Branch data Line data Source code
1 : : /*
2 : : * fs/eventpoll.c (Efficient event retrieval implementation)
3 : : * Copyright (C) 2001,...,2009 Davide Libenzi
4 : : *
5 : : * This program is free software; you can redistribute it and/or modify
6 : : * it under the terms of the GNU General Public License as published by
7 : : * the Free Software Foundation; either version 2 of the License, or
8 : : * (at your option) any later version.
9 : : *
10 : : * Davide Libenzi <davidel@xmailserver.org>
11 : : *
12 : : */
13 : :
14 : : #include <linux/init.h>
15 : : #include <linux/kernel.h>
16 : : #include <linux/sched.h>
17 : : #include <linux/fs.h>
18 : : #include <linux/file.h>
19 : : #include <linux/signal.h>
20 : : #include <linux/errno.h>
21 : : #include <linux/mm.h>
22 : : #include <linux/slab.h>
23 : : #include <linux/poll.h>
24 : : #include <linux/string.h>
25 : : #include <linux/list.h>
26 : : #include <linux/hash.h>
27 : : #include <linux/spinlock.h>
28 : : #include <linux/syscalls.h>
29 : : #include <linux/rbtree.h>
30 : : #include <linux/wait.h>
31 : : #include <linux/eventpoll.h>
32 : : #include <linux/mount.h>
33 : : #include <linux/bitops.h>
34 : : #include <linux/mutex.h>
35 : : #include <linux/anon_inodes.h>
36 : : #include <linux/device.h>
37 : : #include <linux/freezer.h>
38 : : #include <asm/uaccess.h>
39 : : #include <asm/io.h>
40 : : #include <asm/mman.h>
41 : : #include <linux/atomic.h>
42 : : #include <linux/proc_fs.h>
43 : : #include <linux/seq_file.h>
44 : : #include <linux/compat.h>
45 : : #include <linux/rculist.h>
46 : :
47 : : /*
48 : : * LOCKING:
49 : : * There are three level of locking required by epoll :
50 : : *
51 : : * 1) epmutex (mutex)
52 : : * 2) ep->mtx (mutex)
53 : : * 3) ep->lock (spinlock)
54 : : *
55 : : * The acquire order is the one listed above, from 1 to 3.
56 : : * We need a spinlock (ep->lock) because we manipulate objects
57 : : * from inside the poll callback, that might be triggered from
58 : : * a wake_up() that in turn might be called from IRQ context.
59 : : * So we can't sleep inside the poll callback and hence we need
60 : : * a spinlock. During the event transfer loop (from kernel to
61 : : * user space) we could end up sleeping due a copy_to_user(), so
62 : : * we need a lock that will allow us to sleep. This lock is a
63 : : * mutex (ep->mtx). It is acquired during the event transfer loop,
64 : : * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
65 : : * Then we also need a global mutex to serialize eventpoll_release_file()
66 : : * and ep_free().
67 : : * This mutex is acquired by ep_free() during the epoll file
68 : : * cleanup path and it is also acquired by eventpoll_release_file()
69 : : * if a file has been pushed inside an epoll set and it is then
70 : : * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
71 : : * It is also acquired when inserting an epoll fd onto another epoll
72 : : * fd. We do this so that we walk the epoll tree and ensure that this
73 : : * insertion does not create a cycle of epoll file descriptors, which
74 : : * could lead to deadlock. We need a global mutex to prevent two
75 : : * simultaneous inserts (A into B and B into A) from racing and
76 : : * constructing a cycle without either insert observing that it is
77 : : * going to.
78 : : * It is necessary to acquire multiple "ep->mtx"es at once in the
79 : : * case when one epoll fd is added to another. In this case, we
80 : : * always acquire the locks in the order of nesting (i.e. after
81 : : * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
82 : : * before e2->mtx). Since we disallow cycles of epoll file
83 : : * descriptors, this ensures that the mutexes are well-ordered. In
84 : : * order to communicate this nesting to lockdep, when walking a tree
85 : : * of epoll file descriptors, we use the current recursion depth as
86 : : * the lockdep subkey.
87 : : * It is possible to drop the "ep->mtx" and to use the global
88 : : * mutex "epmutex" (together with "ep->lock") to have it working,
89 : : * but having "ep->mtx" will make the interface more scalable.
90 : : * Events that require holding "epmutex" are very rare, while for
91 : : * normal operations the epoll private "ep->mtx" will guarantee
92 : : * a better scalability.
93 : : */
94 : :
95 : : /* Epoll private bits inside the event mask */
96 : : #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET)
97 : :
98 : : /* Maximum number of nesting allowed inside epoll sets */
99 : : #define EP_MAX_NESTS 4
100 : :
101 : : #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
102 : :
103 : : #define EP_UNACTIVE_PTR ((void *) -1L)
104 : :
105 : : #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
106 : :
107 : : struct epoll_filefd {
108 : : struct file *file;
109 : : int fd;
110 : : } __packed;
111 : :
112 : : /*
113 : : * Structure used to track possible nested calls, for too deep recursions
114 : : * and loop cycles.
115 : : */
116 : : struct nested_call_node {
117 : : struct list_head llink;
118 : : void *cookie;
119 : : void *ctx;
120 : : };
121 : :
122 : : /*
123 : : * This structure is used as collector for nested calls, to check for
124 : : * maximum recursion dept and loop cycles.
125 : : */
126 : : struct nested_calls {
127 : : struct list_head tasks_call_list;
128 : : spinlock_t lock;
129 : : };
130 : :
131 : : /*
132 : : * Each file descriptor added to the eventpoll interface will
133 : : * have an entry of this type linked to the "rbr" RB tree.
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.
136 : : */
137 : : struct epitem {
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 : : };
144 : :
145 : : /* List header used to link this structure to the eventpoll ready list */
146 : : struct list_head rdllink;
147 : :
148 : : /*
149 : : * Works together "struct eventpoll"->ovflist in keeping the
150 : : * single linked chain of items.
151 : : */
152 : : struct epitem *next;
153 : :
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 : :
166 : : /* List header used to link this item to the "struct file" items list */
167 : : struct list_head fllink;
168 : :
169 : : /* wakeup_source used when EPOLLWAKEUP is set */
170 : : struct wakeup_source __rcu *ws;
171 : :
172 : : /* The structure that describe the interested events and the source fd */
173 : : struct epoll_event event;
174 : : };
175 : :
176 : : /*
177 : : * This structure is stored inside the "private_data" member of the file
178 : : * structure and represents the main data structure for the eventpoll
179 : : * interface.
180 : : */
181 : : struct eventpoll {
182 : : /* Protect the access to this structure */
183 : : spinlock_t lock;
184 : :
185 : : /*
186 : : * This mutex is used to ensure that files are not removed
187 : : * while epoll is using them. This is held during the event
188 : : * collection loop, the file cleanup path, the epoll file exit
189 : : * code and the ctl operations.
190 : : */
191 : : struct mutex mtx;
192 : :
193 : : /* Wait queue used by sys_epoll_wait() */
194 : : wait_queue_head_t wq;
195 : :
196 : : /* Wait queue used by file->poll() */
197 : : wait_queue_head_t poll_wait;
198 : :
199 : : /* List of ready file descriptors */
200 : : struct list_head rdllist;
201 : :
202 : : /* RB tree root used to store monitored fd structs */
203 : : struct rb_root rbr;
204 : :
205 : : /*
206 : : * This is a single linked list that chains all the "struct epitem" that
207 : : * happened while transferring ready events to userspace w/out
208 : : * holding ->lock.
209 : : */
210 : : struct epitem *ovflist;
211 : :
212 : : /* wakeup_source used when ep_scan_ready_list is running */
213 : : struct wakeup_source *ws;
214 : :
215 : : /* The user that created the eventpoll descriptor */
216 : : struct user_struct *user;
217 : :
218 : : struct file *file;
219 : :
220 : : /* used to optimize loop detection check */
221 : : int visited;
222 : : struct list_head visited_list_link;
223 : : };
224 : :
225 : : /* Wait structure used by the poll hooks */
226 : : struct eppoll_entry {
227 : : /* List header used to link this structure to the "struct epitem" */
228 : : struct list_head llink;
229 : :
230 : : /* The "base" pointer is set to the container "struct epitem" */
231 : : struct epitem *base;
232 : :
233 : : /*
234 : : * Wait queue item that will be linked to the target file wait
235 : : * queue head.
236 : : */
237 : : wait_queue_t wait;
238 : :
239 : : /* The wait queue head that linked the "wait" wait queue item */
240 : : wait_queue_head_t *whead;
241 : : };
242 : :
243 : : /* Wrapper struct used by poll queueing */
244 : : struct ep_pqueue {
245 : : poll_table pt;
246 : : struct epitem *epi;
247 : : };
248 : :
249 : : /* Used by the ep_send_events() function as callback private data */
250 : : struct ep_send_events_data {
251 : : int maxevents;
252 : : struct epoll_event __user *events;
253 : : };
254 : :
255 : : /*
256 : : * Configuration options available inside /proc/sys/fs/epoll/
257 : : */
258 : : /* Maximum number of epoll watched descriptors, per user */
259 : : static long max_user_watches __read_mostly;
260 : :
261 : : /*
262 : : * This mutex is used to serialize ep_free() and eventpoll_release_file().
