blob: aabdfc38cf2499817f150eace1efe8e8f08f1f8f [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 Sharma600634972011-07-26 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.
Nelson Elhaged8805e62011-10-31 17:13:14 -070073 * It is necessary to acquire multiple "ep->mtx"es at once in the
74 * case when one epoll fd is added to another. In this case, we
75 * always acquire the locks in the order of nesting (i.e. after
76 * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
77 * before e2->mtx). Since we disallow cycles of epoll file
78 * descriptors, this ensures that the mutexes are well-ordered. In
79 * order to communicate this nesting to lockdep, when walking a tree
80 * of epoll file descriptors, we use the current recursion depth as
81 * the lockdep subkey.
Davide Libenzid47de162007-05-15 01:40:41 -070082 * It is possible to drop the "ep->mtx" and to use the global
83 * mutex "epmutex" (together with "ep->lock") to have it working,
84 * but having "ep->mtx" will make the interface more scalable.
Arjan van de Ven144efe32006-03-23 03:00:32 -080085 * Events that require holding "epmutex" are very rare, while for
Davide Libenzid47de162007-05-15 01:40:41 -070086 * normal operations the epoll private "ep->mtx" will guarantee
87 * a better scalability.
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 */
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/* Epoll private bits inside the event mask */
91#define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET)
92
Davide Libenzi5071f972009-03-31 15:24:10 -070093/* Maximum number of nesting allowed inside epoll sets */
94#define EP_MAX_NESTS 4
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Davide Libenzib6119672006-10-11 01:21:44 -070096#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
97
Davide Libenzid47de162007-05-15 01:40:41 -070098#define EP_UNACTIVE_PTR ((void *) -1L)
99
Davide Libenzi7ef99642008-12-01 13:13:55 -0800100#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102struct epoll_filefd {
103 struct file *file;
104 int fd;
105};
106
107/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700108 * Structure used to track possible nested calls, for too deep recursions
109 * and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700111struct nested_call_node {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 struct list_head llink;
Davide Libenzi5071f972009-03-31 15:24:10 -0700113 void *cookie;
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700114 void *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115};
116
117/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700118 * This structure is used as collector for nested calls, to check for
119 * maximum recursion dept and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700121struct nested_calls {
122 struct list_head tasks_call_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 spinlock_t lock;
124};
125
126/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 * Each file descriptor added to the eventpoll interface will
Davide Libenzi6192bd52007-05-08 00:25:41 -0700128 * have an entry of this type linked to the "rbr" RB tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 */
130struct epitem {
Davide Libenzi67647d02007-05-15 01:40:52 -0700131 /* RB tree node used to link this structure to the eventpoll RB tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 struct rb_node rbn;
133
134 /* List header used to link this structure to the eventpoll ready list */
135 struct list_head rdllink;
136
Davide Libenzic7ea7632007-05-15 01:40:47 -0700137 /*
138 * Works together "struct eventpoll"->ovflist in keeping the
139 * single linked chain of items.
140 */
141 struct epitem *next;
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 /* The file descriptor information this item refers to */
144 struct epoll_filefd ffd;
145
146 /* Number of active wait queue attached to poll operations */
147 int nwait;
148
149 /* List containing poll wait queues */
150 struct list_head pwqlist;
151
152 /* The "container" of this item */
153 struct eventpoll *ep;
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 /* List header used to link this item to the "struct file" items list */
156 struct list_head fllink;
Davide Libenzid47de162007-05-15 01:40:41 -0700157
Davide Libenzic7ea7632007-05-15 01:40:47 -0700158 /* The structure that describe the interested events and the source fd */
159 struct epoll_event event;
Davide Libenzid47de162007-05-15 01:40:41 -0700160};
161
162/*
163 * This structure is stored inside the "private_data" member of the file
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200164 * structure and represents the main data structure for the eventpoll
Davide Libenzid47de162007-05-15 01:40:41 -0700165 * interface.
166 */
167struct eventpoll {
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200168 /* Protect the access to this structure */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700169 spinlock_t lock;
Davide Libenzid47de162007-05-15 01:40:41 -0700170
171 /*
172 * This mutex is used to ensure that files are not removed
173 * while epoll is using them. This is held during the event
174 * collection loop, the file cleanup path, the epoll file exit
175 * code and the ctl operations.
176 */
177 struct mutex mtx;
178
179 /* Wait queue used by sys_epoll_wait() */
180 wait_queue_head_t wq;
181
182 /* Wait queue used by file->poll() */
183 wait_queue_head_t poll_wait;
184
185 /* List of ready file descriptors */
186 struct list_head rdllist;
187
Davide Libenzi67647d02007-05-15 01:40:52 -0700188 /* RB tree root used to store monitored fd structs */
Davide Libenzid47de162007-05-15 01:40:41 -0700189 struct rb_root rbr;
190
191 /*
192 * This is a single linked list that chains all the "struct epitem" that
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300193 * happened while transferring ready events to userspace w/out
Davide Libenzid47de162007-05-15 01:40:41 -0700194 * holding ->lock.
195 */
196 struct epitem *ovflist;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800197
198 /* The user that created the eventpoll descriptor */
199 struct user_struct *user;
Jason Baron28d82dc2012-01-12 17:17:43 -0800200
201 struct file *file;
202
203 /* used to optimize loop detection check */
204 int visited;
205 struct list_head visited_list_link;
Davide Libenzid47de162007-05-15 01:40:41 -0700206};
207
208/* Wait structure used by the poll hooks */
209struct eppoll_entry {
210 /* List header used to link this structure to the "struct epitem" */
211 struct list_head llink;
212
213 /* The "base" pointer is set to the container "struct epitem" */
Tony Battersby4f0989d2009-03-31 15:24:16 -0700214 struct epitem *base;
Davide Libenzid47de162007-05-15 01:40:41 -0700215
216 /*
217 * Wait queue item that will be linked to the target file wait
218 * queue head.
