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 "internal.h"
  21
  22/*
  23 * Inode locking rules:
  24 *
  25 * inode->i_lock protects:
  26 *   inode->i_state, inode->i_hash, __iget()
  27 * inode->i_sb->s_inode_lru_lock protects:
  28 *   inode->i_sb->s_inode_lru, inode->i_lru
  29 * inode_sb_list_lock protects:
  30 *   sb->s_inodes, inode->i_sb_list
  31 * bdi->wb.list_lock protects:
  32 *   bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
  33 * inode_hash_lock protects:
  34 *   inode_hashtable, inode->i_hash
  35 *
  36 * Lock ordering:
  37 *
  38 * inode_sb_list_lock
  39 *   inode->i_lock
  40 *     inode->i_sb->s_inode_lru_lock
  41 *
  42 * bdi->wb.list_lock
  43 *   inode->i_lock
  44 *
  45 * inode_hash_lock
  46 *   inode_sb_list_lock
  47 *   inode->i_lock
  48 *
  49 * iunique_lock
  50 *   inode_hash_lock
  51 */
  52
  53static unsigned int i_hash_mask __read_mostly;
  54static unsigned int i_hash_shift __read_mostly;
  55static struct hlist_head *inode_hashtable __read_mostly;
  56static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  57
  58__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
  59
  60/*
  61 * Empty aops. Can be used for the cases where the user does not
  62 * define any of the address_space operations.
  63 */
  64const struct address_space_operations empty_aops = {
  65};
  66EXPORT_SYMBOL(empty_aops);
  67
  68/*
  69 * Statistics gathering..
  70 */
  71struct inodes_stat_t inodes_stat;
  72
  73static DEFINE_PER_CPU(unsigned int, nr_inodes);
  74static DEFINE_PER_CPU(unsigned int, nr_unused);
  75
  76static struct kmem_cache *inode_cachep __read_mostly;
  77
  78static int get_nr_inodes(void)
  79{
  80        int i;
  81        int sum = 0;
  82        for_each_possible_cpu(i)
  83                sum += per_cpu(nr_inodes, i);
  84        return sum < 0 ? 0 : sum;
  85}
  86
  87static inline int get_nr_inodes_unused(void)
  88{
  89        int i;
  90        int sum = 0;
  91        for_each_possible_cpu(i)
  92                sum += per_cpu(nr_unused, i);
  93        return sum < 0 ? 0 : sum;
  94}
  95
  96int get_nr_dirty_inodes(void)
  97{
  98        /* not actually dirty inodes, but a wild approximation */
  99        int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
 100        return nr_dirty > 0 ? nr_dirty : 0;
 101}
 102
 103/*
 104 * Handle nr_inode sysctl
 105 */
 106#ifdef CONFIG_SYSCTL
 107int proc_nr_inodes(ctl_table *table, int write,
 108                   void __user *buffer, size_t *lenp, loff_t *ppos)
 109{
 110        inodes_stat.nr_inodes = get_nr_inodes();
 111        inodes_stat.nr_unused = get_nr_inodes_unused();
 112        return proc_dointvec(table, write, buffer, lenp, ppos);
 113}
 114#endif
 115
 116/**
 117 * inode_init_always - perform inode structure intialisation
 118 * @sb: superblock inode belongs to
 119 * @inode: inode to initialise
 120 *
 121 * These are initializations that need to be done on every inode
 122 * allocation as the fields are not initialised by slab allocation.
 123 */
 124int inode_init_always(struct super_block *sb, struct inode *inode)
 125{
 126        static const struct inode_operations empty_iops;
 127        static const struct file_operations empty_fops;
 128        struct address_space *const mapping = &inode->i_data;
 129
 130        inode->i_sb = sb;
 131        inode->i_blkbits = sb->s_blocksize_bits;
 132        inode->i_flags = 0;
 133        atomic_set(&inode->i_count, 1);
 134        inode->i_op = &empty_iops;
 135        inode->i_fop = &empty_fops;
 136        inode->__i_nlink = 1;
 137        inode->i_opflags = 0;
 138        i_uid_write(inode, 0);
 139        i_gid_write(inode, 0);
 140        atomic_set(&inode->i_writecount, 0);
 141        inode->i_size = 0;
 142        inode->i_blocks = 0;
 143        inode->i_bytes = 0;
 144        inode->i_generation = 0;
 145#ifdef CONFIG_QUOTA
 146        memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
 147#endif
 148        inode->i_pipe = NULL;
 149        inode->i_bdev = NULL;
 150        inode->i_cdev = NULL;
 151        inode->i_rdev = 0;
 152        inode->dirtied_when = 0;
 153
 154        if (security_inode_alloc(inode))
 155                goto out;
 156        spin_lock_init(&inode->i_lock);
 157        lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 158
 159        mutex_init(&inode->i_mutex);
 160        lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
 161
 162        atomic_set(&inode->i_dio_count, 0);
 163
 164        mapping->a_ops = &empty_aops;
 165        mapping->host = inode;
 166        mapping->flags = 0;
 167        mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 168        mapping->private_data = NULL;
 169        mapping->backing_dev_info = &default_backing_dev_info;
 170        mapping->writeback_index = 0;
 171
 172        /*
 173         * If the block_device provides a backing_dev_info for client
 174         * inodes then use that.  Otherwise the inode share the bdev's
 175         * backing_dev_info.
 176         */
 177        if (sb->s_bdev) {
 178                struct backing_dev_info *bdi;
 179
 180                bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
 181                mapping->backing_dev_info = bdi;
 182        }
 183        inode->i_private = NULL;
 184        inode->i_mapping = mapping;
 185        INIT_HLIST_HEAD(&inode->i_dentry);      /* buggered by rcu freeing */
 186#ifdef CONFIG_FS_POSIX_ACL
 187        inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 188#endif
 189
 190#ifdef CONFIG_FSNOTIFY
 191        inode->i_fsnotify_mask = 0;
 192#endif
 193
 194        this_cpu_inc(nr_inodes);
 195
 196        return 0;
 197out:
 198        return -ENOMEM;
 199}
 200EXPORT_SYMBOL(inode_init_always);
 201
 202static struct inode *alloc_inode(struct super_block *sb)
 203{
 204        struct inode *inode;
 205
 206        if (sb->s_op->alloc_inode)
 207                inode = sb->s_op->alloc_inode(sb);
 208        else
 209                inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
 210
 211        if (!inode)
 212                return NULL;
 213
 214        if (unlikely(inode_init_always(sb, inode))) {
 215                if (inode->i_sb->s_op->destroy_inode)
 216                        inode->i_sb->s_op->destroy_inode(inode);
 217                else
 218                        kmem_cache_free(inode_cachep, inode);
 219                return NULL;
 220        }
 221
 222        return inode;
 223}
 224
 225void free_inode_nonrcu(struct inode *inode)
 226{
 227        kmem_cache_free(inode_cachep, inode);
 228}
 229EXPORT_SYMBOL(free_inode_nonrcu);
 230
 231void __destroy_inode(struct inode *inode)
 232{
 233        BUG_ON(inode_has_buffers(inode));
 234        security_inode_free(inode);
 235        fsnotify_inode_delete(inode);
 236        if (!inode->i_nlink) {
 237                WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
 238                atomic_long_dec(&inode->i_sb->s_remove_count);
 239        }
 240
 241#ifdef CONFIG_FS_POSIX_ACL
 242        if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
 243                posix_acl_release(inode->i_acl);
 244        if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
 245                posix_acl_release(inode->i_default_acl);
 246#endif
 247        this_cpu_dec(nr_inodes);
 248}
 249EXPORT_SYMBOL(__destroy_inode);
 250
 251static void i_callback(struct rcu_head *head)
 252{
 253        struct inode *inode = container_of(head, struct inode, i_rcu);
 254        kmem_cache_free(inode_cachep, inode);
 255}
 256
 257static void destroy_inode(struct inode *inode)
 258{
 259        BUG_ON(!list_empty(&inode->i_lru));
 260        __destroy_inode(inode);
 261        if (inode->i_sb->s_op->destroy_inode)
 262                inode->i_sb->s_op->destroy_inode(inode);
 263        else
 264                call_rcu(&inode->i_rcu, i_callback);
 265}
 266
 267/**
 268 * drop_nlink - directly drop an inode's link count
 269 * @inode: inode
 270 *
 271 * This is a low-level filesystem helper to replace any
 272 * direct filesystem manipulation of i_nlink.  In cases
 273 * where we are attempting to track writes to the
 274 * filesystem, a decrement to zero means an imminent
 275 * write when the file is truncated and actually unlinked
 276 * on the filesystem.
