1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * fs/eventfd.c 3 * fs/eventfd.c
4 * 4 *
5 * Copyright (C) 2007 Davide Libenzi <davide 5 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
6 * 6 *
7 */ 7 */
8 8
9 #include <linux/file.h> 9 #include <linux/file.h>
10 #include <linux/poll.h> 10 #include <linux/poll.h>
11 #include <linux/init.h> 11 #include <linux/init.h>
12 #include <linux/fs.h> 12 #include <linux/fs.h>
13 #include <linux/sched/signal.h> 13 #include <linux/sched/signal.h>
14 #include <linux/kernel.h> 14 #include <linux/kernel.h>
15 #include <linux/slab.h> 15 #include <linux/slab.h>
16 #include <linux/list.h> 16 #include <linux/list.h>
17 #include <linux/spinlock.h> 17 #include <linux/spinlock.h>
18 #include <linux/anon_inodes.h> 18 #include <linux/anon_inodes.h>
19 #include <linux/syscalls.h> 19 #include <linux/syscalls.h>
20 #include <linux/export.h> 20 #include <linux/export.h>
21 #include <linux/kref.h> 21 #include <linux/kref.h>
22 #include <linux/eventfd.h> 22 #include <linux/eventfd.h>
23 #include <linux/proc_fs.h> 23 #include <linux/proc_fs.h>
24 #include <linux/seq_file.h> 24 #include <linux/seq_file.h>
25 #include <linux/idr.h> 25 #include <linux/idr.h>
26 #include <linux/uio.h> 26 #include <linux/uio.h>
27 27
28 static DEFINE_IDA(eventfd_ida); 28 static DEFINE_IDA(eventfd_ida);
29 29
30 struct eventfd_ctx { 30 struct eventfd_ctx {
31 struct kref kref; 31 struct kref kref;
32 wait_queue_head_t wqh; 32 wait_queue_head_t wqh;
33 /* 33 /*
34 * Every time that a write(2) is perfo 34 * Every time that a write(2) is performed on an eventfd, the
35 * value of the __u64 being written is 35 * value of the __u64 being written is added to "count" and a
36 * wakeup is performed on "wqh". If EF !! 36 * wakeup is performed on "wqh". A read(2) will return the "count"
37 * specified, a read(2) will return th !! 37 * value to userspace, and will reset "count" to zero. The kernel
38 * and will reset "count" to zero. The !! 38 * side eventfd_signal() also, adds to the "count" counter and
39 * also, adds to the "count" counter a !! 39 * issue a wakeup.
40 */ 40 */
41 __u64 count; 41 __u64 count;
42 unsigned int flags; 42 unsigned int flags;
43 int id; 43 int id;
44 }; 44 };
45 45
46 /** 46 /**
47 * eventfd_signal_mask - Increment the event c !! 47 * eventfd_signal - Adds @n to the eventfd counter.
48 * @ctx: [in] Pointer to the eventfd context. 48 * @ctx: [in] Pointer to the eventfd context.
49 * @mask: [in] poll mask !! 49 * @n: [in] Value of the counter to be added to the eventfd internal counter.
>> 50 * The value cannot be negative.
50 * 51 *
51 * This function is supposed to be called by t 52 * This function is supposed to be called by the kernel in paths that do not
52 * allow sleeping. In this function we allow t 53 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
53 * value, and we signal this as overflow condi 54 * value, and we signal this as overflow condition by returning a EPOLLERR
54 * to poll(2). 55 * to poll(2).
>> 56 *
>> 57 * Returns the amount by which the counter was incremented. This will be less
>> 58 * than @n if the counter has overflowed.
55 */ 59 */
56 void eventfd_signal_mask(struct eventfd_ctx *c !! 60 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
57 { 61 {
58 unsigned long flags; 62 unsigned long flags;
59 63
60 /* 64 /*
61 * Deadlock or stack overflow issues c 65 * Deadlock or stack overflow issues can happen if we recurse here
62 * through waitqueue wakeup handlers. 66 * through waitqueue wakeup handlers. If the caller users potentially
63 * nested waitqueues with custom wakeu 67 * nested waitqueues with custom wakeup handlers, then it should
64 * check eventfd_signal_allowed() befo 68 * check eventfd_signal_allowed() before calling this function. If
65 * it returns false, the eventfd_signa 69 * it returns false, the eventfd_signal() call should be deferred to a
66 * safe context. 70 * safe context.
