2 * fs/eventpoll.c (Efficient event retrieval implementation)
3 * Copyright (C) 2001,...,2009 Davide Libenzi
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.
10 * Davide Libenzi <davidel@xmailserver.org>
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/sched.h>
18 #include <linux/file.h>
19 #include <linux/signal.h>
20 #include <linux/errno.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 <asm/uaccess.h>
37 #include <asm/system.h>
40 #include <linux/atomic.h>
44 * There are three level of locking required by epoll :
48 * 3) ep->lock (spinlock)
50 * The acquire order is the one listed above, from 1 to 3.
51 * We need a spinlock (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()
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
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
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.
90 /* Epoll private bits inside the event mask */
91 #define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET)
93 /* Maximum number of nesting allowed inside epoll sets */
94 #define EP_MAX_NESTS 4
96 #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
98 #define EP_UNACTIVE_PTR ((void *) -1L)
100 #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
102 struct epoll_filefd
{
108 * Structure used to track possible nested calls, for too deep recursions
111 struct nested_call_node
{
112 struct list_head llink
;
118 * This structure is used as collector for nested calls, to check for
119 * maximum recursion dept and loop cycles.
121 struct nested_calls
{
122 struct list_head tasks_call_list
;
127 * Each file descriptor added to the eventpoll interface will
128 * have an entry of this type linked to the "rbr" RB tree.
131 /* RB tree node used to link this structure to the eventpoll RB tree */
134 /* List header used to link this structure to the eventpoll ready list */
135 struct list_head rdllink
;
138 * Works together "struct eventpoll"->ovflist in keeping the
139 * single linked chain of items.
143 /* The file descriptor information this item refers to */
144 struct epoll_filefd ffd
;
146 /* Number of active wait queue attached to poll operations */
149 /* List containing poll wait queues */
150 struct list_head pwqlist
;
152 /* The "container" of this item */
153 struct eventpoll
*ep
;
155 /* List header used to link this item to the "struct file" items list */
156 struct list_head fllink
;
158 /* The structure that describe the interested events and the source fd */
159 struct epoll_event event
;
163 * This structure is stored inside the "private_data" member of the file
164 * structure and represents the main data structure for the eventpoll
168 /* Protect the access to this structure */
172 * This mutex is used to ensure that files are not removed
173 * while epoll is using them. This is held during the event
174 * collection loop, the file cleanup path, the epoll file exit
175 * code and the ctl operations.
179 /* Wait queue used by sys_epoll_wait() */
180 wait_queue_head_t wq
;
182 /* Wait queue used by file->poll() */
183 wait_queue_head_t poll_wait
;
185 /* List of ready file descriptors */
186 struct list_head rdllist
;
188 /* RB tree root used to store monitored fd structs */
192 * This is a single linked list that chains all the "struct epitem" that
193 * happened while transferring ready events to userspace w/out
196 struct epitem
*ovflist
;
198 /* The user that created the eventpoll descriptor */
199 struct user_struct
*user
;
203 /* used to optimize loop detection check */
205 struct list_head visited_list_link
;
208 /* Wait structure used by the poll hooks */
209 struct eppoll_entry
{
210 /* List header used to link this structure to the "struct epitem" */
211 struct list_head llink
;
213 /* The "base" pointer is set to the container "struct epitem" */
217 * Wait queue item that will be linked to the target file wait
222 /* The wait queue head that linked the "wait" wait queue item */
223 wait_queue_head_t
*whead
;
226 /* Wrapper struct used by poll queueing */
232 /* Used by the ep_send_events() function as callback private data */
233 struct ep_send_events_data
{
235 struct epoll_event __user
*events
;
239 * Configuration options available inside /proc/sys/fs/epoll/
241 /* Maximum number of epoll watched descriptors, per user */
242 static long max_user_watches __read_mostly
;
245 * This mutex is used to serialize ep_free() and eventpoll_release_file().
247 static DEFINE_MUTEX(epmutex
);
249 /* Used to check for epoll file descriptor inclusion loops */
250 static struct nested_calls poll_loop_ncalls
;
252 /* Used for safe wake up implementation */
253 static struct nested_calls poll_safewake_ncalls
;
255 /* Used to call file's f_op->poll() under the nested calls boundaries */
256 static struct nested_calls poll_readywalk_ncalls
;
258 /* Slab cache used to allocate "struct epitem" */
259 static struct kmem_cache
*epi_cache __read_mostly
;
261 /* Slab cache used to allocate "struct eppoll_entry" */
262 static struct kmem_cache
*pwq_cache __read_mostly
;
264 /* Visited nodes during ep_loop_check(), so we can unset them when we finish */
265 static LIST_HEAD(visited_list
);
268 * List of files with newly added links, where we may need to limit the number
269 * of emanating paths. Protected by the epmutex.
271 static LIST_HEAD(tfile_check_list
);
275 #include <linux/sysctl.h>
278 static long long_max
= LONG_MAX
;
280 ctl_table epoll_table
[] = {
282 .procname
= "max_user_watches",
283 .data
= &max_user_watches
,
284 .maxlen
= sizeof(max_user_watches
),
286 .proc_handler
= proc_doulongvec_minmax
,
292 #endif /* CONFIG_SYSCTL */
294 static const struct file_operations eventpoll_fops
;
296 static inline int is_file_epoll(struct file
*f
)
298 return f
->f_op
== &eventpoll_fops
;
301 /* Setup the structure that is used as key for the RB tree */
302 static inline void ep_set_ffd(struct epoll_filefd
*ffd
,
303 struct file
*file
, int fd
)
309 /* Compare RB tree keys */
310 static inline int ep_cmp_ffd(struct epoll_filefd
*p1
,
311 struct epoll_filefd
*p2
)
313 return (p1
->file
> p2
->file
? +1:
314 (p1
->file
< p2
->file
? -1 : p1
->fd
- p2
->fd
));
317 /* Tells us if the item is currently linked */
318 static inline int ep_is_linked(struct list_head
*p
)
320 return !list_empty(p
);
323 static inline struct eppoll_entry
*ep_pwq_from_wait(wait_queue_t
*p
)
325 return container_of(p
, struct eppoll_entry
, wait
);
328 /* Get the "struct epitem" from a wait queue pointer */
329 static inline struct epitem
*ep_item_from_wait(wait_queue_t
*p
)
331 return container_of(p
, struct eppoll_entry
, wait
)->base
;
334 /* Get the "struct epitem" from an epoll queue wrapper */
335 static inline struct epitem
*ep_item_from_epqueue(poll_table
*p
)
337 return container_of(p
, struct ep_pqueue
, pt
)->epi
;
340 /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
341 static inline int ep_op_has_event(int op
)
343 return op
!= EPOLL_CTL_DEL
;
346 /* Initialize the poll safe wake up structure */
347 static void ep_nested_calls_init(struct nested_calls
*ncalls
)
349 INIT_LIST_HEAD(&ncalls
->tasks_call_list
);
350 spin_lock_init(&ncalls
->lock
);
354 * ep_events_available - Checks if ready events might be available.
