linux/fs/inode.c
<<
>>
Prefs
   1/*
   2 * (C) 1997 Linus Torvalds
   3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
   4 */
   5#include <linux/fs.h>
   6#include <linux/mm.h>
   7#include <linux/dcache.h>
   8#include <linux/init.h>
   9#include <linux/slab.h>
  10#include <linux/writeback.h>
  11#include <linux/module.h>
  12#include <linux/backing-dev.h>
  13#include <linux/wait.h>
  14#include <linux/rwsem.h>
  15#include <linux/hash.h>
  16#include <linux/swap.h>
  17#include <linux/security.h>
  18#include <linux/pagemap.h>
  19#include <linux/cdev.h>
  20#include <linux/bootmem.h>
  21#include <linux/fsnotify.h>
  22#include <linux/mount.h>
  23#include <linux/async.h>
  24#include <linux/posix_acl.h>
  25#include <linux/prefetch.h>
  26#include <linux/ima.h>
  27#include <linux/cred.h>
  28#include <linux/buffer_head.h> /* for inode_has_buffers */
  29#include "internal.h"
  30
  31/*
  32 * Inode locking rules:
  33 *
  34 * inode->i_lock protects:
  35 *   inode->i_state, inode->i_hash, __iget()
  36 * inode->i_sb->s_inode_lru_lock protects:
  37 *   inode->i_sb->s_inode_lru, inode->i_lru
  38 * inode_sb_list_lock protects:
  39 *   sb->s_inodes, inode->i_sb_list
  40 * bdi->wb.list_lock protects:
  41 *   bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
  42 * inode_hash_lock protects:
  43 *   inode_hashtable, inode->i_hash
  44 *
  45 * Lock ordering:
  46 *
  47 * inode_sb_list_lock
  48 *   inode->i_lock
  49 *     inode->i_sb->s_inode_lru_lock
  50 *
  51 * bdi->wb.list_lock
  52 *   inode->i_lock
  53 *
  54 * inode_hash_lock
  55 *   inode_sb_list_lock
  56 *   inode->i_lock
  57 *
  58 * iunique_lock
  59 *   inode_hash_lock
  60 */
  61
  62static unsigned int i_hash_mask __read_mostly;
  63static unsigned int i_hash_shift __read_mostly;
  64static struct hlist_head *inode_hashtable __read_mostly;
  65static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  66
  67__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
  68
  69/*
  70 * Empty aops. Can be used for the cases where the user does not
  71 * define any of the address_space operations.
  72 */
  73const struct address_space_operations empty_aops = {
  74};
  75EXPORT_SYMBOL(empty_aops);
  76
  77/*
  78 * Statistics gathering..
  79 */
  80struct inodes_stat_t inodes_stat;
  81
  82static DEFINE_PER_CPU(unsigned int, nr_inodes);
  83static DEFINE_PER_CPU(unsigned int, nr_unused);
  84
  85static struct kmem_cache *inode_cachep __read_mostly;
  86
  87static int get_nr_inodes(void)
  88{
  89        int i;
  90        int sum = 0;
  91        for_each_possible_cpu(i)
  92                sum += per_cpu(nr_inodes, i);
  93        return sum < 0 ? 0 : sum;
  94}
  95
  96static inline int get_nr_inodes_unused(void)
  97{
  98        int i;
  99        int sum = 0;
 100        for_each_possible_cpu(i)
 101                sum += per_cpu(nr_unused, i);
 102        return sum < 0 ? 0 : sum;
 103}
 104
 105int get_nr_dirty_inodes(void)
 106{
 107        /* not actually dirty inodes, but a wild approximation */
 108        int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
 109        return nr_dirty > 0 ? nr_dirty : 0;
 110}
 111
 112/*
 113 * Handle nr_inode sysctl
 114 */
 115#ifdef CONFIG_SYSCTL
 116int proc_nr_inodes(ctl_table *table, int write,
 117                   void __user *buffer, size_t *lenp, loff_t *ppos)
 118{
 119        inodes_stat.nr_inodes = get_nr_inodes();
 120        inodes_stat.nr_unused = get_nr_inodes_unused();
 121        return proc_dointvec(table, write, buffer, lenp, ppos);
 122}
 123#endif
 124
 125/**
 126 * inode_init_always - perform inode structure intialisation
 127 * @sb: superblock inode belongs to
 128 * @inode: inode to initialise
 129 *
 130 * These are initializations that need to be done on every inode
 131 * allocation as the fields are not initialised by slab allocation.
 132 */
 133int inode_init_always(struct super_block *sb, struct inode *inode)
 134{
 135        static const struct inode_operations empty_iops;
 136        static const struct file_operations empty_fops;
 137        struct address_space *const mapping = &inode->i_data;
 138
 139        inode->i_sb = sb;
 140        inode->i_blkbits = sb->s_blocksize_bits;
 141        inode->i_flags = 0;
 142        atomic_set(&inode->i_count, 1);
 143        inode->i_op = &empty_iops;
 144        inode->i_fop = &empty_fops;
 145        inode->i_nlink = 1;
 146        inode->i_opflags = 0;
 147        inode->i_uid = 0;
 148        inode->i_gid = 0;
 149        atomic_set(&inode->i_writecount, 0);
 150        inode->i_size = 0;
 151        inode->i_blocks = 0;
 152        inode->i_bytes = 0;
 153        inode->i_generation = 0;
 154#ifdef CONFIG_QUOTA
 155        memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
 156#endif
 157        inode->i_pipe = NULL;
 158        inode->i_bdev = NULL;
 159        inode->i_cdev = NULL;
 160        inode->i_rdev = 0;
 161        inode->dirtied_when = 0;
 162
 163        if (security_inode_alloc(inode))
 164                goto out;
 165        spin_lock_init(&inode->i_lock);
 166        lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 167
 168        mutex_init(&inode->i_mutex);
 169        lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
 170
 171        atomic_set(&inode->i_dio_count, 0);
 172
 173        mapping->a_ops = &empty_aops;
 174        mapping->host = inode;
 175        mapping->flags = 0;
 176        mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 177        mapping->assoc_mapping = NULL;
 178        mapping->backing_dev_info = &default_backing_dev_info;
 179        mapping->writeback_index = 0;
 180
 181        /*
 182         * If the block_device provides a backing_dev_info for client
 183         * inodes then use that.  Otherwise the inode share the bdev's
 184         * backing_dev_info.
