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>
38 int sysctl_vfs_cache_pressure __read_mostly = 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 __read_mostly;
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 __read_mostly;
62 static unsigned int d_hash_shift __read_mostly;
63 static struct hlist_head *dentry_hashtable __read_mostly;
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_u.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 call_rcu(&dentry->d_u.d_rcu, d_callback);
92 * Release the dentry's inode, using the filesystem
93 * d_iput() operation if defined.
94 * Called with dcache_lock and per dentry lock held, drops both.
96 static void dentry_iput(struct dentry * dentry)
98 struct inode *inode = dentry->d_inode;
100 dentry->d_inode = NULL;
101 list_del_init(&dentry->d_alias);
102 spin_unlock(&dentry->d_lock);
103 spin_unlock(&dcache_lock);
105 fsnotify_inoderemove(inode);
106 if (dentry->d_op && dentry->d_op->d_iput)
107 dentry->d_op->d_iput(dentry, inode);
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
119 * This is complicated by the fact that we do not want to put
120 * dentries that are no longer on any hash chain on the unused
121 * list: we'd much rather just get rid of them immediately.
123 * However, that implies that we have to traverse the dentry
124 * tree upwards to the parents which might _also_ now be
125 * scheduled for deletion (it may have been only waiting for
126 * its last child to go away).
128 * This tail recursion is done by hand as we don't want to depend
129 * on the compiler to always get this right (gcc generally doesn't).
130 * Real recursion would eat up our stack space.
134 * dput - release a dentry
135 * @dentry: dentry to release
137 * Release a dentry. This will drop the usage count and if appropriate
138 * call the dentry unlink method as well as removing it from the queues and
139 * releasing its resources. If the parent dentries were scheduled for release
140 * they too may now get deleted.
142 * no dcache lock, please.
145 void dput(struct dentry *dentry)
151 if (atomic_read(&dentry->d_count) == 1)
153 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
156 spin_lock(&dentry->d_lock);
157 if (atomic_read(&dentry->d_count)) {
158 spin_unlock(&dentry->d_lock);
159 spin_unlock(&dcache_lock);
164 * AV: ->d_delete() is _NOT_ allowed to block now.
166 if (dentry->d_op && dentry->d_op->d_delete) {
167 if (dentry->d_op->d_delete(dentry))
170 /* Unreachable? Get rid of it */
171 if (d_unhashed(dentry))
173 if (list_empty(&dentry->d_lru)) {
174 dentry->d_flags |= DCACHE_REFERENCED;
175 list_add(&dentry->d_lru, &dentry_unused);
176 dentry_stat.nr_unused++;
178 spin_unlock(&dentry->d_lock);
179 spin_unlock(&dcache_lock);
186 struct dentry *parent;
188 /* If dentry was on d_lru list
189 * delete it from there
191 if (!list_empty(&dentry->d_lru)) {
192 list_del(&dentry->d_lru);
193 dentry_stat.nr_unused--;
195 list_del(&dentry->d_u.d_child);
196 dentry_stat.nr_dentry--; /* For d_free, below */
197 /*drops the locks, at that point nobody can reach this dentry */
199 parent = dentry->d_parent;
201 if (dentry == parent)
209 * d_invalidate - invalidate a dentry
210 * @dentry: dentry to invalidate
212 * Try to invalidate the dentry if it turns out to be
213 * possible. If there are other dentries that can be
214 * reached through this one we can't delete it and we
215 * return -EBUSY. On success we return 0.
220 int d_invalidate(struct dentry * dentry)
223 * If it's already been dropped, return OK.
225 spin_lock(&dcache_lock);
226 if (d_unhashed(dentry)) {
227 spin_unlock(&dcache_lock);
231 * Check whether to do a partial shrink_dcache
232 * to get rid of unused child entries.
234 if (!list_empty(&dentry->d_subdirs)) {
235 spin_unlock(&dcache_lock);
236 shrink_dcache_parent(dentry);
237 spin_lock(&dcache_lock);
241 * Somebody else still using it?
243 * If it's a directory, we can't drop it
244 * for fear of somebody re-populating it
245 * with children (even though dropping it
246 * would make it unreachable from the root,
247 * we might still populate it if it was a
248 * working directory or similar).
250 spin_lock(&dentry->d_lock);
251 if (atomic_read(&dentry->d_count) > 1) {
252 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
253 spin_unlock(&dentry->d_lock);
254 spin_unlock(&dcache_lock);
260 spin_unlock(&dentry->d_lock);
261 spin_unlock(&dcache_lock);
265 /* This should be called _only_ with dcache_lock held */
267 static inline struct dentry * __dget_locked(struct dentry *dentry)
269 atomic_inc(&dentry->d_count);
270 if (!list_empty(&dentry->d_lru)) {
271 dentry_stat.nr_unused--;
272 list_del_init(&dentry->d_lru);
277 struct dentry * dget_locked(struct dentry *dentry)
279 return __dget_locked(dentry);
283 * d_find_alias - grab a hashed alias of inode
284 * @inode: inode in question
285 * @want_discon: flag, used by d_splice_alias, to request
286 * that only a DISCONNECTED alias be returned.
