blob: ea54cdef04ddc94d282c22f8c2b721169ee77293 [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
Oleg Nesterov971316f2012-02-24 20:07:29 +0100323static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
324{
325 return container_of(p, struct eppoll_entry, wait);
326}
327
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700328/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700329static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700330{
331 return container_of(p, struct eppoll_entry, wait)->base;
332}
333
334/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700335static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700336{
337 return container_of(p, struct ep_pqueue, pt)->epi;
338}
339
340/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700341static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700342{
343 return op != EPOLL_CTL_DEL;
344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700347static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Davide Libenzi5071f972009-03-31 15:24:10 -0700349 INIT_LIST_HEAD(&ncalls->tasks_call_list);
350 spin_lock_init(&ncalls->lock);
351}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Davide Libenzi5071f972009-03-31 15:24:10 -0700353/**
Davide Libenzi3fb0e582011-03-22 16:34:46 -0700354 * ep_events_available - Checks if ready events might be available.
355 *
356 * @ep: Pointer to the eventpoll context.
357 *
358 * Returns: Returns a value different than zero if ready events are available,
359 * or zero otherwise.
360 */
361static inline int ep_events_available(struct eventpoll *ep)
362{
363 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
364}
365
366/**
Davide Libenzi5071f972009-03-31 15:24:10 -0700367 * ep_call_nested - Perform a bound (possibly) nested call, by checking
368 * that the recursion limit is not exceeded, and that
369 * the same nested call (by the meaning of same cookie) is
370 * no re-entered.
371 *
372 * @ncalls: Pointer to the nested_calls structure to be used for this call.
373 * @max_nests: Maximum number of allowed nesting calls.
374 * @nproc: Nested call core function pointer.
375 * @priv: Opaque data to be passed to the @nproc callback.
376 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700377 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700378 *
379 * Returns: Returns the code returned by the @nproc callback, or -1 if
380 * the maximum recursion limit has been exceeded.
381 */
382static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
383 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700384 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700385{
386 int error, call_nests = 0;
387 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700388 struct list_head *lsthead = &ncalls->tasks_call_list;
389 struct nested_call_node *tncur;
390 struct nested_call_node tnode;
391
392 spin_lock_irqsave(&ncalls->lock, flags);
393
394 /*
395 * Try to see if the current task is already inside this wakeup call.
396 * We use a list here, since the population inside this set is always
397 * very much limited.
398 */
399 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700400 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700401 (tncur->cookie == cookie || ++call_nests > max_nests)) {
402 /*
403 * Ops ... loop detected or maximum nest level reached.
404 * We abort this wake by breaking the cycle itself.
405 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700406 error = -1;
407 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700408 }
409 }
410
411 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700412 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700413 tnode.cookie = cookie;
414 list_add(&tnode.llink, lsthead);
415
416 spin_unlock_irqrestore(&ncalls->lock, flags);
417
418 /* Call the nested function */
419 error = (*nproc)(priv, cookie, call_nests);
420
421 /* Remove the current task from the list */
422 spin_lock_irqsave(&ncalls->lock, flags);
423 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700424out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700425 spin_unlock_irqrestore(&ncalls->lock, flags);
426
427 return error;
428}
429
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700430#ifdef CONFIG_DEBUG_LOCK_ALLOC
431static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
432 unsigned long events, int subclass)
433{
434 unsigned long flags;
435
436 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
437 wake_up_locked_poll(wqueue, events);
438 spin_unlock_irqrestore(&wqueue->lock, flags);
439}
440#else
441static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
442 unsigned long events, int subclass)
443{
444 wake_up_poll(wqueue, events);
445}
446#endif
447
Davide Libenzi5071f972009-03-31 15:24:10 -0700448static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
449{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700450 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
451 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700452 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455/*
456 * Perform a safe wake up of the poll wait list. The problem is that
457 * with the new callback'd wake up system, it is possible that the
458 * poll callback is reentered from inside the call to wake_up() done
459 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700460 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * and we cannot reenter the same wait queue head at all. This will
462 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700463 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700465static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700467 int this_cpu = get_cpu();
468
Davide Libenzi5071f972009-03-31 15:24:10 -0700469 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700470 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
471
472 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
Oleg Nesterov971316f2012-02-24 20:07:29 +0100475static void ep_remove_wait_queue(struct eppoll_entry *pwq)
476{
477 wait_queue_head_t *whead;
478
479 rcu_read_lock();
480 /* If it is cleared by POLLFREE, it should be rcu-safe */
481 whead = rcu_dereference(pwq->whead);
482 if (whead)
483 remove_wait_queue(whead, &pwq->wait);
484 rcu_read_unlock();
485}
486
Davide Libenzi7699acd2007-05-10 22:23:23 -0700487/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700488 * This function unregisters poll callbacks from the associated file
489 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
490 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700491 */
492static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
493{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700494 struct list_head *lsthead = &epi->pwqlist;
495 struct eppoll_entry *pwq;
496
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700497 while (!list_empty(lsthead)) {
498 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700499
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700500 list_del(&pwq->llink);
Oleg Nesterov971316f2012-02-24 20:07:29 +0100501 ep_remove_wait_queue(pwq);
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700502 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700503 }
504}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Davide Libenzi5071f972009-03-31 15:24:10 -0700506/**
507 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
508 * the scan code, to call f_op->poll(). Also allows for
509 * O(NumReady) performance.
510 *
511 * @ep: Pointer to the epoll private data structure.
512 * @sproc: Pointer to the scan callback.
513 * @priv: Private opaque data passed to the @sproc callback.
Nelson Elhaged8805e62011-10-31 17:13:14 -0700514 * @depth: The current depth of recursive f_op->poll calls.
Davide Libenzi5071f972009-03-31 15:24:10 -0700515 *
516 * Returns: The same integer error code returned by the @sproc callback.
