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 <linux/device.h>
37 #include <asm/uaccess.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 (EPOLLWAKEUP | 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 /* wakeup_source used when EPOLLWAKEUP is set */
159 struct wakeup_source *ws;
161 /* The structure that describe the interested events and the source fd */
162 struct epoll_event event;
166 * This structure is stored inside the "private_data" member of the file
167 * structure and represents the main data structure for the eventpoll
171 /* Protect the access to this structure */
175 * This mutex is used to ensure that files are not removed
176 * while epoll is using them. This is held during the event
177 * collection loop, the file cleanup path, the epoll file exit
178 * code and the ctl operations.
182 /* Wait queue used by sys_epoll_wait() */
183 wait_queue_head_t wq;
185 /* Wait queue used by file->poll() */
186 wait_queue_head_t poll_wait;
188 /* List of ready file descriptors */
189 struct list_head rdllist;
191 /* RB tree root used to store monitored fd structs */
195 * This is a single linked list that chains all the "struct epitem" that
196 * happened while transferring ready events to userspace w/out
199 struct epitem *ovflist;
201 /* wakeup_source used when ep_scan_ready_list is running */
202 struct wakeup_source *ws;
204 /* The user that created the eventpoll descriptor */
205 struct user_struct *user;
209 /* used to optimize loop detection check */
211 struct list_head visited_list_link;
214 /* Wait structure used by the poll hooks */
215 struct eppoll_entry {
216 /* List header used to link this structure to the "struct epitem" */
217 struct list_head llink;
219 /* The "base" pointer is set to the container "struct epitem" */
223 * Wait queue item that will be linked to the target file wait
228 /* The wait queue head that linked the "wait" wait queue item */
229 wait_queue_head_t *whead;
232 /* Wrapper struct used by poll queueing */
238 /* Used by the ep_send_events() function as callback private data */
239 struct ep_send_events_data {
241 struct epoll_event __user *events;
245 * Configuration options available inside /proc/sys/fs/epoll/
247 /* Maximum number of epoll watched descriptors, per user */
248 static long max_user_watches __read_mostly;
251 * This mutex is used to serialize ep_free() and eventpoll_release_file().
253 static DEFINE_MUTEX(epmutex);
255 /* Used to check for epoll file descriptor inclusion loops */
256 static struct nested_calls poll_loop_ncalls;
258 /* Used for safe wake up implementation */
259 static struct nested_calls poll_safewake_ncalls;
261 /* Used to call file's f_op->poll() under the nested calls boundaries */
262 static struct nested_calls poll_readywalk_ncalls;
264 /* Slab cache used to allocate "struct epitem" */
265 static struct kmem_cache *epi_cache __read_mostly;
267 /* Slab cache used to allocate "struct eppoll_entry" */
268 static struct kmem_cache *pwq_cache __read_mostly;
270 /* Visited nodes during ep_loop_check(), so we can unset them when we finish */
271 static LIST_HEAD(visited_list);
274 * List of files with newly added links, where we may need to limit the number
275 * of emanating paths. Protected by the epmutex.
277 static LIST_HEAD(tfile_check_list);
281 #include <linux/sysctl.h>
284 static long long_max = LONG_MAX;
286 ctl_table epoll_table[] = {
288 .procname = "max_user_watches",
289 .data = &max_user_watches,
290 .maxlen = sizeof(max_user_watches),
292 .proc_handler = proc_doulongvec_minmax,
298 #endif /* CONFIG_SYSCTL */
300 static const struct file_operations eventpoll_fops;
302 static inline int is_file_epoll(struct file *f)
304 return f->f_op == &eventpoll_fops;
307 /* Setup the structure that is used as key for the RB tree */
308 static inline void ep_set_ffd(struct epoll_filefd *ffd,
309 struct file *file, int fd)
315 /* Compare RB tree keys */
316 static inline int ep_cmp_ffd(struct epoll_filefd *p1,
317 struct epoll_filefd *p2)
319 return (p1->file > p2->file ? +1:
320 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
323 /* Tells us if the item is currently linked */
324 static inline int ep_is_linked(struct list_head *p)
326 return !list_empty(p);
329 static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
331 return container_of(p, struct eppoll_entry, wait);
334 /* Get the "struct epitem" from a wait queue pointer */
335 static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
337 return container_of(p, struct eppoll_entry, wait)->base;
340 /* Get the "struct epitem" from an epoll queue wrapper */
341 static inline struct epitem *ep_item_from_epqueue(poll_table *p)
343 return container_of(p, struct ep_pqueue, pt)->epi;
346 /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
347 static inline int ep_op_has_event(int op)
349 return op != EPOLL_CTL_DEL;
352 /* Initialize the poll safe wake up structure */
353 static void ep_nested_calls_init(struct nested_calls *ncalls)
355 INIT_LIST_HEAD(&ncalls->tasks_call_list);
356 spin_lock_init(&ncalls->lock);
360 * ep_events_available - Checks if ready events might be available.
362 * @ep: Pointer to the eventpoll context.
364 * Returns: Returns a value different than zero if ready events are available,
367 static inline int ep_events_available(struct eventpoll *ep)
369 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
373 * ep_call_nested - Perform a bound (possibly) nested call, by checking
374 * that the recursion limit is not exceeded, and that
375 * the same nested call (by the meaning of same cookie) is
378 * @ncalls: Pointer to the nested_calls structure to be used for this call.
379 * @max_nests: Maximum number of allowed nesting calls.