263 : : */
264 : : static DEFINE_MUTEX(epmutex);
265 : :
266 : : /* Used to check for epoll file descriptor inclusion loops */
267 : : static struct nested_calls poll_loop_ncalls;
268 : :
269 : : /* Used for safe wake up implementation */
270 : : static struct nested_calls poll_safewake_ncalls;
271 : :
272 : : /* Used to call file's f_op->poll() under the nested calls boundaries */
273 : : static struct nested_calls poll_readywalk_ncalls;
274 : :
275 : : /* Slab cache used to allocate "struct epitem" */
276 : : static struct kmem_cache *epi_cache __read_mostly;
277 : :
278 : : /* Slab cache used to allocate "struct eppoll_entry" */
279 : : static struct kmem_cache *pwq_cache __read_mostly;
280 : :
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 : :
290 : : #ifdef CONFIG_SYSCTL
291 : :
292 : : #include <linux/sysctl.h>
293 : :
294 : : static long zero;
295 : : static long long_max = LONG_MAX;
296 : :
297 : : ctl_table epoll_table[] = {
298 : : {
299 : : .procname = "max_user_watches",
300 : : .data = &max_user_watches,
301 : : .maxlen = sizeof(max_user_watches),
302 : : .mode = 0644,
303 : : .proc_handler = proc_doulongvec_minmax,
304 : : .extra1 = &zero,
305 : : .extra2 = &long_max,
306 : : },
307 : : { }
308 : : };
309 : : #endif /* CONFIG_SYSCTL */
310 : :
311 : : static const struct file_operations eventpoll_fops;
312 : :
313 : : static inline int is_file_epoll(struct file *f)
314 : : {
315 : 0 : return f->f_op == &eventpoll_fops;
316 : : }
317 : :
318 : : /* Setup the structure that is used as key for the RB tree */
319 : : static inline void ep_set_ffd(struct epoll_filefd *ffd,
320 : : struct file *file, int fd)
321 : : {
322 : 29 : ffd->file = file;
323 : 29 : ffd->fd = fd;
324 : : }
325 : :
326 : : /* Compare RB tree keys */
327 : : static inline int ep_cmp_ffd(struct epoll_filefd *p1,
328 : : struct epoll_filefd *p2)
329 : : {
330 [ + - ][ # # ]: 25 : return (p1->file > p2->file ? +1:
331 [ + - ][ # # ]: 25 : (p1->file < p2->file ? -1 : p1->fd - p2->fd));
332 : : }
333 : :
334 : : /* Tells us if the item is currently linked */
335 : : static inline int ep_is_linked(struct list_head *p)
336 : : {
337 : : return !list_empty(p);
338 : : }
339 : :
340 : : static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
341 : : {
342 : : return container_of(p, struct eppoll_entry, wait);
343 : : }
344 : :
345 : : /* Get the "struct epitem" from a wait queue pointer */
346 : : static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
347 : : {
348 : 1093 : return container_of(p, struct eppoll_entry, wait)->base;
349 : : }
350 : :
351 : : /* Get the "struct epitem" from an epoll queue wrapper */
352 : : static inline struct epitem *ep_item_from_epqueue(poll_table *p)
353 : : {
354 : 29 : return container_of(p, struct ep_pqueue, pt)->epi;
355 : : }
356 : :
357 : : /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
358 : : static inline int ep_op_has_event(int op)
359 : : {
360 : : return op != EPOLL_CTL_DEL;
361 : : }
362 : :
363 : : /* Initialize the poll safe wake up structure */
364 : : static void ep_nested_calls_init(struct nested_calls *ncalls)
365 : : {
366 : : INIT_LIST_HEAD(&ncalls->tasks_call_list);
367 : 0 : spin_lock_init(&ncalls->lock);
368 : : }
369 : :
370 : : /**
371 : : * ep_events_available - Checks if ready events might be available.
372 : : *
373 : : * @ep: Pointer to the eventpoll context.
374 : : *
375 : : * Returns: Returns a value different than zero if ready events are available,
376 : : * or zero otherwise.
377 : : */
378 : : static inline int ep_events_available(struct eventpoll *ep)
379 : : {
380 [ + + ][ + - ]: 5954 : return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
[ + + ][ + - ]
[ + + ][ + - ]
381 : : }
382 : :
383 : : /**
384 : : * ep_call_nested - Perform a bound (possibly) nested call, by checking
385 : : * that the recursion limit is not exceeded, and that
386 : : * the same nested call (by the meaning of same cookie) is
387 : : * no re-entered.
388 : : *
389 : : * @ncalls: Pointer to the nested_calls structure to be used for this call.
390 : : * @max_nests: Maximum number of allowed nesting calls.
391 : : * @nproc: Nested call core function pointer.
392 : : * @priv: Opaque data to be passed to the @nproc callback.
393 : : * @cookie: Cookie to be used to identify this nested call.
394 : : * @ctx: This instance context.
395 : : *
396 : : * Returns: Returns the code returned by the @nproc callback, or -1 if
397 : : * the maximum recursion limit has been exceeded.
398 : : */
399 : 0 : static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
400 : : int (*nproc)(void *, void *, int), void *priv,
401 : : void *cookie, void *ctx)
402 : : {
403 : : int error, call_nests = 0;
404 : : unsigned long flags;
405 : 1567 : struct list_head *lsthead = &ncalls->tasks_call_list;
406 : : struct nested_call_node *tncur;
407 : : struct nested_call_node tnode;
408 : :
409 : 1567 : spin_lock_irqsave(&ncalls->lock, flags);
410 : :
411 : : /*
412 : : * Try to see if the current task is already inside this wakeup call.
413 : : * We use a list here, since the population inside this set is always
414 : : * very much limited.
415 : : */
416 [ - + ]: 3134 : list_for_each_entry(tncur, lsthead, llink) {
417 [ # # ][ # # ]: 0 : if (tncur->ctx == ctx &&
418 [ # # ]: 0 : (tncur->cookie == cookie || ++call_nests > max_nests)) {
419 : : /*
420 : : * Ops ... loop detected or maximum nest level reached.
421 : : * We abort this wake by breaking the cycle itself.
422 : : */
423 : : error = -1;
424 : : goto out_unlock;
425 : : }
426 : : }
427 : :
428 : : /* Add the current task and cookie to the list */
429 : 1567 : tnode.ctx = ctx;
430 : 1567 : tnode.cookie = cookie;
431 : : list_add(&tnode.llink, lsthead);
432 : :
433 : : spin_unlock_irqrestore(&ncalls->lock, flags);
434 : :
435 : : /* Call the nested function */
436 : 1567 : error = (*nproc)(priv, cookie, call_nests);
437 : :
438 : : /* Remove the current task from the list */
439 : 1567 : spin_lock_irqsave(&ncalls->lock, flags);
440 : : list_del(&tnode.llink);
441 : : out_unlock:
442 : : spin_unlock_irqrestore(&ncalls->lock, flags);
443 : :
444 : 1567 : return error;
445 : : }
446 : :
447 : : /*
448 : : * As described in commit 0ccf831cb lockdep: annotate epoll
449 : : * the use of wait queues used by epoll is done in a very controlled
450 : : * manner. Wake ups can nest inside each other, but are never done
451 : : * with the same locking. For example:
452 : : *
453 : : * dfd = socket(...);
454 : : * efd1 = epoll_create();
455 : : * efd2 = epoll_create();
456 : : * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
457 : : * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
458 : : *
459 : : * When a packet arrives to the device underneath "dfd", the net code will
460 : : * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
461 : : * callback wakeup entry on that queue, and the wake_up() performed by the
462 : : * "dfd" net code will end up in ep_poll_callback(). At this point epoll
463 : : * (efd1) notices that it may have some event ready, so it needs to wake up
464 : : * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
465 : : * that ends up in another wake_up(), after having checked about the
466 : : * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
467 : : * avoid stack blasting.
468 : : *
469 : : * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
470 : : * this special case of epoll.
471 : : */
472 : : #ifdef CONFIG_DEBUG_LOCK_ALLOC
473 : : static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
474 : : unsigned long events, int subclass)
475 : : {
476 : : unsigned long flags;
477 : :
478 : : spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
479 : : wake_up_locked_poll(wqueue, events);
480 : : spin_unlock_irqrestore(&wqueue->lock, flags);
481 : : }
482 : : #else
483 : : static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
484 : : unsigned long events, int subclass)
485 : : {
486 : 0 : wake_up_poll(wqueue, events);
487 : : }
488 : : #endif
489 : :
490 : 0 : static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
491 : : {
492 : : ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
493 : : 1 + call_nests);
494 : 0 : return 0;
495 : : }
496 : :
497 : : /*
498 : : * Perform a safe wake up of the poll wait list. The problem is that
499 : : * with the new callback'd wake up system, it is possible that the
500 : : * poll callback is reentered from inside the call to wake_up() done
501 : : * on the poll wait queue head. The rule is that we cannot reenter the
502 : : * wake up code from the same task more than EP_MAX_NESTS times,
503 : : * and we cannot reenter the same wait queue head at all. This will
504 : : * enable to have a hierarchy of epoll file descriptor of no more than
505 : : * EP_MAX_NESTS deep.
506 : : */
507 : 0 : static void ep_poll_safewake(wait_queue_head_t *wq)
508 : : {
509 : 0 : int this_cpu = get_cpu();
510 : :
511 : 0 : ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
512 : : ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
513 : :
514 : 0 : put_cpu();
515 : 0 : }
516 : :
517 : : static void ep_remove_wait_queue(struct eppoll_entry *pwq)
518 : : {
519 : : wait_queue_head_t *whead;
520 : :
521 : : rcu_read_lock();
522 : : /* If it is cleared by POLLFREE, it should be rcu-safe */
523 : 29 : whead = rcu_dereference(pwq->whead);
524 [ + - ]: 29 : if (whead)
525 : 29 : remove_wait_queue(whead, &pwq->wait);
526 : : rcu_read_unlock();
527 : : }
528 : :
529 : : /*
530 : : * This function unregisters poll callbacks from the associated file
531 : : * descriptor. Must be called with "mtx" held (or "epmutex" if called from
532 : : * ep_free).
533 : : */
534 : 33 : static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
535 : : {
536 : 33 : struct list_head *lsthead = &epi->pwqlist;
537 : : struct eppoll_entry *pwq;
538 : :
539 [ + + ]: 62 : while (!list_empty(lsthead)) {
540 : : pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
541 : :
542 : : list_del(&pwq->llink);
543 : : ep_remove_wait_queue(pwq);
544 : 29 : kmem_cache_free(pwq_cache, pwq);
545 : : }
546 : 33 : }
547 : :
548 : : /* call only when ep->mtx is held */
549 : : static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
550 : : {
551 : : return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
552 : : }
553 : :
554 : : /* call only when ep->mtx is held */
555 : 0 : static inline void ep_pm_stay_awake(struct epitem *epi)
556 : : {
557 : : struct wakeup_source *ws = ep_wakeup_source(epi);
558 : :
559 [ # # ]: 1041 : if (ws)
[ - + - ]
[ - + ][ # # ]
560 : 2066 : __pm_stay_awake(ws);
561 : : }
562 : :
563 : : static inline bool ep_has_wakeup_source(struct epitem *epi)
564 : : {
565 : : return rcu_access_pointer(epi->ws) ? true : false;
566 : : }
567 : :
568 : : /* call when ep->mtx cannot be held (ep_poll_callback) */
569 : : static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
570 : : {
571 : : struct wakeup_source *ws;
572 : :
573 : : rcu_read_lock();
574 : 1050 : ws = rcu_dereference(epi->ws);
575 [ - + ]: 1050 : if (ws)
576 : 0 : __pm_stay_awake(ws);
577 : : rcu_read_unlock();
578 : : }
579 : :
580 : : /**
581 : : * ep_scan_ready_list - Scans the ready list in a way that makes possible for
582 : : * the scan code, to call f_op->poll(). Also allows for
583 : : * O(NumReady) performance.
584 : : *
585 : : * @ep: Pointer to the epoll private data structure.