517 */
518static int ep_scan_ready_list(struct eventpoll *ep,
519 int (*sproc)(struct eventpoll *,
520 struct list_head *, void *),
Nelson Elhaged8805e62011-10-31 17:13:14 -0700521 void *priv,
522 int depth)
Davide Libenzi5071f972009-03-31 15:24:10 -0700523{
524 int error, pwake = 0;
525 unsigned long flags;
526 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700527 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700528
529 /*
530 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700531 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700532 */
Nelson Elhaged8805e62011-10-31 17:13:14 -0700533 mutex_lock_nested(&ep->mtx, depth);
Davide Libenzi5071f972009-03-31 15:24:10 -0700534
535 /*
536 * Steal the ready list, and re-init the original one to the
537 * empty list. Also, set ep->ovflist to NULL so that events
538 * happening while looping w/out locks, are not lost. We cannot
539 * have the poll callback to queue directly on ep->rdllist,
540 * because we want the "sproc" callback to be able to do it
541 * in a lockless way.
542 */
543 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700544 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700545 ep->ovflist = NULL;
546 spin_unlock_irqrestore(&ep->lock, flags);
547
548 /*
549 * Now call the callback function.
550 */
551 error = (*sproc)(ep, &txlist, priv);
552
553 spin_lock_irqsave(&ep->lock, flags);
554 /*
555 * During the time we spent inside the "sproc" callback, some
556 * other events might have been queued by the poll callback.
557 * We re-insert them inside the main ready-list here.
558 */
559 for (nepi = ep->ovflist; (epi = nepi) != NULL;
560 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
561 /*
562 * We need to check if the item is already in the list.
563 * During the "sproc" callback execution time, items are
564 * queued into ->ovflist but the "txlist" might already
565 * contain them, and the list_splice() below takes care of them.
566 */
567 if (!ep_is_linked(&epi->rdllink))
568 list_add_tail(&epi->rdllink, &ep->rdllist);
569 }
570 /*
571 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
572 * releasing the lock, events will be queued in the normal way inside
573 * ep->rdllist.
574 */
575 ep->ovflist = EP_UNACTIVE_PTR;
576
577 /*
578 * Quickly re-inject items left on "txlist".
579 */
580 list_splice(&txlist, &ep->rdllist);
581
582 if (!list_empty(&ep->rdllist)) {
583 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700584 * Wake up (if active) both the eventpoll wait list and
585 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700586 */
587 if (waitqueue_active(&ep->wq))
588 wake_up_locked(&ep->wq);
589 if (waitqueue_active(&ep->poll_wait))
590 pwake++;
591 }
592 spin_unlock_irqrestore(&ep->lock, flags);
593
594 mutex_unlock(&ep->mtx);
595
596 /* We have to call this outside the lock */
597 if (pwake)
598 ep_poll_safewake(&ep->poll_wait);
599
600 return error;
601}
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700604 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700605 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700606 */
607static int ep_remove(struct eventpoll *ep, struct epitem *epi)
608{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700609 unsigned long flags;
610 struct file *file = epi->ffd.file;
611
612 /*
613 * Removes poll wait queue hooks. We _have_ to do this without holding
614 * the "ep->lock" otherwise a deadlock might occur. This because of the
615 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
616 * queue head lock when unregistering the wait queue. The wakeup callback
617 * will run by holding the wait queue head lock and will call our callback
618 * that will try to get "ep->lock".
619 */
620 ep_unregister_pollwait(ep, epi);
621
622 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700623 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700624 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700625 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700626 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Davide Libenzicdac75e2008-04-29 00:58:34 -0700628 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Davide Libenzic7ea7632007-05-15 01:40:47 -0700630 spin_lock_irqsave(&ep->lock, flags);
631 if (ep_is_linked(&epi->rdllink))
632 list_del_init(&epi->rdllink);
633 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Davide Libenzi7699acd2007-05-10 22:23:23 -0700635 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700636 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Robin Holt52bd19f72011-01-12 17:00:01 -0800638 atomic_long_dec(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800639
Davide Libenzic7ea7632007-05-15 01:40:47 -0700640 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643static void ep_free(struct eventpoll *ep)
644{
645 struct rb_node *rbp;
646 struct epitem *epi;
647
648 /* We need to release all tasks waiting for these file */
649 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700650 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
652 /*
653 * We need to lock this because we could be hit by
654 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700655 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 * is on the way to be removed and no one has references to it
657 * anymore. The only hit might come from eventpoll_release_file() but
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300658 * holding "epmutex" is sufficient here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800660 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
662 /*
663 * Walks through the whole tree by unregistering poll callbacks.
664 */
665 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
666 epi = rb_entry(rbp, struct epitem, rbn);
667
668 ep_unregister_pollwait(ep, epi);
669 }
670
671 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700672 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700674 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 * us during this operation. So we can avoid the lock on "ep->lock".
676 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700677 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 epi = rb_entry(rbp, struct epitem, rbn);
679 ep_remove(ep, epi);
680 }
681
Arjan van de Ven144efe32006-03-23 03:00:32 -0800682 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700683 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800684 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700685 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686}
687
Davide Libenzi7699acd2007-05-10 22:23:23 -0700688static int ep_eventpoll_release(struct inode *inode, struct file *file)
689{
690 struct eventpoll *ep = file->private_data;
691
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700692 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700693 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700694
Davide Libenzi7699acd2007-05-10 22:23:23 -0700695 return 0;
696}
697
Davide Libenzi296e2362009-03-31 15:24:11 -0700698static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
699 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700700{
701 struct epitem *epi, *tmp;
702
703 list_for_each_entry_safe(epi, tmp, head, rdllink) {
704 if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
705 epi->event.events)
706 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700707 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700708 /*
709 * Item has been dropped into the ready list by the poll
710 * callback, but it's not actually ready, as far as
711 * caller requested events goes. We can remove it here.
