blob: a9c0bf8782f5337736df588ad736ae02efa185a0 [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>
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +020036#include <linux/device.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <asm/io.h>
39#include <asm/mman.h>
Arun Sharma600634972011-07-26 16:09:06 -070040#include <linux/atomic.h>
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -080041#include <linux/proc_fs.h>
42#include <linux/seq_file.h>
Al Viro35280bd2013-02-24 14:52:17 -050043#include <linux/compat.h>
Jason Baronae10b2b2013-11-12 15:10:16 -080044#include <linux/rculist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046/*
47 * LOCKING:
48 * There are three level of locking required by epoll :
49 *
Arjan van de Ven144efe32006-03-23 03:00:32 -080050 * 1) epmutex (mutex)
Davide Libenzic7ea7632007-05-15 01:40:47 -070051 * 2) ep->mtx (mutex)
52 * 3) ep->lock (spinlock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 *
54 * The acquire order is the one listed above, from 1 to 3.
55 * We need a spinlock (ep->lock) because we manipulate objects
56 * from inside the poll callback, that might be triggered from
57 * a wake_up() that in turn might be called from IRQ context.
58 * So we can't sleep inside the poll callback and hence we need
59 * a spinlock. During the event transfer loop (from kernel to
60 * user space) we could end up sleeping due a copy_to_user(), so
61 * we need a lock that will allow us to sleep. This lock is a
Davide Libenzid47de162007-05-15 01:40:41 -070062 * mutex (ep->mtx). It is acquired during the event transfer loop,
63 * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
64 * Then we also need a global mutex to serialize eventpoll_release_file()
65 * and ep_free().
66 * This mutex is acquired by ep_free() during the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * cleanup path and it is also acquired by eventpoll_release_file()
68 * if a file has been pushed inside an epoll set and it is then
Daniel Balutabf6a41d2011-01-30 23:42:29 +020069 * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
Davide Libenzi22bacca2011-02-25 14:44:12 -080070 * It is also acquired when inserting an epoll fd onto another epoll
71 * fd. We do this so that we walk the epoll tree and ensure that this
72 * insertion does not create a cycle of epoll file descriptors, which
73 * could lead to deadlock. We need a global mutex to prevent two
74 * simultaneous inserts (A into B and B into A) from racing and
75 * constructing a cycle without either insert observing that it is
76 * going to.
Nelson Elhaged8805e62011-10-31 17:13:14 -070077 * It is necessary to acquire multiple "ep->mtx"es at once in the
78 * case when one epoll fd is added to another. In this case, we
79 * always acquire the locks in the order of nesting (i.e. after
80 * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
81 * before e2->mtx). Since we disallow cycles of epoll file
82 * descriptors, this ensures that the mutexes are well-ordered. In
83 * order to communicate this nesting to lockdep, when walking a tree
84 * of epoll file descriptors, we use the current recursion depth as
85 * the lockdep subkey.
Davide Libenzid47de162007-05-15 01:40:41 -070086 * It is possible to drop the "ep->mtx" and to use the global
87 * mutex "epmutex" (together with "ep->lock") to have it working,
88 * but having "ep->mtx" will make the interface more scalable.
Arjan van de Ven144efe32006-03-23 03:00:32 -080089 * Events that require holding "epmutex" are very rare, while for
Davide Libenzid47de162007-05-15 01:40:41 -070090 * normal operations the epoll private "ep->mtx" will guarantee
91 * a better scalability.
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 */
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/* Epoll private bits inside the event mask */
Jason Barondf0108c2016-01-20 14:59:24 -080095#define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Jason Baronb6a515c2016-02-05 15:37:04 -080097#define EPOLLINOUT_BITS (POLLIN | POLLOUT)
98
99#define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | POLLERR | POLLHUP | \
100 EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE)
101
Davide Libenzi5071f972009-03-31 15:24:10 -0700102/* Maximum number of nesting allowed inside epoll sets */
103#define EP_MAX_NESTS 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
Davide Libenzib6119672006-10-11 01:21:44 -0700105#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
106
Davide Libenzid47de162007-05-15 01:40:41 -0700107#define EP_UNACTIVE_PTR ((void *) -1L)
108
Davide Libenzi7ef99642008-12-01 13:13:55 -0800109#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111struct epoll_filefd {
112 struct file *file;
113 int fd;
Eric Wong39732ca2013-04-30 15:27:38 -0700114} __packed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700117 * Structure used to track possible nested calls, for too deep recursions
118 * and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700120struct nested_call_node {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 struct list_head llink;
Davide Libenzi5071f972009-03-31 15:24:10 -0700122 void *cookie;
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700123 void *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124};
125
126/*
Davide Libenzi5071f972009-03-31 15:24:10 -0700127 * This structure is used as collector for nested calls, to check for
128 * maximum recursion dept and loop cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700130struct nested_calls {
131 struct list_head tasks_call_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 spinlock_t lock;
133};
134
135/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 * Each file descriptor added to the eventpoll interface will
Davide Libenzi6192bd52007-05-08 00:25:41 -0700137 * have an entry of this type linked to the "rbr" RB tree.
Eric Wong39732ca2013-04-30 15:27:38 -0700138 * Avoid increasing the size of this struct, there can be many thousands
139 * of these on a server and we do not want this to take another cache line.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 */
141struct epitem {
Jason Baronae10b2b2013-11-12 15:10:16 -0800142 union {
143 /* RB tree node links this structure to the eventpoll RB tree */
144 struct rb_node rbn;
145 /* Used to free the struct epitem */
146 struct rcu_head rcu;
147 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
149 /* List header used to link this structure to the eventpoll ready list */
150 struct list_head rdllink;
151
Davide Libenzic7ea7632007-05-15 01:40:47 -0700152 /*
153 * Works together "struct eventpoll"->ovflist in keeping the
154 * single linked chain of items.
155 */
156 struct epitem *next;
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 /* The file descriptor information this item refers to */
159 struct epoll_filefd ffd;
160
161 /* Number of active wait queue attached to poll operations */
162 int nwait;
163
164 /* List containing poll wait queues */
165 struct list_head pwqlist;
166
167 /* The "container" of this item */
168 struct eventpoll *ep;
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 /* List header used to link this item to the "struct file" items list */
171 struct list_head fllink;
Davide Libenzid47de162007-05-15 01:40:41 -0700172
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +0200173 /* wakeup_source used when EPOLLWAKEUP is set */
Eric Wongeea1d582013-04-30 15:27:39 -0700174 struct wakeup_source __rcu *ws;
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +0200175
Davide Libenzic7ea7632007-05-15 01:40:47 -0700176 /* The structure that describe the interested events and the source fd */
177 struct epoll_event event;
Davide Libenzid47de162007-05-15 01:40:41 -0700178};
179
180/*
181 * This structure is stored inside the "private_data" member of the file
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200182 * structure and represents the main data structure for the eventpoll
Davide Libenzid47de162007-05-15 01:40:41 -0700183 * interface.
184 */
185struct eventpoll {
Daniel Balutabf6a41d2011-01-30 23:42:29 +0200186 /* Protect the access to this structure */
Davide Libenzic7ea7632007-05-15 01:40:47 -0700187 spinlock_t lock;
Davide Libenzid47de162007-05-15 01:40:41 -0700188
189 /*
190 * This mutex is used to ensure that files are not removed
191 * while epoll is using them. This is held during the event
192 * collection loop, the file cleanup path, the epoll file exit
193 * code and the ctl operations.
194 */
195 struct mutex mtx;
196
197 /* Wait queue used by sys_epoll_wait() */
198 wait_queue_head_t wq;
199
200 /* Wait queue used by file->poll() */
201 wait_queue_head_t poll_wait;
202
203 /* List of ready file descriptors */
204 struct list_head rdllist;
205
Davide Libenzi67647d02007-05-15 01:40:52 -0700206 /* RB tree root used to store monitored fd structs */
Davide Libenzid47de162007-05-15 01:40:41 -0700207 struct rb_root rbr;
208
209 /*
210 * This is a single linked list that chains all the "struct epitem" that
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300211 * happened while transferring ready events to userspace w/out
Davide Libenzid47de162007-05-15 01:40:41 -0700212 * holding ->lock.
213 */
214 struct epitem *ovflist;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800215
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +0200216 /* wakeup_source used when ep_scan_ready_list is running */
217 struct wakeup_source *ws;
218
Davide Libenzi7ef99642008-12-01 13:13:55 -0800219 /* The user that created the eventpoll descriptor */
220 struct user_struct *user;
Jason Baron28d82dc2012-01-12 17:17:43 -0800221
222 struct file *file;
223
224 /* used to optimize loop detection check */
225 int visited;
226 struct list_head visited_list_link;
Davide Libenzid47de162007-05-15 01:40:41 -0700227};
228
229/* Wait structure used by the poll hooks */
230struct eppoll_entry {
231 /* List header used to link this structure to the "struct epitem" */
232 struct list_head llink;
233
234 /* The "base" pointer is set to the container "struct epitem" */
Tony Battersby4f0989d2009-03-31 15:24:16 -0700235 struct epitem *base;
Davide Libenzid47de162007-05-15 01:40:41 -0700236
237 /*
238 * Wait queue item that will be linked to the target file wait
239 * queue head.
240 */
241 wait_queue_t wait;
242
243 /* The wait queue head that linked the "wait" wait queue item */
244 wait_queue_head_t *whead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245};
246
247/* Wrapper struct used by poll queueing */
248struct ep_pqueue {
249 poll_table pt;
250 struct epitem *epi;
251};
252
Davide Libenzi5071f972009-03-31 15:24:10 -0700253/* Used by the ep_send_events() function as callback private data */
254struct ep_send_events_data {
255 int maxevents;
256 struct epoll_event __user *events;
257};
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259/*
Davide Libenzi7ef99642008-12-01 13:13:55 -0800260 * Configuration options available inside /proc/sys/fs/epoll/
261 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800262/* Maximum number of epoll watched descriptors, per user */
Robin Holt52bd19f72011-01-12 17:00:01 -0800263static long max_user_watches __read_mostly;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800264
265/*
Davide Libenzid47de162007-05-15 01:40:41 -0700266 * This mutex is used to serialize ep_free() and eventpoll_release_file().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 */
Davide Libenzi7ef99642008-12-01 13:13:55 -0800268static DEFINE_MUTEX(epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Davide Libenzi22bacca2011-02-25 14:44:12 -0800270/* Used to check for epoll file descriptor inclusion loops */
271static struct nested_calls poll_loop_ncalls;
272
Davide Libenzi5071f972009-03-31 15:24:10 -0700273/* Used for safe wake up implementation */
274static struct nested_calls poll_safewake_ncalls;
275
276/* Used to call file's f_op->poll() under the nested calls boundaries */
277static struct nested_calls poll_readywalk_ncalls;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279/* Slab cache used to allocate "struct epitem" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800280static struct kmem_cache *epi_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282/* Slab cache used to allocate "struct eppoll_entry" */
Christoph Lametere18b8902006-12-06 20:33:20 -0800283static struct kmem_cache *pwq_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Jason Baron28d82dc2012-01-12 17:17:43 -0800285/* Visited nodes during ep_loop_check(), so we can unset them when we finish */
286static LIST_HEAD(visited_list);
287
288/*
289 * List of files with newly added links, where we may need to limit the number
290 * of emanating paths. Protected by the epmutex.
