4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
38 int sysctl_vfs_cache_pressure __read_mostly
= 100;
39 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
41 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lock
);
42 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
44 EXPORT_SYMBOL(dcache_lock
);
46 static struct kmem_cache
*dentry_cache __read_mostly
;
48 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
58 #define D_HASHBITS d_hash_shift
59 #define D_HASHMASK d_hash_mask
61 static unsigned int d_hash_mask __read_mostly
;
62 static unsigned int d_hash_shift __read_mostly
;
63 static struct hlist_head
*dentry_hashtable __read_mostly
;
65 /* Statistics gathering. */
66 struct dentry_stat_t dentry_stat
= {
70 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
72 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
73 static int get_nr_dentry(void)
77 for_each_possible_cpu(i
)
78 sum
+= per_cpu(nr_dentry
, i
);
79 return sum
< 0 ? 0 : sum
;
82 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
83 size_t *lenp
, loff_t
*ppos
)
85 dentry_stat
.nr_dentry
= get_nr_dentry();
86 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
90 static void __d_free(struct rcu_head
*head
)
92 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
94 WARN_ON(!list_empty(&dentry
->d_alias
));
95 if (dname_external(dentry
))
96 kfree(dentry
->d_name
.name
);
97 kmem_cache_free(dentry_cache
, dentry
);
101 * no dcache_lock, please.
103 static void d_free(struct dentry
*dentry
)
105 this_cpu_dec(nr_dentry
);
106 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
107 dentry
->d_op
->d_release(dentry
);
109 /* if dentry was never inserted into hash, immediate free is OK */
110 if (hlist_unhashed(&dentry
->d_hash
))
111 __d_free(&dentry
->d_u
.d_rcu
);
113 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
117 * Release the dentry's inode, using the filesystem
118 * d_iput() operation if defined.
120 static void dentry_iput(struct dentry
* dentry
)
121 __releases(dentry
->d_lock
)
122 __releases(dcache_lock
)
124 struct inode
*inode
= dentry
->d_inode
;
126 dentry
->d_inode
= NULL
;
127 list_del_init(&dentry
->d_alias
);
128 spin_unlock(&dentry
->d_lock
);
129 spin_unlock(&dcache_lock
);
131 fsnotify_inoderemove(inode
);
132 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
133 dentry
->d_op
->d_iput(dentry
, inode
);
137 spin_unlock(&dentry
->d_lock
);
138 spin_unlock(&dcache_lock
);
143 * dentry_lru_(add|del|move_tail) must be called with dcache_lock held.
145 static void dentry_lru_add(struct dentry
*dentry
)
147 if (list_empty(&dentry
->d_lru
)) {
148 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
149 dentry
->d_sb
->s_nr_dentry_unused
++;
150 dentry_stat
.nr_unused
++;
154 static void dentry_lru_del(struct dentry
*dentry
)
156 if (!list_empty(&dentry
->d_lru
)) {
157 list_del_init(&dentry
->d_lru
);
158 dentry
->d_sb
->s_nr_dentry_unused
--;
159 dentry_stat
.nr_unused
--;
163 static void dentry_lru_move_tail(struct dentry
*dentry
)
165 if (list_empty(&dentry
->d_lru
)) {
166 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
167 dentry
->d_sb
->s_nr_dentry_unused
++;
168 dentry_stat
.nr_unused
++;
170 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
175 * d_kill - kill dentry and return parent
176 * @dentry: dentry to kill
178 * The dentry must already be unhashed and removed from the LRU.
180 * If this is the root of the dentry tree, return NULL.
182 static struct dentry
*d_kill(struct dentry
*dentry
)
183 __releases(dentry
->d_lock
)
184 __releases(dcache_lock
)
186 struct dentry
*parent
;
188 list_del(&dentry
->d_u
.d_child
);
189 /*drops the locks, at that point nobody can reach this dentry */
194 parent
= dentry
->d_parent
;
202 * This is complicated by the fact that we do not want to put
203 * dentries that are no longer on any hash chain on the unused
204 * list: we'd much rather just get rid of them immediately.
206 * However, that implies that we have to traverse the dentry
207 * tree upwards to the parents which might _also_ now be
208 * scheduled for deletion (it may have been only waiting for
209 * its last child to go away).
211 * This tail recursion is done by hand as we don't want to depend
212 * on the compiler to always get this right (gcc generally doesn't).
213 * Real recursion would eat up our stack space.
