Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/eventpoll.c ( Efficent event polling implementation ) | |
3419b23a | 3 | * Copyright (C) 2001,...,2006 Davide Libenzi |
1da177e4 LT |
4 | * |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * Davide Libenzi <davidel@xmailserver.org> | |
11 | * | |
12 | */ | |
13 | ||
14 | #include <linux/module.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/kernel.h> | |
17 | #include <linux/sched.h> | |
18 | #include <linux/fs.h> | |
19 | #include <linux/file.h> | |
20 | #include <linux/signal.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/mm.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/poll.h> | |
25 | #include <linux/smp_lock.h> | |
26 | #include <linux/string.h> | |
27 | #include <linux/list.h> | |
28 | #include <linux/hash.h> | |
29 | #include <linux/spinlock.h> | |
30 | #include <linux/syscalls.h> | |
31 | #include <linux/rwsem.h> | |
32 | #include <linux/rbtree.h> | |
33 | #include <linux/wait.h> | |
34 | #include <linux/eventpoll.h> | |
35 | #include <linux/mount.h> | |
36 | #include <linux/bitops.h> | |
144efe3e | 37 | #include <linux/mutex.h> |
1da177e4 LT |
38 | #include <asm/uaccess.h> |
39 | #include <asm/system.h> | |
40 | #include <asm/io.h> | |
41 | #include <asm/mman.h> | |
42 | #include <asm/atomic.h> | |
43 | #include <asm/semaphore.h> | |
44 | ||
45 | ||
46 | /* | |
47 | * LOCKING: | |
48 | * There are three level of locking required by epoll : | |
49 | * | |
144efe3e | 50 | * 1) epmutex (mutex) |
1da177e4 LT |
51 | * 2) ep->sem (rw_semaphore) |
52 | * 3) ep->lock (rw_lock) | |
53 | * | |
54 | * The acquire order is the one listed above, from 1 to 3. | |
55 | * We need a spinlock (ep->lock) because we manipulate objects | |
56 | * from inside the poll callback, that might be triggered from | |
57 | * a wake_up() that in turn might be called from IRQ context. | |
58 | * So we can't sleep inside the poll callback and hence we need | |
59 | * a spinlock. During the event transfer loop (from kernel to | |
60 | * user space) we could end up sleeping due a copy_to_user(), so | |
61 | * we need a lock that will allow us to sleep. This lock is a | |
62 | * read-write semaphore (ep->sem). It is acquired on read during | |
63 | * the event transfer loop and in write during epoll_ctl(EPOLL_CTL_DEL) | |
64 | * and during eventpoll_release_file(). Then we also need a global | |
65 | * semaphore to serialize eventpoll_release_file() and ep_free(). | |
66 | * This semaphore is acquired by ep_free() during the epoll file | |
67 | * cleanup path and it is also acquired by eventpoll_release_file() | |
68 | * if a file has been pushed inside an epoll set and it is then | |
69 | * close()d without a previous call toepoll_ctl(EPOLL_CTL_DEL). | |
70 | * It is possible to drop the "ep->sem" and to use the global | |
144efe3e | 71 | * semaphore "epmutex" (together with "ep->lock") to have it working, |
1da177e4 | 72 | * but having "ep->sem" will make the interface more scalable. |
144efe3e | 73 | * Events that require holding "epmutex" are very rare, while for |
1da177e4 LT |
74 | * normal operations the epoll private "ep->sem" will guarantee |
75 | * a greater scalability. | |
76 | */ | |
77 | ||
78 | ||
79 | #define EVENTPOLLFS_MAGIC 0x03111965 /* My birthday should work for this :) */ | |
80 | ||
81 | #define DEBUG_EPOLL 0 | |
82 | ||
83 | #if DEBUG_EPOLL > 0 | |
84 | #define DPRINTK(x) printk x | |
85 | #define DNPRINTK(n, x) do { if ((n) <= DEBUG_EPOLL) printk x; } while (0) | |
86 | #else /* #if DEBUG_EPOLL > 0 */ | |
87 | #define DPRINTK(x) (void) 0 | |
88 | #define DNPRINTK(n, x) (void) 0 | |
89 | #endif /* #if DEBUG_EPOLL > 0 */ | |
90 | ||
91 | #define DEBUG_EPI 0 | |
92 | ||
93 | #if DEBUG_EPI != 0 | |
94 | #define EPI_SLAB_DEBUG (SLAB_DEBUG_FREE | SLAB_RED_ZONE /* | SLAB_POISON */) | |
95 | #else /* #if DEBUG_EPI != 0 */ | |
96 | #define EPI_SLAB_DEBUG 0 | |
97 | #endif /* #if DEBUG_EPI != 0 */ | |
98 | ||
99 | /* Epoll private bits inside the event mask */ | |
100 | #define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET) | |
101 | ||
102 | /* Maximum number of poll wake up nests we are allowing */ | |
103 | #define EP_MAX_POLLWAKE_NESTS 4 | |
104 | ||
e3306dd5 DL |
105 | /* Maximum msec timeout value storeable in a long int */ |
106 | #define EP_MAX_MSTIMEO min(1000ULL * MAX_SCHEDULE_TIMEOUT / HZ, (LONG_MAX - 999ULL) / HZ) | |
107 | ||
108 | ||
1da177e4 LT |
109 | struct epoll_filefd { |
110 | struct file *file; | |
111 | int fd; | |
112 | }; | |
113 | ||
114 | /* | |
115 | * Node that is linked into the "wake_task_list" member of the "struct poll_safewake". | |
116 | * It is used to keep track on all tasks that are currently inside the wake_up() code | |
117 | * to 1) short-circuit the one coming from the same task and same wait queue head | |
118 | * ( loop ) 2) allow a maximum number of epoll descriptors inclusion nesting | |
119 | * 3) let go the ones coming from other tasks. | |
120 | */ | |
121 | struct wake_task_node { | |
122 | struct list_head llink; | |
36c8b586 | 123 | struct task_struct *task; |
1da177e4 LT |
124 | wait_queue_head_t *wq; |
125 | }; | |
126 | ||
127 | /* | |
128 | * This is used to implement the safe poll wake up avoiding to reenter | |
129 | * the poll callback from inside wake_up(). | |
130 | */ | |
131 | struct poll_safewake { | |
132 | struct list_head wake_task_list; | |
133 | spinlock_t lock; | |
134 | }; | |
135 | ||
136 | /* | |
137 | * This structure is stored inside the "private_data" member of the file | |
138 | * structure and rapresent the main data sructure for the eventpoll | |
139 | * interface. | |
140 | */ | |
141 | struct eventpoll { | |
142 | /* Protect the this structure access */ | |
143 | rwlock_t lock; | |
144 | ||
145 | /* | |
146 | * This semaphore is used to ensure that files are not removed | |
147 | * while epoll is using them. This is read-held during the event | |
148 | * collection loop and it is write-held during the file cleanup | |
149 | * path, the epoll file exit code and the ctl operations. | |
150 | */ | |
151 | struct rw_semaphore sem; | |
152 | ||
153 | /* Wait queue used by sys_epoll_wait() */ | |
154 | wait_queue_head_t wq; | |
155 | ||
156 | /* Wait queue used by file->poll() */ | |
157 | wait_queue_head_t poll_wait; | |
158 | ||
159 | /* List of ready file descriptors */ | |
160 | struct list_head rdllist; | |
161 | ||
162 | /* RB-Tree root used to store monitored fd structs */ | |
163 | struct rb_root rbr; | |
164 | }; | |
165 | ||
166 | /* Wait structure used by the poll hooks */ | |
167 | struct eppoll_entry { | |
168 | /* List header used to link this structure to the "struct epitem" */ | |
169 | struct list_head llink; | |
170 | ||
171 | /* The "base" pointer is set to the container "struct epitem" */ | |
172 | void *base; | |
173 | ||
174 | /* | |
175 | * Wait queue item that will be linked to the target file wait | |
176 | * queue head. | |
177 | */ | |
178 | wait_queue_t wait; | |
179 | ||
180 | /* The wait queue head that linked the "wait" wait queue item */ | |
181 | wait_queue_head_t *whead; | |
182 | }; | |
183 | ||
184 | /* | |
185 | * Each file descriptor added to the eventpoll interface will | |
186 | * have an entry of this type linked to the hash. | |
187 | */ | |
188 | struct epitem { | |
189 | /* RB-Tree node used to link this structure to the eventpoll rb-tree */ | |
190 | struct rb_node rbn; | |
191 | ||
192 | /* List header used to link this structure to the eventpoll ready list */ | |
193 | struct list_head rdllink; | |
194 | ||
195 | /* The file descriptor information this item refers to */ | |
