blob: e24d137c93b6d960a3ef0b9511e726c054ce6c9a [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#define DEBUG_EPOLL 0
75
76#if DEBUG_EPOLL > 0
77#define DPRINTK(x) printk x
78#define DNPRINTK(n, x) do { if ((n) <= DEBUG_EPOLL) printk x; } while (0)
79#else /* #if DEBUG_EPOLL > 0 */
80#define DPRINTK(x) (void) 0
81#define DNPRINTK(n, x) (void) 0
82#endif /* #if DEBUG_EPOLL > 0 */
83
84#define DEBUG_EPI 0
85
86#if DEBUG_EPI != 0
87#define EPI_SLAB_DEBUG (SLAB_DEBUG_FREE | SLAB_RED_ZONE /* | SLAB_POISON */)
88#else /* #if DEBUG_EPI != 0 */
89#define EPI_SLAB_DEBUG 0
90#endif /* #if DEBUG_EPI != 0 */
91
92/* Epoll private bits inside the event mask */
93#define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET)
94
Davide Libenzi5071f972009-03-31 15:24:10 -070095/* Maximum number of nesting allowed inside epoll sets */
96#define EP_MAX_NESTS 4
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Davide Libenzie3306dd2005-09-27 21:45:33 -070098/* Maximum msec timeout value storeable in a long int */
99#define EP_MAX_MSTIMEO min(1000ULL * MAX_SCHEDULE_TIMEOUT / HZ, (LONG_MAX - 999ULL) / HZ)
100
Davide Libenzib6119672006-10-11 01:21:44 -0700101#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
102
Davide Libenzid47de162007-05-15 01:40:41 -0700103#define EP_UNACTIVE_PTR ((void *) -1L)
104
Davide Libenzi7ef99642008-12-01 13:13:55 -0800105#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107struct epoll_filefd {
108 struct file *file;
109 int fd;
110};
111
112/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700113 * Structure used to track possible nested calls, for too deep recursions
114 * and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700116struct nested_call_node {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 struct list_head llink;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700118 struct task_struct *task;
Davide Libenzi5071f972009-03-31 15:24:10 -0700119 void *cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120};
121
122/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700123 * This structure is used as collector for nested calls, to check for
124 * maximum recursion dept and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700126struct nested_calls {
127 struct list_head tasks_call_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 spinlock_t lock;
129};
130
131/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 * Each file descriptor added to the eventpoll interface will
Davide Libenzi6192bd52007-05-08 00:25:41 -0700133 * have an entry of this type linked to the "rbr" RB tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 */
135struct epitem {
Davide Libenzi67647d02007-05-15 01:40:52 -0700136 /* RB tree node used to link this structure to the eventpoll RB tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 struct rb_node rbn;
138
139 /* List header used to link this structure to the eventpoll ready list */
140 struct list_head rdllink;
141
Davide Libenzic7ea7632007-05-15 01:40:47 -0700142 /*
143 * Works together "struct eventpoll"->ovflist in keeping the
144 * single linked chain of items.
145 */
146 struct epitem *next;
147
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 /* The file descriptor information this item refers to */
149 struct epoll_filefd ffd;
150
151 /* Number of active wait queue attached to poll operations */
152 int nwait;
153
154 /* List containing poll wait queues */
155 struct list_head pwqlist;
156
157 /* The "container" of this item */
158 struct eventpoll *ep;
159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 /* List header used to link this item to the "struct file" items list */
161 struct list_head fllink;
Davide Libenzid47de162007-05-15 01:40:41 -0700162
Davide Libenzic7ea7632007-05-15 01:40:47 -0700163 /* The structure that describe the interested events and the source fd */
164 struct epoll_event event;
Davide Libenzid47de162007-05-15 01:40:41 -0700165};
166
167/*
168 * This structure is stored inside the "private_data" member of the file
169 * structure and rapresent the main data sructure for the eventpoll
170 * interface.
171 */
172struct eventpoll {
173 /* Protect the this structure access */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700174 spinlock_t lock;
Davide Libenzid47de162007-05-15 01:40:41 -0700175
176 /*
177 * This mutex is used to ensure that files are not removed
178 * while epoll is using them. This is held during the event
179 * collection loop, the file cleanup path, the epoll file exit
180 * code and the ctl operations.
181 */
182 struct mutex mtx;
183
184 /* Wait queue used by sys_epoll_wait() */
185 wait_queue_head_t wq;
186
187 /* Wait queue used by file->poll() */
188 wait_queue_head_t poll_wait;
189
190 /* List of ready file descriptors */
191 struct list_head rdllist;
192
Davide Libenzi67647d02007-05-15 01:40:52 -0700193 /* RB tree root used to store monitored fd structs */
Davide Libenzid47de162007-05-15 01:40:41 -0700194 struct rb_root rbr;
195
196 /*
197 * This is a single linked list that chains all the "struct epitem" that
198 * happened while transfering ready events to userspace w/out
199 * holding ->lock.
200 */
201 struct epitem *ovflist;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800202
203 /* The user that created the eventpoll descriptor */
204 struct user_struct *user;
Davide Libenzid47de162007-05-15 01:40:41 -0700205};
206
207/* Wait structure used by the poll hooks */
208struct eppoll_entry {
209 /* List header used to link this structure to the "struct epitem" */
210 struct list_head llink;
211
212 /* The "base" pointer is set to the container "struct epitem" */
213 void *base;
214
215 /*
216 * Wait queue item that will be linked to the target file wait
217 * queue head.