712 */
713 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700714 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700715 }
716
717 return 0;
718}
719
720static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
721{
Nelson Elhaged8805e62011-10-31 17:13:14 -0700722 return ep_scan_ready_list(priv, ep_read_events_proc, NULL, call_nests + 1);
Davide Libenzi5071f972009-03-31 15:24:10 -0700723}
724
Davide Libenzi7699acd2007-05-10 22:23:23 -0700725static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
726{
Davide Libenzi5071f972009-03-31 15:24:10 -0700727 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700728 struct eventpoll *ep = file->private_data;
729
730 /* Insert inside our poll wait queue */
731 poll_wait(file, &ep->poll_wait, wait);
732
Davide Libenzi5071f972009-03-31 15:24:10 -0700733 /*
734 * Proceed to find out if wanted events are really available inside
735 * the ready list. This need to be done under ep_call_nested()
736 * supervision, since the call to f_op->poll() done on listed files
737 * could re-enter here.
738 */
739 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700740 ep_poll_readyevents_proc, ep, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700741
Davide Libenzi296e2362009-03-31 15:24:11 -0700742 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700743}
744
745/* File callbacks that implement the eventpoll file behaviour */
746static const struct file_operations eventpoll_fops = {
747 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200748 .poll = ep_eventpoll_poll,
749 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700750};
751
Davide Libenzi7699acd2007-05-10 22:23:23 -0700752/*
753 * This is called from eventpoll_release() to unlink files from the eventpoll
754 * interface. We need to have this facility to cleanup correctly files that are
755 * closed without being removed from the eventpoll interface.
756 */
757void eventpoll_release_file(struct file *file)
758{
759 struct list_head *lsthead = &file->f_ep_links;
760 struct eventpoll *ep;
761 struct epitem *epi;
762
763 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700764 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700765 * necessary. It is not necessary because we're in the "struct file"
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300766 * cleanup path, and this means that no one is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700767 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700768 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700769 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700770 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700771 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700772 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700773 *
774 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700775 */
776 mutex_lock(&epmutex);
777
778 while (!list_empty(lsthead)) {
779 epi = list_first_entry(lsthead, struct epitem, fllink);
780
781 ep = epi->ep;
782 list_del_init(&epi->fllink);
Nelson Elhaged8805e62011-10-31 17:13:14 -0700783 mutex_lock_nested(&ep->mtx, 0);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700784 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700785 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700786 }
787
788 mutex_unlock(&epmutex);
789}
790
791static int ep_alloc(struct eventpoll **pep)
792{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800793 int error;
794 struct user_struct *user;
795 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700796
Davide Libenzi7ef99642008-12-01 13:13:55 -0800797 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800798 error = -ENOMEM;
799 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
800 if (unlikely(!ep))
801 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700802
Davide Libenzic7ea7632007-05-15 01:40:47 -0700803 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700804 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700805 init_waitqueue_head(&ep->wq);
806 init_waitqueue_head(&ep->poll_wait);
807 INIT_LIST_HEAD(&ep->rdllist);
808 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700809 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800810 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700811
812 *pep = ep;
813
Davide Libenzi7699acd2007-05-10 22:23:23 -0700814 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800815
816free_uid:
817 free_uid(user);
818 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700819}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700822 * Search the file inside the eventpoll tree. The RB tree operations
823 * are protected by the "mtx" mutex, and ep_find() must be called with
824 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 */
826static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
827{
828 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 struct rb_node *rbp;
830 struct epitem *epi, *epir = NULL;
831 struct epoll_filefd ffd;
832
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700833 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 for (rbp = ep->rbr.rb_node; rbp; ) {
835 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700836 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (kcmp > 0)
838 rbp = rbp->rb_right;
839 else if (kcmp < 0)
840 rbp = rbp->rb_left;
841 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 epir = epi;
843 break;
844 }
845 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return epir;
848}
849
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700851 * This is the callback that is passed to the wait queue wakeup
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200852 * mechanism. It is called by the stored file descriptors when they
Davide Libenzi7699acd2007-05-10 22:23:23 -0700853 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700855static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700857 int pwake = 0;
858 unsigned long flags;
859 struct epitem *epi = ep_item_from_wait(wait);
860 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
Oleg Nesterov971316f2012-02-24 20:07:29 +0100862 if ((unsigned long)key & POLLFREE) {
863 ep_pwq_from_wait(wait)->whead = NULL;
864 /*
865 * whead = NULL above can race with ep_remove_wait_queue()
866 * which can do another remove_wait_queue() after us, so we
867 * can't use __remove_wait_queue(). whead->lock is held by
868 * the caller.
869 */
Oleg Nesterovd80e7312012-02-24 20:07:11 +0100870 list_del_init(&wait->task_list);
Oleg Nesterov971316f2012-02-24 20:07:29 +0100871 }
Oleg Nesterovd80e7312012-02-24 20:07:11 +0100872
Davide Libenzic7ea7632007-05-15 01:40:47 -0700873 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700874
875 /*
876 * If the event mask does not contain any poll(2) event, we consider the
877 * descriptor to be disabled. This condition is likely the effect of the
878 * EPOLLONESHOT bit that disables the descriptor when an event is received,
879 * until the next EPOLL_CTL_MOD will be issued.
880 */
881 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700882 goto out_unlock;
883
884 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700885 * Check the events coming with the callback. At this stage, not
886 * every device reports the events in the "key" parameter of the
887 * callback. We need to be able to handle both cases here, hence the
888 * test for "key" != NULL before the event match test.
889 */
890 if (key && !((unsigned long) key & epi->event.events))
891 goto out_unlock;
892
893 /*
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200894 * If we are transferring events to userspace, we can hold no locks
Davide Libenzid47de162007-05-15 01:40:41 -0700895 * (because we're accessing user memory, and because of linux f_op->poll()
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200896 * semantics). All the events that happen during that period of time are
Davide Libenzid47de162007-05-15 01:40:41 -0700897 * chained in ep->ovflist and requeued later on.
