4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
38 int sysctl_vfs_cache_pressure __read_mostly
= 100;
39 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
41 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lock
);
42 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
44 EXPORT_SYMBOL(dcache_lock
);
46 static struct kmem_cache
*dentry_cache __read_mostly
;
48 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
58 #define D_HASHBITS d_hash_shift
59 #define D_HASHMASK d_hash_mask
61 static unsigned int d_hash_mask __read_mostly
;
62 static unsigned int d_hash_shift __read_mostly
;
63 static struct hlist_head
*dentry_hashtable __read_mostly
;
65 /* Statistics gathering. */
66 struct dentry_stat_t dentry_stat
= {
70 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
72 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
73 static int get_nr_dentry(void)
77 for_each_possible_cpu(i
)
78 sum
+= per_cpu(nr_dentry
, i
);
79 return sum
< 0 ? 0 : sum
;
82 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
83 size_t *lenp
, loff_t
*ppos
)
85 dentry_stat
.nr_dentry
= get_nr_dentry();
86 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
90 static void __d_free(struct rcu_head
*head
)
92 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
94 WARN_ON(!list_empty(&dentry
->d_alias
));
95 if (dname_external(dentry
))
96 kfree(dentry
->d_name
.name
);
97 kmem_cache_free(dentry_cache
, dentry
);
101 * no dcache_lock, please.
103 static void d_free(struct dentry
*dentry
)
105 this_cpu_dec(nr_dentry
);
106 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
107 dentry
->d_op
->d_release(dentry
);
109 /* if dentry was never inserted into hash, immediate free is OK */
110 if (hlist_unhashed(&dentry
->d_hash
))
111 __d_free(&dentry
->d_u
.d_rcu
);
113 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
117 * Release the dentry's inode, using the filesystem
118 * d_iput() operation if defined.
120 static void dentry_iput(struct dentry
* dentry
)
121 __releases(dentry
->d_lock
)
122 __releases(dcache_lock
)
124 struct inode
*inode
= dentry
->d_inode
;
126 dentry
->d_inode
= NULL
;
127 list_del_init(&dentry
->d_alias
);
128 spin_unlock(&dentry
->d_lock
);
129 spin_unlock(&dcache_lock
);
131 fsnotify_inoderemove(inode
);
132 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
133 dentry
->d_op
->d_iput(dentry
, inode
);
137 spin_unlock(&dentry
->d_lock
);
138 spin_unlock(&dcache_lock
);
143 * dentry_lru_(add|del|move_tail) must be called with dcache_lock held.
145 static void dentry_lru_add(struct dentry
*dentry
)
147 if (list_empty(&dentry
->d_lru
)) {
148 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
149 dentry
->d_sb
->s_nr_dentry_unused
++;
150 dentry_stat
.nr_unused
++;
154 static void dentry_lru_del(struct dentry
*dentry
)
156 if (!list_empty(&dentry
->d_lru
)) {
157 list_del_init(&dentry
->d_lru
);
158 dentry
->d_sb
->s_nr_dentry_unused
--;
159 dentry_stat
.nr_unused
--;
163 static void dentry_lru_move_tail(struct dentry
*dentry
)
165 if (list_empty(&dentry
->d_lru
)) {
166 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
167 dentry
->d_sb
->s_nr_dentry_unused
++;
168 dentry_stat
.nr_unused
++;
170 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
175 * d_kill - kill dentry and return parent
176 * @dentry: dentry to kill
178 * The dentry must already be unhashed and removed from the LRU.
180 * If this is the root of the dentry tree, return NULL.
182 static struct dentry
*d_kill(struct dentry
*dentry
)
183 __releases(dentry
->d_lock
)
184 __releases(dcache_lock
)
186 struct dentry
*parent
;
188 list_del(&dentry
->d_u
.d_child
);
189 /*drops the locks, at that point nobody can reach this dentry */
194 parent
= dentry
->d_parent
;
202 * This is complicated by the fact that we do not want to put
203 * dentries that are no longer on any hash chain on the unused
204 * list: we'd much rather just get rid of them immediately.
206 * However, that implies that we have to traverse the dentry
207 * tree upwards to the parents which might _also_ now be
208 * scheduled for deletion (it may have been only waiting for
209 * its last child to go away).
211 * This tail recursion is done by hand as we don't want to depend
212 * on the compiler to always get this right (gcc generally doesn't).
213 * Real recursion would eat up our stack space.
217 * dput - release a dentry
218 * @dentry: dentry to release
220 * Release a dentry. This will drop the usage count and if appropriate
221 * call the dentry unlink method as well as removing it from the queues and
222 * releasing its resources. If the parent dentries were scheduled for release
223 * they too may now get deleted.
225 * no dcache lock, please.
228 void dput(struct dentry
*dentry
)
234 if (atomic_read(&dentry
->d_count
) == 1)
236 if (!atomic_dec_and_lock(&dentry
->d_count
, &dcache_lock
))
239 spin_lock(&dentry
->d_lock
);
240 if (atomic_read(&dentry
->d_count
)) {
241 spin_unlock(&dentry
->d_lock
);
242 spin_unlock(&dcache_lock
);
247 * AV: ->d_delete() is _NOT_ allowed to block now.
249 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
250 if (dentry
->d_op
->d_delete(dentry
))
254 /* Unreachable? Get rid of it */
255 if (d_unhashed(dentry
))
258 /* Otherwise leave it cached and ensure it's on the LRU */
259 dentry
->d_flags
|= DCACHE_REFERENCED
;
260 dentry_lru_add(dentry
);
262 spin_unlock(&dentry
->d_lock
);
263 spin_unlock(&dcache_lock
);
269 /* if dentry was on the d_lru list delete it from there */
270 dentry_lru_del(dentry
);
271 dentry
= d_kill(dentry
);
278 * d_invalidate - invalidate a dentry
279 * @dentry: dentry to invalidate
281 * Try to invalidate the dentry if it turns out to be
282 * possible. If there are other dentries that can be
283 * reached through this one we can't delete it and we
284 * return -EBUSY. On success we return 0.
289 int d_invalidate(struct dentry
* dentry
)
292 * If it's already been dropped, return OK.
294 spin_lock(&dcache_lock
);
295 if (d_unhashed(dentry
)) {
296 spin_unlock(&dcache_lock
);
300 * Check whether to do a partial shrink_dcache
301 * to get rid of unused child entries.
303 if (!list_empty(&dentry
->d_subdirs
)) {
304 spin_unlock(&dcache_lock
);
305 shrink_dcache_parent(dentry
);
306 spin_lock(&dcache_lock
);
310 * Somebody else still using it?
312 * If it's a directory, we can't drop it
313 * for fear of somebody re-populating it
314 * with children (even though dropping it
315 * would make it unreachable from the root,
316 * we might still populate it if it was a
317 * working directory or similar).
319 spin_lock(&dentry
->d_lock
);
320 if (atomic_read(&dentry
->d_count
) > 1) {
321 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
322 spin_unlock(&dentry
->d_lock
);
323 spin_unlock(&dcache_lock
);
329 spin_unlock(&dentry
->d_lock
);
330 spin_unlock(&dcache_lock
);
333 EXPORT_SYMBOL(d_invalidate
);
335 /* This should be called _only_ with dcache_lock held */
336 static inline struct dentry
* __dget_locked(struct dentry
*dentry
)
338 atomic_inc(&dentry
->d_count
);
339 dentry_lru_del(dentry
);
343 struct dentry
* dget_locked(struct dentry
*dentry
)
345 return __dget_locked(dentry
);
347 EXPORT_SYMBOL(dget_locked
);
350 * d_find_alias - grab a hashed alias of inode
351 * @inode: inode in question
352 * @want_discon: flag, used by d_splice_alias, to request
353 * that only a DISCONNECTED alias be returned.
