]>
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 */ | |
140932e2 | 221 | u64 gen; |
bf3b9f63 SS |
222 | |
223 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
224 | /* used to track busy poll napi_id */ | |
225 | unsigned int napi_id; | |
226 | #endif | |
1da177e4 LT |
227 | }; |
228 | ||
229 | /* Wait structure used by the poll hooks */ | |
230 | struct eppoll_entry { | |
231 | /* List header used to link this structure to the "struct epitem" */ | |
232 | struct list_head llink; | |
233 | ||
234 | /* The "base" pointer is set to the container "struct epitem" */ | |
4f0989db | 235 | struct epitem *base; |
1da177e4 LT |
236 | |
237 | /* | |
238 | * Wait queue item that will be linked to the target file wait | |
239 | * queue head. | |
240 | */ | |
ac6424b9 | 241 | wait_queue_entry_t wait; |
1da177e4 LT |
242 | |
243 | /* The wait queue head that linked the "wait" wait queue item */ | |
244 | wait_queue_head_t *whead; | |
245 | }; | |
246 | ||
1da177e4 LT |
247 | /* Wrapper struct used by poll queueing */ |
248 | struct ep_pqueue { | |
249 | poll_table pt; | |
250 | struct epitem *epi; | |
251 | }; | |
252 | ||
5071f97e DL |
253 | /* Used by the ep_send_events() function as callback private data */ |
254 | struct ep_send_events_data { | |
255 | int maxevents; | |
256 | struct epoll_event __user *events; | |
d7ebbe46 | 257 | int res; |
5071f97e DL |
258 | }; |
259 | ||
7ef9964e DL |
260 | /* |
261 | * Configuration options available inside /proc/sys/fs/epoll/ | |
262 | */ | |
7ef9964e | 263 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 264 | static long max_user_watches __read_mostly; |
7ef9964e | 265 | |
1da177e4 | 266 | /* |
d47de16c | 267 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 268 | */ |
7ef9964e | 269 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 270 | |
140932e2 AV |
271 | static u64 loop_check_gen = 0; |
272 | ||
22bacca4 DL |
273 | /* Used to check for epoll file descriptor inclusion loops */ |
274 | static struct nested_calls poll_loop_ncalls; | |
275 | ||
1da177e4 | 276 | /* Slab cache used to allocate "struct epitem" */ |
e18b890b | 277 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
278 | |
279 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 280 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 281 | |
28d82dc1 JB |
282 | /* |
283 | * List of files with newly added links, where we may need to limit the number | |
284 | * of emanating paths. Protected by the epmutex. | |
285 | */ | |
286 | static LIST_HEAD(tfile_check_list); | |
287 | ||
7ef9964e DL |
288 | #ifdef CONFIG_SYSCTL |
289 | ||
290 | #include <linux/sysctl.h> | |
291 | ||
eec4844f | 292 | static long long_zero; |
52bd19f7 | 293 | static long long_max = LONG_MAX; |
7ef9964e | 294 | |
1f7e0616 | 295 | struct ctl_table epoll_table[] = { |
7ef9964e DL |
296 | { |
297 | .procname = "max_user_watches", | |
298 | .data = &max_user_watches, | |
52bd19f7 | 299 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 300 | .mode = 0644, |
52bd19f7 | 301 | .proc_handler = proc_doulongvec_minmax, |
eec4844f | 302 | .extra1 = &long_zero, |
52bd19f7 | 303 | .extra2 = &long_max, |
7ef9964e | 304 | }, |
ab09203e | 305 | { } |
7ef9964e DL |
306 | }; |
307 | #endif /* CONFIG_SYSCTL */ | |
308 | ||
28d82dc1 JB |
309 | static const struct file_operations eventpoll_fops; |
310 | ||
311 | static inline int is_file_epoll(struct file *f) | |
312 | { | |
313 | return f->f_op == &eventpoll_fops; | |
314 | } | |
b030a4dd | 315 | |
67647d0f | 316 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
317 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
318 | struct file *file, int fd) | |
319 | { | |
320 | ffd->file = file; | |
321 | ffd->fd = fd; | |
322 | } | |
323 | ||
67647d0f | 324 | /* Compare RB tree keys */ |
b030a4dd PE |
325 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
326 | struct epoll_filefd *p2) | |
327 | { | |
328 | return (p1->file > p2->file ? +1: | |
329 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
330 | } | |
331 | ||
b030a4dd | 332 | /* Tells us if the item is currently linked */ |
992991c0 | 333 | static inline int ep_is_linked(struct epitem *epi) |
b030a4dd | 334 | { |
992991c0 | 335 | return !list_empty(&epi->rdllink); |
b030a4dd PE |
336 | } |
337 | ||
ac6424b9 | 338 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
971316f0 ON |
339 | { |
340 | return container_of(p, struct eppoll_entry, wait); | |
341 | } | |
342 | ||
b030a4dd | 343 | /* Get the "struct epitem" from a wait queue pointer */ |
ac6424b9 | 344 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
b030a4dd PE |
345 | { |
346 | return container_of(p, struct eppoll_entry, wait)->base; | |
347 | } | |
348 | ||
349 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 350 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
351 | { |
352 | return container_of(p, struct ep_pqueue, pt)->epi; | |
353 | } | |
354 | ||
355 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
6192bd53 | 356 | static inline int ep_op_has_event(int op) |
b030a4dd | 357 | { |
a80a6b85 | 358 | return op != EPOLL_CTL_DEL; |
b030a4dd PE |
359 | } |
360 | ||
1da177e4 | 361 | /* Initialize the poll safe wake up structure */ |
5071f97e | 362 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 363 | { |
5071f97e DL |
364 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
365 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
366 | } |
367 | ||
3fb0e584 DL |
368 | /** |
369 | * ep_events_available - Checks if ready events might be available. | |
370 | * | |
371 | * @ep: Pointer to the eventpoll context. | |
372 | * | |
373 | * Returns: Returns a value different than zero if ready events are available, | |
374 | * or zero otherwise. | |
375 | */ | |
376 | static inline int ep_events_available(struct eventpoll *ep) | |
377 | { | |
c5a282e9 DB |
378 | return !list_empty_careful(&ep->rdllist) || |
379 | READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR; | |
3fb0e584 DL |
380 | } |
381 | ||
bf3b9f63 SS |
382 | #ifdef CONFIG_NET_RX_BUSY_POLL |
383 | static bool ep_busy_loop_end(void *p, unsigned long start_time) | |
384 | { | |
385 | struct eventpoll *ep = p; | |
386 | ||
387 | return ep_events_available(ep) || busy_loop_timeout(start_time); | |
388 | } | |
bf3b9f63 SS |
389 | |
390 | /* | |
391 | * Busy poll if globally on and supporting sockets found && no events, | |
392 | * busy loop will return if need_resched or ep_events_available. | |
393 | * | |
394 | * we must do our busy polling with irqs enabled | |
395 | */ | |
396 | static void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
397 | { | |
bf3b9f63 SS |
398 | unsigned int napi_id = READ_ONCE(ep->napi_id); |
399 | ||
400 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) | |
401 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep); | |
bf3b9f63 SS |
402 | } |
403 | ||
404 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
405 | { | |
bf3b9f63 SS |
406 | if (ep->napi_id) |
407 | ep->napi_id = 0; | |
bf3b9f63 SS |
408 | } |
409 | ||
410 | /* | |
411 | * Set epoll busy poll NAPI ID from sk. | |
412 | */ | |
413 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
414 | { | |
bf3b9f63 SS |
415 | struct eventpoll *ep; |
416 | unsigned int napi_id; | |
417 | struct socket *sock; | |
418 | struct sock *sk; | |
419 | int err; | |
420 | ||
421 | if (!net_busy_loop_on()) | |
422 | return; | |
423 | ||
424 | sock = sock_from_file(epi->ffd.file, &err); | |
425 | if (!sock) | |
426 | return; | |
427 | ||
428 | sk = sock->sk; | |
429 | if (!sk) | |
430 | return; | |
431 | ||
432 | napi_id = READ_ONCE(sk->sk_napi_id); | |
433 | ep = epi->ep; | |
434 | ||
435 | /* Non-NAPI IDs can be rejected | |
436 | * or | |
437 | * Nothing to do if we already have this ID | |
438 | */ | |
439 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) | |
440 | return; | |
441 | ||
442 | /* record NAPI ID for use in next busy poll */ | |
443 | ep->napi_id = napi_id; | |
bf3b9f63 SS |
444 | } |
445 | ||
514056d5 DB |
446 | #else |
447 | ||
448 | static inline void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
449 | { | |
450 | } | |
451 | ||
452 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
453 | { | |
454 | } | |
455 | ||
456 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
457 | { | |
458 | } | |
459 | ||
460 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
461 | ||
5071f97e DL |
462 | /** |
463 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
464 | * that the recursion limit is not exceeded, and that | |
465 | * the same nested call (by the meaning of same cookie) is | |
466 | * no re-entered. | |
467 | * | |
468 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
5071f97e DL |
469 | * @nproc: Nested call core function pointer. |
470 | * @priv: Opaque data to be passed to the @nproc callback. | |
471 | * @cookie: Cookie to be used to identify this nested call. | |
3fe4a975 | 472 | * @ctx: This instance context. |
5071f97e DL |
473 | * |
474 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
475 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 476 | */ |
74bdc129 | 477 | static int ep_call_nested(struct nested_calls *ncalls, |
5071f97e | 478 | int (*nproc)(void *, void *, int), void *priv, |
3fe4a975 | 479 | void *cookie, void *ctx) |
1da177e4 | 480 | { |
5071f97e | 481 | int error, call_nests = 0; |
1da177e4 | 482 | unsigned long flags; |
5071f97e DL |
483 | struct list_head *lsthead = &ncalls->tasks_call_list; |
484 | struct nested_call_node *tncur; | |
485 | struct nested_call_node tnode; | |
1da177e4 | 486 | |
5071f97e | 487 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 488 | |
5071f97e DL |
489 | /* |
490 | * Try to see if the current task is already inside this wakeup call. | |
491 | * We use a list here, since the population inside this set is always | |
492 | * very much limited. | |
493 | */ | |
b70c3940 | 494 | list_for_each_entry(tncur, lsthead, llink) { |
3fe4a975 | 495 | if (tncur->ctx == ctx && |
74bdc129 | 496 | (tncur->cookie == cookie || ++call_nests > EP_MAX_NESTS)) { |
1da177e4 LT |
497 | /* |
498 | * Ops ... loop detected or maximum nest level reached. | |
499 | * We abort this wake by breaking the cycle itself. | |
500 | */ | |
abff55ce TB |
501 | error = -1; |
502 | goto out_unlock; | |
1da177e4 LT |
503 | } |
504 | } | |
505 | ||
5071f97e | 506 | /* Add the current task and cookie to the list */ |
3fe4a975 | 507 | tnode.ctx = ctx; |
5071f97e | 508 | tnode.cookie = cookie; |
1da177e4 LT |
509 | list_add(&tnode.llink, lsthead); |
510 | ||
5071f97e | 511 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 512 | |
