]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - fs/eventpoll.c
Merge tag 'sound-4.19-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
[mirror_ubuntu-eoan-kernel.git] / fs / eventpoll.c
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/signal.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/uaccess.h>
38 #include <asm/io.h>
39 #include <asm/mman.h>
40 #include <linux/atomic.h>
41 #include <linux/proc_fs.h>
42 #include <linux/seq_file.h>
43 #include <linux/compat.h>
44 #include <linux/rculist.h>
45 #include <net/busy_poll.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->wq.lock (spinlock)
54 *
55 * The acquire order is the one listed above, from 1 to 3.
56 * We need a spinlock (ep->wq.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->wq.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 | EPOLLEXCLUSIVE)
97
98 #define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT)
99
100 #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \
101 EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE)
102
103 /* Maximum number of nesting allowed inside epoll sets */
104 #define EP_MAX_NESTS 4
105
106 #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
107
108 #define EP_UNACTIVE_PTR ((void *) -1L)
109
110 #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
111
112 struct epoll_filefd {
113 struct file *file;
114 int fd;
115 } __packed;
116
117 /*
118 * Structure used to track possible nested calls, for too deep recursions
119 * and loop cycles.
120 */
121 struct nested_call_node {
122 struct list_head llink;
123 void *cookie;
124 void *ctx;
125 };
126
127 /*
128 * This structure is used as collector for nested calls, to check for
129 * maximum recursion dept and loop cycles.
130 */
131 struct nested_calls {
132 struct list_head tasks_call_list;
133 spinlock_t lock;
134 };
135
136 /*
137 * Each file descriptor added to the eventpoll interface will
138 * have an entry of this type linked to the "rbr" RB tree.
139 * Avoid increasing the size of this struct, there can be many thousands
140 * of these on a server and we do not want this to take another cache line.
141 */
142 struct epitem {
143 union {
144 /* RB tree node links this structure to the eventpoll RB tree */
145 struct rb_node rbn;
146 /* Used to free the struct epitem */
147 struct rcu_head rcu;
148 };
149
150 /* List header used to link this structure to the eventpoll ready list */
151 struct list_head rdllink;
152
153 /*
154 * Works together "struct eventpoll"->ovflist in keeping the
155 * single linked chain of items.
156 */
157 struct epitem *next;
158
159 /* The file descriptor information this item refers to */
160 struct epoll_filefd ffd;
161
162 /* Number of active wait queue attached to poll operations */
163 int nwait;
164
165 /* List containing poll wait queues */
166 struct list_head pwqlist;
167
168 /* The "container" of this item */
169 struct eventpoll *ep;
170
171 /* List header used to link this item to the "struct file" items list */
172 struct list_head fllink;
173
174 /* wakeup_source used when EPOLLWAKEUP is set */
175 struct wakeup_source __rcu *ws;
176
177 /* The structure that describe the interested events and the source fd */
178 struct epoll_event event;
179 };
180
181 /*
182 * This structure is stored inside the "private_data" member of the file
183 * structure and represents the main data structure for the eventpoll
184 * interface.
185 *
186 * Access to it is protected by the lock inside wq.
187 */
188 struct eventpoll {
189 /*
190 * This mutex is used to ensure that files are not removed
191 * while epoll is using them. This is held during the event
192 * collection loop, the file cleanup path, the epoll file exit
193 * code and the ctl operations.
194 */
195 struct mutex mtx;
196
197 /* Wait queue used by sys_epoll_wait() */
198 wait_queue_head_t wq;
199
200 /* Wait queue used by file->poll() */
201 wait_queue_head_t poll_wait;
202
203 /* List of ready file descriptors */
204 struct list_head rdllist;
205
206 /* RB tree root used to store monitored fd structs */
207 struct rb_root_cached rbr;
208
209 /*
210 * This is a single linked list that chains all the "struct epitem" that
211 * happened while transferring ready events to userspace w/out
212 * holding ->wq.lock.
213 */
214 struct epitem *ovflist;
215
216 /* wakeup_source used when ep_scan_ready_list is running */
217 struct wakeup_source *ws;
218
219 /* The user that created the eventpoll descriptor */
220 struct user_struct *user;
221
222 struct file *file;
223
224 /* used to optimize loop detection check */
225 int visited;
226 struct list_head visited_list_link;
227
228 #ifdef CONFIG_NET_RX_BUSY_POLL
229 /* used to track busy poll napi_id */
230 unsigned int napi_id;
231 #endif
232 };
233
234 /* Wait structure used by the poll hooks */
235 struct eppoll_entry {
236 /* List header used to link this structure to the "struct epitem" */
237 struct list_head llink;
238
239 /* The "base" pointer is set to the container "struct epitem" */
240 struct epitem *base;
241
242 /*
243 * Wait queue item that will be linked to the target file wait
244 * queue head.
245 */
246 wait_queue_entry_t wait;
247
248 /* The wait queue head that linked the "wait" wait queue item */
249 wait_queue_head_t *whead;
250 };
251
252 /* Wrapper struct used by poll queueing */
253 struct ep_pqueue {
254 poll_table pt;
255 struct epitem *epi;
256 };
257
258 /* Used by the ep_send_events() function as callback private data */
259 struct ep_send_events_data {
260 int maxevents;
261 struct epoll_event __user *events;
262 int res;
263 };
264
265 /*
266 * Configuration options available inside /proc/sys/fs/epoll/
267 */
268 /* Maximum number of epoll watched descriptors, per user */
269 static long max_user_watches __read_mostly;
270
271 /*
272 * This mutex is used to serialize ep_free() and eventpoll_release_file().
273 */
274 static DEFINE_MUTEX(epmutex);
275
276 /* Used to check for epoll file descriptor inclusion loops */
277 static struct nested_calls poll_loop_ncalls;
278
279 /* Slab cache used to allocate "struct epitem" */
280 static struct kmem_cache *epi_cache __read_mostly;
281
282 /* Slab cache used to allocate "struct eppoll_entry" */
283 static struct kmem_cache *pwq_cache __read_mostly;
284
285 /* Visited nodes during ep_loop_check(), so we can unset them when we finish */
286 static LIST_HEAD(visited_list);
287
288 /*
289 * List of files with newly added links, where we may need to limit the number
290 * of emanating paths. Protected by the epmutex.
291 */
292 static LIST_HEAD(tfile_check_list);
293
294 #ifdef CONFIG_SYSCTL
295
296 #include <linux/sysctl.h>
297
298 static long zero;
299 static long long_max = LONG_MAX;
300
301 struct ctl_table epoll_table[] = {
302 {
303 .procname = "max_user_watches",
304 .data = &max_user_watches,
305 .maxlen = sizeof(max_user_watches),
306 .mode = 0644,
307 .proc_handler = proc_doulongvec_minmax,
308 .extra1 = &zero,
309 .extra2 = &long_max,
310 },
311 { }
312 };
313 #endif /* CONFIG_SYSCTL */
314
315 static const struct file_operations eventpoll_fops;
316
317 static inline int is_file_epoll(struct file *f)
318 {
319 return f->f_op == &eventpoll_fops;
320 }
321
322 /* Setup the structure that is used as key for the RB tree */
323 static inline void ep_set_ffd(struct epoll_filefd *ffd,
324 struct file *file, int fd)
325 {
326 ffd->file = file;
327 ffd->fd = fd;
328 }
329
330 /* Compare RB tree keys */
331 static inline int ep_cmp_ffd(struct epoll_filefd *p1,
332 struct epoll_filefd *p2)
333 {
334 return (p1->file > p2->file ? +1:
335 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
336 }
337
338 /* Tells us if the item is currently linked */
339 static inline int ep_is_linked(struct epitem *epi)
340 {
341 return !list_empty(&epi->rdllink);
342 }
343
344 static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p)
345 {
346 return container_of(p, struct eppoll_entry, wait);
347 }
348
349 /* Get the "struct epitem" from a wait queue pointer */
350 static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p)
351 {
352 return container_of(p, struct eppoll_entry, wait)->base;
353 }
354
355 /* Get the "struct epitem" from an epoll queue wrapper */
356 static inline struct epitem *ep_item_from_epqueue(poll_table *p)
357 {
358 return container_of(p, struct ep_pqueue, pt)->epi;
359 }
360
361 /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
362 static inline int ep_op_has_event(int op)
363 {
364 return op != EPOLL_CTL_DEL;
365 }
366
367 /* Initialize the poll safe wake up structure */
368 static void ep_nested_calls_init(struct nested_calls *ncalls)
369 {
370 INIT_LIST_HEAD(&ncalls->tasks_call_list);
371 spin_lock_init(&ncalls->lock);
372 }
373
374 /**
375 * ep_events_available - Checks if ready events might be available.
376 *
377 * @ep: Pointer to the eventpoll context.
378 *
379 * Returns: Returns a value different than zero if ready events are available,
380 * or zero otherwise.
