4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
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.
17 #include <linux/config.h>
18 #include <linux/syscalls.h>
19 #include <linux/string.h>
22 #include <linux/fsnotify.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/smp_lock.h>
26 #include <linux/hash.h>
27 #include <linux/cache.h>
28 #include <linux/module.h>
29 #include <linux/mount.h>
30 #include <linux/file.h>
31 #include <asm/uaccess.h>
32 #include <linux/security.h>
33 #include <linux/seqlock.h>
34 #include <linux/swap.h>
35 #include <linux/bootmem.h>
37 /* #define DCACHE_DEBUG 1 */
39 int sysctl_vfs_cache_pressure = 100;
40 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
42 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
43 static seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
45 EXPORT_SYMBOL(dcache_lock);
47 static kmem_cache_t *dentry_cache;
49 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
52 * This is the single most critical data structure when it comes
53 * to the dcache: the hashtable for lookups. Somebody should try
54 * to make this good - I've just made it work.
56 * This hash-function tries to avoid losing too many bits of hash
57 * information, yet avoid using a prime hash-size or similar.
59 #define D_HASHBITS d_hash_shift
60 #define D_HASHMASK d_hash_mask
62 static unsigned int d_hash_mask;
63 static unsigned int d_hash_shift;
64 static struct hlist_head *dentry_hashtable;
65 static LIST_HEAD(dentry_unused);
67 /* Statistics gathering. */
68 struct dentry_stat_t dentry_stat = {
72 static void d_callback(struct rcu_head *head)
74 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
76 if (dname_external(dentry))
77 kfree(dentry->d_name.name);
78 kmem_cache_free(dentry_cache, dentry);
82 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
85 static void d_free(struct dentry *dentry)
87 if (dentry->d_op && dentry->d_op->d_release)
88 dentry->d_op->d_release(dentry);
89 call_rcu(&dentry->d_u.d_rcu, d_callback);
93 * Release the dentry's inode, using the filesystem
94 * d_iput() operation if defined.
95 * Called with dcache_lock and per dentry lock held, drops both.
97 static inline void dentry_iput(struct dentry * dentry)
99 struct inode *inode = dentry->d_inode;
101 dentry->d_inode = NULL;
102 list_del_init(&dentry->d_alias);
103 spin_unlock(&dentry->d_lock);
104 spin_unlock(&dcache_lock);
106 fsnotify_inoderemove(inode);
107 if (dentry->d_op && dentry->d_op->d_iput)
108 dentry->d_op->d_iput(dentry, inode);
112 spin_unlock(&dentry->d_lock);
113 spin_unlock(&dcache_lock);
120 * This is complicated by the fact that we do not want to put
121 * dentries that are no longer on any hash chain on the unused
122 * list: we'd much rather just get rid of them immediately.
124 * However, that implies that we have to traverse the dentry
125 * tree upwards to the parents which might _also_ now be
126 * scheduled for deletion (it may have been only waiting for
127 * its last child to go away).
129 * This tail recursion is done by hand as we don't want to depend
130 * on the compiler to always get this right (gcc generally doesn't).
131 * Real recursion would eat up our stack space.
135 * dput - release a dentry
136 * @dentry: dentry to release
138 * Release a dentry. This will drop the usage count and if appropriate
139 * call the dentry unlink method as well as removing it from the queues and
140 * releasing its resources. If the parent dentries were scheduled for release
141 * they too may now get deleted.
143 * no dcache lock, please.
146 void dput(struct dentry *dentry)
152 if (atomic_read(&dentry->d_count) == 1)
154 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
157 spin_lock(&dentry->d_lock);
158 if (atomic_read(&dentry->d_count)) {
159 spin_unlock(&dentry->d_lock);
160 spin_unlock(&dcache_lock);
165 * AV: ->d_delete() is _NOT_ allowed to block now.
167 if (dentry->d_op && dentry->d_op->d_delete) {
168 if (dentry->d_op->d_delete(dentry))
171 /* Unreachable? Get rid of it */
172 if (d_unhashed(dentry))
174 if (list_empty(&dentry->d_lru)) {
175 dentry->d_flags |= DCACHE_REFERENCED;
176 list_add(&dentry->d_lru, &dentry_unused);
177 dentry_stat.nr_unused++;
179 spin_unlock(&dentry->d_lock);
180 spin_unlock(&dcache_lock);
187 struct dentry *parent;
189 /* If dentry was on d_lru list
190 * delete it from there
192 if (!list_empty(&dentry->d_lru)) {
193 list_del(&dentry->d_lru);
194 dentry_stat.nr_unused--;
196 list_del(&dentry->d_u.d_child);
197 dentry_stat.nr_dentry--; /* For d_free, below */
198 /*drops the locks, at that point nobody can reach this dentry */
200 parent = dentry->d_parent;
202 if (dentry == parent)
210 * d_invalidate - invalidate a dentry
211 * @dentry: dentry to invalidate
213 * Try to invalidate the dentry if it turns out to be
214 * possible. If there are other dentries that can be
215 * reached through this one we can't delete it and we
216 * return -EBUSY. On success we return 0.
221 int d_invalidate(struct dentry * dentry)
224 * If it's already been dropped, return OK.
