]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 | 2 | /* |
5071f97e DL |
3 | * fs/eventpoll.c (Efficient event retrieval implementation) |
4 | * Copyright (C) 2001,...,2009 Davide Libenzi | |
1da177e4 | 5 | * |
1da177e4 | 6 | * Davide Libenzi <davidel@xmailserver.org> |
1da177e4 LT |
7 | */ |
8 | ||
1da177e4 LT |
9 | #include <linux/init.h> |
10 | #include <linux/kernel.h> | |
174cd4b1 | 11 | #include <linux/sched/signal.h> |
1da177e4 LT |
12 | #include <linux/fs.h> |
13 | #include <linux/file.h> | |
14 | #include <linux/signal.h> | |
15 | #include <linux/errno.h> | |
16 | #include <linux/mm.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/poll.h> | |
1da177e4 LT |
19 | #include <linux/string.h> |
20 | #include <linux/list.h> | |
21 | #include <linux/hash.h> | |
22 | #include <linux/spinlock.h> | |
23 | #include <linux/syscalls.h> | |
1da177e4 LT |
24 | #include <linux/rbtree.h> |
25 | #include <linux/wait.h> | |
26 | #include <linux/eventpoll.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/bitops.h> | |
144efe3e | 29 | #include <linux/mutex.h> |
da66f7cb | 30 | #include <linux/anon_inodes.h> |
4d7e30d9 | 31 | #include <linux/device.h> |
7c0f6ba6 | 32 | #include <linux/uaccess.h> |
1da177e4 LT |
33 | #include <asm/io.h> |
34 | #include <asm/mman.h> | |
60063497 | 35 | #include <linux/atomic.h> |
138d22b5 CG |
36 | #include <linux/proc_fs.h> |
37 | #include <linux/seq_file.h> | |
35280bd4 | 38 | #include <linux/compat.h> |
ae10b2b4 | 39 | #include <linux/rculist.h> |
bf3b9f63 | 40 | #include <net/busy_poll.h> |
1da177e4 | 41 | |
1da177e4 LT |
42 | /* |
43 | * LOCKING: | |
44 | * There are three level of locking required by epoll : | |
45 | * | |
144efe3e | 46 | * 1) epmutex (mutex) |
c7ea7630 | 47 | * 2) ep->mtx (mutex) |
a218cc49 | 48 | * 3) ep->lock (rwlock) |
1da177e4 LT |
49 | * |
50 | * The acquire order is the one listed above, from 1 to 3. | |
a218cc49 | 51 | * We need a rwlock (ep->lock) because we manipulate objects |
1da177e4 LT |
52 | * from inside the poll callback, that might be triggered from |
53 | * a wake_up() that in turn might be called from IRQ context. | |
54 | * So we can't sleep inside the poll callback and hence we need | |
55 | * a spinlock. During the event transfer loop (from kernel to | |
56 | * user space) we could end up sleeping due a copy_to_user(), so | |
57 | * we need a lock that will allow us to sleep. This lock is a | |
d47de16c DL |
58 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
59 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). | |
60 | * Then we also need a global mutex to serialize eventpoll_release_file() | |
61 | * and ep_free(). | |
62 | * This mutex is acquired by ep_free() during the epoll file | |
1da177e4 LT |
63 | * cleanup path and it is also acquired by eventpoll_release_file() |
64 | * if a file has been pushed inside an epoll set and it is then | |
bf6a41db | 65 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
22bacca4 DL |
66 | * It is also acquired when inserting an epoll fd onto another epoll |
67 | * fd. We do this so that we walk the epoll tree and ensure that this | |
68 | * insertion does not create a cycle of epoll file descriptors, which | |
69 | * could lead to deadlock. We need a global mutex to prevent two | |
70 | * simultaneous inserts (A into B and B into A) from racing and | |
71 | * constructing a cycle without either insert observing that it is | |
72 | * going to. | |
d8805e63 NE |
73 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
74 | * case when one epoll fd is added to another. In this case, we | |
75 | * always acquire the locks in the order of nesting (i.e. after | |
76 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired | |
77 | * before e2->mtx). Since we disallow cycles of epoll file | |
78 | * descriptors, this ensures that the mutexes are well-ordered. In | |
79 | * order to communicate this nesting to lockdep, when walking a tree | |
80 | * of epoll file descriptors, we use the current recursion depth as | |
81 | * the lockdep subkey. | |
d47de16c | 82 | * It is possible to drop the "ep->mtx" and to use the global |
a218cc49 | 83 | * mutex "epmutex" (together with "ep->lock") to have it working, |
d47de16c | 84 | * but having "ep->mtx" will make the interface more scalable. |
144efe3e | 85 | * Events that require holding "epmutex" are very rare, while for |
d47de16c DL |
86 | * normal operations the epoll private "ep->mtx" will guarantee |
87 | * a better scalability. | |
1da177e4 LT |
88 | */ |
89 | ||
1da177e4 | 90 | /* Epoll private bits inside the event mask */ |
df0108c5 | 91 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE) |
1da177e4 | 92 | |
a9a08845 | 93 | #define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT) |
b6a515c8 | 94 | |
a9a08845 | 95 | #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \ |
b6a515c8 JB |
96 | EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE) |
97 | ||
5071f97e DL |
98 | /* Maximum number of nesting allowed inside epoll sets */ |
99 | #define EP_MAX_NESTS 4 | |
1da177e4 | 100 | |
b611967d DL |
101 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
102 | ||
d47de16c DL |
103 | #define EP_UNACTIVE_PTR ((void *) -1L) |
104 | ||
7ef9964e DL |
105 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
106 | ||
1da177e4 LT |
107 | struct epoll_filefd { |
108 | struct file *file; | |
109 | int fd; | |
39732ca5 | 110 | } __packed; |
1da177e4 LT |
111 | |
112 | /* | |
5071f97e DL |
113 | * Structure used to track possible nested calls, for too deep recursions |
114 | * and loop cycles. | |
1da177e4 | 115 | */ |
5071f97e | 116 | struct nested_call_node { |
1da177e4 | 117 | struct list_head llink; |
5071f97e | 118 | void *cookie; |
3fe4a975 | 119 | void *ctx; |
1da177e4 LT |
120 | }; |
121 | ||
122 | /* | |
5071f97e DL |
123 | * This structure is used as collector for nested calls, to check for |
124 | * maximum recursion dept and loop cycles. | |
1da177e4 | 125 | */ |
5071f97e DL |
126 | struct nested_calls { |
127 | struct list_head tasks_call_list; | |
1da177e4 LT |
128 | spinlock_t lock; |
129 | }; | |
130 | ||
d47de16c DL |
131 | /* |
132 | * Each file descriptor added to the eventpoll interface will | |
133 | * have an entry of this type linked to the "rbr" RB tree. | |
39732ca5 EW |
134 | * Avoid increasing the size of this struct, there can be many thousands |
135 | * of these on a server and we do not want this to take another cache line. | |
d47de16c DL |
136 | */ |
137 | struct epitem { | |
ae10b2b4 JB |
138 | union { |
139 | /* RB tree node links this structure to the eventpoll RB tree */ | |
140 | struct rb_node rbn; | |
141 | /* Used to free the struct epitem */ | |
142 | struct rcu_head rcu; | |
143 | }; | |
d47de16c DL |
144 | |
145 | /* List header used to link this structure to the eventpoll ready list */ | |
146 | struct list_head rdllink; | |
147 | ||
c7ea7630 DL |
148 | /* |
149 | * Works together "struct eventpoll"->ovflist in keeping the | |
150 | * single linked chain of items. | |
151 | */ | |
152 | struct epitem *next; | |
153 | ||
d47de16c DL |
154 | /* The file descriptor information this item refers to */ |
155 | struct epoll_filefd ffd; | |
156 | ||
157 | /* Number of active wait queue attached to poll operations */ | |
158 | int nwait; | |
159 | ||
160 | /* List containing poll wait queues */ | |
161 | struct list_head pwqlist; | |
162 | ||
163 | /* The "container" of this item */ | |
164 | struct eventpoll *ep; | |
165 | ||
d47de16c DL |
166 | /* List header used to link this item to the "struct file" items list */ |
167 | struct list_head fllink; | |
168 | ||
4d7e30d9 | 169 | /* wakeup_source used when EPOLLWAKEUP is set */ |
eea1d585 | 170 | struct wakeup_source __rcu *ws; |
4d7e30d9 | 171 | |
c7ea7630 DL |
172 | /* The structure that describe the interested events and the source fd */ |
173 | struct epoll_event event; | |
d47de16c DL |
174 | }; |
175 | ||
1da177e4 LT |
176 | /* |
177 | * This structure is stored inside the "private_data" member of the file | |
bf6a41db | 178 | * structure and represents the main data structure for the eventpoll |
1da177e4 LT |
179 | * interface. |
180 | */ | |
181 | struct eventpoll { | |
1da177e4 | 182 | /* |
d47de16c DL |
183 | * This mutex is used to ensure that files are not removed |
184 | * while epoll is using them. This is held during the event | |
185 | * collection loop, the file cleanup path, the epoll file exit | |
186 | * code and the ctl operations. | |
1da177e4 | 187 | */ |
d47de16c | 188 | struct mutex mtx; |
1da177e4 LT |
189 | |
190 | /* Wait queue used by sys_epoll_wait() */ | |
191 | wait_queue_head_t wq; | |
192 | ||
193 | /* Wait queue used by file->poll() */ | |
194 | wait_queue_head_t poll_wait; | |
195 | ||
196 | /* List of ready file descriptors */ | |
197 | struct list_head rdllist; | |
198 | ||
a218cc49 RP |
199 | /* Lock which protects rdllist and ovflist */ |
200 | rwlock_t lock; | |
201 | ||
67647d0f | 202 | /* RB tree root used to store monitored fd structs */ |
b2ac2ea6 | 203 | struct rb_root_cached rbr; |
d47de16c DL |
204 | |
205 | /* | |
206 | * This is a single linked list that chains all the "struct epitem" that | |
25985edc | 207 | * happened while transferring ready events to userspace w/out |
a218cc49 | 208 | * holding ->lock. |
d47de16c DL |
209 | */ |
210 | struct epitem *ovflist; | |
7ef9964e | 211 | |
4d7e30d9 AH |
212 | /* wakeup_source used when ep_scan_ready_list is running */ |
213 | struct wakeup_source *ws; | |
214 | ||
7ef9964e DL |
215 | /* The user that created the eventpoll descriptor */ |
216 | struct user_struct *user; | |
28d82dc1 JB |
217 | |
218 | struct file *file; | |
219 | ||
220 | /* used to optimize loop detection check */ | |
221 | int visited; | |
222 | struct list_head visited_list_link; | |
bf3b9f63 SS |
223 | |
224 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
225 | /* used to track busy poll napi_id */ | |
226 | unsigned int napi_id; | |
227 | #endif | |
1da177e4 LT |
228 | }; |
229 | ||
230 | /* Wait structure used by the poll hooks */ | |
231 | struct eppoll_entry { | |
232 | /* List header used to link this structure to the "struct epitem" */ | |
233 | struct list_head llink; | |
234 | ||
235 | /* The "base" pointer is set to the container "struct epitem" */ | |
4f0989db | 236 | struct epitem *base; |
1da177e4 LT |
237 | |
238 | /* | |
239 | * Wait queue item that will be linked to the target file wait | |
240 | * queue head. | |
241 | */ | |
ac6424b9 | 242 | wait_queue_entry_t wait; |
1da177e4 LT |
243 | |
244 | /* The wait queue head that linked the "wait" wait queue item */ | |
245 | wait_queue_head_t *whead; | |
246 | }; | |
247 | ||
1da177e4 LT |
248 | /* Wrapper struct used by poll queueing */ |
249 | struct ep_pqueue { | |
250 | poll_table pt; | |
251 | struct epitem *epi; | |
252 | }; | |
253 | ||
5071f97e DL |
254 | /* Used by the ep_send_events() function as callback private data */ |
255 | struct ep_send_events_data { | |
256 | int maxevents; | |
257 | struct epoll_event __user *events; | |
d7ebbe46 | 258 | int res; |
5071f97e DL |
259 | }; |
260 | ||
7ef9964e DL |
261 | /* |
262 | * Configuration options available inside /proc/sys/fs/epoll/ | |
263 | */ | |
7ef9964e | 264 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 265 | static long max_user_watches __read_mostly; |
7ef9964e | 266 | |
1da177e4 | 267 | /* |
d47de16c | 268 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 269 | */ |
7ef9964e | 270 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 271 | |
22bacca4 DL |
272 | /* Used to check for epoll file descriptor inclusion loops */ |
