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->assoc_mapping = 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 (WARN_ON(inode->i_nlink == 0))
 337                atomic_long_dec(&inode->i_sb->s_remove_count);
 338
 339        inode->__i_nlink++;
 340}
 341EXPORT_SYMBOL(inc_nlink);
 342
 343void address_space_init_once(struct address_space *mapping)
 344{
 345        memset(mapping, 0, sizeof(*mapping));
 346        INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
 347        spin_lock_init(&mapping->tree_lock);
 348        mutex_init(&mapping->i_mmap_mutex);
 349        INIT_LIST_HEAD(&mapping->private_list);
 350        spin_lock_init(&mapping->private_lock);
 351        INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
 352        INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
 353}
 354EXPORT_SYMBOL(address_space_init_once);
 355
 356/*
 357 * These are initializations that only need to be done
 358 * once, because the fields are idempotent across use
 359 * of the inode, so let the slab aware of that.
 360 */
 361void inode_init_once(struct inode *inode)
 362{
 363        memset(inode, 0, sizeof(*inode));
 364        INIT_HLIST_NODE(&inode->i_hash);
 365        INIT_LIST_HEAD(&inode->i_devices);
 366        INIT_LIST_HEAD(&inode->i_wb_list);
 367        INIT_LIST_HEAD(&inode->i_lru);
 368        address_space_init_once(&inode->i_data);
 369        i_size_ordered_init(inode);
 370#ifdef CONFIG_FSNOTIFY
 371        INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
 372#endif
 373}
 374EXPORT_SYMBOL(inode_init_once);
 375
 376static void init_once(void *foo)
 377{
 378        struct inode *inode = (struct inode *) foo;
 379
 380        inode_init_once(inode);
 381}
 382
 383/*
 384 * inode->i_lock must be held
 385 */
 386void __iget(struct inode *inode)
 387{
 388        atomic_inc(&inode->i_count);
 389}
 390
 391/*
 392 * get additional reference to inode; caller must already hold one.
 393 */
 394void ihold(struct inode *inode)
 395{
 396        WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 397}
 398EXPORT_SYMBOL(ihold);
 399
 400static void inode_lru_list_add(struct inode *inode)
 401{
 402        spin_lock(&inode->i_sb->s_inode_lru_lock);
 403        if (list_empty(&inode->i_lru)) {
 404                list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
 405                inode->i_sb->s_nr_inodes_unused++;
 406                this_cpu_inc(nr_unused);
 407        }
 408        spin_unlock(&inode->i_sb->s_inode_lru_lock);
 409}
 410
 411static void inode_lru_list_del(struct inode *inode)
 412{
 413        spin_lock(&inode->i_sb->s_inode_lru_lock);
 414        if (!list_empty(&inode->i_lru)) {
 415                list_del_init(&inode->i_lru);
 416                inode->i_sb->s_nr_inodes_unused--;
 417                this_cpu_dec(nr_unused);
 418        }
 419        spin_unlock(&inode->i_sb->s_inode_lru_lock);
 420}
 421
 422/**
 423 * inode_sb_list_add - add inode to the superblock list of inodes
 424 * @inode: inode to add
 425 */
 426void inode_sb_list_add(struct inode *inode)
 427{
 428        spin_lock(&inode_sb_list_lock);
 429        list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 430        spin_unlock(&inode_sb_list_lock);
 431}
 432EXPORT_SYMBOL_GPL(inode_sb_list_add);
 433
 434static inline void inode_sb_list_del(struct inode *inode)
 435{
 436        if (!list_empty(&inode->i_sb_list)) {
 437                spin_lock(&inode_sb_list_lock);
 438                list_del_init(&inode->i_sb_list);
 439                spin_unlock(&inode_sb_list_lock);
 440        }
 441}
 442
 443static unsigned long hash(struct super_block *sb, unsigned long hashval)
 444{
 445        unsigned long tmp;
 446
 447        tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 448                        L1_CACHE_BYTES;
 449        tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 450        return tmp & i_hash_mask;
 451}
 452
 453/**
 454 *      __insert_inode_hash - hash an inode
 455 *      @inode: unhashed inode
 456 *      @hashval: unsigned long value used to locate this object in the
 457 *              inode_hashtable.
 458 *
 459 *      Add an inode to the inode hash for this superblock.
 460 */
 461void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 462{
 463        struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 464
 465        spin_lock(&inode_hash_lock);
 466        spin_lock(&inode->i_lock);
 467        hlist_add_head(&inode->i_hash, b);
 468        spin_unlock(&inode->i_lock);
 469        spin_unlock(&inode_hash_lock);
 470}
 471EXPORT_SYMBOL(__insert_inode_hash);
 472
 473/**
 474 *      __remove_inode_hash - remove an inode from the hash
 475 *      @inode: inode to unhash
 476 *
 477 *      Remove an inode from the superblock.
 478 */
 479void __remove_inode_hash(struct inode *inode)
 480{
 481        spin_lock(&inode_hash_lock);
 482        spin_lock(&inode->i_lock);
 483        hlist_del_init(&inode->i_hash);
 484        spin_unlock(&inode->i_lock);
 485        spin_unlock(&inode_hash_lock);
 486}
 487EXPORT_SYMBOL(__remove_inode_hash);
 488
 489void clear_inode(struct inode *inode)
 490{
 491        might_sleep();
 492        /*
 493         * We have to cycle tree_lock here because reclaim can be still in the
 494         * process of removing the last page (in __delete_from_page_cache())
 495         * and we must not free mapping under it.
 496         */
 497        spin_lock_irq(&inode->i_data.tree_lock);
 498        BUG_ON(inode->i_data.nrpages);
 499        spin_unlock_irq(&inode->i_data.tree_lock);
 500        BUG_ON(!list_empty(&inode->i_data.private_list));
 501        BUG_ON(!(inode->i_state & I_FREEING));
 502        BUG_ON(inode->i_state & I_CLEAR);
 503        /* don't need i_lock here, no concurrent mods to i_state */
 504        inode->i_state = I_FREEING | I_CLEAR;
 505}
 506EXPORT_SYMBOL(clear_inode);
 507
 508/*
 509 * Free the inode passed in, removing it from the lists it is still connected
 510 * to. We remove any pages still attached to the inode and wait for any IO that
 511 * is still in progress before finally destroying the inode.
