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 <<
28 static DEFINE_IDA(eventfd_ida); <<
29 24
30 struct eventfd_ctx { 25 struct eventfd_ctx {
31 struct kref kref; 26 struct kref kref;
32 wait_queue_head_t wqh; 27 wait_queue_head_t wqh;
33 /* 28 /*
34 * Every time that a write(2) is perfo 29 * Every time that a write(2) is performed on an eventfd, the
35 * value of the __u64 being written is 30 * value of the __u64 being written is added to "count" and a
36 * wakeup is performed on "wqh". If EF !! 31 * wakeup is performed on "wqh". A read(2) will return the "count"
37 * specified, a read(2) will return th !! 32 * value to userspace, and will reset "count" to zero. The kernel
38 * and will reset "count" to zero. The !! 33 * side eventfd_signal() also, adds to the "count" counter and
39 * also, adds to the "count" counter a !! 34 * issue a wakeup.
40 */ 35 */
41 __u64 count; 36 __u64 count;
42 unsigned int flags; 37 unsigned int flags;
43 int id; <<
44 }; 38 };
45 39
46 /** 40 /**
47 * eventfd_signal_mask - Increment the event c !! 41 * eventfd_signal - Adds @n to the eventfd counter.
48 * @ctx: [in] Pointer to the eventfd context. 42 * @ctx: [in] Pointer to the eventfd context.
49 * @mask: [in] poll mask !! 43 * @n: [in] Value of the counter to be added to the eventfd internal counter.
>> 44 * The value cannot be negative.
50 * 45 *
51 * This function is supposed to be called by t 46 * This function is supposed to be called by the kernel in paths that do not
52 * allow sleeping. In this function we allow t 47 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
53 * value, and we signal this as overflow condi !! 48 * value, and we signal this as overflow condition by returning a POLLERR
54 * to poll(2). 49 * to poll(2).
>> 50 *
>> 51 * Returns the amount by which the counter was incremented. This will be less
>> 52 * than @n if the counter has overflowed.
55 */ 53 */
56 void eventfd_signal_mask(struct eventfd_ctx *c !! 54 __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
57 { 55 {
58 unsigned long flags; 56 unsigned long flags;
59 57
60 /* <<
61 * Deadlock or stack overflow issues c <<
62 * through waitqueue wakeup handlers. <<
63 * nested waitqueues with custom wakeu <<
64 * check eventfd_signal_allowed() befo <<
65 * it returns false, the eventfd_signa <<
66 * safe context. <<
67 */ <<
68 if (WARN_ON_ONCE(current->in_eventfd)) <<
69 return; <<
70 <<
71 spin_lock_irqsave(&ctx->wqh.lock, flag 58 spin_lock_irqsave(&ctx->wqh.lock, flags);
72 current->in_eventfd = 1; !! 59 if (ULLONG_MAX - ctx->count < n)
73 if (ctx->count < ULLONG_MAX) !! 60 n = ULLONG_MAX - ctx->count;
74 ctx->count++; !! 61 ctx->count += n;
75 if (waitqueue_active(&ctx->wqh)) 62 if (waitqueue_active(&ctx->wqh))
76 wake_up_locked_poll(&ctx->wqh, !! 63 wake_up_locked_poll(&ctx->wqh, POLLIN);
77 current->in_eventfd = 0; <<
78 spin_unlock_irqrestore(&ctx->wqh.lock, 64 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
>> 65
>> 66 return n;
79 } 67 }
80 EXPORT_SYMBOL_GPL(eventfd_signal_mask); !! 68 EXPORT_SYMBOL_GPL(eventfd_signal);
81 69
82 static void eventfd_free_ctx(struct eventfd_ct 70 static void eventfd_free_ctx(struct eventfd_ctx *ctx)
83 { 71 {
84 if (ctx->id >= 0) <<
85 ida_free(&eventfd_ida, ctx->id <<
86 kfree(ctx); 72 kfree(ctx);
87 } 73 }
88 74
89 static void eventfd_free(struct kref *kref) 75 static void eventfd_free(struct kref *kref)
90 { 76 {
91 struct eventfd_ctx *ctx = container_of 77 struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
92 78
93 eventfd_free_ctx(ctx); 79 eventfd_free_ctx(ctx);
94 } 80 }
95 81
96 /** 82 /**
>> 83 * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
>> 84 * @ctx: [in] Pointer to the eventfd context.
