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/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.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 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
44 EXPORT_SYMBOL(dcache_lock);
46 static struct kmem_cache *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;
65 /* Statistics gathering. */
66 struct dentry_stat_t dentry_stat = {
70 static void __d_free(struct dentry *dentry)
72 WARN_ON(!list_empty(&dentry->d_alias));
73 if (dname_external(dentry))
74 kfree(dentry->d_name.name);
75 kmem_cache_free(dentry_cache, dentry);
78 static void d_callback(struct rcu_head *head)
80 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
85 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
88 static void d_free(struct dentry *dentry)
90 if (dentry->d_op && dentry->d_op->d_release)
91 dentry->d_op->d_release(dentry);
92 /* if dentry was never inserted into hash, immediate free is OK */
93 if (hlist_unhashed(&dentry->d_hash))
96 call_rcu(&dentry->d_u.d_rcu, d_callback);
100 * Release the dentry's inode, using the filesystem
101 * d_iput() operation if defined.
103 static void dentry_iput(struct dentry * dentry)
104 __releases(dentry->d_lock)
105 __releases(dcache_lock)
107 struct inode *inode = dentry->d_inode;
109 dentry->d_inode = NULL;
110 list_del_init(&dentry->d_alias);
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
114 fsnotify_inoderemove(inode);
115 if (dentry->d_op && dentry->d_op->d_iput)
116 dentry->d_op->d_iput(dentry, inode);
120 spin_unlock(&dentry->d_lock);
121 spin_unlock(&dcache_lock);
126 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
128 static void dentry_lru_add(struct dentry *dentry)
130 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
131 dentry->d_sb->s_nr_dentry_unused++;
132 dentry_stat.nr_unused++;
135 static void dentry_lru_add_tail(struct dentry *dentry)
137 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
138 dentry->d_sb->s_nr_dentry_unused++;
139 dentry_stat.nr_unused++;
142 static void dentry_lru_del(struct dentry *dentry)
144 if (!list_empty(&dentry->d_lru)) {
145 list_del(&dentry->d_lru);
146 dentry->d_sb->s_nr_dentry_unused--;
147 dentry_stat.nr_unused--;
151 static void dentry_lru_del_init(struct dentry *dentry)
153 if (likely(!list_empty(&dentry->d_lru))) {
154 list_del_init(&dentry->d_lru);
155 dentry->d_sb->s_nr_dentry_unused--;
156 dentry_stat.nr_unused--;
161 * d_kill - kill dentry and return parent
162 * @dentry: dentry to kill
164 * The dentry must already be unhashed and removed from the LRU.
166 * If this is the root of the dentry tree, return NULL.
168 static struct dentry *d_kill(struct dentry *dentry)
169 __releases(dentry->d_lock)
170 __releases(dcache_lock)
172 struct dentry *parent;
174 list_del(&dentry->d_u.d_child);
175 dentry_stat.nr_dentry--; /* For d_free, below */
176 /*drops the locks, at that point nobody can reach this dentry */
181 parent = dentry->d_parent;
189 * This is complicated by the fact that we do not want to put
190 * dentries that are no longer on any hash chain on the unused
191 * list: we'd much rather just get rid of them immediately.
193 * However, that implies that we have to traverse the dentry
194 * tree upwards to the parents which might _also_ now be
195 * scheduled for deletion (it may have been only waiting for
196 * its last child to go away).
198 * This tail recursion is done by hand as we don't want to depend
199 * on the compiler to always get this right (gcc generally doesn't).
200 * Real recursion would eat up our stack space.
204 * dput - release a dentry
205 * @dentry: dentry to release
207 * Release a dentry. This will drop the usage count and if appropriate
208 * call the dentry unlink method as well as removing it from the queues and
209 * releasing its resources. If the parent dentries were scheduled for release
210 * they too may now get deleted.
212 * no dcache lock, please.
215 void dput(struct dentry *dentry)
221 if (atomic_read(&dentry->d_count) == 1)
223 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
226 spin_lock(&dentry->d_lock);
227 if (atomic_read(&dentry->d_count)) {
228 spin_unlock(&dentry->d_lock);
229 spin_unlock(&dcache_lock);
234 * AV: ->d_delete() is _NOT_ allowed to block now.
236 if (dentry->d_op && dentry->d_op->d_delete) {
237 if (dentry->d_op->d_delete(dentry))
240 /* Unreachable? Get rid of it */
241 if (d_unhashed(dentry))
243 if (list_empty(&dentry->d_lru)) {
244 dentry->d_flags |= DCACHE_REFERENCED;
245 dentry_lru_add(dentry);
247 spin_unlock(&dentry->d_lock);
248 spin_unlock(&dcache_lock);
254 /* if dentry was on the d_lru list delete it from there */
255 dentry_lru_del(dentry);
256 dentry = d_kill(dentry);
263 * d_invalidate - invalidate a dentry
264 * @dentry: dentry to invalidate
266 * Try to invalidate the dentry if it turns out to be
267 * possible. If there are other dentries that can be
268 * reached through this one we can't delete it and we
269 * return -EBUSY. On success we return 0.
274 int d_invalidate(struct dentry * dentry)
277 * If it's already been dropped, return OK.
279 spin_lock(&dcache_lock);
280 if (d_unhashed(dentry)) {
281 spin_unlock(&dcache_lock);
285 * Check whether to do a partial shrink_dcache
286 * to get rid of unused child entries.
288 if (!list_empty(&dentry->d_subdirs)) {
289 spin_unlock(&dcache_lock);
290 shrink_dcache_parent(dentry);
291 spin_lock(&dcache_lock);
295 * Somebody else still using it?
297 * If it's a directory, we can't drop it
298 * for fear of somebody re-populating it
299 * with children (even though dropping it
300 * would make it unreachable from the root,
301 * we might still populate it if it was a
302 * working directory or similar).
304 spin_lock(&dentry->d_lock);
305 if (atomic_read(&dentry->d_count) > 1) {
306 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
307 spin_unlock(&dentry->d_lock);
308 spin_unlock(&dcache_lock);
314 spin_unlock(&dentry->d_lock);
315 spin_unlock(&dcache_lock);
318 EXPORT_SYMBOL(d_invalidate);
320 /* This should be called _only_ with dcache_lock held */
322 static inline struct dentry * __dget_locked(struct dentry *dentry)
324 atomic_inc(&dentry->d_count);
325 dentry_lru_del_init(dentry);
329 struct dentry * dget_locked(struct dentry *dentry)
331 return __dget_locked(dentry);
333 EXPORT_SYMBOL(dget_locked);
336 * d_find_alias - grab a hashed alias of inode
337 * @inode: inode in question
338 * @want_discon: flag, used by d_splice_alias, to request
339 * that only a DISCONNECTED alias be returned.
341 * If inode has a hashed alias, or is a directory and has any alias,
342 * acquire the reference to alias and return it. Otherwise return NULL.
