Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
5071f97e DL |
2 | * fs/eventpoll.c (Efficient event retrieval implementation) |
3 | * Copyright (C) 2001,...,2009 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 | /* Epoll private bits inside the event mask */ |
75 | #define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET) | |
76 | ||
5071f97e DL |
77 | /* Maximum number of nesting allowed inside epoll sets */ |
78 | #define EP_MAX_NESTS 4 | |
1da177e4 | 79 | |
e3306dd5 DL |
80 | /* Maximum msec timeout value storeable in a long int */ |
81 | #define EP_MAX_MSTIMEO min(1000ULL * MAX_SCHEDULE_TIMEOUT / HZ, (LONG_MAX - 999ULL) / HZ) | |
82 | ||
b611967d DL |
83 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
84 | ||
d47de16c DL |
85 | #define EP_UNACTIVE_PTR ((void *) -1L) |
86 | ||
7ef9964e DL |
87 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
88 | ||
1da177e4 LT |
89 | struct epoll_filefd { |
90 | struct file *file; | |
91 | int fd; | |
92 | }; | |
93 | ||
94 | /* | |
5071f97e DL |
95 | * Structure used to track possible nested calls, for too deep recursions |
96 | * and loop cycles. | |
1da177e4 | 97 | */ |
5071f97e | 98 | struct nested_call_node { |
1da177e4 | 99 | struct list_head llink; |
5071f97e | 100 | void *cookie; |
abff55ce | 101 | int cpu; |
1da177e4 LT |
102 | }; |
103 | ||
104 | /* | |
5071f97e DL |
105 | * This structure is used as collector for nested calls, to check for |
106 | * maximum recursion dept and loop cycles. | |
1da177e4 | 107 | */ |
5071f97e DL |
108 | struct nested_calls { |
109 | struct list_head tasks_call_list; | |
1da177e4 LT |
110 | spinlock_t lock; |
111 | }; | |
112 | ||
d47de16c DL |
113 | /* |
114 | * Each file descriptor added to the eventpoll interface will | |
115 | * have an entry of this type linked to the "rbr" RB tree. | |
116 | */ | |
117 | struct epitem { | |
67647d0f | 118 | /* RB tree node used to link this structure to the eventpoll RB tree */ |
d47de16c DL |
119 | struct rb_node rbn; |
120 | ||
121 | /* List header used to link this structure to the eventpoll ready list */ | |
122 | struct list_head rdllink; | |
123 | ||
c7ea7630 DL |
124 | /* |
125 | * Works together "struct eventpoll"->ovflist in keeping the | |
126 | * single linked chain of items. | |
127 | */ | |
128 | struct epitem *next; | |
129 | ||
d47de16c DL |
130 | /* The file descriptor information this item refers to */ |
131 | struct epoll_filefd ffd; | |
132 | ||
133 | /* Number of active wait queue attached to poll operations */ | |
134 | int nwait; | |
135 | ||
136 | /* List containing poll wait queues */ | |
137 | struct list_head pwqlist; | |
138 | ||
139 | /* The "container" of this item */ | |
140 | struct eventpoll *ep; | |
141 | ||
d47de16c DL |
142 | /* List header used to link this item to the "struct file" items list */ |
143 | struct list_head fllink; | |
144 | ||
c7ea7630 DL |
145 | /* The structure that describe the interested events and the source fd */ |
146 | struct epoll_event event; | |
d47de16c DL |
147 | }; |
148 | ||
1da177e4 LT |
149 | /* |
150 | * This structure is stored inside the "private_data" member of the file | |
151 | * structure and rapresent the main data sructure for the eventpoll | |
152 | * interface. | |
153 | */ | |
154 | struct eventpoll { | |
155 | /* Protect the this structure access */ | |
c7ea7630 | 156 | spinlock_t lock; |
1da177e4 LT |
157 | |
158 | /* | |
d47de16c DL |
159 | * This mutex is used to ensure that files are not removed |
160 | * while epoll is using them. This is held during the event | |
161 | * collection loop, the file cleanup path, the epoll file exit | |
162 | * code and the ctl operations. | |
1da177e4 | 163 | */ |
d47de16c | 164 | struct mutex mtx; |
1da177e4 LT |
165 | |
166 | /* Wait queue used by sys_epoll_wait() */ | |
167 | wait_queue_head_t wq; | |
168 | ||
169 | /* Wait queue used by file->poll() */ | |
170 | wait_queue_head_t poll_wait; | |
171 | ||
172 | /* List of ready file descriptors */ | |
173 | struct list_head rdllist; | |
174 | ||
67647d0f | 175 | /* RB tree root used to store monitored fd structs */ |
1da177e4 | 176 | struct rb_root rbr; |
d47de16c DL |
177 | |
178 | /* | |
179 | * This is a single linked list that chains all the "struct epitem" that | |
180 | * happened while transfering ready events to userspace w/out | |
181 | * holding ->lock. | |
182 | */ | |
183 | struct epitem *ovflist; | |
7ef9964e DL |
184 | |
185 | /* The user that created the eventpoll descriptor */ | |
186 | struct user_struct *user; | |
1da177e4 LT |
187 | }; |
188 | ||
189 | /* Wait structure used by the poll hooks */ | |
190 | struct eppoll_entry { | |
191 | /* List header used to link this structure to the "struct epitem" */ | |
192 | struct list_head llink; | |
193 | ||
194 | /* The "base" pointer is set to the container "struct epitem" */ | |
4f0989db | 195 | struct epitem *base; |
1da177e4 LT |
196 | |
197 | /* | |
198 | * Wait queue item that will be linked to the target file wait | |
199 | * queue head. | |
200 | */ | |
201 | wait_queue_t wait; | |
202 | ||
203 | /* The wait queue head that linked the "wait" wait queue item */ | |
204 | wait_queue_head_t *whead; | |
205 | }; | |
206 | ||
1da177e4 LT |
207 | /* Wrapper struct used by poll queueing */ |
208 | struct ep_pqueue { | |
209 | poll_table pt; | |
210 | struct epitem *epi; | |
211 | }; | |
212 | ||
5071f97e DL |
213 | /* Used by the ep_send_events() function as callback private data */ |
214 | struct ep_send_events_data { | |
215 | int maxevents; | |
216 | struct epoll_event __user *events; | |
217 | }; | |
218 | ||
7ef9964e DL |
219 | /* |
220 | * Configuration options available inside /proc/sys/fs/epoll/ | |
221 | */ | |
7ef9964e DL |
222 | /* Maximum number of epoll watched descriptors, per user */ |
223 | static int max_user_watches __read_mostly; | |
224 | ||
1da177e4 | 225 | /* |
d47de16c | 226 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 227 | */ |
7ef9964e | 228 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 229 | |
5071f97e DL |
230 | /* Used for safe wake up implementation */ |
231 | static struct nested_calls poll_safewake_ncalls; | |
232 | ||
233 | /* Used to call file's f_op->poll() under the nested calls boundaries */ | |
234 | static struct nested_calls poll_readywalk_ncalls; | |
1da177e4 LT |
235 | |
236 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 237 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
238 | |
239 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 240 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 241 | |
