blob: 256bb7bb102a0d2221fe4b9b8d2b533949d069c1 [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 Libenzie3306dd2005-09-27 21:45:33 -070080/* Maximum msec timeout value storeable in a long int */
81#define EP_MAX_MSTIMEO min(1000ULL * MAX_SCHEDULE_TIMEOUT / HZ, (LONG_MAX - 999ULL) / HZ)
82
Davide Libenzib6119672006-10-11 01:21:44 -070083#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
84
Davide Libenzid47de162007-05-15 01:40:41 -070085#define EP_UNACTIVE_PTR ((void *) -1L)
86
Davide Libenzi7ef99642008-12-01 13:13:55 -080087#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089struct epoll_filefd {
90 struct file *file;
91 int fd;
92};
93
94/*
Davide Libenzi5071f972009-03-31 15:24:10 -070095 * Structure used to track possible nested calls, for too deep recursions
96 * and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 */
Davide Libenzi5071f972009-03-31 15:24:10 -070098struct nested_call_node {
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 struct list_head llink;
Davide Libenzi5071f972009-03-31 15:24:10 -0700100 void *cookie;
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700101 void *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102};
103
104/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700105 * This structure is used as collector for nested calls, to check for
106 * maximum recursion dept and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700108struct nested_calls {
109 struct list_head tasks_call_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 spinlock_t lock;
111};
112
113/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * Each file descriptor added to the eventpoll interface will
Davide Libenzi6192bd52007-05-08 00:25:41 -0700115 * have an entry of this type linked to the "rbr" RB tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 */
117struct epitem {
Davide Libenzi67647d02007-05-15 01:40:52 -0700118 /* RB tree node used to link this structure to the eventpoll RB tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 struct rb_node rbn;
120
121 /* List header used to link this structure to the eventpoll ready list */
122 struct list_head rdllink;
123
Davide Libenzic7ea7632007-05-15 01:40:47 -0700124 /*
125 * Works together "struct eventpoll"->ovflist in keeping the
126 * single linked chain of items.
127 */
128 struct epitem *next;
129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 /* The file descriptor information this item refers to */
131 struct epoll_filefd ffd;
132
133 /* Number of active wait queue attached to poll operations */
134 int nwait;
135
136 /* List containing poll wait queues */
137 struct list_head pwqlist;
138
139 /* The "container" of this item */
140 struct eventpoll *ep;
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 /* List header used to link this item to the "struct file" items list */
143 struct list_head fllink;
Davide Libenzid47de162007-05-15 01:40:41 -0700144
Davide Libenzic7ea7632007-05-15 01:40:47 -0700145 /* The structure that describe the interested events and the source fd */
146 struct epoll_event event;
Davide Libenzid47de162007-05-15 01:40:41 -0700147};
148
149/*
150 * This structure is stored inside the "private_data" member of the file
151 * structure and rapresent the main data sructure for the eventpoll
152 * interface.
153 */
154struct eventpoll {
155 /* Protect the this structure access */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700156 spinlock_t lock;
Davide Libenzid47de162007-05-15 01:40:41 -0700157
158 /*
159 * This mutex is used to ensure that files are not removed
160 * while epoll is using them. This is held during the event
161 * collection loop, the file cleanup path, the epoll file exit
162 * code and the ctl operations.
163 */
164 struct mutex mtx;
165
166 /* Wait queue used by sys_epoll_wait() */
167 wait_queue_head_t wq;
168
169 /* Wait queue used by file->poll() */
170 wait_queue_head_t poll_wait;
171
172 /* List of ready file descriptors */
173 struct list_head rdllist;
174
Davide Libenzi67647d02007-05-15 01:40:52 -0700175 /* RB tree root used to store monitored fd structs */
Davide Libenzid47de162007-05-15 01:40:41 -0700176 struct rb_root rbr;
177
178 /*
179 * This is a single linked list that chains all the "struct epitem" that
180 * happened while transfering ready events to userspace w/out
181 * holding ->lock.
182 */
183 struct epitem *ovflist;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800184
185 /* The user that created the eventpoll descriptor */
186 struct user_struct *user;
Davide Libenzid47de162007-05-15 01:40:41 -0700187};
188
189/* Wait structure used by the poll hooks */
190struct eppoll_entry {
191 /* List header used to link this structure to the "struct epitem" */
192 struct list_head llink;
193
194 /* The "base" pointer is set to the container "struct epitem" */
Tony Battersby4f0989d2009-03-31 15:24:16 -0700195 struct epitem *base;
Davide Libenzid47de162007-05-15 01:40:41 -0700196
197 /*
198 * Wait queue item that will be linked to the target file wait
199 * queue head.