355 * If inode has a hashed alias, or is a directory and has any alias,
356 * acquire the reference to alias and return it. Otherwise return NULL.
357 * Notice that if inode is a directory there can be only one alias and
358 * it can be unhashed only if it has no children, or if it is the root
361 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
362 * any other hashed alias over that one unless @want_discon is set,
363 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
366 static struct dentry
* __d_find_alias(struct inode
*inode
, int want_discon
)
368 struct list_head
*head
, *next
, *tmp
;
369 struct dentry
*alias
, *discon_alias
=NULL
;
371 head
= &inode
->i_dentry
;
372 next
= inode
->i_dentry
.next
;
373 while (next
!= head
) {
377 alias
= list_entry(tmp
, struct dentry
, d_alias
);
378 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
379 if (IS_ROOT(alias
) &&
380 (alias
->d_flags
& DCACHE_DISCONNECTED
))
381 discon_alias
= alias
;
382 else if (!want_discon
) {
383 __dget_locked(alias
);
389 __dget_locked(discon_alias
);
393 struct dentry
* d_find_alias(struct inode
*inode
)
395 struct dentry
*de
= NULL
;
397 if (!list_empty(&inode
->i_dentry
)) {
398 spin_lock(&dcache_lock
);
399 de
= __d_find_alias(inode
, 0);
400 spin_unlock(&dcache_lock
);
404 EXPORT_SYMBOL(d_find_alias
);
407 * Try to kill dentries associated with this inode.
408 * WARNING: you must own a reference to inode.
410 void d_prune_aliases(struct inode
*inode
)
412 struct dentry
*dentry
;
414 spin_lock(&dcache_lock
);
415 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
416 spin_lock(&dentry
->d_lock
);
417 if (!atomic_read(&dentry
->d_count
)) {
418 __dget_locked(dentry
);
420 spin_unlock(&dentry
->d_lock
);
421 spin_unlock(&dcache_lock
);
425 spin_unlock(&dentry
->d_lock
);
427 spin_unlock(&dcache_lock
);
429 EXPORT_SYMBOL(d_prune_aliases
);
432 * Throw away a dentry - free the inode, dput the parent. This requires that
433 * the LRU list has already been removed.
435 * Try to prune ancestors as well. This is necessary to prevent
436 * quadratic behavior of shrink_dcache_parent(), but is also expected
437 * to be beneficial in reducing dentry cache fragmentation.
439 static void prune_one_dentry(struct dentry
* dentry
)
440 __releases(dentry
->d_lock
)
441 __releases(dcache_lock
)
442 __acquires(dcache_lock
)
445 dentry
= d_kill(dentry
);
448 * Prune ancestors. Locking is simpler than in dput(),
449 * because dcache_lock needs to be taken anyway.
451 spin_lock(&dcache_lock
);
453 if (!atomic_dec_and_lock(&dentry
->d_count
, &dentry
->d_lock
))
456 dentry_lru_del(dentry
);
458 dentry
= d_kill(dentry
);
459 spin_lock(&dcache_lock
);
463 static void shrink_dentry_list(struct list_head
*list
)
465 struct dentry
*dentry
;
467 while (!list_empty(list
)) {
468 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
469 dentry_lru_del(dentry
);
472 * We found an inuse dentry which was not removed from
473 * the LRU because of laziness during lookup. Do not free
474 * it - just keep it off the LRU list.
476 spin_lock(&dentry
->d_lock
);
477 if (atomic_read(&dentry
->d_count
)) {
478 spin_unlock(&dentry
->d_lock
);
481 prune_one_dentry(dentry
);
482 /* dentry->d_lock was dropped in prune_one_dentry() */
483 cond_resched_lock(&dcache_lock
);
488 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
489 * @sb: superblock to shrink dentry LRU.
490 * @count: number of entries to prune
491 * @flags: flags to control the dentry processing
493 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
495 static void __shrink_dcache_sb(struct super_block
*sb
, int *count
, int flags
)
497 /* called from prune_dcache() and shrink_dcache_parent() */
498 struct dentry
*dentry
;
499 LIST_HEAD(referenced
);
503 spin_lock(&dcache_lock
);
504 while (!list_empty(&sb
->s_dentry_lru
)) {
505 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
506 struct dentry
, d_lru
);
507 BUG_ON(dentry
->d_sb
!= sb
);
510 * If we are honouring the DCACHE_REFERENCED flag and the
511 * dentry has this flag set, don't free it. Clear the flag
512 * and put it back on the LRU.
514 if (flags
& DCACHE_REFERENCED
) {
515 spin_lock(&dentry
->d_lock
);
516 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
517 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
518 list_move(&dentry
->d_lru
, &referenced
);
519 spin_unlock(&dentry
->d_lock
);
520 cond_resched_lock(&dcache_lock
);
523 spin_unlock(&dentry
->d_lock
);
526 list_move_tail(&dentry
->d_lru
, &tmp
);
529 cond_resched_lock(&dcache_lock
);
533 shrink_dentry_list(&tmp
);
535 if (!list_empty(&referenced
))
536 list_splice(&referenced
, &sb
->s_dentry_lru
);
537 spin_unlock(&dcache_lock
);
542 * prune_dcache - shrink the dcache
543 * @count: number of entries to try to free
545 * Shrink the dcache. This is done when we need more memory, or simply when we
546 * need to unmount something (at which point we need to unuse all dentries).
548 * This function may fail to free any resources if all the dentries are in use.
550 static void prune_dcache(int count
)
552 struct super_block
*sb
, *p
= NULL
;
554 int unused
= dentry_stat
.nr_unused
;
558 if (unused
== 0 || count
== 0)
560 spin_lock(&dcache_lock
);
564 prune_ratio
= unused
/ count
;
566 list_for_each_entry(sb
, &super_blocks
, s_list
) {
567 if (list_empty(&sb
->s_instances
))
569 if (sb
->s_nr_dentry_unused
== 0)
572 /* Now, we reclaim unused dentrins with fairness.
573 * We reclaim them same percentage from each superblock.
574 * We calculate number of dentries to scan on this sb
575 * as follows, but the implementation is arranged to avoid
577 * number of dentries to scan on this sb =
578 * count * (number of dentries on this sb /
579 * number of dentries in the machine)
581 spin_unlock(&sb_lock
);
582 if (prune_ratio
!= 1)
583 w_count
= (sb
->s_nr_dentry_unused
/ prune_ratio
) + 1;
585 w_count
= sb
->s_nr_dentry_unused
;
588 * We need to be sure this filesystem isn't being unmounted,
589 * otherwise we could race with generic_shutdown_super(), and
590 * end up holding a reference to an inode while the filesystem
591 * is unmounted. So we try to get s_umount, and make sure
594 if (down_read_trylock(&sb
->s_umount
)) {
595 if ((sb
->s_root
!= NULL
) &&
596 (!list_empty(&sb
->s_dentry_lru
))) {
597 spin_unlock(&dcache_lock
);
598 __shrink_dcache_sb(sb
, &w_count
,
601 spin_lock(&dcache_lock
);
603 up_read(&sb
->s_umount
);
610 /* more work left to do? */
616 spin_unlock(&sb_lock
);
617 spin_unlock(&dcache_lock
);
621 * shrink_dcache_sb - shrink dcache for a superblock
624 * Shrink the dcache for the specified super block. This is used to free
625 * the dcache before unmounting a file system.
