]>
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> | |
174cd4b1 | 16 | #include <linux/sched/signal.h> |
1da177e4 LT |
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> |
7c0f6ba6 | 37 | #include <linux/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> | |
35280bd4 | 43 | #include <linux/compat.h> |
ae10b2b4 | 44 | #include <linux/rculist.h> |
bf3b9f63 | 45 | #include <net/busy_poll.h> |
1da177e4 | 46 | |
1da177e4 LT |
47 | /* |
48 | * LOCKING: | |
49 | * There are three level of locking required by epoll : | |
50 | * | |
144efe3e | 51 | * 1) epmutex (mutex) |
c7ea7630 DL |
52 | * 2) ep->mtx (mutex) |
53 | * 3) ep->lock (spinlock) | |
1da177e4 LT |
54 | * |
55 | * The acquire order is the one listed above, from 1 to 3. | |
56 | * We need a spinlock (ep->lock) because we manipulate objects | |
57 | * from inside the poll callback, that might be triggered from | |
58 | * a wake_up() that in turn might be called from IRQ context. | |
59 | * So we can't sleep inside the poll callback and hence we need | |
60 | * a spinlock. During the event transfer loop (from kernel to | |
61 | * user space) we could end up sleeping due a copy_to_user(), so | |
62 | * we need a lock that will allow us to sleep. This lock is a | |
d47de16c DL |
63 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
64 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). | |
65 | * Then we also need a global mutex to serialize eventpoll_release_file() | |
66 | * and ep_free(). | |
67 | * This mutex is acquired by ep_free() during the epoll file | |
1da177e4 LT |
68 | * cleanup path and it is also acquired by eventpoll_release_file() |
69 | * if a file has been pushed inside an epoll set and it is then | |
bf6a41db | 70 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
22bacca4 DL |
71 | * It is also acquired when inserting an epoll fd onto another epoll |
72 | * fd. We do this so that we walk the epoll tree and ensure that this | |
73 | * insertion does not create a cycle of epoll file descriptors, which | |
74 | * could lead to deadlock. We need a global mutex to prevent two | |
75 | * simultaneous inserts (A into B and B into A) from racing and | |
76 | * constructing a cycle without either insert observing that it is | |
77 | * going to. | |
d8805e63 NE |
78 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
79 | * case when one epoll fd is added to another. In this case, we | |
80 | * always acquire the locks in the order of nesting (i.e. after | |
81 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired | |
82 | * before e2->mtx). Since we disallow cycles of epoll file | |
83 | * descriptors, this ensures that the mutexes are well-ordered. In | |
84 | * order to communicate this nesting to lockdep, when walking a tree | |
85 | * of epoll file descriptors, we use the current recursion depth as | |
86 | * the lockdep subkey. | |
d47de16c DL |
87 | * It is possible to drop the "ep->mtx" and to use the global |
88 | * mutex "epmutex" (together with "ep->lock") to have it working, | |
89 | * but having "ep->mtx" will make the interface more scalable. | |
144efe3e | 90 | * Events that require holding "epmutex" are very rare, while for |
d47de16c DL |
91 | * normal operations the epoll private "ep->mtx" will guarantee |
92 | * a better scalability. | |
1da177e4 LT |
93 | */ |
94 | ||
1da177e4 | 95 | /* Epoll private bits inside the event mask */ |
df0108c5 | 96 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE) |
1da177e4 | 97 | |
b6a515c8 JB |
98 | #define EPOLLINOUT_BITS (POLLIN | POLLOUT) |
99 | ||
100 | #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | POLLERR | POLLHUP | \ | |
101 | EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE) | |
102 | ||
5071f97e DL |
103 | /* Maximum number of nesting allowed inside epoll sets */ |
104 | #define EP_MAX_NESTS 4 | |
1da177e4 | 105 | |
b611967d DL |
106 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
107 | ||
d47de16c DL |
108 | #define EP_UNACTIVE_PTR ((void *) -1L) |
109 | ||
7ef9964e DL |
110 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
111 | ||
1da177e4 LT |
112 | struct epoll_filefd { |
113 | struct file *file; | |
114 | int fd; | |
39732ca5 | 115 | } __packed; |
1da177e4 LT |
116 | |
117 | /* | |
5071f97e DL |
118 | * Structure used to track possible nested calls, for too deep recursions |
119 | * and loop cycles. | |
1da177e4 | 120 | */ |
5071f97e | 121 | struct nested_call_node { |
1da177e4 | 122 | struct list_head llink; |
5071f97e | 123 | void *cookie; |
3fe4a975 | 124 | void *ctx; |
1da177e4 LT |
125 | }; |
126 | ||
127 | /* | |
5071f97e DL |
128 | * This structure is used as collector for nested calls, to check for |
129 | * maximum recursion dept and loop cycles. | |
1da177e4 | 130 | */ |
5071f97e DL |
131 | struct nested_calls { |
132 | struct list_head tasks_call_list; | |
1da177e4 LT |
133 | spinlock_t lock; |
134 | }; | |
135 | ||
d47de16c DL |
136 | /* |
137 | * Each file descriptor added to the eventpoll interface will | |
138 | * have an entry of this type linked to the "rbr" RB tree. | |
39732ca5 EW |
139 | * Avoid increasing the size of this struct, there can be many thousands |
140 | * of these on a server and we do not want this to take another cache line. | |
d47de16c DL |
141 | */ |
142 | struct epitem { | |
ae10b2b4 JB |
143 | union { |
144 | /* RB tree node links this structure to the eventpoll RB tree */ | |
145 | struct rb_node rbn; | |
146 | /* Used to free the struct epitem */ | |
147 | struct rcu_head rcu; | |
148 | }; | |
d47de16c DL |
149 | |
150 | /* List header used to link this structure to the eventpoll ready list */ | |
151 | struct list_head rdllink; | |
152 | ||
c7ea7630 DL |
153 | /* |
154 | * Works together "struct eventpoll"->ovflist in keeping the | |
155 | * single linked chain of items. | |
156 | */ | |
157 | struct epitem *next; | |
158 | ||
d47de16c DL |
159 | /* The file descriptor information this item refers to */ |
160 | struct epoll_filefd ffd; | |
161 | ||
162 | /* Number of active wait queue attached to poll operations */ | |
163 | int nwait; | |
164 | ||
165 | /* List containing poll wait queues */ | |
166 | struct list_head pwqlist; | |
167 | ||
168 | /* The "container" of this item */ | |
169 | struct eventpoll *ep; | |
170 | ||
d47de16c DL |
171 | /* List header used to link this item to the "struct file" items list */ |
172 | struct list_head fllink; | |
173 | ||
4d7e30d9 | 174 | /* wakeup_source used when EPOLLWAKEUP is set */ |
eea1d585 | 175 | struct wakeup_source __rcu *ws; |
4d7e30d9 | 176 | |
c7ea7630 DL |
177 | /* The structure that describe the interested events and the source fd */ |
178 | struct epoll_event event; | |
d47de16c DL |
179 | }; |
180 | ||
1da177e4 LT |
181 | /* |
182 | * This structure is stored inside the "private_data" member of the file | |
bf6a41db | 183 | * structure and represents the main data structure for the eventpoll |
1da177e4 LT |
184 | * interface. |
185 | */ | |
186 | struct eventpoll { | |
bf6a41db | 187 | /* Protect the access to this structure */ |
c7ea7630 | 188 | spinlock_t lock; |
1da177e4 LT |
189 | |
190 | /* | |
d47de16c DL |
191 | * This mutex is used to ensure that files are not removed |
192 | * while epoll is using them. This is held during the event | |
193 | * collection loop, the file cleanup path, the epoll file exit | |
194 | * code and the ctl operations. | |
1da177e4 | 195 | */ |
d47de16c | 196 | struct mutex mtx; |
1da177e4 LT |
197 | |
198 | /* Wait queue used by sys_epoll_wait() */ | |
199 | wait_queue_head_t wq; | |
200 | ||
201 | /* Wait queue used by file->poll() */ | |
202 | wait_queue_head_t poll_wait; | |
203 | ||
204 | /* List of ready file descriptors */ | |
205 | struct list_head rdllist; | |
206 | ||
67647d0f | 207 | /* RB tree root used to store monitored fd structs */ |
1da177e4 | 208 | struct rb_root rbr; |
d47de16c DL |
209 | |
210 | /* | |
211 | * This is a single linked list that chains all the "struct epitem" that | |
25985edc | 212 | * happened while transferring ready events to userspace w/out |
d47de16c DL |
213 | * holding ->lock. |
214 | */ | |
215 | struct epitem *ovflist; | |
7ef9964e | 216 | |
4d7e30d9 AH |
217 | /* wakeup_source used when ep_scan_ready_list is running */ |
218 | struct wakeup_source *ws; | |
219 | ||
7ef9964e DL |
220 | /* The user that created the eventpoll descriptor */ |
221 | struct user_struct *user; | |
28d82dc1 JB |
222 | |
223 | struct file *file; | |
224 | ||
225 | /* used to optimize loop detection check */ | |
226 | int visited; | |
227 | struct list_head visited_list_link; | |
bf3b9f63 SS |
228 | |
229 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
230 | /* used to track busy poll napi_id */ | |
231 | unsigned int napi_id; | |
232 | #endif | |
1da177e4 LT |
233 | }; |
234 | ||
235 | /* Wait structure used by the poll hooks */ | |
236 | struct eppoll_entry { | |
237 | /* List header used to link this structure to the "struct epitem" */ | |
238 | struct list_head llink; | |
239 | ||
240 | /* The "base" pointer is set to the container "struct epitem" */ | |
4f0989db | 241 | struct epitem *base; |
1da177e4 LT |
242 | |
243 | /* | |
244 | * Wait queue item that will be linked to the target file wait | |
245 | * queue head. | |
246 | */ | |
ac6424b9 | 247 | wait_queue_entry_t wait; |
1da177e4 LT |
248 | |
249 | /* The wait queue head that linked the "wait" wait queue item */ | |
250 | wait_queue_head_t *whead; | |
251 | }; | |
252 | ||
1da177e4 LT |
253 | /* Wrapper struct used by poll queueing */ |
254 | struct ep_pqueue { | |
255 | poll_table pt; | |
256 | struct epitem *epi; | |
257 | }; | |
258 | ||
5071f97e DL |
259 | /* Used by the ep_send_events() function as callback private data */ |
260 | struct ep_send_events_data { | |
261 | int maxevents; | |
262 | struct epoll_event __user *events; | |
263 | }; | |
264 | ||
7ef9964e DL |
265 | /* |
266 | * Configuration options available inside /proc/sys/fs/epoll/ | |
267 | */ | |
7ef9964e | 268 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 269 | static long max_user_watches __read_mostly; |
7ef9964e | 270 | |
1da177e4 | 271 | /* |
d47de16c | 272 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 273 | */ |
7ef9964e | 274 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 275 | |
22bacca4 DL |
276 | /* Used to check for epoll file descriptor inclusion loops */ |
277 | static struct nested_calls poll_loop_ncalls; | |
278 | ||
5071f97e DL |
279 | /* Used for safe wake up implementation */ |
280 | static struct nested_calls poll_safewake_ncalls; | |
281 | ||
282 | /* Used to call file's f_op->poll() under the nested calls boundaries */ | |
283 | static struct nested_calls poll_readywalk_ncalls; | |
1da177e4 LT |
284 | |
285 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 286 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
287 | |
288 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 289 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 290 | |
28d82dc1 JB |
291 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
292 | static LIST_HEAD(visited_list); | |
293 | ||
294 | /* | |
295 | * List of files with newly added links, where we may need to limit the number | |
