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/export.h>
   6#include <linux/fs.h>
   7#include <linux/mm.h>
   8#include <linux/backing-dev.h>
   9#include <linux/hash.h>
  10#include <linux/swap.h>
  11#include <linux/security.h>
  12#include <linux/cdev.h>
  13#include <linux/bootmem.h>
  14#include <linux/fsnotify.h>
  15#include <linux/mount.h>
  16#include <linux/posix_acl.h>
  17#include <linux/prefetch.h>
  18#include <linux/buffer_head.h> /* for inode_has_buffers */
  19#include <linux/ratelimit.h>
  20#include <linux/list_lru.h>
  21#include "internal.h"
  22
  23/*
  24 * Inode locking rules:
  25 *
  26 * inode->i_lock protects:
  27 *   inode->i_state, inode->i_hash, __iget()
  28 * Inode LRU list locks protect:
  29 *   inode->i_sb->s_inode_lru, inode->i_lru
  30 * inode_sb_list_lock protects:
  31 *   sb->s_inodes, inode->i_sb_list
  32 * bdi->wb.list_lock protects:
  33 *   bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
  34 * inode_hash_lock protects:
  35 *   inode_hashtable, inode->i_hash
  36 *
  37 * Lock ordering:
  38 *
  39 * inode_sb_list_lock
  40 *   inode->i_lock
  41 *     Inode LRU list locks
  42 *
  43 * bdi->wb.list_lock
  44 *   inode->i_lock
  45 *
  46 * inode_hash_lock
  47 *   inode_sb_list_lock
  48 *   inode->i_lock
  49 *
  50 * iunique_lock
  51 *   inode_hash_lock
  52 */
  53
  54static unsigned int i_hash_mask __read_mostly;
  55static unsigned int i_hash_shift __read_mostly;
  56static struct hlist_head *inode_hashtable __read_mostly;
  57static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  58
  59__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
  60
  61/*
  62 * Empty aops. Can be used for the cases where the user does not
  63 * define any of the address_space operations.
  64 */
  65const struct address_space_operations empty_aops = {
  66};
  67EXPORT_SYMBOL(empty_aops);
  68
  69/*
  70 * Statistics gathering..
  71 */
  72struct inodes_stat_t inodes_stat;
  73
  74static DEFINE_PER_CPU(unsigned long, nr_inodes);
  75static DEFINE_PER_CPU(unsigned long, nr_unused);
  76
  77static struct kmem_cache *inode_cachep __read_mostly;
  78
  79static long get_nr_inodes(void)
  80{
  81        int i;
  82        long sum = 0;
  83        for_each_possible_cpu(i)
  84                sum += per_cpu(nr_inodes, i);
  85        return sum < 0 ? 0 : sum;
  86}
  87
  88static inline long get_nr_inodes_unused(void)
  89{
  90        int i;
  91        long sum = 0;
  92        for_each_possible_cpu(i)
  93                sum += per_cpu(nr_unused, i);
  94        return sum < 0 ? 0 : sum;
  95}
  96
  97long get_nr_dirty_inodes(void)
  98{
  99        /* not actually dirty inodes, but a wild approximation */
 100        long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
 101        return nr_dirty > 0 ? nr_dirty : 0;
 102}
 103
 104/*
 105 * Handle nr_inode sysctl
 106 */
 107#ifdef CONFIG_SYSCTL
 108int proc_nr_inodes(ctl_table *table, int write,
 109                   void __user *buffer, size_t *lenp, loff_t *ppos)
 110{
 111        inodes_stat.nr_inodes = get_nr_inodes();
 112        inodes_stat.nr_unused = get_nr_inodes_unused();
 113        return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 114}
 115#endif
 116
 117/**
 118 * inode_init_always - perform inode structure intialisation
 119 * @sb: superblock inode belongs to
 120 * @inode: inode to initialise
 121 *
 122 * These are initializations that need to be done on every inode
 123 * allocation as the fields are not initialised by slab allocation.
 124 */
 125int inode_init_always(struct super_block *sb, struct inode *inode)
 126{
 127        static const struct inode_operations empty_iops;
 128        static const struct file_operations empty_fops;
 129        struct address_space *const mapping = &inode->i_data;
 130
 131        inode->i_sb = sb;
 132        inode->i_blkbits = sb->s_blocksize_bits;
 133        inode->i_flags = 0;
 134        atomic_set(&inode->i_count, 1);
 135        inode->i_op = &empty_iops;
 136        inode->i_fop = &empty_fops;
 137        inode->__i_nlink = 1;
 138        inode->i_opflags = 0;
 139        i_uid_write(inode, 0);
 140        i_gid_write(inode, 0);
 141        atomic_set(&inode->i_writecount, 0);
 142        inode->i_size = 0;
 143        inode->i_blocks = 0;
 144        inode->i_bytes = 0;
 145        inode->i_generation = 0;
 146#ifdef CONFIG_QUOTA
 147        memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
 148#endif
 149        inode->i_pipe = NULL;
 150        inode->i_bdev = NULL;
 151        inode->i_cdev = NULL;
 152        inode->i_rdev = 0;
 153        inode->dirtied_when = 0;
 154
 155        if (security_inode_alloc(inode))
 156                goto out;
 157        spin_lock_init(&inode->i_lock);
 158        lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 159
 160        mutex_init(&inode->i_mutex);
 161        lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
 162
 163        atomic_set(&inode->i_dio_count, 0);
 164
 165        mapping->a_ops = &empty_aops;
 166        mapping->host = inode;
 167        mapping->flags = 0;
 168        mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 169        mapping->private_data = NULL;
 170        mapping->backing_dev_info = &default_backing_dev_info;
 171        mapping->writeback_index = 0;
 172
 173        /*
 174         * If the block_device provides a backing_dev_info for client
 175         * inodes then use that.  Otherwise the inode share the bdev's
 176         * backing_dev_info.
 177         */
 178        if (sb->s_bdev) {
 179                struct backing_dev_info *bdi;
 180
 181                bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
 182                mapping->backing_dev_info = bdi;
 183        }
 184        inode->i_private = NULL;
 185        inode->i_mapping = mapping;
 186        INIT_HLIST_HEAD(&inode->i_dentry);      /* buggered by rcu freeing */
 187#ifdef CONFIG_FS_POSIX_ACL
 188        inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 189#endif
 190
 191#ifdef CONFIG_FSNOTIFY
 192        inode->i_fsnotify_mask = 0;
 193#endif
 194
 195        this_cpu_inc(nr_inodes);
 196
 197        return 0;
 198out:
 199        return -ENOMEM;
 200}
 201EXPORT_SYMBOL(inode_init_always);
 202
 203static struct inode *alloc_inode(struct super_block *sb)
 204{
 205        struct inode *inode;
 206
 207        if (sb->s_op->alloc_inode)
 208                inode = sb->s_op->alloc_inode(sb);
 209        else
 210                inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
 211
 212        if (!inode)
 213                return NULL;
 214
 215        if (unlikely(inode_init_always(sb, inode))) {
 216                if (inode->i_sb->s_op->destroy_inode)
 217                        inode->i_sb->s_op->destroy_inode(inode);
 218                else
 219                        kmem_cache_free(inode_cachep, inode);
 220                return NULL;
 221        }
 222
 223        return inode;
 224}
 225
 226void free_inode_nonrcu(struct inode *inode)
 227{
 228        kmem_cache_free(inode_cachep, inode);
 229}
 230EXPORT_SYMBOL(free_inode_nonrcu);
 231
 232void __destroy_inode(struct inode *inode)
 233{
 234        BUG_ON(inode_has_buffers(inode));
 235        security_inode_free(inode);
 236        fsnotify_inode_delete(inode);
 237        if (!inode->i_nlink) {
 238                WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
 239                atomic_long_dec(&inode->i_sb->s_remove_count);
 240        }
 241
 242#ifdef CONFIG_FS_POSIX_ACL
 243        if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
 244                posix_acl_release(inode->i_acl);
 245        if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
 246                posix_acl_release(inode->i_default_acl);
 247#endif
 248        this_cpu_dec(nr_inodes);
 249}
 250EXPORT_SYMBOL(__destroy_inode);
 251
 252static void i_callback(struct rcu_head *head)
 253{
 254        struct inode *inode = container_of(head, struct inode, i_rcu);
 255        kmem_cache_free(inode_cachep, inode);
 256}
 257
 258static void destroy_inode(struct inode *inode)
 259{
 260        BUG_ON(!list_empty(&inode->i_lru));
 261        __destroy_inode(inode);
 262        if (inode->i_sb->s_op->destroy_inode)
 263                inode->i_sb->s_op->destroy_inode(inode);
 264        else
 265                call_rcu(&inode->i_rcu, i_callback);
 266}
 267
 268/**
 269 * drop_nlink - directly drop an inode's link count
 270 * @inode: inode
 271 *
 272 * This is a low-level filesystem helper to replace any
 273 * direct filesystem manipulation of i_nlink.  In cases
 274 * where we are attempting to track writes to the
 275 * filesystem, a decrement to zero means an imminent
 276 * write when the file is truncated and actually unlinked
 277 * on the filesystem.