291 */
292static LIST_HEAD(tfile_check_list);
293
Davide Libenzi7ef99642008-12-01 13:13:55 -0800294#ifdef CONFIG_SYSCTL
295
296#include <linux/sysctl.h>
297
Robin Holt52bd19f72011-01-12 17:00:01 -0800298static long zero;
299static long long_max = LONG_MAX;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800300
Joe Perches1f7e0612014-06-06 14:38:05 -0700301struct ctl_table epoll_table[] = {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800302 {
Davide Libenzi7ef99642008-12-01 13:13:55 -0800303 .procname = "max_user_watches",
304 .data = &max_user_watches,
Robin Holt52bd19f72011-01-12 17:00:01 -0800305 .maxlen = sizeof(max_user_watches),
Davide Libenzi7ef99642008-12-01 13:13:55 -0800306 .mode = 0644,
Robin Holt52bd19f72011-01-12 17:00:01 -0800307 .proc_handler = proc_doulongvec_minmax,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800308 .extra1 = &zero,
Robin Holt52bd19f72011-01-12 17:00:01 -0800309 .extra2 = &long_max,
Davide Libenzi7ef99642008-12-01 13:13:55 -0800310 },
Eric W. Biedermanab092032009-11-05 14:25:10 -0800311 { }
Davide Libenzi7ef99642008-12-01 13:13:55 -0800312};
313#endif /* CONFIG_SYSCTL */
314
Jason Baron28d82dc2012-01-12 17:17:43 -0800315static const struct file_operations eventpoll_fops;
316
317static inline int is_file_epoll(struct file *f)
318{
319 return f->f_op == &eventpoll_fops;
320}
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700321
Davide Libenzi67647d02007-05-15 01:40:52 -0700322/* Setup the structure that is used as key for the RB tree */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700323static inline void ep_set_ffd(struct epoll_filefd *ffd,
324 struct file *file, int fd)
325{
326 ffd->file = file;
327 ffd->fd = fd;
328}
329
Davide Libenzi67647d02007-05-15 01:40:52 -0700330/* Compare RB tree keys */
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700331static inline int ep_cmp_ffd(struct epoll_filefd *p1,
332 struct epoll_filefd *p2)
333{
334 return (p1->file > p2->file ? +1:
335 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
336}
337
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700338/* Tells us if the item is currently linked */
339static inline int ep_is_linked(struct list_head *p)
340{
341 return !list_empty(p);
342}
343
Oleg Nesterov971316f2012-02-24 20:07:29 +0100344static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
345{
346 return container_of(p, struct eppoll_entry, wait);
347}
348
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700349/* Get the "struct epitem" from a wait queue pointer */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700350static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700351{
352 return container_of(p, struct eppoll_entry, wait)->base;
353}
354
355/* Get the "struct epitem" from an epoll queue wrapper */
Davide Libenzicdac75e2008-04-29 00:58:34 -0700356static inline struct epitem *ep_item_from_epqueue(poll_table *p)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700357{
358 return container_of(p, struct ep_pqueue, pt)->epi;
359}
360
361/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
Davide Libenzi6192bd52007-05-08 00:25:41 -0700362static inline int ep_op_has_event(int op)
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700363{
Andrew Mortona80a6b82012-11-08 15:53:35 -0800364 return op != EPOLL_CTL_DEL;
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700365}
366
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367/* Initialize the poll safe wake up structure */
Davide Libenzi5071f972009-03-31 15:24:10 -0700368static void ep_nested_calls_init(struct nested_calls *ncalls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
Davide Libenzi5071f972009-03-31 15:24:10 -0700370 INIT_LIST_HEAD(&ncalls->tasks_call_list);
371 spin_lock_init(&ncalls->lock);
372}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Davide Libenzi5071f972009-03-31 15:24:10 -0700374/**
Davide Libenzi3fb0e582011-03-22 16:34:46 -0700375 * ep_events_available - Checks if ready events might be available.
376 *
377 * @ep: Pointer to the eventpoll context.
378 *
379 * Returns: Returns a value different than zero if ready events are available,
380 * or zero otherwise.
381 */
382static inline int ep_events_available(struct eventpoll *ep)
383{
384 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
385}
386
387/**
Davide Libenzi5071f972009-03-31 15:24:10 -0700388 * ep_call_nested - Perform a bound (possibly) nested call, by checking
389 * that the recursion limit is not exceeded, and that
390 * the same nested call (by the meaning of same cookie) is
391 * no re-entered.
392 *
393 * @ncalls: Pointer to the nested_calls structure to be used for this call.
394 * @max_nests: Maximum number of allowed nesting calls.
395 * @nproc: Nested call core function pointer.
396 * @priv: Opaque data to be passed to the @nproc callback.
397 * @cookie: Cookie to be used to identify this nested call.
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700398 * @ctx: This instance context.
Davide Libenzi5071f972009-03-31 15:24:10 -0700399 *
400 * Returns: Returns the code returned by the @nproc callback, or -1 if
401 * the maximum recursion limit has been exceeded.
402 */
403static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
404 int (*nproc)(void *, void *, int), void *priv,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700405 void *cookie, void *ctx)
Davide Libenzi5071f972009-03-31 15:24:10 -0700406{
407 int error, call_nests = 0;
408 unsigned long flags;
Davide Libenzi5071f972009-03-31 15:24:10 -0700409 struct list_head *lsthead = &ncalls->tasks_call_list;
410 struct nested_call_node *tncur;
411 struct nested_call_node tnode;
412
413 spin_lock_irqsave(&ncalls->lock, flags);
414
415 /*
416 * Try to see if the current task is already inside this wakeup call.
417 * We use a list here, since the population inside this set is always
418 * very much limited.
419 */
420 list_for_each_entry(tncur, lsthead, llink) {
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700421 if (tncur->ctx == ctx &&
Davide Libenzi5071f972009-03-31 15:24:10 -0700422 (tncur->cookie == cookie || ++call_nests > max_nests)) {
423 /*
424 * Ops ... loop detected or maximum nest level reached.
425 * We abort this wake by breaking the cycle itself.
426 */
Tony Battersbyabff55c2009-03-31 15:24:13 -0700427 error = -1;
428 goto out_unlock;
Davide Libenzi5071f972009-03-31 15:24:10 -0700429 }
430 }
431
432 /* Add the current task and cookie to the list */
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700433 tnode.ctx = ctx;
Davide Libenzi5071f972009-03-31 15:24:10 -0700434 tnode.cookie = cookie;
435 list_add(&tnode.llink, lsthead);
436
437 spin_unlock_irqrestore(&ncalls->lock, flags);
438
439 /* Call the nested function */
440 error = (*nproc)(priv, cookie, call_nests);
441
442 /* Remove the current task from the list */
443 spin_lock_irqsave(&ncalls->lock, flags);
444 list_del(&tnode.llink);
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700445out_unlock:
Davide Libenzi5071f972009-03-31 15:24:10 -0700446 spin_unlock_irqrestore(&ncalls->lock, flags);
447
448 return error;
449}
450
Steven Rostedt02edc6f2012-03-23 15:02:27 -0700451/*
452 * As described in commit 0ccf831cb lockdep: annotate epoll
453 * the use of wait queues used by epoll is done in a very controlled
454 * manner. Wake ups can nest inside each other, but are never done
455 * with the same locking. For example:
456 *
457 * dfd = socket(...);
458 * efd1 = epoll_create();
459 * efd2 = epoll_create();
460 * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
461 * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
462 *
463 * When a packet arrives to the device underneath "dfd", the net code will
464 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
465 * callback wakeup entry on that queue, and the wake_up() performed by the
466 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
467 * (efd1) notices that it may have some event ready, so it needs to wake up
468 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
469 * that ends up in another wake_up(), after having checked about the
470 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
471 * avoid stack blasting.
472 *
473 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
474 * this special case of epoll.
475 */
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700476#ifdef CONFIG_DEBUG_LOCK_ALLOC
477static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
478 unsigned long events, int subclass)
479{
480 unsigned long flags;
481
482 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
483 wake_up_locked_poll(wqueue, events);
484 spin_unlock_irqrestore(&wqueue->lock, flags);
485}
486#else
487static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
488 unsigned long events, int subclass)
489{
490 wake_up_poll(wqueue, events);
491}
492#endif
493
Davide Libenzi5071f972009-03-31 15:24:10 -0700494static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
495{
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -0700496 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
497 1 + call_nests);
Davide Libenzi5071f972009-03-31 15:24:10 -0700498 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499}
500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501/*
502 * Perform a safe wake up of the poll wait list. The problem is that
503 * with the new callback'd wake up system, it is possible that the
504 * poll callback is reentered from inside the call to wake_up() done
505 * on the poll wait queue head. The rule is that we cannot reenter the
Davide Libenzi5071f972009-03-31 15:24:10 -0700506 * wake up code from the same task more than EP_MAX_NESTS times,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 * and we cannot reenter the same wait queue head at all. This will
508 * enable to have a hierarchy of epoll file descriptor of no more than
Davide Libenzi5071f972009-03-31 15:24:10 -0700509 * EP_MAX_NESTS deep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 */
Davide Libenzi5071f972009-03-31 15:24:10 -0700511static void ep_poll_safewake(wait_queue_head_t *wq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512{
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700513 int this_cpu = get_cpu();
514
Davide Libenzi5071f972009-03-31 15:24:10 -0700515 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
Davide Libenzi3fe4a972009-06-17 16:25:58 -0700516 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
517
518 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
520
Oleg Nesterov971316f2012-02-24 20:07:29 +0100521static void ep_remove_wait_queue(struct eppoll_entry *pwq)
522{
523 wait_queue_head_t *whead;
524
525 rcu_read_lock();
Oleg Nesterovd325f1f2017-09-01 18:55:33 +0200526 /*
527 * If it is cleared by POLLFREE, it should be rcu-safe.
528 * If we read NULL we need a barrier paired with
529 * smp_store_release() in ep_poll_callback(), otherwise
530 * we rely on whead->lock.
531 */
532 whead = smp_load_acquire(&pwq->whead);
Oleg Nesterov971316f2012-02-24 20:07:29 +0100533 if (whead)
534 remove_wait_queue(whead, &pwq->wait);
535 rcu_read_unlock();
536}
537
Davide Libenzi7699acd2007-05-10 22:23:23 -0700538/*
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700539 * This function unregisters poll callbacks from the associated file
540 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
541 * ep_free).
Davide Libenzi7699acd2007-05-10 22:23:23 -0700542 */
543static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
544{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700545 struct list_head *lsthead = &epi->pwqlist;
546 struct eppoll_entry *pwq;
547
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700548 while (!list_empty(lsthead)) {
549 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700550
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700551 list_del(&pwq->llink);
Oleg Nesterov971316f2012-02-24 20:07:29 +0100552 ep_remove_wait_queue(pwq);
Tony Battersbyd1bc90d2009-03-31 15:24:15 -0700553 kmem_cache_free(pwq_cache, pwq);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700554 }
555}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Eric Wongeea1d582013-04-30 15:27:39 -0700557/* call only when ep->mtx is held */
558static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
559{
560 return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
561}
562
563/* call only when ep->mtx is held */
564static inline void ep_pm_stay_awake(struct epitem *epi)
565{
566 struct wakeup_source *ws = ep_wakeup_source(epi);
567
568 if (ws)
569 __pm_stay_awake(ws);
570}
571
572static inline bool ep_has_wakeup_source(struct epitem *epi)
573{
574 return rcu_access_pointer(epi->ws) ? true : false;
575}
576
577/* call when ep->mtx cannot be held (ep_poll_callback) */
578static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
579{
580 struct wakeup_source *ws;
581
582 rcu_read_lock();
583 ws = rcu_dereference(epi->ws);
584 if (ws)
585 __pm_stay_awake(ws);
586 rcu_read_unlock();
587}
588
Davide Libenzi5071f972009-03-31 15:24:10 -0700589/**
590 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
591 * the scan code, to call f_op->poll(). Also allows for
592 * O(NumReady) performance.
593 *
594 * @ep: Pointer to the epoll private data structure.
595 * @sproc: Pointer to the scan callback.
596 * @priv: Private opaque data passed to the @sproc callback.
Nelson Elhaged8805e62011-10-31 17:13:14 -0700597 * @depth: The current depth of recursive f_op->poll calls.
Jason Baron67347fe2013-11-12 15:10:18 -0800598 * @ep_locked: caller already holds ep->mtx
Davide Libenzi5071f972009-03-31 15:24:10 -0700599 *
600 * Returns: The same integer error code returned by the @sproc callback.
601 */
602static int ep_scan_ready_list(struct eventpoll *ep,
603 int (*sproc)(struct eventpoll *,
604 struct list_head *, void *),
Jason Baron67347fe2013-11-12 15:10:18 -0800605 void *priv, int depth, bool ep_locked)
Davide Libenzi5071f972009-03-31 15:24:10 -0700606{
607 int error, pwake = 0;
608 unsigned long flags;
609 struct epitem *epi, *nepi;
Davide Libenzi296e2362009-03-31 15:24:11 -0700610 LIST_HEAD(txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700611
612 /*
613 * We need to lock this because we could be hit by
Tony Battersbye057e152009-03-31 15:24:15 -0700614 * eventpoll_release_file() and epoll_ctl().