 512 *
 513 * An inode must already be marked I_FREEING so that we avoid the inode being
 514 * moved back onto lists if we race with other code that manipulates the lists
 515 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 516 *
 517 * An inode must already be removed from the LRU list before being evicted from
 518 * the cache. This should occur atomically with setting the I_FREEING state
 519 * flag, so no inodes here should ever be on the LRU when being evicted.
 520 */
 521static void evict(struct inode *inode)
 522{
 523        const struct super_operations *op = inode->i_sb->s_op;
 524
 525        BUG_ON(!(inode->i_state & I_FREEING));
 526        BUG_ON(!list_empty(&inode->i_lru));
 527
 528        if (!list_empty(&inode->i_wb_list))
 529                inode_wb_list_del(inode);
 530
 531        inode_sb_list_del(inode);
 532
 533        /*
 534         * Wait for flusher thread to be done with the inode so that filesystem
 535         * does not start destroying it while writeback is still running. Since
 536         * the inode has I_FREEING set, flusher thread won't start new work on
 537         * the inode.  We just have to wait for running writeback to finish.
 538         */
 539        inode_wait_for_writeback(inode);
 540
 541        if (op->evict_inode) {
 542                op->evict_inode(inode);
 543        } else {
 544                if (inode->i_data.nrpages)
 545                        truncate_inode_pages(&inode->i_data, 0);
 546                clear_inode(inode);
 547        }
 548        if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 549                bd_forget(inode);
 550        if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 551                cd_forget(inode);
 552
 553        remove_inode_hash(inode);
 554
 555        spin_lock(&inode->i_lock);
 556        wake_up_bit(&inode->i_state, __I_NEW);
 557        BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 558        spin_unlock(&inode->i_lock);
 559
 560        destroy_inode(inode);
 561}
 562
 563/*
 564 * dispose_list - dispose of the contents of a local list
 565 * @head: the head of the list to free
 566 *
 567 * Dispose-list gets a local list with local inodes in it, so it doesn't
 568 * need to worry about list corruption and SMP locks.
 569 */
 570static void dispose_list(struct list_head *head)
 571{
 572        while (!list_empty(head)) {
 573                struct inode *inode;
 574
 575                inode = list_first_entry(head, struct inode, i_lru);
 576                list_del_init(&inode->i_lru);
 577
 578                evict(inode);
 579        }
 580}
 581
 582/**
 583 * evict_inodes - evict all evictable inodes for a superblock
 584 * @sb:         superblock to operate on
 585 *
 586 * Make sure that no inodes with zero refcount are retained.  This is
 587 * called by superblock shutdown after having MS_ACTIVE flag removed,
 588 * so any inode reaching zero refcount during or after that call will
 589 * be immediately evicted.
 590 */
 591void evict_inodes(struct super_block *sb)
 592{
 593        struct inode *inode, *next;
 594        LIST_HEAD(dispose);
 595
 596        spin_lock(&inode_sb_list_lock);
 597        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 598                if (atomic_read(&inode->i_count))
 599                        continue;
 600
 601                spin_lock(&inode->i_lock);
 602                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 603                        spin_unlock(&inode->i_lock);
 604                        continue;
 605                }
 606
 607                inode->i_state |= I_FREEING;
 608                inode_lru_list_del(inode);
 609                spin_unlock(&inode->i_lock);
 610                list_add(&inode->i_lru, &dispose);
 611        }
 612        spin_unlock(&inode_sb_list_lock);
 613
 614        dispose_list(&dispose);
 615}
 616
 617/**
 618 * invalidate_inodes    - attempt to free all inodes on a superblock
 619 * @sb:         superblock to operate on
 620 * @kill_dirty: flag to guide handling of dirty inodes
 621 *
 622 * Attempts to free all inodes for a given superblock.  If there were any
 623 * busy inodes return a non-zero value, else zero.
 624 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 625 * them as busy.
 626 */
 627int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 628{
 629        int busy = 0;
 630        struct inode *inode, *next;
 631        LIST_HEAD(dispose);
 632
 633        spin_lock(&inode_sb_list_lock);
 634        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 635                spin_lock(&inode->i_lock);
 636                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 637                        spin_unlock(&inode->i_lock);
 638                        continue;
 639                }
 640                if (inode->i_state & I_DIRTY && !kill_dirty) {
 641                        spin_unlock(&inode->i_lock);
 642                        busy = 1;
 643                        continue;
 644                }
 645                if (atomic_read(&inode->i_count)) {
 646                        spin_unlock(&inode->i_lock);
 647                        busy = 1;
 648                        continue;
 649                }
 650
 651                inode->i_state |= I_FREEING;
 652                inode_lru_list_del(inode);
 653                spin_unlock(&inode->i_lock);
 654                list_add(&inode->i_lru, &dispose);
 655        }
 656        spin_unlock(&inode_sb_list_lock);
 657
 658        dispose_list(&dispose);
 659
 660        return busy;
 661}
 662
 663static int can_unuse(struct inode *inode)
 664{
 665        if (inode->i_state & ~I_REFERENCED)
 666                return 0;
 667        if (inode_has_buffers(inode))
 668                return 0;
 669        if (atomic_read(&inode->i_count))
 670                return 0;
 671        if (inode->i_data.nrpages)
 672                return 0;
 673        return 1;
 674}
 675
 676/*
 677 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 678 * This is called from the superblock shrinker function with a number of inodes
 679 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 680 * then are freed outside inode_lock by dispose_list().
 681 *
 682 * Any inodes which are pinned purely because of attached pagecache have their
 683 * pagecache removed.  If the inode has metadata buffers attached to
 684 * mapping->private_list then try to remove them.
 685 *
 686 * If the inode has the I_REFERENCED flag set, then it means that it has been
 687 * used recently - the flag is set in iput_final(). When we encounter such an
 688 * inode, clear the flag and move it to the back of the LRU so it gets another
 689 * pass through the LRU before it gets reclaimed. This is necessary because of
 690 * the fact we are doing lazy LRU updates to minimise lock contention so the
 691 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 692 * with this flag set because they are the inodes that are out of order.
 693 */
 694void prune_icache_sb(struct super_block *sb, int nr_to_scan)
 695{
 696        LIST_HEAD(freeable);
 697        int nr_scanned;
 698        unsigned long reap = 0;
 699
 700        spin_lock(&sb->s_inode_lru_lock);
 701        for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
 702                struct inode *inode;
 703
 704                if (list_empty(&sb->s_inode_lru))
 705                        break;
 706
 707                inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
 708
 709                /*
 710                 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
 711                 * so use a trylock. If we fail to get the lock, just move the
 712                 * inode to the back of the list so we don't spin on it.
