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 | * | |
d4cb626d | 46 | * 1) epnested_mutex (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(). | |
d4cb626d DB |
60 | * The epnested_mutex is acquired when inserting an epoll fd onto another |
61 | * epoll fd. We do this so that we walk the epoll tree and ensure that this | |
22bacca4 DL |
62 | * insertion does not create a cycle of epoll file descriptors, which |
63 | * could lead to deadlock. We need a global mutex to prevent two | |
64 | * simultaneous inserts (A into B and B into A) from racing and | |
65 | * constructing a cycle without either insert observing that it is | |
66 | * going to. | |
d8805e63 NE |
67 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
68 | * case when one epoll fd is added to another. In this case, we | |
69 | * always acquire the locks in the order of nesting (i.e. after | |
70 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired | |
71 | * before e2->mtx). Since we disallow cycles of epoll file | |
72 | * descriptors, this ensures that the mutexes are well-ordered. In | |
73 | * order to communicate this nesting to lockdep, when walking a tree | |
74 | * of epoll file descriptors, we use the current recursion depth as | |
75 | * the lockdep subkey. | |
d47de16c | 76 | * It is possible to drop the "ep->mtx" and to use the global |
d4cb626d | 77 | * mutex "epnested_mutex" (together with "ep->lock") to have it working, |
d47de16c | 78 | * but having "ep->mtx" will make the interface more scalable. |
d4cb626d | 79 | * Events that require holding "epnested_mutex" are very rare, while for |
d47de16c DL |
80 | * normal operations the epoll private "ep->mtx" will guarantee |
81 | * a better scalability. | |
1da177e4 LT |
82 | */ |
83 | ||
1da177e4 | 84 | /* Epoll private bits inside the event mask */ |
df0108c5 | 85 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE) |
1da177e4 | 86 | |
a9a08845 | 87 | #define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT) |
b6a515c8 | 88 | |
a9a08845 | 89 | #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \ |
b6a515c8 JB |
90 | EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE) |
91 | ||
5071f97e DL |
92 | /* Maximum number of nesting allowed inside epoll sets */ |
93 | #define EP_MAX_NESTS 4 | |
1da177e4 | 94 | |
b611967d DL |
95 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
96 | ||
d47de16c DL |
97 | #define EP_UNACTIVE_PTR ((void *) -1L) |
98 | ||
7ef9964e DL |
99 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
100 | ||
1da177e4 LT |
101 | struct epoll_filefd { |
102 | struct file *file; | |
103 | int fd; | |
39732ca5 | 104 | } __packed; |
1da177e4 | 105 | |
80285b75 AV |
106 | /* Wait structure used by the poll hooks */ |
107 | struct eppoll_entry { | |
108 | /* List header used to link this structure to the "struct epitem" */ | |
109 | struct eppoll_entry *next; | |
1da177e4 | 110 | |
80285b75 AV |
111 | /* The "base" pointer is set to the container "struct epitem" */ |
112 | struct epitem *base; | |
113 | ||
114 | /* | |
115 | * Wait queue item that will be linked to the target file wait | |
116 | * queue head. | |
117 | */ | |
118 | wait_queue_entry_t wait; | |
119 | ||
120 | /* The wait queue head that linked the "wait" wait queue item */ | |
121 | wait_queue_head_t *whead; | |
1da177e4 LT |
122 | }; |
123 | ||
d47de16c DL |
124 | /* |
125 | * Each file descriptor added to the eventpoll interface will | |
126 | * have an entry of this type linked to the "rbr" RB tree. | |
39732ca5 EW |
127 | * Avoid increasing the size of this struct, there can be many thousands |
128 | * of these on a server and we do not want this to take another cache line. | |
d47de16c DL |
129 | */ |
130 | struct epitem { | |
ae10b2b4 JB |
131 | union { |
132 | /* RB tree node links this structure to the eventpoll RB tree */ | |
133 | struct rb_node rbn; | |
134 | /* Used to free the struct epitem */ | |
135 | struct rcu_head rcu; | |
136 | }; | |
d47de16c DL |
137 | |
138 | /* List header used to link this structure to the eventpoll ready list */ | |
139 | struct list_head rdllink; | |
140 | ||
c7ea7630 DL |
141 | /* |
142 | * Works together "struct eventpoll"->ovflist in keeping the | |
143 | * single linked chain of items. | |
144 | */ | |
145 | struct epitem *next; | |
146 | ||
d47de16c DL |
147 | /* The file descriptor information this item refers to */ |
148 | struct epoll_filefd ffd; | |
149 | ||
58c9b016 PA |
150 | /* |
151 | * Protected by file->f_lock, true for to-be-released epitem already | |
152 | * removed from the "struct file" items list; together with | |
153 | * eventpoll->refcount orchestrates "struct eventpoll" disposal | |
154 | */ | |
155 | bool dying; | |
156 | ||
d47de16c | 157 | /* List containing poll wait queues */ |
80285b75 | 158 | struct eppoll_entry *pwqlist; |
d47de16c DL |
159 | |
160 | /* The "container" of this item */ | |
161 | struct eventpoll *ep; | |
162 | ||
d47de16c | 163 | /* List header used to link this item to the "struct file" items list */ |
44cdc1d9 | 164 | struct hlist_node fllink; |
d47de16c | 165 | |
4d7e30d9 | 166 | /* wakeup_source used when EPOLLWAKEUP is set */ |
eea1d585 | 167 | struct wakeup_source __rcu *ws; |
4d7e30d9 | 168 | |
c7ea7630 DL |
169 | /* The structure that describe the interested events and the source fd */ |
170 | struct epoll_event event; | |
d47de16c DL |
171 | }; |
172 | ||
1da177e4 LT |
173 | /* |
174 | * This structure is stored inside the "private_data" member of the file | |
bf6a41db | 175 | * structure and represents the main data structure for the eventpoll |
1da177e4 LT |
176 | * interface. |
177 | */ | |
178 | struct eventpoll { | |
1da177e4 | 179 | /* |
d47de16c DL |
180 | * This mutex is used to ensure that files are not removed |
181 | * while epoll is using them. This is held during the event | |
182 | * collection loop, the file cleanup path, the epoll file exit | |
183 | * code and the ctl operations. | |
1da177e4 | 184 | */ |
d47de16c | 185 | struct mutex mtx; |
1da177e4 LT |
186 | |
187 | /* Wait queue used by sys_epoll_wait() */ | |
188 | wait_queue_head_t wq; | |
189 | ||
190 | /* Wait queue used by file->poll() */ | |
191 | wait_queue_head_t poll_wait; | |
192 | ||
193 | /* List of ready file descriptors */ | |
194 | struct list_head rdllist; | |
195 | ||
a218cc49 RP |
196 | /* Lock which protects rdllist and ovflist */ |
197 | rwlock_t lock; | |
198 | ||
67647d0f | 199 | /* RB tree root used to store monitored fd structs */ |
b2ac2ea6 | 200 | struct rb_root_cached rbr; |
d47de16c DL |
201 | |
202 | /* | |
203 | * This is a single linked list that chains all the "struct epitem" that | |
25985edc | 204 | * happened while transferring ready events to userspace w/out |
a218cc49 | 205 | * holding ->lock. |
d47de16c DL |
206 | */ |
207 | struct epitem *ovflist; | |
7ef9964e | 208 | |
4d7e30d9 AH |
209 | /* wakeup_source used when ep_scan_ready_list is running */ |
210 | struct wakeup_source *ws; | |
211 | ||
7ef9964e DL |
212 | /* The user that created the eventpoll descriptor */ |
213 | struct user_struct *user; | |
28d82dc1 JB |
214 | |
215 | struct file *file; | |
216 | ||
217 | /* used to optimize loop detection check */ | |
18306c40 | 218 | u64 gen; |
319c1517 | 219 | struct hlist_head refs; |
bf3b9f63 | 220 | |
58c9b016 PA |
221 | /* |
222 | * usage count, used together with epitem->dying to | |
223 | * orchestrate the disposal of this struct | |
224 | */ | |
225 | refcount_t refcount; | |
226 | ||
bf3b9f63 SS |
227 | #ifdef CONFIG_NET_RX_BUSY_POLL |
228 | /* used to track busy poll napi_id */ | |
229 | unsigned int napi_id; | |
230 | #endif | |
efcdd350 JB |
231 | |
232 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
233 | /* tracks wakeup nests for lockdep validation */ | |
234 | u8 nests; | |
235 | #endif | |
1da177e4 LT |
236 | }; |
237 | ||
1da177e4 LT |
238 | /* Wrapper struct used by poll queueing */ |
239 | struct ep_pqueue { | |
240 | poll_table pt; | |
241 | struct epitem *epi; | |
242 | }; | |
243 | ||
7ef9964e DL |
244 | /* |
245 | * Configuration options available inside /proc/sys/fs/epoll/ | |
246 | */ | |
7ef9964e | 247 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 248 | static long max_user_watches __read_mostly; |
7ef9964e | 249 | |
58c9b016 | 250 | /* Used for cycles detection */ |
d4cb626d | 251 | static DEFINE_MUTEX(epnested_mutex); |
1da177e4 | 252 | |
18306c40 AV |
253 | static u64 loop_check_gen = 0; |
254 | ||
22bacca4 | 255 | /* Used to check for epoll file descriptor inclusion loops */ |
6a3890c4 | 256 | static struct eventpoll *inserting_into; |
22bacca4 | 257 | |
1da177e4 | 258 | /* Slab cache used to allocate "struct epitem" */ |
68279f9c | 259 | static struct kmem_cache *epi_cache __ro_after_init; |
1da177e4 LT |
260 | |
261 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
68279f9c | 262 | static struct kmem_cache *pwq_cache __ro_after_init; |
1da177e4 | 263 | |
28d82dc1 JB |
264 | /* |
265 | * List of files with newly added links, where we may need to limit the number | |
d4cb626d | 266 | * of emanating paths. Protected by the epnested_mutex. |
28d82dc1 | 267 | */ |
319c1517 AV |
268 | struct epitems_head { |
269 | struct hlist_head epitems; | |
270 | struct epitems_head *next; | |
271 | }; | |
272 | static struct epitems_head *tfile_check_list = EP_UNACTIVE_PTR; | |
273 | ||
68279f9c | 274 | static struct kmem_cache *ephead_cache __ro_after_init; |
319c1517 AV |
275 | |
276 | static inline void free_ephead(struct epitems_head *head) | |
277 | { | |
278 | if (head) | |
279 | kmem_cache_free(ephead_cache, head); | |
280 | } | |
281 | ||
282 | static void list_file(struct file *file) | |
283 | { | |
284 | struct epitems_head *head; | |
285 | ||
286 | head = container_of(file->f_ep, struct epitems_head, epitems); | |
287 | if (!head->next) { | |
288 | head->next = tfile_check_list; | |
289 | tfile_check_list = head; | |
290 | } | |
291 | } | |
292 | ||
293 | static void unlist_file(struct epitems_head *head) | |
294 | { | |
295 | struct epitems_head *to_free = head; | |
296 | struct hlist_node *p = rcu_dereference(hlist_first_rcu(&head->epitems)); | |
297 | if (p) { | |
298 | struct epitem *epi= container_of(p, struct epitem, fllink); | |
299 | spin_lock(&epi->ffd.file->f_lock); | |
300 | if (!hlist_empty(&head->epitems)) | |
301 | to_free = NULL; | |
302 | head->next = NULL; | |
303 | spin_unlock(&epi->ffd.file->f_lock); | |
304 | } | |
305 | free_ephead(to_free); | |
306 | } | |
28d82dc1 | 307 | |
7ef9964e DL |
308 | #ifdef CONFIG_SYSCTL |
309 | ||
310 | #include <linux/sysctl.h> | |
311 | ||
eec4844f | 312 | static long long_zero; |
52bd19f7 | 313 | static long long_max = LONG_MAX; |
7ef9964e | 314 | |
a8f5de89 | 315 | static struct ctl_table epoll_table[] = { |
7ef9964e DL |
316 | { |
317 | .procname = "max_user_watches", | |
318 | .data = &max_user_watches, | |
52bd19f7 | 319 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 320 | .mode = 0644, |
52bd19f7 | 321 | .proc_handler = proc_doulongvec_minmax, |
eec4844f | 322 | .extra1 = &long_zero, |
52bd19f7 | 323 | .extra2 = &long_max, |
7ef9964e | 324 | }, |
7ef9964e | 325 | }; |
a8f5de89 XN |
326 | |
327 | static void __init epoll_sysctls_init(void) | |
328 | { | |
329 | register_sysctl("fs/epoll", epoll_table); | |
330 | } | |
331 | #else | |
332 | #define epoll_sysctls_init() do { } while (0) | |
7ef9964e DL |
333 | #endif /* CONFIG_SYSCTL */ |
334 | ||
28d82dc1 JB |
335 | static const struct file_operations eventpoll_fops; |
336 | ||
337 | static inline int is_file_epoll(struct file *f) | |
338 | { | |
339 | return f->f_op == &eventpoll_fops; | |
340 | } | |
b030a4dd | 341 | |
67647d0f | 342 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
343 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
344 | struct file *file, int fd) | |
345 | { | |
346 | ffd->file = file; | |
347 | ffd->fd = fd; | |
348 | } | |
349 | ||
67647d0f | 350 | /* Compare RB tree keys */ |
b030a4dd PE |
351 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
