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 */ |
b2ac2ea6 | 208 | struct rb_root_cached 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 to call file's f_op->poll() under the nested calls boundaries */ |
280 | static struct nested_calls poll_readywalk_ncalls; | |
1da177e4 LT |
281 | |
282 | /* Slab cache used to allocate "struct epitem" */ | |
e18b890b | 283 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
284 | |
285 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 286 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 287 | |
28d82dc1 JB |
288 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
289 | static LIST_HEAD(visited_list); | |
290 | ||
291 | /* | |
292 | * List of files with newly added links, where we may need to limit the number | |
293 | * of emanating paths. Protected by the epmutex. | |
294 | */ | |
295 | static LIST_HEAD(tfile_check_list); | |
296 | ||
7ef9964e DL |
297 | #ifdef CONFIG_SYSCTL |
298 | ||
299 | #include <linux/sysctl.h> | |
300 | ||
52bd19f7 RH |
301 | static long zero; |
302 | static long long_max = LONG_MAX; | |
7ef9964e | 303 | |
1f7e0616 | 304 | struct ctl_table epoll_table[] = { |
7ef9964e DL |
305 | { |
306 | .procname = "max_user_watches", | |
307 | .data = &max_user_watches, | |
52bd19f7 | 308 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 309 | .mode = 0644, |
52bd19f7 | 310 | .proc_handler = proc_doulongvec_minmax, |
7ef9964e | 311 | .extra1 = &zero, |
52bd19f7 | 312 | .extra2 = &long_max, |
7ef9964e | 313 | }, |
ab09203e | 314 | { } |
7ef9964e DL |
315 | }; |
316 | #endif /* CONFIG_SYSCTL */ | |
317 | ||
28d82dc1 JB |
318 | static const struct file_operations eventpoll_fops; |
319 | ||
320 | static inline int is_file_epoll(struct file *f) | |
321 | { | |
322 | return f->f_op == &eventpoll_fops; | |
323 | } | |
b030a4dd | 324 | |
67647d0f | 325 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
326 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
327 | struct file *file, int fd) | |
328 | { | |
329 | ffd->file = file; | |
330 | ffd->fd = fd; | |
331 | } | |
332 | ||
67647d0f | 333 | /* Compare RB tree keys */ |
b030a4dd PE |
334 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
335 | struct epoll_filefd *p2) | |
336 | { | |
337 | return (p1->file > p2->file ? +1: | |
338 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
339 | } | |
340 | ||
b030a4dd PE |
341 | /* Tells us if the item is currently linked */ |
342 | static inline int ep_is_linked(struct list_head *p) | |
343 | { | |
344 | return !list_empty(p); | |
345 | } | |
346 | ||
ac6424b9 | 347 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
971316f0 ON |
348 | { |
349 | return container_of(p, struct eppoll_entry, wait); | |
350 | } | |
351 | ||
b030a4dd | 352 | /* Get the "struct epitem" from a wait queue pointer */ |
ac6424b9 | 353 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
b030a4dd PE |
354 | { |
355 | return container_of(p, struct eppoll_entry, wait)->base; | |
356 | } | |
357 | ||
358 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 359 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
360 | { |
361 | return container_of(p, struct ep_pqueue, pt)->epi; | |
362 | } | |
363 | ||
364 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 365 | static inline int ep_op_has_event(int op) |
b030a4dd | 366 | { |
a80a6b85 | 367 | return op != EPOLL_CTL_DEL; |
b030a4dd PE |
368 | } |
369 | ||
1da177e4 | 370 | /* Initialize the poll safe wake up structure */ |
5071f97e | 371 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 372 | { |
5071f97e DL |
373 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
374 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
375 | } |
376 | ||
3fb0e584 DL |
377 | /** |
378 | * ep_events_available - Checks if ready events might be available. | |
379 | * | |
380 | * @ep: Pointer to the eventpoll context. | |
381 | * | |
382 | * Returns: Returns a value different than zero if ready events are available, | |
383 | * or zero otherwise. | |
384 | */ | |
385 | static inline int ep_events_available(struct eventpoll *ep) | |
386 | { | |
387 | return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; | |
388 | } | |
389 | ||
bf3b9f63 SS |
390 | #ifdef CONFIG_NET_RX_BUSY_POLL |
391 | static bool ep_busy_loop_end(void *p, unsigned long start_time) | |
392 | { | |
393 | struct eventpoll *ep = p; | |
394 | ||
395 | return ep_events_available(ep) || busy_loop_timeout(start_time); | |
396 | } | |
397 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
398 | ||
399 | /* | |
400 | * Busy poll if globally on and supporting sockets found && no events, | |
401 | * busy loop will return if need_resched or ep_events_available. | |
402 | * | |
403 | * we must do our busy polling with irqs enabled | |
404 | */ | |
405 | static void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
406 | { | |
407 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
408 | unsigned int napi_id = READ_ONCE(ep->napi_id); | |
409 | ||
410 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) | |
411 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep); | |
412 | #endif | |
413 | } | |
414 | ||
415 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
416 | { | |
417 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
418 | if (ep->napi_id) | |
419 | ep->napi_id = 0; | |
420 | #endif | |
421 | } | |
422 | ||
423 | /* | |
424 | * Set epoll busy poll NAPI ID from sk. | |
425 | */ | |
426 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
427 | { | |
428 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
429 | struct eventpoll *ep; | |
430 | unsigned int napi_id; | |
431 | struct socket *sock; | |
432 | struct sock *sk; | |
433 | int err; | |
434 | ||
435 | if (!net_busy_loop_on()) | |
436 | return; | |
437 | ||
438 | sock = sock_from_file(epi->ffd.file, &err); | |
439 | if (!sock) | |
440 | return; | |
441 | ||
442 | sk = sock->sk; | |
443 | if (!sk) | |
444 | return; | |
445 | ||
446 | napi_id = READ_ONCE(sk->sk_napi_id); | |
447 | ep = epi->ep; | |
448 | ||
449 | /* Non-NAPI IDs can be rejected | |
450 | * or | |
451 | * Nothing to do if we already have this ID | |
452 | */ | |
453 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) | |
454 | return; | |
455 | ||
456 | /* record NAPI ID for use in next busy poll */ | |
457 | ep->napi_id = napi_id; | |
458 | #endif | |
459 | } | |
460 | ||
5071f97e DL |
461 | /** |
462 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
463 | * that the recursion limit is not exceeded, and that | |
464 | * the same nested call (by the meaning of same cookie) is | |
465 | * no re-entered. | |
466 | * | |
467 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
468 | * @max_nests: Maximum number of allowed nesting calls. | |
469 | * @nproc: Nested call core function pointer. | |
470 | * @priv: Opaque data to be passed to the @nproc callback. | |
471 | * @cookie: Cookie to be used to identify this nested call. | |
3fe4a975 | 472 | * @ctx: This instance context. |
5071f97e DL |
473 | * |
474 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
475 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 476 | */ |
5071f97e DL |
477 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
478 | int (*nproc)(void *, void *, int), void *priv, | |
3fe4a975 | 479 | void *cookie, void *ctx) |
1da177e4 | 480 | { |
5071f97e | 481 | int error, call_nests = 0; |
1da177e4 | 482 | unsigned long flags; |
5071f97e DL |
483 | struct list_head *lsthead = &ncalls->tasks_call_list; |
484 | struct nested_call_node *tncur; | |
485 | struct nested_call_node tnode; | |
1da177e4 | 486 | |
5071f97e | 487 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 488 | |
5071f97e DL |
489 | /* |
490 | * Try to see if the current task is already inside this wakeup call. | |
491 | * We use a list here, since the population inside this set is always | |
492 | * very much limited. | |
493 | */ | |
b70c3940 | 494 | list_for_each_entry(tncur, lsthead, llink) { |
3fe4a975 | 495 | if (tncur->ctx == ctx && |
5071f97e | 496 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
1da177e4 LT |
497 | /* |
498 | * Ops ... loop detected or maximum nest level reached. | |
499 | * We abort this wake by breaking the cycle itself. | |
500 | */ | |
abff55ce TB |
501 | error = -1; |
502 | goto out_unlock; | |
1da177e4 LT |
503 | } |
504 | } | |
505 | ||
5071f97e | 506 | /* Add the current task and cookie to the list */ |
3fe4a975 | 507 | tnode.ctx = ctx; |
5071f97e | 508 | tnode.cookie = cookie; |
1da177e4 LT |
509 | list_add(&tnode.llink, lsthead); |
510 | ||
5071f97e | 511 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 512 | |
5071f97e DL |
513 | /* Call the nested function */ |
514 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
515 | |
516 | /* Remove the current task from the list */ | |
5071f97e | 517 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 518 | list_del(&tnode.llink); |
3fe4a975 | 519 | out_unlock: |
5071f97e DL |
520 | spin_unlock_irqrestore(&ncalls->lock, flags); |
521 | ||
522 | return error; | |
523 | } | |
524 | ||
02edc6fc SR |
525 | /* |
526 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
527 | * the use of wait queues used by epoll is done in a very controlled | |
528 | * manner. Wake ups can nest inside each other, but are never done | |
529 | * with the same locking. For example: | |
530 | * | |
531 | * dfd = socket(...); | |
532 | * efd1 = epoll_create(); | |
533 | * efd2 = epoll_create(); | |
534 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
535 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
536 | * | |
537 | * When a packet arrives to the device underneath "dfd", the net code will | |
538 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
539 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
540 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