Davide Libenzi5071f972009-03-31 15:24:10 -0700615 */
Jason Baron67347fe2013-11-12 15:10:18 -0800616
617 if (!ep_locked)
618 mutex_lock_nested(&ep->mtx, depth);
Davide Libenzi5071f972009-03-31 15:24:10 -0700619
620 /*
621 * Steal the ready list, and re-init the original one to the
622 * empty list. Also, set ep->ovflist to NULL so that events
623 * happening while looping w/out locks, are not lost. We cannot
624 * have the poll callback to queue directly on ep->rdllist,
625 * because we want the "sproc" callback to be able to do it
626 * in a lockless way.
627 */
628 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi296e2362009-03-31 15:24:11 -0700629 list_splice_init(&ep->rdllist, &txlist);
Davide Libenzi5071f972009-03-31 15:24:10 -0700630 ep->ovflist = NULL;
631 spin_unlock_irqrestore(&ep->lock, flags);
632
633 /*
634 * Now call the callback function.
635 */
636 error = (*sproc)(ep, &txlist, priv);
637
638 spin_lock_irqsave(&ep->lock, flags);
639 /*
640 * During the time we spent inside the "sproc" callback, some
641 * other events might have been queued by the poll callback.
642 * We re-insert them inside the main ready-list here.
643 */
644 for (nepi = ep->ovflist; (epi = nepi) != NULL;
645 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
646 /*
647 * We need to check if the item is already in the list.
648 * During the "sproc" callback execution time, items are
649 * queued into ->ovflist but the "txlist" might already
650 * contain them, and the list_splice() below takes care of them.
651 */
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +0200652 if (!ep_is_linked(&epi->rdllink)) {
Davide Libenzi5071f972009-03-31 15:24:10 -0700653 list_add_tail(&epi->rdllink, &ep->rdllist);
Eric Wongeea1d582013-04-30 15:27:39 -0700654 ep_pm_stay_awake(epi);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +0200655 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700656 }
657 /*
658 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
659 * releasing the lock, events will be queued in the normal way inside
660 * ep->rdllist.
661 */
662 ep->ovflist = EP_UNACTIVE_PTR;
663
664 /*
665 * Quickly re-inject items left on "txlist".
666 */
667 list_splice(&txlist, &ep->rdllist);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +0200668 __pm_relax(ep->ws);
Davide Libenzi5071f972009-03-31 15:24:10 -0700669
670 if (!list_empty(&ep->rdllist)) {
671 /*
Davide Libenzi296e2362009-03-31 15:24:11 -0700672 * Wake up (if active) both the eventpoll wait list and
673 * the ->poll() wait list (delayed after we release the lock).
Davide Libenzi5071f972009-03-31 15:24:10 -0700674 */
675 if (waitqueue_active(&ep->wq))
676 wake_up_locked(&ep->wq);
677 if (waitqueue_active(&ep->poll_wait))
678 pwake++;
679 }
680 spin_unlock_irqrestore(&ep->lock, flags);
681
Jason Baron67347fe2013-11-12 15:10:18 -0800682 if (!ep_locked)
683 mutex_unlock(&ep->mtx);
Davide Libenzi5071f972009-03-31 15:24:10 -0700684
685 /* We have to call this outside the lock */
686 if (pwake)
687 ep_poll_safewake(&ep->poll_wait);
688
689 return error;
690}
691
Jason Baronae10b2b2013-11-12 15:10:16 -0800692static void epi_rcu_free(struct rcu_head *head)
693{
694 struct epitem *epi = container_of(head, struct epitem, rcu);
695 kmem_cache_free(epi_cache, epi);
696}
697
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698/*
Davide Libenzi7699acd2007-05-10 22:23:23 -0700699 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
Davide Libenzic7ea7632007-05-15 01:40:47 -0700700 * all the associated resources. Must be called with "mtx" held.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700701 */
702static int ep_remove(struct eventpoll *ep, struct epitem *epi)
703{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700704 unsigned long flags;
705 struct file *file = epi->ffd.file;
706
707 /*
708 * Removes poll wait queue hooks. We _have_ to do this without holding
709 * the "ep->lock" otherwise a deadlock might occur. This because of the
710 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
711 * queue head lock when unregistering the wait queue. The wakeup callback
712 * will run by holding the wait queue head lock and will call our callback
713 * that will try to get "ep->lock".
714 */
715 ep_unregister_pollwait(ep, epi);
716
717 /* Remove the current item from the list of epoll hooks */
Jonathan Corbet68499912009-02-06 13:52:43 -0700718 spin_lock(&file->f_lock);
Jason Baronae10b2b2013-11-12 15:10:16 -0800719 list_del_rcu(&epi->fllink);
Jonathan Corbet68499912009-02-06 13:52:43 -0700720 spin_unlock(&file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Davide Libenzicdac75e2008-04-29 00:58:34 -0700722 rb_erase(&epi->rbn, &ep->rbr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Davide Libenzic7ea7632007-05-15 01:40:47 -0700724 spin_lock_irqsave(&ep->lock, flags);
725 if (ep_is_linked(&epi->rdllink))
726 list_del_init(&epi->rdllink);
727 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Eric Wongeea1d582013-04-30 15:27:39 -0700729 wakeup_source_unregister(ep_wakeup_source(epi));
Jason Baronae10b2b2013-11-12 15:10:16 -0800730 /*
731 * At this point it is safe to free the eventpoll item. Use the union
732 * field epi->rcu, since we are trying to minimize the size of
733 * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
734 * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
735 * use of the rbn field.
736 */
737 call_rcu(&epi->rcu, epi_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
Robin Holt52bd19f72011-01-12 17:00:01 -0800739 atomic_long_dec(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800740
Davide Libenzic7ea7632007-05-15 01:40:47 -0700741 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744static void ep_free(struct eventpoll *ep)
745{
746 struct rb_node *rbp;
747 struct epitem *epi;
748
749 /* We need to release all tasks waiting for these file */
750 if (waitqueue_active(&ep->poll_wait))
Davide Libenzi5071f972009-03-31 15:24:10 -0700751 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753 /*
754 * We need to lock this because we could be hit by
755 * eventpoll_release_file() while we're freeing the "struct eventpoll".
Davide Libenzid47de162007-05-15 01:40:41 -0700756 * We do not need to hold "ep->mtx" here because the epoll file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 * is on the way to be removed and no one has references to it
758 * anymore. The only hit might come from eventpoll_release_file() but
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300759 * holding "epmutex" is sufficient here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 */
Arjan van de Ven144efe32006-03-23 03:00:32 -0800761 mutex_lock(&epmutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
763 /*
764 * Walks through the whole tree by unregistering poll callbacks.
765 */
766 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
767 epi = rb_entry(rbp, struct epitem, rbn);
768
769 ep_unregister_pollwait(ep, epi);
Eric Dumazet91cf5ab2013-09-11 14:24:06 -0700770 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 }
772
773 /*
Davide Libenzi6192bd52007-05-08 00:25:41 -0700774 * Walks through the whole tree by freeing each "struct epitem". At this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 * point we are sure no poll callbacks will be lingering around, and also by
Davide Libenzid47de162007-05-15 01:40:41 -0700776 * holding "epmutex" we can be sure that no file cleanup code will hit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 * us during this operation. So we can avoid the lock on "ep->lock".
Eric Wongddf676c2013-04-30 15:27:40 -0700778 * We do not need to lock ep->mtx, either, we only do it to prevent
779 * a lockdep warning.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 */
Eric Wongddf676c2013-04-30 15:27:40 -0700781 mutex_lock(&ep->mtx);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700782 while ((rbp = rb_first(&ep->rbr)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 epi = rb_entry(rbp, struct epitem, rbn);
784 ep_remove(ep, epi);
Eric Dumazet91cf5ab2013-09-11 14:24:06 -0700785 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
Eric Wongddf676c2013-04-30 15:27:40 -0700787 mutex_unlock(&ep->mtx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Arjan van de Ven144efe32006-03-23 03:00:32 -0800789 mutex_unlock(&epmutex);
Davide Libenzid47de162007-05-15 01:40:41 -0700790 mutex_destroy(&ep->mtx);
Davide Libenzi7ef99642008-12-01 13:13:55 -0800791 free_uid(ep->user);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +0200792 wakeup_source_unregister(ep->ws);
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700793 kfree(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794}
795
Davide Libenzi7699acd2007-05-10 22:23:23 -0700796static int ep_eventpoll_release(struct inode *inode, struct file *file)
797{
798 struct eventpoll *ep = file->private_data;
799
Davide Libenzif0ee9aa2007-05-15 01:40:57 -0700800 if (ep)
Davide Libenzi7699acd2007-05-10 22:23:23 -0700801 ep_free(ep);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700802
Davide Libenzi7699acd2007-05-10 22:23:23 -0700803 return 0;
804}
805
Eric Wong450d89e2013-04-30 15:27:42 -0700806static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt)
807{
808 pt->_key = epi->event.events;
809
810 return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events;
811}
812
Davide Libenzi296e2362009-03-31 15:24:11 -0700813static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
814 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -0700815{
816 struct epitem *epi, *tmp;
Hans Verkuil626cf232012-03-23 15:02:27 -0700817 poll_table pt;
Davide Libenzi5071f972009-03-31 15:24:10 -0700818
Hans Verkuil626cf232012-03-23 15:02:27 -0700819 init_poll_funcptr(&pt, NULL);
Eric Wong450d89e2013-04-30 15:27:42 -0700820
Davide Libenzi5071f972009-03-31 15:24:10 -0700821 list_for_each_entry_safe(epi, tmp, head, rdllink) {
Eric Wong450d89e2013-04-30 15:27:42 -0700822 if (ep_item_poll(epi, &pt))
Davide Libenzi5071f972009-03-31 15:24:10 -0700823 return POLLIN | POLLRDNORM;
Davide Libenzi296e2362009-03-31 15:24:11 -0700824 else {
Davide Libenzi5071f972009-03-31 15:24:10 -0700825 /*
826 * Item has been dropped into the ready list by the poll
827 * callback, but it's not actually ready, as far as
828 * caller requested events goes. We can remove it here.
829 */
Eric Wongeea1d582013-04-30 15:27:39 -0700830 __pm_relax(ep_wakeup_source(epi));
Davide Libenzi5071f972009-03-31 15:24:10 -0700831 list_del_init(&epi->rdllink);
Davide Libenzi296e2362009-03-31 15:24:11 -0700832 }
Davide Libenzi5071f972009-03-31 15:24:10 -0700833 }
834
835 return 0;
836}
837
Jason Baron67347fe2013-11-12 15:10:18 -0800838static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
839 poll_table *pt);
840
841struct readyevents_arg {
842 struct eventpoll *ep;
843 bool locked;
844};
845
Davide Libenzi5071f972009-03-31 15:24:10 -0700846static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
847{
Jason Baron67347fe2013-11-12 15:10:18 -0800848 struct readyevents_arg *arg = priv;
849
850 return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL,
851 call_nests + 1, arg->locked);
Davide Libenzi5071f972009-03-31 15:24:10 -0700852}
853
Davide Libenzi7699acd2007-05-10 22:23:23 -0700854static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
855{
Davide Libenzi5071f972009-03-31 15:24:10 -0700856 int pollflags;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700857 struct eventpoll *ep = file->private_data;
Jason Baron67347fe2013-11-12 15:10:18 -0800858 struct readyevents_arg arg;
859
860 /*
861 * During ep_insert() we already hold the ep->mtx for the tfile.
862 * Prevent re-aquisition.
863 */
864 arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc);
865 arg.ep = ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700866
867 /* Insert inside our poll wait queue */
868 poll_wait(file, &ep->poll_wait, wait);
869
Davide Libenzi5071f972009-03-31 15:24:10 -0700870 /*
871 * Proceed to find out if wanted events are really available inside
872 * the ready list. This need to be done under ep_call_nested()
873 * supervision, since the call to f_op->poll() done on listed files
874 * could re-enter here.