196 | struct epoll_filefd ffd; | |
197 | ||
198 | /* Number of active wait queue attached to poll operations */ | |
199 | int nwait; | |
200 | ||
201 | /* List containing poll wait queues */ | |
202 | struct list_head pwqlist; | |
203 | ||
204 | /* The "container" of this item */ | |
205 | struct eventpoll *ep; | |
206 | ||
207 | /* The structure that describe the interested events and the source fd */ | |
208 | struct epoll_event event; | |
209 | ||
210 | /* | |
211 | * Used to keep track of the usage count of the structure. This avoids | |
212 | * that the structure will desappear from underneath our processing. | |
213 | */ | |
214 | atomic_t usecnt; | |
215 | ||
216 | /* List header used to link this item to the "struct file" items list */ | |
217 | struct list_head fllink; | |
218 | ||
219 | /* List header used to link the item to the transfer list */ | |
220 | struct list_head txlink; | |
221 | ||
222 | /* | |
223 | * This is used during the collection/transfer of events to userspace | |
224 | * to pin items empty events set. | |
225 | */ | |
226 | unsigned int revents; | |
227 | }; | |
228 | ||
229 | /* Wrapper struct used by poll queueing */ | |
230 | struct ep_pqueue { | |
231 | poll_table pt; | |
232 | struct epitem *epi; | |
233 | }; | |
234 | ||
235 | ||
236 | ||
237 | static void ep_poll_safewake_init(struct poll_safewake *psw); | |
238 | static void ep_poll_safewake(struct poll_safewake *psw, wait_queue_head_t *wq); | |
53d2be79 DL |
239 | static int ep_getfd(int *efd, struct inode **einode, struct file **efile, |
240 | struct eventpoll *ep); | |
241 | static int ep_alloc(struct eventpoll **pep); | |
1da177e4 LT |
242 | static void ep_free(struct eventpoll *ep); |
243 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd); | |
244 | static void ep_use_epitem(struct epitem *epi); | |
245 | static void ep_release_epitem(struct epitem *epi); | |
246 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
247 | poll_table *pt); | |
248 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi); | |
249 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, | |
250 | struct file *tfile, int fd); | |
251 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, | |
252 | struct epoll_event *event); | |
253 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi); | |
254 | static int ep_unlink(struct eventpoll *ep, struct epitem *epi); | |
255 | static int ep_remove(struct eventpoll *ep, struct epitem *epi); | |
256 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key); | |
257 | static int ep_eventpoll_close(struct inode *inode, struct file *file); | |
258 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait); | |
259 | static int ep_collect_ready_items(struct eventpoll *ep, | |
260 | struct list_head *txlist, int maxevents); | |
261 | static int ep_send_events(struct eventpoll *ep, struct list_head *txlist, | |
262 | struct epoll_event __user *events); | |
263 | static void ep_reinject_items(struct eventpoll *ep, struct list_head *txlist); | |
264 | static int ep_events_transfer(struct eventpoll *ep, | |
265 | struct epoll_event __user *events, | |
266 | int maxevents); | |
267 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, | |
268 | int maxevents, long timeout); | |
269 | static int eventpollfs_delete_dentry(struct dentry *dentry); | |
270 | static struct inode *ep_eventpoll_inode(void); | |
454e2398 DH |
271 | static int eventpollfs_get_sb(struct file_system_type *fs_type, |
272 | int flags, const char *dev_name, | |
273 | void *data, struct vfsmount *mnt); | |
1da177e4 LT |
274 | |
275 | /* | |
276 | * This semaphore is used to serialize ep_free() and eventpoll_release_file(). | |
277 | */ | |
144efe3e | 278 | static struct mutex epmutex; |
1da177e4 LT |
279 | |
280 | /* Safe wake up implementation */ | |
281 | static struct poll_safewake psw; | |
282 | ||
283 | /* Slab cache used to allocate "struct epitem" */ | |
fa3536cc | 284 | static kmem_cache_t *epi_cache __read_mostly; |
1da177e4 LT |
285 | |
286 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
fa3536cc | 287 | static kmem_cache_t *pwq_cache __read_mostly; |
1da177e4 LT |
288 | |
289 | /* Virtual fs used to allocate inodes for eventpoll files */ | |
fa3536cc | 290 | static struct vfsmount *eventpoll_mnt __read_mostly; |
1da177e4 LT |
291 | |
292 | /* File callbacks that implement the eventpoll file behaviour */ | |
4b6f5d20 | 293 | static const struct file_operations eventpoll_fops = { |
1da177e4 LT |
294 | .release = ep_eventpoll_close, |
295 | .poll = ep_eventpoll_poll | |
296 | }; | |
297 | ||
298 | /* | |
299 | * This is used to register the virtual file system from where | |
300 | * eventpoll inodes are allocated. | |
301 | */ | |
302 | static struct file_system_type eventpoll_fs_type = { | |
303 | .name = "eventpollfs", | |
304 | .get_sb = eventpollfs_get_sb, | |
305 | .kill_sb = kill_anon_super, | |
306 | }; | |
307 | ||
308 | /* Very basic directory entry operations for the eventpoll virtual file system */ | |
309 | static struct dentry_operations eventpollfs_dentry_operations = { | |
310 | .d_delete = eventpollfs_delete_dentry, | |
311 | }; | |
312 | ||
313 | ||
314 | ||
b030a4dd PE |
315 | /* Fast test to see if the file is an evenpoll file */ |
316 | static inline int is_file_epoll(struct file *f) | |
317 | { | |
318 | return f->f_op == &eventpoll_fops; | |
319 | } | |
320 | ||
321 | /* Setup the structure that is used as key for the rb-tree */ | |
322 | static inline void ep_set_ffd(struct epoll_filefd *ffd, | |
323 | struct file *file, int fd) | |
324 | { | |
325 | ffd->file = file; | |
326 | ffd->fd = fd; | |
327 | } | |
328 | ||
329 | /* Compare rb-tree keys */ | |
330 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, | |
331 | struct epoll_filefd *p2) | |
332 | { | |
333 | return (p1->file > p2->file ? +1: | |
334 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
335 | } | |
336 | ||
337 | /* Special initialization for the rb-tree node to detect linkage */ | |
338 | static inline void ep_rb_initnode(struct rb_node *n) | |
339 | { | |
c569882b | 340 | rb_set_parent(n, n); |
b030a4dd PE |
341 | } |
342 | ||
343 | /* Removes a node from the rb-tree and marks it for a fast is-linked check */ | |
344 | static inline void ep_rb_erase(struct rb_node *n, struct rb_root *r) | |
345 | { | |
346 | rb_erase(n, r); | |
c569882b | 347 | rb_set_parent(n, n); |
b030a4dd PE |
348 | } |
349 | ||
350 | /* Fast check to verify that the item is linked to the main rb-tree */ | |
351 | static inline int ep_rb_linked(struct rb_node *n) | |
352 | { | |
c569882b | 353 | return rb_parent(n) != n; |
b030a4dd PE |
354 | } |
355 | ||
356 | /* | |
357 | * Remove the item from the list and perform its initialization. | |
358 | * This is useful for us because we can test if the item is linked | |
359 | * using "ep_is_linked(p)". | |
360 | */ | |
361 | static inline void ep_list_del(struct list_head *p) | |
362 | { | |
363 | list_del(p); | |
364 | INIT_LIST_HEAD(p); | |
365 | } | |
366 | ||
367 | /* Tells us if the item is currently linked */ | |
368 | static inline int ep_is_linked(struct list_head *p) | |
369 | { | |
370 | return !list_empty(p); | |
371 | } | |
372 | ||
373 | /* Get the "struct epitem" from a wait queue pointer */ | |
374 | static inline struct epitem * ep_item_from_wait(wait_queue_t *p) | |
375 | { | |
376 | return container_of(p, struct eppoll_entry, wait)->base; | |
377 | } | |
378 | ||
379 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
380 | static inline struct epitem * ep_item_from_epqueue(poll_table *p) | |
381 | { | |
382 | return container_of(p, struct ep_pqueue, pt)->epi; | |