296 | * of emanating paths. Protected by the epmutex. | |
297 | */ | |
298 | static LIST_HEAD(tfile_check_list); | |
299 | ||
7ef9964e DL |
300 | #ifdef CONFIG_SYSCTL |
301 | ||
302 | #include <linux/sysctl.h> | |
303 | ||
52bd19f7 RH |
304 | static long zero; |
305 | static long long_max = LONG_MAX; | |
7ef9964e | 306 | |
1f7e0616 | 307 | struct ctl_table epoll_table[] = { |
7ef9964e DL |
308 | { |
309 | .procname = "max_user_watches", | |
310 | .data = &max_user_watches, | |
52bd19f7 | 311 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 312 | .mode = 0644, |
52bd19f7 | 313 | .proc_handler = proc_doulongvec_minmax, |
7ef9964e | 314 | .extra1 = &zero, |
52bd19f7 | 315 | .extra2 = &long_max, |
7ef9964e | 316 | }, |
ab09203e | 317 | { } |
7ef9964e DL |
318 | }; |
319 | #endif /* CONFIG_SYSCTL */ | |
320 | ||
28d82dc1 JB |
321 | static const struct file_operations eventpoll_fops; |
322 | ||
323 | static inline int is_file_epoll(struct file *f) | |
324 | { | |
325 | return f->f_op == &eventpoll_fops; | |
326 | } | |
b030a4dd | 327 | |
67647d0f | 328 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
329 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
330 | struct file *file, int fd) | |
331 | { | |
332 | ffd->file = file; | |
333 | ffd->fd = fd; | |
334 | } | |
335 | ||
67647d0f | 336 | /* Compare RB tree keys */ |
b030a4dd PE |
337 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
338 | struct epoll_filefd *p2) | |
339 | { | |
340 | return (p1->file > p2->file ? +1: | |
341 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
342 | } | |
343 | ||
b030a4dd PE |
344 | /* Tells us if the item is currently linked */ |
345 | static inline int ep_is_linked(struct list_head *p) | |
346 | { | |
347 | return !list_empty(p); | |
348 | } | |
349 | ||
ac6424b9 | 350 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
971316f0 ON |
351 | { |
352 | return container_of(p, struct eppoll_entry, wait); | |
353 | } | |
354 | ||
b030a4dd | 355 | /* Get the "struct epitem" from a wait queue pointer */ |
ac6424b9 | 356 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
b030a4dd PE |
357 | { |
358 | return container_of(p, struct eppoll_entry, wait)->base; | |
359 | } | |
360 | ||
361 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 362 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
363 | { |
364 | return container_of(p, struct ep_pqueue, pt)->epi; | |
365 | } | |
366 | ||
367 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 368 | static inline int ep_op_has_event(int op) |
b030a4dd | 369 | { |
a80a6b85 | 370 | return op != EPOLL_CTL_DEL; |
b030a4dd PE |
371 | } |
372 | ||
1da177e4 | 373 | /* Initialize the poll safe wake up structure */ |
5071f97e | 374 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 375 | { |
5071f97e DL |
376 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
377 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
378 | } |
379 | ||
3fb0e584 DL |
380 | /** |
381 | * ep_events_available - Checks if ready events might be available. | |
382 | * | |
383 | * @ep: Pointer to the eventpoll context. | |
384 | * | |
385 | * Returns: Returns a value different than zero if ready events are available, | |
386 | * or zero otherwise. | |
387 | */ | |
388 | static inline int ep_events_available(struct eventpoll *ep) | |
389 | { | |
390 | return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; | |
391 | } | |
392 | ||
bf3b9f63 SS |
393 | #ifdef CONFIG_NET_RX_BUSY_POLL |
394 | static bool ep_busy_loop_end(void *p, unsigned long start_time) | |
395 | { | |
396 | struct eventpoll *ep = p; | |
397 | ||
398 | return ep_events_available(ep) || busy_loop_timeout(start_time); | |
399 | } | |
400 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
401 | ||
402 | /* | |
403 | * Busy poll if globally on and supporting sockets found && no events, | |
404 | * busy loop will return if need_resched or ep_events_available. | |
405 | * | |
406 | * we must do our busy polling with irqs enabled | |
407 | */ | |
408 | static void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
409 | { | |
410 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
411 | unsigned int napi_id = READ_ONCE(ep->napi_id); | |
412 | ||
413 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) | |
414 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep); | |
415 | #endif | |
416 | } | |
417 | ||
418 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
419 | { | |
420 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
421 | if (ep->napi_id) | |
422 | ep->napi_id = 0; | |
423 | #endif | |
424 | } | |
425 | ||
426 | /* | |
427 | * Set epoll busy poll NAPI ID from sk. | |
428 | */ | |
429 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
430 | { | |
431 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
432 | struct eventpoll *ep; | |
433 | unsigned int napi_id; | |
434 | struct socket *sock; | |
435 | struct sock *sk; | |
436 | int err; | |
437 | ||
438 | if (!net_busy_loop_on()) | |
439 | return; | |
440 | ||
441 | sock = sock_from_file(epi->ffd.file, &err); | |
442 | if (!sock) | |
443 | return; | |
444 | ||
445 | sk = sock->sk; | |
446 | if (!sk) | |
447 | return; | |
448 | ||
449 | napi_id = READ_ONCE(sk->sk_napi_id); | |
450 | ep = epi->ep; | |
451 | ||
452 | /* Non-NAPI IDs can be rejected | |
453 | * or | |
454 | * Nothing to do if we already have this ID | |
455 | */ | |
456 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) | |
457 | return; | |
458 | ||
459 | /* record NAPI ID for use in next busy poll */ | |
460 | ep->napi_id = napi_id; | |
461 | #endif | |
462 | } | |
463 | ||
5071f97e DL |
464 | /** |
465 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
466 | * that the recursion limit is not exceeded, and that | |
467 | * the same nested call (by the meaning of same cookie) is | |
468 | * no re-entered. | |
469 | * | |
470 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
471 | * @max_nests: Maximum number of allowed nesting calls. | |
472 | * @nproc: Nested call core function pointer. | |
473 | * @priv: Opaque data to be passed to the @nproc callback. | |
474 | * @cookie: Cookie to be used to identify this nested call. | |
3fe4a975 | 475 | * @ctx: This instance context. |
5071f97e DL |
476 | * |
477 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
478 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 479 | */ |
5071f97e DL |
480 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
481 | int (*nproc)(void *, void *, int), void *priv, | |
3fe4a975 | 482 | void *cookie, void *ctx) |
1da177e4 | 483 | { |
5071f97e | 484 | int error, call_nests = 0; |
1da177e4 | 485 | unsigned long flags; |
5071f97e DL |
486 | struct list_head *lsthead = &ncalls->tasks_call_list; |
487 | struct nested_call_node *tncur; | |
488 | struct nested_call_node tnode; | |
1da177e4 | 489 | |
5071f97e | 490 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 491 | |
5071f97e DL |
492 | /* |
493 | * Try to see if the current task is already inside this wakeup call. | |
494 | * We use a list here, since the population inside this set is always | |
495 | * very much limited. | |
496 | */ | |
b70c3940 | 497 | list_for_each_entry(tncur, lsthead, llink) { |
3fe4a975 | 498 | if (tncur->ctx == ctx && |
5071f97e | 499 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
1da177e4 LT |
500 | /* |
501 | * Ops ... loop detected or maximum nest level reached. | |
502 | * We abort this wake by breaking the cycle itself. | |
503 | */ | |
abff55ce TB |
504 | error = -1; |
505 | goto out_unlock; | |
1da177e4 LT |
506 | } |
507 | } | |
508 | ||
5071f97e | 509 | /* Add the current task and cookie to the list */ |
3fe4a975 | 510 | tnode.ctx = ctx; |
5071f97e | 511 | tnode.cookie = cookie; |
1da177e4 LT |
512 | list_add(&tnode.llink, lsthead); |
513 | ||
5071f97e | 514 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 515 | |
5071f97e DL |
516 | /* Call the nested function */ |
517 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
518 | |
519 | /* Remove the current task from the list */ | |
5071f97e | 520 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 521 | list_del(&tnode.llink); |
3fe4a975 | 522 | out_unlock: |
5071f97e DL |
523 | spin_unlock_irqrestore(&ncalls->lock, flags); |
524 | ||
525 | return error; | |
526 | } | |
527 | ||
02edc6fc SR |
528 | /* |
529 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
530 | * the use of wait queues used by epoll is done in a very controlled | |
531 | * manner. Wake ups can nest inside each other, but are never done | |
532 | * with the same locking. For example: | |
533 | * | |
534 | * dfd = socket(...); | |
535 | * efd1 = epoll_create(); | |
536 | * efd2 = epoll_create(); | |
537 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
538 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
539 | * | |
540 | * When a packet arrives to the device underneath "dfd", the net code will | |
541 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
542 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
543 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
544 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
545 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
546 | * that ends up in another wake_up(), after having checked about the | |
547 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
548 | * avoid stack blasting. | |
549 | * | |
550 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
551 | * this special case of epoll. | |
552 | */ | |
2dfa4eea DL |
553 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
554 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
555 | unsigned long events, int subclass) | |
556 | { | |
557 | unsigned long flags; | |
558 | ||
559 | spin_lock_irqsave_nested(&wqueue->lock, flags, subclass); | |
560 | wake_up_locked_poll(wqueue, events); | |
561 | spin_unlock_irqrestore(&wqueue->lock, flags); | |
562 | } | |
563 | #else | |
564 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, | |
565 | unsigned long events, int subclass) | |
566 | { | |
567 | wake_up_poll(wqueue, events); | |
568 | } | |
569 | #endif | |
570 | ||
5071f97e DL |
571 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
572 | { | |
2dfa4eea DL |
573 | ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN, |
574 | 1 + call_nests); | |
5071f97e DL |
575 | return 0; |
576 | } | |
577 | ||
578 | /* | |
579 | * Perform a safe wake up of the poll wait list. The problem is that | |
580 | * with the new callback'd wake up system, it is possible that the | |
581 | * poll callback is reentered from inside the call to wake_up() done | |
582 | * on the poll wait queue head. The rule is that we cannot reenter the | |
583 | * wake up code from the same task more than EP_MAX_NESTS times, | |
584 | * and we cannot reenter the same wait queue head at all. This will | |
585 | * enable to have a hierarchy of epoll file descriptor of no more than | |
586 | * EP_MAX_NESTS deep. | |
587 | */ | |
588 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
589 | { | |
3fe4a975 DL |
590 | int this_cpu = get_cpu(); |
591 | ||
5071f97e | 592 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
3fe4a975 DL |
