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/memblock.h>
  14#include <linux/fsnotify.h>
  15#include <linux/mount.h>
  16#include <linux/posix_acl.h>
  17#include <linux/prefetch.h>
  18#include <linux/buffer_head.h> /* for inode_has_buffers */
  19#include <linux/ratelimit.h>
  20#include <linux/list_lru.h>
  21#include <linux/iversion.h>
  22#include <trace/events/writeback.h>
  23#include "internal.h"
  24
  25/*
  26 * Inode locking rules:
  27 *
  28 * inode->i_lock protects:
  29 *   inode->i_state, inode->i_hash, __iget()
  30 * Inode LRU list locks protect:
  31 *   inode->i_sb->s_inode_lru, inode->i_lru
  32 * inode->i_sb->s_inode_list_lock protects:
  33 *   inode->i_sb->s_inodes, inode->i_sb_list
  34 * bdi->wb.list_lock protects:
  35 *   bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
  36 * inode_hash_lock protects:
  37 *   inode_hashtable, inode->i_hash
  38 *
  39 * Lock ordering:
  40 *
  41 * inode->i_sb->s_inode_list_lock
  42 *   inode->i_lock
  43 *     Inode LRU list locks
  44 *
  45 * bdi->wb.list_lock
  46 *   inode->i_lock
  47 *
  48 * inode_hash_lock
  49 *   inode->i_sb->s_inode_list_lock
  50 *   inode->i_lock
  51 *
  52 * iunique_lock
  53 *   inode_hash_lock
  54 */
  55
  56static unsigned int i_hash_mask __read_mostly;
  57static unsigned int i_hash_shift __read_mostly;
  58static struct hlist_head *inode_hashtable __read_mostly;
  59static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  60
  61/*
  62 * Empty aops. Can be used for the cases where the user does not
  63 * define any of the address_space operations.
  64 */
  65const struct address_space_operations empty_aops = {
  66};
  67EXPORT_SYMBOL(empty_aops);
  68
  69/*
  70 * Statistics gathering..
  71 */
  72struct inodes_stat_t inodes_stat;
  73
  74static DEFINE_PER_CPU(unsigned long, nr_inodes);
  75static DEFINE_PER_CPU(unsigned long, nr_unused);
  76
  77static struct kmem_cache *inode_cachep __read_mostly;
  78
  79static long get_nr_inodes(void)
  80{
  81        int i;
  82        long sum = 0;
  83        for_each_possible_cpu(i)
  84                sum += per_cpu(nr_inodes, i);
  85        return sum < 0 ? 0 : sum;
  86}
  87
  88static inline long get_nr_inodes_unused(void)
  89{
  90        int i;
  91        long sum = 0;
  92        for_each_possible_cpu(i)
  93                sum += per_cpu(nr_unused, i);
  94        return sum < 0 ? 0 : sum;
  95}
  96
  97long get_nr_dirty_inodes(void)
  98{
  99        /* not actually dirty inodes, but a wild approximation */
 100        long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
 101        return nr_dirty > 0 ? nr_dirty : 0;
 102}
 103
 104/*
 105 * Handle nr_inode sysctl
 106 */
 107#ifdef CONFIG_SYSCTL
 108int proc_nr_inodes(struct ctl_table *table, int write,
 109                   void __user *buffer, size_t *lenp, loff_t *ppos)
 110{
 111        inodes_stat.nr_inodes = get_nr_inodes();
 112        inodes_stat.nr_unused = get_nr_inodes_unused();
 113        return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 114}
 115#endif
 116
 117static int no_open(struct inode *inode, struct file *file)
 118{
 119        return -ENXIO;
 120}
 121
 122/**
 123 * inode_init_always - perform inode structure initialisation
 124 * @sb: superblock inode belongs to
 125 * @inode: inode to initialise
 126 *
 127 * These are initializations that need to be done on every inode
 128 * allocation as the fields are not initialised by slab allocation.
 129 */
 130int inode_init_always(struct super_block *sb, struct inode *inode)
 131{
 132        static const struct inode_operations empty_iops;
 133        static const struct file_operations no_open_fops = {.open = no_open};
 134        struct address_space *const mapping = &inode->i_data;
 135
 136        inode->i_sb = sb;
 137        inode->i_blkbits = sb->s_blocksize_bits;
 138        inode->i_flags = 0;
 139        atomic_set(&inode->i_count, 1);
 140        inode->i_op = &empty_iops;
 141        inode->i_fop = &no_open_fops;
 142        inode->__i_nlink = 1;
 143        inode->i_opflags = 0;
 144        if (sb->s_xattr)
 145                inode->i_opflags |= IOP_XATTR;
 146        i_uid_write(inode, 0);
 147        i_gid_write(inode, 0);
 148        atomic_set(&inode->i_writecount, 0);
 149        inode->i_size = 0;
 150        inode->i_write_hint = WRITE_LIFE_NOT_SET;
 151        inode->i_blocks = 0;
 152        inode->i_bytes = 0;
 153        inode->i_generation = 0;
 154        inode->i_pipe = NULL;
 155        inode->i_bdev = NULL;
 156        inode->i_cdev = NULL;
 157        inode->i_link = NULL;
 158        inode->i_dir_seq = 0;
 159        inode->i_rdev = 0;
 160        inode->dirtied_when = 0;
 161
 162#ifdef CONFIG_CGROUP_WRITEBACK
 163        inode->i_wb_frn_winner = 0;
 164        inode->i_wb_frn_avg_time = 0;
 165        inode->i_wb_frn_history = 0;
 166#endif
 167
 168        if (security_inode_alloc(inode))
 169                goto out;
 170        spin_lock_init(&inode->i_lock);
 171        lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 172
 173        init_rwsem(&inode->i_rwsem);
 174        lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
 175
 176        atomic_set(&inode->i_dio_count, 0);
 177
 178        mapping->a_ops = &empty_aops;
 179        mapping->host = inode;
 180        mapping->flags = 0;
 181        mapping->wb_err = 0;
 182        atomic_set(&mapping->i_mmap_writable, 0);
 183        mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 184        mapping->private_data = NULL;
 185        mapping->writeback_index = 0;
 186        inode->i_private = NULL;
 187        inode->i_mapping = mapping;
 188        INIT_HLIST_HEAD(&inode->i_dentry);      /* buggered by rcu freeing */
 189#ifdef CONFIG_FS_POSIX_ACL
 190        inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 191#endif
 192
 193#ifdef CONFIG_FSNOTIFY
 194        inode->i_fsnotify_mask = 0;
 195#endif
 196        inode->i_flctx = NULL;
 197        this_cpu_inc(nr_inodes);
 198
 199        return 0;
 200out:
 201        return -ENOMEM;
 202}
 203EXPORT_SYMBOL(inode_init_always);
 204
 205static struct inode *alloc_inode(struct super_block *sb)
 206{
 207        struct inode *inode;
 208
 209        if (sb->s_op->alloc_inode)
 210                inode = sb->s_op->alloc_inode(sb);
 211        else
 212                inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
 213
 214        if (!inode)
 215                return NULL;
 216
 217        if (unlikely(inode_init_always(sb, inode))) {
 218                if (inode->i_sb->s_op->destroy_inode)
 219                        inode->i_sb->s_op->destroy_inode(inode);
 220                else
 221                        kmem_cache_free(inode_cachep, inode);
 222                return NULL;
 223        }
 224
 225        return inode;
 226}
 227
 228void free_inode_nonrcu(struct inode *inode)
 229{
 230        kmem_cache_free(inode_cachep, inode);
 231}
 232EXPORT_SYMBOL(free_inode_nonrcu);
 233
 234void __destroy_inode(struct inode *inode)
 235{
 236        BUG_ON(inode_has_buffers(inode));
 237        inode_detach_wb(inode);
 238        security_inode_free(inode);
 239        fsnotify_inode_delete(inode);
 240        locks_free_lock_context(inode);
 241        if (!inode->i_nlink) {
 242                WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
 243                atomic_long_dec(&inode->i_sb->s_remove_count);
 244        }
 245
 246#ifdef CONFIG_FS_POSIX_ACL
 247        if (inode->i_acl && !is_uncached_acl(inode->i_acl))
 248                posix_acl_release(inode->i_acl);
 249        if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
 250                posix_acl_release(inode->i_default_acl);
 251#endif
 252        this_cpu_dec(nr_inodes);
 253}
 254EXPORT_SYMBOL(__destroy_inode);
 255
 256static void i_callback(struct rcu_head *head)
 257{
 258        struct inode *inode = container_of(head, struct inode, i_rcu);
 259        kmem_cache_free(inode_cachep, inode);
 260}
 261
 262static void destroy_inode(struct inode *inode)
 263{
 264        BUG_ON(!list_empty(&inode->i_lru));
 265        __destroy_inode(inode);
 266        if (inode->i_sb->s_op->destroy_inode)
 267                inode->i_sb->s_op->destroy_inode(inode);
 268        else
 269                call_rcu(&inode->i_rcu, i_callback);
 270}
 271
 272/**
 273 * drop_nlink - directly drop an inode's link count
 274 * @inode: inode
 275 *
 276 * This is a low-level filesystem helper to replace any
 277 * direct filesystem manipulation of i_nlink.  In cases
 278 * where we are attempting to track writes to the
 279 * filesystem, a decrement to zero means an imminent
 280 * write when the file is truncated and actually unlinked
 281 * on the filesystem.
