linux-old/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/config.h>
  18#include <linux/string.h>
  19#include <linux/mm.h>
  20#include <linux/fs.h>
  21#include <linux/slab.h>
  22#include <linux/init.h>
  23#include <linux/smp_lock.h>
  24#include <linux/cache.h>
  25#include <linux/module.h>
  26
  27#include <asm/uaccess.h>
  28
  29#define DCACHE_PARANOIA 1
  30/* #define DCACHE_DEBUG 1 */
  31
  32spinlock_t dcache_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
  33
  34/* Right now the dcache depends on the kernel lock */
  35#define check_lock()    if (!kernel_locked()) BUG()
  36
  37static kmem_cache_t *dentry_cache; 
  38
  39/*
  40 * This is the single most critical data structure when it comes
  41 * to the dcache: the hashtable for lookups. Somebody should try
  42 * to make this good - I've just made it work.
  43 *
  44 * This hash-function tries to avoid losing too many bits of hash
  45 * information, yet avoid using a prime hash-size or similar.
  46 */
  47#define D_HASHBITS     d_hash_shift
  48#define D_HASHMASK     d_hash_mask
  49
  50static unsigned int d_hash_mask;
  51static unsigned int d_hash_shift;
  52static struct list_head *dentry_hashtable;
  53static LIST_HEAD(dentry_unused);
  54
  55/* Statistics gathering. */
  56struct dentry_stat_t dentry_stat = {0, 0, 45, 0,};
  57
  58/*
  59 * no dcache_lock, please.  The caller must decrement dentry_stat.nr_dentry
  60 * inside dcache_lock.
  61 */
  62static inline void d_free(struct dentry *dentry)
  63{
  64        if (dentry->d_op && dentry->d_op->d_release)
  65                dentry->d_op->d_release(dentry);
  66        if (dname_external(dentry)) 
  67                kfree(dentry->d_name.name);
  68        kmem_cache_free(dentry_cache, dentry); 
  69}
  70
  71/*
  72 * Release the dentry's inode, using the filesystem
  73 * d_iput() operation if defined.
  74 * Called with dcache_lock held, drops it.
  75 */
  76static inline void dentry_iput(struct dentry * dentry)
  77{
  78        struct inode *inode = dentry->d_inode;
  79        if (inode) {
  80                dentry->d_inode = NULL;
  81                list_del_init(&dentry->d_alias);
  82                spin_unlock(&dcache_lock);
  83                if (dentry->d_op && dentry->d_op->d_iput)
  84                        dentry->d_op->d_iput(dentry, inode);
  85                else
  86                        iput(inode);
  87        } else
  88                spin_unlock(&dcache_lock);
  89}
  90
  91/* 
  92 * This is dput
  93 *
  94 * This is complicated by the fact that we do not want to put
  95 * dentries that are no longer on any hash chain on the unused
  96 * list: we'd much rather just get rid of them immediately.
  97 *
  98 * However, that implies that we have to traverse the dentry
  99 * tree upwards to the parents which might _also_ now be
 100 * scheduled for deletion (it may have been only waiting for
 101 * its last child to go away).
 102 *
 103 * This tail recursion is done by hand as we don't want to depend
 104 * on the compiler to always get this right (gcc generally doesn't).
 105 * Real recursion would eat up our stack space.
 106 */
 107
 108/*
 109 * dput - release a dentry
 110 * @dentry: dentry to release 
 111 *
 112 * Release a dentry. This will drop the usage count and if appropriate
 113 * call the dentry unlink method as well as removing it from the queues and
 114 * releasing its resources. If the parent dentries were scheduled for release
 115 * they too may now get deleted.
 116 *
 117 * no dcache lock, please.
 118 */
 119
 120void dput(struct dentry *dentry)
 121{
 122        if (!dentry)
 123                return;
 124
 125repeat:
 126        if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
 127                return;
 128
 129        /* dput on a free dentry? */
 130        if (!list_empty(&dentry->d_lru))
 131                BUG();
 132        /*
 133         * AV: ->d_delete() is _NOT_ allowed to block now.
 134         */
 135        if (dentry->d_op && dentry->d_op->d_delete) {
 136                if (dentry->d_op->d_delete(dentry))
 137                        goto unhash_it;
 138        }
 139        /* Unreachable? Get rid of it */
 140        if (list_empty(&dentry->d_hash))
 141                goto kill_it;
 142        list_add(&dentry->d_lru, &dentry_unused);
 143        dentry_stat.nr_unused++;
 144        spin_unlock(&dcache_lock);
 145        return;
 146
 147unhash_it:
 148        list_del_init(&dentry->d_hash);
 149
 150kill_it: {
 151                struct dentry *parent;
 152                list_del(&dentry->d_child);
 153                dentry_stat.nr_dentry--;        /* For d_free, below */
 154                /* drops the lock, at that point nobody can reach this dentry */
 155                dentry_iput(dentry);
 156                parent = dentry->d_parent;
 157                d_free(dentry);
 158                if (dentry == parent)
 159                        return;
 160                dentry = parent;
 161                goto repeat;
 162        }
 163}
 164
 165/**
 166 * d_invalidate - invalidate a dentry
 167 * @dentry: dentry to invalidate
 168 *
 169 * Try to invalidate the dentry if it turns out to be
 170 * possible. If there are other dentries that can be
 171 * reached through this one we can't delete it and we
 172 * return -EBUSY. On success we return 0.
 173 *
 174 * no dcache lock.
