1	// SPDX-License-Identifier: GPL-2.0-or-later
2	/*
3	* fs/eventpoll.c (Efficient event retrieval implementation)
4	* Copyright (C) 2001,...,2009 Davide Libenzi
5	*
6	* Davide Libenzi <davidel@xmailserver.org>
7	*/
8
9	#include <linux/init.h>
10	#include <linux/kernel.h>
11	#include <linux/sched/signal.h>
12	#include <linux/fs.h>
13	#include <linux/file.h>
14	#include <linux/signal.h>
15	#include <linux/errno.h>
16	#include <linux/mm.h>
17	#include <linux/slab.h>
18	#include <linux/poll.h>
19	#include <linux/string.h>
20	#include <linux/list.h>
21	#include <linux/hash.h>
22	#include <linux/spinlock.h>
23	#include <linux/syscalls.h>
24	#include <linux/rbtree.h>
25	#include <linux/wait.h>
26	#include <linux/eventpoll.h>
27	#include <linux/mount.h>
28	#include <linux/bitops.h>
29	#include <linux/mutex.h>
30	#include <linux/anon_inodes.h>
31	#include <linux/device.h>
32	#include <linux/uaccess.h>
33	#include <asm/io.h>
34	#include <asm/mman.h>
35	#include <linux/atomic.h>
36	#include <linux/proc_fs.h>
37	#include <linux/seq_file.h>
38	#include <linux/compat.h>
39	#include <linux/rculist.h>
40	#include <net/busy_poll.h>
41
42	/*
43	* LOCKING:
44	* There are three level of locking required by epoll :
45	*
46	* 1) epmutex (mutex)
47	* 2) ep->mtx (mutex)
48	* 3) ep->lock (rwlock)
49	*
50	* The acquire order is the one listed above, from 1 to 3.
51	* We need a rwlock (ep->lock) because we manipulate objects
52	* from inside the poll callback, that might be triggered from
53	* a wake_up() that in turn might be called from IRQ context.
54	* So we can't sleep inside the poll callback and hence we need
55	* a spinlock. During the event transfer loop (from kernel to
56	* user space) we could end up sleeping due a copy_to_user(), so
57	* we need a lock that will allow us to sleep. This lock is a
58	* mutex (ep->mtx). It is acquired during the event transfer loop,
59	* during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
60	* Then we also need a global mutex to serialize eventpoll_release_file()
61	* and ep_free().
62	* This mutex is acquired by ep_free() during the epoll file
63	* cleanup path and it is also acquired by eventpoll_release_file()
64	* if a file has been pushed inside an epoll set and it is then
65	* close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
66	* It is also acquired when inserting an epoll fd onto another epoll
67	* fd. We do this so that we walk the epoll tree and ensure that this
68	* insertion does not create a cycle of epoll file descriptors, which
69	* could lead to deadlock. We need a global mutex to prevent two
70	* simultaneous inserts (A into B and B into A) from racing and
71	* constructing a cycle without either insert observing that it is
72	* going to.
73	* It is necessary to acquire multiple "ep->mtx"es at once in the
74	* case when one epoll fd is added to another. In this case, we
75	* always acquire the locks in the order of nesting (i.e. after
76	* epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
77	* before e2->mtx). Since we disallow cycles of epoll file
78	* descriptors, this ensures that the mutexes are well-ordered. In
79	* order to communicate this nesting to lockdep, when walking a tree
80	* of epoll file descriptors, we use the current recursion depth as
81	* the lockdep subkey.
82	* It is possible to drop the "ep->mtx" and to use the global
83	* mutex "epmutex" (together with "ep->lock") to have it working,
84	* but having "ep->mtx" will make the interface more scalable.
85	* Events that require holding "epmutex" are very rare, while for
86	* normal operations the epoll private "ep->mtx" will guarantee
87	* a better scalability.
88	*/
89
90	/* Epoll private bits inside the event mask */
91	#define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE)
92
93	#define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT)
94
95	#define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \
96	EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE)
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	/* Wait structure used by the poll hooks */
113	struct eppoll_entry {
114	/* List header used to link this structure to the "struct epitem" */
115	struct eppoll_entry *next;
116
117	/* The "base" pointer is set to the container "struct epitem" */
118	struct epitem *base;
119
120	/*
121	* Wait queue item that will be linked to the target file wait
122	* queue head.
123	*/
124	wait_queue_entry_t wait;
125
126	/* The wait queue head that linked the "wait" wait queue item */
127	wait_queue_head_t *whead;
128	};
129
130	/*
131	* Each file descriptor added to the eventpoll interface will
132	* have an entry of this type linked to the "rbr" RB tree.
133	* Avoid increasing the size of this struct, there can be many thousands
134	* of these on a server and we do not want this to take another cache line.
135	*/
136	struct epitem {
137	union {
138	/* RB tree node links this structure to the eventpoll RB tree */
139	struct rb_node rbn;
140	/* Used to free the struct epitem */
141	struct rcu_head rcu;
142	};
143
144	/* List header used to link this structure to the eventpoll ready list */
145	struct list_head rdllink;
146
147	/*
148	* Works together "struct eventpoll"->ovflist in keeping the
149	* single linked chain of items.
150	*/
151	struct epitem *next;
152
153	/* The file descriptor information this item refers to */
154	struct epoll_filefd ffd;
155
156	/* List containing poll wait queues */
157	struct eppoll_entry *pwqlist;
158
159	/* The "container" of this item */
160	struct eventpoll *ep;
161
162	/* List header used to link this item to the "struct file" items list */
163	struct hlist_node fllink;
164
165	/* wakeup_source used when EPOLLWAKEUP is set */
166	struct wakeup_source __rcu *ws;
167
168	/* The structure that describe the interested events and the source fd */
169	struct epoll_event event;
170	};
171
172	/*
173	* This structure is stored inside the "private_data" member of the file
174	* structure and represents the main data structure for the eventpoll
175	* interface.
176	*/
177	struct eventpoll {
178	/*
179	* This mutex is used to ensure that files are not removed
180	* while epoll is using them. This is held during the event
181	* collection loop, the file cleanup path, the epoll file exit
182	* code and the ctl operations.
183	*/
184	struct mutex mtx;
185
186	/* Wait queue used by sys_epoll_wait() */
187	wait_queue_head_t wq;
188
189	/* Wait queue used by file->poll() */
190	wait_queue_head_t poll_wait;
191
192	/* List of ready file descriptors */
193	struct list_head rdllist;
194
195	/* Lock which protects rdllist and ovflist */
196	rwlock_t lock;
197
198	/* RB tree root used to store monitored fd structs */
199	struct rb_root_cached rbr;
200
201	/*
202	* This is a single linked list that chains all the "struct epitem" that
203	* happened while transferring ready events to userspace w/out
204	* holding ->lock.
205	*/
206	struct epitem *ovflist;
207
208	/* wakeup_source used when ep_scan_ready_list is running */
209	struct wakeup_source *ws;
210
211	/* The user that created the eventpoll descriptor */
212	struct user_struct *user;
213
214	struct file *file;
215
216	/* used to optimize loop detection check */
217	u64 gen;
218	struct hlist_head refs;
219
220	#ifdef CONFIG_NET_RX_BUSY_POLL
221	/* used to track busy poll napi_id */
222	unsigned int napi_id;
223	#endif
224
225	#ifdef CONFIG_DEBUG_LOCK_ALLOC
226	/* tracks wakeup nests for lockdep validation */
227	u8 nests;
228	#endif
229	};
230
231	/* Wrapper struct used by poll queueing */
232	struct ep_pqueue {
233	poll_table pt;
234	struct epitem *epi;
235	};
236
237	/*
238	* Configuration options available inside /proc/sys/fs/epoll/
239	*/
240	/* Maximum number of epoll watched descriptors, per user */
241	static long max_user_watches __read_mostly;
242
243	/*
244	* This mutex is used to serialize ep_free() and eventpoll_release_file().
245	*/
246	static DEFINE_MUTEX(epmutex);
247
248	static u64 loop_check_gen = 0;
249
250	/* Used to check for epoll file descriptor inclusion loops */
251	static struct eventpoll *inserting_into;
252
253	/* Slab cache used to allocate "struct epitem" */
254	static struct kmem_cache *epi_cache __read_mostly;
255
256	/* Slab cache used to allocate "struct eppoll_entry" */
257	static struct kmem_cache *pwq_cache __read_mostly;
258
259	/*
260	* List of files with newly added links, where we may need to limit the number
261	* of emanating paths. Protected by the epmutex.
