Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 | 2 | /* |
5071f97e DL |
3 | * fs/eventpoll.c (Efficient event retrieval implementation) |
4 | * Copyright (C) 2001,...,2009 Davide Libenzi | |
1da177e4 | 5 | * |
1da177e4 | 6 | * Davide Libenzi <davidel@xmailserver.org> |
1da177e4 LT |
7 | */ |
8 | ||
1da177e4 LT |
9 | #include <linux/init.h> |
10 | #include <linux/kernel.h> | |
174cd4b1 | 11 | #include <linux/sched/signal.h> |
1da177e4 LT |
12 | #include <linux/fs.h> |
13 | #include <linux/file.h> | |
14 | #include <linux/signal.h> | |
15 | #include <linux/errno.h> | |
16 | #include <linux/mm.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/poll.h> | |
1da177e4 LT |
19 | #include <linux/string.h> |
20 | #include <linux/list.h> | |
21 | #include <linux/hash.h> | |
22 | #include <linux/spinlock.h> | |
23 | #include <linux/syscalls.h> | |
1da177e4 LT |
24 | #include <linux/rbtree.h> |
25 | #include <linux/wait.h> | |
26 | #include <linux/eventpoll.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/bitops.h> | |
144efe3e | 29 | #include <linux/mutex.h> |
da66f7cb | 30 | #include <linux/anon_inodes.h> |
4d7e30d9 | 31 | #include <linux/device.h> |
7c0f6ba6 | 32 | #include <linux/uaccess.h> |
1da177e4 LT |
33 | #include <asm/io.h> |
34 | #include <asm/mman.h> | |
60063497 | 35 | #include <linux/atomic.h> |
138d22b5 CG |
36 | #include <linux/proc_fs.h> |
37 | #include <linux/seq_file.h> | |
35280bd4 | 38 | #include <linux/compat.h> |
ae10b2b4 | 39 | #include <linux/rculist.h> |
bf3b9f63 | 40 | #include <net/busy_poll.h> |
1da177e4 | 41 | |
1da177e4 LT |
42 | /* |
43 | * LOCKING: | |
44 | * There are three level of locking required by epoll : | |
45 | * | |
144efe3e | 46 | * 1) epmutex (mutex) |
c7ea7630 | 47 | * 2) ep->mtx (mutex) |
a218cc49 | 48 | * 3) ep->lock (rwlock) |
1da177e4 LT |
49 | * |
50 | * The acquire order is the one listed above, from 1 to 3. | |
a218cc49 | 51 | * We need a rwlock (ep->lock) because we manipulate objects |
1da177e4 LT |
52 | * from inside the poll callback, that might be triggered from |
53 | * a wake_up() that in turn might be called from IRQ context. | |
54 | * So we can't sleep inside the poll callback and hence we need | |
55 | * a spinlock. During the event transfer loop (from kernel to | |
56 | * user space) we could end up sleeping due a copy_to_user(), so | |
57 | * we need a lock that will allow us to sleep. This lock is a | |
d47de16c DL |
58 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
59 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). | |
60 | * Then we also need a global mutex to serialize eventpoll_release_file() | |
61 | * and ep_free(). | |
62 | * This mutex is acquired by ep_free() during the epoll file | |
1da177e4 LT |
63 | * cleanup path and it is also acquired by eventpoll_release_file() |
64 | * if a file has been pushed inside an epoll set and it is then | |
bf6a41db | 65 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
22bacca4 DL |
66 | * It is also acquired when inserting an epoll fd onto another epoll |
67 | * fd. We do this so that we walk the epoll tree and ensure that this | |
68 | * insertion does not create a cycle of epoll file descriptors, which | |
69 | * could lead to deadlock. We need a global mutex to prevent two | |
70 | * simultaneous inserts (A into B and B into A) from racing and | |
71 | * constructing a cycle without either insert observing that it is | |
72 | * going to. | |
d8805e63 NE |
73 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
74 | * case when one epoll fd is added to another. In this case, we | |
75 | * always acquire the locks in the order of nesting (i.e. after | |
76 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired | |
77 | * before e2->mtx). Since we disallow cycles of epoll file | |
78 | * descriptors, this ensures that the mutexes are well-ordered. In | |
79 | * order to communicate this nesting to lockdep, when walking a tree | |
80 | * of epoll file descriptors, we use the current recursion depth as | |
81 | * the lockdep subkey. | |
d47de16c | 82 | * It is possible to drop the "ep->mtx" and to use the global |
a218cc49 | 83 | * mutex "epmutex" (together with "ep->lock") to have it working, |
d47de16c | 84 | * but having "ep->mtx" will make the interface more scalable. |
144efe3e | 85 | * Events that require holding "epmutex" are very rare, while for |
d47de16c DL |
86 | * normal operations the epoll private "ep->mtx" will guarantee |
87 | * a better scalability. | |
1da177e4 LT |
88 | */ |
89 | ||
1da177e4 | 90 | /* Epoll private bits inside the event mask */ |
df0108c5 | 91 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE) |
1da177e4 | 92 | |
a9a08845 | 93 | #define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT) |
b6a515c8 | 94 | |
a9a08845 | 95 | #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \ |
b6a515c8 JB |
96 | EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE) |
97 | ||
5071f97e DL |
98 | /* Maximum number of nesting allowed inside epoll sets */ |
99 | #define EP_MAX_NESTS 4 | |
1da177e4 | 100 | |
b611967d DL |
101 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
102 | ||
d47de16c DL |
103 | #define EP_UNACTIVE_PTR ((void *) -1L) |
104 | ||
7ef9964e DL |
105 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
106 | ||
1da177e4 LT |
107 | struct epoll_filefd { |
108 | struct file *file; | |
109 | int fd; | |
39732ca5 | 110 | } __packed; |
1da177e4 | 111 | |
80285b75 AV |
112 | /* Wait structure used by the poll hooks */ |
113 | struct eppoll_entry { | |
114 | /* List header used to link this structure to the "struct epitem" */ | |
115 | struct eppoll_entry *next; | |
1da177e4 | 116 | |
80285b75 AV |
117 | /* The "base" pointer is set to the container "struct epitem" */ |
118 | struct epitem *base; | |
119 | ||
120 | /* | |
121 | * Wait queue item that will be linked to the target file wait | |
122 | * queue head. | |
123 | */ | |
124 | wait_queue_entry_t wait; | |
125 | ||
126 | /* The wait queue head that linked the "wait" wait queue item */ | |
127 | wait_queue_head_t *whead; | |
1da177e4 LT |
128 | }; |
129 | ||
d47de16c DL |
130 | /* |
131 | * Each file descriptor added to the eventpoll interface will | |
132 | * have an entry of this type linked to the "rbr" RB tree. | |
39732ca5 EW |
133 | * Avoid increasing the size of this struct, there can be many thousands |
134 | * of these on a server and we do not want this to take another cache line. | |
d47de16c DL |
135 | */ |
136 | struct epitem { | |
ae10b2b4 JB |
137 | union { |
138 | /* RB tree node links this structure to the eventpoll RB tree */ | |
139 | struct rb_node rbn; | |
140 | /* Used to free the struct epitem */ | |
141 | struct rcu_head rcu; | |
142 | }; | |
d47de16c DL |
143 | |
144 | /* List header used to link this structure to the eventpoll ready list */ | |
145 | struct list_head rdllink; | |
146 | ||
c7ea7630 DL |
147 | /* |
148 | * Works together "struct eventpoll"->ovflist in keeping the | |
149 | * single linked chain of items. | |
150 | */ | |
151 | struct epitem *next; | |
152 | ||
d47de16c DL |
153 | /* The file descriptor information this item refers to */ |
154 | struct epoll_filefd ffd; | |
155 | ||
d47de16c | 156 | /* List containing poll wait queues */ |
80285b75 | 157 | struct eppoll_entry *pwqlist; |
d47de16c DL |
158 | |
159 | /* The "container" of this item */ | |
160 | struct eventpoll *ep; | |
161 | ||
d47de16c | 162 | /* List header used to link this item to the "struct file" items list */ |
44cdc1d9 | 163 | struct hlist_node fllink; |
d47de16c | 164 | |
4d7e30d9 | 165 | /* wakeup_source used when EPOLLWAKEUP is set */ |
eea1d585 | 166 | struct wakeup_source __rcu *ws; |
4d7e30d9 | 167 | |
c7ea7630 DL |
168 | /* The structure that describe the interested events and the source fd */ |
169 | struct epoll_event event; | |
d47de16c DL |
170 | }; |
171 | ||
1da177e4 LT |
172 | /* |
173 | * This structure is stored inside the "private_data" member of the file | |
bf6a41db | 174 | * structure and represents the main data structure for the eventpoll |
1da177e4 LT |
175 | * interface. |
176 | */ | |
177 | struct eventpoll { | |
1da177e4 | 178 | /* |
d47de16c DL |
179 | * This mutex is used to ensure that files are not removed |
180 | * while epoll is using them. This is held during the event | |
181 | * collection loop, the file cleanup path, the epoll file exit | |
182 | * code and the ctl operations. | |
1da177e4 | 183 | */ |
d47de16c | 184 | struct mutex mtx; |
1da177e4 LT |
185 | |
186 | /* Wait queue used by sys_epoll_wait() */ | |
187 | wait_queue_head_t wq; | |
188 | ||
189 | /* Wait queue used by file->poll() */ | |
190 | wait_queue_head_t poll_wait; | |
191 | ||
192 | /* List of ready file descriptors */ | |
193 | struct list_head rdllist; | |
194 | ||
a218cc49 RP |
195 | /* Lock which protects rdllist and ovflist */ |
196 | rwlock_t lock; | |
197 | ||
67647d0f | 198 | /* RB tree root used to store monitored fd structs */ |
b2ac2ea6 | 199 | struct rb_root_cached rbr; |
d47de16c DL |
200 | |
201 | /* | |
202 | * This is a single linked list that chains all the "struct epitem" that | |
25985edc | 203 | * happened while transferring ready events to userspace w/out |
a218cc49 | 204 | * holding ->lock. |
d47de16c DL |
205 | */ |
206 | struct epitem *ovflist; | |
7ef9964e | 207 | |
4d7e30d9 AH |
208 | /* wakeup_source used when ep_scan_ready_list is running */ |
209 | struct wakeup_source *ws; | |
210 | ||
7ef9964e DL |
211 | /* The user that created the eventpoll descriptor */ |
212 | struct user_struct *user; | |
28d82dc1 JB |
213 | |
214 | struct file *file; | |
215 | ||
216 | /* used to optimize loop detection check */ | |
18306c40 | 217 | u64 gen; |
319c1517 | 218 | struct hlist_head refs; |
bf3b9f63 SS |
219 | |
220 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
221 | /* used to track busy poll napi_id */ | |
222 | unsigned int napi_id; | |
223 | #endif | |
efcdd350 JB |
224 | |
225 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
226 | /* tracks wakeup nests for lockdep validation */ | |
227 | u8 nests; | |
228 | #endif | |
1da177e4 LT |
229 | }; |
230 | ||
1da177e4 LT |
231 | /* Wrapper struct used by poll queueing */ |
232 | struct ep_pqueue { | |
233 | poll_table pt; | |
234 | struct epitem *epi; | |
235 | }; | |
236 | ||
7ef9964e DL |
237 | /* |
238 | * Configuration options available inside /proc/sys/fs/epoll/ | |
239 | */ | |
7ef9964e | 240 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 241 | static long max_user_watches __read_mostly; |
7ef9964e | 242 | |
1da177e4 | 243 | /* |
d47de16c | 244 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 245 | */ |
7ef9964e | 246 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 247 | |
18306c40 AV |
248 | static u64 loop_check_gen = 0; |
249 | ||
22bacca4 | 250 | /* Used to check for epoll file descriptor inclusion loops */ |
6a3890c4 | 251 | static struct eventpoll *inserting_into; |
22bacca4 | 252 | |
1da177e4 | 253 | /* Slab cache used to allocate "struct epitem" */ |
e18b890b | 254 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
255 | |
256 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 257 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 258 | |
28d82dc1 JB |
259 | /* |
260 | * List of files with newly added links, where we may need to limit the number | |
261 | * of emanating paths. Protected by the epmutex. | |
262 | */ | |
319c1517 AV |
263 | struct epitems_head { |
264 | struct hlist_head epitems; | |
265 | struct epitems_head *next; | |
266 | }; | |
267 | static struct epitems_head *tfile_check_list = EP_UNACTIVE_PTR; | |
268 | ||
269 | static struct kmem_cache *ephead_cache __read_mostly; | |
270 | ||
271 | static inline void free_ephead(struct epitems_head *head) | |
272 | { | |
273 | if (head) | |
274 | kmem_cache_free(ephead_cache, head); | |
275 | } | |
276 | ||
277 | static void list_file(struct file *file) | |
278 | { | |
279 | struct epitems_head *head; | |
280 | ||