 175 */
 176 
 177int d_invalidate(struct dentry * dentry)
 178{
 179        /*
 180         * If it's already been dropped, return OK.
 181         */
 182        spin_lock(&dcache_lock);
 183        if (list_empty(&dentry->d_hash)) {
 184                spin_unlock(&dcache_lock);
 185                return 0;
 186        }
 187        /*
 188         * Check whether to do a partial shrink_dcache
 189         * to get rid of unused child entries.
 190         */
 191        if (!list_empty(&dentry->d_subdirs)) {
 192                spin_unlock(&dcache_lock);
 193                shrink_dcache_parent(dentry);
 194                spin_lock(&dcache_lock);
 195        }
 196
 197        /*
 198         * Somebody else still using it?
 199         *
 200         * If it's a directory, we can't drop it
 201         * for fear of somebody re-populating it
 202         * with children (even though dropping it
 203         * would make it unreachable from the root,
 204         * we might still populate it if it was a
 205         * working directory or similar).
 206         */
 207        if (atomic_read(&dentry->d_count) > 1) {
 208                if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 209                        spin_unlock(&dcache_lock);
 210                        return -EBUSY;
 211                }
 212        }
 213
 214        list_del_init(&dentry->d_hash);
 215        spin_unlock(&dcache_lock);
 216        return 0;
 217}
 218
 219/* This should be called _only_ with dcache_lock held */
 220
 221static inline struct dentry * __dget_locked(struct dentry *dentry)
 222{
 223        atomic_inc(&dentry->d_count);
 224        if (atomic_read(&dentry->d_count) == 1) {
 225                dentry_stat.nr_unused--;
 226                list_del_init(&dentry->d_lru);
 227        }
 228        return dentry;
 229}
 230
 231struct dentry * dget_locked(struct dentry *dentry)
 232{
 233        return __dget_locked(dentry);
 234}
 235
 236/**
 237 * d_find_alias - grab a hashed alias of inode
 238 * @inode: inode in question
 239 *
 240 * If inode has a hashed alias - acquire the reference to alias and
 241 * return it. Otherwise return NULL. Notice that if inode is a directory
 242 * there can be only one alias and it can be unhashed only if it has
 243 * no children.
 244 */
 245
 246struct dentry * d_find_alias(struct inode *inode)
 247{
 248        struct list_head *head, *next, *tmp;
 249        struct dentry *alias;
 250
 251        spin_lock(&dcache_lock);
 252        head = &inode->i_dentry;
 253        next = inode->i_dentry.next;
 254        while (next != head) {
 255                tmp = next;
 256                next = tmp->next;
 257                alias = list_entry(tmp, struct dentry, d_alias);
 258                if (!list_empty(&alias->d_hash)) {
 259                        __dget_locked(alias);
 260                        spin_unlock(&dcache_lock);
 261                        return alias;
 262                }
 263        }
 264        spin_unlock(&dcache_lock);
 265        return NULL;
 266}
 267
 268/*
 269 *      Try to kill dentries associated with this inode.
 270 * WARNING: you must own a reference to inode.
 271 */
 272void d_prune_aliases(struct inode *inode)
 273{
 274        struct list_head *tmp, *head = &inode->i_dentry;
 275restart:
 276        spin_lock(&dcache_lock);
 277        tmp = head;
 278        while ((tmp = tmp->next) != head) {
 279                struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
 280                if (!atomic_read(&dentry->d_count)) {
 281                        __dget_locked(dentry);
 282                        spin_unlock(&dcache_lock);
 283                        d_drop(dentry);
 284                        dput(dentry);
 285                        goto restart;
 286                }
 287        }
 288        spin_unlock(&dcache_lock);
 289}
 290
 291/*
 292 * Throw away a dentry - free the inode, dput the parent.
 293 * This requires that the LRU list has already been
 294 * removed.
 295 * Called with dcache_lock, drops it and then regains.
 296 */
 297static inline void prune_one_dentry(struct dentry * dentry)
 298{
 299        struct dentry * parent;
 300
 301        list_del_init(&dentry->d_hash);
 302        list_del(&dentry->d_child);
 303        dentry_stat.nr_dentry--;        /* For d_free, below */
 304        dentry_iput(dentry);
 305        parent = dentry->d_parent;
 306        d_free(dentry);
 307        if (parent != dentry)
 308                dput(parent);
 309        spin_lock(&dcache_lock);
 310}
 311
 312/**
 313 * prune_dcache - shrink the dcache
 314 * @count: number of entries to try and free
 315 *
 316 * Shrink the dcache. This is done when we need
 317 * more memory, or simply when we need to unmount
 318 * something (at which point we need to unuse
 319 * all dentries).
 320 *
 321 * This function may fail to free any resources if
 322 * all the dentries are in use.
 323 */
 324 
 325void prune_dcache(int count)
 326{
 327        spin_lock(&dcache_lock);
 328        for (;;) {
 329                struct dentry *dentry;
 330                struct list_head *tmp;
 331
 332                tmp = dentry_unused.prev;
 333
 334                if (tmp == &dentry_unused)
 335                        break;
 336                list_del_init(tmp);
 337                dentry = list_entry(tmp, struct dentry, d_lru);
 338
 339                /* If the dentry was recently referenced, don't free it. */
 340                if (dentry->d_vfs_flags & DCACHE_REFERENCED) {
 341                        dentry->d_vfs_flags &= ~DCACHE_REFERENCED;
 342                        list_add(&dentry->d_lru, &dentry_unused);
 343                        continue;
 344                }
 345                dentry_stat.nr_unused--;
 346
 347                /* Unused dentry with a count? */
 348                if (atomic_read(&dentry->d_count))
 349                        BUG();
 350
 351                prune_one_dentry(dentry);
 352                if (!--count)
 353                        break;
 354        }
 355        spin_unlock(&dcache_lock);
 356}
 357
 358/*
 359 * Shrink the dcache for the specified super block.