>> 85 *
>> 86 * Returns: In case of success, returns a pointer to the eventfd context.
>> 87 */
>> 88 struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
>> 89 {
>> 90 kref_get(&ctx->kref);
>> 91 return ctx;
>> 92 }
>> 93 EXPORT_SYMBOL_GPL(eventfd_ctx_get);
>> 94
>> 95 /**
97 * eventfd_ctx_put - Releases a reference to t 96 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
98 * @ctx: [in] Pointer to eventfd context. 97 * @ctx: [in] Pointer to eventfd context.
99 * 98 *
100 * The eventfd context reference must have bee 99 * The eventfd context reference must have been previously acquired either
101 * with eventfd_ctx_fdget() or eventfd_ctx_fil !! 100 * with eventfd_ctx_get() or eventfd_ctx_fdget().
102 */ 101 */
103 void eventfd_ctx_put(struct eventfd_ctx *ctx) 102 void eventfd_ctx_put(struct eventfd_ctx *ctx)
104 { 103 {
105 kref_put(&ctx->kref, eventfd_free); 104 kref_put(&ctx->kref, eventfd_free);
106 } 105 }
107 EXPORT_SYMBOL_GPL(eventfd_ctx_put); 106 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
108 107
109 static int eventfd_release(struct inode *inode 108 static int eventfd_release(struct inode *inode, struct file *file)
110 { 109 {
111 struct eventfd_ctx *ctx = file->privat 110 struct eventfd_ctx *ctx = file->private_data;
112 111
113 wake_up_poll(&ctx->wqh, EPOLLHUP); !! 112 wake_up_poll(&ctx->wqh, POLLHUP);
114 eventfd_ctx_put(ctx); 113 eventfd_ctx_put(ctx);
115 return 0; 114 return 0;
116 } 115 }
117 116
118 static __poll_t eventfd_poll(struct file *file !! 117 static unsigned int eventfd_poll(struct file *file, poll_table *wait)
119 { 118 {
120 struct eventfd_ctx *ctx = file->privat 119 struct eventfd_ctx *ctx = file->private_data;
121 __poll_t events = 0; !! 120 unsigned int events = 0;
122 u64 count; 121 u64 count;
123 122
124 poll_wait(file, &ctx->wqh, wait); 123 poll_wait(file, &ctx->wqh, wait);
125 124
126 /* 125 /*
127 * All writes to ctx->count occur with 126 * All writes to ctx->count occur within ctx->wqh.lock. This read
128 * can be done outside ctx->wqh.lock b 127 * can be done outside ctx->wqh.lock because we know that poll_wait
129 * takes that lock (through add_wait_q 128 * takes that lock (through add_wait_queue) if our caller will sleep.
130 * 129 *
131 * The read _can_ therefore seep into 130 * The read _can_ therefore seep into add_wait_queue's critical
132 * section, but cannot move above it! 131 * section, but cannot move above it! add_wait_queue's spin_lock acts
133 * as an acquire barrier and ensures t 132 * as an acquire barrier and ensures that the read be ordered properly
134 * against the writes. The following 133 * against the writes. The following CAN happen and is safe:
135 * 134 *
136 * poll 135 * poll write
137 * ----------------- 136 * ----------------- ------------
138 * lock ctx->wqh.lock (in poll_wai 137 * lock ctx->wqh.lock (in poll_wait)
139 * count = ctx->count 138 * count = ctx->count
140 * __add_wait_queue 139 * __add_wait_queue
141 * unlock ctx->wqh.lock 140 * unlock ctx->wqh.lock
142 * 141 * lock ctx->qwh.lock
143 * 142 * ctx->count += n
144 * 143 * if (waitqueue_active)
145 * 144 * wake_up_locked_poll
146 * 145 * unlock ctx->qwh.lock
147 * eventfd_poll returns 0 146 * eventfd_poll returns 0
148 * 147 *
149 * but the following, which would miss 148 * but the following, which would miss a wakeup, cannot happen:
150 * 149 *
151 * poll 150 * poll write
152 * ----------------- 151 * ----------------- ------------
153 * count = ctx->count (INVALID!) 152 * count = ctx->count (INVALID!)
