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 | ||
7ef9964e DL |
105 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
106 | ||
1da177e4 LT |
107 | struct epoll_filefd { |
108 | struct file *file; | |
109 | int fd; | |
110 | }; | |
111 | ||
112 | /* | |
113 | * Node that is linked into the "wake_task_list" member of the "struct poll_safewake". | |
114 | * It is used to keep track on all tasks that are currently inside the wake_up() code | |
115 | * to 1) short-circuit the one coming from the same task and same wait queue head | |
d47de16c | 116 | * (loop) 2) allow a maximum number of epoll descriptors inclusion nesting |
1da177e4 LT |
117 | * 3) let go the ones coming from other tasks. |
118 | */ | |
119 | struct wake_task_node { | |
120 | struct list_head llink; | |
36c8b586 | 121 | struct task_struct *task; |
1da177e4 LT |
122 | wait_queue_head_t *wq; |
123 | }; | |
124 | ||
125 | /* | |
126 | * This is used to implement the safe poll wake up avoiding to reenter | |
127 | * the poll callback from inside wake_up(). | |
128 | */ | |
129 | struct poll_safewake { | |
130 | struct list_head wake_task_list; | |
131 | spinlock_t lock; | |
132 | }; | |
133 | ||
d47de16c DL |
134 | /* |
135 | * Each file descriptor added to the eventpoll interface will | |
136 | * have an entry of this type linked to the "rbr" RB tree. | |
137 | */ | |
138 | struct epitem { | |
67647d0f | 139 | /* RB tree node used to link this structure to the eventpoll RB tree */ |
d47de16c DL |
140 | struct rb_node rbn; |
141 | ||
142 | /* List header used to link this structure to the eventpoll ready list */ | |
143 | struct list_head rdllink; | |
144 | ||
c7ea7630 DL |
145 | /* |
146 | * Works together "struct eventpoll"->ovflist in keeping the | |
147 | * single linked chain of items. | |
148 | */ | |
149 | struct epitem *next; | |
150 | ||
d47de16c DL |
151 | /* The file descriptor information this item refers to */ |
152 | struct epoll_filefd ffd; | |
153 | ||
154 | /* Number of active wait queue attached to poll operations */ | |
155 | int nwait; | |
156 | ||
157 | /* List containing poll wait queues */ | |
158 | struct list_head pwqlist; | |
159 | ||
160 | /* The "container" of this item */ | |
161 | struct eventpoll *ep; | |
162 | ||
d47de16c DL |
163 | /* List header used to link this item to the "struct file" items list */ |
164 | struct list_head fllink; | |
165 | ||
c7ea7630 DL |
166 | /* The structure that describe the interested events and the source fd */ |
167 | struct epoll_event event; | |
d47de16c DL |
168 | }; |
169 | ||
1da177e4 LT |
170 | /* |
171 | * This structure is stored inside the "private_data" member of the file | |
172 | * structure and rapresent the main data sructure for the eventpoll | |
173 | * interface. | |
174 | */ | |
175 | struct eventpoll { | |
176 | /* Protect the this structure access */ | |
c7ea7630 | 177 | spinlock_t lock; |
1da177e4 LT |
178 | |
179 | /* | |
d47de16c DL |
180 | * This mutex is used to ensure that files are not removed |
181 | * while epoll is using them. This is held during the event | |
182 | * collection loop, the file cleanup path, the epoll file exit | |
183 | * code and the ctl operations. | |
1da177e4 | 184 | */ |
d47de16c | 185 | struct mutex mtx; |
1da177e4 LT |
186 | |
187 | /* Wait queue used by sys_epoll_wait() */ | |
188 | wait_queue_head_t wq; | |
189 | ||
190 | /* Wait queue used by file->poll() */ | |
191 | wait_queue_head_t poll_wait; | |
192 | ||
193 | /* List of ready file descriptors */ | |
194 | struct list_head rdllist; | |
195 | ||
67647d0f | 196 | /* RB tree root used to store monitored fd structs */ |
1da177e4 | 197 | struct rb_root rbr; |
d47de16c DL |
198 | |
199 | /* | |
200 | * This is a single linked list that chains all the "struct epitem" that | |
201 | * happened while transfering ready events to userspace w/out | |
202 | * holding ->lock. | |
203 | */ | |
204 | struct epitem *ovflist; | |
7ef9964e DL |
205 | |
206 | /* The user that created the eventpoll descriptor */ | |
207 | struct user_struct *user; | |
1da177e4 LT |
208 | }; |
209 | ||
210 | /* Wait structure used by the poll hooks */ | |
211 | struct eppoll_entry { | |
212 | /* List header used to link this structure to the "struct epitem" */ | |
213 | struct list_head llink; | |
214 | ||
215 | /* The "base" pointer is set to the container "struct epitem" */ | |
216 | void *base; | |
217 | ||
218 | /* | |
219 | * Wait queue item that will be linked to the target file wait | |
220 | * queue head. | |
221 | */ | |
222 | wait_queue_t wait; | |
223 | ||
224 | /* The wait queue head that linked the "wait" wait queue item */ | |
225 | wait_queue_head_t *whead; | |
226 | }; | |
227 | ||
1da177e4 LT |
228 | /* Wrapper struct used by poll queueing */ |
229 | struct ep_pqueue { | |
230 | poll_table pt; | |
231 | struct epitem *epi; | |
232 | }; | |
233 | ||
7ef9964e DL |
234 | /* |
235 | * Configuration options available inside /proc/sys/fs/epoll/ | |
236 | */ | |
237 | /* Maximum number of epoll devices, per user */ | |
238 | static int max_user_instances __read_mostly; | |
239 | /* Maximum number of epoll watched descriptors, per user */ | |
240 | static int max_user_watches __read_mostly; | |
241 | ||
1da177e4 | 242 | /* |
d47de16c | 243 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 244 | */ |
7ef9964e | 245 | static DEFINE_MUTEX(epmutex); |
1da177e4 LT |
246 | |
247 | /* Safe wake up implementation */ | |
248 | static struct poll_safewake psw; | |
249 | ||
250 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 251 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
252 | |
253 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 254 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 255 | |
7ef9964e DL |
256 | #ifdef CONFIG_SYSCTL |
257 | ||
258 | #include <linux/sysctl.h> | |
259 | ||
260 | static int zero; | |
261 | ||
262 | ctl_table epoll_table[] = { | |
263 | { | |
264 | .procname = "max_user_instances", | |
265 | .data = &max_user_instances, | |
266 | .maxlen = sizeof(int), | |
267 | .mode = 0644, | |
268 | .proc_handler = &proc_dointvec_minmax, | |
269 | .extra1 = &zero, | |
