Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 | 2 | /* |
5071f97e DL |
3 | * fs/eventpoll.c (Efficient event retrieval implementation) |
4 | * Copyright (C) 2001,...,2009 Davide Libenzi | |
1da177e4 | 5 | * |
1da177e4 | 6 | * Davide Libenzi <davidel@xmailserver.org> |
1da177e4 LT |
7 | */ |
8 | ||
1da177e4 LT |
9 | #include <linux/init.h> |
10 | #include <linux/kernel.h> | |
174cd4b1 | 11 | #include <linux/sched/signal.h> |
1da177e4 LT |
12 | #include <linux/fs.h> |
13 | #include <linux/file.h> | |
14 | #include <linux/signal.h> | |
15 | #include <linux/errno.h> | |
16 | #include <linux/mm.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/poll.h> | |
1da177e4 LT |
19 | #include <linux/string.h> |
20 | #include <linux/list.h> | |
21 | #include <linux/hash.h> | |
22 | #include <linux/spinlock.h> | |
23 | #include <linux/syscalls.h> | |
1da177e4 LT |
24 | #include <linux/rbtree.h> |
25 | #include <linux/wait.h> | |
26 | #include <linux/eventpoll.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/bitops.h> | |
144efe3e | 29 | #include <linux/mutex.h> |
da66f7cb | 30 | #include <linux/anon_inodes.h> |
4d7e30d9 | 31 | #include <linux/device.h> |
7c0f6ba6 | 32 | #include <linux/uaccess.h> |
1da177e4 LT |
33 | #include <asm/io.h> |
34 | #include <asm/mman.h> | |
60063497 | 35 | #include <linux/atomic.h> |
138d22b5 CG |
36 | #include <linux/proc_fs.h> |
37 | #include <linux/seq_file.h> | |
35280bd4 | 38 | #include <linux/compat.h> |
ae10b2b4 | 39 | #include <linux/rculist.h> |
bf3b9f63 | 40 | #include <net/busy_poll.h> |
1da177e4 | 41 | |
1da177e4 LT |
42 | /* |
43 | * LOCKING: | |
44 | * There are three level of locking required by epoll : | |
45 | * | |
144efe3e | 46 | * 1) epmutex (mutex) |
c7ea7630 | 47 | * 2) ep->mtx (mutex) |
a218cc49 | 48 | * 3) ep->lock (rwlock) |
1da177e4 LT |
49 | * |
50 | * The acquire order is the one listed above, from 1 to 3. | |
a218cc49 | 51 | * We need a rwlock (ep->lock) because we manipulate objects |
1da177e4 LT |
52 | * from inside the poll callback, that might be triggered from |
53 | * a wake_up() that in turn might be called from IRQ context. | |
54 | * So we can't sleep inside the poll callback and hence we need | |
55 | * a spinlock. During the event transfer loop (from kernel to | |
56 | * user space) we could end up sleeping due a copy_to_user(), so | |
57 | * we need a lock that will allow us to sleep. This lock is a | |
d47de16c DL |
58 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
59 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). | |
60 | * Then we also need a global mutex to serialize eventpoll_release_file() | |
61 | * and ep_free(). | |
62 | * This mutex is acquired by ep_free() during the epoll file | |
1da177e4 LT |
63 | * cleanup path and it is also acquired by eventpoll_release_file() |
64 | * if a file has been pushed inside an epoll set and it is then | |
bf6a41db | 65 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
22bacca4 DL |
66 | * It is also acquired when inserting an epoll fd onto another epoll |
67 | * fd. We do this so that we walk the epoll tree and ensure that this | |
68 | * insertion does not create a cycle of epoll file descriptors, which | |
69 | * could lead to deadlock. We need a global mutex to prevent two | |
70 | * simultaneous inserts (A into B and B into A) from racing and | |
71 | * constructing a cycle without either insert observing that it is | |
72 | * going to. | |
d8805e63 NE |
73 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
74 | * case when one epoll fd is added to another. In this case, we | |
75 | * always acquire the locks in the order of nesting (i.e. after | |
76 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired | |
77 | * before e2->mtx). Since we disallow cycles of epoll file | |
78 | * descriptors, this ensures that the mutexes are well-ordered. In | |
79 | * order to communicate this nesting to lockdep, when walking a tree | |
80 | * of epoll file descriptors, we use the current recursion depth as | |
81 | * the lockdep subkey. | |
d47de16c | 82 | * It is possible to drop the "ep->mtx" and to use the global |
a218cc49 | 83 | * mutex "epmutex" (together with "ep->lock") to have it working, |
d47de16c | 84 | * but having "ep->mtx" will make the interface more scalable. |
144efe3e | 85 | * Events that require holding "epmutex" are very rare, while for |
d47de16c DL |
86 | * normal operations the epoll private "ep->mtx" will guarantee |
87 | * a better scalability. | |
1da177e4 LT |
88 | */ |
89 | ||
1da177e4 | 90 | /* Epoll private bits inside the event mask */ |
df0108c5 | 91 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE) |
1da177e4 | 92 | |
a9a08845 | 93 | #define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT) |
b6a515c8 | 94 | |
a9a08845 | 95 | #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \ |
b6a515c8 JB |
96 | EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE) |
97 | ||
5071f97e DL |
98 | /* Maximum number of nesting allowed inside epoll sets */ |
99 | #define EP_MAX_NESTS 4 | |
1da177e4 | 100 | |
b611967d DL |
101 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
102 | ||
d47de16c DL |
103 | #define EP_UNACTIVE_PTR ((void *) -1L) |
104 | ||
7ef9964e DL |
105 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
106 | ||
1da177e4 LT |
107 | struct epoll_filefd { |
108 | struct file *file; | |
109 | int fd; | |
39732ca5 | 110 | } __packed; |
1da177e4 LT |
111 | |
112 | /* | |
5071f97e DL |
113 | * Structure used to track possible nested calls, for too deep recursions |
114 | * and loop cycles. | |
1da177e4 | 115 | */ |
5071f97e | 116 | struct nested_call_node { |
1da177e4 | 117 | struct list_head llink; |
5071f97e | 118 | void *cookie; |
3fe4a975 | 119 | void *ctx; |
1da177e4 LT |
120 | }; |
121 | ||
122 | /* | |
5071f97e DL |
123 | * This structure is used as collector for nested calls, to check for |
124 | * maximum recursion dept and loop cycles. | |
1da177e4 | 125 | */ |
5071f97e DL |
126 | struct nested_calls { |
127 | struct list_head tasks_call_list; | |
1da177e4 LT |
128 | spinlock_t lock; |
129 | }; | |
130 | ||
d47de16c DL |
131 | /* |
132 | * Each file descriptor added to the eventpoll interface will | |
133 | * have an entry of this type linked to the "rbr" RB tree. | |
39732ca5 EW |
134 | * Avoid increasing the size of this struct, there can be many thousands |
135 | * of these on a server and we do not want this to take another cache line. | |
d47de16c DL |
136 | */ |
137 | struct epitem { | |
ae10b2b4 JB |
138 | union { |
139 | /* RB tree node links this structure to the eventpoll RB tree */ | |
140 | struct rb_node rbn; | |
141 | /* Used to free the struct epitem */ | |
142 | struct rcu_head rcu; | |
143 | }; | |
d47de16c DL |
144 | |
145 | /* List header used to link this structure to the eventpoll ready list */ | |
146 | struct list_head rdllink; | |
147 | ||
c7ea7630 DL |
148 | /* |
149 | * Works together "struct eventpoll"->ovflist in keeping the | |
150 | * single linked chain of items. | |
151 | */ | |
152 | struct epitem *next; | |
153 | ||
d47de16c DL |
154 | /* The file descriptor information this item refers to */ |
155 | struct epoll_filefd ffd; | |
156 | ||
157 | /* Number of active wait queue attached to poll operations */ | |
158 | int nwait; | |
159 | ||
160 | /* List containing poll wait queues */ | |
161 | struct list_head pwqlist; | |
162 | ||
163 | /* The "container" of this item */ | |
164 | struct eventpoll *ep; | |
165 | ||
d47de16c DL |
166 | /* List header used to link this item to the "struct file" items list */ |
167 | struct list_head fllink; | |
168 | ||
4d7e30d9 | 169 | /* wakeup_source used when EPOLLWAKEUP is set */ |
eea1d585 | 170 | struct wakeup_source __rcu *ws; |
4d7e30d9 | 171 | |
c7ea7630 DL |
172 | /* The structure that describe the interested events and the source fd */ |
173 | struct epoll_event event; | |
d47de16c DL |
174 | }; |
175 | ||
1da177e4 LT |
176 | /* |
177 | * This structure is stored inside the "private_data" member of the file | |
bf6a41db | 178 | * structure and represents the main data structure for the eventpoll |
1da177e4 LT |
179 | * interface. |
180 | */ | |
181 | struct eventpoll { | |
1da177e4 | 182 | /* |
d47de16c DL |
183 | * This mutex is used to ensure that files are not removed |
184 | * while epoll is using them. This is held during the event | |
185 | * collection loop, the file cleanup path, the epoll file exit | |
186 | * code and the ctl operations. | |
1da177e4 | 187 | */ |
d47de16c | 188 | struct mutex mtx; |
1da177e4 LT |
189 | |
190 | /* Wait queue used by sys_epoll_wait() */ | |
191 | wait_queue_head_t wq; | |
192 | ||
193 | /* Wait queue used by file->poll() */ | |
194 | wait_queue_head_t poll_wait; | |
195 | ||
196 | /* List of ready file descriptors */ | |
197 | struct list_head rdllist; | |
198 | ||
a218cc49 RP |
199 | /* Lock which protects rdllist and ovflist */ |
200 | rwlock_t lock; | |
201 | ||
67647d0f | 202 | /* RB tree root used to store monitored fd structs */ |
b2ac2ea6 | 203 | struct rb_root_cached rbr; |
d47de16c DL |
204 | |
205 | /* | |
206 | * This is a single linked list that chains all the "struct epitem" that | |
25985edc | 207 | * happened while transferring ready events to userspace w/out |
a218cc49 | 208 | * holding ->lock. |
d47de16c DL |
209 | */ |
210 | struct epitem *ovflist; | |
7ef9964e | 211 | |
4d7e30d9 AH |
212 | /* wakeup_source used when ep_scan_ready_list is running */ |
213 | struct wakeup_source *ws; | |
214 | ||
7ef9964e DL |
215 | /* The user that created the eventpoll descriptor */ |
216 | struct user_struct *user; | |
28d82dc1 JB |
217 | |
218 | struct file *file; | |
219 | ||
220 | /* used to optimize loop detection check */ | |
221 | int visited; | |
222 | struct list_head visited_list_link; | |
bf3b9f63 SS |
223 | |
224 | #ifdef CONFIG_NET_RX_BUSY_POLL | |
225 | /* used to track busy poll napi_id */ | |
226 | unsigned int napi_id; | |
227 | #endif | |
1da177e4 LT |
228 | }; |
229 | ||
230 | /* Wait structure used by the poll hooks */ | |
231 | struct eppoll_entry { | |
232 | /* List header used to link this structure to the "struct epitem" */ | |
233 | struct list_head llink; | |
234 | ||
235 | /* The "base" pointer is set to the container "struct epitem" */ | |
4f0989db | 236 | struct epitem *base; |
1da177e4 LT |
237 | |
238 | /* | |
239 | * Wait queue item that will be linked to the target file wait | |
240 | * queue head. | |
241 | */ | |
ac6424b9 | 242 | wait_queue_entry_t wait; |
1da177e4 LT |
243 | |
244 | /* The wait queue head that linked the "wait" wait queue item */ | |
245 | wait_queue_head_t *whead; | |
246 | }; | |
247 | ||
1da177e4 LT |
248 | /* Wrapper struct used by poll queueing */ |
249 | struct ep_pqueue { | |
250 | poll_table pt; | |
251 | struct epitem *epi; | |
252 | }; | |
253 | ||
5071f97e DL |
254 | /* Used by the ep_send_events() function as callback private data */ |
255 | struct ep_send_events_data { | |
256 | int maxevents; | |
257 | struct epoll_event __user *events; | |
d7ebbe46 | 258 | int res; |
5071f97e DL |
259 | }; |
260 | ||
7ef9964e DL |
261 | /* |
262 | * Configuration options available inside /proc/sys/fs/epoll/ | |
263 | */ | |
7ef9964e | 264 | /* Maximum number of epoll watched descriptors, per user */ |
52bd19f7 | 265 | static long max_user_watches __read_mostly; |
7ef9964e | 266 | |
1da177e4 | 267 | /* |
d47de16c | 268 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
1da177e4 | 269 | */ |
7ef9964e | 270 | static DEFINE_MUTEX(epmutex); |
1da177e4 | 271 | |
22bacca4 DL |
272 | /* Used to check for epoll file descriptor inclusion loops */ |
273 | static struct nested_calls poll_loop_ncalls; | |
274 | ||
1da177e4 | 275 | /* Slab cache used to allocate "struct epitem" */ |
e18b890b | 276 | static struct kmem_cache *epi_cache __read_mostly; |
1da177e4 LT |
277 | |
278 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
e18b890b | 279 | static struct kmem_cache *pwq_cache __read_mostly; |