273 | static struct nested_calls poll_loop_ncalls; | |
274 | ||
1da177e4 | 275 | /* Slab cache used to allocate "struct epitem" */ |
e18b890b | 276 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
277 | |
278 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 279 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 280 | |
28d82dc1 JB |
281 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
282 | static LIST_HEAD(visited_list); | |
283 | ||
284 | /* | |
285 | * List of files with newly added links, where we may need to limit the number | |
286 | * of emanating paths. Protected by the epmutex. | |
287 | */ | |
288 | static LIST_HEAD(tfile_check_list); | |
289 | ||
7ef9964e DL |
290 | #ifdef CONFIG_SYSCTL |
291 | ||
292 | #include <linux/sysctl.h> | |
293 | ||
eec4844f | 294 | static long long_zero; |
52bd19f7 | 295 | static long long_max = LONG_MAX; |
7ef9964e | 296 | |
1f7e0616 | 297 | struct ctl_table epoll_table[] = { |
7ef9964e DL |
298 | { |
299 | .procname = "max_user_watches", | |
300 | .data = &max_user_watches, | |
52bd19f7 | 301 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 302 | .mode = 0644, |
52bd19f7 | 303 | .proc_handler = proc_doulongvec_minmax, |
eec4844f | 304 | .extra1 = &long_zero, |
52bd19f7 | 305 | .extra2 = &long_max, |
7ef9964e | 306 | }, |
ab09203e | 307 | { } |
7ef9964e DL |
308 | }; |
309 | #endif /* CONFIG_SYSCTL */ | |
310 | ||
28d82dc1 JB |
311 | static const struct file_operations eventpoll_fops; |
312 | ||
313 | static inline int is_file_epoll(struct file *f) | |
314 | { | |
315 | return f->f_op == &eventpoll_fops; | |
316 | } | |
b030a4dd | 317 | |
67647d0f | 318 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
319 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
320 | struct file *file, int fd) | |
321 | { | |
322 | ffd->file = file; | |
323 | ffd->fd = fd; | |
324 | } | |
325 | ||
67647d0f | 326 | /* Compare RB tree keys */ |
b030a4dd PE |
327 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
328 | struct epoll_filefd *p2) | |
329 | { | |
330 | return (p1->file > p2->file ? +1: | |
331 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
332 | } | |
333 | ||
b030a4dd | 334 | /* Tells us if the item is currently linked */ |
992991c0 | 335 | static inline int ep_is_linked(struct epitem *epi) |
b030a4dd | 336 | { |
992991c0 | 337 | return !list_empty(&epi->rdllink); |
b030a4dd PE |
338 | } |
339 | ||
ac6424b9 | 340 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
971316f0 ON |
341 | { |
342 | return container_of(p, struct eppoll_entry, wait); | |
343 | } | |
344 | ||
b030a4dd | 345 | /* Get the "struct epitem" from a wait queue pointer */ |
ac6424b9 | 346 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
b030a4dd PE |
347 | { |
348 | return container_of(p, struct eppoll_entry, wait)->base; | |
349 | } | |
350 | ||
351 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 352 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
353 | { |
354 | return container_of(p, struct ep_pqueue, pt)->epi; | |
355 | } | |
356 | ||
357 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 358 | static inline int ep_op_has_event(int op) |
b030a4dd | 359 | { |
a80a6b85 | 360 | return op != EPOLL_CTL_DEL; |
b030a4dd PE |
361 | } |
362 | ||
1da177e4 | 363 | /* Initialize the poll safe wake up structure */ |
5071f97e | 364 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 365 | { |
5071f97e DL |
366 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
367 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
368 | } |
369 | ||
3fb0e584 DL |
370 | /** |
371 | * ep_events_available - Checks if ready events might be available. | |
372 | * | |
373 | * @ep: Pointer to the eventpoll context. | |
374 | * | |
375 | * Returns: Returns a value different than zero if ready events are available, | |
376 | * or zero otherwise. | |
377 | */ | |
378 | static inline int ep_events_available(struct eventpoll *ep) | |
379 | { | |
c5a282e9 DB |
380 | return !list_empty_careful(&ep->rdllist) || |
381 | READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR; | |
3fb0e584 DL |
382 | } |
383 | ||
bf3b9f63 SS |
384 | #ifdef CONFIG_NET_RX_BUSY_POLL |
385 | static bool ep_busy_loop_end(void *p, unsigned long start_time) | |
386 | { | |
387 | struct eventpoll *ep = p; | |
388 | ||
389 | return ep_events_available(ep) || busy_loop_timeout(start_time); | |
390 | } | |
bf3b9f63 SS |
391 | |
392 | /* | |
393 | * Busy poll if globally on and supporting sockets found && no events, | |
394 | * busy loop will return if need_resched or ep_events_available. | |
395 | * | |
396 | * we must do our busy polling with irqs enabled | |
397 | */ | |
398 | static void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
399 | { | |
bf3b9f63 SS |
400 | unsigned int napi_id = READ_ONCE(ep->napi_id); |
401 | ||
402 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) | |
403 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep); | |
bf3b9f63 SS |
404 | } |
405 | ||
406 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
407 | { | |
bf3b9f63 SS |
408 | if (ep->napi_id) |
409 | ep->napi_id = 0; | |
bf3b9f63 SS |
410 | } |
411 | ||
412 | /* | |
413 | * Set epoll busy poll NAPI ID from sk. | |
414 | */ | |
415 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
416 | { | |
bf3b9f63 SS |
417 | struct eventpoll *ep; |
418 | unsigned int napi_id; | |
419 | struct socket *sock; | |
420 | struct sock *sk; | |
421 | int err; | |
422 | ||
423 | if (!net_busy_loop_on()) | |
424 | return; | |
425 | ||
426 | sock = sock_from_file(epi->ffd.file, &err); | |
427 | if (!sock) | |
428 | return; | |
429 | ||
430 | sk = sock->sk; | |
431 | if (!sk) | |
432 | return; | |
433 | ||
434 | napi_id = READ_ONCE(sk->sk_napi_id); | |
435 | ep = epi->ep; | |
436 | ||
437 | /* Non-NAPI IDs can be rejected | |
438 | * or | |
439 | * Nothing to do if we already have this ID | |
440 | */ | |
441 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) | |
442 | return; | |
443 | ||
444 | /* record NAPI ID for use in next busy poll */ | |
445 | ep->napi_id = napi_id; | |
bf3b9f63 SS |
446 | } |
447 | ||
514056d5 DB |
448 | #else |
449 | ||
450 | static inline void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
451 | { | |
452 | } | |
453 | ||
454 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
455 | { | |
456 | } | |
457 | ||
458 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
459 | { | |
460 | } | |
461 | ||
462 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
463 | ||
5071f97e DL |
464 | /** |
465 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
466 | * that the recursion limit is not exceeded, and that | |
467 | * the same nested call (by the meaning of same cookie) is | |
468 | * no re-entered. | |
469 | * | |
470 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
5071f97e DL |
471 | * @nproc: Nested call core function pointer. |
472 | * @priv: Opaque data to be passed to the @nproc callback. | |
473 | * @cookie: Cookie to be used to identify this nested call. | |
3fe4a975 | 474 | * @ctx: This instance context. |
5071f97e DL |
475 | * |
476 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
477 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 478 | */ |
74bdc129 | 479 | static int ep_call_nested(struct nested_calls *ncalls, |
5071f97e | 480 | int (*nproc)(void *, void *, int), void *priv, |
3fe4a975 | 481 | void *cookie, void *ctx) |
1da177e4 | 482 | { |
5071f97e | 483 | int error, call_nests = 0; |
1da177e4 | 484 | unsigned long flags; |
5071f97e DL |
485 | struct list_head *lsthead = &ncalls->tasks_call_list; |
486 | struct nested_call_node *tncur; | |
487 | struct nested_call_node tnode; | |
1da177e4 | 488 | |
5071f97e | 489 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 490 | |
5071f97e DL |
491 | /* |
492 | * Try to see if the current task is already inside this wakeup call. | |
493 | * We use a list here, since the population inside this set is always | |
494 | * very much limited. | |
495 | */ | |
b70c3940 | 496 | list_for_each_entry(tncur, lsthead, llink) { |
3fe4a975 | 497 | if (tncur->ctx == ctx && |
74bdc129 | 498 | (tncur->cookie == cookie || ++call_nests > EP_MAX_NESTS)) { |
1da177e4 LT |
499 | /* |
500 | * Ops ... loop detected or maximum nest level reached. | |
501 | * We abort this wake by breaking the cycle itself. | |
502 | */ | |
abff55ce TB |
503 | error = -1; |
504 | goto out_unlock; | |
1da177e4 LT |
505 | } |
506 | } | |
507 | ||
5071f97e | 508 | /* Add the current task and cookie to the list */ |
3fe4a975 | 509 | tnode.ctx = ctx; |
5071f97e | 510 | tnode.cookie = cookie; |
1da177e4 LT |
511 | list_add(&tnode.llink, lsthead); |
512 | ||
5071f97e | 513 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 514 | |
5071f97e DL |
515 | /* Call the nested function */ |
516 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
517 | |
518 | /* Remove the current task from the list */ | |
5071f97e | 519 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 520 | list_del(&tnode.llink); |
3fe4a975 | 521 | out_unlock: |
5071f97e DL |
522 | spin_unlock_irqrestore(&ncalls->lock, flags); |
523 | ||
524 | return error; | |
525 | } | |
526 | ||
02edc6fc SR |
527 | /* |
528 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
529 | * the use of wait queues used by epoll is done in a very controlled | |
530 | * manner. Wake ups can nest inside each other, but are never done | |
531 | * with the same locking. For example: | |
532 | * | |
533 | * dfd = socket(...); | |
534 | * efd1 = epoll_create(); | |
535 | * efd2 = epoll_create(); | |
536 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
537 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
538 | * | |
539 | * When a packet arrives to the device underneath "dfd", the net code will | |
540 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
541 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
542 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
543 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
544 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
545 | * that ends up in another wake_up(), after having checked about the | |
546 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
547 | * avoid stack blasting. | |
548 | * | |
549 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
550 | * this special case of epoll. | |
551 | */ | |
2dfa4eea | 552 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
57a173bd | 553 | |
f6520c52 | 554 | static DEFINE_PER_CPU(int, wakeup_nest); |
5071f97e | 555 | |
5071f97e DL |
556 | static void ep_poll_safewake(wait_queue_head_t *wq) |
557 | { | |
f6520c52 JB |
558 | unsigned long flags; |
559 | int subclass; | |
3fe4a975 | 560 | |
f6520c52 JB |
561 | local_irq_save(flags); |
562 | preempt_disable(); | |
563 | subclass = __this_cpu_read(wakeup_nest); | |
564 | spin_lock_nested(&wq->lock, subclass + 1); | |
565 | __this_cpu_inc(wakeup_nest); | |
566 | wake_up_locked_poll(wq, POLLIN); | |
567 | __this_cpu_dec(wakeup_nest); | |
568 | spin_unlock(&wq->lock); | |
569 | local_irq_restore(flags); | |
570 | preempt_enable(); | |
1da177e4 LT |
571 | } |
572 | ||
57a173bd JB |
573 | #else |
574 | ||
575 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
576 | { | |
a9a08845 | 577 | wake_up_poll(wq, EPOLLIN); |
57a173bd JB |
578 | } |
579 | ||
580 | #endif | |
581 | ||
971316f0 ON |
582 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
583 | { | |
584 | wait_queue_head_t *whead; | |
585 | ||
586 | rcu_read_lock(); | |
138e4ad6 ON |
587 | /* |
588 | * If it is cleared by POLLFREE, it should be rcu-safe. | |
589 | * If we read NULL we need a barrier paired with | |