875 */
876 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
Jason Baron67347fe2013-11-12 15:10:18 -0800877 ep_poll_readyevents_proc, &arg, ep, current);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700878
Davide Libenzi296e2362009-03-31 15:24:11 -0700879 return pollflags != -1 ? pollflags : 0;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700880}
881
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800882#ifdef CONFIG_PROC_FS
Joe Perchesa3816ab2014-09-29 16:08:25 -0700883static void ep_show_fdinfo(struct seq_file *m, struct file *f)
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800884{
885 struct eventpoll *ep = f->private_data;
886 struct rb_node *rbp;
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800887
888 mutex_lock(&ep->mtx);
889 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
890 struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
891
Joe Perchesa3816ab2014-09-29 16:08:25 -0700892 seq_printf(m, "tfd: %8d events: %8x data: %16llx\n",
893 epi->ffd.fd, epi->event.events,
894 (long long)epi->event.data);
895 if (seq_has_overflowed(m))
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800896 break;
897 }
898 mutex_unlock(&ep->mtx);
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800899}
900#endif
901
Davide Libenzi7699acd2007-05-10 22:23:23 -0700902/* File callbacks that implement the eventpoll file behaviour */
903static const struct file_operations eventpoll_fops = {
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800904#ifdef CONFIG_PROC_FS
905 .show_fdinfo = ep_show_fdinfo,
906#endif
Davide Libenzi7699acd2007-05-10 22:23:23 -0700907 .release = ep_eventpoll_release,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200908 .poll = ep_eventpoll_poll,
909 .llseek = noop_llseek,
Davide Libenzi7699acd2007-05-10 22:23:23 -0700910};
911
Davide Libenzi7699acd2007-05-10 22:23:23 -0700912/*
913 * This is called from eventpoll_release() to unlink files from the eventpoll
914 * interface. We need to have this facility to cleanup correctly files that are
915 * closed without being removed from the eventpoll interface.
916 */
917void eventpoll_release_file(struct file *file)
918{
Davide Libenzi7699acd2007-05-10 22:23:23 -0700919 struct eventpoll *ep;
Konstantin Khlebnikovebe06182014-06-17 06:58:05 +0400920 struct epitem *epi, *next;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700921
922 /*
Jonathan Corbet68499912009-02-06 13:52:43 -0700923 * We don't want to get "file->f_lock" because it is not
Davide Libenzi7699acd2007-05-10 22:23:23 -0700924 * necessary. It is not necessary because we're in the "struct file"
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300925 * cleanup path, and this means that no one is using this file anymore.
Davide Libenzi5071f972009-03-31 15:24:10 -0700926 * So, for example, epoll_ctl() cannot hit here since if we reach this
Davide Libenzi67647d02007-05-15 01:40:52 -0700927 * point, the file counter already went to zero and fget() would fail.
Davide Libenzid47de162007-05-15 01:40:41 -0700928 * The only hit might come from ep_free() but by holding the mutex
Davide Libenzi7699acd2007-05-10 22:23:23 -0700929 * will correctly serialize the operation. We do need to acquire
Davide Libenzid47de162007-05-15 01:40:41 -0700930 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
Davide Libenzi7699acd2007-05-10 22:23:23 -0700931 * from anywhere but ep_free().
Jonathan Corbet68499912009-02-06 13:52:43 -0700932 *
933 * Besides, ep_remove() acquires the lock, so we can't hold it here.
Davide Libenzi7699acd2007-05-10 22:23:23 -0700934 */
935 mutex_lock(&epmutex);
Konstantin Khlebnikovebe06182014-06-17 06:58:05 +0400936 list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) {
Davide Libenzi7699acd2007-05-10 22:23:23 -0700937 ep = epi->ep;
Nelson Elhaged8805e62011-10-31 17:13:14 -0700938 mutex_lock_nested(&ep->mtx, 0);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700939 ep_remove(ep, epi);
Davide Libenzid47de162007-05-15 01:40:41 -0700940 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700941 }
Davide Libenzi7699acd2007-05-10 22:23:23 -0700942 mutex_unlock(&epmutex);
943}
944
945static int ep_alloc(struct eventpoll **pep)
946{
Davide Libenzi7ef99642008-12-01 13:13:55 -0800947 int error;
948 struct user_struct *user;
949 struct eventpoll *ep;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700950
Davide Libenzi7ef99642008-12-01 13:13:55 -0800951 user = get_current_user();
Davide Libenzi7ef99642008-12-01 13:13:55 -0800952 error = -ENOMEM;
953 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
954 if (unlikely(!ep))
955 goto free_uid;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700956
Davide Libenzic7ea7632007-05-15 01:40:47 -0700957 spin_lock_init(&ep->lock);
Davide Libenzid47de162007-05-15 01:40:41 -0700958 mutex_init(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -0700959 init_waitqueue_head(&ep->wq);
960 init_waitqueue_head(&ep->poll_wait);
961 INIT_LIST_HEAD(&ep->rdllist);
962 ep->rbr = RB_ROOT;
Davide Libenzid47de162007-05-15 01:40:41 -0700963 ep->ovflist = EP_UNACTIVE_PTR;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800964 ep->user = user;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700965
966 *pep = ep;
967
Davide Libenzi7699acd2007-05-10 22:23:23 -0700968 return 0;
Davide Libenzi7ef99642008-12-01 13:13:55 -0800969
970free_uid:
971 free_uid(user);
972 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -0700973}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975/*
Davide Libenzic7ea7632007-05-15 01:40:47 -0700976 * Search the file inside the eventpoll tree. The RB tree operations
977 * are protected by the "mtx" mutex, and ep_find() must be called with
978 * "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 */
980static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
981{
982 int kcmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 struct rb_node *rbp;
984 struct epitem *epi, *epir = NULL;
985 struct epoll_filefd ffd;
986
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700987 ep_set_ffd(&ffd, file, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 for (rbp = ep->rbr.rb_node; rbp; ) {
989 epi = rb_entry(rbp, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -0700990 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 if (kcmp > 0)
992 rbp = rbp->rb_right;
993 else if (kcmp < 0)
994 rbp = rbp->rb_left;
995 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 epir = epi;
997 break;
998 }
999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 return epir;
1002}
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004/*
Davide Libenzi7699acd2007-05-10 22:23:23 -07001005 * This is the callback that is passed to the wait queue wakeup
Daniel Balutabf6a41d2011-01-30 23:42:29 +02001006 * mechanism. It is called by the stored file descriptors when they
Davide Libenzi7699acd2007-05-10 22:23:23 -07001007 * have events to report.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 */
Davide Libenzi7699acd2007-05-10 22:23:23 -07001009static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010{
Davide Libenzi7699acd2007-05-10 22:23:23 -07001011 int pwake = 0;
1012 unsigned long flags;
1013 struct epitem *epi = ep_item_from_wait(wait);
1014 struct eventpoll *ep = epi->ep;
Jason Barondf0108c2016-01-20 14:59:24 -08001015 int ewake = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Davide Libenzic7ea7632007-05-15 01:40:47 -07001017 spin_lock_irqsave(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001018
1019 /*
1020 * If the event mask does not contain any poll(2) event, we consider the
1021 * descriptor to be disabled. This condition is likely the effect of the
1022 * EPOLLONESHOT bit that disables the descriptor when an event is received,
1023 * until the next EPOLL_CTL_MOD will be issued.
1024 */
1025 if (!(epi->event.events & ~EP_PRIVATE_BITS))
Davide Libenzid47de162007-05-15 01:40:41 -07001026 goto out_unlock;
1027
1028 /*
Davide Libenzi2dfa4ee2009-03-31 15:24:22 -07001029 * Check the events coming with the callback. At this stage, not
1030 * every device reports the events in the "key" parameter of the
1031 * callback. We need to be able to handle both cases here, hence the
1032 * test for "key" != NULL before the event match test.
1033 */
1034 if (key && !((unsigned long) key & epi->event.events))
1035 goto out_unlock;
1036
1037 /*
Daniel Balutabf6a41d2011-01-30 23:42:29 +02001038 * If we are transferring events to userspace, we can hold no locks
Davide Libenzid47de162007-05-15 01:40:41 -07001039 * (because we're accessing user memory, and because of linux f_op->poll()
Daniel Balutabf6a41d2011-01-30 23:42:29 +02001040 * semantics). All the events that happen during that period of time are
Davide Libenzid47de162007-05-15 01:40:41 -07001041 * chained in ep->ovflist and requeued later on.
1042 */
Davidlohr Buesocf96e9632019-01-03 15:27:09 -08001043 if (ep->ovflist != EP_UNACTIVE_PTR) {
Davide Libenzid47de162007-05-15 01:40:41 -07001044 if (epi->next == EP_UNACTIVE_PTR) {
1045 epi->next = ep->ovflist;
1046 ep->ovflist = epi;
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001047 if (epi->ws) {
1048 /*
1049 * Activate ep->ws since epi->ws may get
1050 * deactivated at any time.
1051 */
1052 __pm_stay_awake(ep->ws);
1053 }
1054
Davide Libenzid47de162007-05-15 01:40:41 -07001055 }
1056 goto out_unlock;
1057 }
Davide Libenzi7699acd2007-05-10 22:23:23 -07001058
1059 /* If this file is already in the ready list we exit soon */
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001060 if (!ep_is_linked(&epi->rdllink)) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001061 list_add_tail(&epi->rdllink, &ep->rdllist);
Eric Wongeea1d582013-04-30 15:27:39 -07001062 ep_pm_stay_awake_rcu(epi);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001063 }
Davide Libenzi7699acd2007-05-10 22:23:23 -07001064
Davide Libenzi7699acd2007-05-10 22:23:23 -07001065 /*
1066 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1067 * wait list.
1068 */
Jason Barondf0108c2016-01-20 14:59:24 -08001069 if (waitqueue_active(&ep->wq)) {
Jason Baronb6a515c2016-02-05 15:37:04 -08001070 if ((epi->event.events & EPOLLEXCLUSIVE) &&
1071 !((unsigned long)key & POLLFREE)) {
1072 switch ((unsigned long)key & EPOLLINOUT_BITS) {
1073 case POLLIN:
1074 if (epi->event.events & POLLIN)
1075 ewake = 1;
1076 break;
1077 case POLLOUT:
1078 if (epi->event.events & POLLOUT)
1079 ewake = 1;
1080 break;
1081 case 0:
1082 ewake = 1;
1083 break;
1084 }
1085 }
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -04001086 wake_up_locked(&ep->wq);
Jason Barondf0108c2016-01-20 14:59:24 -08001087 }
Davide Libenzi7699acd2007-05-10 22:23:23 -07001088 if (waitqueue_active(&ep->poll_wait))
1089 pwake++;
1090
Davide Libenzid47de162007-05-15 01:40:41 -07001091out_unlock:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001092 spin_unlock_irqrestore(&ep->lock, flags);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001093
1094 /* We have to call this outside the lock */
1095 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001096 ep_poll_safewake(&ep->poll_wait);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001097
Oleg Nesterovd325f1f2017-09-01 18:55:33 +02001098 if (!(epi->event.events & EPOLLEXCLUSIVE))
1099 ewake = 1;
Jason Barondf0108c2016-01-20 14:59:24 -08001100
Oleg Nesterovd325f1f2017-09-01 18:55:33 +02001101 if ((unsigned long)key & POLLFREE) {
1102 /*
1103 * If we race with ep_remove_wait_queue() it can miss
1104 * ->whead = NULL and do another remove_wait_queue() after
1105 * us, so we can't use __remove_wait_queue().
1106 */
1107 list_del_init(&wait->task_list);
1108 /*
1109 * ->whead != NULL protects us from the race with ep_free()
1110 * or ep_remove(), ep_remove_wait_queue() takes whead->lock
1111 * held by the caller. Once we nullify it, nothing protects
1112 * ep/epi or even wait.
1113 */
1114 smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL);
1115 }
1116
1117 return ewake;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120/*
1121 * This is the callback that is used to add our wait queue to the
1122 * target file wakeup lists.
