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 | #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 | ||
5071f97e DL |
95 | /* Maximum number of nesting allowed inside epoll sets */ |
96 | #define EP_MAX_NESTS 4 | |
1da177e4 | 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 | /* | |
5071f97e DL |
113 | * Structure used to track possible nested calls, for too deep recursions |
114 | * and loop cycles. | |
1da177e4 | 115 | */ |
5071f97e | 116 | struct nested_call_node { |
1da177e4 | 117 | struct list_head llink; |
36c8b586 | 118 | struct task_struct *task; |
5071f97e | 119 | void *cookie; |
1da177e4 LT |
120 | }; |
121 | ||
122 | /* | |
5071f97e DL |
123 | * This structure is used as collector for nested calls, to check for |
124 | * maximum recursion dept and loop cycles. | |
1da177e4 | 125 | */ |
5071f97e DL |
126 | struct nested_calls { |
127 | struct list_head tasks_call_list; | |
1da177e4 LT |
128 | spinlock_t lock; |
129 | }; | |
130 | ||
d47de16c DL |
131 | /* |
132 | * Each file descriptor added to the eventpoll interface will | |
133 | * have an entry of this type linked to the "rbr" RB tree. | |
134 | */ | |
135 | struct epitem { | |
67647d0f | 136 | /* RB tree node used to link this structure to the eventpoll RB tree */ |
d47de16c DL |
137 | struct rb_node rbn; |
138 | ||
139 | /* List header used to link this structure to the eventpoll ready list */ | |
140 | struct list_head rdllink; | |
141 | ||
c7ea7630 DL |
142 | /* |
143 | * Works together "struct eventpoll"->ovflist in keeping the | |
144 | * single linked chain of items. | |
145 | */ | |
146 | struct epitem *next; | |
147 | ||
d47de16c DL |
148 | /* The file descriptor information this item refers to */ |
149 | struct epoll_filefd ffd; | |
150 | ||
151 | /* Number of active wait queue attached to poll operations */ | |
152 | int nwait; | |
153 | ||
154 | /* List containing poll wait queues */ | |
155 | struct list_head pwqlist; | |
156 | ||
157 | /* The "container" of this item */ | |
158 | struct eventpoll *ep; | |
159 | ||
d47de16c DL |
160 | /* List header used to link this item to the "struct file" items list */ |
161 | struct list_head fllink; | |
162 | ||
c7ea7630 DL |
163 | /* The structure that describe the interested events and the source fd */ |
164 | struct epoll_event event; | |
d47de16c DL |
165 | }; |
166 | ||
1da177e4 LT |
167 | /* |
168 | * This structure is stored inside the "private_data" member of the file | |
169 | * structure and rapresent the main data sructure for the eventpoll | |
170 | * interface. | |
171 | */ | |
172 | struct eventpoll { | |
173 | /* Protect the this structure access */ | |
c7ea7630 | 174 | spinlock_t lock; |
1da177e4 LT |
175 | |
176 | /* | |
d47de16c DL |
177 | * This mutex is used to ensure that files are not removed |
178 | * while epoll is using them. This is held during the event | |
179 | * collection loop, the file cleanup path, the epoll file exit | |
180 | * code and the ctl operations. | |
1da177e4 | 181 | */ |
d47de16c | 182 | struct mutex mtx; |
1da177e4 LT |
183 | |
184 | /* Wait queue used by sys_epoll_wait() */ | |
185 | wait_queue_head_t wq; | |
186 | ||
187 | /* Wait queue used by file->poll() */ | |
188 | wait_queue_head_t poll_wait; | |
189 | ||
190 | /* List of ready file descriptors */ | |
191 | struct list_head rdllist; | |
192 | ||
67647d0f | 193 | /* RB tree root used to store monitored fd structs */ |
1da177e4 | 194 | struct rb_root rbr; |
d47de16c DL |
195 | |
196 | /* | |
197 | * This is a single linked list that chains all the "struct epitem" that | |
198 | * happened while transfering ready events to userspace w/out | |
199 | * holding ->lock. | |
200 | */ | |
201 | struct epitem *ovflist; | |
7ef9964e DL |
202 | |
203 | /* The user that created the eventpoll descriptor */ | |
204 | struct user_struct *user; | |
1da177e4 LT |
205 | }; |
206 | ||
207 | /* Wait structure used by the poll hooks */ | |
208 | struct eppoll_entry { | |
209 | /* List header used to link this structure to the "struct epitem" */ | |
210 | struct list_head llink; | |
211 | ||
212 | /* The "base" pointer is set to the container "struct epitem" */ | |
213 | void *base; | |
214 | ||
215 | /* | |
216 | * Wait queue item that will be linked to the target file wait | |
217 | * queue head. | |
218 | */ | |
219 | wait_queue_t wait; | |
220 | ||
221 | /* The wait queue head that linked the "wait" wait queue item */ | |
222 | wait_queue_head_t *whead; | |
223 | }; | |
224 | ||
1da177e4 LT |
225 | /* Wrapper struct used by poll queueing */ |
226 | struct ep_pqueue { | |
227 | poll_table pt; | |
228 | struct epitem *epi; | |
229 | }; | |
230 | ||
5071f97e DL |
231 | /* Used by the ep_send_events() function as callback private data */ |
232 | struct ep_send_events_data { | |
233 | int maxevents; | |
234 | struct epoll_event __user *events; | |
235 | }; | |
236 | ||
7ef9964e DL |
237 | /* |
238 | * Configuration options available inside /proc/sys/fs/epoll/ | |
239 | */ | |
7ef9964e DL |
240 | /* Maximum number of epoll watched descriptors, per user */ |
241 | static int max_user_watches __read_mostly; | |
242 | ||
1da177e4 | 243 | /* |
d47de16c | 244 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 245 | */ |
7ef9964e | 246 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 247 | |
5071f97e DL |
248 | /* Used for safe wake up implementation */ |
249 | static struct nested_calls poll_safewake_ncalls; | |
250 | ||
251 | /* Used to call file's f_op->poll() under the nested calls boundaries */ | |
252 | static struct nested_calls poll_readywalk_ncalls; | |
1da177e4 LT |
253 | |
254 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 255 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
256 | |
257 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 258 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 259 | |
7ef9964e DL |
260 | #ifdef CONFIG_SYSCTL |
261 | ||
262 | #include <linux/sysctl.h> | |
263 | ||
264 | static int zero; | |
265 | ||
266 | ctl_table epoll_table[] = { | |
7ef9964e DL |
267 | { |
268 | .procname = "max_user_watches", | |
269 | .data = &max_user_watches, | |
270 | .maxlen = sizeof(int), | |
271 | .mode = 0644, | |
272 | .proc_handler = &proc_dointvec_minmax, | |
273 | .extra1 = &zero, | |
274 | }, | |
275 | { .ctl_name = 0 } | |
276 | }; | |
277 | #endif /* CONFIG_SYSCTL */ | |
278 | ||
b030a4dd | 279 | |
67647d0f | 280 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
281 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