898 */
899 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
900 if (epi->next == EP_UNACTIVE_PTR) {
901 epi->next = ep->ovflist;
902 ep->ovflist = epi;
903 }
904 goto out_unlock;
905 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700906
907 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700908 if (!ep_is_linked(&epi->rdllink))
909 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700910
Davide Libenzi7699acd2007-05-10 22:23:23 -0700911 /*
912 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
913 * wait list.
914 */
915 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400916 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700917 if (waitqueue_active(&ep->poll_wait))
918 pwake++;
919
Davide Libenzid47de162007-05-15 01:40:41 -0700920out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700921 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700922
923 /* We have to call this outside the lock */
924 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700925 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700926
927 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930/*
931 * This is the callback that is used to add our wait queue to the
932 * target file wakeup lists.
933 */
934static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
935 poll_table *pt)
936{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700937 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 struct eppoll_entry *pwq;
939
Christoph Lametere94b1762006-12-06 20:33:17 -0800940 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
942 pwq->whead = whead;
943 pwq->base = epi;
944 add_wait_queue(whead, &pwq->wait);
945 list_add_tail(&pwq->llink, &epi->pwqlist);
946 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700947 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 /* We have to signal that an error occurred */
949 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
954{
955 int kcmp;
956 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
957 struct epitem *epic;
958
959 while (*p) {
960 parent = *p;
961 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700962 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 if (kcmp > 0)
964 p = &parent->rb_right;
965 else
966 p = &parent->rb_left;
967 }
968 rb_link_node(&epi->rbn, parent, p);
969 rb_insert_color(&epi->rbn, &ep->rbr);
970}
971
Jason Baron28d82dc2012-01-12 17:17:43 -0800972
973
974#define PATH_ARR_SIZE 5
975/*
976 * These are the number paths of length 1 to 5, that we are allowing to emanate
977 * from a single file of interest. For example, we allow 1000 paths of length
978 * 1, to emanate from each file of interest. This essentially represents the
979 * potential wakeup paths, which need to be limited in order to avoid massive
980 * uncontrolled wakeup storms. The common use case should be a single ep which
981 * is connected to n file sources. In this case each file source has 1 path
982 * of length 1. Thus, the numbers below should be more than sufficient. These
983 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
984 * and delete can't add additional paths. Protected by the epmutex.
985 */
986static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
987static int path_count[PATH_ARR_SIZE];
988
989static int path_count_inc(int nests)
990{
991 if (++path_count[nests] > path_limits[nests])
992 return -1;
993 return 0;
994}
995
996static void path_count_init(void)
997{
998 int i;
999
1000 for (i = 0; i < PATH_ARR_SIZE; i++)
1001 path_count[i] = 0;
1002}
1003
1004static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1005{
1006 int error = 0;
1007 struct file *file = priv;
1008 struct file *child_file;
1009 struct epitem *epi;
1010
1011 list_for_each_entry(epi, &file->f_ep_links, fllink) {
1012 child_file = epi->ep->file;
1013 if (is_file_epoll(child_file)) {
1014 if (list_empty(&child_file->f_ep_links)) {
1015 if (path_count_inc(call_nests)) {
1016 error = -1;
1017 break;
1018 }
1019 } else {
1020 error = ep_call_nested(&poll_loop_ncalls,
1021 EP_MAX_NESTS,
1022 reverse_path_check_proc,
1023 child_file, child_file,
1024 current);
1025 }
1026 if (error != 0)
1027 break;
1028 } else {
1029 printk(KERN_ERR "reverse_path_check_proc: "
1030 "file is not an ep!\n");
1031 }
1032 }
1033 return error;
1034}
1035
1036/**
1037 * reverse_path_check - The tfile_check_list is list of file *, which have
1038 * links that are proposed to be newly added. We need to
1039 * make sure that those added links don't add too many
1040 * paths such that we will spend all our time waking up
1041 * eventpoll objects.
1042 *
1043 * Returns: Returns zero if the proposed links don't create too many paths,
1044 * -1 otherwise.
1045 */
1046static int reverse_path_check(void)
1047{
1048 int length = 0;
1049 int error = 0;
1050 struct file *current_file;
1051
1052 /* let's call this for all tfiles */
1053 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1054 length++;
1055 path_count_init();
1056 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1057 reverse_path_check_proc, current_file,
1058 current_file, current);
1059 if (error)
1060 break;
1061 }
1062 return error;
1063}
1064
Davide Libenzic7ea7632007-05-15 01:40:47 -07001065/*
1066 * Must be called with "mtx" held.
1067 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1069 struct file *tfile, int fd)
1070{
1071 int error, revents, pwake = 0;
1072 unsigned long flags;
Robin Holt52bd19f72011-01-12 17:00:01 -08001073 long user_watches;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 struct epitem *epi;
1075 struct ep_pqueue epq;
1076
Robin Holt52bd19f72011-01-12 17:00:01 -08001077 user_watches = atomic_long_read(&ep->user->epoll_watches);
1078 if (unlikely(user_watches >= max_user_watches))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001079 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -08001080 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001081 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
1083 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 INIT_LIST_HEAD(&epi->rdllink);
1085 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 INIT_LIST_HEAD(&epi->pwqlist);
1087 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001088 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -07001091 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093 /* Initialize the poll table using the queue callback */
1094 epq.epi = epi;
1095 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1096
1097 /*
1098 * Attach the item to the poll hooks and get current event bits.
1099 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -07001100 * been increased by the caller of this function. Note that after
1101 * this operation completes, the poll callback can start hitting
1102 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 */
1104 revents = tfile->f_op->poll(tfile, &epq.pt);
1105
1106 /*
1107 * We have to check if something went wrong during the poll wait queue
1108 * install process. Namely an allocation for a wait queue failed due
1109 * high memory pressure.