226 spin_lock(&dcache_lock);
227 if (d_unhashed(dentry)) {
228 spin_unlock(&dcache_lock);
232 * Check whether to do a partial shrink_dcache
233 * to get rid of unused child entries.
235 if (!list_empty(&dentry->d_subdirs)) {
236 spin_unlock(&dcache_lock);
237 shrink_dcache_parent(dentry);
238 spin_lock(&dcache_lock);
242 * Somebody else still using it?
244 * If it's a directory, we can't drop it
245 * for fear of somebody re-populating it
246 * with children (even though dropping it
247 * would make it unreachable from the root,
248 * we might still populate it if it was a
249 * working directory or similar).
251 spin_lock(&dentry->d_lock);
252 if (atomic_read(&dentry->d_count) > 1) {
253 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
254 spin_unlock(&dentry->d_lock);
255 spin_unlock(&dcache_lock);
261 spin_unlock(&dentry->d_lock);
262 spin_unlock(&dcache_lock);
266 /* This should be called _only_ with dcache_lock held */
268 static inline struct dentry * __dget_locked(struct dentry *dentry)
270 atomic_inc(&dentry->d_count);
271 if (!list_empty(&dentry->d_lru)) {
272 dentry_stat.nr_unused--;
273 list_del_init(&dentry->d_lru);
278 struct dentry * dget_locked(struct dentry *dentry)
280 return __dget_locked(dentry);
284 * d_find_alias - grab a hashed alias of inode
285 * @inode: inode in question
286 * @want_discon: flag, used by d_splice_alias, to request
287 * that only a DISCONNECTED alias be returned.
289 * If inode has a hashed alias, or is a directory and has any alias,
290 * acquire the reference to alias and return it. Otherwise return NULL.
291 * Notice that if inode is a directory there can be only one alias and
292 * it can be unhashed only if it has no children, or if it is the root
295 * If the inode has a DCACHE_DISCONNECTED alias, then prefer
296 * any other hashed alias over that one unless @want_discon is set,
297 * in which case only return a DCACHE_DISCONNECTED alias.
300 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
302 struct list_head *head, *next, *tmp;
303 struct dentry *alias, *discon_alias=NULL;
305 head = &inode->i_dentry;
306 next = inode->i_dentry.next;
307 while (next != head) {
311 alias = list_entry(tmp, struct dentry, d_alias);
312 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
313 if (alias->d_flags & DCACHE_DISCONNECTED)
314 discon_alias = alias;
315 else if (!want_discon) {
316 __dget_locked(alias);
322 __dget_locked(discon_alias);
326 struct dentry * d_find_alias(struct inode *inode)
329 spin_lock(&dcache_lock);
330 de = __d_find_alias(inode, 0);
331 spin_unlock(&dcache_lock);
336 * Try to kill dentries associated with this inode.
337 * WARNING: you must own a reference to inode.
339 void d_prune_aliases(struct inode *inode)
341 struct dentry *dentry;
343 spin_lock(&dcache_lock);
344 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
345 spin_lock(&dentry->d_lock);
346 if (!atomic_read(&dentry->d_count)) {
347 __dget_locked(dentry);
349 spin_unlock(&dentry->d_lock);
350 spin_unlock(&dcache_lock);
354 spin_unlock(&dentry->d_lock);
356 spin_unlock(&dcache_lock);
360 * Throw away a dentry - free the inode, dput the parent.
361 * This requires that the LRU list has already been
363 * Called with dcache_lock, drops it and then regains.
365 static inline void prune_one_dentry(struct dentry * dentry)
367 struct dentry * parent;
370 list_del(&dentry->d_u.d_child);
371 dentry_stat.nr_dentry--; /* For d_free, below */
373 parent = dentry->d_parent;
375 if (parent != dentry)
377 spin_lock(&dcache_lock);
381 * prune_dcache - shrink the dcache
382 * @count: number of entries to try and free
384 * Shrink the dcache. This is done when we need
385 * more memory, or simply when we need to unmount
386 * something (at which point we need to unuse
389 * This function may fail to free any resources if
390 * all the dentries are in use.
393 static void prune_dcache(int count)
395 spin_lock(&dcache_lock);
396 for (; count ; count--) {
397 struct dentry *dentry;
398 struct list_head *tmp;
400 cond_resched_lock(&dcache_lock);
402 tmp = dentry_unused.prev;
403 if (tmp == &dentry_unused)
406 prefetch(dentry_unused.prev);
407 dentry_stat.nr_unused--;
408 dentry = list_entry(tmp, struct dentry, d_lru);
410 spin_lock(&dentry->d_lock);
412 * We found an inuse dentry which was not removed from
413 * dentry_unused because of laziness during lookup. Do not free
414 * it - just keep it off the dentry_unused list.
416 if (atomic_read(&dentry->d_count)) {
417 spin_unlock(&dentry->d_lock);
420 /* If the dentry was recently referenced, don't free it. */
421 if (dentry->d_flags & DCACHE_REFERENCED) {
422 dentry->d_flags &= ~DCACHE_REFERENCED;
423 list_add(&dentry->d_lru, &dentry_unused);
424 dentry_stat.nr_unused++;
425 spin_unlock(&dentry->d_lock);
428 prune_one_dentry(dentry);
430 spin_unlock(&dcache_lock);
434 * Shrink the dcache for the specified super block.