381 */
382 static inline int ep_events_available(struct eventpoll *ep)
383 {
384 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
385 }
386
387 #ifdef CONFIG_NET_RX_BUSY_POLL
388 static bool ep_busy_loop_end(void *p, unsigned long start_time)
389 {
390 struct eventpoll *ep = p;
391
392 return ep_events_available(ep) || busy_loop_timeout(start_time);
393 }
394
395 /*
396 * Busy poll if globally on and supporting sockets found && no events,
397 * busy loop will return if need_resched or ep_events_available.
398 *
399 * we must do our busy polling with irqs enabled
400 */
401 static void ep_busy_loop(struct eventpoll *ep, int nonblock)
402 {
403 unsigned int napi_id = READ_ONCE(ep->napi_id);
404
405 if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on())
406 napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep);
407 }
408
409 static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep)
410 {
411 if (ep->napi_id)
412 ep->napi_id = 0;
413 }
414
415 /*
416 * Set epoll busy poll NAPI ID from sk.
417 */
418 static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
419 {
420 struct eventpoll *ep;
421 unsigned int napi_id;
422 struct socket *sock;
423 struct sock *sk;
424 int err;
425
426 if (!net_busy_loop_on())
427 return;
428
429 sock = sock_from_file(epi->ffd.file, &err);
430 if (!sock)
431 return;
432
433 sk = sock->sk;
434 if (!sk)
435 return;
436
437 napi_id = READ_ONCE(sk->sk_napi_id);
438 ep = epi->ep;
439
440 /* Non-NAPI IDs can be rejected
441 * or
442 * Nothing to do if we already have this ID
443 */
444 if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id)
445 return;
446
447 /* record NAPI ID for use in next busy poll */
448 ep->napi_id = napi_id;
449 }
450
451 #else
452
453 static inline void ep_busy_loop(struct eventpoll *ep, int nonblock)
454 {
455 }
456
457 static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep)
458 {
459 }
460
461 static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
462 {
463 }
464
465 #endif /* CONFIG_NET_RX_BUSY_POLL */
466
467 /**
468 * ep_call_nested - Perform a bound (possibly) nested call, by checking
469 * that the recursion limit is not exceeded, and that
470 * the same nested call (by the meaning of same cookie) is
471 * no re-entered.
472 *
473 * @ncalls: Pointer to the nested_calls structure to be used for this call.
474 * @max_nests: Maximum number of allowed nesting calls.
475 * @nproc: Nested call core function pointer.
476 * @priv: Opaque data to be passed to the @nproc callback.
477 * @cookie: Cookie to be used to identify this nested call.
478 * @ctx: This instance context.
479 *
480 * Returns: Returns the code returned by the @nproc callback, or -1 if
481 * the maximum recursion limit has been exceeded.
482 */
483 static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
484 int (*nproc)(void *, void *, int), void *priv,
485 void *cookie, void *ctx)
486 {
487 int error, call_nests = 0;
488 unsigned long flags;
489 struct list_head *lsthead = &ncalls->tasks_call_list;
490 struct nested_call_node *tncur;
491 struct nested_call_node tnode;
492
493 spin_lock_irqsave(&ncalls->lock, flags);
494
495 /*
496 * Try to see if the current task is already inside this wakeup call.
497 * We use a list here, since the population inside this set is always
498 * very much limited.
499 */
500 list_for_each_entry(tncur, lsthead, llink) {
501 if (tncur->ctx == ctx &&
502 (tncur->cookie == cookie || ++call_nests > max_nests)) {
503 /*
504 * Ops ... loop detected or maximum nest level reached.
505 * We abort this wake by breaking the cycle itself.
506 */
507 error = -1;
508 goto out_unlock;
509 }
510 }
511
512 /* Add the current task and cookie to the list */
513 tnode.ctx = ctx;
514 tnode.cookie = cookie;
515 list_add(&tnode.llink, lsthead);
516
517 spin_unlock_irqrestore(&ncalls->lock, flags);
518
519 /* Call the nested function */
520 error = (*nproc)(priv, cookie, call_nests);
521
522 /* Remove the current task from the list */
523 spin_lock_irqsave(&ncalls->lock, flags);
524 list_del(&tnode.llink);
525 out_unlock:
526 spin_unlock_irqrestore(&ncalls->lock, flags);
527
528 return error;
529 }
530
531 /*
532 * As described in commit 0ccf831cb lockdep: annotate epoll
533 * the use of wait queues used by epoll is done in a very controlled
534 * manner. Wake ups can nest inside each other, but are never done
535 * with the same locking. For example:
536 *
537 * dfd = socket(...);
538 * efd1 = epoll_create();
539 * efd2 = epoll_create();
540 * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
541 * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
542 *
543 * When a packet arrives to the device underneath "dfd", the net code will
544 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
545 * callback wakeup entry on that queue, and the wake_up() performed by the
546 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
547 * (efd1) notices that it may have some event ready, so it needs to wake up
548 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
549 * that ends up in another wake_up(), after having checked about the
550 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
551 * avoid stack blasting.
552 *
553 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
554 * this special case of epoll.
555 */
556 #ifdef CONFIG_DEBUG_LOCK_ALLOC
557
558 static struct nested_calls poll_safewake_ncalls;
559
560 static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
561 {
562 unsigned long flags;
563 wait_queue_head_t *wqueue = (wait_queue_head_t *)cookie;
564
565 spin_lock_irqsave_nested(&wqueue->lock, flags, call_nests + 1);
566 wake_up_locked_poll(wqueue, EPOLLIN);
567 spin_unlock_irqrestore(&wqueue->lock, flags);
568
569 return 0;
570 }
571
572 static void ep_poll_safewake(wait_queue_head_t *wq)
573 {
574 int this_cpu = get_cpu();
575
576 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
577 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
578
579 put_cpu();
580 }
581
582 #else
583
584 static void ep_poll_safewake(wait_queue_head_t *wq)
585 {
586 wake_up_poll(wq, EPOLLIN);
587 }
588
589 #endif
590
591 static void ep_remove_wait_queue(struct eppoll_entry *pwq)
592 {
593 wait_queue_head_t *whead;
594
595 rcu_read_lock();
596 /*
597 * If it is cleared by POLLFREE, it should be rcu-safe.
598 * If we read NULL we need a barrier paired with
599 * smp_store_release() in ep_poll_callback(), otherwise
600 * we rely on whead->lock.
601 */
602 whead = smp_load_acquire(&pwq->whead);
603 if (whead)
604 remove_wait_queue(whead, &pwq->wait);
605 rcu_read_unlock();
606 }
607
608 /*
609 * This function unregisters poll callbacks from the associated file
610 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
611 * ep_free).
612 */
613 static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
614 {
615 struct list_head *lsthead = &epi->pwqlist;
616 struct eppoll_entry *pwq;
617
618 while (!list_empty(lsthead)) {
619 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
620
621 list_del(&pwq->llink);
622 ep_remove_wait_queue(pwq);
623 kmem_cache_free(pwq_cache, pwq);
624 }
625 }
626
627 /* call only when ep->mtx is held */
628 static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
629 {
630 return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
631 }
632
633 /* call only when ep->mtx is held */
634 static inline void ep_pm_stay_awake(struct epitem *epi)
635 {
636 struct wakeup_source *ws = ep_wakeup_source(epi);
637
638 if (ws)
639 __pm_stay_awake(ws);
640 }
641
642 static inline bool ep_has_wakeup_source(struct epitem *epi)
643 {
644 return rcu_access_pointer(epi->ws) ? true : false;
645 }
646
647 /* call when ep->mtx cannot be held (ep_poll_callback) */
648 static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
649 {
650 struct wakeup_source *ws;
651
652 rcu_read_lock();
653 ws = rcu_dereference(epi->ws);
654 if (ws)
655 __pm_stay_awake(ws);
656 rcu_read_unlock();
657 }
658
659 /**
660 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
661 * the scan code, to call f_op->poll(). Also allows for
662 * O(NumReady) performance.
663 *
664 * @ep: Pointer to the epoll private data structure.
665 * @sproc: Pointer to the scan callback.
666 * @priv: Private opaque data passed to the @sproc callback.
667 * @depth: The current depth of recursive f_op->poll calls.
668 * @ep_locked: caller already holds ep->mtx
669 *
670 * Returns: The same integer error code returned by the @sproc callback.
671 */
672 static __poll_t ep_scan_ready_list(struct eventpoll *ep,
673 __poll_t (*sproc)(struct eventpoll *,
674 struct list_head *, void *),
675 void *priv, int depth, bool ep_locked)
676 {
677 __poll_t res;
678 int pwake = 0;
679 struct epitem *epi, *nepi;
680 LIST_HEAD(txlist);
681
682 lockdep_assert_irqs_enabled();
683
684 /*
685 * We need to lock this because we could be hit by
686 * eventpoll_release_file() and epoll_ctl().
687 */
688
689 if (!ep_locked)
690 mutex_lock_nested(&ep->mtx, depth);
691
692 /*
693 * Steal the ready list, and re-init the original one to the
694 * empty list. Also, set ep->ovflist to NULL so that events
695 * happening while looping w/out locks, are not lost. We cannot
696 * have the poll callback to queue directly on ep->rdllist,
697 * because we want the "sproc" callback to be able to do it
698 * in a lockless way.