 713                 */
 714                if (!spin_trylock(&inode->i_lock)) {
 715                        list_move_tail(&inode->i_lru, &sb->s_inode_lru);
 716                        continue;
 717                }
 718
 719                /*
 720                 * Referenced or dirty inodes are still in use. Give them
 721                 * another pass through the LRU as we canot reclaim them now.
 722                 */
 723                if (atomic_read(&inode->i_count) ||
 724                    (inode->i_state & ~I_REFERENCED)) {
 725                        list_del_init(&inode->i_lru);
 726                        spin_unlock(&inode->i_lock);
 727                        sb->s_nr_inodes_unused--;
 728                        this_cpu_dec(nr_unused);
 729                        continue;
 730                }
 731
 732                /* recently referenced inodes get one more pass */
 733                if (inode->i_state & I_REFERENCED) {
 734                        inode->i_state &= ~I_REFERENCED;
 735                        list_move(&inode->i_lru, &sb->s_inode_lru);
 736                        spin_unlock(&inode->i_lock);
 737                        continue;
 738                }
 739                if (inode_has_buffers(inode) || inode->i_data.nrpages) {
 740                        __iget(inode);
 741                        spin_unlock(&inode->i_lock);
 742                        spin_unlock(&sb->s_inode_lru_lock);
 743                        if (remove_inode_buffers(inode))
 744                                reap += invalidate_mapping_pages(&inode->i_data,
 745                                                                0, -1);
 746                        iput(inode);
 747                        spin_lock(&sb->s_inode_lru_lock);
 748
 749                        if (inode != list_entry(sb->s_inode_lru.next,
 750                                                struct inode, i_lru))
 751                                continue;       /* wrong inode or list_empty */
 752                        /* avoid lock inversions with trylock */
 753                        if (!spin_trylock(&inode->i_lock))
 754                                continue;
 755                        if (!can_unuse(inode)) {
 756                                spin_unlock(&inode->i_lock);
 757                                continue;
 758                        }
 759                }
 760                WARN_ON(inode->i_state & I_NEW);
 761                inode->i_state |= I_FREEING;
 762                spin_unlock(&inode->i_lock);
 763
 764                list_move(&inode->i_lru, &freeable);
 765                sb->s_nr_inodes_unused--;
 766                this_cpu_dec(nr_unused);
 767        }
 768        if (current_is_kswapd())
 769                __count_vm_events(KSWAPD_INODESTEAL, reap);
 770        else
 771                __count_vm_events(PGINODESTEAL, reap);
 772        spin_unlock(&sb->s_inode_lru_lock);
 773        if (current->reclaim_state)
 774                current->reclaim_state->reclaimed_slab += reap;
 775
 776        dispose_list(&freeable);
 777}
 778
 779static void __wait_on_freeing_inode(struct inode *inode);
 780/*
 781 * Called with the inode lock held.
 782 */
 783static struct inode *find_inode(struct super_block *sb,
 784                                struct hlist_head *head,
 785                                int (*test)(struct inode *, void *),
 786                                void *data)
 787{
 788        struct hlist_node *node;
 789        struct inode *inode = NULL;
 790
 791repeat:
 792        hlist_for_each_entry(inode, node, head, i_hash) {
 793                spin_lock(&inode->i_lock);
 794                if (inode->i_sb != sb) {
 795                        spin_unlock(&inode->i_lock);
 796                        continue;
 797                }
 798                if (!test(inode, data)) {
 799                        spin_unlock(&inode->i_lock);
 800                        continue;
 801                }
 802                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 803                        __wait_on_freeing_inode(inode);
 804                        goto repeat;
 805                }
 806                __iget(inode);
 807                spin_unlock(&inode->i_lock);
 808                return inode;
 809        }
 810        return NULL;
 811}
 812
 813/*
 814 * find_inode_fast is the fast path version of find_inode, see the comment at
 815 * iget_locked for details.
 816 */
 817static struct inode *find_inode_fast(struct super_block *sb,
 818                                struct hlist_head *head, unsigned long ino)
 819{
 820        struct hlist_node *node;
 821        struct inode *inode = NULL;
 822
 823repeat:
 824        hlist_for_each_entry(inode, node, head, i_hash) {
 825                spin_lock(&inode->i_lock);
 826                if (inode->i_ino != ino) {
 827                        spin_unlock(&inode->i_lock);
 828                        continue;
 829                }
 830                if (inode->i_sb != sb) {
 831                        spin_unlock(&inode->i_lock);
 832                        continue;
 833                }
 834                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 835                        __wait_on_freeing_inode(inode);
 836                        goto repeat;
 837                }
 838                __iget(inode);
 839                spin_unlock(&inode->i_lock);
 840                return inode;
 841        }
 842        return NULL;
 843}
 844
 845/*
 846 * Each cpu owns a range of LAST_INO_BATCH numbers.
 847 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 848 * to renew the exhausted range.
 849 *
 850 * This does not significantly increase overflow rate because every CPU can
 851 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 852 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 853 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 854 * overflow rate by 2x, which does not seem too significant.
 855 *
 856 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 857 * error if st_ino won't fit in target struct field. Use 32bit counter
 858 * here to attempt to avoid that.
 859 */
 860#define LAST_INO_BATCH 1024
 861static DEFINE_PER_CPU(unsigned int, last_ino);
 862
 863unsigned int get_next_ino(void)
 864{
 865        unsigned int *p = &get_cpu_var(last_ino);
 866        unsigned int res = *p;
 867
 868#ifdef CONFIG_SMP
 869        if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 870                static atomic_t shared_last_ino;
 871                int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 872
 873                res = next - LAST_INO_BATCH;
 874        }
 875#endif
 876
 877        *p = ++res;
 878        put_cpu_var(last_ino);
 879        return res;
 880}
 881EXPORT_SYMBOL(get_next_ino);
 882
 883/**
 884 *      new_inode_pseudo        - obtain an inode
 885 *      @sb: superblock
 886 *
 887 *      Allocates a new inode for given superblock.
