blob: 4a09af9e9a6387869bf398dbda88723b97d319cf [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>
40#include <asm/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
65 * close()d without a previous call toepoll_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
155 * structure and rapresent the main data sructure for the eventpoll
156 * interface.
157 */
158struct eventpoll {
159 /* Protect the this structure access */
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
184 * happened while transfering ready events to userspace w/out
185 * 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/**
319 * ep_call_nested - Perform a bound (possibly) nested call, by checking
320 * that the recursion limit is not exceeded, and that
321 * the same nested call (by the meaning of same cookie) is
322 * no re-entered.
323 *
324 * @ncalls: Pointer to the nested_calls structure to be used for this call.
325 * @max_nests: Maximum number of allowed nesting calls.
326 * @nproc: Nested call core function pointer.
327 * @priv: Opaque data to be passed to the @nproc callback.
328 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700329 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700330 *
331 * Returns: Returns the code returned by the @nproc callback, or -1 if
332 * the maximum recursion limit has been exceeded.
333 */
334static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
335 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700336 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700337{
338 int error, call_nests = 0;
339 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700340 struct list_head *lsthead = &ncalls->tasks_call_list;
341 struct nested_call_node *tncur;
342 struct nested_call_node tnode;
343
344 spin_lock_irqsave(&ncalls->lock, flags);
345
346 /*
347 * Try to see if the current task is already inside this wakeup call.
348 * We use a list here, since the population inside this set is always
349 * very much limited.
350 */
351 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700352 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700353 (tncur->cookie == cookie || ++call_nests > max_nests)) {
354 /*
355 * Ops ... loop detected or maximum nest level reached.
356 * We abort this wake by breaking the cycle itself.
357 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700358 error = -1;
359 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700360 }
361 }
362
363 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700364 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700365 tnode.cookie = cookie;
366 list_add(&tnode.llink, lsthead);
367
368 spin_unlock_irqrestore(&ncalls->lock, flags);
369
370 /* Call the nested function */
371 error = (*nproc)(priv, cookie, call_nests);
372
373 /* Remove the current task from the list */
374 spin_lock_irqsave(&ncalls->lock, flags);
375 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700376out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700377 spin_unlock_irqrestore(&ncalls->lock, flags);
378
379 return error;
380}
381
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700382#ifdef CONFIG_DEBUG_LOCK_ALLOC
383static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
384 unsigned long events, int subclass)
385{
386 unsigned long flags;
387
388 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
389 wake_up_locked_poll(wqueue, events);
390 spin_unlock_irqrestore(&wqueue->lock, flags);
391}
392#else
393static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
394 unsigned long events, int subclass)
395{
396 wake_up_poll(wqueue, events);
397}
398#endif
399
Davide Libenzi5071f972009-03-31 15:24:10 -0700400static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
401{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700402 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
403 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700404 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405}
406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407/*
408 * Perform a safe wake up of the poll wait list. The problem is that
409 * with the new callback'd wake up system, it is possible that the
410 * poll callback is reentered from inside the call to wake_up() done
411 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700412 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 * and we cannot reenter the same wait queue head at all. This will
414 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700415 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700417static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700419 int this_cpu = get_cpu();
420
Davide Libenzi5071f972009-03-31 15:24:10 -0700421 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700422 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
423
424 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425}
426
Davide Libenzi7699acd2007-05-10 22:23:23 -0700427/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700428 * This function unregisters poll callbacks from the associated file
429 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
430 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700431 */
432static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
433{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700434 struct list_head *lsthead = &epi->pwqlist;
435 struct eppoll_entry *pwq;
436
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700437 while (!list_empty(lsthead)) {
438 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700439
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700440 list_del(&pwq->llink);
441 remove_wait_queue(pwq->whead, &pwq->wait);
442 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700443 }
444}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Davide Libenzi5071f972009-03-31 15:24:10 -0700446/**
447 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
448 * the scan code, to call f_op->poll(). Also allows for
449 * O(NumReady) performance.
450 *
451 * @ep: Pointer to the epoll private data structure.
452 * @sproc: Pointer to the scan callback.
453 * @priv: Private opaque data passed to the @sproc callback.
454 *
455 * Returns: The same integer error code returned by the @sproc callback.