380 * @nproc: Nested call core function pointer.
381 * @priv: Opaque data to be passed to the @nproc callback.
382 * @cookie: Cookie to be used to identify this nested call.
383 * @ctx: This instance context.
385 * Returns: Returns the code returned by the @nproc callback, or -1 if
386 * the maximum recursion limit has been exceeded.
388 static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
389 int (*nproc)(void *, void *, int), void *priv,
390 void *cookie, void *ctx)
392 int error, call_nests = 0;
394 struct list_head *lsthead = &ncalls->tasks_call_list;
395 struct nested_call_node *tncur;
396 struct nested_call_node tnode;
398 spin_lock_irqsave(&ncalls->lock, flags);
401 * Try to see if the current task is already inside this wakeup call.
402 * We use a list here, since the population inside this set is always
405 list_for_each_entry(tncur, lsthead, llink) {
406 if (tncur->ctx == ctx &&
407 (tncur->cookie == cookie || ++call_nests > max_nests)) {
409 * Ops ... loop detected or maximum nest level reached.
410 * We abort this wake by breaking the cycle itself.
417 /* Add the current task and cookie to the list */
419 tnode.cookie = cookie;
420 list_add(&tnode.llink, lsthead);
422 spin_unlock_irqrestore(&ncalls->lock, flags);
424 /* Call the nested function */
425 error = (*nproc)(priv, cookie, call_nests);
427 /* Remove the current task from the list */
428 spin_lock_irqsave(&ncalls->lock, flags);
429 list_del(&tnode.llink);
431 spin_unlock_irqrestore(&ncalls->lock, flags);
437 * As described in commit 0ccf831cb lockdep: annotate epoll
438 * the use of wait queues used by epoll is done in a very controlled
439 * manner. Wake ups can nest inside each other, but are never done
440 * with the same locking. For example:
443 * efd1 = epoll_create();
444 * efd2 = epoll_create();
445 * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
446 * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
448 * When a packet arrives to the device underneath "dfd", the net code will
449 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
450 * callback wakeup entry on that queue, and the wake_up() performed by the
451 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
452 * (efd1) notices that it may have some event ready, so it needs to wake up
453 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
454 * that ends up in another wake_up(), after having checked about the
455 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
456 * avoid stack blasting.
458 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
459 * this special case of epoll.
461 #ifdef CONFIG_DEBUG_LOCK_ALLOC
462 static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
463 unsigned long events, int subclass)
467 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
468 wake_up_locked_poll(wqueue, events);
469 spin_unlock_irqrestore(&wqueue->lock, flags);
472 static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
473 unsigned long events, int subclass)
475 wake_up_poll(wqueue, events);
479 static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
481 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
487 * Perform a safe wake up of the poll wait list. The problem is that
488 * with the new callback'd wake up system, it is possible that the
489 * poll callback is reentered from inside the call to wake_up() done
490 * on the poll wait queue head. The rule is that we cannot reenter the
491 * wake up code from the same task more than EP_MAX_NESTS times,
492 * and we cannot reenter the same wait queue head at all. This will
493 * enable to have a hierarchy of epoll file descriptor of no more than
496 static void ep_poll_safewake(wait_queue_head_t *wq)
498 int this_cpu = get_cpu();
500 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
501 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
506 static void ep_remove_wait_queue(struct eppoll_entry *pwq)
508 wait_queue_head_t *whead;
511 /* If it is cleared by POLLFREE, it should be rcu-safe */
512 whead = rcu_dereference(pwq->whead);
514 remove_wait_queue(whead, &pwq->wait);
519 * This function unregisters poll callbacks from the associated file
520 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
523 static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
525 struct list_head *lsthead = &epi->pwqlist;
526 struct eppoll_entry *pwq;
528 while (!list_empty(lsthead)) {
529 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
531 list_del(&pwq->llink);
532 ep_remove_wait_queue(pwq);
533 kmem_cache_free(pwq_cache, pwq);
538 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
539 * the scan code, to call f_op->poll(). Also allows for
540 * O(NumReady) performance.
542 * @ep: Pointer to the epoll private data structure.
543 * @sproc: Pointer to the scan callback.
544 * @priv: Private opaque data passed to the @sproc callback.
545 * @depth: The current depth of recursive f_op->poll calls.
547 * Returns: The same integer error code returned by the @sproc callback.
549 static int ep_scan_ready_list(struct eventpoll *ep,
550 int (*sproc)(struct eventpoll *,
551 struct list_head *, void *),
555 int error, pwake = 0;
557 struct epitem *epi, *nepi;
561 * We need to lock this because we could be hit by
562 * eventpoll_release_file() and epoll_ctl().
564 mutex_lock_nested(&ep->mtx, depth);
567 * Steal the ready list, and re-init the original one to the
568 * empty list. Also, set ep->ovflist to NULL so that events
569 * happening while looping w/out locks, are not lost. We cannot
570 * have the poll callback to queue directly on ep->rdllist,
571 * because we want the "sproc" callback to be able to do it
574 spin_lock_irqsave(&ep->lock, flags);
575 list_splice_init(&ep->rdllist, &txlist);
577 spin_unlock_irqrestore(&ep->lock, flags);
580 * Now call the callback function.
582 error = (*sproc)(ep, &txlist, priv);
584 spin_lock_irqsave(&ep->lock, flags);
586 * During the time we spent inside the "sproc" callback, some
587 * other events might have been queued by the poll callback.