356 * @ep: Pointer to the eventpoll context.
358 * Returns: Returns a value different than zero if ready events are available,
361 static inline int ep_events_available(struct eventpoll
*ep
)
363 return !list_empty(&ep
->rdllist
) || ep
->ovflist
!= EP_UNACTIVE_PTR
;
367 * ep_call_nested - Perform a bound (possibly) nested call, by checking
368 * that the recursion limit is not exceeded, and that
369 * the same nested call (by the meaning of same cookie) is
372 * @ncalls: Pointer to the nested_calls structure to be used for this call.
373 * @max_nests: Maximum number of allowed nesting calls.
374 * @nproc: Nested call core function pointer.
375 * @priv: Opaque data to be passed to the @nproc callback.
376 * @cookie: Cookie to be used to identify this nested call.
377 * @ctx: This instance context.
379 * Returns: Returns the code returned by the @nproc callback, or -1 if
380 * the maximum recursion limit has been exceeded.
382 static int ep_call_nested(struct nested_calls
*ncalls
, int max_nests
,
383 int (*nproc
)(void *, void *, int), void *priv
,
384 void *cookie
, void *ctx
)
386 int error
, call_nests
= 0;
388 struct list_head
*lsthead
= &ncalls
->tasks_call_list
;
389 struct nested_call_node
*tncur
;
390 struct nested_call_node tnode
;
392 spin_lock_irqsave(&ncalls
->lock
, flags
);
395 * Try to see if the current task is already inside this wakeup call.
396 * We use a list here, since the population inside this set is always
399 list_for_each_entry(tncur
, lsthead
, llink
) {
400 if (tncur
->ctx
== ctx
&&
401 (tncur
->cookie
== cookie
|| ++call_nests
> max_nests
)) {
403 * Ops ... loop detected or maximum nest level reached.
404 * We abort this wake by breaking the cycle itself.
411 /* Add the current task and cookie to the list */
413 tnode
.cookie
= cookie
;
414 list_add(&tnode
.llink
, lsthead
);
416 spin_unlock_irqrestore(&ncalls
->lock
, flags
);
418 /* Call the nested function */
419 error
= (*nproc
)(priv
, cookie
, call_nests
);
421 /* Remove the current task from the list */
422 spin_lock_irqsave(&ncalls
->lock
, flags
);
423 list_del(&tnode
.llink
);
425 spin_unlock_irqrestore(&ncalls
->lock
, flags
);
430 #ifdef CONFIG_DEBUG_LOCK_ALLOC
431 static inline void ep_wake_up_nested(wait_queue_head_t
*wqueue
,
432 unsigned long events
, int subclass
)
436 spin_lock_irqsave_nested(&wqueue
->lock
, flags
, subclass
);
437 wake_up_locked_poll(wqueue
, events
);
438 spin_unlock_irqrestore(&wqueue
->lock
, flags
);
441 static inline void ep_wake_up_nested(wait_queue_head_t
*wqueue
,
442 unsigned long events
, int subclass
)
444 wake_up_poll(wqueue
, events
);
448 static int ep_poll_wakeup_proc(void *priv
, void *cookie
, int call_nests
)
450 ep_wake_up_nested((wait_queue_head_t
*) cookie
, POLLIN
,
456 * Perform a safe wake up of the poll wait list. The problem is that
457 * with the new callback'd wake up system, it is possible that the
458 * poll callback is reentered from inside the call to wake_up() done
459 * on the poll wait queue head. The rule is that we cannot reenter the
460 * wake up code from the same task more than EP_MAX_NESTS times,
461 * and we cannot reenter the same wait queue head at all. This will
462 * enable to have a hierarchy of epoll file descriptor of no more than
465 static void ep_poll_safewake(wait_queue_head_t
*wq
)
467 int this_cpu
= get_cpu();
469 ep_call_nested(&poll_safewake_ncalls
, EP_MAX_NESTS
,
470 ep_poll_wakeup_proc
, NULL
, wq
, (void *) (long) this_cpu
);
475 static void ep_remove_wait_queue(struct eppoll_entry
*pwq
)
477 wait_queue_head_t
*whead
;
480 /* If it is cleared by POLLFREE, it should be rcu-safe */
481 whead
= rcu_dereference(pwq
->whead
);
483 remove_wait_queue(whead
, &pwq
->wait
);
488 * This function unregisters poll callbacks from the associated file
489 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
492 static void ep_unregister_pollwait(struct eventpoll
*ep
, struct epitem
*epi
)
494 struct list_head
*lsthead
= &epi
->pwqlist
;
495 struct eppoll_entry
*pwq
;
497 while (!list_empty(lsthead
)) {
498 pwq
= list_first_entry(lsthead
, struct eppoll_entry
, llink
);
500 list_del(&pwq
->llink
);
501 ep_remove_wait_queue(pwq
);
502 kmem_cache_free(pwq_cache
, pwq
);
507 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
508 * the scan code, to call f_op->poll(). Also allows for
509 * O(NumReady) performance.
511 * @ep: Pointer to the epoll private data structure.
512 * @sproc: Pointer to the scan callback.
513 * @priv: Private opaque data passed to the @sproc callback.
514 * @depth: The current depth of recursive f_op->poll calls.
516 * Returns: The same integer error code returned by the @sproc callback.
518 static int ep_scan_ready_list(struct eventpoll
*ep
,
519 int (*sproc
)(struct eventpoll
*,
520 struct list_head
*, void *),
524 int error
, pwake
= 0;
526 struct epitem
*epi
, *nepi
;
530 * We need to lock this because we could be hit by
531 * eventpoll_release_file() and epoll_ctl().