1110 */
Davide Libenzi7ef99642008-12-01 13:13:55 -08001111 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001113 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
1115 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -07001116 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -07001118 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Davide Libenzic7ea7632007-05-15 01:40:47 -07001120 /*
1121 * Add the current item to the RB tree. All RB tree operations are
1122 * protected by "mtx", and ep_insert() is called with "mtx" held.
1123 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 ep_rbtree_insert(ep, epi);
1125
Jason Baron28d82dc2012-01-12 17:17:43 -08001126 /* now check if we've created too many backpaths */
1127 error = -EINVAL;
1128 if (reverse_path_check())
1129 goto error_remove_epi;
1130
Davide Libenzic7ea7632007-05-15 01:40:47 -07001131 /* We have to drop the new item inside our item list to keep track of it */
1132 spin_lock_irqsave(&ep->lock, flags);
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001135 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 list_add_tail(&epi->rdllink, &ep->rdllist);
1137
1138 /* Notify waiting tasks that events are available */
1139 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001140 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 if (waitqueue_active(&ep->poll_wait))
1142 pwake++;
1143 }
1144
Davide Libenzic7ea7632007-05-15 01:40:47 -07001145 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Robin Holt52bd19f72011-01-12 17:00:01 -08001147 atomic_long_inc(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 /* We have to call this outside the lock */
1150 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001151 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 return 0;
1154
Jason Baron28d82dc2012-01-12 17:17:43 -08001155error_remove_epi:
1156 spin_lock(&tfile->f_lock);
1157 if (ep_is_linked(&epi->fllink))
1158 list_del_init(&epi->fllink);
1159 spin_unlock(&tfile->f_lock);
1160
1161 rb_erase(&epi->rbn, &ep->rbr);
1162
Davide Libenzi7699acd2007-05-10 22:23:23 -07001163error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 ep_unregister_pollwait(ep, epi);
1165
1166 /*
1167 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -07001168 * allocated wait queue. Note that we don't care about the ep->ovflist
1169 * list, since that is used/cleaned only inside a section bound by "mtx".
1170 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001172 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001173 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -07001174 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001175 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001177 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 return error;
1180}
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182/*
1183 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001184 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 */
1186static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1187{
1188 int pwake = 0;
1189 unsigned int revents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001192 * Set the new event interest mask before calling f_op->poll();
1193 * otherwise we might miss an event that happens between the
1194 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 */
1196 epi->event.events = event->events;
Tony Battersbye057e152009-03-31 15:24:15 -07001197 epi->event.data = event->data; /* protected by mtx */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 /*
1200 * Get current event bits. We can safely use the file* here because
1201 * its usage count has been increased by the caller of this function.
1202 */
1203 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL);
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001206 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001207 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001209 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001210 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001211 if (!ep_is_linked(&epi->rdllink)) {
1212 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Davide Libenzic7ea7632007-05-15 01:40:47 -07001214 /* Notify waiting tasks that events are available */
1215 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001216 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001217 if (waitqueue_active(&ep->poll_wait))
1218 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 }
Tony Battersbye057e152009-03-31 15:24:15 -07001220 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
1223 /* We have to call this outside the lock */
1224 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001225 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 return 0;
1228}
1229
Davide Libenzi296e2362009-03-31 15:24:11 -07001230static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1231 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001232{
1233 struct ep_send_events_data *esed = priv;
1234 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001235 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001236 struct epitem *epi;
1237 struct epoll_event __user *uevent;
1238
Davide Libenzi296e2362009-03-31 15:24:11 -07001239 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001240 * We can loop without lock because we are passed a task private list.
1241 * Items cannot vanish during the loop because ep_scan_ready_list() is
1242 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001243 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001244 for (eventcnt = 0, uevent = esed->events;
1245 !list_empty(head) && eventcnt < esed->maxevents;) {
1246 epi = list_first_entry(head, struct epitem, rdllink);
1247
1248 list_del_init(&epi->rdllink);
1249
Davide Libenzi296e2362009-03-31 15:24:11 -07001250 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
1251 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001252
Davide Libenzi296e2362009-03-31 15:24:11 -07001253 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001254 * If the event mask intersect the caller-requested one,
1255 * deliver the event to userspace. Again, ep_scan_ready_list()
1256 * is holding "mtx", so no operations coming from userspace
1257 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001258 */
1259 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001260 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001261 __put_user(epi->event.data, &uevent->data)) {
1262 list_add(&epi->rdllink, head);
Davide Libenzi296e2362009-03-31 15:24:11 -07001263 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001264 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001265 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001266 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001267 if (epi->event.events & EPOLLONESHOT)
1268 epi->event.events &= EP_PRIVATE_BITS;
1269 else if (!(epi->event.events & EPOLLET)) {
1270 /*
1271 * If this file has been added with Level
1272 * Trigger mode, we need to insert back inside
1273 * the ready list, so that the next call to
1274 * epoll_wait() will check again the events
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001275 * availability. At this point, no one can insert
Davide Libenzi296e2362009-03-31 15:24:11 -07001276 * into ep->rdllist besides us. The epoll_ctl()
1277 * callers are locked out by
1278 * ep_scan_ready_list() holding "mtx" and the
1279 * poll callback will queue them in ep->ovflist.
