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 kmem_cache_t
*dentry_cache __read_mostly
;
48 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
58 #define D_HASHBITS d_hash_shift
59 #define D_HASHMASK d_hash_mask
61 static unsigned int d_hash_mask __read_mostly
;
62 static unsigned int d_hash_shift __read_mostly
;
63 static struct hlist_head
*dentry_hashtable __read_mostly
;
64 static LIST_HEAD(dentry_unused
);
66 /* Statistics gathering. */
67 struct dentry_stat_t dentry_stat
= {
71 static void d_callback(struct rcu_head
*head
)
73 struct dentry
* dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
75 if (dname_external(dentry
))
76 kfree(dentry
->d_name
.name
);
77 kmem_cache_free(dentry_cache
, dentry
);
81 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
84 static void d_free(struct dentry
*dentry
)
86 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
87 dentry
->d_op
->d_release(dentry
);
88 call_rcu(&dentry
->d_u
.d_rcu
, d_callback
);
92 * Release the dentry's inode, using the filesystem
93 * d_iput() operation if defined.
94 * Called with dcache_lock and per dentry lock held, drops both.
96 static void dentry_iput(struct dentry
* dentry
)
98 struct inode
*inode
= dentry
->d_inode
;
100 dentry
->d_inode
= NULL
;
101 list_del_init(&dentry
->d_alias
);
102 spin_unlock(&dentry
->d_lock
);
103 spin_unlock(&dcache_lock
);
105 fsnotify_inoderemove(inode
);
106 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
107 dentry
->d_op
->d_iput(dentry
, inode
);
111 spin_unlock(&dentry
->d_lock
);
112 spin_unlock(&dcache_lock
);
119 * This is complicated by the fact that we do not want to put
120 * dentries that are no longer on any hash chain on the unused
121 * list: we'd much rather just get rid of them immediately.
123 * However, that implies that we have to traverse the dentry
124 * tree upwards to the parents which might _also_ now be
125 * scheduled for deletion (it may have been only waiting for
126 * its last child to go away).
128 * This tail recursion is done by hand as we don't want to depend
129 * on the compiler to always get this right (gcc generally doesn't).
130 * Real recursion would eat up our stack space.
134 * dput - release a dentry
135 * @dentry: dentry to release
137 * Release a dentry. This will drop the usage count and if appropriate
138 * call the dentry unlink method as well as removing it from the queues and
139 * releasing its resources. If the parent dentries were scheduled for release
140 * they too may now get deleted.
142 * no dcache lock, please.
145 void dput(struct dentry
*dentry
)
151 if (atomic_read(&dentry
->d_count
) == 1)
153 if (!atomic_dec_and_lock(&dentry
->d_count
, &dcache_lock
))
156 spin_lock(&dentry
->d_lock
);
157 if (atomic_read(&dentry
->d_count
)) {
158 spin_unlock(&dentry
->d_lock
);
159 spin_unlock(&dcache_lock
);
164 * AV: ->d_delete() is _NOT_ allowed to block now.
166 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
167 if (dentry
->d_op
->d_delete(dentry
))
170 /* Unreachable? Get rid of it */
171 if (d_unhashed(dentry
))
173 if (list_empty(&dentry
->d_lru
)) {
174 dentry
->d_flags
|= DCACHE_REFERENCED
;
175 list_add(&dentry
->d_lru
, &dentry_unused
);
176 dentry_stat
.nr_unused
++;
178 spin_unlock(&dentry
->d_lock
);
179 spin_unlock(&dcache_lock
);
186 struct dentry
*parent
;
188 /* If dentry was on d_lru list
189 * delete it from there
191 if (!list_empty(&dentry
->d_lru
)) {
192 list_del(&dentry
->d_lru
);
193 dentry_stat
.nr_unused
--;
195 list_del(&dentry
->d_u
.d_child
);
196 dentry_stat
.nr_dentry
--; /* For d_free, below */
197 /*drops the locks, at that point nobody can reach this dentry */
199 parent
= dentry
->d_parent
;
201 if (dentry
== parent
)
209 * d_invalidate - invalidate a dentry
210 * @dentry: dentry to invalidate
212 * Try to invalidate the dentry if it turns out to be
213 * possible. If there are other dentries that can be
214 * reached through this one we can't delete it and we
215 * return -EBUSY. On success we return 0.
220 int d_invalidate(struct dentry
* dentry
)
223 * If it's already been dropped, return OK.
225 spin_lock(&dcache_lock
);
226 if (d_unhashed(dentry
)) {
227 spin_unlock(&dcache_lock
);
231 * Check whether to do a partial shrink_dcache
232 * to get rid of unused child entries.
234 if (!list_empty(&dentry
->d_subdirs
)) {
235 spin_unlock(&dcache_lock
);
236 shrink_dcache_parent(dentry
);
237 spin_lock(&dcache_lock
);
241 * Somebody else still using it?
243 * If it's a directory, we can't drop it
244 * for fear of somebody re-populating it
245 * with children (even though dropping it
246 * would make it unreachable from the root,
247 * we might still populate it if it was a
248 * working directory or similar).
250 spin_lock(&dentry
->d_lock
);
251 if (atomic_read(&dentry
->d_count
) > 1) {
252 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
253 spin_unlock(&dentry
->d_lock
);
254 spin_unlock(&dcache_lock
);
260 spin_unlock(&dentry
->d_lock
);
261 spin_unlock(&dcache_lock
);
265 /* This should be called _only_ with dcache_lock held */
267 static inline struct dentry
* __dget_locked(struct dentry
*dentry
)
269 atomic_inc(&dentry
->d_count
);
270 if (!list_empty(&dentry
->d_lru
)) {
271 dentry_stat
.nr_unused
--;
272 list_del_init(&dentry
->d_lru
);
277 struct dentry
* dget_locked(struct dentry
*dentry
)
279 return __dget_locked(dentry
);
283 * d_find_alias - grab a hashed alias of inode
284 * @inode: inode in question
285 * @want_discon: flag, used by d_splice_alias, to request
286 * that only a DISCONNECTED alias be returned.
288 * If inode has a hashed alias, or is a directory and has any alias,
289 * acquire the reference to alias and return it. Otherwise return NULL.