 185         */
 186        if (sb->s_bdev) {
 187                struct backing_dev_info *bdi;
 188
 189                bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
 190                mapping->backing_dev_info = bdi;
 191        }
 192        inode->i_private = NULL;
 193        inode->i_mapping = mapping;
 194#ifdef CONFIG_FS_POSIX_ACL
 195        inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 196#endif
 197
 198#ifdef CONFIG_FSNOTIFY
 199        inode->i_fsnotify_mask = 0;
 200#endif
 201
 202        this_cpu_inc(nr_inodes);
 203
 204        return 0;
 205out:
 206        return -ENOMEM;
 207}
 208EXPORT_SYMBOL(inode_init_always);
 209
 210static struct inode *alloc_inode(struct super_block *sb)
 211{
 212        struct inode *inode;
 213
 214        if (sb->s_op->alloc_inode)
 215                inode = sb->s_op->alloc_inode(sb);
 216        else
 217                inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
 218
 219        if (!inode)
 220                return NULL;
 221
 222        if (unlikely(inode_init_always(sb, inode))) {
 223                if (inode->i_sb->s_op->destroy_inode)
 224                        inode->i_sb->s_op->destroy_inode(inode);
 225                else
 226                        kmem_cache_free(inode_cachep, inode);
 227                return NULL;
 228        }
 229
 230        return inode;
 231}
 232
 233void free_inode_nonrcu(struct inode *inode)
 234{
 235        kmem_cache_free(inode_cachep, inode);
 236}
 237EXPORT_SYMBOL(free_inode_nonrcu);
 238
 239void __destroy_inode(struct inode *inode)
 240{
 241        BUG_ON(inode_has_buffers(inode));
 242        security_inode_free(inode);
 243        fsnotify_inode_delete(inode);
 244#ifdef CONFIG_FS_POSIX_ACL
 245        if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
 246                posix_acl_release(inode->i_acl);
 247        if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
 248                posix_acl_release(inode->i_default_acl);
 249#endif
 250        this_cpu_dec(nr_inodes);
 251}
 252EXPORT_SYMBOL(__destroy_inode);
 253
 254static void i_callback(struct rcu_head *head)
 255{
 256        struct inode *inode = container_of(head, struct inode, i_rcu);
 257        INIT_LIST_HEAD(&inode->i_dentry);
 258        kmem_cache_free(inode_cachep, inode);
 259}
 260
 261static void destroy_inode(struct inode *inode)
 262{
 263        BUG_ON(!list_empty(&inode->i_lru));
 264        __destroy_inode(inode);
 265        if (inode->i_sb->s_op->destroy_inode)
 266                inode->i_sb->s_op->destroy_inode(inode);
 267        else
 268                call_rcu(&inode->i_rcu, i_callback);
 269}
 270
 271void address_space_init_once(struct address_space *mapping)
 272{
 273        memset(mapping, 0, sizeof(*mapping));
 274        INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
 275        spin_lock_init(&mapping->tree_lock);
 276        mutex_init(&mapping->i_mmap_mutex);
 277        INIT_LIST_HEAD(&mapping->private_list);
 278        spin_lock_init(&mapping->private_lock);
 279        INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
 280        INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
 281}
 282EXPORT_SYMBOL(address_space_init_once);
 283
 284/*
 285 * These are initializations that only need to be done
 286 * once, because the fields are idempotent across use
 287 * of the inode, so let the slab aware of that.
 288 */
 289void inode_init_once(struct inode *inode)
 290{
 291        memset(inode, 0, sizeof(*inode));
 292        INIT_HLIST_NODE(&inode->i_hash);
 293        INIT_LIST_HEAD(&inode->i_dentry);
 294        INIT_LIST_HEAD(&inode->i_devices);
 295        INIT_LIST_HEAD(&inode->i_wb_list);
 296        INIT_LIST_HEAD(&inode->i_lru);
 297        address_space_init_once(&inode->i_data);
 298        i_size_ordered_init(inode);
 299#ifdef CONFIG_FSNOTIFY
 300        INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
 301#endif
 302}
 303EXPORT_SYMBOL(inode_init_once);
 304
 305static void init_once(void *foo)
 306{
 307        struct inode *inode = (struct inode *) foo;
 308
 309        inode_init_once(inode);
 310}
 311
 312/*
 313 * inode->i_lock must be held
 314 */
 315void __iget(struct inode *inode)
 316{
 317        atomic_inc(&inode->i_count);
 318}
 319
 320/*
 321 * get additional reference to inode; caller must already hold one.
 322 */
 323void ihold(struct inode *inode)
 324{
 325        WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 326}
 327EXPORT_SYMBOL(ihold);
 328
 329static void inode_lru_list_add(struct inode *inode)
 330{
 331        spin_lock(&inode->i_sb->s_inode_lru_lock);
 332        if (list_empty(&inode->i_lru)) {
 333                list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
 334                inode->i_sb->s_nr_inodes_unused++;
 335                this_cpu_inc(nr_unused);
 336        }
 337        spin_unlock(&inode->i_sb->s_inode_lru_lock);
 338}
 339
 340static void inode_lru_list_del(struct inode *inode)
 341{
 342        spin_lock(&inode->i_sb->s_inode_lru_lock);
 343        if (!list_empty(&inode->i_lru)) {
 344                list_del_init(&inode->i_lru);
 345                inode->i_sb->s_nr_inodes_unused--;
 346                this_cpu_dec(nr_unused);
 347        }
 348        spin_unlock(&inode->i_sb->s_inode_lru_lock);
 349}
 350
 351/**
 352 * inode_sb_list_add - add inode to the superblock list of inodes
 353 * @inode: inode to add
 354 */
 355void inode_sb_list_add(struct inode *inode)
 356{
 357        spin_lock(&inode_sb_list_lock);
 358        list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 359        spin_unlock(&inode_sb_list_lock);
 360}
 361EXPORT_SYMBOL_GPL(inode_sb_list_add);
 362
 363static inline void inode_sb_list_del(struct inode *inode)
 364{
 365        if (!list_empty(&inode->i_sb_list)) {
 366                spin_lock(&inode_sb_list_lock);
 367                list_del_init(&inode->i_sb_list);
 368                spin_unlock(&inode_sb_list_lock);
 369        }
 370}
 371
 372static unsigned long hash(struct super_block *sb, unsigned long hashval)
 373{
 374        unsigned long tmp;
 375
 376        tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 377                        L1_CACHE_BYTES;
 378        tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 379        return tmp & i_hash_mask;
 380}
 381
 382/**
 383 *      __insert_inode_hash - hash an inode
 384 *      @inode: unhashed inode
 385 *      @hashval: unsigned long value used to locate this object in the
 386 *              inode_hashtable.
 387 *
 388 *      Add an inode to the inode hash for this superblock.
 389 */
 390void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 391{
 392        struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 393
 394        spin_lock(&inode_hash_lock);
 395        spin_lock(&inode->i_lock);
 396        hlist_add_head(&inode->i_hash, b);
 397        spin_unlock(&inode->i_lock);
 398        spin_unlock(&inode_hash_lock);
 399}
 400EXPORT_SYMBOL(__insert_inode_hash);
 401
 402/**
 403 *      __remove_inode_hash - remove an inode from the hash
 404 *      @inode: inode to unhash
 405 *
 406 *      Remove an inode from the superblock.
 407 */
 408void __remove_inode_hash(struct inode *inode)
 409{
 410        spin_lock(&inode_hash_lock);
 411        spin_lock(&inode->i_lock);
 412        hlist_del_init(&inode->i_hash);
 413        spin_unlock(&inode->i_lock);
 414        spin_unlock(&inode_hash_lock);
 415}
 416EXPORT_SYMBOL(__remove_inode_hash);
 417
 418void end_writeback(struct inode *inode)
 419{
 420        might_sleep();
 421        /*
 422         * We have to cycle tree_lock here because reclaim can be still in the
 423         * process of removing the last page (in __delete_from_page_cache())
 424         * and we must not free mapping under it.