593 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
594 | ||
595 | put_cpu(); | |
1da177e4 LT |
596 | } |
597 | ||
971316f0 ON |
598 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
599 | { | |
600 | wait_queue_head_t *whead; | |
601 | ||
602 | rcu_read_lock(); | |
138e4ad6 ON |
603 | /* |
604 | * If it is cleared by POLLFREE, it should be rcu-safe. | |
605 | * If we read NULL we need a barrier paired with | |
606 | * smp_store_release() in ep_poll_callback(), otherwise | |
607 | * we rely on whead->lock. | |
608 | */ | |
609 | whead = smp_load_acquire(&pwq->whead); | |
971316f0 ON |
610 | if (whead) |
611 | remove_wait_queue(whead, &pwq->wait); | |
612 | rcu_read_unlock(); | |
613 | } | |
614 | ||
1da177e4 | 615 | /* |
d1bc90dd TB |
616 | * This function unregisters poll callbacks from the associated file |
617 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
618 | * ep_free). | |
1da177e4 | 619 | */ |
7699acd1 | 620 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 621 | { |
7699acd1 DL |
622 | struct list_head *lsthead = &epi->pwqlist; |
623 | struct eppoll_entry *pwq; | |
1da177e4 | 624 | |
d1bc90dd TB |
625 | while (!list_empty(lsthead)) { |
626 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 627 | |
d1bc90dd | 628 | list_del(&pwq->llink); |
971316f0 | 629 | ep_remove_wait_queue(pwq); |
d1bc90dd | 630 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 631 | } |
1da177e4 LT |
632 | } |
633 | ||
eea1d585 EW |
634 | /* call only when ep->mtx is held */ |
635 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) | |
636 | { | |
637 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); | |
638 | } | |
639 | ||
640 | /* call only when ep->mtx is held */ | |
641 | static inline void ep_pm_stay_awake(struct epitem *epi) | |
642 | { | |
643 | struct wakeup_source *ws = ep_wakeup_source(epi); | |
644 | ||
645 | if (ws) | |
646 | __pm_stay_awake(ws); | |
647 | } | |
648 | ||
649 | static inline bool ep_has_wakeup_source(struct epitem *epi) | |
650 | { | |
651 | return rcu_access_pointer(epi->ws) ? true : false; | |
652 | } | |
653 | ||
654 | /* call when ep->mtx cannot be held (ep_poll_callback) */ | |
655 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) | |
656 | { | |
657 | struct wakeup_source *ws; | |
658 | ||
659 | rcu_read_lock(); | |
660 | ws = rcu_dereference(epi->ws); | |
661 | if (ws) | |
662 | __pm_stay_awake(ws); | |
663 | rcu_read_unlock(); | |
664 | } | |
665 | ||
5071f97e DL |
666 | /** |
667 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
668 | * the scan code, to call f_op->poll(). Also allows for | |
669 | * O(NumReady) performance. | |
670 | * | |
671 | * @ep: Pointer to the epoll private data structure. | |
672 | * @sproc: Pointer to the scan callback. | |
673 | * @priv: Private opaque data passed to the @sproc callback. | |
d8805e63 | 674 | * @depth: The current depth of recursive f_op->poll calls. |
67347fe4 | 675 | * @ep_locked: caller already holds ep->mtx |
5071f97e DL |
676 | * |
677 | * Returns: The same integer error code returned by the @sproc callback. | |
678 | */ | |
679 | static int ep_scan_ready_list(struct eventpoll *ep, | |
680 | int (*sproc)(struct eventpoll *, | |
681 | struct list_head *, void *), | |
67347fe4 | 682 | void *priv, int depth, bool ep_locked) |
5071f97e DL |
683 | { |
684 | int error, pwake = 0; | |
685 | unsigned long flags; | |
686 | struct epitem *epi, *nepi; | |
296e236e | 687 | LIST_HEAD(txlist); |
5071f97e DL |
688 | |
689 | /* | |
690 | * We need to lock this because we could be hit by | |
e057e15f | 691 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e | 692 | */ |
67347fe4 JB |
693 | |
694 | if (!ep_locked) | |
695 | mutex_lock_nested(&ep->mtx, depth); | |
5071f97e DL |
696 | |
697 | /* | |
698 | * Steal the ready list, and re-init the original one to the | |
699 | * empty list. Also, set ep->ovflist to NULL so that events | |
700 | * happening while looping w/out locks, are not lost. We cannot | |
701 | * have the poll callback to queue directly on ep->rdllist, | |
702 | * because we want the "sproc" callback to be able to do it | |
703 | * in a lockless way. | |
704 | */ | |
705 | spin_lock_irqsave(&ep->lock, flags); | |
296e236e | 706 | list_splice_init(&ep->rdllist, &txlist); |
5071f97e DL |
707 | ep->ovflist = NULL; |
708 | spin_unlock_irqrestore(&ep->lock, flags); | |
709 | ||
710 | /* | |
711 | * Now call the callback function. | |
712 | */ | |
713 | error = (*sproc)(ep, &txlist, priv); | |
714 | ||
715 | spin_lock_irqsave(&ep->lock, flags); | |
716 | /* | |
717 | * During the time we spent inside the "sproc" callback, some | |
718 | * other events might have been queued by the poll callback. | |
719 | * We re-insert them inside the main ready-list here. | |
720 | */ | |
721 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
722 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { | |
723 | /* | |
724 | * We need to check if the item is already in the list. | |
725 | * During the "sproc" callback execution time, items are | |
726 | * queued into ->ovflist but the "txlist" might already | |
727 | * contain them, and the list_splice() below takes care of them. | |
728 | */ | |
4d7e30d9 | 729 | if (!ep_is_linked(&epi->rdllink)) { |
5071f97e | 730 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 731 | ep_pm_stay_awake(epi); |
4d7e30d9 | 732 | } |
5071f97e DL |
733 | } |
734 | /* | |
735 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
736 | * releasing the lock, events will be queued in the normal way inside | |
737 | * ep->rdllist. | |
738 | */ | |
739 | ep->ovflist = EP_UNACTIVE_PTR; | |
740 | ||
741 | /* | |
742 | * Quickly re-inject items left on "txlist". | |
743 | */ | |
744 | list_splice(&txlist, &ep->rdllist); | |
4d7e30d9 | 745 | __pm_relax(ep->ws); |
5071f97e DL |
746 | |
747 | if (!list_empty(&ep->rdllist)) { | |
748 | /* | |
296e236e DL |
749 | * Wake up (if active) both the eventpoll wait list and |
750 | * the ->poll() wait list (delayed after we release the lock). | |
5071f97e DL |
751 | */ |
752 | if (waitqueue_active(&ep->wq)) | |
753 | wake_up_locked(&ep->wq); | |
754 | if (waitqueue_active(&ep->poll_wait)) | |
755 | pwake++; | |
756 | } | |
757 | spin_unlock_irqrestore(&ep->lock, flags); | |
758 | ||
67347fe4 JB |
759 | if (!ep_locked) |
760 | mutex_unlock(&ep->mtx); | |
5071f97e DL |
761 | |
762 | /* We have to call this outside the lock */ | |
763 | if (pwake) | |
764 | ep_poll_safewake(&ep->poll_wait); | |
765 | ||
766 | return error; | |
767 | } | |
768 | ||
ae10b2b4 JB |
769 | static void epi_rcu_free(struct rcu_head *head) |
770 | { | |
771 | struct epitem *epi = container_of(head, struct epitem, rcu); | |
772 | kmem_cache_free(epi_cache, epi); | |
773 | } | |
774 | ||
7699acd1 DL |
775 | /* |
776 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 777 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
778 | */ |
779 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
780 | { | |
7699acd1 DL |
781 | unsigned long flags; |
782 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
783 | |
784 | /* | |
7699acd1 DL |
785 | * Removes poll wait queue hooks. We _have_ to do this without holding |
786 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
787 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
788 | * queue head lock when unregistering the wait queue. The wakeup callback | |
789 | * will run by holding the wait queue head lock and will call our callback | |
790 | * that will try to get "ep->lock". | |
1da177e4 | 791 | */ |
7699acd1 | 792 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 793 | |
7699acd1 | 794 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 795 | spin_lock(&file->f_lock); |
ae10b2b4 | 796 | list_del_rcu(&epi->fllink); |
68499914 | 797 | spin_unlock(&file->f_lock); |
1da177e4 | 798 | |
cdac75e6 | 799 | rb_erase(&epi->rbn, &ep->rbr); |
1da177e4 | 800 | |
c7ea7630 DL |
801 | spin_lock_irqsave(&ep->lock, flags); |
802 | if (ep_is_linked(&epi->rdllink)) | |
803 | list_del_init(&epi->rdllink); | |
804 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 805 | |
eea1d585 | 806 | wakeup_source_unregister(ep_wakeup_source(epi)); |
ae10b2b4 JB |
807 | /* |
808 | * At this point it is safe to free the eventpoll item. Use the union | |
809 | * field epi->rcu, since we are trying to minimize the size of | |
810 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by | |
811 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make | |
812 | * use of the rbn field. | |
813 | */ | |
814 | call_rcu(&epi->rcu, epi_rcu_free); | |
1da177e4 | 815 | |
52bd19f7 | 816 | atomic_long_dec(&ep->user->epoll_watches); |
7ef9964e | 817 | |
c7ea7630 | 818 | return 0; |
1da177e4 LT |
819 | } |
820 | ||
7699acd1 | 821 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 822 | { |
7699acd1 DL |
823 | struct rb_node *rbp; |
824 | struct epitem *epi; | |
1da177e4 | 825 | |
7699acd1 DL |
826 | /* We need to release all tasks waiting for these file */ |
827 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 828 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 829 | |
7699acd1 DL |
830 | /* |
831 | * We need to lock this because we could be hit by | |
832 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 833 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
834 | * is on the way to be removed and no one has references to it |
835 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 836 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
837 | */ |
838 | mutex_lock(&epmutex); | |
1da177e4 LT |
839 | |
840 | /* | |
7699acd1 | 841 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 842 | */ |
7699acd1 DL |
843 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
844 | epi = rb_entry(rbp, struct epitem, rbn); | |
845 | ||
846 | ep_unregister_pollwait(ep, epi); | |
91cf5ab6 | 847 | cond_resched(); |
7699acd1 | 848 | } |
1da177e4 LT |
849 | |
850 | /* | |
7699acd1 DL |
851 | * Walks through the whole tree by freeing each "struct epitem". At this |
852 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 853 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 854 | * us during this operation. So we can avoid the lock on "ep->lock". |
ddf676c3 EW |
855 | * We do not need to lock ep->mtx, either, we only do it to prevent |
856 | * a lockdep warning. | |
1da177e4 | 857 | */ |
ddf676c3 | 858 | mutex_lock(&ep->mtx); |
c80544dc | 859 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
7699acd1 DL |
860 | epi = rb_entry(rbp, struct epitem, rbn); |
861 | ep_remove(ep, epi); | |
91cf5ab6 | 862 | cond_resched(); |
7699acd1 | 863 | } |
ddf676c3 | 864 | mutex_unlock(&ep->mtx); |
1da177e4 | 865 | |
7699acd1 | 866 | mutex_unlock(&epmutex); |
d47de16c | 867 | mutex_destroy(&ep->mtx); |
7ef9964e | 868 | free_uid(ep->user); |
4d7e30d9 | 869 | wakeup_source_unregister(ep->ws); |
f0ee9aab | 870 | kfree(ep); |
7699acd1 | 871 | } |
1da177e4 | 872 | |
7699acd1 DL |
873 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
874 | { | |
875 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 876 | |
f0ee9aab | 877 | if (ep) |
7699acd1 | 878 | ep_free(ep); |
7699acd1 | 879 | |
7699acd1 | 880 | return 0; |
1da177e4 LT |
881 | } |
882 | ||
450d89ec EW |
