4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
38 int sysctl_vfs_cache_pressure __read_mostly
= 100;
39 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
41 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lock
);
42 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
44 EXPORT_SYMBOL(dcache_lock
);
46 static struct kmem_cache
*dentry_cache __read_mostly
;
48 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
58 #define D_HASHBITS d_hash_shift
59 #define D_HASHMASK d_hash_mask
61 static unsigned int d_hash_mask __read_mostly
;
62 static unsigned int d_hash_shift __read_mostly
;
63 static struct hlist_head
*dentry_hashtable __read_mostly
;
65 /* Statistics gathering. */
66 struct dentry_stat_t dentry_stat
= {
70 static void __d_free(struct dentry
*dentry
)
72 WARN_ON(!list_empty(&dentry
->d_alias
));
73 if (dname_external(dentry
))
74 kfree(dentry
->d_name
.name
);
75 kmem_cache_free(dentry_cache
, dentry
);
78 static void d_callback(struct rcu_head
*head
)
80 struct dentry
* dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
85 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
88 static void d_free(struct dentry
*dentry
)
90 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
91 dentry
->d_op
->d_release(dentry
);
92 /* if dentry was never inserted into hash, immediate free is OK */
93 if (hlist_unhashed(&dentry
->d_hash
))
96 call_rcu(&dentry
->d_u
.d_rcu
, d_callback
);
100 * Release the dentry's inode, using the filesystem
101 * d_iput() operation if defined.
103 static void dentry_iput(struct dentry
* dentry
)
104 __releases(dentry
->d_lock
)
105 __releases(dcache_lock
)
107 struct inode
*inode
= dentry
->d_inode
;
109 dentry
->d_inode
= NULL
;
110 list_del_init(&dentry
->d_alias
);
111 spin_unlock(&dentry
->d_lock
);
112 spin_unlock(&dcache_lock
);
114 fsnotify_inoderemove(inode
);
115 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
116 dentry
->d_op
->d_iput(dentry
, inode
);
120 spin_unlock(&dentry
->d_lock
);
121 spin_unlock(&dcache_lock
);
126 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
128 static void dentry_lru_add(struct dentry
*dentry
)
130 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
131 dentry
->d_sb
->s_nr_dentry_unused
++;
132 dentry_stat
.nr_unused
++;
135 static void dentry_lru_add_tail(struct dentry
*dentry
)
137 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
138 dentry
->d_sb
->s_nr_dentry_unused
++;
139 dentry_stat
.nr_unused
++;
142 static void dentry_lru_del(struct dentry
*dentry
)
144 if (!list_empty(&dentry
->d_lru
)) {
145 list_del(&dentry
->d_lru
);
146 dentry
->d_sb
->s_nr_dentry_unused
--;
147 dentry_stat
.nr_unused
--;
151 static void dentry_lru_del_init(struct dentry
*dentry
)
153 if (likely(!list_empty(&dentry
->d_lru
))) {
154 list_del_init(&dentry
->d_lru
);
155 dentry
->d_sb
->s_nr_dentry_unused
--;
156 dentry_stat
.nr_unused
--;
161 * d_kill - kill dentry and return parent
162 * @dentry: dentry to kill
164 * The dentry must already be unhashed and removed from the LRU.
166 * If this is the root of the dentry tree, return NULL.
168 static struct dentry
*d_kill(struct dentry
*dentry
)
169 __releases(dentry
->d_lock
)
170 __releases(dcache_lock
)
172 struct dentry
*parent
;
174 list_del(&dentry
->d_u
.d_child
);
175 dentry_stat
.nr_dentry
--; /* For d_free, below */
176 /*drops the locks, at that point nobody can reach this dentry */
181 parent
= dentry
->d_parent
;
189 * This is complicated by the fact that we do not want to put
190 * dentries that are no longer on any hash chain on the unused
191 * list: we'd much rather just get rid of them immediately.
193 * However, that implies that we have to traverse the dentry
194 * tree upwards to the parents which might _also_ now be
195 * scheduled for deletion (it may have been only waiting for
196 * its last child to go away).
198 * This tail recursion is done by hand as we don't want to depend
199 * on the compiler to always get this right (gcc generally doesn't).
200 * Real recursion would eat up our stack space.
204 * dput - release a dentry
205 * @dentry: dentry to release
207 * Release a dentry. This will drop the usage count and if appropriate
208 * call the dentry unlink method as well as removing it from the queues and
209 * releasing its resources. If the parent dentries were scheduled for release
210 * they too may now get deleted.
212 * no dcache lock, please.
215 void dput(struct dentry
*dentry
)
220 WARN_ON_ONCE(!atomic_read(&dentry
->d_count
));
222 if (atomic_read(&dentry
->d_count
) == 1)
224 if (!atomic_dec_and_lock(&dentry
->d_count
, &dcache_lock
))
227 spin_lock(&dentry
->d_lock
);
228 if (atomic_read(&dentry
->d_count
)) {
229 spin_unlock(&dentry
->d_lock
);
230 spin_unlock(&dcache_lock
);
235 * AV: ->d_delete() is _NOT_ allowed to block now.
237 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
238 if (dentry
->d_op
->d_delete(dentry
))
241 /* Unreachable? Get rid of it */
242 if (d_unhashed(dentry
))
244 if (list_empty(&dentry
->d_lru
)) {
245 dentry
->d_flags
|= DCACHE_REFERENCED
;
246 dentry_lru_add(dentry
);
248 spin_unlock(&dentry
->d_lock
);
249 spin_unlock(&dcache_lock
);
255 /* if dentry was on the d_lru list delete it from there */
256 dentry_lru_del(dentry
);
257 dentry
= d_kill(dentry
);
263 * d_invalidate - invalidate a dentry
264 * @dentry: dentry to invalidate
266 * Try to invalidate the dentry if it turns out to be
267 * possible. If there are other dentries that can be
268 * reached through this one we can't delete it and we
269 * return -EBUSY. On success we return 0.
274 int d_invalidate(struct dentry
* dentry
)
277 * If it's already been dropped, return OK.
279 spin_lock(&dcache_lock
);
280 if (d_unhashed(dentry
)) {
281 spin_unlock(&dcache_lock
);
285 * Check whether to do a partial shrink_dcache
286 * to get rid of unused child entries.
288 if (!list_empty(&dentry
->d_subdirs
)) {
289 spin_unlock(&dcache_lock
);
290 shrink_dcache_parent(dentry
);
291 spin_lock(&dcache_lock
);
295 * Somebody else still using it?
297 * If it's a directory, we can't drop it
298 * for fear of somebody re-populating it
299 * with children (even though dropping it
300 * would make it unreachable from the root,
301 * we might still populate it if it was a
302 * working directory or similar).
304 spin_lock(&dentry
->d_lock
);
305 if (atomic_read(&dentry
->d_count
) > 1) {
306 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
307 spin_unlock(&dentry
->d_lock
);
308 spin_unlock(&dcache_lock
);
314 spin_unlock(&dentry
->d_lock
);
315 spin_unlock(&dcache_lock
);
319 /* This should be called _only_ with dcache_lock held */
321 static inline struct dentry
* __dget_locked(struct dentry
*dentry
)
323 atomic_inc(&dentry
->d_count
);
324 dentry_lru_del_init(dentry
);
328 struct dentry
* dget_locked(struct dentry
*dentry
)
330 return __dget_locked(dentry
);
334 * d_find_alias - grab a hashed alias of inode
335 * @inode: inode in question
336 * @want_discon: flag, used by d_splice_alias, to request
337 * that only a DISCONNECTED alias be returned.
