]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/eventpoll.c
bus: brcmstb_gisb: update to support new revision
[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>
174cd4b1 16#include <linux/sched/signal.h>
1da177e4
LT
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>
7c0f6ba6 37#include <linux/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>
bf3b9f63 45#include <net/busy_poll.h>
1da177e4 46
1da177e4
LT
47/*
48 * LOCKING:
49 * There are three level of locking required by epoll :
50 *
144efe3e 51 * 1) epmutex (mutex)
c7ea7630
DL
52 * 2) ep->mtx (mutex)
53 * 3) ep->lock (spinlock)
1da177e4
LT
54 *
55 * The acquire order is the one listed above, from 1 to 3.
56 * We need a spinlock (ep->lock) because we manipulate objects
57 * from inside the poll callback, that might be triggered from
58 * a wake_up() that in turn might be called from IRQ context.
59 * So we can't sleep inside the poll callback and hence we need
60 * a spinlock. During the event transfer loop (from kernel to
61 * user space) we could end up sleeping due a copy_to_user(), so
62 * we need a lock that will allow us to sleep. This lock is a
d47de16c
DL
63 * mutex (ep->mtx). It is acquired during the event transfer loop,
64 * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
65 * Then we also need a global mutex to serialize eventpoll_release_file()
66 * and ep_free().
67 * This mutex is acquired by ep_free() during the epoll file
1da177e4
LT
68 * cleanup path and it is also acquired by eventpoll_release_file()
69 * if a file has been pushed inside an epoll set and it is then
bf6a41db 70 * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
22bacca4
DL
71 * It is also acquired when inserting an epoll fd onto another epoll
72 * fd. We do this so that we walk the epoll tree and ensure that this
73 * insertion does not create a cycle of epoll file descriptors, which
74 * could lead to deadlock. We need a global mutex to prevent two
75 * simultaneous inserts (A into B and B into A) from racing and
76 * constructing a cycle without either insert observing that it is
77 * going to.
d8805e63
NE
78 * It is necessary to acquire multiple "ep->mtx"es at once in the
79 * case when one epoll fd is added to another. In this case, we
80 * always acquire the locks in the order of nesting (i.e. after
81 * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
82 * before e2->mtx). Since we disallow cycles of epoll file
83 * descriptors, this ensures that the mutexes are well-ordered. In
84 * order to communicate this nesting to lockdep, when walking a tree
85 * of epoll file descriptors, we use the current recursion depth as
86 * the lockdep subkey.
d47de16c
DL
87 * It is possible to drop the "ep->mtx" and to use the global
88 * mutex "epmutex" (together with "ep->lock") to have it working,
89 * but having "ep->mtx" will make the interface more scalable.
144efe3e 90 * Events that require holding "epmutex" are very rare, while for
d47de16c
DL
91 * normal operations the epoll private "ep->mtx" will guarantee
92 * a better scalability.
1da177e4
LT
93 */
94
1da177e4 95/* Epoll private bits inside the event mask */
df0108c5 96#define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE)
1da177e4 97
b6a515c8
JB
98#define EPOLLINOUT_BITS (POLLIN | POLLOUT)
99
100#define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | POLLERR | POLLHUP | \
101 EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE)
102
5071f97e
DL
103/* Maximum number of nesting allowed inside epoll sets */
104#define EP_MAX_NESTS 4
1da177e4 105
b611967d
DL
106#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
107
d47de16c
DL
108#define EP_UNACTIVE_PTR ((void *) -1L)
109
7ef9964e
DL
110#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
111
1da177e4
LT
112struct epoll_filefd {
113 struct file *file;
114 int fd;
39732ca5 115} __packed;
1da177e4
LT
116
117/*
5071f97e
DL
118 * Structure used to track possible nested calls, for too deep recursions
119 * and loop cycles.
1da177e4 120 */
5071f97e 121struct nested_call_node {
1da177e4 122 struct list_head llink;
5071f97e 123 void *cookie;
3fe4a975 124 void *ctx;
1da177e4
LT
125};
126
127/*
5071f97e
DL
128 * This structure is used as collector for nested calls, to check for
129 * maximum recursion dept and loop cycles.
1da177e4 130 */
5071f97e
DL
131struct nested_calls {
132 struct list_head tasks_call_list;
1da177e4
LT
133 spinlock_t lock;
134};
135
d47de16c
DL
136/*
137 * Each file descriptor added to the eventpoll interface will
138 * have an entry of this type linked to the "rbr" RB tree.
39732ca5
EW
139 * Avoid increasing the size of this struct, there can be many thousands
140 * of these on a server and we do not want this to take another cache line.
d47de16c
DL
141 */
142struct epitem {
ae10b2b4
JB
143 union {
144 /* RB tree node links this structure to the eventpoll RB tree */
145 struct rb_node rbn;
146 /* Used to free the struct epitem */
147 struct rcu_head rcu;
148 };
d47de16c
DL
149
150 /* List header used to link this structure to the eventpoll ready list */
151 struct list_head rdllink;
152
c7ea7630
DL
153 /*
154 * Works together "struct eventpoll"->ovflist in keeping the
155 * single linked chain of items.
156 */
157 struct epitem *next;
158
d47de16c
DL
159 /* The file descriptor information this item refers to */
160 struct epoll_filefd ffd;
161
162 /* Number of active wait queue attached to poll operations */
163 int nwait;
164
165 /* List containing poll wait queues */
166 struct list_head pwqlist;
167
168 /* The "container" of this item */
169 struct eventpoll *ep;
170
d47de16c
DL
171 /* List header used to link this item to the "struct file" items list */
172 struct list_head fllink;
173
4d7e30d9 174 /* wakeup_source used when EPOLLWAKEUP is set */
eea1d585 175 struct wakeup_source __rcu *ws;
4d7e30d9 176
c7ea7630
DL
177 /* The structure that describe the interested events and the source fd */
178 struct epoll_event event;
d47de16c
DL
179};
180
1da177e4
LT
181/*
182 * This structure is stored inside the "private_data" member of the file
bf6a41db 183 * structure and represents the main data structure for the eventpoll
1da177e4
LT
184 * interface.
185 */
186struct eventpoll {
bf6a41db 187 /* Protect the access to this structure */
c7ea7630 188 spinlock_t lock;
1da177e4
LT
189
190 /*
d47de16c
DL
191 * This mutex is used to ensure that files are not removed
192 * while epoll is using them. This is held during the event
193 * collection loop, the file cleanup path, the epoll file exit
194 * code and the ctl operations.
1da177e4 195 */
d47de16c 196 struct mutex mtx;
1da177e4
LT
197
198 /* Wait queue used by sys_epoll_wait() */
199 wait_queue_head_t wq;
200
201 /* Wait queue used by file->poll() */
202 wait_queue_head_t poll_wait;
203
204 /* List of ready file descriptors */
205 struct list_head rdllist;
206
67647d0f 207 /* RB tree root used to store monitored fd structs */
1da177e4 208 struct rb_root rbr;
d47de16c
DL
209
210 /*
211 * This is a single linked list that chains all the "struct epitem" that
25985edc 212 * happened while transferring ready events to userspace w/out
d47de16c
DL
213 * holding ->lock.
214 */
215 struct epitem *ovflist;
7ef9964e 216
4d7e30d9
AH
217 /* wakeup_source used when ep_scan_ready_list is running */
218 struct wakeup_source *ws;
219
7ef9964e
DL
220 /* The user that created the eventpoll descriptor */
221 struct user_struct *user;
28d82dc1
JB
222
223 struct file *file;
224
225 /* used to optimize loop detection check */
226 int visited;
227 struct list_head visited_list_link;
bf3b9f63
SS
228
229#ifdef CONFIG_NET_RX_BUSY_POLL
230 /* used to track busy poll napi_id */
231 unsigned int napi_id;
232#endif
1da177e4
LT
233};
234
235/* Wait structure used by the poll hooks */
236struct eppoll_entry {
237 /* List header used to link this structure to the "struct epitem" */
238 struct list_head llink;
239
240 /* The "base" pointer is set to the container "struct epitem" */
4f0989db 241 struct epitem *base;
1da177e4
LT
242
243 /*
244 * Wait queue item that will be linked to the target file wait
245 * queue head.
246 */
247 wait_queue_t wait;
248
249 /* The wait queue head that linked the "wait" wait queue item */
250 wait_queue_head_t *whead;
251};
252
1da177e4
LT
253/* Wrapper struct used by poll queueing */
254struct ep_pqueue {
255 poll_table pt;
256 struct epitem *epi;
257};
258
5071f97e
DL
259/* Used by the ep_send_events() function as callback private data */
260struct ep_send_events_data {
261 int maxevents;
262 struct epoll_event __user *events;
263};
264
7ef9964e
DL
265/*
266 * Configuration options available inside /proc/sys/fs/epoll/
267 */
7ef9964e 268/* Maximum number of epoll watched descriptors, per user */
52bd19f7 269static long max_user_watches __read_mostly;
7ef9964e 270
1da177e4 271/*
d47de16c 272 * This mutex is used to serialize ep_free() and eventpoll_release_file().
1da177e4 273 */
7ef9964e 274static DEFINE_MUTEX(epmutex);
1da177e4 275
22bacca4
DL
276/* Used to check for epoll file descriptor inclusion loops */
277static struct nested_calls poll_loop_ncalls;
278
5071f97e
DL
279/* Used for safe wake up implementation */
280static struct nested_calls poll_safewake_ncalls;
281
282/* Used to call file's f_op->poll() under the nested calls boundaries */
283static struct nested_calls poll_readywalk_ncalls;
1da177e4
LT
284
285/* Slab cache used to allocate "struct epitem" */
e18b890b 286static struct kmem_cache *epi_cache __read_mostly;
1da177e4
LT
287
288/* Slab cache used to allocate "struct eppoll_entry" */
e18b890b 289static struct kmem_cache *pwq_cache __read_mostly;
1da177e4 290
28d82dc1
JB
291/* Visited nodes during ep_loop_check(), so we can unset them when we finish */
292static LIST_HEAD(visited_list);
293
294/*
295 * List of files with newly added links, where we may need to limit the number
296 * of emanating paths. Protected by the epmutex.