282 | struct file *file, int fd) | |
283 | { | |
284 | ffd->file = file; | |
285 | ffd->fd = fd; | |
286 | } | |
287 | ||
67647d0f | 288 | /* Compare RB tree keys */ |
b030a4dd PE |
289 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
290 | struct epoll_filefd *p2) | |
291 | { | |
292 | return (p1->file > p2->file ? +1: | |
293 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
294 | } | |
295 | ||
b030a4dd PE |
296 | /* Tells us if the item is currently linked */ |
297 | static inline int ep_is_linked(struct list_head *p) | |
298 | { | |
299 | return !list_empty(p); | |
300 | } | |
301 | ||
302 | /* Get the "struct epitem" from a wait queue pointer */ | |
cdac75e6 | 303 | static inline struct epitem *ep_item_from_wait(wait_queue_t *p) |
b030a4dd PE |
304 | { |
305 | return container_of(p, struct eppoll_entry, wait)->base; | |
306 | } | |
307 | ||
308 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 309 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
310 | { |
311 | return container_of(p, struct ep_pqueue, pt)->epi; | |
312 | } | |
313 | ||
314 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 315 | static inline int ep_op_has_event(int op) |
b030a4dd PE |
316 | { |
317 | return op != EPOLL_CTL_DEL; | |
318 | } | |
319 | ||
1da177e4 | 320 | /* Initialize the poll safe wake up structure */ |
5071f97e | 321 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 322 | { |
5071f97e DL |
323 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
324 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
325 | } |
326 | ||
5071f97e DL |
327 | /** |
328 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
329 | * that the recursion limit is not exceeded, and that | |
330 | * the same nested call (by the meaning of same cookie) is | |
331 | * no re-entered. | |
332 | * | |
333 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
334 | * @max_nests: Maximum number of allowed nesting calls. | |
335 | * @nproc: Nested call core function pointer. | |
336 | * @priv: Opaque data to be passed to the @nproc callback. | |
337 | * @cookie: Cookie to be used to identify this nested call. | |
338 | * | |
339 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
340 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 341 | */ |
5071f97e DL |
342 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
343 | int (*nproc)(void *, void *, int), void *priv, | |
344 | void *cookie) | |
1da177e4 | 345 | { |
5071f97e | 346 | int error, call_nests = 0; |
1da177e4 | 347 | unsigned long flags; |
36c8b586 | 348 | struct task_struct *this_task = current; |
5071f97e DL |
349 | struct list_head *lsthead = &ncalls->tasks_call_list; |
350 | struct nested_call_node *tncur; | |
351 | struct nested_call_node tnode; | |
1da177e4 | 352 | |
5071f97e | 353 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 354 | |
5071f97e DL |
355 | /* |
356 | * Try to see if the current task is already inside this wakeup call. | |
357 | * We use a list here, since the population inside this set is always | |
358 | * very much limited. | |
359 | */ | |
b70c3940 | 360 | list_for_each_entry(tncur, lsthead, llink) { |
5071f97e DL |
361 | if (tncur->task == this_task && |
362 | (tncur->cookie == cookie || ++call_nests > max_nests)) { | |
1da177e4 LT |
363 | /* |
364 | * Ops ... loop detected or maximum nest level reached. | |
365 | * We abort this wake by breaking the cycle itself. | |
366 | */ | |
5071f97e DL |
367 | spin_unlock_irqrestore(&ncalls->lock, flags); |
368 | ||
369 | return -1; | |
1da177e4 LT |
370 | } |
371 | } | |
372 | ||
5071f97e | 373 | /* Add the current task and cookie to the list */ |
1da177e4 | 374 | tnode.task = this_task; |
5071f97e | 375 | tnode.cookie = cookie; |
1da177e4 LT |
376 | list_add(&tnode.llink, lsthead); |
377 | ||
5071f97e | 378 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 379 | |
5071f97e DL |
380 | /* Call the nested function */ |
381 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
382 | |
383 | /* Remove the current task from the list */ | |
5071f97e | 384 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 385 | list_del(&tnode.llink); |
5071f97e DL |
386 | spin_unlock_irqrestore(&ncalls->lock, flags); |
387 | ||
388 | return error; | |
389 | } | |
390 | ||
391 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) | |
392 | { | |
393 | wake_up_nested((wait_queue_head_t *) cookie, 1 + call_nests); | |
394 | return 0; | |
395 | } | |
396 | ||
397 | /* | |
398 | * Perform a safe wake up of the poll wait list. The problem is that | |
399 | * with the new callback'd wake up system, it is possible that the | |
400 | * poll callback is reentered from inside the call to wake_up() done | |
401 | * on the poll wait queue head. The rule is that we cannot reenter the | |
402 | * wake up code from the same task more than EP_MAX_NESTS times, | |
403 | * and we cannot reenter the same wait queue head at all. This will | |
404 | * enable to have a hierarchy of epoll file descriptor of no more than | |
405 | * EP_MAX_NESTS deep. | |
406 | */ | |
407 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
408 | { | |
409 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, | |
410 | ep_poll_wakeup_proc, NULL, wq); | |
1da177e4 LT |
411 | } |
412 | ||
1da177e4 | 413 | /* |
7699acd1 DL |
414 | * This function unregister poll callbacks from the associated file descriptor. |
415 | * Since this must be called without holding "ep->lock" the atomic exchange trick | |
416 | * will protect us from multiple unregister. | |
1da177e4 | 417 | */ |
7699acd1 | 418 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 419 | { |
7699acd1 DL |
420 | int nwait; |
421 | struct list_head *lsthead = &epi->pwqlist; | |
422 | struct eppoll_entry *pwq; | |
1da177e4 | 423 | |
7699acd1 DL |
424 | /* This is called without locks, so we need the atomic exchange */ |
425 | nwait = xchg(&epi->nwait, 0); | |
1da177e4 | 426 | |
7699acd1 DL |
427 | if (nwait) { |
428 | while (!list_empty(lsthead)) { | |
429 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 430 | |
7699acd1 DL |
431 | list_del_init(&pwq->llink); |
432 | remove_wait_queue(pwq->whead, &pwq->wait); | |
433 | kmem_cache_free(pwq_cache, pwq); | |
434 | } | |
1da177e4 | 435 | } |
1da177e4 LT |
436 | } |
437 | ||
5071f97e DL |
438 | /** |
439 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
440 | * the scan code, to call f_op->poll(). Also allows for | |
441 | * O(NumReady) performance. | |
442 | * | |
443 | * @ep: Pointer to the epoll private data structure. | |