5071f97e DL |
513 | /* Call the nested function */ |
514 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
515 | |
516 | /* Remove the current task from the list */ | |
5071f97e | 517 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 518 | list_del(&tnode.llink); |
3fe4a975 | 519 | out_unlock: |
5071f97e DL |
520 | spin_unlock_irqrestore(&ncalls->lock, flags); |
521 | ||
522 | return error; | |
523 | } | |
524 | ||
02edc6fc SR |
525 | /* |
526 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
527 | * the use of wait queues used by epoll is done in a very controlled | |
528 | * manner. Wake ups can nest inside each other, but are never done | |
529 | * with the same locking. For example: | |
530 | * | |
531 | * dfd = socket(...); | |
532 | * efd1 = epoll_create(); | |
533 | * efd2 = epoll_create(); | |
534 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
535 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
536 | * | |
537 | * When a packet arrives to the device underneath "dfd", the net code will | |
538 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
539 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
540 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
541 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
542 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
543 | * that ends up in another wake_up(), after having checked about the | |
544 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
545 | * avoid stack blasting. | |
546 | * | |
547 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
548 | * this special case of epoll. | |
549 | */ | |
2dfa4eea | 550 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
57a173bd JB |
551 | |
552 | static struct nested_calls poll_safewake_ncalls; | |
553 | ||
554 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) | |
2dfa4eea DL |
555 | { |
556 | unsigned long flags; | |
57a173bd | 557 | wait_queue_head_t *wqueue = (wait_queue_head_t *)cookie; |
2dfa4eea | 558 | |
57a173bd | 559 | spin_lock_irqsave_nested(&wqueue->lock, flags, call_nests + 1); |
a9a08845 | 560 | wake_up_locked_poll(wqueue, EPOLLIN); |
2dfa4eea | 561 | spin_unlock_irqrestore(&wqueue->lock, flags); |
2dfa4eea | 562 | |
5071f97e DL |
563 | return 0; |
564 | } | |
565 | ||
5071f97e DL |
566 | static void ep_poll_safewake(wait_queue_head_t *wq) |
567 | { | |
3fe4a975 DL |
568 | int this_cpu = get_cpu(); |
569 | ||
74bdc129 | 570 | ep_call_nested(&poll_safewake_ncalls, |
3fe4a975 DL |
571 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
572 | ||
573 | put_cpu(); | |
1da177e4 LT |
574 | } |
575 | ||
57a173bd JB |
576 | #else |
577 | ||
578 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
579 | { | |
a9a08845 | 580 | wake_up_poll(wq, EPOLLIN); |
57a173bd JB |
581 | } |
582 | ||
583 | #endif | |
584 | ||
971316f0 ON |
585 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
586 | { | |
587 | wait_queue_head_t *whead; | |
588 | ||
589 | rcu_read_lock(); | |
138e4ad6 ON |
590 | /* |
591 | * If it is cleared by POLLFREE, it should be rcu-safe. | |
592 | * If we read NULL we need a barrier paired with | |
593 | * smp_store_release() in ep_poll_callback(), otherwise | |
594 | * we rely on whead->lock. | |
595 | */ | |
596 | whead = smp_load_acquire(&pwq->whead); | |
971316f0 ON |
597 | if (whead) |
598 | remove_wait_queue(whead, &pwq->wait); | |
599 | rcu_read_unlock(); | |
600 | } | |
601 | ||
1da177e4 | 602 | /* |
d1bc90dd TB |
603 | * This function unregisters poll callbacks from the associated file |
604 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
605 | * ep_free). | |
1da177e4 | 606 | */ |
7699acd1 | 607 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 608 | { |
7699acd1 DL |
609 | struct list_head *lsthead = &epi->pwqlist; |
610 | struct eppoll_entry *pwq; | |
1da177e4 | 611 | |
d1bc90dd TB |
612 | while (!list_empty(lsthead)) { |
613 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 614 | |
d1bc90dd | 615 | list_del(&pwq->llink); |
971316f0 | 616 | ep_remove_wait_queue(pwq); |
d1bc90dd | 617 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 618 | } |
1da177e4 LT |
619 | } |
620 | ||
eea1d585 EW |
621 | /* call only when ep->mtx is held */ |
622 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) | |
623 | { | |
624 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); | |
625 | } | |
626 | ||
627 | /* call only when ep->mtx is held */ | |
628 | static inline void ep_pm_stay_awake(struct epitem *epi) | |
629 | { | |
630 | struct wakeup_source *ws = ep_wakeup_source(epi); | |
631 | ||
632 | if (ws) | |
633 | __pm_stay_awake(ws); | |
634 | } | |
635 | ||
636 | static inline bool ep_has_wakeup_source(struct epitem *epi) | |
637 | { | |
638 | return rcu_access_pointer(epi->ws) ? true : false; | |
639 | } | |
640 | ||
641 | /* call when ep->mtx cannot be held (ep_poll_callback) */ | |
642 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) | |
643 | { | |
644 | struct wakeup_source *ws; | |
645 | ||
646 | rcu_read_lock(); | |
647 | ws = rcu_dereference(epi->ws); | |
648 | if (ws) | |
649 | __pm_stay_awake(ws); | |
650 | rcu_read_unlock(); | |
651 | } | |
652 | ||
5071f97e DL |
653 | /** |
654 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
655 | * the scan code, to call f_op->poll(). Also allows for | |
656 | * O(NumReady) performance. | |
657 | * | |
658 | * @ep: Pointer to the epoll private data structure. | |
659 | * @sproc: Pointer to the scan callback. | |
660 | * @priv: Private opaque data passed to the @sproc callback. | |
d8805e63 | 661 | * @depth: The current depth of recursive f_op->poll calls. |
67347fe4 | 662 | * @ep_locked: caller already holds ep->mtx |
5071f97e DL |
663 | * |
664 | * Returns: The same integer error code returned by the @sproc callback. | |
665 | */ | |
d85e2aa2 AV |
666 | static __poll_t ep_scan_ready_list(struct eventpoll *ep, |
667 | __poll_t (*sproc)(struct eventpoll *, | |
5071f97e | 668 | struct list_head *, void *), |
67347fe4 | 669 | void *priv, int depth, bool ep_locked) |
5071f97e | 670 | { |
d85e2aa2 AV |
671 | __poll_t res; |
672 | int pwake = 0; | |
5071f97e | 673 | struct epitem *epi, *nepi; |
296e236e | 674 | LIST_HEAD(txlist); |
5071f97e | 675 | |
92e64178 DB |
676 | lockdep_assert_irqs_enabled(); |
677 | ||
5071f97e DL |
678 | /* |
679 | * We need to lock this because we could be hit by | |
e057e15f | 680 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e | 681 | */ |
67347fe4 JB |
682 | |
683 | if (!ep_locked) | |
684 | mutex_lock_nested(&ep->mtx, depth); | |
5071f97e DL |
685 | |
686 | /* | |
687 | * Steal the ready list, and re-init the original one to the | |
688 | * empty list. Also, set ep->ovflist to NULL so that events | |
689 | * happening while looping w/out locks, are not lost. We cannot | |
690 | * have the poll callback to queue directly on ep->rdllist, | |
691 | * because we want the "sproc" callback to be able to do it | |
692 | * in a lockless way. | |
693 | */ | |
a218cc49 | 694 | write_lock_irq(&ep->lock); |
296e236e | 695 | list_splice_init(&ep->rdllist, &txlist); |
c5a282e9 | 696 | WRITE_ONCE(ep->ovflist, NULL); |
a218cc49 | 697 | write_unlock_irq(&ep->lock); |
5071f97e DL |
698 | |
699 | /* | |
700 | * Now call the callback function. | |
701 | */ | |
d85e2aa2 | 702 | res = (*sproc)(ep, &txlist, priv); |
5071f97e | 703 | |
a218cc49 | 704 | write_lock_irq(&ep->lock); |
5071f97e DL |
705 | /* |
706 | * During the time we spent inside the "sproc" callback, some | |
707 | * other events might have been queued by the poll callback. | |
708 | * We re-insert them inside the main ready-list here. | |
709 | */ | |
c5a282e9 | 710 | for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL; |
5071f97e DL |
711 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
712 | /* | |
713 | * We need to check if the item is already in the list. | |
714 | * During the "sproc" callback execution time, items are | |
715 | * queued into ->ovflist but the "txlist" might already | |
716 | * contain them, and the list_splice() below takes care of them. | |
717 | */ | |
992991c0 | 718 | if (!ep_is_linked(epi)) { |
c141175d RP |
719 | /* |
720 | * ->ovflist is LIFO, so we have to reverse it in order | |
721 | * to keep in FIFO. | |
722 | */ | |
723 | list_add(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 724 | ep_pm_stay_awake(epi); |
4d7e30d9 | 725 | } |
5071f97e DL |
726 | } |
727 | /* | |
728 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
729 | * releasing the lock, events will be queued in the normal way inside | |
730 | * ep->rdllist. | |
731 | */ | |
c5a282e9 | 732 | WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR); |
5071f97e DL |
733 | |
734 | /* | |
735 | * Quickly re-inject items left on "txlist". | |
736 | */ | |
737 | list_splice(&txlist, &ep->rdllist); | |
4d7e30d9 | 738 | __pm_relax(ep->ws); |
5071f97e DL |
739 | |
740 | if (!list_empty(&ep->rdllist)) { | |
741 | /* | |
296e236e DL |
742 | * Wake up (if active) both the eventpoll wait list and |
743 | * the ->poll() wait list (delayed after we release the lock). | |
5071f97e DL |
744 | */ |
745 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 746 | wake_up(&ep->wq); |
5071f97e DL |
747 | if (waitqueue_active(&ep->poll_wait)) |
748 | pwake++; | |
749 | } | |
a218cc49 | 750 | write_unlock_irq(&ep->lock); |
5071f97e | 751 | |
67347fe4 JB |
752 | if (!ep_locked) |
753 | mutex_unlock(&ep->mtx); | |
5071f97e DL |
754 | |
755 | /* We have to call this outside the lock */ | |
756 | if (pwake) | |
757 | ep_poll_safewake(&ep->poll_wait); | |
758 | ||
d85e2aa2 | 759 | return res; |
5071f97e DL |
760 | } |
761 | ||
ae10b2b4 JB |
762 | static void epi_rcu_free(struct rcu_head *head) |
763 | { | |
764 | struct epitem *epi = container_of(head, struct epitem, rcu); | |
765 | kmem_cache_free(epi_cache, epi); | |
766 | } | |
767 | ||
7699acd1 DL |
768 | /* |
769 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 770 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
771 | */ |
772 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
773 | { | |
7699acd1 | 774 | struct file *file = epi->ffd.file; |
1da177e4 | 775 | |
92e64178 DB |
776 | lockdep_assert_irqs_enabled(); |
777 | ||
1da177e4 | 778 | /* |
ee8ef0a4 | 779 | * Removes poll wait queue hooks. |
1da177e4 | 780 | */ |
7699acd1 | 781 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 782 | |
7699acd1 | 783 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 784 | spin_lock(&file->f_lock); |
ae10b2b4 | 785 | list_del_rcu(&epi->fllink); |
68499914 | 786 | spin_unlock(&file->f_lock); |
