~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/eventpoll.c

Version: ~ [ linux-5.15-rc7 ] ~ [ linux-5.14.14 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.75 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.155 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.213 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.252 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.287 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.289 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  fs/eventpoll.c (Efficient event retrieval implementation)
  3  *  Copyright (C) 2001,...,2009  Davide Libenzi
  4  *
  5  *  This program is free software; you can redistribute it and/or modify
  6  *  it under the terms of the GNU General Public License as published by
  7  *  the Free Software Foundation; either version 2 of the License, or
  8  *  (at your option) any later version.
  9  *
 10  *  Davide Libenzi <davidel@xmailserver.org>
 11  *
 12  */
 13 
 14 #include <linux/init.h>
 15 #include <linux/kernel.h>
 16 #include <linux/sched.h>
 17 #include <linux/fs.h>
 18 #include <linux/file.h>
 19 #include <linux/signal.h>
 20 #include <linux/errno.h>
 21 #include <linux/mm.h>
 22 #include <linux/slab.h>
 23 #include <linux/poll.h>
 24 #include <linux/string.h>
 25 #include <linux/list.h>
 26 #include <linux/hash.h>
 27 #include <linux/spinlock.h>
 28 #include <linux/syscalls.h>
 29 #include <linux/rbtree.h>
 30 #include <linux/wait.h>
 31 #include <linux/eventpoll.h>
 32 #include <linux/mount.h>
 33 #include <linux/bitops.h>
 34 #include <linux/mutex.h>
 35 #include <linux/anon_inodes.h>
 36 #include <linux/device.h>
 37 #include <asm/uaccess.h>
 38 #include <asm/io.h>
 39 #include <asm/mman.h>
 40 #include <linux/atomic.h>
 41 #include <linux/proc_fs.h>
 42 #include <linux/seq_file.h>
 43 #include <linux/compat.h>
 44 #include <linux/rculist.h>
 45 
 46 /*
 47  * LOCKING:
 48  * There are three level of locking required by epoll :
 49  *
 50  * 1) epmutex (mutex)
 51  * 2) ep->mtx (mutex)
 52  * 3) ep->lock (spinlock)
 53  *
 54  * The acquire order is the one listed above, from 1 to 3.
 55  * We need a spinlock (ep->lock) because we manipulate objects
 56  * from inside the poll callback, that might be triggered from
 57  * a wake_up() that in turn might be called from IRQ context.
 58  * So we can't sleep inside the poll callback and hence we need
 59  * a spinlock. During the event transfer loop (from kernel to
 60  * user space) we could end up sleeping due a copy_to_user(), so
 61  * we need a lock that will allow us to sleep. This lock is a
 62  * mutex (ep->mtx). It is acquired during the event transfer loop,
 63  * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
 64  * Then we also need a global mutex to serialize eventpoll_release_file()
 65  * and ep_free().
 66  * This mutex is acquired by ep_free() during the epoll file
 67  * cleanup path and it is also acquired by eventpoll_release_file()
 68  * if a file has been pushed inside an epoll set and it is then
 69  * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
 70  * It is also acquired when inserting an epoll fd onto another epoll
 71  * fd. We do this so that we walk the epoll tree and ensure that this
 72  * insertion does not create a cycle of epoll file descriptors, which
 73  * could lead to deadlock. We need a global mutex to prevent two
 74  * simultaneous inserts (A into B and B into A) from racing and
 75  * constructing a cycle without either insert observing that it is
 76  * going to.
 77  * It is necessary to acquire multiple "ep->mtx"es at once in the
 78  * case when one epoll fd is added to another. In this case, we
 79  * always acquire the locks in the order of nesting (i.e. after
 80  * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
 81  * before e2->mtx). Since we disallow cycles of epoll file
 82  * descriptors, this ensures that the mutexes are well-ordered. In
 83  * order to communicate this nesting to lockdep, when walking a tree
 84  * of epoll file descriptors, we use the current recursion depth as
 85  * the lockdep subkey.
 86  * It is possible to drop the "ep->mtx" and to use the global
 87  * mutex "epmutex" (together with "ep->lock") to have it working,
 88  * but having "ep->mtx" will make the interface more scalable.
 89  * Events that require holding "epmutex" are very rare, while for
 90  * normal operations the epoll private "ep->mtx" will guarantee
 91  * a better scalability.
 92  */
 93 
 94 /* Epoll private bits inside the event mask */
 95 #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET)
 96 
 97 /* Maximum number of nesting allowed inside epoll sets */
 98 #define EP_MAX_NESTS 4
 99 
100 #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
101 
102 #define EP_UNACTIVE_PTR ((void *) -1L)
103 
104 #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
105 
106 struct epoll_filefd {
107         struct file *file;
108         int fd;
109 } __packed;
110 
111 /*
112  * Structure used to track possible nested calls, for too deep recursions
113  * and loop cycles.
114  */
115 struct nested_call_node {
116         struct list_head llink;
117         void *cookie;
118         void *ctx;
119 };
120 
121 /*
122  * This structure is used as collector for nested calls, to check for
123  * maximum recursion dept and loop cycles.
124  */
125 struct nested_calls {
126         struct list_head tasks_call_list;
127         spinlock_t lock;
128 };
129 
130 /*
131  * Each file descriptor added to the eventpoll interface will
132  * have an entry of this type linked to the "rbr" RB tree.
133  * Avoid increasing the size of this struct, there can be many thousands
134  * of these on a server and we do not want this to take another cache line.
135  */
136 struct epitem {
137         union {
138                 /* RB tree node links this structure to the eventpoll RB tree */
139                 struct rb_node rbn;
140                 /* Used to free the struct epitem */
141                 struct rcu_head rcu;
142         };
143 
144         /* List header used to link this structure to the eventpoll ready list */
145         struct list_head rdllink;
146 
147         /*
148          * Works together "struct eventpoll"->ovflist in keeping the
149          * single linked chain of items.
150          */
151         struct epitem *next;
152 
153         /* The file descriptor information this item refers to */
154         struct epoll_filefd ffd;
155 
156         /* Number of active wait queue attached to poll operations */
157         int nwait;
158 
159         /* List containing poll wait queues */
160         struct list_head pwqlist;
161 
162         /* The "container" of this item */
163         struct eventpoll *ep;
164 
165         /* List header used to link this item to the "struct file" items list */
166         struct list_head fllink;
167 
168         /* wakeup_source used when EPOLLWAKEUP is set */
169         struct wakeup_source __rcu *ws;
170 
171         /* The structure that describe the interested events and the source fd */
172         struct epoll_event event;
173 };
174 
175 /*
176  * This structure is stored inside the "private_data" member of the file
177  * structure and represents the main data structure for the eventpoll
178  * interface.
179  */
180 struct eventpoll {
181         /* Protect the access to this structure */
182         spinlock_t lock;
183 
184         /*
185          * This mutex is used to ensure that files are not removed
186          * while epoll is using them. This is held during the event
187          * collection loop, the file cleanup path, the epoll file exit
188          * code and the ctl operations.
189          */
190         struct mutex mtx;
191 
192         /* Wait queue used by sys_epoll_wait() */
193         wait_queue_head_t wq;
194 
195         /* Wait queue used by file->poll() */
196         wait_queue_head_t poll_wait;
197 
198         /* List of ready file descriptors */
199         struct list_head rdllist;
200 
201         /* RB tree root used to store monitored fd structs */
202         struct rb_root rbr;
203 
204         /*
205          * This is a single linked list that chains all the "struct epitem" that
206          * happened while transferring ready events to userspace w/out
207          * holding ->lock.
208          */
209         struct epitem *ovflist;
210 
211         /* wakeup_source used when ep_scan_ready_list is running */
212         struct wakeup_source *ws;
213 
214         /* The user that created the eventpoll descriptor */
215         struct user_struct *user;
216 
217         struct file *file;
218 
219         /* used to optimize loop detection check */
220         int visited;
221         struct list_head visited_list_link;
222 };
223 
224 /* Wait structure used by the poll hooks */
225 struct eppoll_entry {
226         /* List header used to link this structure to the "struct epitem" */
227         struct list_head llink;
228 
229         /* The "base" pointer is set to the container "struct epitem" */
230         struct epitem *base;
231 
232         /*
233          * Wait queue item that will be linked to the target file wait
234          * queue head.
235          */
236         wait_queue_t wait;
237 
238         /* The wait queue head that linked the "wait" wait queue item */
239         wait_queue_head_t *whead;
240 };
241 
242 /* Wrapper struct used by poll queueing */
243 struct ep_pqueue {
244         poll_table pt;
245         struct epitem *epi;
246 };
247 
248 /* Used by the ep_send_events() function as callback private data */
249 struct ep_send_events_data {
250         int maxevents;
251         struct epoll_event __user *events;
252 };
253 
254 /*
255  * Configuration options available inside /proc/sys/fs/epoll/
256  */
257 /* Maximum number of epoll watched descriptors, per user */
258 static long max_user_watches __read_mostly;
259 
260 /*
261  * This mutex is used to serialize ep_free() and eventpoll_release_file().