588 * We re-insert them inside the main ready-list here.
590 for (nepi = ep->ovflist; (epi = nepi) != NULL;
591 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
593 * We need to check if the item is already in the list.
594 * During the "sproc" callback execution time, items are
595 * queued into ->ovflist but the "txlist" might already
596 * contain them, and the list_splice() below takes care of them.
598 if (!ep_is_linked(&epi->rdllink)) {
599 list_add_tail(&epi->rdllink, &ep->rdllist);
600 __pm_stay_awake(epi->ws);
604 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
605 * releasing the lock, events will be queued in the normal way inside
608 ep->ovflist = EP_UNACTIVE_PTR;
611 * Quickly re-inject items left on "txlist".
613 list_splice(&txlist, &ep->rdllist);
616 if (!list_empty(&ep->rdllist)) {
618 * Wake up (if active) both the eventpoll wait list and
619 * the ->poll() wait list (delayed after we release the lock).
621 if (waitqueue_active(&ep->wq))
622 wake_up_locked(&ep->wq);
623 if (waitqueue_active(&ep->poll_wait))
626 spin_unlock_irqrestore(&ep->lock, flags);
628 mutex_unlock(&ep->mtx);
630 /* We have to call this outside the lock */
632 ep_poll_safewake(&ep->poll_wait);
638 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
639 * all the associated resources. Must be called with "mtx" held.
641 static int ep_remove(struct eventpoll *ep, struct epitem *epi)
644 struct file *file = epi->ffd.file;
647 * Removes poll wait queue hooks. We _have_ to do this without holding
648 * the "ep->lock" otherwise a deadlock might occur. This because of the
649 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
650 * queue head lock when unregistering the wait queue. The wakeup callback
651 * will run by holding the wait queue head lock and will call our callback
652 * that will try to get "ep->lock".
654 ep_unregister_pollwait(ep, epi);
656 /* Remove the current item from the list of epoll hooks */
657 spin_lock(&file->f_lock);
658 if (ep_is_linked(&epi->fllink))
659 list_del_init(&epi->fllink);
660 spin_unlock(&file->f_lock);
662 rb_erase(&epi->rbn, &ep->rbr);
664 spin_lock_irqsave(&ep->lock, flags);
665 if (ep_is_linked(&epi->rdllink))
666 list_del_init(&epi->rdllink);
667 spin_unlock_irqrestore(&ep->lock, flags);
669 wakeup_source_unregister(epi->ws);
671 /* At this point it is safe to free the eventpoll item */
672 kmem_cache_free(epi_cache, epi);
674 atomic_long_dec(&ep->user->epoll_watches);
679 static void ep_free(struct eventpoll *ep)
684 /* We need to release all tasks waiting for these file */
685 if (waitqueue_active(&ep->poll_wait))
686 ep_poll_safewake(&ep->poll_wait);
689 * We need to lock this because we could be hit by
690 * eventpoll_release_file() while we're freeing the "struct eventpoll".
691 * We do not need to hold "ep->mtx" here because the epoll file
692 * is on the way to be removed and no one has references to it
693 * anymore. The only hit might come from eventpoll_release_file() but
694 * holding "epmutex" is sufficient here.
696 mutex_lock(&epmutex);
699 * Walks through the whole tree by unregistering poll callbacks.
701 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
702 epi = rb_entry(rbp, struct epitem, rbn);
704 ep_unregister_pollwait(ep, epi);
708 * Walks through the whole tree by freeing each "struct epitem". At this
709 * point we are sure no poll callbacks will be lingering around, and also by
710 * holding "epmutex" we can be sure that no file cleanup code will hit
711 * us during this operation. So we can avoid the lock on "ep->lock".
713 while ((rbp = rb_first(&ep->rbr)) != NULL) {
714 epi = rb_entry(rbp, struct epitem, rbn);
718 mutex_unlock(&epmutex);
719 mutex_destroy(&ep->mtx);
721 wakeup_source_unregister(ep->ws);
725 static int ep_eventpoll_release(struct inode *inode, struct file *file)
727 struct eventpoll *ep = file->private_data;
735 static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
738 struct epitem *epi, *tmp;
741 init_poll_funcptr(&pt, NULL);
742 list_for_each_entry_safe(epi, tmp, head, rdllink) {
743 pt._key = epi->event.events;
744 if (epi->ffd.file->f_op->poll(epi->ffd.file, &pt) &
746 return POLLIN | POLLRDNORM;
749 * Item has been dropped into the ready list by the poll
750 * callback, but it's not actually ready, as far as
751 * caller requested events goes. We can remove it here.
754 list_del_init(&epi->rdllink);
761 static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
763 return ep_scan_ready_list(priv, ep_read_events_proc, NULL, call_nests + 1);
766 static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
769 struct eventpoll *ep = file->private_data;
771 /* Insert inside our poll wait queue */
772 poll_wait(file, &ep->poll_wait, wait);
775 * Proceed to find out if wanted events are really available inside
776 * the ready list. This need to be done under ep_call_nested()
777 * supervision, since the call to f_op->poll() done on listed files
778 * could re-enter here.
780 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
781 ep_poll_readyevents_proc, ep, ep, current);
783 return pollflags != -1 ? pollflags : 0;
786 /* File callbacks that implement the eventpoll file behaviour */
787 static const struct file_operations eventpoll_fops = {
788 .release = ep_eventpoll_release,
789 .poll = ep_eventpoll_poll,
790 .llseek = noop_llseek,
794 * This is called from eventpoll_release() to unlink files from the eventpoll
795 * interface. We need to have this facility to cleanup correctly files that are
796 * closed without being removed from the eventpoll interface.