 278 */
 279void drop_nlink(struct inode *inode)
 280{
 281        WARN_ON(inode->i_nlink == 0);
 282        inode->__i_nlink--;
 283        if (!inode->i_nlink)
 284                atomic_long_inc(&inode->i_sb->s_remove_count);
 285}
 286EXPORT_SYMBOL(drop_nlink);
 287
 288/**
 289 * clear_nlink - directly zero an inode's link count
 290 * @inode: inode
 291 *
 292 * This is a low-level filesystem helper to replace any
 293 * direct filesystem manipulation of i_nlink.  See
 294 * drop_nlink() for why we care about i_nlink hitting zero.
 295 */
 296void clear_nlink(struct inode *inode)
 297{
 298        if (inode->i_nlink) {
 299                inode->__i_nlink = 0;
 300                atomic_long_inc(&inode->i_sb->s_remove_count);
 301        }
 302}
 303EXPORT_SYMBOL(clear_nlink);
 304
 305/**
 306 * set_nlink - directly set an inode's link count
 307 * @inode: inode
 308 * @nlink: new nlink (should be non-zero)
 309 *
 310 * This is a low-level filesystem helper to replace any
 311 * direct filesystem manipulation of i_nlink.
 312 */
 313void set_nlink(struct inode *inode, unsigned int nlink)
 314{
 315        if (!nlink) {
 316                clear_nlink(inode);
 317        } else {
 318                /* Yes, some filesystems do change nlink from zero to one */
 319                if (inode->i_nlink == 0)
 320                        atomic_long_dec(&inode->i_sb->s_remove_count);
 321
 322                inode->__i_nlink = nlink;
 323        }
 324}
 325EXPORT_SYMBOL(set_nlink);
 326
 327/**
 328 * inc_nlink - directly increment an inode's link count
 329 * @inode: inode
 330 *
 331 * This is a low-level filesystem helper to replace any
 332 * direct filesystem manipulation of i_nlink.  Currently,
 333 * it is only here for parity with dec_nlink().
 334 */
 335void inc_nlink(struct inode *inode)
 336{
 337        if (unlikely(inode->i_nlink == 0)) {
 338                WARN_ON(!(inode->i_state & I_LINKABLE));
 339                atomic_long_dec(&inode->i_sb->s_remove_count);
 340        }
 341
 342        inode->__i_nlink++;
 343}
 344EXPORT_SYMBOL(inc_nlink);
 345
 346void address_space_init_once(struct address_space *mapping)
 347{
 348        memset(mapping, 0, sizeof(*mapping));
 349        INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
 350        spin_lock_init(&mapping->tree_lock);
 351        mutex_init(&mapping->i_mmap_mutex);
 352        INIT_LIST_HEAD(&mapping->private_list);
 353        spin_lock_init(&mapping->private_lock);
 354        mapping->i_mmap = RB_ROOT;
 355        INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
 356}
 357EXPORT_SYMBOL(address_space_init_once);
 358
 359/*
 360 * These are initializations that only need to be done
 361 * once, because the fields are idempotent across use
 362 * of the inode, so let the slab aware of that.
 363 */
 364void inode_init_once(struct inode *inode)
 365{
 366        memset(inode, 0, sizeof(*inode));
 367        INIT_HLIST_NODE(&inode->i_hash);
 368        INIT_LIST_HEAD(&inode->i_devices);
 369        INIT_LIST_HEAD(&inode->i_wb_list);
 370        INIT_LIST_HEAD(&inode->i_lru);
 371        address_space_init_once(&inode->i_data);
 372        i_size_ordered_init(inode);
 373#ifdef CONFIG_FSNOTIFY
 374        INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
 375#endif
 376}
 377EXPORT_SYMBOL(inode_init_once);
 378
 379static void init_once(void *foo)
 380{
 381        struct inode *inode = (struct inode *) foo;
 382
 383        inode_init_once(inode);
 384}
 385
 386/*
 387 * inode->i_lock must be held
 388 */
 389void __iget(struct inode *inode)
 390{
 391        atomic_inc(&inode->i_count);
 392}
 393
 394/*
 395 * get additional reference to inode; caller must already hold one.
 396 */
 397void ihold(struct inode *inode)
 398{
 399        WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 400}
 401EXPORT_SYMBOL(ihold);
 402
 403static void inode_lru_list_add(struct inode *inode)
 404{
 405        if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
 406                this_cpu_inc(nr_unused);
 407}
 408
 409/*
 410 * Add inode to LRU if needed (inode is unused and clean).
 411 *
 412 * Needs inode->i_lock held.
 413 */
 414void inode_add_lru(struct inode *inode)
 415{
 416        if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
 417            !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
 418                inode_lru_list_add(inode);
 419}
 420
 421
 422static void inode_lru_list_del(struct inode *inode)
 423{
 424
 425        if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
 426                this_cpu_dec(nr_unused);
 427}
 428
 429/**
 430 * inode_sb_list_add - add inode to the superblock list of inodes
 431 * @inode: inode to add
 432 */
 433void inode_sb_list_add(struct inode *inode)
 434{
 435        spin_lock(&inode_sb_list_lock);
 436        list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 437        spin_unlock(&inode_sb_list_lock);
 438}
 439EXPORT_SYMBOL_GPL(inode_sb_list_add);
 440
 441static inline void inode_sb_list_del(struct inode *inode)
 442{
 443        if (!list_empty(&inode->i_sb_list)) {
 444                spin_lock(&inode_sb_list_lock);
 445                list_del_init(&inode->i_sb_list);
 446                spin_unlock(&inode_sb_list_lock);
 447        }
 448}
 449
 450static unsigned long hash(struct super_block *sb, unsigned long hashval)
 451{
 452        unsigned long tmp;
 453
 454        tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 455                        L1_CACHE_BYTES;
 456        tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 457        return tmp & i_hash_mask;
 458}
 459
 460/**
 461 *      __insert_inode_hash - hash an inode
 462 *      @inode: unhashed inode
 463 *      @hashval: unsigned long value used to locate this object in the
 464 *              inode_hashtable.
 465 *
 466 *      Add an inode to the inode hash for this superblock.