343 * Notice that if inode is a directory there can be only one alias and
344 * it can be unhashed only if it has no children, or if it is the root
347 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
348 * any other hashed alias over that one unless @want_discon is set,
349 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
352 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
354 struct list_head *head, *next, *tmp;
355 struct dentry *alias, *discon_alias=NULL;
357 head = &inode->i_dentry;
358 next = inode->i_dentry.next;
359 while (next != head) {
363 alias = list_entry(tmp, struct dentry, d_alias);
364 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
365 if (IS_ROOT(alias) &&
366 (alias->d_flags & DCACHE_DISCONNECTED))
367 discon_alias = alias;
368 else if (!want_discon) {
369 __dget_locked(alias);
375 __dget_locked(discon_alias);
379 struct dentry * d_find_alias(struct inode *inode)
381 struct dentry *de = NULL;
383 if (!list_empty(&inode->i_dentry)) {
384 spin_lock(&dcache_lock);
385 de = __d_find_alias(inode, 0);
386 spin_unlock(&dcache_lock);
390 EXPORT_SYMBOL(d_find_alias);
393 * Try to kill dentries associated with this inode.
394 * WARNING: you must own a reference to inode.
396 void d_prune_aliases(struct inode *inode)
398 struct dentry *dentry;
400 spin_lock(&dcache_lock);
401 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
402 spin_lock(&dentry->d_lock);
403 if (!atomic_read(&dentry->d_count)) {
404 __dget_locked(dentry);
406 spin_unlock(&dentry->d_lock);
407 spin_unlock(&dcache_lock);
411 spin_unlock(&dentry->d_lock);
413 spin_unlock(&dcache_lock);
415 EXPORT_SYMBOL(d_prune_aliases);
418 * Throw away a dentry - free the inode, dput the parent. This requires that
419 * the LRU list has already been removed.
421 * Try to prune ancestors as well. This is necessary to prevent
422 * quadratic behavior of shrink_dcache_parent(), but is also expected
423 * to be beneficial in reducing dentry cache fragmentation.
425 static void prune_one_dentry(struct dentry * dentry)
426 __releases(dentry->d_lock)
427 __releases(dcache_lock)
428 __acquires(dcache_lock)
431 dentry = d_kill(dentry);
434 * Prune ancestors. Locking is simpler than in dput(),
435 * because dcache_lock needs to be taken anyway.
437 spin_lock(&dcache_lock);
439 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
442 if (dentry->d_op && dentry->d_op->d_delete)
443 dentry->d_op->d_delete(dentry);
444 dentry_lru_del_init(dentry);
446 dentry = d_kill(dentry);
447 spin_lock(&dcache_lock);
452 * Shrink the dentry LRU on a given superblock.
453 * @sb : superblock to shrink dentry LRU.
454 * @count: If count is NULL, we prune all dentries on superblock.
455 * @flags: If flags is non-zero, we need to do special processing based on
456 * which flags are set. This means we don't need to maintain multiple
457 * similar copies of this loop.
459 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
461 LIST_HEAD(referenced);
463 struct dentry *dentry;
467 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
468 spin_lock(&dcache_lock);
470 /* called from prune_dcache() and shrink_dcache_parent() */
474 list_splice_init(&sb->s_dentry_lru, &tmp);
476 while (!list_empty(&sb->s_dentry_lru)) {
477 dentry = list_entry(sb->s_dentry_lru.prev,
478 struct dentry, d_lru);
479 BUG_ON(dentry->d_sb != sb);
481 spin_lock(&dentry->d_lock);
483 * If we are honouring the DCACHE_REFERENCED flag and
484 * the dentry has this flag set, don't free it. Clear
485 * the flag and put it back on the LRU.
487 if ((flags & DCACHE_REFERENCED)
488 && (dentry->d_flags & DCACHE_REFERENCED)) {
489 dentry->d_flags &= ~DCACHE_REFERENCED;
490 list_move(&dentry->d_lru, &referenced);
491 spin_unlock(&dentry->d_lock);
493 list_move_tail(&dentry->d_lru, &tmp);
494 spin_unlock(&dentry->d_lock);
499 cond_resched_lock(&dcache_lock);
502 while (!list_empty(&tmp)) {
503 dentry = list_entry(tmp.prev, struct dentry, d_lru);
504 dentry_lru_del_init(dentry);
505 spin_lock(&dentry->d_lock);
507 * We found an inuse dentry which was not removed from
508 * the LRU because of laziness during lookup. Do not free
509 * it - just keep it off the LRU list.
511 if (atomic_read(&dentry->d_count)) {
512 spin_unlock(&dentry->d_lock);
515 prune_one_dentry(dentry);
516 /* dentry->d_lock was dropped in prune_one_dentry() */
517 cond_resched_lock(&dcache_lock);
519 if (count == NULL && !list_empty(&sb->s_dentry_lru))
523 if (!list_empty(&referenced))
524 list_splice(&referenced, &sb->s_dentry_lru);
525 spin_unlock(&dcache_lock);
529 * prune_dcache - shrink the dcache
530 * @count: number of entries to try to free
532 * Shrink the dcache. This is done when we need more memory, or simply when we
533 * need to unmount something (at which point we need to unuse all dentries).
535 * This function may fail to free any resources if all the dentries are in use.
537 static void prune_dcache(int count)
539 struct super_block *sb;
541 int unused = dentry_stat.nr_unused;
545 if (unused == 0 || count == 0)
547 spin_lock(&dcache_lock);
552 prune_ratio = unused / count;
554 list_for_each_entry(sb, &super_blocks, s_list) {
555 if (list_empty(&sb->s_instances))
557 if (sb->s_nr_dentry_unused == 0)
560 /* Now, we reclaim unused dentrins with fairness.
561 * We reclaim them same percentage from each superblock.
562 * We calculate number of dentries to scan on this sb
563 * as follows, but the implementation is arranged to avoid
565 * number of dentries to scan on this sb =
566 * count * (number of dentries on this sb /
567 * number of dentries in the machine)
569 spin_unlock(&sb_lock);
570 if (prune_ratio != 1)
571 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
573 w_count = sb->s_nr_dentry_unused;
576 * We need to be sure this filesystem isn't being unmounted,
577 * otherwise we could race with generic_shutdown_super(), and
578 * end up holding a reference to an inode while the filesystem
579 * is unmounted. So we try to get s_umount, and make sure
582 if (down_read_trylock(&sb->s_umount)) {
583 if ((sb->s_root != NULL) &&
584 (!list_empty(&sb->s_dentry_lru))) {
585 spin_unlock(&dcache_lock);
586 __shrink_dcache_sb(sb, &w_count,
589 spin_lock(&dcache_lock);
591 up_read(&sb->s_umount);
596 * restart only when sb is no longer on the list and
597 * we have more work to do.