219 */
220 wait_queue_t wait;
221
222 /* The wait queue head that linked the "wait" wait queue item */
223 wait_queue_head_t *whead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224};
225
226/* Wrapper struct used by poll queueing */
227struct ep_pqueue {
228 poll_table pt;
229 struct epitem *epi;
230};
231
Davide Libenzi5071f972009-03-31 15:24:10 -0700232/* Used by the ep_send_events() function as callback private data */
233struct ep_send_events_data {
234 int maxevents;
235 struct epoll_event __user *events;
236};
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238/*
Davide Libenzi7ef99642008-12-01 13:13:55 -0800239 * Configuration options available inside /proc/sys/fs/epoll/
240 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800241/* Maximum number of epoll watched descriptors, per user */
Robin Holt52bd19f72011-01-12 17:00:01 -0800242static long max_user_watches __read_mostly;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800243
244/*
Davide Libenzid47de162007-05-15 01:40:41 -0700245 * This mutex is used to serialize ep_free() and eventpoll_release_file().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800247static DEFINE_MUTEX(epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Davide Libenzi22bacca2011-02-25 14:44:12 -0800249/* Used to check for epoll file descriptor inclusion loops */
250static struct nested_calls poll_loop_ncalls;
251
Davide Libenzi5071f972009-03-31 15:24:10 -0700252/* Used for safe wake up implementation */
253static struct nested_calls poll_safewake_ncalls;
254
255/* Used to call file's f_op->poll() under the nested calls boundaries */
256static struct nested_calls poll_readywalk_ncalls;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258/* Slab cache used to allocate "struct epitem" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800259static struct kmem_cache *epi_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261/* Slab cache used to allocate "struct eppoll_entry" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800262static struct kmem_cache *pwq_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
Jason Baron28d82dc2012-01-12 17:17:43 -0800264/* Visited nodes during ep_loop_check(), so we can unset them when we finish */
265static LIST_HEAD(visited_list);
266
267/*
268 * List of files with newly added links, where we may need to limit the number
269 * of emanating paths. Protected by the epmutex.
270 */
271static LIST_HEAD(tfile_check_list);
272
Davide Libenzi7ef99642008-12-01 13:13:55 -0800273#ifdef CONFIG_SYSCTL
274
275#include <linux/sysctl.h>
276
Robin Holt52bd19f72011-01-12 17:00:01 -0800277static long zero;
278static long long_max = LONG_MAX;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800279
280ctl_table epoll_table[] = {
281 {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800282 .procname = "max_user_watches",
283 .data = &max_user_watches,
Robin Holt52bd19f72011-01-12 17:00:01 -0800284 .maxlen = sizeof(max_user_watches),
Davide Libenzi7ef99642008-12-01 13:13:55 -0800285 .mode = 0644,
Robin Holt52bd19f72011-01-12 17:00:01 -0800286 .proc_handler = proc_doulongvec_minmax,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800287 .extra1 = &zero,
Robin Holt52bd19f72011-01-12 17:00:01 -0800288 .extra2 = &long_max,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800289 },
Eric W. Biedermanab092032009-11-05 14:25:10 -0800290 { }
Davide Libenzi7ef99642008-12-01 13:13:55 -0800291};
292#endif /* CONFIG_SYSCTL */
293
Jason Baron28d82dc2012-01-12 17:17:43 -0800294static const struct file_operations eventpoll_fops;
295
296static inline int is_file_epoll(struct file *f)
297{
298 return f->f_op == &eventpoll_fops;
299}
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700300
Davide Libenzi67647d02007-05-15 01:40:52 -0700301/* Setup the structure that is used as key for the RB tree */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700302static inline void ep_set_ffd(struct epoll_filefd *ffd,
303 struct file *file, int fd)
304{
305 ffd->file = file;
306 ffd->fd = fd;
307}
308
Davide Libenzi67647d02007-05-15 01:40:52 -0700309/* Compare RB tree keys */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700310static inline int ep_cmp_ffd(struct epoll_filefd *p1,
311 struct epoll_filefd *p2)
312{
313 return (p1->file > p2->file ? +1:
314 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
315}
316
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700317/* Tells us if the item is currently linked */
318static inline int ep_is_linked(struct list_head *p)
319{
320 return !list_empty(p);
321}
322
323/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700324static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700325{
326 return container_of(p, struct eppoll_entry, wait)->base;
327}
328
329/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700330static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700331{
332 return container_of(p, struct ep_pqueue, pt)->epi;
333}
334
335/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700336static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700337{
338 return op != EPOLL_CTL_DEL;
339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700342static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Davide Libenzi5071f972009-03-31 15:24:10 -0700344 INIT_LIST_HEAD(&ncalls->tasks_call_list);
345 spin_lock_init(&ncalls->lock);
346}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Davide Libenzi5071f972009-03-31 15:24:10 -0700348/**
Davide Libenzi3fb0e582011-03-22 16:34:46 -0700349 * ep_events_available - Checks if ready events might be available.
350 *
351 * @ep: Pointer to the eventpoll context.
352 *
353 * Returns: Returns a value different than zero if ready events are available,
354 * or zero otherwise.
355 */
356static inline int ep_events_available(struct eventpoll *ep)
357{
358 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
359}
360
361/**
Davide Libenzi5071f972009-03-31 15:24:10 -0700362 * ep_call_nested - Perform a bound (possibly) nested call, by checking
363 * that the recursion limit is not exceeded, and that
364 * the same nested call (by the meaning of same cookie) is
365 * no re-entered.
366 *
367 * @ncalls: Pointer to the nested_calls structure to be used for this call.
368 * @max_nests: Maximum number of allowed nesting calls.
369 * @nproc: Nested call core function pointer.
370 * @priv: Opaque data to be passed to the @nproc callback.
371 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700372 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700373 *
374 * Returns: Returns the code returned by the @nproc callback, or -1 if
375 * the maximum recursion limit has been exceeded.
376 */
377static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
378 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700379 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700380{
381 int error, call_nests = 0;
382 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700383 struct list_head *lsthead = &ncalls->tasks_call_list;
384 struct nested_call_node *tncur;
385 struct nested_call_node tnode;
386
387 spin_lock_irqsave(&ncalls->lock, flags);
388
389 /*
390 * Try to see if the current task is already inside this wakeup call.
391 * We use a list here, since the population inside this set is always
392 * very much limited.
393 */
394 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700395 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700396 (tncur->cookie == cookie || ++call_nests > max_nests)) {
397 /*
398 * Ops ... loop detected or maximum nest level reached.
399 * We abort this wake by breaking the cycle itself.
400 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700401 error = -1;
402 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700403 }
404 }
405
406 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700407 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700408 tnode.cookie = cookie;
409 list_add(&tnode.llink, lsthead);
410
411 spin_unlock_irqrestore(&ncalls->lock, flags);
412
413 /* Call the nested function */
414 error = (*nproc)(priv, cookie, call_nests);
415
416 /* Remove the current task from the list */
417 spin_lock_irqsave(&ncalls->lock, flags);
418 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700419out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700420 spin_unlock_irqrestore(&ncalls->lock, flags);
421
422 return error;
423}
424
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700425#ifdef CONFIG_DEBUG_LOCK_ALLOC
426static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
427 unsigned long events, int subclass)
428{
429 unsigned long flags;
430
431 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
432 wake_up_locked_poll(wqueue, events);
433 spin_unlock_irqrestore(&wqueue->lock, flags);
434}
435#else
436static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
437 unsigned long events, int subclass)
438{
439 wake_up_poll(wqueue, events);
440}
441#endif
442
Davide Libenzi5071f972009-03-31 15:24:10 -0700443static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
444{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700445 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
446 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700447 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450/*
451 * Perform a safe wake up of the poll wait list. The problem is that
452 * with the new callback'd wake up system, it is possible that the
453 * poll callback is reentered from inside the call to wake_up() done
454 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700455 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 * and we cannot reenter the same wait queue head at all. This will
457 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700458 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700460static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700462 int this_cpu = get_cpu();
463
Davide Libenzi5071f972009-03-31 15:24:10 -0700464 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700465 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
466
467 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
Davide Libenzi7699acd2007-05-10 22:23:23 -0700470/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700471 * This function unregisters poll callbacks from the associated file
472 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
473 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700474 */
475static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
476{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700477 struct list_head *lsthead = &epi->pwqlist;
478 struct eppoll_entry *pwq;
479
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700480 while (!list_empty(lsthead)) {
481 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700482
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700483 list_del(&pwq->llink);
484 remove_wait_queue(pwq->whead, &pwq->wait);
485 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700486 }
487}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Davide Libenzi5071f972009-03-31 15:24:10 -0700489/**
490 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
491 * the scan code, to call f_op->poll(). Also allows for
492 * O(NumReady) performance.
