blob: 8cf07242067de739fb4c39a9bf607c855a785101 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Davide Libenzi5071f972009-03-31 15:24:10 -07002 * fs/eventpoll.c (Efficient event retrieval implementation)
3 * Copyright (C) 2001,...,2009 Davide Libenzi
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * Davide Libenzi <davidel@xmailserver.org>
11 *
12 */
13
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/init.h>
15#include <linux/kernel.h>
16#include <linux/sched.h>
17#include <linux/fs.h>
18#include <linux/file.h>
19#include <linux/signal.h>
20#include <linux/errno.h>
21#include <linux/mm.h>
22#include <linux/slab.h>
23#include <linux/poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/string.h>
25#include <linux/list.h>
26#include <linux/hash.h>
27#include <linux/spinlock.h>
28#include <linux/syscalls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/rbtree.h>
30#include <linux/wait.h>
31#include <linux/eventpoll.h>
32#include <linux/mount.h>
33#include <linux/bitops.h>
Arjan van de Ven144efe32006-03-23 03:00:32 -080034#include <linux/mutex.h>
Davide Libenzida66f7c2007-05-10 22:23:21 -070035#include <linux/anon_inodes.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37#include <asm/system.h>
38#include <asm/io.h>
39#include <asm/mman.h>
40#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * LOCKING:
44 * There are three level of locking required by epoll :
45 *
Arjan van de Ven144efe32006-03-23 03:00:32 -080046 * 1) epmutex (mutex)
Davide Libenzic7ea7632007-05-15 01:40:47 -070047 * 2) ep->mtx (mutex)
48 * 3) ep->lock (spinlock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 *
50 * The acquire order is the one listed above, from 1 to 3.
51 * We need a spinlock (ep->lock) because we manipulate objects
52 * from inside the poll callback, that might be triggered from
53 * a wake_up() that in turn might be called from IRQ context.
54 * So we can't sleep inside the poll callback and hence we need
55 * a spinlock. During the event transfer loop (from kernel to
56 * user space) we could end up sleeping due a copy_to_user(), so
57 * we need a lock that will allow us to sleep. This lock is a
Davide Libenzid47de162007-05-15 01:40:41 -070058 * mutex (ep->mtx). It is acquired during the event transfer loop,
59 * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
60 * Then we also need a global mutex to serialize eventpoll_release_file()
61 * and ep_free().
62 * This mutex is acquired by ep_free() during the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 * cleanup path and it is also acquired by eventpoll_release_file()
64 * if a file has been pushed inside an epoll set and it is then
65 * close()d without a previous call toepoll_ctl(EPOLL_CTL_DEL).
Davide Libenzid47de162007-05-15 01:40:41 -070066 * It is possible to drop the "ep->mtx" and to use the global
67 * mutex "epmutex" (together with "ep->lock") to have it working,
68 * but having "ep->mtx" will make the interface more scalable.
Arjan van de Ven144efe32006-03-23 03:00:32 -080069 * Events that require holding "epmutex" are very rare, while for
Davide Libenzid47de162007-05-15 01:40:41 -070070 * normal operations the epoll private "ep->mtx" will guarantee
71 * a better scalability.
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 */
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074/* Epoll private bits inside the event mask */
75#define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET)
76
Davide Libenzi5071f972009-03-31 15:24:10 -070077/* Maximum number of nesting allowed inside epoll sets */
78#define EP_MAX_NESTS 4
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Davide Libenzib6119672006-10-11 01:21:44 -070080#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
81
Davide Libenzid47de162007-05-15 01:40:41 -070082#define EP_UNACTIVE_PTR ((void *) -1L)
83
Davide Libenzi7ef99642008-12-01 13:13:55 -080084#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
85
Linus Torvalds1da177e2005-04-16 15:20:36 -070086struct epoll_filefd {
87 struct file *file;
88 int fd;
89};
90
91/*
Davide Libenzi5071f972009-03-31 15:24:10 -070092 * Structure used to track possible nested calls, for too deep recursions
93 * and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Davide Libenzi5071f972009-03-31 15:24:10 -070095struct nested_call_node {
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 struct list_head llink;
Davide Libenzi5071f972009-03-31 15:24:10 -070097 void *cookie;
Davide Libenzi3fe4a972009-06-17 16:25:58 -070098 void *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099};
100
101/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700102 * This structure is used as collector for nested calls, to check for
103 * maximum recursion dept and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700105struct nested_calls {
106 struct list_head tasks_call_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 spinlock_t lock;
108};
109
110/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 * Each file descriptor added to the eventpoll interface will
Davide Libenzi6192bd52007-05-08 00:25:41 -0700112 * have an entry of this type linked to the "rbr" RB tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 */
114struct epitem {
Davide Libenzi67647d02007-05-15 01:40:52 -0700115 /* RB tree node used to link this structure to the eventpoll RB tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 struct rb_node rbn;
117
118 /* List header used to link this structure to the eventpoll ready list */
119 struct list_head rdllink;
120
Davide Libenzic7ea7632007-05-15 01:40:47 -0700121 /*
122 * Works together "struct eventpoll"->ovflist in keeping the
123 * single linked chain of items.