798 void eventpoll_release_file(struct file *file)
800 struct list_head *lsthead = &file->f_ep_links;
801 struct eventpoll *ep;
805 * We don't want to get "file->f_lock" because it is not
806 * necessary. It is not necessary because we're in the "struct file"
807 * cleanup path, and this means that no one is using this file anymore.
808 * So, for example, epoll_ctl() cannot hit here since if we reach this
809 * point, the file counter already went to zero and fget() would fail.
810 * The only hit might come from ep_free() but by holding the mutex
811 * will correctly serialize the operation. We do need to acquire
812 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
813 * from anywhere but ep_free().
815 * Besides, ep_remove() acquires the lock, so we can't hold it here.
817 mutex_lock(&epmutex);
819 while (!list_empty(lsthead)) {
820 epi = list_first_entry(lsthead, struct epitem, fllink);
823 list_del_init(&epi->fllink);
824 mutex_lock_nested(&ep->mtx, 0);
826 mutex_unlock(&ep->mtx);
829 mutex_unlock(&epmutex);
832 static int ep_alloc(struct eventpoll **pep)
835 struct user_struct *user;
836 struct eventpoll *ep;
838 user = get_current_user();
840 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
844 spin_lock_init(&ep->lock);
845 mutex_init(&ep->mtx);
846 init_waitqueue_head(&ep->wq);
847 init_waitqueue_head(&ep->poll_wait);
848 INIT_LIST_HEAD(&ep->rdllist);
850 ep->ovflist = EP_UNACTIVE_PTR;
863 * Search the file inside the eventpoll tree. The RB tree operations
864 * are protected by the "mtx" mutex, and ep_find() must be called with
867 static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
871 struct epitem *epi, *epir = NULL;
872 struct epoll_filefd ffd;
874 ep_set_ffd(&ffd, file, fd);
875 for (rbp = ep->rbr.rb_node; rbp; ) {
876 epi = rb_entry(rbp, struct epitem, rbn);
877 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
892 * This is the callback that is passed to the wait queue wakeup
893 * mechanism. It is called by the stored file descriptors when they
894 * have events to report.
896 static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
900 struct epitem *epi = ep_item_from_wait(wait);
901 struct eventpoll *ep = epi->ep;
903 if ((unsigned long)key & POLLFREE) {
904 ep_pwq_from_wait(wait)->whead = NULL;
906 * whead = NULL above can race with ep_remove_wait_queue()
907 * which can do another remove_wait_queue() after us, so we
908 * can't use __remove_wait_queue(). whead->lock is held by
911 list_del_init(&wait->task_list);
914 spin_lock_irqsave(&ep->lock, flags);
917 * If the event mask does not contain any poll(2) event, we consider the
918 * descriptor to be disabled. This condition is likely the effect of the
919 * EPOLLONESHOT bit that disables the descriptor when an event is received,
920 * until the next EPOLL_CTL_MOD will be issued.
922 if (!(epi->event.events & ~EP_PRIVATE_BITS))
926 * Check the events coming with the callback. At this stage, not
927 * every device reports the events in the "key" parameter of the
928 * callback. We need to be able to handle both cases here, hence the
929 * test for "key" != NULL before the event match test.
931 if (key && !((unsigned long) key & epi->event.events))
935 * If we are transferring events to userspace, we can hold no locks
936 * (because we're accessing user memory, and because of linux f_op->poll()
937 * semantics). All the events that happen during that period of time are
938 * chained in ep->ovflist and requeued later on.
940 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
941 if (epi->next == EP_UNACTIVE_PTR) {
942 epi->next = ep->ovflist;
946 * Activate ep->ws since epi->ws may get
947 * deactivated at any time.
949 __pm_stay_awake(ep->ws);
956 /* If this file is already in the ready list we exit soon */
957 if (!ep_is_linked(&epi->rdllink)) {
958 list_add_tail(&epi->rdllink, &ep->rdllist);
959 __pm_stay_awake(epi->ws);
963 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
966 if (waitqueue_active(&ep->wq))
967 wake_up_locked(&ep->wq);
968 if (waitqueue_active(&ep->poll_wait))
972 spin_unlock_irqrestore(&ep->lock, flags);
974 /* We have to call this outside the lock */
976 ep_poll_safewake(&ep->poll_wait);
982 * This is the callback that is used to add our wait queue to the
983 * target file wakeup lists.
985 static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
988 struct epitem *epi = ep_item_from_epqueue(pt);
989 struct eppoll_entry *pwq;
991 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
992 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
995 add_wait_queue(whead, &pwq->wait);
996 list_add_tail(&pwq->llink, &epi->pwqlist);
999 /* We have to signal that an error occurred */
1004 static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1007 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
1008 struct epitem *epic;
1012 epic = rb_entry(parent, struct epitem, rbn);
1013 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1015 p = &parent->rb_right;
1017 p = &parent->rb_left;
1019 rb_link_node(&epi->rbn, parent, p);
1020 rb_insert_color(&epi->rbn, &ep->rbr);
1025 #define PATH_ARR_SIZE 5
1027 * These are the number paths of length 1 to 5, that we are allowing to emanate
1028 * from a single file of interest. For example, we allow 1000 paths of length
1029 * 1, to emanate from each file of interest. This essentially represents the
1030 * potential wakeup paths, which need to be limited in order to avoid massive
1031 * uncontrolled wakeup storms. The common use case should be a single ep which
1032 * is connected to n file sources. In this case each file source has 1 path
1033 * of length 1. Thus, the numbers below should be more than sufficient. These
1034 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1035 * and delete can't add additional paths. Protected by the epmutex.