262  */
263 static DEFINE_MUTEX(epmutex);
264 
265 /* Used to check for epoll file descriptor inclusion loops */
266 static struct nested_calls poll_loop_ncalls;
267 
268 /* Used for safe wake up implementation */
269 static struct nested_calls poll_safewake_ncalls;
270 
271 /* Used to call file's f_op->poll() under the nested calls boundaries */
272 static struct nested_calls poll_readywalk_ncalls;
273 
274 /* Slab cache used to allocate "struct epitem" */
275 static struct kmem_cache *epi_cache __read_mostly;
276 
277 /* Slab cache used to allocate "struct eppoll_entry" */
278 static struct kmem_cache *pwq_cache __read_mostly;
279 
280 /* Visited nodes during ep_loop_check(), so we can unset them when we finish */
281 static LIST_HEAD(visited_list);
282 
283 /*
284  * List of files with newly added links, where we may need to limit the number
285  * of emanating paths. Protected by the epmutex.
286  */
287 static LIST_HEAD(tfile_check_list);
288 
289 #ifdef CONFIG_SYSCTL
290 
291 #include <linux/sysctl.h>
292 
293 static long zero;
294 static long long_max = LONG_MAX;
295 
296 struct ctl_table epoll_table[] = {
297         {
298                 .procname       = "max_user_watches",
299                 .data           = &max_user_watches,
300                 .maxlen         = sizeof(max_user_watches),
301                 .mode           = 0644,
302                 .proc_handler   = proc_doulongvec_minmax,
303                 .extra1         = &zero,
304                 .extra2         = &long_max,
305         },
306         { }
307 };
308 #endif /* CONFIG_SYSCTL */
309 
310 static const struct file_operations eventpoll_fops;
311 
312 static inline int is_file_epoll(struct file *f)
313 {
314         return f->f_op == &eventpoll_fops;
315 }
316 
317 /* Setup the structure that is used as key for the RB tree */
318 static inline void ep_set_ffd(struct epoll_filefd *ffd,
319                               struct file *file, int fd)
320 {
321         ffd->file = file;
322         ffd->fd = fd;
323 }
324 
325 /* Compare RB tree keys */
326 static inline int ep_cmp_ffd(struct epoll_filefd *p1,
327                              struct epoll_filefd *p2)
328 {
329         return (p1->file > p2->file ? +1:
330                 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
331 }
332 
333 /* Tells us if the item is currently linked */
334 static inline int ep_is_linked(struct list_head *p)
335 {
336         return !list_empty(p);
337 }
338 
339 static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
340 {
341         return container_of(p, struct eppoll_entry, wait);
342 }
343 
344 /* Get the "struct epitem" from a wait queue pointer */
345 static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
346 {
347         return container_of(p, struct eppoll_entry, wait)->base;
348 }
349 
350 /* Get the "struct epitem" from an epoll queue wrapper */
351 static inline struct epitem *ep_item_from_epqueue(poll_table *p)
352 {
353         return container_of(p, struct ep_pqueue, pt)->epi;
354 }
355 
356 /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
357 static inline int ep_op_has_event(int op)
358 {
359         return op != EPOLL_CTL_DEL;
360 }
361 
362 /* Initialize the poll safe wake up structure */
363 static void ep_nested_calls_init(struct nested_calls *ncalls)
364 {
365         INIT_LIST_HEAD(&ncalls->tasks_call_list);
366         spin_lock_init(&ncalls->lock);
367 }
368 
369 /**
370  * ep_events_available - Checks if ready events might be available.
371  *
372  * @ep: Pointer to the eventpoll context.
373  *
374  * Returns: Returns a value different than zero if ready events are available,
375  *          or zero otherwise.
376  */
377 static inline int ep_events_available(struct eventpoll *ep)
378 {
379         return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
380 }
381 
382 /**
383  * ep_call_nested - Perform a bound (possibly) nested call, by checking
384  *                  that the recursion limit is not exceeded, and that
385  *                  the same nested call (by the meaning of same cookie) is
386  *                  no re-entered.
387  *
388  * @ncalls: Pointer to the nested_calls structure to be used for this call.
389  * @max_nests: Maximum number of allowed nesting calls.
390  * @nproc: Nested call core function pointer.
391  * @priv: Opaque data to be passed to the @nproc callback.
392  * @cookie: Cookie to be used to identify this nested call.
393  * @ctx: This instance context.
394  *
395  * Returns: Returns the code returned by the @nproc callback, or -1 if
396  *          the maximum recursion limit has been exceeded.
397  */
398 static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
399                           int (*nproc)(void *, void *, int), void *priv,
400                           void *cookie, void *ctx)
401 {
402         int error, call_nests = 0;
403         unsigned long flags;
404         struct list_head *lsthead = &ncalls->tasks_call_list;
405         struct nested_call_node *tncur;
406         struct nested_call_node tnode;
407 
408         spin_lock_irqsave(&ncalls->lock, flags);
409 
410         /*
411          * Try to see if the current task is already inside this wakeup call.
412          * We use a list here, since the population inside this set is always
413          * very much limited.
414          */
415         list_for_each_entry(tncur, lsthead, llink) {
416                 if (tncur->ctx == ctx &&
417                     (tncur->cookie == cookie || ++call_nests > max_nests)) {
418                         /*
419                          * Ops ... loop detected or maximum nest level reached.
420                          * We abort this wake by breaking the cycle itself.
421                          */
422                         error = -1;
423                         goto out_unlock;
424                 }
425         }
426 
427         /* Add the current task and cookie to the list */
428         tnode.ctx = ctx;
429         tnode.cookie = cookie;
430         list_add(&tnode.llink, lsthead);
431 
432         spin_unlock_irqrestore(&ncalls->lock, flags);
433 
434         /* Call the nested function */
435         error = (*nproc)(priv, cookie, call_nests);
436 
437         /* Remove the current task from the list */
438         spin_lock_irqsave(&ncalls->lock, flags);
439         list_del(&tnode.llink);
440 out_unlock:
441         spin_unlock_irqrestore(&ncalls->lock, flags);
442 
443         return error;
444 }
445 
446 /*
447  * As described in commit 0ccf831cb lockdep: annotate epoll
448  * the use of wait queues used by epoll is done in a very controlled
449  * manner. Wake ups can nest inside each other, but are never done
450  * with the same locking. For example:
451  *
452  *   dfd = socket(...);
453  *   efd1 = epoll_create();
454  *   efd2 = epoll_create();
455  *   epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
456  *   epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
457  *
458  * When a packet arrives to the device underneath "dfd", the net code will
459  * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
460  * callback wakeup entry on that queue, and the wake_up() performed by the
461  * "dfd" net code will end up in ep_poll_callback(). At this point epoll
462  * (efd1) notices that it may have some event ready, so it needs to wake up
463  * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
464  * that ends up in another wake_up(), after having checked about the
465  * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
466  * avoid stack blasting.
467  *
468  * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
469  * this special case of epoll.
470  */
471 #ifdef CONFIG_DEBUG_LOCK_ALLOC
472 static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
473                                      unsigned long events, int subclass)
474 {
475         unsigned long flags;
476 
477         spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
478         wake_up_locked_poll(wqueue, events);
479         spin_unlock_irqrestore(&wqueue->lock, flags);
480 }
481 #else
482 static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
483                                      unsigned long events, int subclass)
484 {
485         wake_up_poll(wqueue, events);
486 }
487 #endif
488 
489 static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
490 {
491         ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
492                           1 + call_nests);
493         return 0;
494 }
495 
496 /*
497  * Perform a safe wake up of the poll wait list. The problem is that
498  * with the new callback'd wake up system, it is possible that the
499  * poll callback is reentered from inside the call to wake_up() done
500  * on the poll wait queue head. The rule is that we cannot reenter the
501  * wake up code from the same task more than EP_MAX_NESTS times,
502  * and we cannot reenter the same wait queue head at all. This will
503  * enable to have a hierarchy of epoll file descriptor of no more than
504  * EP_MAX_NESTS deep.
505  */
506 static void ep_poll_safewake(wait_queue_head_t *wq)
507 {
508         int this_cpu = get_cpu();
509 
510         ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
511                        ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
512 
513         put_cpu();
514 }
515 
516 static void ep_remove_wait_queue(struct eppoll_entry *pwq)
517 {
518         wait_queue_head_t *whead;
519 
520         rcu_read_lock();
521         /*
522          * If it is cleared by POLLFREE, it should be rcu-safe.
523          * If we read NULL we need a barrier paired with
524          * smp_store_release() in ep_poll_callback(), otherwise
525          * we rely on whead->lock.
526          */
527         whead = smp_load_acquire(&pwq->whead);
528         if (whead)
529                 remove_wait_queue(whead, &pwq->wait);
530         rcu_read_unlock();
531 }
532 
533 /*
534  * This function unregisters poll callbacks from the associated file
535  * descriptor.  Must be called with "mtx" held (or "epmutex" if called from
536  * ep_free).