217 * dput - release a dentry
218 * @dentry: dentry to release
220 * Release a dentry. This will drop the usage count and if appropriate
221 * call the dentry unlink method as well as removing it from the queues and
222 * releasing its resources. If the parent dentries were scheduled for release
223 * they too may now get deleted.
225 * no dcache lock, please.
228 void dput(struct dentry
*dentry
)
234 if (atomic_read(&dentry
->d_count
) == 1)
236 if (!atomic_dec_and_lock(&dentry
->d_count
, &dcache_lock
))
239 spin_lock(&dentry
->d_lock
);
240 if (atomic_read(&dentry
->d_count
)) {
241 spin_unlock(&dentry
->d_lock
);
242 spin_unlock(&dcache_lock
);
247 * AV: ->d_delete() is _NOT_ allowed to block now.
249 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
250 if (dentry
->d_op
->d_delete(dentry
))
254 /* Unreachable? Get rid of it */
255 if (d_unhashed(dentry
))
258 /* Otherwise leave it cached and ensure it's on the LRU */
259 dentry
->d_flags
|= DCACHE_REFERENCED
;
260 dentry_lru_add(dentry
);
262 spin_unlock(&dentry
->d_lock
);
263 spin_unlock(&dcache_lock
);
269 /* if dentry was on the d_lru list delete it from there */
270 dentry_lru_del(dentry
);
271 dentry
= d_kill(dentry
);
278 * d_invalidate - invalidate a dentry
279 * @dentry: dentry to invalidate
281 * Try to invalidate the dentry if it turns out to be
282 * possible. If there are other dentries that can be
283 * reached through this one we can't delete it and we
284 * return -EBUSY. On success we return 0.
289 int d_invalidate(struct dentry
* dentry
)
292 * If it's already been dropped, return OK.
294 spin_lock(&dcache_lock
);
295 if (d_unhashed(dentry
)) {
296 spin_unlock(&dcache_lock
);
300 * Check whether to do a partial shrink_dcache
301 * to get rid of unused child entries.
303 if (!list_empty(&dentry
->d_subdirs
)) {
304 spin_unlock(&dcache_lock
);
305 shrink_dcache_parent(dentry
);
306 spin_lock(&dcache_lock
);
310 * Somebody else still using it?
312 * If it's a directory, we can't drop it
313 * for fear of somebody re-populating it
314 * with children (even though dropping it
315 * would make it unreachable from the root,
316 * we might still populate it if it was a
317 * working directory or similar).
319 spin_lock(&dentry
->d_lock
);
320 if (atomic_read(&dentry
->d_count
) > 1) {
321 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
322 spin_unlock(&dentry
->d_lock
);
323 spin_unlock(&dcache_lock
);
329 spin_unlock(&dentry
->d_lock
);
330 spin_unlock(&dcache_lock
);
333 EXPORT_SYMBOL(d_invalidate
);
335 /* This should be called _only_ with dcache_lock held */
336 static inline struct dentry
* __dget_locked(struct dentry
*dentry
)
338 atomic_inc(&dentry
->d_count
);
339 dentry_lru_del(dentry
);
343 struct dentry
* dget_locked(struct dentry
*dentry
)
345 return __dget_locked(dentry
);
347 EXPORT_SYMBOL(dget_locked
);
350 * d_find_alias - grab a hashed alias of inode
351 * @inode: inode in question
352 * @want_discon: flag, used by d_splice_alias, to request
353 * that only a DISCONNECTED alias be returned.
355 * If inode has a hashed alias, or is a directory and has any alias,
356 * acquire the reference to alias and return it. Otherwise return NULL.
357 * Notice that if inode is a directory there can be only one alias and
358 * it can be unhashed only if it has no children, or if it is the root
361 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
362 * any other hashed alias over that one unless @want_discon is set,
363 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
366 static struct dentry
* __d_find_alias(struct inode
*inode
, int want_discon
)
368 struct list_head
*head
, *next
, *tmp
;
369 struct dentry
*alias
, *discon_alias
=NULL
;
371 head
= &inode
->i_dentry
;
372 next
= inode
->i_dentry
.next
;
373 while (next
!= head
) {
377 alias
= list_entry(tmp
, struct dentry
, d_alias
);
378 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
379 if (IS_ROOT(alias
) &&
380 (alias
->d_flags
& DCACHE_DISCONNECTED
))
381 discon_alias
= alias
;
382 else if (!want_discon
) {
383 __dget_locked(alias
);
389 __dget_locked(discon_alias
);
393 struct dentry
* d_find_alias(struct inode
*inode
)
395 struct dentry
*de
= NULL
;
397 if (!list_empty(&inode
->i_dentry
)) {
398 spin_lock(&dcache_lock
);
399 de
= __d_find_alias(inode
, 0);
400 spin_unlock(&dcache_lock
);
404 EXPORT_SYMBOL(d_find_alias
);
407 * Try to kill dentries associated with this inode.