1123 */
1124static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1125 poll_table *pt)
1126{
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001127 struct epitem *epi = ep_item_from_epqueue(pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 struct eppoll_entry *pwq;
1129
Christoph Lametere94b1762006-12-06 20:33:17 -08001130 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1132 pwq->whead = whead;
1133 pwq->base = epi;
Jason Barondf0108c2016-01-20 14:59:24 -08001134 if (epi->event.events & EPOLLEXCLUSIVE)
1135 add_wait_queue_exclusive(whead, &pwq->wait);
1136 else
1137 add_wait_queue(whead, &pwq->wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 list_add_tail(&pwq->llink, &epi->pwqlist);
1139 epi->nwait++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001140 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 /* We have to signal that an error occurred */
1142 epi->nwait = -1;
Davide Libenzi296e2362009-03-31 15:24:11 -07001143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144}
1145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1147{
1148 int kcmp;
1149 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
1150 struct epitem *epic;
1151
1152 while (*p) {
1153 parent = *p;
1154 epic = rb_entry(parent, struct epitem, rbn);
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001155 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (kcmp > 0)
1157 p = &parent->rb_right;
1158 else
1159 p = &parent->rb_left;
1160 }
1161 rb_link_node(&epi->rbn, parent, p);
1162 rb_insert_color(&epi->rbn, &ep->rbr);
1163}
1164
Andrew Mortona80a6b82012-11-08 15:53:35 -08001165
1166
Jason Baron28d82dc2012-01-12 17:17:43 -08001167#define PATH_ARR_SIZE 5
1168/*
1169 * These are the number paths of length 1 to 5, that we are allowing to emanate
1170 * from a single file of interest. For example, we allow 1000 paths of length
1171 * 1, to emanate from each file of interest. This essentially represents the
1172 * potential wakeup paths, which need to be limited in order to avoid massive
1173 * uncontrolled wakeup storms. The common use case should be a single ep which
1174 * is connected to n file sources. In this case each file source has 1 path
1175 * of length 1. Thus, the numbers below should be more than sufficient. These
1176 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1177 * and delete can't add additional paths. Protected by the epmutex.
1178 */
1179static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1180static int path_count[PATH_ARR_SIZE];
1181
1182static int path_count_inc(int nests)
1183{
Jason Baron93dc6102012-03-16 16:34:03 -04001184 /* Allow an arbitrary number of depth 1 paths */
1185 if (nests == 0)
1186 return 0;
1187
Jason Baron28d82dc2012-01-12 17:17:43 -08001188 if (++path_count[nests] > path_limits[nests])
1189 return -1;
1190 return 0;
1191}
1192
1193static void path_count_init(void)
1194{
1195 int i;
1196
1197 for (i = 0; i < PATH_ARR_SIZE; i++)
1198 path_count[i] = 0;
1199}
1200
1201static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1202{
1203 int error = 0;
1204 struct file *file = priv;
1205 struct file *child_file;
1206 struct epitem *epi;
1207
Jason Baronae10b2b2013-11-12 15:10:16 -08001208 /* CTL_DEL can remove links here, but that can't increase our count */
1209 rcu_read_lock();
1210 list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
Jason Baron28d82dc2012-01-12 17:17:43 -08001211 child_file = epi->ep->file;
1212 if (is_file_epoll(child_file)) {
1213 if (list_empty(&child_file->f_ep_links)) {
1214 if (path_count_inc(call_nests)) {
1215 error = -1;
1216 break;
1217 }
1218 } else {
1219 error = ep_call_nested(&poll_loop_ncalls,
1220 EP_MAX_NESTS,
1221 reverse_path_check_proc,
1222 child_file, child_file,
1223 current);
1224 }
1225 if (error != 0)
1226 break;
1227 } else {
1228 printk(KERN_ERR "reverse_path_check_proc: "
1229 "file is not an ep!\n");
1230 }
1231 }
Jason Baronae10b2b2013-11-12 15:10:16 -08001232 rcu_read_unlock();
Jason Baron28d82dc2012-01-12 17:17:43 -08001233 return error;
1234}
1235
1236/**
1237 * reverse_path_check - The tfile_check_list is list of file *, which have
1238 * links that are proposed to be newly added. We need to
1239 * make sure that those added links don't add too many
1240 * paths such that we will spend all our time waking up
1241 * eventpoll objects.
1242 *
1243 * Returns: Returns zero if the proposed links don't create too many paths,
1244 * -1 otherwise.
1245 */
1246static int reverse_path_check(void)
1247{
Jason Baron28d82dc2012-01-12 17:17:43 -08001248 int error = 0;
1249 struct file *current_file;
1250
1251 /* let's call this for all tfiles */
1252 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
Jason Baron28d82dc2012-01-12 17:17:43 -08001253 path_count_init();
1254 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1255 reverse_path_check_proc, current_file,
1256 current_file, current);
1257 if (error)
1258 break;
1259 }
1260 return error;
1261}
1262
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001263static int ep_create_wakeup_source(struct epitem *epi)
1264{
1265 const char *name;
Eric Wongeea1d582013-04-30 15:27:39 -07001266 struct wakeup_source *ws;
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001267
1268 if (!epi->ep->ws) {
1269 epi->ep->ws = wakeup_source_register("eventpoll");
1270 if (!epi->ep->ws)
1271 return -ENOMEM;
1272 }
1273
1274 name = epi->ffd.file->f_path.dentry->d_name.name;
Eric Wongeea1d582013-04-30 15:27:39 -07001275 ws = wakeup_source_register(name);
1276
1277 if (!ws)
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001278 return -ENOMEM;
Eric Wongeea1d582013-04-30 15:27:39 -07001279 rcu_assign_pointer(epi->ws, ws);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001280
1281 return 0;
1282}
1283
Eric Wongeea1d582013-04-30 15:27:39 -07001284/* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1285static noinline void ep_destroy_wakeup_source(struct epitem *epi)
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001286{
Eric Wongeea1d582013-04-30 15:27:39 -07001287 struct wakeup_source *ws = ep_wakeup_source(epi);
1288
Eric Wongd6d67e72013-04-30 15:27:43 -07001289 RCU_INIT_POINTER(epi->ws, NULL);
Eric Wongeea1d582013-04-30 15:27:39 -07001290
1291 /*
1292 * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1293 * used internally by wakeup_source_remove, too (called by
1294 * wakeup_source_unregister), so we cannot use call_rcu
1295 */
1296 synchronize_rcu();
1297 wakeup_source_unregister(ws);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001298}
1299
Davide Libenzic7ea7632007-05-15 01:40:47 -07001300/*
1301 * Must be called with "mtx" held.
1302 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
Jason Baron67347fe2013-11-12 15:10:18 -08001304 struct file *tfile, int fd, int full_check)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305{
1306 int error, revents, pwake = 0;
1307 unsigned long flags;
Robin Holt52bd19f72011-01-12 17:00:01 -08001308 long user_watches;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 struct epitem *epi;
1310 struct ep_pqueue epq;
1311
Robin Holt52bd19f72011-01-12 17:00:01 -08001312 user_watches = atomic_long_read(&ep->user->epoll_watches);
1313 if (unlikely(user_watches >= max_user_watches))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001314 return -ENOSPC;
Christoph Lametere94b1762006-12-06 20:33:17 -08001315 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
Davide Libenzi7ef99642008-12-01 13:13:55 -08001316 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 /* Item initialization follow here ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 INIT_LIST_HEAD(&epi->rdllink);
1320 INIT_LIST_HEAD(&epi->fllink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 INIT_LIST_HEAD(&epi->pwqlist);
1322 epi->ep = ep;
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001323 ep_set_ffd(&epi->ffd, tfile, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 epi->event = *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 epi->nwait = 0;
Davide Libenzid47de162007-05-15 01:40:41 -07001326 epi->next = EP_UNACTIVE_PTR;
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001327 if (epi->event.events & EPOLLWAKEUP) {
1328 error = ep_create_wakeup_source(epi);
1329 if (error)
1330 goto error_create_wakeup_source;
1331 } else {
Eric Wongeea1d582013-04-30 15:27:39 -07001332 RCU_INIT_POINTER(epi->ws, NULL);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 /* Initialize the poll table using the queue callback */
1336 epq.epi = epi;
1337 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1338
1339 /*
1340 * Attach the item to the poll hooks and get current event bits.
1341 * We can safely use the file* here because its usage count has
Davide Libenzic7ea7632007-05-15 01:40:47 -07001342 * been increased by the caller of this function. Note that after
1343 * this operation completes, the poll callback can start hitting
1344 * the new item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 */
Eric Wong450d89e2013-04-30 15:27:42 -07001346 revents = ep_item_poll(epi, &epq.pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348 /*
1349 * We have to check if something went wrong during the poll wait queue
1350 * install process. Namely an allocation for a wait queue failed due
1351 * high memory pressure.
1352 */
Davide Libenzi7ef99642008-12-01 13:13:55 -08001353 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 if (epi->nwait < 0)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001355 goto error_unregister;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
1357 /* Add the current item to the list of active epoll hook for this file */
Jonathan Corbet68499912009-02-06 13:52:43 -07001358 spin_lock(&tfile->f_lock);
Jason Baronae10b2b2013-11-12 15:10:16 -08001359 list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
Jonathan Corbet68499912009-02-06 13:52:43 -07001360 spin_unlock(&tfile->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Davide Libenzic7ea7632007-05-15 01:40:47 -07001362 /*
1363 * Add the current item to the RB tree. All RB tree operations are
1364 * protected by "mtx", and ep_insert() is called with "mtx" held.
1365 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 ep_rbtree_insert(ep, epi);
1367
Jason Baron28d82dc2012-01-12 17:17:43 -08001368 /* now check if we've created too many backpaths */
1369 error = -EINVAL;
Jason Baron67347fe2013-11-12 15:10:18 -08001370 if (full_check && reverse_path_check())
Jason Baron28d82dc2012-01-12 17:17:43 -08001371 goto error_remove_epi;
1372
Davide Libenzic7ea7632007-05-15 01:40:47 -07001373 /* We have to drop the new item inside our item list to keep track of it */
1374 spin_lock_irqsave(&ep->lock, flags);
1375
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 /* If the file is already "ready" we drop it inside the ready list */
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001377 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 list_add_tail(&epi->rdllink, &ep->rdllist);
Eric Wongeea1d582013-04-30 15:27:39 -07001379 ep_pm_stay_awake(epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 /* Notify waiting tasks that events are available */
1382 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -04001383 wake_up_locked(&ep->wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 if (waitqueue_active(&ep->poll_wait))
1385 pwake++;
1386 }
1387
Davide Libenzic7ea7632007-05-15 01:40:47 -07001388 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Robin Holt52bd19f72011-01-12 17:00:01 -08001390 atomic_long_inc(&ep->user->epoll_watches);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001391
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 /* We have to call this outside the lock */
1393 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001394 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 return 0;
1397
Jason Baron28d82dc2012-01-12 17:17:43 -08001398error_remove_epi:
1399 spin_lock(&tfile->f_lock);
Jason Baronae10b2b2013-11-12 15:10:16 -08001400 list_del_rcu(&epi->fllink);
Jason Baron28d82dc2012-01-12 17:17:43 -08001401 spin_unlock(&tfile->f_lock);
1402
1403 rb_erase(&epi->rbn, &ep->rbr);
1404
Davide Libenzi7699acd2007-05-10 22:23:23 -07001405error_unregister:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 ep_unregister_pollwait(ep, epi);
1407
1408 /*
1409 * We need to do this because an event could have been arrived on some
Davide Libenzi67647d02007-05-15 01:40:52 -07001410 * allocated wait queue. Note that we don't care about the ep->ovflist
1411 * list, since that is used/cleaned only inside a section bound by "mtx".
