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/slab.h>
23 #include <linux/init.h>
24 #include <linux/smp_lock.h>
25 #include <linux/hash.h>
26 #include <linux/cache.h>
27 #include <linux/module.h>
28 #include <linux/mount.h>
29 #include <linux/file.h>
30 #include <asm/uaccess.h>
31 #include <linux/security.h>
32 #include <linux/seqlock.h>
33 #include <linux/swap.h>
34 #include <linux/bootmem.h>
36 /* #define DCACHE_DEBUG 1 */
38 int sysctl_vfs_cache_pressure = 100;
39 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
42 static seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
44 EXPORT_SYMBOL(dcache_lock);
46 static kmem_cache_t *dentry_cache;
48 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
58 #define D_HASHBITS d_hash_shift
59 #define D_HASHMASK d_hash_mask
61 static unsigned int d_hash_mask;
62 static unsigned int d_hash_shift;
63 static struct hlist_head *dentry_hashtable;
64 static LIST_HEAD(dentry_unused);
66 /* Statistics gathering. */
67 struct dentry_stat_t dentry_stat = {
71 static void d_callback(struct rcu_head *head)
73 struct dentry * dentry = container_of(head, struct dentry, d_rcu);
75 if (dname_external(dentry))
76 kfree(dentry->d_name.name);
77 kmem_cache_free(dentry_cache, dentry);
81 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
84 static void d_free(struct dentry *dentry)
86 if (dentry->d_op && dentry->d_op->d_release)
87 dentry->d_op->d_release(dentry);
88 if (dentry->d_extra_attributes) {
89 kfree(dentry->d_extra_attributes);
90 dentry->d_extra_attributes = NULL;
92 call_rcu(&dentry->d_rcu, d_callback);
96 * Release the dentry's inode, using the filesystem
97 * d_iput() operation if defined.
98 * Called with dcache_lock and per dentry lock held, drops both.
100 static inline void dentry_iput(struct dentry * dentry)
102 struct inode *inode = dentry->d_inode;
104 dentry->d_inode = NULL;
105 list_del_init(&dentry->d_alias);
106 spin_unlock(&dentry->d_lock);
107 spin_unlock(&dcache_lock);
108 if (dentry->d_op && dentry->d_op->d_iput)
109 dentry->d_op->d_iput(dentry, inode);
113 spin_unlock(&dentry->d_lock);
114 spin_unlock(&dcache_lock);
121 * This is complicated by the fact that we do not want to put
122 * dentries that are no longer on any hash chain on the unused
123 * list: we'd much rather just get rid of them immediately.
125 * However, that implies that we have to traverse the dentry
126 * tree upwards to the parents which might _also_ now be
127 * scheduled for deletion (it may have been only waiting for
128 * its last child to go away).
130 * This tail recursion is done by hand as we don't want to depend
131 * on the compiler to always get this right (gcc generally doesn't).
132 * Real recursion would eat up our stack space.
136 * dput - release a dentry
137 * @dentry: dentry to release
139 * Release a dentry. This will drop the usage count and if appropriate
140 * call the dentry unlink method as well as removing it from the queues and
141 * releasing its resources. If the parent dentries were scheduled for release
142 * they too may now get deleted.
144 * no dcache lock, please.
147 void dput(struct dentry *dentry)
153 if (atomic_read(&dentry->d_count) == 1)
155 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
158 spin_lock(&dentry->d_lock);
159 if (atomic_read(&dentry->d_count)) {
160 spin_unlock(&dentry->d_lock);
161 spin_unlock(&dcache_lock);
166 * AV: ->d_delete() is _NOT_ allowed to block now.
168 if (dentry->d_op && dentry->d_op->d_delete) {
169 if (dentry->d_op->d_delete(dentry))
172 /* Unreachable? Get rid of it */
173 if (d_unhashed(dentry))
175 if (list_empty(&dentry->d_lru)) {
176 dentry->d_flags |= DCACHE_REFERENCED;
177 list_add(&dentry->d_lru, &dentry_unused);
178 dentry_stat.nr_unused++;
180 spin_unlock(&dentry->d_lock);
181 spin_unlock(&dcache_lock);
188 struct dentry *parent;
190 /* If dentry was on d_lru list
191 * delete it from there
193 if (!list_empty(&dentry->d_lru)) {
194 list_del(&dentry->d_lru);
195 dentry_stat.nr_unused--;
197 list_del(&dentry->d_child);
198 dentry_stat.nr_dentry--; /* For d_free, below */
199 /*drops the locks, at that point nobody can reach this dentry */
201 parent = dentry->d_parent;
203 if (dentry == parent)
211 * d_invalidate - invalidate a dentry
212 * @dentry: dentry to invalidate
214 * Try to invalidate the dentry if it turns out to be
215 * possible. If there are other dentries that can be
216 * reached through this one we can't delete it and we
217 * return -EBUSY. On success we return 0.
222 int d_invalidate(struct dentry * dentry)
225 * If it's already been dropped, return OK.
227 spin_lock(&dcache_lock);
228 if (d_unhashed(dentry)) {
229 spin_unlock(&dcache_lock);
233 * Check whether to do a partial shrink_dcache
234 * to get rid of unused child entries.
236 if (!list_empty(&dentry->d_subdirs)) {
237 spin_unlock(&dcache_lock);
238 shrink_dcache_parent(dentry);
239 spin_lock(&dcache_lock);
243 * Somebody else still using it?