7ef9964e DL |
242 | #ifdef CONFIG_SYSCTL |
243 | ||
244 | #include <linux/sysctl.h> | |
245 | ||
246 | static int zero; | |
247 | ||
248 | ctl_table epoll_table[] = { | |
7ef9964e DL |
249 | { |
250 | .procname = "max_user_watches", | |
251 | .data = &max_user_watches, | |
252 | .maxlen = sizeof(int), | |
253 | .mode = 0644, | |
254 | .proc_handler = &proc_dointvec_minmax, | |
255 | .extra1 = &zero, | |
256 | }, | |
257 | { .ctl_name = 0 } | |
258 | }; | |
259 | #endif /* CONFIG_SYSCTL */ | |
260 | ||
b030a4dd | 261 | |
67647d0f | 262 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
263 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
264 | struct file *file, int fd) | |
265 | { | |
266 | ffd->file = file; | |
267 | ffd->fd = fd; | |
268 | } | |
269 | ||
67647d0f | 270 | /* Compare RB tree keys */ |
b030a4dd PE |
271 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
272 | struct epoll_filefd *p2) | |
273 | { | |
274 | return (p1->file > p2->file ? +1: | |
275 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
276 | } | |
277 | ||
b030a4dd PE |
278 | /* Tells us if the item is currently linked */ |
279 | static inline int ep_is_linked(struct list_head *p) | |
280 | { | |
281 | return !list_empty(p); | |
282 | } | |
283 | ||
284 | /* Get the "struct epitem" from a wait queue pointer */ | |
cdac75e6 | 285 | static inline struct epitem *ep_item_from_wait(wait_queue_t *p) |
b030a4dd PE |
286 | { |
287 | return container_of(p, struct eppoll_entry, wait)->base; | |
288 | } | |
289 | ||
290 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 291 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
292 | { |
293 | return container_of(p, struct ep_pqueue, pt)->epi; | |
294 | } | |
295 | ||
296 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 297 | static inline int ep_op_has_event(int op) |
b030a4dd PE |
298 | { |
299 | return op != EPOLL_CTL_DEL; | |
300 | } | |
301 | ||
1da177e4 | 302 | /* Initialize the poll safe wake up structure */ |
5071f97e | 303 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 304 | { |
5071f97e DL |
305 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
306 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
307 | } |
308 | ||
5071f97e DL |
309 | /** |
310 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
311 | * that the recursion limit is not exceeded, and that | |
312 | * the same nested call (by the meaning of same cookie) is | |
313 | * no re-entered. | |
314 | * | |
315 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
316 | * @max_nests: Maximum number of allowed nesting calls. | |
317 | * @nproc: Nested call core function pointer. | |
318 | * @priv: Opaque data to be passed to the @nproc callback. | |
319 | * @cookie: Cookie to be used to identify this nested call. | |
320 | * | |
321 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
322 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 323 | */ |
5071f97e DL |
324 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
325 | int (*nproc)(void *, void *, int), void *priv, | |
326 | void *cookie) | |
1da177e4 | 327 | { |
5071f97e | 328 | int error, call_nests = 0; |
1da177e4 | 329 | unsigned long flags; |
abff55ce | 330 | int this_cpu = get_cpu(); |
5071f97e DL |
331 | struct list_head *lsthead = &ncalls->tasks_call_list; |
332 | struct nested_call_node *tncur; | |
333 | struct nested_call_node tnode; | |
1da177e4 | 334 | |
5071f97e | 335 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 336 | |
5071f97e DL |
337 | /* |
338 | * Try to see if the current task is already inside this wakeup call. | |
339 | * We use a list here, since the population inside this set is always | |
340 | * very much limited. | |
341 | */ | |
b70c3940 | 342 | list_for_each_entry(tncur, lsthead, llink) { |
abff55ce | 343 | if (tncur->cpu == this_cpu && |
5071f97e | 344 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
1da177e4 LT |
345 | /* |
346 | * Ops ... loop detected or maximum nest level reached. | |
347 | * We abort this wake by breaking the cycle itself. | |
348 | */ | |
abff55ce TB |
349 | error = -1; |
350 | goto out_unlock; | |
1da177e4 LT |
351 | } |
352 | } | |
353 | ||
5071f97e | 354 | /* Add the current task and cookie to the list */ |
abff55ce | 355 | tnode.cpu = this_cpu; |
5071f97e | 356 | tnode.cookie = cookie; |
1da177e4 LT |
357 | list_add(&tnode.llink, lsthead); |
358 | ||
5071f97e | 359 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 360 | |
5071f97e DL |
361 | /* Call the nested function */ |
362 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
363 | |
364 | /* Remove the current task from the list */ | |
5071f97e | 365 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 366 | list_del(&tnode.llink); |
abff55ce | 367 | out_unlock: |
5071f97e DL |
368 | spin_unlock_irqrestore(&ncalls->lock, flags); |
369 | ||
abff55ce | 370 | put_cpu(); |
5071f97e DL |
371 | return error; |
372 | } | |
373 | ||
2dfa4eea DL |
374 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
375 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
376 | unsigned long events, int subclass) | |
377 | { | |
378 | unsigned long flags; | |
379 | ||
380 | spin_lock_irqsave_nested(&wqueue->lock, flags, subclass); | |
381 | wake_up_locked_poll(wqueue, events); | |
382 | spin_unlock_irqrestore(&wqueue->lock, flags); | |
383 | } | |
384 | #else | |
385 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
386 | unsigned long events, int subclass) | |
387 | { | |
388 | wake_up_poll(wqueue, events); | |
389 | } | |
390 | #endif | |
391 | ||
5071f97e DL |
392 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
393 | { | |
2dfa4eea DL |
394 | ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN, |
395 | 1 + call_nests); | |
5071f97e DL |
396 | return 0; |
397 | } | |
398 | ||
399 | /* | |
400 | * Perform a safe wake up of the poll wait list. The problem is that | |
401 | * with the new callback'd wake up system, it is possible that the | |
402 | * poll callback is reentered from inside the call to wake_up() done | |
403 | * on the poll wait queue head. The rule is that we cannot reenter the | |
404 | * wake up code from the same task more than EP_MAX_NESTS times, | |
405 | * and we cannot reenter the same wait queue head at all. This will | |
406 | * enable to have a hierarchy of epoll file descriptor of no more than | |
407 | * EP_MAX_NESTS deep. | |
408 | */ | |
409 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
410 | { | |