67 */ 71 */
68 if (WARN_ON_ONCE(current->in_eventfd)) !! 72 if (WARN_ON_ONCE(current->in_eventfd_signal))
69 return; !! 73 return 0;
70 74
71 spin_lock_irqsave(&ctx->wqh.lock, flag 75 spin_lock_irqsave(&ctx->wqh.lock, flags);
72 current->in_eventfd = 1; !! 76 current->in_eventfd_signal = 1;
73 if (ctx->count < ULLONG_MAX) !! 77 if (ULLONG_MAX - ctx->count < n)
74 ctx->count++; !! 78 n = ULLONG_MAX - ctx->count;
>> 79 ctx->count += n;
75 if (waitqueue_active(&ctx->wqh)) 80 if (waitqueue_active(&ctx->wqh))
76 wake_up_locked_poll(&ctx->wqh, !! 81 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
77 current->in_eventfd = 0; !! 82 current->in_eventfd_signal = 0;
78 spin_unlock_irqrestore(&ctx->wqh.lock, 83 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
>> 84
>> 85 return n;
79 } 86 }
80 EXPORT_SYMBOL_GPL(eventfd_signal_mask); !! 87 EXPORT_SYMBOL_GPL(eventfd_signal);
81 88
82 static void eventfd_free_ctx(struct eventfd_ct 89 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
83 { 90 {
84 if (ctx->id >= 0) 91 if (ctx->id >= 0)
85 ida_free(&eventfd_ida, ctx->id !! 92 ida_simple_remove(&eventfd_ida, ctx->id);
86 kfree(ctx); 93 kfree(ctx);
87 } 94 }
88 95
89 static void eventfd_free(struct kref *kref) 96 static void eventfd_free(struct kref *kref)
90 { 97 {
91 struct eventfd_ctx *ctx = container_of 98 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
92 99
93 eventfd_free_ctx(ctx); 100 eventfd_free_ctx(ctx);
94 } 101 }
95 102
96 /** 103 /**
97 * eventfd_ctx_put - Releases a reference to t 104 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
98 * @ctx: [in] Pointer to eventfd context. 105 * @ctx: [in] Pointer to eventfd context.
99 * 106 *
100 * The eventfd context reference must have bee 107 * The eventfd context reference must have been previously acquired either
101 * with eventfd_ctx_fdget() or eventfd_ctx_fil 108 * with eventfd_ctx_fdget() or eventfd_ctx_fileget().
102 */ 109 */
103 void eventfd_ctx_put(struct eventfd_ctx *ctx) 110 void eventfd_ctx_put(struct eventfd_ctx *ctx)
104 { 111 {
105 kref_put(&ctx->kref, eventfd_free); 112 kref_put(&ctx->kref, eventfd_free);
106 } 113 }
107 EXPORT_SYMBOL_GPL(eventfd_ctx_put); 114 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
108 115
109 static int eventfd_release(struct inode *inode 116 static int eventfd_release(struct inode *inode, struct file *file)
110 { 117 {
111 struct eventfd_ctx *ctx = file->privat 118 struct eventfd_ctx *ctx = file->private_data;
112 119
113 wake_up_poll(&ctx->wqh, EPOLLHUP); 120 wake_up_poll(&ctx->wqh, EPOLLHUP);
114 eventfd_ctx_put(ctx); 121 eventfd_ctx_put(ctx);
115 return 0; 122 return 0;
116 } 123 }
117 124
118 static __poll_t eventfd_poll(struct file *file 125 static __poll_t eventfd_poll(struct file *file, poll_table *wait)
119 { 126 {
120 struct eventfd_ctx *ctx = file->privat 127 struct eventfd_ctx *ctx = file->private_data;
121 __poll_t events = 0; 128 __poll_t events = 0;
122 u64 count; 129 u64 count;
123 130
124 poll_wait(file, &ctx->wqh, wait); 131 poll_wait(file, &ctx->wqh, wait);
125 132
126 /* 133 /*
127 * All writes to ctx->count occur with 134 * All writes to ctx->count occur within ctx->wqh.lock. This read
128 * can be done outside ctx->wqh.lock b 135 * can be done outside ctx->wqh.lock because we know that poll_wait
129 * takes that lock (through add_wait_q 136 * takes that lock (through add_wait_queue) if our caller will sleep.