297 */
298static LIST_HEAD(tfile_check_list);
299
7ef9964e
DL
300#ifdef CONFIG_SYSCTL
301
302#include <linux/sysctl.h>
303
52bd19f7
RH
304static long zero;
305static long long_max = LONG_MAX;
7ef9964e 306
1f7e0616 307struct ctl_table epoll_table[] = {
7ef9964e
DL
308 {
309 .procname = "max_user_watches",
310 .data = &max_user_watches,
52bd19f7 311 .maxlen = sizeof(max_user_watches),
7ef9964e 312 .mode = 0644,
52bd19f7 313 .proc_handler = proc_doulongvec_minmax,
7ef9964e 314 .extra1 = &zero,
52bd19f7 315 .extra2 = &long_max,
7ef9964e 316 },
ab09203e 317 { }
7ef9964e
DL
318};
319#endif /* CONFIG_SYSCTL */
320
28d82dc1
JB
321static const struct file_operations eventpoll_fops;
322
323static inline int is_file_epoll(struct file *f)
324{
325 return f->f_op == &eventpoll_fops;
326}
b030a4dd 327
67647d0f 328/* Setup the structure that is used as key for the RB tree */
b030a4dd
PE
329static inline void ep_set_ffd(struct epoll_filefd *ffd,
330 struct file *file, int fd)
331{
332 ffd->file = file;
333 ffd->fd = fd;
334}
335
67647d0f 336/* Compare RB tree keys */
b030a4dd
PE
337static inline int ep_cmp_ffd(struct epoll_filefd *p1,
338 struct epoll_filefd *p2)
339{
340 return (p1->file > p2->file ? +1:
341 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
342}
343
b030a4dd
PE
344/* Tells us if the item is currently linked */
345static inline int ep_is_linked(struct list_head *p)
346{
347 return !list_empty(p);
348}
349
971316f0
ON
350static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
351{
352 return container_of(p, struct eppoll_entry, wait);
353}
354
b030a4dd 355/* Get the "struct epitem" from a wait queue pointer */
cdac75e6 356static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
b030a4dd
PE
357{
358 return container_of(p, struct eppoll_entry, wait)->base;
359}
360
361/* Get the "struct epitem" from an epoll queue wrapper */
cdac75e6 362static inline struct epitem *ep_item_from_epqueue(poll_table *p)
b030a4dd
PE
363{
364 return container_of(p, struct ep_pqueue, pt)->epi;
365}
366
367/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
6192bd53 368static inline int ep_op_has_event(int op)
b030a4dd 369{
a80a6b85 370 return op != EPOLL_CTL_DEL;
b030a4dd
PE
371}
372
1da177e4 373/* Initialize the poll safe wake up structure */
5071f97e 374static void ep_nested_calls_init(struct nested_calls *ncalls)
1da177e4 375{
5071f97e
DL
376 INIT_LIST_HEAD(&ncalls->tasks_call_list);
377 spin_lock_init(&ncalls->lock);
1da177e4
LT
378}
379
3fb0e584
DL
380/**
381 * ep_events_available - Checks if ready events might be available.
382 *
383 * @ep: Pointer to the eventpoll context.
384 *
385 * Returns: Returns a value different than zero if ready events are available,
386 * or zero otherwise.
387 */
388static inline int ep_events_available(struct eventpoll *ep)
389{
390 return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
391}
392
bf3b9f63
SS
393#ifdef CONFIG_NET_RX_BUSY_POLL
394static bool ep_busy_loop_end(void *p, unsigned long start_time)
395{
396 struct eventpoll *ep = p;
397
398 return ep_events_available(ep) || busy_loop_timeout(start_time);
399}
400#endif /* CONFIG_NET_RX_BUSY_POLL */
401
402/*
403 * Busy poll if globally on and supporting sockets found && no events,
404 * busy loop will return if need_resched or ep_events_available.
405 *
406 * we must do our busy polling with irqs enabled
407 */
408static void ep_busy_loop(struct eventpoll *ep, int nonblock)
409{
410#ifdef CONFIG_NET_RX_BUSY_POLL
411 unsigned int napi_id = READ_ONCE(ep->napi_id);
412
413 if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on())
414 napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep);
415#endif
416}
417
418static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep)
419{
420#ifdef CONFIG_NET_RX_BUSY_POLL
421 if (ep->napi_id)
422 ep->napi_id = 0;
423#endif
424}
425
426/*
427 * Set epoll busy poll NAPI ID from sk.
428 */
429static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
430{
431#ifdef CONFIG_NET_RX_BUSY_POLL
432 struct eventpoll *ep;
433 unsigned int napi_id;
434 struct socket *sock;
435 struct sock *sk;
436 int err;
437
438 if (!net_busy_loop_on())
439 return;
440
441 sock = sock_from_file(epi->ffd.file, &err);
442 if (!sock)
443 return;
444
445 sk = sock->sk;
446 if (!sk)
447 return;
448
449 napi_id = READ_ONCE(sk->sk_napi_id);
450 ep = epi->ep;
451
452 /* Non-NAPI IDs can be rejected
453 * or
454 * Nothing to do if we already have this ID
455 */
456 if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id)
457 return;
458
459 /* record NAPI ID for use in next busy poll */
460 ep->napi_id = napi_id;
461#endif
462}
463
5071f97e
DL
464/**
465 * ep_call_nested - Perform a bound (possibly) nested call, by checking
466 * that the recursion limit is not exceeded, and that
467 * the same nested call (by the meaning of same cookie) is
468 * no re-entered.
469 *
470 * @ncalls: Pointer to the nested_calls structure to be used for this call.
471 * @max_nests: Maximum number of allowed nesting calls.
472 * @nproc: Nested call core function pointer.
473 * @priv: Opaque data to be passed to the @nproc callback.
474 * @cookie: Cookie to be used to identify this nested call.
3fe4a975 475 * @ctx: This instance context.
5071f97e
DL
476 *
477 * Returns: Returns the code returned by the @nproc callback, or -1 if
478 * the maximum recursion limit has been exceeded.
1da177e4 479 */
5071f97e
DL
480static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
481 int (*nproc)(void *, void *, int), void *priv,
3fe4a975 482 void *cookie, void *ctx)
1da177e4 483{
5071f97e 484 int error, call_nests = 0;
1da177e4 485 unsigned long flags;
5071f97e
DL
486 struct list_head *lsthead = &ncalls->tasks_call_list;
487 struct nested_call_node *tncur;
488 struct nested_call_node tnode;
1da177e4 489
5071f97e 490 spin_lock_irqsave(&ncalls->lock, flags);
1da177e4 491
5071f97e
DL
492 /*
493 * Try to see if the current task is already inside this wakeup call.
494 * We use a list here, since the population inside this set is always
495 * very much limited.
496 */
b70c3940 497 list_for_each_entry(tncur, lsthead, llink) {
3fe4a975 498 if (tncur->ctx == ctx &&
5071f97e 499 (tncur->cookie == cookie || ++call_nests > max_nests)) {
1da177e4
LT
500 /*
501 * Ops ... loop detected or maximum nest level reached.
502 * We abort this wake by breaking the cycle itself.
503 */
abff55ce
TB
504 error = -1;
505 goto out_unlock;
1da177e4
LT
506 }
507 }
508
5071f97e 509 /* Add the current task and cookie to the list */
3fe4a975 510 tnode.ctx = ctx;
5071f97e 511 tnode.cookie = cookie;
1da177e4
LT
512 list_add(&tnode.llink, lsthead);
513
5071f97e 514 spin_unlock_irqrestore(&ncalls->lock, flags);
1da177e4 515
5071f97e
DL
516 /* Call the nested function */
517 error = (*nproc)(priv, cookie, call_nests);
1da177e4
LT
518
519 /* Remove the current task from the list */
5071f97e 520 spin_lock_irqsave(&ncalls->lock, flags);
1da177e4 521 list_del(&tnode.llink);
3fe4a975 522out_unlock:
5071f97e
DL
523 spin_unlock_irqrestore(&ncalls->lock, flags);
524
525 return error;
526}
527
02edc6fc
SR
528/*
529 * As described in commit 0ccf831cb lockdep: annotate epoll
530 * the use of wait queues used by epoll is done in a very controlled
531 * manner. Wake ups can nest inside each other, but are never done
532 * with the same locking. For example:
533 *
534 * dfd = socket(...);
535 * efd1 = epoll_create();
536 * efd2 = epoll_create();
537 * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
538 * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
539 *
540 * When a packet arrives to the device underneath "dfd", the net code will
541 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
542 * callback wakeup entry on that queue, and the wake_up() performed by the
543 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
544 * (efd1) notices that it may have some event ready, so it needs to wake up
545 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
546 * that ends up in another wake_up(), after having checked about the
547 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
548 * avoid stack blasting.
549 *
550 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
551 * this special case of epoll.
552 */
2dfa4eea
DL
553#ifdef CONFIG_DEBUG_LOCK_ALLOC
554static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
555 unsigned long events, int subclass)
556{
557 unsigned long flags;
558
559 spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
560 wake_up_locked_poll(wqueue, events);
561 spin_unlock_irqrestore(&wqueue->lock, flags);
562}
563#else
564static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
565 unsigned long events, int subclass)
566{
567 wake_up_poll(wqueue, events);
568}
569#endif
570
5071f97e
DL
571static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
572{
2dfa4eea
DL
573 ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
574 1 + call_nests);
5071f97e
DL
575 return 0;
576}
577
578/*
579 * Perform a safe wake up of the poll wait list. The problem is that
580 * with the new callback'd wake up system, it is possible that the
581 * poll callback is reentered from inside the call to wake_up() done
582 * on the poll wait queue head. The rule is that we cannot reenter the
583 * wake up code from the same task more than EP_MAX_NESTS times,
584 * and we cannot reenter the same wait queue head at all. This will
585 * enable to have a hierarchy of epoll file descriptor of no more than
586 * EP_MAX_NESTS deep.
587 */
588static void ep_poll_safewake(wait_queue_head_t *wq)
589{
3fe4a975
DL
590 int this_cpu = get_cpu();
591
5071f97e 592 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
3fe4a975
DL
593 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
594
595 put_cpu();
1da177e4
LT
596}
597
971316f0
ON
598static void ep_remove_wait_queue(struct eppoll_entry *pwq)
599{
600 wait_queue_head_t *whead;
601
602 rcu_read_lock();
603 /* If it is cleared by POLLFREE, it should be rcu-safe */
604 whead = rcu_dereference(pwq->whead);
605 if (whead)
606 remove_wait_queue(whead, &pwq->wait);
607 rcu_read_unlock();
608}
609
1da177e4 610/*
d1bc90dd
TB
611 * This function unregisters poll callbacks from the associated file
612 * descriptor. Must be called with "mtx" held (or "epmutex" if called from
613 * ep_free).