 277 */
 278void drop_nlink(struct inode *inode)
 279{
 280        WARN_ON(inode->i_nlink == 0);
 281        inode->__i_nlink--;
 282        if (!inode->i_nlink)
 283                atomic_long_inc(&inode->i_sb->s_remove_count);
 284}
 285EXPORT_SYMBOL(drop_nlink);
 286
 287/**
 288 * clear_nlink - directly zero an inode's link count
 289 * @inode: inode
 290 *
 291 * This is a low-level filesystem helper to replace any
 292 * direct filesystem manipulation of i_nlink.  See
 293 * drop_nlink() for why we care about i_nlink hitting zero.
 294 */
 295void clear_nlink(struct inode *inode)
 296{
 297        if (inode->i_nlink) {
 298                inode->__i_nlink = 0;
 299                atomic_long_inc(&inode->i_sb->s_remove_count);
 300        }
 301}
 302EXPORT_SYMBOL(clear_nlink);
 303
 304/**
 305 * set_nlink - directly set an inode's link count
 306 * @inode: inode
 307 * @nlink: new nlink (should be non-zero)
 308 *
 309 * This is a low-level filesystem helper to replace any
 310 * direct filesystem manipulation of i_nlink.
 311 */
 312void set_nlink(struct inode *inode, unsigned int nlink)
 313{
 314        if (!nlink) {
 315                clear_nlink(inode);
 316        } else {
 317                /* Yes, some filesystems do change nlink from zero to one */
 318                if (inode->i_nlink == 0)
 319                        atomic_long_dec(&inode->i_sb->s_remove_count);
 320
 321                inode->__i_nlink = nlink;
 322        }
 323}
 324EXPORT_SYMBOL(set_nlink);
 325
 326/**
 327 * inc_nlink - directly increment an inode's link count
 328 * @inode: inode
 329 *
 330 * This is a low-level filesystem helper to replace any
 331 * direct filesystem manipulation of i_nlink.  Currently,
 332 * it is only here for parity with dec_nlink().
 333 */
 334void inc_nlink(struct inode *inode)
 335{
 336        if (unlikely(inode->i_nlink == 0)) {
 337                WARN_ON(!(inode->i_state & I_LINKABLE));
 338                atomic_long_dec(&inode->i_sb->s_remove_count);
 339        }
 340
 341        inode->__i_nlink++;
 342}
 343EXPORT_SYMBOL(inc_nlink);
 344
 345void address_space_init_once(struct address_space *mapping)
 346{
 347        memset(mapping, 0, sizeof(*mapping));
 348        INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
 349        spin_lock_init(&mapping->tree_lock);
 350        mutex_init(&mapping->i_mmap_mutex);
 351        INIT_LIST_HEAD(&mapping->private_list);
 352        spin_lock_init(&mapping->private_lock);
 353        mapping->i_mmap = RB_ROOT;
 354        INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
 355}
 356EXPORT_SYMBOL(address_space_init_once);
 357
 358/*
 359 * These are initializations that only need to be done
 360 * once, because the fields are idempotent across use
 361 * of the inode, so let the slab aware of that.
 362 */
 363void inode_init_once(struct inode *inode)
 364{
 365        memset(inode, 0, sizeof(*inode));
 366        INIT_HLIST_NODE(&inode->i_hash);
 367        INIT_LIST_HEAD(&inode->i_devices);
 368        INIT_LIST_HEAD(&inode->i_wb_list);
 369        INIT_LIST_HEAD(&inode->i_lru);
 370        address_space_init_once(&inode->i_data);
 371        i_size_ordered_init(inode);
 372#ifdef CONFIG_FSNOTIFY
 373        INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
 374#endif
 375}
 376EXPORT_SYMBOL(inode_init_once);
 377
 378static void init_once(void *foo)
 379{
 380        struct inode *inode = (struct inode *) foo;
 381
 382        inode_init_once(inode);
 383}
 384
 385/*
 386 * inode->i_lock must be held
 387 */
 388void __iget(struct inode *inode)
 389{
 390        atomic_inc(&inode->i_count);
 391}
 392
 393/*
 394 * get additional reference to inode; caller must already hold one.
 395 */
 396void ihold(struct inode *inode)
 397{
 398        WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 399}
 400EXPORT_SYMBOL(ihold);
 401
 402static void inode_lru_list_add(struct inode *inode)
 403{
 404        spin_lock(&inode->i_sb->s_inode_lru_lock);
 405        if (list_empty(&inode->i_lru)) {
 406                list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
 407                inode->i_sb->s_nr_inodes_unused++;
 408                this_cpu_inc(nr_unused);
 409        }
 410        spin_unlock(&inode->i_sb->s_inode_lru_lock);
 411}
 412
 413/*
 414 * Add inode to LRU if needed (inode is unused and clean).
 415 *
 416 * Needs inode->i_lock held.
 417 */
 418void inode_add_lru(struct inode *inode)
 419{
 420        if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
 421            !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
 422                inode_lru_list_add(inode);
 423}
 424
 425
 426static void inode_lru_list_del(struct inode *inode)
 427{
 428        spin_lock(&inode->i_sb->s_inode_lru_lock);
 429        if (!list_empty(&inode->i_lru)) {
 430                list_del_init(&inode->i_lru);
 431                inode->i_sb->s_nr_inodes_unused--;
 432                this_cpu_dec(nr_unused);
 433        }
 434        spin_unlock(&inode->i_sb->s_inode_lru_lock);
 435}
 436
 437/**
 438 * inode_sb_list_add - add inode to the superblock list of inodes
 439 * @inode: inode to add
 440 */
 441void inode_sb_list_add(struct inode *inode)
 442{
 443        spin_lock(&inode_sb_list_lock);
 444        list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 445        spin_unlock(&inode_sb_list_lock);
 446}
 447EXPORT_SYMBOL_GPL(inode_sb_list_add);
 448
 449static inline void inode_sb_list_del(struct inode *inode)
 450{
 451        if (!list_empty(&inode->i_sb_list)) {
 452                spin_lock(&inode_sb_list_lock);
 453                list_del_init(&inode->i_sb_list);
 454                spin_unlock(&inode_sb_list_lock);
 455        }
 456}
 457
 458static unsigned long hash(struct super_block *sb, unsigned long hashval)
 459{
 460        unsigned long tmp;
 461
 462        tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 463                        L1_CACHE_BYTES;
 464        tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 465        return tmp & i_hash_mask;
 466}
 467
 468/**
 469 *      __insert_inode_hash - hash an inode
 470 *      @inode: unhashed inode
 471 *      @hashval: unsigned long value used to locate this object in the
 472 *              inode_hashtable.
 473 *
 474 *      Add an inode to the inode hash for this superblock.