1da177e4 | 280 | |
28d82dc1 JB |
281 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
282 | static LIST_HEAD(visited_list); | |
283 | ||
284 | /* | |
285 | * List of files with newly added links, where we may need to limit the number | |
286 | * of emanating paths. Protected by the epmutex. | |
287 | */ | |
288 | static LIST_HEAD(tfile_check_list); | |
289 | ||
7ef9964e DL |
290 | #ifdef CONFIG_SYSCTL |
291 | ||
292 | #include <linux/sysctl.h> | |
293 | ||
eec4844f | 294 | static long long_zero; |
52bd19f7 | 295 | static long long_max = LONG_MAX; |
7ef9964e | 296 | |
1f7e0616 | 297 | struct ctl_table epoll_table[] = { |
7ef9964e DL |
298 | { |
299 | .procname = "max_user_watches", | |
300 | .data = &max_user_watches, | |
52bd19f7 | 301 | .maxlen = sizeof(max_user_watches), |
7ef9964e | 302 | .mode = 0644, |
52bd19f7 | 303 | .proc_handler = proc_doulongvec_minmax, |
eec4844f | 304 | .extra1 = &long_zero, |
52bd19f7 | 305 | .extra2 = &long_max, |
7ef9964e | 306 | }, |
ab09203e | 307 | { } |
7ef9964e DL |
308 | }; |
309 | #endif /* CONFIG_SYSCTL */ | |
310 | ||
28d82dc1 JB |
311 | static const struct file_operations eventpoll_fops; |
312 | ||
313 | static inline int is_file_epoll(struct file *f) | |
314 | { | |
315 | return f->f_op == &eventpoll_fops; | |
316 | } | |
b030a4dd | 317 | |
67647d0f | 318 | /* Setup the structure that is used as key for the RB tree */ |
b030a4dd PE |
319 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
320 | struct file *file, int fd) | |
321 | { | |
322 | ffd->file = file; | |
323 | ffd->fd = fd; | |
324 | } | |
325 | ||
67647d0f | 326 | /* Compare RB tree keys */ |
b030a4dd PE |
327 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
328 | struct epoll_filefd *p2) | |
329 | { | |
330 | return (p1->file > p2->file ? +1: | |
331 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); | |
332 | } | |
333 | ||
b030a4dd | 334 | /* Tells us if the item is currently linked */ |
992991c0 | 335 | static inline int ep_is_linked(struct epitem *epi) |
b030a4dd | 336 | { |
992991c0 | 337 | return !list_empty(&epi->rdllink); |
b030a4dd PE |
338 | } |
339 | ||
ac6424b9 | 340 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
971316f0 ON |
341 | { |
342 | return container_of(p, struct eppoll_entry, wait); | |
343 | } | |
344 | ||
b030a4dd | 345 | /* Get the "struct epitem" from a wait queue pointer */ |
ac6424b9 | 346 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
b030a4dd PE |
347 | { |
348 | return container_of(p, struct eppoll_entry, wait)->base; | |
349 | } | |
350 | ||
351 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
cdac75e6 | 352 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
b030a4dd PE |
353 | { |
354 | return container_of(p, struct ep_pqueue, pt)->epi; | |
355 | } | |
356 | ||
1da177e4 | 357 | /* Initialize the poll safe wake up structure */ |
5071f97e | 358 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
1da177e4 | 359 | { |
5071f97e DL |
360 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
361 | spin_lock_init(&ncalls->lock); | |
1da177e4 LT |
362 | } |
363 | ||
3fb0e584 DL |
364 | /** |
365 | * ep_events_available - Checks if ready events might be available. | |
366 | * | |
367 | * @ep: Pointer to the eventpoll context. | |
368 | * | |
369 | * Returns: Returns a value different than zero if ready events are available, | |
370 | * or zero otherwise. | |
371 | */ | |
372 | static inline int ep_events_available(struct eventpoll *ep) | |
373 | { | |
c5a282e9 DB |
374 | return !list_empty_careful(&ep->rdllist) || |
375 | READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR; | |
3fb0e584 DL |
376 | } |
377 | ||
bf3b9f63 SS |
378 | #ifdef CONFIG_NET_RX_BUSY_POLL |
379 | static bool ep_busy_loop_end(void *p, unsigned long start_time) | |
380 | { | |
381 | struct eventpoll *ep = p; | |
382 | ||
383 | return ep_events_available(ep) || busy_loop_timeout(start_time); | |
384 | } | |
bf3b9f63 SS |
385 | |
386 | /* | |
387 | * Busy poll if globally on and supporting sockets found && no events, | |
388 | * busy loop will return if need_resched or ep_events_available. | |
389 | * | |
390 | * we must do our busy polling with irqs enabled | |
391 | */ | |
392 | static void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
393 | { | |
bf3b9f63 SS |
394 | unsigned int napi_id = READ_ONCE(ep->napi_id); |
395 | ||
396 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) | |
397 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep); | |
bf3b9f63 SS |
398 | } |
399 | ||
400 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
401 | { | |
bf3b9f63 SS |
402 | if (ep->napi_id) |
403 | ep->napi_id = 0; | |
bf3b9f63 SS |
404 | } |
405 | ||
406 | /* | |
407 | * Set epoll busy poll NAPI ID from sk. | |
408 | */ | |
409 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
410 | { | |
bf3b9f63 SS |
411 | struct eventpoll *ep; |
412 | unsigned int napi_id; | |
413 | struct socket *sock; | |
414 | struct sock *sk; | |
415 | int err; | |
416 | ||
417 | if (!net_busy_loop_on()) | |
418 | return; | |
419 | ||
420 | sock = sock_from_file(epi->ffd.file, &err); | |
421 | if (!sock) | |
422 | return; | |
423 | ||
424 | sk = sock->sk; | |
425 | if (!sk) | |
426 | return; | |
427 | ||
428 | napi_id = READ_ONCE(sk->sk_napi_id); | |
429 | ep = epi->ep; | |
430 | ||
431 | /* Non-NAPI IDs can be rejected | |
432 | * or | |
433 | * Nothing to do if we already have this ID | |
434 | */ | |
435 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) | |
436 | return; | |
437 | ||
438 | /* record NAPI ID for use in next busy poll */ | |
439 | ep->napi_id = napi_id; | |
bf3b9f63 SS |
440 | } |
441 | ||
514056d5 DB |
442 | #else |
443 | ||
444 | static inline void ep_busy_loop(struct eventpoll *ep, int nonblock) | |
445 | { | |
446 | } | |
447 | ||
448 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) | |
449 | { | |
450 | } | |
451 | ||
452 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) | |
453 | { | |
454 | } | |
455 | ||
456 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
457 | ||
5071f97e DL |
458 | /** |
459 | * ep_call_nested - Perform a bound (possibly) nested call, by checking | |
460 | * that the recursion limit is not exceeded, and that | |
461 | * the same nested call (by the meaning of same cookie) is | |
462 | * no re-entered. | |
463 | * | |
464 | * @ncalls: Pointer to the nested_calls structure to be used for this call. | |
5071f97e DL |
465 | * @nproc: Nested call core function pointer. |
466 | * @priv: Opaque data to be passed to the @nproc callback. | |
467 | * @cookie: Cookie to be used to identify this nested call. | |
3fe4a975 | 468 | * @ctx: This instance context. |
5071f97e DL |
469 | * |
470 | * Returns: Returns the code returned by the @nproc callback, or -1 if | |
471 | * the maximum recursion limit has been exceeded. | |
1da177e4 | 472 | */ |
74bdc129 | 473 | static int ep_call_nested(struct nested_calls *ncalls, |
5071f97e | 474 | int (*nproc)(void *, void *, int), void *priv, |
3fe4a975 | 475 | void *cookie, void *ctx) |
1da177e4 | 476 | { |
5071f97e | 477 | int error, call_nests = 0; |
1da177e4 | 478 | unsigned long flags; |
5071f97e DL |
479 | struct list_head *lsthead = &ncalls->tasks_call_list; |
480 | struct nested_call_node *tncur; | |
481 | struct nested_call_node tnode; | |
1da177e4 | 482 | |
5071f97e | 483 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 484 | |
5071f97e DL |
485 | /* |
486 | * Try to see if the current task is already inside this wakeup call. | |
487 | * We use a list here, since the population inside this set is always | |
488 | * very much limited. | |
489 | */ | |
b70c3940 | 490 | list_for_each_entry(tncur, lsthead, llink) { |
3fe4a975 | 491 | if (tncur->ctx == ctx && |
74bdc129 | 492 | (tncur->cookie == cookie || ++call_nests > EP_MAX_NESTS)) { |
1da177e4 LT |
493 | /* |
494 | * Ops ... loop detected or maximum nest level reached. | |
495 | * We abort this wake by breaking the cycle itself. | |
496 | */ | |
abff55ce TB |
497 | error = -1; |
498 | goto out_unlock; | |
1da177e4 LT |
499 | } |
500 | } | |
501 | ||
5071f97e | 502 | /* Add the current task and cookie to the list */ |
3fe4a975 | 503 | tnode.ctx = ctx; |
5071f97e | 504 | tnode.cookie = cookie; |
1da177e4 LT |
505 | list_add(&tnode.llink, lsthead); |
506 | ||
5071f97e | 507 | spin_unlock_irqrestore(&ncalls->lock, flags); |
1da177e4 | 508 | |
5071f97e DL |
509 | /* Call the nested function */ |
510 | error = (*nproc)(priv, cookie, call_nests); | |
1da177e4 LT |
511 | |
512 | /* Remove the current task from the list */ | |
5071f97e | 513 | spin_lock_irqsave(&ncalls->lock, flags); |
1da177e4 | 514 | list_del(&tnode.llink); |
3fe4a975 | 515 | out_unlock: |
5071f97e DL |
516 | spin_unlock_irqrestore(&ncalls->lock, flags); |
517 | ||
518 | return error; | |
519 | } | |
520 | ||
02edc6fc SR |
521 | /* |
522 | * As described in commit 0ccf831cb lockdep: annotate epoll | |
523 | * the use of wait queues used by epoll is done in a very controlled | |
524 | * manner. Wake ups can nest inside each other, but are never done | |
525 | * with the same locking. For example: | |
526 | * | |
527 | * dfd = socket(...); | |
528 | * efd1 = epoll_create(); | |
529 | * efd2 = epoll_create(); | |
530 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); | |
531 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); | |
532 | * | |
533 | * When a packet arrives to the device underneath "dfd", the net code will | |
534 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a | |
535 | * callback wakeup entry on that queue, and the wake_up() performed by the | |
536 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll | |
537 | * (efd1) notices that it may have some event ready, so it needs to wake up | |
538 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() | |
539 | * that ends up in another wake_up(), after having checked about the | |
540 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to | |
541 | * avoid stack blasting. | |
542 | * | |
543 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle | |
544 | * this special case of epoll. | |
545 | */ | |
2dfa4eea | 546 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
57a173bd | 547 | |
f6520c52 | 548 | static DEFINE_PER_CPU(int, wakeup_nest); |
5071f97e | 549 | |
5071f97e DL |
550 | static void ep_poll_safewake(wait_queue_head_t *wq) |
551 | { | |
f6520c52 JB |
552 | unsigned long flags; |
553 | int subclass; | |
3fe4a975 | 554 | |
f6520c52 JB |
555 | local_irq_save(flags); |
556 | preempt_disable(); | |
557 | subclass = __this_cpu_read(wakeup_nest); | |
558 | spin_lock_nested(&wq->lock, subclass + 1); | |
559 | __this_cpu_inc(wakeup_nest); | |
560 | wake_up_locked_poll(wq, POLLIN); | |
561 | __this_cpu_dec(wakeup_nest); | |
562 | spin_unlock(&wq->lock); | |
563 | local_irq_restore(flags); | |
564 | preempt_enable(); | |
1da177e4 LT |
565 | } |
566 | ||
57a173bd JB |
567 | #else |
568 | ||
569 | static void ep_poll_safewake(wait_queue_head_t *wq) | |
570 | { | |
a9a08845 | 571 | wake_up_poll(wq, EPOLLIN); |
57a173bd JB |
572 | } |
573 | ||
574 | #endif | |
575 | ||
971316f0 ON |
576 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
577 | { | |
578 | wait_queue_head_t *whead; | |
579 | ||
580 | rcu_read_lock(); | |
138e4ad6 ON |
581 | /* |
582 | * If it is cleared by POLLFREE, it should be rcu-safe. | |
583 | * If we read NULL we need a barrier paired with | |
584 | * smp_store_release() in ep_poll_callback(), otherwise | |
585 | * we rely on whead->lock. | |
586 | */ | |
587 | whead = smp_load_acquire(&pwq->whead); | |
971316f0 ON |
588 | if (whead) |
589 | remove_wait_queue(whead, &pwq->wait); | |
590 | rcu_read_unlock(); | |
591 | } | |
592 | ||
1da177e4 | 593 | /* |
d1bc90dd TB |
594 | * This function unregisters poll callbacks from the associated file |
595 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from | |
596 | * ep_free). | |
1da177e4 | 597 | */ |