124 */
125 struct epitem *next;
126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 /* The file descriptor information this item refers to */
128 struct epoll_filefd ffd;
129
130 /* Number of active wait queue attached to poll operations */
131 int nwait;
132
133 /* List containing poll wait queues */
134 struct list_head pwqlist;
135
136 /* The "container" of this item */
137 struct eventpoll *ep;
138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 /* List header used to link this item to the "struct file" items list */
140 struct list_head fllink;
Davide Libenzid47de162007-05-15 01:40:41 -0700141
Davide Libenzic7ea7632007-05-15 01:40:47 -0700142 /* The structure that describe the interested events and the source fd */
143 struct epoll_event event;
Davide Libenzid47de162007-05-15 01:40:41 -0700144};
145
146/*
147 * This structure is stored inside the "private_data" member of the file
148 * structure and rapresent the main data sructure for the eventpoll
149 * interface.
150 */
151struct eventpoll {
152 /* Protect the this structure access */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700153 spinlock_t lock;
Davide Libenzid47de162007-05-15 01:40:41 -0700154
155 /*
156 * This mutex is used to ensure that files are not removed
157 * while epoll is using them. This is held during the event
158 * collection loop, the file cleanup path, the epoll file exit
159 * code and the ctl operations.
160 */
161 struct mutex mtx;
162
163 /* Wait queue used by sys_epoll_wait() */
164 wait_queue_head_t wq;
165
166 /* Wait queue used by file->poll() */
167 wait_queue_head_t poll_wait;
168
169 /* List of ready file descriptors */
170 struct list_head rdllist;
171
Davide Libenzi67647d02007-05-15 01:40:52 -0700172 /* RB tree root used to store monitored fd structs */
Davide Libenzid47de162007-05-15 01:40:41 -0700173 struct rb_root rbr;
174
175 /*
176 * This is a single linked list that chains all the "struct epitem" that
177 * happened while transfering ready events to userspace w/out
178 * holding ->lock.
179 */
180 struct epitem *ovflist;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800181
182 /* The user that created the eventpoll descriptor */
183 struct user_struct *user;
Davide Libenzid47de162007-05-15 01:40:41 -0700184};
185
186/* Wait structure used by the poll hooks */
187struct eppoll_entry {
188 /* List header used to link this structure to the "struct epitem" */
189 struct list_head llink;
190
191 /* The "base" pointer is set to the container "struct epitem" */
Tony Battersby4f0989d2009-03-31 15:24:16 -0700192 struct epitem *base;
Davide Libenzid47de162007-05-15 01:40:41 -0700193
194 /*
195 * Wait queue item that will be linked to the target file wait
196 * queue head.
197 */
198 wait_queue_t wait;
199
200 /* The wait queue head that linked the "wait" wait queue item */
201 wait_queue_head_t *whead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202};
203
204/* Wrapper struct used by poll queueing */
205struct ep_pqueue {
206 poll_table pt;
207 struct epitem *epi;
208};
209
Davide Libenzi5071f972009-03-31 15:24:10 -0700210/* Used by the ep_send_events() function as callback private data */
211struct ep_send_events_data {
212 int maxevents;
213 struct epoll_event __user *events;
214};
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216/*
Davide Libenzi7ef99642008-12-01 13:13:55 -0800217 * Configuration options available inside /proc/sys/fs/epoll/
218 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800219/* Maximum number of epoll watched descriptors, per user */
220static int max_user_watches __read_mostly;
221
222/*
Davide Libenzid47de162007-05-15 01:40:41 -0700223 * This mutex is used to serialize ep_free() and eventpoll_release_file().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800225static DEFINE_MUTEX(epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Davide Libenzi5071f972009-03-31 15:24:10 -0700227/* Used for safe wake up implementation */
228static struct nested_calls poll_safewake_ncalls;
229
230/* Used to call file's f_op->poll() under the nested calls boundaries */
231static struct nested_calls poll_readywalk_ncalls;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233/* Slab cache used to allocate "struct epitem" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800234static struct kmem_cache *epi_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
236/* Slab cache used to allocate "struct eppoll_entry" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800237static struct kmem_cache *pwq_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Davide Libenzi7ef99642008-12-01 13:13:55 -0800239#ifdef CONFIG_SYSCTL
240
241#include <linux/sysctl.h>
242
243static int zero;
244
245ctl_table epoll_table[] = {
246 {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800247 .procname = "max_user_watches",
248 .data = &max_user_watches,
249 .maxlen = sizeof(int),
250 .mode = 0644,
Eric W. Biederman6d456112009-11-16 03:11:48 -0800251 .proc_handler = proc_dointvec_minmax,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800252 .extra1 = &zero,
253 },
Eric W. Biedermanab092032009-11-05 14:25:10 -0800254 { }
Davide Libenzi7ef99642008-12-01 13:13:55 -0800255};
256#endif /* CONFIG_SYSCTL */
257
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700258
Davide Libenzi67647d02007-05-15 01:40:52 -0700259/* Setup the structure that is used as key for the RB tree */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700260static inline void ep_set_ffd(struct epoll_filefd *ffd,
261 struct file *file, int fd)
262{
263 ffd->file = file;
264 ffd->fd = fd;
265}
266
Davide Libenzi67647d02007-05-15 01:40:52 -0700267/* Compare RB tree keys */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700268static inline int ep_cmp_ffd(struct epoll_filefd *p1,
269 struct epoll_filefd *p2)
270{
271 return (p1->file > p2->file ? +1:
272 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
273}
274
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700275/* Tells us if the item is currently linked */
276static inline int ep_is_linked(struct list_head *p)
277{
278 return !list_empty(p);
279}
280
281/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700282static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700283{
284 return container_of(p, struct eppoll_entry, wait)->base;
285}
286
287/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700288static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700289{
290 return container_of(p, struct ep_pqueue, pt)->epi;
291}
292
293/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700294static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700295{
296 return op != EPOLL_CTL_DEL;
297}
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700300static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Davide Libenzi5071f972009-03-31 15:24:10 -0700302 INIT_LIST_HEAD(&ncalls->tasks_call_list);
303 spin_lock_init(&ncalls->lock);
304}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Davide Libenzi5071f972009-03-31 15:24:10 -0700306/**
307 * ep_call_nested - Perform a bound (possibly) nested call, by checking
308 * that the recursion limit is not exceeded, and that
309 * the same nested call (by the meaning of same cookie) is
310 * no re-entered.