339 * If inode has a hashed alias, or is a directory and has any alias,
340 * acquire the reference to alias and return it. Otherwise return NULL.
341 * Notice that if inode is a directory there can be only one alias and
342 * it can be unhashed only if it has no children, or if it is the root
345 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
346 * any other hashed alias over that one unless @want_discon is set,
347 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
350 static struct dentry
* __d_find_alias(struct inode
*inode
, int want_discon
)
352 struct list_head
*head
, *next
, *tmp
;
353 struct dentry
*alias
, *discon_alias
=NULL
;
355 head
= &inode
->i_dentry
;
356 next
= inode
->i_dentry
.next
;
357 while (next
!= head
) {
361 alias
= list_entry(tmp
, struct dentry
, d_alias
);
362 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
363 if (IS_ROOT(alias
) &&
364 (alias
->d_flags
& DCACHE_DISCONNECTED
))
365 discon_alias
= alias
;
366 else if (!want_discon
) {
367 __dget_locked(alias
);
373 __dget_locked(discon_alias
);
377 struct dentry
* d_find_alias(struct inode
*inode
)
379 struct dentry
*de
= NULL
;
381 if (!list_empty(&inode
->i_dentry
)) {
382 spin_lock(&dcache_lock
);
383 de
= __d_find_alias(inode
, 0);
384 spin_unlock(&dcache_lock
);
390 * Try to kill dentries associated with this inode.
391 * WARNING: you must own a reference to inode.
393 void d_prune_aliases(struct inode
*inode
)
395 struct dentry
*dentry
;
397 spin_lock(&dcache_lock
);
398 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
399 spin_lock(&dentry
->d_lock
);
400 if (!atomic_read(&dentry
->d_count
)) {
401 __dget_locked(dentry
);
403 spin_unlock(&dentry
->d_lock
);
404 spin_unlock(&dcache_lock
);
408 spin_unlock(&dentry
->d_lock
);
410 spin_unlock(&dcache_lock
);
414 * Throw away a dentry - free the inode, dput the parent. This requires that
415 * the LRU list has already been removed.
417 * Try to prune ancestors as well. This is necessary to prevent
418 * quadratic behavior of shrink_dcache_parent(), but is also expected
419 * to be beneficial in reducing dentry cache fragmentation.
421 static void prune_one_dentry(struct dentry
* dentry
)
422 __releases(dentry
->d_lock
)
423 __releases(dcache_lock
)
424 __acquires(dcache_lock
)
427 dentry
= d_kill(dentry
);
430 * Prune ancestors. Locking is simpler than in dput(),
431 * because dcache_lock needs to be taken anyway.
433 spin_lock(&dcache_lock
);
435 if (!atomic_dec_and_lock(&dentry
->d_count
, &dentry
->d_lock
))
438 if (dentry
->d_op
&& dentry
->d_op
->d_delete
)
439 dentry
->d_op
->d_delete(dentry
);
440 dentry_lru_del_init(dentry
);
442 dentry
= d_kill(dentry
);
443 spin_lock(&dcache_lock
);
448 * Shrink the dentry LRU on a given superblock.
449 * @sb : superblock to shrink dentry LRU.
450 * @count: If count is NULL, we prune all dentries on superblock.
451 * @flags: If flags is non-zero, we need to do special processing based on
452 * which flags are set. This means we don't need to maintain multiple
453 * similar copies of this loop.
455 static void __shrink_dcache_sb(struct super_block
*sb
, int *count
, int flags
)
457 LIST_HEAD(referenced
);
459 struct dentry
*dentry
;
463 BUG_ON((flags
& DCACHE_REFERENCED
) && count
== NULL
);
464 spin_lock(&dcache_lock
);
466 /* called from prune_dcache() and shrink_dcache_parent() */
470 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
472 while (!list_empty(&sb
->s_dentry_lru
)) {
473 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
474 struct dentry
, d_lru
);
475 BUG_ON(dentry
->d_sb
!= sb
);
477 spin_lock(&dentry
->d_lock
);
479 * If we are honouring the DCACHE_REFERENCED flag and
480 * the dentry has this flag set, don't free it. Clear
481 * the flag and put it back on the LRU.
483 if ((flags
& DCACHE_REFERENCED
)
484 && (dentry
->d_flags
& DCACHE_REFERENCED
)) {
485 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
486 list_move(&dentry
->d_lru
, &referenced
);
487 spin_unlock(&dentry
->d_lock
);
489 list_move_tail(&dentry
->d_lru
, &tmp
);
490 spin_unlock(&dentry
->d_lock
);
495 cond_resched_lock(&dcache_lock
);
498 while (!list_empty(&tmp
)) {
499 dentry
= list_entry(tmp
.prev
, struct dentry
, d_lru
);
500 dentry_lru_del_init(dentry
);
501 spin_lock(&dentry
->d_lock
);
503 * We found an inuse dentry which was not removed from
504 * the LRU because of laziness during lookup. Do not free
505 * it - just keep it off the LRU list.
507 if (atomic_read(&dentry
->d_count
)) {
508 spin_unlock(&dentry
->d_lock
);
511 prune_one_dentry(dentry
);
512 /* dentry->d_lock was dropped in prune_one_dentry() */
513 cond_resched_lock(&dcache_lock
);
515 if (count
== NULL
&& !list_empty(&sb
->s_dentry_lru
))
519 if (!list_empty(&referenced
))
520 list_splice(&referenced
, &sb
->s_dentry_lru
);
521 spin_unlock(&dcache_lock
);
525 * prune_dcache - shrink the dcache
526 * @count: number of entries to try to free
528 * Shrink the dcache. This is done when we need more memory, or simply when we
529 * need to unmount something (at which point we need to unuse all dentries).
531 * This function may fail to free any resources if all the dentries are in use.
533 static void prune_dcache(int count
)
535 struct super_block
*sb
;
537 int unused
= dentry_stat
.nr_unused
;
541 if (unused
== 0 || count
== 0)
543 spin_lock(&dcache_lock
);
548 prune_ratio
= unused
/ count
;
550 list_for_each_entry(sb
, &super_blocks
, s_list
) {
551 if (sb
->s_nr_dentry_unused
== 0)
554 /* Now, we reclaim unused dentrins with fairness.
555 * We reclaim them same percentage from each superblock.