7699acd1 | 598 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
1da177e4 | 599 | { |
7699acd1 DL |
600 | struct list_head *lsthead = &epi->pwqlist; |
601 | struct eppoll_entry *pwq; | |
1da177e4 | 602 | |
d1bc90dd TB |
603 | while (!list_empty(lsthead)) { |
604 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); | |
1da177e4 | 605 | |
d1bc90dd | 606 | list_del(&pwq->llink); |
971316f0 | 607 | ep_remove_wait_queue(pwq); |
d1bc90dd | 608 | kmem_cache_free(pwq_cache, pwq); |
1da177e4 | 609 | } |
1da177e4 LT |
610 | } |
611 | ||
eea1d585 EW |
612 | /* call only when ep->mtx is held */ |
613 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) | |
614 | { | |
615 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); | |
616 | } | |
617 | ||
618 | /* call only when ep->mtx is held */ | |
619 | static inline void ep_pm_stay_awake(struct epitem *epi) | |
620 | { | |
621 | struct wakeup_source *ws = ep_wakeup_source(epi); | |
622 | ||
623 | if (ws) | |
624 | __pm_stay_awake(ws); | |
625 | } | |
626 | ||
627 | static inline bool ep_has_wakeup_source(struct epitem *epi) | |
628 | { | |
629 | return rcu_access_pointer(epi->ws) ? true : false; | |
630 | } | |
631 | ||
632 | /* call when ep->mtx cannot be held (ep_poll_callback) */ | |
633 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) | |
634 | { | |
635 | struct wakeup_source *ws; | |
636 | ||
637 | rcu_read_lock(); | |
638 | ws = rcu_dereference(epi->ws); | |
639 | if (ws) | |
640 | __pm_stay_awake(ws); | |
641 | rcu_read_unlock(); | |
642 | } | |
643 | ||
5071f97e DL |
644 | /** |
645 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for | |
646 | * the scan code, to call f_op->poll(). Also allows for | |
647 | * O(NumReady) performance. | |
648 | * | |
649 | * @ep: Pointer to the epoll private data structure. | |
650 | * @sproc: Pointer to the scan callback. | |
651 | * @priv: Private opaque data passed to the @sproc callback. | |
d8805e63 | 652 | * @depth: The current depth of recursive f_op->poll calls. |
67347fe4 | 653 | * @ep_locked: caller already holds ep->mtx |
5071f97e DL |
654 | * |
655 | * Returns: The same integer error code returned by the @sproc callback. | |
656 | */ | |
d85e2aa2 AV |
657 | static __poll_t ep_scan_ready_list(struct eventpoll *ep, |
658 | __poll_t (*sproc)(struct eventpoll *, | |
5071f97e | 659 | struct list_head *, void *), |
67347fe4 | 660 | void *priv, int depth, bool ep_locked) |
5071f97e | 661 | { |
d85e2aa2 | 662 | __poll_t res; |
5071f97e | 663 | struct epitem *epi, *nepi; |
296e236e | 664 | LIST_HEAD(txlist); |
5071f97e | 665 | |
92e64178 DB |
666 | lockdep_assert_irqs_enabled(); |
667 | ||
5071f97e DL |
668 | /* |
669 | * We need to lock this because we could be hit by | |
e057e15f | 670 | * eventpoll_release_file() and epoll_ctl(). |
5071f97e | 671 | */ |
67347fe4 JB |
672 | |
673 | if (!ep_locked) | |
674 | mutex_lock_nested(&ep->mtx, depth); | |
5071f97e DL |
675 | |
676 | /* | |
677 | * Steal the ready list, and re-init the original one to the | |
678 | * empty list. Also, set ep->ovflist to NULL so that events | |
679 | * happening while looping w/out locks, are not lost. We cannot | |
680 | * have the poll callback to queue directly on ep->rdllist, | |
681 | * because we want the "sproc" callback to be able to do it | |
682 | * in a lockless way. | |
683 | */ | |
a218cc49 | 684 | write_lock_irq(&ep->lock); |
296e236e | 685 | list_splice_init(&ep->rdllist, &txlist); |
c5a282e9 | 686 | WRITE_ONCE(ep->ovflist, NULL); |
a218cc49 | 687 | write_unlock_irq(&ep->lock); |
5071f97e DL |
688 | |
689 | /* | |
690 | * Now call the callback function. | |
691 | */ | |
d85e2aa2 | 692 | res = (*sproc)(ep, &txlist, priv); |
5071f97e | 693 | |
a218cc49 | 694 | write_lock_irq(&ep->lock); |
5071f97e DL |
695 | /* |
696 | * During the time we spent inside the "sproc" callback, some | |
697 | * other events might have been queued by the poll callback. | |
698 | * We re-insert them inside the main ready-list here. | |
699 | */ | |
c5a282e9 | 700 | for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL; |
5071f97e DL |
701 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
702 | /* | |
703 | * We need to check if the item is already in the list. | |
704 | * During the "sproc" callback execution time, items are | |
705 | * queued into ->ovflist but the "txlist" might already | |
706 | * contain them, and the list_splice() below takes care of them. | |
707 | */ | |
992991c0 | 708 | if (!ep_is_linked(epi)) { |
c141175d RP |
709 | /* |
710 | * ->ovflist is LIFO, so we have to reverse it in order | |
711 | * to keep in FIFO. | |
712 | */ | |
713 | list_add(&epi->rdllink, &ep->rdllist); | |
eea1d585 | 714 | ep_pm_stay_awake(epi); |
4d7e30d9 | 715 | } |
5071f97e DL |
716 | } |
717 | /* | |
718 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after | |
719 | * releasing the lock, events will be queued in the normal way inside | |
720 | * ep->rdllist. | |
721 | */ | |
c5a282e9 | 722 | WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR); |
5071f97e DL |
723 | |
724 | /* | |
725 | * Quickly re-inject items left on "txlist". | |
726 | */ | |
727 | list_splice(&txlist, &ep->rdllist); | |
4d7e30d9 | 728 | __pm_relax(ep->ws); |
a218cc49 | 729 | write_unlock_irq(&ep->lock); |
5071f97e | 730 | |
67347fe4 JB |
731 | if (!ep_locked) |
732 | mutex_unlock(&ep->mtx); | |
5071f97e | 733 | |
d85e2aa2 | 734 | return res; |
5071f97e DL |
735 | } |
736 | ||
ae10b2b4 JB |
737 | static void epi_rcu_free(struct rcu_head *head) |
738 | { | |
739 | struct epitem *epi = container_of(head, struct epitem, rcu); | |
740 | kmem_cache_free(epi_cache, epi); | |
741 | } | |
742 | ||
7699acd1 DL |
743 | /* |
744 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates | |
c7ea7630 | 745 | * all the associated resources. Must be called with "mtx" held. |
7699acd1 DL |
746 | */ |
747 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
748 | { | |
7699acd1 | 749 | struct file *file = epi->ffd.file; |
1da177e4 | 750 | |
92e64178 DB |
751 | lockdep_assert_irqs_enabled(); |
752 | ||
1da177e4 | 753 | /* |
ee8ef0a4 | 754 | * Removes poll wait queue hooks. |
1da177e4 | 755 | */ |
7699acd1 | 756 | ep_unregister_pollwait(ep, epi); |
1da177e4 | 757 | |
7699acd1 | 758 | /* Remove the current item from the list of epoll hooks */ |
68499914 | 759 | spin_lock(&file->f_lock); |
ae10b2b4 | 760 | list_del_rcu(&epi->fllink); |
68499914 | 761 | spin_unlock(&file->f_lock); |
1da177e4 | 762 | |
b2ac2ea6 | 763 | rb_erase_cached(&epi->rbn, &ep->rbr); |
1da177e4 | 764 | |
a218cc49 | 765 | write_lock_irq(&ep->lock); |
992991c0 | 766 | if (ep_is_linked(epi)) |
c7ea7630 | 767 | list_del_init(&epi->rdllink); |
a218cc49 | 768 | write_unlock_irq(&ep->lock); |
1da177e4 | 769 | |
eea1d585 | 770 | wakeup_source_unregister(ep_wakeup_source(epi)); |
ae10b2b4 JB |
771 | /* |
772 | * At this point it is safe to free the eventpoll item. Use the union | |
773 | * field epi->rcu, since we are trying to minimize the size of | |
774 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by | |
775 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make | |
776 | * use of the rbn field. | |
777 | */ | |
778 | call_rcu(&epi->rcu, epi_rcu_free); | |
1da177e4 | 779 | |
52bd19f7 | 780 | atomic_long_dec(&ep->user->epoll_watches); |
7ef9964e | 781 | |
c7ea7630 | 782 | return 0; |
1da177e4 LT |
783 | } |
784 | ||
7699acd1 | 785 | static void ep_free(struct eventpoll *ep) |
1da177e4 | 786 | { |
7699acd1 DL |
787 | struct rb_node *rbp; |
788 | struct epitem *epi; | |
1da177e4 | 789 | |
7699acd1 DL |
790 | /* We need to release all tasks waiting for these file */ |
791 | if (waitqueue_active(&ep->poll_wait)) | |
5071f97e | 792 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 793 | |
7699acd1 DL |
794 | /* |
795 | * We need to lock this because we could be hit by | |
796 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
d47de16c | 797 | * We do not need to hold "ep->mtx" here because the epoll file |
7699acd1 DL |
798 | * is on the way to be removed and no one has references to it |
799 | * anymore. The only hit might come from eventpoll_release_file() but | |
25985edc | 800 | * holding "epmutex" is sufficient here. |
7699acd1 DL |
801 | */ |
802 | mutex_lock(&epmutex); | |
1da177e4 LT |
803 | |
804 | /* | |
7699acd1 | 805 | * Walks through the whole tree by unregistering poll callbacks. |
1da177e4 | 806 | */ |
b2ac2ea6 | 807 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
7699acd1 DL |
808 | epi = rb_entry(rbp, struct epitem, rbn); |
809 | ||
810 | ep_unregister_pollwait(ep, epi); | |
91cf5ab6 | 811 | cond_resched(); |
7699acd1 | 812 | } |
1da177e4 LT |
813 | |
814 | /* | |
7699acd1 DL |
815 | * Walks through the whole tree by freeing each "struct epitem". At this |
816 | * point we are sure no poll callbacks will be lingering around, and also by | |
d47de16c | 817 | * holding "epmutex" we can be sure that no file cleanup code will hit |
a218cc49 | 818 | * us during this operation. So we can avoid the lock on "ep->lock". |
ddf676c3 EW |
819 | * We do not need to lock ep->mtx, either, we only do it to prevent |
820 | * a lockdep warning. | |
1da177e4 | 821 | */ |
ddf676c3 | 822 | mutex_lock(&ep->mtx); |
b2ac2ea6 | 823 | while ((rbp = rb_first_cached(&ep->rbr)) != NULL) { |
7699acd1 DL |
824 | epi = rb_entry(rbp, struct epitem, rbn); |
825 | ep_remove(ep, epi); | |
91cf5ab6 | 826 | cond_resched(); |
7699acd1 | 827 | } |
ddf676c3 | 828 | mutex_unlock(&ep->mtx); |
1da177e4 | 829 | |
7699acd1 | 830 | mutex_unlock(&epmutex); |
d47de16c | 831 | mutex_destroy(&ep->mtx); |
7ef9964e | 832 | free_uid(ep->user); |
4d7e30d9 | 833 | wakeup_source_unregister(ep->ws); |
f0ee9aab | 834 | kfree(ep); |
7699acd1 | 835 | } |
1da177e4 | 836 | |
7699acd1 DL |
837 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
838 | { | |
839 | struct eventpoll *ep = file->private_data; | |
1da177e4 | 840 | |
f0ee9aab | 841 | if (ep) |
7699acd1 | 842 | ep_free(ep); |
7699acd1 | 843 | |
7699acd1 | 844 | return 0; |
1da177e4 LT |
845 | } |
846 | ||
d85e2aa2 | 847 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
37b5e521 JB |
848 | void *priv); |
849 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
850 | poll_table *pt); | |
851 | ||
852 | /* | |
853 | * Differs from ep_eventpoll_poll() in that internal callers already have | |
854 | * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested() | |
855 | * is correctly annotated. | |
856 | */ | |
d85e2aa2 | 857 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, |
bec1a502 | 858 | int depth) |
450d89ec | 859 | { |
37b5e521 JB |
860 | struct eventpoll *ep; |
861 | bool locked; | |
862 | ||
450d89ec | 863 | pt->_key = epi->event.events; |
37b5e521 | 864 | if (!is_file_epoll(epi->ffd.file)) |
9965ed17 | 865 | return vfs_poll(epi->ffd.file, pt) & epi->event.events; |
450d89ec | 866 | |
37b5e521 JB |
867 | ep = epi->ffd.file->private_data; |
868 | poll_wait(epi->ffd.file, &ep->poll_wait, pt); | |
869 | locked = pt && (pt->_qproc == ep_ptable_queue_proc); | |
450d89ec | 870 | |
37b5e521 JB |
871 | return ep_scan_ready_list(epi->ffd.file->private_data, |
872 | ep_read_events_proc, &depth, depth, | |
873 | locked) & epi->event.events; | |
450d89ec EW |
874 | } |
875 | ||
d85e2aa2 | 876 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
296e236e | 877 | void *priv) |
5071f97e DL |
878 | { |
879 | struct epitem *epi, *tmp; | |
626cf236 | 880 | poll_table pt; |
37b5e521 | 881 | int depth = *(int *)priv; |
5071f97e | 882 | |
626cf236 | 883 | init_poll_funcptr(&pt, NULL); |
37b5e521 | 884 | depth++; |
450d89ec | 885 | |
5071f97e | 886 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
37b5e521 | 887 | if (ep_item_poll(epi, &pt, depth)) { |
a9a08845 | 888 | return EPOLLIN | EPOLLRDNORM; |
37b5e521 | 889 | } else { |
5071f97e DL |
890 | /* |
891 | * Item has been dropped into the ready list by the poll | |
892 | * callback, but it's not actually ready, as far as | |