537  */
538 static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
539 {
540         struct list_head *lsthead = &epi->pwqlist;
541         struct eppoll_entry *pwq;
542 
543         while (!list_empty(lsthead)) {
544                 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
545 
546                 list_del(&pwq->llink);
547                 ep_remove_wait_queue(pwq);
548                 kmem_cache_free(pwq_cache, pwq);
549         }
550 }
551 
552 /* call only when ep->mtx is held */
553 static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
554 {
555         return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
556 }
557 
558 /* call only when ep->mtx is held */
559 static inline void ep_pm_stay_awake(struct epitem *epi)
560 {
561         struct wakeup_source *ws = ep_wakeup_source(epi);
562 
563         if (ws)
564                 __pm_stay_awake(ws);
565 }
566 
567 static inline bool ep_has_wakeup_source(struct epitem *epi)
568 {
569         return rcu_access_pointer(epi->ws) ? true : false;
570 }
571 
572 /* call when ep->mtx cannot be held (ep_poll_callback) */
573 static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
574 {
575         struct wakeup_source *ws;
576 
577         rcu_read_lock();
578         ws = rcu_dereference(epi->ws);
579         if (ws)
580                 __pm_stay_awake(ws);
581         rcu_read_unlock();
582 }
583 
584 /**
585  * ep_scan_ready_list - Scans the ready list in a way that makes possible for
586  *                      the scan code, to call f_op->poll(). Also allows for
587  *                      O(NumReady) performance.
588  *
589  * @ep: Pointer to the epoll private data structure.
590  * @sproc: Pointer to the scan callback.
591  * @priv: Private opaque data passed to the @sproc callback.
592  * @depth: The current depth of recursive f_op->poll calls.
593  * @ep_locked: caller already holds ep->mtx
594  *
595  * Returns: The same integer error code returned by the @sproc callback.
596  */
597 static int ep_scan_ready_list(struct eventpoll *ep,
598                               int (*sproc)(struct eventpoll *,
599                                            struct list_head *, void *),
600                               void *priv, int depth, bool ep_locked)
601 {
602         int error, pwake = 0;
603         unsigned long flags;
604         struct epitem *epi, *nepi;
605         LIST_HEAD(txlist);
606 
607         /*
608          * We need to lock this because we could be hit by
609          * eventpoll_release_file() and epoll_ctl().
610          */
611 
612         if (!ep_locked)
613                 mutex_lock_nested(&ep->mtx, depth);
614 
615         /*
616          * Steal the ready list, and re-init the original one to the
617          * empty list. Also, set ep->ovflist to NULL so that events
618          * happening while looping w/out locks, are not lost. We cannot
619          * have the poll callback to queue directly on ep->rdllist,
620          * because we want the "sproc" callback to be able to do it
621          * in a lockless way.
622          */
623         spin_lock_irqsave(&ep->lock, flags);
624         list_splice_init(&ep->rdllist, &txlist);
625         ep->ovflist = NULL;
626         spin_unlock_irqrestore(&ep->lock, flags);
627 
628         /*
629          * Now call the callback function.
630          */
631         error = (*sproc)(ep, &txlist, priv);
632 
633         spin_lock_irqsave(&ep->lock, flags);
634         /*
635          * During the time we spent inside the "sproc" callback, some
636          * other events might have been queued by the poll callback.
637          * We re-insert them inside the main ready-list here.
638          */
639         for (nepi = ep->ovflist; (epi = nepi) != NULL;
640              nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
641                 /*
642                  * We need to check if the item is already in the list.
643                  * During the "sproc" callback execution time, items are
644                  * queued into ->ovflist but the "txlist" might already
645                  * contain them, and the list_splice() below takes care of them.
646                  */
647                 if (!ep_is_linked(&epi->rdllink)) {
648                         list_add_tail(&epi->rdllink, &ep->rdllist);
649                         ep_pm_stay_awake(epi);
650                 }
651         }
652         /*
653          * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
654          * releasing the lock, events will be queued in the normal way inside
655          * ep->rdllist.
656          */
657         ep->ovflist = EP_UNACTIVE_PTR;
658 
659         /*
660          * Quickly re-inject items left on "txlist".
661          */
662         list_splice(&txlist, &ep->rdllist);
663         __pm_relax(ep->ws);
664 
665         if (!list_empty(&ep->rdllist)) {
666                 /*
667                  * Wake up (if active) both the eventpoll wait list and
668                  * the ->poll() wait list (delayed after we release the lock).
669                  */
670                 if (waitqueue_active(&ep->wq))
671                         wake_up_locked(&ep->wq);
672                 if (waitqueue_active(&ep->poll_wait))
673                         pwake++;
674         }
675         spin_unlock_irqrestore(&ep->lock, flags);
676 
677         if (!ep_locked)
678                 mutex_unlock(&ep->mtx);
679 
680         /* We have to call this outside the lock */
681         if (pwake)
682                 ep_poll_safewake(&ep->poll_wait);
683 
684         return error;
685 }
686 
687 static void epi_rcu_free(struct rcu_head *head)
688 {
689         struct epitem *epi = container_of(head, struct epitem, rcu);
690         kmem_cache_free(epi_cache, epi);
691 }
692 
693 /*
694  * Removes a "struct epitem" from the eventpoll RB tree and deallocates
695  * all the associated resources. Must be called with "mtx" held.
696  */
697 static int ep_remove(struct eventpoll *ep, struct epitem *epi)
698 {
699         unsigned long flags;
700         struct file *file = epi->ffd.file;
701 
702         /*
703          * Removes poll wait queue hooks. We _have_ to do this without holding
704          * the "ep->lock" otherwise a deadlock might occur. This because of the
705          * sequence of the lock acquisition. Here we do "ep->lock" then the wait
706          * queue head lock when unregistering the wait queue. The wakeup callback
707          * will run by holding the wait queue head lock and will call our callback
708          * that will try to get "ep->lock".
709          */
710         ep_unregister_pollwait(ep, epi);
711 
712         /* Remove the current item from the list of epoll hooks */
713         spin_lock(&file->f_lock);
714         list_del_rcu(&epi->fllink);
715         spin_unlock(&file->f_lock);
716 
717         rb_erase(&epi->rbn, &ep->rbr);
718 
719         spin_lock_irqsave(&ep->lock, flags);
720         if (ep_is_linked(&epi->rdllink))
721                 list_del_init(&epi->rdllink);
722         spin_unlock_irqrestore(&ep->lock, flags);
723 
724         wakeup_source_unregister(ep_wakeup_source(epi));
725         /*
726          * At this point it is safe to free the eventpoll item. Use the union
727          * field epi->rcu, since we are trying to minimize the size of
728          * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
729          * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
730          * use of the rbn field.
731          */
732         call_rcu(&epi->rcu, epi_rcu_free);
733 
734         atomic_long_dec(&ep->user->epoll_watches);
735 
736         return 0;
737 }
738 
739 static void ep_free(struct eventpoll *ep)
740 {
741         struct rb_node *rbp;
742         struct epitem *epi;
743 
744         /* We need to release all tasks waiting for these file */
745         if (waitqueue_active(&ep->poll_wait))
746                 ep_poll_safewake(&ep->poll_wait);
747 
748         /*
749          * We need to lock this because we could be hit by
750          * eventpoll_release_file() while we're freeing the "struct eventpoll".
751          * We do not need to hold "ep->mtx" here because the epoll file
752          * is on the way to be removed and no one has references to it
753          * anymore. The only hit might come from eventpoll_release_file() but
754          * holding "epmutex" is sufficient here.
755          */
756         mutex_lock(&epmutex);
757 
758         /*
759          * Walks through the whole tree by unregistering poll callbacks.
760          */
761         for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
762                 epi = rb_entry(rbp, struct epitem, rbn);
763 
764                 ep_unregister_pollwait(ep, epi);
765                 cond_resched();
766         }
767 
768         /*
769          * Walks through the whole tree by freeing each "struct epitem". At this
770          * point we are sure no poll callbacks will be lingering around, and also by
771          * holding "epmutex" we can be sure that no file cleanup code will hit
772          * us during this operation. So we can avoid the lock on "ep->lock".
773          * We do not need to lock ep->mtx, either, we only do it to prevent
774          * a lockdep warning.
775          */
776         mutex_lock(&ep->mtx);
777         while ((rbp = rb_first(&ep->rbr)) != NULL) {
778                 epi = rb_entry(rbp, struct epitem, rbn);
779                 ep_remove(ep, epi);
780                 cond_resched();
781         }
782         mutex_unlock(&ep->mtx);
783 
784         mutex_unlock(&epmutex);
785         mutex_destroy(&ep->mtx);
786         free_uid(ep->user);
787         wakeup_source_unregister(ep->ws);
788         kfree(ep);
789 }
790 
791 static int ep_eventpoll_release(struct inode *inode, struct file *file)
792 {
793         struct eventpoll *ep = file->private_data;
794 
795         if (ep)
796                 ep_free(ep);
797 
798         return 0;
799 }
800 
801 static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt)
802 {
803         pt->_key = epi->event.events;
804 
805         return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events;
806 }
807 
808 static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
809                                void *priv)
810 {
811         struct epitem *epi, *tmp;
812         poll_table pt;
813 
814         init_poll_funcptr(&pt, NULL);
815 
816         list_for_each_entry_safe(epi, tmp, head, rdllink) {
817                 if (ep_item_poll(epi, &pt))
818                         return POLLIN | POLLRDNORM;
819                 else {
820                         /*
821                          * Item has been dropped into the ready list by the poll
822                          * callback, but it's not actually ready, as far as
823                          * caller requested events goes. We can remove it here.