352 | struct epoll_filefd *p2) | |
353 | { | |
354 | return (p1->file > p2->file ? +1: | |
355 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
356 | } | |
357 | ||
b030a4dd | 358 | /* Tells us if the item is currently linked */ |
992991c0 | 359 | static inline int ep_is_linked(struct epitem *epi) |
b030a4dd | 360 | { |
992991c0 | 361 | return !list_empty(&epi->rdllink); |
b030a4dd PE |
362 | } |
363 | ||
ac6424b9 | 364 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
971316f0 ON |
365 | { |
366 | return container_of(p, struct eppoll_entry, wait); | |
367 | } | |
368 | ||
b030a4dd | 369 | /* Get the "struct epitem" from a wait queue pointer */ |
ac6424b9 | 370 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
b030a4dd PE |
371 | { |
372 | return container_of(p, struct eppoll_entry, wait)->base; | |
373 | } | |
374 | ||
3fb0e584 DL |
375 | /** |
376 | * ep_events_available - Checks if ready events might be available. | |
377 | * | |
378 | * @ep: Pointer to the eventpoll context. | |
379 | * | |
a6c67fee RD |
380 | * Return: a value different than %zero if ready events are available, |
381 | * or %zero otherwise. | |
3fb0e584 DL |
382 | */ |
383 | static inline int ep_events_available(struct eventpoll *ep) | |
384 | { | |
c5a282e9 DB |
385 | return !list_empty_careful(&ep->rdllist) || |
386 | READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR; | |
3fb0e584 DL |
387 | } |
388 | ||
bf3b9f63 SS |
389 | #ifdef CONFIG_NET_RX_BUSY_POLL |
390 | static bool ep_busy_loop_end(void *p, unsigned long start_time) | |
391 | { | |
392 | struct eventpoll *ep = p; | |
393 | ||
394 | return ep_events_available(ep) || busy_loop_timeout(start_time); | |
395 | } | |
bf3b9f63 SS |
396 | |
397 | /* | |
398 | * Busy poll if globally on and supporting sockets found && no events, | |
399 | * busy loop will return if need_resched or ep_events_available. | |
400 | * | |
401 | * we must do our busy polling with irqs enabled | |
402 | */ | |
1493c47f | 403 | static bool ep_busy_loop(struct eventpoll *ep, int nonblock) |
bf3b9f63 | 404 | { |
bf3b9f63 SS |
405 | unsigned int napi_id = READ_ONCE(ep->napi_id); |
406 | ||
1493c47f | 407 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) { |
7c951caf BT |
408 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep, false, |
409 | BUSY_POLL_BUDGET); | |
1493c47f SHY |
410 | if (ep_events_available(ep)) |
411 | return true; | |
412 | /* | |
413 | * Busy poll timed out. Drop NAPI ID for now, we can add | |
414 | * it back in when we have moved a socket with a valid NAPI | |
415 | * ID onto the ready list. | |
416 | */ | |
bf3b9f63 | 417 | ep->napi_id = 0; |
1493c47f SHY |
418 | return false; |
419 | } | |
420 | return false; | |
bf3b9f63 SS |
421 | } |
422 | ||
423 | /* | |
424 | * Set epoll busy poll NAPI ID from sk. | |
425 | */ | |
426 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
427 | { | |
bf3b9f63 SS |
428 | struct eventpoll *ep; |
429 | unsigned int napi_id; | |
430 | struct socket *sock; | |
431 | struct sock *sk; | |
bf3b9f63 SS |
432 | |
433 | if (!net_busy_loop_on()) | |
434 | return; | |
435 | ||
dba4a925 | 436 | sock = sock_from_file(epi->ffd.file); |
bf3b9f63 SS |
437 | if (!sock) |
438 | return; | |
439 | ||
440 | sk = sock->sk; | |
441 | if (!sk) | |
442 | return; | |
443 | ||
444 | napi_id = READ_ONCE(sk->sk_napi_id); | |
445 | ep = epi->ep; | |
446 | ||
447 | /* Non-NAPI IDs can be rejected | |
448 | * or | |
449 | * Nothing to do if we already have this ID | |
450 | */ | |
451 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) | |
452 | return; | |
453 | ||
454 | /* record NAPI ID for use in next busy poll */ | |
455 | ep->napi_id = napi_id; | |
bf3b9f63 SS |
456 | } |
457 | ||
514056d5 DB |
458 | #else |
459 | ||
1493c47f | 460 | static inline bool ep_busy_loop(struct eventpoll *ep, int nonblock) |
514056d5 | 461 | { |
1493c47f | 462 | return false; |
514056d5 DB |
463 | } |
464 | ||
465 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
466 | { | |
467 | } | |
468 | ||
469 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
470 | ||
02edc6fc SR |
471 | /* |
472 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
473 | * the use of wait queues used by epoll is done in a very controlled | |
474 | * manner. Wake ups can nest inside each other, but are never done | |
475 | * with the same locking. For example: | |
476 | * | |
477 | * dfd = socket(...); | |
478 | * efd1 = epoll_create(); | |
479 | * efd2 = epoll_create(); | |
480 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
481 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
482 | * | |
483 | * When a packet arrives to the device underneath "dfd", the net code will | |
484 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
485 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
486 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
487 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
488 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
489 | * that ends up in another wake_up(), after having checked about the | |
7059a9aa CL |
490 | * recursion constraints. That are, no more than EP_MAX_NESTS, to avoid |
491 | * stack blasting. | |
02edc6fc SR |
492 | * |
493 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
494 | * this special case of epoll. | |
495 | */ | |
2dfa4eea | 496 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
57a173bd | 497 | |
caf1aeaf JA |
498 | static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi, |
499 | unsigned pollflags) | |
5071f97e | 500 | { |
efcdd350 | 501 | struct eventpoll *ep_src; |
f6520c52 | 502 | unsigned long flags; |
efcdd350 JB |
503 | u8 nests = 0; |
504 | ||
505 | /* | |
506 | * To set the subclass or nesting level for spin_lock_irqsave_nested() | |
507 | * it might be natural to create a per-cpu nest count. However, since | |
508 | * we can recurse on ep->poll_wait.lock, and a non-raw spinlock can | |
509 | * schedule() in the -rt kernel, the per-cpu variable are no longer | |
510 | * protected. Thus, we are introducing a per eventpoll nest field. | |
511 | * If we are not being call from ep_poll_callback(), epi is NULL and | |
512 | * we are at the first level of nesting, 0. Otherwise, we are being | |
513 | * called from ep_poll_callback() and if a previous wakeup source is | |
514 | * not an epoll file itself, we are at depth 1 since the wakeup source | |
515 | * is depth 0. If the wakeup source is a previous epoll file in the | |
516 | * wakeup chain then we use its nests value and record ours as | |
517 | * nests + 1. The previous epoll file nests value is stable since its | |
518 | * already holding its own poll_wait.lock. | |
519 | */ | |
520 | if (epi) { | |
521 | if ((is_file_epoll(epi->ffd.file))) { | |
522 | ep_src = epi->ffd.file->private_data; | |
523 | nests = ep_src->nests; | |
524 | } else { | |
525 | nests = 1; | |
526 | } | |
527 | } | |
528 | spin_lock_irqsave_nested(&ep->poll_wait.lock, flags, nests); | |
529 | ep->nests = nests + 1; | |
caf1aeaf | 530 | wake_up_locked_poll(&ep->poll_wait, EPOLLIN | pollflags); |
efcdd350 JB |
531 | ep->nests = 0; |
532 | spin_unlock_irqrestore(&ep->poll_wait.lock, flags); | |
1da177e4 LT |
533 | } |
534 | ||
57a173bd JB |
535 | #else |
536 | ||
caf1aeaf | 537 | static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi, |
38f1755a | 538 | __poll_t pollflags) |
57a173bd | 539 | { |
caf1aeaf | 540 | wake_up_poll(&ep->poll_wait, EPOLLIN | pollflags); |
57a173bd JB |
541 | } |
542 | ||
543 | #endif | |
544 | ||
971316f0 ON |
545 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
546 | { | |
547 | wait_queue_head_t *whead; | |
548 | ||
549 | rcu_read_lock(); | |
138e4ad6 ON |
550 | /* |
551 | * If it is cleared by POLLFREE, it should be rcu-safe. | |
552 | * If we read NULL we need a barrier paired with | |
553 | * smp_store_release() in ep_poll_callback(), otherwise | |
554 | * we rely on whead->lock. | |
555 | */ | |
556 | whead = smp_load_acquire(&pwq->whead); | |
971316f0 ON |
557 | if (whead) |
558 | remove_wait_queue(whead, &pwq->wait); | |
559 | rcu_read_unlock(); | |
560 | } | |
561 | ||
1da177e4 | 562 | /* |
d1bc90dd | 563 | * This function unregisters poll callbacks from the associated file |
58c9b016 | 564 | * descriptor. Must be called with "mtx" held. |
1da177e4 | 565 | */ |
7699acd1 | 566 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 567 | { |
80285b75 | 568 | struct eppoll_entry **p = &epi->pwqlist; |
7699acd1 | 569 | struct eppoll_entry *pwq; |
1da177e4 | 570 | |
80285b75 AV |
571 | while ((pwq = *p) != NULL) { |
572 | *p = pwq->next; | |
971316f0 | 573 | ep_remove_wait_queue(pwq); |
d1bc90dd | 574 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 575 | } |
1da177e4 LT |
576 | } |
577 | ||
eea1d585 EW |
578 | /* call only when ep->mtx is held */ |
579 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) | |
580 | { | |
581 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); | |
582 | } | |
583 | ||
584 | /* call only when ep->mtx is held */ | |
585 | static inline void ep_pm_stay_awake(struct epitem *epi) | |
586 | { | |
587 | struct wakeup_source *ws = ep_wakeup_source(epi); | |
588 | ||
589 | if (ws) | |
590 | __pm_stay_awake(ws); | |
591 | } | |
592 | ||
593 | static inline bool ep_has_wakeup_source(struct epitem *epi) | |
594 | { | |
595 | return rcu_access_pointer(epi->ws) ? true : false; | |
596 | } | |
597 | ||
598 | /* call when ep->mtx cannot be held (ep_poll_callback) */ | |
599 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) | |
600 | { | |
601 | struct wakeup_source *ws; | |
602 | ||
603 | rcu_read_lock(); | |
604 | ws = rcu_dereference(epi->ws); | |
605 | if (ws) | |
606 | __pm_stay_awake(ws); | |
607 | rcu_read_unlock(); | |
608 | } | |
609 | ||
5071f97e | 610 | |
57804b1c AV |
611 | /* |
612 | * ep->mutex needs to be held because we could be hit by | |
613 | * eventpoll_release_file() and epoll_ctl(). | |
5071f97e | 614 | */ |
57804b1c | 615 | static void ep_start_scan(struct eventpoll *ep, struct list_head *txlist) |
5071f97e | 616 | { |
5071f97e DL |
617 | /* |
618 | * Steal the ready list, and re-init the original one to the | |
619 | * empty list. Also, set ep->ovflist to NULL so that events | |
620 | * happening while looping w/out locks, are not lost. We cannot | |
621 | * have the poll callback to queue directly on ep->rdllist, | |
622 | * because we want the "sproc" callback to be able to do it | |
623 | * in a lockless way. | |
624 | */ | |
57804b1c | 625 | lockdep_assert_irqs_enabled(); |
a218cc49 | 626 | write_lock_irq(&ep->lock); |
db502f8a | 627 | list_splice_init(&ep->rdllist, txlist); |
c5a282e9 | 628 | WRITE_ONCE(ep->ovflist, NULL); |
a218cc49 | 629 | write_unlock_irq(&ep->lock); |
db502f8a | 630 | } |
5071f97e | 631 | |
db502f8a | 632 | static void ep_done_scan(struct eventpoll *ep, |
db502f8a AV |
633 | struct list_head *txlist) |
634 | { | |
635 | struct epitem *epi, *nepi; | |
5071f97e | 636 | |
a218cc49 | 637 | write_lock_irq(&ep->lock); |
5071f97e DL |
638 | /* |
639 | * During the time we spent inside the "sproc" callback, some | |
640 | * other events might have been queued by the poll callback. | |
641 | * We re-insert them inside the main ready-list here. | |
642 | */ | |
c5a282e9 | 643 | for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL; |
5071f97e DL |
644 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
645 | /* | |
646 | * We need to check if the item is already in the list. | |
647 | * During the "sproc" callback execution time, items are | |
648 | * queued into ->ovflist but the "txlist" might already | |
649 | * contain them, and the list_splice() below takes care of them. | |
650 | */ | |
992991c0 | 651 | if (!ep_is_linked(epi)) { |
c141175d RP |
652 | /* |
653 | * ->ovflist is LIFO, so we have to reverse it in order | |
654 | * to keep in FIFO. | |
655 | */ | |
656 | list_add(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 657 | ep_pm_stay_awake(epi); |
4d7e30d9 | 658 | } |