281 | head = container_of(file->f_ep, struct epitems_head, epitems); | |
282 | if (!head->next) { | |
283 | head->next = tfile_check_list; | |
284 | tfile_check_list = head; | |
285 | } | |
286 | } | |
287 | ||
288 | static void unlist_file(struct epitems_head *head) | |
289 | { | |
290 | struct epitems_head *to_free = head; | |
291 | struct hlist_node *p = rcu_dereference(hlist_first_rcu(&head->epitems)); | |
292 | if (p) { | |
293 | struct epitem *epi= container_of(p, struct epitem, fllink); | |
294 | spin_lock(&epi->ffd.file->f_lock); | |
295 | if (!hlist_empty(&head->epitems)) | |
296 | to_free = NULL; | |
297 | head->next = NULL; | |
298 | spin_unlock(&epi->ffd.file->f_lock); | |
299 | } | |
300 | free_ephead(to_free); | |
301 | } | |
28d82dc1 | 302 | |
7ef9964e DL |
303 | #ifdef CONFIG_SYSCTL |
304 | ||
305 | #include <linux/sysctl.h> | |
306 | ||
eec4844f | 307 | static long long_zero; |
52bd19f7 | 308 | static long long_max = LONG_MAX; |
7ef9964e | 309 | |
a8f5de89 | 310 | static struct ctl_table epoll_table[] = { |
7ef9964e DL |
311 | { |
312 | .procname = "max_user_watches", | |
313 | .data = &max_user_watches, | |
52bd19f7 | 314 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 315 | .mode = 0644, |
52bd19f7 | 316 | .proc_handler = proc_doulongvec_minmax, |
eec4844f | 317 | .extra1 = &long_zero, |
52bd19f7 | 318 | .extra2 = &long_max, |
7ef9964e | 319 | }, |
ab09203e | 320 | { } |
7ef9964e | 321 | }; |
a8f5de89 XN |
322 | |
323 | static void __init epoll_sysctls_init(void) | |
324 | { | |
325 | register_sysctl("fs/epoll", epoll_table); | |
326 | } | |
327 | #else | |
328 | #define epoll_sysctls_init() do { } while (0) | |
7ef9964e DL |
329 | #endif /* CONFIG_SYSCTL */ |
330 | ||
28d82dc1 JB |
331 | static const struct file_operations eventpoll_fops; |
332 | ||
333 | static inline int is_file_epoll(struct file *f) | |
334 | { | |
335 | return f->f_op == &eventpoll_fops; | |
336 | } | |
b030a4dd | 337 | |
67647d0f | 338 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
339 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
340 | struct file *file, int fd) | |
341 | { | |
342 | ffd->file = file; | |
343 | ffd->fd = fd; | |
344 | } | |
345 | ||
67647d0f | 346 | /* Compare RB tree keys */ |
b030a4dd PE |
347 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
348 | struct epoll_filefd *p2) | |
349 | { | |
350 | return (p1->file > p2->file ? +1: | |
351 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
352 | } | |
353 | ||
b030a4dd | 354 | /* Tells us if the item is currently linked */ |
992991c0 | 355 | static inline int ep_is_linked(struct epitem *epi) |
b030a4dd | 356 | { |
992991c0 | 357 | return !list_empty(&epi->rdllink); |
b030a4dd PE |
358 | } |
359 | ||
ac6424b9 | 360 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
971316f0 ON |
361 | { |
362 | return container_of(p, struct eppoll_entry, wait); | |
363 | } | |
364 | ||
b030a4dd | 365 | /* Get the "struct epitem" from a wait queue pointer */ |
ac6424b9 | 366 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
b030a4dd PE |
367 | { |
368 | return container_of(p, struct eppoll_entry, wait)->base; | |
369 | } | |
370 | ||
3fb0e584 DL |
371 | /** |
372 | * ep_events_available - Checks if ready events might be available. | |
373 | * | |
374 | * @ep: Pointer to the eventpoll context. | |
375 | * | |
a6c67fee RD |
376 | * Return: a value different than %zero if ready events are available, |
377 | * or %zero otherwise. | |
3fb0e584 DL |
378 | */ |
379 | static inline int ep_events_available(struct eventpoll *ep) | |
380 | { | |
c5a282e9 DB |
381 | return !list_empty_careful(&ep->rdllist) || |
382 | READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR; | |
3fb0e584 DL |
383 | } |
384 | ||
bf3b9f63 SS |
385 | #ifdef CONFIG_NET_RX_BUSY_POLL |
386 | static bool ep_busy_loop_end(void *p, unsigned long start_time) | |
387 | { | |
388 | struct eventpoll *ep = p; | |
389 | ||
390 | return ep_events_available(ep) || busy_loop_timeout(start_time); | |
391 | } | |
bf3b9f63 SS |
392 | |
393 | /* | |
394 | * Busy poll if globally on and supporting sockets found && no events, | |
395 | * busy loop will return if need_resched or ep_events_available. | |
396 | * | |
397 | * we must do our busy polling with irqs enabled | |
398 | */ | |
1493c47f | 399 | static bool ep_busy_loop(struct eventpoll *ep, int nonblock) |
bf3b9f63 | 400 | { |
bf3b9f63 SS |
401 | unsigned int napi_id = READ_ONCE(ep->napi_id); |
402 | ||
1493c47f | 403 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) { |
7c951caf BT |
404 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep, false, |
405 | BUSY_POLL_BUDGET); | |
1493c47f SHY |
406 | if (ep_events_available(ep)) |
407 | return true; | |
408 | /* | |
409 | * Busy poll timed out. Drop NAPI ID for now, we can add | |
410 | * it back in when we have moved a socket with a valid NAPI | |
411 | * ID onto the ready list. | |
412 | */ | |
bf3b9f63 | 413 | ep->napi_id = 0; |
1493c47f SHY |
414 | return false; |
415 | } | |
416 | return false; | |
bf3b9f63 SS |
417 | } |
418 | ||
419 | /* | |
420 | * Set epoll busy poll NAPI ID from sk. | |
421 | */ | |
422 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
423 | { | |
bf3b9f63 SS |
424 | struct eventpoll *ep; |
425 | unsigned int napi_id; | |
426 | struct socket *sock; | |
427 | struct sock *sk; | |
bf3b9f63 SS |
428 | |
429 | if (!net_busy_loop_on()) | |
430 | return; | |
431 | ||
dba4a925 | 432 | sock = sock_from_file(epi->ffd.file); |
bf3b9f63 SS |
433 | if (!sock) |
434 | return; | |
435 | ||
436 | sk = sock->sk; | |
437 | if (!sk) | |
438 | return; | |
439 | ||
440 | napi_id = READ_ONCE(sk->sk_napi_id); | |
441 | ep = epi->ep; | |
442 | ||
443 | /* Non-NAPI IDs can be rejected | |
444 | * or | |
445 | * Nothing to do if we already have this ID | |
446 | */ | |
447 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) | |
448 | return; | |
449 | ||
450 | /* record NAPI ID for use in next busy poll */ | |
451 | ep->napi_id = napi_id; | |
bf3b9f63 SS |
452 | } |
453 | ||
514056d5 DB |
454 | #else |
455 | ||
1493c47f | 456 | static inline bool ep_busy_loop(struct eventpoll *ep, int nonblock) |
514056d5 | 457 | { |
1493c47f | 458 | return false; |
514056d5 DB |
459 | } |
460 | ||
461 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
462 | { | |
463 | } | |
464 | ||
465 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
466 | ||
02edc6fc SR |
467 | /* |
468 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
469 | * the use of wait queues used by epoll is done in a very controlled | |
470 | * manner. Wake ups can nest inside each other, but are never done | |
471 | * with the same locking. For example: | |
472 | * | |
473 | * dfd = socket(...); | |
474 | * efd1 = epoll_create(); | |
475 | * efd2 = epoll_create(); | |
476 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
477 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
478 | * | |
479 | * When a packet arrives to the device underneath "dfd", the net code will | |
480 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
481 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
482 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
483 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
484 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
485 | * that ends up in another wake_up(), after having checked about the | |
486 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
487 | * avoid stack blasting. | |
488 | * | |
489 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
490 | * this special case of epoll. | |
491 | */ | |
2dfa4eea | 492 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
57a173bd | 493 | |
caf1aeaf JA |
494 | static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi, |
495 | unsigned pollflags) | |
5071f97e | 496 | { |
efcdd350 | 497 | struct eventpoll *ep_src; |
f6520c52 | 498 | unsigned long flags; |
efcdd350 JB |
499 | u8 nests = 0; |
500 | ||
501 | /* | |
502 | * To set the subclass or nesting level for spin_lock_irqsave_nested() | |
503 | * it might be natural to create a per-cpu nest count. However, since | |
504 | * we can recurse on ep->poll_wait.lock, and a non-raw spinlock can | |
505 | * schedule() in the -rt kernel, the per-cpu variable are no longer | |
506 | * protected. Thus, we are introducing a per eventpoll nest field. | |
507 | * If we are not being call from ep_poll_callback(), epi is NULL and | |
508 | * we are at the first level of nesting, 0. Otherwise, we are being | |
509 | * called from ep_poll_callback() and if a previous wakeup source is | |
510 | * not an epoll file itself, we are at depth 1 since the wakeup source | |
511 | * is depth 0. If the wakeup source is a previous epoll file in the | |
512 | * wakeup chain then we use its nests value and record ours as | |
513 | * nests + 1. The previous epoll file nests value is stable since its | |
514 | * already holding its own poll_wait.lock. | |
515 | */ | |
516 | if (epi) { | |
517 | if ((is_file_epoll(epi->ffd.file))) { | |
518 | ep_src = epi->ffd.file->private_data; | |
519 | nests = ep_src->nests; | |
520 | } else { | |
521 | nests = 1; | |
522 | } | |
523 | } | |
524 | spin_lock_irqsave_nested(&ep->poll_wait.lock, flags, nests); | |
525 | ep->nests = nests + 1; | |
caf1aeaf | 526 | wake_up_locked_poll(&ep->poll_wait, EPOLLIN | pollflags); |
efcdd350 JB |
527 | ep->nests = 0; |
528 | spin_unlock_irqrestore(&ep->poll_wait.lock, flags); | |
1da177e4 LT |
529 | } |
530 | ||
57a173bd JB |
531 | #else |
532 | ||
caf1aeaf JA |
533 | static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi, |
534 | unsigned pollflags) | |
57a173bd | 535 | { |
caf1aeaf | 536 | wake_up_poll(&ep->poll_wait, EPOLLIN | pollflags); |
57a173bd JB |
537 | } |
538 | ||
539 | #endif | |
540 | ||
971316f0 ON |
541 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
542 | { | |
543 | wait_queue_head_t *whead; | |
544 | ||
545 | rcu_read_lock(); | |
138e4ad6 ON |
546 | /* |
547 | * If it is cleared by POLLFREE, it should be rcu-safe. | |
548 | * If we read NULL we need a barrier paired with | |
549 | * smp_store_release() in ep_poll_callback(), otherwise | |
550 | * we rely on whead->lock. | |
551 | */ | |
552 | whead = smp_load_acquire(&pwq->whead); | |
971316f0 ON |
553 | if (whead) |
554 | remove_wait_queue(whead, &pwq->wait); | |
555 | rcu_read_unlock(); | |
556 | } | |
557 | ||
1da177e4 | 558 | /* |
d1bc90dd TB |
559 | * This function unregisters poll callbacks from the associated file |
560 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
561 | * ep_free). | |
1da177e4 | 562 | */ |
7699acd1 | 563 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 564 | { |
80285b75 | 565 | struct eppoll_entry **p = &epi->pwqlist; |
7699acd1 | 566 | struct eppoll_entry *pwq; |
1da177e4 | 567 | |
80285b75 AV |
568 | while ((pwq = *p) != NULL) { |
569 | *p = pwq->next; | |
971316f0 | 570 | ep_remove_wait_queue(pwq); |
d1bc90dd | 571 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 572 | } |
1da177e4 LT |
573 | } |
574 | ||
eea1d585 EW |
575 | /* call only when ep->mtx is held */ |
576 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) | |
577 | { | |
578 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); | |
579 | } | |
580 | ||
581 | /* call only when ep->mtx is held */ | |
582 | static inline void ep_pm_stay_awake(struct epitem *epi) | |
583 | { | |
584 | struct wakeup_source *ws = ep_wakeup_source(epi); | |
585 | ||
586 | if (ws) | |
587 | __pm_stay_awake(ws); | |
588 | } | |
589 | ||
590 | static inline bool ep_has_wakeup_source(struct epitem *epi) | |
591 | { | |
592 | return rcu_access_pointer(epi->ws) ? true : false; | |
593 | } | |
594 | ||
595 | /* call when ep->mtx cannot be held (ep_poll_callback) */ | |
596 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) | |
597 | { | |
598 | struct wakeup_source *ws; | |
599 | ||
600 | rcu_read_lock(); | |
601 | ws = rcu_dereference(epi->ws); | |
602 | if (ws) | |
603 | __pm_stay_awake(ws); | |
604 | rcu_read_unlock(); | |
605 | } | |
606 | ||
