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> |
4d7e30d9 | 36 | #include <linux/device.h> |
1da177e4 | 37 | #include <asm/uaccess.h> |
1da177e4 LT |
38 | #include <asm/io.h> |
39 | #include <asm/mman.h> | |
60063497 | 40 | #include <linux/atomic.h> |
138d22b5 CG |
41 | #include <linux/proc_fs.h> |
42 | #include <linux/seq_file.h> | |
1da177e4 | 43 | |
1da177e4 LT |
44 | /* |
45 | * LOCKING: | |
46 | * There are three level of locking required by epoll : | |
47 | * | |
144efe3e | 48 | * 1) epmutex (mutex) |
c7ea7630 DL |
49 | * 2) ep->mtx (mutex) |
50 | * 3) ep->lock (spinlock) | |
1da177e4 LT |
51 | * |
52 | * The acquire order is the one listed above, from 1 to 3. | |
53 | * We need a spinlock (ep->lock) because we manipulate objects | |
54 | * from inside the poll callback, that might be triggered from | |
55 | * a wake_up() that in turn might be called from IRQ context. | |
56 | * So we can't sleep inside the poll callback and hence we need | |
57 | * a spinlock. During the event transfer loop (from kernel to | |
58 | * user space) we could end up sleeping due a copy_to_user(), so | |
59 | * we need a lock that will allow us to sleep. This lock is a | |
d47de16c DL |
60 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
61 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). | |
62 | * Then we also need a global mutex to serialize eventpoll_release_file() | |
63 | * and ep_free(). | |
64 | * This mutex is acquired by ep_free() during the epoll file | |
1da177e4 LT |
65 | * cleanup path and it is also acquired by eventpoll_release_file() |
66 | * if a file has been pushed inside an epoll set and it is then | |
bf6a41db | 67 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
22bacca4 DL |
68 | * It is also acquired when inserting an epoll fd onto another epoll |
69 | * fd. We do this so that we walk the epoll tree and ensure that this | |
70 | * insertion does not create a cycle of epoll file descriptors, which | |
71 | * could lead to deadlock. We need a global mutex to prevent two | |
72 | * simultaneous inserts (A into B and B into A) from racing and | |
73 | * constructing a cycle without either insert observing that it is | |
74 | * going to. | |
d8805e63 NE |
75 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
76 | * case when one epoll fd is added to another. In this case, we | |
77 | * always acquire the locks in the order of nesting (i.e. after | |
78 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired | |
79 | * before e2->mtx). Since we disallow cycles of epoll file | |
80 | * descriptors, this ensures that the mutexes are well-ordered. In | |
81 | * order to communicate this nesting to lockdep, when walking a tree | |
82 | * of epoll file descriptors, we use the current recursion depth as | |
83 | * the lockdep subkey. | |
d47de16c DL |
84 | * It is possible to drop the "ep->mtx" and to use the global |
85 | * mutex "epmutex" (together with "ep->lock") to have it working, | |
86 | * but having "ep->mtx" will make the interface more scalable. | |
144efe3e | 87 | * Events that require holding "epmutex" are very rare, while for |
d47de16c DL |
88 | * normal operations the epoll private "ep->mtx" will guarantee |
89 | * a better scalability. | |
1da177e4 LT |
90 | */ |
91 | ||
1da177e4 | 92 | /* Epoll private bits inside the event mask */ |
4d7e30d9 | 93 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET) |
1da177e4 | 94 | |
5071f97e DL |
95 | /* Maximum number of nesting allowed inside epoll sets */ |
96 | #define EP_MAX_NESTS 4 | |
1da177e4 | 97 | |
b611967d DL |
98 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
99 | ||
d47de16c DL |
100 | #define EP_UNACTIVE_PTR ((void *) -1L) |
101 | ||
7ef9964e DL |
102 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
103 | ||
1da177e4 LT |
104 | struct epoll_filefd { |
105 | struct file *file; | |
106 | int fd; | |
107 | }; | |
108 | ||
109 | /* | |
5071f97e DL |
110 | * Structure used to track possible nested calls, for too deep recursions |
111 | * and loop cycles. | |
1da177e4 | 112 | */ |
5071f97e | 113 | struct nested_call_node { |
1da177e4 | 114 | struct list_head llink; |
5071f97e | 115 | void *cookie; |
3fe4a975 | 116 | void *ctx; |
1da177e4 LT |
117 | }; |
118 | ||
119 | /* | |
5071f97e DL |
120 | * This structure is used as collector for nested calls, to check for |
121 | * maximum recursion dept and loop cycles. | |
1da177e4 | 122 | */ |
5071f97e DL |
123 | struct nested_calls { |
124 | struct list_head tasks_call_list; | |
1da177e4 LT |
125 | spinlock_t lock; |
126 | }; | |
127 | ||
d47de16c DL |
128 | /* |
129 | * Each file descriptor added to the eventpoll interface will | |
130 | * have an entry of this type linked to the "rbr" RB tree. | |
131 | */ | |
132 | struct epitem { | |
67647d0f | 133 | /* RB tree node used to link this structure to the eventpoll RB tree */ |
d47de16c DL |
134 | struct rb_node rbn; |
135 | ||
136 | /* List header used to link this structure to the eventpoll ready list */ | |
137 | struct list_head rdllink; | |
138 | ||
c7ea7630 DL |
139 | /* |
140 | * Works together "struct eventpoll"->ovflist in keeping the | |
141 | * single linked chain of items. | |
142 | */ | |
143 | struct epitem *next; | |
144 | ||
d47de16c DL |
145 | /* The file descriptor information this item refers to */ |
146 | struct epoll_filefd ffd; | |
147 | ||
148 | /* Number of active wait queue attached to poll operations */ | |
149 | int nwait; | |
150 | ||
151 | /* List containing poll wait queues */ | |
152 | struct list_head pwqlist; | |
153 | ||
154 | /* The "container" of this item */ | |
155 | struct eventpoll *ep; | |
156 | ||
d47de16c DL |
157 | /* List header used to link this item to the "struct file" items list */ |
158 | struct list_head fllink; | |
159 | ||
4d7e30d9 AH |
160 | /* wakeup_source used when EPOLLWAKEUP is set */ |
161 | struct wakeup_source *ws; | |
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 | |
bf6a41db | 169 | * structure and represents the main data structure for the eventpoll |
1da177e4 LT |
170 | * interface. |
171 | */ | |
172 | struct eventpoll { | |
bf6a41db | 173 | /* Protect the access to this structure */ |
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 | |
25985edc | 198 | * happened while transferring ready events to userspace w/out |
d47de16c DL |
199 | * holding ->lock. |
200 | */ | |
201 | struct epitem *ovflist; | |
7ef9964e | 202 | |
4d7e30d9 AH |
203 | /* wakeup_source used when ep_scan_ready_list is running */ |
204 | struct wakeup_source *ws; | |
205 | ||
7ef9964e DL |
206 | /* The user that created the eventpoll descriptor */ |
207 | struct user_struct *user; | |
28d82dc1 JB |
208 | |
209 | struct file *file; | |
210 | ||
211 | /* used to optimize loop detection check */ | |
212 | int visited; | |
213 | struct list_head visited_list_link; | |
1da177e4 LT |
214 | }; |
215 | ||
216 | /* Wait structure used by the poll hooks */ | |
217 | struct eppoll_entry { | |
218 | /* List header used to link this structure to the "struct epitem" */ | |
219 | struct list_head llink; | |
220 | ||
221 | /* The "base" pointer is set to the container "struct epitem" */ | |
4f0989db | 222 | struct epitem *base; |
1da177e4 LT |
223 | |
224 | /* | |
225 | * Wait queue item that will be linked to the target file wait | |
226 | * queue head. | |
227 | */ | |
228 | wait_queue_t wait; | |
229 | ||
230 | /* The wait queue head that linked the "wait" wait queue item */ | |
231 | wait_queue_head_t *whead; | |
232 | }; | |
233 | ||
1da177e4 LT |
234 | /* Wrapper struct used by poll queueing */ |
235 | struct ep_pqueue { | |
236 | poll_table pt; | |
237 | struct epitem *epi; | |
238 | }; | |
239 | ||
5071f97e DL |
240 | /* Used by the ep_send_events() function as callback private data */ |
241 | struct ep_send_events_data { | |
242 | int maxevents; | |
243 | struct epoll_event __user *events; | |
244 | }; | |
245 | ||
7ef9964e DL |
246 | /* |
247 | * Configuration options available inside /proc/sys/fs/epoll/ | |
248 | */ | |
7ef9964e | 249 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 250 | static long max_user_watches __read_mostly; |
7ef9964e | 251 | |
1da177e4 | 252 | /* |
d47de16c | 253 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 254 | */ |
7ef9964e | 255 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 256 | |
22bacca4 DL |
257 | /* Used to check for epoll file descriptor inclusion loops */ |
258 | static struct nested_calls poll_loop_ncalls; | |
259 | ||
5071f97e DL |
260 | /* Used for safe wake up implementation */ |
261 | static struct nested_calls poll_safewake_ncalls; | |
262 | ||
263 | /* Used to call file's f_op->poll() under the nested calls boundaries */ | |
264 | static struct nested_calls poll_readywalk_ncalls; | |
1da177e4 LT |
265 | |
266 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 267 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
268 | |
269 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 270 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 271 | |
28d82dc1 JB |
272 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
273 | static LIST_HEAD(visited_list); | |
274 | ||
275 | /* | |
276 | * List of files with newly added links, where we may need to limit the number | |