541 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
542 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
543 | * that ends up in another wake_up(), after having checked about the | |
544 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
545 | * avoid stack blasting. | |
546 | * | |
547 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
548 | * this special case of epoll. | |
549 | */ | |
2dfa4eea | 550 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
57a173bd JB |
551 | |
552 | static struct nested_calls poll_safewake_ncalls; | |
553 | ||
554 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) | |
2dfa4eea DL |
555 | { |
556 | unsigned long flags; | |
57a173bd | 557 | wait_queue_head_t *wqueue = (wait_queue_head_t *)cookie; |
2dfa4eea | 558 | |
57a173bd JB |
559 | spin_lock_irqsave_nested(&wqueue->lock, flags, call_nests + 1); |
560 | wake_up_locked_poll(wqueue, POLLIN); | |
2dfa4eea | 561 | spin_unlock_irqrestore(&wqueue->lock, flags); |
2dfa4eea | 562 | |
5071f97e DL |
563 | return 0; |
564 | } | |
565 | ||
5071f97e DL |
566 | static void ep_poll_safewake(wait_queue_head_t *wq) |
567 | { | |
3fe4a975 DL |
568 | int this_cpu = get_cpu(); |
569 | ||
5071f97e | 570 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
3fe4a975 DL |
571 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
572 | ||
573 | put_cpu(); | |
1da177e4 LT |
574 | } |
575 | ||
57a173bd JB |
576 | #else |
577 | ||
578 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
579 | { | |
580 | wake_up_poll(wq, POLLIN); | |
581 | } | |
582 | ||
583 | #endif | |
584 | ||
971316f0 ON |
585 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
586 | { | |
587 | wait_queue_head_t *whead; | |
588 | ||
589 | rcu_read_lock(); | |
138e4ad6 ON |
590 | /* |
591 | * If it is cleared by POLLFREE, it should be rcu-safe. | |
592 | * If we read NULL we need a barrier paired with | |
593 | * smp_store_release() in ep_poll_callback(), otherwise | |
594 | * we rely on whead->lock. | |
595 | */ | |
596 | whead = smp_load_acquire(&pwq->whead); | |
971316f0 ON |
597 | if (whead) |
598 | remove_wait_queue(whead, &pwq->wait); | |
599 | rcu_read_unlock(); | |
600 | } | |
601 | ||
1da177e4 | 602 | /* |
d1bc90dd TB |
603 | * This function unregisters poll callbacks from the associated file |
604 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
605 | * ep_free). | |
1da177e4 | 606 | */ |
7699acd1 | 607 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 608 | { |
7699acd1 DL |
609 | struct list_head *lsthead = &epi->pwqlist; |
610 | struct eppoll_entry *pwq; | |
1da177e4 | 611 | |
d1bc90dd TB |
612 | while (!list_empty(lsthead)) { |
613 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 614 | |
d1bc90dd | 615 | list_del(&pwq->llink); |
971316f0 | 616 | ep_remove_wait_queue(pwq); |
d1bc90dd | 617 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 618 | } |
1da177e4 LT |
619 | } |
620 | ||
eea1d585 EW |
621 | /* call only when ep->mtx is held */ |
622 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) | |
623 | { | |
624 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); | |
625 | } | |
626 | ||
627 | /* call only when ep->mtx is held */ | |
628 | static inline void ep_pm_stay_awake(struct epitem *epi) | |
629 | { | |
630 | struct wakeup_source *ws = ep_wakeup_source(epi); | |
631 | ||
632 | if (ws) | |
633 | __pm_stay_awake(ws); | |
634 | } | |
635 | ||
636 | static inline bool ep_has_wakeup_source(struct epitem *epi) | |
637 | { | |
638 | return rcu_access_pointer(epi->ws) ? true : false; | |
639 | } | |
640 | ||
641 | /* call when ep->mtx cannot be held (ep_poll_callback) */ | |
642 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) | |
643 | { | |
644 | struct wakeup_source *ws; | |
645 | ||
646 | rcu_read_lock(); | |
647 | ws = rcu_dereference(epi->ws); | |
648 | if (ws) | |
649 | __pm_stay_awake(ws); | |
650 | rcu_read_unlock(); | |
651 | } | |
652 | ||
5071f97e DL |
653 | /** |
654 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
655 | * the scan code, to call f_op->poll(). Also allows for | |
656 | * O(NumReady) performance. | |
657 | * | |
658 | * @ep: Pointer to the epoll private data structure. | |
659 | * @sproc: Pointer to the scan callback. | |
660 | * @priv: Private opaque data passed to the @sproc callback. | |
d8805e63 | 661 | * @depth: The current depth of recursive f_op->poll calls. |
67347fe4 | 662 | * @ep_locked: caller already holds ep->mtx |
5071f97e DL |
663 | * |
664 | * Returns: The same integer error code returned by the @sproc callback. | |
665 | */ | |
666 | static int ep_scan_ready_list(struct eventpoll *ep, | |
667 | int (*sproc)(struct eventpoll *, | |
668 | struct list_head *, void *), | |
67347fe4 | 669 | void *priv, int depth, bool ep_locked) |
5071f97e DL |
670 | { |
671 | int error, pwake = 0; | |
672 | unsigned long flags; | |
673 | struct epitem *epi, *nepi; | |
296e236e | 674 | LIST_HEAD(txlist); |
5071f97e DL |
675 | |
676 | /* | |
677 | * We need to lock this because we could be hit by | |
e057e15f | 678 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e | 679 | */ |
67347fe4 JB |
680 | |
681 | if (!ep_locked) | |
682 | mutex_lock_nested(&ep->mtx, depth); | |
5071f97e DL |
683 | |
684 | /* | |
685 | * Steal the ready list, and re-init the original one to the | |
686 | * empty list. Also, set ep->ovflist to NULL so that events | |
687 | * happening while looping w/out locks, are not lost. We cannot | |
688 | * have the poll callback to queue directly on ep->rdllist, | |
689 | * because we want the "sproc" callback to be able to do it | |
690 | * in a lockless way. | |
691 | */ | |
692 | spin_lock_irqsave(&ep->lock, flags); | |
296e236e | 693 | list_splice_init(&ep->rdllist, &txlist); |
5071f97e DL |
694 | ep->ovflist = NULL; |
695 | spin_unlock_irqrestore(&ep->lock, flags); | |
696 | ||
697 | /* | |
698 | * Now call the callback function. | |
699 | */ | |
700 | error = (*sproc)(ep, &txlist, priv); | |
701 | ||
702 | spin_lock_irqsave(&ep->lock, flags); | |
703 | /* | |
704 | * During the time we spent inside the "sproc" callback, some | |
705 | * other events might have been queued by the poll callback. | |
706 | * We re-insert them inside the main ready-list here. | |
707 | */ | |
708 | for (nepi = ep->ovflist; (epi = nepi) != NULL; | |
709 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { | |
710 | /* | |
711 | * We need to check if the item is already in the list. | |
712 | * During the "sproc" callback execution time, items are | |
713 | * queued into ->ovflist but the "txlist" might already | |
714 | * contain them, and the list_splice() below takes care of them. | |
715 | */ | |
4d7e30d9 | 716 | if (!ep_is_linked(&epi->rdllink)) { |
5071f97e | 717 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 718 | ep_pm_stay_awake(epi); |
4d7e30d9 | 719 | } |
5071f97e DL |
720 | } |
721 | /* | |
722 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
723 | * releasing the lock, events will be queued in the normal way inside | |
724 | * ep->rdllist. | |
725 | */ | |
726 | ep->ovflist = EP_UNACTIVE_PTR; | |
727 | ||
728 | /* | |
729 | * Quickly re-inject items left on "txlist". | |
730 | */ | |
731 | list_splice(&txlist, &ep->rdllist); | |
4d7e30d9 | 732 | __pm_relax(ep->ws); |
5071f97e DL |
733 | |
734 | if (!list_empty(&ep->rdllist)) { | |
735 | /* | |
296e236e DL |
736 | * Wake up (if active) both the eventpoll wait list and |
737 | * the ->poll() wait list (delayed after we release the lock). | |
5071f97e DL |
738 | */ |
739 | if (waitqueue_active(&ep->wq)) | |
740 | wake_up_locked(&ep->wq); | |
741 | if (waitqueue_active(&ep->poll_wait)) | |
742 | pwake++; | |
743 | } | |
744 | spin_unlock_irqrestore(&ep->lock, flags); | |
745 | ||
67347fe4 JB |
746 | if (!ep_locked) |
747 | mutex_unlock(&ep->mtx); | |
5071f97e DL |
748 | |
749 | /* We have to call this outside the lock */ | |
750 | if (pwake) | |
751 | ep_poll_safewake(&ep->poll_wait); | |
752 | ||
753 | return error; | |
754 | } | |
755 | ||
ae10b2b4 JB |
756 | static void epi_rcu_free(struct rcu_head *head) |
757 | { | |
758 | struct epitem *epi = container_of(head, struct epitem, rcu); | |
759 | kmem_cache_free(epi_cache, epi); | |
760 | } | |
761 | ||
7699acd1 DL |
762 | /* |
763 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 764 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
765 | */ |
766 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
767 | { | |
7699acd1 DL |
768 | unsigned long flags; |
769 | struct file *file = epi->ffd.file; | |
1da177e4 LT |
770 | |
771 | /* | |
7699acd1 DL |
772 | * Removes poll wait queue hooks. We _have_ to do this without holding |
773 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
774 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
775 | * queue head lock when unregistering the wait queue. The wakeup callback | |
776 | * will run by holding the wait queue head lock and will call our callback | |
777 | * that will try to get "ep->lock". | |
1da177e4 | 778 | */ |
7699acd1 | 779 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 780 | |
7699acd1 | 781 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 782 | spin_lock(&file->f_lock); |
ae10b2b4 | 783 | list_del_rcu(&epi->fllink); |
68499914 | 784 | spin_unlock(&file->f_lock); |
1da177e4 | 785 | |
b2ac2ea6 | 786 | rb_erase_cached(&epi->rbn, &ep->rbr); |
1da177e4 | 787 | |
c7ea7630 DL |
788 | spin_lock_irqsave(&ep->lock, flags); |
789 | if (ep_is_linked(&epi->rdllink)) | |
790 | list_del_init(&epi->rdllink); | |
791 | spin_unlock_irqrestore(&ep->lock, flags); | |
1da177e4 | 792 | |
eea1d585 | 793 | wakeup_source_unregister(ep_wakeup_source(epi)); |
ae10b2b4 JB |
794 | /* |
795 | * At this point it is safe to free the eventpoll item. Use the union | |
796 | * field epi->rcu, since we are trying to minimize the size of | |
797 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by | |