586 : : * @sproc: Pointer to the scan callback.
587 : : * @priv: Private opaque data passed to the @sproc callback.
588 : : * @depth: The current depth of recursive f_op->poll calls.
589 : : * @ep_locked: caller already holds ep->mtx
590 : : *
591 : : * Returns: The same integer error code returned by the @sproc callback.
592 : : */
593 : 3557 : static int ep_scan_ready_list(struct eventpoll *ep,
594 : : int (*sproc)(struct eventpoll *,
595 : : struct list_head *, void *),
596 : : void *priv, int depth, bool ep_locked)
597 : : {
598 : : int error, pwake = 0;
599 : : unsigned long flags;
600 : : struct epitem *epi, *nepi;
601 : 3557 : LIST_HEAD(txlist);
602 : :
603 : : /*
604 : : * We need to lock this because we could be hit by
605 : : * eventpoll_release_file() and epoll_ctl().
606 : : */
607 : :
608 [ + - ]: 3557 : if (!ep_locked)
609 : 3557 : mutex_lock_nested(&ep->mtx, depth);
610 : :
611 : : /*
612 : : * Steal the ready list, and re-init the original one to the
613 : : * empty list. Also, set ep->ovflist to NULL so that events
614 : : * happening while looping w/out locks, are not lost. We cannot
615 : : * have the poll callback to queue directly on ep->rdllist,
616 : : * because we want the "sproc" callback to be able to do it
617 : : * in a lockless way.
618 : : */
619 : 3557 : spin_lock_irqsave(&ep->lock, flags);
620 : 3557 : list_splice_init(&ep->rdllist, &txlist);
621 : 3557 : ep->ovflist = NULL;
622 : : spin_unlock_irqrestore(&ep->lock, flags);
623 : :
624 : : /*
625 : : * Now call the callback function.
626 : : */
627 : 3557 : error = (*sproc)(ep, &txlist, priv);
628 : :
629 : 3557 : spin_lock_irqsave(&ep->lock, flags);
630 : : /*
631 : : * During the time we spent inside the "sproc" callback, some
632 : : * other events might have been queued by the poll callback.
633 : : * We re-insert them inside the main ready-list here.
634 : : */
635 [ + + ]: 3558 : for (nepi = ep->ovflist; (epi = nepi) != NULL;
636 : 1 : nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
637 : : /*
638 : : * We need to check if the item is already in the list.
639 : : * During the "sproc" callback execution time, items are
640 : : * queued into ->ovflist but the "txlist" might already
641 : : * contain them, and the list_splice() below takes care of them.
642 : : */
643 [ + - ]: 1 : if (!ep_is_linked(&epi->rdllink)) {
644 : : list_add_tail(&epi->rdllink, &ep->rdllist);
645 : : ep_pm_stay_awake(epi);
646 : : }
647 : : }
648 : : /*
649 : : * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
650 : : * releasing the lock, events will be queued in the normal way inside
651 : : * ep->rdllist.
652 : : */
653 : 3557 : ep->ovflist = EP_UNACTIVE_PTR;
654 : :
655 : : /*
656 : : * Quickly re-inject items left on "txlist".
657 : : */
658 : : list_splice(&txlist, &ep->rdllist);
659 : 3557 : __pm_relax(ep->ws);
660 : :
661 [ + + ]: 3557 : if (!list_empty(&ep->rdllist)) {
662 : : /*
663 : : * Wake up (if active) both the eventpoll wait list and
664 : : * the ->poll() wait list (delayed after we release the lock).
665 : : */
666 [ - + ]: 1036 : if (waitqueue_active(&ep->wq))
667 : 0 : wake_up_locked(&ep->wq);
668 [ - + ]: 4593 : if (waitqueue_active(&ep->poll_wait))
669 : : pwake++;
670 : : }
671 : : spin_unlock_irqrestore(&ep->lock, flags);
672 : :
673 [ + - ]: 3557 : if (!ep_locked)
674 : 3557 : mutex_unlock(&ep->mtx);
675 : :
676 : : /* We have to call this outside the lock */
677 [ - + ]: 3557 : if (pwake)
678 : 0 : ep_poll_safewake(&ep->poll_wait);
679 : :
680 : 3557 : return error;
681 : : }
682 : :
683 : 0 : static void epi_rcu_free(struct rcu_head *head)
684 : : {
685 : : struct epitem *epi = container_of(head, struct epitem, rcu);
686 : 29 : kmem_cache_free(epi_cache, epi);
687 : 29 : }
688 : :
689 : : /*
690 : : * Removes a "struct epitem" from the eventpoll RB tree and deallocates
691 : : * all the associated resources. Must be called with "mtx" held.
692 : : */
693 : 0 : static int ep_remove(struct eventpoll *ep, struct epitem *epi)
694 : : {
695 : : unsigned long flags;
696 : 29 : struct file *file = epi->ffd.file;
697 : :
698 : : /*
699 : : * Removes poll wait queue hooks. We _have_ to do this without holding
700 : : * the "ep->lock" otherwise a deadlock might occur. This because of the
701 : : * sequence of the lock acquisition. Here we do "ep->lock" then the wait
702 : : * queue head lock when unregistering the wait queue. The wakeup callback
703 : : * will run by holding the wait queue head lock and will call our callback
704 : : * that will try to get "ep->lock".
705 : : */
706 : 29 : ep_unregister_pollwait(ep, epi);
707 : :
708 : : /* Remove the current item from the list of epoll hooks */
709 : : spin_lock(&file->f_lock);
710 : : list_del_rcu(&epi->fllink);
711 : : spin_unlock(&file->f_lock);
712 : :
713 : 29 : rb_erase(&epi->rbn, &ep->rbr);
714 : :
715 : 29 : spin_lock_irqsave(&ep->lock, flags);
716 [ + + ]: 29 : if (ep_is_linked(&epi->rdllink))
717 : : list_del_init(&epi->rdllink);
718 : : spin_unlock_irqrestore(&ep->lock, flags);
719 : :
720 : 29 : wakeup_source_unregister(ep_wakeup_source(epi));
721 : : /*
722 : : * At this point it is safe to free the eventpoll item. Use the union
723 : : * field epi->rcu, since we are trying to minimize the size of
724 : : * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
725 : : * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
726 : : * use of the rbn field.
727 : : */
728 : 29 : call_rcu(&epi->rcu, epi_rcu_free);
729 : :
730 : 29 : atomic_long_dec(&ep->user->epoll_watches);
731 : :
732 : 29 : return 0;
733 : : }
734 : :
735 : 0 : static void ep_free(struct eventpoll *ep)
736 : : {
737 : : struct rb_node *rbp;
738 : : struct epitem *epi;
739 : :
740 : : /* We need to release all tasks waiting for these file */
741 [ - + ]: 64 : if (waitqueue_active(&ep->poll_wait))
742 : 0 : ep_poll_safewake(&ep->poll_wait);
743 : :
744 : : /*
745 : : * We need to lock this because we could be hit by
746 : : * eventpoll_release_file() while we're freeing the "struct eventpoll".
747 : : * We do not need to hold "ep->mtx" here because the epoll file
748 : : * is on the way to be removed and no one has references to it
749 : : * anymore. The only hit might come from eventpoll_release_file() but
750 : : * holding "epmutex" is sufficient here.
751 : : */
752 : 64 : mutex_lock(&epmutex);
753 : :
754 : : /*
755 : : * Walks through the whole tree by unregistering poll callbacks.
756 : : */
757 [ + + ]: 68 : for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
758 : : epi = rb_entry(rbp, struct epitem, rbn);
759 : :
760 : 4 : ep_unregister_pollwait(ep, epi);
761 : 4 : cond_resched();
762 : : }
763 : :
764 : : /*
765 : : * Walks through the whole tree by freeing each "struct epitem". At this
766 : : * point we are sure no poll callbacks will be lingering around, and also by
767 : : * holding "epmutex" we can be sure that no file cleanup code will hit
768 : : * us during this operation. So we can avoid the lock on "ep->lock".
769 : : * We do not need to lock ep->mtx, either, we only do it to prevent
770 : : * a lockdep warning.
771 : : */
772 : 64 : mutex_lock(&ep->mtx);
773 [ + + ]: 68 : while ((rbp = rb_first(&ep->rbr)) != NULL) {
774 : : epi = rb_entry(rbp, struct epitem, rbn);
775 : 4 : ep_remove(ep, epi);
776 : 4 : cond_resched();
777 : : }
778 : 64 : mutex_unlock(&ep->mtx);
779 : :
780 : 64 : mutex_unlock(&epmutex);
781 : : mutex_destroy(&ep->mtx);
782 : 64 : free_uid(ep->user);
783 : 64 : wakeup_source_unregister(ep->ws);
784 : 64 : kfree(ep);
785 : 64 : }
786 : :
787 : 0 : static int ep_eventpoll_release(struct inode *inode, struct file *file)
788 : : {
789 : 64 : struct eventpoll *ep = file->private_data;
790 : :
791 [ + - ]: 64 : if (ep)
792 : 64 : ep_free(ep);
793 : :
794 : 0 : return 0;
795 : : }
796 : :
797 : : static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt)
798 : : {
799 : 2094 : pt->_key = epi->event.events;
800 : :
801 : 2094 : return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events;
802 : : }
803 : :
804 : 0 : static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
805 : : void *priv)
806 : : {
807 : 0 : struct epitem *epi, *tmp;
808 : : poll_table pt;
809 : :
810 : : init_poll_funcptr(&pt, NULL);
811 : :
812 [ - + ]: 1567 : list_for_each_entry_safe(epi, tmp, head, rdllink) {
813 [ # # ]: 0 : if (ep_item_poll(epi, &pt))
814 : : return POLLIN | POLLRDNORM;
815 : : else {
816 : : /*
817 : : * Item has been dropped into the ready list by the poll
818 : : * callback, but it's not actually ready, as far as
819 : : * caller requested events goes. We can remove it here.
820 : : */
821 : 0 : __pm_relax(ep_wakeup_source(epi));
822 : : list_del_init(&epi->rdllink);
823 : : }
824 : : }
825 : :
826 : : return 0;
827 : : }
828 : :
829 : : static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
830 : : poll_table *pt);
831 : :
832 : : struct readyevents_arg {
833 : : struct eventpoll *ep;
834 : : bool locked;
835 : : };
836 : :
837 : 0 : static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
838 : : {
839 : : struct readyevents_arg *arg = priv;
840 : :
841 : 1567 : return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL,
842 : : call_nests + 1, arg->locked);
843 : : }
844 : :
845 : 0 : static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
846 : : {
847 : : int pollflags;
848 : 1567 : struct eventpoll *ep = file->private_data;
849 : : struct readyevents_arg arg;
850 : :
851 : : /*
852 : : * During ep_insert() we already hold the ep->mtx for the tfile.