 475 */
 476void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 477{
 478        struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 479
 480        spin_lock(&inode_hash_lock);
 481        spin_lock(&inode->i_lock);
 482        hlist_add_head(&inode->i_hash, b);
 483        spin_unlock(&inode->i_lock);
 484        spin_unlock(&inode_hash_lock);
 485}
 486EXPORT_SYMBOL(__insert_inode_hash);
 487
 488/**
 489 *      __remove_inode_hash - remove an inode from the hash
 490 *      @inode: inode to unhash
 491 *
 492 *      Remove an inode from the superblock.
 493 */
 494void __remove_inode_hash(struct inode *inode)
 495{
 496        spin_lock(&inode_hash_lock);
 497        spin_lock(&inode->i_lock);
 498        hlist_del_init(&inode->i_hash);
 499        spin_unlock(&inode->i_lock);
 500        spin_unlock(&inode_hash_lock);
 501}
 502EXPORT_SYMBOL(__remove_inode_hash);
 503
 504void clear_inode(struct inode *inode)
 505{
 506        might_sleep();
 507        /*
 508         * We have to cycle tree_lock here because reclaim can be still in the
 509         * process of removing the last page (in __delete_from_page_cache())
 510         * and we must not free mapping under it.
 511         */
 512        spin_lock_irq(&inode->i_data.tree_lock);
 513        BUG_ON(inode->i_data.nrpages);
 514        spin_unlock_irq(&inode->i_data.tree_lock);
 515        BUG_ON(!list_empty(&inode->i_data.private_list));
 516        BUG_ON(!(inode->i_state & I_FREEING));
 517        BUG_ON(inode->i_state & I_CLEAR);
 518        /* don't need i_lock here, no concurrent mods to i_state */
 519        inode->i_state = I_FREEING | I_CLEAR;
 520}
 521EXPORT_SYMBOL(clear_inode);
 522
 523/*
 524 * Free the inode passed in, removing it from the lists it is still connected
 525 * to. We remove any pages still attached to the inode and wait for any IO that
 526 * is still in progress before finally destroying the inode.
 527 *
 528 * An inode must already be marked I_FREEING so that we avoid the inode being
 529 * moved back onto lists if we race with other code that manipulates the lists
 530 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 531 *
 532 * An inode must already be removed from the LRU list before being evicted from
 533 * the cache. This should occur atomically with setting the I_FREEING state
 534 * flag, so no inodes here should ever be on the LRU when being evicted.
 535 */
 536static void evict(struct inode *inode)
 537{
 538        const struct super_operations *op = inode->i_sb->s_op;
 539
 540        BUG_ON(!(inode->i_state & I_FREEING));
 541        BUG_ON(!list_empty(&inode->i_lru));
 542
 543        if (!list_empty(&inode->i_wb_list))
 544                inode_wb_list_del(inode);
 545
 546        inode_sb_list_del(inode);
 547
 548        /*
 549         * Wait for flusher thread to be done with the inode so that filesystem
 550         * does not start destroying it while writeback is still running. Since
 551         * the inode has I_FREEING set, flusher thread won't start new work on
 552         * the inode.  We just have to wait for running writeback to finish.
 553         */
 554        inode_wait_for_writeback(inode);
 555
 556        if (op->evict_inode) {
 557                op->evict_inode(inode);
 558        } else {
 559                if (inode->i_data.nrpages)
 560                        truncate_inode_pages(&inode->i_data, 0);
 561                clear_inode(inode);
 562        }
 563        if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 564                bd_forget(inode);
 565        if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 566                cd_forget(inode);
 567
 568        remove_inode_hash(inode);
 569
 570        spin_lock(&inode->i_lock);
 571        wake_up_bit(&inode->i_state, __I_NEW);
 572        BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 573        spin_unlock(&inode->i_lock);
 574
 575        destroy_inode(inode);
 576}
 577
 578/*
 579 * dispose_list - dispose of the contents of a local list
 580 * @head: the head of the list to free
 581 *
 582 * Dispose-list gets a local list with local inodes in it, so it doesn't
 583 * need to worry about list corruption and SMP locks.
 584 */
 585static void dispose_list(struct list_head *head)
 586{
 587        while (!list_empty(head)) {
 588                struct inode *inode;
 589
 590                inode = list_first_entry(head, struct inode, i_lru);
 591                list_del_init(&inode->i_lru);
 592
 593                evict(inode);
 594        }
 595}
 596
 597/**
 598 * evict_inodes - evict all evictable inodes for a superblock
 599 * @sb:         superblock to operate on
 600 *
 601 * Make sure that no inodes with zero refcount are retained.  This is
 602 * called by superblock shutdown after having MS_ACTIVE flag removed,
 603 * so any inode reaching zero refcount during or after that call will
 604 * be immediately evicted.
 605 */
 606void evict_inodes(struct super_block *sb)
 607{
 608        struct inode *inode, *next;
 609        LIST_HEAD(dispose);
 610
 611        spin_lock(&inode_sb_list_lock);
 612        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 613                if (atomic_read(&inode->i_count))
 614                        continue;
 615
 616                spin_lock(&inode->i_lock);
 617                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 618                        spin_unlock(&inode->i_lock);
 619                        continue;
 620                }
 621
 622                inode->i_state |= I_FREEING;
 623                inode_lru_list_del(inode);
 624                spin_unlock(&inode->i_lock);
 625                list_add(&inode->i_lru, &dispose);
 626        }
 627        spin_unlock(&inode_sb_list_lock);
 628
 629        dispose_list(&dispose);
 630}
 631
 632/**
 633 * invalidate_inodes    - attempt to free all inodes on a superblock
 634 * @sb:         superblock to operate on
 635 * @kill_dirty: flag to guide handling of dirty inodes
 636 *
 637 * Attempts to free all inodes for a given superblock.  If there were any
 638 * busy inodes return a non-zero value, else zero.
 639 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 640 * them as busy.
 641 */
 642int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 643{
 644        int busy = 0;
 645        struct inode *inode, *next;
 646        LIST_HEAD(dispose);
 647
 648        spin_lock(&inode_sb_list_lock);
 649        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 650                spin_lock(&inode->i_lock);
 651                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 652                        spin_unlock(&inode->i_lock);
 653                        continue;
 654                }
 655                if (inode->i_state & I_DIRTY && !kill_dirty) {
 656                        spin_unlock(&inode->i_lock);
 657                        busy = 1;
 658                        continue;
 659                }
 660                if (atomic_read(&inode->i_count)) {
 661                        spin_unlock(&inode->i_lock);
 662                        busy = 1;
 663                        continue;
 664                }
 665
 666                inode->i_state |= I_FREEING;
 667                inode_lru_list_del(inode);
 668                spin_unlock(&inode->i_lock);
 669                list_add(&inode->i_lru, &dispose);
 670        }
 671        spin_unlock(&inode_sb_list_lock);
 672
 673        dispose_list(&dispose);
 674
 675        return busy;
 676}
 677
 678static int can_unuse(struct inode *inode)
 679{
 680        if (inode->i_state & ~I_REFERENCED)
 681                return 0;
 682        if (inode_has_buffers(inode))
 683                return 0;
 684        if (atomic_read(&inode->i_count))
 685                return 0;
 686        if (inode->i_data.nrpages)
 687                return 0;
 688        return 1;
 689}
 690
 691/*
 692 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 693 * This is called from the superblock shrinker function with a number of inodes
 694 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 695 * then are freed outside inode_lock by dispose_list().
 696 *
 697 * Any inodes which are pinned purely because of attached pagecache have their
 698 * pagecache removed.  If the inode has metadata buffers attached to
 699 * mapping->private_list then try to remove them.