290 * Notice that if inode is a directory there can be only one alias and
291 * it can be unhashed only if it has no children, or if it is the root
294 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
295 * any other hashed alias over that one unless @want_discon is set,
296 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
299 static struct dentry
* __d_find_alias(struct inode
*inode
, int want_discon
)
301 struct list_head
*head
, *next
, *tmp
;
302 struct dentry
*alias
, *discon_alias
=NULL
;
304 head
= &inode
->i_dentry
;
305 next
= inode
->i_dentry
.next
;
306 while (next
!= head
) {
310 alias
= list_entry(tmp
, struct dentry
, d_alias
);
311 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
312 if (IS_ROOT(alias
) &&
313 (alias
->d_flags
& DCACHE_DISCONNECTED
))
314 discon_alias
= alias
;
315 else if (!want_discon
) {
316 __dget_locked(alias
);
322 __dget_locked(discon_alias
);
326 struct dentry
* d_find_alias(struct inode
*inode
)
328 struct dentry
*de
= NULL
;
330 if (!list_empty(&inode
->i_dentry
)) {
331 spin_lock(&dcache_lock
);
332 de
= __d_find_alias(inode
, 0);
333 spin_unlock(&dcache_lock
);
339 * Try to kill dentries associated with this inode.
340 * WARNING: you must own a reference to inode.
342 void d_prune_aliases(struct inode
*inode
)
344 struct dentry
*dentry
;
346 spin_lock(&dcache_lock
);
347 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
348 spin_lock(&dentry
->d_lock
);
349 if (!atomic_read(&dentry
->d_count
)) {
350 __dget_locked(dentry
);
352 spin_unlock(&dentry
->d_lock
);
353 spin_unlock(&dcache_lock
);
357 spin_unlock(&dentry
->d_lock
);
359 spin_unlock(&dcache_lock
);
363 * Throw away a dentry - free the inode, dput the parent. This requires that
364 * the LRU list has already been removed.
366 * Called with dcache_lock, drops it and then regains.
367 * Called with dentry->d_lock held, drops it.
369 static void prune_one_dentry(struct dentry
* dentry
)
371 struct dentry
* parent
;
374 list_del(&dentry
->d_u
.d_child
);
375 dentry_stat
.nr_dentry
--; /* For d_free, below */
377 parent
= dentry
->d_parent
;
379 if (parent
!= dentry
)
381 spin_lock(&dcache_lock
);
385 * prune_dcache - shrink the dcache
386 * @count: number of entries to try and free
387 * @sb: if given, ignore dentries for other superblocks
388 * which are being unmounted.
390 * Shrink the dcache. This is done when we need
391 * more memory, or simply when we need to unmount
392 * something (at which point we need to unuse
395 * This function may fail to free any resources if
396 * all the dentries are in use.
399 static void prune_dcache(int count
, struct super_block
*sb
)
401 spin_lock(&dcache_lock
);
402 for (; count
; count
--) {
403 struct dentry
*dentry
;
404 struct list_head
*tmp
;
405 struct rw_semaphore
*s_umount
;
407 cond_resched_lock(&dcache_lock
);
409 tmp
= dentry_unused
.prev
;
411 /* Try to find a dentry for this sb, but don't try
412 * too hard, if they aren't near the tail they will
413 * be moved down again soon
416 while (skip
&& tmp
!= &dentry_unused
&&
417 list_entry(tmp
, struct dentry
, d_lru
)->d_sb
!= sb
) {
422 if (tmp
== &dentry_unused
)
425 prefetch(dentry_unused
.prev
);
426 dentry_stat
.nr_unused
--;
427 dentry
= list_entry(tmp
, struct dentry
, d_lru
);
429 spin_lock(&dentry
->d_lock
);
431 * We found an inuse dentry which was not removed from
432 * dentry_unused because of laziness during lookup. Do not free
433 * it - just keep it off the dentry_unused list.
435 if (atomic_read(&dentry
->d_count
)) {
436 spin_unlock(&dentry
->d_lock
);
439 /* If the dentry was recently referenced, don't free it. */
440 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
441 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
442 list_add(&dentry
->d_lru
, &dentry_unused
);
443 dentry_stat
.nr_unused
++;
444 spin_unlock(&dentry
->d_lock
);
448 * If the dentry is not DCACHED_REFERENCED, it is time
449 * to remove it from the dcache, provided the super block is
450 * NULL (which means we are trying to reclaim memory)
451 * or this dentry belongs to the same super block that
455 * If this dentry is for "my" filesystem, then I can prune it
456 * without taking the s_umount lock (I already hold it).
458 if (sb
&& dentry
->d_sb
== sb
) {
459 prune_one_dentry(dentry
);
463 * ...otherwise we need to be sure this filesystem isn't being
464 * unmounted, otherwise we could race with
465 * generic_shutdown_super(), and end up holding a reference to
466 * an inode while the filesystem is unmounted.
467 * So we try to get s_umount, and make sure s_root isn't NULL.
468 * (Take a local copy of s_umount to avoid a use-after-free of
471 s_umount
= &dentry
->d_sb
->s_umount
;
472 if (down_read_trylock(s_umount
)) {
473 if (dentry
->d_sb
->s_root
!= NULL
) {
474 prune_one_dentry(dentry
);
480 spin_unlock(&dentry
->d_lock
);
482 * Insert dentry at the head of the list as inserting at the
483 * tail leads to a cycle.
485 list_add(&dentry
->d_lru
, &dentry_unused
);
486 dentry_stat
.nr_unused
++;
488 spin_unlock(&dcache_lock
);
492 * Shrink the dcache for the specified super block.
493 * This allows us to unmount a device without disturbing
494 * the dcache for the other devices.
496 * This implementation makes just two traversals of the
497 * unused list. On the first pass we move the selected
498 * dentries to the most recent end, and on the second
499 * pass we free them. The second pass must restart after
500 * each dput(), but since the target dentries are all at
501 * the end, it's really just a single traversal.
505 * shrink_dcache_sb - shrink dcache for a superblock
508 * Shrink the dcache for the specified super block. This
509 * is used to free the dcache before unmounting a file
513 void shrink_dcache_sb(struct super_block
* sb
)
515 struct list_head
*tmp
, *next
;
516 struct dentry
*dentry
;
519 * Pass one ... move the dentries for the specified
520 * superblock to the most recent end of the unused list.
522 spin_lock(&dcache_lock
);
523 list_for_each_safe(tmp
, next
, &dentry_unused
) {
524 dentry
= list_entry(tmp
, struct dentry
, d_lru
);
525 if (dentry
->d_sb
!= sb
)
527 list_move(tmp
, &dentry_unused
);
531 * Pass two ... free the dentries for this superblock.