599 if (__put_super_and_need_restart(sb) && count > 0) {
600 spin_unlock(&sb_lock);
604 spin_unlock(&sb_lock);
605 spin_unlock(&dcache_lock);
609 * shrink_dcache_sb - shrink dcache for a superblock
612 * Shrink the dcache for the specified super block. This
613 * is used to free the dcache before unmounting a file
616 void shrink_dcache_sb(struct super_block * sb)
618 __shrink_dcache_sb(sb, NULL, 0);
620 EXPORT_SYMBOL(shrink_dcache_sb);
623 * destroy a single subtree of dentries for unmount
624 * - see the comments on shrink_dcache_for_umount() for a description of the
627 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
629 struct dentry *parent;
630 unsigned detached = 0;
632 BUG_ON(!IS_ROOT(dentry));
634 /* detach this root from the system */
635 spin_lock(&dcache_lock);
636 dentry_lru_del_init(dentry);
638 spin_unlock(&dcache_lock);
641 /* descend to the first leaf in the current subtree */
642 while (!list_empty(&dentry->d_subdirs)) {
645 /* this is a branch with children - detach all of them
646 * from the system in one go */
647 spin_lock(&dcache_lock);
648 list_for_each_entry(loop, &dentry->d_subdirs,
650 dentry_lru_del_init(loop);
652 cond_resched_lock(&dcache_lock);
654 spin_unlock(&dcache_lock);
656 /* move to the first child */
657 dentry = list_entry(dentry->d_subdirs.next,
658 struct dentry, d_u.d_child);
661 /* consume the dentries from this leaf up through its parents
662 * until we find one with children or run out altogether */
666 if (atomic_read(&dentry->d_count) != 0) {
668 "BUG: Dentry %p{i=%lx,n=%s}"
670 " [unmount of %s %s]\n",
673 dentry->d_inode->i_ino : 0UL,
675 atomic_read(&dentry->d_count),
676 dentry->d_sb->s_type->name,
684 parent = dentry->d_parent;
685 atomic_dec(&parent->d_count);
688 list_del(&dentry->d_u.d_child);
691 inode = dentry->d_inode;
693 dentry->d_inode = NULL;
694 list_del_init(&dentry->d_alias);
695 if (dentry->d_op && dentry->d_op->d_iput)
696 dentry->d_op->d_iput(dentry, inode);
703 /* finished when we fall off the top of the tree,
704 * otherwise we ascend to the parent and move to the
705 * next sibling if there is one */
711 } while (list_empty(&dentry->d_subdirs));
713 dentry = list_entry(dentry->d_subdirs.next,
714 struct dentry, d_u.d_child);
717 /* several dentries were freed, need to correct nr_dentry */
718 spin_lock(&dcache_lock);
719 dentry_stat.nr_dentry -= detached;
720 spin_unlock(&dcache_lock);
724 * destroy the dentries attached to a superblock on unmounting
725 * - we don't need to use dentry->d_lock, and only need dcache_lock when
726 * removing the dentry from the system lists and hashes because:
727 * - the superblock is detached from all mountings and open files, so the
728 * dentry trees will not be rearranged by the VFS
729 * - s_umount is write-locked, so the memory pressure shrinker will ignore
730 * any dentries belonging to this superblock that it comes across
731 * - the filesystem itself is no longer permitted to rearrange the dentries
734 void shrink_dcache_for_umount(struct super_block *sb)
736 struct dentry *dentry;
738 if (down_read_trylock(&sb->s_umount))
743 atomic_dec(&dentry->d_count);
744 shrink_dcache_for_umount_subtree(dentry);
746 while (!hlist_empty(&sb->s_anon)) {
747 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
748 shrink_dcache_for_umount_subtree(dentry);
753 * Search for at least 1 mount point in the dentry's subdirs.
754 * We descend to the next level whenever the d_subdirs
755 * list is non-empty and continue searching.
759 * have_submounts - check for mounts over a dentry
760 * @parent: dentry to check.
762 * Return true if the parent or its subdirectories contain
766 int have_submounts(struct dentry *parent)
768 struct dentry *this_parent = parent;
769 struct list_head *next;
771 spin_lock(&dcache_lock);
772 if (d_mountpoint(parent))
775 next = this_parent->d_subdirs.next;
777 while (next != &this_parent->d_subdirs) {
778 struct list_head *tmp = next;
779 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
781 /* Have we found a mount point ? */
782 if (d_mountpoint(dentry))
784 if (!list_empty(&dentry->d_subdirs)) {
785 this_parent = dentry;
790 * All done at this level ... ascend and resume the search.
792 if (this_parent != parent) {
793 next = this_parent->d_u.d_child.next;
794 this_parent = this_parent->d_parent;
797 spin_unlock(&dcache_lock);
798 return 0; /* No mount points found in tree */
800 spin_unlock(&dcache_lock);
803 EXPORT_SYMBOL(have_submounts);
806 * Search the dentry child list for the specified parent,
807 * and move any unused dentries to the end of the unused
808 * list for prune_dcache(). We descend to the next level
809 * whenever the d_subdirs list is non-empty and continue
812 * It returns zero iff there are no unused children,
813 * otherwise it returns the number of children moved to
814 * the end of the unused list. This may not be the total
815 * number of unused children, because select_parent can
816 * drop the lock and return early due to latency
819 static int select_parent(struct dentry * parent)
821 struct dentry *this_parent = parent;
822 struct list_head *next;
825 spin_lock(&dcache_lock);
827 next = this_parent->d_subdirs.next;
829 while (next != &this_parent->d_subdirs) {
830 struct list_head *tmp = next;
831 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
834 dentry_lru_del_init(dentry);
836 * move only zero ref count dentries to the end
837 * of the unused list for prune_dcache
839 if (!atomic_read(&dentry->d_count)) {
840 dentry_lru_add_tail(dentry);
845 * We can return to the caller if we have found some (this
846 * ensures forward progress). We'll be coming back to find
849 if (found && need_resched())
853 * Descend a level if the d_subdirs list is non-empty.
855 if (!list_empty(&dentry->d_subdirs)) {
856 this_parent = dentry;
861 * All done at this level ... ascend and resume the search.
863 if (this_parent != parent) {
864 next = this_parent->d_u.d_child.next;
865 this_parent = this_parent->d_parent;
869 spin_unlock(&dcache_lock);
874 * shrink_dcache_parent - prune dcache
875 * @parent: parent of entries to prune
877 * Prune the dcache to remove unused children of the parent dentry.
880 void shrink_dcache_parent(struct dentry * parent)
882 struct super_block *sb = parent->d_sb;
885 while ((found = select_parent(parent)) != 0)
886 __shrink_dcache_sb(sb, &found, 0);
888 EXPORT_SYMBOL(shrink_dcache_parent);
891 * Scan `nr' dentries and return the number which remain.