883 | static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt) |
884 | { | |
885 | pt->_key = epi->event.events; | |
886 | ||
887 | return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events; | |
888 | } | |
889 | ||
296e236e DL |
890 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
891 | void *priv) | |
5071f97e DL |
892 | { |
893 | struct epitem *epi, *tmp; | |
626cf236 | 894 | poll_table pt; |
5071f97e | 895 | |
626cf236 | 896 | init_poll_funcptr(&pt, NULL); |
450d89ec | 897 | |
5071f97e | 898 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
450d89ec | 899 | if (ep_item_poll(epi, &pt)) |
5071f97e | 900 | return POLLIN | POLLRDNORM; |
296e236e | 901 | else { |
5071f97e DL |
902 | /* |
903 | * Item has been dropped into the ready list by the poll | |
904 | * callback, but it's not actually ready, as far as | |
905 | * caller requested events goes. We can remove it here. | |
906 | */ | |
eea1d585 | 907 | __pm_relax(ep_wakeup_source(epi)); |
5071f97e | 908 | list_del_init(&epi->rdllink); |
296e236e | 909 | } |
5071f97e DL |
910 | } |
911 | ||
912 | return 0; | |
913 | } | |
914 | ||
67347fe4 JB |
915 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
916 | poll_table *pt); | |
917 | ||
918 | struct readyevents_arg { | |
919 | struct eventpoll *ep; | |
920 | bool locked; | |
921 | }; | |
922 | ||
5071f97e DL |
923 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) |
924 | { | |
67347fe4 JB |
925 | struct readyevents_arg *arg = priv; |
926 | ||
927 | return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL, | |
928 | call_nests + 1, arg->locked); | |
5071f97e DL |
929 | } |
930 | ||
7699acd1 DL |
931 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
932 | { | |
5071f97e | 933 | int pollflags; |
7699acd1 | 934 | struct eventpoll *ep = file->private_data; |
67347fe4 JB |
935 | struct readyevents_arg arg; |
936 | ||
937 | /* | |
938 | * During ep_insert() we already hold the ep->mtx for the tfile. | |
939 | * Prevent re-aquisition. | |
940 | */ | |
941 | arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc); | |
942 | arg.ep = ep; | |
1da177e4 | 943 | |
7699acd1 DL |
944 | /* Insert inside our poll wait queue */ |
945 | poll_wait(file, &ep->poll_wait, wait); | |
946 | ||
5071f97e DL |
947 | /* |
948 | * Proceed to find out if wanted events are really available inside | |
949 | * the ready list. This need to be done under ep_call_nested() | |
950 | * supervision, since the call to f_op->poll() done on listed files | |
951 | * could re-enter here. | |
952 | */ | |
953 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, | |
67347fe4 | 954 | ep_poll_readyevents_proc, &arg, ep, current); |
7699acd1 | 955 | |
296e236e | 956 | return pollflags != -1 ? pollflags : 0; |
7699acd1 DL |
957 | } |
958 | ||
138d22b5 | 959 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 960 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
138d22b5 CG |
961 | { |
962 | struct eventpoll *ep = f->private_data; | |
963 | struct rb_node *rbp; | |
138d22b5 CG |
964 | |
965 | mutex_lock(&ep->mtx); | |
966 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { | |
967 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); | |
77493f04 | 968 | struct inode *inode = file_inode(epi->ffd.file); |
138d22b5 | 969 | |
77493f04 CG |
970 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
971 | " pos:%lli ino:%lx sdev:%x\n", | |
a3816ab0 | 972 | epi->ffd.fd, epi->event.events, |
77493f04 CG |
973 | (long long)epi->event.data, |
974 | (long long)epi->ffd.file->f_pos, | |
975 | inode->i_ino, inode->i_sb->s_dev); | |
a3816ab0 | 976 | if (seq_has_overflowed(m)) |
138d22b5 CG |
977 | break; |
978 | } | |
979 | mutex_unlock(&ep->mtx); | |
138d22b5 CG |
980 | } |
981 | #endif | |
982 | ||
7699acd1 DL |
983 | /* File callbacks that implement the eventpoll file behaviour */ |
984 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
985 | #ifdef CONFIG_PROC_FS |
986 | .show_fdinfo = ep_show_fdinfo, | |
987 | #endif | |
7699acd1 | 988 | .release = ep_eventpoll_release, |
6038f373 AB |
989 | .poll = ep_eventpoll_poll, |
990 | .llseek = noop_llseek, | |
7699acd1 DL |
991 | }; |
992 | ||
b611967d | 993 | /* |
7699acd1 DL |
994 | * This is called from eventpoll_release() to unlink files from the eventpoll |
995 | * interface. We need to have this facility to cleanup correctly files that are | |
996 | * closed without being removed from the eventpoll interface. | |
b611967d | 997 | */ |
7699acd1 | 998 | void eventpoll_release_file(struct file *file) |
b611967d | 999 | { |
7699acd1 | 1000 | struct eventpoll *ep; |
ebe06187 | 1001 | struct epitem *epi, *next; |
b611967d DL |
1002 | |
1003 | /* | |
68499914 | 1004 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 1005 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 1006 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 1007 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 1008 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 1009 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 1010 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 1011 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 1012 | * from anywhere but ep_free(). |
68499914 JC |
1013 | * |
1014 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 1015 | */ |
7699acd1 | 1016 | mutex_lock(&epmutex); |
ebe06187 | 1017 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
7699acd1 | 1018 | ep = epi->ep; |
d8805e63 | 1019 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 1020 | ep_remove(ep, epi); |
d47de16c | 1021 | mutex_unlock(&ep->mtx); |
b611967d | 1022 | } |
7699acd1 | 1023 | mutex_unlock(&epmutex); |
b611967d DL |
1024 | } |
1025 | ||
53d2be79 | 1026 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 1027 | { |
7ef9964e DL |
1028 | int error; |
1029 | struct user_struct *user; | |
1030 | struct eventpoll *ep; | |
1da177e4 | 1031 | |
7ef9964e | 1032 | user = get_current_user(); |
7ef9964e DL |
1033 | error = -ENOMEM; |
1034 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
1035 | if (unlikely(!ep)) | |
1036 | goto free_uid; | |
1da177e4 | 1037 | |
c7ea7630 | 1038 | spin_lock_init(&ep->lock); |
d47de16c | 1039 | mutex_init(&ep->mtx); |
1da177e4 LT |
1040 | init_waitqueue_head(&ep->wq); |
1041 | init_waitqueue_head(&ep->poll_wait); | |
1042 | INIT_LIST_HEAD(&ep->rdllist); | |
1043 | ep->rbr = RB_ROOT; | |
d47de16c | 1044 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 1045 | ep->user = user; |
1da177e4 | 1046 | |
53d2be79 | 1047 | *pep = ep; |
1da177e4 | 1048 | |
1da177e4 | 1049 | return 0; |
7ef9964e DL |
1050 | |
1051 | free_uid: | |
1052 | free_uid(user); | |
1053 | return error; | |
1da177e4 LT |
1054 | } |
1055 | ||
1da177e4 | 1056 | /* |
c7ea7630 DL |
1057 | * Search the file inside the eventpoll tree. The RB tree operations |
1058 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
1059 | * "mtx" held. | |
1da177e4 LT |
1060 | */ |
1061 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
1062 | { | |
1063 | int kcmp; | |
1da177e4 LT |
1064 | struct rb_node *rbp; |
1065 | struct epitem *epi, *epir = NULL; | |
1066 | struct epoll_filefd ffd; | |
1067 | ||
b030a4dd | 1068 | ep_set_ffd(&ffd, file, fd); |
1da177e4 LT |
1069 | for (rbp = ep->rbr.rb_node; rbp; ) { |
1070 | epi = rb_entry(rbp, struct epitem, rbn); | |
b030a4dd | 1071 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
1072 | if (kcmp > 0) |
1073 | rbp = rbp->rb_right; | |
1074 | else if (kcmp < 0) | |
1075 | rbp = rbp->rb_left; | |
1076 | else { | |
1da177e4 LT |
1077 | epir = epi; |
1078 | break; | |
1079 | } | |
1080 | } | |
1da177e4 | 1081 | |
1da177e4 LT |
1082 | return epir; |
1083 | } | |
1084 | ||
92ef6da3 | 1085 | #ifdef CONFIG_CHECKPOINT_RESTORE |
0791e364 CG |
1086 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
1087 | { | |
1088 | struct rb_node *rbp; | |
1089 | struct epitem *epi; | |
1090 | ||
1091 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { | |
1092 | epi = rb_entry(rbp, struct epitem, rbn); | |
1093 | if (epi->ffd.fd == tfd) { | |
1094 | if (toff == 0) | |
1095 | return epi; | |
1096 | else | |
1097 | toff--; | |
1098 | } | |
1099 | cond_resched(); | |
1100 | } | |
1101 | ||
1102 | return NULL; | |
1103 | } | |
1104 | ||
1105 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, | |
1106 | unsigned long toff) | |
1107 | { | |
1108 | struct file *file_raw; | |
1109 | struct eventpoll *ep; | |
1110 | struct epitem *epi; | |
1111 | ||
1112 | if (!is_file_epoll(file)) | |
1113 | return ERR_PTR(-EINVAL); | |
1114 | ||
1115 | ep = file->private_data; | |
1116 | ||
1117 | mutex_lock(&ep->mtx); | |
1118 | epi = ep_find_tfd(ep, tfd, toff); | |
1119 | if (epi) | |
1120 | file_raw = epi->ffd.file; | |
1121 | else | |
1122 | file_raw = ERR_PTR(-ENOENT); | |
1123 | mutex_unlock(&ep->mtx); | |
1124 | ||
1125 | return file_raw; | |
1126 | } | |
92ef6da3 | 1127 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
0791e364 | 1128 | |
1da177e4 | 1129 | /* |
7699acd1 | 1130 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 1131 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 1132 | * have events to report. |
1da177e4 | 1133 | */ |
ac6424b9 | 1134 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 1135 | { |
7699acd1 DL |
1136 | int pwake = 0; |
1137 | unsigned long flags; | |
1138 | struct epitem *epi = ep_item_from_wait(wait); | |
1139 | struct eventpoll *ep = epi->ep; | |
df0108c5 | 1140 | int ewake = 0; |
1da177e4 | 1141 | |
c7ea7630 | 1142 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1143 | |
bf3b9f63 SS |
1144 | ep_set_busy_poll_napi_id(epi); |
1145 | ||
7699acd1 DL |
1146 | /* |
1147 | * If the event mask does not contain any poll(2) event, we consider the | |
1148 | * descriptor to be disabled. This condition is likely the effect of the | |
1149 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1150 | * until the next EPOLL_CTL_MOD will be issued. | |
1151 | */ | |
1152 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
1153 | goto out_unlock; |
1154 | ||
2dfa4eea DL |
1155 | /* |
1156 | * Check the events coming with the callback. At this stage, not | |
1157 | * every device reports the events in the "key" parameter of the | |
1158 | * callback. We need to be able to handle both cases here, hence the | |
1159 | * test for "key" != NULL before the event match test. | |
1160 | */ | |
1161 | if (key && !((unsigned long) key & epi->event.events)) | |
1162 | goto out_unlock; | |
1163 | ||
d47de16c | 1164 | /* |
bf6a41db | 1165 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 1166 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 1167 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
1168 | * chained in ep->ovflist and requeued later on. |
1169 | */ | |
1170 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
1171 | if (epi->next == EP_UNACTIVE_PTR) { | |
1172 | epi->next = ep->ovflist; | |
1173 | ep->ovflist = epi; | |
4d7e30d9 AH |
1174 | if (epi->ws) { |
1175 | /* | |
1176 | * Activate ep->ws since epi->ws may get | |
1177 | * deactivated at any time. | |
1178 | */ | |
1179 | __pm_stay_awake(ep->ws); | |
1180 | } | |
1181 | ||
d47de16c DL |
1182 | } |
1183 | goto out_unlock; | |
1184 | } | |
1da177e4 | 1185 | |