 467 */
 468void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 469{
 470        struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 471
 472        spin_lock(&inode_hash_lock);
 473        spin_lock(&inode->i_lock);
 474        hlist_add_head(&inode->i_hash, b);
 475        spin_unlock(&inode->i_lock);
 476        spin_unlock(&inode_hash_lock);
 477}
 478EXPORT_SYMBOL(__insert_inode_hash);
 479
 480/**
 481 *      __remove_inode_hash - remove an inode from the hash
 482 *      @inode: inode to unhash
 483 *
 484 *      Remove an inode from the superblock.
 485 */
 486void __remove_inode_hash(struct inode *inode)
 487{
 488        spin_lock(&inode_hash_lock);
 489        spin_lock(&inode->i_lock);
 490        hlist_del_init(&inode->i_hash);
 491        spin_unlock(&inode->i_lock);
 492        spin_unlock(&inode_hash_lock);
 493}
 494EXPORT_SYMBOL(__remove_inode_hash);
 495
 496void clear_inode(struct inode *inode)
 497{
 498        might_sleep();
 499        /*
 500         * We have to cycle tree_lock here because reclaim can be still in the
 501         * process of removing the last page (in __delete_from_page_cache())
 502         * and we must not free mapping under it.
 503         */
 504        spin_lock_irq(&inode->i_data.tree_lock);
 505        BUG_ON(inode->i_data.nrpages);
 506        spin_unlock_irq(&inode->i_data.tree_lock);
 507        BUG_ON(!list_empty(&inode->i_data.private_list));
 508        BUG_ON(!(inode->i_state & I_FREEING));
 509        BUG_ON(inode->i_state & I_CLEAR);
 510        /* don't need i_lock here, no concurrent mods to i_state */
 511        inode->i_state = I_FREEING | I_CLEAR;
 512}
 513EXPORT_SYMBOL(clear_inode);
 514
 515/*
 516 * Free the inode passed in, removing it from the lists it is still connected
 517 * to. We remove any pages still attached to the inode and wait for any IO that
 518 * is still in progress before finally destroying the inode.
 519 *
 520 * An inode must already be marked I_FREEING so that we avoid the inode being
 521 * moved back onto lists if we race with other code that manipulates the lists
 522 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 523 *
 524 * An inode must already be removed from the LRU list before being evicted from
 525 * the cache. This should occur atomically with setting the I_FREEING state
 526 * flag, so no inodes here should ever be on the LRU when being evicted.
 527 */
 528static void evict(struct inode *inode)
 529{
 530        const struct super_operations *op = inode->i_sb->s_op;
 531
 532        BUG_ON(!(inode->i_state & I_FREEING));
 533        BUG_ON(!list_empty(&inode->i_lru));
 534
 535        if (!list_empty(&inode->i_wb_list))
 536                inode_wb_list_del(inode);
 537
 538        inode_sb_list_del(inode);
 539
 540        /*
 541         * Wait for flusher thread to be done with the inode so that filesystem
 542         * does not start destroying it while writeback is still running. Since
 543         * the inode has I_FREEING set, flusher thread won't start new work on
 544         * the inode.  We just have to wait for running writeback to finish.
 545         */
 546        inode_wait_for_writeback(inode);
 547
 548        if (op->evict_inode) {
 549                op->evict_inode(inode);
 550        } else {
 551                if (inode->i_data.nrpages)
 552                        truncate_inode_pages(&inode->i_data, 0);
 553                clear_inode(inode);
 554        }
 555        if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 556                bd_forget(inode);
 557        if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 558                cd_forget(inode);
 559
 560        remove_inode_hash(inode);
 561
 562        spin_lock(&inode->i_lock);
 563        wake_up_bit(&inode->i_state, __I_NEW);
 564        BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 565        spin_unlock(&inode->i_lock);
 566
 567        destroy_inode(inode);
 568}
 569
 570/*
 571 * dispose_list - dispose of the contents of a local list
 572 * @head: the head of the list to free
 573 *
 574 * Dispose-list gets a local list with local inodes in it, so it doesn't
 575 * need to worry about list corruption and SMP locks.
 576 */
 577static void dispose_list(struct list_head *head)
 578{
 579        while (!list_empty(head)) {
 580                struct inode *inode;
 581
 582                inode = list_first_entry(head, struct inode, i_lru);
 583                list_del_init(&inode->i_lru);
 584
 585                evict(inode);
 586        }
 587}
 588
 589/**
 590 * evict_inodes - evict all evictable inodes for a superblock
 591 * @sb:         superblock to operate on
 592 *
 593 * Make sure that no inodes with zero refcount are retained.  This is
 594 * called by superblock shutdown after having MS_ACTIVE flag removed,
 595 * so any inode reaching zero refcount during or after that call will
 596 * be immediately evicted.
 597 */
 598void evict_inodes(struct super_block *sb)
 599{
 600        struct inode *inode, *next;
 601        LIST_HEAD(dispose);
 602
 603        spin_lock(&inode_sb_list_lock);
 604        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 605                if (atomic_read(&inode->i_count))
 606                        continue;
 607
 608                spin_lock(&inode->i_lock);
 609                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 610                        spin_unlock(&inode->i_lock);
 611                        continue;
 612                }
 613
 614                inode->i_state |= I_FREEING;
 615                inode_lru_list_del(inode);
 616                spin_unlock(&inode->i_lock);
 617                list_add(&inode->i_lru, &dispose);
 618        }
 619        spin_unlock(&inode_sb_list_lock);
 620
 621        dispose_list(&dispose);
 622}
 623
 624/**
 625 * invalidate_inodes    - attempt to free all inodes on a superblock
 626 * @sb:         superblock to operate on
 627 * @kill_dirty: flag to guide handling of dirty inodes
 628 *
 629 * Attempts to free all inodes for a given superblock.  If there were any
 630 * busy inodes return a non-zero value, else zero.
 631 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 632 * them as busy.
 633 */
 634int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 635{
 636        int busy = 0;
 637        struct inode *inode, *next;
 638        LIST_HEAD(dispose);
 639
 640        spin_lock(&inode_sb_list_lock);
 641        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 642                spin_lock(&inode->i_lock);
 643                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 644                        spin_unlock(&inode->i_lock);
 645                        continue;
 646                }
 647                if (inode->i_state & I_DIRTY && !kill_dirty) {
 648                        spin_unlock(&inode->i_lock);
 649                        busy = 1;
 650                        continue;
 651                }
 652                if (atomic_read(&inode->i_count)) {
 653                        spin_unlock(&inode->i_lock);
 654                        busy = 1;
 655                        continue;
 656                }
 657
 658                inode->i_state |= I_FREEING;
 659                inode_lru_list_del(inode);
 660                spin_unlock(&inode->i_lock);
 661                list_add(&inode->i_lru, &dispose);
 662        }
 663        spin_unlock(&inode_sb_list_lock);
 664
 665        dispose_list(&dispose);
 666
 667        return busy;
 668}
 669
 670/*
 671 * Isolate the inode from the LRU in preparation for freeing it.
 672 *
 673 * Any inodes which are pinned purely because of attached pagecache have their
 674 * pagecache removed.  If the inode has metadata buffers attached to
 675 * mapping->private_list then try to remove them.
 676 *
 677 * If the inode has the I_REFERENCED flag set, then it means that it has been
 678 * used recently - the flag is set in iput_final(). When we encounter such an
 679 * inode, clear the flag and move it to the back of the LRU so it gets another
 680 * pass through the LRU before it gets reclaimed. This is necessary because of
 681 * the fact we are doing lazy LRU updates to minimise lock contention so the
 682 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 683 * with this flag set because they are the inodes that are out of order.
 684 */
 685static enum lru_status
 686inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
 687{
 688        struct list_head *freeable = arg;
 689        struct inode    *inode = container_of(item, struct inode, i_lru);
 690
 691        /*
 692         * we are inverting the lru lock/inode->i_lock here, so use a trylock.