 425         */
 426        spin_lock_irq(&inode->i_data.tree_lock);
 427        BUG_ON(inode->i_data.nrpages);
 428        spin_unlock_irq(&inode->i_data.tree_lock);
 429        BUG_ON(!list_empty(&inode->i_data.private_list));
 430        BUG_ON(!(inode->i_state & I_FREEING));
 431        BUG_ON(inode->i_state & I_CLEAR);
 432        inode_sync_wait(inode);
 433        /* don't need i_lock here, no concurrent mods to i_state */
 434        inode->i_state = I_FREEING | I_CLEAR;
 435}
 436EXPORT_SYMBOL(end_writeback);
 437
 438/*
 439 * Free the inode passed in, removing it from the lists it is still connected
 440 * to. We remove any pages still attached to the inode and wait for any IO that
 441 * is still in progress before finally destroying the inode.
 442 *
 443 * An inode must already be marked I_FREEING so that we avoid the inode being
 444 * moved back onto lists if we race with other code that manipulates the lists
 445 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 446 *
 447 * An inode must already be removed from the LRU list before being evicted from
 448 * the cache. This should occur atomically with setting the I_FREEING state
 449 * flag, so no inodes here should ever be on the LRU when being evicted.
 450 */
 451static void evict(struct inode *inode)
 452{
 453        const struct super_operations *op = inode->i_sb->s_op;
 454
 455        BUG_ON(!(inode->i_state & I_FREEING));
 456        BUG_ON(!list_empty(&inode->i_lru));
 457
 458        if (!list_empty(&inode->i_wb_list))
 459                inode_wb_list_del(inode);
 460
 461        inode_sb_list_del(inode);
 462
 463        if (op->evict_inode) {
 464                op->evict_inode(inode);
 465        } else {
 466                if (inode->i_data.nrpages)
 467                        truncate_inode_pages(&inode->i_data, 0);
 468                end_writeback(inode);
 469        }
 470        if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 471                bd_forget(inode);
 472        if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 473                cd_forget(inode);
 474
 475        remove_inode_hash(inode);
 476
 477        spin_lock(&inode->i_lock);
 478        wake_up_bit(&inode->i_state, __I_NEW);
 479        BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 480        spin_unlock(&inode->i_lock);
 481
 482        destroy_inode(inode);
 483}
 484
 485/*
 486 * dispose_list - dispose of the contents of a local list
 487 * @head: the head of the list to free
 488 *
 489 * Dispose-list gets a local list with local inodes in it, so it doesn't
 490 * need to worry about list corruption and SMP locks.
 491 */
 492static void dispose_list(struct list_head *head)
 493{
 494        while (!list_empty(head)) {
 495                struct inode *inode;
 496
 497                inode = list_first_entry(head, struct inode, i_lru);
 498                list_del_init(&inode->i_lru);
 499
 500                evict(inode);
 501        }
 502}
 503
 504/**
 505 * evict_inodes - evict all evictable inodes for a superblock
 506 * @sb:         superblock to operate on
 507 *
 508 * Make sure that no inodes with zero refcount are retained.  This is
 509 * called by superblock shutdown after having MS_ACTIVE flag removed,
 510 * so any inode reaching zero refcount during or after that call will
 511 * be immediately evicted.
 512 */
 513void evict_inodes(struct super_block *sb)
 514{
 515        struct inode *inode, *next;
 516        LIST_HEAD(dispose);
 517
 518        spin_lock(&inode_sb_list_lock);
 519        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 520                if (atomic_read(&inode->i_count))
 521                        continue;
 522
 523                spin_lock(&inode->i_lock);
 524                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 525                        spin_unlock(&inode->i_lock);
 526                        continue;
 527                }
 528
 529                inode->i_state |= I_FREEING;
 530                inode_lru_list_del(inode);
 531                spin_unlock(&inode->i_lock);
 532                list_add(&inode->i_lru, &dispose);
 533        }
 534        spin_unlock(&inode_sb_list_lock);
 535
 536        dispose_list(&dispose);
 537}
 538
 539/**
 540 * invalidate_inodes    - attempt to free all inodes on a superblock
 541 * @sb:         superblock to operate on
 542 * @kill_dirty: flag to guide handling of dirty inodes
 543 *
 544 * Attempts to free all inodes for a given superblock.  If there were any
 545 * busy inodes return a non-zero value, else zero.
 546 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 547 * them as busy.
 548 */
 549int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 550{
 551        int busy = 0;
 552        struct inode *inode, *next;
 553        LIST_HEAD(dispose);
 554
 555        spin_lock(&inode_sb_list_lock);
 556        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 557                spin_lock(&inode->i_lock);
 558                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 559                        spin_unlock(&inode->i_lock);
 560                        continue;
 561                }
 562                if (inode->i_state & I_DIRTY && !kill_dirty) {
 563                        spin_unlock(&inode->i_lock);
 564                        busy = 1;
 565                        continue;
 566                }
 567                if (atomic_read(&inode->i_count)) {
 568                        spin_unlock(&inode->i_lock);
 569                        busy = 1;
 570                        continue;
 571                }
 572
 573                inode->i_state |= I_FREEING;
 574                inode_lru_list_del(inode);
 575                spin_unlock(&inode->i_lock);
 576                list_add(&inode->i_lru, &dispose);
 577        }
 578        spin_unlock(&inode_sb_list_lock);
 579
 580        dispose_list(&dispose);
 581
 582        return busy;
 583}
 584
 585static int can_unuse(struct inode *inode)
 586{
 587        if (inode->i_state & ~I_REFERENCED)
 588                return 0;
 589        if (inode_has_buffers(inode))
 590                return 0;
 591        if (atomic_read(&inode->i_count))
 592                return 0;
 593        if (inode->i_data.nrpages)
 594                return 0;
 595        return 1;
 596}
 597
 598/*
 599 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 600 * This is called from the superblock shrinker function with a number of inodes
 601 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 602 * then are freed outside inode_lock by dispose_list().
 603 *
 604 * Any inodes which are pinned purely because of attached pagecache have their
 605 * pagecache removed.  If the inode has metadata buffers attached to
 606 * mapping->private_list then try to remove them.
 607 *
 608 * If the inode has the I_REFERENCED flag set, then it means that it has been
 609 * used recently - the flag is set in iput_final(). When we encounter such an
 610 * inode, clear the flag and move it to the back of the LRU so it gets another
 611 * pass through the LRU before it gets reclaimed. This is necessary because of
 612 * the fact we are doing lazy LRU updates to minimise lock contention so the
 613 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 614 * with this flag set because they are the inodes that are out of order.