444 | * @sproc: Pointer to the scan callback. | |
445 | * @priv: Private opaque data passed to the @sproc callback. | |
446 | * | |
447 | * Returns: The same integer error code returned by the @sproc callback. | |
448 | */ | |
449 | static int ep_scan_ready_list(struct eventpoll *ep, | |
450 | int (*sproc)(struct eventpoll *, | |
451 | struct list_head *, void *), | |
452 | void *priv) | |
453 | { | |
454 | int error, pwake = 0; | |
455 | unsigned long flags; | |
456 | struct epitem *epi, *nepi; | |
457 | struct list_head txlist; | |
458 | ||
459 | INIT_LIST_HEAD(&txlist); | |
460 | ||
461 | /* | |
462 | * We need to lock this because we could be hit by | |
463 | * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL). | |
464 | */ | |
465 | mutex_lock(&ep->mtx); | |
466 | ||
467 | /* | |
468 | * Steal the ready list, and re-init the original one to the | |
469 | * empty list. Also, set ep->ovflist to NULL so that events | |
470 | * happening while looping w/out locks, are not lost. We cannot | |
471 | * have the poll callback to queue directly on ep->rdllist, | |
472 | * because we want the "sproc" callback to be able to do it | |
473 | * in a lockless way. | |
474 | */ | |
475 | spin_lock_irqsave(&ep->lock, flags); | |
476 | list_splice(&ep->rdllist, &txlist); | |
477 | INIT_LIST_HEAD(&ep->rdllist); | |
478 | ep->ovflist = NULL; | |
479 | spin_unlock_irqrestore(&ep->lock, flags); | |
480 | ||
481 | /* | |
482 | * Now call the callback function. | |
483 | */ | |
484 | error = (*sproc)(ep, &txlist, priv); | |
485 | ||
486 | spin_lock_irqsave(&ep->lock, flags); | |
487 | /* | |
488 | * During the time we spent inside the "sproc" callback, some | |
489 | * other events might have been queued by the poll callback. | |
490 | * We re-insert them inside the main ready-list here. | |
491 | */ | |
492 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
493 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { | |
494 | /* | |
495 | * We need to check if the item is already in the list. | |
496 | * During the "sproc" callback execution time, items are | |
497 | * queued into ->ovflist but the "txlist" might already | |
498 | * contain them, and the list_splice() below takes care of them. | |
499 | */ | |
500 | if (!ep_is_linked(&epi->rdllink)) | |
501 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
502 | } | |
503 | /* | |
504 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
505 | * releasing the lock, events will be queued in the normal way inside | |
506 | * ep->rdllist. | |
507 | */ | |
508 | ep->ovflist = EP_UNACTIVE_PTR; | |
509 | ||
510 | /* | |
511 | * Quickly re-inject items left on "txlist". | |
512 | */ | |
513 | list_splice(&txlist, &ep->rdllist); | |
514 | ||
515 | if (!list_empty(&ep->rdllist)) { | |
516 | /* | |
517 | * Wake up (if active) both the eventpoll wait list and the ->poll() | |
518 | * wait list (delayed after we release the lock). | |
519 | */ | |
520 | if (waitqueue_active(&ep->wq)) | |
521 | wake_up_locked(&ep->wq); | |
522 | if (waitqueue_active(&ep->poll_wait)) | |
523 | pwake++; | |
524 | } | |
525 | spin_unlock_irqrestore(&ep->lock, flags); | |
526 | ||
527 | mutex_unlock(&ep->mtx); | |
528 | ||
529 | /* We have to call this outside the lock */ | |
530 | if (pwake) | |
531 | ep_poll_safewake(&ep->poll_wait); | |
532 | ||
533 | return error; | |
534 | } | |
535 | ||
7699acd1 DL |
536 | /* |
537 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 538 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
539 | */ |
540 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
541 | { | |
7699acd1 DL |
542 | unsigned long flags; |
543 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
544 | |
545 | /* | |
7699acd1 DL |
546 | * Removes poll wait queue hooks. We _have_ to do this without holding |
547 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
548 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
549 | * queue head lock when unregistering the wait queue. The wakeup callback | |
550 | * will run by holding the wait queue head lock and will call our callback | |
551 | * that will try to get "ep->lock". | |
1da177e4 | 552 | */ |
7699acd1 | 553 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 554 | |
7699acd1 | 555 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 556 | spin_lock(&file->f_lock); |
7699acd1 DL |
557 | if (ep_is_linked(&epi->fllink)) |
558 | list_del_init(&epi->fllink); | |
68499914 | 559 | spin_unlock(&file->f_lock); |
1da177e4 | 560 | |
cdac75e6 | 561 | rb_erase(&epi->rbn, &ep->rbr); |
1da177e4 | 562 | |
c7ea7630 DL |
563 | spin_lock_irqsave(&ep->lock, flags); |
564 | if (ep_is_linked(&epi->rdllink)) | |
565 | list_del_init(&epi->rdllink); | |
566 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 567 | |
7699acd1 | 568 | /* At this point it is safe to free the eventpoll item */ |
c7ea7630 | 569 | kmem_cache_free(epi_cache, epi); |
1da177e4 | 570 | |
7ef9964e DL |
571 | atomic_dec(&ep->user->epoll_watches); |
572 | ||
c7ea7630 DL |
573 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_remove(%p, %p)\n", |
574 | current, ep, file)); | |
1da177e4 | 575 | |
c7ea7630 | 576 | return 0; |
1da177e4 LT |
577 | } |
578 | ||
7699acd1 | 579 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 580 | { |
7699acd1 DL |
581 | struct rb_node *rbp; |
582 | struct epitem *epi; | |
1da177e4 | 583 | |
7699acd1 DL |
584 | /* We need to release all tasks waiting for these file */ |
585 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 586 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 587 | |
7699acd1 DL |
588 | /* |
589 | * We need to lock this because we could be hit by | |
590 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 591 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
592 | * is on the way to be removed and no one has references to it |
593 | * anymore. The only hit might come from eventpoll_release_file() but | |
594 | * holding "epmutex" is sufficent here. | |
595 | */ | |
596 | mutex_lock(&epmutex); | |
1da177e4 LT |
597 | |
598 | /* | |
7699acd1 | 599 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 600 | */ |
7699acd1 DL |
601 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
602 | epi = rb_entry(rbp, struct epitem, rbn); | |
603 | ||
604 | ep_unregister_pollwait(ep, epi); | |
605 | } | |
1da177e4 LT |
606 | |
607 | /* | |
7699acd1 DL |
608 | * Walks through the whole tree by freeing each "struct epitem". At this |