853 : : * Prevent re-aquisition.
854 : : */
855 [ + - ][ + - ]: 1567 : arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc);
856 : 1567 : arg.ep = ep;
857 : :
858 : : /* Insert inside our poll wait queue */
859 : 1567 : poll_wait(file, &ep->poll_wait, wait);
860 : :
861 : : /*
862 : : * Proceed to find out if wanted events are really available inside
863 : : * the ready list. This need to be done under ep_call_nested()
864 : : * supervision, since the call to f_op->poll() done on listed files
865 : : * could re-enter here.
866 : : */
867 : 1567 : pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
868 : 0 : ep_poll_readyevents_proc, &arg, ep, current);
869 : :
870 [ + - ]: 1567 : return pollflags != -1 ? pollflags : 0;
871 : : }
872 : :
873 : : #ifdef CONFIG_PROC_FS
874 : 0 : static int ep_show_fdinfo(struct seq_file *m, struct file *f)
875 : : {
876 : 20 : struct eventpoll *ep = f->private_data;
877 : : struct rb_node *rbp;
878 : : int ret = 0;
879 : :
880 : 20 : mutex_lock(&ep->mtx);
881 [ + + ]: 60 : for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
882 : : struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
883 : :
884 : 40 : ret = seq_printf(m, "tfd: %8d events: %8x data: %16llx\n",
885 : : epi->ffd.fd, epi->event.events,
886 : 40 : (long long)epi->event.data);
887 [ + - ]: 40 : if (ret)
888 : : break;
889 : : }
890 : 20 : mutex_unlock(&ep->mtx);
891 : :
892 : 20 : return ret;
893 : : }
894 : : #endif
895 : :
896 : : /* File callbacks that implement the eventpoll file behaviour */
897 : : static const struct file_operations eventpoll_fops = {
898 : : #ifdef CONFIG_PROC_FS
899 : : .show_fdinfo = ep_show_fdinfo,
900 : : #endif
901 : : .release = ep_eventpoll_release,
902 : : .poll = ep_eventpoll_poll,
903 : : .llseek = noop_llseek,
904 : : };
905 : :
906 : : /*
907 : : * This is called from eventpoll_release() to unlink files from the eventpoll
908 : : * interface. We need to have this facility to cleanup correctly files that are
909 : : * closed without being removed from the eventpoll interface.
910 : : */
911 : 0 : void eventpoll_release_file(struct file *file)
912 : : {
913 : : struct eventpoll *ep;
914 : : struct epitem *epi;
915 : :
916 : : /*
917 : : * We don't want to get "file->f_lock" because it is not
918 : : * necessary. It is not necessary because we're in the "struct file"
919 : : * cleanup path, and this means that no one is using this file anymore.
920 : : * So, for example, epoll_ctl() cannot hit here since if we reach this
921 : : * point, the file counter already went to zero and fget() would fail.
922 : : * The only hit might come from ep_free() but by holding the mutex
923 : : * will correctly serialize the operation. We do need to acquire
924 : : * "ep->mtx" after "epmutex" because ep_remove() requires it when called
925 : : * from anywhere but ep_free().
926 : : *
927 : : * Besides, ep_remove() acquires the lock, so we can't hold it here.
928 : : */
929 : 0 : mutex_lock(&epmutex);
930 [ # # ]: 0 : list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
931 : 0 : ep = epi->ep;
932 : 0 : mutex_lock_nested(&ep->mtx, 0);
933 : 0 : ep_remove(ep, epi);
934 : 0 : mutex_unlock(&ep->mtx);
935 : : }
936 : 0 : mutex_unlock(&epmutex);
937 : 0 : }
938 : :
939 : 0 : static int ep_alloc(struct eventpoll **pep)
940 : : {
941 : : int error;
942 : : struct user_struct *user;
943 : : struct eventpoll *ep;
944 : :
945 : 64 : user = get_current_user();
946 : : error = -ENOMEM;
947 : : ep = kzalloc(sizeof(*ep), GFP_KERNEL);
948 [ + - ]: 64 : if (unlikely(!ep))
949 : : goto free_uid;
950 : :
951 : 64 : spin_lock_init(&ep->lock);
952 : 64 : mutex_init(&ep->mtx);
953 : 64 : init_waitqueue_head(&ep->wq);
954 : 64 : init_waitqueue_head(&ep->poll_wait);
955 : 64 : INIT_LIST_HEAD(&ep->rdllist);
956 : 64 : ep->rbr = RB_ROOT;
957 : 64 : ep->ovflist = EP_UNACTIVE_PTR;
958 : 64 : ep->user = user;
959 : :
960 : 64 : *pep = ep;
961 : :
962 : 64 : return 0;
963 : :
964 : : free_uid:
965 : 0 : free_uid(user);
966 : 0 : return error;
967 : : }
968 : :
969 : : /*
970 : : * Search the file inside the eventpoll tree. The RB tree operations
971 : : * are protected by the "mtx" mutex, and ep_find() must be called with
972 : : * "mtx" held.
973 : : */
974 : 54 : static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
975 : : {
976 : : int kcmp;
977 : : struct rb_node *rbp;
978 : : struct epitem *epi, *epir = NULL;
979 : : struct epoll_filefd ffd;
980 : :
981 : : ep_set_ffd(&ffd, file, fd);
982 [ + + ]: 54 : for (rbp = ep->rbr.rb_node; rbp; ) {
983 : : epi = rb_entry(rbp, struct epitem, rbn);
984 : : kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
985 [ - + ]: 25 : if (kcmp > 0)
986 : 0 : rbp = rbp->rb_right;
987 [ - ]: 25 : else if (kcmp < 0)
988 : 0 : rbp = rbp->rb_left;
989 : : else {
990 : : epir = epi;
991 : : break;
992 : : }
993 : : }
994 : :
995 : 0 : return epir;
996 : : }
997 : :
998 : : /*
999 : : * This is the callback that is passed to the wait queue wakeup
1000 : : * mechanism. It is called by the stored file descriptors when they
1001 : : * have events to report.
1002 : : */
1003 : 0 : static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
1004 : : {
1005 : : int pwake = 0;
1006 : : unsigned long flags;
1007 : : struct epitem *epi = ep_item_from_wait(wait);
1008 : 1093 : struct eventpoll *ep = epi->ep;
1009 : :
1010 [ - + ]: 1093 : if ((unsigned long)key & POLLFREE) {
1011 : 0 : ep_pwq_from_wait(wait)->whead = NULL;
1012 : : /*
1013 : : * whead = NULL above can race with ep_remove_wait_queue()
1014 : : * which can do another remove_wait_queue() after us, so we
1015 : : * can't use __remove_wait_queue(). whead->lock is held by
1016 : : * the caller.
1017 : : */
1018 : 0 : list_del_init(&wait->task_list);
1019 : : }
1020 : :
1021 : 1093 : spin_lock_irqsave(&ep->lock, flags);
1022 : :
1023 : : /*
1024 : : * If the event mask does not contain any poll(2) event, we consider the
1025 : : * descriptor to be disabled. This condition is likely the effect of the
1026 : : * EPOLLONESHOT bit that disables the descriptor when an event is received,
1027 : : * until the next EPOLL_CTL_MOD will be issued.
1028 : : */
1029 [ + - ]: 1093 : if (!(epi->event.events & ~EP_PRIVATE_BITS))
1030 : : goto out_unlock;
1031 : :
1032 : : /*
1033 : : * Check the events coming with the callback. At this stage, not
1034 : : * every device reports the events in the "key" parameter of the
1035 : : * callback. We need to be able to handle both cases here, hence the
1036 : : * test for "key" != NULL before the event match test.
1037 : : */
1038 [ + + ][ + + ]: 1093 : if (key && !((unsigned long) key & epi->event.events))
1039 : : goto out_unlock;
1040 : :
1041 : : /*
1042 : : * If we are transferring events to userspace, we can hold no locks
1043 : : * (because we're accessing user memory, and because of linux f_op->poll()
1044 : : * semantics). All the events that happen during that period of time are
1045 : : * chained in ep->ovflist and requeued later on.
1046 : : */
1047 [ + + ]: 1074 : if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
1048 [ + - ]: 1 : if (epi->next == EP_UNACTIVE_PTR) {
1049 : 1 : epi->next = ep->ovflist;
1050 : 1 : ep->ovflist = epi;
1051 [ - + ]: 1 : if (epi->ws) {
1052 : : /*
1053 : : * Activate ep->ws since epi->ws may get
1054 : : * deactivated at any time.
1055 : : */
1056 : 0 : __pm_stay_awake(ep->ws);
1057 : : }
1058 : :
1059 : : }
1060 : : goto out_unlock;
1061 : : }
1062 : :
1063 : : /* If this file is already in the ready list we exit soon */
1064 [ + + ]: 1073 : if (!ep_is_linked(&epi->rdllink)) {
1065 : 1050 : list_add_tail(&epi->rdllink, &ep->rdllist);
1066 : : ep_pm_stay_awake_rcu(epi);
1067 : : }
1068 : :
1069 : : /*
1070 : : * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1071 : : * wait list.
1072 : : */
1073 [ + + ]: 2166 : if (waitqueue_active(&ep->wq))
1074 : 992 : wake_up_locked(&ep->wq);
1075 [ - + ]: 1073 : if (waitqueue_active(&ep->poll_wait))
1076 : : pwake++;
1077 : :
1078 : : out_unlock:
1079 : : spin_unlock_irqrestore(&ep->lock, flags);
1080 : :
1081 : : /* We have to call this outside the lock */
1082 [ - + ]: 1093 : if (pwake)
1083 : 0 : ep_poll_safewake(&ep->poll_wait);
1084 : :
1085 : 1093 : return 1;
1086 : : }
1087 : :
1088 : : /*
1089 : : * This is the callback that is used to add our wait queue to the
1090 : : * target file wakeup lists.
