Branch data Line data Source code
1 : : // SPDX-License-Identifier: GPL-2.0-or-later
2 : : /*
3 : : * fs/eventpoll.c (Efficient event retrieval implementation)
4 : : * Copyright (C) 2001,...,2009 Davide Libenzi
5 : : *
6 : : * Davide Libenzi <davidel@xmailserver.org>
7 : : */
8 : :
9 : : #include <linux/init.h>
10 : : #include <linux/kernel.h>
11 : : #include <linux/sched/signal.h>
12 : : #include <linux/fs.h>
13 : : #include <linux/file.h>
14 : : #include <linux/signal.h>
15 : : #include <linux/errno.h>
16 : : #include <linux/mm.h>
17 : : #include <linux/slab.h>
18 : : #include <linux/poll.h>
19 : : #include <linux/string.h>
20 : : #include <linux/list.h>
21 : : #include <linux/hash.h>
22 : : #include <linux/spinlock.h>
23 : : #include <linux/syscalls.h>
24 : : #include <linux/rbtree.h>
25 : : #include <linux/wait.h>
26 : : #include <linux/eventpoll.h>
27 : : #include <linux/mount.h>
28 : : #include <linux/bitops.h>
29 : : #include <linux/mutex.h>
30 : : #include <linux/anon_inodes.h>
31 : : #include <linux/device.h>
32 : : #include <linux/uaccess.h>
33 : : #include <asm/io.h>
34 : : #include <asm/mman.h>
35 : : #include <linux/atomic.h>
36 : : #include <linux/proc_fs.h>
37 : : #include <linux/seq_file.h>
38 : : #include <linux/compat.h>
39 : : #include <linux/rculist.h>
40 : : #include <net/busy_poll.h>
41 : :
42 : : /*
43 : : * LOCKING:
44 : : * There are three level of locking required by epoll :
45 : : *
46 : : * 1) epmutex (mutex)
47 : : * 2) ep->mtx (mutex)
48 : : * 3) ep->lock (rwlock)
49 : : *
50 : : * The acquire order is the one listed above, from 1 to 3.
51 : : * We need a rwlock (ep->lock) because we manipulate objects
52 : : * from inside the poll callback, that might be triggered from
53 : : * a wake_up() that in turn might be called from IRQ context.
54 : : * So we can't sleep inside the poll callback and hence we need
55 : : * a spinlock. During the event transfer loop (from kernel to
56 : : * user space) we could end up sleeping due a copy_to_user(), so
57 : : * we need a lock that will allow us to sleep. This lock is a
58 : : * mutex (ep->mtx). It is acquired during the event transfer loop,
59 : : * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
60 : : * Then we also need a global mutex to serialize eventpoll_release_file()
61 : : * and ep_free().
62 : : * This mutex is acquired by ep_free() during the epoll file
63 : : * cleanup path and it is also acquired by eventpoll_release_file()
64 : : * if a file has been pushed inside an epoll set and it is then
65 : : * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
66 : : * It is also acquired when inserting an epoll fd onto another epoll
67 : : * fd. We do this so that we walk the epoll tree and ensure that this
68 : : * insertion does not create a cycle of epoll file descriptors, which
69 : : * could lead to deadlock. We need a global mutex to prevent two
70 : : * simultaneous inserts (A into B and B into A) from racing and
71 : : * constructing a cycle without either insert observing that it is
72 : : * going to.
73 : : * It is necessary to acquire multiple "ep->mtx"es at once in the
74 : : * case when one epoll fd is added to another. In this case, we
75 : : * always acquire the locks in the order of nesting (i.e. after
76 : : * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
77 : : * before e2->mtx). Since we disallow cycles of epoll file
78 : : * descriptors, this ensures that the mutexes are well-ordered. In
79 : : * order to communicate this nesting to lockdep, when walking a tree
80 : : * of epoll file descriptors, we use the current recursion depth as
81 : : * the lockdep subkey.
82 : : * It is possible to drop the "ep->mtx" and to use the global
83 : : * mutex "epmutex" (together with "ep->lock") to have it working,
84 : : * but having "ep->mtx" will make the interface more scalable.
85 : : * Events that require holding "epmutex" are very rare, while for
86 : : * normal operations the epoll private "ep->mtx" will guarantee
87 : : * a better scalability.
88 : : */
89 : :
90 : : /* Epoll private bits inside the event mask */
91 : : #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE)
92 : :
93 : : #define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT)
94 : :
95 : : #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \
96 : : EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE)
97 : :
98 : : /* Maximum number of nesting allowed inside epoll sets */
99 : : #define EP_MAX_NESTS 4
100 : :
101 : : #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
102 : :
103 : : #define EP_UNACTIVE_PTR ((void *) -1L)
104 : :
105 : : #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
106 : :
107 : : struct epoll_filefd {
108 : : struct file *file;
109 : : int fd;
110 : : } __packed;
111 : :
112 : : /*
113 : : * Structure used to track possible nested calls, for too deep recursions
114 : : * and loop cycles.
115 : : */
116 : : struct nested_call_node {
117 : : struct list_head llink;
118 : : void *cookie;
119 : : void *ctx;
120 : : };
121 : :
122 : : /*
123 : : * This structure is used as collector for nested calls, to check for
124 : : * maximum recursion dept and loop cycles.
125 : : */
126 : : struct nested_calls {
127 : : struct list_head tasks_call_list;
128 : : spinlock_t lock;
129 : : };
130 : :
131 : : /*
132 : : * Each file descriptor added to the eventpoll interface will
133 : : * have an entry of this type linked to the "rbr" RB tree.
134 : : * Avoid increasing the size of this struct, there can be many thousands
135 : : * of these on a server and we do not want this to take another cache line.
136 : : */
137 : : struct epitem {
138 : : union {
139 : : /* RB tree node links this structure to the eventpoll RB tree */
140 : : struct rb_node rbn;
141 : : /* Used to free the struct epitem */
142 : : struct rcu_head rcu;
143 : : };
144 : :
145 : : /* List header used to link this structure to the eventpoll ready list */
146 : : struct list_head rdllink;
147 : :
148 : : /*
149 : : * Works together "struct eventpoll"->ovflist in keeping the
150 : : * single linked chain of items.
151 : : */
152 : : struct epitem *next;
153 : :
154 : : /* The file descriptor information this item refers to */
155 : : struct epoll_filefd ffd;
156 : :
157 : : /* Number of active wait queue attached to poll operations */
158 : : int nwait;
159 : :
160 : : /* List containing poll wait queues */
161 : : struct list_head pwqlist;
162 : :
163 : : /* The "container" of this item */
164 : : struct eventpoll *ep;
165 : :
166 : : /* List header used to link this item to the "struct file" items list */
167 : : struct list_head fllink;
168 : :
169 : : /* wakeup_source used when EPOLLWAKEUP is set */
170 : : struct wakeup_source __rcu *ws;
171 : :
172 : : /* The structure that describe the interested events and the source fd */
173 : : struct epoll_event event;
174 : : };
175 : :
176 : : /*
177 : : * This structure is stored inside the "private_data" member of the file
178 : : * structure and represents the main data structure for the eventpoll
179 : : * interface.
180 : : */
181 : : struct eventpoll {
182 : : /*
183 : : * This mutex is used to ensure that files are not removed
184 : : * while epoll is using them. This is held during the event
185 : : * collection loop, the file cleanup path, the epoll file exit
186 : : * code and the ctl operations.
187 : : */
188 : : struct mutex mtx;
189 : :
190 : : /* Wait queue used by sys_epoll_wait() */
191 : : wait_queue_head_t wq;
192 : :
193 : : /* Wait queue used by file->poll() */
194 : : wait_queue_head_t poll_wait;
195 : :
196 : : /* List of ready file descriptors */
197 : : struct list_head rdllist;
198 : :
199 : : /* Lock which protects rdllist and ovflist */
200 : : rwlock_t lock;
201 : :
202 : : /* RB tree root used to store monitored fd structs */
203 : : struct rb_root_cached rbr;
204 : :
205 : : /*
206 : : * This is a single linked list that chains all the "struct epitem" that
207 : : * happened while transferring ready events to userspace w/out
208 : : * holding ->lock.
209 : : */
210 : : struct epitem *ovflist;
211 : :
212 : : /* wakeup_source used when ep_scan_ready_list is running */
213 : : struct wakeup_source *ws;
214 : :
215 : : /* The user that created the eventpoll descriptor */
216 : : struct user_struct *user;
217 : :
218 : : struct file *file;
219 : :
220 : : /* used to optimize loop detection check */
221 : : int visited;
222 : : struct list_head visited_list_link;
223 : :
224 : : #ifdef CONFIG_NET_RX_BUSY_POLL
225 : : /* used to track busy poll napi_id */
226 : : unsigned int napi_id;
227 : : #endif
228 : : };
229 : :
230 : : /* Wait structure used by the poll hooks */
231 : : struct eppoll_entry {
232 : : /* List header used to link this structure to the "struct epitem" */
233 : : struct list_head llink;
234 : :
235 : : /* The "base" pointer is set to the container "struct epitem" */
236 : : struct epitem *base;
237 : :
238 : : /*
239 : : * Wait queue item that will be linked to the target file wait
240 : : * queue head.
241 : : */
242 : : wait_queue_entry_t wait;
243 : :
244 : : /* The wait queue head that linked the "wait" wait queue item */
245 : : wait_queue_head_t *whead;
246 : : };
247 : :
248 : : /* Wrapper struct used by poll queueing */
249 : : struct ep_pqueue {
250 : : poll_table pt;
251 : : struct epitem *epi;
252 : : };
253 : :
254 : : /* Used by the ep_send_events() function as callback private data */
255 : : struct ep_send_events_data {
256 : : int maxevents;
257 : : struct epoll_event __user *events;
258 : : int res;
259 : : };
260 : :
261 : : /*
262 : : * Configuration options available inside /proc/sys/fs/epoll/
263 : : */
264 : : /* Maximum number of epoll watched descriptors, per user */
265 : : static long max_user_watches __read_mostly;
266 : :
267 : : /*
268 : : * This mutex is used to serialize ep_free() and eventpoll_release_file().
269 : : */
270 : : static DEFINE_MUTEX(epmutex);
271 : :
272 : : /* Used to check for epoll file descriptor inclusion loops */
273 : : static struct nested_calls poll_loop_ncalls;
274 : :
275 : : /* Slab cache used to allocate "struct epitem" */
276 : : static struct kmem_cache *epi_cache __read_mostly;
277 : :
278 : : /* Slab cache used to allocate "struct eppoll_entry" */
279 : : static struct kmem_cache *pwq_cache __read_mostly;
280 : :
281 : : /* Visited nodes during ep_loop_check(), so we can unset them when we finish */
282 : : static LIST_HEAD(visited_list);
283 : :
284 : : /*
285 : : * List of files with newly added links, where we may need to limit the number
286 : : * of emanating paths. Protected by the epmutex.
287 : : */
288 : : static LIST_HEAD(tfile_check_list);
289 : :
290 : : #ifdef CONFIG_SYSCTL
291 : :
292 : : #include <linux/sysctl.h>
293 : :
294 : : static long long_zero;
295 : : static long long_max = LONG_MAX;
296 : :
297 : : struct ctl_table epoll_table[] = {
298 : : {
299 : : .procname = "max_user_watches",
300 : : .data = &max_user_watches,
301 : : .maxlen = sizeof(max_user_watches),
302 : : .mode = 0644,
303 : : .proc_handler = proc_doulongvec_minmax,
304 : : .extra1 = &long_zero,
305 : : .extra2 = &long_max,
306 : : },
307 : : { }
308 : : };
309 : : #endif /* CONFIG_SYSCTL */
310 : :
311 : : static const struct file_operations eventpoll_fops;
312 : :
313 : 727230 : static inline int is_file_epoll(struct file *f)
314 : : {
315 : 727230 : return f->f_op == &eventpoll_fops;
316 : : }
317 : :
318 : : /* Setup the structure that is used as key for the RB tree */
319 : 39114 : static inline void ep_set_ffd(struct epoll_filefd *ffd,
320 : : struct file *file, int fd)
321 : : {
322 : 39114 : ffd->file = file;
323 : 39114 : ffd->fd = fd;
324 : : }
325 : :
326 : : /* Compare RB tree keys */
327 : 120441 : static inline int ep_cmp_ffd(struct epoll_filefd *p1,
328 : : struct epoll_filefd *p2)
329 : : {
330 : 120441 : return (p1->file > p2->file ? +1:
331 [ + + - + ]: 69220 : (p1->file < p2->file ? -1 : p1->fd - p2->fd));
332 : : }
333 : :
334 : : /* Tells us if the item is currently linked */
335 : 168684 : static inline int ep_is_linked(struct epitem *epi)
336 : : {
337 [ + - ]: 4579 : return !list_empty(&epi->rdllink);
338 : : }
339 : :
340 : 0 : static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p)
341 : : {
342 : 0 : return container_of(p, struct eppoll_entry, wait);
343 : : }
344 : :
345 : : /* Get the "struct epitem" from a wait queue pointer */
346 : 164579 : static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p)
347 : : {
348 : 164579 : return container_of(p, struct eppoll_entry, wait)->base;
349 : : }
350 : :
351 : : /* Get the "struct epitem" from an epoll queue wrapper */
352 : 15330 : static inline struct epitem *ep_item_from_epqueue(poll_table *p)
353 : : {
354 : 15330 : return container_of(p, struct ep_pqueue, pt)->epi;
355 : : }
356 : :
357 : : /* Initialize the poll safe wake up structure */
358 : 78 : static void ep_nested_calls_init(struct nested_calls *ncalls)
359 : : {
360 : 78 : INIT_LIST_HEAD(&ncalls->tasks_call_list);
361 : 78 : spin_lock_init(&ncalls->lock);
362 : : }
363 : :
364 : : /**
365 : : * ep_events_available - Checks if ready events might be available.