456 */
457static int ep_scan_ready_list(struct eventpoll *ep,
458 int (*sproc)(struct eventpoll *,
459 struct list_head *, void *),
460 void *priv)
461{
462 int error, pwake = 0;
463 unsigned long flags;
464 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700465 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700466
467 /*
468 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700469 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700470 */
471 mutex_lock(&ep->mtx);
472
473 /*
474 * Steal the ready list, and re-init the original one to the
475 * empty list. Also, set ep->ovflist to NULL so that events
476 * happening while looping w/out locks, are not lost. We cannot
477 * have the poll callback to queue directly on ep->rdllist,
478 * because we want the "sproc" callback to be able to do it
479 * in a lockless way.
480 */
481 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700482 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700483 ep->ovflist = NULL;
484 spin_unlock_irqrestore(&ep->lock, flags);
485
486 /*
487 * Now call the callback function.
488 */
489 error = (*sproc)(ep, &txlist, priv);
490
491 spin_lock_irqsave(&ep->lock, flags);
492 /*
493 * During the time we spent inside the "sproc" callback, some
494 * other events might have been queued by the poll callback.
495 * We re-insert them inside the main ready-list here.
496 */
497 for (nepi = ep->ovflist; (epi = nepi) != NULL;
498 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
499 /*
500 * We need to check if the item is already in the list.
501 * During the "sproc" callback execution time, items are
502 * queued into ->ovflist but the "txlist" might already
503 * contain them, and the list_splice() below takes care of them.
504 */
505 if (!ep_is_linked(&epi->rdllink))
506 list_add_tail(&epi->rdllink, &ep->rdllist);
507 }
508 /*
509 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
510 * releasing the lock, events will be queued in the normal way inside
511 * ep->rdllist.
512 */
513 ep->ovflist = EP_UNACTIVE_PTR;
514
515 /*
516 * Quickly re-inject items left on "txlist".
517 */
518 list_splice(&txlist, &ep->rdllist);
519
520 if (!list_empty(&ep->rdllist)) {
521 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700522 * Wake up (if active) both the eventpoll wait list and
523 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700524 */
525 if (waitqueue_active(&ep->wq))
526 wake_up_locked(&ep->wq);
527 if (waitqueue_active(&ep->poll_wait))
528 pwake++;
529 }
530 spin_unlock_irqrestore(&ep->lock, flags);
531
532 mutex_unlock(&ep->mtx);
533
534 /* We have to call this outside the lock */
535 if (pwake)
536 ep_poll_safewake(&ep->poll_wait);
537
538 return error;
539}
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700542 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700543 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700544 */
545static int ep_remove(struct eventpoll *ep, struct epitem *epi)
546{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700547 unsigned long flags;
548 struct file *file = epi->ffd.file;
549
550 /*
551 * Removes poll wait queue hooks. We _have_ to do this without holding
552 * the "ep->lock" otherwise a deadlock might occur. This because of the
553 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
554 * queue head lock when unregistering the wait queue. The wakeup callback
555 * will run by holding the wait queue head lock and will call our callback
556 * that will try to get "ep->lock".
557 */
558 ep_unregister_pollwait(ep, epi);
559
560 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700561 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700562 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700563 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700564 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Davide Libenzicdac75e2008-04-29 00:58:34 -0700566 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Davide Libenzic7ea7632007-05-15 01:40:47 -0700568 spin_lock_irqsave(&ep->lock, flags);
569 if (ep_is_linked(&epi->rdllink))
570 list_del_init(&epi->rdllink);
571 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Davide Libenzi7699acd2007-05-10 22:23:23 -0700573 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700574 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
Robin Holt52bd19f72011-01-12 17:00:01 -0800576 atomic_long_dec(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800577
Davide Libenzic7ea7632007-05-15 01:40:47 -0700578 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579}
580
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581static void ep_free(struct eventpoll *ep)
582{
583 struct rb_node *rbp;
584 struct epitem *epi;
585
586 /* We need to release all tasks waiting for these file */
587 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700588 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 /*
591 * We need to lock this because we could be hit by
592 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700593 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 * is on the way to be removed and no one has references to it
595 * anymore. The only hit might come from eventpoll_release_file() but
Arjan van de Ven144efe32006-03-23 03:00:32 -0800596 * holding "epmutex" is sufficent here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800598 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600 /*
601 * Walks through the whole tree by unregistering poll callbacks.
602 */
603 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
604 epi = rb_entry(rbp, struct epitem, rbn);
605
606 ep_unregister_pollwait(ep, epi);
607 }
608
609 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700610 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700612 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 * us during this operation. So we can avoid the lock on "ep->lock".