699 */
700 spin_lock_irq(&ep->wq.lock);
701 list_splice_init(&ep->rdllist, &txlist);
702 ep->ovflist = NULL;
703 spin_unlock_irq(&ep->wq.lock);
704
705 /*
706 * Now call the callback function.
707 */
708 res = (*sproc)(ep, &txlist, priv);
709
710 spin_lock_irq(&ep->wq.lock);
711 /*
712 * During the time we spent inside the "sproc" callback, some
713 * other events might have been queued by the poll callback.
714 * We re-insert them inside the main ready-list here.
715 */
716 for (nepi = ep->ovflist; (epi = nepi) != NULL;
717 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
718 /*
719 * We need to check if the item is already in the list.
720 * During the "sproc" callback execution time, items are
721 * queued into ->ovflist but the "txlist" might already
722 * contain them, and the list_splice() below takes care of them.
723 */
724 if (!ep_is_linked(epi)) {
725 list_add_tail(&epi->rdllink, &ep->rdllist);
726 ep_pm_stay_awake(epi);
727 }
728 }
729 /*
730 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
731 * releasing the lock, events will be queued in the normal way inside
732 * ep->rdllist.
733 */
734 ep->ovflist = EP_UNACTIVE_PTR;
735
736 /*
737 * Quickly re-inject items left on "txlist".
738 */
739 list_splice(&txlist, &ep->rdllist);
740 __pm_relax(ep->ws);
741
742 if (!list_empty(&ep->rdllist)) {
743 /*
744 * Wake up (if active) both the eventpoll wait list and
745 * the ->poll() wait list (delayed after we release the lock).
746 */
747 if (waitqueue_active(&ep->wq))
748 wake_up_locked(&ep->wq);
749 if (waitqueue_active(&ep->poll_wait))
750 pwake++;
751 }
752 spin_unlock_irq(&ep->wq.lock);
753
754 if (!ep_locked)
755 mutex_unlock(&ep->mtx);
756
757 /* We have to call this outside the lock */
758 if (pwake)
759 ep_poll_safewake(&ep->poll_wait);
760
761 return res;
762 }
763
764 static void epi_rcu_free(struct rcu_head *head)
765 {
766 struct epitem *epi = container_of(head, struct epitem, rcu);
767 kmem_cache_free(epi_cache, epi);
768 }
769
770 /*
771 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
772 * all the associated resources. Must be called with "mtx" held.
773 */
774 static int ep_remove(struct eventpoll *ep, struct epitem *epi)
775 {
776 struct file *file = epi->ffd.file;
777
778 lockdep_assert_irqs_enabled();
779
780 /*
781 * Removes poll wait queue hooks.
782 */
783 ep_unregister_pollwait(ep, epi);
784
785 /* Remove the current item from the list of epoll hooks */
786 spin_lock(&file->f_lock);
787 list_del_rcu(&epi->fllink);
788 spin_unlock(&file->f_lock);
789
790 rb_erase_cached(&epi->rbn, &ep->rbr);
791
792 spin_lock_irq(&ep->wq.lock);
793 if (ep_is_linked(epi))
794 list_del_init(&epi->rdllink);
795 spin_unlock_irq(&ep->wq.lock);
796
797 wakeup_source_unregister(ep_wakeup_source(epi));
798 /*
799 * At this point it is safe to free the eventpoll item. Use the union
800 * field epi->rcu, since we are trying to minimize the size of
801 * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
802 * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
803 * use of the rbn field.
804 */
805 call_rcu(&epi->rcu, epi_rcu_free);
806
807 atomic_long_dec(&ep->user->epoll_watches);
808
809 return 0;
810 }
811
812 static void ep_free(struct eventpoll *ep)
813 {
814 struct rb_node *rbp;
815 struct epitem *epi;
816
817 /* We need to release all tasks waiting for these file */
818 if (waitqueue_active(&ep->poll_wait))
819 ep_poll_safewake(&ep->poll_wait);
820
821 /*
822 * We need to lock this because we could be hit by
823 * eventpoll_release_file() while we're freeing the "struct eventpoll".
824 * We do not need to hold "ep->mtx" here because the epoll file
825 * is on the way to be removed and no one has references to it
826 * anymore. The only hit might come from eventpoll_release_file() but
827 * holding "epmutex" is sufficient here.
828 */
829 mutex_lock(&epmutex);
830
831 /*
832 * Walks through the whole tree by unregistering poll callbacks.
833 */
834 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
835 epi = rb_entry(rbp, struct epitem, rbn);
836
837 ep_unregister_pollwait(ep, epi);
838 cond_resched();
839 }
840
841 /*
842 * Walks through the whole tree by freeing each "struct epitem". At this
843 * point we are sure no poll callbacks will be lingering around, and also by
844 * holding "epmutex" we can be sure that no file cleanup code will hit
845 * us during this operation. So we can avoid the lock on "ep->wq.lock".
846 * We do not need to lock ep->mtx, either, we only do it to prevent
847 * a lockdep warning.
848 */
849 mutex_lock(&ep->mtx);
850 while ((rbp = rb_first_cached(&ep->rbr)) != NULL) {
851 epi = rb_entry(rbp, struct epitem, rbn);
852 ep_remove(ep, epi);
853 cond_resched();
854 }
855 mutex_unlock(&ep->mtx);
856
857 mutex_unlock(&epmutex);
858 mutex_destroy(&ep->mtx);
859 free_uid(ep->user);
860 wakeup_source_unregister(ep->ws);
861 kfree(ep);
862 }
863
864 static int ep_eventpoll_release(struct inode *inode, struct file *file)
865 {
866 struct eventpoll *ep = file->private_data;
867
868 if (ep)
869 ep_free(ep);
870
871 return 0;
872 }
873
874 static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
875 void *priv);
876 static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
877 poll_table *pt);
878
879 /*
880 * Differs from ep_eventpoll_poll() in that internal callers already have
881 * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested()
882 * is correctly annotated.
883 */
884 static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt,
885 int depth)
886 {
887 struct eventpoll *ep;
888 bool locked;
889
890 pt->_key = epi->event.events;
891 if (!is_file_epoll(epi->ffd.file))
892 return vfs_poll(epi->ffd.file, pt) & epi->event.events;
893
894 ep = epi->ffd.file->private_data;
895 poll_wait(epi->ffd.file, &ep->poll_wait, pt);
896 locked = pt && (pt->_qproc == ep_ptable_queue_proc);
897
898 return ep_scan_ready_list(epi->ffd.file->private_data,
899 ep_read_events_proc, &depth, depth,
900 locked) & epi->event.events;
901 }
902
903 static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
904 void *priv)
905 {
906 struct epitem *epi, *tmp;
907 poll_table pt;
908 int depth = *(int *)priv;
909
910 init_poll_funcptr(&pt, NULL);
911 depth++;
912
913 list_for_each_entry_safe(epi, tmp, head, rdllink) {
914 if (ep_item_poll(epi, &pt, depth)) {
915 return EPOLLIN | EPOLLRDNORM;
916 } else {
917 /*
918 * Item has been dropped into the ready list by the poll
919 * callback, but it's not actually ready, as far as
920 * caller requested events goes. We can remove it here.
921 */
922 __pm_relax(ep_wakeup_source(epi));
923 list_del_init(&epi->rdllink);
924 }
925 }
926
927 return 0;
928 }
929
930 static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait)
931 {
932 struct eventpoll *ep = file->private_data;
933 int depth = 0;
934
935 /* Insert inside our poll wait queue */
936 poll_wait(file, &ep->poll_wait, wait);
937
938 /*
939 * Proceed to find out if wanted events are really available inside
940 * the ready list.
941 */
942 return ep_scan_ready_list(ep, ep_read_events_proc,
943 &depth, depth, false);
944 }
945
946 #ifdef CONFIG_PROC_FS
947 static void ep_show_fdinfo(struct seq_file *m, struct file *f)
948 {
949 struct eventpoll *ep = f->private_data;
950 struct rb_node *rbp;
951
952 mutex_lock(&ep->mtx);
953 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
954 struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
955 struct inode *inode = file_inode(epi->ffd.file);
956
957 seq_printf(m, "tfd: %8d events: %8x data: %16llx "
958 " pos:%lli ino:%lx sdev:%x\n",
959 epi->ffd.fd, epi->event.events,
960 (long long)epi->event.data,
961 (long long)epi->ffd.file->f_pos,
962 inode->i_ino, inode->i_sb->s_dev);
963 if (seq_has_overflowed(m))
964 break;
965 }
966 mutex_unlock(&ep->mtx);
967 }
968 #endif
969
970 /* File callbacks that implement the eventpoll file behaviour */
971 static const struct file_operations eventpoll_fops = {
972 #ifdef CONFIG_PROC_FS
973 .show_fdinfo = ep_show_fdinfo,
974 #endif
975 .release = ep_eventpoll_release,
976 .poll = ep_eventpoll_poll,
977 .llseek = noop_llseek,
978 };
979
980 /*
981 * This is called from eventpoll_release() to unlink files from the eventpoll
982 * interface. We need to have this facility to cleanup correctly files that are
983 * closed without being removed from the eventpoll interface.