311 *
312 * @ncalls: Pointer to the nested_calls structure to be used for this call.
313 * @max_nests: Maximum number of allowed nesting calls.
314 * @nproc: Nested call core function pointer.
315 * @priv: Opaque data to be passed to the @nproc callback.
316 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700317 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700318 *
319 * Returns: Returns the code returned by the @nproc callback, or -1 if
320 * the maximum recursion limit has been exceeded.
321 */
322static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
323 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700324 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700325{
326 int error, call_nests = 0;
327 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700328 struct list_head *lsthead = &ncalls->tasks_call_list;
329 struct nested_call_node *tncur;
330 struct nested_call_node tnode;
331
332 spin_lock_irqsave(&ncalls->lock, flags);
333
334 /*
335 * Try to see if the current task is already inside this wakeup call.
336 * We use a list here, since the population inside this set is always
337 * very much limited.
338 */
339 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700340 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700341 (tncur->cookie == cookie || ++call_nests > max_nests)) {
342 /*
343 * Ops ... loop detected or maximum nest level reached.
344 * We abort this wake by breaking the cycle itself.
345 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700346 error = -1;
347 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700348 }
349 }
350
351 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700352 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700353 tnode.cookie = cookie;
354 list_add(&tnode.llink, lsthead);
355
356 spin_unlock_irqrestore(&ncalls->lock, flags);
357
358 /* Call the nested function */
359 error = (*nproc)(priv, cookie, call_nests);
360
361 /* Remove the current task from the list */
362 spin_lock_irqsave(&ncalls->lock, flags);
363 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700364out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700365 spin_unlock_irqrestore(&ncalls->lock, flags);
366
367 return error;
368}
369
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700370#ifdef CONFIG_DEBUG_LOCK_ALLOC
371static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
372 unsigned long events, int subclass)
373{
374 unsigned long flags;
375
376 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
377 wake_up_locked_poll(wqueue, events);
378 spin_unlock_irqrestore(&wqueue->lock, flags);
379}
380#else
381static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
382 unsigned long events, int subclass)
383{
384 wake_up_poll(wqueue, events);
385}
386#endif
387
Davide Libenzi5071f972009-03-31 15:24:10 -0700388static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
389{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700390 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
391 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700392 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393}
394
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395/*
396 * Perform a safe wake up of the poll wait list. The problem is that
397 * with the new callback'd wake up system, it is possible that the
398 * poll callback is reentered from inside the call to wake_up() done
399 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700400 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 * and we cannot reenter the same wait queue head at all. This will
402 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700403 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700405static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700407 int this_cpu = get_cpu();
408
Davide Libenzi5071f972009-03-31 15:24:10 -0700409 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700410 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
411
412 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413}
414
Davide Libenzi7699acd2007-05-10 22:23:23 -0700415/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700416 * This function unregisters poll callbacks from the associated file
417 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
418 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700419 */
420static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
421{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700422 struct list_head *lsthead = &epi->pwqlist;
423 struct eppoll_entry *pwq;
424
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700425 while (!list_empty(lsthead)) {
426 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700427
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700428 list_del(&pwq->llink);
429 remove_wait_queue(pwq->whead, &pwq->wait);
430 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700431 }
432}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
Davide Libenzi5071f972009-03-31 15:24:10 -0700434/**
435 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
436 * the scan code, to call f_op->poll(). Also allows for
437 * O(NumReady) performance.
438 *
439 * @ep: Pointer to the epoll private data structure.
440 * @sproc: Pointer to the scan callback.
441 * @priv: Private opaque data passed to the @sproc callback.
442 *
443 * Returns: The same integer error code returned by the @sproc callback.
444 */
445static int ep_scan_ready_list(struct eventpoll *ep,
446 int (*sproc)(struct eventpoll *,
447 struct list_head *, void *),
448 void *priv)
449{
450 int error, pwake = 0;
451 unsigned long flags;
452 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700453 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700454
455 /*
456 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700457 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700458 */
459 mutex_lock(&ep->mtx);
460
461 /*
462 * Steal the ready list, and re-init the original one to the
463 * empty list. Also, set ep->ovflist to NULL so that events
464 * happening while looping w/out locks, are not lost. We cannot
465 * have the poll callback to queue directly on ep->rdllist,
466 * because we want the "sproc" callback to be able to do it
467 * in a lockless way.
468 */
469 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700470 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700471 ep->ovflist = NULL;
472 spin_unlock_irqrestore(&ep->lock, flags);
473
474 /*
475 * Now call the callback function.