824                          */
825                         __pm_relax(ep_wakeup_source(epi));
826                         list_del_init(&epi->rdllink);
827                 }
828         }
829 
830         return 0;
831 }
832 
833 static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
834                                  poll_table *pt);
835 
836 struct readyevents_arg {
837         struct eventpoll *ep;
838         bool locked;
839 };
840 
841 static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
842 {
843         struct readyevents_arg *arg = priv;
844 
845         return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL,
846                                   call_nests + 1, arg->locked);
847 }
848 
849 static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
850 {
851         int pollflags;
852         struct eventpoll *ep = file->private_data;
853         struct readyevents_arg arg;
854 
855         /*
856          * During ep_insert() we already hold the ep->mtx for the tfile.
857          * Prevent re-aquisition.
858          */
859         arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc);
860         arg.ep = ep;
861 
862         /* Insert inside our poll wait queue */
863         poll_wait(file, &ep->poll_wait, wait);
864 
865         /*
866          * Proceed to find out if wanted events are really available inside
867          * the ready list. This need to be done under ep_call_nested()
868          * supervision, since the call to f_op->poll() done on listed files
869          * could re-enter here.
870          */
871         pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
872                                    ep_poll_readyevents_proc, &arg, ep, current);
873 
874         return pollflags != -1 ? pollflags : 0;
875 }
876 
877 #ifdef CONFIG_PROC_FS
878 static int ep_show_fdinfo(struct seq_file *m, struct file *f)
879 {
880         struct eventpoll *ep = f->private_data;
881         struct rb_node *rbp;
882         int ret = 0;
883 
884         mutex_lock(&ep->mtx);
885         for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
886                 struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
887 
888                 ret = seq_printf(m, "tfd: %8d events: %8x data: %16llx\n",
889                                  epi->ffd.fd, epi->event.events,
890                                  (long long)epi->event.data);
891                 if (ret)
892                         break;
893         }
894         mutex_unlock(&ep->mtx);
895 
896         return ret;
897 }
898 #endif
899 
900 /* File callbacks that implement the eventpoll file behaviour */
901 static const struct file_operations eventpoll_fops = {
902 #ifdef CONFIG_PROC_FS
903         .show_fdinfo    = ep_show_fdinfo,
904 #endif
905         .release        = ep_eventpoll_release,
906         .poll           = ep_eventpoll_poll,
907         .llseek         = noop_llseek,
908 };
909 
910 /*
911  * This is called from eventpoll_release() to unlink files from the eventpoll
912  * interface. We need to have this facility to cleanup correctly files that are
913  * closed without being removed from the eventpoll interface.
914  */
915 void eventpoll_release_file(struct file *file)
916 {
917         struct eventpoll *ep;
918         struct epitem *epi, *next;
919 
920         /*
921          * We don't want to get "file->f_lock" because it is not
922          * necessary. It is not necessary because we're in the "struct file"
923          * cleanup path, and this means that no one is using this file anymore.
924          * So, for example, epoll_ctl() cannot hit here since if we reach this
925          * point, the file counter already went to zero and fget() would fail.
926          * The only hit might come from ep_free() but by holding the mutex
927          * will correctly serialize the operation. We do need to acquire
928          * "ep->mtx" after "epmutex" because ep_remove() requires it when called
929          * from anywhere but ep_free().
930          *
931          * Besides, ep_remove() acquires the lock, so we can't hold it here.
932          */
933         mutex_lock(&epmutex);
934         list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) {
935                 ep = epi->ep;
936                 mutex_lock_nested(&ep->mtx, 0);
937                 ep_remove(ep, epi);
938                 mutex_unlock(&ep->mtx);
939         }
940         mutex_unlock(&epmutex);
941 }
942 
943 static int ep_alloc(struct eventpoll **pep)
944 {
945         int error;
946         struct user_struct *user;
947         struct eventpoll *ep;
948 
949         user = get_current_user();
950         error = -ENOMEM;
951         ep = kzalloc(sizeof(*ep), GFP_KERNEL);
952         if (unlikely(!ep))
953                 goto free_uid;
954 
955         spin_lock_init(&ep->lock);
956         mutex_init(&ep->mtx);
957         init_waitqueue_head(&ep->wq);
958         init_waitqueue_head(&ep->poll_wait);
959         INIT_LIST_HEAD(&ep->rdllist);
960         ep->rbr = RB_ROOT;
961         ep->ovflist = EP_UNACTIVE_PTR;
962         ep->user = user;
963 
964         *pep = ep;
965 
966         return 0;
967 
968 free_uid:
969         free_uid(user);
970         return error;
971 }
972 
973 /*
974  * Search the file inside the eventpoll tree. The RB tree operations
975  * are protected by the "mtx" mutex, and ep_find() must be called with
976  * "mtx" held.
977  */
978 static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
979 {
980         int kcmp;
981         struct rb_node *rbp;
982         struct epitem *epi, *epir = NULL;
983         struct epoll_filefd ffd;
984 
985         ep_set_ffd(&ffd, file, fd);
986         for (rbp = ep->rbr.rb_node; rbp; ) {
987                 epi = rb_entry(rbp, struct epitem, rbn);
988                 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
989                 if (kcmp > 0)
990                         rbp = rbp->rb_right;
991                 else if (kcmp < 0)
992                         rbp = rbp->rb_left;
993                 else {
994                         epir = epi;
995                         break;
996                 }
997         }
998 
999         return epir;
1000 }
1001 
1002 /*
1003  * This is the callback that is passed to the wait queue wakeup
1004  * mechanism. It is called by the stored file descriptors when they
1005  * have events to report.
1006  */
1007 static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
1008 {
1009         int pwake = 0;
1010         unsigned long flags;
1011         struct epitem *epi = ep_item_from_wait(wait);
1012         struct eventpoll *ep = epi->ep;
1013 
1014         spin_lock_irqsave(&ep->lock, flags);
1015 
1016         /*
1017          * If the event mask does not contain any poll(2) event, we consider the
1018          * descriptor to be disabled. This condition is likely the effect of the
1019          * EPOLLONESHOT bit that disables the descriptor when an event is received,
1020          * until the next EPOLL_CTL_MOD will be issued.
1021          */
1022         if (!(epi->event.events & ~EP_PRIVATE_BITS))
1023                 goto out_unlock;
1024 
1025         /*
1026          * Check the events coming with the callback. At this stage, not
1027          * every device reports the events in the "key" parameter of the
1028          * callback. We need to be able to handle both cases here, hence the
1029          * test for "key" != NULL before the event match test.
1030          */
1031         if (key && !((unsigned long) key & epi->event.events))
1032                 goto out_unlock;
1033 
1034         /*
1035          * If we are transferring events to userspace, we can hold no locks
1036          * (because we're accessing user memory, and because of linux f_op->poll()
1037          * semantics). All the events that happen during that period of time are
1038          * chained in ep->ovflist and requeued later on.
1039          */
1040         if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
1041                 if (epi->next == EP_UNACTIVE_PTR) {
1042                         epi->next = ep->ovflist;
1043                         ep->ovflist = epi;
1044                         if (epi->ws) {
1045                                 /*
1046                                  * Activate ep->ws since epi->ws may get
1047                                  * deactivated at any time.
1048                                  */
1049                                 __pm_stay_awake(ep->ws);
1050                         }
1051 
1052                 }
1053                 goto out_unlock;
1054         }
1055 
1056         /* If this file is already in the ready list we exit soon */
1057         if (!ep_is_linked(&epi->rdllink)) {
1058                 list_add_tail(&epi->rdllink, &ep->rdllist);
1059                 ep_pm_stay_awake_rcu(epi);
1060         }
1061 
1062         /*
1063          * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1064          * wait list.
1065          */
1066         if (waitqueue_active(&ep->wq))
1067                 wake_up_locked(&ep->wq);
1068         if (waitqueue_active(&ep->poll_wait))
1069                 pwake++;
1070 
1071 out_unlock:
1072         spin_unlock_irqrestore(&ep->lock, flags);
1073 
1074         /* We have to call this outside the lock */
1075         if (pwake)
1076                 ep_poll_safewake(&ep->poll_wait);
1077 
1078         if ((unsigned long)key & POLLFREE) {
1079                 /*
1080                  * If we race with ep_remove_wait_queue() it can miss
1081                  * ->whead = NULL and do another remove_wait_queue() after
1082                  * us, so we can't use __remove_wait_queue().
