]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
c7ea7630 DL |
2 | * fs/eventpoll.c (Efficent event polling implementation) |
3 | * Copyright (C) 2001,...,2007 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 | ||
1da177e4 LT |
14 | #include <linux/init.h> |
15 | #include <linux/kernel.h> | |
16 | #include <linux/sched.h> | |
17 | #include <linux/fs.h> | |
18 | #include <linux/file.h> | |
19 | #include <linux/signal.h> | |
20 | #include <linux/errno.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/poll.h> | |
1da177e4 LT |
24 | #include <linux/string.h> |
25 | #include <linux/list.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/spinlock.h> | |
28 | #include <linux/syscalls.h> | |
1da177e4 LT |
29 | #include <linux/rbtree.h> |
30 | #include <linux/wait.h> | |
31 | #include <linux/eventpoll.h> | |
32 | #include <linux/mount.h> | |
33 | #include <linux/bitops.h> | |
144efe3e | 34 | #include <linux/mutex.h> |
da66f7cb | 35 | #include <linux/anon_inodes.h> |
1da177e4 LT |
36 | #include <asm/uaccess.h> |
37 | #include <asm/system.h> | |
38 | #include <asm/io.h> | |
39 | #include <asm/mman.h> | |
40 | #include <asm/atomic.h> | |
1da177e4 | 41 | |
1da177e4 LT |
42 | /* |
43 | * LOCKING: | |
44 | * There are three level of locking required by epoll : | |
45 | * | |
144efe3e | 46 | * 1) epmutex (mutex) |
c7ea7630 DL |
47 | * 2) ep->mtx (mutex) |
48 | * 3) ep->lock (spinlock) | |
1da177e4 LT |
49 | * |
50 | * The acquire order is the one listed above, from 1 to 3. | |
51 | * We need a spinlock (ep->lock) because we manipulate objects | |
52 | * from inside the poll callback, that might be triggered from | |
53 | * a wake_up() that in turn might be called from IRQ context. | |
54 | * So we can't sleep inside the poll callback and hence we need | |
55 | * a spinlock. During the event transfer loop (from kernel to | |
56 | * user space) we could end up sleeping due a copy_to_user(), so | |
57 | * we need a lock that will allow us to sleep. This lock is a | |
d47de16c DL |
58 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
59 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). | |
60 | * Then we also need a global mutex to serialize eventpoll_release_file() | |
61 | * and ep_free(). | |
62 | * This mutex is acquired by ep_free() during the epoll file | |
1da177e4 LT |
63 | * cleanup path and it is also acquired by eventpoll_release_file() |
64 | * if a file has been pushed inside an epoll set and it is then | |
65 | * close()d without a previous call toepoll_ctl(EPOLL_CTL_DEL). | |
d47de16c DL |
66 | * It is possible to drop the "ep->mtx" and to use the global |
67 | * mutex "epmutex" (together with "ep->lock") to have it working, | |
68 | * but having "ep->mtx" will make the interface more scalable. | |
144efe3e | 69 | * Events that require holding "epmutex" are very rare, while for |
d47de16c DL |
70 | * normal operations the epoll private "ep->mtx" will guarantee |
71 | * a better scalability. | |
1da177e4 LT |
72 | */ |
73 | ||
1da177e4 LT |
74 | #define DEBUG_EPOLL 0 |
75 | ||
76 | #if DEBUG_EPOLL > 0 | |
77 | #define DPRINTK(x) printk x | |
78 | #define DNPRINTK(n, x) do { if ((n) <= DEBUG_EPOLL) printk x; } while (0) | |
79 | #else /* #if DEBUG_EPOLL > 0 */ | |
80 | #define DPRINTK(x) (void) 0 | |
81 | #define DNPRINTK(n, x) (void) 0 | |
82 | #endif /* #if DEBUG_EPOLL > 0 */ | |
83 | ||
84 | #define DEBUG_EPI 0 | |
85 | ||
86 | #if DEBUG_EPI != 0 | |
87 | #define EPI_SLAB_DEBUG (SLAB_DEBUG_FREE | SLAB_RED_ZONE /* | SLAB_POISON */) | |
88 | #else /* #if DEBUG_EPI != 0 */ | |
89 | #define EPI_SLAB_DEBUG 0 | |
90 | #endif /* #if DEBUG_EPI != 0 */ | |
91 | ||
92 | /* Epoll private bits inside the event mask */ | |
93 | #define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET) | |
94 | ||
95 | /* Maximum number of poll wake up nests we are allowing */ | |
96 | #define EP_MAX_POLLWAKE_NESTS 4 | |
97 | ||
e3306dd5 DL |
98 | /* Maximum msec timeout value storeable in a long int */ |
99 | #define EP_MAX_MSTIMEO min(1000ULL * MAX_SCHEDULE_TIMEOUT / HZ, (LONG_MAX - 999ULL) / HZ) | |
100 | ||
b611967d DL |
101 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
102 | ||
d47de16c DL |
103 | #define EP_UNACTIVE_PTR ((void *) -1L) |
104 | ||
1da177e4 LT |
105 | struct epoll_filefd { |
106 | struct file *file; | |
107 | int fd; | |
108 | }; | |
109 | ||
110 | /* | |
111 | * Node that is linked into the "wake_task_list" member of the "struct poll_safewake". | |
112 | * It is used to keep track on all tasks that are currently inside the wake_up() code | |
113 | * to 1) short-circuit the one coming from the same task and same wait queue head | |
d47de16c | 114 | * (loop) 2) allow a maximum number of epoll descriptors inclusion nesting |
1da177e4 LT |
115 | * 3) let go the ones coming from other tasks. |
116 | */ | |
117 | struct wake_task_node { | |
118 | struct list_head llink; | |
36c8b586 | 119 | struct task_struct *task; |
1da177e4 LT |
120 | wait_queue_head_t *wq; |
121 | }; | |
122 | ||
123 | /* | |
124 | * This is used to implement the safe poll wake up avoiding to reenter | |
125 | * the poll callback from inside wake_up(). | |
126 | */ | |
127 | struct poll_safewake { | |
128 | struct list_head wake_task_list; | |
129 | spinlock_t lock; | |
130 | }; | |
131 | ||
d47de16c DL |
132 | /* |
133 | * Each file descriptor added to the eventpoll interface will | |
134 | * have an entry of this type linked to the "rbr" RB tree. | |
135 | */ | |
136 | struct epitem { | |
137 | /* RB-Tree node used to link this structure to the eventpoll rb-tree */ | |
138 | struct rb_node rbn; | |
139 | ||
140 | /* List header used to link this structure to the eventpoll ready list */ | |
141 | struct list_head rdllink; | |
142 | ||
c7ea7630 DL |
143 | /* |
144 | * Works together "struct eventpoll"->ovflist in keeping the | |
145 | * single linked chain of items. | |
146 | */ | |
147 | struct epitem *next; | |
148 | ||
d47de16c DL |
149 | /* The file descriptor information this item refers to */ |
150 | struct epoll_filefd ffd; | |
151 | ||
152 | /* Number of active wait queue attached to poll operations */ | |
153 | int nwait; | |
154 | ||
155 | /* List containing poll wait queues */ | |
156 | struct list_head pwqlist; | |
157 | ||
158 | /* The "container" of this item */ | |
159 | struct eventpoll *ep; | |
160 | ||
d47de16c DL |