366 : : *
367 : : * @ep: Pointer to the eventpoll context.
368 : : *
369 : : * Returns: Returns a value different than zero if ready events are available,
370 : : * or zero otherwise.
371 : : */
372 : 1491532 : static inline int ep_events_available(struct eventpoll *ep)
373 : : {
374 [ - + + - : 192009 : return !list_empty_careful(&ep->rdllist) ||
+ - + - -
- ]
375 [ - + + - : 192009 : READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR;
+ - + - -
- ]
376 : : }
377 : :
378 : : #ifdef CONFIG_NET_RX_BUSY_POLL
379 : 0 : static bool ep_busy_loop_end(void *p, unsigned long start_time)
380 : : {
381 : 0 : struct eventpoll *ep = p;
382 : :
383 [ # # # # ]: 0 : return ep_events_available(ep) || busy_loop_timeout(start_time);
384 : : }
385 : :
386 : : /*
387 : : * Busy poll if globally on and supporting sockets found && no events,
388 : : * busy loop will return if need_resched or ep_events_available.
389 : : *
390 : : * we must do our busy polling with irqs enabled
391 : : */
392 : 63527 : static void ep_busy_loop(struct eventpoll *ep, int nonblock)
393 : : {
394 [ - + ]: 63527 : unsigned int napi_id = READ_ONCE(ep->napi_id);
395 : :
396 [ - + - - ]: 63527 : if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on())
397 [ # # ]: 0 : napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep);
398 : 63527 : }
399 : :
400 : 63527 : static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep)
401 : : {
402 : 63527 : if (ep->napi_id)
403 : 0 : ep->napi_id = 0;
404 : : }
405 : :
406 : : /*
407 : : * Set epoll busy poll NAPI ID from sk.
408 : : */
409 : : static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
410 : : {
411 : : struct eventpoll *ep;
412 : : unsigned int napi_id;
413 : : struct socket *sock;
414 : : struct sock *sk;
415 : : int err;
416 : :
417 : : if (!net_busy_loop_on())
418 : : return;
419 : :
420 : : sock = sock_from_file(epi->ffd.file, &err);
421 : : if (!sock)
422 : : return;
423 : :
424 : : sk = sock->sk;
425 : : if (!sk)
426 : : return;
427 : :
428 : : napi_id = READ_ONCE(sk->sk_napi_id);
429 : : ep = epi->ep;
430 : :
431 : : /* Non-NAPI IDs can be rejected
432 : : * or
433 : : * Nothing to do if we already have this ID
434 : : */
435 : : if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id)
436 : : return;
437 : :
438 : : /* record NAPI ID for use in next busy poll */
439 : : ep->napi_id = napi_id;
440 : : }
441 : :
442 : : #else
443 : :
444 : : static inline void ep_busy_loop(struct eventpoll *ep, int nonblock)
445 : : {
446 : : }
447 : :
448 : : static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep)
449 : : {
450 : : }
451 : :
452 : : static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
453 : : {
454 : : }
455 : :
456 : : #endif /* CONFIG_NET_RX_BUSY_POLL */
457 : :
458 : : /**
459 : : * ep_call_nested - Perform a bound (possibly) nested call, by checking
460 : : * that the recursion limit is not exceeded, and that
461 : : * the same nested call (by the meaning of same cookie) is
462 : : * no re-entered.
463 : : *
464 : : * @ncalls: Pointer to the nested_calls structure to be used for this call.
465 : : * @nproc: Nested call core function pointer.
466 : : * @priv: Opaque data to be passed to the @nproc callback.
467 : : * @cookie: Cookie to be used to identify this nested call.
468 : : * @ctx: This instance context.
469 : : *
470 : : * Returns: Returns the code returned by the @nproc callback, or -1 if
471 : : * the maximum recursion limit has been exceeded.
472 : : */
473 : 390 : static int ep_call_nested(struct nested_calls *ncalls,
474 : : int (*nproc)(void *, void *, int), void *priv,
475 : : void *cookie, void *ctx)
476 : : {
477 : 390 : int error, call_nests = 0;
478 : 390 : unsigned long flags;
479 : 390 : struct list_head *lsthead = &ncalls->tasks_call_list;
480 : 390 : struct nested_call_node *tncur;
481 : 390 : struct nested_call_node tnode;
482 : :
483 : 390 : spin_lock_irqsave(&ncalls->lock, flags);
484 : :
485 : : /*
486 : : * Try to see if the current task is already inside this wakeup call.
487 : : * We use a list here, since the population inside this set is always
488 : : * very much limited.
489 : : */
490 [ + + ]: 546 : list_for_each_entry(tncur, lsthead, llink) {
491 [ + - ]: 156 : if (tncur->ctx == ctx &&
492 [ + - - + ]: 156 : (tncur->cookie == cookie || ++call_nests > EP_MAX_NESTS)) {
493 : : /*
494 : : * Ops ... loop detected or maximum nest level reached.
495 : : * We abort this wake by breaking the cycle itself.
496 : : */
497 : 0 : error = -1;
498 : 0 : goto out_unlock;
499 : : }
500 : : }
501 : :
502 : : /* Add the current task and cookie to the list */
503 : 390 : tnode.ctx = ctx;
504 : 390 : tnode.cookie = cookie;
505 : 390 : list_add(&tnode.llink, lsthead);
506 : :
507 : 390 : spin_unlock_irqrestore(&ncalls->lock, flags);
508 : :
509 : : /* Call the nested function */
510 : 390 : error = (*nproc)(priv, cookie, call_nests);
511 : :
512 : : /* Remove the current task from the list */
513 : 390 : spin_lock_irqsave(&ncalls->lock, flags);
514 : 390 : list_del(&tnode.llink);
515 : 390 : out_unlock:
516 : 390 : spin_unlock_irqrestore(&ncalls->lock, flags);
517 : :
518 : 390 : return error;
519 : : }
520 : :
521 : : /*
522 : : * As described in commit 0ccf831cb lockdep: annotate epoll
523 : : * the use of wait queues used by epoll is done in a very controlled
524 : : * manner. Wake ups can nest inside each other, but are never done
525 : : * with the same locking. For example:
526 : : *
527 : : * dfd = socket(...);
528 : : * efd1 = epoll_create();
529 : : * efd2 = epoll_create();
530 : : * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
531 : : * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
532 : : *
533 : : * When a packet arrives to the device underneath "dfd", the net code will
534 : : * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
535 : : * callback wakeup entry on that queue, and the wake_up() performed by the
536 : : * "dfd" net code will end up in ep_poll_callback(). At this point epoll
537 : : * (efd1) notices that it may have some event ready, so it needs to wake up
538 : : * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
539 : : * that ends up in another wake_up(), after having checked about the
540 : : * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
541 : : * avoid stack blasting.
542 : : *
543 : : * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
544 : : * this special case of epoll.
545 : : */
546 : : #ifdef CONFIG_DEBUG_LOCK_ALLOC
547 : :
548 : : static DEFINE_PER_CPU(int, wakeup_nest);
549 : :
550 : : static void ep_poll_safewake(wait_queue_head_t *wq)
551 : : {
552 : : unsigned long flags;
553 : : int subclass;
554 : :
555 : : local_irq_save(flags);
556 : : preempt_disable();
557 : : subclass = __this_cpu_read(wakeup_nest);
558 : : spin_lock_nested(&wq->lock, subclass + 1);
559 : : __this_cpu_inc(wakeup_nest);
560 : : wake_up_locked_poll(wq, POLLIN);
561 : : __this_cpu_dec(wakeup_nest);
562 : : spin_unlock(&wq->lock);
563 : : local_irq_restore(flags);
564 : : preempt_enable();
565 : : }
566 : :
567 : : #else
568 : :
569 : 1014 : static void ep_poll_safewake(wait_queue_head_t *wq)
570 : : {
571 : 1014 : wake_up_poll(wq, EPOLLIN);
572 : 1014 : }
573 : :
574 : : #endif
575 : :
576 : : static void ep_remove_wait_queue(struct eppoll_entry *pwq)
577 : : {
578 : : wait_queue_head_t *whead;
579 : :
580 : : rcu_read_lock();
581 : : /*
582 : : * If it is cleared by POLLFREE, it should be rcu-safe.
583 : : * If we read NULL we need a barrier paired with
584 : : * smp_store_release() in ep_poll_callback(), otherwise
585 : : * we rely on whead->lock.
586 : : */
587 : : whead = smp_load_acquire(&pwq->whead);
588 : : if (whead)
589 : : remove_wait_queue(whead, &pwq->wait);
590 : : rcu_read_unlock();
591 : : }
592 : :
593 : : /*
594 : : * This function unregisters poll callbacks from the associated file
595 : : * descriptor. Must be called with "mtx" held (or "epmutex" if called from
596 : : * ep_free).
597 : : */
598 : : static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
599 : : {
600 : : struct list_head *lsthead = &epi->pwqlist;
601 : : struct eppoll_entry *pwq;
602 : :
603 : : while (!list_empty(lsthead)) {
604 : : pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
605 : :
606 : : list_del(&pwq->llink);
607 : : ep_remove_wait_queue(pwq);
608 : : kmem_cache_free(pwq_cache, pwq);
609 : : }
610 : : }
611 : :
612 : : /* call only when ep->mtx is held */
613 : 1568922 : static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
614 : : {
615 : 1568922 : return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
616 : : }
617 : :
618 : : /* call only when ep->mtx is held */
619 : 742484 : static inline void ep_pm_stay_awake(struct epitem *epi)
620 : : {
621 : 742484 : struct wakeup_source *ws = ep_wakeup_source(epi);
622 : :
623 [ - - - + : 742484 : if (ws)
- + - + -
+ ]
624 : 0 : __pm_stay_awake(ws);
625 : : }
626 : :
627 : 834 : static inline bool ep_has_wakeup_source(struct epitem *epi)
628 : : {
629 : 834 : return rcu_access_pointer(epi->ws) ? true : false;
630 : : }
631 : :
632 : : /* call when ep->mtx cannot be held (ep_poll_callback) */
633 : 82061 : static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
634 : : {
635 : 82061 : struct wakeup_source *ws;
636 : :
637 : 82061 : rcu_read_lock();
638 [ - + - + ]: 82061 : ws = rcu_dereference(epi->ws);
639 [ - + - + ]: 82061 : if (ws)
640 : 0 : __pm_stay_awake(ws);
641 : 82061 : rcu_read_unlock();
642 : 82061 : }
643 : :
644 : : /**
645 : : * ep_scan_ready_list - Scans the ready list in a way that makes possible for
646 : : * the scan code, to call f_op->poll(). Also allows for
647 : : * O(NumReady) performance.
648 : : *
649 : : * @ep: Pointer to the epoll private data structure.
650 : : * @sproc: Pointer to the scan callback.
651 : : * @priv: Private opaque data passed to the @sproc callback.
652 : : * @depth: The current depth of recursive f_op->poll calls.