200 */
201 wait_queue_t wait;
202
203 /* The wait queue head that linked the "wait" wait queue item */
204 wait_queue_head_t *whead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205};
206
207/* Wrapper struct used by poll queueing */
208struct ep_pqueue {
209 poll_table pt;
210 struct epitem *epi;
211};
212
Davide Libenzi5071f972009-03-31 15:24:10 -0700213/* Used by the ep_send_events() function as callback private data */
214struct ep_send_events_data {
215 int maxevents;
216 struct epoll_event __user *events;
217};
218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219/*
Davide Libenzi7ef99642008-12-01 13:13:55 -0800220 * Configuration options available inside /proc/sys/fs/epoll/
221 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800222/* Maximum number of epoll watched descriptors, per user */
223static int max_user_watches __read_mostly;
224
225/*
Davide Libenzid47de162007-05-15 01:40:41 -0700226 * This mutex is used to serialize ep_free() and eventpoll_release_file().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800228static DEFINE_MUTEX(epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Davide Libenzi5071f972009-03-31 15:24:10 -0700230/* Used for safe wake up implementation */
231static struct nested_calls poll_safewake_ncalls;
232
233/* Used to call file's f_op->poll() under the nested calls boundaries */
234static struct nested_calls poll_readywalk_ncalls;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
236/* Slab cache used to allocate "struct epitem" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800237static struct kmem_cache *epi_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239/* Slab cache used to allocate "struct eppoll_entry" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800240static struct kmem_cache *pwq_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Davide Libenzi7ef99642008-12-01 13:13:55 -0800242#ifdef CONFIG_SYSCTL
243
244#include <linux/sysctl.h>
245
246static int zero;
247
248ctl_table epoll_table[] = {
249 {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800250 .procname = "max_user_watches",
251 .data = &max_user_watches,
252 .maxlen = sizeof(int),
253 .mode = 0644,
Eric W. Biederman6d456112009-11-16 03:11:48 -0800254 .proc_handler = proc_dointvec_minmax,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800255 .extra1 = &zero,
256 },
Eric W. Biedermanab092032009-11-05 14:25:10 -0800257 { }
Davide Libenzi7ef99642008-12-01 13:13:55 -0800258};
259#endif /* CONFIG_SYSCTL */
260
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700261
Davide Libenzi67647d02007-05-15 01:40:52 -0700262/* Setup the structure that is used as key for the RB tree */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700263static inline void ep_set_ffd(struct epoll_filefd *ffd,
264 struct file *file, int fd)
265{
266 ffd->file = file;
267 ffd->fd = fd;
268}
269
Davide Libenzi67647d02007-05-15 01:40:52 -0700270/* Compare RB tree keys */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700271static inline int ep_cmp_ffd(struct epoll_filefd *p1,
272 struct epoll_filefd *p2)
273{
274 return (p1->file > p2->file ? +1:
275 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
276}
277
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700278/* Tells us if the item is currently linked */
279static inline int ep_is_linked(struct list_head *p)
280{
281 return !list_empty(p);
282}
283
284/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700285static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700286{
287 return container_of(p, struct eppoll_entry, wait)->base;
288}
289
290/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700291static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700292{
293 return container_of(p, struct ep_pqueue, pt)->epi;
294}
295
296/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700297static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700298{
299 return op != EPOLL_CTL_DEL;
300}
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700303static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
Davide Libenzi5071f972009-03-31 15:24:10 -0700305 INIT_LIST_HEAD(&ncalls->tasks_call_list);
306 spin_lock_init(&ncalls->lock);
307}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Davide Libenzi5071f972009-03-31 15:24:10 -0700309/**
310 * ep_call_nested - Perform a bound (possibly) nested call, by checking
311 * that the recursion limit is not exceeded, and that
312 * the same nested call (by the meaning of same cookie) is
313 * no re-entered.
314 *
315 * @ncalls: Pointer to the nested_calls structure to be used for this call.
316 * @max_nests: Maximum number of allowed nesting calls.
317 * @nproc: Nested call core function pointer.
318 * @priv: Opaque data to be passed to the @nproc callback.
319 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700320 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700321 *
322 * Returns: Returns the code returned by the @nproc callback, or -1 if
323 * the maximum recursion limit has been exceeded.
324 */
325static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
326 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700327 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700328{
329 int error, call_nests = 0;
330 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700331 struct list_head *lsthead = &ncalls->tasks_call_list;
332 struct nested_call_node *tncur;
333 struct nested_call_node tnode;
334
335 spin_lock_irqsave(&ncalls->lock, flags);
336
337 /*
338 * Try to see if the current task is already inside this wakeup call.
339 * We use a list here, since the population inside this set is always
340 * very much limited.
341 */
342 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700343 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700344 (tncur->cookie == cookie || ++call_nests > max_nests)) {
345 /*
346 * Ops ... loop detected or maximum nest level reached.
347 * We abort this wake by breaking the cycle itself.
