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>
20 #include <linux/fdtable.h>
22 #include <linux/fsnotify.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/hash.h>
26 #include <linux/cache.h>
27 #include <linux/module.h>
28 #include <linux/mount.h>
29 #include <linux/file.h>
30 #include <asm/uaccess.h>
31 #include <linux/security.h>
32 #include <linux/seqlock.h>
33 #include <linux/swap.h>
34 #include <linux/bootmem.h>
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 if (dname_external(dentry))
73 kfree(dentry->d_name.name);
74 kmem_cache_free(dentry_cache, dentry);
77 static void d_callback(struct rcu_head *head)
79 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
84 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
87 static void d_free(struct dentry *dentry)
89 if (dentry->d_op && dentry->d_op->d_release)
90 dentry->d_op->d_release(dentry);
91 /* if dentry was never inserted into hash, immediate free is OK */
92 if (hlist_unhashed(&dentry->d_hash))
95 call_rcu(&dentry->d_u.d_rcu, d_callback);
99 * Release the dentry's inode, using the filesystem
100 * d_iput() operation if defined.
102 static void dentry_iput(struct dentry * dentry)
103 __releases(dentry->d_lock)
104 __releases(dcache_lock)
106 struct inode *inode = dentry->d_inode;
108 dentry->d_inode = NULL;
109 list_del_init(&dentry->d_alias);
110 spin_unlock(&dentry->d_lock);
111 spin_unlock(&dcache_lock);
113 fsnotify_inoderemove(inode);
114 if (dentry->d_op && dentry->d_op->d_iput)
115 dentry->d_op->d_iput(dentry, inode);
119 spin_unlock(&dentry->d_lock);
120 spin_unlock(&dcache_lock);
125 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
127 static void dentry_lru_add(struct dentry *dentry)
129 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
130 dentry->d_sb->s_nr_dentry_unused++;
131 dentry_stat.nr_unused++;
134 static void dentry_lru_add_tail(struct dentry *dentry)
136 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
137 dentry->d_sb->s_nr_dentry_unused++;
138 dentry_stat.nr_unused++;
141 static void dentry_lru_del(struct dentry *dentry)
143 if (!list_empty(&dentry->d_lru)) {
144 list_del(&dentry->d_lru);
145 dentry->d_sb->s_nr_dentry_unused--;
146 dentry_stat.nr_unused--;
150 static void dentry_lru_del_init(struct dentry *dentry)
152 if (likely(!list_empty(&dentry->d_lru))) {
153 list_del_init(&dentry->d_lru);
154 dentry->d_sb->s_nr_dentry_unused--;
155 dentry_stat.nr_unused--;
160 * d_kill - kill dentry and return parent
161 * @dentry: dentry to kill
163 * The dentry must already be unhashed and removed from the LRU.
165 * If this is the root of the dentry tree, return NULL.
167 static struct dentry *d_kill(struct dentry *dentry)
168 __releases(dentry->d_lock)
169 __releases(dcache_lock)
171 struct dentry *parent;
173 list_del(&dentry->d_u.d_child);
174 dentry_stat.nr_dentry--; /* For d_free, below */
175 /*drops the locks, at that point nobody can reach this dentry */
180 parent = dentry->d_parent;
188 * This is complicated by the fact that we do not want to put
189 * dentries that are no longer on any hash chain on the unused
190 * list: we'd much rather just get rid of them immediately.
192 * However, that implies that we have to traverse the dentry
193 * tree upwards to the parents which might _also_ now be
194 * scheduled for deletion (it may have been only waiting for
195 * its last child to go away).
197 * This tail recursion is done by hand as we don't want to depend
198 * on the compiler to always get this right (gcc generally doesn't).
199 * Real recursion would eat up our stack space.
203 * dput - release a dentry
204 * @dentry: dentry to release
206 * Release a dentry. This will drop the usage count and if appropriate
207 * call the dentry unlink method as well as removing it from the queues and
208 * releasing its resources. If the parent dentries were scheduled for release
209 * they too may now get deleted.
211 * no dcache lock, please.
214 void dput(struct dentry *dentry)
220 if (atomic_read(&dentry->d_count) == 1)
222 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
225 spin_lock(&dentry->d_lock);
226 if (atomic_read(&dentry->d_count)) {
227 spin_unlock(&dentry->d_lock);
228 spin_unlock(&dcache_lock);
233 * AV: ->d_delete() is _NOT_ allowed to block now.
235 if (dentry->d_op && dentry->d_op->d_delete) {
236 if (dentry->d_op->d_delete(dentry))
239 /* Unreachable? Get rid of it */
240 if (d_unhashed(dentry))
242 if (list_empty(&dentry->d_lru)) {
243 dentry->d_flags |= DCACHE_REFERENCED;
244 dentry_lru_add(dentry);
246 spin_unlock(&dentry->d_lock);
247 spin_unlock(&dcache_lock);
253 /* if dentry was on the d_lru list delete it from there */
254 dentry_lru_del(dentry);
255 dentry = d_kill(dentry);
261 * d_invalidate - invalidate a dentry
262 * @dentry: dentry to invalidate
264 * Try to invalidate the dentry if it turns out to be
265 * possible. If there are other dentries that can be
266 * reached through this one we can't delete it and we
267 * return -EBUSY. On success we return 0.
272 int d_invalidate(struct dentry * dentry)
275 * If it's already been dropped, return OK.
277 spin_lock(&dcache_lock);
278 if (d_unhashed(dentry)) {
279 spin_unlock(&dcache_lock);
283 * Check whether to do a partial shrink_dcache
284 * to get rid of unused child entries.
286 if (!list_empty(&dentry->d_subdirs)) {
287 spin_unlock(&dcache_lock);
288 shrink_dcache_parent(dentry);
289 spin_lock(&dcache_lock);
293 * Somebody else still using it?
295 * If it's a directory, we can't drop it
296 * for fear of somebody re-populating it
297 * with children (even though dropping it
298 * would make it unreachable from the root,
299 * we might still populate it if it was a
300 * working directory or similar).
302 spin_lock(&dentry->d_lock);
303 if (atomic_read(&dentry->d_count) > 1) {
304 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
305 spin_unlock(&dentry->d_lock);
306 spin_unlock(&dcache_lock);
312 spin_unlock(&dentry->d_lock);
313 spin_unlock(&dcache_lock);
317 /* This should be called _only_ with dcache_lock held */
319 static inline struct dentry * __dget_locked(struct dentry *dentry)
321 atomic_inc(&dentry->d_count);
322 dentry_lru_del_init(dentry);
326 struct dentry * dget_locked(struct dentry *dentry)
328 return __dget_locked(dentry);
332 * d_find_alias - grab a hashed alias of inode
333 * @inode: inode in question
334 * @want_discon: flag, used by d_splice_alias, to request
335 * that only a DISCONNECTED alias be returned.