627 void shrink_dcache_sb(struct super_block
*sb
)
631 spin_lock(&dcache_lock
);
632 while (!list_empty(&sb
->s_dentry_lru
)) {
633 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
634 shrink_dentry_list(&tmp
);
636 spin_unlock(&dcache_lock
);
638 EXPORT_SYMBOL(shrink_dcache_sb
);
641 * destroy a single subtree of dentries for unmount
642 * - see the comments on shrink_dcache_for_umount() for a description of the
645 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
647 struct dentry
*parent
;
648 unsigned detached
= 0;
650 BUG_ON(!IS_ROOT(dentry
));
652 /* detach this root from the system */
653 spin_lock(&dcache_lock
);
654 dentry_lru_del(dentry
);
656 spin_unlock(&dcache_lock
);
659 /* descend to the first leaf in the current subtree */
660 while (!list_empty(&dentry
->d_subdirs
)) {
663 /* this is a branch with children - detach all of them
664 * from the system in one go */
665 spin_lock(&dcache_lock
);
666 list_for_each_entry(loop
, &dentry
->d_subdirs
,
668 dentry_lru_del(loop
);
670 cond_resched_lock(&dcache_lock
);
672 spin_unlock(&dcache_lock
);
674 /* move to the first child */
675 dentry
= list_entry(dentry
->d_subdirs
.next
,
676 struct dentry
, d_u
.d_child
);
679 /* consume the dentries from this leaf up through its parents
680 * until we find one with children or run out altogether */
684 if (atomic_read(&dentry
->d_count
) != 0) {
686 "BUG: Dentry %p{i=%lx,n=%s}"
688 " [unmount of %s %s]\n",
691 dentry
->d_inode
->i_ino
: 0UL,
693 atomic_read(&dentry
->d_count
),
694 dentry
->d_sb
->s_type
->name
,
702 parent
= dentry
->d_parent
;
703 atomic_dec(&parent
->d_count
);
706 list_del(&dentry
->d_u
.d_child
);
709 inode
= dentry
->d_inode
;
711 dentry
->d_inode
= NULL
;
712 list_del_init(&dentry
->d_alias
);
713 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
714 dentry
->d_op
->d_iput(dentry
, inode
);
721 /* finished when we fall off the top of the tree,
722 * otherwise we ascend to the parent and move to the
723 * next sibling if there is one */
727 } while (list_empty(&dentry
->d_subdirs
));
729 dentry
= list_entry(dentry
->d_subdirs
.next
,
730 struct dentry
, d_u
.d_child
);
735 * destroy the dentries attached to a superblock on unmounting
736 * - we don't need to use dentry->d_lock, and only need dcache_lock when
737 * removing the dentry from the system lists and hashes because:
738 * - the superblock is detached from all mountings and open files, so the
739 * dentry trees will not be rearranged by the VFS
740 * - s_umount is write-locked, so the memory pressure shrinker will ignore
741 * any dentries belonging to this superblock that it comes across
742 * - the filesystem itself is no longer permitted to rearrange the dentries
745 void shrink_dcache_for_umount(struct super_block
*sb
)
747 struct dentry
*dentry
;
749 if (down_read_trylock(&sb
->s_umount
))
754 atomic_dec(&dentry
->d_count
);
755 shrink_dcache_for_umount_subtree(dentry
);
757 while (!hlist_empty(&sb
->s_anon
)) {
758 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
759 shrink_dcache_for_umount_subtree(dentry
);
764 * Search for at least 1 mount point in the dentry's subdirs.
765 * We descend to the next level whenever the d_subdirs
766 * list is non-empty and continue searching.
770 * have_submounts - check for mounts over a dentry
771 * @parent: dentry to check.
773 * Return true if the parent or its subdirectories contain
777 int have_submounts(struct dentry
*parent
)
779 struct dentry
*this_parent
= parent
;
780 struct list_head
*next
;
782 spin_lock(&dcache_lock
);
783 if (d_mountpoint(parent
))
786 next
= this_parent
->d_subdirs
.next
;
788 while (next
!= &this_parent
->d_subdirs
) {
789 struct list_head
*tmp
= next
;
790 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
792 /* Have we found a mount point ? */
793 if (d_mountpoint(dentry
))
795 if (!list_empty(&dentry
->d_subdirs
)) {
796 this_parent
= dentry
;
801 * All done at this level ... ascend and resume the search.
803 if (this_parent
!= parent
) {
804 next
= this_parent
->d_u
.d_child
.next
;
805 this_parent
= this_parent
->d_parent
;
808 spin_unlock(&dcache_lock
);
809 return 0; /* No mount points found in tree */
811 spin_unlock(&dcache_lock
);
814 EXPORT_SYMBOL(have_submounts
);
817 * Search the dentry child list for the specified parent,
818 * and move any unused dentries to the end of the unused
819 * list for prune_dcache(). We descend to the next level
820 * whenever the d_subdirs list is non-empty and continue
823 * It returns zero iff there are no unused children,
824 * otherwise it returns the number of children moved to
825 * the end of the unused list. This may not be the total
826 * number of unused children, because select_parent can
827 * drop the lock and return early due to latency
830 static int select_parent(struct dentry
* parent
)
832 struct dentry
*this_parent
= parent
;
833 struct list_head
*next
;
836 spin_lock(&dcache_lock
);
838 next
= this_parent
->d_subdirs
.next
;
840 while (next
!= &this_parent
->d_subdirs
) {
841 struct list_head
*tmp
= next
;
842 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
846 * move only zero ref count dentries to the end
847 * of the unused list for prune_dcache
849 if (!atomic_read(&dentry
->d_count
)) {
850 dentry_lru_move_tail(dentry
);
853 dentry_lru_del(dentry
);
857 * We can return to the caller if we have found some (this
858 * ensures forward progress). We'll be coming back to find
861 if (found
&& need_resched())
865 * Descend a level if the d_subdirs list is non-empty.
867 if (!list_empty(&dentry
->d_subdirs
)) {
868 this_parent
= dentry
;
873 * All done at this level ... ascend and resume the search.
875 if (this_parent
!= parent
) {
876 next
= this_parent
->d_u
.d_child
.next
;
877 this_parent
= this_parent
->d_parent
;
881 spin_unlock(&dcache_lock
);
886 * shrink_dcache_parent - prune dcache
887 * @parent: parent of entries to prune
889 * Prune the dcache to remove unused children of the parent dentry.
892 void shrink_dcache_parent(struct dentry
* parent
)
894 struct super_block
*sb
= parent
->d_sb
;
897 while ((found
= select_parent(parent
)) != 0)
898 __shrink_dcache_sb(sb
, &found
, 0);
900 EXPORT_SYMBOL(shrink_dcache_parent
);
903 * Scan `nr' dentries and return the number which remain.
905 * We need to avoid reentering the filesystem if the caller is performing a
906 * GFP_NOFS allocation attempt. One example deadlock is:
908 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
909 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
910 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
912 * In this case we return -1 to tell the caller that we baled.
914 static int shrink_dcache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
917 if (!(gfp_mask
& __GFP_FS
))
922 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
925 static struct shrinker dcache_shrinker
= {
926 .shrink
= shrink_dcache_memory
,
927 .seeks
= DEFAULT_SEEKS
,
931 * d_alloc - allocate a dcache entry
932 * @parent: parent of entry to allocate
933 * @name: qstr of the name
935 * Allocates a dentry. It returns %NULL if there is insufficient memory
936 * available. On a success the dentry is returned. The name passed in is
937 * copied and the copy passed in may be reused after this call.