476 */
477 error = (*sproc)(ep, &txlist, priv);
478
479 spin_lock_irqsave(&ep->lock, flags);
480 /*
481 * During the time we spent inside the "sproc" callback, some
482 * other events might have been queued by the poll callback.
483 * We re-insert them inside the main ready-list here.
484 */
485 for (nepi = ep->ovflist; (epi = nepi) != NULL;
486 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
487 /*
488 * We need to check if the item is already in the list.
489 * During the "sproc" callback execution time, items are
490 * queued into ->ovflist but the "txlist" might already
491 * contain them, and the list_splice() below takes care of them.
492 */
493 if (!ep_is_linked(&epi->rdllink))
494 list_add_tail(&epi->rdllink, &ep->rdllist);
495 }
496 /*
497 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
498 * releasing the lock, events will be queued in the normal way inside
499 * ep->rdllist.
500 */
501 ep->ovflist = EP_UNACTIVE_PTR;
502
503 /*
504 * Quickly re-inject items left on "txlist".
505 */
506 list_splice(&txlist, &ep->rdllist);
507
508 if (!list_empty(&ep->rdllist)) {
509 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700510 * Wake up (if active) both the eventpoll wait list and
511 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700512 */
513 if (waitqueue_active(&ep->wq))
514 wake_up_locked(&ep->wq);
515 if (waitqueue_active(&ep->poll_wait))
516 pwake++;
517 }
518 spin_unlock_irqrestore(&ep->lock, flags);
519
520 mutex_unlock(&ep->mtx);
521
522 /* We have to call this outside the lock */
523 if (pwake)
524 ep_poll_safewake(&ep->poll_wait);
525
526 return error;
527}
528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700530 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700531 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700532 */
533static int ep_remove(struct eventpoll *ep, struct epitem *epi)
534{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700535 unsigned long flags;
536 struct file *file = epi->ffd.file;
537
538 /*
539 * Removes poll wait queue hooks. We _have_ to do this without holding
540 * the "ep->lock" otherwise a deadlock might occur. This because of the
541 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
542 * queue head lock when unregistering the wait queue. The wakeup callback
543 * will run by holding the wait queue head lock and will call our callback
544 * that will try to get "ep->lock".
545 */
546 ep_unregister_pollwait(ep, epi);
547
548 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700549 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700550 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700551 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700552 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Davide Libenzicdac75e2008-04-29 00:58:34 -0700554 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Davide Libenzic7ea7632007-05-15 01:40:47 -0700556 spin_lock_irqsave(&ep->lock, flags);
557 if (ep_is_linked(&epi->rdllink))
558 list_del_init(&epi->rdllink);
559 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Davide Libenzi7699acd2007-05-10 22:23:23 -0700561 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700562 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Davide Libenzi7ef99642008-12-01 13:13:55 -0800564 atomic_dec(&ep->user->epoll_watches);
565
Davide Libenzic7ea7632007-05-15 01:40:47 -0700566 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569static void ep_free(struct eventpoll *ep)
570{
571 struct rb_node *rbp;
572 struct epitem *epi;
573
574 /* We need to release all tasks waiting for these file */
575 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700576 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 /*
579 * We need to lock this because we could be hit by
580 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700581 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * is on the way to be removed and no one has references to it
583 * anymore. The only hit might come from eventpoll_release_file() but
Arjan van de Ven144efe32006-03-23 03:00:32 -0800584 * holding "epmutex" is sufficent here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800586 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 /*
589 * Walks through the whole tree by unregistering poll callbacks.
590 */
591 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
592 epi = rb_entry(rbp, struct epitem, rbn);
593
594 ep_unregister_pollwait(ep, epi);
595 }
596
597 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700598 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700600 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 * us during this operation. So we can avoid the lock on "ep->lock".
602 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700603 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 epi = rb_entry(rbp, struct epitem, rbn);
605 ep_remove(ep, epi);
606 }
607
Arjan van de Ven144efe32006-03-23 03:00:32 -0800608 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700609 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800610 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700611 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
Davide Libenzi7699acd2007-05-10 22:23:23 -0700614static int ep_eventpoll_release(struct inode *inode, struct file *file)
615{
616 struct eventpoll *ep = file->private_data;
617
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700618 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700619 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700620
Davide Libenzi7699acd2007-05-10 22:23:23 -0700621 return 0;
622}
623
Davide Libenzi296e2362009-03-31 15:24:11 -0700624static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
625 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700626{
627 struct epitem *epi, *tmp;
628
629 list_for_each_entry_safe(epi, tmp, head, rdllink) {
630 if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
631 epi->event.events)
632 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700633 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700634 /*
635 * Item has been dropped into the ready list by the poll
636 * callback, but it's not actually ready, as far as
637 * caller requested events goes. We can remove it here.
638 */
639 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700640 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700641 }
642
643 return 0;
644}
645
646static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
647{
648 return ep_scan_ready_list(priv, ep_read_events_proc, NULL);
649}
650
Davide Libenzi7699acd2007-05-10 22:23:23 -0700651static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
652{
Davide Libenzi5071f972009-03-31 15:24:10 -0700653 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700654 struct eventpoll *ep = file->private_data;
655
656 /* Insert inside our poll wait queue */
657 poll_wait(file, &ep->poll_wait, wait);
658
Davide Libenzi5071f972009-03-31 15:24:10 -0700659 /*
660 * Proceed to find out if wanted events are really available inside
661 * the ready list. This need to be done under ep_call_nested()
662 * supervision, since the call to f_op->poll() done on listed files
663 * could re-enter here.