1037 static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1038 static int path_count[PATH_ARR_SIZE];
1040 static int path_count_inc(int nests)
1042 /* Allow an arbitrary number of depth 1 paths */
1046 if (++path_count[nests] > path_limits[nests])
1051 static void path_count_init(void)
1055 for (i = 0; i < PATH_ARR_SIZE; i++)
1059 static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1062 struct file *file = priv;
1063 struct file *child_file;
1066 list_for_each_entry(epi, &file->f_ep_links, fllink) {
1067 child_file = epi->ep->file;
1068 if (is_file_epoll(child_file)) {
1069 if (list_empty(&child_file->f_ep_links)) {
1070 if (path_count_inc(call_nests)) {
1075 error = ep_call_nested(&poll_loop_ncalls,
1077 reverse_path_check_proc,
1078 child_file, child_file,
1084 printk(KERN_ERR "reverse_path_check_proc: "
1085 "file is not an ep!\n");
1092 * reverse_path_check - The tfile_check_list is list of file *, which have
1093 * links that are proposed to be newly added. We need to
1094 * make sure that those added links don't add too many
1095 * paths such that we will spend all our time waking up
1096 * eventpoll objects.
1098 * Returns: Returns zero if the proposed links don't create too many paths,
1101 static int reverse_path_check(void)
1104 struct file *current_file;
1106 /* let's call this for all tfiles */
1107 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1109 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1110 reverse_path_check_proc, current_file,
1111 current_file, current);
1118 static int ep_create_wakeup_source(struct epitem *epi)
1123 epi->ep->ws = wakeup_source_register("eventpoll");
1128 name = epi->ffd.file->f_path.dentry->d_name.name;
1129 epi->ws = wakeup_source_register(name);
1136 static void ep_destroy_wakeup_source(struct epitem *epi)
1138 wakeup_source_unregister(epi->ws);
1143 * Must be called with "mtx" held.
1145 static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1146 struct file *tfile, int fd)
1148 int error, revents, pwake = 0;
1149 unsigned long flags;
1152 struct ep_pqueue epq;
1154 user_watches = atomic_long_read(&ep->user->epoll_watches);
1155 if (unlikely(user_watches >= max_user_watches))
1157 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
1160 /* Item initialization follow here ... */
1161 INIT_LIST_HEAD(&epi->rdllink);
1162 INIT_LIST_HEAD(&epi->fllink);
1163 INIT_LIST_HEAD(&epi->pwqlist);
1165 ep_set_ffd(&epi->ffd, tfile, fd);
1166 epi->event = *event;
1168 epi->next = EP_UNACTIVE_PTR;
1169 if (epi->event.events & EPOLLWAKEUP) {
1170 error = ep_create_wakeup_source(epi);
1172 goto error_create_wakeup_source;
1177 /* Initialize the poll table using the queue callback */
1179 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1180 epq.pt._key = event->events;
1183 * Attach the item to the poll hooks and get current event bits.
1184 * We can safely use the file* here because its usage count has
1185 * been increased by the caller of this function. Note that after
1186 * this operation completes, the poll callback can start hitting
1189 revents = tfile->f_op->poll(tfile, &epq.pt);
1192 * We have to check if something went wrong during the poll wait queue
1193 * install process. Namely an allocation for a wait queue failed due
1194 * high memory pressure.
1198 goto error_unregister;
1200 /* Add the current item to the list of active epoll hook for this file */
1201 spin_lock(&tfile->f_lock);
1202 list_add_tail(&epi->fllink, &tfile->f_ep_links);
1203 spin_unlock(&tfile->f_lock);
1206 * Add the current item to the RB tree. All RB tree operations are
1207 * protected by "mtx", and ep_insert() is called with "mtx" held.
1209 ep_rbtree_insert(ep, epi);
1211 /* now check if we've created too many backpaths */
1213 if (reverse_path_check())
1214 goto error_remove_epi;
1216 /* We have to drop the new item inside our item list to keep track of it */
1217 spin_lock_irqsave(&ep->lock, flags);
1219 /* If the file is already "ready" we drop it inside the ready list */
1220 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
1221 list_add_tail(&epi->rdllink, &ep->rdllist);
1222 __pm_stay_awake(epi->ws);
1224 /* Notify waiting tasks that events are available */
1225 if (waitqueue_active(&ep->wq))
1226 wake_up_locked(&ep->wq);
1227 if (waitqueue_active(&ep->poll_wait))
1231 spin_unlock_irqrestore(&ep->lock, flags);
1233 atomic_long_inc(&ep->user->epoll_watches);
1235 /* We have to call this outside the lock */
1237 ep_poll_safewake(&ep->poll_wait);
1242 spin_lock(&tfile->f_lock);
1243 if (ep_is_linked(&epi->fllink))
1244 list_del_init(&epi->fllink);
1245 spin_unlock(&tfile->f_lock);
1247 rb_erase(&epi->rbn, &ep->rbr);
1250 ep_unregister_pollwait(ep, epi);
1253 * We need to do this because an event could have been arrived on some
1254 * allocated wait queue. Note that we don't care about the ep->ovflist
1255 * list, since that is used/cleaned only inside a section bound by "mtx".