 282 */
 283void drop_nlink(struct inode *inode)
 284{
 285        WARN_ON(inode->i_nlink == 0);
 286        inode->__i_nlink--;
 287        if (!inode->i_nlink)
 288                atomic_long_inc(&inode->i_sb->s_remove_count);
 289}
 290EXPORT_SYMBOL(drop_nlink);
 291
 292/**
 293 * clear_nlink - directly zero an inode's link count
 294 * @inode: inode
 295 *
 296 * This is a low-level filesystem helper to replace any
 297 * direct filesystem manipulation of i_nlink.  See
 298 * drop_nlink() for why we care about i_nlink hitting zero.
 299 */
 300void clear_nlink(struct inode *inode)
 301{
 302        if (inode->i_nlink) {
 303                inode->__i_nlink = 0;
 304                atomic_long_inc(&inode->i_sb->s_remove_count);
 305        }
 306}
 307EXPORT_SYMBOL(clear_nlink);
 308
 309/**
 310 * set_nlink - directly set an inode's link count
 311 * @inode: inode
 312 * @nlink: new nlink (should be non-zero)
 313 *
 314 * This is a low-level filesystem helper to replace any
 315 * direct filesystem manipulation of i_nlink.
 316 */
 317void set_nlink(struct inode *inode, unsigned int nlink)
 318{
 319        if (!nlink) {
 320                clear_nlink(inode);
 321        } else {
 322                /* Yes, some filesystems do change nlink from zero to one */
 323                if (inode->i_nlink == 0)
 324                        atomic_long_dec(&inode->i_sb->s_remove_count);
 325
 326                inode->__i_nlink = nlink;
 327        }
 328}
 329EXPORT_SYMBOL(set_nlink);
 330
 331/**
 332 * inc_nlink - directly increment an inode's link count
 333 * @inode: inode
 334 *
 335 * This is a low-level filesystem helper to replace any
 336 * direct filesystem manipulation of i_nlink.  Currently,
 337 * it is only here for parity with dec_nlink().
 338 */
 339void inc_nlink(struct inode *inode)
 340{
 341        if (unlikely(inode->i_nlink == 0)) {
 342                WARN_ON(!(inode->i_state & I_LINKABLE));
 343                atomic_long_dec(&inode->i_sb->s_remove_count);
 344        }
 345
 346        inode->__i_nlink++;
 347}
 348EXPORT_SYMBOL(inc_nlink);
 349
 350static void __address_space_init_once(struct address_space *mapping)
 351{
 352        xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ);
 353        init_rwsem(&mapping->i_mmap_rwsem);
 354        INIT_LIST_HEAD(&mapping->private_list);
 355        spin_lock_init(&mapping->private_lock);
 356        mapping->i_mmap = RB_ROOT_CACHED;
 357}
 358
 359void address_space_init_once(struct address_space *mapping)
 360{
 361        memset(mapping, 0, sizeof(*mapping));
 362        __address_space_init_once(mapping);
 363}
 364EXPORT_SYMBOL(address_space_init_once);
 365
 366/*
 367 * These are initializations that only need to be done
 368 * once, because the fields are idempotent across use
 369 * of the inode, so let the slab aware of that.
 370 */
 371void inode_init_once(struct inode *inode)
 372{
 373        memset(inode, 0, sizeof(*inode));
 374        INIT_HLIST_NODE(&inode->i_hash);
 375        INIT_LIST_HEAD(&inode->i_devices);
 376        INIT_LIST_HEAD(&inode->i_io_list);
 377        INIT_LIST_HEAD(&inode->i_wb_list);
 378        INIT_LIST_HEAD(&inode->i_lru);
 379        __address_space_init_once(&inode->i_data);
 380        i_size_ordered_init(inode);
 381}
 382EXPORT_SYMBOL(inode_init_once);
 383
 384static void init_once(void *foo)
 385{
 386        struct inode *inode = (struct inode *) foo;
 387
 388        inode_init_once(inode);
 389}
 390
 391/*
 392 * inode->i_lock must be held
 393 */
 394void __iget(struct inode *inode)
 395{
 396        atomic_inc(&inode->i_count);
 397}
 398
 399/*
 400 * get additional reference to inode; caller must already hold one.
 401 */
 402void ihold(struct inode *inode)
 403{
 404        WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 405}
 406EXPORT_SYMBOL(ihold);
 407
 408static void inode_lru_list_add(struct inode *inode)
 409{
 410        if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
 411                this_cpu_inc(nr_unused);
 412        else
 413                inode->i_state |= I_REFERENCED;
 414}
 415
 416/*
 417 * Add inode to LRU if needed (inode is unused and clean).
 418 *
 419 * Needs inode->i_lock held.
 420 */
 421void inode_add_lru(struct inode *inode)
 422{
 423        if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
 424                                I_FREEING | I_WILL_FREE)) &&
 425            !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
 426                inode_lru_list_add(inode);
 427}
 428
 429
 430static void inode_lru_list_del(struct inode *inode)
 431{
 432
 433        if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
 434                this_cpu_dec(nr_unused);
 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->i_sb->s_inode_list_lock);
 444        list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 445        spin_unlock(&inode->i_sb->s_inode_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->i_sb->s_inode_list_lock);
 453                list_del_init(&inode->i_sb_list);
 454                spin_unlock(&inode->i_sb->s_inode_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        /*
 507         * We have to cycle the i_pages lock here because reclaim can be in the
 508         * process of removing the last page (in __delete_from_page_cache())
 509         * and we must not free the mapping under it.
 510         */
 511        xa_lock_irq(&inode->i_data.i_pages);
 512        BUG_ON(inode->i_data.nrpages);
 513        BUG_ON(inode->i_data.nrexceptional);
 514        xa_unlock_irq(&inode->i_data.i_pages);
 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        BUG_ON(!list_empty(&inode->i_wb_list));
 519        /* don't need i_lock here, no concurrent mods to i_state */
 520        inode->i_state = I_FREEING | I_CLEAR;
 521}
 522EXPORT_SYMBOL(clear_inode);
 523
 524/*
 525 * Free the inode passed in, removing it from the lists it is still connected
 526 * to. We remove any pages still attached to the inode and wait for any IO that
 527 * is still in progress before finally destroying the inode.
 528 *
 529 * An inode must already be marked I_FREEING so that we avoid the inode being
 530 * moved back onto lists if we race with other code that manipulates the lists
 531 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 532 *
 533 * An inode must already be removed from the LRU list before being evicted from
 534 * the cache. This should occur atomically with setting the I_FREEING state
 535 * flag, so no inodes here should ever be on the LRU when being evicted.
 536 */
 537static void evict(struct inode *inode)
 538{
 539        const struct super_operations *op = inode->i_sb->s_op;
 540
 541        BUG_ON(!(inode->i_state & I_FREEING));
 542        BUG_ON(!list_empty(&inode->i_lru));
 543
 544        if (!list_empty(&inode->i_io_list))
 545                inode_io_list_del(inode);
 546
 547        inode_sb_list_del(inode);
 548
 549        /*
 550         * Wait for flusher thread to be done with the inode so that filesystem
 551         * does not start destroying it while writeback is still running. Since
 552         * the inode has I_FREEING set, flusher thread won't start new work on
 553         * the inode.  We just have to wait for running writeback to finish.
 554         */
 555        inode_wait_for_writeback(inode);
 556
 557        if (op->evict_inode) {
 558                op->evict_inode(inode);
 559        } else {
 560                truncate_inode_pages_final(&inode->i_data);
 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                cond_resched();
 595        }
 596}
 597
 598/**
 599 * evict_inodes - evict all evictable inodes for a superblock
 600 * @sb:         superblock to operate on
 601 *
 602 * Make sure that no inodes with zero refcount are retained.  This is
 603 * called by superblock shutdown after having SB_ACTIVE flag removed,
 604 * so any inode reaching zero refcount during or after that call will
 605 * be immediately evicted.