984 */
985 void eventpoll_release_file(struct file *file)
986 {
987 struct eventpoll *ep;
988 struct epitem *epi, *next;
989
990 /*
991 * We don't want to get "file->f_lock" because it is not
992 * necessary. It is not necessary because we're in the "struct file"
993 * cleanup path, and this means that no one is using this file anymore.
994 * So, for example, epoll_ctl() cannot hit here since if we reach this
995 * point, the file counter already went to zero and fget() would fail.
996 * The only hit might come from ep_free() but by holding the mutex
997 * will correctly serialize the operation. We do need to acquire
998 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
999 * from anywhere but ep_free().
1000 *
1001 * Besides, ep_remove() acquires the lock, so we can't hold it here.
1002 */
1003 mutex_lock(&epmutex);
1004 list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) {
1005 ep = epi->ep;
1006 mutex_lock_nested(&ep->mtx, 0);
1007 ep_remove(ep, epi);
1008 mutex_unlock(&ep->mtx);
1009 }
1010 mutex_unlock(&epmutex);
1011 }
1012
1013 static int ep_alloc(struct eventpoll **pep)
1014 {
1015 int error;
1016 struct user_struct *user;
1017 struct eventpoll *ep;
1018
1019 user = get_current_user();
1020 error = -ENOMEM;
1021 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
1022 if (unlikely(!ep))
1023 goto free_uid;
1024
1025 mutex_init(&ep->mtx);
1026 init_waitqueue_head(&ep->wq);
1027 init_waitqueue_head(&ep->poll_wait);
1028 INIT_LIST_HEAD(&ep->rdllist);
1029 ep->rbr = RB_ROOT_CACHED;
1030 ep->ovflist = EP_UNACTIVE_PTR;
1031 ep->user = user;
1032
1033 *pep = ep;
1034
1035 return 0;
1036
1037 free_uid:
1038 free_uid(user);
1039 return error;
1040 }
1041
1042 /*
1043 * Search the file inside the eventpoll tree. The RB tree operations
1044 * are protected by the "mtx" mutex, and ep_find() must be called with
1045 * "mtx" held.
1046 */
1047 static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
1048 {
1049 int kcmp;
1050 struct rb_node *rbp;
1051 struct epitem *epi, *epir = NULL;
1052 struct epoll_filefd ffd;
1053
1054 ep_set_ffd(&ffd, file, fd);
1055 for (rbp = ep->rbr.rb_root.rb_node; rbp; ) {
1056 epi = rb_entry(rbp, struct epitem, rbn);
1057 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
1058 if (kcmp > 0)
1059 rbp = rbp->rb_right;
1060 else if (kcmp < 0)
1061 rbp = rbp->rb_left;
1062 else {
1063 epir = epi;
1064 break;
1065 }
1066 }
1067
1068 return epir;
1069 }
1070
1071 #ifdef CONFIG_CHECKPOINT_RESTORE
1072 static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff)
1073 {
1074 struct rb_node *rbp;
1075 struct epitem *epi;
1076
1077 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1078 epi = rb_entry(rbp, struct epitem, rbn);
1079 if (epi->ffd.fd == tfd) {
1080 if (toff == 0)
1081 return epi;
1082 else
1083 toff--;
1084 }
1085 cond_resched();
1086 }
1087
1088 return NULL;
1089 }
1090
1091 struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd,
1092 unsigned long toff)
1093 {
1094 struct file *file_raw;
1095 struct eventpoll *ep;
1096 struct epitem *epi;
1097
1098 if (!is_file_epoll(file))
1099 return ERR_PTR(-EINVAL);
1100
1101 ep = file->private_data;
1102
1103 mutex_lock(&ep->mtx);
1104 epi = ep_find_tfd(ep, tfd, toff);
1105 if (epi)
1106 file_raw = epi->ffd.file;
1107 else
1108 file_raw = ERR_PTR(-ENOENT);
1109 mutex_unlock(&ep->mtx);
1110
1111 return file_raw;
1112 }
1113 #endif /* CONFIG_CHECKPOINT_RESTORE */
1114
1115 /*
1116 * This is the callback that is passed to the wait queue wakeup
1117 * mechanism. It is called by the stored file descriptors when they
1118 * have events to report.
1119 */
1120 static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1121 {
1122 int pwake = 0;
1123 unsigned long flags;
1124 struct epitem *epi = ep_item_from_wait(wait);
1125 struct eventpoll *ep = epi->ep;
1126 __poll_t pollflags = key_to_poll(key);
1127 int ewake = 0;
1128
1129 spin_lock_irqsave(&ep->wq.lock, flags);
1130
1131 ep_set_busy_poll_napi_id(epi);
1132
1133 /*
1134 * If the event mask does not contain any poll(2) event, we consider the
1135 * descriptor to be disabled. This condition is likely the effect of the
1136 * EPOLLONESHOT bit that disables the descriptor when an event is received,
1137 * until the next EPOLL_CTL_MOD will be issued.
1138 */
1139 if (!(epi->event.events & ~EP_PRIVATE_BITS))
1140 goto out_unlock;
1141
1142 /*
1143 * Check the events coming with the callback. At this stage, not
1144 * every device reports the events in the "key" parameter of the
1145 * callback. We need to be able to handle both cases here, hence the
1146 * test for "key" != NULL before the event match test.
1147 */
1148 if (pollflags && !(pollflags & epi->event.events))
1149 goto out_unlock;
1150
1151 /*
1152 * If we are transferring events to userspace, we can hold no locks
1153 * (because we're accessing user memory, and because of linux f_op->poll()
1154 * semantics). All the events that happen during that period of time are
1155 * chained in ep->ovflist and requeued later on.
1156 */
1157 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
1158 if (epi->next == EP_UNACTIVE_PTR) {
1159 epi->next = ep->ovflist;
1160 ep->ovflist = epi;
1161 if (epi->ws) {
1162 /*
1163 * Activate ep->ws since epi->ws may get
1164 * deactivated at any time.
1165 */
1166 __pm_stay_awake(ep->ws);
1167 }
1168
1169 }
1170 goto out_unlock;
1171 }
1172
1173 /* If this file is already in the ready list we exit soon */
1174 if (!ep_is_linked(epi)) {
1175 list_add_tail(&epi->rdllink, &ep->rdllist);
1176 ep_pm_stay_awake_rcu(epi);
1177 }
1178
1179 /*
1180 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1181 * wait list.
1182 */
1183 if (waitqueue_active(&ep->wq)) {
1184 if ((epi->event.events & EPOLLEXCLUSIVE) &&
1185 !(pollflags & POLLFREE)) {
1186 switch (pollflags & EPOLLINOUT_BITS) {
1187 case EPOLLIN:
1188 if (epi->event.events & EPOLLIN)
1189 ewake = 1;
1190 break;
1191 case EPOLLOUT:
1192 if (epi->event.events & EPOLLOUT)
1193 ewake = 1;
1194 break;
1195 case 0:
1196 ewake = 1;
1197 break;
1198 }
1199 }
1200 wake_up_locked(&ep->wq);
1201 }
1202 if (waitqueue_active(&ep->poll_wait))
1203 pwake++;
1204
1205 out_unlock:
1206 spin_unlock_irqrestore(&ep->wq.lock, flags);
1207
1208 /* We have to call this outside the lock */
1209 if (pwake)
1210 ep_poll_safewake(&ep->poll_wait);
1211
1212 if (!(epi->event.events & EPOLLEXCLUSIVE))
1213 ewake = 1;
1214
1215 if (pollflags & POLLFREE) {
1216 /*
1217 * If we race with ep_remove_wait_queue() it can miss
1218 * ->whead = NULL and do another remove_wait_queue() after
1219 * us, so we can't use __remove_wait_queue().
1220 */
1221 list_del_init(&wait->entry);
1222 /*
1223 * ->whead != NULL protects us from the race with ep_free()
1224 * or ep_remove(), ep_remove_wait_queue() takes whead->lock
1225 * held by the caller. Once we nullify it, nothing protects
1226 * ep/epi or even wait.
1227 */
1228 smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL);
1229 }
1230
1231 return ewake;
1232 }
1233
1234 /*
1235 * This is the callback that is used to add our wait queue to the
1236 * target file wakeup lists.
