eventpoll.c 59.5 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
D
Davide Libenzi 已提交
2 3
 *  fs/eventpoll.c (Efficient event retrieval implementation)
 *  Copyright (C) 2001,...,2009	 Davide Libenzi
L
Linus Torvalds 已提交
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  Davide Libenzi <davidel@xmailserver.org>
 *
 */

#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/signal.h>
#include <linux/errno.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/poll.h>
#include <linux/string.h>
#include <linux/list.h>
#include <linux/hash.h>
#include <linux/spinlock.h>
#include <linux/syscalls.h>
#include <linux/rbtree.h>
#include <linux/wait.h>
#include <linux/eventpoll.h>
#include <linux/mount.h>
#include <linux/bitops.h>
A
Arjan van de Ven 已提交
34
#include <linux/mutex.h>
D
Davide Libenzi 已提交
35
#include <linux/anon_inodes.h>
36
#include <linux/device.h>
L
Linus Torvalds 已提交
37 38 39
#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/mman.h>
A
Arun Sharma 已提交
40
#include <linux/atomic.h>
41 42
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
43
#include <linux/compat.h>
44
#include <linux/rculist.h>
L
Linus Torvalds 已提交
45 46 47 48 49

/*
 * LOCKING:
 * There are three level of locking required by epoll :
 *
A
Arjan van de Ven 已提交
50
 * 1) epmutex (mutex)
51 52
 * 2) ep->mtx (mutex)
 * 3) ep->lock (spinlock)
L
Linus Torvalds 已提交
53 54 55 56 57 58 59 60 61
 *
 * The acquire order is the one listed above, from 1 to 3.
 * We need a spinlock (ep->lock) because we manipulate objects
 * from inside the poll callback, that might be triggered from
 * a wake_up() that in turn might be called from IRQ context.
 * So we can't sleep inside the poll callback and hence we need
 * a spinlock. During the event transfer loop (from kernel to
 * user space) we could end up sleeping due a copy_to_user(), so
 * we need a lock that will allow us to sleep. This lock is a
62 63 64 65 66
 * mutex (ep->mtx). It is acquired during the event transfer loop,
 * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
 * Then we also need a global mutex to serialize eventpoll_release_file()
 * and ep_free().
 * This mutex is acquired by ep_free() during the epoll file
L
Linus Torvalds 已提交
67 68
 * cleanup path and it is also acquired by eventpoll_release_file()
 * if a file has been pushed inside an epoll set and it is then
D
Daniel Baluta 已提交
69
 * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
70 71 72 73 74 75 76
 * It is also acquired when inserting an epoll fd onto another epoll
 * fd. We do this so that we walk the epoll tree and ensure that this
 * insertion does not create a cycle of epoll file descriptors, which
 * could lead to deadlock. We need a global mutex to prevent two
 * simultaneous inserts (A into B and B into A) from racing and
 * constructing a cycle without either insert observing that it is
 * going to.
77 78 79 80 81 82 83 84 85
 * It is necessary to acquire multiple "ep->mtx"es at once in the
 * case when one epoll fd is added to another. In this case, we
 * always acquire the locks in the order of nesting (i.e. after
 * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
 * before e2->mtx). Since we disallow cycles of epoll file
 * descriptors, this ensures that the mutexes are well-ordered. In
 * order to communicate this nesting to lockdep, when walking a tree
 * of epoll file descriptors, we use the current recursion depth as
 * the lockdep subkey.
86 87 88
 * It is possible to drop the "ep->mtx" and to use the global
 * mutex "epmutex" (together with "ep->lock") to have it working,
 * but having "ep->mtx" will make the interface more scalable.
A
Arjan van de Ven 已提交
89
 * Events that require holding "epmutex" are very rare, while for
90 91
 * normal operations the epoll private "ep->mtx" will guarantee
 * a better scalability.
L
Linus Torvalds 已提交
92 93 94
 */

/* Epoll private bits inside the event mask */
J
Jason Baron 已提交
95
#define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE)
L
Linus Torvalds 已提交
96

D
Davide Libenzi 已提交
97 98
/* Maximum number of nesting allowed inside epoll sets */
#define EP_MAX_NESTS 4
L
Linus Torvalds 已提交
99

D
Davide Libenzi 已提交
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))

L
Linus Torvalds 已提交
106 107 108
struct epoll_filefd {
	struct file *file;
	int fd;
E
Eric Wong 已提交
109
} __packed;
L
Linus Torvalds 已提交
110 111

/*
D
Davide Libenzi 已提交
112 113
 * Structure used to track possible nested calls, for too deep recursions
 * and loop cycles.
L
Linus Torvalds 已提交
114
 */
D
Davide Libenzi 已提交
115
struct nested_call_node {
L
Linus Torvalds 已提交
116
	struct list_head llink;
D
Davide Libenzi 已提交
117
	void *cookie;
D
Davide Libenzi 已提交
118
	void *ctx;
L
Linus Torvalds 已提交
119 120 121
};

/*
D
Davide Libenzi 已提交
122 123
 * This structure is used as collector for nested calls, to check for
 * maximum recursion dept and loop cycles.
L
Linus Torvalds 已提交
124
 */
D
Davide Libenzi 已提交
125 126
struct nested_calls {
	struct list_head tasks_call_list;
L
Linus Torvalds 已提交
127 128 129
	spinlock_t lock;
};

130 131 132
/*
 * Each file descriptor added to the eventpoll interface will
 * have an entry of this type linked to the "rbr" RB tree.
E
Eric Wong 已提交
133 134
 * Avoid increasing the size of this struct, there can be many thousands
 * of these on a server and we do not want this to take another cache line.
135 136
 */
struct epitem {
137 138 139 140 141 142
	union {
		/* RB tree node links this structure to the eventpoll RB tree */
		struct rb_node rbn;
		/* Used to free the struct epitem */
		struct rcu_head rcu;
	};
143 144 145 146

	/* List header used to link this structure to the eventpoll ready list */
	struct list_head rdllink;

147 148 149 150 151 152
	/*
	 * Works together "struct eventpoll"->ovflist in keeping the
	 * single linked chain of items.
	 */
	struct epitem *next;

153 154 155 156 157 158 159 160 161 162 163 164 165 166 167
	/* The file descriptor information this item refers to */
	struct epoll_filefd ffd;

	/* Number of active wait queue attached to poll operations */
	int nwait;

	/* List containing poll wait queues */
	struct list_head pwqlist;

	/* The "container" of this item */
	struct eventpoll *ep;

	/* List header used to link this item to the "struct file" items list */
	struct list_head fllink;

168
	/* wakeup_source used when EPOLLWAKEUP is set */
169
	struct wakeup_source __rcu *ws;
170

171 172
	/* The structure that describe the interested events and the source fd */
	struct epoll_event event;
173 174
};

L
Linus Torvalds 已提交
175 176
/*
 * This structure is stored inside the "private_data" member of the file
D
Daniel Baluta 已提交
177
 * structure and represents the main data structure for the eventpoll
L
Linus Torvalds 已提交
178 179 180
 * interface.
 */
struct eventpoll {
D
Daniel Baluta 已提交
181
	/* Protect the access to this structure */
182
	spinlock_t lock;
L
Linus Torvalds 已提交
183 184

	/*
185 186 187 188
	 * This mutex is used to ensure that files are not removed
	 * while epoll is using them. This is held during the event
	 * collection loop, the file cleanup path, the epoll file exit
	 * code and the ctl operations.
L
Linus Torvalds 已提交
189
	 */
190
	struct mutex mtx;
L
Linus Torvalds 已提交
191 192 193 194 195 196 197 198 199 200

	/* Wait queue used by sys_epoll_wait() */
	wait_queue_head_t wq;

	/* Wait queue used by file->poll() */
	wait_queue_head_t poll_wait;

	/* List of ready file descriptors */
	struct list_head rdllist;

D
Davide Libenzi 已提交
201
	/* RB tree root used to store monitored fd structs */
L
Linus Torvalds 已提交
202
	struct rb_root rbr;
203 204 205

	/*
	 * This is a single linked list that chains all the "struct epitem" that
L
Lucas De Marchi 已提交
206
	 * happened while transferring ready events to userspace w/out
207 208 209
	 * holding ->lock.
	 */
	struct epitem *ovflist;
210

211 212 213
	/* wakeup_source used when ep_scan_ready_list is running */
	struct wakeup_source *ws;

214 215
	/* The user that created the eventpoll descriptor */
	struct user_struct *user;
J
Jason Baron 已提交
216 217 218 219 220 221

	struct file *file;

	/* used to optimize loop detection check */
	int visited;
	struct list_head visited_list_link;
L
Linus Torvalds 已提交
222 223 224 225 226 227 228 229
};

/* Wait structure used by the poll hooks */
struct eppoll_entry {
	/* List header used to link this structure to the "struct epitem" */
	struct list_head llink;

	/* The "base" pointer is set to the container "struct epitem" */
230
	struct epitem *base;
L
Linus Torvalds 已提交
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247

	/*
	 * Wait queue item that will be linked to the target file wait
	 * queue head.
	 */
	wait_queue_t wait;

	/* The wait queue head that linked the "wait" wait queue item */
	wait_queue_head_t *whead;
};

/* Wrapper struct used by poll queueing */
struct ep_pqueue {
	poll_table pt;
	struct epitem *epi;
};

D
Davide Libenzi 已提交
248 249 250 251 252 253
/* Used by the ep_send_events() function as callback private data */
struct ep_send_events_data {
	int maxevents;
	struct epoll_event __user *events;
};

254 255 256 257
/*
 * Configuration options available inside /proc/sys/fs/epoll/
 */
/* Maximum number of epoll watched descriptors, per user */
258
static long max_user_watches __read_mostly;
259

L
Linus Torvalds 已提交
260
/*
261
 * This mutex is used to serialize ep_free() and eventpoll_release_file().
L
Linus Torvalds 已提交
262
 */
263
static DEFINE_MUTEX(epmutex);
L
Linus Torvalds 已提交
264

265 266 267
/* Used to check for epoll file descriptor inclusion loops */
static struct nested_calls poll_loop_ncalls;

D
Davide Libenzi 已提交
268 269 270 271 272
/* Used for safe wake up implementation */
static struct nested_calls poll_safewake_ncalls;

/* Used to call file's f_op->poll() under the nested calls boundaries */
static struct nested_calls poll_readywalk_ncalls;
L
Linus Torvalds 已提交
273 274