1da177e4 | 787 | |
b2ac2ea6 | 788 | rb_erase_cached(&epi->rbn, &ep->rbr); |
1da177e4 | 789 | |
a218cc49 | 790 | write_lock_irq(&ep->lock); |
992991c0 | 791 | if (ep_is_linked(epi)) |
c7ea7630 | 792 | list_del_init(&epi->rdllink); |
a218cc49 | 793 | write_unlock_irq(&ep->lock); |
1da177e4 | 794 | |
eea1d585 | 795 | wakeup_source_unregister(ep_wakeup_source(epi)); |
ae10b2b4 JB |
796 | /* |
797 | * At this point it is safe to free the eventpoll item. Use the union | |
798 | * field epi->rcu, since we are trying to minimize the size of | |
799 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by | |
800 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make | |
801 | * use of the rbn field. | |
802 | */ | |
803 | call_rcu(&epi->rcu, epi_rcu_free); | |
1da177e4 | 804 | |
52bd19f7 | 805 | atomic_long_dec(&ep->user->epoll_watches); |
7ef9964e | 806 | |
c7ea7630 | 807 | return 0; |
1da177e4 LT |
808 | } |
809 | ||
7699acd1 | 810 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 811 | { |
7699acd1 DL |
812 | struct rb_node *rbp; |
813 | struct epitem *epi; | |
1da177e4 | 814 | |
7699acd1 DL |
815 | /* We need to release all tasks waiting for these file */ |
816 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 817 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 818 | |
7699acd1 DL |
819 | /* |
820 | * We need to lock this because we could be hit by | |
821 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 822 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
823 | * is on the way to be removed and no one has references to it |
824 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 825 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
826 | */ |
827 | mutex_lock(&epmutex); | |
1da177e4 LT |
828 | |
829 | /* | |
7699acd1 | 830 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 831 | */ |
b2ac2ea6 | 832 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
7699acd1 DL |
833 | epi = rb_entry(rbp, struct epitem, rbn); |
834 | ||
835 | ep_unregister_pollwait(ep, epi); | |
91cf5ab6 | 836 | cond_resched(); |
7699acd1 | 837 | } |
1da177e4 LT |
838 | |
839 | /* | |
7699acd1 DL |
840 | * Walks through the whole tree by freeing each "struct epitem". At this |
841 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 842 | * holding "epmutex" we can be sure that no file cleanup code will hit |
a218cc49 | 843 | * us during this operation. So we can avoid the lock on "ep->lock". |
ddf676c3 EW |
844 | * We do not need to lock ep->mtx, either, we only do it to prevent |
845 | * a lockdep warning. | |
1da177e4 | 846 | */ |
ddf676c3 | 847 | mutex_lock(&ep->mtx); |
b2ac2ea6 | 848 | while ((rbp = rb_first_cached(&ep->rbr)) != NULL) { |
7699acd1 DL |
849 | epi = rb_entry(rbp, struct epitem, rbn); |
850 | ep_remove(ep, epi); | |
91cf5ab6 | 851 | cond_resched(); |
7699acd1 | 852 | } |
ddf676c3 | 853 | mutex_unlock(&ep->mtx); |
1da177e4 | 854 | |
7699acd1 | 855 | mutex_unlock(&epmutex); |
d47de16c | 856 | mutex_destroy(&ep->mtx); |
7ef9964e | 857 | free_uid(ep->user); |
4d7e30d9 | 858 | wakeup_source_unregister(ep->ws); |
f0ee9aab | 859 | kfree(ep); |
7699acd1 | 860 | } |
1da177e4 | 861 | |
7699acd1 DL |
862 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
863 | { | |
864 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 865 | |
f0ee9aab | 866 | if (ep) |
7699acd1 | 867 | ep_free(ep); |
7699acd1 | 868 | |
7699acd1 | 869 | return 0; |
1da177e4 LT |
870 | } |
871 | ||
d85e2aa2 | 872 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
37b5e521 JB |
873 | void *priv); |
874 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
875 | poll_table *pt); | |
876 | ||
877 | /* | |
878 | * Differs from ep_eventpoll_poll() in that internal callers already have | |
879 | * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested() | |
880 | * is correctly annotated. | |
881 | */ | |
d85e2aa2 | 882 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, |
bec1a502 | 883 | int depth) |
450d89ec | 884 | { |
37b5e521 JB |
885 | struct eventpoll *ep; |
886 | bool locked; | |
887 | ||
450d89ec | 888 | pt->_key = epi->event.events; |
37b5e521 | 889 | if (!is_file_epoll(epi->ffd.file)) |
9965ed17 | 890 | return vfs_poll(epi->ffd.file, pt) & epi->event.events; |
450d89ec | 891 | |
37b5e521 JB |
892 | ep = epi->ffd.file->private_data; |
893 | poll_wait(epi->ffd.file, &ep->poll_wait, pt); | |
894 | locked = pt && (pt->_qproc == ep_ptable_queue_proc); | |
450d89ec | 895 | |
37b5e521 JB |
896 | return ep_scan_ready_list(epi->ffd.file->private_data, |
897 | ep_read_events_proc, &depth, depth, | |
898 | locked) & epi->event.events; | |
450d89ec EW |
899 | } |
900 | ||
d85e2aa2 | 901 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
296e236e | 902 | void *priv) |
5071f97e DL |
903 | { |
904 | struct epitem *epi, *tmp; | |
626cf236 | 905 | poll_table pt; |
37b5e521 | 906 | int depth = *(int *)priv; |
5071f97e | 907 | |
626cf236 | 908 | init_poll_funcptr(&pt, NULL); |
37b5e521 | 909 | depth++; |
450d89ec | 910 | |
5071f97e | 911 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
37b5e521 | 912 | if (ep_item_poll(epi, &pt, depth)) { |
a9a08845 | 913 | return EPOLLIN | EPOLLRDNORM; |
37b5e521 | 914 | } else { |
5071f97e DL |
915 | /* |
916 | * Item has been dropped into the ready list by the poll | |
917 | * callback, but it's not actually ready, as far as | |
918 | * caller requested events goes. We can remove it here. | |
919 | */ | |
eea1d585 | 920 | __pm_relax(ep_wakeup_source(epi)); |
5071f97e | 921 | list_del_init(&epi->rdllink); |
296e236e | 922 | } |
5071f97e DL |
923 | } |
924 | ||
925 | return 0; | |
926 | } | |
927 | ||
a11e1d43 | 928 | static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait) |
11c5ad0e BN |
929 | { |
930 | struct eventpoll *ep = file->private_data; | |
931 | int depth = 0; | |
7699acd1 | 932 | |
a11e1d43 LT |
933 | /* Insert inside our poll wait queue */ |
934 | poll_wait(file, &ep->poll_wait, wait); | |
935 | ||
5071f97e DL |
936 | /* |
937 | * Proceed to find out if wanted events are really available inside | |
37b5e521 | 938 | * the ready list. |
5071f97e | 939 | */ |
37b5e521 JB |
940 | return ep_scan_ready_list(ep, ep_read_events_proc, |
941 | &depth, depth, false); | |
7699acd1 DL |
942 | } |
943 | ||
138d22b5 | 944 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 945 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
138d22b5 CG |
946 | { |
947 | struct eventpoll *ep = f->private_data; | |
948 | struct rb_node *rbp; | |
138d22b5 CG |
949 | |
950 | mutex_lock(&ep->mtx); | |
b2ac2ea6 | 951 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
138d22b5 | 952 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
77493f04 | 953 | struct inode *inode = file_inode(epi->ffd.file); |
138d22b5 | 954 | |
77493f04 CG |
955 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
956 | " pos:%lli ino:%lx sdev:%x\n", | |
a3816ab0 | 957 | epi->ffd.fd, epi->event.events, |
77493f04 CG |
958 | (long long)epi->event.data, |
959 | (long long)epi->ffd.file->f_pos, | |
960 | inode->i_ino, inode->i_sb->s_dev); | |
a3816ab0 | 961 | if (seq_has_overflowed(m)) |
138d22b5 CG |
962 | break; |
963 | } | |
964 | mutex_unlock(&ep->mtx); | |
138d22b5 CG |
965 | } |
966 | #endif | |
967 | ||
7699acd1 DL |
968 | /* File callbacks that implement the eventpoll file behaviour */ |
969 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
970 | #ifdef CONFIG_PROC_FS |
971 | .show_fdinfo = ep_show_fdinfo, | |
972 | #endif | |
7699acd1 | 973 | .release = ep_eventpoll_release, |
a11e1d43 | 974 | .poll = ep_eventpoll_poll, |
6038f373 | 975 | .llseek = noop_llseek, |
7699acd1 DL |
976 | }; |
977 | ||
b611967d | 978 | /* |
7699acd1 DL |
979 | * This is called from eventpoll_release() to unlink files from the eventpoll |
980 | * interface. We need to have this facility to cleanup correctly files that are | |
981 | * closed without being removed from the eventpoll interface. | |
b611967d | 982 | */ |
7699acd1 | 983 | void eventpoll_release_file(struct file *file) |
b611967d | 984 | { |
7699acd1 | 985 | struct eventpoll *ep; |
ebe06187 | 986 | struct epitem *epi, *next; |
b611967d DL |
987 | |
988 | /* | |
68499914 | 989 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 990 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 991 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 992 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 993 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 994 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 995 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 996 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 997 | * from anywhere but ep_free(). |
68499914 JC |
998 | * |
999 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 1000 | */ |
7699acd1 | 1001 | mutex_lock(&epmutex); |
ebe06187 | 1002 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
7699acd1 | 1003 | ep = epi->ep; |
d8805e63 | 1004 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 1005 | ep_remove(ep, epi); |
d47de16c | 1006 | mutex_unlock(&ep->mtx); |
b611967d | 1007 | } |
7699acd1 | 1008 | mutex_unlock(&epmutex); |
b611967d DL |
1009 | } |
1010 | ||
53d2be79 | 1011 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 1012 | { |
7ef9964e DL |
1013 | int error; |
1014 | struct user_struct *user; | |
1015 | struct eventpoll *ep; | |
1da177e4 | 1016 | |
7ef9964e | 1017 | user = get_current_user(); |
7ef9964e DL |
1018 | error = -ENOMEM; |
1019 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
1020 | if (unlikely(!ep)) | |
1021 | goto free_uid; | |
1da177e4 | 1022 | |
d47de16c | 1023 | mutex_init(&ep->mtx); |
a218cc49 | 1024 | rwlock_init(&ep->lock); |
1da177e4 LT |
1025 | init_waitqueue_head(&ep->wq); |
1026 | init_waitqueue_head(&ep->poll_wait); | |
1027 | INIT_LIST_HEAD(&ep->rdllist); | |
b2ac2ea6 | 1028 | ep->rbr = RB_ROOT_CACHED; |
d47de16c | 1029 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 1030 | ep->user = user; |
1da177e4 | 1031 | |
53d2be79 | 1032 | *pep = ep; |
1da177e4 | 1033 | |
1da177e4 | 1034 | return 0; |
7ef9964e DL |
1035 | |
1036 | free_uid: | |
1037 | free_uid(user); | |
1038 | return error; | |
1da177e4 LT |
1039 | } |
1040 | ||
1da177e4 | 1041 | /* |
c7ea7630 DL |
1042 | * Search the file inside the eventpoll tree. The RB tree operations |
1043 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
1044 | * "mtx" held. | |
1da177e4 LT |
1045 | */ |
1046 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
1047 | { | |
1048 | int kcmp; | |
1da177e4 LT |
1049 | struct rb_node *rbp; |
1050 | struct epitem *epi, *epir = NULL; | |