411 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, | |
412 | ep_poll_wakeup_proc, NULL, wq); | |
1da177e4 LT |
413 | } |
414 | ||
1da177e4 | 415 | /* |
d1bc90dd TB |
416 | * This function unregisters poll callbacks from the associated file |
417 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
418 | * ep_free). | |
1da177e4 | 419 | */ |
7699acd1 | 420 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 421 | { |
7699acd1 DL |
422 | struct list_head *lsthead = &epi->pwqlist; |
423 | struct eppoll_entry *pwq; | |
1da177e4 | 424 | |
d1bc90dd TB |
425 | while (!list_empty(lsthead)) { |
426 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 427 | |
d1bc90dd TB |
428 | list_del(&pwq->llink); |
429 | remove_wait_queue(pwq->whead, &pwq->wait); | |
430 | kmem_cache_free(pwq_cache, pwq); | |
1da177e4 | 431 | } |
1da177e4 LT |
432 | } |
433 | ||
5071f97e DL |
434 | /** |
435 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
436 | * the scan code, to call f_op->poll(). Also allows for | |
437 | * O(NumReady) performance. | |
438 | * | |
439 | * @ep: Pointer to the epoll private data structure. | |
440 | * @sproc: Pointer to the scan callback. | |
441 | * @priv: Private opaque data passed to the @sproc callback. | |
442 | * | |
443 | * Returns: The same integer error code returned by the @sproc callback. | |
444 | */ | |
445 | static int ep_scan_ready_list(struct eventpoll *ep, | |
446 | int (*sproc)(struct eventpoll *, | |
447 | struct list_head *, void *), | |
448 | void *priv) | |
449 | { | |
450 | int error, pwake = 0; | |
451 | unsigned long flags; | |
452 | struct epitem *epi, *nepi; | |
296e236e | 453 | LIST_HEAD(txlist); |
5071f97e DL |
454 | |
455 | /* | |
456 | * We need to lock this because we could be hit by | |
e057e15f | 457 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e DL |
458 | */ |
459 | mutex_lock(&ep->mtx); | |
460 | ||
461 | /* | |
462 | * Steal the ready list, and re-init the original one to the | |
463 | * empty list. Also, set ep->ovflist to NULL so that events | |
464 | * happening while looping w/out locks, are not lost. We cannot | |
465 | * have the poll callback to queue directly on ep->rdllist, | |
466 | * because we want the "sproc" callback to be able to do it | |
467 | * in a lockless way. | |
468 | */ | |
469 | spin_lock_irqsave(&ep->lock, flags); | |
296e236e | 470 | list_splice_init(&ep->rdllist, &txlist); |
5071f97e DL |
471 | ep->ovflist = NULL; |
472 | spin_unlock_irqrestore(&ep->lock, flags); | |
473 | ||
474 | /* | |
475 | * Now call the callback function. | |
476 | */ | |
477 | error = (*sproc)(ep, &txlist, priv); | |
478 | ||
479 | spin_lock_irqsave(&ep->lock, flags); | |
480 | /* | |
481 | * During the time we spent inside the "sproc" callback, some | |
482 | * other events might have been queued by the poll callback. | |
483 | * We re-insert them inside the main ready-list here. | |
484 | */ | |
485 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
486 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { | |
487 | /* | |
488 | * We need to check if the item is already in the list. | |
489 | * During the "sproc" callback execution time, items are | |
490 | * queued into ->ovflist but the "txlist" might already | |
491 | * contain them, and the list_splice() below takes care of them. | |
492 | */ | |
493 | if (!ep_is_linked(&epi->rdllink)) | |
494 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
495 | } | |
496 | /* | |
497 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
498 | * releasing the lock, events will be queued in the normal way inside | |
499 | * ep->rdllist. | |
500 | */ | |
501 | ep->ovflist = EP_UNACTIVE_PTR; | |
502 | ||
503 | /* | |
504 | * Quickly re-inject items left on "txlist". | |
505 | */ | |
506 | list_splice(&txlist, &ep->rdllist); | |
507 | ||
508 | if (!list_empty(&ep->rdllist)) { | |
509 | /* | |
296e236e DL |
510 | * Wake up (if active) both the eventpoll wait list and |
511 | * the ->poll() wait list (delayed after we release the lock). | |
5071f97e DL |
512 | */ |
513 | if (waitqueue_active(&ep->wq)) | |
514 | wake_up_locked(&ep->wq); | |
515 | if (waitqueue_active(&ep->poll_wait)) | |
516 | pwake++; | |
517 | } | |
518 | spin_unlock_irqrestore(&ep->lock, flags); | |
519 | ||
520 | mutex_unlock(&ep->mtx); | |
521 | ||
522 | /* We have to call this outside the lock */ | |
523 | if (pwake) | |
524 | ep_poll_safewake(&ep->poll_wait); | |
525 | ||
526 | return error; | |
527 | } | |
528 | ||
7699acd1 DL |
529 | /* |
530 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 531 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
532 | */ |
533 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
534 | { | |
7699acd1 DL |
535 | unsigned long flags; |
536 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
537 | |
538 | /* | |
7699acd1 DL |
539 | * Removes poll wait queue hooks. We _have_ to do this without holding |
540 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
541 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
542 | * queue head lock when unregistering the wait queue. The wakeup callback | |
543 | * will run by holding the wait queue head lock and will call our callback | |
544 | * that will try to get "ep->lock". | |
1da177e4 | 545 | */ |
7699acd1 | 546 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 547 | |
7699acd1 | 548 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 549 | spin_lock(&file->f_lock); |
7699acd1 DL |
550 | if (ep_is_linked(&epi->fllink)) |
551 | list_del_init(&epi->fllink); | |
68499914 | 552 | spin_unlock(&file->f_lock); |
1da177e4 | 553 | |
cdac75e6 | 554 | rb_erase(&epi->rbn, &ep->rbr); |
1da177e4 | 555 | |
c7ea7630 DL |
556 | spin_lock_irqsave(&ep->lock, flags); |
557 | if (ep_is_linked(&epi->rdllink)) | |
558 | list_del_init(&epi->rdllink); | |
559 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 560 | |
7699acd1 | 561 | /* At this point it is safe to free the eventpoll item */ |
c7ea7630 | 562 | kmem_cache_free(epi_cache, epi); |
1da177e4 | 563 | |
7ef9964e DL |
564 | atomic_dec(&ep->user->epoll_watches); |
565 | ||
c7ea7630 | 566 | return 0; |
1da177e4 LT |
567 | } |
568 | ||
7699acd1 | 569 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 570 | { |
7699acd1 DL |
571 | struct rb_node *rbp; |
572 | struct epitem *epi; | |
1da177e4 | 573 | |
7699acd1 DL |
574 | /* We need to release all tasks waiting for these file */ |
575 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 576 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 577 | |
7699acd1 DL |