130 * 137 *
131 * The read _can_ therefore seep into 138 * The read _can_ therefore seep into add_wait_queue's critical
132 * section, but cannot move above it! 139 * section, but cannot move above it! add_wait_queue's spin_lock acts
133 * as an acquire barrier and ensures t 140 * as an acquire barrier and ensures that the read be ordered properly
134 * against the writes. The following 141 * against the writes. The following CAN happen and is safe:
135 * 142 *
136 * poll 143 * poll write
137 * ----------------- 144 * ----------------- ------------
138 * lock ctx->wqh.lock (in poll_wai 145 * lock ctx->wqh.lock (in poll_wait)
139 * count = ctx->count 146 * count = ctx->count
140 * __add_wait_queue 147 * __add_wait_queue
141 * unlock ctx->wqh.lock 148 * unlock ctx->wqh.lock
142 * 149 * lock ctx->qwh.lock
143 * 150 * ctx->count += n
144 * 151 * if (waitqueue_active)
145 * 152 * wake_up_locked_poll
146 * 153 * unlock ctx->qwh.lock
147 * eventfd_poll returns 0 154 * eventfd_poll returns 0
148 * 155 *
149 * but the following, which would miss 156 * but the following, which would miss a wakeup, cannot happen:
150 * 157 *
151 * poll 158 * poll write
152 * ----------------- 159 * ----------------- ------------
153 * count = ctx->count (INVALID!) 160 * count = ctx->count (INVALID!)
154 * 161 * lock ctx->qwh.lock
155 * 162 * ctx->count += n
156 * 163 * **waitqueue_active is false**
157 * 164 * **no wake_up_locked_poll!**
158 * 165 * unlock ctx->qwh.lock
159 * lock ctx->wqh.lock (in poll_wai 166 * lock ctx->wqh.lock (in poll_wait)
160 * __add_wait_queue 167 * __add_wait_queue
161 * unlock ctx->wqh.lock 168 * unlock ctx->wqh.lock
162 * eventfd_poll returns 0 169 * eventfd_poll returns 0
163 */ 170 */
164 count = READ_ONCE(ctx->count); 171 count = READ_ONCE(ctx->count);
165 172
166 if (count > 0) 173 if (count > 0)
167 events |= EPOLLIN; 174 events |= EPOLLIN;
168 if (count == ULLONG_MAX) 175 if (count == ULLONG_MAX)
169 events |= EPOLLERR; 176 events |= EPOLLERR;
170 if (ULLONG_MAX - 1 > count) 177 if (ULLONG_MAX - 1 > count)
171 events |= EPOLLOUT; 178 events |= EPOLLOUT;
172 179
173 return events; 180 return events;
174 } 181 }
175 182
176 void eventfd_ctx_do_read(struct eventfd_ctx *c 183 void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
177 { 184 {
178 lockdep_assert_held(&ctx->wqh.lock); 185 lockdep_assert_held(&ctx->wqh.lock);
179 186
180 *cnt = ((ctx->flags & EFD_SEMAPHORE) & !! 187 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
181 ctx->count -= *cnt; 188 ctx->count -= *cnt;
182 } 189 }
183 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read); 190 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read);
184 191
185 /** 192 /**
186 * eventfd_ctx_remove_wait_queue - Read the cu 193 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
187 * @ctx: [in] Pointer to eventfd context. 194 * @ctx: [in] Pointer to eventfd context.
188 * @wait: [in] Wait queue to be removed. 195 * @wait: [in] Wait queue to be removed.
189 * @cnt: [out] Pointer to the 64-bit counter v 196 * @cnt: [out] Pointer to the 64-bit counter value.