435 * This allows us to unmount a device without disturbing
436 * the dcache for the other devices.
438 * This implementation makes just two traversals of the
439 * unused list. On the first pass we move the selected
440 * dentries to the most recent end, and on the second
441 * pass we free them. The second pass must restart after
442 * each dput(), but since the target dentries are all at
443 * the end, it's really just a single traversal.
447 * shrink_dcache_sb - shrink dcache for a superblock
450 * Shrink the dcache for the specified super block. This
451 * is used to free the dcache before unmounting a file
455 void shrink_dcache_sb(struct super_block * sb)
457 struct list_head *tmp, *next;
458 struct dentry *dentry;
461 * Pass one ... move the dentries for the specified
462 * superblock to the most recent end of the unused list.
464 spin_lock(&dcache_lock);
465 list_for_each_safe(tmp, next, &dentry_unused) {
466 dentry = list_entry(tmp, struct dentry, d_lru);
467 if (dentry->d_sb != sb)
470 list_add(tmp, &dentry_unused);
474 * Pass two ... free the dentries for this superblock.
477 list_for_each_safe(tmp, next, &dentry_unused) {
478 dentry = list_entry(tmp, struct dentry, d_lru);
479 if (dentry->d_sb != sb)
481 dentry_stat.nr_unused--;
483 spin_lock(&dentry->d_lock);
484 if (atomic_read(&dentry->d_count)) {
485 spin_unlock(&dentry->d_lock);
488 prune_one_dentry(dentry);
491 spin_unlock(&dcache_lock);
495 * Search for at least 1 mount point in the dentry's subdirs.
496 * We descend to the next level whenever the d_subdirs
497 * list is non-empty and continue searching.
501 * have_submounts - check for mounts over a dentry
502 * @parent: dentry to check.
504 * Return true if the parent or its subdirectories contain
508 int have_submounts(struct dentry *parent)
510 struct dentry *this_parent = parent;
511 struct list_head *next;
513 spin_lock(&dcache_lock);
514 if (d_mountpoint(parent))
517 next = this_parent->d_subdirs.next;
519 while (next != &this_parent->d_subdirs) {
520 struct list_head *tmp = next;
521 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
523 /* Have we found a mount point ? */
524 if (d_mountpoint(dentry))
526 if (!list_empty(&dentry->d_subdirs)) {
527 this_parent = dentry;
532 * All done at this level ... ascend and resume the search.
534 if (this_parent != parent) {
535 next = this_parent->d_u.d_child.next;
536 this_parent = this_parent->d_parent;
539 spin_unlock(&dcache_lock);
540 return 0; /* No mount points found in tree */
542 spin_unlock(&dcache_lock);
547 * Search the dentry child list for the specified parent,
548 * and move any unused dentries to the end of the unused
549 * list for prune_dcache(). We descend to the next level
550 * whenever the d_subdirs list is non-empty and continue
553 * It returns zero iff there are no unused children,
554 * otherwise it returns the number of children moved to
555 * the end of the unused list. This may not be the total
556 * number of unused children, because select_parent can
557 * drop the lock and return early due to latency
560 static int select_parent(struct dentry * parent)
562 struct dentry *this_parent = parent;
563 struct list_head *next;
566 spin_lock(&dcache_lock);
568 next = this_parent->d_subdirs.next;
570 while (next != &this_parent->d_subdirs) {
571 struct list_head *tmp = next;
572 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
575 if (!list_empty(&dentry->d_lru)) {
576 dentry_stat.nr_unused--;
577 list_del_init(&dentry->d_lru);
580 * move only zero ref count dentries to the end
581 * of the unused list for prune_dcache
583 if (!atomic_read(&dentry->d_count)) {
584 list_add(&dentry->d_lru, dentry_unused.prev);
585 dentry_stat.nr_unused++;
590 * We can return to the caller if we have found some (this
591 * ensures forward progress). We'll be coming back to find
594 if (found && need_resched())
598 * Descend a level if the d_subdirs list is non-empty.
600 if (!list_empty(&dentry->d_subdirs)) {
601 this_parent = dentry;
603 printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n",
604 dentry->d_parent->d_name.name, dentry->d_name.name, found);
610 * All done at this level ... ascend and resume the search.
612 if (this_parent != parent) {
613 next = this_parent->d_u.d_child.next;
614 this_parent = this_parent->d_parent;
616 printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n",
617 this_parent->d_parent->d_name.name, this_parent->d_name.name, found);
622 spin_unlock(&dcache_lock);
627 * shrink_dcache_parent - prune dcache
628 * @parent: parent of entries to prune
630 * Prune the dcache to remove unused children of the parent dentry.
633 void shrink_dcache_parent(struct dentry * parent)
637 while ((found = select_parent(parent)) != 0)
642 * shrink_dcache_anon - further prune the cache
643 * @head: head of d_hash list of dentries to prune
645 * Prune the dentries that are anonymous
647 * parsing d_hash list does not hlist_for_each_entry_rcu() as it
648 * done under dcache_lock.