 615 */
 616void prune_icache_sb(struct super_block *sb, int nr_to_scan)
 617{
 618        LIST_HEAD(freeable);
 619        int nr_scanned;
 620        unsigned long reap = 0;
 621
 622        spin_lock(&sb->s_inode_lru_lock);
 623        for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
 624                struct inode *inode;
 625
 626                if (list_empty(&sb->s_inode_lru))
 627                        break;
 628
 629                inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
 630
 631                /*
 632                 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
 633                 * so use a trylock. If we fail to get the lock, just move the
 634                 * inode to the back of the list so we don't spin on it.
 635                 */
 636                if (!spin_trylock(&inode->i_lock)) {
 637                        list_move(&inode->i_lru, &sb->s_inode_lru);
 638                        continue;
 639                }
 640
 641                /*
 642                 * Referenced or dirty inodes are still in use. Give them
 643                 * another pass through the LRU as we canot reclaim them now.
 644                 */
 645                if (atomic_read(&inode->i_count) ||
 646                    (inode->i_state & ~I_REFERENCED)) {
 647                        list_del_init(&inode->i_lru);
 648                        spin_unlock(&inode->i_lock);
 649                        sb->s_nr_inodes_unused--;
 650                        this_cpu_dec(nr_unused);
 651                        continue;
 652                }
 653
 654                /* recently referenced inodes get one more pass */
 655                if (inode->i_state & I_REFERENCED) {
 656                        inode->i_state &= ~I_REFERENCED;
 657                        list_move(&inode->i_lru, &sb->s_inode_lru);
 658                        spin_unlock(&inode->i_lock);
 659                        continue;
 660                }
 661                if (inode_has_buffers(inode) || inode->i_data.nrpages) {
 662                        __iget(inode);
 663                        spin_unlock(&inode->i_lock);
 664                        spin_unlock(&sb->s_inode_lru_lock);
 665                        if (remove_inode_buffers(inode))
 666                                reap += invalidate_mapping_pages(&inode->i_data,
 667                                                                0, -1);
 668                        iput(inode);
 669                        spin_lock(&sb->s_inode_lru_lock);
 670
 671                        if (inode != list_entry(sb->s_inode_lru.next,
 672                                                struct inode, i_lru))
 673                                continue;       /* wrong inode or list_empty */
 674                        /* avoid lock inversions with trylock */
 675                        if (!spin_trylock(&inode->i_lock))
 676                                continue;
 677                        if (!can_unuse(inode)) {
 678                                spin_unlock(&inode->i_lock);
 679                                continue;
 680                        }
 681                }
 682                WARN_ON(inode->i_state & I_NEW);
 683                inode->i_state |= I_FREEING;
 684                spin_unlock(&inode->i_lock);
 685
 686                list_move(&inode->i_lru, &freeable);
 687                sb->s_nr_inodes_unused--;
 688                this_cpu_dec(nr_unused);
 689        }
 690        if (current_is_kswapd())
 691                __count_vm_events(KSWAPD_INODESTEAL, reap);
 692        else
 693                __count_vm_events(PGINODESTEAL, reap);
 694        spin_unlock(&sb->s_inode_lru_lock);
 695
 696        dispose_list(&freeable);
 697}
 698
 699static void __wait_on_freeing_inode(struct inode *inode);
 700/*
 701 * Called with the inode lock held.
 702 */
 703static struct inode *find_inode(struct super_block *sb,
 704                                struct hlist_head *head,
 705                                int (*test)(struct inode *, void *),
 706                                void *data)
 707{
 708        struct hlist_node *node;
 709        struct inode *inode = NULL;
 710
 711repeat:
 712        hlist_for_each_entry(inode, node, head, i_hash) {
 713                spin_lock(&inode->i_lock);
 714                if (inode->i_sb != sb) {
 715                        spin_unlock(&inode->i_lock);
 716                        continue;
 717                }
 718                if (!test(inode, data)) {
 719                        spin_unlock(&inode->i_lock);
 720                        continue;
 721                }
 722                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 723                        __wait_on_freeing_inode(inode);
 724                        goto repeat;
 725                }
 726                __iget(inode);
 727                spin_unlock(&inode->i_lock);
 728                return inode;
 729        }
 730        return NULL;
 731}
 732
 733/*
 734 * find_inode_fast is the fast path version of find_inode, see the comment at
 735 * iget_locked for details.
 736 */
 737static struct inode *find_inode_fast(struct super_block *sb,
 738                                struct hlist_head *head, unsigned long ino)
 739{
 740        struct hlist_node *node;
 741        struct inode *inode = NULL;
 742
 743repeat:
 744        hlist_for_each_entry(inode, node, head, i_hash) {
 745                spin_lock(&inode->i_lock);
 746                if (inode->i_ino != ino) {
 747                        spin_unlock(&inode->i_lock);
 748                        continue;
 749                }
 750                if (inode->i_sb != sb) {
 751                        spin_unlock(&inode->i_lock);
 752                        continue;
 753                }
 754                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 755                        __wait_on_freeing_inode(inode);
 756                        goto repeat;
 757                }
 758                __iget(inode);
 759                spin_unlock(&inode->i_lock);
 760                return inode;
 761        }
 762        return NULL;
 763}
 764
 765/*
 766 * Each cpu owns a range of LAST_INO_BATCH numbers.
 767 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 768 * to renew the exhausted range.
 769 *
 770 * This does not significantly increase overflow rate because every CPU can
 771 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 772 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 773 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 774 * overflow rate by 2x, which does not seem too significant.
 775 *
 776 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 777 * error if st_ino won't fit in target struct field. Use 32bit counter
 778 * here to attempt to avoid that.
 779 */
 780#define LAST_INO_BATCH 1024
 781static DEFINE_PER_CPU(unsigned int, last_ino);
 782
 783unsigned int get_next_ino(void)
 784{
 785        unsigned int *p = &get_cpu_var(last_ino);
 786        unsigned int res = *p;
 787
 788#ifdef CONFIG_SMP
 789        if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 790                static atomic_t shared_last_ino;
 791                int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 792
 793                res = next - LAST_INO_BATCH;
 794        }
 795#endif
 796
 797        *p = ++res;
 798        put_cpu_var(last_ino);
 799        return res;
 800}
 801EXPORT_SYMBOL(get_next_ino);
 802
 803/**
 804 *      new_inode_pseudo        - obtain an inode
 805 *      @sb: superblock
 806 *
 807 *      Allocates a new inode for given superblock.