 700 *
 701 * If the inode has the I_REFERENCED flag set, then it means that it has been
 702 * used recently - the flag is set in iput_final(). When we encounter such an
 703 * inode, clear the flag and move it to the back of the LRU so it gets another
 704 * pass through the LRU before it gets reclaimed. This is necessary because of
 705 * the fact we are doing lazy LRU updates to minimise lock contention so the
 706 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 707 * with this flag set because they are the inodes that are out of order.
 708 */
 709void prune_icache_sb(struct super_block *sb, int nr_to_scan)
 710{
 711        LIST_HEAD(freeable);
 712        int nr_scanned;
 713        unsigned long reap = 0;
 714
 715        spin_lock(&sb->s_inode_lru_lock);
 716        for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
 717                struct inode *inode;
 718
 719                if (list_empty(&sb->s_inode_lru))
 720                        break;
 721
 722                inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
 723
 724                /*
 725                 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
 726                 * so use a trylock. If we fail to get the lock, just move the
 727                 * inode to the back of the list so we don't spin on it.
 728                 */
 729                if (!spin_trylock(&inode->i_lock)) {
 730                        list_move(&inode->i_lru, &sb->s_inode_lru);
 731                        continue;
 732                }
 733
 734                /*
 735                 * Referenced or dirty inodes are still in use. Give them
 736                 * another pass through the LRU as we canot reclaim them now.
 737                 */
 738                if (atomic_read(&inode->i_count) ||
 739                    (inode->i_state & ~I_REFERENCED)) {
 740                        list_del_init(&inode->i_lru);
 741                        spin_unlock(&inode->i_lock);
 742                        sb->s_nr_inodes_unused--;
 743                        this_cpu_dec(nr_unused);
 744                        continue;
 745                }
 746
 747                /* recently referenced inodes get one more pass */
 748                if (inode->i_state & I_REFERENCED) {
 749                        inode->i_state &= ~I_REFERENCED;
 750                        list_move(&inode->i_lru, &sb->s_inode_lru);
 751                        spin_unlock(&inode->i_lock);
 752                        continue;
 753                }
 754                if (inode_has_buffers(inode) || inode->i_data.nrpages) {
 755                        __iget(inode);
 756                        spin_unlock(&inode->i_lock);
 757                        spin_unlock(&sb->s_inode_lru_lock);
 758                        if (remove_inode_buffers(inode))
 759                                reap += invalidate_mapping_pages(&inode->i_data,
 760                                                                0, -1);
 761                        iput(inode);
 762                        spin_lock(&sb->s_inode_lru_lock);
 763
 764                        if (inode != list_entry(sb->s_inode_lru.next,
 765                                                struct inode, i_lru))
 766                                continue;       /* wrong inode or list_empty */
 767                        /* avoid lock inversions with trylock */
 768                        if (!spin_trylock(&inode->i_lock))
 769                                continue;
 770                        if (!can_unuse(inode)) {
 771                                spin_unlock(&inode->i_lock);
 772                                continue;
 773                        }
 774                }
 775                WARN_ON(inode->i_state & I_NEW);
 776                inode->i_state |= I_FREEING;
 777                spin_unlock(&inode->i_lock);
 778
 779                list_move(&inode->i_lru, &freeable);
 780                sb->s_nr_inodes_unused--;
 781                this_cpu_dec(nr_unused);
 782        }
 783        if (current_is_kswapd())
 784                __count_vm_events(KSWAPD_INODESTEAL, reap);
 785        else
 786                __count_vm_events(PGINODESTEAL, reap);
 787        spin_unlock(&sb->s_inode_lru_lock);
 788        if (current->reclaim_state)
 789                current->reclaim_state->reclaimed_slab += reap;
 790
 791        dispose_list(&freeable);
 792}
 793
 794static void __wait_on_freeing_inode(struct inode *inode);
 795/*
 796 * Called with the inode lock held.
 797 */
 798static struct inode *find_inode(struct super_block *sb,
 799                                struct hlist_head *head,
 800                                int (*test)(struct inode *, void *),
 801                                void *data)
 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_sb != sb) {
 809                        spin_unlock(&inode->i_lock);
 810                        continue;
 811                }
 812                if (!test(inode, data)) {
 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 * find_inode_fast is the fast path version of find_inode, see the comment at
 829 * iget_locked for details.
 830 */
 831static struct inode *find_inode_fast(struct super_block *sb,
 832                                struct hlist_head *head, unsigned long ino)
 833{
 834        struct inode *inode = NULL;
 835
 836repeat:
 837        hlist_for_each_entry(inode, head, i_hash) {
 838                spin_lock(&inode->i_lock);
 839                if (inode->i_ino != ino) {
 840                        spin_unlock(&inode->i_lock);
 841                        continue;
 842                }
 843                if (inode->i_sb != sb) {
 844                        spin_unlock(&inode->i_lock);
 845                        continue;
 846                }
 847                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 848                        __wait_on_freeing_inode(inode);
 849                        goto repeat;
 850                }
 851                __iget(inode);
 852                spin_unlock(&inode->i_lock);
 853                return inode;
 854        }
 855        return NULL;
 856}
 857
 858/*
 859 * Each cpu owns a range of LAST_INO_BATCH numbers.
 860 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 861 * to renew the exhausted range.
 862 *
 863 * This does not significantly increase overflow rate because every CPU can
 864 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 865 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 866 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 867 * overflow rate by 2x, which does not seem too significant.
 868 *
 869 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 870 * error if st_ino won't fit in target struct field. Use 32bit counter
 871 * here to attempt to avoid that.
 872 */
 873#define LAST_INO_BATCH 1024
 874static DEFINE_PER_CPU(unsigned int, last_ino);
 875
 876unsigned int get_next_ino(void)
 877{
 878        unsigned int *p = &get_cpu_var(last_ino);
 879        unsigned int res = *p;
 880
 881#ifdef CONFIG_SMP
 882        if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 883                static atomic_t shared_last_ino;
 884                int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 885
 886                res = next - LAST_INO_BATCH;
 887        }
 888#endif
 889
 890        *p = ++res;
 891        put_cpu_var(last_ino);
 892        return res;
 893}
 894EXPORT_SYMBOL(get_next_ino);
 895
 896/**
 897 *      new_inode_pseudo        - obtain an inode
 898 *      @sb: superblock
 899 *
 900 *      Allocates a new inode for given superblock.
 901 *      Inode wont be chained in superblock s_inodes list
 902 *      This means :
 903 *      - fs can't be unmount
 904 *      - quotas, fsnotify, writeback can't work
 905 */
 906struct inode *new_inode_pseudo(struct super_block *sb)
 907{
 908        struct inode *inode = alloc_inode(sb);
 909
 910        if (inode) {
 911                spin_lock(&inode->i_lock);
 912                inode->i_state = 0;
 913                spin_unlock(&inode->i_lock);
 914                INIT_LIST_HEAD(&inode->i_sb_list);
 915        }
 916        return inode;
 917}
 918
 919/**
 920 *      new_inode       - obtain an inode
 921 *      @sb: superblock
 922 *
 923 *      Allocates a new inode for given superblock. The default gfp_mask
 924 *      for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
 925 *      If HIGHMEM pages are unsuitable or it is known that pages allocated
 926 *      for the page cache are not reclaimable or migratable,
 927 *      mapping_set_gfp_mask() must be called with suitable flags on the
 928 *      newly created inode's mapping
 929 *
 930 */
 931struct inode *new_inode(struct super_block *sb)
 932{
 933        struct inode *inode;
 934
 935        spin_lock_prefetch(&inode_sb_list_lock);
 936
 937        inode = new_inode_pseudo(sb);
 938        if (inode)
 939                inode_sb_list_add(inode);
 940        return inode;
 941}
 942EXPORT_SYMBOL(new_inode);
 943
 944#ifdef CONFIG_DEBUG_LOCK_ALLOC
 945void lockdep_annotate_inode_mutex_key(struct inode *inode)
 946{
 947        if (S_ISDIR(inode->i_mode)) {
 948                struct file_system_type *type = inode->i_sb->s_type;
 949
 950                /* Set new key only if filesystem hasn't already changed it */
 951                if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
 952                        /*
 953                         * ensure nobody is actually holding i_mutex
 954                         */
 955                        mutex_destroy(&inode->i_mutex);
 956                        mutex_init(&inode->i_mutex);
 957                        lockdep_set_class(&inode->i_mutex,
 958                                          &type->i_mutex_dir_key);
 959                }
 960        }
 961}
 962EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
 963#endif
 964
 965/**
 966 * unlock_new_inode - clear the I_NEW state and wake up any waiters
 967 * @inode:      new inode to unlock
 968 *
 969 * Called when the inode is fully initialised to clear the new state of the
 970 * inode and wake up anyone waiting for the inode to finish initialisation.
