linux/fs/eventpoll.c
<<
>>
Prefs
   1/*
   2 *  fs/eventpoll.c (Efficient event retrieval implementation)
   3 *  Copyright (C) 2001,...,2009  Davide Libenzi
   4 *
   5 *  This program is free software; you can redistribute it and/or modify
   6 *  it under the terms of the GNU General Public License as published by
   7 *  the Free Software Foundation; either version 2 of the License, or
   8 *  (at your option) any later version.
   9 *
  10 *  Davide Libenzi <davidel@xmailserver.org>
  11 *
  12 */
  13
  14#include <linux/init.h>
  15#include <linux/kernel.h>
  16#include <linux/sched.h>
  17#include <linux/fs.h>
  18#include <linux/file.h>
  19#include <linux/signal.h>
  20#include <linux/errno.h>
  21#include <linux/mm.h>
  22#include <linux/slab.h>
  23#include <linux/poll.h>
  24#include <linux/string.h>
  25#include <linux/list.h>
  26#include <linux/hash.h>
  27#include <linux/spinlock.h>
  28#include <linux/syscalls.h>
  29#include <linux/rbtree.h>
  30#include <linux/wait.h>
  31#include <linux/eventpoll.h>
  32#include <linux/mount.h>
  33#include <linux/bitops.h>
  34#include <linux/mutex.h>
  35#include <linux/anon_inodes.h>
  36#include <linux/device.h>
  37#include <asm/uaccess.h>
  38#include <asm/io.h>
  39#include <asm/mman.h>
  40#include <linux/atomic.h>
  41#include <linux/proc_fs.h>
  42#include <linux/seq_file.h>
  43#include <linux/compat.h>
  44
  45/*
  46 * LOCKING:
  47 * There are three level of locking required by epoll :
  48 *
  49 * 1) epmutex (mutex)
  50 * 2) ep->mtx (mutex)
  51 * 3) ep->lock (spinlock)
  52 *
  53 * The acquire order is the one listed above, from 1 to 3.
  54 * We need a spinlock (ep->lock) because we manipulate objects
  55 * from inside the poll callback, that might be triggered from
  56 * a wake_up() that in turn might be called from IRQ context.
  57 * So we can't sleep inside the poll callback and hence we need
  58 * a spinlock. During the event transfer loop (from kernel to
  59 * user space) we could end up sleeping due a copy_to_user(), so
  60 * we need a lock that will allow us to sleep. This lock is a
  61 * mutex (ep->mtx). It is acquired during the event transfer loop,
  62 * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file().
  63 * Then we also need a global mutex to serialize eventpoll_release_file()
  64 * and ep_free().
  65 * This mutex is acquired by ep_free() during the epoll file
  66 * cleanup path and it is also acquired by eventpoll_release_file()
  67 * if a file has been pushed inside an epoll set and it is then
  68 * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL).
  69 * It is also acquired when inserting an epoll fd onto another epoll
  70 * fd. We do this so that we walk the epoll tree and ensure that this
  71 * insertion does not create a cycle of epoll file descriptors, which
  72 * could lead to deadlock. We need a global mutex to prevent two
  73 * simultaneous inserts (A into B and B into A) from racing and
  74 * constructing a cycle without either insert observing that it is
  75 * going to.
  76 * It is necessary to acquire multiple "ep->mtx"es at once in the
  77 * case when one epoll fd is added to another. In this case, we
  78 * always acquire the locks in the order of nesting (i.e. after
  79 * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired
  80 * before e2->mtx). Since we disallow cycles of epoll file
  81 * descriptors, this ensures that the mutexes are well-ordered. In
  82 * order to communicate this nesting to lockdep, when walking a tree
  83 * of epoll file descriptors, we use the current recursion depth as
  84 * the lockdep subkey.
  85 * It is possible to drop the "ep->mtx" and to use the global
  86 * mutex "epmutex" (together with "ep->lock") to have it working,
  87 * but having "ep->mtx" will make the interface more scalable.
  88 * Events that require holding "epmutex" are very rare, while for
  89 * normal operations the epoll private "ep->mtx" will guarantee
  90 * a better scalability.
  91 */
  92
  93/* Epoll private bits inside the event mask */
  94#define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET)
  95
  96/* Maximum number of nesting allowed inside epoll sets */
  97#define EP_MAX_NESTS 4
  98
  99#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
 100
 101#define EP_UNACTIVE_PTR ((void *) -1L)
 102
 103#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
 104
 105struct epoll_filefd {
 106        struct file *file;
 107        int fd;
 108} __packed;
 109
 110/*
 111 * Structure used to track possible nested calls, for too deep recursions
 112 * and loop cycles.
 113 */
 114struct nested_call_node {
 115        struct list_head llink;
 116        void *cookie;
 117        void *ctx;
 118};
 119
 120/*
 121 * This structure is used as collector for nested calls, to check for
 122 * maximum recursion dept and loop cycles.
 123 */
 124struct nested_calls {
 125        struct list_head tasks_call_list;
 126        spinlock_t lock;
 127};
 128
 129/*
 130 * Each file descriptor added to the eventpoll interface will
 131 * have an entry of this type linked to the "rbr" RB tree.
 132 * Avoid increasing the size of this struct, there can be many thousands
 133 * of these on a server and we do not want this to take another cache line.
 134 */
 135struct epitem {
 136        /* RB tree node used to link this structure to the eventpoll RB tree */
 137        struct rb_node rbn;
 138
 139        /* List header used to link this structure to the eventpoll ready list */
 140        struct list_head rdllink;
 141
 142        /*
 143         * Works together "struct eventpoll"->ovflist in keeping the
 144         * single linked chain of items.
 145         */
 146        struct epitem *next;
 147
 148        /* The file descriptor information this item refers to */
 149        struct epoll_filefd ffd;
 150
 151        /* Number of active wait queue attached to poll operations */
 152        int nwait;
 153
 154        /* List containing poll wait queues */
 155        struct list_head pwqlist;
 156
 157        /* The "container" of this item */
 158        struct eventpoll *ep;
 159
 160        /* List header used to link this item to the "struct file" items list */
 161        struct list_head fllink;
 162
 163        /* wakeup_source used when EPOLLWAKEUP is set */
 164        struct wakeup_source __rcu *ws;
 165
 166        /* The structure that describe the interested events and the source fd */
 167        struct epoll_event event;
 168};
 169
 170/*
 171 * This structure is stored inside the "private_data" member of the file
 172 * structure and represents the main data structure for the eventpoll
 173 * interface.
 174 */
 175struct eventpoll {
 176        /* Protect the access to this structure */
 177        spinlock_t lock;
 178
 179        /*
 180         * This mutex is used to ensure that files are not removed
 181         * while epoll is using them. This is held during the event
 182         * collection loop, the file cleanup path, the epoll file exit
 183         * code and the ctl operations.
 184         */
 185        struct mutex mtx;
 186
 187        /* Wait queue used by sys_epoll_wait() */
 188        wait_queue_head_t wq;
 189
 190        /* Wait queue used by file->poll() */
 191        wait_queue_head_t poll_wait;
 192
 193        /* List of ready file descriptors */
 194        struct list_head rdllist;
 195
 196        /* RB tree root used to store monitored fd structs */
 197        struct rb_root rbr;
 198
 199        /*
 200         * This is a single linked list that chains all the "struct epitem" that
 201         * happened while transferring ready events to userspace w/out
 202         * holding ->lock.
 203         */
 204        struct epitem *ovflist;
 205
 206        /* wakeup_source used when ep_scan_ready_list is running */
 207        struct wakeup_source *ws;
 208
 209        /* The user that created the eventpoll descriptor */
 210        struct user_struct *user;
 211
 212        struct file *file;
 213
 214        /* used to optimize loop detection check */
 215        int visited;
 216        struct list_head visited_list_link;
 217};
 218
 219/* Wait structure used by the poll hooks */
 220struct eppoll_entry {
 221        /* List header used to link this structure to the "struct epitem" */
 222        struct list_head llink;
 223
 224        /* The "base" pointer is set to the container "struct epitem" */
 225        struct epitem *base;
 226
 227        /*
 228         * Wait queue item that will be linked to the target file wait
 229         * queue head.
 230         */
 231        wait_queue_t wait;
 232
 233        /* The wait queue head that linked the "wait" wait queue item */
 234        wait_queue_head_t *whead;
 235};
 236
 237/* Wrapper struct used by poll queueing */
 238struct ep_pqueue {
 239        poll_table pt;
 240        struct epitem *epi;
 241};
 242
 243/* Used by the ep_send_events() function as callback private data */
 244struct ep_send_events_data {
 245        int maxevents;
 246        struct epoll_event __user *events;
 247};
 248
 249/*
 250 * Configuration options available inside /proc/sys/fs/epoll/
 251 */
 252/* Maximum number of epoll watched descriptors, per user */
 253static long max_user_watches __read_mostly;
 254
 255/*
 256 * This mutex is used to serialize ep_free() and eventpoll_release_file().