653 : : * @ep_locked: caller already holds ep->mtx
654 : : *
655 : : * Returns: The same integer error code returned by the @sproc callback.
656 : : */
657 : 741224 : static __poll_t ep_scan_ready_list(struct eventpoll *ep,
658 : : __poll_t (*sproc)(struct eventpoll *,
659 : : struct list_head *, void *),
660 : : void *priv, int depth, bool ep_locked)
661 : : {
662 : 741224 : __poll_t res;
663 : 741224 : struct epitem *epi, *nepi;
664 : 741224 : LIST_HEAD(txlist);
665 : :
666 : 741224 : lockdep_assert_irqs_enabled();
667 : :
668 : : /*
669 : : * We need to lock this because we could be hit by
670 : : * eventpoll_release_file() and epoll_ctl().
671 : : */
672 : :
673 [ + + ]: 741224 : if (!ep_locked)
674 : 741146 : mutex_lock_nested(&ep->mtx, depth);
675 : :
676 : : /*
677 : : * Steal the ready list, and re-init the original one to the
678 : : * empty list. Also, set ep->ovflist to NULL so that events
679 : : * happening while looping w/out locks, are not lost. We cannot
680 : : * have the poll callback to queue directly on ep->rdllist,
681 : : * because we want the "sproc" callback to be able to do it
682 : : * in a lockless way.
683 : : */
684 : 741224 : write_lock_irq(&ep->lock);
685 [ + + ]: 741224 : list_splice_init(&ep->rdllist, &txlist);
686 : 741224 : WRITE_ONCE(ep->ovflist, NULL);
687 : 741224 : write_unlock_irq(&ep->lock);
688 : :
689 : : /*
690 : : * Now call the callback function.
691 : : */
692 : 741224 : res = (*sproc)(ep, &txlist, priv);
693 : :
694 : 741224 : write_lock_irq(&ep->lock);
695 : : /*
696 : : * During the time we spent inside the "sproc" callback, some
697 : : * other events might have been queued by the poll callback.
698 : : * We re-insert them inside the main ready-list here.
699 : : */
700 [ + + ]: 741229 : for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL;
701 : 5 : nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
702 : : /*
703 : : * We need to check if the item is already in the list.
704 : : * During the "sproc" callback execution time, items are
705 : : * queued into ->ovflist but the "txlist" might already
706 : : * contain them, and the list_splice() below takes care of them.
707 : : */
708 [ + - ]: 5 : if (!ep_is_linked(epi)) {
709 : : /*
710 : : * ->ovflist is LIFO, so we have to reverse it in order
711 : : * to keep in FIFO.
712 : : */
713 [ - + ]: 5 : list_add(&epi->rdllink, &ep->rdllist);
714 [ - + ]: 5 : ep_pm_stay_awake(epi);
715 : : }
716 : : }
717 : : /*
718 : : * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
719 : : * releasing the lock, events will be queued in the normal way inside
720 : : * ep->rdllist.
721 : : */
722 [ + + ]: 741224 : WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR);
723 : :
724 : : /*
725 : : * Quickly re-inject items left on "txlist".
726 : : */
727 [ + + ]: 741224 : list_splice(&txlist, &ep->rdllist);
728 : 741224 : __pm_relax(ep->ws);
729 : 741224 : write_unlock_irq(&ep->lock);
730 : :
731 [ + + ]: 741224 : if (!ep_locked)
732 : 741146 : mutex_unlock(&ep->mtx);
733 : :
734 : 741224 : return res;
735 : : }
736 : :
737 : 9412 : static void epi_rcu_free(struct rcu_head *head)
738 : : {
739 : 9412 : struct epitem *epi = container_of(head, struct epitem, rcu);
740 : 9412 : kmem_cache_free(epi_cache, epi);
741 : 9412 : }
742 : :
743 : : /*
744 : : * Removes a "struct epitem" from the eventpoll RB tree and deallocates
745 : : * all the associated resources. Must be called with "mtx" held.
746 : : */
747 : 9412 : static int ep_remove(struct eventpoll *ep, struct epitem *epi)
748 : : {
749 : 9412 : struct file *file = epi->ffd.file;
750 : :
751 : 9412 : lockdep_assert_irqs_enabled();
752 : :
753 : : /*
754 : : * Removes poll wait queue hooks.
755 : : */
756 : 9412 : ep_unregister_pollwait(ep, epi);
757 : :
758 : : /* Remove the current item from the list of epoll hooks */
759 : 9412 : spin_lock(&file->f_lock);
760 : 9412 : list_del_rcu(&epi->fllink);
761 : 9412 : spin_unlock(&file->f_lock);
762 : :
763 : 9412 : rb_erase_cached(&epi->rbn, &ep->rbr);
764 : :
765 : 9412 : write_lock_irq(&ep->lock);
766 [ + + ]: 9412 : if (ep_is_linked(epi))
767 : 7514 : list_del_init(&epi->rdllink);
768 : 9412 : write_unlock_irq(&ep->lock);
769 : :
770 : 9412 : wakeup_source_unregister(ep_wakeup_source(epi));
771 : : /*
772 : : * At this point it is safe to free the eventpoll item. Use the union
773 : : * field epi->rcu, since we are trying to minimize the size of
774 : : * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
775 : : * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
776 : : * use of the rbn field.
777 : : */
778 : 9412 : call_rcu(&epi->rcu, epi_rcu_free);
779 : :
780 : 9412 : atomic_long_dec(&ep->user->epoll_watches);
781 : :
782 : 9412 : return 0;
783 : : }
784 : :
785 : 818 : static void ep_free(struct eventpoll *ep)
786 : : {
787 : 818 : struct rb_node *rbp;
788 : 818 : struct epitem *epi;
789 : :
790 : : /* We need to release all tasks waiting for these file */
791 [ - + ]: 818 : if (waitqueue_active(&ep->poll_wait))
792 : 0 : ep_poll_safewake(&ep->poll_wait);
793 : :
794 : : /*
795 : : * We need to lock this because we could be hit by
796 : : * eventpoll_release_file() while we're freeing the "struct eventpoll".
797 : : * We do not need to hold "ep->mtx" here because the epoll file
798 : : * is on the way to be removed and no one has references to it
799 : : * anymore. The only hit might come from eventpoll_release_file() but
800 : : * holding "epmutex" is sufficient here.
801 : : */
802 : 818 : mutex_lock(&epmutex);
803 : :
804 : : /*
805 : : * Walks through the whole tree by unregistering poll callbacks.
806 : : */
807 [ + + ]: 1586 : for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
808 : 768 : epi = rb_entry(rbp, struct epitem, rbn);
809 : :
810 : 768 : ep_unregister_pollwait(ep, epi);
811 : 768 : cond_resched();
812 : : }
813 : :
814 : : /*
815 : : * Walks through the whole tree by freeing each "struct epitem". At this
816 : : * point we are sure no poll callbacks will be lingering around, and also by
817 : : * holding "epmutex" we can be sure that no file cleanup code will hit
818 : : * us during this operation. So we can avoid the lock on "ep->lock".
819 : : * We do not need to lock ep->mtx, either, we only do it to prevent
820 : : * a lockdep warning.
821 : : */
822 : 818 : mutex_lock(&ep->mtx);
823 [ + + ]: 1586 : while ((rbp = rb_first_cached(&ep->rbr)) != NULL) {
824 : 768 : epi = rb_entry(rbp, struct epitem, rbn);
825 : 768 : ep_remove(ep, epi);
826 : 768 : cond_resched();
827 : : }
828 : 818 : mutex_unlock(&ep->mtx);
829 : :
830 : 818 : mutex_unlock(&epmutex);
831 : 818 : mutex_destroy(&ep->mtx);
832 : 818 : free_uid(ep->user);
833 : 818 : wakeup_source_unregister(ep->ws);
834 : 818 : kfree(ep);
835 : 818 : }
836 : :
837 : 818 : static int ep_eventpoll_release(struct inode *inode, struct file *file)
838 : : {
839 : 818 : struct eventpoll *ep = file->private_data;
840 : :
841 [ + - ]: 818 : if (ep)
842 : 818 : ep_free(ep);
843 : :
844 : 818 : return 0;
845 : : }
846 : :
847 : : static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
848 : : void *priv);
849 : : static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
850 : : poll_table *pt);
851 : :
852 : : /*
853 : : * Differs from ep_eventpoll_poll() in that internal callers already have
854 : : * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested()
855 : : * is correctly annotated.
856 : : */
857 : : static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt,
858 : : int depth)
859 : : {
860 : : struct eventpoll *ep;
861 : : bool locked;
862 : :
863 : : pt->_key = epi->event.events;
864 : : if (!is_file_epoll(epi->ffd.file))
865 : : return vfs_poll(epi->ffd.file, pt) & epi->event.events;
866 : :
867 : : ep = epi->ffd.file->private_data;
868 : : poll_wait(epi->ffd.file, &ep->poll_wait, pt);
869 : : locked = pt && (pt->_qproc == ep_ptable_queue_proc);
870 : :
871 : : return ep_scan_ready_list(epi->ffd.file->private_data,
872 : : ep_read_events_proc, &depth, depth,
873 : : locked) & epi->event.events;
874 : : }
875 : :
876 : 540 : static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
877 : : void *priv)
878 : : {
879 : 540 : struct epitem *epi, *tmp;
880 : 540 : poll_table pt;
881 : 540 : int depth = *(int *)priv;
882 : :
883 : 540 : init_poll_funcptr(&pt, NULL);
884 : 540 : depth++;
885 : :
886 [ + + ]: 540 : list_for_each_entry_safe(epi, tmp, head, rdllink) {
887 [ - + ]: 234 : if (ep_item_poll(epi, &pt, depth)) {
888 : : return EPOLLIN | EPOLLRDNORM;
889 : : } else {
890 : : /*
891 : : * Item has been dropped into the ready list by the poll
892 : : * callback, but it's not actually ready, as far as
893 : : * caller requested events goes. We can remove it here.
894 : : */
895 : 0 : __pm_relax(ep_wakeup_source(epi));
896 : 0 : list_del_init(&epi->rdllink);
897 : : }
898 : : }
899 : :
900 : : return 0;
901 : : }
902 : :
903 : 0 : static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait)
904 : : {
905 : 0 : struct eventpoll *ep = file->private_data;
906 : 0 : int depth = 0;
907 : :
908 : : /* Insert inside our poll wait queue */
909 [ # # ]: 0 : poll_wait(file, &ep->poll_wait, wait);
910 : :
911 : : /*
912 : : * Proceed to find out if wanted events are really available inside
913 : : * the ready list.
914 : : */
915 : 0 : return ep_scan_ready_list(ep, ep_read_events_proc,
916 : : &depth, depth, false);
917 : : }
918 : :
919 : : #ifdef CONFIG_PROC_FS
920 : 0 : static void ep_show_fdinfo(struct seq_file *m, struct file *f)
921 : : {
922 : 0 : struct eventpoll *ep = f->private_data;
923 : 0 : struct rb_node *rbp;
924 : :
925 : 0 : mutex_lock(&ep->mtx);
926 [ # # ]: 0 : for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
927 : 0 : struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
928 : 0 : struct inode *inode = file_inode(epi->ffd.file);
929 : :
930 : 0 : seq_printf(m, "tfd: %8d events: %8x data: %16llx "
931 : : " pos:%lli ino:%lx sdev:%x\n",
932 : : epi->ffd.fd, epi->event.events,
933 : 0 : (long long)epi->event.data,
934 : 0 : (long long)epi->ffd.file->f_pos,
935 : 0 : inode->i_ino, inode->i_sb->s_dev);
936 [ # # ]: 0 : if (seq_has_overflowed(m))
937 : : break;
938 : : }
939 : 0 : mutex_unlock(&ep->mtx);
940 : 0 : }
941 : : #endif
942 : :
943 : : /* File callbacks that implement the eventpoll file behaviour */
944 : : static const struct file_operations eventpoll_fops = {
945 : : #ifdef CONFIG_PROC_FS
946 : : .show_fdinfo = ep_show_fdinfo,
947 : : #endif
948 : : .release = ep_eventpoll_release,
949 : : .poll = ep_eventpoll_poll,
950 : : .llseek = noop_llseek,
951 : : };
952 : :
953 : : /*
954 : : * This is called from eventpoll_release() to unlink files from the eventpoll
955 : : * interface. We need to have this facility to cleanup correctly files that are
956 : : * closed without being removed from the eventpoll interface.
