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