493 *
494 * @ep: Pointer to the epoll private data structure.
495 * @sproc: Pointer to the scan callback.
496 * @priv: Private opaque data passed to the @sproc callback.
Nelson Elhaged8805e62011-10-31 17:13:14 -0700497 * @depth: The current depth of recursive f_op->poll calls.
Davide Libenzi5071f972009-03-31 15:24:10 -0700498 *
499 * Returns: The same integer error code returned by the @sproc callback.
500 */
501static int ep_scan_ready_list(struct eventpoll *ep,
502 int (*sproc)(struct eventpoll *,
503 struct list_head *, void *),
Nelson Elhaged8805e62011-10-31 17:13:14 -0700504 void *priv,
505 int depth)
Davide Libenzi5071f972009-03-31 15:24:10 -0700506{
507 int error, pwake = 0;
508 unsigned long flags;
509 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700510 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700511
512 /*
513 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700514 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700515 */
Nelson Elhaged8805e62011-10-31 17:13:14 -0700516 mutex_lock_nested(&ep->mtx, depth);
Davide Libenzi5071f972009-03-31 15:24:10 -0700517
518 /*
519 * Steal the ready list, and re-init the original one to the
520 * empty list. Also, set ep->ovflist to NULL so that events
521 * happening while looping w/out locks, are not lost. We cannot
522 * have the poll callback to queue directly on ep->rdllist,
523 * because we want the "sproc" callback to be able to do it
524 * in a lockless way.
525 */
526 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700527 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700528 ep->ovflist = NULL;
529 spin_unlock_irqrestore(&ep->lock, flags);
530
531 /*
532 * Now call the callback function.
533 */
534 error = (*sproc)(ep, &txlist, priv);
535
536 spin_lock_irqsave(&ep->lock, flags);
537 /*
538 * During the time we spent inside the "sproc" callback, some
539 * other events might have been queued by the poll callback.
540 * We re-insert them inside the main ready-list here.
541 */
542 for (nepi = ep->ovflist; (epi = nepi) != NULL;
543 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
544 /*
545 * We need to check if the item is already in the list.
546 * During the "sproc" callback execution time, items are
547 * queued into ->ovflist but the "txlist" might already
548 * contain them, and the list_splice() below takes care of them.
549 */
550 if (!ep_is_linked(&epi->rdllink))
551 list_add_tail(&epi->rdllink, &ep->rdllist);
552 }
553 /*
554 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
555 * releasing the lock, events will be queued in the normal way inside
556 * ep->rdllist.
557 */
558 ep->ovflist = EP_UNACTIVE_PTR;
559
560 /*
561 * Quickly re-inject items left on "txlist".
562 */
563 list_splice(&txlist, &ep->rdllist);
564
565 if (!list_empty(&ep->rdllist)) {
566 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700567 * Wake up (if active) both the eventpoll wait list and
568 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700569 */
570 if (waitqueue_active(&ep->wq))
571 wake_up_locked(&ep->wq);
572 if (waitqueue_active(&ep->poll_wait))
573 pwake++;
574 }
575 spin_unlock_irqrestore(&ep->lock, flags);
576
577 mutex_unlock(&ep->mtx);
578
579 /* We have to call this outside the lock */
580 if (pwake)
581 ep_poll_safewake(&ep->poll_wait);
582
583 return error;
584}
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700587 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700588 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700589 */
590static int ep_remove(struct eventpoll *ep, struct epitem *epi)
591{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700592 unsigned long flags;
593 struct file *file = epi->ffd.file;
594
595 /*
596 * Removes poll wait queue hooks. We _have_ to do this without holding
597 * the "ep->lock" otherwise a deadlock might occur. This because of the
598 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
599 * queue head lock when unregistering the wait queue. The wakeup callback
600 * will run by holding the wait queue head lock and will call our callback
601 * that will try to get "ep->lock".
602 */
603 ep_unregister_pollwait(ep, epi);
604
605 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700606 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700607 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700608 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700609 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
Davide Libenzicdac75e2008-04-29 00:58:34 -0700611 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Davide Libenzic7ea7632007-05-15 01:40:47 -0700613 spin_lock_irqsave(&ep->lock, flags);
614 if (ep_is_linked(&epi->rdllink))
615 list_del_init(&epi->rdllink);
616 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Davide Libenzi7699acd2007-05-10 22:23:23 -0700618 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700619 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Robin Holt52bd19f72011-01-12 17:00:01 -0800621 atomic_long_dec(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800622
Davide Libenzic7ea7632007-05-15 01:40:47 -0700623 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624}
625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626static void ep_free(struct eventpoll *ep)
627{
628 struct rb_node *rbp;
629 struct epitem *epi;
630
631 /* We need to release all tasks waiting for these file */
632 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700633 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635 /*
636 * We need to lock this because we could be hit by
637 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700638 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 * is on the way to be removed and no one has references to it
640 * anymore. The only hit might come from eventpoll_release_file() but
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300641 * holding "epmutex" is sufficient here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800643 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
645 /*
646 * Walks through the whole tree by unregistering poll callbacks.
647 */
648 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
649 epi = rb_entry(rbp, struct epitem, rbn);
650
651 ep_unregister_pollwait(ep, epi);
652 }
653
654 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700655 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700657 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 * us during this operation. So we can avoid the lock on "ep->lock".
659 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700660 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 epi = rb_entry(rbp, struct epitem, rbn);
662 ep_remove(ep, epi);
663 }
664
Arjan van de Ven144efe32006-03-23 03:00:32 -0800665 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700666 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800667 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700668 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669}
670
Davide Libenzi7699acd2007-05-10 22:23:23 -0700671static int ep_eventpoll_release(struct inode *inode, struct file *file)
672{
673 struct eventpoll *ep = file->private_data;
674
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700675 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700676 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700677
Davide Libenzi7699acd2007-05-10 22:23:23 -0700678 return 0;
679}
680
Davide Libenzi296e2362009-03-31 15:24:11 -0700681static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
682 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700683{
684 struct epitem *epi, *tmp;
685
686 list_for_each_entry_safe(epi, tmp, head, rdllink) {
687 if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
688 epi->event.events)
689 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700690 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700691 /*
692 * Item has been dropped into the ready list by the poll
693 * callback, but it's not actually ready, as far as
694 * caller requested events goes. We can remove it here.