1083                  */
1084                 list_del_init(&wait->task_list);
1085                 /*
1086                  * ->whead != NULL protects us from the race with ep_free()
1087                  * or ep_remove(), ep_remove_wait_queue() takes whead->lock
1088                  * held by the caller. Once we nullify it, nothing protects
1089                  * ep/epi or even wait.
1090                  */
1091                 smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL);
1092         }
1093 
1094         return 1;
1095 }
1096 
1097 /*
1098  * This is the callback that is used to add our wait queue to the
1099  * target file wakeup lists.
1100  */
1101 static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1102                                  poll_table *pt)
1103 {
1104         struct epitem *epi = ep_item_from_epqueue(pt);
1105         struct eppoll_entry *pwq;
1106 
1107         if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
1108                 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1109                 pwq->whead = whead;
1110                 pwq->base = epi;
1111                 add_wait_queue(whead, &pwq->wait);
1112                 list_add_tail(&pwq->llink, &epi->pwqlist);
1113                 epi->nwait++;
1114         } else {
1115                 /* We have to signal that an error occurred */
1116                 epi->nwait = -1;
1117         }
1118 }
1119 
1120 static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1121 {
1122         int kcmp;
1123         struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
1124         struct epitem *epic;
1125 
1126         while (*p) {
1127                 parent = *p;
1128                 epic = rb_entry(parent, struct epitem, rbn);
1129                 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1130                 if (kcmp > 0)
1131                         p = &parent->rb_right;
1132                 else
1133                         p = &parent->rb_left;
1134         }
1135         rb_link_node(&epi->rbn, parent, p);
1136         rb_insert_color(&epi->rbn, &ep->rbr);
1137 }
1138 
1139 
1140 
1141 #define PATH_ARR_SIZE 5
1142 /*
1143  * These are the number paths of length 1 to 5, that we are allowing to emanate
1144  * from a single file of interest. For example, we allow 1000 paths of length
1145  * 1, to emanate from each file of interest. This essentially represents the
1146  * potential wakeup paths, which need to be limited in order to avoid massive
1147  * uncontrolled wakeup storms. The common use case should be a single ep which
1148  * is connected to n file sources. In this case each file source has 1 path
1149  * of length 1. Thus, the numbers below should be more than sufficient. These
1150  * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1151  * and delete can't add additional paths. Protected by the epmutex.
1152  */
1153 static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1154 static int path_count[PATH_ARR_SIZE];
1155 
1156 static int path_count_inc(int nests)
1157 {
1158         /* Allow an arbitrary number of depth 1 paths */
1159         if (nests == 0)
1160                 return 0;
1161 
1162         if (++path_count[nests] > path_limits[nests])
1163                 return -1;
1164         return 0;
1165 }
1166 
1167 static void path_count_init(void)
1168 {
1169         int i;
1170 
1171         for (i = 0; i < PATH_ARR_SIZE; i++)
1172                 path_count[i] = 0;
1173 }
1174 
1175 static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1176 {
1177         int error = 0;
1178         struct file *file = priv;
1179         struct file *child_file;
1180         struct epitem *epi;
1181 
1182         /* CTL_DEL can remove links here, but that can't increase our count */
1183         rcu_read_lock();
1184         list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
1185                 child_file = epi->ep->file;
1186                 if (is_file_epoll(child_file)) {
1187                         if (list_empty(&child_file->f_ep_links)) {
1188                                 if (path_count_inc(call_nests)) {
1189                                         error = -1;
1190                                         break;
1191                                 }
1192                         } else {
1193                                 error = ep_call_nested(&poll_loop_ncalls,
1194                                                         EP_MAX_NESTS,
1195                                                         reverse_path_check_proc,
1196                                                         child_file, child_file,
1197                                                         current);
1198                         }
1199                         if (error != 0)
1200                                 break;
1201                 } else {
1202                         printk(KERN_ERR "reverse_path_check_proc: "
1203                                 "file is not an ep!\n");
1204                 }
1205         }
1206         rcu_read_unlock();
1207         return error;
1208 }
1209 
1210 /**
1211  * reverse_path_check - The tfile_check_list is list of file *, which have
1212  *                      links that are proposed to be newly added. We need to
1213  *                      make sure that those added links don't add too many
1214  *                      paths such that we will spend all our time waking up
1215  *                      eventpoll objects.
1216  *
1217  * Returns: Returns zero if the proposed links don't create too many paths,
1218  *          -1 otherwise.
1219  */
1220 static int reverse_path_check(void)
1221 {
1222         int error = 0;
1223         struct file *current_file;
1224 
1225         /* let's call this for all tfiles */
1226         list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1227                 path_count_init();
1228                 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1229                                         reverse_path_check_proc, current_file,
1230                                         current_file, current);
1231                 if (error)
1232                         break;
1233         }
1234         return error;
1235 }
1236 
1237 static int ep_create_wakeup_source(struct epitem *epi)
1238 {
1239         const char *name;
1240         struct wakeup_source *ws;
1241 
1242         if (!epi->ep->ws) {
1243                 epi->ep->ws = wakeup_source_register("eventpoll");
1244                 if (!epi->ep->ws)
1245                         return -ENOMEM;
1246         }
1247 
1248         name = epi->ffd.file->f_path.dentry->d_name.name;
1249         ws = wakeup_source_register(name);
1250 
1251         if (!ws)
1252                 return -ENOMEM;
1253         rcu_assign_pointer(epi->ws, ws);
1254 
1255         return 0;
1256 }
1257 
1258 /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1259 static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1260 {
1261         struct wakeup_source *ws = ep_wakeup_source(epi);
1262 
1263         RCU_INIT_POINTER(epi->ws, NULL);
1264 
1265         /*
1266          * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1267          * used internally by wakeup_source_remove, too (called by
1268          * wakeup_source_unregister), so we cannot use call_rcu
1269          */
1270         synchronize_rcu();
1271         wakeup_source_unregister(ws);
1272 }
1273 
1274 /*
1275  * Must be called with "mtx" held.
1276  */
1277 static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1278                      struct file *tfile, int fd, int full_check)
1279 {
1280         int error, revents, pwake = 0;
1281         unsigned long flags;
1282         long user_watches;
1283         struct epitem *epi;
1284         struct ep_pqueue epq;
1285 
1286         user_watches = atomic_long_read(&ep->user->epoll_watches);
1287         if (unlikely(user_watches >= max_user_watches))
1288                 return -ENOSPC;
1289         if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
1290                 return -ENOMEM;
1291 
1292         /* Item initialization follow here ... */
1293         INIT_LIST_HEAD(&epi->rdllink);
1294         INIT_LIST_HEAD(&epi->fllink);
1295         INIT_LIST_HEAD(&epi->pwqlist);
1296         epi->ep = ep;
1297         ep_set_ffd(&epi->ffd, tfile, fd);
1298         epi->event = *event;
1299         epi->nwait = 0;
1300         epi->next = EP_UNACTIVE_PTR;
1301         if (epi->event.events & EPOLLWAKEUP) {
1302                 error = ep_create_wakeup_source(epi);
1303                 if (error)
1304                         goto error_create_wakeup_source;
1305         } else {
1306                 RCU_INIT_POINTER(epi->ws, NULL);
1307         }
1308 
1309         /* Initialize the poll table using the queue callback */
1310         epq.epi = epi;
1311         init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1312 
1313         /*
1314          * Attach the item to the poll hooks and get current event bits.
1315          * We can safely use the file* here because its usage count has
1316          * been increased by the caller of this function. Note that after
1317          * this operation completes, the poll callback can start hitting
1318          * the new item.
1319          */
1320         revents = ep_item_poll(epi, &epq.pt);
1321 
1322         /*
1323          * We have to check if something went wrong during the poll wait queue
1324          * install process. Namely an allocation for a wait queue failed due
1325          * high memory pressure.
1326          */
1327         error = -ENOMEM;
1328         if (epi->nwait < 0)
1329                 goto error_unregister;
1330 
1331         /* Add the current item to the list of active epoll hook for this file */
1332         spin_lock(&tfile->f_lock);
1333         list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
1334         spin_unlock(&tfile->f_lock);
1335 
1336         /*
1337          * Add the current item to the RB tree. All RB tree operations are
1338          * protected by "mtx", and ep_insert() is called with "mtx" held.
1339          */
1340         ep_rbtree_insert(ep, epi);
1341 
1342         /* now check if we've created too many backpaths */
1343         error = -EINVAL;
1344         if (full_check && reverse_path_check())
1345                 goto error_remove_epi;
1346 
1347         /* We have to drop the new item inside our item list to keep track of it */
1348         spin_lock_irqsave(&ep->lock, flags);
1349 
1350         /* If the file is already "ready" we drop it inside the ready list */
1351         if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
1352                 list_add_tail(&epi->rdllink, &ep->rdllist);
1353                 ep_pm_stay_awake(epi);
1354 
1355                 /* Notify waiting tasks that events are available */
1356                 if (waitqueue_active(&ep->wq))
1357                         wake_up_locked(&ep->wq);
1358                 if (waitqueue_active(&ep->poll_wait))
1359                         pwake++;
1360         }
1361 
1362         spin_unlock_irqrestore(&ep->lock, flags);
1363 
1364         atomic_long_inc(&ep->user->epoll_watches);
1365 
1366         /* We have to call this outside the lock */
1367         if (pwake)
1368                 ep_poll_safewake(&ep->poll_wait);
1369 
1370         return 0;
1371 
1372 error_remove_epi:
1373         spin_lock(&tfile->f_lock);
1374         list_del_rcu(&epi->fllink);
1375         spin_unlock(&tfile->f_lock);
1376 
1377         rb_erase(&epi->rbn, &ep->rbr);
1378 
1379 error_unregister:
1380         ep_unregister_pollwait(ep, epi);
1381 
1382         /*
1383          * We need to do this because an event could have been arrived on some
1384          * allocated wait queue. Note that we don't care about the ep->ovflist
1385          * list, since that is used/cleaned only inside a section bound by "mtx".