614 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700615 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 epi = rb_entry(rbp, struct epitem, rbn);
617 ep_remove(ep, epi);
618 }
619
Arjan van de Ven144efe32006-03-23 03:00:32 -0800620 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700621 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800622 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700623 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624}
625
Davide Libenzi7699acd2007-05-10 22:23:23 -0700626static int ep_eventpoll_release(struct inode *inode, struct file *file)
627{
628 struct eventpoll *ep = file->private_data;
629
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700630 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700631 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700632
Davide Libenzi7699acd2007-05-10 22:23:23 -0700633 return 0;
634}
635
Davide Libenzi296e2362009-03-31 15:24:11 -0700636static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
637 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700638{
639 struct epitem *epi, *tmp;
640
641 list_for_each_entry_safe(epi, tmp, head, rdllink) {
642 if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
643 epi->event.events)
644 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700645 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700646 /*
647 * Item has been dropped into the ready list by the poll
648 * callback, but it's not actually ready, as far as
649 * caller requested events goes. We can remove it here.
650 */
651 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700652 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700653 }
654
655 return 0;
656}
657
658static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
659{
660 return ep_scan_ready_list(priv, ep_read_events_proc, NULL);
661}
662
Davide Libenzi7699acd2007-05-10 22:23:23 -0700663static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
664{
Davide Libenzi5071f972009-03-31 15:24:10 -0700665 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700666 struct eventpoll *ep = file->private_data;
667
668 /* Insert inside our poll wait queue */
669 poll_wait(file, &ep->poll_wait, wait);
670
Davide Libenzi5071f972009-03-31 15:24:10 -0700671 /*
672 * Proceed to find out if wanted events are really available inside
673 * the ready list. This need to be done under ep_call_nested()
674 * supervision, since the call to f_op->poll() done on listed files
675 * could re-enter here.
676 */
677 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700678 ep_poll_readyevents_proc, ep, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700679
Davide Libenzi296e2362009-03-31 15:24:11 -0700680 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700681}
682
683/* File callbacks that implement the eventpoll file behaviour */
684static const struct file_operations eventpoll_fops = {
685 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200686 .poll = ep_eventpoll_poll,
687 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700688};
689
690/* Fast test to see if the file is an evenpoll file */
691static inline int is_file_epoll(struct file *f)
692{
693 return f->f_op == &eventpoll_fops;
694}
695
696/*
697 * This is called from eventpoll_release() to unlink files from the eventpoll
698 * interface. We need to have this facility to cleanup correctly files that are
699 * closed without being removed from the eventpoll interface.
700 */
701void eventpoll_release_file(struct file *file)
702{
703 struct list_head *lsthead = &file->f_ep_links;
704 struct eventpoll *ep;
705 struct epitem *epi;
706
707 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700708 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700709 * necessary. It is not necessary because we're in the "struct file"
710 * cleanup path, and this means that noone is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700711 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700712 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700713 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700714 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700715 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700716 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700717 *
718 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700719 */
720 mutex_lock(&epmutex);
721
722 while (!list_empty(lsthead)) {
723 epi = list_first_entry(lsthead, struct epitem, fllink);
724
725 ep = epi->ep;
726 list_del_init(&epi->fllink);
Davide Libenzid47de162007-05-15 01:40:41 -0700727 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700728 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700729 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700730 }
731
732 mutex_unlock(&epmutex);
733}
734
735static int ep_alloc(struct eventpoll **pep)
736{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800737 int error;
738 struct user_struct *user;
739 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700740
Davide Libenzi7ef99642008-12-01 13:13:55 -0800741 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800742 error = -ENOMEM;
743 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
744 if (unlikely(!ep))
745 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700746
Davide Libenzic7ea7632007-05-15 01:40:47 -0700747 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700748 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700749 init_waitqueue_head(&ep->wq);
750 init_waitqueue_head(&ep->poll_wait);
751 INIT_LIST_HEAD(&ep->rdllist);
752 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700753 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800754 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700755
756 *pep = ep;
757
Davide Libenzi7699acd2007-05-10 22:23:23 -0700758 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800759
760free_uid:
761 free_uid(user);
762 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700763}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700766 * Search the file inside the eventpoll tree. The RB tree operations
767 * are protected by the "mtx" mutex, and ep_find() must be called with
768 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 */
770static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
771{
772 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 struct rb_node *rbp;
774 struct epitem *epi, *epir = NULL;
775 struct epoll_filefd ffd;
776
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700777 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 for (rbp = ep->rbr.rb_node; rbp; ) {
779 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700780 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 if (kcmp > 0)
782 rbp = rbp->rb_right;
783 else if (kcmp < 0)
784 rbp = rbp->rb_left;
785 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 epir = epi;
787 break;
788 }
789 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return epir;
792}
793
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700795 * This is the callback that is passed to the wait queue wakeup
796 * machanism. It is called by the stored file descriptors when they
797 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700799static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700801 int pwake = 0;
802 unsigned long flags;
803 struct epitem *epi = ep_item_from_wait(wait);
804 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Davide Libenzic7ea7632007-05-15 01:40:47 -0700806 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700807
808 /*
809 * If the event mask does not contain any poll(2) event, we consider the
810 * descriptor to be disabled. This condition is likely the effect of the
811 * EPOLLONESHOT bit that disables the descriptor when an event is received,
812 * until the next EPOLL_CTL_MOD will be issued.