288 * If inode has a hashed alias, or is a directory and has any alias,
289 * acquire the reference to alias and return it. Otherwise return NULL.
290 * Notice that if inode is a directory there can be only one alias and
291 * it can be unhashed only if it has no children, or if it is the root
294 * If the inode has a DCACHE_DISCONNECTED alias, then prefer
295 * any other hashed alias over that one unless @want_discon is set,
296 * in which case only return a DCACHE_DISCONNECTED alias.
299 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
301 struct list_head *head, *next, *tmp;
302 struct dentry *alias, *discon_alias=NULL;
304 head = &inode->i_dentry;
305 next = inode->i_dentry.next;
306 while (next != head) {
310 alias = list_entry(tmp, struct dentry, d_alias);
311 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
312 if (alias->d_flags & DCACHE_DISCONNECTED)
313 discon_alias = alias;
314 else if (!want_discon) {
315 __dget_locked(alias);
321 __dget_locked(discon_alias);
325 struct dentry * d_find_alias(struct inode *inode)
327 struct dentry *de = NULL;
329 if (!list_empty(&inode->i_dentry)) {
330 spin_lock(&dcache_lock);
331 de = __d_find_alias(inode, 0);
332 spin_unlock(&dcache_lock);
338 * Try to kill dentries associated with this inode.
339 * WARNING: you must own a reference to inode.
341 void d_prune_aliases(struct inode *inode)
343 struct dentry *dentry;
345 spin_lock(&dcache_lock);
346 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
347 spin_lock(&dentry->d_lock);
348 if (!atomic_read(&dentry->d_count)) {
349 __dget_locked(dentry);
351 spin_unlock(&dentry->d_lock);
352 spin_unlock(&dcache_lock);
356 spin_unlock(&dentry->d_lock);
358 spin_unlock(&dcache_lock);
362 * Throw away a dentry - free the inode, dput the parent.
363 * This requires that the LRU list has already been
365 * Called with dcache_lock, drops it and then regains.
367 static inline void prune_one_dentry(struct dentry * dentry)
369 struct dentry * parent;
372 list_del(&dentry->d_u.d_child);
373 dentry_stat.nr_dentry--; /* For d_free, below */
375 parent = dentry->d_parent;
377 if (parent != dentry)
379 spin_lock(&dcache_lock);
383 * prune_dcache - shrink the dcache
384 * @count: number of entries to try and free
386 * Shrink the dcache. This is done when we need
387 * more memory, or simply when we need to unmount
388 * something (at which point we need to unuse
391 * This function may fail to free any resources if
392 * all the dentries are in use.
395 static void prune_dcache(int count)
397 spin_lock(&dcache_lock);
398 for (; count ; count--) {
399 struct dentry *dentry;
400 struct list_head *tmp;
402 cond_resched_lock(&dcache_lock);
404 tmp = dentry_unused.prev;
405 if (tmp == &dentry_unused)
408 prefetch(dentry_unused.prev);
409 dentry_stat.nr_unused--;
410 dentry = list_entry(tmp, struct dentry, d_lru);
412 spin_lock(&dentry->d_lock);
414 * We found an inuse dentry which was not removed from
415 * dentry_unused because of laziness during lookup. Do not free
416 * it - just keep it off the dentry_unused list.
418 if (atomic_read(&dentry->d_count)) {
419 spin_unlock(&dentry->d_lock);
422 /* If the dentry was recently referenced, don't free it. */
423 if (dentry->d_flags & DCACHE_REFERENCED) {
424 dentry->d_flags &= ~DCACHE_REFERENCED;
425 list_add(&dentry->d_lru, &dentry_unused);
426 dentry_stat.nr_unused++;
427 spin_unlock(&dentry->d_lock);
430 prune_one_dentry(dentry);
432 spin_unlock(&dcache_lock);
436 * Shrink the dcache for the specified super block.
437 * This allows us to unmount a device without disturbing
438 * the dcache for the other devices.
440 * This implementation makes just two traversals of the
441 * unused list. On the first pass we move the selected
442 * dentries to the most recent end, and on the second
443 * pass we free them. The second pass must restart after
444 * each dput(), but since the target dentries are all at
445 * the end, it's really just a single traversal.
449 * shrink_dcache_sb - shrink dcache for a superblock
452 * Shrink the dcache for the specified super block. This
453 * is used to free the dcache before unmounting a file
457 void shrink_dcache_sb(struct super_block * sb)
459 struct list_head *tmp, *next;
460 struct dentry *dentry;
463 * Pass one ... move the dentries for the specified
464 * superblock to the most recent end of the unused list.
466 spin_lock(&dcache_lock);
467 list_for_each_safe(tmp, next, &dentry_unused) {
468 dentry = list_entry(tmp, struct dentry, d_lru);
469 if (dentry->d_sb != sb)
472 list_add(tmp, &dentry_unused);
476 * Pass two ... free the dentries for this superblock.