651 void shrink_dcache_anon(struct hlist_head *head)
653 struct hlist_node *lp;
657 spin_lock(&dcache_lock);
658 hlist_for_each(lp, head) {
659 struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
660 if (!list_empty(&this->d_lru)) {
661 dentry_stat.nr_unused--;
662 list_del_init(&this->d_lru);
666 * move only zero ref count dentries to the end
667 * of the unused list for prune_dcache
669 if (!atomic_read(&this->d_count)) {
670 list_add_tail(&this->d_lru, &dentry_unused);
671 dentry_stat.nr_unused++;
675 spin_unlock(&dcache_lock);
681 * Scan `nr' dentries and return the number which remain.
683 * We need to avoid reentering the filesystem if the caller is performing a
684 * GFP_NOFS allocation attempt. One example deadlock is:
686 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
687 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
688 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
690 * In this case we return -1 to tell the caller that we baled.
692 static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
695 if (!(gfp_mask & __GFP_FS))
699 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
703 * d_alloc - allocate a dcache entry
704 * @parent: parent of entry to allocate
705 * @name: qstr of the name
707 * Allocates a dentry. It returns %NULL if there is insufficient memory
708 * available. On a success the dentry is returned. The name passed in is
709 * copied and the copy passed in may be reused after this call.
712 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
714 struct dentry *dentry;
717 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
721 if (name->len > DNAME_INLINE_LEN-1) {
722 dname = kmalloc(name->len + 1, GFP_KERNEL);
724 kmem_cache_free(dentry_cache, dentry);
728 dname = dentry->d_iname;
730 dentry->d_name.name = dname;
732 dentry->d_name.len = name->len;
733 dentry->d_name.hash = name->hash;
734 memcpy(dname, name->name, name->len);
735 dname[name->len] = 0;
737 atomic_set(&dentry->d_count, 1);
738 dentry->d_flags = DCACHE_UNHASHED;
739 spin_lock_init(&dentry->d_lock);
740 dentry->d_inode = NULL;
741 dentry->d_parent = NULL;
744 dentry->d_fsdata = NULL;
745 dentry->d_mounted = 0;
746 dentry->d_cookie = NULL;
747 INIT_HLIST_NODE(&dentry->d_hash);
748 INIT_LIST_HEAD(&dentry->d_lru);
749 INIT_LIST_HEAD(&dentry->d_subdirs);
750 INIT_LIST_HEAD(&dentry->d_alias);
753 dentry->d_parent = dget(parent);
754 dentry->d_sb = parent->d_sb;
756 INIT_LIST_HEAD(&dentry->d_u.d_child);
759 spin_lock(&dcache_lock);
761 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
762 dentry_stat.nr_dentry++;
763 spin_unlock(&dcache_lock);
768 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
773 q.len = strlen(name);
774 q.hash = full_name_hash(q.name, q.len);
775 return d_alloc(parent, &q);
779 * d_instantiate - fill in inode information for a dentry
780 * @entry: dentry to complete
781 * @inode: inode to attach to this dentry
783 * Fill in inode information in the entry.
785 * This turns negative dentries into productive full members
788 * NOTE! This assumes that the inode count has been incremented
789 * (or otherwise set) by the caller to indicate that it is now
790 * in use by the dcache.
793 void d_instantiate(struct dentry *entry, struct inode * inode)
795 if (!list_empty(&entry->d_alias)) BUG();
796 spin_lock(&dcache_lock);
798 list_add(&entry->d_alias, &inode->i_dentry);
799 entry->d_inode = inode;
800 spin_unlock(&dcache_lock);
801 security_d_instantiate(entry, inode);
805 * d_instantiate_unique - instantiate a non-aliased dentry
806 * @entry: dentry to instantiate
807 * @inode: inode to attach to this dentry
809 * Fill in inode information in the entry. On success, it returns NULL.
810 * If an unhashed alias of "entry" already exists, then we return the
811 * aliased dentry instead.
813 * Note that in order to avoid conflicts with rename() etc, the caller
814 * had better be holding the parent directory semaphore.
816 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
818 struct dentry *alias;
819 int len = entry->d_name.len;
820 const char *name = entry->d_name.name;
821 unsigned int hash = entry->d_name.hash;
823 BUG_ON(!list_empty(&entry->d_alias));
824 spin_lock(&dcache_lock);
827 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
828 struct qstr *qstr = &alias->d_name;
830 if (qstr->hash != hash)
832 if (alias->d_parent != entry->d_parent)
834 if (qstr->len != len)
836 if (memcmp(qstr->name, name, len))
839 spin_unlock(&dcache_lock);
840 BUG_ON(!d_unhashed(alias));
843 list_add(&entry->d_alias, &inode->i_dentry);
845 entry->d_inode = inode;
846 spin_unlock(&dcache_lock);
847 security_d_instantiate(entry, inode);
850 EXPORT_SYMBOL(d_instantiate_unique);
853 * d_alloc_root - allocate root dentry
854 * @root_inode: inode to allocate the root for
856 * Allocate a root ("/") dentry for the inode given. The inode is
857 * instantiated and returned. %NULL is returned if there is insufficient
858 * memory or the inode passed is %NULL.
