]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/eventpoll.c
Merge remote-tracking branches 'regulator/fix/as3722', 'regulator/fix/ltc3589' and...
[mirror_ubuntu-bionic-kernel.git] / fs / eventpoll.c
CommitLineData
1da177e4 1/*
5071f97e
DL
2 * fs/eventpoll.c (Efficient event retrieval implementation)
3 * Copyright (C) 2001,...,2009 Davide Libenzi
1da177e4
LT
4 *
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
1da177e4
LT
14#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>
1da177e4
LT
24#include <linux/string.h>
25#include <linux/list.h>
26#include <linux/hash.h>
27#include <linux/spinlock.h>
28#include <linux/syscalls.h>
1da177e4
LT
29#include <linux/rbtree.h>
30#include <linux/wait.h>
31#include <linux/eventpoll.h>
32#include <linux/mount.h>
33#include <linux/bitops.h>
144efe3e 34#include <linux/mutex.h>
da66f7cb 35#include <linux/anon_inodes.h>
4d7e30d9 36#include <linux/device.h>
1da177e4 37#include <asm/uaccess.h>
1da177e4
LT
38#include <asm/io.h>
39#include <asm/mman.h>
60063497 40#include <linux/atomic.h>
138d22b5
CG
41#include <linux/proc_fs.h>
42#include <linux/seq_file.h>
35280bd4 43#include <linux/compat.h>
ae10b2b4 44#include <linux/rculist.h>
1da177e4 45
1da177e4
LT
46/*
47 * LOCKING:
48 * There are three level of locking required by epoll :
49 *
144efe3e 50 * 1) epmutex (mutex)
c7ea7630
DL
51 * 2) ep->mtx (mutex)
52 * 3) ep->lock (spinlock)
1da177e4
LT
53 *
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
d47de16c
DL
62 * 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
1da177e4
LT
67 * 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
bf6a41db 69 * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
22bacca4
DL
70 * 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.
d8805e63
NE
77 * 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.
d47de16c
DL
86 * 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.
144efe3e 89 * Events that require holding "epmutex" are very rare, while for
d47de16c
DL
90 * normal operations the epoll private "ep->mtx" will guarantee
91 * a better scalability.
1da177e4
LT
92 */
93
1da177e4 94/* Epoll private bits inside the event mask */
4d7e30d9 95#define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET)
1da177e4 96
5071f97e
DL
97/* Maximum number of nesting allowed inside epoll sets */
98#define EP_MAX_NESTS 4
1da177e4 99
b611967d
DL
100#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
101
d47de16c
DL
102#define EP_UNACTIVE_PTR ((void *) -1L)
103
7ef9964e
DL
104#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
105
1da177e4
LT
106struct epoll_filefd {
107 struct file *file;
108 int fd;
39732ca5 109} __packed;
1da177e4
LT
110
111/*
5071f97e
DL
112 * Structure used to track possible nested calls, for too deep recursions
113 * and loop cycles.
1da177e4 114 */
5071f97e 115struct nested_call_node {
1da177e4 116 struct list_head llink;
5071f97e 117 void *cookie;
3fe4a975 118 void *ctx;
1da177e4
LT
119};
120
121/*
5071f97e
DL
122 * This structure is used as collector for nested calls, to check for
123 * maximum recursion dept and loop cycles.
1da177e4 124 */
5071f97e
DL
125struct nested_calls {
126 struct list_head tasks_call_list;
1da177e4
LT
127 spinlock_t lock;
128};
129
d47de16c
DL
130/*
131 * Each file descriptor added to the eventpoll interface will
132 * have an entry of this type linked to the "rbr" RB tree.
39732ca5
EW
133 * Avoid increasing the size of this struct, there can be many thousands
134 * of these on a server and we do not want this to take another cache line.
d47de16c
DL
135 */
136struct epitem {
ae10b2b4
JB
137 union {
138 /* RB tree node links this structure to the eventpoll RB tree */
139 struct rb_node rbn;
140 /* Used to free the struct epitem */
141 struct rcu_head rcu;
142 };
d47de16c
DL
143
144 /* List header used to link this structure to the eventpoll ready list */
145 struct list_head rdllink;
146
c7ea7630
DL
147 /*
148 * Works together "struct eventpoll"->ovflist in keeping the
149 * single linked chain of items.
150 */
151 struct epitem *next;
152
d47de16c
DL
153 /* The file descriptor information this item refers to */
154 struct epoll_filefd ffd;
155
156 /* Number of active wait queue attached to poll operations */
157 int nwait;
158
159 /* List containing poll wait queues */
160 struct list_head pwqlist;
161
162 /* The "container" of this item */
163 struct eventpoll *ep;
164
d47de16c
DL
165 /* List header used to link this item to the "struct file" items list */
166 struct list_head fllink;
167
4d7e30d9 168 /* wakeup_source used when EPOLLWAKEUP is set */
eea1d585 169 struct wakeup_source __rcu *ws;
4d7e30d9 170
c7ea7630
DL
171 /* The structure that describe the interested events and the source fd */
172 struct epoll_event event;
d47de16c
DL
173};
174
1da177e4
LT
175/*
176 * This structure is stored inside the "private_data" member of the file
bf6a41db 177 * structure and represents the main data structure for the eventpoll
1da177e4
LT
178 * interface.
179 */
180struct eventpoll {
bf6a41db 181 /* Protect the access to this structure */
c7ea7630 182 spinlock_t lock;
1da177e4
LT
183
184 /*
d47de16c
DL
185 * This mutex is used to ensure that files are not removed
186 * while epoll is using them. This is held during the event
187 * collection loop, the file cleanup path, the epoll file exit
188 * code and the ctl operations.
1da177e4 189 */
d47de16c 190 struct mutex mtx;
1da177e4
LT
191
192 /* Wait queue used by sys_epoll_wait() */
193 wait_queue_head_t wq;
194
195 /* Wait queue used by file->poll() */
196 wait_queue_head_t poll_wait;
197
198 /* List of ready file descriptors */
199 struct list_head rdllist;
200
67647d0f 201 /* RB tree root used to store monitored fd structs */
1da177e4 202 struct rb_root rbr;
d47de16c
DL
203
204 /*
205 * This is a single linked list that chains all the "struct epitem" that
25985edc 206 * happened while transferring ready events to userspace w/out
d47de16c
DL
207 * holding ->lock.
208 */
209 struct epitem *ovflist;
7ef9964e 210
4d7e30d9
AH
211 /* wakeup_source used when ep_scan_ready_list is running */
212 struct wakeup_source *ws;
213
7ef9964e
DL
214 /* The user that created the eventpoll descriptor */
215 struct user_struct *user;
28d82dc1
JB
216
217 struct file *file;
218
219 /* used to optimize loop detection check */
220 int visited;
221 struct list_head visited_list_link;
1da177e4
LT
222};
223
224/* Wait structure used by the poll hooks */
225struct eppoll_entry {
226 /* List header used to link this structure to the "struct epitem" */
227 struct list_head llink;
228
229 /* The "base" pointer is set to the container "struct epitem" */
4f0989db 230 struct epitem *base;
1da177e4
LT
231
232 /*
233 * Wait queue item that will be linked to the target file wait
234 * queue head.
235 */
236 wait_queue_t wait;
237
238 /* The wait queue head that linked the "wait" wait queue item */
239 wait_queue_head_t *whead;
240};
241
1da177e4
LT
242/* Wrapper struct used by poll queueing */
243struct ep_pqueue {
244 poll_table pt;
245 struct epitem *epi;
246};
247
5071f97e
DL
248/* Used by the ep_send_events() function as callback private data */
249struct ep_send_events_data {
250 int maxevents;
251 struct epoll_event __user *events;
252};
253
7ef9964e
DL
254/*
255 * Configuration options available inside /proc/sys/fs/epoll/
256 */
7ef9964e 257/* Maximum number of epoll watched descriptors, per user */
52bd19f7 258static long max_user_watches __read_mostly;
7ef9964e 259
1da177e4 260/*
d47de16c 261 * This mutex is used to serialize ep_free() and eventpoll_release_file().
1da177e4 262 */
7ef9964e 263static DEFINE_MUTEX(epmutex);
1da177e4 264
22bacca4
DL
265/* Used to check for epoll file descriptor inclusion loops */
266static struct nested_calls poll_loop_ncalls;
267
5071f97e
DL
268/* Used for safe wake up implementation */
269static struct nested_calls poll_safewake_ncalls;
270
271/* Used to call file's f_op->poll() under the nested calls boundaries */
272static struct nested_calls poll_readywalk_ncalls;
1da177e4
LT
273
274/* Slab cache used to allocate "struct epitem" */
e18b890b 275static struct kmem_cache *epi_cache __read_mostly;
1da177e4
LT
276
277/* Slab cache used to allocate "struct eppoll_entry" */
e18b890b 278static struct kmem_cache *pwq_cache __read_mostly;
1da177e4 279
28d82dc1
JB
280/* Visited nodes during ep_loop_check(), so we can unset them when we finish */
281static LIST_HEAD(visited_list);
282
283/*
284 * List of files with newly added links, where we may need to limit the number
285 * of emanating paths. Protected by the epmutex.