940 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
942 struct dentry
*dentry
;
945 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
949 if (name
->len
> DNAME_INLINE_LEN
-1) {
950 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
952 kmem_cache_free(dentry_cache
, dentry
);
956 dname
= dentry
->d_iname
;
958 dentry
->d_name
.name
= dname
;
960 dentry
->d_name
.len
= name
->len
;
961 dentry
->d_name
.hash
= name
->hash
;
962 memcpy(dname
, name
->name
, name
->len
);
963 dname
[name
->len
] = 0;
965 atomic_set(&dentry
->d_count
, 1);
966 dentry
->d_flags
= DCACHE_UNHASHED
;
967 spin_lock_init(&dentry
->d_lock
);
968 dentry
->d_inode
= NULL
;
969 dentry
->d_parent
= NULL
;
972 dentry
->d_fsdata
= NULL
;
973 dentry
->d_mounted
= 0;
974 INIT_HLIST_NODE(&dentry
->d_hash
);
975 INIT_LIST_HEAD(&dentry
->d_lru
);
976 INIT_LIST_HEAD(&dentry
->d_subdirs
);
977 INIT_LIST_HEAD(&dentry
->d_alias
);
980 dentry
->d_parent
= dget(parent
);
981 dentry
->d_sb
= parent
->d_sb
;
983 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
986 spin_lock(&dcache_lock
);
988 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
989 spin_unlock(&dcache_lock
);
991 this_cpu_inc(nr_dentry
);
995 EXPORT_SYMBOL(d_alloc
);
997 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1002 q
.len
= strlen(name
);
1003 q
.hash
= full_name_hash(q
.name
, q
.len
);
1004 return d_alloc(parent
, &q
);
1006 EXPORT_SYMBOL(d_alloc_name
);
1008 /* the caller must hold dcache_lock */
1009 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1012 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1013 dentry
->d_inode
= inode
;
1014 fsnotify_d_instantiate(dentry
, inode
);
1018 * d_instantiate - fill in inode information for a dentry
1019 * @entry: dentry to complete
1020 * @inode: inode to attach to this dentry
1022 * Fill in inode information in the entry.
1024 * This turns negative dentries into productive full members
1027 * NOTE! This assumes that the inode count has been incremented
1028 * (or otherwise set) by the caller to indicate that it is now
1029 * in use by the dcache.
1032 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1034 BUG_ON(!list_empty(&entry
->d_alias
));
1035 spin_lock(&dcache_lock
);
1036 __d_instantiate(entry
, inode
);
1037 spin_unlock(&dcache_lock
);
1038 security_d_instantiate(entry
, inode
);
1040 EXPORT_SYMBOL(d_instantiate
);
1043 * d_instantiate_unique - instantiate a non-aliased dentry
1044 * @entry: dentry to instantiate
1045 * @inode: inode to attach to this dentry
1047 * Fill in inode information in the entry. On success, it returns NULL.
1048 * If an unhashed alias of "entry" already exists, then we return the
1049 * aliased dentry instead and drop one reference to inode.
1051 * Note that in order to avoid conflicts with rename() etc, the caller
1052 * had better be holding the parent directory semaphore.
1054 * This also assumes that the inode count has been incremented
1055 * (or otherwise set) by the caller to indicate that it is now
1056 * in use by the dcache.
1058 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1059 struct inode
*inode
)
1061 struct dentry
*alias
;
1062 int len
= entry
->d_name
.len
;
1063 const char *name
= entry
->d_name
.name
;
1064 unsigned int hash
= entry
->d_name
.hash
;
1067 __d_instantiate(entry
, NULL
);
1071 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1072 struct qstr
*qstr
= &alias
->d_name
;
1074 if (qstr
->hash
!= hash
)
1076 if (alias
->d_parent
!= entry
->d_parent
)
1078 if (qstr
->len
!= len
)
1080 if (memcmp(qstr
->name
, name
, len
))
1086 __d_instantiate(entry
, inode
);
1090 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1092 struct dentry
*result
;
1094 BUG_ON(!list_empty(&entry
->d_alias
));
1096 spin_lock(&dcache_lock
);
1097 result
= __d_instantiate_unique(entry
, inode
);
1098 spin_unlock(&dcache_lock
);
1101 security_d_instantiate(entry
, inode
);
1105 BUG_ON(!d_unhashed(result
));
1110 EXPORT_SYMBOL(d_instantiate_unique
);
1113 * d_alloc_root - allocate root dentry
1114 * @root_inode: inode to allocate the root for
1116 * Allocate a root ("/") dentry for the inode given. The inode is
1117 * instantiated and returned. %NULL is returned if there is insufficient
1118 * memory or the inode passed is %NULL.
1121 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1123 struct dentry
*res
= NULL
;
1126 static const struct qstr name
= { .name
= "/", .len
= 1 };
1128 res
= d_alloc(NULL
, &name
);
1130 res
->d_sb
= root_inode
->i_sb
;
1131 res
->d_parent
= res
;
1132 d_instantiate(res
, root_inode
);
1137 EXPORT_SYMBOL(d_alloc_root
);
1139 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1142 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1143 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1144 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1148 * d_obtain_alias - find or allocate a dentry for a given inode
1149 * @inode: inode to allocate the dentry for
1151 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1152 * similar open by handle operations. The returned dentry may be anonymous,
1153 * or may have a full name (if the inode was already in the cache).
1155 * When called on a directory inode, we must ensure that the inode only ever
1156 * has one dentry. If a dentry is found, that is returned instead of
1157 * allocating a new one.
1159 * On successful return, the reference to the inode has been transferred
1160 * to the dentry. In case of an error the reference on the inode is released.
1161 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1162 * be passed in and will be the error will be propagate to the return value,
1163 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1165 struct dentry
*d_obtain_alias(struct inode
*inode
)
1167 static const struct qstr anonstring
= { .name
= "" };
1172 return ERR_PTR(-ESTALE
);
1174 return ERR_CAST(inode
);
1176 res
= d_find_alias(inode
);
1180 tmp
= d_alloc(NULL
, &anonstring
);
1182 res
= ERR_PTR(-ENOMEM
);
1185 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1187 spin_lock(&dcache_lock
);
1188 res
= __d_find_alias(inode
, 0);
1190 spin_unlock(&dcache_lock
);
1195 /* attach a disconnected dentry */
1196 spin_lock(&tmp
->d_lock
);
1197 tmp
->d_sb
= inode
->i_sb
;
1198 tmp
->d_inode
= inode
;
1199 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1200 tmp
->d_flags
&= ~DCACHE_UNHASHED
;
1201 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1202 hlist_add_head(&tmp
->d_hash
, &inode
->i_sb
->s_anon
);
1203 spin_unlock(&tmp
->d_lock
);
1205 spin_unlock(&dcache_lock
);
1212 EXPORT_SYMBOL(d_obtain_alias
);
1215 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1216 * @inode: the inode which may have a disconnected dentry
1217 * @dentry: a negative dentry which we want to point to the inode.
1219 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1220 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1221 * and return it, else simply d_add the inode to the dentry and return NULL.
1223 * This is needed in the lookup routine of any filesystem that is exportable
1224 * (via knfsd) so that we can build dcache paths to directories effectively.
1226 * If a dentry was found and moved, then it is returned. Otherwise NULL
1227 * is returned. This matches the expected return value of ->lookup.