893 | * caller requested events goes. We can remove it here. | |
894 | */ | |
eea1d585 | 895 | __pm_relax(ep_wakeup_source(epi)); |
5071f97e | 896 | list_del_init(&epi->rdllink); |
296e236e | 897 | } |
5071f97e DL |
898 | } |
899 | ||
900 | return 0; | |
901 | } | |
902 | ||
a11e1d43 | 903 | static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait) |
11c5ad0e BN |
904 | { |
905 | struct eventpoll *ep = file->private_data; | |
906 | int depth = 0; | |
7699acd1 | 907 | |
a11e1d43 LT |
908 | /* Insert inside our poll wait queue */ |
909 | poll_wait(file, &ep->poll_wait, wait); | |
910 | ||
5071f97e DL |
911 | /* |
912 | * Proceed to find out if wanted events are really available inside | |
37b5e521 | 913 | * the ready list. |
5071f97e | 914 | */ |
37b5e521 JB |
915 | return ep_scan_ready_list(ep, ep_read_events_proc, |
916 | &depth, depth, false); | |
7699acd1 DL |
917 | } |
918 | ||
138d22b5 | 919 | #ifdef CONFIG_PROC_FS |
a3816ab0 | 920 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
138d22b5 CG |
921 | { |
922 | struct eventpoll *ep = f->private_data; | |
923 | struct rb_node *rbp; | |
138d22b5 CG |
924 | |
925 | mutex_lock(&ep->mtx); | |
b2ac2ea6 | 926 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
138d22b5 | 927 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
77493f04 | 928 | struct inode *inode = file_inode(epi->ffd.file); |
138d22b5 | 929 | |
77493f04 CG |
930 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
931 | " pos:%lli ino:%lx sdev:%x\n", | |
a3816ab0 | 932 | epi->ffd.fd, epi->event.events, |
77493f04 CG |
933 | (long long)epi->event.data, |
934 | (long long)epi->ffd.file->f_pos, | |
935 | inode->i_ino, inode->i_sb->s_dev); | |
a3816ab0 | 936 | if (seq_has_overflowed(m)) |
138d22b5 CG |
937 | break; |
938 | } | |
939 | mutex_unlock(&ep->mtx); | |
138d22b5 CG |
940 | } |
941 | #endif | |
942 | ||
7699acd1 DL |
943 | /* File callbacks that implement the eventpoll file behaviour */ |
944 | static const struct file_operations eventpoll_fops = { | |
138d22b5 CG |
945 | #ifdef CONFIG_PROC_FS |
946 | .show_fdinfo = ep_show_fdinfo, | |
947 | #endif | |
7699acd1 | 948 | .release = ep_eventpoll_release, |
a11e1d43 | 949 | .poll = ep_eventpoll_poll, |
6038f373 | 950 | .llseek = noop_llseek, |
7699acd1 DL |
951 | }; |
952 | ||
b611967d | 953 | /* |
7699acd1 DL |
954 | * This is called from eventpoll_release() to unlink files from the eventpoll |
955 | * interface. We need to have this facility to cleanup correctly files that are | |
956 | * closed without being removed from the eventpoll interface. | |
b611967d | 957 | */ |
7699acd1 | 958 | void eventpoll_release_file(struct file *file) |
b611967d | 959 | { |
7699acd1 | 960 | struct eventpoll *ep; |
ebe06187 | 961 | struct epitem *epi, *next; |
b611967d DL |
962 | |
963 | /* | |
68499914 | 964 | * We don't want to get "file->f_lock" because it is not |
7699acd1 | 965 | * necessary. It is not necessary because we're in the "struct file" |
25985edc | 966 | * cleanup path, and this means that no one is using this file anymore. |
5071f97e | 967 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
67647d0f | 968 | * point, the file counter already went to zero and fget() would fail. |
d47de16c | 969 | * The only hit might come from ep_free() but by holding the mutex |
7699acd1 | 970 | * will correctly serialize the operation. We do need to acquire |
d47de16c | 971 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
7699acd1 | 972 | * from anywhere but ep_free(). |
68499914 JC |
973 | * |
974 | * Besides, ep_remove() acquires the lock, so we can't hold it here. | |
b611967d | 975 | */ |
7699acd1 | 976 | mutex_lock(&epmutex); |
ebe06187 | 977 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
7699acd1 | 978 | ep = epi->ep; |
d8805e63 | 979 | mutex_lock_nested(&ep->mtx, 0); |
7699acd1 | 980 | ep_remove(ep, epi); |
d47de16c | 981 | mutex_unlock(&ep->mtx); |
b611967d | 982 | } |
7699acd1 | 983 | mutex_unlock(&epmutex); |
b611967d DL |
984 | } |
985 | ||
53d2be79 | 986 | static int ep_alloc(struct eventpoll **pep) |
1da177e4 | 987 | { |
7ef9964e DL |
988 | int error; |
989 | struct user_struct *user; | |
990 | struct eventpoll *ep; | |
1da177e4 | 991 | |
7ef9964e | 992 | user = get_current_user(); |
7ef9964e DL |
993 | error = -ENOMEM; |
994 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); | |
995 | if (unlikely(!ep)) | |
996 | goto free_uid; | |
1da177e4 | 997 | |
d47de16c | 998 | mutex_init(&ep->mtx); |
a218cc49 | 999 | rwlock_init(&ep->lock); |
1da177e4 LT |
1000 | init_waitqueue_head(&ep->wq); |
1001 | init_waitqueue_head(&ep->poll_wait); | |
1002 | INIT_LIST_HEAD(&ep->rdllist); | |
b2ac2ea6 | 1003 | ep->rbr = RB_ROOT_CACHED; |
d47de16c | 1004 | ep->ovflist = EP_UNACTIVE_PTR; |
7ef9964e | 1005 | ep->user = user; |
1da177e4 | 1006 | |
53d2be79 | 1007 | *pep = ep; |
1da177e4 | 1008 | |
1da177e4 | 1009 | return 0; |
7ef9964e DL |
1010 | |
1011 | free_uid: | |
1012 | free_uid(user); | |
1013 | return error; | |
1da177e4 LT |
1014 | } |
1015 | ||
1da177e4 | 1016 | /* |
c7ea7630 DL |
1017 | * Search the file inside the eventpoll tree. The RB tree operations |
1018 | * are protected by the "mtx" mutex, and ep_find() must be called with | |
1019 | * "mtx" held. | |
1da177e4 LT |
1020 | */ |
1021 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
1022 | { | |
1023 | int kcmp; | |
1da177e4 LT |
1024 | struct rb_node *rbp; |
1025 | struct epitem *epi, *epir = NULL; | |
1026 | struct epoll_filefd ffd; | |
1027 | ||
b030a4dd | 1028 | ep_set_ffd(&ffd, file, fd); |
b2ac2ea6 | 1029 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
1da177e4 | 1030 | epi = rb_entry(rbp, struct epitem, rbn); |
b030a4dd | 1031 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
1da177e4 LT |
1032 | if (kcmp > 0) |
1033 | rbp = rbp->rb_right; | |
1034 | else if (kcmp < 0) | |
1035 | rbp = rbp->rb_left; | |
1036 | else { | |
1da177e4 LT |
1037 | epir = epi; |
1038 | break; | |
1039 | } | |
1040 | } | |
1da177e4 | 1041 | |
1da177e4 LT |
1042 | return epir; |
1043 | } | |
1044 | ||
92ef6da3 | 1045 | #ifdef CONFIG_CHECKPOINT_RESTORE |
0791e364 CG |
1046 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
1047 | { | |
1048 | struct rb_node *rbp; | |
1049 | struct epitem *epi; | |
1050 | ||
b2ac2ea6 | 1051 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
0791e364 CG |
1052 | epi = rb_entry(rbp, struct epitem, rbn); |
1053 | if (epi->ffd.fd == tfd) { | |
1054 | if (toff == 0) | |
1055 | return epi; | |
1056 | else | |
1057 | toff--; | |
1058 | } | |
1059 | cond_resched(); | |
1060 | } | |
1061 | ||
1062 | return NULL; | |
1063 | } | |
1064 | ||
1065 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, | |
1066 | unsigned long toff) | |
1067 | { | |
1068 | struct file *file_raw; | |
1069 | struct eventpoll *ep; | |
1070 | struct epitem *epi; | |
1071 | ||
1072 | if (!is_file_epoll(file)) | |
1073 | return ERR_PTR(-EINVAL); | |
1074 | ||
1075 | ep = file->private_data; | |
1076 | ||
1077 | mutex_lock(&ep->mtx); | |
1078 | epi = ep_find_tfd(ep, tfd, toff); | |
1079 | if (epi) | |
1080 | file_raw = epi->ffd.file; | |
1081 | else | |
1082 | file_raw = ERR_PTR(-ENOENT); | |
1083 | mutex_unlock(&ep->mtx); | |
1084 | ||
1085 | return file_raw; | |
1086 | } | |
92ef6da3 | 1087 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
0791e364 | 1088 | |
a218cc49 RP |
1089 | /** |
1090 | * Adds a new entry to the tail of the list in a lockless way, i.e. | |
1091 | * multiple CPUs are allowed to call this function concurrently. | |
1092 | * | |
1093 | * Beware: it is necessary to prevent any other modifications of the | |
1094 | * existing list until all changes are completed, in other words | |
1095 | * concurrent list_add_tail_lockless() calls should be protected | |
1096 | * with a read lock, where write lock acts as a barrier which | |
1097 | * makes sure all list_add_tail_lockless() calls are fully | |
1098 | * completed. | |
1099 | * | |
1100 | * Also an element can be locklessly added to the list only in one | |
1101 | * direction i.e. either to the tail either to the head, otherwise | |
1102 | * concurrent access will corrupt the list. | |
1103 | * | |
1104 | * Returns %false if element has been already added to the list, %true | |
1105 | * otherwise. | |
1106 | */ | |
1107 | static inline bool list_add_tail_lockless(struct list_head *new, | |
1108 | struct list_head *head) | |
1109 | { | |
1110 | struct list_head *prev; | |
1111 | ||
1112 | /* | |
1113 | * This is simple 'new->next = head' operation, but cmpxchg() | |
1114 | * is used in order to detect that same element has been just | |
1115 | * added to the list from another CPU: the winner observes | |
1116 | * new->next == new. | |
1117 | */ | |
1118 | if (cmpxchg(&new->next, new, head) != new) | |
1119 | return false; | |
1120 | ||
1121 | /* | |
1122 | * Initially ->next of a new element must be updated with the head | |
1123 | * (we are inserting to the tail) and only then pointers are atomically | |
1124 | * exchanged. XCHG guarantees memory ordering, thus ->next should be | |
1125 | * updated before pointers are actually swapped and pointers are | |
1126 | * swapped before prev->next is updated. | |
1127 | */ | |
1128 | ||
1129 | prev = xchg(&head->prev, new); | |
1130 | ||
1131 | /* | |
1132 | * It is safe to modify prev->next and new->prev, because a new element | |
1133 | * is added only to the tail and new->next is updated before XCHG. | |
1134 | */ | |
1135 | ||
1136 | prev->next = new; | |
1137 | new->prev = prev; | |
1138 | ||
1139 | return true; | |
1140 | } | |
1141 | ||
1142 | /** | |
1143 | * Chains a new epi entry to the tail of the ep->ovflist in a lockless way, | |
1144 | * i.e. multiple CPUs are allowed to call this function concurrently. | |
1145 | * | |
1146 | * Returns %false if epi element has been already chained, %true otherwise. | |
1147 | */ | |
1148 | static inline bool chain_epi_lockless(struct epitem *epi) | |
1149 | { | |
1150 | struct eventpoll *ep = epi->ep; | |
1151 | ||
1152 | /* Check that the same epi has not been just chained from another CPU */ | |
1153 | if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR) | |
1154 | return false; | |
1155 | ||
1156 | /* Atomically exchange tail */ | |
1157 | epi->next = xchg(&ep->ovflist, epi); | |
1158 | ||
1159 | return true; | |
1160 | } | |
1161 | ||
1da177e4 | 1162 | /* |
7699acd1 | 1163 | * This is the callback that is passed to the wait queue wakeup |
bf6a41db | 1164 | * mechanism. It is called by the stored file descriptors when they |
7699acd1 | 1165 | * have events to report. |
a218cc49 RP |
1166 | * |
1167 | * This callback takes a read lock in order not to content with concurrent | |
1168 | * events from another file descriptors, thus all modifications to ->rdllist | |
1169 | * or ->ovflist are lockless. Read lock is paired with the write lock from | |
1170 | * ep_scan_ready_list(), which stops all list modifications and guarantees | |
1171 | * that lists state is seen correctly. | |
1172 | * | |
1173 | * Another thing worth to mention is that ep_poll_callback() can be called | |
1174 | * concurrently for the same @epi from different CPUs if poll table was inited | |
1175 | * with several wait queues entries. Plural wakeup from different CPUs of a | |
1176 | * single wait queue is serialized by wq.lock, but the case when multiple wait | |
1177 | * queues are used should be detected accordingly. This is detected using | |
1178 | * cmpxchg() operation. | |
1da177e4 | 1179 | */ |
ac6424b9 | 1180 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
1da177e4 | 1181 | { |
7699acd1 | 1182 | int pwake = 0; |
7699acd1 DL |
1183 | struct epitem *epi = ep_item_from_wait(wait); |
1184 | struct eventpoll *ep = epi->ep; | |
3ad6f93e | 1185 | __poll_t pollflags = key_to_poll(key); |
a218cc49 | 1186 | unsigned long flags; |
df0108c5 | 1187 | int ewake = 0; |
1da177e4 | 1188 | |
a218cc49 | 1189 | read_lock_irqsave(&ep->lock, flags); |
1da177e4 | 1190 | |