190 * 197 *
191 * Returns %0 if successful, or the following 198 * Returns %0 if successful, or the following error codes:
192 * 199 *
193 * -EAGAIN : The operation would have blo 200 * -EAGAIN : The operation would have blocked.
194 * 201 *
195 * This is used to atomically remove a wait qu 202 * This is used to atomically remove a wait queue entry from the eventfd wait
196 * queue head, and read/reset the counter valu 203 * queue head, and read/reset the counter value.
197 */ 204 */
198 int eventfd_ctx_remove_wait_queue(struct event 205 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
199 __u64 *cnt) 206 __u64 *cnt)
200 { 207 {
201 unsigned long flags; 208 unsigned long flags;
202 209
203 spin_lock_irqsave(&ctx->wqh.lock, flag 210 spin_lock_irqsave(&ctx->wqh.lock, flags);
204 eventfd_ctx_do_read(ctx, cnt); 211 eventfd_ctx_do_read(ctx, cnt);
205 __remove_wait_queue(&ctx->wqh, wait); 212 __remove_wait_queue(&ctx->wqh, wait);
206 if (*cnt != 0 && waitqueue_active(&ctx 213 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
207 wake_up_locked_poll(&ctx->wqh, 214 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
208 spin_unlock_irqrestore(&ctx->wqh.lock, 215 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
209 216
210 return *cnt != 0 ? 0 : -EAGAIN; 217 return *cnt != 0 ? 0 : -EAGAIN;
211 } 218 }
212 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queu 219 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
213 220
214 static ssize_t eventfd_read(struct kiocb *iocb 221 static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to)
215 { 222 {
216 struct file *file = iocb->ki_filp; 223 struct file *file = iocb->ki_filp;
217 struct eventfd_ctx *ctx = file->privat 224 struct eventfd_ctx *ctx = file->private_data;
218 __u64 ucnt = 0; 225 __u64 ucnt = 0;
>> 226 DECLARE_WAITQUEUE(wait, current);
219 227
220 if (iov_iter_count(to) < sizeof(ucnt)) 228 if (iov_iter_count(to) < sizeof(ucnt))
221 return -EINVAL; 229 return -EINVAL;
222 spin_lock_irq(&ctx->wqh.lock); 230 spin_lock_irq(&ctx->wqh.lock);
223 if (!ctx->count) { 231 if (!ctx->count) {
224 if ((file->f_flags & O_NONBLOC 232 if ((file->f_flags & O_NONBLOCK) ||
225 (iocb->ki_flags & IOCB_NOW 233 (iocb->ki_flags & IOCB_NOWAIT)) {
226 spin_unlock_irq(&ctx-> 234 spin_unlock_irq(&ctx->wqh.lock);
227 return -EAGAIN; 235 return -EAGAIN;
228 } 236 }
229 !! 237 __add_wait_queue(&ctx->wqh, &wait);
230 if (wait_event_interruptible_l !! 238 for (;;) {
>> 239 set_current_state(TASK_INTERRUPTIBLE);
>> 240 if (ctx->count)
>> 241 break;
>> 242 if (signal_pending(current)) {
>> 243 __remove_wait_queue(&ctx->wqh, &wait);
>> 244 __set_current_state(TASK_RUNNING);
>> 245 spin_unlock_irq(&ctx->wqh.lock);
>> 246 return -ERESTARTSYS;
>> 247 }
231 spin_unlock_irq(&ctx-> 248 spin_unlock_irq(&ctx->wqh.lock);
232 return -ERESTARTSYS; !! 249 schedule();
>> 250 spin_lock_irq(&ctx->wqh.lock);
233 } 251 }
>> 252 __remove_wait_queue(&ctx->wqh, &wait);
>> 253 __set_current_state(TASK_RUNNING);
234 } 254 }
235 eventfd_ctx_do_read(ctx, &ucnt); 255 eventfd_ctx_do_read(ctx, &ucnt);
236 current->in_eventfd = 1; <<
237 if (waitqueue_active(&ctx->wqh)) 256 if (waitqueue_active(&ctx->wqh))