695 */
696 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700697 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700698 }
699
700 return 0;
701}
702
703static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
704{
Nelson Elhaged8805e62011-10-31 17:13:14 -0700705 return ep_scan_ready_list(priv, ep_read_events_proc, NULL, call_nests + 1);
Davide Libenzi5071f972009-03-31 15:24:10 -0700706}
707
Davide Libenzi7699acd2007-05-10 22:23:23 -0700708static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
709{
Davide Libenzi5071f972009-03-31 15:24:10 -0700710 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700711 struct eventpoll *ep = file->private_data;
712
713 /* Insert inside our poll wait queue */
714 poll_wait(file, &ep->poll_wait, wait);
715
Davide Libenzi5071f972009-03-31 15:24:10 -0700716 /*
717 * Proceed to find out if wanted events are really available inside
718 * the ready list. This need to be done under ep_call_nested()
719 * supervision, since the call to f_op->poll() done on listed files
720 * could re-enter here.
721 */
722 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700723 ep_poll_readyevents_proc, ep, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700724
Davide Libenzi296e2362009-03-31 15:24:11 -0700725 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700726}
727
728/* File callbacks that implement the eventpoll file behaviour */
729static const struct file_operations eventpoll_fops = {
730 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200731 .poll = ep_eventpoll_poll,
732 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700733};
734
Davide Libenzi7699acd2007-05-10 22:23:23 -0700735/*
736 * This is called from eventpoll_release() to unlink files from the eventpoll
737 * interface. We need to have this facility to cleanup correctly files that are
738 * closed without being removed from the eventpoll interface.
739 */
740void eventpoll_release_file(struct file *file)
741{
742 struct list_head *lsthead = &file->f_ep_links;
743 struct eventpoll *ep;
744 struct epitem *epi;
745
746 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700747 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700748 * necessary. It is not necessary because we're in the "struct file"
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300749 * cleanup path, and this means that no one is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700750 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700751 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700752 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700753 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700754 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700755 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700756 *
757 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700758 */
759 mutex_lock(&epmutex);
760
761 while (!list_empty(lsthead)) {
762 epi = list_first_entry(lsthead, struct epitem, fllink);
763
764 ep = epi->ep;
765 list_del_init(&epi->fllink);
Nelson Elhaged8805e62011-10-31 17:13:14 -0700766 mutex_lock_nested(&ep->mtx, 0);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700767 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700768 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700769 }
770
771 mutex_unlock(&epmutex);
772}
773
774static int ep_alloc(struct eventpoll **pep)
775{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800776 int error;
777 struct user_struct *user;
778 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700779
Davide Libenzi7ef99642008-12-01 13:13:55 -0800780 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800781 error = -ENOMEM;
782 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
783 if (unlikely(!ep))
784 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700785
Davide Libenzic7ea7632007-05-15 01:40:47 -0700786 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700787 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700788 init_waitqueue_head(&ep->wq);
789 init_waitqueue_head(&ep->poll_wait);
790 INIT_LIST_HEAD(&ep->rdllist);
791 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700792 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800793 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700794
795 *pep = ep;
796
Davide Libenzi7699acd2007-05-10 22:23:23 -0700797 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800798
799free_uid:
800 free_uid(user);
801 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700802}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700805 * Search the file inside the eventpoll tree. The RB tree operations
806 * are protected by the "mtx" mutex, and ep_find() must be called with
807 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 */
809static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
810{
811 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 struct rb_node *rbp;
813 struct epitem *epi, *epir = NULL;
814 struct epoll_filefd ffd;
815
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700816 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 for (rbp = ep->rbr.rb_node; rbp; ) {
818 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700819 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 if (kcmp > 0)
821 rbp = rbp->rb_right;
822 else if (kcmp < 0)
823 rbp = rbp->rb_left;
824 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 epir = epi;
826 break;
827 }
828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 return epir;
831}
832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700834 * This is the callback that is passed to the wait queue wakeup
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200835 * mechanism. It is called by the stored file descriptors when they
Davide Libenzi7699acd2007-05-10 22:23:23 -0700836 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700838static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700840 int pwake = 0;
841 unsigned long flags;
842 struct epitem *epi = ep_item_from_wait(wait);
843 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Davide Libenzic7ea7632007-05-15 01:40:47 -0700845 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700846
847 /*
848 * If the event mask does not contain any poll(2) event, we consider the
849 * descriptor to be disabled. This condition is likely the effect of the
850 * EPOLLONESHOT bit that disables the descriptor when an event is received,
851 * until the next EPOLL_CTL_MOD will be issued.
852 */
853 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700854 goto out_unlock;
855
856 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700857 * Check the events coming with the callback. At this stage, not
858 * every device reports the events in the "key" parameter of the
859 * callback. We need to be able to handle both cases here, hence the
860 * test for "key" != NULL before the event match test.
861 */
862 if (key && !((unsigned long) key & epi->event.events))
863 goto out_unlock;
864
865 /*
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200866 * If we are transferring events to userspace, we can hold no locks
Davide Libenzid47de162007-05-15 01:40:41 -0700867 * (because we're accessing user memory, and because of linux f_op->poll()
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200868 * semantics). All the events that happen during that period of time are
Davide Libenzid47de162007-05-15 01:40:41 -0700869 * chained in ep->ovflist and requeued later on.
870 */
871 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
872 if (epi->next == EP_UNACTIVE_PTR) {
873 epi->next = ep->ovflist;
874 ep->ovflist = epi;
875 }
876 goto out_unlock;
877 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700878
879 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700880 if (!ep_is_linked(&epi->rdllink))
881 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700882
Davide Libenzi7699acd2007-05-10 22:23:23 -0700883 /*
884 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
885 * wait list.
886 */
887 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400888 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700889 if (waitqueue_active(&ep->poll_wait))
890 pwake++;
891
Davide Libenzid47de162007-05-15 01:40:41 -0700892out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700893 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700894
895 /* We have to call this outside the lock */
896 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700897 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700898
899 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900}
901
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902/*
903 * This is the callback that is used to add our wait queue to the
904 * target file wakeup lists.