245 * If it's a directory, we can't drop it
246 * for fear of somebody re-populating it
247 * with children (even though dropping it
248 * would make it unreachable from the root,
249 * we might still populate it if it was a
250 * working directory or similar).
252 spin_lock(&dentry->d_lock);
253 if (atomic_read(&dentry->d_count) > 1) {
254 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
255 spin_unlock(&dentry->d_lock);
256 spin_unlock(&dcache_lock);
262 spin_unlock(&dentry->d_lock);
263 spin_unlock(&dcache_lock);
267 /* This should be called _only_ with dcache_lock held */
269 static inline struct dentry * __dget_locked(struct dentry *dentry)
271 atomic_inc(&dentry->d_count);
272 if (!list_empty(&dentry->d_lru)) {
273 dentry_stat.nr_unused--;
274 list_del_init(&dentry->d_lru);
279 struct dentry * dget_locked(struct dentry *dentry)
281 return __dget_locked(dentry);
285 * d_find_alias - grab a hashed alias of inode
286 * @inode: inode in question
287 * @want_discon: flag, used by d_splice_alias, to request
288 * that only a DISCONNECTED alias be returned.
290 * If inode has a hashed alias, or is a directory and has any alias,
291 * acquire the reference to alias and return it. Otherwise return NULL.
292 * Notice that if inode is a directory there can be only one alias and
293 * it can be unhashed only if it has no children, or if it is the root
296 * If the inode has a DCACHE_DISCONNECTED alias, then prefer
297 * any other hashed alias over that one unless @want_discon is set,
298 * in which case only return a DCACHE_DISCONNECTED alias.
301 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
303 struct list_head *head, *next, *tmp;
304 struct dentry *alias, *discon_alias=NULL;
306 head = &inode->i_dentry;
307 next = inode->i_dentry.next;
308 while (next != head) {
312 alias = list_entry(tmp, struct dentry, d_alias);
313 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
314 if (alias->d_flags & DCACHE_DISCONNECTED)
315 discon_alias = alias;
316 else if (!want_discon) {
317 __dget_locked(alias);
323 __dget_locked(discon_alias);
327 struct dentry * d_find_alias(struct inode *inode)
330 spin_lock(&dcache_lock);
331 de = __d_find_alias(inode, 0);
332 spin_unlock(&dcache_lock);
337 * Try to kill dentries associated with this inode.
338 * WARNING: you must own a reference to inode.
340 void d_prune_aliases(struct inode *inode)
342 struct list_head *tmp, *head = &inode->i_dentry;
344 spin_lock(&dcache_lock);
346 while ((tmp = tmp->next) != head) {
347 struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
348 spin_lock(&dentry->d_lock);
349 if (!atomic_read(&dentry->d_count)) {
350 __dget_locked(dentry);
352 spin_unlock(&dentry->d_lock);
353 spin_unlock(&dcache_lock);
357 spin_unlock(&dentry->d_lock);
359 spin_unlock(&dcache_lock);
363 * Throw away a dentry - free the inode, dput the parent.
364 * This requires that the LRU list has already been
366 * Called with dcache_lock, drops it and then regains.
368 static inline void prune_one_dentry(struct dentry * dentry)
370 struct dentry * parent;
373 list_del(&dentry->d_child);
374 dentry_stat.nr_dentry--; /* For d_free, below */
376 parent = dentry->d_parent;
378 if (parent != dentry)
380 spin_lock(&dcache_lock);
384 * prune_dcache - shrink the dcache
385 * @count: number of entries to try and free
387 * Shrink the dcache. This is done when we need
388 * more memory, or simply when we need to unmount
389 * something (at which point we need to unuse
392 * This function may fail to free any resources if
393 * all the dentries are in use.
396 static void prune_dcache(int count)
398 spin_lock(&dcache_lock);
399 for (; count ; count--) {
400 struct dentry *dentry;
401 struct list_head *tmp;
403 cond_resched_lock(&dcache_lock);
405 tmp = dentry_unused.prev;
406 if (tmp == &dentry_unused)
409 prefetch(dentry_unused.prev);
410 dentry_stat.nr_unused--;
411 dentry = list_entry(tmp, struct dentry, d_lru);
413 spin_lock(&dentry->d_lock);
415 * We found an inuse dentry which was not removed from
416 * dentry_unused because of laziness during lookup. Do not free
417 * it - just keep it off the dentry_unused list.
419 if (atomic_read(&dentry->d_count)) {
420 spin_unlock(&dentry->d_lock);
423 /* If the dentry was recently referenced, don't free it. */
424 if (dentry->d_flags & DCACHE_REFERENCED) {
425 dentry->d_flags &= ~DCACHE_REFERENCED;
426 list_add(&dentry->d_lru, &dentry_unused);
427 dentry_stat.nr_unused++;
428 spin_unlock(&dentry->d_lock);
431 prune_one_dentry(dentry);
433 spin_unlock(&dcache_lock);
437 * Shrink the dcache for the specified super block.
438 * This allows us to unmount a device without disturbing
439 * the dcache for the other devices.
441 * This implementation makes just two traversals of the
442 * unused list. On the first pass we move the selected
443 * dentries to the most recent end, and on the second
444 * pass we free them. The second pass must restart after
445 * each dput(), but since the target dentries are all at
446 * the end, it's really just a single traversal.