7699acd1 | 1186 | /* If this file is already in the ready list we exit soon */ |
4d7e30d9 | 1187 | if (!ep_is_linked(&epi->rdllink)) { |
5071f97e | 1188 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1189 | ep_pm_stay_awake_rcu(epi); |
4d7e30d9 | 1190 | } |
7699acd1 | 1191 | |
7699acd1 DL |
1192 | /* |
1193 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1194 | * wait list. | |
1195 | */ | |
df0108c5 | 1196 | if (waitqueue_active(&ep->wq)) { |
b6a515c8 JB |
1197 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
1198 | !((unsigned long)key & POLLFREE)) { | |
1199 | switch ((unsigned long)key & EPOLLINOUT_BITS) { | |
1200 | case POLLIN: | |
1201 | if (epi->event.events & POLLIN) | |
1202 | ewake = 1; | |
1203 | break; | |
1204 | case POLLOUT: | |
1205 | if (epi->event.events & POLLOUT) | |
1206 | ewake = 1; | |
1207 | break; | |
1208 | case 0: | |
1209 | ewake = 1; | |
1210 | break; | |
1211 | } | |
1212 | } | |
4a6e9e2c | 1213 | wake_up_locked(&ep->wq); |
df0108c5 | 1214 | } |
7699acd1 DL |
1215 | if (waitqueue_active(&ep->poll_wait)) |
1216 | pwake++; | |
1217 | ||
d47de16c | 1218 | out_unlock: |
c7ea7630 | 1219 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1220 | |
7699acd1 DL |
1221 | /* We have to call this outside the lock */ |
1222 | if (pwake) | |
5071f97e | 1223 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 | 1224 | |
138e4ad6 ON |
1225 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
1226 | ewake = 1; | |
1227 | ||
1228 | if ((unsigned long)key & POLLFREE) { | |
1229 | /* | |
1230 | * If we race with ep_remove_wait_queue() it can miss | |
1231 | * ->whead = NULL and do another remove_wait_queue() after | |
1232 | * us, so we can't use __remove_wait_queue(). | |
1233 | */ | |
1234 | list_del_init(&wait->entry); | |
1235 | /* | |
1236 | * ->whead != NULL protects us from the race with ep_free() | |
1237 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock | |
1238 | * held by the caller. Once we nullify it, nothing protects | |
1239 | * ep/epi or even wait. | |
1240 | */ | |
1241 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); | |
1242 | } | |
df0108c5 | 1243 | |
138e4ad6 | 1244 | return ewake; |
7699acd1 | 1245 | } |
1da177e4 LT |
1246 | |
1247 | /* | |
1248 | * This is the callback that is used to add our wait queue to the | |
1249 | * target file wakeup lists. | |
1250 | */ | |
1251 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1252 | poll_table *pt) | |
1253 | { | |
b030a4dd | 1254 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
1255 | struct eppoll_entry *pwq; |
1256 | ||
e94b1766 | 1257 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
1258 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
1259 | pwq->whead = whead; | |
1260 | pwq->base = epi; | |
df0108c5 JB |
1261 | if (epi->event.events & EPOLLEXCLUSIVE) |
1262 | add_wait_queue_exclusive(whead, &pwq->wait); | |
1263 | else | |
1264 | add_wait_queue(whead, &pwq->wait); | |
1da177e4 LT |
1265 | list_add_tail(&pwq->llink, &epi->pwqlist); |
1266 | epi->nwait++; | |
296e236e | 1267 | } else { |
1da177e4 LT |
1268 | /* We have to signal that an error occurred */ |
1269 | epi->nwait = -1; | |
296e236e | 1270 | } |
1da177e4 LT |
1271 | } |
1272 | ||
1da177e4 LT |
1273 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1274 | { | |
1275 | int kcmp; | |
1276 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
1277 | struct epitem *epic; | |
1278 | ||
1279 | while (*p) { | |
1280 | parent = *p; | |
1281 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1282 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
1da177e4 LT |
1283 | if (kcmp > 0) |
1284 | p = &parent->rb_right; | |
1285 | else | |
1286 | p = &parent->rb_left; | |
1287 | } | |
1288 | rb_link_node(&epi->rbn, parent, p); | |
1289 | rb_insert_color(&epi->rbn, &ep->rbr); | |
1290 | } | |
1291 | ||
a80a6b85 AM |
1292 | |
1293 | ||
28d82dc1 JB |
1294 | #define PATH_ARR_SIZE 5 |
1295 | /* | |
1296 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1297 | * from a single file of interest. For example, we allow 1000 paths of length | |
1298 | * 1, to emanate from each file of interest. This essentially represents the | |
1299 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1300 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1301 | * is connected to n file sources. In this case each file source has 1 path | |
1302 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1303 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1304 | * and delete can't add additional paths. Protected by the epmutex. | |
1305 | */ | |
1306 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1307 | static int path_count[PATH_ARR_SIZE]; | |
1308 | ||
1309 | static int path_count_inc(int nests) | |
1310 | { | |
93dc6107 JB |
1311 | /* Allow an arbitrary number of depth 1 paths */ |
1312 | if (nests == 0) | |
1313 | return 0; | |
1314 | ||
28d82dc1 JB |
1315 | if (++path_count[nests] > path_limits[nests]) |
1316 | return -1; | |
1317 | return 0; | |
1318 | } | |
1319 | ||
1320 | static void path_count_init(void) | |
1321 | { | |
1322 | int i; | |
1323 | ||
1324 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1325 | path_count[i] = 0; | |
1326 | } | |
1327 | ||
1328 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) | |
1329 | { | |
1330 | int error = 0; | |
1331 | struct file *file = priv; | |
1332 | struct file *child_file; | |
1333 | struct epitem *epi; | |
1334 | ||
ae10b2b4 JB |
1335 | /* CTL_DEL can remove links here, but that can't increase our count */ |
1336 | rcu_read_lock(); | |
1337 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { | |
28d82dc1 JB |
1338 | child_file = epi->ep->file; |
1339 | if (is_file_epoll(child_file)) { | |
1340 | if (list_empty(&child_file->f_ep_links)) { | |
1341 | if (path_count_inc(call_nests)) { | |
1342 | error = -1; | |
1343 | break; | |
1344 | } | |
1345 | } else { | |
1346 | error = ep_call_nested(&poll_loop_ncalls, | |
1347 | EP_MAX_NESTS, | |
1348 | reverse_path_check_proc, | |
1349 | child_file, child_file, | |
1350 | current); | |
1351 | } | |
1352 | if (error != 0) | |
1353 | break; | |
1354 | } else { | |
1355 | printk(KERN_ERR "reverse_path_check_proc: " | |
1356 | "file is not an ep!\n"); | |
1357 | } | |
1358 | } | |
ae10b2b4 | 1359 | rcu_read_unlock(); |
28d82dc1 JB |
1360 | return error; |
1361 | } | |
1362 | ||
1363 | /** | |
1364 | * reverse_path_check - The tfile_check_list is list of file *, which have | |
1365 | * links that are proposed to be newly added. We need to | |
1366 | * make sure that those added links don't add too many | |
1367 | * paths such that we will spend all our time waking up | |
1368 | * eventpoll objects. | |
1369 | * | |
1370 | * Returns: Returns zero if the proposed links don't create too many paths, | |
1371 | * -1 otherwise. | |
1372 | */ | |
1373 | static int reverse_path_check(void) | |
1374 | { | |
28d82dc1 JB |
1375 | int error = 0; |
1376 | struct file *current_file; | |
1377 | ||
1378 | /* let's call this for all tfiles */ | |
1379 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { | |
28d82dc1 JB |
1380 | path_count_init(); |
1381 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
1382 | reverse_path_check_proc, current_file, | |
1383 | current_file, current); | |
1384 | if (error) | |
1385 | break; | |
1386 | } | |
1387 | return error; | |
1388 | } | |
1389 | ||
4d7e30d9 AH |
1390 | static int ep_create_wakeup_source(struct epitem *epi) |
1391 | { | |
1392 | const char *name; | |
eea1d585 | 1393 | struct wakeup_source *ws; |
4d7e30d9 AH |
1394 | |
1395 | if (!epi->ep->ws) { | |
1396 | epi->ep->ws = wakeup_source_register("eventpoll"); | |
1397 | if (!epi->ep->ws) | |
1398 | return -ENOMEM; | |
1399 | } | |
1400 | ||
1401 | name = epi->ffd.file->f_path.dentry->d_name.name; | |
eea1d585 EW |
1402 | ws = wakeup_source_register(name); |
1403 | ||
1404 | if (!ws) | |
4d7e30d9 | 1405 | return -ENOMEM; |
eea1d585 | 1406 | rcu_assign_pointer(epi->ws, ws); |
4d7e30d9 AH |
1407 | |
1408 | return 0; | |
1409 | } | |
1410 | ||
eea1d585 EW |
1411 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
1412 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) | |
4d7e30d9 | 1413 | { |
eea1d585 EW |
1414 | struct wakeup_source *ws = ep_wakeup_source(epi); |
1415 | ||
d6d67e72 | 1416 | RCU_INIT_POINTER(epi->ws, NULL); |
eea1d585 EW |
1417 | |
1418 | /* | |
1419 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is | |
1420 | * used internally by wakeup_source_remove, too (called by | |
1421 | * wakeup_source_unregister), so we cannot use call_rcu | |
1422 | */ | |
1423 | synchronize_rcu(); | |
1424 | wakeup_source_unregister(ws); | |
4d7e30d9 AH |
1425 | } |
1426 | ||
c7ea7630 DL |
1427 | /* |
1428 | * Must be called with "mtx" held. | |
1429 | */ | |
1da177e4 | 1430 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
67347fe4 | 1431 | struct file *tfile, int fd, int full_check) |
1da177e4 LT |
1432 | { |
1433 | int error, revents, pwake = 0; | |
1434 | unsigned long flags; | |
52bd19f7 | 1435 | long user_watches; |
1da177e4 LT |
1436 | struct epitem *epi; |
1437 | struct ep_pqueue epq; | |
1438 | ||
52bd19f7 RH |
1439 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
1440 | if (unlikely(user_watches >= max_user_watches)) | |
7ef9964e | 1441 | return -ENOSPC; |
e94b1766 | 1442 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 1443 | return -ENOMEM; |
1da177e4 LT |
1444 | |
1445 | /* Item initialization follow here ... */ | |
1da177e4 LT |
1446 | INIT_LIST_HEAD(&epi->rdllink); |
1447 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
1448 | INIT_LIST_HEAD(&epi->pwqlist); |
1449 | epi->ep = ep; | |
b030a4dd | 1450 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1451 | epi->event = *event; |
1da177e4 | 1452 | epi->nwait = 0; |
d47de16c | 1453 | epi->next = EP_UNACTIVE_PTR; |
4d7e30d9 AH |
1454 | if (epi->event.events & EPOLLWAKEUP) { |
1455 | error = ep_create_wakeup_source(epi); | |
1456 | if (error) | |
1457 | goto error_create_wakeup_source; | |
1458 | } else { | |
eea1d585 | 1459 | RCU_INIT_POINTER(epi->ws, NULL); |
4d7e30d9 | 1460 | } |
1da177e4 LT |
1461 | |
1462 | /* Initialize the poll table using the queue callback */ | |
1463 | epq.epi = epi; | |
1464 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
1465 | ||
1466 | /* | |
1467 | * Attach the item to the poll hooks and get current event bits. | |
1468 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1469 | * been increased by the caller of this function. Note that after |
1470 | * this operation completes, the poll callback can start hitting | |
1471 | * the new item. | |
1da177e4 | 1472 | */ |
450d89ec | 1473 | revents = ep_item_poll(epi, &epq.pt); |
1da177e4 LT |
1474 | |
1475 | /* | |
1476 | * We have to check if something went wrong during the poll wait queue | |
1477 | * install process. Namely an allocation for a wait queue failed due | |
1478 | * high memory pressure. | |
1479 | */ | |
7ef9964e | 1480 | error = -ENOMEM; |
1da177e4 | 1481 | if (epi->nwait < 0) |
7699acd1 | 1482 | goto error_unregister; |
1da177e4 LT |
1483 | |
1484 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 1485 | spin_lock(&tfile->f_lock); |