 257 */
 258static DEFINE_MUTEX(epmutex);
 259
 260/* Used to check for epoll file descriptor inclusion loops */
 261static struct nested_calls poll_loop_ncalls;
 262
 263/* Used for safe wake up implementation */
 264static struct nested_calls poll_safewake_ncalls;
 265
 266/* Used to call file's f_op->poll() under the nested calls boundaries */
 267static struct nested_calls poll_readywalk_ncalls;
 268
 269/* Slab cache used to allocate "struct epitem" */
 270static struct kmem_cache *epi_cache __read_mostly;
 271
 272/* Slab cache used to allocate "struct eppoll_entry" */
 273static struct kmem_cache *pwq_cache __read_mostly;
 274
 275/* Visited nodes during ep_loop_check(), so we can unset them when we finish */
 276static LIST_HEAD(visited_list);
 277
 278/*
 279 * List of files with newly added links, where we may need to limit the number
 280 * of emanating paths. Protected by the epmutex.
 281 */
 282static LIST_HEAD(tfile_check_list);
 283
 284#ifdef CONFIG_SYSCTL
 285
 286#include <linux/sysctl.h>
 287
 288static long zero;
 289static long long_max = LONG_MAX;
 290
 291ctl_table epoll_table[] = {
 292        {
 293                .procname       = "max_user_watches",
 294                .data           = &max_user_watches,
 295                .maxlen         = sizeof(max_user_watches),
 296                .mode           = 0644,
 297                .proc_handler   = proc_doulongvec_minmax,
 298                .extra1         = &zero,
 299                .extra2         = &long_max,
 300        },
 301        { }
 302};
 303#endif /* CONFIG_SYSCTL */
 304
 305static const struct file_operations eventpoll_fops;
 306
 307static inline int is_file_epoll(struct file *f)
 308{
 309        return f->f_op == &eventpoll_fops;
 310}
 311
 312/* Setup the structure that is used as key for the RB tree */
 313static inline void ep_set_ffd(struct epoll_filefd *ffd,
 314                              struct file *file, int fd)
 315{
 316        ffd->file = file;
 317        ffd->fd = fd;
 318}
 319
 320/* Compare RB tree keys */
 321static inline int ep_cmp_ffd(struct epoll_filefd *p1,
 322                             struct epoll_filefd *p2)
 323{
 324        return (p1->file > p2->file ? +1:
 325                (p1->file < p2->file ? -1 : p1->fd - p2->fd));
 326}
 327
 328/* Tells us if the item is currently linked */
 329static inline int ep_is_linked(struct list_head *p)
 330{
 331        return !list_empty(p);
 332}
 333
 334static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p)
 335{
 336        return container_of(p, struct eppoll_entry, wait);
 337}
 338
 339/* Get the "struct epitem" from a wait queue pointer */
 340static inline struct epitem *ep_item_from_wait(wait_queue_t *p)
 341{
 342        return container_of(p, struct eppoll_entry, wait)->base;
 343}
 344
 345/* Get the "struct epitem" from an epoll queue wrapper */
 346static inline struct epitem *ep_item_from_epqueue(poll_table *p)
 347{
 348        return container_of(p, struct ep_pqueue, pt)->epi;
 349}
 350
 351/* Tells if the epoll_ctl(2) operation needs an event copy from userspace */
 352static inline int ep_op_has_event(int op)
 353{
 354        return op != EPOLL_CTL_DEL;
 355}
 356
 357/* Initialize the poll safe wake up structure */
 358static void ep_nested_calls_init(struct nested_calls *ncalls)
 359{
 360        INIT_LIST_HEAD(&ncalls->tasks_call_list);
 361        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 */
 372static inline int ep_events_available(struct eventpoll *ep)
 373{
 374        return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR;
 375}
 376
 377/**
 378 * ep_call_nested - Perform a bound (possibly) nested call, by checking
 379 *                  that the recursion limit is not exceeded, and that
 380 *                  the same nested call (by the meaning of same cookie) is
 381 *                  no re-entered.
 382 *
 383 * @ncalls: Pointer to the nested_calls structure to be used for this call.
 384 * @max_nests: Maximum number of allowed nesting calls.
 385 * @nproc: Nested call core function pointer.
 386 * @priv: Opaque data to be passed to the @nproc callback.
 387 * @cookie: Cookie to be used to identify this nested call.
 388 * @ctx: This instance context.
 389 *
 390 * Returns: Returns the code returned by the @nproc callback, or -1 if
 391 *          the maximum recursion limit has been exceeded.
 392 */
 393static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
 394                          int (*nproc)(void *, void *, int), void *priv,
 395                          void *cookie, void *ctx)
 396{
 397        int error, call_nests = 0;
 398        unsigned long flags;
 399        struct list_head *lsthead = &ncalls->tasks_call_list;
 400        struct nested_call_node *tncur;
 401        struct nested_call_node tnode;
 402
 403        spin_lock_irqsave(&ncalls->lock, flags);
 404
 405        /*
 406         * Try to see if the current task is already inside this wakeup call.
 407         * We use a list here, since the population inside this set is always
 408         * very much limited.
 409         */
 410        list_for_each_entry(tncur, lsthead, llink) {
 411                if (tncur->ctx == ctx &&
 412                    (tncur->cookie == cookie || ++call_nests > max_nests)) {
 413                        /*
 414                         * Ops ... loop detected or maximum nest level reached.
 415                         * We abort this wake by breaking the cycle itself.
 416                         */
 417                        error = -1;
 418                        goto out_unlock;
 419                }
 420        }
 421
 422        /* Add the current task and cookie to the list */
 423        tnode.ctx = ctx;
 424        tnode.cookie = cookie;
 425        list_add(&tnode.llink, lsthead);
 426
 427        spin_unlock_irqrestore(&ncalls->lock, flags);
 428
 429        /* Call the nested function */
 430        error = (*nproc)(priv, cookie, call_nests);
 431
 432        /* Remove the current task from the list */
 433        spin_lock_irqsave(&ncalls->lock, flags);
 434        list_del(&tnode.llink);
 435out_unlock:
 436        spin_unlock_irqrestore(&ncalls->lock, flags);
 437
 438        return error;
 439}
 440
 441/*
 442 * As described in commit 0ccf831cb lockdep: annotate epoll
 443 * the use of wait queues used by epoll is done in a very controlled
 444 * manner. Wake ups can nest inside each other, but are never done
 445 * with the same locking. For example:
 446 *
 447 *   dfd = socket(...);
 448 *   efd1 = epoll_create();
 449 *   efd2 = epoll_create();
 450 *   epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
 451 *   epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
 452 *
 453 * When a packet arrives to the device underneath "dfd", the net code will
 454 * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
 455 * callback wakeup entry on that queue, and the wake_up() performed by the
 456 * "dfd" net code will end up in ep_poll_callback(). At this point epoll
 457 * (efd1) notices that it may have some event ready, so it needs to wake up
 458 * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
 459 * that ends up in another wake_up(), after having checked about the
 460 * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
 461 * avoid stack blasting.
 462 *
 463 * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle
 464 * this special case of epoll.
 465 */
 466#ifdef CONFIG_DEBUG_LOCK_ALLOC
 467static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
 468                                     unsigned long events, int subclass)
 469{
 470        unsigned long flags;
 471
 472        spin_lock_irqsave_nested(&wqueue->lock, flags, subclass);
 473        wake_up_locked_poll(wqueue, events);
 474        spin_unlock_irqrestore(&wqueue->lock, flags);
 475}
 476#else
 477static inline void ep_wake_up_nested(wait_queue_head_t *wqueue,
 478                                     unsigned long events, int subclass)
 479{
 480        wake_up_poll(wqueue, events);
 481}
 482#endif
 483
 484static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
 485{
 486        ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN,
 487                          1 + call_nests);
 488        return 0;
 489}
 490
 491/*
 492 * Perform a safe wake up of the poll wait list. The problem is that
 493 * with the new callback'd wake up system, it is possible that the
 494 * poll callback is reentered from inside the call to wake_up() done
 495 * on the poll wait queue head. The rule is that we cannot reenter the
 496 * wake up code from the same task more than EP_MAX_NESTS times,
 497 * and we cannot reenter the same wait queue head at all. This will
 498 * enable to have a hierarchy of epoll file descriptor of no more than
 499 * EP_MAX_NESTS deep.
 500 */
 501static void ep_poll_safewake(wait_queue_head_t *wq)
 502{
 503        int this_cpu = get_cpu();
 504
 505        ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
 506                       ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
 507
 508        put_cpu();
 509}
 510
 511static void ep_remove_wait_queue(struct eppoll_entry *pwq)
 512{
 513        wait_queue_head_t *whead;
 514
 515        rcu_read_lock();
 516        /* If it is cleared by POLLFREE, it should be rcu-safe */
 517        whead = rcu_dereference(pwq->whead);
 518        if (whead)
 519                remove_wait_queue(whead, &pwq->wait);
 520        rcu_read_unlock();
 521}
 522
 523/*
 524 * This function unregisters poll callbacks from the associated file
 525 * descriptor.  Must be called with "mtx" held (or "epmutex" if called from
 526 * ep_free).