1da177e4 614 */
7699acd1 615static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
1da177e4 616{
7699acd1
DL
617 struct list_head *lsthead = &epi->pwqlist;
618 struct eppoll_entry *pwq;
1da177e4 619
d1bc90dd
TB
620 while (!list_empty(lsthead)) {
621 pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
1da177e4 622
d1bc90dd 623 list_del(&pwq->llink);
971316f0 624 ep_remove_wait_queue(pwq);
d1bc90dd 625 kmem_cache_free(pwq_cache, pwq);
1da177e4 626 }
1da177e4
LT
627}
628
eea1d585
EW
629/* call only when ep->mtx is held */
630static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
631{
632 return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
633}
634
635/* call only when ep->mtx is held */
636static inline void ep_pm_stay_awake(struct epitem *epi)
637{
638 struct wakeup_source *ws = ep_wakeup_source(epi);
639
640 if (ws)
641 __pm_stay_awake(ws);
642}
643
644static inline bool ep_has_wakeup_source(struct epitem *epi)
645{
646 return rcu_access_pointer(epi->ws) ? true : false;
647}
648
649/* call when ep->mtx cannot be held (ep_poll_callback) */
650static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
651{
652 struct wakeup_source *ws;
653
654 rcu_read_lock();
655 ws = rcu_dereference(epi->ws);
656 if (ws)
657 __pm_stay_awake(ws);
658 rcu_read_unlock();
659}
660
5071f97e
DL
661/**
662 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
663 * the scan code, to call f_op->poll(). Also allows for
664 * O(NumReady) performance.
665 *
666 * @ep: Pointer to the epoll private data structure.
667 * @sproc: Pointer to the scan callback.
668 * @priv: Private opaque data passed to the @sproc callback.
d8805e63 669 * @depth: The current depth of recursive f_op->poll calls.
67347fe4 670 * @ep_locked: caller already holds ep->mtx
5071f97e
DL
671 *
672 * Returns: The same integer error code returned by the @sproc callback.
673 */
674static int ep_scan_ready_list(struct eventpoll *ep,
675 int (*sproc)(struct eventpoll *,
676 struct list_head *, void *),
67347fe4 677 void *priv, int depth, bool ep_locked)
5071f97e
DL
678{
679 int error, pwake = 0;
680 unsigned long flags;
681 struct epitem *epi, *nepi;
296e236e 682 LIST_HEAD(txlist);
5071f97e
DL
683
684 /*
685 * We need to lock this because we could be hit by
e057e15f 686 * eventpoll_release_file() and epoll_ctl().
5071f97e 687 */
67347fe4
JB
688
689 if (!ep_locked)
690 mutex_lock_nested(&ep->mtx, depth);
5071f97e
DL
691
692 /*
693 * Steal the ready list, and re-init the original one to the
694 * empty list. Also, set ep->ovflist to NULL so that events
695 * happening while looping w/out locks, are not lost. We cannot
696 * have the poll callback to queue directly on ep->rdllist,
697 * because we want the "sproc" callback to be able to do it
698 * in a lockless way.
699 */
700 spin_lock_irqsave(&ep->lock, flags);
296e236e 701 list_splice_init(&ep->rdllist, &txlist);
5071f97e
DL
702 ep->ovflist = NULL;
703 spin_unlock_irqrestore(&ep->lock, flags);
704
705 /*
706 * Now call the callback function.
707 */
708 error = (*sproc)(ep, &txlist, priv);
709
710 spin_lock_irqsave(&ep->lock, flags);
711 /*
712 * During the time we spent inside the "sproc" callback, some
713 * other events might have been queued by the poll callback.
714 * We re-insert them inside the main ready-list here.
715 */
716 for (nepi = ep->ovflist; (epi = nepi) != NULL;
717 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
718 /*
719 * We need to check if the item is already in the list.
720 * During the "sproc" callback execution time, items are
721 * queued into ->ovflist but the "txlist" might already
722 * contain them, and the list_splice() below takes care of them.
723 */
4d7e30d9 724 if (!ep_is_linked(&epi->rdllink)) {
5071f97e 725 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 726 ep_pm_stay_awake(epi);
4d7e30d9 727 }
5071f97e
DL
728 }
729 /*
730 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
731 * releasing the lock, events will be queued in the normal way inside
732 * ep->rdllist.
733 */
734 ep->ovflist = EP_UNACTIVE_PTR;
735
736 /*
737 * Quickly re-inject items left on "txlist".
738 */
739 list_splice(&txlist, &ep->rdllist);
4d7e30d9 740 __pm_relax(ep->ws);
5071f97e
DL
741
742 if (!list_empty(&ep->rdllist)) {
743 /*
296e236e
DL
744 * Wake up (if active) both the eventpoll wait list and
745 * the ->poll() wait list (delayed after we release the lock).
5071f97e
DL
746 */
747 if (waitqueue_active(&ep->wq))
748 wake_up_locked(&ep->wq);
749 if (waitqueue_active(&ep->poll_wait))
750 pwake++;
751 }
752 spin_unlock_irqrestore(&ep->lock, flags);
753
67347fe4
JB
754 if (!ep_locked)
755 mutex_unlock(&ep->mtx);
5071f97e
DL
756
757 /* We have to call this outside the lock */
758 if (pwake)
759 ep_poll_safewake(&ep->poll_wait);
760
761 return error;
762}
763
ae10b2b4
JB
764static void epi_rcu_free(struct rcu_head *head)
765{
766 struct epitem *epi = container_of(head, struct epitem, rcu);
767 kmem_cache_free(epi_cache, epi);
768}
769
7699acd1
DL
770/*
771 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
c7ea7630 772 * all the associated resources. Must be called with "mtx" held.
7699acd1
DL
773 */
774static int ep_remove(struct eventpoll *ep, struct epitem *epi)
775{
7699acd1
DL
776 unsigned long flags;
777 struct file *file = epi->ffd.file;
1da177e4
LT
778
779 /*
7699acd1
DL
780 * Removes poll wait queue hooks. We _have_ to do this without holding
781 * the "ep->lock" otherwise a deadlock might occur. This because of the
782 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
783 * queue head lock when unregistering the wait queue. The wakeup callback
784 * will run by holding the wait queue head lock and will call our callback
785 * that will try to get "ep->lock".
1da177e4 786 */
7699acd1 787 ep_unregister_pollwait(ep, epi);
1da177e4 788
7699acd1 789 /* Remove the current item from the list of epoll hooks */
68499914 790 spin_lock(&file->f_lock);
ae10b2b4 791 list_del_rcu(&epi->fllink);
68499914 792 spin_unlock(&file->f_lock);
1da177e4 793
cdac75e6 794 rb_erase(&epi->rbn, &ep->rbr);
1da177e4 795
c7ea7630
DL
796 spin_lock_irqsave(&ep->lock, flags);
797 if (ep_is_linked(&epi->rdllink))
798 list_del_init(&epi->rdllink);
799 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 800
eea1d585 801 wakeup_source_unregister(ep_wakeup_source(epi));
ae10b2b4
JB
802 /*
803 * At this point it is safe to free the eventpoll item. Use the union
804 * field epi->rcu, since we are trying to minimize the size of
805 * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
806 * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
807 * use of the rbn field.
808 */
809 call_rcu(&epi->rcu, epi_rcu_free);
1da177e4 810
52bd19f7 811 atomic_long_dec(&ep->user->epoll_watches);
7ef9964e 812
c7ea7630 813 return 0;
1da177e4
LT
814}
815
7699acd1 816static void ep_free(struct eventpoll *ep)
1da177e4 817{
7699acd1
DL
818 struct rb_node *rbp;
819 struct epitem *epi;
1da177e4 820
7699acd1
DL
821 /* We need to release all tasks waiting for these file */
822 if (waitqueue_active(&ep->poll_wait))
5071f97e 823 ep_poll_safewake(&ep->poll_wait);
1da177e4 824
7699acd1
DL
825 /*
826 * We need to lock this because we could be hit by
827 * eventpoll_release_file() while we're freeing the "struct eventpoll".
d47de16c 828 * We do not need to hold "ep->mtx" here because the epoll file
7699acd1
DL
829 * is on the way to be removed and no one has references to it
830 * anymore. The only hit might come from eventpoll_release_file() but
25985edc 831 * holding "epmutex" is sufficient here.
7699acd1
DL
832 */
833 mutex_lock(&epmutex);
1da177e4
LT
834
835 /*
7699acd1 836 * Walks through the whole tree by unregistering poll callbacks.
1da177e4 837 */
7699acd1
DL
838 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
839 epi = rb_entry(rbp, struct epitem, rbn);
840
841 ep_unregister_pollwait(ep, epi);
91cf5ab6 842 cond_resched();
7699acd1 843 }
1da177e4
LT
844
845 /*
7699acd1
DL
846 * Walks through the whole tree by freeing each "struct epitem". At this
847 * point we are sure no poll callbacks will be lingering around, and also by
d47de16c 848 * holding "epmutex" we can be sure that no file cleanup code will hit
7699acd1 849 * us during this operation. So we can avoid the lock on "ep->lock".
ddf676c3
EW
850 * We do not need to lock ep->mtx, either, we only do it to prevent
851 * a lockdep warning.
1da177e4 852 */
ddf676c3 853 mutex_lock(&ep->mtx);
c80544dc 854 while ((rbp = rb_first(&ep->rbr)) != NULL) {
7699acd1
DL
855 epi = rb_entry(rbp, struct epitem, rbn);
856 ep_remove(ep, epi);
91cf5ab6 857 cond_resched();
7699acd1 858 }
ddf676c3 859 mutex_unlock(&ep->mtx);
1da177e4 860
7699acd1 861 mutex_unlock(&epmutex);
d47de16c 862 mutex_destroy(&ep->mtx);
7ef9964e 863 free_uid(ep->user);
4d7e30d9 864 wakeup_source_unregister(ep->ws);
f0ee9aab 865 kfree(ep);
7699acd1 866}
1da177e4 867
7699acd1
DL
868static int ep_eventpoll_release(struct inode *inode, struct file *file)
869{
870 struct eventpoll *ep = file->private_data;
1da177e4 871
f0ee9aab 872 if (ep)
7699acd1 873 ep_free(ep);
7699acd1 874
7699acd1 875 return 0;
1da177e4
LT
876}
877
450d89ec
EW
878static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt)
879{
880 pt->_key = epi->event.events;
881
882 return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events;
883}
884
296e236e
DL
885static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
886 void *priv)
5071f97e
DL
887{
888 struct epitem *epi, *tmp;
626cf236 889 poll_table pt;
5071f97e 890
626cf236 891 init_poll_funcptr(&pt, NULL);
450d89ec 892
5071f97e 893 list_for_each_entry_safe(epi, tmp, head, rdllink) {
450d89ec 894 if (ep_item_poll(epi, &pt))
5071f97e 895 return POLLIN | POLLRDNORM;
296e236e 896 else {
5071f97e
DL
897 /*
898 * Item has been dropped into the ready list by the poll
899 * callback, but it's not actually ready, as far as
900 * caller requested events goes. We can remove it here.