533 mutex_lock_nested(&ep
->mtx
, depth
);
536 * Steal the ready list, and re-init the original one to the
537 * empty list. Also, set ep->ovflist to NULL so that events
538 * happening while looping w/out locks, are not lost. We cannot
539 * have the poll callback to queue directly on ep->rdllist,
540 * because we want the "sproc" callback to be able to do it
543 spin_lock_irqsave(&ep
->lock
, flags
);
544 list_splice_init(&ep
->rdllist
, &txlist
);
546 spin_unlock_irqrestore(&ep
->lock
, flags
);
549 * Now call the callback function.
551 error
= (*sproc
)(ep
, &txlist
, priv
);
553 spin_lock_irqsave(&ep
->lock
, flags
);
555 * During the time we spent inside the "sproc" callback, some
556 * other events might have been queued by the poll callback.
557 * We re-insert them inside the main ready-list here.
559 for (nepi
= ep
->ovflist
; (epi
= nepi
) != NULL
;
560 nepi
= epi
->next
, epi
->next
= EP_UNACTIVE_PTR
) {
562 * We need to check if the item is already in the list.
563 * During the "sproc" callback execution time, items are
564 * queued into ->ovflist but the "txlist" might already
565 * contain them, and the list_splice() below takes care of them.
567 if (!ep_is_linked(&epi
->rdllink
))
568 list_add_tail(&epi
->rdllink
, &ep
->rdllist
);
571 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
572 * releasing the lock, events will be queued in the normal way inside
575 ep
->ovflist
= EP_UNACTIVE_PTR
;
578 * Quickly re-inject items left on "txlist".
580 list_splice(&txlist
, &ep
->rdllist
);
582 if (!list_empty(&ep
->rdllist
)) {
584 * Wake up (if active) both the eventpoll wait list and
585 * the ->poll() wait list (delayed after we release the lock).
587 if (waitqueue_active(&ep
->wq
))
588 wake_up_locked(&ep
->wq
);
589 if (waitqueue_active(&ep
->poll_wait
))
592 spin_unlock_irqrestore(&ep
->lock
, flags
);
594 mutex_unlock(&ep
->mtx
);
596 /* We have to call this outside the lock */
598 ep_poll_safewake(&ep
->poll_wait
);
604 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
605 * all the associated resources. Must be called with "mtx" held.
607 static int ep_remove(struct eventpoll
*ep
, struct epitem
*epi
)
610 struct file
*file
= epi
->ffd
.file
;
613 * Removes poll wait queue hooks. We _have_ to do this without holding
614 * the "ep->lock" otherwise a deadlock might occur. This because of the
615 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
616 * queue head lock when unregistering the wait queue. The wakeup callback
617 * will run by holding the wait queue head lock and will call our callback
618 * that will try to get "ep->lock".
620 ep_unregister_pollwait(ep
, epi
);
622 /* Remove the current item from the list of epoll hooks */
623 spin_lock(&file
->f_lock
);
624 if (ep_is_linked(&epi
->fllink
))
625 list_del_init(&epi
->fllink
);
626 spin_unlock(&file
->f_lock
);
628 rb_erase(&epi
->rbn
, &ep
->rbr
);
630 spin_lock_irqsave(&ep
->lock
, flags
);
631 if (ep_is_linked(&epi
->rdllink
))
632 list_del_init(&epi
->rdllink
);
633 spin_unlock_irqrestore(&ep
->lock
, flags
);
635 /* At this point it is safe to free the eventpoll item */
636 kmem_cache_free(epi_cache
, epi
);
638 atomic_long_dec(&ep
->user
->epoll_watches
);
643 static void ep_free(struct eventpoll
*ep
)
648 /* We need to release all tasks waiting for these file */
649 if (waitqueue_active(&ep
->poll_wait
))
650 ep_poll_safewake(&ep
->poll_wait
);
653 * We need to lock this because we could be hit by
654 * eventpoll_release_file() while we're freeing the "struct eventpoll".
655 * We do not need to hold "ep->mtx" here because the epoll file
656 * is on the way to be removed and no one has references to it
657 * anymore. The only hit might come from eventpoll_release_file() but
658 * holding "epmutex" is sufficient here.
660 mutex_lock(&epmutex
);
663 * Walks through the whole tree by unregistering poll callbacks.
665 for (rbp
= rb_first(&ep
->rbr
); rbp
; rbp
= rb_next(rbp
)) {
666 epi
= rb_entry(rbp
, struct epitem
, rbn
);
668 ep_unregister_pollwait(ep
, epi
);
672 * Walks through the whole tree by freeing each "struct epitem". At this
673 * point we are sure no poll callbacks will be lingering around, and also by
674 * holding "epmutex" we can be sure that no file cleanup code will hit
675 * us during this operation. So we can avoid the lock on "ep->lock".
677 while ((rbp
= rb_first(&ep
->rbr
)) != NULL
) {
678 epi
= rb_entry(rbp
, struct epitem
, rbn
);
682 mutex_unlock(&epmutex
);
683 mutex_destroy(&ep
->mtx
);
688 static int ep_eventpoll_release(struct inode
*inode
, struct file
*file
)
690 struct eventpoll
*ep
= file
->private_data
;
698 static int ep_read_events_proc(struct eventpoll
*ep
, struct list_head
*head
,
701 struct epitem
*epi
, *tmp
;
704 init_poll_funcptr(&pt
, NULL
);
705 list_for_each_entry_safe(epi
, tmp
, head
, rdllink
) {
706 pt
._key
= epi
->event
.events
;
707 if (epi
->ffd
.file
->f_op
->poll(epi
->ffd
.file
, &pt
) &
709 return POLLIN
| POLLRDNORM
;
712 * Item has been dropped into the ready list by the poll
713 * callback, but it's not actually ready, as far as
714 * caller requested events goes. We can remove it here.
716 list_del_init(&epi
->rdllink
);
723 static int ep_poll_readyevents_proc(void *priv
, void *cookie
, int call_nests
)
725 return ep_scan_ready_list(priv
, ep_read_events_proc
, NULL
, call_nests
+ 1);
728 static unsigned int ep_eventpoll_poll(struct file
*file
, poll_table
*wait
)
731 struct eventpoll
*ep
= file
->private_data
;
733 /* Insert inside our poll wait queue */
734 poll_wait(file
, &ep
->poll_wait
, wait
);
737 * Proceed to find out if wanted events are really available inside
738 * the ready list. This need to be done under ep_call_nested()
739 * supervision, since the call to f_op->poll() done on listed files
740 * could re-enter here.
