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 rcu_head
*head
)
72 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
74 WARN_ON(!list_empty(&dentry
->d_alias
));
75 if (dname_external(dentry
))
76 kfree(dentry
->d_name
.name
);
77 kmem_cache_free(dentry_cache
, dentry
);
81 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
84 static void d_free(struct dentry
*dentry
)
86 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
87 dentry
->d_op
->d_release(dentry
);
88 /* if dentry was never inserted into hash, immediate free is OK */
89 if (hlist_unhashed(&dentry
->d_hash
))
90 __d_free(&dentry
->d_u
.d_rcu
);
92 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
96 * Release the dentry's inode, using the filesystem
97 * d_iput() operation if defined.
99 static void dentry_iput(struct dentry
* dentry
)
100 __releases(dentry
->d_lock
)
101 __releases(dcache_lock
)
103 struct inode
*inode
= dentry
->d_inode
;
105 dentry
->d_inode
= NULL
;
106 list_del_init(&dentry
->d_alias
);
107 spin_unlock(&dentry
->d_lock
);
108 spin_unlock(&dcache_lock
);
110 fsnotify_inoderemove(inode
);
111 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
112 dentry
->d_op
->d_iput(dentry
, inode
);
116 spin_unlock(&dentry
->d_lock
);
117 spin_unlock(&dcache_lock
);
122 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
124 static void dentry_lru_add(struct dentry
*dentry
)
126 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
127 dentry
->d_sb
->s_nr_dentry_unused
++;
128 dentry_stat
.nr_unused
++;
131 static void dentry_lru_add_tail(struct dentry
*dentry
)
133 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
134 dentry
->d_sb
->s_nr_dentry_unused
++;
135 dentry_stat
.nr_unused
++;
138 static void dentry_lru_del(struct dentry
*dentry
)
140 if (!list_empty(&dentry
->d_lru
)) {
141 list_del(&dentry
->d_lru
);
142 dentry
->d_sb
->s_nr_dentry_unused
--;
143 dentry_stat
.nr_unused
--;
147 static void dentry_lru_del_init(struct dentry
*dentry
)
149 if (likely(!list_empty(&dentry
->d_lru
))) {
150 list_del_init(&dentry
->d_lru
);
151 dentry
->d_sb
->s_nr_dentry_unused
--;
152 dentry_stat
.nr_unused
--;
157 * d_kill - kill dentry and return parent
158 * @dentry: dentry to kill
160 * The dentry must already be unhashed and removed from the LRU.
162 * If this is the root of the dentry tree, return NULL.
164 static struct dentry
*d_kill(struct dentry
*dentry
)
165 __releases(dentry
->d_lock
)
166 __releases(dcache_lock
)
168 struct dentry
*parent
;
170 list_del(&dentry
->d_u
.d_child
);
171 dentry_stat
.nr_dentry
--; /* For d_free, below */
172 /*drops the locks, at that point nobody can reach this dentry */
177 parent
= dentry
->d_parent
;
185 * This is complicated by the fact that we do not want to put
186 * dentries that are no longer on any hash chain on the unused
187 * list: we'd much rather just get rid of them immediately.
189 * However, that implies that we have to traverse the dentry
190 * tree upwards to the parents which might _also_ now be
191 * scheduled for deletion (it may have been only waiting for
192 * its last child to go away).
194 * This tail recursion is done by hand as we don't want to depend
195 * on the compiler to always get this right (gcc generally doesn't).
196 * Real recursion would eat up our stack space.
200 * dput - release a dentry
201 * @dentry: dentry to release
203 * Release a dentry. This will drop the usage count and if appropriate
204 * call the dentry unlink method as well as removing it from the queues and
205 * releasing its resources. If the parent dentries were scheduled for release
206 * they too may now get deleted.
208 * no dcache lock, please.
211 void dput(struct dentry
*dentry
)
217 if (atomic_read(&dentry
->d_count
) == 1)
219 if (!atomic_dec_and_lock(&dentry
->d_count
, &dcache_lock
))
222 spin_lock(&dentry
->d_lock
);
223 if (atomic_read(&dentry
->d_count
)) {
224 spin_unlock(&dentry
->d_lock
);
225 spin_unlock(&dcache_lock
);
230 * AV: ->d_delete() is _NOT_ allowed to block now.
232 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
233 if (dentry
->d_op
->d_delete(dentry
))
236 /* Unreachable? Get rid of it */
237 if (d_unhashed(dentry
))
239 if (list_empty(&dentry
->d_lru
)) {
240 dentry
->d_flags
|= DCACHE_REFERENCED
;
241 dentry_lru_add(dentry
);
243 spin_unlock(&dentry
->d_lock
);
244 spin_unlock(&dcache_lock
);
250 /* if dentry was on the d_lru list delete it from there */
251 dentry_lru_del(dentry
);
252 dentry
= d_kill(dentry
);
259 * d_invalidate - invalidate a dentry
260 * @dentry: dentry to invalidate
262 * Try to invalidate the dentry if it turns out to be
263 * possible. If there are other dentries that can be
264 * reached through this one we can't delete it and we
265 * return -EBUSY. On success we return 0.
270 int d_invalidate(struct dentry
* dentry
)
273 * If it's already been dropped, return OK.
275 spin_lock(&dcache_lock
);
276 if (d_unhashed(dentry
)) {
277 spin_unlock(&dcache_lock
);
281 * Check whether to do a partial shrink_dcache
282 * to get rid of unused child entries.
284 if (!list_empty(&dentry
->d_subdirs
)) {
285 spin_unlock(&dcache_lock
);
286 shrink_dcache_parent(dentry
);
287 spin_lock(&dcache_lock
);
291 * Somebody else still using it?
293 * If it's a directory, we can't drop it
294 * for fear of somebody re-populating it
295 * with children (even though dropping it
296 * would make it unreachable from the root,
297 * we might still populate it if it was a
298 * working directory or similar).
300 spin_lock(&dentry
->d_lock
);
301 if (atomic_read(&dentry
->d_count
) > 1) {
302 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
303 spin_unlock(&dentry
->d_lock
);
304 spin_unlock(&dcache_lock
);
310 spin_unlock(&dentry
->d_lock
);
311 spin_unlock(&dcache_lock
);
314 EXPORT_SYMBOL(d_invalidate
);
316 /* This should be called _only_ with dcache_lock held */
318 static inline struct dentry
* __dget_locked(struct dentry
*dentry
)
320 atomic_inc(&dentry
->d_count
);
321 dentry_lru_del_init(dentry
);
325 struct dentry
* dget_locked(struct dentry
*dentry
)
327 return __dget_locked(dentry
);
329 EXPORT_SYMBOL(dget_locked
);
332 * d_find_alias - grab a hashed alias of inode
333 * @inode: inode in question
334 * @want_discon: flag, used by d_splice_alias, to request
335 * that only a DISCONNECTED alias be returned.
337 * If inode has a hashed alias, or is a directory and has any alias,
338 * acquire the reference to alias and return it. Otherwise return NULL.