 360 * This allows us to unmount a device without disturbing
 361 * the dcache for the other devices.
 362 *
 363 * This implementation makes just two traversals of the
 364 * unused list.  On the first pass we move the selected
 365 * dentries to the most recent end, and on the second
 366 * pass we free them.  The second pass must restart after
 367 * each dput(), but since the target dentries are all at
 368 * the end, it's really just a single traversal.
 369 */
 370
 371/**
 372 * shrink_dcache_sb - shrink dcache for a superblock
 373 * @sb: superblock
 374 *
 375 * Shrink the dcache for the specified super block. This
 376 * is used to free the dcache before unmounting a file
 377 * system
 378 */
 379
 380void shrink_dcache_sb(struct super_block * sb)
 381{
 382        struct list_head *tmp, *next;
 383        struct dentry *dentry;
 384
 385        /*
 386         * Pass one ... move the dentries for the specified
 387         * superblock to the most recent end of the unused list.
 388         */
 389        spin_lock(&dcache_lock);
 390        next = dentry_unused.next;
 391        while (next != &dentry_unused) {
 392                tmp = next;
 393                next = tmp->next;
 394                dentry = list_entry(tmp, struct dentry, d_lru);
 395                if (dentry->d_sb != sb)
 396                        continue;
 397                list_del(tmp);
 398                list_add(tmp, &dentry_unused);
 399        }
 400
 401        /*
 402         * Pass two ... free the dentries for this superblock.
 403         */
 404repeat:
 405        next = dentry_unused.next;
 406        while (next != &dentry_unused) {
 407                tmp = next;
 408                next = tmp->next;
 409                dentry = list_entry(tmp, struct dentry, d_lru);
 410                if (dentry->d_sb != sb)
 411                        continue;
 412                if (atomic_read(&dentry->d_count))
 413                        continue;
 414                dentry_stat.nr_unused--;
 415                list_del_init(tmp);
 416                prune_one_dentry(dentry);
 417                goto repeat;
 418        }
 419        spin_unlock(&dcache_lock);
 420}
 421
 422/*
 423 * Search for at least 1 mount point in the dentry's subdirs.
 424 * We descend to the next level whenever the d_subdirs
 425 * list is non-empty and continue searching.
 426 */
 427 
 428/**
 429 * have_submounts - check for mounts over a dentry
 430 * @parent: dentry to check.
 431 *
 432 * Return true if the parent or its subdirectories contain
 433 * a mount point
 434 */
 435 
 436int have_submounts(struct dentry *parent)
 437{
 438        struct dentry *this_parent = parent;
 439        struct list_head *next;
 440
 441        spin_lock(&dcache_lock);
 442        if (d_mountpoint(parent))
 443                goto positive;
 444repeat:
 445        next = this_parent->d_subdirs.next;
 446resume:
 447        while (next != &this_parent->d_subdirs) {
 448                struct list_head *tmp = next;
 449                struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
 450                next = tmp->next;
 451                /* Have we found a mount point ? */
 452                if (d_mountpoint(dentry))
 453                        goto positive;
 454                if (!list_empty(&dentry->d_subdirs)) {
 455                        this_parent = dentry;
 456                        goto repeat;
 457                }
 458        }
 459        /*
 460         * All done at this level ... ascend and resume the search.
 461         */
 462        if (this_parent != parent) {
 463                next = this_parent->d_child.next; 
 464                this_parent = this_parent->d_parent;
 465                goto resume;
 466        }
 467        spin_unlock(&dcache_lock);
 468        return 0; /* No mount points found in tree */
 469positive:
 470        spin_unlock(&dcache_lock);
 471        return 1;
 472}
 473
 474/*
 475 * Search the dentry child list for the specified parent,
 476 * and move any unused dentries to the end of the unused
 477 * list for prune_dcache(). We descend to the next level
 478 * whenever the d_subdirs list is non-empty and continue
 479 * searching.
 480 */
 481static int select_parent(struct dentry * parent)
 482{
 483        struct dentry *this_parent = parent;
 484        struct list_head *next;
 485        int found = 0;
 486
 487        spin_lock(&dcache_lock);
 488repeat:
 489        next = this_parent->d_subdirs.next;
 490resume:
 491        while (next != &this_parent->d_subdirs) {
 492                struct list_head *tmp = next;
 493                struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
 494                next = tmp->next;
 495                if (!atomic_read(&dentry->d_count)) {
 496                        list_del(&dentry->d_lru);
 497                        list_add(&dentry->d_lru, dentry_unused.prev);
 498                        found++;
 499                }
 500                /*
 501                 * Descend a level if the d_subdirs list is non-empty.
 502                 */
 503                if (!list_empty(&dentry->d_subdirs)) {
 504                        this_parent = dentry;
 505#ifdef DCACHE_DEBUG
 506printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n",
 507dentry->d_parent->d_name.name, dentry->d_name.name, found);
 508#endif
 509                        goto repeat;
 510                }
 511        }
 512        /*
 513         * All done at this level ... ascend and resume the search.