bf3b9f63 SS |
1191 | ep_set_busy_poll_napi_id(epi); |
1192 | ||
7699acd1 DL |
1193 | /* |
1194 | * If the event mask does not contain any poll(2) event, we consider the | |
1195 | * descriptor to be disabled. This condition is likely the effect of the | |
1196 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1197 | * until the next EPOLL_CTL_MOD will be issued. | |
1198 | */ | |
1199 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
d47de16c DL |
1200 | goto out_unlock; |
1201 | ||
2dfa4eea DL |
1202 | /* |
1203 | * Check the events coming with the callback. At this stage, not | |
1204 | * every device reports the events in the "key" parameter of the | |
1205 | * callback. We need to be able to handle both cases here, hence the | |
1206 | * test for "key" != NULL before the event match test. | |
1207 | */ | |
3ad6f93e | 1208 | if (pollflags && !(pollflags & epi->event.events)) |
2dfa4eea DL |
1209 | goto out_unlock; |
1210 | ||
d47de16c | 1211 | /* |
bf6a41db | 1212 | * If we are transferring events to userspace, we can hold no locks |
d47de16c | 1213 | * (because we're accessing user memory, and because of linux f_op->poll() |
bf6a41db | 1214 | * semantics). All the events that happen during that period of time are |
d47de16c DL |
1215 | * chained in ep->ovflist and requeued later on. |
1216 | */ | |
c5a282e9 | 1217 | if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) { |
a218cc49 RP |
1218 | if (epi->next == EP_UNACTIVE_PTR && |
1219 | chain_epi_lockless(epi)) | |
c3e320b6 | 1220 | ep_pm_stay_awake_rcu(epi); |
d47de16c DL |
1221 | goto out_unlock; |
1222 | } | |
1da177e4 | 1223 | |
7699acd1 | 1224 | /* If this file is already in the ready list we exit soon */ |
a218cc49 RP |
1225 | if (!ep_is_linked(epi) && |
1226 | list_add_tail_lockless(&epi->rdllink, &ep->rdllist)) { | |
eea1d585 | 1227 | ep_pm_stay_awake_rcu(epi); |
4d7e30d9 | 1228 | } |
7699acd1 | 1229 | |
7699acd1 DL |
1230 | /* |
1231 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1232 | * wait list. | |
1233 | */ | |
df0108c5 | 1234 | if (waitqueue_active(&ep->wq)) { |
b6a515c8 | 1235 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
3ad6f93e AV |
1236 | !(pollflags & POLLFREE)) { |
1237 | switch (pollflags & EPOLLINOUT_BITS) { | |
a9a08845 LT |
1238 | case EPOLLIN: |
1239 | if (epi->event.events & EPOLLIN) | |
b6a515c8 JB |
1240 | ewake = 1; |
1241 | break; | |
a9a08845 LT |
1242 | case EPOLLOUT: |
1243 | if (epi->event.events & EPOLLOUT) | |
b6a515c8 JB |
1244 | ewake = 1; |
1245 | break; | |
1246 | case 0: | |
1247 | ewake = 1; | |
1248 | break; | |
1249 | } | |
1250 | } | |
a218cc49 | 1251 | wake_up(&ep->wq); |
df0108c5 | 1252 | } |
7699acd1 DL |
1253 | if (waitqueue_active(&ep->poll_wait)) |
1254 | pwake++; | |
1255 | ||
d47de16c | 1256 | out_unlock: |
a218cc49 | 1257 | read_unlock_irqrestore(&ep->lock, flags); |
1da177e4 | 1258 | |
7699acd1 DL |
1259 | /* We have to call this outside the lock */ |
1260 | if (pwake) | |
5071f97e | 1261 | ep_poll_safewake(&ep->poll_wait); |
7699acd1 | 1262 | |
138e4ad6 ON |
1263 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
1264 | ewake = 1; | |
1265 | ||
3ad6f93e | 1266 | if (pollflags & POLLFREE) { |
138e4ad6 ON |
1267 | /* |
1268 | * If we race with ep_remove_wait_queue() it can miss | |
1269 | * ->whead = NULL and do another remove_wait_queue() after | |
1270 | * us, so we can't use __remove_wait_queue(). | |
1271 | */ | |
1272 | list_del_init(&wait->entry); | |
1273 | /* | |
1274 | * ->whead != NULL protects us from the race with ep_free() | |
1275 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock | |
1276 | * held by the caller. Once we nullify it, nothing protects | |
1277 | * ep/epi or even wait. | |
1278 | */ | |
1279 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); | |
1280 | } | |
df0108c5 | 1281 | |
138e4ad6 | 1282 | return ewake; |
7699acd1 | 1283 | } |
1da177e4 LT |
1284 | |
1285 | /* | |
1286 | * This is the callback that is used to add our wait queue to the | |
1287 | * target file wakeup lists. | |
1288 | */ | |
1289 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
1290 | poll_table *pt) | |
1291 | { | |
b030a4dd | 1292 | struct epitem *epi = ep_item_from_epqueue(pt); |
1da177e4 LT |
1293 | struct eppoll_entry *pwq; |
1294 | ||
e94b1766 | 1295 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
1da177e4 LT |
1296 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
1297 | pwq->whead = whead; | |
1298 | pwq->base = epi; | |
df0108c5 JB |
1299 | if (epi->event.events & EPOLLEXCLUSIVE) |
1300 | add_wait_queue_exclusive(whead, &pwq->wait); | |
1301 | else | |
1302 | add_wait_queue(whead, &pwq->wait); | |
1da177e4 LT |
1303 | list_add_tail(&pwq->llink, &epi->pwqlist); |
1304 | epi->nwait++; | |
296e236e | 1305 | } else { |
1da177e4 LT |
1306 | /* We have to signal that an error occurred */ |
1307 | epi->nwait = -1; | |
296e236e | 1308 | } |
1da177e4 LT |
1309 | } |
1310 | ||
1da177e4 LT |
1311 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
1312 | { | |
1313 | int kcmp; | |
b2ac2ea6 | 1314 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
1da177e4 | 1315 | struct epitem *epic; |
b2ac2ea6 | 1316 | bool leftmost = true; |
1da177e4 LT |
1317 | |
1318 | while (*p) { | |
1319 | parent = *p; | |
1320 | epic = rb_entry(parent, struct epitem, rbn); | |
b030a4dd | 1321 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
b2ac2ea6 | 1322 | if (kcmp > 0) { |
1da177e4 | 1323 | p = &parent->rb_right; |
b2ac2ea6 DB |
1324 | leftmost = false; |
1325 | } else | |
1da177e4 LT |
1326 | p = &parent->rb_left; |
1327 | } | |
1328 | rb_link_node(&epi->rbn, parent, p); | |
b2ac2ea6 | 1329 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
1da177e4 LT |
1330 | } |
1331 | ||
a80a6b85 AM |
1332 | |
1333 | ||
28d82dc1 JB |
1334 | #define PATH_ARR_SIZE 5 |
1335 | /* | |
1336 | * These are the number paths of length 1 to 5, that we are allowing to emanate | |
1337 | * from a single file of interest. For example, we allow 1000 paths of length | |
1338 | * 1, to emanate from each file of interest. This essentially represents the | |
1339 | * potential wakeup paths, which need to be limited in order to avoid massive | |
1340 | * uncontrolled wakeup storms. The common use case should be a single ep which | |
1341 | * is connected to n file sources. In this case each file source has 1 path | |
1342 | * of length 1. Thus, the numbers below should be more than sufficient. These | |
1343 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify | |
1344 | * and delete can't add additional paths. Protected by the epmutex. | |
1345 | */ | |
1346 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; | |
1347 | static int path_count[PATH_ARR_SIZE]; | |
1348 | ||
1349 | static int path_count_inc(int nests) | |
1350 | { | |
93dc6107 JB |
1351 | /* Allow an arbitrary number of depth 1 paths */ |
1352 | if (nests == 0) | |
1353 | return 0; | |
1354 | ||
28d82dc1 JB |
1355 | if (++path_count[nests] > path_limits[nests]) |
1356 | return -1; | |
1357 | return 0; | |
1358 | } | |
1359 | ||
1360 | static void path_count_init(void) | |
1361 | { | |
1362 | int i; | |
1363 | ||
1364 | for (i = 0; i < PATH_ARR_SIZE; i++) | |
1365 | path_count[i] = 0; | |
1366 | } | |
1367 | ||
1368 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) | |
1369 | { | |
1370 | int error = 0; | |
1371 | struct file *file = priv; | |
1372 | struct file *child_file; | |
1373 | struct epitem *epi; | |
1374 | ||
ae10b2b4 JB |
1375 | /* CTL_DEL can remove links here, but that can't increase our count */ |
1376 | rcu_read_lock(); | |
1377 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { | |
28d82dc1 JB |
1378 | child_file = epi->ep->file; |
1379 | if (is_file_epoll(child_file)) { | |
1380 | if (list_empty(&child_file->f_ep_links)) { | |
1381 | if (path_count_inc(call_nests)) { | |
1382 | error = -1; | |
1383 | break; | |
1384 | } | |
1385 | } else { | |
1386 | error = ep_call_nested(&poll_loop_ncalls, | |
28d82dc1 JB |
1387 | reverse_path_check_proc, |
1388 | child_file, child_file, | |
1389 | current); | |
1390 | } | |
1391 | if (error != 0) | |
1392 | break; | |
1393 | } else { | |
1394 | printk(KERN_ERR "reverse_path_check_proc: " | |
1395 | "file is not an ep!\n"); | |
1396 | } | |
1397 | } | |
ae10b2b4 | 1398 | rcu_read_unlock(); |
28d82dc1 JB |
1399 | return error; |
1400 | } | |
1401 | ||
1402 | /** | |
1403 | * reverse_path_check - The tfile_check_list is list of file *, which have | |
1404 | * links that are proposed to be newly added. We need to | |
1405 | * make sure that those added links don't add too many | |
1406 | * paths such that we will spend all our time waking up | |
1407 | * eventpoll objects. | |
1408 | * | |
1409 | * Returns: Returns zero if the proposed links don't create too many paths, | |
1410 | * -1 otherwise. | |
1411 | */ | |
1412 | static int reverse_path_check(void) | |
1413 | { | |
28d82dc1 JB |
1414 | int error = 0; |
1415 | struct file *current_file; | |
1416 | ||
1417 | /* let's call this for all tfiles */ | |
1418 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { | |
28d82dc1 | 1419 | path_count_init(); |
74bdc129 | 1420 | error = ep_call_nested(&poll_loop_ncalls, |
28d82dc1 JB |
1421 | reverse_path_check_proc, current_file, |
1422 | current_file, current); | |
1423 | if (error) | |
1424 | break; | |
1425 | } | |
1426 | return error; | |
1427 | } | |
1428 | ||
4d7e30d9 AH |
1429 | static int ep_create_wakeup_source(struct epitem *epi) |
1430 | { | |
1431 | const char *name; | |
eea1d585 | 1432 | struct wakeup_source *ws; |
4d7e30d9 AH |
1433 | |
1434 | if (!epi->ep->ws) { | |
c8377adf | 1435 | epi->ep->ws = wakeup_source_register(NULL, "eventpoll"); |
4d7e30d9 AH |
1436 | if (!epi->ep->ws) |
1437 | return -ENOMEM; | |
1438 | } | |
1439 | ||
1440 | name = epi->ffd.file->f_path.dentry->d_name.name; | |
c8377adf | 1441 | ws = wakeup_source_register(NULL, name); |
eea1d585 EW |
1442 | |
1443 | if (!ws) | |
4d7e30d9 | 1444 | return -ENOMEM; |
eea1d585 | 1445 | rcu_assign_pointer(epi->ws, ws); |
4d7e30d9 AH |
1446 | |
1447 | return 0; | |
1448 | } | |
1449 | ||
eea1d585 EW |
1450 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
1451 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) | |
4d7e30d9 | 1452 | { |
eea1d585 EW |
1453 | struct wakeup_source *ws = ep_wakeup_source(epi); |
1454 | ||
d6d67e72 | 1455 | RCU_INIT_POINTER(epi->ws, NULL); |
eea1d585 EW |
1456 | |
1457 | /* | |
1458 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is | |
1459 | * used internally by wakeup_source_remove, too (called by | |
1460 | * wakeup_source_unregister), so we cannot use call_rcu | |
1461 | */ | |
1462 | synchronize_rcu(); | |
1463 | wakeup_source_unregister(ws); | |
4d7e30d9 AH |
1464 | } |
1465 | ||
c7ea7630 DL |
1466 | /* |
1467 | * Must be called with "mtx" held. | |
1468 | */ | |
bec1a502 | 1469 | static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, |
67347fe4 | 1470 | struct file *tfile, int fd, int full_check) |
1da177e4 | 1471 | { |
d85e2aa2 AV |
1472 | int error, pwake = 0; |
1473 | __poll_t revents; | |
52bd19f7 | 1474 | long user_watches; |
1da177e4 LT |
1475 | struct epitem *epi; |
1476 | struct ep_pqueue epq; | |
1477 | ||
92e64178 DB |
1478 | lockdep_assert_irqs_enabled(); |
1479 | ||
52bd19f7 RH |
1480 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
1481 | if (unlikely(user_watches >= max_user_watches)) | |
7ef9964e | 1482 | return -ENOSPC; |
e94b1766 | 1483 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
7ef9964e | 1484 | return -ENOMEM; |
1da177e4 LT |
1485 | |
1486 | /* Item initialization follow here ... */ | |
1da177e4 LT |
1487 | INIT_LIST_HEAD(&epi->rdllink); |
1488 | INIT_LIST_HEAD(&epi->fllink); | |
1da177e4 LT |
1489 | INIT_LIST_HEAD(&epi->pwqlist); |
1490 | epi->ep = ep; | |
b030a4dd | 1491 | ep_set_ffd(&epi->ffd, tfile, fd); |
1da177e4 | 1492 | epi->event = *event; |
1da177e4 | 1493 | epi->nwait = 0; |
d47de16c | 1494 | epi->next = EP_UNACTIVE_PTR; |