383 | } | |
384 | ||
385 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
386 | static inline int ep_op_hash_event(int op) | |
387 | { | |
388 | return op != EPOLL_CTL_DEL; | |
389 | } | |
390 | ||
1da177e4 LT |
391 | /* Initialize the poll safe wake up structure */ |
392 | static void ep_poll_safewake_init(struct poll_safewake *psw) | |
393 | { | |
394 | ||
395 | INIT_LIST_HEAD(&psw->wake_task_list); | |
396 | spin_lock_init(&psw->lock); | |
397 | } | |
398 | ||
399 | ||
400 | /* | |
401 | * Perform a safe wake up of the poll wait list. The problem is that | |
402 | * with the new callback'd wake up system, it is possible that the | |
403 | * poll callback is reentered from inside the call to wake_up() done | |
404 | * on the poll wait queue head. The rule is that we cannot reenter the | |
405 | * wake up code from the same task more than EP_MAX_POLLWAKE_NESTS times, | |
406 | * and we cannot reenter the same wait queue head at all. This will | |
407 | * enable to have a hierarchy of epoll file descriptor of no more than | |
408 | * EP_MAX_POLLWAKE_NESTS deep. We need the irq version of the spin lock | |
409 | * because this one gets called by the poll callback, that in turn is called | |
410 | * from inside a wake_up(), that might be called from irq context. | |
411 | */ | |
412 | static void ep_poll_safewake(struct poll_safewake *psw, wait_queue_head_t *wq) | |
413 | { | |
414 | int wake_nests = 0; | |
415 | unsigned long flags; | |
36c8b586 | 416 | struct task_struct *this_task = current; |
1da177e4 LT |
417 | struct list_head *lsthead = &psw->wake_task_list, *lnk; |
418 | struct wake_task_node *tncur; | |
419 | struct wake_task_node tnode; | |
420 | ||
421 | spin_lock_irqsave(&psw->lock, flags); | |
422 | ||
423 | /* Try to see if the current task is already inside this wakeup call */ | |
424 | list_for_each(lnk, lsthead) { | |
425 | tncur = list_entry(lnk, struct wake_task_node, llink); | |
426 | ||
427 | if (tncur->wq == wq || | |
428 | (tncur->task == this_task && ++wake_nests > EP_MAX_POLLWAKE_NESTS)) { | |
429 | /* | |
430 | * Ops ... loop detected or maximum nest level reached. | |
431 | * We abort this wake by breaking the cycle itself. | |
432 | */ | |
433 | spin_unlock_irqrestore(&psw->lock, flags); | |
434 | return; | |
435 | } | |
436 | } | |
437 | ||
438 | /* Add the current task to the list */ | |
439 | tnode.task = this_task; | |
440 | tnode.wq = wq; | |
441 | list_add(&tnode.llink, lsthead); | |
442 | ||
443 | spin_unlock_irqrestore(&psw->lock, flags); | |
444 | ||
445 | /* Do really wake up now */ | |
446 | wake_up(wq); | |
447 | ||
448 | /* Remove the current task from the list */ | |
449 | spin_lock_irqsave(&psw->lock, flags); | |
450 | list_del(&tnode.llink); | |
451 | spin_unlock_irqrestore(&psw->lock, flags); | |
452 | } | |
453 | ||
454 | ||
1da177e4 LT |
455 | /* |
456 | * This is called from eventpoll_release() to unlink files from the eventpoll | |
457 | * interface. We need to have this facility to cleanup correctly files that are | |
458 | * closed without being removed from the eventpoll interface. | |
459 | */ | |
460 | void eventpoll_release_file(struct file *file) | |
461 | { | |
462 | struct list_head *lsthead = &file->f_ep_links; | |
463 | struct eventpoll *ep; | |
464 | struct epitem *epi; | |
465 | ||
466 | /* | |
467 | * We don't want to get "file->f_ep_lock" because it is not | |
468 | * necessary. It is not necessary because we're in the "struct file" | |
469 | * cleanup path, and this means that noone is using this file anymore. | |
470 | * The only hit might come from ep_free() but by holding the semaphore | |
471 | * will correctly serialize the operation. We do need to acquire | |
144efe3e | 472 | * "ep->sem" after "epmutex" because ep_remove() requires it when called |
1da177e4 LT |
473 | * from anywhere but ep_free(). |
474 | */ | |
144efe3e | 475 | mutex_lock(&epmutex); |
1da177e4 LT |
476 | |
477 | while (!list_empty(lsthead)) { | |
478 | epi = list_entry(lsthead->next, struct epitem, fllink); | |
479 | ||
480 | ep = epi->ep; | |
b030a4dd | 481 | ep_list_del(&epi->fllink); |
1da177e4 LT |
482 | down_write(&ep->sem); |
483 | ep_remove(ep, epi); | |
484 | up_write(&ep->sem); | |
485 | } | |
486 | ||
144efe3e | 487 | mutex_unlock(&epmutex); |
1da177e4 LT |
488 | } |
489 | ||
490 | ||
491 | /* | |
492 | * It opens an eventpoll file descriptor by suggesting a storage of "size" | |
493 | * file descriptors. The size parameter is just an hint about how to size | |
494 | * data structures. It won't prevent the user to store more than "size" | |
495 | * file descriptors inside the epoll interface. It is the kernel part of | |
496 | * the userspace epoll_create(2). | |
497 | */ | |
498 | asmlinkage long sys_epoll_create(int size) | |
499 | { | |
500 | int error, fd; | |
53d2be79 | 501 | struct eventpoll *ep; |
1da177e4 LT |
502 | struct inode *inode; |
503 | struct file *file; | |
504 | ||
505 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d)\n", | |
506 | current, size)); | |
507 | ||
53d2be79 DL |
508 | /* |
509 | * Sanity check on the size parameter, and create the internal data | |
510 | * structure ( "struct eventpoll" ). | |
511 | */ | |
1da177e4 | 512 | error = -EINVAL; |
53d2be79 | 513 | if (size <= 0 || (error = ep_alloc(&ep)) != 0) |
1da177e4 LT |
514 | goto eexit_1; |
515 | ||
516 | /* | |
517 | * Creates all the items needed to setup an eventpoll file. That is, | |
518 | * a file structure, and inode and a free file descriptor. | |
519 | */ | |
53d2be79 | 520 | error = ep_getfd(&fd, &inode, &file, ep); |
1da177e4 LT |
521 | if (error) |
522 | goto eexit_2; | |
523 | ||
1da177e4 LT |
524 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", |
525 | current, size, fd)); | |
526 | ||
527 | return fd; | |
528 | ||
529 | eexit_2: | |
53d2be79 DL |
530 | ep_free(ep); |
531 | kfree(ep); | |
1da177e4 LT |
532 | eexit_1: |
533 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", | |
534 | current, size, error)); | |
535 | return error; | |
536 | } | |
537 | ||
538 | ||
539 | /* | |
540 | * The following function implements the controller interface for | |
541 | * the eventpoll file that enables the insertion/removal/change of | |
542 | * file descriptors inside the interest set. It represents | |
543 | * the kernel part of the user space epoll_ctl(2). | |
544 | */ | |
545 | asmlinkage long | |
546 | sys_epoll_ctl(int epfd, int op, int fd, struct epoll_event __user *event) | |
547 | { | |
548 | int error; | |
549 | struct file *file, *tfile; | |
550 | struct eventpoll *ep; | |
551 | struct epitem *epi; | |
552 | struct epoll_event epds; | |
553 | ||
554 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p)\n", | |
555 | current, epfd, op, fd, event)); | |
556 | ||
557 | error = -EFAULT; | |
b030a4dd | 558 | if (ep_op_hash_event(op) && |
1da177e4 LT |
559 | copy_from_user(&epds, event, sizeof(struct epoll_event))) |
560 | goto eexit_1; | |
561 | ||
562 | /* Get the "struct file *" for the eventpoll file */ | |
563 | error = -EBADF; | |
564 | file = fget(epfd); | |
565 | if (!file) | |
566 | goto eexit_1; | |
567 | ||
568 | /* Get the "struct file *" for the target file */ | |
569 | tfile = fget(fd); | |
570 | if (!tfile) | |
571 | goto eexit_2; | |
572 | ||
573 | /* The target file descriptor must support poll */ | |
574 | error = -EPERM; | |
575 | if (!tfile->f_op || !tfile->f_op->poll) | |
576 | goto eexit_3; | |
577 | ||
578 | /* | |
579 | * We have to check that the file structure underneath the file descriptor | |
580 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