556 * We calculate number of dentries to scan on this sb
557 * as follows, but the implementation is arranged to avoid
559 * number of dentries to scan on this sb =
560 * count * (number of dentries on this sb /
561 * number of dentries in the machine)
563 spin_unlock(&sb_lock
);
564 if (prune_ratio
!= 1)
565 w_count
= (sb
->s_nr_dentry_unused
/ prune_ratio
) + 1;
567 w_count
= sb
->s_nr_dentry_unused
;
570 * We need to be sure this filesystem isn't being unmounted,
571 * otherwise we could race with generic_shutdown_super(), and
572 * end up holding a reference to an inode while the filesystem
573 * is unmounted. So we try to get s_umount, and make sure
576 if (down_read_trylock(&sb
->s_umount
)) {
577 if ((sb
->s_root
!= NULL
) &&
578 (!list_empty(&sb
->s_dentry_lru
))) {
579 spin_unlock(&dcache_lock
);
580 __shrink_dcache_sb(sb
, &w_count
,
583 spin_lock(&dcache_lock
);
585 up_read(&sb
->s_umount
);
590 * restart only when sb is no longer on the list and
591 * we have more work to do.
593 if (__put_super_and_need_restart(sb
) && count
> 0) {
594 spin_unlock(&sb_lock
);
598 spin_unlock(&sb_lock
);
599 spin_unlock(&dcache_lock
);
603 * shrink_dcache_sb - shrink dcache for a superblock
606 * Shrink the dcache for the specified super block. This
607 * is used to free the dcache before unmounting a file
610 void shrink_dcache_sb(struct super_block
* sb
)
612 __shrink_dcache_sb(sb
, NULL
, 0);
616 * destroy a single subtree of dentries for unmount
617 * - see the comments on shrink_dcache_for_umount() for a description of the
620 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
622 struct dentry
*parent
;
623 unsigned detached
= 0;
625 BUG_ON(!IS_ROOT(dentry
));
627 /* detach this root from the system */
628 spin_lock(&dcache_lock
);
629 dentry_lru_del_init(dentry
);
631 spin_unlock(&dcache_lock
);
634 /* descend to the first leaf in the current subtree */
635 while (!list_empty(&dentry
->d_subdirs
)) {
638 /* this is a branch with children - detach all of them
639 * from the system in one go */
640 spin_lock(&dcache_lock
);
641 list_for_each_entry(loop
, &dentry
->d_subdirs
,
643 dentry_lru_del_init(loop
);
645 cond_resched_lock(&dcache_lock
);
647 spin_unlock(&dcache_lock
);
649 /* move to the first child */
650 dentry
= list_entry(dentry
->d_subdirs
.next
,
651 struct dentry
, d_u
.d_child
);
654 /* consume the dentries from this leaf up through its parents
655 * until we find one with children or run out altogether */
659 if (atomic_read(&dentry
->d_count
) != 0) {
661 "BUG: Dentry %p{i=%lx,n=%s}"
663 " [unmount of %s %s]\n",
666 dentry
->d_inode
->i_ino
: 0UL,
668 atomic_read(&dentry
->d_count
),
669 dentry
->d_sb
->s_type
->name
,
677 parent
= dentry
->d_parent
;
678 atomic_dec(&parent
->d_count
);
681 list_del(&dentry
->d_u
.d_child
);
684 inode
= dentry
->d_inode
;
686 dentry
->d_inode
= NULL
;
687 list_del_init(&dentry
->d_alias
);
688 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
689 dentry
->d_op
->d_iput(dentry
, inode
);
696 /* finished when we fall off the top of the tree,
697 * otherwise we ascend to the parent and move to the
698 * next sibling if there is one */
704 } while (list_empty(&dentry
->d_subdirs
));
706 dentry
= list_entry(dentry
->d_subdirs
.next
,
707 struct dentry
, d_u
.d_child
);
710 /* several dentries were freed, need to correct nr_dentry */
711 spin_lock(&dcache_lock
);
712 dentry_stat
.nr_dentry
-= detached
;
713 spin_unlock(&dcache_lock
);
717 * destroy the dentries attached to a superblock on unmounting
718 * - we don't need to use dentry->d_lock, and only need dcache_lock when
719 * removing the dentry from the system lists and hashes because:
720 * - the superblock is detached from all mountings and open files, so the
721 * dentry trees will not be rearranged by the VFS
722 * - s_umount is write-locked, so the memory pressure shrinker will ignore
723 * any dentries belonging to this superblock that it comes across
724 * - the filesystem itself is no longer permitted to rearrange the dentries
727 void shrink_dcache_for_umount(struct super_block
*sb
)
729 struct dentry
*dentry
;
731 if (down_read_trylock(&sb
->s_umount
))
736 atomic_dec(&dentry
->d_count
);
737 shrink_dcache_for_umount_subtree(dentry
);
739 while (!hlist_empty(&sb
->s_anon
)) {
740 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
741 shrink_dcache_for_umount_subtree(dentry
);
746 * Search for at least 1 mount point in the dentry's subdirs.
747 * We descend to the next level whenever the d_subdirs
748 * list is non-empty and continue searching.
752 * have_submounts - check for mounts over a dentry
753 * @parent: dentry to check.
755 * Return true if the parent or its subdirectories contain
759 int have_submounts(struct dentry
*parent
)
761 struct dentry
*this_parent
= parent
;
762 struct list_head
*next
;
764 spin_lock(&dcache_lock
);
765 if (d_mountpoint(parent
))
768 next
= this_parent
->d_subdirs
.next
;
770 while (next
!= &this_parent
->d_subdirs
) {
771 struct list_head
*tmp
= next
;
772 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
774 /* Have we found a mount point ? */
775 if (d_mountpoint(dentry
))
777 if (!list_empty(&dentry
->d_subdirs
)) {
778 this_parent
= dentry
;
783 * All done at this level ... ascend and resume the search.
785 if (this_parent
!= parent
) {
786 next
= this_parent
->d_u
.d_child
.next
;
787 this_parent
= this_parent
->d_parent
;
790 spin_unlock(&dcache_lock
);
791 return 0; /* No mount points found in tree */
793 spin_unlock(&dcache_lock
);
798 * Search the dentry child list for the specified parent,
799 * and move any unused dentries to the end of the unused
800 * list for prune_dcache(). We descend to the next level
801 * whenever the d_subdirs list is non-empty and continue
804 * It returns zero iff there are no unused children,
805 * otherwise it returns the number of children moved to
806 * the end of the unused list. This may not be the total
807 * number of unused children, because select_parent can
808 * drop the lock and return early due to latency
811 static int select_parent(struct dentry
* parent
)
813 struct dentry
*this_parent
= parent
;
814 struct list_head
*next
;
817 spin_lock(&dcache_lock
);
819 next
= this_parent
->d_subdirs
.next
;
821 while (next
!= &this_parent
->d_subdirs
) {
822 struct list_head
*tmp
= next
;
823 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
826 dentry_lru_del_init(dentry
);
828 * move only zero ref count dentries to the end
829 * of the unused list for prune_dcache
831 if (!atomic_read(&dentry
->d_count
)) {
832 dentry_lru_add_tail(dentry
);
837 * We can return to the caller if we have found some (this
838 * ensures forward progress). We'll be coming back to find
841 if (found
&& need_resched())
845 * Descend a level if the d_subdirs list is non-empty.
847 if (!list_empty(&dentry
->d_subdirs
)) {
848 this_parent
= dentry
;
853 * All done at this level ... ascend and resume the search.