 971 */
 972void unlock_new_inode(struct inode *inode)
 973{
 974        lockdep_annotate_inode_mutex_key(inode);
 975        spin_lock(&inode->i_lock);
 976        WARN_ON(!(inode->i_state & I_NEW));
 977        inode->i_state &= ~I_NEW;
 978        smp_mb();
 979        wake_up_bit(&inode->i_state, __I_NEW);
 980        spin_unlock(&inode->i_lock);
 981}
 982EXPORT_SYMBOL(unlock_new_inode);
 983
 984/**
 985 * iget5_locked - obtain an inode from a mounted file system
 986 * @sb:         super block of file system
 987 * @hashval:    hash value (usually inode number) to get
 988 * @test:       callback used for comparisons between inodes
 989 * @set:        callback used to initialize a new struct inode
 990 * @data:       opaque data pointer to pass to @test and @set
 991 *
 992 * Search for the inode specified by @hashval and @data in the inode cache,
 993 * and if present it is return it with an increased reference count. This is
 994 * a generalized version of iget_locked() for file systems where the inode
 995 * number is not sufficient for unique identification of an inode.
 996 *
 997 * If the inode is not in cache, allocate a new inode and return it locked,
 998 * hashed, and with the I_NEW flag set. The file system gets to fill it in
 999 * before unlocking it via unlock_new_inode().
1000 *
1001 * Note both @test and @set are called with the inode_hash_lock held, so can't
1002 * sleep.
1003 */
1004struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1005                int (*test)(struct inode *, void *),
1006                int (*set)(struct inode *, void *), void *data)
1007{
1008        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1009        struct inode *inode;
1010
1011        spin_lock(&inode_hash_lock);
1012        inode = find_inode(sb, head, test, data);
1013        spin_unlock(&inode_hash_lock);
1014
1015        if (inode) {
1016                wait_on_inode(inode);
1017                return inode;
1018        }
1019
1020        inode = alloc_inode(sb);
1021        if (inode) {
1022                struct inode *old;
1023
1024                spin_lock(&inode_hash_lock);
1025                /* We released the lock, so.. */
1026                old = find_inode(sb, head, test, data);
1027                if (!old) {
1028                        if (set(inode, data))
1029                                goto set_failed;
1030
1031                        spin_lock(&inode->i_lock);
1032                        inode->i_state = I_NEW;
1033                        hlist_add_head(&inode->i_hash, head);
1034                        spin_unlock(&inode->i_lock);
1035                        inode_sb_list_add(inode);
1036                        spin_unlock(&inode_hash_lock);
1037
1038                        /* Return the locked inode with I_NEW set, the
1039                         * caller is responsible for filling in the contents
1040                         */
1041                        return inode;
1042                }
1043
1044                /*
1045                 * Uhhuh, somebody else created the same inode under
1046                 * us. Use the old inode instead of the one we just
1047                 * allocated.
1048                 */
1049                spin_unlock(&inode_hash_lock);
1050                destroy_inode(inode);
1051                inode = old;
1052                wait_on_inode(inode);
1053        }
1054        return inode;
1055
1056set_failed:
1057        spin_unlock(&inode_hash_lock);
1058        destroy_inode(inode);
1059        return NULL;
1060}
1061EXPORT_SYMBOL(iget5_locked);
1062
1063/**
1064 * iget_locked - obtain an inode from a mounted file system
1065 * @sb:         super block of file system
1066 * @ino:        inode number to get
1067 *
1068 * Search for the inode specified by @ino in the inode cache and if present
1069 * return it with an increased reference count. This is for file systems
1070 * where the inode number is sufficient for unique identification of an inode.
1071 *
1072 * If the inode is not in cache, allocate a new inode and return it locked,
1073 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1074 * before unlocking it via unlock_new_inode().
1075 */
1076struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1077{
1078        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1079        struct inode *inode;
1080
1081        spin_lock(&inode_hash_lock);
1082        inode = find_inode_fast(sb, head, ino);
1083        spin_unlock(&inode_hash_lock);
1084        if (inode) {
1085                wait_on_inode(inode);
1086                return inode;
1087        }
1088
1089        inode = alloc_inode(sb);
1090        if (inode) {
1091                struct inode *old;
1092
1093                spin_lock(&inode_hash_lock);
1094                /* We released the lock, so.. */
1095                old = find_inode_fast(sb, head, ino);
1096                if (!old) {
1097                        inode->i_ino = ino;
1098                        spin_lock(&inode->i_lock);
1099                        inode->i_state = I_NEW;
1100                        hlist_add_head(&inode->i_hash, head);
1101                        spin_unlock(&inode->i_lock);
1102                        inode_sb_list_add(inode);
1103                        spin_unlock(&inode_hash_lock);
1104
1105                        /* Return the locked inode with I_NEW set, the
1106                         * caller is responsible for filling in the contents
1107                         */
1108                        return inode;
1109                }
1110
1111                /*
1112                 * Uhhuh, somebody else created the same inode under
1113                 * us. Use the old inode instead of the one we just
1114                 * allocated.
1115                 */
1116                spin_unlock(&inode_hash_lock);
1117                destroy_inode(inode);
1118                inode = old;
1119                wait_on_inode(inode);
1120        }
1121        return inode;
1122}
1123EXPORT_SYMBOL(iget_locked);
1124
1125/*
1126 * search the inode cache for a matching inode number.
1127 * If we find one, then the inode number we are trying to
1128 * allocate is not unique and so we should not use it.
1129 *
1130 * Returns 1 if the inode number is unique, 0 if it is not.
1131 */
1132static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1133{
1134        struct hlist_head *b = inode_hashtable + hash(sb, ino);
1135        struct inode *inode;
1136
1137        spin_lock(&inode_hash_lock);
1138        hlist_for_each_entry(inode, b, i_hash) {
1139                if (inode->i_ino == ino && inode->i_sb == sb) {
1140                        spin_unlock(&inode_hash_lock);
1141                        return 0;
1142                }
1143        }
1144        spin_unlock(&inode_hash_lock);
1145
1146        return 1;
1147}
1148
1149/**
1150 *      iunique - get a unique inode number
1151 *      @sb: superblock
1152 *      @max_reserved: highest reserved inode number
1153 *
1154 *      Obtain an inode number that is unique on the system for a given
1155 *      superblock. This is used by file systems that have no natural
1156 *      permanent inode numbering system. An inode number is returned that
1157 *      is higher than the reserved limit but unique.
1158 *
1159 *      BUGS:
1160 *      With a large number of inodes live on the file system this function
1161 *      currently becomes quite slow.