1091 : : */
1092 : 0 : static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1093 : : poll_table *pt)
1094 : : {
1095 : : struct epitem *epi = ep_item_from_epqueue(pt);
1096 : : struct eppoll_entry *pwq;
1097 : :
1098 [ + - ][ + - ]: 29 : if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
1099 : : init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1100 : 29 : pwq->whead = whead;
1101 : 29 : pwq->base = epi;
1102 : 29 : add_wait_queue(whead, &pwq->wait);
1103 : 29 : list_add_tail(&pwq->llink, &epi->pwqlist);
1104 : 29 : epi->nwait++;
1105 : : } else {
1106 : : /* We have to signal that an error occurred */
1107 : 0 : epi->nwait = -1;
1108 : : }
1109 : 29 : }
1110 : :
1111 : 0 : static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1112 : : {
1113 : : int kcmp;
1114 : 29 : struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
1115 : : struct epitem *epic;
1116 : :
1117 [ - + ]: 29 : while (*p) {
1118 : : parent = *p;
1119 : : epic = rb_entry(parent, struct epitem, rbn);
1120 : : kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1121 [ # # ]: 0 : if (kcmp > 0)
1122 : 0 : p = &parent->rb_right;
1123 : : else
1124 : 0 : p = &parent->rb_left;
1125 : : }
1126 : 29 : rb_link_node(&epi->rbn, parent, p);
1127 : 29 : rb_insert_color(&epi->rbn, &ep->rbr);
1128 : 29 : }
1129 : :
1130 : :
1131 : :
1132 : : #define PATH_ARR_SIZE 5
1133 : : /*
1134 : : * These are the number paths of length 1 to 5, that we are allowing to emanate
1135 : : * from a single file of interest. For example, we allow 1000 paths of length
1136 : : * 1, to emanate from each file of interest. This essentially represents the
1137 : : * potential wakeup paths, which need to be limited in order to avoid massive
1138 : : * uncontrolled wakeup storms. The common use case should be a single ep which
1139 : : * is connected to n file sources. In this case each file source has 1 path
1140 : : * of length 1. Thus, the numbers below should be more than sufficient. These
1141 : : * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1142 : : * and delete can't add additional paths. Protected by the epmutex.
1143 : : */
1144 : : static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1145 : : static int path_count[PATH_ARR_SIZE];
1146 : :
1147 : : static int path_count_inc(int nests)
1148 : : {
1149 : : /* Allow an arbitrary number of depth 1 paths */
1150 [ # # ]: 0 : if (nests == 0)
1151 : : return 0;
1152 : :
1153 [ # # ]: 0 : if (++path_count[nests] > path_limits[nests])
1154 : : return -1;
1155 : : return 0;
1156 : : }
1157 : :
1158 : : static void path_count_init(void)
1159 : : {
1160 : : int i;
1161 : :
1162 [ # # ]: 0 : for (i = 0; i < PATH_ARR_SIZE; i++)
1163 : 0 : path_count[i] = 0;
1164 : : }
1165 : :
1166 : 0 : static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1167 : : {
1168 : : int error = 0;
1169 : : struct file *file = priv;
1170 : 0 : struct file *child_file;
1171 : : struct epitem *epi;
1172 : :
1173 : : /* CTL_DEL can remove links here, but that can't increase our count */
1174 : : rcu_read_lock();
1175 [ # # ]: 0 : list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
1176 : 0 : child_file = epi->ep->file;
1177 [ # # ]: 0 : if (is_file_epoll(child_file)) {
1178 [ # # ]: 0 : if (list_empty(&child_file->f_ep_links)) {
1179 [ # # ]: 0 : if (path_count_inc(call_nests)) {
1180 : : error = -1;
1181 : : break;
1182 : : }
1183 : : } else {
1184 : 0 : error = ep_call_nested(&poll_loop_ncalls,
1185 : : EP_MAX_NESTS,
1186 : : reverse_path_check_proc,
1187 : : child_file, child_file,
1188 : 0 : current);
1189 : : }
1190 [ # # ]: 0 : if (error != 0)
1191 : : break;
1192 : : } else {
1193 : 0 : printk(KERN_ERR "reverse_path_check_proc: "
1194 : : "file is not an ep!\n");
1195 : : }
1196 : : }
1197 : : rcu_read_unlock();
1198 : 0 : return error;
1199 : : }
1200 : :
1201 : : /**
1202 : : * reverse_path_check - The tfile_check_list is list of file *, which have
1203 : : * links that are proposed to be newly added. We need to
1204 : : * make sure that those added links don't add too many
1205 : : * paths such that we will spend all our time waking up
1206 : : * eventpoll objects.
1207 : : *
1208 : : * Returns: Returns zero if the proposed links don't create too many paths,
1209 : : * -1 otherwise.
1210 : : */
1211 : 0 : static int reverse_path_check(void)
1212 : : {
1213 : : int error = 0;
1214 : : struct file *current_file;
1215 : :
1216 : : /* let's call this for all tfiles */
1217 [ # # ]: 0 : list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1218 : : path_count_init();
1219 : 0 : error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1220 : : reverse_path_check_proc, current_file,
1221 : 0 : current_file, current);
1222 [ # # ]: 0 : if (error)
1223 : : break;
1224 : : }
1225 : 0 : return error;
1226 : : }
1227 : :
1228 : 0 : static int ep_create_wakeup_source(struct epitem *epi)
1229 : : {
1230 : : const char *name;
1231 : : struct wakeup_source *ws;
1232 : :
1233 [ # # ]: 0 : if (!epi->ep->ws) {
1234 : 0 : epi->ep->ws = wakeup_source_register("eventpoll");
1235 [ # # ]: 0 : if (!epi->ep->ws)
1236 : : return -ENOMEM;
1237 : : }
1238 : :
1239 : 0 : name = epi->ffd.file->f_path.dentry->d_name.name;
1240 : 0 : ws = wakeup_source_register(name);
1241 : :
1242 [ # # ]: 0 : if (!ws)
1243 : : return -ENOMEM;
1244 : 0 : rcu_assign_pointer(epi->ws, ws);
1245 : :
1246 : 0 : return 0;
1247 : : }
1248 : :
1249 : : /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1250 : 0 : static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1251 : : {
1252 : : struct wakeup_source *ws = ep_wakeup_source(epi);
1253 : :
1254 : 0 : RCU_INIT_POINTER(epi->ws, NULL);
1255 : :
1256 : : /*
1257 : : * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1258 : : * used internally by wakeup_source_remove, too (called by
1259 : : * wakeup_source_unregister), so we cannot use call_rcu
1260 : : */
1261 : : synchronize_rcu();
1262 : 0 : wakeup_source_unregister(ws);
1263 : 0 : }
1264 : :
1265 : : /*
1266 : : * Must be called with "mtx" held.
1267 : : */
1268 : 0 : static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1269 : : struct file *tfile, int fd, int full_check)
1270 : : {
1271 : : int error, revents, pwake = 0;
1272 : : unsigned long flags;
1273 : : long user_watches;
1274 : 0 : struct epitem *epi;
1275 : : struct ep_pqueue epq;
1276 : :
1277 : 29 : user_watches = atomic_long_read(&ep->user->epoll_watches);
1278 [ + - ]: 29 : if (unlikely(user_watches >= max_user_watches))
1279 : : return -ENOSPC;
1280 [ + - ]: 29 : if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
1281 : : return -ENOMEM;
1282 : :
1283 : : /* Item initialization follow here ... */
1284 : 29 : INIT_LIST_HEAD(&epi->rdllink);
1285 : 29 : INIT_LIST_HEAD(&epi->fllink);
1286 : 29 : INIT_LIST_HEAD(&epi->pwqlist);
1287 : 29 : epi->ep = ep;
1288 : : ep_set_ffd(&epi->ffd, tfile, fd);
1289 : 29 : epi->event = *event;
1290 : 29 : epi->nwait = 0;
1291 : 29 : epi->next = EP_UNACTIVE_PTR;
1292 [ - + ]: 29 : if (epi->event.events & EPOLLWAKEUP) {
1293 : 0 : error = ep_create_wakeup_source(epi);
1294 [ # # ]: 0 : if (error)
1295 : : goto error_create_wakeup_source;
1296 : : } else {
1297 : 29 : RCU_INIT_POINTER(epi->ws, NULL);
1298 : : }
1299 : :
1300 : : /* Initialize the poll table using the queue callback */
1301 : 29 : epq.epi = epi;
1302 : : init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1303 : :
1304 : : /*
1305 : : * Attach the item to the poll hooks and get current event bits.
1306 : : * We can safely use the file* here because its usage count has
1307 : : * been increased by the caller of this function. Note that after
1308 : : * this operation completes, the poll callback can start hitting
1309 : : * the new item.
1310 : : */
1311 : : revents = ep_item_poll(epi, &epq.pt);
1312 : :
1313 : : /*
1314 : : * We have to check if something went wrong during the poll wait queue
1315 : : * install process. Namely an allocation for a wait queue failed due
1316 : : * high memory pressure.
1317 : : */
1318 : : error = -ENOMEM;
1319 [ + - ]: 29 : if (epi->nwait < 0)
1320 : : goto error_unregister;
1321 : :
1322 : : /* Add the current item to the list of active epoll hook for this file */
1323 : : spin_lock(&tfile->f_lock);
1324 : 29 : list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
1325 : : spin_unlock(&tfile->f_lock);
1326 : :
1327 : : /*
1328 : : * Add the current item to the RB tree. All RB tree operations are
1329 : : * protected by "mtx", and ep_insert() is called with "mtx" held.
1330 : : */
1331 : 29 : ep_rbtree_insert(ep, epi);
1332 : :
1333 : : /* now check if we've created too many backpaths */
1334 : : error = -EINVAL;
1335 [ - + ][ # # ]: 29 : if (full_check && reverse_path_check())
1336 : : goto error_remove_epi;
1337 : :
1338 : : /* We have to drop the new item inside our item list to keep track of it */
1339 : 29 : spin_lock_irqsave(&ep->lock, flags);
1340 : :
1341 : : /* If the file is already "ready" we drop it inside the ready list */
1342 [ - + ][ # # ]: 29 : if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
1343 : 0 : list_add_tail(&epi->rdllink, &ep->rdllist);
1344 : : ep_pm_stay_awake(epi);
1345 : :
1346 : : /* Notify waiting tasks that events are available */
1347 [ # # ]: 29 : if (waitqueue_active(&ep->wq))
1348 : 0 : wake_up_locked(&ep->wq);
1349 [ # # ]: 0 : if (waitqueue_active(&ep->poll_wait))
1350 : : pwake++;
1351 : : }
1352 : :
1353 : : spin_unlock_irqrestore(&ep->lock, flags);
1354 : :
1355 : 29 : atomic_long_inc(&ep->user->epoll_watches);
1356 : :
1357 : : /* We have to call this outside the lock */
1358 [ - + ]: 29 : if (pwake)
1359 : 0 : ep_poll_safewake(&ep->poll_wait);
1360 : :
1361 : : return 0;
1362 : :
1363 : : error_remove_epi:
1364 : : spin_lock(&tfile->f_lock);
1365 : : list_del_rcu(&epi->fllink);
1366 : : spin_unlock(&tfile->f_lock);
1367 : :
1368 : 0 : rb_erase(&epi->rbn, &ep->rbr);
1369 : :
1370 : : error_unregister:
1371 : 0 : ep_unregister_pollwait(ep, epi);
1372 : :
1373 : : /*
1374 : : * We need to do this because an event could have been arrived on some
1375 : : * allocated wait queue. Note that we don't care about the ep->ovflist
1376 : : * list, since that is used/cleaned only inside a section bound by "mtx".