813 */
814 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700815 goto out_unlock;
816
817 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700818 * Check the events coming with the callback. At this stage, not
819 * every device reports the events in the "key" parameter of the
820 * callback. We need to be able to handle both cases here, hence the
821 * test for "key" != NULL before the event match test.
822 */
823 if (key && !((unsigned long) key & epi->event.events))
824 goto out_unlock;
825
826 /*
Davide Libenzid47de162007-05-15 01:40:41 -0700827 * If we are trasfering events to userspace, we can hold no locks
828 * (because we're accessing user memory, and because of linux f_op->poll()
829 * semantics). All the events that happens during that period of time are
830 * chained in ep->ovflist and requeued later on.
831 */
832 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
833 if (epi->next == EP_UNACTIVE_PTR) {
834 epi->next = ep->ovflist;
835 ep->ovflist = epi;
836 }
837 goto out_unlock;
838 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700839
840 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700841 if (!ep_is_linked(&epi->rdllink))
842 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700843
Davide Libenzi7699acd2007-05-10 22:23:23 -0700844 /*
845 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
846 * wait list.
847 */
848 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400849 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700850 if (waitqueue_active(&ep->poll_wait))
851 pwake++;
852
Davide Libenzid47de162007-05-15 01:40:41 -0700853out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700854 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700855
856 /* We have to call this outside the lock */
857 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700858 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700859
860 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861}
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863/*
864 * This is the callback that is used to add our wait queue to the
865 * target file wakeup lists.
866 */
867static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
868 poll_table *pt)
869{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700870 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 struct eppoll_entry *pwq;
872
Christoph Lametere94b1762006-12-06 20:33:17 -0800873 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
875 pwq->whead = whead;
876 pwq->base = epi;
877 add_wait_queue(whead, &pwq->wait);
878 list_add_tail(&pwq->llink, &epi->pwqlist);
879 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700880 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 /* We have to signal that an error occurred */
882 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
887{
888 int kcmp;
889 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
890 struct epitem *epic;
891
892 while (*p) {
893 parent = *p;
894 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700895 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (kcmp > 0)
897 p = &parent->rb_right;
898 else
899 p = &parent->rb_left;
900 }
901 rb_link_node(&epi->rbn, parent, p);
902 rb_insert_color(&epi->rbn, &ep->rbr);
903}
904
Davide Libenzic7ea7632007-05-15 01:40:47 -0700905/*
906 * Must be called with "mtx" held.
907 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
909 struct file *tfile, int fd)
910{
911 int error, revents, pwake = 0;
912 unsigned long flags;
Robin Holt52bd19f72011-01-12 17:00:01 -0800913 long user_watches;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 struct epitem *epi;
915 struct ep_pqueue epq;
916
Robin Holt52bd19f72011-01-12 17:00:01 -0800917 user_watches = atomic_long_read(&ep->user->epoll_watches);
918 if (unlikely(user_watches >= max_user_watches))
Davide Libenzi7ef99642008-12-01 13:13:55 -0800919 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -0800920 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -0800921 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 INIT_LIST_HEAD(&epi->rdllink);
925 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 INIT_LIST_HEAD(&epi->pwqlist);
927 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700928 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -0700931 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933 /* Initialize the poll table using the queue callback */
934 epq.epi = epi;
935 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
936
937 /*
938 * Attach the item to the poll hooks and get current event bits.
939 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -0700940 * been increased by the caller of this function. Note that after
941 * this operation completes, the poll callback can start hitting
942 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 */
944 revents = tfile->f_op->poll(tfile, &epq.pt);
945
946 /*
947 * We have to check if something went wrong during the poll wait queue
948 * install process. Namely an allocation for a wait queue failed due
949 * high memory pressure.