 808 *      Inode wont be chained in superblock s_inodes list
 809 *      This means :
 810 *      - fs can't be unmount
 811 *      - quotas, fsnotify, writeback can't work
 812 */
 813struct inode *new_inode_pseudo(struct super_block *sb)
 814{
 815        struct inode *inode = alloc_inode(sb);
 816
 817        if (inode) {
 818                spin_lock(&inode->i_lock);
 819                inode->i_state = 0;
 820                spin_unlock(&inode->i_lock);
 821                INIT_LIST_HEAD(&inode->i_sb_list);
 822        }
 823        return inode;
 824}
 825
 826/**
 827 *      new_inode       - obtain an inode
 828 *      @sb: superblock
 829 *
 830 *      Allocates a new inode for given superblock. The default gfp_mask
 831 *      for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
 832 *      If HIGHMEM pages are unsuitable or it is known that pages allocated
 833 *      for the page cache are not reclaimable or migratable,
 834 *      mapping_set_gfp_mask() must be called with suitable flags on the
 835 *      newly created inode's mapping
 836 *
 837 */
 838struct inode *new_inode(struct super_block *sb)
 839{
 840        struct inode *inode;
 841
 842        spin_lock_prefetch(&inode_sb_list_lock);
 843
 844        inode = new_inode_pseudo(sb);
 845        if (inode)
 846                inode_sb_list_add(inode);
 847        return inode;
 848}
 849EXPORT_SYMBOL(new_inode);
 850
 851#ifdef CONFIG_DEBUG_LOCK_ALLOC
 852void lockdep_annotate_inode_mutex_key(struct inode *inode)
 853{
 854        if (S_ISDIR(inode->i_mode)) {
 855                struct file_system_type *type = inode->i_sb->s_type;
 856
 857                /* Set new key only if filesystem hasn't already changed it */
 858                if (!lockdep_match_class(&inode->i_mutex,
 859                    &type->i_mutex_key)) {
 860                        /*
 861                         * ensure nobody is actually holding i_mutex
 862                         */
 863                        mutex_destroy(&inode->i_mutex);
 864                        mutex_init(&inode->i_mutex);
 865                        lockdep_set_class(&inode->i_mutex,
 866                                          &type->i_mutex_dir_key);
 867                }
 868        }
 869}
 870EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
 871#endif
 872
 873/**
 874 * unlock_new_inode - clear the I_NEW state and wake up any waiters
 875 * @inode:      new inode to unlock
 876 *
 877 * Called when the inode is fully initialised to clear the new state of the
 878 * inode and wake up anyone waiting for the inode to finish initialisation.
 879 */
 880void unlock_new_inode(struct inode *inode)
 881{
 882        lockdep_annotate_inode_mutex_key(inode);
 883        spin_lock(&inode->i_lock);
 884        WARN_ON(!(inode->i_state & I_NEW));
 885        inode->i_state &= ~I_NEW;
 886        wake_up_bit(&inode->i_state, __I_NEW);
 887        spin_unlock(&inode->i_lock);
 888}
 889EXPORT_SYMBOL(unlock_new_inode);
 890
 891/**
 892 * iget5_locked - obtain an inode from a mounted file system
 893 * @sb:         super block of file system
 894 * @hashval:    hash value (usually inode number) to get
 895 * @test:       callback used for comparisons between inodes
 896 * @set:        callback used to initialize a new struct inode
 897 * @data:       opaque data pointer to pass to @test and @set
 898 *
 899 * Search for the inode specified by @hashval and @data in the inode cache,
 900 * and if present it is return it with an increased reference count. This is
 901 * a generalized version of iget_locked() for file systems where the inode
 902 * number is not sufficient for unique identification of an inode.
 903 *
 904 * If the inode is not in cache, allocate a new inode and return it locked,
 905 * hashed, and with the I_NEW flag set. The file system gets to fill it in
 906 * before unlocking it via unlock_new_inode().
 907 *
 908 * Note both @test and @set are called with the inode_hash_lock held, so can't
 909 * sleep.
 910 */
 911struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
 912                int (*test)(struct inode *, void *),
 913                int (*set)(struct inode *, void *), void *data)
 914{
 915        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
 916        struct inode *inode;
 917
 918        spin_lock(&inode_hash_lock);
 919        inode = find_inode(sb, head, test, data);
 920        spin_unlock(&inode_hash_lock);
 921
 922        if (inode) {
 923                wait_on_inode(inode);
 924                return inode;
 925        }
 926
 927        inode = alloc_inode(sb);
 928        if (inode) {
 929                struct inode *old;
 930
 931                spin_lock(&inode_hash_lock);
 932                /* We released the lock, so.. */
 933                old = find_inode(sb, head, test, data);
 934                if (!old) {
 935                        if (set(inode, data))
 936                                goto set_failed;
 937
 938                        spin_lock(&inode->i_lock);
 939                        inode->i_state = I_NEW;
 940                        hlist_add_head(&inode->i_hash, head);
 941                        spin_unlock(&inode->i_lock);
 942                        inode_sb_list_add(inode);
 943                        spin_unlock(&inode_hash_lock);
 944
 945                        /* Return the locked inode with I_NEW set, the
 946                         * caller is responsible for filling in the contents
 947                         */
 948                        return inode;
 949                }
 950
 951                /*
 952                 * Uhhuh, somebody else created the same inode under
 953                 * us. Use the old inode instead of the one we just
 954                 * allocated.
 955                 */
 956                spin_unlock(&inode_hash_lock);
 957                destroy_inode(inode);
 958                inode = old;
 959                wait_on_inode(inode);
 960        }
 961        return inode;
 962
 963set_failed:
 964        spin_unlock(&inode_hash_lock);
 965        destroy_inode(inode);
 966        return NULL;
 967}
 968EXPORT_SYMBOL(iget5_locked);
 969
 970/**
 971 * iget_locked - obtain an inode from a mounted file system
 972 * @sb:         super block of file system
 973 * @ino:        inode number to get
 974 *
 975 * Search for the inode specified by @ino in the inode cache and if present
 976 * return it with an increased reference count. This is for file systems
 977 * where the inode number is sufficient for unique identification of an inode.
 978 *
 979 * If the inode is not in cache, allocate a new inode and return it locked,
 980 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
 981 * before unlocking it via unlock_new_inode().
 982 */
 983struct inode *iget_locked(struct super_block *sb, unsigned long ino)
 984{
 985        struct hlist_head *head = inode_hashtable + hash(sb, ino);
 986        struct inode *inode;
 987
 988        spin_lock(&inode_hash_lock);
 989        inode = find_inode_fast(sb, head, ino);
 990        spin_unlock(&inode_hash_lock);
 991        if (inode) {
 992                wait_on_inode(inode);
 993                return inode;
 994        }
 995
 996        inode = alloc_inode(sb);
 997        if (inode) {
 998                struct inode *old;
 999
1000                spin_lock(&inode_hash_lock);
1001                /* We released the lock, so.. */
1002                old = find_inode_fast(sb, head, ino);
1003                if (!old) {
1004                        inode->i_ino = ino;
1005                        spin_lock(&inode->i_lock);
1006                        inode->i_state = I_NEW;
1007                        hlist_add_head(&inode->i_hash, head);
1008                        spin_unlock(&inode->i_lock);
1009                        inode_sb_list_add(inode);
1010                        spin_unlock(&inode_hash_lock);
1011
1012                        /* Return the locked inode with I_NEW set, the
1013                         * caller is responsible for filling in the contents
1014                         */
1015                        return inode;
1016                }
1017
1018                /*
1019                 * Uhhuh, somebody else created the same inode under
1020                 * us. Use the old inode instead of the one we just
1021                 * allocated.