534 list_for_each_safe(tmp
, next
, &dentry_unused
) {
535 dentry
= list_entry(tmp
, struct dentry
, d_lru
);
536 if (dentry
->d_sb
!= sb
)
538 dentry_stat
.nr_unused
--;
540 spin_lock(&dentry
->d_lock
);
541 if (atomic_read(&dentry
->d_count
)) {
542 spin_unlock(&dentry
->d_lock
);
545 prune_one_dentry(dentry
);
546 cond_resched_lock(&dcache_lock
);
549 spin_unlock(&dcache_lock
);
553 * destroy a single subtree of dentries for unmount
554 * - see the comments on shrink_dcache_for_umount() for a description of the
557 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
559 struct dentry
*parent
;
561 BUG_ON(!IS_ROOT(dentry
));
563 /* detach this root from the system */
564 spin_lock(&dcache_lock
);
565 if (!list_empty(&dentry
->d_lru
)) {
566 dentry_stat
.nr_unused
--;
567 list_del_init(&dentry
->d_lru
);
570 spin_unlock(&dcache_lock
);
573 /* descend to the first leaf in the current subtree */
574 while (!list_empty(&dentry
->d_subdirs
)) {
577 /* this is a branch with children - detach all of them
578 * from the system in one go */
579 spin_lock(&dcache_lock
);
580 list_for_each_entry(loop
, &dentry
->d_subdirs
,
582 if (!list_empty(&loop
->d_lru
)) {
583 dentry_stat
.nr_unused
--;
584 list_del_init(&loop
->d_lru
);
588 cond_resched_lock(&dcache_lock
);
590 spin_unlock(&dcache_lock
);
592 /* move to the first child */
593 dentry
= list_entry(dentry
->d_subdirs
.next
,
594 struct dentry
, d_u
.d_child
);
597 /* consume the dentries from this leaf up through its parents
598 * until we find one with children or run out altogether */
602 if (atomic_read(&dentry
->d_count
) != 0) {
604 "BUG: Dentry %p{i=%lx,n=%s}"
606 " [unmount of %s %s]\n",
609 dentry
->d_inode
->i_ino
: 0UL,
611 atomic_read(&dentry
->d_count
),
612 dentry
->d_sb
->s_type
->name
,
617 parent
= dentry
->d_parent
;
618 if (parent
== dentry
)
621 atomic_dec(&parent
->d_count
);
623 list_del(&dentry
->d_u
.d_child
);
624 dentry_stat
.nr_dentry
--; /* For d_free, below */
626 inode
= dentry
->d_inode
;
628 dentry
->d_inode
= NULL
;
629 list_del_init(&dentry
->d_alias
);
630 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
631 dentry
->d_op
->d_iput(dentry
, inode
);
638 /* finished when we fall off the top of the tree,
639 * otherwise we ascend to the parent and move to the
640 * next sibling if there is one */
646 } while (list_empty(&dentry
->d_subdirs
));
648 dentry
= list_entry(dentry
->d_subdirs
.next
,
649 struct dentry
, d_u
.d_child
);
654 * destroy the dentries attached to a superblock on unmounting
655 * - we don't need to use dentry->d_lock, and only need dcache_lock when
656 * removing the dentry from the system lists and hashes because:
657 * - the superblock is detached from all mountings and open files, so the
658 * dentry trees will not be rearranged by the VFS
659 * - s_umount is write-locked, so the memory pressure shrinker will ignore
660 * any dentries belonging to this superblock that it comes across
661 * - the filesystem itself is no longer permitted to rearrange the dentries
664 void shrink_dcache_for_umount(struct super_block
*sb
)
666 struct dentry
*dentry
;
668 if (down_read_trylock(&sb
->s_umount
))
673 atomic_dec(&dentry
->d_count
);
674 shrink_dcache_for_umount_subtree(dentry
);
676 while (!hlist_empty(&sb
->s_anon
)) {
677 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
678 shrink_dcache_for_umount_subtree(dentry
);
683 * Search for at least 1 mount point in the dentry's subdirs.
684 * We descend to the next level whenever the d_subdirs
685 * list is non-empty and continue searching.
689 * have_submounts - check for mounts over a dentry
690 * @parent: dentry to check.
692 * Return true if the parent or its subdirectories contain
696 int have_submounts(struct dentry
*parent
)
698 struct dentry
*this_parent
= parent
;
699 struct list_head
*next
;
701 spin_lock(&dcache_lock
);
702 if (d_mountpoint(parent
))
705 next
= this_parent
->d_subdirs
.next
;
707 while (next
!= &this_parent
->d_subdirs
) {
708 struct list_head
*tmp
= next
;
709 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
711 /* Have we found a mount point ? */
712 if (d_mountpoint(dentry
))
714 if (!list_empty(&dentry
->d_subdirs
)) {
715 this_parent
= dentry
;
720 * All done at this level ... ascend and resume the search.
722 if (this_parent
!= parent
) {
723 next
= this_parent
->d_u
.d_child
.next
;
724 this_parent
= this_parent
->d_parent
;
727 spin_unlock(&dcache_lock
);
728 return 0; /* No mount points found in tree */
730 spin_unlock(&dcache_lock
);
735 * Search the dentry child list for the specified parent,
736 * and move any unused dentries to the end of the unused
737 * list for prune_dcache(). We descend to the next level
738 * whenever the d_subdirs list is non-empty and continue
741 * It returns zero iff there are no unused children,
742 * otherwise it returns the number of children moved to
743 * the end of the unused list. This may not be the total
744 * number of unused children, because select_parent can
745 * drop the lock and return early due to latency
748 static int select_parent(struct dentry
* parent
)
750 struct dentry
*this_parent
= parent
;
751 struct list_head
*next
;
754 spin_lock(&dcache_lock
);
756 next
= this_parent
->d_subdirs
.next
;
758 while (next
!= &this_parent
->d_subdirs
) {
759 struct list_head
*tmp
= next
;
760 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
763 if (!list_empty(&dentry
->d_lru
)) {
764 dentry_stat
.nr_unused
--;
765 list_del_init(&dentry
->d_lru
);
768 * move only zero ref count dentries to the end
769 * of the unused list for prune_dcache
771 if (!atomic_read(&dentry
->d_count
)) {
772 list_add_tail(&dentry
->d_lru
, &dentry_unused
);
773 dentry_stat
.nr_unused
++;
778 * We can return to the caller if we have found some (this
779 * ensures forward progress). We'll be coming back to find
782 if (found
&& need_resched())
786 * Descend a level if the d_subdirs list is non-empty.
788 if (!list_empty(&dentry
->d_subdirs
)) {
789 this_parent
= dentry
;
794 * All done at this level ... ascend and resume the search.
796 if (this_parent
!= parent
) {
797 next
= this_parent
->d_u
.d_child
.next
;
798 this_parent
= this_parent
->d_parent
;
802 spin_unlock(&dcache_lock
);
807 * shrink_dcache_parent - prune dcache
808 * @parent: parent of entries to prune
810 * Prune the dcache to remove unused children of the parent dentry.