154 * 153 * lock ctx->qwh.lock
155 * 154 * ctx->count += n
156 * 155 * **waitqueue_active is false**
157 * 156 * **no wake_up_locked_poll!**
158 * 157 * unlock ctx->qwh.lock
159 * lock ctx->wqh.lock (in poll_wai 158 * lock ctx->wqh.lock (in poll_wait)
160 * __add_wait_queue 159 * __add_wait_queue
161 * unlock ctx->wqh.lock 160 * unlock ctx->wqh.lock
162 * eventfd_poll returns 0 161 * eventfd_poll returns 0
163 */ 162 */
164 count = READ_ONCE(ctx->count); 163 count = READ_ONCE(ctx->count);
165 164
166 if (count > 0) 165 if (count > 0)
167 events |= EPOLLIN; !! 166 events |= POLLIN;
168 if (count == ULLONG_MAX) 167 if (count == ULLONG_MAX)
169 events |= EPOLLERR; !! 168 events |= POLLERR;
170 if (ULLONG_MAX - 1 > count) 169 if (ULLONG_MAX - 1 > count)
171 events |= EPOLLOUT; !! 170 events |= POLLOUT;
172 171
173 return events; 172 return events;
174 } 173 }
175 174
176 void eventfd_ctx_do_read(struct eventfd_ctx *c !! 175 static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
177 { 176 {
178 lockdep_assert_held(&ctx->wqh.lock); !! 177 *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
179 <<
180 *cnt = ((ctx->flags & EFD_SEMAPHORE) & <<
181 ctx->count -= *cnt; 178 ctx->count -= *cnt;
182 } 179 }
183 EXPORT_SYMBOL_GPL(eventfd_ctx_do_read); <<
184 180
185 /** 181 /**
186 * eventfd_ctx_remove_wait_queue - Read the cu 182 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
187 * @ctx: [in] Pointer to eventfd context. 183 * @ctx: [in] Pointer to eventfd context.
188 * @wait: [in] Wait queue to be removed. 184 * @wait: [in] Wait queue to be removed.
189 * @cnt: [out] Pointer to the 64-bit counter v 185 * @cnt: [out] Pointer to the 64-bit counter value.
190 * 186 *
191 * Returns %0 if successful, or the following 187 * Returns %0 if successful, or the following error codes:
192 * 188 *
193 * -EAGAIN : The operation would have blo 189 * -EAGAIN : The operation would have blocked.
194 * 190 *
195 * This is used to atomically remove a wait qu 191 * This is used to atomically remove a wait queue entry from the eventfd wait
196 * queue head, and read/reset the counter valu 192 * queue head, and read/reset the counter value.
197 */ 193 */
198 int eventfd_ctx_remove_wait_queue(struct event !! 194 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_t *wait,
199 __u64 *cnt) 195 __u64 *cnt)
200 { 196 {
201 unsigned long flags; 197 unsigned long flags;
202 198
203 spin_lock_irqsave(&ctx->wqh.lock, flag 199 spin_lock_irqsave(&ctx->wqh.lock, flags);
204 eventfd_ctx_do_read(ctx, cnt); 200 eventfd_ctx_do_read(ctx, cnt);
205 __remove_wait_queue(&ctx->wqh, wait); 201 __remove_wait_queue(&ctx->wqh, wait);
206 if (*cnt != 0 && waitqueue_active(&ctx 202 if (*cnt != 0 && waitqueue_active(&ctx->wqh))
207 wake_up_locked_poll(&ctx->wqh, !! 203 wake_up_locked_poll(&ctx->wqh, POLLOUT);
208 spin_unlock_irqrestore(&ctx->wqh.lock, 204 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
209 205
210 return *cnt != 0 ? 0 : -EAGAIN; 206 return *cnt != 0 ? 0 : -EAGAIN;
211 } 207 }
212 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queu 208 EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
213 209
214 static ssize_t eventfd_read(struct kiocb *iocb !! 210 /**
>> 211 * eventfd_ctx_read - Reads the eventfd counter or wait if it is zero.