479 list_for_each_safe(tmp, next, &dentry_unused) {
480 dentry = list_entry(tmp, struct dentry, d_lru);
481 if (dentry->d_sb != sb)
483 dentry_stat.nr_unused--;
485 spin_lock(&dentry->d_lock);
486 if (atomic_read(&dentry->d_count)) {
487 spin_unlock(&dentry->d_lock);
490 prune_one_dentry(dentry);
491 cond_resched_lock(&dcache_lock);
494 spin_unlock(&dcache_lock);
498 * Search for at least 1 mount point in the dentry's subdirs.
499 * We descend to the next level whenever the d_subdirs
500 * list is non-empty and continue searching.
504 * have_submounts - check for mounts over a dentry
505 * @parent: dentry to check.
507 * Return true if the parent or its subdirectories contain
511 int have_submounts(struct dentry *parent)
513 struct dentry *this_parent = parent;
514 struct list_head *next;
516 spin_lock(&dcache_lock);
517 if (d_mountpoint(parent))
520 next = this_parent->d_subdirs.next;
522 while (next != &this_parent->d_subdirs) {
523 struct list_head *tmp = next;
524 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
526 /* Have we found a mount point ? */
527 if (d_mountpoint(dentry))
529 if (!list_empty(&dentry->d_subdirs)) {
530 this_parent = dentry;
535 * All done at this level ... ascend and resume the search.
537 if (this_parent != parent) {
538 next = this_parent->d_u.d_child.next;
539 this_parent = this_parent->d_parent;
542 spin_unlock(&dcache_lock);
543 return 0; /* No mount points found in tree */
545 spin_unlock(&dcache_lock);
550 * Search the dentry child list for the specified parent,
551 * and move any unused dentries to the end of the unused
552 * list for prune_dcache(). We descend to the next level
553 * whenever the d_subdirs list is non-empty and continue
556 * It returns zero iff there are no unused children,
557 * otherwise it returns the number of children moved to
558 * the end of the unused list. This may not be the total
559 * number of unused children, because select_parent can
560 * drop the lock and return early due to latency
563 static int select_parent(struct dentry * parent)
565 struct dentry *this_parent = parent;
566 struct list_head *next;
569 spin_lock(&dcache_lock);
571 next = this_parent->d_subdirs.next;
573 while (next != &this_parent->d_subdirs) {
574 struct list_head *tmp = next;
575 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
578 if (!list_empty(&dentry->d_lru)) {
579 dentry_stat.nr_unused--;
580 list_del_init(&dentry->d_lru);
583 * move only zero ref count dentries to the end
584 * of the unused list for prune_dcache
586 if (!atomic_read(&dentry->d_count)) {
587 list_add(&dentry->d_lru, dentry_unused.prev);
588 dentry_stat.nr_unused++;
593 * We can return to the caller if we have found some (this
594 * ensures forward progress). We'll be coming back to find
597 if (found && need_resched())
601 * Descend a level if the d_subdirs list is non-empty.
603 if (!list_empty(&dentry->d_subdirs)) {
604 this_parent = dentry;
609 * All done at this level ... ascend and resume the search.
611 if (this_parent != parent) {
612 next = this_parent->d_u.d_child.next;
613 this_parent = this_parent->d_parent;
617 spin_unlock(&dcache_lock);
622 * shrink_dcache_parent - prune dcache
623 * @parent: parent of entries to prune
625 * Prune the dcache to remove unused children of the parent dentry.
628 void shrink_dcache_parent(struct dentry * parent)
632 while ((found = select_parent(parent)) != 0)
637 * shrink_dcache_anon - further prune the cache
638 * @head: head of d_hash list of dentries to prune
640 * Prune the dentries that are anonymous
642 * parsing d_hash list does not hlist_for_each_entry_rcu() as it
643 * done under dcache_lock.
646 void shrink_dcache_anon(struct hlist_head *head)
648 struct hlist_node *lp;
652 spin_lock(&dcache_lock);
653 hlist_for_each(lp, head) {
654 struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
655 if (!list_empty(&this->d_lru)) {
656 dentry_stat.nr_unused--;
657 list_del_init(&this->d_lru);
661 * move only zero ref count dentries to the end
662 * of the unused list for prune_dcache
664 if (!atomic_read(&this->d_count)) {
665 list_add_tail(&this->d_lru, &dentry_unused);
666 dentry_stat.nr_unused++;
670 spin_unlock(&dcache_lock);
676 * Scan `nr' dentries and return the number which remain.
678 * We need to avoid reentering the filesystem if the caller is performing a
679 * GFP_NOFS allocation attempt. One example deadlock is:
681 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
682 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
683 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
685 * In this case we return -1 to tell the caller that we baled.
687 static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
690 if (!(gfp_mask & __GFP_FS))
694 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
698 * d_alloc - allocate a dcache entry
699 * @parent: parent of entry to allocate
700 * @name: qstr of the name
702 * Allocates a dentry. It returns %NULL if there is insufficient memory
703 * available. On a success the dentry is returned. The name passed in is
704 * copied and the copy passed in may be reused after this call.