161 | /* List header used to link this item to the "struct file" items list */ |
162 | struct list_head fllink; | |
163 | ||
c7ea7630 DL |
164 | /* The structure that describe the interested events and the source fd */ |
165 | struct epoll_event event; | |
d47de16c DL |
166 | }; |
167 | ||
1da177e4 LT |
168 | /* |
169 | * This structure is stored inside the "private_data" member of the file | |
170 | * structure and rapresent the main data sructure for the eventpoll | |
171 | * interface. | |
172 | */ | |
173 | struct eventpoll { | |
174 | /* Protect the this structure access */ | |
c7ea7630 | 175 | spinlock_t lock; |
1da177e4 LT |
176 | |
177 | /* | |
d47de16c DL |
178 | * This mutex is used to ensure that files are not removed |
179 | * while epoll is using them. This is held during the event | |
180 | * collection loop, the file cleanup path, the epoll file exit | |
181 | * code and the ctl operations. | |
1da177e4 | 182 | */ |
d47de16c | 183 | struct mutex mtx; |
1da177e4 LT |
184 | |
185 | /* Wait queue used by sys_epoll_wait() */ | |
186 | wait_queue_head_t wq; | |
187 | ||
188 | /* Wait queue used by file->poll() */ | |
189 | wait_queue_head_t poll_wait; | |
190 | ||
191 | /* List of ready file descriptors */ | |
192 | struct list_head rdllist; | |
193 | ||
194 | /* RB-Tree root used to store monitored fd structs */ | |
195 | struct rb_root rbr; | |
d47de16c DL |
196 | |
197 | /* | |
198 | * This is a single linked list that chains all the "struct epitem" that | |
199 | * happened while transfering ready events to userspace w/out | |
200 | * holding ->lock. | |
201 | */ | |
202 | struct epitem *ovflist; | |
1da177e4 LT |
203 | }; |
204 | ||
205 | /* Wait structure used by the poll hooks */ | |
206 | struct eppoll_entry { | |
207 | /* List header used to link this structure to the "struct epitem" */ | |
208 | struct list_head llink; | |
209 | ||
210 | /* The "base" pointer is set to the container "struct epitem" */ | |
211 | void *base; | |
212 | ||
213 | /* | |
214 | * Wait queue item that will be linked to the target file wait | |
215 | * queue head. | |
216 | */ | |
217 | wait_queue_t wait; | |
218 | ||
219 | /* The wait queue head that linked the "wait" wait queue item */ | |
220 | wait_queue_head_t *whead; | |
221 | }; | |
222 | ||
1da177e4 LT |
223 | /* Wrapper struct used by poll queueing */ |
224 | struct ep_pqueue { | |
225 | poll_table pt; | |
226 | struct epitem *epi; | |
227 | }; | |
228 | ||
1da177e4 | 229 | /* |
d47de16c | 230 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 231 | */ |
144efe3e | 232 | static struct mutex epmutex; |
1da177e4 LT |
233 | |
234 | /* Safe wake up implementation */ | |
235 | static struct poll_safewake psw; | |
236 | ||
237 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 238 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
239 | |
240 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 241 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 242 | |
b030a4dd PE |
243 | |
244 | /* Setup the structure that is used as key for the rb-tree */ | |
245 | static inline void ep_set_ffd(struct epoll_filefd *ffd, | |
246 | struct file *file, int fd) | |
247 | { | |
248 | ffd->file = file; | |
249 | ffd->fd = fd; | |
250 | } | |
251 | ||
252 | /* Compare rb-tree keys */ | |
253 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, | |
254 | struct epoll_filefd *p2) | |
255 | { | |
256 | return (p1->file > p2->file ? +1: | |
257 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
258 | } | |
259 | ||
260 | /* Special initialization for the rb-tree node to detect linkage */ | |
261 | static inline void ep_rb_initnode(struct rb_node *n) | |
262 | { | |
c569882b | 263 | rb_set_parent(n, n); |
b030a4dd PE |
264 | } |
265 | ||
266 | /* Removes a node from the rb-tree and marks it for a fast is-linked check */ | |
267 | static inline void ep_rb_erase(struct rb_node *n, struct rb_root *r) | |
268 | { | |
269 | rb_erase(n, r); | |
c569882b | 270 | rb_set_parent(n, n); |
b030a4dd PE |
271 | } |
272 | ||
273 | /* Fast check to verify that the item is linked to the main rb-tree */ | |
274 | static inline int ep_rb_linked(struct rb_node *n) | |
275 | { | |
c569882b | 276 | return rb_parent(n) != n; |
b030a4dd PE |
277 | } |
278 | ||
b030a4dd PE |
279 | /* Tells us if the item is currently linked */ |
280 | static inline int ep_is_linked(struct list_head *p) | |
281 | { | |
282 | return !list_empty(p); | |
283 | } | |
284 | ||
285 | /* Get the "struct epitem" from a wait queue pointer */ | |
286 | static inline struct epitem * ep_item_from_wait(wait_queue_t *p) | |
287 | { | |
288 | return container_of(p, struct eppoll_entry, wait)->base; | |
289 | } | |
290 | ||
291 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
292 | static inline struct epitem * ep_item_from_epqueue(poll_table *p) | |
293 | { | |
294 | return container_of(p, struct ep_pqueue, pt)->epi; | |
295 | } | |
296 | ||
297 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 298 | static inline int ep_op_has_event(int op) |
b030a4dd PE |
299 | { |
300 | return op != EPOLL_CTL_DEL; | |
301 | } | |
302 | ||
1da177e4 LT |
303 | /* Initialize the poll safe wake up structure */ |
304 | static void ep_poll_safewake_init(struct poll_safewake *psw) | |
305 | { | |
306 | ||
307 | INIT_LIST_HEAD(&psw->wake_task_list); | |
308 | spin_lock_init(&psw->lock); | |
309 | } | |
310 | ||
1da177e4 LT |
311 | /* |
312 | * Perform a safe wake up of the poll wait list. The problem is that | |
313 | * with the new callback'd wake up system, it is possible that the | |
314 | * poll callback is reentered from inside the call to wake_up() done | |
315 | * on the poll wait queue head. The rule is that we cannot reenter the | |
316 | * wake up code from the same task more than EP_MAX_POLLWAKE_NESTS times, | |
317 | * and we cannot reenter the same wait queue head at all. This will | |
318 | * enable to have a hierarchy of epoll file descriptor of no more than | |
319 | * EP_MAX_POLLWAKE_NESTS deep. We need the irq version of the spin lock | |
320 | * because this one gets called by the poll callback, that in turn is called | |
321 | * from inside a wake_up(), that might be called from irq context. | |
322 | */ | |
323 | static void ep_poll_safewake(struct poll_safewake *psw, wait_queue_head_t *wq) | |
324 | { | |
325 | int wake_nests = 0; | |
326 | unsigned long flags; | |