 606 */
 607void evict_inodes(struct super_block *sb)
 608{
 609        struct inode *inode, *next;
 610        LIST_HEAD(dispose);
 611
 612again:
 613        spin_lock(&sb->s_inode_list_lock);
 614        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 615                if (atomic_read(&inode->i_count))
 616                        continue;
 617
 618                spin_lock(&inode->i_lock);
 619                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 620                        spin_unlock(&inode->i_lock);
 621                        continue;
 622                }
 623
 624                inode->i_state |= I_FREEING;
 625                inode_lru_list_del(inode);
 626                spin_unlock(&inode->i_lock);
 627                list_add(&inode->i_lru, &dispose);
 628
 629                /*
 630                 * We can have a ton of inodes to evict at unmount time given
 631                 * enough memory, check to see if we need to go to sleep for a
 632                 * bit so we don't livelock.
 633                 */
 634                if (need_resched()) {
 635                        spin_unlock(&sb->s_inode_list_lock);
 636                        cond_resched();
 637                        dispose_list(&dispose);
 638                        goto again;
 639                }
 640        }
 641        spin_unlock(&sb->s_inode_list_lock);
 642
 643        dispose_list(&dispose);
 644}
 645EXPORT_SYMBOL_GPL(evict_inodes);
 646
 647/**
 648 * invalidate_inodes    - attempt to free all inodes on a superblock
 649 * @sb:         superblock to operate on
 650 * @kill_dirty: flag to guide handling of dirty inodes
 651 *
 652 * Attempts to free all inodes for a given superblock.  If there were any
 653 * busy inodes return a non-zero value, else zero.
 654 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 655 * them as busy.
 656 */
 657int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 658{
 659        int busy = 0;
 660        struct inode *inode, *next;
 661        LIST_HEAD(dispose);
 662
 663        spin_lock(&sb->s_inode_list_lock);
 664        list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 665                spin_lock(&inode->i_lock);
 666                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 667                        spin_unlock(&inode->i_lock);
 668                        continue;
 669                }
 670                if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
 671                        spin_unlock(&inode->i_lock);
 672                        busy = 1;
 673                        continue;
 674                }
 675                if (atomic_read(&inode->i_count)) {
 676                        spin_unlock(&inode->i_lock);
 677                        busy = 1;
 678                        continue;
 679                }
 680
 681                inode->i_state |= I_FREEING;
 682                inode_lru_list_del(inode);
 683                spin_unlock(&inode->i_lock);
 684                list_add(&inode->i_lru, &dispose);
 685        }
 686        spin_unlock(&sb->s_inode_list_lock);
 687
 688        dispose_list(&dispose);
 689
 690        return busy;
 691}
 692
 693/*
 694 * Isolate the inode from the LRU in preparation for freeing it.
 695 *
 696 * Any inodes which are pinned purely because of attached pagecache have their
 697 * pagecache removed.  If the inode has metadata buffers attached to
 698 * mapping->private_list then try to remove them.
 699 *
 700 * If the inode has the I_REFERENCED flag set, then it means that it has been
 701 * used recently - the flag is set in iput_final(). When we encounter such an
 702 * inode, clear the flag and move it to the back of the LRU so it gets another
 703 * pass through the LRU before it gets reclaimed. This is necessary because of
 704 * the fact we are doing lazy LRU updates to minimise lock contention so the
 705 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 706 * with this flag set because they are the inodes that are out of order.
 707 */
 708static enum lru_status inode_lru_isolate(struct list_head *item,
 709                struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
 710{
 711        struct list_head *freeable = arg;
 712        struct inode    *inode = container_of(item, struct inode, i_lru);
 713
 714        /*
 715         * we are inverting the lru lock/inode->i_lock here, so use a trylock.
 716         * If we fail to get the lock, just skip it.
 717         */
 718        if (!spin_trylock(&inode->i_lock))
 719                return LRU_SKIP;
 720
 721        /*
 722         * Referenced or dirty inodes are still in use. Give them another pass
 723         * through the LRU as we canot reclaim them now.
 724         */
 725        if (atomic_read(&inode->i_count) ||
 726            (inode->i_state & ~I_REFERENCED)) {
 727                list_lru_isolate(lru, &inode->i_lru);
 728                spin_unlock(&inode->i_lock);
 729                this_cpu_dec(nr_unused);
 730                return LRU_REMOVED;
 731        }
 732
 733        /*
 734         * Recently referenced inodes and inodes with many attached pages
 735         * get one more pass.
 736         */
 737        if (inode->i_state & I_REFERENCED || inode->i_data.nrpages > 1) {
 738                inode->i_state &= ~I_REFERENCED;
 739                spin_unlock(&inode->i_lock);
 740                return LRU_ROTATE;
 741        }
 742
 743        if (inode_has_buffers(inode) || inode->i_data.nrpages) {
 744                __iget(inode);
 745                spin_unlock(&inode->i_lock);
 746                spin_unlock(lru_lock);
 747                if (remove_inode_buffers(inode)) {
 748                        unsigned long reap;
 749                        reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
 750                        if (current_is_kswapd())
 751                                __count_vm_events(KSWAPD_INODESTEAL, reap);
 752                        else
 753                                __count_vm_events(PGINODESTEAL, reap);
 754                        if (current->reclaim_state)
 755                                current->reclaim_state->reclaimed_slab += reap;
 756                }
 757                iput(inode);
 758                spin_lock(lru_lock);
 759                return LRU_RETRY;
 760        }
 761
 762        WARN_ON(inode->i_state & I_NEW);
 763        inode->i_state |= I_FREEING;
 764        list_lru_isolate_move(lru, &inode->i_lru, freeable);
 765        spin_unlock(&inode->i_lock);
 766
 767        this_cpu_dec(nr_unused);
 768        return LRU_REMOVED;
 769}
 770
 771/*
 772 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 773 * This is called from the superblock shrinker function with a number of inodes
 774 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 775 * then are freed outside inode_lock by dispose_list().
 776 */
 777long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
 778{
 779        LIST_HEAD(freeable);
 780        long freed;
 781
 782        freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
 783                                     inode_lru_isolate, &freeable);
 784        dispose_list(&freeable);
 785        return freed;
 786}
 787
 788static void __wait_on_freeing_inode(struct inode *inode);
 789/*
 790 * Called with the inode lock held.
 791 */
 792static struct inode *find_inode(struct super_block *sb,
 793                                struct hlist_head *head,
 794                                int (*test)(struct inode *, void *),
 795                                void *data)
 796{
 797        struct inode *inode = NULL;
 798
 799repeat:
 800        hlist_for_each_entry(inode, head, i_hash) {
 801                if (inode->i_sb != sb)
 802                        continue;
 803                if (!test(inode, data))
 804                        continue;
 805                spin_lock(&inode->i_lock);
 806                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 807                        __wait_on_freeing_inode(inode);
 808                        goto repeat;
 809                }
 810                if (unlikely(inode->i_state & I_CREATING)) {
 811                        spin_unlock(&inode->i_lock);
 812                        return ERR_PTR(-ESTALE);
 813                }
 814                __iget(inode);
 815                spin_unlock(&inode->i_lock);
 816                return inode;
 817        }
 818        return NULL;
 819}
 820
 821/*
 822 * find_inode_fast is the fast path version of find_inode, see the comment at
 823 * iget_locked for details.
 824 */
 825static struct inode *find_inode_fast(struct super_block *sb,
 826                                struct hlist_head *head, unsigned long ino)
 827{
 828        struct inode *inode = NULL;
 829
 830repeat:
 831        hlist_for_each_entry(inode, head, i_hash) {
 832                if (inode->i_ino != ino)
 833                        continue;
 834                if (inode->i_sb != sb)
 835                        continue;
 836                spin_lock(&inode->i_lock);
 837                if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 838                        __wait_on_freeing_inode(inode);
 839                        goto repeat;
 840                }
 841                if (unlikely(inode->i_state & I_CREATING)) {
 842                        spin_unlock(&inode->i_lock);
 843                        return ERR_PTR(-ESTALE);
 844                }
 845                __iget(inode);
 846                spin_unlock(&inode->i_lock);
 847                return inode;
 848        }
 849        return NULL;
 850}
 851
 852/*
 853 * Each cpu owns a range of LAST_INO_BATCH numbers.
 854 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 855 * to renew the exhausted range.
 856 *
 857 * This does not significantly increase overflow rate because every CPU can
 858 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 859 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 860 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 861 * overflow rate by 2x, which does not seem too significant.