339 * Notice that if inode is a directory there can be only one alias and
340 * it can be unhashed only if it has no children, or if it is the root
343 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
344 * any other hashed alias over that one unless @want_discon is set,
345 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
348 static struct dentry
* __d_find_alias(struct inode
*inode
, int want_discon
)
350 struct list_head
*head
, *next
, *tmp
;
351 struct dentry
*alias
, *discon_alias
=NULL
;
353 head
= &inode
->i_dentry
;
354 next
= inode
->i_dentry
.next
;
355 while (next
!= head
) {
359 alias
= list_entry(tmp
, struct dentry
, d_alias
);
360 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
361 if (IS_ROOT(alias
) &&
362 (alias
->d_flags
& DCACHE_DISCONNECTED
))
363 discon_alias
= alias
;
364 else if (!want_discon
) {
365 __dget_locked(alias
);
371 __dget_locked(discon_alias
);
375 struct dentry
* d_find_alias(struct inode
*inode
)
377 struct dentry
*de
= NULL
;
379 if (!list_empty(&inode
->i_dentry
)) {
380 spin_lock(&dcache_lock
);
381 de
= __d_find_alias(inode
, 0);
382 spin_unlock(&dcache_lock
);
386 EXPORT_SYMBOL(d_find_alias
);
389 * Try to kill dentries associated with this inode.
390 * WARNING: you must own a reference to inode.
392 void d_prune_aliases(struct inode
*inode
)
394 struct dentry
*dentry
;
396 spin_lock(&dcache_lock
);
397 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
398 spin_lock(&dentry
->d_lock
);
399 if (!atomic_read(&dentry
->d_count
)) {
400 __dget_locked(dentry
);
402 spin_unlock(&dentry
->d_lock
);
403 spin_unlock(&dcache_lock
);
407 spin_unlock(&dentry
->d_lock
);
409 spin_unlock(&dcache_lock
);
411 EXPORT_SYMBOL(d_prune_aliases
);
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
, *p
= NULL
;
537 int unused
= dentry_stat
.nr_unused
;
541 if (unused
== 0 || count
== 0)
543 spin_lock(&dcache_lock
);
547 prune_ratio
= unused
/ count
;
549 list_for_each_entry(sb
, &super_blocks
, s_list
) {
550 if (list_empty(&sb
->s_instances
))
552 if (sb
->s_nr_dentry_unused
== 0)
555 /* Now, we reclaim unused dentrins with fairness.
556 * We reclaim them same percentage from each superblock.
557 * We calculate number of dentries to scan on this sb
558 * as follows, but the implementation is arranged to avoid
560 * number of dentries to scan on this sb =
561 * count * (number of dentries on this sb /
562 * number of dentries in the machine)
564 spin_unlock(&sb_lock
);
565 if (prune_ratio
!= 1)
566 w_count
= (sb
->s_nr_dentry_unused
/ prune_ratio
) + 1;
568 w_count
= sb
->s_nr_dentry_unused
;
571 * We need to be sure this filesystem isn't being unmounted,
572 * otherwise we could race with generic_shutdown_super(), and
573 * end up holding a reference to an inode while the filesystem
574 * is unmounted. So we try to get s_umount, and make sure
577 if (down_read_trylock(&sb
->s_umount
)) {
578 if ((sb
->s_root
!= NULL
) &&
579 (!list_empty(&sb
->s_dentry_lru
))) {
580 spin_unlock(&dcache_lock
);
581 __shrink_dcache_sb(sb
, &w_count
,
584 spin_lock(&dcache_lock
);
586 up_read(&sb
->s_umount
);
593 /* more work left to do? */
599 spin_unlock(&sb_lock
);
600 spin_unlock(&dcache_lock
);
604 * shrink_dcache_sb - shrink dcache for a superblock
607 * Shrink the dcache for the specified super block. This
608 * is used to free the dcache before unmounting a file
611 void shrink_dcache_sb(struct super_block
* sb
)
613 __shrink_dcache_sb(sb
, NULL
, 0);
615 EXPORT_SYMBOL(shrink_dcache_sb
);
618 * destroy a single subtree of dentries for unmount
619 * - see the comments on shrink_dcache_for_umount() for a description of the
622 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
624 struct dentry
*parent
;
625 unsigned detached
= 0;
627 BUG_ON(!IS_ROOT(dentry
));
629 /* detach this root from the system */
630 spin_lock(&dcache_lock
);
631 dentry_lru_del_init(dentry
);
633 spin_unlock(&dcache_lock
);
636 /* descend to the first leaf in the current subtree */
637 while (!list_empty(&dentry
->d_subdirs
)) {
640 /* this is a branch with children - detach all of them
641 * from the system in one go */
642 spin_lock(&dcache_lock
);
643 list_for_each_entry(loop
, &dentry
->d_subdirs
,
645 dentry_lru_del_init(loop
);
647 cond_resched_lock(&dcache_lock
);
649 spin_unlock(&dcache_lock
);
651 /* move to the first child */
652 dentry
= list_entry(dentry
->d_subdirs
.next
,
653 struct dentry
, d_u
.d_child
);
656 /* consume the dentries from this leaf up through its parents
657 * until we find one with children or run out altogether */
661 if (atomic_read(&dentry
->d_count
) != 0) {
663 "BUG: Dentry %p{i=%lx,n=%s}"
665 " [unmount of %s %s]\n",
668 dentry
->d_inode
->i_ino
: 0UL,
670 atomic_read(&dentry
->d_count
),
671 dentry
->d_sb
->s_type
->name
,
679 parent
= dentry
->d_parent
;
680 atomic_dec(&parent
->d_count
);
683 list_del(&dentry
->d_u
.d_child
);
686 inode
= dentry
->d_inode
;
688 dentry
->d_inode
= NULL
;
689 list_del_init(&dentry
->d_alias
);
690 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
691 dentry
->d_op
->d_iput(dentry
, inode
);
698 /* finished when we fall off the top of the tree,
699 * otherwise we ascend to the parent and move to the
700 * next sibling if there is one */
706 } while (list_empty(&dentry
->d_subdirs
));
708 dentry
= list_entry(dentry
->d_subdirs
.next
,
709 struct dentry
, d_u
.d_child
);
712 /* several dentries were freed, need to correct nr_dentry */
713 spin_lock(&dcache_lock
);
714 dentry_stat
.nr_dentry
-= detached
;
715 spin_unlock(&dcache_lock
);
719 * destroy the dentries attached to a superblock on unmounting
720 * - we don't need to use dentry->d_lock, and only need dcache_lock when
721 * removing the dentry from the system lists and hashes because:
722 * - the superblock is detached from all mountings and open files, so the
723 * dentry trees will not be rearranged by the VFS
724 * - s_umount is write-locked, so the memory pressure shrinker will ignore
725 * any dentries belonging to this superblock that it comes across
726 * - the filesystem itself is no longer permitted to rearrange the dentries
729 void shrink_dcache_for_umount(struct super_block
*sb
)
731 struct dentry
*dentry
;
733 if (down_read_trylock(&sb
->s_umount
))
738 atomic_dec(&dentry
->d_count
);
739 shrink_dcache_for_umount_subtree(dentry
);
741 while (!hlist_empty(&sb
->s_anon
)) {
742 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
743 shrink_dcache_for_umount_subtree(dentry
);
748 * Search for at least 1 mount point in the dentry's subdirs.