742 pollflags
= ep_call_nested(&poll_readywalk_ncalls
, EP_MAX_NESTS
,
743 ep_poll_readyevents_proc
, ep
, ep
, current
);
745 return pollflags
!= -1 ? pollflags
: 0;
748 /* File callbacks that implement the eventpoll file behaviour */
749 static const struct file_operations eventpoll_fops
= {
750 .release
= ep_eventpoll_release
,
751 .poll
= ep_eventpoll_poll
,
752 .llseek
= noop_llseek
,
756 * This is called from eventpoll_release() to unlink files from the eventpoll
757 * interface. We need to have this facility to cleanup correctly files that are
758 * closed without being removed from the eventpoll interface.
760 void eventpoll_release_file(struct file
*file
)
762 struct list_head
*lsthead
= &file
->f_ep_links
;
763 struct eventpoll
*ep
;
767 * We don't want to get "file->f_lock" because it is not
768 * necessary. It is not necessary because we're in the "struct file"
769 * cleanup path, and this means that no one is using this file anymore.
770 * So, for example, epoll_ctl() cannot hit here since if we reach this
771 * point, the file counter already went to zero and fget() would fail.
772 * The only hit might come from ep_free() but by holding the mutex
773 * will correctly serialize the operation. We do need to acquire
774 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
775 * from anywhere but ep_free().
777 * Besides, ep_remove() acquires the lock, so we can't hold it here.
779 mutex_lock(&epmutex
);
781 while (!list_empty(lsthead
)) {
782 epi
= list_first_entry(lsthead
, struct epitem
, fllink
);
785 list_del_init(&epi
->fllink
);
786 mutex_lock_nested(&ep
->mtx
, 0);
788 mutex_unlock(&ep
->mtx
);
791 mutex_unlock(&epmutex
);
794 static int ep_alloc(struct eventpoll
**pep
)
797 struct user_struct
*user
;
798 struct eventpoll
*ep
;
800 user
= get_current_user();
802 ep
= kzalloc(sizeof(*ep
), GFP_KERNEL
);
806 spin_lock_init(&ep
->lock
);
807 mutex_init(&ep
->mtx
);
808 init_waitqueue_head(&ep
->wq
);
809 init_waitqueue_head(&ep
->poll_wait
);
810 INIT_LIST_HEAD(&ep
->rdllist
);
812 ep
->ovflist
= EP_UNACTIVE_PTR
;
825 * Search the file inside the eventpoll tree. The RB tree operations
826 * are protected by the "mtx" mutex, and ep_find() must be called with
829 static struct epitem
*ep_find(struct eventpoll
*ep
, struct file
*file
, int fd
)
833 struct epitem
*epi
, *epir
= NULL
;
834 struct epoll_filefd ffd
;
836 ep_set_ffd(&ffd
, file
, fd
);
837 for (rbp
= ep
->rbr
.rb_node
; rbp
; ) {
838 epi
= rb_entry(rbp
, struct epitem
, rbn
);
839 kcmp
= ep_cmp_ffd(&ffd
, &epi
->ffd
);
854 * This is the callback that is passed to the wait queue wakeup
855 * mechanism. It is called by the stored file descriptors when they
856 * have events to report.
858 static int ep_poll_callback(wait_queue_t
*wait
, unsigned mode
, int sync
, void *key
)
862 struct epitem
*epi
= ep_item_from_wait(wait
);
863 struct eventpoll
*ep
= epi
->ep
;
865 if ((unsigned long)key
& POLLFREE
) {
866 ep_pwq_from_wait(wait
)->whead
= NULL
;
868 * whead = NULL above can race with ep_remove_wait_queue()
869 * which can do another remove_wait_queue() after us, so we
870 * can't use __remove_wait_queue(). whead->lock is held by
873 list_del_init(&wait
->task_list
);
876 spin_lock_irqsave(&ep
->lock
, flags
);
879 * If the event mask does not contain any poll(2) event, we consider the
880 * descriptor to be disabled. This condition is likely the effect of the
881 * EPOLLONESHOT bit that disables the descriptor when an event is received,
882 * until the next EPOLL_CTL_MOD will be issued.
884 if (!(epi
->event
.events
& ~EP_PRIVATE_BITS
))
888 * Check the events coming with the callback. At this stage, not
889 * every device reports the events in the "key" parameter of the
890 * callback. We need to be able to handle both cases here, hence the
891 * test for "key" != NULL before the event match test.
893 if (key
&& !((unsigned long) key
& epi
->event
.events
))
897 * If we are transferring events to userspace, we can hold no locks
898 * (because we're accessing user memory, and because of linux f_op->poll()
899 * semantics). All the events that happen during that period of time are
900 * chained in ep->ovflist and requeued later on.
902 if (unlikely(ep
->ovflist
!= EP_UNACTIVE_PTR
)) {
903 if (epi
->next
== EP_UNACTIVE_PTR
) {
904 epi
->next
= ep
->ovflist
;
910 /* If this file is already in the ready list we exit soon */
911 if (!ep_is_linked(&epi
->rdllink
))
912 list_add_tail(&epi
->rdllink
, &ep
->rdllist
);
915 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
918 if (waitqueue_active(&ep
->wq
))
919 wake_up_locked(&ep
->wq
);
920 if (waitqueue_active(&ep
->poll_wait
))
924 spin_unlock_irqrestore(&ep
->lock
, flags
);
926 /* We have to call this outside the lock */
928 ep_poll_safewake(&ep
->poll_wait
);
934 * This is the callback that is used to add our wait queue to the
935 * target file wakeup lists.
