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 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 and drop one reference to inode.
813 * Note that in order to avoid conflicts with rename() etc, the caller
814 * had better be holding the parent directory semaphore.
816 * This also assumes that the inode count has been incremented
817 * (or otherwise set) by the caller to indicate that it is now
818 * in use by the dcache.
820 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
822 struct dentry *alias;
823 int len = entry->d_name.len;
824 const char *name = entry->d_name.name;
825 unsigned int hash = entry->d_name.hash;
827 BUG_ON(!list_empty(&entry->d_alias));
828 spin_lock(&dcache_lock);
831 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
832 struct qstr *qstr = &alias->d_name;
834 if (qstr->hash != hash)
836 if (alias->d_parent != entry->d_parent)
838 if (qstr->len != len)
840 if (memcmp(qstr->name, name, len))
843 spin_unlock(&dcache_lock);
844 BUG_ON(!d_unhashed(alias));
848 list_add(&entry->d_alias, &inode->i_dentry);
850 entry->d_inode = inode;
851 spin_unlock(&dcache_lock);
852 security_d_instantiate(entry, inode);
855 EXPORT_SYMBOL(d_instantiate_unique);
858 * d_alloc_root - allocate root dentry
859 * @root_inode: inode to allocate the root for
861 * Allocate a root ("/") dentry for the inode given. The inode is
862 * instantiated and returned. %NULL is returned if there is insufficient
863 * memory or the inode passed is %NULL.
866 struct dentry * d_alloc_root(struct inode * root_inode)
868 struct dentry *res = NULL;
871 static const struct qstr name = { .name = "/", .len = 1 };
873 res = d_alloc(NULL, &name);
875 res->d_sb = root_inode->i_sb;
877 d_instantiate(res, root_inode);
883 static inline struct hlist_head *d_hash(struct dentry *parent,
886 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
887 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
888 return dentry_hashtable + (hash & D_HASHMASK);
892 * d_alloc_anon - allocate an anonymous dentry
893 * @inode: inode to allocate the dentry for
895 * This is similar to d_alloc_root. It is used by filesystems when
896 * creating a dentry for a given inode, often in the process of
897 * mapping a filehandle to a dentry. The returned dentry may be
898 * anonymous, or may have a full name (if the inode was already
899 * in the cache). The file system may need to make further
900 * efforts to connect this dentry into the dcache properly.
902 * When called on a directory inode, we must ensure that
903 * the inode only ever has one dentry. If a dentry is
904 * found, that is returned instead of allocating a new one.
906 * On successful return, the reference to the inode has been transferred
907 * to the dentry. If %NULL is returned (indicating kmalloc failure),
908 * the reference on the inode has not been released.
911 struct dentry * d_alloc_anon(struct inode *inode)
913 static const struct qstr anonstring = { .name = "" };
917 if ((res = d_find_alias(inode))) {
922 tmp = d_alloc(NULL, &anonstring);
926 tmp->d_parent = tmp; /* make sure dput doesn't croak */
928 spin_lock(&dcache_lock);
929 res = __d_find_alias(inode, 0);
931 /* attach a disconnected dentry */
934 spin_lock(&res->d_lock);
935 res->d_sb = inode->i_sb;
937 res->d_inode = inode;
938 res->d_flags |= DCACHE_DISCONNECTED;
939 res->d_flags &= ~DCACHE_UNHASHED;
940 list_add(&res->d_alias, &inode->i_dentry);
941 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
942 spin_unlock(&res->d_lock);
944 inode = NULL; /* don't drop reference */
946 spin_unlock(&dcache_lock);
957 * d_splice_alias - splice a disconnected dentry into the tree if one exists
958 * @inode: the inode which may have a disconnected dentry
959 * @dentry: a negative dentry which we want to point to the inode.
961 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
962 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
963 * and return it, else simply d_add the inode to the dentry and return NULL.
965 * This is needed in the lookup routine of any filesystem that is exportable
966 * (via knfsd) so that we can build dcache paths to directories effectively.
968 * If a dentry was found and moved, then it is returned. Otherwise NULL
969 * is returned. This matches the expected return value of ->lookup.
972 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
974 struct dentry *new = NULL;
977 spin_lock(&dcache_lock);
978 new = __d_find_alias(inode, 1);
980 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
981 spin_unlock(&dcache_lock);
982 security_d_instantiate(new, inode);
987 /* d_instantiate takes dcache_lock, so we do it by hand */
988 list_add(&dentry->d_alias, &inode->i_dentry);
989 dentry->d_inode = inode;
990 spin_unlock(&dcache_lock);
991 security_d_instantiate(dentry, inode);
995 d_add(dentry, inode);
1001 * d_lookup - search for a dentry
1002 * @parent: parent dentry
1003 * @name: qstr of name we wish to find
1005 * Searches the children of the parent dentry for the name in question. If
1006 * the dentry is found its reference count is incremented and the dentry
1007 * is returned. The caller must use d_put to free the entry when it has
1008 * finished using it. %NULL is returned on failure.