813 void shrink_dcache_parent(struct dentry
* parent
)
817 while ((found
= select_parent(parent
)) != 0)
818 prune_dcache(found
, parent
->d_sb
);
822 * Scan `nr' dentries and return the number which remain.
824 * We need to avoid reentering the filesystem if the caller is performing a
825 * GFP_NOFS allocation attempt. One example deadlock is:
827 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
828 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
829 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
831 * In this case we return -1 to tell the caller that we baled.
833 static int shrink_dcache_memory(int nr
, gfp_t gfp_mask
)
836 if (!(gfp_mask
& __GFP_FS
))
838 prune_dcache(nr
, NULL
);
840 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
844 * d_alloc - allocate a dcache entry
845 * @parent: parent of entry to allocate
846 * @name: qstr of the name
848 * Allocates a dentry. It returns %NULL if there is insufficient memory
849 * available. On a success the dentry is returned. The name passed in is
850 * copied and the copy passed in may be reused after this call.
853 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
855 struct dentry
*dentry
;
858 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
862 if (name
->len
> DNAME_INLINE_LEN
-1) {
863 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
865 kmem_cache_free(dentry_cache
, dentry
);
869 dname
= dentry
->d_iname
;
871 dentry
->d_name
.name
= dname
;
873 dentry
->d_name
.len
= name
->len
;
874 dentry
->d_name
.hash
= name
->hash
;
875 memcpy(dname
, name
->name
, name
->len
);
876 dname
[name
->len
] = 0;
878 atomic_set(&dentry
->d_count
, 1);
879 dentry
->d_flags
= DCACHE_UNHASHED
;
880 spin_lock_init(&dentry
->d_lock
);
881 dentry
->d_inode
= NULL
;
882 dentry
->d_parent
= NULL
;
885 dentry
->d_fsdata
= NULL
;
886 dentry
->d_mounted
= 0;
887 #ifdef CONFIG_PROFILING
888 dentry
->d_cookie
= NULL
;
890 INIT_HLIST_NODE(&dentry
->d_hash
);
891 INIT_LIST_HEAD(&dentry
->d_lru
);
892 INIT_LIST_HEAD(&dentry
->d_subdirs
);
893 INIT_LIST_HEAD(&dentry
->d_alias
);
896 dentry
->d_parent
= dget(parent
);
897 dentry
->d_sb
= parent
->d_sb
;
899 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
902 spin_lock(&dcache_lock
);
904 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
905 dentry_stat
.nr_dentry
++;
906 spin_unlock(&dcache_lock
);
911 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
916 q
.len
= strlen(name
);
917 q
.hash
= full_name_hash(q
.name
, q
.len
);
918 return d_alloc(parent
, &q
);
922 * d_instantiate - fill in inode information for a dentry
923 * @entry: dentry to complete
924 * @inode: inode to attach to this dentry
926 * Fill in inode information in the entry.
928 * This turns negative dentries into productive full members
931 * NOTE! This assumes that the inode count has been incremented
932 * (or otherwise set) by the caller to indicate that it is now
933 * in use by the dcache.
936 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
938 BUG_ON(!list_empty(&entry
->d_alias
));
939 spin_lock(&dcache_lock
);
941 list_add(&entry
->d_alias
, &inode
->i_dentry
);
942 entry
->d_inode
= inode
;
943 fsnotify_d_instantiate(entry
, inode
);
944 spin_unlock(&dcache_lock
);
945 security_d_instantiate(entry
, inode
);
949 * d_instantiate_unique - instantiate a non-aliased dentry
950 * @entry: dentry to instantiate
951 * @inode: inode to attach to this dentry
953 * Fill in inode information in the entry. On success, it returns NULL.
954 * If an unhashed alias of "entry" already exists, then we return the
955 * aliased dentry instead and drop one reference to inode.
957 * Note that in order to avoid conflicts with rename() etc, the caller
958 * had better be holding the parent directory semaphore.
960 * This also assumes that the inode count has been incremented
961 * (or otherwise set) by the caller to indicate that it is now
962 * in use by the dcache.
964 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
967 struct dentry
*alias
;
968 int len
= entry
->d_name
.len
;
969 const char *name
= entry
->d_name
.name
;
970 unsigned int hash
= entry
->d_name
.hash
;
973 entry
->d_inode
= NULL
;
977 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
978 struct qstr
*qstr
= &alias
->d_name
;
980 if (qstr
->hash
!= hash
)
982 if (alias
->d_parent
!= entry
->d_parent
)
984 if (qstr
->len
!= len
)
986 if (memcmp(qstr
->name
, name
, len
))
992 list_add(&entry
->d_alias
, &inode
->i_dentry
);
993 entry
->d_inode
= inode
;
994 fsnotify_d_instantiate(entry
, inode
);
998 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1000 struct dentry
*result
;
1002 BUG_ON(!list_empty(&entry
->d_alias
));
1004 spin_lock(&dcache_lock
);
1005 result
= __d_instantiate_unique(entry
, inode
);
1006 spin_unlock(&dcache_lock
);
1009 security_d_instantiate(entry
, inode
);
1013 BUG_ON(!d_unhashed(result
));
1018 EXPORT_SYMBOL(d_instantiate_unique
);
1021 * d_alloc_root - allocate root dentry
1022 * @root_inode: inode to allocate the root for
1024 * Allocate a root ("/") dentry for the inode given. The inode is
1025 * instantiated and returned. %NULL is returned if there is insufficient
1026 * memory or the inode passed is %NULL.
1029 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1031 struct dentry
*res
= NULL
;
1034 static const struct qstr name
= { .name
= "/", .len
= 1 };
1036 res
= d_alloc(NULL
, &name
);
1038 res
->d_sb
= root_inode
->i_sb
;
1039 res
->d_parent
= res
;
1040 d_instantiate(res
, root_inode
);
1046 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1049 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1050 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1051 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1055 * d_alloc_anon - allocate an anonymous dentry
1056 * @inode: inode to allocate the dentry for
1058 * This is similar to d_alloc_root. It is used by filesystems when
1059 * creating a dentry for a given inode, often in the process of
1060 * mapping a filehandle to a dentry. The returned dentry may be
1061 * anonymous, or may have a full name (if the inode was already
1062 * in the cache). The file system may need to make further
1063 * efforts to connect this dentry into the dcache properly.
1065 * When called on a directory inode, we must ensure that
1066 * the inode only ever has one dentry. If a dentry is
1067 * found, that is returned instead of allocating a new one.