>> 212 * @ctx: [in] Pointer to eventfd context.
>> 213 * @no_wait: [in] Different from zero if the operation should not block.
>> 214 * @cnt: [out] Pointer to the 64-bit counter value.
>> 215 *
>> 216 * Returns %0 if successful, or the following error codes:
>> 217 *
>> 218 * -EAGAIN : The operation would have blocked but @no_wait was non-zero.
>> 219 * -ERESTARTSYS : A signal interrupted the wait operation.
>> 220 *
>> 221 * If @no_wait is zero, the function might sleep until the eventfd internal
>> 222 * counter becomes greater than zero.
>> 223 */
>> 224 ssize_t eventfd_ctx_read(struct eventfd_ctx *ctx, int no_wait, __u64 *cnt)
215 { 225 {
216 struct file *file = iocb->ki_filp; !! 226 ssize_t res;
217 struct eventfd_ctx *ctx = file->privat !! 227 DECLARE_WAITQUEUE(wait, current);
218 __u64 ucnt = 0; <<
219 228
220 if (iov_iter_count(to) < sizeof(ucnt)) <<
221 return -EINVAL; <<
222 spin_lock_irq(&ctx->wqh.lock); 229 spin_lock_irq(&ctx->wqh.lock);
223 if (!ctx->count) { !! 230 *cnt = 0;
224 if ((file->f_flags & O_NONBLOC !! 231 res = -EAGAIN;
225 (iocb->ki_flags & IOCB_NOW !! 232 if (ctx->count > 0)
226 spin_unlock_irq(&ctx-> !! 233 res = 0;
227 return -EAGAIN; !! 234 else if (!no_wait) {
228 } !! 235 __add_wait_queue(&ctx->wqh, &wait);
229 !! 236 for (;;) {
230 if (wait_event_interruptible_l !! 237 set_current_state(TASK_INTERRUPTIBLE);
>> 238 if (ctx->count > 0) {
>> 239 res = 0;
>> 240 break;
>> 241 }
>> 242 if (signal_pending(current)) {
>> 243 res = -ERESTARTSYS;
>> 244 break;
>> 245 }
231 spin_unlock_irq(&ctx-> 246 spin_unlock_irq(&ctx->wqh.lock);
232 return -ERESTARTSYS; !! 247 schedule();
>> 248 spin_lock_irq(&ctx->wqh.lock);
233 } 249 }
>> 250 __remove_wait_queue(&ctx->wqh, &wait);
>> 251 __set_current_state(TASK_RUNNING);
>> 252 }
>> 253 if (likely(res == 0)) {
>> 254 eventfd_ctx_do_read(ctx, cnt);
>> 255 if (waitqueue_active(&ctx->wqh))
>> 256 wake_up_locked_poll(&ctx->wqh, POLLOUT);
234 } 257 }
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); 258 spin_unlock_irq(&ctx->wqh.lock);
241 if (unlikely(copy_to_iter(&ucnt, sizeo <<
242 return -EFAULT; <<
243 259
244 return sizeof(ucnt); !! 260 return res;
>> 261 }
>> 262 EXPORT_SYMBOL_GPL(eventfd_ctx_read);
>> 263
>> 264 static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
>> 265 loff_t *ppos)
>> 266 {
>> 267 struct eventfd_ctx *ctx = file->private_data;
>> 268 ssize_t res;
>> 269 __u64 cnt;
>> 270
>> 271 if (count < sizeof(cnt))
>> 272 return -EINVAL;
>> 273 res = eventfd_ctx_read(ctx, file->f_flags & O_NONBLOCK, &cnt);
>> 274 if (res < 0)
>> 275 return res;
>> 276
>> 277 return put_user(cnt, (__u64 __user *) buf) ? -EFAULT : sizeof(cnt);
245 } 278 }
246 279
247 static ssize_t eventfd_write(struct file *file 280 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
248 loff_t *ppos) 281 loff_t *ppos)
249 { 282 {
250 struct eventfd_ctx *ctx = file->privat 283 struct eventfd_ctx *ctx = file->private_data;
251 ssize_t res; 284 ssize_t res;
252 __u64 ucnt; 285 __u64 ucnt;
>> 286 DECLARE_WAITQUEUE(wait, current);
253 287