270 | }, | |
271 | { | |
272 | .procname = "max_user_watches", | |
273 | .data = &max_user_watches, | |
274 | .maxlen = sizeof(int), | |
275 | .mode = 0644, | |
276 | .proc_handler = &proc_dointvec_minmax, | |
277 | .extra1 = &zero, | |
278 | }, | |
279 | { .ctl_name = 0 } | |
280 | }; | |
281 | #endif /* CONFIG_SYSCTL */ | |
282 | ||
b030a4dd | 283 | |
67647d0f | 284 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
285 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
286 | struct file *file, int fd) | |
287 | { | |
288 | ffd->file = file; | |
289 | ffd->fd = fd; | |
290 | } | |
291 | ||
67647d0f | 292 | /* Compare RB tree keys */ |
b030a4dd PE |
293 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
294 | struct epoll_filefd *p2) | |
295 | { | |
296 | return (p1->file > p2->file ? +1: | |
297 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
298 | } | |
299 | ||
b030a4dd PE |
300 | /* Tells us if the item is currently linked */ |
301 | static inline int ep_is_linked(struct list_head *p) | |
302 | { | |
303 | return !list_empty(p); | |
304 | } | |
305 | ||
306 | /* Get the "struct epitem" from a wait queue pointer */ | |
cdac75e6 | 307 | static inline struct epitem *ep_item_from_wait(wait_queue_t *p) |
b030a4dd PE |
308 | { |
309 | return container_of(p, struct eppoll_entry, wait)->base; | |
310 | } | |
311 | ||
312 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 313 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
314 | { |
315 | return container_of(p, struct ep_pqueue, pt)->epi; | |
316 | } | |
317 | ||
318 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 319 | static inline int ep_op_has_event(int op) |
b030a4dd PE |
320 | { |
321 | return op != EPOLL_CTL_DEL; | |
322 | } | |
323 | ||
1da177e4 LT |
324 | /* Initialize the poll safe wake up structure */ |
325 | static void ep_poll_safewake_init(struct poll_safewake *psw) | |
326 | { | |
327 | ||
328 | INIT_LIST_HEAD(&psw->wake_task_list); | |
329 | spin_lock_init(&psw->lock); | |
330 | } | |
331 | ||
1da177e4 LT |
332 | /* |
333 | * Perform a safe wake up of the poll wait list. The problem is that | |
334 | * with the new callback'd wake up system, it is possible that the | |
335 | * poll callback is reentered from inside the call to wake_up() done | |
336 | * on the poll wait queue head. The rule is that we cannot reenter the | |
337 | * wake up code from the same task more than EP_MAX_POLLWAKE_NESTS times, | |
338 | * and we cannot reenter the same wait queue head at all. This will | |
339 | * enable to have a hierarchy of epoll file descriptor of no more than | |
340 | * EP_MAX_POLLWAKE_NESTS deep. We need the irq version of the spin lock | |
341 | * because this one gets called by the poll callback, that in turn is called | |
342 | * from inside a wake_up(), that might be called from irq context. | |
343 | */ | |
344 | static void ep_poll_safewake(struct poll_safewake *psw, wait_queue_head_t *wq) | |
345 | { | |
346 | int wake_nests = 0; | |
347 | unsigned long flags; | |
36c8b586 | 348 | struct task_struct *this_task = current; |
b70c3940 | 349 | struct list_head *lsthead = &psw->wake_task_list; |
1da177e4 LT |
350 | struct wake_task_node *tncur; |
351 | struct wake_task_node tnode; | |
352 | ||
353 | spin_lock_irqsave(&psw->lock, flags); | |
354 | ||
355 | /* Try to see if the current task is already inside this wakeup call */ | |
b70c3940 | 356 | list_for_each_entry(tncur, lsthead, llink) { |
1da177e4 LT |
357 | |
358 | if (tncur->wq == wq || | |
359 | (tncur->task == this_task && ++wake_nests > EP_MAX_POLLWAKE_NESTS)) { | |
360 | /* | |
361 | * Ops ... loop detected or maximum nest level reached. | |
362 | * We abort this wake by breaking the cycle itself. | |
363 | */ | |
364 | spin_unlock_irqrestore(&psw->lock, flags); | |
365 | return; | |
366 | } | |
367 | } | |
368 | ||
369 | /* Add the current task to the list */ | |
370 | tnode.task = this_task; | |
371 | tnode.wq = wq; | |
372 | list_add(&tnode.llink, lsthead); | |
373 | ||
374 | spin_unlock_irqrestore(&psw->lock, flags); | |
375 | ||
376 | /* Do really wake up now */ | |
0ccf831c | 377 | wake_up_nested(wq, 1 + wake_nests); |
1da177e4 LT |
378 | |
379 | /* Remove the current task from the list */ | |
380 | spin_lock_irqsave(&psw->lock, flags); | |
381 | list_del(&tnode.llink); | |
382 | spin_unlock_irqrestore(&psw->lock, flags); | |
383 | } | |
384 | ||
1da177e4 | 385 | /* |
7699acd1 DL |
386 | * This function unregister poll callbacks from the associated file descriptor. |
387 | * Since this must be called without holding "ep->lock" the atomic exchange trick | |
388 | * will protect us from multiple unregister. | |
1da177e4 | 389 | */ |
7699acd1 | 390 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 391 | { |
7699acd1 DL |
392 | int nwait; |
393 | struct list_head *lsthead = &epi->pwqlist; | |
394 | struct eppoll_entry *pwq; | |
1da177e4 | 395 | |
7699acd1 DL |
396 | /* This is called without locks, so we need the atomic exchange */ |
397 | nwait = xchg(&epi->nwait, 0); | |
1da177e4 | 398 | |
7699acd1 DL |
399 | if (nwait) { |
400 | while (!list_empty(lsthead)) { | |
401 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 402 | |
7699acd1 DL |
403 | list_del_init(&pwq->llink); |
404 | remove_wait_queue(pwq->whead, &pwq->wait); | |
405 | kmem_cache_free(pwq_cache, pwq); | |
406 | } | |
1da177e4 | 407 | } |
1da177e4 LT |
408 | } |
409 | ||
7699acd1 DL |
410 | /* |
411 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 412 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
413 | */ |
414 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
415 | { | |
7699acd1 DL |
416 | unsigned long flags; |
417 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
418 | |
419 | /* | |
7699acd1 DL |
420 | * Removes poll wait queue hooks. We _have_ to do this without holding |
421 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
422 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
423 | * queue head lock when unregistering the wait queue. The wakeup callback | |
424 | * will run by holding the wait queue head lock and will call our callback | |