36c8b586 | 327 | struct task_struct *this_task = current; |
1da177e4 LT |
328 | struct list_head *lsthead = &psw->wake_task_list, *lnk; |
329 | struct wake_task_node *tncur; | |
330 | struct wake_task_node tnode; | |
331 | ||
332 | spin_lock_irqsave(&psw->lock, flags); | |
333 | ||
334 | /* Try to see if the current task is already inside this wakeup call */ | |
335 | list_for_each(lnk, lsthead) { | |
336 | tncur = list_entry(lnk, struct wake_task_node, llink); | |
337 | ||
338 | if (tncur->wq == wq || | |
339 | (tncur->task == this_task && ++wake_nests > EP_MAX_POLLWAKE_NESTS)) { | |
340 | /* | |
341 | * Ops ... loop detected or maximum nest level reached. | |
342 | * We abort this wake by breaking the cycle itself. | |
343 | */ | |
344 | spin_unlock_irqrestore(&psw->lock, flags); | |
345 | return; | |
346 | } | |
347 | } | |
348 | ||
349 | /* Add the current task to the list */ | |
350 | tnode.task = this_task; | |
351 | tnode.wq = wq; | |
352 | list_add(&tnode.llink, lsthead); | |
353 | ||
354 | spin_unlock_irqrestore(&psw->lock, flags); | |
355 | ||
356 | /* Do really wake up now */ | |
357 | wake_up(wq); | |
358 | ||
359 | /* Remove the current task from the list */ | |
360 | spin_lock_irqsave(&psw->lock, flags); | |
361 | list_del(&tnode.llink); | |
362 | spin_unlock_irqrestore(&psw->lock, flags); | |
363 | } | |
364 | ||
1da177e4 | 365 | /* |
7699acd1 DL |
366 | * This function unregister poll callbacks from the associated file descriptor. |
367 | * Since this must be called without holding "ep->lock" the atomic exchange trick | |
368 | * will protect us from multiple unregister. | |
1da177e4 | 369 | */ |
7699acd1 | 370 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 371 | { |
7699acd1 DL |
372 | int nwait; |
373 | struct list_head *lsthead = &epi->pwqlist; | |
374 | struct eppoll_entry *pwq; | |
1da177e4 | 375 | |
7699acd1 DL |
376 | /* This is called without locks, so we need the atomic exchange */ |
377 | nwait = xchg(&epi->nwait, 0); | |
1da177e4 | 378 | |
7699acd1 DL |
379 | if (nwait) { |
380 | while (!list_empty(lsthead)) { | |
381 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 382 | |
7699acd1 DL |
383 | list_del_init(&pwq->llink); |
384 | remove_wait_queue(pwq->whead, &pwq->wait); | |
385 | kmem_cache_free(pwq_cache, pwq); | |
386 | } | |
1da177e4 | 387 | } |
1da177e4 LT |
388 | } |
389 | ||
7699acd1 DL |
390 | /* |
391 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 392 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
393 | */ |
394 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
395 | { | |
7699acd1 DL |
396 | unsigned long flags; |
397 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
398 | |
399 | /* | |
7699acd1 DL |
400 | * Removes poll wait queue hooks. We _have_ to do this without holding |
401 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
402 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
403 | * queue head lock when unregistering the wait queue. The wakeup callback | |
404 | * will run by holding the wait queue head lock and will call our callback | |
405 | * that will try to get "ep->lock". | |
1da177e4 | 406 | */ |
7699acd1 | 407 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 408 | |
7699acd1 DL |
409 | /* Remove the current item from the list of epoll hooks */ |
410 | spin_lock(&file->f_ep_lock); | |
411 | if (ep_is_linked(&epi->fllink)) | |
412 | list_del_init(&epi->fllink); | |
413 | spin_unlock(&file->f_ep_lock); | |
1da177e4 | 414 | |
c7ea7630 DL |
415 | if (ep_rb_linked(&epi->rbn)) |
416 | ep_rb_erase(&epi->rbn, &ep->rbr); | |
1da177e4 | 417 | |
c7ea7630 DL |
418 | spin_lock_irqsave(&ep->lock, flags); |
419 | if (ep_is_linked(&epi->rdllink)) | |
420 | list_del_init(&epi->rdllink); | |
421 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 422 | |
7699acd1 | 423 | /* At this point it is safe to free the eventpoll item */ |
c7ea7630 | 424 | kmem_cache_free(epi_cache, epi); |
1da177e4 | 425 | |
c7ea7630 DL |
426 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_remove(%p, %p)\n", |
427 | current, ep, file)); | |
1da177e4 | 428 | |
c7ea7630 | 429 | return 0; |
1da177e4 LT |
430 | } |
431 | ||
7699acd1 | 432 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 433 | { |
7699acd1 DL |
434 | struct rb_node *rbp; |
435 | struct epitem *epi; | |
1da177e4 | 436 | |
7699acd1 DL |
437 | /* We need to release all tasks waiting for these file */ |
438 | if (waitqueue_active(&ep->poll_wait)) | |
439 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1da177e4 | 440 | |
7699acd1 DL |
441 | /* |
442 | * We need to lock this because we could be hit by | |
443 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 444 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
445 | * is on the way to be removed and no one has references to it |
446 | * anymore. The only hit might come from eventpoll_release_file() but | |
447 | * holding "epmutex" is sufficent here. | |
448 | */ | |
449 | mutex_lock(&epmutex); | |
1da177e4 LT |
450 | |
451 | /* | |
7699acd1 | 452 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 453 | */ |
7699acd1 DL |
454 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
455 | epi = rb_entry(rbp, struct epitem, rbn); | |
456 | ||
457 | ep_unregister_pollwait(ep, epi); | |
458 | } | |
1da177e4 LT |
459 | |
460 | /* | |
7699acd1 DL |
461 | * Walks through the whole tree by freeing each "struct epitem". At this |
462 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 463 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 464 | * us during this operation. So we can avoid the lock on "ep->lock". |
1da177e4 | 465 | */ |
7699acd1 DL |
466 | while ((rbp = rb_first(&ep->rbr)) != 0) { |
467 | epi = rb_entry(rbp, struct epitem, rbn); | |
468 | ep_remove(ep, epi); | |
469 | } | |
1da177e4 | 470 | |
7699acd1 | 471 | mutex_unlock(&epmutex); |
d47de16c DL |
472 | |
473 | mutex_destroy(&ep->mtx); | |
7699acd1 | 474 | } |
1da177e4 | 475 | |
7699acd1 DL |
476 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
477 | { | |
478 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 479 | |
7699acd1 DL |
480 | if (ep) { |
481 | ep_free(ep); | |
482 | kfree(ep); | |
483 | } | |
484 | ||
485 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: close() ep=%p\n", current, ep)); | |
486 | return 0; | |
1da177e4 LT |
487 | } |
488 | ||
7699acd1 DL |
489 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