957 : : */
958 : 868 : void eventpoll_release_file(struct file *file)
959 : : {
960 : 868 : struct eventpoll *ep;
961 : 868 : struct epitem *epi, *next;
962 : :
963 : : /*
964 : : * We don't want to get "file->f_lock" because it is not
965 : : * necessary. It is not necessary because we're in the "struct file"
966 : : * cleanup path, and this means that no one is using this file anymore.
967 : : * So, for example, epoll_ctl() cannot hit here since if we reach this
968 : : * point, the file counter already went to zero and fget() would fail.
969 : : * The only hit might come from ep_free() but by holding the mutex
970 : : * will correctly serialize the operation. We do need to acquire
971 : : * "ep->mtx" after "epmutex" because ep_remove() requires it when called
972 : : * from anywhere but ep_free().
973 : : *
974 : : * Besides, ep_remove() acquires the lock, so we can't hold it here.
975 : : */
976 : 868 : mutex_lock(&epmutex);
977 [ + + ]: 1736 : list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) {
978 : 868 : ep = epi->ep;
979 : 868 : mutex_lock_nested(&ep->mtx, 0);
980 : 868 : ep_remove(ep, epi);
981 : 868 : mutex_unlock(&ep->mtx);
982 : : }
983 : 868 : mutex_unlock(&epmutex);
984 : 868 : }
985 : :
986 : 1983 : static int ep_alloc(struct eventpoll **pep)
987 : : {
988 : 1983 : int error;
989 : 1983 : struct user_struct *user;
990 : 1983 : struct eventpoll *ep;
991 : :
992 : 1983 : user = get_current_user();
993 : 1983 : error = -ENOMEM;
994 : 1983 : ep = kzalloc(sizeof(*ep), GFP_KERNEL);
995 [ - + ]: 1983 : if (unlikely(!ep))
996 : 0 : goto free_uid;
997 : :
998 : 1983 : mutex_init(&ep->mtx);
999 : 1983 : rwlock_init(&ep->lock);
1000 : 1983 : init_waitqueue_head(&ep->wq);
1001 : 1983 : init_waitqueue_head(&ep->poll_wait);
1002 : 1983 : INIT_LIST_HEAD(&ep->rdllist);
1003 : 1983 : ep->rbr = RB_ROOT_CACHED;
1004 : 1983 : ep->ovflist = EP_UNACTIVE_PTR;
1005 : 1983 : ep->user = user;
1006 : :
1007 : 1983 : *pep = ep;
1008 : :
1009 : 1983 : return 0;
1010 : :
1011 : : free_uid:
1012 : 0 : free_uid(user);
1013 : 0 : return error;
1014 : : }
1015 : :
1016 : : /*
1017 : : * Search the file inside the eventpoll tree. The RB tree operations
1018 : : * are protected by the "mtx" mutex, and ep_find() must be called with
1019 : : * "mtx" held.
1020 : : */
1021 : 23862 : static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
1022 : : {
1023 : 23862 : int kcmp;
1024 : 23862 : struct rb_node *rbp;
1025 : 23862 : struct epitem *epi, *epir = NULL;
1026 : 23862 : struct epoll_filefd ffd;
1027 : :
1028 : 23862 : ep_set_ffd(&ffd, file, fd);
1029 [ + + ]: 92561 : for (rbp = ep->rbr.rb_root.rb_node; rbp; ) {
1030 : 77309 : epi = rb_entry(rbp, struct epitem, rbn);
1031 [ + + ]: 77309 : kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
1032 [ - + ]: 8610 : if (kcmp > 0)
1033 : 31068 : rbp = rbp->rb_right;
1034 [ - + ]: 8610 : else if (kcmp < 0)
1035 : 37631 : rbp = rbp->rb_left;
1036 : : else {
1037 : : epir = epi;
1038 : : break;
1039 : : }
1040 : : }
1041 : :
1042 : 23862 : return epir;
1043 : : }
1044 : :
1045 : : #ifdef CONFIG_CHECKPOINT_RESTORE
1046 : : static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff)
1047 : : {
1048 : : struct rb_node *rbp;
1049 : : struct epitem *epi;
1050 : :
1051 : : for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1052 : : epi = rb_entry(rbp, struct epitem, rbn);
1053 : : if (epi->ffd.fd == tfd) {
1054 : : if (toff == 0)
1055 : : return epi;
1056 : : else
1057 : : toff--;
1058 : : }
1059 : : cond_resched();
1060 : : }
1061 : :
1062 : : return NULL;
1063 : : }
1064 : :
1065 : : struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd,
1066 : : unsigned long toff)
1067 : : {
1068 : : struct file *file_raw;
1069 : : struct eventpoll *ep;
1070 : : struct epitem *epi;
1071 : :
1072 : : if (!is_file_epoll(file))
1073 : : return ERR_PTR(-EINVAL);
1074 : :
1075 : : ep = file->private_data;
1076 : :
1077 : : mutex_lock(&ep->mtx);
1078 : : epi = ep_find_tfd(ep, tfd, toff);
1079 : : if (epi)
1080 : : file_raw = epi->ffd.file;
1081 : : else
1082 : : file_raw = ERR_PTR(-ENOENT);
1083 : : mutex_unlock(&ep->mtx);
1084 : :
1085 : : return file_raw;
1086 : : }
1087 : : #endif /* CONFIG_CHECKPOINT_RESTORE */
1088 : :
1089 : : /**
1090 : : * Adds a new entry to the tail of the list in a lockless way, i.e.
1091 : : * multiple CPUs are allowed to call this function concurrently.
1092 : : *
1093 : : * Beware: it is necessary to prevent any other modifications of the
1094 : : * existing list until all changes are completed, in other words
1095 : : * concurrent list_add_tail_lockless() calls should be protected
1096 : : * with a read lock, where write lock acts as a barrier which
1097 : : * makes sure all list_add_tail_lockless() calls are fully
1098 : : * completed.
1099 : : *
1100 : : * Also an element can be locklessly added to the list only in one
1101 : : * direction i.e. either to the tail either to the head, otherwise
1102 : : * concurrent access will corrupt the list.
1103 : : *
1104 : : * Returns %false if element has been already added to the list, %true
1105 : : * otherwise.
1106 : : */
1107 : 82056 : static inline bool list_add_tail_lockless(struct list_head *new,
1108 : : struct list_head *head)
1109 : : {
1110 : 82056 : struct list_head *prev;
1111 : :
1112 : : /*
1113 : : * This is simple 'new->next = head' operation, but cmpxchg()
1114 : : * is used in order to detect that same element has been just
1115 : : * added to the list from another CPU: the winner observes
1116 : : * new->next == new.
1117 : : */
1118 [ + - ]: 82056 : if (cmpxchg(&new->next, new, head) != new)
1119 : : return false;
1120 : :
1121 : : /*
1122 : : * Initially ->next of a new element must be updated with the head
1123 : : * (we are inserting to the tail) and only then pointers are atomically
1124 : : * exchanged. XCHG guarantees memory ordering, thus ->next should be
1125 : : * updated before pointers are actually swapped and pointers are
1126 : : * swapped before prev->next is updated.
1127 : : */
1128 : :
1129 : 82056 : prev = xchg(&head->prev, new);
1130 : :
1131 : : /*
1132 : : * It is safe to modify prev->next and new->prev, because a new element
1133 : : * is added only to the tail and new->next is updated before XCHG.
1134 : : */
1135 : :
1136 : 82056 : prev->next = new;
1137 : 82056 : new->prev = prev;
1138 : :
1139 : 82056 : return true;
1140 : : }
1141 : :
1142 : : /**
1143 : : * Chains a new epi entry to the tail of the ep->ovflist in a lockless way,
1144 : : * i.e. multiple CPUs are allowed to call this function concurrently.
1145 : : *
1146 : : * Returns %false if epi element has been already chained, %true otherwise.
1147 : : */
1148 : 5 : static inline bool chain_epi_lockless(struct epitem *epi)
1149 : : {
1150 : 5 : struct eventpoll *ep = epi->ep;
1151 : :
1152 : : /* Check that the same epi has not been just chained from another CPU */
1153 [ + - ]: 5 : if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR)
1154 : : return false;
1155 : :
1156 : : /* Atomically exchange tail */
1157 : 5 : epi->next = xchg(&ep->ovflist, epi);
1158 : :
1159 : 5 : return true;
1160 : : }
1161 : :
1162 : : /*
1163 : : * This is the callback that is passed to the wait queue wakeup
1164 : : * mechanism. It is called by the stored file descriptors when they
1165 : : * have events to report.
1166 : : *
1167 : : * This callback takes a read lock in order not to content with concurrent
1168 : : * events from another file descriptors, thus all modifications to ->rdllist
1169 : : * or ->ovflist are lockless. Read lock is paired with the write lock from
1170 : : * ep_scan_ready_list(), which stops all list modifications and guarantees
1171 : : * that lists state is seen correctly.
1172 : : *
1173 : : * Another thing worth to mention is that ep_poll_callback() can be called
1174 : : * concurrently for the same @epi from different CPUs if poll table was inited
1175 : : * with several wait queues entries. Plural wakeup from different CPUs of a
1176 : : * single wait queue is serialized by wq.lock, but the case when multiple wait
1177 : : * queues are used should be detected accordingly. This is detected using
1178 : : * cmpxchg() operation.
1179 : : */
1180 : 164579 : static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1181 : : {
1182 : 164579 : int pwake = 0;
1183 : 164579 : struct epitem *epi = ep_item_from_wait(wait);
1184 : 164579 : struct eventpoll *ep = epi->ep;
1185 : 164579 : __poll_t pollflags = key_to_poll(key);
1186 : 164579 : unsigned long flags;
1187 : 164579 : int ewake = 0;
1188 : :
1189 : 164579 : read_lock_irqsave(&ep->lock, flags);
1190 : :
1191 : 164579 : ep_set_busy_poll_napi_id(epi);
1192 : :
1193 : : /*
1194 : : * If the event mask does not contain any poll(2) event, we consider the
1195 : : * descriptor to be disabled. This condition is likely the effect of the
1196 : : * EPOLLONESHOT bit that disables the descriptor when an event is received,
1197 : : * until the next EPOLL_CTL_MOD will be issued.
1198 : : */
1199 [ - + ]: 164579 : if (!(epi->event.events & ~EP_PRIVATE_BITS))
1200 : 0 : goto out_unlock;
1201 : :
1202 : : /*
1203 : : * Check the events coming with the callback. At this stage, not
1204 : : * every device reports the events in the "key" parameter of the
1205 : : * callback. We need to be able to handle both cases here, hence the
1206 : : * test for "key" != NULL before the event match test.
1207 : : */
1208 [ + + + + ]: 164579 : if (pollflags && !(pollflags & epi->event.events))
1209 : 10445 : goto out_unlock;
1210 : :
1211 : : /*
1212 : : * If we are transferring events to userspace, we can hold no locks
1213 : : * (because we're accessing user memory, and because of linux f_op->poll()
1214 : : * semantics). All the events that happen during that period of time are
1215 : : * chained in ep->ovflist and requeued later on.
1216 : : */
1217 [ + + ]: 154134 : if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) {
1218 [ + - + - ]: 10 : if (epi->next == EP_UNACTIVE_PTR &&
1219 : 5 : chain_epi_lockless(epi))
1220 : 5 : ep_pm_stay_awake_rcu(epi);
1221 : 5 : goto out_unlock;
1222 : : }
1223 : :
1224 : : /* If this file is already in the ready list we exit soon */
1225 [ + + + - ]: 236185 : if (!ep_is_linked(epi) &&
1226 : 82056 : list_add_tail_lockless(&epi->rdllink, &ep->rdllist)) {
1227 : 82056 : ep_pm_stay_awake_rcu(epi);
1228 : : }
1229 : :
1230 : : /*
1231 : : * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1232 : : * wait list.