425 | * that will try to get "ep->lock". | |
1da177e4 | 426 | */ |
7699acd1 | 427 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 428 | |
7699acd1 DL |
429 | /* Remove the current item from the list of epoll hooks */ |
430 | spin_lock(&file->f_ep_lock); | |
431 | if (ep_is_linked(&epi->fllink)) | |
432 | list_del_init(&epi->fllink); | |
433 | spin_unlock(&file->f_ep_lock); | |
1da177e4 | 434 | |
cdac75e6 | 435 | rb_erase(&epi->rbn, &ep->rbr); |
1da177e4 | 436 | |
c7ea7630 DL |
437 | spin_lock_irqsave(&ep->lock, flags); |
438 | if (ep_is_linked(&epi->rdllink)) | |
439 | list_del_init(&epi->rdllink); | |
440 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 441 | |
7699acd1 | 442 | /* At this point it is safe to free the eventpoll item */ |
c7ea7630 | 443 | kmem_cache_free(epi_cache, epi); |
1da177e4 | 444 | |
7ef9964e DL |
445 | atomic_dec(&ep->user->epoll_watches); |
446 | ||
c7ea7630 DL |
447 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_remove(%p, %p)\n", |
448 | current, ep, file)); | |
1da177e4 | 449 | |
c7ea7630 | 450 | return 0; |
1da177e4 LT |
451 | } |
452 | ||
7699acd1 | 453 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 454 | { |
7699acd1 DL |
455 | struct rb_node *rbp; |
456 | struct epitem *epi; | |
1da177e4 | 457 | |
7699acd1 DL |
458 | /* We need to release all tasks waiting for these file */ |
459 | if (waitqueue_active(&ep->poll_wait)) | |
460 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1da177e4 | 461 | |
7699acd1 DL |
462 | /* |
463 | * We need to lock this because we could be hit by | |
464 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 465 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
466 | * is on the way to be removed and no one has references to it |
467 | * anymore. The only hit might come from eventpoll_release_file() but | |
468 | * holding "epmutex" is sufficent here. | |
469 | */ | |
470 | mutex_lock(&epmutex); | |
1da177e4 LT |
471 | |
472 | /* | |
7699acd1 | 473 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 474 | */ |
7699acd1 DL |
475 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
476 | epi = rb_entry(rbp, struct epitem, rbn); | |
477 | ||
478 | ep_unregister_pollwait(ep, epi); | |
479 | } | |
1da177e4 LT |
480 | |
481 | /* | |
7699acd1 DL |
482 | * Walks through the whole tree by freeing each "struct epitem". At this |
483 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 484 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 485 | * us during this operation. So we can avoid the lock on "ep->lock". |
1da177e4 | 486 | */ |
c80544dc | 487 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
7699acd1 DL |
488 | epi = rb_entry(rbp, struct epitem, rbn); |
489 | ep_remove(ep, epi); | |
490 | } | |
1da177e4 | 491 | |
7699acd1 | 492 | mutex_unlock(&epmutex); |
d47de16c | 493 | mutex_destroy(&ep->mtx); |
7ef9964e DL |
494 | atomic_dec(&ep->user->epoll_devs); |
495 | free_uid(ep->user); | |
f0ee9aab | 496 | kfree(ep); |
7699acd1 | 497 | } |
1da177e4 | 498 | |
7699acd1 DL |
499 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
500 | { | |
501 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 502 | |
f0ee9aab | 503 | if (ep) |
7699acd1 | 504 | ep_free(ep); |
7699acd1 DL |
505 | |
506 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: close() ep=%p\n", current, ep)); | |
507 | return 0; | |
1da177e4 LT |
508 | } |
509 | ||
7699acd1 DL |
510 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
511 | { | |
512 | unsigned int pollflags = 0; | |
513 | unsigned long flags; | |
514 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 515 | |
7699acd1 DL |
516 | /* Insert inside our poll wait queue */ |
517 | poll_wait(file, &ep->poll_wait, wait); | |
518 | ||
519 | /* Check our condition */ | |
c7ea7630 | 520 | spin_lock_irqsave(&ep->lock, flags); |
7699acd1 DL |
521 | if (!list_empty(&ep->rdllist)) |
522 | pollflags = POLLIN | POLLRDNORM; | |
c7ea7630 | 523 | spin_unlock_irqrestore(&ep->lock, flags); |
7699acd1 DL |
524 | |
525 | return pollflags; | |
526 | } | |
527 | ||
528 | /* File callbacks that implement the eventpoll file behaviour */ | |
529 | static const struct file_operations eventpoll_fops = { | |
530 | .release = ep_eventpoll_release, | |
531 | .poll = ep_eventpoll_poll | |
532 | }; | |
533 | ||
534 | /* Fast test to see if the file is an evenpoll file */ | |
535 | static inline int is_file_epoll(struct file *f) | |
536 | { | |
537 | return f->f_op == &eventpoll_fops; | |
538 | } | |
b611967d DL |
539 | |
540 | /* | |
7699acd1 DL |
541 | * This is called from eventpoll_release() to unlink files from the eventpoll |
542 | * interface. We need to have this facility to cleanup correctly files that are | |
543 | * closed without being removed from the eventpoll interface. | |
b611967d | 544 | */ |
7699acd1 | 545 | void eventpoll_release_file(struct file *file) |
b611967d | 546 | { |
7699acd1 DL |
547 | struct list_head *lsthead = &file->f_ep_links; |
548 | struct eventpoll *ep; | |
549 | struct epitem *epi; | |
b611967d DL |
550 | |
551 | /* | |
7699acd1 DL |
552 | * We don't want to get "file->f_ep_lock" because it is not |
553 | * necessary. It is not necessary because we're in the "struct file" | |
554 | * cleanup path, and this means that noone is using this file anymore. | |
67647d0f DL |
555 | * So, for example, epoll_ctl() cannot hit here sicne if we reach this |
556 | * point, the file counter already went to zero and fget() would fail. | |
d47de16c | 557 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 558 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 559 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 560 | * from anywhere but ep_free(). |
b611967d | 561 | */ |
7699acd1 | 562 | mutex_lock(&epmutex); |
b611967d | 563 | |
7699acd1 DL |
564 | while (!list_empty(lsthead)) { |
565 | epi = list_first_entry(lsthead, struct epitem, fllink); | |
b611967d | 566 | |
7699acd1 DL |
567 | ep = epi->ep; |
568 | list_del_init(&epi->fllink); | |
d47de16c | 569 | mutex_lock(&ep->mtx); |
7699acd1 | 570 | ep_remove(ep, epi); |
d47de16c | 571 | mutex_unlock(&ep->mtx); |
b611967d DL |