1377 : : * And ep_insert() is called with "mtx" held.
1378 : : */
1379 : 0 : spin_lock_irqsave(&ep->lock, flags);
1380 [ # # ]: 0 : if (ep_is_linked(&epi->rdllink))
1381 : : list_del_init(&epi->rdllink);
1382 : : spin_unlock_irqrestore(&ep->lock, flags);
1383 : :
1384 : 0 : wakeup_source_unregister(ep_wakeup_source(epi));
1385 : :
1386 : : error_create_wakeup_source:
1387 : 0 : kmem_cache_free(epi_cache, epi);
1388 : :
1389 : 0 : return error;
1390 : : }
1391 : :
1392 : : /*
1393 : : * Modify the interest event mask by dropping an event if the new mask
1394 : : * has a match in the current file status. Must be called with "mtx" held.
1395 : : */
1396 : 0 : static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1397 : : {
1398 : : int pwake = 0;
1399 : : unsigned int revents;
1400 : : poll_table pt;
1401 : :
1402 : : init_poll_funcptr(&pt, NULL);
1403 : :
1404 : : /*
1405 : : * Set the new event interest mask before calling f_op->poll();
1406 : : * otherwise we might miss an event that happens between the
1407 : : * f_op->poll() call and the new event set registering.
1408 : : */
1409 : 0 : epi->event.events = event->events; /* need barrier below */
1410 : 0 : epi->event.data = event->data; /* protected by mtx */
1411 [ # # ]: 0 : if (epi->event.events & EPOLLWAKEUP) {
1412 [ # # ]: 0 : if (!ep_has_wakeup_source(epi))
1413 : 0 : ep_create_wakeup_source(epi);
1414 [ # # ]: 0 : } else if (ep_has_wakeup_source(epi)) {
1415 : 0 : ep_destroy_wakeup_source(epi);
1416 : : }
1417 : :
1418 : : /*
1419 : : * The following barrier has two effects:
1420 : : *
1421 : : * 1) Flush epi changes above to other CPUs. This ensures
1422 : : * we do not miss events from ep_poll_callback if an
1423 : : * event occurs immediately after we call f_op->poll().
1424 : : * We need this because we did not take ep->lock while
1425 : : * changing epi above (but ep_poll_callback does take
1426 : : * ep->lock).
1427 : : *
1428 : : * 2) We also need to ensure we do not miss _past_ events
1429 : : * when calling f_op->poll(). This barrier also
1430 : : * pairs with the barrier in wq_has_sleeper (see
1431 : : * comments for wq_has_sleeper).
1432 : : *
1433 : : * This barrier will now guarantee ep_poll_callback or f_op->poll
1434 : : * (or both) will notice the readiness of an item.
1435 : : */
1436 : 0 : smp_mb();
1437 : :
1438 : : /*
1439 : : * Get current event bits. We can safely use the file* here because
1440 : : * its usage count has been increased by the caller of this function.
1441 : : */
1442 : : revents = ep_item_poll(epi, &pt);
1443 : :
1444 : : /*
1445 : : * If the item is "hot" and it is not registered inside the ready
1446 : : * list, push it inside.
1447 : : */
1448 [ # # ]: 0 : if (revents & event->events) {
1449 : : spin_lock_irq(&ep->lock);
1450 [ # # ]: 0 : if (!ep_is_linked(&epi->rdllink)) {
1451 : 0 : list_add_tail(&epi->rdllink, &ep->rdllist);
1452 : : ep_pm_stay_awake(epi);
1453 : :
1454 : : /* Notify waiting tasks that events are available */
1455 [ # # ]: 0 : if (waitqueue_active(&ep->wq))
1456 : 0 : wake_up_locked(&ep->wq);
1457 [ # # ]: 0 : if (waitqueue_active(&ep->poll_wait))
1458 : : pwake++;
1459 : : }
1460 : : spin_unlock_irq(&ep->lock);
1461 : : }
1462 : :
1463 : : /* We have to call this outside the lock */
1464 [ # # ]: 0 : if (pwake)
1465 : 0 : ep_poll_safewake(&ep->poll_wait);
1466 : :
1467 : 0 : return 0;
1468 : : }
1469 : :
1470 : 0 : static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1471 : : void *priv)
1472 : : {
1473 : : struct ep_send_events_data *esed = priv;
1474 : : int eventcnt;
1475 : : unsigned int revents;
1476 : 2065 : struct epitem *epi;
1477 : : struct epoll_event __user *uevent;
1478 : : struct wakeup_source *ws;
1479 : : poll_table pt;
1480 : :
1481 : : init_poll_funcptr(&pt, NULL);
1482 : :
1483 : : /*
1484 : : * We can loop without lock because we are passed a task private list.
1485 : : * Items cannot vanish during the loop because ep_scan_ready_list() is
1486 : : * holding "mtx" during this call.
1487 : : */
1488 [ + + ]: 4056 : for (eventcnt = 0, uevent = esed->events;
1489 [ + + ]: 2066 : !list_empty(head) && eventcnt < esed->maxevents;) {
1490 : : epi = list_first_entry(head, struct epitem, rdllink);
1491 : :
1492 : : /*
1493 : : * Activate ep->ws before deactivating epi->ws to prevent
1494 : : * triggering auto-suspend here (in case we reactive epi->ws
1495 : : * below).
1496 : : *
1497 : : * This could be rearranged to delay the deactivation of epi->ws
1498 : : * instead, but then epi->ws would temporarily be out of sync
1499 : : * with ep_is_linked().
1500 : : */
1501 : : ws = ep_wakeup_source(epi);
1502 [ - + ]: 2065 : if (ws) {
1503 [ # # ]: 0 : if (ws->active)
1504 : 0 : __pm_stay_awake(ep->ws);
1505 : 0 : __pm_relax(ws);
1506 : : }
1507 : :
1508 : 2065 : list_del_init(&epi->rdllink);
1509 : :
1510 : : revents = ep_item_poll(epi, &pt);
1511 : :
1512 : : /*
1513 : : * If the event mask intersect the caller-requested one,
1514 : : * deliver the event to userspace. Again, ep_scan_ready_list()
1515 : : * is holding "mtx", so no operations coming from userspace
1516 : : * can change the item.
1517 : : */
1518 [ + + ]: 2066 : if (revents) {
1519 [ + - - + ]: 2080 : if (__put_user(revents, &uevent->events) ||
1520 : 1040 : __put_user(epi->event.data, &uevent->data)) {
1521 : : list_add(&epi->rdllink, head);
1522 : : ep_pm_stay_awake(epi);
1523 [ # # ]: 0 : return eventcnt ? eventcnt : -EFAULT;
1524 : : }
1525 : 1040 : eventcnt++;
1526 : 1040 : uevent++;
1527 [ - + ]: 1040 : if (epi->event.events & EPOLLONESHOT)
1528 : 0 : epi->event.events &= EP_PRIVATE_BITS;
1529 [ + - ]: 1040 : else if (!(epi->event.events & EPOLLET)) {
1530 : : /*
1531 : : * If this file has been added with Level
1532 : : * Trigger mode, we need to insert back inside
1533 : : * the ready list, so that the next call to
1534 : : * epoll_wait() will check again the events
1535 : : * availability. At this point, no one can insert
1536 : : * into ep->rdllist besides us. The epoll_ctl()
1537 : : * callers are locked out by
1538 : : * ep_scan_ready_list() holding "mtx" and the
1539 : : * poll callback will queue them in ep->ovflist.
1540 : : */
1541 : 1040 : list_add_tail(&epi->rdllink, &ep->rdllist);
1542 : : ep_pm_stay_awake(epi);
1543 : : }
1544 : : }
1545 : : }
1546 : :
1547 : : return eventcnt;
1548 : : }
1549 : :
1550 : : static int ep_send_events(struct eventpoll *ep,
1551 : : struct epoll_event __user *events, int maxevents)
1552 : : {
1553 : : struct ep_send_events_data esed;
1554 : :
1555 : 1990 : esed.maxevents = maxevents;
1556 : 1990 : esed.events = events;
1557 : :
1558 : 1990 : return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
1559 : : }
1560 : :
1561 : : static inline struct timespec ep_set_mstimeout(long ms)
1562 : : {
1563 : 88 : struct timespec now, ts = {
1564 : 44 : .tv_sec = ms / MSEC_PER_SEC,
1565 : 44 : .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1566 : : };
1567 : :
1568 : 44 : ktime_get_ts(&now);
1569 : 44 : return timespec_add_safe(now, ts);
1570 : : }
1571 : :
1572 : : /**
1573 : : * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1574 : : * event buffer.
1575 : : *
1576 : : * @ep: Pointer to the eventpoll context.
1577 : : * @events: Pointer to the userspace buffer where the ready events should be
1578 : : * stored.
1579 : : * @maxevents: Size (in terms of number of events) of the caller event buffer.
1580 : : * @timeout: Maximum timeout for the ready events fetch operation, in
1581 : : * milliseconds. If the @timeout is zero, the function will not block,
1582 : : * while if the @timeout is less than zero, the function will block
1583 : : * until at least one event has been retrieved (or an error
1584 : : * occurred).
1585 : : *
1586 : : * Returns: Returns the number of ready events which have been fetched, or an
1587 : : * error code, in case of error.
1588 : : */
1589 : 0 : static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1590 : : int maxevents, long timeout)
1591 : : {
1592 : : int res = 0, eavail, timed_out = 0;
1593 : : unsigned long flags;
1594 : : long slack = 0;
1595 : : wait_queue_t wait;
1596 : : ktime_t expires, *to = NULL;
1597 : :
1598 [ + + ]: 1039 : if (timeout > 0) {
1599 : : struct timespec end_time = ep_set_mstimeout(timeout);
1600 : :
1601 : 44 : slack = select_estimate_accuracy(&end_time);
1602 : : to = &expires;
1603 : 44 : *to = timespec_to_ktime(end_time);
1604 [ - + ]: 995 : } else if (timeout == 0) {
1605 : : /*
1606 : : * Avoid the unnecessary trip to the wait queue loop, if the
1607 : : * caller specified a non blocking operation.
1608 : : */
1609 : : timed_out = 1;
1610 : 0 : spin_lock_irqsave(&ep->lock, flags);
1611 : 0 : goto check_events;
1612 : : }
1613 : :
1614 : : fetch_events:
1615 : 1994 : spin_lock_irqsave(&ep->lock, flags);
1616 : :
1617 [ + + ]: 1994 : if (!ep_events_available(ep)) {
1618 : : /*
1619 : : * We don't have any available event to return to the caller.