609 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 610 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 611 | * us during this operation. So we can avoid the lock on "ep->lock". |
1da177e4 | 612 | */ |
c80544dc | 613 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
7699acd1 DL |
614 | epi = rb_entry(rbp, struct epitem, rbn); |
615 | ep_remove(ep, epi); | |
616 | } | |
1da177e4 | 617 | |
7699acd1 | 618 | mutex_unlock(&epmutex); |
d47de16c | 619 | mutex_destroy(&ep->mtx); |
7ef9964e | 620 | free_uid(ep->user); |
f0ee9aab | 621 | kfree(ep); |
7699acd1 | 622 | } |
1da177e4 | 623 | |
7699acd1 DL |
624 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
625 | { | |
626 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 627 | |
f0ee9aab | 628 | if (ep) |
7699acd1 | 629 | ep_free(ep); |
7699acd1 DL |
630 | |
631 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: close() ep=%p\n", current, ep)); | |
632 | return 0; | |
1da177e4 LT |
633 | } |
634 | ||
5071f97e DL |
635 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, void *priv) |
636 | { | |
637 | struct epitem *epi, *tmp; | |
638 | ||
639 | list_for_each_entry_safe(epi, tmp, head, rdllink) { | |
640 | if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) & | |
641 | epi->event.events) | |
642 | return POLLIN | POLLRDNORM; | |
643 | else | |
644 | /* | |
645 | * Item has been dropped into the ready list by the poll | |
646 | * callback, but it's not actually ready, as far as | |
647 | * caller requested events goes. We can remove it here. | |
648 | */ | |
649 | list_del_init(&epi->rdllink); | |
650 | } | |
651 | ||
652 | return 0; | |
653 | } | |
654 | ||
655 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) | |
656 | { | |
657 | return ep_scan_ready_list(priv, ep_read_events_proc, NULL); | |
658 | } | |
659 | ||
7699acd1 DL |
660 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
661 | { | |
5071f97e | 662 | int pollflags; |
7699acd1 | 663 | struct eventpoll *ep = file->private_data; |
1da177e4 | 664 | |
7699acd1 DL |
665 | /* Insert inside our poll wait queue */ |
666 | poll_wait(file, &ep->poll_wait, wait); | |
667 | ||
5071f97e DL |
668 | /* |
669 | * Proceed to find out if wanted events are really available inside | |
670 | * the ready list. This need to be done under ep_call_nested() | |
671 | * supervision, since the call to f_op->poll() done on listed files | |
672 | * could re-enter here. | |
673 | */ | |
674 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, | |
675 | ep_poll_readyevents_proc, ep, ep); | |
7699acd1 | 676 | |
5071f97e | 677 | return pollflags != -1 ? pollflags: 0; |
7699acd1 DL |
678 | } |
679 | ||
680 | /* File callbacks that implement the eventpoll file behaviour */ | |
681 | static const struct file_operations eventpoll_fops = { | |
682 | .release = ep_eventpoll_release, | |
683 | .poll = ep_eventpoll_poll | |
684 | }; | |
685 | ||
686 | /* Fast test to see if the file is an evenpoll file */ | |
687 | static inline int is_file_epoll(struct file *f) | |
688 | { | |
689 | return f->f_op == &eventpoll_fops; | |
690 | } | |
b611967d DL |
691 | |
692 | /* | |
7699acd1 DL |
693 | * This is called from eventpoll_release() to unlink files from the eventpoll |
694 | * interface. We need to have this facility to cleanup correctly files that are | |
695 | * closed without being removed from the eventpoll interface. | |
b611967d | 696 | */ |
7699acd1 | 697 | void eventpoll_release_file(struct file *file) |
b611967d | 698 | { |
7699acd1 DL |
699 | struct list_head *lsthead = &file->f_ep_links; |
700 | struct eventpoll *ep; | |
701 | struct epitem *epi; | |
b611967d DL |
702 | |
703 | /* | |
68499914 | 704 | * We don't want to get "file->f_lock" because it is not |
7699acd1 DL |
705 | * necessary. It is not necessary because we're in the "struct file" |
706 | * cleanup path, and this means that noone is using this file anymore. | |
5071f97e | 707 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 708 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 709 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 710 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 711 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 712 | * from anywhere but ep_free(). |
68499914 JC |
713 | * |
714 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 715 | */ |
7699acd1 | 716 | mutex_lock(&epmutex); |
b611967d | 717 | |
7699acd1 DL |
718 | while (!list_empty(lsthead)) { |
719 | epi = list_first_entry(lsthead, struct epitem, fllink); | |
b611967d | 720 | |
7699acd1 DL |
721 | ep = epi->ep; |
722 | list_del_init(&epi->fllink); | |
d47de16c | 723 | mutex_lock(&ep->mtx); |
7699acd1 | 724 | ep_remove(ep, epi); |
d47de16c | 725 | mutex_unlock(&ep->mtx); |
b611967d DL |
726 | } |
727 | ||
7699acd1 | 728 | mutex_unlock(&epmutex); |
b611967d DL |
729 | } |
730 | ||
53d2be79 | 731 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 732 | { |
7ef9964e DL |
733 | int error; |
734 | struct user_struct *user; | |
735 | struct eventpoll *ep; | |
1da177e4 | 736 | |
7ef9964e | 737 | user = get_current_user(); |
7ef9964e DL |
738 | error = -ENOMEM; |
739 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
740 | if (unlikely(!ep)) | |
741 | goto free_uid; | |
1da177e4 | 742 | |
c7ea7630 | 743 | spin_lock_init(&ep->lock); |
d47de16c | 744 | mutex_init(&ep->mtx); |
1da177e4 LT |
745 | init_waitqueue_head(&ep->wq); |
746 | init_waitqueue_head(&ep->poll_wait); | |
747 | INIT_LIST_HEAD(&ep->rdllist); | |
748 | ep->rbr = RB_ROOT; | |
d47de16c | 749 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 750 | ep->user = user; |
1da177e4 | 751 | |
53d2be79 | 752 | *pep = ep; |
1da177e4 | 753 | |
53d2be79 | 754 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_alloc() ep=%p\n", |
1da177e4 LT |
755 | current, ep)); |
756 | return 0; | |
7ef9964e DL |
757 | |
758 | free_uid: | |
759 | free_uid(user); | |
760 | return error; | |
1da177e4 LT |
761 | } |
762 | ||
1da177e4 | 763 | /* |
c7ea7630 DL |
764 | * Search the file inside the eventpoll tree. The RB tree operations |
765 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
766 | * "mtx" held. | |
1da177e4 LT |
767 | */ |
768 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
769 | { | |
770 | int kcmp; | |
1da177e4 LT |
771 | struct rb_node *rbp; |
772 | struct epitem *epi, *epir = NULL; | |
773 | struct epoll_filefd ffd; | |
774 | ||
b030a4dd | 775 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
776 | for (rbp = ep->rbr.rb_node; rbp; ) { |
