]>
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> | |
60063497 | 40 | #include <linux/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 | |
bf6a41db | 65 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
22bacca4 DL |
66 | * It is also acquired when inserting an epoll fd onto another epoll |
67 | * fd. We do this so that we walk the epoll tree and ensure that this | |
68 | * insertion does not create a cycle of epoll file descriptors, which | |
69 | * could lead to deadlock. We need a global mutex to prevent two | |
70 | * simultaneous inserts (A into B and B into A) from racing and | |
71 | * constructing a cycle without either insert observing that it is | |
72 | * going to. | |
d8805e63 NE |
73 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
74 | * case when one epoll fd is added to another. In this case, we | |
75 | * always acquire the locks in the order of nesting (i.e. after | |
76 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired | |
77 | * before e2->mtx). Since we disallow cycles of epoll file | |
78 | * descriptors, this ensures that the mutexes are well-ordered. In | |
79 | * order to communicate this nesting to lockdep, when walking a tree | |
80 | * of epoll file descriptors, we use the current recursion depth as | |
81 | * the lockdep subkey. | |
d47de16c DL |
82 | * It is possible to drop the "ep->mtx" and to use the global |
83 | * mutex "epmutex" (together with "ep->lock") to have it working, | |
84 | * but having "ep->mtx" will make the interface more scalable. | |
144efe3e | 85 | * Events that require holding "epmutex" are very rare, while for |
d47de16c DL |
86 | * normal operations the epoll private "ep->mtx" will guarantee |
87 | * a better scalability. | |
1da177e4 LT |
88 | */ |
89 | ||
1da177e4 LT |
90 | /* Epoll private bits inside the event mask */ |
91 | #define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET) | |
92 | ||
5071f97e DL |
93 | /* Maximum number of nesting allowed inside epoll sets */ |
94 | #define EP_MAX_NESTS 4 | |
1da177e4 | 95 | |
b611967d DL |
96 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
97 | ||
d47de16c DL |
98 | #define EP_UNACTIVE_PTR ((void *) -1L) |
99 | ||
7ef9964e DL |
100 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
101 | ||
1da177e4 LT |
102 | struct epoll_filefd { |
103 | struct file *file; | |
104 | int fd; | |
105 | }; | |
106 | ||
107 | /* | |
5071f97e DL |
108 | * Structure used to track possible nested calls, for too deep recursions |
109 | * and loop cycles. | |
1da177e4 | 110 | */ |
5071f97e | 111 | struct nested_call_node { |
1da177e4 | 112 | struct list_head llink; |
5071f97e | 113 | void *cookie; |
3fe4a975 | 114 | void *ctx; |
1da177e4 LT |
115 | }; |
116 | ||
117 | /* | |
5071f97e DL |
118 | * This structure is used as collector for nested calls, to check for |
119 | * maximum recursion dept and loop cycles. | |
1da177e4 | 120 | */ |
5071f97e DL |
121 | struct nested_calls { |
122 | struct list_head tasks_call_list; | |
1da177e4 LT |
123 | spinlock_t lock; |
124 | }; | |
125 | ||
d47de16c DL |
126 | /* |
127 | * Each file descriptor added to the eventpoll interface will | |
128 | * have an entry of this type linked to the "rbr" RB tree. | |
129 | */ | |
130 | struct epitem { | |
67647d0f | 131 | /* RB tree node used to link this structure to the eventpoll RB tree */ |
d47de16c DL |
132 | struct rb_node rbn; |
133 | ||
134 | /* List header used to link this structure to the eventpoll ready list */ | |
135 | struct list_head rdllink; | |
136 | ||
c7ea7630 DL |
137 | /* |
138 | * Works together "struct eventpoll"->ovflist in keeping the | |
139 | * single linked chain of items. | |
140 | */ | |
141 | struct epitem *next; | |
142 | ||
d47de16c DL |
143 | /* The file descriptor information this item refers to */ |
144 | struct epoll_filefd ffd; | |
145 | ||
146 | /* Number of active wait queue attached to poll operations */ | |
147 | int nwait; | |
148 | ||
149 | /* List containing poll wait queues */ | |
150 | struct list_head pwqlist; | |
151 | ||
152 | /* The "container" of this item */ | |
153 | struct eventpoll *ep; | |
154 | ||
d47de16c DL |
155 | /* List header used to link this item to the "struct file" items list */ |
156 | struct list_head fllink; | |
157 | ||
c7ea7630 DL |
158 | /* The structure that describe the interested events and the source fd */ |
159 | struct epoll_event event; | |
d47de16c DL |
160 | }; |
161 | ||
1da177e4 LT |
162 | /* |
163 | * This structure is stored inside the "private_data" member of the file | |
bf6a41db | 164 | * structure and represents the main data structure for the eventpoll |
1da177e4 LT |
165 | * interface. |
166 | */ | |
167 | struct eventpoll { | |
bf6a41db | 168 | /* Protect the access to this structure */ |
c7ea7630 | 169 | spinlock_t lock; |
1da177e4 LT |
170 | |
171 | /* | |
d47de16c DL |
172 | * This mutex is used to ensure that files are not removed |
173 | * while epoll is using them. This is held during the event | |
174 | * collection loop, the file cleanup path, the epoll file exit | |
175 | * code and the ctl operations. | |
1da177e4 | 176 | */ |
d47de16c | 177 | struct mutex mtx; |
1da177e4 LT |
178 | |
179 | /* Wait queue used by sys_epoll_wait() */ | |
180 | wait_queue_head_t wq; | |
181 | ||
182 | /* Wait queue used by file->poll() */ | |
183 | wait_queue_head_t poll_wait; | |
184 | ||
185 | /* List of ready file descriptors */ | |
186 | struct list_head rdllist; | |
187 | ||
67647d0f | 188 | /* RB tree root used to store monitored fd structs */ |
1da177e4 | 189 | struct rb_root rbr; |
d47de16c DL |
190 | |
191 | /* | |
192 | * This is a single linked list that chains all the "struct epitem" that | |
25985edc | 193 | * happened while transferring ready events to userspace w/out |
d47de16c DL |
194 | * holding ->lock. |
195 | */ | |
196 | struct epitem *ovflist; | |
7ef9964e DL |
197 | |
198 | /* The user that created the eventpoll descriptor */ | |
199 | struct user_struct *user; | |
28d82dc1 JB |
200 | |
201 | struct file *file; | |
202 | ||
203 | /* used to optimize loop detection check */ | |
204 | int visited; | |
205 | struct list_head visited_list_link; | |
1da177e4 LT |
206 | }; |
207 | ||
208 | /* Wait structure used by the poll hooks */ | |
209 | struct eppoll_entry { | |
210 | /* List header used to link this structure to the "struct epitem" */ | |
211 | struct list_head llink; | |
212 | ||
213 | /* The "base" pointer is set to the container "struct epitem" */ | |
4f0989db | 214 | struct epitem *base; |
1da177e4 LT |
215 | |
216 | /* | |
217 | * Wait queue item that will be linked to the target file wait | |
218 | * queue head. | |
219 | */ | |
220 | wait_queue_t wait; | |
221 | ||
222 | /* The wait queue head that linked the "wait" wait queue item */ | |
223 | wait_queue_head_t *whead; | |
224 | }; | |
225 | ||
1da177e4 LT |
226 | /* Wrapper struct used by poll queueing */ |
227 | struct ep_pqueue { | |
228 | poll_table pt; | |
229 | struct epitem *epi; | |
230 | }; | |
231 | ||
5071f97e DL |
232 | /* Used by the ep_send_events() function as callback private data */ |
233 | struct ep_send_events_data { | |
234 | int maxevents; | |
235 | struct epoll_event __user *events; | |
236 | }; | |
237 | ||
7ef9964e DL |
238 | /* |
239 | * Configuration options available inside /proc/sys/fs/epoll/ | |
240 | */ | |
7ef9964e | 241 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 242 | static long max_user_watches __read_mostly; |
7ef9964e | 243 | |
1da177e4 | 244 | /* |
d47de16c | 245 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 246 | */ |
7ef9964e | 247 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 248 | |
22bacca4 DL |
249 | /* Used to check for epoll file descriptor inclusion loops */ |
250 | static struct nested_calls poll_loop_ncalls; | |
251 | ||
5071f97e DL |
252 | /* Used for safe wake up implementation */ |
253 | static struct nested_calls poll_safewake_ncalls; | |
254 | ||
255 | /* Used to call file's f_op->poll() under the nested calls boundaries */ | |
256 | static struct nested_calls poll_readywalk_ncalls; | |
1da177e4 LT |
257 | |
258 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 259 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
260 | |
261 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 262 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 263 | |
28d82dc1 JB |
264 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
265 | static LIST_HEAD(visited_list); | |
266 | ||
267 | /* | |
268 | * List of files with newly added links, where we may need to limit the number | |
269 | * of emanating paths. Protected by the epmutex. | |
270 | */ | |
271 | static LIST_HEAD(tfile_check_list); | |
272 | ||