1010 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1011 * Memory barriers are used while updating and doing lockless traversal.
1012 * To avoid races with d_move while rename is happening, d_lock is used.
1014 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1015 * and name pointer in one structure pointed by d_qstr.
1017 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1018 * lookup is going on.
1020 * dentry_unused list is not updated even if lookup finds the required dentry
1021 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1022 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1025 * d_lookup() is protected against the concurrent renames in some unrelated
1026 * directory using the seqlockt_t rename_lock.
1029 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1031 struct dentry * dentry = NULL;
1035 seq = read_seqbegin(&rename_lock);
1036 dentry = __d_lookup(parent, name);
1039 } while (read_seqretry(&rename_lock, seq));
1043 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1045 unsigned int len = name->len;
1046 unsigned int hash = name->hash;
1047 const unsigned char *str = name->name;
1048 struct hlist_head *head = d_hash(parent,hash);
1049 struct dentry *found = NULL;
1050 struct hlist_node *node;
1051 struct dentry *dentry;
1055 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1058 if (dentry->d_name.hash != hash)
1060 if (dentry->d_parent != parent)
1063 spin_lock(&dentry->d_lock);
1066 * Recheck the dentry after taking the lock - d_move may have
1067 * changed things. Don't bother checking the hash because we're
1068 * about to compare the whole name anyway.
1070 if (dentry->d_parent != parent)
1074 * It is safe to compare names since d_move() cannot
1075 * change the qstr (protected by d_lock).
1077 qstr = &dentry->d_name;
1078 if (parent->d_op && parent->d_op->d_compare) {
1079 if (parent->d_op->d_compare(parent, qstr, name))
1082 if (qstr->len != len)
1084 if (memcmp(qstr->name, str, len))
1088 if (!d_unhashed(dentry)) {
1089 atomic_inc(&dentry->d_count);
1092 spin_unlock(&dentry->d_lock);
1095 spin_unlock(&dentry->d_lock);
1103 * d_validate - verify dentry provided from insecure source
1104 * @dentry: The dentry alleged to be valid child of @dparent
1105 * @dparent: The parent dentry (known to be valid)
1106 * @hash: Hash of the dentry
1107 * @len: Length of the name
1109 * An insecure source has sent us a dentry, here we verify it and dget() it.
1110 * This is used by ncpfs in its readdir implementation.
1111 * Zero is returned in the dentry is invalid.
1114 int d_validate(struct dentry *dentry, struct dentry *dparent)
1116 struct hlist_head *base;
1117 struct hlist_node *lhp;
1119 /* Check whether the ptr might be valid at all.. */
1120 if (!kmem_ptr_validate(dentry_cache, dentry))
1123 if (dentry->d_parent != dparent)
1126 spin_lock(&dcache_lock);
1127 base = d_hash(dparent, dentry->d_name.hash);
1128 hlist_for_each(lhp,base) {
1129 /* hlist_for_each_entry_rcu() not required for d_hash list
1130 * as it is parsed under dcache_lock
1132 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1133 __dget_locked(dentry);
1134 spin_unlock(&dcache_lock);
1138 spin_unlock(&dcache_lock);
1144 * When a file is deleted, we have two options:
1145 * - turn this dentry into a negative dentry
1146 * - unhash this dentry and free it.
1148 * Usually, we want to just turn this into
1149 * a negative dentry, but if anybody else is
1150 * currently using the dentry or the inode
1151 * we can't do that and we fall back on removing
1152 * it from the hash queues and waiting for
1153 * it to be deleted later when it has no users
1157 * d_delete - delete a dentry
1158 * @dentry: The dentry to delete
1160 * Turn the dentry into a negative dentry if possible, otherwise
1161 * remove it from the hash queues so it can be deleted later
1164 void d_delete(struct dentry * dentry)
1168 * Are we the only user?