408 * WARNING: you must own a reference to inode.
410 void d_prune_aliases(struct inode
*inode
)
412 struct dentry
*dentry
;
414 spin_lock(&dcache_lock
);
415 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
416 spin_lock(&dentry
->d_lock
);
417 if (!atomic_read(&dentry
->d_count
)) {
418 __dget_locked(dentry
);
420 spin_unlock(&dentry
->d_lock
);
421 spin_unlock(&dcache_lock
);
425 spin_unlock(&dentry
->d_lock
);
427 spin_unlock(&dcache_lock
);
429 EXPORT_SYMBOL(d_prune_aliases
);
432 * Throw away a dentry - free the inode, dput the parent. This requires that
433 * the LRU list has already been removed.
435 * Try to prune ancestors as well. This is necessary to prevent
436 * quadratic behavior of shrink_dcache_parent(), but is also expected
437 * to be beneficial in reducing dentry cache fragmentation.
439 static void prune_one_dentry(struct dentry
* dentry
)
440 __releases(dentry
->d_lock
)
441 __releases(dcache_lock
)
442 __acquires(dcache_lock
)
445 dentry
= d_kill(dentry
);
448 * Prune ancestors. Locking is simpler than in dput(),
449 * because dcache_lock needs to be taken anyway.
451 spin_lock(&dcache_lock
);
453 if (!atomic_dec_and_lock(&dentry
->d_count
, &dentry
->d_lock
))
456 if (dentry
->d_op
&& dentry
->d_op
->d_delete
)
457 dentry
->d_op
->d_delete(dentry
);
458 dentry_lru_del(dentry
);
460 dentry
= d_kill(dentry
);
461 spin_lock(&dcache_lock
);
465 static void shrink_dentry_list(struct list_head
*list
)
467 struct dentry
*dentry
;
469 while (!list_empty(list
)) {
470 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
471 dentry_lru_del(dentry
);
474 * We found an inuse dentry which was not removed from
475 * the LRU because of laziness during lookup. Do not free
476 * it - just keep it off the LRU list.
478 spin_lock(&dentry
->d_lock
);
479 if (atomic_read(&dentry
->d_count
)) {
480 spin_unlock(&dentry
->d_lock
);
483 prune_one_dentry(dentry
);
484 /* dentry->d_lock was dropped in prune_one_dentry() */
485 cond_resched_lock(&dcache_lock
);
490 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
491 * @sb: superblock to shrink dentry LRU.
492 * @count: number of entries to prune
493 * @flags: flags to control the dentry processing
495 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
497 static void __shrink_dcache_sb(struct super_block
*sb
, int *count
, int flags
)
499 /* called from prune_dcache() and shrink_dcache_parent() */
500 struct dentry
*dentry
;
501 LIST_HEAD(referenced
);
505 spin_lock(&dcache_lock
);
506 while (!list_empty(&sb
->s_dentry_lru
)) {
507 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
508 struct dentry
, d_lru
);
509 BUG_ON(dentry
->d_sb
!= sb
);
512 * If we are honouring the DCACHE_REFERENCED flag and the
513 * dentry has this flag set, don't free it. Clear the flag
514 * and put it back on the LRU.
516 if (flags
& DCACHE_REFERENCED
) {
517 spin_lock(&dentry
->d_lock
);
518 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
519 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
520 list_move(&dentry
->d_lru
, &referenced
);
521 spin_unlock(&dentry
->d_lock
);
522 cond_resched_lock(&dcache_lock
);
525 spin_unlock(&dentry
->d_lock
);
528 list_move_tail(&dentry
->d_lru
, &tmp
);
531 cond_resched_lock(&dcache_lock
);
535 shrink_dentry_list(&tmp
);
537 if (!list_empty(&referenced
))
538 list_splice(&referenced
, &sb
->s_dentry_lru
);
539 spin_unlock(&dcache_lock
);
544 * prune_dcache - shrink the dcache
545 * @count: number of entries to try to free
547 * Shrink the dcache. This is done when we need more memory, or simply when we
548 * need to unmount something (at which point we need to unuse all dentries).