1237 */
1238 static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1239 poll_table *pt)
1240 {
1241 struct epitem *epi = ep_item_from_epqueue(pt);
1242 struct eppoll_entry *pwq;
1243
1244 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
1245 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1246 pwq->whead = whead;
1247 pwq->base = epi;
1248 if (epi->event.events & EPOLLEXCLUSIVE)
1249 add_wait_queue_exclusive(whead, &pwq->wait);
1250 else
1251 add_wait_queue(whead, &pwq->wait);
1252 list_add_tail(&pwq->llink, &epi->pwqlist);
1253 epi->nwait++;
1254 } else {
1255 /* We have to signal that an error occurred */
1256 epi->nwait = -1;
1257 }
1258 }
1259
1260 static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1261 {
1262 int kcmp;
1263 struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL;
1264 struct epitem *epic;
1265 bool leftmost = true;
1266
1267 while (*p) {
1268 parent = *p;
1269 epic = rb_entry(parent, struct epitem, rbn);
1270 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1271 if (kcmp > 0) {
1272 p = &parent->rb_right;
1273 leftmost = false;
1274 } else
1275 p = &parent->rb_left;
1276 }
1277 rb_link_node(&epi->rbn, parent, p);
1278 rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost);
1279 }
1280
1281
1282
1283 #define PATH_ARR_SIZE 5
1284 /*
1285 * These are the number paths of length 1 to 5, that we are allowing to emanate
1286 * from a single file of interest. For example, we allow 1000 paths of length
1287 * 1, to emanate from each file of interest. This essentially represents the
1288 * potential wakeup paths, which need to be limited in order to avoid massive
1289 * uncontrolled wakeup storms. The common use case should be a single ep which
1290 * is connected to n file sources. In this case each file source has 1 path
1291 * of length 1. Thus, the numbers below should be more than sufficient. These
1292 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1293 * and delete can't add additional paths. Protected by the epmutex.
1294 */
1295 static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1296 static int path_count[PATH_ARR_SIZE];
1297
1298 static int path_count_inc(int nests)
1299 {
1300 /* Allow an arbitrary number of depth 1 paths */
1301 if (nests == 0)
1302 return 0;
1303
1304 if (++path_count[nests] > path_limits[nests])
1305 return -1;
1306 return 0;
1307 }
1308
1309 static void path_count_init(void)
1310 {
1311 int i;
1312
1313 for (i = 0; i < PATH_ARR_SIZE; i++)
1314 path_count[i] = 0;
1315 }
1316
1317 static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1318 {
1319 int error = 0;
1320 struct file *file = priv;
1321 struct file *child_file;
1322 struct epitem *epi;
1323
1324 /* CTL_DEL can remove links here, but that can't increase our count */
1325 rcu_read_lock();
1326 list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
1327 child_file = epi->ep->file;
1328 if (is_file_epoll(child_file)) {
1329 if (list_empty(&child_file->f_ep_links)) {
1330 if (path_count_inc(call_nests)) {
1331 error = -1;
1332 break;
1333 }
1334 } else {
1335 error = ep_call_nested(&poll_loop_ncalls,
1336 EP_MAX_NESTS,
1337 reverse_path_check_proc,
1338 child_file, child_file,
1339 current);
1340 }
1341 if (error != 0)
1342 break;
1343 } else {
1344 printk(KERN_ERR "reverse_path_check_proc: "
1345 "file is not an ep!\n");
1346 }
1347 }
1348 rcu_read_unlock();
1349 return error;
1350 }
1351
1352 /**
1353 * reverse_path_check - The tfile_check_list is list of file *, which have
1354 * links that are proposed to be newly added. We need to
1355 * make sure that those added links don't add too many
1356 * paths such that we will spend all our time waking up
1357 * eventpoll objects.
1358 *
1359 * Returns: Returns zero if the proposed links don't create too many paths,
1360 * -1 otherwise.
1361 */
1362 static int reverse_path_check(void)
1363 {
1364 int error = 0;
1365 struct file *current_file;
1366
1367 /* let's call this for all tfiles */
1368 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1369 path_count_init();
1370 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1371 reverse_path_check_proc, current_file,
1372 current_file, current);
1373 if (error)
1374 break;
1375 }
1376 return error;
1377 }
1378
1379 static int ep_create_wakeup_source(struct epitem *epi)
1380 {
1381 const char *name;
1382 struct wakeup_source *ws;
1383
1384 if (!epi->ep->ws) {
1385 epi->ep->ws = wakeup_source_register("eventpoll");
1386 if (!epi->ep->ws)
1387 return -ENOMEM;
1388 }
1389
1390 name = epi->ffd.file->f_path.dentry->d_name.name;
1391 ws = wakeup_source_register(name);
1392
1393 if (!ws)
1394 return -ENOMEM;
1395 rcu_assign_pointer(epi->ws, ws);
1396
1397 return 0;
1398 }
1399
1400 /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1401 static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1402 {
1403 struct wakeup_source *ws = ep_wakeup_source(epi);
1404
1405 RCU_INIT_POINTER(epi->ws, NULL);
1406
1407 /*
1408 * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1409 * used internally by wakeup_source_remove, too (called by
1410 * wakeup_source_unregister), so we cannot use call_rcu
1411 */
1412 synchronize_rcu();
1413 wakeup_source_unregister(ws);
1414 }
1415
1416 /*
1417 * Must be called with "mtx" held.
1418 */
1419 static int ep_insert(struct eventpoll *ep, const struct epoll_event *event,
1420 struct file *tfile, int fd, int full_check)
1421 {
1422 int error, pwake = 0;
1423 __poll_t revents;
1424 long user_watches;
1425 struct epitem *epi;
1426 struct ep_pqueue epq;
1427
1428 lockdep_assert_irqs_enabled();
1429
1430 user_watches = atomic_long_read(&ep->user->epoll_watches);
1431 if (unlikely(user_watches >= max_user_watches))
1432 return -ENOSPC;
1433 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
1434 return -ENOMEM;
1435
1436 /* Item initialization follow here ... */
1437 INIT_LIST_HEAD(&epi->rdllink);
1438 INIT_LIST_HEAD(&epi->fllink);
1439 INIT_LIST_HEAD(&epi->pwqlist);
1440 epi->ep = ep;
1441 ep_set_ffd(&epi->ffd, tfile, fd);
1442 epi->event = *event;
1443 epi->nwait = 0;
1444 epi->next = EP_UNACTIVE_PTR;
1445 if (epi->event.events & EPOLLWAKEUP) {
1446 error = ep_create_wakeup_source(epi);
1447 if (error)
1448 goto error_create_wakeup_source;
1449 } else {
1450 RCU_INIT_POINTER(epi->ws, NULL);
1451 }
1452
1453 /* Initialize the poll table using the queue callback */
1454 epq.epi = epi;
1455 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1456
1457 /*
1458 * Attach the item to the poll hooks and get current event bits.
1459 * We can safely use the file* here because its usage count has
1460 * been increased by the caller of this function. Note that after
1461 * this operation completes, the poll callback can start hitting
1462 * the new item.
1463 */
1464 revents = ep_item_poll(epi, &epq.pt, 1);
1465
1466 /*
1467 * We have to check if something went wrong during the poll wait queue
1468 * install process. Namely an allocation for a wait queue failed due
1469 * high memory pressure.
1470 */
1471 error = -ENOMEM;
1472 if (epi->nwait < 0)
1473 goto error_unregister;
1474
1475 /* Add the current item to the list of active epoll hook for this file */
1476 spin_lock(&tfile->f_lock);
1477 list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
1478 spin_unlock(&tfile->f_lock);
1479
1480 /*
1481 * Add the current item to the RB tree. All RB tree operations are
1482 * protected by "mtx", and ep_insert() is called with "mtx" held.
1483 */
1484 ep_rbtree_insert(ep, epi);
1485
1486 /* now check if we've created too many backpaths */
1487 error = -EINVAL;
1488 if (full_check && reverse_path_check())
1489 goto error_remove_epi;
1490
1491 /* We have to drop the new item inside our item list to keep track of it */
1492 spin_lock_irq(&ep->wq.lock);
1493
1494 /* record NAPI ID of new item if present */
1495 ep_set_busy_poll_napi_id(epi);
1496
1497 /* If the file is already "ready" we drop it inside the ready list */
1498 if (revents && !ep_is_linked(epi)) {
1499 list_add_tail(&epi->rdllink, &ep->rdllist);
1500 ep_pm_stay_awake(epi);
1501
1502 /* Notify waiting tasks that events are available */
1503 if (waitqueue_active(&ep->wq))
1504 wake_up_locked(&ep->wq);
1505 if (waitqueue_active(&ep->poll_wait))
1506 pwake++;
1507 }
1508
1509 spin_unlock_irq(&ep->wq.lock);
1510
1511 atomic_long_inc(&ep->user->epoll_watches);
1512
1513 /* We have to call this outside the lock */
1514 if (pwake)
1515 ep_poll_safewake(&ep->poll_wait);
1516
1517 return 0;
1518
1519 error_remove_epi:
1520 spin_lock(&tfile->f_lock);
1521 list_del_rcu(&epi->fllink);
1522 spin_unlock(&tfile->f_lock);
1523
1524 rb_erase_cached(&epi->rbn, &ep->rbr);
1525
1526 error_unregister:
1527 ep_unregister_pollwait(ep, epi);
1528
1529 /*
1530 * We need to do this because an event could have been arrived on some
1531 * allocated wait queue. Note that we don't care about the ep->ovflist
1532 * list, since that is used/cleaned only inside a section bound by "mtx".