1256 * And ep_insert() is called with "mtx" held.
1258 spin_lock_irqsave(&ep->lock, flags);
1259 if (ep_is_linked(&epi->rdllink))
1260 list_del_init(&epi->rdllink);
1261 spin_unlock_irqrestore(&ep->lock, flags);
1263 wakeup_source_unregister(epi->ws);
1265 error_create_wakeup_source:
1266 kmem_cache_free(epi_cache, epi);
1272 * Modify the interest event mask by dropping an event if the new mask
1273 * has a match in the current file status. Must be called with "mtx" held.
1275 static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1278 unsigned int revents;
1281 init_poll_funcptr(&pt, NULL);
1284 * Set the new event interest mask before calling f_op->poll();
1285 * otherwise we might miss an event that happens between the
1286 * f_op->poll() call and the new event set registering.
1288 epi->event.events = event->events;
1289 pt._key = event->events;
1290 epi->event.data = event->data; /* protected by mtx */
1291 if (epi->event.events & EPOLLWAKEUP) {
1293 ep_create_wakeup_source(epi);
1294 } else if (epi->ws) {
1295 ep_destroy_wakeup_source(epi);
1299 * Get current event bits. We can safely use the file* here because
1300 * its usage count has been increased by the caller of this function.
1302 revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt);
1305 * If the item is "hot" and it is not registered inside the ready
1306 * list, push it inside.
1308 if (revents & event->events) {
1309 spin_lock_irq(&ep->lock);
1310 if (!ep_is_linked(&epi->rdllink)) {
1311 list_add_tail(&epi->rdllink, &ep->rdllist);
1312 __pm_stay_awake(epi->ws);
1314 /* Notify waiting tasks that events are available */
1315 if (waitqueue_active(&ep->wq))
1316 wake_up_locked(&ep->wq);
1317 if (waitqueue_active(&ep->poll_wait))
1320 spin_unlock_irq(&ep->lock);
1323 /* We have to call this outside the lock */
1325 ep_poll_safewake(&ep->poll_wait);
1330 static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1333 struct ep_send_events_data *esed = priv;
1335 unsigned int revents;
1337 struct epoll_event __user *uevent;
1340 init_poll_funcptr(&pt, NULL);
1343 * We can loop without lock because we are passed a task private list.
1344 * Items cannot vanish during the loop because ep_scan_ready_list() is
1345 * holding "mtx" during this call.
1347 for (eventcnt = 0, uevent = esed->events;
1348 !list_empty(head) && eventcnt < esed->maxevents;) {
1349 epi = list_first_entry(head, struct epitem, rdllink);
1352 * Activate ep->ws before deactivating epi->ws to prevent
1353 * triggering auto-suspend here (in case we reactive epi->ws
1356 * This could be rearranged to delay the deactivation of epi->ws
1357 * instead, but then epi->ws would temporarily be out of sync
1358 * with ep_is_linked().
1360 if (epi->ws && epi->ws->active)
1361 __pm_stay_awake(ep->ws);
1362 __pm_relax(epi->ws);
1363 list_del_init(&epi->rdllink);
1365 pt._key = epi->event.events;
1366 revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt) &
1370 * If the event mask intersect the caller-requested one,
1371 * deliver the event to userspace. Again, ep_scan_ready_list()
1372 * is holding "mtx", so no operations coming from userspace
1373 * can change the item.
1376 if (__put_user(revents, &uevent->events) ||
1377 __put_user(epi->event.data, &uevent->data)) {
1378 list_add(&epi->rdllink, head);
1379 __pm_stay_awake(epi->ws);
1380 return eventcnt ? eventcnt : -EFAULT;
1384 if (epi->event.events & EPOLLONESHOT)
1385 epi->event.events &= EP_PRIVATE_BITS;
1386 else if (!(epi->event.events & EPOLLET)) {
1388 * If this file has been added with Level
1389 * Trigger mode, we need to insert back inside
1390 * the ready list, so that the next call to
1391 * epoll_wait() will check again the events
1392 * availability. At this point, no one can insert
1393 * into ep->rdllist besides us. The epoll_ctl()
1394 * callers are locked out by
1395 * ep_scan_ready_list() holding "mtx" and the
1396 * poll callback will queue them in ep->ovflist.
1398 list_add_tail(&epi->rdllink, &ep->rdllist);
1399 __pm_stay_awake(epi->ws);
1407 static int ep_send_events(struct eventpoll *ep,
1408 struct epoll_event __user *events, int maxevents)
1410 struct ep_send_events_data esed;
1412 esed.maxevents = maxevents;
1413 esed.events = events;
1415 return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0);
1418 static inline struct timespec ep_set_mstimeout(long ms)
1420 struct timespec now, ts = {
1421 .tv_sec = ms / MSEC_PER_SEC,
1422 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1426 return timespec_add_safe(now, ts);
1430 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1433 * @ep: Pointer to the eventpoll context.
1434 * @events: Pointer to the userspace buffer where the ready events should be
1436 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1437 * @timeout: Maximum timeout for the ready events fetch operation, in
1438 * milliseconds. If the @timeout is zero, the function will not block,
1439 * while if the @timeout is less than zero, the function will block
1440 * until at least one event has been retrieved (or an error
1443 * Returns: Returns the number of ready events which have been fetched, or an
1444 * error code, in case of error.
