blob: fe047d966dc5d9af1430dbb7596a511ce57131cb [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Davide Libenzi5071f972009-03-31 15:24:10 -07002 * fs/eventpoll.c (Efficient event retrieval implementation)
3 * Copyright (C) 2001,...,2009 Davide Libenzi
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/string.h>
25#include <linux/list.h>
26#include <linux/hash.h>
27#include <linux/spinlock.h>
28#include <linux/syscalls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/rbtree.h>
30#include <linux/wait.h>
31#include <linux/eventpoll.h>
32#include <linux/mount.h>
33#include <linux/bitops.h>
Arjan van de Ven144efe32006-03-23 03:00:32 -080034#include <linux/mutex.h>
Davide Libenzida66f7c2007-05-10 22:23:21 -070035#include <linux/anon_inodes.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37#include <asm/system.h>
38#include <asm/io.h>
39#include <asm/mman.h>
Arun Sharma60063492011-07-26 16:09:06 -070040#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * LOCKING:
44 * There are three level of locking required by epoll :
45 *
Arjan van de Ven144efe32006-03-23 03:00:32 -080046 * 1) epmutex (mutex)
Davide Libenzic7ea7632007-05-15 01:40:47 -070047 * 2) ep->mtx (mutex)
48 * 3) ep->lock (spinlock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 *
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
Davide Libenzid47de162007-05-15 01:40:41 -070058 * mutex (ep->mtx). It is acquired during the event transfer loop,
59 * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
60 * Then we also need a global mutex to serialize eventpoll_release_file()
61 * and ep_free().
62 * This mutex is acquired by ep_free() during the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 * 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
Daniel Balutabf6a41d2011-01-30 23:42:29 +020065 * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
Davide Libenzi22bacca2011-02-25 14:44:12 -080066 * It is also acquired when inserting an epoll fd onto another epoll
67 * fd. We do this so that we walk the epoll tree and ensure that this
68 * insertion does not create a cycle of epoll file descriptors, which
69 * could lead to deadlock. We need a global mutex to prevent two
70 * simultaneous inserts (A into B and B into A) from racing and
71 * constructing a cycle without either insert observing that it is
72 * going to.
Davide Libenzid47de162007-05-15 01:40:41 -070073 * It is possible to drop the "ep->mtx" and to use the global
74 * mutex "epmutex" (together with "ep->lock") to have it working,
75 * but having "ep->mtx" will make the interface more scalable.
Arjan van de Ven144efe32006-03-23 03:00:32 -080076 * Events that require holding "epmutex" are very rare, while for
Davide Libenzid47de162007-05-15 01:40:41 -070077 * normal operations the epoll private "ep->mtx" will guarantee
78 * a better scalability.
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 */
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/* Epoll private bits inside the event mask */
82#define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET)
83
Davide Libenzi5071f972009-03-31 15:24:10 -070084/* Maximum number of nesting allowed inside epoll sets */
85#define EP_MAX_NESTS 4
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Davide Libenzib6119672006-10-11 01:21:44 -070087#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
88
Davide Libenzid47de162007-05-15 01:40:41 -070089#define EP_UNACTIVE_PTR ((void *) -1L)
90
Davide Libenzi7ef99642008-12-01 13:13:55 -080091#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093struct epoll_filefd {
94 struct file *file;
95 int fd;
96};
97
98/*
Davide Libenzi5071f972009-03-31 15:24:10 -070099 * Structure used to track possible nested calls, for too deep recursions
100 * and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700102struct nested_call_node {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 struct list_head llink;
Davide Libenzi5071f972009-03-31 15:24:10 -0700104 void *cookie;
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700105 void *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106};
107
108/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700109 * This structure is used as collector for nested calls, to check for
110 * maximum recursion dept and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700112struct nested_calls {
113 struct list_head tasks_call_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 spinlock_t lock;
115};
116
117/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 * Each file descriptor added to the eventpoll interface will
Davide Libenzi6192bd52007-05-08 00:25:41 -0700119 * have an entry of this type linked to the "rbr" RB tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 */
121struct epitem {
Davide Libenzi67647d02007-05-15 01:40:52 -0700122 /* RB tree node used to link this structure to the eventpoll RB tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 struct rb_node rbn;
124
125 /* List header used to link this structure to the eventpoll ready list */
126 struct list_head rdllink;
127
Davide Libenzic7ea7632007-05-15 01:40:47 -0700128 /*
129 * Works together "struct eventpoll"->ovflist in keeping the
130 * single linked chain of items.
131 */
132 struct epitem *next;
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 /* The file descriptor information this item refers to */
135 struct epoll_filefd ffd;
136
137 /* Number of active wait queue attached to poll operations */
138 int nwait;
139
140 /* List containing poll wait queues */
141 struct list_head pwqlist;
142
143 /* The "container" of this item */
144 struct eventpoll *ep;
145
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 /* List header used to link this item to the "struct file" items list */
147 struct list_head fllink;
Davide Libenzid47de162007-05-15 01:40:41 -0700148
Davide Libenzic7ea7632007-05-15 01:40:47 -0700149 /* The structure that describe the interested events and the source fd */
150 struct epoll_event event;
Davide Libenzid47de162007-05-15 01:40:41 -0700151};
152
153/*
154 * This structure is stored inside the "private_data" member of the file
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200155 * structure and represents the main data structure for the eventpoll
Davide Libenzid47de162007-05-15 01:40:41 -0700156 * interface.
157 */
158struct eventpoll {
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200159 /* Protect the access to this structure */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700160 spinlock_t lock;
Davide Libenzid47de162007-05-15 01:40:41 -0700161
162 /*
163 * This mutex is used to ensure that files are not removed
164 * while epoll is using them. This is held during the event
165 * collection loop, the file cleanup path, the epoll file exit
166 * code and the ctl operations.