5071f97e | 607 | |
57804b1c AV |
608 | /* |
609 | * ep->mutex needs to be held because we could be hit by | |
610 | * eventpoll_release_file() and epoll_ctl(). | |
5071f97e | 611 | */ |
57804b1c | 612 | static void ep_start_scan(struct eventpoll *ep, struct list_head *txlist) |
5071f97e | 613 | { |
5071f97e DL |
614 | /* |
615 | * Steal the ready list, and re-init the original one to the | |
616 | * empty list. Also, set ep->ovflist to NULL so that events | |
617 | * happening while looping w/out locks, are not lost. We cannot | |
618 | * have the poll callback to queue directly on ep->rdllist, | |
619 | * because we want the "sproc" callback to be able to do it | |
620 | * in a lockless way. | |
621 | */ | |
57804b1c | 622 | lockdep_assert_irqs_enabled(); |
a218cc49 | 623 | write_lock_irq(&ep->lock); |
db502f8a | 624 | list_splice_init(&ep->rdllist, txlist); |
c5a282e9 | 625 | WRITE_ONCE(ep->ovflist, NULL); |
a218cc49 | 626 | write_unlock_irq(&ep->lock); |
db502f8a | 627 | } |
5071f97e | 628 | |
db502f8a | 629 | static void ep_done_scan(struct eventpoll *ep, |
db502f8a AV |
630 | struct list_head *txlist) |
631 | { | |
632 | struct epitem *epi, *nepi; | |
5071f97e | 633 | |
a218cc49 | 634 | write_lock_irq(&ep->lock); |
5071f97e DL |
635 | /* |
636 | * During the time we spent inside the "sproc" callback, some | |
637 | * other events might have been queued by the poll callback. | |
638 | * We re-insert them inside the main ready-list here. | |
639 | */ | |
c5a282e9 | 640 | for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL; |
5071f97e DL |
641 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
642 | /* | |
643 | * We need to check if the item is already in the list. | |
644 | * During the "sproc" callback execution time, items are | |
645 | * queued into ->ovflist but the "txlist" might already | |
646 | * contain them, and the list_splice() below takes care of them. | |
647 | */ | |
992991c0 | 648 | if (!ep_is_linked(epi)) { |
c141175d RP |
649 | /* |
650 | * ->ovflist is LIFO, so we have to reverse it in order | |
651 | * to keep in FIFO. | |
652 | */ | |
653 | list_add(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 654 | ep_pm_stay_awake(epi); |
4d7e30d9 | 655 | } |
5071f97e DL |
656 | } |
657 | /* | |
658 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
659 | * releasing the lock, events will be queued in the normal way inside | |
660 | * ep->rdllist. | |
661 | */ | |
c5a282e9 | 662 | WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR); |
5071f97e DL |
663 | |
664 | /* | |
665 | * Quickly re-inject items left on "txlist". | |
666 | */ | |
db502f8a | 667 | list_splice(txlist, &ep->rdllist); |
4d7e30d9 | 668 | __pm_relax(ep->ws); |
7fab29e3 DB |
669 | |
670 | if (!list_empty(&ep->rdllist)) { | |
671 | if (waitqueue_active(&ep->wq)) | |
672 | wake_up(&ep->wq); | |
673 | } | |
674 | ||
a218cc49 | 675 | write_unlock_irq(&ep->lock); |
5071f97e DL |
676 | } |
677 | ||
ae10b2b4 JB |
678 | static void epi_rcu_free(struct rcu_head *head) |
679 | { | |
680 | struct epitem *epi = container_of(head, struct epitem, rcu); | |
681 | kmem_cache_free(epi_cache, epi); | |
682 | } | |
683 | ||
7699acd1 DL |
684 | /* |
685 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 686 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
687 | */ |
688 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
689 | { | |
7699acd1 | 690 | struct file *file = epi->ffd.file; |
319c1517 AV |
691 | struct epitems_head *to_free; |
692 | struct hlist_head *head; | |
1da177e4 | 693 | |
92e64178 DB |
694 | lockdep_assert_irqs_enabled(); |
695 | ||
1da177e4 | 696 | /* |
ee8ef0a4 | 697 | * Removes poll wait queue hooks. |
1da177e4 | 698 | */ |
7699acd1 | 699 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 700 | |
7699acd1 | 701 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 702 | spin_lock(&file->f_lock); |
319c1517 AV |
703 | to_free = NULL; |
704 | head = file->f_ep; | |
705 | if (head->first == &epi->fllink && !epi->fllink.next) { | |
706 | file->f_ep = NULL; | |
707 | if (!is_file_epoll(file)) { | |
708 | struct epitems_head *v; | |
709 | v = container_of(head, struct epitems_head, epitems); | |
710 | if (!smp_load_acquire(&v->next)) | |
711 | to_free = v; | |
712 | } | |
713 | } | |
44cdc1d9 | 714 | hlist_del_rcu(&epi->fllink); |
68499914 | 715 | spin_unlock(&file->f_lock); |
319c1517 | 716 | free_ephead(to_free); |
1da177e4 | 717 | |
b2ac2ea6 | 718 | rb_erase_cached(&epi->rbn, &ep->rbr); |
1da177e4 | 719 | |
a218cc49 | 720 | write_lock_irq(&ep->lock); |
992991c0 | 721 | if (ep_is_linked(epi)) |
c7ea7630 | 722 | list_del_init(&epi->rdllink); |
a218cc49 | 723 | write_unlock_irq(&ep->lock); |
1da177e4 | 724 | |
eea1d585 | 725 | wakeup_source_unregister(ep_wakeup_source(epi)); |
ae10b2b4 JB |
726 | /* |
727 | * At this point it is safe to free the eventpoll item. Use the union | |
728 | * field epi->rcu, since we are trying to minimize the size of | |
729 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by | |
730 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make | |
731 | * use of the rbn field. | |
732 | */ | |
733 | call_rcu(&epi->rcu, epi_rcu_free); | |
1da177e4 | 734 | |
1e1c1583 | 735 | percpu_counter_dec(&ep->user->epoll_watches); |
7ef9964e | 736 | |
c7ea7630 | 737 | return 0; |
1da177e4 LT |
738 | } |
739 | ||
7699acd1 | 740 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 741 | { |
7699acd1 DL |
742 | struct rb_node *rbp; |
743 | struct epitem *epi; | |
1da177e4 | 744 | |
7699acd1 DL |
745 | /* We need to release all tasks waiting for these file */ |
746 | if (waitqueue_active(&ep->poll_wait)) | |
caf1aeaf | 747 | ep_poll_safewake(ep, NULL, 0); |
1da177e4 | 748 | |
7699acd1 DL |
749 | /* |
750 | * We need to lock this because we could be hit by | |
751 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 752 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
753 | * is on the way to be removed and no one has references to it |
754 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 755 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
756 | */ |
757 | mutex_lock(&epmutex); | |
1da177e4 LT |
758 | |
759 | /* | |
7699acd1 | 760 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 761 | */ |
b2ac2ea6 | 762 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
7699acd1 DL |
763 | epi = rb_entry(rbp, struct epitem, rbn); |
764 | ||
765 | ep_unregister_pollwait(ep, epi); | |
91cf5ab6 | 766 | cond_resched(); |
7699acd1 | 767 | } |
1da177e4 LT |
768 | |
769 | /* | |
7699acd1 DL |
770 | * Walks through the whole tree by freeing each "struct epitem". At this |
771 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 772 | * holding "epmutex" we can be sure that no file cleanup code will hit |
a218cc49 | 773 | * us during this operation. So we can avoid the lock on "ep->lock". |
ddf676c3 EW |
774 | * We do not need to lock ep->mtx, either, we only do it to prevent |
775 | * a lockdep warning. | |
1da177e4 | 776 | */ |
ddf676c3 | 777 | mutex_lock(&ep->mtx); |
b2ac2ea6 | 778 | while ((rbp = rb_first_cached(&ep->rbr)) != NULL) { |
7699acd1 DL |
779 | epi = rb_entry(rbp, struct epitem, rbn); |
780 | ep_remove(ep, epi); | |
91cf5ab6 | 781 | cond_resched(); |
7699acd1 | 782 | } |
ddf676c3 | 783 | mutex_unlock(&ep->mtx); |
1da177e4 | 784 | |
7699acd1 | 785 | mutex_unlock(&epmutex); |
d47de16c | 786 | mutex_destroy(&ep->mtx); |
7ef9964e | 787 | free_uid(ep->user); |
4d7e30d9 | 788 | wakeup_source_unregister(ep->ws); |
f0ee9aab | 789 | kfree(ep); |
7699acd1 | 790 | } |
1da177e4 | 791 | |
7699acd1 DL |
792 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
793 | { | |
794 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 795 | |
f0ee9aab | 796 | if (ep) |
7699acd1 | 797 | ep_free(ep); |
7699acd1 | 798 | |
7699acd1 | 799 | return 0; |
1da177e4 LT |
800 | } |
801 | ||
2c0b71c1 | 802 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, int depth); |
37b5e521 | 803 | |
ad9366b1 | 804 | static __poll_t __ep_eventpoll_poll(struct file *file, poll_table *wait, int depth) |
5071f97e | 805 | { |
ad9366b1 AV |
806 | struct eventpoll *ep = file->private_data; |
807 | LIST_HEAD(txlist); | |
5071f97e | 808 | struct epitem *epi, *tmp; |
626cf236 | 809 | poll_table pt; |
2c0b71c1 | 810 | __poll_t res = 0; |
5071f97e | 811 | |
626cf236 | 812 | init_poll_funcptr(&pt, NULL); |
450d89ec | 813 | |
ad9366b1 AV |
814 | /* Insert inside our poll wait queue */ |
815 | poll_wait(file, &ep->poll_wait, wait); | |
816 | ||
817 | /* | |
818 | * Proceed to find out if wanted events are really available inside | |
819 | * the ready list. | |
820 | */ | |
821 | mutex_lock_nested(&ep->mtx, depth); | |
822 | ep_start_scan(ep, &txlist); | |
2c0b71c1 AV |
823 | list_for_each_entry_safe(epi, tmp, &txlist, rdllink) { |
824 | if (ep_item_poll(epi, &pt, depth + 1)) { | |
825 | res = EPOLLIN | EPOLLRDNORM; | |
826 | break; | |
37b5e521 | 827 | } else { |
5071f97e DL |
828 | /* |
829 | * Item has been dropped into the ready list by the poll | |
830 | * callback, but it's not actually ready, as far as | |
831 | * caller requested events goes. We can remove it here. | |
832 | */ | |
eea1d585 | 833 | __pm_relax(ep_wakeup_source(epi)); |
5071f97e | 834 | list_del_init(&epi->rdllink); |
296e236e | 835 | } |
5071f97e | 836 | } |
ad9366b1 AV |
837 | ep_done_scan(ep, &txlist); |
838 | mutex_unlock(&ep->mtx); | |
839 | return res; | |
5071f97e DL |
840 | } |
841 | ||
37b5e521 JB |
842 | /* |
843 | * Differs from ep_eventpoll_poll() in that internal callers already have | |
844 | * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested() | |
845 | * is correctly annotated. | |
846 | */ | |
d85e2aa2 | 847 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, |
bec1a502 | 848 | int depth) |
11c5ad0e | 849 | { |
ad9366b1 | 850 | struct file *file = epi->ffd.file; |
1ec09974 | 851 | __poll_t res; |
7699acd1 | 852 | |
450d89ec | 853 | pt->_key = epi->event.events; |
ad9366b1 AV |
854 | if (!is_file_epoll(file)) |
855 | res = vfs_poll(file, pt); | |
856 | else | |
857 | res = __ep_eventpoll_poll(file, pt, depth); | |
1ec09974 | 858 | return res & epi->event.events; |
450d89ec | 859 | } |
a11e1d43 | 860 | |
a11e1d43 | 861 | static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait) |
11c5ad0e | 862 | { |
ad9366b1 | 863 | return __ep_eventpoll_poll(file, wait, 0); |
7699acd1 DL |
864 | } |
865 | ||
138d22b5 | 866 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 867 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
138d22b5 CG |
868 | { |
869 | struct eventpoll *ep = f->private_data; | |
870 | struct rb_node *rbp; | |
138d22b5 CG |
871 | |
872 | mutex_lock(&ep->mtx); | |
b2ac2ea6 | 873 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
138d22b5 | 874 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
77493f04 | 875 | struct inode *inode = file_inode(epi->ffd.file); |
138d22b5 | 876 | |
77493f04 CG |
877 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
878 | " pos:%lli ino:%lx sdev:%x\n", | |
a3816ab0 | 879 | epi->ffd.fd, epi->event.events, |
77493f04 CG |
880 | (long long)epi->event.data, |
881 | (long long)epi->ffd.file->f_pos, | |
882 | inode->i_ino, inode->i_sb->s_dev); | |
a3816ab0 | 883 | if (seq_has_overflowed(m)) |
138d22b5 CG |
884 | break; |
885 | } | |
886 | mutex_unlock(&ep->mtx); | |
138d22b5 CG |
887 | } |
888 | #endif | |
889 | ||
7699acd1 DL |
890 | /* File callbacks that implement the eventpoll file behaviour */ |
891 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
892 | #ifdef CONFIG_PROC_FS |
893 | .show_fdinfo = ep_show_fdinfo, | |
894 | #endif | |
7699acd1 | 895 | .release = ep_eventpoll_release, |
a11e1d43 | 896 | .poll = ep_eventpoll_poll, |
6038f373 | 897 | .llseek = noop_llseek, |
7699acd1 DL |
898 | }; |
899 | ||
b611967d | 900 | /* |
7699acd1 DL |
901 | * This is called from eventpoll_release() to unlink files from the eventpoll |
902 | * interface. We need to have this facility to cleanup correctly files that are | |