5071f97e DL |
659 | } |
660 | /* | |
661 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
662 | * releasing the lock, events will be queued in the normal way inside | |
663 | * ep->rdllist. | |
664 | */ | |
c5a282e9 | 665 | WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR); |
5071f97e DL |
666 | |
667 | /* | |
668 | * Quickly re-inject items left on "txlist". | |
669 | */ | |
db502f8a | 670 | list_splice(txlist, &ep->rdllist); |
4d7e30d9 | 671 | __pm_relax(ep->ws); |
7fab29e3 DB |
672 | |
673 | if (!list_empty(&ep->rdllist)) { | |
674 | if (waitqueue_active(&ep->wq)) | |
675 | wake_up(&ep->wq); | |
676 | } | |
677 | ||
a218cc49 | 678 | write_unlock_irq(&ep->lock); |
5071f97e DL |
679 | } |
680 | ||
ae10b2b4 JB |
681 | static void epi_rcu_free(struct rcu_head *head) |
682 | { | |
683 | struct epitem *epi = container_of(head, struct epitem, rcu); | |
684 | kmem_cache_free(epi_cache, epi); | |
685 | } | |
686 | ||
58c9b016 PA |
687 | static void ep_get(struct eventpoll *ep) |
688 | { | |
689 | refcount_inc(&ep->refcount); | |
690 | } | |
691 | ||
692 | /* | |
693 | * Returns true if the event poll can be disposed | |
694 | */ | |
695 | static bool ep_refcount_dec_and_test(struct eventpoll *ep) | |
696 | { | |
697 | if (!refcount_dec_and_test(&ep->refcount)) | |
698 | return false; | |
699 | ||
700 | WARN_ON_ONCE(!RB_EMPTY_ROOT(&ep->rbr.rb_root)); | |
701 | return true; | |
702 | } | |
703 | ||
704 | static void ep_free(struct eventpoll *ep) | |
705 | { | |
706 | mutex_destroy(&ep->mtx); | |
707 | free_uid(ep->user); | |
708 | wakeup_source_unregister(ep->ws); | |
709 | kfree(ep); | |
710 | } | |
711 | ||
7699acd1 DL |
712 | /* |
713 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 714 | * all the associated resources. Must be called with "mtx" held. |
58c9b016 PA |
715 | * If the dying flag is set, do the removal only if force is true. |
716 | * This prevents ep_clear_and_put() from dropping all the ep references | |
717 | * while running concurrently with eventpoll_release_file(). | |
718 | * Returns true if the eventpoll can be disposed. | |
7699acd1 | 719 | */ |
58c9b016 | 720 | static bool __ep_remove(struct eventpoll *ep, struct epitem *epi, bool force) |
7699acd1 | 721 | { |
7699acd1 | 722 | struct file *file = epi->ffd.file; |
319c1517 AV |
723 | struct epitems_head *to_free; |
724 | struct hlist_head *head; | |
1da177e4 | 725 | |
92e64178 DB |
726 | lockdep_assert_irqs_enabled(); |
727 | ||
1da177e4 | 728 | /* |
ee8ef0a4 | 729 | * Removes poll wait queue hooks. |
1da177e4 | 730 | */ |
7699acd1 | 731 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 732 | |
7699acd1 | 733 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 734 | spin_lock(&file->f_lock); |
58c9b016 PA |
735 | if (epi->dying && !force) { |
736 | spin_unlock(&file->f_lock); | |
737 | return false; | |
738 | } | |
739 | ||
319c1517 AV |
740 | to_free = NULL; |
741 | head = file->f_ep; | |
742 | if (head->first == &epi->fllink && !epi->fllink.next) { | |
743 | file->f_ep = NULL; | |
744 | if (!is_file_epoll(file)) { | |
745 | struct epitems_head *v; | |
746 | v = container_of(head, struct epitems_head, epitems); | |
747 | if (!smp_load_acquire(&v->next)) | |
748 | to_free = v; | |
749 | } | |
750 | } | |
44cdc1d9 | 751 | hlist_del_rcu(&epi->fllink); |
68499914 | 752 | spin_unlock(&file->f_lock); |
319c1517 | 753 | free_ephead(to_free); |
1da177e4 | 754 | |
b2ac2ea6 | 755 | rb_erase_cached(&epi->rbn, &ep->rbr); |
1da177e4 | 756 | |
a218cc49 | 757 | write_lock_irq(&ep->lock); |
992991c0 | 758 | if (ep_is_linked(epi)) |
c7ea7630 | 759 | list_del_init(&epi->rdllink); |
a218cc49 | 760 | write_unlock_irq(&ep->lock); |
1da177e4 | 761 | |
eea1d585 | 762 | wakeup_source_unregister(ep_wakeup_source(epi)); |
ae10b2b4 JB |
763 | /* |
764 | * At this point it is safe to free the eventpoll item. Use the union | |
765 | * field epi->rcu, since we are trying to minimize the size of | |
766 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by | |
767 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make | |
768 | * use of the rbn field. | |
769 | */ | |
770 | call_rcu(&epi->rcu, epi_rcu_free); | |
1da177e4 | 771 | |
1e1c1583 | 772 | percpu_counter_dec(&ep->user->epoll_watches); |
58c9b016 PA |
773 | return ep_refcount_dec_and_test(ep); |
774 | } | |
7ef9964e | 775 | |
58c9b016 PA |
776 | /* |
777 | * ep_remove variant for callers owing an additional reference to the ep | |
778 | */ | |
779 | static void ep_remove_safe(struct eventpoll *ep, struct epitem *epi) | |
780 | { | |
781 | WARN_ON_ONCE(__ep_remove(ep, epi, false)); | |
1da177e4 LT |
782 | } |
783 | ||
58c9b016 | 784 | static void ep_clear_and_put(struct eventpoll *ep) |
1da177e4 | 785 | { |
58c9b016 | 786 | struct rb_node *rbp, *next; |
7699acd1 | 787 | struct epitem *epi; |
58c9b016 | 788 | bool dispose; |
1da177e4 | 789 | |
7699acd1 DL |
790 | /* We need to release all tasks waiting for these file */ |
791 | if (waitqueue_active(&ep->poll_wait)) | |
caf1aeaf | 792 | ep_poll_safewake(ep, NULL, 0); |
1da177e4 | 793 | |
58c9b016 | 794 | mutex_lock(&ep->mtx); |
1da177e4 LT |
795 | |
796 | /* | |
7699acd1 | 797 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 798 | */ |
b2ac2ea6 | 799 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
7699acd1 DL |
800 | epi = rb_entry(rbp, struct epitem, rbn); |
801 | ||
802 | ep_unregister_pollwait(ep, epi); | |
91cf5ab6 | 803 | cond_resched(); |
7699acd1 | 804 | } |
1da177e4 LT |
805 | |
806 | /* | |
58c9b016 PA |
807 | * Walks through the whole tree and try to free each "struct epitem". |
808 | * Note that ep_remove_safe() will not remove the epitem in case of a | |
809 | * racing eventpoll_release_file(); the latter will do the removal. | |
810 | * At this point we are sure no poll callbacks will be lingering around. | |
811 | * Since we still own a reference to the eventpoll struct, the loop can't | |
812 | * dispose it. | |
1da177e4 | 813 | */ |
58c9b016 PA |
814 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = next) { |
815 | next = rb_next(rbp); | |
7699acd1 | 816 | epi = rb_entry(rbp, struct epitem, rbn); |
58c9b016 | 817 | ep_remove_safe(ep, epi); |
91cf5ab6 | 818 | cond_resched(); |
7699acd1 | 819 | } |
58c9b016 PA |
820 | |
821 | dispose = ep_refcount_dec_and_test(ep); | |
ddf676c3 | 822 | mutex_unlock(&ep->mtx); |
1da177e4 | 823 | |
58c9b016 PA |
824 | if (dispose) |
825 | ep_free(ep); | |
7699acd1 | 826 | } |
1da177e4 | 827 | |
7699acd1 DL |
828 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
829 | { | |
830 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 831 | |
f0ee9aab | 832 | if (ep) |
58c9b016 | 833 | ep_clear_and_put(ep); |
7699acd1 | 834 | |
7699acd1 | 835 | return 0; |
1da177e4 LT |
836 | } |
837 | ||
2c0b71c1 | 838 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, int depth); |
37b5e521 | 839 | |
ad9366b1 | 840 | static __poll_t __ep_eventpoll_poll(struct file *file, poll_table *wait, int depth) |
5071f97e | 841 | { |
ad9366b1 AV |
842 | struct eventpoll *ep = file->private_data; |
843 | LIST_HEAD(txlist); | |
5071f97e | 844 | struct epitem *epi, *tmp; |
626cf236 | 845 | poll_table pt; |
2c0b71c1 | 846 | __poll_t res = 0; |
5071f97e | 847 | |
626cf236 | 848 | init_poll_funcptr(&pt, NULL); |
450d89ec | 849 | |
ad9366b1 AV |
850 | /* Insert inside our poll wait queue */ |
851 | poll_wait(file, &ep->poll_wait, wait); | |
852 | ||
853 | /* | |
854 | * Proceed to find out if wanted events are really available inside | |
855 | * the ready list. | |
856 | */ | |
857 | mutex_lock_nested(&ep->mtx, depth); | |
858 | ep_start_scan(ep, &txlist); | |
2c0b71c1 AV |
859 | list_for_each_entry_safe(epi, tmp, &txlist, rdllink) { |
860 | if (ep_item_poll(epi, &pt, depth + 1)) { | |
861 | res = EPOLLIN | EPOLLRDNORM; | |
862 | break; | |
37b5e521 | 863 | } else { |
5071f97e DL |
864 | /* |
865 | * Item has been dropped into the ready list by the poll | |
866 | * callback, but it's not actually ready, as far as | |
867 | * caller requested events goes. We can remove it here. | |
868 | */ | |
eea1d585 | 869 | __pm_relax(ep_wakeup_source(epi)); |
5071f97e | 870 | list_del_init(&epi->rdllink); |
296e236e | 871 | } |
5071f97e | 872 | } |
ad9366b1 AV |
873 | ep_done_scan(ep, &txlist); |
874 | mutex_unlock(&ep->mtx); | |
875 | return res; | |
5071f97e DL |
876 | } |
877 | ||
37b5e521 JB |
878 | /* |
879 | * Differs from ep_eventpoll_poll() in that internal callers already have | |
880 | * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested() | |
881 | * is correctly annotated. | |
882 | */ | |
d85e2aa2 | 883 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, |
bec1a502 | 884 | int depth) |
11c5ad0e | 885 | { |
ad9366b1 | 886 | struct file *file = epi->ffd.file; |
1ec09974 | 887 | __poll_t res; |
7699acd1 | 888 | |
450d89ec | 889 | pt->_key = epi->event.events; |
ad9366b1 AV |
890 | if (!is_file_epoll(file)) |
891 | res = vfs_poll(file, pt); | |
892 | else | |
893 | res = __ep_eventpoll_poll(file, pt, depth); | |
1ec09974 | 894 | return res & epi->event.events; |
450d89ec | 895 | } |
a11e1d43 | 896 | |
a11e1d43 | 897 | static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait) |
11c5ad0e | 898 | { |
ad9366b1 | 899 | return __ep_eventpoll_poll(file, wait, 0); |
7699acd1 DL |
900 | } |
901 | ||
138d22b5 | 902 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 903 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
138d22b5 CG |
904 | { |
905 | struct eventpoll *ep = f->private_data; | |
906 | struct rb_node *rbp; | |
138d22b5 CG |
907 | |
908 | mutex_lock(&ep->mtx); | |
b2ac2ea6 | 909 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
138d22b5 | 910 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
77493f04 | 911 | struct inode *inode = file_inode(epi->ffd.file); |
138d22b5 | 912 | |
77493f04 CG |
913 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
914 | " pos:%lli ino:%lx sdev:%x\n", | |
a3816ab0 | 915 | epi->ffd.fd, epi->event.events, |
77493f04 CG |
916 | (long long)epi->event.data, |
917 | (long long)epi->ffd.file->f_pos, | |
918 | inode->i_ino, inode->i_sb->s_dev); | |
a3816ab0 | 919 | if (seq_has_overflowed(m)) |
138d22b5 CG |
920 | break; |
921 | } | |
922 | mutex_unlock(&ep->mtx); | |
138d22b5 CG |
923 | } |
924 | #endif | |
925 | ||
7699acd1 DL |
926 | /* File callbacks that implement the eventpoll file behaviour */ |
927 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
928 | #ifdef CONFIG_PROC_FS |
929 | .show_fdinfo = ep_show_fdinfo, | |
930 | #endif | |
7699acd1 | 931 | .release = ep_eventpoll_release, |
a11e1d43 | 932 | .poll = ep_eventpoll_poll, |
6038f373 | 933 | .llseek = noop_llseek, |
7699acd1 DL |
934 | }; |
935 | ||
b611967d | 936 | /* |
7699acd1 DL |
937 | * This is called from eventpoll_release() to unlink files from the eventpoll |
938 | * interface. We need to have this facility to cleanup correctly files that are | |
939 | * closed without being removed from the eventpoll interface. | |
b611967d | 940 | */ |
7699acd1 | 941 | void eventpoll_release_file(struct file *file) |
b611967d | 942 | { |
7699acd1 | 943 | struct eventpoll *ep; |
44cdc1d9 | 944 | struct epitem *epi; |
58c9b016 | 945 | bool dispose; |
b611967d DL |
946 | |
947 | /* | |
58c9b016 PA |
948 | * Use the 'dying' flag to prevent a concurrent ep_clear_and_put() from |
949 | * touching the epitems list before eventpoll_release_file() can access | |
950 | * the ep->mtx. | |
b611967d | 951 | */ |
58c9b016 PA |
952 | again: |
953 | spin_lock(&file->f_lock); | |
954 | if (file->f_ep && file->f_ep->first) { | |
955 | epi = hlist_entry(file->f_ep->first, struct epitem, fllink); | |
956 | epi->dying = true; | |
957 | spin_unlock(&file->f_lock); | |
958 | ||
959 | /* | |
960 | * ep access is safe as we still own a reference to the ep | |
961 | * struct | |
962 | */ | |
7699acd1 | 963 | ep = epi->ep; |
58c9b016 PA |
964 | mutex_lock(&ep->mtx); |
965 | dispose = __ep_remove(ep, epi, true); | |
d47de16c | 966 | mutex_unlock(&ep->mtx); |
58c9b016 PA |