905 */
906static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
907 poll_table *pt)
908{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700909 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct eppoll_entry *pwq;
911
Christoph Lametere94b1762006-12-06 20:33:17 -0800912 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
914 pwq->whead = whead;
915 pwq->base = epi;
916 add_wait_queue(whead, &pwq->wait);
917 list_add_tail(&pwq->llink, &epi->pwqlist);
918 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700919 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 /* We have to signal that an error occurred */
921 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
926{
927 int kcmp;
928 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
929 struct epitem *epic;
930
931 while (*p) {
932 parent = *p;
933 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700934 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 if (kcmp > 0)
936 p = &parent->rb_right;
937 else
938 p = &parent->rb_left;
939 }
940 rb_link_node(&epi->rbn, parent, p);
941 rb_insert_color(&epi->rbn, &ep->rbr);
942}
943
Jason Baron28d82dc2012-01-12 17:17:43 -0800944
945
946#define PATH_ARR_SIZE 5
947/*
948 * These are the number paths of length 1 to 5, that we are allowing to emanate
949 * from a single file of interest. For example, we allow 1000 paths of length
950 * 1, to emanate from each file of interest. This essentially represents the
951 * potential wakeup paths, which need to be limited in order to avoid massive
952 * uncontrolled wakeup storms. The common use case should be a single ep which
953 * is connected to n file sources. In this case each file source has 1 path
954 * of length 1. Thus, the numbers below should be more than sufficient. These
955 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
956 * and delete can't add additional paths. Protected by the epmutex.
957 */
958static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
959static int path_count[PATH_ARR_SIZE];
960
961static int path_count_inc(int nests)
962{
963 if (++path_count[nests] > path_limits[nests])
964 return -1;
965 return 0;
966}
967
968static void path_count_init(void)
969{
970 int i;
971
972 for (i = 0; i < PATH_ARR_SIZE; i++)
973 path_count[i] = 0;
974}
975
976static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
977{
978 int error = 0;
979 struct file *file = priv;
980 struct file *child_file;
981 struct epitem *epi;
982
983 list_for_each_entry(epi, &file->f_ep_links, fllink) {
984 child_file = epi->ep->file;
985 if (is_file_epoll(child_file)) {
986 if (list_empty(&child_file->f_ep_links)) {
987 if (path_count_inc(call_nests)) {
988 error = -1;
989 break;
990 }
991 } else {
992 error = ep_call_nested(&poll_loop_ncalls,
993 EP_MAX_NESTS,
994 reverse_path_check_proc,
995 child_file, child_file,
996 current);
997 }
998 if (error != 0)
999 break;
1000 } else {
1001 printk(KERN_ERR "reverse_path_check_proc: "
1002 "file is not an ep!\n");
1003 }
1004 }
1005 return error;
1006}
1007
1008/**
1009 * reverse_path_check - The tfile_check_list is list of file *, which have
1010 * links that are proposed to be newly added. We need to
1011 * make sure that those added links don't add too many
1012 * paths such that we will spend all our time waking up
1013 * eventpoll objects.
1014 *
1015 * Returns: Returns zero if the proposed links don't create too many paths,
1016 * -1 otherwise.
1017 */
1018static int reverse_path_check(void)
1019{
1020 int length = 0;
1021 int error = 0;
1022 struct file *current_file;
1023
1024 /* let's call this for all tfiles */
1025 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1026 length++;
1027 path_count_init();
1028 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1029 reverse_path_check_proc, current_file,
1030 current_file, current);
1031 if (error)
1032 break;
1033 }
1034 return error;
1035}
1036
Davide Libenzic7ea7632007-05-15 01:40:47 -07001037/*
1038 * Must be called with "mtx" held.
1039 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1041 struct file *tfile, int fd)
1042{
1043 int error, revents, pwake = 0;
1044 unsigned long flags;
Robin Holt52bd19f72011-01-12 17:00:01 -08001045 long user_watches;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 struct epitem *epi;
1047 struct ep_pqueue epq;
1048
Robin Holt52bd19f72011-01-12 17:00:01 -08001049 user_watches = atomic_long_read(&ep->user->epoll_watches);
1050 if (unlikely(user_watches >= max_user_watches))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001051 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -08001052 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001053 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 INIT_LIST_HEAD(&epi->rdllink);
1057 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 INIT_LIST_HEAD(&epi->pwqlist);
1059 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001060 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -07001063 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065 /* Initialize the poll table using the queue callback */
1066 epq.epi = epi;
1067 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1068
1069 /*
1070 * Attach the item to the poll hooks and get current event bits.
1071 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -07001072 * been increased by the caller of this function. Note that after
1073 * this operation completes, the poll callback can start hitting
1074 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 */
1076 revents = tfile->f_op->poll(tfile, &epq.pt);
1077
1078 /*
1079 * We have to check if something went wrong during the poll wait queue
1080 * install process. Namely an allocation for a wait queue failed due
1081 * high memory pressure.
1082 */
Davide Libenzi7ef99642008-12-01 13:13:55 -08001083 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001085 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -07001088 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -07001090 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
Davide Libenzic7ea7632007-05-15 01:40:47 -07001092 /*
1093 * Add the current item to the RB tree. All RB tree operations are
1094 * protected by "mtx", and ep_insert() is called with "mtx" held.
1095 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 ep_rbtree_insert(ep, epi);
1097
Jason Baron28d82dc2012-01-12 17:17:43 -08001098 /* now check if we've created too many backpaths */
1099 error = -EINVAL;
1100 if (reverse_path_check())
1101 goto error_remove_epi;
1102
Davide Libenzic7ea7632007-05-15 01:40:47 -07001103 /* We have to drop the new item inside our item list to keep track of it */
1104 spin_lock_irqsave(&ep->lock, flags);
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001107 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 list_add_tail(&epi->rdllink, &ep->rdllist);
1109
1110 /* Notify waiting tasks that events are available */
1111 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001112 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 if (waitqueue_active(&ep->poll_wait))
1114 pwake++;
1115 }
1116
Davide Libenzic7ea7632007-05-15 01:40:47 -07001117 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Robin Holt52bd19f72011-01-12 17:00:01 -08001119 atomic_long_inc(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001120
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 /* We have to call this outside the lock */
1122 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001123 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return 0;
1126
Jason Baron28d82dc2012-01-12 17:17:43 -08001127error_remove_epi:
1128 spin_lock(&tfile->f_lock);
1129 if (ep_is_linked(&epi->fllink))
1130 list_del_init(&epi->fllink);
1131 spin_unlock(&tfile->f_lock);
1132
1133 rb_erase(&epi->rbn, &ep->rbr);
1134
Davide Libenzi7699acd2007-05-10 22:23:23 -07001135error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 ep_unregister_pollwait(ep, epi);
1137
1138 /*
1139 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -07001140 * allocated wait queue. Note that we don't care about the ep->ovflist
1141 * list, since that is used/cleaned only inside a section bound by "mtx".
1142 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001144 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001145 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -07001146 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001147 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001149 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001150
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 return error;
1152}
1153
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154/*
1155 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001156 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 */
1158static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1159{
1160 int pwake = 0;
1161 unsigned int revents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
1163 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001164 * Set the new event interest mask before calling f_op->poll();
1165 * otherwise we might miss an event that happens between the
1166 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 */
1168 epi->event.events = event->events;
Tony Battersbye057e152009-03-31 15:24:15 -07001169 epi->event.data = event->data; /* protected by mtx */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
1171 /*
1172 * Get current event bits. We can safely use the file* here because
1173 * its usage count has been increased by the caller of this function.