1280 */
1281 list_add_tail(&epi->rdllink, &ep->rdllist);
1282 }
1283 }
1284 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001285
1286 return eventcnt;
1287}
1288
Davide Libenzi296e2362009-03-31 15:24:11 -07001289static int ep_send_events(struct eventpoll *ep,
1290 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291{
Davide Libenzi5071f972009-03-31 15:24:10 -07001292 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001293
Davide Libenzi5071f972009-03-31 15:24:10 -07001294 esed.maxevents = maxevents;
1295 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001296
Nelson Elhaged8805e62011-10-31 17:13:14 -07001297 return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298}
1299
Eric Dumazet0781b902011-02-01 15:52:35 -08001300static inline struct timespec ep_set_mstimeout(long ms)
1301{
1302 struct timespec now, ts = {
1303 .tv_sec = ms / MSEC_PER_SEC,
1304 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1305 };
1306
1307 ktime_get_ts(&now);
1308 return timespec_add_safe(now, ts);
1309}
1310
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001311/**
1312 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1313 * event buffer.
1314 *
1315 * @ep: Pointer to the eventpoll context.
1316 * @events: Pointer to the userspace buffer where the ready events should be
1317 * stored.
1318 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1319 * @timeout: Maximum timeout for the ready events fetch operation, in
1320 * milliseconds. If the @timeout is zero, the function will not block,
1321 * while if the @timeout is less than zero, the function will block
1322 * until at least one event has been retrieved (or an error
1323 * occurred).
1324 *
1325 * Returns: Returns the number of ready events which have been fetched, or an
1326 * error code, in case of error.
1327 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1329 int maxevents, long timeout)
1330{
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001331 int res = 0, eavail, timed_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 unsigned long flags;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001333 long slack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 wait_queue_t wait;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001335 ktime_t expires, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001337 if (timeout > 0) {
Eric Dumazet0781b902011-02-01 15:52:35 -08001338 struct timespec end_time = ep_set_mstimeout(timeout);
1339
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001340 slack = select_estimate_accuracy(&end_time);
1341 to = &expires;
1342 *to = timespec_to_ktime(end_time);
1343 } else if (timeout == 0) {
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001344 /*
1345 * Avoid the unnecessary trip to the wait queue loop, if the
1346 * caller specified a non blocking operation.
1347 */
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001348 timed_out = 1;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001349 spin_lock_irqsave(&ep->lock, flags);
1350 goto check_events;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001353fetch_events:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001354 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001356 if (!ep_events_available(ep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 /*
1358 * We don't have any available event to return to the caller.
1359 * We need to sleep here, and we will be wake up by
1360 * ep_poll_callback() when events will become available.
1361 */
1362 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001363 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365 for (;;) {
1366 /*
1367 * We don't want to sleep if the ep_poll_callback() sends us
1368 * a wakeup in between. That's why we set the task state
1369 * to TASK_INTERRUPTIBLE before doing the checks.
1370 */
1371 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001372 if (ep_events_available(ep) || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 break;
1374 if (signal_pending(current)) {
1375 res = -EINTR;
1376 break;
1377 }
1378
Davide Libenzic7ea7632007-05-15 01:40:47 -07001379 spin_unlock_irqrestore(&ep->lock, flags);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001380 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1381 timed_out = 1;
1382
Davide Libenzic7ea7632007-05-15 01:40:47 -07001383 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001385 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
1387 set_current_state(TASK_RUNNING);
1388 }
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001389check_events:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 /* Is it worth to try to dig for events ? */
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001391 eavail = ep_events_available(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Davide Libenzic7ea7632007-05-15 01:40:47 -07001393 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 /*
1396 * Try to transfer events to user space. In case we get 0 events and
1397 * there's still timeout left over, we go trying again in search of
1398 * more luck.
1399 */
1400 if (!res && eavail &&
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001401 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001402 goto fetch_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
1404 return res;
1405}
1406
Davide Libenzi22bacca2011-02-25 14:44:12 -08001407/**
1408 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1409 * API, to verify that adding an epoll file inside another
1410 * epoll structure, does not violate the constraints, in
1411 * terms of closed loops, or too deep chains (which can
1412 * result in excessive stack usage).
1413 *
1414 * @priv: Pointer to the epoll file to be currently checked.
1415 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1416 * data structure pointer.
1417 * @call_nests: Current dept of the @ep_call_nested() call stack.
1418 *
1419 * Returns: Returns zero if adding the epoll @file inside current epoll
1420 * structure @ep does not violate the constraints, or -1 otherwise.
1421 */
1422static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1423{
1424 int error = 0;
1425 struct file *file = priv;
1426 struct eventpoll *ep = file->private_data;
Jason Baron28d82dc2012-01-12 17:17:43 -08001427 struct eventpoll *ep_tovisit;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001428 struct rb_node *rbp;
1429 struct epitem *epi;
1430
Nelson Elhaged8805e62011-10-31 17:13:14 -07001431 mutex_lock_nested(&ep->mtx, call_nests + 1);
Jason Baron28d82dc2012-01-12 17:17:43 -08001432 ep->visited = 1;
1433 list_add(&ep->visited_list_link, &visited_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001434 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1435 epi = rb_entry(rbp, struct epitem, rbn);
1436 if (unlikely(is_file_epoll(epi->ffd.file))) {
Jason Baron28d82dc2012-01-12 17:17:43 -08001437 ep_tovisit = epi->ffd.file->private_data;
1438 if (ep_tovisit->visited)
1439 continue;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001440 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Jason Baron28d82dc2012-01-12 17:17:43 -08001441 ep_loop_check_proc, epi->ffd.file,
1442 ep_tovisit, current);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001443 if (error != 0)
1444 break;
Jason Baron28d82dc2012-01-12 17:17:43 -08001445 } else {
1446 /*
1447 * If we've reached a file that is not associated with
1448 * an ep, then we need to check if the newly added
1449 * links are going to add too many wakeup paths. We do
1450 * this by adding it to the tfile_check_list, if it's
1451 * not already there, and calling reverse_path_check()
1452 * during ep_insert().
1453 */
1454 if (list_empty(&epi->ffd.file->f_tfile_llink))
1455 list_add(&epi->ffd.file->f_tfile_llink,
1456 &tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001457 }
1458 }
1459 mutex_unlock(&ep->mtx);
1460
1461 return error;
1462}
1463
1464/**
1465 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1466 * another epoll file (represented by @ep) does not create
1467 * closed loops or too deep chains.