238 wake_up_locked_poll(&ctx->wqh, 257 wake_up_locked_poll(&ctx->wqh, EPOLLOUT);
239 current->in_eventfd = 0; <<
240 spin_unlock_irq(&ctx->wqh.lock); 258 spin_unlock_irq(&ctx->wqh.lock);
241 if (unlikely(copy_to_iter(&ucnt, sizeo 259 if (unlikely(copy_to_iter(&ucnt, sizeof(ucnt), to) != sizeof(ucnt)))
242 return -EFAULT; 260 return -EFAULT;
243 261
244 return sizeof(ucnt); 262 return sizeof(ucnt);
245 } 263 }
246 264
247 static ssize_t eventfd_write(struct file *file 265 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
248 loff_t *ppos) 266 loff_t *ppos)
249 { 267 {
250 struct eventfd_ctx *ctx = file->privat 268 struct eventfd_ctx *ctx = file->private_data;
251 ssize_t res; 269 ssize_t res;
252 __u64 ucnt; 270 __u64 ucnt;
>> 271 DECLARE_WAITQUEUE(wait, current);
253 272
254 if (count != sizeof(ucnt)) !! 273 if (count < sizeof(ucnt))
255 return -EINVAL; 274 return -EINVAL;
256 if (copy_from_user(&ucnt, buf, sizeof( 275 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
257 return -EFAULT; 276 return -EFAULT;
258 if (ucnt == ULLONG_MAX) 277 if (ucnt == ULLONG_MAX)
259 return -EINVAL; 278 return -EINVAL;
260 spin_lock_irq(&ctx->wqh.lock); 279 spin_lock_irq(&ctx->wqh.lock);
261 res = -EAGAIN; 280 res = -EAGAIN;
262 if (ULLONG_MAX - ctx->count > ucnt) 281 if (ULLONG_MAX - ctx->count > ucnt)
263 res = sizeof(ucnt); 282 res = sizeof(ucnt);
264 else if (!(file->f_flags & O_NONBLOCK) 283 else if (!(file->f_flags & O_NONBLOCK)) {
265 res = wait_event_interruptible !! 284 __add_wait_queue(&ctx->wqh, &wait);
266 ULLONG_MAX - c !! 285 for (res = 0;;) {
267 if (!res) !! 286 set_current_state(TASK_INTERRUPTIBLE);
268 res = sizeof(ucnt); !! 287 if (ULLONG_MAX - ctx->count > ucnt) {
>> 288 res = sizeof(ucnt);
>> 289 break;
>> 290 }
>> 291 if (signal_pending(current)) {
>> 292 res = -ERESTARTSYS;
>> 293 break;
>> 294 }
>> 295 spin_unlock_irq(&ctx->wqh.lock);
>> 296 schedule();
>> 297 spin_lock_irq(&ctx->wqh.lock);
>> 298 }
>> 299 __remove_wait_queue(&ctx->wqh, &wait);
>> 300 __set_current_state(TASK_RUNNING);
269 } 301 }
270 if (likely(res > 0)) { 302 if (likely(res > 0)) {
271 ctx->count += ucnt; 303 ctx->count += ucnt;
272 current->in_eventfd = 1; <<
273 if (waitqueue_active(&ctx->wqh 304 if (waitqueue_active(&ctx->wqh))
274 wake_up_locked_poll(&c 305 wake_up_locked_poll(&ctx->wqh, EPOLLIN);
275 current->in_eventfd = 0; <<
276 } 306 }
277 spin_unlock_irq(&ctx->wqh.lock); 307 spin_unlock_irq(&ctx->wqh.lock);
278 308
279 return res; 309 return res;
280 } 310 }
281 311
282 #ifdef CONFIG_PROC_FS 312 #ifdef CONFIG_PROC_FS
283 static void eventfd_show_fdinfo(struct seq_fil 313 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
284 { 314 {
285 struct eventfd_ctx *ctx = f->private_d 315 struct eventfd_ctx *ctx = f->private_data;
286 __u64 cnt; <<
287 316
288 spin_lock_irq(&ctx->wqh.lock); 317 spin_lock_irq(&ctx->wqh.lock);
289 cnt = ctx->count; !! 318 seq_printf(m, "eventfd-count: %16llx\n",
>> 319 (unsigned long long)ctx->count);
290 spin_unlock_irq(&ctx->wqh.lock); 320 spin_unlock_irq(&ctx->wqh.lock);
291 !! 321 seq_printf(m, "eventfd-id: %d\n", ctx->id);
292 seq_printf(m, <<
293 "eventfd-count: %16llx\n" <<
294 "eventfd-id: %d\n" <<
295 "eventfd-semaphore: %d\n", <<
296 cnt, <<