167 */
168 struct mutex mtx;
169
170 /* Wait queue used by sys_epoll_wait() */
171 wait_queue_head_t wq;
172
173 /* Wait queue used by file->poll() */
174 wait_queue_head_t poll_wait;
175
176 /* List of ready file descriptors */
177 struct list_head rdllist;
178
Davide Libenzi67647d02007-05-15 01:40:52 -0700179 /* RB tree root used to store monitored fd structs */
Davide Libenzid47de162007-05-15 01:40:41 -0700180 struct rb_root rbr;
181
182 /*
183 * This is a single linked list that chains all the "struct epitem" that
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300184 * happened while transferring ready events to userspace w/out
Davide Libenzid47de162007-05-15 01:40:41 -0700185 * holding ->lock.
186 */
187 struct epitem *ovflist;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800188
189 /* The user that created the eventpoll descriptor */
190 struct user_struct *user;
Davide Libenzid47de162007-05-15 01:40:41 -0700191};
192
193/* Wait structure used by the poll hooks */
194struct eppoll_entry {
195 /* List header used to link this structure to the "struct epitem" */
196 struct list_head llink;
197
198 /* The "base" pointer is set to the container "struct epitem" */
Tony Battersby4f0989d2009-03-31 15:24:16 -0700199 struct epitem *base;
Davide Libenzid47de162007-05-15 01:40:41 -0700200
201 /*
202 * Wait queue item that will be linked to the target file wait
203 * queue head.
204 */
205 wait_queue_t wait;
206
207 /* The wait queue head that linked the "wait" wait queue item */
208 wait_queue_head_t *whead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209};
210
211/* Wrapper struct used by poll queueing */
212struct ep_pqueue {
213 poll_table pt;
214 struct epitem *epi;
215};
216
Davide Libenzi5071f972009-03-31 15:24:10 -0700217/* Used by the ep_send_events() function as callback private data */
218struct ep_send_events_data {
219 int maxevents;
220 struct epoll_event __user *events;
221};
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223/*
Davide Libenzi7ef99642008-12-01 13:13:55 -0800224 * Configuration options available inside /proc/sys/fs/epoll/
225 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800226/* Maximum number of epoll watched descriptors, per user */
Robin Holt52bd19f72011-01-12 17:00:01 -0800227static long max_user_watches __read_mostly;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800228
229/*
Davide Libenzid47de162007-05-15 01:40:41 -0700230 * This mutex is used to serialize ep_free() and eventpoll_release_file().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800232static DEFINE_MUTEX(epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Davide Libenzi22bacca2011-02-25 14:44:12 -0800234/* Used to check for epoll file descriptor inclusion loops */
235static struct nested_calls poll_loop_ncalls;
236
Davide Libenzi5071f972009-03-31 15:24:10 -0700237/* Used for safe wake up implementation */
238static struct nested_calls poll_safewake_ncalls;
239
240/* Used to call file's f_op->poll() under the nested calls boundaries */
241static struct nested_calls poll_readywalk_ncalls;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243/* Slab cache used to allocate "struct epitem" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800244static struct kmem_cache *epi_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246/* Slab cache used to allocate "struct eppoll_entry" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800247static struct kmem_cache *pwq_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Davide Libenzi7ef99642008-12-01 13:13:55 -0800249#ifdef CONFIG_SYSCTL
250
251#include <linux/sysctl.h>
252
Robin Holt52bd19f72011-01-12 17:00:01 -0800253static long zero;
254static long long_max = LONG_MAX;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800255
256ctl_table epoll_table[] = {
257 {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800258 .procname = "max_user_watches",
259 .data = &max_user_watches,
Robin Holt52bd19f72011-01-12 17:00:01 -0800260 .maxlen = sizeof(max_user_watches),
Davide Libenzi7ef99642008-12-01 13:13:55 -0800261 .mode = 0644,
Robin Holt52bd19f72011-01-12 17:00:01 -0800262 .proc_handler = proc_doulongvec_minmax,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800263 .extra1 = &zero,
Robin Holt52bd19f72011-01-12 17:00:01 -0800264 .extra2 = &long_max,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800265 },
Eric W. Biedermanab092032009-11-05 14:25:10 -0800266 { }
Davide Libenzi7ef99642008-12-01 13:13:55 -0800267};
268#endif /* CONFIG_SYSCTL */
269
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700270
Davide Libenzi67647d02007-05-15 01:40:52 -0700271/* Setup the structure that is used as key for the RB tree */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700272static inline void ep_set_ffd(struct epoll_filefd *ffd,
273 struct file *file, int fd)
274{
275 ffd->file = file;
276 ffd->fd = fd;
277}
278
Davide Libenzi67647d02007-05-15 01:40:52 -0700279/* Compare RB tree keys */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700280static inline int ep_cmp_ffd(struct epoll_filefd *p1,
281 struct epoll_filefd *p2)
282{
283 return (p1->file > p2->file ? +1:
284 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
285}
286
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700287/* Tells us if the item is currently linked */
288static inline int ep_is_linked(struct list_head *p)
289{
290 return !list_empty(p);
291}
292
293/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700294static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700295{
296 return container_of(p, struct eppoll_entry, wait)->base;
297}
298
299/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700300static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700301{
302 return container_of(p, struct ep_pqueue, pt)->epi;
303}
304
305/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700306static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700307{
308 return op != EPOLL_CTL_DEL;
309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700312static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Davide Libenzi5071f972009-03-31 15:24:10 -0700314 INIT_LIST_HEAD(&ncalls->tasks_call_list);
315 spin_lock_init(&ncalls->lock);
316}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Davide Libenzi5071f972009-03-31 15:24:10 -0700318/**
Davide Libenzi3fb0e582011-03-22 16:34:46 -0700319 * ep_events_available - Checks if ready events might be available.
320 *
321 * @ep: Pointer to the eventpoll context.
322 *
323 * Returns: Returns a value different than zero if ready events are available,
324 * or zero otherwise.
325 */
326static inline int ep_events_available(struct eventpoll *ep)
327{
328 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
329}
330
331/**
Davide Libenzi5071f972009-03-31 15:24:10 -0700332 * ep_call_nested - Perform a bound (possibly) nested call, by checking
333 * that the recursion limit is not exceeded, and that
334 * the same nested call (by the meaning of same cookie) is
335 * no re-entered.