4d7e30d9 AH |
1495 | if (epi->event.events & EPOLLWAKEUP) { |
1496 | error = ep_create_wakeup_source(epi); | |
1497 | if (error) | |
1498 | goto error_create_wakeup_source; | |
1499 | } else { | |
eea1d585 | 1500 | RCU_INIT_POINTER(epi->ws, NULL); |
4d7e30d9 | 1501 | } |
1da177e4 LT |
1502 | |
1503 | /* Initialize the poll table using the queue callback */ | |
1504 | epq.epi = epi; | |
1505 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
1506 | ||
1507 | /* | |
1508 | * Attach the item to the poll hooks and get current event bits. | |
1509 | * We can safely use the file* here because its usage count has | |
c7ea7630 DL |
1510 | * been increased by the caller of this function. Note that after |
1511 | * this operation completes, the poll callback can start hitting | |
1512 | * the new item. | |
1da177e4 | 1513 | */ |
37b5e521 | 1514 | revents = ep_item_poll(epi, &epq.pt, 1); |
1da177e4 LT |
1515 | |
1516 | /* | |
1517 | * We have to check if something went wrong during the poll wait queue | |
1518 | * install process. Namely an allocation for a wait queue failed due | |
1519 | * high memory pressure. | |
1520 | */ | |
7ef9964e | 1521 | error = -ENOMEM; |
1da177e4 | 1522 | if (epi->nwait < 0) |
7699acd1 | 1523 | goto error_unregister; |
1da177e4 LT |
1524 | |
1525 | /* Add the current item to the list of active epoll hook for this file */ | |
68499914 | 1526 | spin_lock(&tfile->f_lock); |
ae10b2b4 | 1527 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); |
68499914 | 1528 | spin_unlock(&tfile->f_lock); |
1da177e4 | 1529 | |
c7ea7630 DL |
1530 | /* |
1531 | * Add the current item to the RB tree. All RB tree operations are | |
1532 | * protected by "mtx", and ep_insert() is called with "mtx" held. | |
1533 | */ | |
1da177e4 LT |
1534 | ep_rbtree_insert(ep, epi); |
1535 | ||
28d82dc1 JB |
1536 | /* now check if we've created too many backpaths */ |
1537 | error = -EINVAL; | |
67347fe4 | 1538 | if (full_check && reverse_path_check()) |
28d82dc1 JB |
1539 | goto error_remove_epi; |
1540 | ||
c7ea7630 | 1541 | /* We have to drop the new item inside our item list to keep track of it */ |
a218cc49 | 1542 | write_lock_irq(&ep->lock); |
c7ea7630 | 1543 | |
bf3b9f63 SS |
1544 | /* record NAPI ID of new item if present */ |
1545 | ep_set_busy_poll_napi_id(epi); | |
1546 | ||
1da177e4 | 1547 | /* If the file is already "ready" we drop it inside the ready list */ |
992991c0 | 1548 | if (revents && !ep_is_linked(epi)) { |
1da177e4 | 1549 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1550 | ep_pm_stay_awake(epi); |
1da177e4 LT |
1551 | |
1552 | /* Notify waiting tasks that events are available */ | |
1553 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1554 | wake_up(&ep->wq); |
1da177e4 LT |
1555 | if (waitqueue_active(&ep->poll_wait)) |
1556 | pwake++; | |
1557 | } | |
1558 | ||
a218cc49 | 1559 | write_unlock_irq(&ep->lock); |
1da177e4 | 1560 | |
52bd19f7 | 1561 | atomic_long_inc(&ep->user->epoll_watches); |
7ef9964e | 1562 | |
1da177e4 LT |
1563 | /* We have to call this outside the lock */ |
1564 | if (pwake) | |
5071f97e | 1565 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1566 | |
1da177e4 LT |
1567 | return 0; |
1568 | ||
28d82dc1 JB |
1569 | error_remove_epi: |
1570 | spin_lock(&tfile->f_lock); | |
ae10b2b4 | 1571 | list_del_rcu(&epi->fllink); |
28d82dc1 JB |
1572 | spin_unlock(&tfile->f_lock); |
1573 | ||
b2ac2ea6 | 1574 | rb_erase_cached(&epi->rbn, &ep->rbr); |
28d82dc1 | 1575 | |
7699acd1 | 1576 | error_unregister: |
1da177e4 LT |
1577 | ep_unregister_pollwait(ep, epi); |
1578 | ||
1579 | /* | |
1580 | * We need to do this because an event could have been arrived on some | |
67647d0f DL |
1581 | * allocated wait queue. Note that we don't care about the ep->ovflist |
1582 | * list, since that is used/cleaned only inside a section bound by "mtx". | |
1583 | * And ep_insert() is called with "mtx" held. | |
1da177e4 | 1584 | */ |
a218cc49 | 1585 | write_lock_irq(&ep->lock); |
992991c0 | 1586 | if (ep_is_linked(epi)) |
6192bd53 | 1587 | list_del_init(&epi->rdllink); |
a218cc49 | 1588 | write_unlock_irq(&ep->lock); |
1da177e4 | 1589 | |
eea1d585 | 1590 | wakeup_source_unregister(ep_wakeup_source(epi)); |
4d7e30d9 AH |
1591 | |
1592 | error_create_wakeup_source: | |
b030a4dd | 1593 | kmem_cache_free(epi_cache, epi); |
7ef9964e | 1594 | |
1da177e4 LT |
1595 | return error; |
1596 | } | |
1597 | ||
1da177e4 LT |
1598 | /* |
1599 | * Modify the interest event mask by dropping an event if the new mask | |
c7ea7630 | 1600 | * has a match in the current file status. Must be called with "mtx" held. |
1da177e4 | 1601 | */ |
bec1a502 AV |
1602 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, |
1603 | const struct epoll_event *event) | |
1da177e4 LT |
1604 | { |
1605 | int pwake = 0; | |
626cf236 HV |
1606 | poll_table pt; |
1607 | ||
92e64178 DB |
1608 | lockdep_assert_irqs_enabled(); |
1609 | ||
626cf236 | 1610 | init_poll_funcptr(&pt, NULL); |
1da177e4 LT |
1611 | |
1612 | /* | |
e057e15f TB |
1613 | * Set the new event interest mask before calling f_op->poll(); |
1614 | * otherwise we might miss an event that happens between the | |
1615 | * f_op->poll() call and the new event set registering. | |
1da177e4 | 1616 | */ |
128dd175 | 1617 | epi->event.events = event->events; /* need barrier below */ |
e057e15f | 1618 | epi->event.data = event->data; /* protected by mtx */ |
4d7e30d9 | 1619 | if (epi->event.events & EPOLLWAKEUP) { |
eea1d585 | 1620 | if (!ep_has_wakeup_source(epi)) |
4d7e30d9 | 1621 | ep_create_wakeup_source(epi); |
eea1d585 | 1622 | } else if (ep_has_wakeup_source(epi)) { |
4d7e30d9 AH |
1623 | ep_destroy_wakeup_source(epi); |
1624 | } | |
1da177e4 | 1625 | |
128dd175 EW |
1626 | /* |
1627 | * The following barrier has two effects: | |
1628 | * | |
1629 | * 1) Flush epi changes above to other CPUs. This ensures | |
1630 | * we do not miss events from ep_poll_callback if an | |
1631 | * event occurs immediately after we call f_op->poll(). | |
a218cc49 | 1632 | * We need this because we did not take ep->lock while |
128dd175 | 1633 | * changing epi above (but ep_poll_callback does take |
a218cc49 | 1634 | * ep->lock). |
128dd175 EW |
1635 | * |
1636 | * 2) We also need to ensure we do not miss _past_ events | |
1637 | * when calling f_op->poll(). This barrier also | |
1638 | * pairs with the barrier in wq_has_sleeper (see | |
1639 | * comments for wq_has_sleeper). | |
1640 | * | |
1641 | * This barrier will now guarantee ep_poll_callback or f_op->poll | |
1642 | * (or both) will notice the readiness of an item. | |
1643 | */ | |
1644 | smp_mb(); | |
1645 | ||
1da177e4 LT |
1646 | /* |
1647 | * Get current event bits. We can safely use the file* here because | |
1648 | * its usage count has been increased by the caller of this function. | |
c7ea7630 | 1649 | * If the item is "hot" and it is not registered inside the ready |
67647d0f | 1650 | * list, push it inside. |
1da177e4 | 1651 | */ |
69112736 | 1652 | if (ep_item_poll(epi, &pt, 1)) { |
a218cc49 | 1653 | write_lock_irq(&ep->lock); |
992991c0 | 1654 | if (!ep_is_linked(epi)) { |
c7ea7630 | 1655 | list_add_tail(&epi->rdllink, &ep->rdllist); |
eea1d585 | 1656 | ep_pm_stay_awake(epi); |
c7ea7630 DL |
1657 | |
1658 | /* Notify waiting tasks that events are available */ | |
1659 | if (waitqueue_active(&ep->wq)) | |
a218cc49 | 1660 | wake_up(&ep->wq); |
c7ea7630 DL |
1661 | if (waitqueue_active(&ep->poll_wait)) |
1662 | pwake++; | |
7699acd1 | 1663 | } |
a218cc49 | 1664 | write_unlock_irq(&ep->lock); |
7699acd1 | 1665 | } |
1da177e4 | 1666 | |
7699acd1 DL |
1667 | /* We have to call this outside the lock */ |
1668 | if (pwake) | |
5071f97e | 1669 | ep_poll_safewake(&ep->poll_wait); |
1da177e4 | 1670 | |
7699acd1 | 1671 | return 0; |
1da177e4 LT |
1672 | } |
1673 | ||
d85e2aa2 | 1674 | static __poll_t ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
296e236e | 1675 | void *priv) |
1da177e4 | 1676 | { |
5071f97e | 1677 | struct ep_send_events_data *esed = priv; |
d85e2aa2 | 1678 | __poll_t revents; |
4e0982a0 DB |
1679 | struct epitem *epi, *tmp; |
1680 | struct epoll_event __user *uevent = esed->events; | |
eea1d585 | 1681 | struct wakeup_source *ws; |
626cf236 HV |
1682 | poll_table pt; |
1683 | ||
1684 | init_poll_funcptr(&pt, NULL); | |
4e0982a0 | 1685 | esed->res = 0; |
1da177e4 | 1686 | |
296e236e | 1687 | /* |
5071f97e DL |
1688 | * We can loop without lock because we are passed a task private list. |
1689 | * Items cannot vanish during the loop because ep_scan_ready_list() is | |
1690 | * holding "mtx" during this call. | |
296e236e | 1691 | */ |
21877e1a DB |
1692 | lockdep_assert_held(&ep->mtx); |
1693 | ||
4e0982a0 DB |
1694 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
1695 | if (esed->res >= esed->maxevents) | |
1696 | break; | |
d47de16c | 1697 | |
4d7e30d9 AH |
1698 | /* |
1699 | * Activate ep->ws before deactivating epi->ws to prevent | |
1700 | * triggering auto-suspend here (in case we reactive epi->ws | |
1701 | * below). | |
1702 | * | |
1703 | * This could be rearranged to delay the deactivation of epi->ws | |
1704 | * instead, but then epi->ws would temporarily be out of sync | |
1705 | * with ep_is_linked(). | |
1706 | */ | |
eea1d585 EW |
1707 | ws = ep_wakeup_source(epi); |
1708 | if (ws) { | |
1709 | if (ws->active) | |
1710 | __pm_stay_awake(ep->ws); | |
1711 | __pm_relax(ws); | |
1712 | } | |
1713 | ||
d47de16c | 1714 | list_del_init(&epi->rdllink); |
1da177e4 | 1715 | |
296e236e | 1716 | /* |
5071f97e DL |
1717 | * If the event mask intersect the caller-requested one, |
1718 | * deliver the event to userspace. Again, ep_scan_ready_list() | |
4e0982a0 | 1719 | * is holding ep->mtx, so no operations coming from userspace |
5071f97e | 1720 | * can change the item. |
296e236e | 1721 | */ |
4e0982a0 DB |
1722 | revents = ep_item_poll(epi, &pt, 1); |
1723 | if (!revents) | |
1724 | continue; | |
1725 | ||
1726 | if (__put_user(revents, &uevent->events) || | |
1727 | __put_user(epi->event.data, &uevent->data)) { | |
1728 | list_add(&epi->rdllink, head); | |
1729 | ep_pm_stay_awake(epi); | |
1730 | if (!esed->res) | |
1731 | esed->res = -EFAULT; | |
1732 | return 0; | |
1733 | } | |
1734 | esed->res++; | |
1735 | uevent++; | |
1736 | if (epi->event.events & EPOLLONESHOT) | |
1737 | epi->event.events &= EP_PRIVATE_BITS; | |
1738 | else if (!(epi->event.events & EPOLLET)) { | |
1739 | /* | |
1740 | * If this file has been added with Level | |
1741 | * Trigger mode, we need to insert back inside | |
1742 | * the ready list, so that the next call to | |
1743 | * epoll_wait() will check again the events | |
1744 | * availability. At this point, no one can insert | |
1745 | * into ep->rdllist besides us. The epoll_ctl() | |
1746 | * callers are locked out by | |
1747 | * ep_scan_ready_list() holding "mtx" and the | |
1748 | * poll callback will queue them in ep->ovflist. | |
1749 | */ | |
1750 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1751 | ep_pm_stay_awake(epi); | |
296e236e DL |
1752 | } |
1753 | } | |
5071f97e | 1754 | |
d7ebbe46 | 1755 | return 0; |
5071f97e | 1756 | } |
d47de16c | 1757 | |
296e236e DL |
1758 | static int ep_send_events(struct eventpoll *ep, |
1759 | struct epoll_event __user *events, int maxevents) | |
5071f97e DL |
1760 | { |
1761 | struct ep_send_events_data esed; | |
1da177e4 | 1762 | |
5071f97e DL |
1763 | esed.maxevents = maxevents; |
1764 | esed.events = events; | |
6192bd53 | 1765 | |
d7ebbe46 AV |
1766 | ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false); |
1767 | return esed.res; | |
1da177e4 LT |
1768 | } |
1769 | ||
766b9f92 | 1770 | static inline struct timespec64 ep_set_mstimeout(long ms) |
0781b909 | 1771 | { |
766b9f92 | 1772 | struct timespec64 now, ts = { |
0781b909 ED |
1773 | .tv_sec = ms / MSEC_PER_SEC, |
1774 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), | |
1775 | }; | |
1776 | ||
766b9f92 DD |
1777 | ktime_get_ts64(&now); |