855 if (this_parent
!= parent
) {
856 next
= this_parent
->d_u
.d_child
.next
;
857 this_parent
= this_parent
->d_parent
;
861 spin_unlock(&dcache_lock
);
866 * shrink_dcache_parent - prune dcache
867 * @parent: parent of entries to prune
869 * Prune the dcache to remove unused children of the parent dentry.
872 void shrink_dcache_parent(struct dentry
* parent
)
874 struct super_block
*sb
= parent
->d_sb
;
877 while ((found
= select_parent(parent
)) != 0)
878 __shrink_dcache_sb(sb
, &found
, 0);
882 * Scan `nr' dentries and return the number which remain.
884 * We need to avoid reentering the filesystem if the caller is performing a
885 * GFP_NOFS allocation attempt. One example deadlock is:
887 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
888 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
889 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
891 * In this case we return -1 to tell the caller that we baled.
893 static int shrink_dcache_memory(int nr
, gfp_t gfp_mask
)
896 if (!(gfp_mask
& __GFP_FS
))
900 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
903 static struct shrinker dcache_shrinker
= {
904 .shrink
= shrink_dcache_memory
,
905 .seeks
= DEFAULT_SEEKS
,
909 * d_alloc - allocate a dcache entry
910 * @parent: parent of entry to allocate
911 * @name: qstr of the name
913 * Allocates a dentry. It returns %NULL if there is insufficient memory
914 * available. On a success the dentry is returned. The name passed in is
915 * copied and the copy passed in may be reused after this call.
918 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
920 struct dentry
*dentry
;
923 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
927 if (name
->len
> DNAME_INLINE_LEN
-1) {
928 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
930 kmem_cache_free(dentry_cache
, dentry
);
934 dname
= dentry
->d_iname
;
936 dentry
->d_name
.name
= dname
;
938 dentry
->d_name
.len
= name
->len
;
939 dentry
->d_name
.hash
= name
->hash
;
940 memcpy(dname
, name
->name
, name
->len
);
941 dname
[name
->len
] = 0;
943 atomic_set(&dentry
->d_count
, 1);
944 dentry
->d_flags
= DCACHE_UNHASHED
;
945 spin_lock_init(&dentry
->d_lock
);
946 dentry
->d_inode
= NULL
;
947 dentry
->d_parent
= NULL
;
950 dentry
->d_fsdata
= NULL
;
951 dentry
->d_mounted
= 0;
952 INIT_HLIST_NODE(&dentry
->d_hash
);
953 INIT_LIST_HEAD(&dentry
->d_lru
);
954 INIT_LIST_HEAD(&dentry
->d_subdirs
);
955 INIT_LIST_HEAD(&dentry
->d_alias
);
958 dentry
->d_parent
= dget(parent
);
959 dentry
->d_sb
= parent
->d_sb
;
961 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
964 spin_lock(&dcache_lock
);
966 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
967 dentry_stat
.nr_dentry
++;
968 spin_unlock(&dcache_lock
);
973 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
978 q
.len
= strlen(name
);
979 q
.hash
= full_name_hash(q
.name
, q
.len
);
980 return d_alloc(parent
, &q
);
983 /* the caller must hold dcache_lock */
984 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
987 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
988 dentry
->d_inode
= inode
;
989 fsnotify_d_instantiate(dentry
, inode
);
993 * d_instantiate - fill in inode information for a dentry
994 * @entry: dentry to complete
995 * @inode: inode to attach to this dentry
997 * Fill in inode information in the entry.
999 * This turns negative dentries into productive full members
1002 * NOTE! This assumes that the inode count has been incremented
1003 * (or otherwise set) by the caller to indicate that it is now
1004 * in use by the dcache.
1007 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1009 BUG_ON(!list_empty(&entry
->d_alias
));
1010 spin_lock(&dcache_lock
);
1011 __d_instantiate(entry
, inode
);
1012 spin_unlock(&dcache_lock
);
1013 security_d_instantiate(entry
, inode
);
1017 * d_instantiate_unique - instantiate a non-aliased dentry
1018 * @entry: dentry to instantiate
1019 * @inode: inode to attach to this dentry
1021 * Fill in inode information in the entry. On success, it returns NULL.
1022 * If an unhashed alias of "entry" already exists, then we return the
1023 * aliased dentry instead and drop one reference to inode.
1025 * Note that in order to avoid conflicts with rename() etc, the caller
1026 * had better be holding the parent directory semaphore.
1028 * This also assumes that the inode count has been incremented
1029 * (or otherwise set) by the caller to indicate that it is now
1030 * in use by the dcache.
1032 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1033 struct inode
*inode
)
1035 struct dentry
*alias
;
1036 int len
= entry
->d_name
.len
;
1037 const char *name
= entry
->d_name
.name
;
1038 unsigned int hash
= entry
->d_name
.hash
;
1041 __d_instantiate(entry
, NULL
);
1045 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1046 struct qstr
*qstr
= &alias
->d_name
;
1048 if (qstr
->hash
!= hash
)
1050 if (alias
->d_parent
!= entry
->d_parent
)
1052 if (qstr
->len
!= len
)
1054 if (memcmp(qstr
->name
, name
, len
))
1060 __d_instantiate(entry
, inode
);
1064 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1066 struct dentry
*result
;
1068 BUG_ON(!list_empty(&entry
->d_alias
));
1070 spin_lock(&dcache_lock
);
1071 result
= __d_instantiate_unique(entry
, inode
);
1072 spin_unlock(&dcache_lock
);
1075 security_d_instantiate(entry
, inode
);
1079 BUG_ON(!d_unhashed(result
));
1084 EXPORT_SYMBOL(d_instantiate_unique
);
1087 * d_alloc_root - allocate root dentry
1088 * @root_inode: inode to allocate the root for
1090 * Allocate a root ("/") dentry for the inode given. The inode is
1091 * instantiated and returned. %NULL is returned if there is insufficient
1092 * memory or the inode passed is %NULL.
1095 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1097 struct dentry
*res
= NULL
;
1100 static const struct qstr name
= { .name
= "/", .len
= 1 };
1102 res
= d_alloc(NULL
, &name
);
1104 res
->d_sb
= root_inode
->i_sb
;
1105 res
->d_parent
= res
;
1106 d_instantiate(res
, root_inode
);
1112 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1115 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1116 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1117 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1121 * d_obtain_alias - find or allocate a dentry for a given inode
1122 * @inode: inode to allocate the dentry for
1124 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1125 * similar open by handle operations. The returned dentry may be anonymous,
1126 * or may have a full name (if the inode was already in the cache).
1128 * When called on a directory inode, we must ensure that the inode only ever
1129 * has one dentry. If a dentry is found, that is returned instead of
1130 * allocating a new one.
1132 * On successful return, the reference to the inode has been transferred
1133 * to the dentry. In case of an error the reference on the inode is released.
