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);
  87const struct qstr dotdot_name = QSTR_INIT("..", 2);
  88EXPORT_SYMBOL(dotdot_name);
  89
  90/*
  91 * This is the single most critical data structure when it comes
  92 * to the dcache: the hashtable for lookups. Somebody should try
  93 * to make this good - I've just made it work.
  94 *
  95 * This hash-function tries to avoid losing too many bits of hash
  96 * information, yet avoid using a prime hash-size or similar.
  97 */
  98
  99static unsigned int d_hash_shift __read_mostly;
 100
 101static struct hlist_bl_head *dentry_hashtable __read_mostly;
 102
 103static inline struct hlist_bl_head *d_hash(unsigned int hash)
 104{
 105        return dentry_hashtable + (hash >> d_hash_shift);
 106}
 107
 108#define IN_LOOKUP_SHIFT 10
 109static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
 110
 111static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
 112                                        unsigned int hash)
 113{
 114        hash += (unsigned long) parent / L1_CACHE_BYTES;
 115        return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
 116}
 117
 118
 119/* Statistics gathering. */
 120struct dentry_stat_t dentry_stat = {
 121        .age_limit = 45,
 122};
 123
 124static DEFINE_PER_CPU(long, nr_dentry);
 125static DEFINE_PER_CPU(long, nr_dentry_unused);
 126static DEFINE_PER_CPU(long, nr_dentry_negative);
 127
 128#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
 129
 130/*
 131 * Here we resort to our own counters instead of using generic per-cpu counters
 132 * for consistency with what the vfs inode code does. We are expected to harvest
 133 * better code and performance by having our own specialized counters.
 134 *
 135 * Please note that the loop is done over all possible CPUs, not over all online
 136 * CPUs. The reason for this is that we don't want to play games with CPUs going
 137 * on and off. If one of them goes off, we will just keep their counters.
 138 *
 139 * glommer: See cffbc8a for details, and if you ever intend to change this,
 140 * please update all vfs counters to match.
 141 */
 142static long get_nr_dentry(void)
 143{
 144        int i;
 145        long sum = 0;
 146        for_each_possible_cpu(i)
 147                sum += per_cpu(nr_dentry, i);
 148        return sum < 0 ? 0 : sum;
 149}
 150
 151static long get_nr_dentry_unused(void)
 152{
 153        int i;
 154        long sum = 0;
 155        for_each_possible_cpu(i)
 156                sum += per_cpu(nr_dentry_unused, i);
 157        return sum < 0 ? 0 : sum;
 158}
 159
 160static long get_nr_dentry_negative(void)
 161{
 162        int i;
 163        long sum = 0;
 164
 165        for_each_possible_cpu(i)
 166                sum += per_cpu(nr_dentry_negative, i);
 167        return sum < 0 ? 0 : sum;
 168}
 169
 170int proc_nr_dentry(struct ctl_table *table, int write, void *buffer,
 171                   size_t *lenp, loff_t *ppos)
 172{
 173        dentry_stat.nr_dentry = get_nr_dentry();
 174        dentry_stat.nr_unused = get_nr_dentry_unused();
 175        dentry_stat.nr_negative = get_nr_dentry_negative();
 176        return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 177}
 178#endif
 179
 180/*
 181 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
 182 * The strings are both count bytes long, and count is non-zero.
 183 */
 184#ifdef CONFIG_DCACHE_WORD_ACCESS
 185
 186#include <asm/word-at-a-time.h>
 187/*
 188 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
 189 * aligned allocation for this particular component. We don't
 190 * strictly need the load_unaligned_zeropad() safety, but it
 191 * doesn't hurt either.
 192 *
 193 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
 194 * need the careful unaligned handling.
 195 */
 196static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
 197{
 198        unsigned long a,b,mask;
 199
 200        for (;;) {
 201                a = read_word_at_a_time(cs);
 202                b = load_unaligned_zeropad(ct);
 203                if (tcount < sizeof(unsigned long))
 204                        break;
 205                if (unlikely(a != b))
 206                        return 1;
 207                cs += sizeof(unsigned long);
 208                ct += sizeof(unsigned long);
 209                tcount -= sizeof(unsigned long);
 210                if (!tcount)
 211                        return 0;
 212        }
 213        mask = bytemask_from_count(tcount);
 214        return unlikely(!!((a ^ b) & mask));
 215}
 216
 217#else
 218
 219static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
 220{
 221        do {
 222                if (*cs != *ct)
 223                        return 1;
 224                cs++;
 225                ct++;
 226                tcount--;
 227        } while (tcount);
 228        return 0;
 229}
 230
 231#endif
 232
 233static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
 234{
 235        /*
 236         * Be careful about RCU walk racing with rename:
 237         * use 'READ_ONCE' to fetch the name pointer.
 238         *
 239         * NOTE! Even if a rename will mean that the length
 240         * was not loaded atomically, we don't care. The
 241         * RCU walk will check the sequence count eventually,
 242         * and catch it. And we won't overrun the buffer,
 243         * because we're reading the name pointer atomically,
 244         * and a dentry name is guaranteed to be properly
 245         * terminated with a NUL byte.
 246         *
 247         * End result: even if 'len' is wrong, we'll exit
 248         * early because the data cannot match (there can
 249         * be no NUL in the ct/tcount data)
 250         */
 251        const unsigned char *cs = READ_ONCE(dentry->d_name.name);
 252
 253        return dentry_string_cmp(cs, ct, tcount);
 254}
 255
 256struct external_name {
 257        union {
 258                atomic_t count;
 259                struct rcu_head head;
 260        } u;
 261        unsigned char name[];
 262};
 263
 264static inline struct external_name *external_name(struct dentry *dentry)
 265{
 266        return container_of(dentry->d_name.name, struct external_name, name[0]);
 267}
 268
 269static void __d_free(struct rcu_head *head)
 270{
 271        struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
 272
 273        kmem_cache_free(dentry_cache, dentry); 
 274}
 275
 276static void __d_free_external(struct rcu_head *head)
 277{
 278        struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
 279        kfree(external_name(dentry));
 280        kmem_cache_free(dentry_cache, dentry);
 281}
 282
 283static inline int dname_external(const struct dentry *dentry)
 284{
 285        return dentry->d_name.name != dentry->d_iname;
 286}
 287
 288void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
 289{
 290        spin_lock(&dentry->d_lock);
 291        name->name = dentry->d_name;
 292        if (unlikely(dname_external(dentry))) {
 293                atomic_inc(&external_name(dentry)->u.count);
 294        } else {
 295                memcpy(name->inline_name, dentry->d_iname,
 296                       dentry->d_name.len + 1);
 297                name->name.name = name->inline_name;
 298        }
 299        spin_unlock(&dentry->d_lock);
 300}
 301EXPORT_SYMBOL(take_dentry_name_snapshot);
 302
 303void release_dentry_name_snapshot(struct name_snapshot *name)
 304{
 305        if (unlikely(name->name.name != name->inline_name)) {
 306                struct external_name *p;
 307                p = container_of(name->name.name, struct external_name, name[0]);
 308                if (unlikely(atomic_dec_and_test(&p->u.count)))
 309                        kfree_rcu(p, u.head);
 310        }
 311}
 312EXPORT_SYMBOL(release_dentry_name_snapshot);
 313
 314static inline void __d_set_inode_and_type(struct dentry *dentry,
 315                                          struct inode *inode,
 316                                          unsigned type_flags)
 317{
 318        unsigned flags;
 319
 320        dentry->d_inode = inode;
 321        flags = READ_ONCE(dentry->d_flags);
 322        flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
 323        flags |= type_flags;
 324        smp_store_release(&dentry->d_flags, flags);
 325}
 326
 327static inline void __d_clear_type_and_inode(struct dentry *dentry)
 328{
 329        unsigned flags = READ_ONCE(dentry->d_flags);
 330
 331        flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
 332        WRITE_ONCE(dentry->d_flags, flags);
 333        dentry->d_inode = NULL;
 334        if (dentry->d_flags & DCACHE_LRU_LIST)
 335                this_cpu_inc(nr_dentry_negative);
 336}
 337
 338static void dentry_free(struct dentry *dentry)
 339{
 340        WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
 341        if (unlikely(dname_external(dentry))) {
 342                struct external_name *p = external_name(dentry);
 343                if (likely(atomic_dec_and_test(&p->u.count))) {
 344                        call_rcu(&dentry->d_u.d_rcu, __d_free_external);
 345                        return;
 346                }
 347        }
 348        /* if dentry was never visible to RCU, immediate free is OK */
 349        if (dentry->d_flags & DCACHE_NORCU)
 350                __d_free(&dentry->d_u.d_rcu);
 351        else
 352                call_rcu(&dentry->d_u.d_rcu, __d_free);
 353}
 354
 355/*
 356 * Release the dentry's inode, using the filesystem
 357 * d_iput() operation if defined.
 358 */
 359static void dentry_unlink_inode(struct dentry * dentry)
 360        __releases(dentry->d_lock)
 361        __releases(dentry->d_inode->i_lock)
 362{
 363        struct inode *inode = dentry->d_inode;
 364
 365        raw_write_seqcount_begin(&dentry->d_seq);
 366        __d_clear_type_and_inode(dentry);
 367        hlist_del_init(&dentry->d_u.d_alias);
 368        raw_write_seqcount_end(&dentry->d_seq);
 369        spin_unlock(&dentry->d_lock);
 370        spin_unlock(&inode->i_lock);
 371        if (!inode->i_nlink)
 372                fsnotify_inoderemove(inode);
 373        if (dentry->d_op && dentry->d_op->d_iput)
 374                dentry->d_op->d_iput(dentry, inode);
 375        else
 376                iput(inode);
 377}
 378
 379/*
 380 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
 381 * is in use - which includes both the "real" per-superblock
 382 * LRU list _and_ the DCACHE_SHRINK_LIST use.
 383 *
 384 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
 385 * on the shrink list (ie not on the superblock LRU list).
 386 *
 387 * The per-cpu "nr_dentry_unused" counters are updated with
 388 * the DCACHE_LRU_LIST bit.
 389 *
 390 * The per-cpu "nr_dentry_negative" counters are only updated
 391 * when deleted from or added to the per-superblock LRU list, not
 392 * from/to the shrink list. That is to avoid an unneeded dec/inc
 393 * pair when moving from LRU to shrink list in select_collect().
 394 *
 395 * These helper functions make sure we always follow the
 396 * rules. d_lock must be held by the caller.
 397 */
 398#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
 399static void d_lru_add(struct dentry *dentry)
 400{
 401        D_FLAG_VERIFY(dentry, 0);
 402        dentry->d_flags |= DCACHE_LRU_LIST;
 403        this_cpu_inc(nr_dentry_unused);
 404        if (d_is_negative(dentry))
 405                this_cpu_inc(nr_dentry_negative);
 406        WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
 407}
 408
 409static void d_lru_del(struct dentry *dentry)
 410{
 411        D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
 412        dentry->d_flags &= ~DCACHE_LRU_LIST;
 413        this_cpu_dec(nr_dentry_unused);
 414        if (d_is_negative(dentry))
 415                this_cpu_dec(nr_dentry_negative);
 416        WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
 417}
 418
 419static void d_shrink_del(struct dentry *dentry)
 420{
 421        D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
 422        list_del_init(&dentry->d_lru);
 423        dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
 424        this_cpu_dec(nr_dentry_unused);
 425}
 426
 427static void d_shrink_add(struct dentry *dentry, struct list_head *list)
 428{
 429        D_FLAG_VERIFY(dentry, 0);
 430        list_add(&dentry->d_lru, list);
 431        dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
 432        this_cpu_inc(nr_dentry_unused);
 433}
 434
 435/*
 436 * These can only be called under the global LRU lock, ie during the
 437 * callback for freeing the LRU list. "isolate" removes it from the
 438 * LRU lists entirely, while shrink_move moves it to the indicated
 439 * private list.
 440 */
 441static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
 442{
 443        D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
 444        dentry->d_flags &= ~DCACHE_LRU_LIST;
 445        this_cpu_dec(nr_dentry_unused);
 446        if (d_is_negative(dentry))
 447                this_cpu_dec(nr_dentry_negative);
 448        list_lru_isolate(lru, &dentry->d_lru);
 449}
 450
 451static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
 452                              struct list_head *list)
 453{
 454        D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
 455        dentry->d_flags |= DCACHE_SHRINK_LIST;
 456        if (d_is_negative(dentry))
 457                this_cpu_dec(nr_dentry_negative);
 458        list_lru_isolate_move(lru, &dentry->d_lru, list);
 459}
 460
 461static void ___d_drop(struct dentry *dentry)
 462{
 463        struct hlist_bl_head *b;
 464        /*
 465         * Hashed dentries are normally on the dentry hashtable,
 466         * with the exception of those newly allocated by
 467         * d_obtain_root, which are always IS_ROOT:
 468         */
 469        if (unlikely(IS_ROOT(dentry)))
 470                b = &dentry->d_sb->s_roots;
 471        else
 472                b = d_hash(dentry->d_name.hash);
 473
 474        hlist_bl_lock(b);
 475        __hlist_bl_del(&dentry->d_hash);
 476        hlist_bl_unlock(b);
 477}
 478
 479void __d_drop(struct dentry *dentry)
 480{
 481        if (!d_unhashed(dentry)) {
 482                ___d_drop(dentry);
 483                dentry->d_hash.pprev = NULL;
 484                write_seqcount_invalidate(&dentry->d_seq);
 485        }
 486}
 487EXPORT_SYMBOL(__d_drop);
 488
 489/**
 490 * d_drop - drop a dentry
 491 * @dentry: dentry to drop
 492 *
 493 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
 494 * be found through a VFS lookup any more. Note that this is different from
 495 * deleting the dentry - d_delete will try to mark the dentry negative if
 496 * possible, giving a successful _negative_ lookup, while d_drop will
 497 * just make the cache lookup fail.