893 * We need to avoid reentering the filesystem if the caller is performing a
894 * GFP_NOFS allocation attempt. One example deadlock is:
896 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
897 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
898 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
900 * In this case we return -1 to tell the caller that we baled.
902 static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
905 if (!(gfp_mask & __GFP_FS))
909 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
912 static struct shrinker dcache_shrinker = {
913 .shrink = shrink_dcache_memory,
914 .seeks = DEFAULT_SEEKS,
918 * d_alloc - allocate a dcache entry
919 * @parent: parent of entry to allocate
920 * @name: qstr of the name
922 * Allocates a dentry. It returns %NULL if there is insufficient memory
923 * available. On a success the dentry is returned. The name passed in is
924 * copied and the copy passed in may be reused after this call.
927 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
929 struct dentry *dentry;
932 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
936 if (name->len > DNAME_INLINE_LEN-1) {
937 dname = kmalloc(name->len + 1, GFP_KERNEL);
939 kmem_cache_free(dentry_cache, dentry);
943 dname = dentry->d_iname;
945 dentry->d_name.name = dname;
947 dentry->d_name.len = name->len;
948 dentry->d_name.hash = name->hash;
949 memcpy(dname, name->name, name->len);
950 dname[name->len] = 0;
952 atomic_set(&dentry->d_count, 1);
953 dentry->d_flags = DCACHE_UNHASHED;
954 spin_lock_init(&dentry->d_lock);
955 dentry->d_inode = NULL;
956 dentry->d_parent = NULL;
959 dentry->d_fsdata = NULL;
960 dentry->d_mounted = 0;
961 INIT_HLIST_NODE(&dentry->d_hash);
962 INIT_LIST_HEAD(&dentry->d_lru);
963 INIT_LIST_HEAD(&dentry->d_subdirs);
964 INIT_LIST_HEAD(&dentry->d_alias);
967 dentry->d_parent = dget(parent);
968 dentry->d_sb = parent->d_sb;
970 INIT_LIST_HEAD(&dentry->d_u.d_child);
973 spin_lock(&dcache_lock);
975 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
976 dentry_stat.nr_dentry++;
977 spin_unlock(&dcache_lock);
981 EXPORT_SYMBOL(d_alloc);
983 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
988 q.len = strlen(name);
989 q.hash = full_name_hash(q.name, q.len);
990 return d_alloc(parent, &q);
992 EXPORT_SYMBOL(d_alloc_name);
994 /* the caller must hold dcache_lock */
995 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
998 list_add(&dentry->d_alias, &inode->i_dentry);
999 dentry->d_inode = inode;
1000 fsnotify_d_instantiate(dentry, inode);
1004 * d_instantiate - fill in inode information for a dentry
1005 * @entry: dentry to complete
1006 * @inode: inode to attach to this dentry
1008 * Fill in inode information in the entry.
1010 * This turns negative dentries into productive full members
1013 * NOTE! This assumes that the inode count has been incremented
1014 * (or otherwise set) by the caller to indicate that it is now
1015 * in use by the dcache.
1018 void d_instantiate(struct dentry *entry, struct inode * inode)
1020 BUG_ON(!list_empty(&entry->d_alias));
1021 spin_lock(&dcache_lock);
1022 __d_instantiate(entry, inode);
1023 spin_unlock(&dcache_lock);
1024 security_d_instantiate(entry, inode);
1026 EXPORT_SYMBOL(d_instantiate);
1029 * d_instantiate_unique - instantiate a non-aliased dentry
1030 * @entry: dentry to instantiate
1031 * @inode: inode to attach to this dentry
1033 * Fill in inode information in the entry. On success, it returns NULL.
1034 * If an unhashed alias of "entry" already exists, then we return the
1035 * aliased dentry instead and drop one reference to inode.
1037 * Note that in order to avoid conflicts with rename() etc, the caller
1038 * had better be holding the parent directory semaphore.
1040 * This also assumes that the inode count has been incremented
1041 * (or otherwise set) by the caller to indicate that it is now
1042 * in use by the dcache.
1044 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1045 struct inode *inode)
1047 struct dentry *alias;
1048 int len = entry->d_name.len;
1049 const char *name = entry->d_name.name;
1050 unsigned int hash = entry->d_name.hash;
1053 __d_instantiate(entry, NULL);
1057 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1058 struct qstr *qstr = &alias->d_name;
1060 if (qstr->hash != hash)
1062 if (alias->d_parent != entry->d_parent)
1064 if (qstr->len != len)
1066 if (memcmp(qstr->name, name, len))
1072 __d_instantiate(entry, inode);
1076 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1078 struct dentry *result;
1080 BUG_ON(!list_empty(&entry->d_alias));
1082 spin_lock(&dcache_lock);
1083 result = __d_instantiate_unique(entry, inode);
1084 spin_unlock(&dcache_lock);
1087 security_d_instantiate(entry, inode);
1091 BUG_ON(!d_unhashed(result));
1096 EXPORT_SYMBOL(d_instantiate_unique);
1099 * d_alloc_root - allocate root dentry
1100 * @root_inode: inode to allocate the root for
1102 * Allocate a root ("/") dentry for the inode given. The inode is
1103 * instantiated and returned. %NULL is returned if there is insufficient
1104 * memory or the inode passed is %NULL.
1107 struct dentry * d_alloc_root(struct inode * root_inode)
1109 struct dentry *res = NULL;
1112 static const struct qstr name = { .name = "/", .len = 1 };
1114 res = d_alloc(NULL, &name);
1116 res->d_sb = root_inode->i_sb;
1117 res->d_parent = res;
1118 d_instantiate(res, root_inode);
1123 EXPORT_SYMBOL(d_alloc_root);
1125 static inline struct hlist_head *d_hash(struct dentry *parent,
1128 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1129 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1130 return dentry_hashtable + (hash & D_HASHMASK);
1134 * d_obtain_alias - find or allocate a dentry for a given inode
1135 * @inode: inode to allocate the dentry for
1137 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1138 * similar open by handle operations. The returned dentry may be anonymous,
1139 * or may have a full name (if the inode was already in the cache).
1141 * When called on a directory inode, we must ensure that the inode only ever
1142 * has one dentry. If a dentry is found, that is returned instead of
1143 * allocating a new one.
1145 * On successful return, the reference to the inode has been transferred
1146 * to the dentry. In case of an error the reference on the inode is released.