1233 : : */
1234 [ + + ]: 154129 : if (waitqueue_active(&ep->wq)) {
1235 [ + - ]: 64656 : if ((epi->event.events & EPOLLEXCLUSIVE) &&
1236 [ # # ]: 0 : !(pollflags & POLLFREE)) {
1237 [ # # # # ]: 0 : switch (pollflags & EPOLLINOUT_BITS) {
1238 : 0 : case EPOLLIN:
1239 : 0 : if (epi->event.events & EPOLLIN)
1240 : : ewake = 1;
1241 : : break;
1242 : 0 : case EPOLLOUT:
1243 [ # # ]: 0 : if (epi->event.events & EPOLLOUT)
1244 : 0 : ewake = 1;
1245 : : break;
1246 : 0 : case 0:
1247 : 0 : ewake = 1;
1248 : 0 : break;
1249 : : }
1250 : 64656 : }
1251 : 64656 : wake_up(&ep->wq);
1252 : : }
1253 [ + + ]: 154129 : if (waitqueue_active(&ep->poll_wait))
1254 : 1014 : pwake++;
1255 : :
1256 : 153115 : out_unlock:
1257 : 164579 : read_unlock_irqrestore(&ep->lock, flags);
1258 : :
1259 : : /* We have to call this outside the lock */
1260 [ + + ]: 164579 : if (pwake)
1261 : 1014 : ep_poll_safewake(&ep->poll_wait);
1262 : :
1263 [ + - ]: 164579 : if (!(epi->event.events & EPOLLEXCLUSIVE))
1264 : 164579 : ewake = 1;
1265 : :
1266 [ - + ]: 164579 : if (pollflags & POLLFREE) {
1267 : : /*
1268 : : * If we race with ep_remove_wait_queue() it can miss
1269 : : * ->whead = NULL and do another remove_wait_queue() after
1270 : : * us, so we can't use __remove_wait_queue().
1271 : : */
1272 : 0 : list_del_init(&wait->entry);
1273 : : /*
1274 : : * ->whead != NULL protects us from the race with ep_free()
1275 : : * or ep_remove(), ep_remove_wait_queue() takes whead->lock
1276 : : * held by the caller. Once we nullify it, nothing protects
1277 : : * ep/epi or even wait.
1278 : : */
1279 : 0 : smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL);
1280 : : }
1281 : :
1282 : 164579 : return ewake;
1283 : : }
1284 : :
1285 : : /*
1286 : : * This is the callback that is used to add our wait queue to the
1287 : : * target file wakeup lists.
1288 : : */
1289 : 15330 : static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1290 : : poll_table *pt)
1291 : : {
1292 : 15330 : struct epitem *epi = ep_item_from_epqueue(pt);
1293 : 15330 : struct eppoll_entry *pwq;
1294 : :
1295 [ + - + - ]: 15330 : if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
1296 [ - + ]: 15330 : init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1297 : 15330 : pwq->whead = whead;
1298 : 15330 : pwq->base = epi;
1299 [ - + ]: 15330 : if (epi->event.events & EPOLLEXCLUSIVE)
1300 : 0 : add_wait_queue_exclusive(whead, &pwq->wait);
1301 : : else
1302 : 15330 : add_wait_queue(whead, &pwq->wait);
1303 : 15330 : list_add_tail(&pwq->llink, &epi->pwqlist);
1304 : 15330 : epi->nwait++;
1305 : : } else {
1306 : : /* We have to signal that an error occurred */
1307 : 0 : epi->nwait = -1;
1308 : : }
1309 : 15330 : }
1310 : :
1311 : 15252 : static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1312 : : {
1313 : 15252 : int kcmp;
1314 : 15252 : struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL;
1315 : 15252 : struct epitem *epic;
1316 : 15252 : bool leftmost = true;
1317 : :
1318 [ + + ]: 58384 : while (*p) {
1319 : 43132 : parent = *p;
1320 : 43132 : epic = rb_entry(parent, struct epitem, rbn);
1321 [ + + ]: 43132 : kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1322 [ # # ]: 0 : if (kcmp > 0) {
1323 : 20153 : p = &parent->rb_right;
1324 : 20153 : leftmost = false;
1325 : : } else
1326 : 22979 : p = &parent->rb_left;
1327 : : }
1328 [ + + ]: 15252 : rb_link_node(&epi->rbn, parent, p);
1329 [ + + ]: 15252 : rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost);
1330 : 15252 : }
1331 : :
1332 : :
1333 : :
1334 : : #define PATH_ARR_SIZE 5
1335 : : /*
1336 : : * These are the number paths of length 1 to 5, that we are allowing to emanate
1337 : : * from a single file of interest. For example, we allow 1000 paths of length
1338 : : * 1, to emanate from each file of interest. This essentially represents the
1339 : : * potential wakeup paths, which need to be limited in order to avoid massive
1340 : : * uncontrolled wakeup storms. The common use case should be a single ep which
1341 : : * is connected to n file sources. In this case each file source has 1 path
1342 : : * of length 1. Thus, the numbers below should be more than sufficient. These
1343 : : * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1344 : : * and delete can't add additional paths. Protected by the epmutex.
1345 : : */
1346 : : static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1347 : : static int path_count[PATH_ARR_SIZE];
1348 : :
1349 : 156 : static int path_count_inc(int nests)
1350 : : {
1351 : : /* Allow an arbitrary number of depth 1 paths */
1352 : 156 : if (nests == 0)
1353 : : return 0;
1354 : :
1355 [ + - ]: 156 : if (++path_count[nests] > path_limits[nests])
1356 : : return -1;
1357 : : return 0;
1358 : : }
1359 : :
1360 : : static void path_count_init(void)
1361 : : {
1362 : : int i;
1363 : :
1364 [ + + ]: 936 : for (i = 0; i < PATH_ARR_SIZE; i++)
1365 : 780 : path_count[i] = 0;
1366 : : }
1367 : :
1368 : 312 : static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1369 : : {
1370 : 312 : int error = 0;
1371 : 312 : struct file *file = priv;
1372 : 312 : struct file *child_file;
1373 : 312 : struct epitem *epi;
1374 : :
1375 : : /* CTL_DEL can remove links here, but that can't increase our count */
1376 : 312 : rcu_read_lock();
1377 [ + + ]: 624 : list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
1378 : 312 : child_file = epi->ep->file;
1379 [ + - ]: 312 : if (is_file_epoll(child_file)) {
1380 [ + + ]: 312 : if (list_empty(&child_file->f_ep_links)) {
1381 [ + - ]: 156 : if (path_count_inc(call_nests)) {
1382 : : error = -1;
1383 : : break;
1384 : : }
1385 : : } else {
1386 : 156 : error = ep_call_nested(&poll_loop_ncalls,
1387 : : reverse_path_check_proc,
1388 : : child_file, child_file,
1389 : : current);
1390 : : }
1391 [ + - ]: 312 : if (error != 0)
1392 : : break;
1393 : : } else {
1394 : 0 : printk(KERN_ERR "reverse_path_check_proc: "
1395 : : "file is not an ep!\n");
1396 : : }
1397 : : }
1398 : 312 : rcu_read_unlock();
1399 : 312 : return error;
1400 : : }
1401 : :
1402 : : /**
1403 : : * reverse_path_check - The tfile_check_list is list of file *, which have
1404 : : * links that are proposed to be newly added. We need to
1405 : : * make sure that those added links don't add too many
1406 : : * paths such that we will spend all our time waking up
1407 : : * eventpoll objects.
1408 : : *
1409 : : * Returns: Returns zero if the proposed links don't create too many paths,
1410 : : * -1 otherwise.
1411 : : */
1412 : 78 : static int reverse_path_check(void)
1413 : : {
1414 : 78 : int error = 0;
1415 : 78 : struct file *current_file;
1416 : :
1417 : : /* let's call this for all tfiles */
1418 [ + + ]: 234 : list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1419 : : path_count_init();
1420 : 156 : error = ep_call_nested(&poll_loop_ncalls,
1421 : : reverse_path_check_proc, current_file,
1422 : : current_file, current);
1423 [ + - ]: 156 : if (error)
1424 : : break;
1425 : : }
1426 : 78 : return error;
1427 : : }
1428 : :
1429 : 0 : static int ep_create_wakeup_source(struct epitem *epi)
1430 : : {
1431 : 0 : const char *name;
1432 : 0 : struct wakeup_source *ws;
1433 : :
1434 [ # # ]: 0 : if (!epi->ep->ws) {
1435 : 0 : epi->ep->ws = wakeup_source_register(NULL, "eventpoll");
1436 [ # # ]: 0 : if (!epi->ep->ws)
1437 : : return -ENOMEM;
1438 : : }
1439 : :
1440 : 0 : name = epi->ffd.file->f_path.dentry->d_name.name;
1441 : 0 : ws = wakeup_source_register(NULL, name);
1442 : :
1443 [ # # ]: 0 : if (!ws)
1444 : : return -ENOMEM;
1445 : 0 : rcu_assign_pointer(epi->ws, ws);
1446 : :
1447 : 0 : return 0;
1448 : : }
1449 : :
1450 : : /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1451 : 0 : static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1452 : : {
1453 : 0 : struct wakeup_source *ws = ep_wakeup_source(epi);
1454 : :
1455 : 0 : RCU_INIT_POINTER(epi->ws, NULL);
1456 : :
1457 : : /*
1458 : : * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1459 : : * used internally by wakeup_source_remove, too (called by
1460 : : * wakeup_source_unregister), so we cannot use call_rcu
1461 : : */
1462 : 0 : synchronize_rcu();
1463 : 0 : wakeup_source_unregister(ws);
1464 : 0 : }
1465 : :
1466 : : /*
1467 : : * Must be called with "mtx" held.
1468 : : */
1469 : 15252 : static int ep_insert(struct eventpoll *ep, const struct epoll_event *event,
1470 : : struct file *tfile, int fd, int full_check)
1471 : : {
1472 : 15252 : int error, pwake = 0;
1473 : 15252 : __poll_t revents;
1474 : 15252 : long user_watches;
1475 : 15252 : struct epitem *epi;
1476 : 15252 : struct ep_pqueue epq;
1477 : :
1478 : 15252 : lockdep_assert_irqs_enabled();
1479 : :
1480 : 15252 : user_watches = atomic_long_read(&ep->user->epoll_watches);
1481 [ + - ]: 15252 : if (unlikely(user_watches >= max_user_watches))
1482 : : return -ENOSPC;
1483 [ + - ]: 15252 : if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
1484 : : return -ENOMEM;
1485 : :
1486 : : /* Item initialization follow here ... */
1487 [ - + ]: 15252 : INIT_LIST_HEAD(&epi->rdllink);
1488 : 15252 : INIT_LIST_HEAD(&epi->fllink);
1489 : 15252 : INIT_LIST_HEAD(&epi->pwqlist);
1490 : 15252 : epi->ep = ep;
1491 : 15252 : ep_set_ffd(&epi->ffd, tfile, fd);
1492 : 15252 : epi->event = *event;
1493 : 15252 : epi->nwait = 0;
1494 : 15252 : epi->next = EP_UNACTIVE_PTR;
1495 [ - + ]: 15252 : if (epi->event.events & EPOLLWAKEUP) {
1496 : 0 : error = ep_create_wakeup_source(epi);
1497 [ # # ]: 0 : if (error)
1498 : 0 : goto error_create_wakeup_source;
1499 : : } else {
1500 : 15252 : RCU_INIT_POINTER(epi->ws, NULL);
1501 : : }
1502 : :
1503 : : /* Initialize the poll table using the queue callback */
1504 : 15252 : epq.epi = epi;
1505 : 15252 : init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1506 : :
1507 : : /*
1508 : : * Attach the item to the poll hooks and get current event bits.
1509 : : * We can safely use the file* here because its usage count has
1510 : : * been increased by the caller of this function. Note that after
1511 : : * this operation completes, the poll callback can start hitting
1512 : : * the new item.
1513 : : */
1514 : 15252 : revents = ep_item_poll(epi, &epq.pt, 1);
1515 : :
1516 : : /*
1517 : : * We have to check if something went wrong during the poll wait queue
1518 : : * install process. Namely an allocation for a wait queue failed due
1519 : : * high memory pressure.
1520 : : */
1521 : 15252 : error = -ENOMEM;
1522 [ - + ]: 15252 : if (epi->nwait < 0)
1523 : 0 : goto error_unregister;
1524 : :
1525 : : /* Add the current item to the list of active epoll hook for this file */
1526 : 15252 : spin_lock(&tfile->f_lock);
1527 : 15252 : list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
1528 : 15252 : spin_unlock(&tfile->f_lock);
1529 : :
1530 : : /*
1531 : : * Add the current item to the RB tree. All RB tree operations are
1532 : : * protected by "mtx", and ep_insert() is called with "mtx" held.