1778 | return timespec64_add_safe(now, ts); | |
0781b909 ED |
1779 | } |
1780 | ||
f4d93ad7 SB |
1781 | /** |
1782 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied | |
1783 | * event buffer. | |
1784 | * | |
1785 | * @ep: Pointer to the eventpoll context. | |
1786 | * @events: Pointer to the userspace buffer where the ready events should be | |
1787 | * stored. | |
1788 | * @maxevents: Size (in terms of number of events) of the caller event buffer. | |
1789 | * @timeout: Maximum timeout for the ready events fetch operation, in | |
1790 | * milliseconds. If the @timeout is zero, the function will not block, | |
1791 | * while if the @timeout is less than zero, the function will block | |
1792 | * until at least one event has been retrieved (or an error | |
1793 | * occurred). | |
1794 | * | |
1795 | * Returns: Returns the number of ready events which have been fetched, or an | |
1796 | * error code, in case of error. | |
1797 | */ | |
1da177e4 LT |
1798 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
1799 | int maxevents, long timeout) | |
1800 | { | |
f4d93ad7 | 1801 | int res = 0, eavail, timed_out = 0; |
da8b44d5 | 1802 | u64 slack = 0; |
86c05179 | 1803 | bool waiter = false; |
ac6424b9 | 1804 | wait_queue_entry_t wait; |
95aac7b1 SB |
1805 | ktime_t expires, *to = NULL; |
1806 | ||
679abf38 DB |
1807 | lockdep_assert_irqs_enabled(); |
1808 | ||
95aac7b1 | 1809 | if (timeout > 0) { |
766b9f92 | 1810 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
0781b909 | 1811 | |
95aac7b1 SB |
1812 | slack = select_estimate_accuracy(&end_time); |
1813 | to = &expires; | |
766b9f92 | 1814 | *to = timespec64_to_ktime(end_time); |
95aac7b1 | 1815 | } else if (timeout == 0) { |
f4d93ad7 SB |
1816 | /* |
1817 | * Avoid the unnecessary trip to the wait queue loop, if the | |
c5a282e9 DB |
1818 | * caller specified a non blocking operation. We still need |
1819 | * lock because we could race and not see an epi being added | |
1820 | * to the ready list while in irq callback. Thus incorrectly | |
1821 | * returning 0 back to userspace. | |
f4d93ad7 | 1822 | */ |
95aac7b1 | 1823 | timed_out = 1; |
c5a282e9 | 1824 | |
a218cc49 | 1825 | write_lock_irq(&ep->lock); |
c5a282e9 | 1826 | eavail = ep_events_available(ep); |
a218cc49 | 1827 | write_unlock_irq(&ep->lock); |
c5a282e9 | 1828 | |
35cff1a6 | 1829 | goto send_events; |
95aac7b1 | 1830 | } |
1da177e4 | 1831 | |
f4d93ad7 | 1832 | fetch_events: |
bf3b9f63 SS |
1833 | |
1834 | if (!ep_events_available(ep)) | |
1835 | ep_busy_loop(ep, timed_out); | |
1836 | ||
c5a282e9 DB |
1837 | eavail = ep_events_available(ep); |
1838 | if (eavail) | |
35cff1a6 | 1839 | goto send_events; |
1da177e4 | 1840 | |
c5a282e9 DB |
1841 | /* |
1842 | * Busy poll timed out. Drop NAPI ID for now, we can add | |
1843 | * it back in when we have moved a socket with a valid NAPI | |
1844 | * ID onto the ready list. | |
1845 | */ | |
1846 | ep_reset_busy_poll_napi_id(ep); | |
bf3b9f63 | 1847 | |
c5a282e9 | 1848 | /* |
86c05179 DB |
1849 | * We don't have any available event to return to the caller. We need |
1850 | * to sleep here, and we will be woken by ep_poll_callback() when events | |
1851 | * become available. | |
c5a282e9 | 1852 | */ |
86c05179 DB |
1853 | if (!waiter) { |
1854 | waiter = true; | |
1da177e4 | 1855 | init_waitqueue_entry(&wait, current); |
1da177e4 | 1856 | |
86c05179 DB |
1857 | spin_lock_irq(&ep->wq.lock); |
1858 | __add_wait_queue_exclusive(&ep->wq, &wait); | |
1859 | spin_unlock_irq(&ep->wq.lock); | |
1860 | } | |
1da177e4 | 1861 | |
c5a282e9 | 1862 | for (;;) { |
bf3b9f63 | 1863 | /* |
c5a282e9 DB |
1864 | * We don't want to sleep if the ep_poll_callback() sends us |
1865 | * a wakeup in between. That's why we set the task state | |
1866 | * to TASK_INTERRUPTIBLE before doing the checks. | |
bf3b9f63 | 1867 | */ |
c5a282e9 | 1868 | set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 | 1869 | /* |
c5a282e9 DB |
1870 | * Always short-circuit for fatal signals to allow |
1871 | * threads to make a timely exit without the chance of | |
1872 | * finding more events available and fetching | |
1873 | * repeatedly. | |
1da177e4 | 1874 | */ |
c5a282e9 DB |
1875 | if (fatal_signal_pending(current)) { |
1876 | res = -EINTR; | |
1877 | break; | |
1878 | } | |
95aac7b1 | 1879 | |
abc610e0 DB |
1880 | eavail = ep_events_available(ep); |
1881 | if (eavail) | |
c5a282e9 DB |
1882 | break; |
1883 | if (signal_pending(current)) { | |
1884 | res = -EINTR; | |
1885 | break; | |
1da177e4 | 1886 | } |
1da177e4 | 1887 | |
abc610e0 | 1888 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) { |
c5a282e9 | 1889 | timed_out = 1; |
abc610e0 DB |
1890 | break; |
1891 | } | |
1da177e4 | 1892 | } |
1da177e4 | 1893 | |
c5a282e9 | 1894 | __set_current_state(TASK_RUNNING); |
1da177e4 | 1895 | |
35cff1a6 | 1896 | send_events: |
1da177e4 LT |
1897 | /* |
1898 | * Try to transfer events to user space. In case we get 0 events and | |
1899 | * there's still timeout left over, we go trying again in search of | |
1900 | * more luck. | |
1901 | */ | |
1902 | if (!res && eavail && | |
95aac7b1 | 1903 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
f4d93ad7 | 1904 | goto fetch_events; |
1da177e4 | 1905 | |
86c05179 DB |
1906 | if (waiter) { |
1907 | spin_lock_irq(&ep->wq.lock); | |
1908 | __remove_wait_queue(&ep->wq, &wait); | |
1909 | spin_unlock_irq(&ep->wq.lock); | |
1910 | } | |
1911 | ||
1da177e4 LT |
1912 | return res; |
1913 | } | |
1914 | ||
22bacca4 DL |
1915 | /** |
1916 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() | |
1917 | * API, to verify that adding an epoll file inside another | |
1918 | * epoll structure, does not violate the constraints, in | |
1919 | * terms of closed loops, or too deep chains (which can | |
1920 | * result in excessive stack usage). | |
1921 | * | |
1922 | * @priv: Pointer to the epoll file to be currently checked. | |
1923 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll | |
1924 | * data structure pointer. | |
1925 | * @call_nests: Current dept of the @ep_call_nested() call stack. | |
1926 | * | |
1927 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1928 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1929 | */ | |
1930 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) | |
1931 | { | |
1932 | int error = 0; | |
1933 | struct file *file = priv; | |
1934 | struct eventpoll *ep = file->private_data; | |
28d82dc1 | 1935 | struct eventpoll *ep_tovisit; |
22bacca4 DL |
1936 | struct rb_node *rbp; |
1937 | struct epitem *epi; | |
1938 | ||
d8805e63 | 1939 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
28d82dc1 JB |
1940 | ep->visited = 1; |
1941 | list_add(&ep->visited_list_link, &visited_list); | |
b2ac2ea6 | 1942 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
22bacca4 DL |
1943 | epi = rb_entry(rbp, struct epitem, rbn); |
1944 | if (unlikely(is_file_epoll(epi->ffd.file))) { | |
28d82dc1 JB |
1945 | ep_tovisit = epi->ffd.file->private_data; |
1946 | if (ep_tovisit->visited) | |
1947 | continue; | |
74bdc129 | 1948 | error = ep_call_nested(&poll_loop_ncalls, |
28d82dc1 JB |
1949 | ep_loop_check_proc, epi->ffd.file, |
1950 | ep_tovisit, current); | |
22bacca4 DL |
1951 | if (error != 0) |
1952 | break; | |
28d82dc1 JB |
1953 | } else { |
1954 | /* | |
1955 | * If we've reached a file that is not associated with | |
1956 | * an ep, then we need to check if the newly added | |
1957 | * links are going to add too many wakeup paths. We do | |
1958 | * this by adding it to the tfile_check_list, if it's | |
1959 | * not already there, and calling reverse_path_check() | |
1960 | * during ep_insert(). | |
1961 | */ | |
1962 | if (list_empty(&epi->ffd.file->f_tfile_llink)) | |
1963 | list_add(&epi->ffd.file->f_tfile_llink, | |
1964 | &tfile_check_list); | |
22bacca4 DL |
1965 | } |
1966 | } | |
1967 | mutex_unlock(&ep->mtx); | |
1968 | ||
1969 | return error; | |
1970 | } | |
1971 | ||
1972 | /** | |
1973 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) | |
1974 | * another epoll file (represented by @ep) does not create | |
1975 | * closed loops or too deep chains. | |
1976 | * | |
1977 | * @ep: Pointer to the epoll private data structure. | |
1978 | * @file: Pointer to the epoll file to be checked. | |
1979 | * | |
1980 | * Returns: Returns zero if adding the epoll @file inside current epoll | |
1981 | * structure @ep does not violate the constraints, or -1 otherwise. | |
1982 | */ | |
1983 | static int ep_loop_check(struct eventpoll *ep, struct file *file) | |
1984 | { | |
28d82dc1 JB |
1985 | int ret; |
1986 | struct eventpoll *ep_cur, *ep_next; | |
1987 | ||
74bdc129 | 1988 | ret = ep_call_nested(&poll_loop_ncalls, |
22bacca4 | 1989 | ep_loop_check_proc, file, ep, current); |
28d82dc1 JB |
1990 | /* clear visited list */ |
1991 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, | |
1992 | visited_list_link) { | |
1993 | ep_cur->visited = 0; | |
1994 | list_del(&ep_cur->visited_list_link); | |
1995 | } | |
1996 | return ret; | |
1997 | } | |
1998 | ||
1999 | static void clear_tfile_check_list(void) | |
2000 | { | |
2001 | struct file *file; | |
2002 | ||
2003 | /* first clear the tfile_check_list */ | |
2004 | while (!list_empty(&tfile_check_list)) { | |
2005 | file = list_first_entry(&tfile_check_list, struct file, | |
2006 | f_tfile_llink); | |
2007 | list_del_init(&file->f_tfile_llink); | |
2008 | } | |
2009 | INIT_LIST_HEAD(&tfile_check_list); | |
22bacca4 DL |
2010 | } |
2011 | ||
7699acd1 | 2012 | /* |
523723bb | 2013 | * Open an eventpoll file descriptor. |
7699acd1 | 2014 | */ |
791eb22e | 2015 | static int do_epoll_create(int flags) |
7699acd1 | 2016 | { |
28d82dc1 | 2017 | int error, fd; |
bb57c3ed | 2018 | struct eventpoll *ep = NULL; |
28d82dc1 | 2019 | struct file *file; |
7699acd1 | 2020 | |
e38b36f3 UD |
2021 | /* Check the EPOLL_* constant for consistency. */ |
2022 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); | |
2023 | ||
296e236e DL |
2024 | if (flags & ~EPOLL_CLOEXEC) |
2025 | return -EINVAL; | |
7699acd1 | 2026 | /* |
bb57c3ed | 2027 | * Create the internal data structure ("struct eventpoll"). |
7699acd1 | 2028 | */ |
9fe5ad9c | 2029 | error = ep_alloc(&ep); |
bb57c3ed DL |
2030 | if (error < 0) |
2031 | return error; | |
7699acd1 DL |
2032 | /* |
2033 | * Creates all the items needed to setup an eventpoll file. That is, | |
2030a42c | 2034 | * a file structure and a free file descriptor. |
7699acd1 | 2035 | */ |
28d82dc1 JB |
2036 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
2037 | if (fd < 0) { | |
2038 | error = fd; | |
2039 | goto out_free_ep; | |
2040 | } | |
2041 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, | |
628ff7c1 | 2042 | O_RDWR | (flags & O_CLOEXEC)); |
28d82dc1 JB |
2043 | if (IS_ERR(file)) { |
2044 | error = PTR_ERR(file); | |
2045 | goto out_free_fd; | |
2046 | } | |
28d82dc1 | 2047 | ep->file = file; |
98022748 | 2048 | fd_install(fd, file); |
28d82dc1 JB |
2049 | return fd; |
2050 | ||
2051 | out_free_fd: | |
2052 | put_unused_fd(fd); | |
2053 | out_free_ep: | |
2054 | ep_free(ep); | |
bb57c3ed | 2055 | return error; |
7699acd1 DL |
2056 | } |
2057 | ||
791eb22e DB |
2058 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
2059 | { | |
2060 | return do_epoll_create(flags); | |
2061 | } | |
2062 | ||
5a8a82b1 | 2063 | SYSCALL_DEFINE1(epoll_create, int, size) |
a0998b50 | 2064 | { |
bfe3891a | 2065 | if (size <= 0) |
9fe5ad9c UD |
2066 | return -EINVAL; |
2067 | ||
791eb22e | 2068 | return do_epoll_create(0); |
a0998b50 UD |
2069 | } |
2070 | ||
39220e8d JA |
2071 | static inline int epoll_mutex_lock(struct mutex *mutex, int depth, |
2072 | bool nonblock) | |
2073 | { | |
2074 | if (!nonblock) { | |
2075 | mutex_lock_nested(mutex, depth); | |
2076 | return 0; | |
2077 | } | |
2078 | if (mutex_trylock(mutex)) | |
2079 | return 0; | |
2080 | return -EAGAIN; | |
2081 | } | |
2082 | ||
2083 | int do_epoll_ctl(int epfd, int op, int fd, struct epoll_event *epds, | |