1386          * And ep_insert() is called with "mtx" held.
1387          */
1388         spin_lock_irqsave(&ep->lock, flags);
1389         if (ep_is_linked(&epi->rdllink))
1390                 list_del_init(&epi->rdllink);
1391         spin_unlock_irqrestore(&ep->lock, flags);
1392 
1393         wakeup_source_unregister(ep_wakeup_source(epi));
1394 
1395 error_create_wakeup_source:
1396         kmem_cache_free(epi_cache, epi);
1397 
1398         return error;
1399 }
1400 
1401 /*
1402  * Modify the interest event mask by dropping an event if the new mask
1403  * has a match in the current file status. Must be called with "mtx" held.
1404  */
1405 static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1406 {
1407         int pwake = 0;
1408         unsigned int revents;
1409         poll_table pt;
1410 
1411         init_poll_funcptr(&pt, NULL);
1412 
1413         /*
1414          * Set the new event interest mask before calling f_op->poll();
1415          * otherwise we might miss an event that happens between the
1416          * f_op->poll() call and the new event set registering.
1417          */
1418         epi->event.events = event->events; /* need barrier below */
1419         epi->event.data = event->data; /* protected by mtx */
1420         if (epi->event.events & EPOLLWAKEUP) {
1421                 if (!ep_has_wakeup_source(epi))
1422                         ep_create_wakeup_source(epi);
1423         } else if (ep_has_wakeup_source(epi)) {
1424                 ep_destroy_wakeup_source(epi);
1425         }
1426 
1427         /*
1428          * The following barrier has two effects:
1429          *
1430          * 1) Flush epi changes above to other CPUs.  This ensures
1431          *    we do not miss events from ep_poll_callback if an
1432          *    event occurs immediately after we call f_op->poll().
1433          *    We need this because we did not take ep->lock while
1434          *    changing epi above (but ep_poll_callback does take
1435          *    ep->lock).
1436          *
1437          * 2) We also need to ensure we do not miss _past_ events
1438          *    when calling f_op->poll().  This barrier also
1439          *    pairs with the barrier in wq_has_sleeper (see
1440          *    comments for wq_has_sleeper).
1441          *
1442          * This barrier will now guarantee ep_poll_callback or f_op->poll
1443          * (or both) will notice the readiness of an item.
1444          */
1445         smp_mb();
1446 
1447         /*
1448          * Get current event bits. We can safely use the file* here because
1449          * its usage count has been increased by the caller of this function.
1450          */
1451         revents = ep_item_poll(epi, &pt);
1452 
1453         /*
1454          * If the item is "hot" and it is not registered inside the ready
1455          * list, push it inside.
1456          */
1457         if (revents & event->events) {
1458                 spin_lock_irq(&ep->lock);
1459                 if (!ep_is_linked(&epi->rdllink)) {
1460                         list_add_tail(&epi->rdllink, &ep->rdllist);
1461                         ep_pm_stay_awake(epi);
1462 
1463                         /* Notify waiting tasks that events are available */
1464                         if (waitqueue_active(&ep->wq))
1465                                 wake_up_locked(&ep->wq);
1466                         if (waitqueue_active(&ep->poll_wait))
1467                                 pwake++;
1468                 }
1469                 spin_unlock_irq(&ep->lock);
1470         }
1471 
1472         /* We have to call this outside the lock */
1473         if (pwake)
1474                 ep_poll_safewake(&ep->poll_wait);
1475 
1476         return 0;
1477 }
1478 
1479 static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1480                                void *priv)
1481 {
1482         struct ep_send_events_data *esed = priv;
1483         int eventcnt;
1484         unsigned int revents;
1485         struct epitem *epi;
1486         struct epoll_event __user *uevent;
1487         struct wakeup_source *ws;
1488         poll_table pt;
1489 
1490         init_poll_funcptr(&pt, NULL);
1491 
1492         /*
1493          * We can loop without lock because we are passed a task private list.
1494          * Items cannot vanish during the loop because ep_scan_ready_list() is
1495          * holding "mtx" during this call.
1496          */
1497         for (eventcnt = 0, uevent = esed->events;
1498              !list_empty(head) && eventcnt < esed->maxevents;) {
1499                 epi = list_first_entry(head, struct epitem, rdllink);
1500 
1501                 /*
1502                  * Activate ep->ws before deactivating epi->ws to prevent
1503                  * triggering auto-suspend here (in case we reactive epi->ws
1504                  * below).
1505                  *
1506                  * This could be rearranged to delay the deactivation of epi->ws
1507                  * instead, but then epi->ws would temporarily be out of sync
1508                  * with ep_is_linked().
1509                  */
1510                 ws = ep_wakeup_source(epi);
1511                 if (ws) {
1512                         if (ws->active)
1513                                 __pm_stay_awake(ep->ws);
1514                         __pm_relax(ws);
1515                 }
1516 
1517                 list_del_init(&epi->rdllink);
1518 
1519                 revents = ep_item_poll(epi, &pt);
1520 
1521                 /*
1522                  * If the event mask intersect the caller-requested one,
1523                  * deliver the event to userspace. Again, ep_scan_ready_list()
1524                  * is holding "mtx", so no operations coming from userspace
1525                  * can change the item.
1526                  */
1527                 if (revents) {
1528                         if (__put_user(revents, &uevent->events) ||
1529                             __put_user(epi->event.data, &uevent->data)) {
1530                                 list_add(&epi->rdllink, head);
1531                                 ep_pm_stay_awake(epi);
1532                                 return eventcnt ? eventcnt : -EFAULT;
1533                         }
1534                         eventcnt++;
1535                         uevent++;
1536                         if (epi->event.events & EPOLLONESHOT)
1537                                 epi->event.events &= EP_PRIVATE_BITS;
1538                         else if (!(epi->event.events & EPOLLET)) {
1539                                 /*
1540                                  * If this file has been added with Level
1541                                  * Trigger mode, we need to insert back inside
1542                                  * the ready list, so that the next call to
1543                                  * epoll_wait() will check again the events
1544                                  * availability. At this point, no one can insert
1545                                  * into ep->rdllist besides us. The epoll_ctl()
1546                                  * callers are locked out by
1547                                  * ep_scan_ready_list() holding "mtx" and the
1548                                  * poll callback will queue them in ep->ovflist.
1549                                  */
1550                                 list_add_tail(&epi->rdllink, &ep->rdllist);
1551                                 ep_pm_stay_awake(epi);
1552                         }
1553                 }
1554         }
1555 
1556         return eventcnt;
1557 }
1558 
1559 static int ep_send_events(struct eventpoll *ep,
1560                           struct epoll_event __user *events, int maxevents)
1561 {
1562         struct ep_send_events_data esed;
1563 
1564         esed.maxevents = maxevents;
1565         esed.events = events;
1566 
1567         return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
1568 }
1569 
1570 static inline struct timespec ep_set_mstimeout(long ms)
1571 {
1572         struct timespec now, ts = {
1573                 .tv_sec = ms / MSEC_PER_SEC,
1574                 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1575         };
1576 
1577         ktime_get_ts(&now);
1578         return timespec_add_safe(now, ts);
1579 }
1580 
1581 /**
1582  * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1583  *           event buffer.
1584  *
1585  * @ep: Pointer to the eventpoll context.
1586  * @events: Pointer to the userspace buffer where the ready events should be
1587  *          stored.
1588  * @maxevents: Size (in terms of number of events) of the caller event buffer.
1589  * @timeout: Maximum timeout for the ready events fetch operation, in
1590  *           milliseconds. If the @timeout is zero, the function will not block,
1591  *           while if the @timeout is less than zero, the function will block
1592  *           until at least one event has been retrieved (or an error
1593  *           occurred).
1594  *
1595  * Returns: Returns the number of ready events which have been fetched, or an
1596  *          error code, in case of error.
1597  */
1598 static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1599                    int maxevents, long timeout)
1600 {
1601         int res = 0, eavail, timed_out = 0;
1602         unsigned long flags;
1603         long slack = 0;
1604         wait_queue_t wait;
1605         ktime_t expires, *to = NULL;
1606 
1607         if (timeout > 0) {
1608                 struct timespec end_time = ep_set_mstimeout(timeout);
1609 
1610                 slack = select_estimate_accuracy(&end_time);
1611                 to = &expires;
1612                 *to = timespec_to_ktime(end_time);
1613         } else if (timeout == 0) {
1614                 /*
1615                  * Avoid the unnecessary trip to the wait queue loop, if the
1616                  * caller specified a non blocking operation.