336 *
337 * @ncalls: Pointer to the nested_calls structure to be used for this call.
338 * @max_nests: Maximum number of allowed nesting calls.
339 * @nproc: Nested call core function pointer.
340 * @priv: Opaque data to be passed to the @nproc callback.
341 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700342 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700343 *
344 * Returns: Returns the code returned by the @nproc callback, or -1 if
345 * the maximum recursion limit has been exceeded.
346 */
347static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
348 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700349 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700350{
351 int error, call_nests = 0;
352 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700353 struct list_head *lsthead = &ncalls->tasks_call_list;
354 struct nested_call_node *tncur;
355 struct nested_call_node tnode;
356
357 spin_lock_irqsave(&ncalls->lock, flags);
358
359 /*
360 * Try to see if the current task is already inside this wakeup call.
361 * We use a list here, since the population inside this set is always
362 * very much limited.
363 */
364 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700365 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700366 (tncur->cookie == cookie || ++call_nests > max_nests)) {
367 /*
368 * Ops ... loop detected or maximum nest level reached.
369 * We abort this wake by breaking the cycle itself.
370 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700371 error = -1;
372 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700373 }
374 }
375
376 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700377 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700378 tnode.cookie = cookie;
379 list_add(&tnode.llink, lsthead);
380
381 spin_unlock_irqrestore(&ncalls->lock, flags);
382
383 /* Call the nested function */
384 error = (*nproc)(priv, cookie, call_nests);
385
386 /* Remove the current task from the list */
387 spin_lock_irqsave(&ncalls->lock, flags);
388 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700389out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700390 spin_unlock_irqrestore(&ncalls->lock, flags);
391
392 return error;
393}
394
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700395#ifdef CONFIG_DEBUG_LOCK_ALLOC
396static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
397 unsigned long events, int subclass)
398{
399 unsigned long flags;
400
401 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
402 wake_up_locked_poll(wqueue, events);
403 spin_unlock_irqrestore(&wqueue->lock, flags);
404}
405#else
406static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
407 unsigned long events, int subclass)
408{
409 wake_up_poll(wqueue, events);
410}
411#endif
412
Davide Libenzi5071f972009-03-31 15:24:10 -0700413static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
414{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700415 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
416 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700417 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420/*
421 * Perform a safe wake up of the poll wait list. The problem is that
422 * with the new callback'd wake up system, it is possible that the
423 * poll callback is reentered from inside the call to wake_up() done
424 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700425 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 * and we cannot reenter the same wait queue head at all. This will
427 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700428 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700430static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700432 int this_cpu = get_cpu();
433
Davide Libenzi5071f972009-03-31 15:24:10 -0700434 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700435 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
436
437 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438}
439
Davide Libenzi7699acd2007-05-10 22:23:23 -0700440/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700441 * This function unregisters poll callbacks from the associated file
442 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
443 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700444 */
445static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
446{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700447 struct list_head *lsthead = &epi->pwqlist;
448 struct eppoll_entry *pwq;
449
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700450 while (!list_empty(lsthead)) {
451 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700452
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700453 list_del(&pwq->llink);
454 remove_wait_queue(pwq->whead, &pwq->wait);
455 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700456 }
457}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Davide Libenzi5071f972009-03-31 15:24:10 -0700459/**
460 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
461 * the scan code, to call f_op->poll(). Also allows for
462 * O(NumReady) performance.
463 *
464 * @ep: Pointer to the epoll private data structure.
465 * @sproc: Pointer to the scan callback.
466 * @priv: Private opaque data passed to the @sproc callback.
467 *
468 * Returns: The same integer error code returned by the @sproc callback.
469 */
470static int ep_scan_ready_list(struct eventpoll *ep,
471 int (*sproc)(struct eventpoll *,
472 struct list_head *, void *),
473 void *priv)
474{
475 int error, pwake = 0;
476 unsigned long flags;
477 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700478 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700479
480 /*
481 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700482 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700483 */
484 mutex_lock(&ep->mtx);
485
486 /*
487 * Steal the ready list, and re-init the original one to the
488 * empty list. Also, set ep->ovflist to NULL so that events
489 * happening while looping w/out locks, are not lost. We cannot
490 * have the poll callback to queue directly on ep->rdllist,
491 * because we want the "sproc" callback to be able to do it
492 * in a lockless way.
493 */
494 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700495 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700496 ep->ovflist = NULL;
497 spin_unlock_irqrestore(&ep->lock, flags);
498
499 /*
500 * Now call the callback function.
501 */
502 error = (*sproc)(ep, &txlist, priv);
503
504 spin_lock_irqsave(&ep->lock, flags);
505 /*
506 * During the time we spent inside the "sproc" callback, some
507 * other events might have been queued by the poll callback.
508 * We re-insert them inside the main ready-list here.
509 */
510 for (nepi = ep->ovflist; (epi = nepi) != NULL;
511 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
512 /*
513 * We need to check if the item is already in the list.
514 * During the "sproc" callback execution time, items are
515 * queued into ->ovflist but the "txlist" might already
516 * contain them, and the list_splice() below takes care of them.
517 */
518 if (!ep_is_linked(&epi->rdllink))
519 list_add_tail(&epi->rdllink, &ep->rdllist);
520 }
521 /*
522 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
523 * releasing the lock, events will be queued in the normal way inside
524 * ep->rdllist.
525 */
526 ep->ovflist = EP_UNACTIVE_PTR;
527
528 /*
529 * Quickly re-inject items left on "txlist".