262	*/
263	struct epitems_head {
264	struct hlist_head epitems;
265	struct epitems_head *next;
266	};
267	static struct epitems_head *tfile_check_list = EP_UNACTIVE_PTR;
268
269	static struct kmem_cache *ephead_cache __read_mostly;
270
271	static inline void free_ephead(struct epitems_head *head)
272	{
273	if (head)
274	kmem_cache_free(ephead_cache, head);
275	}
276
277	static void list_file(struct file *file)
278	{
279	struct epitems_head *head;
280
281	head = container_of(file->f_ep, struct epitems_head, epitems);
282	if (!head->next) {
283	head->next = tfile_check_list;
284	tfile_check_list = head;
285	}
286	}
287
288	static void unlist_file(struct epitems_head *head)
289	{
290	struct epitems_head *to_free = head;
291	struct hlist_node *p = rcu_dereference(hlist_first_rcu(&head->epitems));
292	if (p) {
293	struct epitem *epi= container_of(p, struct epitem, fllink);
294	spin_lock(&epi->ffd.file->f_lock);
295	if (!hlist_empty(&head->epitems))
296	to_free = NULL;
297	head->next = NULL;
298	spin_unlock(&epi->ffd.file->f_lock);
299	}
300	free_ephead(to_free);
301	}
302
303	#ifdef CONFIG_SYSCTL
304
305	#include <linux/sysctl.h>
306
307	static long long_zero;
308	static long long_max = LONG_MAX;
309
310	static struct ctl_table epoll_table[] = {
311	{
312	.procname = "max_user_watches",
313	.data = &max_user_watches,
314	.maxlen = sizeof(max_user_watches),
315	.mode = 0644,
316	.proc_handler = proc_doulongvec_minmax,
317	.extra1 = &long_zero,
318	.extra2 = &long_max,
319	},
320	{ }
321	};
322
323	static void __init epoll_sysctls_init(void)
324	{
325	register_sysctl("fs/epoll", epoll_table);
326	}
327	#else
328	#define epoll_sysctls_init() do { } while (0)
329	#endif /* CONFIG_SYSCTL */
330
331	static const struct file_operations eventpoll_fops;
332
333	static inline int is_file_epoll(struct file *f)
334	{
335	return f->f_op == &eventpoll_fops;
336	}
337
338	/* Setup the structure that is used as key for the RB tree */
339	static inline void ep_set_ffd(struct epoll_filefd *ffd,
340	struct file *file, int fd)
341	{
342	ffd->file = file;
343	ffd->fd = fd;
344	}
345
346	/* Compare RB tree keys */
347	static inline int ep_cmp_ffd(struct epoll_filefd *p1,
348	struct epoll_filefd *p2)
349	{
350	return (p1->file > p2->file ? +1:
351	(p1->file < p2->file ? -1 : p1->fd - p2->fd));
352	}
353
354	/* Tells us if the item is currently linked */
355	static inline int ep_is_linked(struct epitem *epi)
356	{
357	return !list_empty(&epi->rdllink);
358	}
359
360	static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p)
361	{
362	return container_of(p, struct eppoll_entry, wait);
363	}
364
365	/* Get the "struct epitem" from a wait queue pointer */
366	static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p)
367	{
368	return container_of(p, struct eppoll_entry, wait)->base;
369	}
370
371	/**
372	* ep_events_available - Checks if ready events might be available.
373	*
374	* @ep: Pointer to the eventpoll context.
375	*
376	* Return: a value different than %zero if ready events are available,
377	* or %zero otherwise.
378	*/
379	static inline int ep_events_available(struct eventpoll *ep)
380	{
381	return !list_empty_careful(&ep->rdllist) ||
382	READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR;
383	}
384
385	#ifdef CONFIG_NET_RX_BUSY_POLL
386	static bool ep_busy_loop_end(void *p, unsigned long start_time)
387	{
388	struct eventpoll *ep = p;
389
390	return ep_events_available(ep) || busy_loop_timeout(start_time);
391	}
392
393	/*
394	* Busy poll if globally on and supporting sockets found && no events,
395	* busy loop will return if need_resched or ep_events_available.
396	*
397	* we must do our busy polling with irqs enabled
398	*/
399	static bool ep_busy_loop(struct eventpoll *ep, int nonblock)
400	{
401	unsigned int napi_id = READ_ONCE(ep->napi_id);
402
403	if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) {
404	napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep, false,
405	BUSY_POLL_BUDGET);
406	if (ep_events_available(ep))
407	return true;
408	/*
409	* Busy poll timed out. Drop NAPI ID for now, we can add
410	* it back in when we have moved a socket with a valid NAPI
411	* ID onto the ready list.
412	*/
413	ep->napi_id = 0;
414	return false;
415	}
416	return false;
417	}
418
419	/*
420	* Set epoll busy poll NAPI ID from sk.
421	*/
422	static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
423	{
424	struct eventpoll *ep;
425	unsigned int napi_id;
426	struct socket *sock;
427	struct sock *sk;
428
429	if (!net_busy_loop_on())
430	return;
431
432	sock = sock_from_file(epi->ffd.file);
433	if (!sock)
434	return;
435
436	sk = sock->sk;
437	if (!sk)
438	return;
439
440	napi_id = READ_ONCE(sk->sk_napi_id);
441	ep = epi->ep;
442
443	/* Non-NAPI IDs can be rejected
444	* or
445	* Nothing to do if we already have this ID
446	*/
447	if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id)
448	return;
449
450	/* record NAPI ID for use in next busy poll */
451	ep->napi_id = napi_id;
452	}
453
454	#else
455
456	static inline bool ep_busy_loop(struct eventpoll *ep, int nonblock)
457	{
458	return false;
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	* As described in commit 0ccf831cb lockdep: annotate epoll
469	* the use of wait queues used by epoll is done in a very controlled
470	* manner. Wake ups can nest inside each other, but are never done
471	* with the same locking. For example:
472	*
473	* dfd = socket(...);
474	* efd1 = epoll_create();
475	* efd2 = epoll_create();
476	* epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
477	* epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
478	*
479	* When a packet arrives to the device underneath "dfd", the net code will
480	* issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
481	* callback wakeup entry on that queue, and the wake_up() performed by the
482	* "dfd" net code will end up in ep_poll_callback(). At this point epoll
483	* (efd1) notices that it may have some event ready, so it needs to wake up
484	* the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
485	* that ends up in another wake_up(), after having checked about the
486	* recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
487	* avoid stack blasting.
488	*
489	* When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
490	* this special case of epoll.
491	*/
492	#ifdef CONFIG_DEBUG_LOCK_ALLOC
493
494	static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi)
495	{
496	struct eventpoll *ep_src;
497	unsigned long flags;
498	u8 nests = 0;
499
500	/*
501	* To set the subclass or nesting level for spin_lock_irqsave_nested()
502	* it might be natural to create a per-cpu nest count. However, since
503	* we can recurse on ep->poll_wait.lock, and a non-raw spinlock can
504	* schedule() in the -rt kernel, the per-cpu variable are no longer
505	* protected. Thus, we are introducing a per eventpoll nest field.
506	* If we are not being call from ep_poll_callback(), epi is NULL and
507	* we are at the first level of nesting, 0. Otherwise, we are being
508	* called from ep_poll_callback() and if a previous wakeup source is
509	* not an epoll file itself, we are at depth 1 since the wakeup source
510	* is depth 0. If the wakeup source is a previous epoll file in the
511	* wakeup chain then we use its nests value and record ours as
512	* nests + 1. The previous epoll file nests value is stable since its
513	* already holding its own poll_wait.lock.
514	*/
515	if (epi) {
516	if ((is_file_epoll(epi->ffd.file))) {
517	ep_src = epi->ffd.file->private_data;
518	nests = ep_src->nests;
519	} else {
520	nests = 1;
521	}
522	}
523	spin_lock_irqsave_nested(&ep->poll_wait.lock, flags, nests);
524	ep->nests = nests + 1;
525	wake_up_locked_poll(&ep->poll_wait, EPOLLIN);
526	ep->nests = 0;
527	spin_unlock_irqrestore(&ep->poll_wait.lock, flags);
528	}
529
530	#else
531
532	static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi)
533	{
534	wake_up_poll(&ep->poll_wait, EPOLLIN);
535	}
536
537	#endif
538
539	static void ep_remove_wait_queue(struct eppoll_entry *pwq)
540	{
541	wait_queue_head_t *whead;
542
543	rcu_read_lock();
544	/*
545	* If it is cleared by POLLFREE, it should be rcu-safe.
546	* If we read NULL we need a barrier paired with
547	* smp_store_release() in ep_poll_callback(), otherwise
548	* we rely on whead->lock.
549	*/
550	whead = smp_load_acquire(&pwq->whead);
551	if (whead)
552	remove_wait_queue(whead, &pwq->wait);
553	rcu_read_unlock();
554	}
555
556	/*
557	* This function unregisters poll callbacks from the associated file
558	* descriptor. Must be called with "mtx" held (or "epmutex" if called from
559	* ep_free).
560	*/
561	static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
562	{
563	struct eppoll_entry **p = &epi->pwqlist;
564	struct eppoll_entry *pwq;
565
566	while ((pwq = *p) != NULL) {
567	*p = pwq->next;
568	ep_remove_wait_queue(pwq);
569	kmem_cache_free(pwq_cache, pwq);
570	}
571	}
572
573	/* call only when ep->mtx is held */
574	static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
575	{
576	return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
577	}
578
579	/* call only when ep->mtx is held */
580	static inline void ep_pm_stay_awake(struct epitem *epi)
581	{
582	struct wakeup_source *ws = ep_wakeup_source(epi);
583
584	if (ws)
585	__pm_stay_awake(ws);
586	}
587
588	static inline bool ep_has_wakeup_source(struct epitem *epi)
589	{
590	return rcu_access_pointer(epi->ws) ? true : false;
591	}
592
593	/* call when ep->mtx cannot be held (ep_poll_callback) */
594	static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
595	{
596	struct wakeup_source *ws;
597
598	rcu_read_lock();
599	ws = rcu_dereference(epi->ws);
600	if (ws)
601	__pm_stay_awake(ws);
602	rcu_read_unlock();
603	}
604
605
606	/*
607	* ep->mutex needs to be held because we could be hit by
608	* eventpoll_release_file() and epoll_ctl().
609	*/
610	static void ep_start_scan(struct eventpoll *ep, struct list_head *txlist)
611	{
612	/*
613	* Steal the ready list, and re-init the original one to the
614	* empty list. Also, set ep->ovflist to NULL so that events
615	* happening while looping w/out locks, are not lost. We cannot
616	* have the poll callback to queue directly on ep->rdllist,
617	* because we want the "sproc" callback to be able to do it
618	* in a lockless way.
619	*/
620	lockdep_assert_irqs_enabled();
621	write_lock_irq(&ep->lock);
622	list_splice_init(&ep->rdllist, txlist);
623	WRITE_ONCE(ep->ovflist, NULL);
624	write_unlock_irq(&ep->lock);
625	}
626
627	static void ep_done_scan(struct eventpoll *ep,
628	struct list_head *txlist)
629	{
630	struct epitem *epi, *nepi;
631
632	write_lock_irq(&ep->lock);
633	/*
634	* During the time we spent inside the "sproc" callback, some
635	* other events might have been queued by the poll callback.
636	* We re-insert them inside the main ready-list here.
637	*/
638	for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL;
639	nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
640	/*
641	* We need to check if the item is already in the list.
642	* During the "sproc" callback execution time, items are
643	* queued into ->ovflist but the "txlist" might already
644	* contain them, and the list_splice() below takes care of them.