 527 */
 528static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
 529{
 530        struct list_head *lsthead = &epi->pwqlist;
 531        struct eppoll_entry *pwq;
 532
 533        while (!list_empty(lsthead)) {
 534                pwq = list_first_entry(lsthead, struct eppoll_entry, llink);
 535
 536                list_del(&pwq->llink);
 537                ep_remove_wait_queue(pwq);
 538                kmem_cache_free(pwq_cache, pwq);
 539        }
 540}
 541
 542/* call only when ep->mtx is held */
 543static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
 544{
 545        return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
 546}
 547
 548/* call only when ep->mtx is held */
 549static inline void ep_pm_stay_awake(struct epitem *epi)
 550{
 551        struct wakeup_source *ws = ep_wakeup_source(epi);
 552
 553        if (ws)
 554                __pm_stay_awake(ws);
 555}
 556
 557static inline bool ep_has_wakeup_source(struct epitem *epi)
 558{
 559        return rcu_access_pointer(epi->ws) ? true : false;
 560}
 561
 562/* call when ep->mtx cannot be held (ep_poll_callback) */
 563static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
 564{
 565        struct wakeup_source *ws;
 566
 567        rcu_read_lock();
 568        ws = rcu_dereference(epi->ws);
 569        if (ws)
 570                __pm_stay_awake(ws);
 571        rcu_read_unlock();
 572}
 573
 574/**
 575 * ep_scan_ready_list - Scans the ready list in a way that makes possible for
 576 *                      the scan code, to call f_op->poll(). Also allows for
 577 *                      O(NumReady) performance.
 578 *
 579 * @ep: Pointer to the epoll private data structure.
 580 * @sproc: Pointer to the scan callback.
 581 * @priv: Private opaque data passed to the @sproc callback.
 582 * @depth: The current depth of recursive f_op->poll calls.
 583 *
 584 * Returns: The same integer error code returned by the @sproc callback.
 585 */
 586static int ep_scan_ready_list(struct eventpoll *ep,
 587                              int (*sproc)(struct eventpoll *,
 588                                           struct list_head *, void *),
 589                              void *priv,
 590                              int depth)
 591{
 592        int error, pwake = 0;
 593        unsigned long flags;
 594        struct epitem *epi, *nepi;
 595        LIST_HEAD(txlist);
 596
 597        /*
 598         * We need to lock this because we could be hit by
 599         * eventpoll_release_file() and epoll_ctl().
 600         */
 601        mutex_lock_nested(&ep->mtx, depth);
 602
 603        /*
 604         * Steal the ready list, and re-init the original one to the
 605         * empty list. Also, set ep->ovflist to NULL so that events
 606         * happening while looping w/out locks, are not lost. We cannot
 607         * have the poll callback to queue directly on ep->rdllist,
 608         * because we want the "sproc" callback to be able to do it
 609         * in a lockless way.
 610         */
 611        spin_lock_irqsave(&ep->lock, flags);
 612        list_splice_init(&ep->rdllist, &txlist);
 613        ep->ovflist = NULL;
 614        spin_unlock_irqrestore(&ep->lock, flags);
 615
 616        /*
 617         * Now call the callback function.
 618         */
 619        error = (*sproc)(ep, &txlist, priv);
 620
 621        spin_lock_irqsave(&ep->lock, flags);
 622        /*
 623         * During the time we spent inside the "sproc" callback, some
 624         * other events might have been queued by the poll callback.
 625         * We re-insert them inside the main ready-list here.
 626         */
 627        for (nepi = ep->ovflist; (epi = nepi) != NULL;
 628             nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
 629                /*
 630                 * We need to check if the item is already in the list.
 631                 * During the "sproc" callback execution time, items are
 632                 * queued into ->ovflist but the "txlist" might already
 633                 * contain them, and the list_splice() below takes care of them.
 634                 */
 635                if (!ep_is_linked(&epi->rdllink)) {
 636                        list_add_tail(&epi->rdllink, &ep->rdllist);
 637                        ep_pm_stay_awake(epi);
 638                }
 639        }
 640        /*
 641         * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
 642         * releasing the lock, events will be queued in the normal way inside
 643         * ep->rdllist.
 644         */
 645        ep->ovflist = EP_UNACTIVE_PTR;
 646
 647        /*
 648         * Quickly re-inject items left on "txlist".
 649         */
 650        list_splice(&txlist, &ep->rdllist);
 651        __pm_relax(ep->ws);
 652
 653        if (!list_empty(&ep->rdllist)) {
 654                /*
 655                 * Wake up (if active) both the eventpoll wait list and
 656                 * the ->poll() wait list (delayed after we release the lock).
 657                 */
 658                if (waitqueue_active(&ep->wq))
 659                        wake_up_locked(&ep->wq);
 660                if (waitqueue_active(&ep->poll_wait))
 661                        pwake++;
 662        }
 663        spin_unlock_irqrestore(&ep->lock, flags);
 664
 665        mutex_unlock(&ep->mtx);
 666
 667        /* We have to call this outside the lock */
 668        if (pwake)
 669                ep_poll_safewake(&ep->poll_wait);
 670
 671        return error;
 672}
 673
 674/*
 675 * Removes a "struct epitem" from the eventpoll RB tree and deallocates
 676 * all the associated resources. Must be called with "mtx" held.
 677 */
 678static int ep_remove(struct eventpoll *ep, struct epitem *epi)
 679{
 680        unsigned long flags;
 681        struct file *file = epi->ffd.file;
 682
 683        /*
 684         * Removes poll wait queue hooks. We _have_ to do this without holding
 685         * the "ep->lock" otherwise a deadlock might occur. This because of the
 686         * sequence of the lock acquisition. Here we do "ep->lock" then the wait
 687         * queue head lock when unregistering the wait queue. The wakeup callback
 688         * will run by holding the wait queue head lock and will call our callback
 689         * that will try to get "ep->lock".
 690         */
 691        ep_unregister_pollwait(ep, epi);
 692
 693        /* Remove the current item from the list of epoll hooks */
 694        spin_lock(&file->f_lock);
 695        if (ep_is_linked(&epi->fllink))
 696                list_del_init(&epi->fllink);
 697        spin_unlock(&file->f_lock);
 698
 699        rb_erase(&epi->rbn, &ep->rbr);
 700
 701        spin_lock_irqsave(&ep->lock, flags);
 702        if (ep_is_linked(&epi->rdllink))
 703                list_del_init(&epi->rdllink);
 704        spin_unlock_irqrestore(&ep->lock, flags);
 705
 706        wakeup_source_unregister(ep_wakeup_source(epi));
 707
 708        /* At this point it is safe to free the eventpoll item */
 709        kmem_cache_free(epi_cache, epi);
 710
 711        atomic_long_dec(&ep->user->epoll_watches);
 712
 713        return 0;
 714}
 715
 716static void ep_free(struct eventpoll *ep)
 717{
 718        struct rb_node *rbp;
 719        struct epitem *epi;
 720
 721        /* We need to release all tasks waiting for these file */
 722        if (waitqueue_active(&ep->poll_wait))
 723                ep_poll_safewake(&ep->poll_wait);
 724
 725        /*
 726         * We need to lock this because we could be hit by
 727         * eventpoll_release_file() while we're freeing the "struct eventpoll".
 728         * We do not need to hold "ep->mtx" here because the epoll file
 729         * is on the way to be removed and no one has references to it
 730         * anymore. The only hit might come from eventpoll_release_file() but
 731         * holding "epmutex" is sufficient here.
 732         */
 733        mutex_lock(&epmutex);
 734
 735        /*
 736         * Walks through the whole tree by unregistering poll callbacks.
 737         */
 738        for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
 739                epi = rb_entry(rbp, struct epitem, rbn);
 740
 741                ep_unregister_pollwait(ep, epi);
 742        }
 743
 744        /*
 745         * Walks through the whole tree by freeing each "struct epitem". At this
 746         * point we are sure no poll callbacks will be lingering around, and also by
 747         * holding "epmutex" we can be sure that no file cleanup code will hit
 748         * us during this operation. So we can avoid the lock on "ep->lock".
 749         * We do not need to lock ep->mtx, either, we only do it to prevent
 750         * a lockdep warning.