 693         * If we fail to get the lock, just skip it.
 694         */
 695        if (!spin_trylock(&inode->i_lock))
 696                return LRU_SKIP;
 697
 698        /*
 699         * Referenced or dirty inodes are still in use. Give them another pass
 700         * through the LRU as we canot reclaim them now.
 701         */
 702        if (atomic_read(&inode->i_count) ||
 703            (inode->i_state & ~I_REFERENCED)) {
 704                list_del_init(&inode->i_lru);
 705                spin_unlock(&inode->i_lock);
 706                this_cpu_dec(nr_unused);
 707                return LRU_REMOVED;
 708        }
 709
 710        /* recently referenced inodes get one more pass */
 711        if (inode->i_state & I_REFERENCED) {
 712                inode->i_state &= ~I_REFERENCED;
 713                spin_unlock(&inode->i_lock);
 714                return LRU_ROTATE;
 715        }
 716
 717        if (inode_has_buffers(inode) || inode->i_data.nrpages) {
 718                __iget(inode);
 719                spin_unlock(&inode->i_lock);
 720                spin_unlock(lru_lock);
 721                if (remove_inode_buffers(inode)) {
 722                        unsigned long reap;
 723                        reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
 724                        if (current_is_kswapd())
 725                                __count_vm_events(KSWAPD_INODESTEAL, reap);
 726                        else
 727                                __count_vm_events(PGINODESTEAL, reap);
 728                        if (current->reclaim_state)
 729                                current->reclaim_state->reclaimed_slab += reap;
 730                }
 731                iput(inode);
 732                spin_lock(lru_lock);
 733                return LRU_RETRY;
 734        }
 735
 736        WARN_ON(inode->i_state & I_NEW);
 737        inode->i_state |= I_FREEING;
 738        list_move(&inode->i_lru, freeable);
 739        spin_unlock(&inode->i_lock);
 740
 741        this_cpu_dec(nr_unused);
 742        return LRU_REMOVED;
 743}
 744
 745/*
 746 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 747 * This is called from the superblock shrinker function with a number of inodes
 748 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 749 * then are freed outside inode_lock by dispose_list().
 750 */
 751long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
 752                     int nid)
 753{
 754        LIST_HEAD(freeable);
 755        long freed;
 756
 757        freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
 758                                       &freeable, &nr_to_scan);
 759        dispose_list(&freeable);
 760        return freed;
 761}
 762
 763static void __wait_on_freeing_inode(struct inode *inode);
 764/*
 765 * Called with the inode lock held.
 766 */
 767static struct inode *find_inode(struct super_block *sb,
 768                                struct hlist_head *head,
 769                                int (*test)(struct inode *, void *),
 770                                void *data)
 771{
 772        struct inode *inode = NULL;
 773
 774repeat:
 775        hlist_for_each_entry(inode, head, i_hash) {
 776                spin_lock(&inode->i_lock);
 777                if (inode->i_sb != sb) {
 778                        spin_unlock(&inode->i_lock);
 779                        continue;
 780                }
 781                if (!test(inode, data)) {
 782                        spin_unlock(&inode->i_lock);
 783                        continue;
 784                }
 785                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 786                        __wait_on_freeing_inode(inode);
 787                        goto repeat;
 788                }
 789                __iget(inode);
 790                spin_unlock(&inode->i_lock);
 791                return inode;
 792        }
 793        return NULL;
 794}
 795
 796/*
 797 * find_inode_fast is the fast path version of find_inode, see the comment at
 798 * iget_locked for details.
 799 */
 800static struct inode *find_inode_fast(struct super_block *sb,
 801                                struct hlist_head *head, unsigned long ino)
 802{
 803        struct inode *inode = NULL;
 804
 805repeat:
 806        hlist_for_each_entry(inode, head, i_hash) {
 807                spin_lock(&inode->i_lock);
 808                if (inode->i_ino != ino) {
 809                        spin_unlock(&inode->i_lock);
 810                        continue;
 811                }
 812                if (inode->i_sb != sb) {
 813                        spin_unlock(&inode->i_lock);
 814                        continue;
 815                }
 816                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 817                        __wait_on_freeing_inode(inode);
 818                        goto repeat;
 819                }
 820                __iget(inode);
 821                spin_unlock(&inode->i_lock);
 822                return inode;
 823        }
 824        return NULL;
 825}
 826
 827/*
 828 * Each cpu owns a range of LAST_INO_BATCH numbers.
 829 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 830 * to renew the exhausted range.
 831 *
 832 * This does not significantly increase overflow rate because every CPU can
 833 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 834 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 835 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 836 * overflow rate by 2x, which does not seem too significant.
 837 *
 838 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 839 * error if st_ino won't fit in target struct field. Use 32bit counter
 840 * here to attempt to avoid that.
 841 */
 842#define LAST_INO_BATCH 1024
 843static DEFINE_PER_CPU(unsigned int, last_ino);
 844
 845unsigned int get_next_ino(void)
 846{
 847        unsigned int *p = &get_cpu_var(last_ino);
 848        unsigned int res = *p;
 849
 850#ifdef CONFIG_SMP
 851        if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 852                static atomic_t shared_last_ino;
 853                int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 854
 855                res = next - LAST_INO_BATCH;
 856        }
 857#endif
 858
 859        *p = ++res;
 860        put_cpu_var(last_ino);
 861        return res;
 862}
 863EXPORT_SYMBOL(get_next_ino);
 864
 865/**
 866 *      new_inode_pseudo        - obtain an inode
 867 *      @sb: superblock
 868 *
 869 *      Allocates a new inode for given superblock.
 870 *      Inode wont be chained in superblock s_inodes list
 871 *      This means :
 872 *      - fs can't be unmount
 873 *      - quotas, fsnotify, writeback can't work
 874 */
 875struct inode *new_inode_pseudo(struct super_block *sb)
 876{
 877        struct inode *inode = alloc_inode(sb);
 878
 879        if (inode) {
 880                spin_lock(&inode->i_lock);
 881                inode->i_state = 0;
 882                spin_unlock(&inode->i_lock);
 883                INIT_LIST_HEAD(&inode->i_sb_list);
 884        }
 885        return inode;
 886}
 887
 888/**
 889 *      new_inode       - obtain an inode
 890 *      @sb: superblock
 891 *
 892 *      Allocates a new inode for given superblock. The default gfp_mask
 893 *      for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
 894 *      If HIGHMEM pages are unsuitable or it is known that pages allocated
 895 *      for the page cache are not reclaimable or migratable,
 896 *      mapping_set_gfp_mask() must be called with suitable flags on the
 897 *      newly created inode's mapping
 898 *
 899 */
 900struct inode *new_inode(struct super_block *sb)
 901{
 902        struct inode *inode;
 903
 904        spin_lock_prefetch(&inode_sb_list_lock);
 905
 906        inode = new_inode_pseudo(sb);
 907        if (inode)
 908                inode_sb_list_add(inode);
 909        return inode;
 910}
 911EXPORT_SYMBOL(new_inode);
 912
 913#ifdef CONFIG_DEBUG_LOCK_ALLOC
 914void lockdep_annotate_inode_mutex_key(struct inode *inode)
 915{
 916        if (S_ISDIR(inode->i_mode)) {
 917                struct file_system_type *type = inode->i_sb->s_type;
 918
 919                /* Set new key only if filesystem hasn't already changed it */
 920                if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
 921                        /*
 922                         * ensure nobody is actually holding i_mutex
 923                         */
 924                        mutex_destroy(&inode->i_mutex);
 925                        mutex_init(&inode->i_mutex);
 926                        lockdep_set_class(&inode->i_mutex,
 927                                          &type->i_mutex_dir_key);
 928                }
 929        }
 930}
 931EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
 932#endif
 933
 934/**
 935 * unlock_new_inode - clear the I_NEW state and wake up any waiters
 936 * @inode:      new inode to unlock
 937 *
 938 * Called when the inode is fully initialised to clear the new state of the
 939 * inode and wake up anyone waiting for the inode to finish initialisation.