277 | * of emanating paths. Protected by the epmutex. | |
278 | */ | |
279 | static LIST_HEAD(tfile_check_list); | |
280 | ||
7ef9964e DL |
281 | #ifdef CONFIG_SYSCTL |
282 | ||
283 | #include <linux/sysctl.h> | |
284 | ||
52bd19f7 RH |
285 | static long zero; |
286 | static long long_max = LONG_MAX; | |
7ef9964e DL |
287 | |
288 | ctl_table epoll_table[] = { | |
7ef9964e DL |
289 | { |
290 | .procname = "max_user_watches", | |
291 | .data = &max_user_watches, | |
52bd19f7 | 292 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 293 | .mode = 0644, |
52bd19f7 | 294 | .proc_handler = proc_doulongvec_minmax, |
7ef9964e | 295 | .extra1 = &zero, |
52bd19f7 | 296 | .extra2 = &long_max, |
7ef9964e | 297 | }, |
ab09203e | 298 | { } |
7ef9964e DL |
299 | }; |
300 | #endif /* CONFIG_SYSCTL */ | |
301 | ||
28d82dc1 JB |
302 | static const struct file_operations eventpoll_fops; |
303 | ||
304 | static inline int is_file_epoll(struct file *f) | |
305 | { | |
306 | return f->f_op == &eventpoll_fops; | |
307 | } | |
b030a4dd | 308 | |
67647d0f | 309 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
310 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
311 | struct file *file, int fd) | |
312 | { | |
313 | ffd->file = file; | |
314 | ffd->fd = fd; | |
315 | } | |
316 | ||
67647d0f | 317 | /* Compare RB tree keys */ |
b030a4dd PE |
318 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
319 | struct epoll_filefd *p2) | |
320 | { | |
321 | return (p1->file > p2->file ? +1: | |
322 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
323 | } | |
324 | ||
b030a4dd PE |
325 | /* Tells us if the item is currently linked */ |
326 | static inline int ep_is_linked(struct list_head *p) | |
327 | { | |
328 | return !list_empty(p); | |
329 | } | |
330 | ||
971316f0 ON |
331 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p) |
332 | { | |
333 | return container_of(p, struct eppoll_entry, wait); | |
334 | } | |
335 | ||
b030a4dd | 336 | /* Get the "struct epitem" from a wait queue pointer */ |
cdac75e6 | 337 | static inline struct epitem *ep_item_from_wait(wait_queue_t *p) |
b030a4dd PE |
338 | { |
339 | return container_of(p, struct eppoll_entry, wait)->base; | |
340 | } | |
341 | ||
342 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 343 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
344 | { |
345 | return container_of(p, struct ep_pqueue, pt)->epi; | |
346 | } | |
347 | ||
348 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 349 | static inline int ep_op_has_event(int op) |
b030a4dd | 350 | { |
a80a6b85 | 351 | return op != EPOLL_CTL_DEL; |
b030a4dd PE |
352 | } |
353 | ||
1da177e4 | 354 | /* Initialize the poll safe wake up structure */ |
5071f97e | 355 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 356 | { |
5071f97e DL |
357 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
358 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
359 | } |
360 | ||
3fb0e584 DL |
361 | /** |
362 | * ep_events_available - Checks if ready events might be available. | |
363 | * | |
364 | * @ep: Pointer to the eventpoll context. | |
365 | * | |
366 | * Returns: Returns a value different than zero if ready events are available, | |
367 | * or zero otherwise. | |
368 | */ | |
369 | static inline int ep_events_available(struct eventpoll *ep) | |
370 | { | |
371 | return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; | |
372 | } | |
373 | ||
5071f97e DL |
374 | /** |
375 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
376 | * that the recursion limit is not exceeded, and that | |
377 | * the same nested call (by the meaning of same cookie) is | |
378 | * no re-entered. | |
379 | * | |
380 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
381 | * @max_nests: Maximum number of allowed nesting calls. | |
382 | * @nproc: Nested call core function pointer. | |
383 | * @priv: Opaque data to be passed to the @nproc callback. | |
384 | * @cookie: Cookie to be used to identify this nested call. | |
3fe4a975 | 385 | * @ctx: This instance context. |
5071f97e DL |
386 | * |
387 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
388 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 389 | */ |
5071f97e DL |
390 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
391 | int (*nproc)(void *, void *, int), void *priv, | |
3fe4a975 | 392 | void *cookie, void *ctx) |
1da177e4 | 393 | { |
5071f97e | 394 | int error, call_nests = 0; |
1da177e4 | 395 | unsigned long flags; |
5071f97e DL |
396 | struct list_head *lsthead = &ncalls->tasks_call_list; |
397 | struct nested_call_node *tncur; | |
398 | struct nested_call_node tnode; | |
1da177e4 | 399 | |
5071f97e | 400 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 401 | |
5071f97e DL |
402 | /* |
403 | * Try to see if the current task is already inside this wakeup call. | |
404 | * We use a list here, since the population inside this set is always | |
405 | * very much limited. | |
406 | */ | |
b70c3940 | 407 | list_for_each_entry(tncur, lsthead, llink) { |
3fe4a975 | 408 | if (tncur->ctx == ctx && |
5071f97e | 409 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
1da177e4 LT |
410 | /* |
411 | * Ops ... loop detected or maximum nest level reached. | |
412 | * We abort this wake by breaking the cycle itself. | |
413 | */ | |
abff55ce TB |
414 | error = -1; |
415 | goto out_unlock; | |
1da177e4 LT |
416 | } |
417 | } | |
418 | ||
5071f97e | 419 | /* Add the current task and cookie to the list */ |
3fe4a975 | 420 | tnode.ctx = ctx; |
5071f97e | 421 | tnode.cookie = cookie; |
1da177e4 LT |
422 | list_add(&tnode.llink, lsthead); |
423 | ||
5071f97e | 424 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 425 | |
5071f97e DL |
426 | /* Call the nested function */ |
427 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
428 | |
429 | /* Remove the current task from the list */ | |
5071f97e | 430 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 431 | list_del(&tnode.llink); |
3fe4a975 | 432 | out_unlock: |
5071f97e DL |
433 | spin_unlock_irqrestore(&ncalls->lock, flags); |
434 | ||
435 | return error; | |
436 | } | |
437 | ||
02edc6fc SR |
438 | /* |
439 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
440 | * the use of wait queues used by epoll is done in a very controlled | |
441 | * manner. Wake ups can nest inside each other, but are never done | |
442 | * with the same locking. For example: | |
443 | * | |
444 | * dfd = socket(...); | |
445 | * efd1 = epoll_create(); | |
446 | * efd2 = epoll_create(); | |
447 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
448 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
449 | * | |
450 | * When a packet arrives to the device underneath "dfd", the net code will | |
451 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
452 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
453 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
454 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
455 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
456 | * that ends up in another wake_up(), after having checked about the | |
457 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
458 | * avoid stack blasting. | |
459 | * | |
460 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
461 | * this special case of epoll. | |
462 | */ | |
2dfa4eea DL |
463 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
464 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
465 | unsigned long events, int subclass) | |
466 | { | |
467 | unsigned long flags; | |
468 | ||
469 | spin_lock_irqsave_nested(&wqueue->lock, flags, subclass); | |
470 | wake_up_locked_poll(wqueue, events); | |
471 | spin_unlock_irqrestore(&wqueue->lock, flags); | |
472 | } | |
473 | #else | |
474 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
475 | unsigned long events, int subclass) | |
476 | { | |
477 | wake_up_poll(wqueue, events); | |
478 | } | |
479 | #endif | |
480 | ||
5071f97e DL |
481 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
482 | { | |
2dfa4eea DL |
483 | ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN, |
484 | 1 + call_nests); | |
5071f97e DL |
485 | return 0; |
486 | } | |
487 | ||
488 | /* | |
489 | * Perform a safe wake up of the poll wait list. The problem is that | |
490 | * with the new callback'd wake up system, it is possible that the | |
491 | * poll callback is reentered from inside the call to wake_up() done | |
492 | * on the poll wait queue head. The rule is that we cannot reenter the | |
493 | * wake up code from the same task more than EP_MAX_NESTS times, | |
494 | * and we cannot reenter the same wait queue head at all. This will | |
495 | * enable to have a hierarchy of epoll file descriptor of no more than | |
496 | * EP_MAX_NESTS deep. | |
497 | */ | |
498 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
499 | { | |
3fe4a975 DL |
500 | int this_cpu = get_cpu(); |
501 | ||
5071f97e | 502 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
3fe4a975 DL |
503 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
504 | ||
505 | put_cpu(); | |
1da177e4 LT |
506 | } |
507 | ||
971316f0 ON |
508 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
509 | { | |
510 | wait_queue_head_t *whead; | |
511 | ||