645	*/
646	if (!ep_is_linked(epi)) {
647	/*
648	* ->ovflist is LIFO, so we have to reverse it in order
649	* to keep in FIFO.
650	*/
651	list_add(&epi->rdllink, &ep->rdllist);
652	ep_pm_stay_awake(epi);
653	}
654	}
655	/*
656	* We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
657	* releasing the lock, events will be queued in the normal way inside
658	* ep->rdllist.
659	*/
660	WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR);
661
662	/*
663	* Quickly re-inject items left on "txlist".
664	*/
665	list_splice(txlist, &ep->rdllist);
666	__pm_relax(ep->ws);
667
668	if (!list_empty(&ep->rdllist)) {
669	if (waitqueue_active(&ep->wq))
670	wake_up(&ep->wq);
671	}
672
673	write_unlock_irq(&ep->lock);
674	}
675
676	static void epi_rcu_free(struct rcu_head *head)
677	{
678	struct epitem *epi = container_of(head, struct epitem, rcu);
679	kmem_cache_free(epi_cache, epi);
680	}
681
682	/*
683	* Removes a "struct epitem" from the eventpoll RB tree and deallocates
684	* all the associated resources. Must be called with "mtx" held.
685	*/
686	static int ep_remove(struct eventpoll *ep, struct epitem *epi)
687	{
688	struct file *file = epi->ffd.file;
689	struct epitems_head *to_free;
690	struct hlist_head *head;
691
692	lockdep_assert_irqs_enabled();
693
694	/*
695	* Removes poll wait queue hooks.
696	*/
697	ep_unregister_pollwait(ep, epi);
698
699	/* Remove the current item from the list of epoll hooks */
700	spin_lock(&file->f_lock);
701	to_free = NULL;
702	head = file->f_ep;
703	if (head->first == &epi->fllink && !epi->fllink.next) {
704	file->f_ep = NULL;
705	if (!is_file_epoll(file)) {
706	struct epitems_head *v;
707	v = container_of(head, struct epitems_head, epitems);
708	if (!smp_load_acquire(&v->next))
709	to_free = v;
710	}
711	}
712	hlist_del_rcu(&epi->fllink);
713	spin_unlock(&file->f_lock);
714	free_ephead(to_free);
715
716	rb_erase_cached(&epi->rbn, &ep->rbr);
717
718	write_lock_irq(&ep->lock);
719	if (ep_is_linked(epi))
720	list_del_init(&epi->rdllink);
721	write_unlock_irq(&ep->lock);
722
723	wakeup_source_unregister(ep_wakeup_source(epi));
724	/*
725	* At this point it is safe to free the eventpoll item. Use the union
726	* field epi->rcu, since we are trying to minimize the size of
727	* 'struct epitem'. The 'rbn' field is no longer in use. Protected by
728	* ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
729	* use of the rbn field.
730	*/
731	call_rcu(&epi->rcu, epi_rcu_free);
732
733	percpu_counter_dec(&ep->user->epoll_watches);
734
735	return 0;
736	}
737
738	static void ep_free(struct eventpoll *ep)
739	{
740	struct rb_node *rbp;
741	struct epitem *epi;
742
743	/* We need to release all tasks waiting for these file */
744	if (waitqueue_active(&ep->poll_wait))
745	ep_poll_safewake(ep, NULL);
746
747	/*
748	* We need to lock this because we could be hit by
749	* eventpoll_release_file() while we're freeing the "struct eventpoll".
750	* We do not need to hold "ep->mtx" here because the epoll file
751	* is on the way to be removed and no one has references to it
752	* anymore. The only hit might come from eventpoll_release_file() but
753	* holding "epmutex" is sufficient here.
754	*/
755	mutex_lock(&epmutex);
756
757	/*
758	* Walks through the whole tree by unregistering poll callbacks.
759	*/
760	for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
761	epi = rb_entry(rbp, struct epitem, rbn);
762
763	ep_unregister_pollwait(ep, epi);
764	cond_resched();
765	}
766
767	/*
768	* Walks through the whole tree by freeing each "struct epitem". At this
769	* point we are sure no poll callbacks will be lingering around, and also by
770	* holding "epmutex" we can be sure that no file cleanup code will hit
771	* us during this operation. So we can avoid the lock on "ep->lock".
772	* We do not need to lock ep->mtx, either, we only do it to prevent
773	* a lockdep warning.
774	*/
775	mutex_lock(&ep->mtx);
776	while ((rbp = rb_first_cached(&ep->rbr)) != NULL) {
777	epi = rb_entry(rbp, struct epitem, rbn);
778	ep_remove(ep, epi);
779	cond_resched();
780	}
781	mutex_unlock(&ep->mtx);
782
783	mutex_unlock(&epmutex);
784	mutex_destroy(&ep->mtx);
785	free_uid(ep->user);
786	wakeup_source_unregister(ep->ws);
787	kfree(ep);
788	}
789
790	static int ep_eventpoll_release(struct inode *inode, struct file *file)
791	{
792	struct eventpoll *ep = file->private_data;
793
794	if (ep)
795	ep_free(ep);
796
797	return 0;
798	}
799
800	static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, int depth);
801
802	static __poll_t __ep_eventpoll_poll(struct file *file, poll_table *wait, int depth)
803	{
804	struct eventpoll *ep = file->private_data;
805	LIST_HEAD(txlist);
806	struct epitem *epi, *tmp;
807	poll_table pt;
808	__poll_t res = 0;
809
810	init_poll_funcptr(&pt, NULL);
811
812	/* Insert inside our poll wait queue */
813	poll_wait(file, &ep->poll_wait, wait);
814
815	/*
816	* Proceed to find out if wanted events are really available inside
817	* the ready list.
818	*/
819	mutex_lock_nested(&ep->mtx, depth);
820	ep_start_scan(ep, &txlist);
821	list_for_each_entry_safe(epi, tmp, &txlist, rdllink) {
822	if (ep_item_poll(epi, &pt, depth + 1)) {
823	res = EPOLLIN | EPOLLRDNORM;
824	break;
825	} else {
826	/*
827	* Item has been dropped into the ready list by the poll
828	* callback, but it's not actually ready, as far as
829	* caller requested events goes. We can remove it here.
830	*/
831	__pm_relax(ep_wakeup_source(epi));
832	list_del_init(&epi->rdllink);
833	}
834	}
835	ep_done_scan(ep, &txlist);
836	mutex_unlock(&ep->mtx);
837	return res;
838	}
839
840	/*
841	* Differs from ep_eventpoll_poll() in that internal callers already have
842	* the ep->mtx so we need to start from depth=1, such that mutex_lock_nested()
843	* is correctly annotated.
844	*/
845	static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt,
846	int depth)
847	{
848	struct file *file = epi->ffd.file;
849	__poll_t res;
850
851	pt->_key = epi->event.events;
852	if (!is_file_epoll(file))
853	res = vfs_poll(file, pt);
854	else
855	res = __ep_eventpoll_poll(file, pt, depth);
856	return res & epi->event.events;
857	}
858
859	static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait)
860	{
861	return __ep_eventpoll_poll(file, wait, 0);
862	}
863
864	#ifdef CONFIG_PROC_FS
865	static void ep_show_fdinfo(struct seq_file *m, struct file *f)
866	{
867	struct eventpoll *ep = f->private_data;
868	struct rb_node *rbp;
869
870	mutex_lock(&ep->mtx);
871	for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
872	struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
873	struct inode *inode = file_inode(epi->ffd.file);
874
875	seq_printf(m, "tfd: %8d events: %8x data: %16llx "
876	" pos:%lli ino:%lx sdev:%x\n",
877	epi->ffd.fd, epi->event.events,
878	(long long)epi->event.data,
879	(long long)epi->ffd.file->f_pos,
880	inode->i_ino, inode->i_sb->s_dev);
881	if (seq_has_overflowed(m))
882	break;
883	}
884	mutex_unlock(&ep->mtx);
885	}
886	#endif
887
888	/* File callbacks that implement the eventpoll file behaviour */
889	static const struct file_operations eventpoll_fops = {
890	#ifdef CONFIG_PROC_FS
891	.show_fdinfo = ep_show_fdinfo,
892	#endif
893	.release = ep_eventpoll_release,
894	.poll = ep_eventpoll_poll,
895	.llseek = noop_llseek,
896	};
897
898	/*
899	* This is called from eventpoll_release() to unlink files from the eventpoll
900	* interface. We need to have this facility to cleanup correctly files that are
901	* closed without being removed from the eventpoll interface.
902	*/
903	void eventpoll_release_file(struct file *file)
904	{
905	struct eventpoll *ep;
906	struct epitem *epi;
907	struct hlist_node *next;
908
909	/*
910	* We don't want to get "file->f_lock" because it is not
911	* necessary. It is not necessary because we're in the "struct file"
912	* cleanup path, and this means that no one is using this file anymore.
913	* So, for example, epoll_ctl() cannot hit here since if we reach this
914	* point, the file counter already went to zero and fget() would fail.
915	* The only hit might come from ep_free() but by holding the mutex
916	* will correctly serialize the operation. We do need to acquire
917	* "ep->mtx" after "epmutex" because ep_remove() requires it when called
918	* from anywhere but ep_free().
919	*
920	* Besides, ep_remove() acquires the lock, so we can't hold it here.