1617                  */
1618                 timed_out = 1;
1619                 spin_lock_irqsave(&ep->lock, flags);
1620                 goto check_events;
1621         }
1622 
1623 fetch_events:
1624         spin_lock_irqsave(&ep->lock, flags);
1625 
1626         if (!ep_events_available(ep)) {
1627                 /*
1628                  * We don't have any available event to return to the caller.
1629                  * We need to sleep here, and we will be wake up by
1630                  * ep_poll_callback() when events will become available.
1631                  */
1632                 init_waitqueue_entry(&wait, current);
1633                 __add_wait_queue_exclusive(&ep->wq, &wait);
1634 
1635                 for (;;) {
1636                         /*
1637                          * We don't want to sleep if the ep_poll_callback() sends us
1638                          * a wakeup in between. That's why we set the task state
1639                          * to TASK_INTERRUPTIBLE before doing the checks.
1640                          */
1641                         set_current_state(TASK_INTERRUPTIBLE);
1642                         if (ep_events_available(ep) || timed_out)
1643                                 break;
1644                         if (signal_pending(current)) {
1645                                 res = -EINTR;
1646                                 break;
1647                         }
1648 
1649                         spin_unlock_irqrestore(&ep->lock, flags);
1650                         if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1651                                 timed_out = 1;
1652 
1653                         spin_lock_irqsave(&ep->lock, flags);
1654                 }
1655                 __remove_wait_queue(&ep->wq, &wait);
1656 
1657                 set_current_state(TASK_RUNNING);
1658         }
1659 check_events:
1660         /* Is it worth to try to dig for events ? */
1661         eavail = ep_events_available(ep);
1662 
1663         spin_unlock_irqrestore(&ep->lock, flags);
1664 
1665         /*
1666          * Try to transfer events to user space. In case we get 0 events and
1667          * there's still timeout left over, we go trying again in search of
1668          * more luck.
1669          */
1670         if (!res && eavail &&
1671             !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
1672                 goto fetch_events;
1673 
1674         return res;
1675 }
1676 
1677 /**
1678  * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1679  *                      API, to verify that adding an epoll file inside another
1680  *                      epoll structure, does not violate the constraints, in
1681  *                      terms of closed loops, or too deep chains (which can
1682  *                      result in excessive stack usage).
1683  *
1684  * @priv: Pointer to the epoll file to be currently checked.
1685  * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1686  *          data structure pointer.
1687  * @call_nests: Current dept of the @ep_call_nested() call stack.
1688  *
1689  * Returns: Returns zero if adding the epoll @file inside current epoll
1690  *          structure @ep does not violate the constraints, or -1 otherwise.
1691  */
1692 static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1693 {
1694         int error = 0;
1695         struct file *file = priv;
1696         struct eventpoll *ep = file->private_data;
1697         struct eventpoll *ep_tovisit;
1698         struct rb_node *rbp;
1699         struct epitem *epi;
1700 
1701         mutex_lock_nested(&ep->mtx, call_nests + 1);
1702         ep->visited = 1;
1703         list_add(&ep->visited_list_link, &visited_list);
1704         for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1705                 epi = rb_entry(rbp, struct epitem, rbn);
1706                 if (unlikely(is_file_epoll(epi->ffd.file))) {
1707                         ep_tovisit = epi->ffd.file->private_data;
1708                         if (ep_tovisit->visited)
1709                                 continue;
1710                         error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1711                                         ep_loop_check_proc, epi->ffd.file,
1712                                         ep_tovisit, current);
1713                         if (error != 0)
1714                                 break;
1715                 } else {
1716                         /*
1717                          * If we've reached a file that is not associated with
1718                          * an ep, then we need to check if the newly added
1719                          * links are going to add too many wakeup paths. We do
1720                          * this by adding it to the tfile_check_list, if it's
1721                          * not already there, and calling reverse_path_check()
1722                          * during ep_insert().
1723                          */
1724                         if (list_empty(&epi->ffd.file->f_tfile_llink))
1725                                 list_add(&epi->ffd.file->f_tfile_llink,
1726                                          &tfile_check_list);
1727                 }
1728         }
1729         mutex_unlock(&ep->mtx);
1730 
1731         return error;
1732 }
1733 
1734 /**
1735  * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1736  *                 another epoll file (represented by @ep) does not create
1737  *                 closed loops or too deep chains.
1738  *
1739  * @ep: Pointer to the epoll private data structure.
1740  * @file: Pointer to the epoll file to be checked.
1741  *
1742  * Returns: Returns zero if adding the epoll @file inside current epoll
1743  *          structure @ep does not violate the constraints, or -1 otherwise.
1744  */
1745 static int ep_loop_check(struct eventpoll *ep, struct file *file)
1746 {
1747         int ret;
1748         struct eventpoll *ep_cur, *ep_next;
1749 
1750         ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1751                               ep_loop_check_proc, file, ep, current);
1752         /* clear visited list */
1753         list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1754                                                         visited_list_link) {
1755                 ep_cur->visited = 0;
1756                 list_del(&ep_cur->visited_list_link);
1757         }
1758         return ret;
1759 }
1760 
1761 static void clear_tfile_check_list(void)
1762 {
1763         struct file *file;
1764 
1765         /* first clear the tfile_check_list */
1766         while (!list_empty(&tfile_check_list)) {
1767                 file = list_first_entry(&tfile_check_list, struct file,
1768                                         f_tfile_llink);
1769                 list_del_init(&file->f_tfile_llink);
1770         }
1771         INIT_LIST_HEAD(&tfile_check_list);
1772 }
1773 
1774 /*
1775  * Open an eventpoll file descriptor.
1776  */
1777 SYSCALL_DEFINE1(epoll_create1, int, flags)
1778 {
1779         int error, fd;
1780         struct eventpoll *ep = NULL;
1781         struct file *file;
1782 
1783         /* Check the EPOLL_* constant for consistency.  */
1784         BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1785 
1786         if (flags & ~EPOLL_CLOEXEC)
1787                 return -EINVAL;
1788         /*
1789          * Create the internal data structure ("struct eventpoll").
1790          */
1791         error = ep_alloc(&ep);
1792         if (error < 0)
1793                 return error;
1794         /*
1795          * Creates all the items needed to setup an eventpoll file. That is,
1796          * a file structure and a free file descriptor.
1797          */
1798         fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1799         if (fd < 0) {
1800                 error = fd;
1801                 goto out_free_ep;
1802         }
1803         file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
1804                                  O_RDWR | (flags & O_CLOEXEC));
1805         if (IS_ERR(file)) {
1806                 error = PTR_ERR(file);
1807                 goto out_free_fd;
1808         }
1809         ep->file = file;
1810         fd_install(fd, file);
1811         return fd;
1812 
1813 out_free_fd:
1814         put_unused_fd(fd);
1815 out_free_ep:
1816         ep_free(ep);
1817         return error;
1818 }
1819 
1820 SYSCALL_DEFINE1(epoll_create, int, size)
1821 {
1822         if (size <= 0)
1823                 return -EINVAL;
1824 
1825         return sys_epoll_create1(0);
1826 }
1827 
1828 /*
1829  * The following function implements the controller interface for
1830  * the eventpoll file that enables the insertion/removal/change of
1831  * file descriptors inside the interest set.
1832  */
1833 SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1834                 struct epoll_event __user *, event)
1835 {
1836         int error;
1837         int full_check = 0;
1838         struct fd f, tf;
1839         struct eventpoll *ep;
1840         struct epitem *epi;
1841         struct epoll_event epds;
1842         struct eventpoll *tep = NULL;
1843 
1844         error = -EFAULT;
1845         if (ep_op_has_event(op) &&
1846             copy_from_user(&epds, event, sizeof(struct epoll_event)))
1847                 goto error_return;
1848 
1849         error = -EBADF;
1850         f = fdget(epfd);
1851         if (!f.file)
1852                 goto error_return;
1853 
1854         /* Get the "struct file *" for the target file */
1855         tf = fdget(fd);
1856         if (!tf.file)
1857                 goto error_fput;
1858 
1859         /* The target file descriptor must support poll */
1860         error = -EPERM;
1861         if (!tf.file->f_op->poll)
1862                 goto error_tgt_fput;
1863 
1864         /* Check if EPOLLWAKEUP is allowed */
1865         if (ep_op_has_event(op))
1866                 ep_take_care_of_epollwakeup(&epds);
1867 
1868         /*
1869          * We have to check that the file structure underneath the file descriptor
1870          * the user passed to us _is_ an eventpoll file. And also we do not permit
1871          * adding an epoll file descriptor inside itself.
1872          */
1873         error = -EINVAL;
1874         if (f.file == tf.file || !is_file_epoll(f.file))
1875                 goto error_tgt_fput;
1876 
1877         /*
1878          * At this point it is safe to assume that the "private_data" contains
1879          * our own data structure.