937 static void ep_ptable_queue_proc(struct file
*file
, wait_queue_head_t
*whead
,
940 struct epitem
*epi
= ep_item_from_epqueue(pt
);
941 struct eppoll_entry
*pwq
;
943 if (epi
->nwait
>= 0 && (pwq
= kmem_cache_alloc(pwq_cache
, GFP_KERNEL
))) {
944 init_waitqueue_func_entry(&pwq
->wait
, ep_poll_callback
);
947 add_wait_queue(whead
, &pwq
->wait
);
948 list_add_tail(&pwq
->llink
, &epi
->pwqlist
);
951 /* We have to signal that an error occurred */
956 static void ep_rbtree_insert(struct eventpoll
*ep
, struct epitem
*epi
)
959 struct rb_node
**p
= &ep
->rbr
.rb_node
, *parent
= NULL
;
964 epic
= rb_entry(parent
, struct epitem
, rbn
);
965 kcmp
= ep_cmp_ffd(&epi
->ffd
, &epic
->ffd
);
967 p
= &parent
->rb_right
;
969 p
= &parent
->rb_left
;
971 rb_link_node(&epi
->rbn
, parent
, p
);
972 rb_insert_color(&epi
->rbn
, &ep
->rbr
);
977 #define PATH_ARR_SIZE 5
979 * These are the number paths of length 1 to 5, that we are allowing to emanate
980 * from a single file of interest. For example, we allow 1000 paths of length
981 * 1, to emanate from each file of interest. This essentially represents the
982 * potential wakeup paths, which need to be limited in order to avoid massive
983 * uncontrolled wakeup storms. The common use case should be a single ep which
984 * is connected to n file sources. In this case each file source has 1 path
985 * of length 1. Thus, the numbers below should be more than sufficient. These
986 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
987 * and delete can't add additional paths. Protected by the epmutex.
989 static const int path_limits
[PATH_ARR_SIZE
] = { 1000, 500, 100, 50, 10 };
990 static int path_count
[PATH_ARR_SIZE
];
992 static int path_count_inc(int nests
)
994 /* Allow an arbitrary number of depth 1 paths */
998 if (++path_count
[nests
] > path_limits
[nests
])
1003 static void path_count_init(void)
1007 for (i
= 0; i
< PATH_ARR_SIZE
; i
++)
1011 static int reverse_path_check_proc(void *priv
, void *cookie
, int call_nests
)
1014 struct file
*file
= priv
;
1015 struct file
*child_file
;
1018 list_for_each_entry(epi
, &file
->f_ep_links
, fllink
) {
1019 child_file
= epi
->ep
->file
;
1020 if (is_file_epoll(child_file
)) {
1021 if (list_empty(&child_file
->f_ep_links
)) {
1022 if (path_count_inc(call_nests
)) {
1027 error
= ep_call_nested(&poll_loop_ncalls
,
1029 reverse_path_check_proc
,
1030 child_file
, child_file
,
1036 printk(KERN_ERR
"reverse_path_check_proc: "
1037 "file is not an ep!\n");
1044 * reverse_path_check - The tfile_check_list is list of file *, which have
1045 * links that are proposed to be newly added. We need to
1046 * make sure that those added links don't add too many
1047 * paths such that we will spend all our time waking up
1048 * eventpoll objects.
1050 * Returns: Returns zero if the proposed links don't create too many paths,
1053 static int reverse_path_check(void)
1057 struct file
*current_file
;
1059 /* let's call this for all tfiles */
1060 list_for_each_entry(current_file
, &tfile_check_list
, f_tfile_llink
) {
1063 error
= ep_call_nested(&poll_loop_ncalls
, EP_MAX_NESTS
,
1064 reverse_path_check_proc
, current_file
,
1065 current_file
, current
);
1073 * Must be called with "mtx" held.
1075 static int ep_insert(struct eventpoll
*ep
, struct epoll_event
*event
,
1076 struct file
*tfile
, int fd
)
1078 int error
, revents
, pwake
= 0;
1079 unsigned long flags
;
1082 struct ep_pqueue epq
;
1084 user_watches
= atomic_long_read(&ep
->user
->epoll_watches
);
1085 if (unlikely(user_watches
>= max_user_watches
))
1087 if (!(epi
= kmem_cache_alloc(epi_cache
, GFP_KERNEL
)))
1090 /* Item initialization follow here ... */
1091 INIT_LIST_HEAD(&epi
->rdllink
);
1092 INIT_LIST_HEAD(&epi
->fllink
);
1093 INIT_LIST_HEAD(&epi
->pwqlist
);
1095 ep_set_ffd(&epi
->ffd
, tfile
, fd
);
1096 epi
->event
= *event
;
1098 epi
->next
= EP_UNACTIVE_PTR
;
1100 /* Initialize the poll table using the queue callback */
1102 init_poll_funcptr(&epq
.pt
, ep_ptable_queue_proc
);
1103 epq
.pt
._key
= event
->events
;
1106 * Attach the item to the poll hooks and get current event bits.
1107 * We can safely use the file* here because its usage count has
1108 * been increased by the caller of this function. Note that after
1109 * this operation completes, the poll callback can start hitting
1112 revents
= tfile
->f_op
->poll(tfile
, &epq
.pt
);
1115 * We have to check if something went wrong during the poll wait queue
1116 * install process. Namely an allocation for a wait queue failed due
1117 * high memory pressure.
1121 goto error_unregister
;
1123 /* Add the current item to the list of active epoll hook for this file */
1124 spin_lock(&tfile
->f_lock
);
1125 list_add_tail(&epi
->fllink
, &tfile
->f_ep_links
);
1126 spin_unlock(&tfile
->f_lock
);
1129 * Add the current item to the RB tree. All RB tree operations are
1130 * protected by "mtx", and ep_insert() is called with "mtx" held.
1132 ep_rbtree_insert(ep
, epi
);
1134 /* now check if we've created too many backpaths */
1136 if (reverse_path_check())
1137 goto error_remove_epi
;
1139 /* We have to drop the new item inside our item list to keep track of it */
1140 spin_lock_irqsave(&ep
->lock
, flags
);
1142 /* If the file is already "ready" we drop it inside the ready list */
1143 if ((revents
& event
->events
) && !ep_is_linked(&epi
->rdllink
)) {
1144 list_add_tail(&epi
->rdllink
, &ep
->rdllist
);
1146 /* Notify waiting tasks that events are available */
1147 if (waitqueue_active(&ep
->wq
))
1148 wake_up_locked(&ep
->wq
);
1149 if (waitqueue_active(&ep
->poll_wait
))
1153 spin_unlock_irqrestore(&ep
->lock
, flags
);
1155 atomic_long_inc(&ep
->user
->epoll_watches
);
1157 /* We have to call this outside the lock */
1159 ep_poll_safewake(&ep
->poll_wait
);
1164 spin_lock(&tfile
->f_lock
);
1165 if (ep_is_linked(&epi
->fllink
))
1166 list_del_init(&epi
->fllink
);
1167 spin_unlock(&tfile
->f_lock
);
1169 rb_erase(&epi
->rbn
, &ep
->rbr
);
1172 ep_unregister_pollwait(ep
, epi
);
1175 * We need to do this because an event could have been arrived on some
1176 * allocated wait queue. Note that we don't care about the ep->ovflist
1177 * list, since that is used/cleaned only inside a section bound by "mtx".