901 */
eea1d585 902 __pm_relax(ep_wakeup_source(epi));
5071f97e 903 list_del_init(&epi->rdllink);
296e236e 904 }
5071f97e
DL
905 }
906
907 return 0;
908}
909
67347fe4
JB
910static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
911 poll_table *pt);
912
913struct readyevents_arg {
914 struct eventpoll *ep;
915 bool locked;
916};
917
5071f97e
DL
918static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
919{
67347fe4
JB
920 struct readyevents_arg *arg = priv;
921
922 return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL,
923 call_nests + 1, arg->locked);
5071f97e
DL
924}
925
7699acd1
DL
926static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
927{
5071f97e 928 int pollflags;
7699acd1 929 struct eventpoll *ep = file->private_data;
67347fe4
JB
930 struct readyevents_arg arg;
931
932 /*
933 * During ep_insert() we already hold the ep->mtx for the tfile.
934 * Prevent re-aquisition.
935 */
936 arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc);
937 arg.ep = ep;
1da177e4 938
7699acd1
DL
939 /* Insert inside our poll wait queue */
940 poll_wait(file, &ep->poll_wait, wait);
941
5071f97e
DL
942 /*
943 * Proceed to find out if wanted events are really available inside
944 * the ready list. This need to be done under ep_call_nested()
945 * supervision, since the call to f_op->poll() done on listed files
946 * could re-enter here.
947 */
948 pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
67347fe4 949 ep_poll_readyevents_proc, &arg, ep, current);
7699acd1 950
296e236e 951 return pollflags != -1 ? pollflags : 0;
7699acd1
DL
952}
953
138d22b5 954#ifdef CONFIG_PROC_FS
a3816ab0 955static void ep_show_fdinfo(struct seq_file *m, struct file *f)
138d22b5
CG
956{
957 struct eventpoll *ep = f->private_data;
958 struct rb_node *rbp;
138d22b5
CG
959
960 mutex_lock(&ep->mtx);
961 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
962 struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
963
a3816ab0
JP
964 seq_printf(m, "tfd: %8d events: %8x data: %16llx\n",
965 epi->ffd.fd, epi->event.events,
966 (long long)epi->event.data);
967 if (seq_has_overflowed(m))
138d22b5
CG
968 break;
969 }
970 mutex_unlock(&ep->mtx);
138d22b5
CG
971}
972#endif
973
7699acd1
DL
974/* File callbacks that implement the eventpoll file behaviour */
975static const struct file_operations eventpoll_fops = {
138d22b5
CG
976#ifdef CONFIG_PROC_FS
977 .show_fdinfo = ep_show_fdinfo,
978#endif
7699acd1 979 .release = ep_eventpoll_release,
6038f373
AB
980 .poll = ep_eventpoll_poll,
981 .llseek = noop_llseek,
7699acd1
DL
982};
983
b611967d 984/*
7699acd1
DL
985 * This is called from eventpoll_release() to unlink files from the eventpoll
986 * interface. We need to have this facility to cleanup correctly files that are
987 * closed without being removed from the eventpoll interface.
b611967d 988 */
7699acd1 989void eventpoll_release_file(struct file *file)
b611967d 990{
7699acd1 991 struct eventpoll *ep;
ebe06187 992 struct epitem *epi, *next;
b611967d
DL
993
994 /*
68499914 995 * We don't want to get "file->f_lock" because it is not
7699acd1 996 * necessary. It is not necessary because we're in the "struct file"
25985edc 997 * cleanup path, and this means that no one is using this file anymore.
5071f97e 998 * So, for example, epoll_ctl() cannot hit here since if we reach this
67647d0f 999 * point, the file counter already went to zero and fget() would fail.
d47de16c 1000 * The only hit might come from ep_free() but by holding the mutex
7699acd1 1001 * will correctly serialize the operation. We do need to acquire
d47de16c 1002 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
7699acd1 1003 * from anywhere but ep_free().
68499914
JC
1004 *
1005 * Besides, ep_remove() acquires the lock, so we can't hold it here.
b611967d 1006 */
7699acd1 1007 mutex_lock(&epmutex);
ebe06187 1008 list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) {
7699acd1 1009 ep = epi->ep;
d8805e63 1010 mutex_lock_nested(&ep->mtx, 0);
7699acd1 1011 ep_remove(ep, epi);
d47de16c 1012 mutex_unlock(&ep->mtx);
b611967d 1013 }
7699acd1 1014 mutex_unlock(&epmutex);
b611967d
DL
1015}
1016
53d2be79 1017static int ep_alloc(struct eventpoll **pep)
1da177e4 1018{
7ef9964e
DL
1019 int error;
1020 struct user_struct *user;
1021 struct eventpoll *ep;
1da177e4 1022
7ef9964e 1023 user = get_current_user();
7ef9964e
DL
1024 error = -ENOMEM;
1025 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
1026 if (unlikely(!ep))
1027 goto free_uid;
1da177e4 1028
c7ea7630 1029 spin_lock_init(&ep->lock);
d47de16c 1030 mutex_init(&ep->mtx);
1da177e4
LT
1031 init_waitqueue_head(&ep->wq);
1032 init_waitqueue_head(&ep->poll_wait);
1033 INIT_LIST_HEAD(&ep->rdllist);
1034 ep->rbr = RB_ROOT;
d47de16c 1035 ep->ovflist = EP_UNACTIVE_PTR;
7ef9964e 1036 ep->user = user;
1da177e4 1037
53d2be79 1038 *pep = ep;
1da177e4 1039
1da177e4 1040 return 0;
7ef9964e
DL
1041
1042free_uid:
1043 free_uid(user);
1044 return error;
1da177e4
LT
1045}
1046
1da177e4 1047/*
c7ea7630
DL
1048 * Search the file inside the eventpoll tree. The RB tree operations
1049 * are protected by the "mtx" mutex, and ep_find() must be called with
1050 * "mtx" held.
1da177e4
LT
1051 */
1052static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
1053{
1054 int kcmp;
1da177e4
LT
1055 struct rb_node *rbp;
1056 struct epitem *epi, *epir = NULL;
1057 struct epoll_filefd ffd;
1058
b030a4dd 1059 ep_set_ffd(&ffd, file, fd);
1da177e4
LT
1060 for (rbp = ep->rbr.rb_node; rbp; ) {
1061 epi = rb_entry(rbp, struct epitem, rbn);
b030a4dd 1062 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
1da177e4
LT
1063 if (kcmp > 0)
1064 rbp = rbp->rb_right;
1065 else if (kcmp < 0)
1066 rbp = rbp->rb_left;
1067 else {
1da177e4
LT
1068 epir = epi;
1069 break;
1070 }
1071 }
1da177e4 1072
1da177e4
LT
1073 return epir;
1074}
1075
1da177e4 1076/*
7699acd1 1077 * This is the callback that is passed to the wait queue wakeup
bf6a41db 1078 * mechanism. It is called by the stored file descriptors when they
7699acd1 1079 * have events to report.
1da177e4 1080 */
7699acd1 1081static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
1da177e4 1082{
7699acd1
DL
1083 int pwake = 0;
1084 unsigned long flags;
1085 struct epitem *epi = ep_item_from_wait(wait);
1086 struct eventpoll *ep = epi->ep;
df0108c5 1087 int ewake = 0;
1da177e4 1088
971316f0
ON
1089 if ((unsigned long)key & POLLFREE) {
1090 ep_pwq_from_wait(wait)->whead = NULL;
1091 /*
1092 * whead = NULL above can race with ep_remove_wait_queue()
1093 * which can do another remove_wait_queue() after us, so we
1094 * can't use __remove_wait_queue(). whead->lock is held by
1095 * the caller.
1096 */
d80e731e 1097 list_del_init(&wait->task_list);
971316f0 1098 }
d80e731e 1099
c7ea7630 1100 spin_lock_irqsave(&ep->lock, flags);
1da177e4 1101
bf3b9f63
SS
1102 ep_set_busy_poll_napi_id(epi);
1103
7699acd1
DL
1104 /*
1105 * If the event mask does not contain any poll(2) event, we consider the
1106 * descriptor to be disabled. This condition is likely the effect of the
1107 * EPOLLONESHOT bit that disables the descriptor when an event is received,
1108 * until the next EPOLL_CTL_MOD will be issued.
1109 */
1110 if (!(epi->event.events & ~EP_PRIVATE_BITS))
d47de16c
DL
1111 goto out_unlock;
1112
2dfa4eea
DL
1113 /*
1114 * Check the events coming with the callback. At this stage, not
1115 * every device reports the events in the "key" parameter of the
1116 * callback. We need to be able to handle both cases here, hence the
1117 * test for "key" != NULL before the event match test.
1118 */
1119 if (key && !((unsigned long) key & epi->event.events))
1120 goto out_unlock;
1121
d47de16c 1122 /*
bf6a41db 1123 * If we are transferring events to userspace, we can hold no locks
d47de16c 1124 * (because we're accessing user memory, and because of linux f_op->poll()
bf6a41db 1125 * semantics). All the events that happen during that period of time are
d47de16c
DL
1126 * chained in ep->ovflist and requeued later on.
1127 */
1128 if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
1129 if (epi->next == EP_UNACTIVE_PTR) {
1130 epi->next = ep->ovflist;
1131 ep->ovflist = epi;
4d7e30d9
AH
1132 if (epi->ws) {
1133 /*
1134 * Activate ep->ws since epi->ws may get
1135 * deactivated at any time.
1136 */
1137 __pm_stay_awake(ep->ws);
1138 }
1139
d47de16c
DL
1140 }
1141 goto out_unlock;
1142 }
1da177e4 1143
7699acd1 1144 /* If this file is already in the ready list we exit soon */
4d7e30d9 1145 if (!ep_is_linked(&epi->rdllink)) {
5071f97e 1146 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1147 ep_pm_stay_awake_rcu(epi);
4d7e30d9 1148 }
7699acd1 1149
7699acd1
DL
1150 /*
1151 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1152 * wait list.
1153 */
df0108c5 1154 if (waitqueue_active(&ep->wq)) {
b6a515c8
JB
1155 if ((epi->event.events & EPOLLEXCLUSIVE) &&
1156 !((unsigned long)key & POLLFREE)) {
1157 switch ((unsigned long)key & EPOLLINOUT_BITS) {
1158 case POLLIN:
1159 if (epi->event.events & POLLIN)
1160 ewake = 1;
1161 break;
1162 case POLLOUT:
1163 if (epi->event.events & POLLOUT)
1164 ewake = 1;
1165 break;
1166 case 0:
1167 ewake = 1;
1168 break;
1169 }
1170 }
4a6e9e2c 1171 wake_up_locked(&ep->wq);
df0108c5 1172 }
7699acd1
DL
1173 if (waitqueue_active(&ep->poll_wait))
1174 pwake++;
1175
d47de16c 1176out_unlock:
c7ea7630 1177 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 1178
7699acd1
DL
1179 /* We have to call this outside the lock */
1180 if (pwake)
5071f97e 1181 ep_poll_safewake(&ep->poll_wait);
7699acd1 1182
df0108c5
JB
1183 if (epi->event.events & EPOLLEXCLUSIVE)
1184 return ewake;
1185
7699acd1
DL
1186 return 1;
1187}
1da177e4
LT
1188
1189/*
1190 * This is the callback that is used to add our wait queue to the
1191 * target file wakeup lists.