967 | |
968 | if (dispose) | |
969 | ep_free(ep); | |
970 | goto again; | |
b611967d | 971 | } |
58c9b016 | 972 | spin_unlock(&file->f_lock); |
b611967d DL |
973 | } |
974 | ||
53d2be79 | 975 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 976 | { |
7ef9964e | 977 | struct eventpoll *ep; |
1da177e4 | 978 | |
7ef9964e DL |
979 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); |
980 | if (unlikely(!ep)) | |
05f26f86 | 981 | return -ENOMEM; |
1da177e4 | 982 | |
d47de16c | 983 | mutex_init(&ep->mtx); |
a218cc49 | 984 | rwlock_init(&ep->lock); |
1da177e4 LT |
985 | init_waitqueue_head(&ep->wq); |
986 | init_waitqueue_head(&ep->poll_wait); | |
987 | INIT_LIST_HEAD(&ep->rdllist); | |
b2ac2ea6 | 988 | ep->rbr = RB_ROOT_CACHED; |
d47de16c | 989 | ep->ovflist = EP_UNACTIVE_PTR; |
05f26f86 | 990 | ep->user = get_current_user(); |
58c9b016 | 991 | refcount_set(&ep->refcount, 1); |
1da177e4 | 992 | |
53d2be79 | 993 | *pep = ep; |
1da177e4 | 994 | |
1da177e4 LT |
995 | return 0; |
996 | } | |
997 | ||
1da177e4 | 998 | /* |
c7ea7630 DL |
999 | * Search the file inside the eventpoll tree. The RB tree operations |
1000 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
1001 | * "mtx" held. | |
1da177e4 LT |
1002 | */ |
1003 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
1004 | { | |
1005 | int kcmp; | |
1da177e4 LT |
1006 | struct rb_node *rbp; |
1007 | struct epitem *epi, *epir = NULL; | |
1008 | struct epoll_filefd ffd; | |
1009 | ||
b030a4dd | 1010 | ep_set_ffd(&ffd, file, fd); |
b2ac2ea6 | 1011 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
1da177e4 | 1012 | epi = rb_entry(rbp, struct epitem, rbn); |
b030a4dd | 1013 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
1014 | if (kcmp > 0) |
1015 | rbp = rbp->rb_right; | |
1016 | else if (kcmp < 0) | |
1017 | rbp = rbp->rb_left; | |
1018 | else { | |
1da177e4 LT |
1019 | epir = epi; |
1020 | break; | |
1021 | } | |
1022 | } | |
1da177e4 | 1023 | |
1da177e4 LT |
1024 | return epir; |
1025 | } | |
1026 | ||
bfe3911a | 1027 | #ifdef CONFIG_KCMP |
0791e364 CG |
1028 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
1029 | { | |
1030 | struct rb_node *rbp; | |
1031 | struct epitem *epi; | |
1032 | ||
b2ac2ea6 | 1033 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
0791e364 CG |
1034 | epi = rb_entry(rbp, struct epitem, rbn); |
1035 | if (epi->ffd.fd == tfd) { | |
1036 | if (toff == 0) | |
1037 | return epi; | |
1038 | else | |
1039 | toff--; | |
1040 | } | |
1041 | cond_resched(); | |
1042 | } | |
1043 | ||
1044 | return NULL; | |
1045 | } | |
1046 | ||
1047 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, | |
1048 | unsigned long toff) | |
1049 | { | |
1050 | struct file *file_raw; | |
1051 | struct eventpoll *ep; | |
1052 | struct epitem *epi; | |
1053 | ||
1054 | if (!is_file_epoll(file)) | |
1055 | return ERR_PTR(-EINVAL); | |
1056 | ||
1057 | ep = file->private_data; | |
1058 | ||
1059 | mutex_lock(&ep->mtx); | |
1060 | epi = ep_find_tfd(ep, tfd, toff); | |
1061 | if (epi) | |
1062 | file_raw = epi->ffd.file; | |
1063 | else | |
1064 | file_raw = ERR_PTR(-ENOENT); | |
1065 | mutex_unlock(&ep->mtx); | |
1066 | ||
1067 | return file_raw; | |
1068 | } | |
bfe3911a | 1069 | #endif /* CONFIG_KCMP */ |
0791e364 | 1070 | |
a6c67fee | 1071 | /* |
a218cc49 RP |
1072 | * Adds a new entry to the tail of the list in a lockless way, i.e. |
1073 | * multiple CPUs are allowed to call this function concurrently. | |
1074 | * | |
1075 | * Beware: it is necessary to prevent any other modifications of the | |
1076 | * existing list until all changes are completed, in other words | |
1077 | * concurrent list_add_tail_lockless() calls should be protected | |
1078 | * with a read lock, where write lock acts as a barrier which | |
1079 | * makes sure all list_add_tail_lockless() calls are fully | |
1080 | * completed. | |
1081 | * | |
1082 | * Also an element can be locklessly added to the list only in one | |
a6c67fee | 1083 | * direction i.e. either to the tail or to the head, otherwise |
a218cc49 RP |
1084 | * concurrent access will corrupt the list. |
1085 | * | |
a6c67fee | 1086 | * Return: %false if element has been already added to the list, %true |
a218cc49 RP |
1087 | * otherwise. |
1088 | */ | |
1089 | static inline bool list_add_tail_lockless(struct list_head *new, | |
1090 | struct list_head *head) | |
1091 | { | |
1092 | struct list_head *prev; | |
1093 | ||
1094 | /* | |
1095 | * This is simple 'new->next = head' operation, but cmpxchg() | |
1096 | * is used in order to detect that same element has been just | |
1097 | * added to the list from another CPU: the winner observes | |
1098 | * new->next == new. | |
1099 | */ | |
693fc06e | 1100 | if (!try_cmpxchg(&new->next, &new, head)) |
a218cc49 RP |
1101 | return false; |
1102 | ||
1103 | /* | |
1104 | * Initially ->next of a new element must be updated with the head | |
1105 | * (we are inserting to the tail) and only then pointers are atomically | |
1106 | * exchanged. XCHG guarantees memory ordering, thus ->next should be | |
1107 | * updated before pointers are actually swapped and pointers are | |
1108 | * swapped before prev->next is updated. | |
1109 | */ | |
1110 | ||
1111 | prev = xchg(&head->prev, new); | |
1112 | ||
1113 | /* | |
1114 | * It is safe to modify prev->next and new->prev, because a new element | |
1115 | * is added only to the tail and new->next is updated before XCHG. | |
1116 | */ | |
1117 | ||
1118 | prev->next = new; | |
1119 | new->prev = prev; | |
1120 | ||
1121 | return true; | |
1122 | } | |
1123 | ||
a6c67fee | 1124 | /* |
a218cc49 RP |
1125 | * Chains a new epi entry to the tail of the ep->ovflist in a lockless way, |
1126 | * i.e. multiple CPUs are allowed to call this function concurrently. | |
1127 | * | |
a6c67fee | 1128 | * Return: %false if epi element has been already chained, %true otherwise. |
a218cc49 RP |
1129 | */ |
1130 | static inline bool chain_epi_lockless(struct epitem *epi) | |
1131 | { | |
1132 | struct eventpoll *ep = epi->ep; | |
1133 | ||
0c54a6a4 KK |
1134 | /* Fast preliminary check */ |
1135 | if (epi->next != EP_UNACTIVE_PTR) | |
1136 | return false; | |
1137 | ||
a218cc49 RP |
1138 | /* Check that the same epi has not been just chained from another CPU */ |
1139 | if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR) | |
1140 | return false; | |
1141 | ||
1142 | /* Atomically exchange tail */ | |
1143 | epi->next = xchg(&ep->ovflist, epi); | |
1144 | ||
1145 | return true; | |
1146 | } | |
1147 | ||
1da177e4 | 1148 | /* |
7699acd1 | 1149 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 1150 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 1151 | * have events to report. |
a218cc49 | 1152 | * |
a6c67fee RD |
1153 | * This callback takes a read lock in order not to contend with concurrent |
1154 | * events from another file descriptor, thus all modifications to ->rdllist | |
a218cc49 RP |
1155 | * or ->ovflist are lockless. Read lock is paired with the write lock from |
1156 | * ep_scan_ready_list(), which stops all list modifications and guarantees | |
1157 | * that lists state is seen correctly. | |
1158 | * | |
1159 | * Another thing worth to mention is that ep_poll_callback() can be called | |
1160 | * concurrently for the same @epi from different CPUs if poll table was inited | |
1161 | * with several wait queues entries. Plural wakeup from different CPUs of a | |
1162 | * single wait queue is serialized by wq.lock, but the case when multiple wait | |
1163 | * queues are used should be detected accordingly. This is detected using | |
1164 | * cmpxchg() operation. | |
1da177e4 | 1165 | */ |
ac6424b9 | 1166 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 1167 | { |
7699acd1 | 1168 | int pwake = 0; |
7699acd1 DL |
1169 | struct epitem *epi = ep_item_from_wait(wait); |
1170 | struct eventpoll *ep = epi->ep; | |
3ad6f93e | 1171 | __poll_t pollflags = key_to_poll(key); |
a218cc49 | 1172 | unsigned long flags; |
df0108c5 | 1173 | int ewake = 0; |
1da177e4 | 1174 | |
a218cc49 | 1175 | read_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1176 | |
bf3b9f63 SS |
1177 | ep_set_busy_poll_napi_id(epi); |
1178 | ||
7699acd1 DL |
1179 | /* |
1180 | * If the event mask does not contain any poll(2) event, we consider the | |
1181 | * descriptor to be disabled. This condition is likely the effect of the | |
1182 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1183 | * until the next EPOLL_CTL_MOD will be issued. | |
1184 | */ | |
1185 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
1186 | goto out_unlock; |
1187 | ||
2dfa4eea DL |
1188 | /* |
1189 | * Check the events coming with the callback. At this stage, not | |
1190 | * every device reports the events in the "key" parameter of the | |
1191 | * callback. We need to be able to handle both cases here, hence the | |
1192 | * test for "key" != NULL before the event match test. | |
1193 | */ | |
3ad6f93e | 1194 | if (pollflags && !(pollflags & epi->event.events)) |
2dfa4eea DL |
1195 | goto out_unlock; |
1196 | ||
d47de16c | 1197 | /* |
bf6a41db | 1198 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 1199 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 1200 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
1201 | * chained in ep->ovflist and requeued later on. |
1202 | */ | |
c5a282e9 | 1203 | if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) { |
0c54a6a4 KK |
1204 | if (chain_epi_lockless(epi)) |
1205 | ep_pm_stay_awake_rcu(epi); | |
1206 | } else if (!ep_is_linked(epi)) { | |
1207 | /* In the usual case, add event to ready list. */ | |
1208 | if (list_add_tail_lockless(&epi->rdllink, &ep->rdllist)) | |
c3e320b6 | 1209 | ep_pm_stay_awake_rcu(epi); |
4d7e30d9 | 1210 | } |
7699acd1 | 1211 | |
7699acd1 DL |
1212 | /* |
1213 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1214 | * wait list. | |
1215 | */ | |
df0108c5 | 1216 | if (waitqueue_active(&ep->wq)) { |
b6a515c8 | 1217 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
3ad6f93e AV |
1218 | !(pollflags & POLLFREE)) { |
1219 | switch (pollflags & EPOLLINOUT_BITS) { | |
a9a08845 LT |
1220 | case EPOLLIN: |
1221 | if (epi->event.events & EPOLLIN) | |
b6a515c8 JB |
1222 | ewake = 1; |
1223 | break; | |
a9a08845 LT |
1224 | case EPOLLOUT: |
1225 | if (epi->event.events & EPOLLOUT) | |
b6a515c8 JB |
1226 | ewake = 1; |
1227 | break; | |
1228 | case 0: | |
1229 | ewake = 1; | |
1230 | break; | |
1231 | } | |
1232 | } | |
a218cc49 | 1233 | wake_up(&ep->wq); |
df0108c5 | 1234 | } |
7699acd1 DL |
1235 | if (waitqueue_active(&ep->poll_wait)) |
1236 | pwake++; | |
1237 | ||
d47de16c | 1238 | out_unlock: |
a218cc49 | 1239 | read_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1240 | |
7699acd1 DL |
1241 | /* We have to call this outside the lock */ |
1242 | if (pwake) | |
caf1aeaf | 1243 | ep_poll_safewake(ep, epi, pollflags & EPOLL_URING_WAKE); |
7699acd1 | 1244 | |
138e4ad6 ON |
1245 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
1246 | ewake = 1; | |
1247 | ||
3ad6f93e | 1248 | if (pollflags & POLLFREE) { |
138e4ad6 ON |
1249 | /* |
1250 | * If we race with ep_remove_wait_queue() it can miss | |
1251 | * ->whead = NULL and do another remove_wait_queue() after | |
1252 | * us, so we can't use __remove_wait_queue(). | |
1253 | */ | |
1254 | list_del_init(&wait->entry); | |
1255 | /* | |
58c9b016 PA |
1256 | * ->whead != NULL protects us from the race with |
1257 | * ep_clear_and_put() or ep_remove(), ep_remove_wait_queue() | |
1258 | * takes whead->lock held by the caller. Once we nullify it, | |
1259 | * nothing protects ep/epi or even wait. | |
138e4ad6 ON |
1260 | */ |
1261 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); | |
1262 | } | |
df0108c5 | 1263 | |
138e4ad6 | 1264 | return ewake; |
7699acd1 | 1265 | } |
1da177e4 LT |
1266 | |
1267 | /* | |
1268 | * This is the callback that is used to add our wait queue to the | |
1269 | * target file wakeup lists. | |
1270 | */ | |