590 | * smp_store_release() in ep_poll_callback(), otherwise | |
591 | * we rely on whead->lock. | |
592 | */ | |
593 | whead = smp_load_acquire(&pwq->whead); | |
971316f0 ON |
594 | if (whead) |
595 | remove_wait_queue(whead, &pwq->wait); | |
596 | rcu_read_unlock(); | |
597 | } | |
598 | ||
1da177e4 | 599 | /* |
d1bc90dd TB |
600 | * This function unregisters poll callbacks from the associated file |
601 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
602 | * ep_free). | |
1da177e4 | 603 | */ |
7699acd1 | 604 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 605 | { |
7699acd1 DL |
606 | struct list_head *lsthead = &epi->pwqlist; |
607 | struct eppoll_entry *pwq; | |
1da177e4 | 608 | |
d1bc90dd TB |
609 | while (!list_empty(lsthead)) { |
610 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 611 | |
d1bc90dd | 612 | list_del(&pwq->llink); |
971316f0 | 613 | ep_remove_wait_queue(pwq); |
d1bc90dd | 614 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 615 | } |
1da177e4 LT |
616 | } |
617 | ||
eea1d585 EW |
618 | /* call only when ep->mtx is held */ |
619 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) | |
620 | { | |
621 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); | |
622 | } | |
623 | ||
624 | /* call only when ep->mtx is held */ | |
625 | static inline void ep_pm_stay_awake(struct epitem *epi) | |
626 | { | |
627 | struct wakeup_source *ws = ep_wakeup_source(epi); | |
628 | ||
629 | if (ws) | |
630 | __pm_stay_awake(ws); | |
631 | } | |
632 | ||
633 | static inline bool ep_has_wakeup_source(struct epitem *epi) | |
634 | { | |
635 | return rcu_access_pointer(epi->ws) ? true : false; | |
636 | } | |
637 | ||
638 | /* call when ep->mtx cannot be held (ep_poll_callback) */ | |
639 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) | |
640 | { | |
641 | struct wakeup_source *ws; | |
642 | ||
643 | rcu_read_lock(); | |
644 | ws = rcu_dereference(epi->ws); | |
645 | if (ws) | |
646 | __pm_stay_awake(ws); | |
647 | rcu_read_unlock(); | |
648 | } | |
649 | ||
5071f97e DL |
650 | /** |
651 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
652 | * the scan code, to call f_op->poll(). Also allows for | |
653 | * O(NumReady) performance. | |
654 | * | |
655 | * @ep: Pointer to the epoll private data structure. | |
656 | * @sproc: Pointer to the scan callback. | |
657 | * @priv: Private opaque data passed to the @sproc callback. | |
d8805e63 | 658 | * @depth: The current depth of recursive f_op->poll calls. |
67347fe4 | 659 | * @ep_locked: caller already holds ep->mtx |
5071f97e DL |
660 | * |
661 | * Returns: The same integer error code returned by the @sproc callback. | |
662 | */ | |
d85e2aa2 AV |
663 | static __poll_t ep_scan_ready_list(struct eventpoll *ep, |
664 | __poll_t (*sproc)(struct eventpoll *, | |
5071f97e | 665 | struct list_head *, void *), |
67347fe4 | 666 | void *priv, int depth, bool ep_locked) |
5071f97e | 667 | { |
d85e2aa2 | 668 | __poll_t res; |
5071f97e | 669 | struct epitem *epi, *nepi; |
296e236e | 670 | LIST_HEAD(txlist); |
5071f97e | 671 | |
92e64178 DB |
672 | lockdep_assert_irqs_enabled(); |
673 | ||
5071f97e DL |
674 | /* |
675 | * We need to lock this because we could be hit by | |
e057e15f | 676 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e | 677 | */ |
67347fe4 JB |
678 | |
679 | if (!ep_locked) | |
680 | mutex_lock_nested(&ep->mtx, depth); | |
5071f97e DL |
681 | |
682 | /* | |
683 | * Steal the ready list, and re-init the original one to the | |
684 | * empty list. Also, set ep->ovflist to NULL so that events | |
685 | * happening while looping w/out locks, are not lost. We cannot | |
686 | * have the poll callback to queue directly on ep->rdllist, | |
687 | * because we want the "sproc" callback to be able to do it | |
688 | * in a lockless way. | |
689 | */ | |
a218cc49 | 690 | write_lock_irq(&ep->lock); |
296e236e | 691 | list_splice_init(&ep->rdllist, &txlist); |
c5a282e9 | 692 | WRITE_ONCE(ep->ovflist, NULL); |
a218cc49 | 693 | write_unlock_irq(&ep->lock); |
5071f97e DL |
694 | |
695 | /* | |
696 | * Now call the callback function. | |
697 | */ | |
d85e2aa2 | 698 | res = (*sproc)(ep, &txlist, priv); |
5071f97e | 699 | |
a218cc49 | 700 | write_lock_irq(&ep->lock); |
5071f97e DL |
701 | /* |
702 | * During the time we spent inside the "sproc" callback, some | |
703 | * other events might have been queued by the poll callback. | |
704 | * We re-insert them inside the main ready-list here. | |
705 | */ | |
c5a282e9 | 706 | for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL; |
5071f97e DL |
707 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
708 | /* | |
709 | * We need to check if the item is already in the list. | |
710 | * During the "sproc" callback execution time, items are | |
711 | * queued into ->ovflist but the "txlist" might already | |
712 | * contain them, and the list_splice() below takes care of them. | |
713 | */ | |
992991c0 | 714 | if (!ep_is_linked(epi)) { |
c141175d RP |
715 | /* |
716 | * ->ovflist is LIFO, so we have to reverse it in order | |
717 | * to keep in FIFO. | |
718 | */ | |
719 | list_add(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 720 | ep_pm_stay_awake(epi); |
4d7e30d9 | 721 | } |
5071f97e DL |
722 | } |
723 | /* | |
724 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
725 | * releasing the lock, events will be queued in the normal way inside | |
726 | * ep->rdllist. | |
727 | */ | |
c5a282e9 | 728 | WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR); |
5071f97e DL |
729 | |
730 | /* | |
731 | * Quickly re-inject items left on "txlist". | |
732 | */ | |
733 | list_splice(&txlist, &ep->rdllist); | |
4d7e30d9 | 734 | __pm_relax(ep->ws); |
a218cc49 | 735 | write_unlock_irq(&ep->lock); |
5071f97e | 736 | |
67347fe4 JB |
737 | if (!ep_locked) |
738 | mutex_unlock(&ep->mtx); | |
5071f97e | 739 | |
d85e2aa2 | 740 | return res; |
5071f97e DL |
741 | } |
742 | ||
ae10b2b4 JB |
743 | static void epi_rcu_free(struct rcu_head *head) |
744 | { | |
745 | struct epitem *epi = container_of(head, struct epitem, rcu); | |
746 | kmem_cache_free(epi_cache, epi); | |
747 | } | |
748 | ||
7699acd1 DL |
749 | /* |
750 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 751 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
752 | */ |
753 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
754 | { | |
7699acd1 | 755 | struct file *file = epi->ffd.file; |
1da177e4 | 756 | |
92e64178 DB |
757 | lockdep_assert_irqs_enabled(); |
758 | ||
1da177e4 | 759 | /* |
ee8ef0a4 | 760 | * Removes poll wait queue hooks. |
1da177e4 | 761 | */ |
7699acd1 | 762 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 763 | |
7699acd1 | 764 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 765 | spin_lock(&file->f_lock); |
ae10b2b4 | 766 | list_del_rcu(&epi->fllink); |
68499914 | 767 | spin_unlock(&file->f_lock); |
1da177e4 | 768 | |
b2ac2ea6 | 769 | rb_erase_cached(&epi->rbn, &ep->rbr); |
1da177e4 | 770 | |
a218cc49 | 771 | write_lock_irq(&ep->lock); |
992991c0 | 772 | if (ep_is_linked(epi)) |
c7ea7630 | 773 | list_del_init(&epi->rdllink); |
a218cc49 | 774 | write_unlock_irq(&ep->lock); |
1da177e4 | 775 | |
eea1d585 | 776 | wakeup_source_unregister(ep_wakeup_source(epi)); |
ae10b2b4 JB |
777 | /* |
778 | * At this point it is safe to free the eventpoll item. Use the union | |
779 | * field epi->rcu, since we are trying to minimize the size of | |
780 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by | |
781 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make | |
782 | * use of the rbn field. | |
783 | */ | |
784 | call_rcu(&epi->rcu, epi_rcu_free); | |
1da177e4 | 785 | |
52bd19f7 | 786 | atomic_long_dec(&ep->user->epoll_watches); |
7ef9964e | 787 | |
c7ea7630 | 788 | return 0; |
1da177e4 LT |
789 | } |
790 | ||
7699acd1 | 791 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 792 | { |
7699acd1 DL |
793 | struct rb_node *rbp; |
794 | struct epitem *epi; | |
1da177e4 | 795 | |
7699acd1 DL |
796 | /* We need to release all tasks waiting for these file */ |
797 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 798 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 799 | |
7699acd1 DL |
800 | /* |
801 | * We need to lock this because we could be hit by | |
802 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 803 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
804 | * is on the way to be removed and no one has references to it |
805 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 806 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
807 | */ |
808 | mutex_lock(&epmutex); | |
1da177e4 LT |
809 | |
810 | /* | |
7699acd1 | 811 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 812 | */ |
b2ac2ea6 | 813 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
7699acd1 DL |
814 | epi = rb_entry(rbp, struct epitem, rbn); |
815 | ||
816 | ep_unregister_pollwait(ep, epi); | |
91cf5ab6 | 817 | cond_resched(); |
7699acd1 | 818 | } |
1da177e4 LT |
819 | |
820 | /* | |
7699acd1 DL |
821 | * Walks through the whole tree by freeing each "struct epitem". At this |
822 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 823 | * holding "epmutex" we can be sure that no file cleanup code will hit |
a218cc49 | 824 | * us during this operation. So we can avoid the lock on "ep->lock". |
ddf676c3 EW |
825 | * We do not need to lock ep->mtx, either, we only do it to prevent |
826 | * a lockdep warning. | |
1da177e4 | 827 | */ |
ddf676c3 | 828 | mutex_lock(&ep->mtx); |
b2ac2ea6 | 829 | while ((rbp = rb_first_cached(&ep->rbr)) != NULL) { |
7699acd1 DL |
830 | epi = rb_entry(rbp, struct epitem, rbn); |
831 | ep_remove(ep, epi); | |
91cf5ab6 | 832 | cond_resched(); |
7699acd1 | 833 | } |
ddf676c3 | 834 | mutex_unlock(&ep->mtx); |
1da177e4 | 835 | |
7699acd1 | 836 | mutex_unlock(&epmutex); |
d47de16c | 837 | mutex_destroy(&ep->mtx); |
7ef9964e | 838 | free_uid(ep->user); |
4d7e30d9 | 839 | wakeup_source_unregister(ep->ws); |
f0ee9aab | 840 | kfree(ep); |
7699acd1 | 841 | } |
1da177e4 | 842 | |
7699acd1 DL |
843 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
844 | { | |
845 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 846 | |
f0ee9aab | 847 | if (ep) |
7699acd1 | 848 | ep_free(ep); |
7699acd1 | 849 | |
7699acd1 | 850 | return 0; |
1da177e4 LT |
851 | } |
852 | ||
d85e2aa2 | 853 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
37b5e521 JB |
854 | void *priv); |
855 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
856 | poll_table *pt); | |
857 | ||
858 | /* | |
859 | * Differs from ep_eventpoll_poll() in that internal callers already have | |
860 | * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested() | |
861 | * is correctly annotated. | |
862 | */ | |
d85e2aa2 | 863 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, |
bec1a502 | 864 | int depth) |
450d89ec | 865 | { |
37b5e521 JB |
866 | struct eventpoll *ep; |
867 | bool locked; | |
868 | ||
450d89ec | 869 | pt->_key = epi->event.events; |
37b5e521 | 870 | if (!is_file_epoll(epi->ffd.file)) |
9965ed17 | 871 | return vfs_poll(epi->ffd.file, pt) & epi->event.events; |
450d89ec | 872 | |
37b5e521 JB |
873 | ep = epi->ffd.file->private_data; |
874 | poll_wait(epi->ffd.file, &ep->poll_wait, pt); | |
875 | locked = pt && (pt->_qproc == ep_ptable_queue_proc); | |
450d89ec | 876 | |
37b5e521 JB |
877 | return ep_scan_ready_list(epi->ffd.file->private_data, |
878 | ep_read_events_proc, &depth, depth, | |
879 | locked) & epi->event.events; | |
450d89ec EW |
880 | } |
881 | ||
d85e2aa2 | 882 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
296e236e | 883 | void *priv) |
5071f97e DL |
884 | { |
885 | struct epitem *epi, *tmp; | |
