blob: c0b3c70ee87a2b8e0e46c01a87d63ac692aecc71 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <asm/io.h>
38#include <asm/mman.h>
Arun Sharma600634972011-07-26 16:09:06 -070039#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/*
42 * LOCKING:
43 * There are three level of locking required by epoll :
44 *
Arjan van de Ven144efe32006-03-23 03:00:32 -080045 * 1) epmutex (mutex)
Davide Libenzic7ea7632007-05-15 01:40:47 -070046 * 2) ep->mtx (mutex)
47 * 3) ep->lock (spinlock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 *
49 * The acquire order is the one listed above, from 1 to 3.
50 * We need a spinlock (ep->lock) because we manipulate objects
51 * from inside the poll callback, that might be triggered from
52 * a wake_up() that in turn might be called from IRQ context.
53 * So we can't sleep inside the poll callback and hence we need
54 * a spinlock. During the event transfer loop (from kernel to
55 * user space) we could end up sleeping due a copy_to_user(), so
56 * we need a lock that will allow us to sleep. This lock is a
Davide Libenzid47de162007-05-15 01:40:41 -070057 * mutex (ep->mtx). It is acquired during the event transfer loop,
58 * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
59 * Then we also need a global mutex to serialize eventpoll_release_file()
60 * and ep_free().
61 * This mutex is acquired by ep_free() during the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * cleanup path and it is also acquired by eventpoll_release_file()
63 * if a file has been pushed inside an epoll set and it is then
Daniel Balutabf6a41d2011-01-30 23:42:29 +020064 * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
Davide Libenzi22bacca2011-02-25 14:44:12 -080065 * It is also acquired when inserting an epoll fd onto another epoll
66 * fd. We do this so that we walk the epoll tree and ensure that this
67 * insertion does not create a cycle of epoll file descriptors, which
68 * could lead to deadlock. We need a global mutex to prevent two
69 * simultaneous inserts (A into B and B into A) from racing and
70 * constructing a cycle without either insert observing that it is
71 * going to.
Nelson Elhaged8805e62011-10-31 17:13:14 -070072 * It is necessary to acquire multiple "ep->mtx"es at once in the
73 * case when one epoll fd is added to another. In this case, we
74 * always acquire the locks in the order of nesting (i.e. after
75 * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
76 * before e2->mtx). Since we disallow cycles of epoll file
77 * descriptors, this ensures that the mutexes are well-ordered. In
78 * order to communicate this nesting to lockdep, when walking a tree
79 * of epoll file descriptors, we use the current recursion depth as
80 * the lockdep subkey.
Davide Libenzid47de162007-05-15 01:40:41 -070081 * It is possible to drop the "ep->mtx" and to use the global
82 * mutex "epmutex" (together with "ep->lock") to have it working,
83 * but having "ep->mtx" will make the interface more scalable.
Arjan van de Ven144efe32006-03-23 03:00:32 -080084 * Events that require holding "epmutex" are very rare, while for
Davide Libenzid47de162007-05-15 01:40:41 -070085 * normal operations the epoll private "ep->mtx" will guarantee
86 * a better scalability.
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 */
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/* Epoll private bits inside the event mask */
90#define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET)
91
Davide Libenzi5071f972009-03-31 15:24:10 -070092/* Maximum number of nesting allowed inside epoll sets */
93#define EP_MAX_NESTS 4
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Davide Libenzib6119672006-10-11 01:21:44 -070095#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
96
Davide Libenzid47de162007-05-15 01:40:41 -070097#define EP_UNACTIVE_PTR ((void *) -1L)
98
Davide Libenzi7ef99642008-12-01 13:13:55 -080099#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101struct epoll_filefd {
102 struct file *file;
103 int fd;
104};
105
106/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700107 * Structure used to track possible nested calls, for too deep recursions
108 * and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700110struct nested_call_node {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 struct list_head llink;
Davide Libenzi5071f972009-03-31 15:24:10 -0700112 void *cookie;
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700113 void *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114};
115
116/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700117 * This structure is used as collector for nested calls, to check for
118 * maximum recursion dept and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700120struct nested_calls {
121 struct list_head tasks_call_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 spinlock_t lock;
123};
124
125/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 * Each file descriptor added to the eventpoll interface will
Davide Libenzi6192bd52007-05-08 00:25:41 -0700127 * have an entry of this type linked to the "rbr" RB tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 */
129struct epitem {
Davide Libenzi67647d02007-05-15 01:40:52 -0700130 /* RB tree node used to link this structure to the eventpoll RB tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 struct rb_node rbn;
132
133 /* List header used to link this structure to the eventpoll ready list */
134 struct list_head rdllink;
135
Davide Libenzic7ea7632007-05-15 01:40:47 -0700136 /*
137 * Works together "struct eventpoll"->ovflist in keeping the
138 * single linked chain of items.
139 */
140 struct epitem *next;
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 /* The file descriptor information this item refers to */
143 struct epoll_filefd ffd;
144
145 /* Number of active wait queue attached to poll operations */
146 int nwait;
147
148 /* List containing poll wait queues */
149 struct list_head pwqlist;
150
151 /* The "container" of this item */
152 struct eventpoll *ep;
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 /* List header used to link this item to the "struct file" items list */
155 struct list_head fllink;
Davide Libenzid47de162007-05-15 01:40:41 -0700156
Davide Libenzic7ea7632007-05-15 01:40:47 -0700157 /* The structure that describe the interested events and the source fd */
158 struct epoll_event event;
Davide Libenzid47de162007-05-15 01:40:41 -0700159};
160
161/*
162 * This structure is stored inside the "private_data" member of the file
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200163 * structure and represents the main data structure for the eventpoll
Davide Libenzid47de162007-05-15 01:40:41 -0700164 * interface.
165 */
166struct eventpoll {
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200167 /* Protect the access to this structure */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700168 spinlock_t lock;
Davide Libenzid47de162007-05-15 01:40:41 -0700169
170 /*
171 * This mutex is used to ensure that files are not removed
172 * while epoll is using them. This is held during the event
173 * collection loop, the file cleanup path, the epoll file exit
174 * code and the ctl operations.
175 */
176 struct mutex mtx;
177
178 /* Wait queue used by sys_epoll_wait() */
179 wait_queue_head_t wq;
180
181 /* Wait queue used by file->poll() */
182 wait_queue_head_t poll_wait;
183
184 /* List of ready file descriptors */
185 struct list_head rdllist;
186
Davide Libenzi67647d02007-05-15 01:40:52 -0700187 /* RB tree root used to store monitored fd structs */
Davide Libenzid47de162007-05-15 01:40:41 -0700188 struct rb_root rbr;
189
190 /*
191 * This is a single linked list that chains all the "struct epitem" that
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300192 * happened while transferring ready events to userspace w/out
Davide Libenzid47de162007-05-15 01:40:41 -0700193 * holding ->lock.
194 */
195 struct epitem *ovflist;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800196
197 /* The user that created the eventpoll descriptor */
198 struct user_struct *user;
Jason Baron28d82dc2012-01-12 17:17:43 -0800199
200 struct file *file;
201
202 /* used to optimize loop detection check */
203 int visited;
204 struct list_head visited_list_link;
Davide Libenzid47de162007-05-15 01:40:41 -0700205};
206
207/* Wait structure used by the poll hooks */
208struct eppoll_entry {
209 /* List header used to link this structure to the "struct epitem" */
210 struct list_head llink;
211
212 /* The "base" pointer is set to the container "struct epitem" */
Tony Battersby4f0989d2009-03-31 15:24:16 -0700213 struct epitem *base;
Davide Libenzid47de162007-05-15 01:40:41 -0700214
215 /*
216 * Wait queue item that will be linked to the target file wait
217 * queue head.