348 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700349 error = -1;
350 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700351 }
352 }
353
354 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700355 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700356 tnode.cookie = cookie;
357 list_add(&tnode.llink, lsthead);
358
359 spin_unlock_irqrestore(&ncalls->lock, flags);
360
361 /* Call the nested function */
362 error = (*nproc)(priv, cookie, call_nests);
363
364 /* Remove the current task from the list */
365 spin_lock_irqsave(&ncalls->lock, flags);
366 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700367out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700368 spin_unlock_irqrestore(&ncalls->lock, flags);
369
370 return error;
371}
372
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700373#ifdef CONFIG_DEBUG_LOCK_ALLOC
374static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
375 unsigned long events, int subclass)
376{
377 unsigned long flags;
378
379 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
380 wake_up_locked_poll(wqueue, events);
381 spin_unlock_irqrestore(&wqueue->lock, flags);
382}
383#else
384static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
385 unsigned long events, int subclass)
386{
387 wake_up_poll(wqueue, events);
388}
389#endif
390
Davide Libenzi5071f972009-03-31 15:24:10 -0700391static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
392{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700393 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
394 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700395 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
397
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398/*
399 * Perform a safe wake up of the poll wait list. The problem is that
400 * with the new callback'd wake up system, it is possible that the
401 * poll callback is reentered from inside the call to wake_up() done
402 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700403 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 * and we cannot reenter the same wait queue head at all. This will
405 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700406 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700408static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700410 int this_cpu = get_cpu();
411
Davide Libenzi5071f972009-03-31 15:24:10 -0700412 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700413 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
414
415 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416}
417
Davide Libenzi7699acd2007-05-10 22:23:23 -0700418/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700419 * This function unregisters poll callbacks from the associated file
420 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
421 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700422 */
423static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
424{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700425 struct list_head *lsthead = &epi->pwqlist;
426 struct eppoll_entry *pwq;
427
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700428 while (!list_empty(lsthead)) {
429 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700430
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700431 list_del(&pwq->llink);
432 remove_wait_queue(pwq->whead, &pwq->wait);
433 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700434 }
435}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Davide Libenzi5071f972009-03-31 15:24:10 -0700437/**
438 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
439 * the scan code, to call f_op->poll(). Also allows for
440 * O(NumReady) performance.
441 *
442 * @ep: Pointer to the epoll private data structure.
443 * @sproc: Pointer to the scan callback.
444 * @priv: Private opaque data passed to the @sproc callback.
445 *
446 * Returns: The same integer error code returned by the @sproc callback.
447 */
448static int ep_scan_ready_list(struct eventpoll *ep,
449 int (*sproc)(struct eventpoll *,
450 struct list_head *, void *),
451 void *priv)
452{
453 int error, pwake = 0;
454 unsigned long flags;
455 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700456 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700457
458 /*
459 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700460 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700461 */
462 mutex_lock(&ep->mtx);
463
464 /*
465 * Steal the ready list, and re-init the original one to the
466 * empty list. Also, set ep->ovflist to NULL so that events
467 * happening while looping w/out locks, are not lost. We cannot
468 * have the poll callback to queue directly on ep->rdllist,
469 * because we want the "sproc" callback to be able to do it
470 * in a lockless way.
471 */
472 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700473 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700474 ep->ovflist = NULL;
475 spin_unlock_irqrestore(&ep->lock, flags);
476
477 /*
478 * Now call the callback function.
479 */
480 error = (*sproc)(ep, &txlist, priv);
481
482 spin_lock_irqsave(&ep->lock, flags);
483 /*
484 * During the time we spent inside the "sproc" callback, some
485 * other events might have been queued by the poll callback.
486 * We re-insert them inside the main ready-list here.
487 */
488 for (nepi = ep->ovflist; (epi = nepi) != NULL;
489 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
490 /*
491 * We need to check if the item is already in the list.
492 * During the "sproc" callback execution time, items are
493 * queued into ->ovflist but the "txlist" might already
494 * contain them, and the list_splice() below takes care of them.
495 */
496 if (!ep_is_linked(&epi->rdllink))
497 list_add_tail(&epi->rdllink, &ep->rdllist);
498 }
499 /*
500 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
501 * releasing the lock, events will be queued in the normal way inside
502 * ep->rdllist.
503 */
504 ep->ovflist = EP_UNACTIVE_PTR;
505
506 /*
507 * Quickly re-inject items left on "txlist".
508 */
509 list_splice(&txlist, &ep->rdllist);
510
511 if (!list_empty(&ep->rdllist)) {
512 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700513 * Wake up (if active) both the eventpoll wait list and
514 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700515 */
516 if (waitqueue_active(&ep->wq))
517 wake_up_locked(&ep->wq);
518 if (waitqueue_active(&ep->poll_wait))
519 pwake++;
520 }
521 spin_unlock_irqrestore(&ep->lock, flags);
522
523 mutex_unlock(&ep->mtx);
524
525 /* We have to call this outside the lock */
526 if (pwake)
527 ep_poll_safewake(&ep->poll_wait);
528
529 return error;
530}
531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700533 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700534 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700535 */
536static int ep_remove(struct eventpoll *ep, struct epitem *epi)
537{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700538 unsigned long flags;
539 struct file *file = epi->ffd.file;
540
541 /*
542 * Removes poll wait queue hooks. We _have_ to do this without holding
543 * the "ep->lock" otherwise a deadlock might occur. This because of the
544 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
545 * queue head lock when unregistering the wait queue. The wakeup callback
546 * will run by holding the wait queue head lock and will call our callback
547 * that will try to get "ep->lock".