512 | rcu_read_lock(); | |
513 | /* If it is cleared by POLLFREE, it should be rcu-safe */ | |
514 | whead = rcu_dereference(pwq->whead); | |
515 | if (whead) | |
516 | remove_wait_queue(whead, &pwq->wait); | |
517 | rcu_read_unlock(); | |
518 | } | |
519 | ||
1da177e4 | 520 | /* |
d1bc90dd TB |
521 | * This function unregisters poll callbacks from the associated file |
522 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
523 | * ep_free). | |
1da177e4 | 524 | */ |
7699acd1 | 525 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 526 | { |
7699acd1 DL |
527 | struct list_head *lsthead = &epi->pwqlist; |
528 | struct eppoll_entry *pwq; | |
1da177e4 | 529 | |
d1bc90dd TB |
530 | while (!list_empty(lsthead)) { |
531 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 532 | |
d1bc90dd | 533 | list_del(&pwq->llink); |
971316f0 | 534 | ep_remove_wait_queue(pwq); |
d1bc90dd | 535 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 536 | } |
1da177e4 LT |
537 | } |
538 | ||
5071f97e DL |
539 | /** |
540 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
541 | * the scan code, to call f_op->poll(). Also allows for | |
542 | * O(NumReady) performance. | |
543 | * | |
544 | * @ep: Pointer to the epoll private data structure. | |
545 | * @sproc: Pointer to the scan callback. | |
546 | * @priv: Private opaque data passed to the @sproc callback. | |
d8805e63 | 547 | * @depth: The current depth of recursive f_op->poll calls. |
5071f97e DL |
548 | * |
549 | * Returns: The same integer error code returned by the @sproc callback. | |
550 | */ | |
551 | static int ep_scan_ready_list(struct eventpoll *ep, | |
552 | int (*sproc)(struct eventpoll *, | |
553 | struct list_head *, void *), | |
d8805e63 NE |
554 | void *priv, |
555 | int depth) | |
5071f97e DL |
556 | { |
557 | int error, pwake = 0; | |
558 | unsigned long flags; | |
559 | struct epitem *epi, *nepi; | |
296e236e | 560 | LIST_HEAD(txlist); |
5071f97e DL |
561 | |
562 | /* | |
563 | * We need to lock this because we could be hit by | |
e057e15f | 564 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e | 565 | */ |
d8805e63 | 566 | mutex_lock_nested(&ep->mtx, depth); |
5071f97e DL |
567 | |
568 | /* | |
569 | * Steal the ready list, and re-init the original one to the | |
570 | * empty list. Also, set ep->ovflist to NULL so that events | |
571 | * happening while looping w/out locks, are not lost. We cannot | |
572 | * have the poll callback to queue directly on ep->rdllist, | |
573 | * because we want the "sproc" callback to be able to do it | |
574 | * in a lockless way. | |
575 | */ | |
576 | spin_lock_irqsave(&ep->lock, flags); | |
296e236e | 577 | list_splice_init(&ep->rdllist, &txlist); |
5071f97e DL |
578 | ep->ovflist = NULL; |
579 | spin_unlock_irqrestore(&ep->lock, flags); | |
580 | ||
581 | /* | |
582 | * Now call the callback function. | |
583 | */ | |
584 | error = (*sproc)(ep, &txlist, priv); | |
585 | ||
586 | spin_lock_irqsave(&ep->lock, flags); | |
587 | /* | |
588 | * During the time we spent inside the "sproc" callback, some | |
589 | * other events might have been queued by the poll callback. | |
590 | * We re-insert them inside the main ready-list here. | |
591 | */ | |
592 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
593 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { | |
594 | /* | |
595 | * We need to check if the item is already in the list. | |
596 | * During the "sproc" callback execution time, items are | |
597 | * queued into ->ovflist but the "txlist" might already | |
598 | * contain them, and the list_splice() below takes care of them. | |
599 | */ | |
4d7e30d9 | 600 | if (!ep_is_linked(&epi->rdllink)) { |
5071f97e | 601 | list_add_tail(&epi->rdllink, &ep->rdllist); |
4d7e30d9 AH |
602 | __pm_stay_awake(epi->ws); |
603 | } | |
5071f97e DL |
604 | } |
605 | /* | |
606 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
607 | * releasing the lock, events will be queued in the normal way inside | |
608 | * ep->rdllist. | |
609 | */ | |
610 | ep->ovflist = EP_UNACTIVE_PTR; | |
611 | ||
612 | /* | |
613 | * Quickly re-inject items left on "txlist". | |
614 | */ | |
615 | list_splice(&txlist, &ep->rdllist); | |
4d7e30d9 | 616 | __pm_relax(ep->ws); |
5071f97e DL |
617 | |
618 | if (!list_empty(&ep->rdllist)) { | |
619 | /* | |
296e236e DL |
620 | * Wake up (if active) both the eventpoll wait list and |
621 | * the ->poll() wait list (delayed after we release the lock). | |
5071f97e DL |
622 | */ |
623 | if (waitqueue_active(&ep->wq)) | |
624 | wake_up_locked(&ep->wq); | |
625 | if (waitqueue_active(&ep->poll_wait)) | |
626 | pwake++; | |
627 | } | |
628 | spin_unlock_irqrestore(&ep->lock, flags); | |
629 | ||
630 | mutex_unlock(&ep->mtx); | |
631 | ||
632 | /* We have to call this outside the lock */ | |
633 | if (pwake) | |
634 | ep_poll_safewake(&ep->poll_wait); | |
635 | ||
636 | return error; | |
637 | } | |
638 | ||
7699acd1 DL |
639 | /* |
640 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 641 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
642 | */ |
643 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
644 | { | |
7699acd1 DL |
645 | unsigned long flags; |
646 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
647 | |
648 | /* | |
7699acd1 DL |
649 | * Removes poll wait queue hooks. We _have_ to do this without holding |
650 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
651 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
652 | * queue head lock when unregistering the wait queue. The wakeup callback | |
653 | * will run by holding the wait queue head lock and will call our callback | |
654 | * that will try to get "ep->lock". | |
1da177e4 | 655 | */ |
7699acd1 | 656 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 657 | |
7699acd1 | 658 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 659 | spin_lock(&file->f_lock); |
7699acd1 DL |
660 | if (ep_is_linked(&epi->fllink)) |
661 | list_del_init(&epi->fllink); | |
68499914 | 662 | spin_unlock(&file->f_lock); |
1da177e4 | 663 | |
cdac75e6 | 664 | rb_erase(&epi->rbn, &ep->rbr); |
1da177e4 | 665 | |
c7ea7630 DL |
666 | spin_lock_irqsave(&ep->lock, flags); |
667 | if (ep_is_linked(&epi->rdllink)) | |
668 | list_del_init(&epi->rdllink); | |
669 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 670 | |
4d7e30d9 AH |
671 | wakeup_source_unregister(epi->ws); |
672 | ||
7699acd1 | 673 | /* At this point it is safe to free the eventpoll item */ |
c7ea7630 | 674 | kmem_cache_free(epi_cache, epi); |
1da177e4 | 675 | |
52bd19f7 | 676 | atomic_long_dec(&ep->user->epoll_watches); |
7ef9964e | 677 | |
c7ea7630 | 678 | return 0; |
1da177e4 LT |
679 | } |
680 | ||
7699acd1 | 681 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 682 | { |
7699acd1 DL |
683 | struct rb_node *rbp; |
684 | struct epitem *epi; | |
1da177e4 | 685 | |
7699acd1 DL |
686 | /* We need to release all tasks waiting for these file */ |
687 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 688 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 689 | |
7699acd1 DL |
690 | /* |
691 | * We need to lock this because we could be hit by | |
692 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 693 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
694 | * is on the way to be removed and no one has references to it |
695 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 696 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
697 | */ |
698 | mutex_lock(&epmutex); | |
1da177e4 LT |
699 | |
700 | /* | |
7699acd1 | 701 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 702 | */ |
7699acd1 DL |
703 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
704 | epi = rb_entry(rbp, struct epitem, rbn); | |
705 | ||
706 | ep_unregister_pollwait(ep, epi); | |
707 | } | |
1da177e4 LT |
708 | |
709 | /* | |
7699acd1 DL |
710 | * Walks through the whole tree by freeing each "struct epitem". At this |
711 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 712 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 713 | * us during this operation. So we can avoid the lock on "ep->lock". |
1da177e4 | 714 | */ |
c80544dc | 715 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
7699acd1 DL |
716 | epi = rb_entry(rbp, struct epitem, rbn); |
717 | ep_remove(ep, epi); | |
718 | } | |
1da177e4 | 719 | |
7699acd1 | 720 | mutex_unlock(&epmutex); |
d47de16c | 721 | mutex_destroy(&ep->mtx); |
7ef9964e | 722 | free_uid(ep->user); |
4d7e30d9 | 723 | wakeup_source_unregister(ep->ws); |
f0ee9aab | 724 | kfree(ep); |
7699acd1 | 725 | } |
1da177e4 | 726 | |
7699acd1 DL |
727 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
728 | { | |
729 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 730 | |
f0ee9aab | 731 | if (ep) |
7699acd1 | 732 | ep_free(ep); |
7699acd1 | 733 | |
7699acd1 | 734 | return 0; |
1da177e4 LT |
735 | } |
736 | ||
296e236e DL |
737 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
738 | void *priv) | |
5071f97e DL |
739 | { |
740 | struct epitem *epi, *tmp; | |
626cf236 | 741 | poll_table pt; |
5071f97e | 742 | |
626cf236 | 743 | init_poll_funcptr(&pt, NULL); |
5071f97e | 744 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
626cf236 HV |
745 | pt._key = epi->event.events; |
