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/smp_lock.h>
25 #include <linux/hash.h>
26 #include <linux/cache.h>
27 #include <linux/module.h>
28 #include <linux/mount.h>
29 #include <linux/file.h>
30 #include <asm/uaccess.h>
31 #include <linux/security.h>
32 #include <linux/seqlock.h>
33 #include <linux/swap.h>
34 #include <linux/bootmem.h>
38 int sysctl_vfs_cache_pressure __read_mostly
= 100;
39 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
41 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lock
);
42 static __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
;
64 static LIST_HEAD(dentry_unused
);
66 /* Statistics gathering. */
67 struct dentry_stat_t dentry_stat
= {
71 static void __d_free(struct dentry
*dentry
)
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 (dentry
->d_hash
.pprev
== NULL
)
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.
102 * Called with dcache_lock and per dentry lock held, drops both.
104 static void dentry_iput(struct dentry
* dentry
)
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 * d_kill - kill dentry and return parent
126 * @dentry: dentry to kill
128 * Called with dcache_lock and d_lock, releases both. The dentry must
129 * already be unhashed and removed from the LRU.
131 * If this is the root of the dentry tree, return NULL.
133 static struct dentry
*d_kill(struct dentry
*dentry
)
135 struct dentry
*parent
;
137 list_del(&dentry
->d_u
.d_child
);
138 dentry_stat
.nr_dentry
--; /* For d_free, below */
139 /*drops the locks, at that point nobody can reach this dentry */
141 parent
= dentry
->d_parent
;
143 return dentry
== parent
? NULL
: parent
;
149 * This is complicated by the fact that we do not want to put
150 * dentries that are no longer on any hash chain on the unused
151 * list: we'd much rather just get rid of them immediately.
153 * However, that implies that we have to traverse the dentry
154 * tree upwards to the parents which might _also_ now be
155 * scheduled for deletion (it may have been only waiting for
156 * its last child to go away).
158 * This tail recursion is done by hand as we don't want to depend
159 * on the compiler to always get this right (gcc generally doesn't).
160 * Real recursion would eat up our stack space.
164 * dput - release a dentry
165 * @dentry: dentry to release
167 * Release a dentry. This will drop the usage count and if appropriate
168 * call the dentry unlink method as well as removing it from the queues and
169 * releasing its resources. If the parent dentries were scheduled for release
170 * they too may now get deleted.
172 * no dcache lock, please.
175 void dput(struct dentry
*dentry
)
181 if (atomic_read(&dentry
->d_count
) == 1)
183 if (!atomic_dec_and_lock(&dentry
->d_count
, &dcache_lock
))
186 spin_lock(&dentry
->d_lock
);
187 if (atomic_read(&dentry
->d_count
)) {
188 spin_unlock(&dentry
->d_lock
);
189 spin_unlock(&dcache_lock
);
194 * AV: ->d_delete() is _NOT_ allowed to block now.
196 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
197 if (dentry
->d_op
->d_delete(dentry
))
200 /* Unreachable? Get rid of it */
201 if (d_unhashed(dentry
))
203 if (list_empty(&dentry
->d_lru
)) {
204 dentry
->d_flags
|= DCACHE_REFERENCED
;
205 list_add(&dentry
->d_lru
, &dentry_unused
);
206 dentry_stat
.nr_unused
++;
208 spin_unlock(&dentry
->d_lock
);
209 spin_unlock(&dcache_lock
);
215 /* If dentry was on d_lru list
216 * delete it from there
218 if (!list_empty(&dentry
->d_lru
)) {
219 list_del(&dentry
->d_lru
);
220 dentry_stat
.nr_unused
--;
222 dentry
= d_kill(dentry
);
228 * d_invalidate - invalidate a dentry
229 * @dentry: dentry to invalidate
231 * Try to invalidate the dentry if it turns out to be
232 * possible. If there are other dentries that can be
233 * reached through this one we can't delete it and we
234 * return -EBUSY. On success we return 0.
239 int d_invalidate(struct dentry
* dentry
)
242 * If it's already been dropped, return OK.
244 spin_lock(&dcache_lock
);
245 if (d_unhashed(dentry
)) {
246 spin_unlock(&dcache_lock
);
250 * Check whether to do a partial shrink_dcache
251 * to get rid of unused child entries.
253 if (!list_empty(&dentry
->d_subdirs
)) {
254 spin_unlock(&dcache_lock
);
255 shrink_dcache_parent(dentry
);
256 spin_lock(&dcache_lock
);
260 * Somebody else still using it?
262 * If it's a directory, we can't drop it
263 * for fear of somebody re-populating it
264 * with children (even though dropping it
265 * would make it unreachable from the root,
266 * we might still populate it if it was a
267 * working directory or similar).
269 spin_lock(&dentry
->d_lock
);
270 if (atomic_read(&dentry
->d_count
) > 1) {
271 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
272 spin_unlock(&dentry
->d_lock
);
273 spin_unlock(&dcache_lock
);
279 spin_unlock(&dentry
->d_lock
);
280 spin_unlock(&dcache_lock
);
284 /* This should be called _only_ with dcache_lock held */
286 static inline struct dentry
* __dget_locked(struct dentry
*dentry
)
288 atomic_inc(&dentry
->d_count
);
289 if (!list_empty(&dentry
->d_lru
)) {
290 dentry_stat
.nr_unused
--;
291 list_del_init(&dentry
->d_lru
);
296 struct dentry
* dget_locked(struct dentry
*dentry
)
298 return __dget_locked(dentry
);
302 * d_find_alias - grab a hashed alias of inode
303 * @inode: inode in question
304 * @want_discon: flag, used by d_splice_alias, to request
305 * that only a DISCONNECTED alias be returned.
307 * If inode has a hashed alias, or is a directory and has any alias,
308 * acquire the reference to alias and return it. Otherwise return NULL.
309 * Notice that if inode is a directory there can be only one alias and
310 * it can be unhashed only if it has no children, or if it is the root
313 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
314 * any other hashed alias over that one unless @want_discon is set,
315 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
318 static struct dentry
* __d_find_alias(struct inode
*inode
, int want_discon
)
320 struct list_head
*head
, *next
, *tmp
;
321 struct dentry
*alias
, *discon_alias
=NULL
;
323 head
= &inode
->i_dentry
;
324 next
= inode
->i_dentry
.next
;
325 while (next
!= head
) {
329 alias
= list_entry(tmp
, struct dentry
, d_alias
);
330 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
331 if (IS_ROOT(alias
) &&
332 (alias
->d_flags
& DCACHE_DISCONNECTED
))
333 discon_alias
= alias
;
334 else if (!want_discon
) {
335 __dget_locked(alias
);
341 __dget_locked(discon_alias
);
345 struct dentry
* d_find_alias(struct inode
*inode
)
347 struct dentry
*de
= NULL
;
349 if (!list_empty(&inode
->i_dentry
)) {
350 spin_lock(&dcache_lock
);
351 de
= __d_find_alias(inode
, 0);
352 spin_unlock(&dcache_lock
);
358 * Try to kill dentries associated with this inode.