1051 | struct epoll_filefd ffd; | |
1052 | ||
b030a4dd | 1053 | ep_set_ffd(&ffd, file, fd); |
b2ac2ea6 | 1054 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
1da177e4 | 1055 | epi = rb_entry(rbp, struct epitem, rbn); |
b030a4dd | 1056 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
1057 | if (kcmp > 0) |
1058 | rbp = rbp->rb_right; | |
1059 | else if (kcmp < 0) | |
1060 | rbp = rbp->rb_left; | |
1061 | else { | |
1da177e4 LT |
1062 | epir = epi; |
1063 | break; | |
1064 | } | |
1065 | } | |
1da177e4 | 1066 | |
1da177e4 LT |
1067 | return epir; |
1068 | } | |
1069 | ||
92ef6da3 | 1070 | #ifdef CONFIG_CHECKPOINT_RESTORE |
0791e364 CG |
1071 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
1072 | { | |
1073 | struct rb_node *rbp; | |
1074 | struct epitem *epi; | |
1075 | ||
b2ac2ea6 | 1076 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
0791e364 CG |
1077 | epi = rb_entry(rbp, struct epitem, rbn); |
1078 | if (epi->ffd.fd == tfd) { | |
1079 | if (toff == 0) | |
1080 | return epi; | |
1081 | else | |
1082 | toff--; | |
1083 | } | |
1084 | cond_resched(); | |
1085 | } | |
1086 | ||
1087 | return NULL; | |
1088 | } | |
1089 | ||
1090 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, | |
1091 | unsigned long toff) | |
1092 | { | |
1093 | struct file *file_raw; | |
1094 | struct eventpoll *ep; | |
1095 | struct epitem *epi; | |
1096 | ||
1097 | if (!is_file_epoll(file)) | |
1098 | return ERR_PTR(-EINVAL); | |
1099 | ||
1100 | ep = file->private_data; | |
1101 | ||
1102 | mutex_lock(&ep->mtx); | |
1103 | epi = ep_find_tfd(ep, tfd, toff); | |
1104 | if (epi) | |
1105 | file_raw = epi->ffd.file; | |
1106 | else | |
1107 | file_raw = ERR_PTR(-ENOENT); | |
1108 | mutex_unlock(&ep->mtx); | |
1109 | ||
1110 | return file_raw; | |
1111 | } | |
92ef6da3 | 1112 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
0791e364 | 1113 | |
a218cc49 RP |
1114 | /** |
1115 | * Adds a new entry to the tail of the list in a lockless way, i.e. | |
1116 | * multiple CPUs are allowed to call this function concurrently. | |
1117 | * | |
1118 | * Beware: it is necessary to prevent any other modifications of the | |
1119 | * existing list until all changes are completed, in other words | |
1120 | * concurrent list_add_tail_lockless() calls should be protected | |
1121 | * with a read lock, where write lock acts as a barrier which | |
1122 | * makes sure all list_add_tail_lockless() calls are fully | |
1123 | * completed. | |
1124 | * | |
1125 | * Also an element can be locklessly added to the list only in one | |
1126 | * direction i.e. either to the tail either to the head, otherwise | |
1127 | * concurrent access will corrupt the list. | |
1128 | * | |
1129 | * Returns %false if element has been already added to the list, %true | |
1130 | * otherwise. | |
1131 | */ | |
1132 | static inline bool list_add_tail_lockless(struct list_head *new, | |
1133 | struct list_head *head) | |
1134 | { | |
1135 | struct list_head *prev; | |
1136 | ||
1137 | /* | |
1138 | * This is simple 'new->next = head' operation, but cmpxchg() | |
1139 | * is used in order to detect that same element has been just | |
1140 | * added to the list from another CPU: the winner observes | |
1141 | * new->next == new. | |
1142 | */ | |
1143 | if (cmpxchg(&new->next, new, head) != new) | |
1144 | return false; | |
1145 | ||
1146 | /* | |
1147 | * Initially ->next of a new element must be updated with the head | |
1148 | * (we are inserting to the tail) and only then pointers are atomically | |
1149 | * exchanged. XCHG guarantees memory ordering, thus ->next should be | |
1150 | * updated before pointers are actually swapped and pointers are | |
1151 | * swapped before prev->next is updated. | |
1152 | */ | |
1153 | ||
1154 | prev = xchg(&head->prev, new); | |
1155 | ||
1156 | /* | |
1157 | * It is safe to modify prev->next and new->prev, because a new element | |
1158 | * is added only to the tail and new->next is updated before XCHG. | |
1159 | */ | |
1160 | ||
1161 | prev->next = new; | |
1162 | new->prev = prev; | |
1163 | ||
1164 | return true; | |
1165 | } | |
1166 | ||
1167 | /** | |
1168 | * Chains a new epi entry to the tail of the ep->ovflist in a lockless way, | |
1169 | * i.e. multiple CPUs are allowed to call this function concurrently. | |
1170 | * | |
1171 | * Returns %false if epi element has been already chained, %true otherwise. | |
1172 | */ | |
1173 | static inline bool chain_epi_lockless(struct epitem *epi) | |
1174 | { | |
1175 | struct eventpoll *ep = epi->ep; | |
1176 | ||
bc789a86 KK |
1177 | /* Fast preliminary check */ |
1178 | if (epi->next != EP_UNACTIVE_PTR) | |
1179 | return false; | |
1180 | ||
a218cc49 RP |
1181 | /* Check that the same epi has not been just chained from another CPU */ |
1182 | if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR) | |
1183 | return false; | |
1184 | ||
1185 | /* Atomically exchange tail */ | |
1186 | epi->next = xchg(&ep->ovflist, epi); | |
1187 | ||
1188 | return true; | |
1189 | } | |
1190 | ||
1da177e4 | 1191 | /* |
7699acd1 | 1192 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 1193 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 1194 | * have events to report. |
a218cc49 RP |
1195 | * |
1196 | * This callback takes a read lock in order not to content with concurrent | |
1197 | * events from another file descriptors, thus all modifications to ->rdllist | |
1198 | * or ->ovflist are lockless. Read lock is paired with the write lock from | |
1199 | * ep_scan_ready_list(), which stops all list modifications and guarantees | |
1200 | * that lists state is seen correctly. | |
1201 | * | |
1202 | * Another thing worth to mention is that ep_poll_callback() can be called | |
1203 | * concurrently for the same @epi from different CPUs if poll table was inited | |
1204 | * with several wait queues entries. Plural wakeup from different CPUs of a | |
1205 | * single wait queue is serialized by wq.lock, but the case when multiple wait | |
1206 | * queues are used should be detected accordingly. This is detected using | |
1207 | * cmpxchg() operation. | |
1da177e4 | 1208 | */ |
ac6424b9 | 1209 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 1210 | { |
7699acd1 | 1211 | int pwake = 0; |
7699acd1 DL |
1212 | struct epitem *epi = ep_item_from_wait(wait); |
1213 | struct eventpoll *ep = epi->ep; | |
3ad6f93e | 1214 | __poll_t pollflags = key_to_poll(key); |
a218cc49 | 1215 | unsigned long flags; |
df0108c5 | 1216 | int ewake = 0; |
1da177e4 | 1217 | |
a218cc49 | 1218 | read_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1219 | |
bf3b9f63 SS |
1220 | ep_set_busy_poll_napi_id(epi); |
1221 | ||
7699acd1 DL |
1222 | /* |
1223 | * If the event mask does not contain any poll(2) event, we consider the | |
1224 | * descriptor to be disabled. This condition is likely the effect of the | |
1225 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1226 | * until the next EPOLL_CTL_MOD will be issued. | |
1227 | */ | |
1228 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
1229 | goto out_unlock; |
1230 | ||
2dfa4eea DL |
1231 | /* |
1232 | * Check the events coming with the callback. At this stage, not | |
1233 | * every device reports the events in the "key" parameter of the | |
1234 | * callback. We need to be able to handle both cases here, hence the | |
1235 | * test for "key" != NULL before the event match test. | |
1236 | */ | |
3ad6f93e | 1237 | if (pollflags && !(pollflags & epi->event.events)) |
2dfa4eea DL |
1238 | goto out_unlock; |
1239 | ||
d47de16c | 1240 | /* |
bf6a41db | 1241 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 1242 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 1243 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
1244 | * chained in ep->ovflist and requeued later on. |
1245 | */ | |
c5a282e9 | 1246 | if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) { |
bc789a86 KK |
1247 | if (chain_epi_lockless(epi)) |
1248 | ep_pm_stay_awake_rcu(epi); | |
1249 | } else if (!ep_is_linked(epi)) { | |
1250 | /* In the usual case, add event to ready list. */ | |
1251 | if (list_add_tail_lockless(&epi->rdllink, &ep->rdllist)) | |
c3e320b6 | 1252 | ep_pm_stay_awake_rcu(epi); |
4d7e30d9 | 1253 | } |
7699acd1 | 1254 | |
7699acd1 DL |
1255 | /* |
1256 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1257 | * wait list. | |
1258 | */ | |
df0108c5 | 1259 | if (waitqueue_active(&ep->wq)) { |
b6a515c8 | 1260 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
3ad6f93e AV |
1261 | !(pollflags & POLLFREE)) { |
1262 | switch (pollflags & EPOLLINOUT_BITS) { | |
a9a08845 LT |
1263 | case EPOLLIN: |
1264 | if (epi->event.events & EPOLLIN) | |
b6a515c8 JB |
1265 | ewake = 1; |
1266 | break; | |
a9a08845 LT |
1267 | case EPOLLOUT: |
1268 | if (epi->event.events & EPOLLOUT) | |
b6a515c8 JB |
1269 | ewake = 1; |
1270 | break; | |
1271 | case 0: | |
1272 | ewake = 1; | |
1273 | break; | |
1274 | } | |
1275 | } | |
a218cc49 | 1276 | wake_up(&ep->wq); |
df0108c5 | 1277 | } |
7699acd1 DL |
1278 | if (waitqueue_active(&ep->poll_wait)) |
1279 | pwake++; | |
1280 | ||
d47de16c | 1281 | out_unlock: |
a218cc49 | 1282 | read_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1283 | |
7699acd1 DL |
1284 | /* We have to call this outside the lock */ |
1285 | if (pwake) | |
5071f97e | 1286 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 | 1287 | |
138e4ad6 ON |
1288 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
1289 | ewake = 1; | |
1290 | ||
3ad6f93e | 1291 | if (pollflags & POLLFREE) { |
138e4ad6 ON |
1292 | /* |
1293 | * If we race with ep_remove_wait_queue() it can miss | |
1294 | * ->whead = NULL and do another remove_wait_queue() after | |
1295 | * us, so we can't use __remove_wait_queue(). | |
1296 | */ | |
1297 | list_del_init(&wait->entry); | |
1298 | /* | |
1299 | * ->whead != NULL protects us from the race with ep_free() | |
1300 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock | |
1301 | * held by the caller. Once we nullify it, nothing protects | |
1302 | * ep/epi or even wait. | |
1303 | */ | |
1304 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); | |
1305 | } | |
df0108c5 | 1306 | |
138e4ad6 | 1307 | return ewake; |
7699acd1 | 1308 | } |
1da177e4 LT |
1309 | |
1310 | /* | |
1311 | * This is the callback that is used to add our wait queue to the | |
1312 | * target file wakeup lists. | |
1313 | */ | |
1314 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1315 | poll_table *pt) | |
1316 | { | |
b030a4dd | 1317 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
1318 | struct eppoll_entry *pwq; |
1319 | ||