 751         */
 752        mutex_lock(&ep->mtx);
 753        while ((rbp = rb_first(&ep->rbr)) != NULL) {
 754                epi = rb_entry(rbp, struct epitem, rbn);
 755                ep_remove(ep, epi);
 756        }
 757        mutex_unlock(&ep->mtx);
 758
 759        mutex_unlock(&epmutex);
 760        mutex_destroy(&ep->mtx);
 761        free_uid(ep->user);
 762        wakeup_source_unregister(ep->ws);
 763        kfree(ep);
 764}
 765
 766static int ep_eventpoll_release(struct inode *inode, struct file *file)
 767{
 768        struct eventpoll *ep = file->private_data;
 769
 770        if (ep)
 771                ep_free(ep);
 772
 773        return 0;
 774}
 775
 776static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt)
 777{
 778        pt->_key = epi->event.events;
 779
 780        return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events;
 781}
 782
 783static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head,
 784                               void *priv)
 785{
 786        struct epitem *epi, *tmp;
 787        poll_table pt;
 788
 789        init_poll_funcptr(&pt, NULL);
 790
 791        list_for_each_entry_safe(epi, tmp, head, rdllink) {
 792                if (ep_item_poll(epi, &pt))
 793                        return POLLIN | POLLRDNORM;
 794                else {
 795                        /*
 796                         * Item has been dropped into the ready list by the poll
 797                         * callback, but it's not actually ready, as far as
 798                         * caller requested events goes. We can remove it here.
 799                         */
 800                        __pm_relax(ep_wakeup_source(epi));
 801                        list_del_init(&epi->rdllink);
 802                }
 803        }
 804
 805        return 0;
 806}
 807
 808static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests)
 809{
 810        return ep_scan_ready_list(priv, ep_read_events_proc, NULL, call_nests + 1);
 811}
 812
 813static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
 814{
 815        int pollflags;
 816        struct eventpoll *ep = file->private_data;
 817
 818        /* Insert inside our poll wait queue */
 819        poll_wait(file, &ep->poll_wait, wait);
 820
 821        /*
 822         * Proceed to find out if wanted events are really available inside
 823         * the ready list. This need to be done under ep_call_nested()
 824         * supervision, since the call to f_op->poll() done on listed files
 825         * could re-enter here.
 826         */
 827        pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
 828                                   ep_poll_readyevents_proc, ep, ep, current);
 829
 830        return pollflags != -1 ? pollflags : 0;
 831}
 832
 833#ifdef CONFIG_PROC_FS
 834static int ep_show_fdinfo(struct seq_file *m, struct file *f)
 835{
 836        struct eventpoll *ep = f->private_data;
 837        struct rb_node *rbp;
 838        int ret = 0;
 839
 840        mutex_lock(&ep->mtx);
 841        for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
 842                struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
 843
 844                ret = seq_printf(m, "tfd: %8d events: %8x data: %16llx\n",
 845                                 epi->ffd.fd, epi->event.events,
 846                                 (long long)epi->event.data);
 847                if (ret)
 848                        break;
 849        }
 850        mutex_unlock(&ep->mtx);
 851
 852        return ret;
 853}
 854#endif
 855
 856/* File callbacks that implement the eventpoll file behaviour */
 857static const struct file_operations eventpoll_fops = {
 858#ifdef CONFIG_PROC_FS
 859        .show_fdinfo    = ep_show_fdinfo,
 860#endif
 861        .release        = ep_eventpoll_release,
 862        .poll           = ep_eventpoll_poll,
 863        .llseek         = noop_llseek,
 864};
 865
 866/*
 867 * This is called from eventpoll_release() to unlink files from the eventpoll
 868 * interface. We need to have this facility to cleanup correctly files that are
 869 * closed without being removed from the eventpoll interface.
 870 */
 871void eventpoll_release_file(struct file *file)
 872{
 873        struct list_head *lsthead = &file->f_ep_links;
 874        struct eventpoll *ep;
 875        struct epitem *epi;
 876
 877        /*
 878         * We don't want to get "file->f_lock" because it is not
 879         * necessary. It is not necessary because we're in the "struct file"
 880         * cleanup path, and this means that no one is using this file anymore.
 881         * So, for example, epoll_ctl() cannot hit here since if we reach this
 882         * point, the file counter already went to zero and fget() would fail.
 883         * The only hit might come from ep_free() but by holding the mutex
 884         * will correctly serialize the operation. We do need to acquire
 885         * "ep->mtx" after "epmutex" because ep_remove() requires it when called
 886         * from anywhere but ep_free().
 887         *
 888         * Besides, ep_remove() acquires the lock, so we can't hold it here.
 889         */
 890        mutex_lock(&epmutex);
 891
 892        while (!list_empty(lsthead)) {
 893                epi = list_first_entry(lsthead, struct epitem, fllink);
 894
 895                ep = epi->ep;
 896                list_del_init(&epi->fllink);
 897                mutex_lock_nested(&ep->mtx, 0);
 898                ep_remove(ep, epi);
 899                mutex_unlock(&ep->mtx);
 900        }
 901
 902        mutex_unlock(&epmutex);
 903}
 904
 905static int ep_alloc(struct eventpoll **pep)
 906{
 907        int error;
 908        struct user_struct *user;
 909        struct eventpoll *ep;
 910
 911        user = get_current_user();
 912        error = -ENOMEM;
 913        ep = kzalloc(sizeof(*ep), GFP_KERNEL);
 914        if (unlikely(!ep))
 915                goto free_uid;
 916
 917        spin_lock_init(&ep->lock);
 918        mutex_init(&ep->mtx);
 919        init_waitqueue_head(&ep->wq);
 920        init_waitqueue_head(&ep->poll_wait);
 921        INIT_LIST_HEAD(&ep->rdllist);
 922        ep->rbr = RB_ROOT;
 923        ep->ovflist = EP_UNACTIVE_PTR;
 924        ep->user = user;
 925
 926        *pep = ep;
 927
 928        return 0;
 929
 930free_uid:
 931        free_uid(user);
 932        return error;
 933}
 934
 935/*
 936 * Search the file inside the eventpoll tree. The RB tree operations
 937 * are protected by the "mtx" mutex, and ep_find() must be called with
 938 * "mtx" held.
 939 */
 940static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
 941{
 942        int kcmp;
 943        struct rb_node *rbp;
 944        struct epitem *epi, *epir = NULL;
 945        struct epoll_filefd ffd;
 946
 947        ep_set_ffd(&ffd, file, fd);
 948        for (rbp = ep->rbr.rb_node; rbp; ) {
 949                epi = rb_entry(rbp, struct epitem, rbn);
 950                kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
 951                if (kcmp > 0)
 952                        rbp = rbp->rb_right;
 953                else if (kcmp < 0)
 954                        rbp = rbp->rb_left;
 955                else {
 956                        epir = epi;
 957                        break;
 958                }
 959        }
 960
 961        return epir;
 962}
 963
 964/*
 965 * This is the callback that is passed to the wait queue wakeup
 966 * mechanism. It is called by the stored file descriptors when they
 967 * have events to report.
 968 */
 969static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key)
 970{
 971        int pwake = 0;
 972        unsigned long flags;
 973        struct epitem *epi = ep_item_from_wait(wait);
 974        struct eventpoll *ep = epi->ep;
 975
 976        if ((unsigned long)key & POLLFREE) {
 977                ep_pwq_from_wait(wait)->whead = NULL;
 978                /*
 979                 * whead = NULL above can race with ep_remove_wait_queue()
 980                 * which can do another remove_wait_queue() after us, so we
 981                 * can't use __remove_wait_queue(). whead->lock is held by
 982                 * the caller.
 983                 */
 984                list_del_init(&wait->task_list);
 985        }
 986
 987        spin_lock_irqsave(&ep->lock, flags);
 988
 989        /*
 990         * If the event mask does not contain any poll(2) event, we consider the
 991         * descriptor to be disabled. This condition is likely the effect of the
 992         * EPOLLONESHOT bit that disables the descriptor when an event is received,
 993         * until the next EPOLL_CTL_MOD will be issued.
 994         */
 995        if (!(epi->event.events & ~EP_PRIVATE_BITS))
 996                goto out_unlock;
 997
 998        /*
 999         * Check the events coming with the callback. At this stage, not
1000         * every device reports the events in the "key" parameter of the
1001         * callback. We need to be able to handle both cases here, hence the
1002         * test for "key" != NULL before the event match test.
1003         */
1004        if (key && !((unsigned long) key & epi->event.events))
1005                goto out_unlock;
1006
1007        /*
1008         * If we are transferring events to userspace, we can hold no locks
1009         * (because we're accessing user memory, and because of linux f_op->poll()
1010         * semantics). All the events that happen during that period of time are
1011         * chained in ep->ovflist and requeued later on.
1012         */
1013        if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) {
1014                if (epi->next == EP_UNACTIVE_PTR) {
1015                        epi->next = ep->ovflist;
1016                        ep->ovflist = epi;
1017                        if (epi->ws) {
1018                                /*
1019                                 * Activate ep->ws since epi->ws may get
1020                                 * deactivated at any time.