861 struct dentry * d_alloc_root(struct inode * root_inode)
863 struct dentry *res = NULL;
866 static const struct qstr name = { .name = "/", .len = 1 };
868 res = d_alloc(NULL, &name);
870 res->d_sb = root_inode->i_sb;
872 d_instantiate(res, root_inode);
878 static inline struct hlist_head *d_hash(struct dentry *parent,
881 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
882 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
883 return dentry_hashtable + (hash & D_HASHMASK);
887 * d_alloc_anon - allocate an anonymous dentry
888 * @inode: inode to allocate the dentry for
890 * This is similar to d_alloc_root. It is used by filesystems when
891 * creating a dentry for a given inode, often in the process of
892 * mapping a filehandle to a dentry. The returned dentry may be
893 * anonymous, or may have a full name (if the inode was already
894 * in the cache). The file system may need to make further
895 * efforts to connect this dentry into the dcache properly.
897 * When called on a directory inode, we must ensure that
898 * the inode only ever has one dentry. If a dentry is
899 * found, that is returned instead of allocating a new one.
901 * On successful return, the reference to the inode has been transferred
902 * to the dentry. If %NULL is returned (indicating kmalloc failure),
903 * the reference on the inode has not been released.
906 struct dentry * d_alloc_anon(struct inode *inode)
908 static const struct qstr anonstring = { .name = "" };
912 if ((res = d_find_alias(inode))) {
917 tmp = d_alloc(NULL, &anonstring);
921 tmp->d_parent = tmp; /* make sure dput doesn't croak */
923 spin_lock(&dcache_lock);
924 res = __d_find_alias(inode, 0);
926 /* attach a disconnected dentry */
929 spin_lock(&res->d_lock);
930 res->d_sb = inode->i_sb;
932 res->d_inode = inode;
933 res->d_flags |= DCACHE_DISCONNECTED;
934 res->d_flags &= ~DCACHE_UNHASHED;
935 list_add(&res->d_alias, &inode->i_dentry);
936 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
937 spin_unlock(&res->d_lock);
939 inode = NULL; /* don't drop reference */
941 spin_unlock(&dcache_lock);
952 * d_splice_alias - splice a disconnected dentry into the tree if one exists
953 * @inode: the inode which may have a disconnected dentry
954 * @dentry: a negative dentry which we want to point to the inode.
956 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
957 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
958 * and return it, else simply d_add the inode to the dentry and return NULL.
960 * This is needed in the lookup routine of any filesystem that is exportable
961 * (via knfsd) so that we can build dcache paths to directories effectively.
963 * If a dentry was found and moved, then it is returned. Otherwise NULL
964 * is returned. This matches the expected return value of ->lookup.
967 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
969 struct dentry *new = NULL;
972 spin_lock(&dcache_lock);
973 new = __d_find_alias(inode, 1);
975 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
976 spin_unlock(&dcache_lock);
977 security_d_instantiate(new, inode);
982 /* d_instantiate takes dcache_lock, so we do it by hand */
983 list_add(&dentry->d_alias, &inode->i_dentry);
984 dentry->d_inode = inode;
985 spin_unlock(&dcache_lock);
986 security_d_instantiate(dentry, inode);
990 d_add(dentry, inode);
996 * d_lookup - search for a dentry
997 * @parent: parent dentry
998 * @name: qstr of name we wish to find
1000 * Searches the children of the parent dentry for the name in question. If
1001 * the dentry is found its reference count is incremented and the dentry
1002 * is returned. The caller must use d_put to free the entry when it has
1003 * finished using it. %NULL is returned on failure.
1005 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1006 * Memory barriers are used while updating and doing lockless traversal.
1007 * To avoid races with d_move while rename is happening, d_lock is used.
1009 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1010 * and name pointer in one structure pointed by d_qstr.
1012 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1013 * lookup is going on.
1015 * dentry_unused list is not updated even if lookup finds the required dentry
1016 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1017 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1020 * d_lookup() is protected against the concurrent renames in some unrelated
1021 * directory using the seqlockt_t rename_lock.
1024 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1026 struct dentry * dentry = NULL;
1030 seq = read_seqbegin(&rename_lock);
1031 dentry = __d_lookup(parent, name);
1034 } while (read_seqretry(&rename_lock, seq));
1038 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1040 unsigned int len = name->len;
1041 unsigned int hash = name->hash;
1042 const unsigned char *str = name->name;
1043 struct hlist_head *head = d_hash(parent,hash);
1044 struct dentry *found = NULL;
1045 struct hlist_node *node;
1046 struct dentry *dentry;
1050 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1053 if (dentry->d_name.hash != hash)
1055 if (dentry->d_parent != parent)
1058 spin_lock(&dentry->d_lock);
1061 * Recheck the dentry after taking the lock - d_move may have
1062 * changed things. Don't bother checking the hash because we're
1063 * about to compare the whole name anyway.
1065 if (dentry->d_parent != parent)
1069 * It is safe to compare names since d_move() cannot
1070 * change the qstr (protected by d_lock).
1072 qstr = &dentry->d_name;
1073 if (parent->d_op && parent->d_op->d_compare) {
1074 if (parent->d_op->d_compare(parent, qstr, name))
1077 if (qstr->len != len)
1079 if (memcmp(qstr->name, str, len))
1083 if (!d_unhashed(dentry)) {
1084 atomic_inc(&dentry->d_count);
1087 spin_unlock(&dentry->d_lock);
1090 spin_unlock(&dentry->d_lock);
1098 * d_validate - verify dentry provided from insecure source
1099 * @dentry: The dentry alleged to be valid child of @dparent
1100 * @dparent: The parent dentry (known to be valid)
1101 * @hash: Hash of the dentry
1102 * @len: Length of the name
1104 * An insecure source has sent us a dentry, here we verify it and dget() it.