550 * This function may fail to free any resources if all the dentries are in use.
552 static void prune_dcache(int count
)
554 struct super_block
*sb
, *p
= NULL
;
556 int unused
= dentry_stat
.nr_unused
;
560 if (unused
== 0 || count
== 0)
562 spin_lock(&dcache_lock
);
566 prune_ratio
= unused
/ count
;
568 list_for_each_entry(sb
, &super_blocks
, s_list
) {
569 if (list_empty(&sb
->s_instances
))
571 if (sb
->s_nr_dentry_unused
== 0)
574 /* Now, we reclaim unused dentrins with fairness.
575 * We reclaim them same percentage from each superblock.
576 * We calculate number of dentries to scan on this sb
577 * as follows, but the implementation is arranged to avoid
579 * number of dentries to scan on this sb =
580 * count * (number of dentries on this sb /
581 * number of dentries in the machine)
583 spin_unlock(&sb_lock
);
584 if (prune_ratio
!= 1)
585 w_count
= (sb
->s_nr_dentry_unused
/ prune_ratio
) + 1;
587 w_count
= sb
->s_nr_dentry_unused
;
590 * We need to be sure this filesystem isn't being unmounted,
591 * otherwise we could race with generic_shutdown_super(), and
592 * end up holding a reference to an inode while the filesystem
593 * is unmounted. So we try to get s_umount, and make sure
596 if (down_read_trylock(&sb
->s_umount
)) {
597 if ((sb
->s_root
!= NULL
) &&
598 (!list_empty(&sb
->s_dentry_lru
))) {
599 spin_unlock(&dcache_lock
);
600 __shrink_dcache_sb(sb
, &w_count
,
603 spin_lock(&dcache_lock
);
605 up_read(&sb
->s_umount
);
612 /* more work left to do? */
618 spin_unlock(&sb_lock
);
619 spin_unlock(&dcache_lock
);
623 * shrink_dcache_sb - shrink dcache for a superblock
626 * Shrink the dcache for the specified super block. This is used to free
627 * the dcache before unmounting a file system.
629 void shrink_dcache_sb(struct super_block
*sb
)
633 spin_lock(&dcache_lock
);
634 while (!list_empty(&sb
->s_dentry_lru
)) {
635 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
636 shrink_dentry_list(&tmp
);
638 spin_unlock(&dcache_lock
);
640 EXPORT_SYMBOL(shrink_dcache_sb
);
643 * destroy a single subtree of dentries for unmount
644 * - see the comments on shrink_dcache_for_umount() for a description of the
647 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
649 struct dentry
*parent
;
650 unsigned detached
= 0;
652 BUG_ON(!IS_ROOT(dentry
));
654 /* detach this root from the system */
655 spin_lock(&dcache_lock
);
656 dentry_lru_del(dentry
);
658 spin_unlock(&dcache_lock
);
661 /* descend to the first leaf in the current subtree */
662 while (!list_empty(&dentry
->d_subdirs
)) {
665 /* this is a branch with children - detach all of them
666 * from the system in one go */
667 spin_lock(&dcache_lock
);
668 list_for_each_entry(loop
, &dentry
->d_subdirs
,
670 dentry_lru_del(loop
);
672 cond_resched_lock(&dcache_lock
);
674 spin_unlock(&dcache_lock
);
676 /* move to the first child */
677 dentry
= list_entry(dentry
->d_subdirs
.next
,
678 struct dentry
, d_u
.d_child
);
681 /* consume the dentries from this leaf up through its parents
682 * until we find one with children or run out altogether */
686 if (atomic_read(&dentry
->d_count
) != 0) {
688 "BUG: Dentry %p{i=%lx,n=%s}"
690 " [unmount of %s %s]\n",
693 dentry
->d_inode
->i_ino
: 0UL,
695 atomic_read(&dentry
->d_count
),
696 dentry
->d_sb
->s_type
->name
,
704 parent
= dentry
->d_parent
;
705 atomic_dec(&parent
->d_count
);
708 list_del(&dentry
->d_u
.d_child
);
711 inode
= dentry
->d_inode
;
713 dentry
->d_inode
= NULL
;
714 list_del_init(&dentry
->d_alias
);
715 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
716 dentry
->d_op
->d_iput(dentry
, inode
);
723 /* finished when we fall off the top of the tree,
724 * otherwise we ascend to the parent and move to the
725 * next sibling if there is one */
729 } while (list_empty(&dentry
->d_subdirs
));
731 dentry
= list_entry(dentry
->d_subdirs
.next