490 | { | |
491 | unsigned int pollflags = 0; | |
492 | unsigned long flags; | |
493 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 494 | |
7699acd1 DL |
495 | /* Insert inside our poll wait queue */ |
496 | poll_wait(file, &ep->poll_wait, wait); | |
497 | ||
498 | /* Check our condition */ | |
c7ea7630 | 499 | spin_lock_irqsave(&ep->lock, flags); |
7699acd1 DL |
500 | if (!list_empty(&ep->rdllist)) |
501 | pollflags = POLLIN | POLLRDNORM; | |
c7ea7630 | 502 | spin_unlock_irqrestore(&ep->lock, flags); |
7699acd1 DL |
503 | |
504 | return pollflags; | |
505 | } | |
506 | ||
507 | /* File callbacks that implement the eventpoll file behaviour */ | |
508 | static const struct file_operations eventpoll_fops = { | |
509 | .release = ep_eventpoll_release, | |
510 | .poll = ep_eventpoll_poll | |
511 | }; | |
512 | ||
513 | /* Fast test to see if the file is an evenpoll file */ | |
514 | static inline int is_file_epoll(struct file *f) | |
515 | { | |
516 | return f->f_op == &eventpoll_fops; | |
517 | } | |
b611967d DL |
518 | |
519 | /* | |
7699acd1 DL |
520 | * This is called from eventpoll_release() to unlink files from the eventpoll |
521 | * interface. We need to have this facility to cleanup correctly files that are | |
522 | * closed without being removed from the eventpoll interface. | |
b611967d | 523 | */ |
7699acd1 | 524 | void eventpoll_release_file(struct file *file) |
b611967d | 525 | { |
7699acd1 DL |
526 | struct list_head *lsthead = &file->f_ep_links; |
527 | struct eventpoll *ep; | |
528 | struct epitem *epi; | |
b611967d DL |
529 | |
530 | /* | |
7699acd1 DL |
531 | * We don't want to get "file->f_ep_lock" because it is not |
532 | * necessary. It is not necessary because we're in the "struct file" | |
533 | * cleanup path, and this means that noone is using this file anymore. | |
d47de16c | 534 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 535 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 536 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 537 | * from anywhere but ep_free(). |
b611967d | 538 | */ |
7699acd1 | 539 | mutex_lock(&epmutex); |
b611967d | 540 | |
7699acd1 DL |
541 | while (!list_empty(lsthead)) { |
542 | epi = list_first_entry(lsthead, struct epitem, fllink); | |
b611967d | 543 | |
7699acd1 DL |
544 | ep = epi->ep; |
545 | list_del_init(&epi->fllink); | |
d47de16c | 546 | mutex_lock(&ep->mtx); |
7699acd1 | 547 | ep_remove(ep, epi); |
d47de16c | 548 | mutex_unlock(&ep->mtx); |
b611967d DL |
549 | } |
550 | ||
7699acd1 | 551 | mutex_unlock(&epmutex); |
b611967d DL |
552 | } |
553 | ||
53d2be79 | 554 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 555 | { |
53d2be79 | 556 | struct eventpoll *ep = kzalloc(sizeof(*ep), GFP_KERNEL); |
1da177e4 | 557 | |
53d2be79 | 558 | if (!ep) |
1da177e4 LT |
559 | return -ENOMEM; |
560 | ||
c7ea7630 | 561 | spin_lock_init(&ep->lock); |
d47de16c | 562 | mutex_init(&ep->mtx); |
1da177e4 LT |
563 | init_waitqueue_head(&ep->wq); |
564 | init_waitqueue_head(&ep->poll_wait); | |
565 | INIT_LIST_HEAD(&ep->rdllist); | |
566 | ep->rbr = RB_ROOT; | |
d47de16c | 567 | ep->ovflist = EP_UNACTIVE_PTR; |
1da177e4 | 568 | |
53d2be79 | 569 | *pep = ep; |
1da177e4 | 570 | |
53d2be79 | 571 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_alloc() ep=%p\n", |
1da177e4 LT |
572 | current, ep)); |
573 | return 0; | |
574 | } | |
575 | ||
1da177e4 | 576 | /* |
c7ea7630 DL |
577 | * Search the file inside the eventpoll tree. The RB tree operations |
578 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
579 | * "mtx" held. | |
1da177e4 LT |
580 | */ |
581 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
582 | { | |
583 | int kcmp; | |
1da177e4 LT |
584 | struct rb_node *rbp; |
585 | struct epitem *epi, *epir = NULL; | |
586 | struct epoll_filefd ffd; | |
587 | ||
b030a4dd | 588 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
589 | for (rbp = ep->rbr.rb_node; rbp; ) { |
590 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 591 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
592 | if (kcmp > 0) |
593 | rbp = rbp->rb_right; | |
594 | else if (kcmp < 0) | |
595 | rbp = rbp->rb_left; | |
596 | else { | |
1da177e4 LT |
597 | epir = epi; |
598 | break; | |
599 | } | |
600 | } | |
1da177e4 LT |
601 | |
602 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_find(%p) -> %p\n", | |
603 | current, file, epir)); | |
604 | ||
605 | return epir; | |
606 | } | |
607 | ||
1da177e4 | 608 | /* |
7699acd1 DL |
609 | * This is the callback that is passed to the wait queue wakeup |
610 | * machanism. It is called by the stored file descriptors when they | |
611 | * have events to report. | |
1da177e4 | 612 | */ |
7699acd1 | 613 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 614 | { |
7699acd1 DL |
615 | int pwake = 0; |
616 | unsigned long flags; | |
617 | struct epitem *epi = ep_item_from_wait(wait); | |
618 | struct eventpoll *ep = epi->ep; | |
1da177e4 | 619 | |
7699acd1 DL |
620 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: poll_callback(%p) epi=%p ep=%p\n", |
621 | current, epi->ffd.file, epi, ep)); | |
1da177e4 | 622 | |
c7ea7630 | 623 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 624 | |
7699acd1 DL |
625 | /* |
626 | * If the event mask does not contain any poll(2) event, we consider the | |
627 | * descriptor to be disabled. This condition is likely the effect of the | |
628 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
629 | * until the next EPOLL_CTL_MOD will be issued. | |
630 | */ | |
631 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
632 | goto out_unlock; |
633 | ||
634 | /* | |
635 | * If we are trasfering events to userspace, we can hold no locks | |
636 | * (because we're accessing user memory, and because of linux f_op->poll() | |
637 | * semantics). All the events that happens during that period of time are | |
638 | * chained in ep->ovflist and requeued later on. | |
639 | */ | |
640 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
641 | if (epi->next == EP_UNACTIVE_PTR) { | |
642 | epi->next = ep->ovflist; | |
643 | ep->ovflist = epi; | |
644 | } | |
645 | goto out_unlock; | |
646 | } | |
1da177e4 | 647 | |
7699acd1 DL |
648 | /* If this file is already in the ready list we exit soon */ |
649 | if (ep_is_linked(&epi->rdllink)) | |
650 | goto is_linked; | |
651 | ||
652 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