1105 * This is used by ncpfs in its readdir implementation.
1106 * Zero is returned in the dentry is invalid.
1109 int d_validate(struct dentry *dentry, struct dentry *dparent)
1111 struct hlist_head *base;
1112 struct hlist_node *lhp;
1114 /* Check whether the ptr might be valid at all.. */
1115 if (!kmem_ptr_validate(dentry_cache, dentry))
1118 if (dentry->d_parent != dparent)
1121 spin_lock(&dcache_lock);
1122 base = d_hash(dparent, dentry->d_name.hash);
1123 hlist_for_each(lhp,base) {
1124 /* hlist_for_each_entry_rcu() not required for d_hash list
1125 * as it is parsed under dcache_lock
1127 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1128 __dget_locked(dentry);
1129 spin_unlock(&dcache_lock);
1133 spin_unlock(&dcache_lock);
1139 * When a file is deleted, we have two options:
1140 * - turn this dentry into a negative dentry
1141 * - unhash this dentry and free it.
1143 * Usually, we want to just turn this into
1144 * a negative dentry, but if anybody else is
1145 * currently using the dentry or the inode
1146 * we can't do that and we fall back on removing
1147 * it from the hash queues and waiting for
1148 * it to be deleted later when it has no users
1152 * d_delete - delete a dentry
1153 * @dentry: The dentry to delete
1155 * Turn the dentry into a negative dentry if possible, otherwise
1156 * remove it from the hash queues so it can be deleted later
1159 void d_delete(struct dentry * dentry)
1163 * Are we the only user?
1165 spin_lock(&dcache_lock);
1166 spin_lock(&dentry->d_lock);
1167 isdir = S_ISDIR(dentry->d_inode->i_mode);
1168 if (atomic_read(&dentry->d_count) == 1) {
1169 dentry_iput(dentry);
1170 fsnotify_nameremove(dentry, isdir);
1174 if (!d_unhashed(dentry))
1177 spin_unlock(&dentry->d_lock);
1178 spin_unlock(&dcache_lock);
1180 fsnotify_nameremove(dentry, isdir);
1183 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1186 entry->d_flags &= ~DCACHE_UNHASHED;
1187 hlist_add_head_rcu(&entry->d_hash, list);
1191 * d_rehash - add an entry back to the hash
1192 * @entry: dentry to add to the hash
1194 * Adds a dentry to the hash according to its name.
1197 void d_rehash(struct dentry * entry)
1199 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1201 spin_lock(&dcache_lock);
1202 spin_lock(&entry->d_lock);
1203 __d_rehash(entry, list);
1204 spin_unlock(&entry->d_lock);
1205 spin_unlock(&dcache_lock);
1208 #define do_switch(x,y) do { \
1209 __typeof__ (x) __tmp = x; \
1210 x = y; y = __tmp; } while (0)
1213 * When switching names, the actual string doesn't strictly have to
1214 * be preserved in the target - because we're dropping the target
1215 * anyway. As such, we can just do a simple memcpy() to copy over
1216 * the new name before we switch.
1218 * Note that we have to be a lot more careful about getting the hash
1219 * switched - we have to switch the hash value properly even if it
1220 * then no longer matches the actual (corrupted) string of the target.
1221 * The hash value has to match the hash queue that the dentry is on..
1223 static void switch_names(struct dentry *dentry, struct dentry *target)
1225 if (dname_external(target)) {
1226 if (dname_external(dentry)) {
1228 * Both external: swap the pointers
1230 do_switch(target->d_name.name, dentry->d_name.name);
1233 * dentry:internal, target:external. Steal target's
1234 * storage and make target internal.
1236 dentry->d_name.name = target->d_name.name;
1237 target->d_name.name = target->d_iname;
1240 if (dname_external(dentry)) {
1242 * dentry:external, target:internal. Give dentry's
1243 * storage to target and make dentry internal
1245 memcpy(dentry->d_iname, target->d_name.name,
1246 target->d_name.len + 1);
1247 target->d_name.name = dentry->d_name.name;
1248 dentry->d_name.name = dentry->d_iname;
1251 * Both are internal. Just copy target to dentry
1253 memcpy(dentry->d_iname, target->d_name.name,
1254 target->d_name.len + 1);
1260 * We cannibalize "target" when moving dentry on top of it,
1261 * because it's going to be thrown away anyway. We could be more
1262 * polite about it, though.
1264 * This forceful removal will result in ugly /proc output if
1265 * somebody holds a file open that got deleted due to a rename.
1266 * We could be nicer about the deleted file, and let it show
1267 * up under the name it got deleted rather than the name that
1272 * d_move - move a dentry
1273 * @dentry: entry to move
1274 * @target: new dentry
1276 * Update the dcache to reflect the move of a file name. Negative
1277 * dcache entries should not be moved in this way.
1280 void d_move(struct dentry * dentry, struct dentry * target)
1282 struct hlist_head *list;
1284 if (!dentry->d_inode)
1285 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1287 spin_lock(&dcache_lock);
1288 write_seqlock(&rename_lock);
1290 * XXXX: do we really need to take target->d_lock?