921	*/
922	mutex_lock(&epmutex);
923	if (unlikely(!file->f_ep)) {
924	mutex_unlock(&epmutex);
925	return;
926	}
927	hlist_for_each_entry_safe(epi, next, file->f_ep, fllink) {
928	ep = epi->ep;
929	mutex_lock_nested(&ep->mtx, 0);
930	ep_remove(ep, epi);
931	mutex_unlock(&ep->mtx);
932	}
933	mutex_unlock(&epmutex);
934	}
935
936	static int ep_alloc(struct eventpoll **pep)
937	{
938	int error;
939	struct user_struct *user;
940	struct eventpoll *ep;
941
942	user = get_current_user();
943	error = -ENOMEM;
944	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
945	if (unlikely(!ep))
946	goto free_uid;
947
948	mutex_init(&ep->mtx);
949	rwlock_init(&ep->lock);
950	init_waitqueue_head(&ep->wq);
951	init_waitqueue_head(&ep->poll_wait);
952	INIT_LIST_HEAD(&ep->rdllist);
953	ep->rbr = RB_ROOT_CACHED;
954	ep->ovflist = EP_UNACTIVE_PTR;
955	ep->user = user;
956
957	*pep = ep;
958
959	return 0;
960
961	free_uid:
962	free_uid(user);
963	return error;
964	}
965
966	/*
967	* Search the file inside the eventpoll tree. The RB tree operations
968	* are protected by the "mtx" mutex, and ep_find() must be called with
969	* "mtx" held.
970	*/
971	static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
972	{
973	int kcmp;
974	struct rb_node *rbp;
975	struct epitem *epi, *epir = NULL;
976	struct epoll_filefd ffd;
977
978	ep_set_ffd(&ffd, file, fd);
979	for (rbp = ep->rbr.rb_root.rb_node; rbp; ) {
980	epi = rb_entry(rbp, struct epitem, rbn);
981	kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
982	if (kcmp > 0)
983	rbp = rbp->rb_right;
984	else if (kcmp < 0)
985	rbp = rbp->rb_left;
986	else {
987	epir = epi;
988	break;
989	}
990	}
991
992	return epir;
993	}
994
995	#ifdef CONFIG_KCMP
996	static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff)
997	{
998	struct rb_node *rbp;
999	struct epitem *epi;
1000
1001	for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1002	epi = rb_entry(rbp, struct epitem, rbn);
1003	if (epi->ffd.fd == tfd) {
1004	if (toff == 0)
1005	return epi;
1006	else
1007	toff--;
1008	}
1009	cond_resched();
1010	}
1011
1012	return NULL;
1013	}
1014
1015	struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd,
1016	unsigned long toff)
1017	{
1018	struct file *file_raw;
1019	struct eventpoll *ep;
1020	struct epitem *epi;
1021
1022	if (!is_file_epoll(file))
1023	return ERR_PTR(-EINVAL);
1024
1025	ep = file->private_data;
1026
1027	mutex_lock(&ep->mtx);
1028	epi = ep_find_tfd(ep, tfd, toff);
1029	if (epi)
1030	file_raw = epi->ffd.file;
1031	else
1032	file_raw = ERR_PTR(-ENOENT);
1033	mutex_unlock(&ep->mtx);
1034
1035	return file_raw;
1036	}
1037	#endif /* CONFIG_KCMP */
1038
1039	/*
1040	* Adds a new entry to the tail of the list in a lockless way, i.e.
1041	* multiple CPUs are allowed to call this function concurrently.
1042	*
1043	* Beware: it is necessary to prevent any other modifications of the
1044	* existing list until all changes are completed, in other words
1045	* concurrent list_add_tail_lockless() calls should be protected
1046	* with a read lock, where write lock acts as a barrier which
1047	* makes sure all list_add_tail_lockless() calls are fully
1048	* completed.
1049	*
1050	* Also an element can be locklessly added to the list only in one
1051	* direction i.e. either to the tail or to the head, otherwise
1052	* concurrent access will corrupt the list.
1053	*
1054	* Return: %false if element has been already added to the list, %true
1055	* otherwise.
1056	*/
1057	static inline bool list_add_tail_lockless(struct list_head *new,
1058	struct list_head *head)
1059	{
1060	struct list_head *prev;
1061
1062	/*
1063	* This is simple 'new->next = head' operation, but cmpxchg()
1064	* is used in order to detect that same element has been just
1065	* added to the list from another CPU: the winner observes
1066	* new->next == new.
1067	*/
1068	if (cmpxchg(&new->next, new, head) != new)
1069	return false;
1070
1071	/*
1072	* Initially ->next of a new element must be updated with the head
1073	* (we are inserting to the tail) and only then pointers are atomically
1074	* exchanged. XCHG guarantees memory ordering, thus ->next should be
1075	* updated before pointers are actually swapped and pointers are
1076	* swapped before prev->next is updated.
1077	*/
1078
1079	prev = xchg(&head->prev, new);
1080
1081	/*
1082	* It is safe to modify prev->next and new->prev, because a new element
1083	* is added only to the tail and new->next is updated before XCHG.
1084	*/
1085
1086	prev->next = new;
1087	new->prev = prev;
1088
1089	return true;
1090	}
1091
1092	/*
1093	* Chains a new epi entry to the tail of the ep->ovflist in a lockless way,
1094	* i.e. multiple CPUs are allowed to call this function concurrently.
1095	*
1096	* Return: %false if epi element has been already chained, %true otherwise.
1097	*/
1098	static inline bool chain_epi_lockless(struct epitem *epi)
1099	{
1100	struct eventpoll *ep = epi->ep;
1101
1102	/* Fast preliminary check */
1103	if (epi->next != EP_UNACTIVE_PTR)
1104	return false;
1105
1106	/* Check that the same epi has not been just chained from another CPU */
1107	if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR)
1108	return false;
1109
1110	/* Atomically exchange tail */
1111	epi->next = xchg(&ep->ovflist, epi);
1112
1113	return true;
1114	}
1115
1116	/*
1117	* This is the callback that is passed to the wait queue wakeup
1118	* mechanism. It is called by the stored file descriptors when they
1119	* have events to report.
1120	*
1121	* This callback takes a read lock in order not to contend with concurrent
1122	* events from another file descriptor, thus all modifications to ->rdllist
1123	* or ->ovflist are lockless. Read lock is paired with the write lock from
1124	* ep_scan_ready_list(), which stops all list modifications and guarantees
1125	* that lists state is seen correctly.
1126	*
1127	* Another thing worth to mention is that ep_poll_callback() can be called
1128	* concurrently for the same @epi from different CPUs if poll table was inited
1129	* with several wait queues entries. Plural wakeup from different CPUs of a
1130	* single wait queue is serialized by wq.lock, but the case when multiple wait
1131	* queues are used should be detected accordingly. This is detected using
1132	* cmpxchg() operation.
1133	*/
1134	static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1135	{
1136	int pwake = 0;
1137	struct epitem *epi = ep_item_from_wait(wait);
1138	struct eventpoll *ep = epi->ep;
1139	__poll_t pollflags = key_to_poll(key);
1140	unsigned long flags;
1141	int ewake = 0;
1142
1143	read_lock_irqsave(&ep->lock, flags);
1144
1145	ep_set_busy_poll_napi_id(epi);
1146
1147	/*
1148	* If the event mask does not contain any poll(2) event, we consider the
1149	* descriptor to be disabled. This condition is likely the effect of the
1150	* EPOLLONESHOT bit that disables the descriptor when an event is received,
1151	* until the next EPOLL_CTL_MOD will be issued.
1152	*/
1153	if (!(epi->event.events & ~EP_PRIVATE_BITS))
1154	goto out_unlock;
1155
1156	/*
1157	* Check the events coming with the callback. At this stage, not
1158	* every device reports the events in the "key" parameter of the
1159	* callback. We need to be able to handle both cases here, hence the
1160	* test for "key" != NULL before the event match test.
1161	*/
1162	if (pollflags && !(pollflags & epi->event.events))
1163	goto out_unlock;
1164
1165	/*
1166	* If we are transferring events to userspace, we can hold no locks
1167	* (because we're accessing user memory, and because of linux f_op->poll()
1168	* semantics). All the events that happen during that period of time are
1169	* chained in ep->ovflist and requeued later on.
1170	*/
1171	if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) {
1172	if (chain_epi_lockless(epi))
1173	ep_pm_stay_awake_rcu(epi);
1174	} else if (!ep_is_linked(epi)) {
1175	/* In the usual case, add event to ready list. */
1176	if (list_add_tail_lockless(&epi->rdllink, &ep->rdllist))
1177	ep_pm_stay_awake_rcu(epi);
1178	}
1179
1180	/*
1181	* Wake up ( if active ) both the eventpoll wait list and the ->poll()
1182	* wait list.
1183	*/
1184	if (waitqueue_active(&ep->wq)) {
1185	if ((epi->event.events & EPOLLEXCLUSIVE) &&
1186	!(pollflags & POLLFREE)) {
1187	switch (pollflags & EPOLLINOUT_BITS) {
1188	case EPOLLIN:
1189	if (epi->event.events & EPOLLIN)
1190	ewake = 1;
1191	break;
1192	case EPOLLOUT:
1193	if (epi->event.events & EPOLLOUT)
1194	ewake = 1;
1195	break;
1196	case 0:
1197	ewake = 1;
1198	break;
1199	}
1200	}
1201	wake_up(&ep->wq);
1202	}
1203	if (waitqueue_active(&ep->poll_wait))
1204	pwake++;
1205
1206	out_unlock:
1207	read_unlock_irqrestore(&ep->lock, flags);
1208
1209	/* We have to call this outside the lock */
1210	if (pwake)
1211	ep_poll_safewake(ep, epi);
1212
1213	if (!(epi->event.events & EPOLLEXCLUSIVE))
1214	ewake = 1;
1215
1216	if (pollflags & POLLFREE) {
1217	/*
1218	* If we race with ep_remove_wait_queue() it can miss
1219	* ->whead = NULL and do another remove_wait_queue() after
1220	* us, so we can't use __remove_wait_queue().
1221	*/
1222	list_del_init(&wait->entry);
1223	/*
1224	* ->whead != NULL protects us from the race with ep_free()
1225	* or ep_remove(), ep_remove_wait_queue() takes whead->lock
1226	* held by the caller. Once we nullify it, nothing protects
1227	* ep/epi or even wait.
1228	*/
1229	smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL);
1230	}
1231
1232	return ewake;
1233	}
1234
1235	/*
1236	* This is the callback that is used to add our wait queue to the
1237	* target file wakeup lists.