,
732 struct dentry
, d_u
.d_child
);
737 * destroy the dentries attached to a superblock on unmounting
738 * - we don't need to use dentry->d_lock, and only need dcache_lock when
739 * removing the dentry from the system lists and hashes because:
740 * - the superblock is detached from all mountings and open files, so the
741 * dentry trees will not be rearranged by the VFS
742 * - s_umount is write-locked, so the memory pressure shrinker will ignore
743 * any dentries belonging to this superblock that it comes across
744 * - the filesystem itself is no longer permitted to rearrange the dentries
747 void shrink_dcache_for_umount(struct super_block
*sb
)
749 struct dentry
*dentry
;
751 if (down_read_trylock(&sb
->s_umount
))
756 atomic_dec(&dentry
->d_count
);
757 shrink_dcache_for_umount_subtree(dentry
);
759 while (!hlist_empty(&sb
->s_anon
)) {
760 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
761 shrink_dcache_for_umount_subtree(dentry
);
766 * Search for at least 1 mount point in the dentry's subdirs.
767 * We descend to the next level whenever the d_subdirs
768 * list is non-empty and continue searching.
772 * have_submounts - check for mounts over a dentry
773 * @parent: dentry to check.
775 * Return true if the parent or its subdirectories contain
779 int have_submounts(struct dentry
*parent
)
781 struct dentry
*this_parent
= parent
;
782 struct list_head
*next
;
784 spin_lock(&dcache_lock
);
785 if (d_mountpoint(parent
))
788 next
= this_parent
->d_subdirs
.next
;
790 while (next
!= &this_parent
->d_subdirs
) {
791 struct list_head
*tmp
= next
;
792 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
794 /* Have we found a mount point ? */
795 if (d_mountpoint(dentry
))
797 if (!list_empty(&dentry
->d_subdirs
)) {
798 this_parent
= dentry
;
803 * All done at this level ... ascend and resume the search.
805 if (this_parent
!= parent
) {
806 next
= this_parent
->d_u
.d_child
.next
;
807 this_parent
= this_parent
->d_parent
;
810 spin_unlock(&dcache_lock
);
811 return 0; /* No mount points found in tree */
813 spin_unlock(&dcache_lock
);
816 EXPORT_SYMBOL(have_submounts
);
819 * Search the dentry child list for the specified parent,
820 * and move any unused dentries to the end of the unused
821 * list for prune_dcache(). We descend to the next level
822 * whenever the d_subdirs list is non-empty and continue
825 * It returns zero iff there are no unused children,
826 * otherwise it returns the number of children moved to
827 * the end of the unused list. This may not be the total
828 * number of unused children, because select_parent can
829 * drop the lock and return early due to latency
832 static int select_parent(struct dentry
* parent
)
834 struct dentry
*this_parent
= parent
;
835 struct list_head
*next
;
838 spin_lock(&dcache_lock
);
840 next
= this_parent
->d_subdirs
.next
;
842 while (next
!= &this_parent
->d_subdirs
) {
843 struct list_head
*tmp
= next
;
844 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
848 * move only zero ref count dentries to the end
849 * of the unused list for prune_dcache
851 if (!atomic_read(&dentry
->d_count
)) {
852 dentry_lru_move_tail(dentry
);
855 dentry_lru_del(dentry
);
859 * We can return to the caller if we have found some (this
860 * ensures forward progress). We'll be coming back to find
863 if (found
&& need_resched())
867 * Descend a level if the d_subdirs list is non-empty.
869 if (!list_empty(&dentry
->d_subdirs
)) {
870 this_parent
= dentry
;
875 * All done at this level ... ascend and resume the search.
877 if (this_parent
!= parent
) {
878 next
= this_parent
->d_u
.d_child
.next
;
879 this_parent
= this_parent
->d_parent
;
883 spin_unlock(&dcache_lock
);
888 * shrink_dcache_parent - prune dcache
889 * @parent: parent of entries to prune
891 * Prune the dcache to remove unused children of the parent dentry.