337 * If inode has a hashed alias, or is a directory and has any alias,
338 * acquire the reference to alias and return it. Otherwise return NULL.
339 * Notice that if inode is a directory there can be only one alias and
340 * it can be unhashed only if it has no children, or if it is the root
343 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
344 * any other hashed alias over that one unless @want_discon is set,
345 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
348 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
350 struct list_head *head, *next, *tmp;
351 struct dentry *alias, *discon_alias=NULL;
353 head = &inode->i_dentry;
354 next = inode->i_dentry.next;
355 while (next != head) {
359 alias = list_entry(tmp, struct dentry, d_alias);
360 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
361 if (IS_ROOT(alias) &&
362 (alias->d_flags & DCACHE_DISCONNECTED))
363 discon_alias = alias;
364 else if (!want_discon) {
365 __dget_locked(alias);
371 __dget_locked(discon_alias);
375 struct dentry * d_find_alias(struct inode *inode)
377 struct dentry *de = NULL;
379 if (!list_empty(&inode->i_dentry)) {
380 spin_lock(&dcache_lock);
381 de = __d_find_alias(inode, 0);
382 spin_unlock(&dcache_lock);
388 * Try to kill dentries associated with this inode.
389 * WARNING: you must own a reference to inode.
391 void d_prune_aliases(struct inode *inode)
393 struct dentry *dentry;
395 spin_lock(&dcache_lock);
396 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
397 spin_lock(&dentry->d_lock);
398 if (!atomic_read(&dentry->d_count)) {
399 __dget_locked(dentry);
401 spin_unlock(&dentry->d_lock);
402 spin_unlock(&dcache_lock);
406 spin_unlock(&dentry->d_lock);
408 spin_unlock(&dcache_lock);
412 * Throw away a dentry - free the inode, dput the parent. This requires that
413 * the LRU list has already been removed.
415 * Try to prune ancestors as well. This is necessary to prevent
416 * quadratic behavior of shrink_dcache_parent(), but is also expected
417 * to be beneficial in reducing dentry cache fragmentation.
419 static void prune_one_dentry(struct dentry * dentry)
420 __releases(dentry->d_lock)
421 __releases(dcache_lock)
422 __acquires(dcache_lock)
425 dentry = d_kill(dentry);
428 * Prune ancestors. Locking is simpler than in dput(),
429 * because dcache_lock needs to be taken anyway.
431 spin_lock(&dcache_lock);
433 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
436 if (dentry->d_op && dentry->d_op->d_delete)
437 dentry->d_op->d_delete(dentry);
438 dentry_lru_del_init(dentry);
440 dentry = d_kill(dentry);
441 spin_lock(&dcache_lock);
446 * Shrink the dentry LRU on a given superblock.
447 * @sb : superblock to shrink dentry LRU.
448 * @count: If count is NULL, we prune all dentries on superblock.
449 * @flags: If flags is non-zero, we need to do special processing based on
450 * which flags are set. This means we don't need to maintain multiple
451 * similar copies of this loop.
453 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
455 LIST_HEAD(referenced);
457 struct dentry *dentry;
461 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
462 spin_lock(&dcache_lock);
464 /* called from prune_dcache() and shrink_dcache_parent() */
468 list_splice_init(&sb->s_dentry_lru, &tmp);
470 while (!list_empty(&sb->s_dentry_lru)) {
471 dentry = list_entry(sb->s_dentry_lru.prev,
472 struct dentry, d_lru);
473 BUG_ON(dentry->d_sb != sb);
475 spin_lock(&dentry->d_lock);
477 * If we are honouring the DCACHE_REFERENCED flag and
478 * the dentry has this flag set, don't free it. Clear
479 * the flag and put it back on the LRU.
481 if ((flags & DCACHE_REFERENCED)
482 && (dentry->d_flags & DCACHE_REFERENCED)) {
483 dentry->d_flags &= ~DCACHE_REFERENCED;
484 list_move_tail(&dentry->d_lru, &referenced);
485 spin_unlock(&dentry->d_lock);
487 list_move_tail(&dentry->d_lru, &tmp);
488 spin_unlock(&dentry->d_lock);
493 cond_resched_lock(&dcache_lock);
496 while (!list_empty(&tmp)) {
497 dentry = list_entry(tmp.prev, struct dentry, d_lru);
498 dentry_lru_del_init(dentry);
499 spin_lock(&dentry->d_lock);
501 * We found an inuse dentry which was not removed from
502 * the LRU because of laziness during lookup. Do not free
503 * it - just keep it off the LRU list.
505 if (atomic_read(&dentry->d_count)) {
506 spin_unlock(&dentry->d_lock);
509 prune_one_dentry(dentry);
510 /* dentry->d_lock was dropped in prune_one_dentry() */
511 cond_resched_lock(&dcache_lock);
513 if (count == NULL && !list_empty(&sb->s_dentry_lru))
517 if (!list_empty(&referenced))
518 list_splice(&referenced, &sb->s_dentry_lru);
519 spin_unlock(&dcache_lock);
523 * prune_dcache - shrink the dcache
524 * @count: number of entries to try to free
526 * Shrink the dcache. This is done when we need more memory, or simply when we
527 * need to unmount something (at which point we need to unuse all dentries).
529 * This function may fail to free any resources if all the dentries are in use.
531 static void prune_dcache(int count)
533 struct super_block *sb;
535 int unused = dentry_stat.nr_unused;
539 if (unused == 0 || count == 0)
541 spin_lock(&dcache_lock);
546 prune_ratio = unused / count;
548 list_for_each_entry(sb, &super_blocks, s_list) {
549 if (sb->s_nr_dentry_unused == 0)
552 /* Now, we reclaim unused dentrins with fairness.
553 * We reclaim them same percentage from each superblock.
554 * We calculate number of dentries to scan on this sb
555 * as follows, but the implementation is arranged to avoid
557 * number of dentries to scan on this sb =
558 * count * (number of dentries on this sb /
559 * number of dentries in the machine)
561 spin_unlock(&sb_lock);
562 if (prune_ratio != 1)
563 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
565 w_count = sb->s_nr_dentry_unused;
568 * We need to be sure this filesystem isn't being unmounted,
569 * otherwise we could race with generic_shutdown_super(), and
570 * end up holding a reference to an inode while the filesystem
571 * is unmounted. So we try to get s_umount, and make sure
574 if (down_read_trylock(&sb->s_umount)) {
575 if ((sb->s_root != NULL) &&
576 (!list_empty(&sb->s_dentry_lru))) {
577 spin_unlock(&dcache_lock);
578 __shrink_dcache_sb(sb, &w_count,
581 spin_lock(&dcache_lock);
583 up_read(&sb->s_umount);
588 * restart only when sb is no longer on the list and
589 * we have more work to do.