1230 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1232 struct dentry
*new = NULL
;
1234 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1235 spin_lock(&dcache_lock
);
1236 new = __d_find_alias(inode
, 1);
1238 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1239 spin_unlock(&dcache_lock
);
1240 security_d_instantiate(new, inode
);
1241 d_move(new, dentry
);
1244 /* already taking dcache_lock, so d_add() by hand */
1245 __d_instantiate(dentry
, inode
);
1246 spin_unlock(&dcache_lock
);
1247 security_d_instantiate(dentry
, inode
);
1251 d_add(dentry
, inode
);
1254 EXPORT_SYMBOL(d_splice_alias
);
1257 * d_add_ci - lookup or allocate new dentry with case-exact name
1258 * @inode: the inode case-insensitive lookup has found
1259 * @dentry: the negative dentry that was passed to the parent's lookup func
1260 * @name: the case-exact name to be associated with the returned dentry
1262 * This is to avoid filling the dcache with case-insensitive names to the
1263 * same inode, only the actual correct case is stored in the dcache for
1264 * case-insensitive filesystems.
1266 * For a case-insensitive lookup match and if the the case-exact dentry
1267 * already exists in in the dcache, use it and return it.
1269 * If no entry exists with the exact case name, allocate new dentry with
1270 * the exact case, and return the spliced entry.
1272 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1276 struct dentry
*found
;
1280 * First check if a dentry matching the name already exists,
1281 * if not go ahead and create it now.
1283 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1285 new = d_alloc(dentry
->d_parent
, name
);
1291 found
= d_splice_alias(inode
, new);
1300 * If a matching dentry exists, and it's not negative use it.
1302 * Decrement the reference count to balance the iget() done
1305 if (found
->d_inode
) {
1306 if (unlikely(found
->d_inode
!= inode
)) {
1307 /* This can't happen because bad inodes are unhashed. */
1308 BUG_ON(!is_bad_inode(inode
));
1309 BUG_ON(!is_bad_inode(found
->d_inode
));
1316 * Negative dentry: instantiate it unless the inode is a directory and
1317 * already has a dentry.
1319 spin_lock(&dcache_lock
);
1320 if (!S_ISDIR(inode
->i_mode
) || list_empty(&inode
->i_dentry
)) {
1321 __d_instantiate(found
, inode
);
1322 spin_unlock(&dcache_lock
);
1323 security_d_instantiate(found
, inode
);
1328 * In case a directory already has a (disconnected) entry grab a
1329 * reference to it, move it in place and use it.
1331 new = list_entry(inode
->i_dentry
.next
, struct dentry
, d_alias
);
1333 spin_unlock(&dcache_lock
);
1334 security_d_instantiate(found
, inode
);
1342 return ERR_PTR(error
);
1344 EXPORT_SYMBOL(d_add_ci
);
1347 * d_lookup - search for a dentry
1348 * @parent: parent dentry
1349 * @name: qstr of name we wish to find
1350 * Returns: dentry, or NULL
1352 * d_lookup searches the children of the parent dentry for the name in
1353 * question. If the dentry is found its reference count is incremented and the
1354 * dentry is returned. The caller must use dput to free the entry when it has
1355 * finished using it. %NULL is returned if the dentry does not exist.
1357 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1359 struct dentry
* dentry
= NULL
;
1363 seq
= read_seqbegin(&rename_lock
);
1364 dentry
= __d_lookup(parent
, name
);
1367 } while (read_seqretry(&rename_lock
, seq
));
1370 EXPORT_SYMBOL(d_lookup
);
1373 * __d_lookup - search for a dentry (racy)
1374 * @parent: parent dentry
1375 * @name: qstr of name we wish to find
1376 * Returns: dentry, or NULL
1378 * __d_lookup is like d_lookup, however it may (rarely) return a
1379 * false-negative result due to unrelated rename activity.
1381 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1382 * however it must be used carefully, eg. with a following d_lookup in
1383 * the case of failure.
1385 * __d_lookup callers must be commented.
1387 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1389 unsigned int len
= name
->len
;
1390 unsigned int hash
= name
->hash
;
1391 const unsigned char *str
= name
->name
;
1392 struct hlist_head
*head
= d_hash(parent
,hash
);
1393 struct dentry
*found
= NULL
;
1394 struct hlist_node
*node
;
1395 struct dentry
*dentry
;
1398 * The hash list is protected using RCU.
1400 * Take d_lock when comparing a candidate dentry, to avoid races
1403 * It is possible that concurrent renames can mess up our list
1404 * walk here and result in missing our dentry, resulting in the
1405 * false-negative result. d_lookup() protects against concurrent
1406 * renames using rename_lock seqlock.
1408 * See Documentation/vfs/dcache-locking.txt for more details.
1412 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1415 if (dentry
->d_name
.hash
!= hash
)
1417 if (dentry
->d_parent
!= parent
)
1420 spin_lock(&dentry
->d_lock
);
1423 * Recheck the dentry after taking the lock - d_move may have
1424 * changed things. Don't bother checking the hash because
1425 * we're about to compare the whole name anyway.
1427 if (dentry
->d_parent
!= parent
)
1430 /* non-existing due to RCU? */
1431 if (d_unhashed(dentry
))
1435 * It is safe to compare names since d_move() cannot
1436 * change the qstr (protected by d_lock).
1438 qstr
= &dentry
->d_name
;
1439 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1440 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1441 dentry
, dentry
->d_inode
,
1442 qstr
->len
, qstr
->name
, name
))
1445 if (qstr
->len
!= len
)
1447 if (memcmp(qstr
->name
, str
, len
))
1451 atomic_inc(&dentry
->d_count
);
1453 spin_unlock(&dentry
->d_lock
);
1456 spin_unlock(&dentry
->d_lock
);
1464 * d_hash_and_lookup - hash the qstr then search for a dentry
1465 * @dir: Directory to search in
1466 * @name: qstr of name we wish to find
1468 * On hash failure or on lookup failure NULL is returned.
1470 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1472 struct dentry
*dentry
= NULL
;
1475 * Check for a fs-specific hash function. Note that we must
1476 * calculate the standard hash first, as the d_op->d_hash()
1477 * routine may choose to leave the hash value unchanged.
1479 name
->hash
= full_name_hash(name
->name
, name
->len
);
1480 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1481 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1484 dentry
= d_lookup(dir
, name
);
1490 * d_validate - verify dentry provided from insecure source (deprecated)
1491 * @dentry: The dentry alleged to be valid child of @dparent
1492 * @dparent: The parent dentry (known to be valid)
1494 * An insecure source has sent us a dentry, here we verify it and dget() it.
1495 * This is used by ncpfs in its readdir implementation.
1496 * Zero is returned in the dentry is invalid.
1498 * This function is slow for big directories, and deprecated, do not use it.
1500 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1502 struct dentry
*child
;
1504 spin_lock(&dcache_lock
);
1505 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1506 if (dentry
== child
) {
1507 __dget_locked(dentry
);
1508 spin_unlock(&dcache_lock
);
1512 spin_unlock(&dcache_lock
);
1516 EXPORT_SYMBOL(d_validate
);
1519 * When a file is deleted, we have two options:
1520 * - turn this dentry into a negative dentry
1521 * - unhash this dentry and free it.