777 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 778 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
779 | if (kcmp > 0) |
780 | rbp = rbp->rb_right; | |
781 | else if (kcmp < 0) | |
782 | rbp = rbp->rb_left; | |
783 | else { | |
1da177e4 LT |
784 | epir = epi; |
785 | break; | |
786 | } | |
787 | } | |
1da177e4 LT |
788 | |
789 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_find(%p) -> %p\n", | |
790 | current, file, epir)); | |
791 | ||
792 | return epir; | |
793 | } | |
794 | ||
1da177e4 | 795 | /* |
7699acd1 DL |
796 | * This is the callback that is passed to the wait queue wakeup |
797 | * machanism. It is called by the stored file descriptors when they | |
798 | * have events to report. | |
1da177e4 | 799 | */ |
7699acd1 | 800 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 801 | { |
7699acd1 DL |
802 | int pwake = 0; |
803 | unsigned long flags; | |
804 | struct epitem *epi = ep_item_from_wait(wait); | |
805 | struct eventpoll *ep = epi->ep; | |
1da177e4 | 806 | |
7699acd1 DL |
807 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: poll_callback(%p) epi=%p ep=%p\n", |
808 | current, epi->ffd.file, epi, ep)); | |
1da177e4 | 809 | |
c7ea7630 | 810 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 811 | |
7699acd1 DL |
812 | /* |
813 | * If the event mask does not contain any poll(2) event, we consider the | |
814 | * descriptor to be disabled. This condition is likely the effect of the | |
815 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
816 | * until the next EPOLL_CTL_MOD will be issued. | |
817 | */ | |
818 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
819 | goto out_unlock; |
820 | ||
821 | /* | |
822 | * If we are trasfering events to userspace, we can hold no locks | |
823 | * (because we're accessing user memory, and because of linux f_op->poll() | |
824 | * semantics). All the events that happens during that period of time are | |
825 | * chained in ep->ovflist and requeued later on. | |
826 | */ | |
827 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
828 | if (epi->next == EP_UNACTIVE_PTR) { | |
829 | epi->next = ep->ovflist; | |
830 | ep->ovflist = epi; | |
831 | } | |
832 | goto out_unlock; | |
833 | } | |
1da177e4 | 834 | |
7699acd1 | 835 | /* If this file is already in the ready list we exit soon */ |
5071f97e DL |
836 | if (!ep_is_linked(&epi->rdllink)) |
837 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
7699acd1 | 838 | |
7699acd1 DL |
839 | /* |
840 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
841 | * wait list. | |
842 | */ | |
843 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 844 | wake_up_locked(&ep->wq); |
7699acd1 DL |
845 | if (waitqueue_active(&ep->poll_wait)) |
846 | pwake++; | |
847 | ||
d47de16c | 848 | out_unlock: |
c7ea7630 | 849 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 850 | |
7699acd1 DL |
851 | /* We have to call this outside the lock */ |
852 | if (pwake) | |
5071f97e | 853 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 DL |
854 | |
855 | return 1; | |
856 | } | |
1da177e4 LT |
857 | |
858 | /* | |
859 | * This is the callback that is used to add our wait queue to the | |
860 | * target file wakeup lists. | |
861 | */ | |
862 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
863 | poll_table *pt) | |
864 | { | |
b030a4dd | 865 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
866 | struct eppoll_entry *pwq; |
867 | ||
e94b1766 | 868 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
869 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
870 | pwq->whead = whead; | |
871 | pwq->base = epi; | |
872 | add_wait_queue(whead, &pwq->wait); | |
873 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
874 | epi->nwait++; | |
5071f97e | 875 | } else |
1da177e4 LT |
876 | /* We have to signal that an error occurred */ |
877 | epi->nwait = -1; | |
1da177e4 LT |
878 | } |
879 | ||
1da177e4 LT |
880 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
881 | { | |
882 | int kcmp; | |
883 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
884 | struct epitem *epic; | |
885 | ||
886 | while (*p) { | |
887 | parent = *p; | |
888 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 889 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
890 | if (kcmp > 0) |
891 | p = &parent->rb_right; | |
892 | else | |
893 | p = &parent->rb_left; | |
894 | } | |
895 | rb_link_node(&epi->rbn, parent, p); | |
896 | rb_insert_color(&epi->rbn, &ep->rbr); | |
897 | } | |
898 | ||
c7ea7630 DL |
899 | /* |
900 | * Must be called with "mtx" held. | |
901 | */ | |
1da177e4 LT |
902 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
903 | struct file *tfile, int fd) | |
904 | { | |
905 | int error, revents, pwake = 0; | |
906 | unsigned long flags; | |
907 | struct epitem *epi; | |
908 | struct ep_pqueue epq; | |
909 | ||
7ef9964e DL |
910 | if (unlikely(atomic_read(&ep->user->epoll_watches) >= |
911 | max_user_watches)) | |
912 | return -ENOSPC; | |
e94b1766 | 913 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 914 | return -ENOMEM; |
1da177e4 LT |
915 | |
916 | /* Item initialization follow here ... */ | |
1da177e4 LT |
917 | INIT_LIST_HEAD(&epi->rdllink); |
918 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
919 | INIT_LIST_HEAD(&epi->pwqlist); |
920 | epi->ep = ep; | |
b030a4dd | 921 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 922 | epi->event = *event; |
1da177e4 | 923 | epi->nwait = 0; |
d47de16c | 924 | epi->next = EP_UNACTIVE_PTR; |
1da177e4 LT |
925 | |
926 | /* Initialize the poll table using the queue callback */ | |
927 | epq.epi = epi; | |
928 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
929 | ||
930 | /* | |
931 | * Attach the item to the poll hooks and get current event bits. | |
932 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
933 | * been increased by the caller of this function. Note that after |
934 | * this operation completes, the poll callback can start hitting | |
935 | * the new item. | |
1da177e4 LT |
936 | */ |
937 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
938 | ||
939 | /* | |
940 | * We have to check if something went wrong during the poll wait queue | |
941 | * install process. Namely an allocation for a wait queue failed due | |
942 | * high memory pressure. | |
943 | */ | |
7ef9964e | 944 | error = -ENOMEM; |
1da177e4 | 945 | if (epi->nwait < 0) |
7699acd1 | 946 | goto error_unregister; |
1da177e4 LT |
947 | |
948 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 949 | spin_lock(&tfile->f_lock); |