7ef9964e DL |
273 | #ifdef CONFIG_SYSCTL |
274 | ||
275 | #include <linux/sysctl.h> | |
276 | ||
52bd19f7 RH |
277 | static long zero; |
278 | static long long_max = LONG_MAX; | |
7ef9964e DL |
279 | |
280 | ctl_table epoll_table[] = { | |
7ef9964e DL |
281 | { |
282 | .procname = "max_user_watches", | |
283 | .data = &max_user_watches, | |
52bd19f7 | 284 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 285 | .mode = 0644, |
52bd19f7 | 286 | .proc_handler = proc_doulongvec_minmax, |
7ef9964e | 287 | .extra1 = &zero, |
52bd19f7 | 288 | .extra2 = &long_max, |
7ef9964e | 289 | }, |
ab09203e | 290 | { } |
7ef9964e DL |
291 | }; |
292 | #endif /* CONFIG_SYSCTL */ | |
293 | ||
28d82dc1 JB |
294 | static const struct file_operations eventpoll_fops; |
295 | ||
296 | static inline int is_file_epoll(struct file *f) | |
297 | { | |
298 | return f->f_op == &eventpoll_fops; | |
299 | } | |
b030a4dd | 300 | |
67647d0f | 301 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
302 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
303 | struct file *file, int fd) | |
304 | { | |
305 | ffd->file = file; | |
306 | ffd->fd = fd; | |
307 | } | |
308 | ||
67647d0f | 309 | /* Compare RB tree keys */ |
b030a4dd PE |
310 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
311 | struct epoll_filefd *p2) | |
312 | { | |
313 | return (p1->file > p2->file ? +1: | |
314 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
315 | } | |
316 | ||
b030a4dd PE |
317 | /* Tells us if the item is currently linked */ |
318 | static inline int ep_is_linked(struct list_head *p) | |
319 | { | |
320 | return !list_empty(p); | |
321 | } | |
322 | ||
971316f0 ON |
323 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p) |
324 | { | |
325 | return container_of(p, struct eppoll_entry, wait); | |
326 | } | |
327 | ||
b030a4dd | 328 | /* Get the "struct epitem" from a wait queue pointer */ |
cdac75e6 | 329 | static inline struct epitem *ep_item_from_wait(wait_queue_t *p) |
b030a4dd PE |
330 | { |
331 | return container_of(p, struct eppoll_entry, wait)->base; | |
332 | } | |
333 | ||
334 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 335 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
336 | { |
337 | return container_of(p, struct ep_pqueue, pt)->epi; | |
338 | } | |
339 | ||
340 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 341 | static inline int ep_op_has_event(int op) |
b030a4dd PE |
342 | { |
343 | return op != EPOLL_CTL_DEL; | |
344 | } | |
345 | ||
1da177e4 | 346 | /* Initialize the poll safe wake up structure */ |
5071f97e | 347 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 348 | { |
5071f97e DL |
349 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
350 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
351 | } |
352 | ||
3fb0e584 DL |
353 | /** |
354 | * ep_events_available - Checks if ready events might be available. | |
355 | * | |
356 | * @ep: Pointer to the eventpoll context. | |
357 | * | |
358 | * Returns: Returns a value different than zero if ready events are available, | |
359 | * or zero otherwise. | |
360 | */ | |
361 | static inline int ep_events_available(struct eventpoll *ep) | |
362 | { | |
363 | return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; | |
364 | } | |
365 | ||
5071f97e DL |
366 | /** |
367 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
368 | * that the recursion limit is not exceeded, and that | |
369 | * the same nested call (by the meaning of same cookie) is | |
370 | * no re-entered. | |
371 | * | |
372 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
373 | * @max_nests: Maximum number of allowed nesting calls. | |
374 | * @nproc: Nested call core function pointer. | |
375 | * @priv: Opaque data to be passed to the @nproc callback. | |
376 | * @cookie: Cookie to be used to identify this nested call. | |
3fe4a975 | 377 | * @ctx: This instance context. |
5071f97e DL |
378 | * |
379 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
380 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 381 | */ |
5071f97e DL |
382 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
383 | int (*nproc)(void *, void *, int), void *priv, | |
3fe4a975 | 384 | void *cookie, void *ctx) |
1da177e4 | 385 | { |
5071f97e | 386 | int error, call_nests = 0; |
1da177e4 | 387 | unsigned long flags; |
5071f97e DL |
388 | struct list_head *lsthead = &ncalls->tasks_call_list; |
389 | struct nested_call_node *tncur; | |
390 | struct nested_call_node tnode; | |
1da177e4 | 391 | |
5071f97e | 392 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 393 | |
5071f97e DL |
394 | /* |
395 | * Try to see if the current task is already inside this wakeup call. | |
396 | * We use a list here, since the population inside this set is always | |
397 | * very much limited. | |
398 | */ | |
b70c3940 | 399 | list_for_each_entry(tncur, lsthead, llink) { |
3fe4a975 | 400 | if (tncur->ctx == ctx && |
5071f97e | 401 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
1da177e4 LT |
402 | /* |
403 | * Ops ... loop detected or maximum nest level reached. | |
404 | * We abort this wake by breaking the cycle itself. | |
405 | */ | |
abff55ce TB |
406 | error = -1; |
407 | goto out_unlock; | |
1da177e4 LT |
408 | } |
409 | } | |
410 | ||
5071f97e | 411 | /* Add the current task and cookie to the list */ |
3fe4a975 | 412 | tnode.ctx = ctx; |
5071f97e | 413 | tnode.cookie = cookie; |
1da177e4 LT |
414 | list_add(&tnode.llink, lsthead); |
415 | ||
5071f97e | 416 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 417 | |
5071f97e DL |
418 | /* Call the nested function */ |
419 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
420 | |
421 | /* Remove the current task from the list */ | |
5071f97e | 422 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 423 | list_del(&tnode.llink); |
3fe4a975 | 424 | out_unlock: |
5071f97e DL |
425 | spin_unlock_irqrestore(&ncalls->lock, flags); |
426 | ||
427 | return error; | |
428 | } | |
429 | ||
02edc6fc SR |
430 | /* |
431 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
432 | * the use of wait queues used by epoll is done in a very controlled | |
433 | * manner. Wake ups can nest inside each other, but are never done | |
434 | * with the same locking. For example: | |
435 | * | |
436 | * dfd = socket(...); | |
437 | * efd1 = epoll_create(); | |
438 | * efd2 = epoll_create(); | |
439 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
440 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
441 | * | |
442 | * When a packet arrives to the device underneath "dfd", the net code will | |
443 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
444 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
445 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
446 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
447 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
448 | * that ends up in another wake_up(), after having checked about the | |
449 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
450 | * avoid stack blasting. | |
451 | * | |
452 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
453 | * this special case of epoll. | |
454 | */ | |
2dfa4eea DL |
455 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
456 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
457 | unsigned long events, int subclass) | |
458 | { | |
459 | unsigned long flags; | |
460 | ||
461 | spin_lock_irqsave_nested(&wqueue->lock, flags, subclass); | |
462 | wake_up_locked_poll(wqueue, events); | |
463 | spin_unlock_irqrestore(&wqueue->lock, flags); | |
464 | } | |
465 | #else | |
466 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
467 | unsigned long events, int subclass) | |
468 | { | |
469 | wake_up_poll(wqueue, events); | |
470 | } | |
471 | #endif | |
472 | ||
5071f97e DL |
473 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
474 | { | |
2dfa4eea DL |
475 | ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN, |
476 | 1 + call_nests); | |
5071f97e DL |
477 | return 0; |
478 | } | |
479 | ||
480 | /* | |
481 | * Perform a safe wake up of the poll wait list. The problem is that | |
482 | * with the new callback'd wake up system, it is possible that the | |
483 | * poll callback is reentered from inside the call to wake_up() done | |
484 | * on the poll wait queue head. The rule is that we cannot reenter the | |
485 | * wake up code from the same task more than EP_MAX_NESTS times, | |
486 | * and we cannot reenter the same wait queue head at all. This will | |
487 | * enable to have a hierarchy of epoll file descriptor of no more than | |
488 | * EP_MAX_NESTS deep. | |
489 | */ | |
490 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
491 | { | |
3fe4a975 DL |