254 if (count != sizeof(ucnt)) !! 288 if (count < sizeof(ucnt))
255 return -EINVAL; 289 return -EINVAL;
256 if (copy_from_user(&ucnt, buf, sizeof( 290 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
257 return -EFAULT; 291 return -EFAULT;
258 if (ucnt == ULLONG_MAX) 292 if (ucnt == ULLONG_MAX)
259 return -EINVAL; 293 return -EINVAL;
260 spin_lock_irq(&ctx->wqh.lock); 294 spin_lock_irq(&ctx->wqh.lock);
261 res = -EAGAIN; 295 res = -EAGAIN;
262 if (ULLONG_MAX - ctx->count > ucnt) 296 if (ULLONG_MAX - ctx->count > ucnt)
263 res = sizeof(ucnt); 297 res = sizeof(ucnt);
264 else if (!(file->f_flags & O_NONBLOCK) 298 else if (!(file->f_flags & O_NONBLOCK)) {
265 res = wait_event_interruptible !! 299 __add_wait_queue(&ctx->wqh, &wait);
266 ULLONG_MAX - c !! 300 for (res = 0;;) {
267 if (!res) !! 301 set_current_state(TASK_INTERRUPTIBLE);
268 res = sizeof(ucnt); !! 302 if (ULLONG_MAX - ctx->count > ucnt) {
>> 303 res = sizeof(ucnt);
>> 304 break;
>> 305 }
>> 306 if (signal_pending(current)) {
>> 307 res = -ERESTARTSYS;
>> 308 break;
>> 309 }
>> 310 spin_unlock_irq(&ctx->wqh.lock);
>> 311 schedule();
>> 312 spin_lock_irq(&ctx->wqh.lock);
>> 313 }
>> 314 __remove_wait_queue(&ctx->wqh, &wait);
>> 315 __set_current_state(TASK_RUNNING);
269 } 316 }
270 if (likely(res > 0)) { 317 if (likely(res > 0)) {
271 ctx->count += ucnt; 318 ctx->count += ucnt;
272 current->in_eventfd = 1; <<
273 if (waitqueue_active(&ctx->wqh 319 if (waitqueue_active(&ctx->wqh))
274 wake_up_locked_poll(&c !! 320 wake_up_locked_poll(&ctx->wqh, POLLIN);
275 current->in_eventfd = 0; <<
276 } 321 }
277 spin_unlock_irq(&ctx->wqh.lock); 322 spin_unlock_irq(&ctx->wqh.lock);
278 323
279 return res; 324 return res;
280 } 325 }
281 326
282 #ifdef CONFIG_PROC_FS 327 #ifdef CONFIG_PROC_FS
283 static void eventfd_show_fdinfo(struct seq_fil 328 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
284 { 329 {
285 struct eventfd_ctx *ctx = f->private_d 330 struct eventfd_ctx *ctx = f->private_data;
286 __u64 cnt; <<
287 331
288 spin_lock_irq(&ctx->wqh.lock); 332 spin_lock_irq(&ctx->wqh.lock);
289 cnt = ctx->count; !! 333 seq_printf(m, "eventfd-count: %16llx\n",
>> 334 (unsigned long long)ctx->count);
290 spin_unlock_irq(&ctx->wqh.lock); 335 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 } 336 }
300 #endif 337 #endif
301 338
302 static const struct file_operations eventfd_fo 339 static const struct file_operations eventfd_fops = {
303 #ifdef CONFIG_PROC_FS 340 #ifdef CONFIG_PROC_FS
304 .show_fdinfo = eventfd_show_fdinfo, 341 .show_fdinfo = eventfd_show_fdinfo,
305 #endif 342 #endif
306 .release = eventfd_release, 343 .release = eventfd_release,
307 .poll = eventfd_poll, 344 .poll = eventfd_poll,
308 .read_iter = eventfd_read, !! 345 .read = eventfd_read,
309 .write = eventfd_write, 346 .write = eventfd_write,
310 .llseek = noop_llseek, 347 .llseek = noop_llseek,
311 }; 348 };
312 349
313 /** 350 /**
314 * eventfd_fget - Acquire a reference of an ev 351 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
315 * @fd: [in] Eventfd file descriptor. 352 * @fd: [in] Eventfd file descriptor.