894 void shrink_dcache_parent(struct dentry
* parent
)
896 struct super_block
*sb
= parent
->d_sb
;
899 while ((found
= select_parent(parent
)) != 0)
900 __shrink_dcache_sb(sb
, &found
, 0);
902 EXPORT_SYMBOL(shrink_dcache_parent
);
905 * Scan `nr' dentries and return the number which remain.
907 * We need to avoid reentering the filesystem if the caller is performing a
908 * GFP_NOFS allocation attempt. One example deadlock is:
910 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
911 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
912 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
914 * In this case we return -1 to tell the caller that we baled.
916 static int shrink_dcache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
919 if (!(gfp_mask
& __GFP_FS
))
924 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
927 static struct shrinker dcache_shrinker
= {
928 .shrink
= shrink_dcache_memory
,
929 .seeks
= DEFAULT_SEEKS
,
933 * d_alloc - allocate a dcache entry
934 * @parent: parent of entry to allocate
935 * @name: qstr of the name
937 * Allocates a dentry. It returns %NULL if there is insufficient memory
938 * available. On a success the dentry is returned. The name passed in is
939 * copied and the copy passed in may be reused after this call.
942 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
944 struct dentry
*dentry
;
947 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
951 if (name
->len
> DNAME_INLINE_LEN
-1) {
952 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
954 kmem_cache_free(dentry_cache
, dentry
);
958 dname
= dentry
->d_iname
;
960 dentry
->d_name
.name
= dname
;
962 dentry
->d_name
.len
= name
->len
;
963 dentry
->d_name
.hash
= name
->hash
;
964 memcpy(dname
, name
->name
, name
->len
);
965 dname
[name
->len
] = 0;
967 atomic_set(&dentry
->d_count
, 1);
968 dentry
->d_flags
= DCACHE_UNHASHED
;
969 spin_lock_init(&dentry
->d_lock
);
970 dentry
->d_inode
= NULL
;
971 dentry
->d_parent
= NULL
;
974 dentry
->d_fsdata
= NULL
;
975 dentry
->d_mounted
= 0;
976 INIT_HLIST_NODE(&dentry
->d_hash
);
977 INIT_LIST_HEAD(&dentry
->d_lru
);
978 INIT_LIST_HEAD(&dentry
->d_subdirs
);
979 INIT_LIST_HEAD(&dentry
->d_alias
);
982 dentry
->d_parent
= dget(parent
);
983 dentry
->d_sb
= parent
->d_sb
;
985 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
988 spin_lock(&dcache_lock
);
990 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
991 spin_unlock(&dcache_lock
);
993 this_cpu_inc(nr_dentry
);
997 EXPORT_SYMBOL(d_alloc
);
999 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1004 q
.len
= strlen(name
);
1005 q
.hash
= full_name_hash(q
.name
, q
.len
);
1006 return d_alloc(parent
, &q
);
1008 EXPORT_SYMBOL(d_alloc_name
);
1010 /* the caller must hold dcache_lock */
1011 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1014 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1015 dentry
->d_inode
= inode
;
1016 fsnotify_d_instantiate(dentry
, inode
);
1020 * d_instantiate - fill in inode information for a dentry
1021 * @entry: dentry to complete
1022 * @inode: inode to attach to this dentry
1024 * Fill in inode information in the entry.
1026 * This turns negative dentries into productive full members
1029 * NOTE! This assumes that the inode count has been incremented
1030 * (or otherwise set) by the caller to indicate that it is now
1031 * in use by the dcache.
1034 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1036 BUG_ON(!list_empty(&entry
->d_alias
));
1037 spin_lock(&dcache_lock
);
1038 __d_instantiate(entry
, inode
);
1039 spin_unlock(&dcache_lock
);
1040 security_d_instantiate(entry
, inode
);
1042 EXPORT_SYMBOL(d_instantiate
);
1045 * d_instantiate_unique - instantiate a non-aliased dentry
1046 * @entry: dentry to instantiate
1047 * @inode: inode to attach to this dentry
1049 * Fill in inode information in the entry. On success, it returns NULL.
1050 * If an unhashed alias of "entry" already exists, then we return the
1051 * aliased dentry instead and drop one reference to inode.
1053 * Note that in order to avoid conflicts with rename() etc, the caller
1054 * had better be holding the parent directory semaphore.
1056 * This also assumes that the inode count has been incremented
1057 * (or otherwise set) by the caller to indicate that it is now
1058 * in use by the dcache.