1238	*/
1239	static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1240	poll_table *pt)
1241	{
1242	struct ep_pqueue *epq = container_of(pt, struct ep_pqueue, pt);
1243	struct epitem *epi = epq->epi;
1244	struct eppoll_entry *pwq;
1245
1246	if (unlikely(!epi)) // an earlier allocation has failed
1247	return;
1248
1249	pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL);
1250	if (unlikely(!pwq)) {
1251	epq->epi = NULL;
1252	return;
1253	}
1254
1255	init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1256	pwq->whead = whead;
1257	pwq->base = epi;
1258	if (epi->event.events & EPOLLEXCLUSIVE)
1259	add_wait_queue_exclusive(whead, &pwq->wait);
1260	else
1261	add_wait_queue(whead, &pwq->wait);
1262	pwq->next = epi->pwqlist;
1263	epi->pwqlist = pwq;
1264	}
1265
1266	static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1267	{
1268	int kcmp;
1269	struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL;
1270	struct epitem *epic;
1271	bool leftmost = true;
1272
1273	while (*p) {
1274	parent = *p;
1275	epic = rb_entry(parent, struct epitem, rbn);
1276	kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1277	if (kcmp > 0) {
1278	p = &parent->rb_right;
1279	leftmost = false;
1280	} else
1281	p = &parent->rb_left;
1282	}
1283	rb_link_node(&epi->rbn, parent, p);
1284	rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost);
1285	}
1286
1287
1288
1289	#define PATH_ARR_SIZE 5
1290	/*
1291	* These are the number paths of length 1 to 5, that we are allowing to emanate
1292	* from a single file of interest. For example, we allow 1000 paths of length
1293	* 1, to emanate from each file of interest. This essentially represents the
1294	* potential wakeup paths, which need to be limited in order to avoid massive
1295	* uncontrolled wakeup storms. The common use case should be a single ep which
1296	* is connected to n file sources. In this case each file source has 1 path
1297	* of length 1. Thus, the numbers below should be more than sufficient. These
1298	* path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1299	* and delete can't add additional paths. Protected by the epmutex.
1300	*/
1301	static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1302	static int path_count[PATH_ARR_SIZE];
1303
1304	static int path_count_inc(int nests)
1305	{
1306	/* Allow an arbitrary number of depth 1 paths */
1307	if (nests == 0)
1308	return 0;
1309
1310	if (++path_count[nests] > path_limits[nests])
1311	return -1;
1312	return 0;
1313	}
1314
1315	static void path_count_init(void)
1316	{
1317	int i;
1318
1319	for (i = 0; i < PATH_ARR_SIZE; i++)
1320	path_count[i] = 0;
1321	}
1322
1323	static int reverse_path_check_proc(struct hlist_head *refs, int depth)
1324	{
1325	int error = 0;
1326	struct epitem *epi;
1327
1328	if (depth > EP_MAX_NESTS) /* too deep nesting */
1329	return -1;
1330
1331	/* CTL_DEL can remove links here, but that can't increase our count */
1332	hlist_for_each_entry_rcu(epi, refs, fllink) {
1333	struct hlist_head *refs = &epi->ep->refs;
1334	if (hlist_empty(refs))
1335	error = path_count_inc(depth);
1336	else
1337	error = reverse_path_check_proc(refs, depth + 1);
1338	if (error != 0)
1339	break;
1340	}
1341	return error;
1342	}
1343
1344	/**
1345	* reverse_path_check - The tfile_check_list is list of epitem_head, which have
1346	* links that are proposed to be newly added. We need to
1347	* make sure that those added links don't add too many
1348	* paths such that we will spend all our time waking up
1349	* eventpoll objects.
1350	*
1351	* Return: %zero if the proposed links don't create too many paths,
1352	* %-1 otherwise.
1353	*/
1354	static int reverse_path_check(void)
1355	{
1356	struct epitems_head *p;
1357
1358	for (p = tfile_check_list; p != EP_UNACTIVE_PTR; p = p->next) {
1359	int error;
1360	path_count_init();
1361	rcu_read_lock();
1362	error = reverse_path_check_proc(&p->epitems, 0);
1363	rcu_read_unlock();
1364	if (error)
1365	return error;
1366	}
1367	return 0;
1368	}
1369
1370	static int ep_create_wakeup_source(struct epitem *epi)
1371	{
1372	struct name_snapshot n;
1373	struct wakeup_source *ws;
1374
1375	if (!epi->ep->ws) {
1376	epi->ep->ws = wakeup_source_register(NULL, "eventpoll");
1377	if (!epi->ep->ws)
1378	return -ENOMEM;
1379	}
1380
1381	take_dentry_name_snapshot(&n, epi->ffd.file->f_path.dentry);
1382	ws = wakeup_source_register(NULL, n.name.name);
1383	release_dentry_name_snapshot(&n);
1384
1385	if (!ws)
1386	return -ENOMEM;
1387	rcu_assign_pointer(epi->ws, ws);
1388
1389	return 0;
1390	}
1391
1392	/* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1393	static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1394	{
1395	struct wakeup_source *ws = ep_wakeup_source(epi);
1396
1397	RCU_INIT_POINTER(epi->ws, NULL);
1398
1399	/*
1400	* wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1401	* used internally by wakeup_source_remove, too (called by
1402	* wakeup_source_unregister), so we cannot use call_rcu
1403	*/
1404	synchronize_rcu();
1405	wakeup_source_unregister(ws);
1406	}
1407
1408	static int attach_epitem(struct file *file, struct epitem *epi)
1409	{
1410	struct epitems_head *to_free = NULL;
1411	struct hlist_head *head = NULL;
1412	struct eventpoll *ep = NULL;
1413
1414	if (is_file_epoll(file))
1415	ep = file->private_data;
1416
1417	if (ep) {
1418	head = &ep->refs;
1419	} else if (!READ_ONCE(file->f_ep)) {
1420	allocate:
1421	to_free = kmem_cache_zalloc(ephead_cache, GFP_KERNEL);
1422	if (!to_free)
1423	return -ENOMEM;
1424	head = &to_free->epitems;
1425	}
1426	spin_lock(&file->f_lock);
1427	if (!file->f_ep) {
1428	if (unlikely(!head)) {
1429	spin_unlock(&file->f_lock);
1430	goto allocate;
1431	}
1432	file->f_ep = head;
1433	to_free = NULL;
1434	}
1435	hlist_add_head_rcu(&epi->fllink, file->f_ep);
1436	spin_unlock(&file->f_lock);
1437	free_ephead(to_free);
1438	return 0;
1439	}
1440
1441	/*
1442	* Must be called with "mtx" held.
1443	*/
1444	static int ep_insert(struct eventpoll *ep, const struct epoll_event *event,
1445	struct file *tfile, int fd, int full_check)
1446	{
1447	int error, pwake = 0;
1448	__poll_t revents;
1449	struct epitem *epi;
1450	struct ep_pqueue epq;
1451	struct eventpoll *tep = NULL;
1452
1453	if (is_file_epoll(tfile))
1454	tep = tfile->private_data;
1455
1456	lockdep_assert_irqs_enabled();
1457
1458	if (unlikely(percpu_counter_compare(&ep->user->epoll_watches,
1459	max_user_watches) >= 0))
1460	return -ENOSPC;
1461	percpu_counter_inc(&ep->user->epoll_watches);
1462
1463	if (!(epi = kmem_cache_zalloc(epi_cache, GFP_KERNEL))) {
1464	percpu_counter_dec(&ep->user->epoll_watches);
1465	return -ENOMEM;
1466	}
1467
1468	/* Item initialization follow here ... */
1469	INIT_LIST_HEAD(&epi->rdllink);
1470	epi->ep = ep;
1471	ep_set_ffd(&epi->ffd, tfile, fd);
1472	epi->event = *event;
1473	epi->next = EP_UNACTIVE_PTR;
1474
1475	if (tep)
1476	mutex_lock_nested(&tep->mtx, 1);
1477	/* Add the current item to the list of active epoll hook for this file */
1478	if (unlikely(attach_epitem(tfile, epi) < 0)) {
1479	if (tep)
1480	mutex_unlock(&tep->mtx);
1481	kmem_cache_free(epi_cache, epi);
1482	percpu_counter_dec(&ep->user->epoll_watches);
1483	return -ENOMEM;
1484	}
1485
1486	if (full_check && !tep)
1487	list_file(tfile);
1488
1489	/*
1490	* Add the current item to the RB tree. All RB tree operations are
1491	* protected by "mtx", and ep_insert() is called with "mtx" held.
1492	*/
1493	ep_rbtree_insert(ep, epi);
1494	if (tep)
1495	mutex_unlock(&tep->mtx);
1496
1497	/* now check if we've created too many backpaths */
1498	if (unlikely(full_check && reverse_path_check())) {
1499	ep_remove(ep, epi);
1500	return -EINVAL;
1501	}
1502
1503	if (epi->event.events & EPOLLWAKEUP) {
1504	error = ep_create_wakeup_source(epi);
1505	if (error) {
1506	ep_remove(ep, epi);
1507	return error;
1508	}
1509	}
1510
1511	/* Initialize the poll table using the queue callback */
1512	epq.epi = epi;
1513	init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1514
1515	/*
1516	* Attach the item to the poll hooks and get current event bits.
1517	* We can safely use the file* here because its usage count has
1518	* been increased by the caller of this function. Note that after
1519	* this operation completes, the poll callback can start hitting
1520	* the new item.
1521	*/
1522	revents = ep_item_poll(epi, &epq.pt, 1);
1523
1524	/*
1525	* We have to check if something went wrong during the poll wait queue
1526	* install process. Namely an allocation for a wait queue failed due
1527	* high memory pressure.