707 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
709 struct dentry *dentry;
712 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
716 if (name->len > DNAME_INLINE_LEN-1) {
717 dname = kmalloc(name->len + 1, GFP_KERNEL);
719 kmem_cache_free(dentry_cache, dentry);
723 dname = dentry->d_iname;
725 dentry->d_name.name = dname;
727 dentry->d_name.len = name->len;
728 dentry->d_name.hash = name->hash;
729 memcpy(dname, name->name, name->len);
730 dname[name->len] = 0;
732 atomic_set(&dentry->d_count, 1);
733 dentry->d_flags = DCACHE_UNHASHED;
734 spin_lock_init(&dentry->d_lock);
735 dentry->d_inode = NULL;
736 dentry->d_parent = NULL;
739 dentry->d_fsdata = NULL;
740 dentry->d_mounted = 0;
741 #ifdef CONFIG_PROFILING
742 dentry->d_cookie = NULL;
744 INIT_HLIST_NODE(&dentry->d_hash);
745 INIT_LIST_HEAD(&dentry->d_lru);
746 INIT_LIST_HEAD(&dentry->d_subdirs);
747 INIT_LIST_HEAD(&dentry->d_alias);
750 dentry->d_parent = dget(parent);
751 dentry->d_sb = parent->d_sb;
753 INIT_LIST_HEAD(&dentry->d_u.d_child);
756 spin_lock(&dcache_lock);
758 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
759 dentry_stat.nr_dentry++;
760 spin_unlock(&dcache_lock);
765 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
770 q.len = strlen(name);
771 q.hash = full_name_hash(q.name, q.len);
772 return d_alloc(parent, &q);
776 * d_instantiate - fill in inode information for a dentry
777 * @entry: dentry to complete
778 * @inode: inode to attach to this dentry
780 * Fill in inode information in the entry.
782 * This turns negative dentries into productive full members
785 * NOTE! This assumes that the inode count has been incremented
786 * (or otherwise set) by the caller to indicate that it is now
787 * in use by the dcache.
790 void d_instantiate(struct dentry *entry, struct inode * inode)
792 BUG_ON(!list_empty(&entry->d_alias));
793 spin_lock(&dcache_lock);
795 list_add(&entry->d_alias, &inode->i_dentry);
796 entry->d_inode = inode;
797 fsnotify_d_instantiate(entry, inode);
798 spin_unlock(&dcache_lock);
799 security_d_instantiate(entry, inode);
803 * d_instantiate_unique - instantiate a non-aliased dentry
804 * @entry: dentry to instantiate
805 * @inode: inode to attach to this dentry
807 * Fill in inode information in the entry. On success, it returns NULL.
808 * If an unhashed alias of "entry" already exists, then we return the
809 * aliased dentry instead and drop one reference to inode.
811 * Note that in order to avoid conflicts with rename() etc, the caller
812 * had better be holding the parent directory semaphore.
814 * This also assumes that the inode count has been incremented
815 * (or otherwise set) by the caller to indicate that it is now
816 * in use by the dcache.
818 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
820 struct dentry *alias;
821 int len = entry->d_name.len;
822 const char *name = entry->d_name.name;
823 unsigned int hash = entry->d_name.hash;
825 BUG_ON(!list_empty(&entry->d_alias));
826 spin_lock(&dcache_lock);
829 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
830 struct qstr *qstr = &alias->d_name;
832 if (qstr->hash != hash)
834 if (alias->d_parent != entry->d_parent)
836 if (qstr->len != len)
838 if (memcmp(qstr->name, name, len))
841 spin_unlock(&dcache_lock);
842 BUG_ON(!d_unhashed(alias));
846 list_add(&entry->d_alias, &inode->i_dentry);
848 entry->d_inode = inode;
849 fsnotify_d_instantiate(entry, inode);
850 spin_unlock(&dcache_lock);
851 security_d_instantiate(entry, inode);
854 EXPORT_SYMBOL(d_instantiate_unique);
857 * d_alloc_root - allocate root dentry
858 * @root_inode: inode to allocate the root for
860 * Allocate a root ("/") dentry for the inode given. The inode is
861 * instantiated and returned. %NULL is returned if there is insufficient
862 * memory or the inode passed is %NULL.
865 struct dentry * d_alloc_root(struct inode * root_inode)
867 struct dentry *res = NULL;
870 static const struct qstr name = { .name = "/", .len = 1 };
872 res = d_alloc(NULL, &name);
874 res->d_sb = root_inode->i_sb;
876 d_instantiate(res, root_inode);
882 static inline struct hlist_head *d_hash(struct dentry *parent,
885 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
886 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
887 return dentry_hashtable + (hash & D_HASHMASK);
891 * d_alloc_anon - allocate an anonymous dentry
892 * @inode: inode to allocate the dentry for
894 * This is similar to d_alloc_root. It is used by filesystems when
895 * creating a dentry for a given inode, often in the process of
896 * mapping a filehandle to a dentry. The returned dentry may be
897 * anonymous, or may have a full name (if the inode was already
898 * in the cache). The file system may need to make further
899 * efforts to connect this dentry into the dcache properly.
901 * When called on a directory inode, we must ensure that
902 * the inode only ever has one dentry. If a dentry is
903 * found, that is returned instead of allocating a new one.
905 * On successful return, the reference to the inode has been transferred
906 * to the dentry. If %NULL is returned (indicating kmalloc failure),
907 * the reference on the inode has not been released.