1da177e4 | 950 | list_add_tail(&epi->fllink, &tfile->f_ep_links); |
68499914 | 951 | spin_unlock(&tfile->f_lock); |
1da177e4 | 952 | |
c7ea7630 DL |
953 | /* |
954 | * Add the current item to the RB tree. All RB tree operations are | |
955 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
956 | */ | |
1da177e4 LT |
957 | ep_rbtree_insert(ep, epi); |
958 | ||
c7ea7630 DL |
959 | /* We have to drop the new item inside our item list to keep track of it */ |
960 | spin_lock_irqsave(&ep->lock, flags); | |
961 | ||
1da177e4 | 962 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 963 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 LT |
964 | list_add_tail(&epi->rdllink, &ep->rdllist); |
965 | ||
966 | /* Notify waiting tasks that events are available */ | |
967 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 968 | wake_up_locked(&ep->wq); |
1da177e4 LT |
969 | if (waitqueue_active(&ep->poll_wait)) |
970 | pwake++; | |
971 | } | |
972 | ||
c7ea7630 | 973 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 974 | |
7ef9964e DL |
975 | atomic_inc(&ep->user->epoll_watches); |
976 | ||
1da177e4 LT |
977 | /* We have to call this outside the lock */ |
978 | if (pwake) | |
5071f97e | 979 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 LT |
980 | |
981 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_insert(%p, %p, %d)\n", | |
982 | current, ep, tfile, fd)); | |
983 | ||
984 | return 0; | |
985 | ||
7699acd1 | 986 | error_unregister: |
1da177e4 LT |
987 | ep_unregister_pollwait(ep, epi); |
988 | ||
989 | /* | |
990 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
991 | * allocated wait queue. Note that we don't care about the ep->ovflist |
992 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
993 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 994 | */ |
c7ea7630 | 995 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 996 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 997 | list_del_init(&epi->rdllink); |
c7ea7630 | 998 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 999 | |
b030a4dd | 1000 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1001 | |
1da177e4 LT |
1002 | return error; |
1003 | } | |
1004 | ||
1da177e4 LT |
1005 | /* |
1006 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1007 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
1008 | */ |
1009 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1010 | { | |
1011 | int pwake = 0; | |
1012 | unsigned int revents; | |
1013 | unsigned long flags; | |
1014 | ||
1015 | /* | |
1016 | * Set the new event interest mask before calling f_op->poll(), otherwise | |
1017 | * a potential race might occur. In fact if we do this operation inside | |
1018 | * the lock, an event might happen between the f_op->poll() call and the | |
1019 | * new event set registering. | |
1020 | */ | |
1021 | epi->event.events = event->events; | |
1022 | ||
1023 | /* | |
1024 | * Get current event bits. We can safely use the file* here because | |
1025 | * its usage count has been increased by the caller of this function. | |
1026 | */ | |
1027 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
1028 | ||
c7ea7630 | 1029 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 LT |
1030 | |
1031 | /* Copy the data member from inside the lock */ | |
1032 | epi->event.data = event->data; | |
1033 | ||
1034 | /* | |
c7ea7630 | 1035 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1036 | * list, push it inside. |
1da177e4 | 1037 | */ |
c7ea7630 DL |
1038 | if (revents & event->events) { |
1039 | if (!ep_is_linked(&epi->rdllink)) { | |
1040 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1041 | ||
1042 | /* Notify waiting tasks that events are available */ | |
1043 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1044 | wake_up_locked(&ep->wq); |
c7ea7630 DL |
1045 | if (waitqueue_active(&ep->poll_wait)) |
1046 | pwake++; | |
7699acd1 DL |
1047 | } |
1048 | } | |
c7ea7630 | 1049 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1050 | |
7699acd1 DL |
1051 | /* We have to call this outside the lock */ |
1052 | if (pwake) | |
5071f97e | 1053 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1054 | |
7699acd1 | 1055 | return 0; |
1da177e4 LT |
1056 | } |
1057 | ||
5071f97e | 1058 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, void *priv) |
1da177e4 | 1059 | { |
5071f97e DL |
1060 | struct ep_send_events_data *esed = priv; |
1061 | int eventcnt; | |
1062 | unsigned int revents; | |
1063 | struct epitem *epi; | |
1064 | struct epoll_event __user *uevent; | |
1da177e4 | 1065 | |
5071f97e DL |
1066 | /* |
1067 | * We can loop without lock because we are passed a task private list. | |
1068 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1069 | * holding "mtx" during this call. | |
1070 | */ | |
1071 | for (eventcnt = 0, uevent = esed->events; | |
1072 | !list_empty(head) && eventcnt < esed->maxevents;) { | |
1073 | epi = list_first_entry(head, struct epitem, rdllink); | |
d47de16c DL |
1074 | |
1075 | list_del_init(&epi->rdllink); | |
1da177e4 | 1076 | |
5071f97e DL |
1077 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) & |
1078 | epi->event.events; | |
1079 | ||
1080 | /* | |
1081 | * If the event mask intersect the caller-requested one, | |
1082 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
1083 | * is holding "mtx", so no operations coming from userspace | |
1084 | * can change the item. | |
1085 | */ | |
1086 | if (revents) { | |
1087 | if (__put_user(revents, &uevent->events) || | |
1088 | __put_user(epi->event.data, &uevent->data)) | |
1089 | return eventcnt ? eventcnt: -EFAULT; | |
1090 | eventcnt++; | |
1091 | uevent++; | |
1092 | if (epi->event.events & EPOLLONESHOT) | |
1093 | epi->event.events &= EP_PRIVATE_BITS; | |
1094 | else if (!(epi->event.events & EPOLLET)) | |
1095 | /* | |
1096 | * If this file has been added with Level Trigger | |
1097 | * mode, we need to insert back inside the ready | |
1098 | * list, so that the next call to epoll_wait() | |
1099 | * will check again the events availability. | |
1100 | * At this point, noone can insert into ep->rdllist | |
1101 | * besides us. The epoll_ctl() callers are locked | |
1102 | * out by ep_scan_ready_list() holding "mtx" and | |
1103 | * the poll callback will queue them in ep->ovflist. | |
1104 | */ | |
1105 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | return eventcnt; | |
1110 | } | |
d47de16c | 1111 | |
5071f97e DL |
1112 | static int ep_send_events(struct eventpoll *ep, struct epoll_event __user *events, |