1533 * And ep_insert() is called with "mtx" held.
1534 */
1535 spin_lock_irq(&ep->wq.lock);
1536 if (ep_is_linked(epi))
1537 list_del_init(&epi->rdllink);
1538 spin_unlock_irq(&ep->wq.lock);
1539
1540 wakeup_source_unregister(ep_wakeup_source(epi));
1541
1542 error_create_wakeup_source:
1543 kmem_cache_free(epi_cache, epi);
1544
1545 return error;
1546 }
1547
1548 /*
1549 * Modify the interest event mask by dropping an event if the new mask
1550 * has a match in the current file status. Must be called with "mtx" held.
1551 */
1552 static int ep_modify(struct eventpoll *ep, struct epitem *epi,
1553 const struct epoll_event *event)
1554 {
1555 int pwake = 0;
1556 poll_table pt;
1557
1558 lockdep_assert_irqs_enabled();
1559
1560 init_poll_funcptr(&pt, NULL);
1561
1562 /*
1563 * Set the new event interest mask before calling f_op->poll();
1564 * otherwise we might miss an event that happens between the
1565 * f_op->poll() call and the new event set registering.
1566 */
1567 epi->event.events = event->events; /* need barrier below */
1568 epi->event.data = event->data; /* protected by mtx */
1569 if (epi->event.events & EPOLLWAKEUP) {
1570 if (!ep_has_wakeup_source(epi))
1571 ep_create_wakeup_source(epi);
1572 } else if (ep_has_wakeup_source(epi)) {
1573 ep_destroy_wakeup_source(epi);
1574 }
1575
1576 /*
1577 * The following barrier has two effects:
1578 *
1579 * 1) Flush epi changes above to other CPUs. This ensures
1580 * we do not miss events from ep_poll_callback if an
1581 * event occurs immediately after we call f_op->poll().
1582 * We need this because we did not take ep->wq.lock while
1583 * changing epi above (but ep_poll_callback does take
1584 * ep->wq.lock).
1585 *
1586 * 2) We also need to ensure we do not miss _past_ events
1587 * when calling f_op->poll(). This barrier also
1588 * pairs with the barrier in wq_has_sleeper (see
1589 * comments for wq_has_sleeper).
1590 *
1591 * This barrier will now guarantee ep_poll_callback or f_op->poll
1592 * (or both) will notice the readiness of an item.
1593 */
1594 smp_mb();
1595
1596 /*
1597 * Get current event bits. We can safely use the file* here because
1598 * its usage count has been increased by the caller of this function.
1599 * If the item is "hot" and it is not registered inside the ready
1600 * list, push it inside.
1601 */
1602 if (ep_item_poll(epi, &pt, 1)) {
1603 spin_lock_irq(&ep->wq.lock);
1604 if (!ep_is_linked(epi)) {
1605 list_add_tail(&epi->rdllink, &ep->rdllist);
1606 ep_pm_stay_awake(epi);
1607
1608 /* Notify waiting tasks that events are available */
1609 if (waitqueue_active(&ep->wq))
1610 wake_up_locked(&ep->wq);
1611 if (waitqueue_active(&ep->poll_wait))
1612 pwake++;
1613 }
1614 spin_unlock_irq(&ep->wq.lock);
1615 }
1616
1617 /* We have to call this outside the lock */
1618 if (pwake)
1619 ep_poll_safewake(&ep->poll_wait);
1620
1621 return 0;
1622 }
1623
1624 static __poll_t ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1625 void *priv)
1626 {
1627 struct ep_send_events_data *esed = priv;
1628 __poll_t revents;
1629 struct epitem *epi;
1630 struct epoll_event __user *uevent;
1631 struct wakeup_source *ws;
1632 poll_table pt;
1633
1634 init_poll_funcptr(&pt, NULL);
1635
1636 /*
1637 * We can loop without lock because we are passed a task private list.
1638 * Items cannot vanish during the loop because ep_scan_ready_list() is
1639 * holding "mtx" during this call.
1640 */
1641 for (esed->res = 0, uevent = esed->events;
1642 !list_empty(head) && esed->res < esed->maxevents;) {
1643 epi = list_first_entry(head, struct epitem, rdllink);
1644
1645 /*
1646 * Activate ep->ws before deactivating epi->ws to prevent
1647 * triggering auto-suspend here (in case we reactive epi->ws
1648 * below).
1649 *
1650 * This could be rearranged to delay the deactivation of epi->ws
1651 * instead, but then epi->ws would temporarily be out of sync
1652 * with ep_is_linked().
1653 */
1654 ws = ep_wakeup_source(epi);
1655 if (ws) {
1656 if (ws->active)
1657 __pm_stay_awake(ep->ws);
1658 __pm_relax(ws);
1659 }
1660
1661 list_del_init(&epi->rdllink);
1662
1663 revents = ep_item_poll(epi, &pt, 1);
1664
1665 /*
1666 * If the event mask intersect the caller-requested one,
1667 * deliver the event to userspace. Again, ep_scan_ready_list()
1668 * is holding "mtx", so no operations coming from userspace
1669 * can change the item.
1670 */
1671 if (revents) {
1672 if (__put_user(revents, &uevent->events) ||
1673 __put_user(epi->event.data, &uevent->data)) {
1674 list_add(&epi->rdllink, head);
1675 ep_pm_stay_awake(epi);
1676 if (!esed->res)
1677 esed->res = -EFAULT;
1678 return 0;
1679 }
1680 esed->res++;
1681 uevent++;
1682 if (epi->event.events & EPOLLONESHOT)
1683 epi->event.events &= EP_PRIVATE_BITS;
1684 else if (!(epi->event.events & EPOLLET)) {
1685 /*
1686 * If this file has been added with Level
1687 * Trigger mode, we need to insert back inside
1688 * the ready list, so that the next call to
1689 * epoll_wait() will check again the events
1690 * availability. At this point, no one can insert
1691 * into ep->rdllist besides us. The epoll_ctl()
1692 * callers are locked out by
1693 * ep_scan_ready_list() holding "mtx" and the
1694 * poll callback will queue them in ep->ovflist.
1695 */
1696 list_add_tail(&epi->rdllink, &ep->rdllist);
1697 ep_pm_stay_awake(epi);
1698 }
1699 }
1700 }
1701
1702 return 0;
1703 }
1704
1705 static int ep_send_events(struct eventpoll *ep,
1706 struct epoll_event __user *events, int maxevents)
1707 {
1708 struct ep_send_events_data esed;
1709
1710 esed.maxevents = maxevents;
1711 esed.events = events;
1712
1713 ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
1714 return esed.res;
1715 }
1716
1717 static inline struct timespec64 ep_set_mstimeout(long ms)
1718 {
1719 struct timespec64 now, ts = {
1720 .tv_sec = ms / MSEC_PER_SEC,
1721 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1722 };
1723
1724 ktime_get_ts64(&now);
1725 return timespec64_add_safe(now, ts);
1726 }
1727
1728 /**
1729 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1730 * event buffer.
1731 *
1732 * @ep: Pointer to the eventpoll context.
1733 * @events: Pointer to the userspace buffer where the ready events should be
1734 * stored.
1735 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1736 * @timeout: Maximum timeout for the ready events fetch operation, in
1737 * milliseconds. If the @timeout is zero, the function will not block,
1738 * while if the @timeout is less than zero, the function will block
1739 * until at least one event has been retrieved (or an error
1740 * occurred).
1741 *
1742 * Returns: Returns the number of ready events which have been fetched, or an
1743 * error code, in case of error.
1744 */
1745 static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1746 int maxevents, long timeout)
1747 {
1748 int res = 0, eavail, timed_out = 0;
1749 u64 slack = 0;
1750 wait_queue_entry_t wait;
1751 ktime_t expires, *to = NULL;
1752
1753 lockdep_assert_irqs_enabled();
1754
1755 if (timeout > 0) {
1756 struct timespec64 end_time = ep_set_mstimeout(timeout);
1757
1758 slack = select_estimate_accuracy(&end_time);
1759 to = &expires;
1760 *to = timespec64_to_ktime(end_time);
1761 } else if (timeout == 0) {
1762 /*
1763 * Avoid the unnecessary trip to the wait queue loop, if the
1764 * caller specified a non blocking operation.
1765 */
1766 timed_out = 1;
1767 spin_lock_irq(&ep->wq.lock);
1768 goto check_events;
1769 }
1770
1771 fetch_events:
1772
1773 if (!ep_events_available(ep))
1774 ep_busy_loop(ep, timed_out);
1775
1776 spin_lock_irq(&ep->wq.lock);
1777
1778 if (!ep_events_available(ep)) {
1779 /*
1780 * Busy poll timed out. Drop NAPI ID for now, we can add
1781 * it back in when we have moved a socket with a valid NAPI
1782 * ID onto the ready list.