910 struct dentry * d_alloc_anon(struct inode *inode)
912 static const struct qstr anonstring = { .name = "" };
916 if ((res = d_find_alias(inode))) {
921 tmp = d_alloc(NULL, &anonstring);
925 tmp->d_parent = tmp; /* make sure dput doesn't croak */
927 spin_lock(&dcache_lock);
928 res = __d_find_alias(inode, 0);
930 /* attach a disconnected dentry */
933 spin_lock(&res->d_lock);
934 res->d_sb = inode->i_sb;
936 res->d_inode = inode;
937 res->d_flags |= DCACHE_DISCONNECTED;
938 res->d_flags &= ~DCACHE_UNHASHED;
939 list_add(&res->d_alias, &inode->i_dentry);
940 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
941 spin_unlock(&res->d_lock);
943 inode = NULL; /* don't drop reference */
945 spin_unlock(&dcache_lock);
956 * d_splice_alias - splice a disconnected dentry into the tree if one exists
957 * @inode: the inode which may have a disconnected dentry
958 * @dentry: a negative dentry which we want to point to the inode.
960 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
961 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
962 * and return it, else simply d_add the inode to the dentry and return NULL.
964 * This is needed in the lookup routine of any filesystem that is exportable
965 * (via knfsd) so that we can build dcache paths to directories effectively.
967 * If a dentry was found and moved, then it is returned. Otherwise NULL
968 * is returned. This matches the expected return value of ->lookup.
971 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
973 struct dentry *new = NULL;
976 spin_lock(&dcache_lock);
977 new = __d_find_alias(inode, 1);
979 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
980 fsnotify_d_instantiate(new, inode);
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 fsnotify_d_instantiate(dentry, inode);
991 spin_unlock(&dcache_lock);
992 security_d_instantiate(dentry, inode);
996 d_add(dentry, inode);
1002 * d_lookup - search for a dentry
1003 * @parent: parent dentry
1004 * @name: qstr of name we wish to find
1006 * Searches the children of the parent dentry for the name in question. If
1007 * the dentry is found its reference count is incremented and the dentry
1008 * is returned. The caller must use d_put to free the entry when it has
1009 * finished using it. %NULL is returned on failure.
1011 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1012 * Memory barriers are used while updating and doing lockless traversal.
1013 * To avoid races with d_move while rename is happening, d_lock is used.
1015 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1016 * and name pointer in one structure pointed by d_qstr.
1018 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1019 * lookup is going on.
1021 * dentry_unused list is not updated even if lookup finds the required dentry
1022 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1023 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1026 * d_lookup() is protected against the concurrent renames in some unrelated
1027 * directory using the seqlockt_t rename_lock.
1030 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1032 struct dentry * dentry = NULL;
1036 seq = read_seqbegin(&rename_lock);
1037 dentry = __d_lookup(parent, name);
1040 } while (read_seqretry(&rename_lock, seq));
1044 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1046 unsigned int len = name->len;
1047 unsigned int hash = name->hash;
1048 const unsigned char *str = name->name;
1049 struct hlist_head *head = d_hash(parent,hash);
1050 struct dentry *found = NULL;
1051 struct hlist_node *node;
1052 struct dentry *dentry;
1056 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1059 if (dentry->d_name.hash != hash)
1061 if (dentry->d_parent != parent)
1064 spin_lock(&dentry->d_lock);
1067 * Recheck the dentry after taking the lock - d_move may have
1068 * changed things. Don't bother checking the hash because we're
1069 * about to compare the whole name anyway.
1071 if (dentry->d_parent != parent)
1075 * It is safe to compare names since d_move() cannot
1076 * change the qstr (protected by d_lock).
1078 qstr = &dentry->d_name;
1079 if (parent->d_op && parent->d_op->d_compare) {
1080 if (parent->d_op->d_compare(parent, qstr, name))
1083 if (qstr->len != len)
1085 if (memcmp(qstr->name, str, len))
1089 if (!d_unhashed(dentry)) {
1090 atomic_inc(&dentry->d_count);
1093 spin_unlock(&dentry->d_lock);
1096 spin_unlock(&dentry->d_lock);
1104 * d_hash_and_lookup - hash the qstr then search for a dentry
1105 * @dir: Directory to search in
1106 * @name: qstr of name we wish to find
1108 * On hash failure or on lookup failure NULL is returned.
1110 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1112 struct dentry *dentry = NULL;
1115 * Check for a fs-specific hash function. Note that we must
1116 * calculate the standard hash first, as the d_op->d_hash()
1117 * routine may choose to leave the hash value unchanged.
1119 name->hash = full_name_hash(name->name, name->len);
1120 if (dir->d_op && dir->d_op->d_hash) {
1121 if (dir->d_op->d_hash(dir, name) < 0)
1124 dentry = d_lookup(dir, name);
1130 * d_validate - verify dentry provided from insecure source
1131 * @dentry: The dentry alleged to be valid child of @dparent
1132 * @dparent: The parent dentry (known to be valid)
1133 * @hash: Hash of the dentry
1134 * @len: Length of the name
1136 * An insecure source has sent us a dentry, here we verify it and dget() it.