626cf236 | 886 | poll_table pt; |
37b5e521 | 887 | int depth = *(int *)priv; |
5071f97e | 888 | |
626cf236 | 889 | init_poll_funcptr(&pt, NULL); |
37b5e521 | 890 | depth++; |
450d89ec | 891 | |
5071f97e | 892 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
37b5e521 | 893 | if (ep_item_poll(epi, &pt, depth)) { |
a9a08845 | 894 | return EPOLLIN | EPOLLRDNORM; |
37b5e521 | 895 | } else { |
5071f97e DL |
896 | /* |
897 | * Item has been dropped into the ready list by the poll | |
898 | * callback, but it's not actually ready, as far as | |
899 | * caller requested events goes. We can remove it here. | |
900 | */ | |
eea1d585 | 901 | __pm_relax(ep_wakeup_source(epi)); |
5071f97e | 902 | list_del_init(&epi->rdllink); |
296e236e | 903 | } |
5071f97e DL |
904 | } |
905 | ||
906 | return 0; | |
907 | } | |
908 | ||
a11e1d43 | 909 | static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait) |
11c5ad0e BN |
910 | { |
911 | struct eventpoll *ep = file->private_data; | |
912 | int depth = 0; | |
7699acd1 | 913 | |
a11e1d43 LT |
914 | /* Insert inside our poll wait queue */ |
915 | poll_wait(file, &ep->poll_wait, wait); | |
916 | ||
5071f97e DL |
917 | /* |
918 | * Proceed to find out if wanted events are really available inside | |
37b5e521 | 919 | * the ready list. |
5071f97e | 920 | */ |
37b5e521 JB |
921 | return ep_scan_ready_list(ep, ep_read_events_proc, |
922 | &depth, depth, false); | |
7699acd1 DL |
923 | } |
924 | ||
138d22b5 | 925 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 926 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
138d22b5 CG |
927 | { |
928 | struct eventpoll *ep = f->private_data; | |
929 | struct rb_node *rbp; | |
138d22b5 CG |
930 | |
931 | mutex_lock(&ep->mtx); | |
b2ac2ea6 | 932 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
138d22b5 | 933 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
77493f04 | 934 | struct inode *inode = file_inode(epi->ffd.file); |
138d22b5 | 935 | |
77493f04 CG |
936 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
937 | " pos:%lli ino:%lx sdev:%x\n", | |
a3816ab0 | 938 | epi->ffd.fd, epi->event.events, |
77493f04 CG |
939 | (long long)epi->event.data, |
940 | (long long)epi->ffd.file->f_pos, | |
941 | inode->i_ino, inode->i_sb->s_dev); | |
a3816ab0 | 942 | if (seq_has_overflowed(m)) |
138d22b5 CG |
943 | break; |
944 | } | |
945 | mutex_unlock(&ep->mtx); | |
138d22b5 CG |
946 | } |
947 | #endif | |
948 | ||
7699acd1 DL |
949 | /* File callbacks that implement the eventpoll file behaviour */ |
950 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
951 | #ifdef CONFIG_PROC_FS |
952 | .show_fdinfo = ep_show_fdinfo, | |
953 | #endif | |
7699acd1 | 954 | .release = ep_eventpoll_release, |
a11e1d43 | 955 | .poll = ep_eventpoll_poll, |
6038f373 | 956 | .llseek = noop_llseek, |
7699acd1 DL |
957 | }; |
958 | ||
b611967d | 959 | /* |
7699acd1 DL |
960 | * This is called from eventpoll_release() to unlink files from the eventpoll |
961 | * interface. We need to have this facility to cleanup correctly files that are | |
962 | * closed without being removed from the eventpoll interface. | |
b611967d | 963 | */ |
7699acd1 | 964 | void eventpoll_release_file(struct file *file) |
b611967d | 965 | { |
7699acd1 | 966 | struct eventpoll *ep; |
ebe06187 | 967 | struct epitem *epi, *next; |
b611967d DL |
968 | |
969 | /* | |
68499914 | 970 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 971 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 972 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 973 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 974 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 975 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 976 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 977 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 978 | * from anywhere but ep_free(). |
68499914 JC |
979 | * |
980 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 981 | */ |
7699acd1 | 982 | mutex_lock(&epmutex); |
ebe06187 | 983 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
7699acd1 | 984 | ep = epi->ep; |
d8805e63 | 985 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 986 | ep_remove(ep, epi); |
d47de16c | 987 | mutex_unlock(&ep->mtx); |
b611967d | 988 | } |
7699acd1 | 989 | mutex_unlock(&epmutex); |
b611967d DL |
990 | } |
991 | ||
53d2be79 | 992 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 993 | { |
7ef9964e DL |
994 | int error; |
995 | struct user_struct *user; | |
996 | struct eventpoll *ep; | |
1da177e4 | 997 | |
7ef9964e | 998 | user = get_current_user(); |
7ef9964e DL |
999 | error = -ENOMEM; |
1000 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
1001 | if (unlikely(!ep)) | |
1002 | goto free_uid; | |
1da177e4 | 1003 | |
d47de16c | 1004 | mutex_init(&ep->mtx); |
a218cc49 | 1005 | rwlock_init(&ep->lock); |
1da177e4 LT |
1006 | init_waitqueue_head(&ep->wq); |
1007 | init_waitqueue_head(&ep->poll_wait); | |
1008 | INIT_LIST_HEAD(&ep->rdllist); | |
b2ac2ea6 | 1009 | ep->rbr = RB_ROOT_CACHED; |
d47de16c | 1010 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 1011 | ep->user = user; |
1da177e4 | 1012 | |
53d2be79 | 1013 | *pep = ep; |
1da177e4 | 1014 | |
1da177e4 | 1015 | return 0; |
7ef9964e DL |
1016 | |
1017 | free_uid: | |
1018 | free_uid(user); | |
1019 | return error; | |
1da177e4 LT |
1020 | } |
1021 | ||
1da177e4 | 1022 | /* |
c7ea7630 DL |
1023 | * Search the file inside the eventpoll tree. The RB tree operations |
1024 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
1025 | * "mtx" held. | |
1da177e4 LT |
1026 | */ |
1027 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
1028 | { | |
1029 | int kcmp; | |
1da177e4 LT |
1030 | struct rb_node *rbp; |
1031 | struct epitem *epi, *epir = NULL; | |
1032 | struct epoll_filefd ffd; | |
1033 | ||
b030a4dd | 1034 | ep_set_ffd(&ffd, file, fd); |
b2ac2ea6 | 1035 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
1da177e4 | 1036 | epi = rb_entry(rbp, struct epitem, rbn); |
b030a4dd | 1037 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
1038 | if (kcmp > 0) |
1039 | rbp = rbp->rb_right; | |
1040 | else if (kcmp < 0) | |
1041 | rbp = rbp->rb_left; | |
1042 | else { | |
1da177e4 LT |
1043 | epir = epi; |
1044 | break; | |
1045 | } | |
1046 | } | |
1da177e4 | 1047 | |
1da177e4 LT |
1048 | return epir; |
1049 | } | |
1050 | ||
92ef6da3 | 1051 | #ifdef CONFIG_CHECKPOINT_RESTORE |
0791e364 CG |
1052 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
1053 | { | |
1054 | struct rb_node *rbp; | |
1055 | struct epitem *epi; | |
1056 | ||
b2ac2ea6 | 1057 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
0791e364 CG |
1058 | epi = rb_entry(rbp, struct epitem, rbn); |
1059 | if (epi->ffd.fd == tfd) { | |
1060 | if (toff == 0) | |
1061 | return epi; | |
1062 | else | |
1063 | toff--; | |
1064 | } | |
1065 | cond_resched(); | |
1066 | } | |
1067 | ||
1068 | return NULL; | |
1069 | } | |
1070 | ||
1071 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, | |
1072 | unsigned long toff) | |
1073 | { | |
1074 | struct file *file_raw; | |
1075 | struct eventpoll *ep; | |
1076 | struct epitem *epi; | |
1077 | ||
1078 | if (!is_file_epoll(file)) | |
1079 | return ERR_PTR(-EINVAL); | |
1080 | ||
1081 | ep = file->private_data; | |
1082 | ||
1083 | mutex_lock(&ep->mtx); | |
1084 | epi = ep_find_tfd(ep, tfd, toff); | |
1085 | if (epi) | |
1086 | file_raw = epi->ffd.file; | |
1087 | else | |
1088 | file_raw = ERR_PTR(-ENOENT); | |
1089 | mutex_unlock(&ep->mtx); | |
1090 | ||
1091 | return file_raw; | |
1092 | } | |
92ef6da3 | 1093 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
0791e364 | 1094 | |
a218cc49 RP |
1095 | /** |
1096 | * Adds a new entry to the tail of the list in a lockless way, i.e. | |
1097 | * multiple CPUs are allowed to call this function concurrently. | |
1098 | * | |
1099 | * Beware: it is necessary to prevent any other modifications of the | |
1100 | * existing list until all changes are completed, in other words | |
1101 | * concurrent list_add_tail_lockless() calls should be protected | |
1102 | * with a read lock, where write lock acts as a barrier which | |
1103 | * makes sure all list_add_tail_lockless() calls are fully | |
1104 | * completed. | |
1105 | * | |
1106 | * Also an element can be locklessly added to the list only in one | |
1107 | * direction i.e. either to the tail either to the head, otherwise | |
1108 | * concurrent access will corrupt the list. | |
1109 | * | |
1110 | * Returns %false if element has been already added to the list, %true | |
1111 | * otherwise. | |
1112 | */ | |
1113 | static inline bool list_add_tail_lockless(struct list_head *new, | |
1114 | struct list_head *head) | |
1115 | { | |
1116 | struct list_head *prev; | |
1117 | ||
1118 | /* | |
1119 | * This is simple 'new->next = head' operation, but cmpxchg() | |
1120 | * is used in order to detect that same element has been just | |
1121 | * added to the list from another CPU: the winner observes | |
1122 | * new->next == new. | |
1123 | */ | |
1124 | if (cmpxchg(&new->next, new, head) != new) | |
1125 | return false; | |
1126 | ||
1127 | /* | |
1128 | * Initially ->next of a new element must be updated with the head | |
1129 | * (we are inserting to the tail) and only then pointers are atomically | |
1130 | * exchanged. XCHG guarantees memory ordering, thus ->next should be | |
1131 | * updated before pointers are actually swapped and pointers are | |
1132 | * swapped before prev->next is updated. | |
1133 | */ | |
1134 | ||
1135 | prev = xchg(&head->prev, new); | |
1136 | ||
1137 | /* | |
1138 | * It is safe to modify prev->next and new->prev, because a new element | |
1139 | * is added only to the tail and new->next is updated before XCHG. | |
1140 | */ | |
1141 | ||
1142 | prev->next = new; | |
1143 | new->prev = prev; | |
1144 | ||
1145 | return true; | |
1146 | } | |
1147 | ||
1148 | /** | |
1149 | * Chains a new epi entry to the tail of the ep->ovflist in a lockless way, | |
1150 | * i.e. multiple CPUs are allowed to call this function concurrently. | |
1151 | * | |
1152 | * Returns %false if epi element has been already chained, %true otherwise. | |
1153 | */ | |
1154 | static inline bool chain_epi_lockless(struct epitem *epi) | |
1155 | { | |
1156 | struct eventpoll *ep = epi->ep; | |
1157 | ||
1158 | /* Check that the same epi has not been just chained from another CPU */ | |
1159 | if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR) | |
1160 | return false; | |
1161 | ||
1162 | /* Atomically exchange tail */ | |
1163 | epi->next = xchg(&ep->ovflist, epi); | |
1164 | ||
1165 | return true; | |
1166 | } | |
1167 | ||
1da177e4 | 1168 | /* |
7699acd1 | 1169 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 1170 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 1171 | * have events to report. |
a218cc49 RP |
1172 | * |
1173 | * This callback takes a read lock in order not to content with concurrent | |
1174 | * events from another file descriptors, thus all modifications to ->rdllist | |
1175 | * or ->ovflist are lockless. Read lock is paired with the write lock from | |
1176 | * ep_scan_ready_list(), which stops all list modifications and guarantees | |
1177 | * that lists state is seen correctly. | |
1178 | * | |
1179 | * Another thing worth to mention is that ep_poll_callback() can be called | |
1180 | * concurrently for the same @epi from different CPUs if poll table was inited | |
1181 | * with several wait queues entries. Plural wakeup from different CPUs of a | |
1182 | * single wait queue is serialized by wq.lock, but the case when multiple wait | |