218 */
219 wait_queue_t wait;
220
221 /* The wait queue head that linked the "wait" wait queue item */
222 wait_queue_head_t *whead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223};
224
225/* Wrapper struct used by poll queueing */
226struct ep_pqueue {
227 poll_table pt;
228 struct epitem *epi;
229};
230
Davide Libenzi5071f972009-03-31 15:24:10 -0700231/* Used by the ep_send_events() function as callback private data */
232struct ep_send_events_data {
233 int maxevents;
234 struct epoll_event __user *events;
235};
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237/*
Davide Libenzi7ef99642008-12-01 13:13:55 -0800238 * Configuration options available inside /proc/sys/fs/epoll/
239 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800240/* Maximum number of epoll watched descriptors, per user */
241static int max_user_watches __read_mostly;
242
243/*
Davide Libenzid47de162007-05-15 01:40:41 -0700244 * This mutex is used to serialize ep_free() and eventpoll_release_file().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800246static DEFINE_MUTEX(epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Davide Libenzi5071f972009-03-31 15:24:10 -0700248/* Used for safe wake up implementation */
249static struct nested_calls poll_safewake_ncalls;
250
251/* Used to call file's f_op->poll() under the nested calls boundaries */
252static struct nested_calls poll_readywalk_ncalls;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
254/* Slab cache used to allocate "struct epitem" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800255static struct kmem_cache *epi_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257/* Slab cache used to allocate "struct eppoll_entry" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800258static struct kmem_cache *pwq_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Davide Libenzi7ef99642008-12-01 13:13:55 -0800260#ifdef CONFIG_SYSCTL
261
262#include <linux/sysctl.h>
263
264static int zero;
265
266ctl_table epoll_table[] = {
267 {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800268 .procname = "max_user_watches",
269 .data = &max_user_watches,
270 .maxlen = sizeof(int),
271 .mode = 0644,
272 .proc_handler = &proc_dointvec_minmax,
273 .extra1 = &zero,
274 },
275 { .ctl_name = 0 }
276};
277#endif /* CONFIG_SYSCTL */
278
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700279
Davide Libenzi67647d02007-05-15 01:40:52 -0700280/* Setup the structure that is used as key for the RB tree */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700281static inline void ep_set_ffd(struct epoll_filefd *ffd,
282 struct file *file, int fd)
283{
284 ffd->file = file;
285 ffd->fd = fd;
286}
287
Davide Libenzi67647d02007-05-15 01:40:52 -0700288/* Compare RB tree keys */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700289static inline int ep_cmp_ffd(struct epoll_filefd *p1,
290 struct epoll_filefd *p2)
291{
292 return (p1->file > p2->file ? +1:
293 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
294}
295
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700296/* Tells us if the item is currently linked */
297static inline int ep_is_linked(struct list_head *p)
298{
299 return !list_empty(p);
300}
301
302/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700303static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700304{
305 return container_of(p, struct eppoll_entry, wait)->base;
306}
307
308/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700309static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700310{
311 return container_of(p, struct ep_pqueue, pt)->epi;
312}
313
314/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700315static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700316{
317 return op != EPOLL_CTL_DEL;
318}
319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700321static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
Davide Libenzi5071f972009-03-31 15:24:10 -0700323 INIT_LIST_HEAD(&ncalls->tasks_call_list);
324 spin_lock_init(&ncalls->lock);
325}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Davide Libenzi5071f972009-03-31 15:24:10 -0700327/**
328 * ep_call_nested - Perform a bound (possibly) nested call, by checking
329 * that the recursion limit is not exceeded, and that
330 * the same nested call (by the meaning of same cookie) is
331 * no re-entered.
332 *
333 * @ncalls: Pointer to the nested_calls structure to be used for this call.
334 * @max_nests: Maximum number of allowed nesting calls.
335 * @nproc: Nested call core function pointer.
336 * @priv: Opaque data to be passed to the @nproc callback.
337 * @cookie: Cookie to be used to identify this nested call.
338 *
339 * Returns: Returns the code returned by the @nproc callback, or -1 if
340 * the maximum recursion limit has been exceeded.
341 */
342static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
343 int (*nproc)(void *, void *, int), void *priv,
344 void *cookie)
345{
346 int error, call_nests = 0;
347 unsigned long flags;
348 struct task_struct *this_task = current;
349 struct list_head *lsthead = &ncalls->tasks_call_list;
350 struct nested_call_node *tncur;
351 struct nested_call_node tnode;
352
353 spin_lock_irqsave(&ncalls->lock, flags);
354
355 /*
356 * Try to see if the current task is already inside this wakeup call.
357 * We use a list here, since the population inside this set is always
358 * very much limited.
359 */
360 list_for_each_entry(tncur, lsthead, llink) {
361 if (tncur->task == this_task &&
362 (tncur->cookie == cookie || ++call_nests > max_nests)) {
363 /*
364 * Ops ... loop detected or maximum nest level reached.
365 * We abort this wake by breaking the cycle itself.