591 if (__put_super_and_need_restart(sb) && count > 0) {
592 spin_unlock(&sb_lock);
596 spin_unlock(&sb_lock);
597 spin_unlock(&dcache_lock);
601 * shrink_dcache_sb - shrink dcache for a superblock
604 * Shrink the dcache for the specified super block. This
605 * is used to free the dcache before unmounting a file
608 void shrink_dcache_sb(struct super_block * sb)
610 __shrink_dcache_sb(sb, NULL, 0);
614 * destroy a single subtree of dentries for unmount
615 * - see the comments on shrink_dcache_for_umount() for a description of the
618 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
620 struct dentry *parent;
621 unsigned detached = 0;
623 BUG_ON(!IS_ROOT(dentry));
625 /* detach this root from the system */
626 spin_lock(&dcache_lock);
627 dentry_lru_del_init(dentry);
629 spin_unlock(&dcache_lock);
632 /* descend to the first leaf in the current subtree */
633 while (!list_empty(&dentry->d_subdirs)) {
636 /* this is a branch with children - detach all of them
637 * from the system in one go */
638 spin_lock(&dcache_lock);
639 list_for_each_entry(loop, &dentry->d_subdirs,
641 dentry_lru_del_init(loop);
643 cond_resched_lock(&dcache_lock);
645 spin_unlock(&dcache_lock);
647 /* move to the first child */
648 dentry = list_entry(dentry->d_subdirs.next,
649 struct dentry, d_u.d_child);
652 /* consume the dentries from this leaf up through its parents
653 * until we find one with children or run out altogether */
657 if (atomic_read(&dentry->d_count) != 0) {
659 "BUG: Dentry %p{i=%lx,n=%s}"
661 " [unmount of %s %s]\n",
664 dentry->d_inode->i_ino : 0UL,
666 atomic_read(&dentry->d_count),
667 dentry->d_sb->s_type->name,
675 parent = dentry->d_parent;
676 atomic_dec(&parent->d_count);
679 list_del(&dentry->d_u.d_child);
682 inode = dentry->d_inode;
684 dentry->d_inode = NULL;
685 list_del_init(&dentry->d_alias);
686 if (dentry->d_op && dentry->d_op->d_iput)
687 dentry->d_op->d_iput(dentry, inode);
694 /* finished when we fall off the top of the tree,
695 * otherwise we ascend to the parent and move to the
696 * next sibling if there is one */
702 } while (list_empty(&dentry->d_subdirs));
704 dentry = list_entry(dentry->d_subdirs.next,
705 struct dentry, d_u.d_child);
708 /* several dentries were freed, need to correct nr_dentry */
709 spin_lock(&dcache_lock);
710 dentry_stat.nr_dentry -= detached;
711 spin_unlock(&dcache_lock);
715 * destroy the dentries attached to a superblock on unmounting
716 * - we don't need to use dentry->d_lock, and only need dcache_lock when
717 * removing the dentry from the system lists and hashes because:
718 * - the superblock is detached from all mountings and open files, so the
719 * dentry trees will not be rearranged by the VFS
720 * - s_umount is write-locked, so the memory pressure shrinker will ignore
721 * any dentries belonging to this superblock that it comes across
722 * - the filesystem itself is no longer permitted to rearrange the dentries
725 void shrink_dcache_for_umount(struct super_block *sb)
727 struct dentry *dentry;
729 if (down_read_trylock(&sb->s_umount))
734 atomic_dec(&dentry->d_count);
735 shrink_dcache_for_umount_subtree(dentry);
737 while (!hlist_empty(&sb->s_anon)) {
738 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
739 shrink_dcache_for_umount_subtree(dentry);
744 * Search for at least 1 mount point in the dentry's subdirs.
745 * We descend to the next level whenever the d_subdirs
746 * list is non-empty and continue searching.
750 * have_submounts - check for mounts over a dentry
751 * @parent: dentry to check.
753 * Return true if the parent or its subdirectories contain
757 int have_submounts(struct dentry *parent)
759 struct dentry *this_parent = parent;
760 struct list_head *next;
762 spin_lock(&dcache_lock);
763 if (d_mountpoint(parent))
766 next = this_parent->d_subdirs.next;
768 while (next != &this_parent->d_subdirs) {
769 struct list_head *tmp = next;
770 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
772 /* Have we found a mount point ? */
773 if (d_mountpoint(dentry))
775 if (!list_empty(&dentry->d_subdirs)) {
776 this_parent = dentry;
781 * All done at this level ... ascend and resume the search.
783 if (this_parent != parent) {
784 next = this_parent->d_u.d_child.next;
785 this_parent = this_parent->d_parent;
788 spin_unlock(&dcache_lock);
789 return 0; /* No mount points found in tree */
791 spin_unlock(&dcache_lock);
796 * Search the dentry child list for the specified parent,
797 * and move any unused dentries to the end of the unused
798 * list for prune_dcache(). We descend to the next level
799 * whenever the d_subdirs list is non-empty and continue
802 * It returns zero iff there are no unused children,
803 * otherwise it returns the number of children moved to
804 * the end of the unused list. This may not be the total
805 * number of unused children, because select_parent can
806 * drop the lock and return early due to latency
809 static int select_parent(struct dentry * parent)
811 struct dentry *this_parent = parent;
812 struct list_head *next;
815 spin_lock(&dcache_lock);
817 next = this_parent->d_subdirs.next;
819 while (next != &this_parent->d_subdirs) {
820 struct list_head *tmp = next;
821 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
824 dentry_lru_del_init(dentry);
826 * move only zero ref count dentries to the end
827 * of the unused list for prune_dcache
829 if (!atomic_read(&dentry->d_count)) {
830 dentry_lru_add_tail(dentry);
835 * We can return to the caller if we have found some (this
836 * ensures forward progress). We'll be coming back to find
839 if (found && need_resched())
843 * Descend a level if the d_subdirs list is non-empty.
845 if (!list_empty(&dentry->d_subdirs)) {
846 this_parent = dentry;
851 * All done at this level ... ascend and resume the search.
853 if (this_parent != parent) {
854 next = this_parent->d_u.d_child.next;
855 this_parent = this_parent->d_parent;
859 spin_unlock(&dcache_lock);
864 * shrink_dcache_parent - prune dcache
865 * @parent: parent of entries to prune
867 * Prune the dcache to remove unused children of the parent dentry.