359 * WARNING: you must own a reference to inode.
361 void d_prune_aliases(struct inode
*inode
)
363 struct dentry
*dentry
;
365 spin_lock(&dcache_lock
);
366 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
367 spin_lock(&dentry
->d_lock
);
368 if (!atomic_read(&dentry
->d_count
)) {
369 __dget_locked(dentry
);
371 spin_unlock(&dentry
->d_lock
);
372 spin_unlock(&dcache_lock
);
376 spin_unlock(&dentry
->d_lock
);
378 spin_unlock(&dcache_lock
);
382 * Throw away a dentry - free the inode, dput the parent. This requires that
383 * the LRU list has already been removed.
385 * If prune_parents is true, try to prune ancestors as well.
387 * Called with dcache_lock, drops it and then regains.
388 * Called with dentry->d_lock held, drops it.
390 static void prune_one_dentry(struct dentry
* dentry
, int prune_parents
)
393 dentry
= d_kill(dentry
);
394 if (!prune_parents
) {
396 spin_lock(&dcache_lock
);
401 * Prune ancestors. Locking is simpler than in dput(),
402 * because dcache_lock needs to be taken anyway.
404 spin_lock(&dcache_lock
);
406 if (!atomic_dec_and_lock(&dentry
->d_count
, &dentry
->d_lock
))
409 if (dentry
->d_op
&& dentry
->d_op
->d_delete
)
410 dentry
->d_op
->d_delete(dentry
);
411 if (!list_empty(&dentry
->d_lru
)) {
412 list_del(&dentry
->d_lru
);
413 dentry_stat
.nr_unused
--;
416 dentry
= d_kill(dentry
);
417 spin_lock(&dcache_lock
);
422 * prune_dcache - shrink the dcache
423 * @count: number of entries to try and free
424 * @sb: if given, ignore dentries for other superblocks
425 * which are being unmounted.
426 * @prune_parents: if true, try to prune ancestors as well in one go
428 * Shrink the dcache. This is done when we need
429 * more memory, or simply when we need to unmount
430 * something (at which point we need to unuse
433 * This function may fail to free any resources if
434 * all the dentries are in use.
437 static void prune_dcache(int count
, struct super_block
*sb
, int prune_parents
)
439 spin_lock(&dcache_lock
);
440 for (; count
; count
--) {
441 struct dentry
*dentry
;
442 struct list_head
*tmp
;
443 struct rw_semaphore
*s_umount
;
445 cond_resched_lock(&dcache_lock
);
447 tmp
= dentry_unused
.prev
;
449 /* Try to find a dentry for this sb, but don't try
450 * too hard, if they aren't near the tail they will
451 * be moved down again soon
454 while (skip
&& tmp
!= &dentry_unused
&&
455 list_entry(tmp
, struct dentry
, d_lru
)->d_sb
!= sb
) {
460 if (tmp
== &dentry_unused
)
463 prefetch(dentry_unused
.prev
);
464 dentry_stat
.nr_unused
--;
465 dentry
= list_entry(tmp
, struct dentry
, d_lru
);
467 spin_lock(&dentry
->d_lock
);
469 * We found an inuse dentry which was not removed from
470 * dentry_unused because of laziness during lookup. Do not free
471 * it - just keep it off the dentry_unused list.
473 if (atomic_read(&dentry
->d_count
)) {
474 spin_unlock(&dentry
->d_lock
);
477 /* If the dentry was recently referenced, don't free it. */
478 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
479 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
480 list_add(&dentry
->d_lru
, &dentry_unused
);
481 dentry_stat
.nr_unused
++;
482 spin_unlock(&dentry
->d_lock
);
486 * If the dentry is not DCACHED_REFERENCED, it is time
487 * to remove it from the dcache, provided the super block is
488 * NULL (which means we are trying to reclaim memory)
489 * or this dentry belongs to the same super block that
493 * If this dentry is for "my" filesystem, then I can prune it
494 * without taking the s_umount lock (I already hold it).
496 if (sb
&& dentry
->d_sb
== sb
) {
497 prune_one_dentry(dentry
, prune_parents
);
501 * ...otherwise we need to be sure this filesystem isn't being
502 * unmounted, otherwise we could race with
503 * generic_shutdown_super(), and end up holding a reference to
504 * an inode while the filesystem is unmounted.
505 * So we try to get s_umount, and make sure s_root isn't NULL.
506 * (Take a local copy of s_umount to avoid a use-after-free of
509 s_umount
= &dentry
->d_sb
->s_umount
;
510 if (down_read_trylock(s_umount
)) {
511 if (dentry
->d_sb
->s_root
!= NULL
) {
512 prune_one_dentry(dentry
, prune_parents
);
518 spin_unlock(&dentry
->d_lock
);
520 * Insert dentry at the head of the list as inserting at the
521 * tail leads to a cycle.
523 list_add(&dentry
->d_lru
, &dentry_unused
);
524 dentry_stat
.nr_unused
++;
526 spin_unlock(&dcache_lock
);
530 * Shrink the dcache for the specified super block.
531 * This allows us to unmount a device without disturbing
532 * the dcache for the other devices.
534 * This implementation makes just two traversals of the
535 * unused list. On the first pass we move the selected
536 * dentries to the most recent end, and on the second
537 * pass we free them. The second pass must restart after
538 * each dput(), but since the target dentries are all at
539 * the end, it's really just a single traversal.
543 * shrink_dcache_sb - shrink dcache for a superblock
546 * Shrink the dcache for the specified super block. This
547 * is used to free the dcache before unmounting a file
551 void shrink_dcache_sb(struct super_block
* sb
)
553 struct list_head
*tmp
, *next
;
554 struct dentry
*dentry
;
557 * Pass one ... move the dentries for the specified
558 * superblock to the most recent end of the unused list.
560 spin_lock(&dcache_lock
);
561 list_for_each_safe(tmp
, next
, &dentry_unused
) {
562 dentry
= list_entry(tmp
, struct dentry
, d_lru
);
563 if (dentry
->d_sb
!= sb
)
565 list_move(tmp
, &dentry_unused
);
569 * Pass two ... free the dentries for this superblock.