218 */
219 wait_queue_t wait;
220
221 /* The wait queue head that linked the "wait" wait queue item */
222 wait_queue_head_t *whead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223};
224
225/* Wrapper struct used by poll queueing */
226struct ep_pqueue {
227 poll_table pt;
228 struct epitem *epi;
229};
230
Davide Libenzi5071f972009-03-31 15:24:10 -0700231/* Used by the ep_send_events() function as callback private data */
232struct ep_send_events_data {
233 int maxevents;
234 struct epoll_event __user *events;
235};
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237/*
Davide Libenzi7ef99642008-12-01 13:13:55 -0800238 * Configuration options available inside /proc/sys/fs/epoll/
239 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800240/* Maximum number of epoll watched descriptors, per user */
Robin Holt52bd19f72011-01-12 17:00:01 -0800241static long max_user_watches __read_mostly;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800242
243/*
Davide Libenzid47de162007-05-15 01:40:41 -0700244 * This mutex is used to serialize ep_free() and eventpoll_release_file().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800246static DEFINE_MUTEX(epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Davide Libenzi22bacca2011-02-25 14:44:12 -0800248/* Used to check for epoll file descriptor inclusion loops */
249static struct nested_calls poll_loop_ncalls;
250
Davide Libenzi5071f972009-03-31 15:24:10 -0700251/* Used for safe wake up implementation */
252static struct nested_calls poll_safewake_ncalls;
253
254/* Used to call file's f_op->poll() under the nested calls boundaries */
255static struct nested_calls poll_readywalk_ncalls;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257/* Slab cache used to allocate "struct epitem" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800258static struct kmem_cache *epi_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260/* Slab cache used to allocate "struct eppoll_entry" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800261static struct kmem_cache *pwq_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
Jason Baron28d82dc2012-01-12 17:17:43 -0800263/* Visited nodes during ep_loop_check(), so we can unset them when we finish */
264static LIST_HEAD(visited_list);
265
266/*
267 * List of files with newly added links, where we may need to limit the number
268 * of emanating paths. Protected by the epmutex.
269 */
270static LIST_HEAD(tfile_check_list);
271
Davide Libenzi7ef99642008-12-01 13:13:55 -0800272#ifdef CONFIG_SYSCTL
273
274#include <linux/sysctl.h>
275
Robin Holt52bd19f72011-01-12 17:00:01 -0800276static long zero;
277static long long_max = LONG_MAX;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800278
279ctl_table epoll_table[] = {
280 {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800281 .procname = "max_user_watches",
282 .data = &max_user_watches,
Robin Holt52bd19f72011-01-12 17:00:01 -0800283 .maxlen = sizeof(max_user_watches),
Davide Libenzi7ef99642008-12-01 13:13:55 -0800284 .mode = 0644,
Robin Holt52bd19f72011-01-12 17:00:01 -0800285 .proc_handler = proc_doulongvec_minmax,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800286 .extra1 = &zero,
Robin Holt52bd19f72011-01-12 17:00:01 -0800287 .extra2 = &long_max,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800288 },
Eric W. Biedermanab092032009-11-05 14:25:10 -0800289 { }
Davide Libenzi7ef99642008-12-01 13:13:55 -0800290};
291#endif /* CONFIG_SYSCTL */
292
Jason Baron28d82dc2012-01-12 17:17:43 -0800293static const struct file_operations eventpoll_fops;
294
295static inline int is_file_epoll(struct file *f)
296{
297 return f->f_op == &eventpoll_fops;
298}
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700299
Davide Libenzi67647d02007-05-15 01:40:52 -0700300/* Setup the structure that is used as key for the RB tree */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700301static inline void ep_set_ffd(struct epoll_filefd *ffd,
302 struct file *file, int fd)
303{
304 ffd->file = file;
305 ffd->fd = fd;
306}
307
Davide Libenzi67647d02007-05-15 01:40:52 -0700308/* Compare RB tree keys */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700309static inline int ep_cmp_ffd(struct epoll_filefd *p1,
310 struct epoll_filefd *p2)
311{
312 return (p1->file > p2->file ? +1:
313 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
314}
315
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700316/* Tells us if the item is currently linked */
317static inline int ep_is_linked(struct list_head *p)
318{
319 return !list_empty(p);
320}
321
Oleg Nesterov971316f2012-02-24 20:07:29 +0100322static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
323{
324 return container_of(p, struct eppoll_entry, wait);
325}
326
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700327/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700328static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700329{
330 return container_of(p, struct eppoll_entry, wait)->base;
331}
332
333/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700334static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700335{
336 return container_of(p, struct ep_pqueue, pt)->epi;
337}
338
339/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700340static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700341{
342 return op != EPOLL_CTL_DEL;
343}
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700346static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
Davide Libenzi5071f972009-03-31 15:24:10 -0700348 INIT_LIST_HEAD(&ncalls->tasks_call_list);
349 spin_lock_init(&ncalls->lock);
350}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Davide Libenzi5071f972009-03-31 15:24:10 -0700352/**
Davide Libenzi3fb0e582011-03-22 16:34:46 -0700353 * ep_events_available - Checks if ready events might be available.
354 *
355 * @ep: Pointer to the eventpoll context.
356 *
357 * Returns: Returns a value different than zero if ready events are available,
358 * or zero otherwise.
359 */
360static inline int ep_events_available(struct eventpoll *ep)
361{
362 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
363}
364
365/**
Davide Libenzi5071f972009-03-31 15:24:10 -0700366 * ep_call_nested - Perform a bound (possibly) nested call, by checking
367 * that the recursion limit is not exceeded, and that
368 * the same nested call (by the meaning of same cookie) is
369 * no re-entered.
370 *
371 * @ncalls: Pointer to the nested_calls structure to be used for this call.
372 * @max_nests: Maximum number of allowed nesting calls.
373 * @nproc: Nested call core function pointer.
374 * @priv: Opaque data to be passed to the @nproc callback.
375 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700376 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700377 *
378 * Returns: Returns the code returned by the @nproc callback, or -1 if
379 * the maximum recursion limit has been exceeded.
380 */
381static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
382 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700383 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700384{
385 int error, call_nests = 0;
386 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700387 struct list_head *lsthead = &ncalls->tasks_call_list;
388 struct nested_call_node *tncur;
389 struct nested_call_node tnode;
390
391 spin_lock_irqsave(&ncalls->lock, flags);
392
393 /*
394 * Try to see if the current task is already inside this wakeup call.
395 * We use a list here, since the population inside this set is always
396 * very much limited.
397 */
398 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700399 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700400 (tncur->cookie == cookie || ++call_nests > max_nests)) {
401 /*
402 * Ops ... loop detected or maximum nest level reached.
403 * We abort this wake by breaking the cycle itself.
404 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700405 error = -1;
406 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700407 }
408 }
409
410 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700411 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700412 tnode.cookie = cookie;
413 list_add(&tnode.llink, lsthead);
414
415 spin_unlock_irqrestore(&ncalls->lock, flags);
416
417 /* Call the nested function */
418 error = (*nproc)(priv, cookie, call_nests);
419
420 /* Remove the current task from the list */
421 spin_lock_irqsave(&ncalls->lock, flags);
422 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700423out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700424 spin_unlock_irqrestore(&ncalls->lock, flags);
425
426 return error;
427}
428
Steven Rostedt02edc6f2012-03-23 15:02:27 -0700429/*
430 * As described in commit 0ccf831cb lockdep: annotate epoll
431 * the use of wait queues used by epoll is done in a very controlled
432 * manner. Wake ups can nest inside each other, but are never done
433 * with the same locking. For example:
434 *
435 * dfd = socket(...);
436 * efd1 = epoll_create();
437 * efd2 = epoll_create();
438 * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
439 * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
440 *
441 * When a packet arrives to the device underneath "dfd", the net code will
442 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
443 * callback wakeup entry on that queue, and the wake_up() performed by the
444 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
445 * (efd1) notices that it may have some event ready, so it needs to wake up
446 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
447 * that ends up in another wake_up(), after having checked about the
448 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
449 * avoid stack blasting.