1292 if (target < dentry) {
1293 spin_lock(&target->d_lock);
1294 spin_lock(&dentry->d_lock);
1296 spin_lock(&dentry->d_lock);
1297 spin_lock(&target->d_lock);
1300 /* Move the dentry to the target hash queue, if on different bucket */
1301 if (dentry->d_flags & DCACHE_UNHASHED)
1302 goto already_unhashed;
1304 hlist_del_rcu(&dentry->d_hash);
1307 list = d_hash(target->d_parent, target->d_name.hash);
1308 __d_rehash(dentry, list);
1310 /* Unhash the target: dput() will then get rid of it */
1313 list_del(&dentry->d_u.d_child);
1314 list_del(&target->d_u.d_child);
1316 /* Switch the names.. */
1317 switch_names(dentry, target);
1318 do_switch(dentry->d_name.len, target->d_name.len);
1319 do_switch(dentry->d_name.hash, target->d_name.hash);
1321 /* ... and switch the parents */
1322 if (IS_ROOT(dentry)) {
1323 dentry->d_parent = target->d_parent;
1324 target->d_parent = target;
1325 INIT_LIST_HEAD(&target->d_u.d_child);
1327 do_switch(dentry->d_parent, target->d_parent);
1329 /* And add them back to the (new) parent lists */
1330 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1333 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1334 spin_unlock(&target->d_lock);
1335 spin_unlock(&dentry->d_lock);
1336 write_sequnlock(&rename_lock);
1337 spin_unlock(&dcache_lock);
1341 * d_path - return the path of a dentry
1342 * @dentry: dentry to report
1343 * @vfsmnt: vfsmnt to which the dentry belongs
1344 * @root: root dentry
1345 * @rootmnt: vfsmnt to which the root dentry belongs
1346 * @buffer: buffer to return value in
1347 * @buflen: buffer length
1349 * Convert a dentry into an ASCII path name. If the entry has been deleted
1350 * the string " (deleted)" is appended. Note that this is ambiguous.
1352 * Returns the buffer or an error code if the path was too long.
1354 * "buflen" should be positive. Caller holds the dcache_lock.
1356 static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1357 struct dentry *root, struct vfsmount *rootmnt,
1358 char *buffer, int buflen)
1360 char * end = buffer+buflen;
1366 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1371 memcpy(end, " (deleted)", 10);
1381 struct dentry * parent;
1383 if (dentry == root && vfsmnt == rootmnt)
1385 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1387 spin_lock(&vfsmount_lock);
1388 if (vfsmnt->mnt_parent == vfsmnt) {
1389 spin_unlock(&vfsmount_lock);
1392 dentry = vfsmnt->mnt_mountpoint;
1393 vfsmnt = vfsmnt->mnt_parent;
1394 spin_unlock(&vfsmount_lock);
1397 parent = dentry->d_parent;
1399 namelen = dentry->d_name.len;
1400 buflen -= namelen + 1;
1404 memcpy(end, dentry->d_name.name, namelen);
1413 namelen = dentry->d_name.len;
1417 retval -= namelen-1; /* hit the slash */
1418 memcpy(retval, dentry->d_name.name, namelen);
1421 return ERR_PTR(-ENAMETOOLONG);
1424 /* write full pathname into buffer and return start of pathname */
1425 char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1426 char *buf, int buflen)
1429 struct vfsmount *rootmnt;
1430 struct dentry *root;
1432 read_lock(¤t->fs->lock);
1433 rootmnt = mntget(current->fs->rootmnt);
1434 root = dget(current->fs->root);
1435 read_unlock(¤t->fs->lock);
1436 spin_lock(&dcache_lock);
1437 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1438 spin_unlock(&dcache_lock);
1445 * NOTE! The user-level library version returns a
1446 * character pointer. The kernel system call just
1447 * returns the length of the buffer filled (which
1448 * includes the ending '\0' character), or a negative
1449 * error value. So libc would do something like
1451 * char *getcwd(char * buf, size_t size)
1455 * retval = sys_getcwd(buf, size);
1462 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1465 struct vfsmount *pwdmnt, *rootmnt;
1466 struct dentry *pwd, *root;
1467 char *page = (char *) __get_free_page(GFP_USER);
1472 read_lock(¤t->fs->lock);
1473 pwdmnt = mntget(current->fs->pwdmnt);
1474 pwd = dget(current->fs->pwd);
1475 rootmnt = mntget(current->fs->rootmnt);
1476 root = dget(current->fs->root);
1477 read_unlock(¤t->fs->lock);
1480 /* Has the current directory has been unlinked? */
1481 spin_lock(&dcache_lock);
1482 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1486 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1487 spin_unlock(&dcache_lock);
1489 error = PTR_ERR(cwd);
1494 len = PAGE_SIZE + page - cwd;
1497 if (copy_to_user(buf, cwd, len))
1501 spin_unlock(&dcache_lock);
1508 free_page((unsigned long) page);
1513 * Test whether new_dentry is a subdirectory of old_dentry.
1515 * Trivially implemented using the dcache structure
1519 * is_subdir - is new dentry a subdirectory of old_dentry
1520 * @new_dentry: new dentry
1521 * @old_dentry: old dentry
1523 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1524 * Returns 0 otherwise.