366 */
367 spin_unlock_irqrestore(&ncalls->lock, flags);
368
369 return -1;
370 }
371 }
372
373 /* Add the current task and cookie to the list */
374 tnode.task = this_task;
375 tnode.cookie = cookie;
376 list_add(&tnode.llink, lsthead);
377
378 spin_unlock_irqrestore(&ncalls->lock, flags);
379
380 /* Call the nested function */
381 error = (*nproc)(priv, cookie, call_nests);
382
383 /* Remove the current task from the list */
384 spin_lock_irqsave(&ncalls->lock, flags);
385 list_del(&tnode.llink);
386 spin_unlock_irqrestore(&ncalls->lock, flags);
387
388 return error;
389}
390
391static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
392{
393 wake_up_nested((wait_queue_head_t *) cookie, 1 + call_nests);
394 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397/*
398 * Perform a safe wake up of the poll wait list. The problem is that
399 * with the new callback'd wake up system, it is possible that the
400 * poll callback is reentered from inside the call to wake_up() done
401 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700402 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 * and we cannot reenter the same wait queue head at all. This will
404 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700405 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700407static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408{
Davide Libenzi5071f972009-03-31 15:24:10 -0700409 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
410 ep_poll_wakeup_proc, NULL, wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
412
Davide Libenzi7699acd2007-05-10 22:23:23 -0700413/*
414 * This function unregister poll callbacks from the associated file descriptor.
415 * Since this must be called without holding "ep->lock" the atomic exchange trick
416 * will protect us from multiple unregister.
417 */
418static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
419{
420 int nwait;
421 struct list_head *lsthead = &epi->pwqlist;
422 struct eppoll_entry *pwq;
423
424 /* This is called without locks, so we need the atomic exchange */
425 nwait = xchg(&epi->nwait, 0);
426
427 if (nwait) {
428 while (!list_empty(lsthead)) {
429 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
430
431 list_del_init(&pwq->llink);
432 remove_wait_queue(pwq->whead, &pwq->wait);
433 kmem_cache_free(pwq_cache, pwq);
434 }
435 }
436}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Davide Libenzi5071f972009-03-31 15:24:10 -0700438/**
439 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
440 * the scan code, to call f_op->poll(). Also allows for
441 * O(NumReady) performance.
442 *
443 * @ep: Pointer to the epoll private data structure.
444 * @sproc: Pointer to the scan callback.
445 * @priv: Private opaque data passed to the @sproc callback.
446 *
447 * Returns: The same integer error code returned by the @sproc callback.
448 */
449static int ep_scan_ready_list(struct eventpoll *ep,
450 int (*sproc)(struct eventpoll *,
451 struct list_head *, void *),
452 void *priv)
453{
454 int error, pwake = 0;
455 unsigned long flags;
456 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700457 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700458
459 /*
460 * We need to lock this because we could be hit by
461 * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL).
462 */
463 mutex_lock(&ep->mtx);
464
465 /*
466 * Steal the ready list, and re-init the original one to the
467 * empty list. Also, set ep->ovflist to NULL so that events
468 * happening while looping w/out locks, are not lost. We cannot
469 * have the poll callback to queue directly on ep->rdllist,
470 * because we want the "sproc" callback to be able to do it
471 * in a lockless way.
472 */
473 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700474 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700475 ep->ovflist = NULL;
476 spin_unlock_irqrestore(&ep->lock, flags);
477
478 /*
479 * Now call the callback function.
480 */
481 error = (*sproc)(ep, &txlist, priv);
482
483 spin_lock_irqsave(&ep->lock, flags);
484 /*
485 * During the time we spent inside the "sproc" callback, some
486 * other events might have been queued by the poll callback.
487 * We re-insert them inside the main ready-list here.
488 */
489 for (nepi = ep->ovflist; (epi = nepi) != NULL;
490 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
491 /*
492 * We need to check if the item is already in the list.
493 * During the "sproc" callback execution time, items are
494 * queued into ->ovflist but the "txlist" might already
495 * contain them, and the list_splice() below takes care of them.
496 */
497 if (!ep_is_linked(&epi->rdllink))
498 list_add_tail(&epi->rdllink, &ep->rdllist);
499 }
500 /*
501 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
502 * releasing the lock, events will be queued in the normal way inside
503 * ep->rdllist.
504 */
505 ep->ovflist = EP_UNACTIVE_PTR;
506
507 /*
508 * Quickly re-inject items left on "txlist".
509 */
510 list_splice(&txlist, &ep->rdllist);
511
512 if (!list_empty(&ep->rdllist)) {
513 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700514 * Wake up (if active) both the eventpoll wait list and
515 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700516 */
517 if (waitqueue_active(&ep->wq))
518 wake_up_locked(&ep->wq);
519 if (waitqueue_active(&ep->poll_wait))
520 pwake++;
521 }
522 spin_unlock_irqrestore(&ep->lock, flags);
523
524 mutex_unlock(&ep->mtx);
525
526 /* We have to call this outside the lock */
527 if (pwake)
528 ep_poll_safewake(&ep->poll_wait);
529
530 return error;
531}
532
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700534 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700535 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700536 */
537static int ep_remove(struct eventpoll *ep, struct epitem *epi)
538{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700539 unsigned long flags;
540 struct file *file = epi->ffd.file;
541
542 /*
543 * Removes poll wait queue hooks. We _have_ to do this without holding
544 * the "ep->lock" otherwise a deadlock might occur. This because of the
545 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
546 * queue head lock when unregistering the wait queue. The wakeup callback
547 * will run by holding the wait queue head lock and will call our callback
548 * that will try to get "ep->lock".