1192 */
1193static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1194 poll_table *pt)
1195{
b030a4dd 1196 struct epitem *epi = ep_item_from_epqueue(pt);
1da177e4
LT
1197 struct eppoll_entry *pwq;
1198
e94b1766 1199 if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
1da177e4
LT
1200 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1201 pwq->whead = whead;
1202 pwq->base = epi;
df0108c5
JB
1203 if (epi->event.events & EPOLLEXCLUSIVE)
1204 add_wait_queue_exclusive(whead, &pwq->wait);
1205 else
1206 add_wait_queue(whead, &pwq->wait);
1da177e4
LT
1207 list_add_tail(&pwq->llink, &epi->pwqlist);
1208 epi->nwait++;
296e236e 1209 } else {
1da177e4
LT
1210 /* We have to signal that an error occurred */
1211 epi->nwait = -1;
296e236e 1212 }
1da177e4
LT
1213}
1214
1da177e4
LT
1215static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1216{
1217 int kcmp;
1218 struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
1219 struct epitem *epic;
1220
1221 while (*p) {
1222 parent = *p;
1223 epic = rb_entry(parent, struct epitem, rbn);
b030a4dd 1224 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1da177e4
LT
1225 if (kcmp > 0)
1226 p = &parent->rb_right;
1227 else
1228 p = &parent->rb_left;
1229 }
1230 rb_link_node(&epi->rbn, parent, p);
1231 rb_insert_color(&epi->rbn, &ep->rbr);
1232}
1233
a80a6b85
AM
1234
1235
28d82dc1
JB
1236#define PATH_ARR_SIZE 5
1237/*
1238 * These are the number paths of length 1 to 5, that we are allowing to emanate
1239 * from a single file of interest. For example, we allow 1000 paths of length
1240 * 1, to emanate from each file of interest. This essentially represents the
1241 * potential wakeup paths, which need to be limited in order to avoid massive
1242 * uncontrolled wakeup storms. The common use case should be a single ep which
1243 * is connected to n file sources. In this case each file source has 1 path
1244 * of length 1. Thus, the numbers below should be more than sufficient. These
1245 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1246 * and delete can't add additional paths. Protected by the epmutex.
1247 */
1248static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1249static int path_count[PATH_ARR_SIZE];
1250
1251static int path_count_inc(int nests)
1252{
93dc6107
JB
1253 /* Allow an arbitrary number of depth 1 paths */
1254 if (nests == 0)
1255 return 0;
1256
28d82dc1
JB
1257 if (++path_count[nests] > path_limits[nests])
1258 return -1;
1259 return 0;
1260}
1261
1262static void path_count_init(void)
1263{
1264 int i;
1265
1266 for (i = 0; i < PATH_ARR_SIZE; i++)
1267 path_count[i] = 0;
1268}
1269
1270static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1271{
1272 int error = 0;
1273 struct file *file = priv;
1274 struct file *child_file;
1275 struct epitem *epi;
1276
ae10b2b4
JB
1277 /* CTL_DEL can remove links here, but that can't increase our count */
1278 rcu_read_lock();
1279 list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
28d82dc1
JB
1280 child_file = epi->ep->file;
1281 if (is_file_epoll(child_file)) {
1282 if (list_empty(&child_file->f_ep_links)) {
1283 if (path_count_inc(call_nests)) {
1284 error = -1;
1285 break;
1286 }
1287 } else {
1288 error = ep_call_nested(&poll_loop_ncalls,
1289 EP_MAX_NESTS,
1290 reverse_path_check_proc,
1291 child_file, child_file,
1292 current);
1293 }
1294 if (error != 0)
1295 break;
1296 } else {
1297 printk(KERN_ERR "reverse_path_check_proc: "
1298 "file is not an ep!\n");
1299 }
1300 }
ae10b2b4 1301 rcu_read_unlock();
28d82dc1
JB
1302 return error;
1303}
1304
1305/**
1306 * reverse_path_check - The tfile_check_list is list of file *, which have
1307 * links that are proposed to be newly added. We need to
1308 * make sure that those added links don't add too many
1309 * paths such that we will spend all our time waking up
1310 * eventpoll objects.
1311 *
1312 * Returns: Returns zero if the proposed links don't create too many paths,
1313 * -1 otherwise.
1314 */
1315static int reverse_path_check(void)
1316{
28d82dc1
JB
1317 int error = 0;
1318 struct file *current_file;
1319
1320 /* let's call this for all tfiles */
1321 list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
28d82dc1
JB
1322 path_count_init();
1323 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1324 reverse_path_check_proc, current_file,
1325 current_file, current);
1326 if (error)
1327 break;
1328 }
1329 return error;
1330}
1331
4d7e30d9
AH
1332static int ep_create_wakeup_source(struct epitem *epi)
1333{
1334 const char *name;
eea1d585 1335 struct wakeup_source *ws;
4d7e30d9
AH
1336
1337 if (!epi->ep->ws) {
1338 epi->ep->ws = wakeup_source_register("eventpoll");
1339 if (!epi->ep->ws)
1340 return -ENOMEM;
1341 }
1342
1343 name = epi->ffd.file->f_path.dentry->d_name.name;
eea1d585
EW
1344 ws = wakeup_source_register(name);
1345
1346 if (!ws)
4d7e30d9 1347 return -ENOMEM;
eea1d585 1348 rcu_assign_pointer(epi->ws, ws);
4d7e30d9
AH
1349
1350 return 0;
1351}
1352
eea1d585
EW
1353/* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1354static noinline void ep_destroy_wakeup_source(struct epitem *epi)
4d7e30d9 1355{
eea1d585
EW
1356 struct wakeup_source *ws = ep_wakeup_source(epi);
1357
d6d67e72 1358 RCU_INIT_POINTER(epi->ws, NULL);
eea1d585
EW
1359
1360 /*
1361 * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1362 * used internally by wakeup_source_remove, too (called by
1363 * wakeup_source_unregister), so we cannot use call_rcu
1364 */
1365 synchronize_rcu();
1366 wakeup_source_unregister(ws);
4d7e30d9
AH
1367}
1368
c7ea7630
DL
1369/*
1370 * Must be called with "mtx" held.
1371 */
1da177e4 1372static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
67347fe4 1373 struct file *tfile, int fd, int full_check)
1da177e4
LT
1374{
1375 int error, revents, pwake = 0;
1376 unsigned long flags;
52bd19f7 1377 long user_watches;
1da177e4
LT
1378 struct epitem *epi;
1379 struct ep_pqueue epq;
1380
52bd19f7
RH
1381 user_watches = atomic_long_read(&ep->user->epoll_watches);
1382 if (unlikely(user_watches >= max_user_watches))
7ef9964e 1383 return -ENOSPC;
e94b1766 1384 if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
7ef9964e 1385 return -ENOMEM;
1da177e4
LT
1386
1387 /* Item initialization follow here ... */
1da177e4
LT
1388 INIT_LIST_HEAD(&epi->rdllink);
1389 INIT_LIST_HEAD(&epi->fllink);
1da177e4
LT
1390 INIT_LIST_HEAD(&epi->pwqlist);
1391 epi->ep = ep;
b030a4dd 1392 ep_set_ffd(&epi->ffd, tfile, fd);
1da177e4 1393 epi->event = *event;
1da177e4 1394 epi->nwait = 0;
d47de16c 1395 epi->next = EP_UNACTIVE_PTR;
4d7e30d9
AH
1396 if (epi->event.events & EPOLLWAKEUP) {
1397 error = ep_create_wakeup_source(epi);
1398 if (error)
1399 goto error_create_wakeup_source;
1400 } else {
eea1d585 1401 RCU_INIT_POINTER(epi->ws, NULL);
4d7e30d9 1402 }
1da177e4
LT
1403
1404 /* Initialize the poll table using the queue callback */
1405 epq.epi = epi;
1406 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1407
1408 /*
1409 * Attach the item to the poll hooks and get current event bits.
1410 * We can safely use the file* here because its usage count has
c7ea7630
DL
1411 * been increased by the caller of this function. Note that after
1412 * this operation completes, the poll callback can start hitting
1413 * the new item.
1da177e4 1414 */
450d89ec 1415 revents = ep_item_poll(epi, &epq.pt);
1da177e4
LT
1416
1417 /*
1418 * We have to check if something went wrong during the poll wait queue
1419 * install process. Namely an allocation for a wait queue failed due
1420 * high memory pressure.
1421 */
7ef9964e 1422 error = -ENOMEM;
1da177e4 1423 if (epi->nwait < 0)
7699acd1 1424 goto error_unregister;
1da177e4
LT
1425
1426 /* Add the current item to the list of active epoll hook for this file */
68499914 1427 spin_lock(&tfile->f_lock);
ae10b2b4 1428 list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
68499914 1429 spin_unlock(&tfile->f_lock);
1da177e4 1430
c7ea7630
DL
1431 /*
1432 * Add the current item to the RB tree. All RB tree operations are
1433 * protected by "mtx", and ep_insert() is called with "mtx" held.
1434 */
1da177e4
LT
1435 ep_rbtree_insert(ep, epi);
1436
28d82dc1
JB
1437 /* now check if we've created too many backpaths */
1438 error = -EINVAL;
67347fe4 1439 if (full_check && reverse_path_check())
28d82dc1
JB
1440 goto error_remove_epi;
1441
c7ea7630
DL
1442 /* We have to drop the new item inside our item list to keep track of it */
1443 spin_lock_irqsave(&ep->lock, flags);
1444
bf3b9f63
SS
1445 /* record NAPI ID of new item if present */
1446 ep_set_busy_poll_napi_id(epi);
1447
1da177e4 1448 /* If the file is already "ready" we drop it inside the ready list */
b030a4dd 1449 if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
1da177e4 1450 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1451 ep_pm_stay_awake(epi);
1da177e4
LT
1452
1453 /* Notify waiting tasks that events are available */
1454 if (waitqueue_active(&ep->wq))
4a6e9e2c 1455 wake_up_locked(&ep->wq);
1da177e4
LT
1456 if (waitqueue_active(&ep->poll_wait))
1457 pwake++;
1458 }
1459
c7ea7630 1460 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 1461
52bd19f7 1462 atomic_long_inc(&ep->user->epoll_watches);
7ef9964e 1463
1da177e4
LT
1464 /* We have to call this outside the lock */
1465 if (pwake)
5071f97e 1466 ep_poll_safewake(&ep->poll_wait);
1da177e4 1467
1da177e4
LT
1468 return 0;
1469
28d82dc1
JB
1470error_remove_epi:
1471 spin_lock(&tfile->f_lock);
ae10b2b4 1472 list_del_rcu(&epi->fllink);
28d82dc1
JB
1473 spin_unlock(&tfile->f_lock);
1474
1475 rb_erase(&epi->rbn, &ep->rbr);
1476
7699acd1 1477error_unregister:
1da177e4
LT
1478 ep_unregister_pollwait(ep, epi);
1479
1480 /*
1481 * We need to do this because an event could have been arrived on some
67647d0f
DL
1482 * allocated wait queue. Note that we don't care about the ep->ovflist
1483 * list, since that is used/cleaned only inside a section bound by "mtx".