746 | if (epi->ffd.file->f_op->poll(epi->ffd.file, &pt) & | |
5071f97e DL |
747 | epi->event.events) |
748 | return POLLIN | POLLRDNORM; | |
296e236e | 749 | else { |
5071f97e DL |
750 | /* |
751 | * Item has been dropped into the ready list by the poll | |
752 | * callback, but it's not actually ready, as far as | |
753 | * caller requested events goes. We can remove it here. | |
754 | */ | |
4d7e30d9 | 755 | __pm_relax(epi->ws); |
5071f97e | 756 | list_del_init(&epi->rdllink); |
296e236e | 757 | } |
5071f97e DL |
758 | } |
759 | ||
760 | return 0; | |
761 | } | |
762 | ||
763 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) | |
764 | { | |
d8805e63 | 765 | return ep_scan_ready_list(priv, ep_read_events_proc, NULL, call_nests + 1); |
5071f97e DL |
766 | } |
767 | ||
7699acd1 DL |
768 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
769 | { | |
5071f97e | 770 | int pollflags; |
7699acd1 | 771 | struct eventpoll *ep = file->private_data; |
1da177e4 | 772 | |
7699acd1 DL |
773 | /* Insert inside our poll wait queue */ |
774 | poll_wait(file, &ep->poll_wait, wait); | |
775 | ||
5071f97e DL |
776 | /* |
777 | * Proceed to find out if wanted events are really available inside | |
778 | * the ready list. This need to be done under ep_call_nested() | |
779 | * supervision, since the call to f_op->poll() done on listed files | |
780 | * could re-enter here. | |
781 | */ | |
782 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, | |
3fe4a975 | 783 | ep_poll_readyevents_proc, ep, ep, current); |
7699acd1 | 784 | |
296e236e | 785 | return pollflags != -1 ? pollflags : 0; |
7699acd1 DL |
786 | } |
787 | ||
138d22b5 CG |
788 | #ifdef CONFIG_PROC_FS |
789 | static int ep_show_fdinfo(struct seq_file *m, struct file *f) | |
790 | { | |
791 | struct eventpoll *ep = f->private_data; | |
792 | struct rb_node *rbp; | |
793 | int ret = 0; | |
794 | ||
795 | mutex_lock(&ep->mtx); | |
796 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { | |
797 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); | |
798 | ||
799 | ret = seq_printf(m, "tfd: %8d events: %8x data: %16llx\n", | |
800 | epi->ffd.fd, epi->event.events, | |
801 | (long long)epi->event.data); | |
802 | if (ret) | |
803 | break; | |
804 | } | |
805 | mutex_unlock(&ep->mtx); | |
806 | ||
807 | return ret; | |
808 | } | |
809 | #endif | |
810 | ||
7699acd1 DL |
811 | /* File callbacks that implement the eventpoll file behaviour */ |
812 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
813 | #ifdef CONFIG_PROC_FS |
814 | .show_fdinfo = ep_show_fdinfo, | |
815 | #endif | |
7699acd1 | 816 | .release = ep_eventpoll_release, |
6038f373 AB |
817 | .poll = ep_eventpoll_poll, |
818 | .llseek = noop_llseek, | |
7699acd1 DL |
819 | }; |
820 | ||
b611967d | 821 | /* |
7699acd1 DL |
822 | * This is called from eventpoll_release() to unlink files from the eventpoll |
823 | * interface. We need to have this facility to cleanup correctly files that are | |
824 | * closed without being removed from the eventpoll interface. | |
b611967d | 825 | */ |
7699acd1 | 826 | void eventpoll_release_file(struct file *file) |
b611967d | 827 | { |
7699acd1 DL |
828 | struct list_head *lsthead = &file->f_ep_links; |
829 | struct eventpoll *ep; | |
830 | struct epitem *epi; | |
b611967d DL |
831 | |
832 | /* | |
68499914 | 833 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 834 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 835 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 836 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 837 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 838 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 839 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 840 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 841 | * from anywhere but ep_free(). |
68499914 JC |
842 | * |
843 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 844 | */ |
7699acd1 | 845 | mutex_lock(&epmutex); |
b611967d | 846 | |
7699acd1 DL |
847 | while (!list_empty(lsthead)) { |
848 | epi = list_first_entry(lsthead, struct epitem, fllink); | |
b611967d | 849 | |
7699acd1 DL |
850 | ep = epi->ep; |
851 | list_del_init(&epi->fllink); | |
d8805e63 | 852 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 853 | ep_remove(ep, epi); |
d47de16c | 854 | mutex_unlock(&ep->mtx); |
b611967d DL |
855 | } |
856 | ||
7699acd1 | 857 | mutex_unlock(&epmutex); |
b611967d DL |
858 | } |
859 | ||
53d2be79 | 860 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 861 | { |
7ef9964e DL |
862 | int error; |
863 | struct user_struct *user; | |
864 | struct eventpoll *ep; | |
1da177e4 | 865 | |
7ef9964e | 866 | user = get_current_user(); |
7ef9964e DL |
867 | error = -ENOMEM; |
868 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
869 | if (unlikely(!ep)) | |
870 | goto free_uid; | |
1da177e4 | 871 | |
c7ea7630 | 872 | spin_lock_init(&ep->lock); |
d47de16c | 873 | mutex_init(&ep->mtx); |
1da177e4 LT |
874 | init_waitqueue_head(&ep->wq); |
875 | init_waitqueue_head(&ep->poll_wait); | |
876 | INIT_LIST_HEAD(&ep->rdllist); | |
877 | ep->rbr = RB_ROOT; | |
d47de16c | 878 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 879 | ep->user = user; |
1da177e4 | 880 | |
53d2be79 | 881 | *pep = ep; |
1da177e4 | 882 | |
1da177e4 | 883 | return 0; |
7ef9964e DL |
884 | |
885 | free_uid: | |
886 | free_uid(user); | |
887 | return error; | |
1da177e4 LT |
888 | } |
889 | ||
1da177e4 | 890 | /* |
c7ea7630 DL |
891 | * Search the file inside the eventpoll tree. The RB tree operations |
892 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
893 | * "mtx" held. | |
1da177e4 LT |
894 | */ |
895 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
896 | { | |
897 | int kcmp; | |
1da177e4 LT |
898 | struct rb_node *rbp; |
899 | struct epitem *epi, *epir = NULL; | |
900 | struct epoll_filefd ffd; | |
901 | ||
b030a4dd | 902 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
903 | for (rbp = ep->rbr.rb_node; rbp; ) { |
904 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 905 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
906 | if (kcmp > 0) |
907 | rbp = rbp->rb_right; | |
908 | else if (kcmp < 0) | |
909 | rbp = rbp->rb_left; | |
910 | else { | |
1da177e4 LT |
911 | epir = epi; |
912 | break; | |
913 | } | |
914 | } | |
1da177e4 | 915 | |
1da177e4 LT |
916 | return epir; |
917 | } | |
918 | ||
1da177e4 | 919 | /* |
7699acd1 | 920 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 921 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 922 | * have events to report. |
1da177e4 | 923 | */ |
7699acd1 | 924 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 925 | { |
7699acd1 DL |
926 | int pwake = 0; |
927 | unsigned long flags; | |
928 | struct epitem *epi = ep_item_from_wait(wait); | |
929 | struct eventpoll *ep = epi->ep; | |
1da177e4 | 930 | |
971316f0 ON |
931 | if ((unsigned long)key & POLLFREE) { |
932 | ep_pwq_from_wait(wait)->whead = NULL; | |
933 | /* | |
934 | * whead = NULL above can race with ep_remove_wait_queue() | |
935 | * which can do another remove_wait_queue() after us, so we | |
936 | * can't use __remove_wait_queue(). whead->lock is held by | |
937 | * the caller. | |
938 | */ | |
d80e731e | 939 | list_del_init(&wait->task_list); |
971316f0 | 940 | } |
d80e731e | 941 | |
c7ea7630 | 942 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 943 | |
7699acd1 DL |
944 | /* |
945 | * If the event mask does not contain any poll(2) event, we consider the | |
946 | * descriptor to be disabled. This condition is likely the effect of the | |
947 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
948 | * until the next EPOLL_CTL_MOD will be issued. | |
949 | */ | |
950 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
951 | goto out_unlock; |
952 | ||
2dfa4eea DL |
953 | /* |
954 | * Check the events coming with the callback. At this stage, not | |
955 | * every device reports the events in the "key" parameter of the | |
956 | * callback. We need to be able to handle both cases here, hence the | |
957 | * test for "key" != NULL before the event match test. | |
958 | */ | |
959 | if (key && !((unsigned long) key & epi->event.events)) | |
960 | goto out_unlock; | |
961 | ||
d47de16c | 962 | /* |
bf6a41db | 963 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 964 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 965 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
966 | * chained in ep->ovflist and requeued later on. |
967 | */ | |
968 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
969 | if (epi->next == EP_UNACTIVE_PTR) { | |
970 | epi->next = ep->ovflist; | |
971 | ep->ovflist = epi; | |
4d7e30d9 AH |
972 | if (epi->ws) { |
973 | /* | |
974 | * Activate ep->ws since epi->ws may get | |
975 | * deactivated at any time. | |
976 | */ | |
977 | __pm_stay_awake(ep->ws); | |
978 | } | |
979 | ||
d47de16c DL |
980 | } |
981 | goto out_unlock; | |
982 | } | |
1da177e4 | 983 | |
7699acd1 | 984 | /* If this file is already in the ready list we exit soon */ |
4d7e30d9 | 985 | if (!ep_is_linked(&epi->rdllink)) { |
5071f97e | 986 | list_add_tail(&epi->rdllink, &ep->rdllist); |
4d7e30d9 AH |
987 | __pm_stay_awake(epi->ws); |
988 | } | |
7699acd1 | 989 | |
7699acd1 DL |
990 | /* |