1069 * On successful return, the reference to the inode has been transferred
1070 * to the dentry. If %NULL is returned (indicating kmalloc failure),
1071 * the reference on the inode has not been released.
1074 struct dentry
* d_alloc_anon(struct inode
*inode
)
1076 static const struct qstr anonstring
= { .name
= "" };
1080 if ((res
= d_find_alias(inode
))) {
1085 tmp
= d_alloc(NULL
, &anonstring
);
1089 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1091 spin_lock(&dcache_lock
);
1092 res
= __d_find_alias(inode
, 0);
1094 /* attach a disconnected dentry */
1097 spin_lock(&res
->d_lock
);
1098 res
->d_sb
= inode
->i_sb
;
1099 res
->d_parent
= res
;
1100 res
->d_inode
= inode
;
1101 res
->d_flags
|= DCACHE_DISCONNECTED
;
1102 res
->d_flags
&= ~DCACHE_UNHASHED
;
1103 list_add(&res
->d_alias
, &inode
->i_dentry
);
1104 hlist_add_head(&res
->d_hash
, &inode
->i_sb
->s_anon
);
1105 spin_unlock(&res
->d_lock
);
1107 inode
= NULL
; /* don't drop reference */
1109 spin_unlock(&dcache_lock
);
1120 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1121 * @inode: the inode which may have a disconnected dentry
1122 * @dentry: a negative dentry which we want to point to the inode.
1124 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1125 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1126 * and return it, else simply d_add the inode to the dentry and return NULL.
1128 * This is needed in the lookup routine of any filesystem that is exportable
1129 * (via knfsd) so that we can build dcache paths to directories effectively.
1131 * If a dentry was found and moved, then it is returned. Otherwise NULL
1132 * is returned. This matches the expected return value of ->lookup.
1135 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1137 struct dentry
*new = NULL
;
1139 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1140 spin_lock(&dcache_lock
);
1141 new = __d_find_alias(inode
, 1);
1143 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1144 fsnotify_d_instantiate(new, inode
);
1145 spin_unlock(&dcache_lock
);
1146 security_d_instantiate(new, inode
);
1148 d_move(new, dentry
);
1151 /* d_instantiate takes dcache_lock, so we do it by hand */
1152 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1153 dentry
->d_inode
= inode
;
1154 fsnotify_d_instantiate(dentry
, inode
);
1155 spin_unlock(&dcache_lock
);
1156 security_d_instantiate(dentry
, inode
);
1160 d_add(dentry
, inode
);
1166 * d_lookup - search for a dentry
1167 * @parent: parent dentry
1168 * @name: qstr of name we wish to find
1170 * Searches the children of the parent dentry for the name in question. If
1171 * the dentry is found its reference count is incremented and the dentry
1172 * is returned. The caller must use d_put to free the entry when it has
1173 * finished using it. %NULL is returned on failure.
1175 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1176 * Memory barriers are used while updating and doing lockless traversal.
1177 * To avoid races with d_move while rename is happening, d_lock is used.
1179 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1180 * and name pointer in one structure pointed by d_qstr.
1182 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1183 * lookup is going on.
1185 * dentry_unused list is not updated even if lookup finds the required dentry
1186 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1187 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1190 * d_lookup() is protected against the concurrent renames in some unrelated
1191 * directory using the seqlockt_t rename_lock.
1194 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1196 struct dentry
* dentry
= NULL
;
1200 seq
= read_seqbegin(&rename_lock
);
1201 dentry
= __d_lookup(parent
, name
);
1204 } while (read_seqretry(&rename_lock
, seq
));
1208 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1210 unsigned int len
= name
->len
;
1211 unsigned int hash
= name
->hash
;
1212 const unsigned char *str
= name
->name
;
1213 struct hlist_head
*head
= d_hash(parent
,hash
);
1214 struct dentry
*found
= NULL
;
1215 struct hlist_node
*node
;
1216 struct dentry
*dentry
;
1220 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1223 if (dentry
->d_name
.hash
!= hash
)
1225 if (dentry
->d_parent
!= parent
)
1228 spin_lock(&dentry
->d_lock
);
1231 * Recheck the dentry after taking the lock - d_move may have
1232 * changed things. Don't bother checking the hash because we're
1233 * about to compare the whole name anyway.
1235 if (dentry
->d_parent
!= parent
)
1239 * It is safe to compare names since d_move() cannot
1240 * change the qstr (protected by d_lock).
1242 qstr
= &dentry
->d_name
;
1243 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1244 if (parent
->d_op
->d_compare(parent
, qstr
, name
))
1247 if (qstr
->len
!= len
)
1249 if (memcmp(qstr
->name
, str
, len
))
1253 if (!d_unhashed(dentry
)) {
1254 atomic_inc(&dentry
->d_count
);
1257 spin_unlock(&dentry
->d_lock
);
1260 spin_unlock(&dentry
->d_lock
);
1268 * d_hash_and_lookup - hash the qstr then search for a dentry
1269 * @dir: Directory to search in
1270 * @name: qstr of name we wish to find
1272 * On hash failure or on lookup failure NULL is returned.
1274 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1276 struct dentry
*dentry
= NULL
;
1279 * Check for a fs-specific hash function. Note that we must
1280 * calculate the standard hash first, as the d_op->d_hash()
1281 * routine may choose to leave the hash value unchanged.
1283 name
->hash
= full_name_hash(name
->name
, name
->len
);
1284 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1285 if (dir
->d_op
->d_hash(dir
, name
) < 0)
1288 dentry
= d_lookup(dir
, name
);
1294 * d_validate - verify dentry provided from insecure source
1295 * @dentry: The dentry alleged to be valid child of @dparent
1296 * @dparent: The parent dentry (known to be valid)
1297 * @hash: Hash of the dentry
1298 * @len: Length of the name
1300 * An insecure source has sent us a dentry, here we verify it and dget() it.
1301 * This is used by ncpfs in its readdir implementation.
1302 * Zero is returned in the dentry is invalid.
1305 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1307 struct hlist_head
*base
;
1308 struct hlist_node
*lhp
;
1310 /* Check whether the ptr might be valid at all.. */
1311 if (!kmem_ptr_validate(dentry_cache
, dentry
))
1314 if (dentry
->d_parent
!= dparent
)
1317 spin_lock(&dcache_lock
);
1318 base
= d_hash(dparent
, dentry
->d_name
.hash
);
1319 hlist_for_each(lhp
,base
) {
1320 /* hlist_for_each_entry_rcu() not required for d_hash list
1321 * as it is parsed under dcache_lock
1323 if (dentry
== hlist_entry(lhp
, struct dentry
, d_hash
)) {
1324 __dget_locked(dentry
);
1325 spin_unlock(&dcache_lock
);
1329 spin_unlock(&dcache_lock
);
1335 * When a file is deleted, we have two options:
1336 * - turn this dentry into a negative dentry
1337 * - unhash this dentry and free it.