1412 * And ep_insert() is called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001414 spin_lock_irqsave(&ep->lock, flags);
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001415 if (ep_is_linked(&epi->rdllink))
Davide Libenzi6192bd52007-05-08 00:25:41 -07001416 list_del_init(&epi->rdllink);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001417 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Eric Wongeea1d582013-04-30 15:27:39 -07001419 wakeup_source_unregister(ep_wakeup_source(epi));
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001420
1421error_create_wakeup_source:
Pekka Enbergb030a4d2005-06-23 00:10:03 -07001422 kmem_cache_free(epi_cache, epi);
Davide Libenzi7ef99642008-12-01 13:13:55 -08001423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 return error;
1425}
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427/*
1428 * Modify the interest event mask by dropping an event if the new mask
Davide Libenzic7ea7632007-05-15 01:40:47 -07001429 * has a match in the current file status. Must be called with "mtx" held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 */
1431static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1432{
1433 int pwake = 0;
1434 unsigned int revents;
Hans Verkuil626cf232012-03-23 15:02:27 -07001435 poll_table pt;
1436
1437 init_poll_funcptr(&pt, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
1439 /*
Tony Battersbye057e152009-03-31 15:24:15 -07001440 * Set the new event interest mask before calling f_op->poll();
1441 * otherwise we might miss an event that happens between the
1442 * f_op->poll() call and the new event set registering.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 */
Eric Wong128dd172013-01-01 21:20:27 +00001444 epi->event.events = event->events; /* need barrier below */
Tony Battersbye057e152009-03-31 15:24:15 -07001445 epi->event.data = event->data; /* protected by mtx */
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001446 if (epi->event.events & EPOLLWAKEUP) {
Eric Wongeea1d582013-04-30 15:27:39 -07001447 if (!ep_has_wakeup_source(epi))
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001448 ep_create_wakeup_source(epi);
Eric Wongeea1d582013-04-30 15:27:39 -07001449 } else if (ep_has_wakeup_source(epi)) {
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001450 ep_destroy_wakeup_source(epi);
1451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 /*
Eric Wong128dd172013-01-01 21:20:27 +00001454 * The following barrier has two effects:
1455 *
1456 * 1) Flush epi changes above to other CPUs. This ensures
1457 * we do not miss events from ep_poll_callback if an
1458 * event occurs immediately after we call f_op->poll().
1459 * We need this because we did not take ep->lock while
1460 * changing epi above (but ep_poll_callback does take
1461 * ep->lock).
1462 *
1463 * 2) We also need to ensure we do not miss _past_ events
1464 * when calling f_op->poll(). This barrier also
1465 * pairs with the barrier in wq_has_sleeper (see
1466 * comments for wq_has_sleeper).
1467 *
1468 * This barrier will now guarantee ep_poll_callback or f_op->poll
1469 * (or both) will notice the readiness of an item.
1470 */
1471 smp_mb();
1472
1473 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 * Get current event bits. We can safely use the file* here because
1475 * its usage count has been increased by the caller of this function.
1476 */
Eric Wong450d89e2013-04-30 15:27:42 -07001477 revents = ep_item_poll(epi, &pt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 /*
Davide Libenzic7ea7632007-05-15 01:40:47 -07001480 * If the item is "hot" and it is not registered inside the ready
Davide Libenzi67647d02007-05-15 01:40:52 -07001481 * list, push it inside.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 */
Davide Libenzic7ea7632007-05-15 01:40:47 -07001483 if (revents & event->events) {
Tony Battersbye057e152009-03-31 15:24:15 -07001484 spin_lock_irq(&ep->lock);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001485 if (!ep_is_linked(&epi->rdllink)) {
1486 list_add_tail(&epi->rdllink, &ep->rdllist);
Eric Wongeea1d582013-04-30 15:27:39 -07001487 ep_pm_stay_awake(epi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Davide Libenzic7ea7632007-05-15 01:40:47 -07001489 /* Notify waiting tasks that events are available */
1490 if (waitqueue_active(&ep->wq))
Matthew Wilcox4a6e9e22007-08-30 16:10:22 -04001491 wake_up_locked(&ep->wq);
Davide Libenzic7ea7632007-05-15 01:40:47 -07001492 if (waitqueue_active(&ep->poll_wait))
1493 pwake++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 }
Tony Battersbye057e152009-03-31 15:24:15 -07001495 spin_unlock_irq(&ep->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498 /* We have to call this outside the lock */
1499 if (pwake)
Davide Libenzi5071f972009-03-31 15:24:10 -07001500 ep_poll_safewake(&ep->poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
1502 return 0;
1503}
1504
Davide Libenzi296e2362009-03-31 15:24:11 -07001505static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1506 void *priv)
Davide Libenzi5071f972009-03-31 15:24:10 -07001507{
1508 struct ep_send_events_data *esed = priv;
1509 int eventcnt;
Davide Libenzi296e2362009-03-31 15:24:11 -07001510 unsigned int revents;
Davide Libenzi5071f972009-03-31 15:24:10 -07001511 struct epitem *epi;
1512 struct epoll_event __user *uevent;
Eric Wongeea1d582013-04-30 15:27:39 -07001513 struct wakeup_source *ws;
Hans Verkuil626cf232012-03-23 15:02:27 -07001514 poll_table pt;
1515
1516 init_poll_funcptr(&pt, NULL);
Davide Libenzi5071f972009-03-31 15:24:10 -07001517
Davide Libenzi296e2362009-03-31 15:24:11 -07001518 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001519 * We can loop without lock because we are passed a task private list.
1520 * Items cannot vanish during the loop because ep_scan_ready_list() is
1521 * holding "mtx" during this call.
Davide Libenzi296e2362009-03-31 15:24:11 -07001522 */
Davide Libenzi5071f972009-03-31 15:24:10 -07001523 for (eventcnt = 0, uevent = esed->events;
1524 !list_empty(head) && eventcnt < esed->maxevents;) {
1525 epi = list_first_entry(head, struct epitem, rdllink);
1526
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001527 /*
1528 * Activate ep->ws before deactivating epi->ws to prevent
1529 * triggering auto-suspend here (in case we reactive epi->ws
1530 * below).
1531 *
1532 * This could be rearranged to delay the deactivation of epi->ws
1533 * instead, but then epi->ws would temporarily be out of sync
1534 * with ep_is_linked().
1535 */
Eric Wongeea1d582013-04-30 15:27:39 -07001536 ws = ep_wakeup_source(epi);
1537 if (ws) {
1538 if (ws->active)
1539 __pm_stay_awake(ep->ws);
1540 __pm_relax(ws);
1541 }
1542
Davide Libenzi5071f972009-03-31 15:24:10 -07001543 list_del_init(&epi->rdllink);
1544
Eric Wong450d89e2013-04-30 15:27:42 -07001545 revents = ep_item_poll(epi, &pt);
Davide Libenzi5071f972009-03-31 15:24:10 -07001546
Davide Libenzi296e2362009-03-31 15:24:11 -07001547 /*
Davide Libenzi5071f972009-03-31 15:24:10 -07001548 * If the event mask intersect the caller-requested one,
1549 * deliver the event to userspace. Again, ep_scan_ready_list()
1550 * is holding "mtx", so no operations coming from userspace
1551 * can change the item.
Davide Libenzi296e2362009-03-31 15:24:11 -07001552 */
1553 if (revents) {
Davide Libenzi5071f972009-03-31 15:24:10 -07001554 if (__put_user(revents, &uevent->events) ||
Tony Battersbyd0305882009-03-31 15:24:14 -07001555 __put_user(epi->event.data, &uevent->data)) {
1556 list_add(&epi->rdllink, head);
Eric Wongeea1d582013-04-30 15:27:39 -07001557 ep_pm_stay_awake(epi);
Davide Libenzi296e2362009-03-31 15:24:11 -07001558 return eventcnt ? eventcnt : -EFAULT;
Tony Battersbyd0305882009-03-31 15:24:14 -07001559 }
Davide Libenzi296e2362009-03-31 15:24:11 -07001560 eventcnt++;
Davide Libenzi5071f972009-03-31 15:24:10 -07001561 uevent++;
Davide Libenzi296e2362009-03-31 15:24:11 -07001562 if (epi->event.events & EPOLLONESHOT)
1563 epi->event.events &= EP_PRIVATE_BITS;
1564 else if (!(epi->event.events & EPOLLET)) {
1565 /*
1566 * If this file has been added with Level
1567 * Trigger mode, we need to insert back inside
1568 * the ready list, so that the next call to
1569 * epoll_wait() will check again the events
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001570 * availability. At this point, no one can insert
Davide Libenzi296e2362009-03-31 15:24:11 -07001571 * into ep->rdllist besides us. The epoll_ctl()
1572 * callers are locked out by
1573 * ep_scan_ready_list() holding "mtx" and the
1574 * poll callback will queue them in ep->ovflist.
1575 */
1576 list_add_tail(&epi->rdllink, &ep->rdllist);
Eric Wongeea1d582013-04-30 15:27:39 -07001577 ep_pm_stay_awake(epi);
Davide Libenzi296e2362009-03-31 15:24:11 -07001578 }
1579 }
1580 }
Davide Libenzi5071f972009-03-31 15:24:10 -07001581
1582 return eventcnt;
1583}
1584
Davide Libenzi296e2362009-03-31 15:24:11 -07001585static int ep_send_events(struct eventpoll *ep,
1586 struct epoll_event __user *events, int maxevents)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587{
Davide Libenzi5071f972009-03-31 15:24:10 -07001588 struct ep_send_events_data esed;
Davide Libenzi6192bd52007-05-08 00:25:41 -07001589
Davide Libenzi5071f972009-03-31 15:24:10 -07001590 esed.maxevents = maxevents;
1591 esed.events = events;
Davide Libenzid47de162007-05-15 01:40:41 -07001592
Jason Baron67347fe2013-11-12 15:10:18 -08001593 return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594}
1595
Deepa Dinamani766b9f92016-05-19 17:09:05 -07001596static inline struct timespec64 ep_set_mstimeout(long ms)
Eric Dumazet0781b902011-02-01 15:52:35 -08001597{
Deepa Dinamani766b9f92016-05-19 17:09:05 -07001598 struct timespec64 now, ts = {
Eric Dumazet0781b902011-02-01 15:52:35 -08001599 .tv_sec = ms / MSEC_PER_SEC,
1600 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1601 };
1602
Deepa Dinamani766b9f92016-05-19 17:09:05 -07001603 ktime_get_ts64(&now);
1604 return timespec64_add_safe(now, ts);
Eric Dumazet0781b902011-02-01 15:52:35 -08001605}
1606
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001607/**
1608 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1609 * event buffer.
1610 *
1611 * @ep: Pointer to the eventpoll context.
1612 * @events: Pointer to the userspace buffer where the ready events should be
1613 * stored.
1614 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1615 * @timeout: Maximum timeout for the ready events fetch operation, in
1616 * milliseconds. If the @timeout is zero, the function will not block,
1617 * while if the @timeout is less than zero, the function will block
1618 * until at least one event has been retrieved (or an error
1619 * occurred).
1620 *
1621 * Returns: Returns the number of ready events which have been fetched, or an
1622 * error code, in case of error.
1623 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1625 int maxevents, long timeout)
1626{
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001627 int res = 0, eavail, timed_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 unsigned long flags;
John Stultzda8b44d2016-03-17 14:20:51 -07001629 u64 slack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 wait_queue_t wait;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001631 ktime_t expires, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001633 if (timeout > 0) {
Deepa Dinamani766b9f92016-05-19 17:09:05 -07001634 struct timespec64 end_time = ep_set_mstimeout(timeout);
Eric Dumazet0781b902011-02-01 15:52:35 -08001635
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001636 slack = select_estimate_accuracy(&end_time);
1637 to = &expires;
Deepa Dinamani766b9f92016-05-19 17:09:05 -07001638 *to = timespec64_to_ktime(end_time);
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001639 } else if (timeout == 0) {
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001640 /*
1641 * Avoid the unnecessary trip to the wait queue loop, if the
1642 * caller specified a non blocking operation.
1643 */
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001644 timed_out = 1;
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001645 spin_lock_irqsave(&ep->lock, flags);
1646 goto check_events;
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001647 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001649fetch_events:
Davide Libenzic7ea7632007-05-15 01:40:47 -07001650 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001652 if (!ep_events_available(ep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 /*
1654 * We don't have any available event to return to the caller.
1655 * We need to sleep here, and we will be wake up by
1656 * ep_poll_callback() when events will become available.
1657 */
1658 init_waitqueue_entry(&wait, current);
Changli Gaoa93d2f12010-05-07 14:33:26 +08001659 __add_wait_queue_exclusive(&ep->wq, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661 for (;;) {
1662 /*
1663 * We don't want to sleep if the ep_poll_callback() sends us
1664 * a wakeup in between. That's why we set the task state
1665 * to TASK_INTERRUPTIBLE before doing the checks.
