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