 514         */
 515        if (this_parent != parent) {
 516                next = this_parent->d_child.next; 
 517                this_parent = this_parent->d_parent;
 518#ifdef DCACHE_DEBUG
 519printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n",
 520this_parent->d_parent->d_name.name, this_parent->d_name.name, found);
 521#endif
 522                goto resume;
 523        }
 524        spin_unlock(&dcache_lock);
 525        return found;
 526}
 527
 528/**
 529 * shrink_dcache_parent - prune dcache
 530 * @parent: parent of entries to prune
 531 *
 532 * Prune the dcache to remove unused children of the parent dentry.
 533 */
 534 
 535void shrink_dcache_parent(struct dentry * parent)
 536{
 537        int found;
 538
 539        while ((found = select_parent(parent)) != 0)
 540                prune_dcache(found);
 541}
 542
 543/*
 544 * This is called from kswapd when we think we need some
 545 * more memory, but aren't really sure how much. So we
 546 * carefully try to free a _bit_ of our dcache, but not
 547 * too much.
 548 *
 549 * Priority:
 550 *   0 - very urgent: shrink everything
 551 *  ...
 552 *   6 - base-level: try to shrink a bit.
 553 */
 554int shrink_dcache_memory(int priority, unsigned int gfp_mask)
 555{
 556        int count = 0;
 557
 558        /*
 559         * Nasty deadlock avoidance.
 560         *
 561         * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
 562         * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->
 563         * put_inode->ext2_discard_prealloc->ext2_free_blocks->lock_super->
 564         * DEADLOCK.
 565         *
 566         * We should make sure we don't hold the superblock lock over
 567         * block allocations, but for now:
 568         */
 569        if (!(gfp_mask & __GFP_FS))
 570                return 0;
 571
 572        count = dentry_stat.nr_unused / priority;
 573
 574        prune_dcache(count);
 575        return kmem_cache_shrink(dentry_cache);
 576}
 577
 578#define NAME_ALLOC_LEN(len)     ((len+16) & ~15)
 579
 580/**
 581 * d_alloc      -       allocate a dcache entry
 582 * @parent: parent of entry to allocate
 583 * @name: qstr of the name
 584 *
 585 * Allocates a dentry. It returns %NULL if there is insufficient memory
 586 * available. On a success the dentry is returned. The name passed in is
 587 * copied and the copy passed in may be reused after this call.
 588 */
 589 
 590struct dentry * d_alloc(struct dentry * parent, const struct qstr *name)
 591{
 592        char * str;
 593        struct dentry *dentry;
 594
 595        dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); 
 596        if (!dentry)
 597                return NULL;
 598
 599        if (name->len > DNAME_INLINE_LEN-1) {
 600                str = kmalloc(NAME_ALLOC_LEN(name->len), GFP_KERNEL);
 601                if (!str) {
 602                        kmem_cache_free(dentry_cache, dentry); 
 603                        return NULL;
 604                }
 605        } else
 606                str = dentry->d_iname; 
 607
 608        memcpy(str, name->name, name->len);
 609        str[name->len] = 0;
 610
 611        atomic_set(&dentry->d_count, 1);
 612        dentry->d_vfs_flags = 0;
 613        dentry->d_flags = 0;
 614        dentry->d_inode = NULL;
 615        dentry->d_parent = NULL;
 616        dentry->d_sb = NULL;
 617        dentry->d_name.name = str;
 618        dentry->d_name.len = name->len;
 619        dentry->d_name.hash = name->hash;
 620        dentry->d_op = NULL;
 621        dentry->d_fsdata = NULL;
 622        dentry->d_mounted = 0;
 623        INIT_LIST_HEAD(&dentry->d_hash);
 624        INIT_LIST_HEAD(&dentry->d_lru);
 625        INIT_LIST_HEAD(&dentry->d_subdirs);
 626        INIT_LIST_HEAD(&dentry->d_alias);
 627        if (parent) {
 628                dentry->d_parent = dget(parent);
 629                dentry->d_sb = parent->d_sb;
 630        } else
 631                INIT_LIST_HEAD(&dentry->d_child);
 632
 633        spin_lock(&dcache_lock);
 634        if (parent)
 635                list_add(&dentry->d_child, &parent->d_subdirs);
 636        dentry_stat.nr_dentry++;
 637        spin_unlock(&dcache_lock);
 638
 639        return dentry;
 640}
 641
 642/**
 643 * d_instantiate - fill in inode information for a dentry
 644 * @entry: dentry to complete
 645 * @inode: inode to attach to this dentry
 646 *
 647 * Fill in inode information in the entry.
 648 *
 649 * This turns negative dentries into productive full members
 650 * of society.
 651 *
 652 * NOTE! This assumes that the inode count has been incremented
 653 * (or otherwise set) by the caller to indicate that it is now
 654 * in use by the dcache.
 655 */
 656 
 657void d_instantiate(struct dentry *entry, struct inode * inode)
 658{
 659        if (!list_empty(&entry->d_alias)) BUG();
 660        spin_lock(&dcache_lock);
 661        if (inode)
 662                list_add(&entry->d_alias, &inode->i_dentry);
 663        entry->d_inode = inode;
 664        spin_unlock(&dcache_lock);
 665}
 666
 667/**
 668 * d_alloc_root - allocate root dentry
 669 * @root_inode: inode to allocate the root for
 670 *
 671 * Allocate a root ("/") dentry for the inode given. The inode is
 672 * instantiated and returned. %NULL is returned if there is insufficient
 673 * memory or the inode passed is %NULL.