548 */
549 ep_unregister_pollwait(ep, epi);
550
551 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700552 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700553 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700554 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700555 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Davide Libenzicdac75e2008-04-29 00:58:34 -0700557 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Davide Libenzic7ea7632007-05-15 01:40:47 -0700559 spin_lock_irqsave(&ep->lock, flags);
560 if (ep_is_linked(&epi->rdllink))
561 list_del_init(&epi->rdllink);
562 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Davide Libenzi7699acd2007-05-10 22:23:23 -0700564 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700565 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Davide Libenzi7ef99642008-12-01 13:13:55 -0800567 atomic_dec(&ep->user->epoll_watches);
568
Davide Libenzic7ea7632007-05-15 01:40:47 -0700569 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572static void ep_free(struct eventpoll *ep)
573{
574 struct rb_node *rbp;
575 struct epitem *epi;
576
577 /* We need to release all tasks waiting for these file */
578 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700579 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 /*
582 * We need to lock this because we could be hit by
583 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700584 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 * is on the way to be removed and no one has references to it
586 * anymore. The only hit might come from eventpoll_release_file() but
Arjan van de Ven144efe32006-03-23 03:00:32 -0800587 * holding "epmutex" is sufficent here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800589 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 /*
592 * Walks through the whole tree by unregistering poll callbacks.
593 */
594 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
595 epi = rb_entry(rbp, struct epitem, rbn);
596
597 ep_unregister_pollwait(ep, epi);
598 }
599
600 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700601 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700603 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 * us during this operation. So we can avoid the lock on "ep->lock".
605 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700606 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 epi = rb_entry(rbp, struct epitem, rbn);
608 ep_remove(ep, epi);
609 }
610
Arjan van de Ven144efe32006-03-23 03:00:32 -0800611 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700612 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800613 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700614 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
616
Davide Libenzi7699acd2007-05-10 22:23:23 -0700617static int ep_eventpoll_release(struct inode *inode, struct file *file)
618{
619 struct eventpoll *ep = file->private_data;
620
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700621 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700622 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700623
Davide Libenzi7699acd2007-05-10 22:23:23 -0700624 return 0;
625}
626
Davide Libenzi296e2362009-03-31 15:24:11 -0700627static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
628 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700629{
630 struct epitem *epi, *tmp;
631
632 list_for_each_entry_safe(epi, tmp, head, rdllink) {
633 if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
634 epi->event.events)
635 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700636 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700637 /*
638 * Item has been dropped into the ready list by the poll
639 * callback, but it's not actually ready, as far as
640 * caller requested events goes. We can remove it here.
641 */
642 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700643 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700644 }
645
646 return 0;
647}
648
649static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
650{
651 return ep_scan_ready_list(priv, ep_read_events_proc, NULL);
652}
653
Davide Libenzi7699acd2007-05-10 22:23:23 -0700654static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
655{
Davide Libenzi5071f972009-03-31 15:24:10 -0700656 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700657 struct eventpoll *ep = file->private_data;
658
659 /* Insert inside our poll wait queue */
660 poll_wait(file, &ep->poll_wait, wait);
661
Davide Libenzi5071f972009-03-31 15:24:10 -0700662 /*
663 * Proceed to find out if wanted events are really available inside
664 * the ready list. This need to be done under ep_call_nested()
665 * supervision, since the call to f_op->poll() done on listed files
666 * could re-enter here.
667 */
668 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700669 ep_poll_readyevents_proc, ep, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700670
Davide Libenzi296e2362009-03-31 15:24:11 -0700671 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700672}
673
674/* File callbacks that implement the eventpoll file behaviour */
675static const struct file_operations eventpoll_fops = {
676 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200677 .poll = ep_eventpoll_poll,
678 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700679};
680
681/* Fast test to see if the file is an evenpoll file */
682static inline int is_file_epoll(struct file *f)
683{
684 return f->f_op == &eventpoll_fops;
685}
686
687/*
688 * This is called from eventpoll_release() to unlink files from the eventpoll
689 * interface. We need to have this facility to cleanup correctly files that are
690 * closed without being removed from the eventpoll interface.