950 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800951 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700953 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
955 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -0700956 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -0700958 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Davide Libenzic7ea7632007-05-15 01:40:47 -0700960 /*
961 * Add the current item to the RB tree. All RB tree operations are
962 * protected by "mtx", and ep_insert() is called with "mtx" held.
963 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 ep_rbtree_insert(ep, epi);
965
Davide Libenzic7ea7632007-05-15 01:40:47 -0700966 /* We have to drop the new item inside our item list to keep track of it */
967 spin_lock_irqsave(&ep->lock, flags);
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700970 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 list_add_tail(&epi->rdllink, &ep->rdllist);
972
973 /* Notify waiting tasks that events are available */
974 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400975 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (waitqueue_active(&ep->poll_wait))
977 pwake++;
978 }
979
Davide Libenzic7ea7632007-05-15 01:40:47 -0700980 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Robin Holt52bd19f72011-01-12 17:00:01 -0800982 atomic_long_inc(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /* We have to call this outside the lock */
985 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700986 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 return 0;
989
Davide Libenzi7699acd2007-05-10 22:23:23 -0700990error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 ep_unregister_pollwait(ep, epi);
992
993 /*
994 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -0700995 * allocated wait queue. Note that we don't care about the ep->ovflist
996 * list, since that is used/cleaned only inside a section bound by "mtx".
997 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700999 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001000 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -07001001 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001002 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001004 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001005
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 return error;
1007}
1008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009/*
1010 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001011 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 */
1013static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1014{
1015 int pwake = 0;
1016 unsigned int revents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001019 * Set the new event interest mask before calling f_op->poll();
1020 * otherwise we might miss an event that happens between the
1021 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 */
1023 epi->event.events = event->events;
Tony Battersbye057e152009-03-31 15:24:15 -07001024 epi->event.data = event->data; /* protected by mtx */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026 /*
1027 * Get current event bits. We can safely use the file* here because
1028 * its usage count has been increased by the caller of this function.
1029 */
1030 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL);
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001033 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001034 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001036 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001037 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001038 if (!ep_is_linked(&epi->rdllink)) {
1039 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Davide Libenzic7ea7632007-05-15 01:40:47 -07001041 /* Notify waiting tasks that events are available */
1042 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001043 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001044 if (waitqueue_active(&ep->poll_wait))
1045 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 }
Tony Battersbye057e152009-03-31 15:24:15 -07001047 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
1050 /* We have to call this outside the lock */
1051 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001052 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
1054 return 0;
1055}
1056
Davide Libenzi296e2362009-03-31 15:24:11 -07001057static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1058 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001059{
1060 struct ep_send_events_data *esed = priv;
1061 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001062 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001063 struct epitem *epi;
1064 struct epoll_event __user *uevent;
1065
Davide Libenzi296e2362009-03-31 15:24:11 -07001066 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001067 * We can loop without lock because we are passed a task private list.
1068 * Items cannot vanish during the loop because ep_scan_ready_list() is
1069 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001070 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001071 for (eventcnt = 0, uevent = esed->events;
1072 !list_empty(head) && eventcnt < esed->maxevents;) {
1073 epi = list_first_entry(head, struct epitem, rdllink);
1074
1075 list_del_init(&epi->rdllink);
1076
Davide Libenzi296e2362009-03-31 15:24:11 -07001077 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
1078 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001079
Davide Libenzi296e2362009-03-31 15:24:11 -07001080 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001081 * If the event mask intersect the caller-requested one,
1082 * deliver the event to userspace. Again, ep_scan_ready_list()
1083 * is holding "mtx", so no operations coming from userspace
1084 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001085 */
1086 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001087 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001088 __put_user(epi->event.data, &uevent->data)) {
1089 list_add(&epi->rdllink, head);
Davide Libenzi296e2362009-03-31 15:24:11 -07001090 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001091 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001092 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001093 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001094 if (epi->event.events & EPOLLONESHOT)
1095 epi->event.events &= EP_PRIVATE_BITS;
1096 else if (!(epi->event.events & EPOLLET)) {
1097 /*
1098 * If this file has been added with Level
1099 * Trigger mode, we need to insert back inside
1100 * the ready list, so that the next call to
1101 * epoll_wait() will check again the events
1102 * availability. At this point, noone can insert
1103 * into ep->rdllist besides us. The epoll_ctl()
1104 * callers are locked out by
1105 * ep_scan_ready_list() holding "mtx" and the
1106 * poll callback will queue them in ep->ovflist.