549 */
550 ep_unregister_pollwait(ep, epi);
551
552 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700553 spin_lock(&file->f_lock);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700554 if (ep_is_linked(&epi->fllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700555 list_del_init(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700556 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Davide Libenzicdac75e2008-04-29 00:58:34 -0700558 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Davide Libenzic7ea7632007-05-15 01:40:47 -0700560 spin_lock_irqsave(&ep->lock, flags);
561 if (ep_is_linked(&epi->rdllink))
562 list_del_init(&epi->rdllink);
563 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Davide Libenzi7699acd2007-05-10 22:23:23 -0700565 /* At this point it is safe to free the eventpoll item */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700566 kmem_cache_free(epi_cache, epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Davide Libenzi7ef99642008-12-01 13:13:55 -0800568 atomic_dec(&ep->user->epoll_watches);
569
Davide Libenzic7ea7632007-05-15 01:40:47 -0700570 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_remove(%p, %p)\n",
571 current, ep, file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Davide Libenzic7ea7632007-05-15 01:40:47 -0700573 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576static void ep_free(struct eventpoll *ep)
577{
578 struct rb_node *rbp;
579 struct epitem *epi;
580
581 /* We need to release all tasks waiting for these file */
582 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700583 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 /*
586 * We need to lock this because we could be hit by
587 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700588 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 * is on the way to be removed and no one has references to it
590 * anymore. The only hit might come from eventpoll_release_file() but
Arjan van de Ven144efe32006-03-23 03:00:32 -0800591 * holding "epmutex" is sufficent here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800593 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595 /*
596 * Walks through the whole tree by unregistering poll callbacks.
597 */
598 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
599 epi = rb_entry(rbp, struct epitem, rbn);
600
601 ep_unregister_pollwait(ep, epi);
602 }
603
604 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700605 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700607 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 * us during this operation. So we can avoid the lock on "ep->lock".
609 */
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700610 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 epi = rb_entry(rbp, struct epitem, rbn);
612 ep_remove(ep, epi);
613 }
614
Arjan van de Ven144efe32006-03-23 03:00:32 -0800615 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700616 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800617 free_uid(ep->user);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700618 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
620
Davide Libenzi7699acd2007-05-10 22:23:23 -0700621static int ep_eventpoll_release(struct inode *inode, struct file *file)
622{
623 struct eventpoll *ep = file->private_data;
624
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700625 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700626 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700627
628 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: close() ep=%p\n", current, ep));
629 return 0;
630}
631
Davide Libenzi296e2362009-03-31 15:24:11 -0700632static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
633 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700634{
635 struct epitem *epi, *tmp;
636
637 list_for_each_entry_safe(epi, tmp, head, rdllink) {
638 if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
639 epi->event.events)
640 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700641 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700642 /*
643 * Item has been dropped into the ready list by the poll
644 * callback, but it's not actually ready, as far as
645 * caller requested events goes. We can remove it here.
646 */
647 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700648 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700649 }
650
651 return 0;
652}
653
654static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
655{
656 return ep_scan_ready_list(priv, ep_read_events_proc, NULL);
657}
658
Davide Libenzi7699acd2007-05-10 22:23:23 -0700659static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
660{
Davide Libenzi5071f972009-03-31 15:24:10 -0700661 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700662 struct eventpoll *ep = file->private_data;
663
664 /* Insert inside our poll wait queue */
665 poll_wait(file, &ep->poll_wait, wait);
666
Davide Libenzi5071f972009-03-31 15:24:10 -0700667 /*
668 * Proceed to find out if wanted events are really available inside
669 * the ready list. This need to be done under ep_call_nested()
670 * supervision, since the call to f_op->poll() done on listed files
671 * could re-enter here.
672 */
673 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
674 ep_poll_readyevents_proc, ep, ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700675
Davide Libenzi296e2362009-03-31 15:24:11 -0700676 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700677}
678
679/* File callbacks that implement the eventpoll file behaviour */
680static const struct file_operations eventpoll_fops = {
681 .release = ep_eventpoll_release,
682 .poll = ep_eventpoll_poll
683};
684
685/* Fast test to see if the file is an evenpoll file */
686static inline int is_file_epoll(struct file *f)
687{
688 return f->f_op == &eventpoll_fops;
689}
690
691/*
692 * This is called from eventpoll_release() to unlink files from the eventpoll
693 * interface. We need to have this facility to cleanup correctly files that are
694 * closed without being removed from the eventpoll interface.
695 */
696void eventpoll_release_file(struct file *file)
697{
698 struct list_head *lsthead = &file->f_ep_links;
699 struct eventpoll *ep;
700 struct epitem *epi;
701
702 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700703 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700704 * necessary. It is not necessary because we're in the "struct file"
705 * cleanup path, and this means that noone is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700706 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700707 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700708 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700709 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700710 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700711 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700712 *
713 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700714 */
715 mutex_lock(&epmutex);
716
717 while (!list_empty(lsthead)) {
718 epi = list_first_entry(lsthead, struct epitem, fllink);
719
720 ep = epi->ep;
721 list_del_init(&epi->fllink);
Davide Libenzid47de162007-05-15 01:40:41 -0700722 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700723 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700724 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700725 }
726
727 mutex_unlock(&epmutex);
728}
729
730static int ep_alloc(struct eventpoll **pep)
731{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800732 int error;
733 struct user_struct *user;
734 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700735
Davide Libenzi7ef99642008-12-01 13:13:55 -0800736 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800737 error = -ENOMEM;
738 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
739 if (unlikely(!ep))
740 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700741
Davide Libenzic7ea7632007-05-15 01:40:47 -0700742 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700743 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700744 init_waitqueue_head(&ep->wq);
745 init_waitqueue_head(&ep->poll_wait);
746 INIT_LIST_HEAD(&ep->rdllist);
747 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700748 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800749 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700750
751 *pep = ep;
752
753 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_alloc() ep=%p\n",
754 current, ep));
755 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800756
757free_uid:
758 free_uid(user);
759 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700760}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
762/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700763 * Search the file inside the eventpoll tree. The RB tree operations
764 * are protected by the "mtx" mutex, and ep_find() must be called with
765 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 */
767static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
768{
769 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 struct rb_node *rbp;
771 struct epitem *epi, *epir = NULL;
772 struct epoll_filefd ffd;
773
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700774 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 for (rbp = ep->rbr.rb_node; rbp; ) {
776 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700777 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 if (kcmp > 0)
779 rbp = rbp->rb_right;
780 else if (kcmp < 0)
781 rbp = rbp->rb_left;
782 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 epir = epi;
784 break;
785 }
786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
788 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_find(%p) -> %p\n",
789 current, file, epir));
790
791 return epir;
792}
793
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700795 * This is the callback that is passed to the wait queue wakeup
796 * machanism. It is called by the stored file descriptors when they
797 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 */
Davide Libenzi7699acd2007-05-10 22:23:23 -0700799static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700801 int pwake = 0;
802 unsigned long flags;
803 struct epitem *epi = ep_item_from_wait(wait);
804 struct eventpoll *ep = epi->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Davide Libenzi7699acd2007-05-10 22:23:23 -0700806 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: poll_callback(%p) epi=%p ep=%p\n",
807 current, epi->ffd.file, epi, ep));
808
Davide Libenzic7ea7632007-05-15 01:40:47 -0700809 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700810
811 /*
812 * If the event mask does not contain any poll(2) event, we consider the
813 * descriptor to be disabled. This condition is likely the effect of the
814 * EPOLLONESHOT bit that disables the descriptor when an event is received,
815 * until the next EPOLL_CTL_MOD will be issued.