691 */
692void eventpoll_release_file(struct file *file)
693{
694 struct list_head *lsthead = &file->f_ep_links;
695 struct eventpoll *ep;
696 struct epitem *epi;
697
698 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700699 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700700 * necessary. It is not necessary because we're in the "struct file"
701 * cleanup path, and this means that noone is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700702 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700703 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700704 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700705 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700706 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700707 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700708 *
709 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700710 */
711 mutex_lock(&epmutex);
712
713 while (!list_empty(lsthead)) {
714 epi = list_first_entry(lsthead, struct epitem, fllink);
715
716 ep = epi->ep;
717 list_del_init(&epi->fllink);
Davide Libenzid47de162007-05-15 01:40:41 -0700718 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700719 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700720 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700721 }
722
723 mutex_unlock(&epmutex);
724}
725
726static int ep_alloc(struct eventpoll **pep)
727{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800728 int error;
729 struct user_struct *user;
730 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700731
Davide Libenzi7ef99642008-12-01 13:13:55 -0800732 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800733 error = -ENOMEM;
734 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
735 if (unlikely(!ep))
736 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700737
Davide Libenzic7ea7632007-05-15 01:40:47 -0700738 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700739 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700740 init_waitqueue_head(&ep->wq);
741 init_waitqueue_head(&ep->poll_wait);
742 INIT_LIST_HEAD(&ep->rdllist);
743 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700744 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800745 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700746
747 *pep = ep;
748
Davide Libenzi7699acd2007-05-10 22:23:23 -0700749 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800750
751free_uid:
752 free_uid(user);
753 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700754}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
756/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700757 * Search the file inside the eventpoll tree. The RB tree operations
758 * are protected by the "mtx" mutex, and ep_find() must be called with
759 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 */
761static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
762{
763 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 struct rb_node *rbp;
765 struct epitem *epi, *epir = NULL;
766 struct epoll_filefd ffd;
767
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700768 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 for (rbp = ep->rbr.rb_node; rbp; ) {
770 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700771 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 if (kcmp > 0)
773 rbp = rbp->rb_right;
774 else if (kcmp < 0)
775 rbp = rbp->rb_left;
776 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 epir = epi;
778 break;
779 }
780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 return epir;
783}
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700786 * This is the callback that is passed to the wait queue wakeup
787 * machanism. It is called by the stored file descriptors when they
788 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700790static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700792 int pwake = 0;
793 unsigned long flags;
794 struct epitem *epi = ep_item_from_wait(wait);
795 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Davide Libenzic7ea7632007-05-15 01:40:47 -0700797 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700798
799 /*
800 * If the event mask does not contain any poll(2) event, we consider the
801 * descriptor to be disabled. This condition is likely the effect of the
802 * EPOLLONESHOT bit that disables the descriptor when an event is received,
803 * until the next EPOLL_CTL_MOD will be issued.
804 */
805 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700806 goto out_unlock;
807
808 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700809 * Check the events coming with the callback. At this stage, not
810 * every device reports the events in the "key" parameter of the
811 * callback. We need to be able to handle both cases here, hence the
812 * test for "key" != NULL before the event match test.
813 */
814 if (key && !((unsigned long) key & epi->event.events))
815 goto out_unlock;
816
817 /*
Davide Libenzid47de162007-05-15 01:40:41 -0700818 * If we are trasfering events to userspace, we can hold no locks
819 * (because we're accessing user memory, and because of linux f_op->poll()
820 * semantics). All the events that happens during that period of time are
821 * chained in ep->ovflist and requeued later on.
822 */
823 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
824 if (epi->next == EP_UNACTIVE_PTR) {
825 epi->next = ep->ovflist;
826 ep->ovflist = epi;
827 }
828 goto out_unlock;
829 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700830
831 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700832 if (!ep_is_linked(&epi->rdllink))
833 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700834
Davide Libenzi7699acd2007-05-10 22:23:23 -0700835 /*
836 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
837 * wait list.
838 */
839 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400840 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700841 if (waitqueue_active(&ep->poll_wait))
842 pwake++;
843
Davide Libenzid47de162007-05-15 01:40:41 -0700844out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700845 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700846
847 /* We have to call this outside the lock */
848 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700849 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700850
851 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852}
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854/*
855 * This is the callback that is used to add our wait queue to the
856 * target file wakeup lists.
857 */
858static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
859 poll_table *pt)
860{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700861 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 struct eppoll_entry *pwq;
863
Christoph Lametere94b1762006-12-06 20:33:17 -0800864 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
866 pwq->whead = whead;
867 pwq->base = epi;
868 add_wait_queue(whead, &pwq->wait);
869 list_add_tail(&pwq->llink, &epi->pwqlist);
870 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700871 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 /* We have to signal that an error occurred */
873 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
878{
879 int kcmp;
880 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
881 struct epitem *epic;
882
883 while (*p) {
884 parent = *p;
885 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700886 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 if (kcmp > 0)
888 p = &parent->rb_right;
889 else
890 p = &parent->rb_left;
891 }
892 rb_link_node(&epi->rbn, parent, p);
893 rb_insert_color(&epi->rbn, &ep->rbr);
894}
895
Davide Libenzic7ea7632007-05-15 01:40:47 -0700896/*
897 * Must be called with "mtx" held.
898 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
900 struct file *tfile, int fd)
901{
902 int error, revents, pwake = 0;
903 unsigned long flags;
904 struct epitem *epi;
905 struct ep_pqueue epq;
906
Davide Libenzi7ef99642008-12-01 13:13:55 -0800907 if (unlikely(atomic_read(&ep->user->epoll_watches) >=
908 max_user_watches))
909 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -0800910 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -0800911 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 INIT_LIST_HEAD(&epi->rdllink);
915 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 INIT_LIST_HEAD(&epi->pwqlist);
917 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700918 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -0700921 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 /* Initialize the poll table using the queue callback */
924 epq.epi = epi;
925 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
926
927 /*
928 * Attach the item to the poll hooks and get current event bits.