749 * We descend to the next level whenever the d_subdirs
750 * list is non-empty and continue searching.
754 * have_submounts - check for mounts over a dentry
755 * @parent: dentry to check.
757 * Return true if the parent or its subdirectories contain
761 int have_submounts(struct dentry
*parent
)
763 struct dentry
*this_parent
= parent
;
764 struct list_head
*next
;
766 spin_lock(&dcache_lock
);
767 if (d_mountpoint(parent
))
770 next
= this_parent
->d_subdirs
.next
;
772 while (next
!= &this_parent
->d_subdirs
) {
773 struct list_head
*tmp
= next
;
774 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
776 /* Have we found a mount point ? */
777 if (d_mountpoint(dentry
))
779 if (!list_empty(&dentry
->d_subdirs
)) {
780 this_parent
= dentry
;
785 * All done at this level ... ascend and resume the search.
787 if (this_parent
!= parent
) {
788 next
= this_parent
->d_u
.d_child
.next
;
789 this_parent
= this_parent
->d_parent
;
792 spin_unlock(&dcache_lock
);
793 return 0; /* No mount points found in tree */
795 spin_unlock(&dcache_lock
);
798 EXPORT_SYMBOL(have_submounts
);
801 * Search the dentry child list for the specified parent,
802 * and move any unused dentries to the end of the unused
803 * list for prune_dcache(). We descend to the next level
804 * whenever the d_subdirs list is non-empty and continue
807 * It returns zero iff there are no unused children,
808 * otherwise it returns the number of children moved to
809 * the end of the unused list. This may not be the total
810 * number of unused children, because select_parent can
811 * drop the lock and return early due to latency
814 static int select_parent(struct dentry
* parent
)
816 struct dentry
*this_parent
= parent
;
817 struct list_head
*next
;
820 spin_lock(&dcache_lock
);
822 next
= this_parent
->d_subdirs
.next
;
824 while (next
!= &this_parent
->d_subdirs
) {
825 struct list_head
*tmp
= next
;
826 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
829 dentry_lru_del_init(dentry
);
831 * move only zero ref count dentries to the end
832 * of the unused list for prune_dcache
834 if (!atomic_read(&dentry
->d_count
)) {
835 dentry_lru_add_tail(dentry
);
840 * We can return to the caller if we have found some (this
841 * ensures forward progress). We'll be coming back to find
844 if (found
&& need_resched())
848 * Descend a level if the d_subdirs list is non-empty.
850 if (!list_empty(&dentry
->d_subdirs
)) {
851 this_parent
= dentry
;
856 * All done at this level ... ascend and resume the search.
858 if (this_parent
!= parent
) {
859 next
= this_parent
->d_u
.d_child
.next
;
860 this_parent
= this_parent
->d_parent
;
864 spin_unlock(&dcache_lock
);
869 * shrink_dcache_parent - prune dcache
870 * @parent: parent of entries to prune
872 * Prune the dcache to remove unused children of the parent dentry.
875 void shrink_dcache_parent(struct dentry
* parent
)
877 struct super_block
*sb
= parent
->d_sb
;
880 while ((found
= select_parent(parent
)) != 0)
881 __shrink_dcache_sb(sb
, &found
, 0);
883 EXPORT_SYMBOL(shrink_dcache_parent
);
886 * Scan `nr' dentries and return the number which remain.
888 * We need to avoid reentering the filesystem if the caller is performing a
889 * GFP_NOFS allocation attempt. One example deadlock is:
891 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
892 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
893 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
895 * In this case we return -1 to tell the caller that we baled.
897 static int shrink_dcache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
900 if (!(gfp_mask
& __GFP_FS
))
904 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
907 static struct shrinker dcache_shrinker
= {
908 .shrink
= shrink_dcache_memory
,
909 .seeks
= DEFAULT_SEEKS
,
913 * d_alloc - allocate a dcache entry
914 * @parent: parent of entry to allocate
915 * @name: qstr of the name
917 * Allocates a dentry. It returns %NULL if there is insufficient memory
918 * available. On a success the dentry is returned. The name passed in is
919 * copied and the copy passed in may be reused after this call.
922 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
924 struct dentry
*dentry
;
927 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
931 if (name
->len
> DNAME_INLINE_LEN
-1) {
932 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
934 kmem_cache_free(dentry_cache
, dentry
);
938 dname
= dentry
->d_iname
;
940 dentry
->d_name
.name
= dname
;
942 dentry
->d_name
.len
= name
->len
;
943 dentry
->d_name
.hash
= name
->hash
;
944 memcpy(dname
, name
->name
, name
->len
);
945 dname
[name
->len
] = 0;
947 atomic_set(&dentry
->d_count
, 1);
948 dentry
->d_flags
= DCACHE_UNHASHED
;
949 spin_lock_init(&dentry
->d_lock
);
950 dentry
->d_inode
= NULL
;
951 dentry
->d_parent
= NULL
;
954 dentry
->d_fsdata
= NULL
;
955 dentry
->d_mounted
= 0;
956 INIT_HLIST_NODE(&dentry
->d_hash
);
957 INIT_LIST_HEAD(&dentry
->d_lru
);
958 INIT_LIST_HEAD(&dentry
->d_subdirs
);
959 INIT_LIST_HEAD(&dentry
->d_alias
);
962 dentry
->d_parent
= dget(parent
);
963 dentry
->d_sb
= parent
->d_sb
;
965 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
968 spin_lock(&dcache_lock
);
970 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
971 dentry_stat
.nr_dentry
++;
972 spin_unlock(&dcache_lock
);
976 EXPORT_SYMBOL(d_alloc
);
978 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
983 q
.len
= strlen(name
);
984 q
.hash
= full_name_hash(q
.name
, q
.len
);
985 return d_alloc(parent
, &q
);
987 EXPORT_SYMBOL(d_alloc_name
);
989 /* the caller must hold dcache_lock */
990 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
993 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
994 dentry
->d_inode
= inode
;
995 fsnotify_d_instantiate(dentry
, inode
);
999 * d_instantiate - fill in inode information for a dentry
1000 * @entry: dentry to complete
1001 * @inode: inode to attach to this dentry
1003 * Fill in inode information in the entry.
1005 * This turns negative dentries into productive full members
1008 * NOTE! This assumes that the inode count has been incremented
1009 * (or otherwise set) by the caller to indicate that it is now
1010 * in use by the dcache.