286 */
287static LIST_HEAD(tfile_check_list);
288
7ef9964e
DL
289#ifdef CONFIG_SYSCTL
290
291#include <linux/sysctl.h>
292
52bd19f7
RH
293static long zero;
294static long long_max = LONG_MAX;
7ef9964e 295
1f7e0616 296struct ctl_table epoll_table[] = {
7ef9964e
DL
297 {
298 .procname = "max_user_watches",
299 .data = &max_user_watches,
52bd19f7 300 .maxlen = sizeof(max_user_watches),
7ef9964e 301 .mode = 0644,
52bd19f7 302 .proc_handler = proc_doulongvec_minmax,
7ef9964e 303 .extra1 = &zero,
52bd19f7 304 .extra2 = &long_max,
7ef9964e 305 },
ab09203e 306 { }
7ef9964e
DL
307};
308#endif /* CONFIG_SYSCTL */
309
28d82dc1
JB
310static const struct file_operations eventpoll_fops;
311
312static inline int is_file_epoll(struct file *f)
313{
314 return f->f_op == &eventpoll_fops;
315}
b030a4dd 316
67647d0f 317/* Setup the structure that is used as key for the RB tree */
b030a4dd
PE
318static inline void ep_set_ffd(struct epoll_filefd *ffd,
319 struct file *file, int fd)
320{
321 ffd->file = file;
322 ffd->fd = fd;
323}
324
67647d0f 325/* Compare RB tree keys */
b030a4dd
PE
326static inline int ep_cmp_ffd(struct epoll_filefd *p1,
327 struct epoll_filefd *p2)
328{
329 return (p1->file > p2->file ? +1:
330 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
331}
332
b030a4dd
PE
333/* Tells us if the item is currently linked */
334static inline int ep_is_linked(struct list_head *p)
335{
336 return !list_empty(p);
337}
338
971316f0
ON
339static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
340{
341 return container_of(p, struct eppoll_entry, wait);
342}
343
b030a4dd 344/* Get the "struct epitem" from a wait queue pointer */
cdac75e6 345static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
b030a4dd
PE
346{
347 return container_of(p, struct eppoll_entry, wait)->base;
348}
349
350/* Get the "struct epitem" from an epoll queue wrapper */
cdac75e6 351static inline struct epitem *ep_item_from_epqueue(poll_table *p)
b030a4dd
PE
352{
353 return container_of(p, struct ep_pqueue, pt)->epi;
354}
355
356/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
6192bd53 357static inline int ep_op_has_event(int op)
b030a4dd 358{
a80a6b85 359 return op != EPOLL_CTL_DEL;
b030a4dd
PE
360}
361
1da177e4 362/* Initialize the poll safe wake up structure */
5071f97e 363static void ep_nested_calls_init(struct nested_calls *ncalls)
1da177e4 364{
5071f97e
DL
365 INIT_LIST_HEAD(&ncalls->tasks_call_list);
366 spin_lock_init(&ncalls->lock);
1da177e4
LT
367}
368
3fb0e584
DL
369/**
370 * ep_events_available - Checks if ready events might be available.
371 *
372 * @ep: Pointer to the eventpoll context.
373 *
374 * Returns: Returns a value different than zero if ready events are available,
375 * or zero otherwise.
376 */
377static inline int ep_events_available(struct eventpoll *ep)
378{
379 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
380}
381
5071f97e
DL
382/**
383 * ep_call_nested - Perform a bound (possibly) nested call, by checking
384 * that the recursion limit is not exceeded, and that
385 * the same nested call (by the meaning of same cookie) is
386 * no re-entered.
387 *
388 * @ncalls: Pointer to the nested_calls structure to be used for this call.
389 * @max_nests: Maximum number of allowed nesting calls.
390 * @nproc: Nested call core function pointer.
391 * @priv: Opaque data to be passed to the @nproc callback.
392 * @cookie: Cookie to be used to identify this nested call.
3fe4a975 393 * @ctx: This instance context.
5071f97e
DL
394 *
395 * Returns: Returns the code returned by the @nproc callback, or -1 if
396 * the maximum recursion limit has been exceeded.
1da177e4 397 */
5071f97e
DL
398static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
399 int (*nproc)(void *, void *, int), void *priv,
3fe4a975 400 void *cookie, void *ctx)
1da177e4 401{
5071f97e 402 int error, call_nests = 0;
1da177e4 403 unsigned long flags;
5071f97e
DL
404 struct list_head *lsthead = &ncalls->tasks_call_list;
405 struct nested_call_node *tncur;
406 struct nested_call_node tnode;
1da177e4 407
5071f97e 408 spin_lock_irqsave(&ncalls->lock, flags);
1da177e4 409
5071f97e
DL
410 /*
411 * Try to see if the current task is already inside this wakeup call.
412 * We use a list here, since the population inside this set is always
413 * very much limited.
414 */
b70c3940 415 list_for_each_entry(tncur, lsthead, llink) {
3fe4a975 416 if (tncur->ctx == ctx &&
5071f97e 417 (tncur->cookie == cookie || ++call_nests > max_nests)) {
1da177e4
LT
418 /*
419 * Ops ... loop detected or maximum nest level reached.
420 * We abort this wake by breaking the cycle itself.
421 */
abff55ce
TB
422 error = -1;
423 goto out_unlock;
1da177e4
LT
424 }
425 }
426
5071f97e 427 /* Add the current task and cookie to the list */
3fe4a975 428 tnode.ctx = ctx;
5071f97e 429 tnode.cookie = cookie;
1da177e4
LT
430 list_add(&tnode.llink, lsthead);
431
5071f97e 432 spin_unlock_irqrestore(&ncalls->lock, flags);
1da177e4 433
5071f97e
DL
434 /* Call the nested function */
435 error = (*nproc)(priv, cookie, call_nests);
1da177e4
LT
436
437 /* Remove the current task from the list */
5071f97e 438 spin_lock_irqsave(&ncalls->lock, flags);
1da177e4 439 list_del(&tnode.llink);
3fe4a975 440out_unlock:
5071f97e
DL
441 spin_unlock_irqrestore(&ncalls->lock, flags);
442
443 return error;
444}
445
02edc6fc
SR
446/*
447 * As described in commit 0ccf831cb lockdep: annotate epoll
448 * the use of wait queues used by epoll is done in a very controlled
449 * manner. Wake ups can nest inside each other, but are never done
450 * with the same locking. For example:
451 *
452 * dfd = socket(...);
453 * efd1 = epoll_create();
454 * efd2 = epoll_create();
455 * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
456 * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
457 *
458 * When a packet arrives to the device underneath "dfd", the net code will
459 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
460 * callback wakeup entry on that queue, and the wake_up() performed by the
461 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
462 * (efd1) notices that it may have some event ready, so it needs to wake up
463 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
464 * that ends up in another wake_up(), after having checked about the
465 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
466 * avoid stack blasting.
467 *
468 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
469 * this special case of epoll.
470 */
2dfa4eea
DL
471#ifdef CONFIG_DEBUG_LOCK_ALLOC
472static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
473 unsigned long events, int subclass)
474{
475 unsigned long flags;
476
477 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
478 wake_up_locked_poll(wqueue, events);
479 spin_unlock_irqrestore(&wqueue->lock, flags);
480}
481#else
482static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
483 unsigned long events, int subclass)
484{
485 wake_up_poll(wqueue, events);
486}
487#endif
488
5071f97e
DL
489static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
490{
2dfa4eea
DL
491 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
492 1 + call_nests);
5071f97e
DL
493 return 0;
494}
495
496/*
497 * Perform a safe wake up of the poll wait list. The problem is that
498 * with the new callback'd wake up system, it is possible that the
499 * poll callback is reentered from inside the call to wake_up() done
500 * on the poll wait queue head. The rule is that we cannot reenter the
501 * wake up code from the same task more than EP_MAX_NESTS times,
502 * and we cannot reenter the same wait queue head at all. This will
503 * enable to have a hierarchy of epoll file descriptor of no more than
504 * EP_MAX_NESTS deep.
505 */
506static void ep_poll_safewake(wait_queue_head_t *wq)
507{
3fe4a975
DL
508 int this_cpu = get_cpu();
509
5071f97e 510 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
3fe4a975
DL
511 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
512
513 put_cpu();
1da177e4
LT
514}
515
971316f0
ON
516static void ep_remove_wait_queue(struct eppoll_entry *pwq)
517{
518 wait_queue_head_t *whead;
519
520 rcu_read_lock();
521 /* If it is cleared by POLLFREE, it should be rcu-safe */
522 whead = rcu_dereference(pwq->whead);
523 if (whead)
524 remove_wait_queue(whead, &pwq->wait);
525 rcu_read_unlock();
526}
527
1da177e4 528/*
d1bc90dd
TB
529 * This function unregisters poll callbacks from the associated file
530 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
531 * ep_free).
1da177e4 532 */
7699acd1 533static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
1da177e4 534{
7699acd1
DL
535 struct list_head *lsthead = &epi->pwqlist;
536 struct eppoll_entry *pwq;
1da177e4 537
d1bc90dd
TB
538 while (!list_empty(lsthead)) {
539 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
1da177e4 540
d1bc90dd 541 list_del(&pwq->llink);
971316f0 542 ep_remove_wait_queue(pwq);
d1bc90dd 543 kmem_cache_free(pwq_cache, pwq);
1da177e4 544 }
1da177e4
LT
545}
546
eea1d585
EW
547/* call only when ep->mtx is held */
548static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
549{
550 return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
551}
552
553/* call only when ep->mtx is held */
554static inline void ep_pm_stay_awake(struct epitem *epi)
555{
556 struct wakeup_source *ws = ep_wakeup_source(epi);
557
558 if (ws)
559 __pm_stay_awake(ws);
560}
561
562static inline bool ep_has_wakeup_source(struct epitem *epi)
563{
564 return rcu_access_pointer(epi->ws) ? true : false;
565}
566
567/* call when ep->mtx cannot be held (ep_poll_callback) */
568static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
569{
570 struct wakeup_source *ws;
571
572 rcu_read_lock();
573 ws = rcu_dereference(epi->ws);
574 if (ws)
575 __pm_stay_awake(ws);
576 rcu_read_unlock();
577}
578
5071f97e
DL
579/**
580 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
581 * the scan code, to call f_op->poll(). Also allows for
582 * O(NumReady) performance.
583 *
584 * @ep: Pointer to the epoll private data structure.
585 * @sproc: Pointer to the scan callback.
586 * @priv: Private opaque data passed to the @sproc callback.
d8805e63 587 * @depth: The current depth of recursive f_op->poll calls.
67347fe4 588 * @ep_locked: caller already holds ep->mtx
5071f97e
DL
589 *
590 * Returns: The same integer error code returned by the @sproc callback.