870 void shrink_dcache_parent(struct dentry * parent)
872 struct super_block *sb = parent->d_sb;
875 while ((found = select_parent(parent)) != 0)
876 __shrink_dcache_sb(sb, &found, 0);
880 * Scan `nr' dentries and return the number which remain.
882 * We need to avoid reentering the filesystem if the caller is performing a
883 * GFP_NOFS allocation attempt. One example deadlock is:
885 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
886 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
887 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
889 * In this case we return -1 to tell the caller that we baled.
891 static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
894 if (!(gfp_mask & __GFP_FS))
898 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
901 static struct shrinker dcache_shrinker = {
902 .shrink = shrink_dcache_memory,
903 .seeks = DEFAULT_SEEKS,
907 * d_alloc - allocate a dcache entry
908 * @parent: parent of entry to allocate
909 * @name: qstr of the name
911 * Allocates a dentry. It returns %NULL if there is insufficient memory
912 * available. On a success the dentry is returned. The name passed in is
913 * copied and the copy passed in may be reused after this call.
916 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
918 struct dentry *dentry;
921 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
925 if (name->len > DNAME_INLINE_LEN-1) {
926 dname = kmalloc(name->len + 1, GFP_KERNEL);
928 kmem_cache_free(dentry_cache, dentry);
932 dname = dentry->d_iname;
934 dentry->d_name.name = dname;
936 dentry->d_name.len = name->len;
937 dentry->d_name.hash = name->hash;
938 memcpy(dname, name->name, name->len);
939 dname[name->len] = 0;
941 atomic_set(&dentry->d_count, 1);
942 dentry->d_flags = DCACHE_UNHASHED;
943 spin_lock_init(&dentry->d_lock);
944 dentry->d_inode = NULL;
945 dentry->d_parent = NULL;
948 dentry->d_fsdata = NULL;
949 dentry->d_mounted = 0;
950 #ifdef CONFIG_PROFILING
951 dentry->d_cookie = NULL;
953 INIT_HLIST_NODE(&dentry->d_hash);
954 INIT_LIST_HEAD(&dentry->d_lru);
955 INIT_LIST_HEAD(&dentry->d_subdirs);
956 INIT_LIST_HEAD(&dentry->d_alias);
959 dentry->d_parent = dget(parent);
960 dentry->d_sb = parent->d_sb;
962 INIT_LIST_HEAD(&dentry->d_u.d_child);
965 spin_lock(&dcache_lock);
967 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
968 dentry_stat.nr_dentry++;
969 spin_unlock(&dcache_lock);
974 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
979 q.len = strlen(name);
980 q.hash = full_name_hash(q.name, q.len);
981 return d_alloc(parent, &q);
984 /* the caller must hold dcache_lock */
985 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
988 list_add(&dentry->d_alias, &inode->i_dentry);
989 dentry->d_inode = inode;
990 fsnotify_d_instantiate(dentry, inode);
994 * d_instantiate - fill in inode information for a dentry
995 * @entry: dentry to complete
996 * @inode: inode to attach to this dentry
998 * Fill in inode information in the entry.
1000 * This turns negative dentries into productive full members
1003 * NOTE! This assumes that the inode count has been incremented
1004 * (or otherwise set) by the caller to indicate that it is now
1005 * in use by the dcache.
1008 void d_instantiate(struct dentry *entry, struct inode * inode)
1010 BUG_ON(!list_empty(&entry->d_alias));
1011 spin_lock(&dcache_lock);
1012 __d_instantiate(entry, inode);
1013 spin_unlock(&dcache_lock);
1014 security_d_instantiate(entry, inode);
1018 * d_instantiate_unique - instantiate a non-aliased dentry
1019 * @entry: dentry to instantiate
1020 * @inode: inode to attach to this dentry
1022 * Fill in inode information in the entry. On success, it returns NULL.
1023 * If an unhashed alias of "entry" already exists, then we return the
1024 * aliased dentry instead and drop one reference to inode.
1026 * Note that in order to avoid conflicts with rename() etc, the caller
1027 * had better be holding the parent directory semaphore.
1029 * This also assumes that the inode count has been incremented
1030 * (or otherwise set) by the caller to indicate that it is now
1031 * in use by the dcache.
1033 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1034 struct inode *inode)
1036 struct dentry *alias;
1037 int len = entry->d_name.len;
1038 const char *name = entry->d_name.name;
1039 unsigned int hash = entry->d_name.hash;
1042 __d_instantiate(entry, NULL);
1046 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1047 struct qstr *qstr = &alias->d_name;
1049 if (qstr->hash != hash)
1051 if (alias->d_parent != entry->d_parent)
1053 if (qstr->len != len)
1055 if (memcmp(qstr->name, name, len))
1061 __d_instantiate(entry, inode);
1065 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1067 struct dentry *result;
1069 BUG_ON(!list_empty(&entry->d_alias));
1071 spin_lock(&dcache_lock);
1072 result = __d_instantiate_unique(entry, inode);
1073 spin_unlock(&dcache_lock);
1076 security_d_instantiate(entry, inode);
1080 BUG_ON(!d_unhashed(result));
1085 EXPORT_SYMBOL(d_instantiate_unique);
1088 * d_alloc_root - allocate root dentry
1089 * @root_inode: inode to allocate the root for
1091 * Allocate a root ("/") dentry for the inode given. The inode is
1092 * instantiated and returned. %NULL is returned if there is insufficient
1093 * memory or the inode passed is %NULL.
1096 struct dentry * d_alloc_root(struct inode * root_inode)
1098 struct dentry *res = NULL;
1101 static const struct qstr name = { .name = "/", .len = 1 };
1103 res = d_alloc(NULL, &name);
1105 res->d_sb = root_inode->i_sb;
1106 res->d_parent = res;
1107 d_instantiate(res, root_inode);
1113 static inline struct hlist_head *d_hash(struct dentry *parent,
1116 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1117 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1118 return dentry_hashtable + (hash & D_HASHMASK);
1122 * d_obtain_alias - find or allocate a dentry for a given inode
1123 * @inode: inode to allocate the dentry for
1125 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1126 * similar open by handle operations. The returned dentry may be anonymous,
1127 * or may have a full name (if the inode was already in the cache).
1129 * When called on a directory inode, we must ensure that the inode only ever
1130 * has one dentry. If a dentry is found, that is returned instead of
1131 * allocating a new one.
1133 * On successful return, the reference to the inode has been transferred
1134 * to the dentry. In case of an error the reference on the inode is released.