798 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make | |
799 | * use of the rbn field. | |
800 | */ | |
801 | call_rcu(&epi->rcu, epi_rcu_free); | |
1da177e4 | 802 | |
52bd19f7 | 803 | atomic_long_dec(&ep->user->epoll_watches); |
7ef9964e | 804 | |
c7ea7630 | 805 | return 0; |
1da177e4 LT |
806 | } |
807 | ||
7699acd1 | 808 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 809 | { |
7699acd1 DL |
810 | struct rb_node *rbp; |
811 | struct epitem *epi; | |
1da177e4 | 812 | |
7699acd1 DL |
813 | /* We need to release all tasks waiting for these file */ |
814 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 815 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 816 | |
7699acd1 DL |
817 | /* |
818 | * We need to lock this because we could be hit by | |
819 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 820 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
821 | * is on the way to be removed and no one has references to it |
822 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 823 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
824 | */ |
825 | mutex_lock(&epmutex); | |
1da177e4 LT |
826 | |
827 | /* | |
7699acd1 | 828 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 829 | */ |
b2ac2ea6 | 830 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
7699acd1 DL |
831 | epi = rb_entry(rbp, struct epitem, rbn); |
832 | ||
833 | ep_unregister_pollwait(ep, epi); | |
91cf5ab6 | 834 | cond_resched(); |
7699acd1 | 835 | } |
1da177e4 LT |
836 | |
837 | /* | |
7699acd1 DL |
838 | * Walks through the whole tree by freeing each "struct epitem". At this |
839 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 840 | * holding "epmutex" we can be sure that no file cleanup code will hit |
7699acd1 | 841 | * us during this operation. So we can avoid the lock on "ep->lock". |
ddf676c3 EW |
842 | * We do not need to lock ep->mtx, either, we only do it to prevent |
843 | * a lockdep warning. | |
1da177e4 | 844 | */ |
ddf676c3 | 845 | mutex_lock(&ep->mtx); |
b2ac2ea6 | 846 | while ((rbp = rb_first_cached(&ep->rbr)) != NULL) { |
7699acd1 DL |
847 | epi = rb_entry(rbp, struct epitem, rbn); |
848 | ep_remove(ep, epi); | |
91cf5ab6 | 849 | cond_resched(); |
7699acd1 | 850 | } |
ddf676c3 | 851 | mutex_unlock(&ep->mtx); |
1da177e4 | 852 | |
7699acd1 | 853 | mutex_unlock(&epmutex); |
d47de16c | 854 | mutex_destroy(&ep->mtx); |
7ef9964e | 855 | free_uid(ep->user); |
4d7e30d9 | 856 | wakeup_source_unregister(ep->ws); |
f0ee9aab | 857 | kfree(ep); |
7699acd1 | 858 | } |
1da177e4 | 859 | |
7699acd1 DL |
860 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
861 | { | |
862 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 863 | |
f0ee9aab | 864 | if (ep) |
7699acd1 | 865 | ep_free(ep); |
7699acd1 | 866 | |
7699acd1 | 867 | return 0; |
1da177e4 LT |
868 | } |
869 | ||
450d89ec EW |
870 | static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt) |
871 | { | |
872 | pt->_key = epi->event.events; | |
873 | ||
874 | return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events; | |
875 | } | |
876 | ||
296e236e DL |
877 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
878 | void *priv) | |
5071f97e DL |
879 | { |
880 | struct epitem *epi, *tmp; | |
626cf236 | 881 | poll_table pt; |
5071f97e | 882 | |
626cf236 | 883 | init_poll_funcptr(&pt, NULL); |
450d89ec | 884 | |
5071f97e | 885 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
450d89ec | 886 | if (ep_item_poll(epi, &pt)) |
5071f97e | 887 | return POLLIN | POLLRDNORM; |
296e236e | 888 | else { |
5071f97e DL |
889 | /* |
890 | * Item has been dropped into the ready list by the poll | |
891 | * callback, but it's not actually ready, as far as | |
892 | * caller requested events goes. We can remove it here. | |
893 | */ | |
eea1d585 | 894 | __pm_relax(ep_wakeup_source(epi)); |
5071f97e | 895 | list_del_init(&epi->rdllink); |
296e236e | 896 | } |
5071f97e DL |
897 | } |
898 | ||
899 | return 0; | |
900 | } | |
901 | ||
67347fe4 JB |
902 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
903 | poll_table *pt); | |
904 | ||
905 | struct readyevents_arg { | |
906 | struct eventpoll *ep; | |
907 | bool locked; | |
908 | }; | |
909 | ||
5071f97e DL |
910 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) |
911 | { | |
67347fe4 JB |
912 | struct readyevents_arg *arg = priv; |
913 | ||
914 | return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL, | |
915 | call_nests + 1, arg->locked); | |
5071f97e DL |
916 | } |
917 | ||
7699acd1 DL |
918 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
919 | { | |
5071f97e | 920 | int pollflags; |
7699acd1 | 921 | struct eventpoll *ep = file->private_data; |
67347fe4 JB |
922 | struct readyevents_arg arg; |
923 | ||
924 | /* | |
925 | * During ep_insert() we already hold the ep->mtx for the tfile. | |
926 | * Prevent re-aquisition. | |
927 | */ | |
928 | arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc); | |
929 | arg.ep = ep; | |
1da177e4 | 930 | |
7699acd1 DL |
931 | /* Insert inside our poll wait queue */ |
932 | poll_wait(file, &ep->poll_wait, wait); | |
933 | ||
5071f97e DL |
934 | /* |
935 | * Proceed to find out if wanted events are really available inside | |
936 | * the ready list. This need to be done under ep_call_nested() | |
937 | * supervision, since the call to f_op->poll() done on listed files | |
938 | * could re-enter here. | |
939 | */ | |
940 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, | |
67347fe4 | 941 | ep_poll_readyevents_proc, &arg, ep, current); |
7699acd1 | 942 | |
296e236e | 943 | return pollflags != -1 ? pollflags : 0; |
7699acd1 DL |
944 | } |
945 | ||
138d22b5 | 946 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 947 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
138d22b5 CG |
948 | { |
949 | struct eventpoll *ep = f->private_data; | |
950 | struct rb_node *rbp; | |
138d22b5 CG |
951 | |
952 | mutex_lock(&ep->mtx); | |
b2ac2ea6 | 953 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
138d22b5 | 954 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
77493f04 | 955 | struct inode *inode = file_inode(epi->ffd.file); |
138d22b5 | 956 | |
77493f04 CG |
957 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
958 | " pos:%lli ino:%lx sdev:%x\n", | |
a3816ab0 | 959 | epi->ffd.fd, epi->event.events, |
77493f04 CG |
960 | (long long)epi->event.data, |
961 | (long long)epi->ffd.file->f_pos, | |
962 | inode->i_ino, inode->i_sb->s_dev); | |
a3816ab0 | 963 | if (seq_has_overflowed(m)) |
138d22b5 CG |
964 | break; |
965 | } | |
966 | mutex_unlock(&ep->mtx); | |
138d22b5 CG |
967 | } |
968 | #endif | |
969 | ||
7699acd1 DL |
970 | /* File callbacks that implement the eventpoll file behaviour */ |
971 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
972 | #ifdef CONFIG_PROC_FS |
973 | .show_fdinfo = ep_show_fdinfo, | |
974 | #endif | |
7699acd1 | 975 | .release = ep_eventpoll_release, |
6038f373 AB |
976 | .poll = ep_eventpoll_poll, |
977 | .llseek = noop_llseek, | |
7699acd1 DL |
978 | }; |
979 | ||
b611967d | 980 | /* |
7699acd1 DL |
981 | * This is called from eventpoll_release() to unlink files from the eventpoll |
982 | * interface. We need to have this facility to cleanup correctly files that are | |
983 | * closed without being removed from the eventpoll interface. | |
b611967d | 984 | */ |
7699acd1 | 985 | void eventpoll_release_file(struct file *file) |
b611967d | 986 | { |
7699acd1 | 987 | struct eventpoll *ep; |
ebe06187 | 988 | struct epitem *epi, *next; |
b611967d DL |
989 | |
990 | /* | |
68499914 | 991 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 992 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 993 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 994 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 995 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 996 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 997 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 998 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 999 | * from anywhere but ep_free(). |
68499914 JC |
1000 | * |
1001 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 1002 | */ |
7699acd1 | 1003 | mutex_lock(&epmutex); |
ebe06187 | 1004 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
7699acd1 | 1005 | ep = epi->ep; |
d8805e63 | 1006 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 1007 | ep_remove(ep, epi); |
d47de16c | 1008 | mutex_unlock(&ep->mtx); |
b611967d | 1009 | } |
7699acd1 | 1010 | mutex_unlock(&epmutex); |
b611967d DL |
1011 | } |
1012 | ||
53d2be79 | 1013 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 1014 | { |
7ef9964e DL |
1015 | int error; |
1016 | struct user_struct *user; | |
1017 | struct eventpoll *ep; | |
1da177e4 | 1018 | |
7ef9964e | 1019 | user = get_current_user(); |
7ef9964e DL |
1020 | error = -ENOMEM; |
1021 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
1022 | if (unlikely(!ep)) | |
1023 | goto free_uid; | |
1da177e4 | 1024 | |
c7ea7630 | 1025 | spin_lock_init(&ep->lock); |
d47de16c | 1026 | mutex_init(&ep->mtx); |
1da177e4 LT |
1027 | init_waitqueue_head(&ep->wq); |
1028 | init_waitqueue_head(&ep->poll_wait); | |
1029 | INIT_LIST_HEAD(&ep->rdllist); | |
b2ac2ea6 | 1030 | ep->rbr = RB_ROOT_CACHED; |
d47de16c | 1031 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 1032 | ep->user = user; |
1da177e4 | 1033 | |
53d2be79 | 1034 | *pep = ep; |
1da177e4 | 1035 | |
1da177e4 | 1036 | return 0; |
7ef9964e DL |
1037 | |
1038 | free_uid: | |
1039 | free_uid(user); | |
1040 | return error; | |
1da177e4 LT |
1041 | } |
1042 | ||
1da177e4 | 1043 | /* |
c7ea7630 DL |
1044 | * Search the file inside the eventpoll tree. The RB tree operations |
1045 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
1046 | * "mtx" held. | |
1da177e4 LT |
1047 | */ |
1048 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
1049 | { | |
1050 | int kcmp; | |
1da177e4 LT |