903 | * closed without being removed from the eventpoll interface. | |
b611967d | 904 | */ |
7699acd1 | 905 | void eventpoll_release_file(struct file *file) |
b611967d | 906 | { |
7699acd1 | 907 | struct eventpoll *ep; |
44cdc1d9 AV |
908 | struct epitem *epi; |
909 | struct hlist_node *next; | |
b611967d DL |
910 | |
911 | /* | |
68499914 | 912 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 913 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 914 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 915 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 916 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 917 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 918 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 919 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 920 | * from anywhere but ep_free(). |
68499914 JC |
921 | * |
922 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 923 | */ |
7699acd1 | 924 | mutex_lock(&epmutex); |
319c1517 AV |
925 | if (unlikely(!file->f_ep)) { |
926 | mutex_unlock(&epmutex); | |
927 | return; | |
928 | } | |
929 | hlist_for_each_entry_safe(epi, next, file->f_ep, fllink) { | |
7699acd1 | 930 | ep = epi->ep; |
d8805e63 | 931 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 932 | ep_remove(ep, epi); |
d47de16c | 933 | mutex_unlock(&ep->mtx); |
b611967d | 934 | } |
7699acd1 | 935 | mutex_unlock(&epmutex); |
b611967d DL |
936 | } |
937 | ||
53d2be79 | 938 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 939 | { |
7ef9964e DL |
940 | int error; |
941 | struct user_struct *user; | |
942 | struct eventpoll *ep; | |
1da177e4 | 943 | |
7ef9964e | 944 | user = get_current_user(); |
7ef9964e DL |
945 | error = -ENOMEM; |
946 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
947 | if (unlikely(!ep)) | |
948 | goto free_uid; | |
1da177e4 | 949 | |
d47de16c | 950 | mutex_init(&ep->mtx); |
a218cc49 | 951 | rwlock_init(&ep->lock); |
1da177e4 LT |
952 | init_waitqueue_head(&ep->wq); |
953 | init_waitqueue_head(&ep->poll_wait); | |
954 | INIT_LIST_HEAD(&ep->rdllist); | |
b2ac2ea6 | 955 | ep->rbr = RB_ROOT_CACHED; |
d47de16c | 956 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 957 | ep->user = user; |
1da177e4 | 958 | |
53d2be79 | 959 | *pep = ep; |
1da177e4 | 960 | |
1da177e4 | 961 | return 0; |
7ef9964e DL |
962 | |
963 | free_uid: | |
964 | free_uid(user); | |
965 | return error; | |
1da177e4 LT |
966 | } |
967 | ||
1da177e4 | 968 | /* |
c7ea7630 DL |
969 | * Search the file inside the eventpoll tree. The RB tree operations |
970 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
971 | * "mtx" held. | |
1da177e4 LT |
972 | */ |
973 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
974 | { | |
975 | int kcmp; | |
1da177e4 LT |
976 | struct rb_node *rbp; |
977 | struct epitem *epi, *epir = NULL; | |
978 | struct epoll_filefd ffd; | |
979 | ||
b030a4dd | 980 | ep_set_ffd(&ffd, file, fd); |
b2ac2ea6 | 981 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
1da177e4 | 982 | epi = rb_entry(rbp, struct epitem, rbn); |
b030a4dd | 983 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
984 | if (kcmp > 0) |
985 | rbp = rbp->rb_right; | |
986 | else if (kcmp < 0) | |
987 | rbp = rbp->rb_left; | |
988 | else { | |
1da177e4 LT |
989 | epir = epi; |
990 | break; | |
991 | } | |
992 | } | |
1da177e4 | 993 | |
1da177e4 LT |
994 | return epir; |
995 | } | |
996 | ||
bfe3911a | 997 | #ifdef CONFIG_KCMP |
0791e364 CG |
998 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
999 | { | |
1000 | struct rb_node *rbp; | |
1001 | struct epitem *epi; | |
1002 | ||
b2ac2ea6 | 1003 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
0791e364 CG |
1004 | epi = rb_entry(rbp, struct epitem, rbn); |
1005 | if (epi->ffd.fd == tfd) { | |
1006 | if (toff == 0) | |
1007 | return epi; | |
1008 | else | |
1009 | toff--; | |
1010 | } | |
1011 | cond_resched(); | |
1012 | } | |
1013 | ||
1014 | return NULL; | |
1015 | } | |
1016 | ||
1017 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, | |
1018 | unsigned long toff) | |
1019 | { | |
1020 | struct file *file_raw; | |
1021 | struct eventpoll *ep; | |
1022 | struct epitem *epi; | |
1023 | ||
1024 | if (!is_file_epoll(file)) | |
1025 | return ERR_PTR(-EINVAL); | |
1026 | ||
1027 | ep = file->private_data; | |
1028 | ||
1029 | mutex_lock(&ep->mtx); | |
1030 | epi = ep_find_tfd(ep, tfd, toff); | |
1031 | if (epi) | |
1032 | file_raw = epi->ffd.file; | |
1033 | else | |
1034 | file_raw = ERR_PTR(-ENOENT); | |
1035 | mutex_unlock(&ep->mtx); | |
1036 | ||
1037 | return file_raw; | |
1038 | } | |
bfe3911a | 1039 | #endif /* CONFIG_KCMP */ |
0791e364 | 1040 | |
a6c67fee | 1041 | /* |
a218cc49 RP |
1042 | * Adds a new entry to the tail of the list in a lockless way, i.e. |
1043 | * multiple CPUs are allowed to call this function concurrently. | |
1044 | * | |
1045 | * Beware: it is necessary to prevent any other modifications of the | |
1046 | * existing list until all changes are completed, in other words | |
1047 | * concurrent list_add_tail_lockless() calls should be protected | |
1048 | * with a read lock, where write lock acts as a barrier which | |
1049 | * makes sure all list_add_tail_lockless() calls are fully | |
1050 | * completed. | |
1051 | * | |
1052 | * Also an element can be locklessly added to the list only in one | |
a6c67fee | 1053 | * direction i.e. either to the tail or to the head, otherwise |
a218cc49 RP |
1054 | * concurrent access will corrupt the list. |
1055 | * | |
a6c67fee | 1056 | * Return: %false if element has been already added to the list, %true |
a218cc49 RP |
1057 | * otherwise. |
1058 | */ | |
1059 | static inline bool list_add_tail_lockless(struct list_head *new, | |
1060 | struct list_head *head) | |
1061 | { | |
1062 | struct list_head *prev; | |
1063 | ||
1064 | /* | |
1065 | * This is simple 'new->next = head' operation, but cmpxchg() | |
1066 | * is used in order to detect that same element has been just | |
1067 | * added to the list from another CPU: the winner observes | |
1068 | * new->next == new. | |
1069 | */ | |
693fc06e | 1070 | if (!try_cmpxchg(&new->next, &new, head)) |
a218cc49 RP |
1071 | return false; |
1072 | ||
1073 | /* | |
1074 | * Initially ->next of a new element must be updated with the head | |
1075 | * (we are inserting to the tail) and only then pointers are atomically | |
1076 | * exchanged. XCHG guarantees memory ordering, thus ->next should be | |
1077 | * updated before pointers are actually swapped and pointers are | |
1078 | * swapped before prev->next is updated. | |
1079 | */ | |
1080 | ||
1081 | prev = xchg(&head->prev, new); | |
1082 | ||
1083 | /* | |
1084 | * It is safe to modify prev->next and new->prev, because a new element | |
1085 | * is added only to the tail and new->next is updated before XCHG. | |
1086 | */ | |
1087 | ||
1088 | prev->next = new; | |
1089 | new->prev = prev; | |
1090 | ||
1091 | return true; | |
1092 | } | |
1093 | ||
a6c67fee | 1094 | /* |
a218cc49 RP |
1095 | * Chains a new epi entry to the tail of the ep->ovflist in a lockless way, |
1096 | * i.e. multiple CPUs are allowed to call this function concurrently. | |
1097 | * | |
a6c67fee | 1098 | * Return: %false if epi element has been already chained, %true otherwise. |
a218cc49 RP |
1099 | */ |
1100 | static inline bool chain_epi_lockless(struct epitem *epi) | |
1101 | { | |
1102 | struct eventpoll *ep = epi->ep; | |
1103 | ||
0c54a6a4 KK |
1104 | /* Fast preliminary check */ |
1105 | if (epi->next != EP_UNACTIVE_PTR) | |
1106 | return false; | |
1107 | ||
a218cc49 RP |
1108 | /* Check that the same epi has not been just chained from another CPU */ |
1109 | if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR) | |
1110 | return false; | |
1111 | ||
1112 | /* Atomically exchange tail */ | |
1113 | epi->next = xchg(&ep->ovflist, epi); | |
1114 | ||
1115 | return true; | |
1116 | } | |
1117 | ||
1da177e4 | 1118 | /* |
7699acd1 | 1119 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 1120 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 1121 | * have events to report. |
a218cc49 | 1122 | * |
a6c67fee RD |
1123 | * This callback takes a read lock in order not to contend with concurrent |
1124 | * events from another file descriptor, thus all modifications to ->rdllist | |
a218cc49 RP |
1125 | * or ->ovflist are lockless. Read lock is paired with the write lock from |
1126 | * ep_scan_ready_list(), which stops all list modifications and guarantees | |
1127 | * that lists state is seen correctly. | |
1128 | * | |
1129 | * Another thing worth to mention is that ep_poll_callback() can be called | |
1130 | * concurrently for the same @epi from different CPUs if poll table was inited | |
1131 | * with several wait queues entries. Plural wakeup from different CPUs of a | |
1132 | * single wait queue is serialized by wq.lock, but the case when multiple wait | |
1133 | * queues are used should be detected accordingly. This is detected using | |
1134 | * cmpxchg() operation. | |
1da177e4 | 1135 | */ |
ac6424b9 | 1136 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 1137 | { |
7699acd1 | 1138 | int pwake = 0; |
7699acd1 DL |
1139 | struct epitem *epi = ep_item_from_wait(wait); |
1140 | struct eventpoll *ep = epi->ep; | |
3ad6f93e | 1141 | __poll_t pollflags = key_to_poll(key); |
a218cc49 | 1142 | unsigned long flags; |
df0108c5 | 1143 | int ewake = 0; |
1da177e4 | 1144 | |
a218cc49 | 1145 | read_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1146 | |
bf3b9f63 SS |
1147 | ep_set_busy_poll_napi_id(epi); |
1148 | ||
7699acd1 DL |
1149 | /* |
1150 | * If the event mask does not contain any poll(2) event, we consider the | |
1151 | * descriptor to be disabled. This condition is likely the effect of the | |
1152 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1153 | * until the next EPOLL_CTL_MOD will be issued. | |
1154 | */ | |
1155 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
1156 | goto out_unlock; |
1157 | ||
2dfa4eea DL |
1158 | /* |
1159 | * Check the events coming with the callback. At this stage, not | |
1160 | * every device reports the events in the "key" parameter of the | |
1161 | * callback. We need to be able to handle both cases here, hence the | |
1162 | * test for "key" != NULL before the event match test. | |
1163 | */ | |
3ad6f93e | 1164 | if (pollflags && !(pollflags & epi->event.events)) |
2dfa4eea DL |
1165 | goto out_unlock; |
1166 | ||
d47de16c | 1167 | /* |
bf6a41db | 1168 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 1169 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 1170 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
1171 | * chained in ep->ovflist and requeued later on. |
1172 | */ | |
c5a282e9 | 1173 | if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) { |
0c54a6a4 KK |
1174 | if (chain_epi_lockless(epi)) |
1175 | ep_pm_stay_awake_rcu(epi); | |
1176 | } else if (!ep_is_linked(epi)) { | |
1177 | /* In the usual case, add event to ready list. */ | |
1178 | if (list_add_tail_lockless(&epi->rdllink, &ep->rdllist)) | |
c3e320b6 | 1179 | ep_pm_stay_awake_rcu(epi); |
4d7e30d9 | 1180 | } |
7699acd1 | 1181 | |
7699acd1 DL |
1182 | /* |
1183 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1184 | * wait list. | |
1185 | */ | |
df0108c5 | 1186 | if (waitqueue_active(&ep->wq)) { |
b6a515c8 | 1187 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
3ad6f93e AV |
1188 | !(pollflags & POLLFREE)) { |
1189 | switch (pollflags & EPOLLINOUT_BITS) { | |
a9a08845 LT |
1190 | case EPOLLIN: |
1191 | if (epi->event.events & EPOLLIN) | |
b6a515c8 JB |
1192 | ewake = 1; |
1193 | break; | |
a9a08845 LT |
1194 | case EPOLLOUT: |
1195 | if (epi->event.events & EPOLLOUT) | |
b6a515c8 JB |
1196 | ewake = 1; |
1197 | break; | |
1198 | case 0: | |
1199 | ewake = 1; | |
1200 | break; | |
1201 | } | |
1202 | } | |
a218cc49 | 1203 | wake_up(&ep->wq); |
df0108c5 | 1204 | } |
7699acd1 DL |
1205 | if (waitqueue_active(&ep->poll_wait)) |
1206 | pwake++; | |
1207 | ||
d47de16c | 1208 | out_unlock: |