1137 * This is used by ncpfs in its readdir implementation.
1138 * Zero is returned in the dentry is invalid.
1141 int d_validate(struct dentry *dentry, struct dentry *dparent)
1143 struct hlist_head *base;
1144 struct hlist_node *lhp;
1146 /* Check whether the ptr might be valid at all.. */
1147 if (!kmem_ptr_validate(dentry_cache, dentry))
1150 if (dentry->d_parent != dparent)
1153 spin_lock(&dcache_lock);
1154 base = d_hash(dparent, dentry->d_name.hash);
1155 hlist_for_each(lhp,base) {
1156 /* hlist_for_each_entry_rcu() not required for d_hash list
1157 * as it is parsed under dcache_lock
1159 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1160 __dget_locked(dentry);
1161 spin_unlock(&dcache_lock);
1165 spin_unlock(&dcache_lock);
1171 * When a file is deleted, we have two options:
1172 * - turn this dentry into a negative dentry
1173 * - unhash this dentry and free it.
1175 * Usually, we want to just turn this into
1176 * a negative dentry, but if anybody else is
1177 * currently using the dentry or the inode
1178 * we can't do that and we fall back on removing
1179 * it from the hash queues and waiting for
1180 * it to be deleted later when it has no users
1184 * d_delete - delete a dentry
1185 * @dentry: The dentry to delete
1187 * Turn the dentry into a negative dentry if possible, otherwise
1188 * remove it from the hash queues so it can be deleted later
1191 void d_delete(struct dentry * dentry)
1195 * Are we the only user?
1197 spin_lock(&dcache_lock);
1198 spin_lock(&dentry->d_lock);
1199 isdir = S_ISDIR(dentry->d_inode->i_mode);
1200 if (atomic_read(&dentry->d_count) == 1) {
1201 dentry_iput(dentry);
1202 fsnotify_nameremove(dentry, isdir);
1204 /* remove this and other inotify debug checks after 2.6.18 */
1205 dentry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
1209 if (!d_unhashed(dentry))
1212 spin_unlock(&dentry->d_lock);
1213 spin_unlock(&dcache_lock);
1215 fsnotify_nameremove(dentry, isdir);
1218 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1221 entry->d_flags &= ~DCACHE_UNHASHED;
1222 hlist_add_head_rcu(&entry->d_hash, list);
1226 * d_rehash - add an entry back to the hash
1227 * @entry: dentry to add to the hash
1229 * Adds a dentry to the hash according to its name.
1232 void d_rehash(struct dentry * entry)
1234 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1236 spin_lock(&dcache_lock);
1237 spin_lock(&entry->d_lock);
1238 __d_rehash(entry, list);
1239 spin_unlock(&entry->d_lock);
1240 spin_unlock(&dcache_lock);
1243 #define do_switch(x,y) do { \
1244 __typeof__ (x) __tmp = x; \
1245 x = y; y = __tmp; } while (0)
1248 * When switching names, the actual string doesn't strictly have to
1249 * be preserved in the target - because we're dropping the target
1250 * anyway. As such, we can just do a simple memcpy() to copy over
1251 * the new name before we switch.
1253 * Note that we have to be a lot more careful about getting the hash
1254 * switched - we have to switch the hash value properly even if it
1255 * then no longer matches the actual (corrupted) string of the target.
1256 * The hash value has to match the hash queue that the dentry is on..
1258 static void switch_names(struct dentry *dentry, struct dentry *target)
1260 if (dname_external(target)) {
1261 if (dname_external(dentry)) {
1263 * Both external: swap the pointers
1265 do_switch(target->d_name.name, dentry->d_name.name);
1268 * dentry:internal, target:external. Steal target's
1269 * storage and make target internal.
1271 dentry->d_name.name = target->d_name.name;
1272 target->d_name.name = target->d_iname;
1275 if (dname_external(dentry)) {
1277 * dentry:external, target:internal. Give dentry's
1278 * storage to target and make dentry internal
1280 memcpy(dentry->d_iname, target->d_name.name,
1281 target->d_name.len + 1);
1282 target->d_name.name = dentry->d_name.name;
1283 dentry->d_name.name = dentry->d_iname;
1286 * Both are internal. Just copy target to dentry
1288 memcpy(dentry->d_iname, target->d_name.name,
1289 target->d_name.len + 1);
1295 * We cannibalize "target" when moving dentry on top of it,
1296 * because it's going to be thrown away anyway. We could be more
1297 * polite about it, though.
1299 * This forceful removal will result in ugly /proc output if
1300 * somebody holds a file open that got deleted due to a rename.
1301 * We could be nicer about the deleted file, and let it show
1302 * up under the name it got deleted rather than the name that
1307 * d_move - move a dentry
1308 * @dentry: entry to move
1309 * @target: new dentry
1311 * Update the dcache to reflect the move of a file name. Negative
1312 * dcache entries should not be moved in this way.
1315 void d_move(struct dentry * dentry, struct dentry * target)
1317 struct hlist_head *list;
1319 if (!dentry->d_inode)
1320 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1322 spin_lock(&dcache_lock);
1323 write_seqlock(&rename_lock);
1325 * XXXX: do we really need to take target->d_lock?