 498 *
 499 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
 500 * reason (NFS timeouts or autofs deletes).
 501 *
 502 * __d_drop requires dentry->d_lock
 503 *
 504 * ___d_drop doesn't mark dentry as "unhashed"
 505 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
 506 */
 507void d_drop(struct dentry *dentry)
 508{
 509        spin_lock(&dentry->d_lock);
 510        __d_drop(dentry);
 511        spin_unlock(&dentry->d_lock);
 512}
 513EXPORT_SYMBOL(d_drop);
 514
 515static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
 516{
 517        struct dentry *next;
 518        /*
 519         * Inform d_walk() and shrink_dentry_list() that we are no longer
 520         * attached to the dentry tree
 521         */
 522        dentry->d_flags |= DCACHE_DENTRY_KILLED;
 523        if (unlikely(list_empty(&dentry->d_child)))
 524                return;
 525        __list_del_entry(&dentry->d_child);
 526        /*
 527         * Cursors can move around the list of children.  While we'd been
 528         * a normal list member, it didn't matter - ->d_child.next would've
 529         * been updated.  However, from now on it won't be and for the
 530         * things like d_walk() it might end up with a nasty surprise.
 531         * Normally d_walk() doesn't care about cursors moving around -
 532         * ->d_lock on parent prevents that and since a cursor has no children
 533         * of its own, we get through it without ever unlocking the parent.
 534         * There is one exception, though - if we ascend from a child that
 535         * gets killed as soon as we unlock it, the next sibling is found
 536         * using the value left in its ->d_child.next.  And if _that_
 537         * pointed to a cursor, and cursor got moved (e.g. by lseek())
 538         * before d_walk() regains parent->d_lock, we'll end up skipping
 539         * everything the cursor had been moved past.
 540         *
 541         * Solution: make sure that the pointer left behind in ->d_child.next
 542         * points to something that won't be moving around.  I.e. skip the
 543         * cursors.
 544         */
 545        while (dentry->d_child.next != &parent->d_subdirs) {
 546                next = list_entry(dentry->d_child.next, struct dentry, d_child);
 547                if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
 548                        break;
 549                dentry->d_child.next = next->d_child.next;
 550        }
 551}
 552
 553static void __dentry_kill(struct dentry *dentry)
 554{
 555        struct dentry *parent = NULL;
 556        bool can_free = true;
 557        if (!IS_ROOT(dentry))
 558                parent = dentry->d_parent;
 559
 560        /*
 561         * The dentry is now unrecoverably dead to the world.
 562         */
 563        lockref_mark_dead(&dentry->d_lockref);
 564
 565        /*
 566         * inform the fs via d_prune that this dentry is about to be
 567         * unhashed and destroyed.
 568         */
 569        if (dentry->d_flags & DCACHE_OP_PRUNE)
 570                dentry->d_op->d_prune(dentry);
 571
 572        if (dentry->d_flags & DCACHE_LRU_LIST) {
 573                if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
 574                        d_lru_del(dentry);
 575        }
 576        /* if it was on the hash then remove it */
 577        __d_drop(dentry);
 578        dentry_unlist(dentry, parent);
 579        if (parent)
 580                spin_unlock(&parent->d_lock);
 581        if (dentry->d_inode)
 582                dentry_unlink_inode(dentry);
 583        else
 584                spin_unlock(&dentry->d_lock);
 585        this_cpu_dec(nr_dentry);
 586        if (dentry->d_op && dentry->d_op->d_release)
 587                dentry->d_op->d_release(dentry);
 588
 589        spin_lock(&dentry->d_lock);
 590        if (dentry->d_flags & DCACHE_SHRINK_LIST) {
 591                dentry->d_flags |= DCACHE_MAY_FREE;
 592                can_free = false;
 593        }
 594        spin_unlock(&dentry->d_lock);
 595        if (likely(can_free))
 596                dentry_free(dentry);
 597        cond_resched();
 598}
 599
 600static struct dentry *__lock_parent(struct dentry *dentry)
 601{
 602        struct dentry *parent;
 603        rcu_read_lock();
 604        spin_unlock(&dentry->d_lock);
 605again:
 606        parent = READ_ONCE(dentry->d_parent);
 607        spin_lock(&parent->d_lock);
 608        /*
 609         * We can't blindly lock dentry until we are sure
 610         * that we won't violate the locking order.
 611         * Any changes of dentry->d_parent must have
 612         * been done with parent->d_lock held, so
 613         * spin_lock() above is enough of a barrier
 614         * for checking if it's still our child.
 615         */
 616        if (unlikely(parent != dentry->d_parent)) {
 617                spin_unlock(&parent->d_lock);
 618                goto again;
 619        }
 620        rcu_read_unlock();
 621        if (parent != dentry)
 622                spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 623        else
 624                parent = NULL;
 625        return parent;
 626}
 627
 628static inline struct dentry *lock_parent(struct dentry *dentry)
 629{
 630        struct dentry *parent = dentry->d_parent;
 631        if (IS_ROOT(dentry))
 632                return NULL;
 633        if (likely(spin_trylock(&parent->d_lock)))
 634                return parent;
 635        return __lock_parent(dentry);
 636}
 637
 638static inline bool retain_dentry(struct dentry *dentry)
 639{
 640        WARN_ON(d_in_lookup(dentry));
 641
 642        /* Unreachable? Get rid of it */
 643        if (unlikely(d_unhashed(dentry)))
 644                return false;
 645
 646        if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
 647                return false;
 648
 649        if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
 650                if (dentry->d_op->d_delete(dentry))
 651                        return false;
 652        }
 653
 654        if (unlikely(dentry->d_flags & DCACHE_DONTCACHE))
 655                return false;
 656
 657        /* retain; LRU fodder */
 658        dentry->d_lockref.count--;
 659        if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
 660                d_lru_add(dentry);
 661        else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
 662                dentry->d_flags |= DCACHE_REFERENCED;
 663        return true;
 664}
 665
 666void d_mark_dontcache(struct inode *inode)
 667{
 668        struct dentry *de;
 669
 670        spin_lock(&inode->i_lock);
 671        hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
 672                spin_lock(&de->d_lock);
 673                de->d_flags |= DCACHE_DONTCACHE;
 674                spin_unlock(&de->d_lock);
 675        }
 676        inode->i_state |= I_DONTCACHE;
 677        spin_unlock(&inode->i_lock);
 678}
 679EXPORT_SYMBOL(d_mark_dontcache);
 680
 681/*
 682 * Finish off a dentry we've decided to kill.
 683 * dentry->d_lock must be held, returns with it unlocked.
 684 * Returns dentry requiring refcount drop, or NULL if we're done.
 685 */
 686static struct dentry *dentry_kill(struct dentry *dentry)
 687        __releases(dentry->d_lock)
 688{
 689        struct inode *inode = dentry->d_inode;
 690        struct dentry *parent = NULL;
 691
 692        if (inode && unlikely(!spin_trylock(&inode->i_lock)))
 693                goto slow_positive;
 694
 695        if (!IS_ROOT(dentry)) {
 696                parent = dentry->d_parent;
 697                if (unlikely(!spin_trylock(&parent->d_lock))) {
 698                        parent = __lock_parent(dentry);
 699                        if (likely(inode || !dentry->d_inode))
 700                                goto got_locks;
 701                        /* negative that became positive */
 702                        if (parent)
 703                                spin_unlock(&parent->d_lock);
 704                        inode = dentry->d_inode;
 705                        goto slow_positive;
 706                }
 707        }
 708        __dentry_kill(dentry);
 709        return parent;
 710
 711slow_positive:
 712        spin_unlock(&dentry->d_lock);
 713        spin_lock(&inode->i_lock);
 714        spin_lock(&dentry->d_lock);
 715        parent = lock_parent(dentry);
 716got_locks:
 717        if (unlikely(dentry->d_lockref.count != 1)) {
 718                dentry->d_lockref.count--;
 719        } else if (likely(!retain_dentry(dentry))) {
 720                __dentry_kill(dentry);
 721                return parent;
 722        }
 723        /* we are keeping it, after all */
 724        if (inode)
 725                spin_unlock(&inode->i_lock);
 726        if (parent)
 727                spin_unlock(&parent->d_lock);
 728        spin_unlock(&dentry->d_lock);
 729        return NULL;
 730}
 731
 732/*
 733 * Try to do a lockless dput(), and return whether that was successful.
 734 *
 735 * If unsuccessful, we return false, having already taken the dentry lock.
 736 *
 737 * The caller needs to hold the RCU read lock, so that the dentry is
 738 * guaranteed to stay around even if the refcount goes down to zero!
 739 */
 740static inline bool fast_dput(struct dentry *dentry)
 741{
 742        int ret;
 743        unsigned int d_flags;
 744
 745        /*
 746         * If we have a d_op->d_delete() operation, we sould not
 747         * let the dentry count go to zero, so use "put_or_lock".
 748         */
 749        if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
 750                return lockref_put_or_lock(&dentry->d_lockref);
 751
 752        /*
 753         * .. otherwise, we can try to just decrement the
 754         * lockref optimistically.
 755         */
 756        ret = lockref_put_return(&dentry->d_lockref);
 757
 758        /*
 759         * If the lockref_put_return() failed due to the lock being held
 760         * by somebody else, the fast path has failed. We will need to
 761         * get the lock, and then check the count again.
 762         */
 763        if (unlikely(ret < 0)) {
 764                spin_lock(&dentry->d_lock);
 765                if (dentry->d_lockref.count > 1) {
 766                        dentry->d_lockref.count--;
 767                        spin_unlock(&dentry->d_lock);
 768                        return true;
 769                }
 770                return false;
 771        }
 772
 773        /*
 774         * If we weren't the last ref, we're done.
 775         */
 776        if (ret)
 777                return true;
 778
 779        /*
 780         * Careful, careful. The reference count went down
 781         * to zero, but we don't hold the dentry lock, so
 782         * somebody else could get it again, and do another
 783         * dput(), and we need to not race with that.
 784         *
 785         * However, there is a very special and common case
 786         * where we don't care, because there is nothing to
 787         * do: the dentry is still hashed, it does not have
 788         * a 'delete' op, and it's referenced and already on
 789         * the LRU list.
 790         *
 791         * NOTE! Since we aren't locked, these values are
 792         * not "stable". However, it is sufficient that at
 793         * some point after we dropped the reference the
 794         * dentry was hashed and the flags had the proper
 795         * value. Other dentry users may have re-gotten
 796         * a reference to the dentry and change that, but
 797         * our work is done - we can leave the dentry
 798         * around with a zero refcount.
 799         *
 800         * Nevertheless, there are two cases that we should kill
 801         * the dentry anyway.
 802         * 1. free disconnected dentries as soon as their refcount
 803         *    reached zero.
 804         * 2. free dentries if they should not be cached.
 805         */
 806        smp_rmb();
 807        d_flags = READ_ONCE(dentry->d_flags);
 808        d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
 809                        DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
 810
 811        /* Nothing to do? Dropping the reference was all we needed? */
 812        if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
 813                return true;
 814
 815        /*
 816         * Not the fast normal case? Get the lock. We've already decremented
 817         * the refcount, but we'll need to re-check the situation after
 818         * getting the lock.
 819         */
 820        spin_lock(&dentry->d_lock);
 821
 822        /*
 823         * Did somebody else grab a reference to it in the meantime, and
 824         * we're no longer the last user after all? Alternatively, somebody
 825         * else could have killed it and marked it dead. Either way, we
 826         * don't need to do anything else.
 827         */
 828        if (dentry->d_lockref.count) {
 829                spin_unlock(&dentry->d_lock);
 830                return true;
 831        }
 832
 833        /*
 834         * Re-get the reference we optimistically dropped. We hold the
 835         * lock, and we just tested that it was zero, so we can just
 836         * set it to 1.
 837         */
 838        dentry->d_lockref.count = 1;
 839        return false;
 840}
 841
 842
 843/* 
 844 * This is dput
 845 *
 846 * This is complicated by the fact that we do not want to put
 847 * dentries that are no longer on any hash chain on the unused
 848 * list: we'd much rather just get rid of them immediately.
 849 *
 850 * However, that implies that we have to traverse the dentry
 851 * tree upwards to the parents which might _also_ now be
 852 * scheduled for deletion (it may have been only waiting for
 853 * its last child to go away).
 854 *
 855 * This tail recursion is done by hand as we don't want to depend
 856 * on the compiler to always get this right (gcc generally doesn't).
 857 * Real recursion would eat up our stack space.
 858 */
 859
 860/*
 861 * dput - release a dentry
 862 * @dentry: dentry to release 
 863 *
 864 * Release a dentry. This will drop the usage count and if appropriate
 865 * call the dentry unlink method as well as removing it from the queues and
 866 * releasing its resources. If the parent dentries were scheduled for release
 867 * they too may now get deleted.
 868 */
 869void dput(struct dentry *dentry)
 870{
 871        while (dentry) {
 872                might_sleep();
 873
 874                rcu_read_lock();
 875                if (likely(fast_dput(dentry))) {
 876                        rcu_read_unlock();
 877                        return;
 878                }
 879
 880                /* Slow case: now with the dentry lock held */
 881                rcu_read_unlock();
 882
 883                if (likely(retain_dentry(dentry))) {
 884                        spin_unlock(&dentry->d_lock);
 885                        return;
 886                }
 887
 888                dentry = dentry_kill(dentry);
 889        }
 890}
 891EXPORT_SYMBOL(dput);
 892
 893static void __dput_to_list(struct dentry *dentry, struct list_head *list)
 894__must_hold(&dentry->d_lock)
 895{
 896        if (dentry->d_flags & DCACHE_SHRINK_LIST) {
 897                /* let the owner of the list it's on deal with it */
 898                --dentry->d_lockref.count;
 899        } else {
 900                if (dentry->d_flags & DCACHE_LRU_LIST)
 901                        d_lru_del(dentry);
 902                if (!--dentry->d_lockref.count)
 903                        d_shrink_add(dentry, list);
 904        }
 905}
 906
 907void dput_to_list(struct dentry *dentry, struct list_head *list)
 908{
 909        rcu_read_lock();
 910        if (likely(fast_dput(dentry))) {
 911                rcu_read_unlock();
 912                return;
 913        }
 914        rcu_read_unlock();
 915        if (!retain_dentry(dentry))
 916                __dput_to_list(dentry, list);
 917        spin_unlock(&dentry->d_lock);
 918}
 919
 920/* This must be called with d_lock held */
 921static inline void __dget_dlock(struct dentry *dentry)
 922{
 923        dentry->d_lockref.count++;
 924}
 925
 926static inline void __dget(struct dentry *dentry)
 927{
 928        lockref_get(&dentry->d_lockref);
 929}
 930
 931struct dentry *dget_parent(struct dentry *dentry)
 932{
 933        int gotref;
 934        struct dentry *ret;
 935        unsigned seq;
 936
 937        /*
 938         * Do optimistic parent lookup without any
 939         * locking.
 940         */
 941        rcu_read_lock();
 942        seq = raw_seqcount_begin(&dentry->d_seq);
 943        ret = READ_ONCE(dentry->d_parent);
 944        gotref = lockref_get_not_zero(&ret->d_lockref);
 945        rcu_read_unlock();
 946        if (likely(gotref)) {
 947                if (!read_seqcount_retry(&dentry->d_seq, seq))
 948                        return ret;
 949                dput(ret);
 950        }
 951
 952repeat:
 953        /*
 954         * Don't need rcu_dereference because we re-check it was correct under
 955         * the lock.
 956         */
 957        rcu_read_lock();
 958        ret = dentry->d_parent;
 959        spin_lock(&ret->d_lock);
 960        if (unlikely(ret != dentry->d_parent)) {
 961                spin_unlock(&ret->d_lock);
 962                rcu_read_unlock();
 963                goto repeat;
 964        }
 965        rcu_read_unlock();
 966        BUG_ON(!ret->d_lockref.count);
 967        ret->d_lockref.count++;
 968        spin_unlock(&ret->d_lock);
 969        return ret;
 970}
 971EXPORT_SYMBOL(dget_parent);
 972
 973static struct dentry * __d_find_any_alias(struct inode *inode)
 974{
 975        struct dentry *alias;
 976
 977        if (hlist_empty(&inode->i_dentry))
 978                return NULL;
 979        alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
 980        __dget(alias);
 981        return alias;
 982}
 983
 984/**
 985 * d_find_any_alias - find any alias for a given inode
 986 * @inode: inode to find an alias for
 987 *
 988 * If any aliases exist for the given inode, take and return a
 989 * reference for one of them.  If no aliases exist, return %NULL.
 990 */
 991struct dentry *d_find_any_alias(struct inode *inode)
 992{
 993        struct dentry *de;
 994
 995        spin_lock(&inode->i_lock);
 996        de = __d_find_any_alias(inode);
 997        spin_unlock(&inode->i_lock);
 998        return de;
 999}
1000EXPORT_SYMBOL(d_find_any_alias);
1001
1002static struct dentry *__d_find_alias(struct inode *inode)
1003{
1004        struct dentry *alias;
1005
1006        if (S_ISDIR(inode->i_mode))
1007                return __d_find_any_alias(inode);
1008
1009        hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1010                spin_lock(&alias->d_lock);
1011                if (!d_unhashed(alias)) {
1012                        __dget_dlock(alias);
1013                        spin_unlock(&alias->d_lock);
1014                        return alias;
1015                }
1016                spin_unlock(&alias->d_lock);
1017        }
1018        return NULL;
1019}
1020
1021/**
1022 * d_find_alias - grab a hashed alias of inode
1023 * @inode: inode in question
1024 *
1025 * If inode has a hashed alias, or is a directory and has any alias,
1026 * acquire the reference to alias and return it. Otherwise return NULL.
1027 * Notice that if inode is a directory there can be only one alias and
1028 * it can be unhashed only if it has no children, or if it is the root
1029 * of a filesystem, or if the directory was renamed and d_revalidate
1030 * was the first vfs operation to notice.
1031 *
1032 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1033 * any other hashed alias over that one.
1034 */
1035struct dentry *d_find_alias(struct inode *inode)
1036{
1037        struct dentry *de = NULL;
1038
1039        if (!hlist_empty(&inode->i_dentry)) {
1040                spin_lock(&inode->i_lock);
1041                de = __d_find_alias(inode);
1042                spin_unlock(&inode->i_lock);
1043        }
1044        return de;
1045}
1046EXPORT_SYMBOL(d_find_alias);
1047
1048/*
1049 *  Caller MUST be holding rcu_read_lock() and be guaranteed
1050 *  that inode won't get freed until rcu_read_unlock().
1051 */
1052struct dentry *d_find_alias_rcu(struct inode *inode)
1053{
1054        struct hlist_head *l = &inode->i_dentry;
1055        struct dentry *de = NULL;
1056
1057        spin_lock(&inode->i_lock);
1058        // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1059        // used without having I_FREEING set, which means no aliases left
1060        if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
1061                if (S_ISDIR(inode->i_mode)) {
1062                        de = hlist_entry(l->first, struct dentry, d_u.d_alias);
1063                } else {
1064                        hlist_for_each_entry(de, l, d_u.d_alias)
1065                                if (!d_unhashed(de))
1066                                        break;
1067                }
1068        }
1069        spin_unlock(&inode->i_lock);
1070        return de;
1071}
1072
1073/*
1074 *      Try to kill dentries associated with this inode.
1075 * WARNING: you must own a reference to inode.
1076 */
1077void d_prune_aliases(struct inode *inode)
1078{
1079        struct dentry *dentry;
1080restart:
1081        spin_lock(&inode->i_lock);
1082        hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1083                spin_lock(&dentry->d_lock);
1084                if (!dentry->d_lockref.count) {
1085                        struct dentry *parent = lock_parent(dentry);
1086                        if (likely(!dentry->d_lockref.count)) {
1087                                __dentry_kill(dentry);
1088                                dput(parent);
1089                                goto restart;
1090                        }
1091                        if (parent)
1092                                spin_unlock(&parent->d_lock);
1093                }
1094                spin_unlock(&dentry->d_lock);
1095        }
1096        spin_unlock(&inode->i_lock);
1097}
1098EXPORT_SYMBOL(d_prune_aliases);
1099
1100/*
1101 * Lock a dentry from shrink list.
1102 * Called under rcu_read_lock() and dentry->d_lock; the former
1103 * guarantees that nothing we access will be freed under us.
1104 * Note that dentry is *not* protected from concurrent dentry_kill(),
1105 * d_delete(), etc.
1106 *
1107 * Return false if dentry has been disrupted or grabbed, leaving
1108 * the caller to kick it off-list.  Otherwise, return true and have
1109 * that dentry's inode and parent both locked.
1110 */
1111static bool shrink_lock_dentry(struct dentry *dentry)
1112{
1113        struct inode *inode;
1114        struct dentry *parent;
1115
1116        if (dentry->d_lockref.count)
1117                return false;
1118
1119        inode = dentry->d_inode;
1120        if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1121                spin_unlock(&dentry->d_lock);
1122                spin_lock(&inode->i_lock);
1123                spin_lock(&dentry->d_lock);
1124                if (unlikely(dentry->d_lockref.count))
1125                        goto out;
1126                /* changed inode means that somebody had grabbed it */
1127                if (unlikely(inode != dentry->d_inode))
1128                        goto out;
1129        }
1130
1131        parent = dentry->d_parent;
1132        if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1133                return true;
1134
1135        spin_unlock(&dentry->d_lock);
1136        spin_lock(&parent->d_lock);
1137        if (unlikely(parent != dentry->d_parent)) {
1138                spin_unlock(&parent->d_lock);
1139                spin_lock(&dentry->d_lock);
1140                goto out;
1141        }
1142        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1143        if (likely(!dentry->d_lockref.count))
1144                return true;
1145        spin_unlock(&parent->d_lock);
1146out:
1147        if (inode)
1148                spin_unlock(&inode->i_lock);
1149        return false;
1150}
1151
1152void shrink_dentry_list(struct list_head *list)
1153{
1154        while (!list_empty(list)) {
1155                struct dentry *dentry, *parent;
1156
1157                dentry = list_entry(list->prev, struct dentry, d_lru);
1158                spin_lock(&dentry->d_lock);
1159                rcu_read_lock();
1160                if (!shrink_lock_dentry(dentry)) {
1161                        bool can_free = false;
1162                        rcu_read_unlock();
1163                        d_shrink_del(dentry);
1164                        if (dentry->d_lockref.count < 0)
1165                                can_free = dentry->d_flags & DCACHE_MAY_FREE;
1166                        spin_unlock(&dentry->d_lock);
1167                        if (can_free)
1168                                dentry_free(dentry);
1169                        continue;
1170                }
1171                rcu_read_unlock();
1172                d_shrink_del(dentry);
1173                parent = dentry->d_parent;
1174                if (parent != dentry)
1175                        __dput_to_list(parent, list);
1176                __dentry_kill(dentry);
1177        }
1178}
1179
1180static enum lru_status dentry_lru_isolate(struct list_head *item,
1181                struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1182{
1183        struct list_head *freeable = arg;
1184        struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1185
1186
1187        /*
1188         * we are inverting the lru lock/dentry->d_lock here,
1189         * so use a trylock. If we fail to get the lock, just skip
1190         * it
1191         */
1192        if (!spin_trylock(&dentry->d_lock))
1193                return LRU_SKIP;
1194
1195        /*
1196         * Referenced dentries are still in use. If they have active
1197         * counts, just remove them from the LRU. Otherwise give them
1198         * another pass through the LRU.
1199         */
1200        if (dentry->d_lockref.count) {
1201                d_lru_isolate(lru, dentry);
1202                spin_unlock(&dentry->d_lock);
1203                return LRU_REMOVED;
1204        }
1205
1206        if (dentry->d_flags & DCACHE_REFERENCED) {
1207                dentry->d_flags &= ~DCACHE_REFERENCED;
1208                spin_unlock(&dentry->d_lock);
1209
1210                /*
1211                 * The list move itself will be made by the common LRU code. At
1212                 * this point, we've dropped the dentry->d_lock but keep the
1213                 * lru lock. This is safe to do, since every list movement is
1214                 * protected by the lru lock even if both locks are held.
1215                 *
1216                 * This is guaranteed by the fact that all LRU management
1217                 * functions are intermediated by the LRU API calls like
1218                 * list_lru_add and list_lru_del. List movement in this file
1219                 * only ever occur through this functions or through callbacks
1220                 * like this one, that are called from the LRU API.
1221                 *
1222                 * The only exceptions to this are functions like
1223                 * shrink_dentry_list, and code that first checks for the
1224                 * DCACHE_SHRINK_LIST flag.  Those are guaranteed to be
1225                 * operating only with stack provided lists after they are
1226                 * properly isolated from the main list.  It is thus, always a
1227                 * local access.
1228                 */
1229                return LRU_ROTATE;
1230        }
1231
1232        d_lru_shrink_move(lru, dentry, freeable);
1233        spin_unlock(&dentry->d_lock);
1234
1235        return LRU_REMOVED;
1236}
1237
1238/**
1239 * prune_dcache_sb - shrink the dcache
1240 * @sb: superblock
1241 * @sc: shrink control, passed to list_lru_shrink_walk()
1242 *
1243 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1244 * is done when we need more memory and called from the superblock shrinker
1245 * function.
1246 *
1247 * This function may fail to free any resources if all the dentries are in
1248 * use.
1249 */
1250long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1251{
1252        LIST_HEAD(dispose);
1253        long freed;
1254
1255        freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1256                                     dentry_lru_isolate, &dispose);
1257        shrink_dentry_list(&dispose);
1258        return freed;
1259}
1260
1261static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1262                struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1263{
1264        struct list_head *freeable = arg;
1265        struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1266
1267        /*
1268         * we are inverting the lru lock/dentry->d_lock here,
1269         * so use a trylock. If we fail to get the lock, just skip
1270         * it
1271         */
1272        if (!spin_trylock(&dentry->d_lock))
1273                return LRU_SKIP;
1274
1275        d_lru_shrink_move(lru, dentry, freeable);
1276        spin_unlock(&dentry->d_lock);
1277
1278        return LRU_REMOVED;
1279}
1280
1281
1282/**
1283 * shrink_dcache_sb - shrink dcache for a superblock
1284 * @sb: superblock
1285 *
1286 * Shrink the dcache for the specified super block. This is used to free
1287 * the dcache before unmounting a file system.
1288 */
1289void shrink_dcache_sb(struct super_block *sb)
1290{
1291        do {
1292                LIST_HEAD(dispose);
1293
1294                list_lru_walk(&sb->s_dentry_lru,
1295                        dentry_lru_isolate_shrink, &dispose, 1024);
1296                shrink_dentry_list(&dispose);
1297        } while (list_lru_count(&sb->s_dentry_lru) > 0);
1298}
1299EXPORT_SYMBOL(shrink_dcache_sb);
1300
1301/**
1302 * enum d_walk_ret - action to talke during tree walk
1303 * @D_WALK_CONTINUE:    contrinue walk
1304 * @D_WALK_QUIT:        quit walk
1305 * @D_WALK_NORETRY:     quit when retry is needed
1306 * @D_WALK_SKIP:        skip this dentry and its children
1307 */
1308enum d_walk_ret {
1309        D_WALK_CONTINUE,
1310        D_WALK_QUIT,
1311        D_WALK_NORETRY,
1312        D_WALK_SKIP,
1313};
1314
1315/**
1316 * d_walk - walk the dentry tree
1317 * @parent:     start of walk
1318 * @data:       data passed to @enter() and @finish()
1319 * @enter:      callback when first entering the dentry
1320 *
1321 * The @enter() callbacks are called with d_lock held.
1322 */
1323static void d_walk(struct dentry *parent, void *data,
1324                   enum d_walk_ret (*enter)(void *, struct dentry *))
1325{
1326        struct dentry *this_parent;
1327        struct list_head *next;
1328        unsigned seq = 0;
1329        enum d_walk_ret ret;
1330        bool retry = true;
1331
1332again:
1333        read_seqbegin_or_lock(&rename_lock, &seq);
1334        this_parent = parent;
1335        spin_lock(&this_parent->d_lock);
1336
1337        ret = enter(data, this_parent);
1338        switch (ret) {
1339        case D_WALK_CONTINUE:
1340                break;
1341        case D_WALK_QUIT:
1342        case D_WALK_SKIP:
1343                goto out_unlock;
1344        case D_WALK_NORETRY:
1345                retry = false;
1346                break;
1347        }
1348repeat:
1349        next = this_parent->d_subdirs.next;
1350resume:
1351        while (next != &this_parent->d_subdirs) {
1352                struct list_head *tmp = next;
1353                struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1354                next = tmp->next;
1355
1356                if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1357                        continue;
1358
1359                spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1360
1361                ret = enter(data, dentry);
1362                switch (ret) {
1363                case D_WALK_CONTINUE:
1364                        break;
1365                case D_WALK_QUIT:
1366                        spin_unlock(&dentry->d_lock);
1367                        goto out_unlock;
1368                case D_WALK_NORETRY:
1369                        retry = false;
1370                        break;
1371                case D_WALK_SKIP:
1372                        spin_unlock(&dentry->d_lock);
1373                        continue;
1374                }
1375
1376                if (!list_empty(&dentry->d_subdirs)) {
1377                        spin_unlock(&this_parent->d_lock);
1378                        spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1379                        this_parent = dentry;
1380                        spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1381                        goto repeat;
1382                }
1383                spin_unlock(&dentry->d_lock);
1384        }
1385        /*
1386         * All done at this level ... ascend and resume the search.
1387         */
1388        rcu_read_lock();
1389ascend:
1390        if (this_parent != parent) {
1391                struct dentry *child = this_parent;
1392                this_parent = child->d_parent;
1393
1394                spin_unlock(&child->d_lock);
1395                spin_lock(&this_parent->d_lock);
1396
1397                /* might go back up the wrong parent if we have had a rename. */
1398                if (need_seqretry(&rename_lock, seq))
1399                        goto rename_retry;
1400                /* go into the first sibling still alive */
1401                do {
1402                        next = child->d_child.next;
1403                        if (next == &this_parent->d_subdirs)
1404                                goto ascend;
1405                        child = list_entry(next, struct dentry, d_child);
1406                } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1407                rcu_read_unlock();
1408                goto resume;
1409        }
1410        if (need_seqretry(&rename_lock, seq))
1411                goto rename_retry;
1412        rcu_read_unlock();
1413
1414out_unlock:
1415        spin_unlock(&this_parent->d_lock);
1416        done_seqretry(&rename_lock, seq);
1417        return;
1418
1419rename_retry:
1420        spin_unlock(&this_parent->d_lock);
1421        rcu_read_unlock();
1422        BUG_ON(seq & 1);
1423        if (!retry)
1424                return;
1425        seq = 1;
1426        goto again;
1427}
1428
1429struct check_mount {
1430        struct vfsmount *mnt;
1431        unsigned int mounted;
1432};
1433
1434static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1435{
1436        struct check_mount *info = data;
1437        struct path path = { .mnt = info->mnt, .dentry = dentry };
1438
1439        if (likely(!d_mountpoint(dentry)))
1440                return D_WALK_CONTINUE;
1441        if (__path_is_mountpoint(&path)) {
1442                info->mounted = 1;
1443                return D_WALK_QUIT;
1444        }
1445        return D_WALK_CONTINUE;
1446}
1447
1448/**
1449 * path_has_submounts - check for mounts over a dentry in the
1450 *                      current namespace.
1451 * @parent: path to check.
1452 *
1453 * Return true if the parent or its subdirectories contain
1454 * a mount point in the current namespace.
1455 */
1456int path_has_submounts(const struct path *parent)
1457{
1458        struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1459
1460        read_seqlock_excl(&mount_lock);
1461        d_walk(parent->dentry, &data, path_check_mount);
1462        read_sequnlock_excl(&mount_lock);
1463
1464        return data.mounted;
1465}
1466EXPORT_SYMBOL(path_has_submounts);
1467
1468/*
1469 * Called by mount code to set a mountpoint and check if the mountpoint is
1470 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1471 * subtree can become unreachable).
1472 *
1473 * Only one of d_invalidate() and d_set_mounted() must succeed.  For
1474 * this reason take rename_lock and d_lock on dentry and ancestors.
1475 */
1476int d_set_mounted(struct dentry *dentry)
1477{
1478        struct dentry *p;
1479        int ret = -ENOENT;
1480        write_seqlock(&rename_lock);
1481        for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1482                /* Need exclusion wrt. d_invalidate() */
1483                spin_lock(&p->d_lock);
1484                if (unlikely(d_unhashed(p))) {
1485                        spin_unlock(&p->d_lock);
1486                        goto out;
1487                }
1488                spin_unlock(&p->d_lock);
1489        }
1490        spin_lock(&dentry->d_lock);
1491        if (!d_unlinked(dentry)) {
1492                ret = -EBUSY;
1493                if (!d_mountpoint(dentry)) {
1494                        dentry->d_flags |= DCACHE_MOUNTED;
1495                        ret = 0;
1496                }
1497        }
1498        spin_unlock(&dentry->d_lock);
1499out:
1500        write_sequnlock(&rename_lock);
1501        return ret;
1502}
1503
1504/*
1505 * Search the dentry child list of the specified parent,
1506 * and move any unused dentries to the end of the unused
1507 * list for prune_dcache(). We descend to the next level
1508 * whenever the d_subdirs list is non-empty and continue
1509 * searching.
1510 *
1511 * It returns zero iff there are no unused children,
1512 * otherwise  it returns the number of children moved to
1513 * the end of the unused list. This may not be the total
1514 * number of unused children, because select_parent can
1515 * drop the lock and return early due to latency
1516 * constraints.
1517 */
1518
1519struct select_data {
1520        struct dentry *start;
1521        union {
1522                long found;
1523                struct dentry *victim;
1524        };
1525        struct list_head dispose;
1526};
1527
1528static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1529{
1530        struct select_data *data = _data;
1531        enum d_walk_ret ret = D_WALK_CONTINUE;
1532
1533        if (data->start == dentry)
1534                goto out;
1535
1536        if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1537                data->found++;
1538        } else {
1539                if (dentry->d_flags & DCACHE_LRU_LIST)
1540                        d_lru_del(dentry);
1541                if (!dentry->d_lockref.count) {
1542                        d_shrink_add(dentry, &data->dispose);
1543                        data->found++;
1544                }
1545        }
1546        /*
1547         * We can return to the caller if we have found some (this
1548         * ensures forward progress). We'll be coming back to find
1549         * the rest.
1550         */
1551        if (!list_empty(&data->dispose))
1552                ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1553out:
1554        return ret;
1555}
1556
1557static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1558{
1559        struct select_data *data = _data;
1560        enum d_walk_ret ret = D_WALK_CONTINUE;
1561
1562        if (data->start == dentry)
1563                goto out;
1564
1565        if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1566                if (!dentry->d_lockref.count) {
1567                        rcu_read_lock();
1568                        data->victim = dentry;
1569                        return D_WALK_QUIT;
1570                }
1571        } else {
1572                if (dentry->d_flags & DCACHE_LRU_LIST)
1573                        d_lru_del(dentry);
1574                if (!dentry->d_lockref.count)
1575                        d_shrink_add(dentry, &data->dispose);
1576        }
1577        /*
1578         * We can return to the caller if we have found some (this
1579         * ensures forward progress). We'll be coming back to find
1580         * the rest.
1581         */
1582        if (!list_empty(&data->dispose))
1583                ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1584out:
1585        return ret;
1586}
1587
1588/**
1589 * shrink_dcache_parent - prune dcache
1590 * @parent: parent of entries to prune
1591 *
1592 * Prune the dcache to remove unused children of the parent dentry.
1593 */
1594void shrink_dcache_parent(struct dentry *parent)
1595{
1596        for (;;) {
1597                struct select_data data = {.start = parent};
1598
1599                INIT_LIST_HEAD(&data.dispose);
1600                d_walk(parent, &data, select_collect);
1601
1602                if (!list_empty(&data.dispose)) {
1603                        shrink_dentry_list(&data.dispose);
1604                        continue;
1605                }
1606
1607                cond_resched();
1608                if (!data.found)
1609                        break;
1610                data.victim = NULL;
1611                d_walk(parent, &data, select_collect2);
1612                if (data.victim) {
1613                        struct dentry *parent;
1614                        spin_lock(&data.victim->d_lock);
1615                        if (!shrink_lock_dentry(data.victim)) {
1616                                spin_unlock(&data.victim->d_lock);
1617                                rcu_read_unlock();
1618                        } else {
1619                                rcu_read_unlock();
1620                                parent = data.victim->d_parent;
1621                                if (parent != data.victim)
1622                                        __dput_to_list(parent, &data.dispose);
1623                                __dentry_kill(data.victim);
1624                        }
1625                }
1626                if (!list_empty(&data.dispose))
1627                        shrink_dentry_list(&data.dispose);
1628        }
1629}
1630EXPORT_SYMBOL(shrink_dcache_parent);
1631
1632static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1633{
1634        /* it has busy descendents; complain about those instead */
1635        if (!list_empty(&dentry->d_subdirs))
1636                return D_WALK_CONTINUE;
1637
1638        /* root with refcount 1 is fine */
1639        if (dentry == _data && dentry->d_lockref.count == 1)
1640                return D_WALK_CONTINUE;
1641
1642        printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1643                        " still in use (%d) [unmount of %s %s]\n",
1644                       dentry,
1645                       dentry->d_inode ?
1646                       dentry->d_inode->i_ino : 0UL,
1647                       dentry,
1648                       dentry->d_lockref.count,
1649                       dentry->d_sb->s_type->name,
1650                       dentry->d_sb->s_id);
1651        WARN_ON(1);
1652        return D_WALK_CONTINUE;
1653}
1654
1655static void do_one_tree(struct dentry *dentry)
1656{
1657        shrink_dcache_parent(dentry);
1658        d_walk(dentry, dentry, umount_check);
1659        d_drop(dentry);
1660        dput(dentry);
1661}
1662
1663/*
1664 * destroy the dentries attached to a superblock on unmounting
1665 */
1666void shrink_dcache_for_umount(struct super_block *sb)
1667{
1668        struct dentry *dentry;
1669
1670        WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1671
1672        dentry = sb->s_root;
1673        sb->s_root = NULL;
1674        do_one_tree(dentry);
1675
1676        while (!hlist_bl_empty(&sb->s_roots)) {
1677                dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1678                do_one_tree(dentry);
1679        }
1680}
1681
1682static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1683{
1684        struct dentry **victim = _data;
1685        if (d_mountpoint(dentry)) {
1686                __dget_dlock(dentry);
1687                *victim = dentry;
1688                return D_WALK_QUIT;
1689        }
1690        return D_WALK_CONTINUE;
1691}
1692
1693/**
1694 * d_invalidate - detach submounts, prune dcache, and drop
1695 * @dentry: dentry to invalidate (aka detach, prune and drop)
1696 */
1697void d_invalidate(struct dentry *dentry)
1698{
1699        bool had_submounts = false;
1700        spin_lock(&dentry->d_lock);
1701        if (d_unhashed(dentry)) {
1702                spin_unlock(&dentry->d_lock);
1703                return;
1704        }
1705        __d_drop(dentry);
1706        spin_unlock(&dentry->d_lock);
1707
1708        /* Negative dentries can be dropped without further checks */
1709        if (!dentry->d_inode)
1710                return;
1711
1712        shrink_dcache_parent(dentry);
1713        for (;;) {
1714                struct dentry *victim = NULL;
1715                d_walk(dentry, &victim, find_submount);
1716                if (!victim) {
1717                        if (had_submounts)
1718                                shrink_dcache_parent(dentry);
1719                        return;
1720                }
1721                had_submounts = true;
1722                detach_mounts(victim);
1723                dput(victim);
1724        }
1725}
1726EXPORT_SYMBOL(d_invalidate);
1727
1728/**
1729 * __d_alloc    -       allocate a dcache entry
1730 * @sb: filesystem it will belong to
1731 * @name: qstr of the name
1732 *
1733 * Allocates a dentry. It returns %NULL if there is insufficient memory
1734 * available. On a success the dentry is returned. The name passed in is
1735 * copied and the copy passed in may be reused after this call.
1736 */
1737 
1738static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1739{
1740        struct dentry *dentry;
1741        char *dname;
1742        int err;
1743
1744        dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1745        if (!dentry)
1746                return NULL;
1747
1748        /*
1749         * We guarantee that the inline name is always NUL-terminated.
1750         * This way the memcpy() done by the name switching in rename
1751         * will still always have a NUL at the end, even if we might
1752         * be overwriting an internal NUL character
1753         */
1754        dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1755        if (unlikely(!name)) {
1756                name = &slash_name;
1757                dname = dentry->d_iname;
1758        } else if (name->len > DNAME_INLINE_LEN-1) {
1759                size_t size = offsetof(struct external_name, name[1]);
1760                struct external_name *p = kmalloc(size + name->len,
1761                                                  GFP_KERNEL_ACCOUNT |
1762                                                  __GFP_RECLAIMABLE);
1763                if (!p) {
1764                        kmem_cache_free(dentry_cache, dentry); 
1765                        return NULL;
1766                }
1767                atomic_set(&p->u.count, 1);
1768                dname = p->name;
1769        } else  {
1770                dname = dentry->d_iname;
1771        }       
1772
1773        dentry->d_name.len = name->len;
1774        dentry->d_name.hash = name->hash;
1775        memcpy(dname, name->name, name->len);
1776        dname[name->len] = 0;
1777
1778        /* Make sure we always see the terminating NUL character */
1779        smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1780
1781        dentry->d_lockref.count = 1;
1782        dentry->d_flags = 0;
1783        spin_lock_init(&dentry->d_lock);
1784        seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1785        dentry->d_inode = NULL;
1786        dentry->d_parent = dentry;
1787        dentry->d_sb = sb;
1788        dentry->d_op = NULL;
1789        dentry->d_fsdata = NULL;
1790        INIT_HLIST_BL_NODE(&dentry->d_hash);
1791        INIT_LIST_HEAD(&dentry->d_lru);
1792        INIT_LIST_HEAD(&dentry->d_subdirs);
1793        INIT_HLIST_NODE(&dentry->d_u.d_alias);
1794        INIT_LIST_HEAD(&dentry->d_child);
1795        d_set_d_op(dentry, dentry->d_sb->s_d_op);
1796
1797        if (dentry->d_op && dentry->d_op->d_init) {
1798                err = dentry->d_op->d_init(dentry);
1799                if (err) {
1800                        if (dname_external(dentry))
1801                                kfree(external_name(dentry));
1802                        kmem_cache_free(dentry_cache, dentry);
1803                        return NULL;
1804                }
1805        }
1806
1807        this_cpu_inc(nr_dentry);
1808
1809        return dentry;
1810}
1811
1812/**
1813 * d_alloc      -       allocate a dcache entry
1814 * @parent: parent of entry to allocate
1815 * @name: qstr of the name
1816 *
1817 * Allocates a dentry. It returns %NULL if there is insufficient memory
1818 * available. On a success the dentry is returned. The name passed in is
1819 * copied and the copy passed in may be reused after this call.
1820 */
1821struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1822{
1823        struct dentry *dentry = __d_alloc(parent->d_sb, name);
1824        if (!dentry)
1825                return NULL;
1826        spin_lock(&parent->d_lock);
1827        /*
1828         * don't need child lock because it is not subject
1829         * to concurrency here
1830         */
1831        __dget_dlock(parent);
1832        dentry->d_parent = parent;
1833        list_add(&dentry->d_child, &parent->d_subdirs);
1834        spin_unlock(&parent->d_lock);
1835
1836        return dentry;
1837}
1838EXPORT_SYMBOL(d_alloc);
1839
1840struct dentry *d_alloc_anon(struct super_block *sb)
1841{
1842        return __d_alloc(sb, NULL);
1843}
1844EXPORT_SYMBOL(d_alloc_anon);
1845
1846struct dentry *d_alloc_cursor(struct dentry * parent)
1847{
1848        struct dentry *dentry = d_alloc_anon(parent->d_sb);
1849        if (dentry) {
1850                dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1851                dentry->d_parent = dget(parent);
1852        }
1853        return dentry;
1854}
1855
1856/**
1857 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1858 * @sb: the superblock
1859 * @name: qstr of the name
1860 *
1861 * For a filesystem that just pins its dentries in memory and never
1862 * performs lookups at all, return an unhashed IS_ROOT dentry.
1863 * This is used for pipes, sockets et.al. - the stuff that should
1864 * never be anyone's children or parents.  Unlike all other
1865 * dentries, these will not have RCU delay between dropping the
1866 * last reference and freeing them.
1867 *
1868 * The only user is alloc_file_pseudo() and that's what should
1869 * be considered a public interface.  Don't use directly.
1870 */
1871struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1872{
1873        struct dentry *dentry = __d_alloc(sb, name);
1874        if (likely(dentry))
1875                dentry->d_flags |= DCACHE_NORCU;
1876        return dentry;
1877}
1878
1879struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1880{
1881        struct qstr q;
1882
1883        q.name = name;
1884        q.hash_len = hashlen_string(parent, name);
1885        return d_alloc(parent, &q);
1886}
1887EXPORT_SYMBOL(d_alloc_name);
1888
1889void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1890{
1891        WARN_ON_ONCE(dentry->d_op);
1892        WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH  |
1893                                DCACHE_OP_COMPARE       |
1894                                DCACHE_OP_REVALIDATE    |
1895                                DCACHE_OP_WEAK_REVALIDATE       |
1896                                DCACHE_OP_DELETE        |
1897                                DCACHE_OP_REAL));
1898        dentry->d_op = op;
1899        if (!op)
1900                return;
1901        if (op->d_hash)
1902                dentry->d_flags |= DCACHE_OP_HASH;
1903        if (op->d_compare)
1904                dentry->d_flags |= DCACHE_OP_COMPARE;
1905        if (op->d_revalidate)
1906                dentry->d_flags |= DCACHE_OP_REVALIDATE;
1907        if (op->d_weak_revalidate)
1908                dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1909        if (op->d_delete)
1910                dentry->d_flags |= DCACHE_OP_DELETE;
1911        if (op->d_prune)
1912                dentry->d_flags |= DCACHE_OP_PRUNE;
1913        if (op->d_real)
1914                dentry->d_flags |= DCACHE_OP_REAL;
1915
1916}
1917EXPORT_SYMBOL(d_set_d_op);
1918
1919
1920/*
1921 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1922 * @dentry - The dentry to mark
1923 *
1924 * Mark a dentry as falling through to the lower layer (as set with
1925 * d_pin_lower()).  This flag may be recorded on the medium.
1926 */
1927void d_set_fallthru(struct dentry *dentry)
1928{
1929        spin_lock(&dentry->d_lock);
1930        dentry->d_flags |= DCACHE_FALLTHRU;
1931        spin_unlock(&dentry->d_lock);
1932}
1933EXPORT_SYMBOL(d_set_fallthru);
1934
1935static unsigned d_flags_for_inode(struct inode *inode)
1936{
1937        unsigned add_flags = DCACHE_REGULAR_TYPE;
1938
1939        if (!inode)
1940                return DCACHE_MISS_TYPE;
1941
1942        if (S_ISDIR(inode->i_mode)) {
1943                add_flags = DCACHE_DIRECTORY_TYPE;
1944                if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1945                        if (unlikely(!inode->i_op->lookup))
1946                                add_flags = DCACHE_AUTODIR_TYPE;
1947                        else
1948                                inode->i_opflags |= IOP_LOOKUP;
1949                }
1950                goto type_determined;
1951        }
1952
1953        if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1954                if (unlikely(inode->i_op->get_link)) {
1955                        add_flags = DCACHE_SYMLINK_TYPE;
1956                        goto type_determined;
1957                }
1958                inode->i_opflags |= IOP_NOFOLLOW;
1959        }
1960
1961        if (unlikely(!S_ISREG(inode->i_mode)))
1962                add_flags = DCACHE_SPECIAL_TYPE;
1963
1964type_determined:
1965        if (unlikely(IS_AUTOMOUNT(inode)))
1966                add_flags |= DCACHE_NEED_AUTOMOUNT;
1967        return add_flags;
1968}
1969
1970static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1971{
1972        unsigned add_flags = d_flags_for_inode(inode);
1973        WARN_ON(d_in_lookup(dentry));
1974
1975        spin_lock(&dentry->d_lock);
1976        /*
1977         * Decrement negative dentry count if it was in the LRU list.
1978         */
1979        if (dentry->d_flags & DCACHE_LRU_LIST)
1980                this_cpu_dec(nr_dentry_negative);
1981        hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1982        raw_write_seqcount_begin(&dentry->d_seq);
1983        __d_set_inode_and_type(dentry, inode, add_flags);
1984        raw_write_seqcount_end(&dentry->d_seq);
1985        fsnotify_update_flags(dentry);
1986        spin_unlock(&dentry->d_lock);
1987}
1988
1989/**
1990 * d_instantiate - fill in inode information for a dentry
1991 * @entry: dentry to complete
1992 * @inode: inode to attach to this dentry
1993 *
1994 * Fill in inode information in the entry.
1995 *
1996 * This turns negative dentries into productive full members
1997 * of society.
1998 *
1999 * NOTE! This assumes that the inode count has been incremented
2000 * (or otherwise set) by the caller to indicate that it is now
2001 * in use by the dcache.
2002 */
2003 
2004void d_instantiate(struct dentry *entry, struct inode * inode)
2005{
2006        BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
2007        if (inode) {
2008                security_d_instantiate(entry, inode);
2009                spin_lock(&inode->i_lock);
2010                __d_instantiate(entry, inode);
2011                spin_unlock(&inode->i_lock);
2012        }
2013}
2014EXPORT_SYMBOL(d_instantiate);
2015
2016/*
2017 * This should be equivalent to d_instantiate() + unlock_new_inode(),
2018 * with lockdep-related part of unlock_new_inode() done before
2019 * anything else.  Use that instead of open-coding d_instantiate()/
2020 * unlock_new_inode() combinations.