e94b1766 | 1320 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
1321 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
1322 | pwq->whead = whead; | |
1323 | pwq->base = epi; | |
df0108c5 JB |
1324 | if (epi->event.events & EPOLLEXCLUSIVE) |
1325 | add_wait_queue_exclusive(whead, &pwq->wait); | |
1326 | else | |
1327 | add_wait_queue(whead, &pwq->wait); | |
1da177e4 LT |
1328 | list_add_tail(&pwq->llink, &epi->pwqlist); |
1329 | epi->nwait++; | |
296e236e | 1330 | } else { |
1da177e4 LT |
1331 | /* We have to signal that an error occurred */ |
1332 | epi->nwait = -1; | |
296e236e | 1333 | } |
1da177e4 LT |
1334 | } |
1335 | ||
1da177e4 LT |
1336 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1337 | { | |
1338 | int kcmp; | |
b2ac2ea6 | 1339 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
1da177e4 | 1340 | struct epitem *epic; |
b2ac2ea6 | 1341 | bool leftmost = true; |
1da177e4 LT |
1342 | |
1343 | while (*p) { | |
1344 | parent = *p; | |
1345 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1346 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
b2ac2ea6 | 1347 | if (kcmp > 0) { |
1da177e4 | 1348 | p = &parent->rb_right; |
b2ac2ea6 DB |
1349 | leftmost = false; |
1350 | } else | |
1da177e4 LT |
1351 | p = &parent->rb_left; |
1352 | } | |
1353 | rb_link_node(&epi->rbn, parent, p); | |
b2ac2ea6 | 1354 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
1da177e4 LT |
1355 | } |
1356 | ||
a80a6b85 AM |
1357 | |
1358 | ||
28d82dc1 JB |
1359 | #define PATH_ARR_SIZE 5 |
1360 | /* | |
1361 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1362 | * from a single file of interest. For example, we allow 1000 paths of length | |
1363 | * 1, to emanate from each file of interest. This essentially represents the | |
1364 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1365 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1366 | * is connected to n file sources. In this case each file source has 1 path | |
1367 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1368 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1369 | * and delete can't add additional paths. Protected by the epmutex. | |
1370 | */ | |
1371 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1372 | static int path_count[PATH_ARR_SIZE]; | |
1373 | ||
1374 | static int path_count_inc(int nests) | |
1375 | { | |
93dc6107 JB |
1376 | /* Allow an arbitrary number of depth 1 paths */ |
1377 | if (nests == 0) | |
1378 | return 0; | |
1379 | ||
28d82dc1 JB |
1380 | if (++path_count[nests] > path_limits[nests]) |
1381 | return -1; | |
1382 | return 0; | |
1383 | } | |
1384 | ||
1385 | static void path_count_init(void) | |
1386 | { | |
1387 | int i; | |
1388 | ||
1389 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1390 | path_count[i] = 0; | |
1391 | } | |
1392 | ||
1393 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) | |
1394 | { | |
1395 | int error = 0; | |
1396 | struct file *file = priv; | |
1397 | struct file *child_file; | |
1398 | struct epitem *epi; | |
1399 | ||
ae10b2b4 JB |
1400 | /* CTL_DEL can remove links here, but that can't increase our count */ |
1401 | rcu_read_lock(); | |
1402 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { | |
28d82dc1 JB |
1403 | child_file = epi->ep->file; |
1404 | if (is_file_epoll(child_file)) { | |
1405 | if (list_empty(&child_file->f_ep_links)) { | |
1406 | if (path_count_inc(call_nests)) { | |
1407 | error = -1; | |
1408 | break; | |
1409 | } | |
1410 | } else { | |
1411 | error = ep_call_nested(&poll_loop_ncalls, | |
28d82dc1 JB |
1412 | reverse_path_check_proc, |
1413 | child_file, child_file, | |
1414 | current); | |
1415 | } | |
1416 | if (error != 0) | |
1417 | break; | |
1418 | } else { | |
1419 | printk(KERN_ERR "reverse_path_check_proc: " | |
1420 | "file is not an ep!\n"); | |
1421 | } | |
1422 | } | |
ae10b2b4 | 1423 | rcu_read_unlock(); |
28d82dc1 JB |
1424 | return error; |
1425 | } | |
1426 | ||
1427 | /** | |
1428 | * reverse_path_check - The tfile_check_list is list of file *, which have | |
1429 | * links that are proposed to be newly added. We need to | |
1430 | * make sure that those added links don't add too many | |
1431 | * paths such that we will spend all our time waking up | |
1432 | * eventpoll objects. | |
1433 | * | |
1434 | * Returns: Returns zero if the proposed links don't create too many paths, | |
1435 | * -1 otherwise. | |
1436 | */ | |
1437 | static int reverse_path_check(void) | |
1438 | { | |
28d82dc1 JB |
1439 | int error = 0; |
1440 | struct file *current_file; | |
1441 | ||
1442 | /* let's call this for all tfiles */ | |
1443 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { | |
28d82dc1 | 1444 | path_count_init(); |
74bdc129 | 1445 | error = ep_call_nested(&poll_loop_ncalls, |
28d82dc1 JB |
1446 | reverse_path_check_proc, current_file, |
1447 | current_file, current); | |
1448 | if (error) | |
1449 | break; | |
1450 | } | |
1451 | return error; | |
1452 | } | |
1453 | ||
4d7e30d9 AH |
1454 | static int ep_create_wakeup_source(struct epitem *epi) |
1455 | { | |
cfe6d3cf | 1456 | struct name_snapshot n; |
eea1d585 | 1457 | struct wakeup_source *ws; |
4d7e30d9 AH |
1458 | |
1459 | if (!epi->ep->ws) { | |
c8377adf | 1460 | epi->ep->ws = wakeup_source_register(NULL, "eventpoll"); |
4d7e30d9 AH |
1461 | if (!epi->ep->ws) |
1462 | return -ENOMEM; | |
1463 | } | |
1464 | ||
cfe6d3cf AV |
1465 | take_dentry_name_snapshot(&n, epi->ffd.file->f_path.dentry); |
1466 | ws = wakeup_source_register(NULL, n.name.name); | |
1467 | release_dentry_name_snapshot(&n); | |
eea1d585 EW |
1468 | |
1469 | if (!ws) | |
4d7e30d9 | 1470 | return -ENOMEM; |
eea1d585 | 1471 | rcu_assign_pointer(epi->ws, ws); |
4d7e30d9 AH |
1472 | |
1473 | return 0; | |
1474 | } | |
1475 | ||
eea1d585 EW |
1476 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
1477 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) | |
4d7e30d9 | 1478 | { |
eea1d585 EW |
1479 | struct wakeup_source *ws = ep_wakeup_source(epi); |
1480 | ||
d6d67e72 | 1481 | RCU_INIT_POINTER(epi->ws, NULL); |
eea1d585 EW |
1482 | |
1483 | /* | |
1484 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is | |
1485 | * used internally by wakeup_source_remove, too (called by | |
1486 | * wakeup_source_unregister), so we cannot use call_rcu | |
1487 | */ | |
1488 | synchronize_rcu(); | |
1489 | wakeup_source_unregister(ws); | |
4d7e30d9 AH |
1490 | } |
1491 | ||
c7ea7630 DL |
1492 | /* |
1493 | * Must be called with "mtx" held. | |
1494 | */ | |
bec1a502 | 1495 | static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, |
67347fe4 | 1496 | struct file *tfile, int fd, int full_check) |
1da177e4 | 1497 | { |
d85e2aa2 AV |
1498 | int error, pwake = 0; |
1499 | __poll_t revents; | |
52bd19f7 | 1500 | long user_watches; |
1da177e4 LT |
1501 | struct epitem *epi; |
1502 | struct ep_pqueue epq; | |
1503 | ||
92e64178 DB |
1504 | lockdep_assert_irqs_enabled(); |
1505 | ||
52bd19f7 RH |
1506 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
1507 | if (unlikely(user_watches >= max_user_watches)) | |
7ef9964e | 1508 | return -ENOSPC; |
e94b1766 | 1509 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 1510 | return -ENOMEM; |
1da177e4 LT |
1511 | |
1512 | /* Item initialization follow here ... */ | |
1da177e4 LT |
1513 | INIT_LIST_HEAD(&epi->rdllink); |
1514 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
1515 | INIT_LIST_HEAD(&epi->pwqlist); |
1516 | epi->ep = ep; | |
b030a4dd | 1517 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1518 | epi->event = *event; |
1da177e4 | 1519 | epi->nwait = 0; |
d47de16c | 1520 | epi->next = EP_UNACTIVE_PTR; |
4d7e30d9 AH |
1521 | if (epi->event.events & EPOLLWAKEUP) { |
1522 | error = ep_create_wakeup_source(epi); | |
1523 | if (error) | |
1524 | goto error_create_wakeup_source; | |
1525 | } else { | |
eea1d585 | 1526 | RCU_INIT_POINTER(epi->ws, NULL); |
4d7e30d9 | 1527 | } |
1da177e4 | 1528 | |
eba46f9b AV |
1529 | /* Add the current item to the list of active epoll hook for this file */ |
1530 | spin_lock(&tfile->f_lock); | |
1531 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); | |
1532 | spin_unlock(&tfile->f_lock); | |
1533 | ||
1534 | /* | |
1535 | * Add the current item to the RB tree. All RB tree operations are | |
1536 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1537 | */ | |
1538 | ep_rbtree_insert(ep, epi); | |
1539 | ||
1540 | /* now check if we've created too many backpaths */ | |
1541 | error = -EINVAL; | |
1542 | if (full_check && reverse_path_check()) | |
1543 | goto error_remove_epi; | |
1544 | ||
1da177e4 LT |
1545 | /* Initialize the poll table using the queue callback */ |
1546 | epq.epi = epi; | |
1547 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
1548 | ||
1549 | /* | |
1550 | * Attach the item to the poll hooks and get current event bits. | |
1551 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1552 | * been increased by the caller of this function. Note that after |
1553 | * this operation completes, the poll callback can start hitting | |
1554 | * the new item. | |
1da177e4 | 1555 | */ |
37b5e521 | 1556 | revents = ep_item_poll(epi, &epq.pt, 1); |
1da177e4 LT |
1557 | |
1558 | /* | |
1559 | * We have to check if something went wrong during the poll wait queue | |
1560 | * install process. Namely an allocation for a wait queue failed due | |
1561 | * high memory pressure. | |
1562 | */ | |
7ef9964e | 1563 | error = -ENOMEM; |
1da177e4 | 1564 | if (epi->nwait < 0) |
7699acd1 | 1565 | goto error_unregister; |
1da177e4 | 1566 | |
c7ea7630 | 1567 | /* We have to drop the new item inside our item list to keep track of it */ |
a218cc49 | 1568 | write_lock_irq(&ep->lock); |
c7ea7630 | 1569 | |
bf3b9f63 SS |
1570 | /* record NAPI ID of new item if present */ |
1571 | ep_set_busy_poll_napi_id(epi); | |
1572 | ||
1da177e4 | 1573 | /* If the file is already "ready" we drop it inside the ready list */ |
992991c0 | 1574 | if (revents && !ep_is_linked(epi)) { |
1da177e4 | 1575 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1576 | ep_pm_stay_awake(epi); |
1da177e4 LT |
1577 | |
1578 | /* Notify waiting tasks that events are available */ | |
1579 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1580 | wake_up(&ep->wq); |
1da177e4 LT |
1581 | if (waitqueue_active(&ep->poll_wait)) |
1582 | pwake++; | |
1583 | } | |
1584 | ||
a218cc49 | 1585 | write_unlock_irq(&ep->lock); |
1da177e4 | 1586 | |
52bd19f7 | 1587 | atomic_long_inc(&ep->user->epoll_watches); |
7ef9964e | 1588 | |
1da177e4 LT |
1589 | /* We have to call this outside the lock */ |
1590 | if (pwake) | |