a218cc49 | 1209 | read_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1210 | |
7699acd1 DL |
1211 | /* We have to call this outside the lock */ |
1212 | if (pwake) | |
caf1aeaf | 1213 | ep_poll_safewake(ep, epi, pollflags & EPOLL_URING_WAKE); |
7699acd1 | 1214 | |
138e4ad6 ON |
1215 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
1216 | ewake = 1; | |
1217 | ||
3ad6f93e | 1218 | if (pollflags & POLLFREE) { |
138e4ad6 ON |
1219 | /* |
1220 | * If we race with ep_remove_wait_queue() it can miss | |
1221 | * ->whead = NULL and do another remove_wait_queue() after | |
1222 | * us, so we can't use __remove_wait_queue(). | |
1223 | */ | |
1224 | list_del_init(&wait->entry); | |
1225 | /* | |
1226 | * ->whead != NULL protects us from the race with ep_free() | |
1227 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock | |
1228 | * held by the caller. Once we nullify it, nothing protects | |
1229 | * ep/epi or even wait. | |
1230 | */ | |
1231 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); | |
1232 | } | |
df0108c5 | 1233 | |
138e4ad6 | 1234 | return ewake; |
7699acd1 | 1235 | } |
1da177e4 LT |
1236 | |
1237 | /* | |
1238 | * This is the callback that is used to add our wait queue to the | |
1239 | * target file wakeup lists. | |
1240 | */ | |
1241 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1242 | poll_table *pt) | |
1243 | { | |
364f374f AV |
1244 | struct ep_pqueue *epq = container_of(pt, struct ep_pqueue, pt); |
1245 | struct epitem *epi = epq->epi; | |
1da177e4 LT |
1246 | struct eppoll_entry *pwq; |
1247 | ||
364f374f AV |
1248 | if (unlikely(!epi)) // an earlier allocation has failed |
1249 | return; | |
1250 | ||
1251 | pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL); | |
1252 | if (unlikely(!pwq)) { | |
1253 | epq->epi = NULL; | |
1254 | return; | |
296e236e | 1255 | } |
364f374f AV |
1256 | |
1257 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); | |
1258 | pwq->whead = whead; | |
1259 | pwq->base = epi; | |
1260 | if (epi->event.events & EPOLLEXCLUSIVE) | |
1261 | add_wait_queue_exclusive(whead, &pwq->wait); | |
1262 | else | |
1263 | add_wait_queue(whead, &pwq->wait); | |
1264 | pwq->next = epi->pwqlist; | |
1265 | epi->pwqlist = pwq; | |
1da177e4 LT |
1266 | } |
1267 | ||
1da177e4 LT |
1268 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1269 | { | |
1270 | int kcmp; | |
b2ac2ea6 | 1271 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
1da177e4 | 1272 | struct epitem *epic; |
b2ac2ea6 | 1273 | bool leftmost = true; |
1da177e4 LT |
1274 | |
1275 | while (*p) { | |
1276 | parent = *p; | |
1277 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1278 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
b2ac2ea6 | 1279 | if (kcmp > 0) { |
1da177e4 | 1280 | p = &parent->rb_right; |
b2ac2ea6 DB |
1281 | leftmost = false; |
1282 | } else | |
1da177e4 LT |
1283 | p = &parent->rb_left; |
1284 | } | |
1285 | rb_link_node(&epi->rbn, parent, p); | |
b2ac2ea6 | 1286 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
1da177e4 LT |
1287 | } |
1288 | ||
a80a6b85 AM |
1289 | |
1290 | ||
28d82dc1 JB |
1291 | #define PATH_ARR_SIZE 5 |
1292 | /* | |
1293 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1294 | * from a single file of interest. For example, we allow 1000 paths of length | |
1295 | * 1, to emanate from each file of interest. This essentially represents the | |
1296 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1297 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1298 | * is connected to n file sources. In this case each file source has 1 path | |
1299 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1300 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1301 | * and delete can't add additional paths. Protected by the epmutex. | |
1302 | */ | |
1303 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1304 | static int path_count[PATH_ARR_SIZE]; | |
1305 | ||
1306 | static int path_count_inc(int nests) | |
1307 | { | |
93dc6107 JB |
1308 | /* Allow an arbitrary number of depth 1 paths */ |
1309 | if (nests == 0) | |
1310 | return 0; | |
1311 | ||
28d82dc1 JB |
1312 | if (++path_count[nests] > path_limits[nests]) |
1313 | return -1; | |
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | static void path_count_init(void) | |
1318 | { | |
1319 | int i; | |
1320 | ||
1321 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1322 | path_count[i] = 0; | |
1323 | } | |
1324 | ||
319c1517 | 1325 | static int reverse_path_check_proc(struct hlist_head *refs, int depth) |
28d82dc1 JB |
1326 | { |
1327 | int error = 0; | |
28d82dc1 JB |
1328 | struct epitem *epi; |
1329 | ||
0c320f77 | 1330 | if (depth > EP_MAX_NESTS) /* too deep nesting */ |
99d84d43 AV |
1331 | return -1; |
1332 | ||
ae10b2b4 | 1333 | /* CTL_DEL can remove links here, but that can't increase our count */ |
319c1517 AV |
1334 | hlist_for_each_entry_rcu(epi, refs, fllink) { |
1335 | struct hlist_head *refs = &epi->ep->refs; | |
1336 | if (hlist_empty(refs)) | |
d16312a4 AV |
1337 | error = path_count_inc(depth); |
1338 | else | |
319c1517 | 1339 | error = reverse_path_check_proc(refs, depth + 1); |
d16312a4 AV |
1340 | if (error != 0) |
1341 | break; | |
28d82dc1 JB |
1342 | } |
1343 | return error; | |
1344 | } | |
1345 | ||
1346 | /** | |
319c1517 | 1347 | * reverse_path_check - The tfile_check_list is list of epitem_head, which have |
28d82dc1 JB |
1348 | * links that are proposed to be newly added. We need to |
1349 | * make sure that those added links don't add too many | |
1350 | * paths such that we will spend all our time waking up | |
1351 | * eventpoll objects. | |
1352 | * | |
a6c67fee RD |
1353 | * Return: %zero if the proposed links don't create too many paths, |
1354 | * %-1 otherwise. | |
28d82dc1 JB |
1355 | */ |
1356 | static int reverse_path_check(void) | |
1357 | { | |
319c1517 | 1358 | struct epitems_head *p; |
28d82dc1 | 1359 | |
319c1517 AV |
1360 | for (p = tfile_check_list; p != EP_UNACTIVE_PTR; p = p->next) { |
1361 | int error; | |
28d82dc1 | 1362 | path_count_init(); |
b62d2706 | 1363 | rcu_read_lock(); |
319c1517 | 1364 | error = reverse_path_check_proc(&p->epitems, 0); |
b62d2706 | 1365 | rcu_read_unlock(); |
28d82dc1 | 1366 | if (error) |
319c1517 | 1367 | return error; |
28d82dc1 | 1368 | } |
319c1517 | 1369 | return 0; |
28d82dc1 JB |
1370 | } |
1371 | ||
4d7e30d9 AH |
1372 | static int ep_create_wakeup_source(struct epitem *epi) |
1373 | { | |
3701cb59 | 1374 | struct name_snapshot n; |
eea1d585 | 1375 | struct wakeup_source *ws; |
4d7e30d9 AH |
1376 | |
1377 | if (!epi->ep->ws) { | |
c8377adf | 1378 | epi->ep->ws = wakeup_source_register(NULL, "eventpoll"); |
4d7e30d9 AH |
1379 | if (!epi->ep->ws) |
1380 | return -ENOMEM; | |
1381 | } | |
1382 | ||
3701cb59 AV |
1383 | take_dentry_name_snapshot(&n, epi->ffd.file->f_path.dentry); |
1384 | ws = wakeup_source_register(NULL, n.name.name); | |
1385 | release_dentry_name_snapshot(&n); | |
eea1d585 EW |
1386 | |
1387 | if (!ws) | |
4d7e30d9 | 1388 | return -ENOMEM; |
eea1d585 | 1389 | rcu_assign_pointer(epi->ws, ws); |
4d7e30d9 AH |
1390 | |
1391 | return 0; | |
1392 | } | |
1393 | ||
eea1d585 EW |
1394 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
1395 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) | |
4d7e30d9 | 1396 | { |
eea1d585 EW |
1397 | struct wakeup_source *ws = ep_wakeup_source(epi); |
1398 | ||
d6d67e72 | 1399 | RCU_INIT_POINTER(epi->ws, NULL); |
eea1d585 EW |
1400 | |
1401 | /* | |
1402 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is | |
1403 | * used internally by wakeup_source_remove, too (called by | |
1404 | * wakeup_source_unregister), so we cannot use call_rcu | |
1405 | */ | |
1406 | synchronize_rcu(); | |
1407 | wakeup_source_unregister(ws); | |
4d7e30d9 AH |
1408 | } |
1409 | ||
319c1517 AV |
1410 | static int attach_epitem(struct file *file, struct epitem *epi) |
1411 | { | |
1412 | struct epitems_head *to_free = NULL; | |
1413 | struct hlist_head *head = NULL; | |
1414 | struct eventpoll *ep = NULL; | |
1415 | ||
1416 | if (is_file_epoll(file)) | |
1417 | ep = file->private_data; | |
1418 | ||
1419 | if (ep) { | |
1420 | head = &ep->refs; | |
1421 | } else if (!READ_ONCE(file->f_ep)) { | |
1422 | allocate: | |
1423 | to_free = kmem_cache_zalloc(ephead_cache, GFP_KERNEL); | |
1424 | if (!to_free) | |
1425 | return -ENOMEM; | |
1426 | head = &to_free->epitems; | |
1427 | } | |
1428 | spin_lock(&file->f_lock); | |
1429 | if (!file->f_ep) { | |
1430 | if (unlikely(!head)) { | |
1431 | spin_unlock(&file->f_lock); | |
1432 | goto allocate; | |
1433 | } | |
1434 | file->f_ep = head; | |
1435 | to_free = NULL; | |
1436 | } | |
1437 | hlist_add_head_rcu(&epi->fllink, file->f_ep); | |
1438 | spin_unlock(&file->f_lock); | |
1439 | free_ephead(to_free); | |
1440 | return 0; | |
1441 | } | |
1442 | ||
c7ea7630 DL |
1443 | /* |
1444 | * Must be called with "mtx" held. | |
1445 | */ | |
bec1a502 | 1446 | static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, |
67347fe4 | 1447 | struct file *tfile, int fd, int full_check) |
1da177e4 | 1448 | { |
d85e2aa2 AV |
1449 | int error, pwake = 0; |
1450 | __poll_t revents; | |
1da177e4 LT |
1451 | struct epitem *epi; |
1452 | struct ep_pqueue epq; | |
85353e91 AV |
1453 | struct eventpoll *tep = NULL; |
1454 | ||
1455 | if (is_file_epoll(tfile)) | |
1456 | tep = tfile->private_data; | |
1da177e4 | 1457 | |
92e64178 DB |
1458 | lockdep_assert_irqs_enabled(); |
1459 | ||
1e1c1583 NP |
1460 | if (unlikely(percpu_counter_compare(&ep->user->epoll_watches, |
1461 | max_user_watches) >= 0)) | |
7ef9964e | 1462 | return -ENOSPC; |
1e1c1583 NP |
1463 | percpu_counter_inc(&ep->user->epoll_watches); |
1464 | ||
1465 | if (!(epi = kmem_cache_zalloc(epi_cache, GFP_KERNEL))) { | |
1466 | percpu_counter_dec(&ep->user->epoll_watches); | |
7ef9964e | 1467 | return -ENOMEM; |
1e1c1583 | 1468 | } |
1da177e4 LT |
1469 | |
1470 | /* Item initialization follow here ... */ | |
1da177e4 | 1471 | INIT_LIST_HEAD(&epi->rdllink); |
1da177e4 | 1472 | epi->ep = ep; |
b030a4dd | 1473 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1474 | epi->event = *event; |
d47de16c | 1475 | epi->next = EP_UNACTIVE_PTR; |
1da177e4 | 1476 | |
85353e91 AV |
1477 | if (tep) |
1478 | mutex_lock_nested(&tep->mtx, 1); | |
f8d4f44d | 1479 | /* Add the current item to the list of active epoll hook for this file */ |
319c1517 | 1480 | if (unlikely(attach_epitem(tfile, epi) < 0)) { |
319c1517 AV |
1481 | if (tep) |
1482 | mutex_unlock(&tep->mtx); | |
1e1c1583 NP |
1483 | kmem_cache_free(epi_cache, epi); |
1484 | percpu_counter_dec(&ep->user->epoll_watches); | |
319c1517 | 1485 | return -ENOMEM; |
d9f41e3c | 1486 | } |
f8d4f44d | 1487 | |
319c1517 AV |
1488 | if (full_check && !tep) |
1489 | list_file(tfile); | |
1490 | ||
f8d4f44d AV |
1491 | /* |
1492 | * Add the current item to the RB tree. All RB tree operations are | |
1493 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1494 | */ | |
1495 | ep_rbtree_insert(ep, epi); | |
85353e91 AV |
1496 | if (tep) |
1497 | mutex_unlock(&tep->mtx); | |
f8d4f44d AV |
1498 | |
1499 | /* now check if we've created too many backpaths */ | |
e3e096e7 AV |
1500 | if (unlikely(full_check && reverse_path_check())) { |
1501 | ep_remove(ep, epi); | |
1502 | return -EINVAL; | |
1503 | } | |
f8d4f44d | 1504 | |
d1ec50ad AV |
1505 | if (epi->event.events & EPOLLWAKEUP) { |
1506 | error = ep_create_wakeup_source(epi); | |
1507 | if (error) { | |
1508 | ep_remove(ep, epi); | |
1509 | return error; | |
1510 | } | |
1511 | } | |
f8d4f44d | 1512 | |
1da177e4 LT |
1513 | /* Initialize the poll table using the queue callback */ |
1514 | epq.epi = epi; | |
1515 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
1516 | ||
1517 | /* | |
1518 | * Attach the item to the poll hooks and get current event bits. | |
1519 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1520 | * been increased by the caller of this function. Note that after |
1521 | * this operation completes, the poll callback can start hitting | |