2021 */
2022void d_instantiate_new(struct dentry *entry, struct inode *inode)
2023{
2024        BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
2025        BUG_ON(!inode);
2026        lockdep_annotate_inode_mutex_key(inode);
2027        security_d_instantiate(entry, inode);
2028        spin_lock(&inode->i_lock);
2029        __d_instantiate(entry, inode);
2030        WARN_ON(!(inode->i_state & I_NEW));
2031        inode->i_state &= ~I_NEW & ~I_CREATING;
2032        smp_mb();
2033        wake_up_bit(&inode->i_state, __I_NEW);
2034        spin_unlock(&inode->i_lock);
2035}
2036EXPORT_SYMBOL(d_instantiate_new);
2037
2038struct dentry *d_make_root(struct inode *root_inode)
2039{
2040        struct dentry *res = NULL;
2041
2042        if (root_inode) {
2043                res = d_alloc_anon(root_inode->i_sb);
2044                if (res)
2045                        d_instantiate(res, root_inode);
2046                else
2047                        iput(root_inode);
2048        }
2049        return res;
2050}
2051EXPORT_SYMBOL(d_make_root);
2052
2053static struct dentry *__d_instantiate_anon(struct dentry *dentry,
2054                                           struct inode *inode,
2055                                           bool disconnected)
2056{
2057        struct dentry *res;
2058        unsigned add_flags;
2059
2060        security_d_instantiate(dentry, inode);
2061        spin_lock(&inode->i_lock);
2062        res = __d_find_any_alias(inode);
2063        if (res) {
2064                spin_unlock(&inode->i_lock);
2065                dput(dentry);
2066                goto out_iput;
2067        }
2068
2069        /* attach a disconnected dentry */
2070        add_flags = d_flags_for_inode(inode);
2071
2072        if (disconnected)
2073                add_flags |= DCACHE_DISCONNECTED;
2074
2075        spin_lock(&dentry->d_lock);
2076        __d_set_inode_and_type(dentry, inode, add_flags);
2077        hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2078        if (!disconnected) {
2079                hlist_bl_lock(&dentry->d_sb->s_roots);
2080                hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
2081                hlist_bl_unlock(&dentry->d_sb->s_roots);
2082        }
2083        spin_unlock(&dentry->d_lock);
2084        spin_unlock(&inode->i_lock);
2085
2086        return dentry;
2087
2088 out_iput:
2089        iput(inode);
2090        return res;
2091}
2092
2093struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
2094{
2095        return __d_instantiate_anon(dentry, inode, true);
2096}
2097EXPORT_SYMBOL(d_instantiate_anon);
2098
2099static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
2100{
2101        struct dentry *tmp;
2102        struct dentry *res;
2103
2104        if (!inode)
2105                return ERR_PTR(-ESTALE);
2106        if (IS_ERR(inode))
2107                return ERR_CAST(inode);
2108
2109        res = d_find_any_alias(inode);
2110        if (res)
2111                goto out_iput;
2112
2113        tmp = d_alloc_anon(inode->i_sb);
2114        if (!tmp) {
2115                res = ERR_PTR(-ENOMEM);
2116                goto out_iput;
2117        }
2118
2119        return __d_instantiate_anon(tmp, inode, disconnected);
2120
2121out_iput:
2122        iput(inode);
2123        return res;
2124}
2125
2126/**
2127 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2128 * @inode: inode to allocate the dentry for
2129 *
2130 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2131 * similar open by handle operations.  The returned dentry may be anonymous,
2132 * or may have a full name (if the inode was already in the cache).
2133 *
2134 * When called on a directory inode, we must ensure that the inode only ever
2135 * has one dentry.  If a dentry is found, that is returned instead of
2136 * allocating a new one.
2137 *
2138 * On successful return, the reference to the inode has been transferred
2139 * to the dentry.  In case of an error the reference on the inode is released.
2140 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2141 * be passed in and the error will be propagated to the return value,
2142 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2143 */
2144struct dentry *d_obtain_alias(struct inode *inode)
2145{
2146        return __d_obtain_alias(inode, true);
2147}
2148EXPORT_SYMBOL(d_obtain_alias);
2149
2150/**
2151 * d_obtain_root - find or allocate a dentry for a given inode
2152 * @inode: inode to allocate the dentry for
2153 *
2154 * Obtain an IS_ROOT dentry for the root of a filesystem.
2155 *
2156 * We must ensure that directory inodes only ever have one dentry.  If a
2157 * dentry is found, that is returned instead of allocating a new one.
2158 *
2159 * On successful return, the reference to the inode has been transferred
2160 * to the dentry.  In case of an error the reference on the inode is
2161 * released.  A %NULL or IS_ERR inode may be passed in and will be the
2162 * error will be propagate to the return value, with a %NULL @inode
2163 * replaced by ERR_PTR(-ESTALE).
2164 */
2165struct dentry *d_obtain_root(struct inode *inode)
2166{
2167        return __d_obtain_alias(inode, false);
2168}
2169EXPORT_SYMBOL(d_obtain_root);
2170
2171/**
2172 * d_add_ci - lookup or allocate new dentry with case-exact name
2173 * @inode:  the inode case-insensitive lookup has found
2174 * @dentry: the negative dentry that was passed to the parent's lookup func
2175 * @name:   the case-exact name to be associated with the returned dentry
2176 *
2177 * This is to avoid filling the dcache with case-insensitive names to the
2178 * same inode, only the actual correct case is stored in the dcache for
2179 * case-insensitive filesystems.
2180 *
2181 * For a case-insensitive lookup match and if the case-exact dentry
2182 * already exists in the dcache, use it and return it.
2183 *
2184 * If no entry exists with the exact case name, allocate new dentry with
2185 * the exact case, and return the spliced entry.
2186 */
2187struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2188                        struct qstr *name)
2189{
2190        struct dentry *found, *res;
2191
2192        /*
2193         * First check if a dentry matching the name already exists,
2194         * if not go ahead and create it now.
2195         */
2196        found = d_hash_and_lookup(dentry->d_parent, name);
2197        if (found) {
2198                iput(inode);
2199                return found;
2200        }
2201        if (d_in_lookup(dentry)) {
2202                found = d_alloc_parallel(dentry->d_parent, name,
2203                                        dentry->d_wait);
2204                if (IS_ERR(found) || !d_in_lookup(found)) {
2205                        iput(inode);
2206                        return found;
2207                }
2208        } else {
2209                found = d_alloc(dentry->d_parent, name);
2210                if (!found) {
2211                        iput(inode);
2212                        return ERR_PTR(-ENOMEM);
2213                } 
2214        }
2215        res = d_splice_alias(inode, found);
2216        if (res) {
2217                dput(found);
2218                return res;
2219        }
2220        return found;
2221}
2222EXPORT_SYMBOL(d_add_ci);
2223
2224
2225static inline bool d_same_name(const struct dentry *dentry,
2226                                const struct dentry *parent,
2227                                const struct qstr *name)
2228{
2229        if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2230                if (dentry->d_name.len != name->len)
2231                        return false;
2232                return dentry_cmp(dentry, name->name, name->len) == 0;
2233        }
2234        return parent->d_op->d_compare(dentry,
2235                                       dentry->d_name.len, dentry->d_name.name,
2236                                       name) == 0;
2237}
2238
2239/**
2240 * __d_lookup_rcu - search for a dentry (racy, store-free)
2241 * @parent: parent dentry
2242 * @name: qstr of name we wish to find
2243 * @seqp: returns d_seq value at the point where the dentry was found
2244 * Returns: dentry, or NULL
2245 *
2246 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2247 * resolution (store-free path walking) design described in
2248 * Documentation/filesystems/path-lookup.txt.
2249 *
2250 * This is not to be used outside core vfs.
2251 *
2252 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2253 * held, and rcu_read_lock held. The returned dentry must not be stored into
2254 * without taking d_lock and checking d_seq sequence count against @seq
2255 * returned here.
2256 *
2257 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2258 * function.
2259 *
2260 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2261 * the returned dentry, so long as its parent's seqlock is checked after the
2262 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2263 * is formed, giving integrity down the path walk.
2264 *
2265 * NOTE! The caller *has* to check the resulting dentry against the sequence
2266 * number we've returned before using any of the resulting dentry state!
2267 */
2268struct dentry *__d_lookup_rcu(const struct dentry *parent,
2269                                const struct qstr *name,
2270                                unsigned *seqp)
2271{
2272        u64 hashlen = name->hash_len;
2273        const unsigned char *str = name->name;
2274        struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2275        struct hlist_bl_node *node;
2276        struct dentry *dentry;
2277
2278        /*
2279         * Note: There is significant duplication with __d_lookup_rcu which is
2280         * required to prevent single threaded performance regressions
2281         * especially on architectures where smp_rmb (in seqcounts) are costly.
2282         * Keep the two functions in sync.
2283         */
2284
2285        /*
2286         * The hash list is protected using RCU.
2287         *
2288         * Carefully use d_seq when comparing a candidate dentry, to avoid
2289         * races with d_move().
2290         *
2291         * It is possible that concurrent renames can mess up our list
2292         * walk here and result in missing our dentry, resulting in the
2293         * false-negative result. d_lookup() protects against concurrent
2294         * renames using rename_lock seqlock.
2295         *
2296         * See Documentation/filesystems/path-lookup.txt for more details.
2297         */
2298        hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2299                unsigned seq;
2300
2301seqretry:
2302                /*
2303                 * The dentry sequence count protects us from concurrent
2304                 * renames, and thus protects parent and name fields.
2305                 *
2306                 * The caller must perform a seqcount check in order
2307                 * to do anything useful with the returned dentry.
2308                 *
2309                 * NOTE! We do a "raw" seqcount_begin here. That means that
2310                 * we don't wait for the sequence count to stabilize if it
2311                 * is in the middle of a sequence change. If we do the slow
2312                 * dentry compare, we will do seqretries until it is stable,
2313                 * and if we end up with a successful lookup, we actually
2314                 * want to exit RCU lookup anyway.
2315                 *
2316                 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2317                 * we are still guaranteed NUL-termination of ->d_name.name.
2318                 */
2319                seq = raw_seqcount_begin(&dentry->d_seq);
2320                if (dentry->d_parent != parent)
2321                        continue;
2322                if (d_unhashed(dentry))
2323                        continue;
2324
2325                if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2326                        int tlen;
2327                        const char *tname;
2328                        if (dentry->d_name.hash != hashlen_hash(hashlen))
2329                                continue;
2330                        tlen = dentry->d_name.len;
2331                        tname = dentry->d_name.name;
2332                        /* we want a consistent (name,len) pair */
2333                        if (read_seqcount_retry(&dentry->d_seq, seq)) {
2334                                cpu_relax();
2335                                goto seqretry;
2336                        }
2337                        if (parent->d_op->d_compare(dentry,
2338                                                    tlen, tname, name) != 0)
2339                                continue;
2340                } else {
2341                        if (dentry->d_name.hash_len != hashlen)
2342                                continue;
2343                        if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2344                                continue;
2345                }
2346                *seqp = seq;
2347                return dentry;
2348        }
2349        return NULL;
2350}
2351
2352/**
2353 * d_lookup - search for a dentry
2354 * @parent: parent dentry
2355 * @name: qstr of name we wish to find
2356 * Returns: dentry, or NULL
2357 *
2358 * d_lookup searches the children of the parent dentry for the name in
2359 * question. If the dentry is found its reference count is incremented and the
2360 * dentry is returned. The caller must use dput to free the entry when it has
2361 * finished using it. %NULL is returned if the dentry does not exist.
2362 */
2363struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2364{
2365        struct dentry *dentry;
2366        unsigned seq;
2367
2368        do {
2369                seq = read_seqbegin(&rename_lock);
2370                dentry = __d_lookup(parent, name);
2371                if (dentry)
2372                        break;
2373        } while (read_seqretry(&rename_lock, seq));
2374        return dentry;
2375}
2376EXPORT_SYMBOL(d_lookup);
2377
2378/**
2379 * __d_lookup - search for a dentry (racy)
2380 * @parent: parent dentry
2381 * @name: qstr of name we wish to find
2382 * Returns: dentry, or NULL
2383 *
2384 * __d_lookup is like d_lookup, however it may (rarely) return a
2385 * false-negative result due to unrelated rename activity.
2386 *
2387 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2388 * however it must be used carefully, eg. with a following d_lookup in
2389 * the case of failure.
2390 *
2391 * __d_lookup callers must be commented.
2392 */
2393struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2394{
2395        unsigned int hash = name->hash;
2396        struct hlist_bl_head *b = d_hash(hash);
2397        struct hlist_bl_node *node;
2398        struct dentry *found = NULL;
2399        struct dentry *dentry;
2400
2401        /*
2402         * Note: There is significant duplication with __d_lookup_rcu which is
2403         * required to prevent single threaded performance regressions
2404         * especially on architectures where smp_rmb (in seqcounts) are costly.
2405         * Keep the two functions in sync.
2406         */
2407
2408        /*
2409         * The hash list is protected using RCU.
2410         *
2411         * Take d_lock when comparing a candidate dentry, to avoid races
2412         * with d_move().
2413         *
2414         * It is possible that concurrent renames can mess up our list
2415         * walk here and result in missing our dentry, resulting in the
2416         * false-negative result. d_lookup() protects against concurrent
2417         * renames using rename_lock seqlock.
2418         *
2419         * See Documentation/filesystems/path-lookup.txt for more details.
2420         */
2421        rcu_read_lock();
2422        
2423        hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2424
2425                if (dentry->d_name.hash != hash)
2426                        continue;
2427
2428                spin_lock(&dentry->d_lock);
2429                if (dentry->d_parent != parent)
2430                        goto next;
2431                if (d_unhashed(dentry))
2432                        goto next;
2433
2434                if (!d_same_name(dentry, parent, name))
2435                        goto next;
2436
2437                dentry->d_lockref.count++;
2438                found = dentry;
2439                spin_unlock(&dentry->d_lock);
2440                break;
2441next:
2442                spin_unlock(&dentry->d_lock);
2443        }
2444        rcu_read_unlock();
2445
2446        return found;
2447}
2448
2449/**
2450 * d_hash_and_lookup - hash the qstr then search for a dentry
2451 * @dir: Directory to search in
2452 * @name: qstr of name we wish to find
2453 *
2454 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2455 */
2456struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2457{
2458        /*
2459         * Check for a fs-specific hash function. Note that we must
2460         * calculate the standard hash first, as the d_op->d_hash()
2461         * routine may choose to leave the hash value unchanged.