1134 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1135 * be passed in and will be the error will be propagate to the return value,
1136 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1138 struct dentry
*d_obtain_alias(struct inode
*inode
)
1140 static const struct qstr anonstring
= { .name
= "" };
1145 return ERR_PTR(-ESTALE
);
1147 return ERR_CAST(inode
);
1149 res
= d_find_alias(inode
);
1153 tmp
= d_alloc(NULL
, &anonstring
);
1155 res
= ERR_PTR(-ENOMEM
);
1158 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1160 spin_lock(&dcache_lock
);
1161 res
= __d_find_alias(inode
, 0);
1163 spin_unlock(&dcache_lock
);
1168 /* attach a disconnected dentry */
1169 spin_lock(&tmp
->d_lock
);
1170 tmp
->d_sb
= inode
->i_sb
;
1171 tmp
->d_inode
= inode
;
1172 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1173 tmp
->d_flags
&= ~DCACHE_UNHASHED
;
1174 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1175 hlist_add_head(&tmp
->d_hash
, &inode
->i_sb
->s_anon
);
1176 spin_unlock(&tmp
->d_lock
);
1178 spin_unlock(&dcache_lock
);
1185 EXPORT_SYMBOL(d_obtain_alias
);
1188 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1189 * @inode: the inode which may have a disconnected dentry
1190 * @dentry: a negative dentry which we want to point to the inode.
1192 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1193 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1194 * and return it, else simply d_add the inode to the dentry and return NULL.
1196 * This is needed in the lookup routine of any filesystem that is exportable
1197 * (via knfsd) so that we can build dcache paths to directories effectively.
1199 * If a dentry was found and moved, then it is returned. Otherwise NULL
1200 * is returned. This matches the expected return value of ->lookup.
1203 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1205 struct dentry
*new = NULL
;
1207 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1208 spin_lock(&dcache_lock
);
1209 new = __d_find_alias(inode
, 1);
1211 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1212 spin_unlock(&dcache_lock
);
1213 security_d_instantiate(new, inode
);
1215 d_move(new, dentry
);
1218 /* already taking dcache_lock, so d_add() by hand */
1219 __d_instantiate(dentry
, inode
);
1220 spin_unlock(&dcache_lock
);
1221 security_d_instantiate(dentry
, inode
);
1225 d_add(dentry
, inode
);
1230 * d_add_ci - lookup or allocate new dentry with case-exact name
1231 * @inode: the inode case-insensitive lookup has found
1232 * @dentry: the negative dentry that was passed to the parent's lookup func
1233 * @name: the case-exact name to be associated with the returned dentry
1235 * This is to avoid filling the dcache with case-insensitive names to the
1236 * same inode, only the actual correct case is stored in the dcache for
1237 * case-insensitive filesystems.
1239 * For a case-insensitive lookup match and if the the case-exact dentry
1240 * already exists in in the dcache, use it and return it.
1242 * If no entry exists with the exact case name, allocate new dentry with
1243 * the exact case, and return the spliced entry.
1245 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1249 struct dentry
*found
;
1253 * First check if a dentry matching the name already exists,
1254 * if not go ahead and create it now.
1256 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1258 new = d_alloc(dentry
->d_parent
, name
);
1264 found
= d_splice_alias(inode
, new);
1273 * If a matching dentry exists, and it's not negative use it.
1275 * Decrement the reference count to balance the iget() done
1278 if (found
->d_inode
) {
1279 if (unlikely(found
->d_inode
!= inode
)) {
1280 /* This can't happen because bad inodes are unhashed. */
1281 BUG_ON(!is_bad_inode(inode
));
1282 BUG_ON(!is_bad_inode(found
->d_inode
));
1289 * Negative dentry: instantiate it unless the inode is a directory and
1290 * already has a dentry.
1292 spin_lock(&dcache_lock
);
1293 if (!S_ISDIR(inode
->i_mode
) || list_empty(&inode
->i_dentry
)) {
1294 __d_instantiate(found
, inode
);
1295 spin_unlock(&dcache_lock
);
1296 security_d_instantiate(found
, inode
);
1301 * In case a directory already has a (disconnected) entry grab a
1302 * reference to it, move it in place and use it.
1304 new = list_entry(inode
->i_dentry
.next
, struct dentry
, d_alias
);
1306 spin_unlock(&dcache_lock
);
1307 security_d_instantiate(found
, inode
);
1315 return ERR_PTR(error
);
1319 * d_lookup - search for a dentry
1320 * @parent: parent dentry
1321 * @name: qstr of name we wish to find
1323 * Searches the children of the parent dentry for the name in question. If
1324 * the dentry is found its reference count is incremented and the dentry
1325 * is returned. The caller must use dput to free the entry when it has
1326 * finished using it. %NULL is returned on failure.
1328 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1329 * Memory barriers are used while updating and doing lockless traversal.
1330 * To avoid races with d_move while rename is happening, d_lock is used.
1332 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1333 * and name pointer in one structure pointed by d_qstr.
1335 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1336 * lookup is going on.
1338 * The dentry unused LRU is not updated even if lookup finds the required dentry
1339 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1340 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1343 * d_lookup() is protected against the concurrent renames in some unrelated
1344 * directory using the seqlockt_t rename_lock.
1347 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1349 struct dentry
* dentry
= NULL
;
1353 seq
= read_seqbegin(&rename_lock
);
1354 dentry
= __d_lookup(parent
, name
);
1357 } while (read_seqretry(&rename_lock
, seq
));
1361 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1363 unsigned int len
= name
->len
;
1364 unsigned int hash
= name
->hash
;
1365 const unsigned char *str
= name
->name
;
1366 struct hlist_head
*head
= d_hash(parent
,hash
);
1367 struct dentry
*found
= NULL
;
1368 struct hlist_node
*node
;
1369 struct dentry
*dentry
;
1373 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1376 if (dentry
->d_name
.hash
!= hash
)
1378 if (dentry
->d_parent
!= parent
)
1381 spin_lock(&dentry
->d_lock
);
1384 * Recheck the dentry after taking the lock - d_move may have
1385 * changed things. Don't bother checking the hash because we're
1386 * about to compare the whole name anyway.
1388 if (dentry
->d_parent
!= parent
)
1391 /* non-existing due to RCU? */
1392 if (d_unhashed(dentry
))
1396 * It is safe to compare names since d_move() cannot
1397 * change the qstr (protected by d_lock).
1399 qstr
= &dentry
->d_name
;
1400 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1401 if (parent
->d_op
->d_compare(parent
, qstr
, name
))
1404 if (qstr
->len
!= len
)
1406 if (memcmp(qstr
->name
, str
, len
))
1410 atomic_inc(&dentry
->d_count
);
1412 spin_unlock(&dentry
->d_lock
);
1415 spin_unlock(&dentry
->d_lock
);
1423 * d_hash_and_lookup - hash the qstr then search for a dentry
1424 * @dir: Directory to search in
1425 * @name: qstr of name we wish to find
1427 * On hash failure or on lookup failure NULL is returned.
1429 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1431 struct dentry
*dentry
= NULL
;
1434 * Check for a fs-specific hash function. Note that we must
1435 * calculate the standard hash first, as the d_op->d_hash()
1436 * routine may choose to leave the hash value unchanged.
1438 name
->hash
= full_name_hash(name
->name
, name
->len
);
1439 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1440 if (dir
->d_op
->d_hash(dir
, name
) < 0)
1443 dentry
= d_lookup(dir
, name
);
1449 * d_validate - verify dentry provided from insecure source
1450 * @dentry: The dentry alleged to be valid child of @dparent
1451 * @dparent: The parent dentry (known to be valid)
1453 * An insecure source has sent us a dentry, here we verify it and dget() it.