450 *
451 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
452 * this special case of epoll.
453 */
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700454#ifdef CONFIG_DEBUG_LOCK_ALLOC
455static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
456 unsigned long events, int subclass)
457{
458 unsigned long flags;
459
460 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
461 wake_up_locked_poll(wqueue, events);
462 spin_unlock_irqrestore(&wqueue->lock, flags);
463}
464#else
465static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
466 unsigned long events, int subclass)
467{
468 wake_up_poll(wqueue, events);
469}
470#endif
471
Davide Libenzi5071f972009-03-31 15:24:10 -0700472static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
473{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700474 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
475 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700476 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479/*
480 * Perform a safe wake up of the poll wait list. The problem is that
481 * with the new callback'd wake up system, it is possible that the
482 * poll callback is reentered from inside the call to wake_up() done
483 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700484 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 * and we cannot reenter the same wait queue head at all. This will
486 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700487 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700489static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700491 int this_cpu = get_cpu();
492
Davide Libenzi5071f972009-03-31 15:24:10 -0700493 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700494 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
495
496 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
Oleg Nesterov971316f2012-02-24 20:07:29 +0100499static void ep_remove_wait_queue(struct eppoll_entry *pwq)
500{
501 wait_queue_head_t *whead;
502
503 rcu_read_lock();
504 /* If it is cleared by POLLFREE, it should be rcu-safe */
505 whead = rcu_dereference(pwq->whead);
506 if (whead)
507 remove_wait_queue(whead, &pwq->wait);
508 rcu_read_unlock();
509}
510
Davide Libenzi7699acd2007-05-10 22:23:23 -0700511/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700512 * This function unregisters poll callbacks from the associated file
513 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
514 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700515 */
516static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
517{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700518 struct list_head *lsthead = &epi->pwqlist;
519 struct eppoll_entry *pwq;
520
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700521 while (!list_empty(lsthead)) {
522 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700523
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700524 list_del(&pwq->llink);
Oleg Nesterov971316f2012-02-24 20:07:29 +0100525 ep_remove_wait_queue(pwq);
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700526 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700527 }
528}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Davide Libenzi5071f972009-03-31 15:24:10 -0700530/**
531 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
532 * the scan code, to call f_op->poll(). Also allows for
533 * O(NumReady) performance.
534 *
535 * @ep: Pointer to the epoll private data structure.
536 * @sproc: Pointer to the scan callback.
537 * @priv: Private opaque data passed to the @sproc callback.
Nelson Elhaged8805e62011-10-31 17:13:14 -0700538 * @depth: The current depth of recursive f_op->poll calls.
Davide Libenzi5071f972009-03-31 15:24:10 -0700539 *
540 * Returns: The same integer error code returned by the @sproc callback.
541 */
542static int ep_scan_ready_list(struct eventpoll *ep,
543 int (*sproc)(struct eventpoll *,
544 struct list_head *, void *),
Nelson Elhaged8805e62011-10-31 17:13:14 -0700545 void *priv,
546 int depth)
Davide Libenzi5071f972009-03-31 15:24:10 -0700547{
548 int error, pwake = 0;
549 unsigned long flags;
550 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700551 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700552
553 /*
554 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700555 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700556 */
Nelson Elhaged8805e62011-10-31 17:13:14 -0700557 mutex_lock_nested(&ep->mtx, depth);
Davide Libenzi5071f972009-03-31 15:24:10 -0700558
559 /*
560 * Steal the ready list, and re-init the original one to the
561 * empty list. Also, set ep->ovflist to NULL so that events
562 * happening while looping w/out locks, are not lost. We cannot
563 * have the poll callback to queue directly on ep->rdllist,
564 * because we want the "sproc" callback to be able to do it
565 * in a lockless way.
566 */
567 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700568 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700569 ep->ovflist = NULL;
570 spin_unlock_irqrestore(&ep->lock, flags);
571
572 /*
573 * Now call the callback function.
574 */
575 error = (*sproc)(ep, &txlist, priv);
576
577 spin_lock_irqsave(&ep->lock, flags);
578 /*
579 * During the time we spent inside the "sproc" callback, some
580 * other events might have been queued by the poll callback.
581 * We re-insert them inside the main ready-list here.
582 */
583 for (nepi = ep->ovflist; (epi = nepi) != NULL;
584 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
585 /*
586 * We need to check if the item is already in the list.
587 * During the "sproc" callback execution time, items are
588 * queued into ->ovflist but the "txlist" might already
589 * contain them, and the list_splice() below takes care of them.
590 */
591 if (!ep_is_linked(&epi->rdllink))
592 list_add_tail(&epi->rdllink, &ep->rdllist);
593 }
594 /*
595 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
596 * releasing the lock, events will be queued in the normal way inside
597 * ep->rdllist.
598 */
599 ep->ovflist = EP_UNACTIVE_PTR;
600
601 /*
602 * Quickly re-inject items left on "txlist".
603 */
604 list_splice(&txlist, &ep->rdllist);
605
606 if (!list_empty(&ep->rdllist)) {
607 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700608 * Wake up (if active) both the eventpoll wait list and
609 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700610 */
611 if (waitqueue_active(&ep->wq))
612 wake_up_locked(&ep->wq);
613 if (waitqueue_active(&ep->poll_wait))
614 pwake++;
615 }
616 spin_unlock_irqrestore(&ep->lock, flags);
617
618 mutex_unlock(&ep->mtx);
619
620 /* We have to call this outside the lock */
621 if (pwake)
622 ep_poll_safewake(&ep->poll_wait);
623
624 return error;
625}
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700628 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700629 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700630 */
631static int ep_remove(struct eventpoll *ep, struct epitem *epi)
632{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700633 unsigned long flags;
634 struct file *file = epi->ffd.file;
635
636 /*
637 * Removes poll wait queue hooks. We _have_ to do this without holding
638 * the "ep->lock" otherwise a deadlock might occur. This because of the
639 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
640 * queue head lock when unregistering the wait queue. The wakeup callback
641 * will run by holding the wait queue head lock and will call our callback
642 * that will try to get "ep->lock".
643 */
644 ep_unregister_pollwait(ep, epi);
645
646 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700647 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700648 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700649 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700650 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Davide Libenzicdac75e2008-04-29 00:58:34 -0700652 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Davide Libenzic7ea7632007-05-15 01:40:47 -0700654 spin_lock_irqsave(&ep->lock, flags);
655 if (ep_is_linked(&epi->rdllink))
656 list_del_init(&epi->rdllink);
657 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Davide Libenzi7699acd2007-05-10 22:23:23 -0700659 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700660 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Robin Holt52bd19f72011-01-12 17:00:01 -0800662 atomic_long_dec(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800663
Davide Libenzic7ea7632007-05-15 01:40:47 -0700664 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667static void ep_free(struct eventpoll *ep)
668{
669 struct rb_node *rbp;
670 struct epitem *epi;
671
672 /* We need to release all tasks waiting for these file */
673 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700674 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
676 /*
677 * We need to lock this because we could be hit by
678 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700679 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 * is on the way to be removed and no one has references to it
681 * anymore. The only hit might come from eventpoll_release_file() but
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300682 * holding "epmutex" is sufficient here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800684 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
686 /*
687 * Walks through the whole tree by unregistering poll callbacks.