1528	*/
1529	if (unlikely(!epq.epi)) {
1530	ep_remove(ep, epi);
1531	return -ENOMEM;
1532	}
1533
1534	/* We have to drop the new item inside our item list to keep track of it */
1535	write_lock_irq(&ep->lock);
1536
1537	/* record NAPI ID of new item if present */
1538	ep_set_busy_poll_napi_id(epi);
1539
1540	/* If the file is already "ready" we drop it inside the ready list */
1541	if (revents && !ep_is_linked(epi)) {
1542	list_add_tail(&epi->rdllink, &ep->rdllist);
1543	ep_pm_stay_awake(epi);
1544
1545	/* Notify waiting tasks that events are available */
1546	if (waitqueue_active(&ep->wq))
1547	wake_up(&ep->wq);
1548	if (waitqueue_active(&ep->poll_wait))
1549	pwake++;
1550	}
1551
1552	write_unlock_irq(&ep->lock);
1553
1554	/* We have to call this outside the lock */
1555	if (pwake)
1556	ep_poll_safewake(ep, NULL);
1557
1558	return 0;
1559	}
1560
1561	/*
1562	* Modify the interest event mask by dropping an event if the new mask
1563	* has a match in the current file status. Must be called with "mtx" held.
1564	*/
1565	static int ep_modify(struct eventpoll *ep, struct epitem *epi,
1566	const struct epoll_event *event)
1567	{
1568	int pwake = 0;
1569	poll_table pt;
1570
1571	lockdep_assert_irqs_enabled();
1572
1573	init_poll_funcptr(&pt, NULL);
1574
1575	/*
1576	* Set the new event interest mask before calling f_op->poll();
1577	* otherwise we might miss an event that happens between the
1578	* f_op->poll() call and the new event set registering.
1579	*/
1580	epi->event.events = event->events; /* need barrier below */
1581	epi->event.data = event->data; /* protected by mtx */
1582	if (epi->event.events & EPOLLWAKEUP) {
1583	if (!ep_has_wakeup_source(epi))
1584	ep_create_wakeup_source(epi);
1585	} else if (ep_has_wakeup_source(epi)) {
1586	ep_destroy_wakeup_source(epi);
1587	}
1588
1589	/*
1590	* The following barrier has two effects:
1591	*
1592	* 1) Flush epi changes above to other CPUs. This ensures
1593	* we do not miss events from ep_poll_callback if an
1594	* event occurs immediately after we call f_op->poll().
1595	* We need this because we did not take ep->lock while
1596	* changing epi above (but ep_poll_callback does take
1597	* ep->lock).
1598	*
1599	* 2) We also need to ensure we do not miss _past_ events
1600	* when calling f_op->poll(). This barrier also
1601	* pairs with the barrier in wq_has_sleeper (see
1602	* comments for wq_has_sleeper).
1603	*
1604	* This barrier will now guarantee ep_poll_callback or f_op->poll
1605	* (or both) will notice the readiness of an item.
1606	*/
1607	smp_mb();
1608
1609	/*
1610	* Get current event bits. We can safely use the file* here because
1611	* its usage count has been increased by the caller of this function.
1612	* If the item is "hot" and it is not registered inside the ready
1613	* list, push it inside.
1614	*/
1615	if (ep_item_poll(epi, &pt, 1)) {
1616	write_lock_irq(&ep->lock);
1617	if (!ep_is_linked(epi)) {
1618	list_add_tail(&epi->rdllink, &ep->rdllist);
1619	ep_pm_stay_awake(epi);
1620
1621	/* Notify waiting tasks that events are available */
1622	if (waitqueue_active(&ep->wq))
1623	wake_up(&ep->wq);
1624	if (waitqueue_active(&ep->poll_wait))
1625	pwake++;
1626	}
1627	write_unlock_irq(&ep->lock);
1628	}
1629
1630	/* We have to call this outside the lock */
1631	if (pwake)
1632	ep_poll_safewake(ep, NULL);
1633
1634	return 0;
1635	}
1636
1637	static int ep_send_events(struct eventpoll *ep,
1638	struct epoll_event __user *events, int maxevents)
1639	{
1640	struct epitem *epi, *tmp;
1641	LIST_HEAD(txlist);
1642	poll_table pt;
1643	int res = 0;
1644
1645	/*
1646	* Always short-circuit for fatal signals to allow threads to make a
1647	* timely exit without the chance of finding more events available and
1648	* fetching repeatedly.
1649	*/
1650	if (fatal_signal_pending(current))
1651	return -EINTR;
1652
1653	init_poll_funcptr(&pt, NULL);
1654
1655	mutex_lock(&ep->mtx);
1656	ep_start_scan(ep, &txlist);
1657
1658	/*
1659	* We can loop without lock because we are passed a task private list.
1660	* Items cannot vanish during the loop we are holding ep->mtx.
1661	*/
1662	list_for_each_entry_safe(epi, tmp, &txlist, rdllink) {
1663	struct wakeup_source *ws;
1664	__poll_t revents;
1665
1666	if (res >= maxevents)
1667	break;
1668
1669	/*
1670	* Activate ep->ws before deactivating epi->ws to prevent
1671	* triggering auto-suspend here (in case we reactive epi->ws
1672	* below).
1673	*
1674	* This could be rearranged to delay the deactivation of epi->ws
1675	* instead, but then epi->ws would temporarily be out of sync
1676	* with ep_is_linked().
1677	*/
1678	ws = ep_wakeup_source(epi);
1679	if (ws) {
1680	if (ws->active)
1681	__pm_stay_awake(ep->ws);
1682	__pm_relax(ws);
1683	}
1684
1685	list_del_init(&epi->rdllink);
1686
1687	/*
1688	* If the event mask intersect the caller-requested one,
1689	* deliver the event to userspace. Again, we are holding ep->mtx,
1690	* so no operations coming from userspace can change the item.
1691	*/
1692	revents = ep_item_poll(epi, &pt, 1);
1693	if (!revents)
1694	continue;
1695
1696	events = epoll_put_uevent(revents, epi->event.data, events);
1697	if (!events) {
1698	list_add(&epi->rdllink, &txlist);
1699	ep_pm_stay_awake(epi);
1700	if (!res)
1701	res = -EFAULT;
1702	break;
1703	}
1704	res++;
1705	if (epi->event.events & EPOLLONESHOT)
1706	epi->event.events &= EP_PRIVATE_BITS;
1707	else if (!(epi->event.events & EPOLLET)) {
1708	/*
1709	* If this file has been added with Level
1710	* Trigger mode, we need to insert back inside
1711	* the ready list, so that the next call to
1712	* epoll_wait() will check again the events
1713	* availability. At this point, no one can insert
1714	* into ep->rdllist besides us. The epoll_ctl()
1715	* callers are locked out by
1716	* ep_scan_ready_list() holding "mtx" and the
1717	* poll callback will queue them in ep->ovflist.
1718	*/
1719	list_add_tail(&epi->rdllink, &ep->rdllist);
1720	ep_pm_stay_awake(epi);
1721	}
1722	}
1723	ep_done_scan(ep, &txlist);
1724	mutex_unlock(&ep->mtx);
1725
1726	return res;
1727	}
1728
1729	static struct timespec64 *ep_timeout_to_timespec(struct timespec64 *to, long ms)
1730	{
1731	struct timespec64 now;
1732
1733	if (ms < 0)
1734	return NULL;
1735
1736	if (!ms) {
1737	to->tv_sec = 0;
1738	to->tv_nsec = 0;
1739	return to;
1740	}
1741
1742	to->tv_sec = ms / MSEC_PER_SEC;
1743	to->tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC);
1744
1745	ktime_get_ts64(&now);
1746	*to = timespec64_add_safe(now, *to);
1747	return to;
1748	}
1749
1750	/*
1751	* autoremove_wake_function, but remove even on failure to wake up, because we
1752	* know that default_wake_function/ttwu will only fail if the thread is already
1753	* woken, and in that case the ep_poll loop will remove the entry anyways, not
1754	* try to reuse it.
1755	*/
1756	static int ep_autoremove_wake_function(struct wait_queue_entry *wq_entry,
1757	unsigned int mode, int sync, void *key)
1758	{
1759	int ret = default_wake_function(wq_entry, mode, sync, key);
1760
1761	list_del_init(&wq_entry->entry);
1762	return ret;
1763	}
1764
1765	/**
1766	* ep_poll - Retrieves ready events, and delivers them to the caller-supplied
1767	* event buffer.
1768	*
1769	* @ep: Pointer to the eventpoll context.
1770	* @events: Pointer to the userspace buffer where the ready events should be
1771	* stored.
1772	* @maxevents: Size (in terms of number of events) of the caller event buffer.
1773	* @timeout: Maximum timeout for the ready events fetch operation, in
1774	* timespec. If the timeout is zero, the function will not block,
1775	* while if the @timeout ptr is NULL, the function will block
1776	* until at least one event has been retrieved (or an error
1777	* occurred).
1778	*
1779	* Return: the number of ready events which have been fetched, or an
1780	* error code, in case of error.
1781	*/
1782	static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1783	int maxevents, struct timespec64 *timeout)
1784	{
1785	int res, eavail, timed_out = 0;
1786	u64 slack = 0;
1787	wait_queue_entry_t wait;
1788	ktime_t expires, *to = NULL;
1789
1790	lockdep_assert_irqs_enabled();
1791
1792	if (timeout && (timeout->tv_sec | timeout->tv_nsec)) {
1793	slack = select_estimate_accuracy(timeout);
1794	to = &expires;
1795	*to = timespec64_to_ktime(*timeout);
1796	} else if (timeout) {
1797	/*
1798	* Avoid the unnecessary trip to the wait queue loop, if the
1799	* caller specified a non blocking operation.
1800	*/
1801	timed_out = 1;
1802	}
1803
1804	/*
1805	* This call is racy: We may or may not see events that are being added
1806	* to the ready list under the lock (e.g., in IRQ callbacks). For cases
1807	* with a non-zero timeout, this thread will check the ready list under
1808	* lock and will add to the wait queue. For cases with a zero
1809	* timeout, the user by definition should not care and will have to
1810	* recheck again.
1811	*/
1812	eavail = ep_events_available(ep);
1813
1814	while (1) {
1815	if (eavail) {
1816	/*
1817	* Try to transfer events to user space. In case we get
1818	* 0 events and there's still timeout left over, we go
1819	* trying again in search of more luck.