5071f97e | 1591 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1592 | |
1da177e4 LT |
1593 | return 0; |
1594 | ||
eba46f9b AV |
1595 | error_unregister: |
1596 | ep_unregister_pollwait(ep, epi); | |
28d82dc1 JB |
1597 | error_remove_epi: |
1598 | spin_lock(&tfile->f_lock); | |
ae10b2b4 | 1599 | list_del_rcu(&epi->fllink); |
28d82dc1 JB |
1600 | spin_unlock(&tfile->f_lock); |
1601 | ||
b2ac2ea6 | 1602 | rb_erase_cached(&epi->rbn, &ep->rbr); |
28d82dc1 | 1603 | |
1da177e4 LT |
1604 | /* |
1605 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
1606 | * allocated wait queue. Note that we don't care about the ep->ovflist |
1607 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
1608 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 1609 | */ |
a218cc49 | 1610 | write_lock_irq(&ep->lock); |
992991c0 | 1611 | if (ep_is_linked(epi)) |
6192bd53 | 1612 | list_del_init(&epi->rdllink); |
a218cc49 | 1613 | write_unlock_irq(&ep->lock); |
1da177e4 | 1614 | |
eea1d585 | 1615 | wakeup_source_unregister(ep_wakeup_source(epi)); |
4d7e30d9 AH |
1616 | |
1617 | error_create_wakeup_source: | |
b030a4dd | 1618 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1619 | |
1da177e4 LT |
1620 | return error; |
1621 | } | |
1622 | ||
1da177e4 LT |
1623 | /* |
1624 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1625 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 | 1626 | */ |
bec1a502 AV |
1627 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, |
1628 | const struct epoll_event *event) | |
1da177e4 LT |
1629 | { |
1630 | int pwake = 0; | |
626cf236 HV |
1631 | poll_table pt; |
1632 | ||
92e64178 DB |
1633 | lockdep_assert_irqs_enabled(); |
1634 | ||
626cf236 | 1635 | init_poll_funcptr(&pt, NULL); |
1da177e4 LT |
1636 | |
1637 | /* | |
e057e15f TB |
1638 | * Set the new event interest mask before calling f_op->poll(); |
1639 | * otherwise we might miss an event that happens between the | |
1640 | * f_op->poll() call and the new event set registering. | |
1da177e4 | 1641 | */ |
128dd175 | 1642 | epi->event.events = event->events; /* need barrier below */ |
e057e15f | 1643 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 | 1644 | if (epi->event.events & EPOLLWAKEUP) { |
eea1d585 | 1645 | if (!ep_has_wakeup_source(epi)) |
4d7e30d9 | 1646 | ep_create_wakeup_source(epi); |
eea1d585 | 1647 | } else if (ep_has_wakeup_source(epi)) { |
4d7e30d9 AH |
1648 | ep_destroy_wakeup_source(epi); |
1649 | } | |
1da177e4 | 1650 | |
128dd175 EW |
1651 | /* |
1652 | * The following barrier has two effects: | |
1653 | * | |
1654 | * 1) Flush epi changes above to other CPUs. This ensures | |
1655 | * we do not miss events from ep_poll_callback if an | |
1656 | * event occurs immediately after we call f_op->poll(). | |
a218cc49 | 1657 | * We need this because we did not take ep->lock while |
128dd175 | 1658 | * changing epi above (but ep_poll_callback does take |
a218cc49 | 1659 | * ep->lock). |
128dd175 EW |
1660 | * |
1661 | * 2) We also need to ensure we do not miss _past_ events | |
1662 | * when calling f_op->poll(). This barrier also | |
1663 | * pairs with the barrier in wq_has_sleeper (see | |
1664 | * comments for wq_has_sleeper). | |
1665 | * | |
1666 | * This barrier will now guarantee ep_poll_callback or f_op->poll | |
1667 | * (or both) will notice the readiness of an item. | |
1668 | */ | |
1669 | smp_mb(); | |
1670 | ||
1da177e4 LT |
1671 | /* |
1672 | * Get current event bits. We can safely use the file* here because | |
1673 | * its usage count has been increased by the caller of this function. | |
c7ea7630 | 1674 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1675 | * list, push it inside. |
1da177e4 | 1676 | */ |
69112736 | 1677 | if (ep_item_poll(epi, &pt, 1)) { |
a218cc49 | 1678 | write_lock_irq(&ep->lock); |
992991c0 | 1679 | if (!ep_is_linked(epi)) { |
c7ea7630 | 1680 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1681 | ep_pm_stay_awake(epi); |
c7ea7630 DL |
1682 | |
1683 | /* Notify waiting tasks that events are available */ | |
1684 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1685 | wake_up(&ep->wq); |
c7ea7630 DL |
1686 | if (waitqueue_active(&ep->poll_wait)) |
1687 | pwake++; | |
7699acd1 | 1688 | } |
a218cc49 | 1689 | write_unlock_irq(&ep->lock); |
7699acd1 | 1690 | } |
1da177e4 | 1691 | |
7699acd1 DL |
1692 | /* We have to call this outside the lock */ |
1693 | if (pwake) | |
5071f97e | 1694 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1695 | |
7699acd1 | 1696 | return 0; |
1da177e4 LT |
1697 | } |
1698 | ||
d85e2aa2 | 1699 | static __poll_t ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
296e236e | 1700 | void *priv) |
1da177e4 | 1701 | { |
5071f97e | 1702 | struct ep_send_events_data *esed = priv; |
d85e2aa2 | 1703 | __poll_t revents; |
4e0982a0 DB |
1704 | struct epitem *epi, *tmp; |
1705 | struct epoll_event __user *uevent = esed->events; | |
eea1d585 | 1706 | struct wakeup_source *ws; |
626cf236 HV |
1707 | poll_table pt; |
1708 | ||
1709 | init_poll_funcptr(&pt, NULL); | |
4e0982a0 | 1710 | esed->res = 0; |
1da177e4 | 1711 | |
296e236e | 1712 | /* |
5071f97e DL |
1713 | * We can loop without lock because we are passed a task private list. |
1714 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1715 | * holding "mtx" during this call. | |
296e236e | 1716 | */ |
21877e1a DB |
1717 | lockdep_assert_held(&ep->mtx); |
1718 | ||
4e0982a0 DB |
1719 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
1720 | if (esed->res >= esed->maxevents) | |
1721 | break; | |
d47de16c | 1722 | |
4d7e30d9 AH |
1723 | /* |
1724 | * Activate ep->ws before deactivating epi->ws to prevent | |
1725 | * triggering auto-suspend here (in case we reactive epi->ws | |
1726 | * below). | |
1727 | * | |
1728 | * This could be rearranged to delay the deactivation of epi->ws | |
1729 | * instead, but then epi->ws would temporarily be out of sync | |
1730 | * with ep_is_linked(). | |
1731 | */ | |
eea1d585 EW |
1732 | ws = ep_wakeup_source(epi); |
1733 | if (ws) { | |
1734 | if (ws->active) | |
1735 | __pm_stay_awake(ep->ws); | |
1736 | __pm_relax(ws); | |
1737 | } | |
1738 | ||
d47de16c | 1739 | list_del_init(&epi->rdllink); |
1da177e4 | 1740 | |
296e236e | 1741 | /* |
5071f97e DL |
1742 | * If the event mask intersect the caller-requested one, |
1743 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
4e0982a0 | 1744 | * is holding ep->mtx, so no operations coming from userspace |
5071f97e | 1745 | * can change the item. |
296e236e | 1746 | */ |
4e0982a0 DB |
1747 | revents = ep_item_poll(epi, &pt, 1); |
1748 | if (!revents) | |
1749 | continue; | |
1750 | ||
1751 | if (__put_user(revents, &uevent->events) || | |
1752 | __put_user(epi->event.data, &uevent->data)) { | |
1753 | list_add(&epi->rdllink, head); | |
1754 | ep_pm_stay_awake(epi); | |
1755 | if (!esed->res) | |
1756 | esed->res = -EFAULT; | |
1757 | return 0; | |
1758 | } | |
1759 | esed->res++; | |
1760 | uevent++; | |
1761 | if (epi->event.events & EPOLLONESHOT) | |
1762 | epi->event.events &= EP_PRIVATE_BITS; | |
1763 | else if (!(epi->event.events & EPOLLET)) { | |
1764 | /* | |
1765 | * If this file has been added with Level | |
1766 | * Trigger mode, we need to insert back inside | |
1767 | * the ready list, so that the next call to | |
1768 | * epoll_wait() will check again the events | |
1769 | * availability. At this point, no one can insert | |
1770 | * into ep->rdllist besides us. The epoll_ctl() | |
1771 | * callers are locked out by | |
1772 | * ep_scan_ready_list() holding "mtx" and the | |
1773 | * poll callback will queue them in ep->ovflist. | |
1774 | */ | |
1775 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1776 | ep_pm_stay_awake(epi); | |
296e236e DL |
1777 | } |
1778 | } | |
5071f97e | 1779 | |
d7ebbe46 | 1780 | return 0; |
5071f97e | 1781 | } |
d47de16c | 1782 | |
296e236e DL |
1783 | static int ep_send_events(struct eventpoll *ep, |
1784 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1785 | { |
1786 | struct ep_send_events_data esed; | |
1da177e4 | 1787 | |
5071f97e DL |
1788 | esed.maxevents = maxevents; |
1789 | esed.events = events; | |
6192bd53 | 1790 | |
d7ebbe46 AV |
1791 | ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false); |
1792 | return esed.res; | |
1da177e4 LT |
1793 | } |
1794 | ||
766b9f92 | 1795 | static inline struct timespec64 ep_set_mstimeout(long ms) |
0781b909 | 1796 | { |
766b9f92 | 1797 | struct timespec64 now, ts = { |
0781b909 ED |
1798 | .tv_sec = ms / MSEC_PER_SEC, |
1799 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), | |
1800 | }; | |
1801 | ||
766b9f92 DD |
1802 | ktime_get_ts64(&now); |
1803 | return timespec64_add_safe(now, ts); | |
0781b909 ED |
1804 | } |
1805 | ||
f4d93ad7 SB |
1806 | /** |
1807 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied | |
1808 | * event buffer. | |
1809 | * | |
1810 | * @ep: Pointer to the eventpoll context. | |
1811 | * @events: Pointer to the userspace buffer where the ready events should be | |
1812 | * stored. | |
1813 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1814 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
1815 | * milliseconds. If the @timeout is zero, the function will not block, | |
1816 | * while if the @timeout is less than zero, the function will block | |
1817 | * until at least one event has been retrieved (or an error | |
1818 | * occurred). | |
1819 | * | |
1820 | * Returns: Returns the number of ready events which have been fetched, or an | |
1821 | * error code, in case of error. | |
1822 | */ | |
1da177e4 LT |
1823 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1824 | int maxevents, long timeout) | |
1825 | { | |
f4d93ad7 | 1826 | int res = 0, eavail, timed_out = 0; |
da8b44d5 | 1827 | u64 slack = 0; |
ac6424b9 | 1828 | wait_queue_entry_t wait; |
95aac7b1 SB |
1829 | ktime_t expires, *to = NULL; |
1830 | ||
679abf38 DB |
1831 | lockdep_assert_irqs_enabled(); |
1832 | ||
95aac7b1 | 1833 | if (timeout > 0) { |
766b9f92 | 1834 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
0781b909 | 1835 | |
95aac7b1 SB |
1836 | slack = select_estimate_accuracy(&end_time); |
1837 | to = &expires; | |
766b9f92 | 1838 | *to = timespec64_to_ktime(end_time); |
95aac7b1 | 1839 | } else if (timeout == 0) { |
f4d93ad7 SB |
1840 | /* |
1841 | * Avoid the unnecessary trip to the wait queue loop, if the | |
c5a282e9 DB |
1842 | * caller specified a non blocking operation. We still need |
1843 | * lock because we could race and not see an epi being added | |
1844 | * to the ready list while in irq callback. Thus incorrectly | |
1845 | * returning 0 back to userspace. | |