 862 *
 863 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 864 * error if st_ino won't fit in target struct field. Use 32bit counter
 865 * here to attempt to avoid that.
 866 */
 867#define LAST_INO_BATCH 1024
 868static DEFINE_PER_CPU(unsigned int, last_ino);
 869
 870unsigned int get_next_ino(void)
 871{
 872        unsigned int *p = &get_cpu_var(last_ino);
 873        unsigned int res = *p;
 874
 875#ifdef CONFIG_SMP
 876        if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 877                static atomic_t shared_last_ino;
 878                int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 879
 880                res = next - LAST_INO_BATCH;
 881        }
 882#endif
 883
 884        res++;
 885        /* get_next_ino should not provide a 0 inode number */
 886        if (unlikely(!res))
 887                res++;
 888        *p = res;
 889        put_cpu_var(last_ino);
 890        return res;
 891}
 892EXPORT_SYMBOL(get_next_ino);
 893
 894/**
 895 *      new_inode_pseudo        - obtain an inode
 896 *      @sb: superblock
 897 *
 898 *      Allocates a new inode for given superblock.
 899 *      Inode wont be chained in superblock s_inodes list
 900 *      This means :
 901 *      - fs can't be unmount
 902 *      - quotas, fsnotify, writeback can't work
 903 */
 904struct inode *new_inode_pseudo(struct super_block *sb)
 905{
 906        struct inode *inode = alloc_inode(sb);
 907
 908        if (inode) {
 909                spin_lock(&inode->i_lock);
 910                inode->i_state = 0;
 911                spin_unlock(&inode->i_lock);
 912                INIT_LIST_HEAD(&inode->i_sb_list);
 913        }
 914        return inode;
 915}
 916
 917/**
 918 *      new_inode       - obtain an inode
 919 *      @sb: superblock
 920 *
 921 *      Allocates a new inode for given superblock. The default gfp_mask
 922 *      for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
 923 *      If HIGHMEM pages are unsuitable or it is known that pages allocated
 924 *      for the page cache are not reclaimable or migratable,
 925 *      mapping_set_gfp_mask() must be called with suitable flags on the
 926 *      newly created inode's mapping
 927 *
 928 */
 929struct inode *new_inode(struct super_block *sb)
 930{
 931        struct inode *inode;
 932
 933        spin_lock_prefetch(&sb->s_inode_list_lock);
 934
 935        inode = new_inode_pseudo(sb);
 936        if (inode)
 937                inode_sb_list_add(inode);
 938        return inode;
 939}
 940EXPORT_SYMBOL(new_inode);
 941
 942#ifdef CONFIG_DEBUG_LOCK_ALLOC
 943void lockdep_annotate_inode_mutex_key(struct inode *inode)
 944{
 945        if (S_ISDIR(inode->i_mode)) {
 946                struct file_system_type *type = inode->i_sb->s_type;
 947
 948                /* Set new key only if filesystem hasn't already changed it */
 949                if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
 950                        /*
 951                         * ensure nobody is actually holding i_mutex
 952                         */
 953                        // mutex_destroy(&inode->i_mutex);
 954                        init_rwsem(&inode->i_rwsem);
 955                        lockdep_set_class(&inode->i_rwsem,
 956                                          &type->i_mutex_dir_key);
 957                }
 958        }
 959}
 960EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
 961#endif
 962
 963/**
 964 * unlock_new_inode - clear the I_NEW state and wake up any waiters
 965 * @inode:      new inode to unlock
 966 *
 967 * Called when the inode is fully initialised to clear the new state of the
 968 * inode and wake up anyone waiting for the inode to finish initialisation.
 969 */
 970void unlock_new_inode(struct inode *inode)
 971{
 972        lockdep_annotate_inode_mutex_key(inode);
 973        spin_lock(&inode->i_lock);
 974        WARN_ON(!(inode->i_state & I_NEW));
 975        inode->i_state &= ~I_NEW & ~I_CREATING;
 976        smp_mb();
 977        wake_up_bit(&inode->i_state, __I_NEW);
 978        spin_unlock(&inode->i_lock);
 979}
 980EXPORT_SYMBOL(unlock_new_inode);
 981
 982void discard_new_inode(struct inode *inode)
 983{
 984        lockdep_annotate_inode_mutex_key(inode);
 985        spin_lock(&inode->i_lock);
 986        WARN_ON(!(inode->i_state & I_NEW));
 987        inode->i_state &= ~I_NEW;
 988        smp_mb();
 989        wake_up_bit(&inode->i_state, __I_NEW);
 990        spin_unlock(&inode->i_lock);
 991        iput(inode);
 992}
 993EXPORT_SYMBOL(discard_new_inode);
 994
 995/**
 996 * lock_two_nondirectories - take two i_mutexes on non-directory objects
 997 *
 998 * Lock any non-NULL argument that is not a directory.
 999 * Zero, one or two objects may be locked by this function.
1000 *
1001 * @inode1: first inode to lock
1002 * @inode2: second inode to lock
1003 */
1004void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1005{
1006        if (inode1 > inode2)
1007                swap(inode1, inode2);
1008
1009        if (inode1 && !S_ISDIR(inode1->i_mode))
1010                inode_lock(inode1);
1011        if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1012                inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1013}
1014EXPORT_SYMBOL(lock_two_nondirectories);
1015
1016/**
1017 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1018 * @inode1: first inode to unlock
1019 * @inode2: second inode to unlock
1020 */
1021void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1022{
1023        if (inode1 && !S_ISDIR(inode1->i_mode))
1024                inode_unlock(inode1);
1025        if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1026                inode_unlock(inode2);
1027}
1028EXPORT_SYMBOL(unlock_two_nondirectories);
1029
1030/**
1031 * inode_insert5 - obtain an inode from a mounted file system
1032 * @inode:      pre-allocated inode to use for insert to cache
1033 * @hashval:    hash value (usually inode number) to get
1034 * @test:       callback used for comparisons between inodes
1035 * @set:        callback used to initialize a new struct inode
1036 * @data:       opaque data pointer to pass to @test and @set
1037 *
1038 * Search for the inode specified by @hashval and @data in the inode cache,
1039 * and if present it is return it with an increased reference count. This is
1040 * a variant of iget5_locked() for callers that don't want to fail on memory
1041 * allocation of inode.
1042 *
1043 * If the inode is not in cache, insert the pre-allocated inode to cache and
1044 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1045 * to fill it in before unlocking it via unlock_new_inode().
1046 *
1047 * Note both @test and @set are called with the inode_hash_lock held, so can't
1048 * sleep.
1049 */
1050struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1051                            int (*test)(struct inode *, void *),
1052                            int (*set)(struct inode *, void *), void *data)
1053{
1054        struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1055        struct inode *old;
1056        bool creating = inode->i_state & I_CREATING;
1057
1058again:
1059        spin_lock(&inode_hash_lock);
1060        old = find_inode(inode->i_sb, head, test, data);
1061        if (unlikely(old)) {
1062                /*
1063                 * Uhhuh, somebody else created the same inode under us.
1064                 * Use the old inode instead of the preallocated one.
1065                 */
1066                spin_unlock(&inode_hash_lock);
1067                if (IS_ERR(old))
1068                        return NULL;
1069                wait_on_inode(old);
1070                if (unlikely(inode_unhashed(old))) {
1071                        iput(old);
1072                        goto again;
1073                }
1074                return old;
1075        }
1076
1077        if (set && unlikely(set(inode, data))) {
1078                inode = NULL;
1079                goto unlock;
1080        }
1081
1082        /*
1083         * Return the locked inode with I_NEW set, the
1084         * caller is responsible for filling in the contents
1085         */
1086        spin_lock(&inode->i_lock);
1087        inode->i_state |= I_NEW;
1088        hlist_add_head(&inode->i_hash, head);
1089        spin_unlock(&inode->i_lock);
1090        if (!creating)
1091                inode_sb_list_add(inode);
1092unlock:
1093        spin_unlock(&inode_hash_lock);
1094
1095        return inode;
1096}
1097EXPORT_SYMBOL(inode_insert5);
1098
1099/**
1100 * iget5_locked - obtain an inode from a mounted file system
1101 * @sb:         super block of file system
1102 * @hashval:    hash value (usually inode number) to get
1103 * @test:       callback used for comparisons between inodes
1104 * @set:        callback used to initialize a new struct inode
1105 * @data:       opaque data pointer to pass to @test and @set
1106 *
1107 * Search for the inode specified by @hashval and @data in the inode cache,
1108 * and if present it is return it with an increased reference count. This is
1109 * a generalized version of iget_locked() for file systems where the inode
1110 * number is not sufficient for unique identification of an inode.