581 | * adding an epoll file descriptor inside itself. | |
582 | */ | |
583 | error = -EINVAL; | |
b030a4dd | 584 | if (file == tfile || !is_file_epoll(file)) |
1da177e4 LT |
585 | goto eexit_3; |
586 | ||
587 | /* | |
588 | * At this point it is safe to assume that the "private_data" contains | |
589 | * our own data structure. | |
590 | */ | |
591 | ep = file->private_data; | |
592 | ||
593 | down_write(&ep->sem); | |
594 | ||
595 | /* Try to lookup the file inside our hash table */ | |
596 | epi = ep_find(ep, tfile, fd); | |
597 | ||
598 | error = -EINVAL; | |
599 | switch (op) { | |
600 | case EPOLL_CTL_ADD: | |
601 | if (!epi) { | |
2395140e | 602 | epds.events |= POLLERR | POLLHUP; |
1da177e4 LT |
603 | |
604 | error = ep_insert(ep, &epds, tfile, fd); | |
605 | } else | |
606 | error = -EEXIST; | |
607 | break; | |
608 | case EPOLL_CTL_DEL: | |
609 | if (epi) | |
610 | error = ep_remove(ep, epi); | |
611 | else | |
612 | error = -ENOENT; | |
613 | break; | |
614 | case EPOLL_CTL_MOD: | |
615 | if (epi) { | |
2395140e | 616 | epds.events |= POLLERR | POLLHUP; |
1da177e4 LT |
617 | error = ep_modify(ep, epi, &epds); |
618 | } else | |
619 | error = -ENOENT; | |
620 | break; | |
621 | } | |
622 | ||
623 | /* | |
624 | * The function ep_find() increments the usage count of the structure | |
625 | * so, if this is not NULL, we need to release it. | |
626 | */ | |
627 | if (epi) | |
628 | ep_release_epitem(epi); | |
629 | ||
630 | up_write(&ep->sem); | |
631 | ||
632 | eexit_3: | |
633 | fput(tfile); | |
634 | eexit_2: | |
635 | fput(file); | |
636 | eexit_1: | |
637 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p) = %d\n", | |
638 | current, epfd, op, fd, event, error)); | |
639 | ||
640 | return error; | |
641 | } | |
642 | ||
643 | #define MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) | |
644 | ||
645 | /* | |
646 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
647 | * part of the user space epoll_wait(2). | |
648 | */ | |
649 | asmlinkage long sys_epoll_wait(int epfd, struct epoll_event __user *events, | |
650 | int maxevents, int timeout) | |
651 | { | |
652 | int error; | |
653 | struct file *file; | |
654 | struct eventpoll *ep; | |
655 | ||
656 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d)\n", | |
657 | current, epfd, events, maxevents, timeout)); | |
658 | ||
659 | /* The maximum number of event must be greater than zero */ | |
660 | if (maxevents <= 0 || maxevents > MAX_EVENTS) | |
661 | return -EINVAL; | |
662 | ||
663 | /* Verify that the area passed by the user is writeable */ | |
664 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { | |
665 | error = -EFAULT; | |
666 | goto eexit_1; | |
667 | } | |
668 | ||
669 | /* Get the "struct file *" for the eventpoll file */ | |
670 | error = -EBADF; | |
671 | file = fget(epfd); | |
672 | if (!file) | |
673 | goto eexit_1; | |
674 | ||
675 | /* | |
676 | * We have to check that the file structure underneath the fd | |
677 | * the user passed to us _is_ an eventpoll file. | |
678 | */ | |
679 | error = -EINVAL; | |
b030a4dd | 680 | if (!is_file_epoll(file)) |
1da177e4 LT |
681 | goto eexit_2; |
682 | ||
683 | /* | |
684 | * At this point it is safe to assume that the "private_data" contains | |
685 | * our own data structure. | |
686 | */ | |
687 | ep = file->private_data; | |
688 | ||
689 | /* Time to fish for events ... */ | |
690 | error = ep_poll(ep, events, maxevents, timeout); | |
691 | ||
692 | eexit_2: | |
693 | fput(file); | |
694 | eexit_1: | |
695 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d) = %d\n", | |
696 | current, epfd, events, maxevents, timeout, error)); | |
697 | ||
698 | return error; | |
699 | } | |
700 | ||
701 | ||
702 | /* | |
703 | * Creates the file descriptor to be used by the epoll interface. | |
704 | */ | |
53d2be79 DL |
705 | static int ep_getfd(int *efd, struct inode **einode, struct file **efile, |
706 | struct eventpoll *ep) | |
1da177e4 LT |
707 | { |
708 | struct qstr this; | |
709 | char name[32]; | |
710 | struct dentry *dentry; | |
711 | struct inode *inode; | |
712 | struct file *file; | |
713 | int error, fd; | |
714 | ||
715 | /* Get an ready to use file */ | |
716 | error = -ENFILE; | |
717 | file = get_empty_filp(); | |
718 | if (!file) | |
719 | goto eexit_1; | |
720 | ||
721 | /* Allocates an inode from the eventpoll file system */ | |
722 | inode = ep_eventpoll_inode(); | |
c3b65713 JG |
723 | if (IS_ERR(inode)) { |
724 | error = PTR_ERR(inode); | |
1da177e4 | 725 | goto eexit_2; |
c3b65713 | 726 | } |
1da177e4 LT |
727 | |
728 | /* Allocates a free descriptor to plug the file onto */ | |
729 | error = get_unused_fd(); | |
730 | if (error < 0) | |
731 | goto eexit_3; | |
732 | fd = error; | |
733 | ||
734 | /* | |
735 | * Link the inode to a directory entry by creating a unique name | |
736 | * using the inode number. | |
737 | */ | |
738 | error = -ENOMEM; | |
739 | sprintf(name, "[%lu]", inode->i_ino); | |
740 | this.name = name; | |
741 | this.len = strlen(name); | |
742 | this.hash = inode->i_ino; | |
743 | dentry = d_alloc(eventpoll_mnt->mnt_sb->s_root, &this); | |
744 | if (!dentry) | |
745 | goto eexit_4; | |
746 | dentry->d_op = &eventpollfs_dentry_operations; | |
747 | d_add(dentry, inode); | |
748 | file->f_vfsmnt = mntget(eventpoll_mnt); | |
749 | file->f_dentry = dentry; | |
750 | file->f_mapping = inode->i_mapping; | |
751 | ||
752 | file->f_pos = 0; | |
753 | file->f_flags = O_RDONLY; | |
754 | file->f_op = &eventpoll_fops; | |
755 | file->f_mode = FMODE_READ; | |
756 | file->f_version = 0; | |
53d2be79 | 757 | file->private_data = ep; |
1da177e4 LT |
758 | |
759 | /* Install the new setup file into the allocated fd. */ | |
760 | fd_install(fd, file); | |
761 | ||
762 | *efd = fd; | |
763 | *einode = inode; | |
764 | *efile = file; | |
765 | return 0; | |
766 | ||
767 | eexit_4: | |
768 | put_unused_fd(fd); | |
769 | eexit_3: | |
770 | iput(inode); | |
771 | eexit_2: | |
772 | put_filp(file); | |
773 | eexit_1: | |
774 | return error; | |
775 | } | |
776 | ||
777 | ||
53d2be79 | 778 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 779 | { |
53d2be79 | 780 | struct eventpoll *ep = kzalloc(sizeof(*ep), GFP_KERNEL); |
1da177e4 | 781 | |
53d2be79 | 782 | if (!ep) |
1da177e4 LT |
783 | return -ENOMEM; |
784 | ||
1da177e4 LT |
785 | rwlock_init(&ep->lock); |
786 | init_rwsem(&ep->sem); | |
787 | init_waitqueue_head(&ep->wq); | |
788 | init_waitqueue_head(&ep->poll_wait); | |
789 | INIT_LIST_HEAD(&ep->rdllist); | |
790 | ep->rbr = RB_ROOT; | |
791 | ||
53d2be79 | 792 | *pep = ep; |
1da177e4 | 793 | |
53d2be79 | 794 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_alloc() ep=%p\n", |
1da177e4 LT |
795 | current, ep)); |
796 | return 0; | |
797 | } | |
798 | ||
799 | ||
800 | static void ep_free(struct eventpoll *ep) | |
801 | { | |
802 | struct rb_node *rbp; | |
803 | struct epitem *epi; | |
804 | ||
805 | /* We need to release all tasks waiting for these file */ | |
806 | if (waitqueue_active(&ep->poll_wait)) | |
807 | ep_poll_safewake(&psw, &ep->poll_wait); | |
808 | ||
809 | /* | |
810 | * We need to lock this because we could be hit by | |
811 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
812 | * We do not need to hold "ep->sem" here because the epoll file | |
813 | * is on the way to be removed and no one has references to it | |
814 | * anymore. The only hit might come from eventpoll_release_file() but | |
144efe3e | 815 | * holding "epmutex" is sufficent here. |
1da177e4 | 816 | */ |
144efe3e | 817 | mutex_lock(&epmutex); |
1da177e4 LT |
818 | |
819 | /* | |