/* Slab cache used to allocate "struct epitem" */
275
static struct kmem_cache *epi_cache __read_mostly;
L
Linus Torvalds 已提交
276 277

/* Slab cache used to allocate "struct eppoll_entry" */
278
static struct kmem_cache *pwq_cache __read_mostly;
L
Linus Torvalds 已提交
279

J
Jason Baron 已提交
280 281 282 283 284 285 286 287 288
/* Visited nodes during ep_loop_check(), so we can unset them when we finish */
static LIST_HEAD(visited_list);

/*
 * List of files with newly added links, where we may need to limit the number
 * of emanating paths. Protected by the epmutex.
 */
static LIST_HEAD(tfile_check_list);

289 290 291 292
#ifdef CONFIG_SYSCTL

#include <linux/sysctl.h>

293 294
static long zero;
static long long_max = LONG_MAX;
295

296
struct ctl_table epoll_table[] = {
297 298 299
	{
		.procname	= "max_user_watches",
		.data		= &max_user_watches,
300
		.maxlen		= sizeof(max_user_watches),
301
		.mode		= 0644,
302
		.proc_handler	= proc_doulongvec_minmax,
303
		.extra1		= &zero,
304
		.extra2		= &long_max,
305
	},
306
	{ }
307 308 309
};
#endif /* CONFIG_SYSCTL */

J
Jason Baron 已提交
310 311 312 313 314 315
static const struct file_operations eventpoll_fops;

static inline int is_file_epoll(struct file *f)
{
	return f->f_op == &eventpoll_fops;
}
316

D
Davide Libenzi 已提交
317
/* Setup the structure that is used as key for the RB tree */
318 319 320 321 322 323 324
static inline void ep_set_ffd(struct epoll_filefd *ffd,
			      struct file *file, int fd)
{
	ffd->file = file;
	ffd->fd = fd;
}

D
Davide Libenzi 已提交
325
/* Compare RB tree keys */
326 327 328 329 330 331 332 333 334 335 336 337 338
static inline int ep_cmp_ffd(struct epoll_filefd *p1,
			     struct epoll_filefd *p2)
{
	return (p1->file > p2->file ? +1:
	        (p1->file < p2->file ? -1 : p1->fd - p2->fd));
}

/* Tells us if the item is currently linked */
static inline int ep_is_linked(struct list_head *p)
{
	return !list_empty(p);
}

339 340 341 342 343
static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
{
	return container_of(p, struct eppoll_entry, wait);
}

344
/* Get the "struct epitem" from a wait queue pointer */
D
Davide Libenzi 已提交
345
static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
346 347 348 349 350
{
	return container_of(p, struct eppoll_entry, wait)->base;
}

/* Get the "struct epitem" from an epoll queue wrapper */
D
Davide Libenzi 已提交
351
static inline struct epitem *ep_item_from_epqueue(poll_table *p)
352 353 354 355 356
{
	return container_of(p, struct ep_pqueue, pt)->epi;
}

/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
357
static inline int ep_op_has_event(int op)
358
{
359
	return op != EPOLL_CTL_DEL;
360 361
}

L
Linus Torvalds 已提交
362
/* Initialize the poll safe wake up structure */
D
Davide Libenzi 已提交
363
static void ep_nested_calls_init(struct nested_calls *ncalls)
L
Linus Torvalds 已提交
364
{
D
Davide Libenzi 已提交
365 366
	INIT_LIST_HEAD(&ncalls->tasks_call_list);
	spin_lock_init(&ncalls->lock);
L
Linus Torvalds 已提交
367 368
}

369 370 371 372 373 374 375 376 377 378 379 380 381
/**
 * ep_events_available - Checks if ready events might be available.
 *
 * @ep: Pointer to the eventpoll context.
 *
 * Returns: Returns a value different than zero if ready events are available,
 *          or zero otherwise.
 */
static inline int ep_events_available(struct eventpoll *ep)
{
	return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
}

D
Davide Libenzi 已提交
382 383 384 385 386 387 388 389 390 391 392
/**
 * ep_call_nested - Perform a bound (possibly) nested call, by checking
 *                  that the recursion limit is not exceeded, and that
 *                  the same nested call (by the meaning of same cookie) is
 *                  no re-entered.
 *
 * @ncalls: Pointer to the nested_calls structure to be used for this call.
 * @max_nests: Maximum number of allowed nesting calls.
 * @nproc: Nested call core function pointer.
 * @priv: Opaque data to be passed to the @nproc callback.
 * @cookie: Cookie to be used to identify this nested call.
D
Davide Libenzi 已提交
393
 * @ctx: This instance context.
D
Davide Libenzi 已提交
394 395 396
 *
 * Returns: Returns the code returned by the @nproc callback, or -1 if
 *          the maximum recursion limit has been exceeded.
L
Linus Torvalds 已提交
397
 */
D
Davide Libenzi 已提交
398 399
static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
			  int (*nproc)(void *, void *, int), void *priv,
D
Davide Libenzi 已提交
400
			  void *cookie, void *ctx)
L
Linus Torvalds 已提交
401
{
D
Davide Libenzi 已提交
402
	int error, call_nests = 0;
L
Linus Torvalds 已提交
403
	unsigned long flags;
D
Davide Libenzi 已提交
404 405 406
	struct list_head *lsthead = &ncalls->tasks_call_list;
	struct nested_call_node *tncur;
	struct nested_call_node tnode;
L
Linus Torvalds 已提交
407

D
Davide Libenzi 已提交
408
	spin_lock_irqsave(&ncalls->lock, flags);
L
Linus Torvalds 已提交
409

D
Davide Libenzi 已提交
410 411 412 413 414
	/*
	 * Try to see if the current task is already inside this wakeup call.
	 * We use a list here, since the population inside this set is always
	 * very much limited.
	 */
415
	list_for_each_entry(tncur, lsthead, llink) {
D
Davide Libenzi 已提交
416
		if (tncur->ctx == ctx &&
D
Davide Libenzi 已提交
417
		    (tncur->cookie == cookie || ++call_nests > max_nests)) {
L
Linus Torvalds 已提交
418 419 420 421
			/*
			 * Ops ... loop detected or maximum nest level reached.
			 * We abort this wake by breaking the cycle itself.
			 */
422 423
			error = -1;
			goto out_unlock;
L
Linus Torvalds 已提交
424 425 426
		}
	}

D
Davide Libenzi 已提交
427
	/* Add the current task and cookie to the list */
D
Davide Libenzi 已提交
428
	tnode.ctx = ctx;
D
Davide Libenzi 已提交
429
	tnode.cookie = cookie;
L
Linus Torvalds 已提交
430 431
	list_add(&tnode.llink, lsthead);

D
Davide Libenzi 已提交
432
	spin_unlock_irqrestore(&ncalls->lock, flags);
L
Linus Torvalds 已提交
433

D
Davide Libenzi 已提交
434 435
	/* Call the nested function */
	error = (*nproc)(priv, cookie, call_nests);
L
Linus Torvalds 已提交
436 437

	/* Remove the current task from the list */
D
Davide Libenzi 已提交
438
	spin_lock_irqsave(&ncalls->lock, flags);
L
Linus Torvalds 已提交
439
	list_del(&tnode.llink);
D
Davide Libenzi 已提交
440
out_unlock:
D
Davide Libenzi 已提交
441 442 443 444 445
	spin_unlock_irqrestore(&ncalls->lock, flags);

	return error;
}

S
Steven Rostedt 已提交
446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470
/*
 * As described in commit 0ccf831cb lockdep: annotate epoll
 * the use of wait queues used by epoll is done in a very controlled
 * manner. Wake ups can nest inside each other, but are never done
 * with the same locking. For example:
 *
 *   dfd = socket(...);
 *   efd1 = epoll_create();
 *   efd2 = epoll_create();
 *   epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
 *   epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
 *
 * When a packet arrives to the device underneath "dfd", the net code will
 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
 * callback wakeup entry on that queue, and the wake_up() performed by the
 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
 * (efd1) notices that it may have some event ready, so it needs to wake up
 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
 * that ends up in another wake_up(), after having checked about the
 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
 * avoid stack blasting.
 *
 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
 * this special case of epoll.
 */
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488
#ifdef CONFIG_DEBUG_LOCK_ALLOC
static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
				     unsigned long events, int subclass)
{
	unsigned long flags;

	spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
	wake_up_locked_poll(wqueue, events);
	spin_unlock_irqrestore(&wqueue->lock, flags);
}
#else
static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
				     unsigned long events, int subclass)
{
	wake_up_poll(wqueue, events);
}
#endif

D
Davide Libenzi 已提交
489 490
static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
{
491 492
	ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
			  1 + call_nests);
D
Davide Libenzi 已提交
493 494 495 496 497 498 499 500 501 502 503 504 505 506 507
	return 0;
}

/*
 * Perform a safe wake up of the poll wait list. The problem is that
 * with the new callback'd wake up system, it is possible that the
 * poll callback is reentered from inside the call to wake_up() done
 * on the poll wait queue head. The rule is that we cannot reenter the
 * wake up code from the same task more than EP_MAX_NESTS times,
 * and we cannot reenter the same wait queue head at all. This will
 * enable to have a hierarchy of epoll file descriptor of no more than
 * EP_MAX_NESTS deep.
 */
static void ep_poll_safewake(wait_queue_head_t *wq)
{
D
Davide Libenzi 已提交
508 509
	int this_cpu = get_cpu();

D
Davide Libenzi 已提交
510
	ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
D
Davide Libenzi 已提交
511 512 513
		       ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);

	put_cpu();
L
Linus Torvalds 已提交
514 515
}

516 517 518 519 520 521 522 523 524 525 526 527
static void ep_remove_wait_queue(struct eppoll_entry *pwq)
{
	wait_queue_head_t *whead;

	rcu_read_lock();
	/* If it is cleared by POLLFREE, it should be rcu-safe */
	whead = rcu_dereference(pwq->whead);
	if (whead)
		remove_wait_queue(whead, &pwq->wait);
	rcu_read_unlock();
}

L
Linus Torvalds 已提交
528
/*
T
Tony Battersby 已提交
529 530 531
 * This function unregisters poll callbacks from the associated file
 * descriptor.  Must be called with "mtx" held (or "epmutex" if called from
 * ep_free).
L
Linus Torvalds 已提交
532
 */
533
static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
L
Linus Torvalds 已提交
534
{
535 536
	struct list_head *lsthead = &epi->pwqlist;
	struct eppoll_entry *pwq;
L
Linus Torvalds 已提交
537