1468 *
1469 * @ep: Pointer to the epoll private data structure.
1470 * @file: Pointer to the epoll file to be checked.
1471 *
1472 * Returns: Returns zero if adding the epoll @file inside current epoll
1473 * structure @ep does not violate the constraints, or -1 otherwise.
1474 */
1475static int ep_loop_check(struct eventpoll *ep, struct file *file)
1476{
Jason Baron28d82dc2012-01-12 17:17:43 -08001477 int ret;
1478 struct eventpoll *ep_cur, *ep_next;
1479
1480 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Davide Libenzi22bacca2011-02-25 14:44:12 -08001481 ep_loop_check_proc, file, ep, current);
Jason Baron28d82dc2012-01-12 17:17:43 -08001482 /* clear visited list */
1483 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1484 visited_list_link) {
1485 ep_cur->visited = 0;
1486 list_del(&ep_cur->visited_list_link);
1487 }
1488 return ret;
1489}
1490
1491static void clear_tfile_check_list(void)
1492{
1493 struct file *file;
1494
1495 /* first clear the tfile_check_list */
1496 while (!list_empty(&tfile_check_list)) {
1497 file = list_first_entry(&tfile_check_list, struct file,
1498 f_tfile_llink);
1499 list_del_init(&file->f_tfile_llink);
1500 }
1501 INIT_LIST_HEAD(&tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001502}
1503
Davide Libenzi7699acd2007-05-10 22:23:23 -07001504/*
Andrew Morton523723b2008-08-12 15:09:01 -07001505 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001506 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001507SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001508{
Jason Baron28d82dc2012-01-12 17:17:43 -08001509 int error, fd;
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001510 struct eventpoll *ep = NULL;
Jason Baron28d82dc2012-01-12 17:17:43 -08001511 struct file *file;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001512
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001513 /* Check the EPOLL_* constant for consistency. */
1514 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1515
Davide Libenzi296e2362009-03-31 15:24:11 -07001516 if (flags & ~EPOLL_CLOEXEC)
1517 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001518 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001519 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001520 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001521 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001522 if (error < 0)
1523 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001524 /*
1525 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001526 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001527 */
Jason Baron28d82dc2012-01-12 17:17:43 -08001528 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1529 if (fd < 0) {
1530 error = fd;
1531 goto out_free_ep;
1532 }
1533 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001534 O_RDWR | (flags & O_CLOEXEC));
Jason Baron28d82dc2012-01-12 17:17:43 -08001535 if (IS_ERR(file)) {
1536 error = PTR_ERR(file);
1537 goto out_free_fd;
1538 }
1539 fd_install(fd, file);
1540 ep->file = file;
1541 return fd;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001542
Jason Baron28d82dc2012-01-12 17:17:43 -08001543out_free_fd:
1544 put_unused_fd(fd);
1545out_free_ep:
1546 ep_free(ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001547 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001548}
1549
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001550SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001551{
Davide Libenzibfe38912009-05-12 13:19:44 -07001552 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001553 return -EINVAL;
1554
1555 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001556}
1557
Davide Libenzi7699acd2007-05-10 22:23:23 -07001558/*
1559 * The following function implements the controller interface for
1560 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001561 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001562 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001563SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1564 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001565{
1566 int error;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001567 int did_lock_epmutex = 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001568 struct file *file, *tfile;
1569 struct eventpoll *ep;
1570 struct epitem *epi;
1571 struct epoll_event epds;
1572
Davide Libenzi7699acd2007-05-10 22:23:23 -07001573 error = -EFAULT;
1574 if (ep_op_has_event(op) &&
1575 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1576 goto error_return;
1577
1578 /* Get the "struct file *" for the eventpoll file */
1579 error = -EBADF;
1580 file = fget(epfd);
1581 if (!file)
1582 goto error_return;
1583
1584 /* Get the "struct file *" for the target file */
1585 tfile = fget(fd);
1586 if (!tfile)
1587 goto error_fput;
1588
1589 /* The target file descriptor must support poll */
1590 error = -EPERM;
1591 if (!tfile->f_op || !tfile->f_op->poll)
1592 goto error_tgt_fput;
1593
1594 /*
1595 * We have to check that the file structure underneath the file descriptor
1596 * the user passed to us _is_ an eventpoll file. And also we do not permit
1597 * adding an epoll file descriptor inside itself.
1598 */
1599 error = -EINVAL;
1600 if (file == tfile || !is_file_epoll(file))
1601 goto error_tgt_fput;
1602
1603 /*
1604 * At this point it is safe to assume that the "private_data" contains
1605 * our own data structure.
1606 */
1607 ep = file->private_data;
1608
Davide Libenzi22bacca2011-02-25 14:44:12 -08001609 /*
1610 * When we insert an epoll file descriptor, inside another epoll file
1611 * descriptor, there is the change of creating closed loops, which are
Jason Baron28d82dc2012-01-12 17:17:43 -08001612 * better be handled here, than in more critical paths. While we are
1613 * checking for loops we also determine the list of files reachable
1614 * and hang them on the tfile_check_list, so we can check that we
1615 * haven't created too many possible wakeup paths.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001616 *
Jason Baron28d82dc2012-01-12 17:17:43 -08001617 * We need to hold the epmutex across both ep_insert and ep_remove
1618 * b/c we want to make sure we are looking at a coherent view of
1619 * epoll network.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001620 */
Jason Baron28d82dc2012-01-12 17:17:43 -08001621 if (op == EPOLL_CTL_ADD || op == EPOLL_CTL_DEL) {
Davide Libenzi22bacca2011-02-25 14:44:12 -08001622 mutex_lock(&epmutex);
1623 did_lock_epmutex = 1;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001624 }
Jason Baron28d82dc2012-01-12 17:17:43 -08001625 if (op == EPOLL_CTL_ADD) {
1626 if (is_file_epoll(tfile)) {
1627 error = -ELOOP;
1628 if (ep_loop_check(ep, tfile) != 0)
1629 goto error_tgt_fput;
1630 } else
1631 list_add(&tfile->f_tfile_llink, &tfile_check_list);
1632 }
Davide Libenzi22bacca2011-02-25 14:44:12 -08001633
Nelson Elhaged8805e62011-10-31 17:13:14 -07001634 mutex_lock_nested(&ep->mtx, 0);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001635
Davide Libenzi67647d02007-05-15 01:40:52 -07001636 /*
1637 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1638 * above, we can be sure to be able to use the item looked up by
1639 * ep_find() till we release the mutex.