572 list_for_each_safe(tmp
, next
, &dentry_unused
) {
573 dentry
= list_entry(tmp
, struct dentry
, d_lru
);
574 if (dentry
->d_sb
!= sb
)
576 dentry_stat
.nr_unused
--;
578 spin_lock(&dentry
->d_lock
);
579 if (atomic_read(&dentry
->d_count
)) {
580 spin_unlock(&dentry
->d_lock
);
583 prune_one_dentry(dentry
, 1);
584 cond_resched_lock(&dcache_lock
);
587 spin_unlock(&dcache_lock
);
591 * destroy a single subtree of dentries for unmount
592 * - see the comments on shrink_dcache_for_umount() for a description of the
595 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
597 struct dentry
*parent
;
598 unsigned detached
= 0;
600 BUG_ON(!IS_ROOT(dentry
));
602 /* detach this root from the system */
603 spin_lock(&dcache_lock
);
604 if (!list_empty(&dentry
->d_lru
)) {
605 dentry_stat
.nr_unused
--;
606 list_del_init(&dentry
->d_lru
);
609 spin_unlock(&dcache_lock
);
612 /* descend to the first leaf in the current subtree */
613 while (!list_empty(&dentry
->d_subdirs
)) {
616 /* this is a branch with children - detach all of them
617 * from the system in one go */
618 spin_lock(&dcache_lock
);
619 list_for_each_entry(loop
, &dentry
->d_subdirs
,
621 if (!list_empty(&loop
->d_lru
)) {
622 dentry_stat
.nr_unused
--;
623 list_del_init(&loop
->d_lru
);
627 cond_resched_lock(&dcache_lock
);
629 spin_unlock(&dcache_lock
);
631 /* move to the first child */
632 dentry
= list_entry(dentry
->d_subdirs
.next
,
633 struct dentry
, d_u
.d_child
);
636 /* consume the dentries from this leaf up through its parents
637 * until we find one with children or run out altogether */
641 if (atomic_read(&dentry
->d_count
) != 0) {
643 "BUG: Dentry %p{i=%lx,n=%s}"
645 " [unmount of %s %s]\n",
648 dentry
->d_inode
->i_ino
: 0UL,
650 atomic_read(&dentry
->d_count
),
651 dentry
->d_sb
->s_type
->name
,
656 parent
= dentry
->d_parent
;
657 if (parent
== dentry
)
660 atomic_dec(&parent
->d_count
);
662 list_del(&dentry
->d_u
.d_child
);
665 inode
= dentry
->d_inode
;
667 dentry
->d_inode
= NULL
;
668 list_del_init(&dentry
->d_alias
);
669 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
670 dentry
->d_op
->d_iput(dentry
, inode
);
677 /* finished when we fall off the top of the tree,
678 * otherwise we ascend to the parent and move to the
679 * next sibling if there is one */
685 } while (list_empty(&dentry
->d_subdirs
));
687 dentry
= list_entry(dentry
->d_subdirs
.next
,
688 struct dentry
, d_u
.d_child
);
691 /* several dentries were freed, need to correct nr_dentry */
692 spin_lock(&dcache_lock
);
693 dentry_stat
.nr_dentry
-= detached
;
694 spin_unlock(&dcache_lock
);
698 * destroy the dentries attached to a superblock on unmounting
699 * - we don't need to use dentry->d_lock, and only need dcache_lock when
700 * removing the dentry from the system lists and hashes because:
701 * - the superblock is detached from all mountings and open files, so the
702 * dentry trees will not be rearranged by the VFS
703 * - s_umount is write-locked, so the memory pressure shrinker will ignore
704 * any dentries belonging to this superblock that it comes across
705 * - the filesystem itself is no longer permitted to rearrange the dentries
708 void shrink_dcache_for_umount(struct super_block
*sb
)
710 struct dentry
*dentry
;
712 if (down_read_trylock(&sb
->s_umount
))
717 atomic_dec(&dentry
->d_count
);
718 shrink_dcache_for_umount_subtree(dentry
);
720 while (!hlist_empty(&sb
->s_anon
)) {
721 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
722 shrink_dcache_for_umount_subtree(dentry
);
727 * Search for at least 1 mount point in the dentry's subdirs.
728 * We descend to the next level whenever the d_subdirs
729 * list is non-empty and continue searching.
733 * have_submounts - check for mounts over a dentry
734 * @parent: dentry to check.
736 * Return true if the parent or its subdirectories contain
740 int have_submounts(struct dentry
*parent
)
742 struct dentry
*this_parent
= parent
;
743 struct list_head
*next
;
745 spin_lock(&dcache_lock
);
746 if (d_mountpoint(parent
))
749 next
= this_parent
->d_subdirs
.next
;
751 while (next
!= &this_parent
->d_subdirs
) {
752 struct list_head
*tmp
= next
;
753 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
755 /* Have we found a mount point ? */
756 if (d_mountpoint(dentry
))
758 if (!list_empty(&dentry
->d_subdirs
)) {
759 this_parent
= dentry
;
764 * All done at this level ... ascend and resume the search.
766 if (this_parent
!= parent
) {
767 next
= this_parent
->d_u
.d_child
.next
;
768 this_parent
= this_parent
->d_parent
;
771 spin_unlock(&dcache_lock
);
772 return 0; /* No mount points found in tree */
774 spin_unlock(&dcache_lock
);
779 * Search the dentry child list for the specified parent,
780 * and move any unused dentries to the end of the unused
781 * list for prune_dcache(). We descend to the next level
782 * whenever the d_subdirs list is non-empty and continue
785 * It returns zero iff there are no unused children,
786 * otherwise it returns the number of children moved to
787 * the end of the unused list. This may not be the total
788 * number of unused children, because select_parent can
789 * drop the lock and return early due to latency
792 static int select_parent(struct dentry
* parent
)
794 struct dentry
*this_parent
= parent
;
795 struct list_head
*next
;
798 spin_lock(&dcache_lock
);
800 next
= this_parent
->d_subdirs
.next
;
802 while (next
!= &this_parent
->d_subdirs
) {
803 struct list_head
*tmp
= next
;
804 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
807 if (!list_empty(&dentry
->d_lru
)) {
808 dentry_stat
.nr_unused
--;
809 list_del_init(&dentry
->d_lru
);
812 * move only zero ref count dentries to the end
813 * of the unused list for prune_dcache
815 if (!atomic_read(&dentry
->d_count
)) {
816 list_add_tail(&dentry
->d_lru
, &dentry_unused
);
817 dentry_stat
.nr_unused
++;
822 * We can return to the caller if we have found some (this
823 * ensures forward progress). We'll be coming back to find
826 if (found
&& need_resched())
830 * Descend a level if the d_subdirs list is non-empty.
832 if (!list_empty(&dentry
->d_subdirs
)) {
833 this_parent
= dentry
;
838 * All done at this level ... ascend and resume the search.