1162 */
1163ino_t iunique(struct super_block *sb, ino_t max_reserved)
1164{
1165        /*
1166         * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1167         * error if st_ino won't fit in target struct field. Use 32bit counter
1168         * here to attempt to avoid that.
1169         */
1170        static DEFINE_SPINLOCK(iunique_lock);
1171        static unsigned int counter;
1172        ino_t res;
1173
1174        spin_lock(&iunique_lock);
1175        do {
1176                if (counter <= max_reserved)
1177                        counter = max_reserved + 1;
1178                res = counter++;
1179        } while (!test_inode_iunique(sb, res));
1180        spin_unlock(&iunique_lock);
1181
1182        return res;
1183}
1184EXPORT_SYMBOL(iunique);
1185
1186struct inode *igrab(struct inode *inode)
1187{
1188        spin_lock(&inode->i_lock);
1189        if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1190                __iget(inode);
1191                spin_unlock(&inode->i_lock);
1192        } else {
1193                spin_unlock(&inode->i_lock);
1194                /*
1195                 * Handle the case where s_op->clear_inode is not been
1196                 * called yet, and somebody is calling igrab
1197                 * while the inode is getting freed.
1198                 */
1199                inode = NULL;
1200        }
1201        return inode;
1202}
1203EXPORT_SYMBOL(igrab);
1204
1205/**
1206 * ilookup5_nowait - search for an inode in the inode cache
1207 * @sb:         super block of file system to search
1208 * @hashval:    hash value (usually inode number) to search for
1209 * @test:       callback used for comparisons between inodes
1210 * @data:       opaque data pointer to pass to @test
1211 *
1212 * Search for the inode specified by @hashval and @data in the inode cache.
1213 * If the inode is in the cache, the inode is returned with an incremented
1214 * reference count.
1215 *
1216 * Note: I_NEW is not waited upon so you have to be very careful what you do
1217 * with the returned inode.  You probably should be using ilookup5() instead.
1218 *
1219 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1220 */
1221struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1222                int (*test)(struct inode *, void *), void *data)
1223{
1224        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1225        struct inode *inode;
1226
1227        spin_lock(&inode_hash_lock);
1228        inode = find_inode(sb, head, test, data);
1229        spin_unlock(&inode_hash_lock);
1230
1231        return inode;
1232}
1233EXPORT_SYMBOL(ilookup5_nowait);
1234
1235/**
1236 * ilookup5 - search for an inode in the inode cache
1237 * @sb:         super block of file system to search
1238 * @hashval:    hash value (usually inode number) to search for
1239 * @test:       callback used for comparisons between inodes
1240 * @data:       opaque data pointer to pass to @test
1241 *
1242 * Search for the inode specified by @hashval and @data in the inode cache,
1243 * and if the inode is in the cache, return the inode with an incremented
1244 * reference count.  Waits on I_NEW before returning the inode.
1245 * returned with an incremented reference count.
1246 *
1247 * This is a generalized version of ilookup() for file systems where the
1248 * inode number is not sufficient for unique identification of an inode.
1249 *
1250 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1251 */
1252struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1253                int (*test)(struct inode *, void *), void *data)
1254{
1255        struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1256
1257        if (inode)
1258                wait_on_inode(inode);
1259        return inode;
1260}
1261EXPORT_SYMBOL(ilookup5);
1262
1263/**
1264 * ilookup - search for an inode in the inode cache
1265 * @sb:         super block of file system to search
1266 * @ino:        inode number to search for
1267 *
1268 * Search for the inode @ino in the inode cache, and if the inode is in the
1269 * cache, the inode is returned with an incremented reference count.
1270 */
1271struct inode *ilookup(struct super_block *sb, unsigned long ino)
1272{
1273        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1274        struct inode *inode;
1275
1276        spin_lock(&inode_hash_lock);
1277        inode = find_inode_fast(sb, head, ino);
1278        spin_unlock(&inode_hash_lock);
1279
1280        if (inode)
1281                wait_on_inode(inode);
1282        return inode;
1283}
1284EXPORT_SYMBOL(ilookup);
1285
1286int insert_inode_locked(struct inode *inode)
1287{
1288        struct super_block *sb = inode->i_sb;
1289        ino_t ino = inode->i_ino;
1290        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1291
1292        while (1) {
1293                struct inode *old = NULL;
1294                spin_lock(&inode_hash_lock);
1295                hlist_for_each_entry(old, head, i_hash) {
1296                        if (old->i_ino != ino)
1297                                continue;
1298                        if (old->i_sb != sb)
1299                                continue;
1300                        spin_lock(&old->i_lock);
1301                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1302                                spin_unlock(&old->i_lock);
1303                                continue;
1304                        }
1305                        break;
1306                }
1307                if (likely(!old)) {
1308                        spin_lock(&inode->i_lock);
1309                        inode->i_state |= I_NEW;
1310                        hlist_add_head(&inode->i_hash, head);
1311                        spin_unlock(&inode->i_lock);
1312                        spin_unlock(&inode_hash_lock);
1313                        return 0;
1314                }
1315                __iget(old);
1316                spin_unlock(&old->i_lock);
1317                spin_unlock(&inode_hash_lock);
1318                wait_on_inode(old);
1319                if (unlikely(!inode_unhashed(old))) {
1320                        iput(old);
1321                        return -EBUSY;
1322                }
1323                iput(old);
1324        }
1325}
1326EXPORT_SYMBOL(insert_inode_locked);
1327
1328int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1329                int (*test)(struct inode *, void *), void *data)
1330{
1331        struct super_block *sb = inode->i_sb;
1332        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1333
1334        while (1) {
1335                struct inode *old = NULL;
1336
1337                spin_lock(&inode_hash_lock);
1338                hlist_for_each_entry(old, head, i_hash) {
1339                        if (old->i_sb != sb)
1340                                continue;
1341                        if (!test(old, data))
1342                                continue;
1343                        spin_lock(&old->i_lock);
1344                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1345                                spin_unlock(&old->i_lock);
1346                                continue;
1347                        }
1348                        break;
1349                }
1350                if (likely(!old)) {
1351                        spin_lock(&inode->i_lock);
1352                        inode->i_state |= I_NEW;
1353                        hlist_add_head(&inode->i_hash, head);
1354                        spin_unlock(&inode->i_lock);
1355                        spin_unlock(&inode_hash_lock);
1356                        return 0;
1357                }
1358                __iget(old);
1359                spin_unlock(&old->i_lock);
1360                spin_unlock(&inode_hash_lock);
1361                wait_on_inode(old);
1362                if (unlikely(!inode_unhashed(old))) {
1363                        iput(old);
1364                        return -EBUSY;
1365                }
1366                iput(old);
1367        }
1368}
1369EXPORT_SYMBOL(insert_inode_locked4);
1370
1371
1372int generic_delete_inode(struct inode *inode)
1373{
1374        return 1;
1375}
1376EXPORT_SYMBOL(generic_delete_inode);
1377
1378/*
1379 * Called when we're dropping the last reference
1380 * to an inode.
1381 *
1382 * Call the FS "drop_inode()" function, defaulting to
1383 * the legacy UNIX filesystem behaviour.  If it tells
1384 * us to evict inode, do so.  Otherwise, retain inode
1385 * in cache if fs is alive, sync and evict if fs is
1386 * shutting down.