578 | /* |
579 | * We need to lock this because we could be hit by | |
580 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 581 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
582 | * is on the way to be removed and no one has references to it |
583 | * anymore. The only hit might come from eventpoll_release_file() but | |
584 | * holding "epmutex" is sufficent here. | |
585 | */ | |
586 | mutex_lock(&epmutex); | |
1da177e4 LT |
587 | |
588 | /* | |
7699acd1 | 589 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 590 | */ |
7699acd1 DL |
591 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
592 | epi = rb_entry(rbp, struct epitem, rbn); | |
593 | ||
594 | ep_unregister_pollwait(ep, epi); | |
595 | } | |
1da177e4 LT |
596 | |
597 | /* | |
7699acd1 DL |
598 | * Walks through the whole tree by freeing each "struct epitem". At this |
599 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 600 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 601 | * us during this operation. So we can avoid the lock on "ep->lock". |
1da177e4 | 602 | */ |
c80544dc | 603 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
7699acd1 DL |
604 | epi = rb_entry(rbp, struct epitem, rbn); |
605 | ep_remove(ep, epi); | |
606 | } | |
1da177e4 | 607 | |
7699acd1 | 608 | mutex_unlock(&epmutex); |
d47de16c | 609 | mutex_destroy(&ep->mtx); |
7ef9964e | 610 | free_uid(ep->user); |
f0ee9aab | 611 | kfree(ep); |
7699acd1 | 612 | } |
1da177e4 | 613 | |
7699acd1 DL |
614 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
615 | { | |
616 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 617 | |
f0ee9aab | 618 | if (ep) |
7699acd1 | 619 | ep_free(ep); |
7699acd1 | 620 | |
7699acd1 | 621 | return 0; |
1da177e4 LT |
622 | } |
623 | ||
296e236e DL |
624 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
625 | void *priv) | |
5071f97e DL |
626 | { |
627 | struct epitem *epi, *tmp; | |
628 | ||
629 | list_for_each_entry_safe(epi, tmp, head, rdllink) { | |
630 | if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) & | |
631 | epi->event.events) | |
632 | return POLLIN | POLLRDNORM; | |
296e236e | 633 | else { |
5071f97e DL |
634 | /* |
635 | * Item has been dropped into the ready list by the poll | |
636 | * callback, but it's not actually ready, as far as | |
637 | * caller requested events goes. We can remove it here. | |
638 | */ | |
639 | list_del_init(&epi->rdllink); | |
296e236e | 640 | } |
5071f97e DL |
641 | } |
642 | ||
643 | return 0; | |
644 | } | |
645 | ||
646 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) | |
647 | { | |
648 | return ep_scan_ready_list(priv, ep_read_events_proc, NULL); | |
649 | } | |
650 | ||
7699acd1 DL |
651 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
652 | { | |
5071f97e | 653 | int pollflags; |
7699acd1 | 654 | struct eventpoll *ep = file->private_data; |
1da177e4 | 655 | |
7699acd1 DL |
656 | /* Insert inside our poll wait queue */ |
657 | poll_wait(file, &ep->poll_wait, wait); | |
658 | ||
5071f97e DL |
659 | /* |
660 | * Proceed to find out if wanted events are really available inside | |
661 | * the ready list. This need to be done under ep_call_nested() | |
662 | * supervision, since the call to f_op->poll() done on listed files | |
663 | * could re-enter here. | |
664 | */ | |
665 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, | |
666 | ep_poll_readyevents_proc, ep, ep); | |
7699acd1 | 667 | |
296e236e | 668 | return pollflags != -1 ? pollflags : 0; |
7699acd1 DL |
669 | } |
670 | ||
671 | /* File callbacks that implement the eventpoll file behaviour */ | |
672 | static const struct file_operations eventpoll_fops = { | |
673 | .release = ep_eventpoll_release, | |
674 | .poll = ep_eventpoll_poll | |
675 | }; | |
676 | ||
677 | /* Fast test to see if the file is an evenpoll file */ | |
678 | static inline int is_file_epoll(struct file *f) | |
679 | { | |
680 | return f->f_op == &eventpoll_fops; | |
681 | } | |
b611967d DL |
682 | |
683 | /* | |
7699acd1 DL |
684 | * This is called from eventpoll_release() to unlink files from the eventpoll |
685 | * interface. We need to have this facility to cleanup correctly files that are | |
686 | * closed without being removed from the eventpoll interface. | |
b611967d | 687 | */ |
7699acd1 | 688 | void eventpoll_release_file(struct file *file) |
b611967d | 689 | { |
7699acd1 DL |
690 | struct list_head *lsthead = &file->f_ep_links; |
691 | struct eventpoll *ep; | |
692 | struct epitem *epi; | |
b611967d DL |
693 | |
694 | /* | |
68499914 | 695 | * We don't want to get "file->f_lock" because it is not |
7699acd1 DL |
696 | * necessary. It is not necessary because we're in the "struct file" |
697 | * cleanup path, and this means that noone is using this file anymore. | |
5071f97e | 698 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 699 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 700 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 701 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 702 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 703 | * from anywhere but ep_free(). |
68499914 JC |
704 | * |
705 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 706 | */ |
7699acd1 | 707 | mutex_lock(&epmutex); |
b611967d | 708 | |
7699acd1 DL |
709 | while (!list_empty(lsthead)) { |
710 | epi = list_first_entry(lsthead, struct epitem, fllink); | |
b611967d | 711 | |
7699acd1 DL |
712 | ep = epi->ep; |
713 | list_del_init(&epi->fllink); | |
d47de16c | 714 | mutex_lock(&ep->mtx); |
7699acd1 | 715 | ep_remove(ep, epi); |
d47de16c | 716 | mutex_unlock(&ep->mtx); |
b611967d DL |
717 | } |
718 | ||
7699acd1 | 719 | mutex_unlock(&epmutex); |
b611967d DL |
720 | } |
721 | ||
53d2be79 | 722 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 723 | { |
7ef9964e DL |
724 | int error; |
725 | struct user_struct *user; | |
726 | struct eventpoll *ep; | |
1da177e4 | 727 | |
7ef9964e | 728 | user = get_current_user(); |
7ef9964e DL |
729 | error = -ENOMEM; |
730 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
731 | if (unlikely(!ep)) | |
732 | goto free_uid; | |
1da177e4 | 733 | |
c7ea7630 | 734 | spin_lock_init(&ep->lock); |
d47de16c | 735 | mutex_init(&ep->mtx); |
1da177e4 LT |
736 | init_waitqueue_head(&ep->wq); |
737 | init_waitqueue_head(&ep->poll_wait); | |
738 | INIT_LIST_HEAD(&ep->rdllist); | |
739 | ep->rbr = RB_ROOT; | |
d47de16c | 740 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 741 | ep->user = user; |
1da177e4 | 742 | |
53d2be79 | 743 | *pep = ep; |
1da177e4 | 744 | |