 888 *      Inode wont be chained in superblock s_inodes list
 889 *      This means :
 890 *      - fs can't be unmount
 891 *      - quotas, fsnotify, writeback can't work
 892 */
 893struct inode *new_inode_pseudo(struct super_block *sb)
 894{
 895        struct inode *inode = alloc_inode(sb);
 896
 897        if (inode) {
 898                spin_lock(&inode->i_lock);
 899                inode->i_state = 0;
 900                spin_unlock(&inode->i_lock);
 901                INIT_LIST_HEAD(&inode->i_sb_list);
 902        }
 903        return inode;
 904}
 905
 906/**
 907 *      new_inode       - obtain an inode
 908 *      @sb: superblock
 909 *
 910 *      Allocates a new inode for given superblock. The default gfp_mask
 911 *      for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
 912 *      If HIGHMEM pages are unsuitable or it is known that pages allocated
 913 *      for the page cache are not reclaimable or migratable,
 914 *      mapping_set_gfp_mask() must be called with suitable flags on the
 915 *      newly created inode's mapping
 916 *
 917 */
 918struct inode *new_inode(struct super_block *sb)
 919{
 920        struct inode *inode;
 921
 922        spin_lock_prefetch(&inode_sb_list_lock);
 923
 924        inode = new_inode_pseudo(sb);
 925        if (inode)
 926                inode_sb_list_add(inode);
 927        return inode;
 928}
 929EXPORT_SYMBOL(new_inode);
 930
 931#ifdef CONFIG_DEBUG_LOCK_ALLOC
 932void lockdep_annotate_inode_mutex_key(struct inode *inode)
 933{
 934        if (S_ISDIR(inode->i_mode)) {
 935                struct file_system_type *type = inode->i_sb->s_type;
 936
 937                /* Set new key only if filesystem hasn't already changed it */
 938                if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
 939                        /*
 940                         * ensure nobody is actually holding i_mutex
 941                         */
 942                        mutex_destroy(&inode->i_mutex);
 943                        mutex_init(&inode->i_mutex);
 944                        lockdep_set_class(&inode->i_mutex,
 945                                          &type->i_mutex_dir_key);
 946                }
 947        }
 948}
 949EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
 950#endif
 951
 952/**
 953 * unlock_new_inode - clear the I_NEW state and wake up any waiters
 954 * @inode:      new inode to unlock
 955 *
 956 * Called when the inode is fully initialised to clear the new state of the
 957 * inode and wake up anyone waiting for the inode to finish initialisation.
 958 */
 959void unlock_new_inode(struct inode *inode)
 960{
 961        lockdep_annotate_inode_mutex_key(inode);
 962        spin_lock(&inode->i_lock);
 963        WARN_ON(!(inode->i_state & I_NEW));
 964        inode->i_state &= ~I_NEW;
 965        smp_mb();
 966        wake_up_bit(&inode->i_state, __I_NEW);
 967        spin_unlock(&inode->i_lock);
 968}
 969EXPORT_SYMBOL(unlock_new_inode);
 970
 971/**
 972 * iget5_locked - obtain an inode from a mounted file system
 973 * @sb:         super block of file system
 974 * @hashval:    hash value (usually inode number) to get
 975 * @test:       callback used for comparisons between inodes
 976 * @set:        callback used to initialize a new struct inode
 977 * @data:       opaque data pointer to pass to @test and @set
 978 *
 979 * Search for the inode specified by @hashval and @data in the inode cache,
 980 * and if present it is return it with an increased reference count. This is
 981 * a generalized version of iget_locked() for file systems where the inode
 982 * number is not sufficient for unique identification of an inode.
 983 *
 984 * If the inode is not in cache, allocate a new inode and return it locked,
 985 * hashed, and with the I_NEW flag set. The file system gets to fill it in
 986 * before unlocking it via unlock_new_inode().
 987 *
 988 * Note both @test and @set are called with the inode_hash_lock held, so can't
 989 * sleep.
 990 */
 991struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
 992                int (*test)(struct inode *, void *),
 993                int (*set)(struct inode *, void *), void *data)
 994{
 995        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
 996        struct inode *inode;
 997
 998        spin_lock(&inode_hash_lock);
 999        inode = find_inode(sb, head, test, data);
1000        spin_unlock(&inode_hash_lock);
1001
1002        if (inode) {
1003                wait_on_inode(inode);
1004                return inode;
1005        }
1006
1007        inode = alloc_inode(sb);
1008        if (inode) {
1009                struct inode *old;
1010
1011                spin_lock(&inode_hash_lock);
1012                /* We released the lock, so.. */
1013                old = find_inode(sb, head, test, data);
1014                if (!old) {
1015                        if (set(inode, data))
1016                                goto set_failed;
1017
1018                        spin_lock(&inode->i_lock);
1019                        inode->i_state = I_NEW;
1020                        hlist_add_head(&inode->i_hash, head);
1021                        spin_unlock(&inode->i_lock);
1022                        inode_sb_list_add(inode);
1023                        spin_unlock(&inode_hash_lock);
1024
1025                        /* Return the locked inode with I_NEW set, the
1026                         * caller is responsible for filling in the contents
1027                         */
1028                        return inode;
1029                }
1030
1031                /*
1032                 * Uhhuh, somebody else created the same inode under
1033                 * us. Use the old inode instead of the one we just
1034                 * allocated.
1035                 */
1036                spin_unlock(&inode_hash_lock);
1037                destroy_inode(inode);
1038                inode = old;
1039                wait_on_inode(inode);
1040        }
1041        return inode;
1042
1043set_failed:
1044        spin_unlock(&inode_hash_lock);
1045        destroy_inode(inode);
1046        return NULL;
1047}
1048EXPORT_SYMBOL(iget5_locked);
1049
1050/**
1051 * iget_locked - obtain an inode from a mounted file system
1052 * @sb:         super block of file system
1053 * @ino:        inode number to get
1054 *
1055 * Search for the inode specified by @ino in the inode cache and if present
1056 * return it with an increased reference count. This is for file systems
1057 * where the inode number is sufficient for unique identification of an inode.
1058 *
1059 * If the inode is not in cache, allocate a new inode and return it locked,
1060 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1061 * before unlocking it via unlock_new_inode().
1062 */
1063struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1064{
1065        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1066        struct inode *inode;
1067
1068        spin_lock(&inode_hash_lock);
1069        inode = find_inode_fast(sb, head, ino);
1070        spin_unlock(&inode_hash_lock);
1071        if (inode) {
1072                wait_on_inode(inode);
1073                return inode;
1074        }
1075
1076        inode = alloc_inode(sb);
1077        if (inode) {
1078                struct inode *old;
1079
1080                spin_lock(&inode_hash_lock);
1081                /* We released the lock, so.. */
1082                old = find_inode_fast(sb, head, ino);
1083                if (!old) {
1084                        inode->i_ino = ino;
1085                        spin_lock(&inode->i_lock);
1086                        inode->i_state = I_NEW;
1087                        hlist_add_head(&inode->i_hash, head);
1088                        spin_unlock(&inode->i_lock);
1089                        inode_sb_list_add(inode);
1090                        spin_unlock(&inode_hash_lock);
1091
1092                        /* Return the locked inode with I_NEW set, the
1093                         * caller is responsible for filling in the contents
1094                         */
1095                        return inode;
1096                }
1097
1098                /*
1099                 * Uhhuh, somebody else created the same inode under
1100                 * us. Use the old inode instead of the one we just
1101                 * allocated.