T
Tony Battersby 已提交
538 539
	while (!list_empty(lsthead)) {
		pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
L
Linus Torvalds 已提交
540

T
Tony Battersby 已提交
541
		list_del(&pwq->llink);
542
		ep_remove_wait_queue(pwq);
T
Tony Battersby 已提交
543
		kmem_cache_free(pwq_cache, pwq);
L
Linus Torvalds 已提交
544 545 546
	}
}

547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578
/* call only when ep->mtx is held */
static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
{
	return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
}

/* call only when ep->mtx is held */
static inline void ep_pm_stay_awake(struct epitem *epi)
{
	struct wakeup_source *ws = ep_wakeup_source(epi);

	if (ws)
		__pm_stay_awake(ws);
}

static inline bool ep_has_wakeup_source(struct epitem *epi)
{
	return rcu_access_pointer(epi->ws) ? true : false;
}

/* call when ep->mtx cannot be held (ep_poll_callback) */
static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
{
	struct wakeup_source *ws;

	rcu_read_lock();
	ws = rcu_dereference(epi->ws);
	if (ws)
		__pm_stay_awake(ws);
	rcu_read_unlock();
}

D
Davide Libenzi 已提交
579 580 581 582 583 584 585 586
/**
 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
 *                      the scan code, to call f_op->poll(). Also allows for
 *                      O(NumReady) performance.
 *
 * @ep: Pointer to the epoll private data structure.
 * @sproc: Pointer to the scan callback.
 * @priv: Private opaque data passed to the @sproc callback.
587
 * @depth: The current depth of recursive f_op->poll calls.
588
 * @ep_locked: caller already holds ep->mtx
D
Davide Libenzi 已提交
589 590 591 592 593 594
 *
 * Returns: The same integer error code returned by the @sproc callback.
 */
static int ep_scan_ready_list(struct eventpoll *ep,
			      int (*sproc)(struct eventpoll *,
					   struct list_head *, void *),
595
			      void *priv, int depth, bool ep_locked)
D
Davide Libenzi 已提交
596 597 598 599
{
	int error, pwake = 0;
	unsigned long flags;
	struct epitem *epi, *nepi;
600
	LIST_HEAD(txlist);
D
Davide Libenzi 已提交
601 602 603

	/*
	 * We need to lock this because we could be hit by
T
Tony Battersby 已提交
604
	 * eventpoll_release_file() and epoll_ctl().
D
Davide Libenzi 已提交
605
	 */
606 607 608

	if (!ep_locked)
		mutex_lock_nested(&ep->mtx, depth);
D
Davide Libenzi 已提交
609 610 611 612 613 614 615 616 617 618

	/*
	 * Steal the ready list, and re-init the original one to the
	 * empty list. Also, set ep->ovflist to NULL so that events
	 * happening while looping w/out locks, are not lost. We cannot
	 * have the poll callback to queue directly on ep->rdllist,
	 * because we want the "sproc" callback to be able to do it
	 * in a lockless way.
	 */
	spin_lock_irqsave(&ep->lock, flags);
619
	list_splice_init(&ep->rdllist, &txlist);
D
Davide Libenzi 已提交
620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641
	ep->ovflist = NULL;
	spin_unlock_irqrestore(&ep->lock, flags);

	/*
	 * Now call the callback function.
	 */
	error = (*sproc)(ep, &txlist, priv);

	spin_lock_irqsave(&ep->lock, flags);
	/*
	 * During the time we spent inside the "sproc" callback, some
	 * other events might have been queued by the poll callback.
	 * We re-insert them inside the main ready-list here.
	 */
	for (nepi = ep->ovflist; (epi = nepi) != NULL;
	     nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
		/*
		 * We need to check if the item is already in the list.
		 * During the "sproc" callback execution time, items are
		 * queued into ->ovflist but the "txlist" might already
		 * contain them, and the list_splice() below takes care of them.
		 */
642
		if (!ep_is_linked(&epi->rdllink)) {
D
Davide Libenzi 已提交
643
			list_add_tail(&epi->rdllink, &ep->rdllist);
644
			ep_pm_stay_awake(epi);
645
		}
D
Davide Libenzi 已提交
646 647 648 649 650 651 652 653 654 655 656 657
	}
	/*
	 * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
	 * releasing the lock, events will be queued in the normal way inside
	 * ep->rdllist.
	 */
	ep->ovflist = EP_UNACTIVE_PTR;

	/*
	 * Quickly re-inject items left on "txlist".
	 */
	list_splice(&txlist, &ep->rdllist);
658
	__pm_relax(ep->ws);
D
Davide Libenzi 已提交
659 660 661

	if (!list_empty(&ep->rdllist)) {
		/*
662 663
		 * Wake up (if active) both the eventpoll wait list and
		 * the ->poll() wait list (delayed after we release the lock).
D
Davide Libenzi 已提交
664 665 666 667 668 669 670 671
		 */
		if (waitqueue_active(&ep->wq))
			wake_up_locked(&ep->wq);
		if (waitqueue_active(&ep->poll_wait))
			pwake++;
	}
	spin_unlock_irqrestore(&ep->lock, flags);

672 673
	if (!ep_locked)
		mutex_unlock(&ep->mtx);
D
Davide Libenzi 已提交
674 675 676 677 678 679 680 681

	/* We have to call this outside the lock */
	if (pwake)
		ep_poll_safewake(&ep->poll_wait);

	return error;
}

682 683 684 685 686 687
static void epi_rcu_free(struct rcu_head *head)
{
	struct epitem *epi = container_of(head, struct epitem, rcu);
	kmem_cache_free(epi_cache, epi);
}

688 689
/*
 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
690
 * all the associated resources. Must be called with "mtx" held.
691 692 693 694 695
 */
static int ep_remove(struct eventpoll *ep, struct epitem *epi)
{
	unsigned long flags;
	struct file *file = epi->ffd.file;
L
Linus Torvalds 已提交
696 697

	/*
698 699 700 701 702 703
	 * Removes poll wait queue hooks. We _have_ to do this without holding
	 * the "ep->lock" otherwise a deadlock might occur. This because of the
	 * sequence of the lock acquisition. Here we do "ep->lock" then the wait
	 * queue head lock when unregistering the wait queue. The wakeup callback
	 * will run by holding the wait queue head lock and will call our callback
	 * that will try to get "ep->lock".
L
Linus Torvalds 已提交
704
	 */
705
	ep_unregister_pollwait(ep, epi);
L
Linus Torvalds 已提交
706

707
	/* Remove the current item from the list of epoll hooks */
J
Jonathan Corbet 已提交
708
	spin_lock(&file->f_lock);
709
	list_del_rcu(&epi->fllink);
J
Jonathan Corbet 已提交
710
	spin_unlock(&file->f_lock);
L
Linus Torvalds 已提交
711

D
Davide Libenzi 已提交
712
	rb_erase(&epi->rbn, &ep->rbr);
L
Linus Torvalds 已提交
713

714 715 716 717
	spin_lock_irqsave(&ep->lock, flags);
	if (ep_is_linked(&epi->rdllink))
		list_del_init(&epi->rdllink);
	spin_unlock_irqrestore(&ep->lock, flags);
L
Linus Torvalds 已提交
718

719
	wakeup_source_unregister(ep_wakeup_source(epi));
720 721 722 723 724 725 726 727
	/*
	 * At this point it is safe to free the eventpoll item. Use the union
	 * field epi->rcu, since we are trying to minimize the size of
	 * 'struct epitem'. The 'rbn' field is no longer in use. Protected by
	 * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make
	 * use of the rbn field.
	 */
	call_rcu(&epi->rcu, epi_rcu_free);
L
Linus Torvalds 已提交
728

729
	atomic_long_dec(&ep->user->epoll_watches);
730

731
	return 0;
L
Linus Torvalds 已提交
732 733
}

734
static void ep_free(struct eventpoll *ep)
L
Linus Torvalds 已提交
735
{
736 737
	struct rb_node *rbp;
	struct epitem *epi;
L
Linus Torvalds 已提交
738

739 740
	/* We need to release all tasks waiting for these file */
	if (waitqueue_active(&ep->poll_wait))
D
Davide Libenzi 已提交
741
		ep_poll_safewake(&ep->poll_wait);
L
Linus Torvalds 已提交
742

743 744 745
	/*
	 * We need to lock this because we could be hit by
	 * eventpoll_release_file() while we're freeing the "struct eventpoll".
746
	 * We do not need to hold "ep->mtx" here because the epoll file
747 748
	 * is on the way to be removed and no one has references to it
	 * anymore. The only hit might come from eventpoll_release_file() but
L
Lucas De Marchi 已提交
749
	 * holding "epmutex" is sufficient here.
750 751
	 */
	mutex_lock(&epmutex);
L
Linus Torvalds 已提交
752 753

	/*
754
	 * Walks through the whole tree by unregistering poll callbacks.
L
Linus Torvalds 已提交
755
	 */
756 757 758 759
	for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
		epi = rb_entry(rbp, struct epitem, rbn);

		ep_unregister_pollwait(ep, epi);
760
		cond_resched();
761
	}
L
Linus Torvalds 已提交
762 763

	/*
764 765
	 * Walks through the whole tree by freeing each "struct epitem". At this
	 * point we are sure no poll callbacks will be lingering around, and also by
766
	 * holding "epmutex" we can be sure that no file cleanup code will hit
767
	 * us during this operation. So we can avoid the lock on "ep->lock".
768 769
	 * We do not need to lock ep->mtx, either, we only do it to prevent
	 * a lockdep warning.
L
Linus Torvalds 已提交
770
	 */
771
	mutex_lock(&ep->mtx);
772
	while ((rbp = rb_first(&ep->rbr)) != NULL) {
773 774
		epi = rb_entry(rbp, struct epitem, rbn);
		ep_remove(ep, epi);
775
		cond_resched();
776
	}
777
	mutex_unlock(&ep->mtx);
L
Linus Torvalds 已提交
778

779
	mutex_unlock(&epmutex);
780
	mutex_destroy(&ep->mtx);
781
	free_uid(ep->user);
782
	wakeup_source_unregister(ep->ws);
783
	kfree(ep);
784
}
L
Linus Torvalds 已提交
785

786 787 788
static int ep_eventpoll_release(struct inode *inode, struct file *file)
{
	struct eventpoll *ep = file->private_data;
L
Linus Torvalds 已提交
789