1013 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1015 BUG_ON(!list_empty(&entry
->d_alias
));
1016 spin_lock(&dcache_lock
);
1017 __d_instantiate(entry
, inode
);
1018 spin_unlock(&dcache_lock
);
1019 security_d_instantiate(entry
, inode
);
1021 EXPORT_SYMBOL(d_instantiate
);
1024 * d_instantiate_unique - instantiate a non-aliased dentry
1025 * @entry: dentry to instantiate
1026 * @inode: inode to attach to this dentry
1028 * Fill in inode information in the entry. On success, it returns NULL.
1029 * If an unhashed alias of "entry" already exists, then we return the
1030 * aliased dentry instead and drop one reference to inode.
1032 * Note that in order to avoid conflicts with rename() etc, the caller
1033 * had better be holding the parent directory semaphore.
1035 * This also assumes that the inode count has been incremented
1036 * (or otherwise set) by the caller to indicate that it is now
1037 * in use by the dcache.
1039 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1040 struct inode
*inode
)
1042 struct dentry
*alias
;
1043 int len
= entry
->d_name
.len
;
1044 const char *name
= entry
->d_name
.name
;
1045 unsigned int hash
= entry
->d_name
.hash
;
1048 __d_instantiate(entry
, NULL
);
1052 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1053 struct qstr
*qstr
= &alias
->d_name
;
1055 if (qstr
->hash
!= hash
)
1057 if (alias
->d_parent
!= entry
->d_parent
)
1059 if (qstr
->len
!= len
)
1061 if (memcmp(qstr
->name
, name
, len
))
1067 __d_instantiate(entry
, inode
);
1071 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1073 struct dentry
*result
;
1075 BUG_ON(!list_empty(&entry
->d_alias
));
1077 spin_lock(&dcache_lock
);
1078 result
= __d_instantiate_unique(entry
, inode
);
1079 spin_unlock(&dcache_lock
);
1082 security_d_instantiate(entry
, inode
);
1086 BUG_ON(!d_unhashed(result
));
1091 EXPORT_SYMBOL(d_instantiate_unique
);
1094 * d_alloc_root - allocate root dentry
1095 * @root_inode: inode to allocate the root for
1097 * Allocate a root ("/") dentry for the inode given. The inode is
1098 * instantiated and returned. %NULL is returned if there is insufficient
1099 * memory or the inode passed is %NULL.
1102 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1104 struct dentry
*res
= NULL
;
1107 static const struct qstr name
= { .name
= "/", .len
= 1 };
1109 res
= d_alloc(NULL
, &name
);
1111 res
->d_sb
= root_inode
->i_sb
;
1112 res
->d_parent
= res
;
1113 d_instantiate(res
, root_inode
);
1118 EXPORT_SYMBOL(d_alloc_root
);
1120 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1123 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1124 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1125 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1129 * d_obtain_alias - find or allocate a dentry for a given inode
1130 * @inode: inode to allocate the dentry for
1132 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1133 * similar open by handle operations. The returned dentry may be anonymous,
1134 * or may have a full name (if the inode was already in the cache).
1136 * When called on a directory inode, we must ensure that the inode only ever
1137 * has one dentry. If a dentry is found, that is returned instead of
1138 * allocating a new one.
1140 * On successful return, the reference to the inode has been transferred
1141 * to the dentry. In case of an error the reference on the inode is released.
1142 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1143 * be passed in and will be the error will be propagate to the return value,
1144 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1146 struct dentry
*d_obtain_alias(struct inode
*inode
)
1148 static const struct qstr anonstring
= { .name
= "" };
1153 return ERR_PTR(-ESTALE
);
1155 return ERR_CAST(inode
);
1157 res
= d_find_alias(inode
);
1161 tmp
= d_alloc(NULL
, &anonstring
);
1163 res
= ERR_PTR(-ENOMEM
);
1166 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1168 spin_lock(&dcache_lock
);
1169 res
= __d_find_alias(inode
, 0);
1171 spin_unlock(&dcache_lock
);
1176 /* attach a disconnected dentry */
1177 spin_lock(&tmp
->d_lock
);
1178 tmp
->d_sb
= inode
->i_sb
;
1179 tmp
->d_inode
= inode
;
1180 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1181 tmp
->d_flags
&= ~DCACHE_UNHASHED
;
1182 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1183 hlist_add_head(&tmp
->d_hash
, &inode
->i_sb
->s_anon
);
1184 spin_unlock(&tmp
->d_lock
);
1186 spin_unlock(&dcache_lock
);
1193 EXPORT_SYMBOL(d_obtain_alias
);
1196 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1197 * @inode: the inode which may have a disconnected dentry
1198 * @dentry: a negative dentry which we want to point to the inode.
1200 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1201 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1202 * and return it, else simply d_add the inode to the dentry and return NULL.
1204 * This is needed in the lookup routine of any filesystem that is exportable
1205 * (via knfsd) so that we can build dcache paths to directories effectively.
1207 * If a dentry was found and moved, then it is returned. Otherwise NULL
1208 * is returned. This matches the expected return value of ->lookup.
1211 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1213 struct dentry
*new = NULL
;
1215 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1216 spin_lock(&dcache_lock
);
1217 new = __d_find_alias(inode
, 1);
1219 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1220 spin_unlock(&dcache_lock
);
1221 security_d_instantiate(new, inode
);
1222 d_move(new, dentry
);
1225 /* already taking dcache_lock, so d_add() by hand */
1226 __d_instantiate(dentry
, inode
);
1227 spin_unlock(&dcache_lock
);
1228 security_d_instantiate(dentry
, inode
);
1232 d_add(dentry
, inode
);
1235 EXPORT_SYMBOL(d_splice_alias
);
1238 * d_add_ci - lookup or allocate new dentry with case-exact name
1239 * @inode: the inode case-insensitive lookup has found
1240 * @dentry: the negative dentry that was passed to the parent's lookup func
1241 * @name: the case-exact name to be associated with the returned dentry
1243 * This is to avoid filling the dcache with case-insensitive names to the
1244 * same inode, only the actual correct case is stored in the dcache for
1245 * case-insensitive filesystems.
1247 * For a case-insensitive lookup match and if the the case-exact dentry
1248 * already exists in in the dcache, use it and return it.
1250 * If no entry exists with the exact case name, allocate new dentry with
1251 * the exact case, and return the spliced entry.
1253 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1257 struct dentry
*found
;
1261 * First check if a dentry matching the name already exists,
1262 * if not go ahead and create it now.
1264 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1266 new = d_alloc(dentry
->d_parent
, name
);
1272 found
= d_splice_alias(inode
, new);
1281 * If a matching dentry exists, and it's not negative use it.
1283 * Decrement the reference count to balance the iget() done
1286 if (found
->d_inode
) {
1287 if (unlikely(found
->d_inode
!= inode
)) {
1288 /* This can't happen because bad inodes are unhashed. */
1289 BUG_ON(!is_bad_inode(inode
));
1290 BUG_ON(!is_bad_inode(found
->d_inode
));
1297 * Negative dentry: instantiate it unless the inode is a directory and
1298 * already has a dentry.