 674 */
 675 
 676struct dentry * d_alloc_root(struct inode * root_inode)
 677{
 678        struct dentry *res = NULL;
 679
 680        if (root_inode) {
 681                res = d_alloc(NULL, &(const struct qstr) { "/", 1, 0 });
 682                if (res) {
 683                        res->d_sb = root_inode->i_sb;
 684                        res->d_parent = res;
 685                        d_instantiate(res, root_inode);
 686                }
 687        }
 688        return res;
 689}
 690
 691static inline struct list_head * d_hash(struct dentry * parent, unsigned long hash)
 692{
 693        hash += (unsigned long) parent / L1_CACHE_BYTES;
 694        hash = hash ^ (hash >> D_HASHBITS);
 695        return dentry_hashtable + (hash & D_HASHMASK);
 696}
 697
 698/**
 699 * d_lookup - search for a dentry
 700 * @parent: parent dentry
 701 * @name: qstr of name we wish to find
 702 *
 703 * Searches the children of the parent dentry for the name in question. If
 704 * the dentry is found its reference count is incremented and the dentry
 705 * is returned. The caller must use d_put to free the entry when it has
 706 * finished using it. %NULL is returned on failure.
 707 */
 708 
 709struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
 710{
 711        unsigned int len = name->len;
 712        unsigned int hash = name->hash;
 713        const unsigned char *str = name->name;
 714        struct list_head *head = d_hash(parent,hash);
 715        struct list_head *tmp;
 716
 717        spin_lock(&dcache_lock);
 718        tmp = head->next;
 719        for (;;) {
 720                struct dentry * dentry = list_entry(tmp, struct dentry, d_hash);
 721                if (tmp == head)
 722                        break;
 723                tmp = tmp->next;
 724                if (dentry->d_name.hash != hash)
 725                        continue;
 726                if (dentry->d_parent != parent)
 727                        continue;
 728                if (parent->d_op && parent->d_op->d_compare) {
 729                        if (parent->d_op->d_compare(parent, &dentry->d_name, name))
 730                                continue;
 731                } else {
 732                        if (dentry->d_name.len != len)
 733                                continue;
 734                        if (memcmp(dentry->d_name.name, str, len))
 735                                continue;
 736                }
 737                __dget_locked(dentry);
 738                dentry->d_vfs_flags |= DCACHE_REFERENCED;
 739                spin_unlock(&dcache_lock);
 740                return dentry;
 741        }
 742        spin_unlock(&dcache_lock);
 743        return NULL;
 744}
 745
 746/**
 747 * d_validate - verify dentry provided from insecure source
 748 * @dentry: The dentry alleged to be valid child of @dparent
 749 * @dparent: The parent dentry (known to be valid)
 750 * @hash: Hash of the dentry
 751 * @len: Length of the name
 752 *
 753 * An insecure source has sent us a dentry, here we verify it and dget() it.
 754 * This is used by ncpfs in its readdir implementation.
 755 * Zero is returned in the dentry is invalid.
 756 */
 757 
 758int d_validate(struct dentry *dentry, struct dentry *dparent)
 759{
 760        unsigned long dent_addr = (unsigned long) dentry;
 761        unsigned long min_addr = PAGE_OFFSET;
 762        unsigned long align_mask = 0x0F;
 763        struct list_head *base, *lhp;
 764
 765        if (dent_addr < min_addr)
 766                goto out;
 767        if (dent_addr > (unsigned long)high_memory - sizeof(struct dentry))
 768                goto out;
 769        if (dent_addr & align_mask)
 770                goto out;
 771        if ((!kern_addr_valid(dent_addr)) || (!kern_addr_valid(dent_addr -1 +
 772                                                sizeof(struct dentry))))
 773                goto out;
 774
 775        if (dentry->d_parent != dparent)
 776                goto out;
 777
 778        spin_lock(&dcache_lock);
 779        lhp = base = d_hash(dparent, dentry->d_name.hash);
 780        while ((lhp = lhp->next) != base) {
 781                if (dentry == list_entry(lhp, struct dentry, d_hash)) {
 782                        __dget_locked(dentry);
 783                        spin_unlock(&dcache_lock);
 784                        return 1;
 785                }
 786        }
 787        spin_unlock(&dcache_lock);
 788out:
 789        return 0;
 790}
 791
 792/*
 793 * When a file is deleted, we have two options:
 794 * - turn this dentry into a negative dentry
 795 * - unhash this dentry and free it.
 796 *
 797 * Usually, we want to just turn this into
 798 * a negative dentry, but if anybody else is
 799 * currently using the dentry or the inode
 800 * we can't do that and we fall back on removing
 801 * it from the hash queues and waiting for
 802 * it to be deleted later when it has no users
 803 */
 804 
 805/**
 806 * d_delete - delete a dentry
 807 * @dentry: The dentry to delete
 808 *
 809 * Turn the dentry into a negative dentry if possible, otherwise
 810 * remove it from the hash queues so it can be deleted later
 811 */
 812 
 813void d_delete(struct dentry * dentry)
 814{
 815        /*
 816         * Are we the only user?
 817         */
 818        spin_lock(&dcache_lock);
 819        if (atomic_read(&dentry->d_count) == 1) {
 820                dentry_iput(dentry);
 821                return;
 822        }
 823        spin_unlock(&dcache_lock);
 824
 825        /*
 826         * If not, just drop the dentry and let dput
 827         * pick up the tab..