1533 : : */
1534 : 15252 : ep_rbtree_insert(ep, epi);
1535 : :
1536 : : /* now check if we've created too many backpaths */
1537 : 15252 : error = -EINVAL;
1538 [ + + - + ]: 15252 : if (full_check && reverse_path_check())
1539 : 0 : goto error_remove_epi;
1540 : :
1541 : : /* We have to drop the new item inside our item list to keep track of it */
1542 : 15252 : write_lock_irq(&ep->lock);
1543 : :
1544 : : /* record NAPI ID of new item if present */
1545 : 15252 : ep_set_busy_poll_napi_id(epi);
1546 : :
1547 : : /* If the file is already "ready" we drop it inside the ready list */
1548 [ + + + - ]: 15252 : if (revents && !ep_is_linked(epi)) {
1549 [ - + ]: 4579 : list_add_tail(&epi->rdllink, &ep->rdllist);
1550 [ - + ]: 4579 : ep_pm_stay_awake(epi);
1551 : :
1552 : : /* Notify waiting tasks that events are available */
1553 [ - + ]: 4579 : if (waitqueue_active(&ep->wq))
1554 : 0 : wake_up(&ep->wq);
1555 [ - + ]: 4579 : if (waitqueue_active(&ep->poll_wait))
1556 : 0 : pwake++;
1557 : : }
1558 : :
1559 : 15252 : write_unlock_irq(&ep->lock);
1560 : :
1561 : 15252 : atomic_long_inc(&ep->user->epoll_watches);
1562 : :
1563 : : /* We have to call this outside the lock */
1564 [ - + ]: 15252 : if (pwake)
1565 : 0 : ep_poll_safewake(&ep->poll_wait);
1566 : :
1567 : : return 0;
1568 : :
1569 : : error_remove_epi:
1570 : 0 : spin_lock(&tfile->f_lock);
1571 : 0 : list_del_rcu(&epi->fllink);
1572 : 0 : spin_unlock(&tfile->f_lock);
1573 : :
1574 : 0 : rb_erase_cached(&epi->rbn, &ep->rbr);
1575 : :
1576 : 0 : error_unregister:
1577 : 0 : ep_unregister_pollwait(ep, epi);
1578 : :
1579 : : /*
1580 : : * We need to do this because an event could have been arrived on some
1581 : : * allocated wait queue. Note that we don't care about the ep->ovflist
1582 : : * list, since that is used/cleaned only inside a section bound by "mtx".
1583 : : * And ep_insert() is called with "mtx" held.
1584 : : */
1585 : 0 : write_lock_irq(&ep->lock);
1586 [ # # ]: 0 : if (ep_is_linked(epi))
1587 : 0 : list_del_init(&epi->rdllink);
1588 : 0 : write_unlock_irq(&ep->lock);
1589 : :
1590 : 0 : wakeup_source_unregister(ep_wakeup_source(epi));
1591 : :
1592 : 0 : error_create_wakeup_source:
1593 : 0 : kmem_cache_free(epi_cache, epi);
1594 : :
1595 : 0 : return error;
1596 : : }
1597 : :
1598 : : /*
1599 : : * Modify the interest event mask by dropping an event if the new mask
1600 : : * has a match in the current file status. Must be called with "mtx" held.
1601 : : */
1602 : 834 : static int ep_modify(struct eventpoll *ep, struct epitem *epi,
1603 : : const struct epoll_event *event)
1604 : : {
1605 : 834 : int pwake = 0;
1606 : 834 : poll_table pt;
1607 : :
1608 : 834 : lockdep_assert_irqs_enabled();
1609 : :
1610 [ - + ]: 834 : init_poll_funcptr(&pt, NULL);
1611 : :
1612 : : /*
1613 : : * Set the new event interest mask before calling f_op->poll();
1614 : : * otherwise we might miss an event that happens between the
1615 : : * f_op->poll() call and the new event set registering.
1616 : : */
1617 : 834 : epi->event.events = event->events; /* need barrier below */
1618 : 834 : epi->event.data = event->data; /* protected by mtx */
1619 [ - + ]: 834 : if (epi->event.events & EPOLLWAKEUP) {
1620 [ # # ]: 0 : if (!ep_has_wakeup_source(epi))
1621 : 0 : ep_create_wakeup_source(epi);
1622 [ - + ]: 834 : } else if (ep_has_wakeup_source(epi)) {
1623 : 0 : ep_destroy_wakeup_source(epi);
1624 : : }
1625 : :
1626 : : /*
1627 : : * The following barrier has two effects:
1628 : : *
1629 : : * 1) Flush epi changes above to other CPUs. This ensures
1630 : : * we do not miss events from ep_poll_callback if an
1631 : : * event occurs immediately after we call f_op->poll().
1632 : : * We need this because we did not take ep->lock while
1633 : : * changing epi above (but ep_poll_callback does take
1634 : : * ep->lock).
1635 : : *
1636 : : * 2) We also need to ensure we do not miss _past_ events
1637 : : * when calling f_op->poll(). This barrier also
1638 : : * pairs with the barrier in wq_has_sleeper (see
1639 : : * comments for wq_has_sleeper).
1640 : : *
1641 : : * This barrier will now guarantee ep_poll_callback or f_op->poll
1642 : : * (or both) will notice the readiness of an item.
1643 : : */
1644 : 834 : smp_mb();
1645 : :
1646 : : /*
1647 : : * Get current event bits. We can safely use the file* here because
1648 : : * its usage count has been increased by the caller of this function.
1649 : : * If the item is "hot" and it is not registered inside the ready
1650 : : * list, push it inside.
1651 : : */
1652 [ + + ]: 834 : if (ep_item_poll(epi, &pt, 1)) {
1653 : 559 : write_lock_irq(&ep->lock);
1654 [ + + ]: 559 : if (!ep_is_linked(epi)) {
1655 [ - + ]: 102 : list_add_tail(&epi->rdllink, &ep->rdllist);
1656 [ - + ]: 102 : ep_pm_stay_awake(epi);
1657 : :
1658 : : /* Notify waiting tasks that events are available */
1659 [ - + ]: 102 : if (waitqueue_active(&ep->wq))
1660 : 0 : wake_up(&ep->wq);
1661 [ - + ]: 102 : if (waitqueue_active(&ep->poll_wait))
1662 : 0 : pwake++;
1663 : : }
1664 : 559 : write_unlock_irq(&ep->lock);
1665 : : }
1666 : :
1667 : : /* We have to call this outside the lock */
1668 [ - + ]: 559 : if (pwake)
1669 : 0 : ep_poll_safewake(&ep->poll_wait);
1670 : :
1671 : 834 : return 0;
1672 : : }
1673 : :
1674 : 740684 : static __poll_t ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1675 : : void *priv)
1676 : : {
1677 : 740684 : struct ep_send_events_data *esed = priv;
1678 : 740684 : __poll_t revents;
1679 : 740684 : struct epitem *epi, *tmp;
1680 : 740684 : struct epoll_event __user *uevent = esed->events;
1681 : 740684 : struct wakeup_source *ws;
1682 : 740684 : poll_table pt;
1683 : :
1684 : 740684 : init_poll_funcptr(&pt, NULL);
1685 : 740684 : esed->res = 0;
1686 : :
1687 : : /*
1688 : : * We can loop without lock because we are passed a task private list.
1689 : : * Items cannot vanish during the loop because ep_scan_ready_list() is
1690 : : * holding "mtx" during this call.
1691 : : */
1692 : 740684 : lockdep_assert_held(&ep->mtx);
1693 : :
1694 [ + + ]: 1557710 : list_for_each_entry_safe(epi, tmp, head, rdllink) {
1695 [ + + ]: 817160 : if (esed->res >= esed->maxevents)
1696 : : break;
1697 : :
1698 : : /*
1699 : : * Activate ep->ws before deactivating epi->ws to prevent
1700 : : * triggering auto-suspend here (in case we reactive epi->ws
1701 : : * below).
1702 : : *
1703 : : * This could be rearranged to delay the deactivation of epi->ws
1704 : : * instead, but then epi->ws would temporarily be out of sync
1705 : : * with ep_is_linked().
1706 : : */
1707 [ - + ]: 817026 : ws = ep_wakeup_source(epi);
1708 [ - + ]: 817026 : if (ws) {
1709 [ # # ]: 0 : if (ws->active)
1710 : 0 : __pm_stay_awake(ep->ws);
1711 : 0 : __pm_relax(ws);
1712 : : }
1713 : :
1714 : 817026 : list_del_init(&epi->rdllink);
1715 : :
1716 : : /*
1717 : : * If the event mask intersect the caller-requested one,
1718 : : * deliver the event to userspace. Again, ep_scan_ready_list()
1719 : : * is holding ep->mtx, so no operations coming from userspace
1720 : : * can change the item.
1721 : : */
1722 : 817026 : revents = ep_item_poll(epi, &pt, 1);
1723 [ + + ]: 817026 : if (!revents)
1724 : 78916 : continue;
1725 : :
1726 [ + - + - ]: 738110 : if (__put_user(revents, &uevent->events) ||
1727 [ + - - + ]: 738110 : __put_user(epi->event.data, &uevent->data)) {
1728 [ # # ]: 0 : list_add(&epi->rdllink, head);
1729 [ # # ]: 0 : ep_pm_stay_awake(epi);
1730 [ # # ]: 0 : if (!esed->res)
1731 : 0 : esed->res = -EFAULT;
1732 : 0 : return 0;
1733 : : }
1734 : 738110 : esed->res++;
1735 : 738110 : uevent++;
1736 [ - + ]: 738110 : if (epi->event.events & EPOLLONESHOT)
1737 : 0 : epi->event.events &= EP_PRIVATE_BITS;
1738 [ + + ]: 738110 : else if (!(epi->event.events & EPOLLET)) {
1739 : : /*
1740 : : * If this file has been added with Level
1741 : : * Trigger mode, we need to insert back inside
1742 : : * the ready list, so that the next call to
1743 : : * epoll_wait() will check again the events
1744 : : * availability. At this point, no one can insert
1745 : : * into ep->rdllist besides us. The epoll_ctl()
1746 : : * callers are locked out by
1747 : : * ep_scan_ready_list() holding "mtx" and the
1748 : : * poll callback will queue them in ep->ovflist.
1749 : : */
1750 [ - + ]: 737798 : list_add_tail(&epi->rdllink, &ep->rdllist);
1751 [ - + ]: 737798 : ep_pm_stay_awake(epi);
1752 : : }
1753 : : }
1754 : :
1755 : : return 0;
1756 : : }
1757 : :
1758 : 740684 : static int ep_send_events(struct eventpoll *ep,
1759 : : struct epoll_event __user *events, int maxevents)
1760 : : {
1761 : 740684 : struct ep_send_events_data esed;
1762 : :
1763 : 740684 : esed.maxevents = maxevents;
1764 : 740684 : esed.events = events;
1765 : :
1766 : 740684 : ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
1767 : 740684 : return esed.res;
1768 : : }
1769 : :
1770 : 0 : static inline struct timespec64 ep_set_mstimeout(long ms)
1771 : : {
1772 : 0 : struct timespec64 now, ts = {
1773 : 0 : .tv_sec = ms / MSEC_PER_SEC,
1774 : 0 : .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1775 : : };
1776 : :
1777 : 0 : ktime_get_ts64(&now);
1778 : 0 : return timespec64_add_safe(now, ts);
1779 : : }
1780 : :
1781 : : /**
1782 : : * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1783 : : * event buffer.
1784 : : *
1785 : : * @ep: Pointer to the eventpoll context.
1786 : : * @events: Pointer to the userspace buffer where the ready events should be
1787 : : * stored.
1788 : : * @maxevents: Size (in terms of number of events) of the caller event buffer.
1789 : : * @timeout: Maximum timeout for the ready events fetch operation, in
1790 : : * milliseconds. If the @timeout is zero, the function will not block,
1791 : : * while if the @timeout is less than zero, the function will block
1792 : : * until at least one event has been retrieved (or an error
1793 : : * occurred).
1794 : : *
1795 : : * Returns: Returns the number of ready events which have been fetched, or an
1796 : : * error code, in case of error.