664 */
665 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700666 ep_poll_readyevents_proc, ep, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700667
Davide Libenzi296e2362009-03-31 15:24:11 -0700668 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700669}
670
671/* File callbacks that implement the eventpoll file behaviour */
672static const struct file_operations eventpoll_fops = {
673 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200674 .poll = ep_eventpoll_poll,
675 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700676};
677
678/* Fast test to see if the file is an evenpoll file */
679static inline int is_file_epoll(struct file *f)
680{
681 return f->f_op == &eventpoll_fops;
682}
683
684/*
685 * This is called from eventpoll_release() to unlink files from the eventpoll
686 * interface. We need to have this facility to cleanup correctly files that are
687 * closed without being removed from the eventpoll interface.
688 */
689void eventpoll_release_file(struct file *file)
690{
691 struct list_head *lsthead = &file->f_ep_links;
692 struct eventpoll *ep;
693 struct epitem *epi;
694
695 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700696 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700697 * necessary. It is not necessary because we're in the "struct file"
698 * cleanup path, and this means that noone is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700699 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700700 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700701 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700702 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700703 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700704 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700705 *
706 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700707 */
708 mutex_lock(&epmutex);
709
710 while (!list_empty(lsthead)) {
711 epi = list_first_entry(lsthead, struct epitem, fllink);
712
713 ep = epi->ep;
714 list_del_init(&epi->fllink);
Davide Libenzid47de162007-05-15 01:40:41 -0700715 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700716 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700717 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700718 }
719
720 mutex_unlock(&epmutex);
721}
722
723static int ep_alloc(struct eventpoll **pep)
724{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800725 int error;
726 struct user_struct *user;
727 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700728
Davide Libenzi7ef99642008-12-01 13:13:55 -0800729 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800730 error = -ENOMEM;
731 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
732 if (unlikely(!ep))
733 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700734
Davide Libenzic7ea7632007-05-15 01:40:47 -0700735 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700736 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700737 init_waitqueue_head(&ep->wq);
738 init_waitqueue_head(&ep->poll_wait);
739 INIT_LIST_HEAD(&ep->rdllist);
740 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700741 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800742 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700743
744 *pep = ep;
745
Davide Libenzi7699acd2007-05-10 22:23:23 -0700746 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800747
748free_uid:
749 free_uid(user);
750 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700751}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700754 * Search the file inside the eventpoll tree. The RB tree operations
755 * are protected by the "mtx" mutex, and ep_find() must be called with
756 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 */
758static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
759{
760 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 struct rb_node *rbp;
762 struct epitem *epi, *epir = NULL;
763 struct epoll_filefd ffd;
764
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700765 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 for (rbp = ep->rbr.rb_node; rbp; ) {
767 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700768 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 if (kcmp > 0)
770 rbp = rbp->rb_right;
771 else if (kcmp < 0)
772 rbp = rbp->rb_left;
773 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 epir = epi;
775 break;
776 }
777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return epir;
780}
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700783 * This is the callback that is passed to the wait queue wakeup
784 * machanism. It is called by the stored file descriptors when they
785 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700787static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700789 int pwake = 0;
790 unsigned long flags;
791 struct epitem *epi = ep_item_from_wait(wait);
792 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Davide Libenzic7ea7632007-05-15 01:40:47 -0700794 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700795
796 /*
797 * If the event mask does not contain any poll(2) event, we consider the
798 * descriptor to be disabled. This condition is likely the effect of the
799 * EPOLLONESHOT bit that disables the descriptor when an event is received,
800 * until the next EPOLL_CTL_MOD will be issued.
801 */
802 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700803 goto out_unlock;
804
805 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700806 * Check the events coming with the callback. At this stage, not
807 * every device reports the events in the "key" parameter of the
808 * callback. We need to be able to handle both cases here, hence the
809 * test for "key" != NULL before the event match test.
810 */
811 if (key && !((unsigned long) key & epi->event.events))
812 goto out_unlock;
813
814 /*
Davide Libenzid47de162007-05-15 01:40:41 -0700815 * If we are trasfering events to userspace, we can hold no locks
816 * (because we're accessing user memory, and because of linux f_op->poll()
817 * semantics). All the events that happens during that period of time are
818 * chained in ep->ovflist and requeued later on.
819 */
820 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
821 if (epi->next == EP_UNACTIVE_PTR) {
822 epi->next = ep->ovflist;
823 ep->ovflist = epi;
824 }
825 goto out_unlock;
826 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700827
828 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700829 if (!ep_is_linked(&epi->rdllink))
830 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700831
Davide Libenzi7699acd2007-05-10 22:23:23 -0700832 /*
833 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
834 * wait list.
835 */
836 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400837 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700838 if (waitqueue_active(&ep->poll_wait))
839 pwake++;
840
Davide Libenzid47de162007-05-15 01:40:41 -0700841out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700842 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700843
844 /* We have to call this outside the lock */
845 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700846 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700847
848 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849}
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851/*
852 * This is the callback that is used to add our wait queue to the
853 * target file wakeup lists.