1666 */
1667 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001668 if (ep_events_available(ep) || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 break;
1670 if (signal_pending(current)) {
1671 res = -EINTR;
1672 break;
1673 }
1674
Davide Libenzic7ea7632007-05-15 01:40:47 -07001675 spin_unlock_irqrestore(&ep->lock, flags);
Rafael J. Wysockic5118512013-10-29 13:12:56 +01001676 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001677 timed_out = 1;
1678
Davide Libenzic7ea7632007-05-15 01:40:47 -07001679 spin_lock_irqsave(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
Davidlohr Bueso4d5755b2015-02-13 14:39:08 -08001682 __remove_wait_queue(&ep->wq, &wait);
1683 __set_current_state(TASK_RUNNING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001685check_events:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 /* Is it worth to try to dig for events ? */
Davide Libenzi3fb0e582011-03-22 16:34:46 -07001687 eavail = ep_events_available(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
Davide Libenzic7ea7632007-05-15 01:40:47 -07001689 spin_unlock_irqrestore(&ep->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691 /*
1692 * Try to transfer events to user space. In case we get 0 events and
1693 * there's still timeout left over, we go trying again in search of
1694 * more luck.
1695 */
1696 if (!res && eavail &&
Shawn Bohrer95aac7b2010-10-27 15:34:54 -07001697 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
Shawn Bohrerf4d93ad2011-03-22 16:34:47 -07001698 goto fetch_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
1700 return res;
1701}
1702
Davide Libenzi22bacca2011-02-25 14:44:12 -08001703/**
1704 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1705 * API, to verify that adding an epoll file inside another
1706 * epoll structure, does not violate the constraints, in
1707 * terms of closed loops, or too deep chains (which can
1708 * result in excessive stack usage).
1709 *
1710 * @priv: Pointer to the epoll file to be currently checked.
1711 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1712 * data structure pointer.
1713 * @call_nests: Current dept of the @ep_call_nested() call stack.
1714 *
1715 * Returns: Returns zero if adding the epoll @file inside current epoll
1716 * structure @ep does not violate the constraints, or -1 otherwise.
1717 */
1718static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1719{
1720 int error = 0;
1721 struct file *file = priv;
1722 struct eventpoll *ep = file->private_data;
Jason Baron28d82dc2012-01-12 17:17:43 -08001723 struct eventpoll *ep_tovisit;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001724 struct rb_node *rbp;
1725 struct epitem *epi;
1726
Nelson Elhaged8805e62011-10-31 17:13:14 -07001727 mutex_lock_nested(&ep->mtx, call_nests + 1);
Jason Baron28d82dc2012-01-12 17:17:43 -08001728 ep->visited = 1;
1729 list_add(&ep->visited_list_link, &visited_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001730 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1731 epi = rb_entry(rbp, struct epitem, rbn);
1732 if (unlikely(is_file_epoll(epi->ffd.file))) {
Jason Baron28d82dc2012-01-12 17:17:43 -08001733 ep_tovisit = epi->ffd.file->private_data;
1734 if (ep_tovisit->visited)
1735 continue;
Davide Libenzi22bacca2011-02-25 14:44:12 -08001736 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Jason Baron28d82dc2012-01-12 17:17:43 -08001737 ep_loop_check_proc, epi->ffd.file,
1738 ep_tovisit, current);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001739 if (error != 0)
1740 break;
Jason Baron28d82dc2012-01-12 17:17:43 -08001741 } else {
1742 /*
1743 * If we've reached a file that is not associated with
1744 * an ep, then we need to check if the newly added
1745 * links are going to add too many wakeup paths. We do
1746 * this by adding it to the tfile_check_list, if it's
1747 * not already there, and calling reverse_path_check()
1748 * during ep_insert().
1749 */
1750 if (list_empty(&epi->ffd.file->f_tfile_llink))
1751 list_add(&epi->ffd.file->f_tfile_llink,
1752 &tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001753 }
1754 }
1755 mutex_unlock(&ep->mtx);
1756
1757 return error;
1758}
1759
1760/**
1761 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1762 * another epoll file (represented by @ep) does not create
1763 * closed loops or too deep chains.
1764 *
1765 * @ep: Pointer to the epoll private data structure.
1766 * @file: Pointer to the epoll file to be checked.
1767 *
1768 * Returns: Returns zero if adding the epoll @file inside current epoll
1769 * structure @ep does not violate the constraints, or -1 otherwise.
1770 */
1771static int ep_loop_check(struct eventpoll *ep, struct file *file)
1772{
Jason Baron28d82dc2012-01-12 17:17:43 -08001773 int ret;
1774 struct eventpoll *ep_cur, *ep_next;
1775
1776 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
Davide Libenzi22bacca2011-02-25 14:44:12 -08001777 ep_loop_check_proc, file, ep, current);
Jason Baron28d82dc2012-01-12 17:17:43 -08001778 /* clear visited list */
1779 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1780 visited_list_link) {
1781 ep_cur->visited = 0;
1782 list_del(&ep_cur->visited_list_link);
1783 }
1784 return ret;
1785}
1786
1787static void clear_tfile_check_list(void)
1788{
1789 struct file *file;
1790
1791 /* first clear the tfile_check_list */
1792 while (!list_empty(&tfile_check_list)) {
1793 file = list_first_entry(&tfile_check_list, struct file,
1794 f_tfile_llink);
1795 list_del_init(&file->f_tfile_llink);
1796 }
1797 INIT_LIST_HEAD(&tfile_check_list);
Davide Libenzi22bacca2011-02-25 14:44:12 -08001798}
1799
Davide Libenzi7699acd2007-05-10 22:23:23 -07001800/*
Andrew Morton523723b2008-08-12 15:09:01 -07001801 * Open an eventpoll file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001802 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001803SYSCALL_DEFINE1(epoll_create1, int, flags)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001804{
Jason Baron28d82dc2012-01-12 17:17:43 -08001805 int error, fd;
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001806 struct eventpoll *ep = NULL;
Jason Baron28d82dc2012-01-12 17:17:43 -08001807 struct file *file;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001808
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001809 /* Check the EPOLL_* constant for consistency. */
1810 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1811
Davide Libenzi296e2362009-03-31 15:24:11 -07001812 if (flags & ~EPOLL_CLOEXEC)
1813 return -EINVAL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001814 /*
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001815 * Create the internal data structure ("struct eventpoll").
Davide Libenzi7699acd2007-05-10 22:23:23 -07001816 */
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001817 error = ep_alloc(&ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001818 if (error < 0)
1819 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001820 /*
1821 * Creates all the items needed to setup an eventpoll file. That is,
Al Viro2030a422008-02-23 06:46:49 -05001822 * a file structure and a free file descriptor.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001823 */
Jason Baron28d82dc2012-01-12 17:17:43 -08001824 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1825 if (fd < 0) {
1826 error = fd;
1827 goto out_free_ep;
1828 }
1829 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
Roland Dreier628ff7c2009-12-18 09:41:24 -08001830 O_RDWR | (flags & O_CLOEXEC));
Jason Baron28d82dc2012-01-12 17:17:43 -08001831 if (IS_ERR(file)) {
1832 error = PTR_ERR(file);
1833 goto out_free_fd;
1834 }
Jason Baron28d82dc2012-01-12 17:17:43 -08001835 ep->file = file;
Al Viro98022742012-08-17 22:42:36 -04001836 fd_install(fd, file);
Jason Baron28d82dc2012-01-12 17:17:43 -08001837 return fd;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001838
Jason Baron28d82dc2012-01-12 17:17:43 -08001839out_free_fd:
1840 put_unused_fd(fd);
1841out_free_ep:
1842 ep_free(ep);
Davide Libenzibb57c3e2009-03-31 15:24:12 -07001843 return error;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001844}
1845
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001846SYSCALL_DEFINE1(epoll_create, int, size)
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001847{
Davide Libenzibfe38912009-05-12 13:19:44 -07001848 if (size <= 0)
Ulrich Drepper9fe5ad92008-07-23 21:29:43 -07001849 return -EINVAL;
1850
1851 return sys_epoll_create1(0);
Ulrich Dreppera0998b52008-07-23 21:29:27 -07001852}
1853
Davide Libenzi7699acd2007-05-10 22:23:23 -07001854/*
1855 * The following function implements the controller interface for
1856 * the eventpoll file that enables the insertion/removal/change of
Davide Libenzi67647d02007-05-15 01:40:52 -07001857 * file descriptors inside the interest set.
Davide Libenzi7699acd2007-05-10 22:23:23 -07001858 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001859SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1860 struct epoll_event __user *, event)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001861{
1862 int error;
Jason Baron67347fe2013-11-12 15:10:18 -08001863 int full_check = 0;
Al Viro7e3fb582013-08-30 12:47:21 -04001864 struct fd f, tf;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001865 struct eventpoll *ep;
1866 struct epitem *epi;
1867 struct epoll_event epds;
Jason Baron67347fe2013-11-12 15:10:18 -08001868 struct eventpoll *tep = NULL;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001869
Davide Libenzi7699acd2007-05-10 22:23:23 -07001870 error = -EFAULT;
1871 if (ep_op_has_event(op) &&
1872 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1873 goto error_return;
1874
Davide Libenzi7699acd2007-05-10 22:23:23 -07001875 error = -EBADF;
Al Viro7e3fb582013-08-30 12:47:21 -04001876 f = fdget(epfd);
1877 if (!f.file)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001878 goto error_return;
1879
1880 /* Get the "struct file *" for the target file */
Al Viro7e3fb582013-08-30 12:47:21 -04001881 tf = fdget(fd);
1882 if (!tf.file)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001883 goto error_fput;
1884
1885 /* The target file descriptor must support poll */
1886 error = -EPERM;
Al Viro72c2d532013-09-22 16:27:52 -04001887 if (!tf.file->f_op->poll)
Davide Libenzi7699acd2007-05-10 22:23:23 -07001888 goto error_tgt_fput;
1889
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001890 /* Check if EPOLLWAKEUP is allowed */
Nicolas Ioossc680e412014-09-09 14:50:51 -07001891 if (ep_op_has_event(op))
1892 ep_take_care_of_epollwakeup(&epds);
Arve Hjønnevåg4d7e30d2012-05-01 21:33:34 +02001893
Davide Libenzi7699acd2007-05-10 22:23:23 -07001894 /*
1895 * We have to check that the file structure underneath the file descriptor
1896 * the user passed to us _is_ an eventpoll file. And also we do not permit
1897 * adding an epoll file descriptor inside itself.
1898 */
1899 error = -EINVAL;
Al Viro7e3fb582013-08-30 12:47:21 -04001900 if (f.file == tf.file || !is_file_epoll(f.file))
Davide Libenzi7699acd2007-05-10 22:23:23 -07001901 goto error_tgt_fput;
1902
1903 /*
Jason Barondf0108c2016-01-20 14:59:24 -08001904 * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only,
1905 * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation.
1906 * Also, we do not currently supported nested exclusive wakeups.
1907 */
Jason Baronb6a515c2016-02-05 15:37:04 -08001908 if (epds.events & EPOLLEXCLUSIVE) {
1909 if (op == EPOLL_CTL_MOD)
1910 goto error_tgt_fput;
1911 if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) ||
1912 (epds.events & ~EPOLLEXCLUSIVE_OK_BITS)))
1913 goto error_tgt_fput;
1914 }
Jason Barondf0108c2016-01-20 14:59:24 -08001915
1916 /*
Davide Libenzi7699acd2007-05-10 22:23:23 -07001917 * At this point it is safe to assume that the "private_data" contains
1918 * our own data structure.