297 ctx->id, <<
298 !!(ctx->flags & EFD_SEMAPHO <<
299 } 322 }
300 #endif 323 #endif
301 324
302 static const struct file_operations eventfd_fo 325 static const struct file_operations eventfd_fops = {
303 #ifdef CONFIG_PROC_FS 326 #ifdef CONFIG_PROC_FS
304 .show_fdinfo = eventfd_show_fdinfo, 327 .show_fdinfo = eventfd_show_fdinfo,
305 #endif 328 #endif
306 .release = eventfd_release, 329 .release = eventfd_release,
307 .poll = eventfd_poll, 330 .poll = eventfd_poll,
308 .read_iter = eventfd_read, 331 .read_iter = eventfd_read,
309 .write = eventfd_write, 332 .write = eventfd_write,
310 .llseek = noop_llseek, 333 .llseek = noop_llseek,
311 }; 334 };
312 335
313 /** 336 /**
314 * eventfd_fget - Acquire a reference of an ev 337 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
315 * @fd: [in] Eventfd file descriptor. 338 * @fd: [in] Eventfd file descriptor.
316 * 339 *
317 * Returns a pointer to the eventfd file struc 340 * Returns a pointer to the eventfd file structure in case of success, or the
318 * following error pointer: 341 * following error pointer:
319 * 342 *
320 * -EBADF : Invalid @fd file descriptor. 343 * -EBADF : Invalid @fd file descriptor.
321 * -EINVAL : The @fd file descriptor is not 344 * -EINVAL : The @fd file descriptor is not an eventfd file.
322 */ 345 */
323 struct file *eventfd_fget(int fd) 346 struct file *eventfd_fget(int fd)
324 { 347 {
325 struct file *file; 348 struct file *file;
326 349
327 file = fget(fd); 350 file = fget(fd);
328 if (!file) 351 if (!file)
329 return ERR_PTR(-EBADF); 352 return ERR_PTR(-EBADF);
330 if (file->f_op != &eventfd_fops) { 353 if (file->f_op != &eventfd_fops) {
331 fput(file); 354 fput(file);
332 return ERR_PTR(-EINVAL); 355 return ERR_PTR(-EINVAL);
333 } 356 }
334 357
335 return file; 358 return file;
336 } 359 }
337 EXPORT_SYMBOL_GPL(eventfd_fget); 360 EXPORT_SYMBOL_GPL(eventfd_fget);
338 361
339 /** 362 /**
340 * eventfd_ctx_fdget - Acquires a reference to 363 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
341 * @fd: [in] Eventfd file descriptor. 364 * @fd: [in] Eventfd file descriptor.
342 * 365 *
343 * Returns a pointer to the internal eventfd c 366 * Returns a pointer to the internal eventfd context, otherwise the error
344 * pointers returned by the following function 367 * pointers returned by the following functions:
345 * 368 *
346 * eventfd_fget 369 * eventfd_fget
347 */ 370 */
348 struct eventfd_ctx *eventfd_ctx_fdget(int fd) 371 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
349 { 372 {
350 struct eventfd_ctx *ctx; 373 struct eventfd_ctx *ctx;
351 struct fd f = fdget(fd); 374 struct fd f = fdget(fd);
352 if (!f.file) 375 if (!f.file)
353 return ERR_PTR(-EBADF); 376 return ERR_PTR(-EBADF);
354 ctx = eventfd_ctx_fileget(f.file); 377 ctx = eventfd_ctx_fileget(f.file);
355 fdput(f); 378 fdput(f);
356 return ctx; 379 return ctx;
357 } 380 }
358 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget); 381 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
359 382
360 /** 383 /**
361 * eventfd_ctx_fileget - Acquires a reference 384 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
362 * @file: [in] Eventfd file pointer. 385 * @file: [in] Eventfd file pointer.