450 * shrink_dcache_sb - shrink dcache for a superblock
453 * Shrink the dcache for the specified super block. This
454 * is used to free the dcache before unmounting a file
458 void shrink_dcache_sb(struct super_block * sb)
460 struct list_head *tmp, *next;
461 struct dentry *dentry;
464 * Pass one ... move the dentries for the specified
465 * superblock to the most recent end of the unused list.
467 spin_lock(&dcache_lock);
468 next = dentry_unused.next;
469 while (next != &dentry_unused) {
472 dentry = list_entry(tmp, struct dentry, d_lru);
473 if (dentry->d_sb != sb)
476 list_add(tmp, &dentry_unused);
480 * Pass two ... free the dentries for this superblock.
483 next = dentry_unused.next;
484 while (next != &dentry_unused) {
487 dentry = list_entry(tmp, struct dentry, d_lru);
488 if (dentry->d_sb != sb)
490 dentry_stat.nr_unused--;
492 spin_lock(&dentry->d_lock);
493 if (atomic_read(&dentry->d_count)) {
494 spin_unlock(&dentry->d_lock);
497 prune_one_dentry(dentry);
500 spin_unlock(&dcache_lock);
504 * Search for at least 1 mount point in the dentry's subdirs.
505 * We descend to the next level whenever the d_subdirs
506 * list is non-empty and continue searching.
510 * have_submounts - check for mounts over a dentry
511 * @parent: dentry to check.
513 * Return true if the parent or its subdirectories contain
517 int have_submounts(struct dentry *parent)
519 struct dentry *this_parent = parent;
520 struct list_head *next;
522 spin_lock(&dcache_lock);
523 if (d_mountpoint(parent))
526 next = this_parent->d_subdirs.next;
528 while (next != &this_parent->d_subdirs) {
529 struct list_head *tmp = next;
530 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
532 /* Have we found a mount point ? */
533 if (d_mountpoint(dentry))
535 if (!list_empty(&dentry->d_subdirs)) {
536 this_parent = dentry;
541 * All done at this level ... ascend and resume the search.
543 if (this_parent != parent) {
544 next = this_parent->d_child.next;
545 this_parent = this_parent->d_parent;
548 spin_unlock(&dcache_lock);
549 return 0; /* No mount points found in tree */
551 spin_unlock(&dcache_lock);
556 * Search the dentry child list for the specified parent,
557 * and move any unused dentries to the end of the unused
558 * list for prune_dcache(). We descend to the next level
559 * whenever the d_subdirs list is non-empty and continue
562 * It returns zero iff there are no unused children,
563 * otherwise it returns the number of children moved to
564 * the end of the unused list. This may not be the total
565 * number of unused children, because select_parent can
566 * drop the lock and return early due to latency
569 static int select_parent(struct dentry * parent)
571 struct dentry *this_parent = parent;
572 struct list_head *next;
575 spin_lock(&dcache_lock);
577 next = this_parent->d_subdirs.next;
579 while (next != &this_parent->d_subdirs) {
580 struct list_head *tmp = next;
581 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
584 if (!list_empty(&dentry->d_lru)) {
585 dentry_stat.nr_unused--;
586 list_del_init(&dentry->d_lru);
589 * move only zero ref count dentries to the end
590 * of the unused list for prune_dcache
592 if (!atomic_read(&dentry->d_count)) {
593 list_add(&dentry->d_lru, dentry_unused.prev);
594 dentry_stat.nr_unused++;
599 * We can return to the caller if we have found some (this
600 * ensures forward progress). We'll be coming back to find
603 if (found && need_resched())
607 * Descend a level if the d_subdirs list is non-empty.
609 if (!list_empty(&dentry->d_subdirs)) {
610 this_parent = dentry;
612 printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n",
613 dentry->d_parent->d_name.name, dentry->d_name.name, found);
619 * All done at this level ... ascend and resume the search.
621 if (this_parent != parent) {
622 next = this_parent->d_child.next;
623 this_parent = this_parent->d_parent;
625 printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n",
626 this_parent->d_parent->d_name.name, this_parent->d_name.name, found);
631 spin_unlock(&dcache_lock);
636 * shrink_dcache_parent - prune dcache
637 * @parent: parent of entries to prune
639 * Prune the dcache to remove unused children of the parent dentry.
642 void shrink_dcache_parent(struct dentry * parent)
646 while ((found = select_parent(parent)) != 0)
651 * shrink_dcache_anon - further prune the cache
652 * @head: head of d_hash list of dentries to prune
654 * Prune the dentries that are anonymous
656 * parsing d_hash list does not hlist_for_each_rcu() as it
657 * done under dcache_lock.
660 void shrink_dcache_anon(struct hlist_head *head)
662 struct hlist_node *lp;
666 spin_lock(&dcache_lock);
667 hlist_for_each(lp, head) {
668 struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
669 if (!list_empty(&this->d_lru)) {
670 dentry_stat.nr_unused--;
671 list_del_init(&this->d_lru);
675 * move only zero ref count dentries to the end
676 * of the unused list for prune_dcache
678 if (!atomic_read(&this->d_count)) {
679 list_add_tail(&this->d_lru, &dentry_unused);
680 dentry_stat.nr_unused++;
684 spin_unlock(&dcache_lock);
690 * Scan `nr' dentries and return the number which remain.