2462         */
2463        name->hash = full_name_hash(dir, name->name, name->len);
2464        if (dir->d_flags & DCACHE_OP_HASH) {
2465                int err = dir->d_op->d_hash(dir, name);
2466                if (unlikely(err < 0))
2467                        return ERR_PTR(err);
2468        }
2469        return d_lookup(dir, name);
2470}
2471EXPORT_SYMBOL(d_hash_and_lookup);
2472
2473/*
2474 * When a file is deleted, we have two options:
2475 * - turn this dentry into a negative dentry
2476 * - unhash this dentry and free it.
2477 *
2478 * Usually, we want to just turn this into
2479 * a negative dentry, but if anybody else is
2480 * currently using the dentry or the inode
2481 * we can't do that and we fall back on removing
2482 * it from the hash queues and waiting for
2483 * it to be deleted later when it has no users
2484 */
2485 
2486/**
2487 * d_delete - delete a dentry
2488 * @dentry: The dentry to delete
2489 *
2490 * Turn the dentry into a negative dentry if possible, otherwise
2491 * remove it from the hash queues so it can be deleted later
2492 */
2493 
2494void d_delete(struct dentry * dentry)
2495{
2496        struct inode *inode = dentry->d_inode;
2497
2498        spin_lock(&inode->i_lock);
2499        spin_lock(&dentry->d_lock);
2500        /*
2501         * Are we the only user?
2502         */
2503        if (dentry->d_lockref.count == 1) {
2504                dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2505                dentry_unlink_inode(dentry);
2506        } else {
2507                __d_drop(dentry);
2508                spin_unlock(&dentry->d_lock);
2509                spin_unlock(&inode->i_lock);
2510        }
2511}
2512EXPORT_SYMBOL(d_delete);
2513
2514static void __d_rehash(struct dentry *entry)
2515{
2516        struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2517
2518        hlist_bl_lock(b);
2519        hlist_bl_add_head_rcu(&entry->d_hash, b);
2520        hlist_bl_unlock(b);
2521}
2522
2523/**
2524 * d_rehash     - add an entry back to the hash
2525 * @entry: dentry to add to the hash
2526 *
2527 * Adds a dentry to the hash according to its name.
2528 */
2529 
2530void d_rehash(struct dentry * entry)
2531{
2532        spin_lock(&entry->d_lock);
2533        __d_rehash(entry);
2534        spin_unlock(&entry->d_lock);
2535}
2536EXPORT_SYMBOL(d_rehash);
2537
2538static inline unsigned start_dir_add(struct inode *dir)
2539{
2540
2541        for (;;) {
2542                unsigned n = dir->i_dir_seq;
2543                if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2544                        return n;
2545                cpu_relax();
2546        }
2547}
2548
2549static inline void end_dir_add(struct inode *dir, unsigned n)
2550{
2551        smp_store_release(&dir->i_dir_seq, n + 2);
2552}
2553
2554static void d_wait_lookup(struct dentry *dentry)
2555{
2556        if (d_in_lookup(dentry)) {
2557                DECLARE_WAITQUEUE(wait, current);
2558                add_wait_queue(dentry->d_wait, &wait);
2559                do {
2560                        set_current_state(TASK_UNINTERRUPTIBLE);
2561                        spin_unlock(&dentry->d_lock);
2562                        schedule();
2563                        spin_lock(&dentry->d_lock);
2564                } while (d_in_lookup(dentry));
2565        }
2566}
2567
2568struct dentry *d_alloc_parallel(struct dentry *parent,
2569                                const struct qstr *name,
2570                                wait_queue_head_t *wq)
2571{
2572        unsigned int hash = name->hash;
2573        struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2574        struct hlist_bl_node *node;
2575        struct dentry *new = d_alloc(parent, name);
2576        struct dentry *dentry;
2577        unsigned seq, r_seq, d_seq;
2578
2579        if (unlikely(!new))
2580                return ERR_PTR(-ENOMEM);
2581
2582retry:
2583        rcu_read_lock();
2584        seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2585        r_seq = read_seqbegin(&rename_lock);
2586        dentry = __d_lookup_rcu(parent, name, &d_seq);
2587        if (unlikely(dentry)) {
2588                if (!lockref_get_not_dead(&dentry->d_lockref)) {
2589                        rcu_read_unlock();
2590                        goto retry;
2591                }
2592                if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2593                        rcu_read_unlock();
2594                        dput(dentry);
2595                        goto retry;
2596                }
2597                rcu_read_unlock();
2598                dput(new);
2599                return dentry;
2600        }
2601        if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2602                rcu_read_unlock();
2603                goto retry;
2604        }
2605
2606        if (unlikely(seq & 1)) {
2607                rcu_read_unlock();
2608                goto retry;
2609        }
2610
2611        hlist_bl_lock(b);
2612        if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2613                hlist_bl_unlock(b);
2614                rcu_read_unlock();
2615                goto retry;
2616        }
2617        /*
2618         * No changes for the parent since the beginning of d_lookup().
2619         * Since all removals from the chain happen with hlist_bl_lock(),
2620         * any potential in-lookup matches are going to stay here until
2621         * we unlock the chain.  All fields are stable in everything
2622         * we encounter.
2623         */
2624        hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2625                if (dentry->d_name.hash != hash)
2626                        continue;
2627                if (dentry->d_parent != parent)
2628                        continue;
2629                if (!d_same_name(dentry, parent, name))
2630                        continue;
2631                hlist_bl_unlock(b);
2632                /* now we can try to grab a reference */
2633                if (!lockref_get_not_dead(&dentry->d_lockref)) {
2634                        rcu_read_unlock();
2635                        goto retry;
2636                }
2637
2638                rcu_read_unlock();
2639                /*
2640                 * somebody is likely to be still doing lookup for it;
2641                 * wait for them to finish
2642                 */
2643                spin_lock(&dentry->d_lock);
2644                d_wait_lookup(dentry);
2645                /*
2646                 * it's not in-lookup anymore; in principle we should repeat
2647                 * everything from dcache lookup, but it's likely to be what
2648                 * d_lookup() would've found anyway.  If it is, just return it;
2649                 * otherwise we really have to repeat the whole thing.
2650                 */
2651                if (unlikely(dentry->d_name.hash != hash))
2652                        goto mismatch;
2653                if (unlikely(dentry->d_parent != parent))
2654                        goto mismatch;
2655                if (unlikely(d_unhashed(dentry)))
2656                        goto mismatch;
2657                if (unlikely(!d_same_name(dentry, parent, name)))
2658                        goto mismatch;
2659                /* OK, it *is* a hashed match; return it */
2660                spin_unlock(&dentry->d_lock);
2661                dput(new);
2662                return dentry;
2663        }
2664        rcu_read_unlock();
2665        /* we can't take ->d_lock here; it's OK, though. */
2666        new->d_flags |= DCACHE_PAR_LOOKUP;
2667        new->d_wait = wq;
2668        hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2669        hlist_bl_unlock(b);
2670        return new;
2671mismatch:
2672        spin_unlock(&dentry->d_lock);
2673        dput(dentry);
2674        goto retry;
2675}
2676EXPORT_SYMBOL(d_alloc_parallel);
2677
2678void __d_lookup_done(struct dentry *dentry)
2679{
2680        struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2681                                                 dentry->d_name.hash);
2682        hlist_bl_lock(b);
2683        dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2684        __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2685        wake_up_all(dentry->d_wait);
2686        dentry->d_wait = NULL;
2687        hlist_bl_unlock(b);
2688        INIT_HLIST_NODE(&dentry->d_u.d_alias);
2689        INIT_LIST_HEAD(&dentry->d_lru);
2690}
2691EXPORT_SYMBOL(__d_lookup_done);
2692
2693/* inode->i_lock held if inode is non-NULL */
2694
2695static inline void __d_add(struct dentry *dentry, struct inode *inode)
2696{
2697        struct inode *dir = NULL;
2698        unsigned n;
2699        spin_lock(&dentry->d_lock);
2700        if (unlikely(d_in_lookup(dentry))) {
2701                dir = dentry->d_parent->d_inode;
2702                n = start_dir_add(dir);
2703                __d_lookup_done(dentry);
2704        }
2705        if (inode) {
2706                unsigned add_flags = d_flags_for_inode(inode);
2707                hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2708                raw_write_seqcount_begin(&dentry->d_seq);
2709                __d_set_inode_and_type(dentry, inode, add_flags);
2710                raw_write_seqcount_end(&dentry->d_seq);
2711                fsnotify_update_flags(dentry);
2712        }
2713        __d_rehash(dentry);
2714        if (dir)
2715                end_dir_add(dir, n);
2716        spin_unlock(&dentry->d_lock);
2717        if (inode)
2718                spin_unlock(&inode->i_lock);
2719}
2720
2721/**
2722 * d_add - add dentry to hash queues
2723 * @entry: dentry to add
2724 * @inode: The inode to attach to this dentry
2725 *
2726 * This adds the entry to the hash queues and initializes @inode.
2727 * The entry was actually filled in earlier during d_alloc().
2728 */
2729
2730void d_add(struct dentry *entry, struct inode *inode)
2731{
2732        if (inode) {
2733                security_d_instantiate(entry, inode);
2734                spin_lock(&inode->i_lock);
2735        }
2736        __d_add(entry, inode);
2737}
2738EXPORT_SYMBOL(d_add);
2739
2740/**
2741 * d_exact_alias - find and hash an exact unhashed alias
2742 * @entry: dentry to add
2743 * @inode: The inode to go with this dentry
2744 *
2745 * If an unhashed dentry with the same name/parent and desired
2746 * inode already exists, hash and return it.  Otherwise, return
2747 * NULL.
2748 *
2749 * Parent directory should be locked.
2750 */
2751struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2752{
2753        struct dentry *alias;
2754        unsigned int hash = entry->d_name.hash;
2755
2756        spin_lock(&inode->i_lock);
2757        hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2758                /*
2759                 * Don't need alias->d_lock here, because aliases with
2760                 * d_parent == entry->d_parent are not subject to name or
2761                 * parent changes, because the parent inode i_mutex is held.
2762                 */
2763                if (alias->d_name.hash != hash)
2764                        continue;
2765                if (alias->d_parent != entry->d_parent)
2766                        continue;
2767                if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2768                        continue;
2769                spin_lock(&alias->d_lock);
2770                if (!d_unhashed(alias)) {
2771                        spin_unlock(&alias->d_lock);
2772                        alias = NULL;
2773                } else {
2774                        __dget_dlock(alias);
2775                        __d_rehash(alias);
2776                        spin_unlock(&alias->d_lock);
2777                }
2778                spin_unlock(&inode->i_lock);
2779                return alias;
2780        }
2781        spin_unlock(&inode->i_lock);
2782        return NULL;
2783}
2784EXPORT_SYMBOL(d_exact_alias);
2785
2786static void swap_names(struct dentry *dentry, struct dentry *target)
2787{
2788        if (unlikely(dname_external(target))) {
2789                if (unlikely(dname_external(dentry))) {
2790                        /*
2791                         * Both external: swap the pointers
2792                         */
2793                        swap(target->d_name.name, dentry->d_name.name);
2794                } else {
2795                        /*
2796                         * dentry:internal, target:external.  Steal target's
2797                         * storage and make target internal.
2798                         */
2799                        memcpy(target->d_iname, dentry->d_name.name,
2800                                        dentry->d_name.len + 1);
2801                        dentry->d_name.name = target->d_name.name;
2802                        target->d_name.name = target->d_iname;
2803                }
2804        } else {
2805                if (unlikely(dname_external(dentry))) {
2806                        /*
2807                         * dentry:external, target:internal.  Give dentry's
2808                         * storage to target and make dentry internal
2809                         */
2810                        memcpy(dentry->d_iname, target->d_name.name,
2811                                        target->d_name.len + 1);
2812                        target->d_name.name = dentry->d_name.name;
2813                        dentry->d_name.name = dentry->d_iname;
2814                } else {
2815                        /*
2816                         * Both are internal.
2817                         */
2818                        unsigned int i;
2819                        BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2820                        for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2821                                swap(((long *) &dentry->d_iname)[i],
2822                                     ((long *) &target->d_iname)[i]);
2823                        }
2824                }
2825        }
2826        swap(dentry->d_name.hash_len, target->d_name.hash_len);
2827}
2828
2829static void copy_name(struct dentry *dentry, struct dentry *target)
2830{
2831        struct external_name *old_name = NULL;
2832        if (unlikely(dname_external(dentry)))
2833                old_name = external_name(dentry);
2834        if (unlikely(dname_external(target))) {
2835                atomic_inc(&external_name(target)->u.count);
2836                dentry->d_name = target->d_name;
2837        } else {
2838                memcpy(dentry->d_iname, target->d_name.name,
2839                                target->d_name.len + 1);
2840                dentry->d_name.name = dentry->d_iname;
2841                dentry->d_name.hash_len = target->d_name.hash_len;
2842        }
2843        if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2844                kfree_rcu(old_name, u.head);
2845}
2846
2847/*
2848 * __d_move - move a dentry
2849 * @dentry: entry to move
2850 * @target: new dentry
2851 * @exchange: exchange the two dentries
2852 *
2853 * Update the dcache to reflect the move of a file name. Negative
2854 * dcache entries should not be moved in this way. Caller must hold
2855 * rename_lock, the i_mutex of the source and target directories,
2856 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2857 */
2858static void __d_move(struct dentry *dentry, struct dentry *target,
2859                     bool exchange)
2860{
2861        struct dentry *old_parent, *p;
2862        struct inode *dir = NULL;
2863        unsigned n;
2864
2865        WARN_ON(!dentry->d_inode);
2866        if (WARN_ON(dentry == target))
2867                return;
2868
2869        BUG_ON(d_ancestor(target, dentry));
2870        old_parent = dentry->d_parent;
2871        p = d_ancestor(old_parent, target);
2872        if (IS_ROOT(dentry)) {
2873                BUG_ON(p);
2874                spin_lock(&target->d_parent->d_lock);
2875        } else if (!p) {
2876                /* target is not a descendent of dentry->d_parent */
2877                spin_lock(&target->d_parent->d_lock);
2878                spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2879        } else {
2880                BUG_ON(p == dentry);
2881                spin_lock(&old_parent->d_lock);
2882                if (p != target)
2883                        spin_lock_nested(&target->d_parent->d_lock,
2884                                        DENTRY_D_LOCK_NESTED);
2885        }
2886        spin_lock_nested(&dentry->d_lock, 2);
2887        spin_lock_nested(&target->d_lock, 3);
2888
2889        if (unlikely(d_in_lookup(target))) {
2890                dir = target->d_parent->d_inode;
2891                n = start_dir_add(dir);
2892                __d_lookup_done(target);
2893        }
2894
2895        write_seqcount_begin(&dentry->d_seq);
2896        write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2897
2898        /* unhash both */
2899        if (!d_unhashed(dentry))
2900                ___d_drop(dentry);
2901        if (!d_unhashed(target))
2902                ___d_drop(target);
2903
2904        /* ... and switch them in the tree */
2905        dentry->d_parent = target->d_parent;
2906        if (!exchange) {
2907                copy_name(dentry, target);
2908                target->d_hash.pprev = NULL;
2909                dentry->d_parent->d_lockref.count++;
2910                if (dentry != old_parent) /* wasn't IS_ROOT */
2911                        WARN_ON(!--old_parent->d_lockref.count);
2912        } else {
2913                target->d_parent = old_parent;
2914                swap_names(dentry, target);
2915                list_move(&target->d_child, &target->d_parent->d_subdirs);
2916                __d_rehash(target);
2917                fsnotify_update_flags(target);
2918        }
2919        list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2920        __d_rehash(dentry);
2921        fsnotify_update_flags(dentry);
2922        fscrypt_handle_d_move(dentry);
2923
2924        write_seqcount_end(&target->d_seq);
2925        write_seqcount_end(&dentry->d_seq);
2926
2927        if (dir)
2928                end_dir_add(dir, n);
2929
2930        if (dentry->d_parent != old_parent)
2931                spin_unlock(&dentry->d_parent->d_lock);
2932        if (dentry != old_parent)
2933                spin_unlock(&old_parent->d_lock);
2934        spin_unlock(&target->d_lock);
2935        spin_unlock(&dentry->d_lock);
2936}
2937
2938/*
2939 * d_move - move a dentry
2940 * @dentry: entry to move
2941 * @target: new dentry
2942 *
2943 * Update the dcache to reflect the move of a file name. Negative
2944 * dcache entries should not be moved in this way. See the locking
2945 * requirements for __d_move.
2946 */
2947void d_move(struct dentry *dentry, struct dentry *target)
2948{
2949        write_seqlock(&rename_lock);
2950        __d_move(dentry, target, false);
2951        write_sequnlock(&rename_lock);
2952}
2953EXPORT_SYMBOL(d_move);
2954
2955/*
2956 * d_exchange - exchange two dentries
2957 * @dentry1: first dentry
2958 * @dentry2: second dentry
2959 */
2960void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2961{
2962        write_seqlock(&rename_lock);
2963
2964        WARN_ON(!dentry1->d_inode);
2965        WARN_ON(!dentry2->d_inode);
2966        WARN_ON(IS_ROOT(dentry1));
2967        WARN_ON(IS_ROOT(dentry2));
2968
2969        __d_move(dentry1, dentry2, true);
2970
2971        write_sequnlock(&rename_lock);
2972}
2973
2974/**
2975 * d_ancestor - search for an ancestor
2976 * @p1: ancestor dentry
2977 * @p2: child dentry
2978 *
2979 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2980 * an ancestor of p2, else NULL.
2981 */
2982struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2983{
2984        struct dentry *p;
2985
2986        for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2987                if (p->d_parent == p1)
2988                        return p;
2989        }
2990        return NULL;
2991}
2992
2993/*
2994 * This helper attempts to cope with remotely renamed directories
2995 *
2996 * It assumes that the caller is already holding
2997 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2998 *
2999 * Note: If ever the locking in lock_rename() changes, then please
3000 * remember to update this too...
3001 */
3002static int __d_unalias(struct inode *inode,
3003                struct dentry *dentry, struct dentry *alias)
3004{
3005        struct mutex *m1 = NULL;
3006        struct rw_semaphore *m2 = NULL;
3007        int ret = -ESTALE;
3008
3009        /* If alias and dentry share a parent, then no extra locks required */
3010        if (alias->d_parent == dentry->d_parent)
3011                goto out_unalias;
3012
3013        /* See lock_rename() */
3014        if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
3015                goto out_err;
3016        m1 = &dentry->d_sb->s_vfs_rename_mutex;
3017        if (!inode_trylock_shared(alias->d_parent->d_inode))
3018                goto out_err;
3019        m2 = &alias->d_parent->d_inode->i_rwsem;
3020out_unalias:
3021        __d_move(alias, dentry, false);
3022        ret = 0;
3023out_err:
3024        if (m2)
3025                up_read(m2);
3026        if (m1)
3027                mutex_unlock(m1);
3028        return ret;
3029}
3030
3031/**
3032 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3033 * @inode:  the inode which may have a disconnected dentry
3034 * @dentry: a negative dentry which we want to point to the inode.
3035 *
3036 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3037 * place of the given dentry and return it, else simply d_add the inode
3038 * to the dentry and return NULL.
3039 *
3040 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3041 * we should error out: directories can't have multiple aliases.
3042 *
3043 * This is needed in the lookup routine of any filesystem that is exportable
3044 * (via knfsd) so that we can build dcache paths to directories effectively.
3045 *
3046 * If a dentry was found and moved, then it is returned.  Otherwise NULL
3047 * is returned.  This matches the expected return value of ->lookup.
3048 *
3049 * Cluster filesystems may call this function with a negative, hashed dentry.
3050 * In that case, we know that the inode will be a regular file, and also this
3051 * will only occur during atomic_open. So we need to check for the dentry
3052 * being already hashed only in the final case.
3053 */
3054struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3055{
3056        if (IS_ERR(inode))
3057                return ERR_CAST(inode);
3058
3059        BUG_ON(!d_unhashed(dentry));
3060
3061        if (!inode)
3062                goto out;
3063
3064        security_d_instantiate(dentry, inode);
3065        spin_lock(&inode->i_lock);
3066        if (S_ISDIR(inode->i_mode)) {
3067                struct dentry *new = __d_find_any_alias(inode);
3068                if (unlikely(new)) {
3069                        /* The reference to new ensures it remains an alias */
3070                        spin_unlock(&inode->i_lock);
3071                        write_seqlock(&rename_lock);
3072                        if (unlikely(d_ancestor(new, dentry))) {
3073                                write_sequnlock(&rename_lock);
3074                                dput(new);
3075                                new = ERR_PTR(-ELOOP);
3076                                pr_warn_ratelimited(
3077                                        "VFS: Lookup of '%s' in %s %s"
3078                                        " would have caused loop\n",
3079                                        dentry->d_name.name,
3080                                        inode->i_sb->s_type->name,
3081                                        inode->i_sb->s_id);
3082                        } else if (!IS_ROOT(new)) {
3083                                struct dentry *old_parent = dget(new->d_parent);
3084                                int err = __d_unalias(inode, dentry, new);
3085                                write_sequnlock(&rename_lock);
3086                                if (err) {
3087                                        dput(new);
3088                                        new = ERR_PTR(err);
3089                                }
3090                                dput(old_parent);
3091                        } else {
3092                                __d_move(new, dentry, false);
3093                                write_sequnlock(&rename_lock);
3094                        }
3095                        iput(inode);
3096                        return new;
3097                }
3098        }
3099out:
3100        __d_add(dentry, inode);
3101        return NULL;
3102}
3103EXPORT_SYMBOL(d_splice_alias);
3104
3105/*
3106 * Test whether new_dentry is a subdirectory of old_dentry.
3107 *
3108 * Trivially implemented using the dcache structure
3109 */
3110
3111/**
3112 * is_subdir - is new dentry a subdirectory of old_dentry
3113 * @new_dentry: new dentry
3114 * @old_dentry: old dentry
3115 *
3116 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3117 * Returns false otherwise.
3118 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3119 */
3120  
3121bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3122{
3123        bool result;
3124        unsigned seq;
3125
3126        if (new_dentry == old_dentry)
3127                return true;
3128
3129        do {
3130                /* for restarting inner loop in case of seq retry */
3131                seq = read_seqbegin(&rename_lock);
3132                /*
3133                 * Need rcu_readlock to protect against the d_parent trashing
3134                 * due to d_move
3135                 */
3136                rcu_read_lock();
3137                if (d_ancestor(old_dentry, new_dentry))
3138                        result = true;
3139                else
3140                        result = false;
3141                rcu_read_unlock();
3142        } while (read_seqretry(&rename_lock, seq));
3143
3144        return result;
3145}
3146EXPORT_SYMBOL(is_subdir);
3147
3148static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3149{
3150        struct dentry *root = data;
3151        if (dentry != root) {
3152                if (d_unhashed(dentry) || !dentry->d_inode)
3153                        return D_WALK_SKIP;
3154
3155                if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3156                        dentry->d_flags |= DCACHE_GENOCIDE;
3157                        dentry->d_lockref.count--;
3158                }
3159        }
3160        return D_WALK_CONTINUE;
3161}
3162
3163void d_genocide(struct dentry *parent)
3164{
3165        d_walk(parent, parent, d_genocide_kill);
3166}
3167
3168EXPORT_SYMBOL(d_genocide);
3169
3170void d_tmpfile(struct dentry *dentry, struct inode *inode)
3171{
3172        inode_dec_link_count(inode);
3173        BUG_ON(dentry->d_name.name != dentry->d_iname ||
3174                !hlist_unhashed(&dentry->d_u.d_alias) ||
3175                !d_unlinked(dentry));
3176        spin_lock(&dentry->d_parent->d_lock);
3177        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3178        dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3179                                (unsigned long long)inode->i_ino);
3180        spin_unlock(&dentry->d_lock);
3181        spin_unlock(&dentry->d_parent->d_lock);
3182        d_instantiate(dentry, inode);
3183}
3184EXPORT_SYMBOL(d_tmpfile);
3185
3186static __initdata unsigned long dhash_entries;
3187static int __init set_dhash_entries(char *str)
3188{
3189        if (!str)
3190                return 0;
3191        dhash_entries = simple_strtoul(str, &str, 0);
3192        return 1;
3193}
3194__setup("dhash_entries=", set_dhash_entries);
3195
3196static void __init dcache_init_early(void)
3197{
3198        /* If hashes are distributed across NUMA nodes, defer
3199         * hash allocation until vmalloc space is available.
3200         */
3201        if (hashdist)
3202                return;
3203
3204        dentry_hashtable =
3205                alloc_large_system_hash("Dentry cache",
3206                                        sizeof(struct hlist_bl_head),
3207                                        dhash_entries,
3208                                        13,
3209                                        HASH_EARLY | HASH_ZERO,
3210                                        &d_hash_shift,
3211                                        NULL,
3212                                        0,
3213                                        0);
3214        d_hash_shift = 32 - d_hash_shift;
3215}
3216
3217static void __init dcache_init(void)
3218{
3219        /*
3220         * A constructor could be added for stable state like the lists,
3221         * but it is probably not worth it because of the cache nature
3222         * of the dcache.
3223         */
3224        dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3225                SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3226                d_iname);
3227
3228        /* Hash may have been set up in dcache_init_early */
3229        if (!hashdist)
3230                return;
3231
3232        dentry_hashtable =
3233                alloc_large_system_hash("Dentry cache",
3234                                        sizeof(struct hlist_bl_head),
3235                                        dhash_entries,
3236                                        13,
3237                                        HASH_ZERO,
3238                                        &d_hash_shift,
3239                                        NULL,
3240                                        0,
3241                                        0);
3242        d_hash_shift = 32 - d_hash_shift;
3243}
3244
3245/* SLAB cache for __getname() consumers */
3246struct kmem_cache *names_cachep __read_mostly;
3247EXPORT_SYMBOL(names_cachep);
3248
3249void __init vfs_caches_init_early(void)
3250{
3251        int i;
3252
3253        for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3254                INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3255
3256        dcache_init_early();
3257        inode_init_early();
3258}
3259
3260void __init vfs_caches_init(void)
3261{
3262        names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3263                        SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3264
3265        dcache_init();
3266        inode_init();
3267        files_init();
3268        files_maxfiles_init();
3269        mnt_init();
3270        bdev_cache_init();
3271        chrdev_init();
3272}
3273
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.