1446 static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1447 int maxevents, long timeout)
1449 int res = 0, eavail, timed_out = 0;
1450 unsigned long flags;
1453 ktime_t expires, *to = NULL;
1456 struct timespec end_time = ep_set_mstimeout(timeout);
1458 slack = select_estimate_accuracy(&end_time);
1460 *to = timespec_to_ktime(end_time);
1461 } else if (timeout == 0) {
1463 * Avoid the unnecessary trip to the wait queue loop, if the
1464 * caller specified a non blocking operation.
1467 spin_lock_irqsave(&ep->lock, flags);
1472 spin_lock_irqsave(&ep->lock, flags);
1474 if (!ep_events_available(ep)) {
1476 * We don't have any available event to return to the caller.
1477 * We need to sleep here, and we will be wake up by
1478 * ep_poll_callback() when events will become available.
1480 init_waitqueue_entry(&wait, current);
1481 __add_wait_queue_exclusive(&ep->wq, &wait);
1485 * We don't want to sleep if the ep_poll_callback() sends us
1486 * a wakeup in between. That's why we set the task state
1487 * to TASK_INTERRUPTIBLE before doing the checks.
1489 set_current_state(TASK_INTERRUPTIBLE);
1490 if (ep_events_available(ep) || timed_out)
1492 if (signal_pending(current)) {
1497 spin_unlock_irqrestore(&ep->lock, flags);
1498 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1501 spin_lock_irqsave(&ep->lock, flags);
1503 __remove_wait_queue(&ep->wq, &wait);
1505 set_current_state(TASK_RUNNING);
1508 /* Is it worth to try to dig for events ? */
1509 eavail = ep_events_available(ep);
1511 spin_unlock_irqrestore(&ep->lock, flags);
1514 * Try to transfer events to user space. In case we get 0 events and
1515 * there's still timeout left over, we go trying again in search of
1518 if (!res && eavail &&
1519 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
1526 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1527 * API, to verify that adding an epoll file inside another
1528 * epoll structure, does not violate the constraints, in
1529 * terms of closed loops, or too deep chains (which can
1530 * result in excessive stack usage).
1532 * @priv: Pointer to the epoll file to be currently checked.
1533 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1534 * data structure pointer.
1535 * @call_nests: Current dept of the @ep_call_nested() call stack.
1537 * Returns: Returns zero if adding the epoll @file inside current epoll
1538 * structure @ep does not violate the constraints, or -1 otherwise.
1540 static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1543 struct file *file = priv;
1544 struct eventpoll *ep = file->private_data;
1545 struct eventpoll *ep_tovisit;
1546 struct rb_node *rbp;
1549 mutex_lock_nested(&ep->mtx, call_nests + 1);
1551 list_add(&ep->visited_list_link, &visited_list);
1552 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1553 epi = rb_entry(rbp, struct epitem, rbn);
1554 if (unlikely(is_file_epoll(epi->ffd.file))) {
1555 ep_tovisit = epi->ffd.file->private_data;
1556 if (ep_tovisit->visited)
1558 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1559 ep_loop_check_proc, epi->ffd.file,
1560 ep_tovisit, current);
1565 * If we've reached a file that is not associated with
1566 * an ep, then we need to check if the newly added
1567 * links are going to add too many wakeup paths. We do
1568 * this by adding it to the tfile_check_list, if it's
1569 * not already there, and calling reverse_path_check()
1570 * during ep_insert().
1572 if (list_empty(&epi->ffd.file->f_tfile_llink))
1573 list_add(&epi->ffd.file->f_tfile_llink,
1577 mutex_unlock(&ep->mtx);
1583 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1584 * another epoll file (represented by @ep) does not create
1585 * closed loops or too deep chains.
1587 * @ep: Pointer to the epoll private data structure.
1588 * @file: Pointer to the epoll file to be checked.
1590 * Returns: Returns zero if adding the epoll @file inside current epoll
1591 * structure @ep does not violate the constraints, or -1 otherwise.
1593 static int ep_loop_check(struct eventpoll *ep, struct file *file)
1596 struct eventpoll *ep_cur, *ep_next;
1598 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1599 ep_loop_check_proc, file, ep, current);
1600 /* clear visited list */
1601 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1602 visited_list_link) {
1603 ep_cur->visited = 0;
1604 list_del(&ep_cur->visited_list_link);
1609 static void clear_tfile_check_list(void)
1613 /* first clear the tfile_check_list */
1614 while (!list_empty(&tfile_check_list)) {
1615 file = list_first_entry(&tfile_check_list, struct file,
1617 list_del_init(&file->f_tfile_llink);
1619 INIT_LIST_HEAD(&tfile_check_list);
1623 * Open an eventpoll file descriptor.
1625 SYSCALL_DEFINE1(epoll_create1, int, flags)
1628 struct eventpoll *ep = NULL;
1631 /* Check the EPOLL_* constant for consistency. */
1632 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1634 if (flags & ~EPOLL_CLOEXEC)
1637 * Create the internal data structure ("struct eventpoll").
1639 error = ep_alloc(&ep);
1643 * Creates all the items needed to setup an eventpoll file. That is,
1644 * a file structure and a free file descriptor.