1300 spin_lock(&dcache_lock
);
1301 if (!S_ISDIR(inode
->i_mode
) || list_empty(&inode
->i_dentry
)) {
1302 __d_instantiate(found
, inode
);
1303 spin_unlock(&dcache_lock
);
1304 security_d_instantiate(found
, inode
);
1309 * In case a directory already has a (disconnected) entry grab a
1310 * reference to it, move it in place and use it.
1312 new = list_entry(inode
->i_dentry
.next
, struct dentry
, d_alias
);
1314 spin_unlock(&dcache_lock
);
1315 security_d_instantiate(found
, inode
);
1323 return ERR_PTR(error
);
1325 EXPORT_SYMBOL(d_add_ci
);
1328 * d_lookup - search for a dentry
1329 * @parent: parent dentry
1330 * @name: qstr of name we wish to find
1331 * Returns: dentry, or NULL
1333 * d_lookup searches the children of the parent dentry for the name in
1334 * question. If the dentry is found its reference count is incremented and the
1335 * dentry is returned. The caller must use dput to free the entry when it has
1336 * finished using it. %NULL is returned if the dentry does not exist.
1338 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1340 struct dentry
* dentry
= NULL
;
1344 seq
= read_seqbegin(&rename_lock
);
1345 dentry
= __d_lookup(parent
, name
);
1348 } while (read_seqretry(&rename_lock
, seq
));
1351 EXPORT_SYMBOL(d_lookup
);
1354 * __d_lookup - search for a dentry (racy)
1355 * @parent: parent dentry
1356 * @name: qstr of name we wish to find
1357 * Returns: dentry, or NULL
1359 * __d_lookup is like d_lookup, however it may (rarely) return a
1360 * false-negative result due to unrelated rename activity.
1362 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1363 * however it must be used carefully, eg. with a following d_lookup in
1364 * the case of failure.
1366 * __d_lookup callers must be commented.
1368 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1370 unsigned int len
= name
->len
;
1371 unsigned int hash
= name
->hash
;
1372 const unsigned char *str
= name
->name
;
1373 struct hlist_head
*head
= d_hash(parent
,hash
);
1374 struct dentry
*found
= NULL
;
1375 struct hlist_node
*node
;
1376 struct dentry
*dentry
;
1379 * The hash list is protected using RCU.
1381 * Take d_lock when comparing a candidate dentry, to avoid races
1384 * It is possible that concurrent renames can mess up our list
1385 * walk here and result in missing our dentry, resulting in the
1386 * false-negative result. d_lookup() protects against concurrent
1387 * renames using rename_lock seqlock.
1389 * See Documentation/vfs/dcache-locking.txt for more details.
1393 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1396 if (dentry
->d_name
.hash
!= hash
)
1398 if (dentry
->d_parent
!= parent
)
1401 spin_lock(&dentry
->d_lock
);
1404 * Recheck the dentry after taking the lock - d_move may have
1405 * changed things. Don't bother checking the hash because
1406 * we're about to compare the whole name anyway.
1408 if (dentry
->d_parent
!= parent
)
1411 /* non-existing due to RCU? */
1412 if (d_unhashed(dentry
))
1416 * It is safe to compare names since d_move() cannot
1417 * change the qstr (protected by d_lock).
1419 qstr
= &dentry
->d_name
;
1420 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1421 if (parent
->d_op
->d_compare(parent
, qstr
, name
))
1424 if (qstr
->len
!= len
)
1426 if (memcmp(qstr
->name
, str
, len
))
1430 atomic_inc(&dentry
->d_count
);
1432 spin_unlock(&dentry
->d_lock
);
1435 spin_unlock(&dentry
->d_lock
);
1443 * d_hash_and_lookup - hash the qstr then search for a dentry
1444 * @dir: Directory to search in
1445 * @name: qstr of name we wish to find
1447 * On hash failure or on lookup failure NULL is returned.
1449 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1451 struct dentry
*dentry
= NULL
;
1454 * Check for a fs-specific hash function. Note that we must
1455 * calculate the standard hash first, as the d_op->d_hash()
1456 * routine may choose to leave the hash value unchanged.
1458 name
->hash
= full_name_hash(name
->name
, name
->len
);
1459 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1460 if (dir
->d_op
->d_hash(dir
, name
) < 0)
1463 dentry
= d_lookup(dir
, name
);
1469 * d_validate - verify dentry provided from insecure source
1470 * @dentry: The dentry alleged to be valid child of @dparent
1471 * @dparent: The parent dentry (known to be valid)
1473 * An insecure source has sent us a dentry, here we verify it and dget() it.
1474 * This is used by ncpfs in its readdir implementation.
1475 * Zero is returned in the dentry is invalid.
1478 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1480 struct hlist_head
*base
;
1481 struct hlist_node
*lhp
;
1483 /* Check whether the ptr might be valid at all.. */
1484 if (!kmem_ptr_validate(dentry_cache
, dentry
))
1487 if (dentry
->d_parent
!= dparent
)
1490 spin_lock(&dcache_lock
);
1491 base
= d_hash(dparent
, dentry
->d_name
.hash
);
1492 hlist_for_each(lhp
,base
) {
1493 /* hlist_for_each_entry_rcu() not required for d_hash list
1494 * as it is parsed under dcache_lock
1496 if (dentry
== hlist_entry(lhp
, struct dentry
, d_hash
)) {
1497 __dget_locked(dentry
);
1498 spin_unlock(&dcache_lock
);
1502 spin_unlock(&dcache_lock
);
1506 EXPORT_SYMBOL(d_validate
);
1509 * When a file is deleted, we have two options:
1510 * - turn this dentry into a negative dentry
1511 * - unhash this dentry and free it.
1513 * Usually, we want to just turn this into
1514 * a negative dentry, but if anybody else is
1515 * currently using the dentry or the inode
1516 * we can't do that and we fall back on removing
1517 * it from the hash queues and waiting for
1518 * it to be deleted later when it has no users
1522 * d_delete - delete a dentry
1523 * @dentry: The dentry to delete
1525 * Turn the dentry into a negative dentry if possible, otherwise
1526 * remove it from the hash queues so it can be deleted later
1529 void d_delete(struct dentry
* dentry
)
1533 * Are we the only user?
1535 spin_lock(&dcache_lock
);
1536 spin_lock(&dentry
->d_lock
);
1537 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1538 if (atomic_read(&dentry
->d_count
) == 1) {
1539 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1540 dentry_iput(dentry
);
1541 fsnotify_nameremove(dentry
, isdir
);
1545 if (!d_unhashed(dentry
))
1548 spin_unlock(&dentry
->d_lock
);
1549 spin_unlock(&dcache_lock
);
1551 fsnotify_nameremove(dentry
, isdir
);
1553 EXPORT_SYMBOL(d_delete
);
1555 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1558 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1559 hlist_add_head_rcu(&entry
->d_hash
, list
);
1562 static void _d_rehash(struct dentry
* entry
)
1564 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1568 * d_rehash - add an entry back to the hash
1569 * @entry: dentry to add to the hash
1571 * Adds a dentry to the hash according to its name.