1051 | struct rb_node *rbp; |
1052 | struct epitem *epi, *epir = NULL; | |
1053 | struct epoll_filefd ffd; | |
1054 | ||
b030a4dd | 1055 | ep_set_ffd(&ffd, file, fd); |
b2ac2ea6 | 1056 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
1da177e4 | 1057 | epi = rb_entry(rbp, struct epitem, rbn); |
b030a4dd | 1058 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
1059 | if (kcmp > 0) |
1060 | rbp = rbp->rb_right; | |
1061 | else if (kcmp < 0) | |
1062 | rbp = rbp->rb_left; | |
1063 | else { | |
1da177e4 LT |
1064 | epir = epi; |
1065 | break; | |
1066 | } | |
1067 | } | |
1da177e4 | 1068 | |
1da177e4 LT |
1069 | return epir; |
1070 | } | |
1071 | ||
92ef6da3 | 1072 | #ifdef CONFIG_CHECKPOINT_RESTORE |
0791e364 CG |
1073 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
1074 | { | |
1075 | struct rb_node *rbp; | |
1076 | struct epitem *epi; | |
1077 | ||
b2ac2ea6 | 1078 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
0791e364 CG |
1079 | epi = rb_entry(rbp, struct epitem, rbn); |
1080 | if (epi->ffd.fd == tfd) { | |
1081 | if (toff == 0) | |
1082 | return epi; | |
1083 | else | |
1084 | toff--; | |
1085 | } | |
1086 | cond_resched(); | |
1087 | } | |
1088 | ||
1089 | return NULL; | |
1090 | } | |
1091 | ||
1092 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, | |
1093 | unsigned long toff) | |
1094 | { | |
1095 | struct file *file_raw; | |
1096 | struct eventpoll *ep; | |
1097 | struct epitem *epi; | |
1098 | ||
1099 | if (!is_file_epoll(file)) | |
1100 | return ERR_PTR(-EINVAL); | |
1101 | ||
1102 | ep = file->private_data; | |
1103 | ||
1104 | mutex_lock(&ep->mtx); | |
1105 | epi = ep_find_tfd(ep, tfd, toff); | |
1106 | if (epi) | |
1107 | file_raw = epi->ffd.file; | |
1108 | else | |
1109 | file_raw = ERR_PTR(-ENOENT); | |
1110 | mutex_unlock(&ep->mtx); | |
1111 | ||
1112 | return file_raw; | |
1113 | } | |
92ef6da3 | 1114 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
0791e364 | 1115 | |
1da177e4 | 1116 | /* |
7699acd1 | 1117 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 1118 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 1119 | * have events to report. |
1da177e4 | 1120 | */ |
ac6424b9 | 1121 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 1122 | { |
7699acd1 DL |
1123 | int pwake = 0; |
1124 | unsigned long flags; | |
1125 | struct epitem *epi = ep_item_from_wait(wait); | |
1126 | struct eventpoll *ep = epi->ep; | |
df0108c5 | 1127 | int ewake = 0; |
1da177e4 | 1128 | |
c7ea7630 | 1129 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1130 | |
bf3b9f63 SS |
1131 | ep_set_busy_poll_napi_id(epi); |
1132 | ||
7699acd1 DL |
1133 | /* |
1134 | * If the event mask does not contain any poll(2) event, we consider the | |
1135 | * descriptor to be disabled. This condition is likely the effect of the | |
1136 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1137 | * until the next EPOLL_CTL_MOD will be issued. | |
1138 | */ | |
1139 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
1140 | goto out_unlock; |
1141 | ||
2dfa4eea DL |
1142 | /* |
1143 | * Check the events coming with the callback. At this stage, not | |
1144 | * every device reports the events in the "key" parameter of the | |
1145 | * callback. We need to be able to handle both cases here, hence the | |
1146 | * test for "key" != NULL before the event match test. | |
1147 | */ | |
1148 | if (key && !((unsigned long) key & epi->event.events)) | |
1149 | goto out_unlock; | |
1150 | ||
d47de16c | 1151 | /* |
bf6a41db | 1152 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 1153 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 1154 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
1155 | * chained in ep->ovflist and requeued later on. |
1156 | */ | |
1157 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { | |
1158 | if (epi->next == EP_UNACTIVE_PTR) { | |
1159 | epi->next = ep->ovflist; | |
1160 | ep->ovflist = epi; | |
4d7e30d9 AH |
1161 | if (epi->ws) { |
1162 | /* | |
1163 | * Activate ep->ws since epi->ws may get | |
1164 | * deactivated at any time. | |
1165 | */ | |
1166 | __pm_stay_awake(ep->ws); | |
1167 | } | |
1168 | ||
d47de16c DL |
1169 | } |
1170 | goto out_unlock; | |
1171 | } | |
1da177e4 | 1172 | |
7699acd1 | 1173 | /* If this file is already in the ready list we exit soon */ |
4d7e30d9 | 1174 | if (!ep_is_linked(&epi->rdllink)) { |
5071f97e | 1175 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1176 | ep_pm_stay_awake_rcu(epi); |
4d7e30d9 | 1177 | } |
7699acd1 | 1178 | |
7699acd1 DL |
1179 | /* |
1180 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1181 | * wait list. | |
1182 | */ | |
df0108c5 | 1183 | if (waitqueue_active(&ep->wq)) { |
b6a515c8 JB |
1184 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
1185 | !((unsigned long)key & POLLFREE)) { | |
1186 | switch ((unsigned long)key & EPOLLINOUT_BITS) { | |
1187 | case POLLIN: | |
1188 | if (epi->event.events & POLLIN) | |
1189 | ewake = 1; | |
1190 | break; | |
1191 | case POLLOUT: | |
1192 | if (epi->event.events & POLLOUT) | |
1193 | ewake = 1; | |
1194 | break; | |
1195 | case 0: | |
1196 | ewake = 1; | |
1197 | break; | |
1198 | } | |
1199 | } | |
4a6e9e2c | 1200 | wake_up_locked(&ep->wq); |
df0108c5 | 1201 | } |
7699acd1 DL |
1202 | if (waitqueue_active(&ep->poll_wait)) |
1203 | pwake++; | |
1204 | ||
d47de16c | 1205 | out_unlock: |
c7ea7630 | 1206 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1207 | |
7699acd1 DL |
1208 | /* We have to call this outside the lock */ |
1209 | if (pwake) | |
5071f97e | 1210 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 | 1211 | |
138e4ad6 ON |
1212 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
1213 | ewake = 1; | |
1214 | ||
1215 | if ((unsigned long)key & POLLFREE) { | |
1216 | /* | |
1217 | * If we race with ep_remove_wait_queue() it can miss | |
1218 | * ->whead = NULL and do another remove_wait_queue() after | |
1219 | * us, so we can't use __remove_wait_queue(). | |
1220 | */ | |
1221 | list_del_init(&wait->entry); | |
1222 | /* | |
1223 | * ->whead != NULL protects us from the race with ep_free() | |
1224 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock | |
1225 | * held by the caller. Once we nullify it, nothing protects | |
1226 | * ep/epi or even wait. | |
1227 | */ | |
1228 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); | |
1229 | } | |
df0108c5 | 1230 | |
138e4ad6 | 1231 | return ewake; |
7699acd1 | 1232 | } |
1da177e4 LT |
1233 | |
1234 | /* | |
1235 | * This is the callback that is used to add our wait queue to the | |
1236 | * target file wakeup lists. | |
1237 | */ | |
1238 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1239 | poll_table *pt) | |
1240 | { | |
b030a4dd | 1241 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
1242 | struct eppoll_entry *pwq; |
1243 | ||
e94b1766 | 1244 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
1245 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
1246 | pwq->whead = whead; | |
1247 | pwq->base = epi; | |
df0108c5 JB |
1248 | if (epi->event.events & EPOLLEXCLUSIVE) |
1249 | add_wait_queue_exclusive(whead, &pwq->wait); | |
1250 | else | |
1251 | add_wait_queue(whead, &pwq->wait); | |
1da177e4 LT |
1252 | list_add_tail(&pwq->llink, &epi->pwqlist); |
1253 | epi->nwait++; | |
296e236e | 1254 | } else { |
1da177e4 LT |
1255 | /* We have to signal that an error occurred */ |
1256 | epi->nwait = -1; | |
296e236e | 1257 | } |
1da177e4 LT |
1258 | } |
1259 | ||
1da177e4 LT |
1260 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1261 | { | |
1262 | int kcmp; | |
b2ac2ea6 | 1263 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
1da177e4 | 1264 | struct epitem *epic; |
b2ac2ea6 | 1265 | bool leftmost = true; |
1da177e4 LT |
1266 | |
1267 | while (*p) { | |
1268 | parent = *p; | |
1269 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1270 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
b2ac2ea6 | 1271 | if (kcmp > 0) { |
1da177e4 | 1272 | p = &parent->rb_right; |
b2ac2ea6 DB |
1273 | leftmost = false; |
1274 | } else | |
1da177e4 LT |
1275 | p = &parent->rb_left; |
1276 | } | |
1277 | rb_link_node(&epi->rbn, parent, p); | |
b2ac2ea6 | 1278 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
1da177e4 LT |
1279 | } |
1280 | ||
a80a6b85 AM |
1281 | |
1282 | ||
28d82dc1 JB |
1283 | #define PATH_ARR_SIZE 5 |
1284 | /* | |
1285 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1286 | * from a single file of interest. For example, we allow 1000 paths of length | |
1287 | * 1, to emanate from each file of interest. This essentially represents the | |
1288 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1289 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1290 | * is connected to n file sources. In this case each file source has 1 path | |
1291 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1292 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1293 | * and delete can't add additional paths. Protected by the epmutex. | |
1294 | */ | |
1295 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1296 | static int path_count[PATH_ARR_SIZE]; | |
1297 | ||
1298 | static int path_count_inc(int nests) | |
1299 | { | |
93dc6107 JB |
1300 | /* Allow an arbitrary number of depth 1 paths */ |
1301 | if (nests == 0) | |
1302 | return 0; | |
1303 | ||
28d82dc1 JB |
1304 | if (++path_count[nests] > path_limits[nests]) |
1305 | return -1; | |
1306 | return 0; | |
1307 | } | |
1308 | ||
1309 | static void path_count_init(void) | |
1310 | { | |
1311 | int i; | |
1312 | ||
1313 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1314 | path_count[i] = 0; | |
1315 | } | |
1316 | ||
1317 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) | |
1318 | { | |
1319 | int error = 0; | |
1320 | struct file *file = priv; | |
1321 | struct file *child_file; | |
1322 | struct epitem *epi; | |
1323 | ||
ae10b2b4 JB |