1135 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1136 * be passed in and will be the error will be propagate to the return value,
1137 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1139 struct dentry *d_obtain_alias(struct inode *inode)
1141 static const struct qstr anonstring = { .name = "" };
1146 return ERR_PTR(-ESTALE);
1148 return ERR_CAST(inode);
1150 res = d_find_alias(inode);
1154 tmp = d_alloc(NULL, &anonstring);
1156 res = ERR_PTR(-ENOMEM);
1159 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1161 spin_lock(&dcache_lock);
1162 res = __d_find_alias(inode, 0);
1164 spin_unlock(&dcache_lock);
1169 /* attach a disconnected dentry */
1170 spin_lock(&tmp->d_lock);
1171 tmp->d_sb = inode->i_sb;
1172 tmp->d_inode = inode;
1173 tmp->d_flags |= DCACHE_DISCONNECTED;
1174 tmp->d_flags &= ~DCACHE_UNHASHED;
1175 list_add(&tmp->d_alias, &inode->i_dentry);
1176 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1177 spin_unlock(&tmp->d_lock);
1179 spin_unlock(&dcache_lock);
1186 EXPORT_SYMBOL_GPL(d_obtain_alias);
1189 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1190 * @inode: the inode which may have a disconnected dentry
1191 * @dentry: a negative dentry which we want to point to the inode.
1193 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1194 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1195 * and return it, else simply d_add the inode to the dentry and return NULL.
1197 * This is needed in the lookup routine of any filesystem that is exportable
1198 * (via knfsd) so that we can build dcache paths to directories effectively.
1200 * If a dentry was found and moved, then it is returned. Otherwise NULL
1201 * is returned. This matches the expected return value of ->lookup.
1204 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1206 struct dentry *new = NULL;
1208 if (inode && S_ISDIR(inode->i_mode)) {
1209 spin_lock(&dcache_lock);
1210 new = __d_find_alias(inode, 1);
1212 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1213 fsnotify_d_instantiate(new, inode);
1214 spin_unlock(&dcache_lock);
1215 security_d_instantiate(new, inode);
1217 d_move(new, dentry);
1220 /* already taking dcache_lock, so d_add() by hand */
1221 __d_instantiate(dentry, inode);
1222 spin_unlock(&dcache_lock);
1223 security_d_instantiate(dentry, inode);
1227 d_add(dentry, inode);
1232 * d_add_ci - lookup or allocate new dentry with case-exact name
1233 * @inode: the inode case-insensitive lookup has found
1234 * @dentry: the negative dentry that was passed to the parent's lookup func
1235 * @name: the case-exact name to be associated with the returned dentry
1237 * This is to avoid filling the dcache with case-insensitive names to the
1238 * same inode, only the actual correct case is stored in the dcache for
1239 * case-insensitive filesystems.
1241 * For a case-insensitive lookup match and if the the case-exact dentry
1242 * already exists in in the dcache, use it and return it.
1244 * If no entry exists with the exact case name, allocate new dentry with
1245 * the exact case, and return the spliced entry.
1247 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1251 struct dentry *found;
1254 /* Does a dentry matching the name exist already? */
1255 found = d_hash_and_lookup(dentry->d_parent, name);
1256 /* If not, create it now and return */
1258 new = d_alloc(dentry->d_parent, name);
1263 found = d_splice_alias(inode, new);
1270 /* Matching dentry exists, check if it is negative. */
1271 if (found->d_inode) {
1272 if (unlikely(found->d_inode != inode)) {
1273 /* This can't happen because bad inodes are unhashed. */
1274 BUG_ON(!is_bad_inode(inode));
1275 BUG_ON(!is_bad_inode(found->d_inode));
1278 * Already have the inode and the dentry attached, decrement
1279 * the reference count to balance the iget() done
1280 * earlier on. We found the dentry using d_lookup() so it
1281 * cannot be disconnected and thus we do not need to worry
1282 * about any NFS/disconnectedness issues here.
1288 * Negative dentry: instantiate it unless the inode is a directory and
1289 * has a 'disconnected' dentry (i.e. IS_ROOT and DCACHE_DISCONNECTED),
1290 * in which case d_move() that in place of the found dentry.
1292 if (!S_ISDIR(inode->i_mode)) {
1293 /* Not a directory; everything is easy. */
1294 d_instantiate(found, inode);
1297 spin_lock(&dcache_lock);
1298 if (list_empty(&inode->i_dentry)) {
1300 * Directory without a 'disconnected' dentry; we need to do
1301 * d_instantiate() by hand because it takes dcache_lock which
1304 __d_instantiate(found, inode);
1305 spin_unlock(&dcache_lock);
1306 security_d_instantiate(found, inode);
1310 * Directory with a 'disconnected' dentry; get a reference to the
1311 * 'disconnected' dentry.
1313 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1315 spin_unlock(&dcache_lock);
1316 /* Do security vodoo. */
1317 security_d_instantiate(found, inode);
1318 /* Move new in place of found. */
1320 /* Balance the iget() we did above. */
1322 /* Throw away found. */
1324 /* Use new as the actual dentry. */
1329 return ERR_PTR(error);
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 d_put 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));
1375 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1377 unsigned int len = name->len;
1378 unsigned int hash = name->hash;
1379 const unsigned char *str = name->name;
1380 struct hlist_head *head = d_hash(parent,hash);
1381 struct dentry *found = NULL;
1382 struct hlist_node *node;
1383 struct dentry *dentry;
1387 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1390 if (dentry->d_name.hash != hash)
1392 if (dentry->d_parent != parent)
1395 spin_lock(&dentry->d_lock);
1398 * Recheck the dentry after taking the lock - d_move may have
1399 * changed things. Don't bother checking the hash because we're
1400 * about to compare the whole name anyway.
1402 if (dentry->d_parent != parent)
1405 /* non-existing due to RCU? */
1406 if (d_unhashed(dentry))
1410 * It is safe to compare names since d_move() cannot
1411 * change the qstr (protected by d_lock).
1413 qstr = &dentry->d_name;
1414 if (parent->d_op && parent->d_op->d_compare) {
1415 if (parent->d_op->d_compare(parent, qstr, name))
1418 if (qstr->len != len)
1420 if (memcmp(qstr->name, str, len))
1424 atomic_inc(&dentry->d_count);
1426 spin_unlock(&dentry->d_lock);
1429 spin_unlock(&dentry->d_lock);
1437 * d_hash_and_lookup - hash the qstr then search for a dentry
1438 * @dir: Directory to search in
1439 * @name: qstr of name we wish to find
1441 * On hash failure or on lookup failure NULL is returned.
1443 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1445 struct dentry *dentry = NULL;
1448 * Check for a fs-specific hash function. Note that we must
1449 * calculate the standard hash first, as the d_op->d_hash()
1450 * routine may choose to leave the hash value unchanged.