790
	if (ep)
791 792 793
		ep_free(ep);

	return 0;
L
Linus Torvalds 已提交
794 795
}

796 797 798 799 800 801 802
static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt)
{
	pt->_key = epi->event.events;

	return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events;
}

803 804
static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
			       void *priv)
D
Davide Libenzi 已提交
805 806
{
	struct epitem *epi, *tmp;
807
	poll_table pt;
D
Davide Libenzi 已提交
808

809
	init_poll_funcptr(&pt, NULL);
810

D
Davide Libenzi 已提交
811
	list_for_each_entry_safe(epi, tmp, head, rdllink) {
812
		if (ep_item_poll(epi, &pt))
D
Davide Libenzi 已提交
813
			return POLLIN | POLLRDNORM;
814
		else {
D
Davide Libenzi 已提交
815 816 817 818 819
			/*
			 * Item has been dropped into the ready list by the poll
			 * callback, but it's not actually ready, as far as
			 * caller requested events goes. We can remove it here.
			 */
820
			__pm_relax(ep_wakeup_source(epi));
D
Davide Libenzi 已提交
821
			list_del_init(&epi->rdllink);
822
		}
D
Davide Libenzi 已提交
823 824 825 826 827
	}

	return 0;
}

828 829 830 831 832 833 834 835
static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
				 poll_table *pt);

struct readyevents_arg {
	struct eventpoll *ep;
	bool locked;
};

D
Davide Libenzi 已提交
836 837
static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
{
838 839 840 841
	struct readyevents_arg *arg = priv;

	return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL,
				  call_nests + 1, arg->locked);
D
Davide Libenzi 已提交
842 843
}

844 845
static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
{
D
Davide Libenzi 已提交
846
	int pollflags;
847
	struct eventpoll *ep = file->private_data;
848 849 850 851 852 853 854 855
	struct readyevents_arg arg;

	/*
	 * During ep_insert() we already hold the ep->mtx for the tfile.
	 * Prevent re-aquisition.
	 */
	arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc);
	arg.ep = ep;
L
Linus Torvalds 已提交
856

857 858 859
	/* Insert inside our poll wait queue */
	poll_wait(file, &ep->poll_wait, wait);

D
Davide Libenzi 已提交
860 861 862 863 864 865 866
	/*
	 * Proceed to find out if wanted events are really available inside
	 * the ready list. This need to be done under ep_call_nested()
	 * supervision, since the call to f_op->poll() done on listed files
	 * could re-enter here.
	 */
	pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
867
				   ep_poll_readyevents_proc, &arg, ep, current);
868

869
	return pollflags != -1 ? pollflags : 0;
870 871
}

872
#ifdef CONFIG_PROC_FS
873
static void ep_show_fdinfo(struct seq_file *m, struct file *f)
874 875 876 877 878 879 880 881
{
	struct eventpoll *ep = f->private_data;
	struct rb_node *rbp;

	mutex_lock(&ep->mtx);
	for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
		struct epitem *epi = rb_entry(rbp, struct epitem, rbn);

882 883 884 885
		seq_printf(m, "tfd: %8d events: %8x data: %16llx\n",
			   epi->ffd.fd, epi->event.events,
			   (long long)epi->event.data);
		if (seq_has_overflowed(m))
886 887 888 889 890 891
			break;
	}
	mutex_unlock(&ep->mtx);
}
#endif

892 893
/* File callbacks that implement the eventpoll file behaviour */
static const struct file_operations eventpoll_fops = {
894 895 896
#ifdef CONFIG_PROC_FS
	.show_fdinfo	= ep_show_fdinfo,
#endif
897
	.release	= ep_eventpoll_release,
898 899
	.poll		= ep_eventpoll_poll,
	.llseek		= noop_llseek,
900 901
};

D
Davide Libenzi 已提交
902
/*
903 904 905
 * This is called from eventpoll_release() to unlink files from the eventpoll
 * interface. We need to have this facility to cleanup correctly files that are
 * closed without being removed from the eventpoll interface.
D
Davide Libenzi 已提交
906
 */
907
void eventpoll_release_file(struct file *file)
D
Davide Libenzi 已提交
908
{
909
	struct eventpoll *ep;
910
	struct epitem *epi, *next;
D
Davide Libenzi 已提交
911 912

	/*
J
Jonathan Corbet 已提交
913
	 * We don't want to get "file->f_lock" because it is not
914
	 * necessary. It is not necessary because we're in the "struct file"
L
Lucas De Marchi 已提交
915
	 * cleanup path, and this means that no one is using this file anymore.
D
Davide Libenzi 已提交
916
	 * So, for example, epoll_ctl() cannot hit here since if we reach this
D
Davide Libenzi 已提交
917
	 * point, the file counter already went to zero and fget() would fail.
918
	 * The only hit might come from ep_free() but by holding the mutex
919
	 * will correctly serialize the operation. We do need to acquire
920
	 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
921
	 * from anywhere but ep_free().
J
Jonathan Corbet 已提交
922 923
	 *
	 * Besides, ep_remove() acquires the lock, so we can't hold it here.
D
Davide Libenzi 已提交
924
	 */
925
	mutex_lock(&epmutex);
926
	list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) {
927
		ep = epi->ep;
928
		mutex_lock_nested(&ep->mtx, 0);
929
		ep_remove(ep, epi);
930
		mutex_unlock(&ep->mtx);
D
Davide Libenzi 已提交
931
	}
932
	mutex_unlock(&epmutex);
D
Davide Libenzi 已提交
933 934
}

935
static int ep_alloc(struct eventpoll **pep)
L
Linus Torvalds 已提交
936
{
937 938 939
	int error;
	struct user_struct *user;
	struct eventpoll *ep;
L
Linus Torvalds 已提交
940

941 942 943 944 945
	user = get_current_user();
	error = -ENOMEM;
	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
	if (unlikely(!ep))
		goto free_uid;
L
Linus Torvalds 已提交
946

947
	spin_lock_init(&ep->lock);
948
	mutex_init(&ep->mtx);
L
Linus Torvalds 已提交
949 950 951 952
	init_waitqueue_head(&ep->wq);
	init_waitqueue_head(&ep->poll_wait);
	INIT_LIST_HEAD(&ep->rdllist);
	ep->rbr = RB_ROOT;
953
	ep->ovflist = EP_UNACTIVE_PTR;
954
	ep->user = user;
L
Linus Torvalds 已提交
955

956
	*pep = ep;
L
Linus Torvalds 已提交
957 958

	return 0;
959 960 961 962

free_uid:
	free_uid(user);
	return error;
L
Linus Torvalds 已提交
963 964 965
}

/*
966 967 968
 * Search the file inside the eventpoll tree. The RB tree operations
 * are protected by the "mtx" mutex, and ep_find() must be called with
 * "mtx" held.
L
Linus Torvalds 已提交
969 970 971 972 973 974 975 976
 */
static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
{
	int kcmp;
	struct rb_node *rbp;
	struct epitem *epi, *epir = NULL;
	struct epoll_filefd ffd;

977
	ep_set_ffd(&ffd, file, fd);
L
Linus Torvalds 已提交
978 979
	for (rbp = ep->rbr.rb_node; rbp; ) {
		epi = rb_entry(rbp, struct epitem, rbn);
980
		kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
L
Linus Torvalds 已提交
981 982 983 984 985 986 987 988 989 990 991 992 993 994
		if (kcmp > 0)
			rbp = rbp->rb_right;
		else if (kcmp < 0)
			rbp = rbp->rb_left;
		else {
			epir = epi;
			break;
		}
	}

	return epir;
}

/*
995
 * This is the callback that is passed to the wait queue wakeup
D
Daniel Baluta 已提交
996
 * mechanism. It is called by the stored file descriptors when they
997
 * have events to report.
L
Linus Torvalds 已提交
998
 */
999
static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
L
Linus Torvalds 已提交
1000
{
1001 1002 1003 1004
	int pwake = 0;
	unsigned long flags;
	struct epitem *epi = ep_item_from_wait(wait);
	struct eventpoll *ep = epi->ep;
J
Jason Baron 已提交
1005
	int ewake = 0;
L
Linus Torvalds 已提交
1006

1007 1008 1009 1010 1011 1012 1013 1014
	if ((unsigned long)key & POLLFREE) {
		ep_pwq_from_wait(wait)->whead = NULL;
		/*
		 * whead = NULL above can race with ep_remove_wait_queue()
		 * which can do another remove_wait_queue() after us, so we
		 * can't use __remove_wait_queue(). whead->lock is held by
		 * the caller.
		 */
1015
		list_del_init(&wait->task_list);
1016
	}
1017

1018
	spin_lock_irqsave(&ep->lock, flags);
L
Linus Torvalds 已提交
1019

1020 1021 1022 1023 1024 1025 1026
	/*
	 * If the event mask does not contain any poll(2) event, we consider the
	 * descriptor to be disabled. This condition is likely the effect of the
	 * EPOLLONESHOT bit that disables the descriptor when an event is received,
	 * until the next EPOLL_CTL_MOD will be issued.
	 */
	if (!(epi->event.events & ~EP_PRIVATE_BITS))
1027 1028
		goto out_unlock;

1029 1030 1031 1032 1033 1034 1035 1036 1037
	/*
	 * Check the events coming with the callback. At this stage, not
	 * every device reports the events in the "key" parameter of the
	 * callback. We need to be able to handle both cases here, hence the
	 * test for "key" != NULL before the event match test.
	 */
	if (key && !((unsigned long) key & epi->event.events))
		goto out_unlock;

1038
	/*
D
Daniel Baluta 已提交
1039
	 * If we are transferring events to userspace, we can hold no locks
1040
	 * (because we're accessing user memory, and because of linux f_op->poll()
D
Daniel Baluta 已提交
1041
	 * semantics). All the events that happen during that period of time are
1042 1043 1044 1045 1046 1047
	 * chained in ep->ovflist and requeued later on.
	 */
	if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
		if (epi->next == EP_UNACTIVE_PTR) {
			epi->next = ep->ovflist;
			ep->ovflist = epi;
1048 1049 1050 1051 1052 1053 1054 1055
			if (epi->ws) {
				/*
				 * Activate ep->ws since epi->ws may get
				 * deactivated at any time.
				 */
				__pm_stay_awake(ep->ws);
			}