1646 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1651 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
1652 O_RDWR | (flags & O_CLOEXEC));
1654 error = PTR_ERR(file);
1657 fd_install(fd, file);
1668 SYSCALL_DEFINE1(epoll_create, int, size)
1673 return sys_epoll_create1(0);
1677 * The following function implements the controller interface for
1678 * the eventpoll file that enables the insertion/removal/change of
1679 * file descriptors inside the interest set.
1681 SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1682 struct epoll_event __user *, event)
1685 int did_lock_epmutex = 0;
1686 struct file *file, *tfile;
1687 struct eventpoll *ep;
1689 struct epoll_event epds;
1692 if (ep_op_has_event(op) &&
1693 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1696 /* Get the "struct file *" for the eventpoll file */
1702 /* Get the "struct file *" for the target file */
1707 /* The target file descriptor must support poll */
1709 if (!tfile->f_op || !tfile->f_op->poll)
1710 goto error_tgt_fput;
1712 /* Check if EPOLLWAKEUP is allowed */
1713 if ((epds.events & EPOLLWAKEUP) && !capable(CAP_BLOCK_SUSPEND))
1714 epds.events &= ~EPOLLWAKEUP;
1717 * We have to check that the file structure underneath the file descriptor
1718 * the user passed to us _is_ an eventpoll file. And also we do not permit
1719 * adding an epoll file descriptor inside itself.
1722 if (file == tfile || !is_file_epoll(file))
1723 goto error_tgt_fput;
1726 * At this point it is safe to assume that the "private_data" contains
1727 * our own data structure.
1729 ep = file->private_data;
1732 * When we insert an epoll file descriptor, inside another epoll file
1733 * descriptor, there is the change of creating closed loops, which are
1734 * better be handled here, than in more critical paths. While we are
1735 * checking for loops we also determine the list of files reachable
1736 * and hang them on the tfile_check_list, so we can check that we
1737 * haven't created too many possible wakeup paths.
1739 * We need to hold the epmutex across both ep_insert and ep_remove
1740 * b/c we want to make sure we are looking at a coherent view of
1743 if (op == EPOLL_CTL_ADD || op == EPOLL_CTL_DEL) {
1744 mutex_lock(&epmutex);
1745 did_lock_epmutex = 1;
1747 if (op == EPOLL_CTL_ADD) {
1748 if (is_file_epoll(tfile)) {
1750 if (ep_loop_check(ep, tfile) != 0) {
1751 clear_tfile_check_list();
1752 goto error_tgt_fput;
1755 list_add(&tfile->f_tfile_llink, &tfile_check_list);
1758 mutex_lock_nested(&ep->mtx, 0);
1761 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1762 * above, we can be sure to be able to use the item looked up by
1763 * ep_find() till we release the mutex.
1765 epi = ep_find(ep, tfile, fd);
1771 epds.events |= POLLERR | POLLHUP;
1772 error = ep_insert(ep, &epds, tfile, fd);
1775 clear_tfile_check_list();
1779 error = ep_remove(ep, epi);
1785 epds.events |= POLLERR | POLLHUP;
1786 error = ep_modify(ep, epi, &epds);
1791 mutex_unlock(&ep->mtx);
1794 if (did_lock_epmutex)
1795 mutex_unlock(&epmutex);
1806 * Implement the event wait interface for the eventpoll file. It is the kernel
1807 * part of the user space epoll_wait(2).
1809 SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1810 int, maxevents, int, timeout)
1814 struct eventpoll *ep;
1816 /* The maximum number of event must be greater than zero */
1817 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1820 /* Verify that the area passed by the user is writeable */
1821 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1826 /* Get the "struct file *" for the eventpoll file */
1833 * We have to check that the file structure underneath the fd
1834 * the user passed to us _is_ an eventpoll file.
1837 if (!is_file_epoll(file))
1841 * At this point it is safe to assume that the "private_data" contains
1842 * our own data structure.
1844 ep = file->private_data;
1846 /* Time to fish for events ... */
1847 error = ep_poll(ep, events, maxevents, timeout);
1857 * Implement the event wait interface for the eventpoll file. It is the kernel
1858 * part of the user space epoll_pwait(2).
1860 SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1861 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1865 sigset_t ksigmask, sigsaved;
1868 * If the caller wants a certain signal mask to be set during the wait,
1872 if (sigsetsize != sizeof(sigset_t))
1874 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1876 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1877 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1880 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1883 * If we changed the signal mask, we need to restore the original one.
1884 * In case we've got a signal while waiting, we do not restore the
1885 * signal mask yet, and we allow do_signal() to deliver the signal on
1886 * the way back to userspace, before the signal mask is restored.
1889 if (error == -EINTR) {
1890 memcpy(¤t->saved_sigmask, &sigsaved,
1892 set_restore_sigmask();
1894 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1900 static int __init eventpoll_init(void)
1906 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1908 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
1910 BUG_ON(max_user_watches < 0);
1913 * Initialize the structure used to perform epoll file descriptor
1914 * inclusion loops checks.
1916 ep_nested_calls_init(&poll_loop_ncalls);
1918 /* Initialize the structure used to perform safe poll wait head wake ups */
1919 ep_nested_calls_init(&poll_safewake_ncalls);
1921 /* Initialize the structure used to perform file's f_op->poll() calls */
1922 ep_nested_calls_init(&poll_readywalk_ncalls);
1924 /* Allocates slab cache used to allocate "struct epitem" items */
1925 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
1926 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1928 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1929 pwq_cache = kmem_cache_create("eventpoll_pwq",
1930 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
1934 fs_initcall(eventpoll_init);