1620 : : * We need to sleep here, and we will be wake up by
1621 : : * ep_poll_callback() when events will become available.
1622 : : */
1623 : 983 : init_waitqueue_entry(&wait, current);
1624 : : __add_wait_queue_exclusive(&ep->wq, &wait);
1625 : :
1626 : : for (;;) {
1627 : : /*
1628 : : * We don't want to sleep if the ep_poll_callback() sends us
1629 : : * a wakeup in between. That's why we set the task state
1630 : : * to TASK_INTERRUPTIBLE before doing the checks.
1631 : : */
1632 : 1966 : set_current_state(TASK_INTERRUPTIBLE);
1633 [ + + ][ + - ]: 1966 : if (ep_events_available(ep) || timed_out)
1634 : : break;
1635 [ + + ]: 987 : if (signal_pending(current)) {
1636 : : res = -EINTR;
1637 : : break;
1638 : : }
1639 : :
1640 : : spin_unlock_irqrestore(&ep->lock, flags);
1641 [ - + ]: 983 : if (!freezable_schedule_hrtimeout_range(to, slack,
1642 : : HRTIMER_MODE_ABS))
1643 : : timed_out = 1;
1644 : :
1645 : 983 : spin_lock_irqsave(&ep->lock, flags);
1646 : 983 : }
1647 : : __remove_wait_queue(&ep->wq, &wait);
1648 : :
1649 : 983 : set_current_state(TASK_RUNNING);
1650 : : }
1651 : : check_events:
1652 : : /* Is it worth to try to dig for events ? */
1653 : : eavail = ep_events_available(ep);
1654 : :
1655 : : spin_unlock_irqrestore(&ep->lock, flags);
1656 : :
1657 : : /*
1658 : : * Try to transfer events to user space. In case we get 0 events and
1659 : : * there's still timeout left over, we go trying again in search of
1660 : : * more luck.
1661 : : */
1662 [ + + + + ]: 3984 : if (!res && eavail &&
1663 [ + - ]: 955 : !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
1664 : : goto fetch_events;
1665 : :
1666 : 1039 : return res;
1667 : : }
1668 : :
1669 : : /**
1670 : : * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1671 : : * API, to verify that adding an epoll file inside another
1672 : : * epoll structure, does not violate the constraints, in
1673 : : * terms of closed loops, or too deep chains (which can
1674 : : * result in excessive stack usage).
1675 : : *
1676 : : * @priv: Pointer to the epoll file to be currently checked.
1677 : : * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1678 : : * data structure pointer.
1679 : : * @call_nests: Current dept of the @ep_call_nested() call stack.
1680 : : *
1681 : : * Returns: Returns zero if adding the epoll @file inside current epoll
1682 : : * structure @ep does not violate the constraints, or -1 otherwise.
1683 : : */
1684 : 0 : static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1685 : : {
1686 : : int error = 0;
1687 : : struct file *file = priv;
1688 : 0 : struct eventpoll *ep = file->private_data;
1689 : : struct eventpoll *ep_tovisit;
1690 : : struct rb_node *rbp;
1691 : : struct epitem *epi;
1692 : :
1693 : 0 : mutex_lock_nested(&ep->mtx, call_nests + 1);
1694 : 0 : ep->visited = 1;
1695 : 0 : list_add(&ep->visited_list_link, &visited_list);
1696 [ # # ]: 0 : for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1697 : : epi = rb_entry(rbp, struct epitem, rbn);
1698 [ # # ]: 0 : if (unlikely(is_file_epoll(epi->ffd.file))) {
1699 : 0 : ep_tovisit = epi->ffd.file->private_data;
1700 [ # # ]: 0 : if (ep_tovisit->visited)
1701 : 0 : continue;
1702 : 0 : error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1703 : : ep_loop_check_proc, epi->ffd.file,
1704 : 0 : ep_tovisit, current);
1705 [ # # ]: 0 : if (error != 0)
1706 : : break;
1707 : : } else {
1708 : : /*
1709 : : * If we've reached a file that is not associated with
1710 : : * an ep, then we need to check if the newly added
1711 : : * links are going to add too many wakeup paths. We do
1712 : : * this by adding it to the tfile_check_list, if it's
1713 : : * not already there, and calling reverse_path_check()
1714 : : * during ep_insert().
1715 : : */
1716 [ # # ]: 0 : if (list_empty(&epi->ffd.file->f_tfile_llink))
1717 : : list_add(&epi->ffd.file->f_tfile_llink,
1718 : : &tfile_check_list);
1719 : : }
1720 : : }
1721 : 0 : mutex_unlock(&ep->mtx);
1722 : :
1723 : 0 : return error;
1724 : : }
1725 : :
1726 : : /**
1727 : : * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1728 : : * another epoll file (represented by @ep) does not create
1729 : : * closed loops or too deep chains.
1730 : : *
1731 : : * @ep: Pointer to the epoll private data structure.
1732 : : * @file: Pointer to the epoll file to be checked.
1733 : : *
1734 : : * Returns: Returns zero if adding the epoll @file inside current epoll
1735 : : * structure @ep does not violate the constraints, or -1 otherwise.
1736 : : */
1737 : 0 : static int ep_loop_check(struct eventpoll *ep, struct file *file)
1738 : : {
1739 : : int ret;
1740 : : struct eventpoll *ep_cur, *ep_next;
1741 : :
1742 : 0 : ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1743 : 0 : ep_loop_check_proc, file, ep, current);
1744 : : /* clear visited list */
1745 [ # # ]: 0 : list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1746 : : visited_list_link) {
1747 : 0 : ep_cur->visited = 0;
1748 : : list_del(&ep_cur->visited_list_link);
1749 : : }
1750 : 0 : return ret;
1751 : : }
1752 : :
1753 : 0 : static void clear_tfile_check_list(void)
1754 : : {
1755 : : struct file *file;
1756 : :
1757 : : /* first clear the tfile_check_list */
1758 [ # # ]: 0 : while (!list_empty(&tfile_check_list)) {
1759 : : file = list_first_entry(&tfile_check_list, struct file,
1760 : : f_tfile_llink);
1761 : 0 : list_del_init(&file->f_tfile_llink);
1762 : : }
1763 : : INIT_LIST_HEAD(&tfile_check_list);
1764 : 0 : }
1765 : :
1766 : : /*
1767 : : * Open an eventpoll file descriptor.
1768 : : */
1769 : 0 : SYSCALL_DEFINE1(epoll_create1, int, flags)
1770 : : {
1771 : : int error, fd;
1772 : 64 : struct eventpoll *ep = NULL;
1773 : : struct file *file;
1774 : :
1775 : : /* Check the EPOLL_* constant for consistency. */
1776 : : BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1777 : :
1778 [ + - ]: 64 : if (flags & ~EPOLL_CLOEXEC)
1779 : : return -EINVAL;
1780 : : /*
1781 : : * Create the internal data structure ("struct eventpoll").
1782 : : */
1783 : 64 : error = ep_alloc(&ep);
1784 [ + - ]: 64 : if (error < 0)
1785 : : return error;
1786 : : /*
1787 : : * Creates all the items needed to setup an eventpoll file. That is,
1788 : : * a file structure and a free file descriptor.
1789 : : */
1790 : 64 : fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1791 [ + - ]: 64 : if (fd < 0) {
1792 : : error = fd;
1793 : : goto out_free_ep;
1794 : : }
1795 : 64 : file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
1796 : : O_RDWR | (flags & O_CLOEXEC));
1797 [ - + ]: 64 : if (IS_ERR(file)) {
1798 : : error = PTR_ERR(file);
1799 : : goto out_free_fd;
1800 : : }
1801 : 64 : ep->file = file;
1802 : 64 : fd_install(fd, file);
1803 : : return fd;
1804 : :
1805 : : out_free_fd:
1806 : 0 : put_unused_fd(fd);
1807 : : out_free_ep:
1808 : 0 : ep_free(ep);
1809 : : return error;
1810 : : }
1811 : :
1812 : 0 : SYSCALL_DEFINE1(epoll_create, int, size)
1813 : : {
1814 [ + + ]: 34 : if (size <= 0)
1815 : : return -EINVAL;
1816 : :
1817 : 33 : return sys_epoll_create1(0);
1818 : : }
1819 : :
1820 : : /*
1821 : : * The following function implements the controller interface for
1822 : : * the eventpoll file that enables the insertion/removal/change of
1823 : : * file descriptors inside the interest set.
1824 : : */
1825 : 0 : SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1826 : : struct epoll_event __user *, event)
1827 : : {
1828 : : int error;
1829 : : int full_check = 0;
1830 : : struct fd f, tf;
1831 : 54 : struct eventpoll *ep;
1832 : : struct epitem *epi;
1833 : : struct epoll_event epds;
1834 : : struct eventpoll *tep = NULL;
1835 : :
1836 : : error = -EFAULT;
1837 [ + + ][ + + ]: 23123 : if (ep_op_has_event(op) &&
1838 : : copy_from_user(&epds, event, sizeof(struct epoll_event)))
1839 : : goto error_return;
1840 : :
1841 : : error = -EBADF;
1842 : 10806 : f = fdget(epfd);
1843 [ + + ]: 10806 : if (!f.file)
1844 : : goto error_return;
1845 : :
1846 : : /* Get the "struct file *" for the target file */
1847 : 3894 : tf = fdget(fd);
1848 [ + + ]: 3894 : if (!tf.file)
1849 : : goto error_fput;
1850 : :
1851 : : /* The target file descriptor must support poll */
1852 : : error = -EPERM;
1853 [ + - ]: 54 : if (!tf.file->f_op->poll)
1854 : : goto error_tgt_fput;
1855 : :
1856 : : /* Check if EPOLLWAKEUP is allowed */
1857 : : ep_take_care_of_epollwakeup(&epds);
1858 : :
1859 : : /*
1860 : : * We have to check that the file structure underneath the file descriptor
1861 : : * the user passed to us _is_ an eventpoll file. And also we do not permit
1862 : : * adding an epoll file descriptor inside itself.
1863 : : */
1864 : : error = -EINVAL;
1865 [ + - ][ + - ]: 54 : if (f.file == tf.file || !is_file_epoll(f.file))
1866 : : goto error_tgt_fput;
1867 : :
1868 : : /*
1869 : : * At this point it is safe to assume that the "private_data" contains
1870 : : * our own data structure.