688 */
689 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
690 epi = rb_entry(rbp, struct epitem, rbn);
691
692 ep_unregister_pollwait(ep, epi);
693 }
694
695 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700696 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700698 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 * us during this operation. So we can avoid the lock on "ep->lock".
700 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700701 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 epi = rb_entry(rbp, struct epitem, rbn);
703 ep_remove(ep, epi);
704 }
705
Arjan van de Ven144efe32006-03-23 03:00:32 -0800706 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700707 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800708 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700709 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Davide Libenzi7699acd2007-05-10 22:23:23 -0700712static int ep_eventpoll_release(struct inode *inode, struct file *file)
713{
714 struct eventpoll *ep = file->private_data;
715
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700716 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700717 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700718
Davide Libenzi7699acd2007-05-10 22:23:23 -0700719 return 0;
720}
721
Davide Libenzi296e2362009-03-31 15:24:11 -0700722static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
723 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700724{
725 struct epitem *epi, *tmp;
Hans Verkuil626cf232012-03-23 15:02:27 -0700726 poll_table pt;
Davide Libenzi5071f972009-03-31 15:24:10 -0700727
Hans Verkuil626cf232012-03-23 15:02:27 -0700728 init_poll_funcptr(&pt, NULL);
Davide Libenzi5071f972009-03-31 15:24:10 -0700729 list_for_each_entry_safe(epi, tmp, head, rdllink) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700730 pt._key = epi->event.events;
731 if (epi->ffd.file->f_op->poll(epi->ffd.file, &pt) &
Davide Libenzi5071f972009-03-31 15:24:10 -0700732 epi->event.events)
733 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700734 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700735 /*
736 * Item has been dropped into the ready list by the poll
737 * callback, but it's not actually ready, as far as
738 * caller requested events goes. We can remove it here.
739 */
740 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700741 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700742 }
743
744 return 0;
745}
746
747static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
748{
Nelson Elhaged8805e62011-10-31 17:13:14 -0700749 return ep_scan_ready_list(priv, ep_read_events_proc, NULL, call_nests + 1);
Davide Libenzi5071f972009-03-31 15:24:10 -0700750}
751
Davide Libenzi7699acd2007-05-10 22:23:23 -0700752static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
753{
Davide Libenzi5071f972009-03-31 15:24:10 -0700754 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700755 struct eventpoll *ep = file->private_data;
756
757 /* Insert inside our poll wait queue */
758 poll_wait(file, &ep->poll_wait, wait);
759
Davide Libenzi5071f972009-03-31 15:24:10 -0700760 /*
761 * Proceed to find out if wanted events are really available inside
762 * the ready list. This need to be done under ep_call_nested()
763 * supervision, since the call to f_op->poll() done on listed files
764 * could re-enter here.
765 */
766 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700767 ep_poll_readyevents_proc, ep, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700768
Davide Libenzi296e2362009-03-31 15:24:11 -0700769 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700770}
771
772/* File callbacks that implement the eventpoll file behaviour */
773static const struct file_operations eventpoll_fops = {
774 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200775 .poll = ep_eventpoll_poll,
776 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700777};
778
Davide Libenzi7699acd2007-05-10 22:23:23 -0700779/*
780 * This is called from eventpoll_release() to unlink files from the eventpoll
781 * interface. We need to have this facility to cleanup correctly files that are
782 * closed without being removed from the eventpoll interface.
783 */
784void eventpoll_release_file(struct file *file)
785{
786 struct list_head *lsthead = &file->f_ep_links;
787 struct eventpoll *ep;
788 struct epitem *epi;
789
790 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700791 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700792 * necessary. It is not necessary because we're in the "struct file"
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300793 * cleanup path, and this means that no one is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700794 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700795 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700796 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700797 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700798 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700799 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700800 *
801 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700802 */
803 mutex_lock(&epmutex);
804
805 while (!list_empty(lsthead)) {
806 epi = list_first_entry(lsthead, struct epitem, fllink);
807
808 ep = epi->ep;
809 list_del_init(&epi->fllink);
Nelson Elhaged8805e62011-10-31 17:13:14 -0700810 mutex_lock_nested(&ep->mtx, 0);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700811 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700812 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700813 }
814
815 mutex_unlock(&epmutex);
816}
817
818static int ep_alloc(struct eventpoll **pep)
819{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800820 int error;
821 struct user_struct *user;
822 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700823
Davide Libenzi7ef99642008-12-01 13:13:55 -0800824 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800825 error = -ENOMEM;
826 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
827 if (unlikely(!ep))
828 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700829
Davide Libenzic7ea7632007-05-15 01:40:47 -0700830 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700831 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700832 init_waitqueue_head(&ep->wq);
833 init_waitqueue_head(&ep->poll_wait);
834 INIT_LIST_HEAD(&ep->rdllist);
835 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700836 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800837 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700838
839 *pep = ep;
840
Davide Libenzi7699acd2007-05-10 22:23:23 -0700841 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800842
843free_uid:
844 free_uid(user);
845 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700846}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700849 * Search the file inside the eventpoll tree. The RB tree operations
850 * are protected by the "mtx" mutex, and ep_find() must be called with
851 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 */
853static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
854{
855 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 struct rb_node *rbp;
857 struct epitem *epi, *epir = NULL;
858 struct epoll_filefd ffd;
859
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700860 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 for (rbp = ep->rbr.rb_node; rbp; ) {
862 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700863 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 if (kcmp > 0)
865 rbp = rbp->rb_right;
866 else if (kcmp < 0)
867 rbp = rbp->rb_left;
868 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 epir = epi;
870 break;
871 }
872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 return epir;
875}
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700878 * This is the callback that is passed to the wait queue wakeup
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200879 * mechanism. It is called by the stored file descriptors when they
Davide Libenzi7699acd2007-05-10 22:23:23 -0700880 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700882static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700884 int pwake = 0;
885 unsigned long flags;
886 struct epitem *epi = ep_item_from_wait(wait);
887 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Oleg Nesterov971316f2012-02-24 20:07:29 +0100889 if ((unsigned long)key & POLLFREE) {
890 ep_pwq_from_wait(wait)->whead = NULL;
891 /*
892 * whead = NULL above can race with ep_remove_wait_queue()
893 * which can do another remove_wait_queue() after us, so we
894 * can't use __remove_wait_queue(). whead->lock is held by
895 * the caller.
896 */
Oleg Nesterovd80e7312012-02-24 20:07:11 +0100897 list_del_init(&wait->task_list);
Oleg Nesterov971316f2012-02-24 20:07:29 +0100898 }
Oleg Nesterovd80e7312012-02-24 20:07:11 +0100899
Davide Libenzic7ea7632007-05-15 01:40:47 -0700900 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700901
902 /*
903 * If the event mask does not contain any poll(2) event, we consider the
904 * descriptor to be disabled. This condition is likely the effect of the
905 * EPOLLONESHOT bit that disables the descriptor when an event is received,
906 * until the next EPOLL_CTL_MOD will be issued.
907 */
908 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700909 goto out_unlock;
910
911 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700912 * Check the events coming with the callback. At this stage, not
913 * every device reports the events in the "key" parameter of the
914 * callback. We need to be able to handle both cases here, hence the
915 * test for "key" != NULL before the event match test.
916 */
917 if (key && !((unsigned long) key & epi->event.events))
918 goto out_unlock;
919
920 /*
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200921 * If we are transferring events to userspace, we can hold no locks
Davide Libenzid47de162007-05-15 01:40:41 -0700922 * (because we're accessing user memory, and because of linux f_op->poll()
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200923 * semantics). All the events that happen during that period of time are
Davide Libenzid47de162007-05-15 01:40:41 -0700924 * chained in ep->ovflist and requeued later on.