1454 * This is used by ncpfs in its readdir implementation.
1455 * Zero is returned in the dentry is invalid.
1458 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1460 struct hlist_head
*base
;
1461 struct hlist_node
*lhp
;
1463 /* Check whether the ptr might be valid at all.. */
1464 if (!kmem_ptr_validate(dentry_cache
, dentry
))
1467 if (dentry
->d_parent
!= dparent
)
1470 spin_lock(&dcache_lock
);
1471 base
= d_hash(dparent
, dentry
->d_name
.hash
);
1472 hlist_for_each(lhp
,base
) {
1473 /* hlist_for_each_entry_rcu() not required for d_hash list
1474 * as it is parsed under dcache_lock
1476 if (dentry
== hlist_entry(lhp
, struct dentry
, d_hash
)) {
1477 __dget_locked(dentry
);
1478 spin_unlock(&dcache_lock
);
1482 spin_unlock(&dcache_lock
);
1488 * When a file is deleted, we have two options:
1489 * - turn this dentry into a negative dentry
1490 * - unhash this dentry and free it.
1492 * Usually, we want to just turn this into
1493 * a negative dentry, but if anybody else is
1494 * currently using the dentry or the inode
1495 * we can't do that and we fall back on removing
1496 * it from the hash queues and waiting for
1497 * it to be deleted later when it has no users
1501 * d_delete - delete a dentry
1502 * @dentry: The dentry to delete
1504 * Turn the dentry into a negative dentry if possible, otherwise
1505 * remove it from the hash queues so it can be deleted later
1508 void d_delete(struct dentry
* dentry
)
1512 * Are we the only user?
1514 spin_lock(&dcache_lock
);
1515 spin_lock(&dentry
->d_lock
);
1516 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1517 if (atomic_read(&dentry
->d_count
) == 1) {
1518 dentry_iput(dentry
);
1519 fsnotify_nameremove(dentry
, isdir
);
1523 if (!d_unhashed(dentry
))
1526 spin_unlock(&dentry
->d_lock
);
1527 spin_unlock(&dcache_lock
);
1529 fsnotify_nameremove(dentry
, isdir
);
1532 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1535 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1536 hlist_add_head_rcu(&entry
->d_hash
, list
);
1539 static void _d_rehash(struct dentry
* entry
)
1541 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1545 * d_rehash - add an entry back to the hash
1546 * @entry: dentry to add to the hash
1548 * Adds a dentry to the hash according to its name.
1551 void d_rehash(struct dentry
* entry
)
1553 spin_lock(&dcache_lock
);
1554 spin_lock(&entry
->d_lock
);
1556 spin_unlock(&entry
->d_lock
);
1557 spin_unlock(&dcache_lock
);
1561 * When switching names, the actual string doesn't strictly have to
1562 * be preserved in the target - because we're dropping the target
1563 * anyway. As such, we can just do a simple memcpy() to copy over
1564 * the new name before we switch.
1566 * Note that we have to be a lot more careful about getting the hash
1567 * switched - we have to switch the hash value properly even if it
1568 * then no longer matches the actual (corrupted) string of the target.
1569 * The hash value has to match the hash queue that the dentry is on..
1571 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1573 if (dname_external(target
)) {
1574 if (dname_external(dentry
)) {
1576 * Both external: swap the pointers
1578 swap(target
->d_name
.name
, dentry
->d_name
.name
);
1581 * dentry:internal, target:external. Steal target's
1582 * storage and make target internal.
1584 memcpy(target
->d_iname
, dentry
->d_name
.name
,
1585 dentry
->d_name
.len
+ 1);
1586 dentry
->d_name
.name
= target
->d_name
.name
;
1587 target
->d_name
.name
= target
->d_iname
;
1590 if (dname_external(dentry
)) {
1592 * dentry:external, target:internal. Give dentry's
1593 * storage to target and make dentry internal
1595 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1596 target
->d_name
.len
+ 1);
1597 target
->d_name
.name
= dentry
->d_name
.name
;
1598 dentry
->d_name
.name
= dentry
->d_iname
;
1601 * Both are internal. Just copy target to dentry
1603 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1604 target
->d_name
.len
+ 1);
1605 dentry
->d_name
.len
= target
->d_name
.len
;
1609 swap(dentry
->d_name
.len
, target
->d_name
.len
);
1613 * We cannibalize "target" when moving dentry on top of it,
1614 * because it's going to be thrown away anyway. We could be more
1615 * polite about it, though.
1617 * This forceful removal will result in ugly /proc output if
1618 * somebody holds a file open that got deleted due to a rename.
1619 * We could be nicer about the deleted file, and let it show
1620 * up under the name it had before it was deleted rather than
1621 * under the original name of the file that was moved on top of it.
1625 * d_move_locked - move a dentry
1626 * @dentry: entry to move
1627 * @target: new dentry
1629 * Update the dcache to reflect the move of a file name. Negative
1630 * dcache entries should not be moved in this way.
1632 static void d_move_locked(struct dentry
* dentry
, struct dentry
* target
)
1634 struct hlist_head
*list
;
1636 if (!dentry
->d_inode
)
1637 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1639 write_seqlock(&rename_lock
);
1641 * XXXX: do we really need to take target->d_lock?
1643 if (target
< dentry
) {
1644 spin_lock(&target
->d_lock
);
1645 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1647 spin_lock(&dentry
->d_lock
);
1648 spin_lock_nested(&target
->d_lock
, DENTRY_D_LOCK_NESTED
);
1651 /* Move the dentry to the target hash queue, if on different bucket */
1652 if (d_unhashed(dentry
))
1653 goto already_unhashed
;
1655 hlist_del_rcu(&dentry
->d_hash
);
1658 list
= d_hash(target
->d_parent
, target
->d_name
.hash
);
1659 __d_rehash(dentry
, list
);
1661 /* Unhash the target: dput() will then get rid of it */
1664 list_del(&dentry
->d_u
.d_child
);
1665 list_del(&target
->d_u
.d_child
);
1667 /* Switch the names.. */
1668 switch_names(dentry
, target
);
1669 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
1671 /* ... and switch the parents */
1672 if (IS_ROOT(dentry
)) {
1673 dentry
->d_parent
= target
->d_parent
;
1674 target
->d_parent
= target
;
1675 INIT_LIST_HEAD(&target
->d_u
.d_child
);
1677 swap(dentry
->d_parent
, target
->d_parent
);
1679 /* And add them back to the (new) parent lists */
1680 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
1683 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1684 spin_unlock(&target
->d_lock
);
1685 fsnotify_d_move(dentry
);
1686 spin_unlock(&dentry
->d_lock
);
1687 write_sequnlock(&rename_lock
);
1691 * d_move - move a dentry
1692 * @dentry: entry to move
1693 * @target: new dentry
1695 * Update the dcache to reflect the move of a file name. Negative
1696 * dcache entries should not be moved in this way.