1147 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1148 * be passed in and will be the error will be propagate to the return value,
1149 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1151 struct dentry *d_obtain_alias(struct inode *inode)
1153 static const struct qstr anonstring = { .name = "" };
1158 return ERR_PTR(-ESTALE);
1160 return ERR_CAST(inode);
1162 res = d_find_alias(inode);
1166 tmp = d_alloc(NULL, &anonstring);
1168 res = ERR_PTR(-ENOMEM);
1171 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1173 spin_lock(&dcache_lock);
1174 res = __d_find_alias(inode, 0);
1176 spin_unlock(&dcache_lock);
1181 /* attach a disconnected dentry */
1182 spin_lock(&tmp->d_lock);
1183 tmp->d_sb = inode->i_sb;
1184 tmp->d_inode = inode;
1185 tmp->d_flags |= DCACHE_DISCONNECTED;
1186 tmp->d_flags &= ~DCACHE_UNHASHED;
1187 list_add(&tmp->d_alias, &inode->i_dentry);
1188 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1189 spin_unlock(&tmp->d_lock);
1191 spin_unlock(&dcache_lock);
1198 EXPORT_SYMBOL(d_obtain_alias);
1201 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1202 * @inode: the inode which may have a disconnected dentry
1203 * @dentry: a negative dentry which we want to point to the inode.
1205 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1206 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1207 * and return it, else simply d_add the inode to the dentry and return NULL.
1209 * This is needed in the lookup routine of any filesystem that is exportable
1210 * (via knfsd) so that we can build dcache paths to directories effectively.
1212 * If a dentry was found and moved, then it is returned. Otherwise NULL
1213 * is returned. This matches the expected return value of ->lookup.
1216 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1218 struct dentry *new = NULL;
1220 if (inode && S_ISDIR(inode->i_mode)) {
1221 spin_lock(&dcache_lock);
1222 new = __d_find_alias(inode, 1);
1224 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1225 spin_unlock(&dcache_lock);
1226 security_d_instantiate(new, inode);
1227 d_move(new, dentry);
1230 /* already taking dcache_lock, so d_add() by hand */
1231 __d_instantiate(dentry, inode);
1232 spin_unlock(&dcache_lock);
1233 security_d_instantiate(dentry, inode);
1237 d_add(dentry, inode);
1240 EXPORT_SYMBOL(d_splice_alias);
1243 * d_add_ci - lookup or allocate new dentry with case-exact name
1244 * @inode: the inode case-insensitive lookup has found
1245 * @dentry: the negative dentry that was passed to the parent's lookup func
1246 * @name: the case-exact name to be associated with the returned dentry
1248 * This is to avoid filling the dcache with case-insensitive names to the
1249 * same inode, only the actual correct case is stored in the dcache for
1250 * case-insensitive filesystems.
1252 * For a case-insensitive lookup match and if the the case-exact dentry
1253 * already exists in in the dcache, use it and return it.
1255 * If no entry exists with the exact case name, allocate new dentry with
1256 * the exact case, and return the spliced entry.
1258 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1262 struct dentry *found;
1266 * First check if a dentry matching the name already exists,
1267 * if not go ahead and create it now.
1269 found = d_hash_and_lookup(dentry->d_parent, name);
1271 new = d_alloc(dentry->d_parent, name);
1277 found = d_splice_alias(inode, new);
1286 * If a matching dentry exists, and it's not negative use it.
1288 * Decrement the reference count to balance the iget() done
1291 if (found->d_inode) {
1292 if (unlikely(found->d_inode != inode)) {
1293 /* This can't happen because bad inodes are unhashed. */
1294 BUG_ON(!is_bad_inode(inode));
1295 BUG_ON(!is_bad_inode(found->d_inode));
1302 * Negative dentry: instantiate it unless the inode is a directory and
1303 * already has a dentry.
1305 spin_lock(&dcache_lock);
1306 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1307 __d_instantiate(found, inode);
1308 spin_unlock(&dcache_lock);
1309 security_d_instantiate(found, inode);
1314 * In case a directory already has a (disconnected) entry grab a
1315 * reference to it, move it in place and use it.
1317 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1319 spin_unlock(&dcache_lock);
1320 security_d_instantiate(found, inode);
1328 return ERR_PTR(error);
1330 EXPORT_SYMBOL(d_add_ci);
1333 * d_lookup - search for a dentry
1334 * @parent: parent dentry
1335 * @name: qstr of name we wish to find
1337 * Searches the children of the parent dentry for the name in question. If
1338 * the dentry is found its reference count is incremented and the dentry
1339 * is returned. The caller must use dput to free the entry when it has
1340 * finished using it. %NULL is returned on failure.
1342 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1343 * Memory barriers are used while updating and doing lockless traversal.
1344 * To avoid races with d_move while rename is happening, d_lock is used.
1346 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1347 * and name pointer in one structure pointed by d_qstr.
1349 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1350 * lookup is going on.
1352 * The dentry unused LRU is not updated even if lookup finds the required dentry
1353 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1354 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1357 * d_lookup() is protected against the concurrent renames in some unrelated
1358 * directory using the seqlockt_t rename_lock.
1361 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1363 struct dentry * dentry = NULL;
1367 seq = read_seqbegin(&rename_lock);
1368 dentry = __d_lookup(parent, name);
1371 } while (read_seqretry(&rename_lock, seq));
1374 EXPORT_SYMBOL(d_lookup);
1376 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1378 unsigned int len = name->len;
1379 unsigned int hash = name->hash;
1380 const unsigned char *str = name->name;
1381 struct hlist_head *head = d_hash(parent,hash);
1382 struct dentry *found = NULL;
1383 struct hlist_node *node;
1384 struct dentry *dentry;
1388 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1391 if (dentry->d_name.hash != hash)
1393 if (dentry->d_parent != parent)
1396 spin_lock(&dentry->d_lock);
1399 * Recheck the dentry after taking the lock - d_move may have
1400 * changed things. Don't bother checking the hash because we're
1401 * about to compare the whole name anyway.
1403 if (dentry->d_parent != parent)
1406 /* non-existing due to RCU? */
1407 if (d_unhashed(dentry))
1411 * It is safe to compare names since d_move() cannot
1412 * change the qstr (protected by d_lock).
1414 qstr = &dentry->d_name;
1415 if (parent->d_op && parent->d_op->d_compare) {
1416 if (parent->d_op->d_compare(parent, qstr, name))
1419 if (qstr->len != len)
1421 if (memcmp(qstr->name, str, len))
1425 atomic_inc(&dentry->d_count);
1427 spin_unlock(&dentry->d_lock);
1430 spin_unlock(&dentry->d_lock);
1438 * d_hash_and_lookup - hash the qstr then search for a dentry
1439 * @dir: Directory to search in
1440 * @name: qstr of name we wish to find
1442 * On hash failure or on lookup failure NULL is returned.
1444 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1446 struct dentry *dentry = NULL;
1449 * Check for a fs-specific hash function. Note that we must
1450 * calculate the standard hash first, as the d_op->d_hash()
1451 * routine may choose to leave the hash value unchanged.
1453 name->hash = full_name_hash(name->name, name->len);
1454 if (dir->d_op && dir->d_op->d_hash) {
1455 if (dir->d_op->d_hash(dir, name) < 0)
1458 dentry = d_lookup(dir, name);
1464 * d_validate - verify dentry provided from insecure source
1465 * @dentry: The dentry alleged to be valid child of @dparent
1466 * @dparent: The parent dentry (known to be valid)
1468 * An insecure source has sent us a dentry, here we verify it and dget() it.