820 | * Walks through the whole tree by unregistering poll callbacks. | |
821 | */ | |
822 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { | |
823 | epi = rb_entry(rbp, struct epitem, rbn); | |
824 | ||
825 | ep_unregister_pollwait(ep, epi); | |
826 | } | |
827 | ||
828 | /* | |
829 | * Walks through the whole hash by freeing each "struct epitem". At this | |
830 | * point we are sure no poll callbacks will be lingering around, and also by | |
831 | * write-holding "sem" we can be sure that no file cleanup code will hit | |
832 | * us during this operation. So we can avoid the lock on "ep->lock". | |
833 | */ | |
834 | while ((rbp = rb_first(&ep->rbr)) != 0) { | |
835 | epi = rb_entry(rbp, struct epitem, rbn); | |
836 | ep_remove(ep, epi); | |
837 | } | |
838 | ||
144efe3e | 839 | mutex_unlock(&epmutex); |
1da177e4 LT |
840 | } |
841 | ||
842 | ||
843 | /* | |
844 | * Search the file inside the eventpoll hash. It add usage count to | |
845 | * the returned item, so the caller must call ep_release_epitem() | |
846 | * after finished using the "struct epitem". | |
847 | */ | |
848 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
849 | { | |
850 | int kcmp; | |
851 | unsigned long flags; | |
852 | struct rb_node *rbp; | |
853 | struct epitem *epi, *epir = NULL; | |
854 | struct epoll_filefd ffd; | |
855 | ||
b030a4dd | 856 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
857 | read_lock_irqsave(&ep->lock, flags); |
858 | for (rbp = ep->rbr.rb_node; rbp; ) { | |
859 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 860 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
861 | if (kcmp > 0) |
862 | rbp = rbp->rb_right; | |
863 | else if (kcmp < 0) | |
864 | rbp = rbp->rb_left; | |
865 | else { | |
866 | ep_use_epitem(epi); | |
867 | epir = epi; | |
868 | break; | |
869 | } | |
870 | } | |
871 | read_unlock_irqrestore(&ep->lock, flags); | |
872 | ||
873 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_find(%p) -> %p\n", | |
874 | current, file, epir)); | |
875 | ||
876 | return epir; | |
877 | } | |
878 | ||
879 | ||
880 | /* | |
881 | * Increment the usage count of the "struct epitem" making it sure | |
882 | * that the user will have a valid pointer to reference. | |
883 | */ | |
884 | static void ep_use_epitem(struct epitem *epi) | |
885 | { | |
886 | ||
887 | atomic_inc(&epi->usecnt); | |
888 | } | |
889 | ||
890 | ||
891 | /* | |
892 | * Decrement ( release ) the usage count by signaling that the user | |
893 | * has finished using the structure. It might lead to freeing the | |
894 | * structure itself if the count goes to zero. | |
895 | */ | |
896 | static void ep_release_epitem(struct epitem *epi) | |
897 | { | |
898 | ||
899 | if (atomic_dec_and_test(&epi->usecnt)) | |
b030a4dd | 900 | kmem_cache_free(epi_cache, epi); |
1da177e4 LT |
901 | } |
902 | ||
903 | ||
904 | /* | |
905 | * This is the callback that is used to add our wait queue to the | |
906 | * target file wakeup lists. | |
907 | */ | |
908 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
909 | poll_table *pt) | |
910 | { | |
b030a4dd | 911 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
912 | struct eppoll_entry *pwq; |
913 | ||
b030a4dd | 914 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, SLAB_KERNEL))) { |
1da177e4 LT |
915 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
916 | pwq->whead = whead; | |
917 | pwq->base = epi; | |
918 | add_wait_queue(whead, &pwq->wait); | |
919 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
920 | epi->nwait++; | |
921 | } else { | |
922 | /* We have to signal that an error occurred */ | |
923 | epi->nwait = -1; | |
924 | } | |
925 | } | |
926 | ||
927 | ||
928 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) | |
929 | { | |
930 | int kcmp; | |
931 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
932 | struct epitem *epic; | |
933 | ||
934 | while (*p) { | |
935 | parent = *p; | |
936 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 937 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
938 | if (kcmp > 0) |
939 | p = &parent->rb_right; | |
940 | else | |
941 | p = &parent->rb_left; | |
942 | } | |
943 | rb_link_node(&epi->rbn, parent, p); | |
944 | rb_insert_color(&epi->rbn, &ep->rbr); | |
945 | } | |
946 | ||
947 | ||
948 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, | |
949 | struct file *tfile, int fd) | |
950 | { | |
951 | int error, revents, pwake = 0; | |
952 | unsigned long flags; | |
953 | struct epitem *epi; | |
954 | struct ep_pqueue epq; | |
955 | ||
956 | error = -ENOMEM; | |
b030a4dd | 957 | if (!(epi = kmem_cache_alloc(epi_cache, SLAB_KERNEL))) |
1da177e4 LT |
958 | goto eexit_1; |
959 | ||
960 | /* Item initialization follow here ... */ | |
b030a4dd | 961 | ep_rb_initnode(&epi->rbn); |
1da177e4 LT |
962 | INIT_LIST_HEAD(&epi->rdllink); |
963 | INIT_LIST_HEAD(&epi->fllink); | |
964 | INIT_LIST_HEAD(&epi->txlink); | |
965 | INIT_LIST_HEAD(&epi->pwqlist); | |
966 | epi->ep = ep; | |
b030a4dd | 967 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 LT |
968 | epi->event = *event; |
969 | atomic_set(&epi->usecnt, 1); | |
970 | epi->nwait = 0; | |
971 | ||
972 | /* Initialize the poll table using the queue callback */ | |
973 | epq.epi = epi; | |
974 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
975 | ||
976 | /* | |
977 | * Attach the item to the poll hooks and get current event bits. | |
978 | * We can safely use the file* here because its usage count has | |
979 | * been increased by the caller of this function. | |
980 | */ | |
981 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
982 | ||
983 | /* | |
984 | * We have to check if something went wrong during the poll wait queue | |
985 | * install process. Namely an allocation for a wait queue failed due | |
986 | * high memory pressure. | |
987 | */ | |
988 | if (epi->nwait < 0) | |
989 | goto eexit_2; | |
990 | ||
991 | /* Add the current item to the list of active epoll hook for this file */ | |
992 | spin_lock(&tfile->f_ep_lock); | |
993 | list_add_tail(&epi->fllink, &tfile->f_ep_links); | |
994 | spin_unlock(&tfile->f_ep_lock); | |
995 | ||
996 | /* We have to drop the new item inside our item list to keep track of it */ | |
997 | write_lock_irqsave(&ep->lock, flags); | |
998 | ||
999 | /* Add the current item to the rb-tree */ | |
1000 | ep_rbtree_insert(ep, epi); | |
1001 | ||
1002 | /* If the file is already "ready" we drop it inside the ready list */ | |
b030a4dd | 1003 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 LT |
1004 | list_add_tail(&epi->rdllink, &ep->rdllist); |
1005 | ||
1006 | /* Notify waiting tasks that events are available */ | |
1007 | if (waitqueue_active(&ep->wq)) | |
3419b23a | 1008 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE); |
1da177e4 LT |
1009 | if (waitqueue_active(&ep->poll_wait)) |
1010 | pwake++; | |
1011 | } | |
1012 | ||
1013 | write_unlock_irqrestore(&ep->lock, flags); | |
1014 | ||
1015 | /* We have to call this outside the lock */ | |
1016 | if (pwake) | |
1017 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1018 | ||
1019 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_insert(%p, %p, %d)\n", | |
1020 | current, ep, tfile, fd)); | |
1021 | ||
1022 | return 0; | |
1023 | ||
1024 | eexit_2: | |
1025 | ep_unregister_pollwait(ep, epi); | |
1026 | ||
1027 | /* | |
1028 | * We need to do this because an event could have been arrived on some | |
1029 | * allocated wait queue. | |
1030 | */ | |
1031 | write_lock_irqsave(&ep->lock, flags); | |
b030a4dd PE |
1032 | if (ep_is_linked(&epi->rdllink)) |
1033 | ep_list_del(&epi->rdllink); | |
1da177e4 LT |
1034 | write_unlock_irqrestore(&ep->lock, flags); |