653 | ||
654 | is_linked: | |
655 | /* | |
656 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
657 | * wait list. | |
658 | */ | |
659 | if (waitqueue_active(&ep->wq)) | |
660 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | | |
661 | TASK_INTERRUPTIBLE); | |
662 | if (waitqueue_active(&ep->poll_wait)) | |
663 | pwake++; | |
664 | ||
d47de16c | 665 | out_unlock: |
c7ea7630 | 666 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 667 | |
7699acd1 DL |
668 | /* We have to call this outside the lock */ |
669 | if (pwake) | |
670 | ep_poll_safewake(&psw, &ep->poll_wait); | |
671 | ||
672 | return 1; | |
673 | } | |
1da177e4 LT |
674 | |
675 | /* | |
676 | * This is the callback that is used to add our wait queue to the | |
677 | * target file wakeup lists. | |
678 | */ | |
679 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
680 | poll_table *pt) | |
681 | { | |
b030a4dd | 682 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
683 | struct eppoll_entry *pwq; |
684 | ||
e94b1766 | 685 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
686 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
687 | pwq->whead = whead; | |
688 | pwq->base = epi; | |
689 | add_wait_queue(whead, &pwq->wait); | |
690 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
691 | epi->nwait++; | |
692 | } else { | |
693 | /* We have to signal that an error occurred */ | |
694 | epi->nwait = -1; | |
695 | } | |
696 | } | |
697 | ||
1da177e4 LT |
698 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
699 | { | |
700 | int kcmp; | |
701 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
702 | struct epitem *epic; | |
703 | ||
704 | while (*p) { | |
705 | parent = *p; | |
706 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 707 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
708 | if (kcmp > 0) |
709 | p = &parent->rb_right; | |
710 | else | |
711 | p = &parent->rb_left; | |
712 | } | |
713 | rb_link_node(&epi->rbn, parent, p); | |
714 | rb_insert_color(&epi->rbn, &ep->rbr); | |
715 | } | |
716 | ||
c7ea7630 DL |
717 | /* |
718 | * Must be called with "mtx" held. | |
719 | */ | |
1da177e4 LT |
720 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
721 | struct file *tfile, int fd) | |
722 | { | |
723 | int error, revents, pwake = 0; | |
724 | unsigned long flags; | |
725 | struct epitem *epi; | |
726 | struct ep_pqueue epq; | |
727 | ||
728 | error = -ENOMEM; | |
e94b1766 | 729 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7699acd1 | 730 | goto error_return; |
1da177e4 LT |
731 | |
732 | /* Item initialization follow here ... */ | |
b030a4dd | 733 | ep_rb_initnode(&epi->rbn); |
1da177e4 LT |
734 | INIT_LIST_HEAD(&epi->rdllink); |
735 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
736 | INIT_LIST_HEAD(&epi->pwqlist); |
737 | epi->ep = ep; | |
b030a4dd | 738 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 739 | epi->event = *event; |
1da177e4 | 740 | epi->nwait = 0; |
d47de16c | 741 | epi->next = EP_UNACTIVE_PTR; |
1da177e4 LT |
742 | |
743 | /* Initialize the poll table using the queue callback */ | |
744 | epq.epi = epi; | |
745 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
746 | ||
747 | /* | |
748 | * Attach the item to the poll hooks and get current event bits. | |
749 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
750 | * been increased by the caller of this function. Note that after |
751 | * this operation completes, the poll callback can start hitting | |
752 | * the new item. | |
1da177e4 LT |
753 | */ |
754 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
755 | ||
756 | /* | |
757 | * We have to check if something went wrong during the poll wait queue | |
758 | * install process. Namely an allocation for a wait queue failed due | |
759 | * high memory pressure. | |
760 | */ | |
761 | if (epi->nwait < 0) | |
7699acd1 | 762 | goto error_unregister; |
1da177e4 LT |
763 | |
764 | /* Add the current item to the list of active epoll hook for this file */ | |
765 | spin_lock(&tfile->f_ep_lock); | |
766 | list_add_tail(&epi->fllink, &tfile->f_ep_links); | |
767 | spin_unlock(&tfile->f_ep_lock); | |
768 | ||
c7ea7630 DL |
769 | /* |
770 | * Add the current item to the RB tree. All RB tree operations are | |
771 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
772 | */ | |
1da177e4 LT |
773 | ep_rbtree_insert(ep, epi); |
774 | ||
c7ea7630 DL |
775 | /* We have to drop the new item inside our item list to keep track of it */ |
776 | spin_lock_irqsave(&ep->lock, flags); | |
777 | ||
1da177e4 | 778 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 779 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 LT |
780 | list_add_tail(&epi->rdllink, &ep->rdllist); |
781 | ||
782 | /* Notify waiting tasks that events are available */ | |
783 | if (waitqueue_active(&ep->wq)) | |
3419b23a | 784 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE); |
1da177e4 LT |
785 | if (waitqueue_active(&ep->poll_wait)) |
786 | pwake++; | |
787 | } | |
788 | ||
c7ea7630 | 789 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
790 | |
791 | /* We have to call this outside the lock */ | |
792 | if (pwake) | |
793 | ep_poll_safewake(&psw, &ep->poll_wait); | |
794 | ||
795 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_insert(%p, %p, %d)\n", | |
796 | current, ep, tfile, fd)); | |
797 | ||
798 | return 0; | |
799 | ||
7699acd1 | 800 | error_unregister: |
1da177e4 LT |
801 | ep_unregister_pollwait(ep, epi); |
802 | ||
803 | /* | |
804 | * We need to do this because an event could have been arrived on some | |
805 | * allocated wait queue. | |
806 | */ | |
c7ea7630 | 807 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 808 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 809 | list_del_init(&epi->rdllink); |
c7ea7630 | 810 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 811 | |
b030a4dd | 812 | kmem_cache_free(epi_cache, epi); |
7699acd1 | 813 | error_return: |
1da177e4 LT |
814 | return error; |
815 | } | |
816 | ||
1da177e4 LT |
817 | /* |
818 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 819 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
820 | */ |
821 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
822 | { | |
823 | int pwake = 0; | |
824 | unsigned int revents; | |
825 | unsigned long flags; | |
826 | ||
827 | /* | |
828 | * Set the new event interest mask before calling f_op->poll(), otherwise | |