1102                 */
1103                spin_unlock(&inode_hash_lock);
1104                destroy_inode(inode);
1105                inode = old;
1106                wait_on_inode(inode);
1107        }
1108        return inode;
1109}
1110EXPORT_SYMBOL(iget_locked);
1111
1112/*
1113 * search the inode cache for a matching inode number.
1114 * If we find one, then the inode number we are trying to
1115 * allocate is not unique and so we should not use it.
1116 *
1117 * Returns 1 if the inode number is unique, 0 if it is not.
1118 */
1119static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1120{
1121        struct hlist_head *b = inode_hashtable + hash(sb, ino);
1122        struct hlist_node *node;
1123        struct inode *inode;
1124
1125        spin_lock(&inode_hash_lock);
1126        hlist_for_each_entry(inode, node, b, i_hash) {
1127                if (inode->i_ino == ino && inode->i_sb == sb) {
1128                        spin_unlock(&inode_hash_lock);
1129                        return 0;
1130                }
1131        }
1132        spin_unlock(&inode_hash_lock);
1133
1134        return 1;
1135}
1136
1137/**
1138 *      iunique - get a unique inode number
1139 *      @sb: superblock
1140 *      @max_reserved: highest reserved inode number
1141 *
1142 *      Obtain an inode number that is unique on the system for a given
1143 *      superblock. This is used by file systems that have no natural
1144 *      permanent inode numbering system. An inode number is returned that
1145 *      is higher than the reserved limit but unique.
1146 *
1147 *      BUGS:
1148 *      With a large number of inodes live on the file system this function
1149 *      currently becomes quite slow.
1150 */
1151ino_t iunique(struct super_block *sb, ino_t max_reserved)
1152{
1153        /*
1154         * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1155         * error if st_ino won't fit in target struct field. Use 32bit counter
1156         * here to attempt to avoid that.
1157         */
1158        static DEFINE_SPINLOCK(iunique_lock);
1159        static unsigned int counter;
1160        ino_t res;
1161
1162        spin_lock(&iunique_lock);
1163        do {
1164                if (counter <= max_reserved)
1165                        counter = max_reserved + 1;
1166                res = counter++;
1167        } while (!test_inode_iunique(sb, res));
1168        spin_unlock(&iunique_lock);
1169
1170        return res;
1171}
1172EXPORT_SYMBOL(iunique);
1173
1174struct inode *igrab(struct inode *inode)
1175{
1176        spin_lock(&inode->i_lock);
1177        if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1178                __iget(inode);
1179                spin_unlock(&inode->i_lock);
1180        } else {
1181                spin_unlock(&inode->i_lock);
1182                /*
1183                 * Handle the case where s_op->clear_inode is not been
1184                 * called yet, and somebody is calling igrab
1185                 * while the inode is getting freed.
1186                 */
1187                inode = NULL;
1188        }
1189        return inode;
1190}
1191EXPORT_SYMBOL(igrab);
1192
1193/**
1194 * ilookup5_nowait - search for an inode in the inode cache
1195 * @sb:         super block of file system to search
1196 * @hashval:    hash value (usually inode number) to search for
1197 * @test:       callback used for comparisons between inodes
1198 * @data:       opaque data pointer to pass to @test
1199 *
1200 * Search for the inode specified by @hashval and @data in the inode cache.
1201 * If the inode is in the cache, the inode is returned with an incremented
1202 * reference count.
1203 *
1204 * Note: I_NEW is not waited upon so you have to be very careful what you do
1205 * with the returned inode.  You probably should be using ilookup5() instead.
1206 *
1207 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1208 */
1209struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1210                int (*test)(struct inode *, void *), void *data)
1211{
1212        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1213        struct inode *inode;
1214
1215        spin_lock(&inode_hash_lock);
1216        inode = find_inode(sb, head, test, data);
1217        spin_unlock(&inode_hash_lock);
1218
1219        return inode;
1220}
1221EXPORT_SYMBOL(ilookup5_nowait);
1222
1223/**
1224 * ilookup5 - search for an inode in the inode cache
1225 * @sb:         super block of file system to search
1226 * @hashval:    hash value (usually inode number) to search for
1227 * @test:       callback used for comparisons between inodes
1228 * @data:       opaque data pointer to pass to @test
1229 *
1230 * Search for the inode specified by @hashval and @data in the inode cache,
1231 * and if the inode is in the cache, return the inode with an incremented
1232 * reference count.  Waits on I_NEW before returning the inode.
1233 * returned with an incremented reference count.
1234 *
1235 * This is a generalized version of ilookup() for file systems where the
1236 * inode number is not sufficient for unique identification of an inode.
1237 *
1238 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1239 */
1240struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1241                int (*test)(struct inode *, void *), void *data)
1242{
1243        struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1244
1245        if (inode)
1246                wait_on_inode(inode);
1247        return inode;
1248}
1249EXPORT_SYMBOL(ilookup5);
1250
1251/**
1252 * ilookup - search for an inode in the inode cache
1253 * @sb:         super block of file system to search
1254 * @ino:        inode number to search for
1255 *
1256 * Search for the inode @ino in the inode cache, and if the inode is in the
1257 * cache, the inode is returned with an incremented reference count.