1111 *
1112 * If the inode is not in cache, allocate a new inode and return it locked,
1113 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1114 * before unlocking it via unlock_new_inode().
1115 *
1116 * Note both @test and @set are called with the inode_hash_lock held, so can't
1117 * sleep.
1118 */
1119struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1120                int (*test)(struct inode *, void *),
1121                int (*set)(struct inode *, void *), void *data)
1122{
1123        struct inode *inode = ilookup5(sb, hashval, test, data);
1124
1125        if (!inode) {
1126                struct inode *new = alloc_inode(sb);
1127
1128                if (new) {
1129                        new->i_state = 0;
1130                        inode = inode_insert5(new, hashval, test, set, data);
1131                        if (unlikely(inode != new))
1132                                destroy_inode(new);
1133                }
1134        }
1135        return inode;
1136}
1137EXPORT_SYMBOL(iget5_locked);
1138
1139/**
1140 * iget_locked - obtain an inode from a mounted file system
1141 * @sb:         super block of file system
1142 * @ino:        inode number to get
1143 *
1144 * Search for the inode specified by @ino in the inode cache and if present
1145 * return it with an increased reference count. This is for file systems
1146 * where the inode number is sufficient for unique identification of an inode.
1147 *
1148 * If the inode is not in cache, allocate a new inode and return it locked,
1149 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1150 * before unlocking it via unlock_new_inode().
1151 */
1152struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1153{
1154        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1155        struct inode *inode;
1156again:
1157        spin_lock(&inode_hash_lock);
1158        inode = find_inode_fast(sb, head, ino);
1159        spin_unlock(&inode_hash_lock);
1160        if (inode) {
1161                if (IS_ERR(inode))
1162                        return NULL;
1163                wait_on_inode(inode);
1164                if (unlikely(inode_unhashed(inode))) {
1165                        iput(inode);
1166                        goto again;
1167                }
1168                return inode;
1169        }
1170
1171        inode = alloc_inode(sb);
1172        if (inode) {
1173                struct inode *old;
1174
1175                spin_lock(&inode_hash_lock);
1176                /* We released the lock, so.. */
1177                old = find_inode_fast(sb, head, ino);
1178                if (!old) {
1179                        inode->i_ino = ino;
1180                        spin_lock(&inode->i_lock);
1181                        inode->i_state = I_NEW;
1182                        hlist_add_head(&inode->i_hash, head);
1183                        spin_unlock(&inode->i_lock);
1184                        inode_sb_list_add(inode);
1185                        spin_unlock(&inode_hash_lock);
1186
1187                        /* Return the locked inode with I_NEW set, the
1188                         * caller is responsible for filling in the contents
1189                         */
1190                        return inode;
1191                }
1192
1193                /*
1194                 * Uhhuh, somebody else created the same inode under
1195                 * us. Use the old inode instead of the one we just
1196                 * allocated.
1197                 */
1198                spin_unlock(&inode_hash_lock);
1199                destroy_inode(inode);
1200                if (IS_ERR(old))
1201                        return NULL;
1202                inode = old;
1203                wait_on_inode(inode);
1204                if (unlikely(inode_unhashed(inode))) {
1205                        iput(inode);
1206                        goto again;
1207                }
1208        }
1209        return inode;
1210}
1211EXPORT_SYMBOL(iget_locked);
1212
1213/*
1214 * search the inode cache for a matching inode number.
1215 * If we find one, then the inode number we are trying to
1216 * allocate is not unique and so we should not use it.
1217 *
1218 * Returns 1 if the inode number is unique, 0 if it is not.
1219 */
1220static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1221{
1222        struct hlist_head *b = inode_hashtable + hash(sb, ino);
1223        struct inode *inode;
1224
1225        spin_lock(&inode_hash_lock);
1226        hlist_for_each_entry(inode, b, i_hash) {
1227                if (inode->i_ino == ino && inode->i_sb == sb) {
1228                        spin_unlock(&inode_hash_lock);
1229                        return 0;
1230                }
1231        }
1232        spin_unlock(&inode_hash_lock);
1233
1234        return 1;
1235}
1236
1237/**
1238 *      iunique - get a unique inode number
1239 *      @sb: superblock
1240 *      @max_reserved: highest reserved inode number
1241 *
1242 *      Obtain an inode number that is unique on the system for a given
1243 *      superblock. This is used by file systems that have no natural
1244 *      permanent inode numbering system. An inode number is returned that
1245 *      is higher than the reserved limit but unique.
1246 *
1247 *      BUGS:
1248 *      With a large number of inodes live on the file system this function
1249 *      currently becomes quite slow.
1250 */
1251ino_t iunique(struct super_block *sb, ino_t max_reserved)
1252{
1253        /*
1254         * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1255         * error if st_ino won't fit in target struct field. Use 32bit counter
1256         * here to attempt to avoid that.
1257         */
1258        static DEFINE_SPINLOCK(iunique_lock);
1259        static unsigned int counter;
1260        ino_t res;
1261
1262        spin_lock(&iunique_lock);
1263        do {
1264                if (counter <= max_reserved)
1265                        counter = max_reserved + 1;
1266                res = counter++;
1267        } while (!test_inode_iunique(sb, res));
1268        spin_unlock(&iunique_lock);
1269
1270        return res;
1271}
1272EXPORT_SYMBOL(iunique);
1273
1274struct inode *igrab(struct inode *inode)
1275{
1276        spin_lock(&inode->i_lock);
1277        if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1278                __iget(inode);
1279                spin_unlock(&inode->i_lock);
1280        } else {
1281                spin_unlock(&inode->i_lock);
1282                /*
1283                 * Handle the case where s_op->clear_inode is not been
1284                 * called yet, and somebody is calling igrab
1285                 * while the inode is getting freed.
1286                 */
1287                inode = NULL;
1288        }
1289        return inode;
1290}
1291EXPORT_SYMBOL(igrab);
1292
1293/**
1294 * ilookup5_nowait - search for an inode in the inode cache
1295 * @sb:         super block of file system to search
1296 * @hashval:    hash value (usually inode number) to search for
1297 * @test:       callback used for comparisons between inodes
1298 * @data:       opaque data pointer to pass to @test
1299 *
1300 * Search for the inode specified by @hashval and @data in the inode cache.
1301 * If the inode is in the cache, the inode is returned with an incremented
1302 * reference count.
1303 *
1304 * Note: I_NEW is not waited upon so you have to be very careful what you do
1305 * with the returned inode.  You probably should be using ilookup5() instead.
1306 *
1307 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1308 */
1309struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1310                int (*test)(struct inode *, void *), void *data)
1311{
1312        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1313        struct inode *inode;
1314
1315        spin_lock(&inode_hash_lock);
1316        inode = find_inode(sb, head, test, data);
1317        spin_unlock(&inode_hash_lock);
1318
1319        return IS_ERR(inode) ? NULL : inode;
1320}
1321EXPORT_SYMBOL(ilookup5_nowait);
1322
1323/**
1324 * ilookup5 - search for an inode in the inode cache
1325 * @sb:         super block of file system to search
1326 * @hashval:    hash value (usually inode number) to search for
1327 * @test:       callback used for comparisons between inodes
1328 * @data:       opaque data pointer to pass to @test
1329 *
1330 * Search for the inode specified by @hashval and @data in the inode cache,
1331 * and if the inode is in the cache, return the inode with an incremented
1332 * reference count.  Waits on I_NEW before returning the inode.
1333 * returned with an incremented reference count.
1334 *
1335 * This is a generalized version of ilookup() for file systems where the
1336 * inode number is not sufficient for unique identification of an inode.
1337 *
1338 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1339 */
1340struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1341                int (*test)(struct inode *, void *), void *data)
1342{
1343        struct inode *inode;
1344again:
1345        inode = ilookup5_nowait(sb, hashval, test, data);
1346        if (inode) {
1347                wait_on_inode(inode);
1348                if (unlikely(inode_unhashed(inode))) {
1349                        iput(inode);
1350                        goto again;
1351                }
1352        }
1353        return inode;
1354}
1355EXPORT_SYMBOL(ilookup5);
1356
1357/**
1358 * ilookup - search for an inode in the inode cache
1359 * @sb:         super block of file system to search
1360 * @ino:        inode number to search for
1361 *
1362 * Search for the inode @ino in the inode cache, and if the inode is in the
1363 * cache, the inode is returned with an incremented reference count.