1452 name->hash = full_name_hash(name->name, name->len);
1453 if (dir->d_op && dir->d_op->d_hash) {
1454 if (dir->d_op->d_hash(dir, name) < 0)
1457 dentry = d_lookup(dir, name);
1463 * d_validate - verify dentry provided from insecure source
1464 * @dentry: The dentry alleged to be valid child of @dparent
1465 * @dparent: The parent dentry (known to be valid)
1466 * @hash: Hash of the dentry
1467 * @len: Length of the name
1469 * An insecure source has sent us a dentry, here we verify it and dget() it.
1470 * This is used by ncpfs in its readdir implementation.
1471 * Zero is returned in the dentry is invalid.
1474 int d_validate(struct dentry *dentry, struct dentry *dparent)
1476 struct hlist_head *base;
1477 struct hlist_node *lhp;
1479 /* Check whether the ptr might be valid at all.. */
1480 if (!kmem_ptr_validate(dentry_cache, dentry))
1483 if (dentry->d_parent != dparent)
1486 spin_lock(&dcache_lock);
1487 base = d_hash(dparent, dentry->d_name.hash);
1488 hlist_for_each(lhp,base) {
1489 /* hlist_for_each_entry_rcu() not required for d_hash list
1490 * as it is parsed under dcache_lock
1492 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1493 __dget_locked(dentry);
1494 spin_unlock(&dcache_lock);
1498 spin_unlock(&dcache_lock);
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_iput(dentry);
1535 fsnotify_nameremove(dentry, isdir);
1539 if (!d_unhashed(dentry))
1542 spin_unlock(&dentry->d_lock);
1543 spin_unlock(&dcache_lock);
1545 fsnotify_nameremove(dentry, isdir);
1548 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1551 entry->d_flags &= ~DCACHE_UNHASHED;
1552 hlist_add_head_rcu(&entry->d_hash, list);
1555 static void _d_rehash(struct dentry * entry)
1557 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1561 * d_rehash - add an entry back to the hash
1562 * @entry: dentry to add to the hash
1564 * Adds a dentry to the hash according to its name.
1567 void d_rehash(struct dentry * entry)
1569 spin_lock(&dcache_lock);
1570 spin_lock(&entry->d_lock);
1572 spin_unlock(&entry->d_lock);
1573 spin_unlock(&dcache_lock);
1576 #define do_switch(x,y) do { \
1577 __typeof__ (x) __tmp = x; \
1578 x = y; y = __tmp; } while (0)
1581 * When switching names, the actual string doesn't strictly have to
1582 * be preserved in the target - because we're dropping the target
1583 * anyway. As such, we can just do a simple memcpy() to copy over
1584 * the new name before we switch.
1586 * Note that we have to be a lot more careful about getting the hash
1587 * switched - we have to switch the hash value properly even if it
1588 * then no longer matches the actual (corrupted) string of the target.
1589 * The hash value has to match the hash queue that the dentry is on..
1591 static void switch_names(struct dentry *dentry, struct dentry *target)
1593 if (dname_external(target)) {
1594 if (dname_external(dentry)) {
1596 * Both external: swap the pointers
1598 do_switch(target->d_name.name, dentry->d_name.name);
1601 * dentry:internal, target:external. Steal target's
1602 * storage and make target internal.
1604 memcpy(target->d_iname, dentry->d_name.name,
1605 dentry->d_name.len + 1);
1606 dentry->d_name.name = target->d_name.name;
1607 target->d_name.name = target->d_iname;
1610 if (dname_external(dentry)) {
1612 * dentry:external, target:internal. Give dentry's
1613 * storage to target and make dentry internal
1615 memcpy(dentry->d_iname, target->d_name.name,
1616 target->d_name.len + 1);
1617 target->d_name.name = dentry->d_name.name;
1618 dentry->d_name.name = dentry->d_iname;
1621 * Both are internal. Just copy target to dentry
1623 memcpy(dentry->d_iname, target->d_name.name,
1624 target->d_name.len + 1);
1630 * We cannibalize "target" when moving dentry on top of it,
1631 * because it's going to be thrown away anyway. We could be more
1632 * polite about it, though.
1634 * This forceful removal will result in ugly /proc output if
1635 * somebody holds a file open that got deleted due to a rename.
1636 * We could be nicer about the deleted file, and let it show
1637 * up under the name it had before it was deleted rather than
1638 * under the original name of the file that was moved on top of it.
1642 * d_move_locked - move a dentry
1643 * @dentry: entry to move
1644 * @target: new dentry
1646 * Update the dcache to reflect the move of a file name. Negative
1647 * dcache entries should not be moved in this way.
1649 static void d_move_locked(struct dentry * dentry, struct dentry * target)
1651 struct hlist_head *list;
1653 if (!dentry->d_inode)
1654 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1656 write_seqlock(&rename_lock);
1658 * XXXX: do we really need to take target->d_lock?
1660 if (target < dentry) {
1661 spin_lock(&target->d_lock);
1662 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1664 spin_lock(&dentry->d_lock);
1665 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
1668 /* Move the dentry to the target hash queue, if on different bucket */
1669 if (d_unhashed(dentry))
1670 goto already_unhashed;
1672 hlist_del_rcu(&dentry->d_hash);
1675 list = d_hash(target->d_parent, target->d_name.hash);
1676 __d_rehash(dentry, list);
1678 /* Unhash the target: dput() will then get rid of it */
1681 list_del(&dentry->d_u.d_child);
1682 list_del(&target->d_u.d_child);
1684 /* Switch the names.. */
1685 switch_names(dentry, target);
1686 do_switch(dentry->d_name.len, target->d_name.len);
1687 do_switch(dentry->d_name.hash, target->d_name.hash);
1689 /* ... and switch the parents */
1690 if (IS_ROOT(dentry)) {
1691 dentry->d_parent = target->d_parent;
1692 target->d_parent = target;
1693 INIT_LIST_HEAD(&target->d_u.d_child);
1695 do_switch(dentry->d_parent, target->d_parent);
1697 /* And add them back to the (new) parent lists */
1698 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1701 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1702 spin_unlock(&target->d_lock);
1703 fsnotify_d_move(dentry);
1704 spin_unlock(&dentry->d_lock);
1705 write_sequnlock(&rename_lock);
1709 * d_move - move a dentry
1710 * @dentry: entry to move
1711 * @target: new dentry
1713 * Update the dcache to reflect the move of a file name. Negative
1714 * dcache entries should not be moved in this way.
1717 void d_move(struct dentry * dentry, struct dentry * target)
1719 spin_lock(&dcache_lock);
1720 d_move_locked(dentry, target);
1721 spin_unlock(&dcache_lock);
1725 * d_ancestor - search for an ancestor
1726 * @p1: ancestor dentry
1729 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1730 * an ancestor of p2, else NULL.