530 */
531 list_splice(&txlist, &ep->rdllist);
532
533 if (!list_empty(&ep->rdllist)) {
534 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700535 * Wake up (if active) both the eventpoll wait list and
536 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700537 */
538 if (waitqueue_active(&ep->wq))
539 wake_up_locked(&ep->wq);
540 if (waitqueue_active(&ep->poll_wait))
541 pwake++;
542 }
543 spin_unlock_irqrestore(&ep->lock, flags);
544
545 mutex_unlock(&ep->mtx);
546
547 /* We have to call this outside the lock */
548 if (pwake)
549 ep_poll_safewake(&ep->poll_wait);
550
551 return error;
552}
553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700555 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700556 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700557 */
558static int ep_remove(struct eventpoll *ep, struct epitem *epi)
559{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700560 unsigned long flags;
561 struct file *file = epi->ffd.file;
562
563 /*
564 * Removes poll wait queue hooks. We _have_ to do this without holding
565 * the "ep->lock" otherwise a deadlock might occur. This because of the
566 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
567 * queue head lock when unregistering the wait queue. The wakeup callback
568 * will run by holding the wait queue head lock and will call our callback
569 * that will try to get "ep->lock".
570 */
571 ep_unregister_pollwait(ep, epi);
572
573 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700574 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700575 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700576 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700577 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Davide Libenzicdac75e2008-04-29 00:58:34 -0700579 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Davide Libenzic7ea7632007-05-15 01:40:47 -0700581 spin_lock_irqsave(&ep->lock, flags);
582 if (ep_is_linked(&epi->rdllink))
583 list_del_init(&epi->rdllink);
584 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Davide Libenzi7699acd2007-05-10 22:23:23 -0700586 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700587 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Robin Holt52bd19f72011-01-12 17:00:01 -0800589 atomic_long_dec(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800590
Davide Libenzic7ea7632007-05-15 01:40:47 -0700591 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592}
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594static void ep_free(struct eventpoll *ep)
595{
596 struct rb_node *rbp;
597 struct epitem *epi;
598
599 /* We need to release all tasks waiting for these file */
600 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700601 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
603 /*
604 * We need to lock this because we could be hit by
605 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700606 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 * is on the way to be removed and no one has references to it
608 * anymore. The only hit might come from eventpoll_release_file() but
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300609 * holding "epmutex" is sufficient here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800611 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
613 /*
614 * Walks through the whole tree by unregistering poll callbacks.
615 */
616 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
617 epi = rb_entry(rbp, struct epitem, rbn);
618
619 ep_unregister_pollwait(ep, epi);
620 }
621
622 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700623 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700625 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 * us during this operation. So we can avoid the lock on "ep->lock".
627 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700628 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 epi = rb_entry(rbp, struct epitem, rbn);
630 ep_remove(ep, epi);
631 }
632
Arjan van de Ven144efe32006-03-23 03:00:32 -0800633 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700634 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800635 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700636 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
Davide Libenzi7699acd2007-05-10 22:23:23 -0700639static int ep_eventpoll_release(struct inode *inode, struct file *file)
640{
641 struct eventpoll *ep = file->private_data;
642
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700643 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700644 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700645
Davide Libenzi7699acd2007-05-10 22:23:23 -0700646 return 0;
647}
648
Davide Libenzi296e2362009-03-31 15:24:11 -0700649static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
650 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700651{
652 struct epitem *epi, *tmp;
653
654 list_for_each_entry_safe(epi, tmp, head, rdllink) {
655 if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
656 epi->event.events)
657 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700658 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700659 /*
660 * Item has been dropped into the ready list by the poll
661 * callback, but it's not actually ready, as far as
662 * caller requested events goes. We can remove it here.
663 */
664 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700665 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700666 }
667
668 return 0;
669}
670
671static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
672{
673 return ep_scan_ready_list(priv, ep_read_events_proc, NULL);
674}
675
Davide Libenzi7699acd2007-05-10 22:23:23 -0700676static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
677{
Davide Libenzi5071f972009-03-31 15:24:10 -0700678 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700679 struct eventpoll *ep = file->private_data;
680
681 /* Insert inside our poll wait queue */
682 poll_wait(file, &ep->poll_wait, wait);
683
Davide Libenzi5071f972009-03-31 15:24:10 -0700684 /*
685 * Proceed to find out if wanted events are really available inside
686 * the ready list. This need to be done under ep_call_nested()
687 * supervision, since the call to f_op->poll() done on listed files
688 * could re-enter here.
689 */
690 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700691 ep_poll_readyevents_proc, ep, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700692
Davide Libenzi296e2362009-03-31 15:24:11 -0700693 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700694}
695
696/* File callbacks that implement the eventpoll file behaviour */
697static const struct file_operations eventpoll_fops = {
698 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200699 .poll = ep_eventpoll_poll,
700 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700701};
702
703/* Fast test to see if the file is an evenpoll file */
704static inline int is_file_epoll(struct file *f)
705{
706 return f->f_op == &eventpoll_fops;
707}
708
709/*
710 * This is called from eventpoll_release() to unlink files from the eventpoll
711 * interface. We need to have this facility to cleanup correctly files that are
712 * closed without being removed from the eventpoll interface.