925 */
926 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
927 if (epi->next == EP_UNACTIVE_PTR) {
928 epi->next = ep->ovflist;
929 ep->ovflist = epi;
930 }
931 goto out_unlock;
932 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700933
934 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700935 if (!ep_is_linked(&epi->rdllink))
936 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700937
Davide Libenzi7699acd2007-05-10 22:23:23 -0700938 /*
939 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
940 * wait list.
941 */
942 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -0400943 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700944 if (waitqueue_active(&ep->poll_wait))
945 pwake++;
946
Davide Libenzid47de162007-05-15 01:40:41 -0700947out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700948 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700949
950 /* We have to call this outside the lock */
951 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700952 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700953
954 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957/*
958 * This is the callback that is used to add our wait queue to the
959 * target file wakeup lists.
960 */
961static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
962 poll_table *pt)
963{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700964 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 struct eppoll_entry *pwq;
966
Christoph Lametere94b1762006-12-06 20:33:17 -0800967 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
969 pwq->whead = whead;
970 pwq->base = epi;
971 add_wait_queue(whead, &pwq->wait);
972 list_add_tail(&pwq->llink, &epi->pwqlist);
973 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700974 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 /* We have to signal that an error occurred */
976 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700977 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978}
979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
981{
982 int kcmp;
983 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
984 struct epitem *epic;
985
986 while (*p) {
987 parent = *p;
988 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700989 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 if (kcmp > 0)
991 p = &parent->rb_right;
992 else
993 p = &parent->rb_left;
994 }
995 rb_link_node(&epi->rbn, parent, p);
996 rb_insert_color(&epi->rbn, &ep->rbr);
997}
998
Jason Baron28d82dc2012-01-12 17:17:43 -0800999
1000
1001#define PATH_ARR_SIZE 5
1002/*
1003 * These are the number paths of length 1 to 5, that we are allowing to emanate
1004 * from a single file of interest. For example, we allow 1000 paths of length
1005 * 1, to emanate from each file of interest. This essentially represents the
1006 * potential wakeup paths, which need to be limited in order to avoid massive
1007 * uncontrolled wakeup storms. The common use case should be a single ep which
1008 * is connected to n file sources. In this case each file source has 1 path
1009 * of length 1. Thus, the numbers below should be more than sufficient. These
1010 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1011 * and delete can't add additional paths. Protected by the epmutex.
1012 */
1013static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1014static int path_count[PATH_ARR_SIZE];
1015
1016static int path_count_inc(int nests)
1017{
Jason Baron93dc6102012-03-16 16:34:03 -04001018 /* Allow an arbitrary number of depth 1 paths */
1019 if (nests == 0)
1020 return 0;
1021
Jason Baron28d82dc2012-01-12 17:17:43 -08001022 if (++path_count[nests] > path_limits[nests])
1023 return -1;
1024 return 0;
1025}
1026
1027static void path_count_init(void)
1028{
1029 int i;
1030
1031 for (i = 0; i < PATH_ARR_SIZE; i++)
1032 path_count[i] = 0;
1033}
1034
1035static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1036{
1037 int error = 0;
1038 struct file *file = priv;
1039 struct file *child_file;
1040 struct epitem *epi;
1041
1042 list_for_each_entry(epi, &file->f_ep_links, fllink) {
1043 child_file = epi->ep->file;
1044 if (is_file_epoll(child_file)) {
1045 if (list_empty(&child_file->f_ep_links)) {
1046 if (path_count_inc(call_nests)) {
1047 error = -1;
1048 break;
1049 }
1050 } else {
1051 error = ep_call_nested(&poll_loop_ncalls,
1052 EP_MAX_NESTS,
1053 reverse_path_check_proc,
1054 child_file, child_file,
1055 current);
1056 }
1057 if (error != 0)
1058 break;
1059 } else {
1060 printk(KERN_ERR "reverse_path_check_proc: "
1061 "file is not an ep!\n");
1062 }
1063 }
1064 return error;
1065}
1066
1067/**
1068 * reverse_path_check - The tfile_check_list is list of file *, which have
1069 * links that are proposed to be newly added. We need to
1070 * make sure that those added links don't add too many
1071 * paths such that we will spend all our time waking up
1072 * eventpoll objects.
1073 *
1074 * Returns: Returns zero if the proposed links don't create too many paths,
1075 * -1 otherwise.
1076 */
1077static int reverse_path_check(void)
1078{
Jason Baron28d82dc2012-01-12 17:17:43 -08001079 int error = 0;
1080 struct file *current_file;
1081
1082 /* let's call this for all tfiles */
1083 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
Jason Baron28d82dc2012-01-12 17:17:43 -08001084 path_count_init();
1085 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1086 reverse_path_check_proc, current_file,
1087 current_file, current);
1088 if (error)
1089 break;
1090 }
1091 return error;
1092}
1093
Davide Libenzic7ea7632007-05-15 01:40:47 -07001094/*
1095 * Must be called with "mtx" held.
1096 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1098 struct file *tfile, int fd)
1099{
1100 int error, revents, pwake = 0;
1101 unsigned long flags;
Robin Holt52bd19f72011-01-12 17:00:01 -08001102 long user_watches;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 struct epitem *epi;
1104 struct ep_pqueue epq;
1105
Robin Holt52bd19f72011-01-12 17:00:01 -08001106 user_watches = atomic_long_read(&ep->user->epoll_watches);
1107 if (unlikely(user_watches >= max_user_watches))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001108 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -08001109 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001110 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 INIT_LIST_HEAD(&epi->rdllink);
1114 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 INIT_LIST_HEAD(&epi->pwqlist);
1116 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001117 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -07001120 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
1122 /* Initialize the poll table using the queue callback */
1123 epq.epi = epi;
1124 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
Hans Verkuil626cf232012-03-23 15:02:27 -07001125 epq.pt._key = event->events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 /*
1128 * Attach the item to the poll hooks and get current event bits.
1129 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -07001130 * been increased by the caller of this function. Note that after
1131 * this operation completes, the poll callback can start hitting
1132 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 */
1134 revents = tfile->f_op->poll(tfile, &epq.pt);
1135
1136 /*
1137 * We have to check if something went wrong during the poll wait queue
1138 * install process. Namely an allocation for a wait queue failed due
1139 * high memory pressure.
1140 */
Davide Libenzi7ef99642008-12-01 13:13:55 -08001141 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001143 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
1145 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -07001146 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -07001148 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Davide Libenzic7ea7632007-05-15 01:40:47 -07001150 /*
1151 * Add the current item to the RB tree. All RB tree operations are
1152 * protected by "mtx", and ep_insert() is called with "mtx" held.