1469 * This is used by ncpfs in its readdir implementation.
1470 * Zero is returned in the dentry is invalid.
1473 int d_validate(struct dentry *dentry, struct dentry *dparent)
1475 struct hlist_head *base;
1476 struct hlist_node *lhp;
1478 /* Check whether the ptr might be valid at all.. */
1479 if (!kmem_ptr_validate(dentry_cache, dentry))
1482 if (dentry->d_parent != dparent)
1485 spin_lock(&dcache_lock);
1486 base = d_hash(dparent, dentry->d_name.hash);
1487 hlist_for_each(lhp,base) {
1488 /* hlist_for_each_entry_rcu() not required for d_hash list
1489 * as it is parsed under dcache_lock
1491 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1492 __dget_locked(dentry);
1493 spin_unlock(&dcache_lock);
1497 spin_unlock(&dcache_lock);
1501 EXPORT_SYMBOL(d_validate);
1504 * When a file is deleted, we have two options:
1505 * - turn this dentry into a negative dentry
1506 * - unhash this dentry and free it.
1508 * Usually, we want to just turn this into
1509 * a negative dentry, but if anybody else is
1510 * currently using the dentry or the inode
1511 * we can't do that and we fall back on removing
1512 * it from the hash queues and waiting for
1513 * it to be deleted later when it has no users
1517 * d_delete - delete a dentry
1518 * @dentry: The dentry to delete
1520 * Turn the dentry into a negative dentry if possible, otherwise
1521 * remove it from the hash queues so it can be deleted later
1524 void d_delete(struct dentry * dentry)
1528 * Are we the only user?
1530 spin_lock(&dcache_lock);
1531 spin_lock(&dentry->d_lock);
1532 isdir = S_ISDIR(dentry->d_inode->i_mode);
1533 if (atomic_read(&dentry->d_count) == 1) {
1534 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1535 dentry_iput(dentry);
1536 fsnotify_nameremove(dentry, isdir);
1540 if (!d_unhashed(dentry))
1543 spin_unlock(&dentry->d_lock);
1544 spin_unlock(&dcache_lock);
1546 fsnotify_nameremove(dentry, isdir);
1548 EXPORT_SYMBOL(d_delete);
1550 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1553 entry->d_flags &= ~DCACHE_UNHASHED;
1554 hlist_add_head_rcu(&entry->d_hash, list);
1557 static void _d_rehash(struct dentry * entry)
1559 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1563 * d_rehash - add an entry back to the hash
1564 * @entry: dentry to add to the hash
1566 * Adds a dentry to the hash according to its name.
1569 void d_rehash(struct dentry * entry)
1571 spin_lock(&dcache_lock);
1572 spin_lock(&entry->d_lock);
1574 spin_unlock(&entry->d_lock);
1575 spin_unlock(&dcache_lock);
1577 EXPORT_SYMBOL(d_rehash);
1580 * When switching names, the actual string doesn't strictly have to
1581 * be preserved in the target - because we're dropping the target
1582 * anyway. As such, we can just do a simple memcpy() to copy over
1583 * the new name before we switch.
1585 * Note that we have to be a lot more careful about getting the hash
1586 * switched - we have to switch the hash value properly even if it
1587 * then no longer matches the actual (corrupted) string of the target.
1588 * The hash value has to match the hash queue that the dentry is on..
1590 static void switch_names(struct dentry *dentry, struct dentry *target)
1592 if (dname_external(target)) {
1593 if (dname_external(dentry)) {
1595 * Both external: swap the pointers
1597 swap(target->d_name.name, dentry->d_name.name);
1600 * dentry:internal, target:external. Steal target's
1601 * storage and make target internal.
1603 memcpy(target->d_iname, dentry->d_name.name,
1604 dentry->d_name.len + 1);
1605 dentry->d_name.name = target->d_name.name;
1606 target->d_name.name = target->d_iname;
1609 if (dname_external(dentry)) {
1611 * dentry:external, target:internal. Give dentry's
1612 * storage to target and make dentry internal
1614 memcpy(dentry->d_iname, target->d_name.name,
1615 target->d_name.len + 1);
1616 target->d_name.name = dentry->d_name.name;
1617 dentry->d_name.name = dentry->d_iname;
1620 * Both are internal. Just copy target to dentry
1622 memcpy(dentry->d_iname, target->d_name.name,
1623 target->d_name.len + 1);
1624 dentry->d_name.len = target->d_name.len;
1628 swap(dentry->d_name.len, target->d_name.len);
1632 * We cannibalize "target" when moving dentry on top of it,
1633 * because it's going to be thrown away anyway. We could be more
1634 * polite about it, though.
1636 * This forceful removal will result in ugly /proc output if
1637 * somebody holds a file open that got deleted due to a rename.
1638 * We could be nicer about the deleted file, and let it show
1639 * up under the name it had before it was deleted rather than
1640 * under the original name of the file that was moved on top of it.
1644 * d_move_locked - move a dentry
1645 * @dentry: entry to move
1646 * @target: new dentry
1648 * Update the dcache to reflect the move of a file name. Negative
1649 * dcache entries should not be moved in this way.
1651 static void d_move_locked(struct dentry * dentry, struct dentry * target)
1653 struct hlist_head *list;
1655 if (!dentry->d_inode)
1656 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1658 write_seqlock(&rename_lock);
1660 * XXXX: do we really need to take target->d_lock?
1662 if (target < dentry) {
1663 spin_lock(&target->d_lock);
1664 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1666 spin_lock(&dentry->d_lock);
1667 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
1670 /* Move the dentry to the target hash queue, if on different bucket */
1671 if (d_unhashed(dentry))
1672 goto already_unhashed;
1674 hlist_del_rcu(&dentry->d_hash);
1677 list = d_hash(target->d_parent, target->d_name.hash);
1678 __d_rehash(dentry, list);
1680 /* Unhash the target: dput() will then get rid of it */
1683 list_del(&dentry->d_u.d_child);
1684 list_del(&target->d_u.d_child);
1686 /* Switch the names.. */
1687 switch_names(dentry, target);
1688 swap(dentry->d_name.hash, target->d_name.hash);
1690 /* ... and switch the parents */
1691 if (IS_ROOT(dentry)) {
1692 dentry->d_parent = target->d_parent;
1693 target->d_parent = target;
1694 INIT_LIST_HEAD(&target->d_u.d_child);
1696 swap(dentry->d_parent, target->d_parent);
1698 /* And add them back to the (new) parent lists */
1699 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1702 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1703 spin_unlock(&target->d_lock);
1704 fsnotify_d_move(dentry);
1705 spin_unlock(&dentry->d_lock);
1706 write_sequnlock(&rename_lock);
1710 * d_move - move a dentry
1711 * @dentry: entry to move
1712 * @target: new dentry
1714 * Update the dcache to reflect the move of a file name. Negative
1715 * dcache entries should not be moved in this way.
1718 void d_move(struct dentry * dentry, struct dentry * target)
1720 spin_lock(&dcache_lock);
1721 d_move_locked(dentry, target);
1722 spin_unlock(&dcache_lock);
1724 EXPORT_SYMBOL(d_move);
1727 * d_ancestor - search for an ancestor
1728 * @p1: ancestor dentry
1731 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1732 * an ancestor of p2, else NULL.
1734 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
1738 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
1739 if (p->d_parent == p1)
1746 * This helper attempts to cope with remotely renamed directories
1748 * It assumes that the caller is already holding
1749 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1751 * Note: If ever the locking in lock_rename() changes, then please
1752 * remember to update this too...
1754 static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
1755 __releases(dcache_lock)
1757 struct mutex *m1 = NULL, *m2 = NULL;
1760 /* If alias and dentry share a parent, then no extra locks required */
1761 if (alias->d_parent == dentry->d_parent)
1764 /* Check for loops */
1765 ret = ERR_PTR(-ELOOP);
1766 if (d_ancestor(alias, dentry))
1769 /* See lock_rename() */
1770 ret = ERR_PTR(-EBUSY);
1771 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1773 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1774 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1776 m2 = &alias->d_parent->d_inode->i_mutex;
1778 d_move_locked(alias, dentry);
1781 spin_unlock(&dcache_lock);
1790 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1791 * named dentry in place of the dentry to be replaced.
1793 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1795 struct dentry *dparent, *aparent;
1797 switch_names(dentry, anon);
1798 swap(dentry->d_name.hash, anon->d_name.hash);
1800 dparent = dentry->d_parent;
1801 aparent = anon->d_parent;
1803 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1804 list_del(&dentry->d_u.d_child);
1805 if (!IS_ROOT(dentry))
1806 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1808 INIT_LIST_HEAD(&dentry->d_u.d_child);
1810 anon->d_parent = (dparent == dentry) ? anon : dparent;
1811 list_del(&anon->d_u.d_child);
1813 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1815 INIT_LIST_HEAD(&anon->d_u.d_child);
1817 anon->d_flags &= ~DCACHE_DISCONNECTED;
1821 * d_materialise_unique - introduce an inode into the tree
1822 * @dentry: candidate dentry
1823 * @inode: inode to bind to the dentry, to which aliases may be attached
1825 * Introduces an dentry into the tree, substituting an extant disconnected
1826 * root directory alias in its place if there is one
1828 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1830 struct dentry *actual;
1832 BUG_ON(!d_unhashed(dentry));
1834 spin_lock(&dcache_lock);
1838 __d_instantiate(dentry, NULL);
1842 if (S_ISDIR(inode->i_mode)) {
1843 struct dentry *alias;
1845 /* Does an aliased dentry already exist? */
1846 alias = __d_find_alias(inode, 0);
1849 /* Is this an anonymous mountpoint that we could splice
1851 if (IS_ROOT(alias)) {
1852 spin_lock(&alias->d_lock);
1853 __d_materialise_dentry(dentry, alias);
1857 /* Nope, but we must(!) avoid directory aliasing */
1858 actual = __d_unalias(dentry, alias);
1865 /* Add a unique reference */
1866 actual = __d_instantiate_unique(dentry, inode);
1869 else if (unlikely(!d_unhashed(actual)))
1870 goto shouldnt_be_hashed;
1873 spin_lock(&actual->d_lock);
1876 spin_unlock(&actual->d_lock);
1877 spin_unlock(&dcache_lock);
1879 if (actual == dentry) {
1880 security_d_instantiate(dentry, inode);
1888 spin_unlock(&dcache_lock);
1891 EXPORT_SYMBOL_GPL(d_materialise_unique);
1893 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
1897 return -ENAMETOOLONG;
1899 memcpy(*buffer, str, namelen);
1903 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1905 return prepend(buffer, buflen, name->name, name->len);
1909 * __d_path - return the path of a dentry
1910 * @path: the dentry/vfsmount to report
1911 * @root: root vfsmnt/dentry (may be modified by this function)
1912 * @buffer: buffer to return value in
1913 * @buflen: buffer length
1915 * Convert a dentry into an ASCII path name. If the entry has been deleted
1916 * the string " (deleted)" is appended. Note that this is ambiguous.
1918 * Returns a pointer into the buffer or an error code if the
1919 * path was too long.
1921 * "buflen" should be positive. Caller holds the dcache_lock.
1923 * If path is not reachable from the supplied root, then the value of
1924 * root is changed (without modifying refcounts).
1926 char *__d_path(const struct path *path, struct path *root,
1927 char *buffer, int buflen)
1929 struct dentry *dentry = path->dentry;
1930 struct vfsmount *vfsmnt = path->mnt;
1931 char *end = buffer + buflen;
1934 spin_lock(&vfsmount_lock);
1935 prepend(&end, &buflen, "\0", 1);
1936 if (d_unlinked(dentry) &&
1937 (prepend(&end, &buflen, " (deleted)", 10) != 0))
1947 struct dentry * parent;
1949 if (dentry == root->dentry && vfsmnt == root->mnt)
1951 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1953 if (vfsmnt->mnt_parent == vfsmnt) {
1956 dentry = vfsmnt->mnt_mountpoint;
1957 vfsmnt = vfsmnt->mnt_parent;
1960 parent = dentry->d_parent;
1962 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
1963 (prepend(&end, &buflen, "/", 1) != 0))
1970 spin_unlock(&vfsmount_lock);
1974 retval += 1; /* hit the slash */
1975 if (prepend_name(&retval, &buflen, &dentry->d_name) != 0)
1978 root->dentry = dentry;
1982 retval = ERR_PTR(-ENAMETOOLONG);
1987 * d_path - return the path of a dentry
1988 * @path: path to report
1989 * @buf: buffer to return value in
1990 * @buflen: buffer length
1992 * Convert a dentry into an ASCII path name. If the entry has been deleted
1993 * the string " (deleted)" is appended. Note that this is ambiguous.
1995 * Returns a pointer into the buffer or an error code if the path was
1996 * too long. Note: Callers should use the returned pointer, not the passed
1997 * in buffer, to use the name! The implementation often starts at an offset
1998 * into the buffer, and may leave 0 bytes at the start.
2000 * "buflen" should be positive.