991 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
992 | * wait list. | |
993 | */ | |
994 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 995 | wake_up_locked(&ep->wq); |
7699acd1 DL |
996 | if (waitqueue_active(&ep->poll_wait)) |
997 | pwake++; | |
998 | ||
d47de16c | 999 | out_unlock: |
c7ea7630 | 1000 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1001 | |
7699acd1 DL |
1002 | /* We have to call this outside the lock */ |
1003 | if (pwake) | |
5071f97e | 1004 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 DL |
1005 | |
1006 | return 1; | |
1007 | } | |
1da177e4 LT |
1008 | |
1009 | /* | |
1010 | * This is the callback that is used to add our wait queue to the | |
1011 | * target file wakeup lists. | |
1012 | */ | |
1013 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1014 | poll_table *pt) | |
1015 | { | |
b030a4dd | 1016 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
1017 | struct eppoll_entry *pwq; |
1018 | ||
e94b1766 | 1019 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
1020 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
1021 | pwq->whead = whead; | |
1022 | pwq->base = epi; | |
1023 | add_wait_queue(whead, &pwq->wait); | |
1024 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
1025 | epi->nwait++; | |
296e236e | 1026 | } else { |
1da177e4 LT |
1027 | /* We have to signal that an error occurred */ |
1028 | epi->nwait = -1; | |
296e236e | 1029 | } |
1da177e4 LT |
1030 | } |
1031 | ||
1da177e4 LT |
1032 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1033 | { | |
1034 | int kcmp; | |
1035 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
1036 | struct epitem *epic; | |
1037 | ||
1038 | while (*p) { | |
1039 | parent = *p; | |
1040 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1041 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
1042 | if (kcmp > 0) |
1043 | p = &parent->rb_right; | |
1044 | else | |
1045 | p = &parent->rb_left; | |
1046 | } | |
1047 | rb_link_node(&epi->rbn, parent, p); | |
1048 | rb_insert_color(&epi->rbn, &ep->rbr); | |
1049 | } | |
1050 | ||
a80a6b85 AM |
1051 | |
1052 | ||
28d82dc1 JB |
1053 | #define PATH_ARR_SIZE 5 |
1054 | /* | |
1055 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1056 | * from a single file of interest. For example, we allow 1000 paths of length | |
1057 | * 1, to emanate from each file of interest. This essentially represents the | |
1058 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1059 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1060 | * is connected to n file sources. In this case each file source has 1 path | |
1061 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1062 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1063 | * and delete can't add additional paths. Protected by the epmutex. | |
1064 | */ | |
1065 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1066 | static int path_count[PATH_ARR_SIZE]; | |
1067 | ||
1068 | static int path_count_inc(int nests) | |
1069 | { | |
93dc6107 JB |
1070 | /* Allow an arbitrary number of depth 1 paths */ |
1071 | if (nests == 0) | |
1072 | return 0; | |
1073 | ||
28d82dc1 JB |
1074 | if (++path_count[nests] > path_limits[nests]) |
1075 | return -1; | |
1076 | return 0; | |
1077 | } | |
1078 | ||
1079 | static void path_count_init(void) | |
1080 | { | |
1081 | int i; | |
1082 | ||
1083 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1084 | path_count[i] = 0; | |
1085 | } | |
1086 | ||
1087 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) | |
1088 | { | |
1089 | int error = 0; | |
1090 | struct file *file = priv; | |
1091 | struct file *child_file; | |
1092 | struct epitem *epi; | |
1093 | ||
1094 | list_for_each_entry(epi, &file->f_ep_links, fllink) { | |
1095 | child_file = epi->ep->file; | |
1096 | if (is_file_epoll(child_file)) { | |
1097 | if (list_empty(&child_file->f_ep_links)) { | |
1098 | if (path_count_inc(call_nests)) { | |
1099 | error = -1; | |
1100 | break; | |
1101 | } | |
1102 | } else { | |
1103 | error = ep_call_nested(&poll_loop_ncalls, | |
1104 | EP_MAX_NESTS, | |
1105 | reverse_path_check_proc, | |
1106 | child_file, child_file, | |
1107 | current); | |
1108 | } | |
1109 | if (error != 0) | |
1110 | break; | |
1111 | } else { | |
1112 | printk(KERN_ERR "reverse_path_check_proc: " | |
1113 | "file is not an ep!\n"); | |
1114 | } | |
1115 | } | |
1116 | return error; | |
1117 | } | |
1118 | ||
1119 | /** | |
1120 | * reverse_path_check - The tfile_check_list is list of file *, which have | |
1121 | * links that are proposed to be newly added. We need to | |
1122 | * make sure that those added links don't add too many | |
1123 | * paths such that we will spend all our time waking up | |
1124 | * eventpoll objects. | |
1125 | * | |
1126 | * Returns: Returns zero if the proposed links don't create too many paths, | |
1127 | * -1 otherwise. | |
1128 | */ | |
1129 | static int reverse_path_check(void) | |
1130 | { | |
28d82dc1 JB |
1131 | int error = 0; |
1132 | struct file *current_file; | |
1133 | ||
1134 | /* let's call this for all tfiles */ | |
1135 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { | |
28d82dc1 JB |
1136 | path_count_init(); |
1137 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
1138 | reverse_path_check_proc, current_file, | |
1139 | current_file, current); | |
1140 | if (error) | |
1141 | break; | |
1142 | } | |
1143 | return error; | |
1144 | } | |
1145 | ||
4d7e30d9 AH |
1146 | static int ep_create_wakeup_source(struct epitem *epi) |
1147 | { | |
1148 | const char *name; | |
1149 | ||
1150 | if (!epi->ep->ws) { | |
1151 | epi->ep->ws = wakeup_source_register("eventpoll"); | |
1152 | if (!epi->ep->ws) | |
1153 | return -ENOMEM; | |
1154 | } | |
1155 | ||
1156 | name = epi->ffd.file->f_path.dentry->d_name.name; | |
1157 | epi->ws = wakeup_source_register(name); | |
1158 | if (!epi->ws) | |
1159 | return -ENOMEM; | |
1160 | ||
1161 | return 0; | |
1162 | } | |
1163 | ||
1164 | static void ep_destroy_wakeup_source(struct epitem *epi) | |
1165 | { | |
1166 | wakeup_source_unregister(epi->ws); | |
1167 | epi->ws = NULL; | |
1168 | } | |
1169 | ||
c7ea7630 DL |
1170 | /* |
1171 | * Must be called with "mtx" held. | |
1172 | */ | |
1da177e4 LT |
1173 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
1174 | struct file *tfile, int fd) | |
1175 | { | |
1176 | int error, revents, pwake = 0; | |
1177 | unsigned long flags; | |
52bd19f7 | 1178 | long user_watches; |
1da177e4 LT |
1179 | struct epitem *epi; |
1180 | struct ep_pqueue epq; | |
1181 | ||
52bd19f7 RH |
1182 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
1183 | if (unlikely(user_watches >= max_user_watches)) | |
7ef9964e | 1184 | return -ENOSPC; |
e94b1766 | 1185 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 1186 | return -ENOMEM; |
1da177e4 LT |
1187 | |
1188 | /* Item initialization follow here ... */ | |
1da177e4 LT |
1189 | INIT_LIST_HEAD(&epi->rdllink); |
1190 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
1191 | INIT_LIST_HEAD(&epi->pwqlist); |
1192 | epi->ep = ep; | |
b030a4dd | 1193 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1194 | epi->event = *event; |
1da177e4 | 1195 | epi->nwait = 0; |
d47de16c | 1196 | epi->next = EP_UNACTIVE_PTR; |
4d7e30d9 AH |
1197 | if (epi->event.events & EPOLLWAKEUP) { |
1198 | error = ep_create_wakeup_source(epi); | |
1199 | if (error) | |
1200 | goto error_create_wakeup_source; | |
1201 | } else { | |
1202 | epi->ws = NULL; | |
1203 | } | |
1da177e4 LT |
1204 | |
1205 | /* Initialize the poll table using the queue callback */ | |
1206 | epq.epi = epi; | |
1207 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
626cf236 | 1208 | epq.pt._key = event->events; |
1da177e4 LT |
1209 | |
1210 | /* | |
1211 | * Attach the item to the poll hooks and get current event bits. | |
1212 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1213 | * been increased by the caller of this function. Note that after |
1214 | * this operation completes, the poll callback can start hitting | |
1215 | * the new item. | |
1da177e4 LT |
1216 | */ |
1217 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
1218 | ||
1219 | /* | |
1220 | * We have to check if something went wrong during the poll wait queue | |
1221 | * install process. Namely an allocation for a wait queue failed due | |
1222 | * high memory pressure. | |
1223 | */ | |
7ef9964e | 1224 | error = -ENOMEM; |
1da177e4 | 1225 | if (epi->nwait < 0) |
7699acd1 | 1226 | goto error_unregister; |
1da177e4 LT |
1227 | |
1228 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 1229 | spin_lock(&tfile->f_lock); |
1da177e4 | 1230 | list_add_tail(&epi->fllink, &tfile->f_ep_links); |
68499914 | 1231 | spin_unlock(&tfile->f_lock); |
1da177e4 | 1232 | |
c7ea7630 DL |
1233 | /* |
1234 | * Add the current item to the RB tree. All RB tree operations are | |
1235 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1236 | */ | |
1da177e4 LT |
1237 | ep_rbtree_insert(ep, epi); |
1238 | ||
28d82dc1 JB |
1239 | /* now check if we've created too many backpaths */ |
1240 | error = -EINVAL; | |
1241 | if (reverse_path_check()) | |
1242 | goto error_remove_epi; | |
1243 | ||
c7ea7630 DL |
1244 | /* We have to drop the new item inside our item list to keep track of it */ |