1183 | * queues are used should be detected accordingly. This is detected using | |
1184 | * cmpxchg() operation. | |
1da177e4 | 1185 | */ |
ac6424b9 | 1186 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 1187 | { |
7699acd1 | 1188 | int pwake = 0; |
7699acd1 DL |
1189 | struct epitem *epi = ep_item_from_wait(wait); |
1190 | struct eventpoll *ep = epi->ep; | |
3ad6f93e | 1191 | __poll_t pollflags = key_to_poll(key); |
a218cc49 | 1192 | unsigned long flags; |
df0108c5 | 1193 | int ewake = 0; |
1da177e4 | 1194 | |
a218cc49 | 1195 | read_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1196 | |
bf3b9f63 SS |
1197 | ep_set_busy_poll_napi_id(epi); |
1198 | ||
7699acd1 DL |
1199 | /* |
1200 | * If the event mask does not contain any poll(2) event, we consider the | |
1201 | * descriptor to be disabled. This condition is likely the effect of the | |
1202 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1203 | * until the next EPOLL_CTL_MOD will be issued. | |
1204 | */ | |
1205 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
1206 | goto out_unlock; |
1207 | ||
2dfa4eea DL |
1208 | /* |
1209 | * Check the events coming with the callback. At this stage, not | |
1210 | * every device reports the events in the "key" parameter of the | |
1211 | * callback. We need to be able to handle both cases here, hence the | |
1212 | * test for "key" != NULL before the event match test. | |
1213 | */ | |
3ad6f93e | 1214 | if (pollflags && !(pollflags & epi->event.events)) |
2dfa4eea DL |
1215 | goto out_unlock; |
1216 | ||
d47de16c | 1217 | /* |
bf6a41db | 1218 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 1219 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 1220 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
1221 | * chained in ep->ovflist and requeued later on. |
1222 | */ | |
c5a282e9 | 1223 | if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) { |
a218cc49 RP |
1224 | if (epi->next == EP_UNACTIVE_PTR && |
1225 | chain_epi_lockless(epi)) | |
c3e320b6 | 1226 | ep_pm_stay_awake_rcu(epi); |
d47de16c DL |
1227 | goto out_unlock; |
1228 | } | |
1da177e4 | 1229 | |
7699acd1 | 1230 | /* If this file is already in the ready list we exit soon */ |
a218cc49 RP |
1231 | if (!ep_is_linked(epi) && |
1232 | list_add_tail_lockless(&epi->rdllink, &ep->rdllist)) { | |
eea1d585 | 1233 | ep_pm_stay_awake_rcu(epi); |
4d7e30d9 | 1234 | } |
7699acd1 | 1235 | |
7699acd1 DL |
1236 | /* |
1237 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1238 | * wait list. | |
1239 | */ | |
df0108c5 | 1240 | if (waitqueue_active(&ep->wq)) { |
b6a515c8 | 1241 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
3ad6f93e AV |
1242 | !(pollflags & POLLFREE)) { |
1243 | switch (pollflags & EPOLLINOUT_BITS) { | |
a9a08845 LT |
1244 | case EPOLLIN: |
1245 | if (epi->event.events & EPOLLIN) | |
b6a515c8 JB |
1246 | ewake = 1; |
1247 | break; | |
a9a08845 LT |
1248 | case EPOLLOUT: |
1249 | if (epi->event.events & EPOLLOUT) | |
b6a515c8 JB |
1250 | ewake = 1; |
1251 | break; | |
1252 | case 0: | |
1253 | ewake = 1; | |
1254 | break; | |
1255 | } | |
1256 | } | |
a218cc49 | 1257 | wake_up(&ep->wq); |
df0108c5 | 1258 | } |
7699acd1 DL |
1259 | if (waitqueue_active(&ep->poll_wait)) |
1260 | pwake++; | |
1261 | ||
d47de16c | 1262 | out_unlock: |
a218cc49 | 1263 | read_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1264 | |
7699acd1 DL |
1265 | /* We have to call this outside the lock */ |
1266 | if (pwake) | |
5071f97e | 1267 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 | 1268 | |
138e4ad6 ON |
1269 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
1270 | ewake = 1; | |
1271 | ||
3ad6f93e | 1272 | if (pollflags & POLLFREE) { |
138e4ad6 ON |
1273 | /* |
1274 | * If we race with ep_remove_wait_queue() it can miss | |
1275 | * ->whead = NULL and do another remove_wait_queue() after | |
1276 | * us, so we can't use __remove_wait_queue(). | |
1277 | */ | |
1278 | list_del_init(&wait->entry); | |
1279 | /* | |
1280 | * ->whead != NULL protects us from the race with ep_free() | |
1281 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock | |
1282 | * held by the caller. Once we nullify it, nothing protects | |
1283 | * ep/epi or even wait. | |
1284 | */ | |
1285 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); | |
1286 | } | |
df0108c5 | 1287 | |
138e4ad6 | 1288 | return ewake; |
7699acd1 | 1289 | } |
1da177e4 LT |
1290 | |
1291 | /* | |
1292 | * This is the callback that is used to add our wait queue to the | |
1293 | * target file wakeup lists. | |
1294 | */ | |
1295 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1296 | poll_table *pt) | |
1297 | { | |
b030a4dd | 1298 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
1299 | struct eppoll_entry *pwq; |
1300 | ||
e94b1766 | 1301 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
1302 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
1303 | pwq->whead = whead; | |
1304 | pwq->base = epi; | |
df0108c5 JB |
1305 | if (epi->event.events & EPOLLEXCLUSIVE) |
1306 | add_wait_queue_exclusive(whead, &pwq->wait); | |
1307 | else | |
1308 | add_wait_queue(whead, &pwq->wait); | |
1da177e4 LT |
1309 | list_add_tail(&pwq->llink, &epi->pwqlist); |
1310 | epi->nwait++; | |
296e236e | 1311 | } else { |
1da177e4 LT |
1312 | /* We have to signal that an error occurred */ |
1313 | epi->nwait = -1; | |
296e236e | 1314 | } |
1da177e4 LT |
1315 | } |
1316 | ||
1da177e4 LT |
1317 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1318 | { | |
1319 | int kcmp; | |
b2ac2ea6 | 1320 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
1da177e4 | 1321 | struct epitem *epic; |
b2ac2ea6 | 1322 | bool leftmost = true; |
1da177e4 LT |
1323 | |
1324 | while (*p) { | |
1325 | parent = *p; | |
1326 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1327 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
b2ac2ea6 | 1328 | if (kcmp > 0) { |
1da177e4 | 1329 | p = &parent->rb_right; |
b2ac2ea6 DB |
1330 | leftmost = false; |
1331 | } else | |
1da177e4 LT |
1332 | p = &parent->rb_left; |
1333 | } | |
1334 | rb_link_node(&epi->rbn, parent, p); | |
b2ac2ea6 | 1335 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
1da177e4 LT |
1336 | } |
1337 | ||
a80a6b85 AM |
1338 | |
1339 | ||
28d82dc1 JB |
1340 | #define PATH_ARR_SIZE 5 |
1341 | /* | |
1342 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1343 | * from a single file of interest. For example, we allow 1000 paths of length | |
1344 | * 1, to emanate from each file of interest. This essentially represents the | |
1345 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1346 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1347 | * is connected to n file sources. In this case each file source has 1 path | |
1348 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1349 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1350 | * and delete can't add additional paths. Protected by the epmutex. | |
1351 | */ | |
1352 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1353 | static int path_count[PATH_ARR_SIZE]; | |
1354 | ||
1355 | static int path_count_inc(int nests) | |
1356 | { | |
93dc6107 JB |
1357 | /* Allow an arbitrary number of depth 1 paths */ |
1358 | if (nests == 0) | |
1359 | return 0; | |
1360 | ||
28d82dc1 JB |
1361 | if (++path_count[nests] > path_limits[nests]) |
1362 | return -1; | |
1363 | return 0; | |
1364 | } | |
1365 | ||
1366 | static void path_count_init(void) | |
1367 | { | |
1368 | int i; | |
1369 | ||
1370 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1371 | path_count[i] = 0; | |
1372 | } | |
1373 | ||
1374 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) | |
1375 | { | |
1376 | int error = 0; | |
1377 | struct file *file = priv; | |
1378 | struct file *child_file; | |
1379 | struct epitem *epi; | |
1380 | ||
ae10b2b4 JB |
1381 | /* CTL_DEL can remove links here, but that can't increase our count */ |
1382 | rcu_read_lock(); | |
1383 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { | |
28d82dc1 JB |
1384 | child_file = epi->ep->file; |
1385 | if (is_file_epoll(child_file)) { | |
1386 | if (list_empty(&child_file->f_ep_links)) { | |
1387 | if (path_count_inc(call_nests)) { | |
1388 | error = -1; | |
1389 | break; | |
1390 | } | |
1391 | } else { | |
1392 | error = ep_call_nested(&poll_loop_ncalls, | |
28d82dc1 JB |
1393 | reverse_path_check_proc, |
1394 | child_file, child_file, | |
1395 | current); | |
1396 | } | |
1397 | if (error != 0) | |
1398 | break; | |
1399 | } else { | |
1400 | printk(KERN_ERR "reverse_path_check_proc: " | |
1401 | "file is not an ep!\n"); | |
1402 | } | |
1403 | } | |
ae10b2b4 | 1404 | rcu_read_unlock(); |
28d82dc1 JB |
1405 | return error; |
1406 | } | |
1407 | ||
1408 | /** | |
1409 | * reverse_path_check - The tfile_check_list is list of file *, which have | |
1410 | * links that are proposed to be newly added. We need to | |
1411 | * make sure that those added links don't add too many | |
1412 | * paths such that we will spend all our time waking up | |
1413 | * eventpoll objects. | |
1414 | * | |
1415 | * Returns: Returns zero if the proposed links don't create too many paths, | |
1416 | * -1 otherwise. | |
1417 | */ | |
1418 | static int reverse_path_check(void) | |
1419 | { | |
28d82dc1 JB |
1420 | int error = 0; |
1421 | struct file *current_file; | |
1422 | ||
1423 | /* let's call this for all tfiles */ | |
1424 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { | |
28d82dc1 | 1425 | path_count_init(); |
74bdc129 | 1426 | error = ep_call_nested(&poll_loop_ncalls, |
28d82dc1 JB |
1427 | reverse_path_check_proc, current_file, |
1428 | current_file, current); | |
1429 | if (error) | |
1430 | break; | |
1431 | } | |
1432 | return error; | |
1433 | } | |
1434 | ||
4d7e30d9 AH |
1435 | static int ep_create_wakeup_source(struct epitem *epi) |
1436 | { | |
1437 | const char *name; | |
eea1d585 | 1438 | struct wakeup_source *ws; |
4d7e30d9 AH |
1439 | |
1440 | if (!epi->ep->ws) { | |
c8377adf | 1441 | epi->ep->ws = wakeup_source_register(NULL, "eventpoll"); |
4d7e30d9 AH |
1442 | if (!epi->ep->ws) |
1443 | return -ENOMEM; | |
1444 | } | |
1445 | ||
1446 | name = epi->ffd.file->f_path.dentry->d_name.name; | |
c8377adf | 1447 | ws = wakeup_source_register(NULL, name); |
eea1d585 EW |
1448 | |
1449 | if (!ws) | |
4d7e30d9 | 1450 | return -ENOMEM; |
eea1d585 | 1451 | rcu_assign_pointer(epi->ws, ws); |
4d7e30d9 AH |
1452 | |
1453 | return 0; | |
1454 | } | |
1455 | ||
eea1d585 EW |
1456 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
1457 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) | |
4d7e30d9 | 1458 | { |
eea1d585 EW |
1459 | struct wakeup_source *ws = ep_wakeup_source(epi); |
1460 | ||
d6d67e72 | 1461 | RCU_INIT_POINTER(epi->ws, NULL); |
eea1d585 EW |
1462 | |
1463 | /* | |
1464 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is | |
1465 | * used internally by wakeup_source_remove, too (called by | |
1466 | * wakeup_source_unregister), so we cannot use call_rcu | |
1467 | */ | |
1468 | synchronize_rcu(); | |
1469 | wakeup_source_unregister(ws); | |
4d7e30d9 AH |
1470 | } |
1471 | ||
c7ea7630 DL |
1472 | /* |
1473 | * Must be called with "mtx" held. | |
1474 | */ | |
bec1a502 | 1475 | static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, |
67347fe4 | 1476 | struct file *tfile, int fd, int full_check) |
1da177e4 | 1477 | { |
d85e2aa2 AV |
1478 | int error, pwake = 0; |
1479 | __poll_t revents; | |
52bd19f7 | 1480 | long user_watches; |
1da177e4 LT |
1481 | struct epitem *epi; |
1482 | struct ep_pqueue epq; | |
1483 | ||
92e64178 DB |
1484 | lockdep_assert_irqs_enabled(); |
1485 | ||
52bd19f7 RH |
1486 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