1364 */
1365struct inode *ilookup(struct super_block *sb, unsigned long ino)
1366{
1367        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1368        struct inode *inode;
1369again:
1370        spin_lock(&inode_hash_lock);
1371        inode = find_inode_fast(sb, head, ino);
1372        spin_unlock(&inode_hash_lock);
1373
1374        if (inode) {
1375                if (IS_ERR(inode))
1376                        return NULL;
1377                wait_on_inode(inode);
1378                if (unlikely(inode_unhashed(inode))) {
1379                        iput(inode);
1380                        goto again;
1381                }
1382        }
1383        return inode;
1384}
1385EXPORT_SYMBOL(ilookup);
1386
1387/**
1388 * find_inode_nowait - find an inode in the inode cache
1389 * @sb:         super block of file system to search
1390 * @hashval:    hash value (usually inode number) to search for
1391 * @match:      callback used for comparisons between inodes
1392 * @data:       opaque data pointer to pass to @match
1393 *
1394 * Search for the inode specified by @hashval and @data in the inode
1395 * cache, where the helper function @match will return 0 if the inode
1396 * does not match, 1 if the inode does match, and -1 if the search
1397 * should be stopped.  The @match function must be responsible for
1398 * taking the i_lock spin_lock and checking i_state for an inode being
1399 * freed or being initialized, and incrementing the reference count
1400 * before returning 1.  It also must not sleep, since it is called with
1401 * the inode_hash_lock spinlock held.
1402 *
1403 * This is a even more generalized version of ilookup5() when the
1404 * function must never block --- find_inode() can block in
1405 * __wait_on_freeing_inode() --- or when the caller can not increment
1406 * the reference count because the resulting iput() might cause an
1407 * inode eviction.  The tradeoff is that the @match funtion must be
1408 * very carefully implemented.
1409 */
1410struct inode *find_inode_nowait(struct super_block *sb,
1411                                unsigned long hashval,
1412                                int (*match)(struct inode *, unsigned long,
1413                                             void *),
1414                                void *data)
1415{
1416        struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1417        struct inode *inode, *ret_inode = NULL;
1418        int mval;
1419
1420        spin_lock(&inode_hash_lock);
1421        hlist_for_each_entry(inode, head, i_hash) {
1422                if (inode->i_sb != sb)
1423                        continue;
1424                mval = match(inode, hashval, data);
1425                if (mval == 0)
1426                        continue;
1427                if (mval == 1)
1428                        ret_inode = inode;
1429                goto out;
1430        }
1431out:
1432        spin_unlock(&inode_hash_lock);
1433        return ret_inode;
1434}
1435EXPORT_SYMBOL(find_inode_nowait);
1436
1437int insert_inode_locked(struct inode *inode)
1438{
1439        struct super_block *sb = inode->i_sb;
1440        ino_t ino = inode->i_ino;
1441        struct hlist_head *head = inode_hashtable + hash(sb, ino);
1442
1443        while (1) {
1444                struct inode *old = NULL;
1445                spin_lock(&inode_hash_lock);
1446                hlist_for_each_entry(old, head, i_hash) {
1447                        if (old->i_ino != ino)
1448                                continue;
1449                        if (old->i_sb != sb)
1450                                continue;
1451                        spin_lock(&old->i_lock);
1452                        if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1453                                spin_unlock(&old->i_lock);
1454                                continue;
1455                        }
1456                        break;
1457                }
1458                if (likely(!old)) {
1459                        spin_lock(&inode->i_lock);
1460                        inode->i_state |= I_NEW | I_CREATING;
1461                        hlist_add_head(&inode->i_hash, head);
1462                        spin_unlock(&inode->i_lock);
1463                        spin_unlock(&inode_hash_lock);
1464                        return 0;
1465                }
1466                if (unlikely(old->i_state & I_CREATING)) {
1467                        spin_unlock(&old->i_lock);
1468                        spin_unlock(&inode_hash_lock);
1469                        return -EBUSY;
1470                }
1471                __iget(old);
1472                spin_unlock(&old->i_lock);
1473                spin_unlock(&inode_hash_lock);
1474                wait_on_inode(old);
1475                if (unlikely(!inode_unhashed(old))) {
1476                        iput(old);
1477                        return -EBUSY;
1478                }
1479                iput(old);
1480        }
1481}
1482EXPORT_SYMBOL(insert_inode_locked);
1483
1484int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1485                int (*test)(struct inode *, void *), void *data)
1486{
1487        struct inode *old;
1488
1489        inode->i_state |= I_CREATING;
1490        old = inode_insert5(inode, hashval, test, NULL, data);
1491
1492        if (old != inode) {
1493                iput(old);
1494                return -EBUSY;
1495        }
1496        return 0;
1497}
1498EXPORT_SYMBOL(insert_inode_locked4);
1499
1500
1501int generic_delete_inode(struct inode *inode)
1502{
1503        return 1;
1504}
1505EXPORT_SYMBOL(generic_delete_inode);
1506
1507/*
1508 * Called when we're dropping the last reference
1509 * to an inode.
1510 *
1511 * Call the FS "drop_inode()" function, defaulting to
1512 * the legacy UNIX filesystem behaviour.  If it tells
1513 * us to evict inode, do so.  Otherwise, retain inode
1514 * in cache if fs is alive, sync and evict if fs is
1515 * shutting down.
1516 */
1517static void iput_final(struct inode *inode)
1518{
1519        struct super_block *sb = inode->i_sb;
1520        const struct super_operations *op = inode->i_sb->s_op;
1521        int drop;
1522
1523        WARN_ON(inode->i_state & I_NEW);
1524
1525        if (op->drop_inode)
1526                drop = op->drop_inode(inode);
1527        else
1528                drop = generic_drop_inode(inode);
1529
1530        if (!drop && (sb->s_flags & SB_ACTIVE)) {
1531                inode_add_lru(inode);
1532                spin_unlock(&inode->i_lock);
1533                return;
1534        }
1535
1536        if (!drop) {
1537                inode->i_state |= I_WILL_FREE;
1538                spin_unlock(&inode->i_lock);
1539                write_inode_now(inode, 1);
1540                spin_lock(&inode->i_lock);
1541                WARN_ON(inode->i_state & I_NEW);
1542                inode->i_state &= ~I_WILL_FREE;
1543        }
1544
1545        inode->i_state |= I_FREEING;
1546        if (!list_empty(&inode->i_lru))
1547                inode_lru_list_del(inode);
1548        spin_unlock(&inode->i_lock);
1549
1550        evict(inode);
1551}
1552
1553/**
1554 *      iput    - put an inode
1555 *      @inode: inode to put
1556 *
1557 *      Puts an inode, dropping its usage count. If the inode use count hits
1558 *      zero, the inode is then freed and may also be destroyed.
1559 *
1560 *      Consequently, iput() can sleep.
1561 */
1562void iput(struct inode *inode)
1563{
1564        if (!inode)
1565                return;
1566        BUG_ON(inode->i_state & I_CLEAR);
1567retry:
1568        if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1569                if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1570                        atomic_inc(&inode->i_count);
1571                        spin_unlock(&inode->i_lock);
1572                        trace_writeback_lazytime_iput(inode);
1573                        mark_inode_dirty_sync(inode);
1574                        goto retry;
1575                }
1576                iput_final(inode);
1577        }
1578}
1579EXPORT_SYMBOL(iput);
1580
1581/**
1582 *      bmap    - find a block number in a file
1583 *      @inode: inode of file
1584 *      @block: block to find
1585 *
1586 *      Returns the block number on the device holding the inode that
1587 *      is the disk block number for the block of the file requested.
1588 *      That is, asked for block 4 of inode 1 the function will return the
1589 *      disk block relative to the disk start that holds that block of the
1590 *      file.
1591 */
1592sector_t bmap(struct inode *inode, sector_t block)
1593{
1594        sector_t res = 0;
1595        if (inode->i_mapping->a_ops->bmap)
1596                res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1597        return res;
1598}
1599EXPORT_SYMBOL(bmap);
1600
1601/*
1602 * With relative atime, only update atime if the previous atime is
1603 * earlier than either the ctime or mtime or if at least a day has
1604 * passed since the last atime update.