1327 if (target < dentry) {
1328 spin_lock(&target->d_lock);
1329 spin_lock(&dentry->d_lock);
1331 spin_lock(&dentry->d_lock);
1332 spin_lock(&target->d_lock);
1335 /* Move the dentry to the target hash queue, if on different bucket */
1336 if (dentry->d_flags & DCACHE_UNHASHED)
1337 goto already_unhashed;
1339 hlist_del_rcu(&dentry->d_hash);
1342 list = d_hash(target->d_parent, target->d_name.hash);
1343 __d_rehash(dentry, list);
1345 /* Unhash the target: dput() will then get rid of it */
1348 list_del(&dentry->d_u.d_child);
1349 list_del(&target->d_u.d_child);
1351 /* Switch the names.. */
1352 switch_names(dentry, target);
1353 do_switch(dentry->d_name.len, target->d_name.len);
1354 do_switch(dentry->d_name.hash, target->d_name.hash);
1356 /* ... and switch the parents */
1357 if (IS_ROOT(dentry)) {
1358 dentry->d_parent = target->d_parent;
1359 target->d_parent = target;
1360 INIT_LIST_HEAD(&target->d_u.d_child);
1362 do_switch(dentry->d_parent, target->d_parent);
1364 /* And add them back to the (new) parent lists */
1365 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1368 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1369 spin_unlock(&target->d_lock);
1370 fsnotify_d_move(dentry);
1371 spin_unlock(&dentry->d_lock);
1372 write_sequnlock(&rename_lock);
1373 spin_unlock(&dcache_lock);
1377 * d_path - return the path of a dentry
1378 * @dentry: dentry to report
1379 * @vfsmnt: vfsmnt to which the dentry belongs
1380 * @root: root dentry
1381 * @rootmnt: vfsmnt to which the root dentry belongs
1382 * @buffer: buffer to return value in
1383 * @buflen: buffer length
1385 * Convert a dentry into an ASCII path name. If the entry has been deleted
1386 * the string " (deleted)" is appended. Note that this is ambiguous.
1388 * Returns the buffer or an error code if the path was too long.
1390 * "buflen" should be positive. Caller holds the dcache_lock.
1392 static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1393 struct dentry *root, struct vfsmount *rootmnt,
1394 char *buffer, int buflen)
1396 char * end = buffer+buflen;
1402 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1407 memcpy(end, " (deleted)", 10);
1417 struct dentry * parent;
1419 if (dentry == root && vfsmnt == rootmnt)
1421 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1423 spin_lock(&vfsmount_lock);
1424 if (vfsmnt->mnt_parent == vfsmnt) {
1425 spin_unlock(&vfsmount_lock);
1428 dentry = vfsmnt->mnt_mountpoint;
1429 vfsmnt = vfsmnt->mnt_parent;
1430 spin_unlock(&vfsmount_lock);
1433 parent = dentry->d_parent;
1435 namelen = dentry->d_name.len;
1436 buflen -= namelen + 1;
1440 memcpy(end, dentry->d_name.name, namelen);
1449 namelen = dentry->d_name.len;
1453 retval -= namelen-1; /* hit the slash */
1454 memcpy(retval, dentry->d_name.name, namelen);
1457 return ERR_PTR(-ENAMETOOLONG);
1460 /* write full pathname into buffer and return start of pathname */
1461 char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1462 char *buf, int buflen)
1465 struct vfsmount *rootmnt;
1466 struct dentry *root;
1468 read_lock(¤t->fs->lock);
1469 rootmnt = mntget(current->fs->rootmnt);
1470 root = dget(current->fs->root);
1471 read_unlock(¤t->fs->lock);
1472 spin_lock(&dcache_lock);
1473 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1474 spin_unlock(&dcache_lock);
1481 * NOTE! The user-level library version returns a
1482 * character pointer. The kernel system call just
1483 * returns the length of the buffer filled (which
1484 * includes the ending '\0' character), or a negative
1485 * error value. So libc would do something like
1487 * char *getcwd(char * buf, size_t size)
1491 * retval = sys_getcwd(buf, size);
1498 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1501 struct vfsmount *pwdmnt, *rootmnt;
1502 struct dentry *pwd, *root;
1503 char *page = (char *) __get_free_page(GFP_USER);
1508 read_lock(¤t->fs->lock);
1509 pwdmnt = mntget(current->fs->pwdmnt);
1510 pwd = dget(current->fs->pwd);
1511 rootmnt = mntget(current->fs->rootmnt);
1512 root = dget(current->fs->root);
1513 read_unlock(¤t->fs->lock);
1516 /* Has the current directory has been unlinked? */
1517 spin_lock(&dcache_lock);
1518 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1522 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1523 spin_unlock(&dcache_lock);
1525 error = PTR_ERR(cwd);
1530 len = PAGE_SIZE + page - cwd;
1533 if (copy_to_user(buf, cwd, len))
1537 spin_unlock(&dcache_lock);
1544 free_page((unsigned long) page);
1549 * Test whether new_dentry is a subdirectory of old_dentry.