 940 */
 941void unlock_new_inode(struct inode *inode)
 942{
 943        lockdep_annotate_inode_mutex_key(inode);
 944        spin_lock(&inode->i_lock);
 945        WARN_ON(!(inode->i_state & I_NEW));
 946        inode->i_state &= ~I_NEW;
 947        smp_mb();
 948        wake_up_bit(&inode->i_state, __I_NEW);
 949        spin_unlock(&inode->i_lock);
 950}
 951EXPORT_SYMBOL(unlock_new_inode);
 952
 953/**
 954 * iget5_locked - obtain an inode from a mounted file system
 955 * @sb:         super block of file system
 956 * @hashval:    hash value (usually inode number) to get
 957 * @test:       callback used for comparisons between inodes
 958 * @set:        callback used to initialize a new struct inode
 959 * @data:       opaque data pointer to pass to @test and @set
 960 *
 961 * Search for the inode specified by @hashval and @data in the inode cache,
 962 * and if present it is return it with an increased reference count. This is
 963 * a generalized version of iget_locked() for file systems where the inode
 964 * number is not sufficient for unique identification of an inode.
 965 *
 966 * If the inode is not in cache, allocate a new inode and return it locked,
 967 * hashed, and with the I_NEW flag set. The file system gets to fill it in
 968 * before unlocking it via unlock_new_inode().
 969 *
 970 * Note both @test and @set are called with the inode_hash_lock held, so can't
 971 * sleep.
 972 */
 973struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
 974                int (*test)(struct inode *, void *),
 975                int (*set)(struct inode *, void *), void *data)
 976{
 977        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
 978        struct inode *inode;
 979
 980        spin_lock(&inode_hash_lock);
 981        inode = find_inode(sb, head, test, data);
 982        spin_unlock(&inode_hash_lock);
 983
 984        if (inode) {
 985                wait_on_inode(inode);
 986                return inode;
 987        }
 988
 989        inode = alloc_inode(sb);
 990        if (inode) {
 991                struct inode *old;
 992
 993                spin_lock(&inode_hash_lock);
 994                /* We released the lock, so.. */
 995                old = find_inode(sb, head, test, data);
 996                if (!old) {
 997                        if (set(inode, data))
 998                                goto set_failed;
 999
1000                        spin_lock(&inode->i_lock);
1001                        inode->i_state = I_NEW;
1002                        hlist_add_head(&inode->i_hash, head);
1003                        spin_unlock(&inode->i_lock);
1004                        inode_sb_list_add(inode);
1005                        spin_unlock(&inode_hash_lock);
1006
1007                        /* Return the locked inode with I_NEW set, the
1008                         * caller is responsible for filling in the contents
1009                         */
1010                        return inode;
1011                }
1012
1013                /*
1014                 * Uhhuh, somebody else created the same inode under
1015                 * us. Use the old inode instead of the one we just
1016                 * allocated.
1017                 */
1018                spin_unlock(&inode_hash_lock);
1019                destroy_inode(inode);
1020                inode = old;
1021                wait_on_inode(inode);
1022        }
1023        return inode;
1024
1025set_failed:
1026        spin_unlock(&inode_hash_lock);
1027        destroy_inode(inode);
1028        return NULL;
1029}
1030EXPORT_SYMBOL(iget5_locked);
1031
1032/**
1033 * iget_locked - obtain an inode from a mounted file system
1034 * @sb:         super block of file system
1035 * @ino:        inode number to get
1036 *
1037 * Search for the inode specified by @ino in the inode cache and if present
1038 * return it with an increased reference count. This is for file systems
1039 * where the inode number is sufficient for unique identification of an inode.
1040 *
1041 * If the inode is not in cache, allocate a new inode and return it locked,
1042 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1043 * before unlocking it via unlock_new_inode().
1044 */
1045struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1046{
1047        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1048        struct inode *inode;
1049
1050        spin_lock(&inode_hash_lock);
1051        inode = find_inode_fast(sb, head, ino);
1052        spin_unlock(&inode_hash_lock);
1053        if (inode) {
1054                wait_on_inode(inode);
1055                return inode;
1056        }
1057
1058        inode = alloc_inode(sb);
1059        if (inode) {
1060                struct inode *old;
1061
1062                spin_lock(&inode_hash_lock);
1063                /* We released the lock, so.. */
1064                old = find_inode_fast(sb, head, ino);
1065                if (!old) {
1066                        inode->i_ino = ino;
1067                        spin_lock(&inode->i_lock);
1068                        inode->i_state = I_NEW;
1069                        hlist_add_head(&inode->i_hash, head);
1070                        spin_unlock(&inode->i_lock);
1071                        inode_sb_list_add(inode);
1072                        spin_unlock(&inode_hash_lock);
1073
1074                        /* Return the locked inode with I_NEW set, the
1075                         * caller is responsible for filling in the contents
1076                         */
1077                        return inode;
1078                }
1079
1080                /*
1081                 * Uhhuh, somebody else created the same inode under
1082                 * us. Use the old inode instead of the one we just
1083                 * allocated.
1084                 */
1085                spin_unlock(&inode_hash_lock);
1086                destroy_inode(inode);
1087                inode = old;
1088                wait_on_inode(inode);
1089        }
1090        return inode;
1091}
1092EXPORT_SYMBOL(iget_locked);
1093
1094/*
1095 * search the inode cache for a matching inode number.
1096 * If we find one, then the inode number we are trying to
1097 * allocate is not unique and so we should not use it.
1098 *
1099 * Returns 1 if the inode number is unique, 0 if it is not.
1100 */
1101static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1102{
1103        struct hlist_head *b = inode_hashtable + hash(sb, ino);
1104        struct inode *inode;
1105
1106        spin_lock(&inode_hash_lock);
1107        hlist_for_each_entry(inode, b, i_hash) {
1108                if (inode->i_ino == ino && inode->i_sb == sb) {
1109                        spin_unlock(&inode_hash_lock);
1110                        return 0;
1111                }
1112        }
1113        spin_unlock(&inode_hash_lock);
1114
1115        return 1;
1116}
1117
1118/**
1119 *      iunique - get a unique inode number
1120 *      @sb: superblock
1121 *      @max_reserved: highest reserved inode number
1122 *
1123 *      Obtain an inode number that is unique on the system for a given
1124 *      superblock. This is used by file systems that have no natural
1125 *      permanent inode numbering system. An inode number is returned that
1126 *      is higher than the reserved limit but unique.
1127 *
1128 *      BUGS:
1129 *      With a large number of inodes live on the file system this function
1130 *      currently becomes quite slow.
1131 */
1132ino_t iunique(struct super_block *sb, ino_t max_reserved)
1133{
1134        /*
1135         * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1136         * error if st_ino won't fit in target struct field. Use 32bit counter
1137         * here to attempt to avoid that.