1174 */
1175 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL);
1176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001178 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001179 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001181 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001182 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001183 if (!ep_is_linked(&epi->rdllink)) {
1184 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Davide Libenzic7ea7632007-05-15 01:40:47 -07001186 /* Notify waiting tasks that events are available */
1187 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001188 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001189 if (waitqueue_active(&ep->poll_wait))
1190 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 }
Tony Battersbye057e152009-03-31 15:24:15 -07001192 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 /* We have to call this outside the lock */
1196 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001197 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 return 0;
1200}
1201
Davide Libenzi296e2362009-03-31 15:24:11 -07001202static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1203 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001204{
1205 struct ep_send_events_data *esed = priv;
1206 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001207 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001208 struct epitem *epi;
1209 struct epoll_event __user *uevent;
1210
Davide Libenzi296e2362009-03-31 15:24:11 -07001211 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001212 * We can loop without lock because we are passed a task private list.
1213 * Items cannot vanish during the loop because ep_scan_ready_list() is
1214 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001215 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001216 for (eventcnt = 0, uevent = esed->events;
1217 !list_empty(head) && eventcnt < esed->maxevents;) {
1218 epi = list_first_entry(head, struct epitem, rdllink);
1219
1220 list_del_init(&epi->rdllink);
1221
Davide Libenzi296e2362009-03-31 15:24:11 -07001222 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
1223 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001224
Davide Libenzi296e2362009-03-31 15:24:11 -07001225 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001226 * If the event mask intersect the caller-requested one,
1227 * deliver the event to userspace. Again, ep_scan_ready_list()
1228 * is holding "mtx", so no operations coming from userspace
1229 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001230 */
1231 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001232 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001233 __put_user(epi->event.data, &uevent->data)) {
1234 list_add(&epi->rdllink, head);
Davide Libenzi296e2362009-03-31 15:24:11 -07001235 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001236 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001237 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001238 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001239 if (epi->event.events & EPOLLONESHOT)
1240 epi->event.events &= EP_PRIVATE_BITS;
1241 else if (!(epi->event.events & EPOLLET)) {
1242 /*
1243 * If this file has been added with Level
1244 * Trigger mode, we need to insert back inside
1245 * the ready list, so that the next call to
1246 * epoll_wait() will check again the events
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001247 * availability. At this point, no one can insert
Davide Libenzi296e2362009-03-31 15:24:11 -07001248 * into ep->rdllist besides us. The epoll_ctl()
1249 * callers are locked out by
1250 * ep_scan_ready_list() holding "mtx" and the
1251 * poll callback will queue them in ep->ovflist.
1252 */
1253 list_add_tail(&epi->rdllink, &ep->rdllist);
1254 }
1255 }
1256 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001257
1258 return eventcnt;
1259}
1260
Davide Libenzi296e2362009-03-31 15:24:11 -07001261static int ep_send_events(struct eventpoll *ep,
1262 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
Davide Libenzi5071f972009-03-31 15:24:10 -07001264 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001265
Davide Libenzi5071f972009-03-31 15:24:10 -07001266 esed.maxevents = maxevents;
1267 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001268
Nelson Elhaged8805e62011-10-31 17:13:14 -07001269 return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
1271
Eric Dumazet0781b902011-02-01 15:52:35 -08001272static inline struct timespec ep_set_mstimeout(long ms)
1273{
1274 struct timespec now, ts = {
1275 .tv_sec = ms / MSEC_PER_SEC,
1276 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1277 };
1278
1279 ktime_get_ts(&now);
1280 return timespec_add_safe(now, ts);
1281}
1282
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001283/**
1284 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1285 * event buffer.
1286 *
1287 * @ep: Pointer to the eventpoll context.
1288 * @events: Pointer to the userspace buffer where the ready events should be
1289 * stored.
1290 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1291 * @timeout: Maximum timeout for the ready events fetch operation, in
1292 * milliseconds. If the @timeout is zero, the function will not block,
1293 * while if the @timeout is less than zero, the function will block
1294 * until at least one event has been retrieved (or an error
1295 * occurred).
1296 *
1297 * Returns: Returns the number of ready events which have been fetched, or an
1298 * error code, in case of error.
1299 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1301 int maxevents, long timeout)
1302{
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001303 int res = 0, eavail, timed_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 unsigned long flags;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001305 long slack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 wait_queue_t wait;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001307 ktime_t expires, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001309 if (timeout > 0) {
Eric Dumazet0781b902011-02-01 15:52:35 -08001310 struct timespec end_time = ep_set_mstimeout(timeout);
1311
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001312 slack = select_estimate_accuracy(&end_time);
1313 to = &expires;
1314 *to = timespec_to_ktime(end_time);
1315 } else if (timeout == 0) {
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001316 /*
1317 * Avoid the unnecessary trip to the wait queue loop, if the
1318 * caller specified a non blocking operation.
1319 */
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001320 timed_out = 1;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001321 spin_lock_irqsave(&ep->lock, flags);
1322 goto check_events;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001325fetch_events:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001326 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001328 if (!ep_events_available(ep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 /*
1330 * We don't have any available event to return to the caller.
1331 * We need to sleep here, and we will be wake up by
1332 * ep_poll_callback() when events will become available.
1333 */
1334 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001335 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 for (;;) {
1338 /*
1339 * We don't want to sleep if the ep_poll_callback() sends us
1340 * a wakeup in between. That's why we set the task state
1341 * to TASK_INTERRUPTIBLE before doing the checks.
1342 */
1343 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001344 if (ep_events_available(ep) || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 break;
1346 if (signal_pending(current)) {
1347 res = -EINTR;
1348 break;
1349 }
1350
Davide Libenzic7ea7632007-05-15 01:40:47 -07001351 spin_unlock_irqrestore(&ep->lock, flags);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001352 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1353 timed_out = 1;
1354
Davide Libenzic7ea7632007-05-15 01:40:47 -07001355 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001357 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 set_current_state(TASK_RUNNING);
1360 }
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001361check_events:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 /* Is it worth to try to dig for events ? */
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001363 eavail = ep_events_available(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Davide Libenzic7ea7632007-05-15 01:40:47 -07001365 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 /*
1368 * Try to transfer events to user space. In case we get 0 events and
1369 * there's still timeout left over, we go trying again in search of
1370 * more luck.
1371 */
1372 if (!res && eavail &&
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001373 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001374 goto fetch_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
1376 return res;
1377}
1378
Davide Libenzi22bacca2011-02-25 14:44:12 -08001379/**
1380 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1381 * API, to verify that adding an epoll file inside another
1382 * epoll structure, does not violate the constraints, in
1383 * terms of closed loops, or too deep chains (which can
1384 * result in excessive stack usage).