363 * 386 *
364 * Returns a pointer to the internal eventfd c 387 * Returns a pointer to the internal eventfd context, otherwise the error
365 * pointer: 388 * pointer:
366 * 389 *
367 * -EINVAL : The @fd file descriptor is not 390 * -EINVAL : The @fd file descriptor is not an eventfd file.
368 */ 391 */
369 struct eventfd_ctx *eventfd_ctx_fileget(struct 392 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
370 { 393 {
371 struct eventfd_ctx *ctx; 394 struct eventfd_ctx *ctx;
372 395
373 if (file->f_op != &eventfd_fops) 396 if (file->f_op != &eventfd_fops)
374 return ERR_PTR(-EINVAL); 397 return ERR_PTR(-EINVAL);
375 398
376 ctx = file->private_data; 399 ctx = file->private_data;
377 kref_get(&ctx->kref); 400 kref_get(&ctx->kref);
378 return ctx; 401 return ctx;
379 } 402 }
380 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget); 403 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
381 404
382 static int do_eventfd(unsigned int count, int 405 static int do_eventfd(unsigned int count, int flags)
383 { 406 {
384 struct eventfd_ctx *ctx; 407 struct eventfd_ctx *ctx;
385 struct file *file; 408 struct file *file;
386 int fd; 409 int fd;
387 410
388 /* Check the EFD_* constants for consi 411 /* Check the EFD_* constants for consistency. */
389 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC) 412 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
390 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOC 413 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
391 BUILD_BUG_ON(EFD_SEMAPHORE != (1 << 0) <<
392 414
393 if (flags & ~EFD_FLAGS_SET) 415 if (flags & ~EFD_FLAGS_SET)
394 return -EINVAL; 416 return -EINVAL;
395 417
396 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL 418 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
397 if (!ctx) 419 if (!ctx)
398 return -ENOMEM; 420 return -ENOMEM;
399 421
400 kref_init(&ctx->kref); 422 kref_init(&ctx->kref);
401 init_waitqueue_head(&ctx->wqh); 423 init_waitqueue_head(&ctx->wqh);
402 ctx->count = count; 424 ctx->count = count;
403 ctx->flags = flags; 425 ctx->flags = flags;
404 ctx->id = ida_alloc(&eventfd_ida, GFP_ !! 426 ctx->id = ida_simple_get(&eventfd_ida, 0, 0, GFP_KERNEL);
405 427
406 flags &= EFD_SHARED_FCNTL_FLAGS; 428 flags &= EFD_SHARED_FCNTL_FLAGS;
407 flags |= O_RDWR; 429 flags |= O_RDWR;
408 fd = get_unused_fd_flags(flags); 430 fd = get_unused_fd_flags(flags);
409 if (fd < 0) 431 if (fd < 0)
410 goto err; 432 goto err;
411 433
412 file = anon_inode_getfile("[eventfd]", 434 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags);
413 if (IS_ERR(file)) { 435 if (IS_ERR(file)) {
414 put_unused_fd(fd); 436 put_unused_fd(fd);
415 fd = PTR_ERR(file); 437 fd = PTR_ERR(file);
416 goto err; 438 goto err;
417 } 439 }
418 440
419 file->f_mode |= FMODE_NOWAIT; 441 file->f_mode |= FMODE_NOWAIT;
420 fd_install(fd, file); 442 fd_install(fd, file);
421 return fd; 443 return fd;
422 err: 444 err:
423 eventfd_free_ctx(ctx); 445 eventfd_free_ctx(ctx);
424 return fd; 446 return fd;
425 } 447 }
426 448
427 SYSCALL_DEFINE2(eventfd2, unsigned int, count, 449 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
428 { 450 {
429 return do_eventfd(count, flags); 451 return do_eventfd(count, flags);
430 } 452 }
431 453
432 SYSCALL_DEFINE1(eventfd, unsigned int, count) 454 SYSCALL_DEFINE1(eventfd, unsigned int, count)
433 { 455 {
434 return do_eventfd(count, 0); 456 return do_eventfd(count, 0);
435 } 457 }
436 458
437 459
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.