1484 * And ep_insert() is called with "mtx" held.
1da177e4 1485 */
c7ea7630 1486 spin_lock_irqsave(&ep->lock, flags);
b030a4dd 1487 if (ep_is_linked(&epi->rdllink))
6192bd53 1488 list_del_init(&epi->rdllink);
c7ea7630 1489 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4 1490
eea1d585 1491 wakeup_source_unregister(ep_wakeup_source(epi));
4d7e30d9
AH
1492
1493error_create_wakeup_source:
b030a4dd 1494 kmem_cache_free(epi_cache, epi);
7ef9964e 1495
1da177e4
LT
1496 return error;
1497}
1498
1da177e4
LT
1499/*
1500 * Modify the interest event mask by dropping an event if the new mask
c7ea7630 1501 * has a match in the current file status. Must be called with "mtx" held.
1da177e4
LT
1502 */
1503static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1504{
1505 int pwake = 0;
1506 unsigned int revents;
626cf236
HV
1507 poll_table pt;
1508
1509 init_poll_funcptr(&pt, NULL);
1da177e4
LT
1510
1511 /*
e057e15f
TB
1512 * Set the new event interest mask before calling f_op->poll();
1513 * otherwise we might miss an event that happens between the
1514 * f_op->poll() call and the new event set registering.
1da177e4 1515 */
128dd175 1516 epi->event.events = event->events; /* need barrier below */
e057e15f 1517 epi->event.data = event->data; /* protected by mtx */
4d7e30d9 1518 if (epi->event.events & EPOLLWAKEUP) {
eea1d585 1519 if (!ep_has_wakeup_source(epi))
4d7e30d9 1520 ep_create_wakeup_source(epi);
eea1d585 1521 } else if (ep_has_wakeup_source(epi)) {
4d7e30d9
AH
1522 ep_destroy_wakeup_source(epi);
1523 }
1da177e4 1524
128dd175
EW
1525 /*
1526 * The following barrier has two effects:
1527 *
1528 * 1) Flush epi changes above to other CPUs. This ensures
1529 * we do not miss events from ep_poll_callback if an
1530 * event occurs immediately after we call f_op->poll().
1531 * We need this because we did not take ep->lock while
1532 * changing epi above (but ep_poll_callback does take
1533 * ep->lock).
1534 *
1535 * 2) We also need to ensure we do not miss _past_ events
1536 * when calling f_op->poll(). This barrier also
1537 * pairs with the barrier in wq_has_sleeper (see
1538 * comments for wq_has_sleeper).
1539 *
1540 * This barrier will now guarantee ep_poll_callback or f_op->poll
1541 * (or both) will notice the readiness of an item.
1542 */
1543 smp_mb();
1544
1da177e4
LT
1545 /*
1546 * Get current event bits. We can safely use the file* here because
1547 * its usage count has been increased by the caller of this function.
1548 */
450d89ec 1549 revents = ep_item_poll(epi, &pt);
1da177e4 1550
1da177e4 1551 /*
c7ea7630 1552 * If the item is "hot" and it is not registered inside the ready
67647d0f 1553 * list, push it inside.
1da177e4 1554 */
c7ea7630 1555 if (revents & event->events) {
e057e15f 1556 spin_lock_irq(&ep->lock);
c7ea7630
DL
1557 if (!ep_is_linked(&epi->rdllink)) {
1558 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1559 ep_pm_stay_awake(epi);
c7ea7630
DL
1560
1561 /* Notify waiting tasks that events are available */
1562 if (waitqueue_active(&ep->wq))
4a6e9e2c 1563 wake_up_locked(&ep->wq);
c7ea7630
DL
1564 if (waitqueue_active(&ep->poll_wait))
1565 pwake++;
7699acd1 1566 }
e057e15f 1567 spin_unlock_irq(&ep->lock);
7699acd1 1568 }
1da177e4 1569
7699acd1
DL
1570 /* We have to call this outside the lock */
1571 if (pwake)
5071f97e 1572 ep_poll_safewake(&ep->poll_wait);
1da177e4 1573
7699acd1 1574 return 0;
1da177e4
LT
1575}
1576
296e236e
DL
1577static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1578 void *priv)
1da177e4 1579{
5071f97e
DL
1580 struct ep_send_events_data *esed = priv;
1581 int eventcnt;
296e236e 1582 unsigned int revents;
5071f97e
DL
1583 struct epitem *epi;
1584 struct epoll_event __user *uevent;
eea1d585 1585 struct wakeup_source *ws;
626cf236
HV
1586 poll_table pt;
1587
1588 init_poll_funcptr(&pt, NULL);
1da177e4 1589
296e236e 1590 /*
5071f97e
DL
1591 * We can loop without lock because we are passed a task private list.
1592 * Items cannot vanish during the loop because ep_scan_ready_list() is
1593 * holding "mtx" during this call.
296e236e 1594 */
5071f97e
DL
1595 for (eventcnt = 0, uevent = esed->events;
1596 !list_empty(head) && eventcnt < esed->maxevents;) {
1597 epi = list_first_entry(head, struct epitem, rdllink);
d47de16c 1598
4d7e30d9
AH
1599 /*
1600 * Activate ep->ws before deactivating epi->ws to prevent
1601 * triggering auto-suspend here (in case we reactive epi->ws
1602 * below).
1603 *
1604 * This could be rearranged to delay the deactivation of epi->ws
1605 * instead, but then epi->ws would temporarily be out of sync
1606 * with ep_is_linked().
1607 */
eea1d585
EW
1608 ws = ep_wakeup_source(epi);
1609 if (ws) {
1610 if (ws->active)
1611 __pm_stay_awake(ep->ws);
1612 __pm_relax(ws);
1613 }
1614
d47de16c 1615 list_del_init(&epi->rdllink);
1da177e4 1616
450d89ec 1617 revents = ep_item_poll(epi, &pt);
5071f97e 1618
296e236e 1619 /*
5071f97e
DL
1620 * If the event mask intersect the caller-requested one,
1621 * deliver the event to userspace. Again, ep_scan_ready_list()
1622 * is holding "mtx", so no operations coming from userspace
1623 * can change the item.
296e236e
DL
1624 */
1625 if (revents) {
5071f97e 1626 if (__put_user(revents, &uevent->events) ||
d0305882
TB
1627 __put_user(epi->event.data, &uevent->data)) {
1628 list_add(&epi->rdllink, head);
eea1d585 1629 ep_pm_stay_awake(epi);
296e236e 1630 return eventcnt ? eventcnt : -EFAULT;
d0305882 1631 }
296e236e 1632 eventcnt++;
5071f97e 1633 uevent++;
296e236e
DL
1634 if (epi->event.events & EPOLLONESHOT)
1635 epi->event.events &= EP_PRIVATE_BITS;
1636 else if (!(epi->event.events & EPOLLET)) {
1637 /*
1638 * If this file has been added with Level
1639 * Trigger mode, we need to insert back inside
1640 * the ready list, so that the next call to
1641 * epoll_wait() will check again the events
25985edc 1642 * availability. At this point, no one can insert
296e236e
DL
1643 * into ep->rdllist besides us. The epoll_ctl()
1644 * callers are locked out by
1645 * ep_scan_ready_list() holding "mtx" and the
1646 * poll callback will queue them in ep->ovflist.
1647 */
1648 list_add_tail(&epi->rdllink, &ep->rdllist);
eea1d585 1649 ep_pm_stay_awake(epi);
296e236e
DL
1650 }
1651 }
1652 }
5071f97e
DL
1653
1654 return eventcnt;
1655}
d47de16c 1656
296e236e
DL
1657static int ep_send_events(struct eventpoll *ep,
1658 struct epoll_event __user *events, int maxevents)
5071f97e
DL
1659{
1660 struct ep_send_events_data esed;
1da177e4 1661
5071f97e
DL
1662 esed.maxevents = maxevents;
1663 esed.events = events;
6192bd53 1664
67347fe4 1665 return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
1da177e4
LT
1666}
1667
766b9f92 1668static inline struct timespec64 ep_set_mstimeout(long ms)
0781b909 1669{
766b9f92 1670 struct timespec64 now, ts = {
0781b909
ED
1671 .tv_sec = ms / MSEC_PER_SEC,
1672 .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1673 };
1674
766b9f92
DD
1675 ktime_get_ts64(&now);
1676 return timespec64_add_safe(now, ts);
0781b909
ED
1677}
1678
f4d93ad7
SB
1679/**
1680 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1681 * event buffer.
1682 *
1683 * @ep: Pointer to the eventpoll context.
1684 * @events: Pointer to the userspace buffer where the ready events should be
1685 * stored.
1686 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1687 * @timeout: Maximum timeout for the ready events fetch operation, in
1688 * milliseconds. If the @timeout is zero, the function will not block,
1689 * while if the @timeout is less than zero, the function will block
1690 * until at least one event has been retrieved (or an error
1691 * occurred).
1692 *
1693 * Returns: Returns the number of ready events which have been fetched, or an
1694 * error code, in case of error.
1695 */
1da177e4
LT
1696static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1697 int maxevents, long timeout)
1698{
f4d93ad7 1699 int res = 0, eavail, timed_out = 0;
1da177e4 1700 unsigned long flags;
da8b44d5 1701 u64 slack = 0;
1da177e4 1702 wait_queue_t wait;
95aac7b1
SB
1703 ktime_t expires, *to = NULL;
1704
1705 if (timeout > 0) {
766b9f92 1706 struct timespec64 end_time = ep_set_mstimeout(timeout);
0781b909 1707
95aac7b1
SB
1708 slack = select_estimate_accuracy(&end_time);
1709 to = &expires;
766b9f92 1710 *to = timespec64_to_ktime(end_time);
95aac7b1 1711 } else if (timeout == 0) {
f4d93ad7
SB
1712 /*
1713 * Avoid the unnecessary trip to the wait queue loop, if the
1714 * caller specified a non blocking operation.