1385 *
1386 * @priv: Pointer to the epoll file to be currently checked.
1387 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1388 * data structure pointer.
1389 * @call_nests: Current dept of the @ep_call_nested() call stack.
1390 *
1391 * Returns: Returns zero if adding the epoll @file inside current epoll
1392 * structure @ep does not violate the constraints, or -1 otherwise.
1393 */
1394static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1395{
1396 int error = 0;
1397 struct file *file = priv;
1398 struct eventpoll *ep = file->private_data;
Jason Baron28d82dc2012-01-12 17:17:43 -08001399 struct eventpoll *ep_tovisit;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001400 struct rb_node *rbp;
1401 struct epitem *epi;
1402
Nelson Elhaged8805e62011-10-31 17:13:14 -07001403 mutex_lock_nested(&ep->mtx, call_nests + 1);
Jason Baron28d82dc2012-01-12 17:17:43 -08001404 ep->visited = 1;
1405 list_add(&ep->visited_list_link, &visited_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001406 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1407 epi = rb_entry(rbp, struct epitem, rbn);
1408 if (unlikely(is_file_epoll(epi->ffd.file))) {
Jason Baron28d82dc2012-01-12 17:17:43 -08001409 ep_tovisit = epi->ffd.file->private_data;
1410 if (ep_tovisit->visited)
1411 continue;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001412 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Jason Baron28d82dc2012-01-12 17:17:43 -08001413 ep_loop_check_proc, epi->ffd.file,
1414 ep_tovisit, current);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001415 if (error != 0)
1416 break;
Jason Baron28d82dc2012-01-12 17:17:43 -08001417 } else {
1418 /*
1419 * If we've reached a file that is not associated with
1420 * an ep, then we need to check if the newly added
1421 * links are going to add too many wakeup paths. We do
1422 * this by adding it to the tfile_check_list, if it's
1423 * not already there, and calling reverse_path_check()
1424 * during ep_insert().
1425 */
1426 if (list_empty(&epi->ffd.file->f_tfile_llink))
1427 list_add(&epi->ffd.file->f_tfile_llink,
1428 &tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001429 }
1430 }
1431 mutex_unlock(&ep->mtx);
1432
1433 return error;
1434}
1435
1436/**
1437 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1438 * another epoll file (represented by @ep) does not create
1439 * closed loops or too deep chains.
1440 *
1441 * @ep: Pointer to the epoll private data structure.
1442 * @file: Pointer to the epoll file to be checked.
1443 *
1444 * Returns: Returns zero if adding the epoll @file inside current epoll
1445 * structure @ep does not violate the constraints, or -1 otherwise.
1446 */
1447static int ep_loop_check(struct eventpoll *ep, struct file *file)
1448{
Jason Baron28d82dc2012-01-12 17:17:43 -08001449 int ret;
1450 struct eventpoll *ep_cur, *ep_next;
1451
1452 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Davide Libenzi22bacca2011-02-25 14:44:12 -08001453 ep_loop_check_proc, file, ep, current);
Jason Baron28d82dc2012-01-12 17:17:43 -08001454 /* clear visited list */
1455 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1456 visited_list_link) {
1457 ep_cur->visited = 0;
1458 list_del(&ep_cur->visited_list_link);
1459 }
1460 return ret;
1461}
1462
1463static void clear_tfile_check_list(void)
1464{
1465 struct file *file;
1466
1467 /* first clear the tfile_check_list */
1468 while (!list_empty(&tfile_check_list)) {
1469 file = list_first_entry(&tfile_check_list, struct file,
1470 f_tfile_llink);
1471 list_del_init(&file->f_tfile_llink);
1472 }
1473 INIT_LIST_HEAD(&tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001474}
1475
Davide Libenzi7699acd2007-05-10 22:23:23 -07001476/*
Andrew Morton523723b2008-08-12 15:09:01 -07001477 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001478 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001479SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001480{
Jason Baron28d82dc2012-01-12 17:17:43 -08001481 int error, fd;
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001482 struct eventpoll *ep = NULL;
Jason Baron28d82dc2012-01-12 17:17:43 -08001483 struct file *file;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001484
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001485 /* Check the EPOLL_* constant for consistency. */
1486 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1487
Davide Libenzi296e2362009-03-31 15:24:11 -07001488 if (flags & ~EPOLL_CLOEXEC)
1489 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001490 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001491 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001492 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001493 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001494 if (error < 0)
1495 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001496 /*
1497 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001498 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001499 */
Jason Baron28d82dc2012-01-12 17:17:43 -08001500 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1501 if (fd < 0) {
1502 error = fd;
1503 goto out_free_ep;
1504 }
1505 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001506 O_RDWR | (flags & O_CLOEXEC));
Jason Baron28d82dc2012-01-12 17:17:43 -08001507 if (IS_ERR(file)) {
1508 error = PTR_ERR(file);
1509 goto out_free_fd;
1510 }
1511 fd_install(fd, file);
1512 ep->file = file;
1513 return fd;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001514
Jason Baron28d82dc2012-01-12 17:17:43 -08001515out_free_fd:
1516 put_unused_fd(fd);
1517out_free_ep:
1518 ep_free(ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001519 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001520}
1521
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001522SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001523{
Davide Libenzibfe38912009-05-12 13:19:44 -07001524 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001525 return -EINVAL;
1526
1527 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001528}
1529
Davide Libenzi7699acd2007-05-10 22:23:23 -07001530/*
1531 * The following function implements the controller interface for
1532 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001533 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001534 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001535SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1536 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001537{
1538 int error;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001539 int did_lock_epmutex = 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001540 struct file *file, *tfile;
1541 struct eventpoll *ep;
1542 struct epitem *epi;
1543 struct epoll_event epds;
1544
Davide Libenzi7699acd2007-05-10 22:23:23 -07001545 error = -EFAULT;
1546 if (ep_op_has_event(op) &&
1547 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1548 goto error_return;
1549
1550 /* Get the "struct file *" for the eventpoll file */
1551 error = -EBADF;
1552 file = fget(epfd);
1553 if (!file)
1554 goto error_return;
1555
1556 /* Get the "struct file *" for the target file */
1557 tfile = fget(fd);
1558 if (!tfile)
1559 goto error_fput;
1560
1561 /* The target file descriptor must support poll */
1562 error = -EPERM;
1563 if (!tfile->f_op || !tfile->f_op->poll)
1564 goto error_tgt_fput;
1565
1566 /*
1567 * We have to check that the file structure underneath the file descriptor
1568 * the user passed to us _is_ an eventpoll file. And also we do not permit
1569 * adding an epoll file descriptor inside itself.
1570 */
1571 error = -EINVAL;
1572 if (file == tfile || !is_file_epoll(file))
1573 goto error_tgt_fput;
1574
1575 /*
1576 * At this point it is safe to assume that the "private_data" contains
1577 * our own data structure.