1022                 */
1023                spin_unlock(&inode_hash_lock);
1024                destroy_inode(inode);
1025                inode = old;
1026                wait_on_inode(inode);
1027        }
1028        return inode;
1029}
1030EXPORT_SYMBOL(iget_locked);
1031
1032/*
1033 * search the inode cache for a matching inode number.
1034 * If we find one, then the inode number we are trying to
1035 * allocate is not unique and so we should not use it.
1036 *
1037 * Returns 1 if the inode number is unique, 0 if it is not.
1038 */
1039static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1040{
1041        struct hlist_head *b = inode_hashtable + hash(sb, ino);
1042        struct hlist_node *node;
1043        struct inode *inode;
1044
1045        spin_lock(&inode_hash_lock);
1046        hlist_for_each_entry(inode, node, b, i_hash) {
1047                if (inode->i_ino == ino && inode->i_sb == sb) {
1048                        spin_unlock(&inode_hash_lock);
1049                        return 0;
1050                }
1051        }
1052        spin_unlock(&inode_hash_lock);
1053
1054        return 1;
1055}
1056
1057/**
1058 *      iunique - get a unique inode number
1059 *      @sb: superblock
1060 *      @max_reserved: highest reserved inode number
1061 *
1062 *      Obtain an inode number that is unique on the system for a given
1063 *      superblock. This is used by file systems that have no natural
1064 *      permanent inode numbering system. An inode number is returned that
1065 *      is higher than the reserved limit but unique.
1066 *
1067 *      BUGS:
1068 *      With a large number of inodes live on the file system this function
1069 *      currently becomes quite slow.
1070 */
1071ino_t iunique(struct super_block *sb, ino_t max_reserved)
1072{
1073        /*
1074         * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1075         * error if st_ino won't fit in target struct field. Use 32bit counter
1076         * here to attempt to avoid that.
1077         */
1078        static DEFINE_SPINLOCK(iunique_lock);
1079        static unsigned int counter;
1080        ino_t res;
1081
1082        spin_lock(&iunique_lock);
1083        do {
1084                if (counter <= max_reserved)
1085                        counter = max_reserved + 1;
1086                res = counter++;
1087        } while (!test_inode_iunique(sb, res));
1088        spin_unlock(&iunique_lock);
1089
1090        return res;
1091}
1092EXPORT_SYMBOL(iunique);
1093
1094struct inode *igrab(struct inode *inode)
1095{
1096        spin_lock(&inode->i_lock);
1097        if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1098                __iget(inode);
1099                spin_unlock(&inode->i_lock);
1100        } else {
1101                spin_unlock(&inode->i_lock);
1102                /*
1103                 * Handle the case where s_op->clear_inode is not been
1104                 * called yet, and somebody is calling igrab
1105                 * while the inode is getting freed.
1106                 */
1107                inode = NULL;
1108        }
1109        return inode;
1110}
1111EXPORT_SYMBOL(igrab);
1112
1113/**
1114 * ilookup5_nowait - search for an inode in the inode cache
1115 * @sb:         super block of file system to search
1116 * @hashval:    hash value (usually inode number) to search for
1117 * @test:       callback used for comparisons between inodes
1118 * @data:       opaque data pointer to pass to @test
1119 *
1120 * Search for the inode specified by @hashval and @data in the inode cache.
1121 * If the inode is in the cache, the inode is returned with an incremented
1122 * reference count.
1123 *
1124 * Note: I_NEW is not waited upon so you have to be very careful what you do
1125 * with the returned inode.  You probably should be using ilookup5() instead.
1126 *
1127 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1128 */
1129struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1130                int (*test)(struct inode *, void *), void *data)
1131{
1132        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1133        struct inode *inode;
1134
1135        spin_lock(&inode_hash_lock);
1136        inode = find_inode(sb, head, test, data);
1137        spin_unlock(&inode_hash_lock);
1138
1139        return inode;
1140}
1141EXPORT_SYMBOL(ilookup5_nowait);
1142
1143/**
1144 * ilookup5 - search for an inode in the inode cache
1145 * @sb:         super block of file system to search
1146 * @hashval:    hash value (usually inode number) to search for
1147 * @test:       callback used for comparisons between inodes
1148 * @data:       opaque data pointer to pass to @test
1149 *
1150 * Search for the inode specified by @hashval and @data in the inode cache,
1151 * and if the inode is in the cache, return the inode with an incremented
1152 * reference count.  Waits on I_NEW before returning the inode.
1153 * returned with an incremented reference count.
1154 *
1155 * This is a generalized version of ilookup() for file systems where the
1156 * inode number is not sufficient for unique identification of an inode.
1157 *
1158 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1159 */
1160struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1161                int (*test)(struct inode *, void *), void *data)
1162{
1163        struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1164
1165        if (inode)
1166                wait_on_inode(inode);
1167        return inode;
1168}
1169EXPORT_SYMBOL(ilookup5);
1170
1171/**
1172 * ilookup - search for an inode in the inode cache
1173 * @sb:         super block of file system to search
1174 * @ino:        inode number to search for
1175 *
1176 * Search for the inode @ino in the inode cache, and if the inode is in the
1177 * cache, the inode is returned with an incremented reference count.
1178 */
1179struct inode *ilookup(struct super_block *sb, unsigned long ino)
1180{
1181        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1182        struct inode *inode;
1183
1184        spin_lock(&inode_hash_lock);
1185        inode = find_inode_fast(sb, head, ino);
1186        spin_unlock(&inode_hash_lock);
1187
1188        if (inode)
1189                wait_on_inode(inode);
1190        return inode;
1191}
1192EXPORT_SYMBOL(ilookup);
1193
1194int insert_inode_locked(struct inode *inode)
1195{
1196        struct super_block *sb = inode->i_sb;
1197        ino_t ino = inode->i_ino;
1198        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1199
1200        while (1) {
1201                struct hlist_node *node;
1202                struct inode *old = NULL;
1203                spin_lock(&inode_hash_lock);
1204                hlist_for_each_entry(old, node, head, i_hash) {
1205                        if (old->i_ino != ino)
1206                                continue;
1207                        if (old->i_sb != sb)
1208                                continue;
1209                        spin_lock(&old->i_lock);
1210                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1211                                spin_unlock(&old->i_lock);
1212                                continue;
1213                        }
1214                        break;
1215                }
1216                if (likely(!node)) {
1217                        spin_lock(&inode->i_lock);
1218                        inode->i_state |= I_NEW;
1219                        hlist_add_head(&inode->i_hash, head);
1220                        spin_unlock(&inode->i_lock);
1221                        spin_unlock(&inode_hash_lock);
1222                        return 0;
1223                }
1224                __iget(old);
1225                spin_unlock(&old->i_lock);
1226                spin_unlock(&inode_hash_lock);
1227                wait_on_inode(old);
1228                if (unlikely(!inode_unhashed(old))) {
1229                        iput(old);
1230                        return -EBUSY;
1231                }
1232                iput(old);
1233        }
1234}
1235EXPORT_SYMBOL(insert_inode_locked);
1236
1237int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1238                int (*test)(struct inode *, void *), void *data)
1239{
1240        struct super_block *sb = inode->i_sb;
1241        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1242
1243        while (1) {
1244                struct hlist_node *node;
1245                struct inode *old = NULL;
1246
1247                spin_lock(&inode_hash_lock);
1248                hlist_for_each_entry(old, node, head, i_hash) {
1249                        if (old->i_sb != sb)
1250                                continue;
1251                        if (!test(old, data))
1252                                continue;
1253                        spin_lock(&old->i_lock);
1254                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1255                                spin_unlock(&old->i_lock);
1256                                continue;
1257                        }
1258                        break;
1259                }
1260                if (likely(!node)) {
1261                        spin_lock(&inode->i_lock);
1262                        inode->i_state |= I_NEW;
1263                        hlist_add_head(&inode->i_hash, head);
1264                        spin_unlock(&inode->i_lock);
1265                        spin_unlock(&inode_hash_lock);
1266                        return 0;
1267                }
1268                __iget(old);
1269                spin_unlock(&old->i_lock);
1270                spin_unlock(&inode_hash_lock);
1271                wait_on_inode(old);
1272                if (unlikely(!inode_unhashed(old))) {
1273                        iput(old);
1274                        return -EBUSY;
1275                }
1276                iput(old);
1277        }
1278}
1279EXPORT_SYMBOL(insert_inode_locked4);
1280
1281
1282int generic_delete_inode(struct inode *inode)
1283{
1284        return 1;
1285}
1286EXPORT_SYMBOL(generic_delete_inode);
1287
1288/*
1289 * Normal UNIX filesystem behaviour: delete the
1290 * inode when the usage count drops to zero, and
1291 * i_nlink is zero.