1056 1057 1058
		}
		goto out_unlock;
	}
L
Linus Torvalds 已提交
1059

1060
	/* If this file is already in the ready list we exit soon */
1061
	if (!ep_is_linked(&epi->rdllink)) {
D
Davide Libenzi 已提交
1062
		list_add_tail(&epi->rdllink, &ep->rdllist);
1063
		ep_pm_stay_awake_rcu(epi);
1064
	}
1065 1066 1067 1068 1069

	/*
	 * Wake up ( if active ) both the eventpoll wait list and the ->poll()
	 * wait list.
	 */
J
Jason Baron 已提交
1070 1071
	if (waitqueue_active(&ep->wq)) {
		ewake = 1;
M
Matthew Wilcox 已提交
1072
		wake_up_locked(&ep->wq);
J
Jason Baron 已提交
1073
	}
1074 1075 1076
	if (waitqueue_active(&ep->poll_wait))
		pwake++;

1077
out_unlock:
1078
	spin_unlock_irqrestore(&ep->lock, flags);
L
Linus Torvalds 已提交
1079

1080 1081
	/* We have to call this outside the lock */
	if (pwake)
D
Davide Libenzi 已提交
1082
		ep_poll_safewake(&ep->poll_wait);
1083

J
Jason Baron 已提交
1084 1085 1086
	if (epi->event.events & EPOLLEXCLUSIVE)
		return ewake;

1087 1088
	return 1;
}
L
Linus Torvalds 已提交
1089 1090 1091 1092 1093 1094 1095 1096

/*
 * This is the callback that is used to add our wait queue to the
 * target file wakeup lists.
 */
static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
				 poll_table *pt)
{
1097
	struct epitem *epi = ep_item_from_epqueue(pt);
L
Linus Torvalds 已提交
1098 1099
	struct eppoll_entry *pwq;

1100
	if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
L
Linus Torvalds 已提交
1101 1102 1103
		init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
		pwq->whead = whead;
		pwq->base = epi;
J
Jason Baron 已提交
1104 1105 1106 1107
		if (epi->event.events & EPOLLEXCLUSIVE)
			add_wait_queue_exclusive(whead, &pwq->wait);
		else
			add_wait_queue(whead, &pwq->wait);
L
Linus Torvalds 已提交
1108 1109
		list_add_tail(&pwq->llink, &epi->pwqlist);
		epi->nwait++;
1110
	} else {
L
Linus Torvalds 已提交
1111 1112
		/* We have to signal that an error occurred */
		epi->nwait = -1;
1113
	}
L
Linus Torvalds 已提交
1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124
}

static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
{
	int kcmp;
	struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
	struct epitem *epic;

	while (*p) {
		parent = *p;
		epic = rb_entry(parent, struct epitem, rbn);
1125
		kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
L
Linus Torvalds 已提交
1126 1127 1128 1129 1130 1131 1132 1133 1134
		if (kcmp > 0)
			p = &parent->rb_right;
		else
			p = &parent->rb_left;
	}
	rb_link_node(&epi->rbn, parent, p);
	rb_insert_color(&epi->rbn, &ep->rbr);
}

1135 1136


J
Jason Baron 已提交
1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153
#define PATH_ARR_SIZE 5
/*
 * These are the number paths of length 1 to 5, that we are allowing to emanate
 * from a single file of interest. For example, we allow 1000 paths of length
 * 1, to emanate from each file of interest. This essentially represents the
 * potential wakeup paths, which need to be limited in order to avoid massive
 * uncontrolled wakeup storms. The common use case should be a single ep which
 * is connected to n file sources. In this case each file source has 1 path
 * of length 1. Thus, the numbers below should be more than sufficient. These
 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
 * and delete can't add additional paths. Protected by the epmutex.
 */
static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
static int path_count[PATH_ARR_SIZE];

static int path_count_inc(int nests)
{
J
Jason Baron 已提交
1154 1155 1156 1157
	/* Allow an arbitrary number of depth 1 paths */
	if (nests == 0)
		return 0;

J
Jason Baron 已提交
1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177
	if (++path_count[nests] > path_limits[nests])
		return -1;
	return 0;
}

static void path_count_init(void)
{
	int i;

	for (i = 0; i < PATH_ARR_SIZE; i++)
		path_count[i] = 0;
}

static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
{
	int error = 0;
	struct file *file = priv;
	struct file *child_file;
	struct epitem *epi;

1178 1179 1180
	/* CTL_DEL can remove links here, but that can't increase our count */
	rcu_read_lock();
	list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) {
J
Jason Baron 已提交
1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201
		child_file = epi->ep->file;
		if (is_file_epoll(child_file)) {
			if (list_empty(&child_file->f_ep_links)) {
				if (path_count_inc(call_nests)) {
					error = -1;
					break;
				}
			} else {
				error = ep_call_nested(&poll_loop_ncalls,
							EP_MAX_NESTS,
							reverse_path_check_proc,
							child_file, child_file,
							current);
			}
			if (error != 0)
				break;
		} else {
			printk(KERN_ERR "reverse_path_check_proc: "
				"file is not an ep!\n");
		}
	}
1202
	rcu_read_unlock();
J
Jason Baron 已提交
1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
	return error;
}

/**
 * reverse_path_check - The tfile_check_list is list of file *, which have
 *                      links that are proposed to be newly added. We need to
 *                      make sure that those added links don't add too many
 *                      paths such that we will spend all our time waking up
 *                      eventpoll objects.
 *
 * Returns: Returns zero if the proposed links don't create too many paths,
 *	    -1 otherwise.
 */
static int reverse_path_check(void)
{
	int error = 0;
	struct file *current_file;

	/* let's call this for all tfiles */
	list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
		path_count_init();
		error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
					reverse_path_check_proc, current_file,
					current_file, current);
		if (error)
			break;
	}
	return error;
}

1233 1234 1235
static int ep_create_wakeup_source(struct epitem *epi)
{
	const char *name;
1236
	struct wakeup_source *ws;
1237 1238 1239 1240 1241 1242 1243 1244

	if (!epi->ep->ws) {
		epi->ep->ws = wakeup_source_register("eventpoll");
		if (!epi->ep->ws)
			return -ENOMEM;
	}

	name = epi->ffd.file->f_path.dentry->d_name.name;
1245 1246 1247
	ws = wakeup_source_register(name);

	if (!ws)
1248
		return -ENOMEM;
1249
	rcu_assign_pointer(epi->ws, ws);
1250 1251 1252 1253

	return 0;
}

1254 1255
/* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1256
{
1257 1258
	struct wakeup_source *ws = ep_wakeup_source(epi);

1259
	RCU_INIT_POINTER(epi->ws, NULL);
1260 1261 1262 1263 1264 1265 1266 1267

	/*
	 * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
	 * used internally by wakeup_source_remove, too (called by
	 * wakeup_source_unregister), so we cannot use call_rcu
	 */
	synchronize_rcu();
	wakeup_source_unregister(ws);
1268 1269
}

1270 1271 1272
/*
 * Must be called with "mtx" held.
 */
L
Linus Torvalds 已提交
1273
static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1274
		     struct file *tfile, int fd, int full_check)
L
Linus Torvalds 已提交
1275 1276 1277
{
	int error, revents, pwake = 0;
	unsigned long flags;
1278
	long user_watches;
L
Linus Torvalds 已提交
1279 1280 1281
	struct epitem *epi;
	struct ep_pqueue epq;

1282 1283
	user_watches = atomic_long_read(&ep->user->epoll_watches);
	if (unlikely(user_watches >= max_user_watches))
1284
		return -ENOSPC;
1285
	if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
1286
		return -ENOMEM;
L
Linus Torvalds 已提交
1287 1288 1289 1290 1291 1292

	/* Item initialization follow here ... */
	INIT_LIST_HEAD(&epi->rdllink);
	INIT_LIST_HEAD(&epi->fllink);
	INIT_LIST_HEAD(&epi->pwqlist);
	epi->ep = ep;
1293
	ep_set_ffd(&epi->ffd, tfile, fd);
L
Linus Torvalds 已提交
1294 1295
	epi->event = *event;
	epi->nwait = 0;
1296
	epi->next = EP_UNACTIVE_PTR;
1297 1298 1299 1300 1301
	if (epi->event.events & EPOLLWAKEUP) {
		error = ep_create_wakeup_source(epi);
		if (error)
			goto error_create_wakeup_source;
	} else {
1302
		RCU_INIT_POINTER(epi->ws, NULL);
1303
	}
L
Linus Torvalds 已提交
1304 1305 1306 1307 1308 1309 1310 1311

	/* Initialize the poll table using the queue callback */
	epq.epi = epi;
	init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);

	/*
	 * Attach the item to the poll hooks and get current event bits.
	 * We can safely use the file* here because its usage count has
1312 1313 1314
	 * been increased by the caller of this function. Note that after
	 * this operation completes, the poll callback can start hitting
	 * the new item.
L
Linus Torvalds 已提交
1315
	 */
1316
	revents = ep_item_poll(epi, &epq.pt);
L
Linus Torvalds 已提交
1317 1318 1319 1320 1321 1322

	/*
	 * We have to check if something went wrong during the poll wait queue
	 * install process. Namely an allocation for a wait queue failed due
	 * high memory pressure.
	 */
1323
	error = -ENOMEM;
L
Linus Torvalds 已提交
1324
	if (epi->nwait < 0)
1325
		goto error_unregister;
L
Linus Torvalds 已提交
1326 1327

	/* Add the current item to the list of active epoll hook for this file */
J
Jonathan Corbet 已提交
1328
	spin_lock(&tfile->f_lock);
1329
	list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links);
J
Jonathan Corbet 已提交
1330
	spin_unlock(&tfile->f_lock);
L
Linus Torvalds 已提交
1331

1332 1333 1334 1335
	/*
	 * Add the current item to the RB tree. All RB tree operations are
	 * protected by "mtx", and ep_insert() is called with "mtx" held.
	 */
L
Linus Torvalds 已提交
1336 1337
	ep_rbtree_insert(ep, epi);

J
Jason Baron 已提交
1338 1339
	/* now check if we've created too many backpaths */
	error = -EINVAL;
1340
	if (full_check && reverse_path_check())
J
Jason Baron 已提交
1341 1342
		goto error_remove_epi;

1343 1344 1345
	/* We have to drop the new item inside our item list to keep track of it */
	spin_lock_irqsave(&ep->lock, flags);

L
Linus Torvalds 已提交
1346
	/* If the file is already "ready" we drop it inside the ready list */
1347
	if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
L
Linus Torvalds 已提交
1348
		list_add_tail(&epi->rdllink, &ep->rdllist);
1349
		ep_pm_stay_awake(epi);
L
Linus Torvalds 已提交
1350 1351 1352