1797 : : */
1798 : 687492 : static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1799 : : int maxevents, long timeout)
1800 : : {
1801 : 687492 : int res = 0, eavail, timed_out = 0;
1802 : 687492 : u64 slack = 0;
1803 : 687492 : bool waiter = false;
1804 : 687492 : wait_queue_entry_t wait;
1805 : 687492 : ktime_t expires, *to = NULL;
1806 : :
1807 : 687492 : lockdep_assert_irqs_enabled();
1808 : :
1809 [ - + ]: 687492 : if (timeout > 0) {
1810 : 0 : struct timespec64 end_time = ep_set_mstimeout(timeout);
1811 : :
1812 : 0 : slack = select_estimate_accuracy(&end_time);
1813 : 0 : to = &expires;
1814 [ # # ]: 0 : *to = timespec64_to_ktime(end_time);
1815 [ + + ]: 687492 : } else if (timeout == 0) {
1816 : : /*
1817 : : * Avoid the unnecessary trip to the wait queue loop, if the
1818 : : * caller specified a non blocking operation. We still need
1819 : : * lock because we could race and not see an epi being added
1820 : : * to the ready list while in irq callback. Thus incorrectly
1821 : : * returning 0 back to userspace.
1822 : : */
1823 : 120833 : timed_out = 1;
1824 : :
1825 : 120833 : write_lock_irq(&ep->lock);
1826 [ + + ]: 120833 : eavail = ep_events_available(ep);
1827 : 120833 : write_unlock_irq(&ep->lock);
1828 : :
1829 : 120833 : goto send_events;
1830 : : }
1831 : :
1832 : 566659 : fetch_events:
1833 : :
1834 [ + + ]: 622366 : if (!ep_events_available(ep))
1835 : 63527 : ep_busy_loop(ep, timed_out);
1836 : :
1837 [ + + ]: 622366 : eavail = ep_events_available(ep);
1838 : 63527 : if (eavail)
1839 : 558839 : goto send_events;
1840 : :
1841 : : /*
1842 : : * Busy poll timed out. Drop NAPI ID for now, we can add
1843 : : * it back in when we have moved a socket with a valid NAPI
1844 : : * ID onto the ready list.
1845 : : */
1846 [ - + ]: 63527 : ep_reset_busy_poll_napi_id(ep);
1847 : :
1848 : : /*
1849 : : * We don't have any available event to return to the caller. We need
1850 : : * to sleep here, and we will be woken by ep_poll_callback() when events
1851 : : * become available.
1852 : : */
1853 [ + + ]: 63527 : if (!waiter) {
1854 : 62606 : waiter = true;
1855 : 62606 : init_waitqueue_entry(&wait, current);
1856 : :
1857 : 62606 : write_lock_irq(&ep->lock);
1858 : 62606 : __add_wait_queue_exclusive(&ep->wq, &wait);
1859 : 62606 : write_unlock_irq(&ep->lock);
1860 : : }
1861 : :
1862 : 125967 : for (;;) {
1863 : : /*
1864 : : * We don't want to sleep if the ep_poll_callback() sends us
1865 : : * a wakeup in between. That's why we set the task state
1866 : : * to TASK_INTERRUPTIBLE before doing the checks.
1867 : : */
1868 : 125967 : set_current_state(TASK_INTERRUPTIBLE);
1869 : : /*
1870 : : * Always short-circuit for fatal signals to allow
1871 : : * threads to make a timely exit without the chance of
1872 : : * finding more events available and fetching
1873 : : * repeatedly.
1874 : : */
1875 [ + - ]: 125967 : if (fatal_signal_pending(current)) {
1876 : : res = -EINTR;
1877 : : break;
1878 : : }
1879 : :
1880 [ + + ]: 125967 : eavail = ep_events_available(ep);
1881 : 63527 : if (eavail)
1882 : : break;
1883 [ + - ]: 63527 : if (signal_pending(current)) {
1884 : : res = -EINTR;
1885 : : break;
1886 : : }
1887 : :
1888 [ + - ]: 63527 : if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) {
1889 : : timed_out = 1;
1890 : : break;
1891 : : }
1892 : : }
1893 : :
1894 : 62440 : __set_current_state(TASK_RUNNING);
1895 : :
1896 : 742112 : send_events:
1897 : : /*
1898 : : * Try to transfer events to user space. In case we get 0 events and
1899 : : * there's still timeout left over, we go trying again in search of
1900 : : * more luck.
1901 : : */
1902 [ + + + + ]: 1482796 : if (!res && eavail &&
1903 [ + + ]: 59674 : !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
1904 : 55707 : goto fetch_events;
1905 : :
1906 [ + + ]: 686405 : if (waiter) {
1907 : 61519 : write_lock_irq(&ep->lock);
1908 : 61519 : __remove_wait_queue(&ep->wq, &wait);
1909 : 61519 : write_unlock_irq(&ep->lock);
1910 : : }
1911 : :
1912 : 686405 : return res;
1913 : : }
1914 : :
1915 : : /**
1916 : : * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1917 : : * API, to verify that adding an epoll file inside another
1918 : : * epoll structure, does not violate the constraints, in
1919 : : * terms of closed loops, or too deep chains (which can
1920 : : * result in excessive stack usage).
1921 : : *
1922 : : * @priv: Pointer to the epoll file to be currently checked.
1923 : : * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1924 : : * data structure pointer.
1925 : : * @call_nests: Current dept of the @ep_call_nested() call stack.
1926 : : *
1927 : : * Returns: Returns zero if adding the epoll @file inside current epoll
1928 : : * structure @ep does not violate the constraints, or -1 otherwise.
1929 : : */
1930 : 78 : static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1931 : : {
1932 : 78 : int error = 0;
1933 : 78 : struct file *file = priv;
1934 : 78 : struct eventpoll *ep = file->private_data;
1935 : 78 : struct eventpoll *ep_tovisit;
1936 : 78 : struct rb_node *rbp;
1937 : 78 : struct epitem *epi;
1938 : :
1939 : 78 : mutex_lock_nested(&ep->mtx, call_nests + 1);
1940 : 78 : ep->visited = 1;
1941 : 78 : list_add(&ep->visited_list_link, &visited_list);
1942 [ + + ]: 234 : for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1943 : 156 : epi = rb_entry(rbp, struct epitem, rbn);
1944 [ - + ]: 156 : if (unlikely(is_file_epoll(epi->ffd.file))) {
1945 : 0 : ep_tovisit = epi->ffd.file->private_data;
1946 [ # # ]: 0 : if (ep_tovisit->visited)
1947 : 0 : continue;
1948 : 0 : error = ep_call_nested(&poll_loop_ncalls,
1949 : : ep_loop_check_proc, epi->ffd.file,
1950 : : ep_tovisit, current);
1951 [ # # ]: 0 : if (error != 0)
1952 : : break;
1953 : : } else {
1954 : : /*
1955 : : * If we've reached a file that is not associated with
1956 : : * an ep, then we need to check if the newly added
1957 : : * links are going to add too many wakeup paths. We do
1958 : : * this by adding it to the tfile_check_list, if it's
1959 : : * not already there, and calling reverse_path_check()
1960 : : * during ep_insert().
1961 : : */
1962 [ + - ]: 156 : if (list_empty(&epi->ffd.file->f_tfile_llink))
1963 : 156 : list_add(&epi->ffd.file->f_tfile_llink,
1964 : : &tfile_check_list);
1965 : : }
1966 : : }
1967 : 78 : mutex_unlock(&ep->mtx);
1968 : :
1969 : 78 : return error;
1970 : : }
1971 : :
1972 : : /**
1973 : : * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1974 : : * another epoll file (represented by @ep) does not create
1975 : : * closed loops or too deep chains.
1976 : : *
1977 : : * @ep: Pointer to the epoll private data structure.
1978 : : * @file: Pointer to the epoll file to be checked.
1979 : : *
1980 : : * Returns: Returns zero if adding the epoll @file inside current epoll
1981 : : * structure @ep does not violate the constraints, or -1 otherwise.
1982 : : */
1983 : 78 : static int ep_loop_check(struct eventpoll *ep, struct file *file)
1984 : : {
1985 : 78 : int ret;
1986 : 78 : struct eventpoll *ep_cur, *ep_next;
1987 : :
1988 : 78 : ret = ep_call_nested(&poll_loop_ncalls,
1989 : : ep_loop_check_proc, file, ep, current);
1990 : : /* clear visited list */
1991 [ + + ]: 156 : list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1992 : : visited_list_link) {
1993 : 78 : ep_cur->visited = 0;
1994 : 78 : list_del(&ep_cur->visited_list_link);
1995 : : }
1996 : 78 : return ret;
1997 : : }
1998 : :
1999 : : static void clear_tfile_check_list(void)
2000 : : {
2001 : : struct file *file;
2002 : :
2003 : : /* first clear the tfile_check_list */
2004 [ - - + + ]: 234 : while (!list_empty(&tfile_check_list)) {
2005 : 156 : file = list_first_entry(&tfile_check_list, struct file,
2006 : : f_tfile_llink);
2007 : 156 : list_del_init(&file->f_tfile_llink);
2008 : : }
2009 : 78 : INIT_LIST_HEAD(&tfile_check_list);
2010 : 78 : }
2011 : :
2012 : : /*
2013 : : * Open an eventpoll file descriptor.
2014 : : */
2015 : 1983 : static int do_epoll_create(int flags)
2016 : : {
2017 : 1983 : int error, fd;
2018 : 1983 : struct eventpoll *ep = NULL;
2019 : 1983 : struct file *file;
2020 : :
2021 : : /* Check the EPOLL_* constant for consistency. */
2022 : 1983 : BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
2023 : :
2024 [ + - ]: 1983 : if (flags & ~EPOLL_CLOEXEC)
2025 : : return -EINVAL;
2026 : : /*
2027 : : * Create the internal data structure ("struct eventpoll").
2028 : : */
2029 : 1983 : error = ep_alloc(&ep);
2030 [ + - ]: 1983 : if (error < 0)
2031 : : return error;
2032 : : /*
2033 : : * Creates all the items needed to setup an eventpoll file. That is,
2034 : : * a file structure and a free file descriptor.
2035 : : */
2036 : 1983 : fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
2037 [ - + ]: 1983 : if (fd < 0) {
2038 : 0 : error = fd;
2039 : 0 : goto out_free_ep;
2040 : : }
2041 : 1983 : file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
2042 : : O_RDWR | (flags & O_CLOEXEC));
2043 [ - + ]: 1983 : if (IS_ERR(file)) {
2044 : 0 : error = PTR_ERR(file);
2045 : 0 : goto out_free_fd;
2046 : : }
2047 : 1983 : ep->file = file;
2048 : 1983 : fd_install(fd, file);
2049 : 1983 : return fd;
2050 : :
2051 : : out_free_fd:
2052 : 0 : put_unused_fd(fd);
2053 : 0 : out_free_ep:
2054 : 0 : ep_free(ep);
2055 : 0 : return error;
2056 : : }
2057 : :
2058 : 3966 : SYSCALL_DEFINE1(epoll_create1, int, flags)
2059 : : {
2060 : 1983 : return do_epoll_create(flags);
2061 : : }
2062 : :
2063 : 0 : SYSCALL_DEFINE1(epoll_create, int, size)
2064 : : {
2065 [ # # # # ]: 0 : if (size <= 0)
2066 : : return -EINVAL;
2067 : :
2068 : 0 : return do_epoll_create(0);
2069 : : }
2070 : :
2071 : 24096 : static inline int epoll_mutex_lock(struct mutex *mutex, int depth,
2072 : : bool nonblock)
2073 : : {
2074 : 24096 : if (!nonblock) {
2075 : 24096 : mutex_lock_nested(mutex, depth);
2076 : 24096 : return 0;
2077 : : }
2078 [ # # # # : 0 : if (mutex_trylock(mutex))
# # # # ]
2079 : : return 0;
2080 : : return -EAGAIN;
2081 : : }
2082 : :
2083 : 23862 : int do_epoll_ctl(int epfd, int op, int fd, struct epoll_event *epds,
2084 : : bool nonblock)
2085 : : {
2086 : 23862 : int error;
2087 : 23862 : int full_check = 0;
2088 : 23862 : struct fd f, tf;
2089 : 23862 : struct eventpoll *ep;
2090 : 23862 : struct epitem *epi;
2091 : 23862 : struct eventpoll *tep = NULL;
2092 : :
2093 : 23862 : error = -EBADF;
2094 : 23862 : f = fdget(epfd);
2095 [ - + ]: 23862 : if (!f.file)
2096 : 0 : goto error_return;
2097 : :
2098 : : /* Get the "struct file *" for the target file */
2099 : 23862 : tf = fdget(fd);
2100 [ - + ]: 23862 : if (!tf.file)
2101 : 0 : goto error_fput;
2102 : :
2103 : : /* The target file descriptor must support poll */
2104 : 23862 : error = -EPERM;
2105 [ - + ]: 23862 : if (!file_can_poll(tf.file))
2106 : 0 : goto error_tgt_fput;
2107 : :
2108 : : /* Check if EPOLLWAKEUP is allowed */
2109 [ + + ]: 23862 : if (ep_op_has_event(op))
2110 : 16086 : ep_take_care_of_epollwakeup(epds);
2111 : :
2112 : : /*
2113 : : * We have to check that the file structure underneath the file descriptor
2114 : : * the user passed to us _is_ an eventpoll file. And also we do not permit
2115 : : * adding an epoll file descriptor inside itself.