2084 | bool nonblock) | |
7699acd1 DL |
2085 | { |
2086 | int error; | |
67347fe4 | 2087 | int full_check = 0; |
7e3fb584 | 2088 | struct fd f, tf; |
7699acd1 DL |
2089 | struct eventpoll *ep; |
2090 | struct epitem *epi; | |
67347fe4 | 2091 | struct eventpoll *tep = NULL; |
7699acd1 | 2092 | |
7699acd1 | 2093 | error = -EBADF; |
7e3fb584 AV |
2094 | f = fdget(epfd); |
2095 | if (!f.file) | |
7699acd1 DL |
2096 | goto error_return; |
2097 | ||
2098 | /* Get the "struct file *" for the target file */ | |
7e3fb584 AV |
2099 | tf = fdget(fd); |
2100 | if (!tf.file) | |
7699acd1 DL |
2101 | goto error_fput; |
2102 | ||
2103 | /* The target file descriptor must support poll */ | |
2104 | error = -EPERM; | |
9965ed17 | 2105 | if (!file_can_poll(tf.file)) |
7699acd1 DL |
2106 | goto error_tgt_fput; |
2107 | ||
4d7e30d9 | 2108 | /* Check if EPOLLWAKEUP is allowed */ |
c680e41b | 2109 | if (ep_op_has_event(op)) |
58e41a44 | 2110 | ep_take_care_of_epollwakeup(epds); |
4d7e30d9 | 2111 | |
7699acd1 DL |
2112 | /* |
2113 | * We have to check that the file structure underneath the file descriptor | |
2114 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
2115 | * adding an epoll file descriptor inside itself. | |
2116 | */ | |
2117 | error = -EINVAL; | |
7e3fb584 | 2118 | if (f.file == tf.file || !is_file_epoll(f.file)) |
7699acd1 DL |
2119 | goto error_tgt_fput; |
2120 | ||
df0108c5 JB |
2121 | /* |
2122 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, | |
2123 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. | |
2124 | * Also, we do not currently supported nested exclusive wakeups. | |
2125 | */ | |
58e41a44 | 2126 | if (ep_op_has_event(op) && (epds->events & EPOLLEXCLUSIVE)) { |
b6a515c8 JB |
2127 | if (op == EPOLL_CTL_MOD) |
2128 | goto error_tgt_fput; | |
2129 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || | |
58e41a44 | 2130 | (epds->events & ~EPOLLEXCLUSIVE_OK_BITS))) |
b6a515c8 JB |
2131 | goto error_tgt_fput; |
2132 | } | |
df0108c5 | 2133 | |
7699acd1 DL |
2134 | /* |
2135 | * At this point it is safe to assume that the "private_data" contains | |
2136 | * our own data structure. | |
2137 | */ | |
7e3fb584 | 2138 | ep = f.file->private_data; |
7699acd1 | 2139 | |
22bacca4 DL |
2140 | /* |
2141 | * When we insert an epoll file descriptor, inside another epoll file | |
2142 | * descriptor, there is the change of creating closed loops, which are | |
28d82dc1 JB |
2143 | * better be handled here, than in more critical paths. While we are |
2144 | * checking for loops we also determine the list of files reachable | |
2145 | * and hang them on the tfile_check_list, so we can check that we | |
2146 | * haven't created too many possible wakeup paths. | |
22bacca4 | 2147 | * |
67347fe4 JB |
2148 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
2149 | * the epoll file descriptor is attaching directly to a wakeup source, | |
2150 | * unless the epoll file descriptor is nested. The purpose of taking the | |
2151 | * 'epmutex' on add is to prevent complex toplogies such as loops and | |
2152 | * deep wakeup paths from forming in parallel through multiple | |
2153 | * EPOLL_CTL_ADD operations. | |
22bacca4 | 2154 | */ |
39220e8d JA |
2155 | error = epoll_mutex_lock(&ep->mtx, 0, nonblock); |
2156 | if (error) | |
2157 | goto error_tgt_fput; | |
28d82dc1 | 2158 | if (op == EPOLL_CTL_ADD) { |
67347fe4 JB |
2159 | if (!list_empty(&f.file->f_ep_links) || |
2160 | is_file_epoll(tf.file)) { | |
67347fe4 | 2161 | mutex_unlock(&ep->mtx); |
39220e8d JA |
2162 | error = epoll_mutex_lock(&epmutex, 0, nonblock); |
2163 | if (error) | |
2164 | goto error_tgt_fput; | |
2165 | full_check = 1; | |
67347fe4 JB |
2166 | if (is_file_epoll(tf.file)) { |
2167 | error = -ELOOP; | |
2168 | if (ep_loop_check(ep, tf.file) != 0) { | |
2169 | clear_tfile_check_list(); | |
2170 | goto error_tgt_fput; | |
2171 | } | |
2172 | } else | |
2173 | list_add(&tf.file->f_tfile_llink, | |
2174 | &tfile_check_list); | |
39220e8d JA |
2175 | error = epoll_mutex_lock(&ep->mtx, 0, nonblock); |
2176 | if (error) { | |
2177 | out_del: | |
2178 | list_del(&tf.file->f_tfile_llink); | |
2179 | goto error_tgt_fput; | |
2180 | } | |
67347fe4 JB |
2181 | if (is_file_epoll(tf.file)) { |
2182 | tep = tf.file->private_data; | |
39220e8d JA |
2183 | error = epoll_mutex_lock(&tep->mtx, 1, nonblock); |
2184 | if (error) { | |
2185 | mutex_unlock(&ep->mtx); | |
2186 | goto out_del; | |
2187 | } | |
13d51807 | 2188 | } |
67347fe4 JB |
2189 | } |
2190 | } | |
7699acd1 | 2191 | |
67647d0f DL |
2192 | /* |
2193 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" | |
2194 | * above, we can be sure to be able to use the item looked up by | |
2195 | * ep_find() till we release the mutex. | |
2196 | */ | |
7e3fb584 | 2197 | epi = ep_find(ep, tf.file, fd); |
7699acd1 DL |
2198 | |
2199 | error = -EINVAL; | |
2200 | switch (op) { | |
2201 | case EPOLL_CTL_ADD: | |
2202 | if (!epi) { | |
58e41a44 JA |
2203 | epds->events |= EPOLLERR | EPOLLHUP; |
2204 | error = ep_insert(ep, epds, tf.file, fd, full_check); | |
7699acd1 DL |
2205 | } else |
2206 | error = -EEXIST; | |
67347fe4 JB |
2207 | if (full_check) |
2208 | clear_tfile_check_list(); | |
7699acd1 DL |
2209 | break; |
2210 | case EPOLL_CTL_DEL: | |
2211 | if (epi) | |
2212 | error = ep_remove(ep, epi); | |
2213 | else | |
2214 | error = -ENOENT; | |
2215 | break; | |
2216 | case EPOLL_CTL_MOD: | |
2217 | if (epi) { | |
b6a515c8 | 2218 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
58e41a44 JA |
2219 | epds->events |= EPOLLERR | EPOLLHUP; |
2220 | error = ep_modify(ep, epi, epds); | |
b6a515c8 | 2221 | } |
7699acd1 DL |
2222 | } else |
2223 | error = -ENOENT; | |
2224 | break; | |
2225 | } | |
67347fe4 JB |
2226 | if (tep != NULL) |
2227 | mutex_unlock(&tep->mtx); | |
d47de16c | 2228 | mutex_unlock(&ep->mtx); |
7699acd1 DL |
2229 | |
2230 | error_tgt_fput: | |
67347fe4 | 2231 | if (full_check) |
22bacca4 DL |
2232 | mutex_unlock(&epmutex); |
2233 | ||
7e3fb584 | 2234 | fdput(tf); |
7699acd1 | 2235 | error_fput: |
7e3fb584 | 2236 | fdput(f); |
7699acd1 | 2237 | error_return: |
7699acd1 DL |
2238 | |
2239 | return error; | |
2240 | } | |
2241 | ||
58e41a44 JA |
2242 | /* |
2243 | * The following function implements the controller interface for | |
2244 | * the eventpoll file that enables the insertion/removal/change of | |
2245 | * file descriptors inside the interest set. | |
2246 | */ | |
2247 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, | |
2248 | struct epoll_event __user *, event) | |
2249 | { | |
2250 | struct epoll_event epds; | |
2251 | ||
2252 | if (ep_op_has_event(op) && | |
2253 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
2254 | return -EFAULT; | |
2255 | ||
39220e8d | 2256 | return do_epoll_ctl(epfd, op, fd, &epds, false); |
58e41a44 JA |
2257 | } |
2258 | ||
7699acd1 DL |
2259 | /* |
2260 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2261 | * part of the user space epoll_wait(2). | |
2262 | */ | |
791eb22e DB |
2263 | static int do_epoll_wait(int epfd, struct epoll_event __user *events, |
2264 | int maxevents, int timeout) | |
7699acd1 | 2265 | { |
2903ff01 AV |
2266 | int error; |
2267 | struct fd f; | |
7699acd1 DL |
2268 | struct eventpoll *ep; |
2269 | ||
7699acd1 DL |
2270 | /* The maximum number of event must be greater than zero */ |
2271 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) | |
2272 | return -EINVAL; | |
2273 | ||
2274 | /* Verify that the area passed by the user is writeable */ | |
96d4f267 | 2275 | if (!access_ok(events, maxevents * sizeof(struct epoll_event))) |
2903ff01 | 2276 | return -EFAULT; |
7699acd1 DL |
2277 | |
2278 | /* Get the "struct file *" for the eventpoll file */ | |
2903ff01 AV |
2279 | f = fdget(epfd); |
2280 | if (!f.file) | |
2281 | return -EBADF; | |
7699acd1 DL |
2282 | |
2283 | /* | |
2284 | * We have to check that the file structure underneath the fd | |
2285 | * the user passed to us _is_ an eventpoll file. | |
2286 | */ | |
2287 | error = -EINVAL; | |
2903ff01 | 2288 | if (!is_file_epoll(f.file)) |
7699acd1 DL |
2289 | goto error_fput; |
2290 | ||
2291 | /* | |
2292 | * At this point it is safe to assume that the "private_data" contains | |
2293 | * our own data structure. | |
2294 | */ | |
2903ff01 | 2295 | ep = f.file->private_data; |
7699acd1 DL |
2296 | |
2297 | /* Time to fish for events ... */ | |
2298 | error = ep_poll(ep, events, maxevents, timeout); | |
2299 | ||
2300 | error_fput: | |
2903ff01 | 2301 | fdput(f); |
7699acd1 DL |
2302 | return error; |
2303 | } | |
2304 | ||
791eb22e DB |
2305 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
2306 | int, maxevents, int, timeout) | |
2307 | { | |
2308 | return do_epoll_wait(epfd, events, maxevents, timeout); | |
2309 | } | |
2310 | ||
7699acd1 DL |
2311 | /* |
2312 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
2313 | * part of the user space epoll_pwait(2). | |
2314 | */ | |
5a8a82b1 HC |
2315 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
2316 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, | |
2317 | size_t, sigsetsize) | |
7699acd1 DL |
2318 | { |
2319 | int error; | |
7699acd1 DL |
2320 | |
2321 | /* | |
2322 | * If the caller wants a certain signal mask to be set during the wait, | |
2323 | * we apply it here. | |
2324 | */ | |
b772434b | 2325 | error = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2326 | if (error) |
2327 | return error; | |
7699acd1 | 2328 | |
791eb22e | 2329 | error = do_epoll_wait(epfd, events, maxevents, timeout); |
b772434b | 2330 | restore_saved_sigmask_unless(error == -EINTR); |
7699acd1 DL |
2331 | |
2332 | return error; | |
2333 | } | |
2334 | ||
35280bd4 AV |
2335 | #ifdef CONFIG_COMPAT |
2336 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, | |
2337 | struct epoll_event __user *, events, | |
2338 | int, maxevents, int, timeout, | |
2339 | const compat_sigset_t __user *, sigmask, | |
2340 | compat_size_t, sigsetsize) | |
2341 | { | |
2342 | long err; | |
35280bd4 AV |
2343 | |
2344 | /* | |
2345 | * If the caller wants a certain signal mask to be set during the wait, | |
2346 | * we apply it here. | |
2347 | */ | |
b772434b | 2348 | err = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
2349 | if (err) |
2350 | return err; | |
35280bd4 | 2351 | |
791eb22e | 2352 | err = do_epoll_wait(epfd, events, maxevents, timeout); |
b772434b | 2353 | restore_saved_sigmask_unless(err == -EINTR); |
35280bd4 AV |
2354 | |
2355 | return err; | |
2356 | } | |
2357 | #endif | |
2358 | ||
1da177e4 LT |
2359 | static int __init eventpoll_init(void) |
2360 | { | |
7ef9964e DL |
2361 | struct sysinfo si; |
2362 | ||
2363 | si_meminfo(&si); | |
9df04e1f DL |
2364 | /* |
2365 | * Allows top 4% of lomem to be allocated for epoll watches (per user). | |
2366 | */ | |
2367 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / | |
7ef9964e | 2368 | EP_ITEM_COST; |
52bd19f7 | 2369 | BUG_ON(max_user_watches < 0); |
1da177e4 | 2370 | |
22bacca4 DL |
2371 | /* |
2372 | * Initialize the structure used to perform epoll file descriptor | |
2373 | * inclusion loops checks. | |
2374 | */ | |
2375 | ep_nested_calls_init(&poll_loop_ncalls); | |
2376 | ||
39732ca5 EW |
2377 | /* |
2378 | * We can have many thousands of epitems, so prevent this from | |
2379 | * using an extra cache line on 64-bit (and smaller) CPUs | |
2380 | */ | |
2381 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); | |
2382 | ||
1da177e4 LT |
2383 | /* Allocates slab cache used to allocate "struct epitem" items */ |
2384 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
2ae928a9 | 2385 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 LT |
2386 | |
2387 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
2388 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
2ae928a9 | 2389 | sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 | 2390 | |
1da177e4 | 2391 | return 0; |
1da177e4 | 2392 | } |
cea69241 | 2393 | fs_initcall(eventpoll_init); |