1178 * And ep_insert() is called with "mtx" held.
1180 spin_lock_irqsave(&ep
->lock
, flags
);
1181 if (ep_is_linked(&epi
->rdllink
))
1182 list_del_init(&epi
->rdllink
);
1183 spin_unlock_irqrestore(&ep
->lock
, flags
);
1185 kmem_cache_free(epi_cache
, epi
);
1191 * Modify the interest event mask by dropping an event if the new mask
1192 * has a match in the current file status. Must be called with "mtx" held.
1194 static int ep_modify(struct eventpoll
*ep
, struct epitem
*epi
, struct epoll_event
*event
)
1197 unsigned int revents
;
1200 init_poll_funcptr(&pt
, NULL
);
1203 * Set the new event interest mask before calling f_op->poll();
1204 * otherwise we might miss an event that happens between the
1205 * f_op->poll() call and the new event set registering.
1207 epi
->event
.events
= event
->events
;
1208 pt
._key
= event
->events
;
1209 epi
->event
.data
= event
->data
; /* protected by mtx */
1212 * Get current event bits. We can safely use the file* here because
1213 * its usage count has been increased by the caller of this function.
1215 revents
= epi
->ffd
.file
->f_op
->poll(epi
->ffd
.file
, &pt
);
1218 * If the item is "hot" and it is not registered inside the ready
1219 * list, push it inside.
1221 if (revents
& event
->events
) {
1222 spin_lock_irq(&ep
->lock
);
1223 if (!ep_is_linked(&epi
->rdllink
)) {
1224 list_add_tail(&epi
->rdllink
, &ep
->rdllist
);
1226 /* Notify waiting tasks that events are available */
1227 if (waitqueue_active(&ep
->wq
))
1228 wake_up_locked(&ep
->wq
);
1229 if (waitqueue_active(&ep
->poll_wait
))
1232 spin_unlock_irq(&ep
->lock
);
1235 /* We have to call this outside the lock */
1237 ep_poll_safewake(&ep
->poll_wait
);
1242 static int ep_send_events_proc(struct eventpoll
*ep
, struct list_head
*head
,
1245 struct ep_send_events_data
*esed
= priv
;
1247 unsigned int revents
;
1249 struct epoll_event __user
*uevent
;
1252 init_poll_funcptr(&pt
, NULL
);
1255 * We can loop without lock because we are passed a task private list.
1256 * Items cannot vanish during the loop because ep_scan_ready_list() is
1257 * holding "mtx" during this call.
1259 for (eventcnt
= 0, uevent
= esed
->events
;
1260 !list_empty(head
) && eventcnt
< esed
->maxevents
;) {
1261 epi
= list_first_entry(head
, struct epitem
, rdllink
);
1263 list_del_init(&epi
->rdllink
);
1265 pt
._key
= epi
->event
.events
;
1266 revents
= epi
->ffd
.file
->f_op
->poll(epi
->ffd
.file
, &pt
) &
1270 * If the event mask intersect the caller-requested one,
1271 * deliver the event to userspace. Again, ep_scan_ready_list()
1272 * is holding "mtx", so no operations coming from userspace
1273 * can change the item.
1276 if (__put_user(revents
, &uevent
->events
) ||
1277 __put_user(epi
->event
.data
, &uevent
->data
)) {
1278 list_add(&epi
->rdllink
, head
);
1279 return eventcnt
? eventcnt
: -EFAULT
;
1283 if (epi
->event
.events
& EPOLLONESHOT
)
1284 epi
->event
.events
&= EP_PRIVATE_BITS
;
1285 else if (!(epi
->event
.events
& EPOLLET
)) {
1287 * If this file has been added with Level
1288 * Trigger mode, we need to insert back inside
1289 * the ready list, so that the next call to
1290 * epoll_wait() will check again the events
1291 * availability. At this point, no one can insert
1292 * into ep->rdllist besides us. The epoll_ctl()
1293 * callers are locked out by
1294 * ep_scan_ready_list() holding "mtx" and the
1295 * poll callback will queue them in ep->ovflist.
1297 list_add_tail(&epi
->rdllink
, &ep
->rdllist
);
1305 static int ep_send_events(struct eventpoll
*ep
,
1306 struct epoll_event __user
*events
, int maxevents
)
1308 struct ep_send_events_data esed
;
1310 esed
.maxevents
= maxevents
;
1311 esed
.events
= events
;
1313 return ep_scan_ready_list(ep
, ep_send_events_proc
, &esed
, 0);
1316 static inline struct timespec
ep_set_mstimeout(long ms
)
1318 struct timespec now
, ts
= {
1319 .tv_sec
= ms
/ MSEC_PER_SEC
,
1320 .tv_nsec
= NSEC_PER_MSEC
* (ms
% MSEC_PER_SEC
),
1324 return timespec_add_safe(now
, ts
);
1328 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1331 * @ep: Pointer to the eventpoll context.
1332 * @events: Pointer to the userspace buffer where the ready events should be
1334 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1335 * @timeout: Maximum timeout for the ready events fetch operation, in
1336 * milliseconds. If the @timeout is zero, the function will not block,
1337 * while if the @timeout is less than zero, the function will block
1338 * until at least one event has been retrieved (or an error
1341 * Returns: Returns the number of ready events which have been fetched, or an
1342 * error code, in case of error.
1344 static int ep_poll(struct eventpoll
*ep
, struct epoll_event __user
*events
,
1345 int maxevents
, long timeout
)
1347 int res
= 0, eavail
, timed_out
= 0;
1348 unsigned long flags
;
1351 ktime_t expires
, *to
= NULL
;
1354 struct timespec end_time
= ep_set_mstimeout(timeout
);
1356 slack
= select_estimate_accuracy(&end_time
);
1358 *to
= timespec_to_ktime(end_time
);
1359 } else if (timeout
== 0) {
1361 * Avoid the unnecessary trip to the wait queue loop, if the
1362 * caller specified a non blocking operation.