1138         */
1139        static DEFINE_SPINLOCK(iunique_lock);
1140        static unsigned int counter;
1141        ino_t res;
1142
1143        spin_lock(&iunique_lock);
1144        do {
1145                if (counter <= max_reserved)
1146                        counter = max_reserved + 1;
1147                res = counter++;
1148        } while (!test_inode_iunique(sb, res));
1149        spin_unlock(&iunique_lock);
1150
1151        return res;
1152}
1153EXPORT_SYMBOL(iunique);
1154
1155struct inode *igrab(struct inode *inode)
1156{
1157        spin_lock(&inode->i_lock);
1158        if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1159                __iget(inode);
1160                spin_unlock(&inode->i_lock);
1161        } else {
1162                spin_unlock(&inode->i_lock);
1163                /*
1164                 * Handle the case where s_op->clear_inode is not been
1165                 * called yet, and somebody is calling igrab
1166                 * while the inode is getting freed.
1167                 */
1168                inode = NULL;
1169        }
1170        return inode;
1171}
1172EXPORT_SYMBOL(igrab);
1173
1174/**
1175 * ilookup5_nowait - search for an inode in the inode cache
1176 * @sb:         super block of file system to search
1177 * @hashval:    hash value (usually inode number) to search for
1178 * @test:       callback used for comparisons between inodes
1179 * @data:       opaque data pointer to pass to @test
1180 *
1181 * Search for the inode specified by @hashval and @data in the inode cache.
1182 * If the inode is in the cache, the inode is returned with an incremented
1183 * reference count.
1184 *
1185 * Note: I_NEW is not waited upon so you have to be very careful what you do
1186 * with the returned inode.  You probably should be using ilookup5() instead.
1187 *
1188 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1189 */
1190struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1191                int (*test)(struct inode *, void *), void *data)
1192{
1193        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1194        struct inode *inode;
1195
1196        spin_lock(&inode_hash_lock);
1197        inode = find_inode(sb, head, test, data);
1198        spin_unlock(&inode_hash_lock);
1199
1200        return inode;
1201}
1202EXPORT_SYMBOL(ilookup5_nowait);
1203
1204/**
1205 * ilookup5 - search for an inode in the inode cache
1206 * @sb:         super block of file system to search
1207 * @hashval:    hash value (usually inode number) to search for
1208 * @test:       callback used for comparisons between inodes
1209 * @data:       opaque data pointer to pass to @test
1210 *
1211 * Search for the inode specified by @hashval and @data in the inode cache,
1212 * and if the inode is in the cache, return the inode with an incremented
1213 * reference count.  Waits on I_NEW before returning the inode.
1214 * returned with an incremented reference count.
1215 *
1216 * This is a generalized version of ilookup() for file systems where the
1217 * inode number is not sufficient for unique identification of an inode.
1218 *
1219 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1220 */
1221struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1222                int (*test)(struct inode *, void *), void *data)
1223{
1224        struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1225
1226        if (inode)
1227                wait_on_inode(inode);
1228        return inode;
1229}
1230EXPORT_SYMBOL(ilookup5);
1231
1232/**
1233 * ilookup - search for an inode in the inode cache
1234 * @sb:         super block of file system to search
1235 * @ino:        inode number to search for
1236 *
1237 * Search for the inode @ino in the inode cache, and if the inode is in the
1238 * cache, the inode is returned with an incremented reference count.
1239 */
1240struct inode *ilookup(struct super_block *sb, unsigned long ino)
1241{
1242        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1243        struct inode *inode;
1244
1245        spin_lock(&inode_hash_lock);
1246        inode = find_inode_fast(sb, head, ino);
1247        spin_unlock(&inode_hash_lock);
1248
1249        if (inode)
1250                wait_on_inode(inode);
1251        return inode;
1252}
1253EXPORT_SYMBOL(ilookup);
1254
1255int insert_inode_locked(struct inode *inode)
1256{
1257        struct super_block *sb = inode->i_sb;
1258        ino_t ino = inode->i_ino;
1259        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1260
1261        while (1) {
1262                struct inode *old = NULL;
1263                spin_lock(&inode_hash_lock);
1264                hlist_for_each_entry(old, head, i_hash) {
1265                        if (old->i_ino != ino)
1266                                continue;
1267                        if (old->i_sb != sb)
1268                                continue;
1269                        spin_lock(&old->i_lock);
1270                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1271                                spin_unlock(&old->i_lock);
1272                                continue;
1273                        }
1274                        break;
1275                }
1276                if (likely(!old)) {
1277                        spin_lock(&inode->i_lock);
1278                        inode->i_state |= I_NEW;
1279                        hlist_add_head(&inode->i_hash, head);
1280                        spin_unlock(&inode->i_lock);
1281                        spin_unlock(&inode_hash_lock);
1282                        return 0;
1283                }
1284                __iget(old);
1285                spin_unlock(&old->i_lock);
1286                spin_unlock(&inode_hash_lock);
1287                wait_on_inode(old);
1288                if (unlikely(!inode_unhashed(old))) {
1289                        iput(old);
1290                        return -EBUSY;
1291                }
1292                iput(old);
1293        }
1294}
1295EXPORT_SYMBOL(insert_inode_locked);
1296
1297int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1298                int (*test)(struct inode *, void *), void *data)
1299{
1300        struct super_block *sb = inode->i_sb;
1301        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1302
1303        while (1) {
1304                struct inode *old = NULL;
1305
1306                spin_lock(&inode_hash_lock);
1307                hlist_for_each_entry(old, head, i_hash) {
1308                        if (old->i_sb != sb)
1309                                continue;
1310                        if (!test(old, data))
1311                                continue;
1312                        spin_lock(&old->i_lock);
1313                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1314                                spin_unlock(&old->i_lock);
1315                                continue;
1316                        }
1317                        break;
1318                }
1319                if (likely(!old)) {
1320                        spin_lock(&inode->i_lock);
1321                        inode->i_state |= I_NEW;
1322                        hlist_add_head(&inode->i_hash, head);
1323                        spin_unlock(&inode->i_lock);
1324                        spin_unlock(&inode_hash_lock);
1325                        return 0;
1326                }
1327                __iget(old);
1328                spin_unlock(&old->i_lock);
1329                spin_unlock(&inode_hash_lock);
1330                wait_on_inode(old);
1331                if (unlikely(!inode_unhashed(old))) {
1332                        iput(old);
1333                        return -EBUSY;
1334                }
1335                iput(old);
1336        }
1337}
1338EXPORT_SYMBOL(insert_inode_locked4);
1339
1340
1341int generic_delete_inode(struct inode *inode)
1342{
1343        return 1;
1344}
1345EXPORT_SYMBOL(generic_delete_inode);
1346
1347/*
1348 * Called when we're dropping the last reference
1349 * to an inode.
1350 *
1351 * Call the FS "drop_inode()" function, defaulting to
1352 * the legacy UNIX filesystem behaviour.  If it tells
1353 * us to evict inode, do so.  Otherwise, retain inode
1354 * in cache if fs is alive, sync and evict if fs is
1355 * shutting down.
1356 */
1357static void iput_final(struct inode *inode)
1358{
1359        struct super_block *sb = inode->i_sb;
1360        const struct super_operations *op = inode->i_sb->s_op;
1361        int drop;
1362
1363        WARN_ON(inode->i_state & I_NEW);
1364
1365        if (op->drop_inode)
1366                drop = op->drop_inode(inode);
1367        else
1368                drop = generic_drop_inode(inode);
1369
1370        if (!drop && (sb->s_flags & MS_ACTIVE)) {
1371                inode->i_state |= I_REFERENCED;
1372                inode_add_lru(inode);
1373                spin_unlock(&inode->i_lock);
1374                return;
1375        }
1376
1377        if (!drop) {
1378                inode->i_state |= I_WILL_FREE;
1379                spin_unlock(&inode->i_lock);
1380                write_inode_now(inode, 1);
1381                spin_lock(&inode->i_lock);
1382                WARN_ON(inode->i_state & I_NEW);
1383                inode->i_state &= ~I_WILL_FREE;
1384        }
1385
1386        inode->i_state |= I_FREEING;
1387        if (!list_empty(&inode->i_lru))
1388                inode_lru_list_del(inode);
1389        spin_unlock(&inode->i_lock);
1390
1391        evict(inode);
1392}
1393
1394/**
1395 *      iput    - put an inode
1396 *      @inode: inode to put
1397 *
1398 *      Puts an inode, dropping its usage count. If the inode use count hits
1399 *      zero, the inode is then freed and may also be destroyed.