1271 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1272 | poll_table *pt) | |
1273 | { | |
364f374f AV |
1274 | struct ep_pqueue *epq = container_of(pt, struct ep_pqueue, pt); |
1275 | struct epitem *epi = epq->epi; | |
1da177e4 LT |
1276 | struct eppoll_entry *pwq; |
1277 | ||
364f374f AV |
1278 | if (unlikely(!epi)) // an earlier allocation has failed |
1279 | return; | |
1280 | ||
1281 | pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL); | |
1282 | if (unlikely(!pwq)) { | |
1283 | epq->epi = NULL; | |
1284 | return; | |
296e236e | 1285 | } |
364f374f AV |
1286 | |
1287 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); | |
1288 | pwq->whead = whead; | |
1289 | pwq->base = epi; | |
1290 | if (epi->event.events & EPOLLEXCLUSIVE) | |
1291 | add_wait_queue_exclusive(whead, &pwq->wait); | |
1292 | else | |
1293 | add_wait_queue(whead, &pwq->wait); | |
1294 | pwq->next = epi->pwqlist; | |
1295 | epi->pwqlist = pwq; | |
1da177e4 LT |
1296 | } |
1297 | ||
1da177e4 LT |
1298 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1299 | { | |
1300 | int kcmp; | |
b2ac2ea6 | 1301 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
1da177e4 | 1302 | struct epitem *epic; |
b2ac2ea6 | 1303 | bool leftmost = true; |
1da177e4 LT |
1304 | |
1305 | while (*p) { | |
1306 | parent = *p; | |
1307 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1308 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
b2ac2ea6 | 1309 | if (kcmp > 0) { |
1da177e4 | 1310 | p = &parent->rb_right; |
b2ac2ea6 DB |
1311 | leftmost = false; |
1312 | } else | |
1da177e4 LT |
1313 | p = &parent->rb_left; |
1314 | } | |
1315 | rb_link_node(&epi->rbn, parent, p); | |
b2ac2ea6 | 1316 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
1da177e4 LT |
1317 | } |
1318 | ||
a80a6b85 AM |
1319 | |
1320 | ||
28d82dc1 JB |
1321 | #define PATH_ARR_SIZE 5 |
1322 | /* | |
1323 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1324 | * from a single file of interest. For example, we allow 1000 paths of length | |
1325 | * 1, to emanate from each file of interest. This essentially represents the | |
1326 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1327 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1328 | * is connected to n file sources. In this case each file source has 1 path | |
1329 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1330 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
d4cb626d | 1331 | * and delete can't add additional paths. Protected by the epnested_mutex. |
28d82dc1 JB |
1332 | */ |
1333 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1334 | static int path_count[PATH_ARR_SIZE]; | |
1335 | ||
1336 | static int path_count_inc(int nests) | |
1337 | { | |
93dc6107 JB |
1338 | /* Allow an arbitrary number of depth 1 paths */ |
1339 | if (nests == 0) | |
1340 | return 0; | |
1341 | ||
28d82dc1 JB |
1342 | if (++path_count[nests] > path_limits[nests]) |
1343 | return -1; | |
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | static void path_count_init(void) | |
1348 | { | |
1349 | int i; | |
1350 | ||
1351 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1352 | path_count[i] = 0; | |
1353 | } | |
1354 | ||
319c1517 | 1355 | static int reverse_path_check_proc(struct hlist_head *refs, int depth) |
28d82dc1 JB |
1356 | { |
1357 | int error = 0; | |
28d82dc1 JB |
1358 | struct epitem *epi; |
1359 | ||
0c320f77 | 1360 | if (depth > EP_MAX_NESTS) /* too deep nesting */ |
99d84d43 AV |
1361 | return -1; |
1362 | ||
ae10b2b4 | 1363 | /* CTL_DEL can remove links here, but that can't increase our count */ |
319c1517 AV |
1364 | hlist_for_each_entry_rcu(epi, refs, fllink) { |
1365 | struct hlist_head *refs = &epi->ep->refs; | |
1366 | if (hlist_empty(refs)) | |
d16312a4 AV |
1367 | error = path_count_inc(depth); |
1368 | else | |
319c1517 | 1369 | error = reverse_path_check_proc(refs, depth + 1); |
d16312a4 AV |
1370 | if (error != 0) |
1371 | break; | |
28d82dc1 JB |
1372 | } |
1373 | return error; | |
1374 | } | |
1375 | ||
1376 | /** | |
319c1517 | 1377 | * reverse_path_check - The tfile_check_list is list of epitem_head, which have |
28d82dc1 JB |
1378 | * links that are proposed to be newly added. We need to |
1379 | * make sure that those added links don't add too many | |
1380 | * paths such that we will spend all our time waking up | |
1381 | * eventpoll objects. | |
1382 | * | |
a6c67fee RD |
1383 | * Return: %zero if the proposed links don't create too many paths, |
1384 | * %-1 otherwise. | |
28d82dc1 JB |
1385 | */ |
1386 | static int reverse_path_check(void) | |
1387 | { | |
319c1517 | 1388 | struct epitems_head *p; |
28d82dc1 | 1389 | |
319c1517 AV |
1390 | for (p = tfile_check_list; p != EP_UNACTIVE_PTR; p = p->next) { |
1391 | int error; | |
28d82dc1 | 1392 | path_count_init(); |
b62d2706 | 1393 | rcu_read_lock(); |
319c1517 | 1394 | error = reverse_path_check_proc(&p->epitems, 0); |
b62d2706 | 1395 | rcu_read_unlock(); |
28d82dc1 | 1396 | if (error) |
319c1517 | 1397 | return error; |
28d82dc1 | 1398 | } |
319c1517 | 1399 | return 0; |
28d82dc1 JB |
1400 | } |
1401 | ||
4d7e30d9 AH |
1402 | static int ep_create_wakeup_source(struct epitem *epi) |
1403 | { | |
3701cb59 | 1404 | struct name_snapshot n; |
eea1d585 | 1405 | struct wakeup_source *ws; |
4d7e30d9 AH |
1406 | |
1407 | if (!epi->ep->ws) { | |
c8377adf | 1408 | epi->ep->ws = wakeup_source_register(NULL, "eventpoll"); |
4d7e30d9 AH |
1409 | if (!epi->ep->ws) |
1410 | return -ENOMEM; | |
1411 | } | |
1412 | ||
3701cb59 AV |
1413 | take_dentry_name_snapshot(&n, epi->ffd.file->f_path.dentry); |
1414 | ws = wakeup_source_register(NULL, n.name.name); | |
1415 | release_dentry_name_snapshot(&n); | |
eea1d585 EW |
1416 | |
1417 | if (!ws) | |
4d7e30d9 | 1418 | return -ENOMEM; |
eea1d585 | 1419 | rcu_assign_pointer(epi->ws, ws); |
4d7e30d9 AH |
1420 | |
1421 | return 0; | |
1422 | } | |
1423 | ||
eea1d585 EW |
1424 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
1425 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) | |
4d7e30d9 | 1426 | { |
eea1d585 EW |
1427 | struct wakeup_source *ws = ep_wakeup_source(epi); |
1428 | ||
d6d67e72 | 1429 | RCU_INIT_POINTER(epi->ws, NULL); |
eea1d585 EW |
1430 | |
1431 | /* | |
1432 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is | |
1433 | * used internally by wakeup_source_remove, too (called by | |
1434 | * wakeup_source_unregister), so we cannot use call_rcu | |
1435 | */ | |
1436 | synchronize_rcu(); | |
1437 | wakeup_source_unregister(ws); | |
4d7e30d9 AH |
1438 | } |
1439 | ||
319c1517 AV |
1440 | static int attach_epitem(struct file *file, struct epitem *epi) |
1441 | { | |
1442 | struct epitems_head *to_free = NULL; | |
1443 | struct hlist_head *head = NULL; | |
1444 | struct eventpoll *ep = NULL; | |
1445 | ||
1446 | if (is_file_epoll(file)) | |
1447 | ep = file->private_data; | |
1448 | ||
1449 | if (ep) { | |
1450 | head = &ep->refs; | |
1451 | } else if (!READ_ONCE(file->f_ep)) { | |
1452 | allocate: | |
1453 | to_free = kmem_cache_zalloc(ephead_cache, GFP_KERNEL); | |
1454 | if (!to_free) | |
1455 | return -ENOMEM; | |
1456 | head = &to_free->epitems; | |
1457 | } | |
1458 | spin_lock(&file->f_lock); | |
1459 | if (!file->f_ep) { | |
1460 | if (unlikely(!head)) { | |
1461 | spin_unlock(&file->f_lock); | |
1462 | goto allocate; | |
1463 | } | |
1464 | file->f_ep = head; | |
1465 | to_free = NULL; | |
1466 | } | |
1467 | hlist_add_head_rcu(&epi->fllink, file->f_ep); | |
1468 | spin_unlock(&file->f_lock); | |
1469 | free_ephead(to_free); | |
1470 | return 0; | |
1471 | } | |
1472 | ||
c7ea7630 DL |
1473 | /* |
1474 | * Must be called with "mtx" held. | |
1475 | */ | |
bec1a502 | 1476 | static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, |
67347fe4 | 1477 | struct file *tfile, int fd, int full_check) |
1da177e4 | 1478 | { |
d85e2aa2 AV |
1479 | int error, pwake = 0; |
1480 | __poll_t revents; | |
1da177e4 LT |
1481 | struct epitem *epi; |
1482 | struct ep_pqueue epq; | |
85353e91 AV |
1483 | struct eventpoll *tep = NULL; |
1484 | ||
1485 | if (is_file_epoll(tfile)) | |
1486 | tep = tfile->private_data; | |
1da177e4 | 1487 | |
92e64178 DB |
1488 | lockdep_assert_irqs_enabled(); |
1489 | ||
1e1c1583 NP |
1490 | if (unlikely(percpu_counter_compare(&ep->user->epoll_watches, |
1491 | max_user_watches) >= 0)) | |
7ef9964e | 1492 | return -ENOSPC; |
1e1c1583 NP |
1493 | percpu_counter_inc(&ep->user->epoll_watches); |
1494 | ||
1495 | if (!(epi = kmem_cache_zalloc(epi_cache, GFP_KERNEL))) { | |
1496 | percpu_counter_dec(&ep->user->epoll_watches); | |
7ef9964e | 1497 | return -ENOMEM; |
1e1c1583 | 1498 | } |
1da177e4 LT |
1499 | |
1500 | /* Item initialization follow here ... */ | |
1da177e4 | 1501 | INIT_LIST_HEAD(&epi->rdllink); |
1da177e4 | 1502 | epi->ep = ep; |
b030a4dd | 1503 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1504 | epi->event = *event; |
d47de16c | 1505 | epi->next = EP_UNACTIVE_PTR; |
1da177e4 | 1506 | |
85353e91 AV |
1507 | if (tep) |
1508 | mutex_lock_nested(&tep->mtx, 1); | |
f8d4f44d | 1509 | /* Add the current item to the list of active epoll hook for this file */ |
319c1517 | 1510 | if (unlikely(attach_epitem(tfile, epi) < 0)) { |
319c1517 AV |
1511 | if (tep) |
1512 | mutex_unlock(&tep->mtx); | |
1e1c1583 NP |
1513 | kmem_cache_free(epi_cache, epi); |
1514 | percpu_counter_dec(&ep->user->epoll_watches); | |
319c1517 | 1515 | return -ENOMEM; |
d9f41e3c | 1516 | } |
f8d4f44d | 1517 | |
319c1517 AV |
1518 | if (full_check && !tep) |
1519 | list_file(tfile); | |
1520 | ||
f8d4f44d AV |
1521 | /* |
1522 | * Add the current item to the RB tree. All RB tree operations are | |
1523 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1524 | */ | |
1525 | ep_rbtree_insert(ep, epi); | |
85353e91 AV |
1526 | if (tep) |
1527 | mutex_unlock(&tep->mtx); | |
f8d4f44d | 1528 | |
58c9b016 PA |
1529 | /* |
1530 | * ep_remove_safe() calls in the later error paths can't lead to | |
1531 | * ep_free() as the ep file itself still holds an ep reference. | |
1532 | */ | |
1533 | ep_get(ep); | |
1534 | ||
f8d4f44d | 1535 | /* now check if we've created too many backpaths */ |
e3e096e7 | 1536 | if (unlikely(full_check && reverse_path_check())) { |
58c9b016 | 1537 | ep_remove_safe(ep, epi); |
e3e096e7 AV |
1538 | return -EINVAL; |
1539 | } | |
f8d4f44d | 1540 | |
d1ec50ad AV |
1541 | if (epi->event.events & EPOLLWAKEUP) { |
1542 | error = ep_create_wakeup_source(epi); | |
1543 | if (error) { | |
58c9b016 | 1544 | ep_remove_safe(ep, epi); |
d1ec50ad AV |
1545 | return error; |
1546 | } | |
1547 | } | |
f8d4f44d | 1548 | |
1da177e4 LT |
1549 | /* Initialize the poll table using the queue callback */ |
1550 | epq.epi = epi; | |
1551 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
1552 | ||
1553 | /* | |
1554 | * Attach the item to the poll hooks and get current event bits. | |
1555 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1556 | * been increased by the caller of this function. Note that after |
1557 | * this operation completes, the poll callback can start hitting | |
1558 | * the new item. | |
1da177e4 | 1559 | */ |
37b5e521 | 1560 | revents = ep_item_poll(epi, &epq.pt, 1); |
1da177e4 LT |
1561 | |
1562 | /* | |
1563 | * We have to check if something went wrong during the poll wait queue | |
1564 | * install process. Namely an allocation for a wait queue failed due | |
1565 | * high memory pressure. | |
1566 | */ | |
e3e096e7 | 1567 | if (unlikely(!epq.epi)) { |
58c9b016 | 1568 | ep_remove_safe(ep, epi); |
e3e096e7 AV |
1569 | return -ENOMEM; |
1570 | } | |
1da177e4 | 1571 | |
c7ea7630 | 1572 | /* We have to drop the new item inside our item list to keep track of it */ |
a218cc49 | 1573 | write_lock_irq(&ep->lock); |
c7ea7630 | 1574 | |
bf3b9f63 SS |
1575 | /* record NAPI ID of new item if present */ |
1576 | ep_set_busy_poll_napi_id(epi); | |
1577 | ||
1da177e4 | 1578 | /* If the file is already "ready" we drop it inside the ready list */ |
992991c0 | 1579 | if (revents && !ep_is_linked(epi)) { |
