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