854 */
855static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
856 poll_table *pt)
857{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700858 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 struct eppoll_entry *pwq;
860
Christoph Lametere94b1762006-12-06 20:33:17 -0800861 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
863 pwq->whead = whead;
864 pwq->base = epi;
865 add_wait_queue(whead, &pwq->wait);
866 list_add_tail(&pwq->llink, &epi->pwqlist);
867 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700868 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 /* We have to signal that an error occurred */
870 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872}
873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
875{
876 int kcmp;
877 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
878 struct epitem *epic;
879
880 while (*p) {
881 parent = *p;
882 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700883 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (kcmp > 0)
885 p = &parent->rb_right;
886 else
887 p = &parent->rb_left;
888 }
889 rb_link_node(&epi->rbn, parent, p);
890 rb_insert_color(&epi->rbn, &ep->rbr);
891}
892
Davide Libenzic7ea7632007-05-15 01:40:47 -0700893/*
894 * Must be called with "mtx" held.
895 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
897 struct file *tfile, int fd)
898{
899 int error, revents, pwake = 0;
900 unsigned long flags;
901 struct epitem *epi;
902 struct ep_pqueue epq;
903
Davide Libenzi7ef99642008-12-01 13:13:55 -0800904 if (unlikely(atomic_read(&ep->user->epoll_watches) >=
905 max_user_watches))
906 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -0800907 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -0800908 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
910 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 INIT_LIST_HEAD(&epi->rdllink);
912 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 INIT_LIST_HEAD(&epi->pwqlist);
914 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700915 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -0700918 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 /* Initialize the poll table using the queue callback */
921 epq.epi = epi;
922 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
923
924 /*
925 * Attach the item to the poll hooks and get current event bits.
926 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -0700927 * been increased by the caller of this function. Note that after
928 * this operation completes, the poll callback can start hitting
929 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 */
931 revents = tfile->f_op->poll(tfile, &epq.pt);
932
933 /*
934 * We have to check if something went wrong during the poll wait queue
935 * install process. Namely an allocation for a wait queue failed due
936 * high memory pressure.
937 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800938 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700940 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
942 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -0700943 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -0700945 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Davide Libenzic7ea7632007-05-15 01:40:47 -0700947 /*
948 * Add the current item to the RB tree. All RB tree operations are
949 * protected by "mtx", and ep_insert() is called with "mtx" held.
950 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 ep_rbtree_insert(ep, epi);
952
Davide Libenzic7ea7632007-05-15 01:40:47 -0700953 /* We have to drop the new item inside our item list to keep track of it */
954 spin_lock_irqsave(&ep->lock, flags);
955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700957 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 list_add_tail(&epi->rdllink, &ep->rdllist);
959
960 /* Notify waiting tasks that events are available */
961 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400962 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 if (waitqueue_active(&ep->poll_wait))
964 pwake++;
965 }
966
Davide Libenzic7ea7632007-05-15 01:40:47 -0700967 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Davide Libenzi7ef99642008-12-01 13:13:55 -0800969 atomic_inc(&ep->user->epoll_watches);
970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 /* We have to call this outside the lock */
972 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700973 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 return 0;
976
Davide Libenzi7699acd2007-05-10 22:23:23 -0700977error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 ep_unregister_pollwait(ep, epi);
979
980 /*
981 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -0700982 * allocated wait queue. Note that we don't care about the ep->ovflist
983 * list, since that is used/cleaned only inside a section bound by "mtx".
984 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700986 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700987 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700988 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -0700989 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700991 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 return error;
994}
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996/*
997 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -0700998 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 */
1000static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1001{
1002 int pwake = 0;
1003 unsigned int revents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
1005 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001006 * Set the new event interest mask before calling f_op->poll();
1007 * otherwise we might miss an event that happens between the
1008 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 */
1010 epi->event.events = event->events;
Tony Battersbye057e152009-03-31 15:24:15 -07001011 epi->event.data = event->data; /* protected by mtx */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
1013 /*
1014 * Get current event bits. We can safely use the file* here because
1015 * its usage count has been increased by the caller of this function.
1016 */
1017 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL);
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001020 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001021 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001023 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001024 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001025 if (!ep_is_linked(&epi->rdllink)) {
1026 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
Davide Libenzic7ea7632007-05-15 01:40:47 -07001028 /* Notify waiting tasks that events are available */
1029 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001030 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001031 if (waitqueue_active(&ep->poll_wait))
1032 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 }
Tony Battersbye057e152009-03-31 15:24:15 -07001034 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
1037 /* We have to call this outside the lock */
1038 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001039 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 return 0;
1042}
1043
Davide Libenzi296e2362009-03-31 15:24:11 -07001044static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1045 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001046{
1047 struct ep_send_events_data *esed = priv;
1048 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001049 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001050 struct epitem *epi;
1051 struct epoll_event __user *uevent;
1052
Davide Libenzi296e2362009-03-31 15:24:11 -07001053 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001054 * We can loop without lock because we are passed a task private list.