1365 spin_lock_irqsave(&ep
->lock
, flags
);
1370 spin_lock_irqsave(&ep
->lock
, flags
);
1372 if (!ep_events_available(ep
)) {
1374 * We don't have any available event to return to the caller.
1375 * We need to sleep here, and we will be wake up by
1376 * ep_poll_callback() when events will become available.
1378 init_waitqueue_entry(&wait
, current
);
1379 __add_wait_queue_exclusive(&ep
->wq
, &wait
);
1383 * We don't want to sleep if the ep_poll_callback() sends us
1384 * a wakeup in between. That's why we set the task state
1385 * to TASK_INTERRUPTIBLE before doing the checks.
1387 set_current_state(TASK_INTERRUPTIBLE
);
1388 if (ep_events_available(ep
) || timed_out
)
1390 if (signal_pending(current
)) {
1395 spin_unlock_irqrestore(&ep
->lock
, flags
);
1396 if (!schedule_hrtimeout_range(to
, slack
, HRTIMER_MODE_ABS
))
1399 spin_lock_irqsave(&ep
->lock
, flags
);
1401 __remove_wait_queue(&ep
->wq
, &wait
);
1403 set_current_state(TASK_RUNNING
);
1406 /* Is it worth to try to dig for events ? */
1407 eavail
= ep_events_available(ep
);
1409 spin_unlock_irqrestore(&ep
->lock
, flags
);
1412 * Try to transfer events to user space. In case we get 0 events and
1413 * there's still timeout left over, we go trying again in search of
1416 if (!res
&& eavail
&&
1417 !(res
= ep_send_events(ep
, events
, maxevents
)) && !timed_out
)
1424 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1425 * API, to verify that adding an epoll file inside another
1426 * epoll structure, does not violate the constraints, in
1427 * terms of closed loops, or too deep chains (which can
1428 * result in excessive stack usage).
1430 * @priv: Pointer to the epoll file to be currently checked.
1431 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1432 * data structure pointer.
1433 * @call_nests: Current dept of the @ep_call_nested() call stack.
1435 * Returns: Returns zero if adding the epoll @file inside current epoll
1436 * structure @ep does not violate the constraints, or -1 otherwise.
1438 static int ep_loop_check_proc(void *priv
, void *cookie
, int call_nests
)
1441 struct file
*file
= priv
;
1442 struct eventpoll
*ep
= file
->private_data
;
1443 struct eventpoll
*ep_tovisit
;
1444 struct rb_node
*rbp
;
1447 mutex_lock_nested(&ep
->mtx
, call_nests
+ 1);
1449 list_add(&ep
->visited_list_link
, &visited_list
);
1450 for (rbp
= rb_first(&ep
->rbr
); rbp
; rbp
= rb_next(rbp
)) {
1451 epi
= rb_entry(rbp
, struct epitem
, rbn
);
1452 if (unlikely(is_file_epoll(epi
->ffd
.file
))) {
1453 ep_tovisit
= epi
->ffd
.file
->private_data
;
1454 if (ep_tovisit
->visited
)
1456 error
= ep_call_nested(&poll_loop_ncalls
, EP_MAX_NESTS
,
1457 ep_loop_check_proc
, epi
->ffd
.file
,
1458 ep_tovisit
, current
);
1463 * If we've reached a file that is not associated with
1464 * an ep, then we need to check if the newly added
1465 * links are going to add too many wakeup paths. We do
1466 * this by adding it to the tfile_check_list, if it's
1467 * not already there, and calling reverse_path_check()
1468 * during ep_insert().
1470 if (list_empty(&epi
->ffd
.file
->f_tfile_llink
))
1471 list_add(&epi
->ffd
.file
->f_tfile_llink
,
1475 mutex_unlock(&ep
->mtx
);
1481 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1482 * another epoll file (represented by @ep) does not create
1483 * closed loops or too deep chains.
1485 * @ep: Pointer to the epoll private data structure.
1486 * @file: Pointer to the epoll file to be checked.
1488 * Returns: Returns zero if adding the epoll @file inside current epoll
1489 * structure @ep does not violate the constraints, or -1 otherwise.
1491 static int ep_loop_check(struct eventpoll
*ep
, struct file
*file
)
1494 struct eventpoll
*ep_cur
, *ep_next
;
1496 ret
= ep_call_nested(&poll_loop_ncalls
, EP_MAX_NESTS
,
1497 ep_loop_check_proc
, file
, ep
, current
);
1498 /* clear visited list */
1499 list_for_each_entry_safe(ep_cur
, ep_next
, &visited_list
,
1500 visited_list_link
) {
1501 ep_cur
->visited
= 0;
1502 list_del(&ep_cur
->visited_list_link
);
1507 static void clear_tfile_check_list(void)
1511 /* first clear the tfile_check_list */
1512 while (!list_empty(&tfile_check_list
)) {
1513 file
= list_first_entry(&tfile_check_list
, struct file
,
1515 list_del_init(&file
->f_tfile_llink
);
1517 INIT_LIST_HEAD(&tfile_check_list
);
1521 * Open an eventpoll file descriptor.
1523 SYSCALL_DEFINE1(epoll_create1
, int, flags
)
1526 struct eventpoll
*ep
= NULL
;
1529 /* Check the EPOLL_* constant for consistency. */
1530 BUILD_BUG_ON(EPOLL_CLOEXEC
!= O_CLOEXEC
);
1532 if (flags
& ~EPOLL_CLOEXEC
)
1535 * Create the internal data structure ("struct eventpoll").
1537 error
= ep_alloc(&ep
);
1541 * Creates all the items needed to setup an eventpoll file. That is,
1542 * a file structure and a free file descriptor.
1544 fd
= get_unused_fd_flags(O_RDWR
| (flags
& O_CLOEXEC
));
1549 file
= anon_inode_getfile("[eventpoll]", &eventpoll_fops
, ep
,
1550 O_RDWR
| (flags
& O_CLOEXEC
));
1552 error
= PTR_ERR(file
);
1555 fd_install(fd
, file
);
1566 SYSCALL_DEFINE1(epoll_create
, int, size
)
1571 return sys_epoll_create1(0);
1575 * The following function implements the controller interface for
1576 * the eventpoll file that enables the insertion/removal/change of
1577 * file descriptors inside the interest set.