ae10b2b4 | 1486 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); |
68499914 | 1487 | spin_unlock(&tfile->f_lock); |
1da177e4 | 1488 | |
c7ea7630 DL |
1489 | /* |
1490 | * Add the current item to the RB tree. All RB tree operations are | |
1491 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1492 | */ | |
1da177e4 LT |
1493 | ep_rbtree_insert(ep, epi); |
1494 | ||
28d82dc1 JB |
1495 | /* now check if we've created too many backpaths */ |
1496 | error = -EINVAL; | |
67347fe4 | 1497 | if (full_check && reverse_path_check()) |
28d82dc1 JB |
1498 | goto error_remove_epi; |
1499 | ||
c7ea7630 DL |
1500 | /* We have to drop the new item inside our item list to keep track of it */ |
1501 | spin_lock_irqsave(&ep->lock, flags); | |
1502 | ||
bf3b9f63 SS |
1503 | /* record NAPI ID of new item if present */ |
1504 | ep_set_busy_poll_napi_id(epi); | |
1505 | ||
1da177e4 | 1506 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 1507 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 | 1508 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1509 | ep_pm_stay_awake(epi); |
1da177e4 LT |
1510 | |
1511 | /* Notify waiting tasks that events are available */ | |
1512 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1513 | wake_up_locked(&ep->wq); |
1da177e4 LT |
1514 | if (waitqueue_active(&ep->poll_wait)) |
1515 | pwake++; | |
1516 | } | |
1517 | ||
c7ea7630 | 1518 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1519 | |
52bd19f7 | 1520 | atomic_long_inc(&ep->user->epoll_watches); |
7ef9964e | 1521 | |
1da177e4 LT |
1522 | /* We have to call this outside the lock */ |
1523 | if (pwake) | |
5071f97e | 1524 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1525 | |
1da177e4 LT |
1526 | return 0; |
1527 | ||
28d82dc1 JB |
1528 | error_remove_epi: |
1529 | spin_lock(&tfile->f_lock); | |
ae10b2b4 | 1530 | list_del_rcu(&epi->fllink); |
28d82dc1 JB |
1531 | spin_unlock(&tfile->f_lock); |
1532 | ||
1533 | rb_erase(&epi->rbn, &ep->rbr); | |
1534 | ||
7699acd1 | 1535 | error_unregister: |
1da177e4 LT |
1536 | ep_unregister_pollwait(ep, epi); |
1537 | ||
1538 | /* | |
1539 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
1540 | * allocated wait queue. Note that we don't care about the ep->ovflist |
1541 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
1542 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 1543 | */ |
c7ea7630 | 1544 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 1545 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 1546 | list_del_init(&epi->rdllink); |
c7ea7630 | 1547 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1548 | |
eea1d585 | 1549 | wakeup_source_unregister(ep_wakeup_source(epi)); |
4d7e30d9 AH |
1550 | |
1551 | error_create_wakeup_source: | |
b030a4dd | 1552 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1553 | |
1da177e4 LT |
1554 | return error; |
1555 | } | |
1556 | ||
1da177e4 LT |
1557 | /* |
1558 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1559 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
1560 | */ |
1561 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1562 | { | |
1563 | int pwake = 0; | |
1564 | unsigned int revents; | |
626cf236 HV |
1565 | poll_table pt; |
1566 | ||
1567 | init_poll_funcptr(&pt, NULL); | |
1da177e4 LT |
1568 | |
1569 | /* | |
e057e15f TB |
1570 | * Set the new event interest mask before calling f_op->poll(); |
1571 | * otherwise we might miss an event that happens between the | |
1572 | * f_op->poll() call and the new event set registering. | |
1da177e4 | 1573 | */ |
128dd175 | 1574 | epi->event.events = event->events; /* need barrier below */ |
e057e15f | 1575 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 | 1576 | if (epi->event.events & EPOLLWAKEUP) { |
eea1d585 | 1577 | if (!ep_has_wakeup_source(epi)) |
4d7e30d9 | 1578 | ep_create_wakeup_source(epi); |
eea1d585 | 1579 | } else if (ep_has_wakeup_source(epi)) { |
4d7e30d9 AH |
1580 | ep_destroy_wakeup_source(epi); |
1581 | } | |
1da177e4 | 1582 | |
128dd175 EW |
1583 | /* |
1584 | * The following barrier has two effects: | |
1585 | * | |
1586 | * 1) Flush epi changes above to other CPUs. This ensures | |
1587 | * we do not miss events from ep_poll_callback if an | |
1588 | * event occurs immediately after we call f_op->poll(). | |
1589 | * We need this because we did not take ep->lock while | |
1590 | * changing epi above (but ep_poll_callback does take | |
1591 | * ep->lock). | |
1592 | * | |
1593 | * 2) We also need to ensure we do not miss _past_ events | |
1594 | * when calling f_op->poll(). This barrier also | |
1595 | * pairs with the barrier in wq_has_sleeper (see | |
1596 | * comments for wq_has_sleeper). | |
1597 | * | |
1598 | * This barrier will now guarantee ep_poll_callback or f_op->poll | |
1599 | * (or both) will notice the readiness of an item. | |
1600 | */ | |
1601 | smp_mb(); | |
1602 | ||
1da177e4 LT |
1603 | /* |
1604 | * Get current event bits. We can safely use the file* here because | |
1605 | * its usage count has been increased by the caller of this function. | |
1606 | */ | |
450d89ec | 1607 | revents = ep_item_poll(epi, &pt); |
1da177e4 | 1608 | |
1da177e4 | 1609 | /* |
c7ea7630 | 1610 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1611 | * list, push it inside. |
1da177e4 | 1612 | */ |
c7ea7630 | 1613 | if (revents & event->events) { |
e057e15f | 1614 | spin_lock_irq(&ep->lock); |
c7ea7630 DL |
1615 | if (!ep_is_linked(&epi->rdllink)) { |
1616 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 1617 | ep_pm_stay_awake(epi); |
c7ea7630 DL |
1618 | |
1619 | /* Notify waiting tasks that events are available */ | |
1620 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1621 | wake_up_locked(&ep->wq); |
c7ea7630 DL |
1622 | if (waitqueue_active(&ep->poll_wait)) |
1623 | pwake++; | |
7699acd1 | 1624 | } |
e057e15f | 1625 | spin_unlock_irq(&ep->lock); |
7699acd1 | 1626 | } |
1da177e4 | 1627 | |
7699acd1 DL |
1628 | /* We have to call this outside the lock */ |
1629 | if (pwake) | |
5071f97e | 1630 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1631 | |
7699acd1 | 1632 | return 0; |
1da177e4 LT |
1633 | } |
1634 | ||
296e236e DL |
1635 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
1636 | void *priv) | |
1da177e4 | 1637 | { |
5071f97e DL |
1638 | struct ep_send_events_data *esed = priv; |
1639 | int eventcnt; | |
296e236e | 1640 | unsigned int revents; |
5071f97e DL |
1641 | struct epitem *epi; |
1642 | struct epoll_event __user *uevent; | |
eea1d585 | 1643 | struct wakeup_source *ws; |
626cf236 HV |
1644 | poll_table pt; |
1645 | ||
1646 | init_poll_funcptr(&pt, NULL); | |
1da177e4 | 1647 | |
296e236e | 1648 | /* |
5071f97e DL |
1649 | * We can loop without lock because we are passed a task private list. |
1650 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1651 | * holding "mtx" during this call. | |
296e236e | 1652 | */ |
5071f97e DL |
1653 | for (eventcnt = 0, uevent = esed->events; |
1654 | !list_empty(head) && eventcnt < esed->maxevents;) { | |
1655 | epi = list_first_entry(head, struct epitem, rdllink); | |
d47de16c | 1656 | |
4d7e30d9 AH |
1657 | /* |
1658 | * Activate ep->ws before deactivating epi->ws to prevent | |
1659 | * triggering auto-suspend here (in case we reactive epi->ws | |
1660 | * below). | |
1661 | * | |
1662 | * This could be rearranged to delay the deactivation of epi->ws | |
1663 | * instead, but then epi->ws would temporarily be out of sync | |
1664 | * with ep_is_linked(). | |
1665 | */ | |
eea1d585 EW |
1666 | ws = ep_wakeup_source(epi); |
1667 | if (ws) { | |
1668 | if (ws->active) | |
1669 | __pm_stay_awake(ep->ws); | |
1670 | __pm_relax(ws); | |
1671 | } | |
1672 | ||
d47de16c | 1673 | list_del_init(&epi->rdllink); |
1da177e4 | 1674 | |
450d89ec | 1675 | revents = ep_item_poll(epi, &pt); |
5071f97e | 1676 | |
296e236e | 1677 | /* |
5071f97e DL |
1678 | * If the event mask intersect the caller-requested one, |
1679 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
1680 | * is holding "mtx", so no operations coming from userspace | |
1681 | * can change the item. | |
296e236e DL |
1682 | */ |
1683 | if (revents) { | |
5071f97e | 1684 | if (__put_user(revents, &uevent->events) || |
d0305882 TB |
1685 | __put_user(epi->event.data, &uevent->data)) { |
1686 | list_add(&epi->rdllink, head); | |
eea1d585 | 1687 | ep_pm_stay_awake(epi); |
296e236e | 1688 | return eventcnt ? eventcnt : -EFAULT; |
d0305882 | 1689 | } |
296e236e | 1690 | eventcnt++; |
5071f97e | 1691 | uevent++; |
296e236e DL |
1692 | if (epi->event.events & EPOLLONESHOT) |
1693 | epi->event.events &= EP_PRIVATE_BITS; | |
1694 | else if (!(epi->event.events & EPOLLET)) { | |
1695 | /* | |
1696 | * If this file has been added with Level | |
1697 | * Trigger mode, we need to insert back inside | |
1698 | * the ready list, so that the next call to | |
1699 | * epoll_wait() will check again the events | |
25985edc | 1700 | * availability. At this point, no one can insert |
296e236e DL |
1701 | * into ep->rdllist besides us. The epoll_ctl() |
1702 | * callers are locked out by | |
1703 | * ep_scan_ready_list() holding "mtx" and the | |
1704 | * poll callback will queue them in ep->ovflist. | |
1705 | */ | |
1706 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 1707 | ep_pm_stay_awake(epi); |
296e236e DL |
1708 | } |
1709 | } | |
1710 | } | |
5071f97e DL |
1711 | |
1712 | return eventcnt; | |
1713 | } | |
d47de16c | 1714 | |
296e236e DL |
1715 | static int ep_send_events(struct eventpoll *ep, |
1716 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1717 | { |
1718 | struct ep_send_events_data esed; | |
1da177e4 | 1719 | |
5071f97e DL |
1720 | esed.maxevents = maxevents; |
1721 | esed.events = events; | |
6192bd53 | 1722 | |
67347fe4 | 1723 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false); |
1da177e4 LT |
1724 | } |
1725 | ||
766b9f92 | 1726 | static inline struct timespec64 ep_set_mstimeout(long ms) |
0781b909 | 1727 | { |
766b9f92 | 1728 | struct timespec64 now, ts = { |
0781b909 ED |
1729 | .tv_sec = ms / MSEC_PER_SEC, |
1730 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), | |
1731 | }; | |
1732 | ||
766b9f92 DD |
1733 | ktime_get_ts64(&now); |
1734 | return timespec64_add_safe(now, ts); | |
0781b909 ED |
1735 | } |
1736 | ||
f4d93ad7 SB |
1737 | /** |
1738 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied | |
1739 | * event buffer. | |
1740 | * | |
1741 | * @ep: Pointer to the eventpoll context. | |
1742 | * @events: Pointer to the userspace buffer where the ready events should be | |
1743 | * stored. | |
1744 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1745 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
1746 | * milliseconds. If the @timeout is zero, the function will not block, | |
1747 | * while if the @timeout is less than zero, the function will block | |
1748 | * until at least one event has been retrieved (or an error | |
1749 | * occurred). | |
1750 | * | |
1751 | * Returns: Returns the number of ready events which have been fetched, or an | |
1752 | * error code, in case of error. | |
1753 | */ | |
1da177e4 LT |
1754 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1755 | int maxevents, long timeout) | |
1756 | { | |