2002 char *d_path(const struct path *path, char *buf, int buflen)
2009 * We have various synthetic filesystems that never get mounted. On
2010 * these filesystems dentries are never used for lookup purposes, and
2011 * thus don't need to be hashed. They also don't need a name until a
2012 * user wants to identify the object in /proc/pid/fd/. The little hack
2013 * below allows us to generate a name for these objects on demand:
2015 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2016 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2018 read_lock(¤t->fs->lock);
2019 root = current->fs->root;
2021 read_unlock(¤t->fs->lock);
2022 spin_lock(&dcache_lock);
2024 res = __d_path(path, &tmp, buf, buflen);
2025 spin_unlock(&dcache_lock);
2029 EXPORT_SYMBOL(d_path);
2032 * Helper function for dentry_operations.d_dname() members
2034 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2035 const char *fmt, ...)
2041 va_start(args, fmt);
2042 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2045 if (sz > sizeof(temp) || sz > buflen)
2046 return ERR_PTR(-ENAMETOOLONG);
2048 buffer += buflen - sz;
2049 return memcpy(buffer, temp, sz);
2053 * Write full pathname from the root of the filesystem into the buffer.
2055 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2057 char *end = buf + buflen;
2060 spin_lock(&dcache_lock);
2061 prepend(&end, &buflen, "\0", 1);
2062 if (d_unlinked(dentry) &&
2063 (prepend(&end, &buflen, "//deleted", 9) != 0))
2071 while (!IS_ROOT(dentry)) {
2072 struct dentry *parent = dentry->d_parent;
2075 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
2076 (prepend(&end, &buflen, "/", 1) != 0))
2082 spin_unlock(&dcache_lock);
2085 spin_unlock(&dcache_lock);
2086 return ERR_PTR(-ENAMETOOLONG);
2090 * NOTE! The user-level library version returns a
2091 * character pointer. The kernel system call just
2092 * returns the length of the buffer filled (which
2093 * includes the ending '\0' character), or a negative
2094 * error value. So libc would do something like
2096 * char *getcwd(char * buf, size_t size)
2100 * retval = sys_getcwd(buf, size);
2107 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2110 struct path pwd, root;
2111 char *page = (char *) __get_free_page(GFP_USER);
2116 read_lock(¤t->fs->lock);
2117 pwd = current->fs->pwd;
2119 root = current->fs->root;
2121 read_unlock(¤t->fs->lock);
2124 spin_lock(&dcache_lock);
2125 if (!d_unlinked(pwd.dentry)) {
2127 struct path tmp = root;
2130 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
2131 spin_unlock(&dcache_lock);
2133 error = PTR_ERR(cwd);
2138 len = PAGE_SIZE + page - cwd;
2141 if (copy_to_user(buf, cwd, len))
2145 spin_unlock(&dcache_lock);
2150 free_page((unsigned long) page);
2155 * Test whether new_dentry is a subdirectory of old_dentry.
2157 * Trivially implemented using the dcache structure
2161 * is_subdir - is new dentry a subdirectory of old_dentry
2162 * @new_dentry: new dentry
2163 * @old_dentry: old dentry
2165 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2166 * Returns 0 otherwise.
2167 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2170 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2175 if (new_dentry == old_dentry)
2179 * Need rcu_readlock to protect against the d_parent trashing
2184 /* for restarting inner loop in case of seq retry */
2185 seq = read_seqbegin(&rename_lock);
2186 if (d_ancestor(old_dentry, new_dentry))
2190 } while (read_seqretry(&rename_lock, seq));
2196 int path_is_under(struct path *path1, struct path *path2)
2198 struct vfsmount *mnt = path1->mnt;
2199 struct dentry *dentry = path1->dentry;
2201 spin_lock(&vfsmount_lock);
2202 if (mnt != path2->mnt) {
2204 if (mnt->mnt_parent == mnt) {
2205 spin_unlock(&vfsmount_lock);
2208 if (mnt->mnt_parent == path2->mnt)
2210 mnt = mnt->mnt_parent;
2212 dentry = mnt->mnt_mountpoint;
2214 res = is_subdir(dentry, path2->dentry);
2215 spin_unlock(&vfsmount_lock);
2218 EXPORT_SYMBOL(path_is_under);
2220 void d_genocide(struct dentry *root)
2222 struct dentry *this_parent = root;
2223 struct list_head *next;
2225 spin_lock(&dcache_lock);
2227 next = this_parent->d_subdirs.next;
2229 while (next != &this_parent->d_subdirs) {
2230 struct list_head *tmp = next;
2231 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2233 if (d_unhashed(dentry)||!dentry->d_inode)
2235 if (!list_empty(&dentry->d_subdirs)) {
2236 this_parent = dentry;
2239 atomic_dec(&dentry->d_count);
2241 if (this_parent != root) {
2242 next = this_parent->d_u.d_child.next;
2243 atomic_dec(&this_parent->d_count);
2244 this_parent = this_parent->d_parent;
2247 spin_unlock(&dcache_lock);
2251 * find_inode_number - check for dentry with name
2252 * @dir: directory to check
2253 * @name: Name to find.
2255 * Check whether a dentry already exists for the given name,
2256 * and return the inode number if it has an inode. Otherwise
2259 * This routine is used to post-process directory listings for
2260 * filesystems using synthetic inode numbers, and is necessary
2261 * to keep getcwd() working.
2264 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2266 struct dentry * dentry;
2269 dentry = d_hash_and_lookup(dir, name);
2271 if (dentry->d_inode)
2272 ino = dentry->d_inode->i_ino;
2277 EXPORT_SYMBOL(find_inode_number);
2279 static __initdata unsigned long dhash_entries;
2280 static int __init set_dhash_entries(char *str)
2284 dhash_entries = simple_strtoul(str, &str, 0);
2287 __setup("dhash_entries=", set_dhash_entries);
2289 static void __init dcache_init_early(void)
2293 /* If hashes are distributed across NUMA nodes, defer
2294 * hash allocation until vmalloc space is available.
2300 alloc_large_system_hash("Dentry cache",
2301 sizeof(struct hlist_head),
2309 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2310 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2313 static void __init dcache_init(void)
2318 * A constructor could be added for stable state like the lists,
2319 * but it is probably not worth it because of the cache nature
2322 dentry_cache = KMEM_CACHE(dentry,
2323 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2325 register_shrinker(&dcache_shrinker);
2327 /* Hash may have been set up in dcache_init_early */
2332 alloc_large_system_hash("Dentry cache",
2333 sizeof(struct hlist_head),
2341 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2342 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2345 /* SLAB cache for __getname() consumers */
2346 struct kmem_cache *names_cachep __read_mostly;
2347 EXPORT_SYMBOL(names_cachep);
2349 EXPORT_SYMBOL(d_genocide);
2351 void __init vfs_caches_init_early(void)
2353 dcache_init_early();
2357 void __init vfs_caches_init(unsigned long mempages)
2359 unsigned long reserve;
2361 /* Base hash sizes on available memory, with a reserve equal to
2362 150% of current kernel size */
2364 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2365 mempages -= reserve;
2367 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
2368 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2372 files_init(mempages);