1699 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1701 spin_lock(&dcache_lock
);
1702 d_move_locked(dentry
, target
);
1703 spin_unlock(&dcache_lock
);
1707 * d_ancestor - search for an ancestor
1708 * @p1: ancestor dentry
1711 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1712 * an ancestor of p2, else NULL.
1714 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
1718 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
1719 if (p
->d_parent
== p1
)
1726 * This helper attempts to cope with remotely renamed directories
1728 * It assumes that the caller is already holding
1729 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1731 * Note: If ever the locking in lock_rename() changes, then please
1732 * remember to update this too...
1734 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
1735 __releases(dcache_lock
)
1737 struct mutex
*m1
= NULL
, *m2
= NULL
;
1740 /* If alias and dentry share a parent, then no extra locks required */
1741 if (alias
->d_parent
== dentry
->d_parent
)
1744 /* Check for loops */
1745 ret
= ERR_PTR(-ELOOP
);
1746 if (d_ancestor(alias
, dentry
))
1749 /* See lock_rename() */
1750 ret
= ERR_PTR(-EBUSY
);
1751 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
1753 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
1754 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
1756 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
1758 d_move_locked(alias
, dentry
);
1761 spin_unlock(&dcache_lock
);
1770 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1771 * named dentry in place of the dentry to be replaced.
1773 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
1775 struct dentry
*dparent
, *aparent
;
1777 switch_names(dentry
, anon
);
1778 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
1780 dparent
= dentry
->d_parent
;
1781 aparent
= anon
->d_parent
;
1783 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
1784 list_del(&dentry
->d_u
.d_child
);
1785 if (!IS_ROOT(dentry
))
1786 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1788 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1790 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
1791 list_del(&anon
->d_u
.d_child
);
1793 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
1795 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
1797 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
1801 * d_materialise_unique - introduce an inode into the tree
1802 * @dentry: candidate dentry
1803 * @inode: inode to bind to the dentry, to which aliases may be attached
1805 * Introduces an dentry into the tree, substituting an extant disconnected
1806 * root directory alias in its place if there is one
1808 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
1810 struct dentry
*actual
;
1812 BUG_ON(!d_unhashed(dentry
));
1814 spin_lock(&dcache_lock
);
1818 __d_instantiate(dentry
, NULL
);
1822 if (S_ISDIR(inode
->i_mode
)) {
1823 struct dentry
*alias
;
1825 /* Does an aliased dentry already exist? */
1826 alias
= __d_find_alias(inode
, 0);
1829 /* Is this an anonymous mountpoint that we could splice
1831 if (IS_ROOT(alias
)) {
1832 spin_lock(&alias
->d_lock
);
1833 __d_materialise_dentry(dentry
, alias
);
1837 /* Nope, but we must(!) avoid directory aliasing */
1838 actual
= __d_unalias(dentry
, alias
);
1845 /* Add a unique reference */
1846 actual
= __d_instantiate_unique(dentry
, inode
);
1849 else if (unlikely(!d_unhashed(actual
)))
1850 goto shouldnt_be_hashed
;
1853 spin_lock(&actual
->d_lock
);
1856 spin_unlock(&actual
->d_lock
);
1857 spin_unlock(&dcache_lock
);
1859 if (actual
== dentry
) {
1860 security_d_instantiate(dentry
, inode
);
1868 spin_unlock(&dcache_lock
);
1872 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
1876 return -ENAMETOOLONG
;
1878 memcpy(*buffer
, str
, namelen
);
1882 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
1884 return prepend(buffer
, buflen
, name
->name
, name
->len
);
1888 * __d_path - return the path of a dentry
1889 * @path: the dentry/vfsmount to report
1890 * @root: root vfsmnt/dentry (may be modified by this function)
1891 * @buffer: buffer to return value in
1892 * @buflen: buffer length
1894 * Convert a dentry into an ASCII path name. If the entry has been deleted
1895 * the string " (deleted)" is appended. Note that this is ambiguous.
1897 * Returns a pointer into the buffer or an error code if the
1898 * path was too long.
1900 * "buflen" should be positive. Caller holds the dcache_lock.
1902 * If path is not reachable from the supplied root, then the value of
1903 * root is changed (without modifying refcounts).
1905 char *__d_path(const struct path
*path
, struct path
*root
,
1906 char *buffer
, int buflen
)
1908 struct dentry
*dentry
= path
->dentry
;
1909 struct vfsmount
*vfsmnt
= path
->mnt
;
1910 char *end
= buffer
+ buflen
;
1913 spin_lock(&vfsmount_lock
);
1914 prepend(&end
, &buflen
, "\0", 1);
1915 if (d_unlinked(dentry
) &&
1916 (prepend(&end
, &buflen
, " (deleted)", 10) != 0))
1926 struct dentry
* parent
;
1928 if (dentry
== root
->dentry
&& vfsmnt
== root
->mnt
)
1930 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
1932 if (vfsmnt
->mnt_parent
== vfsmnt
) {
1935 dentry
= vfsmnt
->mnt_mountpoint
;
1936 vfsmnt
= vfsmnt
->mnt_parent
;
1939 parent
= dentry
->d_parent
;
1941 if ((prepend_name(&end
, &buflen
, &dentry
->d_name
) != 0) ||
1942 (prepend(&end
, &buflen
, "/", 1) != 0))
1949 spin_unlock(&vfsmount_lock
);
1953 retval
+= 1; /* hit the slash */
1954 if (prepend_name(&retval
, &buflen
, &dentry
->d_name
) != 0)
1957 root
->dentry
= dentry
;
1961 retval
= ERR_PTR(-ENAMETOOLONG
);
1966 * d_path - return the path of a dentry
1967 * @path: path to report
1968 * @buf: buffer to return value in
1969 * @buflen: buffer length
1971 * Convert a dentry into an ASCII path name. If the entry has been deleted
1972 * the string " (deleted)" is appended. Note that this is ambiguous.
1974 * Returns a pointer into the buffer or an error code if the path was
1975 * too long. Note: Callers should use the returned pointer, not the passed
1976 * in buffer, to use the name! The implementation often starts at an offset
1977 * into the buffer, and may leave 0 bytes at the start.
1979 * "buflen" should be positive.
1981 char *d_path(const struct path
*path
, char *buf
, int buflen
)
1988 * We have various synthetic filesystems that never get mounted. On
1989 * these filesystems dentries are never used for lookup purposes, and
1990 * thus don't need to be hashed. They also don't need a name until a
1991 * user wants to identify the object in /proc/pid/fd/. The little hack
1992 * below allows us to generate a name for these objects on demand:
1994 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
1995 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
1997 read_lock(¤t
->fs
->lock
);
1998 root
= current
->fs
->root
;
2000 read_unlock(¤t
->fs
->lock
);
2001 spin_lock(&dcache_lock
);
2003 res
= __d_path(path
, &tmp
, buf
, buflen
);
2004 spin_unlock(&dcache_lock
);
2010 * Helper function for dentry_operations.d_dname() members
2012 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2013 const char *fmt
, ...)
2019 va_start(args
, fmt
);
2020 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2023 if (sz
> sizeof(temp
) || sz
> buflen
)
2024 return ERR_PTR(-ENAMETOOLONG
);
2026 buffer
+= buflen
- sz
;
2027 return memcpy(buffer
, temp
, sz
);
2031 * Write full pathname from the root of the filesystem into the buffer.