572 | } |
573 | ||
7699acd1 | 574 | mutex_unlock(&epmutex); |
b611967d DL |
575 | } |
576 | ||
53d2be79 | 577 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 578 | { |
7ef9964e DL |
579 | int error; |
580 | struct user_struct *user; | |
581 | struct eventpoll *ep; | |
1da177e4 | 582 | |
7ef9964e DL |
583 | user = get_current_user(); |
584 | error = -EMFILE; | |
585 | if (unlikely(atomic_read(&user->epoll_devs) >= | |
586 | max_user_instances)) | |
587 | goto free_uid; | |
588 | error = -ENOMEM; | |
589 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
590 | if (unlikely(!ep)) | |
591 | goto free_uid; | |
1da177e4 | 592 | |
c7ea7630 | 593 | spin_lock_init(&ep->lock); |
d47de16c | 594 | mutex_init(&ep->mtx); |
1da177e4 LT |
595 | init_waitqueue_head(&ep->wq); |
596 | init_waitqueue_head(&ep->poll_wait); | |
597 | INIT_LIST_HEAD(&ep->rdllist); | |
598 | ep->rbr = RB_ROOT; | |
d47de16c | 599 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 600 | ep->user = user; |
1da177e4 | 601 | |
53d2be79 | 602 | *pep = ep; |
1da177e4 | 603 | |
53d2be79 | 604 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_alloc() ep=%p\n", |
1da177e4 LT |
605 | current, ep)); |
606 | return 0; | |
7ef9964e DL |
607 | |
608 | free_uid: | |
609 | free_uid(user); | |
610 | return error; | |
1da177e4 LT |
611 | } |
612 | ||
1da177e4 | 613 | /* |
c7ea7630 DL |
614 | * Search the file inside the eventpoll tree. The RB tree operations |
615 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
616 | * "mtx" held. | |
1da177e4 LT |
617 | */ |
618 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
619 | { | |
620 | int kcmp; | |
1da177e4 LT |
621 | struct rb_node *rbp; |
622 | struct epitem *epi, *epir = NULL; | |
623 | struct epoll_filefd ffd; | |
624 | ||
b030a4dd | 625 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
626 | for (rbp = ep->rbr.rb_node; rbp; ) { |
627 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 628 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
629 | if (kcmp > 0) |
630 | rbp = rbp->rb_right; | |
631 | else if (kcmp < 0) | |
632 | rbp = rbp->rb_left; | |
633 | else { | |
1da177e4 LT |
634 | epir = epi; |
635 | break; | |
636 | } | |
637 | } | |
1da177e4 LT |
638 | |
639 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_find(%p) -> %p\n", | |
640 | current, file, epir)); | |
641 | ||
642 | return epir; | |
643 | } | |
644 | ||
1da177e4 | 645 | /* |
7699acd1 DL |
646 | * This is the callback that is passed to the wait queue wakeup |
647 | * machanism. It is called by the stored file descriptors when they | |
648 | * have events to report. | |
1da177e4 | 649 | */ |
7699acd1 | 650 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 651 | { |
7699acd1 DL |
652 | int pwake = 0; |
653 | unsigned long flags; | |
654 | struct epitem *epi = ep_item_from_wait(wait); | |
655 | struct eventpoll *ep = epi->ep; | |
1da177e4 | 656 | |
7699acd1 DL |
657 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: poll_callback(%p) epi=%p ep=%p\n", |
658 | current, epi->ffd.file, epi, ep)); | |
1da177e4 | 659 | |
c7ea7630 | 660 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 661 | |
7699acd1 DL |
662 | /* |
663 | * If the event mask does not contain any poll(2) event, we consider the | |
664 | * descriptor to be disabled. This condition is likely the effect of the | |
665 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
666 | * until the next EPOLL_CTL_MOD will be issued. | |
667 | */ | |
668 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
669 | goto out_unlock; |
670 | ||
671 | /* | |
672 | * If we are trasfering events to userspace, we can hold no locks | |
673 | * (because we're accessing user memory, and because of linux f_op->poll() | |
674 | * semantics). All the events that happens during that period of time are | |
675 | * chained in ep->ovflist and requeued later on. | |
676 | */ | |
677 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
678 | if (epi->next == EP_UNACTIVE_PTR) { | |
679 | epi->next = ep->ovflist; | |
680 | ep->ovflist = epi; | |
681 | } | |
682 | goto out_unlock; | |
683 | } | |
1da177e4 | 684 | |
7699acd1 DL |
685 | /* If this file is already in the ready list we exit soon */ |
686 | if (ep_is_linked(&epi->rdllink)) | |
687 | goto is_linked; | |
688 | ||
689 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
690 | ||
691 | is_linked: | |
692 | /* | |
693 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
694 | * wait list. | |
695 | */ | |
696 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 697 | wake_up_locked(&ep->wq); |
7699acd1 DL |
698 | if (waitqueue_active(&ep->poll_wait)) |
699 | pwake++; | |
700 | ||
d47de16c | 701 | out_unlock: |
c7ea7630 | 702 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 703 | |
7699acd1 DL |
704 | /* We have to call this outside the lock */ |
705 | if (pwake) | |
706 | ep_poll_safewake(&psw, &ep->poll_wait); | |
707 | ||
708 | return 1; | |
709 | } | |
1da177e4 LT |
710 | |
711 | /* | |
712 | * This is the callback that is used to add our wait queue to the | |
713 | * target file wakeup lists. | |
714 | */ | |
715 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
716 | poll_table *pt) | |
717 | { | |
b030a4dd | 718 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
719 | struct eppoll_entry *pwq; |
720 | ||
e94b1766 | 721 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
722 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
723 | pwq->whead = whead; | |
724 | pwq->base = epi; | |
725 | add_wait_queue(whead, &pwq->wait); | |
726 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
727 | epi->nwait++; | |
728 | } else { | |
729 | /* We have to signal that an error occurred */ | |
730 | epi->nwait = -1; | |
731 | } | |
732 | } | |
733 | ||
1da177e4 LT |
734 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
735 | { | |
736 | int kcmp; | |
737 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
738 | struct epitem *epic; | |
739 | ||
740 | while (*p) { | |
741 | parent = *p; | |
742 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 743 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