1523 * Usually, we want to just turn this into
1524 * a negative dentry, but if anybody else is
1525 * currently using the dentry or the inode
1526 * we can't do that and we fall back on removing
1527 * it from the hash queues and waiting for
1528 * it to be deleted later when it has no users
1532 * d_delete - delete a dentry
1533 * @dentry: The dentry to delete
1535 * Turn the dentry into a negative dentry if possible, otherwise
1536 * remove it from the hash queues so it can be deleted later
1539 void d_delete(struct dentry
* dentry
)
1543 * Are we the only user?
1545 spin_lock(&dcache_lock
);
1546 spin_lock(&dentry
->d_lock
);
1547 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1548 if (atomic_read(&dentry
->d_count
) == 1) {
1549 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1550 dentry_iput(dentry
);
1551 fsnotify_nameremove(dentry
, isdir
);
1555 if (!d_unhashed(dentry
))
1558 spin_unlock(&dentry
->d_lock
);
1559 spin_unlock(&dcache_lock
);
1561 fsnotify_nameremove(dentry
, isdir
);
1563 EXPORT_SYMBOL(d_delete
);
1565 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1568 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1569 hlist_add_head_rcu(&entry
->d_hash
, list
);
1572 static void _d_rehash(struct dentry
* entry
)
1574 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1578 * d_rehash - add an entry back to the hash
1579 * @entry: dentry to add to the hash
1581 * Adds a dentry to the hash according to its name.
1584 void d_rehash(struct dentry
* entry
)
1586 spin_lock(&dcache_lock
);
1587 spin_lock(&entry
->d_lock
);
1589 spin_unlock(&entry
->d_lock
);
1590 spin_unlock(&dcache_lock
);
1592 EXPORT_SYMBOL(d_rehash
);
1595 * dentry_update_name_case - update case insensitive dentry with a new name
1596 * @dentry: dentry to be updated
1599 * Update a case insensitive dentry with new case of name.
1601 * dentry must have been returned by d_lookup with name @name. Old and new
1602 * name lengths must match (ie. no d_compare which allows mismatched name
1605 * Parent inode i_mutex must be held over d_lookup and into this call (to
1606 * keep renames and concurrent inserts, and readdir(2) away).
1608 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
1610 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
1611 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
1613 spin_lock(&dcache_lock
);
1614 spin_lock(&dentry
->d_lock
);
1615 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
1616 spin_unlock(&dentry
->d_lock
);
1617 spin_unlock(&dcache_lock
);
1619 EXPORT_SYMBOL(dentry_update_name_case
);
1622 * When switching names, the actual string doesn't strictly have to
1623 * be preserved in the target - because we're dropping the target
1624 * anyway. As such, we can just do a simple memcpy() to copy over
1625 * the new name before we switch.
1627 * Note that we have to be a lot more careful about getting the hash
1628 * switched - we have to switch the hash value properly even if it
1629 * then no longer matches the actual (corrupted) string of the target.
1630 * The hash value has to match the hash queue that the dentry is on..
1632 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1634 if (dname_external(target
)) {
1635 if (dname_external(dentry
)) {
1637 * Both external: swap the pointers
1639 swap(target
->d_name
.name
, dentry
->d_name
.name
);
1642 * dentry:internal, target:external. Steal target's
1643 * storage and make target internal.
1645 memcpy(target
->d_iname
, dentry
->d_name
.name
,
1646 dentry
->d_name
.len
+ 1);
1647 dentry
->d_name
.name
= target
->d_name
.name
;
1648 target
->d_name
.name
= target
->d_iname
;
1651 if (dname_external(dentry
)) {
1653 * dentry:external, target:internal. Give dentry's
1654 * storage to target and make dentry internal
1656 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1657 target
->d_name
.len
+ 1);
1658 target
->d_name
.name
= dentry
->d_name
.name
;
1659 dentry
->d_name
.name
= dentry
->d_iname
;
1662 * Both are internal. Just copy target to dentry
1664 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1665 target
->d_name
.len
+ 1);
1666 dentry
->d_name
.len
= target
->d_name
.len
;
1670 swap(dentry
->d_name
.len
, target
->d_name
.len
);
1674 * We cannibalize "target" when moving dentry on top of it,
1675 * because it's going to be thrown away anyway. We could be more
1676 * polite about it, though.
1678 * This forceful removal will result in ugly /proc output if
1679 * somebody holds a file open that got deleted due to a rename.
1680 * We could be nicer about the deleted file, and let it show
1681 * up under the name it had before it was deleted rather than
1682 * under the original name of the file that was moved on top of it.
1686 * d_move_locked - move a dentry
1687 * @dentry: entry to move
1688 * @target: new dentry
1690 * Update the dcache to reflect the move of a file name. Negative
1691 * dcache entries should not be moved in this way.
1693 static void d_move_locked(struct dentry
* dentry
, struct dentry
* target
)
1695 struct hlist_head
*list
;
1697 if (!dentry
->d_inode
)
1698 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1700 write_seqlock(&rename_lock
);
1702 * XXXX: do we really need to take target->d_lock?
1704 if (target
< dentry
) {
1705 spin_lock(&target
->d_lock
);
1706 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1708 spin_lock(&dentry
->d_lock
);
1709 spin_lock_nested(&target
->d_lock
, DENTRY_D_LOCK_NESTED
);
1712 /* Move the dentry to the target hash queue, if on different bucket */
1713 if (d_unhashed(dentry
))
1714 goto already_unhashed
;
1716 hlist_del_rcu(&dentry
->d_hash
);
1719 list
= d_hash(target
->d_parent
, target
->d_name
.hash
);
1720 __d_rehash(dentry
, list
);
1722 /* Unhash the target: dput() will then get rid of it */
1725 list_del(&dentry
->d_u
.d_child
);
1726 list_del(&target
->d_u
.d_child
);
1728 /* Switch the names.. */
1729 switch_names(dentry
, target
);
1730 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
1732 /* ... and switch the parents */
1733 if (IS_ROOT(dentry
)) {
1734 dentry
->d_parent
= target
->d_parent
;
1735 target
->d_parent
= target
;
1736 INIT_LIST_HEAD(&target
->d_u
.d_child
);
1738 swap(dentry
->d_parent
, target
->d_parent
);
1740 /* And add them back to the (new) parent lists */
1741 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
1744 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1745 spin_unlock(&target
->d_lock
);
1746 fsnotify_d_move(dentry
);
1747 spin_unlock(&dentry
->d_lock
);
1748 write_sequnlock(&rename_lock
);
1752 * d_move - move a dentry
1753 * @dentry: entry to move
1754 * @target: new dentry
1756 * Update the dcache to reflect the move of a file name. Negative
1757 * dcache entries should not be moved in this way.
1760 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1762 spin_lock(&dcache_lock
);
1763 d_move_locked(dentry
, target
);
1764 spin_unlock(&dcache_lock
);
1766 EXPORT_SYMBOL(d_move
);
1769 * d_ancestor - search for an ancestor
1770 * @p1: ancestor dentry
1773 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1774 * an ancestor of p2, else NULL.
1776 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
1780 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
1781 if (p
->d_parent
== p1
)
1788 * This helper attempts to cope with remotely renamed directories
1790 * It assumes that the caller is already holding
1791 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1793 * Note: If ever the locking in lock_rename() changes, then please
1794 * remember to update this too...
1796 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
1797 __releases(dcache_lock
)
1799 struct mutex
*m1
= NULL
, *m2
= NULL
;
1802 /* If alias and dentry share a parent, then no extra locks required */
1803 if (alias
->d_parent
== dentry
->d_parent
)
1806 /* Check for loops */
1807 ret
= ERR_PTR(-ELOOP
);
1808 if (d_ancestor(alias
, dentry
))
1811 /* See lock_rename() */
1812 ret
= ERR_PTR(-EBUSY
);
1813 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
1815 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
1816 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
1818 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
1820 d_move_locked(alias
, dentry
);
1823 spin_unlock(&dcache_lock
);
1832 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1833 * named dentry in place of the dentry to be replaced.