713 */
714void eventpoll_release_file(struct file *file)
715{
716 struct list_head *lsthead = &file->f_ep_links;
717 struct eventpoll *ep;
718 struct epitem *epi;
719
720 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700721 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700722 * necessary. It is not necessary because we're in the "struct file"
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300723 * cleanup path, and this means that no one is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700724 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700725 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700726 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700727 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700728 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700729 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700730 *
731 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700732 */
733 mutex_lock(&epmutex);
734
735 while (!list_empty(lsthead)) {
736 epi = list_first_entry(lsthead, struct epitem, fllink);
737
738 ep = epi->ep;
739 list_del_init(&epi->fllink);
Davide Libenzid47de162007-05-15 01:40:41 -0700740 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700741 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700742 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700743 }
744
745 mutex_unlock(&epmutex);
746}
747
748static int ep_alloc(struct eventpoll **pep)
749{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800750 int error;
751 struct user_struct *user;
752 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700753
Davide Libenzi7ef99642008-12-01 13:13:55 -0800754 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800755 error = -ENOMEM;
756 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
757 if (unlikely(!ep))
758 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700759
Davide Libenzic7ea7632007-05-15 01:40:47 -0700760 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700761 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700762 init_waitqueue_head(&ep->wq);
763 init_waitqueue_head(&ep->poll_wait);
764 INIT_LIST_HEAD(&ep->rdllist);
765 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700766 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800767 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700768
769 *pep = ep;
770
Davide Libenzi7699acd2007-05-10 22:23:23 -0700771 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800772
773free_uid:
774 free_uid(user);
775 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700776}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700779 * Search the file inside the eventpoll tree. The RB tree operations
780 * are protected by the "mtx" mutex, and ep_find() must be called with
781 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
783static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
784{
785 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 struct rb_node *rbp;
787 struct epitem *epi, *epir = NULL;
788 struct epoll_filefd ffd;
789
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700790 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 for (rbp = ep->rbr.rb_node; rbp; ) {
792 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700793 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 if (kcmp > 0)
795 rbp = rbp->rb_right;
796 else if (kcmp < 0)
797 rbp = rbp->rb_left;
798 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 epir = epi;
800 break;
801 }
802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 return epir;
805}
806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700808 * This is the callback that is passed to the wait queue wakeup
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200809 * mechanism. It is called by the stored file descriptors when they
Davide Libenzi7699acd2007-05-10 22:23:23 -0700810 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700812static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700814 int pwake = 0;
815 unsigned long flags;
816 struct epitem *epi = ep_item_from_wait(wait);
817 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Davide Libenzic7ea7632007-05-15 01:40:47 -0700819 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700820
821 /*
822 * If the event mask does not contain any poll(2) event, we consider the
823 * descriptor to be disabled. This condition is likely the effect of the
824 * EPOLLONESHOT bit that disables the descriptor when an event is received,
825 * until the next EPOLL_CTL_MOD will be issued.
826 */
827 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700828 goto out_unlock;
829
830 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700831 * Check the events coming with the callback. At this stage, not
832 * every device reports the events in the "key" parameter of the
833 * callback. We need to be able to handle both cases here, hence the
834 * test for "key" != NULL before the event match test.
835 */
836 if (key && !((unsigned long) key & epi->event.events))
837 goto out_unlock;
838
839 /*
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200840 * If we are transferring events to userspace, we can hold no locks
Davide Libenzid47de162007-05-15 01:40:41 -0700841 * (because we're accessing user memory, and because of linux f_op->poll()
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200842 * semantics). All the events that happen during that period of time are
Davide Libenzid47de162007-05-15 01:40:41 -0700843 * chained in ep->ovflist and requeued later on.
844 */
845 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
846 if (epi->next == EP_UNACTIVE_PTR) {
847 epi->next = ep->ovflist;
848 ep->ovflist = epi;
849 }
850 goto out_unlock;
851 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700852
853 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700854 if (!ep_is_linked(&epi->rdllink))
855 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700856
Davide Libenzi7699acd2007-05-10 22:23:23 -0700857 /*
858 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
859 * wait list.
860 */
861 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -0400862 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700863 if (waitqueue_active(&ep->poll_wait))
864 pwake++;
865
Davide Libenzid47de162007-05-15 01:40:41 -0700866out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700867 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700868
869 /* We have to call this outside the lock */
870 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700871 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700872
873 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874}
875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876/*
877 * This is the callback that is used to add our wait queue to the
878 * target file wakeup lists.
879 */
880static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
881 poll_table *pt)
882{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700883 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 struct eppoll_entry *pwq;
885
Christoph Lametere94b1762006-12-06 20:33:17 -0800886 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
888 pwq->whead = whead;
889 pwq->base = epi;
890 add_wait_queue(whead, &pwq->wait);
891 list_add_tail(&pwq->llink, &epi->pwqlist);
892 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700893 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 /* We have to signal that an error occurred */
895 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
900{
901 int kcmp;
902 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
903 struct epitem *epic;
904
905 while (*p) {
906 parent = *p;
907 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700908 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (kcmp > 0)
910 p = &parent->rb_right;
911 else
912 p = &parent->rb_left;
913 }
914 rb_link_node(&epi->rbn, parent, p);
915 rb_insert_color(&epi->rbn, &ep->rbr);
916}
917
Davide Libenzic7ea7632007-05-15 01:40:47 -0700918/*
919 * Must be called with "mtx" held.
920 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
922 struct file *tfile, int fd)
923{
924 int error, revents, pwake = 0;
925 unsigned long flags;
Robin Holt52bd19f72011-01-12 17:00:01 -0800926 long user_watches;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 struct epitem *epi;
928 struct ep_pqueue epq;
929
Robin Holt52bd19f72011-01-12 17:00:01 -0800930 user_watches = atomic_long_read(&ep->user->epoll_watches);
931 if (unlikely(user_watches >= max_user_watches))
Davide Libenzi7ef99642008-12-01 13:13:55 -0800932 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -0800933 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -0800934 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 INIT_LIST_HEAD(&epi->rdllink);
938 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 INIT_LIST_HEAD(&epi->pwqlist);
940 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700941 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -0700944 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
946 /* Initialize the poll table using the queue callback */
947 epq.epi = epi;
948 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
949
950 /*
951 * Attach the item to the poll hooks and get current event bits.
952 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -0700953 * been increased by the caller of this function. Note that after
954 * this operation completes, the poll callback can start hitting
955 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 */
957 revents = tfile->f_op->poll(tfile, &epq.pt);
958
959 /*
960 * We have to check if something went wrong during the poll wait queue
961 * install process. Namely an allocation for a wait queue failed due
962 * high memory pressure.
963 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800964 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700966 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
968 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -0700969 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -0700971 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Davide Libenzic7ea7632007-05-15 01:40:47 -0700973 /*
974 * Add the current item to the RB tree. All RB tree operations are
975 * protected by "mtx", and ep_insert() is called with "mtx" held.