1732 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
1736 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
1737 if (p->d_parent == p1)
1744 * This helper attempts to cope with remotely renamed directories
1746 * It assumes that the caller is already holding
1747 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1749 * Note: If ever the locking in lock_rename() changes, then please
1750 * remember to update this too...
1752 static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
1753 __releases(dcache_lock)
1755 struct mutex *m1 = NULL, *m2 = NULL;
1758 /* If alias and dentry share a parent, then no extra locks required */
1759 if (alias->d_parent == dentry->d_parent)
1762 /* Check for loops */
1763 ret = ERR_PTR(-ELOOP);
1764 if (d_ancestor(alias, dentry))
1767 /* See lock_rename() */
1768 ret = ERR_PTR(-EBUSY);
1769 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1771 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1772 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1774 m2 = &alias->d_parent->d_inode->i_mutex;
1776 d_move_locked(alias, dentry);
1779 spin_unlock(&dcache_lock);
1788 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1789 * named dentry in place of the dentry to be replaced.
1791 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1793 struct dentry *dparent, *aparent;
1795 switch_names(dentry, anon);
1796 do_switch(dentry->d_name.len, anon->d_name.len);
1797 do_switch(dentry->d_name.hash, anon->d_name.hash);
1799 dparent = dentry->d_parent;
1800 aparent = anon->d_parent;
1802 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1803 list_del(&dentry->d_u.d_child);
1804 if (!IS_ROOT(dentry))
1805 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1807 INIT_LIST_HEAD(&dentry->d_u.d_child);
1809 anon->d_parent = (dparent == dentry) ? anon : dparent;
1810 list_del(&anon->d_u.d_child);
1812 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1814 INIT_LIST_HEAD(&anon->d_u.d_child);
1816 anon->d_flags &= ~DCACHE_DISCONNECTED;
1820 * d_materialise_unique - introduce an inode into the tree
1821 * @dentry: candidate dentry
1822 * @inode: inode to bind to the dentry, to which aliases may be attached
1824 * Introduces an dentry into the tree, substituting an extant disconnected
1825 * root directory alias in its place if there is one
1827 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1829 struct dentry *actual;
1831 BUG_ON(!d_unhashed(dentry));
1833 spin_lock(&dcache_lock);
1837 __d_instantiate(dentry, NULL);
1841 if (S_ISDIR(inode->i_mode)) {
1842 struct dentry *alias;
1844 /* Does an aliased dentry already exist? */
1845 alias = __d_find_alias(inode, 0);
1848 /* Is this an anonymous mountpoint that we could splice
1850 if (IS_ROOT(alias)) {
1851 spin_lock(&alias->d_lock);
1852 __d_materialise_dentry(dentry, alias);
1856 /* Nope, but we must(!) avoid directory aliasing */
1857 actual = __d_unalias(dentry, alias);
1864 /* Add a unique reference */
1865 actual = __d_instantiate_unique(dentry, inode);
1868 else if (unlikely(!d_unhashed(actual)))
1869 goto shouldnt_be_hashed;
1872 spin_lock(&actual->d_lock);
1875 spin_unlock(&actual->d_lock);
1876 spin_unlock(&dcache_lock);
1878 if (actual == dentry) {
1879 security_d_instantiate(dentry, inode);
1887 spin_unlock(&dcache_lock);
1891 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
1895 return -ENAMETOOLONG;
1897 memcpy(*buffer, str, namelen);
1901 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1903 return prepend(buffer, buflen, name->name, name->len);
1907 * __d_path - return the path of a dentry
1908 * @path: the dentry/vfsmount to report
1909 * @root: root vfsmnt/dentry (may be modified by this function)
1910 * @buffer: buffer to return value in
1911 * @buflen: buffer length
1913 * Convert a dentry into an ASCII path name. If the entry has been deleted
1914 * the string " (deleted)" is appended. Note that this is ambiguous.
1916 * Returns the buffer or an error code if the path was too long.
1918 * "buflen" should be positive. Caller holds the dcache_lock.
1920 * If path is not reachable from the supplied root, then the value of
1921 * root is changed (without modifying refcounts).
1923 char *__d_path(const struct path *path, struct path *root,
1924 char *buffer, int buflen)
1926 struct dentry *dentry = path->dentry;
1927 struct vfsmount *vfsmnt = path->mnt;
1928 char *end = buffer + buflen;
1931 spin_lock(&vfsmount_lock);
1932 prepend(&end, &buflen, "\0", 1);
1933 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
1934 (prepend(&end, &buflen, " (deleted)", 10) != 0))
1944 struct dentry * parent;
1946 if (dentry == root->dentry && vfsmnt == root->mnt)
1948 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1950 if (vfsmnt->mnt_parent == vfsmnt) {
1953 dentry = vfsmnt->mnt_mountpoint;
1954 vfsmnt = vfsmnt->mnt_parent;
1957 parent = dentry->d_parent;
1959 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
1960 (prepend(&end, &buflen, "/", 1) != 0))
1967 spin_unlock(&vfsmount_lock);
1971 retval += 1; /* hit the slash */
1972 if (prepend_name(&retval, &buflen, &dentry->d_name) != 0)
1975 root->dentry = dentry;
1979 retval = ERR_PTR(-ENAMETOOLONG);
1984 * d_path - return the path of a dentry
1985 * @path: path to report
1986 * @buf: buffer to return value in
1987 * @buflen: buffer length
1989 * Convert a dentry into an ASCII path name. If the entry has been deleted
1990 * the string " (deleted)" is appended. Note that this is ambiguous.
1992 * Returns the buffer or an error code if the path was too long.
1994 * "buflen" should be positive.
1996 char *d_path(const struct path *path, char *buf, int buflen)
2003 * We have various synthetic filesystems that never get mounted. On
2004 * these filesystems dentries are never used for lookup purposes, and
2005 * thus don't need to be hashed. They also don't need a name until a
2006 * user wants to identify the object in /proc/pid/fd/. The little hack
2007 * below allows us to generate a name for these objects on demand:
2009 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2010 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2012 read_lock(¤t->fs->lock);
2013 root = current->fs->root;
2015 read_unlock(¤t->fs->lock);
2016 spin_lock(&dcache_lock);
2018 res = __d_path(path, &tmp, buf, buflen);
2019 spin_unlock(&dcache_lock);
2025 * Helper function for dentry_operations.d_dname() members
2027 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2028 const char *fmt, ...)
2034 va_start(args, fmt);
2035 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2038 if (sz > sizeof(temp) || sz > buflen)
2039 return ERR_PTR(-ENAMETOOLONG);
2041 buffer += buflen - sz;
2042 return memcpy(buffer, temp, sz);
2046 * Write full pathname from the root of the filesystem into the buffer.