1170 spin_lock(&dcache_lock);
1171 spin_lock(&dentry->d_lock);
1172 isdir = S_ISDIR(dentry->d_inode->i_mode);
1173 if (atomic_read(&dentry->d_count) == 1) {
1174 dentry_iput(dentry);
1175 fsnotify_nameremove(dentry, isdir);
1179 if (!d_unhashed(dentry))
1182 spin_unlock(&dentry->d_lock);
1183 spin_unlock(&dcache_lock);
1185 fsnotify_nameremove(dentry, isdir);
1188 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1191 entry->d_flags &= ~DCACHE_UNHASHED;
1192 hlist_add_head_rcu(&entry->d_hash, list);
1196 * d_rehash - add an entry back to the hash
1197 * @entry: dentry to add to the hash
1199 * Adds a dentry to the hash according to its name.
1202 void d_rehash(struct dentry * entry)
1204 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1206 spin_lock(&dcache_lock);
1207 spin_lock(&entry->d_lock);
1208 __d_rehash(entry, list);
1209 spin_unlock(&entry->d_lock);
1210 spin_unlock(&dcache_lock);
1213 #define do_switch(x,y) do { \
1214 __typeof__ (x) __tmp = x; \
1215 x = y; y = __tmp; } while (0)
1218 * When switching names, the actual string doesn't strictly have to
1219 * be preserved in the target - because we're dropping the target
1220 * anyway. As such, we can just do a simple memcpy() to copy over
1221 * the new name before we switch.
1223 * Note that we have to be a lot more careful about getting the hash
1224 * switched - we have to switch the hash value properly even if it
1225 * then no longer matches the actual (corrupted) string of the target.
1226 * The hash value has to match the hash queue that the dentry is on..
1228 static void switch_names(struct dentry *dentry, struct dentry *target)
1230 if (dname_external(target)) {
1231 if (dname_external(dentry)) {
1233 * Both external: swap the pointers
1235 do_switch(target->d_name.name, dentry->d_name.name);
1238 * dentry:internal, target:external. Steal target's
1239 * storage and make target internal.
1241 dentry->d_name.name = target->d_name.name;
1242 target->d_name.name = target->d_iname;
1245 if (dname_external(dentry)) {
1247 * dentry:external, target:internal. Give dentry's
1248 * storage to target and make dentry internal
1250 memcpy(dentry->d_iname, target->d_name.name,
1251 target->d_name.len + 1);
1252 target->d_name.name = dentry->d_name.name;
1253 dentry->d_name.name = dentry->d_iname;
1256 * Both are internal. Just copy target to dentry
1258 memcpy(dentry->d_iname, target->d_name.name,
1259 target->d_name.len + 1);
1265 * We cannibalize "target" when moving dentry on top of it,
1266 * because it's going to be thrown away anyway. We could be more
1267 * polite about it, though.
1269 * This forceful removal will result in ugly /proc output if
1270 * somebody holds a file open that got deleted due to a rename.
1271 * We could be nicer about the deleted file, and let it show
1272 * up under the name it got deleted rather than the name that
1277 * d_move - move a dentry
1278 * @dentry: entry to move
1279 * @target: new dentry
1281 * Update the dcache to reflect the move of a file name. Negative
1282 * dcache entries should not be moved in this way.
1285 void d_move(struct dentry * dentry, struct dentry * target)
1287 struct hlist_head *list;
1289 if (!dentry->d_inode)
1290 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1292 spin_lock(&dcache_lock);
1293 write_seqlock(&rename_lock);
1295 * XXXX: do we really need to take target->d_lock?
1297 if (target < dentry) {
1298 spin_lock(&target->d_lock);
1299 spin_lock(&dentry->d_lock);
1301 spin_lock(&dentry->d_lock);
1302 spin_lock(&target->d_lock);
1305 /* Move the dentry to the target hash queue, if on different bucket */
1306 if (dentry->d_flags & DCACHE_UNHASHED)
1307 goto already_unhashed;
1309 hlist_del_rcu(&dentry->d_hash);
1312 list = d_hash(target->d_parent, target->d_name.hash);
1313 __d_rehash(dentry, list);
1315 /* Unhash the target: dput() will then get rid of it */
1318 list_del(&dentry->d_u.d_child);
1319 list_del(&target->d_u.d_child);
1321 /* Switch the names.. */
1322 switch_names(dentry, target);
1323 do_switch(dentry->d_name.len, target->d_name.len);
1324 do_switch(dentry->d_name.hash, target->d_name.hash);
1326 /* ... and switch the parents */
1327 if (IS_ROOT(dentry)) {
1328 dentry->d_parent = target->d_parent;
1329 target->d_parent = target;
1330 INIT_LIST_HEAD(&target->d_u.d_child);
1332 do_switch(dentry->d_parent, target->d_parent);
1334 /* And add them back to the (new) parent lists */
1335 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1338 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1339 spin_unlock(&target->d_lock);
1340 spin_unlock(&dentry->d_lock);
1341 write_sequnlock(&rename_lock);
1342 spin_unlock(&dcache_lock);
1346 * d_path - return the path of a dentry
1347 * @dentry: dentry to report
1348 * @vfsmnt: vfsmnt to which the dentry belongs
1349 * @root: root dentry
1350 * @rootmnt: vfsmnt to which the root dentry belongs
1351 * @buffer: buffer to return value in
1352 * @buflen: buffer length
1354 * Convert a dentry into an ASCII path name. If the entry has been deleted
1355 * the string " (deleted)" is appended. Note that this is ambiguous.