976 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 ep_rbtree_insert(ep, epi);
978
Davide Libenzic7ea7632007-05-15 01:40:47 -0700979 /* We have to drop the new item inside our item list to keep track of it */
980 spin_lock_irqsave(&ep->lock, flags);
981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700983 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 list_add_tail(&epi->rdllink, &ep->rdllist);
985
986 /* Notify waiting tasks that events are available */
987 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -0400988 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (waitqueue_active(&ep->poll_wait))
990 pwake++;
991 }
992
Davide Libenzic7ea7632007-05-15 01:40:47 -0700993 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Robin Holt52bd19f72011-01-12 17:00:01 -0800995 atomic_long_inc(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 /* We have to call this outside the lock */
998 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700999 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 return 0;
1002
Davide Libenzi7699acd2007-05-10 22:23:23 -07001003error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 ep_unregister_pollwait(ep, epi);
1005
1006 /*
1007 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -07001008 * allocated wait queue. Note that we don't care about the ep->ovflist
1009 * list, since that is used/cleaned only inside a section bound by "mtx".
1010 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001012 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001013 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -07001014 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001015 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001017 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 return error;
1020}
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022/*
1023 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001024 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 */
1026static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1027{
1028 int pwake = 0;
1029 unsigned int revents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
1031 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001032 * Set the new event interest mask before calling f_op->poll();
1033 * otherwise we might miss an event that happens between the
1034 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 */
1036 epi->event.events = event->events;
Tony Battersbye057e152009-03-31 15:24:15 -07001037 epi->event.data = event->data; /* protected by mtx */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039 /*
1040 * Get current event bits. We can safely use the file* here because
1041 * its usage count has been increased by the caller of this function.
1042 */
1043 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL);
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001046 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001047 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001049 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001050 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001051 if (!ep_is_linked(&epi->rdllink)) {
1052 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Davide Libenzic7ea7632007-05-15 01:40:47 -07001054 /* Notify waiting tasks that events are available */
1055 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -04001056 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001057 if (waitqueue_active(&ep->poll_wait))
1058 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 }
Tony Battersbye057e152009-03-31 15:24:15 -07001060 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
1063 /* We have to call this outside the lock */
1064 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001065 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
1067 return 0;
1068}
1069
Davide Libenzi296e2362009-03-31 15:24:11 -07001070static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1071 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001072{
1073 struct ep_send_events_data *esed = priv;
1074 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001075 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001076 struct epitem *epi;
1077 struct epoll_event __user *uevent;
1078
Davide Libenzi296e2362009-03-31 15:24:11 -07001079 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001080 * We can loop without lock because we are passed a task private list.
1081 * Items cannot vanish during the loop because ep_scan_ready_list() is
1082 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001083 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001084 for (eventcnt = 0, uevent = esed->events;
1085 !list_empty(head) && eventcnt < esed->maxevents;) {
1086 epi = list_first_entry(head, struct epitem, rdllink);
1087
1088 list_del_init(&epi->rdllink);
1089
Davide Libenzi296e2362009-03-31 15:24:11 -07001090 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
1091 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001092
Davide Libenzi296e2362009-03-31 15:24:11 -07001093 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001094 * If the event mask intersect the caller-requested one,
1095 * deliver the event to userspace. Again, ep_scan_ready_list()
1096 * is holding "mtx", so no operations coming from userspace
1097 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001098 */
1099 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001100 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001101 __put_user(epi->event.data, &uevent->data)) {
1102 list_add(&epi->rdllink, head);
Davide Libenzi296e2362009-03-31 15:24:11 -07001103 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001104 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001105 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001106 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001107 if (epi->event.events & EPOLLONESHOT)
1108 epi->event.events &= EP_PRIVATE_BITS;
1109 else if (!(epi->event.events & EPOLLET)) {
1110 /*
1111 * If this file has been added with Level
1112 * Trigger mode, we need to insert back inside
1113 * the ready list, so that the next call to
1114 * epoll_wait() will check again the events
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001115 * availability. At this point, no one can insert
Davide Libenzi296e2362009-03-31 15:24:11 -07001116 * into ep->rdllist besides us. The epoll_ctl()
1117 * callers are locked out by
1118 * ep_scan_ready_list() holding "mtx" and the
1119 * poll callback will queue them in ep->ovflist.
1120 */
1121 list_add_tail(&epi->rdllink, &ep->rdllist);
1122 }
1123 }
1124 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001125
1126 return eventcnt;
1127}
1128
Davide Libenzi296e2362009-03-31 15:24:11 -07001129static int ep_send_events(struct eventpoll *ep,
1130 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131{
Davide Libenzi5071f972009-03-31 15:24:10 -07001132 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001133
Davide Libenzi5071f972009-03-31 15:24:10 -07001134 esed.maxevents = maxevents;
1135 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001136
Davide Libenzi5071f972009-03-31 15:24:10 -07001137 return ep_scan_ready_list(ep, ep_send_events_proc, &esed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
Eric Dumazet0781b902011-02-01 15:52:35 -08001140static inline struct timespec ep_set_mstimeout(long ms)
1141{
1142 struct timespec now, ts = {
1143 .tv_sec = ms / MSEC_PER_SEC,
1144 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1145 };
1146
1147 ktime_get_ts(&now);
1148 return timespec_add_safe(now, ts);
1149}
1150
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001151/**
1152 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1153 * event buffer.
1154 *
1155 * @ep: Pointer to the eventpoll context.
1156 * @events: Pointer to the userspace buffer where the ready events should be
1157 * stored.
1158 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1159 * @timeout: Maximum timeout for the ready events fetch operation, in
1160 * milliseconds. If the @timeout is zero, the function will not block,
1161 * while if the @timeout is less than zero, the function will block
1162 * until at least one event has been retrieved (or an error
1163 * occurred).
1164 *
1165 * Returns: Returns the number of ready events which have been fetched, or an
1166 * error code, in case of error.
1167 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1169 int maxevents, long timeout)
1170{
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001171 int res = 0, eavail, timed_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 unsigned long flags;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001173 long slack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 wait_queue_t wait;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001175 ktime_t expires, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001177 if (timeout > 0) {
Eric Dumazet0781b902011-02-01 15:52:35 -08001178 struct timespec end_time = ep_set_mstimeout(timeout);
1179
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001180 slack = select_estimate_accuracy(&end_time);
1181 to = &expires;
1182 *to = timespec_to_ktime(end_time);
1183 } else if (timeout == 0) {
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001184 /*
1185 * Avoid the unnecessary trip to the wait queue loop, if the
1186 * caller specified a non blocking operation.