492 | int this_cpu = get_cpu(); |
493 | ||
5071f97e | 494 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
3fe4a975 DL |
495 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
496 | ||
497 | put_cpu(); | |
1da177e4 LT |
498 | } |
499 | ||
971316f0 ON |
500 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
501 | { | |
502 | wait_queue_head_t *whead; | |
503 | ||
504 | rcu_read_lock(); | |
505 | /* If it is cleared by POLLFREE, it should be rcu-safe */ | |
506 | whead = rcu_dereference(pwq->whead); | |
507 | if (whead) | |
508 | remove_wait_queue(whead, &pwq->wait); | |
509 | rcu_read_unlock(); | |
510 | } | |
511 | ||
1da177e4 | 512 | /* |
d1bc90dd TB |
513 | * This function unregisters poll callbacks from the associated file |
514 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
515 | * ep_free). | |
1da177e4 | 516 | */ |
7699acd1 | 517 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 518 | { |
7699acd1 DL |
519 | struct list_head *lsthead = &epi->pwqlist; |
520 | struct eppoll_entry *pwq; | |
1da177e4 | 521 | |
d1bc90dd TB |
522 | while (!list_empty(lsthead)) { |
523 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 524 | |
d1bc90dd | 525 | list_del(&pwq->llink); |
971316f0 | 526 | ep_remove_wait_queue(pwq); |
d1bc90dd | 527 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 528 | } |
1da177e4 LT |
529 | } |
530 | ||
5071f97e DL |
531 | /** |
532 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
533 | * the scan code, to call f_op->poll(). Also allows for | |
534 | * O(NumReady) performance. | |
535 | * | |
536 | * @ep: Pointer to the epoll private data structure. | |
537 | * @sproc: Pointer to the scan callback. | |
538 | * @priv: Private opaque data passed to the @sproc callback. | |
d8805e63 | 539 | * @depth: The current depth of recursive f_op->poll calls. |
5071f97e DL |
540 | * |
541 | * Returns: The same integer error code returned by the @sproc callback. | |
542 | */ | |
543 | static int ep_scan_ready_list(struct eventpoll *ep, | |
544 | int (*sproc)(struct eventpoll *, | |
545 | struct list_head *, void *), | |
d8805e63 NE |
546 | void *priv, |
547 | int depth) | |
5071f97e DL |
548 | { |
549 | int error, pwake = 0; | |
550 | unsigned long flags; | |
551 | struct epitem *epi, *nepi; | |
296e236e | 552 | LIST_HEAD(txlist); |
5071f97e DL |
553 | |
554 | /* | |
555 | * We need to lock this because we could be hit by | |
e057e15f | 556 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e | 557 | */ |
d8805e63 | 558 | mutex_lock_nested(&ep->mtx, depth); |
5071f97e DL |
559 | |
560 | /* | |
561 | * Steal the ready list, and re-init the original one to the | |
562 | * empty list. Also, set ep->ovflist to NULL so that events | |
563 | * happening while looping w/out locks, are not lost. We cannot | |
564 | * have the poll callback to queue directly on ep->rdllist, | |
565 | * because we want the "sproc" callback to be able to do it | |
566 | * in a lockless way. | |
567 | */ | |
568 | spin_lock_irqsave(&ep->lock, flags); | |
296e236e | 569 | list_splice_init(&ep->rdllist, &txlist); |
5071f97e DL |
570 | ep->ovflist = NULL; |
571 | spin_unlock_irqrestore(&ep->lock, flags); | |
572 | ||
573 | /* | |
574 | * Now call the callback function. | |
575 | */ | |
576 | error = (*sproc)(ep, &txlist, priv); | |
577 | ||
578 | spin_lock_irqsave(&ep->lock, flags); | |
579 | /* | |
580 | * During the time we spent inside the "sproc" callback, some | |
581 | * other events might have been queued by the poll callback. | |
582 | * We re-insert them inside the main ready-list here. | |
583 | */ | |
584 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
585 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { | |
586 | /* | |
587 | * We need to check if the item is already in the list. | |
588 | * During the "sproc" callback execution time, items are | |
589 | * queued into ->ovflist but the "txlist" might already | |
590 | * contain them, and the list_splice() below takes care of them. | |
591 | */ | |
592 | if (!ep_is_linked(&epi->rdllink)) | |
593 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
594 | } | |
595 | /* | |
596 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
597 | * releasing the lock, events will be queued in the normal way inside | |
598 | * ep->rdllist. | |
599 | */ | |
600 | ep->ovflist = EP_UNACTIVE_PTR; | |
601 | ||
602 | /* | |
603 | * Quickly re-inject items left on "txlist". | |
604 | */ | |
605 | list_splice(&txlist, &ep->rdllist); | |
606 | ||
607 | if (!list_empty(&ep->rdllist)) { | |
608 | /* | |
296e236e DL |
609 | * Wake up (if active) both the eventpoll wait list and |
610 | * the ->poll() wait list (delayed after we release the lock). | |
5071f97e DL |
611 | */ |
612 | if (waitqueue_active(&ep->wq)) | |
613 | wake_up_locked(&ep->wq); | |
614 | if (waitqueue_active(&ep->poll_wait)) | |
615 | pwake++; | |
616 | } | |
617 | spin_unlock_irqrestore(&ep->lock, flags); | |
618 | ||
619 | mutex_unlock(&ep->mtx); | |
620 | ||
621 | /* We have to call this outside the lock */ | |
622 | if (pwake) | |
623 | ep_poll_safewake(&ep->poll_wait); | |
624 | ||
625 | return error; | |
626 | } | |
627 | ||
7699acd1 DL |
628 | /* |
629 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 630 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
631 | */ |
632 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
633 | { | |
7699acd1 DL |
634 | unsigned long flags; |
635 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
636 | |
637 | /* | |
7699acd1 DL |
638 | * Removes poll wait queue hooks. We _have_ to do this without holding |
639 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
640 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
641 | * queue head lock when unregistering the wait queue. The wakeup callback | |
642 | * will run by holding the wait queue head lock and will call our callback | |
643 | * that will try to get "ep->lock". | |
1da177e4 | 644 | */ |
7699acd1 | 645 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 646 | |
7699acd1 | 647 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 648 | spin_lock(&file->f_lock); |
7699acd1 DL |
649 | if (ep_is_linked(&epi->fllink)) |
650 | list_del_init(&epi->fllink); | |
68499914 | 651 | spin_unlock(&file->f_lock); |
1da177e4 | 652 | |
cdac75e6 | 653 | rb_erase(&epi->rbn, &ep->rbr); |
1da177e4 | 654 | |
c7ea7630 DL |
655 | spin_lock_irqsave(&ep->lock, flags); |
656 | if (ep_is_linked(&epi->rdllink)) | |
657 | list_del_init(&epi->rdllink); | |
658 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 659 | |
7699acd1 | 660 | /* At this point it is safe to free the eventpoll item */ |
c7ea7630 | 661 | kmem_cache_free(epi_cache, epi); |
1da177e4 | 662 | |
52bd19f7 | 663 | atomic_long_dec(&ep->user->epoll_watches); |
7ef9964e | 664 | |
c7ea7630 | 665 | return 0; |
1da177e4 LT |
666 | } |
667 | ||
7699acd1 | 668 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 669 | { |
7699acd1 DL |
670 | struct rb_node *rbp; |
671 | struct epitem *epi; | |
1da177e4 | 672 | |
7699acd1 DL |
673 | /* We need to release all tasks waiting for these file */ |
674 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 675 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 676 | |
7699acd1 DL |
677 | /* |
678 | * We need to lock this because we could be hit by | |
679 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 680 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
681 | * is on the way to be removed and no one has references to it |
682 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 683 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
684 | */ |
685 | mutex_lock(&epmutex); | |
1da177e4 LT |
686 | |
687 | /* | |
7699acd1 | 688 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 689 | */ |
7699acd1 DL |
690 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
691 | epi = rb_entry(rbp, struct epitem, rbn); | |
692 | ||
693 | ep_unregister_pollwait(ep, epi); | |
694 | } | |
1da177e4 LT |
695 | |
696 | /* | |
7699acd1 DL |
697 | * Walks through the whole tree by freeing each "struct epitem". At this |
698 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 699 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 700 | * us during this operation. So we can avoid the lock on "ep->lock". |
1da177e4 | 701 | */ |
c80544dc | 702 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
7699acd1 DL |
703 | epi = rb_entry(rbp, struct epitem, rbn); |
704 | ep_remove(ep, epi); | |
705 | } | |
1da177e4 | 706 | |
7699acd1 | 707 | mutex_unlock(&epmutex); |
d47de16c | 708 | mutex_destroy(&ep->mtx); |
7ef9964e | 709 | free_uid(ep->user); |
f0ee9aab | 710 | kfree(ep); |
7699acd1 | 711 | } |
1da177e4 | 712 | |
7699acd1 DL |