1880          */
1881         ep = f.file->private_data;
1882 
1883         /*
1884          * When we insert an epoll file descriptor, inside another epoll file
1885          * descriptor, there is the change of creating closed loops, which are
1886          * better be handled here, than in more critical paths. While we are
1887          * checking for loops we also determine the list of files reachable
1888          * and hang them on the tfile_check_list, so we can check that we
1889          * haven't created too many possible wakeup paths.
1890          *
1891          * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
1892          * the epoll file descriptor is attaching directly to a wakeup source,
1893          * unless the epoll file descriptor is nested. The purpose of taking the
1894          * 'epmutex' on add is to prevent complex toplogies such as loops and
1895          * deep wakeup paths from forming in parallel through multiple
1896          * EPOLL_CTL_ADD operations.
1897          */
1898         mutex_lock_nested(&ep->mtx, 0);
1899         if (op == EPOLL_CTL_ADD) {
1900                 if (!list_empty(&f.file->f_ep_links) ||
1901                                                 is_file_epoll(tf.file)) {
1902                         full_check = 1;
1903                         mutex_unlock(&ep->mtx);
1904                         mutex_lock(&epmutex);
1905                         if (is_file_epoll(tf.file)) {
1906                                 error = -ELOOP;
1907                                 if (ep_loop_check(ep, tf.file) != 0) {
1908                                         clear_tfile_check_list();
1909                                         goto error_tgt_fput;
1910                                 }
1911                         } else
1912                                 list_add(&tf.file->f_tfile_llink,
1913                                                         &tfile_check_list);
1914                         mutex_lock_nested(&ep->mtx, 0);
1915                         if (is_file_epoll(tf.file)) {
1916                                 tep = tf.file->private_data;
1917                                 mutex_lock_nested(&tep->mtx, 1);
1918                         }
1919                 }
1920         }
1921 
1922         /*
1923          * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1924          * above, we can be sure to be able to use the item looked up by
1925          * ep_find() till we release the mutex.
1926          */
1927         epi = ep_find(ep, tf.file, fd);
1928 
1929         error = -EINVAL;
1930         switch (op) {
1931         case EPOLL_CTL_ADD:
1932                 if (!epi) {
1933                         epds.events |= POLLERR | POLLHUP;
1934                         error = ep_insert(ep, &epds, tf.file, fd, full_check);
1935                 } else
1936                         error = -EEXIST;
1937                 if (full_check)
1938                         clear_tfile_check_list();
1939                 break;
1940         case EPOLL_CTL_DEL:
1941                 if (epi)
1942                         error = ep_remove(ep, epi);
1943                 else
1944                         error = -ENOENT;
1945                 break;
1946         case EPOLL_CTL_MOD:
1947                 if (epi) {
1948                         epds.events |= POLLERR | POLLHUP;
1949                         error = ep_modify(ep, epi, &epds);
1950                 } else
1951                         error = -ENOENT;
1952                 break;
1953         }
1954         if (tep != NULL)
1955                 mutex_unlock(&tep->mtx);
1956         mutex_unlock(&ep->mtx);
1957 
1958 error_tgt_fput:
1959         if (full_check)
1960                 mutex_unlock(&epmutex);
1961 
1962         fdput(tf);
1963 error_fput:
1964         fdput(f);
1965 error_return:
1966 
1967         return error;
1968 }
1969 
1970 /*
1971  * Implement the event wait interface for the eventpoll file. It is the kernel
1972  * part of the user space epoll_wait(2).
1973  */
1974 SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1975                 int, maxevents, int, timeout)
1976 {
1977         int error;
1978         struct fd f;
1979         struct eventpoll *ep;
1980 
1981         /* The maximum number of event must be greater than zero */
1982         if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1983                 return -EINVAL;
1984 
1985         /* Verify that the area passed by the user is writeable */
1986         if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
1987                 return -EFAULT;
1988 
1989         /* Get the "struct file *" for the eventpoll file */
1990         f = fdget(epfd);
1991         if (!f.file)
1992                 return -EBADF;
1993 
1994         /*
1995          * We have to check that the file structure underneath the fd
1996          * the user passed to us _is_ an eventpoll file.
1997          */
1998         error = -EINVAL;
1999         if (!is_file_epoll(f.file))
2000                 goto error_fput;
2001 
2002         /*
2003          * At this point it is safe to assume that the "private_data" contains
2004          * our own data structure.
2005          */
2006         ep = f.file->private_data;
2007 
2008         /* Time to fish for events ... */
2009         error = ep_poll(ep, events, maxevents, timeout);
2010 
2011 error_fput:
2012         fdput(f);
2013         return error;
2014 }
2015 
2016 /*
2017  * Implement the event wait interface for the eventpoll file. It is the kernel
2018  * part of the user space epoll_pwait(2).
2019  */
2020 SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2021                 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2022                 size_t, sigsetsize)
2023 {
2024         int error;
2025         sigset_t ksigmask, sigsaved;
2026 
2027         /*
2028          * If the caller wants a certain signal mask to be set during the wait,
2029          * we apply it here.
2030          */
2031         if (sigmask) {
2032                 if (sigsetsize != sizeof(sigset_t))
2033                         return -EINVAL;
2034                 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
2035                         return -EFAULT;
2036                 sigsaved = current->blocked;
2037                 set_current_blocked(&ksigmask);
2038         }
2039 
2040         error = sys_epoll_wait(epfd, events, maxevents, timeout);
2041 
2042         /*
2043          * If we changed the signal mask, we need to restore the original one.
2044          * In case we've got a signal while waiting, we do not restore the
2045          * signal mask yet, and we allow do_signal() to deliver the signal on
2046          * the way back to userspace, before the signal mask is restored.
2047          */
2048         if (sigmask) {
2049                 if (error == -EINTR) {
2050                         memcpy(&current->saved_sigmask, &sigsaved,
2051                                sizeof(sigsaved));
2052                         set_restore_sigmask();
2053                 } else
2054                         set_current_blocked(&sigsaved);
2055         }
2056 
2057         return error;
2058 }
2059 
2060 #ifdef CONFIG_COMPAT
2061 COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2062                         struct epoll_event __user *, events,
2063                         int, maxevents, int, timeout,
2064                         const compat_sigset_t __user *, sigmask,
2065                         compat_size_t, sigsetsize)
2066 {
2067         long err;
2068         compat_sigset_t csigmask;
2069         sigset_t ksigmask, sigsaved;
2070 
2071         /*
2072          * If the caller wants a certain signal mask to be set during the wait,
2073          * we apply it here.
2074          */
2075         if (sigmask) {
2076                 if (sigsetsize != sizeof(compat_sigset_t))
2077                         return -EINVAL;
2078                 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
2079                         return -EFAULT;
2080                 sigset_from_compat(&ksigmask, &csigmask);
2081                 sigsaved = current->blocked;
2082                 set_current_blocked(&ksigmask);
2083         }
2084 
2085         err = sys_epoll_wait(epfd, events, maxevents, timeout);
2086 
2087         /*
2088          * If we changed the signal mask, we need to restore the original one.
2089          * In case we've got a signal while waiting, we do not restore the
2090          * signal mask yet, and we allow do_signal() to deliver the signal on
2091          * the way back to userspace, before the signal mask is restored.
2092          */
2093         if (sigmask) {
2094                 if (err == -EINTR) {
2095                         memcpy(&current->saved_sigmask, &sigsaved,
2096                                sizeof(sigsaved));
2097                         set_restore_sigmask();
2098                 } else
2099                         set_current_blocked(&sigsaved);
2100         }
2101 
2102         return err;
2103 }
2104 #endif
2105 
2106 static int __init eventpoll_init(void)
2107 {
2108         struct sysinfo si;
2109 
2110         si_meminfo(&si);
2111         /*
2112          * Allows top 4% of lomem to be allocated for epoll watches (per user).
2113          */
2114         max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2115                 EP_ITEM_COST;
2116         BUG_ON(max_user_watches < 0);
2117 
2118         /*
2119          * Initialize the structure used to perform epoll file descriptor
2120          * inclusion loops checks.
2121          */
2122         ep_nested_calls_init(&poll_loop_ncalls);
2123 
2124         /* Initialize the structure used to perform safe poll wait head wake ups */
2125         ep_nested_calls_init(&poll_safewake_ncalls);
2126 
2127         /* Initialize the structure used to perform file's f_op->poll() calls */
2128         ep_nested_calls_init(&poll_readywalk_ncalls);
2129 
2130         /*
2131          * We can have many thousands of epitems, so prevent this from
2132          * using an extra cache line on 64-bit (and smaller) CPUs
2133          */
2134         BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2135 
2136         /* Allocates slab cache used to allocate "struct epitem" items */
2137         epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
2138                         0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2139 
2140         /* Allocates slab cache used to allocate "struct eppoll_entry" */
2141         pwq_cache = kmem_cache_create("eventpoll_pwq",
2142                         sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
2143 
2144         return 0;
2145 }
2146 fs_initcall(eventpoll_init);
2147 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp