linux/fs/dcache.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * fs/dcache.c
   4 *
   5 * Complete reimplementation
   6 * (C) 1997 Thomas Schoebel-Theuer,
   7 * with heavy changes by Linus Torvalds
   8 */
   9
  10/*
  11 * Notes on the allocation strategy:
  12 *
  13 * The dcache is a master of the icache - whenever a dcache entry
  14 * exists, the inode will always exist. "iput()" is done either when
  15 * the dcache entry is deleted or garbage collected.
  16 */
  17
  18#include <linux/ratelimit.h>
  19#include <linux/string.h>
  20#include <linux/mm.h>
  21#include <linux/fs.h>
  22#include <linux/fscrypt.h>
  23#include <linux/fsnotify.h>
  24#include <linux/slab.h>
  25#include <linux/init.h>
  26#include <linux/hash.h>
  27#include <linux/cache.h>
  28#include <linux/export.h>
  29#include <linux/security.h>
  30#include <linux/seqlock.h>
  31#include <linux/memblock.h>
  32#include <linux/bit_spinlock.h>
  33#include <linux/rculist_bl.h>
  34#include <linux/list_lru.h>
  35#include "internal.h"
  36#include "mount.h"
  37
  38/*
  39 * Usage:
  40 * dcache->d_inode->i_lock protects:
  41 *   - i_dentry, d_u.d_alias, d_inode of aliases
  42 * dcache_hash_bucket lock protects:
  43 *   - the dcache hash table
  44 * s_roots bl list spinlock protects:
  45 *   - the s_roots list (see __d_drop)
  46 * dentry->d_sb->s_dentry_lru_lock protects:
  47 *   - the dcache lru lists and counters
  48 * d_lock protects:
  49 *   - d_flags
  50 *   - d_name
  51 *   - d_lru
  52 *   - d_count
  53 *   - d_unhashed()
  54 *   - d_parent and d_subdirs
  55 *   - childrens' d_child and d_parent
  56 *   - d_u.d_alias, d_inode
  57 *
  58 * Ordering:
  59 * dentry->d_inode->i_lock
  60 *   dentry->d_lock
  61 *     dentry->d_sb->s_dentry_lru_lock
  62 *     dcache_hash_bucket lock
  63 *     s_roots lock
  64 *
  65 * If there is an ancestor relationship:
  66 * dentry->d_parent->...->d_parent->d_lock
  67 *   ...
  68 *     dentry->d_parent->d_lock
  69 *       dentry->d_lock
  70 *
  71 * If no ancestor relationship:
  72 * arbitrary, since it's serialized on rename_lock
  73 */
  74int sysctl_vfs_cache_pressure __read_mostly = 100;
  75EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
  76
  77__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
  78
  79EXPORT_SYMBOL(rename_lock);
  80
  81static struct kmem_cache *dentry_cache __read_mostly;
  82
  83const struct qstr empty_name = QSTR_INIT("", 0);
  84EXPORT_SYMBOL(empty_name);
  85const struct qstr slash_name = QSTR_INIT("/", 1);
  86EXPORT_SYMBOL(slash_name);
  87
  88/*
  89 * This is the single most critical data structure when it comes
  90 * to the dcache: the hashtable for lookups. Somebody should try
  91 * to make this good - I've just made it work.
  92 *
  93 * This hash-function tries to avoid losing too many bits of hash
  94 * information, yet avoid using a prime hash-size or similar.
  95 */
  96
  97static unsigned int d_hash_shift __read_mostly;
  98
  99static struct hlist_bl_head *dentry_hashtable __read_mostly;
 100
 101static inline struct hlist_bl_head *d_hash(unsigned int hash)
 102{
 103        return dentry_hashtable + (hash >> d_hash_shift);
 104}
 105
 106#define IN_LOOKUP_SHIFT 10
 107static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
 108
 109static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
 110                                        unsigned int hash)
 111{
 112        hash += (unsigned long) parent / L1_CACHE_BYTES;
 113        return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
 114}
 115
 116
 117/* Statistics gathering. */
 118struct dentry_stat_t dentry_stat = {
 119        .age_limit = 45,
 120};
 121
 122static DEFINE_PER_CPU(long, nr_dentry);
 123static DEFINE_PER_CPU(long, nr_dentry_unused);
 124static DEFINE_PER_CPU(long, nr_dentry_negative);
 125
 126#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
 127
 128/*
 129 * Here we resort to our own counters instead of using generic per-cpu counters
 130 * for consistency with what the vfs inode code does. We are expected to harvest
 131 * better code and performance by having our own specialized counters.
 132 *
 133 * Please note that the loop is done over all possible CPUs, not over all online
 134 * CPUs. The reason for this is that we don't want to play games with CPUs going
 135 * on and off. If one of them goes off, we will just keep their counters.
 136 *
 137 * glommer: See cffbc8a for details, and if you ever intend to change this,
 138 * please update all vfs counters to match.
 139 */
 140static long get_nr_dentry(void)
 141{
 142        int i;
 143        long sum = 0;
 144        for_each_possible_cpu(i)
 145                sum += per_cpu(nr_dentry, i);
 146        return sum < 0 ? 0 : sum;
 147}
 148
 149static long get_nr_dentry_unused(void)
 150{
 151        int i;
 152        long sum = 0;
 153        for_each_possible_cpu(i)
 154                sum += per_cpu(nr_dentry_unused, i);
 155        return sum < 0 ? 0 : sum;
 156}
 157
 158static long get_nr_dentry_negative(void)
 159{
 160        int i;
 161        long sum = 0;
 162
 163        for_each_possible_cpu(i)
 164                sum += per_cpu(nr_dentry_negative, i);
 165        return sum < 0 ? 0 : sum;
 166}
 167
 168int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
 169                   size_t *lenp, loff_t *ppos)
 170{
 171        dentry_stat.nr_dentry = get_nr_dentry();
 172        dentry_stat.nr_unused = get_nr_dentry_unused();
 173        dentry_stat.nr_negative = get_nr_dentry_negative();
 174        return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 175}
 176#endif
 177
 178/*
 179 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
 180 * The strings are both count bytes long, and count is non-zero.
 181 */
 182#ifdef CONFIG_DCACHE_WORD_ACCESS
 183
 184#include <asm/word-at-a-time.h>
 185/*
 186 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
 187 * aligned allocation for this particular component. We don't
 188 * strictly need the load_unaligned_zeropad() safety, but it
 189 * doesn't hurt either.
 190 *
 191 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
 192 * need the careful unaligned handling.
 193 */
 194static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
 195{
 196        unsigned long a,b,mask;
 197
 198        for (;;) {
 199                a = read_word_at_a_time(cs);
 200                b = load_unaligned_zeropad(ct);
 201                if (tcount < sizeof(unsigned long))
 202                        break;
 203                if (unlikely(a != b))
 204                        return 1;
 205                cs += sizeof(unsigned long);
 206                ct += sizeof(unsigned long);
 207                tcount -= sizeof(unsigned long);
 208                if (!tcount)
 209                        return 0;
 210        }
 211        mask = bytemask_from_count(tcount);
 212        return unlikely(!!((a ^ b) & mask));
 213}
 214
 215#else
 216
 217static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
 218{
 219        do {
 220                if (*cs != *ct)
 221                        return 1;
 222                cs++;
 223                ct++;
 224                tcount--;
 225        } while (tcount);
 226        return 0;
 227}
 228
 229#endif
 230
 231static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
 232{
 233        /*
 234         * Be careful about RCU walk racing with rename:
 235         * use 'READ_ONCE' to fetch the name pointer.
 236         *
 237         * NOTE! Even if a rename will mean that the length
 238         * was not loaded atomically, we don't care. The
 239         * RCU walk will check the sequence count eventually,
 240         * and catch it. And we won't overrun the buffer,
 241         * because we're reading the name pointer atomically,
 242         * and a dentry name is guaranteed to be properly
 243         * terminated with a NUL byte.
 244         *
 245         * End result: even if 'len' is wrong, we'll exit
 246         * early because the data cannot match (there can
 247         * be no NUL in the ct/tcount data)
 248         */
 249        const unsigned char *cs = READ_ONCE(dentry->d_name.name);
 250
 251        return dentry_string_cmp(cs, ct, tcount);
 252}
 253
 254struct external_name {
 255        union {
 256                atomic_t count;
 257                struct rcu_head head;
 258        } u;
 259        unsigned char name[];
 260};
 261
 262static inline struct external_name *external_name(struct dentry *dentry)
 263{
 264        return container_of(dentry->d_name.name, struct external_name, name[0]);
 265}
 266
 267static void __d_free(struct rcu_head *head)
 268{
 269        struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
 270
 271        kmem_cache_free(dentry_cache, dentry); 
 272}
 273
 274static void __d_free_external(struct rcu_head *head)
 275{
 276        struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
 277        kfree(external_name(dentry));
 278        kmem_cache_free(dentry_cache, dentry);
 279}
 280
 281static inline int dname_external(const struct dentry *dentry)
 282{
 283        return dentry->d_name.name != dentry->d_iname;
 284}
 285
 286void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
 287{
 288        spin_lock(&dentry->d_lock);
 289        name->name = dentry->d_name;
 290        if (unlikely(dname_external(dentry))) {
 291                atomic_inc(&external_name(dentry)->u.count);
 292        } else {
 293                memcpy(name->inline_name, dentry->d_iname,
 294                       dentry->d_name.len + 1);
 295                name->name.name = name->inline_name;
 296        }
 297        spin_unlock(&dentry->d_lock);
 298}
 299EXPORT_SYMBOL(take_dentry_name_snapshot);
 300
 301void release_dentry_name_snapshot(struct name_snapshot *name)
 302{
 303        if (unlikely(name->name.name != name->inline_name)) {
 304                struct external_name *p;
 305                p = container_of(name->name.name, struct external_name, name[0]);
 306                if (unlikely(atomic_dec_and_test(&p->u.count)))
 307                        kfree_rcu(p, u.head);
 308        }
 309}
 310EXPORT_SYMBOL(release_dentry_name_snapshot);
 311
 312static inline void __d_set_inode_and_type(struct dentry *dentry,
 313                                          struct inode *inode,
 314                                          unsigned type_flags)
 315{
 316        unsigned flags;
 317
 318        dentry->d_inode = inode;
 319        flags = READ_ONCE(dentry->d_flags);
 320        flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
 321        flags |= type_flags;
 322        WRITE_ONCE(dentry->d_flags, flags);
 323}
 324
 325static inline void __d_clear_type_and_inode(struct dentry *dentry)
 326{
 327        unsigned flags = READ_ONCE(dentry->d_flags);
 328
 329        flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
 330        WRITE_ONCE(dentry->d_flags, flags);
 331        dentry->d_inode = NULL;
 332        if (dentry->d_flags & DCACHE_LRU_LIST)
 333                this_cpu_inc(nr_dentry_negative);
 334}
 335
 336static void dentry_free(struct dentry *dentry)
 337{
 338        WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
 339        if (unlikely(dname_external(dentry))) {
 340                struct external_name *p = external_name(dentry);
 341                if (likely(atomic_dec_and_test(&p->u.count))) {
 342                        call_rcu(&dentry->d_u.d_rcu, __d_free_external);
 343                        return;
 344                }
 345        }
 346        /* if dentry was never visible to RCU, immediate free is OK */
 347        if (dentry->d_flags & DCACHE_NORCU)
 348                __d_free(&dentry->d_u.d_rcu);
 349        else
 350                call_rcu(&dentry->d_u.d_rcu, __d_free);
 351}
 352
 353/*
 354 * Release the dentry's inode, using the filesystem
 355 * d_iput() operation if defined.
 356 */
 357static void dentry_unlink_inode(struct dentry * dentry)
 358        __releases(dentry->d_lock)
 359        __releases(dentry->d_inode->i_lock)
 360{
 361        struct inode *inode = dentry->d_inode;
 362
 363        raw_write_seqcount_begin(&dentry->d_seq);
 364        __d_clear_type_and_inode(dentry);
 365        hlist_del_init(&dentry->d_u.d_alias);
 366        raw_write_seqcount_end(&dentry->d_seq);
 367        spin_unlock(&dentry->d_lock);
 368        spin_unlock(&inode->i_lock);
 369        if (!inode->i_nlink)
 370                fsnotify_inoderemove(inode);
 371        if (dentry->d_op && dentry->d_op->d_iput)
 372                dentry->d_op->d_iput(dentry, inode);
 373        else
 374                iput(inode);
 375}
 376
 377/*
 378 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
 379 * is in use - which includes both the "real" per-superblock
 380 * LRU list _and_ the DCACHE_SHRINK_LIST use.
 381 *
 382 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
 383 * on the shrink list (ie not on the superblock LRU list).
 384 *
 385 * The per-cpu "nr_dentry_unused" counters are updated with
 386 * the DCACHE_LRU_LIST bit.
 387 *
 388 * The per-cpu "nr_dentry_negative" counters are only updated
 389 * when deleted from or added to the per-superblock LRU list, not
 390 * from/to the shrink list. That is to avoid an unneeded dec/inc
 391 * pair when moving from LRU to shrink list in select_collect().
 392 *
 393 * These helper functions make sure we always follow the
 394 * rules. d_lock must be held by the caller.
 395 */
 396#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
 397static void d_lru_add(struct dentry *dentry)
 398{
 399        D_FLAG_VERIFY(dentry, 0);
 400        dentry->d_flags |= DCACHE_LRU_LIST;
 401        this_cpu_inc(nr_dentry_unused);
 402        if (d_is_negative(dentry))
 403                this_cpu_inc(nr_dentry_negative);
 404        WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
 405}
 406
 407static void d_lru_del(struct dentry *dentry)
 408{
 409        D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
 410        dentry->d_flags &= ~DCACHE_LRU_LIST;
 411        this_cpu_dec(nr_dentry_unused);
 412        if (d_is_negative(dentry))
 413                this_cpu_dec(nr_dentry_negative);
 414        WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
 415}
 416
 417static void d_shrink_del(struct dentry *dentry)
 418{
 419        D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
 420        list_del_init(&dentry->d_lru);
 421        dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
 422        this_cpu_dec(nr_dentry_unused);
 423}
 424
 425static void d_shrink_add(struct dentry *dentry, struct list_head *list)
 426{
 427        D_FLAG_VERIFY(dentry, 0);
 428        list_add(&dentry->d_lru, list);
 429        dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
 430        this_cpu_inc(nr_dentry_unused);
 431}
 432
 433/*
 434 * These can only be called under the global LRU lock, ie during the
 435 * callback for freeing the LRU list. "isolate" removes it from the
 436 * LRU lists entirely, while shrink_move moves it to the indicated
 437 * private list.
 438 */
 439static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
 440{
 441        D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
 442        dentry->d_flags &= ~DCACHE_LRU_LIST;
 443        this_cpu_dec(nr_dentry_unused);
 444        if (d_is_negative(dentry))
 445                this_cpu_dec(nr_dentry_negative);
 446        list_lru_isolate(lru, &dentry->d_lru);
 447}
 448
 449static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
 450                              struct list_head *list)
 451{
 452        D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
 453        dentry->d_flags |= DCACHE_SHRINK_LIST;
 454        if (d_is_negative(dentry))
 455                this_cpu_dec(nr_dentry_negative);
 456        list_lru_isolate_move(lru, &dentry->d_lru, list);
 457}
 458
 459/**
 460 * d_drop - drop a dentry
 461 * @dentry: dentry to drop
 462 *
 463 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
 464 * be found through a VFS lookup any more. Note that this is different from
 465 * deleting the dentry - d_delete will try to mark the dentry negative if
 466 * possible, giving a successful _negative_ lookup, while d_drop will
 467 * just make the cache lookup fail.
 468 *
 469 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
 470 * reason (NFS timeouts or autofs deletes).
 471 *
 472 * __d_drop requires dentry->d_lock
 473 * ___d_drop doesn't mark dentry as "unhashed"
 474 *   (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
 475 */
 476static void ___d_drop(struct dentry *dentry)
 477{
 478        struct hlist_bl_head *b;
 479        /*
 480         * Hashed dentries are normally on the dentry hashtable,
 481         * with the exception of those newly allocated by
 482         * d_obtain_root, which are always IS_ROOT:
 483         */
 484        if (unlikely(IS_ROOT(dentry)))
 485                b = &dentry->d_sb->s_roots;
 486        else
 487                b = d_hash(dentry->d_name.hash);
 488
 489        hlist_bl_lock(b);
 490        __hlist_bl_del(&dentry->d_hash);
 491        hlist_bl_unlock(b);
 492}
 493
 494void __d_drop(struct dentry *dentry)
 495{
 496        if (!d_unhashed(dentry)) {
 497                ___d_drop(dentry);
 498                dentry->d_hash.pprev = NULL;
 499                write_seqcount_invalidate(&dentry->d_seq);
 500        }
 501}
 502EXPORT_SYMBOL(__d_drop);
 503
 504void d_drop(struct dentry *dentry)
 505{
 506        spin_lock(&dentry->d_lock);
 507        __d_drop(dentry);
 508        spin_unlock(&dentry->d_lock);
 509}
 510EXPORT_SYMBOL(d_drop);
 511
 512static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
 513{
 514        struct dentry *next;
 515        /*
 516         * Inform d_walk() and shrink_dentry_list() that we are no longer
 517         * attached to the dentry tree
 518         */
 519        dentry->d_flags |= DCACHE_DENTRY_KILLED;
 520        if (unlikely(list_empty(&dentry->d_child)))
 521                return;
 522        __list_del_entry(&dentry->d_child);
 523        /*
 524         * Cursors can move around the list of children.  While we'd been
 525         * a normal list member, it didn't matter - ->d_child.next would've
 526         * been updated.  However, from now on it won't be and for the
 527         * things like d_walk() it might end up with a nasty surprise.
 528         * Normally d_walk() doesn't care about cursors moving around -
 529         * ->d_lock on parent prevents that and since a cursor has no children
 530         * of its own, we get through it without ever unlocking the parent.
 531         * There is one exception, though - if we ascend from a child that
 532         * gets killed as soon as we unlock it, the next sibling is found
 533         * using the value left in its ->d_child.next.  And if _that_
 534         * pointed to a cursor, and cursor got moved (e.g. by lseek())
 535         * before d_walk() regains parent->d_lock, we'll end up skipping
 536         * everything the cursor had been moved past.
 537         *
 538         * Solution: make sure that the pointer left behind in ->d_child.next
 539         * points to something that won't be moving around.  I.e. skip the
 540         * cursors.
 541         */
 542        while (dentry->d_child.next != &parent->d_subdirs) {
 543                next = list_entry(dentry->d_child.next, struct dentry, d_child);
 544                if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
 545                        break;
 546                dentry->d_child.next = next->d_child.next;
 547        }
 548}
 549
 550static void __dentry_kill(struct dentry *dentry)
 551{
 552        struct dentry *parent = NULL;
 553        bool can_free = true;
 554        if (!IS_ROOT(dentry))
 555                parent = dentry->d_parent;
 556
 557        /*
 558         * The dentry is now unrecoverably dead to the world.
 559         */
 560        lockref_mark_dead(&dentry->d_lockref);
 561
 562        /*
 563         * inform the fs via d_prune that this dentry is about to be
 564         * unhashed and destroyed.
 565         */
 566        if (dentry->d_flags & DCACHE_OP_PRUNE)
 567                dentry->d_op->d_prune(dentry);
 568
 569        if (dentry->d_flags & DCACHE_LRU_LIST) {
 570                if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
 571                        d_lru_del(dentry);
 572        }
 573        /* if it was on the hash then remove it */
 574        __d_drop(dentry);
 575        dentry_unlist(dentry, parent);
 576        if (parent)
 577                spin_unlock(&parent->d_lock);
 578        if (dentry->d_inode)
 579                dentry_unlink_inode(dentry);
 580        else
 581                spin_unlock(&dentry->d_lock);
 582        this_cpu_dec(nr_dentry);
 583        if (dentry->d_op && dentry->d_op->d_release)
 584                dentry->d_op->d_release(dentry);
 585
 586        spin_lock(&dentry->d_lock);
 587        if (dentry->d_flags & DCACHE_SHRINK_LIST) {
 588                dentry->d_flags |= DCACHE_MAY_FREE;
 589                can_free = false;
 590        }
 591        spin_unlock(&dentry->d_lock);
 592        if (likely(can_free))
 593                dentry_free(dentry);
 594        cond_resched();
 595}
 596
 597static struct dentry *__lock_parent(struct dentry *dentry)
 598{
 599        struct dentry *parent;
 600        rcu_read_lock();
 601        spin_unlock(&dentry->d_lock);
 602again:
 603        parent = READ_ONCE(dentry->d_parent);
 604        spin_lock(&parent->d_lock);
 605        /*
 606         * We can't blindly lock dentry until we are sure
 607         * that we won't violate the locking order.
 608         * Any changes of dentry->d_parent must have
 609         * been done with parent->d_lock held, so
 610         * spin_lock() above is enough of a barrier
 611         * for checking if it's still our child.
 612         */
 613        if (unlikely(parent != dentry->d_parent)) {
 614                spin_unlock(&parent->d_lock);
 615                goto again;
 616        }
 617        rcu_read_unlock();
 618        if (parent != dentry)
 619                spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 620        else
 621                parent = NULL;
 622        return parent;
 623}
 624
 625static inline struct dentry *lock_parent(struct dentry *dentry)
 626{
 627        struct dentry *parent = dentry->d_parent;
 628        if (IS_ROOT(dentry))
 629                return NULL;
 630        if (likely(spin_trylock(&parent->d_lock)))
 631                return parent;
 632        return __lock_parent(dentry);
 633}
 634
 635static inline bool retain_dentry(struct dentry *dentry)
 636{
 637        WARN_ON(d_in_lookup(dentry));
 638
 639        /* Unreachable? Get rid of it */
 640        if (unlikely(d_unhashed(dentry)))
 641                return false;
 642
 643        if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
 644                return false;
 645
 646        if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
 647                if (dentry->d_op->d_delete(dentry))
 648                        return false;
 649        }
 650        /* retain; LRU fodder */
 651        dentry->d_lockref.count--;
 652        if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
 653                d_lru_add(dentry);
 654        else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
 655                dentry->d_flags |= DCACHE_REFERENCED;
 656        return true;
 657}
 658
 659/*
 660 * Finish off a dentry we've decided to kill.
 661 * dentry->d_lock must be held, returns with it unlocked.
 662 * Returns dentry requiring refcount drop, or NULL if we're done.
 663 */
 664static struct dentry *dentry_kill(struct dentry *dentry)
 665        __releases(dentry->d_lock)
 666{
 667        struct inode *inode = dentry->d_inode;
 668        struct dentry *parent = NULL;
 669
 670        if (inode && unlikely(!spin_trylock(&inode->i_lock)))
 671                goto slow_positive;
 672
 673        if (!IS_ROOT(dentry)) {
 674                parent = dentry->d_parent;
 675                if (unlikely(!spin_trylock(&parent->d_lock))) {
 676                        parent = __lock_parent(dentry);
 677                        if (likely(inode || !dentry->d_inode))
 678                                goto got_locks;
 679                        /* negative that became positive */
 680                        if (parent)
 681                                spin_unlock(&parent->d_lock);
 682                        inode = dentry->d_inode;
 683                        goto slow_positive;
 684                }
 685        }
 686        __dentry_kill(dentry);
 687        return parent;
 688
 689slow_positive:
 690        spin_unlock(&dentry->d_lock);
 691        spin_lock(&inode->i_lock);
 692        spin_lock(&dentry->d_lock);
 693        parent = lock_parent(dentry);
 694got_locks:
 695        if (unlikely(dentry->d_lockref.count != 1)) {
 696                dentry->d_lockref.count--;
 697        } else if (likely(!retain_dentry(dentry))) {
 698                __dentry_kill(dentry);
 699                return parent;
 700        }
 701        /* we are keeping it, after all */
 702        if (inode)
 703                spin_unlock(&inode->i_lock);
 704        if (parent)
 705                spin_unlock(&parent->d_lock);
 706        spin_unlock(&dentry->d_lock);
 707        return NULL;
 708}
 709
 710/*
 711 * Try to do a lockless dput(), and return whether that was successful.
 712 *
 713 * If unsuccessful, we return false, having already taken the dentry lock.
 714 *
 715 * The caller needs to hold the RCU read lock, so that the dentry is
 716 * guaranteed to stay around even if the refcount goes down to zero!
 717 */
 718static inline bool fast_dput(struct dentry *dentry)
 719{
 720        int ret;
 721        unsigned int d_flags;
 722
 723        /*
 724         * If we have a d_op->d_delete() operation, we sould not
 725         * let the dentry count go to zero, so use "put_or_lock".
 726         */
 727        if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
 728                return lockref_put_or_lock(&dentry->d_lockref);
 729
 730        /*
 731         * .. otherwise, we can try to just decrement the
 732         * lockref optimistically.
 733         */
 734        ret = lockref_put_return(&dentry->d_lockref);
 735
 736        /*
 737         * If the lockref_put_return() failed due to the lock being held
 738         * by somebody else, the fast path has failed. We will need to
 739         * get the lock, and then check the count again.
 740         */
 741        if (unlikely(ret < 0)) {
 742                spin_lock(&dentry->d_lock);
 743                if (dentry->d_lockref.count > 1) {
 744                        dentry->d_lockref.count--;
 745                        spin_unlock(&dentry->d_lock);
 746                        return true;
 747                }
 748                return false;
 749        }
 750
 751        /*
 752         * If we weren't the last ref, we're done.
 753         */
 754        if (ret)
 755                return true;
 756
 757        /*
 758         * Careful, careful. The reference count went down
 759         * to zero, but we don't hold the dentry lock, so
 760         * somebody else could get it again, and do another
 761         * dput(), and we need to not race with that.
 762         *
 763         * However, there is a very special and common case
 764         * where we don't care, because there is nothing to
 765         * do: the dentry is still hashed, it does not have
 766         * a 'delete' op, and it's referenced and already on
 767         * the LRU list.
 768         *
 769         * NOTE! Since we aren't locked, these values are
 770         * not "stable". However, it is sufficient that at
 771         * some point after we dropped the reference the
 772         * dentry was hashed and the flags had the proper
 773         * value. Other dentry users may have re-gotten
 774         * a reference to the dentry and change that, but
 775         * our work is done - we can leave the dentry
 776         * around with a zero refcount.
 777         */
 778        smp_rmb();
 779        d_flags = READ_ONCE(dentry->d_flags);
 780        d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
 781
 782        /* Nothing to do? Dropping the reference was all we needed? */
 783        if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
 784                return true;
 785
 786        /*
 787         * Not the fast normal case? Get the lock. We've already decremented
 788         * the refcount, but we'll need to re-check the situation after
 789         * getting the lock.
 790         */
 791        spin_lock(&dentry->d_lock);
 792
 793        /*
 794         * Did somebody else grab a reference to it in the meantime, and
 795         * we're no longer the last user after all? Alternatively, somebody
 796         * else could have killed it and marked it dead. Either way, we
 797         * don't need to do anything else.
 798         */
 799        if (dentry->d_lockref.count) {
 800                spin_unlock(&dentry->d_lock);
 801                return true;
 802        }
 803
 804        /*
 805         * Re-get the reference we optimistically dropped. We hold the
 806         * lock, and we just tested that it was zero, so we can just
 807         * set it to 1.
 808         */
 809        dentry->d_lockref.count = 1;
 810        return false;
 811}
 812
 813
 814/* 
 815 * This is dput
 816 *
 817 * This is complicated by the fact that we do not want to put
 818 * dentries that are no longer on any hash chain on the unused
 819 * list: we'd much rather just get rid of them immediately.
 820 *
 821 * However, that implies that we have to traverse the dentry
 822 * tree upwards to the parents which might _also_ now be
 823 * scheduled for deletion (it may have been only waiting for
 824 * its last child to go away).
 825 *
 826 * This tail recursion is done by hand as we don't want to depend
 827 * on the compiler to always get this right (gcc generally doesn't).
 828 * Real recursion would eat up our stack space.
 829 */
 830
 831/*
 832 * dput - release a dentry
 833 * @dentry: dentry to release 
 834 *
 835 * Release a dentry. This will drop the usage count and if appropriate
 836 * call the dentry unlink method as well as removing it from the queues and
 837 * releasing its resources. If the parent dentries were scheduled for release
 838 * they too may now get deleted.
 839 */
 840void dput(struct dentry *dentry)
 841{
 842        while (dentry) {
 843                might_sleep();
 844
 845                rcu_read_lock();
 846                if (likely(fast_dput(dentry))) {
 847                        rcu_read_unlock();
 848                        return;
 849                }
 850
 851                /* Slow case: now with the dentry lock held */
 852                rcu_read_unlock();
 853
 854                if (likely(retain_dentry(dentry))) {
 855                        spin_unlock(&dentry->d_lock);
 856                        return;
 857                }
 858
 859                dentry = dentry_kill(dentry);
 860        }
 861}
 862EXPORT_SYMBOL(dput);
 863
 864
 865/* This must be called with d_lock held */
 866static inline void __dget_dlock(struct dentry *dentry)
 867{
 868        dentry->d_lockref.count++;
 869}
 870
 871static inline void __dget(struct dentry *dentry)
 872{
 873        lockref_get(&dentry->d_lockref);
 874}
 875
 876struct dentry *dget_parent(struct dentry *dentry)
 877{
 878        int gotref;
 879        struct dentry *ret;
 880
 881        /*
 882         * Do optimistic parent lookup without any
 883         * locking.
 884         */
 885        rcu_read_lock();
 886        ret = READ_ONCE(dentry->d_parent);
 887        gotref = lockref_get_not_zero(&ret->d_lockref);
 888        rcu_read_unlock();
 889        if (likely(gotref)) {
 890                if (likely(ret == READ_ONCE(dentry->d_parent)))
 891                        return ret;
 892                dput(ret);
 893        }
 894
 895repeat:
 896        /*
 897         * Don't need rcu_dereference because we re-check it was correct under
 898         * the lock.
 899         */
 900        rcu_read_lock();
 901        ret = dentry->d_parent;
 902        spin_lock(&ret->d_lock);
 903        if (unlikely(ret != dentry->d_parent)) {
 904                spin_unlock(&ret->d_lock);
 905                rcu_read_unlock();
 906                goto repeat;
 907        }
 908        rcu_read_unlock();
 909        BUG_ON(!ret->d_lockref.count);
 910        ret->d_lockref.count++;
 911        spin_unlock(&ret->d_lock);
 912        return ret;
 913}
 914EXPORT_SYMBOL(dget_parent);
 915
 916static struct dentry * __d_find_any_alias(struct inode *inode)
 917{
 918        struct dentry *alias;
 919
 920        if (hlist_empty(&inode->i_dentry))
 921                return NULL;
 922        alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
 923        __dget(alias);
 924        return alias;
 925}
 926
 927/**
 928 * d_find_any_alias - find any alias for a given inode
 929 * @inode: inode to find an alias for
 930 *
 931 * If any aliases exist for the given inode, take and return a
 932 * reference for one of them.  If no aliases exist, return %NULL.
 933 */
 934struct dentry *d_find_any_alias(struct inode *inode)
 935{
 936        struct dentry *de;
 937
 938        spin_lock(&inode->i_lock);
 939        de = __d_find_any_alias(inode);
 940        spin_unlock(&inode->i_lock);
 941        return de;
 942}
 943EXPORT_SYMBOL(d_find_any_alias);
 944
 945/**
 946 * d_find_alias - grab a hashed alias of inode
 947 * @inode: inode in question
 948 *
 949 * If inode has a hashed alias, or is a directory and has any alias,
 950 * acquire the reference to alias and return it. Otherwise return NULL.
 951 * Notice that if inode is a directory there can be only one alias and
 952 * it can be unhashed only if it has no children, or if it is the root
 953 * of a filesystem, or if the directory was renamed and d_revalidate
 954 * was the first vfs operation to notice.
 955 *
 956 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
 957 * any other hashed alias over that one.
 958 */
 959static struct dentry *__d_find_alias(struct inode *inode)
 960{
 961        struct dentry *alias;
 962
 963        if (S_ISDIR(inode->i_mode))
 964                return __d_find_any_alias(inode);
 965
 966        hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 967                spin_lock(&alias->d_lock);
 968                if (!d_unhashed(alias)) {
 969                        __dget_dlock(alias);
 970                        spin_unlock(&alias->d_lock);
 971                        return alias;
 972                }
 973                spin_unlock(&alias->d_lock);
 974        }
 975        return NULL;
 976}
 977
 978struct dentry *d_find_alias(struct inode *inode)
 979{
 980        struct dentry *de = NULL;
 981
 982        if (!hlist_empty(&inode->i_dentry)) {
 983                spin_lock(&inode->i_lock);
 984                de = __d_find_alias(inode);
 985                spin_unlock(&inode->i_lock);
 986        }
 987        return de;
 988}
 989EXPORT_SYMBOL(d_find_alias);
 990
 991/*
 992 *      Try to kill dentries associated with this inode.
 993 * WARNING: you must own a reference to inode.
 994 */
 995void d_prune_aliases(struct inode *inode)
 996{
 997        struct dentry *dentry;
 998restart:
 999        spin_lock(&inode->i_lock);
1000        hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1001                spin_lock(&dentry->d_lock);
1002                if (!dentry->d_lockref.count) {
1003                        struct dentry *parent = lock_parent(dentry);
1004                        if (likely(!dentry->d_lockref.count)) {
1005                                __dentry_kill(dentry);
1006                                dput(parent);
1007                                goto restart;
1008                        }
1009                        if (parent)
1010                                spin_unlock(&parent->d_lock);
1011                }
1012                spin_unlock(&dentry->d_lock);
1013        }
1014        spin_unlock(&inode->i_lock);
1015}
1016EXPORT_SYMBOL(d_prune_aliases);
1017
1018/*
1019 * Lock a dentry from shrink list.
1020 * Called under rcu_read_lock() and dentry->d_lock; the former
1021 * guarantees that nothing we access will be freed under us.
1022 * Note that dentry is *not* protected from concurrent dentry_kill(),
1023 * d_delete(), etc.
1024 *
1025 * Return false if dentry has been disrupted or grabbed, leaving
1026 * the caller to kick it off-list.  Otherwise, return true and have
1027 * that dentry's inode and parent both locked.
1028 */
1029static bool shrink_lock_dentry(struct dentry *dentry)
1030{
1031        struct inode *inode;
1032        struct dentry *parent;
1033
1034        if (dentry->d_lockref.count)
1035                return false;
1036
1037        inode = dentry->d_inode;
1038        if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1039                spin_unlock(&dentry->d_lock);
1040                spin_lock(&inode->i_lock);
1041                spin_lock(&dentry->d_lock);
1042                if (unlikely(dentry->d_lockref.count))
1043                        goto out;
1044                /* changed inode means that somebody had grabbed it */
1045                if (unlikely(inode != dentry->d_inode))
1046                        goto out;
1047        }
1048
1049        parent = dentry->d_parent;
1050        if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1051                return true;
1052
1053        spin_unlock(&dentry->d_lock);
1054        spin_lock(&parent->d_lock);
1055        if (unlikely(parent != dentry->d_parent)) {
1056                spin_unlock(&parent->d_lock);
1057                spin_lock(&dentry->d_lock);
1058                goto out;
1059        }
1060        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1061        if (likely(!dentry->d_lockref.count))
1062                return true;
1063        spin_unlock(&parent->d_lock);
1064out:
1065        if (inode)
1066                spin_unlock(&inode->i_lock);
1067        return false;
1068}
1069
1070static void shrink_dentry_list(struct list_head *list)
1071{
1072        while (!list_empty(list)) {
1073                struct dentry *dentry, *parent;
1074
1075                dentry = list_entry(list->prev, struct dentry, d_lru);
1076                spin_lock(&dentry->d_lock);
1077                rcu_read_lock();
1078                if (!shrink_lock_dentry(dentry)) {
1079                        bool can_free = false;
1080                        rcu_read_unlock();
1081                        d_shrink_del(dentry);
1082                        if (dentry->d_lockref.count < 0)
1083                                can_free = dentry->d_flags & DCACHE_MAY_FREE;
1084                        spin_unlock(&dentry->d_lock);
1085                        if (can_free)
1086                                dentry_free(dentry);
1087                        continue;
1088                }
1089                rcu_read_unlock();
1090                d_shrink_del(dentry);
1091                parent = dentry->d_parent;
1092                __dentry_kill(dentry);
1093                if (parent == dentry)
1094                        continue;
1095                /*
1096                 * We need to prune ancestors too. This is necessary to prevent
1097                 * quadratic behavior of shrink_dcache_parent(), but is also
1098                 * expected to be beneficial in reducing dentry cache
1099                 * fragmentation.
1100                 */
1101                dentry = parent;
1102                while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
1103                        dentry = dentry_kill(dentry);
1104        }
1105}
1106
1107static enum lru_status dentry_lru_isolate(struct list_head *item,
1108                struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1109{
1110        struct list_head *freeable = arg;
1111        struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1112
1113
1114        /*
1115         * we are inverting the lru lock/dentry->d_lock here,
1116         * so use a trylock. If we fail to get the lock, just skip
1117         * it
1118         */
1119        if (!spin_trylock(&dentry->d_lock))
1120                return LRU_SKIP;
1121
1122        /*
1123         * Referenced dentries are still in use. If they have active
1124         * counts, just remove them from the LRU. Otherwise give them
1125         * another pass through the LRU.
1126         */
1127        if (dentry->d_lockref.count) {
1128                d_lru_isolate(lru, dentry);
1129                spin_unlock(&dentry->d_lock);
1130                return LRU_REMOVED;
1131        }
1132
1133        if (dentry->d_flags & DCACHE_REFERENCED) {
1134                dentry->d_flags &= ~DCACHE_REFERENCED;
1135                spin_unlock(&dentry->d_lock);
1136
1137                /*
1138                 * The list move itself will be made by the common LRU code. At
1139                 * this point, we've dropped the dentry->d_lock but keep the
1140                 * lru lock. This is safe to do, since every list movement is
1141                 * protected by the lru lock even if both locks are held.
1142                 *
1143                 * This is guaranteed by the fact that all LRU management
1144                 * functions are intermediated by the LRU API calls like
1145                 * list_lru_add and list_lru_del. List movement in this file
1146                 * only ever occur through this functions or through callbacks
1147                 * like this one, that are called from the LRU API.
1148                 *
1149                 * The only exceptions to this are functions like
1150                 * shrink_dentry_list, and code that first checks for the
1151                 * DCACHE_SHRINK_LIST flag.  Those are guaranteed to be
1152                 * operating only with stack provided lists after they are
1153                 * properly isolated from the main list.  It is thus, always a
1154                 * local access.
1155                 */
1156                return LRU_ROTATE;
1157        }
1158
1159        d_lru_shrink_move(lru, dentry, freeable);
1160        spin_unlock(&dentry->d_lock);
1161
1162        return LRU_REMOVED;
1163}
1164
1165/**
1166 * prune_dcache_sb - shrink the dcache
1167 * @sb: superblock
1168 * @sc: shrink control, passed to list_lru_shrink_walk()
1169 *
1170 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1171 * is done when we need more memory and called from the superblock shrinker
1172 * function.
1173 *
1174 * This function may fail to free any resources if all the dentries are in
1175 * use.
1176 */
1177long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1178{
1179        LIST_HEAD(dispose);
1180        long freed;
1181
1182        freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1183                                     dentry_lru_isolate, &dispose);
1184        shrink_dentry_list(&dispose);
1185        return freed;
1186}
1187
1188static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1189                struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1190{
1191        struct list_head *freeable = arg;
1192        struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1193
1194        /*
1195         * we are inverting the lru lock/dentry->d_lock here,
1196         * so use a trylock. If we fail to get the lock, just skip
1197         * it
1198         */
1199        if (!spin_trylock(&dentry->d_lock))
1200                return LRU_SKIP;
1201
1202        d_lru_shrink_move(lru, dentry, freeable);
1203        spin_unlock(&dentry->d_lock);
1204
1205        return LRU_REMOVED;
1206}
1207
1208
1209/**
1210 * shrink_dcache_sb - shrink dcache for a superblock
1211 * @sb: superblock
1212 *
1213 * Shrink the dcache for the specified super block. This is used to free
1214 * the dcache before unmounting a file system.
1215 */
1216void shrink_dcache_sb(struct super_block *sb)
1217{
1218        do {
1219                LIST_HEAD(dispose);
1220
1221                list_lru_walk(&sb->s_dentry_lru,
1222                        dentry_lru_isolate_shrink, &dispose, 1024);
1223                shrink_dentry_list(&dispose);
1224        } while (list_lru_count(&sb->s_dentry_lru) > 0);
1225}
1226EXPORT_SYMBOL(shrink_dcache_sb);
1227
1228/**
1229 * enum d_walk_ret - action to talke during tree walk
1230 * @D_WALK_CONTINUE:    contrinue walk
1231 * @D_WALK_QUIT:        quit walk
1232 * @D_WALK_NORETRY:     quit when retry is needed
1233 * @D_WALK_SKIP:        skip this dentry and its children
1234 */
1235enum d_walk_ret {
1236        D_WALK_CONTINUE,
1237        D_WALK_QUIT,
1238        D_WALK_NORETRY,
1239        D_WALK_SKIP,
1240};
1241
1242/**
1243 * d_walk - walk the dentry tree
1244 * @parent:     start of walk
1245 * @data:       data passed to @enter() and @finish()
1246 * @enter:      callback when first entering the dentry
1247 *
1248 * The @enter() callbacks are called with d_lock held.
1249 */
1250static void d_walk(struct dentry *parent, void *data,
1251                   enum d_walk_ret (*enter)(void *, struct dentry *))
1252{
1253        struct dentry *this_parent;
1254        struct list_head *next;
1255        unsigned seq = 0;
1256        enum d_walk_ret ret;
1257        bool retry = true;
1258
1259again:
1260        read_seqbegin_or_lock(&rename_lock, &seq);
1261        this_parent = parent;
1262        spin_lock(&this_parent->d_lock);
1263
1264        ret = enter(data, this_parent);
1265        switch (ret) {
1266        case D_WALK_CONTINUE:
1267                break;
1268        case D_WALK_QUIT:
1269        case D_WALK_SKIP:
1270                goto out_unlock;
1271        case D_WALK_NORETRY:
1272                retry = false;
1273                break;
1274        }
1275repeat:
1276        next = this_parent->d_subdirs.next;
1277resume:
1278        while (next != &this_parent->d_subdirs) {
1279                struct list_head *tmp = next;
1280                struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1281                next = tmp->next;
1282
1283                if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1284                        continue;
1285
1286                spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1287
1288                ret = enter(data, dentry);
1289                switch (ret) {
1290                case D_WALK_CONTINUE:
1291                        break;
1292                case D_WALK_QUIT:
1293                        spin_unlock(&dentry->d_lock);
1294                        goto out_unlock;
1295                case D_WALK_NORETRY:
1296                        retry = false;
1297                        break;
1298                case D_WALK_SKIP:
1299                        spin_unlock(&dentry->d_lock);
1300                        continue;
1301                }
1302
1303                if (!list_empty(&dentry->d_subdirs)) {
1304                        spin_unlock(&this_parent->d_lock);
1305                        spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1306                        this_parent = dentry;
1307                        spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1308                        goto repeat;
1309                }
1310                spin_unlock(&dentry->d_lock);
1311        }
1312        /*
1313         * All done at this level ... ascend and resume the search.
1314         */
1315        rcu_read_lock();
1316ascend:
1317        if (this_parent != parent) {
1318                struct dentry *child = this_parent;
1319                this_parent = child->d_parent;
1320
1321                spin_unlock(&child->d_lock);
1322                spin_lock(&this_parent->d_lock);
1323
1324                /* might go back up the wrong parent if we have had a rename. */
1325                if (need_seqretry(&rename_lock, seq))
1326                        goto rename_retry;
1327                /* go into the first sibling still alive */
1328                do {
1329                        next = child->d_child.next;
1330                        if (next == &this_parent->d_subdirs)
1331                                goto ascend;
1332                        child = list_entry(next, struct dentry, d_child);
1333                } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1334                rcu_read_unlock();
1335                goto resume;
1336        }
1337        if (need_seqretry(&rename_lock, seq))
1338                goto rename_retry;
1339        rcu_read_unlock();
1340
1341out_unlock:
1342        spin_unlock(&this_parent->d_lock);
1343        done_seqretry(&rename_lock, seq);
1344        return;
1345
1346rename_retry:
1347        spin_unlock(&this_parent->d_lock);
1348        rcu_read_unlock();
1349        BUG_ON(seq & 1);
1350        if (!retry)
1351                return;
1352        seq = 1;
1353        goto again;
1354}
1355
1356struct check_mount {
1357        struct vfsmount *mnt;
1358        unsigned int mounted;
1359};
1360
1361static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1362{
1363        struct check_mount *info = data;
1364        struct path path = { .mnt = info->mnt, .dentry = dentry };
1365
1366        if (likely(!d_mountpoint(dentry)))
1367                return D_WALK_CONTINUE;
1368        if (__path_is_mountpoint(&path)) {
1369                info->mounted = 1;
1370                return D_WALK_QUIT;
1371        }
1372        return D_WALK_CONTINUE;
1373}
1374
1375/**
1376 * path_has_submounts - check for mounts over a dentry in the
1377 *                      current namespace.
1378 * @parent: path to check.
1379 *
1380 * Return true if the parent or its subdirectories contain
1381 * a mount point in the current namespace.
1382 */
1383int path_has_submounts(const struct path *parent)
1384{
1385        struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1386
1387        read_seqlock_excl(&mount_lock);
1388        d_walk(parent->dentry, &data, path_check_mount);
1389        read_sequnlock_excl(&mount_lock);
1390
1391        return data.mounted;
1392}
1393EXPORT_SYMBOL(path_has_submounts);
1394
1395/*
1396 * Called by mount code to set a mountpoint and check if the mountpoint is
1397 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1398 * subtree can become unreachable).
1399 *
1400 * Only one of d_invalidate() and d_set_mounted() must succeed.  For
1401 * this reason take rename_lock and d_lock on dentry and ancestors.
1402 */
1403int d_set_mounted(struct dentry *dentry)
1404{
1405        struct dentry *p;
1406        int ret = -ENOENT;
1407        write_seqlock(&rename_lock);
1408        for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1409                /* Need exclusion wrt. d_invalidate() */
1410                spin_lock(&p->d_lock);
1411                if (unlikely(d_unhashed(p))) {
1412                        spin_unlock(&p->d_lock);
1413                        goto out;
1414                }
1415                spin_unlock(&p->d_lock);
1416        }
1417        spin_lock(&dentry->d_lock);
1418        if (!d_unlinked(dentry)) {
1419                ret = -EBUSY;
1420                if (!d_mountpoint(dentry)) {
1421                        dentry->d_flags |= DCACHE_MOUNTED;
1422                        ret = 0;
1423                }
1424        }
1425        spin_unlock(&dentry->d_lock);
1426out:
1427        write_sequnlock(&rename_lock);
1428        return ret;
1429}
1430
1431/*
1432 * Search the dentry child list of the specified parent,
1433 * and move any unused dentries to the end of the unused
1434 * list for prune_dcache(). We descend to the next level
1435 * whenever the d_subdirs list is non-empty and continue
1436 * searching.
1437 *
1438 * It returns zero iff there are no unused children,
1439 * otherwise  it returns the number of children moved to
1440 * the end of the unused list. This may not be the total
1441 * number of unused children, because select_parent can
1442 * drop the lock and return early due to latency
1443 * constraints.
1444 */
1445
1446struct select_data {
1447        struct dentry *start;
1448        struct list_head dispose;
1449        int found;
1450};
1451
1452static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1453{
1454        struct select_data *data = _data;
1455        enum d_walk_ret ret = D_WALK_CONTINUE;
1456
1457        if (data->start == dentry)
1458                goto out;
1459
1460        if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1461                data->found++;
1462        } else {
1463                if (dentry->d_flags & DCACHE_LRU_LIST)
1464                        d_lru_del(dentry);
1465                if (!dentry->d_lockref.count) {
1466                        d_shrink_add(dentry, &data->dispose);
1467                        data->found++;
1468                }
1469        }
1470        /*
1471         * We can return to the caller if we have found some (this
1472         * ensures forward progress). We'll be coming back to find
1473         * the rest.
1474         */
1475        if (!list_empty(&data->dispose))
1476                ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1477out:
1478        return ret;
1479}
1480
1481/**
1482 * shrink_dcache_parent - prune dcache
1483 * @parent: parent of entries to prune
1484 *
1485 * Prune the dcache to remove unused children of the parent dentry.
1486 */
1487void shrink_dcache_parent(struct dentry *parent)
1488{
1489        for (;;) {
1490                struct select_data data;
1491
1492                INIT_LIST_HEAD(&data.dispose);
1493                data.start = parent;
1494                data.found = 0;
1495
1496                d_walk(parent, &data, select_collect);
1497
1498                if (!list_empty(&data.dispose)) {
1499                        shrink_dentry_list(&data.dispose);
1500                        continue;
1501                }
1502
1503                cond_resched();
1504                if (!data.found)
1505                        break;
1506        }
1507}
1508EXPORT_SYMBOL(shrink_dcache_parent);
1509
1510static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1511{
1512        /* it has busy descendents; complain about those instead */
1513        if (!list_empty(&dentry->d_subdirs))
1514                return D_WALK_CONTINUE;
1515
1516        /* root with refcount 1 is fine */
1517        if (dentry == _data && dentry->d_lockref.count == 1)
1518                return D_WALK_CONTINUE;
1519
1520        printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1521                        " still in use (%d) [unmount of %s %s]\n",
1522                       dentry,
1523                       dentry->d_inode ?
1524                       dentry->d_inode->i_ino : 0UL,
1525                       dentry,
1526                       dentry->d_lockref.count,
1527                       dentry->d_sb->s_type->name,
1528                       dentry->d_sb->s_id);
1529        WARN_ON(1);
1530        return D_WALK_CONTINUE;
1531}
1532
1533static void do_one_tree(struct dentry *dentry)
1534{
1535        shrink_dcache_parent(dentry);
1536        d_walk(dentry, dentry, umount_check);
1537        d_drop(dentry);
1538        dput(dentry);
1539}
1540
1541/*
1542 * destroy the dentries attached to a superblock on unmounting
1543 */
1544void shrink_dcache_for_umount(struct super_block *sb)
1545{
1546        struct dentry *dentry;
1547
1548        WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1549
1550        dentry = sb->s_root;
1551        sb->s_root = NULL;
1552        do_one_tree(dentry);
1553
1554        while (!hlist_bl_empty(&sb->s_roots)) {
1555                dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1556                do_one_tree(dentry);
1557        }
1558}
1559
1560static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1561{
1562        struct dentry **victim = _data;
1563        if (d_mountpoint(dentry)) {
1564                __dget_dlock(dentry);
1565                *victim = dentry;
1566                return D_WALK_QUIT;
1567        }
1568        return D_WALK_CONTINUE;
1569}
1570
1571/**
1572 * d_invalidate - detach submounts, prune dcache, and drop
1573 * @dentry: dentry to invalidate (aka detach, prune and drop)
1574 */
1575void d_invalidate(struct dentry *dentry)
1576{
1577        bool had_submounts = false;
1578        spin_lock(&dentry->d_lock);
1579        if (d_unhashed(dentry)) {
1580                spin_unlock(&dentry->d_lock);
1581                return;
1582        }
1583        __d_drop(dentry);
1584        spin_unlock(&dentry->d_lock);
1585
1586        /* Negative dentries can be dropped without further checks */
1587        if (!dentry->d_inode)
1588                return;
1589
1590        shrink_dcache_parent(dentry);
1591        for (;;) {
1592                struct dentry *victim = NULL;
1593                d_walk(dentry, &victim, find_submount);
1594                if (!victim) {
1595                        if (had_submounts)
1596                                shrink_dcache_parent(dentry);
1597                        return;
1598                }
1599                had_submounts = true;
1600                detach_mounts(victim);
1601                dput(victim);
1602        }
1603}
1604EXPORT_SYMBOL(d_invalidate);
1605
1606/**
1607 * __d_alloc    -       allocate a dcache entry
1608 * @sb: filesystem it will belong to
1609 * @name: qstr of the name
1610 *
1611 * Allocates a dentry. It returns %NULL if there is insufficient memory
1612 * available. On a success the dentry is returned. The name passed in is
1613 * copied and the copy passed in may be reused after this call.
1614 */
1615 
1616struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1617{
1618        struct dentry *dentry;
1619        char *dname;
1620        int err;
1621
1622        dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1623        if (!dentry)
1624                return NULL;
1625
1626        /*
1627         * We guarantee that the inline name is always NUL-terminated.
1628         * This way the memcpy() done by the name switching in rename
1629         * will still always have a NUL at the end, even if we might
1630         * be overwriting an internal NUL character
1631         */
1632        dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1633        if (unlikely(!name)) {
1634                name = &slash_name;
1635                dname = dentry->d_iname;
1636        } else if (name->len > DNAME_INLINE_LEN-1) {
1637                size_t size = offsetof(struct external_name, name[1]);
1638                struct external_name *p = kmalloc(size + name->len,
1639                                                  GFP_KERNEL_ACCOUNT |
1640                                                  __GFP_RECLAIMABLE);
1641                if (!p) {
1642                        kmem_cache_free(dentry_cache, dentry); 
1643                        return NULL;
1644                }
1645                atomic_set(&p->u.count, 1);
1646                dname = p->name;
1647        } else  {
1648                dname = dentry->d_iname;
1649        }       
1650
1651        dentry->d_name.len = name->len;
1652        dentry->d_name.hash = name->hash;
1653        memcpy(dname, name->name, name->len);
1654        dname[name->len] = 0;
1655
1656        /* Make sure we always see the terminating NUL character */
1657        smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1658
1659        dentry->d_lockref.count = 1;
1660        dentry->d_flags = 0;
1661        spin_lock_init(&dentry->d_lock);
1662        seqcount_init(&dentry->d_seq);
1663        dentry->d_inode = NULL;
1664        dentry->d_parent = dentry;
1665        dentry->d_sb = sb;
1666        dentry->d_op = NULL;
1667        dentry->d_fsdata = NULL;
1668        INIT_HLIST_BL_NODE(&dentry->d_hash);
1669        INIT_LIST_HEAD(&dentry->d_lru);
1670        INIT_LIST_HEAD(&dentry->d_subdirs);
1671        INIT_HLIST_NODE(&dentry->d_u.d_alias);
1672        INIT_LIST_HEAD(&dentry->d_child);
1673        d_set_d_op(dentry, dentry->d_sb->s_d_op);
1674
1675        if (dentry->d_op && dentry->d_op->d_init) {
1676                err = dentry->d_op->d_init(dentry);
1677                if (err) {
1678                        if (dname_external(dentry))
1679                                kfree(external_name(dentry));
1680                        kmem_cache_free(dentry_cache, dentry);
1681                        return NULL;
1682                }
1683        }
1684
1685        this_cpu_inc(nr_dentry);
1686
1687        return dentry;
1688}
1689
1690/**
1691 * d_alloc      -       allocate a dcache entry
1692 * @parent: parent of entry to allocate
1693 * @name: qstr of the name
1694 *
1695 * Allocates a dentry. It returns %NULL if there is insufficient memory
1696 * available. On a success the dentry is returned. The name passed in is
1697 * copied and the copy passed in may be reused after this call.
1698 */
1699struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1700{
1701        struct dentry *dentry = __d_alloc(parent->d_sb, name);
1702        if (!dentry)
1703                return NULL;
1704        spin_lock(&parent->d_lock);
1705        /*
1706         * don't need child lock because it is not subject
1707         * to concurrency here
1708         */
1709        __dget_dlock(parent);
1710        dentry->d_parent = parent;
1711        list_add(&dentry->d_child, &parent->d_subdirs);
1712        spin_unlock(&parent->d_lock);
1713
1714        return dentry;
1715}
1716EXPORT_SYMBOL(d_alloc);
1717
1718struct dentry *d_alloc_anon(struct super_block *sb)
1719{
1720        return __d_alloc(sb, NULL);
1721}
1722EXPORT_SYMBOL(d_alloc_anon);
1723
1724struct dentry *d_alloc_cursor(struct dentry * parent)
1725{
1726        struct dentry *dentry = d_alloc_anon(parent->d_sb);
1727        if (dentry) {
1728                dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1729                dentry->d_parent = dget(parent);
1730        }
1731        return dentry;
1732}
1733
1734/**
1735 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1736 * @sb: the superblock
1737 * @name: qstr of the name
1738 *
1739 * For a filesystem that just pins its dentries in memory and never
1740 * performs lookups at all, return an unhashed IS_ROOT dentry.
1741 * This is used for pipes, sockets et.al. - the stuff that should
1742 * never be anyone's children or parents.  Unlike all other
1743 * dentries, these will not have RCU delay between dropping the
1744 * last reference and freeing them.
1745 *
1746 * The only user is alloc_file_pseudo() and that's what should
1747 * be considered a public interface.  Don't use directly.
1748 */
1749struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1750{
1751        struct dentry *dentry = __d_alloc(sb, name);
1752        if (likely(dentry))
1753                dentry->d_flags |= DCACHE_NORCU;
1754        return dentry;
1755}
1756
1757struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1758{
1759        struct qstr q;
1760
1761        q.name = name;
1762        q.hash_len = hashlen_string(parent, name);
1763        return d_alloc(parent, &q);
1764}
1765EXPORT_SYMBOL(d_alloc_name);
1766
1767void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1768{
1769        WARN_ON_ONCE(dentry->d_op);
1770        WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH  |
1771                                DCACHE_OP_COMPARE       |
1772                                DCACHE_OP_REVALIDATE    |
1773                                DCACHE_OP_WEAK_REVALIDATE       |
1774                                DCACHE_OP_DELETE        |
1775                                DCACHE_OP_REAL));
1776        dentry->d_op = op;
1777        if (!op)
1778                return;
1779        if (op->d_hash)
1780                dentry->d_flags |= DCACHE_OP_HASH;
1781        if (op->d_compare)
1782                dentry->d_flags |= DCACHE_OP_COMPARE;
1783        if (op->d_revalidate)
1784                dentry->d_flags |= DCACHE_OP_REVALIDATE;
1785        if (op->d_weak_revalidate)
1786                dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1787        if (op->d_delete)
1788                dentry->d_flags |= DCACHE_OP_DELETE;
1789        if (op->d_prune)
1790                dentry->d_flags |= DCACHE_OP_PRUNE;
1791        if (op->d_real)
1792                dentry->d_flags |= DCACHE_OP_REAL;
1793
1794}
1795EXPORT_SYMBOL(d_set_d_op);
1796
1797
1798/*
1799 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1800 * @dentry - The dentry to mark
1801 *
1802 * Mark a dentry as falling through to the lower layer (as set with
1803 * d_pin_lower()).  This flag may be recorded on the medium.
1804 */
1805void d_set_fallthru(struct dentry *dentry)
1806{
1807        spin_lock(&dentry->d_lock);
1808        dentry->d_flags |= DCACHE_FALLTHRU;
1809        spin_unlock(&dentry->d_lock);
1810}
1811EXPORT_SYMBOL(d_set_fallthru);
1812
1813static unsigned d_flags_for_inode(struct inode *inode)
1814{
1815        unsigned add_flags = DCACHE_REGULAR_TYPE;
1816
1817        if (!inode)
1818                return DCACHE_MISS_TYPE;
1819
1820        if (S_ISDIR(inode->i_mode)) {
1821                add_flags = DCACHE_DIRECTORY_TYPE;
1822                if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1823                        if (unlikely(!inode->i_op->lookup))
1824                                add_flags = DCACHE_AUTODIR_TYPE;
1825                        else
1826                                inode->i_opflags |= IOP_LOOKUP;
1827                }
1828                goto type_determined;
1829        }
1830
1831        if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1832                if (unlikely(inode->i_op->get_link)) {
1833                        add_flags = DCACHE_SYMLINK_TYPE;
1834                        goto type_determined;
1835                }
1836                inode->i_opflags |= IOP_NOFOLLOW;
1837        }
1838
1839        if (unlikely(!S_ISREG(inode->i_mode)))
1840                add_flags = DCACHE_SPECIAL_TYPE;
1841
1842type_determined:
1843        if (unlikely(IS_AUTOMOUNT(inode)))
1844                add_flags |= DCACHE_NEED_AUTOMOUNT;
1845        return add_flags;
1846}
1847
1848static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1849{
1850        unsigned add_flags = d_flags_for_inode(inode);
1851        WARN_ON(d_in_lookup(dentry));
1852
1853        spin_lock(&dentry->d_lock);
1854        /*
1855         * Decrement negative dentry count if it was in the LRU list.
1856         */
1857        if (dentry->d_flags & DCACHE_LRU_LIST)
1858                this_cpu_dec(nr_dentry_negative);
1859        hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1860        raw_write_seqcount_begin(&dentry->d_seq);
1861        __d_set_inode_and_type(dentry, inode, add_flags);
1862        raw_write_seqcount_end(&dentry->d_seq);
1863        fsnotify_update_flags(dentry);
1864        spin_unlock(&dentry->d_lock);
1865}
1866
1867/**
1868 * d_instantiate - fill in inode information for a dentry
1869 * @entry: dentry to complete
1870 * @inode: inode to attach to this dentry
1871 *
1872 * Fill in inode information in the entry.
1873 *
1874 * This turns negative dentries into productive full members
1875 * of society.
1876 *
1877 * NOTE! This assumes that the inode count has been incremented
1878 * (or otherwise set) by the caller to indicate that it is now
1879 * in use by the dcache.
1880 */
1881 
1882void d_instantiate(struct dentry *entry, struct inode * inode)
1883{
1884        BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1885        if (inode) {
1886                security_d_instantiate(entry, inode);
1887                spin_lock(&inode->i_lock);
1888                __d_instantiate(entry, inode);
1889                spin_unlock(&inode->i_lock);
1890        }
1891}
1892EXPORT_SYMBOL(d_instantiate);
1893
1894/*
1895 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1896 * with lockdep-related part of unlock_new_inode() done before
1897 * anything else.  Use that instead of open-coding d_instantiate()/
1898 * unlock_new_inode() combinations.
1899 */
1900void d_instantiate_new(struct dentry *entry, struct inode *inode)
1901{
1902        BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1903        BUG_ON(!inode);
1904        lockdep_annotate_inode_mutex_key(inode);
1905        security_d_instantiate(entry, inode);
1906        spin_lock(&inode->i_lock);
1907        __d_instantiate(entry, inode);
1908        WARN_ON(!(inode->i_state & I_NEW));
1909        inode->i_state &= ~I_NEW & ~I_CREATING;
1910        smp_mb();
1911        wake_up_bit(&inode->i_state, __I_NEW);
1912        spin_unlock(&inode->i_lock);
1913}
1914EXPORT_SYMBOL(d_instantiate_new);
1915
1916struct dentry *d_make_root(struct inode *root_inode)
1917{
1918        struct dentry *res = NULL;
1919
1920        if (root_inode) {
1921                res = d_alloc_anon(root_inode->i_sb);
1922                if (res)
1923                        d_instantiate(res, root_inode);
1924                else
1925                        iput(root_inode);
1926        }
1927        return res;
1928}
1929EXPORT_SYMBOL(d_make_root);
1930
1931static struct dentry *__d_instantiate_anon(struct dentry *dentry,
1932                                           struct inode *inode,
1933                                           bool disconnected)
1934{
1935        struct dentry *res;
1936        unsigned add_flags;
1937
1938        security_d_instantiate(dentry, inode);
1939        spin_lock(&inode->i_lock);
1940        res = __d_find_any_alias(inode);
1941        if (res) {
1942                spin_unlock(&inode->i_lock);
1943                dput(dentry);
1944                goto out_iput;
1945        }
1946
1947        /* attach a disconnected dentry */
1948        add_flags = d_flags_for_inode(inode);
1949
1950        if (disconnected)
1951                add_flags |= DCACHE_DISCONNECTED;
1952
1953        spin_lock(&dentry->d_lock);
1954        __d_set_inode_and_type(dentry, inode, add_flags);
1955        hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1956        if (!disconnected) {
1957                hlist_bl_lock(&dentry->d_sb->s_roots);
1958                hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
1959                hlist_bl_unlock(&dentry->d_sb->s_roots);
1960        }
1961        spin_unlock(&dentry->d_lock);
1962        spin_unlock(&inode->i_lock);
1963
1964        return dentry;
1965
1966 out_iput:
1967        iput(inode);
1968        return res;
1969}
1970
1971struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
1972{
1973        return __d_instantiate_anon(dentry, inode, true);
1974}
1975EXPORT_SYMBOL(d_instantiate_anon);
1976
1977static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
1978{
1979        struct dentry *tmp;
1980        struct dentry *res;
1981
1982        if (!inode)
1983                return ERR_PTR(-ESTALE);
1984        if (IS_ERR(inode))
1985                return ERR_CAST(inode);
1986
1987        res = d_find_any_alias(inode);
1988        if (res)
1989                goto out_iput;
1990
1991        tmp = d_alloc_anon(inode->i_sb);
1992        if (!tmp) {
1993                res = ERR_PTR(-ENOMEM);
1994                goto out_iput;
1995        }
1996
1997        return __d_instantiate_anon(tmp, inode, disconnected);
1998
1999out_iput:
2000        iput(inode);
2001        return res;
2002}
2003
2004/**
2005 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2006 * @inode: inode to allocate the dentry for
2007 *
2008 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2009 * similar open by handle operations.  The returned dentry may be anonymous,
2010 * or may have a full name (if the inode was already in the cache).
2011 *
2012 * When called on a directory inode, we must ensure that the inode only ever
2013 * has one dentry.  If a dentry is found, that is returned instead of
2014 * allocating a new one.
2015 *
2016 * On successful return, the reference to the inode has been transferred
2017 * to the dentry.  In case of an error the reference on the inode is released.
2018 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2019 * be passed in and the error will be propagated to the return value,
2020 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2021 */
2022struct dentry *d_obtain_alias(struct inode *inode)
2023{
2024        return __d_obtain_alias(inode, true);
2025}
2026EXPORT_SYMBOL(d_obtain_alias);
2027
2028/**
2029 * d_obtain_root - find or allocate a dentry for a given inode
2030 * @inode: inode to allocate the dentry for
2031 *
2032 * Obtain an IS_ROOT dentry for the root of a filesystem.
2033 *
2034 * We must ensure that directory inodes only ever have one dentry.  If a
2035 * dentry is found, that is returned instead of allocating a new one.
2036 *
2037 * On successful return, the reference to the inode has been transferred
2038 * to the dentry.  In case of an error the reference on the inode is
2039 * released.  A %NULL or IS_ERR inode may be passed in and will be the
2040 * error will be propagate to the return value, with a %NULL @inode
2041 * replaced by ERR_PTR(-ESTALE).
2042 */
2043struct dentry *d_obtain_root(struct inode *inode)
2044{
2045        return __d_obtain_alias(inode, false);
2046}
2047EXPORT_SYMBOL(d_obtain_root);
2048
2049/**
2050 * d_add_ci - lookup or allocate new dentry with case-exact name
2051 * @inode:  the inode case-insensitive lookup has found
2052 * @dentry: the negative dentry that was passed to the parent's lookup func
2053 * @name:   the case-exact name to be associated with the returned dentry
2054 *
2055 * This is to avoid filling the dcache with case-insensitive names to the
2056 * same inode, only the actual correct case is stored in the dcache for
2057 * case-insensitive filesystems.
2058 *
2059 * For a case-insensitive lookup match and if the the case-exact dentry
2060 * already exists in in the dcache, use it and return it.
2061 *
2062 * If no entry exists with the exact case name, allocate new dentry with
2063 * the exact case, and return the spliced entry.
2064 */
2065struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2066                        struct qstr *name)
2067{
2068        struct dentry *found, *res;
2069
2070        /*
2071         * First check if a dentry matching the name already exists,
2072         * if not go ahead and create it now.
2073         */
2074        found = d_hash_and_lookup(dentry->d_parent, name);
2075        if (found) {
2076                iput(inode);
2077                return found;
2078        }
2079        if (d_in_lookup(dentry)) {
2080                found = d_alloc_parallel(dentry->d_parent, name,
2081                                        dentry->d_wait);
2082                if (IS_ERR(found) || !d_in_lookup(found)) {
2083                        iput(inode);
2084                        return found;
2085                }
2086        } else {
2087                found = d_alloc(dentry->d_parent, name);
2088                if (!found) {
2089                        iput(inode);
2090                        return ERR_PTR(-ENOMEM);
2091                } 
2092        }
2093        res = d_splice_alias(inode, found);
2094        if (res) {
2095                dput(found);
2096                return res;
2097        }
2098        return found;
2099}
2100EXPORT_SYMBOL(d_add_ci);
2101
2102
2103static inline bool d_same_name(const struct dentry *dentry,
2104                                const struct dentry *parent,
2105                                const struct qstr *name)
2106{
2107        if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2108                if (dentry->d_name.len != name->len)
2109                        return false;
2110                return dentry_cmp(dentry, name->name, name->len) == 0;
2111        }
2112        return parent->d_op->d_compare(dentry,
2113                                       dentry->d_name.len, dentry->d_name.name,
2114                                       name) == 0;
2115}
2116
2117/**
2118 * __d_lookup_rcu - search for a dentry (racy, store-free)
2119 * @parent: parent dentry
2120 * @name: qstr of name we wish to find
2121 * @seqp: returns d_seq value at the point where the dentry was found
2122 * Returns: dentry, or NULL
2123 *
2124 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2125 * resolution (store-free path walking) design described in
2126 * Documentation/filesystems/path-lookup.txt.
2127 *
2128 * This is not to be used outside core vfs.
2129 *
2130 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2131 * held, and rcu_read_lock held. The returned dentry must not be stored into
2132 * without taking d_lock and checking d_seq sequence count against @seq
2133 * returned here.
2134 *
2135 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2136 * function.
2137 *
2138 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2139 * the returned dentry, so long as its parent's seqlock is checked after the
2140 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2141 * is formed, giving integrity down the path walk.
2142 *
2143 * NOTE! The caller *has* to check the resulting dentry against the sequence
2144 * number we've returned before using any of the resulting dentry state!
2145 */
2146struct dentry *__d_lookup_rcu(const struct dentry *parent,
2147                                const struct qstr *name,
2148                                unsigned *seqp)
2149{
2150        u64 hashlen = name->hash_len;
2151        const unsigned char *str = name->name;
2152        struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2153        struct hlist_bl_node *node;
2154        struct dentry *dentry;
2155
2156        /*
2157         * Note: There is significant duplication with __d_lookup_rcu which is
2158         * required to prevent single threaded performance regressions
2159         * especially on architectures where smp_rmb (in seqcounts) are costly.
2160         * Keep the two functions in sync.
2161         */
2162
2163        /*
2164         * The hash list is protected using RCU.
2165         *
2166         * Carefully use d_seq when comparing a candidate dentry, to avoid
2167         * races with d_move().
2168         *
2169         * It is possible that concurrent renames can mess up our list
2170         * walk here and result in missing our dentry, resulting in the
2171         * false-negative result. d_lookup() protects against concurrent
2172         * renames using rename_lock seqlock.
2173         *
2174         * See Documentation/filesystems/path-lookup.txt for more details.
2175         */
2176        hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2177                unsigned seq;
2178
2179seqretry:
2180                /*
2181                 * The dentry sequence count protects us from concurrent
2182                 * renames, and thus protects parent and name fields.
2183                 *
2184                 * The caller must perform a seqcount check in order
2185                 * to do anything useful with the returned dentry.
2186                 *
2187                 * NOTE! We do a "raw" seqcount_begin here. That means that
2188                 * we don't wait for the sequence count to stabilize if it
2189                 * is in the middle of a sequence change. If we do the slow
2190                 * dentry compare, we will do seqretries until it is stable,
2191                 * and if we end up with a successful lookup, we actually
2192                 * want to exit RCU lookup anyway.
2193                 *
2194                 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2195                 * we are still guaranteed NUL-termination of ->d_name.name.
2196                 */
2197                seq = raw_seqcount_begin(&dentry->d_seq);
2198                if (dentry->d_parent != parent)
2199                        continue;
2200                if (d_unhashed(dentry))
2201                        continue;
2202
2203                if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2204                        int tlen;
2205                        const char *tname;
2206                        if (dentry->d_name.hash != hashlen_hash(hashlen))
2207                                continue;
2208                        tlen = dentry->d_name.len;
2209                        tname = dentry->d_name.name;
2210                        /* we want a consistent (name,len) pair */
2211                        if (read_seqcount_retry(&dentry->d_seq, seq)) {
2212                                cpu_relax();
2213                                goto seqretry;
2214                        }
2215                        if (parent->d_op->d_compare(dentry,
2216                                                    tlen, tname, name) != 0)
2217                                continue;
2218                } else {
2219                        if (dentry->d_name.hash_len != hashlen)
2220                                continue;
2221                        if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2222                                continue;
2223                }
2224                *seqp = seq;
2225                return dentry;
2226        }
2227        return NULL;
2228}
2229
2230/**
2231 * d_lookup - search for a dentry
2232 * @parent: parent dentry
2233 * @name: qstr of name we wish to find
2234 * Returns: dentry, or NULL
2235 *
2236 * d_lookup searches the children of the parent dentry for the name in
2237 * question. If the dentry is found its reference count is incremented and the
2238 * dentry is returned. The caller must use dput to free the entry when it has
2239 * finished using it. %NULL is returned if the dentry does not exist.
2240 */
2241struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2242{
2243        struct dentry *dentry;
2244        unsigned seq;
2245
2246        do {
2247                seq = read_seqbegin(&rename_lock);
2248                dentry = __d_lookup(parent, name);
2249                if (dentry)
2250                        break;
2251        } while (read_seqretry(&rename_lock, seq));
2252        return dentry;
2253}
2254EXPORT_SYMBOL(d_lookup);
2255
2256/**
2257 * __d_lookup - search for a dentry (racy)
2258 * @parent: parent dentry
2259 * @name: qstr of name we wish to find
2260 * Returns: dentry, or NULL
2261 *
2262 * __d_lookup is like d_lookup, however it may (rarely) return a
2263 * false-negative result due to unrelated rename activity.
2264 *
2265 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2266 * however it must be used carefully, eg. with a following d_lookup in
2267 * the case of failure.
2268 *
2269 * __d_lookup callers must be commented.
2270 */
2271struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2272{
2273        unsigned int hash = name->hash;
2274        struct hlist_bl_head *b = d_hash(hash);
2275        struct hlist_bl_node *node;
2276        struct dentry *found = NULL;
2277        struct dentry *dentry;
2278
2279        /*
2280         * Note: There is significant duplication with __d_lookup_rcu which is
2281         * required to prevent single threaded performance regressions
2282         * especially on architectures where smp_rmb (in seqcounts) are costly.
2283         * Keep the two functions in sync.
2284         */
2285
2286        /*
2287         * The hash list is protected using RCU.
2288         *
2289         * Take d_lock when comparing a candidate dentry, to avoid races
2290         * with d_move().
2291         *
2292         * It is possible that concurrent renames can mess up our list
2293         * walk here and result in missing our dentry, resulting in the
2294         * false-negative result. d_lookup() protects against concurrent
2295         * renames using rename_lock seqlock.
2296         *
2297         * See Documentation/filesystems/path-lookup.txt for more details.
2298         */
2299        rcu_read_lock();
2300        
2301        hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2302
2303                if (dentry->d_name.hash != hash)
2304                        continue;
2305
2306                spin_lock(&dentry->d_lock);
2307                if (dentry->d_parent != parent)
2308                        goto next;
2309                if (d_unhashed(dentry))
2310                        goto next;
2311
2312                if (!d_same_name(dentry, parent, name))
2313                        goto next;
2314
2315                dentry->d_lockref.count++;
2316                found = dentry;
2317                spin_unlock(&dentry->d_lock);
2318                break;
2319next:
2320                spin_unlock(&dentry->d_lock);
2321        }
2322        rcu_read_unlock();
2323
2324        return found;
2325}
2326
2327/**
2328 * d_hash_and_lookup - hash the qstr then search for a dentry
2329 * @dir: Directory to search in
2330 * @name: qstr of name we wish to find
2331 *
2332 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2333 */
2334struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2335{
2336        /*
2337         * Check for a fs-specific hash function. Note that we must
2338         * calculate the standard hash first, as the d_op->d_hash()
2339         * routine may choose to leave the hash value unchanged.
2340         */
2341        name->hash = full_name_hash(dir, name->name, name->len);
2342        if (dir->d_flags & DCACHE_OP_HASH) {
2343                int err = dir->d_op->d_hash(dir, name);
2344                if (unlikely(err < 0))
2345                        return ERR_PTR(err);
2346        }
2347        return d_lookup(dir, name);
2348}
2349EXPORT_SYMBOL(d_hash_and_lookup);
2350
2351/*
2352 * When a file is deleted, we have two options:
2353 * - turn this dentry into a negative dentry
2354 * - unhash this dentry and free it.
2355 *
2356 * Usually, we want to just turn this into
2357 * a negative dentry, but if anybody else is
2358 * currently using the dentry or the inode
2359 * we can't do that and we fall back on removing
2360 * it from the hash queues and waiting for
2361 * it to be deleted later when it has no users
2362 */
2363 
2364/**
2365 * d_delete - delete a dentry
2366 * @dentry: The dentry to delete
2367 *
2368 * Turn the dentry into a negative dentry if possible, otherwise
2369 * remove it from the hash queues so it can be deleted later
2370 */
2371 
2372void d_delete(struct dentry * dentry)
2373{
2374        struct inode *inode = dentry->d_inode;
2375        int isdir = d_is_dir(dentry);
2376
2377        spin_lock(&inode->i_lock);
2378        spin_lock(&dentry->d_lock);
2379        /*
2380         * Are we the only user?
2381         */
2382        if (dentry->d_lockref.count == 1) {
2383                dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2384                dentry_unlink_inode(dentry);
2385        } else {
2386                __d_drop(dentry);
2387                spin_unlock(&dentry->d_lock);
2388                spin_unlock(&inode->i_lock);
2389        }
2390        fsnotify_nameremove(dentry, isdir);
2391}
2392EXPORT_SYMBOL(d_delete);
2393
2394static void __d_rehash(struct dentry *entry)
2395{
2396        struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2397
2398        hlist_bl_lock(b);
2399        hlist_bl_add_head_rcu(&entry->d_hash, b);
2400        hlist_bl_unlock(b);
2401}
2402
2403/**
2404 * d_rehash     - add an entry back to the hash
2405 * @entry: dentry to add to the hash
2406 *
2407 * Adds a dentry to the hash according to its name.
2408 */
2409 
2410void d_rehash(struct dentry * entry)
2411{
2412        spin_lock(&entry->d_lock);
2413        __d_rehash(entry);
2414        spin_unlock(&entry->d_lock);
2415}
2416EXPORT_SYMBOL(d_rehash);
2417
2418static inline unsigned start_dir_add(struct inode *dir)
2419{
2420
2421        for (;;) {
2422                unsigned n = dir->i_dir_seq;
2423                if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2424                        return n;
2425                cpu_relax();
2426        }
2427}
2428
2429static inline void end_dir_add(struct inode *dir, unsigned n)
2430{
2431        smp_store_release(&dir->i_dir_seq, n + 2);
2432}
2433
2434static void d_wait_lookup(struct dentry *dentry)
2435{
2436        if (d_in_lookup(dentry)) {
2437                DECLARE_WAITQUEUE(wait, current);
2438                add_wait_queue(dentry->d_wait, &wait);
2439                do {
2440                        set_current_state(TASK_UNINTERRUPTIBLE);
2441                        spin_unlock(&dentry->d_lock);
2442                        schedule();
2443                        spin_lock(&dentry->d_lock);
2444                } while (d_in_lookup(dentry));
2445        }
2446}
2447
2448struct dentry *d_alloc_parallel(struct dentry *parent,
2449                                const struct qstr *name,
2450                                wait_queue_head_t *wq)
2451{
2452        unsigned int hash = name->hash;
2453        struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2454        struct hlist_bl_node *node;
2455        struct dentry *new = d_alloc(parent, name);
2456        struct dentry *dentry;
2457        unsigned seq, r_seq, d_seq;
2458
2459        if (unlikely(!new))
2460                return ERR_PTR(-ENOMEM);
2461
2462retry:
2463        rcu_read_lock();
2464        seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2465        r_seq = read_seqbegin(&rename_lock);
2466        dentry = __d_lookup_rcu(parent, name, &d_seq);
2467        if (unlikely(dentry)) {
2468                if (!lockref_get_not_dead(&dentry->d_lockref)) {
2469                        rcu_read_unlock();
2470                        goto retry;
2471                }
2472                if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2473                        rcu_read_unlock();
2474                        dput(dentry);
2475                        goto retry;
2476                }
2477                rcu_read_unlock();
2478                dput(new);
2479                return dentry;
2480        }
2481        if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2482                rcu_read_unlock();
2483                goto retry;
2484        }
2485
2486        if (unlikely(seq & 1)) {
2487                rcu_read_unlock();
2488                goto retry;
2489        }
2490
2491        hlist_bl_lock(b);
2492        if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2493                hlist_bl_unlock(b);
2494                rcu_read_unlock();
2495                goto retry;
2496        }
2497        /*
2498         * No changes for the parent since the beginning of d_lookup().
2499         * Since all removals from the chain happen with hlist_bl_lock(),
2500         * any potential in-lookup matches are going to stay here until
2501         * we unlock the chain.  All fields are stable in everything
2502         * we encounter.
2503         */
2504        hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2505                if (dentry->d_name.hash != hash)
2506                        continue;
2507                if (dentry->d_parent != parent)
2508                        continue;
2509                if (!d_same_name(dentry, parent, name))
2510                        continue;
2511                hlist_bl_unlock(b);
2512                /* now we can try to grab a reference */
2513                if (!lockref_get_not_dead(&dentry->d_lockref)) {
2514                        rcu_read_unlock();
2515                        goto retry;
2516                }
2517
2518                rcu_read_unlock();
2519                /*
2520                 * somebody is likely to be still doing lookup for it;
2521                 * wait for them to finish
2522                 */
2523                spin_lock(&dentry->d_lock);
2524                d_wait_lookup(dentry);
2525                /*
2526                 * it's not in-lookup anymore; in principle we should repeat
2527                 * everything from dcache lookup, but it's likely to be what
2528                 * d_lookup() would've found anyway.  If it is, just return it;
2529                 * otherwise we really have to repeat the whole thing.
2530                 */
2531                if (unlikely(dentry->d_name.hash != hash))
2532                        goto mismatch;
2533                if (unlikely(dentry->d_parent != parent))
2534                        goto mismatch;
2535                if (unlikely(d_unhashed(dentry)))
2536                        goto mismatch;
2537                if (unlikely(!d_same_name(dentry, parent, name)))
2538                        goto mismatch;
2539                /* OK, it *is* a hashed match; return it */
2540                spin_unlock(&dentry->d_lock);
2541                dput(new);
2542                return dentry;
2543        }
2544        rcu_read_unlock();
2545        /* we can't take ->d_lock here; it's OK, though. */
2546        new->d_flags |= DCACHE_PAR_LOOKUP;
2547        new->d_wait = wq;
2548        hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2549        hlist_bl_unlock(b);
2550        return new;
2551mismatch:
2552        spin_unlock(&dentry->d_lock);
2553        dput(dentry);
2554        goto retry;
2555}
2556EXPORT_SYMBOL(d_alloc_parallel);
2557
2558void __d_lookup_done(struct dentry *dentry)
2559{
2560        struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2561                                                 dentry->d_name.hash);
2562        hlist_bl_lock(b);
2563        dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2564        __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2565        wake_up_all(dentry->d_wait);
2566        dentry->d_wait = NULL;
2567        hlist_bl_unlock(b);
2568        INIT_HLIST_NODE(&dentry->d_u.d_alias);
2569        INIT_LIST_HEAD(&dentry->d_lru);
2570}
2571EXPORT_SYMBOL(__d_lookup_done);
2572
2573/* inode->i_lock held if inode is non-NULL */
2574
2575static inline void __d_add(struct dentry *dentry, struct inode *inode)
2576{
2577        struct inode *dir = NULL;
2578        unsigned n;
2579        spin_lock(&dentry->d_lock);
2580        if (unlikely(d_in_lookup(dentry))) {
2581                dir = dentry->d_parent->d_inode;
2582                n = start_dir_add(dir);
2583                __d_lookup_done(dentry);
2584        }
2585        if (inode) {
2586                unsigned add_flags = d_flags_for_inode(inode);
2587                hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2588                raw_write_seqcount_begin(&dentry->d_seq);
2589                __d_set_inode_and_type(dentry, inode, add_flags);
2590                raw_write_seqcount_end(&dentry->d_seq);
2591                fsnotify_update_flags(dentry);
2592        }
2593        __d_rehash(dentry);
2594        if (dir)
2595                end_dir_add(dir, n);
2596        spin_unlock(&dentry->d_lock);
2597        if (inode)
2598                spin_unlock(&inode->i_lock);
2599}
2600
2601/**
2602 * d_add - add dentry to hash queues
2603 * @entry: dentry to add
2604 * @inode: The inode to attach to this dentry
2605 *
2606 * This adds the entry to the hash queues and initializes @inode.
2607 * The entry was actually filled in earlier during d_alloc().
2608 */
2609
2610void d_add(struct dentry *entry, struct inode *inode)
2611{
2612        if (inode) {
2613                security_d_instantiate(entry, inode);
2614                spin_lock(&inode->i_lock);
2615        }
2616        __d_add(entry, inode);
2617}
2618EXPORT_SYMBOL(d_add);
2619
2620/**
2621 * d_exact_alias - find and hash an exact unhashed alias
2622 * @entry: dentry to add
2623 * @inode: The inode to go with this dentry
2624 *
2625 * If an unhashed dentry with the same name/parent and desired
2626 * inode already exists, hash and return it.  Otherwise, return
2627 * NULL.
2628 *
2629 * Parent directory should be locked.
2630 */
2631struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2632{
2633        struct dentry *alias;
2634        unsigned int hash = entry->d_name.hash;
2635
2636        spin_lock(&inode->i_lock);
2637        hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2638                /*
2639                 * Don't need alias->d_lock here, because aliases with
2640                 * d_parent == entry->d_parent are not subject to name or
2641                 * parent changes, because the parent inode i_mutex is held.
2642                 */
2643                if (alias->d_name.hash != hash)
2644                        continue;
2645                if (alias->d_parent != entry->d_parent)
2646                        continue;
2647                if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2648                        continue;
2649                spin_lock(&alias->d_lock);
2650                if (!d_unhashed(alias)) {
2651                        spin_unlock(&alias->d_lock);
2652                        alias = NULL;
2653                } else {
2654                        __dget_dlock(alias);
2655                        __d_rehash(alias);
2656                        spin_unlock(&alias->d_lock);
2657                }
2658                spin_unlock(&inode->i_lock);
2659                return alias;
2660        }
2661        spin_unlock(&inode->i_lock);
2662        return NULL;
2663}
2664EXPORT_SYMBOL(d_exact_alias);
2665
2666static void swap_names(struct dentry *dentry, struct dentry *target)
2667{
2668        if (unlikely(dname_external(target))) {
2669                if (unlikely(dname_external(dentry))) {
2670                        /*
2671                         * Both external: swap the pointers
2672                         */
2673                        swap(target->d_name.name, dentry->d_name.name);
2674                } else {
2675                        /*
2676                         * dentry:internal, target:external.  Steal target's
2677                         * storage and make target internal.
2678                         */
2679                        memcpy(target->d_iname, dentry->d_name.name,
2680                                        dentry->d_name.len + 1);
2681                        dentry->d_name.name = target->d_name.name;
2682                        target->d_name.name = target->d_iname;
2683                }
2684        } else {
2685                if (unlikely(dname_external(dentry))) {
2686                        /*
2687                         * dentry:external, target:internal.  Give dentry's
2688                         * storage to target and make dentry internal
2689                         */
2690                        memcpy(dentry->d_iname, target->d_name.name,
2691                                        target->d_name.len + 1);
2692                        target->d_name.name = dentry->d_name.name;
2693                        dentry->d_name.name = dentry->d_iname;
2694                } else {
2695                        /*
2696                         * Both are internal.
2697                         */
2698                        unsigned int i;
2699                        BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2700                        for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2701                                swap(((long *) &dentry->d_iname)[i],
2702                                     ((long *) &target->d_iname)[i]);
2703                        }
2704                }
2705        }
2706        swap(dentry->d_name.hash_len, target->d_name.hash_len);
2707}
2708
2709static void copy_name(struct dentry *dentry, struct dentry *target)
2710{
2711        struct external_name *old_name = NULL;
2712        if (unlikely(dname_external(dentry)))
2713                old_name = external_name(dentry);
2714        if (unlikely(dname_external(target))) {
2715                atomic_inc(&external_name(target)->u.count);
2716                dentry->d_name = target->d_name;
2717        } else {
2718                memcpy(dentry->d_iname, target->d_name.name,
2719                                target->d_name.len + 1);
2720                dentry->d_name.name = dentry->d_iname;
2721                dentry->d_name.hash_len = target->d_name.hash_len;
2722        }
2723        if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2724                kfree_rcu(old_name, u.head);
2725}
2726
2727/*
2728 * __d_move - move a dentry
2729 * @dentry: entry to move
2730 * @target: new dentry
2731 * @exchange: exchange the two dentries
2732 *
2733 * Update the dcache to reflect the move of a file name. Negative
2734 * dcache entries should not be moved in this way. Caller must hold
2735 * rename_lock, the i_mutex of the source and target directories,
2736 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2737 */
2738static void __d_move(struct dentry *dentry, struct dentry *target,
2739                     bool exchange)
2740{
2741        struct dentry *old_parent, *p;
2742        struct inode *dir = NULL;
2743        unsigned n;
2744
2745        WARN_ON(!dentry->d_inode);
2746        if (WARN_ON(dentry == target))
2747                return;
2748
2749        BUG_ON(d_ancestor(target, dentry));
2750        old_parent = dentry->d_parent;
2751        p = d_ancestor(old_parent, target);
2752        if (IS_ROOT(dentry)) {
2753                BUG_ON(p);
2754                spin_lock(&target->d_parent->d_lock);
2755        } else if (!p) {
2756                /* target is not a descendent of dentry->d_parent */
2757                spin_lock(&target->d_parent->d_lock);
2758                spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2759        } else {
2760                BUG_ON(p == dentry);
2761                spin_lock(&old_parent->d_lock);
2762                if (p != target)
2763                        spin_lock_nested(&target->d_parent->d_lock,
2764                                        DENTRY_D_LOCK_NESTED);
2765        }
2766        spin_lock_nested(&dentry->d_lock, 2);
2767        spin_lock_nested(&target->d_lock, 3);
2768
2769        if (unlikely(d_in_lookup(target))) {
2770                dir = target->d_parent->d_inode;
2771                n = start_dir_add(dir);
2772                __d_lookup_done(target);
2773        }
2774
2775        write_seqcount_begin(&dentry->d_seq);
2776        write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2777
2778        /* unhash both */
2779        if (!d_unhashed(dentry))
2780                ___d_drop(dentry);
2781        if (!d_unhashed(target))
2782                ___d_drop(target);
2783
2784        /* ... and switch them in the tree */
2785        dentry->d_parent = target->d_parent;
2786        if (!exchange) {
2787                copy_name(dentry, target);
2788                target->d_hash.pprev = NULL;
2789                dentry->d_parent->d_lockref.count++;
2790                if (dentry != old_parent) /* wasn't IS_ROOT */
2791                        WARN_ON(!--old_parent->d_lockref.count);
2792        } else {
2793                target->d_parent = old_parent;
2794                swap_names(dentry, target);
2795                list_move(&target->d_child, &target->d_parent->d_subdirs);
2796                __d_rehash(target);
2797                fsnotify_update_flags(target);
2798        }
2799        list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2800        __d_rehash(dentry);
2801        fsnotify_update_flags(dentry);
2802        fscrypt_handle_d_move(dentry);
2803
2804        write_seqcount_end(&target->d_seq);
2805        write_seqcount_end(&dentry->d_seq);
2806
2807        if (dir)
2808                end_dir_add(dir, n);
2809
2810        if (dentry->d_parent != old_parent)
2811                spin_unlock(&dentry->d_parent->d_lock);
2812        if (dentry != old_parent)
2813                spin_unlock(&old_parent->d_lock);
2814        spin_unlock(&target->d_lock);
2815        spin_unlock(&dentry->d_lock);
2816}
2817
2818/*
2819 * d_move - move a dentry
2820 * @dentry: entry to move
2821 * @target: new dentry
2822 *
2823 * Update the dcache to reflect the move of a file name. Negative
2824 * dcache entries should not be moved in this way. See the locking
2825 * requirements for __d_move.
2826 */
2827void d_move(struct dentry *dentry, struct dentry *target)
2828{
2829        write_seqlock(&rename_lock);
2830        __d_move(dentry, target, false);
2831        write_sequnlock(&rename_lock);
2832}
2833EXPORT_SYMBOL(d_move);
2834
2835/*
2836 * d_exchange - exchange two dentries
2837 * @dentry1: first dentry
2838 * @dentry2: second dentry
2839 */
2840void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2841{
2842        write_seqlock(&rename_lock);
2843
2844        WARN_ON(!dentry1->d_inode);
2845        WARN_ON(!dentry2->d_inode);
2846        WARN_ON(IS_ROOT(dentry1));
2847        WARN_ON(IS_ROOT(dentry2));
2848
2849        __d_move(dentry1, dentry2, true);
2850
2851        write_sequnlock(&rename_lock);
2852}
2853
2854/**
2855 * d_ancestor - search for an ancestor
2856 * @p1: ancestor dentry
2857 * @p2: child dentry
2858 *
2859 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2860 * an ancestor of p2, else NULL.
2861 */
2862struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2863{
2864        struct dentry *p;
2865
2866        for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2867                if (p->d_parent == p1)
2868                        return p;
2869        }
2870        return NULL;
2871}
2872
2873/*
2874 * This helper attempts to cope with remotely renamed directories
2875 *
2876 * It assumes that the caller is already holding
2877 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2878 *
2879 * Note: If ever the locking in lock_rename() changes, then please
2880 * remember to update this too...
2881 */
2882static int __d_unalias(struct inode *inode,
2883                struct dentry *dentry, struct dentry *alias)
2884{
2885        struct mutex *m1 = NULL;
2886        struct rw_semaphore *m2 = NULL;
2887        int ret = -ESTALE;
2888
2889        /* If alias and dentry share a parent, then no extra locks required */
2890        if (alias->d_parent == dentry->d_parent)
2891                goto out_unalias;
2892
2893        /* See lock_rename() */
2894        if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2895                goto out_err;
2896        m1 = &dentry->d_sb->s_vfs_rename_mutex;
2897        if (!inode_trylock_shared(alias->d_parent->d_inode))
2898                goto out_err;
2899        m2 = &alias->d_parent->d_inode->i_rwsem;
2900out_unalias:
2901        __d_move(alias, dentry, false);
2902        ret = 0;
2903out_err:
2904        if (m2)
2905                up_read(m2);
2906        if (m1)
2907                mutex_unlock(m1);
2908        return ret;
2909}
2910
2911/**
2912 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2913 * @inode:  the inode which may have a disconnected dentry
2914 * @dentry: a negative dentry which we want to point to the inode.
2915 *
2916 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2917 * place of the given dentry and return it, else simply d_add the inode
2918 * to the dentry and return NULL.
2919 *
2920 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2921 * we should error out: directories can't have multiple aliases.
2922 *
2923 * This is needed in the lookup routine of any filesystem that is exportable
2924 * (via knfsd) so that we can build dcache paths to directories effectively.
2925 *
2926 * If a dentry was found and moved, then it is returned.  Otherwise NULL
2927 * is returned.  This matches the expected return value of ->lookup.
2928 *
2929 * Cluster filesystems may call this function with a negative, hashed dentry.
2930 * In that case, we know that the inode will be a regular file, and also this
2931 * will only occur during atomic_open. So we need to check for the dentry
2932 * being already hashed only in the final case.
2933 */
2934struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2935{
2936        if (IS_ERR(inode))
2937                return ERR_CAST(inode);
2938
2939        BUG_ON(!d_unhashed(dentry));
2940
2941        if (!inode)
2942                goto out;
2943
2944        security_d_instantiate(dentry, inode);
2945        spin_lock(&inode->i_lock);
2946        if (S_ISDIR(inode->i_mode)) {
2947                struct dentry *new = __d_find_any_alias(inode);
2948                if (unlikely(new)) {
2949                        /* The reference to new ensures it remains an alias */
2950                        spin_unlock(&inode->i_lock);
2951                        write_seqlock(&rename_lock);
2952                        if (unlikely(d_ancestor(new, dentry))) {
2953                                write_sequnlock(&rename_lock);
2954                                dput(new);
2955                                new = ERR_PTR(-ELOOP);
2956                                pr_warn_ratelimited(
2957                                        "VFS: Lookup of '%s' in %s %s"
2958                                        " would have caused loop\n",
2959                                        dentry->d_name.name,
2960                                        inode->i_sb->s_type->name,
2961                                        inode->i_sb->s_id);
2962                        } else if (!IS_ROOT(new)) {
2963                                struct dentry *old_parent = dget(new->d_parent);
2964                                int err = __d_unalias(inode, dentry, new);
2965                                write_sequnlock(&rename_lock);
2966                                if (err) {
2967                                        dput(new);
2968                                        new = ERR_PTR(err);
2969                                }
2970                                dput(old_parent);
2971                        } else {
2972                                __d_move(new, dentry, false);
2973                                write_sequnlock(&rename_lock);
2974                        }
2975                        iput(inode);
2976                        return new;
2977                }
2978        }
2979out:
2980        __d_add(dentry, inode);
2981        return NULL;
2982}
2983EXPORT_SYMBOL(d_splice_alias);
2984
2985/*
2986 * Test whether new_dentry is a subdirectory of old_dentry.
2987 *
2988 * Trivially implemented using the dcache structure
2989 */
2990
2991/**
2992 * is_subdir - is new dentry a subdirectory of old_dentry
2993 * @new_dentry: new dentry
2994 * @old_dentry: old dentry
2995 *
2996 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
2997 * Returns false otherwise.
2998 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2999 */
3000  
3001bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3002{
3003        bool result;
3004        unsigned seq;
3005
3006        if (new_dentry == old_dentry)
3007                return true;
3008
3009        do {
3010                /* for restarting inner loop in case of seq retry */
3011                seq = read_seqbegin(&rename_lock);
3012                /*
3013                 * Need rcu_readlock to protect against the d_parent trashing
3014                 * due to d_move
3015                 */
3016                rcu_read_lock();
3017                if (d_ancestor(old_dentry, new_dentry))
3018                        result = true;
3019                else
3020                        result = false;
3021                rcu_read_unlock();
3022        } while (read_seqretry(&rename_lock, seq));
3023
3024        return result;
3025}
3026EXPORT_SYMBOL(is_subdir);
3027
3028static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3029{
3030        struct dentry *root = data;
3031        if (dentry != root) {
3032                if (d_unhashed(dentry) || !dentry->d_inode)
3033                        return D_WALK_SKIP;
3034
3035                if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3036                        dentry->d_flags |= DCACHE_GENOCIDE;
3037                        dentry->d_lockref.count--;
3038                }
3039        }
3040        return D_WALK_CONTINUE;
3041}
3042
3043void d_genocide(struct dentry *parent)
3044{
3045        d_walk(parent, parent, d_genocide_kill);
3046}
3047
3048EXPORT_SYMBOL(d_genocide);
3049
3050void d_tmpfile(struct dentry *dentry, struct inode *inode)
3051{
3052        inode_dec_link_count(inode);
3053        BUG_ON(dentry->d_name.name != dentry->d_iname ||
3054                !hlist_unhashed(&dentry->d_u.d_alias) ||
3055                !d_unlinked(dentry));
3056        spin_lock(&dentry->d_parent->d_lock);
3057        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3058        dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3059                                (unsigned long long)inode->i_ino);
3060        spin_unlock(&dentry->d_lock);
3061        spin_unlock(&dentry->d_parent->d_lock);
3062        d_instantiate(dentry, inode);
3063}
3064EXPORT_SYMBOL(d_tmpfile);
3065
3066static __initdata unsigned long dhash_entries;
3067static int __init set_dhash_entries(char *str)
3068{
3069        if (!str)
3070                return 0;
3071        dhash_entries = simple_strtoul(str, &str, 0);
3072        return 1;
3073}
3074__setup("dhash_entries=", set_dhash_entries);
3075
3076static void __init dcache_init_early(void)
3077{
3078        /* If hashes are distributed across NUMA nodes, defer
3079         * hash allocation until vmalloc space is available.
3080         */
3081        if (hashdist)
3082                return;
3083
3084        dentry_hashtable =
3085                alloc_large_system_hash("Dentry cache",
3086                                        sizeof(struct hlist_bl_head),
3087                                        dhash_entries,
3088                                        13,
3089                                        HASH_EARLY | HASH_ZERO,
3090                                        &d_hash_shift,
3091                                        NULL,
3092                                        0,
3093                                        0);
3094        d_hash_shift = 32 - d_hash_shift;
3095}
3096
3097static void __init dcache_init(void)
3098{
3099        /*
3100         * A constructor could be added for stable state like the lists,
3101         * but it is probably not worth it because of the cache nature
3102         * of the dcache.
3103         */
3104        dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3105                SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3106                d_iname);
3107
3108        /* Hash may have been set up in dcache_init_early */
3109        if (!hashdist)
3110                return;
3111
3112        dentry_hashtable =
3113                alloc_large_system_hash("Dentry cache",
3114                                        sizeof(struct hlist_bl_head),
3115                                        dhash_entries,
3116                                        13,
3117                                        HASH_ZERO,
3118                                        &d_hash_shift,
3119                                        NULL,
3120                                        0,
3121                                        0);
3122        d_hash_shift = 32 - d_hash_shift;
3123}
3124
3125/* SLAB cache for __getname() consumers */
3126struct kmem_cache *names_cachep __read_mostly;
3127EXPORT_SYMBOL(names_cachep);
3128
3129void __init vfs_caches_init_early(void)
3130{
3131        int i;
3132
3133        for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3134                INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3135
3136        dcache_init_early();
3137        inode_init_early();
3138}
3139
3140void __init vfs_caches_init(void)
3141{
3142        names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3143                        SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3144
3145        dcache_init();
3146        inode_init();
3147        files_init();
3148        files_maxfiles_init();
3149        mnt_init();
3150        bdev_cache_init();
3151        chrdev_init();
3152}
3153