840 if (this_parent
!= parent
) {
841 next
= this_parent
->d_u
.d_child
.next
;
842 this_parent
= this_parent
->d_parent
;
846 spin_unlock(&dcache_lock
);
851 * shrink_dcache_parent - prune dcache
852 * @parent: parent of entries to prune
854 * Prune the dcache to remove unused children of the parent dentry.
857 void shrink_dcache_parent(struct dentry
* parent
)
861 while ((found
= select_parent(parent
)) != 0)
862 prune_dcache(found
, parent
->d_sb
, 1);
866 * Scan `nr' dentries and return the number which remain.
868 * We need to avoid reentering the filesystem if the caller is performing a
869 * GFP_NOFS allocation attempt. One example deadlock is:
871 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
872 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
873 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
875 * In this case we return -1 to tell the caller that we baled.
877 static int shrink_dcache_memory(int nr
, gfp_t gfp_mask
)
880 if (!(gfp_mask
& __GFP_FS
))
882 prune_dcache(nr
, NULL
, 1);
884 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
888 * d_alloc - allocate a dcache entry
889 * @parent: parent of entry to allocate
890 * @name: qstr of the name
892 * Allocates a dentry. It returns %NULL if there is insufficient memory
893 * available. On a success the dentry is returned. The name passed in is
894 * copied and the copy passed in may be reused after this call.
897 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
899 struct dentry
*dentry
;
902 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
906 if (name
->len
> DNAME_INLINE_LEN
-1) {
907 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
909 kmem_cache_free(dentry_cache
, dentry
);
913 dname
= dentry
->d_iname
;
915 dentry
->d_name
.name
= dname
;
917 dentry
->d_name
.len
= name
->len
;
918 dentry
->d_name
.hash
= name
->hash
;
919 memcpy(dname
, name
->name
, name
->len
);
920 dname
[name
->len
] = 0;
922 atomic_set(&dentry
->d_count
, 1);
923 dentry
->d_flags
= DCACHE_UNHASHED
;
924 spin_lock_init(&dentry
->d_lock
);
925 dentry
->d_inode
= NULL
;
926 dentry
->d_parent
= NULL
;
929 dentry
->d_fsdata
= NULL
;
930 dentry
->d_mounted
= 0;
931 #ifdef CONFIG_PROFILING
932 dentry
->d_cookie
= NULL
;
934 INIT_HLIST_NODE(&dentry
->d_hash
);
935 INIT_LIST_HEAD(&dentry
->d_lru
);
936 INIT_LIST_HEAD(&dentry
->d_subdirs
);
937 INIT_LIST_HEAD(&dentry
->d_alias
);
940 dentry
->d_parent
= dget(parent
);
941 dentry
->d_sb
= parent
->d_sb
;
943 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
946 spin_lock(&dcache_lock
);
948 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
949 dentry_stat
.nr_dentry
++;
950 spin_unlock(&dcache_lock
);
955 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
960 q
.len
= strlen(name
);
961 q
.hash
= full_name_hash(q
.name
, q
.len
);
962 return d_alloc(parent
, &q
);
966 * d_instantiate - fill in inode information for a dentry
967 * @entry: dentry to complete
968 * @inode: inode to attach to this dentry
970 * Fill in inode information in the entry.
972 * This turns negative dentries into productive full members
975 * NOTE! This assumes that the inode count has been incremented
976 * (or otherwise set) by the caller to indicate that it is now
977 * in use by the dcache.
980 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
982 BUG_ON(!list_empty(&entry
->d_alias
));
983 spin_lock(&dcache_lock
);
985 list_add(&entry
->d_alias
, &inode
->i_dentry
);
986 entry
->d_inode
= inode
;
987 fsnotify_d_instantiate(entry
, inode
);
988 spin_unlock(&dcache_lock
);
989 security_d_instantiate(entry
, inode
);
993 * d_instantiate_unique - instantiate a non-aliased dentry
994 * @entry: dentry to instantiate
995 * @inode: inode to attach to this dentry
997 * Fill in inode information in the entry. On success, it returns NULL.
998 * If an unhashed alias of "entry" already exists, then we return the
999 * aliased dentry instead and drop one reference to inode.
1001 * Note that in order to avoid conflicts with rename() etc, the caller
1002 * had better be holding the parent directory semaphore.
1004 * This also assumes that the inode count has been incremented
1005 * (or otherwise set) by the caller to indicate that it is now
1006 * in use by the dcache.
1008 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1009 struct inode
*inode
)
1011 struct dentry
*alias
;
1012 int len
= entry
->d_name
.len
;
1013 const char *name
= entry
->d_name
.name
;
1014 unsigned int hash
= entry
->d_name
.hash
;
1017 entry
->d_inode
= NULL
;
1021 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1022 struct qstr
*qstr
= &alias
->d_name
;
1024 if (qstr
->hash
!= hash
)
1026 if (alias
->d_parent
!= entry
->d_parent
)
1028 if (qstr
->len
!= len
)
1030 if (memcmp(qstr
->name
, name
, len
))
1036 list_add(&entry
->d_alias
, &inode
->i_dentry
);
1037 entry
->d_inode
= inode
;
1038 fsnotify_d_instantiate(entry
, inode
);
1042 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1044 struct dentry
*result
;
1046 BUG_ON(!list_empty(&entry
->d_alias
));
1048 spin_lock(&dcache_lock
);
1049 result
= __d_instantiate_unique(entry
, inode
);
1050 spin_unlock(&dcache_lock
);
1053 security_d_instantiate(entry
, inode
);
1057 BUG_ON(!d_unhashed(result
));
1062 EXPORT_SYMBOL(d_instantiate_unique
);
1065 * d_alloc_root - allocate root dentry
1066 * @root_inode: inode to allocate the root for
1068 * Allocate a root ("/") dentry for the inode given. The inode is
1069 * instantiated and returned. %NULL is returned if there is insufficient
1070 * memory or the inode passed is %NULL.
1073 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1075 struct dentry
*res
= NULL
;
1078 static const struct qstr name
= { .name
= "/", .len
= 1 };
1080 res
= d_alloc(NULL
, &name
);
1082 res
->d_sb
= root_inode
->i_sb
;
1083 res
->d_parent
= res
;
1084 d_instantiate(res
, root_inode
);
1090 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1093 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1094 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1095 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1099 * d_alloc_anon - allocate an anonymous dentry
1100 * @inode: inode to allocate the dentry for
1102 * This is similar to d_alloc_root. It is used by filesystems when
1103 * creating a dentry for a given inode, often in the process of
1104 * mapping a filehandle to a dentry. The returned dentry may be
1105 * anonymous, or may have a full name (if the inode was already
1106 * in the cache). The file system may need to make further
1107 * efforts to connect this dentry into the dcache properly.