1245 | spin_lock_irqsave(&ep->lock, flags); | |
1246 | ||
1da177e4 | 1247 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 1248 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 | 1249 | list_add_tail(&epi->rdllink, &ep->rdllist); |
4d7e30d9 | 1250 | __pm_stay_awake(epi->ws); |
1da177e4 LT |
1251 | |
1252 | /* Notify waiting tasks that events are available */ | |
1253 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1254 | wake_up_locked(&ep->wq); |
1da177e4 LT |
1255 | if (waitqueue_active(&ep->poll_wait)) |
1256 | pwake++; | |
1257 | } | |
1258 | ||
c7ea7630 | 1259 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1260 | |
52bd19f7 | 1261 | atomic_long_inc(&ep->user->epoll_watches); |
7ef9964e | 1262 | |
1da177e4 LT |
1263 | /* We have to call this outside the lock */ |
1264 | if (pwake) | |
5071f97e | 1265 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1266 | |
1da177e4 LT |
1267 | return 0; |
1268 | ||
28d82dc1 JB |
1269 | error_remove_epi: |
1270 | spin_lock(&tfile->f_lock); | |
1271 | if (ep_is_linked(&epi->fllink)) | |
1272 | list_del_init(&epi->fllink); | |
1273 | spin_unlock(&tfile->f_lock); | |
1274 | ||
1275 | rb_erase(&epi->rbn, &ep->rbr); | |
1276 | ||
7699acd1 | 1277 | error_unregister: |
1da177e4 LT |
1278 | ep_unregister_pollwait(ep, epi); |
1279 | ||
1280 | /* | |
1281 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
1282 | * allocated wait queue. Note that we don't care about the ep->ovflist |
1283 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
1284 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 1285 | */ |
c7ea7630 | 1286 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 1287 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 1288 | list_del_init(&epi->rdllink); |
c7ea7630 | 1289 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1290 | |
4d7e30d9 AH |
1291 | wakeup_source_unregister(epi->ws); |
1292 | ||
1293 | error_create_wakeup_source: | |
b030a4dd | 1294 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1295 | |
1da177e4 LT |
1296 | return error; |
1297 | } | |
1298 | ||
1da177e4 LT |
1299 | /* |
1300 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1301 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
1302 | */ |
1303 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1304 | { | |
1305 | int pwake = 0; | |
1306 | unsigned int revents; | |
626cf236 HV |
1307 | poll_table pt; |
1308 | ||
1309 | init_poll_funcptr(&pt, NULL); | |
1da177e4 LT |
1310 | |
1311 | /* | |
e057e15f TB |
1312 | * Set the new event interest mask before calling f_op->poll(); |
1313 | * otherwise we might miss an event that happens between the | |
1314 | * f_op->poll() call and the new event set registering. | |
1da177e4 LT |
1315 | */ |
1316 | epi->event.events = event->events; | |
626cf236 | 1317 | pt._key = event->events; |
e057e15f | 1318 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 AH |
1319 | if (epi->event.events & EPOLLWAKEUP) { |
1320 | if (!epi->ws) | |
1321 | ep_create_wakeup_source(epi); | |
1322 | } else if (epi->ws) { | |
1323 | ep_destroy_wakeup_source(epi); | |
1324 | } | |
1da177e4 LT |
1325 | |
1326 | /* | |
1327 | * Get current event bits. We can safely use the file* here because | |
1328 | * its usage count has been increased by the caller of this function. | |
1329 | */ | |
626cf236 | 1330 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt); |
1da177e4 | 1331 | |
1da177e4 | 1332 | /* |
c7ea7630 | 1333 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1334 | * list, push it inside. |
1da177e4 | 1335 | */ |
c7ea7630 | 1336 | if (revents & event->events) { |
e057e15f | 1337 | spin_lock_irq(&ep->lock); |
c7ea7630 DL |
1338 | if (!ep_is_linked(&epi->rdllink)) { |
1339 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
4d7e30d9 | 1340 | __pm_stay_awake(epi->ws); |
c7ea7630 DL |
1341 | |
1342 | /* Notify waiting tasks that events are available */ | |
1343 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1344 | wake_up_locked(&ep->wq); |
c7ea7630 DL |
1345 | if (waitqueue_active(&ep->poll_wait)) |
1346 | pwake++; | |
7699acd1 | 1347 | } |
e057e15f | 1348 | spin_unlock_irq(&ep->lock); |
7699acd1 | 1349 | } |
1da177e4 | 1350 | |
7699acd1 DL |
1351 | /* We have to call this outside the lock */ |
1352 | if (pwake) | |
5071f97e | 1353 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1354 | |
7699acd1 | 1355 | return 0; |
1da177e4 LT |
1356 | } |
1357 | ||
296e236e DL |
1358 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
1359 | void *priv) | |
1da177e4 | 1360 | { |
5071f97e DL |
1361 | struct ep_send_events_data *esed = priv; |
1362 | int eventcnt; | |
296e236e | 1363 | unsigned int revents; |
5071f97e DL |
1364 | struct epitem *epi; |
1365 | struct epoll_event __user *uevent; | |
626cf236 HV |
1366 | poll_table pt; |
1367 | ||
1368 | init_poll_funcptr(&pt, NULL); | |
1da177e4 | 1369 | |
296e236e | 1370 | /* |
5071f97e DL |
1371 | * We can loop without lock because we are passed a task private list. |
1372 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1373 | * holding "mtx" during this call. | |
296e236e | 1374 | */ |
5071f97e DL |
1375 | for (eventcnt = 0, uevent = esed->events; |
1376 | !list_empty(head) && eventcnt < esed->maxevents;) { | |
1377 | epi = list_first_entry(head, struct epitem, rdllink); | |
d47de16c | 1378 | |
4d7e30d9 AH |
1379 | /* |
1380 | * Activate ep->ws before deactivating epi->ws to prevent | |
1381 | * triggering auto-suspend here (in case we reactive epi->ws | |
1382 | * below). | |
1383 | * | |
1384 | * This could be rearranged to delay the deactivation of epi->ws | |
1385 | * instead, but then epi->ws would temporarily be out of sync | |
1386 | * with ep_is_linked(). | |
1387 | */ | |
1388 | if (epi->ws && epi->ws->active) | |
1389 | __pm_stay_awake(ep->ws); | |
1390 | __pm_relax(epi->ws); | |
d47de16c | 1391 | list_del_init(&epi->rdllink); |
1da177e4 | 1392 | |
626cf236 HV |
1393 | pt._key = epi->event.events; |
1394 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt) & | |
296e236e | 1395 | epi->event.events; |
5071f97e | 1396 | |
296e236e | 1397 | /* |
5071f97e DL |
1398 | * If the event mask intersect the caller-requested one, |
1399 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
1400 | * is holding "mtx", so no operations coming from userspace | |
1401 | * can change the item. | |
296e236e DL |
1402 | */ |
1403 | if (revents) { | |
5071f97e | 1404 | if (__put_user(revents, &uevent->events) || |
d0305882 TB |
1405 | __put_user(epi->event.data, &uevent->data)) { |
1406 | list_add(&epi->rdllink, head); | |
4d7e30d9 | 1407 | __pm_stay_awake(epi->ws); |
296e236e | 1408 | return eventcnt ? eventcnt : -EFAULT; |
d0305882 | 1409 | } |
296e236e | 1410 | eventcnt++; |
5071f97e | 1411 | uevent++; |
296e236e DL |
1412 | if (epi->event.events & EPOLLONESHOT) |
1413 | epi->event.events &= EP_PRIVATE_BITS; | |
1414 | else if (!(epi->event.events & EPOLLET)) { | |
1415 | /* | |
1416 | * If this file has been added with Level | |
1417 | * Trigger mode, we need to insert back inside | |
1418 | * the ready list, so that the next call to | |
1419 | * epoll_wait() will check again the events | |
25985edc | 1420 | * availability. At this point, no one can insert |
296e236e DL |
1421 | * into ep->rdllist besides us. The epoll_ctl() |
1422 | * callers are locked out by | |
1423 | * ep_scan_ready_list() holding "mtx" and the | |
1424 | * poll callback will queue them in ep->ovflist. | |
1425 | */ | |
1426 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
4d7e30d9 | 1427 | __pm_stay_awake(epi->ws); |
296e236e DL |
1428 | } |
1429 | } | |
1430 | } | |
5071f97e DL |
1431 | |
1432 | return eventcnt; | |
1433 | } | |
d47de16c | 1434 | |
296e236e DL |
1435 | static int ep_send_events(struct eventpoll *ep, |
1436 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1437 | { |
1438 | struct ep_send_events_data esed; | |
1da177e4 | 1439 | |
5071f97e DL |
1440 | esed.maxevents = maxevents; |
1441 | esed.events = events; | |
6192bd53 | 1442 | |
d8805e63 | 1443 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0); |
1da177e4 LT |
1444 | } |
1445 | ||
0781b909 ED |
1446 | static inline struct timespec ep_set_mstimeout(long ms) |
1447 | { | |
1448 | struct timespec now, ts = { | |
1449 | .tv_sec = ms / MSEC_PER_SEC, | |
1450 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), | |
1451 | }; | |
1452 | ||
1453 | ktime_get_ts(&now); | |
1454 | return timespec_add_safe(now, ts); | |
1455 | } | |
1456 | ||
f4d93ad7 SB |
1457 | /** |
1458 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied | |
1459 | * event buffer. | |
1460 | * | |
1461 | * @ep: Pointer to the eventpoll context. | |
1462 | * @events: Pointer to the userspace buffer where the ready events should be | |
1463 | * stored. | |
1464 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1465 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
1466 | * milliseconds. If the @timeout is zero, the function will not block, | |
1467 | * while if the @timeout is less than zero, the function will block | |
1468 | * until at least one event has been retrieved (or an error | |
1469 | * occurred). | |
1470 | * | |
1471 | * Returns: Returns the number of ready events which have been fetched, or an | |