1835 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
1837 struct dentry
*dparent
, *aparent
;
1839 switch_names(dentry
, anon
);
1840 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
1842 dparent
= dentry
->d_parent
;
1843 aparent
= anon
->d_parent
;
1845 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
1846 list_del(&dentry
->d_u
.d_child
);
1847 if (!IS_ROOT(dentry
))
1848 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1850 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1852 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
1853 list_del(&anon
->d_u
.d_child
);
1855 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
1857 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
1859 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
1863 * d_materialise_unique - introduce an inode into the tree
1864 * @dentry: candidate dentry
1865 * @inode: inode to bind to the dentry, to which aliases may be attached
1867 * Introduces an dentry into the tree, substituting an extant disconnected
1868 * root directory alias in its place if there is one
1870 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
1872 struct dentry
*actual
;
1874 BUG_ON(!d_unhashed(dentry
));
1876 spin_lock(&dcache_lock
);
1880 __d_instantiate(dentry
, NULL
);
1884 if (S_ISDIR(inode
->i_mode
)) {
1885 struct dentry
*alias
;
1887 /* Does an aliased dentry already exist? */
1888 alias
= __d_find_alias(inode
, 0);
1891 /* Is this an anonymous mountpoint that we could splice
1893 if (IS_ROOT(alias
)) {
1894 spin_lock(&alias
->d_lock
);
1895 __d_materialise_dentry(dentry
, alias
);
1899 /* Nope, but we must(!) avoid directory aliasing */
1900 actual
= __d_unalias(dentry
, alias
);
1907 /* Add a unique reference */
1908 actual
= __d_instantiate_unique(dentry
, inode
);
1911 else if (unlikely(!d_unhashed(actual
)))
1912 goto shouldnt_be_hashed
;
1915 spin_lock(&actual
->d_lock
);
1918 spin_unlock(&actual
->d_lock
);
1919 spin_unlock(&dcache_lock
);
1921 if (actual
== dentry
) {
1922 security_d_instantiate(dentry
, inode
);
1930 spin_unlock(&dcache_lock
);
1933 EXPORT_SYMBOL_GPL(d_materialise_unique
);
1935 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
1939 return -ENAMETOOLONG
;
1941 memcpy(*buffer
, str
, namelen
);
1945 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
1947 return prepend(buffer
, buflen
, name
->name
, name
->len
);
1951 * Prepend path string to a buffer
1953 * @path: the dentry/vfsmount to report
1954 * @root: root vfsmnt/dentry (may be modified by this function)
1955 * @buffer: pointer to the end of the buffer
1956 * @buflen: pointer to buffer length
1958 * Caller holds the dcache_lock.
1960 * If path is not reachable from the supplied root, then the value of
1961 * root is changed (without modifying refcounts).
1963 static int prepend_path(const struct path
*path
, struct path
*root
,
1964 char **buffer
, int *buflen
)
1966 struct dentry
*dentry
= path
->dentry
;
1967 struct vfsmount
*vfsmnt
= path
->mnt
;
1971 br_read_lock(vfsmount_lock
);
1972 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
1973 struct dentry
* parent
;
1975 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
1977 if (vfsmnt
->mnt_parent
== vfsmnt
) {
1980 dentry
= vfsmnt
->mnt_mountpoint
;
1981 vfsmnt
= vfsmnt
->mnt_parent
;
1984 parent
= dentry
->d_parent
;
1986 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
1988 error
= prepend(buffer
, buflen
, "/", 1);
1997 if (!error
&& !slash
)
1998 error
= prepend(buffer
, buflen
, "/", 1);
2000 br_read_unlock(vfsmount_lock
);
2005 * Filesystems needing to implement special "root names"
2006 * should do so with ->d_dname()
2008 if (IS_ROOT(dentry
) &&
2009 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2010 WARN(1, "Root dentry has weird name <%.*s>\n",
2011 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2014 root
->dentry
= dentry
;
2019 * __d_path - return the path of a dentry
2020 * @path: the dentry/vfsmount to report
2021 * @root: root vfsmnt/dentry (may be modified by this function)
2022 * @buf: buffer to return value in
2023 * @buflen: buffer length
2025 * Convert a dentry into an ASCII path name.
2027 * Returns a pointer into the buffer or an error code if the
2028 * path was too long.
2030 * "buflen" should be positive.
2032 * If path is not reachable from the supplied root, then the value of
2033 * root is changed (without modifying refcounts).
2035 char *__d_path(const struct path
*path
, struct path
*root
,
2036 char *buf
, int buflen
)
2038 char *res
= buf
+ buflen
;
2041 prepend(&res
, &buflen
, "\0", 1);
2042 spin_lock(&dcache_lock
);
2043 error
= prepend_path(path
, root
, &res
, &buflen
);
2044 spin_unlock(&dcache_lock
);
2047 return ERR_PTR(error
);
2052 * same as __d_path but appends "(deleted)" for unlinked files.
2054 static int path_with_deleted(const struct path
*path
, struct path
*root
,
2055 char **buf
, int *buflen
)
2057 prepend(buf
, buflen
, "\0", 1);
2058 if (d_unlinked(path
->dentry
)) {
2059 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2064 return prepend_path(path
, root
, buf
, buflen
);
2067 static int prepend_unreachable(char **buffer
, int *buflen
)
2069 return prepend(buffer
, buflen
, "(unreachable)", 13);
2073 * d_path - return the path of a dentry
2074 * @path: path to report
2075 * @buf: buffer to return value in
2076 * @buflen: buffer length
2078 * Convert a dentry into an ASCII path name. If the entry has been deleted
2079 * the string " (deleted)" is appended. Note that this is ambiguous.
2081 * Returns a pointer into the buffer or an error code if the path was
2082 * too long. Note: Callers should use the returned pointer, not the passed
2083 * in buffer, to use the name! The implementation often starts at an offset
2084 * into the buffer, and may leave 0 bytes at the start.
2086 * "buflen" should be positive.
2088 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2090 char *res
= buf
+ buflen
;
2096 * We have various synthetic filesystems that never get mounted. On
2097 * these filesystems dentries are never used for lookup purposes, and
2098 * thus don't need to be hashed. They also don't need a name until a
2099 * user wants to identify the object in /proc/pid/fd/. The little hack
2100 * below allows us to generate a name for these objects on demand:
2102 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2103 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2105 get_fs_root(current
->fs
, &root
);
2106 spin_lock(&dcache_lock
);
2108 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2110 res
= ERR_PTR(error
);
2111 spin_unlock(&dcache_lock
);
2115 EXPORT_SYMBOL(d_path
);
2118 * d_path_with_unreachable - return the path of a dentry
2119 * @path: path to report
2120 * @buf: buffer to return value in
2121 * @buflen: buffer length
2123 * The difference from d_path() is that this prepends "(unreachable)"
2124 * to paths which are unreachable from the current process' root.
2126 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2128 char *res
= buf
+ buflen
;
2133 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2134 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2136 get_fs_root(current
->fs
, &root
);
2137 spin_lock(&dcache_lock
);
2139 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2140 if (!error
&& !path_equal(&tmp
, &root
))
2141 error
= prepend_unreachable(&res
, &buflen
);
2142 spin_unlock(&dcache_lock
);
2145 res
= ERR_PTR(error
);
2151 * Helper function for dentry_operations.d_dname() members
2153 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2154 const char *fmt
, ...)