		/* Notify waiting tasks that events are available */
		if (waitqueue_active(&ep->wq))
M
Matthew Wilcox 已提交
1353
			wake_up_locked(&ep->wq);
L
Linus Torvalds 已提交
1354 1355 1356 1357
		if (waitqueue_active(&ep->poll_wait))
			pwake++;
	}

1358
	spin_unlock_irqrestore(&ep->lock, flags);
L
Linus Torvalds 已提交
1359

1360
	atomic_long_inc(&ep->user->epoll_watches);
1361

L
Linus Torvalds 已提交
1362 1363
	/* We have to call this outside the lock */
	if (pwake)
D
Davide Libenzi 已提交
1364
		ep_poll_safewake(&ep->poll_wait);
L
Linus Torvalds 已提交
1365 1366 1367

	return 0;

J
Jason Baron 已提交
1368 1369
error_remove_epi:
	spin_lock(&tfile->f_lock);
1370
	list_del_rcu(&epi->fllink);
J
Jason Baron 已提交
1371 1372 1373 1374
	spin_unlock(&tfile->f_lock);

	rb_erase(&epi->rbn, &ep->rbr);

1375
error_unregister:
L
Linus Torvalds 已提交
1376 1377 1378 1379
	ep_unregister_pollwait(ep, epi);

	/*
	 * We need to do this because an event could have been arrived on some
D
Davide Libenzi 已提交
1380 1381 1382
	 * allocated wait queue. Note that we don't care about the ep->ovflist
	 * list, since that is used/cleaned only inside a section bound by "mtx".
	 * And ep_insert() is called with "mtx" held.
L
Linus Torvalds 已提交
1383
	 */
1384
	spin_lock_irqsave(&ep->lock, flags);
1385
	if (ep_is_linked(&epi->rdllink))
1386
		list_del_init(&epi->rdllink);
1387
	spin_unlock_irqrestore(&ep->lock, flags);
L
Linus Torvalds 已提交
1388

1389
	wakeup_source_unregister(ep_wakeup_source(epi));
1390 1391

error_create_wakeup_source:
1392
	kmem_cache_free(epi_cache, epi);
1393

L
Linus Torvalds 已提交
1394 1395 1396 1397 1398
	return error;
}

/*
 * Modify the interest event mask by dropping an event if the new mask
1399
 * has a match in the current file status. Must be called with "mtx" held.
L
Linus Torvalds 已提交
1400 1401 1402 1403 1404
 */
static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
{
	int pwake = 0;
	unsigned int revents;
1405 1406 1407
	poll_table pt;

	init_poll_funcptr(&pt, NULL);
L
Linus Torvalds 已提交
1408 1409

	/*
T
Tony Battersby 已提交
1410 1411 1412
	 * Set the new event interest mask before calling f_op->poll();
	 * otherwise we might miss an event that happens between the
	 * f_op->poll() call and the new event set registering.
L
Linus Torvalds 已提交
1413
	 */
1414
	epi->event.events = event->events; /* need barrier below */
T
Tony Battersby 已提交
1415
	epi->event.data = event->data; /* protected by mtx */
1416
	if (epi->event.events & EPOLLWAKEUP) {
1417
		if (!ep_has_wakeup_source(epi))
1418
			ep_create_wakeup_source(epi);
1419
	} else if (ep_has_wakeup_source(epi)) {
1420 1421
		ep_destroy_wakeup_source(epi);
	}
L
Linus Torvalds 已提交
1422

1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442
	/*
	 * The following barrier has two effects:
	 *
	 * 1) Flush epi changes above to other CPUs.  This ensures
	 *    we do not miss events from ep_poll_callback if an
	 *    event occurs immediately after we call f_op->poll().
	 *    We need this because we did not take ep->lock while
	 *    changing epi above (but ep_poll_callback does take
	 *    ep->lock).
	 *
	 * 2) We also need to ensure we do not miss _past_ events
	 *    when calling f_op->poll().  This barrier also
	 *    pairs with the barrier in wq_has_sleeper (see
	 *    comments for wq_has_sleeper).
	 *
	 * This barrier will now guarantee ep_poll_callback or f_op->poll
	 * (or both) will notice the readiness of an item.
	 */
	smp_mb();

L
Linus Torvalds 已提交
1443 1444 1445 1446
	/*
	 * Get current event bits. We can safely use the file* here because
	 * its usage count has been increased by the caller of this function.
	 */
1447
	revents = ep_item_poll(epi, &pt);
L
Linus Torvalds 已提交
1448 1449

	/*
1450
	 * If the item is "hot" and it is not registered inside the ready
D
Davide Libenzi 已提交
1451
	 * list, push it inside.
L
Linus Torvalds 已提交
1452
	 */
1453
	if (revents & event->events) {
T
Tony Battersby 已提交
1454
		spin_lock_irq(&ep->lock);
1455 1456
		if (!ep_is_linked(&epi->rdllink)) {
			list_add_tail(&epi->rdllink, &ep->rdllist);
1457
			ep_pm_stay_awake(epi);
1458 1459 1460

			/* Notify waiting tasks that events are available */
			if (waitqueue_active(&ep->wq))
M
Matthew Wilcox 已提交
1461
				wake_up_locked(&ep->wq);
1462 1463
			if (waitqueue_active(&ep->poll_wait))
				pwake++;
1464
		}
T
Tony Battersby 已提交
1465
		spin_unlock_irq(&ep->lock);
1466
	}
L
Linus Torvalds 已提交
1467

1468 1469
	/* We have to call this outside the lock */
	if (pwake)
D
Davide Libenzi 已提交
1470
		ep_poll_safewake(&ep->poll_wait);
L
Linus Torvalds 已提交
1471

1472
	return 0;
L
Linus Torvalds 已提交
1473 1474
}

1475 1476
static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
			       void *priv)
L
Linus Torvalds 已提交
1477
{
D
Davide Libenzi 已提交
1478 1479
	struct ep_send_events_data *esed = priv;
	int eventcnt;
1480
	unsigned int revents;
D
Davide Libenzi 已提交
1481 1482
	struct epitem *epi;
	struct epoll_event __user *uevent;
1483
	struct wakeup_source *ws;
1484 1485 1486
	poll_table pt;

	init_poll_funcptr(&pt, NULL);
L
Linus Torvalds 已提交
1487

1488
	/*
D
Davide Libenzi 已提交
1489 1490 1491
	 * We can loop without lock because we are passed a task private list.
	 * Items cannot vanish during the loop because ep_scan_ready_list() is
	 * holding "mtx" during this call.
1492
	 */
D
Davide Libenzi 已提交
1493 1494 1495
	for (eventcnt = 0, uevent = esed->events;
	     !list_empty(head) && eventcnt < esed->maxevents;) {
		epi = list_first_entry(head, struct epitem, rdllink);
1496

1497 1498 1499 1500 1501 1502 1503 1504 1505
		/*
		 * Activate ep->ws before deactivating epi->ws to prevent
		 * triggering auto-suspend here (in case we reactive epi->ws
		 * below).
		 *
		 * This could be rearranged to delay the deactivation of epi->ws
		 * instead, but then epi->ws would temporarily be out of sync
		 * with ep_is_linked().
		 */
1506 1507 1508 1509 1510 1511 1512
		ws = ep_wakeup_source(epi);
		if (ws) {
			if (ws->active)
				__pm_stay_awake(ep->ws);
			__pm_relax(ws);
		}

1513
		list_del_init(&epi->rdllink);
L
Linus Torvalds 已提交
1514

1515
		revents = ep_item_poll(epi, &pt);
D
Davide Libenzi 已提交
1516

1517
		/*
D
Davide Libenzi 已提交
1518 1519 1520 1521
		 * If the event mask intersect the caller-requested one,
		 * deliver the event to userspace. Again, ep_scan_ready_list()
		 * is holding "mtx", so no operations coming from userspace
		 * can change the item.
1522 1523
		 */
		if (revents) {
D
Davide Libenzi 已提交
1524
			if (__put_user(revents, &uevent->events) ||
1525 1526
			    __put_user(epi->event.data, &uevent->data)) {
				list_add(&epi->rdllink, head);
1527
				ep_pm_stay_awake(epi);
1528
				return eventcnt ? eventcnt : -EFAULT;
1529
			}
1530
			eventcnt++;
D
Davide Libenzi 已提交
1531
			uevent++;
1532 1533 1534 1535 1536 1537 1538 1539
			if (epi->event.events & EPOLLONESHOT)
				epi->event.events &= EP_PRIVATE_BITS;
			else if (!(epi->event.events & EPOLLET)) {
				/*
				 * If this file has been added with Level
				 * Trigger mode, we need to insert back inside
				 * the ready list, so that the next call to
				 * epoll_wait() will check again the events
L
Lucas De Marchi 已提交
1540
				 * availability. At this point, no one can insert
1541 1542 1543 1544 1545 1546
				 * into ep->rdllist besides us. The epoll_ctl()
				 * callers are locked out by
				 * ep_scan_ready_list() holding "mtx" and the
				 * poll callback will queue them in ep->ovflist.
				 */
				list_add_tail(&epi->rdllink, &ep->rdllist);
1547
				ep_pm_stay_awake(epi);
1548 1549 1550
			}
		}
	}
D
Davide Libenzi 已提交
1551 1552 1553

	return eventcnt;
}
1554

1555 1556
static int ep_send_events(struct eventpoll *ep,
			  struct epoll_event __user *events, int maxevents)
D
Davide Libenzi 已提交
1557 1558
{
	struct ep_send_events_data esed;
L
Linus Torvalds 已提交
1559

D
Davide Libenzi 已提交
1560 1561
	esed.maxevents = maxevents;
	esed.events = events;
1562

1563
	return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false);
L
Linus Torvalds 已提交
1564 1565
}