744 | if (kcmp > 0) |
745 | p = &parent->rb_right; | |
746 | else | |
747 | p = &parent->rb_left; | |
748 | } | |
749 | rb_link_node(&epi->rbn, parent, p); | |
750 | rb_insert_color(&epi->rbn, &ep->rbr); | |
751 | } | |
752 | ||
c7ea7630 DL |
753 | /* |
754 | * Must be called with "mtx" held. | |
755 | */ | |
1da177e4 LT |
756 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
757 | struct file *tfile, int fd) | |
758 | { | |
759 | int error, revents, pwake = 0; | |
760 | unsigned long flags; | |
761 | struct epitem *epi; | |
762 | struct ep_pqueue epq; | |
763 | ||
7ef9964e DL |
764 | if (unlikely(atomic_read(&ep->user->epoll_watches) >= |
765 | max_user_watches)) | |
766 | return -ENOSPC; | |
e94b1766 | 767 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 768 | return -ENOMEM; |
1da177e4 LT |
769 | |
770 | /* Item initialization follow here ... */ | |
1da177e4 LT |
771 | INIT_LIST_HEAD(&epi->rdllink); |
772 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
773 | INIT_LIST_HEAD(&epi->pwqlist); |
774 | epi->ep = ep; | |
b030a4dd | 775 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 776 | epi->event = *event; |
1da177e4 | 777 | epi->nwait = 0; |
d47de16c | 778 | epi->next = EP_UNACTIVE_PTR; |
1da177e4 LT |
779 | |
780 | /* Initialize the poll table using the queue callback */ | |
781 | epq.epi = epi; | |
782 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
783 | ||
784 | /* | |
785 | * Attach the item to the poll hooks and get current event bits. | |
786 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
787 | * been increased by the caller of this function. Note that after |
788 | * this operation completes, the poll callback can start hitting | |
789 | * the new item. | |
1da177e4 LT |
790 | */ |
791 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
792 | ||
793 | /* | |
794 | * We have to check if something went wrong during the poll wait queue | |
795 | * install process. Namely an allocation for a wait queue failed due | |
796 | * high memory pressure. | |
797 | */ | |
7ef9964e | 798 | error = -ENOMEM; |
1da177e4 | 799 | if (epi->nwait < 0) |
7699acd1 | 800 | goto error_unregister; |
1da177e4 LT |
801 | |
802 | /* Add the current item to the list of active epoll hook for this file */ | |
803 | spin_lock(&tfile->f_ep_lock); | |
804 | list_add_tail(&epi->fllink, &tfile->f_ep_links); | |
805 | spin_unlock(&tfile->f_ep_lock); | |
806 | ||
c7ea7630 DL |
807 | /* |
808 | * Add the current item to the RB tree. All RB tree operations are | |
809 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
810 | */ | |
1da177e4 LT |
811 | ep_rbtree_insert(ep, epi); |
812 | ||
c7ea7630 DL |
813 | /* We have to drop the new item inside our item list to keep track of it */ |
814 | spin_lock_irqsave(&ep->lock, flags); | |
815 | ||
1da177e4 | 816 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 817 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 LT |
818 | list_add_tail(&epi->rdllink, &ep->rdllist); |
819 | ||
820 | /* Notify waiting tasks that events are available */ | |
821 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 822 | wake_up_locked(&ep->wq); |
1da177e4 LT |
823 | if (waitqueue_active(&ep->poll_wait)) |
824 | pwake++; | |
825 | } | |
826 | ||
c7ea7630 | 827 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 828 | |
7ef9964e DL |
829 | atomic_inc(&ep->user->epoll_watches); |
830 | ||
1da177e4 LT |
831 | /* We have to call this outside the lock */ |
832 | if (pwake) | |
833 | ep_poll_safewake(&psw, &ep->poll_wait); | |
834 | ||
835 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_insert(%p, %p, %d)\n", | |
836 | current, ep, tfile, fd)); | |
837 | ||
838 | return 0; | |
839 | ||
7699acd1 | 840 | error_unregister: |
1da177e4 LT |
841 | ep_unregister_pollwait(ep, epi); |
842 | ||
843 | /* | |
844 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
845 | * allocated wait queue. Note that we don't care about the ep->ovflist |
846 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
847 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 848 | */ |
c7ea7630 | 849 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 850 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 851 | list_del_init(&epi->rdllink); |
c7ea7630 | 852 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 853 | |
b030a4dd | 854 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 855 | |
1da177e4 LT |
856 | return error; |
857 | } | |
858 | ||
1da177e4 LT |
859 | /* |
860 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 861 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
862 | */ |
863 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
864 | { | |
865 | int pwake = 0; | |
866 | unsigned int revents; | |
867 | unsigned long flags; | |
868 | ||
869 | /* | |
870 | * Set the new event interest mask before calling f_op->poll(), otherwise | |
871 | * a potential race might occur. In fact if we do this operation inside | |
872 | * the lock, an event might happen between the f_op->poll() call and the | |
873 | * new event set registering. | |
874 | */ | |
875 | epi->event.events = event->events; | |
876 | ||
877 | /* | |
878 | * Get current event bits. We can safely use the file* here because | |
879 | * its usage count has been increased by the caller of this function. | |
880 | */ | |
881 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
882 | ||
c7ea7630 | 883 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 LT |
884 | |
885 | /* Copy the data member from inside the lock */ | |
886 | epi->event.data = event->data; | |
887 | ||
888 | /* | |
c7ea7630 | 889 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 890 | * list, push it inside. |
1da177e4 | 891 | */ |
c7ea7630 DL |
892 | if (revents & event->events) { |
893 | if (!ep_is_linked(&epi->rdllink)) { | |
894 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
895 | ||
896 | /* Notify waiting tasks that events are available */ | |
897 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 898 | wake_up_locked(&ep->wq); |
c7ea7630 DL |
899 | if (waitqueue_active(&ep->poll_wait)) |
900 | pwake++; | |
7699acd1 DL |