1574 void d_rehash(struct dentry
* entry
)
1576 spin_lock(&dcache_lock
);
1577 spin_lock(&entry
->d_lock
);
1579 spin_unlock(&entry
->d_lock
);
1580 spin_unlock(&dcache_lock
);
1582 EXPORT_SYMBOL(d_rehash
);
1585 * When switching names, the actual string doesn't strictly have to
1586 * be preserved in the target - because we're dropping the target
1587 * anyway. As such, we can just do a simple memcpy() to copy over
1588 * the new name before we switch.
1590 * Note that we have to be a lot more careful about getting the hash
1591 * switched - we have to switch the hash value properly even if it
1592 * then no longer matches the actual (corrupted) string of the target.
1593 * The hash value has to match the hash queue that the dentry is on..
1595 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1597 if (dname_external(target
)) {
1598 if (dname_external(dentry
)) {
1600 * Both external: swap the pointers
1602 swap(target
->d_name
.name
, dentry
->d_name
.name
);
1605 * dentry:internal, target:external. Steal target's
1606 * storage and make target internal.
1608 memcpy(target
->d_iname
, dentry
->d_name
.name
,
1609 dentry
->d_name
.len
+ 1);
1610 dentry
->d_name
.name
= target
->d_name
.name
;
1611 target
->d_name
.name
= target
->d_iname
;
1614 if (dname_external(dentry
)) {
1616 * dentry:external, target:internal. Give dentry's
1617 * storage to target and make dentry internal
1619 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1620 target
->d_name
.len
+ 1);
1621 target
->d_name
.name
= dentry
->d_name
.name
;
1622 dentry
->d_name
.name
= dentry
->d_iname
;
1625 * Both are internal. Just copy target to dentry
1627 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1628 target
->d_name
.len
+ 1);
1629 dentry
->d_name
.len
= target
->d_name
.len
;
1633 swap(dentry
->d_name
.len
, target
->d_name
.len
);
1637 * We cannibalize "target" when moving dentry on top of it,
1638 * because it's going to be thrown away anyway. We could be more
1639 * polite about it, though.
1641 * This forceful removal will result in ugly /proc output if
1642 * somebody holds a file open that got deleted due to a rename.
1643 * We could be nicer about the deleted file, and let it show
1644 * up under the name it had before it was deleted rather than
1645 * under the original name of the file that was moved on top of it.
1649 * d_move_locked - move a dentry
1650 * @dentry: entry to move
1651 * @target: new dentry
1653 * Update the dcache to reflect the move of a file name. Negative
1654 * dcache entries should not be moved in this way.
1656 static void d_move_locked(struct dentry
* dentry
, struct dentry
* target
)
1658 struct hlist_head
*list
;
1660 if (!dentry
->d_inode
)
1661 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1663 write_seqlock(&rename_lock
);
1665 * XXXX: do we really need to take target->d_lock?
1667 if (target
< dentry
) {
1668 spin_lock(&target
->d_lock
);
1669 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1671 spin_lock(&dentry
->d_lock
);
1672 spin_lock_nested(&target
->d_lock
, DENTRY_D_LOCK_NESTED
);
1675 /* Move the dentry to the target hash queue, if on different bucket */
1676 if (d_unhashed(dentry
))
1677 goto already_unhashed
;
1679 hlist_del_rcu(&dentry
->d_hash
);
1682 list
= d_hash(target
->d_parent
, target
->d_name
.hash
);
1683 __d_rehash(dentry
, list
);
1685 /* Unhash the target: dput() will then get rid of it */
1688 list_del(&dentry
->d_u
.d_child
);
1689 list_del(&target
->d_u
.d_child
);
1691 /* Switch the names.. */
1692 switch_names(dentry
, target
);
1693 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
1695 /* ... and switch the parents */
1696 if (IS_ROOT(dentry
)) {
1697 dentry
->d_parent
= target
->d_parent
;
1698 target
->d_parent
= target
;
1699 INIT_LIST_HEAD(&target
->d_u
.d_child
);
1701 swap(dentry
->d_parent
, target
->d_parent
);
1703 /* And add them back to the (new) parent lists */
1704 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
1707 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1708 spin_unlock(&target
->d_lock
);
1709 fsnotify_d_move(dentry
);
1710 spin_unlock(&dentry
->d_lock
);
1711 write_sequnlock(&rename_lock
);
1715 * d_move - move a dentry
1716 * @dentry: entry to move
1717 * @target: new dentry
1719 * Update the dcache to reflect the move of a file name. Negative
1720 * dcache entries should not be moved in this way.
1723 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1725 spin_lock(&dcache_lock
);
1726 d_move_locked(dentry
, target
);
1727 spin_unlock(&dcache_lock
);
1729 EXPORT_SYMBOL(d_move
);
1732 * d_ancestor - search for an ancestor
1733 * @p1: ancestor dentry
1736 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1737 * an ancestor of p2, else NULL.
1739 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
1743 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
1744 if (p
->d_parent
== p1
)
1751 * This helper attempts to cope with remotely renamed directories
1753 * It assumes that the caller is already holding
1754 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1756 * Note: If ever the locking in lock_rename() changes, then please
1757 * remember to update this too...
1759 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
1760 __releases(dcache_lock
)
1762 struct mutex
*m1
= NULL
, *m2
= NULL
;
1765 /* If alias and dentry share a parent, then no extra locks required */
1766 if (alias
->d_parent
== dentry
->d_parent
)
1769 /* Check for loops */
1770 ret
= ERR_PTR(-ELOOP
);
1771 if (d_ancestor(alias
, dentry
))
1774 /* See lock_rename() */
1775 ret
= ERR_PTR(-EBUSY
);
1776 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
1778 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
1779 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
1781 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
1783 d_move_locked(alias
, dentry
);
1786 spin_unlock(&dcache_lock
);
1795 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1796 * named dentry in place of the dentry to be replaced.