1153 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 ep_rbtree_insert(ep, epi);
1155
Jason Baron28d82dc2012-01-12 17:17:43 -08001156 /* now check if we've created too many backpaths */
1157 error = -EINVAL;
1158 if (reverse_path_check())
1159 goto error_remove_epi;
1160
Davide Libenzic7ea7632007-05-15 01:40:47 -07001161 /* We have to drop the new item inside our item list to keep track of it */
1162 spin_lock_irqsave(&ep->lock, flags);
1163
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001165 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 list_add_tail(&epi->rdllink, &ep->rdllist);
1167
1168 /* Notify waiting tasks that events are available */
1169 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -04001170 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 if (waitqueue_active(&ep->poll_wait))
1172 pwake++;
1173 }
1174
Davide Libenzic7ea7632007-05-15 01:40:47 -07001175 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Robin Holt52bd19f72011-01-12 17:00:01 -08001177 atomic_long_inc(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 /* We have to call this outside the lock */
1180 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001181 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 return 0;
1184
Jason Baron28d82dc2012-01-12 17:17:43 -08001185error_remove_epi:
1186 spin_lock(&tfile->f_lock);
1187 if (ep_is_linked(&epi->fllink))
1188 list_del_init(&epi->fllink);
1189 spin_unlock(&tfile->f_lock);
1190
1191 rb_erase(&epi->rbn, &ep->rbr);
1192
Davide Libenzi7699acd2007-05-10 22:23:23 -07001193error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 ep_unregister_pollwait(ep, epi);
1195
1196 /*
1197 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -07001198 * allocated wait queue. Note that we don't care about the ep->ovflist
1199 * list, since that is used/cleaned only inside a section bound by "mtx".
1200 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001202 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001203 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -07001204 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001205 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001207 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 return error;
1210}
1211
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212/*
1213 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001214 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 */
1216static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1217{
1218 int pwake = 0;
1219 unsigned int revents;
Hans Verkuil626cf232012-03-23 15:02:27 -07001220 poll_table pt;
1221
1222 init_poll_funcptr(&pt, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
1224 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001225 * Set the new event interest mask before calling f_op->poll();
1226 * otherwise we might miss an event that happens between the
1227 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 */
1229 epi->event.events = event->events;
Hans Verkuil626cf232012-03-23 15:02:27 -07001230 pt._key = event->events;
Tony Battersbye057e152009-03-31 15:24:15 -07001231 epi->event.data = event->data; /* protected by mtx */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 /*
1234 * Get current event bits. We can safely use the file* here because
1235 * its usage count has been increased by the caller of this function.
1236 */
Hans Verkuil626cf232012-03-23 15:02:27 -07001237 revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001240 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001241 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001243 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001244 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001245 if (!ep_is_linked(&epi->rdllink)) {
1246 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Davide Libenzic7ea7632007-05-15 01:40:47 -07001248 /* Notify waiting tasks that events are available */
1249 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -04001250 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001251 if (waitqueue_active(&ep->poll_wait))
1252 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 }
Tony Battersbye057e152009-03-31 15:24:15 -07001254 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
1257 /* We have to call this outside the lock */
1258 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001259 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261 return 0;
1262}
1263
Davide Libenzi296e2362009-03-31 15:24:11 -07001264static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1265 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001266{
1267 struct ep_send_events_data *esed = priv;
1268 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001269 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001270 struct epitem *epi;
1271 struct epoll_event __user *uevent;
Hans Verkuil626cf232012-03-23 15:02:27 -07001272 poll_table pt;
1273
1274 init_poll_funcptr(&pt, NULL);
Davide Libenzi5071f972009-03-31 15:24:10 -07001275
Davide Libenzi296e2362009-03-31 15:24:11 -07001276 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001277 * We can loop without lock because we are passed a task private list.
1278 * Items cannot vanish during the loop because ep_scan_ready_list() is
1279 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001280 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001281 for (eventcnt = 0, uevent = esed->events;
1282 !list_empty(head) && eventcnt < esed->maxevents;) {
1283 epi = list_first_entry(head, struct epitem, rdllink);
1284
1285 list_del_init(&epi->rdllink);
1286
Hans Verkuil626cf232012-03-23 15:02:27 -07001287 pt._key = epi->event.events;
1288 revents = epi->ffd.file->f_op->poll(epi->ffd.file, &pt) &
Davide Libenzi296e2362009-03-31 15:24:11 -07001289 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001290
Davide Libenzi296e2362009-03-31 15:24:11 -07001291 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001292 * If the event mask intersect the caller-requested one,
1293 * deliver the event to userspace. Again, ep_scan_ready_list()
1294 * is holding "mtx", so no operations coming from userspace
1295 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001296 */
1297 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001298 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001299 __put_user(epi->event.data, &uevent->data)) {
1300 list_add(&epi->rdllink, head);
Davide Libenzi296e2362009-03-31 15:24:11 -07001301 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001302 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001303 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001304 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001305 if (epi->event.events & EPOLLONESHOT)
1306 epi->event.events &= EP_PRIVATE_BITS;
1307 else if (!(epi->event.events & EPOLLET)) {
1308 /*
1309 * If this file has been added with Level
1310 * Trigger mode, we need to insert back inside
1311 * the ready list, so that the next call to
1312 * epoll_wait() will check again the events
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001313 * availability. At this point, no one can insert
Davide Libenzi296e2362009-03-31 15:24:11 -07001314 * into ep->rdllist besides us. The epoll_ctl()
1315 * callers are locked out by
1316 * ep_scan_ready_list() holding "mtx" and the
1317 * poll callback will queue them in ep->ovflist.
1318 */
1319 list_add_tail(&epi->rdllink, &ep->rdllist);
1320 }
1321 }
1322 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001323
1324 return eventcnt;
1325}
1326
Davide Libenzi296e2362009-03-31 15:24:11 -07001327static int ep_send_events(struct eventpoll *ep,
1328 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
Davide Libenzi5071f972009-03-31 15:24:10 -07001330 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001331
Davide Libenzi5071f972009-03-31 15:24:10 -07001332 esed.maxevents = maxevents;
1333 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001334
Nelson Elhaged8805e62011-10-31 17:13:14 -07001335 return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
1337
Eric Dumazet0781b902011-02-01 15:52:35 -08001338static inline struct timespec ep_set_mstimeout(long ms)
1339{
1340 struct timespec now, ts = {
1341 .tv_sec = ms / MSEC_PER_SEC,
1342 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1343 };
1344
1345 ktime_get_ts(&now);
1346 return timespec_add_safe(now, ts);
1347}
1348
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001349/**
1350 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1351 * event buffer.
1352 *
1353 * @ep: Pointer to the eventpoll context.
1354 * @events: Pointer to the userspace buffer where the ready events should be
1355 * stored.
1356 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1357 * @timeout: Maximum timeout for the ready events fetch operation, in
1358 * milliseconds. If the @timeout is zero, the function will not block,
1359 * while if the @timeout is less than zero, the function will block
1360 * until at least one event has been retrieved (or an error
1361 * occurred).
1362 *
1363 * Returns: Returns the number of ready events which have been fetched, or an
1364 * error code, in case of error.
1365 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1367 int maxevents, long timeout)
1368{
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001369 int res = 0, eavail, timed_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 unsigned long flags;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001371 long slack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 wait_queue_t wait;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001373 ktime_t expires, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001375 if (timeout > 0) {
Eric Dumazet0781b902011-02-01 15:52:35 -08001376 struct timespec end_time = ep_set_mstimeout(timeout);
1377
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001378 slack = select_estimate_accuracy(&end_time);
1379 to = &expires;
1380 *to = timespec_to_ktime(end_time);
1381 } else if (timeout == 0) {
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001382 /*
1383 * Avoid the unnecessary trip to the wait queue loop, if the
1384 * caller specified a non blocking operation.
1385 */
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001386 timed_out = 1;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001387 spin_lock_irqsave(&ep->lock, flags);
1388 goto check_events;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001391fetch_events:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001392 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001394 if (!ep_events_available(ep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 /*
1396 * We don't have any available event to return to the caller.
1397 * We need to sleep here, and we will be wake up by
1398 * ep_poll_callback() when events will become available.
1399 */
1400 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001401 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 for (;;) {
1404 /*
1405 * We don't want to sleep if the ep_poll_callback() sends us
1406 * a wakeup in between. That's why we set the task state
1407 * to TASK_INTERRUPTIBLE before doing the checks.