1da177e4 | 745 | return 0; |
7ef9964e DL |
746 | |
747 | free_uid: | |
748 | free_uid(user); | |
749 | return error; | |
1da177e4 LT |
750 | } |
751 | ||
1da177e4 | 752 | /* |
c7ea7630 DL |
753 | * Search the file inside the eventpoll tree. The RB tree operations |
754 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
755 | * "mtx" held. | |
1da177e4 LT |
756 | */ |
757 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
758 | { | |
759 | int kcmp; | |
1da177e4 LT |
760 | struct rb_node *rbp; |
761 | struct epitem *epi, *epir = NULL; | |
762 | struct epoll_filefd ffd; | |
763 | ||
b030a4dd | 764 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
765 | for (rbp = ep->rbr.rb_node; rbp; ) { |
766 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 767 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
768 | if (kcmp > 0) |
769 | rbp = rbp->rb_right; | |
770 | else if (kcmp < 0) | |
771 | rbp = rbp->rb_left; | |
772 | else { | |
1da177e4 LT |
773 | epir = epi; |
774 | break; | |
775 | } | |
776 | } | |
1da177e4 | 777 | |
1da177e4 LT |
778 | return epir; |
779 | } | |
780 | ||
1da177e4 | 781 | /* |
7699acd1 DL |
782 | * This is the callback that is passed to the wait queue wakeup |
783 | * machanism. It is called by the stored file descriptors when they | |
784 | * have events to report. | |
1da177e4 | 785 | */ |
7699acd1 | 786 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 787 | { |
7699acd1 DL |
788 | int pwake = 0; |
789 | unsigned long flags; | |
790 | struct epitem *epi = ep_item_from_wait(wait); | |
791 | struct eventpoll *ep = epi->ep; | |
1da177e4 | 792 | |
c7ea7630 | 793 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 794 | |
7699acd1 DL |
795 | /* |
796 | * If the event mask does not contain any poll(2) event, we consider the | |
797 | * descriptor to be disabled. This condition is likely the effect of the | |
798 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
799 | * until the next EPOLL_CTL_MOD will be issued. | |
800 | */ | |
801 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
802 | goto out_unlock; |
803 | ||
2dfa4eea DL |
804 | /* |
805 | * Check the events coming with the callback. At this stage, not | |
806 | * every device reports the events in the "key" parameter of the | |
807 | * callback. We need to be able to handle both cases here, hence the | |
808 | * test for "key" != NULL before the event match test. | |
809 | */ | |
810 | if (key && !((unsigned long) key & epi->event.events)) | |
811 | goto out_unlock; | |
812 | ||
d47de16c DL |
813 | /* |
814 | * If we are trasfering events to userspace, we can hold no locks | |
815 | * (because we're accessing user memory, and because of linux f_op->poll() | |
816 | * semantics). All the events that happens during that period of time are | |
817 | * chained in ep->ovflist and requeued later on. | |
818 | */ | |
819 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
820 | if (epi->next == EP_UNACTIVE_PTR) { | |
821 | epi->next = ep->ovflist; | |
822 | ep->ovflist = epi; | |
823 | } | |
824 | goto out_unlock; | |
825 | } | |
1da177e4 | 826 | |
7699acd1 | 827 | /* If this file is already in the ready list we exit soon */ |
5071f97e DL |
828 | if (!ep_is_linked(&epi->rdllink)) |
829 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
7699acd1 | 830 | |
7699acd1 DL |
831 | /* |
832 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
833 | * wait list. | |
834 | */ | |
835 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 836 | wake_up_locked(&ep->wq); |
7699acd1 DL |
837 | if (waitqueue_active(&ep->poll_wait)) |
838 | pwake++; | |
839 | ||
d47de16c | 840 | out_unlock: |
c7ea7630 | 841 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 842 | |
7699acd1 DL |
843 | /* We have to call this outside the lock */ |
844 | if (pwake) | |
5071f97e | 845 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 DL |
846 | |
847 | return 1; | |
848 | } | |
1da177e4 LT |
849 | |
850 | /* | |
851 | * This is the callback that is used to add our wait queue to the | |
852 | * target file wakeup lists. | |
853 | */ | |
854 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
855 | poll_table *pt) | |
856 | { | |
b030a4dd | 857 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
858 | struct eppoll_entry *pwq; |
859 | ||
e94b1766 | 860 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
861 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
862 | pwq->whead = whead; | |
863 | pwq->base = epi; | |
864 | add_wait_queue(whead, &pwq->wait); | |
865 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
866 | epi->nwait++; | |
296e236e | 867 | } else { |
1da177e4 LT |
868 | /* We have to signal that an error occurred */ |
869 | epi->nwait = -1; | |
296e236e | 870 | } |
1da177e4 LT |
871 | } |
872 | ||
1da177e4 LT |
873 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
874 | { | |
875 | int kcmp; | |
876 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
877 | struct epitem *epic; | |
878 | ||
879 | while (*p) { | |
880 | parent = *p; | |
881 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 882 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
883 | if (kcmp > 0) |
884 | p = &parent->rb_right; | |
885 | else | |
886 | p = &parent->rb_left; | |
887 | } | |
888 | rb_link_node(&epi->rbn, parent, p); | |
889 | rb_insert_color(&epi->rbn, &ep->rbr); | |
890 | } | |
891 | ||
c7ea7630 DL |
892 | /* |
893 | * Must be called with "mtx" held. | |
894 | */ | |
1da177e4 LT |
895 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
896 | struct file *tfile, int fd) | |
897 | { | |
898 | int error, revents, pwake = 0; | |
899 | unsigned long flags; | |
900 | struct epitem *epi; | |
901 | struct ep_pqueue epq; | |
902 | ||
7ef9964e DL |
903 | if (unlikely(atomic_read(&ep->user->epoll_watches) >= |
904 | max_user_watches)) | |
905 | return -ENOSPC; | |
e94b1766 | 906 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 907 | return -ENOMEM; |
1da177e4 LT |
908 | |
909 | /* Item initialization follow here ... */ | |
1da177e4 LT |
910 | INIT_LIST_HEAD(&epi->rdllink); |
911 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
912 | INIT_LIST_HEAD(&epi->pwqlist); |
913 | epi->ep = ep; | |
b030a4dd | 914 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 915 | epi->event = *event; |
1da177e4 | 916 | epi->nwait = 0; |
d47de16c | 917 | epi->next = EP_UNACTIVE_PTR; |
1da177e4 LT |
918 | |
919 | /* Initialize the poll table using the queue callback */ | |
920 | epq.epi = epi; | |
921 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