816 */
817 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -0700818 goto out_unlock;
819
820 /*
821 * If we are trasfering events to userspace, we can hold no locks
822 * (because we're accessing user memory, and because of linux f_op->poll()
823 * semantics). All the events that happens during that period of time are
824 * chained in ep->ovflist and requeued later on.
825 */
826 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
827 if (epi->next == EP_UNACTIVE_PTR) {
828 epi->next = ep->ovflist;
829 ep->ovflist = epi;
830 }
831 goto out_unlock;
832 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700833
834 /* If this file is already in the ready list we exit soon */
Davide Libenzi5071f972009-03-31 15:24:10 -0700835 if (!ep_is_linked(&epi->rdllink))
836 list_add_tail(&epi->rdllink, &ep->rdllist);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700837
Davide Libenzi7699acd2007-05-10 22:23:23 -0700838 /*
839 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
840 * wait list.
841 */
842 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400843 wake_up_locked(&ep->wq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700844 if (waitqueue_active(&ep->poll_wait))
845 pwake++;
846
Davide Libenzid47de162007-05-15 01:40:41 -0700847out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -0700848 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700849
850 /* We have to call this outside the lock */
851 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700852 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700853
854 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855}
856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857/*
858 * This is the callback that is used to add our wait queue to the
859 * target file wakeup lists.
860 */
861static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
862 poll_table *pt)
863{
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700864 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 struct eppoll_entry *pwq;
866
Christoph Lametere94b1762006-12-06 20:33:17 -0800867 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
869 pwq->whead = whead;
870 pwq->base = epi;
871 add_wait_queue(whead, &pwq->wait);
872 list_add_tail(&pwq->llink, &epi->pwqlist);
873 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -0700874 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 /* We have to signal that an error occurred */
876 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -0700877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
881{
882 int kcmp;
883 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
884 struct epitem *epic;
885
886 while (*p) {
887 parent = *p;
888 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700889 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (kcmp > 0)
891 p = &parent->rb_right;
892 else
893 p = &parent->rb_left;
894 }
895 rb_link_node(&epi->rbn, parent, p);
896 rb_insert_color(&epi->rbn, &ep->rbr);
897}
898
Davide Libenzic7ea7632007-05-15 01:40:47 -0700899/*
900 * Must be called with "mtx" held.
901 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
903 struct file *tfile, int fd)
904{
905 int error, revents, pwake = 0;
906 unsigned long flags;
907 struct epitem *epi;
908 struct ep_pqueue epq;
909
Davide Libenzi7ef99642008-12-01 13:13:55 -0800910 if (unlikely(atomic_read(&ep->user->epoll_watches) >=
911 max_user_watches))
912 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -0800913 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -0800914 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 INIT_LIST_HEAD(&epi->rdllink);
918 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 INIT_LIST_HEAD(&epi->pwqlist);
920 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700921 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -0700924 epi->next = EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926 /* Initialize the poll table using the queue callback */
927 epq.epi = epi;
928 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
929
930 /*
931 * Attach the item to the poll hooks and get current event bits.
932 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -0700933 * been increased by the caller of this function. Note that after
934 * this operation completes, the poll callback can start hitting
935 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 */
937 revents = tfile->f_op->poll(tfile, &epq.pt);
938
939 /*
940 * We have to check if something went wrong during the poll wait queue
941 * install process. Namely an allocation for a wait queue failed due
942 * high memory pressure.
943 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800944 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700946 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -0700949 spin_lock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 list_add_tail(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -0700951 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Davide Libenzic7ea7632007-05-15 01:40:47 -0700953 /*
954 * Add the current item to the RB tree. All RB tree operations are
955 * protected by "mtx", and ep_insert() is called with "mtx" held.