829 | * a potential race might occur. In fact if we do this operation inside | |
830 | * the lock, an event might happen between the f_op->poll() call and the | |
831 | * new event set registering. | |
832 | */ | |
833 | epi->event.events = event->events; | |
834 | ||
835 | /* | |
836 | * Get current event bits. We can safely use the file* here because | |
837 | * its usage count has been increased by the caller of this function. | |
838 | */ | |
839 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
840 | ||
c7ea7630 | 841 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 LT |
842 | |
843 | /* Copy the data member from inside the lock */ | |
844 | epi->event.data = event->data; | |
845 | ||
846 | /* | |
c7ea7630 DL |
847 | * If the item is "hot" and it is not registered inside the ready |
848 | * list, push it inside. If the item is not "hot" and it is currently | |
849 | * registered inside the ready list, unlink it. | |
1da177e4 | 850 | */ |
c7ea7630 DL |
851 | if (revents & event->events) { |
852 | if (!ep_is_linked(&epi->rdllink)) { | |
853 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
854 | ||
855 | /* Notify waiting tasks that events are available */ | |
856 | if (waitqueue_active(&ep->wq)) | |
857 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | | |
858 | TASK_INTERRUPTIBLE); | |
859 | if (waitqueue_active(&ep->poll_wait)) | |
860 | pwake++; | |
7699acd1 DL |
861 | } |
862 | } | |
c7ea7630 | 863 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 864 | |
7699acd1 DL |
865 | /* We have to call this outside the lock */ |
866 | if (pwake) | |
867 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1da177e4 | 868 | |
7699acd1 | 869 | return 0; |
1da177e4 LT |
870 | } |
871 | ||
d47de16c DL |
872 | static int ep_send_events(struct eventpoll *ep, struct epoll_event __user *events, |
873 | int maxevents) | |
1da177e4 | 874 | { |
6192bd53 | 875 | int eventcnt, error = -EFAULT, pwake = 0; |
1da177e4 | 876 | unsigned int revents; |
6192bd53 | 877 | unsigned long flags; |
d47de16c DL |
878 | struct epitem *epi, *nepi; |
879 | struct list_head txlist; | |
880 | ||
881 | INIT_LIST_HEAD(&txlist); | |
6192bd53 | 882 | |
d47de16c DL |
883 | /* |
884 | * We need to lock this because we could be hit by | |
885 | * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL). | |
886 | */ | |
887 | mutex_lock(&ep->mtx); | |
888 | ||
889 | /* | |
890 | * Steal the ready list, and re-init the original one to the | |
891 | * empty list. Also, set ep->ovflist to NULL so that events | |
892 | * happening while looping w/out locks, are not lost. We cannot | |
893 | * have the poll callback to queue directly on ep->rdllist, | |
894 | * because we are doing it in the loop below, in a lockless way. | |
895 | */ | |
c7ea7630 | 896 | spin_lock_irqsave(&ep->lock, flags); |
d47de16c DL |
897 | list_splice(&ep->rdllist, &txlist); |
898 | INIT_LIST_HEAD(&ep->rdllist); | |
899 | ep->ovflist = NULL; | |
c7ea7630 | 900 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
901 | |
902 | /* | |
903 | * We can loop without lock because this is a task private list. | |
6192bd53 | 904 | * We just splice'd out the ep->rdllist in ep_collect_ready_items(). |
d47de16c | 905 | * Items cannot vanish during the loop because we are holding "mtx". |
1da177e4 | 906 | */ |
d47de16c DL |
907 | for (eventcnt = 0; !list_empty(&txlist) && eventcnt < maxevents;) { |
908 | epi = list_first_entry(&txlist, struct epitem, rdllink); | |
909 | ||
910 | list_del_init(&epi->rdllink); | |
1da177e4 LT |
911 | |
912 | /* | |
913 | * Get the ready file event set. We can safely use the file | |
d47de16c DL |
914 | * because we are holding the "mtx" and this will guarantee |
915 | * that both the file and the item will not vanish. | |
1da177e4 LT |
916 | */ |
917 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
6192bd53 | 918 | revents &= epi->event.events; |
1da177e4 LT |
919 | |
920 | /* | |
6192bd53 DL |
921 | * Is the event mask intersect the caller-requested one, |
922 | * deliver the event to userspace. Again, we are holding | |
d47de16c DL |
923 | * "mtx", so no operations coming from userspace can change |
924 | * the item. | |
1da177e4 | 925 | */ |
6192bd53 DL |
926 | if (revents) { |
927 | if (__put_user(revents, | |
1da177e4 LT |
928 | &events[eventcnt].events) || |
929 | __put_user(epi->event.data, | |
930 | &events[eventcnt].data)) | |
6192bd53 | 931 | goto errxit; |
1da177e4 LT |
932 | if (epi->event.events & EPOLLONESHOT) |
933 | epi->event.events &= EP_PRIVATE_BITS; | |
934 | eventcnt++; | |
935 | } | |
1da177e4 | 936 | /* |
d47de16c DL |
937 | * At this point, noone can insert into ep->rdllist besides |
938 | * us. The epoll_ctl() callers are locked out by us holding | |
939 | * "mtx" and the poll callback will queue them in ep->ovflist. | |
1da177e4 | 940 | */ |
6192bd53 | 941 | if (!(epi->event.events & EPOLLET) && |
d47de16c DL |
942 | (revents & epi->event.events)) |
943 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1da177e4 | 944 | } |
6192bd53 | 945 | error = 0; |
1da177e4 | 946 | |
d47de16c | 947 | errxit: |
6192bd53 | 948 | |
c7ea7630 | 949 | spin_lock_irqsave(&ep->lock, flags); |
d47de16c DL |
950 | /* |
951 | * During the time we spent in the loop above, some other events | |
952 | * might have been queued by the poll callback. We re-insert them | |
953 | * here (in case they are not already queued, or they're one-shot). | |
954 | */ | |
955 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
956 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { | |
957 | if (!ep_is_linked(&epi->rdllink) && | |
958 | (epi->event.events & ~EP_PRIVATE_BITS)) | |
959 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
960 | } | |
6192bd53 | 961 | /* |
d47de16c DL |
962 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after |
963 | * releasing the lock, events will be queued in the normal way inside | |
964 | * ep->rdllist. | |
6192bd53 | 965 | */ |
d47de16c | 966 | ep->ovflist = EP_UNACTIVE_PTR; |
6192bd53 | 967 | |
d47de16c DL |
968 | /* |
969 | * In case of error in the event-send loop, we might still have items | |
970 | * inside the "txlist". We need to splice them back inside ep->rdllist. | |
971 | */ | |
972 | list_splice(&txlist, &ep->rdllist); | |
973 | ||
974 | if (!list_empty(&ep->rdllist)) { | |
1da177e4 | 975 | /* |