1258 */
1259struct inode *ilookup(struct super_block *sb, unsigned long ino)
1260{
1261        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1262        struct inode *inode;
1263
1264        spin_lock(&inode_hash_lock);
1265        inode = find_inode_fast(sb, head, ino);
1266        spin_unlock(&inode_hash_lock);
1267
1268        if (inode)
1269                wait_on_inode(inode);
1270        return inode;
1271}
1272EXPORT_SYMBOL(ilookup);
1273
1274int insert_inode_locked(struct inode *inode)
1275{
1276        struct super_block *sb = inode->i_sb;
1277        ino_t ino = inode->i_ino;
1278        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1279
1280        while (1) {
1281                struct hlist_node *node;
1282                struct inode *old = NULL;
1283                spin_lock(&inode_hash_lock);
1284                hlist_for_each_entry(old, node, head, i_hash) {
1285                        if (old->i_ino != ino)
1286                                continue;
1287                        if (old->i_sb != sb)
1288                                continue;
1289                        spin_lock(&old->i_lock);
1290                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1291                                spin_unlock(&old->i_lock);
1292                                continue;
1293                        }
1294                        break;
1295                }
1296                if (likely(!node)) {
1297                        spin_lock(&inode->i_lock);
1298                        inode->i_state |= I_NEW;
1299                        hlist_add_head(&inode->i_hash, head);
1300                        spin_unlock(&inode->i_lock);
1301                        spin_unlock(&inode_hash_lock);
1302                        return 0;
1303                }
1304                __iget(old);
1305                spin_unlock(&old->i_lock);
1306                spin_unlock(&inode_hash_lock);
1307                wait_on_inode(old);
1308                if (unlikely(!inode_unhashed(old))) {
1309                        iput(old);
1310                        return -EBUSY;
1311                }
1312                iput(old);
1313        }
1314}
1315EXPORT_SYMBOL(insert_inode_locked);
1316
1317int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1318                int (*test)(struct inode *, void *), void *data)
1319{
1320        struct super_block *sb = inode->i_sb;
1321        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1322
1323        while (1) {
1324                struct hlist_node *node;
1325                struct inode *old = NULL;
1326
1327                spin_lock(&inode_hash_lock);
1328                hlist_for_each_entry(old, node, head, i_hash) {
1329                        if (old->i_sb != sb)
1330                                continue;
1331                        if (!test(old, data))
1332                                continue;
1333                        spin_lock(&old->i_lock);
1334                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1335                                spin_unlock(&old->i_lock);
1336                                continue;
1337                        }
1338                        break;
1339                }
1340                if (likely(!node)) {
1341                        spin_lock(&inode->i_lock);
1342                        inode->i_state |= I_NEW;
1343                        hlist_add_head(&inode->i_hash, head);
1344                        spin_unlock(&inode->i_lock);
1345                        spin_unlock(&inode_hash_lock);
1346                        return 0;
1347                }
1348                __iget(old);
1349                spin_unlock(&old->i_lock);
1350                spin_unlock(&inode_hash_lock);
1351                wait_on_inode(old);
1352                if (unlikely(!inode_unhashed(old))) {
1353                        iput(old);
1354                        return -EBUSY;
1355                }
1356                iput(old);
1357        }
1358}
1359EXPORT_SYMBOL(insert_inode_locked4);
1360
1361
1362int generic_delete_inode(struct inode *inode)
1363{
1364        return 1;
1365}
1366EXPORT_SYMBOL(generic_delete_inode);
1367
1368/*
1369 * Called when we're dropping the last reference
1370 * to an inode.
1371 *
1372 * Call the FS "drop_inode()" function, defaulting to
1373 * the legacy UNIX filesystem behaviour.  If it tells
1374 * us to evict inode, do so.  Otherwise, retain inode
1375 * in cache if fs is alive, sync and evict if fs is
1376 * shutting down.
1377 */
1378static void iput_final(struct inode *inode)
1379{
1380        struct super_block *sb = inode->i_sb;
1381        const struct super_operations *op = inode->i_sb->s_op;
1382        int drop;
1383
1384        WARN_ON(inode->i_state & I_NEW);
1385
1386        if (op->drop_inode)
1387                drop = op->drop_inode(inode);
1388        else
1389                drop = generic_drop_inode(inode);
1390
1391        if (!drop && (sb->s_flags & MS_ACTIVE)) {
1392                inode->i_state |= I_REFERENCED;
1393                if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1394                        inode_lru_list_add(inode);
1395                spin_unlock(&inode->i_lock);
1396                return;
1397        }
1398
1399        if (!drop) {
1400                inode->i_state |= I_WILL_FREE;
1401                spin_unlock(&inode->i_lock);
1402                write_inode_now(inode, 1);
1403                spin_lock(&inode->i_lock);
1404                WARN_ON(inode->i_state & I_NEW);
1405                inode->i_state &= ~I_WILL_FREE;
1406        }
1407
1408        inode->i_state |= I_FREEING;
1409        if (!list_empty(&inode->i_lru))
1410                inode_lru_list_del(inode);
1411        spin_unlock(&inode->i_lock);
1412
1413        evict(inode);
1414}
1415
1416/**
1417 *      iput    - put an inode
1418 *      @inode: inode to put
1419 *
1420 *      Puts an inode, dropping its usage count. If the inode use count hits
1421 *      zero, the inode is then freed and may also be destroyed.
1422 *
1423 *      Consequently, iput() can sleep.
1424 */
1425void iput(struct inode *inode)
1426{
1427        if (inode) {
1428                BUG_ON(inode->i_state & I_CLEAR);
1429
1430                if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1431                        iput_final(inode);
1432        }
1433}
1434EXPORT_SYMBOL(iput);
1435
1436/**
1437 *      bmap    - find a block number in a file
1438 *      @inode: inode of file
1439 *      @block: block to find
1440 *
1441 *      Returns the block number on the device holding the inode that
1442 *      is the disk block number for the block of the file requested.
1443 *      That is, asked for block 4 of inode 1 the function will return the
1444 *      disk block relative to the disk start that holds that block of the
1445 *      file.
1446 */
1447sector_t bmap(struct inode *inode, sector_t block)
1448{
1449        sector_t res = 0;
1450        if (inode->i_mapping->a_ops->bmap)
1451                res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1452        return res;
1453}
1454EXPORT_SYMBOL(bmap);
1455
1456/*
1457 * With relative atime, only update atime if the previous atime is
1458 * earlier than either the ctime or mtime or if at least a day has
1459 * passed since the last atime update.
1460 */
1461static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1462                             struct timespec now)
1463{
1464
1465        if (!(mnt->mnt_flags & MNT_RELATIME))
1466                return 1;
1467        /*
1468         * Is mtime younger than atime? If yes, update atime:
1469         */
1470        if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1471                return 1;
1472        /*
1473         * Is ctime younger than atime? If yes, update atime:
1474         */
1475        if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1476                return 1;
1477
1478        /*
1479         * Is the previous atime value older than a day? If yes,
1480         * update atime:
1481         */
1482        if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1483                return 1;
1484        /*
1485         * Good, we can skip the atime update:
1486         */
1487        return 0;
1488}
1489
1490/*
1491 * This does the actual work of updating an inodes time or version.  Must have
1492 * had called mnt_want_write() before calling this.