1408 */
1409 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001410 if (ep_events_available(ep) || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 break;
1412 if (signal_pending(current)) {
1413 res = -EINTR;
1414 break;
1415 }
1416
Davide Libenzic7ea7632007-05-15 01:40:47 -07001417 spin_unlock_irqrestore(&ep->lock, flags);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001418 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1419 timed_out = 1;
1420
Davide Libenzic7ea7632007-05-15 01:40:47 -07001421 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001423 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425 set_current_state(TASK_RUNNING);
1426 }
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001427check_events:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 /* Is it worth to try to dig for events ? */
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001429 eavail = ep_events_available(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
Davide Libenzic7ea7632007-05-15 01:40:47 -07001431 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
1433 /*
1434 * Try to transfer events to user space. In case we get 0 events and
1435 * there's still timeout left over, we go trying again in search of
1436 * more luck.
1437 */
1438 if (!res && eavail &&
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001439 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001440 goto fetch_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442 return res;
1443}
1444
Davide Libenzi22bacca2011-02-25 14:44:12 -08001445/**
1446 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1447 * API, to verify that adding an epoll file inside another
1448 * epoll structure, does not violate the constraints, in
1449 * terms of closed loops, or too deep chains (which can
1450 * result in excessive stack usage).
1451 *
1452 * @priv: Pointer to the epoll file to be currently checked.
1453 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1454 * data structure pointer.
1455 * @call_nests: Current dept of the @ep_call_nested() call stack.
1456 *
1457 * Returns: Returns zero if adding the epoll @file inside current epoll
1458 * structure @ep does not violate the constraints, or -1 otherwise.
1459 */
1460static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1461{
1462 int error = 0;
1463 struct file *file = priv;
1464 struct eventpoll *ep = file->private_data;
Jason Baron28d82dc2012-01-12 17:17:43 -08001465 struct eventpoll *ep_tovisit;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001466 struct rb_node *rbp;
1467 struct epitem *epi;
1468
Nelson Elhaged8805e62011-10-31 17:13:14 -07001469 mutex_lock_nested(&ep->mtx, call_nests + 1);
Jason Baron28d82dc2012-01-12 17:17:43 -08001470 ep->visited = 1;
1471 list_add(&ep->visited_list_link, &visited_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001472 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1473 epi = rb_entry(rbp, struct epitem, rbn);
1474 if (unlikely(is_file_epoll(epi->ffd.file))) {
Jason Baron28d82dc2012-01-12 17:17:43 -08001475 ep_tovisit = epi->ffd.file->private_data;
1476 if (ep_tovisit->visited)
1477 continue;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001478 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Jason Baron28d82dc2012-01-12 17:17:43 -08001479 ep_loop_check_proc, epi->ffd.file,
1480 ep_tovisit, current);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001481 if (error != 0)
1482 break;
Jason Baron28d82dc2012-01-12 17:17:43 -08001483 } else {
1484 /*
1485 * If we've reached a file that is not associated with
1486 * an ep, then we need to check if the newly added
1487 * links are going to add too many wakeup paths. We do
1488 * this by adding it to the tfile_check_list, if it's
1489 * not already there, and calling reverse_path_check()
1490 * during ep_insert().
1491 */
1492 if (list_empty(&epi->ffd.file->f_tfile_llink))
1493 list_add(&epi->ffd.file->f_tfile_llink,
1494 &tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001495 }
1496 }
1497 mutex_unlock(&ep->mtx);
1498
1499 return error;
1500}
1501
1502/**
1503 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1504 * another epoll file (represented by @ep) does not create
1505 * closed loops or too deep chains.
1506 *
1507 * @ep: Pointer to the epoll private data structure.
1508 * @file: Pointer to the epoll file to be checked.
1509 *
1510 * Returns: Returns zero if adding the epoll @file inside current epoll
1511 * structure @ep does not violate the constraints, or -1 otherwise.
1512 */
1513static int ep_loop_check(struct eventpoll *ep, struct file *file)
1514{
Jason Baron28d82dc2012-01-12 17:17:43 -08001515 int ret;
1516 struct eventpoll *ep_cur, *ep_next;
1517
1518 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Davide Libenzi22bacca2011-02-25 14:44:12 -08001519 ep_loop_check_proc, file, ep, current);
Jason Baron28d82dc2012-01-12 17:17:43 -08001520 /* clear visited list */
1521 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1522 visited_list_link) {
1523 ep_cur->visited = 0;
1524 list_del(&ep_cur->visited_list_link);
1525 }
1526 return ret;
1527}
1528
1529static void clear_tfile_check_list(void)
1530{
1531 struct file *file;
1532
1533 /* first clear the tfile_check_list */
1534 while (!list_empty(&tfile_check_list)) {
1535 file = list_first_entry(&tfile_check_list, struct file,
1536 f_tfile_llink);
1537 list_del_init(&file->f_tfile_llink);
1538 }
1539 INIT_LIST_HEAD(&tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001540}
1541
Davide Libenzi7699acd2007-05-10 22:23:23 -07001542/*
Andrew Morton523723b2008-08-12 15:09:01 -07001543 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001544 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001545SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001546{
Jason Baron28d82dc2012-01-12 17:17:43 -08001547 int error, fd;
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001548 struct eventpoll *ep = NULL;
Jason Baron28d82dc2012-01-12 17:17:43 -08001549 struct file *file;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001550
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001551 /* Check the EPOLL_* constant for consistency. */
1552 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1553
Davide Libenzi296e2362009-03-31 15:24:11 -07001554 if (flags & ~EPOLL_CLOEXEC)
1555 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001556 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001557 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001558 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001559 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001560 if (error < 0)
1561 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001562 /*
1563 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001564 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001565 */
Jason Baron28d82dc2012-01-12 17:17:43 -08001566 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1567 if (fd < 0) {
1568 error = fd;
1569 goto out_free_ep;
1570 }
1571 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001572 O_RDWR | (flags & O_CLOEXEC));
Jason Baron28d82dc2012-01-12 17:17:43 -08001573 if (IS_ERR(file)) {
1574 error = PTR_ERR(file);
1575 goto out_free_fd;
1576 }
1577 fd_install(fd, file);
1578 ep->file = file;
1579 return fd;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001580
Jason Baron28d82dc2012-01-12 17:17:43 -08001581out_free_fd:
1582 put_unused_fd(fd);
1583out_free_ep:
1584 ep_free(ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001585 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001586}
1587
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001588SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001589{
Davide Libenzibfe38912009-05-12 13:19:44 -07001590 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001591 return -EINVAL;
1592
1593 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001594}
1595
Davide Libenzi7699acd2007-05-10 22:23:23 -07001596/*
1597 * The following function implements the controller interface for
1598 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001599 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001600 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001601SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1602 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001603{
1604 int error;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001605 int did_lock_epmutex = 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001606 struct file *file, *tfile;
1607 struct eventpoll *ep;
1608 struct epitem *epi;
1609 struct epoll_event epds;
1610
Davide Libenzi7699acd2007-05-10 22:23:23 -07001611 error = -EFAULT;
1612 if (ep_op_has_event(op) &&
1613 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1614 goto error_return;
1615
1616 /* Get the "struct file *" for the eventpoll file */
1617 error = -EBADF;
1618 file = fget(epfd);
1619 if (!file)
1620 goto error_return;
1621
1622 /* Get the "struct file *" for the target file */
1623 tfile = fget(fd);
1624 if (!tfile)
1625 goto error_fput;
1626
1627 /* The target file descriptor must support poll */
1628 error = -EPERM;
1629 if (!tfile->f_op || !tfile->f_op->poll)
1630 goto error_tgt_fput;
1631
1632 /*
1633 * We have to check that the file structure underneath the file descriptor
1634 * the user passed to us _is_ an eventpoll file. And also we do not permit
1635 * adding an epoll file descriptor inside itself.