1522 | * the new item. | |
1da177e4 | 1523 | */ |
37b5e521 | 1524 | revents = ep_item_poll(epi, &epq.pt, 1); |
1da177e4 LT |
1525 | |
1526 | /* | |
1527 | * We have to check if something went wrong during the poll wait queue | |
1528 | * install process. Namely an allocation for a wait queue failed due | |
1529 | * high memory pressure. | |
1530 | */ | |
e3e096e7 AV |
1531 | if (unlikely(!epq.epi)) { |
1532 | ep_remove(ep, epi); | |
1533 | return -ENOMEM; | |
1534 | } | |
1da177e4 | 1535 | |
c7ea7630 | 1536 | /* We have to drop the new item inside our item list to keep track of it */ |
a218cc49 | 1537 | write_lock_irq(&ep->lock); |
c7ea7630 | 1538 | |
bf3b9f63 SS |
1539 | /* record NAPI ID of new item if present */ |
1540 | ep_set_busy_poll_napi_id(epi); | |
1541 | ||
1da177e4 | 1542 | /* If the file is already "ready" we drop it inside the ready list */ |
992991c0 | 1543 | if (revents && !ep_is_linked(epi)) { |
1da177e4 | 1544 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1545 | ep_pm_stay_awake(epi); |
1da177e4 LT |
1546 | |
1547 | /* Notify waiting tasks that events are available */ | |
1548 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1549 | wake_up(&ep->wq); |
1da177e4 LT |
1550 | if (waitqueue_active(&ep->poll_wait)) |
1551 | pwake++; | |
1552 | } | |
1553 | ||
a218cc49 | 1554 | write_unlock_irq(&ep->lock); |
1da177e4 LT |
1555 | |
1556 | /* We have to call this outside the lock */ | |
1557 | if (pwake) | |
caf1aeaf | 1558 | ep_poll_safewake(ep, NULL, 0); |
1da177e4 | 1559 | |
1da177e4 | 1560 | return 0; |
1da177e4 LT |
1561 | } |
1562 | ||
1da177e4 LT |
1563 | /* |
1564 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1565 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 | 1566 | */ |
bec1a502 AV |
1567 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, |
1568 | const struct epoll_event *event) | |
1da177e4 LT |
1569 | { |
1570 | int pwake = 0; | |
626cf236 HV |
1571 | poll_table pt; |
1572 | ||
92e64178 DB |
1573 | lockdep_assert_irqs_enabled(); |
1574 | ||
626cf236 | 1575 | init_poll_funcptr(&pt, NULL); |
1da177e4 LT |
1576 | |
1577 | /* | |
e057e15f TB |
1578 | * Set the new event interest mask before calling f_op->poll(); |
1579 | * otherwise we might miss an event that happens between the | |
1580 | * f_op->poll() call and the new event set registering. | |
1da177e4 | 1581 | */ |
128dd175 | 1582 | epi->event.events = event->events; /* need barrier below */ |
e057e15f | 1583 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 | 1584 | if (epi->event.events & EPOLLWAKEUP) { |
eea1d585 | 1585 | if (!ep_has_wakeup_source(epi)) |
4d7e30d9 | 1586 | ep_create_wakeup_source(epi); |
eea1d585 | 1587 | } else if (ep_has_wakeup_source(epi)) { |
4d7e30d9 AH |
1588 | ep_destroy_wakeup_source(epi); |
1589 | } | |
1da177e4 | 1590 | |
128dd175 EW |
1591 | /* |
1592 | * The following barrier has two effects: | |
1593 | * | |
1594 | * 1) Flush epi changes above to other CPUs. This ensures | |
1595 | * we do not miss events from ep_poll_callback if an | |
1596 | * event occurs immediately after we call f_op->poll(). | |
a218cc49 | 1597 | * We need this because we did not take ep->lock while |
128dd175 | 1598 | * changing epi above (but ep_poll_callback does take |
a218cc49 | 1599 | * ep->lock). |
128dd175 EW |
1600 | * |
1601 | * 2) We also need to ensure we do not miss _past_ events | |
1602 | * when calling f_op->poll(). This barrier also | |
1603 | * pairs with the barrier in wq_has_sleeper (see | |
1604 | * comments for wq_has_sleeper). | |
1605 | * | |
1606 | * This barrier will now guarantee ep_poll_callback or f_op->poll | |
1607 | * (or both) will notice the readiness of an item. | |
1608 | */ | |
1609 | smp_mb(); | |
1610 | ||
1da177e4 LT |
1611 | /* |
1612 | * Get current event bits. We can safely use the file* here because | |
1613 | * its usage count has been increased by the caller of this function. | |
c7ea7630 | 1614 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1615 | * list, push it inside. |
1da177e4 | 1616 | */ |
69112736 | 1617 | if (ep_item_poll(epi, &pt, 1)) { |
a218cc49 | 1618 | write_lock_irq(&ep->lock); |
992991c0 | 1619 | if (!ep_is_linked(epi)) { |
c7ea7630 | 1620 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1621 | ep_pm_stay_awake(epi); |
c7ea7630 DL |
1622 | |
1623 | /* Notify waiting tasks that events are available */ | |
1624 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1625 | wake_up(&ep->wq); |
c7ea7630 DL |
1626 | if (waitqueue_active(&ep->poll_wait)) |
1627 | pwake++; | |
7699acd1 | 1628 | } |
a218cc49 | 1629 | write_unlock_irq(&ep->lock); |
7699acd1 | 1630 | } |
1da177e4 | 1631 | |
7699acd1 DL |
1632 | /* We have to call this outside the lock */ |
1633 | if (pwake) | |
caf1aeaf | 1634 | ep_poll_safewake(ep, NULL, 0); |
1da177e4 | 1635 | |
7699acd1 | 1636 | return 0; |
1da177e4 LT |
1637 | } |
1638 | ||
ff07952a AV |
1639 | static int ep_send_events(struct eventpoll *ep, |
1640 | struct epoll_event __user *events, int maxevents) | |
1da177e4 | 1641 | { |
4e0982a0 | 1642 | struct epitem *epi, *tmp; |
ff07952a | 1643 | LIST_HEAD(txlist); |
626cf236 | 1644 | poll_table pt; |
ff07952a | 1645 | int res = 0; |
626cf236 | 1646 | |
cccd29bf SHY |
1647 | /* |
1648 | * Always short-circuit for fatal signals to allow threads to make a | |
1649 | * timely exit without the chance of finding more events available and | |
1650 | * fetching repeatedly. | |
1651 | */ | |
1652 | if (fatal_signal_pending(current)) | |
1653 | return -EINTR; | |
1654 | ||
626cf236 | 1655 | init_poll_funcptr(&pt, NULL); |
ff07952a | 1656 | |
57804b1c AV |
1657 | mutex_lock(&ep->mtx); |
1658 | ep_start_scan(ep, &txlist); | |
1da177e4 | 1659 | |
296e236e | 1660 | /* |
5071f97e | 1661 | * We can loop without lock because we are passed a task private list. |
57804b1c | 1662 | * Items cannot vanish during the loop we are holding ep->mtx. |
296e236e | 1663 | */ |
ff07952a AV |
1664 | list_for_each_entry_safe(epi, tmp, &txlist, rdllink) { |
1665 | struct wakeup_source *ws; | |
1666 | __poll_t revents; | |
21877e1a | 1667 | |
ff07952a | 1668 | if (res >= maxevents) |
4e0982a0 | 1669 | break; |
d47de16c | 1670 | |
4d7e30d9 AH |
1671 | /* |
1672 | * Activate ep->ws before deactivating epi->ws to prevent | |
1673 | * triggering auto-suspend here (in case we reactive epi->ws | |
1674 | * below). | |
1675 | * | |
1676 | * This could be rearranged to delay the deactivation of epi->ws | |
1677 | * instead, but then epi->ws would temporarily be out of sync | |
1678 | * with ep_is_linked(). | |
1679 | */ | |
eea1d585 EW |
1680 | ws = ep_wakeup_source(epi); |
1681 | if (ws) { | |
1682 | if (ws->active) | |
1683 | __pm_stay_awake(ep->ws); | |
1684 | __pm_relax(ws); | |
1685 | } | |
1686 | ||
d47de16c | 1687 | list_del_init(&epi->rdllink); |
1da177e4 | 1688 | |
296e236e | 1689 | /* |
5071f97e | 1690 | * If the event mask intersect the caller-requested one, |
57804b1c AV |
1691 | * deliver the event to userspace. Again, we are holding ep->mtx, |
1692 | * so no operations coming from userspace can change the item. | |
296e236e | 1693 | */ |
4e0982a0 DB |
1694 | revents = ep_item_poll(epi, &pt, 1); |
1695 | if (!revents) | |
1696 | continue; | |
1697 | ||
249dbe74 AB |
1698 | events = epoll_put_uevent(revents, epi->event.data, events); |
1699 | if (!events) { | |
ff07952a | 1700 | list_add(&epi->rdllink, &txlist); |
4e0982a0 | 1701 | ep_pm_stay_awake(epi); |
ff07952a AV |
1702 | if (!res) |
1703 | res = -EFAULT; | |
1704 | break; | |
4e0982a0 | 1705 | } |
ff07952a | 1706 | res++; |
4e0982a0 DB |
1707 | if (epi->event.events & EPOLLONESHOT) |
1708 | epi->event.events &= EP_PRIVATE_BITS; | |
1709 | else if (!(epi->event.events & EPOLLET)) { | |
1710 | /* | |
1711 | * If this file has been added with Level | |
1712 | * Trigger mode, we need to insert back inside | |
1713 | * the ready list, so that the next call to | |
1714 | * epoll_wait() will check again the events | |
1715 | * availability. At this point, no one can insert | |
1716 | * into ep->rdllist besides us. The epoll_ctl() | |
1717 | * callers are locked out by | |
1718 | * ep_scan_ready_list() holding "mtx" and the | |
1719 | * poll callback will queue them in ep->ovflist. | |
1720 | */ | |
1721 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1722 | ep_pm_stay_awake(epi); | |
296e236e DL |
1723 | } |
1724 | } | |
57804b1c AV |
1725 | ep_done_scan(ep, &txlist); |
1726 | mutex_unlock(&ep->mtx); | |
5071f97e | 1727 | |
ff07952a | 1728 | return res; |
1da177e4 LT |
1729 | } |
1730 | ||
7cdf7c20 | 1731 | static struct timespec64 *ep_timeout_to_timespec(struct timespec64 *to, long ms) |
0781b909 | 1732 | { |
7cdf7c20 WB |
1733 | struct timespec64 now; |
1734 | ||
1735 | if (ms < 0) | |
1736 | return NULL; | |
1737 | ||
1738 | if (!ms) { | |
1739 | to->tv_sec = 0; | |
1740 | to->tv_nsec = 0; | |
1741 | return to; | |
1742 | } | |
1743 | ||
1744 | to->tv_sec = ms / MSEC_PER_SEC; | |
1745 | to->tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC); | |
0781b909 | 1746 | |
766b9f92 | 1747 | ktime_get_ts64(&now); |
7cdf7c20 WB |
1748 | *to = timespec64_add_safe(now, *to); |
1749 | return to; | |
0781b909 ED |
1750 | } |
1751 | ||
a16ceb13 BS |
1752 | /* |
1753 | * autoremove_wake_function, but remove even on failure to wake up, because we | |
1754 | * know that default_wake_function/ttwu will only fail if the thread is already | |
1755 | * woken, and in that case the ep_poll loop will remove the entry anyways, not | |
1756 | * try to reuse it. | |
1757 | */ | |
1758 | static int ep_autoremove_wake_function(struct wait_queue_entry *wq_entry, | |
1759 | unsigned int mode, int sync, void *key) | |
1760 | { | |
1761 | int ret = default_wake_function(wq_entry, mode, sync, key); | |
1762 | ||
1763 | list_del_init(&wq_entry->entry); | |
1764 | return ret; | |
1765 | } | |
1766 | ||
f4d93ad7 | 1767 | /** |
a6c67fee | 1768 | * ep_poll - Retrieves ready events, and delivers them to the caller-supplied |
f4d93ad7 SB |
1769 | * event buffer. |
1770 | * | |
1771 | * @ep: Pointer to the eventpoll context. | |
1772 | * @events: Pointer to the userspace buffer where the ready events should be | |
1773 | * stored. | |
1774 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1775 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
7cdf7c20 WB |
1776 | * timespec. If the timeout is zero, the function will not block, |
1777 | * while if the @timeout ptr is NULL, the function will block | |
f4d93ad7 SB |
1778 | * until at least one event has been retrieved (or an error |
1779 | * occurred). | |
1780 | * | |
a6c67fee | 1781 | * Return: the number of ready events which have been fetched, or an |
f4d93ad7 SB |
1782 | * error code, in case of error. |
1783 | */ | |
1da177e4 | 1784 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
7cdf7c20 | 1785 | int maxevents, struct timespec64 *timeout) |
1da177e4 | 1786 | { |
e59d3c64 | 1787 | int res, eavail, timed_out = 0; |
da8b44d5 | 1788 | u64 slack = 0; |
ac6424b9 | 1789 | wait_queue_entry_t wait; |
95aac7b1 SB |
1790 | ktime_t expires, *to = NULL; |
1791 | ||
679abf38 DB |
1792 | lockdep_assert_irqs_enabled(); |
1793 | ||
7cdf7c20 WB |
1794 | if (timeout && (timeout->tv_sec | timeout->tv_nsec)) { |
1795 | slack = select_estimate_accuracy(timeout); | |
95aac7b1 | 1796 | to = &expires; |
7cdf7c20 WB |
1797 | *to = timespec64_to_ktime(*timeout); |
1798 | } else if (timeout) { | |
f4d93ad7 SB |
1799 | /* |
1800 | * Avoid the unnecessary trip to the wait queue loop, if the | |
e59d3c64 | 1801 | * caller specified a non blocking operation. |
f4d93ad7 | 1802 | */ |
95aac7b1 SB |
1803 | timed_out = 1; |
1804 | } | |
1da177e4 | 1805 | |
e59d3c64 SHY |
1806 | /* |
1807 | * This call is racy: We may or may not see events that are being added | |
a6c67fee | 1808 | * to the ready list under the lock (e.g., in IRQ callbacks). For cases |
e59d3c64 | 1809 | * with a non-zero timeout, this thread will check the ready list under |
a6c67fee | 1810 | * lock and will add to the wait queue. For cases with a zero |