1525 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1528 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1531 struct dentry * saved = new_dentry;
1534 /* need rcu_readlock to protect against the d_parent trashing due to
1539 /* for restarting inner loop in case of seq retry */
1542 seq = read_seqbegin(&rename_lock);
1544 if (new_dentry != old_dentry) {
1545 struct dentry * parent = new_dentry->d_parent;
1546 if (parent == new_dentry)
1548 new_dentry = parent;
1554 } while (read_seqretry(&rename_lock, seq));
1560 void d_genocide(struct dentry *root)
1562 struct dentry *this_parent = root;
1563 struct list_head *next;
1565 spin_lock(&dcache_lock);
1567 next = this_parent->d_subdirs.next;
1569 while (next != &this_parent->d_subdirs) {
1570 struct list_head *tmp = next;
1571 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1573 if (d_unhashed(dentry)||!dentry->d_inode)
1575 if (!list_empty(&dentry->d_subdirs)) {
1576 this_parent = dentry;
1579 atomic_dec(&dentry->d_count);
1581 if (this_parent != root) {
1582 next = this_parent->d_u.d_child.next;
1583 atomic_dec(&this_parent->d_count);
1584 this_parent = this_parent->d_parent;
1587 spin_unlock(&dcache_lock);
1591 * find_inode_number - check for dentry with name
1592 * @dir: directory to check
1593 * @name: Name to find.
1595 * Check whether a dentry already exists for the given name,
1596 * and return the inode number if it has an inode. Otherwise
1599 * This routine is used to post-process directory listings for
1600 * filesystems using synthetic inode numbers, and is necessary
1601 * to keep getcwd() working.
1604 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1606 struct dentry * dentry;
1610 * Check for a fs-specific hash function. Note that we must
1611 * calculate the standard hash first, as the d_op->d_hash()
1612 * routine may choose to leave the hash value unchanged.
1614 name->hash = full_name_hash(name->name, name->len);
1615 if (dir->d_op && dir->d_op->d_hash)
1617 if (dir->d_op->d_hash(dir, name) != 0)
1621 dentry = d_lookup(dir, name);
1624 if (dentry->d_inode)
1625 ino = dentry->d_inode->i_ino;
1632 static __initdata unsigned long dhash_entries;
1633 static int __init set_dhash_entries(char *str)
1637 dhash_entries = simple_strtoul(str, &str, 0);
1640 __setup("dhash_entries=", set_dhash_entries);
1642 static void __init dcache_init_early(void)
1646 /* If hashes are distributed across NUMA nodes, defer
1647 * hash allocation until vmalloc space is available.
1653 alloc_large_system_hash("Dentry cache",
1654 sizeof(struct hlist_head),
1662 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1663 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1666 static void __init dcache_init(unsigned long mempages)
1671 * A constructor could be added for stable state like the lists,
1672 * but it is probably not worth it because of the cache nature
1675 dentry_cache = kmem_cache_create("dentry_cache",
1676 sizeof(struct dentry),
1678 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1681 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1683 /* Hash may have been set up in dcache_init_early */
1688 alloc_large_system_hash("Dentry cache",
1689 sizeof(struct hlist_head),
1697 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1698 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1701 /* SLAB cache for __getname() consumers */
1702 kmem_cache_t *names_cachep;
1704 /* SLAB cache for file structures */
1705 kmem_cache_t *filp_cachep;
1707 EXPORT_SYMBOL(d_genocide);
1709 extern void bdev_cache_init(void);
1710 extern void chrdev_init(void);
1712 void __init vfs_caches_init_early(void)
1714 dcache_init_early();
1718 void __init vfs_caches_init(unsigned long mempages)
1720 unsigned long reserve;
1722 /* Base hash sizes on available memory, with a reserve equal to
1723 150% of current kernel size */
1725 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1726 mempages -= reserve;
1728 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1729 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1731 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1732 SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
1734 dcache_init(mempages);
1735 inode_init(mempages);
1736 files_init(mempages);
1742 EXPORT_SYMBOL(d_alloc);
1743 EXPORT_SYMBOL(d_alloc_anon);
1744 EXPORT_SYMBOL(d_alloc_root);
1745 EXPORT_SYMBOL(d_delete);
1746 EXPORT_SYMBOL(d_find_alias);
1747 EXPORT_SYMBOL(d_instantiate);
1748 EXPORT_SYMBOL(d_invalidate);
1749 EXPORT_SYMBOL(d_lookup);
1750 EXPORT_SYMBOL(d_move);
1751 EXPORT_SYMBOL(d_path);
1752 EXPORT_SYMBOL(d_prune_aliases);
1753 EXPORT_SYMBOL(d_rehash);
1754 EXPORT_SYMBOL(d_splice_alias);
1755 EXPORT_SYMBOL(d_validate);
1756 EXPORT_SYMBOL(dget_locked);
1757 EXPORT_SYMBOL(dput);
1758 EXPORT_SYMBOL(find_inode_number);
1759 EXPORT_SYMBOL(have_submounts);
1760 EXPORT_SYMBOL(names_cachep);
1761 EXPORT_SYMBOL(shrink_dcache_parent);
1762 EXPORT_SYMBOL(shrink_dcache_sb);