1357 * Returns the buffer or an error code if the path was too long.
1359 * "buflen" should be positive. Caller holds the dcache_lock.
1361 static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1362 struct dentry *root, struct vfsmount *rootmnt,
1363 char *buffer, int buflen)
1365 char * end = buffer+buflen;
1371 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1376 memcpy(end, " (deleted)", 10);
1386 struct dentry * parent;
1388 if (dentry == root && vfsmnt == rootmnt)
1390 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1392 spin_lock(&vfsmount_lock);
1393 if (vfsmnt->mnt_parent == vfsmnt) {
1394 spin_unlock(&vfsmount_lock);
1397 dentry = vfsmnt->mnt_mountpoint;
1398 vfsmnt = vfsmnt->mnt_parent;
1399 spin_unlock(&vfsmount_lock);
1402 parent = dentry->d_parent;
1404 namelen = dentry->d_name.len;
1405 buflen -= namelen + 1;
1409 memcpy(end, dentry->d_name.name, namelen);
1418 namelen = dentry->d_name.len;
1422 retval -= namelen-1; /* hit the slash */
1423 memcpy(retval, dentry->d_name.name, namelen);
1426 return ERR_PTR(-ENAMETOOLONG);
1429 /* write full pathname into buffer and return start of pathname */
1430 char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1431 char *buf, int buflen)
1434 struct vfsmount *rootmnt;
1435 struct dentry *root;
1437 read_lock(¤t->fs->lock);
1438 rootmnt = mntget(current->fs->rootmnt);
1439 root = dget(current->fs->root);
1440 read_unlock(¤t->fs->lock);
1441 spin_lock(&dcache_lock);
1442 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1443 spin_unlock(&dcache_lock);
1450 * NOTE! The user-level library version returns a
1451 * character pointer. The kernel system call just
1452 * returns the length of the buffer filled (which
1453 * includes the ending '\0' character), or a negative
1454 * error value. So libc would do something like
1456 * char *getcwd(char * buf, size_t size)
1460 * retval = sys_getcwd(buf, size);
1467 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1470 struct vfsmount *pwdmnt, *rootmnt;
1471 struct dentry *pwd, *root;
1472 char *page = (char *) __get_free_page(GFP_USER);
1477 read_lock(¤t->fs->lock);
1478 pwdmnt = mntget(current->fs->pwdmnt);
1479 pwd = dget(current->fs->pwd);
1480 rootmnt = mntget(current->fs->rootmnt);
1481 root = dget(current->fs->root);
1482 read_unlock(¤t->fs->lock);
1485 /* Has the current directory has been unlinked? */
1486 spin_lock(&dcache_lock);
1487 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1491 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1492 spin_unlock(&dcache_lock);
1494 error = PTR_ERR(cwd);
1499 len = PAGE_SIZE + page - cwd;
1502 if (copy_to_user(buf, cwd, len))
1506 spin_unlock(&dcache_lock);
1513 free_page((unsigned long) page);
1518 * Test whether new_dentry is a subdirectory of old_dentry.
1520 * Trivially implemented using the dcache structure
1524 * is_subdir - is new dentry a subdirectory of old_dentry
1525 * @new_dentry: new dentry
1526 * @old_dentry: old dentry
1528 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1529 * Returns 0 otherwise.