956 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 ep_rbtree_insert(ep, epi);
958
Davide Libenzic7ea7632007-05-15 01:40:47 -0700959 /* We have to drop the new item inside our item list to keep track of it */
960 spin_lock_irqsave(&ep->lock, flags);
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700963 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 list_add_tail(&epi->rdllink, &ep->rdllist);
965
966 /* Notify waiting tasks that events are available */
967 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -0400968 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 if (waitqueue_active(&ep->poll_wait))
970 pwake++;
971 }
972
Davide Libenzic7ea7632007-05-15 01:40:47 -0700973 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Davide Libenzi7ef99642008-12-01 13:13:55 -0800975 atomic_inc(&ep->user->epoll_watches);
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 /* We have to call this outside the lock */
978 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -0700979 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_insert(%p, %p, %d)\n",
982 current, ep, tfile, fd));
983
984 return 0;
985
Davide Libenzi7699acd2007-05-10 22:23:23 -0700986error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 ep_unregister_pollwait(ep, epi);
988
989 /*
990 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -0700991 * allocated wait queue. Note that we don't care about the ep->ovflist
992 * list, since that is used/cleaned only inside a section bound by "mtx".
993 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700995 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700996 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -0700997 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -0700998 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001000 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 return error;
1003}
1004
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005/*
1006 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001007 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 */
1009static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1010{
1011 int pwake = 0;
1012 unsigned int revents;
1013 unsigned long flags;
1014
1015 /*
1016 * Set the new event interest mask before calling f_op->poll(), otherwise
1017 * a potential race might occur. In fact if we do this operation inside
1018 * the lock, an event might happen between the f_op->poll() call and the
1019 * new event set registering.
1020 */
1021 epi->event.events = event->events;
1022
1023 /*
1024 * Get current event bits. We can safely use the file* here because
1025 * its usage count has been increased by the caller of this function.
1026 */
1027 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL);
1028
Davide Libenzic7ea7632007-05-15 01:40:47 -07001029 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
1031 /* Copy the data member from inside the lock */
1032 epi->event.data = event->data;
1033
1034 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001035 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001036 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001038 if (revents & event->events) {
1039 if (!ep_is_linked(&epi->rdllink)) {
1040 list_add_tail(&epi->rdllink, &ep->rdllist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
Davide Libenzic7ea7632007-05-15 01:40:47 -07001042 /* Notify waiting tasks that events are available */
1043 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e2c2007-08-30 16:10:22 -04001044 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001045 if (waitqueue_active(&ep->poll_wait))
1046 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
1048 }
Davide Libenzic7ea7632007-05-15 01:40:47 -07001049 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
1051 /* We have to call this outside the lock */
1052 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001053 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055 return 0;
1056}
1057
Davide Libenzi296e2362009-03-31 15:24:11 -07001058static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1059 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001060{
1061 struct ep_send_events_data *esed = priv;
1062 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001063 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001064 struct epitem *epi;
1065 struct epoll_event __user *uevent;
1066
Davide Libenzi296e2362009-03-31 15:24:11 -07001067 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001068 * We can loop without lock because we are passed a task private list.
1069 * Items cannot vanish during the loop because ep_scan_ready_list() is
1070 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001071 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001072 for (eventcnt = 0, uevent = esed->events;
1073 !list_empty(head) && eventcnt < esed->maxevents;) {
1074 epi = list_first_entry(head, struct epitem, rdllink);
1075
1076 list_del_init(&epi->rdllink);
1077
Davide Libenzi296e2362009-03-31 15:24:11 -07001078 revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) &
1079 epi->event.events;
Davide Libenzi5071f972009-03-31 15:24:10 -07001080
Davide Libenzi296e2362009-03-31 15:24:11 -07001081 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001082 * If the event mask intersect the caller-requested one,
1083 * deliver the event to userspace. Again, ep_scan_ready_list()
1084 * is holding "mtx", so no operations coming from userspace
1085 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001086 */
1087 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001088 if (__put_user(revents, &uevent->events) ||
1089 __put_user(epi->event.data, &uevent->data))
Davide Libenzi296e2362009-03-31 15:24:11 -07001090 return eventcnt ? eventcnt : -EFAULT;
1091 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001092 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001093 if (epi->event.events & EPOLLONESHOT)
1094 epi->event.events &= EP_PRIVATE_BITS;
1095 else if (!(epi->event.events & EPOLLET)) {
1096 /*
1097 * If this file has been added with Level
1098 * Trigger mode, we need to insert back inside
1099 * the ready list, so that the next call to
1100 * epoll_wait() will check again the events
1101 * availability. At this point, noone can insert
1102 * into ep->rdllist besides us. The epoll_ctl()
1103 * callers are locked out by
1104 * ep_scan_ready_list() holding "mtx" and the
1105 * poll callback will queue them in ep->ovflist.
1106 */
1107 list_add_tail(&epi->rdllink, &ep->rdllist);
1108 }
1109 }
1110 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001111
1112 return eventcnt;
1113}
1114
Davide Libenzi296e2362009-03-31 15:24:11 -07001115static int ep_send_events(struct eventpoll *ep,
1116 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
Davide Libenzi5071f972009-03-31 15:24:10 -07001118 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001119
Davide Libenzi5071f972009-03-31 15:24:10 -07001120 esed.maxevents = maxevents;
1121 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001122
Davide Libenzi5071f972009-03-31 15:24:10 -07001123 return ep_scan_ready_list(ep, ep_send_events_proc, &esed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124}
1125
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1127 int maxevents, long timeout)
1128{
1129 int res, eavail;
1130 unsigned long flags;
1131 long jtimeout;
1132 wait_queue_t wait;
1133
1134 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001135 * Calculate the timeout by checking for the "infinite" value (-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 * and the overflow condition. The passed timeout is in milliseconds,
1137 * that why (t * HZ) / 1000.