1551 * Trivially implemented using the dcache structure
1555 * is_subdir - is new dentry a subdirectory of old_dentry
1556 * @new_dentry: new dentry
1557 * @old_dentry: old dentry
1559 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1560 * Returns 0 otherwise.
1561 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1564 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1567 struct dentry * saved = new_dentry;
1570 /* need rcu_readlock to protect against the d_parent trashing due to
1575 /* for restarting inner loop in case of seq retry */
1578 seq = read_seqbegin(&rename_lock);
1580 if (new_dentry != old_dentry) {
1581 struct dentry * parent = new_dentry->d_parent;
1582 if (parent == new_dentry)
1584 new_dentry = parent;
1590 } while (read_seqretry(&rename_lock, seq));
1596 void d_genocide(struct dentry *root)
1598 struct dentry *this_parent = root;
1599 struct list_head *next;
1601 spin_lock(&dcache_lock);
1603 next = this_parent->d_subdirs.next;
1605 while (next != &this_parent->d_subdirs) {
1606 struct list_head *tmp = next;
1607 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1609 if (d_unhashed(dentry)||!dentry->d_inode)
1611 if (!list_empty(&dentry->d_subdirs)) {
1612 this_parent = dentry;
1615 atomic_dec(&dentry->d_count);
1617 if (this_parent != root) {
1618 next = this_parent->d_u.d_child.next;
1619 atomic_dec(&this_parent->d_count);
1620 this_parent = this_parent->d_parent;
1623 spin_unlock(&dcache_lock);
1627 * find_inode_number - check for dentry with name
1628 * @dir: directory to check
1629 * @name: Name to find.
1631 * Check whether a dentry already exists for the given name,
1632 * and return the inode number if it has an inode. Otherwise
1635 * This routine is used to post-process directory listings for
1636 * filesystems using synthetic inode numbers, and is necessary
1637 * to keep getcwd() working.
1640 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1642 struct dentry * dentry;
1645 dentry = d_hash_and_lookup(dir, name);
1647 if (dentry->d_inode)
1648 ino = dentry->d_inode->i_ino;
1654 static __initdata unsigned long dhash_entries;
1655 static int __init set_dhash_entries(char *str)
1659 dhash_entries = simple_strtoul(str, &str, 0);
1662 __setup("dhash_entries=", set_dhash_entries);
1664 static void __init dcache_init_early(void)
1668 /* If hashes are distributed across NUMA nodes, defer
1669 * hash allocation until vmalloc space is available.
1675 alloc_large_system_hash("Dentry cache",
1676 sizeof(struct hlist_head),
1684 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1685 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1688 static void __init dcache_init(unsigned long mempages)
1693 * A constructor could be added for stable state like the lists,
1694 * but it is probably not worth it because of the cache nature
1697 dentry_cache = kmem_cache_create("dentry_cache",
1698 sizeof(struct dentry),
1700 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1704 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1706 /* Hash may have been set up in dcache_init_early */
1711 alloc_large_system_hash("Dentry cache",
1712 sizeof(struct hlist_head),
1720 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1721 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1724 /* SLAB cache for __getname() consumers */
1725 kmem_cache_t *names_cachep __read_mostly;
1727 /* SLAB cache for file structures */
1728 kmem_cache_t *filp_cachep __read_mostly;
1730 EXPORT_SYMBOL(d_genocide);
1732 extern void bdev_cache_init(void);
1733 extern void chrdev_init(void);
1735 void __init vfs_caches_init_early(void)
1737 dcache_init_early();
1741 void __init vfs_caches_init(unsigned long mempages)
1743 unsigned long reserve;
1745 /* Base hash sizes on available memory, with a reserve equal to
1746 150% of current kernel size */
1748 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1749 mempages -= reserve;
1751 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1752 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1754 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1755 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1757 dcache_init(mempages);
1758 inode_init(mempages);
1759 files_init(mempages);
1765 EXPORT_SYMBOL(d_alloc);
1766 EXPORT_SYMBOL(d_alloc_anon);
1767 EXPORT_SYMBOL(d_alloc_root);
1768 EXPORT_SYMBOL(d_delete);
1769 EXPORT_SYMBOL(d_find_alias);
1770 EXPORT_SYMBOL(d_instantiate);
1771 EXPORT_SYMBOL(d_invalidate);
1772 EXPORT_SYMBOL(d_lookup);
1773 EXPORT_SYMBOL(d_move);
1774 EXPORT_SYMBOL(d_path);
1775 EXPORT_SYMBOL(d_prune_aliases);
1776 EXPORT_SYMBOL(d_rehash);
1777 EXPORT_SYMBOL(d_splice_alias);
1778 EXPORT_SYMBOL(d_validate);
1779 EXPORT_SYMBOL(dget_locked);
1780 EXPORT_SYMBOL(dput);
1781 EXPORT_SYMBOL(find_inode_number);
1782 EXPORT_SYMBOL(have_submounts);
1783 EXPORT_SYMBOL(names_cachep);
1784 EXPORT_SYMBOL(shrink_dcache_parent);
1785 EXPORT_SYMBOL(shrink_dcache_sb);