1107 */
1108 list_add_tail(&epi->rdllink, &ep->rdllist);
1109 }
1110 }
1111 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001112
1113 return eventcnt;
1114}
1115
Davide Libenzi296e2362009-03-31 15:24:11 -07001116static int ep_send_events(struct eventpoll *ep,
1117 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118{
Davide Libenzi5071f972009-03-31 15:24:10 -07001119 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001120
Davide Libenzi5071f972009-03-31 15:24:10 -07001121 esed.maxevents = maxevents;
1122 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001123
Davide Libenzi5071f972009-03-31 15:24:10 -07001124 return ep_scan_ready_list(ep, ep_send_events_proc, &esed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
1126
Eric Dumazet0781b902011-02-01 15:52:35 -08001127static inline struct timespec ep_set_mstimeout(long ms)
1128{
1129 struct timespec now, ts = {
1130 .tv_sec = ms / MSEC_PER_SEC,
1131 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1132 };
1133
1134 ktime_get_ts(&now);
1135 return timespec_add_safe(now, ts);
1136}
1137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1139 int maxevents, long timeout)
1140{
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001141 int res, eavail, timed_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 unsigned long flags;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001143 long slack;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 wait_queue_t wait;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001145 ktime_t expires, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001147 if (timeout > 0) {
Eric Dumazet0781b902011-02-01 15:52:35 -08001148 struct timespec end_time = ep_set_mstimeout(timeout);
1149
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001150 slack = select_estimate_accuracy(&end_time);
1151 to = &expires;
1152 *to = timespec_to_ktime(end_time);
1153 } else if (timeout == 0) {
1154 timed_out = 1;
1155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
1157retry:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001158 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 res = 0;
1161 if (list_empty(&ep->rdllist)) {
1162 /*
1163 * We don't have any available event to return to the caller.
1164 * We need to sleep here, and we will be wake up by
1165 * ep_poll_callback() when events will become available.
1166 */
1167 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001168 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 for (;;) {
1171 /*
1172 * We don't want to sleep if the ep_poll_callback() sends us
1173 * a wakeup in between. That's why we set the task state
1174 * to TASK_INTERRUPTIBLE before doing the checks.
1175 */
1176 set_current_state(TASK_INTERRUPTIBLE);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001177 if (!list_empty(&ep->rdllist) || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 break;
1179 if (signal_pending(current)) {
1180 res = -EINTR;
1181 break;
1182 }
1183
Davide Libenzic7ea7632007-05-15 01:40:47 -07001184 spin_unlock_irqrestore(&ep->lock, flags);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001185 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1186 timed_out = 1;
1187
Davide Libenzic7ea7632007-05-15 01:40:47 -07001188 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001190 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
1192 set_current_state(TASK_RUNNING);
1193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 /* Is it worth to try to dig for events ? */
Davide Libenzi5071f972009-03-31 15:24:10 -07001195 eavail = !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Davide Libenzic7ea7632007-05-15 01:40:47 -07001197 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 /*
1200 * Try to transfer events to user space. In case we get 0 events and
1201 * there's still timeout left over, we go trying again in search of
1202 * more luck.
1203 */
1204 if (!res && eavail &&
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001205 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 goto retry;
1207
1208 return res;
1209}
1210
Davide Libenzi22bacca2011-02-25 14:44:12 -08001211/**
1212 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1213 * API, to verify that adding an epoll file inside another
1214 * epoll structure, does not violate the constraints, in
1215 * terms of closed loops, or too deep chains (which can
1216 * result in excessive stack usage).
1217 *
1218 * @priv: Pointer to the epoll file to be currently checked.
1219 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1220 * data structure pointer.
1221 * @call_nests: Current dept of the @ep_call_nested() call stack.
1222 *
1223 * Returns: Returns zero if adding the epoll @file inside current epoll
1224 * structure @ep does not violate the constraints, or -1 otherwise.
1225 */
1226static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1227{
1228 int error = 0;
1229 struct file *file = priv;
1230 struct eventpoll *ep = file->private_data;
1231 struct rb_node *rbp;
1232 struct epitem *epi;
1233
1234 mutex_lock(&ep->mtx);
1235 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1236 epi = rb_entry(rbp, struct epitem, rbn);
1237 if (unlikely(is_file_epoll(epi->ffd.file))) {
1238 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1239 ep_loop_check_proc, epi->ffd.file,
1240 epi->ffd.file->private_data, current);
1241 if (error != 0)
1242 break;
1243 }
1244 }
1245 mutex_unlock(&ep->mtx);
1246
1247 return error;
1248}
1249
1250/**
1251 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1252 * another epoll file (represented by @ep) does not create
1253 * closed loops or too deep chains.