1640 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001641 epi = ep_find(ep, tfile, fd);
1642
1643 error = -EINVAL;
1644 switch (op) {
1645 case EPOLL_CTL_ADD:
1646 if (!epi) {
1647 epds.events |= POLLERR | POLLHUP;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001648 error = ep_insert(ep, &epds, tfile, fd);
1649 } else
1650 error = -EEXIST;
Jason Baron28d82dc2012-01-12 17:17:43 -08001651 clear_tfile_check_list();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001652 break;
1653 case EPOLL_CTL_DEL:
1654 if (epi)
1655 error = ep_remove(ep, epi);
1656 else
1657 error = -ENOENT;
1658 break;
1659 case EPOLL_CTL_MOD:
1660 if (epi) {
1661 epds.events |= POLLERR | POLLHUP;
1662 error = ep_modify(ep, epi, &epds);
1663 } else
1664 error = -ENOENT;
1665 break;
1666 }
Davide Libenzid47de162007-05-15 01:40:41 -07001667 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001668
1669error_tgt_fput:
Jason Baron28d82dc2012-01-12 17:17:43 -08001670 if (did_lock_epmutex)
Davide Libenzi22bacca2011-02-25 14:44:12 -08001671 mutex_unlock(&epmutex);
1672
Davide Libenzi7699acd2007-05-10 22:23:23 -07001673 fput(tfile);
1674error_fput:
1675 fput(file);
1676error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001677
1678 return error;
1679}
1680
1681/*
1682 * Implement the event wait interface for the eventpoll file. It is the kernel
1683 * part of the user space epoll_wait(2).
1684 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001685SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1686 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001687{
1688 int error;
1689 struct file *file;
1690 struct eventpoll *ep;
1691
Davide Libenzi7699acd2007-05-10 22:23:23 -07001692 /* The maximum number of event must be greater than zero */
1693 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1694 return -EINVAL;
1695
1696 /* Verify that the area passed by the user is writeable */
1697 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1698 error = -EFAULT;
1699 goto error_return;
1700 }
1701
1702 /* Get the "struct file *" for the eventpoll file */
1703 error = -EBADF;
1704 file = fget(epfd);
1705 if (!file)
1706 goto error_return;
1707
1708 /*
1709 * We have to check that the file structure underneath the fd
1710 * the user passed to us _is_ an eventpoll file.
1711 */
1712 error = -EINVAL;
1713 if (!is_file_epoll(file))
1714 goto error_fput;
1715
1716 /*
1717 * At this point it is safe to assume that the "private_data" contains
1718 * our own data structure.
1719 */
1720 ep = file->private_data;
1721
1722 /* Time to fish for events ... */
1723 error = ep_poll(ep, events, maxevents, timeout);
1724
1725error_fput:
1726 fput(file);
1727error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001728
1729 return error;
1730}
1731
Roland McGrathf3de2722008-04-30 00:53:09 -07001732#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001733
1734/*
1735 * Implement the event wait interface for the eventpoll file. It is the kernel
1736 * part of the user space epoll_pwait(2).
1737 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001738SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1739 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1740 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001741{
1742 int error;
1743 sigset_t ksigmask, sigsaved;
1744
1745 /*
1746 * If the caller wants a certain signal mask to be set during the wait,
1747 * we apply it here.
1748 */
1749 if (sigmask) {
1750 if (sigsetsize != sizeof(sigset_t))
1751 return -EINVAL;
1752 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1753 return -EFAULT;
1754 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1755 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1756 }
1757
1758 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1759
1760 /*
1761 * If we changed the signal mask, we need to restore the original one.
1762 * In case we've got a signal while waiting, we do not restore the
1763 * signal mask yet, and we allow do_signal() to deliver the signal on
1764 * the way back to userspace, before the signal mask is restored.
1765 */
1766 if (sigmask) {
1767 if (error == -EINTR) {
1768 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001769 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001770 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001771 } else
1772 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1773 }
1774
1775 return error;
1776}
1777
Roland McGrathf3de2722008-04-30 00:53:09 -07001778#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780static int __init eventpoll_init(void)
1781{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001782 struct sysinfo si;
1783
1784 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001785 /*
1786 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1787 */
1788 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001789 EP_ITEM_COST;
Robin Holt52bd19f72011-01-12 17:00:01 -08001790 BUG_ON(max_user_watches < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Davide Libenzi22bacca2011-02-25 14:44:12 -08001792 /*
1793 * Initialize the structure used to perform epoll file descriptor
1794 * inclusion loops checks.
1795 */
1796 ep_nested_calls_init(&poll_loop_ncalls);
1797
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001799 ep_nested_calls_init(&poll_safewake_ncalls);
1800
1801 /* Initialize the structure used to perform file's f_op->poll() calls */
1802 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 /* Allocates slab cache used to allocate "struct epitem" items */
1805 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001806 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
1808 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1809 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001810 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
Davide Libenzicea69242007-05-10 22:23:22 -07001814fs_initcall(eventpoll_init);