1783 */
1784 ep_reset_busy_poll_napi_id(ep);
1785
1786 /*
1787 * We don't have any available event to return to the caller.
1788 * We need to sleep here, and we will be wake up by
1789 * ep_poll_callback() when events will become available.
1790 */
1791 init_waitqueue_entry(&wait, current);
1792 __add_wait_queue_exclusive(&ep->wq, &wait);
1793
1794 for (;;) {
1795 /*
1796 * We don't want to sleep if the ep_poll_callback() sends us
1797 * a wakeup in between. That's why we set the task state
1798 * to TASK_INTERRUPTIBLE before doing the checks.
1799 */
1800 set_current_state(TASK_INTERRUPTIBLE);
1801 /*
1802 * Always short-circuit for fatal signals to allow
1803 * threads to make a timely exit without the chance of
1804 * finding more events available and fetching
1805 * repeatedly.
1806 */
1807 if (fatal_signal_pending(current)) {
1808 res = -EINTR;
1809 break;
1810 }
1811 if (ep_events_available(ep) || timed_out)
1812 break;
1813 if (signal_pending(current)) {
1814 res = -EINTR;
1815 break;
1816 }
1817
1818 spin_unlock_irq(&ep->wq.lock);
1819 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1820 timed_out = 1;
1821
1822 spin_lock_irq(&ep->wq.lock);
1823 }
1824
1825 __remove_wait_queue(&ep->wq, &wait);
1826 __set_current_state(TASK_RUNNING);
1827 }
1828 check_events:
1829 /* Is it worth to try to dig for events ? */
1830 eavail = ep_events_available(ep);
1831
1832 spin_unlock_irq(&ep->wq.lock);
1833
1834 /*
1835 * Try to transfer events to user space. In case we get 0 events and
1836 * there's still timeout left over, we go trying again in search of
1837 * more luck.
1838 */
1839 if (!res && eavail &&
1840 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
1841 goto fetch_events;
1842
1843 return res;
1844 }
1845
1846 /**
1847 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1848 * API, to verify that adding an epoll file inside another
1849 * epoll structure, does not violate the constraints, in
1850 * terms of closed loops, or too deep chains (which can
1851 * result in excessive stack usage).
1852 *
1853 * @priv: Pointer to the epoll file to be currently checked.
1854 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1855 * data structure pointer.
1856 * @call_nests: Current dept of the @ep_call_nested() call stack.
1857 *
1858 * Returns: Returns zero if adding the epoll @file inside current epoll
1859 * structure @ep does not violate the constraints, or -1 otherwise.
1860 */
1861 static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1862 {
1863 int error = 0;
1864 struct file *file = priv;
1865 struct eventpoll *ep = file->private_data;
1866 struct eventpoll *ep_tovisit;
1867 struct rb_node *rbp;
1868 struct epitem *epi;
1869
1870 mutex_lock_nested(&ep->mtx, call_nests + 1);
1871 ep->visited = 1;
1872 list_add(&ep->visited_list_link, &visited_list);
1873 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1874 epi = rb_entry(rbp, struct epitem, rbn);
1875 if (unlikely(is_file_epoll(epi->ffd.file))) {
1876 ep_tovisit = epi->ffd.file->private_data;
1877 if (ep_tovisit->visited)
1878 continue;
1879 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1880 ep_loop_check_proc, epi->ffd.file,
1881 ep_tovisit, current);
1882 if (error != 0)
1883 break;
1884 } else {
1885 /*
1886 * If we've reached a file that is not associated with
1887 * an ep, then we need to check if the newly added
1888 * links are going to add too many wakeup paths. We do
1889 * this by adding it to the tfile_check_list, if it's
1890 * not already there, and calling reverse_path_check()
1891 * during ep_insert().
1892 */
1893 if (list_empty(&epi->ffd.file->f_tfile_llink))
1894 list_add(&epi->ffd.file->f_tfile_llink,
1895 &tfile_check_list);
1896 }
1897 }
1898 mutex_unlock(&ep->mtx);
1899
1900 return error;
1901 }
1902
1903 /**
1904 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1905 * another epoll file (represented by @ep) does not create
1906 * closed loops or too deep chains.
1907 *
1908 * @ep: Pointer to the epoll private data structure.
1909 * @file: Pointer to the epoll file to be checked.
1910 *
1911 * Returns: Returns zero if adding the epoll @file inside current epoll
1912 * structure @ep does not violate the constraints, or -1 otherwise.
1913 */
1914 static int ep_loop_check(struct eventpoll *ep, struct file *file)
1915 {
1916 int ret;
1917 struct eventpoll *ep_cur, *ep_next;
1918
1919 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1920 ep_loop_check_proc, file, ep, current);
1921 /* clear visited list */
1922 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1923 visited_list_link) {
1924 ep_cur->visited = 0;
1925 list_del(&ep_cur->visited_list_link);
1926 }
1927 return ret;
1928 }
1929
1930 static void clear_tfile_check_list(void)
1931 {
1932 struct file *file;
1933
1934 /* first clear the tfile_check_list */
1935 while (!list_empty(&tfile_check_list)) {
1936 file = list_first_entry(&tfile_check_list, struct file,
1937 f_tfile_llink);
1938 list_del_init(&file->f_tfile_llink);
1939 }
1940 INIT_LIST_HEAD(&tfile_check_list);
1941 }
1942
1943 /*
1944 * Open an eventpoll file descriptor.
1945 */
1946 static int do_epoll_create(int flags)
1947 {
1948 int error, fd;
1949 struct eventpoll *ep = NULL;
1950 struct file *file;
1951
1952 /* Check the EPOLL_* constant for consistency. */
1953 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1954
1955 if (flags & ~EPOLL_CLOEXEC)
1956 return -EINVAL;
1957 /*
1958 * Create the internal data structure ("struct eventpoll").
1959 */
1960 error = ep_alloc(&ep);
1961 if (error < 0)
1962 return error;
1963 /*
1964 * Creates all the items needed to setup an eventpoll file. That is,
1965 * a file structure and a free file descriptor.
1966 */
1967 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1968 if (fd < 0) {
1969 error = fd;
1970 goto out_free_ep;
1971 }
1972 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
1973 O_RDWR | (flags & O_CLOEXEC));
1974 if (IS_ERR(file)) {
1975 error = PTR_ERR(file);
1976 goto out_free_fd;
1977 }
1978 ep->file = file;
1979 fd_install(fd, file);
1980 return fd;
1981
1982 out_free_fd:
1983 put_unused_fd(fd);
1984 out_free_ep:
1985 ep_free(ep);
1986 return error;
1987 }
1988
1989 SYSCALL_DEFINE1(epoll_create1, int, flags)
1990 {
1991 return do_epoll_create(flags);
1992 }
1993
1994 SYSCALL_DEFINE1(epoll_create, int, size)
1995 {
1996 if (size <= 0)
1997 return -EINVAL;
1998
1999 return do_epoll_create(0);
2000 }
2001
2002 /*
2003 * The following function implements the controller interface for
2004 * the eventpoll file that enables the insertion/removal/change of
2005 * file descriptors inside the interest set.
2006 */
2007 SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
2008 struct epoll_event __user *, event)
2009 {
2010 int error;
2011 int full_check = 0;
2012 struct fd f, tf;
2013 struct eventpoll *ep;
2014 struct epitem *epi;
2015 struct epoll_event epds;
2016 struct eventpoll *tep = NULL;
2017
2018 error = -EFAULT;
2019 if (ep_op_has_event(op) &&
2020 copy_from_user(&epds, event, sizeof(struct epoll_event)))
2021 goto error_return;
2022
2023 error = -EBADF;
2024 f = fdget(epfd);
2025 if (!f.file)
2026 goto error_return;
2027
2028 /* Get the "struct file *" for the target file */
2029 tf = fdget(fd);
2030 if (!tf.file)
2031 goto error_fput;
2032
2033 /* The target file descriptor must support poll */
2034 error = -EPERM;
2035 if (!file_can_poll(tf.file))
2036 goto error_tgt_fput;
2037
2038 /* Check if EPOLLWAKEUP is allowed */
2039 if (ep_op_has_event(op))
2040 ep_take_care_of_epollwakeup(&epds);
2041
2042 /*
2043 * We have to check that the file structure underneath the file descriptor
2044 * the user passed to us _is_ an eventpoll file. And also we do not permit
2045 * adding an epoll file descriptor inside itself.
2046 */
2047 error = -EINVAL;
2048 if (f.file == tf.file || !is_file_epoll(f.file))
2049 goto error_tgt_fput;
2050
2051 /*
2052 * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only,
2053 * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation.
2054 * Also, we do not currently supported nested exclusive wakeups.
2055 */
2056 if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) {
2057 if (op == EPOLL_CTL_MOD)
2058 goto error_tgt_fput;
2059 if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) ||
2060 (epds.events & ~EPOLLEXCLUSIVE_OK_BITS)))
2061 goto error_tgt_fput;
2062 }
2063
2064 /*
2065 * At this point it is safe to assume that the "private_data" contains
2066 * our own data structure.