1820	*/
1821	res = ep_send_events(ep, events, maxevents);
1822	if (res)
1823	return res;
1824	}
1825
1826	if (timed_out)
1827	return 0;
1828
1829	eavail = ep_busy_loop(ep, timed_out);
1830	if (eavail)
1831	continue;
1832
1833	if (signal_pending(current))
1834	return -EINTR;
1835
1836	/*
1837	* Internally init_wait() uses autoremove_wake_function(),
1838	* thus wait entry is removed from the wait queue on each
1839	* wakeup. Why it is important? In case of several waiters
1840	* each new wakeup will hit the next waiter, giving it the
1841	* chance to harvest new event. Otherwise wakeup can be
1842	* lost. This is also good performance-wise, because on
1843	* normal wakeup path no need to call __remove_wait_queue()
1844	* explicitly, thus ep->lock is not taken, which halts the
1845	* event delivery.
1846	*
1847	* In fact, we now use an even more aggressive function that
1848	* unconditionally removes, because we don't reuse the wait
1849	* entry between loop iterations. This lets us also avoid the
1850	* performance issue if a process is killed, causing all of its
1851	* threads to wake up without being removed normally.
1852	*/
1853	init_wait(&wait);
1854	wait.func = ep_autoremove_wake_function;
1855
1856	write_lock_irq(&ep->lock);
1857	/*
1858	* Barrierless variant, waitqueue_active() is called under
1859	* the same lock on wakeup ep_poll_callback() side, so it
1860	* is safe to avoid an explicit barrier.
1861	*/
1862	__set_current_state(TASK_INTERRUPTIBLE);
1863
1864	/*
1865	* Do the final check under the lock. ep_scan_ready_list()
1866	* plays with two lists (->rdllist and ->ovflist) and there
1867	* is always a race when both lists are empty for short
1868	* period of time although events are pending, so lock is
1869	* important.
1870	*/
1871	eavail = ep_events_available(ep);
1872	if (!eavail)
1873	__add_wait_queue_exclusive(&ep->wq, &wait);
1874
1875	write_unlock_irq(&ep->lock);
1876
1877	if (!eavail)
1878	timed_out = !schedule_hrtimeout_range(to, slack,
1879	HRTIMER_MODE_ABS);
1880	__set_current_state(TASK_RUNNING);
1881
1882	/*
1883	* We were woken up, thus go and try to harvest some events.
1884	* If timed out and still on the wait queue, recheck eavail
1885	* carefully under lock, below.
1886	*/
1887	eavail = 1;
1888
1889	if (!list_empty_careful(&wait.entry)) {
1890	write_lock_irq(&ep->lock);
1891	/*
1892	* If the thread timed out and is not on the wait queue,
1893	* it means that the thread was woken up after its
1894	* timeout expired before it could reacquire the lock.
1895	* Thus, when wait.entry is empty, it needs to harvest
1896	* events.
1897	*/
1898	if (timed_out)
1899	eavail = list_empty(&wait.entry);
1900	__remove_wait_queue(&ep->wq, &wait);
1901	write_unlock_irq(&ep->lock);
1902	}
1903	}
1904	}
1905
1906	/**
1907	* ep_loop_check_proc - verify that adding an epoll file inside another
1908	* epoll structure does not violate the constraints, in
1909	* terms of closed loops, or too deep chains (which can
1910	* result in excessive stack usage).
1911	*
1912	* @ep: the &struct eventpoll to be currently checked.
1913	* @depth: Current depth of the path being checked.
1914	*
1915	* Return: %zero if adding the epoll @file inside current epoll
1916	* structure @ep does not violate the constraints, or %-1 otherwise.
1917	*/
1918	static int ep_loop_check_proc(struct eventpoll *ep, int depth)
1919	{
1920	int error = 0;
1921	struct rb_node *rbp;
1922	struct epitem *epi;
1923
1924	mutex_lock_nested(&ep->mtx, depth + 1);
1925	ep->gen = loop_check_gen;
1926	for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1927	epi = rb_entry(rbp, struct epitem, rbn);
1928	if (unlikely(is_file_epoll(epi->ffd.file))) {
1929	struct eventpoll *ep_tovisit;
1930	ep_tovisit = epi->ffd.file->private_data;
1931	if (ep_tovisit->gen == loop_check_gen)
1932	continue;
1933	if (ep_tovisit == inserting_into || depth > EP_MAX_NESTS)
1934	error = -1;
1935	else
1936	error = ep_loop_check_proc(ep_tovisit, depth + 1);
1937	if (error != 0)
1938	break;
1939	} else {
1940	/*
1941	* If we've reached a file that is not associated with
1942	* an ep, then we need to check if the newly added
1943	* links are going to add too many wakeup paths. We do
1944	* this by adding it to the tfile_check_list, if it's
1945	* not already there, and calling reverse_path_check()
1946	* during ep_insert().
1947	*/
1948	list_file(epi->ffd.file);
1949	}
1950	}
1951	mutex_unlock(&ep->mtx);
1952
1953	return error;
1954	}
1955
1956	/**
1957	* ep_loop_check - Performs a check to verify that adding an epoll file (@to)
1958	* into another epoll file (represented by @ep) does not create
1959	* closed loops or too deep chains.
1960	*
1961	* @ep: Pointer to the epoll we are inserting into.
1962	* @to: Pointer to the epoll to be inserted.
1963	*
1964	* Return: %zero if adding the epoll @to inside the epoll @from
1965	* does not violate the constraints, or %-1 otherwise.
1966	*/
1967	static int ep_loop_check(struct eventpoll *ep, struct eventpoll *to)
1968	{
1969	inserting_into = ep;
1970	return ep_loop_check_proc(to, 0);
1971	}
1972
1973	static void clear_tfile_check_list(void)
1974	{
1975	rcu_read_lock();
1976	while (tfile_check_list != EP_UNACTIVE_PTR) {
1977	struct epitems_head *head = tfile_check_list;
1978	tfile_check_list = head->next;
1979	unlist_file(head);
1980	}
1981	rcu_read_unlock();
1982	}
1983
1984	/*
1985	* Open an eventpoll file descriptor.
1986	*/
1987	static int do_epoll_create(int flags)
1988	{
1989	int error, fd;
1990	struct eventpoll *ep = NULL;
1991	struct file *file;
1992
1993	/* Check the EPOLL_* constant for consistency. */
1994	BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1995
1996	if (flags & ~EPOLL_CLOEXEC)
1997	return -EINVAL;
1998	/*
1999	* Create the internal data structure ("struct eventpoll").
2000	*/
2001	error = ep_alloc(&ep);
2002	if (error < 0)
2003	return error;
2004	/*
2005	* Creates all the items needed to setup an eventpoll file. That is,
2006	* a file structure and a free file descriptor.
2007	*/
2008	fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
2009	if (fd < 0) {
2010	error = fd;
2011	goto out_free_ep;
2012	}
2013	file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
2014	O_RDWR | (flags & O_CLOEXEC));
2015	if (IS_ERR(file)) {
2016	error = PTR_ERR(file);
2017	goto out_free_fd;
2018	}
2019	ep->file = file;
2020	fd_install(fd, file);
2021	return fd;
2022
2023	out_free_fd:
2024	put_unused_fd(fd);
2025	out_free_ep:
2026	ep_free(ep);
2027	return error;
2028	}
2029
2030	SYSCALL_DEFINE1(epoll_create1, int, flags)
2031	{
2032	return do_epoll_create(flags);
2033	}
2034
2035	SYSCALL_DEFINE1(epoll_create, int, size)
2036	{
2037	if (size <= 0)
2038	return -EINVAL;
2039
2040	return do_epoll_create(0);
2041	}
2042
2043	static inline int epoll_mutex_lock(struct mutex *mutex, int depth,
2044	bool nonblock)
2045	{
2046	if (!nonblock) {
2047	mutex_lock_nested(mutex, depth);
2048	return 0;
2049	}
2050	if (mutex_trylock(mutex))
2051	return 0;
2052	return -EAGAIN;
2053	}
2054
2055	int do_epoll_ctl(int epfd, int op, int fd, struct epoll_event *epds,
2056	bool nonblock)
2057	{
2058	int error;
2059	int full_check = 0;
2060	struct fd f, tf;
2061	struct eventpoll *ep;
2062	struct epitem *epi;
2063	struct eventpoll *tep = NULL;
2064
2065	error = -EBADF;
2066	f = fdget(epfd);
2067	if (!f.file)
2068	goto error_return;
2069
2070	/* Get the "struct file *" for the target file */
2071	tf = fdget(fd);
2072	if (!tf.file)
2073	goto error_fput;
2074
2075	/* The target file descriptor must support poll */
2076	error = -EPERM;
2077	if (!file_can_poll(tf.file))
2078	goto error_tgt_fput;
2079
2080	/* Check if EPOLLWAKEUP is allowed */
2081	if (ep_op_has_event(op))
2082	ep_take_care_of_epollwakeup(epds);
2083
2084	/*
2085	* We have to check that the file structure underneath the file descriptor
2086	* the user passed to us _is_ an eventpoll file. And also we do not permit
2087	* adding an epoll file descriptor inside itself.
2088	*/
2089	error = -EINVAL;
2090	if (f.file == tf.file || !is_file_epoll(f.file))
2091	goto error_tgt_fput;
2092
2093	/*
2094	* epoll adds to the wakeup queue at EPOLL_CTL_ADD time only,
2095	* so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation.
2096	* Also, we do not currently supported nested exclusive wakeups.
2097	*/
2098	if (ep_op_has_event(op) && (epds->events & EPOLLEXCLUSIVE)) {
2099	if (op == EPOLL_CTL_MOD)
2100	goto error_tgt_fput;
2101	if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) ||
2102	(epds->events & ~EPOLLEXCLUSIVE_OK_BITS)))
2103	goto error_tgt_fput;
2104	}
2105
2106	/*
2107	* At this point it is safe to assume that the "private_data" contains
2108	* our own data structure.