1292 */
1293int generic_drop_inode(struct inode *inode)
1294{
1295        return !inode->i_nlink || inode_unhashed(inode);
1296}
1297EXPORT_SYMBOL_GPL(generic_drop_inode);
1298
1299/*
1300 * Called when we're dropping the last reference
1301 * to an inode.
1302 *
1303 * Call the FS "drop_inode()" function, defaulting to
1304 * the legacy UNIX filesystem behaviour.  If it tells
1305 * us to evict inode, do so.  Otherwise, retain inode
1306 * in cache if fs is alive, sync and evict if fs is
1307 * shutting down.
1308 */
1309static void iput_final(struct inode *inode)
1310{
1311        struct super_block *sb = inode->i_sb;
1312        const struct super_operations *op = inode->i_sb->s_op;
1313        int drop;
1314
1315        WARN_ON(inode->i_state & I_NEW);
1316
1317        if (op->drop_inode)
1318                drop = op->drop_inode(inode);
1319        else
1320                drop = generic_drop_inode(inode);
1321
1322        if (!drop && (sb->s_flags & MS_ACTIVE)) {
1323                inode->i_state |= I_REFERENCED;
1324                if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1325                        inode_lru_list_add(inode);
1326                spin_unlock(&inode->i_lock);
1327                return;
1328        }
1329
1330        if (!drop) {
1331                inode->i_state |= I_WILL_FREE;
1332                spin_unlock(&inode->i_lock);
1333                write_inode_now(inode, 1);
1334                spin_lock(&inode->i_lock);
1335                WARN_ON(inode->i_state & I_NEW);
1336                inode->i_state &= ~I_WILL_FREE;
1337        }
1338
1339        inode->i_state |= I_FREEING;
1340        if (!list_empty(&inode->i_lru))
1341                inode_lru_list_del(inode);
1342        spin_unlock(&inode->i_lock);
1343
1344        evict(inode);
1345}
1346
1347/**
1348 *      iput    - put an inode
1349 *      @inode: inode to put
1350 *
1351 *      Puts an inode, dropping its usage count. If the inode use count hits
1352 *      zero, the inode is then freed and may also be destroyed.
1353 *
1354 *      Consequently, iput() can sleep.
1355 */
1356void iput(struct inode *inode)
1357{
1358        if (inode) {
1359                BUG_ON(inode->i_state & I_CLEAR);
1360
1361                if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1362                        iput_final(inode);
1363        }
1364}
1365EXPORT_SYMBOL(iput);
1366
1367/**
1368 *      bmap    - find a block number in a file
1369 *      @inode: inode of file
1370 *      @block: block to find
1371 *
1372 *      Returns the block number on the device holding the inode that
1373 *      is the disk block number for the block of the file requested.
1374 *      That is, asked for block 4 of inode 1 the function will return the
1375 *      disk block relative to the disk start that holds that block of the
1376 *      file.
1377 */
1378sector_t bmap(struct inode *inode, sector_t block)
1379{
1380        sector_t res = 0;
1381        if (inode->i_mapping->a_ops->bmap)
1382                res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1383        return res;
1384}
1385EXPORT_SYMBOL(bmap);
1386
1387/*
1388 * With relative atime, only update atime if the previous atime is
1389 * earlier than either the ctime or mtime or if at least a day has
1390 * passed since the last atime update.
1391 */
1392static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1393                             struct timespec now)
1394{
1395
1396        if (!(mnt->mnt_flags & MNT_RELATIME))
1397                return 1;
1398        /*
1399         * Is mtime younger than atime? If yes, update atime:
1400         */
1401        if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1402                return 1;
1403        /*
1404         * Is ctime younger than atime? If yes, update atime:
1405         */
1406        if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1407                return 1;
1408
1409        /*
1410         * Is the previous atime value older than a day? If yes,
1411         * update atime:
1412         */
1413        if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1414                return 1;
1415        /*
1416         * Good, we can skip the atime update:
1417         */
1418        return 0;
1419}
1420
1421/**
1422 *      touch_atime     -       update the access time
1423 *      @mnt: mount the inode is accessed on
1424 *      @dentry: dentry accessed
1425 *
1426 *      Update the accessed time on an inode and mark it for writeback.
1427 *      This function automatically handles read only file systems and media,
1428 *      as well as the "noatime" flag and inode specific "noatime" markers.
1429 */
1430void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1431{
1432        struct inode *inode = dentry->d_inode;
1433        struct timespec now;
1434
1435        if (inode->i_flags & S_NOATIME)
1436                return;
1437        if (IS_NOATIME(inode))
1438                return;
1439        if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1440                return;
1441
1442        if (mnt->mnt_flags & MNT_NOATIME)
1443                return;
1444        if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1445                return;
1446
1447        now = current_fs_time(inode->i_sb);
1448
1449        if (!relatime_need_update(mnt, inode, now))
1450                return;
1451
1452        if (timespec_equal(&inode->i_atime, &now))
1453                return;
1454
1455        if (mnt_want_write(mnt))
1456                return;
1457
1458        inode->i_atime = now;
1459        mark_inode_dirty_sync(inode);
1460        mnt_drop_write(mnt);
1461}
1462EXPORT_SYMBOL(touch_atime);
1463
1464/**
1465 *      file_update_time        -       update mtime and ctime time
1466 *      @file: file accessed
1467 *
1468 *      Update the mtime and ctime members of an inode and mark the inode
1469 *      for writeback.  Note that this function is meant exclusively for
1470 *      usage in the file write path of filesystems, and filesystems may
1471 *      choose to explicitly ignore update via this function with the
1472 *      S_NOCMTIME inode flag, e.g. for network filesystem where these
1473 *      timestamps are handled by the server.