1472 | * error code, in case of error. | |
1473 | */ | |
1da177e4 LT |
1474 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1475 | int maxevents, long timeout) | |
1476 | { | |
f4d93ad7 | 1477 | int res = 0, eavail, timed_out = 0; |
1da177e4 | 1478 | unsigned long flags; |
f4d93ad7 | 1479 | long slack = 0; |
1da177e4 | 1480 | wait_queue_t wait; |
95aac7b1 SB |
1481 | ktime_t expires, *to = NULL; |
1482 | ||
1483 | if (timeout > 0) { | |
0781b909 ED |
1484 | struct timespec end_time = ep_set_mstimeout(timeout); |
1485 | ||
95aac7b1 SB |
1486 | slack = select_estimate_accuracy(&end_time); |
1487 | to = &expires; | |
1488 | *to = timespec_to_ktime(end_time); | |
1489 | } else if (timeout == 0) { | |
f4d93ad7 SB |
1490 | /* |
1491 | * Avoid the unnecessary trip to the wait queue loop, if the | |
1492 | * caller specified a non blocking operation. | |
1493 | */ | |
95aac7b1 | 1494 | timed_out = 1; |
f4d93ad7 SB |
1495 | spin_lock_irqsave(&ep->lock, flags); |
1496 | goto check_events; | |
95aac7b1 | 1497 | } |
1da177e4 | 1498 | |
f4d93ad7 | 1499 | fetch_events: |
c7ea7630 | 1500 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1501 | |
3fb0e584 | 1502 | if (!ep_events_available(ep)) { |
1da177e4 LT |
1503 | /* |
1504 | * We don't have any available event to return to the caller. | |
1505 | * We need to sleep here, and we will be wake up by | |
1506 | * ep_poll_callback() when events will become available. | |
1507 | */ | |
1508 | init_waitqueue_entry(&wait, current); | |
a93d2f17 | 1509 | __add_wait_queue_exclusive(&ep->wq, &wait); |
1da177e4 LT |
1510 | |
1511 | for (;;) { | |
1512 | /* | |
1513 | * We don't want to sleep if the ep_poll_callback() sends us | |
1514 | * a wakeup in between. That's why we set the task state | |
1515 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1516 | */ | |
1517 | set_current_state(TASK_INTERRUPTIBLE); | |
3fb0e584 | 1518 | if (ep_events_available(ep) || timed_out) |
1da177e4 LT |
1519 | break; |
1520 | if (signal_pending(current)) { | |
1521 | res = -EINTR; | |
1522 | break; | |
1523 | } | |
1524 | ||
c7ea7630 | 1525 | spin_unlock_irqrestore(&ep->lock, flags); |
95aac7b1 SB |
1526 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) |
1527 | timed_out = 1; | |
1528 | ||
c7ea7630 | 1529 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1530 | } |
3419b23a | 1531 | __remove_wait_queue(&ep->wq, &wait); |
1da177e4 LT |
1532 | |
1533 | set_current_state(TASK_RUNNING); | |
1534 | } | |
f4d93ad7 | 1535 | check_events: |
1da177e4 | 1536 | /* Is it worth to try to dig for events ? */ |
3fb0e584 | 1537 | eavail = ep_events_available(ep); |
1da177e4 | 1538 | |
c7ea7630 | 1539 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1540 | |
1541 | /* | |
1542 | * Try to transfer events to user space. In case we get 0 events and | |
1543 | * there's still timeout left over, we go trying again in search of | |
1544 | * more luck. | |
1545 | */ | |
1546 | if (!res && eavail && | |
95aac7b1 | 1547 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
f4d93ad7 | 1548 | goto fetch_events; |
1da177e4 LT |
1549 | |
1550 | return res; | |
1551 | } | |
1552 | ||
22bacca4 DL |
1553 | /** |
1554 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() | |
1555 | * API, to verify that adding an epoll file inside another | |
1556 | * epoll structure, does not violate the constraints, in | |
1557 | * terms of closed loops, or too deep chains (which can | |
1558 | * result in excessive stack usage). | |
1559 | * | |
1560 | * @priv: Pointer to the epoll file to be currently checked. | |
1561 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll | |
1562 | * data structure pointer. | |
1563 | * @call_nests: Current dept of the @ep_call_nested() call stack. | |
1564 | * | |
1565 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1566 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1567 | */ | |
1568 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) | |
1569 | { | |
1570 | int error = 0; | |
1571 | struct file *file = priv; | |
1572 | struct eventpoll *ep = file->private_data; | |
28d82dc1 | 1573 | struct eventpoll *ep_tovisit; |
22bacca4 DL |
1574 | struct rb_node *rbp; |
1575 | struct epitem *epi; | |
1576 | ||
d8805e63 | 1577 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
28d82dc1 JB |
1578 | ep->visited = 1; |
1579 | list_add(&ep->visited_list_link, &visited_list); | |
22bacca4 DL |
1580 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
1581 | epi = rb_entry(rbp, struct epitem, rbn); | |
1582 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
28d82dc1 JB |
1583 | ep_tovisit = epi->ffd.file->private_data; |
1584 | if (ep_tovisit->visited) | |
1585 | continue; | |
22bacca4 | 1586 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
28d82dc1 JB |
1587 | ep_loop_check_proc, epi->ffd.file, |
1588 | ep_tovisit, current); | |
22bacca4 DL |
1589 | if (error != 0) |
1590 | break; | |
28d82dc1 JB |
1591 | } else { |
1592 | /* | |
1593 | * If we've reached a file that is not associated with | |
1594 | * an ep, then we need to check if the newly added | |
1595 | * links are going to add too many wakeup paths. We do | |
1596 | * this by adding it to the tfile_check_list, if it's | |
1597 | * not already there, and calling reverse_path_check() | |
1598 | * during ep_insert(). | |
1599 | */ | |
1600 | if (list_empty(&epi->ffd.file->f_tfile_llink)) | |
1601 | list_add(&epi->ffd.file->f_tfile_llink, | |
1602 | &tfile_check_list); | |
22bacca4 DL |
1603 | } |
1604 | } | |
1605 | mutex_unlock(&ep->mtx); | |
1606 | ||
1607 | return error; | |
1608 | } | |
1609 | ||
1610 | /** | |
1611 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) | |
1612 | * another epoll file (represented by @ep) does not create | |
1613 | * closed loops or too deep chains. | |
1614 | * | |
1615 | * @ep: Pointer to the epoll private data structure. | |
1616 | * @file: Pointer to the epoll file to be checked. | |
1617 | * | |
1618 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1619 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1620 | */ | |
1621 | static int ep_loop_check(struct eventpoll *ep, struct file *file) | |
1622 | { | |
28d82dc1 JB |
1623 | int ret; |
1624 | struct eventpoll *ep_cur, *ep_next; | |
1625 | ||
1626 | ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
22bacca4 | 1627 | ep_loop_check_proc, file, ep, current); |
28d82dc1 JB |
1628 | /* clear visited list */ |
1629 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, | |
1630 | visited_list_link) { | |
1631 | ep_cur->visited = 0; | |
1632 | list_del(&ep_cur->visited_list_link); | |
1633 | } | |
1634 | return ret; | |
1635 | } | |
1636 | ||
1637 | static void clear_tfile_check_list(void) | |
1638 | { | |
1639 | struct file *file; | |
1640 | ||
1641 | /* first clear the tfile_check_list */ | |
1642 | while (!list_empty(&tfile_check_list)) { | |
1643 | file = list_first_entry(&tfile_check_list, struct file, | |
1644 | f_tfile_llink); | |
1645 | list_del_init(&file->f_tfile_llink); | |
1646 | } | |
1647 | INIT_LIST_HEAD(&tfile_check_list); | |
22bacca4 DL |
1648 | } |
1649 | ||
7699acd1 | 1650 | /* |
523723bb | 1651 | * Open an eventpoll file descriptor. |
7699acd1 | 1652 | */ |
5a8a82b1 | 1653 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
7699acd1 | 1654 | { |
28d82dc1 | 1655 | int error, fd; |
bb57c3ed | 1656 | struct eventpoll *ep = NULL; |
28d82dc1 | 1657 | struct file *file; |
7699acd1 | 1658 | |
e38b36f3 UD |
1659 | /* Check the EPOLL_* constant for consistency. */ |
1660 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1661 | ||
296e236e DL |
1662 | if (flags & ~EPOLL_CLOEXEC) |
1663 | return -EINVAL; | |
7699acd1 | 1664 | /* |
bb57c3ed | 1665 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 1666 | */ |
9fe5ad9c | 1667 | error = ep_alloc(&ep); |
bb57c3ed DL |
1668 | if (error < 0) |
1669 | return error; | |
7699acd1 DL |
1670 | /* |
1671 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 1672 | * a file structure and a free file descriptor. |
7699acd1 | 1673 | */ |
28d82dc1 JB |
1674 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
1675 | if (fd < 0) { | |
1676 | error = fd; | |
1677 | goto out_free_ep; | |
1678 | } | |
1679 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 1680 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
1681 | if (IS_ERR(file)) { |
1682 | error = PTR_ERR(file); | |
1683 | goto out_free_fd; | |
1684 | } | |
28d82dc1 | 1685 | ep->file = file; |
98022748 | 1686 | fd_install(fd, file); |
28d82dc1 JB |
1687 | return fd; |
1688 | ||
1689 | out_free_fd: | |
1690 | put_unused_fd(fd); | |
1691 | out_free_ep: | |
1692 | ep_free(ep); | |
bb57c3ed | 1693 | return error; |
7699acd1 DL |
1694 | } |
1695 | ||
5a8a82b1 | 1696 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 1697 | { |
bfe3891a | 1698 | if (size <= 0) |
9fe5ad9c UD |
1699 | return -EINVAL; |
1700 | ||
1701 | return sys_epoll_create1(0); | |
a0998b50 UD |
1702 | } |
1703 | ||
7699acd1 DL |
1704 | /* |
1705 | * The following function implements the controller interface for | |
1706 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 1707 | * file descriptors inside the interest set. |
7699acd1 | 1708 | */ |
5a8a82b1 HC |
1709 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