1113 | int maxevents) | |
1114 | { | |
1115 | struct ep_send_events_data esed; | |
1da177e4 | 1116 | |
5071f97e DL |
1117 | esed.maxevents = maxevents; |
1118 | esed.events = events; | |
6192bd53 | 1119 | |
5071f97e | 1120 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed); |
1da177e4 LT |
1121 | } |
1122 | ||
1da177e4 LT |
1123 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1124 | int maxevents, long timeout) | |
1125 | { | |
1126 | int res, eavail; | |
1127 | unsigned long flags; | |
1128 | long jtimeout; | |
1129 | wait_queue_t wait; | |
1130 | ||
1131 | /* | |
5071f97e | 1132 | * Calculate the timeout by checking for the "infinite" value (-1) |
1da177e4 LT |
1133 | * and the overflow condition. The passed timeout is in milliseconds, |
1134 | * that why (t * HZ) / 1000. | |
1135 | */ | |
e3306dd5 DL |
1136 | jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ? |
1137 | MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000; | |
1da177e4 LT |
1138 | |
1139 | retry: | |
c7ea7630 | 1140 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 LT |
1141 | |
1142 | res = 0; | |
1143 | if (list_empty(&ep->rdllist)) { | |
1144 | /* | |
1145 | * We don't have any available event to return to the caller. | |
1146 | * We need to sleep here, and we will be wake up by | |
1147 | * ep_poll_callback() when events will become available. | |
1148 | */ | |
1149 | init_waitqueue_entry(&wait, current); | |
d47de16c | 1150 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
3419b23a | 1151 | __add_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1152 | |
1153 | for (;;) { | |
1154 | /* | |
1155 | * We don't want to sleep if the ep_poll_callback() sends us | |
1156 | * a wakeup in between. That's why we set the task state | |
1157 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1158 | */ | |
1159 | set_current_state(TASK_INTERRUPTIBLE); | |
1160 | if (!list_empty(&ep->rdllist) || !jtimeout) | |
1161 | break; | |
1162 | if (signal_pending(current)) { | |
1163 | res = -EINTR; | |
1164 | break; | |
1165 | } | |
1166 | ||
c7ea7630 | 1167 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1168 | jtimeout = schedule_timeout(jtimeout); |
c7ea7630 | 1169 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1170 | } |
3419b23a | 1171 | __remove_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1172 | |
1173 | set_current_state(TASK_RUNNING); | |
1174 | } | |
1da177e4 | 1175 | /* Is it worth to try to dig for events ? */ |
5071f97e | 1176 | eavail = !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; |
1da177e4 | 1177 | |
c7ea7630 | 1178 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1179 | |
1180 | /* | |
1181 | * Try to transfer events to user space. In case we get 0 events and | |
1182 | * there's still timeout left over, we go trying again in search of | |
1183 | * more luck. | |
1184 | */ | |
1185 | if (!res && eavail && | |
d47de16c | 1186 | !(res = ep_send_events(ep, events, maxevents)) && jtimeout) |
1da177e4 LT |
1187 | goto retry; |
1188 | ||
1189 | return res; | |
1190 | } | |
1191 | ||
7699acd1 | 1192 | /* |
523723bb | 1193 | * Open an eventpoll file descriptor. |
7699acd1 | 1194 | */ |
5a8a82b1 | 1195 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
7699acd1 | 1196 | { |
5071f97e DL |
1197 | int error; |
1198 | struct eventpoll *ep = NULL; | |
7699acd1 | 1199 | |
e38b36f3 UD |
1200 | /* Check the EPOLL_* constant for consistency. */ |
1201 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1202 | ||
7699acd1 | 1203 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d)\n", |
9fe5ad9c | 1204 | current, flags)); |
7699acd1 | 1205 | |
5071f97e DL |
1206 | error = -EINVAL; |
1207 | if (flags & ~EPOLL_CLOEXEC) | |
1208 | goto error_return; | |
1209 | ||
7699acd1 | 1210 | /* |
5071f97e | 1211 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 1212 | */ |
9fe5ad9c | 1213 | error = ep_alloc(&ep); |
5071f97e | 1214 | if (error < 0) |
7699acd1 DL |
1215 | goto error_return; |
1216 | ||
1217 | /* | |
1218 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 1219 | * a file structure and a free file descriptor. |
7699acd1 | 1220 | */ |
5071f97e DL |
1221 | error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep, |
1222 | flags & O_CLOEXEC); | |
1223 | if (error < 0) | |
2030a42c | 1224 | ep_free(ep); |
7699acd1 | 1225 | |
2030a42c | 1226 | error_return: |
7699acd1 | 1227 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", |
5071f97e | 1228 | current, flags, error)); |
7699acd1 | 1229 | |
5071f97e | 1230 | return error; |
7699acd1 DL |
1231 | } |
1232 | ||
5a8a82b1 | 1233 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 1234 | { |
9fe5ad9c UD |
1235 | if (size < 0) |
1236 | return -EINVAL; | |
1237 | ||
1238 | return sys_epoll_create1(0); | |
a0998b50 UD |
1239 | } |
1240 | ||
7699acd1 DL |
1241 | /* |
1242 | * The following function implements the controller interface for | |
1243 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 1244 | * file descriptors inside the interest set. |
7699acd1 | 1245 | */ |
5a8a82b1 HC |
1246 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
1247 | struct epoll_event __user *, event) | |
7699acd1 DL |
1248 | { |
1249 | int error; | |
1250 | struct file *file, *tfile; | |
1251 | struct eventpoll *ep; | |
1252 | struct epitem *epi; | |
1253 | struct epoll_event epds; | |
1254 | ||
1255 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p)\n", | |
1256 | current, epfd, op, fd, event)); | |
1257 | ||
1258 | error = -EFAULT; | |
1259 | if (ep_op_has_event(op) && | |
1260 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
1261 | goto error_return; | |
1262 | ||
1263 | /* Get the "struct file *" for the eventpoll file */ | |
1264 | error = -EBADF; | |
1265 | file = fget(epfd); | |
1266 | if (!file) | |
1267 | goto error_return; | |
1268 | ||
1269 | /* Get the "struct file *" for the target file */ | |
1270 | tfile = fget(fd); | |
1271 | if (!tfile) | |
1272 | goto error_fput; | |
1273 | ||
1274 | /* The target file descriptor must support poll */ | |
1275 | error = -EPERM; | |
1276 | if (!tfile->f_op || !tfile->f_op->poll) | |
1277 | goto error_tgt_fput; | |
1278 | ||
1279 | /* | |
1280 | * We have to check that the file structure underneath the file descriptor | |
1281 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
1282 | * adding an epoll file descriptor inside itself. | |
1283 | */ | |
1284 | error = -EINVAL; | |
1285 | if (file == tfile || !is_file_epoll(file)) | |