2160 va_start(args
, fmt
);
2161 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2164 if (sz
> sizeof(temp
) || sz
> buflen
)
2165 return ERR_PTR(-ENAMETOOLONG
);
2167 buffer
+= buflen
- sz
;
2168 return memcpy(buffer
, temp
, sz
);
2172 * Write full pathname from the root of the filesystem into the buffer.
2174 char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2176 char *end
= buf
+ buflen
;
2179 prepend(&end
, &buflen
, "\0", 1);
2186 while (!IS_ROOT(dentry
)) {
2187 struct dentry
*parent
= dentry
->d_parent
;
2190 if ((prepend_name(&end
, &buflen
, &dentry
->d_name
) != 0) ||
2191 (prepend(&end
, &buflen
, "/", 1) != 0))
2199 return ERR_PTR(-ENAMETOOLONG
);
2201 EXPORT_SYMBOL(__dentry_path
);
2203 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2208 spin_lock(&dcache_lock
);
2209 if (d_unlinked(dentry
)) {
2211 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2215 retval
= __dentry_path(dentry
, buf
, buflen
);
2216 spin_unlock(&dcache_lock
);
2217 if (!IS_ERR(retval
) && p
)
2218 *p
= '/'; /* restore '/' overriden with '\0' */
2221 spin_unlock(&dcache_lock
);
2222 return ERR_PTR(-ENAMETOOLONG
);
2226 * NOTE! The user-level library version returns a
2227 * character pointer. The kernel system call just
2228 * returns the length of the buffer filled (which
2229 * includes the ending '\0' character), or a negative
2230 * error value. So libc would do something like
2232 * char *getcwd(char * buf, size_t size)
2236 * retval = sys_getcwd(buf, size);
2243 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2246 struct path pwd
, root
;
2247 char *page
= (char *) __get_free_page(GFP_USER
);
2252 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2255 spin_lock(&dcache_lock
);
2256 if (!d_unlinked(pwd
.dentry
)) {
2258 struct path tmp
= root
;
2259 char *cwd
= page
+ PAGE_SIZE
;
2260 int buflen
= PAGE_SIZE
;
2262 prepend(&cwd
, &buflen
, "\0", 1);
2263 error
= prepend_path(&pwd
, &tmp
, &cwd
, &buflen
);
2264 spin_unlock(&dcache_lock
);
2269 /* Unreachable from current root */
2270 if (!path_equal(&tmp
, &root
)) {
2271 error
= prepend_unreachable(&cwd
, &buflen
);
2277 len
= PAGE_SIZE
+ page
- cwd
;
2280 if (copy_to_user(buf
, cwd
, len
))
2284 spin_unlock(&dcache_lock
);
2289 free_page((unsigned long) page
);
2294 * Test whether new_dentry is a subdirectory of old_dentry.
2296 * Trivially implemented using the dcache structure
2300 * is_subdir - is new dentry a subdirectory of old_dentry
2301 * @new_dentry: new dentry
2302 * @old_dentry: old dentry
2304 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2305 * Returns 0 otherwise.
2306 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2309 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2314 if (new_dentry
== old_dentry
)
2318 * Need rcu_readlock to protect against the d_parent trashing
2323 /* for restarting inner loop in case of seq retry */
2324 seq
= read_seqbegin(&rename_lock
);
2325 if (d_ancestor(old_dentry
, new_dentry
))
2329 } while (read_seqretry(&rename_lock
, seq
));
2335 int path_is_under(struct path
*path1
, struct path
*path2
)
2337 struct vfsmount
*mnt
= path1
->mnt
;
2338 struct dentry
*dentry
= path1
->dentry
;
2341 br_read_lock(vfsmount_lock
);
2342 if (mnt
!= path2
->mnt
) {
2344 if (mnt
->mnt_parent
== mnt
) {
2345 br_read_unlock(vfsmount_lock
);
2348 if (mnt
->mnt_parent
== path2
->mnt
)
2350 mnt
= mnt
->mnt_parent
;
2352 dentry
= mnt
->mnt_mountpoint
;
2354 res
= is_subdir(dentry
, path2
->dentry
);
2355 br_read_unlock(vfsmount_lock
);
2358 EXPORT_SYMBOL(path_is_under
);
2360 void d_genocide(struct dentry
*root
)
2362 struct dentry
*this_parent
= root
;
2363 struct list_head
*next
;
2365 spin_lock(&dcache_lock
);
2367 next
= this_parent
->d_subdirs
.next
;
2369 while (next
!= &this_parent
->d_subdirs
) {
2370 struct list_head
*tmp
= next
;
2371 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2373 if (d_unhashed(dentry
)||!dentry
->d_inode
)
2375 if (!list_empty(&dentry
->d_subdirs
)) {
2376 this_parent
= dentry
;
2379 atomic_dec(&dentry
->d_count
);
2381 if (this_parent
!= root
) {
2382 next
= this_parent
->d_u
.d_child
.next
;
2383 atomic_dec(&this_parent
->d_count
);
2384 this_parent
= this_parent
->d_parent
;
2387 spin_unlock(&dcache_lock
);
2391 * find_inode_number - check for dentry with name
2392 * @dir: directory to check
2393 * @name: Name to find.
2395 * Check whether a dentry already exists for the given name,
2396 * and return the inode number if it has an inode. Otherwise
2399 * This routine is used to post-process directory listings for
2400 * filesystems using synthetic inode numbers, and is necessary
2401 * to keep getcwd() working.
2404 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2406 struct dentry
* dentry
;
2409 dentry
= d_hash_and_lookup(dir
, name
);
2411 if (dentry
->d_inode
)
2412 ino
= dentry
->d_inode
->i_ino
;
2417 EXPORT_SYMBOL(find_inode_number
);
2419 static __initdata
unsigned long dhash_entries
;
2420 static int __init
set_dhash_entries(char *str
)
2424 dhash_entries
= simple_strtoul(str
, &str
, 0);
2427 __setup("dhash_entries=", set_dhash_entries
);
2429 static void __init
dcache_init_early(void)
2433 /* If hashes are distributed across NUMA nodes, defer
2434 * hash allocation until vmalloc space is available.
2440 alloc_large_system_hash("Dentry cache",
2441 sizeof(struct hlist_head
),
2449 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2450 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2453 static void __init
dcache_init(void)
2458 * A constructor could be added for stable state like the lists,
2459 * but it is probably not worth it because of the cache nature
2462 dentry_cache
= KMEM_CACHE(dentry
,
2463 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2465 register_shrinker(&dcache_shrinker
);
2467 /* Hash may have been set up in dcache_init_early */
2472 alloc_large_system_hash("Dentry cache",
2473 sizeof(struct hlist_head
),
2481 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2482 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2485 /* SLAB cache for __getname() consumers */
2486 struct kmem_cache
*names_cachep __read_mostly
;
2487 EXPORT_SYMBOL(names_cachep
);
2489 EXPORT_SYMBOL(d_genocide
);
2491 void __init
vfs_caches_init_early(void)
2493 dcache_init_early();
2497 void __init
vfs_caches_init(unsigned long mempages
)
2499 unsigned long reserve
;
2501 /* Base hash sizes on available memory, with a reserve equal to
2502 150% of current kernel size */
2504 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2505 mempages
-= reserve
;
2507 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2508 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2512 files_init(mempages
);