d47de16c | 976 | * Wake up (if active) both the eventpoll wait list and the ->poll() |
1da177e4 LT |
977 | * wait list. |
978 | */ | |
979 | if (waitqueue_active(&ep->wq)) | |
3419b23a DL |
980 | __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | |
981 | TASK_INTERRUPTIBLE); | |
1da177e4 LT |
982 | if (waitqueue_active(&ep->poll_wait)) |
983 | pwake++; | |
6192bd53 | 984 | } |
c7ea7630 | 985 | spin_unlock_irqrestore(&ep->lock, flags); |
d47de16c DL |
986 | |
987 | mutex_unlock(&ep->mtx); | |
1da177e4 LT |
988 | |
989 | /* We have to call this outside the lock */ | |
990 | if (pwake) | |
991 | ep_poll_safewake(&psw, &ep->poll_wait); | |
6192bd53 DL |
992 | |
993 | return eventcnt == 0 ? error: eventcnt; | |
1da177e4 LT |
994 | } |
995 | ||
1da177e4 LT |
996 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
997 | int maxevents, long timeout) | |
998 | { | |
999 | int res, eavail; | |
1000 | unsigned long flags; | |
1001 | long jtimeout; | |
1002 | wait_queue_t wait; | |
1003 | ||
1004 | /* | |
1005 | * Calculate the timeout by checking for the "infinite" value ( -1 ) | |
1006 | * and the overflow condition. The passed timeout is in milliseconds, | |
1007 | * that why (t * HZ) / 1000. | |
1008 | */ | |
e3306dd5 DL |
1009 | jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ? |
1010 | MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000; | |
1da177e4 LT |
1011 | |
1012 | retry: | |
c7ea7630 | 1013 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 LT |
1014 | |
1015 | res = 0; | |
1016 | if (list_empty(&ep->rdllist)) { | |
1017 | /* | |
1018 | * We don't have any available event to return to the caller. | |
1019 | * We need to sleep here, and we will be wake up by | |
1020 | * ep_poll_callback() when events will become available. | |
1021 | */ | |
1022 | init_waitqueue_entry(&wait, current); | |
d47de16c | 1023 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
3419b23a | 1024 | __add_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1025 | |
1026 | for (;;) { | |
1027 | /* | |
1028 | * We don't want to sleep if the ep_poll_callback() sends us | |
1029 | * a wakeup in between. That's why we set the task state | |
1030 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1031 | */ | |
1032 | set_current_state(TASK_INTERRUPTIBLE); | |
1033 | if (!list_empty(&ep->rdllist) || !jtimeout) | |
1034 | break; | |
1035 | if (signal_pending(current)) { | |
1036 | res = -EINTR; | |
1037 | break; | |
1038 | } | |
1039 | ||
c7ea7630 | 1040 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1041 | jtimeout = schedule_timeout(jtimeout); |
c7ea7630 | 1042 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1043 | } |
3419b23a | 1044 | __remove_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1045 | |
1046 | set_current_state(TASK_RUNNING); | |
1047 | } | |
1048 | ||
1049 | /* Is it worth to try to dig for events ? */ | |
1050 | eavail = !list_empty(&ep->rdllist); | |
1051 | ||
c7ea7630 | 1052 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1053 | |
1054 | /* | |
1055 | * Try to transfer events to user space. In case we get 0 events and | |
1056 | * there's still timeout left over, we go trying again in search of | |
1057 | * more luck. | |
1058 | */ | |
1059 | if (!res && eavail && | |
d47de16c | 1060 | !(res = ep_send_events(ep, events, maxevents)) && jtimeout) |
1da177e4 LT |
1061 | goto retry; |
1062 | ||
1063 | return res; | |
1064 | } | |
1065 | ||
7699acd1 DL |
1066 | /* |
1067 | * It opens an eventpoll file descriptor by suggesting a storage of "size" | |
1068 | * file descriptors. The size parameter is just an hint about how to size | |
1069 | * data structures. It won't prevent the user to store more than "size" | |
1070 | * file descriptors inside the epoll interface. It is the kernel part of | |
1071 | * the userspace epoll_create(2). | |
1072 | */ | |
1073 | asmlinkage long sys_epoll_create(int size) | |
1074 | { | |
1075 | int error, fd = -1; | |
1076 | struct eventpoll *ep; | |
1077 | struct inode *inode; | |
1078 | struct file *file; | |
1079 | ||
1080 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d)\n", | |
1081 | current, size)); | |
1082 | ||
1083 | /* | |
1084 | * Sanity check on the size parameter, and create the internal data | |
1085 | * structure ( "struct eventpoll" ). | |
1086 | */ | |
1087 | error = -EINVAL; | |
1088 | if (size <= 0 || (error = ep_alloc(&ep)) != 0) | |
1089 | goto error_return; | |
1090 | ||
1091 | /* | |
1092 | * Creates all the items needed to setup an eventpoll file. That is, | |
1093 | * a file structure, and inode and a free file descriptor. | |
1094 | */ | |
1095 | error = anon_inode_getfd(&fd, &inode, &file, "[eventpoll]", | |
1096 | &eventpoll_fops, ep); | |
1097 | if (error) | |
1098 | goto error_free; | |
1099 | ||
1100 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", | |
1101 | current, size, fd)); | |
1102 | ||
1103 | return fd; | |
1104 | ||
1105 | error_free: | |
1106 | ep_free(ep); | |
1107 | kfree(ep); | |
1108 | error_return: | |
1109 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", | |
1110 | current, size, error)); | |
1111 | return error; | |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * The following function implements the controller interface for | |
1116 | * the eventpoll file that enables the insertion/removal/change of | |
1117 | * file descriptors inside the interest set. It represents | |
1118 | * the kernel part of the user space epoll_ctl(2). | |
1119 | */ | |
1120 | asmlinkage long sys_epoll_ctl(int epfd, int op, int fd, | |
1121 | struct epoll_event __user *event) | |
1122 | { | |
1123 | int error; | |
1124 | struct file *file, *tfile; | |
1125 | struct eventpoll *ep; | |
1126 | struct epitem *epi; | |
1127 | struct epoll_event epds; | |
1128 | ||
1129 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p)\n", | |
1130 | current, epfd, op, fd, event)); | |
1131 | ||
1132 | error = -EFAULT; | |
1133 | if (ep_op_has_event(op) && | |
1134 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
1135 | goto error_return; | |
1136 | ||
1137 | /* Get the "struct file *" for the eventpoll file */ | |
1138 | error = -EBADF; | |
1139 | file = fget(epfd); | |
1140 | if (!file) | |
1141 | goto error_return; | |
1142 | ||
1143 | /* Get the "struct file *" for the target file */ | |
1144 | tfile = fget(fd); | |
1145 | if (!tfile) | |
1146 | goto error_fput; | |