2067 */
2068 ep = f.file->private_data;
2069
2070 /*
2071 * When we insert an epoll file descriptor, inside another epoll file
2072 * descriptor, there is the change of creating closed loops, which are
2073 * better be handled here, than in more critical paths. While we are
2074 * checking for loops we also determine the list of files reachable
2075 * and hang them on the tfile_check_list, so we can check that we
2076 * haven't created too many possible wakeup paths.
2077 *
2078 * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
2079 * the epoll file descriptor is attaching directly to a wakeup source,
2080 * unless the epoll file descriptor is nested. The purpose of taking the
2081 * 'epmutex' on add is to prevent complex toplogies such as loops and
2082 * deep wakeup paths from forming in parallel through multiple
2083 * EPOLL_CTL_ADD operations.
2084 */
2085 mutex_lock_nested(&ep->mtx, 0);
2086 if (op == EPOLL_CTL_ADD) {
2087 if (!list_empty(&f.file->f_ep_links) ||
2088 is_file_epoll(tf.file)) {
2089 full_check = 1;
2090 mutex_unlock(&ep->mtx);
2091 mutex_lock(&epmutex);
2092 if (is_file_epoll(tf.file)) {
2093 error = -ELOOP;
2094 if (ep_loop_check(ep, tf.file) != 0) {
2095 clear_tfile_check_list();
2096 goto error_tgt_fput;
2097 }
2098 } else
2099 list_add(&tf.file->f_tfile_llink,
2100 &tfile_check_list);
2101 mutex_lock_nested(&ep->mtx, 0);
2102 if (is_file_epoll(tf.file)) {
2103 tep = tf.file->private_data;
2104 mutex_lock_nested(&tep->mtx, 1);
2105 }
2106 }
2107 }
2108
2109 /*
2110 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
2111 * above, we can be sure to be able to use the item looked up by
2112 * ep_find() till we release the mutex.
2113 */
2114 epi = ep_find(ep, tf.file, fd);
2115
2116 error = -EINVAL;
2117 switch (op) {
2118 case EPOLL_CTL_ADD:
2119 if (!epi) {
2120 epds.events |= EPOLLERR | EPOLLHUP;
2121 error = ep_insert(ep, &epds, tf.file, fd, full_check);
2122 } else
2123 error = -EEXIST;
2124 if (full_check)
2125 clear_tfile_check_list();
2126 break;
2127 case EPOLL_CTL_DEL:
2128 if (epi)
2129 error = ep_remove(ep, epi);
2130 else
2131 error = -ENOENT;
2132 break;
2133 case EPOLL_CTL_MOD:
2134 if (epi) {
2135 if (!(epi->event.events & EPOLLEXCLUSIVE)) {
2136 epds.events |= EPOLLERR | EPOLLHUP;
2137 error = ep_modify(ep, epi, &epds);
2138 }
2139 } else
2140 error = -ENOENT;
2141 break;
2142 }
2143 if (tep != NULL)
2144 mutex_unlock(&tep->mtx);
2145 mutex_unlock(&ep->mtx);
2146
2147 error_tgt_fput:
2148 if (full_check)
2149 mutex_unlock(&epmutex);
2150
2151 fdput(tf);
2152 error_fput:
2153 fdput(f);
2154 error_return:
2155
2156 return error;
2157 }
2158
2159 /*
2160 * Implement the event wait interface for the eventpoll file. It is the kernel
2161 * part of the user space epoll_wait(2).
2162 */
2163 static int do_epoll_wait(int epfd, struct epoll_event __user *events,
2164 int maxevents, int timeout)
2165 {
2166 int error;
2167 struct fd f;
2168 struct eventpoll *ep;
2169
2170 /* The maximum number of event must be greater than zero */
2171 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
2172 return -EINVAL;
2173
2174 /* Verify that the area passed by the user is writeable */
2175 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
2176 return -EFAULT;
2177
2178 /* Get the "struct file *" for the eventpoll file */
2179 f = fdget(epfd);
2180 if (!f.file)
2181 return -EBADF;
2182
2183 /*
2184 * We have to check that the file structure underneath the fd
2185 * the user passed to us _is_ an eventpoll file.
2186 */
2187 error = -EINVAL;
2188 if (!is_file_epoll(f.file))
2189 goto error_fput;
2190
2191 /*
2192 * At this point it is safe to assume that the "private_data" contains
2193 * our own data structure.
2194 */
2195 ep = f.file->private_data;
2196
2197 /* Time to fish for events ... */
2198 error = ep_poll(ep, events, maxevents, timeout);
2199
2200 error_fput:
2201 fdput(f);
2202 return error;
2203 }
2204
2205 SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
2206 int, maxevents, int, timeout)
2207 {
2208 return do_epoll_wait(epfd, events, maxevents, timeout);
2209 }
2210
2211 /*
2212 * Implement the event wait interface for the eventpoll file. It is the kernel
2213 * part of the user space epoll_pwait(2).
2214 */
2215 SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2216 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2217 size_t, sigsetsize)
2218 {
2219 int error;
2220 sigset_t ksigmask, sigsaved;
2221
2222 /*
2223 * If the caller wants a certain signal mask to be set during the wait,
2224 * we apply it here.
2225 */
2226 if (sigmask) {
2227 if (sigsetsize != sizeof(sigset_t))
2228 return -EINVAL;
2229 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
2230 return -EFAULT;
2231 sigsaved = current->blocked;
2232 set_current_blocked(&ksigmask);
2233 }
2234
2235 error = do_epoll_wait(epfd, events, maxevents, timeout);
2236
2237 /*
2238 * If we changed the signal mask, we need to restore the original one.
2239 * In case we've got a signal while waiting, we do not restore the
2240 * signal mask yet, and we allow do_signal() to deliver the signal on
2241 * the way back to userspace, before the signal mask is restored.
2242 */
2243 if (sigmask) {
2244 if (error == -EINTR) {
2245 memcpy(&current->saved_sigmask, &sigsaved,
2246 sizeof(sigsaved));
2247 set_restore_sigmask();
2248 } else
2249 set_current_blocked(&sigsaved);
2250 }
2251
2252 return error;
2253 }
2254
2255 #ifdef CONFIG_COMPAT
2256 COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2257 struct epoll_event __user *, events,
2258 int, maxevents, int, timeout,
2259 const compat_sigset_t __user *, sigmask,
2260 compat_size_t, sigsetsize)
2261 {
2262 long err;
2263 sigset_t ksigmask, sigsaved;
2264
2265 /*
2266 * If the caller wants a certain signal mask to be set during the wait,
2267 * we apply it here.
2268 */
2269 if (sigmask) {
2270 if (sigsetsize != sizeof(compat_sigset_t))
2271 return -EINVAL;
2272 if (get_compat_sigset(&ksigmask, sigmask))
2273 return -EFAULT;
2274 sigsaved = current->blocked;
2275 set_current_blocked(&ksigmask);
2276 }
2277
2278 err = do_epoll_wait(epfd, events, maxevents, timeout);
2279
2280 /*
2281 * If we changed the signal mask, we need to restore the original one.
2282 * In case we've got a signal while waiting, we do not restore the
2283 * signal mask yet, and we allow do_signal() to deliver the signal on
2284 * the way back to userspace, before the signal mask is restored.
2285 */
2286 if (sigmask) {
2287 if (err == -EINTR) {
2288 memcpy(&current->saved_sigmask, &sigsaved,
2289 sizeof(sigsaved));
2290 set_restore_sigmask();
2291 } else
2292 set_current_blocked(&sigsaved);
2293 }
2294
2295 return err;
2296 }
2297 #endif
2298
2299 static int __init eventpoll_init(void)
2300 {
2301 struct sysinfo si;
2302
2303 si_meminfo(&si);
2304 /*
2305 * Allows top 4% of lomem to be allocated for epoll watches (per user).
2306 */
2307 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2308 EP_ITEM_COST;
2309 BUG_ON(max_user_watches < 0);
2310
2311 /*
2312 * Initialize the structure used to perform epoll file descriptor
2313 * inclusion loops checks.
2314 */
2315 ep_nested_calls_init(&poll_loop_ncalls);
2316
2317 #ifdef CONFIG_DEBUG_LOCK_ALLOC
2318 /* Initialize the structure used to perform safe poll wait head wake ups */
2319 ep_nested_calls_init(&poll_safewake_ncalls);
2320 #endif
2321
2322 /*
2323 * We can have many thousands of epitems, so prevent this from
2324 * using an extra cache line on 64-bit (and smaller) CPUs
2325 */
2326 BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2327
2328 /* Allocates slab cache used to allocate "struct epitem" items */
2329 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
2330 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
2331
2332 /* Allocates slab cache used to allocate "struct eppoll_entry" */
2333 pwq_cache = kmem_cache_create("eventpoll_pwq",
2334 sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL);
2335
2336 return 0;
2337 }
2338 fs_initcall(eventpoll_init);