1871 : : */
1872 : 54 : ep = f.file->private_data;
1873 : :
1874 : : /*
1875 : : * When we insert an epoll file descriptor, inside another epoll file
1876 : : * descriptor, there is the change of creating closed loops, which are
1877 : : * better be handled here, than in more critical paths. While we are
1878 : : * checking for loops we also determine the list of files reachable
1879 : : * and hang them on the tfile_check_list, so we can check that we
1880 : : * haven't created too many possible wakeup paths.
1881 : : *
1882 : : * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
1883 : : * the epoll file descriptor is attaching directly to a wakeup source,
1884 : : * unless the epoll file descriptor is nested. The purpose of taking the
1885 : : * 'epmutex' on add is to prevent complex toplogies such as loops and
1886 : : * deep wakeup paths from forming in parallel through multiple
1887 : : * EPOLL_CTL_ADD operations.
1888 : : */
1889 : 54 : mutex_lock_nested(&ep->mtx, 0);
1890 [ + + ]: 54 : if (op == EPOLL_CTL_ADD) {
1891 [ + - ][ - + ]: 29 : if (!list_empty(&f.file->f_ep_links) ||
1892 : : is_file_epoll(tf.file)) {
1893 : : full_check = 1;
1894 : 0 : mutex_unlock(&ep->mtx);
1895 : 0 : mutex_lock(&epmutex);
1896 [ # # ]: 0 : if (is_file_epoll(tf.file)) {
1897 : : error = -ELOOP;
1898 [ # # ]: 0 : if (ep_loop_check(ep, tf.file) != 0) {
1899 : 0 : clear_tfile_check_list();
1900 : : goto error_tgt_fput;
1901 : : }
1902 : : } else
1903 : 0 : list_add(&tf.file->f_tfile_llink,
1904 : : &tfile_check_list);
1905 : 0 : mutex_lock_nested(&ep->mtx, 0);
1906 [ # # ]: 0 : if (is_file_epoll(tf.file)) {
1907 : 0 : tep = tf.file->private_data;
1908 : 0 : mutex_lock_nested(&tep->mtx, 1);
1909 : : }
1910 : : }
1911 : : }
1912 : :
1913 : : /*
1914 : : * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1915 : : * above, we can be sure to be able to use the item looked up by
1916 : : * ep_find() till we release the mutex.
1917 : : */
1918 : 54 : epi = ep_find(ep, tf.file, fd);
1919 : :
1920 : : error = -EINVAL;
1921 [ + + - - ]: 54 : switch (op) {
1922 : : case EPOLL_CTL_ADD:
1923 [ + - ]: 29 : if (!epi) {
1924 : 29 : epds.events |= POLLERR | POLLHUP;
1925 : 29 : error = ep_insert(ep, &epds, tf.file, fd, full_check);
1926 : : } else
1927 : : error = -EEXIST;
1928 [ - + ]: 29 : if (full_check)
1929 : 0 : clear_tfile_check_list();
1930 : : break;
1931 : : case EPOLL_CTL_DEL:
1932 [ + - ]: 25 : if (epi)
1933 : 25 : error = ep_remove(ep, epi);
1934 : : else
1935 : : error = -ENOENT;
1936 : : break;
1937 : : case EPOLL_CTL_MOD:
1938 [ # # ]: 0 : if (epi) {
1939 : 0 : epds.events |= POLLERR | POLLHUP;
1940 : 0 : error = ep_modify(ep, epi, &epds);
1941 : : } else
1942 : : error = -ENOENT;
1943 : : break;
1944 : : }
1945 [ - + ]: 54 : if (tep != NULL)
1946 : 0 : mutex_unlock(&tep->mtx);
1947 : 54 : mutex_unlock(&ep->mtx);
1948 : :
1949 : : error_tgt_fput:
1950 [ - + ]: 13932 : if (full_check)
1951 : 0 : mutex_unlock(&epmutex);
1952 : :
1953 : : fdput(tf);
1954 : : error_fput:
1955 : : fdput(f);
1956 : : error_return:
1957 : :
1958 : : return error;
1959 : : }
1960 : :
1961 : : /*
1962 : : * Implement the event wait interface for the eventpoll file. It is the kernel
1963 : : * part of the user space epoll_wait(2).
1964 : : */
1965 : 0 : SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1966 : : int, maxevents, int, timeout)
1967 : : {
1968 : : int error;
1969 : : struct fd f;
1970 : : struct eventpoll *ep;
1971 : :
1972 : : /* The maximum number of event must be greater than zero */
1973 [ + - ]: 1039 : if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1974 : : return -EINVAL;
1975 : :
1976 : : /* Verify that the area passed by the user is writeable */
1977 [ + - ]: 1039 : if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
1978 : : return -EFAULT;
1979 : :
1980 : : /* Get the "struct file *" for the eventpoll file */
1981 : 1039 : f = fdget(epfd);
1982 [ + - ]: 1039 : if (!f.file)
1983 : : return -EBADF;
1984 : :
1985 : : /*
1986 : : * We have to check that the file structure underneath the fd
1987 : : * the user passed to us _is_ an eventpoll file.
1988 : : */
1989 : : error = -EINVAL;
1990 [ + - ]: 1039 : if (!is_file_epoll(f.file))
1991 : : goto error_fput;
1992 : :
1993 : : /*
1994 : : * At this point it is safe to assume that the "private_data" contains
1995 : : * our own data structure.
1996 : : */
1997 : 1039 : ep = f.file->private_data;
1998 : :
1999 : : /* Time to fish for events ... */
2000 : 1039 : error = ep_poll(ep, events, maxevents, timeout);
2001 : :
2002 : : error_fput:
2003 : : fdput(f);
2004 : : return error;
2005 : : }
2006 : :
2007 : : /*
2008 : : * Implement the event wait interface for the eventpoll file. It is the kernel
2009 : : * part of the user space epoll_pwait(2).
2010 : : */
2011 : 0 : SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2012 : : int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2013 : : size_t, sigsetsize)
2014 : : {
2015 : : int error;
2016 : : sigset_t ksigmask, sigsaved;
2017 : :
2018 : : /*
2019 : : * If the caller wants a certain signal mask to be set during the wait,
2020 : : * we apply it here.
2021 : : */
2022 [ # # ]: 0 : if (sigmask) {
2023 [ # # ]: 0 : if (sigsetsize != sizeof(sigset_t))
2024 : : return -EINVAL;
2025 [ # # ]: 0 : if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
2026 : : return -EFAULT;
2027 : 0 : sigsaved = current->blocked;
2028 : 0 : set_current_blocked(&ksigmask);
2029 : : }
2030 : :
2031 : 0 : error = sys_epoll_wait(epfd, events, maxevents, timeout);
2032 : :
2033 : : /*
2034 : : * If we changed the signal mask, we need to restore the original one.
2035 : : * In case we've got a signal while waiting, we do not restore the
2036 : : * signal mask yet, and we allow do_signal() to deliver the signal on
2037 : : * the way back to userspace, before the signal mask is restored.
2038 : : */
2039 [ # # ]: 0 : if (sigmask) {
2040 [ # # ]: 0 : if (error == -EINTR) {
2041 : 0 : memcpy(¤t->saved_sigmask, &sigsaved,
2042 : : sizeof(sigsaved));
2043 : : set_restore_sigmask();
2044 : : } else
2045 : 0 : set_current_blocked(&sigsaved);
2046 : : }
2047 : :
2048 : : return error;
2049 : : }
2050 : :
2051 : : #ifdef CONFIG_COMPAT
2052 : : COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2053 : : struct epoll_event __user *, events,
2054 : : int, maxevents, int, timeout,
2055 : : const compat_sigset_t __user *, sigmask,
2056 : : compat_size_t, sigsetsize)
2057 : : {
2058 : : long err;
2059 : : compat_sigset_t csigmask;
2060 : : sigset_t ksigmask, sigsaved;
2061 : :
2062 : : /*
2063 : : * If the caller wants a certain signal mask to be set during the wait,
2064 : : * we apply it here.
2065 : : */
2066 : : if (sigmask) {
2067 : : if (sigsetsize != sizeof(compat_sigset_t))
2068 : : return -EINVAL;
2069 : : if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
2070 : : return -EFAULT;
2071 : : sigset_from_compat(&ksigmask, &csigmask);
2072 : : sigsaved = current->blocked;
2073 : : set_current_blocked(&ksigmask);
2074 : : }
2075 : :
2076 : : err = sys_epoll_wait(epfd, events, maxevents, timeout);
2077 : :
2078 : : /*
2079 : : * If we changed the signal mask, we need to restore the original one.
2080 : : * In case we've got a signal while waiting, we do not restore the
2081 : : * signal mask yet, and we allow do_signal() to deliver the signal on
2082 : : * the way back to userspace, before the signal mask is restored.
2083 : : */
2084 : : if (sigmask) {
2085 : : if (err == -EINTR) {
2086 : : memcpy(¤t->saved_sigmask, &sigsaved,
2087 : : sizeof(sigsaved));
2088 : : set_restore_sigmask();
2089 : : } else
2090 : : set_current_blocked(&sigsaved);
2091 : : }
2092 : :
2093 : : return err;
2094 : : }
2095 : : #endif
2096 : :
2097 : 0 : static int __init eventpoll_init(void)
2098 : : {
2099 : : struct sysinfo si;
2100 : :
2101 : 0 : si_meminfo(&si);
2102 : : /*
2103 : : * Allows top 4% of lomem to be allocated for epoll watches (per user).
2104 : : */
2105 : 0 : max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2106 : : EP_ITEM_COST;
2107 [ # # ]: 0 : BUG_ON(max_user_watches < 0);
2108 : :
2109 : : /*
2110 : : * Initialize the structure used to perform epoll file descriptor
2111 : : * inclusion loops checks.
2112 : : */
2113 : : ep_nested_calls_init(&poll_loop_ncalls);
2114 : :
2115 : : /* Initialize the structure used to perform safe poll wait head wake ups */
2116 : : ep_nested_calls_init(&poll_safewake_ncalls);
2117 : :
2118 : : /* Initialize the structure used to perform file's f_op->poll() calls */
2119 : : ep_nested_calls_init(&poll_readywalk_ncalls);
2120 : :
2121 : : /*
2122 : : * We can have many thousands of epitems, so prevent this from
2123 : : * using an extra cache line on 64-bit (and smaller) CPUs
2124 : : */
2125 : : BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2126 : :
2127 : : /* Allocates slab cache used to allocate "struct epitem" items */
2128 : 0 : epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
2129 : : 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2130 : :
2131 : : /* Allocates slab cache used to allocate "struct eppoll_entry" */
2132 : 0 : pwq_cache = kmem_cache_create("eventpoll_pwq",
2133 : : sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
2134 : :
2135 : 0 : return 0;
2136 : : }
2137 : : fs_initcall(eventpoll_init);
|