929 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -0700930 * been increased by the caller of this function. Note that after
931 * this operation completes, the poll callback can start hitting
932 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 */
934 revents = tfile->f_op->poll(tfile, &epq.pt);
935
936 /*
937 * We have to check if something went wrong during the poll wait queue
938 * install process. Namely an allocation for a wait queue failed due
939 * high memory pressure.
940 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800941 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700943 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
945 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -0700946 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -0700948 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Davide Libenzic7ea7632007-05-15 01:40:47 -0700950 /*
951 * Add the current item to the RB tree. All RB tree operations are
952 * protected by "mtx", and ep_insert() is called with "mtx" held.
953 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 ep_rbtree_insert(ep, epi);
955
Davide Libenzic7ea7632007-05-15 01:40:47 -0700956 /* We have to drop the new item inside our item list to keep track of it */
957 spin_lock_irqsave(&ep->lock, flags);
958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700960 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 list_add_tail(&epi->rdllink, &ep->rdllist);
962
963 /* Notify waiting tasks that events are available */
964 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400965 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (waitqueue_active(&ep->poll_wait))
967 pwake++;
968 }
969
Davide Libenzic7ea7632007-05-15 01:40:47 -0700970 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Davide Libenzi7ef99642008-12-01 13:13:55 -0800972 atomic_inc(&ep->user->epoll_watches);
973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 /* We have to call this outside the lock */
975 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700976 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 return 0;
979
Davide Libenzi7699acd2007-05-10 22:23:23 -0700980error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 ep_unregister_pollwait(ep, epi);
982
983 /*
984 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -0700985 * allocated wait queue. Note that we don't care about the ep->ovflist
986 * list, since that is used/cleaned only inside a section bound by "mtx".
987 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700989 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700990 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700991 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -0700992 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700994 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 return error;
997}
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999/*
1000 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001001 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 */
1003static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1004{
1005 int pwake = 0;
1006 unsigned int revents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001009 * Set the new event interest mask before calling f_op->poll();
1010 * otherwise we might miss an event that happens between the
1011 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 */
1013 epi->event.events = event->events;
Tony Battersbye057e152009-03-31 15:24:15 -07001014 epi->event.data = event->data; /* protected by mtx */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 /*
1017 * Get current event bits. We can safely use the file* here because
1018 * its usage count has been increased by the caller of this function.
1019 */
1020 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL);
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001023 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001024 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001026 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001027 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001028 if (!ep_is_linked(&epi->rdllink)) {
1029 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Davide Libenzic7ea7632007-05-15 01:40:47 -07001031 /* Notify waiting tasks that events are available */
1032 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001033 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001034 if (waitqueue_active(&ep->poll_wait))
1035 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
Tony Battersbye057e152009-03-31 15:24:15 -07001037 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 /* We have to call this outside the lock */
1041 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001042 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 return 0;
1045}
1046
Davide Libenzi296e2362009-03-31 15:24:11 -07001047static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1048 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001049{
1050 struct ep_send_events_data *esed = priv;
1051 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001052 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001053 struct epitem *epi;
1054 struct epoll_event __user *uevent;
1055
Davide Libenzi296e2362009-03-31 15:24:11 -07001056 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001057 * We can loop without lock because we are passed a task private list.
1058 * Items cannot vanish during the loop because ep_scan_ready_list() is
1059 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001060 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001061 for (eventcnt = 0, uevent = esed->events;
1062 !list_empty(head) && eventcnt < esed->maxevents;) {
1063 epi = list_first_entry(head, struct epitem, rdllink);
1064
1065 list_del_init(&epi->rdllink);
1066
Davide Libenzi296e2362009-03-31 15:24:11 -07001067 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
1068 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001069
Davide Libenzi296e2362009-03-31 15:24:11 -07001070 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001071 * If the event mask intersect the caller-requested one,
1072 * deliver the event to userspace. Again, ep_scan_ready_list()
1073 * is holding "mtx", so no operations coming from userspace
1074 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001075 */
1076 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001077 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001078 __put_user(epi->event.data, &uevent->data)) {
1079 list_add(&epi->rdllink, head);
Davide Libenzi296e2362009-03-31 15:24:11 -07001080 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001081 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001082 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001083 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001084 if (epi->event.events & EPOLLONESHOT)
1085 epi->event.events &= EP_PRIVATE_BITS;
1086 else if (!(epi->event.events & EPOLLET)) {
1087 /*
1088 * If this file has been added with Level
1089 * Trigger mode, we need to insert back inside
1090 * the ready list, so that the next call to
1091 * epoll_wait() will check again the events
1092 * availability. At this point, noone can insert
1093 * into ep->rdllist besides us. The epoll_ctl()
1094 * callers are locked out by
1095 * ep_scan_ready_list() holding "mtx" and the
1096 * poll callback will queue them in ep->ovflist.