1021                                 */
1022                                __pm_stay_awake(ep->ws);
1023                        }
1024
1025                }
1026                goto out_unlock;
1027        }
1028
1029        /* If this file is already in the ready list we exit soon */
1030        if (!ep_is_linked(&epi->rdllink)) {
1031                list_add_tail(&epi->rdllink, &ep->rdllist);
1032                ep_pm_stay_awake_rcu(epi);
1033        }
1034
1035        /*
1036         * Wake up ( if active ) both the eventpoll wait list and the ->poll()
1037         * wait list.
1038         */
1039        if (waitqueue_active(&ep->wq))
1040                wake_up_locked(&ep->wq);
1041        if (waitqueue_active(&ep->poll_wait))
1042                pwake++;
1043
1044out_unlock:
1045        spin_unlock_irqrestore(&ep->lock, flags);
1046
1047        /* We have to call this outside the lock */
1048        if (pwake)
1049                ep_poll_safewake(&ep->poll_wait);
1050
1051        return 1;
1052}
1053
1054/*
1055 * This is the callback that is used to add our wait queue to the
1056 * target file wakeup lists.
1057 */
1058static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1059                                 poll_table *pt)
1060{
1061        struct epitem *epi = ep_item_from_epqueue(pt);
1062        struct eppoll_entry *pwq;
1063
1064        if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
1065                init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1066                pwq->whead = whead;
1067                pwq->base = epi;
1068                add_wait_queue(whead, &pwq->wait);
1069                list_add_tail(&pwq->llink, &epi->pwqlist);
1070                epi->nwait++;
1071        } else {
1072                /* We have to signal that an error occurred */
1073                epi->nwait = -1;
1074        }
1075}
1076
1077static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1078{
1079        int kcmp;
1080        struct rb_node **p = &ep->rbr.rb_node, *parent = NULL;
1081        struct epitem *epic;
1082
1083        while (*p) {
1084                parent = *p;
1085                epic = rb_entry(parent, struct epitem, rbn);
1086                kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1087                if (kcmp > 0)
1088                        p = &parent->rb_right;
1089                else
1090                        p = &parent->rb_left;
1091        }
1092        rb_link_node(&epi->rbn, parent, p);
1093        rb_insert_color(&epi->rbn, &ep->rbr);
1094}
1095
1096
1097
1098#define PATH_ARR_SIZE 5
1099/*
1100 * These are the number paths of length 1 to 5, that we are allowing to emanate
1101 * from a single file of interest. For example, we allow 1000 paths of length
1102 * 1, to emanate from each file of interest. This essentially represents the
1103 * potential wakeup paths, which need to be limited in order to avoid massive
1104 * uncontrolled wakeup storms. The common use case should be a single ep which
1105 * is connected to n file sources. In this case each file source has 1 path
1106 * of length 1. Thus, the numbers below should be more than sufficient. These
1107 * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify
1108 * and delete can't add additional paths. Protected by the epmutex.
1109 */
1110static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1111static int path_count[PATH_ARR_SIZE];
1112
1113static int path_count_inc(int nests)
1114{
1115        /* Allow an arbitrary number of depth 1 paths */
1116        if (nests == 0)
1117                return 0;
1118
1119        if (++path_count[nests] > path_limits[nests])
1120                return -1;
1121        return 0;
1122}
1123
1124static void path_count_init(void)
1125{
1126        int i;
1127
1128        for (i = 0; i < PATH_ARR_SIZE; i++)
1129                path_count[i] = 0;
1130}
1131
1132static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
1133{
1134        int error = 0;
1135        struct file *file = priv;
1136        struct file *child_file;
1137        struct epitem *epi;
1138
1139        list_for_each_entry(epi, &file->f_ep_links, fllink) {
1140                child_file = epi->ep->file;
1141                if (is_file_epoll(child_file)) {
1142                        if (list_empty(&child_file->f_ep_links)) {
1143                                if (path_count_inc(call_nests)) {
1144                                        error = -1;
1145                                        break;
1146                                }
1147                        } else {
1148                                error = ep_call_nested(&poll_loop_ncalls,
1149                                                        EP_MAX_NESTS,
1150                                                        reverse_path_check_proc,
1151                                                        child_file, child_file,
1152                                                        current);
1153                        }
1154                        if (error != 0)
1155                                break;
1156                } else {
1157                        printk(KERN_ERR "reverse_path_check_proc: "
1158                                "file is not an ep!\n");
1159                }
1160        }
1161        return error;
1162}
1163
1164/**
1165 * reverse_path_check - The tfile_check_list is list of file *, which have
1166 *                      links that are proposed to be newly added. We need to
1167 *                      make sure that those added links don't add too many
1168 *                      paths such that we will spend all our time waking up
1169 *                      eventpoll objects.
1170 *
1171 * Returns: Returns zero if the proposed links don't create too many paths,
1172 *          -1 otherwise.
1173 */
1174static int reverse_path_check(void)
1175{
1176        int error = 0;
1177        struct file *current_file;
1178
1179        /* let's call this for all tfiles */
1180        list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) {
1181                path_count_init();
1182                error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1183                                        reverse_path_check_proc, current_file,
1184                                        current_file, current);
1185                if (error)
1186                        break;
1187        }
1188        return error;
1189}
1190
1191static int ep_create_wakeup_source(struct epitem *epi)
1192{
1193        const char *name;
1194        struct wakeup_source *ws;
1195
1196        if (!epi->ep->ws) {
1197                epi->ep->ws = wakeup_source_register("eventpoll");
1198                if (!epi->ep->ws)
1199                        return -ENOMEM;
1200        }
1201
1202        name = epi->ffd.file->f_path.dentry->d_name.name;
1203        ws = wakeup_source_register(name);
1204
1205        if (!ws)
1206                return -ENOMEM;
1207        rcu_assign_pointer(epi->ws, ws);
1208
1209        return 0;
1210}
1211
1212/* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */
1213static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1214{
1215        struct wakeup_source *ws = ep_wakeup_source(epi);
1216
1217        RCU_INIT_POINTER(epi->ws, NULL);
1218
1219        /*
1220         * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is
1221         * used internally by wakeup_source_remove, too (called by
1222         * wakeup_source_unregister), so we cannot use call_rcu
1223         */
1224        synchronize_rcu();
1225        wakeup_source_unregister(ws);
1226}
1227
1228/*
1229 * Must be called with "mtx" held.
1230 */
1231static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
1232                     struct file *tfile, int fd)
1233{
1234        int error, revents, pwake = 0;
1235        unsigned long flags;
1236        long user_watches;
1237        struct epitem *epi;
1238        struct ep_pqueue epq;
1239
1240        user_watches = atomic_long_read(&ep->user->epoll_watches);
1241        if (unlikely(user_watches >= max_user_watches))
1242                return -ENOSPC;
1243        if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
1244                return -ENOMEM;
1245
1246        /* Item initialization follow here ... */
1247        INIT_LIST_HEAD(&epi->rdllink);
1248        INIT_LIST_HEAD(&epi->fllink);
1249        INIT_LIST_HEAD(&epi->pwqlist);
1250        epi->ep = ep;
1251        ep_set_ffd(&epi->ffd, tfile, fd);
1252        epi->event = *event;
1253        epi->nwait = 0;
1254        epi->next = EP_UNACTIVE_PTR;
1255        if (epi->event.events & EPOLLWAKEUP) {
1256                error = ep_create_wakeup_source(epi);
1257                if (error)
1258                        goto error_create_wakeup_source;
1259        } else {
1260                RCU_INIT_POINTER(epi->ws, NULL);
1261        }
1262
1263        /* Initialize the poll table using the queue callback */
1264        epq.epi = epi;
1265        init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1266
1267        /*
1268         * Attach the item to the poll hooks and get current event bits.
1269         * We can safely use the file* here because its usage count has
1270         * been increased by the caller of this function. Note that after
1271         * this operation completes, the poll callback can start hitting
1272         * the new item.
1273         */
1274        revents = ep_item_poll(epi, &epq.pt);
1275
1276        /*
1277         * We have to check if something went wrong during the poll wait queue
1278         * install process. Namely an allocation for a wait queue failed due
1279         * high memory pressure.
1280         */
1281        error = -ENOMEM;
1282        if (epi->nwait < 0)
1283                goto error_unregister;
1284
1285        /* Add the current item to the list of active epoll hook for this file */
1286        spin_lock(&tfile->f_lock);
1287        list_add_tail(&epi->fllink, &tfile->f_ep_links);
1288        spin_unlock(&tfile->f_lock);
1289
1290        /*
1291         * Add the current item to the RB tree. All RB tree operations are
1292         * protected by "mtx", and ep_insert() is called with "mtx" held.