1339 * Usually, we want to just turn this into
1340 * a negative dentry, but if anybody else is
1341 * currently using the dentry or the inode
1342 * we can't do that and we fall back on removing
1343 * it from the hash queues and waiting for
1344 * it to be deleted later when it has no users
1348 * d_delete - delete a dentry
1349 * @dentry: The dentry to delete
1351 * Turn the dentry into a negative dentry if possible, otherwise
1352 * remove it from the hash queues so it can be deleted later
1355 void d_delete(struct dentry
* dentry
)
1359 * Are we the only user?
1361 spin_lock(&dcache_lock
);
1362 spin_lock(&dentry
->d_lock
);
1363 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1364 if (atomic_read(&dentry
->d_count
) == 1) {
1365 dentry_iput(dentry
);
1366 fsnotify_nameremove(dentry
, isdir
);
1368 /* remove this and other inotify debug checks after 2.6.18 */
1369 dentry
->d_flags
&= ~DCACHE_INOTIFY_PARENT_WATCHED
;
1373 if (!d_unhashed(dentry
))
1376 spin_unlock(&dentry
->d_lock
);
1377 spin_unlock(&dcache_lock
);
1379 fsnotify_nameremove(dentry
, isdir
);
1382 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1385 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1386 hlist_add_head_rcu(&entry
->d_hash
, list
);
1389 static void _d_rehash(struct dentry
* entry
)
1391 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1395 * d_rehash - add an entry back to the hash
1396 * @entry: dentry to add to the hash
1398 * Adds a dentry to the hash according to its name.
1401 void d_rehash(struct dentry
* entry
)
1403 spin_lock(&dcache_lock
);
1404 spin_lock(&entry
->d_lock
);
1406 spin_unlock(&entry
->d_lock
);
1407 spin_unlock(&dcache_lock
);
1410 #define do_switch(x,y) do { \
1411 __typeof__ (x) __tmp = x; \
1412 x = y; y = __tmp; } while (0)
1415 * When switching names, the actual string doesn't strictly have to
1416 * be preserved in the target - because we're dropping the target
1417 * anyway. As such, we can just do a simple memcpy() to copy over
1418 * the new name before we switch.
1420 * Note that we have to be a lot more careful about getting the hash
1421 * switched - we have to switch the hash value properly even if it
1422 * then no longer matches the actual (corrupted) string of the target.
1423 * The hash value has to match the hash queue that the dentry is on..
1425 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1427 if (dname_external(target
)) {
1428 if (dname_external(dentry
)) {
1430 * Both external: swap the pointers
1432 do_switch(target
->d_name
.name
, dentry
->d_name
.name
);
1435 * dentry:internal, target:external. Steal target's
1436 * storage and make target internal.
1438 dentry
->d_name
.name
= target
->d_name
.name
;
1439 target
->d_name
.name
= target
->d_iname
;
1442 if (dname_external(dentry
)) {
1444 * dentry:external, target:internal. Give dentry's
1445 * storage to target and make dentry internal
1447 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1448 target
->d_name
.len
+ 1);
1449 target
->d_name
.name
= dentry
->d_name
.name
;
1450 dentry
->d_name
.name
= dentry
->d_iname
;
1453 * Both are internal. Just copy target to dentry
1455 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1456 target
->d_name
.len
+ 1);
1462 * We cannibalize "target" when moving dentry on top of it,
1463 * because it's going to be thrown away anyway. We could be more
1464 * polite about it, though.
1466 * This forceful removal will result in ugly /proc output if
1467 * somebody holds a file open that got deleted due to a rename.
1468 * We could be nicer about the deleted file, and let it show
1469 * up under the name it got deleted rather than the name that
1474 * d_move_locked - move a dentry
1475 * @dentry: entry to move
1476 * @target: new dentry
1478 * Update the dcache to reflect the move of a file name. Negative
1479 * dcache entries should not be moved in this way.
1481 static void d_move_locked(struct dentry
* dentry
, struct dentry
* target
)
1483 struct hlist_head
*list
;
1485 if (!dentry
->d_inode
)
1486 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1488 write_seqlock(&rename_lock
);
1490 * XXXX: do we really need to take target->d_lock?
1492 if (target
< dentry
) {
1493 spin_lock(&target
->d_lock
);
1494 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1496 spin_lock(&dentry
->d_lock
);
1497 spin_lock_nested(&target
->d_lock
, DENTRY_D_LOCK_NESTED
);
1500 /* Move the dentry to the target hash queue, if on different bucket */
1501 if (dentry
->d_flags
& DCACHE_UNHASHED
)
1502 goto already_unhashed
;
1504 hlist_del_rcu(&dentry
->d_hash
);
1507 list
= d_hash(target
->d_parent
, target
->d_name
.hash
);
1508 __d_rehash(dentry
, list
);
1510 /* Unhash the target: dput() will then get rid of it */
1513 list_del(&dentry
->d_u
.d_child
);
1514 list_del(&target
->d_u
.d_child
);
1516 /* Switch the names.. */
1517 switch_names(dentry
, target
);
1518 do_switch(dentry
->d_name
.len
, target
->d_name
.len
);
1519 do_switch(dentry
->d_name
.hash
, target
->d_name
.hash
);
1521 /* ... and switch the parents */
1522 if (IS_ROOT(dentry
)) {
1523 dentry
->d_parent
= target
->d_parent
;
1524 target
->d_parent
= target
;
1525 INIT_LIST_HEAD(&target
->d_u
.d_child
);
1527 do_switch(dentry
->d_parent
, target
->d_parent
);
1529 /* And add them back to the (new) parent lists */
1530 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
1533 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1534 spin_unlock(&target
->d_lock
);
1535 fsnotify_d_move(dentry
);
1536 spin_unlock(&dentry
->d_lock
);
1537 write_sequnlock(&rename_lock
);
1541 * d_move - move a dentry
1542 * @dentry: entry to move
1543 * @target: new dentry
1545 * Update the dcache to reflect the move of a file name. Negative
1546 * dcache entries should not be moved in this way.
1549 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1551 spin_lock(&dcache_lock
);
1552 d_move_locked(dentry
, target
);
1553 spin_unlock(&dcache_lock
);
1557 * Helper that returns 1 if p1 is a parent of p2, else 0
1559 static int d_isparent(struct dentry
*p1
, struct dentry
*p2
)
1563 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1564 if (p
->d_parent
== p1
)
1571 * This helper attempts to cope with remotely renamed directories
1573 * It assumes that the caller is already holding
1574 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1576 * Note: If ever the locking in lock_rename() changes, then please
1577 * remember to update this too...