591 */
592static int ep_scan_ready_list(struct eventpoll *ep,
593 int (*sproc)(struct eventpoll *,
594 struct list_head *, void *),
67347fe4 595 void *priv, int depth, bool ep_locked)
5071f97e
DL
596{
597 int error, pwake = 0;
598 unsigned long flags;
599 struct epitem *epi, *nepi;
296e236e 600 LIST_HEAD(txlist);
5071f97e
DL
601
602 /*
603 * We need to lock this because we could be hit by
e057e15f 604 * eventpoll_release_file() and epoll_ctl().
5071f97e 605 */
67347fe4
JB
606
607 if (!ep_locked)
608 mutex_lock_nested(&ep->mtx, depth);
5071f97e
DL
609
610 /*
611 * Steal the ready list, and re-init the original one to the
612 * empty list. Also, set ep->ovflist to NULL so that events
613 * happening while looping w/out locks, are not lost. We cannot
614 * have the poll callback to queue directly on ep->rdllist,
615 * because we want the "sproc" callback to be able to do it
616 * in a lockless way.
617 */
618 spin_lock_irqsave(&ep->lock, flags);
296e236e 619 list_splice_init(&ep->rdllist, &txlist);
5071f97e
DL
620 ep->ovflist = NULL;
621 spin_unlock_irqrestore(&ep->lock, flags);
622
623 /*
624 * Now call the callback function.
625 */
626 error = (*sproc)(ep, &txlist, priv);
627
628 spin_lock_irqsave(&ep->lock, flags);
629 /*
630 * During the time we spent inside the "sproc" callback, some
631 * other events might have been queued by the poll callback.
632 * We re-insert them inside the main ready-list here.
633 */
634 for (nepi = ep->ovflist; (epi = nepi) != NULL;
635 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
636 /*
637 * We need to check if the item is already in the list.
638 * During the "sproc" callback execution time, items are
639 * queued into ->ovflist but the "txlist" might already
640 * contain them, and the list_splice() below takes care of them.
641 */
4d7e30d9 642 if (!ep_is_linked(&epi->rdllink)) {
5071f97e 643 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 644 ep_pm_stay_awake(epi);
4d7e30d9 645 }
5071f97e
DL
646 }
647 /*
648 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
649 * releasing the lock, events will be queued in the normal way inside
650 * ep->rdllist.
651 */
652 ep->ovflist = EP_UNACTIVE_PTR;
653
654 /*
655 * Quickly re-inject items left on "txlist".
656 */
657 list_splice(&txlist, &ep->rdllist);
4d7e30d9 658 __pm_relax(ep->ws);
5071f97e
DL
659
660 if (!list_empty(&ep->rdllist)) {
661 /*
296e236e
DL
662 * Wake up (if active) both the eventpoll wait list and
663 * the ->poll() wait list (delayed after we release the lock).
5071f97e
DL
664 */
665 if (waitqueue_active(&ep->wq))
666 wake_up_locked(&ep->wq);
667 if (waitqueue_active(&ep->poll_wait))
668 pwake++;
669 }
670 spin_unlock_irqrestore(&ep->lock, flags);
671
67347fe4
JB
672 if (!ep_locked)
673 mutex_unlock(&ep->mtx);
5071f97e
DL
674
675 /* We have to call this outside the lock */
676 if (pwake)
677 ep_poll_safewake(&ep->poll_wait);
678
679 return error;
680}
681
ae10b2b4
JB
682static void epi_rcu_free(struct rcu_head *head)
683{
684 struct epitem *epi = container_of(head, struct epitem, rcu);
685 kmem_cache_free(epi_cache, epi);
686}
687
7699acd1
DL
688/*
689 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
c7ea7630 690 * all the associated resources. Must be called with "mtx" held.
7699acd1
DL
691 */
692static int ep_remove(struct eventpoll *ep, struct epitem *epi)
693{
7699acd1
DL
694 unsigned long flags;
695 struct file *file = epi->ffd.file;
1da177e4
LT
696
697 /*
7699acd1
DL
698 * Removes poll wait queue hooks. We _have_ to do this without holding
699 * the "ep->lock" otherwise a deadlock might occur. This because of the
700 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
701 * queue head lock when unregistering the wait queue. The wakeup callback
702 * will run by holding the wait queue head lock and will call our callback
703 * that will try to get "ep->lock".
1da177e4 704 */
7699acd1 705 ep_unregister_pollwait(ep, epi);
1da177e4 706
7699acd1 707 /* Remove the current item from the list of epoll hooks */
68499914 708 spin_lock(&file->f_lock);
ae10b2b4 709 list_del_rcu(&epi->fllink);
68499914 710 spin_unlock(&file->f_lock);
1da177e4 711
cdac75e6 712 rb_erase(&epi->rbn, &ep->rbr);
1da177e4 713
c7ea7630
DL
714 spin_lock_irqsave(&ep->lock, flags);
715 if (ep_is_linked(&epi->rdllink))
716 list_del_init(&epi->rdllink);
717 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 718
eea1d585 719 wakeup_source_unregister(ep_wakeup_source(epi));
ae10b2b4
JB
720 /*
721 * At this point it is safe to free the eventpoll item. Use the union
722 * field epi->rcu, since we are trying to minimize the size of
723 * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
724 * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
725 * use of the rbn field.
726 */
727 call_rcu(&epi->rcu, epi_rcu_free);
1da177e4 728
52bd19f7 729 atomic_long_dec(&ep->user->epoll_watches);
7ef9964e 730
c7ea7630 731 return 0;
1da177e4
LT
732}
733
7699acd1 734static void ep_free(struct eventpoll *ep)
1da177e4 735{
7699acd1
DL
736 struct rb_node *rbp;
737 struct epitem *epi;
1da177e4 738
7699acd1
DL
739 /* We need to release all tasks waiting for these file */
740 if (waitqueue_active(&ep->poll_wait))
5071f97e 741 ep_poll_safewake(&ep->poll_wait);
1da177e4 742
7699acd1
DL
743 /*
744 * We need to lock this because we could be hit by
745 * eventpoll_release_file() while we're freeing the "struct eventpoll".
d47de16c 746 * We do not need to hold "ep->mtx" here because the epoll file
7699acd1
DL
747 * is on the way to be removed and no one has references to it
748 * anymore. The only hit might come from eventpoll_release_file() but
25985edc 749 * holding "epmutex" is sufficient here.
7699acd1
DL
750 */
751 mutex_lock(&epmutex);
1da177e4
LT
752
753 /*
7699acd1 754 * Walks through the whole tree by unregistering poll callbacks.
1da177e4 755 */
7699acd1
DL
756 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
757 epi = rb_entry(rbp, struct epitem, rbn);
758
759 ep_unregister_pollwait(ep, epi);
91cf5ab6 760 cond_resched();
7699acd1 761 }
1da177e4
LT
762
763 /*
7699acd1
DL
764 * Walks through the whole tree by freeing each "struct epitem". At this
765 * point we are sure no poll callbacks will be lingering around, and also by
d47de16c 766 * holding "epmutex" we can be sure that no file cleanup code will hit
7699acd1 767 * us during this operation. So we can avoid the lock on "ep->lock".
ddf676c3
EW
768 * We do not need to lock ep->mtx, either, we only do it to prevent
769 * a lockdep warning.
1da177e4 770 */
ddf676c3 771 mutex_lock(&ep->mtx);
c80544dc 772 while ((rbp = rb_first(&ep->rbr)) != NULL) {
7699acd1
DL
773 epi = rb_entry(rbp, struct epitem, rbn);
774 ep_remove(ep, epi);
91cf5ab6 775 cond_resched();
7699acd1 776 }
ddf676c3 777 mutex_unlock(&ep->mtx);
1da177e4 778
7699acd1 779 mutex_unlock(&epmutex);
d47de16c 780 mutex_destroy(&ep->mtx);
7ef9964e 781 free_uid(ep->user);
4d7e30d9 782 wakeup_source_unregister(ep->ws);
f0ee9aab 783 kfree(ep);
7699acd1 784}
1da177e4 785
7699acd1
DL
786static int ep_eventpoll_release(struct inode *inode, struct file *file)
787{
788 struct eventpoll *ep = file->private_data;
1da177e4 789
f0ee9aab 790 if (ep)
7699acd1 791 ep_free(ep);
7699acd1 792
7699acd1 793 return 0;
1da177e4
LT
794}
795
450d89ec
EW
796static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt)
797{
798 pt->_key = epi->event.events;
799
800 return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events;
801}
802
296e236e
DL
803static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
804 void *priv)
5071f97e
DL
805{
806 struct epitem *epi, *tmp;
626cf236 807 poll_table pt;
5071f97e 808
626cf236 809 init_poll_funcptr(&pt, NULL);
450d89ec 810
5071f97e 811 list_for_each_entry_safe(epi, tmp, head, rdllink) {
450d89ec 812 if (ep_item_poll(epi, &pt))
5071f97e 813 return POLLIN | POLLRDNORM;
296e236e 814 else {
5071f97e
DL
815 /*
816 * Item has been dropped into the ready list by the poll
817 * callback, but it's not actually ready, as far as
818 * caller requested events goes. We can remove it here.
819 */
eea1d585 820 __pm_relax(ep_wakeup_source(epi));
5071f97e 821 list_del_init(&epi->rdllink);
296e236e 822 }
5071f97e
DL
823 }
824
825 return 0;
826}
827
67347fe4
JB
828static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
829 poll_table *pt);
830
831struct readyevents_arg {
832 struct eventpoll *ep;
833 bool locked;
834};
835
5071f97e
DL
836static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
837{
67347fe4
JB
838 struct readyevents_arg *arg = priv;
839
840 return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL,
841 call_nests + 1, arg->locked);
5071f97e
DL
842}
843
7699acd1
DL
844static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
845{
5071f97e 846 int pollflags;
7699acd1 847 struct eventpoll *ep = file->private_data;
67347fe4
JB
848 struct readyevents_arg arg;
849
850 /*
851 * During ep_insert() we already hold the ep->mtx for the tfile.
852 * Prevent re-aquisition.