1109 * When called on a directory inode, we must ensure that
1110 * the inode only ever has one dentry. If a dentry is
1111 * found, that is returned instead of allocating a new one.
1113 * On successful return, the reference to the inode has been transferred
1114 * to the dentry. If %NULL is returned (indicating kmalloc failure),
1115 * the reference on the inode has not been released.
1118 struct dentry
* d_alloc_anon(struct inode
*inode
)
1120 static const struct qstr anonstring
= { .name
= "" };
1124 if ((res
= d_find_alias(inode
))) {
1129 tmp
= d_alloc(NULL
, &anonstring
);
1133 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1135 spin_lock(&dcache_lock
);
1136 res
= __d_find_alias(inode
, 0);
1138 /* attach a disconnected dentry */
1141 spin_lock(&res
->d_lock
);
1142 res
->d_sb
= inode
->i_sb
;
1143 res
->d_parent
= res
;
1144 res
->d_inode
= inode
;
1145 res
->d_flags
|= DCACHE_DISCONNECTED
;
1146 res
->d_flags
&= ~DCACHE_UNHASHED
;
1147 list_add(&res
->d_alias
, &inode
->i_dentry
);
1148 hlist_add_head(&res
->d_hash
, &inode
->i_sb
->s_anon
);
1149 spin_unlock(&res
->d_lock
);
1151 inode
= NULL
; /* don't drop reference */
1153 spin_unlock(&dcache_lock
);
1164 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1165 * @inode: the inode which may have a disconnected dentry
1166 * @dentry: a negative dentry which we want to point to the inode.
1168 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1169 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1170 * and return it, else simply d_add the inode to the dentry and return NULL.
1172 * This is needed in the lookup routine of any filesystem that is exportable
1173 * (via knfsd) so that we can build dcache paths to directories effectively.
1175 * If a dentry was found and moved, then it is returned. Otherwise NULL
1176 * is returned. This matches the expected return value of ->lookup.
1179 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1181 struct dentry
*new = NULL
;
1183 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1184 spin_lock(&dcache_lock
);
1185 new = __d_find_alias(inode
, 1);
1187 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1188 fsnotify_d_instantiate(new, inode
);
1189 spin_unlock(&dcache_lock
);
1190 security_d_instantiate(new, inode
);
1192 d_move(new, dentry
);
1195 /* d_instantiate takes dcache_lock, so we do it by hand */
1196 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1197 dentry
->d_inode
= inode
;
1198 fsnotify_d_instantiate(dentry
, inode
);
1199 spin_unlock(&dcache_lock
);
1200 security_d_instantiate(dentry
, inode
);
1204 d_add(dentry
, inode
);
1210 * d_lookup - search for a dentry
1211 * @parent: parent dentry
1212 * @name: qstr of name we wish to find
1214 * Searches the children of the parent dentry for the name in question. If
1215 * the dentry is found its reference count is incremented and the dentry
1216 * is returned. The caller must use d_put to free the entry when it has
1217 * finished using it. %NULL is returned on failure.
1219 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1220 * Memory barriers are used while updating and doing lockless traversal.
1221 * To avoid races with d_move while rename is happening, d_lock is used.
1223 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1224 * and name pointer in one structure pointed by d_qstr.
1226 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1227 * lookup is going on.
1229 * dentry_unused list is not updated even if lookup finds the required dentry
1230 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1231 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1234 * d_lookup() is protected against the concurrent renames in some unrelated
1235 * directory using the seqlockt_t rename_lock.
1238 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1240 struct dentry
* dentry
= NULL
;
1244 seq
= read_seqbegin(&rename_lock
);
1245 dentry
= __d_lookup(parent
, name
);
1248 } while (read_seqretry(&rename_lock
, seq
));
1252 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1254 unsigned int len
= name
->len
;
1255 unsigned int hash
= name
->hash
;
1256 const unsigned char *str
= name
->name
;
1257 struct hlist_head
*head
= d_hash(parent
,hash
);
1258 struct dentry
*found
= NULL
;
1259 struct hlist_node
*node
;
1260 struct dentry
*dentry
;
1264 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1267 if (dentry
->d_name
.hash
!= hash
)
1269 if (dentry
->d_parent
!= parent
)
1272 spin_lock(&dentry
->d_lock
);
1275 * Recheck the dentry after taking the lock - d_move may have
1276 * changed things. Don't bother checking the hash because we're
1277 * about to compare the whole name anyway.
1279 if (dentry
->d_parent
!= parent
)
1283 * It is safe to compare names since d_move() cannot
1284 * change the qstr (protected by d_lock).
1286 qstr
= &dentry
->d_name
;
1287 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1288 if (parent
->d_op
->d_compare(parent
, qstr
, name
))
1291 if (qstr
->len
!= len
)
1293 if (memcmp(qstr
->name
, str
, len
))
1297 if (!d_unhashed(dentry
)) {
1298 atomic_inc(&dentry
->d_count
);
1301 spin_unlock(&dentry
->d_lock
);
1304 spin_unlock(&dentry
->d_lock
);
1312 * d_hash_and_lookup - hash the qstr then search for a dentry
1313 * @dir: Directory to search in
1314 * @name: qstr of name we wish to find
1316 * On hash failure or on lookup failure NULL is returned.
1318 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1320 struct dentry
*dentry
= NULL
;
1323 * Check for a fs-specific hash function. Note that we must
1324 * calculate the standard hash first, as the d_op->d_hash()
1325 * routine may choose to leave the hash value unchanged.
1327 name
->hash
= full_name_hash(name
->name
, name
->len
);
1328 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1329 if (dir
->d_op
->d_hash(dir
, name
) < 0)
1332 dentry
= d_lookup(dir
, name
);
1338 * d_validate - verify dentry provided from insecure source
1339 * @dentry: The dentry alleged to be valid child of @dparent
1340 * @dparent: The parent dentry (known to be valid)
1341 * @hash: Hash of the dentry
1342 * @len: Length of the name
1344 * An insecure source has sent us a dentry, here we verify it and dget() it.
1345 * This is used by ncpfs in its readdir implementation.
1346 * Zero is returned in the dentry is invalid.