922 | ||
923 | /* | |
924 | * Attach the item to the poll hooks and get current event bits. | |
925 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
926 | * been increased by the caller of this function. Note that after |
927 | * this operation completes, the poll callback can start hitting | |
928 | * the new item. | |
1da177e4 LT |
929 | */ |
930 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
931 | ||
932 | /* | |
933 | * We have to check if something went wrong during the poll wait queue | |
934 | * install process. Namely an allocation for a wait queue failed due | |
935 | * high memory pressure. | |
936 | */ | |
7ef9964e | 937 | error = -ENOMEM; |
1da177e4 | 938 | if (epi->nwait < 0) |
7699acd1 | 939 | goto error_unregister; |
1da177e4 LT |
940 | |
941 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 942 | spin_lock(&tfile->f_lock); |
1da177e4 | 943 | list_add_tail(&epi->fllink, &tfile->f_ep_links); |
68499914 | 944 | spin_unlock(&tfile->f_lock); |
1da177e4 | 945 | |
c7ea7630 DL |
946 | /* |
947 | * Add the current item to the RB tree. All RB tree operations are | |
948 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
949 | */ | |
1da177e4 LT |
950 | ep_rbtree_insert(ep, epi); |
951 | ||
c7ea7630 DL |
952 | /* We have to drop the new item inside our item list to keep track of it */ |
953 | spin_lock_irqsave(&ep->lock, flags); | |
954 | ||
1da177e4 | 955 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 956 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 LT |
957 | list_add_tail(&epi->rdllink, &ep->rdllist); |
958 | ||
959 | /* Notify waiting tasks that events are available */ | |
960 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 961 | wake_up_locked(&ep->wq); |
1da177e4 LT |
962 | if (waitqueue_active(&ep->poll_wait)) |
963 | pwake++; | |
964 | } | |
965 | ||
c7ea7630 | 966 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 967 | |
7ef9964e DL |
968 | atomic_inc(&ep->user->epoll_watches); |
969 | ||
1da177e4 LT |
970 | /* We have to call this outside the lock */ |
971 | if (pwake) | |
5071f97e | 972 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 973 | |
1da177e4 LT |
974 | return 0; |
975 | ||
7699acd1 | 976 | error_unregister: |
1da177e4 LT |
977 | ep_unregister_pollwait(ep, epi); |
978 | ||
979 | /* | |
980 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
981 | * allocated wait queue. Note that we don't care about the ep->ovflist |
982 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
983 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 984 | */ |
c7ea7630 | 985 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 986 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 987 | list_del_init(&epi->rdllink); |
c7ea7630 | 988 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 989 | |
b030a4dd | 990 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 991 | |
1da177e4 LT |
992 | return error; |
993 | } | |
994 | ||
1da177e4 LT |
995 | /* |
996 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 997 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
998 | */ |
999 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1000 | { | |
1001 | int pwake = 0; | |
1002 | unsigned int revents; | |
1da177e4 LT |
1003 | |
1004 | /* | |
e057e15f TB |
1005 | * Set the new event interest mask before calling f_op->poll(); |
1006 | * otherwise we might miss an event that happens between the | |
1007 | * f_op->poll() call and the new event set registering. | |
1da177e4 LT |
1008 | */ |
1009 | epi->event.events = event->events; | |
e057e15f | 1010 | epi->event.data = event->data; /* protected by mtx */ |
1da177e4 LT |
1011 | |
1012 | /* | |
1013 | * Get current event bits. We can safely use the file* here because | |
1014 | * its usage count has been increased by the caller of this function. | |
1015 | */ | |
1016 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
1017 | ||
1da177e4 | 1018 | /* |
c7ea7630 | 1019 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1020 | * list, push it inside. |
1da177e4 | 1021 | */ |
c7ea7630 | 1022 | if (revents & event->events) { |
e057e15f | 1023 | spin_lock_irq(&ep->lock); |
c7ea7630 DL |
1024 | if (!ep_is_linked(&epi->rdllink)) { |
1025 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1026 | ||
1027 | /* Notify waiting tasks that events are available */ | |
1028 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1029 | wake_up_locked(&ep->wq); |
c7ea7630 DL |
1030 | if (waitqueue_active(&ep->poll_wait)) |
1031 | pwake++; | |
7699acd1 | 1032 | } |
e057e15f | 1033 | spin_unlock_irq(&ep->lock); |
7699acd1 | 1034 | } |
1da177e4 | 1035 | |
7699acd1 DL |
1036 | /* We have to call this outside the lock */ |
1037 | if (pwake) | |
5071f97e | 1038 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1039 | |
7699acd1 | 1040 | return 0; |
1da177e4 LT |
1041 | } |
1042 | ||
296e236e DL |
1043 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
1044 | void *priv) | |
1da177e4 | 1045 | { |
5071f97e DL |
1046 | struct ep_send_events_data *esed = priv; |
1047 | int eventcnt; | |
296e236e | 1048 | unsigned int revents; |
5071f97e DL |
1049 | struct epitem *epi; |
1050 | struct epoll_event __user *uevent; | |
1da177e4 | 1051 | |
296e236e | 1052 | /* |
5071f97e DL |
1053 | * We can loop without lock because we are passed a task private list. |
1054 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1055 | * holding "mtx" during this call. | |
296e236e | 1056 | */ |
5071f97e DL |
1057 | for (eventcnt = 0, uevent = esed->events; |
1058 | !list_empty(head) && eventcnt < esed->maxevents;) { | |
1059 | epi = list_first_entry(head, struct epitem, rdllink); | |
d47de16c DL |
1060 | |
1061 | list_del_init(&epi->rdllink); | |
1da177e4 | 1062 | |
296e236e DL |
1063 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) & |
1064 | epi->event.events; | |
5071f97e | 1065 | |
296e236e | 1066 | /* |
5071f97e DL |
1067 | * If the event mask intersect the caller-requested one, |
1068 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
1069 | * is holding "mtx", so no operations coming from userspace | |
1070 | * can change the item. | |
296e236e DL |
1071 | */ |
1072 | if (revents) { | |
5071f97e | 1073 | if (__put_user(revents, &uevent->events) || |
d0305882 TB |
1074 | __put_user(epi->event.data, &uevent->data)) { |
1075 | list_add(&epi->rdllink, head); | |
296e236e | 1076 | return eventcnt ? eventcnt : -EFAULT; |
d0305882 | 1077 | } |
296e236e | 1078 | eventcnt++; |
5071f97e | 1079 | uevent++; |
296e236e DL |