f4d93ad7 | 1846 | */ |
95aac7b1 | 1847 | timed_out = 1; |
c5a282e9 | 1848 | |
a218cc49 | 1849 | write_lock_irq(&ep->lock); |
c5a282e9 | 1850 | eavail = ep_events_available(ep); |
a218cc49 | 1851 | write_unlock_irq(&ep->lock); |
c5a282e9 | 1852 | |
35cff1a6 | 1853 | goto send_events; |
95aac7b1 | 1854 | } |
1da177e4 | 1855 | |
f4d93ad7 | 1856 | fetch_events: |
bf3b9f63 SS |
1857 | |
1858 | if (!ep_events_available(ep)) | |
1859 | ep_busy_loop(ep, timed_out); | |
1860 | ||
c5a282e9 DB |
1861 | eavail = ep_events_available(ep); |
1862 | if (eavail) | |
35cff1a6 | 1863 | goto send_events; |
1da177e4 | 1864 | |
c5a282e9 DB |
1865 | /* |
1866 | * Busy poll timed out. Drop NAPI ID for now, we can add | |
1867 | * it back in when we have moved a socket with a valid NAPI | |
1868 | * ID onto the ready list. | |
1869 | */ | |
1870 | ep_reset_busy_poll_napi_id(ep); | |
bf3b9f63 | 1871 | |
50da50ad RP |
1872 | do { |
1873 | /* | |
1874 | * Internally init_wait() uses autoremove_wake_function(), | |
1875 | * thus wait entry is removed from the wait queue on each | |
1876 | * wakeup. Why it is important? In case of several waiters | |
1877 | * each new wakeup will hit the next waiter, giving it the | |
1878 | * chance to harvest new event. Otherwise wakeup can be | |
1879 | * lost. This is also good performance-wise, because on | |
1880 | * normal wakeup path no need to call __remove_wait_queue() | |
1881 | * explicitly, thus ep->lock is not taken, which halts the | |
1882 | * event delivery. | |
1883 | */ | |
1884 | init_wait(&wait); | |
f8090469 | 1885 | write_lock_irq(&ep->lock); |
86c05179 | 1886 | __add_wait_queue_exclusive(&ep->wq, &wait); |
f8090469 | 1887 | write_unlock_irq(&ep->lock); |
1da177e4 | 1888 | |
bf3b9f63 | 1889 | /* |
c5a282e9 DB |
1890 | * We don't want to sleep if the ep_poll_callback() sends us |
1891 | * a wakeup in between. That's why we set the task state | |
1892 | * to TASK_INTERRUPTIBLE before doing the checks. | |
bf3b9f63 | 1893 | */ |
c5a282e9 | 1894 | set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 | 1895 | /* |
c5a282e9 DB |
1896 | * Always short-circuit for fatal signals to allow |
1897 | * threads to make a timely exit without the chance of | |
1898 | * finding more events available and fetching | |
1899 | * repeatedly. | |
1da177e4 | 1900 | */ |
c5a282e9 DB |
1901 | if (fatal_signal_pending(current)) { |
1902 | res = -EINTR; | |
1903 | break; | |
1904 | } | |
95aac7b1 | 1905 | |
abc610e0 DB |
1906 | eavail = ep_events_available(ep); |
1907 | if (eavail) | |
c5a282e9 DB |
1908 | break; |
1909 | if (signal_pending(current)) { | |
1910 | res = -EINTR; | |
1911 | break; | |
1da177e4 | 1912 | } |
1da177e4 | 1913 | |
abc610e0 | 1914 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) { |
c5a282e9 | 1915 | timed_out = 1; |
abc610e0 DB |
1916 | break; |
1917 | } | |
50da50ad RP |
1918 | |
1919 | /* We were woken up, thus go and try to harvest some events */ | |
1920 | eavail = 1; | |
1921 | ||
1922 | } while (0); | |
1da177e4 | 1923 | |
c5a282e9 | 1924 | __set_current_state(TASK_RUNNING); |
1da177e4 | 1925 | |
50da50ad RP |
1926 | if (!list_empty_careful(&wait.entry)) { |
1927 | write_lock_irq(&ep->lock); | |
1928 | __remove_wait_queue(&ep->wq, &wait); | |
1929 | write_unlock_irq(&ep->lock); | |
1930 | } | |
1931 | ||
35cff1a6 | 1932 | send_events: |
1da177e4 LT |
1933 | /* |
1934 | * Try to transfer events to user space. In case we get 0 events and | |
1935 | * there's still timeout left over, we go trying again in search of | |
1936 | * more luck. | |
1937 | */ | |
1938 | if (!res && eavail && | |
95aac7b1 | 1939 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
f4d93ad7 | 1940 | goto fetch_events; |
1da177e4 LT |
1941 | |
1942 | return res; | |
1943 | } | |
1944 | ||
22bacca4 DL |
1945 | /** |
1946 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() | |
1947 | * API, to verify that adding an epoll file inside another | |
1948 | * epoll structure, does not violate the constraints, in | |
1949 | * terms of closed loops, or too deep chains (which can | |
1950 | * result in excessive stack usage). | |
1951 | * | |
1952 | * @priv: Pointer to the epoll file to be currently checked. | |
1953 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll | |
1954 | * data structure pointer. | |
1955 | * @call_nests: Current dept of the @ep_call_nested() call stack. | |
1956 | * | |
1957 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1958 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1959 | */ | |
1960 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) | |
1961 | { | |
1962 | int error = 0; | |
1963 | struct file *file = priv; | |
1964 | struct eventpoll *ep = file->private_data; | |
28d82dc1 | 1965 | struct eventpoll *ep_tovisit; |
22bacca4 DL |
1966 | struct rb_node *rbp; |
1967 | struct epitem *epi; | |
1968 | ||
d8805e63 | 1969 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
140932e2 | 1970 | ep->gen = loop_check_gen; |
b2ac2ea6 | 1971 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
22bacca4 DL |
1972 | epi = rb_entry(rbp, struct epitem, rbn); |
1973 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
28d82dc1 | 1974 | ep_tovisit = epi->ffd.file->private_data; |
140932e2 | 1975 | if (ep_tovisit->gen == loop_check_gen) |
28d82dc1 | 1976 | continue; |
74bdc129 | 1977 | error = ep_call_nested(&poll_loop_ncalls, |
28d82dc1 JB |
1978 | ep_loop_check_proc, epi->ffd.file, |
1979 | ep_tovisit, current); | |
22bacca4 DL |
1980 | if (error != 0) |
1981 | break; | |
28d82dc1 JB |
1982 | } else { |
1983 | /* | |
1984 | * If we've reached a file that is not associated with | |
1985 | * an ep, then we need to check if the newly added | |
1986 | * links are going to add too many wakeup paths. We do | |
1987 | * this by adding it to the tfile_check_list, if it's | |
1988 | * not already there, and calling reverse_path_check() | |
1989 | * during ep_insert(). | |
1990 | */ | |
24c0b8ea | 1991 | if (list_empty(&epi->ffd.file->f_tfile_llink)) { |
d6aa2f23 AV |
1992 | if (get_file_rcu(epi->ffd.file)) |
1993 | list_add(&epi->ffd.file->f_tfile_llink, | |
1994 | &tfile_check_list); | |
24c0b8ea | 1995 | } |
22bacca4 DL |
1996 | } |
1997 | } | |
1998 | mutex_unlock(&ep->mtx); | |
1999 | ||
2000 | return error; | |
2001 | } | |
2002 | ||
2003 | /** | |
2004 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) | |
2005 | * another epoll file (represented by @ep) does not create | |
2006 | * closed loops or too deep chains. | |
2007 | * | |
2008 | * @ep: Pointer to the epoll private data structure. | |
2009 | * @file: Pointer to the epoll file to be checked. | |
2010 | * | |
2011 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
2012 | * structure @ep does not violate the constraints, or -1 otherwise. | |
2013 | */ | |
2014 | static int ep_loop_check(struct eventpoll *ep, struct file *file) | |
2015 | { | |
140932e2 | 2016 | return ep_call_nested(&poll_loop_ncalls, |
22bacca4 | 2017 | ep_loop_check_proc, file, ep, current); |
28d82dc1 JB |
2018 | } |
2019 | ||
2020 | static void clear_tfile_check_list(void) | |
2021 | { | |
2022 | struct file *file; | |
2023 | ||
2024 | /* first clear the tfile_check_list */ | |
2025 | while (!list_empty(&tfile_check_list)) { | |
2026 | file = list_first_entry(&tfile_check_list, struct file, | |
2027 | f_tfile_llink); | |
2028 | list_del_init(&file->f_tfile_llink); | |
24c0b8ea | 2029 | fput(file); |
28d82dc1 JB |
2030 | } |
2031 | INIT_LIST_HEAD(&tfile_check_list); | |
22bacca4 DL |
2032 | } |
2033 | ||
7699acd1 | 2034 | /* |
523723bb | 2035 | * Open an eventpoll file descriptor. |
7699acd1 | 2036 | */ |
791eb22e | 2037 | static int do_epoll_create(int flags) |
7699acd1 | 2038 | { |
28d82dc1 | 2039 | int error, fd; |
bb57c3ed | 2040 | struct eventpoll *ep = NULL; |
28d82dc1 | 2041 | struct file *file; |
7699acd1 | 2042 | |
e38b36f3 UD |
2043 | /* Check the EPOLL_* constant for consistency. */ |
2044 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
2045 | ||
296e236e DL |
2046 | if (flags & ~EPOLL_CLOEXEC) |
2047 | return -EINVAL; | |
7699acd1 | 2048 | /* |
bb57c3ed | 2049 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 2050 | */ |
9fe5ad9c | 2051 | error = ep_alloc(&ep); |
bb57c3ed DL |
2052 | if (error < 0) |
2053 | return error; | |
7699acd1 DL |
2054 | /* |
2055 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 2056 | * a file structure and a free file descriptor. |
7699acd1 | 2057 | */ |
28d82dc1 JB |
2058 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
2059 | if (fd < 0) { | |
2060 | error = fd; | |
2061 | goto out_free_ep; | |
2062 | } | |
2063 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 2064 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
2065 | if (IS_ERR(file)) { |
2066 | error = PTR_ERR(file); | |
2067 | goto out_free_fd; | |
2068 | } | |
28d82dc1 | 2069 | ep->file = file; |
98022748 | 2070 | fd_install(fd, file); |
28d82dc1 JB |
2071 | return fd; |
2072 | ||
2073 | out_free_fd: | |
2074 | put_unused_fd(fd); | |
2075 | out_free_ep: | |
2076 | ep_free(ep); | |
bb57c3ed | 2077 | return error; |
7699acd1 DL |
2078 | } |
2079 | ||
791eb22e DB |
2080 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
2081 | { | |
2082 | return do_epoll_create(flags); | |
2083 | } | |
2084 | ||
5a8a82b1 | 2085 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 2086 | { |
bfe3891a | 2087 | if (size <= 0) |
9fe5ad9c UD |
2088 | return -EINVAL; |
2089 | ||
791eb22e | 2090 | return do_epoll_create(0); |
a0998b50 UD |
2091 | } |
2092 | ||
7699acd1 DL |
2093 | /* |
2094 | * The following function implements the controller interface for | |
2095 | * the eventpoll file that enables the insertion/removal/change of | |
67647d0f | 2096 | * file descriptors inside the interest set. |
7699acd1 | 2097 | */ |
5a8a82b1 HC |
2098 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
2099 | struct epoll_event __user *, event) | |
7699acd1 DL |
2100 | { |
2101 | int error; | |
67347fe4 | 2102 | int full_check = 0; |
7e3fb584 | 2103 | struct fd f, tf; |
7699acd1 DL |
2104 | struct eventpoll *ep; |
2105 | struct epitem *epi; | |
2106 | struct epoll_event epds; | |
67347fe4 | 2107 | struct eventpoll *tep = NULL; |
7699acd1 | 2108 | |
7699acd1 DL |
2109 | error = -EFAULT; |
2110 | if (ep_op_has_event(op) && | |
2111 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
2112 | goto error_return; | |
2113 | ||
7699acd1 | 2114 | error = -EBADF; |
7e3fb584 AV |
2115 | f = fdget(epfd); |
2116 | if (!f.file) | |
7699acd1 DL |
2117 | goto error_return; |
2118 | ||
2119 | /* Get the "struct file *" for the target file */ | |