1138 */
Davide Libenzie3306dd2005-09-27 21:45:33 -07001139 jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ?
1140 MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
1142retry:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001143 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
1145 res = 0;
1146 if (list_empty(&ep->rdllist)) {
1147 /*
1148 * We don't have any available event to return to the caller.
1149 * We need to sleep here, and we will be wake up by
1150 * ep_poll_callback() when events will become available.
1151 */
1152 init_waitqueue_entry(&wait, current);
Davide Libenzid47de162007-05-15 01:40:41 -07001153 wait.flags |= WQ_FLAG_EXCLUSIVE;
Davide Libenzi3419b232006-06-25 05:48:14 -07001154 __add_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
1156 for (;;) {
1157 /*
1158 * We don't want to sleep if the ep_poll_callback() sends us
1159 * a wakeup in between. That's why we set the task state
1160 * to TASK_INTERRUPTIBLE before doing the checks.
1161 */
1162 set_current_state(TASK_INTERRUPTIBLE);
1163 if (!list_empty(&ep->rdllist) || !jtimeout)
1164 break;
1165 if (signal_pending(current)) {
1166 res = -EINTR;
1167 break;
1168 }
1169
Davide Libenzic7ea7632007-05-15 01:40:47 -07001170 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 jtimeout = schedule_timeout(jtimeout);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001172 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 }
Davide Libenzi3419b232006-06-25 05:48:14 -07001174 __remove_wait_queue(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
1176 set_current_state(TASK_RUNNING);
1177 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 /* Is it worth to try to dig for events ? */
Davide Libenzi5071f972009-03-31 15:24:10 -07001179 eavail = !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Davide Libenzic7ea7632007-05-15 01:40:47 -07001181 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183 /*
1184 * Try to transfer events to user space. In case we get 0 events and
1185 * there's still timeout left over, we go trying again in search of
1186 * more luck.
1187 */
1188 if (!res && eavail &&
Davide Libenzid47de162007-05-15 01:40:41 -07001189 !(res = ep_send_events(ep, events, maxevents)) && jtimeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 goto retry;
1191
1192 return res;
1193}
1194
Davide Libenzi7699acd2007-05-10 22:23:23 -07001195/*
Andrew Morton523723b2008-08-12 15:09:01 -07001196 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001197 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001198SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001199{
Davide Libenzi296e2362009-03-31 15:24:11 -07001200 int error, fd = -1;
1201 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001202
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001203 /* Check the EPOLL_* constant for consistency. */
1204 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1205
Davide Libenzi296e2362009-03-31 15:24:11 -07001206 if (flags & ~EPOLL_CLOEXEC)
1207 return -EINVAL;
1208
Davide Libenzi7699acd2007-05-10 22:23:23 -07001209 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d)\n",
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001210 current, flags));
Davide Libenzi7699acd2007-05-10 22:23:23 -07001211
1212 /*
Davide Libenzi296e2362009-03-31 15:24:11 -07001213 * Create the internal data structure ( "struct eventpoll" ).
Davide Libenzi7699acd2007-05-10 22:23:23 -07001214 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001215 error = ep_alloc(&ep);
Davide Libenzi296e2362009-03-31 15:24:11 -07001216 if (error < 0) {
1217 fd = error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001218 goto error_return;
Davide Libenzi296e2362009-03-31 15:24:11 -07001219 }
Davide Libenzi7699acd2007-05-10 22:23:23 -07001220
1221 /*
1222 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001223 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001224 */
Davide Libenzi296e2362009-03-31 15:24:11 -07001225 fd = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep,
1226 flags & O_CLOEXEC);
1227 if (fd < 0)
Al Viro2030a422008-02-23 06:46:49 -05001228 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001229
Al Viro2030a422008-02-23 06:46:49 -05001230error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07001231 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n",
Davide Libenzi296e2362009-03-31 15:24:11 -07001232 current, flags, fd));
Davide Libenzi7699acd2007-05-10 22:23:23 -07001233
Davide Libenzi296e2362009-03-31 15:24:11 -07001234 return fd;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001235}
1236
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001237SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001238{
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001239 if (size < 0)
1240 return -EINVAL;
1241
1242 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001243}
1244
Davide Libenzi7699acd2007-05-10 22:23:23 -07001245/*
1246 * The following function implements the controller interface for
1247 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001248 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001249 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001250SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1251 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001252{
1253 int error;
1254 struct file *file, *tfile;
1255 struct eventpoll *ep;
1256 struct epitem *epi;
1257 struct epoll_event epds;
1258
1259 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p)\n",
1260 current, epfd, op, fd, event));
1261
1262 error = -EFAULT;
1263 if (ep_op_has_event(op) &&
1264 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1265 goto error_return;
1266
1267 /* Get the "struct file *" for the eventpoll file */
1268 error = -EBADF;
1269 file = fget(epfd);
1270 if (!file)
1271 goto error_return;
1272
1273 /* Get the "struct file *" for the target file */
1274 tfile = fget(fd);
1275 if (!tfile)
1276 goto error_fput;
1277
1278 /* The target file descriptor must support poll */
1279 error = -EPERM;
1280 if (!tfile->f_op || !tfile->f_op->poll)
1281 goto error_tgt_fput;
1282
1283 /*
1284 * We have to check that the file structure underneath the file descriptor
1285 * the user passed to us _is_ an eventpoll file. And also we do not permit
1286 * adding an epoll file descriptor inside itself.
1287 */
1288 error = -EINVAL;
1289 if (file == tfile || !is_file_epoll(file))
1290 goto error_tgt_fput;
1291
1292 /*
1293 * At this point it is safe to assume that the "private_data" contains
1294 * our own data structure.