1605 */
1606static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1607                             struct timespec now)
1608{
1609
1610        if (!(mnt->mnt_flags & MNT_RELATIME))
1611                return 1;
1612        /*
1613         * Is mtime younger than atime? If yes, update atime:
1614         */
1615        if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1616                return 1;
1617        /*
1618         * Is ctime younger than atime? If yes, update atime:
1619         */
1620        if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1621                return 1;
1622
1623        /*
1624         * Is the previous atime value older than a day? If yes,
1625         * update atime:
1626         */
1627        if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1628                return 1;
1629        /*
1630         * Good, we can skip the atime update:
1631         */
1632        return 0;
1633}
1634
1635int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1636{
1637        int iflags = I_DIRTY_TIME;
1638        bool dirty = false;
1639
1640        if (flags & S_ATIME)
1641                inode->i_atime = *time;
1642        if (flags & S_VERSION)
1643                dirty = inode_maybe_inc_iversion(inode, false);
1644        if (flags & S_CTIME)
1645                inode->i_ctime = *time;
1646        if (flags & S_MTIME)
1647                inode->i_mtime = *time;
1648        if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1649            !(inode->i_sb->s_flags & SB_LAZYTIME))
1650                dirty = true;
1651
1652        if (dirty)
1653                iflags |= I_DIRTY_SYNC;
1654        __mark_inode_dirty(inode, iflags);
1655        return 0;
1656}
1657EXPORT_SYMBOL(generic_update_time);
1658
1659/*
1660 * This does the actual work of updating an inodes time or version.  Must have
1661 * had called mnt_want_write() before calling this.
1662 */
1663static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1664{
1665        int (*update_time)(struct inode *, struct timespec64 *, int);
1666
1667        update_time = inode->i_op->update_time ? inode->i_op->update_time :
1668                generic_update_time;
1669
1670        return update_time(inode, time, flags);
1671}
1672
1673/**
1674 *      touch_atime     -       update the access time
1675 *      @path: the &struct path to update
1676 *      @inode: inode to update
1677 *
1678 *      Update the accessed time on an inode and mark it for writeback.
1679 *      This function automatically handles read only file systems and media,
1680 *      as well as the "noatime" flag and inode specific "noatime" markers.
1681 */
1682bool atime_needs_update(const struct path *path, struct inode *inode)
1683{
1684        struct vfsmount *mnt = path->mnt;
1685        struct timespec64 now;
1686
1687        if (inode->i_flags & S_NOATIME)
1688                return false;
1689
1690        /* Atime updates will likely cause i_uid and i_gid to be written
1691         * back improprely if their true value is unknown to the vfs.
1692         */
1693        if (HAS_UNMAPPED_ID(inode))
1694                return false;
1695
1696        if (IS_NOATIME(inode))
1697                return false;
1698        if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1699                return false;
1700
1701        if (mnt->mnt_flags & MNT_NOATIME)
1702                return false;
1703        if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1704                return false;
1705
1706        now = current_time(inode);
1707
1708        if (!relatime_need_update(mnt, inode, timespec64_to_timespec(now)))
1709                return false;
1710
1711        if (timespec64_equal(&inode->i_atime, &now))
1712                return false;
1713
1714        return true;
1715}
1716
1717void touch_atime(const struct path *path)
1718{
1719        struct vfsmount *mnt = path->mnt;
1720        struct inode *inode = d_inode(path->dentry);
1721        struct timespec64 now;
1722
1723        if (!atime_needs_update(path, inode))
1724                return;
1725
1726        if (!sb_start_write_trylock(inode->i_sb))
1727                return;
1728
1729        if (__mnt_want_write(mnt) != 0)
1730                goto skip_update;
1731        /*
1732         * File systems can error out when updating inodes if they need to
1733         * allocate new space to modify an inode (such is the case for
1734         * Btrfs), but since we touch atime while walking down the path we
1735         * really don't care if we failed to update the atime of the file,
1736         * so just ignore the return value.
1737         * We may also fail on filesystems that have the ability to make parts
1738         * of the fs read only, e.g. subvolumes in Btrfs.
1739         */
1740        now = current_time(inode);
1741        update_time(inode, &now, S_ATIME);
1742        __mnt_drop_write(mnt);
1743skip_update:
1744        sb_end_write(inode->i_sb);
1745}
1746EXPORT_SYMBOL(touch_atime);
1747
1748/*
1749 * The logic we want is
1750 *
1751 *      if suid or (sgid and xgrp)
1752 *              remove privs
1753 */
1754int should_remove_suid(struct dentry *dentry)
1755{
1756        umode_t mode = d_inode(dentry)->i_mode;
1757        int kill = 0;
1758
1759        /* suid always must be killed */
1760        if (unlikely(mode & S_ISUID))
1761                kill = ATTR_KILL_SUID;
1762
1763        /*
1764         * sgid without any exec bits is just a mandatory locking mark; leave
1765         * it alone.  If some exec bits are set, it's a real sgid; kill it.
1766         */
1767        if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1768                kill |= ATTR_KILL_SGID;
1769
1770        if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1771                return kill;
1772
1773        return 0;
1774}
1775EXPORT_SYMBOL(should_remove_suid);
1776
1777/*
1778 * Return mask of changes for notify_change() that need to be done as a
1779 * response to write or truncate. Return 0 if nothing has to be changed.
1780 * Negative value on error (change should be denied).
1781 */
1782int dentry_needs_remove_privs(struct dentry *dentry)
1783{
1784        struct inode *inode = d_inode(dentry);
1785        int mask = 0;
1786        int ret;
1787
1788        if (IS_NOSEC(inode))
1789                return 0;
1790
1791        mask = should_remove_suid(dentry);
1792        ret = security_inode_need_killpriv(dentry);
1793        if (ret < 0)
1794                return ret;
1795        if (ret)
1796                mask |= ATTR_KILL_PRIV;
1797        return mask;
1798}
1799
1800static int __remove_privs(struct dentry *dentry, int kill)
1801{
1802        struct iattr newattrs;
1803
1804        newattrs.ia_valid = ATTR_FORCE | kill;
1805        /*
1806         * Note we call this on write, so notify_change will not
1807         * encounter any conflicting delegations:
1808         */
1809        return notify_change(dentry, &newattrs, NULL);
1810}
1811
1812/*
1813 * Remove special file priviledges (suid, capabilities) when file is written
1814 * to or truncated.
1815 */
1816int file_remove_privs(struct file *file)
1817{
1818        struct dentry *dentry = file_dentry(file);
1819        struct inode *inode = file_inode(file);
1820        int kill;
1821        int error = 0;
1822
1823        /* Fast path for nothing security related */
1824        if (IS_NOSEC(inode))
1825                return 0;
1826
1827        kill = dentry_needs_remove_privs(dentry);
1828        if (kill < 0)
1829                return kill;
1830        if (kill)
1831                error = __remove_privs(dentry, kill);
1832        if (!error)
1833                inode_has_no_xattr(inode);
1834
1835        return error;
1836}
1837EXPORT_SYMBOL(file_remove_privs);
1838
1839/**
1840 *      file_update_time        -       update mtime and ctime time
1841 *      @file: file accessed
1842 *
1843 *      Update the mtime and ctime members of an inode and mark the inode
1844 *      for writeback.  Note that this function is meant exclusively for
1845 *      usage in the file write path of filesystems, and filesystems may
1846 *      choose to explicitly ignore update via this function with the
1847 *      S_NOCMTIME inode flag, e.g. for network filesystem where these
1848 *      timestamps are handled by the server.  This can return an error for
1849 *      file systems who need to allocate space in order to update an inode.
1850 */
1851
1852int file_update_time(struct file *file)
1853{
1854        struct inode *inode = file_inode(file);
1855        struct timespec64 now;
1856        int sync_it = 0;
1857        int ret;
1858
1859        /* First try to exhaust all avenues to not sync */
1860        if (IS_NOCMTIME(inode))
1861                return 0;
1862
1863        now = current_time(inode);
1864        if (!timespec64_equal(&inode->i_mtime, &now))
1865                sync_it = S_MTIME;
1866
1867        if (!timespec64_equal(&inode->i_ctime, &now))
1868                sync_it |= S_CTIME;
1869
1870        if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1871                sync_it |= S_VERSION;
1872
1873        if (!sync_it)
1874                return 0;
1875
1876        /* Finally allowed to write? Takes lock. */
1877        if (__mnt_want_write_file(file))
1878                return 0;
1879
1880        ret = update_time(inode, &now, sync_it);
1881        __mnt_drop_write_file(file);
1882
1883        return ret;
1884}
1885EXPORT_SYMBOL(file_update_time);
1886
1887int inode_needs_sync(struct inode *inode)
1888{
1889        if (IS_SYNC(inode))
1890                return 1;
1891        if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1892                return 1;
1893        return 0;
1894}
1895EXPORT_SYMBOL(inode_needs_sync);
1896
1897/*
1898 * If we try to find an inode in the inode hash while it is being
1899 * deleted, we have to wait until the filesystem completes its
1900 * deletion before reporting that it isn't found.  This function waits
1901 * until the deletion _might_ have completed.  Callers are responsible
1902 * to recheck inode state.