1530 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1533 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1536 struct dentry * saved = new_dentry;
1539 /* need rcu_readlock to protect against the d_parent trashing due to
1544 /* for restarting inner loop in case of seq retry */
1547 seq = read_seqbegin(&rename_lock);
1549 if (new_dentry != old_dentry) {
1550 struct dentry * parent = new_dentry->d_parent;
1551 if (parent == new_dentry)
1553 new_dentry = parent;
1559 } while (read_seqretry(&rename_lock, seq));
1565 void d_genocide(struct dentry *root)
1567 struct dentry *this_parent = root;
1568 struct list_head *next;
1570 spin_lock(&dcache_lock);
1572 next = this_parent->d_subdirs.next;
1574 while (next != &this_parent->d_subdirs) {
1575 struct list_head *tmp = next;
1576 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1578 if (d_unhashed(dentry)||!dentry->d_inode)
1580 if (!list_empty(&dentry->d_subdirs)) {
1581 this_parent = dentry;
1584 atomic_dec(&dentry->d_count);
1586 if (this_parent != root) {
1587 next = this_parent->d_u.d_child.next;
1588 atomic_dec(&this_parent->d_count);
1589 this_parent = this_parent->d_parent;
1592 spin_unlock(&dcache_lock);
1596 * find_inode_number - check for dentry with name
1597 * @dir: directory to check
1598 * @name: Name to find.
1600 * Check whether a dentry already exists for the given name,
1601 * and return the inode number if it has an inode. Otherwise
1604 * This routine is used to post-process directory listings for
1605 * filesystems using synthetic inode numbers, and is necessary
1606 * to keep getcwd() working.
1609 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1611 struct dentry * dentry;
1615 * Check for a fs-specific hash function. Note that we must
1616 * calculate the standard hash first, as the d_op->d_hash()
1617 * routine may choose to leave the hash value unchanged.
1619 name->hash = full_name_hash(name->name, name->len);
1620 if (dir->d_op && dir->d_op->d_hash)
1622 if (dir->d_op->d_hash(dir, name) != 0)
1626 dentry = d_lookup(dir, name);
1629 if (dentry->d_inode)
1630 ino = dentry->d_inode->i_ino;
1637 static __initdata unsigned long dhash_entries;
1638 static int __init set_dhash_entries(char *str)
1642 dhash_entries = simple_strtoul(str, &str, 0);
1645 __setup("dhash_entries=", set_dhash_entries);
1647 static void __init dcache_init_early(void)
1651 /* If hashes are distributed across NUMA nodes, defer
1652 * hash allocation until vmalloc space is available.
1658 alloc_large_system_hash("Dentry cache",
1659 sizeof(struct hlist_head),
1667 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1668 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1671 static void __init dcache_init(unsigned long mempages)
1676 * A constructor could be added for stable state like the lists,
1677 * but it is probably not worth it because of the cache nature
1680 dentry_cache = kmem_cache_create("dentry_cache",
1681 sizeof(struct dentry),
1683 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1686 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1688 /* Hash may have been set up in dcache_init_early */
1693 alloc_large_system_hash("Dentry cache",
1694 sizeof(struct hlist_head),
1702 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1703 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1706 /* SLAB cache for __getname() consumers */
1707 kmem_cache_t *names_cachep;
1709 /* SLAB cache for file structures */
1710 kmem_cache_t *filp_cachep;
1712 EXPORT_SYMBOL(d_genocide);
1714 extern void bdev_cache_init(void);
1715 extern void chrdev_init(void);
1717 void __init vfs_caches_init_early(void)
1719 dcache_init_early();
1723 void __init vfs_caches_init(unsigned long mempages)
1725 unsigned long reserve;
1727 /* Base hash sizes on available memory, with a reserve equal to
1728 150% of current kernel size */
1730 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1731 mempages -= reserve;
1733 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1734 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1736 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1737 SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
1739 dcache_init(mempages);
1740 inode_init(mempages);
1741 files_init(mempages);
1747 EXPORT_SYMBOL(d_alloc);
1748 EXPORT_SYMBOL(d_alloc_anon);
1749 EXPORT_SYMBOL(d_alloc_root);
1750 EXPORT_SYMBOL(d_delete);
1751 EXPORT_SYMBOL(d_find_alias);
1752 EXPORT_SYMBOL(d_instantiate);
1753 EXPORT_SYMBOL(d_invalidate);
1754 EXPORT_SYMBOL(d_lookup);
1755 EXPORT_SYMBOL(d_move);
1756 EXPORT_SYMBOL(d_path);
1757 EXPORT_SYMBOL(d_prune_aliases);
1758 EXPORT_SYMBOL(d_rehash);
1759 EXPORT_SYMBOL(d_splice_alias);
1760 EXPORT_SYMBOL(d_validate);
1761 EXPORT_SYMBOL(dget_locked);
1762 EXPORT_SYMBOL(dput);
1763 EXPORT_SYMBOL(find_inode_number);
1764 EXPORT_SYMBOL(have_submounts);
1765 EXPORT_SYMBOL(names_cachep);
1766 EXPORT_SYMBOL(shrink_dcache_parent);
1767 EXPORT_SYMBOL(shrink_dcache_sb);