853 */
854 arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc);
855 arg.ep = ep;
1da177e4 856
7699acd1
DL
857 /* Insert inside our poll wait queue */
858 poll_wait(file, &ep->poll_wait, wait);
859
5071f97e
DL
860 /*
861 * Proceed to find out if wanted events are really available inside
862 * the ready list. This need to be done under ep_call_nested()
863 * supervision, since the call to f_op->poll() done on listed files
864 * could re-enter here.
865 */
866 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
67347fe4 867 ep_poll_readyevents_proc, &arg, ep, current);
7699acd1 868
296e236e 869 return pollflags != -1 ? pollflags : 0;
7699acd1
DL
870}
871
138d22b5
CG
872#ifdef CONFIG_PROC_FS
873static int ep_show_fdinfo(struct seq_file *m, struct file *f)
874{
875 struct eventpoll *ep = f->private_data;
876 struct rb_node *rbp;
877 int ret = 0;
878
879 mutex_lock(&ep->mtx);
880 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
881 struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
882
883 ret = seq_printf(m, "tfd: %8d events: %8x data: %16llx\n",
884 epi->ffd.fd, epi->event.events,
885 (long long)epi->event.data);
886 if (ret)
887 break;
888 }
889 mutex_unlock(&ep->mtx);
890
891 return ret;
892}
893#endif
894
7699acd1
DL
895/* File callbacks that implement the eventpoll file behaviour */
896static const struct file_operations eventpoll_fops = {
138d22b5
CG
897#ifdef CONFIG_PROC_FS
898 .show_fdinfo = ep_show_fdinfo,
899#endif
7699acd1 900 .release = ep_eventpoll_release,
6038f373
AB
901 .poll = ep_eventpoll_poll,
902 .llseek = noop_llseek,
7699acd1
DL
903};
904
b611967d 905/*
7699acd1
DL
906 * This is called from eventpoll_release() to unlink files from the eventpoll
907 * interface. We need to have this facility to cleanup correctly files that are
908 * closed without being removed from the eventpoll interface.
b611967d 909 */
7699acd1 910void eventpoll_release_file(struct file *file)
b611967d 911{
7699acd1
DL
912 struct eventpoll *ep;
913 struct epitem *epi;
b611967d
DL
914
915 /*
68499914 916 * We don't want to get "file->f_lock" because it is not
7699acd1 917 * necessary. It is not necessary because we're in the "struct file"
25985edc 918 * cleanup path, and this means that no one is using this file anymore.
5071f97e 919 * So, for example, epoll_ctl() cannot hit here since if we reach this
67647d0f 920 * point, the file counter already went to zero and fget() would fail.
d47de16c 921 * The only hit might come from ep_free() but by holding the mutex
7699acd1 922 * will correctly serialize the operation. We do need to acquire
d47de16c 923 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
7699acd1 924 * from anywhere but ep_free().
68499914
JC
925 *
926 * Besides, ep_remove() acquires the lock, so we can't hold it here.
b611967d 927 */
7699acd1 928 mutex_lock(&epmutex);
ae10b2b4 929 list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
7699acd1 930 ep = epi->ep;
d8805e63 931 mutex_lock_nested(&ep->mtx, 0);
7699acd1 932 ep_remove(ep, epi);
d47de16c 933 mutex_unlock(&ep->mtx);
b611967d 934 }
7699acd1 935 mutex_unlock(&epmutex);
b611967d
DL
936}
937
53d2be79 938static int ep_alloc(struct eventpoll **pep)
1da177e4 939{
7ef9964e
DL
940 int error;
941 struct user_struct *user;
942 struct eventpoll *ep;
1da177e4 943
7ef9964e 944 user = get_current_user();
7ef9964e
DL
945 error = -ENOMEM;
946 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
947 if (unlikely(!ep))
948 goto free_uid;
1da177e4 949
c7ea7630 950 spin_lock_init(&ep->lock);
d47de16c 951 mutex_init(&ep->mtx);
1da177e4
LT
952 init_waitqueue_head(&ep->wq);
953 init_waitqueue_head(&ep->poll_wait);
954 INIT_LIST_HEAD(&ep->rdllist);
955 ep->rbr = RB_ROOT;
d47de16c 956 ep->ovflist = EP_UNACTIVE_PTR;
7ef9964e 957 ep->user = user;
1da177e4 958
53d2be79 959 *pep = ep;
1da177e4 960
1da177e4 961 return 0;
7ef9964e
DL
962
963free_uid:
964 free_uid(user);
965 return error;
1da177e4
LT
966}
967
1da177e4 968/*
c7ea7630
DL
969 * Search the file inside the eventpoll tree. The RB tree operations
970 * are protected by the "mtx" mutex, and ep_find() must be called with
971 * "mtx" held.
1da177e4
LT
972 */
973static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
974{
975 int kcmp;
1da177e4
LT
976 struct rb_node *rbp;
977 struct epitem *epi, *epir = NULL;
978 struct epoll_filefd ffd;
979
b030a4dd 980 ep_set_ffd(&ffd, file, fd);
1da177e4
LT
981 for (rbp = ep->rbr.rb_node; rbp; ) {
982 epi = rb_entry(rbp, struct epitem, rbn);
b030a4dd 983 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
1da177e4
LT
984 if (kcmp > 0)
985 rbp = rbp->rb_right;
986 else if (kcmp < 0)
987 rbp = rbp->rb_left;
988 else {
1da177e4
LT
989 epir = epi;
990 break;
991 }
992 }
1da177e4 993
1da177e4
LT
994 return epir;
995}
996
1da177e4 997/*
7699acd1 998 * This is the callback that is passed to the wait queue wakeup
bf6a41db 999 * mechanism. It is called by the stored file descriptors when they
7699acd1 1000 * have events to report.
1da177e4 1001 */
7699acd1 1002static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
1da177e4 1003{
7699acd1
DL
1004 int pwake = 0;
1005 unsigned long flags;
1006 struct epitem *epi = ep_item_from_wait(wait);
1007 struct eventpoll *ep = epi->ep;
1da177e4 1008
971316f0
ON
1009 if ((unsigned long)key & POLLFREE) {
1010 ep_pwq_from_wait(wait)->whead = NULL;
1011 /*
1012 * whead = NULL above can race with ep_remove_wait_queue()
1013 * which can do another remove_wait_queue() after us, so we
1014 * can't use __remove_wait_queue(). whead->lock is held by
1015 * the caller.
1016 */
d80e731e 1017 list_del_init(&wait->task_list);
971316f0 1018 }
d80e731e 1019
c7ea7630 1020 spin_lock_irqsave(&ep->lock, flags);
1da177e4 1021
7699acd1
DL
1022 /*
1023 * If the event mask does not contain any poll(2) event, we consider the
1024 * descriptor to be disabled. This condition is likely the effect of the
1025 * EPOLLONESHOT bit that disables the descriptor when an event is received,
1026 * until the next EPOLL_CTL_MOD will be issued.
1027 */
1028 if (!(epi->event.events & ~EP_PRIVATE_BITS))
d47de16c
DL
1029 goto out_unlock;
1030
2dfa4eea
DL
1031 /*
1032 * Check the events coming with the callback. At this stage, not
1033 * every device reports the events in the "key" parameter of the
1034 * callback. We need to be able to handle both cases here, hence the
1035 * test for "key" != NULL before the event match test.
1036 */
1037 if (key && !((unsigned long) key & epi->event.events))
1038 goto out_unlock;
1039
d47de16c 1040 /*
bf6a41db 1041 * If we are transferring events to userspace, we can hold no locks
d47de16c 1042 * (because we're accessing user memory, and because of linux f_op->poll()
bf6a41db 1043 * semantics). All the events that happen during that period of time are
d47de16c
DL
1044 * chained in ep->ovflist and requeued later on.
1045 */
1046 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
1047 if (epi->next == EP_UNACTIVE_PTR) {
1048 epi->next = ep->ovflist;
1049 ep->ovflist = epi;
4d7e30d9
AH
1050 if (epi->ws) {
1051 /*
1052 * Activate ep->ws since epi->ws may get
1053 * deactivated at any time.
1054 */
1055 __pm_stay_awake(ep->ws);
1056 }
1057
d47de16c
DL
1058 }
1059 goto out_unlock;
1060 }
1da177e4 1061
7699acd1 1062 /* If this file is already in the ready list we exit soon */
4d7e30d9 1063 if (!ep_is_linked(&epi->rdllink)) {
5071f97e 1064 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1065 ep_pm_stay_awake_rcu(epi);
4d7e30d9 1066 }
7699acd1 1067
7699acd1
DL
1068 /*
1069 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1070 * wait list.
1071 */
1072 if (waitqueue_active(&ep->wq))
4a6e9e2c 1073 wake_up_locked(&ep->wq);
7699acd1
DL
1074 if (waitqueue_active(&ep->poll_wait))
1075 pwake++;
1076
d47de16c 1077out_unlock:
c7ea7630 1078 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 1079
7699acd1
DL
1080 /* We have to call this outside the lock */
1081 if (pwake)
5071f97e 1082 ep_poll_safewake(&ep->poll_wait);
7699acd1
DL
1083
1084 return 1;
1085}
1da177e4
LT
1086
1087/*
1088 * This is the callback that is used to add our wait queue to the
1089 * target file wakeup lists.