 828         */
 829        d_drop(dentry);
 830}
 831
 832/**
 833 * d_rehash     - add an entry back to the hash
 834 * @entry: dentry to add to the hash
 835 *
 836 * Adds a dentry to the hash according to its name.
 837 */
 838 
 839void d_rehash(struct dentry * entry)
 840{
 841        struct list_head *list = d_hash(entry->d_parent, entry->d_name.hash);
 842        if (!list_empty(&entry->d_hash)) BUG();
 843        spin_lock(&dcache_lock);
 844        list_add(&entry->d_hash, list);
 845        spin_unlock(&dcache_lock);
 846}
 847
 848#define do_switch(x,y) do { \
 849        __typeof__ (x) __tmp = x; \
 850        x = y; y = __tmp; } while (0)
 851
 852/*
 853 * When switching names, the actual string doesn't strictly have to
 854 * be preserved in the target - because we're dropping the target
 855 * anyway. As such, we can just do a simple memcpy() to copy over
 856 * the new name before we switch.
 857 *
 858 * Note that we have to be a lot more careful about getting the hash
 859 * switched - we have to switch the hash value properly even if it
 860 * then no longer matches the actual (corrupted) string of the target.
 861 * The hash value has to match the hash queue that the dentry is on..
 862 */
 863static inline void switch_names(struct dentry * dentry, struct dentry * target)
 864{
 865        const unsigned char *old_name, *new_name;
 866
 867        check_lock();
 868        memcpy(dentry->d_iname, target->d_iname, DNAME_INLINE_LEN); 
 869        old_name = target->d_name.name;
 870        new_name = dentry->d_name.name;
 871        if (old_name == target->d_iname)
 872                old_name = dentry->d_iname;
 873        if (new_name == dentry->d_iname)
 874                new_name = target->d_iname;
 875        target->d_name.name = new_name;
 876        dentry->d_name.name = old_name;
 877}
 878
 879/*
 880 * We cannibalize "target" when moving dentry on top of it,
 881 * because it's going to be thrown away anyway. We could be more
 882 * polite about it, though.
 883 *
 884 * This forceful removal will result in ugly /proc output if
 885 * somebody holds a file open that got deleted due to a rename.
 886 * We could be nicer about the deleted file, and let it show
 887 * up under the name it got deleted rather than the name that
 888 * deleted it.
 889 *
 890 * Careful with the hash switch. The hash switch depends on
 891 * the fact that any list-entry can be a head of the list.
 892 * Think about it.
 893 */
 894 
 895/**
 896 * d_move - move a dentry
 897 * @dentry: entry to move
 898 * @target: new dentry
 899 *
 900 * Update the dcache to reflect the move of a file name. Negative
 901 * dcache entries should not be moved in this way.
 902 */
 903  
 904void d_move(struct dentry * dentry, struct dentry * target)
 905{
 906        check_lock();
 907
 908        if (!dentry->d_inode)
 909                printk(KERN_WARNING "VFS: moving negative dcache entry\n");
 910
 911        spin_lock(&dcache_lock);
 912        /* Move the dentry to the target hash queue */
 913        list_del(&dentry->d_hash);
 914        list_add(&dentry->d_hash, &target->d_hash);
 915
 916        /* Unhash the target: dput() will then get rid of it */
 917        list_del_init(&target->d_hash);
 918
 919        list_del(&dentry->d_child);
 920        list_del(&target->d_child);
 921
 922        /* Switch the parents and the names.. */
 923        switch_names(dentry, target);
 924        do_switch(dentry->d_parent, target->d_parent);
 925        do_switch(dentry->d_name.len, target->d_name.len);
 926        do_switch(dentry->d_name.hash, target->d_name.hash);
 927
 928        /* And add them back to the (new) parent lists */
 929        list_add(&target->d_child, &target->d_parent->d_subdirs);
 930        list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
 931        spin_unlock(&dcache_lock);
 932}
 933
 934/**
 935 * d_path - return the path of a dentry
 936 * @dentry: dentry to report
 937 * @vfsmnt: vfsmnt to which the dentry belongs
 938 * @root: root dentry
 939 * @rootmnt: vfsmnt to which the root dentry belongs
 940 * @buffer: buffer to return value in
 941 * @buflen: buffer length
 942 *
 943 * Convert a dentry into an ASCII path name. If the entry has been deleted
 944 * the string " (deleted)" is appended. Note that this is ambiguous. Returns
 945 * the buffer.
 946 *
 947 * "buflen" should be %PAGE_SIZE or more. Caller holds the dcache_lock.