2048 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2050 char *end = buf + buflen;
2053 spin_lock(&dcache_lock);
2054 prepend(&end, &buflen, "\0", 1);
2055 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
2056 (prepend(&end, &buflen, "//deleted", 9) != 0))
2064 while (!IS_ROOT(dentry)) {
2065 struct dentry *parent = dentry->d_parent;
2068 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
2069 (prepend(&end, &buflen, "/", 1) != 0))
2075 spin_unlock(&dcache_lock);
2078 spin_unlock(&dcache_lock);
2079 return ERR_PTR(-ENAMETOOLONG);
2083 * NOTE! The user-level library version returns a
2084 * character pointer. The kernel system call just
2085 * returns the length of the buffer filled (which
2086 * includes the ending '\0' character), or a negative
2087 * error value. So libc would do something like
2089 * char *getcwd(char * buf, size_t size)
2093 * retval = sys_getcwd(buf, size);
2100 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
2103 struct path pwd, root;
2104 char *page = (char *) __get_free_page(GFP_USER);
2109 read_lock(¤t->fs->lock);
2110 pwd = current->fs->pwd;
2112 root = current->fs->root;
2114 read_unlock(¤t->fs->lock);
2117 /* Has the current directory has been unlinked? */
2118 spin_lock(&dcache_lock);
2119 if (IS_ROOT(pwd.dentry) || !d_unhashed(pwd.dentry)) {
2121 struct path tmp = root;
2124 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
2125 spin_unlock(&dcache_lock);
2127 error = PTR_ERR(cwd);
2132 len = PAGE_SIZE + page - cwd;
2135 if (copy_to_user(buf, cwd, len))
2139 spin_unlock(&dcache_lock);
2144 free_page((unsigned long) page);
2149 * Test whether new_dentry is a subdirectory of old_dentry.
2151 * Trivially implemented using the dcache structure
2155 * is_subdir - is new dentry a subdirectory of old_dentry
2156 * @new_dentry: new dentry
2157 * @old_dentry: old dentry
2159 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2160 * Returns 0 otherwise.
2161 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2164 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2169 /* FIXME: This is old behavior, needed? Please check callers. */
2170 if (new_dentry == old_dentry)
2174 * Need rcu_readlock to protect against the d_parent trashing
2179 /* for restarting inner loop in case of seq retry */
2180 seq = read_seqbegin(&rename_lock);
2181 if (d_ancestor(old_dentry, new_dentry))
2185 } while (read_seqretry(&rename_lock, seq));
2191 void d_genocide(struct dentry *root)
2193 struct dentry *this_parent = root;
2194 struct list_head *next;
2196 spin_lock(&dcache_lock);
2198 next = this_parent->d_subdirs.next;
2200 while (next != &this_parent->d_subdirs) {
2201 struct list_head *tmp = next;
2202 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2204 if (d_unhashed(dentry)||!dentry->d_inode)
2206 if (!list_empty(&dentry->d_subdirs)) {
2207 this_parent = dentry;
2210 atomic_dec(&dentry->d_count);
2212 if (this_parent != root) {
2213 next = this_parent->d_u.d_child.next;
2214 atomic_dec(&this_parent->d_count);
2215 this_parent = this_parent->d_parent;
2218 spin_unlock(&dcache_lock);
2222 * find_inode_number - check for dentry with name
2223 * @dir: directory to check
2224 * @name: Name to find.
2226 * Check whether a dentry already exists for the given name,
2227 * and return the inode number if it has an inode. Otherwise
2230 * This routine is used to post-process directory listings for
2231 * filesystems using synthetic inode numbers, and is necessary
2232 * to keep getcwd() working.
2235 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2237 struct dentry * dentry;
2240 dentry = d_hash_and_lookup(dir, name);
2242 if (dentry->d_inode)
2243 ino = dentry->d_inode->i_ino;
2249 static __initdata unsigned long dhash_entries;
2250 static int __init set_dhash_entries(char *str)
2254 dhash_entries = simple_strtoul(str, &str, 0);
2257 __setup("dhash_entries=", set_dhash_entries);
2259 static void __init dcache_init_early(void)
2263 /* If hashes are distributed across NUMA nodes, defer
2264 * hash allocation until vmalloc space is available.
2270 alloc_large_system_hash("Dentry cache",
2271 sizeof(struct hlist_head),
2279 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2280 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2283 static void __init dcache_init(void)
2288 * A constructor could be added for stable state like the lists,
2289 * but it is probably not worth it because of the cache nature
2292 dentry_cache = KMEM_CACHE(dentry,
2293 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2295 register_shrinker(&dcache_shrinker);
2297 /* Hash may have been set up in dcache_init_early */
2302 alloc_large_system_hash("Dentry cache",
2303 sizeof(struct hlist_head),
2311 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2312 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2315 /* SLAB cache for __getname() consumers */
2316 struct kmem_cache *names_cachep __read_mostly;
2318 /* SLAB cache for file structures */
2319 struct kmem_cache *filp_cachep __read_mostly;
2321 EXPORT_SYMBOL(d_genocide);
2323 void __init vfs_caches_init_early(void)
2325 dcache_init_early();
2329 void __init vfs_caches_init(unsigned long mempages)
2331 unsigned long reserve;
2333 /* Base hash sizes on available memory, with a reserve equal to
2334 150% of current kernel size */
2336 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2337 mempages -= reserve;
2339 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
2340 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2342 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
2343 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2347 files_init(mempages);
2353 EXPORT_SYMBOL(d_alloc);
2354 EXPORT_SYMBOL(d_alloc_root);
2355 EXPORT_SYMBOL(d_delete);
2356 EXPORT_SYMBOL(d_find_alias);
2357 EXPORT_SYMBOL(d_instantiate);
2358 EXPORT_SYMBOL(d_invalidate);
2359 EXPORT_SYMBOL(d_lookup);
2360 EXPORT_SYMBOL(d_move);
2361 EXPORT_SYMBOL_GPL(d_materialise_unique);
2362 EXPORT_SYMBOL(d_path);
2363 EXPORT_SYMBOL(d_prune_aliases);
2364 EXPORT_SYMBOL(d_rehash);
2365 EXPORT_SYMBOL(d_splice_alias);
2366 EXPORT_SYMBOL(d_add_ci);
2367 EXPORT_SYMBOL(d_validate);
2368 EXPORT_SYMBOL(dget_locked);
2369 EXPORT_SYMBOL(dput);
2370 EXPORT_SYMBOL(find_inode_number);
2371 EXPORT_SYMBOL(have_submounts);
2372 EXPORT_SYMBOL(names_cachep);
2373 EXPORT_SYMBOL(shrink_dcache_parent);
2374 EXPORT_SYMBOL(shrink_dcache_sb);