316 * 353 *
317 * Returns a pointer to the eventfd file struc 354 * Returns a pointer to the eventfd file structure in case of success, or the
318 * following error pointer: 355 * following error pointer:
319 * 356 *
320 * -EBADF : Invalid @fd file descriptor. 357 * -EBADF : Invalid @fd file descriptor.
321 * -EINVAL : The @fd file descriptor is not 358 * -EINVAL : The @fd file descriptor is not an eventfd file.
322 */ 359 */
323 struct file *eventfd_fget(int fd) 360 struct file *eventfd_fget(int fd)
324 { 361 {
325 struct file *file; 362 struct file *file;
326 363
327 file = fget(fd); 364 file = fget(fd);
328 if (!file) 365 if (!file)
329 return ERR_PTR(-EBADF); 366 return ERR_PTR(-EBADF);
330 if (file->f_op != &eventfd_fops) { 367 if (file->f_op != &eventfd_fops) {
331 fput(file); 368 fput(file);
332 return ERR_PTR(-EINVAL); 369 return ERR_PTR(-EINVAL);
333 } 370 }
334 371
335 return file; 372 return file;
336 } 373 }
337 EXPORT_SYMBOL_GPL(eventfd_fget); 374 EXPORT_SYMBOL_GPL(eventfd_fget);
338 375
339 /** 376 /**
340 * eventfd_ctx_fdget - Acquires a reference to 377 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
341 * @fd: [in] Eventfd file descriptor. 378 * @fd: [in] Eventfd file descriptor.
342 * 379 *
343 * Returns a pointer to the internal eventfd c 380 * Returns a pointer to the internal eventfd context, otherwise the error
344 * pointers returned by the following function 381 * pointers returned by the following functions:
345 * 382 *
346 * eventfd_fget 383 * eventfd_fget
347 */ 384 */
348 struct eventfd_ctx *eventfd_ctx_fdget(int fd) 385 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
349 { 386 {
350 struct eventfd_ctx *ctx; 387 struct eventfd_ctx *ctx;
351 struct fd f = fdget(fd); 388 struct fd f = fdget(fd);
352 if (!f.file) 389 if (!f.file)
353 return ERR_PTR(-EBADF); 390 return ERR_PTR(-EBADF);
354 ctx = eventfd_ctx_fileget(f.file); 391 ctx = eventfd_ctx_fileget(f.file);
355 fdput(f); 392 fdput(f);
356 return ctx; 393 return ctx;
357 } 394 }
358 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget); 395 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
359 396
360 /** 397 /**
361 * eventfd_ctx_fileget - Acquires a reference 398 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
362 * @file: [in] Eventfd file pointer. 399 * @file: [in] Eventfd file pointer.
363 * 400 *
364 * Returns a pointer to the internal eventfd c 401 * Returns a pointer to the internal eventfd context, otherwise the error
365 * pointer: 402 * pointer:
366 * 403 *
367 * -EINVAL : The @fd file descriptor is not 404 * -EINVAL : The @fd file descriptor is not an eventfd file.
368 */ 405 */
369 struct eventfd_ctx *eventfd_ctx_fileget(struct 406 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
370 { 407 {
371 struct eventfd_ctx *ctx; <<
372 <<
373 if (file->f_op != &eventfd_fops) 408 if (file->f_op != &eventfd_fops)
374 return ERR_PTR(-EINVAL); 409 return ERR_PTR(-EINVAL);
375 410
376 ctx = file->private_data; !! 411 return eventfd_ctx_get(file->private_data);
377 kref_get(&ctx->kref); <<
378 return ctx; <<
379 } 412 }
380 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget); 413 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
381 414
382 static int do_eventfd(unsigned int count, int !! 415 /**
>> 416 * eventfd_file_create - Creates an eventfd file pointer.