e59d3c64 SHY |
1811 | * timeout, the user by definition should not care and will have to |
1812 | * recheck again. | |
1813 | */ | |
1814 | eavail = ep_events_available(ep); | |
1815 | ||
00b27634 SHY |
1816 | while (1) { |
1817 | if (eavail) { | |
1818 | /* | |
1819 | * Try to transfer events to user space. In case we get | |
1820 | * 0 events and there's still timeout left over, we go | |
1821 | * trying again in search of more luck. | |
1822 | */ | |
1823 | res = ep_send_events(ep, events, maxevents); | |
1824 | if (res) | |
1825 | return res; | |
1826 | } | |
1827 | ||
1828 | if (timed_out) | |
1829 | return 0; | |
1830 | ||
00b27634 | 1831 | eavail = ep_busy_loop(ep, timed_out); |
e8c85328 | 1832 | if (eavail) |
00b27634 | 1833 | continue; |
1da177e4 | 1834 | |
2efdaf76 SHY |
1835 | if (signal_pending(current)) |
1836 | return -EINTR; | |
1837 | ||
412895f0 RP |
1838 | /* |
1839 | * Internally init_wait() uses autoremove_wake_function(), | |
1840 | * thus wait entry is removed from the wait queue on each | |
1841 | * wakeup. Why it is important? In case of several waiters | |
1842 | * each new wakeup will hit the next waiter, giving it the | |
1843 | * chance to harvest new event. Otherwise wakeup can be | |
1844 | * lost. This is also good performance-wise, because on | |
1845 | * normal wakeup path no need to call __remove_wait_queue() | |
1846 | * explicitly, thus ep->lock is not taken, which halts the | |
1847 | * event delivery. | |
a16ceb13 BS |
1848 | * |
1849 | * In fact, we now use an even more aggressive function that | |
1850 | * unconditionally removes, because we don't reuse the wait | |
1851 | * entry between loop iterations. This lets us also avoid the | |
1852 | * performance issue if a process is killed, causing all of its | |
1853 | * threads to wake up without being removed normally. | |
412895f0 RP |
1854 | */ |
1855 | init_wait(&wait); | |
a16ceb13 | 1856 | wait.func = ep_autoremove_wake_function; |
1da177e4 | 1857 | |
65759097 | 1858 | write_lock_irq(&ep->lock); |
bf3b9f63 | 1859 | /* |
65759097 RP |
1860 | * Barrierless variant, waitqueue_active() is called under |
1861 | * the same lock on wakeup ep_poll_callback() side, so it | |
1862 | * is safe to avoid an explicit barrier. | |
bf3b9f63 | 1863 | */ |
65759097 RP |
1864 | __set_current_state(TASK_INTERRUPTIBLE); |
1865 | ||
1da177e4 | 1866 | /* |
65759097 RP |
1867 | * Do the final check under the lock. ep_scan_ready_list() |
1868 | * plays with two lists (->rdllist and ->ovflist) and there | |
1869 | * is always a race when both lists are empty for short | |
1870 | * period of time although events are pending, so lock is | |
1871 | * important. | |
1da177e4 | 1872 | */ |
65759097 | 1873 | eavail = ep_events_available(ep); |
2efdaf76 SHY |
1874 | if (!eavail) |
1875 | __add_wait_queue_exclusive(&ep->wq, &wait); | |
1876 | ||
65759097 | 1877 | write_unlock_irq(&ep->lock); |
95aac7b1 | 1878 | |
2efdaf76 | 1879 | if (!eavail) |
289caf5d SHY |
1880 | timed_out = !schedule_hrtimeout_range(to, slack, |
1881 | HRTIMER_MODE_ABS); | |
e411596d | 1882 | __set_current_state(TASK_RUNNING); |
1da177e4 | 1883 | |
289caf5d SHY |
1884 | /* |
1885 | * We were woken up, thus go and try to harvest some events. | |
1886 | * If timed out and still on the wait queue, recheck eavail | |
1887 | * carefully under lock, below. | |
1888 | */ | |
412895f0 | 1889 | eavail = 1; |
1da177e4 | 1890 | |
e8c85328 SHY |
1891 | if (!list_empty_careful(&wait.entry)) { |
1892 | write_lock_irq(&ep->lock); | |
1893 | /* | |
1894 | * If the thread timed out and is not on the wait queue, | |
1895 | * it means that the thread was woken up after its | |
1896 | * timeout expired before it could reacquire the lock. | |
1897 | * Thus, when wait.entry is empty, it needs to harvest | |
1898 | * events. | |
1899 | */ | |
1900 | if (timed_out) | |
1901 | eavail = list_empty(&wait.entry); | |
1902 | __remove_wait_queue(&ep->wq, &wait); | |
1903 | write_unlock_irq(&ep->lock); | |
1904 | } | |
00b27634 | 1905 | } |
1da177e4 LT |
1906 | } |
1907 | ||
22bacca4 | 1908 | /** |
773318ed | 1909 | * ep_loop_check_proc - verify that adding an epoll file inside another |
a6c67fee | 1910 | * epoll structure does not violate the constraints, in |
22bacca4 DL |
1911 | * terms of closed loops, or too deep chains (which can |
1912 | * result in excessive stack usage). | |
1913 | * | |
a6c67fee | 1914 | * @ep: the &struct eventpoll to be currently checked. |
bde03c4c | 1915 | * @depth: Current depth of the path being checked. |
22bacca4 | 1916 | * |
a6c67fee RD |
1917 | * Return: %zero if adding the epoll @file inside current epoll |
1918 | * structure @ep does not violate the constraints, or %-1 otherwise. | |
22bacca4 | 1919 | */ |
bde03c4c | 1920 | static int ep_loop_check_proc(struct eventpoll *ep, int depth) |
22bacca4 DL |
1921 | { |
1922 | int error = 0; | |
22bacca4 DL |
1923 | struct rb_node *rbp; |
1924 | struct epitem *epi; | |
1925 | ||
773318ed | 1926 | mutex_lock_nested(&ep->mtx, depth + 1); |
18306c40 | 1927 | ep->gen = loop_check_gen; |
b2ac2ea6 | 1928 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
22bacca4 DL |
1929 | epi = rb_entry(rbp, struct epitem, rbn); |
1930 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
bde03c4c | 1931 | struct eventpoll *ep_tovisit; |
28d82dc1 | 1932 | ep_tovisit = epi->ffd.file->private_data; |
18306c40 | 1933 | if (ep_tovisit->gen == loop_check_gen) |
28d82dc1 | 1934 | continue; |
bde03c4c | 1935 | if (ep_tovisit == inserting_into || depth > EP_MAX_NESTS) |
56c428ca | 1936 | error = -1; |
bde03c4c AV |
1937 | else |
1938 | error = ep_loop_check_proc(ep_tovisit, depth + 1); | |
22bacca4 DL |
1939 | if (error != 0) |
1940 | break; | |
28d82dc1 JB |
1941 | } else { |
1942 | /* | |
1943 | * If we've reached a file that is not associated with | |
1944 | * an ep, then we need to check if the newly added | |
1945 | * links are going to add too many wakeup paths. We do | |
1946 | * this by adding it to the tfile_check_list, if it's | |
1947 | * not already there, and calling reverse_path_check() | |
1948 | * during ep_insert(). | |
1949 | */ | |
319c1517 | 1950 | list_file(epi->ffd.file); |
22bacca4 DL |
1951 | } |
1952 | } | |
1953 | mutex_unlock(&ep->mtx); | |
1954 | ||
1955 | return error; | |
1956 | } | |
1957 | ||
1958 | /** | |
bde03c4c | 1959 | * ep_loop_check - Performs a check to verify that adding an epoll file (@to) |
a6c67fee | 1960 | * into another epoll file (represented by @ep) does not create |
22bacca4 DL |
1961 | * closed loops or too deep chains. |
1962 | * | |
a6c67fee | 1963 | * @ep: Pointer to the epoll we are inserting into. |
bde03c4c | 1964 | * @to: Pointer to the epoll to be inserted. |
22bacca4 | 1965 | * |
a6c67fee RD |
1966 | * Return: %zero if adding the epoll @to inside the epoll @from |
1967 | * does not violate the constraints, or %-1 otherwise. | |
22bacca4 | 1968 | */ |
bde03c4c | 1969 | static int ep_loop_check(struct eventpoll *ep, struct eventpoll *to) |
22bacca4 | 1970 | { |
6a3890c4 | 1971 | inserting_into = ep; |
bde03c4c | 1972 | return ep_loop_check_proc(to, 0); |
28d82dc1 JB |
1973 | } |
1974 | ||
1975 | static void clear_tfile_check_list(void) | |
1976 | { | |
319c1517 AV |
1977 | rcu_read_lock(); |
1978 | while (tfile_check_list != EP_UNACTIVE_PTR) { | |
1979 | struct epitems_head *head = tfile_check_list; | |
1980 | tfile_check_list = head->next; | |
1981 | unlist_file(head); | |
28d82dc1 | 1982 | } |
319c1517 | 1983 | rcu_read_unlock(); |
22bacca4 DL |
1984 | } |
1985 | ||
7699acd1 | 1986 | /* |
523723bb | 1987 | * Open an eventpoll file descriptor. |
7699acd1 | 1988 | */ |
791eb22e | 1989 | static int do_epoll_create(int flags) |
7699acd1 | 1990 | { |
28d82dc1 | 1991 | int error, fd; |
bb57c3ed | 1992 | struct eventpoll *ep = NULL; |
28d82dc1 | 1993 | struct file *file; |
7699acd1 | 1994 | |
e38b36f3 UD |
1995 | /* Check the EPOLL_* constant for consistency. */ |
1996 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
1997 | ||
296e236e DL |
1998 | if (flags & ~EPOLL_CLOEXEC) |
1999 | return -EINVAL; | |
7699acd1 | 2000 | /* |
bb57c3ed | 2001 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 2002 | */ |
9fe5ad9c | 2003 | error = ep_alloc(&ep); |
bb57c3ed DL |
2004 | if (error < 0) |
2005 | return error; | |
7699acd1 DL |
2006 | /* |
2007 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 2008 | * a file structure and a free file descriptor. |
7699acd1 | 2009 | */ |
28d82dc1 JB |
2010 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
2011 | if (fd < 0) { | |
2012 | error = fd; | |
2013 | goto out_free_ep; | |
2014 | } | |
2015 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 2016 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
2017 | if (IS_ERR(file)) { |
2018 | error = PTR_ERR(file); | |
2019 | goto out_free_fd; | |
2020 | } | |
28d82dc1 | 2021 | ep->file = file; |
98022748 | 2022 | fd_install(fd, file); |
28d82dc1 JB |
2023 | return fd; |
2024 | ||
2025 | out_free_fd: | |
2026 | put_unused_fd(fd); | |
2027 | out_free_ep: | |
2028 | ep_free(ep); | |
bb57c3ed | 2029 | return error; |
7699acd1 DL |
2030 | } |
2031 | ||
791eb22e DB |
2032 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
2033 | { | |
2034 | return do_epoll_create(flags); | |
2035 | } | |
2036 | ||
5a8a82b1 | 2037 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 2038 | { |
bfe3891a | 2039 | if (size <= 0) |
9fe5ad9c UD |
2040 | return -EINVAL; |
2041 | ||
791eb22e | 2042 | return do_epoll_create(0); |
a0998b50 UD |
2043 | } |
2044 | ||
39220e8d JA |
2045 | static inline int epoll_mutex_lock(struct mutex *mutex, int depth, |
2046 | bool nonblock) | |
2047 | { | |
2048 | if (!nonblock) { | |
2049 | mutex_lock_nested(mutex, depth); | |
2050 | return 0; | |
2051 | } | |
2052 | if (mutex_trylock(mutex)) | |
2053 | return 0; | |
2054 | return -EAGAIN; | |
2055 | } | |
2056 | ||
2057 | int do_epoll_ctl(int epfd, int op, int fd, struct epoll_event *epds, | |
2058 | bool nonblock) | |
7699acd1 DL |
2059 | { |
2060 | int error; | |
67347fe4 | 2061 | int full_check = 0; |
7e3fb584 | 2062 | struct fd f, tf; |
7699acd1 DL |
2063 | struct eventpoll *ep; |
2064 | struct epitem *epi; | |
67347fe4 | 2065 | struct eventpoll *tep = NULL; |
7699acd1 | 2066 | |
7699acd1 | 2067 | error = -EBADF; |
7e3fb584 AV |
2068 | f = fdget(epfd); |
2069 | if (!f.file) | |
7699acd1 DL |
2070 | goto error_return; |
2071 | ||
2072 | /* Get the "struct file *" for the target file */ | |
7e3fb584 AV |
2073 | tf = fdget(fd); |
2074 | if (!tf.file) | |
7699acd1 DL |
2075 | goto error_fput; |
2076 | ||
2077 | /* The target file descriptor must support poll */ | |
2078 | error = -EPERM; | |
9965ed17 | 2079 | if (!file_can_poll(tf.file)) |
7699acd1 DL |
2080 | goto error_tgt_fput; |
2081 | ||
4d7e30d9 | 2082 | /* Check if EPOLLWAKEUP is allowed */ |
c680e41b | 2083 | if (ep_op_has_event(op)) |
58e41a44 | 2084 | ep_take_care_of_epollwakeup(epds); |
4d7e30d9 | 2085 | |
7699acd1 DL |
2086 | /* |
2087 | * We have to check that the file structure underneath the file descriptor | |
2088 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
2089 | * adding an epoll file descriptor inside itself. | |
2090 | */ | |
2091 | error = -EINVAL; | |
7e3fb584 | 2092 | if (f.file == tf.file || !is_file_epoll(f.file)) |
7699acd1 DL |
2093 | goto error_tgt_fput; |
2094 | ||
df0108c5 JB |
2095 | /* |
2096 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, | |
2097 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. | |
2098 | * Also, we do not currently supported nested exclusive wakeups. | |
2099 | */ | |
58e41a44 | 2100 | if (ep_op_has_event(op) && (epds->events & EPOLLEXCLUSIVE)) { |
b6a515c8 JB |
2101 | if (op == EPOLL_CTL_MOD) |
2102 | goto error_tgt_fput; | |
2103 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || | |
58e41a44 | 2104 | (epds->events & ~EPOLLEXCLUSIVE_OK_BITS))) |
b6a515c8 JB |
2105 | goto error_tgt_fput; |
2106 | } | |
df0108c5 | 2107 | |
7699acd1 DL |
2108 | /* |