1324 | /* CTL_DEL can remove links here, but that can't increase our count */ |
1325 | rcu_read_lock(); | |
1326 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { | |
28d82dc1 JB |
1327 | child_file = epi->ep->file; |
1328 | if (is_file_epoll(child_file)) { | |
1329 | if (list_empty(&child_file->f_ep_links)) { | |
1330 | if (path_count_inc(call_nests)) { | |
1331 | error = -1; | |
1332 | break; | |
1333 | } | |
1334 | } else { | |
1335 | error = ep_call_nested(&poll_loop_ncalls, | |
1336 | EP_MAX_NESTS, | |
1337 | reverse_path_check_proc, | |
1338 | child_file, child_file, | |
1339 | current); | |
1340 | } | |
1341 | if (error != 0) | |
1342 | break; | |
1343 | } else { | |
1344 | printk(KERN_ERR "reverse_path_check_proc: " | |
1345 | "file is not an ep!\n"); | |
1346 | } | |
1347 | } | |
ae10b2b4 | 1348 | rcu_read_unlock(); |
28d82dc1 JB |
1349 | return error; |
1350 | } | |
1351 | ||
1352 | /** | |
1353 | * reverse_path_check - The tfile_check_list is list of file *, which have | |
1354 | * links that are proposed to be newly added. We need to | |
1355 | * make sure that those added links don't add too many | |
1356 | * paths such that we will spend all our time waking up | |
1357 | * eventpoll objects. | |
1358 | * | |
1359 | * Returns: Returns zero if the proposed links don't create too many paths, | |
1360 | * -1 otherwise. | |
1361 | */ | |
1362 | static int reverse_path_check(void) | |
1363 | { | |
28d82dc1 JB |
1364 | int error = 0; |
1365 | struct file *current_file; | |
1366 | ||
1367 | /* let's call this for all tfiles */ | |
1368 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { | |
28d82dc1 JB |
1369 | path_count_init(); |
1370 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
1371 | reverse_path_check_proc, current_file, | |
1372 | current_file, current); | |
1373 | if (error) | |
1374 | break; | |
1375 | } | |
1376 | return error; | |
1377 | } | |
1378 | ||
4d7e30d9 AH |
1379 | static int ep_create_wakeup_source(struct epitem *epi) |
1380 | { | |
1381 | const char *name; | |
eea1d585 | 1382 | struct wakeup_source *ws; |
4d7e30d9 AH |
1383 | |
1384 | if (!epi->ep->ws) { | |
1385 | epi->ep->ws = wakeup_source_register("eventpoll"); | |
1386 | if (!epi->ep->ws) | |
1387 | return -ENOMEM; | |
1388 | } | |
1389 | ||
1390 | name = epi->ffd.file->f_path.dentry->d_name.name; | |
eea1d585 EW |
1391 | ws = wakeup_source_register(name); |
1392 | ||
1393 | if (!ws) | |
4d7e30d9 | 1394 | return -ENOMEM; |
eea1d585 | 1395 | rcu_assign_pointer(epi->ws, ws); |
4d7e30d9 AH |
1396 | |
1397 | return 0; | |
1398 | } | |
1399 | ||
eea1d585 EW |
1400 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
1401 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) | |
4d7e30d9 | 1402 | { |
eea1d585 EW |
1403 | struct wakeup_source *ws = ep_wakeup_source(epi); |
1404 | ||
d6d67e72 | 1405 | RCU_INIT_POINTER(epi->ws, NULL); |
eea1d585 EW |
1406 | |
1407 | /* | |
1408 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is | |
1409 | * used internally by wakeup_source_remove, too (called by | |
1410 | * wakeup_source_unregister), so we cannot use call_rcu | |
1411 | */ | |
1412 | synchronize_rcu(); | |
1413 | wakeup_source_unregister(ws); | |
4d7e30d9 AH |
1414 | } |
1415 | ||
c7ea7630 DL |
1416 | /* |
1417 | * Must be called with "mtx" held. | |
1418 | */ | |
1da177e4 | 1419 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
67347fe4 | 1420 | struct file *tfile, int fd, int full_check) |
1da177e4 LT |
1421 | { |
1422 | int error, revents, pwake = 0; | |
1423 | unsigned long flags; | |
52bd19f7 | 1424 | long user_watches; |
1da177e4 LT |
1425 | struct epitem *epi; |
1426 | struct ep_pqueue epq; | |
1427 | ||
52bd19f7 RH |
1428 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
1429 | if (unlikely(user_watches >= max_user_watches)) | |
7ef9964e | 1430 | return -ENOSPC; |
e94b1766 | 1431 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 1432 | return -ENOMEM; |
1da177e4 LT |
1433 | |
1434 | /* Item initialization follow here ... */ | |
1da177e4 LT |
1435 | INIT_LIST_HEAD(&epi->rdllink); |
1436 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
1437 | INIT_LIST_HEAD(&epi->pwqlist); |
1438 | epi->ep = ep; | |
b030a4dd | 1439 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1440 | epi->event = *event; |
1da177e4 | 1441 | epi->nwait = 0; |
d47de16c | 1442 | epi->next = EP_UNACTIVE_PTR; |
4d7e30d9 AH |
1443 | if (epi->event.events & EPOLLWAKEUP) { |
1444 | error = ep_create_wakeup_source(epi); | |
1445 | if (error) | |
1446 | goto error_create_wakeup_source; | |
1447 | } else { | |
eea1d585 | 1448 | RCU_INIT_POINTER(epi->ws, NULL); |
4d7e30d9 | 1449 | } |
1da177e4 LT |
1450 | |
1451 | /* Initialize the poll table using the queue callback */ | |
1452 | epq.epi = epi; | |
1453 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
1454 | ||
1455 | /* | |
1456 | * Attach the item to the poll hooks and get current event bits. | |
1457 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1458 | * been increased by the caller of this function. Note that after |
1459 | * this operation completes, the poll callback can start hitting | |
1460 | * the new item. | |
1da177e4 | 1461 | */ |
450d89ec | 1462 | revents = ep_item_poll(epi, &epq.pt); |
1da177e4 LT |
1463 | |
1464 | /* | |
1465 | * We have to check if something went wrong during the poll wait queue | |
1466 | * install process. Namely an allocation for a wait queue failed due | |
1467 | * high memory pressure. | |
1468 | */ | |
7ef9964e | 1469 | error = -ENOMEM; |
1da177e4 | 1470 | if (epi->nwait < 0) |
7699acd1 | 1471 | goto error_unregister; |
1da177e4 LT |
1472 | |
1473 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 1474 | spin_lock(&tfile->f_lock); |
ae10b2b4 | 1475 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); |
68499914 | 1476 | spin_unlock(&tfile->f_lock); |
1da177e4 | 1477 | |
c7ea7630 DL |
1478 | /* |
1479 | * Add the current item to the RB tree. All RB tree operations are | |
1480 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1481 | */ | |
1da177e4 LT |
1482 | ep_rbtree_insert(ep, epi); |
1483 | ||
28d82dc1 JB |
1484 | /* now check if we've created too many backpaths */ |
1485 | error = -EINVAL; | |
67347fe4 | 1486 | if (full_check && reverse_path_check()) |
28d82dc1 JB |
1487 | goto error_remove_epi; |
1488 | ||
c7ea7630 DL |
1489 | /* We have to drop the new item inside our item list to keep track of it */ |
1490 | spin_lock_irqsave(&ep->lock, flags); | |
1491 | ||
bf3b9f63 SS |
1492 | /* record NAPI ID of new item if present */ |
1493 | ep_set_busy_poll_napi_id(epi); | |
1494 | ||
1da177e4 | 1495 | /* If the file is already "ready" we drop it inside the ready list */ |
b030a4dd | 1496 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
1da177e4 | 1497 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1498 | ep_pm_stay_awake(epi); |
1da177e4 LT |
1499 | |
1500 | /* Notify waiting tasks that events are available */ | |
1501 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1502 | wake_up_locked(&ep->wq); |
1da177e4 LT |
1503 | if (waitqueue_active(&ep->poll_wait)) |
1504 | pwake++; | |
1505 | } | |
1506 | ||
c7ea7630 | 1507 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1508 | |
52bd19f7 | 1509 | atomic_long_inc(&ep->user->epoll_watches); |
7ef9964e | 1510 | |
1da177e4 LT |
1511 | /* We have to call this outside the lock */ |
1512 | if (pwake) | |
5071f97e | 1513 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1514 | |
1da177e4 LT |
1515 | return 0; |
1516 | ||
28d82dc1 JB |
1517 | error_remove_epi: |
1518 | spin_lock(&tfile->f_lock); | |
ae10b2b4 | 1519 | list_del_rcu(&epi->fllink); |
28d82dc1 JB |
1520 | spin_unlock(&tfile->f_lock); |
1521 | ||
b2ac2ea6 | 1522 | rb_erase_cached(&epi->rbn, &ep->rbr); |
28d82dc1 | 1523 | |
7699acd1 | 1524 | error_unregister: |
1da177e4 LT |
1525 | ep_unregister_pollwait(ep, epi); |
1526 | ||
1527 | /* | |
1528 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
1529 | * allocated wait queue. Note that we don't care about the ep->ovflist |
1530 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
1531 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 1532 | */ |
c7ea7630 | 1533 | spin_lock_irqsave(&ep->lock, flags); |
b030a4dd | 1534 | if (ep_is_linked(&epi->rdllink)) |
6192bd53 | 1535 | list_del_init(&epi->rdllink); |
c7ea7630 | 1536 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1537 | |
eea1d585 | 1538 | wakeup_source_unregister(ep_wakeup_source(epi)); |
4d7e30d9 AH |
1539 | |
1540 | error_create_wakeup_source: | |
b030a4dd | 1541 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1542 | |
1da177e4 LT |
1543 | return error; |
1544 | } | |
1545 | ||
1da177e4 LT |
1546 | /* |
1547 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1548 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 LT |
1549 | */ |
1550 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1551 | { | |
1552 | int pwake = 0; | |
1553 | unsigned int revents; | |
626cf236 HV |
1554 | poll_table pt; |
1555 | ||
1556 | init_poll_funcptr(&pt, NULL); | |
1da177e4 LT |
1557 | |
1558 | /* | |
e057e15f TB |
1559 | * Set the new event interest mask before calling f_op->poll(); |
1560 | * otherwise we might miss an event that happens between the | |
1561 | * f_op->poll() call and the new event set registering. | |
1da177e4 | 1562 | */ |
128dd175 | 1563 | epi->event.events = event->events; /* need barrier below */ |
e057e15f | 1564 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 | 1565 | if (epi->event.events & EPOLLWAKEUP) { |
eea1d585 | 1566 | if (!ep_has_wakeup_source(epi)) |
4d7e30d9 | 1567 | ep_create_wakeup_source(epi); |
eea1d585 | 1568 | } else if (ep_has_wakeup_source(epi)) { |
4d7e30d9 AH |
1569 | ep_destroy_wakeup_source(epi); |
1570 | } | |
1da177e4 | 1571 | |
128dd175 EW |
1572 | /* |
1573 | * The following barrier has two effects: | |
1574 | * | |