1349 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1351 struct hlist_head
*base
;
1352 struct hlist_node
*lhp
;
1354 /* Check whether the ptr might be valid at all.. */
1355 if (!kmem_ptr_validate(dentry_cache
, dentry
))
1358 if (dentry
->d_parent
!= dparent
)
1361 spin_lock(&dcache_lock
);
1362 base
= d_hash(dparent
, dentry
->d_name
.hash
);
1363 hlist_for_each(lhp
,base
) {
1364 /* hlist_for_each_entry_rcu() not required for d_hash list
1365 * as it is parsed under dcache_lock
1367 if (dentry
== hlist_entry(lhp
, struct dentry
, d_hash
)) {
1368 __dget_locked(dentry
);
1369 spin_unlock(&dcache_lock
);
1373 spin_unlock(&dcache_lock
);
1379 * When a file is deleted, we have two options:
1380 * - turn this dentry into a negative dentry
1381 * - unhash this dentry and free it.
1383 * Usually, we want to just turn this into
1384 * a negative dentry, but if anybody else is
1385 * currently using the dentry or the inode
1386 * we can't do that and we fall back on removing
1387 * it from the hash queues and waiting for
1388 * it to be deleted later when it has no users
1392 * d_delete - delete a dentry
1393 * @dentry: The dentry to delete
1395 * Turn the dentry into a negative dentry if possible, otherwise
1396 * remove it from the hash queues so it can be deleted later
1399 void d_delete(struct dentry
* dentry
)
1403 * Are we the only user?
1405 spin_lock(&dcache_lock
);
1406 spin_lock(&dentry
->d_lock
);
1407 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1408 if (atomic_read(&dentry
->d_count
) == 1) {
1409 dentry_iput(dentry
);
1410 fsnotify_nameremove(dentry
, isdir
);
1412 /* remove this and other inotify debug checks after 2.6.18 */
1413 dentry
->d_flags
&= ~DCACHE_INOTIFY_PARENT_WATCHED
;
1417 if (!d_unhashed(dentry
))
1420 spin_unlock(&dentry
->d_lock
);
1421 spin_unlock(&dcache_lock
);
1423 fsnotify_nameremove(dentry
, isdir
);
1426 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1429 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1430 hlist_add_head_rcu(&entry
->d_hash
, list
);
1433 static void _d_rehash(struct dentry
* entry
)
1435 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1439 * d_rehash - add an entry back to the hash
1440 * @entry: dentry to add to the hash
1442 * Adds a dentry to the hash according to its name.
1445 void d_rehash(struct dentry
* entry
)
1447 spin_lock(&dcache_lock
);
1448 spin_lock(&entry
->d_lock
);
1450 spin_unlock(&entry
->d_lock
);
1451 spin_unlock(&dcache_lock
);
1454 #define do_switch(x,y) do { \
1455 __typeof__ (x) __tmp = x; \
1456 x = y; y = __tmp; } while (0)
1459 * When switching names, the actual string doesn't strictly have to
1460 * be preserved in the target - because we're dropping the target
1461 * anyway. As such, we can just do a simple memcpy() to copy over
1462 * the new name before we switch.
1464 * Note that we have to be a lot more careful about getting the hash
1465 * switched - we have to switch the hash value properly even if it
1466 * then no longer matches the actual (corrupted) string of the target.
1467 * The hash value has to match the hash queue that the dentry is on..
1469 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1471 if (dname_external(target
)) {
1472 if (dname_external(dentry
)) {
1474 * Both external: swap the pointers
1476 do_switch(target
->d_name
.name
, dentry
->d_name
.name
);
1479 * dentry:internal, target:external. Steal target's
1480 * storage and make target internal.
1482 dentry
->d_name
.name
= target
->d_name
.name
;
1483 target
->d_name
.name
= target
->d_iname
;
1486 if (dname_external(dentry
)) {
1488 * dentry:external, target:internal. Give dentry's
1489 * storage to target and make dentry internal
1491 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1492 target
->d_name
.len
+ 1);
1493 target
->d_name
.name
= dentry
->d_name
.name
;
1494 dentry
->d_name
.name
= dentry
->d_iname
;
1497 * Both are internal. Just copy target to dentry
1499 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1500 target
->d_name
.len
+ 1);
1506 * We cannibalize "target" when moving dentry on top of it,
1507 * because it's going to be thrown away anyway. We could be more
1508 * polite about it, though.
1510 * This forceful removal will result in ugly /proc output if
1511 * somebody holds a file open that got deleted due to a rename.
1512 * We could be nicer about the deleted file, and let it show
1513 * up under the name it got deleted rather than the name that
1518 * d_move_locked - move a dentry
1519 * @dentry: entry to move
1520 * @target: new dentry
1522 * Update the dcache to reflect the move of a file name. Negative
1523 * dcache entries should not be moved in this way.
1525 static void d_move_locked(struct dentry
* dentry
, struct dentry
* target
)
1527 struct hlist_head
*list
;
1529 if (!dentry
->d_inode
)
1530 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1532 write_seqlock(&rename_lock
);
1534 * XXXX: do we really need to take target->d_lock?
1536 if (target
< dentry
) {
1537 spin_lock(&target
->d_lock
);
1538 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1540 spin_lock(&dentry
->d_lock
);
1541 spin_lock_nested(&target
->d_lock
, DENTRY_D_LOCK_NESTED
);
1544 /* Move the dentry to the target hash queue, if on different bucket */
1545 if (dentry
->d_flags
& DCACHE_UNHASHED
)
1546 goto already_unhashed
;
1548 hlist_del_rcu(&dentry
->d_hash
);
1551 list
= d_hash(target
->d_parent
, target
->d_name
.hash
);
1552 __d_rehash(dentry
, list
);
1554 /* Unhash the target: dput() will then get rid of it */
1557 list_del(&dentry
->d_u
.d_child
);
1558 list_del(&target
->d_u
.d_child
);
1560 /* Switch the names.. */
1561 switch_names(dentry
, target
);
1562 do_switch(dentry
->d_name
.len
, target
->d_name
.len
);
1563 do_switch(dentry
->d_name
.hash
, target
->d_name
.hash
);
1565 /* ... and switch the parents */
1566 if (IS_ROOT(dentry
)) {
1567 dentry
->d_parent
= target
->d_parent
;
1568 target
->d_parent
= target
;
1569 INIT_LIST_HEAD(&target
->d_u
.d_child
);
1571 do_switch(dentry
->d_parent
, target
->d_parent
);
1573 /* And add them back to the (new) parent lists */
1574 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
1577 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1578 spin_unlock(&target
->d_lock
);
1579 fsnotify_d_move(dentry
);
1580 spin_unlock(&dentry
->d_lock
);
1581 write_sequnlock(&rename_lock
);
1585 * d_move - move a dentry
1586 * @dentry: entry to move
1587 * @target: new dentry
1589 * Update the dcache to reflect the move of a file name. Negative
1590 * dcache entries should not be moved in this way.