1055 * Items cannot vanish during the loop because ep_scan_ready_list() is
1056 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001057 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001058 for (eventcnt = 0, uevent = esed->events;
1059 !list_empty(head) && eventcnt < esed->maxevents;) {
1060 epi = list_first_entry(head, struct epitem, rdllink);
1061
1062 list_del_init(&epi->rdllink);
1063
Davide Libenzi296e2362009-03-31 15:24:11 -07001064 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
1065 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001066
Davide Libenzi296e2362009-03-31 15:24:11 -07001067 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001068 * If the event mask intersect the caller-requested one,
1069 * deliver the event to userspace. Again, ep_scan_ready_list()
1070 * is holding "mtx", so no operations coming from userspace
1071 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001072 */
1073 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001074 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001075 __put_user(epi->event.data, &uevent->data)) {
1076 list_add(&epi->rdllink, head);
Davide Libenzi296e2362009-03-31 15:24:11 -07001077 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001078 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001079 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001080 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001081 if (epi->event.events & EPOLLONESHOT)
1082 epi->event.events &= EP_PRIVATE_BITS;
1083 else if (!(epi->event.events & EPOLLET)) {
1084 /*
1085 * If this file has been added with Level
1086 * Trigger mode, we need to insert back inside
1087 * the ready list, so that the next call to
1088 * epoll_wait() will check again the events
1089 * availability. At this point, noone can insert
1090 * into ep->rdllist besides us. The epoll_ctl()
1091 * callers are locked out by
1092 * ep_scan_ready_list() holding "mtx" and the
1093 * poll callback will queue them in ep->ovflist.
1094 */
1095 list_add_tail(&epi->rdllink, &ep->rdllist);
1096 }
1097 }
1098 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001099
1100 return eventcnt;
1101}
1102
Davide Libenzi296e2362009-03-31 15:24:11 -07001103static int ep_send_events(struct eventpoll *ep,
1104 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Davide Libenzi5071f972009-03-31 15:24:10 -07001106 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001107
Davide Libenzi5071f972009-03-31 15:24:10 -07001108 esed.maxevents = maxevents;
1109 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001110
Davide Libenzi5071f972009-03-31 15:24:10 -07001111 return ep_scan_ready_list(ep, ep_send_events_proc, &esed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112}
1113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1115 int maxevents, long timeout)
1116{
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001117 int res, eavail, timed_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 unsigned long flags;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001119 long slack;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 wait_queue_t wait;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001121 struct timespec end_time;
1122 ktime_t expires, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001124 if (timeout > 0) {
1125 ktime_get_ts(&end_time);
1126 timespec_add_ns(&end_time, (u64)timeout * NSEC_PER_MSEC);
1127 slack = select_estimate_accuracy(&end_time);
1128 to = &expires;
1129 *to = timespec_to_ktime(end_time);
1130 } else if (timeout == 0) {
1131 timed_out = 1;
1132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
1134retry:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001135 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 res = 0;
1138 if (list_empty(&ep->rdllist)) {
1139 /*
1140 * We don't have any available event to return to the caller.
1141 * We need to sleep here, and we will be wake up by
1142 * ep_poll_callback() when events will become available.
1143 */
1144 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001145 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147 for (;;) {
1148 /*
1149 * We don't want to sleep if the ep_poll_callback() sends us
1150 * a wakeup in between. That's why we set the task state
1151 * to TASK_INTERRUPTIBLE before doing the checks.
1152 */
1153 set_current_state(TASK_INTERRUPTIBLE);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001154 if (!list_empty(&ep->rdllist) || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 break;
1156 if (signal_pending(current)) {
1157 res = -EINTR;
1158 break;
1159 }
1160
Davide Libenzic7ea7632007-05-15 01:40:47 -07001161 spin_unlock_irqrestore(&ep->lock, flags);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001162 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1163 timed_out = 1;
1164
Davide Libenzic7ea7632007-05-15 01:40:47 -07001165 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001167 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 set_current_state(TASK_RUNNING);
1170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 /* Is it worth to try to dig for events ? */
Davide Libenzi5071f972009-03-31 15:24:10 -07001172 eavail = !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Davide Libenzic7ea7632007-05-15 01:40:47 -07001174 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
1176 /*
1177 * Try to transfer events to user space. In case we get 0 events and
1178 * there's still timeout left over, we go trying again in search of
1179 * more luck.
1180 */
1181 if (!res && eavail &&
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001182 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 goto retry;
1184
1185 return res;
1186}
1187
Davide Libenzi7699acd2007-05-10 22:23:23 -07001188/*
Andrew Morton523723b2008-08-12 15:09:01 -07001189 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001190 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001191SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001192{
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001193 int error;
1194 struct eventpoll *ep = NULL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001195
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001196 /* Check the EPOLL_* constant for consistency. */
1197 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1198
Davide Libenzi296e2362009-03-31 15:24:11 -07001199 if (flags & ~EPOLL_CLOEXEC)
1200 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001201 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001202 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001203 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001204 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001205 if (error < 0)
1206 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001207 /*
1208 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001209 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001210 */
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001211 error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001212 O_RDWR | (flags & O_CLOEXEC));
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001213 if (error < 0)
Al Viro2030a422008-02-23 06:46:49 -05001214 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001215
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001216 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001217}
1218
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001219SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001220{
Davide Libenzibfe38912009-05-12 13:19:44 -07001221 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001222 return -EINVAL;
1223
1224 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001225}
1226
Davide Libenzi7699acd2007-05-10 22:23:23 -07001227/*
1228 * The following function implements the controller interface for
1229 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001230 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001231 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001232SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1233 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001234{
1235 int error;
1236 struct file *file, *tfile;
1237 struct eventpoll *ep;
1238 struct epitem *epi;
1239 struct epoll_event epds;
1240
Davide Libenzi7699acd2007-05-10 22:23:23 -07001241 error = -EFAULT;
1242 if (ep_op_has_event(op) &&
1243 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1244 goto error_return;
1245
1246 /* Get the "struct file *" for the eventpoll file */
1247 error = -EBADF;
1248 file = fget(epfd);
1249 if (!file)
1250 goto error_return;
1251
1252 /* Get the "struct file *" for the target file */
1253 tfile = fget(fd);
1254 if (!tfile)
1255 goto error_fput;
1256
1257 /* The target file descriptor must support poll */
1258 error = -EPERM;
1259 if (!tfile->f_op || !tfile->f_op->poll)
1260 goto error_tgt_fput;
1261
1262 /*
1263 * We have to check that the file structure underneath the file descriptor
1264 * the user passed to us _is_ an eventpoll file. And also we do not permit
1265 * adding an epoll file descriptor inside itself.