1097 */
1098 list_add_tail(&epi->rdllink, &ep->rdllist);
1099 }
1100 }
1101 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001102
1103 return eventcnt;
1104}
1105
Davide Libenzi296e2362009-03-31 15:24:11 -07001106static int ep_send_events(struct eventpoll *ep,
1107 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108{
Davide Libenzi5071f972009-03-31 15:24:10 -07001109 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001110
Davide Libenzi5071f972009-03-31 15:24:10 -07001111 esed.maxevents = maxevents;
1112 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001113
Davide Libenzi5071f972009-03-31 15:24:10 -07001114 return ep_scan_ready_list(ep, ep_send_events_proc, &esed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115}
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1118 int maxevents, long timeout)
1119{
1120 int res, eavail;
1121 unsigned long flags;
1122 long jtimeout;
1123 wait_queue_t wait;
1124
1125 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001126 * Calculate the timeout by checking for the "infinite" value (-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 * and the overflow condition. The passed timeout is in milliseconds,
1128 * that why (t * HZ) / 1000.
1129 */
Davide Libenzie3306dd2005-09-27 21:45:33 -07001130 jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ?
1131 MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
1133retry:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001134 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 res = 0;
1137 if (list_empty(&ep->rdllist)) {
1138 /*
1139 * We don't have any available event to return to the caller.
1140 * We need to sleep here, and we will be wake up by
1141 * ep_poll_callback() when events will become available.
1142 */
1143 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001144 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 for (;;) {
1147 /*
1148 * We don't want to sleep if the ep_poll_callback() sends us
1149 * a wakeup in between. That's why we set the task state
1150 * to TASK_INTERRUPTIBLE before doing the checks.
1151 */
1152 set_current_state(TASK_INTERRUPTIBLE);
1153 if (!list_empty(&ep->rdllist) || !jtimeout)
1154 break;
1155 if (signal_pending(current)) {
1156 res = -EINTR;
1157 break;
1158 }
1159
Davide Libenzic7ea7632007-05-15 01:40:47 -07001160 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 jtimeout = schedule_timeout(jtimeout);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001162 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001164 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
1166 set_current_state(TASK_RUNNING);
1167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 /* Is it worth to try to dig for events ? */
Davide Libenzi5071f972009-03-31 15:24:10 -07001169 eavail = !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Davide Libenzic7ea7632007-05-15 01:40:47 -07001171 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173 /*
1174 * Try to transfer events to user space. In case we get 0 events and
1175 * there's still timeout left over, we go trying again in search of
1176 * more luck.
1177 */
1178 if (!res && eavail &&
Davide Libenzid47de162007-05-15 01:40:41 -07001179 !(res = ep_send_events(ep, events, maxevents)) && jtimeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 goto retry;
1181
1182 return res;
1183}
1184
Davide Libenzi7699acd2007-05-10 22:23:23 -07001185/*
Andrew Morton523723b2008-08-12 15:09:01 -07001186 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001187 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001188SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001189{
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001190 int error;
1191 struct eventpoll *ep = NULL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001192
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001193 /* Check the EPOLL_* constant for consistency. */
1194 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1195
Davide Libenzi296e2362009-03-31 15:24:11 -07001196 if (flags & ~EPOLL_CLOEXEC)
1197 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001198 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001199 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001200 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001201 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001202 if (error < 0)
1203 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001204 /*
1205 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001206 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001207 */
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001208 error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001209 O_RDWR | (flags & O_CLOEXEC));
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001210 if (error < 0)
Al Viro2030a422008-02-23 06:46:49 -05001211 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001212
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001213 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001214}
1215
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001216SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001217{
Davide Libenzibfe38912009-05-12 13:19:44 -07001218 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001219 return -EINVAL;
1220
1221 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001222}
1223
Davide Libenzi7699acd2007-05-10 22:23:23 -07001224/*
1225 * The following function implements the controller interface for
1226 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001227 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001228 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001229SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1230 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001231{
1232 int error;
1233 struct file *file, *tfile;
1234 struct eventpoll *ep;
1235 struct epitem *epi;
1236 struct epoll_event epds;
1237
Davide Libenzi7699acd2007-05-10 22:23:23 -07001238 error = -EFAULT;
1239 if (ep_op_has_event(op) &&
1240 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1241 goto error_return;
1242
1243 /* Get the "struct file *" for the eventpoll file */
1244 error = -EBADF;
1245 file = fget(epfd);
1246 if (!file)
1247 goto error_return;
1248
1249 /* Get the "struct file *" for the target file */
1250 tfile = fget(fd);
1251 if (!tfile)
1252 goto error_fput;
1253
1254 /* The target file descriptor must support poll */
1255 error = -EPERM;
1256 if (!tfile->f_op || !tfile->f_op->poll)
1257 goto error_tgt_fput;
1258
1259 /*
1260 * We have to check that the file structure underneath the file descriptor
1261 * the user passed to us _is_ an eventpoll file. And also we do not permit
1262 * adding an epoll file descriptor inside itself.