1286 | goto error_tgt_fput; | |
1287 | ||
1288 | /* | |
1289 | * At this point it is safe to assume that the "private_data" contains | |
1290 | * our own data structure. | |
1291 | */ | |
1292 | ep = file->private_data; | |
1293 | ||
d47de16c | 1294 | mutex_lock(&ep->mtx); |
7699acd1 | 1295 | |
67647d0f DL |
1296 | /* |
1297 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
1298 | * above, we can be sure to be able to use the item looked up by | |
1299 | * ep_find() till we release the mutex. | |
1300 | */ | |
7699acd1 DL |
1301 | epi = ep_find(ep, tfile, fd); |
1302 | ||
1303 | error = -EINVAL; | |
1304 | switch (op) { | |
1305 | case EPOLL_CTL_ADD: | |
1306 | if (!epi) { | |
1307 | epds.events |= POLLERR | POLLHUP; | |
1308 | ||
1309 | error = ep_insert(ep, &epds, tfile, fd); | |
1310 | } else | |
1311 | error = -EEXIST; | |
1312 | break; | |
1313 | case EPOLL_CTL_DEL: | |
1314 | if (epi) | |
1315 | error = ep_remove(ep, epi); | |
1316 | else | |
1317 | error = -ENOENT; | |
1318 | break; | |
1319 | case EPOLL_CTL_MOD: | |
1320 | if (epi) { | |
1321 | epds.events |= POLLERR | POLLHUP; | |
1322 | error = ep_modify(ep, epi, &epds); | |
1323 | } else | |
1324 | error = -ENOENT; | |
1325 | break; | |
1326 | } | |
d47de16c | 1327 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
1328 | |
1329 | error_tgt_fput: | |
1330 | fput(tfile); | |
1331 | error_fput: | |
1332 | fput(file); | |
1333 | error_return: | |
1334 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p) = %d\n", | |
1335 | current, epfd, op, fd, event, error)); | |
1336 | ||
1337 | return error; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1342 | * part of the user space epoll_wait(2). | |
1343 | */ | |
5a8a82b1 HC |
1344 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
1345 | int, maxevents, int, timeout) | |
7699acd1 DL |
1346 | { |
1347 | int error; | |
1348 | struct file *file; | |
1349 | struct eventpoll *ep; | |
1350 | ||
1351 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d)\n", | |
1352 | current, epfd, events, maxevents, timeout)); | |
1353 | ||
1354 | /* The maximum number of event must be greater than zero */ | |
1355 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
1356 | return -EINVAL; | |
1357 | ||
1358 | /* Verify that the area passed by the user is writeable */ | |
1359 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { | |
1360 | error = -EFAULT; | |
1361 | goto error_return; | |
1362 | } | |
1363 | ||
1364 | /* Get the "struct file *" for the eventpoll file */ | |
1365 | error = -EBADF; | |
1366 | file = fget(epfd); | |
1367 | if (!file) | |
1368 | goto error_return; | |
1369 | ||
1370 | /* | |
1371 | * We have to check that the file structure underneath the fd | |
1372 | * the user passed to us _is_ an eventpoll file. | |
1373 | */ | |
1374 | error = -EINVAL; | |
1375 | if (!is_file_epoll(file)) | |
1376 | goto error_fput; | |
1377 | ||
1378 | /* | |
1379 | * At this point it is safe to assume that the "private_data" contains | |
1380 | * our own data structure. | |
1381 | */ | |
1382 | ep = file->private_data; | |
1383 | ||
1384 | /* Time to fish for events ... */ | |
1385 | error = ep_poll(ep, events, maxevents, timeout); | |
1386 | ||
1387 | error_fput: | |
1388 | fput(file); | |
1389 | error_return: | |
1390 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d) = %d\n", | |
1391 | current, epfd, events, maxevents, timeout, error)); | |
1392 | ||
1393 | return error; | |
1394 | } | |
1395 | ||
f3de272b | 1396 | #ifdef HAVE_SET_RESTORE_SIGMASK |
7699acd1 DL |
1397 | |
1398 | /* | |
1399 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1400 | * part of the user space epoll_pwait(2). | |
1401 | */ | |
5a8a82b1 HC |
1402 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
1403 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
1404 | size_t, sigsetsize) | |
7699acd1 DL |
1405 | { |
1406 | int error; | |
1407 | sigset_t ksigmask, sigsaved; | |
1408 | ||
1409 | /* | |
1410 | * If the caller wants a certain signal mask to be set during the wait, | |
1411 | * we apply it here. | |
1412 | */ | |
1413 | if (sigmask) { | |
1414 | if (sigsetsize != sizeof(sigset_t)) | |
1415 | return -EINVAL; | |
1416 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1417 | return -EFAULT; | |
1418 | sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP)); | |
1419 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1420 | } | |
1421 | ||
1422 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
1423 | ||
1424 | /* | |
1425 | * If we changed the signal mask, we need to restore the original one. | |
1426 | * In case we've got a signal while waiting, we do not restore the | |
1427 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
1428 | * the way back to userspace, before the signal mask is restored. | |
1429 | */ | |
1430 | if (sigmask) { | |
1431 | if (error == -EINTR) { | |
1432 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 1433 | sizeof(sigsaved)); |
4e4c22c7 | 1434 | set_restore_sigmask(); |
7699acd1 DL |
1435 | } else |
1436 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1437 | } | |
1438 | ||
1439 | return error; | |
1440 | } | |
1441 | ||
f3de272b | 1442 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
7699acd1 | 1443 | |
1da177e4 LT |
1444 | static int __init eventpoll_init(void) |
1445 | { | |
7ef9964e DL |
1446 | struct sysinfo si; |
1447 | ||
1448 | si_meminfo(&si); | |
9df04e1f DL |
1449 | /* |
1450 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
1451 | */ | |
1452 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 1453 | EP_ITEM_COST; |
1da177e4 LT |
1454 | |
1455 | /* Initialize the structure used to perform safe poll wait head wake ups */ | |
5071f97e DL |
1456 | ep_nested_calls_init(&poll_safewake_ncalls); |
1457 | ||
1458 | /* Initialize the structure used to perform file's f_op->poll() calls */ | |
1459 | ep_nested_calls_init(&poll_readywalk_ncalls); | |
1da177e4 LT |
1460 | |
1461 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1462 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
1463 | 0, SLAB_HWCACHE_ALIGN|EPI_SLAB_DEBUG|SLAB_PANIC, | |
20c2df83 | 1464 | NULL); |
1da177e4 LT |
1465 | |
1466 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1467 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
1468 | sizeof(struct eppoll_entry), 0, | |
20c2df83 | 1469 | EPI_SLAB_DEBUG|SLAB_PANIC, NULL); |
1da177e4 | 1470 | |
1da177e4 | 1471 | return 0; |
1da177e4 | 1472 | } |
cea69241 | 1473 | fs_initcall(eventpoll_init); |