901 | } |
902 | } | |
c7ea7630 | 903 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 904 | |
7699acd1 DL |
905 | /* We have to call this outside the lock */ |
906 | if (pwake) | |
907 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1da177e4 | 908 | |
7699acd1 | 909 | return 0; |
1da177e4 LT |
910 | } |
911 | ||
d47de16c DL |
912 | static int ep_send_events(struct eventpoll *ep, struct epoll_event __user *events, |
913 | int maxevents) | |
1da177e4 | 914 | { |
6192bd53 | 915 | int eventcnt, error = -EFAULT, pwake = 0; |
1da177e4 | 916 | unsigned int revents; |
6192bd53 | 917 | unsigned long flags; |
d47de16c DL |
918 | struct epitem *epi, *nepi; |
919 | struct list_head txlist; | |
920 | ||
921 | INIT_LIST_HEAD(&txlist); | |
6192bd53 | 922 | |
d47de16c DL |
923 | /* |
924 | * We need to lock this because we could be hit by | |
925 | * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL). | |
926 | */ | |
927 | mutex_lock(&ep->mtx); | |
928 | ||
929 | /* | |
930 | * Steal the ready list, and re-init the original one to the | |
931 | * empty list. Also, set ep->ovflist to NULL so that events | |
932 | * happening while looping w/out locks, are not lost. We cannot | |
933 | * have the poll callback to queue directly on ep->rdllist, | |
934 | * because we are doing it in the loop below, in a lockless way. | |
935 | */ | |
c7ea7630 | 936 | spin_lock_irqsave(&ep->lock, flags); |
d47de16c DL |
937 | list_splice(&ep->rdllist, &txlist); |
938 | INIT_LIST_HEAD(&ep->rdllist); | |
939 | ep->ovflist = NULL; | |
c7ea7630 | 940 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
941 | |
942 | /* | |
943 | * We can loop without lock because this is a task private list. | |
6192bd53 | 944 | * We just splice'd out the ep->rdllist in ep_collect_ready_items(). |
d47de16c | 945 | * Items cannot vanish during the loop because we are holding "mtx". |
1da177e4 | 946 | */ |
d47de16c DL |
947 | for (eventcnt = 0; !list_empty(&txlist) && eventcnt < maxevents;) { |
948 | epi = list_first_entry(&txlist, struct epitem, rdllink); | |
949 | ||
950 | list_del_init(&epi->rdllink); | |
1da177e4 LT |
951 | |
952 | /* | |
953 | * Get the ready file event set. We can safely use the file | |
d47de16c DL |
954 | * because we are holding the "mtx" and this will guarantee |
955 | * that both the file and the item will not vanish. | |
1da177e4 LT |
956 | */ |
957 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
6192bd53 | 958 | revents &= epi->event.events; |
1da177e4 LT |
959 | |
960 | /* | |
6192bd53 DL |
961 | * Is the event mask intersect the caller-requested one, |
962 | * deliver the event to userspace. Again, we are holding | |
d47de16c DL |
963 | * "mtx", so no operations coming from userspace can change |
964 | * the item. | |
1da177e4 | 965 | */ |
6192bd53 DL |
966 | if (revents) { |
967 | if (__put_user(revents, | |
1da177e4 LT |
968 | &events[eventcnt].events) || |
969 | __put_user(epi->event.data, | |
970 | &events[eventcnt].data)) | |
6192bd53 | 971 | goto errxit; |
1da177e4 LT |
972 | if (epi->event.events & EPOLLONESHOT) |
973 | epi->event.events &= EP_PRIVATE_BITS; | |
974 | eventcnt++; | |
975 | } | |
1da177e4 | 976 | /* |
d47de16c DL |
977 | * At this point, noone can insert into ep->rdllist besides |
978 | * us. The epoll_ctl() callers are locked out by us holding | |
979 | * "mtx" and the poll callback will queue them in ep->ovflist. | |
1da177e4 | 980 | */ |
6192bd53 | 981 | if (!(epi->event.events & EPOLLET) && |
d47de16c DL |
982 | (revents & epi->event.events)) |
983 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1da177e4 | 984 | } |
6192bd53 | 985 | error = 0; |
1da177e4 | 986 | |
d47de16c | 987 | errxit: |
6192bd53 | 988 | |
c7ea7630 | 989 | spin_lock_irqsave(&ep->lock, flags); |
d47de16c DL |
990 | /* |
991 | * During the time we spent in the loop above, some other events | |
992 | * might have been queued by the poll callback. We re-insert them | |
f337b9c5 | 993 | * inside the main ready-list here. |
d47de16c DL |
994 | */ |
995 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
9ce209d6 DL |
996 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
997 | /* | |
998 | * If the above loop quit with errors, the epoll item might still | |
999 | * be linked to "txlist", and the list_splice() done below will | |
1000 | * take care of those cases. | |
1001 | */ | |
1002 | if (!ep_is_linked(&epi->rdllink)) | |
1003 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1004 | } | |
6192bd53 | 1005 | /* |
d47de16c DL |
1006 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after |
1007 | * releasing the lock, events will be queued in the normal way inside | |
1008 | * ep->rdllist. | |
6192bd53 | 1009 | */ |
d47de16c | 1010 | ep->ovflist = EP_UNACTIVE_PTR; |
6192bd53 | 1011 | |
d47de16c | 1012 | /* |
67647d0f DL |
1013 | * In case of error in the event-send loop, or in case the number of |
1014 | * ready events exceeds the userspace limit, we need to splice the | |
1015 | * "txlist" back inside ep->rdllist. | |
d47de16c DL |
1016 | */ |
1017 | list_splice(&txlist, &ep->rdllist); | |
1018 | ||
1019 | if (!list_empty(&ep->rdllist)) { | |
1da177e4 | 1020 | /* |
d47de16c | 1021 | * Wake up (if active) both the eventpoll wait list and the ->poll() |
67647d0f | 1022 | * wait list (delayed after we release the lock). |
1da177e4 LT |
1023 | */ |
1024 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1025 | wake_up_locked(&ep->wq); |
1da177e4 LT |
1026 | if (waitqueue_active(&ep->poll_wait)) |
1027 | pwake++; | |
6192bd53 | 1028 | } |
c7ea7630 | 1029 | spin_unlock_irqrestore(&ep->lock, flags); |
d47de16c DL |
1030 | |
1031 | mutex_unlock(&ep->mtx); | |
1da177e4 LT |
1032 | |
1033 | /* We have to call this outside the lock */ | |
1034 | if (pwake) | |
1035 | ep_poll_safewake(&psw, &ep->poll_wait); | |
6192bd53 DL |
1036 | |
1037 | return eventcnt == 0 ? error: eventcnt; | |
1da177e4 LT |
1038 | } |
1039 | ||
1da177e4 LT |
1040 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1041 | int maxevents, long timeout) | |
1042 | { | |
1043 | int res, eavail; | |
1044 | unsigned long flags; | |
1045 | long jtimeout; | |