1187 */
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001188 timed_out = 1;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001189 spin_lock_irqsave(&ep->lock, flags);
1190 goto check_events;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001193fetch_events:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001194 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001196 if (!ep_events_available(ep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 /*
1198 * We don't have any available event to return to the caller.
1199 * We need to sleep here, and we will be wake up by
1200 * ep_poll_callback() when events will become available.
1201 */
1202 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001203 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
1205 for (;;) {
1206 /*
1207 * We don't want to sleep if the ep_poll_callback() sends us
1208 * a wakeup in between. That's why we set the task state
1209 * to TASK_INTERRUPTIBLE before doing the checks.
1210 */
1211 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001212 if (ep_events_available(ep) || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 break;
1214 if (signal_pending(current)) {
1215 res = -EINTR;
1216 break;
1217 }
1218
Davide Libenzic7ea7632007-05-15 01:40:47 -07001219 spin_unlock_irqrestore(&ep->lock, flags);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001220 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1221 timed_out = 1;
1222
Davide Libenzic7ea7632007-05-15 01:40:47 -07001223 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001225 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 set_current_state(TASK_RUNNING);
1228 }
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001229check_events:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 /* Is it worth to try to dig for events ? */
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001231 eavail = ep_events_available(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Davide Libenzic7ea7632007-05-15 01:40:47 -07001233 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
1235 /*
1236 * Try to transfer events to user space. In case we get 0 events and
1237 * there's still timeout left over, we go trying again in search of
1238 * more luck.
1239 */
1240 if (!res && eavail &&
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001241 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001242 goto fetch_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
1244 return res;
1245}
1246
Davide Libenzi22bacca2011-02-25 14:44:12 -08001247/**
1248 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1249 * API, to verify that adding an epoll file inside another
1250 * epoll structure, does not violate the constraints, in
1251 * terms of closed loops, or too deep chains (which can
1252 * result in excessive stack usage).
1253 *
1254 * @priv: Pointer to the epoll file to be currently checked.
1255 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1256 * data structure pointer.
1257 * @call_nests: Current dept of the @ep_call_nested() call stack.
1258 *
1259 * Returns: Returns zero if adding the epoll @file inside current epoll
1260 * structure @ep does not violate the constraints, or -1 otherwise.
1261 */
1262static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1263{
1264 int error = 0;
1265 struct file *file = priv;
1266 struct eventpoll *ep = file->private_data;
1267 struct rb_node *rbp;
1268 struct epitem *epi;
1269
1270 mutex_lock(&ep->mtx);
1271 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1272 epi = rb_entry(rbp, struct epitem, rbn);
1273 if (unlikely(is_file_epoll(epi->ffd.file))) {
1274 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1275 ep_loop_check_proc, epi->ffd.file,
1276 epi->ffd.file->private_data, current);
1277 if (error != 0)
1278 break;
1279 }
1280 }
1281 mutex_unlock(&ep->mtx);
1282
1283 return error;
1284}
1285
1286/**
1287 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1288 * another epoll file (represented by @ep) does not create
1289 * closed loops or too deep chains.
1290 *
1291 * @ep: Pointer to the epoll private data structure.
1292 * @file: Pointer to the epoll file to be checked.
1293 *
1294 * Returns: Returns zero if adding the epoll @file inside current epoll
1295 * structure @ep does not violate the constraints, or -1 otherwise.
1296 */
1297static int ep_loop_check(struct eventpoll *ep, struct file *file)
1298{
1299 return ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1300 ep_loop_check_proc, file, ep, current);
1301}
1302
Davide Libenzi7699acd2007-05-10 22:23:23 -07001303/*
Andrew Morton523723b2008-08-12 15:09:01 -07001304 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001305 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001306SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001307{
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001308 int error;
1309 struct eventpoll *ep = NULL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001310
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001311 /* Check the EPOLL_* constant for consistency. */
1312 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1313
Davide Libenzi296e2362009-03-31 15:24:11 -07001314 if (flags & ~EPOLL_CLOEXEC)
1315 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001316 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001317 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001318 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001319 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001320 if (error < 0)
1321 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001322 /*
1323 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001324 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001325 */
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001326 error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001327 O_RDWR | (flags & O_CLOEXEC));
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001328 if (error < 0)
Al Viro2030a422008-02-23 06:46:49 -05001329 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001330
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001331 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001332}
1333
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001334SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001335{
Davide Libenzibfe38912009-05-12 13:19:44 -07001336 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001337 return -EINVAL;
1338
1339 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001340}
1341
Davide Libenzi7699acd2007-05-10 22:23:23 -07001342/*
1343 * The following function implements the controller interface for
1344 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001345 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001346 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001347SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1348 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001349{
1350 int error;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001351 int did_lock_epmutex = 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001352 struct file *file, *tfile;
1353 struct eventpoll *ep;
1354 struct epitem *epi;
1355 struct epoll_event epds;
1356
Davide Libenzi7699acd2007-05-10 22:23:23 -07001357 error = -EFAULT;
1358 if (ep_op_has_event(op) &&
1359 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1360 goto error_return;
1361
1362 /* Get the "struct file *" for the eventpoll file */
1363 error = -EBADF;
1364 file = fget(epfd);
1365 if (!file)
1366 goto error_return;
1367
1368 /* Get the "struct file *" for the target file */
1369 tfile = fget(fd);
1370 if (!tfile)
1371 goto error_fput;
1372
1373 /* The target file descriptor must support poll */
1374 error = -EPERM;
1375 if (!tfile->f_op || !tfile->f_op->poll)
1376 goto error_tgt_fput;
1377
1378 /*
1379 * We have to check that the file structure underneath the file descriptor
1380 * the user passed to us _is_ an eventpoll file. And also we do not permit
1381 * adding an epoll file descriptor inside itself.
1382 */
1383 error = -EINVAL;
1384 if (file == tfile || !is_file_epoll(file))
1385 goto error_tgt_fput;
1386
1387 /*
1388 * At this point it is safe to assume that the "private_data" contains
1389 * our own data structure.