713 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
714 | { | |
715 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 716 | |
f0ee9aab | 717 | if (ep) |
7699acd1 | 718 | ep_free(ep); |
7699acd1 | 719 | |
7699acd1 | 720 | return 0; |
1da177e4 LT |
721 | } |
722 | ||
296e236e DL |
723 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
724 | void *priv) | |
5071f97e DL |
725 | { |
726 | struct epitem *epi, *tmp; | |
626cf236 | 727 | poll_table pt; |
5071f97e | 728 | |
626cf236 | 729 | init_poll_funcptr(&pt, NULL); |
5071f97e | 730 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
626cf236 HV |
731 | pt._key = epi->event.events; |
732 | if (epi->ffd.file->f_op->poll(epi->ffd.file, &pt) & | |
5071f97e DL |
733 | epi->event.events) |
734 | return POLLIN | POLLRDNORM; | |
296e236e | 735 | else { |
5071f97e DL |
736 | /* |
737 | * Item has been dropped into the ready list by the poll | |
738 | * callback, but it's not actually ready, as far as | |
739 | * caller requested events goes. We can remove it here. | |
740 | */ | |
741 | list_del_init(&epi->rdllink); | |
296e236e | 742 | } |
5071f97e DL |
743 | } |
744 | ||
745 | return 0; | |
746 | } | |
747 | ||
748 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) | |
749 | { | |
d8805e63 | 750 | return ep_scan_ready_list(priv, ep_read_events_proc, NULL, call_nests + 1); |
5071f97e DL |
751 | } |
752 | ||
7699acd1 DL |
753 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
754 | { | |
5071f97e | 755 | int pollflags; |
7699acd1 | 756 | struct eventpoll *ep = file->private_data; |
1da177e4 | 757 | |
7699acd1 DL |
758 | /* Insert inside our poll wait queue */ |
759 | poll_wait(file, &ep->poll_wait, wait); | |
760 | ||
5071f97e DL |
761 | /* |
762 | * Proceed to find out if wanted events are really available inside | |
763 | * the ready list. This need to be done under ep_call_nested() | |
764 | * supervision, since the call to f_op->poll() done on listed files | |
765 | * could re-enter here. | |
766 | */ | |
767 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, | |
3fe4a975 | 768 | ep_poll_readyevents_proc, ep, ep, current); |
7699acd1 | 769 | |
296e236e | 770 | return pollflags != -1 ? pollflags : 0; |
7699acd1 DL |
771 | } |
772 | ||
773 | /* File callbacks that implement the eventpoll file behaviour */ | |
774 | static const struct file_operations eventpoll_fops = { | |
775 | .release = ep_eventpoll_release, | |
6038f373 AB |
776 | .poll = ep_eventpoll_poll, |
777 | .llseek = noop_llseek, | |
7699acd1 DL |
778 | }; |
779 | ||
b611967d | 780 | /* |
7699acd1 DL |
781 | * This is called from eventpoll_release() to unlink files from the eventpoll |
782 | * interface. We need to have this facility to cleanup correctly files that are | |
783 | * closed without being removed from the eventpoll interface. | |
b611967d | 784 | */ |
7699acd1 | 785 | void eventpoll_release_file(struct file *file) |
b611967d | 786 | { |
7699acd1 DL |
787 | struct list_head *lsthead = &file->f_ep_links; |
788 | struct eventpoll *ep; | |
789 | struct epitem *epi; | |
b611967d DL |
790 | |
791 | /* | |
68499914 | 792 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 793 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 794 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 795 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 796 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 797 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 798 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 799 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 800 | * from anywhere but ep_free(). |
68499914 JC |
801 | * |
802 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 803 | */ |
7699acd1 | 804 | mutex_lock(&epmutex); |
b611967d | 805 | |
7699acd1 DL |
806 | while (!list_empty(lsthead)) { |
807 | epi = list_first_entry(lsthead, struct epitem, fllink); | |
b611967d | 808 | |
7699acd1 DL |
809 | ep = epi->ep; |
810 | list_del_init(&epi->fllink); | |
d8805e63 | 811 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 812 | ep_remove(ep, epi); |
d47de16c | 813 | mutex_unlock(&ep->mtx); |
b611967d DL |
814 | } |
815 | ||
7699acd1 | 816 | mutex_unlock(&epmutex); |
b611967d DL |
817 | } |
818 | ||
53d2be79 | 819 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 820 | { |
7ef9964e DL |
821 | int error; |
822 | struct user_struct *user; | |
823 | struct eventpoll *ep; | |
1da177e4 | 824 | |
7ef9964e | 825 | user = get_current_user(); |
7ef9964e DL |
826 | error = -ENOMEM; |
827 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
828 | if (unlikely(!ep)) | |
829 | goto free_uid; | |
1da177e4 | 830 | |
c7ea7630 | 831 | spin_lock_init(&ep->lock); |
d47de16c | 832 | mutex_init(&ep->mtx); |
1da177e4 LT |
833 | init_waitqueue_head(&ep->wq); |
834 | init_waitqueue_head(&ep->poll_wait); | |
835 | INIT_LIST_HEAD(&ep->rdllist); | |
836 | ep->rbr = RB_ROOT; | |
d47de16c | 837 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 838 | ep->user = user; |
1da177e4 | 839 | |
53d2be79 | 840 | *pep = ep; |
1da177e4 | 841 | |
1da177e4 | 842 | return 0; |
7ef9964e DL |
843 | |
844 | free_uid: | |
845 | free_uid(user); | |
846 | return error; | |
1da177e4 LT |
847 | } |
848 | ||
1da177e4 | 849 | /* |
c7ea7630 DL |
850 | * Search the file inside the eventpoll tree. The RB tree operations |
851 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
852 | * "mtx" held. | |
1da177e4 LT |
853 | */ |
854 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
855 | { | |
856 | int kcmp; | |
1da177e4 LT |
857 | struct rb_node *rbp; |
858 | struct epitem *epi, *epir = NULL; | |
859 | struct epoll_filefd ffd; | |
860 | ||
b030a4dd | 861 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
862 | for (rbp = ep->rbr.rb_node; rbp; ) { |
863 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 864 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
865 | if (kcmp > 0) |
866 | rbp = rbp->rb_right; | |
867 | else if (kcmp < 0) | |
868 | rbp = rbp->rb_left; | |
869 | else { | |
1da177e4 LT |
870 | epir = epi; |
871 | break; | |
872 | } | |
873 | } | |
1da177e4 | 874 | |
1da177e4 LT |
875 | return epir; |
876 | } | |
877 | ||
1da177e4 | 878 | /* |
7699acd1 | 879 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 880 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 881 | * have events to report. |
1da177e4 | 882 | */ |
7699acd1 | 883 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 884 | { |
7699acd1 DL |
885 | int pwake = 0; |
886 | unsigned long flags; | |
887 | struct epitem *epi = ep_item_from_wait(wait); | |
888 | struct eventpoll *ep = epi->ep; | |
1da177e4 | 889 | |
971316f0 ON |
890 | if ((unsigned long)key & POLLFREE) { |
891 | ep_pwq_from_wait(wait)->whead = NULL; | |
892 | /* | |
893 | * whead = NULL above can race with ep_remove_wait_queue() | |
894 | * which can do another remove_wait_queue() after us, so we | |
895 | * can't use __remove_wait_queue(). whead->lock is held by | |
896 | * the caller. | |
897 | */ | |
d80e731e | 898 | list_del_init(&wait->task_list); |
971316f0 | 899 | } |
d80e731e | 900 | |
c7ea7630 | 901 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 902 | |
7699acd1 DL |
903 | /* |
904 | * If the event mask does not contain any poll(2) event, we consider the | |
905 | * descriptor to be disabled. This condition is likely the effect of the | |
906 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
907 | * until the next EPOLL_CTL_MOD will be issued. | |
908 | */ | |
909 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
910 | goto out_unlock; |
911 | ||
2dfa4eea DL |
912 | /* |
913 | * Check the events coming with the callback. At this stage, not | |
914 | * every device reports the events in the "key" parameter of the | |
915 | * callback. We need to be able to handle both cases here, hence the | |
916 | * test for "key" != NULL before the event match test. | |
917 | */ | |
918 | if (key && !((unsigned long) key & epi->event.events)) | |
919 | goto out_unlock; | |
920 | ||
d47de16c | 921 | /* |
bf6a41db | 922 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 923 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 924 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
925 | * chained in ep->ovflist and requeued later on. |
926 | */ | |
927 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
928 | if (epi->next == EP_UNACTIVE_PTR) { | |
929 | epi->next = ep->ovflist; | |
930 | ep->ovflist = epi; | |
931 | } | |
932 | goto out_unlock; | |
933 | } | |
1da177e4 | 934 | |
7699acd1 | 935 | /* If this file is already in the ready list we exit soon */ |
5071f97e DL |
936 | if (!ep_is_linked(&epi->rdllink)) |
937 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
7699acd1 | 938 | |
7699acd1 DL |
939 | /* |
940 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
941 | * wait list. | |
942 | */ | |
943 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 944 | wake_up_locked(&ep->wq); |
7699acd1 DL |
945 | if (waitqueue_active(&ep->poll_wait)) |
946 | pwake++; | |
947 | ||
d47de16c | 948 | out_unlock: |
c7ea7630 | 949 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 950 | |
7699acd1 DL |
951 | /* We have to call this outside the lock */ |
952 | if (pwake) | |
5071f97e | 953 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 DL |
954 | |
955 | return 1; | |
956 | } | |
1da177e4 LT |
957 | |
958 | /* | |
959 | * This is the callback that is used to add our wait queue to the | |
960 | * target file wakeup lists. | |
961 | */ | |
962 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
963 | poll_table *pt) | |
964 | { | |
b030a4dd | 965 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
966 | struct eppoll_entry *pwq; |
967 | ||
e94b1766 | 968 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
969 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
970 | pwq->whead = whead; | |
971 | pwq->base = epi; | |
972 | add_wait_queue(whead, &pwq->wait); | |
973 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
974 | epi->nwait++; | |
296e236e | 975 | } else { |
1da177e4 LT |
976 | /* We have to signal that an error occurred */ |
977 | epi->nwait = -1; | |
296e236e | 978 | } |
1da177e4 LT |
979 | } |
980 | ||
1da177e4 LT |
981 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
982 | { | |
983 | int kcmp; | |
984 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
985 | struct epitem *epic; | |
986 | ||
987 | while (*p) { | |
988 | parent = *p; | |
989 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 990 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
991 | if (kcmp > 0) |
992 | p = &parent->rb_right; | |
993 | else | |
994 | p = &parent->rb_left; | |
995 | } | |
996 | rb_link_node(&epi->rbn, parent, p); | |
997 | rb_insert_color(&epi->rbn, &ep->rbr); | |
998 | } | |
999 | ||
28d82dc1 JB |
1000 | |
1001 | ||
1002 | #define PATH_ARR_SIZE 5 | |
1003 | /* | |
1004 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1005 | * from a single file of interest. For example, we allow 1000 paths of length | |
1006 | * 1, to emanate from each file of interest. This essentially represents the | |
1007 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1008 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1009 | * is connected to n file sources. In this case each file source has 1 path | |
1010 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1011 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1012 | * and delete can't add additional paths. Protected by the epmutex. | |
1013 | */ | |
1014 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1015 | static int path_count[PATH_ARR_SIZE]; | |
1016 | ||
1017 | static int path_count_inc(int nests) | |
1018 | { | |
93dc6107 JB |
1019 | /* Allow an arbitrary number of depth 1 paths */ |
1020 | if (nests == 0) | |
1021 | return 0; | |
1022 | ||
28d82dc1 JB |
1023 | if (++path_count[nests] > path_limits[nests]) |
1024 | return -1; | |
1025 | return 0; | |
1026 | } | |
1027 | ||
1028 | static void path_count_init(void) | |
1029 | { | |
1030 | int i; | |
1031 | ||
1032 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1033 | path_count[i] = 0; | |
1034 | } | |
1035 | ||
1036 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) | |
1037 | { | |
1038 | int error = 0; | |
1039 | struct file *file = priv; | |
1040 | struct file *child_file; | |
1041 | struct epitem *epi; | |
1042 | ||
1043 | list_for_each_entry(epi, &file->f_ep_links, fllink) { | |
1044 | child_file = epi->ep->file; | |
1045 | if (is_file_epoll(child_file)) { | |
1046 | if (list_empty(&child_file->f_ep_links)) { | |
1047 | if (path_count_inc(call_nests)) { | |
1048 | error = -1; | |
1049 | break; | |
1050 | } | |
1051 | } else { | |
1052 | error = ep_call_nested(&poll_loop_ncalls, | |
1053 | EP_MAX_NESTS, | |
1054 | reverse_path_check_proc, | |
1055 | child_file, child_file, | |
1056 | current); | |
1057 | } | |
1058 | if (error != 0) | |
1059 | break; | |
1060 | } else { | |
1061 | printk(KERN_ERR "reverse_path_check_proc: " | |
1062 | "file is not an ep!\n"); | |
1063 | } | |
1064 | } | |
1065 | return error; | |
1066 | } | |
1067 | ||
1068 | /** | |
1069 | * reverse_path_check - The tfile_check_list is list of file *, which have | |
1070 | * links that are proposed to be newly added. We need to | |
1071 | * make sure that those added links don't add too many | |
1072 | * paths such that we will spend all our time waking up | |
1073 | * eventpoll objects. | |
1074 | * | |
1075 | * Returns: Returns zero if the proposed links don't create too many paths, | |
1076 | * -1 otherwise. | |
1077 | */ | |
1078 | static int reverse_path_check(void) | |
1079 | { | |
28d82dc1 JB |
1080 | int error = 0; |
1081 | struct file *current_file; | |
1082 | ||
1083 | /* let's call this for all tfiles */ | |
1084 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { | |
28d82dc1 JB |
1085 | path_count_init(); |
1086 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
1087 | reverse_path_check_proc, current_file, | |
1088 | current_file, current); | |
1089 | if (error) | |
1090 | break; | |
1091 | } | |
1092 | return error; | |
1093 | } | |
1094 | ||
c7ea7630 DL |
1095 | /* |
1096 | * Must be called with "mtx" held. | |
1097 | */ | |
1da177e4 LT |
1098 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
1099 | struct file *tfile, int fd) | |
1100 | { | |
1101 | int error, revents, pwake = 0; | |
1102 | unsigned long flags; | |
52bd19f7 | 1103 | long user_watches; |
1da177e4 LT |
1104 | struct epitem *epi; |
1105 | struct ep_pqueue epq; | |
1106 | ||
52bd19f7 RH |
1107 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
1108 | if (unlikely(user_watches >= max_user_watches)) | |
7ef9964e | 1109 | return -ENOSPC; |
e94b1766 | 1110 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 1111 | return -ENOMEM; |
1da177e4 LT |
1112 | |
1113 | /* Item initialization follow here ... */ | |
1da177e4 LT |
1114 | INIT_LIST_HEAD(&epi->rdllink); |
1115 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
1116 | INIT_LIST_HEAD(&epi->pwqlist); |
1117 | epi->ep = ep; | |
b030a4dd | 1118 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1119 | epi->event = *event; |
1da177e4 | 1120 | epi->nwait = 0; |
d47de16c | 1121 | epi->next = EP_UNACTIVE_PTR; |
1da177e4 LT |
1122 | |
1123 | /* Initialize the poll table using the queue callback */ | |
1124 | epq.epi = epi; | |
1125 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
626cf236 | 1126 | epq.pt._key = event->events; |
1da177e4 LT |
1127 | |
1128 | /* | |
1129 | * Attach the item to the poll hooks and get current event bits. | |
1130 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1131 | * been increased by the caller of this function. Note that after |
1132 | * this operation completes, the poll callback can start hitting | |
1133 | * the new item. | |
1da177e4 LT |
1134 | */ |
1135 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
1136 | ||
1137 | /* | |
1138 | * We have to check if something went wrong during the poll wait queue | |
1139 | * install process. Namely an allocation for a wait queue failed due | |
1140 | * high memory pressure. | |
1141 | */ | |
7ef9964e | 1142 | error = -ENOMEM; |
1da177e4 | 1143 | if (epi->nwait < 0) |
7699acd1 | 1144 | goto error_unregister; |
1da177e4 LT |
1145 | |
1146 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 1147 | spin_lock(&tfile->f_lock); |
1da177e4 | 1148 | list_add_tail(&epi->fllink, &tfile->f_ep_links); |
68499914 | 1149 | spin_unlock(&tfile->f_lock); |
1da177e4 | 1150 | |
c7ea7630 DL |
1151 | /* |
1152 | * Add the current item to the RB tree. All RB tree operations are | |
1153 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1154 | */ | |
1da177e4 LT |
1155 | ep_rbtree_insert(ep, epi); |
1156 | ||
28d82dc1 JB |
1157 | /* now check if we've created too many backpaths */ |
1158 | error = -EINVAL; | |
1159 | if (reverse_path_check()) | |
1160 | goto error_remove_epi; | |
1161 | ||
c7ea7630 DL |
1162 | /* We have to drop the new item inside our item list to keep track of it */ |
1163 | spin_lock_irqsave(&ep->lock, flags); | |
1164 | ||
1da177e4 | 1165 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 1166 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 LT |
1167 | list_add_tail(&epi->rdllink, &ep->rdllist); |
1168 | ||
1169 | /* Notify waiting tasks that events are available */ | |
1170 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1171 | wake_up_locked(&ep->wq); |
1da177e4 LT |
1172 | if (waitqueue_active(&ep->poll_wait)) |
1173 | pwake++; | |
1174 | } | |
1175 | ||
c7ea7630 | 1176 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1177 | |
52bd19f7 | 1178 | atomic_long_inc(&ep->user->epoll_watches); |
7ef9964e | 1179 | |
1da177e4 LT |
1180 | /* We have to call this outside the lock */ |
1181 | if (pwake) | |
5071f97e | 1182 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1183 | |
1da177e4 LT |
1184 | return 0; |
1185 | ||
28d82dc1 JB |
1186 | error_remove_epi: |
1187 | spin_lock(&tfile->f_lock); | |
1188 | if (ep_is_linked(&epi->fllink)) | |
1189 | list_del_init(&epi->fllink); | |
1190 | spin_unlock(&tfile->f_lock); | |
1191 | ||
1192 | rb_erase(&epi->rbn, &ep->rbr); | |
1193 | ||
7699acd1 | 1194 | error_unregister: |
1da177e4 LT |
1195 | ep_unregister_pollwait(ep, epi); |
1196 | ||
1197 | /* | |
1198 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
1199 | * allocated wait queue. Note that we don't care about the ep->ovflist |
1200 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
1201 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 1202 | */ |
c7ea7630 | 1203 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 1204 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 1205 | list_del_init(&epi->rdllink); |
c7ea7630 | 1206 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1207 | |
b030a4dd | 1208 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1209 | |
1da177e4 LT |
1210 | return error; |
1211 | } | |
1212 | ||
1da177e4 LT |
1213 | /* |
1214 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1215 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
1216 | */ |
1217 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1218 | { | |
1219 | int pwake = 0; | |
1220 | unsigned int revents; | |
626cf236 HV |
1221 | poll_table pt; |
1222 | ||
1223 | init_poll_funcptr(&pt, NULL); | |
1da177e4 LT |
1224 | |
1225 | /* | |
e057e15f TB |
1226 | * Set the new event interest mask before calling f_op->poll(); |
1227 | * otherwise we might miss an event that happens between the | |
1228 | * f_op->poll() call and the new event set registering. | |
1da177e4 LT |
1229 | */ |
1230 | epi->event.events = event->events; | |
626cf236 | 1231 | pt._key = event->events; |
e057e15f | 1232 | epi->event.data = event->data; /* protected by mtx */ |
1da177e4 LT |
1233 | |
1234 | /* | |
1235 | * Get current event bits. We can safely use the file* here because | |
1236 | * its usage count has been increased by the caller of this function. | |
1237 | */ | |
626cf236 | 1238 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt); |
1da177e4 | 1239 | |
1da177e4 | 1240 | /* |
c7ea7630 | 1241 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1242 | * list, push it inside. |
1da177e4 | 1243 | */ |
c7ea7630 | 1244 | if (revents & event->events) { |
e057e15f | 1245 | spin_lock_irq(&ep->lock); |
c7ea7630 DL |
1246 | if (!ep_is_linked(&epi->rdllink)) { |
1247 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1248 | ||
1249 | /* Notify waiting tasks that events are available */ | |
1250 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1251 | wake_up_locked(&ep->wq); |
c7ea7630 DL |
1252 | if (waitqueue_active(&ep->poll_wait)) |
1253 | pwake++; | |
7699acd1 | 1254 | } |
e057e15f | 1255 | spin_unlock_irq(&ep->lock); |
7699acd1 | 1256 | } |
1da177e4 | 1257 | |
7699acd1 DL |
1258 | /* We have to call this outside the lock */ |
1259 | if (pwake) | |
5071f97e | 1260 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1261 | |
7699acd1 | 1262 | return 0; |
1da177e4 LT |
1263 | } |
1264 | ||
296e236e DL |
1265 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
1266 | void *priv) | |
1da177e4 | 1267 | { |
5071f97e DL |
1268 | struct ep_send_events_data *esed = priv; |
1269 | int eventcnt; | |
296e236e | 1270 | unsigned int revents; |
5071f97e DL |
1271 | struct epitem *epi; |
1272 | struct epoll_event __user *uevent; | |
626cf236 HV |
1273 | poll_table pt; |
1274 | ||
1275 | init_poll_funcptr(&pt, NULL); | |
1da177e4 | 1276 | |
296e236e | 1277 | /* |
5071f97e DL |
1278 | * We can loop without lock because we are passed a task private list. |
1279 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1280 | * holding "mtx" during this call. | |
296e236e | 1281 | */ |
5071f97e DL |
1282 | for (eventcnt = 0, uevent = esed->events; |
1283 | !list_empty(head) && eventcnt < esed->maxevents;) { | |
1284 | epi = list_first_entry(head, struct epitem, rdllink); | |
d47de16c DL |
1285 | |
1286 | list_del_init(&epi->rdllink); | |
1da177e4 | 1287 | |
626cf236 HV |
1288 | pt._key = epi->event.events; |
1289 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt) & | |
296e236e | 1290 | epi->event.events; |
5071f97e | 1291 | |
296e236e | 1292 | /* |
5071f97e DL |
1293 | * If the event mask intersect the caller-requested one, |
1294 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
1295 | * is holding "mtx", so no operations coming from userspace | |
1296 | * can change the item. | |
296e236e DL |
1297 | */ |
1298 | if (revents) { | |
5071f97e | 1299 | if (__put_user(revents, &uevent->events) || |
d0305882 TB |
1300 | __put_user(epi->event.data, &uevent->data)) { |
1301 | list_add(&epi->rdllink, head); | |
296e236e | 1302 | return eventcnt ? eventcnt : -EFAULT; |
d0305882 | 1303 | } |
296e236e | 1304 | eventcnt++; |
5071f97e | 1305 | uevent++; |
296e236e DL |
1306 | if (epi->event.events & EPOLLONESHOT) |
1307 | epi->event.events &= EP_PRIVATE_BITS; | |
1308 | else if (!(epi->event.events & EPOLLET)) { | |
1309 | /* | |
1310 | * If this file has been added with Level | |
1311 | * Trigger mode, we need to insert back inside | |
1312 | * the ready list, so that the next call to | |
1313 | * epoll_wait() will check again the events | |
25985edc | 1314 | * availability. At this point, no one can insert |
296e236e DL |
1315 | * into ep->rdllist besides us. The epoll_ctl() |
1316 | * callers are locked out by | |
1317 | * ep_scan_ready_list() holding "mtx" and the | |
1318 | * poll callback will queue them in ep->ovflist. | |
1319 | */ | |
1320 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1321 | } | |
1322 | } | |
1323 | } | |
5071f97e DL |
1324 | |
1325 | return eventcnt; | |
1326 | } | |
d47de16c | 1327 | |
296e236e DL |
1328 | static int ep_send_events(struct eventpoll *ep, |
1329 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1330 | { |
1331 | struct ep_send_events_data esed; | |
1da177e4 | 1332 | |
5071f97e DL |
1333 | esed.maxevents = maxevents; |
1334 | esed.events = events; | |
6192bd53 | 1335 | |
d8805e63 | 1336 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0); |
1da177e4 LT |
1337 | } |
1338 | ||
0781b909 ED |
1339 | static inline struct timespec ep_set_mstimeout(long ms) |
1340 | { | |
1341 | struct timespec now, ts = { | |
1342 | .tv_sec = ms / MSEC_PER_SEC, | |
1343 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), | |
1344 | }; | |
1345 | ||
1346 | ktime_get_ts(&now); | |
1347 | return timespec_add_safe(now, ts); | |
1348 | } | |
1349 | ||
f4d93ad7 SB |
1350 | /** |
1351 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied | |
1352 | * event buffer. | |
1353 | * | |
1354 | * @ep: Pointer to the eventpoll context. | |
1355 | * @events: Pointer to the userspace buffer where the ready events should be | |
1356 | * stored. | |
1357 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1358 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
1359 | * milliseconds. If the @timeout is zero, the function will not block, | |
1360 | * while if the @timeout is less than zero, the function will block | |
1361 | * until at least one event has been retrieved (or an error | |
1362 | * occurred). | |
1363 | * | |
1364 | * Returns: Returns the number of ready events which have been fetched, or an | |
1365 | * error code, in case of error. | |
1366 | */ | |
1da177e4 LT |
1367 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1368 | int maxevents, long timeout) | |
1369 | { | |
f4d93ad7 | 1370 | int res = 0, eavail, timed_out = 0; |
1da177e4 | 1371 | unsigned long flags; |
f4d93ad7 | 1372 | long slack = 0; |
1da177e4 | 1373 | wait_queue_t wait; |
95aac7b1 SB |
1374 | ktime_t expires, *to = NULL; |
1375 | ||
1376 | if (timeout > 0) { | |
0781b909 ED |
1377 | struct timespec end_time = ep_set_mstimeout(timeout); |
1378 | ||
95aac7b1 SB |
1379 | slack = select_estimate_accuracy(&end_time); |
1380 | to = &expires; | |
1381 | *to = timespec_to_ktime(end_time); | |
1382 | } else if (timeout == 0) { | |
f4d93ad7 SB |
1383 | /* |
1384 | * Avoid the unnecessary trip to the wait queue loop, if the | |
1385 | * caller specified a non blocking operation. | |
1386 | */ | |
95aac7b1 | 1387 | timed_out = 1; |
f4d93ad7 SB |
1388 | spin_lock_irqsave(&ep->lock, flags); |
1389 | goto check_events; | |
95aac7b1 | 1390 | } |
1da177e4 | 1391 | |
f4d93ad7 | 1392 | fetch_events: |
c7ea7630 | 1393 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1394 | |
3fb0e584 | 1395 | if (!ep_events_available(ep)) { |
1da177e4 LT |
1396 | /* |
1397 | * We don't have any available event to return to the caller. | |
1398 | * We need to sleep here, and we will be wake up by | |
1399 | * ep_poll_callback() when events will become available. | |
1400 | */ | |
1401 | init_waitqueue_entry(&wait, current); | |
a93d2f17 | 1402 | __add_wait_queue_exclusive(&ep->wq, &wait); |
1da177e4 LT |
1403 | |
1404 | for (;;) { | |
1405 | /* | |
1406 | * We don't want to sleep if the ep_poll_callback() sends us | |
1407 | * a wakeup in between. That's why we set the task state | |
1408 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1409 | */ | |
1410 | set_current_state(TASK_INTERRUPTIBLE); | |
3fb0e584 | 1411 | if (ep_events_available(ep) || timed_out) |
1da177e4 LT |
1412 | break; |
1413 | if (signal_pending(current)) { | |
1414 | res = -EINTR; | |
1415 | break; | |
1416 | } | |
1417 | ||
c7ea7630 | 1418 | spin_unlock_irqrestore(&ep->lock, flags); |
95aac7b1 SB |
1419 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) |
1420 | timed_out = 1; | |
1421 | ||
c7ea7630 | 1422 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1423 | } |
3419b23a | 1424 | __remove_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1425 | |
1426 | set_current_state(TASK_RUNNING); | |
1427 | } | |
f4d93ad7 | 1428 | check_events: |
1da177e4 | 1429 | /* Is it worth to try to dig for events ? */ |
3fb0e584 | 1430 | eavail = ep_events_available(ep); |
1da177e4 | 1431 | |
c7ea7630 | 1432 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1433 | |
1434 | /* | |
1435 | * Try to transfer events to user space. In case we get 0 events and | |
1436 | * there's still timeout left over, we go trying again in search of | |
1437 | * more luck. | |
1438 | */ | |
1439 | if (!res && eavail && | |
95aac7b1 | 1440 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
f4d93ad7 | 1441 | goto fetch_events; |
1da177e4 LT |
1442 | |
1443 | return res; | |
1444 | } | |
1445 | ||
22bacca4 DL |
1446 | /** |
1447 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() | |
1448 | * API, to verify that adding an epoll file inside another | |
1449 | * epoll structure, does not violate the constraints, in | |
1450 | * terms of closed loops, or too deep chains (which can | |
1451 | * result in excessive stack usage). | |
1452 | * | |
1453 | * @priv: Pointer to the epoll file to be currently checked. | |
1454 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll | |
1455 | * data structure pointer. | |
1456 | * @call_nests: Current dept of the @ep_call_nested() call stack. | |
1457 | * | |
1458 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1459 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1460 | */ | |
1461 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) | |
1462 | { | |
1463 | int error = 0; | |
1464 | struct file *file = priv; | |
1465 | struct eventpoll *ep = file->private_data; | |
28d82dc1 | 1466 | struct eventpoll *ep_tovisit; |
22bacca4 DL |
1467 | struct rb_node *rbp; |
1468 | struct epitem *epi; | |
1469 | ||
d8805e63 | 1470 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
28d82dc1 JB |
1471 | ep->visited = 1; |
1472 | list_add(&ep->visited_list_link, &visited_list); | |
22bacca4 DL |
1473 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
1474 | epi = rb_entry(rbp, struct epitem, rbn); | |
1475 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
28d82dc1 JB |
1476 | ep_tovisit = epi->ffd.file->private_data; |
1477 | if (ep_tovisit->visited) | |
1478 | continue; | |
22bacca4 | 1479 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
28d82dc1 JB |
1480 | ep_loop_check_proc, epi->ffd.file, |
1481 | ep_tovisit, current); | |
22bacca4 DL |
1482 | if (error != 0) |
1483 | break; | |
28d82dc1 JB |
1484 | } else { |
1485 | /* | |
1486 | * If we've reached a file that is not associated with | |
1487 | * an ep, then we need to check if the newly added | |
1488 | * links are going to add too many wakeup paths. We do | |
1489 | * this by adding it to the tfile_check_list, if it's | |
1490 | * not already there, and calling reverse_path_check() | |
1491 | * during ep_insert(). | |
1492 | */ | |
1493 | if (list_empty(&epi->ffd.file->f_tfile_llink)) | |
1494 | list_add(&epi->ffd.file->f_tfile_llink, | |
1495 | &tfile_check_list); | |
22bacca4 DL |
1496 | } |
1497 | } | |
1498 | mutex_unlock(&ep->mtx); | |
1499 | ||
1500 | return error; | |
1501 | } | |
1502 | ||
1503 | /** | |
1504 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) | |
1505 | * another epoll file (represented by @ep) does not create | |
1506 | * closed loops or too deep chains. | |
1507 | * | |
1508 | * @ep: Pointer to the epoll private data structure. | |
1509 | * @file: Pointer to the epoll file to be checked. | |
1510 | * | |
1511 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1512 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1513 | */ | |
1514 | static int ep_loop_check(struct eventpoll *ep, struct file *file) | |
1515 | { | |
28d82dc1 JB |
1516 | int ret; |
1517 | struct eventpoll *ep_cur, *ep_next; | |
1518 | ||
1519 | ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
22bacca4 | 1520 | ep_loop_check_proc, file, ep, current); |
28d82dc1 JB |
1521 | /* clear visited list */ |
1522 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, | |
1523 | visited_list_link) { | |
1524 | ep_cur->visited = 0; | |
1525 | list_del(&ep_cur->visited_list_link); | |
1526 | } | |
1527 | return ret; | |
1528 | } | |
1529 | ||
1530 | static void clear_tfile_check_list(void) | |
1531 | { | |
1532 | struct file *file; | |
1533 | ||
1534 | /* first clear the tfile_check_list */ | |
1535 | while (!list_empty(&tfile_check_list)) { | |
1536 | file = list_first_entry(&tfile_check_list, struct file, | |
1537 | f_tfile_llink); | |
1538 | list_del_init(&file->f_tfile_llink); | |
1539 | } | |
1540 | INIT_LIST_HEAD(&tfile_check_list); | |
22bacca4 DL |
1541 | } |
1542 | ||
7699acd1 | 1543 | /* |
523723bb | 1544 | * Open an eventpoll file descriptor. |
7699acd1 | 1545 | */ |
5a8a82b1 | 1546 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
7699acd1 | 1547 | { |
28d82dc1 | 1548 | int error, fd; |
bb57c3ed | 1549 | struct eventpoll *ep = NULL; |
28d82dc1 | 1550 | struct file *file; |
7699acd1 | 1551 | |
e38b36f3 UD |
1552 | /* Check the EPOLL_* constant for consistency. */ |
1553 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1554 | ||
296e236e DL |
1555 | if (flags & ~EPOLL_CLOEXEC) |
1556 | return -EINVAL; | |
7699acd1 | 1557 | /* |
bb57c3ed | 1558 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 1559 | */ |
9fe5ad9c | 1560 | error = ep_alloc(&ep); |
bb57c3ed DL |
1561 | if (error < 0) |
1562 | return error; | |
7699acd1 DL |
1563 | /* |
1564 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 1565 | * a file structure and a free file descriptor. |
7699acd1 | 1566 | */ |
28d82dc1 JB |
1567 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
1568 | if (fd < 0) { | |
1569 | error = fd; | |
1570 | goto out_free_ep; | |
1571 | } | |
1572 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 1573 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
1574 | if (IS_ERR(file)) { |
1575 | error = PTR_ERR(file); | |
1576 | goto out_free_fd; | |
1577 | } | |
1578 | fd_install(fd, file); | |
1579 | ep->file = file; | |
1580 | return fd; | |
1581 | ||
1582 | out_free_fd: | |
1583 | put_unused_fd(fd); | |
1584 | out_free_ep: | |
1585 | ep_free(ep); | |
bb57c3ed | 1586 | return error; |
7699acd1 DL |
1587 | } |
1588 | ||
5a8a82b1 | 1589 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 1590 | { |
bfe3891a | 1591 | if (size <= 0) |
9fe5ad9c UD |
1592 | return -EINVAL; |
1593 | ||
1594 | return sys_epoll_create1(0); | |
a0998b50 UD |
1595 | } |
1596 | ||
7699acd1 DL |
1597 | /* |
1598 | * The following function implements the controller interface for | |
1599 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 1600 | * file descriptors inside the interest set. |
7699acd1 | 1601 | */ |
5a8a82b1 HC |
1602 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
1603 | struct epoll_event __user *, event) | |
7699acd1 DL |
1604 | { |
1605 | int error; | |
22bacca4 | 1606 | int did_lock_epmutex = 0; |
7699acd1 DL |
1607 | struct file *file, *tfile; |
1608 | struct eventpoll *ep; | |
1609 | struct epitem *epi; | |
1610 | struct epoll_event epds; | |
1611 | ||
7699acd1 DL |
1612 | error = -EFAULT; |