1798 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
1800 struct dentry
*dparent
, *aparent
;
1802 switch_names(dentry
, anon
);
1803 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
1805 dparent
= dentry
->d_parent
;
1806 aparent
= anon
->d_parent
;
1808 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
1809 list_del(&dentry
->d_u
.d_child
);
1810 if (!IS_ROOT(dentry
))
1811 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1813 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1815 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
1816 list_del(&anon
->d_u
.d_child
);
1818 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
1820 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
1822 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
1826 * d_materialise_unique - introduce an inode into the tree
1827 * @dentry: candidate dentry
1828 * @inode: inode to bind to the dentry, to which aliases may be attached
1830 * Introduces an dentry into the tree, substituting an extant disconnected
1831 * root directory alias in its place if there is one
1833 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
1835 struct dentry
*actual
;
1837 BUG_ON(!d_unhashed(dentry
));
1839 spin_lock(&dcache_lock
);
1843 __d_instantiate(dentry
, NULL
);
1847 if (S_ISDIR(inode
->i_mode
)) {
1848 struct dentry
*alias
;
1850 /* Does an aliased dentry already exist? */
1851 alias
= __d_find_alias(inode
, 0);
1854 /* Is this an anonymous mountpoint that we could splice
1856 if (IS_ROOT(alias
)) {
1857 spin_lock(&alias
->d_lock
);
1858 __d_materialise_dentry(dentry
, alias
);
1862 /* Nope, but we must(!) avoid directory aliasing */
1863 actual
= __d_unalias(dentry
, alias
);
1870 /* Add a unique reference */
1871 actual
= __d_instantiate_unique(dentry
, inode
);
1874 else if (unlikely(!d_unhashed(actual
)))
1875 goto shouldnt_be_hashed
;
1878 spin_lock(&actual
->d_lock
);
1881 spin_unlock(&actual
->d_lock
);
1882 spin_unlock(&dcache_lock
);
1884 if (actual
== dentry
) {
1885 security_d_instantiate(dentry
, inode
);
1893 spin_unlock(&dcache_lock
);
1896 EXPORT_SYMBOL_GPL(d_materialise_unique
);
1898 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
1902 return -ENAMETOOLONG
;
1904 memcpy(*buffer
, str
, namelen
);
1908 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
1910 return prepend(buffer
, buflen
, name
->name
, name
->len
);
1914 * Prepend path string to a buffer
1916 * @path: the dentry/vfsmount to report
1917 * @root: root vfsmnt/dentry (may be modified by this function)
1918 * @buffer: pointer to the end of the buffer
1919 * @buflen: pointer to buffer length
1921 * Caller holds the dcache_lock.
1923 * If path is not reachable from the supplied root, then the value of
1924 * root is changed (without modifying refcounts).
1926 static int prepend_path(const struct path
*path
, struct path
*root
,
1927 char **buffer
, int *buflen
)
1929 struct dentry
*dentry
= path
->dentry
;
1930 struct vfsmount
*vfsmnt
= path
->mnt
;
1934 br_read_lock(vfsmount_lock
);
1935 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
1936 struct dentry
* parent
;
1938 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
1940 if (vfsmnt
->mnt_parent
== vfsmnt
) {
1943 dentry
= vfsmnt
->mnt_mountpoint
;
1944 vfsmnt
= vfsmnt
->mnt_parent
;
1947 parent
= dentry
->d_parent
;
1949 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
1951 error
= prepend(buffer
, buflen
, "/", 1);
1960 if (!error
&& !slash
)
1961 error
= prepend(buffer
, buflen
, "/", 1);
1963 br_read_unlock(vfsmount_lock
);
1968 * Filesystems needing to implement special "root names"
1969 * should do so with ->d_dname()
1971 if (IS_ROOT(dentry
) &&
1972 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
1973 WARN(1, "Root dentry has weird name <%.*s>\n",
1974 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
1977 root
->dentry
= dentry
;
1982 * __d_path - return the path of a dentry
1983 * @path: the dentry/vfsmount to report
1984 * @root: root vfsmnt/dentry (may be modified by this function)
1985 * @buf: buffer to return value in
1986 * @buflen: buffer length
1988 * Convert a dentry into an ASCII path name.
1990 * Returns a pointer into the buffer or an error code if the
1991 * path was too long.
1993 * "buflen" should be positive.
1995 * If path is not reachable from the supplied root, then the value of
1996 * root is changed (without modifying refcounts).
1998 char *__d_path(const struct path
*path
, struct path
*root
,
1999 char *buf
, int buflen
)
2001 char *res
= buf
+ buflen
;
2004 prepend(&res
, &buflen
, "\0", 1);
2005 spin_lock(&dcache_lock
);
2006 error
= prepend_path(path
, root
, &res
, &buflen
);
2007 spin_unlock(&dcache_lock
);
2010 return ERR_PTR(error
);
2015 * same as __d_path but appends "(deleted)" for unlinked files.
2017 static int path_with_deleted(const struct path
*path
, struct path
*root
,
2018 char **buf
, int *buflen
)
2020 prepend(buf
, buflen
, "\0", 1);
2021 if (d_unlinked(path
->dentry
)) {
2022 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2027 return prepend_path(path
, root
, buf
, buflen
);
2030 static int prepend_unreachable(char **buffer
, int *buflen
)
2032 return prepend(buffer
, buflen
, "(unreachable)", 13);
2036 * d_path - return the path of a dentry
2037 * @path: path to report
2038 * @buf: buffer to return value in
2039 * @buflen: buffer length
2041 * Convert a dentry into an ASCII path name. If the entry has been deleted
2042 * the string " (deleted)" is appended. Note that this is ambiguous.
2044 * Returns a pointer into the buffer or an error code if the path was
2045 * too long. Note: Callers should use the returned pointer, not the passed
2046 * in buffer, to use the name! The implementation often starts at an offset
2047 * into the buffer, and may leave 0 bytes at the start.
2049 * "buflen" should be positive.
2051 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2053 char *res
= buf
+ buflen
;
2059 * We have various synthetic filesystems that never get mounted. On
2060 * these filesystems dentries are never used for lookup purposes, and
2061 * thus don't need to be hashed. They also don't need a name until a
2062 * user wants to identify the object in /proc/pid/fd/. The little hack
2063 * below allows us to generate a name for these objects on demand:
2065 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2066 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2068 get_fs_root(current
->fs
, &root
);
2069 spin_lock(&dcache_lock
);
2071 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2073 res
= ERR_PTR(error
);
2074 spin_unlock(&dcache_lock
);
2078 EXPORT_SYMBOL(d_path
);
2081 * d_path_with_unreachable - return the path of a dentry
2082 * @path: path to report
2083 * @buf: buffer to return value in
2084 * @buflen: buffer length
2086 * The difference from d_path() is that this prepends "(unreachable)"
2087 * to paths which are unreachable from the current process' root.
2089 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2091 char *res
= buf
+ buflen
;
2096 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2097 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2099 get_fs_root(current
->fs
, &root
);
2100 spin_lock(&dcache_lock
);
2102 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2103 if (!error
&& !path_equal(&tmp
, &root
))
2104 error
= prepend_unreachable(&res
, &buflen
);
2105 spin_unlock(&dcache_lock
);
2108 res
= ERR_PTR(error
);
2114 * Helper function for dentry_operations.d_dname() members
2116 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2117 const char *fmt
, ...)
2123 va_start(args
, fmt
);
2124 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2127 if (sz
> sizeof(temp
) || sz
> buflen
)
2128 return ERR_PTR(-ENAMETOOLONG
);
2130 buffer
+= buflen
- sz
;
2131 return memcpy(buffer
, temp
, sz
);
2135 * Write full pathname from the root of the filesystem into the buffer.