1474 */
1475
1476void file_update_time(struct file *file)
1477{
1478        struct inode *inode = file->f_path.dentry->d_inode;
1479        struct timespec now;
1480        enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1481
1482        /* First try to exhaust all avenues to not sync */
1483        if (IS_NOCMTIME(inode))
1484                return;
1485
1486        now = current_fs_time(inode->i_sb);
1487        if (!timespec_equal(&inode->i_mtime, &now))
1488                sync_it = S_MTIME;
1489
1490        if (!timespec_equal(&inode->i_ctime, &now))
1491                sync_it |= S_CTIME;
1492
1493        if (IS_I_VERSION(inode))
1494                sync_it |= S_VERSION;
1495
1496        if (!sync_it)
1497                return;
1498
1499        /* Finally allowed to write? Takes lock. */
1500        if (mnt_want_write_file(file))
1501                return;
1502
1503        /* Only change inode inside the lock region */
1504        if (sync_it & S_VERSION)
1505                inode_inc_iversion(inode);
1506        if (sync_it & S_CTIME)
1507                inode->i_ctime = now;
1508        if (sync_it & S_MTIME)
1509                inode->i_mtime = now;
1510        mark_inode_dirty_sync(inode);
1511        mnt_drop_write(file->f_path.mnt);
1512}
1513EXPORT_SYMBOL(file_update_time);
1514
1515int inode_needs_sync(struct inode *inode)
1516{
1517        if (IS_SYNC(inode))
1518                return 1;
1519        if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1520                return 1;
1521        return 0;
1522}
1523EXPORT_SYMBOL(inode_needs_sync);
1524
1525int inode_wait(void *word)
1526{
1527        schedule();
1528        return 0;
1529}
1530EXPORT_SYMBOL(inode_wait);
1531
1532/*
1533 * If we try to find an inode in the inode hash while it is being
1534 * deleted, we have to wait until the filesystem completes its
1535 * deletion before reporting that it isn't found.  This function waits
1536 * until the deletion _might_ have completed.  Callers are responsible
1537 * to recheck inode state.
1538 *
1539 * It doesn't matter if I_NEW is not set initially, a call to
1540 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1541 * will DTRT.
1542 */
1543static void __wait_on_freeing_inode(struct inode *inode)
1544{
1545        wait_queue_head_t *wq;
1546        DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1547        wq = bit_waitqueue(&inode->i_state, __I_NEW);
1548        prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1549        spin_unlock(&inode->i_lock);
1550        spin_unlock(&inode_hash_lock);
1551        schedule();
1552        finish_wait(wq, &wait.wait);
1553        spin_lock(&inode_hash_lock);
1554}
1555
1556static __initdata unsigned long ihash_entries;
1557static int __init set_ihash_entries(char *str)
1558{
1559        if (!str)
1560                return 0;
1561        ihash_entries = simple_strtoul(str, &str, 0);
1562        return 1;
1563}
1564__setup("ihash_entries=", set_ihash_entries);
1565
1566/*
1567 * Initialize the waitqueues and inode hash table.
1568 */
1569void __init inode_init_early(void)
1570{
1571        int loop;
1572
1573        /* If hashes are distributed across NUMA nodes, defer
1574         * hash allocation until vmalloc space is available.
1575         */
1576        if (hashdist)
1577                return;
1578
1579        inode_hashtable =
1580                alloc_large_system_hash("Inode-cache",
1581                                        sizeof(struct hlist_head),
1582                                        ihash_entries,
1583                                        14,
1584                                        HASH_EARLY,
1585                                        &i_hash_shift,
1586                                        &i_hash_mask,
1587                                        0);
1588
1589        for (loop = 0; loop < (1 << i_hash_shift); loop++)
1590                INIT_HLIST_HEAD(&inode_hashtable[loop]);
1591}
1592
1593void __init inode_init(void)
1594{
1595        int loop;
1596
1597        /* inode slab cache */
1598        inode_cachep = kmem_cache_create("inode_cache",
1599                                         sizeof(struct inode),
1600                                         0,
1601                                         (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1602                                         SLAB_MEM_SPREAD),
1603                                         init_once);
1604
1605        /* Hash may have been set up in inode_init_early */
1606        if (!hashdist)
1607                return;
1608
1609        inode_hashtable =
1610                alloc_large_system_hash("Inode-cache",
1611                                        sizeof(struct hlist_head),
1612                                        ihash_entries,
1613                                        14,
1614                                        0,
1615                                        &i_hash_shift,
1616                                        &i_hash_mask,
1617                                        0);
1618
1619        for (loop = 0; loop < (1 << i_hash_shift); loop++)
1620                INIT_HLIST_HEAD(&inode_hashtable[loop]);
1621}
1622
1623void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1624{
1625        inode->i_mode = mode;
1626        if (S_ISCHR(mode)) {
1627                inode->i_fop = &def_chr_fops;
1628                inode->i_rdev = rdev;
1629        } else if (S_ISBLK(mode)) {
1630                inode->i_fop = &def_blk_fops;
1631                inode->i_rdev = rdev;
1632        } else if (S_ISFIFO(mode))
1633                inode->i_fop = &def_fifo_fops;
1634        else if (S_ISSOCK(mode))
1635                inode->i_fop = &bad_sock_fops;
1636        else
1637                printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1638                                  " inode %s:%lu\n", mode, inode->i_sb->s_id,
1639                                  inode->i_ino);
1640}
1641EXPORT_SYMBOL(init_special_inode);
1642
1643/**
1644 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1645 * @inode: New inode
1646 * @dir: Directory inode
1647 * @mode: mode of the new inode
1648 */
1649void inode_init_owner(struct inode *inode, const struct inode *dir,
1650                        mode_t mode)
1651{
1652        inode->i_uid = current_fsuid();
1653        if (dir && dir->i_mode & S_ISGID) {
1654                inode->i_gid = dir->i_gid;
1655                if (S_ISDIR(mode))
1656                        mode |= S_ISGID;
1657        } else
1658                inode->i_gid = current_fsgid();
1659        inode->i_mode = mode;
1660}
1661EXPORT_SYMBOL(inode_init_owner);
1662
1663/**
1664 * inode_owner_or_capable - check current task permissions to inode
1665 * @inode: inode being checked
1666 *
1667 * Return true if current either has CAP_FOWNER to the inode, or
1668 * owns the file.
1669 */
1670bool inode_owner_or_capable(const struct inode *inode)
1671{
1672        struct user_namespace *ns = inode_userns(inode);
1673
1674        if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1675                return true;
1676        if (ns_capable(ns, CAP_FOWNER))
1677                return true;
1678        return false;
1679}
1680EXPORT_SYMBOL(inode_owner_or_capable);
1681
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.