1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576
static inline struct timespec ep_set_mstimeout(long ms)
{
	struct timespec now, ts = {
		.tv_sec = ms / MSEC_PER_SEC,
		.tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
	};

	ktime_get_ts(&now);
	return timespec_add_safe(now, ts);
}

1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593
/**
 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
 *           event buffer.
 *
 * @ep: Pointer to the eventpoll context.
 * @events: Pointer to the userspace buffer where the ready events should be
 *          stored.
 * @maxevents: Size (in terms of number of events) of the caller event buffer.
 * @timeout: Maximum timeout for the ready events fetch operation, in
 *           milliseconds. If the @timeout is zero, the function will not block,
 *           while if the @timeout is less than zero, the function will block
 *           until at least one event has been retrieved (or an error
 *           occurred).
 *
 * Returns: Returns the number of ready events which have been fetched, or an
 *          error code, in case of error.
 */
L
Linus Torvalds 已提交
1594 1595 1596
static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
		   int maxevents, long timeout)
{
1597
	int res = 0, eavail, timed_out = 0;
L
Linus Torvalds 已提交
1598
	unsigned long flags;
1599
	long slack = 0;
L
Linus Torvalds 已提交
1600
	wait_queue_t wait;
1601 1602 1603
	ktime_t expires, *to = NULL;

	if (timeout > 0) {
1604 1605
		struct timespec end_time = ep_set_mstimeout(timeout);

1606 1607 1608 1609
		slack = select_estimate_accuracy(&end_time);
		to = &expires;
		*to = timespec_to_ktime(end_time);
	} else if (timeout == 0) {
1610 1611 1612 1613
		/*
		 * Avoid the unnecessary trip to the wait queue loop, if the
		 * caller specified a non blocking operation.
		 */
1614
		timed_out = 1;
1615 1616
		spin_lock_irqsave(&ep->lock, flags);
		goto check_events;
1617
	}
L
Linus Torvalds 已提交
1618

1619
fetch_events:
1620
	spin_lock_irqsave(&ep->lock, flags);
L
Linus Torvalds 已提交
1621

1622
	if (!ep_events_available(ep)) {
L
Linus Torvalds 已提交
1623 1624 1625 1626 1627 1628
		/*
		 * We don't have any available event to return to the caller.
		 * We need to sleep here, and we will be wake up by
		 * ep_poll_callback() when events will become available.
		 */
		init_waitqueue_entry(&wait, current);
C
Changli Gao 已提交
1629
		__add_wait_queue_exclusive(&ep->wq, &wait);
L
Linus Torvalds 已提交
1630 1631 1632 1633 1634 1635 1636 1637

		for (;;) {
			/*
			 * We don't want to sleep if the ep_poll_callback() sends us
			 * a wakeup in between. That's why we set the task state
			 * to TASK_INTERRUPTIBLE before doing the checks.
			 */
			set_current_state(TASK_INTERRUPTIBLE);
1638
			if (ep_events_available(ep) || timed_out)
L
Linus Torvalds 已提交
1639 1640 1641 1642 1643 1644
				break;
			if (signal_pending(current)) {
				res = -EINTR;
				break;
			}

1645
			spin_unlock_irqrestore(&ep->lock, flags);
1646
			if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1647 1648
				timed_out = 1;

1649
			spin_lock_irqsave(&ep->lock, flags);
L
Linus Torvalds 已提交
1650 1651
		}

1652 1653
		__remove_wait_queue(&ep->wq, &wait);
		__set_current_state(TASK_RUNNING);
L
Linus Torvalds 已提交
1654
	}
1655
check_events:
L
Linus Torvalds 已提交
1656
	/* Is it worth to try to dig for events ? */
1657
	eavail = ep_events_available(ep);
L
Linus Torvalds 已提交
1658

1659
	spin_unlock_irqrestore(&ep->lock, flags);
L
Linus Torvalds 已提交
1660 1661 1662 1663 1664 1665 1666

	/*
	 * Try to transfer events to user space. In case we get 0 events and
	 * there's still timeout left over, we go trying again in search of
	 * more luck.
	 */
	if (!res && eavail &&
1667
	    !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
1668
		goto fetch_events;
L
Linus Torvalds 已提交
1669 1670 1671 1672

	return res;
}

1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692
/**
 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
 *                      API, to verify that adding an epoll file inside another
 *                      epoll structure, does not violate the constraints, in
 *                      terms of closed loops, or too deep chains (which can
 *                      result in excessive stack usage).
 *
 * @priv: Pointer to the epoll file to be currently checked.
 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
 *          data structure pointer.
 * @call_nests: Current dept of the @ep_call_nested() call stack.
 *
 * Returns: Returns zero if adding the epoll @file inside current epoll
 *          structure @ep does not violate the constraints, or -1 otherwise.
 */
static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
{
	int error = 0;
	struct file *file = priv;
	struct eventpoll *ep = file->private_data;
J
Jason Baron 已提交
1693
	struct eventpoll *ep_tovisit;
1694 1695 1696
	struct rb_node *rbp;
	struct epitem *epi;

1697
	mutex_lock_nested(&ep->mtx, call_nests + 1);
J
Jason Baron 已提交
1698 1699
	ep->visited = 1;
	list_add(&ep->visited_list_link, &visited_list);
1700 1701 1702
	for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
		epi = rb_entry(rbp, struct epitem, rbn);
		if (unlikely(is_file_epoll(epi->ffd.file))) {
J
Jason Baron 已提交
1703 1704 1705
			ep_tovisit = epi->ffd.file->private_data;
			if (ep_tovisit->visited)
				continue;
1706
			error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
J
Jason Baron 已提交
1707 1708
					ep_loop_check_proc, epi->ffd.file,
					ep_tovisit, current);
1709 1710
			if (error != 0)
				break;
J
Jason Baron 已提交
1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722
		} else {
			/*
			 * If we've reached a file that is not associated with
			 * an ep, then we need to check if the newly added
			 * links are going to add too many wakeup paths. We do
			 * this by adding it to the tfile_check_list, if it's
			 * not already there, and calling reverse_path_check()
			 * during ep_insert().
			 */
			if (list_empty(&epi->ffd.file->f_tfile_llink))
				list_add(&epi->ffd.file->f_tfile_llink,
					 &tfile_check_list);
1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742
		}
	}
	mutex_unlock(&ep->mtx);

	return error;
}

/**
 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
 *                 another epoll file (represented by @ep) does not create
 *                 closed loops or too deep chains.
 *
 * @ep: Pointer to the epoll private data structure.
 * @file: Pointer to the epoll file to be checked.
 *
 * Returns: Returns zero if adding the epoll @file inside current epoll
 *          structure @ep does not violate the constraints, or -1 otherwise.
 */
static int ep_loop_check(struct eventpoll *ep, struct file *file)
{
J
Jason Baron 已提交
1743 1744 1745 1746
	int ret;
	struct eventpoll *ep_cur, *ep_next;

	ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1747
			      ep_loop_check_proc, file, ep, current);
J
Jason Baron 已提交
1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767
	/* clear visited list */
	list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
							visited_list_link) {
		ep_cur->visited = 0;
		list_del(&ep_cur->visited_list_link);
	}
	return ret;
}

static void clear_tfile_check_list(void)
{
	struct file *file;

	/* first clear the tfile_check_list */
	while (!list_empty(&tfile_check_list)) {
		file = list_first_entry(&tfile_check_list, struct file,
					f_tfile_llink);
		list_del_init(&file->f_tfile_llink);
	}
	INIT_LIST_HEAD(&tfile_check_list);
1768 1769
}

1770
/*
1771
 * Open an eventpoll file descriptor.
1772
 */
1773
SYSCALL_DEFINE1(epoll_create1, int, flags)
1774
{
J
Jason Baron 已提交
1775
	int error, fd;
D
Davide Libenzi 已提交
1776
	struct eventpoll *ep = NULL;
J
Jason Baron 已提交
1777
	struct file *file;
1778

1779 1780 1781
	/* Check the EPOLL_* constant for consistency.  */
	BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);

1782 1783
	if (flags & ~EPOLL_CLOEXEC)
		return -EINVAL;
1784
	/*
D
Davide Libenzi 已提交
1785
	 * Create the internal data structure ("struct eventpoll").
1786
	 */
1787
	error = ep_alloc(&ep);
D
Davide Libenzi 已提交
1788 1789
	if (error < 0)
		return error;
1790 1791
	/*
	 * Creates all the items needed to setup an eventpoll file. That is,
A
Al Viro 已提交
1792
	 * a file structure and a free file descriptor.
1793
	 */
J
Jason Baron 已提交
1794 1795 1796 1797 1798 1799
	fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
	if (fd < 0) {
		error = fd;
		goto out_free_ep;
	}
	file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
1800
				 O_RDWR | (flags & O_CLOEXEC));
J
Jason Baron 已提交
1801 1802 1803 1804 1805
	if (IS_ERR(file)) {
		error = PTR_ERR(file);
		goto out_free_fd;
	}
	ep->file = file;
1806
	fd_install(fd, file);
J
Jason Baron 已提交
1807 1808 1809 1810 1811 1812
	return fd;

out_free_fd:
	put_unused_fd(fd);
out_free_ep:
	ep_free(ep);
D
Davide Libenzi 已提交
1813
	return error;
1814 1815
}

1816
SYSCALL_DEFINE1(epoll_create, int, size)
U
Ulrich Drepper 已提交
1817
{
1818
	if (size <= 0)
1819 1820 1821
		return -EINVAL;

	return sys_epoll_create1(0);
U
Ulrich Drepper 已提交
1822 1823
}

1824 1825 1826
/*
 * The following function implements the controller interface for
 * the eventpoll file that enables the insertion/removal/change of
D
Davide Libenzi 已提交
1827
 * file descriptors inside the interest set.
1828
 */
1829 1830
SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
		struct epoll_event __user *, event)
1831 1832
{
	int error;
1833
	int full_check = 0;
A
Al Viro 已提交
1834
	struct fd f, tf;
1835 1836 1837
	struct eventpoll *ep;
	struct epitem *epi;
	struct epoll_event epds;
1838
	struct eventpoll *tep = NULL;
1839 1840 1841 1842 1843 1844 1845

	error = -EFAULT;
	if (ep_op_has_event(op) &&
	    copy_from_user(&epds, event, sizeof(struct epoll_event)))
		goto error_return;

	error = -EBADF;
A
Al Viro 已提交
1846 1847
	f = fdget(epfd);
	if (!f.file)
1848 1849 1850
		goto error_return;

	/* Get the "struct file *" for the target file */
A
Al Viro 已提交
1851 1852
	tf = fdget(fd);
	if (!tf.file)
1853 1854 1855 1856
		goto error_fput;

	/* The target file descriptor must support poll */
	error = -EPERM;
A
Al Viro 已提交
1857
	if (!tf.file->f_op->poll)
1858 1859
		goto error_tgt_fput;