1da177e4 | 1580 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1581 | ep_pm_stay_awake(epi); |
1da177e4 LT |
1582 | |
1583 | /* Notify waiting tasks that events are available */ | |
1584 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1585 | wake_up(&ep->wq); |
1da177e4 LT |
1586 | if (waitqueue_active(&ep->poll_wait)) |
1587 | pwake++; | |
1588 | } | |
1589 | ||
a218cc49 | 1590 | write_unlock_irq(&ep->lock); |
1da177e4 LT |
1591 | |
1592 | /* We have to call this outside the lock */ | |
1593 | if (pwake) | |
caf1aeaf | 1594 | ep_poll_safewake(ep, NULL, 0); |
1da177e4 | 1595 | |
1da177e4 | 1596 | return 0; |
1da177e4 LT |
1597 | } |
1598 | ||
1da177e4 LT |
1599 | /* |
1600 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1601 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 | 1602 | */ |
bec1a502 AV |
1603 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, |
1604 | const struct epoll_event *event) | |
1da177e4 LT |
1605 | { |
1606 | int pwake = 0; | |
626cf236 HV |
1607 | poll_table pt; |
1608 | ||
92e64178 DB |
1609 | lockdep_assert_irqs_enabled(); |
1610 | ||
626cf236 | 1611 | init_poll_funcptr(&pt, NULL); |
1da177e4 LT |
1612 | |
1613 | /* | |
e057e15f TB |
1614 | * Set the new event interest mask before calling f_op->poll(); |
1615 | * otherwise we might miss an event that happens between the | |
1616 | * f_op->poll() call and the new event set registering. | |
1da177e4 | 1617 | */ |
128dd175 | 1618 | epi->event.events = event->events; /* need barrier below */ |
e057e15f | 1619 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 | 1620 | if (epi->event.events & EPOLLWAKEUP) { |
eea1d585 | 1621 | if (!ep_has_wakeup_source(epi)) |
4d7e30d9 | 1622 | ep_create_wakeup_source(epi); |
eea1d585 | 1623 | } else if (ep_has_wakeup_source(epi)) { |
4d7e30d9 AH |
1624 | ep_destroy_wakeup_source(epi); |
1625 | } | |
1da177e4 | 1626 | |
128dd175 EW |
1627 | /* |
1628 | * The following barrier has two effects: | |
1629 | * | |
1630 | * 1) Flush epi changes above to other CPUs. This ensures | |
1631 | * we do not miss events from ep_poll_callback if an | |
1632 | * event occurs immediately after we call f_op->poll(). | |
a218cc49 | 1633 | * We need this because we did not take ep->lock while |
128dd175 | 1634 | * changing epi above (but ep_poll_callback does take |
a218cc49 | 1635 | * ep->lock). |
128dd175 EW |
1636 | * |
1637 | * 2) We also need to ensure we do not miss _past_ events | |
1638 | * when calling f_op->poll(). This barrier also | |
1639 | * pairs with the barrier in wq_has_sleeper (see | |
1640 | * comments for wq_has_sleeper). | |
1641 | * | |
1642 | * This barrier will now guarantee ep_poll_callback or f_op->poll | |
1643 | * (or both) will notice the readiness of an item. | |
1644 | */ | |
1645 | smp_mb(); | |
1646 | ||
1da177e4 LT |
1647 | /* |
1648 | * Get current event bits. We can safely use the file* here because | |
1649 | * its usage count has been increased by the caller of this function. | |
c7ea7630 | 1650 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1651 | * list, push it inside. |
1da177e4 | 1652 | */ |
69112736 | 1653 | if (ep_item_poll(epi, &pt, 1)) { |
a218cc49 | 1654 | write_lock_irq(&ep->lock); |
992991c0 | 1655 | if (!ep_is_linked(epi)) { |
c7ea7630 | 1656 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1657 | ep_pm_stay_awake(epi); |
c7ea7630 DL |
1658 | |
1659 | /* Notify waiting tasks that events are available */ | |
1660 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1661 | wake_up(&ep->wq); |
c7ea7630 DL |
1662 | if (waitqueue_active(&ep->poll_wait)) |
1663 | pwake++; | |
7699acd1 | 1664 | } |
a218cc49 | 1665 | write_unlock_irq(&ep->lock); |
7699acd1 | 1666 | } |
1da177e4 | 1667 | |
7699acd1 DL |
1668 | /* We have to call this outside the lock */ |
1669 | if (pwake) | |
caf1aeaf | 1670 | ep_poll_safewake(ep, NULL, 0); |
1da177e4 | 1671 | |
7699acd1 | 1672 | return 0; |
1da177e4 LT |
1673 | } |
1674 | ||
ff07952a AV |
1675 | static int ep_send_events(struct eventpoll *ep, |
1676 | struct epoll_event __user *events, int maxevents) | |
1da177e4 | 1677 | { |
4e0982a0 | 1678 | struct epitem *epi, *tmp; |
ff07952a | 1679 | LIST_HEAD(txlist); |
626cf236 | 1680 | poll_table pt; |
ff07952a | 1681 | int res = 0; |
626cf236 | 1682 | |
cccd29bf SHY |
1683 | /* |
1684 | * Always short-circuit for fatal signals to allow threads to make a | |
1685 | * timely exit without the chance of finding more events available and | |
1686 | * fetching repeatedly. | |
1687 | */ | |
1688 | if (fatal_signal_pending(current)) | |
1689 | return -EINTR; | |
1690 | ||
626cf236 | 1691 | init_poll_funcptr(&pt, NULL); |
ff07952a | 1692 | |
57804b1c AV |
1693 | mutex_lock(&ep->mtx); |
1694 | ep_start_scan(ep, &txlist); | |
1da177e4 | 1695 | |
296e236e | 1696 | /* |
5071f97e | 1697 | * We can loop without lock because we are passed a task private list. |
57804b1c | 1698 | * Items cannot vanish during the loop we are holding ep->mtx. |
296e236e | 1699 | */ |
ff07952a AV |
1700 | list_for_each_entry_safe(epi, tmp, &txlist, rdllink) { |
1701 | struct wakeup_source *ws; | |
1702 | __poll_t revents; | |
21877e1a | 1703 | |
ff07952a | 1704 | if (res >= maxevents) |
4e0982a0 | 1705 | break; |
d47de16c | 1706 | |
4d7e30d9 AH |
1707 | /* |
1708 | * Activate ep->ws before deactivating epi->ws to prevent | |
1709 | * triggering auto-suspend here (in case we reactive epi->ws | |
1710 | * below). | |
1711 | * | |
1712 | * This could be rearranged to delay the deactivation of epi->ws | |
1713 | * instead, but then epi->ws would temporarily be out of sync | |
1714 | * with ep_is_linked(). | |
1715 | */ | |
eea1d585 EW |
1716 | ws = ep_wakeup_source(epi); |
1717 | if (ws) { | |
1718 | if (ws->active) | |
1719 | __pm_stay_awake(ep->ws); | |
1720 | __pm_relax(ws); | |
1721 | } | |
1722 | ||
d47de16c | 1723 | list_del_init(&epi->rdllink); |
1da177e4 | 1724 | |
296e236e | 1725 | /* |
5071f97e | 1726 | * If the event mask intersect the caller-requested one, |
57804b1c AV |
1727 | * deliver the event to userspace. Again, we are holding ep->mtx, |
1728 | * so no operations coming from userspace can change the item. | |
296e236e | 1729 | */ |
4e0982a0 DB |
1730 | revents = ep_item_poll(epi, &pt, 1); |
1731 | if (!revents) | |
1732 | continue; | |
1733 | ||
249dbe74 AB |
1734 | events = epoll_put_uevent(revents, epi->event.data, events); |
1735 | if (!events) { | |
ff07952a | 1736 | list_add(&epi->rdllink, &txlist); |
4e0982a0 | 1737 | ep_pm_stay_awake(epi); |
ff07952a AV |
1738 | if (!res) |
1739 | res = -EFAULT; | |
1740 | break; | |
4e0982a0 | 1741 | } |
ff07952a | 1742 | res++; |
4e0982a0 DB |
1743 | if (epi->event.events & EPOLLONESHOT) |
1744 | epi->event.events &= EP_PRIVATE_BITS; | |
1745 | else if (!(epi->event.events & EPOLLET)) { | |
1746 | /* | |
1747 | * If this file has been added with Level | |
1748 | * Trigger mode, we need to insert back inside | |
1749 | * the ready list, so that the next call to | |
1750 | * epoll_wait() will check again the events | |
1751 | * availability. At this point, no one can insert | |
1752 | * into ep->rdllist besides us. The epoll_ctl() | |
1753 | * callers are locked out by | |
1754 | * ep_scan_ready_list() holding "mtx" and the | |
1755 | * poll callback will queue them in ep->ovflist. | |
1756 | */ | |
1757 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1758 | ep_pm_stay_awake(epi); | |
296e236e DL |
1759 | } |
1760 | } | |
57804b1c AV |
1761 | ep_done_scan(ep, &txlist); |
1762 | mutex_unlock(&ep->mtx); | |
5071f97e | 1763 | |
ff07952a | 1764 | return res; |
1da177e4 LT |
1765 | } |
1766 | ||
7cdf7c20 | 1767 | static struct timespec64 *ep_timeout_to_timespec(struct timespec64 *to, long ms) |
0781b909 | 1768 | { |
7cdf7c20 WB |
1769 | struct timespec64 now; |
1770 | ||
1771 | if (ms < 0) | |
1772 | return NULL; | |
1773 | ||
1774 | if (!ms) { | |
1775 | to->tv_sec = 0; | |
1776 | to->tv_nsec = 0; | |
1777 | return to; | |
1778 | } | |
1779 | ||
1780 | to->tv_sec = ms / MSEC_PER_SEC; | |
1781 | to->tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC); | |
0781b909 | 1782 | |
766b9f92 | 1783 | ktime_get_ts64(&now); |
7cdf7c20 WB |
1784 | *to = timespec64_add_safe(now, *to); |
1785 | return to; | |
0781b909 ED |
1786 | } |
1787 | ||
a16ceb13 BS |
1788 | /* |
1789 | * autoremove_wake_function, but remove even on failure to wake up, because we | |
1790 | * know that default_wake_function/ttwu will only fail if the thread is already | |
1791 | * woken, and in that case the ep_poll loop will remove the entry anyways, not | |
1792 | * try to reuse it. | |
1793 | */ | |
1794 | static int ep_autoremove_wake_function(struct wait_queue_entry *wq_entry, | |
1795 | unsigned int mode, int sync, void *key) | |
1796 | { | |
1797 | int ret = default_wake_function(wq_entry, mode, sync, key); | |
1798 | ||
2192bba0 BS |
1799 | /* |
1800 | * Pairs with list_empty_careful in ep_poll, and ensures future loop | |
1801 | * iterations see the cause of this wakeup. | |
1802 | */ | |
1803 | list_del_init_careful(&wq_entry->entry); | |
a16ceb13 BS |
1804 | return ret; |
1805 | } | |
1806 | ||
f4d93ad7 | 1807 | /** |
a6c67fee | 1808 | * ep_poll - Retrieves ready events, and delivers them to the caller-supplied |
f4d93ad7 SB |
1809 | * event buffer. |
1810 | * | |
1811 | * @ep: Pointer to the eventpoll context. | |
1812 | * @events: Pointer to the userspace buffer where the ready events should be | |
1813 | * stored. | |
1814 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1815 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
7cdf7c20 WB |
1816 | * timespec. If the timeout is zero, the function will not block, |
1817 | * while if the @timeout ptr is NULL, the function will block | |
f4d93ad7 SB |
1818 | * until at least one event has been retrieved (or an error |
1819 | * occurred). | |
1820 | * | |
a6c67fee | 1821 | * Return: the number of ready events which have been fetched, or an |
f4d93ad7 SB |
1822 | * error code, in case of error. |
1823 | */ | |
1da177e4 | 1824 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
7cdf7c20 | 1825 | int maxevents, struct timespec64 *timeout) |
1da177e4 | 1826 | { |
e59d3c64 | 1827 | int res, eavail, timed_out = 0; |
da8b44d5 | 1828 | u64 slack = 0; |
ac6424b9 | 1829 | wait_queue_entry_t wait; |
95aac7b1 SB |
1830 | ktime_t expires, *to = NULL; |
1831 | ||
679abf38 DB |
1832 | lockdep_assert_irqs_enabled(); |
1833 | ||
7cdf7c20 WB |
1834 | if (timeout && (timeout->tv_sec | timeout->tv_nsec)) { |
1835 | slack = select_estimate_accuracy(timeout); | |
95aac7b1 | 1836 | to = &expires; |
7cdf7c20 WB |
1837 | *to = timespec64_to_ktime(*timeout); |
1838 | } else if (timeout) { | |
f4d93ad7 SB |
1839 | /* |
1840 | * Avoid the unnecessary trip to the wait queue loop, if the | |
e59d3c64 | 1841 | * caller specified a non blocking operation. |
f4d93ad7 | 1842 | */ |
95aac7b1 SB |
1843 | timed_out = 1; |
1844 | } | |
1da177e4 | 1845 | |
e59d3c64 SHY |
1846 | /* |
1847 | * This call is racy: We may or may not see events that are being added | |
a6c67fee | 1848 | * to the ready list under the lock (e.g., in IRQ callbacks). For cases |
e59d3c64 | 1849 | * with a non-zero timeout, this thread will check the ready list under |
a6c67fee | 1850 | * lock and will add to the wait queue. For cases with a zero |
e59d3c64 SHY |
1851 | * timeout, the user by definition should not care and will have to |
1852 | * recheck again. | |
1853 | */ | |
1854 | eavail = ep_events_available(ep); | |
1855 | ||
00b27634 SHY |
1856 | while (1) { |
1857 | if (eavail) { | |
1858 | /* | |
1859 | * Try to transfer events to user space. In case we get | |
1860 | * 0 events and there's still timeout left over, we go | |
1861 | * trying again in search of more luck. | |
1862 | */ | |
1863 | res = ep_send_events(ep, events, maxevents); | |
1864 | if (res) | |
1865 | return res; | |
1866 | } | |
1867 | ||
1868 | if (timed_out) | |
1869 | return 0; | |
1870 | ||
00b27634 | 1871 | eavail = ep_busy_loop(ep, timed_out); |
e8c85328 | 1872 | if (eavail) |