1715 */
95aac7b1 1716 timed_out = 1;
f4d93ad7
SB
1717 spin_lock_irqsave(&ep->lock, flags);
1718 goto check_events;
95aac7b1 1719 }
1da177e4 1720
f4d93ad7 1721fetch_events:
bf3b9f63
SS
1722
1723 if (!ep_events_available(ep))
1724 ep_busy_loop(ep, timed_out);
1725
c7ea7630 1726 spin_lock_irqsave(&ep->lock, flags);
1da177e4 1727
3fb0e584 1728 if (!ep_events_available(ep)) {
bf3b9f63
SS
1729 /*
1730 * Busy poll timed out. Drop NAPI ID for now, we can add
1731 * it back in when we have moved a socket with a valid NAPI
1732 * ID onto the ready list.
1733 */
1734 ep_reset_busy_poll_napi_id(ep);
1735
1da177e4
LT
1736 /*
1737 * We don't have any available event to return to the caller.
1738 * We need to sleep here, and we will be wake up by
1739 * ep_poll_callback() when events will become available.
1740 */
1741 init_waitqueue_entry(&wait, current);
a93d2f17 1742 __add_wait_queue_exclusive(&ep->wq, &wait);
1da177e4
LT
1743
1744 for (;;) {
1745 /*
1746 * We don't want to sleep if the ep_poll_callback() sends us
1747 * a wakeup in between. That's why we set the task state
1748 * to TASK_INTERRUPTIBLE before doing the checks.
1749 */
1750 set_current_state(TASK_INTERRUPTIBLE);
3fb0e584 1751 if (ep_events_available(ep) || timed_out)
1da177e4
LT
1752 break;
1753 if (signal_pending(current)) {
1754 res = -EINTR;
1755 break;
1756 }
1757
c7ea7630 1758 spin_unlock_irqrestore(&ep->lock, flags);
c511851d 1759 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
95aac7b1
SB
1760 timed_out = 1;
1761
c7ea7630 1762 spin_lock_irqsave(&ep->lock, flags);
1da177e4 1763 }
1da177e4 1764
4d5755b1
DB
1765 __remove_wait_queue(&ep->wq, &wait);
1766 __set_current_state(TASK_RUNNING);
1da177e4 1767 }
f4d93ad7 1768check_events:
1da177e4 1769 /* Is it worth to try to dig for events ? */
3fb0e584 1770 eavail = ep_events_available(ep);
1da177e4 1771
c7ea7630 1772 spin_unlock_irqrestore(&ep->lock, flags);
1da177e4
LT
1773
1774 /*
1775 * Try to transfer events to user space. In case we get 0 events and
1776 * there's still timeout left over, we go trying again in search of
1777 * more luck.
1778 */
1779 if (!res && eavail &&
95aac7b1 1780 !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
f4d93ad7 1781 goto fetch_events;
1da177e4
LT
1782
1783 return res;
1784}
1785
22bacca4
DL
1786/**
1787 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1788 * API, to verify that adding an epoll file inside another
1789 * epoll structure, does not violate the constraints, in
1790 * terms of closed loops, or too deep chains (which can
1791 * result in excessive stack usage).
1792 *
1793 * @priv: Pointer to the epoll file to be currently checked.
1794 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1795 * data structure pointer.
1796 * @call_nests: Current dept of the @ep_call_nested() call stack.
1797 *
1798 * Returns: Returns zero if adding the epoll @file inside current epoll
1799 * structure @ep does not violate the constraints, or -1 otherwise.
1800 */
1801static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1802{
1803 int error = 0;
1804 struct file *file = priv;
1805 struct eventpoll *ep = file->private_data;
28d82dc1 1806 struct eventpoll *ep_tovisit;
22bacca4
DL
1807 struct rb_node *rbp;
1808 struct epitem *epi;
1809
d8805e63 1810 mutex_lock_nested(&ep->mtx, call_nests + 1);
28d82dc1
JB
1811 ep->visited = 1;
1812 list_add(&ep->visited_list_link, &visited_list);
22bacca4
DL
1813 for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1814 epi = rb_entry(rbp, struct epitem, rbn);
1815 if (unlikely(is_file_epoll(epi->ffd.file))) {
28d82dc1
JB
1816 ep_tovisit = epi->ffd.file->private_data;
1817 if (ep_tovisit->visited)
1818 continue;
22bacca4 1819 error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
28d82dc1
JB
1820 ep_loop_check_proc, epi->ffd.file,
1821 ep_tovisit, current);
22bacca4
DL
1822 if (error != 0)
1823 break;
28d82dc1
JB
1824 } else {
1825 /*
1826 * If we've reached a file that is not associated with
1827 * an ep, then we need to check if the newly added
1828 * links are going to add too many wakeup paths. We do
1829 * this by adding it to the tfile_check_list, if it's
1830 * not already there, and calling reverse_path_check()
1831 * during ep_insert().
1832 */
1833 if (list_empty(&epi->ffd.file->f_tfile_llink))
1834 list_add(&epi->ffd.file->f_tfile_llink,
1835 &tfile_check_list);
22bacca4
DL
1836 }
1837 }
1838 mutex_unlock(&ep->mtx);
1839
1840 return error;
1841}
1842
1843/**
1844 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1845 * another epoll file (represented by @ep) does not create
1846 * closed loops or too deep chains.
1847 *
1848 * @ep: Pointer to the epoll private data structure.
1849 * @file: Pointer to the epoll file to be checked.
1850 *
1851 * Returns: Returns zero if adding the epoll @file inside current epoll
1852 * structure @ep does not violate the constraints, or -1 otherwise.
1853 */
1854static int ep_loop_check(struct eventpoll *ep, struct file *file)
1855{
28d82dc1
JB
1856 int ret;
1857 struct eventpoll *ep_cur, *ep_next;
1858
1859 ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
22bacca4 1860 ep_loop_check_proc, file, ep, current);
28d82dc1
JB
1861 /* clear visited list */
1862 list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1863 visited_list_link) {
1864 ep_cur->visited = 0;
1865 list_del(&ep_cur->visited_list_link);
1866 }
1867 return ret;
1868}
1869
1870static void clear_tfile_check_list(void)
1871{
1872 struct file *file;
1873
1874 /* first clear the tfile_check_list */
1875 while (!list_empty(&tfile_check_list)) {
1876 file = list_first_entry(&tfile_check_list, struct file,
1877 f_tfile_llink);
1878 list_del_init(&file->f_tfile_llink);
1879 }
1880 INIT_LIST_HEAD(&tfile_check_list);
22bacca4
DL
1881}
1882
7699acd1 1883/*
523723bb 1884 * Open an eventpoll file descriptor.
7699acd1 1885 */
5a8a82b1 1886SYSCALL_DEFINE1(epoll_create1, int, flags)
7699acd1 1887{
28d82dc1 1888 int error, fd;
bb57c3ed 1889 struct eventpoll *ep = NULL;
28d82dc1 1890 struct file *file;
7699acd1 1891
e38b36f3
UD
1892 /* Check the EPOLL_* constant for consistency. */
1893 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1894
296e236e
DL
1895 if (flags & ~EPOLL_CLOEXEC)
1896 return -EINVAL;
7699acd1 1897 /*
bb57c3ed 1898 * Create the internal data structure ("struct eventpoll").
7699acd1 1899 */
9fe5ad9c 1900 error = ep_alloc(&ep);
bb57c3ed
DL
1901 if (error < 0)
1902 return error;
7699acd1
DL
1903 /*
1904 * Creates all the items needed to setup an eventpoll file. That is,
2030a42c 1905 * a file structure and a free file descriptor.
7699acd1 1906 */
28d82dc1
JB
1907 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1908 if (fd < 0) {
1909 error = fd;
1910 goto out_free_ep;
1911 }
1912 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
628ff7c1 1913 O_RDWR | (flags & O_CLOEXEC));
28d82dc1
JB
1914 if (IS_ERR(file)) {
1915 error = PTR_ERR(file);
1916 goto out_free_fd;
1917 }
28d82dc1 1918 ep->file = file;
98022748 1919 fd_install(fd, file);
28d82dc1
JB
1920 return fd;
1921
1922out_free_fd:
1923 put_unused_fd(fd);
1924out_free_ep:
1925 ep_free(ep);
bb57c3ed 1926 return error;
7699acd1
DL
1927}
1928
5a8a82b1 1929SYSCALL_DEFINE1(epoll_create, int, size)
a0998b50 1930{
bfe3891a 1931 if (size <= 0)
9fe5ad9c
UD
1932 return -EINVAL;
1933
1934 return sys_epoll_create1(0);
a0998b50
UD
1935}
1936
7699acd1
DL
1937/*
1938 * The following function implements the controller interface for
1939 * the eventpoll file that enables the insertion/removal/change of
67647d0f 1940 * file descriptors inside the interest set.
7699acd1 1941 */
5a8a82b1
HC
1942SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1943 struct epoll_event __user *, event)
7699acd1
DL
1944{
1945 int error;
67347fe4 1946 int full_check = 0;
7e3fb584 1947 struct fd f, tf;
7699acd1
DL
1948 struct eventpoll *ep;
1949 struct epitem *epi;
1950 struct epoll_event epds;
67347fe4 1951 struct eventpoll *tep = NULL;
7699acd1 1952
7699acd1
DL
1953 error = -EFAULT;
1954 if (ep_op_has_event(op) &&
1955 copy_from_user(&epds, event, sizeof(struct epoll_event)))
1956 goto error_return;
1957
7699acd1 1958 error = -EBADF;
7e3fb584
AV
1959 f = fdget(epfd);
1960 if (!f.file)
7699acd1
DL
1961 goto error_return;
1962
1963 /* Get the "struct file *" for the target file */
7e3fb584
AV
1964 tf = fdget(fd);
1965 if (!tf.file)
7699acd1
DL
1966 goto error_fput;
1967
1968 /* The target file descriptor must support poll */
1969 error = -EPERM;
72c2d531 1970 if (!tf.file->f_op->poll)
7699acd1
DL
1971 goto error_tgt_fput;
1972
4d7e30d9 1973 /* Check if EPOLLWAKEUP is allowed */
c680e41b
NI
1974 if (ep_op_has_event(op))
1975 ep_take_care_of_epollwakeup(&epds);
4d7e30d9 1976
7699acd1
DL
1977 /*
1978 * We have to check that the file structure underneath the file descriptor
1979 * the user passed to us _is_ an eventpoll file. And also we do not permit
1980 * adding an epoll file descriptor inside itself.
1981 */
1982 error = -EINVAL;
7e3fb584 1983 if (f.file == tf.file || !is_file_epoll(f.file))
7699acd1
DL
1984 goto error_tgt_fput;
1985
df0108c5
JB
1986 /*
1987 * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only,
1988 * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation.