1400 *
1401 *      Consequently, iput() can sleep.
1402 */
1403void iput(struct inode *inode)
1404{
1405        if (inode) {
1406                BUG_ON(inode->i_state & I_CLEAR);
1407
1408                if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1409                        iput_final(inode);
1410        }
1411}
1412EXPORT_SYMBOL(iput);
1413
1414/**
1415 *      bmap    - find a block number in a file
1416 *      @inode: inode of file
1417 *      @block: block to find
1418 *
1419 *      Returns the block number on the device holding the inode that
1420 *      is the disk block number for the block of the file requested.
1421 *      That is, asked for block 4 of inode 1 the function will return the
1422 *      disk block relative to the disk start that holds that block of the
1423 *      file.
1424 */
1425sector_t bmap(struct inode *inode, sector_t block)
1426{
1427        sector_t res = 0;
1428        if (inode->i_mapping->a_ops->bmap)
1429                res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1430        return res;
1431}
1432EXPORT_SYMBOL(bmap);
1433
1434/*
1435 * With relative atime, only update atime if the previous atime is
1436 * earlier than either the ctime or mtime or if at least a day has
1437 * passed since the last atime update.
1438 */
1439static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1440                             struct timespec now)
1441{
1442
1443        if (!(mnt->mnt_flags & MNT_RELATIME))
1444                return 1;
1445        /*
1446         * Is mtime younger than atime? If yes, update atime:
1447         */
1448        if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1449                return 1;
1450        /*
1451         * Is ctime younger than atime? If yes, update atime:
1452         */
1453        if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1454                return 1;
1455
1456        /*
1457         * Is the previous atime value older than a day? If yes,
1458         * update atime:
1459         */
1460        if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1461                return 1;
1462        /*
1463         * Good, we can skip the atime update:
1464         */
1465        return 0;
1466}
1467
1468/*
1469 * This does the actual work of updating an inodes time or version.  Must have
1470 * had called mnt_want_write() before calling this.
1471 */
1472static int update_time(struct inode *inode, struct timespec *time, int flags)
1473{
1474        if (inode->i_op->update_time)
1475                return inode->i_op->update_time(inode, time, flags);
1476
1477        if (flags & S_ATIME)
1478                inode->i_atime = *time;
1479        if (flags & S_VERSION)
1480                inode_inc_iversion(inode);
1481        if (flags & S_CTIME)
1482                inode->i_ctime = *time;
1483        if (flags & S_MTIME)
1484                inode->i_mtime = *time;
1485        mark_inode_dirty_sync(inode);
1486        return 0;
1487}
1488
1489/**
1490 *      touch_atime     -       update the access time
1491 *      @path: the &struct path to update
1492 *
1493 *      Update the accessed time on an inode and mark it for writeback.
1494 *      This function automatically handles read only file systems and media,
1495 *      as well as the "noatime" flag and inode specific "noatime" markers.
1496 */
1497void touch_atime(const struct path *path)
1498{
1499        struct vfsmount *mnt = path->mnt;
1500        struct inode *inode = path->dentry->d_inode;
1501        struct timespec now;
1502
1503        if (inode->i_flags & S_NOATIME)
1504                return;
1505        if (IS_NOATIME(inode))
1506                return;
1507        if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1508                return;
1509
1510        if (mnt->mnt_flags & MNT_NOATIME)
1511                return;
1512        if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1513                return;
1514
1515        now = current_fs_time(inode->i_sb);
1516
1517        if (!relatime_need_update(mnt, inode, now))
1518                return;
1519
1520        if (timespec_equal(&inode->i_atime, &now))
1521                return;
1522
1523        if (!sb_start_write_trylock(inode->i_sb))
1524                return;
1525
1526        if (__mnt_want_write(mnt))
1527                goto skip_update;
1528        /*
1529         * File systems can error out when updating inodes if they need to
1530         * allocate new space to modify an inode (such is the case for
1531         * Btrfs), but since we touch atime while walking down the path we
1532         * really don't care if we failed to update the atime of the file,
1533         * so just ignore the return value.
1534         * We may also fail on filesystems that have the ability to make parts
1535         * of the fs read only, e.g. subvolumes in Btrfs.
1536         */
1537        update_time(inode, &now, S_ATIME);
1538        __mnt_drop_write(mnt);
1539skip_update:
1540        sb_end_write(inode->i_sb);
1541}
1542EXPORT_SYMBOL(touch_atime);
1543
1544/*
1545 * The logic we want is
1546 *
1547 *      if suid or (sgid and xgrp)
1548 *              remove privs
1549 */
1550int should_remove_suid(struct dentry *dentry)
1551{
1552        umode_t mode = dentry->d_inode->i_mode;
1553        int kill = 0;
1554
1555        /* suid always must be killed */
1556        if (unlikely(mode & S_ISUID))
1557                kill = ATTR_KILL_SUID;
1558
1559        /*
1560         * sgid without any exec bits is just a mandatory locking mark; leave
1561         * it alone.  If some exec bits are set, it's a real sgid; kill it.
1562         */
1563        if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1564                kill |= ATTR_KILL_SGID;
1565
1566        if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1567                return kill;
1568
1569        return 0;
1570}
1571EXPORT_SYMBOL(should_remove_suid);
1572
1573static int __remove_suid(struct dentry *dentry, int kill)
1574{
1575        struct iattr newattrs;
1576
1577        newattrs.ia_valid = ATTR_FORCE | kill;
1578        return notify_change(dentry, &newattrs);
1579}
1580
1581int file_remove_suid(struct file *file)
1582{
1583        struct dentry *dentry = file->f_path.dentry;
1584        struct inode *inode = dentry->d_inode;
1585        int killsuid;
1586        int killpriv;
1587        int error = 0;
1588
1589        /* Fast path for nothing security related */
1590        if (IS_NOSEC(inode))
1591                return 0;
1592
1593        killsuid = should_remove_suid(dentry);
1594        killpriv = security_inode_need_killpriv(dentry);
1595
1596        if (killpriv < 0)
1597                return killpriv;
1598        if (killpriv)
1599                error = security_inode_killpriv(dentry);
1600        if (!error && killsuid)
1601                error = __remove_suid(dentry, killsuid);
1602        if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1603                inode->i_flags |= S_NOSEC;
1604
1605        return error;
1606}
1607EXPORT_SYMBOL(file_remove_suid);
1608
1609/**
1610 *      file_update_time        -       update mtime and ctime time
1611 *      @file: file accessed
1612 *
1613 *      Update the mtime and ctime members of an inode and mark the inode
1614 *      for writeback.  Note that this function is meant exclusively for
1615 *      usage in the file write path of filesystems, and filesystems may
1616 *      choose to explicitly ignore update via this function with the
1617 *      S_NOCMTIME inode flag, e.g. for network filesystem where these
1618 *      timestamps are handled by the server.  This can return an error for
1619 *      file systems who need to allocate space in order to update an inode.