f4d93ad7 | 1757 | int res = 0, eavail, timed_out = 0; |
1da177e4 | 1758 | unsigned long flags; |
da8b44d5 | 1759 | u64 slack = 0; |
ac6424b9 | 1760 | wait_queue_entry_t wait; |
95aac7b1 SB |
1761 | ktime_t expires, *to = NULL; |
1762 | ||
1763 | if (timeout > 0) { | |
766b9f92 | 1764 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
0781b909 | 1765 | |
95aac7b1 SB |
1766 | slack = select_estimate_accuracy(&end_time); |
1767 | to = &expires; | |
766b9f92 | 1768 | *to = timespec64_to_ktime(end_time); |
95aac7b1 | 1769 | } else if (timeout == 0) { |
f4d93ad7 SB |
1770 | /* |
1771 | * Avoid the unnecessary trip to the wait queue loop, if the | |
1772 | * caller specified a non blocking operation. | |
1773 | */ | |
95aac7b1 | 1774 | timed_out = 1; |
f4d93ad7 SB |
1775 | spin_lock_irqsave(&ep->lock, flags); |
1776 | goto check_events; | |
95aac7b1 | 1777 | } |
1da177e4 | 1778 | |
f4d93ad7 | 1779 | fetch_events: |
bf3b9f63 SS |
1780 | |
1781 | if (!ep_events_available(ep)) | |
1782 | ep_busy_loop(ep, timed_out); | |
1783 | ||
c7ea7630 | 1784 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1785 | |
3fb0e584 | 1786 | if (!ep_events_available(ep)) { |
bf3b9f63 SS |
1787 | /* |
1788 | * Busy poll timed out. Drop NAPI ID for now, we can add | |
1789 | * it back in when we have moved a socket with a valid NAPI | |
1790 | * ID onto the ready list. | |
1791 | */ | |
1792 | ep_reset_busy_poll_napi_id(ep); | |
1793 | ||
1da177e4 LT |
1794 | /* |
1795 | * We don't have any available event to return to the caller. | |
1796 | * We need to sleep here, and we will be wake up by | |
1797 | * ep_poll_callback() when events will become available. | |
1798 | */ | |
1799 | init_waitqueue_entry(&wait, current); | |
a93d2f17 | 1800 | __add_wait_queue_exclusive(&ep->wq, &wait); |
1da177e4 LT |
1801 | |
1802 | for (;;) { | |
1803 | /* | |
1804 | * We don't want to sleep if the ep_poll_callback() sends us | |
1805 | * a wakeup in between. That's why we set the task state | |
1806 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1807 | */ | |
1808 | set_current_state(TASK_INTERRUPTIBLE); | |
c257a340 DR |
1809 | /* |
1810 | * Always short-circuit for fatal signals to allow | |
1811 | * threads to make a timely exit without the chance of | |
1812 | * finding more events available and fetching | |
1813 | * repeatedly. | |
1814 | */ | |
1815 | if (fatal_signal_pending(current)) { | |
1816 | res = -EINTR; | |
1817 | break; | |
1818 | } | |
3fb0e584 | 1819 | if (ep_events_available(ep) || timed_out) |
1da177e4 LT |
1820 | break; |
1821 | if (signal_pending(current)) { | |
1822 | res = -EINTR; | |
1823 | break; | |
1824 | } | |
1825 | ||
c7ea7630 | 1826 | spin_unlock_irqrestore(&ep->lock, flags); |
c511851d | 1827 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) |
95aac7b1 SB |
1828 | timed_out = 1; |
1829 | ||
c7ea7630 | 1830 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1831 | } |
1da177e4 | 1832 | |
4d5755b1 DB |
1833 | __remove_wait_queue(&ep->wq, &wait); |
1834 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 1835 | } |
f4d93ad7 | 1836 | check_events: |
1da177e4 | 1837 | /* Is it worth to try to dig for events ? */ |
3fb0e584 | 1838 | eavail = ep_events_available(ep); |
1da177e4 | 1839 | |
c7ea7630 | 1840 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1841 | |
1842 | /* | |
1843 | * Try to transfer events to user space. In case we get 0 events and | |
1844 | * there's still timeout left over, we go trying again in search of | |
1845 | * more luck. | |
1846 | */ | |
1847 | if (!res && eavail && | |
95aac7b1 | 1848 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
f4d93ad7 | 1849 | goto fetch_events; |
1da177e4 LT |
1850 | |
1851 | return res; | |
1852 | } | |
1853 | ||
22bacca4 DL |
1854 | /** |
1855 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() | |
1856 | * API, to verify that adding an epoll file inside another | |
1857 | * epoll structure, does not violate the constraints, in | |
1858 | * terms of closed loops, or too deep chains (which can | |
1859 | * result in excessive stack usage). | |
1860 | * | |
1861 | * @priv: Pointer to the epoll file to be currently checked. | |
1862 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll | |
1863 | * data structure pointer. | |
1864 | * @call_nests: Current dept of the @ep_call_nested() call stack. | |
1865 | * | |
1866 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1867 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1868 | */ | |
1869 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) | |
1870 | { | |
1871 | int error = 0; | |
1872 | struct file *file = priv; | |
1873 | struct eventpoll *ep = file->private_data; | |
28d82dc1 | 1874 | struct eventpoll *ep_tovisit; |
22bacca4 DL |
1875 | struct rb_node *rbp; |
1876 | struct epitem *epi; | |
1877 | ||
d8805e63 | 1878 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
28d82dc1 JB |
1879 | ep->visited = 1; |
1880 | list_add(&ep->visited_list_link, &visited_list); | |
22bacca4 DL |
1881 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
1882 | epi = rb_entry(rbp, struct epitem, rbn); | |
1883 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
28d82dc1 JB |
1884 | ep_tovisit = epi->ffd.file->private_data; |
1885 | if (ep_tovisit->visited) | |
1886 | continue; | |
22bacca4 | 1887 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
28d82dc1 JB |
1888 | ep_loop_check_proc, epi->ffd.file, |
1889 | ep_tovisit, current); | |
22bacca4 DL |
1890 | if (error != 0) |
1891 | break; | |
28d82dc1 JB |
1892 | } else { |
1893 | /* | |
1894 | * If we've reached a file that is not associated with | |
1895 | * an ep, then we need to check if the newly added | |
1896 | * links are going to add too many wakeup paths. We do | |
1897 | * this by adding it to the tfile_check_list, if it's | |
1898 | * not already there, and calling reverse_path_check() | |
1899 | * during ep_insert(). | |
1900 | */ | |
1901 | if (list_empty(&epi->ffd.file->f_tfile_llink)) | |
1902 | list_add(&epi->ffd.file->f_tfile_llink, | |
1903 | &tfile_check_list); | |
22bacca4 DL |
1904 | } |
1905 | } | |
1906 | mutex_unlock(&ep->mtx); | |
1907 | ||
1908 | return error; | |
1909 | } | |
1910 | ||
1911 | /** | |
1912 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) | |
1913 | * another epoll file (represented by @ep) does not create | |
1914 | * closed loops or too deep chains. | |
1915 | * | |
1916 | * @ep: Pointer to the epoll private data structure. | |
1917 | * @file: Pointer to the epoll file to be checked. | |
1918 | * | |
1919 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1920 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1921 | */ | |
1922 | static int ep_loop_check(struct eventpoll *ep, struct file *file) | |
1923 | { | |
28d82dc1 JB |
1924 | int ret; |
1925 | struct eventpoll *ep_cur, *ep_next; | |
1926 | ||
1927 | ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
22bacca4 | 1928 | ep_loop_check_proc, file, ep, current); |
28d82dc1 JB |
1929 | /* clear visited list */ |
1930 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, | |
1931 | visited_list_link) { | |
1932 | ep_cur->visited = 0; | |
1933 | list_del(&ep_cur->visited_list_link); | |
1934 | } | |
1935 | return ret; | |
1936 | } | |
1937 | ||
1938 | static void clear_tfile_check_list(void) | |
1939 | { | |
1940 | struct file *file; | |
1941 | ||
1942 | /* first clear the tfile_check_list */ | |
1943 | while (!list_empty(&tfile_check_list)) { | |
1944 | file = list_first_entry(&tfile_check_list, struct file, | |
1945 | f_tfile_llink); | |
1946 | list_del_init(&file->f_tfile_llink); | |
1947 | } | |
1948 | INIT_LIST_HEAD(&tfile_check_list); | |
22bacca4 DL |
1949 | } |
1950 | ||
7699acd1 | 1951 | /* |
523723bb | 1952 | * Open an eventpoll file descriptor. |
7699acd1 | 1953 | */ |
5a8a82b1 | 1954 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
7699acd1 | 1955 | { |
28d82dc1 | 1956 | int error, fd; |
bb57c3ed | 1957 | struct eventpoll *ep = NULL; |
28d82dc1 | 1958 | struct file *file; |
7699acd1 | 1959 | |
e38b36f3 UD |
1960 | /* Check the EPOLL_* constant for consistency. */ |
1961 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1962 | ||
296e236e DL |
1963 | if (flags & ~EPOLL_CLOEXEC) |
1964 | return -EINVAL; | |
7699acd1 | 1965 | /* |
bb57c3ed | 1966 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 1967 | */ |
9fe5ad9c | 1968 | error = ep_alloc(&ep); |
bb57c3ed DL |
1969 | if (error < 0) |
1970 | return error; | |
7699acd1 DL |
1971 | /* |
1972 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 1973 | * a file structure and a free file descriptor. |
7699acd1 | 1974 | */ |
28d82dc1 JB |
1975 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
1976 | if (fd < 0) { | |
1977 | error = fd; | |
1978 | goto out_free_ep; | |
1979 | } | |
1980 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 1981 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
1982 | if (IS_ERR(file)) { |
1983 | error = PTR_ERR(file); | |
1984 | goto out_free_fd; | |
1985 | } | |
28d82dc1 | 1986 | ep->file = file; |
98022748 | 1987 | fd_install(fd, file); |
28d82dc1 JB |
1988 | return fd; |
1989 | ||
1990 | out_free_fd: | |
1991 | put_unused_fd(fd); | |
1992 | out_free_ep: | |
1993 | ep_free(ep); | |
bb57c3ed | 1994 | return error; |
7699acd1 DL |
1995 | } |
1996 | ||
5a8a82b1 | 1997 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 1998 | { |
bfe3891a | 1999 | if (size <= 0) |
9fe5ad9c UD |
2000 | return -EINVAL; |
2001 | ||
2002 | return sys_epoll_create1(0); | |
a0998b50 UD |
2003 | } |
2004 | ||
7699acd1 DL |
2005 | /* |
2006 | * The following function implements the controller interface for | |
2007 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 2008 | * file descriptors inside the interest set. |
7699acd1 | 2009 | */ |
5a8a82b1 HC |
2010 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
2011 | struct epoll_event __user *, event) | |
7699acd1 DL |
2012 | { |
2013 | int error; | |
67347fe4 | 2014 | int full_check = 0; |
7e3fb584 | 2015 | struct fd f, tf; |
7699acd1 DL |
2016 | struct eventpoll *ep; |
2017 | struct epitem *epi; | |
2018 | struct epoll_event epds; | |
67347fe4 | 2019 | struct eventpoll *tep = NULL; |
7699acd1 | 2020 | |
7699acd1 DL |
2021 | error = -EFAULT; |
2022 | if (ep_op_has_event(op) && | |
2023 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
2024 | goto error_return; | |
2025 | ||
7699acd1 | 2026 | error = -EBADF; |
7e3fb584 AV |
2027 | f = fdget(epfd); |
2028 | if (!f.file) | |
7699acd1 DL |
2029 | goto error_return; |
2030 | ||
2031 | /* Get the "struct file *" for the target file */ | |
7e3fb584 AV |
2032 | tf = fdget(fd); |
2033 | if (!tf.file) | |
7699acd1 DL |
2034 | goto error_fput; |
2035 | ||
2036 | /* The target file descriptor must support poll */ | |
2037 | error = -EPERM; | |
72c2d531 | 2038 | if (!tf.file->f_op->poll) |
7699acd1 DL |
2039 | goto error_tgt_fput; |
2040 | ||
4d7e30d9 | 2041 | /* Check if EPOLLWAKEUP is allowed */ |
c680e41b NI |
2042 | if (ep_op_has_event(op)) |
2043 | ep_take_care_of_epollwakeup(&epds); | |
4d7e30d9 | 2044 | |
7699acd1 DL |
2045 | /* |
2046 | * We have to check that the file structure underneath the file descriptor | |