1989 * Also, we do not currently supported nested exclusive wakeups.
1990 */
c857ab64 1991 if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) {
b6a515c8
JB
1992 if (op == EPOLL_CTL_MOD)
1993 goto error_tgt_fput;
1994 if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) ||
1995 (epds.events & ~EPOLLEXCLUSIVE_OK_BITS)))
1996 goto error_tgt_fput;
1997 }
df0108c5 1998
7699acd1
DL
1999 /*
2000 * At this point it is safe to assume that the "private_data" contains
2001 * our own data structure.
2002 */
7e3fb584 2003 ep = f.file->private_data;
7699acd1 2004
22bacca4
DL
2005 /*
2006 * When we insert an epoll file descriptor, inside another epoll file
2007 * descriptor, there is the change of creating closed loops, which are
28d82dc1
JB
2008 * better be handled here, than in more critical paths. While we are
2009 * checking for loops we also determine the list of files reachable
2010 * and hang them on the tfile_check_list, so we can check that we
2011 * haven't created too many possible wakeup paths.
22bacca4 2012 *
67347fe4
JB
2013 * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
2014 * the epoll file descriptor is attaching directly to a wakeup source,
2015 * unless the epoll file descriptor is nested. The purpose of taking the
2016 * 'epmutex' on add is to prevent complex toplogies such as loops and
2017 * deep wakeup paths from forming in parallel through multiple
2018 * EPOLL_CTL_ADD operations.
22bacca4 2019 */
67347fe4 2020 mutex_lock_nested(&ep->mtx, 0);
28d82dc1 2021 if (op == EPOLL_CTL_ADD) {
67347fe4
JB
2022 if (!list_empty(&f.file->f_ep_links) ||
2023 is_file_epoll(tf.file)) {
2024 full_check = 1;
2025 mutex_unlock(&ep->mtx);
2026 mutex_lock(&epmutex);
2027 if (is_file_epoll(tf.file)) {
2028 error = -ELOOP;
2029 if (ep_loop_check(ep, tf.file) != 0) {
2030 clear_tfile_check_list();
2031 goto error_tgt_fput;
2032 }
2033 } else
2034 list_add(&tf.file->f_tfile_llink,
2035 &tfile_check_list);
2036 mutex_lock_nested(&ep->mtx, 0);
2037 if (is_file_epoll(tf.file)) {
2038 tep = tf.file->private_data;
2039 mutex_lock_nested(&tep->mtx, 1);
13d51807 2040 }
67347fe4
JB
2041 }
2042 }
7699acd1 2043
67647d0f
DL
2044 /*
2045 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
2046 * above, we can be sure to be able to use the item looked up by
2047 * ep_find() till we release the mutex.
2048 */
7e3fb584 2049 epi = ep_find(ep, tf.file, fd);
7699acd1
DL
2050
2051 error = -EINVAL;
2052 switch (op) {
2053 case EPOLL_CTL_ADD:
2054 if (!epi) {
2055 epds.events |= POLLERR | POLLHUP;
67347fe4 2056 error = ep_insert(ep, &epds, tf.file, fd, full_check);
7699acd1
DL
2057 } else
2058 error = -EEXIST;
67347fe4
JB
2059 if (full_check)
2060 clear_tfile_check_list();
7699acd1
DL
2061 break;
2062 case EPOLL_CTL_DEL:
2063 if (epi)
2064 error = ep_remove(ep, epi);
2065 else
2066 error = -ENOENT;
2067 break;
2068 case EPOLL_CTL_MOD:
2069 if (epi) {
b6a515c8
JB
2070 if (!(epi->event.events & EPOLLEXCLUSIVE)) {
2071 epds.events |= POLLERR | POLLHUP;
2072 error = ep_modify(ep, epi, &epds);
2073 }
7699acd1
DL
2074 } else
2075 error = -ENOENT;
2076 break;
2077 }
67347fe4
JB
2078 if (tep != NULL)
2079 mutex_unlock(&tep->mtx);
d47de16c 2080 mutex_unlock(&ep->mtx);
7699acd1
DL
2081
2082error_tgt_fput:
67347fe4 2083 if (full_check)
22bacca4
DL
2084 mutex_unlock(&epmutex);
2085
7e3fb584 2086 fdput(tf);
7699acd1 2087error_fput:
7e3fb584 2088 fdput(f);
7699acd1 2089error_return:
7699acd1
DL
2090
2091 return error;
2092}
2093
2094/*
2095 * Implement the event wait interface for the eventpoll file. It is the kernel
2096 * part of the user space epoll_wait(2).
2097 */
5a8a82b1
HC
2098SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
2099 int, maxevents, int, timeout)
7699acd1 2100{
2903ff01
AV
2101 int error;
2102 struct fd f;
7699acd1
DL
2103 struct eventpoll *ep;
2104
7699acd1
DL
2105 /* The maximum number of event must be greater than zero */
2106 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
2107 return -EINVAL;
2108
2109 /* Verify that the area passed by the user is writeable */
2903ff01
AV
2110 if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
2111 return -EFAULT;
7699acd1
DL
2112
2113 /* Get the "struct file *" for the eventpoll file */
2903ff01
AV
2114 f = fdget(epfd);
2115 if (!f.file)
2116 return -EBADF;
7699acd1
DL
2117
2118 /*
2119 * We have to check that the file structure underneath the fd
2120 * the user passed to us _is_ an eventpoll file.
2121 */
2122 error = -EINVAL;
2903ff01 2123 if (!is_file_epoll(f.file))
7699acd1
DL
2124 goto error_fput;
2125
2126 /*
2127 * At this point it is safe to assume that the "private_data" contains
2128 * our own data structure.
2129 */
2903ff01 2130 ep = f.file->private_data;
7699acd1
DL
2131
2132 /* Time to fish for events ... */
2133 error = ep_poll(ep, events, maxevents, timeout);
2134
2135error_fput:
2903ff01 2136 fdput(f);
7699acd1
DL
2137 return error;
2138}
2139
7699acd1
DL
2140/*
2141 * Implement the event wait interface for the eventpoll file. It is the kernel
2142 * part of the user space epoll_pwait(2).
2143 */
5a8a82b1
HC
2144SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2145 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2146 size_t, sigsetsize)
7699acd1
DL
2147{
2148 int error;
2149 sigset_t ksigmask, sigsaved;
2150
2151 /*
2152 * If the caller wants a certain signal mask to be set during the wait,
2153 * we apply it here.
2154 */
2155 if (sigmask) {
2156 if (sigsetsize != sizeof(sigset_t))
2157 return -EINVAL;
2158 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
2159 return -EFAULT;
77d55918
ON
2160 sigsaved = current->blocked;
2161 set_current_blocked(&ksigmask);
7699acd1
DL
2162 }
2163
2164 error = sys_epoll_wait(epfd, events, maxevents, timeout);
2165
2166 /*
2167 * If we changed the signal mask, we need to restore the original one.
2168 * In case we've got a signal while waiting, we do not restore the
2169 * signal mask yet, and we allow do_signal() to deliver the signal on
2170 * the way back to userspace, before the signal mask is restored.
2171 */
2172 if (sigmask) {
2173 if (error == -EINTR) {
2174 memcpy(&current->saved_sigmask, &sigsaved,
c7ea7630 2175 sizeof(sigsaved));
4e4c22c7 2176 set_restore_sigmask();
7699acd1 2177 } else
77d55918 2178 set_current_blocked(&sigsaved);
7699acd1
DL
2179 }
2180
2181 return error;
2182}
2183
35280bd4
AV
2184#ifdef CONFIG_COMPAT
2185COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2186 struct epoll_event __user *, events,
2187 int, maxevents, int, timeout,
2188 const compat_sigset_t __user *, sigmask,
2189 compat_size_t, sigsetsize)
2190{
2191 long err;
2192 compat_sigset_t csigmask;
2193 sigset_t ksigmask, sigsaved;
2194
2195 /*
2196 * If the caller wants a certain signal mask to be set during the wait,
2197 * we apply it here.
2198 */
2199 if (sigmask) {
2200 if (sigsetsize != sizeof(compat_sigset_t))
2201 return -EINVAL;
2202 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
2203 return -EFAULT;
2204 sigset_from_compat(&ksigmask, &csigmask);
77d55918
ON
2205 sigsaved = current->blocked;
2206 set_current_blocked(&ksigmask);
35280bd4
AV
2207 }
2208
2209 err = sys_epoll_wait(epfd, events, maxevents, timeout);
2210
2211 /*
2212 * If we changed the signal mask, we need to restore the original one.
2213 * In case we've got a signal while waiting, we do not restore the
2214 * signal mask yet, and we allow do_signal() to deliver the signal on
2215 * the way back to userspace, before the signal mask is restored.
2216 */
2217 if (sigmask) {
2218 if (err == -EINTR) {
2219 memcpy(&current->saved_sigmask, &sigsaved,
2220 sizeof(sigsaved));
2221 set_restore_sigmask();
2222 } else
77d55918 2223 set_current_blocked(&sigsaved);
35280bd4
AV
2224 }
2225
2226 return err;
2227}
2228#endif
2229
1da177e4
LT
2230static int __init eventpoll_init(void)
2231{
7ef9964e
DL
2232 struct sysinfo si;
2233
2234 si_meminfo(&si);
9df04e1f
DL
2235 /*
2236 * Allows top 4% of lomem to be allocated for epoll watches (per user).
2237 */
2238 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
7ef9964e 2239 EP_ITEM_COST;
52bd19f7 2240 BUG_ON(max_user_watches < 0);
1da177e4 2241
22bacca4
DL
2242 /*
2243 * Initialize the structure used to perform epoll file descriptor
2244 * inclusion loops checks.
2245 */
2246 ep_nested_calls_init(&poll_loop_ncalls);
2247
1da177e4 2248 /* Initialize the structure used to perform safe poll wait head wake ups */
5071f97e
DL
2249 ep_nested_calls_init(&poll_safewake_ncalls);
2250
2251 /* Initialize the structure used to perform file's f_op->poll() calls */
2252 ep_nested_calls_init(&poll_readywalk_ncalls);
1da177e4 2253
39732ca5
EW
2254 /*
2255 * We can have many thousands of epitems, so prevent this from
2256 * using an extra cache line on 64-bit (and smaller) CPUs
2257 */
2258 BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2259
1da177e4
LT
2260 /* Allocates slab cache used to allocate "struct epitem" items */
2261 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
bb57c3ed 2262 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1da177e4
LT
2263
2264 /* Allocates slab cache used to allocate "struct eppoll_entry" */
2265 pwq_cache = kmem_cache_create("eventpoll_pwq",
bb57c3ed 2266 sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
1da177e4 2267
1da177e4 2268 return 0;
1da177e4 2269}
cea69241 2270fs_initcall(eventpoll_init);