1266 */
1267 error = -EINVAL;
1268 if (file == tfile || !is_file_epoll(file))
1269 goto error_tgt_fput;
1270
1271 /*
1272 * At this point it is safe to assume that the "private_data" contains
1273 * our own data structure.
1274 */
1275 ep = file->private_data;
1276
Davide Libenzid47de162007-05-15 01:40:41 -07001277 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001278
Davide Libenzi67647d02007-05-15 01:40:52 -07001279 /*
1280 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1281 * above, we can be sure to be able to use the item looked up by
1282 * ep_find() till we release the mutex.
1283 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001284 epi = ep_find(ep, tfile, fd);
1285
1286 error = -EINVAL;
1287 switch (op) {
1288 case EPOLL_CTL_ADD:
1289 if (!epi) {
1290 epds.events |= POLLERR | POLLHUP;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001291 error = ep_insert(ep, &epds, tfile, fd);
1292 } else
1293 error = -EEXIST;
1294 break;
1295 case EPOLL_CTL_DEL:
1296 if (epi)
1297 error = ep_remove(ep, epi);
1298 else
1299 error = -ENOENT;
1300 break;
1301 case EPOLL_CTL_MOD:
1302 if (epi) {
1303 epds.events |= POLLERR | POLLHUP;
1304 error = ep_modify(ep, epi, &epds);
1305 } else
1306 error = -ENOENT;
1307 break;
1308 }
Davide Libenzid47de162007-05-15 01:40:41 -07001309 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001310
1311error_tgt_fput:
1312 fput(tfile);
1313error_fput:
1314 fput(file);
1315error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001316
1317 return error;
1318}
1319
1320/*
1321 * Implement the event wait interface for the eventpoll file. It is the kernel
1322 * part of the user space epoll_wait(2).
1323 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001324SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1325 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001326{
1327 int error;
1328 struct file *file;
1329 struct eventpoll *ep;
1330
Davide Libenzi7699acd2007-05-10 22:23:23 -07001331 /* The maximum number of event must be greater than zero */
1332 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1333 return -EINVAL;
1334
1335 /* Verify that the area passed by the user is writeable */
1336 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1337 error = -EFAULT;
1338 goto error_return;
1339 }
1340
1341 /* Get the "struct file *" for the eventpoll file */
1342 error = -EBADF;
1343 file = fget(epfd);
1344 if (!file)
1345 goto error_return;
1346
1347 /*
1348 * We have to check that the file structure underneath the fd
1349 * the user passed to us _is_ an eventpoll file.
1350 */
1351 error = -EINVAL;
1352 if (!is_file_epoll(file))
1353 goto error_fput;
1354
1355 /*
1356 * At this point it is safe to assume that the "private_data" contains
1357 * our own data structure.
1358 */
1359 ep = file->private_data;
1360
1361 /* Time to fish for events ... */
1362 error = ep_poll(ep, events, maxevents, timeout);
1363
1364error_fput:
1365 fput(file);
1366error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001367
1368 return error;
1369}
1370
Roland McGrathf3de2722008-04-30 00:53:09 -07001371#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001372
1373/*
1374 * Implement the event wait interface for the eventpoll file. It is the kernel
1375 * part of the user space epoll_pwait(2).
1376 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001377SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1378 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1379 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001380{
1381 int error;
1382 sigset_t ksigmask, sigsaved;
1383
1384 /*
1385 * If the caller wants a certain signal mask to be set during the wait,
1386 * we apply it here.
1387 */
1388 if (sigmask) {
1389 if (sigsetsize != sizeof(sigset_t))
1390 return -EINVAL;
1391 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1392 return -EFAULT;
1393 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1394 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1395 }
1396
1397 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1398
1399 /*
1400 * If we changed the signal mask, we need to restore the original one.
1401 * In case we've got a signal while waiting, we do not restore the
1402 * signal mask yet, and we allow do_signal() to deliver the signal on
1403 * the way back to userspace, before the signal mask is restored.
1404 */
1405 if (sigmask) {
1406 if (error == -EINTR) {
1407 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001408 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001409 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001410 } else
1411 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1412 }
1413
1414 return error;
1415}
1416
Roland McGrathf3de2722008-04-30 00:53:09 -07001417#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419static int __init eventpoll_init(void)
1420{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001421 struct sysinfo si;
1422
1423 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001424 /*
1425 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1426 */
1427 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001428 EP_ITEM_COST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001431 ep_nested_calls_init(&poll_safewake_ncalls);
1432
1433 /* Initialize the structure used to perform file's f_op->poll() calls */
1434 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 /* Allocates slab cache used to allocate "struct epitem" items */
1437 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001438 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1441 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001442 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445}
Davide Libenzicea69242007-05-10 22:23:22 -07001446fs_initcall(eventpoll_init);