1487 | if (unlikely(user_watches >= max_user_watches)) | |
7ef9964e | 1488 | return -ENOSPC; |
e94b1766 | 1489 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 1490 | return -ENOMEM; |
1da177e4 LT |
1491 | |
1492 | /* Item initialization follow here ... */ | |
1da177e4 LT |
1493 | INIT_LIST_HEAD(&epi->rdllink); |
1494 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
1495 | INIT_LIST_HEAD(&epi->pwqlist); |
1496 | epi->ep = ep; | |
b030a4dd | 1497 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1498 | epi->event = *event; |
1da177e4 | 1499 | epi->nwait = 0; |
d47de16c | 1500 | epi->next = EP_UNACTIVE_PTR; |
4d7e30d9 AH |
1501 | if (epi->event.events & EPOLLWAKEUP) { |
1502 | error = ep_create_wakeup_source(epi); | |
1503 | if (error) | |
1504 | goto error_create_wakeup_source; | |
1505 | } else { | |
eea1d585 | 1506 | RCU_INIT_POINTER(epi->ws, NULL); |
4d7e30d9 | 1507 | } |
1da177e4 LT |
1508 | |
1509 | /* Initialize the poll table using the queue callback */ | |
1510 | epq.epi = epi; | |
1511 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
1512 | ||
1513 | /* | |
1514 | * Attach the item to the poll hooks and get current event bits. | |
1515 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1516 | * been increased by the caller of this function. Note that after |
1517 | * this operation completes, the poll callback can start hitting | |
1518 | * the new item. | |
1da177e4 | 1519 | */ |
37b5e521 | 1520 | revents = ep_item_poll(epi, &epq.pt, 1); |
1da177e4 LT |
1521 | |
1522 | /* | |
1523 | * We have to check if something went wrong during the poll wait queue | |
1524 | * install process. Namely an allocation for a wait queue failed due | |
1525 | * high memory pressure. | |
1526 | */ | |
7ef9964e | 1527 | error = -ENOMEM; |
1da177e4 | 1528 | if (epi->nwait < 0) |
7699acd1 | 1529 | goto error_unregister; |
1da177e4 LT |
1530 | |
1531 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 1532 | spin_lock(&tfile->f_lock); |
ae10b2b4 | 1533 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); |
68499914 | 1534 | spin_unlock(&tfile->f_lock); |
1da177e4 | 1535 | |
c7ea7630 DL |
1536 | /* |
1537 | * Add the current item to the RB tree. All RB tree operations are | |
1538 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1539 | */ | |
1da177e4 LT |
1540 | ep_rbtree_insert(ep, epi); |
1541 | ||
28d82dc1 JB |
1542 | /* now check if we've created too many backpaths */ |
1543 | error = -EINVAL; | |
67347fe4 | 1544 | if (full_check && reverse_path_check()) |
28d82dc1 JB |
1545 | goto error_remove_epi; |
1546 | ||
c7ea7630 | 1547 | /* We have to drop the new item inside our item list to keep track of it */ |
a218cc49 | 1548 | write_lock_irq(&ep->lock); |
c7ea7630 | 1549 | |
bf3b9f63 SS |
1550 | /* record NAPI ID of new item if present */ |
1551 | ep_set_busy_poll_napi_id(epi); | |
1552 | ||
1da177e4 | 1553 | /* If the file is already "ready" we drop it inside the ready list */ |
992991c0 | 1554 | if (revents && !ep_is_linked(epi)) { |
1da177e4 | 1555 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1556 | ep_pm_stay_awake(epi); |
1da177e4 LT |
1557 | |
1558 | /* Notify waiting tasks that events are available */ | |
1559 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1560 | wake_up(&ep->wq); |
1da177e4 LT |
1561 | if (waitqueue_active(&ep->poll_wait)) |
1562 | pwake++; | |
1563 | } | |
1564 | ||
a218cc49 | 1565 | write_unlock_irq(&ep->lock); |
1da177e4 | 1566 | |
52bd19f7 | 1567 | atomic_long_inc(&ep->user->epoll_watches); |
7ef9964e | 1568 | |
1da177e4 LT |
1569 | /* We have to call this outside the lock */ |
1570 | if (pwake) | |
5071f97e | 1571 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1572 | |
1da177e4 LT |
1573 | return 0; |
1574 | ||
28d82dc1 JB |
1575 | error_remove_epi: |
1576 | spin_lock(&tfile->f_lock); | |
ae10b2b4 | 1577 | list_del_rcu(&epi->fllink); |
28d82dc1 JB |
1578 | spin_unlock(&tfile->f_lock); |
1579 | ||
b2ac2ea6 | 1580 | rb_erase_cached(&epi->rbn, &ep->rbr); |
28d82dc1 | 1581 | |
7699acd1 | 1582 | error_unregister: |
1da177e4 LT |
1583 | ep_unregister_pollwait(ep, epi); |
1584 | ||
1585 | /* | |
1586 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
1587 | * allocated wait queue. Note that we don't care about the ep->ovflist |
1588 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
1589 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 1590 | */ |
a218cc49 | 1591 | write_lock_irq(&ep->lock); |
992991c0 | 1592 | if (ep_is_linked(epi)) |
6192bd53 | 1593 | list_del_init(&epi->rdllink); |
a218cc49 | 1594 | write_unlock_irq(&ep->lock); |
1da177e4 | 1595 | |
eea1d585 | 1596 | wakeup_source_unregister(ep_wakeup_source(epi)); |
4d7e30d9 AH |
1597 | |
1598 | error_create_wakeup_source: | |
b030a4dd | 1599 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1600 | |
1da177e4 LT |
1601 | return error; |
1602 | } | |
1603 | ||
1da177e4 LT |
1604 | /* |
1605 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1606 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 | 1607 | */ |
bec1a502 AV |
1608 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, |
1609 | const struct epoll_event *event) | |
1da177e4 LT |
1610 | { |
1611 | int pwake = 0; | |
626cf236 HV |
1612 | poll_table pt; |
1613 | ||
92e64178 DB |
1614 | lockdep_assert_irqs_enabled(); |
1615 | ||
626cf236 | 1616 | init_poll_funcptr(&pt, NULL); |
1da177e4 LT |
1617 | |
1618 | /* | |
e057e15f TB |
1619 | * Set the new event interest mask before calling f_op->poll(); |
1620 | * otherwise we might miss an event that happens between the | |
1621 | * f_op->poll() call and the new event set registering. | |
1da177e4 | 1622 | */ |
128dd175 | 1623 | epi->event.events = event->events; /* need barrier below */ |
e057e15f | 1624 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 | 1625 | if (epi->event.events & EPOLLWAKEUP) { |
eea1d585 | 1626 | if (!ep_has_wakeup_source(epi)) |
4d7e30d9 | 1627 | ep_create_wakeup_source(epi); |
eea1d585 | 1628 | } else if (ep_has_wakeup_source(epi)) { |
4d7e30d9 AH |
1629 | ep_destroy_wakeup_source(epi); |
1630 | } | |
1da177e4 | 1631 | |
128dd175 EW |
1632 | /* |
1633 | * The following barrier has two effects: | |
1634 | * | |
1635 | * 1) Flush epi changes above to other CPUs. This ensures | |
1636 | * we do not miss events from ep_poll_callback if an | |
1637 | * event occurs immediately after we call f_op->poll(). | |
a218cc49 | 1638 | * We need this because we did not take ep->lock while |
128dd175 | 1639 | * changing epi above (but ep_poll_callback does take |
a218cc49 | 1640 | * ep->lock). |
128dd175 EW |
1641 | * |
1642 | * 2) We also need to ensure we do not miss _past_ events | |
1643 | * when calling f_op->poll(). This barrier also | |
1644 | * pairs with the barrier in wq_has_sleeper (see | |
1645 | * comments for wq_has_sleeper). | |
1646 | * | |
1647 | * This barrier will now guarantee ep_poll_callback or f_op->poll | |
1648 | * (or both) will notice the readiness of an item. | |
1649 | */ | |
1650 | smp_mb(); | |
1651 | ||
1da177e4 LT |
1652 | /* |
1653 | * Get current event bits. We can safely use the file* here because | |
1654 | * its usage count has been increased by the caller of this function. | |
c7ea7630 | 1655 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1656 | * list, push it inside. |
1da177e4 | 1657 | */ |
69112736 | 1658 | if (ep_item_poll(epi, &pt, 1)) { |
a218cc49 | 1659 | write_lock_irq(&ep->lock); |
992991c0 | 1660 | if (!ep_is_linked(epi)) { |
c7ea7630 | 1661 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1662 | ep_pm_stay_awake(epi); |
c7ea7630 DL |
1663 | |
1664 | /* Notify waiting tasks that events are available */ | |
1665 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1666 | wake_up(&ep->wq); |
c7ea7630 DL |
1667 | if (waitqueue_active(&ep->poll_wait)) |
1668 | pwake++; | |
7699acd1 | 1669 | } |
a218cc49 | 1670 | write_unlock_irq(&ep->lock); |
7699acd1 | 1671 | } |
1da177e4 | 1672 | |
7699acd1 DL |
1673 | /* We have to call this outside the lock */ |
1674 | if (pwake) | |
5071f97e | 1675 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1676 | |
7699acd1 | 1677 | return 0; |
1da177e4 LT |
1678 | } |
1679 | ||
d85e2aa2 | 1680 | static __poll_t ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
296e236e | 1681 | void *priv) |
1da177e4 | 1682 | { |
5071f97e | 1683 | struct ep_send_events_data *esed = priv; |
d85e2aa2 | 1684 | __poll_t revents; |
4e0982a0 DB |
1685 | struct epitem *epi, *tmp; |
1686 | struct epoll_event __user *uevent = esed->events; | |
eea1d585 | 1687 | struct wakeup_source *ws; |
626cf236 HV |
1688 | poll_table pt; |
1689 | ||
1690 | init_poll_funcptr(&pt, NULL); | |
4e0982a0 | 1691 | esed->res = 0; |
1da177e4 | 1692 | |
296e236e | 1693 | /* |
5071f97e DL |
1694 | * We can loop without lock because we are passed a task private list. |
1695 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1696 | * holding "mtx" during this call. | |
296e236e | 1697 | */ |
21877e1a DB |
1698 | lockdep_assert_held(&ep->mtx); |
1699 | ||
4e0982a0 DB |
1700 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
1701 | if (esed->res >= esed->maxevents) | |
1702 | break; | |
d47de16c | 1703 | |
4d7e30d9 AH |
1704 | /* |
1705 | * Activate ep->ws before deactivating epi->ws to prevent | |
1706 | * triggering auto-suspend here (in case we reactive epi->ws | |
1707 | * below). | |
1708 | * | |
1709 | * This could be rearranged to delay the deactivation of epi->ws | |
1710 | * instead, but then epi->ws would temporarily be out of sync | |
1711 | * with ep_is_linked(). | |
1712 | */ | |
eea1d585 EW |
1713 | ws = ep_wakeup_source(epi); |
1714 | if (ws) { | |
1715 | if (ws->active) | |
1716 | __pm_stay_awake(ep->ws); | |
1717 | __pm_relax(ws); | |
1718 | } | |
1719 | ||
d47de16c | 1720 | list_del_init(&epi->rdllink); |
1da177e4 | 1721 | |
296e236e | 1722 | /* |
5071f97e DL |
1723 | * If the event mask intersect the caller-requested one, |
1724 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
4e0982a0 | 1725 | * is holding ep->mtx, so no operations coming from userspace |
5071f97e | 1726 | * can change the item. |
296e236e | 1727 | */ |
4e0982a0 DB |
1728 | revents = ep_item_poll(epi, &pt, 1); |
1729 | if (!revents) | |
1730 | continue; | |
1731 | ||
1732 | if (__put_user(revents, &uevent->events) || | |
1733 | __put_user(epi->event.data, &uevent->data)) { | |
1734 | list_add(&epi->rdllink, head); | |
1735 | ep_pm_stay_awake(epi); | |
1736 | if (!esed->res) | |
1737 | esed->res = -EFAULT; | |
1738 | return 0; | |
1739 | } | |
1740 | esed->res++; | |
1741 | uevent++; | |
1742 | if (epi->event.events & EPOLLONESHOT) | |
1743 | epi->event.events &= EP_PRIVATE_BITS; | |
1744 | else if (!(epi->event.events & EPOLLET)) { | |
1745 | /* | |
1746 | * If this file has been added with Level | |
1747 | * Trigger mode, we need to insert back inside | |
1748 | * the ready list, so that the next call to | |
1749 | * epoll_wait() will check again the events | |
1750 | * availability. At this point, no one can insert | |
1751 | * into ep->rdllist besides us. The epoll_ctl() | |
1752 | * callers are locked out by | |
1753 | * ep_scan_ready_list() holding "mtx" and the | |
1754 | * poll callback will queue them in ep->ovflist. | |
1755 | */ | |
1756 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1757 | ep_pm_stay_awake(epi); | |
296e236e DL |
1758 | } |
1759 | } | |
5071f97e | 1760 | |
d7ebbe46 | 1761 | return 0; |
5071f97e | 1762 | } |
d47de16c | 1763 | |
296e236e DL |
1764 | static int ep_send_events(struct eventpoll *ep, |
1765 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1766 | { |