1254 *
1255 * @ep: Pointer to the epoll private data structure.
1256 * @file: Pointer to the epoll file to be checked.
1257 *
1258 * Returns: Returns zero if adding the epoll @file inside current epoll
1259 * structure @ep does not violate the constraints, or -1 otherwise.
1260 */
1261static int ep_loop_check(struct eventpoll *ep, struct file *file)
1262{
1263 return ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1264 ep_loop_check_proc, file, ep, current);
1265}
1266
Davide Libenzi7699acd2007-05-10 22:23:23 -07001267/*
Andrew Morton523723b2008-08-12 15:09:01 -07001268 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001269 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001270SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001271{
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001272 int error;
1273 struct eventpoll *ep = NULL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001274
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001275 /* Check the EPOLL_* constant for consistency. */
1276 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1277
Davide Libenzi296e2362009-03-31 15:24:11 -07001278 if (flags & ~EPOLL_CLOEXEC)
1279 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001280 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001281 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001282 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001283 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001284 if (error < 0)
1285 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001286 /*
1287 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001288 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001289 */
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001290 error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001291 O_RDWR | (flags & O_CLOEXEC));
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001292 if (error < 0)
Al Viro2030a422008-02-23 06:46:49 -05001293 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001294
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001295 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001296}
1297
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001298SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001299{
Davide Libenzibfe38912009-05-12 13:19:44 -07001300 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001301 return -EINVAL;
1302
1303 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001304}
1305
Davide Libenzi7699acd2007-05-10 22:23:23 -07001306/*
1307 * The following function implements the controller interface for
1308 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001309 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001310 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001311SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1312 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001313{
1314 int error;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001315 int did_lock_epmutex = 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001316 struct file *file, *tfile;
1317 struct eventpoll *ep;
1318 struct epitem *epi;
1319 struct epoll_event epds;
1320
Davide Libenzi7699acd2007-05-10 22:23:23 -07001321 error = -EFAULT;
1322 if (ep_op_has_event(op) &&
1323 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1324 goto error_return;
1325
1326 /* Get the "struct file *" for the eventpoll file */
1327 error = -EBADF;
1328 file = fget(epfd);
1329 if (!file)
1330 goto error_return;
1331
1332 /* Get the "struct file *" for the target file */
1333 tfile = fget(fd);
1334 if (!tfile)
1335 goto error_fput;
1336
1337 /* The target file descriptor must support poll */
1338 error = -EPERM;
1339 if (!tfile->f_op || !tfile->f_op->poll)
1340 goto error_tgt_fput;
1341
1342 /*
1343 * We have to check that the file structure underneath the file descriptor
1344 * the user passed to us _is_ an eventpoll file. And also we do not permit
1345 * adding an epoll file descriptor inside itself.
1346 */
1347 error = -EINVAL;
1348 if (file == tfile || !is_file_epoll(file))
1349 goto error_tgt_fput;
1350
1351 /*
1352 * At this point it is safe to assume that the "private_data" contains
1353 * our own data structure.
1354 */
1355 ep = file->private_data;
1356
Davide Libenzi22bacca2011-02-25 14:44:12 -08001357 /*
1358 * When we insert an epoll file descriptor, inside another epoll file
1359 * descriptor, there is the change of creating closed loops, which are
1360 * better be handled here, than in more critical paths.
1361 *
1362 * We hold epmutex across the loop check and the insert in this case, in
1363 * order to prevent two separate inserts from racing and each doing the
1364 * insert "at the same time" such that ep_loop_check passes on both
1365 * before either one does the insert, thereby creating a cycle.
1366 */
1367 if (unlikely(is_file_epoll(tfile) && op == EPOLL_CTL_ADD)) {
1368 mutex_lock(&epmutex);
1369 did_lock_epmutex = 1;
1370 error = -ELOOP;
1371 if (ep_loop_check(ep, tfile) != 0)
1372 goto error_tgt_fput;
1373 }
1374
1375
Davide Libenzid47de162007-05-15 01:40:41 -07001376 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001377
Davide Libenzi67647d02007-05-15 01:40:52 -07001378 /*
1379 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1380 * above, we can be sure to be able to use the item looked up by
1381 * ep_find() till we release the mutex.