1919 */
Al Viro7e3fb582013-08-30 12:47:21 -04001920 ep = f.file->private_data;
Davide Libenzi7699acd2007-05-10 22:23:23 -07001921
Davide Libenzi22bacca2011-02-25 14:44:12 -08001922 /*
1923 * When we insert an epoll file descriptor, inside another epoll file
1924 * descriptor, there is the change of creating closed loops, which are
Jason Baron28d82dc2012-01-12 17:17:43 -08001925 * better be handled here, than in more critical paths. While we are
1926 * checking for loops we also determine the list of files reachable
1927 * and hang them on the tfile_check_list, so we can check that we
1928 * haven't created too many possible wakeup paths.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001929 *
Jason Baron67347fe2013-11-12 15:10:18 -08001930 * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
1931 * the epoll file descriptor is attaching directly to a wakeup source,
1932 * unless the epoll file descriptor is nested. The purpose of taking the
1933 * 'epmutex' on add is to prevent complex toplogies such as loops and
1934 * deep wakeup paths from forming in parallel through multiple
1935 * EPOLL_CTL_ADD operations.
Davide Libenzi22bacca2011-02-25 14:44:12 -08001936 */
Nelson Elhaged8805e62011-10-31 17:13:14 -07001937 mutex_lock_nested(&ep->mtx, 0);
Jason Baron67347fe2013-11-12 15:10:18 -08001938 if (op == EPOLL_CTL_ADD) {
1939 if (!list_empty(&f.file->f_ep_links) ||
1940 is_file_epoll(tf.file)) {
1941 full_check = 1;
1942 mutex_unlock(&ep->mtx);
1943 mutex_lock(&epmutex);
1944 if (is_file_epoll(tf.file)) {
1945 error = -ELOOP;
1946 if (ep_loop_check(ep, tf.file) != 0) {
1947 clear_tfile_check_list();
1948 goto error_tgt_fput;
1949 }
1950 } else
1951 list_add(&tf.file->f_tfile_llink,
1952 &tfile_check_list);
1953 mutex_lock_nested(&ep->mtx, 0);
1954 if (is_file_epoll(tf.file)) {
1955 tep = tf.file->private_data;
1956 mutex_lock_nested(&tep->mtx, 1);
1957 }
1958 }
1959 }
Davide Libenzi7699acd2007-05-10 22:23:23 -07001960
Davide Libenzi67647d02007-05-15 01:40:52 -07001961 /*
1962 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1963 * above, we can be sure to be able to use the item looked up by
1964 * ep_find() till we release the mutex.
1965 */
Al Viro7e3fb582013-08-30 12:47:21 -04001966 epi = ep_find(ep, tf.file, fd);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001967
1968 error = -EINVAL;
1969 switch (op) {
1970 case EPOLL_CTL_ADD:
1971 if (!epi) {
1972 epds.events |= POLLERR | POLLHUP;
Jason Baron67347fe2013-11-12 15:10:18 -08001973 error = ep_insert(ep, &epds, tf.file, fd, full_check);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001974 } else
1975 error = -EEXIST;
Jason Baron67347fe2013-11-12 15:10:18 -08001976 if (full_check)
1977 clear_tfile_check_list();
Davide Libenzi7699acd2007-05-10 22:23:23 -07001978 break;
1979 case EPOLL_CTL_DEL:
1980 if (epi)
1981 error = ep_remove(ep, epi);
1982 else
1983 error = -ENOENT;
1984 break;
1985 case EPOLL_CTL_MOD:
1986 if (epi) {
Jason Baronb6a515c2016-02-05 15:37:04 -08001987 if (!(epi->event.events & EPOLLEXCLUSIVE)) {
1988 epds.events |= POLLERR | POLLHUP;
1989 error = ep_modify(ep, epi, &epds);
1990 }
Davide Libenzi7699acd2007-05-10 22:23:23 -07001991 } else
1992 error = -ENOENT;
1993 break;
1994 }
Jason Baron67347fe2013-11-12 15:10:18 -08001995 if (tep != NULL)
1996 mutex_unlock(&tep->mtx);
Davide Libenzid47de162007-05-15 01:40:41 -07001997 mutex_unlock(&ep->mtx);
Davide Libenzi7699acd2007-05-10 22:23:23 -07001998
1999error_tgt_fput:
Jason Baron67347fe2013-11-12 15:10:18 -08002000 if (full_check)
Davide Libenzi22bacca2011-02-25 14:44:12 -08002001 mutex_unlock(&epmutex);
2002
Al Viro7e3fb582013-08-30 12:47:21 -04002003 fdput(tf);
Davide Libenzi7699acd2007-05-10 22:23:23 -07002004error_fput:
Al Viro7e3fb582013-08-30 12:47:21 -04002005 fdput(f);
Davide Libenzi7699acd2007-05-10 22:23:23 -07002006error_return:
Davide Libenzi7699acd2007-05-10 22:23:23 -07002007
2008 return error;
2009}
2010
2011/*
2012 * Implement the event wait interface for the eventpoll file. It is the kernel
2013 * part of the user space epoll_wait(2).
2014 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01002015SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
2016 int, maxevents, int, timeout)
Davide Libenzi7699acd2007-05-10 22:23:23 -07002017{
Al Viro2903ff02012-08-28 12:52:22 -04002018 int error;
2019 struct fd f;
Davide Libenzi7699acd2007-05-10 22:23:23 -07002020 struct eventpoll *ep;
2021
Davide Libenzi7699acd2007-05-10 22:23:23 -07002022 /* The maximum number of event must be greater than zero */
2023 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
2024 return -EINVAL;
2025
2026 /* Verify that the area passed by the user is writeable */
Al Viro2903ff02012-08-28 12:52:22 -04002027 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
2028 return -EFAULT;
Davide Libenzi7699acd2007-05-10 22:23:23 -07002029
2030 /* Get the "struct file *" for the eventpoll file */
Al Viro2903ff02012-08-28 12:52:22 -04002031 f = fdget(epfd);
2032 if (!f.file)
2033 return -EBADF;
Davide Libenzi7699acd2007-05-10 22:23:23 -07002034
2035 /*
2036 * We have to check that the file structure underneath the fd
2037 * the user passed to us _is_ an eventpoll file.
2038 */
2039 error = -EINVAL;
Al Viro2903ff02012-08-28 12:52:22 -04002040 if (!is_file_epoll(f.file))
Davide Libenzi7699acd2007-05-10 22:23:23 -07002041 goto error_fput;
2042
2043 /*
2044 * At this point it is safe to assume that the "private_data" contains
2045 * our own data structure.
2046 */
Al Viro2903ff02012-08-28 12:52:22 -04002047 ep = f.file->private_data;
Davide Libenzi7699acd2007-05-10 22:23:23 -07002048
2049 /* Time to fish for events ... */
2050 error = ep_poll(ep, events, maxevents, timeout);
2051
2052error_fput:
Al Viro2903ff02012-08-28 12:52:22 -04002053 fdput(f);
Davide Libenzi7699acd2007-05-10 22:23:23 -07002054 return error;
2055}
2056
Davide Libenzi7699acd2007-05-10 22:23:23 -07002057/*
2058 * Implement the event wait interface for the eventpoll file. It is the kernel
2059 * part of the user space epoll_pwait(2).
2060 */
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01002061SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2062 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2063 size_t, sigsetsize)
Davide Libenzi7699acd2007-05-10 22:23:23 -07002064{
2065 int error;
2066 sigset_t ksigmask, sigsaved;
2067
2068 /*
2069 * If the caller wants a certain signal mask to be set during the wait,
2070 * we apply it here.
2071 */
2072 if (sigmask) {
2073 if (sigsetsize != sizeof(sigset_t))
2074 return -EINVAL;
2075 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
2076 return -EFAULT;
Oleg Nesterov77d55912013-07-03 15:08:14 -07002077 sigsaved = current->blocked;
2078 set_current_blocked(&ksigmask);
Davide Libenzi7699acd2007-05-10 22:23:23 -07002079 }
2080
2081 error = sys_epoll_wait(epfd, events, maxevents, timeout);
2082
2083 /*
2084 * If we changed the signal mask, we need to restore the original one.
2085 * In case we've got a signal while waiting, we do not restore the
2086 * signal mask yet, and we allow do_signal() to deliver the signal on
2087 * the way back to userspace, before the signal mask is restored.
2088 */
2089 if (sigmask) {
2090 if (error == -EINTR) {
2091 memcpy(&current->saved_sigmask, &sigsaved,
Davide Libenzic7ea7632007-05-15 01:40:47 -07002092 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07002093 set_restore_sigmask();
Davide Libenzi7699acd2007-05-10 22:23:23 -07002094 } else
Oleg Nesterov77d55912013-07-03 15:08:14 -07002095 set_current_blocked(&sigsaved);
Davide Libenzi7699acd2007-05-10 22:23:23 -07002096 }
2097
2098 return error;
2099}
2100
Al Viro35280bd2013-02-24 14:52:17 -05002101#ifdef CONFIG_COMPAT
2102COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2103 struct epoll_event __user *, events,
2104 int, maxevents, int, timeout,
2105 const compat_sigset_t __user *, sigmask,
2106 compat_size_t, sigsetsize)
2107{
2108 long err;
2109 compat_sigset_t csigmask;
2110 sigset_t ksigmask, sigsaved;
2111
2112 /*
2113 * If the caller wants a certain signal mask to be set during the wait,
2114 * we apply it here.
2115 */
2116 if (sigmask) {
2117 if (sigsetsize != sizeof(compat_sigset_t))
2118 return -EINVAL;
2119 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
2120 return -EFAULT;
2121 sigset_from_compat(&ksigmask, &csigmask);
Oleg Nesterov77d55912013-07-03 15:08:14 -07002122 sigsaved = current->blocked;
2123 set_current_blocked(&ksigmask);
Al Viro35280bd2013-02-24 14:52:17 -05002124 }
2125
2126 err = sys_epoll_wait(epfd, events, maxevents, timeout);
2127
2128 /*
2129 * If we changed the signal mask, we need to restore the original one.
2130 * In case we've got a signal while waiting, we do not restore the
2131 * signal mask yet, and we allow do_signal() to deliver the signal on
2132 * the way back to userspace, before the signal mask is restored.
2133 */
2134 if (sigmask) {
2135 if (err == -EINTR) {
2136 memcpy(&current->saved_sigmask, &sigsaved,
2137 sizeof(sigsaved));
2138 set_restore_sigmask();
2139 } else
Oleg Nesterov77d55912013-07-03 15:08:14 -07002140 set_current_blocked(&sigsaved);
Al Viro35280bd2013-02-24 14:52:17 -05002141 }
2142
2143 return err;
2144}
2145#endif
2146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147static int __init eventpoll_init(void)
2148{
Davide Libenzi7ef99642008-12-01 13:13:55 -08002149 struct sysinfo si;
2150
2151 si_meminfo(&si);
Davide Libenzi9df04e12009-01-29 14:25:26 -08002152 /*
2153 * Allows top 4% of lomem to be allocated for epoll watches (per user).
2154 */
2155 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
Davide Libenzi7ef99642008-12-01 13:13:55 -08002156 EP_ITEM_COST;
Robin Holt52bd19f72011-01-12 17:00:01 -08002157 BUG_ON(max_user_watches < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
Davide Libenzi22bacca2011-02-25 14:44:12 -08002159 /*
2160 * Initialize the structure used to perform epoll file descriptor
2161 * inclusion loops checks.
2162 */
2163 ep_nested_calls_init(&poll_loop_ncalls);
2164
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 /* Initialize the structure used to perform safe poll wait head wake ups */
Davide Libenzi5071f972009-03-31 15:24:10 -07002166 ep_nested_calls_init(&poll_safewake_ncalls);
2167
2168 /* Initialize the structure used to perform file's f_op->poll() calls */
2169 ep_nested_calls_init(&poll_readywalk_ncalls);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
Eric Wong39732ca2013-04-30 15:27:38 -07002171 /*
2172 * We can have many thousands of epitems, so prevent this from
2173 * using an extra cache line on 64-bit (and smaller) CPUs
2174 */
2175 BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2176
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 /* Allocates slab cache used to allocate "struct epitem" items */
2178 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
Davide Libenzibb57c3e2009-03-31 15:24:12 -07002179 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180
2181 /* Allocates slab cache used to allocate "struct eppoll_entry" */
2182 pwq_cache = kmem_cache_create("eventpoll_pwq",
Davide Libenzibb57c3e2009-03-31 15:24:12 -07002183 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186}
Davide Libenzicea69242007-05-10 22:23:22 -07002187fs_initcall(eventpoll_init);