1263 */
1264 error = -EINVAL;
1265 if (file == tfile || !is_file_epoll(file))
1266 goto error_tgt_fput;
1267
1268 /*
1269 * At this point it is safe to assume that the "private_data" contains
1270 * our own data structure.
1271 */
1272 ep = file->private_data;
1273
Davide Libenzid47de162007-05-15 01:40:41 -07001274 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001275
Davide Libenzi67647d02007-05-15 01:40:52 -07001276 /*
1277 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1278 * above, we can be sure to be able to use the item looked up by
1279 * ep_find() till we release the mutex.
1280 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001281 epi = ep_find(ep, tfile, fd);
1282
1283 error = -EINVAL;
1284 switch (op) {
1285 case EPOLL_CTL_ADD:
1286 if (!epi) {
1287 epds.events |= POLLERR | POLLHUP;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001288 error = ep_insert(ep, &epds, tfile, fd);
1289 } else
1290 error = -EEXIST;
1291 break;
1292 case EPOLL_CTL_DEL:
1293 if (epi)
1294 error = ep_remove(ep, epi);
1295 else
1296 error = -ENOENT;
1297 break;
1298 case EPOLL_CTL_MOD:
1299 if (epi) {
1300 epds.events |= POLLERR | POLLHUP;
1301 error = ep_modify(ep, epi, &epds);
1302 } else
1303 error = -ENOENT;
1304 break;
1305 }
Davide Libenzid47de162007-05-15 01:40:41 -07001306 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001307
1308error_tgt_fput:
1309 fput(tfile);
1310error_fput:
1311 fput(file);
1312error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001313
1314 return error;
1315}
1316
1317/*
1318 * Implement the event wait interface for the eventpoll file. It is the kernel
1319 * part of the user space epoll_wait(2).
1320 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001321SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1322 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001323{
1324 int error;
1325 struct file *file;
1326 struct eventpoll *ep;
1327
Davide Libenzi7699acd2007-05-10 22:23:23 -07001328 /* The maximum number of event must be greater than zero */
1329 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1330 return -EINVAL;
1331
1332 /* Verify that the area passed by the user is writeable */
1333 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1334 error = -EFAULT;
1335 goto error_return;
1336 }
1337
1338 /* Get the "struct file *" for the eventpoll file */
1339 error = -EBADF;
1340 file = fget(epfd);
1341 if (!file)
1342 goto error_return;
1343
1344 /*
1345 * We have to check that the file structure underneath the fd
1346 * the user passed to us _is_ an eventpoll file.
1347 */
1348 error = -EINVAL;
1349 if (!is_file_epoll(file))
1350 goto error_fput;
1351
1352 /*
1353 * At this point it is safe to assume that the "private_data" contains
1354 * our own data structure.
1355 */
1356 ep = file->private_data;
1357
1358 /* Time to fish for events ... */
1359 error = ep_poll(ep, events, maxevents, timeout);
1360
1361error_fput:
1362 fput(file);
1363error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001364
1365 return error;
1366}
1367
Roland McGrathf3de2722008-04-30 00:53:09 -07001368#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001369
1370/*
1371 * Implement the event wait interface for the eventpoll file. It is the kernel
1372 * part of the user space epoll_pwait(2).
1373 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001374SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1375 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1376 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001377{
1378 int error;
1379 sigset_t ksigmask, sigsaved;
1380
1381 /*
1382 * If the caller wants a certain signal mask to be set during the wait,
1383 * we apply it here.
1384 */
1385 if (sigmask) {
1386 if (sigsetsize != sizeof(sigset_t))
1387 return -EINVAL;
1388 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1389 return -EFAULT;
1390 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1391 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1392 }
1393
1394 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1395
1396 /*
1397 * If we changed the signal mask, we need to restore the original one.
1398 * In case we've got a signal while waiting, we do not restore the
1399 * signal mask yet, and we allow do_signal() to deliver the signal on
1400 * the way back to userspace, before the signal mask is restored.
1401 */
1402 if (sigmask) {
1403 if (error == -EINTR) {
1404 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001405 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001406 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001407 } else
1408 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1409 }
1410
1411 return error;
1412}
1413
Roland McGrathf3de2722008-04-30 00:53:09 -07001414#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416static int __init eventpoll_init(void)
1417{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001418 struct sysinfo si;
1419
1420 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001421 /*
1422 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1423 */
1424 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001425 EP_ITEM_COST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
1427 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001428 ep_nested_calls_init(&poll_safewake_ncalls);
1429
1430 /* Initialize the structure used to perform file's f_op->poll() calls */
1431 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
1433 /* Allocates slab cache used to allocate "struct epitem" items */
1434 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001435 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
1437 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1438 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001439 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
Davide Libenzicea69242007-05-10 22:23:22 -07001443fs_initcall(eventpoll_init);