1046 | wait_queue_t wait; | |
1047 | ||
1048 | /* | |
1049 | * Calculate the timeout by checking for the "infinite" value ( -1 ) | |
1050 | * and the overflow condition. The passed timeout is in milliseconds, | |
1051 | * that why (t * HZ) / 1000. | |
1052 | */ | |
e3306dd5 DL |
1053 | jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ? |
1054 | MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000; | |
1da177e4 LT |
1055 | |
1056 | retry: | |
c7ea7630 | 1057 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 LT |
1058 | |
1059 | res = 0; | |
1060 | if (list_empty(&ep->rdllist)) { | |
1061 | /* | |
1062 | * We don't have any available event to return to the caller. | |
1063 | * We need to sleep here, and we will be wake up by | |
1064 | * ep_poll_callback() when events will become available. | |
1065 | */ | |
1066 | init_waitqueue_entry(&wait, current); | |
d47de16c | 1067 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
3419b23a | 1068 | __add_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1069 | |
1070 | for (;;) { | |
1071 | /* | |
1072 | * We don't want to sleep if the ep_poll_callback() sends us | |
1073 | * a wakeup in between. That's why we set the task state | |
1074 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1075 | */ | |
1076 | set_current_state(TASK_INTERRUPTIBLE); | |
1077 | if (!list_empty(&ep->rdllist) || !jtimeout) | |
1078 | break; | |
1079 | if (signal_pending(current)) { | |
1080 | res = -EINTR; | |
1081 | break; | |
1082 | } | |
1083 | ||
c7ea7630 | 1084 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1085 | jtimeout = schedule_timeout(jtimeout); |
c7ea7630 | 1086 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1087 | } |
3419b23a | 1088 | __remove_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1089 | |
1090 | set_current_state(TASK_RUNNING); | |
1091 | } | |
1092 | ||
1093 | /* Is it worth to try to dig for events ? */ | |
1094 | eavail = !list_empty(&ep->rdllist); | |
1095 | ||
c7ea7630 | 1096 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1097 | |
1098 | /* | |
1099 | * Try to transfer events to user space. In case we get 0 events and | |
1100 | * there's still timeout left over, we go trying again in search of | |
1101 | * more luck. | |
1102 | */ | |
1103 | if (!res && eavail && | |
d47de16c | 1104 | !(res = ep_send_events(ep, events, maxevents)) && jtimeout) |
1da177e4 LT |
1105 | goto retry; |
1106 | ||
1107 | return res; | |
1108 | } | |
1109 | ||
7699acd1 | 1110 | /* |
523723bb | 1111 | * Open an eventpoll file descriptor. |
7699acd1 | 1112 | */ |
9fe5ad9c | 1113 | asmlinkage long sys_epoll_create1(int flags) |
7699acd1 DL |
1114 | { |
1115 | int error, fd = -1; | |
1116 | struct eventpoll *ep; | |
7699acd1 | 1117 | |
e38b36f3 UD |
1118 | /* Check the EPOLL_* constant for consistency. */ |
1119 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1120 | ||
a0998b50 UD |
1121 | if (flags & ~EPOLL_CLOEXEC) |
1122 | return -EINVAL; | |
1123 | ||
7699acd1 | 1124 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d)\n", |
9fe5ad9c | 1125 | current, flags)); |
7699acd1 DL |
1126 | |
1127 | /* | |
9fe5ad9c | 1128 | * Create the internal data structure ( "struct eventpoll" ). |
7699acd1 | 1129 | */ |
9fe5ad9c UD |
1130 | error = ep_alloc(&ep); |
1131 | if (error < 0) { | |
2030a42c | 1132 | fd = error; |
7699acd1 | 1133 | goto error_return; |
2030a42c | 1134 | } |
7699acd1 DL |
1135 | |
1136 | /* | |
1137 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 1138 | * a file structure and a free file descriptor. |
7699acd1 | 1139 | */ |
a0998b50 UD |
1140 | fd = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep, |
1141 | flags & O_CLOEXEC); | |
2030a42c AV |
1142 | if (fd < 0) |
1143 | ep_free(ep); | |
7ef9964e | 1144 | atomic_inc(&ep->user->epoll_devs); |
7699acd1 | 1145 | |
2030a42c | 1146 | error_return: |
7699acd1 | 1147 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", |
9fe5ad9c | 1148 | current, flags, fd)); |
7699acd1 DL |
1149 | |
1150 | return fd; | |
7699acd1 DL |
1151 | } |
1152 | ||
a0998b50 UD |
1153 | asmlinkage long sys_epoll_create(int size) |
1154 | { | |
9fe5ad9c UD |
1155 | if (size < 0) |
1156 | return -EINVAL; | |
1157 | ||
1158 | return sys_epoll_create1(0); | |
a0998b50 UD |
1159 | } |
1160 | ||
7699acd1 DL |
1161 | /* |
1162 | * The following function implements the controller interface for | |
1163 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 1164 | * file descriptors inside the interest set. |
7699acd1 DL |
1165 | */ |
1166 | asmlinkage long sys_epoll_ctl(int epfd, int op, int fd, | |
1167 | struct epoll_event __user *event) | |
1168 | { | |
1169 | int error; | |
1170 | struct file *file, *tfile; | |
1171 | struct eventpoll *ep; | |
1172 | struct epitem *epi; | |
1173 | struct epoll_event epds; | |
1174 | ||
1175 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p)\n", | |
1176 | current, epfd, op, fd, event)); | |
1177 | ||
1178 | error = -EFAULT; | |
1179 | if (ep_op_has_event(op) && | |
1180 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
1181 | goto error_return; | |
1182 | ||
1183 | /* Get the "struct file *" for the eventpoll file */ | |
1184 | error = -EBADF; | |
1185 | file = fget(epfd); | |
1186 | if (!file) | |
1187 | goto error_return; | |
1188 | ||
1189 | /* Get the "struct file *" for the target file */ | |
1190 | tfile = fget(fd); | |
1191 | if (!tfile) | |
1192 | goto error_fput; | |
1193 | ||
1194 | /* The target file descriptor must support poll */ | |
1195 | error = -EPERM; | |
1196 | if (!tfile->f_op || !tfile->f_op->poll) | |
1197 | goto error_tgt_fput; | |
1198 | ||
1199 | /* | |
1200 | * We have to check that the file structure underneath the file descriptor | |
1201 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
1202 | * adding an epoll file descriptor inside itself. | |
1203 | */ | |
1204 | error = -EINVAL; | |
1205 | if (file == tfile || !is_file_epoll(file)) | |
1206 | goto error_tgt_fput; | |
1207 | ||
1208 | /* | |
1209 | * At this point it is safe to assume that the "private_data" contains | |
1210 | * our own data structure. | |