1090 */
1091static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1092 poll_table *pt)
1093{
b030a4dd 1094 struct epitem *epi = ep_item_from_epqueue(pt);
1da177e4
LT
1095 struct eppoll_entry *pwq;
1096
e94b1766 1097 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
1da177e4
LT
1098 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1099 pwq->whead = whead;
1100 pwq->base = epi;
1101 add_wait_queue(whead, &pwq->wait);
1102 list_add_tail(&pwq->llink, &epi->pwqlist);
1103 epi->nwait++;
296e236e 1104 } else {
1da177e4
LT
1105 /* We have to signal that an error occurred */
1106 epi->nwait = -1;
296e236e 1107 }
1da177e4
LT
1108}
1109
1da177e4
LT
1110static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1111{
1112 int kcmp;
1113 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
1114 struct epitem *epic;
1115
1116 while (*p) {
1117 parent = *p;
1118 epic = rb_entry(parent, struct epitem, rbn);
b030a4dd 1119 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1da177e4
LT
1120 if (kcmp > 0)
1121 p = &parent->rb_right;
1122 else
1123 p = &parent->rb_left;
1124 }
1125 rb_link_node(&epi->rbn, parent, p);
1126 rb_insert_color(&epi->rbn, &ep->rbr);
1127}
1128
a80a6b85
AM
1129
1130
28d82dc1
JB
1131#define PATH_ARR_SIZE 5
1132/*
1133 * These are the number paths of length 1 to 5, that we are allowing to emanate
1134 * from a single file of interest. For example, we allow 1000 paths of length
1135 * 1, to emanate from each file of interest. This essentially represents the
1136 * potential wakeup paths, which need to be limited in order to avoid massive
1137 * uncontrolled wakeup storms. The common use case should be a single ep which
1138 * is connected to n file sources. In this case each file source has 1 path
1139 * of length 1. Thus, the numbers below should be more than sufficient. These
1140 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1141 * and delete can't add additional paths. Protected by the epmutex.
1142 */
1143static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1144static int path_count[PATH_ARR_SIZE];
1145
1146static int path_count_inc(int nests)
1147{
93dc6107
JB
1148 /* Allow an arbitrary number of depth 1 paths */
1149 if (nests == 0)
1150 return 0;
1151
28d82dc1
JB
1152 if (++path_count[nests] > path_limits[nests])
1153 return -1;
1154 return 0;
1155}
1156
1157static void path_count_init(void)
1158{
1159 int i;
1160
1161 for (i = 0; i < PATH_ARR_SIZE; i++)
1162 path_count[i] = 0;
1163}
1164
1165static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1166{
1167 int error = 0;
1168 struct file *file = priv;
1169 struct file *child_file;
1170 struct epitem *epi;
1171
ae10b2b4
JB
1172 /* CTL_DEL can remove links here, but that can't increase our count */
1173 rcu_read_lock();
1174 list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
28d82dc1
JB
1175 child_file = epi->ep->file;
1176 if (is_file_epoll(child_file)) {
1177 if (list_empty(&child_file->f_ep_links)) {
1178 if (path_count_inc(call_nests)) {
1179 error = -1;
1180 break;
1181 }
1182 } else {
1183 error = ep_call_nested(&poll_loop_ncalls,
1184 EP_MAX_NESTS,
1185 reverse_path_check_proc,
1186 child_file, child_file,
1187 current);
1188 }
1189 if (error != 0)
1190 break;
1191 } else {
1192 printk(KERN_ERR "reverse_path_check_proc: "
1193 "file is not an ep!\n");
1194 }
1195 }
ae10b2b4 1196 rcu_read_unlock();
28d82dc1
JB
1197 return error;
1198}
1199
1200/**
1201 * reverse_path_check - The tfile_check_list is list of file *, which have
1202 * links that are proposed to be newly added. We need to
1203 * make sure that those added links don't add too many
1204 * paths such that we will spend all our time waking up
1205 * eventpoll objects.
1206 *
1207 * Returns: Returns zero if the proposed links don't create too many paths,
1208 * -1 otherwise.
1209 */
1210static int reverse_path_check(void)
1211{
28d82dc1
JB
1212 int error = 0;
1213 struct file *current_file;
1214
1215 /* let's call this for all tfiles */
1216 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
28d82dc1
JB
1217 path_count_init();
1218 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1219 reverse_path_check_proc, current_file,
1220 current_file, current);
1221 if (error)
1222 break;
1223 }
1224 return error;
1225}
1226
4d7e30d9
AH
1227static int ep_create_wakeup_source(struct epitem *epi)
1228{
1229 const char *name;
eea1d585 1230 struct wakeup_source *ws;
4d7e30d9
AH
1231
1232 if (!epi->ep->ws) {
1233 epi->ep->ws = wakeup_source_register("eventpoll");
1234 if (!epi->ep->ws)
1235 return -ENOMEM;
1236 }
1237
1238 name = epi->ffd.file->f_path.dentry->d_name.name;
eea1d585
EW
1239 ws = wakeup_source_register(name);
1240
1241 if (!ws)
4d7e30d9 1242 return -ENOMEM;
eea1d585 1243 rcu_assign_pointer(epi->ws, ws);
4d7e30d9
AH
1244
1245 return 0;
1246}
1247
eea1d585
EW
1248/* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1249static noinline void ep_destroy_wakeup_source(struct epitem *epi)
4d7e30d9 1250{
eea1d585
EW
1251 struct wakeup_source *ws = ep_wakeup_source(epi);
1252
d6d67e72 1253 RCU_INIT_POINTER(epi->ws, NULL);
eea1d585
EW
1254
1255 /*
1256 * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1257 * used internally by wakeup_source_remove, too (called by
1258 * wakeup_source_unregister), so we cannot use call_rcu
1259 */
1260 synchronize_rcu();
1261 wakeup_source_unregister(ws);
4d7e30d9
AH
1262}
1263
c7ea7630
DL
1264/*
1265 * Must be called with "mtx" held.
1266 */
1da177e4 1267static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
67347fe4 1268 struct file *tfile, int fd, int full_check)
1da177e4
LT
1269{
1270 int error, revents, pwake = 0;
1271 unsigned long flags;
52bd19f7 1272 long user_watches;
1da177e4
LT
1273 struct epitem *epi;
1274 struct ep_pqueue epq;
1275
52bd19f7
RH
1276 user_watches = atomic_long_read(&ep->user->epoll_watches);
1277 if (unlikely(user_watches >= max_user_watches))
7ef9964e 1278 return -ENOSPC;
e94b1766 1279 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
7ef9964e 1280 return -ENOMEM;
1da177e4
LT
1281
1282 /* Item initialization follow here ... */
1da177e4
LT
1283 INIT_LIST_HEAD(&epi->rdllink);
1284 INIT_LIST_HEAD(&epi->fllink);
1da177e4
LT
1285 INIT_LIST_HEAD(&epi->pwqlist);
1286 epi->ep = ep;
b030a4dd 1287 ep_set_ffd(&epi->ffd, tfile, fd);
1da177e4 1288 epi->event = *event;
1da177e4 1289 epi->nwait = 0;
d47de16c 1290 epi->next = EP_UNACTIVE_PTR;
4d7e30d9
AH
1291 if (epi->event.events & EPOLLWAKEUP) {
1292 error = ep_create_wakeup_source(epi);
1293 if (error)
1294 goto error_create_wakeup_source;
1295 } else {
eea1d585 1296 RCU_INIT_POINTER(epi->ws, NULL);
4d7e30d9 1297 }
1da177e4
LT
1298
1299 /* Initialize the poll table using the queue callback */
1300 epq.epi = epi;
1301 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1302
1303 /*
1304 * Attach the item to the poll hooks and get current event bits.
1305 * We can safely use the file* here because its usage count has
c7ea7630
DL
1306 * been increased by the caller of this function. Note that after
1307 * this operation completes, the poll callback can start hitting
1308 * the new item.
1da177e4 1309 */
450d89ec 1310 revents = ep_item_poll(epi, &epq.pt);
1da177e4
LT
1311
1312 /*
1313 * We have to check if something went wrong during the poll wait queue
1314 * install process. Namely an allocation for a wait queue failed due
1315 * high memory pressure.
1316 */
7ef9964e 1317 error = -ENOMEM;
1da177e4 1318 if (epi->nwait < 0)
7699acd1 1319 goto error_unregister;
1da177e4
LT
1320
1321 /* Add the current item to the list of active epoll hook for this file */
68499914 1322 spin_lock(&tfile->f_lock);
ae10b2b4 1323 list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
68499914 1324 spin_unlock(&tfile->f_lock);
1da177e4 1325
c7ea7630
DL
1326 /*
1327 * Add the current item to the RB tree. All RB tree operations are
1328 * protected by "mtx", and ep_insert() is called with "mtx" held.
1329 */
1da177e4
LT
1330 ep_rbtree_insert(ep, epi);
1331
28d82dc1
JB
1332 /* now check if we've created too many backpaths */
1333 error = -EINVAL;
67347fe4 1334 if (full_check && reverse_path_check())
28d82dc1
JB
1335 goto error_remove_epi;
1336
c7ea7630
DL
1337 /* We have to drop the new item inside our item list to keep track of it */
1338 spin_lock_irqsave(&ep->lock, flags);
1339
1da177e4 1340 /* If the file is already "ready" we drop it inside the ready list */
b030a4dd 1341 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
1da177e4 1342 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1343 ep_pm_stay_awake(epi);
1da177e4
LT
1344
1345 /* Notify waiting tasks that events are available */
1346 if (waitqueue_active(&ep->wq))
4a6e9e2c 1347 wake_up_locked(&ep->wq);
1da177e4
LT
1348 if (waitqueue_active(&ep->poll_wait))
1349 pwake++;
1350 }
1351
c7ea7630 1352 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 1353
52bd19f7 1354 atomic_long_inc(&ep->user->epoll_watches);
7ef9964e 1355
1da177e4
LT
1356 /* We have to call this outside the lock */
1357 if (pwake)
5071f97e 1358 ep_poll_safewake(&ep->poll_wait);
1da177e4 1359
1da177e4
LT
1360 return 0;
1361
28d82dc1
JB
1362error_remove_epi:
1363 spin_lock(&tfile->f_lock);
ae10b2b4 1364 list_del_rcu(&epi->fllink);
28d82dc1
JB
1365 spin_unlock(&tfile->f_lock);
1366
1367 rb_erase(&epi->rbn, &ep->rbr);
1368
7699acd1 1369error_unregister:
1da177e4
LT
1370 ep_unregister_pollwait(ep, epi);
1371
1372 /*
1373 * We need to do this because an event could have been arrived on some
67647d0f
DL
1374 * allocated wait queue. Note that we don't care about the ep->ovflist
1375 * list, since that is used/cleaned only inside a section bound by "mtx".