>> 417 * @count: Initial eventfd counter value.
>> 418 * @flags: Flags for the eventfd file.
>> 419 *
>> 420 * This function creates an eventfd file pointer, w/out installing it into
>> 421 * the fd table. This is useful when the eventfd file is used during the
>> 422 * initialization of data structures that require extra setup after the eventfd
>> 423 * creation. So the eventfd creation is split into the file pointer creation
>> 424 * phase, and the file descriptor installation phase.
>> 425 * In this way races with userspace closing the newly installed file descriptor
>> 426 * can be avoided.
>> 427 * Returns an eventfd file pointer, or a proper error pointer.
>> 428 */
>> 429 struct file *eventfd_file_create(unsigned int count, int flags)
383 { 430 {
384 struct eventfd_ctx *ctx; <<
385 struct file *file; 431 struct file *file;
386 int fd; !! 432 struct eventfd_ctx *ctx;
387 433
388 /* Check the EFD_* constants for consi 434 /* Check the EFD_* constants for consistency. */
389 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC) 435 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
390 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOC 436 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
391 BUILD_BUG_ON(EFD_SEMAPHORE != (1 << 0) <<
392 437
393 if (flags & ~EFD_FLAGS_SET) 438 if (flags & ~EFD_FLAGS_SET)
394 return -EINVAL; !! 439 return ERR_PTR(-EINVAL);
395 440
396 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL 441 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
397 if (!ctx) 442 if (!ctx)
398 return -ENOMEM; !! 443 return ERR_PTR(-ENOMEM);
399 444
400 kref_init(&ctx->kref); 445 kref_init(&ctx->kref);
401 init_waitqueue_head(&ctx->wqh); 446 init_waitqueue_head(&ctx->wqh);
402 ctx->count = count; 447 ctx->count = count;
403 ctx->flags = flags; 448 ctx->flags = flags;
404 ctx->id = ida_alloc(&eventfd_ida, GFP_ <<
405 449
406 flags &= EFD_SHARED_FCNTL_FLAGS; !! 450 file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
407 flags |= O_RDWR; !! 451 O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
408 fd = get_unused_fd_flags(flags); !! 452 if (IS_ERR(file))
409 if (fd < 0) !! 453 eventfd_free_ctx(ctx);
410 goto err; !! 454
>> 455 return file;
>> 456 }
>> 457
>> 458 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
>> 459 {
>> 460 int fd, error;
>> 461 struct file *file;
>> 462
>> 463 error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
>> 464 if (error < 0)
>> 465 return error;
>> 466 fd = error;
411 467
412 file = anon_inode_getfile("[eventfd]", !! 468 file = eventfd_file_create(count, flags);
413 if (IS_ERR(file)) { 469 if (IS_ERR(file)) {
414 put_unused_fd(fd); !! 470 error = PTR_ERR(file);
415 fd = PTR_ERR(file); !! 471 goto err_put_unused_fd;
416 goto err; <<
417 } 472 }
418 <<
419 file->f_mode |= FMODE_NOWAIT; <<
420 fd_install(fd, file); 473 fd_install(fd, file);
>> 474
421 return fd; 475 return fd;
422 err: <<
423 eventfd_free_ctx(ctx); <<
424 return fd; <<
425 } <<
426 476
427 SYSCALL_DEFINE2(eventfd2, unsigned int, count, !! 477 err_put_unused_fd:
428 { !! 478 put_unused_fd(fd);
429 return do_eventfd(count, flags); !! 479
>> 480 return error;
430 } 481 }
431 482
432 SYSCALL_DEFINE1(eventfd, unsigned int, count) 483 SYSCALL_DEFINE1(eventfd, unsigned int, count)
433 { 484 {
434 return do_eventfd(count, 0); !! 485 return sys_eventfd2(count, 0);
435 } 486 }
436 487
437 488
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.