2137 char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2139 char *end
= buf
+ buflen
;
2142 prepend(&end
, &buflen
, "\0", 1);
2149 while (!IS_ROOT(dentry
)) {
2150 struct dentry
*parent
= dentry
->d_parent
;
2153 if ((prepend_name(&end
, &buflen
, &dentry
->d_name
) != 0) ||
2154 (prepend(&end
, &buflen
, "/", 1) != 0))
2162 return ERR_PTR(-ENAMETOOLONG
);
2164 EXPORT_SYMBOL(__dentry_path
);
2166 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2171 spin_lock(&dcache_lock
);
2172 if (d_unlinked(dentry
)) {
2174 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2178 retval
= __dentry_path(dentry
, buf
, buflen
);
2179 spin_unlock(&dcache_lock
);
2180 if (!IS_ERR(retval
) && p
)
2181 *p
= '/'; /* restore '/' overriden with '\0' */
2184 spin_unlock(&dcache_lock
);
2185 return ERR_PTR(-ENAMETOOLONG
);
2189 * NOTE! The user-level library version returns a
2190 * character pointer. The kernel system call just
2191 * returns the length of the buffer filled (which
2192 * includes the ending '\0' character), or a negative
2193 * error value. So libc would do something like
2195 * char *getcwd(char * buf, size_t size)
2199 * retval = sys_getcwd(buf, size);
2206 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2209 struct path pwd
, root
;
2210 char *page
= (char *) __get_free_page(GFP_USER
);
2215 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2218 spin_lock(&dcache_lock
);
2219 if (!d_unlinked(pwd
.dentry
)) {
2221 struct path tmp
= root
;
2222 char *cwd
= page
+ PAGE_SIZE
;
2223 int buflen
= PAGE_SIZE
;
2225 prepend(&cwd
, &buflen
, "\0", 1);
2226 error
= prepend_path(&pwd
, &tmp
, &cwd
, &buflen
);
2227 spin_unlock(&dcache_lock
);
2232 /* Unreachable from current root */
2233 if (!path_equal(&tmp
, &root
)) {
2234 error
= prepend_unreachable(&cwd
, &buflen
);
2240 len
= PAGE_SIZE
+ page
- cwd
;
2243 if (copy_to_user(buf
, cwd
, len
))
2247 spin_unlock(&dcache_lock
);
2252 free_page((unsigned long) page
);
2257 * Test whether new_dentry is a subdirectory of old_dentry.
2259 * Trivially implemented using the dcache structure
2263 * is_subdir - is new dentry a subdirectory of old_dentry
2264 * @new_dentry: new dentry
2265 * @old_dentry: old dentry
2267 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2268 * Returns 0 otherwise.
2269 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2272 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2277 if (new_dentry
== old_dentry
)
2281 * Need rcu_readlock to protect against the d_parent trashing
2286 /* for restarting inner loop in case of seq retry */
2287 seq
= read_seqbegin(&rename_lock
);
2288 if (d_ancestor(old_dentry
, new_dentry
))
2292 } while (read_seqretry(&rename_lock
, seq
));
2298 int path_is_under(struct path
*path1
, struct path
*path2
)
2300 struct vfsmount
*mnt
= path1
->mnt
;
2301 struct dentry
*dentry
= path1
->dentry
;
2304 br_read_lock(vfsmount_lock
);
2305 if (mnt
!= path2
->mnt
) {
2307 if (mnt
->mnt_parent
== mnt
) {
2308 br_read_unlock(vfsmount_lock
);
2311 if (mnt
->mnt_parent
== path2
->mnt
)
2313 mnt
= mnt
->mnt_parent
;
2315 dentry
= mnt
->mnt_mountpoint
;
2317 res
= is_subdir(dentry
, path2
->dentry
);
2318 br_read_unlock(vfsmount_lock
);
2321 EXPORT_SYMBOL(path_is_under
);
2323 void d_genocide(struct dentry
*root
)
2325 struct dentry
*this_parent
= root
;
2326 struct list_head
*next
;
2328 spin_lock(&dcache_lock
);
2330 next
= this_parent
->d_subdirs
.next
;
2332 while (next
!= &this_parent
->d_subdirs
) {
2333 struct list_head
*tmp
= next
;
2334 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2336 if (d_unhashed(dentry
)||!dentry
->d_inode
)
2338 if (!list_empty(&dentry
->d_subdirs
)) {
2339 this_parent
= dentry
;
2342 atomic_dec(&dentry
->d_count
);
2344 if (this_parent
!= root
) {
2345 next
= this_parent
->d_u
.d_child
.next
;
2346 atomic_dec(&this_parent
->d_count
);
2347 this_parent
= this_parent
->d_parent
;
2350 spin_unlock(&dcache_lock
);
2354 * find_inode_number - check for dentry with name
2355 * @dir: directory to check
2356 * @name: Name to find.
2358 * Check whether a dentry already exists for the given name,
2359 * and return the inode number if it has an inode. Otherwise
2362 * This routine is used to post-process directory listings for
2363 * filesystems using synthetic inode numbers, and is necessary
2364 * to keep getcwd() working.
2367 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2369 struct dentry
* dentry
;
2372 dentry
= d_hash_and_lookup(dir
, name
);
2374 if (dentry
->d_inode
)
2375 ino
= dentry
->d_inode
->i_ino
;
2380 EXPORT_SYMBOL(find_inode_number
);
2382 static __initdata
unsigned long dhash_entries
;
2383 static int __init
set_dhash_entries(char *str
)
2387 dhash_entries
= simple_strtoul(str
, &str
, 0);
2390 __setup("dhash_entries=", set_dhash_entries
);
2392 static void __init
dcache_init_early(void)
2396 /* If hashes are distributed across NUMA nodes, defer
2397 * hash allocation until vmalloc space is available.
2403 alloc_large_system_hash("Dentry cache",
2404 sizeof(struct hlist_head
),
2412 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2413 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2416 static void __init
dcache_init(void)
2421 * A constructor could be added for stable state like the lists,
2422 * but it is probably not worth it because of the cache nature
2425 dentry_cache
= KMEM_CACHE(dentry
,
2426 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2428 register_shrinker(&dcache_shrinker
);
2430 /* Hash may have been set up in dcache_init_early */
2435 alloc_large_system_hash("Dentry cache",
2436 sizeof(struct hlist_head
),
2444 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2445 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2448 /* SLAB cache for __getname() consumers */
2449 struct kmem_cache
*names_cachep __read_mostly
;
2450 EXPORT_SYMBOL(names_cachep
);
2452 EXPORT_SYMBOL(d_genocide
);
2454 void __init
vfs_caches_init_early(void)
2456 dcache_init_early();
2460 void __init
vfs_caches_init(unsigned long mempages
)
2462 unsigned long reserve
;
2464 /* Base hash sizes on available memory, with a reserve equal to
2465 150% of current kernel size */
2467 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2468 mempages
-= reserve
;
2470 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2471 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2475 files_init(mempages
);