1903 *
1904 * It doesn't matter if I_NEW is not set initially, a call to
1905 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1906 * will DTRT.
1907 */
1908static void __wait_on_freeing_inode(struct inode *inode)
1909{
1910        wait_queue_head_t *wq;
1911        DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1912        wq = bit_waitqueue(&inode->i_state, __I_NEW);
1913        prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1914        spin_unlock(&inode->i_lock);
1915        spin_unlock(&inode_hash_lock);
1916        schedule();
1917        finish_wait(wq, &wait.wq_entry);
1918        spin_lock(&inode_hash_lock);
1919}
1920
1921static __initdata unsigned long ihash_entries;
1922static int __init set_ihash_entries(char *str)
1923{
1924        if (!str)
1925                return 0;
1926        ihash_entries = simple_strtoul(str, &str, 0);
1927        return 1;
1928}
1929__setup("ihash_entries=", set_ihash_entries);
1930
1931/*
1932 * Initialize the waitqueues and inode hash table.
1933 */
1934void __init inode_init_early(void)
1935{
1936        /* If hashes are distributed across NUMA nodes, defer
1937         * hash allocation until vmalloc space is available.
1938         */
1939        if (hashdist)
1940                return;
1941
1942        inode_hashtable =
1943                alloc_large_system_hash("Inode-cache",
1944                                        sizeof(struct hlist_head),
1945                                        ihash_entries,
1946                                        14,
1947                                        HASH_EARLY | HASH_ZERO,
1948                                        &i_hash_shift,
1949                                        &i_hash_mask,
1950                                        0,
1951                                        0);
1952}
1953
1954void __init inode_init(void)
1955{
1956        /* inode slab cache */
1957        inode_cachep = kmem_cache_create("inode_cache",
1958                                         sizeof(struct inode),
1959                                         0,
1960                                         (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1961                                         SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1962                                         init_once);
1963
1964        /* Hash may have been set up in inode_init_early */
1965        if (!hashdist)
1966                return;
1967
1968        inode_hashtable =
1969                alloc_large_system_hash("Inode-cache",
1970                                        sizeof(struct hlist_head),
1971                                        ihash_entries,
1972                                        14,
1973                                        HASH_ZERO,
1974                                        &i_hash_shift,
1975                                        &i_hash_mask,
1976                                        0,
1977                                        0);
1978}
1979
1980void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1981{
1982        inode->i_mode = mode;
1983        if (S_ISCHR(mode)) {
1984                inode->i_fop = &def_chr_fops;
1985                inode->i_rdev = rdev;
1986        } else if (S_ISBLK(mode)) {
1987                inode->i_fop = &def_blk_fops;
1988                inode->i_rdev = rdev;
1989        } else if (S_ISFIFO(mode))
1990                inode->i_fop = &pipefifo_fops;
1991        else if (S_ISSOCK(mode))
1992                ;       /* leave it no_open_fops */
1993        else
1994                printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1995                                  " inode %s:%lu\n", mode, inode->i_sb->s_id,
1996                                  inode->i_ino);
1997}
1998EXPORT_SYMBOL(init_special_inode);
1999
2000/**
2001 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2002 * @inode: New inode
2003 * @dir: Directory inode
2004 * @mode: mode of the new inode
2005 */
2006void inode_init_owner(struct inode *inode, const struct inode *dir,
2007                        umode_t mode)
2008{
2009        inode->i_uid = current_fsuid();
2010        if (dir && dir->i_mode & S_ISGID) {
2011                inode->i_gid = dir->i_gid;
2012
2013                /* Directories are special, and always inherit S_ISGID */
2014                if (S_ISDIR(mode))
2015                        mode |= S_ISGID;
2016                else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2017                         !in_group_p(inode->i_gid) &&
2018                         !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2019                        mode &= ~S_ISGID;
2020        } else
2021                inode->i_gid = current_fsgid();
2022        inode->i_mode = mode;
2023}
2024EXPORT_SYMBOL(inode_init_owner);
2025
2026/**
2027 * inode_owner_or_capable - check current task permissions to inode
2028 * @inode: inode being checked
2029 *
2030 * Return true if current either has CAP_FOWNER in a namespace with the
2031 * inode owner uid mapped, or owns the file.
2032 */
2033bool inode_owner_or_capable(const struct inode *inode)
2034{
2035        struct user_namespace *ns;
2036
2037        if (uid_eq(current_fsuid(), inode->i_uid))
2038                return true;
2039
2040        ns = current_user_ns();
2041        if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2042                return true;
2043        return false;
2044}
2045EXPORT_SYMBOL(inode_owner_or_capable);
2046
2047/*
2048 * Direct i/o helper functions
2049 */
2050static void __inode_dio_wait(struct inode *inode)
2051{
2052        wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2053        DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2054
2055        do {
2056                prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2057                if (atomic_read(&inode->i_dio_count))
2058                        schedule();
2059        } while (atomic_read(&inode->i_dio_count));
2060        finish_wait(wq, &q.wq_entry);
2061}
2062
2063/**
2064 * inode_dio_wait - wait for outstanding DIO requests to finish
2065 * @inode: inode to wait for
2066 *
2067 * Waits for all pending direct I/O requests to finish so that we can
2068 * proceed with a truncate or equivalent operation.
2069 *
2070 * Must be called under a lock that serializes taking new references
2071 * to i_dio_count, usually by inode->i_mutex.
2072 */
2073void inode_dio_wait(struct inode *inode)
2074{
2075        if (atomic_read(&inode->i_dio_count))
2076                __inode_dio_wait(inode);
2077}
2078EXPORT_SYMBOL(inode_dio_wait);
2079
2080/*
2081 * inode_set_flags - atomically set some inode flags
2082 *
2083 * Note: the caller should be holding i_mutex, or else be sure that
2084 * they have exclusive access to the inode structure (i.e., while the
2085 * inode is being instantiated).  The reason for the cmpxchg() loop
2086 * --- which wouldn't be necessary if all code paths which modify
2087 * i_flags actually followed this rule, is that there is at least one
2088 * code path which doesn't today so we use cmpxchg() out of an abundance
2089 * of caution.
2090 *
2091 * In the long run, i_mutex is overkill, and we should probably look
2092 * at using the i_lock spinlock to protect i_flags, and then make sure
2093 * it is so documented in include/linux/fs.h and that all code follows
2094 * the locking convention!!
2095 */
2096void inode_set_flags(struct inode *inode, unsigned int flags,
2097                     unsigned int mask)
2098{
2099        unsigned int old_flags, new_flags;
2100
2101        WARN_ON_ONCE(flags & ~mask);
2102        do {
2103                old_flags = READ_ONCE(inode->i_flags);
2104                new_flags = (old_flags & ~mask) | flags;
2105        } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2106                                  new_flags) != old_flags));
2107}
2108EXPORT_SYMBOL(inode_set_flags);
2109
2110void inode_nohighmem(struct inode *inode)
2111{
2112        mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2113}
2114EXPORT_SYMBOL(inode_nohighmem);
2115
2116/**
2117 * timespec64_trunc - Truncate timespec64 to a granularity
2118 * @t: Timespec64
2119 * @gran: Granularity in ns.
2120 *
2121 * Truncate a timespec64 to a granularity. Always rounds down. gran must
2122 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2123 */
2124struct timespec64 timespec64_trunc(struct timespec64 t, unsigned gran)
2125{
2126        /* Avoid division in the common cases 1 ns and 1 s. */
2127        if (gran == 1) {
2128                /* nothing */
2129        } else if (gran == NSEC_PER_SEC) {
2130                t.tv_nsec = 0;
2131        } else if (gran > 1 && gran < NSEC_PER_SEC) {
2132                t.tv_nsec -= t.tv_nsec % gran;
2133        } else {
2134                WARN(1, "illegal file time granularity: %u", gran);
2135        }
2136        return t;
2137}
2138EXPORT_SYMBOL(timespec64_trunc);
2139
2140/**
2141 * current_time - Return FS time
2142 * @inode: inode.
2143 *
2144 * Return the current time truncated to the time granularity supported by
2145 * the fs.
2146 *
2147 * Note that inode and inode->sb cannot be NULL.
2148 * Otherwise, the function warns and returns time without truncation.
2149 */
2150struct timespec64 current_time(struct inode *inode)
2151{
2152        struct timespec64 now = current_kernel_time64();
2153
2154        if (unlikely(!inode->i_sb)) {
2155                WARN(1, "current_time() called with uninitialized super_block in the inode");
2156                return now;
2157        }
2158
2159        return timespec64_trunc(now, inode->i_sb->s_time_gran);
2160}
2161EXPORT_SYMBOL(current_time);
2162