2033 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2035 char *end
= buf
+ buflen
;
2038 spin_lock(&dcache_lock
);
2039 prepend(&end
, &buflen
, "\0", 1);
2040 if (d_unlinked(dentry
) &&
2041 (prepend(&end
, &buflen
, "//deleted", 9) != 0))
2049 while (!IS_ROOT(dentry
)) {
2050 struct dentry
*parent
= dentry
->d_parent
;
2053 if ((prepend_name(&end
, &buflen
, &dentry
->d_name
) != 0) ||
2054 (prepend(&end
, &buflen
, "/", 1) != 0))
2060 spin_unlock(&dcache_lock
);
2063 spin_unlock(&dcache_lock
);
2064 return ERR_PTR(-ENAMETOOLONG
);
2068 * NOTE! The user-level library version returns a
2069 * character pointer. The kernel system call just
2070 * returns the length of the buffer filled (which
2071 * includes the ending '\0' character), or a negative
2072 * error value. So libc would do something like
2074 * char *getcwd(char * buf, size_t size)
2078 * retval = sys_getcwd(buf, size);
2085 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2088 struct path pwd
, root
;
2089 char *page
= (char *) __get_free_page(GFP_USER
);
2094 read_lock(¤t
->fs
->lock
);
2095 pwd
= current
->fs
->pwd
;
2097 root
= current
->fs
->root
;
2099 read_unlock(¤t
->fs
->lock
);
2102 spin_lock(&dcache_lock
);
2103 if (!d_unlinked(pwd
.dentry
)) {
2105 struct path tmp
= root
;
2108 cwd
= __d_path(&pwd
, &tmp
, page
, PAGE_SIZE
);
2109 spin_unlock(&dcache_lock
);
2111 error
= PTR_ERR(cwd
);
2116 len
= PAGE_SIZE
+ page
- cwd
;
2119 if (copy_to_user(buf
, cwd
, len
))
2123 spin_unlock(&dcache_lock
);
2128 free_page((unsigned long) page
);
2133 * Test whether new_dentry is a subdirectory of old_dentry.
2135 * Trivially implemented using the dcache structure
2139 * is_subdir - is new dentry a subdirectory of old_dentry
2140 * @new_dentry: new dentry
2141 * @old_dentry: old dentry
2143 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2144 * Returns 0 otherwise.
2145 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2148 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2153 if (new_dentry
== old_dentry
)
2157 * Need rcu_readlock to protect against the d_parent trashing
2162 /* for restarting inner loop in case of seq retry */
2163 seq
= read_seqbegin(&rename_lock
);
2164 if (d_ancestor(old_dentry
, new_dentry
))
2168 } while (read_seqretry(&rename_lock
, seq
));
2174 void d_genocide(struct dentry
*root
)
2176 struct dentry
*this_parent
= root
;
2177 struct list_head
*next
;
2179 spin_lock(&dcache_lock
);
2181 next
= this_parent
->d_subdirs
.next
;
2183 while (next
!= &this_parent
->d_subdirs
) {
2184 struct list_head
*tmp
= next
;
2185 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2187 if (d_unhashed(dentry
)||!dentry
->d_inode
)
2189 if (!list_empty(&dentry
->d_subdirs
)) {
2190 this_parent
= dentry
;
2193 atomic_dec(&dentry
->d_count
);
2195 if (this_parent
!= root
) {
2196 next
= this_parent
->d_u
.d_child
.next
;
2197 atomic_dec(&this_parent
->d_count
);
2198 this_parent
= this_parent
->d_parent
;
2201 spin_unlock(&dcache_lock
);
2205 * find_inode_number - check for dentry with name
2206 * @dir: directory to check
2207 * @name: Name to find.
2209 * Check whether a dentry already exists for the given name,
2210 * and return the inode number if it has an inode. Otherwise
2213 * This routine is used to post-process directory listings for
2214 * filesystems using synthetic inode numbers, and is necessary
2215 * to keep getcwd() working.
2218 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2220 struct dentry
* dentry
;
2223 dentry
= d_hash_and_lookup(dir
, name
);
2225 if (dentry
->d_inode
)
2226 ino
= dentry
->d_inode
->i_ino
;
2232 static __initdata
unsigned long dhash_entries
;
2233 static int __init
set_dhash_entries(char *str
)
2237 dhash_entries
= simple_strtoul(str
, &str
, 0);
2240 __setup("dhash_entries=", set_dhash_entries
);
2242 static void __init
dcache_init_early(void)
2246 /* If hashes are distributed across NUMA nodes, defer
2247 * hash allocation until vmalloc space is available.
2253 alloc_large_system_hash("Dentry cache",
2254 sizeof(struct hlist_head
),
2262 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2263 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2266 static void __init
dcache_init(void)
2271 * A constructor could be added for stable state like the lists,
2272 * but it is probably not worth it because of the cache nature
2275 dentry_cache
= KMEM_CACHE(dentry
,
2276 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2278 register_shrinker(&dcache_shrinker
);
2280 /* Hash may have been set up in dcache_init_early */
2285 alloc_large_system_hash("Dentry cache",
2286 sizeof(struct hlist_head
),
2294 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2295 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2298 /* SLAB cache for __getname() consumers */
2299 struct kmem_cache
*names_cachep __read_mostly
;
2301 EXPORT_SYMBOL(d_genocide
);
2303 void __init
vfs_caches_init_early(void)
2305 dcache_init_early();
2309 void __init
vfs_caches_init(unsigned long mempages
)
2311 unsigned long reserve
;
2313 /* Base hash sizes on available memory, with a reserve equal to
2314 150% of current kernel size */
2316 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2317 mempages
-= reserve
;
2319 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2320 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2324 files_init(mempages
);
2330 EXPORT_SYMBOL(d_alloc
);
2331 EXPORT_SYMBOL(d_alloc_root
);
2332 EXPORT_SYMBOL(d_delete
);
2333 EXPORT_SYMBOL(d_find_alias
);
2334 EXPORT_SYMBOL(d_instantiate
);
2335 EXPORT_SYMBOL(d_invalidate
);
2336 EXPORT_SYMBOL(d_lookup
);
2337 EXPORT_SYMBOL(d_move
);
2338 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2339 EXPORT_SYMBOL(d_path
);
2340 EXPORT_SYMBOL(d_prune_aliases
);
2341 EXPORT_SYMBOL(d_rehash
);
2342 EXPORT_SYMBOL(d_splice_alias
);
2343 EXPORT_SYMBOL(d_add_ci
);
2344 EXPORT_SYMBOL(d_validate
);
2345 EXPORT_SYMBOL(dget_locked
);
2346 EXPORT_SYMBOL(dput
);
2347 EXPORT_SYMBOL(find_inode_number
);
2348 EXPORT_SYMBOL(have_submounts
);
2349 EXPORT_SYMBOL(names_cachep
);
2350 EXPORT_SYMBOL(shrink_dcache_parent
);
2351 EXPORT_SYMBOL(shrink_dcache_sb
);