1493 */
1494static int update_time(struct inode *inode, struct timespec *time, int flags)
1495{
1496        if (inode->i_op->update_time)
1497                return inode->i_op->update_time(inode, time, flags);
1498
1499        if (flags & S_ATIME)
1500                inode->i_atime = *time;
1501        if (flags & S_VERSION)
1502                inode_inc_iversion(inode);
1503        if (flags & S_CTIME)
1504                inode->i_ctime = *time;
1505        if (flags & S_MTIME)
1506                inode->i_mtime = *time;
1507        mark_inode_dirty_sync(inode);
1508        return 0;
1509}
1510
1511/**
1512 *      touch_atime     -       update the access time
1513 *      @path: the &struct path to update
1514 *
1515 *      Update the accessed time on an inode and mark it for writeback.
1516 *      This function automatically handles read only file systems and media,
1517 *      as well as the "noatime" flag and inode specific "noatime" markers.
1518 */
1519void touch_atime(struct path *path)
1520{
1521        struct vfsmount *mnt = path->mnt;
1522        struct inode *inode = path->dentry->d_inode;
1523        struct timespec now;
1524
1525        if (inode->i_flags & S_NOATIME)
1526                return;
1527        if (IS_NOATIME(inode))
1528                return;
1529        if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1530                return;
1531
1532        if (mnt->mnt_flags & MNT_NOATIME)
1533                return;
1534        if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1535                return;
1536
1537        now = current_fs_time(inode->i_sb);
1538
1539        if (!relatime_need_update(mnt, inode, now))
1540                return;
1541
1542        if (timespec_equal(&inode->i_atime, &now))
1543                return;
1544
1545        if (!sb_start_write_trylock(inode->i_sb))
1546                return;
1547
1548        if (__mnt_want_write(mnt))
1549                goto skip_update;
1550        /*
1551         * File systems can error out when updating inodes if they need to
1552         * allocate new space to modify an inode (such is the case for
1553         * Btrfs), but since we touch atime while walking down the path we
1554         * really don't care if we failed to update the atime of the file,
1555         * so just ignore the return value.
1556         * We may also fail on filesystems that have the ability to make parts
1557         * of the fs read only, e.g. subvolumes in Btrfs.
1558         */
1559        update_time(inode, &now, S_ATIME);
1560        __mnt_drop_write(mnt);
1561skip_update:
1562        sb_end_write(inode->i_sb);
1563}
1564EXPORT_SYMBOL(touch_atime);
1565
1566/*
1567 * The logic we want is
1568 *
1569 *      if suid or (sgid and xgrp)
1570 *              remove privs
1571 */
1572int should_remove_suid(struct dentry *dentry)
1573{
1574        umode_t mode = dentry->d_inode->i_mode;
1575        int kill = 0;
1576
1577        /* suid always must be killed */
1578        if (unlikely(mode & S_ISUID))
1579                kill = ATTR_KILL_SUID;
1580
1581        /*
1582         * sgid without any exec bits is just a mandatory locking mark; leave
1583         * it alone.  If some exec bits are set, it's a real sgid; kill it.
1584         */
1585        if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1586                kill |= ATTR_KILL_SGID;
1587
1588        if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1589                return kill;
1590
1591        return 0;
1592}
1593EXPORT_SYMBOL(should_remove_suid);
1594
1595static int __remove_suid(struct dentry *dentry, int kill)
1596{
1597        struct iattr newattrs;
1598
1599        newattrs.ia_valid = ATTR_FORCE | kill;
1600        return notify_change(dentry, &newattrs);
1601}
1602
1603int file_remove_suid(struct file *file)
1604{
1605        struct dentry *dentry = file->f_path.dentry;
1606        struct inode *inode = dentry->d_inode;
1607        int killsuid;
1608        int killpriv;
1609        int error = 0;
1610
1611        /* Fast path for nothing security related */
1612        if (IS_NOSEC(inode))
1613                return 0;
1614
1615        killsuid = should_remove_suid(dentry);
1616        killpriv = security_inode_need_killpriv(dentry);
1617
1618        if (killpriv < 0)
1619                return killpriv;
1620        if (killpriv)
1621                error = security_inode_killpriv(dentry);
1622        if (!error && killsuid)
1623                error = __remove_suid(dentry, killsuid);
1624        if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1625                inode->i_flags |= S_NOSEC;
1626
1627        return error;
1628}
1629EXPORT_SYMBOL(file_remove_suid);
1630
1631/**
1632 *      file_update_time        -       update mtime and ctime time
1633 *      @file: file accessed
1634 *
1635 *      Update the mtime and ctime members of an inode and mark the inode
1636 *      for writeback.  Note that this function is meant exclusively for
1637 *      usage in the file write path of filesystems, and filesystems may
1638 *      choose to explicitly ignore update via this function with the
1639 *      S_NOCMTIME inode flag, e.g. for network filesystem where these
1640 *      timestamps are handled by the server.  This can return an error for
1641 *      file systems who need to allocate space in order to update an inode.
1642 */
1643
1644int file_update_time(struct file *file)
1645{
1646        struct inode *inode = file->f_path.dentry->d_inode;
1647        struct timespec now;
1648        int sync_it = 0;
1649        int ret;
1650
1651        /* First try to exhaust all avenues to not sync */
1652        if (IS_NOCMTIME(inode))
1653                return 0;
1654
1655        now = current_fs_time(inode->i_sb);
1656        if (!timespec_equal(&inode->i_mtime, &now))
1657                sync_it = S_MTIME;
1658
1659        if (!timespec_equal(&inode->i_ctime, &now))
1660                sync_it |= S_CTIME;
1661
1662        if (IS_I_VERSION(inode))
1663                sync_it |= S_VERSION;
1664
1665        if (!sync_it)
1666                return 0;
1667
1668        /* Finally allowed to write? Takes lock. */
1669        if (__mnt_want_write_file(file))
1670                return 0;
1671
1672        ret = update_time(inode, &now, sync_it);
1673        __mnt_drop_write_file(file);
1674
1675        return ret;
1676}
1677EXPORT_SYMBOL(file_update_time);
1678
1679int inode_needs_sync(struct inode *inode)
1680{
1681        if (IS_SYNC(inode))
1682                return 1;
1683        if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1684                return 1;
1685        return 0;
1686}
1687EXPORT_SYMBOL(inode_needs_sync);
1688
1689int inode_wait(void *word)
1690{
1691        schedule();
1692        return 0;
1693}
1694EXPORT_SYMBOL(inode_wait);
1695
1696/*
1697 * If we try to find an inode in the inode hash while it is being
1698 * deleted, we have to wait until the filesystem completes its
1699 * deletion before reporting that it isn't found.  This function waits
1700 * until the deletion _might_ have completed.  Callers are responsible
1701 * to recheck inode state.