00b27634 | 1873 | continue; |
1da177e4 | 1874 | |
2efdaf76 SHY |
1875 | if (signal_pending(current)) |
1876 | return -EINTR; | |
1877 | ||
412895f0 RP |
1878 | /* |
1879 | * Internally init_wait() uses autoremove_wake_function(), | |
1880 | * thus wait entry is removed from the wait queue on each | |
1881 | * wakeup. Why it is important? In case of several waiters | |
1882 | * each new wakeup will hit the next waiter, giving it the | |
1883 | * chance to harvest new event. Otherwise wakeup can be | |
1884 | * lost. This is also good performance-wise, because on | |
1885 | * normal wakeup path no need to call __remove_wait_queue() | |
1886 | * explicitly, thus ep->lock is not taken, which halts the | |
1887 | * event delivery. | |
a16ceb13 BS |
1888 | * |
1889 | * In fact, we now use an even more aggressive function that | |
1890 | * unconditionally removes, because we don't reuse the wait | |
1891 | * entry between loop iterations. This lets us also avoid the | |
1892 | * performance issue if a process is killed, causing all of its | |
1893 | * threads to wake up without being removed normally. | |
412895f0 RP |
1894 | */ |
1895 | init_wait(&wait); | |
a16ceb13 | 1896 | wait.func = ep_autoremove_wake_function; |
1da177e4 | 1897 | |
65759097 | 1898 | write_lock_irq(&ep->lock); |
bf3b9f63 | 1899 | /* |
65759097 RP |
1900 | * Barrierless variant, waitqueue_active() is called under |
1901 | * the same lock on wakeup ep_poll_callback() side, so it | |
1902 | * is safe to avoid an explicit barrier. | |
bf3b9f63 | 1903 | */ |
65759097 RP |
1904 | __set_current_state(TASK_INTERRUPTIBLE); |
1905 | ||
1da177e4 | 1906 | /* |
65759097 RP |
1907 | * Do the final check under the lock. ep_scan_ready_list() |
1908 | * plays with two lists (->rdllist and ->ovflist) and there | |
1909 | * is always a race when both lists are empty for short | |
1910 | * period of time although events are pending, so lock is | |
1911 | * important. | |
1da177e4 | 1912 | */ |
65759097 | 1913 | eavail = ep_events_available(ep); |
2efdaf76 SHY |
1914 | if (!eavail) |
1915 | __add_wait_queue_exclusive(&ep->wq, &wait); | |
1916 | ||
65759097 | 1917 | write_unlock_irq(&ep->lock); |
95aac7b1 | 1918 | |
2efdaf76 | 1919 | if (!eavail) |
289caf5d SHY |
1920 | timed_out = !schedule_hrtimeout_range(to, slack, |
1921 | HRTIMER_MODE_ABS); | |
e411596d | 1922 | __set_current_state(TASK_RUNNING); |
1da177e4 | 1923 | |
289caf5d SHY |
1924 | /* |
1925 | * We were woken up, thus go and try to harvest some events. | |
1926 | * If timed out and still on the wait queue, recheck eavail | |
1927 | * carefully under lock, below. | |
1928 | */ | |
412895f0 | 1929 | eavail = 1; |
1da177e4 | 1930 | |
e8c85328 SHY |
1931 | if (!list_empty_careful(&wait.entry)) { |
1932 | write_lock_irq(&ep->lock); | |
1933 | /* | |
1934 | * If the thread timed out and is not on the wait queue, | |
1935 | * it means that the thread was woken up after its | |
1936 | * timeout expired before it could reacquire the lock. | |
1937 | * Thus, when wait.entry is empty, it needs to harvest | |
1938 | * events. | |
1939 | */ | |
1940 | if (timed_out) | |
1941 | eavail = list_empty(&wait.entry); | |
1942 | __remove_wait_queue(&ep->wq, &wait); | |
1943 | write_unlock_irq(&ep->lock); | |
1944 | } | |
00b27634 | 1945 | } |
1da177e4 LT |
1946 | } |
1947 | ||
22bacca4 | 1948 | /** |
773318ed | 1949 | * ep_loop_check_proc - verify that adding an epoll file inside another |
a6c67fee | 1950 | * epoll structure does not violate the constraints, in |
22bacca4 DL |
1951 | * terms of closed loops, or too deep chains (which can |
1952 | * result in excessive stack usage). | |
1953 | * | |
a6c67fee | 1954 | * @ep: the &struct eventpoll to be currently checked. |
bde03c4c | 1955 | * @depth: Current depth of the path being checked. |
22bacca4 | 1956 | * |
a6c67fee RD |
1957 | * Return: %zero if adding the epoll @file inside current epoll |
1958 | * structure @ep does not violate the constraints, or %-1 otherwise. | |
22bacca4 | 1959 | */ |
bde03c4c | 1960 | static int ep_loop_check_proc(struct eventpoll *ep, int depth) |
22bacca4 DL |
1961 | { |
1962 | int error = 0; | |
22bacca4 DL |
1963 | struct rb_node *rbp; |
1964 | struct epitem *epi; | |
1965 | ||
773318ed | 1966 | mutex_lock_nested(&ep->mtx, depth + 1); |
18306c40 | 1967 | ep->gen = loop_check_gen; |
b2ac2ea6 | 1968 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
22bacca4 DL |
1969 | epi = rb_entry(rbp, struct epitem, rbn); |
1970 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
bde03c4c | 1971 | struct eventpoll *ep_tovisit; |
28d82dc1 | 1972 | ep_tovisit = epi->ffd.file->private_data; |
18306c40 | 1973 | if (ep_tovisit->gen == loop_check_gen) |
28d82dc1 | 1974 | continue; |
bde03c4c | 1975 | if (ep_tovisit == inserting_into || depth > EP_MAX_NESTS) |
56c428ca | 1976 | error = -1; |
bde03c4c AV |
1977 | else |
1978 | error = ep_loop_check_proc(ep_tovisit, depth + 1); | |
22bacca4 DL |
1979 | if (error != 0) |
1980 | break; | |
28d82dc1 JB |
1981 | } else { |
1982 | /* | |
1983 | * If we've reached a file that is not associated with | |
1984 | * an ep, then we need to check if the newly added | |
1985 | * links are going to add too many wakeup paths. We do | |
1986 | * this by adding it to the tfile_check_list, if it's | |
1987 | * not already there, and calling reverse_path_check() | |
1988 | * during ep_insert(). | |
1989 | */ | |
319c1517 | 1990 | list_file(epi->ffd.file); |
22bacca4 DL |
1991 | } |
1992 | } | |
1993 | mutex_unlock(&ep->mtx); | |
1994 | ||
1995 | return error; | |
1996 | } | |
1997 | ||
1998 | /** | |
bde03c4c | 1999 | * ep_loop_check - Performs a check to verify that adding an epoll file (@to) |
a6c67fee | 2000 | * into another epoll file (represented by @ep) does not create |
22bacca4 DL |
2001 | * closed loops or too deep chains. |
2002 | * | |
a6c67fee | 2003 | * @ep: Pointer to the epoll we are inserting into. |
bde03c4c | 2004 | * @to: Pointer to the epoll to be inserted. |
22bacca4 | 2005 | * |
a6c67fee RD |
2006 | * Return: %zero if adding the epoll @to inside the epoll @from |
2007 | * does not violate the constraints, or %-1 otherwise. | |
22bacca4 | 2008 | */ |
bde03c4c | 2009 | static int ep_loop_check(struct eventpoll *ep, struct eventpoll *to) |
22bacca4 | 2010 | { |
6a3890c4 | 2011 | inserting_into = ep; |
bde03c4c | 2012 | return ep_loop_check_proc(to, 0); |
28d82dc1 JB |
2013 | } |
2014 | ||
2015 | static void clear_tfile_check_list(void) | |
2016 | { | |
319c1517 AV |
2017 | rcu_read_lock(); |
2018 | while (tfile_check_list != EP_UNACTIVE_PTR) { | |
2019 | struct epitems_head *head = tfile_check_list; | |
2020 | tfile_check_list = head->next; | |
2021 | unlist_file(head); | |
28d82dc1 | 2022 | } |
319c1517 | 2023 | rcu_read_unlock(); |
22bacca4 DL |
2024 | } |
2025 | ||
7699acd1 | 2026 | /* |
523723bb | 2027 | * Open an eventpoll file descriptor. |
7699acd1 | 2028 | */ |
791eb22e | 2029 | static int do_epoll_create(int flags) |
7699acd1 | 2030 | { |
28d82dc1 | 2031 | int error, fd; |
bb57c3ed | 2032 | struct eventpoll *ep = NULL; |
28d82dc1 | 2033 | struct file *file; |
7699acd1 | 2034 | |
e38b36f3 UD |
2035 | /* Check the EPOLL_* constant for consistency. */ |
2036 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
2037 | ||
296e236e DL |
2038 | if (flags & ~EPOLL_CLOEXEC) |
2039 | return -EINVAL; | |
7699acd1 | 2040 | /* |
bb57c3ed | 2041 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 2042 | */ |
9fe5ad9c | 2043 | error = ep_alloc(&ep); |
bb57c3ed DL |
2044 | if (error < 0) |
2045 | return error; | |
7699acd1 DL |
2046 | /* |
2047 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 2048 | * a file structure and a free file descriptor. |
7699acd1 | 2049 | */ |
28d82dc1 JB |
2050 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
2051 | if (fd < 0) { | |
2052 | error = fd; | |
2053 | goto out_free_ep; | |
2054 | } | |
2055 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 2056 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
2057 | if (IS_ERR(file)) { |
2058 | error = PTR_ERR(file); | |
2059 | goto out_free_fd; | |
2060 | } | |
28d82dc1 | 2061 | ep->file = file; |
98022748 | 2062 | fd_install(fd, file); |
28d82dc1 JB |
2063 | return fd; |
2064 | ||
2065 | out_free_fd: | |
2066 | put_unused_fd(fd); | |
2067 | out_free_ep: | |
58c9b016 | 2068 | ep_clear_and_put(ep); |
bb57c3ed | 2069 | return error; |
7699acd1 DL |
2070 | } |
2071 | ||
791eb22e DB |
2072 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
2073 | { | |
2074 | return do_epoll_create(flags); | |
2075 | } | |
2076 | ||
5a8a82b1 | 2077 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 2078 | { |
bfe3891a | 2079 | if (size <= 0) |
9fe5ad9c UD |
2080 | return -EINVAL; |
2081 | ||
791eb22e | 2082 | return do_epoll_create(0); |
a0998b50 UD |
2083 | } |
2084 | ||
063f3ed9 PD |
2085 | #ifdef CONFIG_PM_SLEEP |
2086 | static inline void ep_take_care_of_epollwakeup(struct epoll_event *epev) | |
2087 | { | |
2088 | if ((epev->events & EPOLLWAKEUP) && !capable(CAP_BLOCK_SUSPEND)) | |
2089 | epev->events &= ~EPOLLWAKEUP; | |
2090 | } | |
2091 | #else | |
2092 | static inline void ep_take_care_of_epollwakeup(struct epoll_event *epev) | |
2093 | { | |
2094 | epev->events &= ~EPOLLWAKEUP; | |
2095 | } | |
2096 | #endif | |
2097 | ||
39220e8d JA |
2098 | static inline int epoll_mutex_lock(struct mutex *mutex, int depth, |
2099 | bool nonblock) | |
2100 | { | |
2101 | if (!nonblock) { | |
2102 | mutex_lock_nested(mutex, depth); | |
2103 | return 0; | |
2104 | } | |
2105 | if (mutex_trylock(mutex)) | |
2106 | return 0; | |
2107 | return -EAGAIN; | |
2108 | } | |
2109 | ||
2110 | int do_epoll_ctl(int epfd, int op, int fd, struct epoll_event *epds, | |
2111 | bool nonblock) | |
7699acd1 DL |
2112 | { |
2113 | int error; | |
67347fe4 | 2114 | int full_check = 0; |
7e3fb584 | 2115 | struct fd f, tf; |
7699acd1 DL |
2116 | struct eventpoll *ep; |
2117 | struct epitem *epi; | |
67347fe4 | 2118 | struct eventpoll *tep = NULL; |
7699acd1 | 2119 | |
7699acd1 | 2120 | error = -EBADF; |
7e3fb584 AV |
2121 | f = fdget(epfd); |
2122 | if (!f.file) | |
7699acd1 DL |
2123 | goto error_return; |
2124 | ||
2125 | /* Get the "struct file *" for the target file */ | |
7e3fb584 AV |
2126 | tf = fdget(fd); |
2127 | if (!tf.file) | |
7699acd1 DL |
2128 | goto error_fput; |
2129 | ||
2130 | /* The target file descriptor must support poll */ | |
2131 | error = -EPERM; | |
9965ed17 | 2132 | if (!file_can_poll(tf.file)) |
7699acd1 DL |
2133 | goto error_tgt_fput; |
2134 | ||
4d7e30d9 | 2135 | /* Check if EPOLLWAKEUP is allowed */ |
c680e41b | 2136 | if (ep_op_has_event(op)) |
58e41a44 | 2137 | ep_take_care_of_epollwakeup(epds); |
4d7e30d9 | 2138 | |
7699acd1 DL |
2139 | /* |
2140 | * We have to check that the file structure underneath the file descriptor | |
2141 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
2142 | * adding an epoll file descriptor inside itself. | |
2143 | */ | |
2144 | error = -EINVAL; | |
7e3fb584 | 2145 | if (f.file == tf.file || !is_file_epoll(f.file)) |
7699acd1 DL |
2146 | goto error_tgt_fput; |
2147 | ||
df0108c5 JB |
2148 | /* |
2149 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, | |
2150 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. | |
2151 | * Also, we do not currently supported nested exclusive wakeups. | |
2152 | */ | |
58e41a44 | 2153 | if (ep_op_has_event(op) && (epds->events & EPOLLEXCLUSIVE)) { |
b6a515c8 JB |
2154 | if (op == EPOLL_CTL_MOD) |
2155 | goto error_tgt_fput; | |
2156 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || | |
58e41a44 | 2157 | (epds->events & ~EPOLLEXCLUSIVE_OK_BITS))) |
b6a515c8 JB |
2158 | goto error_tgt_fput; |
2159 | } | |
df0108c5 | 2160 | |
7699acd1 DL |
2161 | /* |
2162 | * At this point it is safe to assume that the "private_data" contains | |
2163 | * our own data structure. | |
2164 | */ | |
7e3fb584 | 2165 | ep = f.file->private_data; |
7699acd1 | 2166 | |
22bacca4 | 2167 | /* |
a6c67fee RD |