692 * We need to avoid reentering the filesystem if the caller is performing a
693 * GFP_NOFS allocation attempt. One example deadlock is:
695 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
696 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
697 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
699 * In this case we return -1 to tell the caller that we baled.
701 static int shrink_dcache_memory(int nr, unsigned int gfp_mask)
704 if (!(gfp_mask & __GFP_FS))
708 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
712 * d_alloc - allocate a dcache entry
713 * @parent: parent of entry to allocate
714 * @name: qstr of the name
716 * Allocates a dentry. It returns %NULL if there is insufficient memory
717 * available. On a success the dentry is returned. The name passed in is
718 * copied and the copy passed in may be reused after this call.
721 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
723 struct dentry *dentry;
726 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
730 if (name->len > DNAME_INLINE_LEN-1) {
731 dname = kmalloc(name->len + 1, GFP_KERNEL);
733 kmem_cache_free(dentry_cache, dentry);
737 dname = dentry->d_iname;
739 dentry->d_name.name = dname;
741 dentry->d_name.len = name->len;
742 dentry->d_name.hash = name->hash;
743 memcpy(dname, name->name, name->len);
744 dname[name->len] = 0;
746 atomic_set(&dentry->d_count, 1);
747 dentry->d_flags = DCACHE_UNHASHED;
748 spin_lock_init(&dentry->d_lock);
749 dentry->d_inode = NULL;
750 dentry->d_parent = NULL;
753 dentry->d_fsdata = NULL;
754 dentry->d_extra_attributes = NULL;
755 dentry->d_mounted = 0;
756 dentry->d_cookie = NULL;
757 INIT_HLIST_NODE(&dentry->d_hash);
758 INIT_LIST_HEAD(&dentry->d_lru);
759 INIT_LIST_HEAD(&dentry->d_subdirs);
760 INIT_LIST_HEAD(&dentry->d_alias);
763 dentry->d_parent = dget(parent);
764 dentry->d_sb = parent->d_sb;
766 INIT_LIST_HEAD(&dentry->d_child);
769 spin_lock(&dcache_lock);
771 list_add(&dentry->d_child, &parent->d_subdirs);
772 dentry_stat.nr_dentry++;
773 spin_unlock(&dcache_lock);
778 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
783 q.len = strlen(name);
784 q.hash = full_name_hash(q.name, q.len);
785 return d_alloc(parent, &q);
789 * d_instantiate - fill in inode information for a dentry
790 * @entry: dentry to complete
791 * @inode: inode to attach to this dentry
793 * Fill in inode information in the entry.
795 * This turns negative dentries into productive full members
798 * NOTE! This assumes that the inode count has been incremented
799 * (or otherwise set) by the caller to indicate that it is now
800 * in use by the dcache.
803 void d_instantiate(struct dentry *entry, struct inode * inode)
805 if (!list_empty(&entry->d_alias)) BUG();
806 spin_lock(&dcache_lock);
808 list_add(&entry->d_alias, &inode->i_dentry);
809 entry->d_inode = inode;
810 spin_unlock(&dcache_lock);
811 security_d_instantiate(entry, inode);
815 * d_instantiate_unique - instantiate a non-aliased dentry
816 * @entry: dentry to instantiate
817 * @inode: inode to attach to this dentry
819 * Fill in inode information in the entry. On success, it returns NULL.
820 * If an unhashed alias of "entry" already exists, then we return the
821 * aliased dentry instead.
823 * Note that in order to avoid conflicts with rename() etc, the caller
824 * had better be holding the parent directory semaphore.
826 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
828 struct dentry *alias;
829 int len = entry->d_name.len;
830 const char *name = entry->d_name.name;
831 unsigned int hash = entry->d_name.hash;
833 BUG_ON(!list_empty(&entry->d_alias));
834 spin_lock(&dcache_lock);
837 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
838 struct qstr *qstr = &alias->d_name;
840 if (qstr->hash != hash)
842 if (alias->d_parent != entry->d_parent)
844 if (qstr->len != len)
846 if (memcmp(qstr->name, name, len))
849 spin_unlock(&dcache_lock);
850 BUG_ON(!d_unhashed(alias));
853 list_add(&entry->d_alias, &inode->i_dentry);
855 entry->d_inode = inode;
856 spin_unlock(&dcache_lock);
857 security_d_instantiate(entry, inode);
860 EXPORT_SYMBOL(d_instantiate_unique);
863 * d_alloc_root - allocate root dentry
864 * @root_inode: inode to allocate the root for
866 * Allocate a root ("/") dentry for the inode given. The inode is
867 * instantiated and returned. %NULL is returned if there is insufficient
868 * memory or the inode passed is %NULL.
871 struct dentry * d_alloc_root(struct inode * root_inode)
873 struct dentry *res = NULL;
876 static const struct qstr name = { .name = "/", .len = 1 };
878 res = d_alloc(NULL, &name);
880 res->d_sb = root_inode->i_sb;
882 d_instantiate(res, root_inode);
888 static inline struct hlist_head *d_hash(struct dentry *parent,
891 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
892 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
893 return dentry_hashtable + (hash & D_HASHMASK);
897 * d_alloc_anon - allocate an anonymous dentry
898 * @inode: inode to allocate the dentry for
900 * This is similar to d_alloc_root. It is used by filesystems when
901 * creating a dentry for a given inode, often in the process of
902 * mapping a filehandle to a dentry. The returned dentry may be
903 * anonymous, or may have a full name (if the inode was already
904 * in the cache). The file system may need to make further
905 * efforts to connect this dentry into the dcache properly.