1387 */
1388static void iput_final(struct inode *inode)
1389{
1390        struct super_block *sb = inode->i_sb;
1391        const struct super_operations *op = inode->i_sb->s_op;
1392        int drop;
1393
1394        WARN_ON(inode->i_state & I_NEW);
1395
1396        if (op->drop_inode)
1397                drop = op->drop_inode(inode);
1398        else
1399                drop = generic_drop_inode(inode);
1400
1401        if (!drop && (sb->s_flags & MS_ACTIVE)) {
1402                inode->i_state |= I_REFERENCED;
1403                inode_add_lru(inode);
1404                spin_unlock(&inode->i_lock);
1405                return;
1406        }
1407
1408        if (!drop) {
1409                inode->i_state |= I_WILL_FREE;
1410                spin_unlock(&inode->i_lock);
1411                write_inode_now(inode, 1);
1412                spin_lock(&inode->i_lock);
1413                WARN_ON(inode->i_state & I_NEW);
1414                inode->i_state &= ~I_WILL_FREE;
1415        }
1416
1417        inode->i_state |= I_FREEING;
1418        if (!list_empty(&inode->i_lru))
1419                inode_lru_list_del(inode);
1420        spin_unlock(&inode->i_lock);
1421
1422        evict(inode);
1423}
1424
1425/**
1426 *      iput    - put an inode
1427 *      @inode: inode to put
1428 *
1429 *      Puts an inode, dropping its usage count. If the inode use count hits
1430 *      zero, the inode is then freed and may also be destroyed.
1431 *
1432 *      Consequently, iput() can sleep.
1433 */
1434void iput(struct inode *inode)
1435{
1436        if (inode) {
1437                BUG_ON(inode->i_state & I_CLEAR);
1438
1439                if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1440                        iput_final(inode);
1441        }
1442}
1443EXPORT_SYMBOL(iput);
1444
1445/**
1446 *      bmap    - find a block number in a file
1447 *      @inode: inode of file
1448 *      @block: block to find
1449 *
1450 *      Returns the block number on the device holding the inode that
1451 *      is the disk block number for the block of the file requested.
1452 *      That is, asked for block 4 of inode 1 the function will return the
1453 *      disk block relative to the disk start that holds that block of the
1454 *      file.
1455 */
1456sector_t bmap(struct inode *inode, sector_t block)
1457{
1458        sector_t res = 0;
1459        if (inode->i_mapping->a_ops->bmap)
1460                res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1461        return res;
1462}
1463EXPORT_SYMBOL(bmap);
1464
1465/*
1466 * With relative atime, only update atime if the previous atime is
1467 * earlier than either the ctime or mtime or if at least a day has
1468 * passed since the last atime update.
1469 */
1470static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1471                             struct timespec now)
1472{
1473
1474        if (!(mnt->mnt_flags & MNT_RELATIME))
1475                return 1;
1476        /*
1477         * Is mtime younger than atime? If yes, update atime:
1478         */
1479        if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1480                return 1;
1481        /*
1482         * Is ctime younger than atime? If yes, update atime:
1483         */
1484        if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1485                return 1;
1486
1487        /*
1488         * Is the previous atime value older than a day? If yes,
1489         * update atime:
1490         */
1491        if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1492                return 1;
1493        /*
1494         * Good, we can skip the atime update:
1495         */
1496        return 0;
1497}
1498
1499/*
1500 * This does the actual work of updating an inodes time or version.  Must have
1501 * had called mnt_want_write() before calling this.
1502 */
1503static int update_time(struct inode *inode, struct timespec *time, int flags)
1504{
1505        if (inode->i_op->update_time)
1506                return inode->i_op->update_time(inode, time, flags);
1507
1508        if (flags & S_ATIME)
1509                inode->i_atime = *time;
1510        if (flags & S_VERSION)
1511                inode_inc_iversion(inode);
1512        if (flags & S_CTIME)
1513                inode->i_ctime = *time;
1514        if (flags & S_MTIME)
1515                inode->i_mtime = *time;
1516        mark_inode_dirty_sync(inode);
1517        return 0;
1518}
1519
1520/**
1521 *      touch_atime     -       update the access time
1522 *      @path: the &struct path to update
1523 *
1524 *      Update the accessed time on an inode and mark it for writeback.
1525 *      This function automatically handles read only file systems and media,
1526 *      as well as the "noatime" flag and inode specific "noatime" markers.
1527 */
1528void touch_atime(struct path *path)
1529{
1530        struct vfsmount *mnt = path->mnt;
1531        struct inode *inode = path->dentry->d_inode;
1532        struct timespec now;
1533
1534        if (inode->i_flags & S_NOATIME)
1535                return;
1536        if (IS_NOATIME(inode))
1537                return;
1538        if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1539                return;
1540
1541        if (mnt->mnt_flags & MNT_NOATIME)
1542                return;
1543        if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1544                return;
1545
1546        now = current_fs_time(inode->i_sb);
1547
1548        if (!relatime_need_update(mnt, inode, now))
1549                return;
1550
1551        if (timespec_equal(&inode->i_atime, &now))
1552                return;
1553
1554        if (!sb_start_write_trylock(inode->i_sb))
1555                return;
1556
1557        if (__mnt_want_write(mnt))
1558                goto skip_update;
1559        /*
1560         * File systems can error out when updating inodes if they need to
1561         * allocate new space to modify an inode (such is the case for
1562         * Btrfs), but since we touch atime while walking down the path we
1563         * really don't care if we failed to update the atime of the file,
1564         * so just ignore the return value.
1565         * We may also fail on filesystems that have the ability to make parts
1566         * of the fs read only, e.g. subvolumes in Btrfs.
1567         */
1568        update_time(inode, &now, S_ATIME);
1569        __mnt_drop_write(mnt);
1570skip_update:
1571        sb_end_write(inode->i_sb);
1572}
1573EXPORT_SYMBOL(touch_atime);
1574
1575/*
1576 * The logic we want is
1577 *
1578 *      if suid or (sgid and xgrp)
1579 *              remove privs
1580 */
1581int should_remove_suid(struct dentry *dentry)
1582{
1583        umode_t mode = dentry->d_inode->i_mode;
1584        int kill = 0;
1585
1586        /* suid always must be killed */
1587        if (unlikely(mode & S_ISUID))
1588                kill = ATTR_KILL_SUID;
1589
1590        /*
1591         * sgid without any exec bits is just a mandatory locking mark; leave
1592         * it alone.  If some exec bits are set, it's a real sgid; kill it.
1593         */
1594        if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1595                kill |= ATTR_KILL_SGID;
1596
1597        if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1598                return kill;
1599
1600        return 0;
1601}
1602EXPORT_SYMBOL(should_remove_suid);
1603
1604static int __remove_suid(struct dentry *dentry, int kill)
1605{
1606        struct iattr newattrs;
1607
1608        newattrs.ia_valid = ATTR_FORCE | kill;
1609        return notify_change(dentry, &newattrs);
1610}
1611
1612int file_remove_suid(struct file *file)
1613{
1614        struct dentry *dentry = file->f_path.dentry;
1615        struct inode *inode = dentry->d_inode;
1616        int killsuid;
1617        int killpriv;
1618        int error = 0;
1619
1620        /* Fast path for nothing security related */
1621        if (IS_NOSEC(inode))
1622                return 0;
1623
1624        killsuid = should_remove_suid(dentry);
1625        killpriv = security_inode_need_killpriv(dentry);
1626
1627        if (killpriv < 0)
1628                return killpriv;
1629        if (killpriv)
1630                error = security_inode_killpriv(dentry);
1631        if (!error && killsuid)
1632                error = __remove_suid(dentry, killsuid);
1633        if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1634                inode->i_flags |= S_NOSEC;
1635
1636        return error;
1637}
1638EXPORT_SYMBOL(file_remove_suid);
1639
1640/**
1641 *      file_update_time        -       update mtime and ctime time
1642 *      @file: file accessed
1643 *
1644 *      Update the mtime and ctime members of an inode and mark the inode
1645 *      for writeback.  Note that this function is meant exclusively for
1646 *      usage in the file write path of filesystems, and filesystems may
1647 *      choose to explicitly ignore update via this function with the
1648 *      S_NOCMTIME inode flag, e.g. for network filesystem where these
1649 *      timestamps are handled by the server.  This can return an error for
1650 *      file systems who need to allocate space in order to update an inode.