1579 SYSCALL_DEFINE4(epoll_ctl
, int, epfd
, int, op
, int, fd
,
1580 struct epoll_event __user
*, event
)
1583 int did_lock_epmutex
= 0;
1584 struct file
*file
, *tfile
;
1585 struct eventpoll
*ep
;
1587 struct epoll_event epds
;
1590 if (ep_op_has_event(op
) &&
1591 copy_from_user(&epds
, event
, sizeof(struct epoll_event
)))
1594 /* Get the "struct file *" for the eventpoll file */
1600 /* Get the "struct file *" for the target file */
1605 /* The target file descriptor must support poll */
1607 if (!tfile
->f_op
|| !tfile
->f_op
->poll
)
1608 goto error_tgt_fput
;
1611 * We have to check that the file structure underneath the file descriptor
1612 * the user passed to us _is_ an eventpoll file. And also we do not permit
1613 * adding an epoll file descriptor inside itself.
1616 if (file
== tfile
|| !is_file_epoll(file
))
1617 goto error_tgt_fput
;
1620 * At this point it is safe to assume that the "private_data" contains
1621 * our own data structure.
1623 ep
= file
->private_data
;
1626 * When we insert an epoll file descriptor, inside another epoll file
1627 * descriptor, there is the change of creating closed loops, which are
1628 * better be handled here, than in more critical paths. While we are
1629 * checking for loops we also determine the list of files reachable
1630 * and hang them on the tfile_check_list, so we can check that we
1631 * haven't created too many possible wakeup paths.
1633 * We need to hold the epmutex across both ep_insert and ep_remove
1634 * b/c we want to make sure we are looking at a coherent view of
1637 if (op
== EPOLL_CTL_ADD
|| op
== EPOLL_CTL_DEL
) {
1638 mutex_lock(&epmutex
);
1639 did_lock_epmutex
= 1;
1641 if (op
== EPOLL_CTL_ADD
) {
1642 if (is_file_epoll(tfile
)) {
1644 if (ep_loop_check(ep
, tfile
) != 0)
1645 goto error_tgt_fput
;
1647 list_add(&tfile
->f_tfile_llink
, &tfile_check_list
);
1650 mutex_lock_nested(&ep
->mtx
, 0);
1653 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1654 * above, we can be sure to be able to use the item looked up by
1655 * ep_find() till we release the mutex.
1657 epi
= ep_find(ep
, tfile
, fd
);
1663 epds
.events
|= POLLERR
| POLLHUP
;
1664 error
= ep_insert(ep
, &epds
, tfile
, fd
);
1667 clear_tfile_check_list();
1671 error
= ep_remove(ep
, epi
);
1677 epds
.events
|= POLLERR
| POLLHUP
;
1678 error
= ep_modify(ep
, epi
, &epds
);
1683 mutex_unlock(&ep
->mtx
);
1686 if (did_lock_epmutex
)
1687 mutex_unlock(&epmutex
);
1698 * Implement the event wait interface for the eventpoll file. It is the kernel
1699 * part of the user space epoll_wait(2).
1701 SYSCALL_DEFINE4(epoll_wait
, int, epfd
, struct epoll_event __user
*, events
,
1702 int, maxevents
, int, timeout
)
1706 struct eventpoll
*ep
;
1708 /* The maximum number of event must be greater than zero */
1709 if (maxevents
<= 0 || maxevents
> EP_MAX_EVENTS
)
1712 /* Verify that the area passed by the user is writeable */
1713 if (!access_ok(VERIFY_WRITE
, events
, maxevents
* sizeof(struct epoll_event
))) {
1718 /* Get the "struct file *" for the eventpoll file */
1725 * We have to check that the file structure underneath the fd
1726 * the user passed to us _is_ an eventpoll file.
1729 if (!is_file_epoll(file
))
1733 * At this point it is safe to assume that the "private_data" contains
1734 * our own data structure.
1736 ep
= file
->private_data
;
1738 /* Time to fish for events ... */
1739 error
= ep_poll(ep
, events
, maxevents
, timeout
);
1748 #ifdef HAVE_SET_RESTORE_SIGMASK
1751 * Implement the event wait interface for the eventpoll file. It is the kernel
1752 * part of the user space epoll_pwait(2).
1754 SYSCALL_DEFINE6(epoll_pwait
, int, epfd
, struct epoll_event __user
*, events
,
1755 int, maxevents
, int, timeout
, const sigset_t __user
*, sigmask
,
1759 sigset_t ksigmask
, sigsaved
;
1762 * If the caller wants a certain signal mask to be set during the wait,
1766 if (sigsetsize
!= sizeof(sigset_t
))
1768 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
1770 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
) | sigmask(SIGSTOP
));
1771 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1774 error
= sys_epoll_wait(epfd
, events
, maxevents
, timeout
);
1777 * If we changed the signal mask, we need to restore the original one.
1778 * In case we've got a signal while waiting, we do not restore the
1779 * signal mask yet, and we allow do_signal() to deliver the signal on
1780 * the way back to userspace, before the signal mask is restored.
1783 if (error
== -EINTR
) {
1784 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1786 set_restore_sigmask();
1788 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1794 #endif /* HAVE_SET_RESTORE_SIGMASK */
1796 static int __init
eventpoll_init(void)
1802 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1804 max_user_watches
= (((si
.totalram
- si
.totalhigh
) / 25) << PAGE_SHIFT
) /
1806 BUG_ON(max_user_watches
< 0);
1809 * Initialize the structure used to perform epoll file descriptor
1810 * inclusion loops checks.
1812 ep_nested_calls_init(&poll_loop_ncalls
);
1814 /* Initialize the structure used to perform safe poll wait head wake ups */
1815 ep_nested_calls_init(&poll_safewake_ncalls
);
1817 /* Initialize the structure used to perform file's f_op->poll() calls */
1818 ep_nested_calls_init(&poll_readywalk_ncalls
);
1820 /* Allocates slab cache used to allocate "struct epitem" items */
1821 epi_cache
= kmem_cache_create("eventpoll_epi", sizeof(struct epitem
),
1822 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1824 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1825 pwq_cache
= kmem_cache_create("eventpoll_pwq",
1826 sizeof(struct eppoll_entry
), 0, SLAB_PANIC
, NULL
);
1830 fs_initcall(eventpoll_init
);