907 * When called on a directory inode, we must ensure that
908 * the inode only ever has one dentry. If a dentry is
909 * found, that is returned instead of allocating a new one.
911 * On successful return, the reference to the inode has been transferred
912 * to the dentry. If %NULL is returned (indicating kmalloc failure),
913 * the reference on the inode has not been released.
916 struct dentry * d_alloc_anon(struct inode *inode)
918 static const struct qstr anonstring = { .name = "" };
922 if ((res = d_find_alias(inode))) {
927 tmp = d_alloc(NULL, &anonstring);
931 tmp->d_parent = tmp; /* make sure dput doesn't croak */
933 spin_lock(&dcache_lock);
934 res = __d_find_alias(inode, 0);
936 /* attach a disconnected dentry */
939 spin_lock(&res->d_lock);
940 res->d_sb = inode->i_sb;
942 res->d_inode = inode;
943 res->d_flags |= DCACHE_DISCONNECTED;
944 res->d_flags &= ~DCACHE_UNHASHED;
945 list_add(&res->d_alias, &inode->i_dentry);
946 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
947 spin_unlock(&res->d_lock);
949 inode = NULL; /* don't drop reference */
951 spin_unlock(&dcache_lock);
962 * d_splice_alias - splice a disconnected dentry into the tree if one exists
963 * @inode: the inode which may have a disconnected dentry
964 * @dentry: a negative dentry which we want to point to the inode.
966 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
967 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
968 * and return it, else simply d_add the inode to the dentry and return NULL.
970 * This is needed in the lookup routine of any filesystem that is exportable
971 * (via knfsd) so that we can build dcache paths to directories effectively.
973 * If a dentry was found and moved, then it is returned. Otherwise NULL
974 * is returned. This matches the expected return value of ->lookup.
977 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
979 struct dentry *new = NULL;
982 spin_lock(&dcache_lock);
983 new = __d_find_alias(inode, 1);
985 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
986 spin_unlock(&dcache_lock);
987 security_d_instantiate(new, inode);
992 /* d_instantiate takes dcache_lock, so we do it by hand */
993 list_add(&dentry->d_alias, &inode->i_dentry);
994 dentry->d_inode = inode;
995 spin_unlock(&dcache_lock);
996 security_d_instantiate(dentry, inode);
1000 d_add(dentry, inode);
1006 * d_lookup - search for a dentry
1007 * @parent: parent dentry
1008 * @name: qstr of name we wish to find
1010 * Searches the children of the parent dentry for the name in question. If
1011 * the dentry is found its reference count is incremented and the dentry
1012 * is returned. The caller must use d_put to free the entry when it has
1013 * finished using it. %NULL is returned on failure.
1015 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1016 * Memory barriers are used while updating and doing lockless traversal.
1017 * To avoid races with d_move while rename is happening, d_lock is used.
1019 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1020 * and name pointer in one structure pointed by d_qstr.
1022 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1023 * lookup is going on.
1025 * dentry_unused list is not updated even if lookup finds the required dentry
1026 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1027 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1030 * d_lookup() is protected against the concurrent renames in some unrelated
1031 * directory using the seqlockt_t rename_lock.
1034 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1036 struct dentry * dentry = NULL;
1040 seq = read_seqbegin(&rename_lock);
1041 dentry = __d_lookup(parent, name);
1044 } while (read_seqretry(&rename_lock, seq));
1048 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1050 unsigned int len = name->len;
1051 unsigned int hash = name->hash;
1052 const unsigned char *str = name->name;
1053 struct hlist_head *head = d_hash(parent,hash);
1054 struct dentry *found = NULL;
1055 struct hlist_node *node;
1059 hlist_for_each_rcu(node, head) {
1060 struct dentry *dentry;
1063 dentry = hlist_entry(node, struct dentry, d_hash);
1065 if (dentry->d_name.hash != hash)
1067 if (dentry->d_parent != parent)
1070 spin_lock(&dentry->d_lock);
1073 * Recheck the dentry after taking the lock - d_move may have
1074 * changed things. Don't bother checking the hash because we're
1075 * about to compare the whole name anyway.
1077 if (dentry->d_parent != parent)
1081 * It is safe to compare names since d_move() cannot
1082 * change the qstr (protected by d_lock).
1084 qstr = &dentry->d_name;
1085 if (parent->d_op && parent->d_op->d_compare) {
1086 if (parent->d_op->d_compare(parent, qstr, name))
1089 if (qstr->len != len)
1091 if (memcmp(qstr->name, str, len))
1095 if (!d_unhashed(dentry)) {
1096 atomic_inc(&dentry->d_count);
1099 spin_unlock(&dentry->d_lock);
1102 spin_unlock(&dentry->d_lock);
1110 * d_validate - verify dentry provided from insecure source
1111 * @dentry: The dentry alleged to be valid child of @dparent
1112 * @dparent: The parent dentry (known to be valid)
1113 * @hash: Hash of the dentry
1114 * @len: Length of the name
1116 * An insecure source has sent us a dentry, here we verify it and dget() it.
1117 * This is used by ncpfs in its readdir implementation.
1118 * Zero is returned in the dentry is invalid.