1593 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1595 spin_lock(&dcache_lock
);
1596 d_move_locked(dentry
, target
);
1597 spin_unlock(&dcache_lock
);
1601 * Helper that returns 1 if p1 is a parent of p2, else 0
1603 static int d_isparent(struct dentry
*p1
, struct dentry
*p2
)
1607 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1608 if (p
->d_parent
== p1
)
1615 * This helper attempts to cope with remotely renamed directories
1617 * It assumes that the caller is already holding
1618 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1620 * Note: If ever the locking in lock_rename() changes, then please
1621 * remember to update this too...
1623 * On return, dcache_lock will have been unlocked.
1625 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
1627 struct mutex
*m1
= NULL
, *m2
= NULL
;
1630 /* If alias and dentry share a parent, then no extra locks required */
1631 if (alias
->d_parent
== dentry
->d_parent
)
1634 /* Check for loops */
1635 ret
= ERR_PTR(-ELOOP
);
1636 if (d_isparent(alias
, dentry
))
1639 /* See lock_rename() */
1640 ret
= ERR_PTR(-EBUSY
);
1641 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
1643 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
1644 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
1646 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
1648 d_move_locked(alias
, dentry
);
1651 spin_unlock(&dcache_lock
);
1660 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1661 * named dentry in place of the dentry to be replaced.
1663 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
1665 struct dentry
*dparent
, *aparent
;
1667 switch_names(dentry
, anon
);
1668 do_switch(dentry
->d_name
.len
, anon
->d_name
.len
);
1669 do_switch(dentry
->d_name
.hash
, anon
->d_name
.hash
);
1671 dparent
= dentry
->d_parent
;
1672 aparent
= anon
->d_parent
;
1674 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
1675 list_del(&dentry
->d_u
.d_child
);
1676 if (!IS_ROOT(dentry
))
1677 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1679 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1681 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
1682 list_del(&anon
->d_u
.d_child
);
1684 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
1686 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
1688 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
1692 * d_materialise_unique - introduce an inode into the tree
1693 * @dentry: candidate dentry
1694 * @inode: inode to bind to the dentry, to which aliases may be attached
1696 * Introduces an dentry into the tree, substituting an extant disconnected
1697 * root directory alias in its place if there is one
1699 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
1701 struct dentry
*actual
;
1703 BUG_ON(!d_unhashed(dentry
));
1705 spin_lock(&dcache_lock
);
1709 dentry
->d_inode
= NULL
;
1713 if (S_ISDIR(inode
->i_mode
)) {
1714 struct dentry
*alias
;
1716 /* Does an aliased dentry already exist? */
1717 alias
= __d_find_alias(inode
, 0);
1720 /* Is this an anonymous mountpoint that we could splice
1722 if (IS_ROOT(alias
)) {
1723 spin_lock(&alias
->d_lock
);
1724 __d_materialise_dentry(dentry
, alias
);
1728 /* Nope, but we must(!) avoid directory aliasing */
1729 actual
= __d_unalias(dentry
, alias
);
1736 /* Add a unique reference */
1737 actual
= __d_instantiate_unique(dentry
, inode
);
1740 else if (unlikely(!d_unhashed(actual
)))
1741 goto shouldnt_be_hashed
;
1744 spin_lock(&actual
->d_lock
);
1747 spin_unlock(&actual
->d_lock
);
1748 spin_unlock(&dcache_lock
);
1750 if (actual
== dentry
) {
1751 security_d_instantiate(dentry
, inode
);
1759 spin_unlock(&dcache_lock
);
1761 goto shouldnt_be_hashed
;
1765 * d_path - return the path of a dentry
1766 * @dentry: dentry to report
1767 * @vfsmnt: vfsmnt to which the dentry belongs
1768 * @root: root dentry
1769 * @rootmnt: vfsmnt to which the root dentry belongs
1770 * @buffer: buffer to return value in
1771 * @buflen: buffer length
1773 * Convert a dentry into an ASCII path name. If the entry has been deleted
1774 * the string " (deleted)" is appended. Note that this is ambiguous.
1776 * Returns the buffer or an error code if the path was too long.
1778 * "buflen" should be positive. Caller holds the dcache_lock.
1780 static char * __d_path( struct dentry
*dentry
, struct vfsmount
*vfsmnt
,
1781 struct dentry
*root
, struct vfsmount
*rootmnt
,
1782 char *buffer
, int buflen
)
1784 char * end
= buffer
+buflen
;
1790 if (!IS_ROOT(dentry
) && d_unhashed(dentry
)) {
1795 memcpy(end
, " (deleted)", 10);
1805 struct dentry
* parent
;
1807 if (dentry
== root
&& vfsmnt
== rootmnt
)
1809 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
1811 spin_lock(&vfsmount_lock
);
1812 if (vfsmnt
->mnt_parent
== vfsmnt
) {
1813 spin_unlock(&vfsmount_lock
);
1816 dentry
= vfsmnt
->mnt_mountpoint
;
1817 vfsmnt
= vfsmnt
->mnt_parent
;
1818 spin_unlock(&vfsmount_lock
);
1821 parent
= dentry
->d_parent
;
1823 namelen
= dentry
->d_name
.len
;
1824 buflen
-= namelen
+ 1;
1828 memcpy(end
, dentry
->d_name
.name
, namelen
);
1837 namelen
= dentry
->d_name
.len
;
1841 retval
-= namelen
-1; /* hit the slash */
1842 memcpy(retval
, dentry
->d_name
.name
, namelen
);
1845 return ERR_PTR(-ENAMETOOLONG
);
1848 /* write full pathname into buffer and return start of pathname */
1849 char * d_path(struct dentry
*dentry
, struct vfsmount
*vfsmnt
,
1850 char *buf
, int buflen
)
1853 struct vfsmount
*rootmnt
;
1854 struct dentry
*root
;
1856 read_lock(¤t
->fs
->lock
);
1857 rootmnt
= mntget(current
->fs
->rootmnt
);
1858 root
= dget(current
->fs
->root
);
1859 read_unlock(¤t
->fs
->lock
);
1860 spin_lock(&dcache_lock
);
1861 res
= __d_path(dentry
, vfsmnt
, root
, rootmnt
, buf
, buflen
);
1862 spin_unlock(&dcache_lock
);
1869 * NOTE! The user-level library version returns a
1870 * character pointer. The kernel system call just
1871 * returns the length of the buffer filled (which
1872 * includes the ending '\0' character), or a negative
1873 * error value. So libc would do something like
1875 * char *getcwd(char * buf, size_t size)
1879 * retval = sys_getcwd(buf, size);
1886 asmlinkage
long sys_getcwd(char __user
*buf
, unsigned long size
)
1889 struct vfsmount
*pwdmnt
, *rootmnt
;
1890 struct dentry
*pwd
, *root
;
1891 char *page
= (char *) __get_free_page(GFP_USER
);
1896 read_lock(¤t
->fs
->lock
);
1897 pwdmnt
= mntget(current
->fs
->pwdmnt
);
1898 pwd
= dget(current
->fs
->pwd
);
1899 rootmnt
= mntget(current
->fs
->rootmnt
);
1900 root
= dget(current
->fs
->root
);
1901 read_unlock(¤t
->fs
->lock
);
1904 /* Has the current directory has been unlinked? */
1905 spin_lock(&dcache_lock
);
1906 if (pwd
->d_parent
== pwd
|| !d_unhashed(pwd
)) {
1910 cwd
= __d_path(pwd
, pwdmnt
, root
, rootmnt
, page
, PAGE_SIZE
);
1911 spin_unlock(&dcache_lock
);
1913 error
= PTR_ERR(cwd
);
1918 len
= PAGE_SIZE
+ page
- cwd
;
1921 if (copy_to_user(buf
, cwd
, len
))
1925 spin_unlock(&dcache_lock
);
1932 free_page((unsigned long) page
);
1937 * Test whether new_dentry is a subdirectory of old_dentry.