1767 | struct ep_send_events_data esed; | |
1da177e4 | 1768 | |
5071f97e DL |
1769 | esed.maxevents = maxevents; |
1770 | esed.events = events; | |
6192bd53 | 1771 | |
d7ebbe46 AV |
1772 | ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false); |
1773 | return esed.res; | |
1da177e4 LT |
1774 | } |
1775 | ||
766b9f92 | 1776 | static inline struct timespec64 ep_set_mstimeout(long ms) |
0781b909 | 1777 | { |
766b9f92 | 1778 | struct timespec64 now, ts = { |
0781b909 ED |
1779 | .tv_sec = ms / MSEC_PER_SEC, |
1780 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), | |
1781 | }; | |
1782 | ||
766b9f92 DD |
1783 | ktime_get_ts64(&now); |
1784 | return timespec64_add_safe(now, ts); | |
0781b909 ED |
1785 | } |
1786 | ||
f4d93ad7 SB |
1787 | /** |
1788 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied | |
1789 | * event buffer. | |
1790 | * | |
1791 | * @ep: Pointer to the eventpoll context. | |
1792 | * @events: Pointer to the userspace buffer where the ready events should be | |
1793 | * stored. | |
1794 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1795 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
1796 | * milliseconds. If the @timeout is zero, the function will not block, | |
1797 | * while if the @timeout is less than zero, the function will block | |
1798 | * until at least one event has been retrieved (or an error | |
1799 | * occurred). | |
1800 | * | |
1801 | * Returns: Returns the number of ready events which have been fetched, or an | |
1802 | * error code, in case of error. | |
1803 | */ | |
1da177e4 LT |
1804 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1805 | int maxevents, long timeout) | |
1806 | { | |
f4d93ad7 | 1807 | int res = 0, eavail, timed_out = 0; |
da8b44d5 | 1808 | u64 slack = 0; |
86c05179 | 1809 | bool waiter = false; |
ac6424b9 | 1810 | wait_queue_entry_t wait; |
95aac7b1 SB |
1811 | ktime_t expires, *to = NULL; |
1812 | ||
679abf38 DB |
1813 | lockdep_assert_irqs_enabled(); |
1814 | ||
95aac7b1 | 1815 | if (timeout > 0) { |
766b9f92 | 1816 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
0781b909 | 1817 | |
95aac7b1 SB |
1818 | slack = select_estimate_accuracy(&end_time); |
1819 | to = &expires; | |
766b9f92 | 1820 | *to = timespec64_to_ktime(end_time); |
95aac7b1 | 1821 | } else if (timeout == 0) { |
f4d93ad7 SB |
1822 | /* |
1823 | * Avoid the unnecessary trip to the wait queue loop, if the | |
c5a282e9 DB |
1824 | * caller specified a non blocking operation. We still need |
1825 | * lock because we could race and not see an epi being added | |
1826 | * to the ready list while in irq callback. Thus incorrectly | |
1827 | * returning 0 back to userspace. | |
f4d93ad7 | 1828 | */ |
95aac7b1 | 1829 | timed_out = 1; |
c5a282e9 | 1830 | |
a218cc49 | 1831 | write_lock_irq(&ep->lock); |
c5a282e9 | 1832 | eavail = ep_events_available(ep); |
a218cc49 | 1833 | write_unlock_irq(&ep->lock); |
c5a282e9 | 1834 | |
35cff1a6 | 1835 | goto send_events; |
95aac7b1 | 1836 | } |
1da177e4 | 1837 | |
f4d93ad7 | 1838 | fetch_events: |
bf3b9f63 SS |
1839 | |
1840 | if (!ep_events_available(ep)) | |
1841 | ep_busy_loop(ep, timed_out); | |
1842 | ||
c5a282e9 DB |
1843 | eavail = ep_events_available(ep); |
1844 | if (eavail) | |
35cff1a6 | 1845 | goto send_events; |
1da177e4 | 1846 | |
c5a282e9 DB |
1847 | /* |
1848 | * Busy poll timed out. Drop NAPI ID for now, we can add | |
1849 | * it back in when we have moved a socket with a valid NAPI | |
1850 | * ID onto the ready list. | |
1851 | */ | |
1852 | ep_reset_busy_poll_napi_id(ep); | |
bf3b9f63 | 1853 | |
c5a282e9 | 1854 | /* |
86c05179 DB |
1855 | * We don't have any available event to return to the caller. We need |
1856 | * to sleep here, and we will be woken by ep_poll_callback() when events | |
1857 | * become available. | |
c5a282e9 | 1858 | */ |
86c05179 DB |
1859 | if (!waiter) { |
1860 | waiter = true; | |
1da177e4 | 1861 | init_waitqueue_entry(&wait, current); |
1da177e4 | 1862 | |
86c05179 DB |
1863 | spin_lock_irq(&ep->wq.lock); |
1864 | __add_wait_queue_exclusive(&ep->wq, &wait); | |
1865 | spin_unlock_irq(&ep->wq.lock); | |
1866 | } | |
1da177e4 | 1867 | |
c5a282e9 | 1868 | for (;;) { |
bf3b9f63 | 1869 | /* |
c5a282e9 DB |
1870 | * We don't want to sleep if the ep_poll_callback() sends us |
1871 | * a wakeup in between. That's why we set the task state | |
1872 | * to TASK_INTERRUPTIBLE before doing the checks. | |
bf3b9f63 | 1873 | */ |
c5a282e9 | 1874 | set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 | 1875 | /* |
c5a282e9 DB |
1876 | * Always short-circuit for fatal signals to allow |
1877 | * threads to make a timely exit without the chance of | |
1878 | * finding more events available and fetching | |
1879 | * repeatedly. | |
1da177e4 | 1880 | */ |
c5a282e9 DB |
1881 | if (fatal_signal_pending(current)) { |
1882 | res = -EINTR; | |
1883 | break; | |
1884 | } | |
95aac7b1 | 1885 | |
abc610e0 DB |
1886 | eavail = ep_events_available(ep); |
1887 | if (eavail) | |
c5a282e9 DB |
1888 | break; |
1889 | if (signal_pending(current)) { | |
1890 | res = -EINTR; | |
1891 | break; | |
1da177e4 | 1892 | } |
1da177e4 | 1893 | |
abc610e0 | 1894 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) { |
c5a282e9 | 1895 | timed_out = 1; |
abc610e0 DB |
1896 | break; |
1897 | } | |
1da177e4 | 1898 | } |
1da177e4 | 1899 | |
c5a282e9 | 1900 | __set_current_state(TASK_RUNNING); |
1da177e4 | 1901 | |
35cff1a6 | 1902 | send_events: |
1da177e4 LT |
1903 | /* |
1904 | * Try to transfer events to user space. In case we get 0 events and | |
1905 | * there's still timeout left over, we go trying again in search of | |
1906 | * more luck. | |
1907 | */ | |
1908 | if (!res && eavail && | |
95aac7b1 | 1909 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
f4d93ad7 | 1910 | goto fetch_events; |
1da177e4 | 1911 | |
86c05179 DB |
1912 | if (waiter) { |
1913 | spin_lock_irq(&ep->wq.lock); | |
1914 | __remove_wait_queue(&ep->wq, &wait); | |
1915 | spin_unlock_irq(&ep->wq.lock); | |
1916 | } | |
1917 | ||
1da177e4 LT |
1918 | return res; |
1919 | } | |
1920 | ||
22bacca4 DL |
1921 | /** |
1922 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() | |
1923 | * API, to verify that adding an epoll file inside another | |
1924 | * epoll structure, does not violate the constraints, in | |
1925 | * terms of closed loops, or too deep chains (which can | |
1926 | * result in excessive stack usage). | |
1927 | * | |
1928 | * @priv: Pointer to the epoll file to be currently checked. | |
1929 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll | |
1930 | * data structure pointer. | |
1931 | * @call_nests: Current dept of the @ep_call_nested() call stack. | |
1932 | * | |
1933 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1934 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1935 | */ | |
1936 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) | |
1937 | { | |
1938 | int error = 0; | |
1939 | struct file *file = priv; | |
1940 | struct eventpoll *ep = file->private_data; | |
28d82dc1 | 1941 | struct eventpoll *ep_tovisit; |
22bacca4 DL |
1942 | struct rb_node *rbp; |
1943 | struct epitem *epi; | |
1944 | ||
d8805e63 | 1945 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
28d82dc1 JB |
1946 | ep->visited = 1; |
1947 | list_add(&ep->visited_list_link, &visited_list); | |
b2ac2ea6 | 1948 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
22bacca4 DL |
1949 | epi = rb_entry(rbp, struct epitem, rbn); |
1950 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
28d82dc1 JB |
1951 | ep_tovisit = epi->ffd.file->private_data; |
1952 | if (ep_tovisit->visited) | |
1953 | continue; | |
74bdc129 | 1954 | error = ep_call_nested(&poll_loop_ncalls, |
28d82dc1 JB |
1955 | ep_loop_check_proc, epi->ffd.file, |
1956 | ep_tovisit, current); | |
22bacca4 DL |
1957 | if (error != 0) |
1958 | break; | |
28d82dc1 JB |
1959 | } else { |
1960 | /* | |
1961 | * If we've reached a file that is not associated with | |
1962 | * an ep, then we need to check if the newly added | |
1963 | * links are going to add too many wakeup paths. We do | |
1964 | * this by adding it to the tfile_check_list, if it's | |
1965 | * not already there, and calling reverse_path_check() | |
1966 | * during ep_insert(). | |
1967 | */ | |
1968 | if (list_empty(&epi->ffd.file->f_tfile_llink)) | |
1969 | list_add(&epi->ffd.file->f_tfile_llink, | |
1970 | &tfile_check_list); | |
22bacca4 DL |
1971 | } |
1972 | } | |
1973 | mutex_unlock(&ep->mtx); | |
1974 | ||
1975 | return error; | |
1976 | } | |
1977 | ||
1978 | /** | |
1979 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) | |
1980 | * another epoll file (represented by @ep) does not create | |
1981 | * closed loops or too deep chains. | |
1982 | * | |
1983 | * @ep: Pointer to the epoll private data structure. | |
1984 | * @file: Pointer to the epoll file to be checked. | |
1985 | * | |
1986 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1987 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1988 | */ | |
1989 | static int ep_loop_check(struct eventpoll *ep, struct file *file) | |
1990 | { | |
28d82dc1 JB |
1991 | int ret; |
1992 | struct eventpoll *ep_cur, *ep_next; | |
1993 | ||
74bdc129 | 1994 | ret = ep_call_nested(&poll_loop_ncalls, |
22bacca4 | 1995 | ep_loop_check_proc, file, ep, current); |
28d82dc1 JB |
1996 | /* clear visited list */ |
1997 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, | |
1998 | visited_list_link) { | |
1999 | ep_cur->visited = 0; | |
2000 | list_del(&ep_cur->visited_list_link); | |
2001 | } | |
2002 | return ret; | |
2003 | } | |
2004 | ||
2005 | static void clear_tfile_check_list(void) | |
2006 | { | |
2007 | struct file *file; | |
2008 | ||
2009 | /* first clear the tfile_check_list */ | |
2010 | while (!list_empty(&tfile_check_list)) { | |
2011 | file = list_first_entry(&tfile_check_list, struct file, | |
2012 | f_tfile_llink); | |
2013 | list_del_init(&file->f_tfile_llink); | |
2014 | } | |
2015 | INIT_LIST_HEAD(&tfile_check_list); | |
22bacca4 DL |
2016 | } |
2017 | ||
7699acd1 | 2018 | /* |
523723bb | 2019 | * Open an eventpoll file descriptor. |
7699acd1 | 2020 | */ |
791eb22e | 2021 | static int do_epoll_create(int flags) |
7699acd1 | 2022 | { |
28d82dc1 | 2023 | int error, fd; |
bb57c3ed | 2024 | struct eventpoll *ep = NULL; |
28d82dc1 | 2025 | struct file *file; |
7699acd1 | 2026 | |
e38b36f3 UD |
2027 | /* Check the EPOLL_* constant for consistency. */ |
2028 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
2029 | ||
296e236e DL |
2030 | if (flags & ~EPOLL_CLOEXEC) |
2031 | return -EINVAL; | |
7699acd1 | 2032 | /* |
bb57c3ed | 2033 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 2034 | */ |
9fe5ad9c | 2035 | error = ep_alloc(&ep); |
bb57c3ed DL |
2036 | if (error < 0) |
2037 | return error; | |
7699acd1 DL |
2038 | /* |
2039 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 2040 | * a file structure and a free file descriptor. |
7699acd1 | 2041 | */ |
28d82dc1 JB |
2042 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
2043 | if (fd < 0) { | |
2044 | error = fd; | |
2045 | goto out_free_ep; | |
2046 | } | |
2047 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 2048 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
2049 | if (IS_ERR(file)) { |
2050 | error = PTR_ERR(file); | |
2051 | goto out_free_fd; | |
2052 | } | |
28d82dc1 | 2053 | ep->file = file; |
98022748 | 2054 | fd_install(fd, file); |
28d82dc1 JB |
2055 | return fd; |
2056 | ||
2057 | out_free_fd: | |
2058 | put_unused_fd(fd); | |
2059 | out_free_ep: | |
2060 | ep_free(ep); | |
bb57c3ed | 2061 | return error; |
7699acd1 DL |
2062 | } |
2063 | ||
791eb22e DB |
