data; |
385 | kref_get(&ctx->kref); | |
386 | return ctx; | |
13389010 DL |
387 | } |
388 | EXPORT_SYMBOL_GPL(eventfd_ctx_fileget); | |
389 | ||
2fc96f83 | 390 | static int do_eventfd(unsigned int count, int flags) |
e1ad7468 | 391 | { |
e1ad7468 | 392 | struct eventfd_ctx *ctx; |
7d815165 | 393 | int fd; |
e1ad7468 | 394 | |
e38b36f3 UD |
395 | /* Check the EFD_* constants for consistency. */ |
396 | BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC); | |
397 | BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK); | |
398 | ||
bcd0b235 | 399 | if (flags & ~EFD_FLAGS_SET) |
7d815165 | 400 | return -EINVAL; |
b087498e | 401 | |
e1ad7468 DL |
402 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
403 | if (!ctx) | |
7d815165 | 404 | return -ENOMEM; |
e1ad7468 | 405 | |
13389010 | 406 | kref_init(&ctx->kref); |
e1ad7468 | 407 | init_waitqueue_head(&ctx->wqh); |
e1ad7468 | 408 | ctx->count = count; |
bcd0b235 | 409 | ctx->flags = flags; |
e1ad7468 | 410 | |
7d815165 EB |
411 | fd = anon_inode_getfd("[eventfd]", &eventfd_fops, ctx, |
412 | O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS)); | |
413 | if (fd < 0) | |
562787a5 DL |
414 | eventfd_free_ctx(ctx); |
415 | ||
2030a42c | 416 | return fd; |
e1ad7468 DL |
417 | } |
418 | ||
2fc96f83 DB |
419 | SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags) |
420 | { | |
421 | return do_eventfd(count, flags); | |
422 | } | |
423 | ||
d4e82042 | 424 | SYSCALL_DEFINE1(eventfd, unsigned int, count) |
b087498e | 425 | { |
2fc96f83 | 426 | return do_eventfd(count, 0); |
b087498e | 427 | } |
bcd0b235 | 428 |