1035 | ||
b030a4dd | 1036 | kmem_cache_free(epi_cache, epi); |
1da177e4 LT |
1037 | eexit_1: |
1038 | return error; | |
1039 | } | |
1040 | ||
1041 | ||
1042 | /* | |
1043 | * Modify the interest event mask by dropping an event if the new mask | |
1044 | * has a match in the current file status. | |
1045 | */ | |
1046 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1047 | { | |
1048 | int pwake = 0; | |
1049 | unsigned int revents; | |
1050 | unsigned long flags; | |
1051 | ||
1052 | /* | |
1053 | * Set the new event interest mask before calling f_op->poll(), otherwise | |
1054 | * a potential race might occur. In fact if we do this operation inside | |
1055 | * the lock, an event might happen between the f_op->poll() call and the | |
1056 | * new event set registering. | |
1057 | */ | |
1058 | epi->event.events = event->events; | |
1059 | ||
1060 | /* | |
1061 | * Get current event bits. We can safely use the file* here because | |
1062 | * its usage count has been increased by the caller of this function. | |
1063 | */ | |
1064 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
1065 | ||
1066 | write_lock_irqsave(&ep->lock, flags); | |
1067 | ||
1068 | /* Copy the data member from inside the lock */ | |
1069 | epi->event.data = event->data; | |
1070 | ||
1071 | /* | |
1072 | * If the item is not linked to the hash it means that it's on its | |
1073 | * way toward the removal. Do nothing in this case. | |
1074 | */ | |
b030a4dd | 1075 | if (ep_rb_linked(&epi->rbn)) { |
1da177e4 LT |
1076 | /* |
1077 | * If the item is "hot" and it is not registered inside the ready | |
1078 | * list, push it inside. If the item is not "hot" and it is currently | |
1079 | * registered inside the ready list, unlink it. | |
1080 | */ | |
1081 | if (revents & event->events) { | |
b030a4dd | 1082 | if (!ep_is_linked(&epi->rdllink)) { |
1da177e4 LT |
1083 | list_add_tail(&epi->rdllink, &ep->rdllist); |
1084 | ||
1085 | /* Notify waiting tasks that events are available */ | |
1086 | if (waitqueue_active(&ep->wq)) | |
3419b23a DL |
1087 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | |
1088 | TASK_INTERRUPTIBLE); | |
1da177e4 LT |
1089 | if (waitqueue_active(&ep->poll_wait)) |
1090 | pwake++; | |
1091 | } | |
1092 | } | |
1093 | } | |
1094 | ||
1095 | write_unlock_irqrestore(&ep->lock, flags); | |
1096 | ||
1097 | /* We have to call this outside the lock */ | |
1098 | if (pwake) | |
1099 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1100 | ||
1101 | return 0; | |
1102 | } | |
1103 | ||
1104 | ||
1105 | /* | |
1106 | * This function unregister poll callbacks from the associated file descriptor. | |
1107 | * Since this must be called without holding "ep->lock" the atomic exchange trick | |
1108 | * will protect us from multiple unregister. | |
1109 | */ | |
1110 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) | |
1111 | { | |
1112 | int nwait; | |
1113 | struct list_head *lsthead = &epi->pwqlist; | |
1114 | struct eppoll_entry *pwq; | |
1115 | ||
1116 | /* This is called without locks, so we need the atomic exchange */ | |
1117 | nwait = xchg(&epi->nwait, 0); | |
1118 | ||
1119 | if (nwait) { | |
1120 | while (!list_empty(lsthead)) { | |
1121 | pwq = list_entry(lsthead->next, struct eppoll_entry, llink); | |
1122 | ||
b030a4dd | 1123 | ep_list_del(&pwq->llink); |
1da177e4 | 1124 | remove_wait_queue(pwq->whead, &pwq->wait); |
b030a4dd | 1125 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 LT |
1126 | } |
1127 | } | |
1128 | } | |
1129 | ||
1130 | ||
1131 | /* | |
1132 | * Unlink the "struct epitem" from all places it might have been hooked up. | |
1133 | * This function must be called with write IRQ lock on "ep->lock". | |
1134 | */ | |
1135 | static int ep_unlink(struct eventpoll *ep, struct epitem *epi) | |
1136 | { | |
1137 | int error; | |
1138 | ||
1139 | /* | |
1140 | * It can happen that this one is called for an item already unlinked. | |
1141 | * The check protect us from doing a double unlink ( crash ). | |
1142 | */ | |
1143 | error = -ENOENT; | |
b030a4dd | 1144 | if (!ep_rb_linked(&epi->rbn)) |
1da177e4 LT |
1145 | goto eexit_1; |
1146 | ||
1147 | /* | |
1148 | * Clear the event mask for the unlinked item. This will avoid item | |
1149 | * notifications to be sent after the unlink operation from inside | |
1150 | * the kernel->userspace event transfer loop. | |
1151 | */ | |
1152 | epi->event.events = 0; | |
1153 | ||
1154 | /* | |
1155 | * At this point is safe to do the job, unlink the item from our rb-tree. | |
1156 | * This operation togheter with the above check closes the door to | |
1157 | * double unlinks. | |
1158 | */ | |
b030a4dd | 1159 | ep_rb_erase(&epi->rbn, &ep->rbr); |
1da177e4 LT |
1160 | |
1161 | /* | |
1162 | * If the item we are going to remove is inside the ready file descriptors | |
1163 | * we want to remove it from this list to avoid stale events. | |
1164 | */ | |
b030a4dd PE |
1165 | if (ep_is_linked(&epi->rdllink)) |
1166 | ep_list_del(&epi->rdllink); | |
1da177e4 LT |
1167 | |
1168 | error = 0; | |
1169 | eexit_1: | |
1170 | ||
1171 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_unlink(%p, %p) = %d\n", | |
45f17e0c | 1172 | current, ep, epi->ffd.file, error)); |
1da177e4 LT |
1173 | |
1174 | return error; | |
1175 | } | |
1176 | ||
1177 | ||
1178 | /* | |
1179 | * Removes a "struct epitem" from the eventpoll hash and deallocates | |
1180 | * all the associated resources. | |
1181 | */ | |
1182 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
1183 | { | |
1184 | int error; | |
1185 | unsigned long flags; | |
1186 | struct file *file = epi->ffd.file; | |
1187 | ||
1188 | /* | |
1189 | * Removes poll wait queue hooks. We _have_ to do this without holding | |
1190 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
1191 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
1192 | * queue head lock when unregistering the wait queue. The wakeup callback | |
1193 | * will run by holding the wait queue head lock and will call our callback | |
1194 | * that will try to get "ep->lock". | |
1195 | */ | |
1196 | ep_unregister_pollwait(ep, epi); | |
1197 | ||
1198 | /* Remove the current item from the list of epoll hooks */ | |
1199 | spin_lock(&file->f_ep_lock); | |
b030a4dd PE |
1200 | if (ep_is_linked(&epi->fllink)) |
1201 | ep_list_del(&epi->fllink); | |
1da177e4 LT |
1202 | spin_unlock(&file->f_ep_lock); |
1203 | ||
1204 | /* We need to acquire the write IRQ lock before calling ep_unlink() */ | |
1205 | write_lock_irqsave(&ep->lock, flags); | |
1206 | ||
1207 | /* Really unlink the item from the hash */ | |
1208 | error = ep_unlink(ep, epi); | |
1209 | ||
1210 | write_unlock_irqrestore(&ep->lock, flags); | |
1211 | ||
1212 | if (error) | |
1213 | goto eexit_1; | |
1214 | ||
1215 | /* At this point it is safe to free the eventpoll item */ | |
1216 | ep_release_epitem(epi); | |
1217 | ||
1218 | error = 0; | |
1219 | eexit_1: | |
1220 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_remove(%p, %p) = %d\n", | |
1221 | current, ep, file, error)); | |
1222 | ||
1223 | return error; | |
1224 | } | |
1225 | ||
1226 | ||
1227 | /* | |
1228 | * This is the callback that is passed to the wait queue wakeup | |
1229 | * machanism. It is called by the stored file descriptors when they | |
1230 | * have events to report. | |
1231 | */ | |
1232 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
1233 | { | |
1234 | int pwake = 0; | |
1235 | unsigned long flags; | |
b030a4dd | 1236 | struct epitem *epi = ep_item_from_wait(wait); |
1da177e4 LT |
1237 | struct eventpoll *ep = epi->ep; |
1238 | ||