1080 | if (epi->event.events & EPOLLONESHOT) |
1081 | epi->event.events &= EP_PRIVATE_BITS; | |
1082 | else if (!(epi->event.events & EPOLLET)) { | |
1083 | /* | |
1084 | * If this file has been added with Level | |
1085 | * Trigger mode, we need to insert back inside | |
1086 | * the ready list, so that the next call to | |
1087 | * epoll_wait() will check again the events | |
1088 | * availability. At this point, noone can insert | |
1089 | * into ep->rdllist besides us. The epoll_ctl() | |
1090 | * callers are locked out by | |
1091 | * ep_scan_ready_list() holding "mtx" and the | |
1092 | * poll callback will queue them in ep->ovflist. | |
1093 | */ | |
1094 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1095 | } | |
1096 | } | |
1097 | } | |
5071f97e DL |
1098 | |
1099 | return eventcnt; | |
1100 | } | |
d47de16c | 1101 | |
296e236e DL |
1102 | static int ep_send_events(struct eventpoll *ep, |
1103 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1104 | { |
1105 | struct ep_send_events_data esed; | |
1da177e4 | 1106 | |
5071f97e DL |
1107 | esed.maxevents = maxevents; |
1108 | esed.events = events; | |
6192bd53 | 1109 | |
5071f97e | 1110 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed); |
1da177e4 LT |
1111 | } |
1112 | ||
1da177e4 LT |
1113 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1114 | int maxevents, long timeout) | |
1115 | { | |
1116 | int res, eavail; | |
1117 | unsigned long flags; | |
1118 | long jtimeout; | |
1119 | wait_queue_t wait; | |
1120 | ||
1121 | /* | |
5071f97e | 1122 | * Calculate the timeout by checking for the "infinite" value (-1) |
1da177e4 LT |
1123 | * and the overflow condition. The passed timeout is in milliseconds, |
1124 | * that why (t * HZ) / 1000. | |
1125 | */ | |
e3306dd5 DL |
1126 | jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ? |
1127 | MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000; | |
1da177e4 LT |
1128 | |
1129 | retry: | |
c7ea7630 | 1130 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 LT |
1131 | |
1132 | res = 0; | |
1133 | if (list_empty(&ep->rdllist)) { | |
1134 | /* | |
1135 | * We don't have any available event to return to the caller. | |
1136 | * We need to sleep here, and we will be wake up by | |
1137 | * ep_poll_callback() when events will become available. | |
1138 | */ | |
1139 | init_waitqueue_entry(&wait, current); | |
d47de16c | 1140 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
3419b23a | 1141 | __add_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1142 | |
1143 | for (;;) { | |
1144 | /* | |
1145 | * We don't want to sleep if the ep_poll_callback() sends us | |
1146 | * a wakeup in between. That's why we set the task state | |
1147 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1148 | */ | |
1149 | set_current_state(TASK_INTERRUPTIBLE); | |
1150 | if (!list_empty(&ep->rdllist) || !jtimeout) | |
1151 | break; | |
1152 | if (signal_pending(current)) { | |
1153 | res = -EINTR; | |
1154 | break; | |
1155 | } | |
1156 | ||
c7ea7630 | 1157 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1158 | jtimeout = schedule_timeout(jtimeout); |
c7ea7630 | 1159 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1160 | } |
3419b23a | 1161 | __remove_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1162 | |
1163 | set_current_state(TASK_RUNNING); | |
1164 | } | |
1da177e4 | 1165 | /* Is it worth to try to dig for events ? */ |
5071f97e | 1166 | eavail = !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; |
1da177e4 | 1167 | |
c7ea7630 | 1168 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1169 | |
1170 | /* | |
1171 | * Try to transfer events to user space. In case we get 0 events and | |
1172 | * there's still timeout left over, we go trying again in search of | |
1173 | * more luck. | |
1174 | */ | |
1175 | if (!res && eavail && | |
d47de16c | 1176 | !(res = ep_send_events(ep, events, maxevents)) && jtimeout) |
1da177e4 LT |
1177 | goto retry; |
1178 | ||
1179 | return res; | |
1180 | } | |
1181 | ||
7699acd1 | 1182 | /* |
523723bb | 1183 | * Open an eventpoll file descriptor. |
7699acd1 | 1184 | */ |
5a8a82b1 | 1185 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
7699acd1 | 1186 | { |
bb57c3ed DL |
1187 | int error; |
1188 | struct eventpoll *ep = NULL; | |
7699acd1 | 1189 | |
e38b36f3 UD |
1190 | /* Check the EPOLL_* constant for consistency. */ |
1191 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1192 | ||
296e236e DL |
1193 | if (flags & ~EPOLL_CLOEXEC) |
1194 | return -EINVAL; | |
7699acd1 | 1195 | /* |
bb57c3ed | 1196 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 1197 | */ |
9fe5ad9c | 1198 | error = ep_alloc(&ep); |
bb57c3ed DL |
1199 | if (error < 0) |
1200 | return error; | |
7699acd1 DL |
1201 | /* |
1202 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 1203 | * a file structure and a free file descriptor. |
7699acd1 | 1204 | */ |
bb57c3ed DL |
1205 | error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep, |
1206 | flags & O_CLOEXEC); | |
1207 | if (error < 0) | |
2030a42c | 1208 | ep_free(ep); |
7699acd1 | 1209 | |
bb57c3ed | 1210 | return error; |
7699acd1 DL |
1211 | } |
1212 | ||
5a8a82b1 | 1213 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 1214 | { |
9fe5ad9c UD |
1215 | if (size < 0) |
1216 | return -EINVAL; | |
1217 | ||
1218 | return sys_epoll_create1(0); | |
a0998b50 UD |
1219 | } |
1220 | ||
7699acd1 DL |
1221 | /* |
1222 | * The following function implements the controller interface for | |
1223 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 1224 | * file descriptors inside the interest set. |
7699acd1 | 1225 | */ |
5a8a82b1 HC |
1226 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
1227 | struct epoll_event __user *, event) | |
7699acd1 DL |
1228 | { |
1229 | int error; | |
1230 | struct file *file, *tfile; | |
1231 | struct eventpoll *ep; | |
1232 | struct epitem *epi; | |
1233 | struct epoll_event epds; | |
1234 | ||
7699acd1 DL |
1235 | error = -EFAULT; |
1236 | if (ep_op_has_event(op) && | |
1237 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
1238 | goto error_return; | |
1239 | ||
1240 | /* Get the "struct file *" for the eventpoll file */ | |
1241 | error = -EBADF; | |
1242 | file = fget(epfd); | |
1243 | if (!file) | |
1244 | goto error_return; | |
1245 | ||
1246 | /* Get the "struct file *" for the target file */ | |
1247 | tfile = fget(fd); | |
1248 | if (!tfile) | |
1249 | goto error_fput; | |
1250 | ||
1251 | /* The target file descriptor must support poll */ | |