1211 | */ | |
1212 | ep = file->private_data; | |
1213 | ||
d47de16c | 1214 | mutex_lock(&ep->mtx); |
7699acd1 | 1215 | |
67647d0f DL |
1216 | /* |
1217 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
1218 | * above, we can be sure to be able to use the item looked up by | |
1219 | * ep_find() till we release the mutex. | |
1220 | */ | |
7699acd1 DL |
1221 | epi = ep_find(ep, tfile, fd); |
1222 | ||
1223 | error = -EINVAL; | |
1224 | switch (op) { | |
1225 | case EPOLL_CTL_ADD: | |
1226 | if (!epi) { | |
1227 | epds.events |= POLLERR | POLLHUP; | |
1228 | ||
1229 | error = ep_insert(ep, &epds, tfile, fd); | |
1230 | } else | |
1231 | error = -EEXIST; | |
1232 | break; | |
1233 | case EPOLL_CTL_DEL: | |
1234 | if (epi) | |
1235 | error = ep_remove(ep, epi); | |
1236 | else | |
1237 | error = -ENOENT; | |
1238 | break; | |
1239 | case EPOLL_CTL_MOD: | |
1240 | if (epi) { | |
1241 | epds.events |= POLLERR | POLLHUP; | |
1242 | error = ep_modify(ep, epi, &epds); | |
1243 | } else | |
1244 | error = -ENOENT; | |
1245 | break; | |
1246 | } | |
d47de16c | 1247 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
1248 | |
1249 | error_tgt_fput: | |
1250 | fput(tfile); | |
1251 | error_fput: | |
1252 | fput(file); | |
1253 | error_return: | |
1254 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p) = %d\n", | |
1255 | current, epfd, op, fd, event, error)); | |
1256 | ||
1257 | return error; | |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1262 | * part of the user space epoll_wait(2). | |
1263 | */ | |
1264 | asmlinkage long sys_epoll_wait(int epfd, struct epoll_event __user *events, | |
1265 | int maxevents, int timeout) | |
1266 | { | |
1267 | int error; | |
1268 | struct file *file; | |
1269 | struct eventpoll *ep; | |
1270 | ||
1271 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d)\n", | |
1272 | current, epfd, events, maxevents, timeout)); | |
1273 | ||
1274 | /* The maximum number of event must be greater than zero */ | |
1275 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
1276 | return -EINVAL; | |
1277 | ||
1278 | /* Verify that the area passed by the user is writeable */ | |
1279 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { | |
1280 | error = -EFAULT; | |
1281 | goto error_return; | |
1282 | } | |
1283 | ||
1284 | /* Get the "struct file *" for the eventpoll file */ | |
1285 | error = -EBADF; | |
1286 | file = fget(epfd); | |
1287 | if (!file) | |
1288 | goto error_return; | |
1289 | ||
1290 | /* | |
1291 | * We have to check that the file structure underneath the fd | |
1292 | * the user passed to us _is_ an eventpoll file. | |
1293 | */ | |
1294 | error = -EINVAL; | |
1295 | if (!is_file_epoll(file)) | |
1296 | goto error_fput; | |
1297 | ||
1298 | /* | |
1299 | * At this point it is safe to assume that the "private_data" contains | |
1300 | * our own data structure. | |
1301 | */ | |
1302 | ep = file->private_data; | |
1303 | ||
1304 | /* Time to fish for events ... */ | |
1305 | error = ep_poll(ep, events, maxevents, timeout); | |
1306 | ||
1307 | error_fput: | |
1308 | fput(file); | |
1309 | error_return: | |
1310 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d) = %d\n", | |
1311 | current, epfd, events, maxevents, timeout, error)); | |
1312 | ||
1313 | return error; | |
1314 | } | |
1315 | ||
f3de272b | 1316 | #ifdef HAVE_SET_RESTORE_SIGMASK |
7699acd1 DL |
1317 | |
1318 | /* | |
1319 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1320 | * part of the user space epoll_pwait(2). | |
1321 | */ | |
1322 | asmlinkage long sys_epoll_pwait(int epfd, struct epoll_event __user *events, | |
1323 | int maxevents, int timeout, const sigset_t __user *sigmask, | |
1324 | size_t sigsetsize) | |
1325 | { | |
1326 | int error; | |
1327 | sigset_t ksigmask, sigsaved; | |
1328 | ||
1329 | /* | |
1330 | * If the caller wants a certain signal mask to be set during the wait, | |
1331 | * we apply it here. | |
1332 | */ | |
1333 | if (sigmask) { | |
1334 | if (sigsetsize != sizeof(sigset_t)) | |
1335 | return -EINVAL; | |
1336 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1337 | return -EFAULT; | |
1338 | sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP)); | |
1339 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1340 | } | |
1341 | ||
1342 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
1343 | ||
1344 | /* | |
1345 | * If we changed the signal mask, we need to restore the original one. | |
1346 | * In case we've got a signal while waiting, we do not restore the | |
1347 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
1348 | * the way back to userspace, before the signal mask is restored. | |
1349 | */ | |
1350 | if (sigmask) { | |
1351 | if (error == -EINTR) { | |
1352 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 1353 | sizeof(sigsaved)); |
4e4c22c7 | 1354 | set_restore_sigmask(); |
7699acd1 DL |
1355 | } else |
1356 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1357 | } | |
1358 | ||
1359 | return error; | |
1360 | } | |
1361 | ||
f3de272b | 1362 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
7699acd1 | 1363 | |
1da177e4 LT |
1364 | static int __init eventpoll_init(void) |
1365 | { | |
7ef9964e DL |
1366 | struct sysinfo si; |
1367 | ||
1368 | si_meminfo(&si); | |
1369 | max_user_instances = 128; | |
1370 | max_user_watches = (((si.totalram - si.totalhigh) / 32) << PAGE_SHIFT) / | |
1371 | EP_ITEM_COST; | |
1da177e4 LT |
1372 | |
1373 | /* Initialize the structure used to perform safe poll wait head wake ups */ | |
1374 | ep_poll_safewake_init(&psw); | |
1375 | ||
1376 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1377 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
1378 | 0, SLAB_HWCACHE_ALIGN|EPI_SLAB_DEBUG|SLAB_PANIC, | |
20c2df83 | 1379 | NULL); |
1da177e4 LT |
1380 | |
1381 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1382 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
1383 | sizeof(struct eppoll_entry), 0, | |
20c2df83 | 1384 | EPI_SLAB_DEBUG|SLAB_PANIC, NULL); |
1da177e4 | 1385 | |
1da177e4 | 1386 | return 0; |
1da177e4 | 1387 | } |
cea69241 | 1388 | fs_initcall(eventpoll_init); |