1147 | ||
1148 | /* The target file descriptor must support poll */ | |
1149 | error = -EPERM; | |
1150 | if (!tfile->f_op || !tfile->f_op->poll) | |
1151 | goto error_tgt_fput; | |
1152 | ||
1153 | /* | |
1154 | * We have to check that the file structure underneath the file descriptor | |
1155 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
1156 | * adding an epoll file descriptor inside itself. | |
1157 | */ | |
1158 | error = -EINVAL; | |
1159 | if (file == tfile || !is_file_epoll(file)) | |
1160 | goto error_tgt_fput; | |
1161 | ||
1162 | /* | |
1163 | * At this point it is safe to assume that the "private_data" contains | |
1164 | * our own data structure. | |
1165 | */ | |
1166 | ep = file->private_data; | |
1167 | ||
d47de16c | 1168 | mutex_lock(&ep->mtx); |
7699acd1 DL |
1169 | |
1170 | /* Try to lookup the file inside our RB tree */ | |
1171 | epi = ep_find(ep, tfile, fd); | |
1172 | ||
1173 | error = -EINVAL; | |
1174 | switch (op) { | |
1175 | case EPOLL_CTL_ADD: | |
1176 | if (!epi) { | |
1177 | epds.events |= POLLERR | POLLHUP; | |
1178 | ||
1179 | error = ep_insert(ep, &epds, tfile, fd); | |
1180 | } else | |
1181 | error = -EEXIST; | |
1182 | break; | |
1183 | case EPOLL_CTL_DEL: | |
1184 | if (epi) | |
1185 | error = ep_remove(ep, epi); | |
1186 | else | |
1187 | error = -ENOENT; | |
1188 | break; | |
1189 | case EPOLL_CTL_MOD: | |
1190 | if (epi) { | |
1191 | epds.events |= POLLERR | POLLHUP; | |
1192 | error = ep_modify(ep, epi, &epds); | |
1193 | } else | |
1194 | error = -ENOENT; | |
1195 | break; | |
1196 | } | |
d47de16c | 1197 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
1198 | |
1199 | error_tgt_fput: | |
1200 | fput(tfile); | |
1201 | error_fput: | |
1202 | fput(file); | |
1203 | error_return: | |
1204 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p) = %d\n", | |
1205 | current, epfd, op, fd, event, error)); | |
1206 | ||
1207 | return error; | |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1212 | * part of the user space epoll_wait(2). | |
1213 | */ | |
1214 | asmlinkage long sys_epoll_wait(int epfd, struct epoll_event __user *events, | |
1215 | int maxevents, int timeout) | |
1216 | { | |
1217 | int error; | |
1218 | struct file *file; | |
1219 | struct eventpoll *ep; | |
1220 | ||
1221 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d)\n", | |
1222 | current, epfd, events, maxevents, timeout)); | |
1223 | ||
1224 | /* The maximum number of event must be greater than zero */ | |
1225 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
1226 | return -EINVAL; | |
1227 | ||
1228 | /* Verify that the area passed by the user is writeable */ | |
1229 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { | |
1230 | error = -EFAULT; | |
1231 | goto error_return; | |
1232 | } | |
1233 | ||
1234 | /* Get the "struct file *" for the eventpoll file */ | |
1235 | error = -EBADF; | |
1236 | file = fget(epfd); | |
1237 | if (!file) | |
1238 | goto error_return; | |
1239 | ||
1240 | /* | |
1241 | * We have to check that the file structure underneath the fd | |
1242 | * the user passed to us _is_ an eventpoll file. | |
1243 | */ | |
1244 | error = -EINVAL; | |
1245 | if (!is_file_epoll(file)) | |
1246 | goto error_fput; | |
1247 | ||
1248 | /* | |
1249 | * At this point it is safe to assume that the "private_data" contains | |
1250 | * our own data structure. | |
1251 | */ | |
1252 | ep = file->private_data; | |
1253 | ||
1254 | /* Time to fish for events ... */ | |
1255 | error = ep_poll(ep, events, maxevents, timeout); | |
1256 | ||
1257 | error_fput: | |
1258 | fput(file); | |
1259 | error_return: | |
1260 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d) = %d\n", | |
1261 | current, epfd, events, maxevents, timeout, error)); | |
1262 | ||
1263 | return error; | |
1264 | } | |
1265 | ||
1266 | #ifdef TIF_RESTORE_SIGMASK | |
1267 | ||
1268 | /* | |
1269 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1270 | * part of the user space epoll_pwait(2). | |
1271 | */ | |
1272 | asmlinkage long sys_epoll_pwait(int epfd, struct epoll_event __user *events, | |
1273 | int maxevents, int timeout, const sigset_t __user *sigmask, | |
1274 | size_t sigsetsize) | |
1275 | { | |
1276 | int error; | |
1277 | sigset_t ksigmask, sigsaved; | |
1278 | ||
1279 | /* | |
1280 | * If the caller wants a certain signal mask to be set during the wait, | |
1281 | * we apply it here. | |
1282 | */ | |
1283 | if (sigmask) { | |
1284 | if (sigsetsize != sizeof(sigset_t)) | |
1285 | return -EINVAL; | |
1286 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1287 | return -EFAULT; | |
1288 | sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP)); | |
1289 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1290 | } | |
1291 | ||
1292 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
1293 | ||
1294 | /* | |
1295 | * If we changed the signal mask, we need to restore the original one. | |
1296 | * In case we've got a signal while waiting, we do not restore the | |
1297 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
1298 | * the way back to userspace, before the signal mask is restored. | |
1299 | */ | |
1300 | if (sigmask) { | |
1301 | if (error == -EINTR) { | |
1302 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 1303 | sizeof(sigsaved)); |
7699acd1 DL |
1304 | set_thread_flag(TIF_RESTORE_SIGMASK); |
1305 | } else | |
1306 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1307 | } | |
1308 | ||
1309 | return error; | |
1310 | } | |
1311 | ||
1312 | #endif /* #ifdef TIF_RESTORE_SIGMASK */ | |
1313 | ||
1da177e4 LT |
1314 | static int __init eventpoll_init(void) |
1315 | { | |
144efe3e | 1316 | mutex_init(&epmutex); |
1da177e4 LT |
1317 | |
1318 | /* Initialize the structure used to perform safe poll wait head wake ups */ | |
1319 | ep_poll_safewake_init(&psw); | |
1320 | ||
1321 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1322 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
1323 | 0, SLAB_HWCACHE_ALIGN|EPI_SLAB_DEBUG|SLAB_PANIC, | |
1324 | NULL, NULL); | |
1325 | ||
1326 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1327 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
1328 | sizeof(struct eppoll_entry), 0, | |
1329 | EPI_SLAB_DEBUG|SLAB_PANIC, NULL, NULL); | |
1330 | ||
1da177e4 | 1331 | return 0; |
1da177e4 | 1332 | } |
cea69241 | 1333 | fs_initcall(eventpoll_init); |
1da177e4 | 1334 |