2168 | * When we insert an epoll file descriptor inside another epoll file |
2169 | * descriptor, there is the chance of creating closed loops, which are | |
28d82dc1 JB |
2170 | * better be handled here, than in more critical paths. While we are |
2171 | * checking for loops we also determine the list of files reachable | |
2172 | * and hang them on the tfile_check_list, so we can check that we | |
2173 | * haven't created too many possible wakeup paths. | |
22bacca4 | 2174 | * |
67347fe4 JB |
2175 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
2176 | * the epoll file descriptor is attaching directly to a wakeup source, | |
2177 | * unless the epoll file descriptor is nested. The purpose of taking the | |
d4cb626d | 2178 | * 'epnested_mutex' on add is to prevent complex toplogies such as loops and |
67347fe4 JB |
2179 | * deep wakeup paths from forming in parallel through multiple |
2180 | * EPOLL_CTL_ADD operations. | |
22bacca4 | 2181 | */ |
39220e8d JA |
2182 | error = epoll_mutex_lock(&ep->mtx, 0, nonblock); |
2183 | if (error) | |
2184 | goto error_tgt_fput; | |
28d82dc1 | 2185 | if (op == EPOLL_CTL_ADD) { |
319c1517 AV |
2186 | if (READ_ONCE(f.file->f_ep) || ep->gen == loop_check_gen || |
2187 | is_file_epoll(tf.file)) { | |
67347fe4 | 2188 | mutex_unlock(&ep->mtx); |
d4cb626d | 2189 | error = epoll_mutex_lock(&epnested_mutex, 0, nonblock); |
39220e8d JA |
2190 | if (error) |
2191 | goto error_tgt_fput; | |
18306c40 | 2192 | loop_check_gen++; |
39220e8d | 2193 | full_check = 1; |
67347fe4 | 2194 | if (is_file_epoll(tf.file)) { |
bde03c4c | 2195 | tep = tf.file->private_data; |
67347fe4 | 2196 | error = -ELOOP; |
bde03c4c | 2197 | if (ep_loop_check(ep, tep) != 0) |
67347fe4 | 2198 | goto error_tgt_fput; |
a9ed4a65 | 2199 | } |
39220e8d | 2200 | error = epoll_mutex_lock(&ep->mtx, 0, nonblock); |
52c47969 | 2201 | if (error) |
39220e8d | 2202 | goto error_tgt_fput; |
67347fe4 JB |
2203 | } |
2204 | } | |
7699acd1 | 2205 | |
67647d0f | 2206 | /* |
a6c67fee | 2207 | * Try to lookup the file inside our RB tree. Since we grabbed "mtx" |
67647d0f DL |
2208 | * above, we can be sure to be able to use the item looked up by |
2209 | * ep_find() till we release the mutex. | |
2210 | */ | |
7e3fb584 | 2211 | epi = ep_find(ep, tf.file, fd); |
7699acd1 DL |
2212 | |
2213 | error = -EINVAL; | |
2214 | switch (op) { | |
2215 | case EPOLL_CTL_ADD: | |
2216 | if (!epi) { | |
58e41a44 JA |
2217 | epds->events |= EPOLLERR | EPOLLHUP; |
2218 | error = ep_insert(ep, epds, tf.file, fd, full_check); | |
7699acd1 DL |
2219 | } else |
2220 | error = -EEXIST; | |
2221 | break; | |
2222 | case EPOLL_CTL_DEL: | |
58c9b016 PA |
2223 | if (epi) { |
2224 | /* | |
2225 | * The eventpoll itself is still alive: the refcount | |
2226 | * can't go to zero here. | |
2227 | */ | |
2228 | ep_remove_safe(ep, epi); | |
2229 | error = 0; | |
2230 | } else { | |
7699acd1 | 2231 | error = -ENOENT; |
58c9b016 | 2232 | } |
7699acd1 DL |
2233 | break; |
2234 | case EPOLL_CTL_MOD: | |
2235 | if (epi) { | |
b6a515c8 | 2236 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
58e41a44 JA |
2237 | epds->events |= EPOLLERR | EPOLLHUP; |
2238 | error = ep_modify(ep, epi, epds); | |
b6a515c8 | 2239 | } |
7699acd1 DL |
2240 | } else |
2241 | error = -ENOENT; | |
2242 | break; | |
2243 | } | |
d47de16c | 2244 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
2245 | |
2246 | error_tgt_fput: | |
52c47969 AV |
2247 | if (full_check) { |
2248 | clear_tfile_check_list(); | |
18306c40 | 2249 | loop_check_gen++; |
d4cb626d | 2250 | mutex_unlock(&epnested_mutex); |
52c47969 | 2251 | } |
22bacca4 | 2252 | |
7e3fb584 | 2253 | fdput(tf); |
7699acd1 | 2254 | error_fput: |
7e3fb584 | 2255 | fdput(f); |
7699acd1 | 2256 | error_return: |
7699acd1 DL |
2257 | |
2258 | return error; | |
2259 | } | |
2260 | ||
58e41a44 JA |
2261 | /* |
2262 | * The following function implements the controller interface for | |
2263 | * the eventpoll file that enables the insertion/removal/change of | |
2264 | * file descriptors inside the interest set. | |
2265 | */ | |
2266 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, | |
2267 | struct epoll_event __user *, event) | |
2268 | { | |
2269 | struct epoll_event epds; | |
2270 | ||
2271 | if (ep_op_has_event(op) && | |
2272 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
2273 | return -EFAULT; | |
2274 | ||
39220e8d | 2275 | return do_epoll_ctl(epfd, op, fd, &epds, false); |
58e41a44 JA |
2276 | } |
2277 | ||
7699acd1 DL |
2278 | /* |
2279 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2280 | * part of the user space epoll_wait(2). | |
2281 | */ | |
791eb22e | 2282 | static int do_epoll_wait(int epfd, struct epoll_event __user *events, |
7cdf7c20 | 2283 | int maxevents, struct timespec64 *to) |
7699acd1 | 2284 | { |
2903ff01 AV |
2285 | int error; |
2286 | struct fd f; | |
7699acd1 DL |
2287 | struct eventpoll *ep; |
2288 | ||
7699acd1 DL |
2289 | /* The maximum number of event must be greater than zero */ |
2290 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
2291 | return -EINVAL; | |
2292 | ||
2293 | /* Verify that the area passed by the user is writeable */ | |
96d4f267 | 2294 | if (!access_ok(events, maxevents * sizeof(struct epoll_event))) |
2903ff01 | 2295 | return -EFAULT; |
7699acd1 DL |
2296 | |
2297 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
2298 | f = fdget(epfd); |
2299 | if (!f.file) | |
2300 | return -EBADF; | |
7699acd1 DL |
2301 | |
2302 | /* | |
2303 | * We have to check that the file structure underneath the fd | |
2304 | * the user passed to us _is_ an eventpoll file. | |
2305 | */ | |
2306 | error = -EINVAL; | |
2903ff01 | 2307 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
2308 | goto error_fput; |
2309 | ||
2310 | /* | |
2311 | * At this point it is safe to assume that the "private_data" contains | |
2312 | * our own data structure. | |
2313 | */ | |
2903ff01 | 2314 | ep = f.file->private_data; |
7699acd1 DL |
2315 | |
2316 | /* Time to fish for events ... */ | |
7cdf7c20 | 2317 | error = ep_poll(ep, events, maxevents, to); |
7699acd1 DL |
2318 | |
2319 | error_fput: | |
2903ff01 | 2320 | fdput(f); |
7699acd1 DL |
2321 | return error; |
2322 | } | |
2323 | ||
791eb22e DB |
2324 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
2325 | int, maxevents, int, timeout) | |
2326 | { | |
7cdf7c20 WB |
2327 | struct timespec64 to; |
2328 | ||
2329 | return do_epoll_wait(epfd, events, maxevents, | |
2330 | ep_timeout_to_timespec(&to, timeout)); | |
791eb22e DB |
2331 | } |
2332 | ||
7699acd1 DL |
2333 | /* |
2334 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2335 | * part of the user space epoll_pwait(2). | |
2336 | */ | |
58169a52 WB |
2337 | static int do_epoll_pwait(int epfd, struct epoll_event __user *events, |
2338 | int maxevents, struct timespec64 *to, | |
2339 | const sigset_t __user *sigmask, size_t sigsetsize) | |
7699acd1 DL |
2340 | { |
2341 | int error; | |
7699acd1 DL |
2342 | |
2343 | /* | |
2344 | * If the caller wants a certain signal mask to be set during the wait, | |
2345 | * we apply it here. | |
2346 | */ | |
b772434b | 2347 | error = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2348 | if (error) |
2349 | return error; | |
7699acd1 | 2350 | |
58169a52 | 2351 | error = do_epoll_wait(epfd, events, maxevents, to); |
7cdf7c20 | 2352 | |
b772434b | 2353 | restore_saved_sigmask_unless(error == -EINTR); |
7699acd1 DL |
2354 | |
2355 | return error; | |
2356 | } | |
2357 | ||
58169a52 WB |
2358 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
2359 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
2360 | size_t, sigsetsize) | |
35280bd4 | 2361 | { |
7cdf7c20 | 2362 | struct timespec64 to; |
58169a52 WB |
2363 | |
2364 | return do_epoll_pwait(epfd, events, maxevents, | |
2365 | ep_timeout_to_timespec(&to, timeout), | |
2366 | sigmask, sigsetsize); | |
2367 | } | |
2368 | ||
2369 | SYSCALL_DEFINE6(epoll_pwait2, int, epfd, struct epoll_event __user *, events, | |
2370 | int, maxevents, const struct __kernel_timespec __user *, timeout, | |
2371 | const sigset_t __user *, sigmask, size_t, sigsetsize) | |
2372 | { | |
2373 | struct timespec64 ts, *to = NULL; | |
2374 | ||
2375 | if (timeout) { | |
2376 | if (get_timespec64(&ts, timeout)) | |
2377 | return -EFAULT; | |
2378 | to = &ts; | |
2379 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
2380 | return -EINVAL; | |
2381 | } | |
2382 | ||
2383 | return do_epoll_pwait(epfd, events, maxevents, to, | |
2384 | sigmask, sigsetsize); | |
2385 | } | |
2386 | ||
2387 | #ifdef CONFIG_COMPAT | |
2388 | static int do_compat_epoll_pwait(int epfd, struct epoll_event __user *events, | |
2389 | int maxevents, struct timespec64 *timeout, | |
2390 | const compat_sigset_t __user *sigmask, | |
2391 | compat_size_t sigsetsize) | |
2392 | { | |
35280bd4 | 2393 | long err; |
35280bd4 AV |
2394 | |
2395 | /* | |
2396 | * If the caller wants a certain signal mask to be set during the wait, | |
2397 | * we apply it here. | |
2398 | */ | |
b772434b | 2399 | err = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2400 | if (err) |
2401 | return err; | |
35280bd4 | 2402 | |
58169a52 | 2403 | err = do_epoll_wait(epfd, events, maxevents, timeout); |
7cdf7c20 | 2404 | |
b772434b | 2405 | restore_saved_sigmask_unless(err == -EINTR); |
35280bd4 AV |
2406 | |
2407 | return err; | |
2408 | } | |
58169a52 WB |
2409 | |
2410 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, | |
2411 | struct epoll_event __user *, events, | |
2412 | int, maxevents, int, timeout, | |
2413 | const compat_sigset_t __user *, sigmask, | |
2414 | compat_size_t, sigsetsize) | |
2415 | { | |
2416 | struct timespec64 to; | |
2417 | ||
2418 | return do_compat_epoll_pwait(epfd, events, maxevents, | |
2419 | ep_timeout_to_timespec(&to, timeout), | |
2420 | sigmask, sigsetsize); | |
2421 | } | |
2422 | ||
2423 | COMPAT_SYSCALL_DEFINE6(epoll_pwait2, int, epfd, | |
2424 | struct epoll_event __user *, events, | |
2425 | int, maxevents, | |
2426 | const struct __kernel_timespec __user *, timeout, | |
2427 | const compat_sigset_t __user *, sigmask, | |
2428 | compat_size_t, sigsetsize) | |
2429 | { | |
2430 | struct timespec64 ts, *to = NULL; | |
2431 | ||
2432 | if (timeout) { | |
2433 | if (get_timespec64(&ts, timeout)) | |
2434 | return -EFAULT; | |
2435 | to = &ts; | |
2436 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
2437 | return -EINVAL; | |
2438 | } | |
2439 | ||
2440 | return do_compat_epoll_pwait(epfd, events, maxevents, to, | |
2441 | sigmask, sigsetsize); | |
2442 | } | |
2443 | ||
35280bd4 AV |
2444 | #endif |
2445 | ||
1da177e4 LT |
2446 | static int __init eventpoll_init(void) |
2447 | { | |
7ef9964e DL |
2448 | struct sysinfo si; |
2449 | ||
2450 | si_meminfo(&si); | |
9df04e1f DL |
2451 | /* |
2452 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
2453 | */ | |
2454 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 2455 | EP_ITEM_COST; |
52bd19f7 | 2456 | BUG_ON(max_user_watches < 0); |
1da177e4 | 2457 | |
39732ca5 EW |
2458 | /* |
2459 | * We can have many thousands of epitems, so prevent this from | |
2460 | * using an extra cache line on 64-bit (and smaller) CPUs | |
2461 | */ | |
2462 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); | |
2463 | ||
1da177e4 LT |
2464 | /* Allocates slab cache used to allocate "struct epitem" items */ |
2465 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
2ae928a9 | 2466 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 LT |
2467 | |
2468 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
2469 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
2ae928a9 | 2470 | sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); |
a8f5de89 | 2471 | epoll_sysctls_init(); |
1da177e4 | 2472 | |
319c1517 AV |
2473 | ephead_cache = kmem_cache_create("ep_head", |
2474 | sizeof(struct epitems_head), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); | |
2475 | ||
1da177e4 | 2476 | return 0; |
1da177e4 | 2477 | } |
cea69241 | 2478 | fs_initcall(eventpoll_init); |