1651 */
1652
1653int file_update_time(struct file *file)
1654{
1655        struct inode *inode = file_inode(file);
1656        struct timespec now;
1657        int sync_it = 0;
1658        int ret;
1659
1660        /* First try to exhaust all avenues to not sync */
1661        if (IS_NOCMTIME(inode))
1662                return 0;
1663
1664        now = current_fs_time(inode->i_sb);
1665        if (!timespec_equal(&inode->i_mtime, &now))
1666                sync_it = S_MTIME;
1667
1668        if (!timespec_equal(&inode->i_ctime, &now))
1669                sync_it |= S_CTIME;
1670
1671        if (IS_I_VERSION(inode))
1672                sync_it |= S_VERSION;
1673
1674        if (!sync_it)
1675                return 0;
1676
1677        /* Finally allowed to write? Takes lock. */
1678        if (__mnt_want_write_file(file))
1679                return 0;
1680
1681        ret = update_time(inode, &now, sync_it);
1682        __mnt_drop_write_file(file);
1683
1684        return ret;
1685}
1686EXPORT_SYMBOL(file_update_time);
1687
1688int inode_needs_sync(struct inode *inode)
1689{
1690        if (IS_SYNC(inode))
1691                return 1;
1692        if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1693                return 1;
1694        return 0;
1695}
1696EXPORT_SYMBOL(inode_needs_sync);
1697
1698int inode_wait(void *word)
1699{
1700        schedule();
1701        return 0;
1702}
1703EXPORT_SYMBOL(inode_wait);
1704
1705/*
1706 * If we try to find an inode in the inode hash while it is being
1707 * deleted, we have to wait until the filesystem completes its
1708 * deletion before reporting that it isn't found.  This function waits
1709 * until the deletion _might_ have completed.  Callers are responsible
1710 * to recheck inode state.
1711 *
1712 * It doesn't matter if I_NEW is not set initially, a call to
1713 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1714 * will DTRT.
1715 */
1716static void __wait_on_freeing_inode(struct inode *inode)
1717{
1718        wait_queue_head_t *wq;
1719        DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1720        wq = bit_waitqueue(&inode->i_state, __I_NEW);
1721        prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1722        spin_unlock(&inode->i_lock);
1723        spin_unlock(&inode_hash_lock);
1724        schedule();
1725        finish_wait(wq, &wait.wait);
1726        spin_lock(&inode_hash_lock);
1727}
1728
1729static __initdata unsigned long ihash_entries;
1730static int __init set_ihash_entries(char *str)
1731{
1732        if (!str)
1733                return 0;
1734        ihash_entries = simple_strtoul(str, &str, 0);
1735        return 1;
1736}
1737__setup("ihash_entries=", set_ihash_entries);
1738
1739/*
1740 * Initialize the waitqueues and inode hash table.
1741 */
1742void __init inode_init_early(void)
1743{
1744        unsigned int loop;
1745
1746        /* If hashes are distributed across NUMA nodes, defer
1747         * hash allocation until vmalloc space is available.
1748         */
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                                        HASH_EARLY,
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 inode_init(void)
1768{
1769        unsigned int loop;
1770
1771        /* inode slab cache */
1772        inode_cachep = kmem_cache_create("inode_cache",
1773                                         sizeof(struct inode),
1774                                         0,
1775                                         (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1776                                         SLAB_MEM_SPREAD),
1777                                         init_once);
1778
1779        /* Hash may have been set up in inode_init_early */
1780        if (!hashdist)
1781                return;
1782
1783        inode_hashtable =
1784                alloc_large_system_hash("Inode-cache",
1785                                        sizeof(struct hlist_head),
1786                                        ihash_entries,
1787                                        14,
1788                                        0,
1789                                        &i_hash_shift,
1790                                        &i_hash_mask,
1791                                        0,
1792                                        0);
1793
1794        for (loop = 0; loop < (1U << i_hash_shift); loop++)
1795                INIT_HLIST_HEAD(&inode_hashtable[loop]);
1796}
1797
1798void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1799{
1800        inode->i_mode = mode;
1801        if (S_ISCHR(mode)) {
1802                inode->i_fop = &def_chr_fops;
1803                inode->i_rdev = rdev;
1804        } else if (S_ISBLK(mode)) {
1805                inode->i_fop = &def_blk_fops;
1806                inode->i_rdev = rdev;
1807        } else if (S_ISFIFO(mode))
1808                inode->i_fop = &pipefifo_fops;
1809        else if (S_ISSOCK(mode))
1810                inode->i_fop = &bad_sock_fops;
1811        else
1812                printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1813                                  " inode %s:%lu\n", mode, inode->i_sb->s_id,
1814                                  inode->i_ino);
1815}
1816EXPORT_SYMBOL(init_special_inode);
1817
1818/**
1819 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1820 * @inode: New inode
1821 * @dir: Directory inode
1822 * @mode: mode of the new inode
1823 */
1824void inode_init_owner(struct inode *inode, const struct inode *dir,
1825                        umode_t mode)
1826{
1827        inode->i_uid = current_fsuid();
1828        if (dir && dir->i_mode & S_ISGID) {
1829                inode->i_gid = dir->i_gid;
1830                if (S_ISDIR(mode))
1831                        mode |= S_ISGID;
1832        } else
1833                inode->i_gid = current_fsgid();
1834        inode->i_mode = mode;
1835}
1836EXPORT_SYMBOL(inode_init_owner);
1837
1838/**
1839 * inode_owner_or_capable - check current task permissions to inode
1840 * @inode: inode being checked
1841 *
1842 * Return true if current either has CAP_FOWNER to the inode, or
1843 * owns the file.
1844 */
1845bool inode_owner_or_capable(const struct inode *inode)
1846{
1847        if (uid_eq(current_fsuid(), inode->i_uid))
1848                return true;
1849        if (inode_capable(inode, CAP_FOWNER))
1850                return true;
1851        return false;
1852}
1853EXPORT_SYMBOL(inode_owner_or_capable);
1854
1855/*
1856 * Direct i/o helper functions
1857 */
1858static void __inode_dio_wait(struct inode *inode)
1859{
1860        wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1861        DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1862
1863        do {
1864                prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1865                if (atomic_read(&inode->i_dio_count))
1866                        schedule();
1867        } while (atomic_read(&inode->i_dio_count));
1868        finish_wait(wq, &q.wait);
1869}
1870
1871/**
1872 * inode_dio_wait - wait for outstanding DIO requests to finish
1873 * @inode: inode to wait for
1874 *
1875 * Waits for all pending direct I/O requests to finish so that we can
1876 * proceed with a truncate or equivalent operation.
1877 *
1878 * Must be called under a lock that serializes taking new references
1879 * to i_dio_count, usually by inode->i_mutex.
1880 */
1881void inode_dio_wait(struct inode *inode)
1882{
1883        if (atomic_read(&inode->i_dio_count))
1884                __inode_dio_wait(inode);
1885}
1886EXPORT_SYMBOL(inode_dio_wait);
1887
1888/*
1889 * inode_dio_done - signal finish of a direct I/O requests
1890 * @inode: inode the direct I/O happens on
1891 *
1892 * This is called once we've finished processing a direct I/O request,
1893 * and is used to wake up callers waiting for direct I/O to be quiesced.
1894 */
1895void inode_dio_done(struct inode *inode)
1896{
1897        if (atomic_dec_and_test(&inode->i_dio_count))
1898                wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1899}
1900EXPORT_SYMBOL(inode_dio_done);
1901
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.