2109	*/
2110	ep = f.file->private_data;
2111
2112	/*
2113	* When we insert an epoll file descriptor inside another epoll file
2114	* descriptor, there is the chance of creating closed loops, which are
2115	* better be handled here, than in more critical paths. While we are
2116	* checking for loops we also determine the list of files reachable
2117	* and hang them on the tfile_check_list, so we can check that we
2118	* haven't created too many possible wakeup paths.
2119	*
2120	* We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
2121	* the epoll file descriptor is attaching directly to a wakeup source,
2122	* unless the epoll file descriptor is nested. The purpose of taking the
2123	* 'epmutex' on add is to prevent complex toplogies such as loops and
2124	* deep wakeup paths from forming in parallel through multiple
2125	* EPOLL_CTL_ADD operations.
2126	*/
2127	error = epoll_mutex_lock(&ep->mtx, 0, nonblock);
2128	if (error)
2129	goto error_tgt_fput;
2130	if (op == EPOLL_CTL_ADD) {
2131	if (READ_ONCE(f.file->f_ep) || ep->gen == loop_check_gen ||
2132	is_file_epoll(tf.file)) {
2133	mutex_unlock(&ep->mtx);
2134	error = epoll_mutex_lock(&epmutex, 0, nonblock);
2135	if (error)
2136	goto error_tgt_fput;
2137	loop_check_gen++;
2138	full_check = 1;
2139	if (is_file_epoll(tf.file)) {
2140	tep = tf.file->private_data;
2141	error = -ELOOP;
2142	if (ep_loop_check(ep, tep) != 0)
2143	goto error_tgt_fput;
2144	}
2145	error = epoll_mutex_lock(&ep->mtx, 0, nonblock);
2146	if (error)
2147	goto error_tgt_fput;
2148	}
2149	}
2150
2151	/*
2152	* Try to lookup the file inside our RB tree. Since we grabbed "mtx"
2153	* above, we can be sure to be able to use the item looked up by
2154	* ep_find() till we release the mutex.
2155	*/
2156	epi = ep_find(ep, tf.file, fd);
2157
2158	error = -EINVAL;
2159	switch (op) {
2160	case EPOLL_CTL_ADD:
2161	if (!epi) {
2162	epds->events |= EPOLLERR | EPOLLHUP;
2163	error = ep_insert(ep, epds, tf.file, fd, full_check);
2164	} else
2165	error = -EEXIST;
2166	break;
2167	case EPOLL_CTL_DEL:
2168	if (epi)
2169	error = ep_remove(ep, epi);
2170	else
2171	error = -ENOENT;
2172	break;
2173	case EPOLL_CTL_MOD:
2174	if (epi) {
2175	if (!(epi->event.events & EPOLLEXCLUSIVE)) {
2176	epds->events |= EPOLLERR | EPOLLHUP;
2177	error = ep_modify(ep, epi, epds);
2178	}
2179	} else
2180	error = -ENOENT;
2181	break;
2182	}
2183	mutex_unlock(&ep->mtx);
2184
2185	error_tgt_fput:
2186	if (full_check) {
2187	clear_tfile_check_list();
2188	loop_check_gen++;
2189	mutex_unlock(&epmutex);
2190	}
2191
2192	fdput(tf);
2193	error_fput:
2194	fdput(f);
2195	error_return:
2196
2197	return error;
2198	}
2199
2200	/*
2201	* The following function implements the controller interface for
2202	* the eventpoll file that enables the insertion/removal/change of
2203	* file descriptors inside the interest set.
2204	*/
2205	SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
2206	struct epoll_event __user *, event)
2207	{
2208	struct epoll_event epds;
2209
2210	if (ep_op_has_event(op) &&
2211	copy_from_user(&epds, event, sizeof(struct epoll_event)))
2212	return -EFAULT;
2213
2214	return do_epoll_ctl(epfd, op, fd, &epds, false);
2215	}
2216
2217	/*
2218	* Implement the event wait interface for the eventpoll file. It is the kernel
2219	* part of the user space epoll_wait(2).
2220	*/
2221	static int do_epoll_wait(int epfd, struct epoll_event __user *events,
2222	int maxevents, struct timespec64 *to)
2223	{
2224	int error;
2225	struct fd f;
2226	struct eventpoll *ep;
2227
2228	/* The maximum number of event must be greater than zero */
2229	if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
2230	return -EINVAL;
2231
2232	/* Verify that the area passed by the user is writeable */
2233	if (!access_ok(events, maxevents * sizeof(struct epoll_event)))
2234	return -EFAULT;
2235
2236	/* Get the "struct file *" for the eventpoll file */
2237	f = fdget(epfd);
2238	if (!f.file)
2239	return -EBADF;
2240
2241	/*
2242	* We have to check that the file structure underneath the fd
2243	* the user passed to us _is_ an eventpoll file.
2244	*/
2245	error = -EINVAL;
2246	if (!is_file_epoll(f.file))
2247	goto error_fput;
2248
2249	/*
2250	* At this point it is safe to assume that the "private_data" contains
2251	* our own data structure.
2252	*/
2253	ep = f.file->private_data;
2254
2255	/* Time to fish for events ... */
2256	error = ep_poll(ep, events, maxevents, to);
2257
2258	error_fput:
2259	fdput(f);
2260	return error;
2261	}
2262
2263	SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
2264	int, maxevents, int, timeout)
2265	{
2266	struct timespec64 to;
2267
2268	return do_epoll_wait(epfd, events, maxevents,
2269	ep_timeout_to_timespec(&to, timeout));
2270	}
2271
2272	/*
2273	* Implement the event wait interface for the eventpoll file. It is the kernel
2274	* part of the user space epoll_pwait(2).
2275	*/
2276	static int do_epoll_pwait(int epfd, struct epoll_event __user *events,
2277	int maxevents, struct timespec64 *to,
2278	const sigset_t __user *sigmask, size_t sigsetsize)
2279	{
2280	int error;
2281
2282	/*
2283	* If the caller wants a certain signal mask to be set during the wait,
2284	* we apply it here.
2285	*/
2286	error = set_user_sigmask(sigmask, sigsetsize);
2287	if (error)
2288	return error;
2289
2290	error = do_epoll_wait(epfd, events, maxevents, to);
2291
2292	restore_saved_sigmask_unless(error == -EINTR);
2293
2294	return error;
2295	}
2296
2297	SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2298	int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2299	size_t, sigsetsize)
2300	{
2301	struct timespec64 to;
2302
2303	return do_epoll_pwait(epfd, events, maxevents,
2304	ep_timeout_to_timespec(&to, timeout),
2305	sigmask, sigsetsize);
2306	}
2307
2308	SYSCALL_DEFINE6(epoll_pwait2, int, epfd, struct epoll_event __user *, events,
2309	int, maxevents, const struct __kernel_timespec __user *, timeout,
2310	const sigset_t __user *, sigmask, size_t, sigsetsize)
2311	{
2312	struct timespec64 ts, *to = NULL;
2313
2314	if (timeout) {
2315	if (get_timespec64(&ts, timeout))
2316	return -EFAULT;
2317	to = &ts;
2318	if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
2319	return -EINVAL;
2320	}
2321
2322	return do_epoll_pwait(epfd, events, maxevents, to,
2323	sigmask, sigsetsize);
2324	}
2325
2326	#ifdef CONFIG_COMPAT
2327	static int do_compat_epoll_pwait(int epfd, struct epoll_event __user *events,
2328	int maxevents, struct timespec64 *timeout,
2329	const compat_sigset_t __user *sigmask,
2330	compat_size_t sigsetsize)
2331	{
2332	long err;
2333
2334	/*
2335	* If the caller wants a certain signal mask to be set during the wait,
2336	* we apply it here.
2337	*/
2338	err = set_compat_user_sigmask(sigmask, sigsetsize);
2339	if (err)
2340	return err;
2341
2342	err = do_epoll_wait(epfd, events, maxevents, timeout);
2343
2344	restore_saved_sigmask_unless(err == -EINTR);
2345
2346	return err;
2347	}
2348
2349	COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2350	struct epoll_event __user *, events,
2351	int, maxevents, int, timeout,
2352	const compat_sigset_t __user *, sigmask,
2353	compat_size_t, sigsetsize)
2354	{
2355	struct timespec64 to;
2356
2357	return do_compat_epoll_pwait(epfd, events, maxevents,
2358	ep_timeout_to_timespec(&to, timeout),
2359	sigmask, sigsetsize);
2360	}
2361
2362	COMPAT_SYSCALL_DEFINE6(epoll_pwait2, int, epfd,
2363	struct epoll_event __user *, events,
2364	int, maxevents,
2365	const struct __kernel_timespec __user *, timeout,
2366	const compat_sigset_t __user *, sigmask,
2367	compat_size_t, sigsetsize)
2368	{
2369	struct timespec64 ts, *to = NULL;
2370
2371	if (timeout) {
2372	if (get_timespec64(&ts, timeout))
2373	return -EFAULT;
2374	to = &ts;
2375	if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
2376	return -EINVAL;
2377	}
2378
2379	return do_compat_epoll_pwait(epfd, events, maxevents, to,
2380	sigmask, sigsetsize);
2381	}
2382
2383	#endif
2384
2385	static int __init eventpoll_init(void)
2386	{
2387	struct sysinfo si;
2388
2389	si_meminfo(&si);
2390	/*
2391	* Allows top 4% of lomem to be allocated for epoll watches (per user).
2392	*/
2393	max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2394	EP_ITEM_COST;
2395	BUG_ON(max_user_watches < 0);
2396
2397	/*
2398	* We can have many thousands of epitems, so prevent this from
2399	* using an extra cache line on 64-bit (and smaller) CPUs
2400	*/
2401	BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2402
2403	/* Allocates slab cache used to allocate "struct epitem" items */
2404	epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
2405	0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
2406
2407	/* Allocates slab cache used to allocate "struct eppoll_entry" */
2408	pwq_cache = kmem_cache_create("eventpoll_pwq",
2409	sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL);
2410	epoll_sysctls_init();
2411
2412	ephead_cache = kmem_cache_create("ep_head",
2413	sizeof(struct epitems_head), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL);
2414
2415	return 0;
2416	}
2417	fs_initcall(eventpoll_init);
2418