1613 | if (ep_op_has_event(op) && | |
1614 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
1615 | goto error_return; | |
1616 | ||
1617 | /* Get the "struct file *" for the eventpoll file */ | |
1618 | error = -EBADF; | |
1619 | file = fget(epfd); | |
1620 | if (!file) | |
1621 | goto error_return; | |
1622 | ||
1623 | /* Get the "struct file *" for the target file */ | |
1624 | tfile = fget(fd); | |
1625 | if (!tfile) | |
1626 | goto error_fput; | |
1627 | ||
1628 | /* The target file descriptor must support poll */ | |
1629 | error = -EPERM; | |
1630 | if (!tfile->f_op || !tfile->f_op->poll) | |
1631 | goto error_tgt_fput; | |
1632 | ||
1633 | /* | |
1634 | * We have to check that the file structure underneath the file descriptor | |
1635 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
1636 | * adding an epoll file descriptor inside itself. | |
1637 | */ | |
1638 | error = -EINVAL; | |
1639 | if (file == tfile || !is_file_epoll(file)) | |
1640 | goto error_tgt_fput; | |
1641 | ||
1642 | /* | |
1643 | * At this point it is safe to assume that the "private_data" contains | |
1644 | * our own data structure. | |
1645 | */ | |
1646 | ep = file->private_data; | |
1647 | ||
22bacca4 DL |
1648 | /* |
1649 | * When we insert an epoll file descriptor, inside another epoll file | |
1650 | * descriptor, there is the change of creating closed loops, which are | |
28d82dc1 JB |
1651 | * better be handled here, than in more critical paths. While we are |
1652 | * checking for loops we also determine the list of files reachable | |
1653 | * and hang them on the tfile_check_list, so we can check that we | |
1654 | * haven't created too many possible wakeup paths. | |
22bacca4 | 1655 | * |
28d82dc1 JB |
1656 | * We need to hold the epmutex across both ep_insert and ep_remove |
1657 | * b/c we want to make sure we are looking at a coherent view of | |
1658 | * epoll network. | |
22bacca4 | 1659 | */ |
28d82dc1 | 1660 | if (op == EPOLL_CTL_ADD || op == EPOLL_CTL_DEL) { |
22bacca4 DL |
1661 | mutex_lock(&epmutex); |
1662 | did_lock_epmutex = 1; | |
22bacca4 | 1663 | } |
28d82dc1 JB |
1664 | if (op == EPOLL_CTL_ADD) { |
1665 | if (is_file_epoll(tfile)) { | |
1666 | error = -ELOOP; | |
1667 | if (ep_loop_check(ep, tfile) != 0) | |
1668 | goto error_tgt_fput; | |
1669 | } else | |
1670 | list_add(&tfile->f_tfile_llink, &tfile_check_list); | |
1671 | } | |
22bacca4 | 1672 | |
d8805e63 | 1673 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 1674 | |
67647d0f DL |
1675 | /* |
1676 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
1677 | * above, we can be sure to be able to use the item looked up by | |
1678 | * ep_find() till we release the mutex. | |
1679 | */ | |
7699acd1 DL |
1680 | epi = ep_find(ep, tfile, fd); |
1681 | ||
1682 | error = -EINVAL; | |
1683 | switch (op) { | |
1684 | case EPOLL_CTL_ADD: | |
1685 | if (!epi) { | |
1686 | epds.events |= POLLERR | POLLHUP; | |
7699acd1 DL |
1687 | error = ep_insert(ep, &epds, tfile, fd); |
1688 | } else | |
1689 | error = -EEXIST; | |
28d82dc1 | 1690 | clear_tfile_check_list(); |
7699acd1 DL |
1691 | break; |
1692 | case EPOLL_CTL_DEL: | |
1693 | if (epi) | |
1694 | error = ep_remove(ep, epi); | |
1695 | else | |
1696 | error = -ENOENT; | |
1697 | break; | |
1698 | case EPOLL_CTL_MOD: | |
1699 | if (epi) { | |
1700 | epds.events |= POLLERR | POLLHUP; | |
1701 | error = ep_modify(ep, epi, &epds); | |
1702 | } else | |
1703 | error = -ENOENT; | |
1704 | break; | |
1705 | } | |
d47de16c | 1706 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
1707 | |
1708 | error_tgt_fput: | |
28d82dc1 | 1709 | if (did_lock_epmutex) |
22bacca4 DL |
1710 | mutex_unlock(&epmutex); |
1711 | ||
7699acd1 DL |
1712 | fput(tfile); |
1713 | error_fput: | |
1714 | fput(file); | |
1715 | error_return: | |
7699acd1 DL |
1716 | |
1717 | return error; | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1722 | * part of the user space epoll_wait(2). | |
1723 | */ | |
5a8a82b1 HC |
1724 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
1725 | int, maxevents, int, timeout) | |
7699acd1 DL |
1726 | { |
1727 | int error; | |
1728 | struct file *file; | |
1729 | struct eventpoll *ep; | |
1730 | ||
7699acd1 DL |
1731 | /* The maximum number of event must be greater than zero */ |
1732 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
1733 | return -EINVAL; | |
1734 | ||
1735 | /* Verify that the area passed by the user is writeable */ | |
1736 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { | |
1737 | error = -EFAULT; | |
1738 | goto error_return; | |
1739 | } | |
1740 | ||
1741 | /* Get the "struct file *" for the eventpoll file */ | |
1742 | error = -EBADF; | |
1743 | file = fget(epfd); | |
1744 | if (!file) | |
1745 | goto error_return; | |
1746 | ||
1747 | /* | |
1748 | * We have to check that the file structure underneath the fd | |
1749 | * the user passed to us _is_ an eventpoll file. | |
1750 | */ | |
1751 | error = -EINVAL; | |
1752 | if (!is_file_epoll(file)) | |
1753 | goto error_fput; | |
1754 | ||
1755 | /* | |
1756 | * At this point it is safe to assume that the "private_data" contains | |
1757 | * our own data structure. | |
1758 | */ | |
1759 | ep = file->private_data; | |
1760 | ||
1761 | /* Time to fish for events ... */ | |
1762 | error = ep_poll(ep, events, maxevents, timeout); | |
1763 | ||
1764 | error_fput: | |
1765 | fput(file); | |
1766 | error_return: | |
7699acd1 DL |
1767 | |
1768 | return error; | |
1769 | } | |
1770 | ||
f3de272b | 1771 | #ifdef HAVE_SET_RESTORE_SIGMASK |
7699acd1 DL |
1772 | |
1773 | /* | |
1774 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1775 | * part of the user space epoll_pwait(2). | |
1776 | */ | |
5a8a82b1 HC |
1777 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
1778 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
1779 | size_t, sigsetsize) | |
7699acd1 DL |
1780 | { |
1781 | int error; | |
1782 | sigset_t ksigmask, sigsaved; | |
1783 | ||
1784 | /* | |
1785 | * If the caller wants a certain signal mask to be set during the wait, | |
1786 | * we apply it here. | |
1787 | */ | |
1788 | if (sigmask) { | |
1789 | if (sigsetsize != sizeof(sigset_t)) | |
1790 | return -EINVAL; | |
1791 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1792 | return -EFAULT; | |
1793 | sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP)); | |
1794 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1795 | } | |
1796 | ||
1797 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
1798 | ||
1799 | /* | |
1800 | * If we changed the signal mask, we need to restore the original one. | |
1801 | * In case we've got a signal while waiting, we do not restore the | |
1802 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
1803 | * the way back to userspace, before the signal mask is restored. | |
1804 | */ | |
1805 | if (sigmask) { | |
1806 | if (error == -EINTR) { | |
1807 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 1808 | sizeof(sigsaved)); |
4e4c22c7 | 1809 | set_restore_sigmask(); |
7699acd1 DL |
1810 | } else |
1811 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1812 | } | |
1813 | ||
1814 | return error; | |
1815 | } | |
1816 | ||
f3de272b | 1817 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
7699acd1 | 1818 | |
1da177e4 LT |
1819 | static int __init eventpoll_init(void) |
1820 | { | |
7ef9964e DL |
1821 | struct sysinfo si; |
1822 | ||
1823 | si_meminfo(&si); | |
9df04e1f DL |
1824 | /* |
1825 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
1826 | */ | |
1827 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 1828 | EP_ITEM_COST; |
52bd19f7 | 1829 | BUG_ON(max_user_watches < 0); |
1da177e4 | 1830 | |
22bacca4 DL |
1831 | /* |
1832 | * Initialize the structure used to perform epoll file descriptor | |
1833 | * inclusion loops checks. | |
1834 | */ | |
1835 | ep_nested_calls_init(&poll_loop_ncalls); | |
1836 | ||
1da177e4 | 1837 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
5071f97e DL |
1838 | ep_nested_calls_init(&poll_safewake_ncalls); |
1839 | ||
1840 | /* Initialize the structure used to perform file's f_op->poll() calls */ | |
1841 | ep_nested_calls_init(&poll_readywalk_ncalls); | |
1da177e4 LT |
1842 | |
1843 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1844 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
bb57c3ed | 1845 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 LT |
1846 | |
1847 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1848 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
bb57c3ed | 1849 | sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL); |
1da177e4 | 1850 | |
1da177e4 | 1851 | return 0; |
1da177e4 | 1852 | } |
cea69241 | 1853 | fs_initcall(eventpoll_init); |