1239 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: poll_callback(%p) epi=%p ep=%p\n", | |
45f17e0c | 1240 | current, epi->ffd.file, epi, ep)); |
1da177e4 LT |
1241 | |
1242 | write_lock_irqsave(&ep->lock, flags); | |
1243 | ||
1244 | /* | |
1245 | * If the event mask does not contain any poll(2) event, we consider the | |
1246 | * descriptor to be disabled. This condition is likely the effect of the | |
1247 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1248 | * until the next EPOLL_CTL_MOD will be issued. | |
1249 | */ | |
1250 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
1251 | goto is_disabled; | |
1252 | ||
1253 | /* If this file is already in the ready list we exit soon */ | |
b030a4dd | 1254 | if (ep_is_linked(&epi->rdllink)) |
1da177e4 LT |
1255 | goto is_linked; |
1256 | ||
1257 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1258 | ||
1259 | is_linked: | |
1260 | /* | |
1261 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1262 | * wait list. | |
1263 | */ | |
1264 | if (waitqueue_active(&ep->wq)) | |
3419b23a DL |
1265 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | |
1266 | TASK_INTERRUPTIBLE); | |
1da177e4 LT |
1267 | if (waitqueue_active(&ep->poll_wait)) |
1268 | pwake++; | |
1269 | ||
1270 | is_disabled: | |
1271 | write_unlock_irqrestore(&ep->lock, flags); | |
1272 | ||
1273 | /* We have to call this outside the lock */ | |
1274 | if (pwake) | |
1275 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1276 | ||
1277 | return 1; | |
1278 | } | |
1279 | ||
1280 | ||
1281 | static int ep_eventpoll_close(struct inode *inode, struct file *file) | |
1282 | { | |
1283 | struct eventpoll *ep = file->private_data; | |
1284 | ||
1285 | if (ep) { | |
1286 | ep_free(ep); | |
1287 | kfree(ep); | |
1288 | } | |
1289 | ||
1290 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: close() ep=%p\n", current, ep)); | |
1291 | return 0; | |
1292 | } | |
1293 | ||
1294 | ||
1295 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) | |
1296 | { | |
1297 | unsigned int pollflags = 0; | |
1298 | unsigned long flags; | |
1299 | struct eventpoll *ep = file->private_data; | |
1300 | ||
1301 | /* Insert inside our poll wait queue */ | |
1302 | poll_wait(file, &ep->poll_wait, wait); | |
1303 | ||
1304 | /* Check our condition */ | |
1305 | read_lock_irqsave(&ep->lock, flags); | |
1306 | if (!list_empty(&ep->rdllist)) | |
1307 | pollflags = POLLIN | POLLRDNORM; | |
1308 | read_unlock_irqrestore(&ep->lock, flags); | |
1309 | ||
1310 | return pollflags; | |
1311 | } | |
1312 | ||
1313 | ||
1314 | /* | |
1315 | * Since we have to release the lock during the __copy_to_user() operation and | |
1316 | * during the f_op->poll() call, we try to collect the maximum number of items | |
1317 | * by reducing the irqlock/irqunlock switching rate. | |
1318 | */ | |
1319 | static int ep_collect_ready_items(struct eventpoll *ep, struct list_head *txlist, int maxevents) | |
1320 | { | |
1321 | int nepi; | |
1322 | unsigned long flags; | |
1323 | struct list_head *lsthead = &ep->rdllist, *lnk; | |
1324 | struct epitem *epi; | |
1325 | ||
1326 | write_lock_irqsave(&ep->lock, flags); | |
1327 | ||
1328 | for (nepi = 0, lnk = lsthead->next; lnk != lsthead && nepi < maxevents;) { | |
1329 | epi = list_entry(lnk, struct epitem, rdllink); | |
1330 | ||
1331 | lnk = lnk->next; | |
1332 | ||
1333 | /* If this file is already in the ready list we exit soon */ | |
b030a4dd | 1334 | if (!ep_is_linked(&epi->txlink)) { |
1da177e4 LT |
1335 | /* |
1336 | * This is initialized in this way so that the default | |
1337 | * behaviour of the reinjecting code will be to push back | |
1338 | * the item inside the ready list. | |
1339 | */ | |
1340 | epi->revents = epi->event.events; | |
1341 | ||
1342 | /* Link the ready item into the transfer list */ | |
1343 | list_add(&epi->txlink, txlist); | |
1344 | nepi++; | |
1345 | ||
1346 | /* | |
1347 | * Unlink the item from the ready list. | |
1348 | */ | |
b030a4dd | 1349 | ep_list_del(&epi->rdllink); |
1da177e4 LT |
1350 | } |
1351 | } | |
1352 | ||
1353 | write_unlock_irqrestore(&ep->lock, flags); | |
1354 | ||
1355 | return nepi; | |
1356 | } | |
1357 | ||
1358 | ||
1359 | /* | |
1360 | * This function is called without holding the "ep->lock" since the call to | |
1361 | * __copy_to_user() might sleep, and also f_op->poll() might reenable the IRQ | |
1362 | * because of the way poll() is traditionally implemented in Linux. | |
1363 | */ | |
1364 | static int ep_send_events(struct eventpoll *ep, struct list_head *txlist, | |
1365 | struct epoll_event __user *events) | |
1366 | { | |
1367 | int eventcnt = 0; | |
1368 | unsigned int revents; | |
1369 | struct list_head *lnk; | |
1370 | struct epitem *epi; | |
1371 | ||
1372 | /* | |
1373 | * We can loop without lock because this is a task private list. | |
1374 | * The test done during the collection loop will guarantee us that | |
1375 | * another task will not try to collect this file. Also, items | |
1376 | * cannot vanish during the loop because we are holding "sem". | |
1377 | */ | |
1378 | list_for_each(lnk, txlist) { | |
1379 | epi = list_entry(lnk, struct epitem, txlink); | |
1380 | ||
1381 | /* | |
1382 | * Get the ready file event set. We can safely use the file | |
1383 | * because we are holding the "sem" in read and this will | |
1384 | * guarantee that both the file and the item will not vanish. | |
1385 | */ | |
1386 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
1387 | ||
1388 | /* | |
1389 | * Set the return event set for the current file descriptor. | |
1390 | * Note that only the task task was successfully able to link | |
1391 | * the item to its "txlist" will write this field. | |
1392 | */ | |
1393 | epi->revents = revents & epi->event.events; | |
1394 | ||
1395 | if (epi->revents) { | |
1396 | if (__put_user(epi->revents, | |
1397 | &events[eventcnt].events) || | |
1398 | __put_user(epi->event.data, | |
1399 | &events[eventcnt].data)) | |
1400 | return -EFAULT; | |
1401 | if (epi->event.events & EPOLLONESHOT) | |
1402 | epi->event.events &= EP_PRIVATE_BITS; | |
1403 | eventcnt++; | |
1404 | } | |
1405 | } | |
1406 | return eventcnt; | |
1407 | } | |
1408 | ||
1409 | ||
1410 | /* | |
1411 | * Walk through the transfer list we collected with ep_collect_ready_items() | |
1412 | * and, if 1) the item is still "alive" 2) its event set is not empty 3) it's | |
1413 | * not already linked, links it to the ready list. Same as above, we are holding | |
1414 | * "sem" so items cannot vanish underneath our nose. | |
1415 | */ | |
1416 | static void ep_reinject_items(struct eventpoll *ep, struct list_head *txlist) | |
1417 | { | |
1418 | int ricnt = 0, pwake = 0; | |
1419 | unsigned long flags; | |
1420 | struct epitem *epi; | |
1421 | ||
1422 | write_lock_irqsave(&ep->lock, flags); | |
1423 | ||
1424 | while (!list_empty(txlist)) { | |
1425 | epi = list_entry(txlist->next, struct epitem, txlink); | |
1426 | ||
1427 | /* Unlink the current item from the transfer list */ | |
b030a4dd | 1428 | ep_list_del(&epi->txlink); |
1da177e4 LT |
1429 | |
1430 | /* | |
1431 | * If the item is no more linked to the interest set, we don't | |
1432 | * have to push it inside the ready list because the following | |
1433 | * ep_release_epitem() is going to drop it. Also, if the current | |
1434 | * item is set to have an Edge Triggered behaviour, we don't have | |
1435 | * to push it back either. | |
1436 | */ | |
b030a4dd PE |
1437 | if (ep_rb_linked(&epi->rbn) && !(epi->event.events & EPOLLET) && |
1438 | (epi->revents & epi->event.events) && !ep_is_linked(&epi->rdllink)) { | |