1939 * Trivially implemented using the dcache structure
1943 * is_subdir - is new dentry a subdirectory of old_dentry
1944 * @new_dentry: new dentry
1945 * @old_dentry: old dentry
1947 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1948 * Returns 0 otherwise.
1949 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1952 int is_subdir(struct dentry
* new_dentry
, struct dentry
* old_dentry
)
1955 struct dentry
* saved
= new_dentry
;
1958 /* need rcu_readlock to protect against the d_parent trashing due to
1963 /* for restarting inner loop in case of seq retry */
1966 seq
= read_seqbegin(&rename_lock
);
1968 if (new_dentry
!= old_dentry
) {
1969 struct dentry
* parent
= new_dentry
->d_parent
;
1970 if (parent
== new_dentry
)
1972 new_dentry
= parent
;
1978 } while (read_seqretry(&rename_lock
, seq
));
1984 void d_genocide(struct dentry
*root
)
1986 struct dentry
*this_parent
= root
;
1987 struct list_head
*next
;
1989 spin_lock(&dcache_lock
);
1991 next
= this_parent
->d_subdirs
.next
;
1993 while (next
!= &this_parent
->d_subdirs
) {
1994 struct list_head
*tmp
= next
;
1995 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1997 if (d_unhashed(dentry
)||!dentry
->d_inode
)
1999 if (!list_empty(&dentry
->d_subdirs
)) {
2000 this_parent
= dentry
;
2003 atomic_dec(&dentry
->d_count
);
2005 if (this_parent
!= root
) {
2006 next
= this_parent
->d_u
.d_child
.next
;
2007 atomic_dec(&this_parent
->d_count
);
2008 this_parent
= this_parent
->d_parent
;
2011 spin_unlock(&dcache_lock
);
2015 * find_inode_number - check for dentry with name
2016 * @dir: directory to check
2017 * @name: Name to find.
2019 * Check whether a dentry already exists for the given name,
2020 * and return the inode number if it has an inode. Otherwise
2023 * This routine is used to post-process directory listings for
2024 * filesystems using synthetic inode numbers, and is necessary
2025 * to keep getcwd() working.
2028 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2030 struct dentry
* dentry
;
2033 dentry
= d_hash_and_lookup(dir
, name
);
2035 if (dentry
->d_inode
)
2036 ino
= dentry
->d_inode
->i_ino
;
2042 static __initdata
unsigned long dhash_entries
;
2043 static int __init
set_dhash_entries(char *str
)
2047 dhash_entries
= simple_strtoul(str
, &str
, 0);
2050 __setup("dhash_entries=", set_dhash_entries
);
2052 static void __init
dcache_init_early(void)
2056 /* If hashes are distributed across NUMA nodes, defer
2057 * hash allocation until vmalloc space is available.
2063 alloc_large_system_hash("Dentry cache",
2064 sizeof(struct hlist_head
),
2072 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2073 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2076 static void __init
dcache_init(unsigned long mempages
)
2081 * A constructor could be added for stable state like the lists,
2082 * but it is probably not worth it because of the cache nature
2085 dentry_cache
= KMEM_CACHE(dentry
,
2086 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2088 set_shrinker(DEFAULT_SEEKS
, shrink_dcache_memory
);
2090 /* Hash may have been set up in dcache_init_early */
2095 alloc_large_system_hash("Dentry cache",
2096 sizeof(struct hlist_head
),
2104 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2105 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2108 /* SLAB cache for __getname() consumers */
2109 struct kmem_cache
*names_cachep __read_mostly
;
2111 /* SLAB cache for file structures */
2112 struct kmem_cache
*filp_cachep __read_mostly
;
2114 EXPORT_SYMBOL(d_genocide
);
2116 void __init
vfs_caches_init_early(void)
2118 dcache_init_early();
2122 void __init
vfs_caches_init(unsigned long mempages
)
2124 unsigned long reserve
;
2126 /* Base hash sizes on available memory, with a reserve equal to
2127 150% of current kernel size */
2129 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2130 mempages
-= reserve
;
2132 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2133 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
2135 filp_cachep
= kmem_cache_create("filp", sizeof(struct file
), 0,
2136 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
2138 dcache_init(mempages
);
2139 inode_init(mempages
);
2140 files_init(mempages
);
2146 EXPORT_SYMBOL(d_alloc
);
2147 EXPORT_SYMBOL(d_alloc_anon
);
2148 EXPORT_SYMBOL(d_alloc_root
);
2149 EXPORT_SYMBOL(d_delete
);
2150 EXPORT_SYMBOL(d_find_alias
);
2151 EXPORT_SYMBOL(d_instantiate
);
2152 EXPORT_SYMBOL(d_invalidate
);
2153 EXPORT_SYMBOL(d_lookup
);
2154 EXPORT_SYMBOL(d_move
);
2155 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2156 EXPORT_SYMBOL(d_path
);
2157 EXPORT_SYMBOL(d_prune_aliases
);
2158 EXPORT_SYMBOL(d_rehash
);
2159 EXPORT_SYMBOL(d_splice_alias
);
2160 EXPORT_SYMBOL(d_validate
);
2161 EXPORT_SYMBOL(dget_locked
);
2162 EXPORT_SYMBOL(dput
);
2163 EXPORT_SYMBOL(find_inode_number
);
2164 EXPORT_SYMBOL(have_submounts
);
2165 EXPORT_SYMBOL(names_cachep
);
2166 EXPORT_SYMBOL(shrink_dcache_parent
);
2167 EXPORT_SYMBOL(shrink_dcache_sb
);