1578 */
1579 ep = file->private_data;
1580
Davide Libenzi22bacca2011-02-25 14:44:12 -08001581 /*
1582 * When we insert an epoll file descriptor, inside another epoll file
1583 * descriptor, there is the change of creating closed loops, which are
Jason Baron28d82dc2012-01-12 17:17:43 -08001584 * better be handled here, than in more critical paths. While we are
1585 * checking for loops we also determine the list of files reachable
1586 * and hang them on the tfile_check_list, so we can check that we
1587 * haven't created too many possible wakeup paths.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001588 *
Jason Baron28d82dc2012-01-12 17:17:43 -08001589 * We need to hold the epmutex across both ep_insert and ep_remove
1590 * b/c we want to make sure we are looking at a coherent view of
1591 * epoll network.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001592 */
Jason Baron28d82dc2012-01-12 17:17:43 -08001593 if (op == EPOLL_CTL_ADD || op == EPOLL_CTL_DEL) {
Davide Libenzi22bacca2011-02-25 14:44:12 -08001594 mutex_lock(&epmutex);
1595 did_lock_epmutex = 1;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001596 }
Jason Baron28d82dc2012-01-12 17:17:43 -08001597 if (op == EPOLL_CTL_ADD) {
1598 if (is_file_epoll(tfile)) {
1599 error = -ELOOP;
1600 if (ep_loop_check(ep, tfile) != 0)
1601 goto error_tgt_fput;
1602 } else
1603 list_add(&tfile->f_tfile_llink, &tfile_check_list);
1604 }
Davide Libenzi22bacca2011-02-25 14:44:12 -08001605
Nelson Elhaged8805e62011-10-31 17:13:14 -07001606 mutex_lock_nested(&ep->mtx, 0);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001607
Davide Libenzi67647d02007-05-15 01:40:52 -07001608 /*
1609 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1610 * above, we can be sure to be able to use the item looked up by
1611 * ep_find() till we release the mutex.
1612 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001613 epi = ep_find(ep, tfile, fd);
1614
1615 error = -EINVAL;
1616 switch (op) {
1617 case EPOLL_CTL_ADD:
1618 if (!epi) {
1619 epds.events |= POLLERR | POLLHUP;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001620 error = ep_insert(ep, &epds, tfile, fd);
1621 } else
1622 error = -EEXIST;
Jason Baron28d82dc2012-01-12 17:17:43 -08001623 clear_tfile_check_list();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001624 break;
1625 case EPOLL_CTL_DEL:
1626 if (epi)
1627 error = ep_remove(ep, epi);
1628 else
1629 error = -ENOENT;
1630 break;
1631 case EPOLL_CTL_MOD:
1632 if (epi) {
1633 epds.events |= POLLERR | POLLHUP;
1634 error = ep_modify(ep, epi, &epds);
1635 } else
1636 error = -ENOENT;
1637 break;
1638 }
Davide Libenzid47de162007-05-15 01:40:41 -07001639 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001640
1641error_tgt_fput:
Jason Baron28d82dc2012-01-12 17:17:43 -08001642 if (did_lock_epmutex)
Davide Libenzi22bacca2011-02-25 14:44:12 -08001643 mutex_unlock(&epmutex);
1644
Davide Libenzi7699acd2007-05-10 22:23:23 -07001645 fput(tfile);
1646error_fput:
1647 fput(file);
1648error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001649
1650 return error;
1651}
1652
1653/*
1654 * Implement the event wait interface for the eventpoll file. It is the kernel
1655 * part of the user space epoll_wait(2).
1656 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001657SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1658 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001659{
1660 int error;
1661 struct file *file;
1662 struct eventpoll *ep;
1663
Davide Libenzi7699acd2007-05-10 22:23:23 -07001664 /* The maximum number of event must be greater than zero */
1665 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1666 return -EINVAL;
1667
1668 /* Verify that the area passed by the user is writeable */
1669 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1670 error = -EFAULT;
1671 goto error_return;
1672 }
1673
1674 /* Get the "struct file *" for the eventpoll file */
1675 error = -EBADF;
1676 file = fget(epfd);
1677 if (!file)
1678 goto error_return;
1679
1680 /*
1681 * We have to check that the file structure underneath the fd
1682 * the user passed to us _is_ an eventpoll file.
1683 */
1684 error = -EINVAL;
1685 if (!is_file_epoll(file))
1686 goto error_fput;
1687
1688 /*
1689 * At this point it is safe to assume that the "private_data" contains
1690 * our own data structure.
1691 */
1692 ep = file->private_data;
1693
1694 /* Time to fish for events ... */
1695 error = ep_poll(ep, events, maxevents, timeout);
1696
1697error_fput:
1698 fput(file);
1699error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001700
1701 return error;
1702}
1703
Roland McGrathf3de2722008-04-30 00:53:09 -07001704#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001705
1706/*
1707 * Implement the event wait interface for the eventpoll file. It is the kernel
1708 * part of the user space epoll_pwait(2).
1709 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001710SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1711 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1712 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001713{
1714 int error;
1715 sigset_t ksigmask, sigsaved;
1716
1717 /*
1718 * If the caller wants a certain signal mask to be set during the wait,
1719 * we apply it here.
1720 */
1721 if (sigmask) {
1722 if (sigsetsize != sizeof(sigset_t))
1723 return -EINVAL;
1724 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1725 return -EFAULT;
1726 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1727 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1728 }
1729
1730 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1731
1732 /*
1733 * If we changed the signal mask, we need to restore the original one.
1734 * In case we've got a signal while waiting, we do not restore the
1735 * signal mask yet, and we allow do_signal() to deliver the signal on
1736 * the way back to userspace, before the signal mask is restored.
1737 */
1738 if (sigmask) {
1739 if (error == -EINTR) {
1740 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001741 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001742 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001743 } else
1744 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1745 }
1746
1747 return error;
1748}
1749
Roland McGrathf3de2722008-04-30 00:53:09 -07001750#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752static int __init eventpoll_init(void)
1753{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001754 struct sysinfo si;
1755
1756 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001757 /*
1758 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1759 */
1760 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001761 EP_ITEM_COST;
Robin Holt52bd19f72011-01-12 17:00:01 -08001762 BUG_ON(max_user_watches < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
Davide Libenzi22bacca2011-02-25 14:44:12 -08001764 /*
1765 * Initialize the structure used to perform epoll file descriptor
1766 * inclusion loops checks.
1767 */
1768 ep_nested_calls_init(&poll_loop_ncalls);
1769
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001771 ep_nested_calls_init(&poll_safewake_ncalls);
1772
1773 /* Initialize the structure used to perform file's f_op->poll() calls */
1774 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
1776 /* Allocates slab cache used to allocate "struct epitem" items */
1777 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001778 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779
1780 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1781 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001782 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785}
Davide Libenzicea69242007-05-10 22:23:22 -07001786fs_initcall(eventpoll_init);