1376 * And ep_insert() is called with "mtx" held.
1da177e4 1377 */
c7ea7630 1378 spin_lock_irqsave(&ep->lock, flags);
b030a4dd 1379 if (ep_is_linked(&epi->rdllink))
6192bd53 1380 list_del_init(&epi->rdllink);
c7ea7630 1381 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 1382
eea1d585 1383 wakeup_source_unregister(ep_wakeup_source(epi));
4d7e30d9
AH
1384
1385error_create_wakeup_source:
b030a4dd 1386 kmem_cache_free(epi_cache, epi);
7ef9964e 1387
1da177e4
LT
1388 return error;
1389}
1390
1da177e4
LT
1391/*
1392 * Modify the interest event mask by dropping an event if the new mask
c7ea7630 1393 * has a match in the current file status. Must be called with "mtx" held.
1da177e4
LT
1394 */
1395static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1396{
1397 int pwake = 0;
1398 unsigned int revents;
626cf236
HV
1399 poll_table pt;
1400
1401 init_poll_funcptr(&pt, NULL);
1da177e4
LT
1402
1403 /*
e057e15f
TB
1404 * Set the new event interest mask before calling f_op->poll();
1405 * otherwise we might miss an event that happens between the
1406 * f_op->poll() call and the new event set registering.
1da177e4 1407 */
128dd175 1408 epi->event.events = event->events; /* need barrier below */
e057e15f 1409 epi->event.data = event->data; /* protected by mtx */
4d7e30d9 1410 if (epi->event.events & EPOLLWAKEUP) {
eea1d585 1411 if (!ep_has_wakeup_source(epi))
4d7e30d9 1412 ep_create_wakeup_source(epi);
eea1d585 1413 } else if (ep_has_wakeup_source(epi)) {
4d7e30d9
AH
1414 ep_destroy_wakeup_source(epi);
1415 }
1da177e4 1416
128dd175
EW
1417 /*
1418 * The following barrier has two effects:
1419 *
1420 * 1) Flush epi changes above to other CPUs. This ensures
1421 * we do not miss events from ep_poll_callback if an
1422 * event occurs immediately after we call f_op->poll().
1423 * We need this because we did not take ep->lock while
1424 * changing epi above (but ep_poll_callback does take
1425 * ep->lock).
1426 *
1427 * 2) We also need to ensure we do not miss _past_ events
1428 * when calling f_op->poll(). This barrier also
1429 * pairs with the barrier in wq_has_sleeper (see
1430 * comments for wq_has_sleeper).
1431 *
1432 * This barrier will now guarantee ep_poll_callback or f_op->poll
1433 * (or both) will notice the readiness of an item.
1434 */
1435 smp_mb();
1436
1da177e4
LT
1437 /*
1438 * Get current event bits. We can safely use the file* here because
1439 * its usage count has been increased by the caller of this function.
1440 */
450d89ec 1441 revents = ep_item_poll(epi, &pt);
1da177e4 1442
1da177e4 1443 /*
c7ea7630 1444 * If the item is "hot" and it is not registered inside the ready
67647d0f 1445 * list, push it inside.
1da177e4 1446 */
c7ea7630 1447 if (revents & event->events) {
e057e15f 1448 spin_lock_irq(&ep->lock);
c7ea7630
DL
1449 if (!ep_is_linked(&epi->rdllink)) {
1450 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1451 ep_pm_stay_awake(epi);
c7ea7630
DL
1452
1453 /* Notify waiting tasks that events are available */
1454 if (waitqueue_active(&ep->wq))
4a6e9e2c 1455 wake_up_locked(&ep->wq);
c7ea7630
DL
1456 if (waitqueue_active(&ep->poll_wait))
1457 pwake++;
7699acd1 1458 }
e057e15f 1459 spin_unlock_irq(&ep->lock);
7699acd1 1460 }
1da177e4 1461
7699acd1
DL
1462 /* We have to call this outside the lock */
1463 if (pwake)
5071f97e 1464 ep_poll_safewake(&ep->poll_wait);
1da177e4 1465
7699acd1 1466 return 0;
1da177e4
LT
1467}
1468
296e236e
DL
1469static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1470 void *priv)
1da177e4 1471{
5071f97e
DL
1472 struct ep_send_events_data *esed = priv;
1473 int eventcnt;
296e236e 1474 unsigned int revents;
5071f97e
DL
1475 struct epitem *epi;
1476 struct epoll_event __user *uevent;
eea1d585 1477 struct wakeup_source *ws;
626cf236
HV
1478 poll_table pt;
1479
1480 init_poll_funcptr(&pt, NULL);
1da177e4 1481
296e236e 1482 /*
5071f97e
DL
1483 * We can loop without lock because we are passed a task private list.
1484 * Items cannot vanish during the loop because ep_scan_ready_list() is
1485 * holding "mtx" during this call.
296e236e 1486 */
5071f97e
DL
1487 for (eventcnt = 0, uevent = esed->events;
1488 !list_empty(head) && eventcnt < esed->maxevents;) {
1489 epi = list_first_entry(head, struct epitem, rdllink);
d47de16c 1490
4d7e30d9
AH
1491 /*
1492 * Activate ep->ws before deactivating epi->ws to prevent
1493 * triggering auto-suspend here (in case we reactive epi->ws
1494 * below).
1495 *
1496 * This could be rearranged to delay the deactivation of epi->ws
1497 * instead, but then epi->ws would temporarily be out of sync
1498 * with ep_is_linked().
1499 */
eea1d585
EW
1500 ws = ep_wakeup_source(epi);
1501 if (ws) {
1502 if (ws->active)
1503 __pm_stay_awake(ep->ws);
1504 __pm_relax(ws);
1505 }
1506
d47de16c 1507 list_del_init(&epi->rdllink);
1da177e4 1508
450d89ec 1509 revents = ep_item_poll(epi, &pt);
5071f97e 1510
296e236e 1511 /*
5071f97e
DL
1512 * If the event mask intersect the caller-requested one,
1513 * deliver the event to userspace. Again, ep_scan_ready_list()
1514 * is holding "mtx", so no operations coming from userspace
1515 * can change the item.
296e236e
DL
1516 */
1517 if (revents) {
5071f97e 1518 if (__put_user(revents, &uevent->events) ||
d0305882
TB
1519 __put_user(epi->event.data, &uevent->data)) {
1520 list_add(&epi->rdllink, head);
eea1d585 1521 ep_pm_stay_awake(epi);
296e236e 1522 return eventcnt ? eventcnt : -EFAULT;
d0305882 1523 }
296e236e 1524 eventcnt++;
5071f97e 1525 uevent++;
296e236e
DL
1526 if (epi->event.events & EPOLLONESHOT)
1527 epi->event.events &= EP_PRIVATE_BITS;
1528 else if (!(epi->event.events & EPOLLET)) {
1529 /*
1530 * If this file has been added with Level
1531 * Trigger mode, we need to insert back inside
1532 * the ready list, so that the next call to
1533 * epoll_wait() will check again the events
25985edc 1534 * availability. At this point, no one can insert
296e236e
DL
1535 * into ep->rdllist besides us. The epoll_ctl()
1536 * callers are locked out by
1537 * ep_scan_ready_list() holding "mtx" and the
1538 * poll callback will queue them in ep->ovflist.
1539 */
1540 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1541 ep_pm_stay_awake(epi);
296e236e
DL
1542 }
1543 }
1544 }
5071f97e
DL
1545
1546 return eventcnt;
1547}
d47de16c 1548
296e236e
DL
1549static int ep_send_events(struct eventpoll *ep,
1550 struct epoll_event __user *events, int maxevents)
5071f97e
DL
1551{
1552 struct ep_send_events_data esed;
1da177e4 1553
5071f97e
DL
1554 esed.maxevents = maxevents;
1555 esed.events = events;
6192bd53 1556
67347fe4 1557 return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
1da177e4
LT
1558}
1559
0781b909
ED
1560static inline struct timespec ep_set_mstimeout(long ms)
1561{
1562 struct timespec now, ts = {
1563 .tv_sec = ms / MSEC_PER_SEC,
1564 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1565 };
1566
1567 ktime_get_ts(&now);
1568 return timespec_add_safe(now, ts);
1569}
1570
f4d93ad7
SB
1571/**
1572 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1573 * event buffer.
1574 *
1575 * @ep: Pointer to the eventpoll context.
1576 * @events: Pointer to the userspace buffer where the ready events should be
1577 * stored.
1578 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1579 * @timeout: Maximum timeout for the ready events fetch operation, in
1580 * milliseconds. If the @timeout is zero, the function will not block,
1581 * while if the @timeout is less than zero, the function will block
1582 * until at least one event has been retrieved (or an error
1583 * occurred).
1584 *
1585 * Returns: Returns the number of ready events which have been fetched, or an
1586 * error code, in case of error.
1587 */
1da177e4
LT
1588static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1589 int maxevents, long timeout)
1590{
f4d93ad7 1591 int res = 0, eavail, timed_out = 0;
1da177e4 1592 unsigned long flags;
f4d93ad7 1593 long slack = 0;
1da177e4 1594 wait_queue_t wait;
95aac7b1
SB
1595 ktime_t expires, *to = NULL;
1596
1597 if (timeout > 0) {
0781b909
ED
1598 struct timespec end_time = ep_set_mstimeout(timeout);
1599
95aac7b1
SB
1600 slack = select_estimate_accuracy(&end_time);
1601 to = &expires;
1602 *to = timespec_to_ktime(end_time);
1603 } else if (timeout == 0) {
f4d93ad7
SB
1604 /*
1605 * Avoid the unnecessary trip to the wait queue loop, if the
1606 * caller specified a non blocking operation.
1607 */
95aac7b1 1608 timed_out = 1;
f4d93ad7
SB
1609 spin_lock_irqsave(&ep->lock, flags);
1610 goto check_events;
95aac7b1 1611 }
1da177e4 1612
f4d93ad7 1613fetch_events:
c7ea7630 1614 spin_lock_irqsave(&ep->lock, flags);
1da177e4 1615
3fb0e584 1616 if (!ep_events_available(ep)) {
1da177e4
LT
1617 /*
1618 * We don't have any available event to return to the caller.