2047 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
2048 | * adding an epoll file descriptor inside itself. | |
2049 | */ | |
2050 | error = -EINVAL; | |
7e3fb584 | 2051 | if (f.file == tf.file || !is_file_epoll(f.file)) |
7699acd1 DL |
2052 | goto error_tgt_fput; |
2053 | ||
df0108c5 JB |
2054 | /* |
2055 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, | |
2056 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. | |
2057 | * Also, we do not currently supported nested exclusive wakeups. | |
2058 | */ | |
c857ab64 | 2059 | if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) { |
b6a515c8 JB |
2060 | if (op == EPOLL_CTL_MOD) |
2061 | goto error_tgt_fput; | |
2062 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || | |
2063 | (epds.events & ~EPOLLEXCLUSIVE_OK_BITS))) | |
2064 | goto error_tgt_fput; | |
2065 | } | |
df0108c5 | 2066 | |
7699acd1 DL |
2067 | /* |
2068 | * At this point it is safe to assume that the "private_data" contains | |
2069 | * our own data structure. | |
2070 | */ | |
7e3fb584 | 2071 | ep = f.file->private_data; |
7699acd1 | 2072 | |
22bacca4 DL |
2073 | /* |
2074 | * When we insert an epoll file descriptor, inside another epoll file | |
2075 | * descriptor, there is the change of creating closed loops, which are | |
28d82dc1 JB |
2076 | * better be handled here, than in more critical paths. While we are |
2077 | * checking for loops we also determine the list of files reachable | |
2078 | * and hang them on the tfile_check_list, so we can check that we | |
2079 | * haven't created too many possible wakeup paths. | |
22bacca4 | 2080 | * |
67347fe4 JB |
2081 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
2082 | * the epoll file descriptor is attaching directly to a wakeup source, | |
2083 | * unless the epoll file descriptor is nested. The purpose of taking the | |
2084 | * 'epmutex' on add is to prevent complex toplogies such as loops and | |
2085 | * deep wakeup paths from forming in parallel through multiple | |
2086 | * EPOLL_CTL_ADD operations. | |
22bacca4 | 2087 | */ |
67347fe4 | 2088 | mutex_lock_nested(&ep->mtx, 0); |
28d82dc1 | 2089 | if (op == EPOLL_CTL_ADD) { |
67347fe4 JB |
2090 | if (!list_empty(&f.file->f_ep_links) || |
2091 | is_file_epoll(tf.file)) { | |
2092 | full_check = 1; | |
2093 | mutex_unlock(&ep->mtx); | |
2094 | mutex_lock(&epmutex); | |
2095 | if (is_file_epoll(tf.file)) { | |
2096 | error = -ELOOP; | |
2097 | if (ep_loop_check(ep, tf.file) != 0) { | |
2098 | clear_tfile_check_list(); | |
2099 | goto error_tgt_fput; | |
2100 | } | |
2101 | } else | |
2102 | list_add(&tf.file->f_tfile_llink, | |
2103 | &tfile_check_list); | |
2104 | mutex_lock_nested(&ep->mtx, 0); | |
2105 | if (is_file_epoll(tf.file)) { | |
2106 | tep = tf.file->private_data; | |
2107 | mutex_lock_nested(&tep->mtx, 1); | |
13d51807 | 2108 | } |
67347fe4 JB |
2109 | } |
2110 | } | |
7699acd1 | 2111 | |
67647d0f DL |
2112 | /* |
2113 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
2114 | * above, we can be sure to be able to use the item looked up by | |
2115 | * ep_find() till we release the mutex. | |
2116 | */ | |
7e3fb584 | 2117 | epi = ep_find(ep, tf.file, fd); |
7699acd1 DL |
2118 | |
2119 | error = -EINVAL; | |
2120 | switch (op) { | |
2121 | case EPOLL_CTL_ADD: | |
2122 | if (!epi) { | |
2123 | epds.events |= POLLERR | POLLHUP; | |
67347fe4 | 2124 | error = ep_insert(ep, &epds, tf.file, fd, full_check); |
7699acd1 DL |
2125 | } else |
2126 | error = -EEXIST; | |
67347fe4 JB |
2127 | if (full_check) |
2128 | clear_tfile_check_list(); | |
7699acd1 DL |
2129 | break; |
2130 | case EPOLL_CTL_DEL: | |
2131 | if (epi) | |
2132 | error = ep_remove(ep, epi); | |
2133 | else | |
2134 | error = -ENOENT; | |
2135 | break; | |
2136 | case EPOLL_CTL_MOD: | |
2137 | if (epi) { | |
b6a515c8 JB |
2138 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
2139 | epds.events |= POLLERR | POLLHUP; | |
2140 | error = ep_modify(ep, epi, &epds); | |
2141 | } | |
7699acd1 DL |
2142 | } else |
2143 | error = -ENOENT; | |
2144 | break; | |
2145 | } | |
67347fe4 JB |
2146 | if (tep != NULL) |
2147 | mutex_unlock(&tep->mtx); | |
d47de16c | 2148 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
2149 | |
2150 | error_tgt_fput: | |
67347fe4 | 2151 | if (full_check) |
22bacca4 DL |
2152 | mutex_unlock(&epmutex); |
2153 | ||
7e3fb584 | 2154 | fdput(tf); |
7699acd1 | 2155 | error_fput: |
7e3fb584 | 2156 | fdput(f); |
7699acd1 | 2157 | error_return: |
7699acd1 DL |
2158 | |
2159 | return error; | |
2160 | } | |
2161 | ||
2162 | /* | |
2163 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2164 | * part of the user space epoll_wait(2). | |
2165 | */ | |
5a8a82b1 HC |
2166 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
2167 | int, maxevents, int, timeout) | |
7699acd1 | 2168 | { |
2903ff01 AV |
2169 | int error; |
2170 | struct fd f; | |
7699acd1 DL |
2171 | struct eventpoll *ep; |
2172 | ||
7699acd1 DL |
2173 | /* The maximum number of event must be greater than zero */ |
2174 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
2175 | return -EINVAL; | |
2176 | ||
2177 | /* Verify that the area passed by the user is writeable */ | |
2903ff01 AV |
2178 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) |
2179 | return -EFAULT; | |
7699acd1 DL |
2180 | |
2181 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
2182 | f = fdget(epfd); |
2183 | if (!f.file) | |
2184 | return -EBADF; | |
7699acd1 DL |
2185 | |
2186 | /* | |
2187 | * We have to check that the file structure underneath the fd | |
2188 | * the user passed to us _is_ an eventpoll file. | |
2189 | */ | |
2190 | error = -EINVAL; | |
2903ff01 | 2191 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
2192 | goto error_fput; |
2193 | ||
2194 | /* | |
2195 | * At this point it is safe to assume that the "private_data" contains | |
2196 | * our own data structure. | |
2197 | */ | |
2903ff01 | 2198 | ep = f.file->private_data; |
7699acd1 DL |
2199 | |
2200 | /* Time to fish for events ... */ | |
2201 | error = ep_poll(ep, events, maxevents, timeout); | |
2202 | ||
2203 | error_fput: | |
2903ff01 | 2204 | fdput(f); |
7699acd1 DL |
2205 | return error; |
2206 | } | |
2207 | ||
7699acd1 DL |
2208 | /* |
2209 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2210 | * part of the user space epoll_pwait(2). | |
2211 | */ | |
5a8a82b1 HC |
2212 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
2213 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
2214 | size_t, sigsetsize) | |
7699acd1 DL |
2215 | { |
2216 | int error; | |
2217 | sigset_t ksigmask, sigsaved; | |
2218 | ||
2219 | /* | |
2220 | * If the caller wants a certain signal mask to be set during the wait, | |
2221 | * we apply it here. | |
2222 | */ | |
2223 | if (sigmask) { | |
2224 | if (sigsetsize != sizeof(sigset_t)) | |
2225 | return -EINVAL; | |
2226 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
2227 | return -EFAULT; | |
77d55918 ON |
2228 | sigsaved = current->blocked; |
2229 | set_current_blocked(&ksigmask); | |
7699acd1 DL |
2230 | } |
2231 | ||
2232 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
2233 | ||
2234 | /* | |
2235 | * If we changed the signal mask, we need to restore the original one. | |
2236 | * In case we've got a signal while waiting, we do not restore the | |
2237 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
2238 | * the way back to userspace, before the signal mask is restored. | |
2239 | */ | |
2240 | if (sigmask) { | |
2241 | if (error == -EINTR) { | |
2242 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 2243 | sizeof(sigsaved)); |
4e4c22c7 | 2244 | set_restore_sigmask(); |
7699acd1 | 2245 | } else |
77d55918 | 2246 | set_current_blocked(&sigsaved); |
7699acd1 DL |
2247 | } |
2248 | ||
2249 | return error; | |
2250 | } | |
2251 | ||
35280bd4 AV |
2252 | #ifdef CONFIG_COMPAT |
2253 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, | |
2254 | struct epoll_event __user *, events, | |
2255 | int, maxevents, int, timeout, | |
2256 | const compat_sigset_t __user *, sigmask, | |
2257 | compat_size_t, sigsetsize) | |
2258 | { | |
2259 | long err; | |
2260 | compat_sigset_t csigmask; | |
2261 | sigset_t ksigmask, sigsaved; | |
2262 | ||
2263 | /* | |
2264 | * If the caller wants a certain signal mask to be set during the wait, | |
2265 | * we apply it here. | |
2266 | */ | |
2267 | if (sigmask) { | |
2268 | if (sigsetsize != sizeof(compat_sigset_t)) | |
2269 | return -EINVAL; | |
2270 | if (copy_from_user(&csigmask, sigmask, sizeof(csigmask))) | |
2271 | return -EFAULT; | |
2272 | sigset_from_compat(&ksigmask, &csigmask); | |
77d55918 ON |
2273 | sigsaved = current->blocked; |
2274 | set_current_blocked(&ksigmask); | |
35280bd4 AV |
2275 | } |
2276 | ||
2277 | err = sys_epoll_wait(epfd, events, maxevents, timeout); | |
2278 | ||
2279 | /* | |
2280 | * If we changed the signal mask, we need to restore the original one. | |
2281 | * In case we've got a signal while waiting, we do not restore the | |
2282 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
2283 | * the way back to userspace, before the signal mask is restored. | |
2284 | */ | |
2285 | if (sigmask) { | |
2286 | if (err == -EINTR) { | |
2287 | memcpy(¤t->saved_sigmask, &sigsaved, | |
2288 | sizeof(sigsaved)); | |
2289 | set_restore_sigmask(); | |
2290 | } else | |
77d55918 | 2291 | set_current_blocked(&sigsaved); |
35280bd4 AV |
2292 | } |
2293 | ||
2294 | return err; | |
2295 | } | |
2296 | #endif | |
2297 | ||
1da177e4 LT |
2298 | static int __init eventpoll_init(void) |
2299 | { | |
7ef9964e DL |
2300 | struct sysinfo si; |
2301 | ||
2302 | si_meminfo(&si); | |
9df04e1f DL |
2303 | /* |
2304 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
2305 | */ | |
2306 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 2307 | EP_ITEM_COST; |
52bd19f7 | 2308 | BUG_ON(max_user_watches < 0); |
1da177e4 | 2309 | |
22bacca4 DL |
2310 | /* |
2311 | * Initialize the structure used to perform epoll file descriptor | |
2312 | * inclusion loops checks. | |
2313 | */ | |
2314 | ep_nested_calls_init(&poll_loop_ncalls); | |
2315 | ||
1da177e4 | 2316 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
5071f97e DL |
2317 | ep_nested_calls_init(&poll_safewake_ncalls); |
2318 | ||
2319 | /* Initialize the structure used to perform file's f_op->poll() calls */ | |
2320 | ep_nested_calls_init(&poll_readywalk_ncalls); | |
1da177e4 | 2321 | |
39732ca5 EW |
2322 | /* |
2323 | * We can have many thousands of epitems, so prevent this from | |
2324 | * using an extra cache line on 64-bit (and smaller) CPUs | |
2325 | */ | |
2326 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); | |
2327 | ||
1da177e4 LT |
2328 | /* Allocates slab cache used to allocate "struct epitem" items */ |
2329 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
bb57c3ed | 2330 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 LT |
2331 | |
2332 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
2333 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
bb57c3ed | 2334 | sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL); |
1da177e4 | 2335 | |
1da177e4 | 2336 | return 0; |
1da177e4 | 2337 | } |
cea69241 | 2338 | fs_initcall(eventpoll_init); |