7e3fb584 AV |
2120 | tf = fdget(fd); |
2121 | if (!tf.file) | |
7699acd1 DL |
2122 | goto error_fput; |
2123 | ||
2124 | /* The target file descriptor must support poll */ | |
2125 | error = -EPERM; | |
9965ed17 | 2126 | if (!file_can_poll(tf.file)) |
7699acd1 DL |
2127 | goto error_tgt_fput; |
2128 | ||
4d7e30d9 | 2129 | /* Check if EPOLLWAKEUP is allowed */ |
c680e41b NI |
2130 | if (ep_op_has_event(op)) |
2131 | ep_take_care_of_epollwakeup(&epds); | |
4d7e30d9 | 2132 | |
7699acd1 DL |
2133 | /* |
2134 | * We have to check that the file structure underneath the file descriptor | |
2135 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
2136 | * adding an epoll file descriptor inside itself. | |
2137 | */ | |
2138 | error = -EINVAL; | |
7e3fb584 | 2139 | if (f.file == tf.file || !is_file_epoll(f.file)) |
7699acd1 DL |
2140 | goto error_tgt_fput; |
2141 | ||
df0108c5 JB |
2142 | /* |
2143 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, | |
2144 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. | |
2145 | * Also, we do not currently supported nested exclusive wakeups. | |
2146 | */ | |
c857ab64 | 2147 | if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) { |
b6a515c8 JB |
2148 | if (op == EPOLL_CTL_MOD) |
2149 | goto error_tgt_fput; | |
2150 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || | |
2151 | (epds.events & ~EPOLLEXCLUSIVE_OK_BITS))) | |
2152 | goto error_tgt_fput; | |
2153 | } | |
df0108c5 | 2154 | |
7699acd1 DL |
2155 | /* |
2156 | * At this point it is safe to assume that the "private_data" contains | |
2157 | * our own data structure. | |
2158 | */ | |
7e3fb584 | 2159 | ep = f.file->private_data; |
7699acd1 | 2160 | |
22bacca4 DL |
2161 | /* |
2162 | * When we insert an epoll file descriptor, inside another epoll file | |
2163 | * descriptor, there is the change of creating closed loops, which are | |
28d82dc1 JB |
2164 | * better be handled here, than in more critical paths. While we are |
2165 | * checking for loops we also determine the list of files reachable | |
2166 | * and hang them on the tfile_check_list, so we can check that we | |
2167 | * haven't created too many possible wakeup paths. | |
22bacca4 | 2168 | * |
67347fe4 JB |
2169 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
2170 | * the epoll file descriptor is attaching directly to a wakeup source, | |
2171 | * unless the epoll file descriptor is nested. The purpose of taking the | |
2172 | * 'epmutex' on add is to prevent complex toplogies such as loops and | |
2173 | * deep wakeup paths from forming in parallel through multiple | |
2174 | * EPOLL_CTL_ADD operations. | |
22bacca4 | 2175 | */ |
67347fe4 | 2176 | mutex_lock_nested(&ep->mtx, 0); |
28d82dc1 | 2177 | if (op == EPOLL_CTL_ADD) { |
67347fe4 | 2178 | if (!list_empty(&f.file->f_ep_links) || |
4afc22d6 | 2179 | ep->gen == loop_check_gen || |
67347fe4 JB |
2180 | is_file_epoll(tf.file)) { |
2181 | full_check = 1; | |
2182 | mutex_unlock(&ep->mtx); | |
2183 | mutex_lock(&epmutex); | |
2184 | if (is_file_epoll(tf.file)) { | |
2185 | error = -ELOOP; | |
9dbadc3c | 2186 | if (ep_loop_check(ep, tf.file) != 0) |
67347fe4 | 2187 | goto error_tgt_fput; |
24c0b8ea MZ |
2188 | } else { |
2189 | get_file(tf.file); | |
67347fe4 JB |
2190 | list_add(&tf.file->f_tfile_llink, |
2191 | &tfile_check_list); | |
24c0b8ea | 2192 | } |
67347fe4 JB |
2193 | mutex_lock_nested(&ep->mtx, 0); |
2194 | if (is_file_epoll(tf.file)) { | |
2195 | tep = tf.file->private_data; | |
2196 | mutex_lock_nested(&tep->mtx, 1); | |
13d51807 | 2197 | } |
67347fe4 JB |
2198 | } |
2199 | } | |
7699acd1 | 2200 | |
67647d0f DL |
2201 | /* |
2202 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
2203 | * above, we can be sure to be able to use the item looked up by | |
2204 | * ep_find() till we release the mutex. | |
2205 | */ | |
7e3fb584 | 2206 | epi = ep_find(ep, tf.file, fd); |
7699acd1 DL |
2207 | |
2208 | error = -EINVAL; | |
2209 | switch (op) { | |
2210 | case EPOLL_CTL_ADD: | |
2211 | if (!epi) { | |
a9a08845 | 2212 | epds.events |= EPOLLERR | EPOLLHUP; |
67347fe4 | 2213 | error = ep_insert(ep, &epds, tf.file, fd, full_check); |
7699acd1 DL |
2214 | } else |
2215 | error = -EEXIST; | |
2216 | break; | |
2217 | case EPOLL_CTL_DEL: | |
2218 | if (epi) | |
2219 | error = ep_remove(ep, epi); | |
2220 | else | |
2221 | error = -ENOENT; | |
2222 | break; | |
2223 | case EPOLL_CTL_MOD: | |
2224 | if (epi) { | |
b6a515c8 | 2225 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
a9a08845 | 2226 | epds.events |= EPOLLERR | EPOLLHUP; |
b6a515c8 JB |
2227 | error = ep_modify(ep, epi, &epds); |
2228 | } | |
7699acd1 DL |
2229 | } else |
2230 | error = -ENOENT; | |
2231 | break; | |
2232 | } | |
67347fe4 JB |
2233 | if (tep != NULL) |
2234 | mutex_unlock(&tep->mtx); | |
d47de16c | 2235 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
2236 | |
2237 | error_tgt_fput: | |
9dbadc3c AV |
2238 | if (full_check) { |
2239 | clear_tfile_check_list(); | |
140932e2 | 2240 | loop_check_gen++; |
22bacca4 | 2241 | mutex_unlock(&epmutex); |
9dbadc3c | 2242 | } |
22bacca4 | 2243 | |
7e3fb584 | 2244 | fdput(tf); |
7699acd1 | 2245 | error_fput: |
7e3fb584 | 2246 | fdput(f); |
7699acd1 | 2247 | error_return: |
7699acd1 DL |
2248 | |
2249 | return error; | |
2250 | } | |
2251 | ||
2252 | /* | |
2253 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2254 | * part of the user space epoll_wait(2). | |
2255 | */ | |
791eb22e DB |
2256 | static int do_epoll_wait(int epfd, struct epoll_event __user *events, |
2257 | int maxevents, int timeout) | |
7699acd1 | 2258 | { |
2903ff01 AV |
2259 | int error; |
2260 | struct fd f; | |
7699acd1 DL |
2261 | struct eventpoll *ep; |
2262 | ||
7699acd1 DL |
2263 | /* The maximum number of event must be greater than zero */ |
2264 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
2265 | return -EINVAL; | |
2266 | ||
2267 | /* Verify that the area passed by the user is writeable */ | |
96d4f267 | 2268 | if (!access_ok(events, maxevents * sizeof(struct epoll_event))) |
2903ff01 | 2269 | return -EFAULT; |
7699acd1 DL |
2270 | |
2271 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
2272 | f = fdget(epfd); |
2273 | if (!f.file) | |
2274 | return -EBADF; | |
7699acd1 DL |
2275 | |
2276 | /* | |
2277 | * We have to check that the file structure underneath the fd | |
2278 | * the user passed to us _is_ an eventpoll file. | |
2279 | */ | |
2280 | error = -EINVAL; | |
2903ff01 | 2281 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
2282 | goto error_fput; |
2283 | ||
2284 | /* | |
2285 | * At this point it is safe to assume that the "private_data" contains | |
2286 | * our own data structure. | |
2287 | */ | |
2903ff01 | 2288 | ep = f.file->private_data; |
7699acd1 DL |
2289 | |
2290 | /* Time to fish for events ... */ | |
2291 | error = ep_poll(ep, events, maxevents, timeout); | |
2292 | ||
2293 | error_fput: | |
2903ff01 | 2294 | fdput(f); |
7699acd1 DL |
2295 | return error; |
2296 | } | |
2297 | ||
791eb22e DB |
2298 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
2299 | int, maxevents, int, timeout) | |
2300 | { | |
2301 | return do_epoll_wait(epfd, events, maxevents, timeout); | |
2302 | } | |
2303 | ||
7699acd1 DL |
2304 | /* |
2305 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2306 | * part of the user space epoll_pwait(2). | |
2307 | */ | |
5a8a82b1 HC |
2308 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
2309 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
2310 | size_t, sigsetsize) | |
7699acd1 DL |
2311 | { |
2312 | int error; | |
7699acd1 DL |
2313 | |
2314 | /* | |
2315 | * If the caller wants a certain signal mask to be set during the wait, | |
2316 | * we apply it here. | |
2317 | */ | |
b772434b | 2318 | error = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2319 | if (error) |
2320 | return error; | |
7699acd1 | 2321 | |
791eb22e | 2322 | error = do_epoll_wait(epfd, events, maxevents, timeout); |
b772434b | 2323 | restore_saved_sigmask_unless(error == -EINTR); |
7699acd1 DL |
2324 | |
2325 | return error; | |
2326 | } | |
2327 | ||
35280bd4 AV |
2328 | #ifdef CONFIG_COMPAT |
2329 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, | |
2330 | struct epoll_event __user *, events, | |
2331 | int, maxevents, int, timeout, | |
2332 | const compat_sigset_t __user *, sigmask, | |
2333 | compat_size_t, sigsetsize) | |
2334 | { | |
2335 | long err; | |
35280bd4 AV |
2336 | |
2337 | /* | |
2338 | * If the caller wants a certain signal mask to be set during the wait, | |
2339 | * we apply it here. | |
2340 | */ | |
b772434b | 2341 | err = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2342 | if (err) |
2343 | return err; | |
35280bd4 | 2344 | |
791eb22e | 2345 | err = do_epoll_wait(epfd, events, maxevents, timeout); |
b772434b | 2346 | restore_saved_sigmask_unless(err == -EINTR); |
35280bd4 AV |
2347 | |
2348 | return err; | |
2349 | } | |
2350 | #endif | |
2351 | ||
1da177e4 LT |
2352 | static int __init eventpoll_init(void) |
2353 | { | |
7ef9964e DL |
2354 | struct sysinfo si; |
2355 | ||
2356 | si_meminfo(&si); | |
9df04e1f DL |
2357 | /* |
2358 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
2359 | */ | |
2360 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 2361 | EP_ITEM_COST; |
52bd19f7 | 2362 | BUG_ON(max_user_watches < 0); |
1da177e4 | 2363 | |
22bacca4 DL |
2364 | /* |
2365 | * Initialize the structure used to perform epoll file descriptor | |
2366 | * inclusion loops checks. | |
2367 | */ | |
2368 | ep_nested_calls_init(&poll_loop_ncalls); | |
2369 | ||
57a173bd | 2370 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1da177e4 | 2371 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
5071f97e | 2372 | ep_nested_calls_init(&poll_safewake_ncalls); |
57a173bd | 2373 | #endif |
1da177e4 | 2374 | |
39732ca5 EW |
2375 | /* |
2376 | * We can have many thousands of epitems, so prevent this from | |
2377 | * using an extra cache line on 64-bit (and smaller) CPUs | |
2378 | */ | |
2379 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); | |
2380 | ||
1da177e4 LT |
2381 | /* Allocates slab cache used to allocate "struct epitem" items */ |
2382 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
2ae928a9 | 2383 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 LT |
2384 | |
2385 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
2386 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
2ae928a9 | 2387 | sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 | 2388 | |
1da177e4 | 2389 | return 0; |
1da177e4 | 2390 | } |
cea69241 | 2391 | fs_initcall(eventpoll_init); |