1da177e4 LT |
1439 | list_add_tail(&epi->rdllink, &ep->rdllist); |
1440 | ricnt++; | |
1441 | } | |
1442 | } | |
1443 | ||
1444 | if (ricnt) { | |
1445 | /* | |
1446 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1447 | * wait list. | |
1448 | */ | |
1449 | if (waitqueue_active(&ep->wq)) | |
3419b23a DL |
1450 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | |
1451 | TASK_INTERRUPTIBLE); | |
1da177e4 LT |
1452 | if (waitqueue_active(&ep->poll_wait)) |
1453 | pwake++; | |
1454 | } | |
1455 | ||
1456 | write_unlock_irqrestore(&ep->lock, flags); | |
1457 | ||
1458 | /* We have to call this outside the lock */ | |
1459 | if (pwake) | |
1460 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1461 | } | |
1462 | ||
1463 | ||
1464 | /* | |
1465 | * Perform the transfer of events to user space. | |
1466 | */ | |
1467 | static int ep_events_transfer(struct eventpoll *ep, | |
1468 | struct epoll_event __user *events, int maxevents) | |
1469 | { | |
1470 | int eventcnt = 0; | |
1471 | struct list_head txlist; | |
1472 | ||
1473 | INIT_LIST_HEAD(&txlist); | |
1474 | ||
1475 | /* | |
1476 | * We need to lock this because we could be hit by | |
1477 | * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL). | |
1478 | */ | |
1479 | down_read(&ep->sem); | |
1480 | ||
1481 | /* Collect/extract ready items */ | |
1482 | if (ep_collect_ready_items(ep, &txlist, maxevents) > 0) { | |
1483 | /* Build result set in userspace */ | |
1484 | eventcnt = ep_send_events(ep, &txlist, events); | |
1485 | ||
1486 | /* Reinject ready items into the ready list */ | |
1487 | ep_reinject_items(ep, &txlist); | |
1488 | } | |
1489 | ||
1490 | up_read(&ep->sem); | |
1491 | ||
1492 | return eventcnt; | |
1493 | } | |
1494 | ||
1495 | ||
1496 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, | |
1497 | int maxevents, long timeout) | |
1498 | { | |
1499 | int res, eavail; | |
1500 | unsigned long flags; | |
1501 | long jtimeout; | |
1502 | wait_queue_t wait; | |
1503 | ||
1504 | /* | |
1505 | * Calculate the timeout by checking for the "infinite" value ( -1 ) | |
1506 | * and the overflow condition. The passed timeout is in milliseconds, | |
1507 | * that why (t * HZ) / 1000. | |
1508 | */ | |
e3306dd5 DL |
1509 | jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ? |
1510 | MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000; | |
1da177e4 LT |
1511 | |
1512 | retry: | |
1513 | write_lock_irqsave(&ep->lock, flags); | |
1514 | ||
1515 | res = 0; | |
1516 | if (list_empty(&ep->rdllist)) { | |
1517 | /* | |
1518 | * We don't have any available event to return to the caller. | |
1519 | * We need to sleep here, and we will be wake up by | |
1520 | * ep_poll_callback() when events will become available. | |
1521 | */ | |
1522 | init_waitqueue_entry(&wait, current); | |
3419b23a | 1523 | __add_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1524 | |
1525 | for (;;) { | |
1526 | /* | |
1527 | * We don't want to sleep if the ep_poll_callback() sends us | |
1528 | * a wakeup in between. That's why we set the task state | |
1529 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1530 | */ | |
1531 | set_current_state(TASK_INTERRUPTIBLE); | |
1532 | if (!list_empty(&ep->rdllist) || !jtimeout) | |
1533 | break; | |
1534 | if (signal_pending(current)) { | |
1535 | res = -EINTR; | |
1536 | break; | |
1537 | } | |
1538 | ||
1539 | write_unlock_irqrestore(&ep->lock, flags); | |
1540 | jtimeout = schedule_timeout(jtimeout); | |
1541 | write_lock_irqsave(&ep->lock, flags); | |
1542 | } | |
3419b23a | 1543 | __remove_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1544 | |
1545 | set_current_state(TASK_RUNNING); | |
1546 | } | |
1547 | ||
1548 | /* Is it worth to try to dig for events ? */ | |
1549 | eavail = !list_empty(&ep->rdllist); | |
1550 | ||
1551 | write_unlock_irqrestore(&ep->lock, flags); | |
1552 | ||
1553 | /* | |
1554 | * Try to transfer events to user space. In case we get 0 events and | |
1555 | * there's still timeout left over, we go trying again in search of | |
1556 | * more luck. | |
1557 | */ | |
1558 | if (!res && eavail && | |
1559 | !(res = ep_events_transfer(ep, events, maxevents)) && jtimeout) | |
1560 | goto retry; | |
1561 | ||
1562 | return res; | |
1563 | } | |
1564 | ||
1565 | ||
1566 | static int eventpollfs_delete_dentry(struct dentry *dentry) | |
1567 | { | |
1568 | ||
1569 | return 1; | |
1570 | } | |
1571 | ||
1572 | ||
1573 | static struct inode *ep_eventpoll_inode(void) | |
1574 | { | |
1575 | int error = -ENOMEM; | |
1576 | struct inode *inode = new_inode(eventpoll_mnt->mnt_sb); | |
1577 | ||
1578 | if (!inode) | |
1579 | goto eexit_1; | |
1580 | ||
1581 | inode->i_fop = &eventpoll_fops; | |
1582 | ||
1583 | /* | |
1584 | * Mark the inode dirty from the very beginning, | |
1585 | * that way it will never be moved to the dirty | |
1586 | * list because mark_inode_dirty() will think | |
1587 | * that it already _is_ on the dirty list. | |
1588 | */ | |
1589 | inode->i_state = I_DIRTY; | |
1590 | inode->i_mode = S_IRUSR | S_IWUSR; | |
1591 | inode->i_uid = current->fsuid; | |
1592 | inode->i_gid = current->fsgid; | |
1593 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
1da177e4 LT |
1594 | return inode; |
1595 | ||
1596 | eexit_1: | |
1597 | return ERR_PTR(error); | |
1598 | } | |
1599 | ||
1600 | ||
454e2398 | 1601 | static int |
1da177e4 | 1602 | eventpollfs_get_sb(struct file_system_type *fs_type, int flags, |
454e2398 | 1603 | const char *dev_name, void *data, struct vfsmount *mnt) |
1da177e4 | 1604 | { |
454e2398 DH |
1605 | return get_sb_pseudo(fs_type, "eventpoll:", NULL, EVENTPOLLFS_MAGIC, |
1606 | mnt); | |
1da177e4 LT |
1607 | } |
1608 | ||
1609 | ||
1610 | static int __init eventpoll_init(void) | |
1611 | { | |
1612 | int error; | |
1613 | ||
144efe3e | 1614 | mutex_init(&epmutex); |
1da177e4 LT |
1615 | |
1616 | /* Initialize the structure used to perform safe poll wait head wake ups */ | |
1617 | ep_poll_safewake_init(&psw); | |
1618 | ||
1619 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1620 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
1621 | 0, SLAB_HWCACHE_ALIGN|EPI_SLAB_DEBUG|SLAB_PANIC, | |
1622 | NULL, NULL); | |
1623 | ||
1624 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1625 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
1626 | sizeof(struct eppoll_entry), 0, | |
1627 | EPI_SLAB_DEBUG|SLAB_PANIC, NULL, NULL); | |
1628 | ||
1629 | /* | |
1630 | * Register the virtual file system that will be the source of inodes | |
1631 | * for the eventpoll files | |
1632 | */ | |
1633 | error = register_filesystem(&eventpoll_fs_type); | |
1634 | if (error) | |
1635 | goto epanic; | |
1636 | ||
1637 | /* Mount the above commented virtual file system */ | |
1638 | eventpoll_mnt = kern_mount(&eventpoll_fs_type); | |
1639 | error = PTR_ERR(eventpoll_mnt); | |
1640 | if (IS_ERR(eventpoll_mnt)) | |
1641 | goto epanic; | |
1642 | ||
1643 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: successfully initialized.\n", | |
1644 | current)); | |
1645 | return 0; | |
1646 | ||
1647 | epanic: | |
1648 | panic("eventpoll_init() failed\n"); | |
1649 | } | |
1650 | ||
1651 | ||
1652 | static void __exit eventpoll_exit(void) | |
1653 | { | |
1654 | /* Undo all operations done inside eventpoll_init() */ | |
1655 | unregister_filesystem(&eventpoll_fs_type); | |
1656 | mntput(eventpoll_mnt); | |
1657 | kmem_cache_destroy(pwq_cache); | |
1658 | kmem_cache_destroy(epi_cache); | |
1659 | } | |
1660 | ||
1661 | module_init(eventpoll_init); | |
1662 | module_exit(eventpoll_exit); | |
1663 | ||
1664 | MODULE_LICENSE("GPL"); |