1710 | struct epoll_event __user *, event) | |
7699acd1 DL |
1711 | { |
1712 | int error; | |
22bacca4 | 1713 | int did_lock_epmutex = 0; |
7699acd1 DL |
1714 | struct file *file, *tfile; |
1715 | struct eventpoll *ep; | |
1716 | struct epitem *epi; | |
1717 | struct epoll_event epds; | |
1718 | ||
7699acd1 DL |
1719 | error = -EFAULT; |
1720 | if (ep_op_has_event(op) && | |
1721 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
1722 | goto error_return; | |
1723 | ||
1724 | /* Get the "struct file *" for the eventpoll file */ | |
1725 | error = -EBADF; | |
1726 | file = fget(epfd); | |
1727 | if (!file) | |
1728 | goto error_return; | |
1729 | ||
1730 | /* Get the "struct file *" for the target file */ | |
1731 | tfile = fget(fd); | |
1732 | if (!tfile) | |
1733 | goto error_fput; | |
1734 | ||
1735 | /* The target file descriptor must support poll */ | |
1736 | error = -EPERM; | |
1737 | if (!tfile->f_op || !tfile->f_op->poll) | |
1738 | goto error_tgt_fput; | |
1739 | ||
4d7e30d9 | 1740 | /* Check if EPOLLWAKEUP is allowed */ |
d9914cf6 | 1741 | if ((epds.events & EPOLLWAKEUP) && !capable(CAP_BLOCK_SUSPEND)) |
a8159414 | 1742 | epds.events &= ~EPOLLWAKEUP; |
4d7e30d9 | 1743 | |
7699acd1 DL |
1744 | /* |
1745 | * We have to check that the file structure underneath the file descriptor | |
1746 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
1747 | * adding an epoll file descriptor inside itself. | |
1748 | */ | |
1749 | error = -EINVAL; | |
1750 | if (file == tfile || !is_file_epoll(file)) | |
1751 | goto error_tgt_fput; | |
1752 | ||
1753 | /* | |
1754 | * At this point it is safe to assume that the "private_data" contains | |
1755 | * our own data structure. | |
1756 | */ | |
1757 | ep = file->private_data; | |
1758 | ||
22bacca4 DL |
1759 | /* |
1760 | * When we insert an epoll file descriptor, inside another epoll file | |
1761 | * descriptor, there is the change of creating closed loops, which are | |
28d82dc1 JB |
1762 | * better be handled here, than in more critical paths. While we are |
1763 | * checking for loops we also determine the list of files reachable | |
1764 | * and hang them on the tfile_check_list, so we can check that we | |
1765 | * haven't created too many possible wakeup paths. | |
22bacca4 | 1766 | * |
28d82dc1 JB |
1767 | * We need to hold the epmutex across both ep_insert and ep_remove |
1768 | * b/c we want to make sure we are looking at a coherent view of | |
1769 | * epoll network. | |
22bacca4 | 1770 | */ |
28d82dc1 | 1771 | if (op == EPOLL_CTL_ADD || op == EPOLL_CTL_DEL) { |
22bacca4 DL |
1772 | mutex_lock(&epmutex); |
1773 | did_lock_epmutex = 1; | |
22bacca4 | 1774 | } |
28d82dc1 JB |
1775 | if (op == EPOLL_CTL_ADD) { |
1776 | if (is_file_epoll(tfile)) { | |
1777 | error = -ELOOP; | |
13d51807 JB |
1778 | if (ep_loop_check(ep, tfile) != 0) { |
1779 | clear_tfile_check_list(); | |
28d82dc1 | 1780 | goto error_tgt_fput; |
13d51807 | 1781 | } |
28d82dc1 JB |
1782 | } else |
1783 | list_add(&tfile->f_tfile_llink, &tfile_check_list); | |
1784 | } | |
22bacca4 | 1785 | |
d8805e63 | 1786 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 1787 | |
67647d0f DL |
1788 | /* |
1789 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
1790 | * above, we can be sure to be able to use the item looked up by | |
1791 | * ep_find() till we release the mutex. | |
1792 | */ | |
7699acd1 DL |
1793 | epi = ep_find(ep, tfile, fd); |
1794 | ||
1795 | error = -EINVAL; | |
1796 | switch (op) { | |
1797 | case EPOLL_CTL_ADD: | |
1798 | if (!epi) { | |
1799 | epds.events |= POLLERR | POLLHUP; | |
7699acd1 DL |
1800 | error = ep_insert(ep, &epds, tfile, fd); |
1801 | } else | |
1802 | error = -EEXIST; | |
28d82dc1 | 1803 | clear_tfile_check_list(); |
7699acd1 DL |
1804 | break; |
1805 | case EPOLL_CTL_DEL: | |
1806 | if (epi) | |
1807 | error = ep_remove(ep, epi); | |
1808 | else | |
1809 | error = -ENOENT; | |
1810 | break; | |
1811 | case EPOLL_CTL_MOD: | |
1812 | if (epi) { | |
1813 | epds.events |= POLLERR | POLLHUP; | |
1814 | error = ep_modify(ep, epi, &epds); | |
1815 | } else | |
1816 | error = -ENOENT; | |
1817 | break; | |
1818 | } | |
d47de16c | 1819 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
1820 | |
1821 | error_tgt_fput: | |
28d82dc1 | 1822 | if (did_lock_epmutex) |
22bacca4 DL |
1823 | mutex_unlock(&epmutex); |
1824 | ||
7699acd1 DL |
1825 | fput(tfile); |
1826 | error_fput: | |
1827 | fput(file); | |
1828 | error_return: | |
7699acd1 DL |
1829 | |
1830 | return error; | |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1835 | * part of the user space epoll_wait(2). | |
1836 | */ | |
5a8a82b1 HC |
1837 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
1838 | int, maxevents, int, timeout) | |
7699acd1 | 1839 | { |
2903ff01 AV |
1840 | int error; |
1841 | struct fd f; | |
7699acd1 DL |
1842 | struct eventpoll *ep; |
1843 | ||
7699acd1 DL |
1844 | /* The maximum number of event must be greater than zero */ |
1845 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
1846 | return -EINVAL; | |
1847 | ||
1848 | /* Verify that the area passed by the user is writeable */ | |
2903ff01 AV |
1849 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) |
1850 | return -EFAULT; | |
7699acd1 DL |
1851 | |
1852 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
1853 | f = fdget(epfd); |
1854 | if (!f.file) | |
1855 | return -EBADF; | |
7699acd1 DL |
1856 | |
1857 | /* | |
1858 | * We have to check that the file structure underneath the fd | |
1859 | * the user passed to us _is_ an eventpoll file. | |
1860 | */ | |
1861 | error = -EINVAL; | |
2903ff01 | 1862 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
1863 | goto error_fput; |
1864 | ||
1865 | /* | |
1866 | * At this point it is safe to assume that the "private_data" contains | |
1867 | * our own data structure. | |
1868 | */ | |
2903ff01 | 1869 | ep = f.file->private_data; |
7699acd1 DL |
1870 | |
1871 | /* Time to fish for events ... */ | |
1872 | error = ep_poll(ep, events, maxevents, timeout); | |
1873 | ||
1874 | error_fput: | |
2903ff01 | 1875 | fdput(f); |
7699acd1 DL |
1876 | return error; |
1877 | } | |
1878 | ||
7699acd1 DL |
1879 | /* |
1880 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
1881 | * part of the user space epoll_pwait(2). | |
1882 | */ | |
5a8a82b1 HC |
1883 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
1884 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
1885 | size_t, sigsetsize) | |
7699acd1 DL |
1886 | { |
1887 | int error; | |
1888 | sigset_t ksigmask, sigsaved; | |
1889 | ||
1890 | /* | |
1891 | * If the caller wants a certain signal mask to be set during the wait, | |
1892 | * we apply it here. | |
1893 | */ | |
1894 | if (sigmask) { | |
1895 | if (sigsetsize != sizeof(sigset_t)) | |
1896 | return -EINVAL; | |
1897 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1898 | return -EFAULT; | |
1899 | sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP)); | |
1900 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1901 | } | |
1902 | ||
1903 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
1904 | ||
1905 | /* | |
1906 | * If we changed the signal mask, we need to restore the original one. | |
1907 | * In case we've got a signal while waiting, we do not restore the | |
1908 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
1909 | * the way back to userspace, before the signal mask is restored. | |
1910 | */ | |
1911 | if (sigmask) { | |
1912 | if (error == -EINTR) { | |
1913 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 1914 | sizeof(sigsaved)); |
4e4c22c7 | 1915 | set_restore_sigmask(); |
7699acd1 DL |
1916 | } else |
1917 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1918 | } | |
1919 | ||
1920 | return error; | |
1921 | } | |
1922 | ||
1da177e4 LT |
1923 | static int __init eventpoll_init(void) |
1924 | { | |
7ef9964e DL |
1925 | struct sysinfo si; |
1926 | ||
1927 | si_meminfo(&si); | |
9df04e1f DL |
1928 | /* |
1929 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
1930 | */ | |
1931 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 1932 | EP_ITEM_COST; |
52bd19f7 | 1933 | BUG_ON(max_user_watches < 0); |
1da177e4 | 1934 | |
22bacca4 DL |
1935 | /* |
1936 | * Initialize the structure used to perform epoll file descriptor | |
1937 | * inclusion loops checks. | |
1938 | */ | |
1939 | ep_nested_calls_init(&poll_loop_ncalls); | |
1940 | ||
1da177e4 | 1941 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
5071f97e DL |
1942 | ep_nested_calls_init(&poll_safewake_ncalls); |
1943 | ||
1944 | /* Initialize the structure used to perform file's f_op->poll() calls */ | |
1945 | ep_nested_calls_init(&poll_readywalk_ncalls); | |
1da177e4 LT |
1946 | |
1947 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1948 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
bb57c3ed | 1949 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 LT |
1950 | |
1951 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1952 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
bb57c3ed | 1953 | sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL); |
1da177e4 | 1954 | |
1da177e4 | 1955 | return 0; |
1da177e4 | 1956 | } |
cea69241 | 1957 | fs_initcall(eventpoll_init); |