1619 * We need to sleep here, and we will be wake up by
1620 * ep_poll_callback() when events will become available.
1621 */
1622 init_waitqueue_entry(&wait, current);
a93d2f17 1623 __add_wait_queue_exclusive(&ep->wq, &wait);
1da177e4
LT
1624
1625 for (;;) {
1626 /*
1627 * We don't want to sleep if the ep_poll_callback() sends us
1628 * a wakeup in between. That's why we set the task state
1629 * to TASK_INTERRUPTIBLE before doing the checks.
1630 */
1631 set_current_state(TASK_INTERRUPTIBLE);
3fb0e584 1632 if (ep_events_available(ep) || timed_out)
1da177e4
LT
1633 break;
1634 if (signal_pending(current)) {
1635 res = -EINTR;
1636 break;
1637 }
1638
c7ea7630 1639 spin_unlock_irqrestore(&ep->lock, flags);
c511851d 1640 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
95aac7b1
SB
1641 timed_out = 1;
1642
c7ea7630 1643 spin_lock_irqsave(&ep->lock, flags);
1da177e4 1644 }
3419b23a 1645 __remove_wait_queue(&ep->wq, &wait);
1da177e4
LT
1646
1647 set_current_state(TASK_RUNNING);
1648 }
f4d93ad7 1649check_events:
1da177e4 1650 /* Is it worth to try to dig for events ? */
3fb0e584 1651 eavail = ep_events_available(ep);
1da177e4 1652
c7ea7630 1653 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4
LT
1654
1655 /*
1656 * Try to transfer events to user space. In case we get 0 events and
1657 * there's still timeout left over, we go trying again in search of
1658 * more luck.
1659 */
1660 if (!res && eavail &&
95aac7b1 1661 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
f4d93ad7 1662 goto fetch_events;
1da177e4
LT
1663
1664 return res;
1665}
1666
22bacca4
DL
1667/**
1668 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1669 * API, to verify that adding an epoll file inside another
1670 * epoll structure, does not violate the constraints, in
1671 * terms of closed loops, or too deep chains (which can
1672 * result in excessive stack usage).
1673 *
1674 * @priv: Pointer to the epoll file to be currently checked.
1675 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1676 * data structure pointer.
1677 * @call_nests: Current dept of the @ep_call_nested() call stack.
1678 *
1679 * Returns: Returns zero if adding the epoll @file inside current epoll
1680 * structure @ep does not violate the constraints, or -1 otherwise.
1681 */
1682static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1683{
1684 int error = 0;
1685 struct file *file = priv;
1686 struct eventpoll *ep = file->private_data;
28d82dc1 1687 struct eventpoll *ep_tovisit;
22bacca4
DL
1688 struct rb_node *rbp;
1689 struct epitem *epi;
1690
d8805e63 1691 mutex_lock_nested(&ep->mtx, call_nests + 1);
28d82dc1
JB
1692 ep->visited = 1;
1693 list_add(&ep->visited_list_link, &visited_list);
22bacca4
DL
1694 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1695 epi = rb_entry(rbp, struct epitem, rbn);
1696 if (unlikely(is_file_epoll(epi->ffd.file))) {
28d82dc1
JB
1697 ep_tovisit = epi->ffd.file->private_data;
1698 if (ep_tovisit->visited)
1699 continue;
22bacca4 1700 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
28d82dc1
JB
1701 ep_loop_check_proc, epi->ffd.file,
1702 ep_tovisit, current);
22bacca4
DL
1703 if (error != 0)
1704 break;
28d82dc1
JB
1705 } else {
1706 /*
1707 * If we've reached a file that is not associated with
1708 * an ep, then we need to check if the newly added
1709 * links are going to add too many wakeup paths. We do
1710 * this by adding it to the tfile_check_list, if it's
1711 * not already there, and calling reverse_path_check()
1712 * during ep_insert().
1713 */
1714 if (list_empty(&epi->ffd.file->f_tfile_llink))
1715 list_add(&epi->ffd.file->f_tfile_llink,
1716 &tfile_check_list);
22bacca4
DL
1717 }
1718 }
1719 mutex_unlock(&ep->mtx);
1720
1721 return error;
1722}
1723
1724/**
1725 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1726 * another epoll file (represented by @ep) does not create
1727 * closed loops or too deep chains.
1728 *
1729 * @ep: Pointer to the epoll private data structure.
1730 * @file: Pointer to the epoll file to be checked.
1731 *
1732 * Returns: Returns zero if adding the epoll @file inside current epoll
1733 * structure @ep does not violate the constraints, or -1 otherwise.
1734 */
1735static int ep_loop_check(struct eventpoll *ep, struct file *file)
1736{
28d82dc1
JB
1737 int ret;
1738 struct eventpoll *ep_cur, *ep_next;
1739
1740 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
22bacca4 1741 ep_loop_check_proc, file, ep, current);
28d82dc1
JB
1742 /* clear visited list */
1743 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1744 visited_list_link) {
1745 ep_cur->visited = 0;
1746 list_del(&ep_cur->visited_list_link);
1747 }
1748 return ret;
1749}
1750
1751static void clear_tfile_check_list(void)
1752{
1753 struct file *file;
1754
1755 /* first clear the tfile_check_list */
1756 while (!list_empty(&tfile_check_list)) {
1757 file = list_first_entry(&tfile_check_list, struct file,
1758 f_tfile_llink);
1759 list_del_init(&file->f_tfile_llink);
1760 }
1761 INIT_LIST_HEAD(&tfile_check_list);
22bacca4
DL
1762}
1763
7699acd1 1764/*
523723bb 1765 * Open an eventpoll file descriptor.
7699acd1 1766 */
5a8a82b1 1767SYSCALL_DEFINE1(epoll_create1, int, flags)
7699acd1 1768{
28d82dc1 1769 int error, fd;
bb57c3ed 1770 struct eventpoll *ep = NULL;
28d82dc1 1771 struct file *file;
7699acd1 1772
e38b36f3
UD
1773 /* Check the EPOLL_* constant for consistency. */
1774 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1775
296e236e
DL
1776 if (flags & ~EPOLL_CLOEXEC)
1777 return -EINVAL;
7699acd1 1778 /*
bb57c3ed 1779 * Create the internal data structure ("struct eventpoll").
7699acd1 1780 */
9fe5ad9c 1781 error = ep_alloc(&ep);
bb57c3ed
DL
1782 if (error < 0)
1783 return error;
7699acd1
DL
1784 /*
1785 * Creates all the items needed to setup an eventpoll file. That is,
2030a42c 1786 * a file structure and a free file descriptor.
7699acd1 1787 */
28d82dc1
JB
1788 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1789 if (fd < 0) {
1790 error = fd;
1791 goto out_free_ep;
1792 }
1793 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
628ff7c1 1794 O_RDWR | (flags & O_CLOEXEC));
28d82dc1
JB
1795 if (IS_ERR(file)) {
1796 error = PTR_ERR(file);
1797 goto out_free_fd;
1798 }
28d82dc1 1799 ep->file = file;
98022748 1800 fd_install(fd, file);
28d82dc1
JB
1801 return fd;
1802
1803out_free_fd:
1804 put_unused_fd(fd);
1805out_free_ep:
1806 ep_free(ep);
bb57c3ed 1807 return error;
7699acd1
DL
1808}
1809
5a8a82b1 1810SYSCALL_DEFINE1(epoll_create, int, size)
a0998b50 1811{
bfe3891a 1812 if (size <= 0)
9fe5ad9c
UD
1813 return -EINVAL;
1814
1815 return sys_epoll_create1(0);
a0998b50
UD
1816}
1817
7699acd1
DL
1818/*
1819 * The following function implements the controller interface for
1820 * the eventpoll file that enables the insertion/removal/change of
67647d0f 1821 * file descriptors inside the interest set.
7699acd1 1822 */
5a8a82b1
HC
1823SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1824 struct epoll_event __user *, event)
7699acd1
DL
1825{
1826 int error;
67347fe4 1827 int full_check = 0;
7e3fb584 1828 struct fd f, tf;
7699acd1
DL
1829 struct eventpoll *ep;
1830 struct epitem *epi;
1831 struct epoll_event epds;
67347fe4 1832 struct eventpoll *tep = NULL;
7699acd1 1833
7699acd1
DL
1834 error = -EFAULT;
1835 if (ep_op_has_event(op) &&
1836 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1837 goto error_return;
1838
7699acd1 1839 error = -EBADF;
7e3fb584
AV
1840 f = fdget(epfd);
1841 if (!f.file)
7699acd1
DL
1842 goto error_return;
1843
1844 /* Get the "struct file *" for the target file */
7e3fb584
AV
1845 tf = fdget(fd);
1846 if (!tf.file)
7699acd1
DL
1847 goto error_fput;
1848
1849 /* The target file descriptor must support poll */
1850 error = -EPERM;
72c2d531 1851 if (!tf.file->f_op->poll)
7699acd1
DL
1852 goto error_tgt_fput;
1853
4d7e30d9 1854 /* Check if EPOLLWAKEUP is allowed */
95f19f65 1855 ep_take_care_of_epollwakeup(&epds);
4d7e30d9 1856
7699acd1
DL
1857 /*
1858 * We have to check that the file structure underneath the file descriptor
1859 * the user passed to us _is_ an eventpoll file. And also we do not permit
1860 * adding an epoll file descriptor inside itself.
1861 */
1862 error = -EINVAL;
7e3fb584 1863 if (f.file == tf.file || !is_file_epoll(f.file))
7699acd1
DL
1864 goto error_tgt_fput;
1865
1866 /*
1867 * At this point it is safe to assume that the "private_data" contains
1868 * our own data structure.