1575 | * 1) Flush epi changes above to other CPUs. This ensures | |
1576 | * we do not miss events from ep_poll_callback if an | |
1577 | * event occurs immediately after we call f_op->poll(). | |
1578 | * We need this because we did not take ep->lock while | |
1579 | * changing epi above (but ep_poll_callback does take | |
1580 | * ep->lock). | |
1581 | * | |
1582 | * 2) We also need to ensure we do not miss _past_ events | |
1583 | * when calling f_op->poll(). This barrier also | |
1584 | * pairs with the barrier in wq_has_sleeper (see | |
1585 | * comments for wq_has_sleeper). | |
1586 | * | |
1587 | * This barrier will now guarantee ep_poll_callback or f_op->poll | |
1588 | * (or both) will notice the readiness of an item. | |
1589 | */ | |
1590 | smp_mb(); | |
1591 | ||
1da177e4 LT |
1592 | /* |
1593 | * Get current event bits. We can safely use the file* here because | |
1594 | * its usage count has been increased by the caller of this function. | |
1595 | */ | |
450d89ec | 1596 | revents = ep_item_poll(epi, &pt); |
1da177e4 | 1597 | |
1da177e4 | 1598 | /* |
c7ea7630 | 1599 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1600 | * list, push it inside. |
1da177e4 | 1601 | */ |
c7ea7630 | 1602 | if (revents & event->events) { |
e057e15f | 1603 | spin_lock_irq(&ep->lock); |
c7ea7630 DL |
1604 | if (!ep_is_linked(&epi->rdllink)) { |
1605 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 1606 | ep_pm_stay_awake(epi); |
c7ea7630 DL |
1607 | |
1608 | /* Notify waiting tasks that events are available */ | |
1609 | if (waitqueue_active(&ep->wq)) | |
4a6e9e2c | 1610 | wake_up_locked(&ep->wq); |
c7ea7630 DL |
1611 | if (waitqueue_active(&ep->poll_wait)) |
1612 | pwake++; | |
7699acd1 | 1613 | } |
e057e15f | 1614 | spin_unlock_irq(&ep->lock); |
7699acd1 | 1615 | } |
1da177e4 | 1616 | |
7699acd1 DL |
1617 | /* We have to call this outside the lock */ |
1618 | if (pwake) | |
5071f97e | 1619 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1620 | |
7699acd1 | 1621 | return 0; |
1da177e4 LT |
1622 | } |
1623 | ||
296e236e DL |
1624 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
1625 | void *priv) | |
1da177e4 | 1626 | { |
5071f97e DL |
1627 | struct ep_send_events_data *esed = priv; |
1628 | int eventcnt; | |
296e236e | 1629 | unsigned int revents; |
5071f97e DL |
1630 | struct epitem *epi; |
1631 | struct epoll_event __user *uevent; | |
eea1d585 | 1632 | struct wakeup_source *ws; |
626cf236 HV |
1633 | poll_table pt; |
1634 | ||
1635 | init_poll_funcptr(&pt, NULL); | |
1da177e4 | 1636 | |
296e236e | 1637 | /* |
5071f97e DL |
1638 | * We can loop without lock because we are passed a task private list. |
1639 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1640 | * holding "mtx" during this call. | |
296e236e | 1641 | */ |
5071f97e DL |
1642 | for (eventcnt = 0, uevent = esed->events; |
1643 | !list_empty(head) && eventcnt < esed->maxevents;) { | |
1644 | epi = list_first_entry(head, struct epitem, rdllink); | |
d47de16c | 1645 | |
4d7e30d9 AH |
1646 | /* |
1647 | * Activate ep->ws before deactivating epi->ws to prevent | |
1648 | * triggering auto-suspend here (in case we reactive epi->ws | |
1649 | * below). | |
1650 | * | |
1651 | * This could be rearranged to delay the deactivation of epi->ws | |
1652 | * instead, but then epi->ws would temporarily be out of sync | |
1653 | * with ep_is_linked(). | |
1654 | */ | |
eea1d585 EW |
1655 | ws = ep_wakeup_source(epi); |
1656 | if (ws) { | |
1657 | if (ws->active) | |
1658 | __pm_stay_awake(ep->ws); | |
1659 | __pm_relax(ws); | |
1660 | } | |
1661 | ||
d47de16c | 1662 | list_del_init(&epi->rdllink); |
1da177e4 | 1663 | |
450d89ec | 1664 | revents = ep_item_poll(epi, &pt); |
5071f97e | 1665 | |
296e236e | 1666 | /* |
5071f97e DL |
1667 | * If the event mask intersect the caller-requested one, |
1668 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
1669 | * is holding "mtx", so no operations coming from userspace | |
1670 | * can change the item. | |
296e236e DL |
1671 | */ |
1672 | if (revents) { | |
5071f97e | 1673 | if (__put_user(revents, &uevent->events) || |
d0305882 TB |
1674 | __put_user(epi->event.data, &uevent->data)) { |
1675 | list_add(&epi->rdllink, head); | |
eea1d585 | 1676 | ep_pm_stay_awake(epi); |
296e236e | 1677 | return eventcnt ? eventcnt : -EFAULT; |
d0305882 | 1678 | } |
296e236e | 1679 | eventcnt++; |
5071f97e | 1680 | uevent++; |
296e236e DL |
1681 | if (epi->event.events & EPOLLONESHOT) |
1682 | epi->event.events &= EP_PRIVATE_BITS; | |
1683 | else if (!(epi->event.events & EPOLLET)) { | |
1684 | /* | |
1685 | * If this file has been added with Level | |
1686 | * Trigger mode, we need to insert back inside | |
1687 | * the ready list, so that the next call to | |
1688 | * epoll_wait() will check again the events | |
25985edc | 1689 | * availability. At this point, no one can insert |
296e236e DL |
1690 | * into ep->rdllist besides us. The epoll_ctl() |
1691 | * callers are locked out by | |
1692 | * ep_scan_ready_list() holding "mtx" and the | |
1693 | * poll callback will queue them in ep->ovflist. | |
1694 | */ | |
1695 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 1696 | ep_pm_stay_awake(epi); |
296e236e DL |
1697 | } |
1698 | } | |
1699 | } | |
5071f97e DL |
1700 | |
1701 | return eventcnt; | |
1702 | } | |
d47de16c | 1703 | |
296e236e DL |
1704 | static int ep_send_events(struct eventpoll *ep, |
1705 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1706 | { |
1707 | struct ep_send_events_data esed; | |
1da177e4 | 1708 | |
5071f97e DL |
1709 | esed.maxevents = maxevents; |
1710 | esed.events = events; | |
6192bd53 | 1711 | |
67347fe4 | 1712 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false); |
1da177e4 LT |
1713 | } |
1714 | ||
766b9f92 | 1715 | static inline struct timespec64 ep_set_mstimeout(long ms) |
0781b909 | 1716 | { |
766b9f92 | 1717 | struct timespec64 now, ts = { |
0781b909 ED |
1718 | .tv_sec = ms / MSEC_PER_SEC, |
1719 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), | |
1720 | }; | |
1721 | ||
766b9f92 DD |
1722 | ktime_get_ts64(&now); |
1723 | return timespec64_add_safe(now, ts); | |
0781b909 ED |
1724 | } |
1725 | ||
f4d93ad7 SB |
1726 | /** |
1727 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied | |
1728 | * event buffer. | |
1729 | * | |
1730 | * @ep: Pointer to the eventpoll context. | |
1731 | * @events: Pointer to the userspace buffer where the ready events should be | |
1732 | * stored. | |
1733 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1734 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
1735 | * milliseconds. If the @timeout is zero, the function will not block, | |
1736 | * while if the @timeout is less than zero, the function will block | |
1737 | * until at least one event has been retrieved (or an error | |
1738 | * occurred). | |
1739 | * | |
1740 | * Returns: Returns the number of ready events which have been fetched, or an | |
1741 | * error code, in case of error. | |
1742 | */ | |
1da177e4 LT |
1743 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1744 | int maxevents, long timeout) | |
1745 | { | |
f4d93ad7 | 1746 | int res = 0, eavail, timed_out = 0; |
1da177e4 | 1747 | unsigned long flags; |
da8b44d5 | 1748 | u64 slack = 0; |
ac6424b9 | 1749 | wait_queue_entry_t wait; |
95aac7b1 SB |
1750 | ktime_t expires, *to = NULL; |
1751 | ||
1752 | if (timeout > 0) { | |
766b9f92 | 1753 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
0781b909 | 1754 | |
95aac7b1 SB |
1755 | slack = select_estimate_accuracy(&end_time); |
1756 | to = &expires; | |
766b9f92 | 1757 | *to = timespec64_to_ktime(end_time); |
95aac7b1 | 1758 | } else if (timeout == 0) { |
f4d93ad7 SB |
1759 | /* |
1760 | * Avoid the unnecessary trip to the wait queue loop, if the | |
1761 | * caller specified a non blocking operation. | |
1762 | */ | |
95aac7b1 | 1763 | timed_out = 1; |
f4d93ad7 SB |
1764 | spin_lock_irqsave(&ep->lock, flags); |
1765 | goto check_events; | |
95aac7b1 | 1766 | } |
1da177e4 | 1767 | |
f4d93ad7 | 1768 | fetch_events: |
bf3b9f63 SS |
1769 | |
1770 | if (!ep_events_available(ep)) | |
1771 | ep_busy_loop(ep, timed_out); | |
1772 | ||
c7ea7630 | 1773 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1774 | |
3fb0e584 | 1775 | if (!ep_events_available(ep)) { |
bf3b9f63 SS |
1776 | /* |
1777 | * Busy poll timed out. Drop NAPI ID for now, we can add | |
1778 | * it back in when we have moved a socket with a valid NAPI | |
1779 | * ID onto the ready list. | |
1780 | */ | |
1781 | ep_reset_busy_poll_napi_id(ep); | |
1782 | ||
1da177e4 LT |
1783 | /* |
1784 | * We don't have any available event to return to the caller. | |
1785 | * We need to sleep here, and we will be wake up by | |
1786 | * ep_poll_callback() when events will become available. | |
1787 | */ | |
1788 | init_waitqueue_entry(&wait, current); | |
a93d2f17 | 1789 | __add_wait_queue_exclusive(&ep->wq, &wait); |
1da177e4 LT |
1790 | |
1791 | for (;;) { | |
1792 | /* | |
1793 | * We don't want to sleep if the ep_poll_callback() sends us | |
1794 | * a wakeup in between. That's why we set the task state | |
1795 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1796 | */ | |
1797 | set_current_state(TASK_INTERRUPTIBLE); | |
c257a340 DR |
1798 | /* |
1799 | * Always short-circuit for fatal signals to allow | |
1800 | * threads to make a timely exit without the chance of | |
1801 | * finding more events available and fetching | |
1802 | * repeatedly. | |
1803 | */ | |
1804 | if (fatal_signal_pending(current)) { | |
1805 | res = -EINTR; | |
1806 | break; | |
1807 | } | |
3fb0e584 | 1808 | if (ep_events_available(ep) || timed_out) |
1da177e4 LT |
1809 | break; |
1810 | if (signal_pending(current)) { | |
1811 | res = -EINTR; | |
1812 | break; | |
1813 | } | |
1814 | ||
c7ea7630 | 1815 | spin_unlock_irqrestore(&ep->lock, flags); |
c511851d | 1816 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) |