1295 */
1296 ep = file->private_data;
1297
Davide Libenzid47de162007-05-15 01:40:41 -07001298 mutex_lock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001299
Davide Libenzi67647d02007-05-15 01:40:52 -07001300 /*
1301 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1302 * above, we can be sure to be able to use the item looked up by
1303 * ep_find() till we release the mutex.
1304 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001305 epi = ep_find(ep, tfile, fd);
1306
1307 error = -EINVAL;
1308 switch (op) {
1309 case EPOLL_CTL_ADD:
1310 if (!epi) {
1311 epds.events |= POLLERR | POLLHUP;
1312
1313 error = ep_insert(ep, &epds, tfile, fd);
1314 } else
1315 error = -EEXIST;
1316 break;
1317 case EPOLL_CTL_DEL:
1318 if (epi)
1319 error = ep_remove(ep, epi);
1320 else
1321 error = -ENOENT;
1322 break;
1323 case EPOLL_CTL_MOD:
1324 if (epi) {
1325 epds.events |= POLLERR | POLLHUP;
1326 error = ep_modify(ep, epi, &epds);
1327 } else
1328 error = -ENOENT;
1329 break;
1330 }
Davide Libenzid47de162007-05-15 01:40:41 -07001331 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001332
1333error_tgt_fput:
1334 fput(tfile);
1335error_fput:
1336 fput(file);
1337error_return:
1338 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p) = %d\n",
1339 current, epfd, op, fd, event, error));
1340
1341 return error;
1342}
1343
1344/*
1345 * Implement the event wait interface for the eventpoll file. It is the kernel
1346 * part of the user space epoll_wait(2).
1347 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001348SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1349 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001350{
1351 int error;
1352 struct file *file;
1353 struct eventpoll *ep;
1354
1355 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d)\n",
1356 current, epfd, events, maxevents, timeout));
1357
1358 /* The maximum number of event must be greater than zero */
1359 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1360 return -EINVAL;
1361
1362 /* Verify that the area passed by the user is writeable */
1363 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) {
1364 error = -EFAULT;
1365 goto error_return;
1366 }
1367
1368 /* Get the "struct file *" for the eventpoll file */
1369 error = -EBADF;
1370 file = fget(epfd);
1371 if (!file)
1372 goto error_return;
1373
1374 /*
1375 * We have to check that the file structure underneath the fd
1376 * the user passed to us _is_ an eventpoll file.
1377 */
1378 error = -EINVAL;
1379 if (!is_file_epoll(file))
1380 goto error_fput;
1381
1382 /*
1383 * At this point it is safe to assume that the "private_data" contains
1384 * our own data structure.
1385 */
1386 ep = file->private_data;
1387
1388 /* Time to fish for events ... */
1389 error = ep_poll(ep, events, maxevents, timeout);
1390
1391error_fput:
1392 fput(file);
1393error_return:
1394 DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d) = %d\n",
1395 current, epfd, events, maxevents, timeout, error));
1396
1397 return error;
1398}
1399
Roland McGrathf3de2722008-04-30 00:53:09 -07001400#ifdef HAVE_SET_RESTORE_SIGMASK
Davide Libenzi7699acd2007-05-10 22:23:23 -07001401
1402/*
1403 * Implement the event wait interface for the eventpoll file. It is the kernel
1404 * part of the user space epoll_pwait(2).
1405 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001406SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1407 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1408 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001409{
1410 int error;
1411 sigset_t ksigmask, sigsaved;
1412
1413 /*
1414 * If the caller wants a certain signal mask to be set during the wait,
1415 * we apply it here.
1416 */
1417 if (sigmask) {
1418 if (sigsetsize != sizeof(sigset_t))
1419 return -EINVAL;
1420 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1421 return -EFAULT;
1422 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1423 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1424 }
1425
1426 error = sys_epoll_wait(epfd, events, maxevents, timeout);
1427
1428 /*
1429 * If we changed the signal mask, we need to restore the original one.
1430 * In case we've got a signal while waiting, we do not restore the
1431 * signal mask yet, and we allow do_signal() to deliver the signal on
1432 * the way back to userspace, before the signal mask is restored.
1433 */
1434 if (sigmask) {
1435 if (error == -EINTR) {
1436 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07001437 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001438 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001439 } else
1440 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1441 }
1442
1443 return error;
1444}
1445
Roland McGrathf3de2722008-04-30 00:53:09 -07001446#endif /* HAVE_SET_RESTORE_SIGMASK */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448static int __init eventpoll_init(void)
1449{
Davide Libenzi7ef99642008-12-01 13:13:55 -08001450 struct sysinfo si;
1451
1452 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08001453 /*
1454 * Allows top 4% of lomem to be allocated for epoll watches (per user).
1455 */
1456 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08001457 EP_ITEM_COST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07001460 ep_nested_calls_init(&poll_safewake_ncalls);
1461
1462 /* Initialize the structure used to perform file's f_op->poll() calls */
1463 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465 /* Allocates slab cache used to allocate "struct epitem" items */
1466 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
1467 0, SLAB_HWCACHE_ALIGN|EPI_SLAB_DEBUG|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09001468 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
1470 /* Allocates slab cache used to allocate "struct eppoll_entry" */
1471 pwq_cache = kmem_cache_create("eventpoll_pwq",
1472 sizeof(struct eppoll_entry), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001473 EPI_SLAB_DEBUG|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476}
Davide Libenzicea69242007-05-10 22:23:22 -07001477fs_initcall(eventpoll_init);