1121 int d_validate(struct dentry *dentry, struct dentry *dparent)
1123 struct hlist_head *base;
1124 struct hlist_node *lhp;
1126 /* Check whether the ptr might be valid at all.. */
1127 if (!kmem_ptr_validate(dentry_cache, dentry))
1130 if (dentry->d_parent != dparent)
1133 spin_lock(&dcache_lock);
1134 base = d_hash(dparent, dentry->d_name.hash);
1135 hlist_for_each(lhp,base) {
1136 /* hlist_for_each_rcu() not required for d_hash list
1137 * as it is parsed under dcache_lock
1139 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1140 __dget_locked(dentry);
1141 spin_unlock(&dcache_lock);
1145 spin_unlock(&dcache_lock);
1151 * When a file is deleted, we have two options:
1152 * - turn this dentry into a negative dentry
1153 * - unhash this dentry and free it.
1155 * Usually, we want to just turn this into
1156 * a negative dentry, but if anybody else is
1157 * currently using the dentry or the inode
1158 * we can't do that and we fall back on removing
1159 * it from the hash queues and waiting for
1160 * it to be deleted later when it has no users
1164 * d_delete - delete a dentry
1165 * @dentry: The dentry to delete
1167 * Turn the dentry into a negative dentry if possible, otherwise
1168 * remove it from the hash queues so it can be deleted later
1171 void d_delete(struct dentry * dentry)
1174 * Are we the only user?
1176 spin_lock(&dcache_lock);
1177 spin_lock(&dentry->d_lock);
1178 if (atomic_read(&dentry->d_count) == 1) {
1179 dentry_iput(dentry);
1183 if (!d_unhashed(dentry))
1186 spin_unlock(&dentry->d_lock);
1187 spin_unlock(&dcache_lock);
1190 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1193 entry->d_flags &= ~DCACHE_UNHASHED;
1194 hlist_add_head_rcu(&entry->d_hash, list);
1198 * d_rehash - add an entry back to the hash
1199 * @entry: dentry to add to the hash
1201 * Adds a dentry to the hash according to its name.
1204 void d_rehash(struct dentry * entry)
1206 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1208 spin_lock(&dcache_lock);
1209 spin_lock(&entry->d_lock);
1210 __d_rehash(entry, list);
1211 spin_unlock(&entry->d_lock);
1212 spin_unlock(&dcache_lock);
1215 #define do_switch(x,y) do { \
1216 __typeof__ (x) __tmp = x; \
1217 x = y; y = __tmp; } while (0)
1220 * When switching names, the actual string doesn't strictly have to
1221 * be preserved in the target - because we're dropping the target
1222 * anyway. As such, we can just do a simple memcpy() to copy over
1223 * the new name before we switch.
1225 * Note that we have to be a lot more careful about getting the hash
1226 * switched - we have to switch the hash value properly even if it
1227 * then no longer matches the actual (corrupted) string of the target.
1228 * The hash value has to match the hash queue that the dentry is on..
1230 static void switch_names(struct dentry *dentry, struct dentry *target)
1232 if (dname_external(target)) {
1233 if (dname_external(dentry)) {
1235 * Both external: swap the pointers
1237 do_switch(target->d_name.name, dentry->d_name.name);
1240 * dentry:internal, target:external. Steal target's
1241 * storage and make target internal.
1243 dentry->d_name.name = target->d_name.name;
1244 target->d_name.name = target->d_iname;
1247 if (dname_external(dentry)) {
1249 * dentry:external, target:internal. Give dentry's
1250 * storage to target and make dentry internal
1252 memcpy(dentry->d_iname, target->d_name.name,
1253 target->d_name.len + 1);
1254 target->d_name.name = dentry->d_name.name;
1255 dentry->d_name.name = dentry->d_iname;
1258 * Both are internal. Just copy target to dentry
1260 memcpy(dentry->d_iname, target->d_name.name,
1261 target->d_name.len + 1);
1267 * We cannibalize "target" when moving dentry on top of it,
1268 * because it's going to be thrown away anyway. We could be more
1269 * polite about it, though.
1271 * This forceful removal will result in ugly /proc output if
1272 * somebody holds a file open that got deleted due to a rename.
1273 * We could be nicer about the deleted file, and let it show
1274 * up under the name it got deleted rather than the name that
1279 * d_move - move a dentry
1280 * @dentry: entry to move
1281 * @target: new dentry
1283 * Update the dcache to reflect the move of a file name. Negative
1284 * dcache entries should not be moved in this way.
1287 void d_move(struct dentry * dentry, struct dentry * target)
1289 struct hlist_head *list;
1291 if (!dentry->d_inode)
1292 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1294 spin_lock(&dcache_lock);
1295 write_seqlock(&rename_lock);
1297 * XXXX: do we really need to take target->d_lock?