1382 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001383 epi = ep_find(ep, tfile, fd);
1384
1385 error = -EINVAL;
1386 switch (op) {
1387 case EPOLL_CTL_ADD:
1388 if (!epi) {
1389 epds.events |= POLLERR | POLLHUP;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001390 error = ep_insert(ep, &epds, tfile, fd);
1391 } else
1392 error = -EEXIST;
1393 break;
1394 case EPOLL_CTL_DEL:
1395 if (epi)
1396 error = ep_remove(ep, epi);
1397 else
1398 error = -ENOENT;
1399 break;
1400 case EPOLL_CTL_MOD:
1401 if (epi) {
1402 epds.events |= POLLERR | POLLHUP;
1403 error = ep_modify(ep, epi, &epds);
1404 } else
1405 error = -ENOENT;
1406 break;
1407 }
Davide Libenzid47de162007-05-15 01:40:41 -07001408 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001409
1410error_tgt_fput:
Davide Libenzi22bacca2011-02-25 14:44:12 -08001411 if (unlikely(did_lock_epmutex))
1412 mutex_unlock(&epmutex);
1413
Davide Libenzi7699acd2007-05-10 22:23:23 -07001414 fput(tfile);
1415error_fput:
1416 fput(file);
1417error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001418
1419 return error;
1420}
1421
1422/*
1423 * Implement the event wait interface for the eventpoll file. It is the kernel
1424 * part of the user space epoll_wait(2).
1425 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001426SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1427 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001428{
1429 int error;
1430 struct file *file;
1431 struct eventpoll *ep;
1432
Davide Libenzi7699acd2007-05-10 22:23:23 -07001433 /* The maximum number of event must be greater than zero */
1434 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1435 return -EINVAL;
1436
1437 /* Verify that the area passed by the user is writeable */
1438 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1439 error = -EFAULT;
1440 goto error_return;
1441 }
1442
1443 /* Get the "struct file *" for the eventpoll file */
1444 error = -EBADF;
1445 file = fget(epfd);
1446 if (!file)
1447 goto error_return;
1448
1449 /*
1450 * We have to check that the file structure underneath the fd
1451 * the user passed to us _is_ an eventpoll file.
1452 */
1453 error = -EINVAL;
1454 if (!is_file_epoll(file))
1455 goto error_fput;
1456
1457 /*
1458 * At this point it is safe to assume that the "private_data" contains
1459 * our own data structure.
1460 */
1461 ep = file->private_data;
1462
1463 /* Time to fish for events ... */
1464 error = ep_poll(ep, events, maxevents, timeout);
1465
1466error_fput:
1467 fput(file);
1468error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001469
1470 return error;
1471}
1472
Roland McGrathf3de2722008-04-30 00:53:09 -07001473#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001474
1475/*
1476 * Implement the event wait interface for the eventpoll file. It is the kernel
1477 * part of the user space epoll_pwait(2).
1478 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001479SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1480 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1481 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001482{
1483 int error;
1484 sigset_t ksigmask, sigsaved;
1485
1486 /*
1487 * If the caller wants a certain signal mask to be set during the wait,
1488 * we apply it here.
1489 */
1490 if (sigmask) {
1491 if (sigsetsize != sizeof(sigset_t))
1492 return -EINVAL;
1493 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1494 return -EFAULT;
1495 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1496 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1497 }
1498
1499 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1500
1501 /*
1502 * If we changed the signal mask, we need to restore the original one.
1503 * In case we've got a signal while waiting, we do not restore the
1504 * signal mask yet, and we allow do_signal() to deliver the signal on
1505 * the way back to userspace, before the signal mask is restored.
1506 */
1507 if (sigmask) {
1508 if (error == -EINTR) {
1509 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001510 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001511 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001512 } else
1513 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1514 }
1515
1516 return error;
1517}
1518
Roland McGrathf3de2722008-04-30 00:53:09 -07001519#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521static int __init eventpoll_init(void)
1522{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001523 struct sysinfo si;
1524
1525 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001526 /*
1527 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1528 */
1529 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001530 EP_ITEM_COST;
Robin Holt52bd19f72011-01-12 17:00:01 -08001531 BUG_ON(max_user_watches < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Davide Libenzi22bacca2011-02-25 14:44:12 -08001533 /*
1534 * Initialize the structure used to perform epoll file descriptor
1535 * inclusion loops checks.
1536 */
1537 ep_nested_calls_init(&poll_loop_ncalls);
1538
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001540 ep_nested_calls_init(&poll_safewake_ncalls);
1541
1542 /* Initialize the structure used to perform file's f_op->poll() calls */
1543 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545 /* Allocates slab cache used to allocate "struct epitem" items */
1546 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001547 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1550 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001551 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554}
Davide Libenzicea69242007-05-10 22:23:22 -07001555fs_initcall(eventpoll_init);