2109 | * At this point it is safe to assume that the "private_data" contains | |
2110 | * our own data structure. | |
2111 | */ | |
7e3fb584 | 2112 | ep = f.file->private_data; |
7699acd1 | 2113 | |
22bacca4 | 2114 | /* |
a6c67fee RD |
2115 | * When we insert an epoll file descriptor inside another epoll file |
2116 | * descriptor, there is the chance of creating closed loops, which are | |
28d82dc1 JB |
2117 | * better be handled here, than in more critical paths. While we are |
2118 | * checking for loops we also determine the list of files reachable | |
2119 | * and hang them on the tfile_check_list, so we can check that we | |
2120 | * haven't created too many possible wakeup paths. | |
22bacca4 | 2121 | * |
67347fe4 JB |
2122 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
2123 | * the epoll file descriptor is attaching directly to a wakeup source, | |
2124 | * unless the epoll file descriptor is nested. The purpose of taking the | |
2125 | * 'epmutex' on add is to prevent complex toplogies such as loops and | |
2126 | * deep wakeup paths from forming in parallel through multiple | |
2127 | * EPOLL_CTL_ADD operations. | |
22bacca4 | 2128 | */ |
39220e8d JA |
2129 | error = epoll_mutex_lock(&ep->mtx, 0, nonblock); |
2130 | if (error) | |
2131 | goto error_tgt_fput; | |
28d82dc1 | 2132 | if (op == EPOLL_CTL_ADD) { |
319c1517 AV |
2133 | if (READ_ONCE(f.file->f_ep) || ep->gen == loop_check_gen || |
2134 | is_file_epoll(tf.file)) { | |
67347fe4 | 2135 | mutex_unlock(&ep->mtx); |
39220e8d JA |
2136 | error = epoll_mutex_lock(&epmutex, 0, nonblock); |
2137 | if (error) | |
2138 | goto error_tgt_fput; | |
18306c40 | 2139 | loop_check_gen++; |
39220e8d | 2140 | full_check = 1; |
67347fe4 | 2141 | if (is_file_epoll(tf.file)) { |
bde03c4c | 2142 | tep = tf.file->private_data; |
67347fe4 | 2143 | error = -ELOOP; |
bde03c4c | 2144 | if (ep_loop_check(ep, tep) != 0) |
67347fe4 | 2145 | goto error_tgt_fput; |
a9ed4a65 | 2146 | } |
39220e8d | 2147 | error = epoll_mutex_lock(&ep->mtx, 0, nonblock); |
52c47969 | 2148 | if (error) |
39220e8d | 2149 | goto error_tgt_fput; |
67347fe4 JB |
2150 | } |
2151 | } | |
7699acd1 | 2152 | |
67647d0f | 2153 | /* |
a6c67fee | 2154 | * Try to lookup the file inside our RB tree. Since we grabbed "mtx" |
67647d0f DL |
2155 | * above, we can be sure to be able to use the item looked up by |
2156 | * ep_find() till we release the mutex. | |
2157 | */ | |
7e3fb584 | 2158 | epi = ep_find(ep, tf.file, fd); |
7699acd1 DL |
2159 | |
2160 | error = -EINVAL; | |
2161 | switch (op) { | |
2162 | case EPOLL_CTL_ADD: | |
2163 | if (!epi) { | |
58e41a44 JA |
2164 | epds->events |= EPOLLERR | EPOLLHUP; |
2165 | error = ep_insert(ep, epds, tf.file, fd, full_check); | |
7699acd1 DL |
2166 | } else |
2167 | error = -EEXIST; | |
2168 | break; | |
2169 | case EPOLL_CTL_DEL: | |
2170 | if (epi) | |
2171 | error = ep_remove(ep, epi); | |
2172 | else | |
2173 | error = -ENOENT; | |
2174 | break; | |
2175 | case EPOLL_CTL_MOD: | |
2176 | if (epi) { | |
b6a515c8 | 2177 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
58e41a44 JA |
2178 | epds->events |= EPOLLERR | EPOLLHUP; |
2179 | error = ep_modify(ep, epi, epds); | |
b6a515c8 | 2180 | } |
7699acd1 DL |
2181 | } else |
2182 | error = -ENOENT; | |
2183 | break; | |
2184 | } | |
d47de16c | 2185 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
2186 | |
2187 | error_tgt_fput: | |
52c47969 AV |
2188 | if (full_check) { |
2189 | clear_tfile_check_list(); | |
18306c40 | 2190 | loop_check_gen++; |
22bacca4 | 2191 | mutex_unlock(&epmutex); |
52c47969 | 2192 | } |
22bacca4 | 2193 | |
7e3fb584 | 2194 | fdput(tf); |
7699acd1 | 2195 | error_fput: |
7e3fb584 | 2196 | fdput(f); |
7699acd1 | 2197 | error_return: |
7699acd1 DL |
2198 | |
2199 | return error; | |
2200 | } | |
2201 | ||
58e41a44 JA |
2202 | /* |
2203 | * The following function implements the controller interface for | |
2204 | * the eventpoll file that enables the insertion/removal/change of | |
2205 | * file descriptors inside the interest set. | |
2206 | */ | |
2207 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, | |
2208 | struct epoll_event __user *, event) | |
2209 | { | |
2210 | struct epoll_event epds; | |
2211 | ||
2212 | if (ep_op_has_event(op) && | |
2213 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
2214 | return -EFAULT; | |
2215 | ||
39220e8d | 2216 | return do_epoll_ctl(epfd, op, fd, &epds, false); |
58e41a44 JA |
2217 | } |
2218 | ||
7699acd1 DL |
2219 | /* |
2220 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2221 | * part of the user space epoll_wait(2). | |
2222 | */ | |
791eb22e | 2223 | static int do_epoll_wait(int epfd, struct epoll_event __user *events, |
7cdf7c20 | 2224 | int maxevents, struct timespec64 *to) |
7699acd1 | 2225 | { |
2903ff01 AV |
2226 | int error; |
2227 | struct fd f; | |
7699acd1 DL |
2228 | struct eventpoll *ep; |
2229 | ||
7699acd1 DL |
2230 | /* The maximum number of event must be greater than zero */ |
2231 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
2232 | return -EINVAL; | |
2233 | ||
2234 | /* Verify that the area passed by the user is writeable */ | |
96d4f267 | 2235 | if (!access_ok(events, maxevents * sizeof(struct epoll_event))) |
2903ff01 | 2236 | return -EFAULT; |
7699acd1 DL |
2237 | |
2238 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
2239 | f = fdget(epfd); |
2240 | if (!f.file) | |
2241 | return -EBADF; | |
7699acd1 DL |
2242 | |
2243 | /* | |
2244 | * We have to check that the file structure underneath the fd | |
2245 | * the user passed to us _is_ an eventpoll file. | |
2246 | */ | |
2247 | error = -EINVAL; | |
2903ff01 | 2248 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
2249 | goto error_fput; |
2250 | ||
2251 | /* | |
2252 | * At this point it is safe to assume that the "private_data" contains | |
2253 | * our own data structure. | |
2254 | */ | |
2903ff01 | 2255 | ep = f.file->private_data; |
7699acd1 DL |
2256 | |
2257 | /* Time to fish for events ... */ | |
7cdf7c20 | 2258 | error = ep_poll(ep, events, maxevents, to); |
7699acd1 DL |
2259 | |
2260 | error_fput: | |
2903ff01 | 2261 | fdput(f); |
7699acd1 DL |
2262 | return error; |
2263 | } | |
2264 | ||
791eb22e DB |
2265 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
2266 | int, maxevents, int, timeout) | |
2267 | { | |
7cdf7c20 WB |
2268 | struct timespec64 to; |
2269 | ||
2270 | return do_epoll_wait(epfd, events, maxevents, | |
2271 | ep_timeout_to_timespec(&to, timeout)); | |
791eb22e DB |
2272 | } |
2273 | ||
7699acd1 DL |
2274 | /* |
2275 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2276 | * part of the user space epoll_pwait(2). | |
2277 | */ | |
58169a52 WB |
2278 | static int do_epoll_pwait(int epfd, struct epoll_event __user *events, |
2279 | int maxevents, struct timespec64 *to, | |
2280 | const sigset_t __user *sigmask, size_t sigsetsize) | |
7699acd1 DL |
2281 | { |
2282 | int error; | |
7699acd1 DL |
2283 | |
2284 | /* | |
2285 | * If the caller wants a certain signal mask to be set during the wait, | |
2286 | * we apply it here. | |
2287 | */ | |
b772434b | 2288 | error = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2289 | if (error) |
2290 | return error; | |
7699acd1 | 2291 | |
58169a52 | 2292 | error = do_epoll_wait(epfd, events, maxevents, to); |
7cdf7c20 | 2293 | |
b772434b | 2294 | restore_saved_sigmask_unless(error == -EINTR); |
7699acd1 DL |
2295 | |
2296 | return error; | |
2297 | } | |
2298 | ||
58169a52 WB |
2299 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
2300 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
2301 | size_t, sigsetsize) | |
35280bd4 | 2302 | { |
7cdf7c20 | 2303 | struct timespec64 to; |
58169a52 WB |
2304 | |
2305 | return do_epoll_pwait(epfd, events, maxevents, | |
2306 | ep_timeout_to_timespec(&to, timeout), | |
2307 | sigmask, sigsetsize); | |
2308 | } | |
2309 | ||
2310 | SYSCALL_DEFINE6(epoll_pwait2, int, epfd, struct epoll_event __user *, events, | |
2311 | int, maxevents, const struct __kernel_timespec __user *, timeout, | |
2312 | const sigset_t __user *, sigmask, size_t, sigsetsize) | |
2313 | { | |
2314 | struct timespec64 ts, *to = NULL; | |
2315 | ||
2316 | if (timeout) { | |
2317 | if (get_timespec64(&ts, timeout)) | |
2318 | return -EFAULT; | |
2319 | to = &ts; | |
2320 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
2321 | return -EINVAL; | |
2322 | } | |
2323 | ||
2324 | return do_epoll_pwait(epfd, events, maxevents, to, | |
2325 | sigmask, sigsetsize); | |
2326 | } | |
2327 | ||
2328 | #ifdef CONFIG_COMPAT | |
2329 | static int do_compat_epoll_pwait(int epfd, struct epoll_event __user *events, | |
2330 | int maxevents, struct timespec64 *timeout, | |
2331 | const compat_sigset_t __user *sigmask, | |
2332 | compat_size_t sigsetsize) | |
2333 | { | |
35280bd4 | 2334 | long err; |
35280bd4 AV |
2335 | |
2336 | /* | |
2337 | * If the caller wants a certain signal mask to be set during the wait, | |
2338 | * we apply it here. | |
2339 | */ | |
b772434b | 2340 | err = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2341 | if (err) |
2342 | return err; | |
35280bd4 | 2343 | |
58169a52 | 2344 | err = do_epoll_wait(epfd, events, maxevents, timeout); |
7cdf7c20 | 2345 | |
b772434b | 2346 | restore_saved_sigmask_unless(err == -EINTR); |
35280bd4 AV |
2347 | |
2348 | return err; | |
2349 | } | |
58169a52 WB |
2350 | |
2351 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, | |
2352 | struct epoll_event __user *, events, | |
2353 | int, maxevents, int, timeout, | |
2354 | const compat_sigset_t __user *, sigmask, | |
2355 | compat_size_t, sigsetsize) | |
2356 | { | |
2357 | struct timespec64 to; | |
2358 | ||
2359 | return do_compat_epoll_pwait(epfd, events, maxevents, | |
2360 | ep_timeout_to_timespec(&to, timeout), | |
2361 | sigmask, sigsetsize); | |
2362 | } | |
2363 | ||
2364 | COMPAT_SYSCALL_DEFINE6(epoll_pwait2, int, epfd, | |
2365 | struct epoll_event __user *, events, | |
2366 | int, maxevents, | |
2367 | const struct __kernel_timespec __user *, timeout, | |
2368 | const compat_sigset_t __user *, sigmask, | |
2369 | compat_size_t, sigsetsize) | |
2370 | { | |
2371 | struct timespec64 ts, *to = NULL; | |
2372 | ||
2373 | if (timeout) { | |
2374 | if (get_timespec64(&ts, timeout)) | |
2375 | return -EFAULT; | |
2376 | to = &ts; | |
2377 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
2378 | return -EINVAL; | |
2379 | } | |
2380 | ||
2381 | return do_compat_epoll_pwait(epfd, events, maxevents, to, | |
2382 | sigmask, sigsetsize); | |
2383 | } | |
2384 | ||
35280bd4 AV |
2385 | #endif |
2386 | ||
1da177e4 LT |
2387 | static int __init eventpoll_init(void) |
2388 | { | |
7ef9964e DL |
2389 | struct sysinfo si; |
2390 | ||
2391 | si_meminfo(&si); | |
9df04e1f DL |
2392 | /* |
2393 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
2394 | */ | |
2395 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 2396 | EP_ITEM_COST; |
52bd19f7 | 2397 | BUG_ON(max_user_watches < 0); |
1da177e4 | 2398 | |
39732ca5 EW |
2399 | /* |
2400 | * We can have many thousands of epitems, so prevent this from | |
2401 | * using an extra cache line on 64-bit (and smaller) CPUs | |
2402 | */ | |
2403 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); | |
2404 | ||
1da177e4 LT |
2405 | /* Allocates slab cache used to allocate "struct epitem" items */ |
2406 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
2ae928a9 | 2407 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 LT |
2408 | |
2409 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
2410 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
2ae928a9 | 2411 | sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); |
a8f5de89 | 2412 | epoll_sysctls_init(); |
1da177e4 | 2413 | |
319c1517 AV |
2414 | ephead_cache = kmem_cache_create("ep_head", |
2415 | sizeof(struct epitems_head), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); | |
2416 | ||
1da177e4 | 2417 | return 0; |
1da177e4 | 2418 | } |
cea69241 | 2419 | fs_initcall(eventpoll_init); |