1869 */
7e3fb584 1870 ep = f.file->private_data;
7699acd1 1871
22bacca4
DL
1872 /*
1873 * When we insert an epoll file descriptor, inside another epoll file
1874 * descriptor, there is the change of creating closed loops, which are
28d82dc1
JB
1875 * better be handled here, than in more critical paths. While we are
1876 * checking for loops we also determine the list of files reachable
1877 * and hang them on the tfile_check_list, so we can check that we
1878 * haven't created too many possible wakeup paths.
22bacca4 1879 *
67347fe4
JB
1880 * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
1881 * the epoll file descriptor is attaching directly to a wakeup source,
1882 * unless the epoll file descriptor is nested. The purpose of taking the
1883 * 'epmutex' on add is to prevent complex toplogies such as loops and
1884 * deep wakeup paths from forming in parallel through multiple
1885 * EPOLL_CTL_ADD operations.
22bacca4 1886 */
67347fe4 1887 mutex_lock_nested(&ep->mtx, 0);
28d82dc1 1888 if (op == EPOLL_CTL_ADD) {
67347fe4
JB
1889 if (!list_empty(&f.file->f_ep_links) ||
1890 is_file_epoll(tf.file)) {
1891 full_check = 1;
1892 mutex_unlock(&ep->mtx);
1893 mutex_lock(&epmutex);
1894 if (is_file_epoll(tf.file)) {
1895 error = -ELOOP;
1896 if (ep_loop_check(ep, tf.file) != 0) {
1897 clear_tfile_check_list();
1898 goto error_tgt_fput;
1899 }
1900 } else
1901 list_add(&tf.file->f_tfile_llink,
1902 &tfile_check_list);
1903 mutex_lock_nested(&ep->mtx, 0);
1904 if (is_file_epoll(tf.file)) {
1905 tep = tf.file->private_data;
1906 mutex_lock_nested(&tep->mtx, 1);
13d51807 1907 }
67347fe4
JB
1908 }
1909 }
7699acd1 1910
67647d0f
DL
1911 /*
1912 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1913 * above, we can be sure to be able to use the item looked up by
1914 * ep_find() till we release the mutex.
1915 */
7e3fb584 1916 epi = ep_find(ep, tf.file, fd);
7699acd1
DL
1917
1918 error = -EINVAL;
1919 switch (op) {
1920 case EPOLL_CTL_ADD:
1921 if (!epi) {
1922 epds.events |= POLLERR | POLLHUP;
67347fe4 1923 error = ep_insert(ep, &epds, tf.file, fd, full_check);
7699acd1
DL
1924 } else
1925 error = -EEXIST;
67347fe4
JB
1926 if (full_check)
1927 clear_tfile_check_list();
7699acd1
DL
1928 break;
1929 case EPOLL_CTL_DEL:
1930 if (epi)
1931 error = ep_remove(ep, epi);
1932 else
1933 error = -ENOENT;
1934 break;
1935 case EPOLL_CTL_MOD:
1936 if (epi) {
1937 epds.events |= POLLERR | POLLHUP;
1938 error = ep_modify(ep, epi, &epds);
1939 } else
1940 error = -ENOENT;
1941 break;
1942 }
67347fe4
JB
1943 if (tep != NULL)
1944 mutex_unlock(&tep->mtx);
d47de16c 1945 mutex_unlock(&ep->mtx);
7699acd1
DL
1946
1947error_tgt_fput:
67347fe4 1948 if (full_check)
22bacca4
DL
1949 mutex_unlock(&epmutex);
1950
7e3fb584 1951 fdput(tf);
7699acd1 1952error_fput:
7e3fb584 1953 fdput(f);
7699acd1 1954error_return:
7699acd1
DL
1955
1956 return error;
1957}
1958
1959/*
1960 * Implement the event wait interface for the eventpoll file. It is the kernel
1961 * part of the user space epoll_wait(2).
1962 */
5a8a82b1
HC
1963SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1964 int, maxevents, int, timeout)
7699acd1 1965{
2903ff01
AV
1966 int error;
1967 struct fd f;
7699acd1
DL
1968 struct eventpoll *ep;
1969
7699acd1
DL
1970 /* The maximum number of event must be greater than zero */
1971 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1972 return -EINVAL;
1973
1974 /* Verify that the area passed by the user is writeable */
2903ff01
AV
1975 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
1976 return -EFAULT;
7699acd1
DL
1977
1978 /* Get the "struct file *" for the eventpoll file */
2903ff01
AV
1979 f = fdget(epfd);
1980 if (!f.file)
1981 return -EBADF;
7699acd1
DL
1982
1983 /*
1984 * We have to check that the file structure underneath the fd
1985 * the user passed to us _is_ an eventpoll file.
1986 */
1987 error = -EINVAL;
2903ff01 1988 if (!is_file_epoll(f.file))
7699acd1
DL
1989 goto error_fput;
1990
1991 /*
1992 * At this point it is safe to assume that the "private_data" contains
1993 * our own data structure.
1994 */
2903ff01 1995 ep = f.file->private_data;
7699acd1
DL
1996
1997 /* Time to fish for events ... */
1998 error = ep_poll(ep, events, maxevents, timeout);
1999
2000error_fput:
2903ff01 2001 fdput(f);
7699acd1
DL
2002 return error;
2003}
2004
7699acd1
DL
2005/*
2006 * Implement the event wait interface for the eventpoll file. It is the kernel
2007 * part of the user space epoll_pwait(2).
2008 */
5a8a82b1
HC
2009SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2010 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2011 size_t, sigsetsize)
7699acd1
DL
2012{
2013 int error;
2014 sigset_t ksigmask, sigsaved;
2015
2016 /*
2017 * If the caller wants a certain signal mask to be set during the wait,
2018 * we apply it here.
2019 */
2020 if (sigmask) {
2021 if (sigsetsize != sizeof(sigset_t))
2022 return -EINVAL;
2023 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
2024 return -EFAULT;
77d55918
ON
2025 sigsaved = current->blocked;
2026 set_current_blocked(&ksigmask);
7699acd1
DL
2027 }
2028
2029 error = sys_epoll_wait(epfd, events, maxevents, timeout);
2030
2031 /*
2032 * If we changed the signal mask, we need to restore the original one.
2033 * In case we've got a signal while waiting, we do not restore the
2034 * signal mask yet, and we allow do_signal() to deliver the signal on
2035 * the way back to userspace, before the signal mask is restored.
2036 */
2037 if (sigmask) {
2038 if (error == -EINTR) {
2039 memcpy(&current->saved_sigmask, &sigsaved,
c7ea7630 2040 sizeof(sigsaved));
4e4c22c7 2041 set_restore_sigmask();
7699acd1 2042 } else
77d55918 2043 set_current_blocked(&sigsaved);
7699acd1
DL
2044 }
2045
2046 return error;
2047}
2048
35280bd4
AV
2049#ifdef CONFIG_COMPAT
2050COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2051 struct epoll_event __user *, events,
2052 int, maxevents, int, timeout,
2053 const compat_sigset_t __user *, sigmask,
2054 compat_size_t, sigsetsize)
2055{
2056 long err;
2057 compat_sigset_t csigmask;
2058 sigset_t ksigmask, sigsaved;
2059
2060 /*
2061 * If the caller wants a certain signal mask to be set during the wait,
2062 * we apply it here.
2063 */
2064 if (sigmask) {
2065 if (sigsetsize != sizeof(compat_sigset_t))
2066 return -EINVAL;
2067 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
2068 return -EFAULT;
2069 sigset_from_compat(&ksigmask, &csigmask);
77d55918
ON
2070 sigsaved = current->blocked;
2071 set_current_blocked(&ksigmask);
35280bd4
AV
2072 }
2073
2074 err = sys_epoll_wait(epfd, events, maxevents, timeout);
2075
2076 /*
2077 * If we changed the signal mask, we need to restore the original one.
2078 * In case we've got a signal while waiting, we do not restore the
2079 * signal mask yet, and we allow do_signal() to deliver the signal on
2080 * the way back to userspace, before the signal mask is restored.
2081 */
2082 if (sigmask) {
2083 if (err == -EINTR) {
2084 memcpy(&current->saved_sigmask, &sigsaved,
2085 sizeof(sigsaved));
2086 set_restore_sigmask();
2087 } else
77d55918 2088 set_current_blocked(&sigsaved);
35280bd4
AV
2089 }
2090
2091 return err;
2092}
2093#endif
2094
1da177e4
LT
2095static int __init eventpoll_init(void)
2096{
7ef9964e
DL
2097 struct sysinfo si;
2098
2099 si_meminfo(&si);
9df04e1f
DL
2100 /*
2101 * Allows top 4% of lomem to be allocated for epoll watches (per user).
2102 */
2103 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
7ef9964e 2104 EP_ITEM_COST;
52bd19f7 2105 BUG_ON(max_user_watches < 0);
1da177e4 2106
22bacca4
DL
2107 /*
2108 * Initialize the structure used to perform epoll file descriptor
2109 * inclusion loops checks.
2110 */
2111 ep_nested_calls_init(&poll_loop_ncalls);
2112
1da177e4 2113 /* Initialize the structure used to perform safe poll wait head wake ups */
5071f97e
DL
2114 ep_nested_calls_init(&poll_safewake_ncalls);
2115
2116 /* Initialize the structure used to perform file's f_op->poll() calls */
2117 ep_nested_calls_init(&poll_readywalk_ncalls);
1da177e4 2118
39732ca5
EW
2119 /*
2120 * We can have many thousands of epitems, so prevent this from
2121 * using an extra cache line on 64-bit (and smaller) CPUs
2122 */
2123 BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2124
1da177e4
LT
2125 /* Allocates slab cache used to allocate "struct epitem" items */
2126 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
bb57c3ed 2127 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1da177e4
LT
2128
2129 /* Allocates slab cache used to allocate "struct eppoll_entry" */
2130 pwq_cache = kmem_cache_create("eventpoll_pwq",
bb57c3ed 2131 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
1da177e4 2132
1da177e4 2133 return 0;
1da177e4 2134}
cea69241 2135fs_initcall(eventpoll_init);