1299 if (target < dentry) {
1300 spin_lock(&target->d_lock);
1301 spin_lock(&dentry->d_lock);
1303 spin_lock(&dentry->d_lock);
1304 spin_lock(&target->d_lock);
1307 /* Move the dentry to the target hash queue, if on different bucket */
1308 if (dentry->d_flags & DCACHE_UNHASHED)
1309 goto already_unhashed;
1311 hlist_del_rcu(&dentry->d_hash);
1314 list = d_hash(target->d_parent, target->d_name.hash);
1315 __d_rehash(dentry, list);
1317 /* Unhash the target: dput() will then get rid of it */
1320 /* flush any possible attributes */
1321 if (dentry->d_extra_attributes) {
1322 kfree(dentry->d_extra_attributes);
1323 dentry->d_extra_attributes = NULL;
1325 if (target->d_extra_attributes) {
1326 kfree(target->d_extra_attributes);
1327 target->d_extra_attributes = NULL;
1330 list_del(&dentry->d_child);
1331 list_del(&target->d_child);
1333 /* Switch the names.. */
1334 switch_names(dentry, target);
1335 do_switch(dentry->d_name.len, target->d_name.len);
1336 do_switch(dentry->d_name.hash, target->d_name.hash);
1338 /* ... and switch the parents */
1339 if (IS_ROOT(dentry)) {
1340 dentry->d_parent = target->d_parent;
1341 target->d_parent = target;
1342 INIT_LIST_HEAD(&target->d_child);
1344 do_switch(dentry->d_parent, target->d_parent);
1346 /* And add them back to the (new) parent lists */
1347 list_add(&target->d_child, &target->d_parent->d_subdirs);
1350 list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
1351 spin_unlock(&target->d_lock);
1352 spin_unlock(&dentry->d_lock);
1353 write_sequnlock(&rename_lock);
1354 spin_unlock(&dcache_lock);
1358 * d_path - return the path of a dentry
1359 * @dentry: dentry to report
1360 * @vfsmnt: vfsmnt to which the dentry belongs
1361 * @root: root dentry
1362 * @rootmnt: vfsmnt to which the root dentry belongs
1363 * @buffer: buffer to return value in
1364 * @buflen: buffer length
1366 * Convert a dentry into an ASCII path name. If the entry has been deleted
1367 * the string " (deleted)" is appended. Note that this is ambiguous.
1369 * Returns the buffer or an error code if the path was too long.
1371 * "buflen" should be positive. Caller holds the dcache_lock.
1373 char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1374 struct dentry *root, struct vfsmount *rootmnt,
1375 char *buffer, int buflen)
1377 char * end = buffer+buflen;
1383 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1388 memcpy(end, " (deleted)", 10);
1398 struct dentry * parent;
1400 if (dentry == root && vfsmnt == rootmnt)
1402 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1404 spin_lock(&vfsmount_lock);
1405 if (vfsmnt->mnt_parent == vfsmnt) {
1406 spin_unlock(&vfsmount_lock);
1409 dentry = vfsmnt->mnt_mountpoint;
1410 vfsmnt = vfsmnt->mnt_parent;
1411 spin_unlock(&vfsmount_lock);
1414 parent = dentry->d_parent;
1416 namelen = dentry->d_name.len;
1417 buflen -= namelen + 1;
1421 memcpy(end, dentry->d_name.name, namelen);
1430 namelen = dentry->d_name.len;
1434 retval -= namelen-1; /* hit the slash */
1435 memcpy(retval, dentry->d_name.name, namelen);
1438 return ERR_PTR(-ENAMETOOLONG);
1441 EXPORT_SYMBOL_GPL(__d_path);
1443 /* write full pathname into buffer and return start of pathname */
1444 char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1445 char *buf, int buflen)
1448 struct vfsmount *rootmnt;
1449 struct dentry *root;
1451 read_lock(¤t->fs->lock);
1452 rootmnt = mntget(current->fs->rootmnt);
1453 root = dget(current->fs->root);
1454 read_unlock(¤t->fs->lock);
1455 spin_lock(&dcache_lock);
1456 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1457 spin_unlock(&dcache_lock);
1464 * NOTE! The user-level library version returns a
1465 * character pointer. The kernel system call just
1466 * returns the length of the buffer filled (which
1467 * includes the ending '\0' character), or a negative
1468 * error value. So libc would do something like
1470 * char *getcwd(char * buf, size_t size)
1474 * retval = sys_getcwd(buf, size);
1481 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1484 struct vfsmount *pwdmnt, *rootmnt;
1485 struct dentry *pwd, *root;
1486 char *page = (char *) __get_free_page(GFP_USER);
1491 read_lock(¤t->fs->lock);
1492 pwdmnt = mntget(current->fs->pwdmnt);
1493 pwd = dget(current->fs->pwd);
1494 rootmnt = mntget(current->fs->rootmnt);
1495 root = dget(current->fs->root);
1496 read_unlock(¤t->fs->lock);
1499 /* Has the current directory has been unlinked? */
1500 spin_lock(&dcache_lock);
1501 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1505 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1506 spin_unlock(&dcache_lock);
1508 error = PTR_ERR(cwd);
1513 len = PAGE_SIZE + page - cwd;
1516 if (copy_to_user(buf, cwd, len))
1520 spin_unlock(&dcache_lock);
1527 free_page((unsigned long) page);
1532 * Test whether new_dentry is a subdirectory of old_dentry.
1534 * Trivially implemented using the dcache structure
1538 * is_subdir - is new dentry a subdirectory of old_dentry
1539 * @new_dentry: new dentry
1540 * @old_dentry: old dentry
1542 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1543 * Returns 0 otherwise.