1293         */
1294        ep_rbtree_insert(ep, epi);
1295
1296        /* now check if we've created too many backpaths */
1297        error = -EINVAL;
1298        if (reverse_path_check())
1299                goto error_remove_epi;
1300
1301        /* We have to drop the new item inside our item list to keep track of it */
1302        spin_lock_irqsave(&ep->lock, flags);
1303
1304        /* If the file is already "ready" we drop it inside the ready list */
1305        if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) {
1306                list_add_tail(&epi->rdllink, &ep->rdllist);
1307                ep_pm_stay_awake(epi);
1308
1309                /* Notify waiting tasks that events are available */
1310                if (waitqueue_active(&ep->wq))
1311                        wake_up_locked(&ep->wq);
1312                if (waitqueue_active(&ep->poll_wait))
1313                        pwake++;
1314        }
1315
1316        spin_unlock_irqrestore(&ep->lock, flags);
1317
1318        atomic_long_inc(&ep->user->epoll_watches);
1319
1320        /* We have to call this outside the lock */
1321        if (pwake)
1322                ep_poll_safewake(&ep->poll_wait);
1323
1324        return 0;
1325
1326error_remove_epi:
1327        spin_lock(&tfile->f_lock);
1328        if (ep_is_linked(&epi->fllink))
1329                list_del_init(&epi->fllink);
1330        spin_unlock(&tfile->f_lock);
1331
1332        rb_erase(&epi->rbn, &ep->rbr);
1333
1334error_unregister:
1335        ep_unregister_pollwait(ep, epi);
1336
1337        /*
1338         * We need to do this because an event could have been arrived on some
1339         * allocated wait queue. Note that we don't care about the ep->ovflist
1340         * list, since that is used/cleaned only inside a section bound by "mtx".
1341         * And ep_insert() is called with "mtx" held.
1342         */
1343        spin_lock_irqsave(&ep->lock, flags);
1344        if (ep_is_linked(&epi->rdllink))
1345                list_del_init(&epi->rdllink);
1346        spin_unlock_irqrestore(&ep->lock, flags);
1347
1348        wakeup_source_unregister(ep_wakeup_source(epi));
1349
1350error_create_wakeup_source:
1351        kmem_cache_free(epi_cache, epi);
1352
1353        return error;
1354}
1355
1356/*
1357 * Modify the interest event mask by dropping an event if the new mask
1358 * has a match in the current file status. Must be called with "mtx" held.
1359 */
1360static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event)
1361{
1362        int pwake = 0;
1363        unsigned int revents;
1364        poll_table pt;
1365
1366        init_poll_funcptr(&pt, NULL);
1367
1368        /*
1369         * Set the new event interest mask before calling f_op->poll();
1370         * otherwise we might miss an event that happens between the
1371         * f_op->poll() call and the new event set registering.
1372         */
1373        epi->event.events = event->events; /* need barrier below */
1374        epi->event.data = event->data; /* protected by mtx */
1375        if (epi->event.events & EPOLLWAKEUP) {
1376                if (!ep_has_wakeup_source(epi))
1377                        ep_create_wakeup_source(epi);
1378        } else if (ep_has_wakeup_source(epi)) {
1379                ep_destroy_wakeup_source(epi);
1380        }
1381
1382        /*
1383         * The following barrier has two effects:
1384         *
1385         * 1) Flush epi changes above to other CPUs.  This ensures
1386         *    we do not miss events from ep_poll_callback if an
1387         *    event occurs immediately after we call f_op->poll().
1388         *    We need this because we did not take ep->lock while
1389         *    changing epi above (but ep_poll_callback does take
1390         *    ep->lock).
1391         *
1392         * 2) We also need to ensure we do not miss _past_ events
1393         *    when calling f_op->poll().  This barrier also
1394         *    pairs with the barrier in wq_has_sleeper (see
1395         *    comments for wq_has_sleeper).
1396         *
1397         * This barrier will now guarantee ep_poll_callback or f_op->poll
1398         * (or both) will notice the readiness of an item.
1399         */
1400        smp_mb();
1401
1402        /*
1403         * Get current event bits. We can safely use the file* here because
1404         * its usage count has been increased by the caller of this function.
1405         */
1406        revents = ep_item_poll(epi, &pt);
1407
1408        /*
1409         * If the item is "hot" and it is not registered inside the ready
1410         * list, push it inside.
1411         */
1412        if (revents & event->events) {
1413                spin_lock_irq(&ep->lock);
1414                if (!ep_is_linked(&epi->rdllink)) {
1415                        list_add_tail(&epi->rdllink, &ep->rdllist);
1416                        ep_pm_stay_awake(epi);
1417
1418                        /* Notify waiting tasks that events are available */
1419                        if (waitqueue_active(&ep->wq))
1420                                wake_up_locked(&ep->wq);
1421                        if (waitqueue_active(&ep->poll_wait))
1422                                pwake++;
1423                }
1424                spin_unlock_irq(&ep->lock);
1425        }
1426
1427        /* We have to call this outside the lock */
1428        if (pwake)
1429                ep_poll_safewake(&ep->poll_wait);
1430
1431        return 0;
1432}
1433
1434static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head,
1435                               void *priv)
1436{
1437        struct ep_send_events_data *esed = priv;
1438        int eventcnt;
1439        unsigned int revents;
1440        struct epitem *epi;
1441        struct epoll_event __user *uevent;
1442        struct wakeup_source *ws;
1443        poll_table pt;
1444
1445        init_poll_funcptr(&pt, NULL);
1446
1447        /*
1448         * We can loop without lock because we are passed a task private list.
1449         * Items cannot vanish during the loop because ep_scan_ready_list() is
1450         * holding "mtx" during this call.
1451         */
1452        for (eventcnt = 0, uevent = esed->events;
1453             !list_empty(head) && eventcnt < esed->maxevents;) {
1454                epi = list_first_entry(head, struct epitem, rdllink);
1455
1456                /*
1457                 * Activate ep->ws before deactivating epi->ws to prevent
1458                 * triggering auto-suspend here (in case we reactive epi->ws
1459                 * below).
1460                 *
1461                 * This could be rearranged to delay the deactivation of epi->ws
1462                 * instead, but then epi->ws would temporarily be out of sync
1463                 * with ep_is_linked().
1464                 */
1465                ws = ep_wakeup_source(epi);
1466                if (ws) {
1467                        if (ws->active)
1468                                __pm_stay_awake(ep->ws);
1469                        __pm_relax(ws);
1470                }
1471
1472                list_del_init(&epi->rdllink);
1473
1474                revents = ep_item_poll(epi, &pt);
1475
1476                /*
1477                 * If the event mask intersect the caller-requested one,
1478                 * deliver the event to userspace. Again, ep_scan_ready_list()
1479                 * is holding "mtx", so no operations coming from userspace
1480                 * can change the item.
1481                 */
1482                if (revents) {
1483                        if (__put_user(revents, &uevent->events) ||
1484                            __put_user(epi->event.data, &uevent->data)) {
1485                                list_add(&epi->rdllink, head);
1486                                ep_pm_stay_awake(epi);
1487                                return eventcnt ? eventcnt : -EFAULT;
1488                        }
1489                        eventcnt++;
1490                        uevent++;
1491                        if (epi->event.events & EPOLLONESHOT)
1492                                epi->event.events &= EP_PRIVATE_BITS;
1493                        else if (!(epi->event.events & EPOLLET)) {
1494                                /*
1495                                 * If this file has been added with Level
1496                                 * Trigger mode, we need to insert back inside
1497                                 * the ready list, so that the next call to
1498                                 * epoll_wait() will check again the events
1499                                 * availability. At this point, no one can insert
1500                                 * into ep->rdllist besides us. The epoll_ctl()
1501                                 * callers are locked out by
1502                                 * ep_scan_ready_list() holding "mtx" and the
1503                                 * poll callback will queue them in ep->ovflist.
1504                                 */
1505                                list_add_tail(&epi->rdllink, &ep->rdllist);
1506                                ep_pm_stay_awake(epi);
1507                        }
1508                }
1509        }
1510
1511        return eventcnt;
1512}
1513
1514static int ep_send_events(struct eventpoll *ep,
1515                          struct epoll_event __user *events, int maxevents)
1516{
1517        struct ep_send_events_data esed;
1518
1519        esed.maxevents = maxevents;
1520        esed.events = events;
1521
1522        return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0);
1523}
1524
1525static inline struct timespec ep_set_mstimeout(long ms)
1526{
1527        struct timespec now, ts = {
1528                .tv_sec = ms / MSEC_PER_SEC,
1529                .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC),
1530        };
1531
1532        ktime_get_ts(&now);
1533        return timespec_add_safe(now, ts);
1534}
1535
1536/**
1537 * ep_poll - Retrieves ready events, and delivers them to the caller supplied
1538 *           event buffer.
1539 *
1540 * @ep: Pointer to the eventpoll context.
1541 * @events: Pointer to the userspace buffer where the ready events should be
1542 *          stored.
1543 * @maxevents: Size (in terms of number of events) of the caller event buffer.
1544 * @timeout: Maximum timeout for the ready events fetch operation, in
1545 *           milliseconds. If the @timeout is zero, the function will not block,
1546 *           while if the @timeout is less than zero, the function will block
1547 *           until at least one event has been retrieved (or an error
1548 *           occurred).