 948 */
 949char * __d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
 950                struct dentry *root, struct vfsmount *rootmnt,
 951                char *buffer, int buflen)
 952{
 953        char * end = buffer+buflen;
 954        char * retval;
 955        int namelen;
 956
 957        *--end = '\0';
 958        buflen--;
 959        if (!IS_ROOT(dentry) && list_empty(&dentry->d_hash)) {
 960                buflen -= 10;
 961                end -= 10;
 962                memcpy(end, " (deleted)", 10);
 963        }
 964
 965        /* Get '/' right */
 966        retval = end-1;
 967        *retval = '/';
 968
 969        for (;;) {
 970                struct dentry * parent;
 971
 972                if (dentry == root && vfsmnt == rootmnt)
 973                        break;
 974                if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
 975                        /* Global root? */
 976                        if (vfsmnt->mnt_parent == vfsmnt)
 977                                goto global_root;
 978                        dentry = vfsmnt->mnt_mountpoint;
 979                        vfsmnt = vfsmnt->mnt_parent;
 980                        continue;
 981                }
 982                parent = dentry->d_parent;
 983                namelen = dentry->d_name.len;
 984                buflen -= namelen + 1;
 985                if (buflen < 0)
 986                        return ERR_PTR(-ENAMETOOLONG);
 987                end -= namelen;
 988                memcpy(end, dentry->d_name.name, namelen);
 989                *--end = '/';
 990                retval = end;
 991                dentry = parent;
 992        }
 993
 994        return retval;
 995
 996global_root:
 997        namelen = dentry->d_name.len;
 998        buflen -= namelen;
 999        if (buflen >= 0) {
1000                retval -= namelen-1;    /* hit the slash */
1001                memcpy(retval, dentry->d_name.name, namelen);
1002        } else
1003                retval = ERR_PTR(-ENAMETOOLONG);
1004        return retval;
1005}
1006
1007/*
1008 * NOTE! The user-level library version returns a
1009 * character pointer. The kernel system call just
1010 * returns the length of the buffer filled (which
1011 * includes the ending '\0' character), or a negative
1012 * error value. So libc would do something like
1013 *
1014 *      char *getcwd(char * buf, size_t size)
1015 *      {
1016 *              int retval;
1017 *
1018 *              retval = sys_getcwd(buf, size);
1019 *              if (retval >= 0)
1020 *                      return buf;
1021 *              errno = -retval;
1022 *              return NULL;
1023 *      }
1024 */
1025asmlinkage long sys_getcwd(char *buf, unsigned long size)
1026{
1027        int error;
1028        struct vfsmount *pwdmnt, *rootmnt;
1029        struct dentry *pwd, *root;
1030        char *page = (char *) __get_free_page(GFP_USER);
1031
1032        if (!page)
1033                return -ENOMEM;
1034
1035        read_lock(&current->fs->lock);
1036        pwdmnt = mntget(current->fs->pwdmnt);
1037        pwd = dget(current->fs->pwd);
1038        rootmnt = mntget(current->fs->rootmnt);
1039        root = dget(current->fs->root);
1040        read_unlock(&current->fs->lock);
1041
1042        error = -ENOENT;
1043        /* Has the current directory has been unlinked? */
1044        spin_lock(&dcache_lock);
1045        if (pwd->d_parent == pwd || !list_empty(&pwd->d_hash)) {
1046                unsigned long len;
1047                char * cwd;
1048
1049                cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1050                spin_unlock(&dcache_lock);
1051
1052                error = PTR_ERR(cwd);
1053                if (IS_ERR(cwd))
1054                        goto out;
1055
1056                error = -ERANGE;
1057                len = PAGE_SIZE + page - cwd;
1058                if (len <= size) {
1059                        error = len;
1060                        if (copy_to_user(buf, cwd, len))
1061                                error = -EFAULT;
1062                }
1063        } else
1064                spin_unlock(&dcache_lock);
1065
1066out:
1067        dput(pwd);
1068        mntput(pwdmnt);
1069        dput(root);
1070        mntput(rootmnt);
1071        free_page((unsigned long) page);
1072        return error;
1073}
1074
1075/*
1076 * Test whether new_dentry is a subdirectory of old_dentry.
1077 *
1078 * Trivially implemented using the dcache structure
1079 */
1080
1081/**
1082 * is_subdir - is new dentry a subdirectory of old_dentry
1083 * @new_dentry: new dentry
1084 * @old_dentry: old dentry
1085 *
1086 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1087 * Returns 0 otherwise.
1088 */
1089  
1090int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1091{
1092        int result;
1093
1094        result = 0;
1095        for (;;) {
1096                if (new_dentry != old_dentry) {
1097                        struct dentry * parent = new_dentry->d_parent;
1098                        if (parent == new_dentry)
1099                                break;
1100                        new_dentry = parent;
1101                        continue;
1102                }
1103                result = 1;
1104                break;
1105        }
1106        return result;
1107}
1108
1109void d_genocide(struct dentry *root)
1110{
1111        struct dentry *this_parent = root;
1112        struct list_head *next;
1113
1114        spin_lock(&dcache_lock);
1115repeat:
1116        next = this_parent->d_subdirs.next;
1117resume:
1118        while (next != &this_parent->d_subdirs) {
1119                struct list_head *tmp = next;
1120                struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1121                next = tmp->next;
1122                if (d_unhashed(dentry)||!dentry->d_inode)
1123                        continue;
1124                if (!list_empty(&dentry->d_subdirs)) {
1125                        this_parent = dentry;
1126                        goto repeat;
1127                }
1128                atomic_dec(&dentry->d_count);
1129        }
1130        if (this_parent != root) {
1131                next = this_parent->d_child.next; 
1132                atomic_dec(&this_parent->d_count);
1133                this_parent = this_parent->d_parent;
1134                goto resume;
1135        }
1136        spin_unlock(&dcache_lock);
1137}
1138
1139/**
1140 * find_inode_number - check for dentry with name
1141 * @dir: directory to check
1142 * @name: Name to find.
1143 *
1144 * Check whether a dentry already exists for the given name,
1145 * and return the inode number if it has an inode. Otherwise
1146 * 0 is returned.
1147 *
1148 * This routine is used to post-process directory listings for
1149 * filesystems using synthetic inode numbers, and is necessary
1150 * to keep getcwd() working.
1151 */
1152 
1153ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1154{
1155        struct dentry * dentry;
1156        ino_t ino = 0;
1157
1158        /*
1159         * Check for a fs-specific hash function. Note that we must
1160         * calculate the standard hash first, as the d_op->d_hash()
1161         * routine may choose to leave the hash value unchanged.