1579 * On return, dcache_lock will have been unlocked.
1581 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
1583 struct mutex
*m1
= NULL
, *m2
= NULL
;
1586 /* If alias and dentry share a parent, then no extra locks required */
1587 if (alias
->d_parent
== dentry
->d_parent
)
1590 /* Check for loops */
1591 ret
= ERR_PTR(-ELOOP
);
1592 if (d_isparent(alias
, dentry
))
1595 /* See lock_rename() */
1596 ret
= ERR_PTR(-EBUSY
);
1597 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
1599 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
1600 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
1602 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
1604 d_move_locked(alias
, dentry
);
1607 spin_unlock(&dcache_lock
);
1616 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1617 * named dentry in place of the dentry to be replaced.
1619 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
1621 struct dentry
*dparent
, *aparent
;
1623 switch_names(dentry
, anon
);
1624 do_switch(dentry
->d_name
.len
, anon
->d_name
.len
);
1625 do_switch(dentry
->d_name
.hash
, anon
->d_name
.hash
);
1627 dparent
= dentry
->d_parent
;
1628 aparent
= anon
->d_parent
;
1630 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
1631 list_del(&dentry
->d_u
.d_child
);
1632 if (!IS_ROOT(dentry
))
1633 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1635 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1637 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
1638 list_del(&anon
->d_u
.d_child
);
1640 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
1642 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
1644 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
1648 * d_materialise_unique - introduce an inode into the tree
1649 * @dentry: candidate dentry
1650 * @inode: inode to bind to the dentry, to which aliases may be attached
1652 * Introduces an dentry into the tree, substituting an extant disconnected
1653 * root directory alias in its place if there is one
1655 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
1657 struct dentry
*actual
;
1659 BUG_ON(!d_unhashed(dentry
));
1661 spin_lock(&dcache_lock
);
1665 dentry
->d_inode
= NULL
;
1669 if (S_ISDIR(inode
->i_mode
)) {
1670 struct dentry
*alias
;
1672 /* Does an aliased dentry already exist? */
1673 alias
= __d_find_alias(inode
, 0);
1676 /* Is this an anonymous mountpoint that we could splice
1678 if (IS_ROOT(alias
)) {
1679 spin_lock(&alias
->d_lock
);
1680 __d_materialise_dentry(dentry
, alias
);
1684 /* Nope, but we must(!) avoid directory aliasing */
1685 actual
= __d_unalias(dentry
, alias
);
1692 /* Add a unique reference */
1693 actual
= __d_instantiate_unique(dentry
, inode
);
1696 else if (unlikely(!d_unhashed(actual
)))
1697 goto shouldnt_be_hashed
;
1700 spin_lock(&actual
->d_lock
);
1703 spin_unlock(&actual
->d_lock
);
1704 spin_unlock(&dcache_lock
);
1706 if (actual
== dentry
) {
1707 security_d_instantiate(dentry
, inode
);
1715 spin_unlock(&dcache_lock
);
1717 goto shouldnt_be_hashed
;
1721 * d_path - return the path of a dentry
1722 * @dentry: dentry to report
1723 * @vfsmnt: vfsmnt to which the dentry belongs
1724 * @root: root dentry
1725 * @rootmnt: vfsmnt to which the root dentry belongs
1726 * @buffer: buffer to return value in
1727 * @buflen: buffer length
1729 * Convert a dentry into an ASCII path name. If the entry has been deleted
1730 * the string " (deleted)" is appended. Note that this is ambiguous.
1732 * Returns the buffer or an error code if the path was too long.
1734 * "buflen" should be positive. Caller holds the dcache_lock.
1736 static char * __d_path( struct dentry
*dentry
, struct vfsmount
*vfsmnt
,
1737 struct dentry
*root
, struct vfsmount
*rootmnt
,
1738 char *buffer
, int buflen
)
1740 char * end
= buffer
+buflen
;
1746 if (!IS_ROOT(dentry
) && d_unhashed(dentry
)) {
1751 memcpy(end
, " (deleted)", 10);
1761 struct dentry
* parent
;
1763 if (dentry
== root
&& vfsmnt
== rootmnt
)
1765 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
1767 spin_lock(&vfsmount_lock
);
1768 if (vfsmnt
->mnt_parent
== vfsmnt
) {
1769 spin_unlock(&vfsmount_lock
);
1772 dentry
= vfsmnt
->mnt_mountpoint
;
1773 vfsmnt
= vfsmnt
->mnt_parent
;
1774 spin_unlock(&vfsmount_lock
);
1777 parent
= dentry
->d_parent
;
1779 namelen
= dentry
->d_name
.len
;
1780 buflen
-= namelen
+ 1;
1784 memcpy(end
, dentry
->d_name
.name
, namelen
);
1793 namelen
= dentry
->d_name
.len
;
1797 retval
-= namelen
-1; /* hit the slash */
1798 memcpy(retval
, dentry
->d_name
.name
, namelen
);
1801 return ERR_PTR(-ENAMETOOLONG
);
1804 /* write full pathname into buffer and return start of pathname */
1805 char * d_path(struct dentry
*dentry
, struct vfsmount
*vfsmnt
,
1806 char *buf
, int buflen
)
1809 struct vfsmount
*rootmnt
;
1810 struct dentry
*root
;
1812 read_lock(¤t
->fs
->lock
);
1813 rootmnt
= mntget(current
->fs
->rootmnt
);
1814 root
= dget(current
->fs
->root
);
1815 read_unlock(¤t
->fs
->lock
);
1816 spin_lock(&dcache_lock
);
1817 res
= __d_path(dentry
, vfsmnt
, root
, rootmnt
, buf
, buflen
);
1818 spin_unlock(&dcache_lock
);
1825 * NOTE! The user-level library version returns a
1826 * character pointer. The kernel system call just
1827 * returns the length of the buffer filled (which
1828 * includes the ending '\0' character), or a negative
1829 * error value. So libc would do something like
1831 * char *getcwd(char * buf, size_t size)
1835 * retval = sys_getcwd(buf, size);
1842 asmlinkage
long sys_getcwd(char __user
*buf
, unsigned long size
)
1845 struct vfsmount
*pwdmnt
, *rootmnt
;
1846 struct dentry
*pwd
, *root
;
1847 char *page
= (char *) __get_free_page(GFP_USER
);
1852 read_lock(¤t
->fs
->lock
);
1853 pwdmnt
= mntget(current
->fs
->pwdmnt
);
1854 pwd
= dget(current
->fs
->pwd
);
1855 rootmnt
= mntget(current
->fs
->rootmnt
);
1856 root
= dget(current
->fs
->root
);
1857 read_unlock(¤t
->fs
->lock
);
1860 /* Has the current directory has been unlinked? */
1861 spin_lock(&dcache_lock
);
1862 if (pwd
->d_parent
== pwd
|| !d_unhashed(pwd
)) {
1866 cwd
= __d_path(pwd
, pwdmnt
, root
, rootmnt
, page
, PAGE_SIZE
);
1867 spin_unlock(&dcache_lock
);
1869 error
= PTR_ERR(cwd
);
1874 len
= PAGE_SIZE
+ page
- cwd
;
1877 if (copy_to_user(buf
, cwd
, len
))
1881 spin_unlock(&dcache_lock
);
1888 free_page((unsigned long) page
);
1893 * Test whether new_dentry is a subdirectory of old_dentry.