2116 : : */
2117 : 23862 : error = -EINVAL;
2118 [ + - - + ]: 23862 : if (f.file == tf.file || !is_file_epoll(f.file))
2119 : 0 : goto error_tgt_fput;
2120 : :
2121 : : /*
2122 : : * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only,
2123 : : * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation.
2124 : : * Also, we do not currently supported nested exclusive wakeups.
2125 : : */
2126 [ + + - + ]: 23862 : if (ep_op_has_event(op) && (epds->events & EPOLLEXCLUSIVE)) {
2127 [ # # ]: 0 : if (op == EPOLL_CTL_MOD)
2128 : 0 : goto error_tgt_fput;
2129 [ # # # # ]: 0 : if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) ||
2130 [ # # ]: 0 : (epds->events & ~EPOLLEXCLUSIVE_OK_BITS)))
2131 : 0 : goto error_tgt_fput;
2132 : : }
2133 : :
2134 : : /*
2135 : : * At this point it is safe to assume that the "private_data" contains
2136 : : * our own data structure.
2137 : : */
2138 : 23862 : ep = f.file->private_data;
2139 : :
2140 : : /*
2141 : : * When we insert an epoll file descriptor, inside another epoll file
2142 : : * descriptor, there is the change of creating closed loops, which are
2143 : : * better be handled here, than in more critical paths. While we are
2144 : : * checking for loops we also determine the list of files reachable
2145 : : * and hang them on the tfile_check_list, so we can check that we
2146 : : * haven't created too many possible wakeup paths.
2147 : : *
2148 : : * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
2149 : : * the epoll file descriptor is attaching directly to a wakeup source,
2150 : : * unless the epoll file descriptor is nested. The purpose of taking the
2151 : : * 'epmutex' on add is to prevent complex toplogies such as loops and
2152 : : * deep wakeup paths from forming in parallel through multiple
2153 : : * EPOLL_CTL_ADD operations.
2154 : : */
2155 [ + - ]: 23862 : error = epoll_mutex_lock(&ep->mtx, 0, nonblock);
2156 : 23862 : if (error)
2157 : 0 : goto error_tgt_fput;
2158 [ + + ]: 23862 : if (op == EPOLL_CTL_ADD) {
2159 [ + - ]: 15252 : if (!list_empty(&f.file->f_ep_links) ||
2160 [ + + ]: 15252 : is_file_epoll(tf.file)) {
2161 : 78 : mutex_unlock(&ep->mtx);
2162 [ + - ]: 78 : error = epoll_mutex_lock(&epmutex, 0, nonblock);
2163 : 78 : if (error)
2164 : 0 : goto error_tgt_fput;
2165 : 78 : full_check = 1;
2166 [ + - ]: 78 : if (is_file_epoll(tf.file)) {
2167 : 78 : error = -ELOOP;
2168 [ - + ]: 78 : if (ep_loop_check(ep, tf.file) != 0) {
2169 : : clear_tfile_check_list();
2170 : 0 : goto error_tgt_fput;
2171 : : }
2172 : : } else
2173 : 0 : list_add(&tf.file->f_tfile_llink,
2174 : : &tfile_check_list);
2175 [ + - ]: 78 : error = epoll_mutex_lock(&ep->mtx, 0, nonblock);
2176 : 78 : if (error) {
2177 : 0 : out_del:
2178 : 0 : list_del(&tf.file->f_tfile_llink);
2179 : 0 : goto error_tgt_fput;
2180 : : }
2181 [ + - ]: 78 : if (is_file_epoll(tf.file)) {
2182 : 78 : tep = tf.file->private_data;
2183 [ + - ]: 78 : error = epoll_mutex_lock(&tep->mtx, 1, nonblock);
2184 : 78 : if (error) {
2185 : 0 : mutex_unlock(&ep->mtx);
2186 : 0 : goto out_del;
2187 : : }
2188 : : }
2189 : : }
2190 : : }
2191 : :
2192 : : /*
2193 : : * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
2194 : : * above, we can be sure to be able to use the item looked up by
2195 : : * ep_find() till we release the mutex.
2196 : : */
2197 : 23862 : epi = ep_find(ep, tf.file, fd);
2198 : :
2199 : 23862 : error = -EINVAL;
2200 [ + + + - ]: 23862 : switch (op) {
2201 : 15252 : case EPOLL_CTL_ADD:
2202 [ + - ]: 15252 : if (!epi) {
2203 : 15252 : epds->events |= EPOLLERR | EPOLLHUP;
2204 : 15252 : error = ep_insert(ep, epds, tf.file, fd, full_check);
2205 : : } else
2206 : : error = -EEXIST;
2207 [ + + ]: 15252 : if (full_check)
2208 : : clear_tfile_check_list();
2209 : : break;
2210 : 7776 : case EPOLL_CTL_DEL:
2211 [ + - ]: 7776 : if (epi)
2212 : 7776 : error = ep_remove(ep, epi);
2213 : : else
2214 : : error = -ENOENT;
2215 : : break;
2216 : 834 : case EPOLL_CTL_MOD:
2217 [ + - ]: 834 : if (epi) {
2218 [ + - ]: 834 : if (!(epi->event.events & EPOLLEXCLUSIVE)) {
2219 : 834 : epds->events |= EPOLLERR | EPOLLHUP;
2220 : 834 : error = ep_modify(ep, epi, epds);
2221 : : }
2222 : : } else
2223 : : error = -ENOENT;
2224 : : break;
2225 : : }
2226 [ + + ]: 23862 : if (tep != NULL)
2227 : 78 : mutex_unlock(&tep->mtx);
2228 : 23862 : mutex_unlock(&ep->mtx);
2229 : :
2230 : : error_tgt_fput:
2231 [ + + ]: 23862 : if (full_check)
2232 : 78 : mutex_unlock(&epmutex);
2233 : :
2234 [ + + ]: 23862 : fdput(tf);
2235 : 23862 : error_fput:
2236 [ + + ]: 23862 : fdput(f);
2237 : 23862 : error_return:
2238 : :
2239 : 23862 : return error;
2240 : : }
2241 : :
2242 : : /*
2243 : : * The following function implements the controller interface for
2244 : : * the eventpoll file that enables the insertion/removal/change of
2245 : : * file descriptors inside the interest set.
2246 : : */
2247 : 47724 : SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
2248 : : struct epoll_event __user *, event)
2249 : : {
2250 : 23862 : struct epoll_event epds;
2251 : :
2252 [ + + + - ]: 39948 : if (ep_op_has_event(op) &&
2253 : : copy_from_user(&epds, event, sizeof(struct epoll_event)))
2254 : : return -EFAULT;
2255 : :
2256 : 23862 : return do_epoll_ctl(epfd, op, fd, &epds, false);
2257 : : }
2258 : :
2259 : : /*
2260 : : * Implement the event wait interface for the eventpoll file. It is the kernel
2261 : : * part of the user space epoll_wait(2).
2262 : : */
2263 : 687492 : static int do_epoll_wait(int epfd, struct epoll_event __user *events,
2264 : : int maxevents, int timeout)
2265 : : {
2266 : 687492 : int error;
2267 : 687492 : struct fd f;
2268 : 687492 : struct eventpoll *ep;
2269 : :
2270 : : /* The maximum number of event must be greater than zero */
2271 [ + - ]: 687492 : if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
2272 : : return -EINVAL;
2273 : :
2274 : : /* Verify that the area passed by the user is writeable */
2275 [ - + + - ]: 1374984 : if (!access_ok(events, maxevents * sizeof(struct epoll_event)))
2276 : : return -EFAULT;
2277 : :
2278 : : /* Get the "struct file *" for the eventpoll file */
2279 : 687492 : f = fdget(epfd);
2280 [ + - ]: 687492 : if (!f.file)
2281 : : return -EBADF;
2282 : :
2283 : : /*
2284 : : * We have to check that the file structure underneath the fd
2285 : : * the user passed to us _is_ an eventpoll file.
2286 : : */
2287 : 687492 : error = -EINVAL;
2288 [ - + ]: 687492 : if (!is_file_epoll(f.file))
2289 : 0 : goto error_fput;
2290 : :
2291 : : /*
2292 : : * At this point it is safe to assume that the "private_data" contains
2293 : : * our own data structure.
2294 : : */
2295 : 687492 : ep = f.file->private_data;
2296 : :
2297 : : /* Time to fish for events ... */
2298 : 687492 : error = ep_poll(ep, events, maxevents, timeout);
2299 : :
2300 : 686405 : error_fput:
2301 [ + + ]: 686405 : fdput(f);
2302 : : return error;
2303 : : }
2304 : :
2305 : 1373897 : SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
2306 : : int, maxevents, int, timeout)
2307 : : {
2308 : 687492 : return do_epoll_wait(epfd, events, maxevents, timeout);
2309 : : }
2310 : :
2311 : : /*
2312 : : * Implement the event wait interface for the eventpoll file. It is the kernel
2313 : : * part of the user space epoll_pwait(2).
2314 : : */
2315 : 0 : SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2316 : : int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2317 : : size_t, sigsetsize)
2318 : : {
2319 : 0 : int error;
2320 : :
2321 : : /*
2322 : : * If the caller wants a certain signal mask to be set during the wait,
2323 : : * we apply it here.
2324 : : */
2325 : 0 : error = set_user_sigmask(sigmask, sigsetsize);
2326 [ # # ]: 0 : if (error)
2327 : 0 : return error;
2328 : :
2329 : 0 : error = do_epoll_wait(epfd, events, maxevents, timeout);
2330 : 0 : restore_saved_sigmask_unless(error == -EINTR);
2331 : :
2332 : 0 : return error;
2333 : : }
2334 : :
2335 : : #ifdef CONFIG_COMPAT
2336 : 0 : COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2337 : : struct epoll_event __user *, events,
2338 : : int, maxevents, int, timeout,
2339 : : const compat_sigset_t __user *, sigmask,
2340 : : compat_size_t, sigsetsize)
2341 : : {
2342 : 0 : long err;
2343 : :
2344 : : /*
2345 : : * If the caller wants a certain signal mask to be set during the wait,
2346 : : * we apply it here.
2347 : : */
2348 : 0 : err = set_compat_user_sigmask(sigmask, sigsetsize);
2349 [ # # ]: 0 : if (err)
2350 : : return err;
2351 : :
2352 : 0 : err = do_epoll_wait(epfd, events, maxevents, timeout);
2353 : 0 : restore_saved_sigmask_unless(err == -EINTR);
2354 : :
2355 : 0 : return err;
2356 : : }
2357 : : #endif
2358 : :
2359 : 78 : static int __init eventpoll_init(void)
2360 : : {
2361 : 78 : struct sysinfo si;
2362 : :
2363 : 78 : si_meminfo(&si);
2364 : : /*
2365 : : * Allows top 4% of lomem to be allocated for epoll watches (per user).
2366 : : */
2367 : 78 : max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2368 : : EP_ITEM_COST;
2369 : 78 : BUG_ON(max_user_watches < 0);
2370 : :
2371 : : /*
2372 : : * Initialize the structure used to perform epoll file descriptor
2373 : : * inclusion loops checks.
2374 : : */
2375 : 78 : ep_nested_calls_init(&poll_loop_ncalls);
2376 : :
2377 : : /*
2378 : : * We can have many thousands of epitems, so prevent this from
2379 : : * using an extra cache line on 64-bit (and smaller) CPUs
2380 : : */
2381 : 78 : BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2382 : :
2383 : : /* Allocates slab cache used to allocate "struct epitem" items */
2384 : 78 : epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
2385 : : 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
2386 : :
2387 : : /* Allocates slab cache used to allocate "struct eppoll_entry" */
2388 : 78 : pwq_cache = kmem_cache_create("eventpoll_pwq",
2389 : : sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL);
2390 : :
2391 : 78 : return 0;
2392 : : }
2393 : : fs_initcall(eventpoll_init);
|