1390 */
1391 ep = file->private_data;
1392
Davide Libenzi22bacca2011-02-25 14:44:12 -08001393 /*
1394 * When we insert an epoll file descriptor, inside another epoll file
1395 * descriptor, there is the change of creating closed loops, which are
1396 * better be handled here, than in more critical paths.
1397 *
1398 * We hold epmutex across the loop check and the insert in this case, in
1399 * order to prevent two separate inserts from racing and each doing the
1400 * insert "at the same time" such that ep_loop_check passes on both
1401 * before either one does the insert, thereby creating a cycle.
1402 */
1403 if (unlikely(is_file_epoll(tfile) && op == EPOLL_CTL_ADD)) {
1404 mutex_lock(&epmutex);
1405 did_lock_epmutex = 1;
1406 error = -ELOOP;
1407 if (ep_loop_check(ep, tfile) != 0)
1408 goto error_tgt_fput;
1409 }
1410
1411
Davide Libenzid47de162007-05-15 01:40:41 -07001412 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001413
Davide Libenzi67647d02007-05-15 01:40:52 -07001414 /*
1415 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1416 * above, we can be sure to be able to use the item looked up by
1417 * ep_find() till we release the mutex.
1418 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001419 epi = ep_find(ep, tfile, fd);
1420
1421 error = -EINVAL;
1422 switch (op) {
1423 case EPOLL_CTL_ADD:
1424 if (!epi) {
1425 epds.events |= POLLERR | POLLHUP;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001426 error = ep_insert(ep, &epds, tfile, fd);
1427 } else
1428 error = -EEXIST;
1429 break;
1430 case EPOLL_CTL_DEL:
1431 if (epi)
1432 error = ep_remove(ep, epi);
1433 else
1434 error = -ENOENT;
1435 break;
1436 case EPOLL_CTL_MOD:
1437 if (epi) {
1438 epds.events |= POLLERR | POLLHUP;
1439 error = ep_modify(ep, epi, &epds);
1440 } else
1441 error = -ENOENT;
1442 break;
1443 }
Davide Libenzid47de162007-05-15 01:40:41 -07001444 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001445
1446error_tgt_fput:
Davide Libenzi22bacca2011-02-25 14:44:12 -08001447 if (unlikely(did_lock_epmutex))
1448 mutex_unlock(&epmutex);
1449
Davide Libenzi7699acd2007-05-10 22:23:23 -07001450 fput(tfile);
1451error_fput:
1452 fput(file);
1453error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001454
1455 return error;
1456}
1457
1458/*
1459 * Implement the event wait interface for the eventpoll file. It is the kernel
1460 * part of the user space epoll_wait(2).
1461 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001462SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1463 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001464{
1465 int error;
1466 struct file *file;
1467 struct eventpoll *ep;
1468
Davide Libenzi7699acd2007-05-10 22:23:23 -07001469 /* The maximum number of event must be greater than zero */
1470 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1471 return -EINVAL;
1472
1473 /* Verify that the area passed by the user is writeable */
1474 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1475 error = -EFAULT;
1476 goto error_return;
1477 }
1478
1479 /* Get the "struct file *" for the eventpoll file */
1480 error = -EBADF;
1481 file = fget(epfd);
1482 if (!file)
1483 goto error_return;
1484
1485 /*
1486 * We have to check that the file structure underneath the fd
1487 * the user passed to us _is_ an eventpoll file.
1488 */
1489 error = -EINVAL;
1490 if (!is_file_epoll(file))
1491 goto error_fput;
1492
1493 /*
1494 * At this point it is safe to assume that the "private_data" contains
1495 * our own data structure.
1496 */
1497 ep = file->private_data;
1498
1499 /* Time to fish for events ... */
1500 error = ep_poll(ep, events, maxevents, timeout);
1501
1502error_fput:
1503 fput(file);
1504error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001505
1506 return error;
1507}
1508
Roland McGrathf3de2722008-04-30 00:53:09 -07001509#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001510
1511/*
1512 * Implement the event wait interface for the eventpoll file. It is the kernel
1513 * part of the user space epoll_pwait(2).
1514 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001515SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1516 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1517 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001518{
1519 int error;
1520 sigset_t ksigmask, sigsaved;
1521
1522 /*
1523 * If the caller wants a certain signal mask to be set during the wait,
1524 * we apply it here.
1525 */
1526 if (sigmask) {
1527 if (sigsetsize != sizeof(sigset_t))
1528 return -EINVAL;
1529 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1530 return -EFAULT;
1531 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1532 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1533 }
1534
1535 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1536
1537 /*
1538 * If we changed the signal mask, we need to restore the original one.
1539 * In case we've got a signal while waiting, we do not restore the
1540 * signal mask yet, and we allow do_signal() to deliver the signal on
1541 * the way back to userspace, before the signal mask is restored.
1542 */
1543 if (sigmask) {
1544 if (error == -EINTR) {
1545 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001546 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001547 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001548 } else
1549 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1550 }
1551
1552 return error;
1553}
1554
Roland McGrathf3de2722008-04-30 00:53:09 -07001555#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557static int __init eventpoll_init(void)
1558{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001559 struct sysinfo si;
1560
1561 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001562 /*
1563 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1564 */
1565 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001566 EP_ITEM_COST;
Robin Holt52bd19f72011-01-12 17:00:01 -08001567 BUG_ON(max_user_watches < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Davide Libenzi22bacca2011-02-25 14:44:12 -08001569 /*
1570 * Initialize the structure used to perform epoll file descriptor
1571 * inclusion loops checks.
1572 */
1573 ep_nested_calls_init(&poll_loop_ncalls);
1574
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001576 ep_nested_calls_init(&poll_safewake_ncalls);
1577
1578 /* Initialize the structure used to perform file's f_op->poll() calls */
1579 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
1581 /* Allocates slab cache used to allocate "struct epitem" items */
1582 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001583 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1586 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001587 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590}
Davide Libenzicea69242007-05-10 22:23:22 -07001591fs_initcall(eventpoll_init);