1162         */
1163        name->hash = full_name_hash(name->name, name->len);
1164        if (dir->d_op && dir->d_op->d_hash)
1165        {
1166                if (dir->d_op->d_hash(dir, name) != 0)
1167                        goto out;
1168        }
1169
1170        dentry = d_lookup(dir, name);
1171        if (dentry)
1172        {
1173                if (dentry->d_inode)
1174                        ino = dentry->d_inode->i_ino;
1175                dput(dentry);
1176        }
1177out:
1178        return ino;
1179}
1180
1181static void __init dcache_init(unsigned long mempages)
1182{
1183        struct list_head *d;
1184        unsigned long order;
1185        unsigned int nr_hash;
1186        int i;
1187
1188        /* 
1189         * A constructor could be added for stable state like the lists,
1190         * but it is probably not worth it because of the cache nature
1191         * of the dcache. 
1192         * If fragmentation is too bad then the SLAB_HWCACHE_ALIGN
1193         * flag could be removed here, to hint to the allocator that
1194         * it should not try to get multiple page regions.  
1195         */
1196        dentry_cache = kmem_cache_create("dentry_cache",
1197                                         sizeof(struct dentry),
1198                                         0,
1199                                         SLAB_HWCACHE_ALIGN,
1200                                         NULL, NULL);
1201        if (!dentry_cache)
1202                panic("Cannot create dentry cache");
1203
1204#if PAGE_SHIFT < 13
1205        mempages >>= (13 - PAGE_SHIFT);
1206#endif
1207        mempages *= sizeof(struct list_head);
1208        for (order = 0; ((1UL << order) << PAGE_SHIFT) < mempages; order++)
1209                ;
1210
1211        do {
1212                unsigned long tmp;
1213
1214                nr_hash = (1UL << order) * PAGE_SIZE /
1215                        sizeof(struct list_head);
1216                d_hash_mask = (nr_hash - 1);
1217
1218                tmp = nr_hash;
1219                d_hash_shift = 0;
1220                while ((tmp >>= 1UL) != 0UL)
1221                        d_hash_shift++;
1222
1223                dentry_hashtable = (struct list_head *)
1224                        __get_free_pages(GFP_ATOMIC, order);
1225        } while (dentry_hashtable == NULL && --order >= 0);
1226
1227        printk(KERN_INFO "Dentry cache hash table entries: %d (order: %ld, %ld bytes)\n",
1228                        nr_hash, order, (PAGE_SIZE << order));
1229
1230        if (!dentry_hashtable)
1231                panic("Failed to allocate dcache hash table\n");
1232
1233        d = dentry_hashtable;
1234        i = nr_hash;
1235        do {
1236                INIT_LIST_HEAD(d);
1237                d++;
1238                i--;
1239        } while (i);
1240}
1241
1242static void init_buffer_head(void * foo, kmem_cache_t * cachep, unsigned long flags)
1243{
1244        if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1245            SLAB_CTOR_CONSTRUCTOR)
1246        {
1247                struct buffer_head * bh = (struct buffer_head *) foo;
1248
1249                memset(bh, 0, sizeof(*bh));
1250                init_waitqueue_head(&bh->b_wait);
1251        }
1252}
1253
1254/* SLAB cache for __getname() consumers */
1255kmem_cache_t *names_cachep;
1256
1257/* SLAB cache for file structures */
1258kmem_cache_t *filp_cachep;
1259
1260/* SLAB cache for dquot structures */
1261kmem_cache_t *dquot_cachep;
1262
1263/* SLAB cache for buffer_head structures */
1264kmem_cache_t *bh_cachep;
1265EXPORT_SYMBOL(bh_cachep);
1266
1267extern void bdev_cache_init(void);
1268extern void cdev_cache_init(void);
1269extern void iobuf_cache_init(void);
1270
1271void __init vfs_caches_init(unsigned long mempages)
1272{
1273        bh_cachep = kmem_cache_create("buffer_head",
1274                        sizeof(struct buffer_head), 0,
1275                        SLAB_HWCACHE_ALIGN, init_buffer_head, NULL);
1276        if(!bh_cachep)
1277                panic("Cannot create buffer head SLAB cache");
1278
1279        names_cachep = kmem_cache_create("names_cache", 
1280                        PATH_MAX, 0, 
1281                        SLAB_HWCACHE_ALIGN, NULL, NULL);
1282        if (!names_cachep)
1283                panic("Cannot create names SLAB cache");
1284
1285        filp_cachep = kmem_cache_create("filp", 
1286                        sizeof(struct file), 0,
1287                        SLAB_HWCACHE_ALIGN, NULL, NULL);
1288        if(!filp_cachep)
1289                panic("Cannot create filp SLAB cache");
1290
1291#if defined (CONFIG_QUOTA)
1292        dquot_cachep = kmem_cache_create("dquot", 
1293                        sizeof(struct dquot), sizeof(unsigned long) * 4,
1294                        SLAB_HWCACHE_ALIGN, NULL, NULL);
1295        if (!dquot_cachep)
1296                panic("Cannot create dquot SLAB cache");
1297#endif
1298
1299        dcache_init(mempages);
1300        inode_init(mempages);
1301        files_init(mempages); 
1302        mnt_init(mempages);
1303        bdev_cache_init();
1304        cdev_cache_init();
1305        iobuf_cache_init();
1306}
1307
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.