1549 *
1550 * Returns: Returns the number of ready events which have been fetched, or an
1551 *          error code, in case of error.
1552 */
1553static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1554                   int maxevents, long timeout)
1555{
1556        int res = 0, eavail, timed_out = 0;
1557        unsigned long flags;
1558        long slack = 0;
1559        wait_queue_t wait;
1560        ktime_t expires, *to = NULL;
1561
1562        if (timeout > 0) {
1563                struct timespec end_time = ep_set_mstimeout(timeout);
1564
1565                slack = select_estimate_accuracy(&end_time);
1566                to = &expires;
1567                *to = timespec_to_ktime(end_time);
1568        } else if (timeout == 0) {
1569                /*
1570                 * Avoid the unnecessary trip to the wait queue loop, if the
1571                 * caller specified a non blocking operation.
1572                 */
1573                timed_out = 1;
1574                spin_lock_irqsave(&ep->lock, flags);
1575                goto check_events;
1576        }
1577
1578fetch_events:
1579        spin_lock_irqsave(&ep->lock, flags);
1580
1581        if (!ep_events_available(ep)) {
1582                /*
1583                 * We don't have any available event to return to the caller.
1584                 * We need to sleep here, and we will be wake up by
1585                 * ep_poll_callback() when events will become available.
1586                 */
1587                init_waitqueue_entry(&wait, current);
1588                __add_wait_queue_exclusive(&ep->wq, &wait);
1589
1590                for (;;) {
1591                        /*
1592                         * We don't want to sleep if the ep_poll_callback() sends us
1593                         * a wakeup in between. That's why we set the task state
1594                         * to TASK_INTERRUPTIBLE before doing the checks.
1595                         */
1596                        set_current_state(TASK_INTERRUPTIBLE);
1597                        if (ep_events_available(ep) || timed_out)
1598                                break;
1599                        if (signal_pending(current)) {
1600                                res = -EINTR;
1601                                break;
1602                        }
1603
1604                        spin_unlock_irqrestore(&ep->lock, flags);
1605                        if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
1606                                timed_out = 1;
1607
1608                        spin_lock_irqsave(&ep->lock, flags);
1609                }
1610                __remove_wait_queue(&ep->wq, &wait);
1611
1612                set_current_state(TASK_RUNNING);
1613        }
1614check_events:
1615        /* Is it worth to try to dig for events ? */
1616        eavail = ep_events_available(ep);
1617
1618        spin_unlock_irqrestore(&ep->lock, flags);
1619
1620        /*
1621         * Try to transfer events to user space. In case we get 0 events and
1622         * there's still timeout left over, we go trying again in search of
1623         * more luck.
1624         */
1625        if (!res && eavail &&
1626            !(res = ep_send_events(ep, events, maxevents)) && !timed_out)
1627                goto fetch_events;
1628
1629        return res;
1630}
1631
1632/**
1633 * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested()
1634 *                      API, to verify that adding an epoll file inside another
1635 *                      epoll structure, does not violate the constraints, in
1636 *                      terms of closed loops, or too deep chains (which can
1637 *                      result in excessive stack usage).
1638 *
1639 * @priv: Pointer to the epoll file to be currently checked.
1640 * @cookie: Original cookie for this call. This is the top-of-the-chain epoll
1641 *          data structure pointer.
1642 * @call_nests: Current dept of the @ep_call_nested() call stack.
1643 *
1644 * Returns: Returns zero if adding the epoll @file inside current epoll
1645 *          structure @ep does not violate the constraints, or -1 otherwise.
1646 */
1647static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
1648{
1649        int error = 0;
1650        struct file *file = priv;
1651        struct eventpoll *ep = file->private_data;
1652        struct eventpoll *ep_tovisit;
1653        struct rb_node *rbp;
1654        struct epitem *epi;
1655
1656        mutex_lock_nested(&ep->mtx, call_nests + 1);
1657        ep->visited = 1;
1658        list_add(&ep->visited_list_link, &visited_list);
1659        for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1660                epi = rb_entry(rbp, struct epitem, rbn);
1661                if (unlikely(is_file_epoll(epi->ffd.file))) {
1662                        ep_tovisit = epi->ffd.file->private_data;
1663                        if (ep_tovisit->visited)
1664                                continue;
1665                        error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1666                                        ep_loop_check_proc, epi->ffd.file,
1667                                        ep_tovisit, current);
1668                        if (error != 0)
1669                                break;
1670                } else {
1671                        /*
1672                         * If we've reached a file that is not associated with
1673                         * an ep, then we need to check if the newly added
1674                         * links are going to add too many wakeup paths. We do
1675                         * this by adding it to the tfile_check_list, if it's
1676                         * not already there, and calling reverse_path_check()
1677                         * during ep_insert().
1678                         */
1679                        if (list_empty(&epi->ffd.file->f_tfile_llink))
1680                                list_add(&epi->ffd.file->f_tfile_llink,
1681                                         &tfile_check_list);
1682                }
1683        }
1684        mutex_unlock(&ep->mtx);
1685
1686        return error;
1687}
1688
1689/**
1690 * ep_loop_check - Performs a check to verify that adding an epoll file (@file)
1691 *                 another epoll file (represented by @ep) does not create
1692 *                 closed loops or too deep chains.
1693 *
1694 * @ep: Pointer to the epoll private data structure.
1695 * @file: Pointer to the epoll file to be checked.
1696 *
1697 * Returns: Returns zero if adding the epoll @file inside current epoll
1698 *          structure @ep does not violate the constraints, or -1 otherwise.
1699 */
1700static int ep_loop_check(struct eventpoll *ep, struct file *file)
1701{
1702        int ret;
1703        struct eventpoll *ep_cur, *ep_next;
1704
1705        ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS,
1706                              ep_loop_check_proc, file, ep, current);
1707        /* clear visited list */
1708        list_for_each_entry_safe(ep_cur, ep_next, &visited_list,
1709                                                        visited_list_link) {
1710                ep_cur->visited = 0;
1711                list_del(&ep_cur->visited_list_link);
1712        }
1713        return ret;
1714}
1715
1716static void clear_tfile_check_list(void)
1717{
1718        struct file *file;
1719
1720        /* first clear the tfile_check_list */
1721        while (!list_empty(&tfile_check_list)) {
1722                file = list_first_entry(&tfile_check_list, struct file,
1723                                        f_tfile_llink);
1724                list_del_init(&file->f_tfile_llink);
1725        }
1726        INIT_LIST_HEAD(&tfile_check_list);
1727}
1728
1729/*
1730 * Open an eventpoll file descriptor.
1731 */
1732SYSCALL_DEFINE1(epoll_create1, int, flags)
1733{
1734        int error, fd;
1735        struct eventpoll *ep = NULL;
1736        struct file *file;
1737
1738        /* Check the EPOLL_* constant for consistency.  */
1739        BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1740
1741        if (flags & ~EPOLL_CLOEXEC)
1742                return -EINVAL;
1743        /*
1744         * Create the internal data structure ("struct eventpoll").
1745         */
1746        error = ep_alloc(&ep);
1747        if (error < 0)
1748                return error;
1749        /*
1750         * Creates all the items needed to setup an eventpoll file. That is,
1751         * a file structure and a free file descriptor.
1752         */
1753        fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1754        if (fd < 0) {
1755                error = fd;
1756                goto out_free_ep;
1757        }
1758        file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
1759                                 O_RDWR | (flags & O_CLOEXEC));
1760        if (IS_ERR(file)) {
1761                error = PTR_ERR(file);
1762                goto out_free_fd;
1763        }
1764        ep->file = file;
1765        fd_install(fd, file);
1766        return fd;
1767
1768out_free_fd:
1769        put_unused_fd(fd);
1770out_free_ep:
1771        ep_free(ep);
1772        return error;
1773}
1774
1775SYSCALL_DEFINE1(epoll_create, int, size)
1776{
1777        if (size <= 0)
1778                return -EINVAL;
1779
1780        return sys_epoll_create1(0);
1781}
1782
1783/*
1784 * The following function implements the controller interface for
1785 * the eventpoll file that enables the insertion/removal/change of
1786 * file descriptors inside the interest set.
1787 */
1788SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
1789                struct epoll_event __user *, event)
1790{
1791        int error;
1792        int did_lock_epmutex = 0;
1793        struct file *file, *tfile;
1794        struct eventpoll *ep;
1795        struct epitem *epi;
1796        struct epoll_event epds;
1797
1798        error = -EFAULT;
1799        if (ep_op_has_event(op) &&
1800            copy_from_user(&epds, event, sizeof(struct epoll_event)))
1801                goto error_return;
1802
1803        /* Get the "struct file *" for the eventpoll file */
1804        error = -EBADF;
1805        file = fget(epfd);
1806        if (!file)
1807                goto error_return;
1808
1809        /* Get the "struct file *" for the target file */
1810        tfile = fget(fd);
1811        if (!tfile)
1812                goto error_fput;
1813
1814        /* The target file descriptor must support poll */
1815        error = -EPERM;
1816        if (!tfile->f_op || !tfile->f_op->poll)
1817                goto error_tgt_fput;
1818
1819        /* Check if EPOLLWAKEUP is allowed */
1820        if ((epds.events & EPOLLWAKEUP) && !capable(CAP_BLOCK_SUSPEND))
1821                epds.events &= ~EPOLLWAKEUP;
1822
1823        /*
1824         * We have to check that the file structure underneath the file descriptor
1825         * the user passed to us _is_ an eventpoll file. And also we do not permit
1826         * adding an epoll file descriptor inside itself.