1702 *
1703 * It doesn't matter if I_NEW is not set initially, a call to
1704 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1705 * will DTRT.
1706 */
1707static void __wait_on_freeing_inode(struct inode *inode)
1708{
1709        wait_queue_head_t *wq;
1710        DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1711        wq = bit_waitqueue(&inode->i_state, __I_NEW);
1712        prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1713        spin_unlock(&inode->i_lock);
1714        spin_unlock(&inode_hash_lock);
1715        schedule();
1716        finish_wait(wq, &wait.wait);
1717        spin_lock(&inode_hash_lock);
1718}
1719
1720static __initdata unsigned long ihash_entries;
1721static int __init set_ihash_entries(char *str)
1722{
1723        if (!str)
1724                return 0;
1725        ihash_entries = simple_strtoul(str, &str, 0);
1726        return 1;
1727}
1728__setup("ihash_entries=", set_ihash_entries);
1729
1730/*
1731 * Initialize the waitqueues and inode hash table.
1732 */
1733void __init inode_init_early(void)
1734{
1735        unsigned int loop;
1736
1737        /* If hashes are distributed across NUMA nodes, defer
1738         * hash allocation until vmalloc space is available.
1739         */
1740        if (hashdist)
1741                return;
1742
1743        inode_hashtable =
1744                alloc_large_system_hash("Inode-cache",
1745                                        sizeof(struct hlist_head),
1746                                        ihash_entries,
1747                                        14,
1748                                        HASH_EARLY,
1749                                        &i_hash_shift,
1750                                        &i_hash_mask,
1751                                        0,
1752                                        0);
1753
1754        for (loop = 0; loop < (1U << i_hash_shift); loop++)
1755                INIT_HLIST_HEAD(&inode_hashtable[loop]);
1756}
1757
1758void __init inode_init(void)
1759{
1760        unsigned int loop;
1761
1762        /* inode slab cache */
1763        inode_cachep = kmem_cache_create("inode_cache",
1764                                         sizeof(struct inode),
1765                                         0,
1766                                         (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1767                                         SLAB_MEM_SPREAD),
1768                                         init_once);
1769
1770        /* Hash may have been set up in inode_init_early */
1771        if (!hashdist)
1772                return;
1773
1774        inode_hashtable =
1775                alloc_large_system_hash("Inode-cache",
1776                                        sizeof(struct hlist_head),
1777                                        ihash_entries,
1778                                        14,
1779                                        0,
1780                                        &i_hash_shift,
1781                                        &i_hash_mask,
1782                                        0,
1783                                        0);
1784
1785        for (loop = 0; loop < (1U << i_hash_shift); loop++)
1786                INIT_HLIST_HEAD(&inode_hashtable[loop]);
1787}
1788
1789void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1790{
1791        inode->i_mode = mode;
1792        if (S_ISCHR(mode)) {
1793                inode->i_fop = &def_chr_fops;
1794                inode->i_rdev = rdev;
1795        } else if (S_ISBLK(mode)) {
1796                inode->i_fop = &def_blk_fops;
1797                inode->i_rdev = rdev;
1798        } else if (S_ISFIFO(mode))
1799                inode->i_fop = &def_fifo_fops;
1800        else if (S_ISSOCK(mode))
1801                inode->i_fop = &bad_sock_fops;
1802        else
1803                printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1804                                  " inode %s:%lu\n", mode, inode->i_sb->s_id,
1805                                  inode->i_ino);
1806}
1807EXPORT_SYMBOL(init_special_inode);
1808
1809/**
1810 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1811 * @inode: New inode
1812 * @dir: Directory inode
1813 * @mode: mode of the new inode
1814 */
1815void inode_init_owner(struct inode *inode, const struct inode *dir,
1816                        umode_t mode)
1817{
1818        inode->i_uid = current_fsuid();
1819        if (dir && dir->i_mode & S_ISGID) {
1820                inode->i_gid = dir->i_gid;
1821                if (S_ISDIR(mode))
1822                        mode |= S_ISGID;
1823        } else
1824                inode->i_gid = current_fsgid();
1825        inode->i_mode = mode;
1826}
1827EXPORT_SYMBOL(inode_init_owner);
1828
1829/**
1830 * inode_owner_or_capable - check current task permissions to inode
1831 * @inode: inode being checked
1832 *
1833 * Return true if current either has CAP_FOWNER to the inode, or
1834 * owns the file.
1835 */
1836bool inode_owner_or_capable(const struct inode *inode)
1837{
1838        if (uid_eq(current_fsuid(), inode->i_uid))
1839                return true;
1840        if (inode_capable(inode, CAP_FOWNER))
1841                return true;
1842        return false;
1843}
1844EXPORT_SYMBOL(inode_owner_or_capable);
1845
1846/*
1847 * Direct i/o helper functions
1848 */
1849static void __inode_dio_wait(struct inode *inode)
1850{
1851        wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1852        DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1853
1854        do {
1855                prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1856                if (atomic_read(&inode->i_dio_count))
1857                        schedule();
1858        } while (atomic_read(&inode->i_dio_count));
1859        finish_wait(wq, &q.wait);
1860}
1861
1862/**
1863 * inode_dio_wait - wait for outstanding DIO requests to finish
1864 * @inode: inode to wait for
1865 *
1866 * Waits for all pending direct I/O requests to finish so that we can
1867 * proceed with a truncate or equivalent operation.
1868 *
1869 * Must be called under a lock that serializes taking new references
1870 * to i_dio_count, usually by inode->i_mutex.
1871 */
1872void inode_dio_wait(struct inode *inode)
1873{
1874        if (atomic_read(&inode->i_dio_count))
1875                __inode_dio_wait(inode);
1876}
1877EXPORT_SYMBOL(inode_dio_wait);
1878
1879/*
1880 * inode_dio_done - signal finish of a direct I/O requests
1881 * @inode: inode the direct I/O happens on
1882 *
1883 * This is called once we've finished processing a direct I/O request,
1884 * and is used to wake up callers waiting for direct I/O to be quiesced.
1885 */
1886void inode_dio_done(struct inode *inode)
1887{
1888        if (atomic_dec_and_test(&inode->i_dio_count))
1889                wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1890}
1891EXPORT_SYMBOL(inode_dio_done);
1892
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.