1860
	/* Check if EPOLLWAKEUP is allowed */
1861 1862
	if (ep_op_has_event(op))
		ep_take_care_of_epollwakeup(&epds);
1863

1864 1865 1866 1867 1868 1869
	/*
	 * We have to check that the file structure underneath the file descriptor
	 * the user passed to us _is_ an eventpoll file. And also we do not permit
	 * adding an epoll file descriptor inside itself.
	 */
	error = -EINVAL;
A
Al Viro 已提交
1870
	if (f.file == tf.file || !is_file_epoll(f.file))
1871 1872
		goto error_tgt_fput;

J
Jason Baron 已提交
1873 1874 1875 1876 1877 1878 1879 1880 1881
	/*
	 * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only,
	 * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation.
	 * Also, we do not currently supported nested exclusive wakeups.
	 */
	if ((epds.events & EPOLLEXCLUSIVE) && (op == EPOLL_CTL_MOD ||
		(op == EPOLL_CTL_ADD && is_file_epoll(tf.file))))
		goto error_tgt_fput;

1882 1883 1884 1885
	/*
	 * At this point it is safe to assume that the "private_data" contains
	 * our own data structure.
	 */
A
Al Viro 已提交
1886
	ep = f.file->private_data;
1887

1888 1889 1890
	/*
	 * When we insert an epoll file descriptor, inside another epoll file
	 * descriptor, there is the change of creating closed loops, which are
J
Jason Baron 已提交
1891 1892 1893 1894
	 * better be handled here, than in more critical paths. While we are
	 * checking for loops we also determine the list of files reachable
	 * and hang them on the tfile_check_list, so we can check that we
	 * haven't created too many possible wakeup paths.
1895
	 *
1896 1897 1898 1899 1900 1901
	 * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when
	 * the epoll file descriptor is attaching directly to a wakeup source,
	 * unless the epoll file descriptor is nested. The purpose of taking the
	 * 'epmutex' on add is to prevent complex toplogies such as loops and
	 * deep wakeup paths from forming in parallel through multiple
	 * EPOLL_CTL_ADD operations.
1902
	 */
1903
	mutex_lock_nested(&ep->mtx, 0);
J
Jason Baron 已提交
1904
	if (op == EPOLL_CTL_ADD) {
1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922
		if (!list_empty(&f.file->f_ep_links) ||
						is_file_epoll(tf.file)) {
			full_check = 1;
			mutex_unlock(&ep->mtx);
			mutex_lock(&epmutex);
			if (is_file_epoll(tf.file)) {
				error = -ELOOP;
				if (ep_loop_check(ep, tf.file) != 0) {
					clear_tfile_check_list();
					goto error_tgt_fput;
				}
			} else
				list_add(&tf.file->f_tfile_llink,
							&tfile_check_list);
			mutex_lock_nested(&ep->mtx, 0);
			if (is_file_epoll(tf.file)) {
				tep = tf.file->private_data;
				mutex_lock_nested(&tep->mtx, 1);
1923
			}
1924 1925
		}
	}
1926

D
Davide Libenzi 已提交
1927 1928 1929 1930 1931
	/*
	 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
	 * above, we can be sure to be able to use the item looked up by
	 * ep_find() till we release the mutex.
	 */
A
Al Viro 已提交
1932
	epi = ep_find(ep, tf.file, fd);
1933 1934 1935 1936 1937 1938

	error = -EINVAL;
	switch (op) {
	case EPOLL_CTL_ADD:
		if (!epi) {
			epds.events |= POLLERR | POLLHUP;
1939
			error = ep_insert(ep, &epds, tf.file, fd, full_check);
1940 1941
		} else
			error = -EEXIST;
1942 1943
		if (full_check)
			clear_tfile_check_list();
1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958
		break;
	case EPOLL_CTL_DEL:
		if (epi)
			error = ep_remove(ep, epi);
		else
			error = -ENOENT;
		break;
	case EPOLL_CTL_MOD:
		if (epi) {
			epds.events |= POLLERR | POLLHUP;
			error = ep_modify(ep, epi, &epds);
		} else
			error = -ENOENT;
		break;
	}
1959 1960
	if (tep != NULL)
		mutex_unlock(&tep->mtx);
1961
	mutex_unlock(&ep->mtx);
1962 1963

error_tgt_fput:
1964
	if (full_check)
1965 1966
		mutex_unlock(&epmutex);

A
Al Viro 已提交
1967
	fdput(tf);
1968
error_fput:
A
Al Viro 已提交
1969
	fdput(f);
1970 1971 1972 1973 1974 1975 1976 1977 1978
error_return:

	return error;
}

/*
 * Implement the event wait interface for the eventpoll file. It is the kernel
 * part of the user space epoll_wait(2).
 */
1979 1980
SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
		int, maxevents, int, timeout)
1981
{
1982 1983
	int error;
	struct fd f;
1984 1985 1986 1987 1988 1989 1990
	struct eventpoll *ep;

	/* The maximum number of event must be greater than zero */
	if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
		return -EINVAL;

	/* Verify that the area passed by the user is writeable */
1991 1992
	if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
		return -EFAULT;
1993 1994

	/* Get the "struct file *" for the eventpoll file */
1995 1996 1997
	f = fdget(epfd);
	if (!f.file)
		return -EBADF;
1998 1999 2000 2001 2002 2003

	/*
	 * We have to check that the file structure underneath the fd
	 * the user passed to us _is_ an eventpoll file.
	 */
	error = -EINVAL;
2004
	if (!is_file_epoll(f.file))
2005 2006 2007 2008 2009 2010
		goto error_fput;

	/*
	 * At this point it is safe to assume that the "private_data" contains
	 * our own data structure.
	 */
2011
	ep = f.file->private_data;
2012 2013 2014 2015 2016

	/* Time to fish for events ... */
	error = ep_poll(ep, events, maxevents, timeout);

error_fput:
2017
	fdput(f);
2018 2019 2020 2021 2022 2023 2024
	return error;
}

/*
 * Implement the event wait interface for the eventpoll file. It is the kernel
 * part of the user space epoll_pwait(2).
 */
2025 2026 2027
SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
		int, maxevents, int, timeout, const sigset_t __user *, sigmask,
		size_t, sigsetsize)
2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040
{
	int error;
	sigset_t ksigmask, sigsaved;

	/*
	 * If the caller wants a certain signal mask to be set during the wait,
	 * we apply it here.
	 */
	if (sigmask) {
		if (sigsetsize != sizeof(sigset_t))
			return -EINVAL;
		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
			return -EFAULT;
2041 2042
		sigsaved = current->blocked;
		set_current_blocked(&ksigmask);
2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055
	}

	error = sys_epoll_wait(epfd, events, maxevents, timeout);

	/*
	 * If we changed the signal mask, we need to restore the original one.
	 * In case we've got a signal while waiting, we do not restore the
	 * signal mask yet, and we allow do_signal() to deliver the signal on
	 * the way back to userspace, before the signal mask is restored.
	 */
	if (sigmask) {
		if (error == -EINTR) {
			memcpy(&current->saved_sigmask, &sigsaved,
2056
			       sizeof(sigsaved));
2057
			set_restore_sigmask();
2058
		} else
2059
			set_current_blocked(&sigsaved);
2060 2061 2062 2063 2064
	}

	return error;
}

2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085
#ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
			struct epoll_event __user *, events,
			int, maxevents, int, timeout,
			const compat_sigset_t __user *, sigmask,
			compat_size_t, sigsetsize)
{
	long err;
	compat_sigset_t csigmask;
	sigset_t ksigmask, sigsaved;

	/*
	 * If the caller wants a certain signal mask to be set during the wait,
	 * we apply it here.
	 */
	if (sigmask) {
		if (sigsetsize != sizeof(compat_sigset_t))
			return -EINVAL;
		if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
			return -EFAULT;
		sigset_from_compat(&ksigmask, &csigmask);
2086 2087
		sigsaved = current->blocked;
		set_current_blocked(&ksigmask);
2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103
	}

	err = sys_epoll_wait(epfd, events, maxevents, timeout);

	/*
	 * If we changed the signal mask, we need to restore the original one.
	 * In case we've got a signal while waiting, we do not restore the
	 * signal mask yet, and we allow do_signal() to deliver the signal on
	 * the way back to userspace, before the signal mask is restored.
	 */
	if (sigmask) {
		if (err == -EINTR) {
			memcpy(&current->saved_sigmask, &sigsaved,
			       sizeof(sigsaved));
			set_restore_sigmask();
		} else
2104
			set_current_blocked(&sigsaved);
2105 2106 2107 2108 2109 2110
	}

	return err;
}
#endif

L
Linus Torvalds 已提交
2111 2112
static int __init eventpoll_init(void)
{
2113 2114 2115
	struct sysinfo si;

	si_meminfo(&si);
2116 2117 2118 2119
	/*
	 * Allows top 4% of lomem to be allocated for epoll watches (per user).
	 */
	max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2120
		EP_ITEM_COST;
2121
	BUG_ON(max_user_watches < 0);
L
Linus Torvalds 已提交
2122

2123 2124 2125 2126 2127 2128
	/*
	 * Initialize the structure used to perform epoll file descriptor
	 * inclusion loops checks.
	 */
	ep_nested_calls_init(&poll_loop_ncalls);

L
Linus Torvalds 已提交
2129
	/* Initialize the structure used to perform safe poll wait head wake ups */
D
Davide Libenzi 已提交
2130 2131 2132 2133
	ep_nested_calls_init(&poll_safewake_ncalls);

	/* Initialize the structure used to perform file's f_op->poll() calls */
	ep_nested_calls_init(&poll_readywalk_ncalls);
L
Linus Torvalds 已提交
2134

E
Eric Wong 已提交
2135 2136 2137 2138 2139 2140
	/*
	 * We can have many thousands of epitems, so prevent this from
	 * using an extra cache line on 64-bit (and smaller) CPUs
	 */
	BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);

L
Linus Torvalds 已提交
2141 2142
	/* Allocates slab cache used to allocate "struct epitem" items */
	epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
D
Davide Libenzi 已提交
2143
			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
L
Linus Torvalds 已提交
2144 2145 2146

	/* Allocates slab cache used to allocate "struct eppoll_entry" */
	pwq_cache = kmem_cache_create("eventpoll_pwq",
D
Davide Libenzi 已提交
2147
			sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
L
Linus Torvalds 已提交
2148 2149 2150

	return 0;
}
D
Davide Libenzi 已提交
2151
fs_initcall(eventpoll_init);