95aac7b1 SB |
1817 | timed_out = 1; |
1818 | ||
c7ea7630 | 1819 | spin_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1820 | } |
1da177e4 | 1821 | |
4d5755b1 DB |
1822 | __remove_wait_queue(&ep->wq, &wait); |
1823 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 1824 | } |
f4d93ad7 | 1825 | check_events: |
1da177e4 | 1826 | /* Is it worth to try to dig for events ? */ |
3fb0e584 | 1827 | eavail = ep_events_available(ep); |
1da177e4 | 1828 | |
c7ea7630 | 1829 | spin_unlock_irqrestore(&ep->lock, flags); |
1da177e4 LT |
1830 | |
1831 | /* | |
1832 | * Try to transfer events to user space. In case we get 0 events and | |
1833 | * there's still timeout left over, we go trying again in search of | |
1834 | * more luck. | |
1835 | */ | |
1836 | if (!res && eavail && | |
95aac7b1 | 1837 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
f4d93ad7 | 1838 | goto fetch_events; |
1da177e4 LT |
1839 | |
1840 | return res; | |
1841 | } | |
1842 | ||
22bacca4 DL |
1843 | /** |
1844 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() | |
1845 | * API, to verify that adding an epoll file inside another | |
1846 | * epoll structure, does not violate the constraints, in | |
1847 | * terms of closed loops, or too deep chains (which can | |
1848 | * result in excessive stack usage). | |
1849 | * | |
1850 | * @priv: Pointer to the epoll file to be currently checked. | |
1851 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll | |
1852 | * data structure pointer. | |
1853 | * @call_nests: Current dept of the @ep_call_nested() call stack. | |
1854 | * | |
1855 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1856 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1857 | */ | |
1858 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) | |
1859 | { | |
1860 | int error = 0; | |
1861 | struct file *file = priv; | |
1862 | struct eventpoll *ep = file->private_data; | |
28d82dc1 | 1863 | struct eventpoll *ep_tovisit; |
22bacca4 DL |
1864 | struct rb_node *rbp; |
1865 | struct epitem *epi; | |
1866 | ||
d8805e63 | 1867 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
28d82dc1 JB |
1868 | ep->visited = 1; |
1869 | list_add(&ep->visited_list_link, &visited_list); | |
b2ac2ea6 | 1870 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
22bacca4 DL |
1871 | epi = rb_entry(rbp, struct epitem, rbn); |
1872 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
28d82dc1 JB |
1873 | ep_tovisit = epi->ffd.file->private_data; |
1874 | if (ep_tovisit->visited) | |
1875 | continue; | |
22bacca4 | 1876 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
28d82dc1 JB |
1877 | ep_loop_check_proc, epi->ffd.file, |
1878 | ep_tovisit, current); | |
22bacca4 DL |
1879 | if (error != 0) |
1880 | break; | |
28d82dc1 JB |
1881 | } else { |
1882 | /* | |
1883 | * If we've reached a file that is not associated with | |
1884 | * an ep, then we need to check if the newly added | |
1885 | * links are going to add too many wakeup paths. We do | |
1886 | * this by adding it to the tfile_check_list, if it's | |
1887 | * not already there, and calling reverse_path_check() | |
1888 | * during ep_insert(). | |
1889 | */ | |
1890 | if (list_empty(&epi->ffd.file->f_tfile_llink)) | |
1891 | list_add(&epi->ffd.file->f_tfile_llink, | |
1892 | &tfile_check_list); | |
22bacca4 DL |
1893 | } |
1894 | } | |
1895 | mutex_unlock(&ep->mtx); | |
1896 | ||
1897 | return error; | |
1898 | } | |
1899 | ||
1900 | /** | |
1901 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) | |
1902 | * another epoll file (represented by @ep) does not create | |
1903 | * closed loops or too deep chains. | |
1904 | * | |
1905 | * @ep: Pointer to the epoll private data structure. | |
1906 | * @file: Pointer to the epoll file to be checked. | |
1907 | * | |
1908 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1909 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1910 | */ | |
1911 | static int ep_loop_check(struct eventpoll *ep, struct file *file) | |
1912 | { | |
28d82dc1 JB |
1913 | int ret; |
1914 | struct eventpoll *ep_cur, *ep_next; | |
1915 | ||
1916 | ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, | |
22bacca4 | 1917 | ep_loop_check_proc, file, ep, current); |
28d82dc1 JB |
1918 | /* clear visited list */ |
1919 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, | |
1920 | visited_list_link) { | |
1921 | ep_cur->visited = 0; | |
1922 | list_del(&ep_cur->visited_list_link); | |
1923 | } | |
1924 | return ret; | |
1925 | } | |
1926 | ||
1927 | static void clear_tfile_check_list(void) | |
1928 | { | |
1929 | struct file *file; | |
1930 | ||
1931 | /* first clear the tfile_check_list */ | |
1932 | while (!list_empty(&tfile_check_list)) { | |
1933 | file = list_first_entry(&tfile_check_list, struct file, | |
1934 | f_tfile_llink); | |
1935 | list_del_init(&file->f_tfile_llink); | |
1936 | } | |
1937 | INIT_LIST_HEAD(&tfile_check_list); | |
22bacca4 DL |
1938 | } |
1939 | ||
7699acd1 | 1940 | /* |
523723bb | 1941 | * Open an eventpoll file descriptor. |
7699acd1 | 1942 | */ |
5a8a82b1 | 1943 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
7699acd1 | 1944 | { |
28d82dc1 | 1945 | int error, fd; |
bb57c3ed | 1946 | struct eventpoll *ep = NULL; |
28d82dc1 | 1947 | struct file *file; |
7699acd1 | 1948 | |
e38b36f3 UD |
1949 | /* Check the EPOLL_* constant for consistency. */ |
1950 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1951 | ||
296e236e DL |
1952 | if (flags & ~EPOLL_CLOEXEC) |
1953 | return -EINVAL; | |
7699acd1 | 1954 | /* |
bb57c3ed | 1955 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 1956 | */ |
9fe5ad9c | 1957 | error = ep_alloc(&ep); |
bb57c3ed DL |
1958 | if (error < 0) |
1959 | return error; | |
7699acd1 DL |
1960 | /* |
1961 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 1962 | * a file structure and a free file descriptor. |
7699acd1 | 1963 | */ |
28d82dc1 JB |
1964 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
1965 | if (fd < 0) { | |
1966 | error = fd; | |
1967 | goto out_free_ep; | |
1968 | } | |
1969 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 1970 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
1971 | if (IS_ERR(file)) { |
1972 | error = PTR_ERR(file); | |
1973 | goto out_free_fd; | |
1974 | } | |
28d82dc1 | 1975 | ep->file = file; |
98022748 | 1976 | fd_install(fd, file); |
28d82dc1 JB |
1977 | return fd; |
1978 | ||
1979 | out_free_fd: | |
1980 | put_unused_fd(fd); | |
1981 | out_free_ep: | |
1982 | ep_free(ep); | |
bb57c3ed | 1983 | return error; |
7699acd1 DL |
1984 | } |
1985 | ||
5a8a82b1 | 1986 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 1987 | { |
bfe3891a | 1988 | if (size <= 0) |
9fe5ad9c UD |
1989 | return -EINVAL; |
1990 | ||
1991 | return sys_epoll_create1(0); | |
a0998b50 UD |
1992 | } |
1993 | ||
7699acd1 DL |
1994 | /* |
1995 | * The following function implements the controller interface for | |
1996 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 1997 | * file descriptors inside the interest set. |
7699acd1 | 1998 | */ |
5a8a82b1 HC |
1999 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
2000 | struct epoll_event __user *, event) | |
7699acd1 DL |
2001 | { |
2002 | int error; | |
67347fe4 | 2003 | int full_check = 0; |
7e3fb584 | 2004 | struct fd f, tf; |
7699acd1 DL |
2005 | struct eventpoll *ep; |
2006 | struct epitem *epi; | |
2007 | struct epoll_event epds; | |
67347fe4 | 2008 | struct eventpoll *tep = NULL; |
7699acd1 | 2009 | |
7699acd1 DL |
2010 | error = -EFAULT; |
2011 | if (ep_op_has_event(op) && | |
2012 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
2013 | goto error_return; | |
2014 | ||
7699acd1 | 2015 | error = -EBADF; |
7e3fb584 AV |
2016 | f = fdget(epfd); |
2017 | if (!f.file) | |
7699acd1 DL |
2018 | goto error_return; |
2019 | ||
2020 | /* Get the "struct file *" for the target file */ | |
7e3fb584 AV |
2021 | tf = fdget(fd); |
2022 | if (!tf.file) | |
7699acd1 DL |
2023 | goto error_fput; |
2024 | ||
2025 | /* The target file descriptor must support poll */ | |
2026 | error = -EPERM; | |
72c2d531 | 2027 | if (!tf.file->f_op->poll) |
7699acd1 DL |
2028 | goto error_tgt_fput; |
2029 | ||
4d7e30d9 | 2030 | /* Check if EPOLLWAKEUP is allowed */ |
c680e41b NI |
2031 | if (ep_op_has_event(op)) |
2032 | ep_take_care_of_epollwakeup(&epds); | |
4d7e30d9 | 2033 | |
7699acd1 DL |
2034 | /* |
2035 | * We have to check that the file structure underneath the file descriptor | |
2036 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
2037 | * adding an epoll file descriptor inside itself. | |
2038 | */ | |
2039 | error = -EINVAL; | |
7e3fb584 | 2040 | if (f.file == tf.file || !is_file_epoll(f.file)) |
7699acd1 DL |
2041 | goto error_tgt_fput; |
2042 | ||
df0108c5 JB |
2043 | /* |
2044 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, | |
2045 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. | |
2046 | * Also, we do not currently supported nested exclusive wakeups. | |
2047 | */ | |
c857ab64 | 2048 | if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) { |
b6a515c8 JB |
2049 | if (op == EPOLL_CTL_MOD) |
2050 | goto error_tgt_fput; | |
2051 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || | |
2052 | (epds.events & ~EPOLLEXCLUSIVE_OK_BITS))) | |
2053 | goto error_tgt_fput; | |
2054 | } | |
df0108c5 | 2055 | |
7699acd1 DL |
2056 | /* |
2057 | * At this point it is safe to assume that the "private_data" contains | |
2058 | * our own data structure. | |
2059 | */ | |
7e3fb584 | 2060 | ep = f.file->private_data; |
7699acd1 | 2061 | |
22bacca4 DL |
2062 | /* |
2063 | * When we insert an epoll file descriptor, inside another epoll file | |
2064 | * descriptor, there is the change of creating closed loops, which are | |
28d82dc1 JB |
2065 | * better be handled here, than in more critical paths. While we are |
2066 | * checking for loops we also determine the list of files reachable | |