1620 */
1621
1622int file_update_time(struct file *file)
1623{
1624        struct inode *inode = file_inode(file);
1625        struct timespec now;
1626        int sync_it = 0;
1627        int ret;
1628
1629        /* First try to exhaust all avenues to not sync */
1630        if (IS_NOCMTIME(inode))
1631                return 0;
1632
1633        now = current_fs_time(inode->i_sb);
1634        if (!timespec_equal(&inode->i_mtime, &now))
1635                sync_it = S_MTIME;
1636
1637        if (!timespec_equal(&inode->i_ctime, &now))
1638                sync_it |= S_CTIME;
1639
1640        if (IS_I_VERSION(inode))
1641                sync_it |= S_VERSION;
1642
1643        if (!sync_it)
1644                return 0;
1645
1646        /* Finally allowed to write? Takes lock. */
1647        if (__mnt_want_write_file(file))
1648                return 0;
1649
1650        ret = update_time(inode, &now, sync_it);
1651        __mnt_drop_write_file(file);
1652
1653        return ret;
1654}
1655EXPORT_SYMBOL(file_update_time);
1656
1657int inode_needs_sync(struct inode *inode)
1658{
1659        if (IS_SYNC(inode))
1660                return 1;
1661        if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1662                return 1;
1663        return 0;
1664}
1665EXPORT_SYMBOL(inode_needs_sync);
1666
1667int inode_wait(void *word)
1668{
1669        schedule();
1670        return 0;
1671}
1672EXPORT_SYMBOL(inode_wait);
1673
1674/*
1675 * If we try to find an inode in the inode hash while it is being
1676 * deleted, we have to wait until the filesystem completes its
1677 * deletion before reporting that it isn't found.  This function waits
1678 * until the deletion _might_ have completed.  Callers are responsible
1679 * to recheck inode state.
1680 *
1681 * It doesn't matter if I_NEW is not set initially, a call to
1682 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1683 * will DTRT.
1684 */
1685static void __wait_on_freeing_inode(struct inode *inode)
1686{
1687        wait_queue_head_t *wq;
1688        DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1689        wq = bit_waitqueue(&inode->i_state, __I_NEW);
1690        prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1691        spin_unlock(&inode->i_lock);
1692        spin_unlock(&inode_hash_lock);
1693        schedule();
1694        finish_wait(wq, &wait.wait);
1695        spin_lock(&inode_hash_lock);
1696}
1697
1698static __initdata unsigned long ihash_entries;
1699static int __init set_ihash_entries(char *str)
1700{
1701        if (!str)
1702                return 0;
1703        ihash_entries = simple_strtoul(str, &str, 0);
1704        return 1;
1705}
1706__setup("ihash_entries=", set_ihash_entries);
1707
1708/*
1709 * Initialize the waitqueues and inode hash table.
1710 */
1711void __init inode_init_early(void)
1712{
1713        unsigned int loop;
1714
1715        /* If hashes are distributed across NUMA nodes, defer
1716         * hash allocation until vmalloc space is available.
1717         */
1718        if (hashdist)
1719                return;
1720
1721        inode_hashtable =
1722                alloc_large_system_hash("Inode-cache",
1723                                        sizeof(struct hlist_head),
1724                                        ihash_entries,
1725                                        14,
1726                                        HASH_EARLY,
1727                                        &i_hash_shift,
1728                                        &i_hash_mask,
1729                                        0,
1730                                        0);
1731
1732        for (loop = 0; loop < (1U << i_hash_shift); loop++)
1733                INIT_HLIST_HEAD(&inode_hashtable[loop]);
1734}
1735
1736void __init inode_init(void)
1737{
1738        unsigned int loop;
1739
1740        /* inode slab cache */
1741        inode_cachep = kmem_cache_create("inode_cache",
1742                                         sizeof(struct inode),
1743                                         0,
1744                                         (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1745                                         SLAB_MEM_SPREAD),
1746                                         init_once);
1747
1748        /* Hash may have been set up in inode_init_early */
1749        if (!hashdist)
1750                return;
1751
1752        inode_hashtable =
1753                alloc_large_system_hash("Inode-cache",
1754                                        sizeof(struct hlist_head),
1755                                        ihash_entries,
1756                                        14,
1757                                        0,
1758                                        &i_hash_shift,
1759                                        &i_hash_mask,
1760                                        0,
1761                                        0);
1762
1763        for (loop = 0; loop < (1U << i_hash_shift); loop++)
1764                INIT_HLIST_HEAD(&inode_hashtable[loop]);
1765}
1766
1767void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1768{
1769        inode->i_mode = mode;
1770        if (S_ISCHR(mode)) {
1771                inode->i_fop = &def_chr_fops;
1772                inode->i_rdev = rdev;
1773        } else if (S_ISBLK(mode)) {
1774                inode->i_fop = &def_blk_fops;
1775                inode->i_rdev = rdev;
1776        } else if (S_ISFIFO(mode))
1777                inode->i_fop = &pipefifo_fops;
1778        else if (S_ISSOCK(mode))
1779                inode->i_fop = &bad_sock_fops;
1780        else
1781                printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1782                                  " inode %s:%lu\n", mode, inode->i_sb->s_id,
1783                                  inode->i_ino);
1784}
1785EXPORT_SYMBOL(init_special_inode);
1786
1787/**
1788 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1789 * @inode: New inode
1790 * @dir: Directory inode
1791 * @mode: mode of the new inode
1792 */
1793void inode_init_owner(struct inode *inode, const struct inode *dir,
1794                        umode_t mode)
1795{
1796        inode->i_uid = current_fsuid();
1797        if (dir && dir->i_mode & S_ISGID) {
1798                inode->i_gid = dir->i_gid;
1799                if (S_ISDIR(mode))
1800                        mode |= S_ISGID;
1801        } else
1802                inode->i_gid = current_fsgid();
1803        inode->i_mode = mode;
1804}
1805EXPORT_SYMBOL(inode_init_owner);
1806
1807/**
1808 * inode_owner_or_capable - check current task permissions to inode
1809 * @inode: inode being checked
1810 *
1811 * Return true if current either has CAP_FOWNER to the inode, or
1812 * owns the file.
1813 */
1814bool inode_owner_or_capable(const struct inode *inode)
1815{
1816        if (uid_eq(current_fsuid(), inode->i_uid))
1817                return true;
1818        if (inode_capable(inode, CAP_FOWNER))
1819                return true;
1820        return false;
1821}
1822EXPORT_SYMBOL(inode_owner_or_capable);
1823
1824/*
1825 * Direct i/o helper functions
1826 */
1827static void __inode_dio_wait(struct inode *inode)
1828{
1829        wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1830        DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1831
1832        do {
1833                prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1834                if (atomic_read(&inode->i_dio_count))
1835                        schedule();
1836        } while (atomic_read(&inode->i_dio_count));
1837        finish_wait(wq, &q.wait);
1838}
1839
1840/**
1841 * inode_dio_wait - wait for outstanding DIO requests to finish
1842 * @inode: inode to wait for
1843 *
1844 * Waits for all pending direct I/O requests to finish so that we can
1845 * proceed with a truncate or equivalent operation.
1846 *
1847 * Must be called under a lock that serializes taking new references
1848 * to i_dio_count, usually by inode->i_mutex.
1849 */
1850void inode_dio_wait(struct inode *inode)
1851{
1852        if (atomic_read(&inode->i_dio_count))
1853                __inode_dio_wait(inode);
1854}
1855EXPORT_SYMBOL(inode_dio_wait);
1856
1857/*
1858 * inode_dio_done - signal finish of a direct I/O requests
1859 * @inode: inode the direct I/O happens on
1860 *
1861 * This is called once we've finished processing a direct I/O request,
1862 * and is used to wake up callers waiting for direct I/O to be quiesced.
1863 */
1864void inode_dio_done(struct inode *inode)
1865{
1866        if (atomic_dec_and_test(&inode->i_dio_count))
1867                wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1868}
1869EXPORT_SYMBOL(inode_dio_done);
1870
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.