1895 * Trivially implemented using the dcache structure
1899 * is_subdir - is new dentry a subdirectory of old_dentry
1900 * @new_dentry: new dentry
1901 * @old_dentry: old dentry
1903 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1904 * Returns 0 otherwise.
1905 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1908 int is_subdir(struct dentry
* new_dentry
, struct dentry
* old_dentry
)
1911 struct dentry
* saved
= new_dentry
;
1914 /* need rcu_readlock to protect against the d_parent trashing due to
1919 /* for restarting inner loop in case of seq retry */
1922 seq
= read_seqbegin(&rename_lock
);
1924 if (new_dentry
!= old_dentry
) {
1925 struct dentry
* parent
= new_dentry
->d_parent
;
1926 if (parent
== new_dentry
)
1928 new_dentry
= parent
;
1934 } while (read_seqretry(&rename_lock
, seq
));
1940 void d_genocide(struct dentry
*root
)
1942 struct dentry
*this_parent
= root
;
1943 struct list_head
*next
;
1945 spin_lock(&dcache_lock
);
1947 next
= this_parent
->d_subdirs
.next
;
1949 while (next
!= &this_parent
->d_subdirs
) {
1950 struct list_head
*tmp
= next
;
1951 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1953 if (d_unhashed(dentry
)||!dentry
->d_inode
)
1955 if (!list_empty(&dentry
->d_subdirs
)) {
1956 this_parent
= dentry
;
1959 atomic_dec(&dentry
->d_count
);
1961 if (this_parent
!= root
) {
1962 next
= this_parent
->d_u
.d_child
.next
;
1963 atomic_dec(&this_parent
->d_count
);
1964 this_parent
= this_parent
->d_parent
;
1967 spin_unlock(&dcache_lock
);
1971 * find_inode_number - check for dentry with name
1972 * @dir: directory to check
1973 * @name: Name to find.
1975 * Check whether a dentry already exists for the given name,
1976 * and return the inode number if it has an inode. Otherwise
1979 * This routine is used to post-process directory listings for
1980 * filesystems using synthetic inode numbers, and is necessary
1981 * to keep getcwd() working.
1984 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
1986 struct dentry
* dentry
;
1989 dentry
= d_hash_and_lookup(dir
, name
);
1991 if (dentry
->d_inode
)
1992 ino
= dentry
->d_inode
->i_ino
;
1998 static __initdata
unsigned long dhash_entries
;
1999 static int __init
set_dhash_entries(char *str
)
2003 dhash_entries
= simple_strtoul(str
, &str
, 0);
2006 __setup("dhash_entries=", set_dhash_entries
);
2008 static void __init
dcache_init_early(void)
2012 /* If hashes are distributed across NUMA nodes, defer
2013 * hash allocation until vmalloc space is available.
2019 alloc_large_system_hash("Dentry cache",
2020 sizeof(struct hlist_head
),
2028 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2029 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2032 static void __init
dcache_init(unsigned long mempages
)
2037 * A constructor could be added for stable state like the lists,
2038 * but it is probably not worth it because of the cache nature
2041 dentry_cache
= kmem_cache_create("dentry_cache",
2042 sizeof(struct dentry
),
2044 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
2048 set_shrinker(DEFAULT_SEEKS
, shrink_dcache_memory
);
2050 /* Hash may have been set up in dcache_init_early */
2055 alloc_large_system_hash("Dentry cache",
2056 sizeof(struct hlist_head
),
2064 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2065 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2068 /* SLAB cache for __getname() consumers */
2069 kmem_cache_t
*names_cachep __read_mostly
;
2071 /* SLAB cache for file structures */
2072 kmem_cache_t
*filp_cachep __read_mostly
;
2074 EXPORT_SYMBOL(d_genocide
);
2076 void __init
vfs_caches_init_early(void)
2078 dcache_init_early();
2082 void __init
vfs_caches_init(unsigned long mempages
)
2084 unsigned long reserve
;
2086 /* Base hash sizes on available memory, with a reserve equal to
2087 150% of current kernel size */
2089 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2090 mempages
-= reserve
;
2092 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2093 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
2095 filp_cachep
= kmem_cache_create("filp", sizeof(struct file
), 0,
2096 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
2098 dcache_init(mempages
);
2099 inode_init(mempages
);
2100 files_init(mempages
);
2106 EXPORT_SYMBOL(d_alloc
);
2107 EXPORT_SYMBOL(d_alloc_anon
);
2108 EXPORT_SYMBOL(d_alloc_root
);
2109 EXPORT_SYMBOL(d_delete
);
2110 EXPORT_SYMBOL(d_find_alias
);
2111 EXPORT_SYMBOL(d_instantiate
);
2112 EXPORT_SYMBOL(d_invalidate
);
2113 EXPORT_SYMBOL(d_lookup
);
2114 EXPORT_SYMBOL(d_move
);
2115 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2116 EXPORT_SYMBOL(d_path
);
2117 EXPORT_SYMBOL(d_prune_aliases
);
2118 EXPORT_SYMBOL(d_rehash
);
2119 EXPORT_SYMBOL(d_splice_alias
);
2120 EXPORT_SYMBOL(d_validate
);
2121 EXPORT_SYMBOL(dget_locked
);
2122 EXPORT_SYMBOL(dput
);
2123 EXPORT_SYMBOL(find_inode_number
);
2124 EXPORT_SYMBOL(have_submounts
);
2125 EXPORT_SYMBOL(names_cachep
);
2126 EXPORT_SYMBOL(shrink_dcache_parent
);
2127 EXPORT_SYMBOL(shrink_dcache_sb
);