1544 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1547 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1550 struct dentry * saved = new_dentry;
1553 /* need rcu_readlock to protect against the d_parent trashing due to
1558 /* for restarting inner loop in case of seq retry */
1561 seq = read_seqbegin(&rename_lock);
1563 if (new_dentry != old_dentry) {
1564 struct dentry * parent = new_dentry->d_parent;
1565 if (parent == new_dentry)
1567 new_dentry = parent;
1573 } while (read_seqretry(&rename_lock, seq));
1579 void d_genocide(struct dentry *root)
1581 struct dentry *this_parent = root;
1582 struct list_head *next;
1584 spin_lock(&dcache_lock);
1586 next = this_parent->d_subdirs.next;
1588 while (next != &this_parent->d_subdirs) {
1589 struct list_head *tmp = next;
1590 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1592 if (d_unhashed(dentry)||!dentry->d_inode)
1594 if (!list_empty(&dentry->d_subdirs)) {
1595 this_parent = dentry;
1598 atomic_dec(&dentry->d_count);
1600 if (this_parent != root) {
1601 next = this_parent->d_child.next;
1602 atomic_dec(&this_parent->d_count);
1603 this_parent = this_parent->d_parent;
1606 spin_unlock(&dcache_lock);
1610 * find_inode_number - check for dentry with name
1611 * @dir: directory to check
1612 * @name: Name to find.
1614 * Check whether a dentry already exists for the given name,
1615 * and return the inode number if it has an inode. Otherwise
1618 * This routine is used to post-process directory listings for
1619 * filesystems using synthetic inode numbers, and is necessary
1620 * to keep getcwd() working.
1623 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1625 struct dentry * dentry;
1629 * Check for a fs-specific hash function. Note that we must
1630 * calculate the standard hash first, as the d_op->d_hash()
1631 * routine may choose to leave the hash value unchanged.
1633 name->hash = full_name_hash(name->name, name->len);
1634 if (dir->d_op && dir->d_op->d_hash)
1636 if (dir->d_op->d_hash(dir, name) != 0)
1640 dentry = d_lookup(dir, name);
1643 if (dentry->d_inode)
1644 ino = dentry->d_inode->i_ino;
1651 static __initdata unsigned long dhash_entries;
1652 static int __init set_dhash_entries(char *str)
1656 dhash_entries = simple_strtoul(str, &str, 0);
1659 __setup("dhash_entries=", set_dhash_entries);
1661 static void __init dcache_init_early(void)
1665 /* If hashes are distributed across NUMA nodes, defer
1666 * hash allocation until vmalloc space is available.
1672 alloc_large_system_hash("Dentry cache",
1673 sizeof(struct hlist_head),
1681 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1682 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1685 void flush_dentry_attributes (void)
1687 struct hlist_node *tmp;
1688 struct dentry *dentry;
1691 spin_lock(&dcache_lock);
1692 for (i = 0; i <= d_hash_mask; i++)
1693 hlist_for_each_entry(dentry, tmp, dentry_hashtable+i, d_hash) {
1694 kfree(dentry->d_extra_attributes);
1695 dentry->d_extra_attributes = NULL;
1697 spin_unlock(&dcache_lock);
1700 EXPORT_SYMBOL_GPL(flush_dentry_attributes);
1702 static void __init dcache_init(unsigned long mempages)
1707 * A constructor could be added for stable state like the lists,
1708 * but it is probably not worth it because of the cache nature
1711 dentry_cache = kmem_cache_create("dentry_cache",
1712 sizeof(struct dentry),
1714 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1717 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1719 /* Hash may have been set up in dcache_init_early */
1724 alloc_large_system_hash("Dentry cache",
1725 sizeof(struct hlist_head),
1733 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1734 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1737 /* SLAB cache for __getname() consumers */
1738 kmem_cache_t *names_cachep;
1740 /* SLAB cache for file structures */
1741 kmem_cache_t *filp_cachep;
1743 EXPORT_SYMBOL(d_genocide);
1745 extern void bdev_cache_init(void);
1746 extern void chrdev_init(void);
1748 void __init vfs_caches_init_early(void)
1750 dcache_init_early();
1754 void __init vfs_caches_init(unsigned long mempages)
1756 unsigned long reserve;
1758 /* Base hash sizes on available memory, with a reserve equal to
1759 150% of current kernel size */
1761 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1762 mempages -= reserve;
1764 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1765 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1767 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1768 SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
1770 dcache_init(mempages);
1771 inode_init(mempages);
1772 files_init(mempages);
1778 EXPORT_SYMBOL(d_alloc);
1779 EXPORT_SYMBOL(d_alloc_anon);
1780 EXPORT_SYMBOL(d_alloc_root);
1781 EXPORT_SYMBOL(d_delete);
1782 EXPORT_SYMBOL(d_find_alias);
1783 EXPORT_SYMBOL(d_instantiate);
1784 EXPORT_SYMBOL(d_invalidate);
1785 EXPORT_SYMBOL(d_lookup);
1786 EXPORT_SYMBOL(d_move);
1787 EXPORT_SYMBOL(d_path);
1788 EXPORT_SYMBOL(d_prune_aliases);
1789 EXPORT_SYMBOL(d_rehash);
1790 EXPORT_SYMBOL(d_splice_alias);
1791 EXPORT_SYMBOL(d_validate);
1792 EXPORT_SYMBOL(dget_locked);
1793 EXPORT_SYMBOL(dput);
1794 EXPORT_SYMBOL(find_inode_number);
1795 EXPORT_SYMBOL(have_submounts);
1796 EXPORT_SYMBOL(names_cachep);
1797 EXPORT_SYMBOL(shrink_dcache_parent);
1798 EXPORT_SYMBOL(shrink_dcache_sb);