1827         */
1828        error = -EINVAL;
1829        if (file == tfile || !is_file_epoll(file))
1830                goto error_tgt_fput;
1831
1832        /*
1833         * At this point it is safe to assume that the "private_data" contains
1834         * our own data structure.
1835         */
1836        ep = file->private_data;
1837
1838        /*
1839         * When we insert an epoll file descriptor, inside another epoll file
1840         * descriptor, there is the change of creating closed loops, which are
1841         * better be handled here, than in more critical paths. While we are
1842         * checking for loops we also determine the list of files reachable
1843         * and hang them on the tfile_check_list, so we can check that we
1844         * haven't created too many possible wakeup paths.
1845         *
1846         * We need to hold the epmutex across both ep_insert and ep_remove
1847         * b/c we want to make sure we are looking at a coherent view of
1848         * epoll network.
1849         */
1850        if (op == EPOLL_CTL_ADD || op == EPOLL_CTL_DEL) {
1851                mutex_lock(&epmutex);
1852                did_lock_epmutex = 1;
1853        }
1854        if (op == EPOLL_CTL_ADD) {
1855                if (is_file_epoll(tfile)) {
1856                        error = -ELOOP;
1857                        if (ep_loop_check(ep, tfile) != 0) {
1858                                clear_tfile_check_list();
1859                                goto error_tgt_fput;
1860                        }
1861                } else
1862                        list_add(&tfile->f_tfile_llink, &tfile_check_list);
1863        }
1864
1865        mutex_lock_nested(&ep->mtx, 0);
1866
1867        /*
1868         * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1869         * above, we can be sure to be able to use the item looked up by
1870         * ep_find() till we release the mutex.
1871         */
1872        epi = ep_find(ep, tfile, fd);
1873
1874        error = -EINVAL;
1875        switch (op) {
1876        case EPOLL_CTL_ADD:
1877                if (!epi) {
1878                        epds.events |= POLLERR | POLLHUP;
1879                        error = ep_insert(ep, &epds, tfile, fd);
1880                } else
1881                        error = -EEXIST;
1882                clear_tfile_check_list();
1883                break;
1884        case EPOLL_CTL_DEL:
1885                if (epi)
1886                        error = ep_remove(ep, epi);
1887                else
1888                        error = -ENOENT;
1889                break;
1890        case EPOLL_CTL_MOD:
1891                if (epi) {
1892                        epds.events |= POLLERR | POLLHUP;
1893                        error = ep_modify(ep, epi, &epds);
1894                } else
1895                        error = -ENOENT;
1896                break;
1897        }
1898        mutex_unlock(&ep->mtx);
1899
1900error_tgt_fput:
1901        if (did_lock_epmutex)
1902                mutex_unlock(&epmutex);
1903
1904        fput(tfile);
1905error_fput:
1906        fput(file);
1907error_return:
1908
1909        return error;
1910}
1911
1912/*
1913 * Implement the event wait interface for the eventpoll file. It is the kernel
1914 * part of the user space epoll_wait(2).
1915 */
1916SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
1917                int, maxevents, int, timeout)
1918{
1919        int error;
1920        struct fd f;
1921        struct eventpoll *ep;
1922
1923        /* The maximum number of event must be greater than zero */
1924        if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
1925                return -EINVAL;
1926
1927        /* Verify that the area passed by the user is writeable */
1928        if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event)))
1929                return -EFAULT;
1930
1931        /* Get the "struct file *" for the eventpoll file */
1932        f = fdget(epfd);
1933        if (!f.file)
1934                return -EBADF;
1935
1936        /*
1937         * We have to check that the file structure underneath the fd
1938         * the user passed to us _is_ an eventpoll file.
1939         */
1940        error = -EINVAL;
1941        if (!is_file_epoll(f.file))
1942                goto error_fput;
1943
1944        /*
1945         * At this point it is safe to assume that the "private_data" contains
1946         * our own data structure.
1947         */
1948        ep = f.file->private_data;
1949
1950        /* Time to fish for events ... */
1951        error = ep_poll(ep, events, maxevents, timeout);
1952
1953error_fput:
1954        fdput(f);
1955        return error;
1956}
1957
1958/*
1959 * Implement the event wait interface for the eventpoll file. It is the kernel
1960 * part of the user space epoll_pwait(2).
1961 */
1962SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
1963                int, maxevents, int, timeout, const sigset_t __user *, sigmask,
1964                size_t, sigsetsize)
1965{
1966        int error;
1967        sigset_t ksigmask, sigsaved;
1968
1969        /*
1970         * If the caller wants a certain signal mask to be set during the wait,
1971         * we apply it here.
1972         */
1973        if (sigmask) {
1974                if (sigsetsize != sizeof(sigset_t))
1975                        return -EINVAL;
1976                if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1977                        return -EFAULT;
1978                sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1979                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1980        }
1981
1982        error = sys_epoll_wait(epfd, events, maxevents, timeout);
1983
1984        /*
1985         * If we changed the signal mask, we need to restore the original one.
1986         * In case we've got a signal while waiting, we do not restore the
1987         * signal mask yet, and we allow do_signal() to deliver the signal on
1988         * the way back to userspace, before the signal mask is restored.
1989         */
1990        if (sigmask) {
1991                if (error == -EINTR) {
1992                        memcpy(&current->saved_sigmask, &sigsaved,
1993                               sizeof(sigsaved));
1994                        set_restore_sigmask();
1995                } else
1996                        sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1997        }
1998
1999        return error;
2000}
2001
2002#ifdef CONFIG_COMPAT
2003COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2004                        struct epoll_event __user *, events,
2005                        int, maxevents, int, timeout,
2006                        const compat_sigset_t __user *, sigmask,
2007                        compat_size_t, sigsetsize)
2008{
2009        long err;
2010        compat_sigset_t csigmask;
2011        sigset_t ksigmask, sigsaved;
2012
2013        /*
2014         * If the caller wants a certain signal mask to be set during the wait,
2015         * we apply it here.
2016         */
2017        if (sigmask) {
2018                if (sigsetsize != sizeof(compat_sigset_t))
2019                        return -EINVAL;
2020                if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
2021                        return -EFAULT;
2022                sigset_from_compat(&ksigmask, &csigmask);
2023                sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
2024                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
2025        }
2026
2027        err = sys_epoll_wait(epfd, events, maxevents, timeout);
2028
2029        /*
2030         * If we changed the signal mask, we need to restore the original one.
2031         * In case we've got a signal while waiting, we do not restore the
2032         * signal mask yet, and we allow do_signal() to deliver the signal on
2033         * the way back to userspace, before the signal mask is restored.
2034         */
2035        if (sigmask) {
2036                if (err == -EINTR) {
2037                        memcpy(&current->saved_sigmask, &sigsaved,
2038                               sizeof(sigsaved));
2039                        set_restore_sigmask();
2040                } else
2041                        sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2042        }
2043
2044        return err;
2045}
2046#endif
2047
2048static int __init eventpoll_init(void)
2049{
2050        struct sysinfo si;
2051
2052        si_meminfo(&si);
2053        /*
2054         * Allows top 4% of lomem to be allocated for epoll watches (per user).
2055         */
2056        max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2057                EP_ITEM_COST;
2058        BUG_ON(max_user_watches < 0);
2059
2060        /*
2061         * Initialize the structure used to perform epoll file descriptor
2062         * inclusion loops checks.
2063         */
2064        ep_nested_calls_init(&poll_loop_ncalls);
2065
2066        /* Initialize the structure used to perform safe poll wait head wake ups */
2067        ep_nested_calls_init(&poll_safewake_ncalls);
2068
2069        /* Initialize the structure used to perform file's f_op->poll() calls */
2070        ep_nested_calls_init(&poll_readywalk_ncalls);
2071
2072        /*
2073         * We can have many thousands of epitems, so prevent this from
2074         * using an extra cache line on 64-bit (and smaller) CPUs
2075         */
2076        BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2077
2078        /* Allocates slab cache used to allocate "struct epitem" items */
2079        epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
2080                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2081
2082        /* Allocates slab cache used to allocate "struct eppoll_entry" */
2083        pwq_cache = kmem_cache_create("eventpoll_pwq",
2084                        sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL);
2085
2086        return 0;
2087}
2088fs_initcall(eventpoll_init);
2089
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.