2064 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
2065 | { | |
2066 | return do_epoll_create(flags); | |
2067 | } | |
2068 | ||
5a8a82b1 | 2069 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 2070 | { |
bfe3891a | 2071 | if (size <= 0) |
9fe5ad9c UD |
2072 | return -EINVAL; |
2073 | ||
791eb22e | 2074 | return do_epoll_create(0); |
a0998b50 UD |
2075 | } |
2076 | ||
7699acd1 DL |
2077 | /* |
2078 | * The following function implements the controller interface for | |
2079 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 2080 | * file descriptors inside the interest set. |
7699acd1 | 2081 | */ |
5a8a82b1 HC |
2082 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
2083 | struct epoll_event __user *, event) | |
7699acd1 DL |
2084 | { |
2085 | int error; | |
67347fe4 | 2086 | int full_check = 0; |
7e3fb584 | 2087 | struct fd f, tf; |
7699acd1 DL |
2088 | struct eventpoll *ep; |
2089 | struct epitem *epi; | |
2090 | struct epoll_event epds; | |
67347fe4 | 2091 | struct eventpoll *tep = NULL; |
7699acd1 | 2092 | |
7699acd1 DL |
2093 | error = -EFAULT; |
2094 | if (ep_op_has_event(op) && | |
2095 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
2096 | goto error_return; | |
2097 | ||
7699acd1 | 2098 | error = -EBADF; |
7e3fb584 AV |
2099 | f = fdget(epfd); |
2100 | if (!f.file) | |
7699acd1 DL |
2101 | goto error_return; |
2102 | ||
2103 | /* Get the "struct file *" for the target file */ | |
7e3fb584 AV |
2104 | tf = fdget(fd); |
2105 | if (!tf.file) | |
7699acd1 DL |
2106 | goto error_fput; |
2107 | ||
2108 | /* The target file descriptor must support poll */ | |
2109 | error = -EPERM; | |
9965ed17 | 2110 | if (!file_can_poll(tf.file)) |
7699acd1 DL |
2111 | goto error_tgt_fput; |
2112 | ||
4d7e30d9 | 2113 | /* Check if EPOLLWAKEUP is allowed */ |
c680e41b NI |
2114 | if (ep_op_has_event(op)) |
2115 | ep_take_care_of_epollwakeup(&epds); | |
4d7e30d9 | 2116 | |
7699acd1 DL |
2117 | /* |
2118 | * We have to check that the file structure underneath the file descriptor | |
2119 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
2120 | * adding an epoll file descriptor inside itself. | |
2121 | */ | |
2122 | error = -EINVAL; | |
7e3fb584 | 2123 | if (f.file == tf.file || !is_file_epoll(f.file)) |
7699acd1 DL |
2124 | goto error_tgt_fput; |
2125 | ||
df0108c5 JB |
2126 | /* |
2127 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, | |
2128 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. | |
2129 | * Also, we do not currently supported nested exclusive wakeups. | |
2130 | */ | |
c857ab64 | 2131 | if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) { |
b6a515c8 JB |
2132 | if (op == EPOLL_CTL_MOD) |
2133 | goto error_tgt_fput; | |
2134 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || | |
2135 | (epds.events & ~EPOLLEXCLUSIVE_OK_BITS))) | |
2136 | goto error_tgt_fput; | |
2137 | } | |
df0108c5 | 2138 | |
7699acd1 DL |
2139 | /* |
2140 | * At this point it is safe to assume that the "private_data" contains | |
2141 | * our own data structure. | |
2142 | */ | |
7e3fb584 | 2143 | ep = f.file->private_data; |
7699acd1 | 2144 | |
22bacca4 DL |
2145 | /* |
2146 | * When we insert an epoll file descriptor, inside another epoll file | |
2147 | * descriptor, there is the change of creating closed loops, which are | |
28d82dc1 JB |
2148 | * better be handled here, than in more critical paths. While we are |
2149 | * checking for loops we also determine the list of files reachable | |
2150 | * and hang them on the tfile_check_list, so we can check that we | |
2151 | * haven't created too many possible wakeup paths. | |
22bacca4 | 2152 | * |
67347fe4 JB |
2153 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
2154 | * the epoll file descriptor is attaching directly to a wakeup source, | |
2155 | * unless the epoll file descriptor is nested. The purpose of taking the | |
2156 | * 'epmutex' on add is to prevent complex toplogies such as loops and | |
2157 | * deep wakeup paths from forming in parallel through multiple | |
2158 | * EPOLL_CTL_ADD operations. | |
22bacca4 | 2159 | */ |
67347fe4 | 2160 | mutex_lock_nested(&ep->mtx, 0); |
28d82dc1 | 2161 | if (op == EPOLL_CTL_ADD) { |
67347fe4 JB |
2162 | if (!list_empty(&f.file->f_ep_links) || |
2163 | is_file_epoll(tf.file)) { | |
2164 | full_check = 1; | |
2165 | mutex_unlock(&ep->mtx); | |
2166 | mutex_lock(&epmutex); | |
2167 | if (is_file_epoll(tf.file)) { | |
2168 | error = -ELOOP; | |
2169 | if (ep_loop_check(ep, tf.file) != 0) { | |
2170 | clear_tfile_check_list(); | |
2171 | goto error_tgt_fput; | |
2172 | } | |
2173 | } else | |
2174 | list_add(&tf.file->f_tfile_llink, | |
2175 | &tfile_check_list); | |
2176 | mutex_lock_nested(&ep->mtx, 0); | |
2177 | if (is_file_epoll(tf.file)) { | |
2178 | tep = tf.file->private_data; | |
2179 | mutex_lock_nested(&tep->mtx, 1); | |
13d51807 | 2180 | } |
67347fe4 JB |
2181 | } |
2182 | } | |
7699acd1 | 2183 | |
67647d0f DL |
2184 | /* |
2185 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
2186 | * above, we can be sure to be able to use the item looked up by | |
2187 | * ep_find() till we release the mutex. | |
2188 | */ | |
7e3fb584 | 2189 | epi = ep_find(ep, tf.file, fd); |
7699acd1 DL |
2190 | |
2191 | error = -EINVAL; | |
2192 | switch (op) { | |
2193 | case EPOLL_CTL_ADD: | |
2194 | if (!epi) { | |
a9a08845 | 2195 | epds.events |= EPOLLERR | EPOLLHUP; |
67347fe4 | 2196 | error = ep_insert(ep, &epds, tf.file, fd, full_check); |
7699acd1 DL |
2197 | } else |
2198 | error = -EEXIST; | |
67347fe4 JB |
2199 | if (full_check) |
2200 | clear_tfile_check_list(); | |
7699acd1 DL |
2201 | break; |
2202 | case EPOLL_CTL_DEL: | |
2203 | if (epi) | |
2204 | error = ep_remove(ep, epi); | |
2205 | else | |
2206 | error = -ENOENT; | |
2207 | break; | |
2208 | case EPOLL_CTL_MOD: | |
2209 | if (epi) { | |
b6a515c8 | 2210 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
a9a08845 | 2211 | epds.events |= EPOLLERR | EPOLLHUP; |
b6a515c8 JB |
2212 | error = ep_modify(ep, epi, &epds); |
2213 | } | |
7699acd1 DL |
2214 | } else |
2215 | error = -ENOENT; | |
2216 | break; | |
2217 | } | |
67347fe4 JB |
2218 | if (tep != NULL) |
2219 | mutex_unlock(&tep->mtx); | |
d47de16c | 2220 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
2221 | |
2222 | error_tgt_fput: | |
67347fe4 | 2223 | if (full_check) |
22bacca4 DL |
2224 | mutex_unlock(&epmutex); |
2225 | ||
7e3fb584 | 2226 | fdput(tf); |
7699acd1 | 2227 | error_fput: |
7e3fb584 | 2228 | fdput(f); |
7699acd1 | 2229 | error_return: |
7699acd1 DL |
2230 | |
2231 | return error; | |
2232 | } | |
2233 | ||
2234 | /* | |
2235 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2236 | * part of the user space epoll_wait(2). | |
2237 | */ | |
791eb22e DB |
2238 | static int do_epoll_wait(int epfd, struct epoll_event __user *events, |
2239 | int maxevents, int timeout) | |
7699acd1 | 2240 | { |
2903ff01 AV |
2241 | int error; |
2242 | struct fd f; | |
7699acd1 DL |
2243 | struct eventpoll *ep; |
2244 | ||
7699acd1 DL |
2245 | /* The maximum number of event must be greater than zero */ |
2246 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
2247 | return -EINVAL; | |
2248 | ||
2249 | /* Verify that the area passed by the user is writeable */ | |
96d4f267 | 2250 | if (!access_ok(events, maxevents * sizeof(struct epoll_event))) |
2903ff01 | 2251 | return -EFAULT; |
7699acd1 DL |
2252 | |
2253 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
2254 | f = fdget(epfd); |
2255 | if (!f.file) | |
2256 | return -EBADF; | |
7699acd1 DL |
2257 | |
2258 | /* | |
2259 | * We have to check that the file structure underneath the fd | |
2260 | * the user passed to us _is_ an eventpoll file. | |
2261 | */ | |
2262 | error = -EINVAL; | |
2903ff01 | 2263 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
2264 | goto error_fput; |
2265 | ||
2266 | /* | |
2267 | * At this point it is safe to assume that the "private_data" contains | |
2268 | * our own data structure. | |
2269 | */ | |
2903ff01 | 2270 | ep = f.file->private_data; |
7699acd1 DL |
2271 | |
2272 | /* Time to fish for events ... */ | |
2273 | error = ep_poll(ep, events, maxevents, timeout); | |
2274 | ||
2275 | error_fput: | |
2903ff01 | 2276 | fdput(f); |
7699acd1 DL |
2277 | return error; |
2278 | } | |
2279 | ||
791eb22e DB |
2280 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
2281 | int, maxevents, int, timeout) | |
2282 | { | |
2283 | return do_epoll_wait(epfd, events, maxevents, timeout); | |
2284 | } | |
2285 | ||
7699acd1 DL |
2286 | /* |
2287 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2288 | * part of the user space epoll_pwait(2). | |
2289 | */ | |
5a8a82b1 HC |
2290 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
2291 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
2292 | size_t, sigsetsize) | |
7699acd1 DL |
2293 | { |
2294 | int error; | |
7699acd1 DL |
2295 | |
2296 | /* | |
2297 | * If the caller wants a certain signal mask to be set during the wait, | |
2298 | * we apply it here. | |
2299 | */ | |
b772434b | 2300 | error = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2301 | if (error) |
2302 | return error; | |
7699acd1 | 2303 | |
791eb22e | 2304 | error = do_epoll_wait(epfd, events, maxevents, timeout); |
b772434b | 2305 | restore_saved_sigmask_unless(error == -EINTR); |
7699acd1 DL |
2306 | |
2307 | return error; | |
2308 | } | |
2309 | ||
35280bd4 AV |
2310 | #ifdef CONFIG_COMPAT |
2311 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, | |
2312 | struct epoll_event __user *, events, | |
2313 | int, maxevents, int, timeout, | |
2314 | const compat_sigset_t __user *, sigmask, | |
2315 | compat_size_t, sigsetsize) | |
2316 | { | |
2317 | long err; | |
35280bd4 AV |
2318 | |
2319 | /* | |
2320 | * If the caller wants a certain signal mask to be set during the wait, | |
2321 | * we apply it here. | |
2322 | */ | |
b772434b | 2323 | err = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2324 | if (err) |
2325 | return err; | |
35280bd4 | 2326 | |
791eb22e | 2327 | err = do_epoll_wait(epfd, events, maxevents, timeout); |
b772434b | 2328 | restore_saved_sigmask_unless(err == -EINTR); |
35280bd4 AV |
2329 | |
2330 | return err; | |
2331 | } | |
2332 | #endif | |
2333 | ||
1da177e4 LT |
2334 | static int __init eventpoll_init(void) |
2335 | { | |
7ef9964e DL |
2336 | struct sysinfo si; |
2337 | ||
2338 | si_meminfo(&si); | |
9df04e1f DL |
2339 | /* |
2340 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
2341 | */ | |
2342 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 2343 | EP_ITEM_COST; |
52bd19f7 | 2344 | BUG_ON(max_user_watches < 0); |
1da177e4 | 2345 | |
22bacca4 DL |
2346 | /* |
2347 | * Initialize the structure used to perform epoll file descriptor | |
2348 | * inclusion loops checks. | |
2349 | */ | |
2350 | ep_nested_calls_init(&poll_loop_ncalls); | |
2351 | ||
39732ca5 EW |
2352 | /* |
2353 | * We can have many thousands of epitems, so prevent this from | |
2354 | * using an extra cache line on 64-bit (and smaller) CPUs | |
2355 | */ | |
2356 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); | |
2357 | ||
1da177e4 LT |
2358 | /* Allocates slab cache used to allocate "struct epitem" items */ |
2359 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
2ae928a9 | 2360 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 LT |
2361 | |
2362 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
2363 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
2ae928a9 | 2364 | sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 | 2365 | |
1da177e4 | 2366 | return 0; |
1da177e4 | 2367 | } |
cea69241 | 2368 | fs_initcall(eventpoll_init); |