1636 */
1637 error = -EINVAL;
1638 if (file == tfile || !is_file_epoll(file))
1639 goto error_tgt_fput;
1640
1641 /*
1642 * At this point it is safe to assume that the "private_data" contains
1643 * our own data structure.
1644 */
1645 ep = file->private_data;
1646
Davide Libenzi22bacca2011-02-25 14:44:12 -08001647 /*
1648 * When we insert an epoll file descriptor, inside another epoll file
1649 * descriptor, there is the change of creating closed loops, which are
Jason Baron28d82dc2012-01-12 17:17:43 -08001650 * better be handled here, than in more critical paths. While we are
1651 * checking for loops we also determine the list of files reachable
1652 * and hang them on the tfile_check_list, so we can check that we
1653 * haven't created too many possible wakeup paths.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001654 *
Jason Baron28d82dc2012-01-12 17:17:43 -08001655 * We need to hold the epmutex across both ep_insert and ep_remove
1656 * b/c we want to make sure we are looking at a coherent view of
1657 * epoll network.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001658 */
Jason Baron28d82dc2012-01-12 17:17:43 -08001659 if (op == EPOLL_CTL_ADD || op == EPOLL_CTL_DEL) {
Davide Libenzi22bacca2011-02-25 14:44:12 -08001660 mutex_lock(&epmutex);
1661 did_lock_epmutex = 1;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001662 }
Jason Baron28d82dc2012-01-12 17:17:43 -08001663 if (op == EPOLL_CTL_ADD) {
1664 if (is_file_epoll(tfile)) {
1665 error = -ELOOP;
Jason Baron13d51802012-04-25 16:01:47 -07001666 if (ep_loop_check(ep, tfile) != 0) {
1667 clear_tfile_check_list();
Jason Baron28d82dc2012-01-12 17:17:43 -08001668 goto error_tgt_fput;
Jason Baron13d51802012-04-25 16:01:47 -07001669 }
Jason Baron28d82dc2012-01-12 17:17:43 -08001670 } else
1671 list_add(&tfile->f_tfile_llink, &tfile_check_list);
1672 }
Davide Libenzi22bacca2011-02-25 14:44:12 -08001673
Nelson Elhaged8805e62011-10-31 17:13:14 -07001674 mutex_lock_nested(&ep->mtx, 0);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001675
Davide Libenzi67647d02007-05-15 01:40:52 -07001676 /*
1677 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1678 * above, we can be sure to be able to use the item looked up by
1679 * ep_find() till we release the mutex.
1680 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001681 epi = ep_find(ep, tfile, fd);
1682
1683 error = -EINVAL;
1684 switch (op) {
1685 case EPOLL_CTL_ADD:
1686 if (!epi) {
1687 epds.events |= POLLERR | POLLHUP;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001688 error = ep_insert(ep, &epds, tfile, fd);
1689 } else
1690 error = -EEXIST;
Jason Baron28d82dc2012-01-12 17:17:43 -08001691 clear_tfile_check_list();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001692 break;
1693 case EPOLL_CTL_DEL:
1694 if (epi)
1695 error = ep_remove(ep, epi);
1696 else
1697 error = -ENOENT;
1698 break;
1699 case EPOLL_CTL_MOD:
1700 if (epi) {
1701 epds.events |= POLLERR | POLLHUP;
1702 error = ep_modify(ep, epi, &epds);
1703 } else
1704 error = -ENOENT;
1705 break;
1706 }
Davide Libenzid47de162007-05-15 01:40:41 -07001707 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001708
1709error_tgt_fput:
Jason Baron28d82dc2012-01-12 17:17:43 -08001710 if (did_lock_epmutex)
Davide Libenzi22bacca2011-02-25 14:44:12 -08001711 mutex_unlock(&epmutex);
1712
Davide Libenzi7699acd2007-05-10 22:23:23 -07001713 fput(tfile);
1714error_fput:
1715 fput(file);
1716error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001717
1718 return error;
1719}
1720
1721/*
1722 * Implement the event wait interface for the eventpoll file. It is the kernel
1723 * part of the user space epoll_wait(2).
1724 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001725SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1726 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001727{
1728 int error;
1729 struct file *file;
1730 struct eventpoll *ep;
1731
Davide Libenzi7699acd2007-05-10 22:23:23 -07001732 /* The maximum number of event must be greater than zero */
1733 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1734 return -EINVAL;
1735
1736 /* Verify that the area passed by the user is writeable */
1737 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1738 error = -EFAULT;
1739 goto error_return;
1740 }
1741
1742 /* Get the "struct file *" for the eventpoll file */
1743 error = -EBADF;
1744 file = fget(epfd);
1745 if (!file)
1746 goto error_return;
1747
1748 /*
1749 * We have to check that the file structure underneath the fd
1750 * the user passed to us _is_ an eventpoll file.
1751 */
1752 error = -EINVAL;
1753 if (!is_file_epoll(file))
1754 goto error_fput;
1755
1756 /*
1757 * At this point it is safe to assume that the "private_data" contains
1758 * our own data structure.
1759 */
1760 ep = file->private_data;
1761
1762 /* Time to fish for events ... */
1763 error = ep_poll(ep, events, maxevents, timeout);
1764
1765error_fput:
1766 fput(file);
1767error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001768
1769 return error;
1770}
1771
Roland McGrathf3de2722008-04-30 00:53:09 -07001772#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001773
1774/*
1775 * Implement the event wait interface for the eventpoll file. It is the kernel
1776 * part of the user space epoll_pwait(2).
1777 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001778SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1779 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1780 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001781{
1782 int error;
1783 sigset_t ksigmask, sigsaved;
1784
1785 /*
1786 * If the caller wants a certain signal mask to be set during the wait,
1787 * we apply it here.
1788 */
1789 if (sigmask) {
1790 if (sigsetsize != sizeof(sigset_t))
1791 return -EINVAL;
1792 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1793 return -EFAULT;
1794 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1795 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1796 }
1797
1798 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1799
1800 /*
1801 * If we changed the signal mask, we need to restore the original one.
1802 * In case we've got a signal while waiting, we do not restore the
1803 * signal mask yet, and we allow do_signal() to deliver the signal on
1804 * the way back to userspace, before the signal mask is restored.
1805 */
1806 if (sigmask) {
1807 if (error == -EINTR) {
1808 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001809 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001810 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001811 } else
1812 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1813 }
1814
1815 return error;
1816}
1817
Roland McGrathf3de2722008-04-30 00:53:09 -07001818#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820static int __init eventpoll_init(void)
1821{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001822 struct sysinfo si;
1823
1824 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001825 /*
1826 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1827 */
1828 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001829 EP_ITEM_COST;
Robin Holt52bd19f72011-01-12 17:00:01 -08001830 BUG_ON(max_user_watches < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
Davide Libenzi22bacca2011-02-25 14:44:12 -08001832 /*
1833 * Initialize the structure used to perform epoll file descriptor
1834 * inclusion loops checks.
1835 */
1836 ep_nested_calls_init(&poll_loop_ncalls);
1837
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001839 ep_nested_calls_init(&poll_safewake_ncalls);
1840
1841 /* Initialize the structure used to perform file's f_op->poll() calls */
1842 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
1844 /* Allocates slab cache used to allocate "struct epitem" items */
1845 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001846 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1849 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001850 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853}
Davide Libenzicea69242007-05-10 22:23:22 -07001854fs_initcall(eventpoll_init);