2067 | * and hang them on the tfile_check_list, so we can check that we | |
2068 | * haven't created too many possible wakeup paths. | |
22bacca4 | 2069 | * |
67347fe4 JB |
2070 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
2071 | * the epoll file descriptor is attaching directly to a wakeup source, | |
2072 | * unless the epoll file descriptor is nested. The purpose of taking the | |
2073 | * 'epmutex' on add is to prevent complex toplogies such as loops and | |
2074 | * deep wakeup paths from forming in parallel through multiple | |
2075 | * EPOLL_CTL_ADD operations. | |
22bacca4 | 2076 | */ |
67347fe4 | 2077 | mutex_lock_nested(&ep->mtx, 0); |
28d82dc1 | 2078 | if (op == EPOLL_CTL_ADD) { |
67347fe4 JB |
2079 | if (!list_empty(&f.file->f_ep_links) || |
2080 | is_file_epoll(tf.file)) { | |
2081 | full_check = 1; | |
2082 | mutex_unlock(&ep->mtx); | |
2083 | mutex_lock(&epmutex); | |
2084 | if (is_file_epoll(tf.file)) { | |
2085 | error = -ELOOP; | |
2086 | if (ep_loop_check(ep, tf.file) != 0) { | |
2087 | clear_tfile_check_list(); | |
2088 | goto error_tgt_fput; | |
2089 | } | |
2090 | } else | |
2091 | list_add(&tf.file->f_tfile_llink, | |
2092 | &tfile_check_list); | |
2093 | mutex_lock_nested(&ep->mtx, 0); | |
2094 | if (is_file_epoll(tf.file)) { | |
2095 | tep = tf.file->private_data; | |
2096 | mutex_lock_nested(&tep->mtx, 1); | |
13d51807 | 2097 | } |
67347fe4 JB |
2098 | } |
2099 | } | |
7699acd1 | 2100 | |
67647d0f DL |
2101 | /* |
2102 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
2103 | * above, we can be sure to be able to use the item looked up by | |
2104 | * ep_find() till we release the mutex. | |
2105 | */ | |
7e3fb584 | 2106 | epi = ep_find(ep, tf.file, fd); |
7699acd1 DL |
2107 | |
2108 | error = -EINVAL; | |
2109 | switch (op) { | |
2110 | case EPOLL_CTL_ADD: | |
2111 | if (!epi) { | |
2112 | epds.events |= POLLERR | POLLHUP; | |
67347fe4 | 2113 | error = ep_insert(ep, &epds, tf.file, fd, full_check); |
7699acd1 DL |
2114 | } else |
2115 | error = -EEXIST; | |
67347fe4 JB |
2116 | if (full_check) |
2117 | clear_tfile_check_list(); | |
7699acd1 DL |
2118 | break; |
2119 | case EPOLL_CTL_DEL: | |
2120 | if (epi) | |
2121 | error = ep_remove(ep, epi); | |
2122 | else | |
2123 | error = -ENOENT; | |
2124 | break; | |
2125 | case EPOLL_CTL_MOD: | |
2126 | if (epi) { | |
b6a515c8 JB |
2127 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
2128 | epds.events |= POLLERR | POLLHUP; | |
2129 | error = ep_modify(ep, epi, &epds); | |
2130 | } | |
7699acd1 DL |
2131 | } else |
2132 | error = -ENOENT; | |
2133 | break; | |
2134 | } | |
67347fe4 JB |
2135 | if (tep != NULL) |
2136 | mutex_unlock(&tep->mtx); | |
d47de16c | 2137 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
2138 | |
2139 | error_tgt_fput: | |
67347fe4 | 2140 | if (full_check) |
22bacca4 DL |
2141 | mutex_unlock(&epmutex); |
2142 | ||
7e3fb584 | 2143 | fdput(tf); |
7699acd1 | 2144 | error_fput: |
7e3fb584 | 2145 | fdput(f); |
7699acd1 | 2146 | error_return: |
7699acd1 DL |
2147 | |
2148 | return error; | |
2149 | } | |
2150 | ||
2151 | /* | |
2152 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2153 | * part of the user space epoll_wait(2). | |
2154 | */ | |
5a8a82b1 HC |
2155 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
2156 | int, maxevents, int, timeout) | |
7699acd1 | 2157 | { |
2903ff01 AV |
2158 | int error; |
2159 | struct fd f; | |
7699acd1 DL |
2160 | struct eventpoll *ep; |
2161 | ||
7699acd1 DL |
2162 | /* The maximum number of event must be greater than zero */ |
2163 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
2164 | return -EINVAL; | |
2165 | ||
2166 | /* Verify that the area passed by the user is writeable */ | |
2903ff01 AV |
2167 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) |
2168 | return -EFAULT; | |
7699acd1 DL |
2169 | |
2170 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
2171 | f = fdget(epfd); |
2172 | if (!f.file) | |
2173 | return -EBADF; | |
7699acd1 DL |
2174 | |
2175 | /* | |
2176 | * We have to check that the file structure underneath the fd | |
2177 | * the user passed to us _is_ an eventpoll file. | |
2178 | */ | |
2179 | error = -EINVAL; | |
2903ff01 | 2180 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
2181 | goto error_fput; |
2182 | ||
2183 | /* | |
2184 | * At this point it is safe to assume that the "private_data" contains | |
2185 | * our own data structure. | |
2186 | */ | |
2903ff01 | 2187 | ep = f.file->private_data; |
7699acd1 DL |
2188 | |
2189 | /* Time to fish for events ... */ | |
2190 | error = ep_poll(ep, events, maxevents, timeout); | |
2191 | ||
2192 | error_fput: | |
2903ff01 | 2193 | fdput(f); |
7699acd1 DL |
2194 | return error; |
2195 | } | |
2196 | ||
7699acd1 DL |
2197 | /* |
2198 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2199 | * part of the user space epoll_pwait(2). | |
2200 | */ | |
5a8a82b1 HC |
2201 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
2202 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
2203 | size_t, sigsetsize) | |
7699acd1 DL |
2204 | { |
2205 | int error; | |
2206 | sigset_t ksigmask, sigsaved; | |
2207 | ||
2208 | /* | |
2209 | * If the caller wants a certain signal mask to be set during the wait, | |
2210 | * we apply it here. | |
2211 | */ | |
2212 | if (sigmask) { | |
2213 | if (sigsetsize != sizeof(sigset_t)) | |
2214 | return -EINVAL; | |
2215 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
2216 | return -EFAULT; | |
77d55918 ON |
2217 | sigsaved = current->blocked; |
2218 | set_current_blocked(&ksigmask); | |
7699acd1 DL |
2219 | } |
2220 | ||
2221 | error = sys_epoll_wait(epfd, events, maxevents, timeout); | |
2222 | ||
2223 | /* | |
2224 | * If we changed the signal mask, we need to restore the original one. | |
2225 | * In case we've got a signal while waiting, we do not restore the | |
2226 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
2227 | * the way back to userspace, before the signal mask is restored. | |
2228 | */ | |
2229 | if (sigmask) { | |
2230 | if (error == -EINTR) { | |
2231 | memcpy(¤t->saved_sigmask, &sigsaved, | |
c7ea7630 | 2232 | sizeof(sigsaved)); |
4e4c22c7 | 2233 | set_restore_sigmask(); |
7699acd1 | 2234 | } else |
77d55918 | 2235 | set_current_blocked(&sigsaved); |
7699acd1 DL |
2236 | } |
2237 | ||
2238 | return error; | |
2239 | } | |
2240 | ||
35280bd4 AV |
2241 | #ifdef CONFIG_COMPAT |
2242 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, | |
2243 | struct epoll_event __user *, events, | |
2244 | int, maxevents, int, timeout, | |
2245 | const compat_sigset_t __user *, sigmask, | |
2246 | compat_size_t, sigsetsize) | |
2247 | { | |
2248 | long err; | |
2249 | compat_sigset_t csigmask; | |
2250 | sigset_t ksigmask, sigsaved; | |
2251 | ||
2252 | /* | |
2253 | * If the caller wants a certain signal mask to be set during the wait, | |
2254 | * we apply it here. | |
2255 | */ | |
2256 | if (sigmask) { | |
2257 | if (sigsetsize != sizeof(compat_sigset_t)) | |
2258 | return -EINVAL; | |
2259 | if (copy_from_user(&csigmask, sigmask, sizeof(csigmask))) | |
2260 | return -EFAULT; | |
2261 | sigset_from_compat(&ksigmask, &csigmask); | |
77d55918 ON |
2262 | sigsaved = current->blocked; |
2263 | set_current_blocked(&ksigmask); | |
35280bd4 AV |
2264 | } |
2265 | ||
2266 | err = sys_epoll_wait(epfd, events, maxevents, timeout); | |
2267 | ||
2268 | /* | |
2269 | * If we changed the signal mask, we need to restore the original one. | |
2270 | * In case we've got a signal while waiting, we do not restore the | |
2271 | * signal mask yet, and we allow do_signal() to deliver the signal on | |
2272 | * the way back to userspace, before the signal mask is restored. | |
2273 | */ | |
2274 | if (sigmask) { | |
2275 | if (err == -EINTR) { | |
2276 | memcpy(¤t->saved_sigmask, &sigsaved, | |
2277 | sizeof(sigsaved)); | |
2278 | set_restore_sigmask(); | |
2279 | } else | |
77d55918 | 2280 | set_current_blocked(&sigsaved); |
35280bd4 AV |
2281 | } |
2282 | ||
2283 | return err; | |
2284 | } | |
2285 | #endif | |
2286 | ||
1da177e4 LT |
2287 | static int __init eventpoll_init(void) |
2288 | { | |
7ef9964e DL |
2289 | struct sysinfo si; |
2290 | ||
2291 | si_meminfo(&si); | |
9df04e1f DL |
2292 | /* |
2293 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
2294 | */ | |
2295 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 2296 | EP_ITEM_COST; |
52bd19f7 | 2297 | BUG_ON(max_user_watches < 0); |
1da177e4 | 2298 | |
22bacca4 DL |
2299 | /* |
2300 | * Initialize the structure used to perform epoll file descriptor | |
2301 | * inclusion loops checks. | |
2302 | */ | |
2303 | ep_nested_calls_init(&poll_loop_ncalls); | |
2304 | ||
57a173bd | 2305 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1da177e4 | 2306 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
5071f97e | 2307 | ep_nested_calls_init(&poll_safewake_ncalls); |
57a173bd | 2308 | #endif |
5071f97e DL |
2309 | |
2310 | /* Initialize the structure used to perform file's f_op->poll() calls */ | |
2311 | ep_nested_calls_init(&poll_readywalk_ncalls); | |
1da177e4 | 2312 | |
39732ca5 EW |
2313 | /* |
2314 | * We can have many thousands of epitems, so prevent this from | |
2315 | * using an extra cache line on 64-bit (and smaller) CPUs | |
2316 | */ | |
2317 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); | |
2318 | ||
1da177e4 LT |
2319 | /* Allocates slab cache used to allocate "struct epitem" items */ |
2320 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
2ae928a9 | 2321 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 LT |
2322 | |
2323 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
2324 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
2ae928a9 | 2325 | sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 | 2326 | |
1da177e4 | 2327 | return 0; |
1da177e4 | 2328 | } |
cea69241 | 2329 | fs_initcall(eventpoll_init); |