1252 | error = -EPERM; | |
1253 | if (!tfile->f_op || !tfile->f_op->poll) | |
1254 | goto error_tgt_fput; | |
1255 | ||
1256 | /* | |
1257 | * We have to check that the file structure underneath the file descriptor | |
1258 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
1259 | * adding an epoll file descriptor inside itself. | |
1260 | */ | |
1261 | error = -EINVAL; | |
1262 | if (file == tfile || !is_file_epoll(file)) | |
1263 | goto error_tgt_fput; | |
1264 | ||
1265 | /* | |
1266 | * At this point it is safe to assume that the "private_data" contains | |
1267 | * our own data structure. | |
1268 | */ | |
1269 | ep = file->private_data; | |
1270 | ||
d47de16c | 1271 | mutex_lock(&ep->mtx); |
7699acd1 | 1272 | |
67647d0f DL |
1273 | /* |
1274 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
1275 | * above, we can be sure to be able to use the item looked up by | |
1276 | * ep_find() till we release the mutex. | |
1277 | */ | |
7699acd1 DL |
1278 | epi = ep_find(ep, tfile, fd); |
1279 | ||
1280 | error = -EINVAL; | |
1281 | switch (op) { | |
1282 | case EPOLL_CTL_ADD: | |
1283 | if (!epi) { | |
1284 | epds.events |= POLLERR | POLLHUP; | |
7699acd1 DL |
1285 | error = ep_insert(ep, &epds, tfile, fd); |
1286 | } else | |
1287 | error = -EEXIST; | |
1288 | break; | |
1289 | case EPOLL_CTL_DEL: | |
1290 | if (epi) | |
1291 | error = ep_remove(ep, epi); | |
1292 | else | |
1293 | error = -ENOENT; | |
1294 | break; | |
1295 | case EPOLL_CTL_MOD: | |
1296 | if (epi) { | |
1297 | epds.events |= POLLERR | POLLHUP; | |
1298 | error = ep_modify(ep, epi, &epds); | |
1299 | } else | |
1300 | error = -ENOENT; | |
1301 | break; | |
1302 | } | |
d47de16c | 1303 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
1304 | |
1305 | error_tgt_fput: | |
1306 | fput(tfile); | |
1307 | error_fput: | |
1308 | fput(file); | |
1309 | error_return: | |
7699acd1 DL |
1310 | |
1311 | return error; | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1316 | * part of the user space epoll_wait(2). | |
1317 | */ | |
5a8a82b1 HC |
1318 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
1319 | int, maxevents, int, timeout) | |
7699acd1 DL |
1320 | { |
1321 | int error; | |
1322 | struct file *file; | |
1323 | struct eventpoll *ep; | |
1324 | ||
7699acd1 DL |
1325 | /* The maximum number of event must be greater than zero */ |
1326 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
1327 | return -EINVAL; | |
1328 | ||
1329 | /* Verify that the area passed by the user is writeable */ | |
1330 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { | |
1331 | error = -EFAULT; | |
1332 | goto error_return; | |
1333 | } | |
1334 | ||
1335 | /* Get the "struct file *" for the eventpoll file */ | |
1336 | error = -EBADF; | |
1337 | file = fget(epfd); | |
1338 | if (!file) | |
1339 | goto error_return; | |
1340 | ||
1341 | /* | |
1342 | * We have to check that the file structure underneath the fd | |
1343 | * the user passed to us _is_ an eventpoll file. | |
1344 | */ | |
1345 | error = -EINVAL; | |
1346 | if (!is_file_epoll(file)) | |
1347 | goto error_fput; | |
1348 | ||
1349 | /* | |
1350 | * At this point it is safe to assume that the "private_data" contains | |
1351 | * our own data structure. | |
1352 | */ | |
1353 | ep = file->private_data; | |
1354 | ||
1355 | /* Time to fish for events ... */ | |
1356 | error = ep_poll(ep, events, maxevents, timeout); | |
1357 | ||
1358 | error_fput: | |
1359 | fput(file); | |
1360 | error_return: | |
7699acd1 DL |
1361 | |
1362 | return error; | |
1363 | } | |
1364 | ||
f3de272b | 1365 | #ifdef HAVE_SET_RESTORE_SIGMASK |
7699acd1 DL |
1366 | |
1367 | /* | |
1368 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1369 | * part of the user space epoll_pwait(2). | |
1370 | */ | |
5a8a82b1 HC |
1371 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
1372 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
1373 | size_t, sigsetsize) | |
7699acd1 DL |
1374 | { |
1375 | int error; | |
1376 | sigset_t ksigmask, sigsaved; | |
1377 | ||
1378 | /* | |
1379 | * If the caller wants a certain signal mask to be set during the wait, | |
1380 | * we apply it here. | |
1381 | */ | |
1382 | if (sigmask) { | |
1383 | if (sigsetsize != sizeof(sigset_t)) | |
1384 | return -EINVAL; | |
1385 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1386 | return -EFAULT; | |
1387 | sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP)); | |
1388 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1389 | } | |
1390 | ||
1391 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
1392 | ||
1393 | /* | |
1394 | * If we changed the signal mask, we need to restore the original one. | |
1395 | * In case we've got a signal while waiting, we do not restore the | |
1396 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
1397 | * the way back to userspace, before the signal mask is restored. | |
1398 | */ | |
1399 | if (sigmask) { | |
1400 | if (error == -EINTR) { | |
1401 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 1402 | sizeof(sigsaved)); |
4e4c22c7 | 1403 | set_restore_sigmask(); |
7699acd1 DL |
1404 | } else |
1405 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1406 | } | |
1407 | ||
1408 | return error; | |
1409 | } | |
1410 | ||
f3de272b | 1411 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
7699acd1 | 1412 | |
1da177e4 LT |
1413 | static int __init eventpoll_init(void) |
1414 | { | |
7ef9964e DL |
1415 | struct sysinfo si; |
1416 | ||
1417 | si_meminfo(&si); | |
9df04e1f DL |
1418 | /* |
1419 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
1420 | */ | |
1421 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 1422 | EP_ITEM_COST; |
1da177e4 LT |
1423 | |
1424 | /* Initialize the structure used to perform safe poll wait head wake ups */ | |
5071f97e DL |
1425 | ep_nested_calls_init(&poll_safewake_ncalls); |
1426 | ||
1427 | /* Initialize the structure used to perform file's f_op->poll() calls */ | |
1428 | ep_nested_calls_init(&poll_readywalk_ncalls); | |
1da177e4 LT |
1429 | |
1430 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1431 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
bb57c3ed | 1432 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 LT |
1433 | |
1434 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1435 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
bb57c3ed | 1436 | sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL); |
1da177e4 | 1437 | |
1da177e4 | 1438 | return 0; |
1da177e4 | 1439 | } |
cea69241 | 1440 | fs_initcall(eventpoll_init); |