1060 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1061 struct inode
*inode
)
1063 struct dentry
*alias
;
1064 int len
= entry
->d_name
.len
;
1065 const char *name
= entry
->d_name
.name
;
1066 unsigned int hash
= entry
->d_name
.hash
;
1069 __d_instantiate(entry
, NULL
);
1073 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1074 struct qstr
*qstr
= &alias
->d_name
;
1076 if (qstr
->hash
!= hash
)
1078 if (alias
->d_parent
!= entry
->d_parent
)
1080 if (qstr
->len
!= len
)
1082 if (memcmp(qstr
->name
, name
, len
))
1088 __d_instantiate(entry
, inode
);
1092 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1094 struct dentry
*result
;
1096 BUG_ON(!list_empty(&entry
->d_alias
));
1098 spin_lock(&dcache_lock
);
1099 result
= __d_instantiate_unique(entry
, inode
);
1100 spin_unlock(&dcache_lock
);
1103 security_d_instantiate(entry
, inode
);
1107 BUG_ON(!d_unhashed(result
));
1112 EXPORT_SYMBOL(d_instantiate_unique
);
1115 * d_alloc_root - allocate root dentry
1116 * @root_inode: inode to allocate the root for
1118 * Allocate a root ("/") dentry for the inode given. The inode is
1119 * instantiated and returned. %NULL is returned if there is insufficient
1120 * memory or the inode passed is %NULL.
1123 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1125 struct dentry
*res
= NULL
;
1128 static const struct qstr name
= { .name
= "/", .len
= 1 };
1130 res
= d_alloc(NULL
, &name
);
1132 res
->d_sb
= root_inode
->i_sb
;
1133 res
->d_parent
= res
;
1134 d_instantiate(res
, root_inode
);
1139 EXPORT_SYMBOL(d_alloc_root
);
1141 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1144 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1145 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1146 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1150 * d_obtain_alias - find or allocate a dentry for a given inode
1151 * @inode: inode to allocate the dentry for
1153 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1154 * similar open by handle operations. The returned dentry may be anonymous,
1155 * or may have a full name (if the inode was already in the cache).
1157 * When called on a directory inode, we must ensure that the inode only ever
1158 * has one dentry. If a dentry is found, that is returned instead of
1159 * allocating a new one.
1161 * On successful return, the reference to the inode has been transferred
1162 * to the dentry. In case of an error the reference on the inode is released.
1163 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1164 * be passed in and will be the error will be propagate to the return value,
1165 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1167 struct dentry
*d_obtain_alias(struct inode
*inode
)
1169 static const struct qstr anonstring
= { .name
= "" };
1174 return ERR_PTR(-ESTALE
);
1176 return ERR_CAST(inode
);
1178 res
= d_find_alias(inode
);
1182 tmp
= d_alloc(NULL
, &anonstring
);
1184 res
= ERR_PTR(-ENOMEM
);
1187 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1189 spin_lock(&dcache_lock
);
1190 res
= __d_find_alias(inode
, 0);
1192 spin_unlock(&dcache_lock
);
1197 /* attach a disconnected dentry */
1198 spin_lock(&tmp
->d_lock
);
1199 tmp
->d_sb
= inode
->i_sb
;
1200 tmp
->d_inode
= inode
;
1201 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1202 tmp
->d_flags
&= ~DCACHE_UNHASHED
;
1203 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1204 hlist_add_head(&tmp
->d_hash
, &inode
->i_sb
->s_anon
);
1205 spin_unlock(&tmp
->d_lock
);
1207 spin_unlock(&dcache_lock
);
1214 EXPORT_SYMBOL(d_obtain_alias
);
1217 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1218 * @inode: the inode which may have a disconnected dentry
1219 * @dentry: a negative dentry which we want to point to the inode.
1221 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1222 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1223 * and return it, else simply d_add the inode to the dentry and return NULL.
1225 * This is needed in the lookup routine of any filesystem that is exportable
1226 * (via knfsd) so that we can build dcache paths to directories effectively.
1228 * If a dentry was found and moved, then it is returned. Otherwise NULL
1229 * is returned. This matches the expected return value of ->lookup.
1232 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1234 struct dentry
*new = NULL
;
1236 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1237 spin_lock(&dcache_lock
);
1238 new = __d_find_alias(inode
, 1);
1240 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1241 spin_unlock(&dcache_lock
);
1242 security_d_instantiate(new, inode
);
1243 d_move(new, dentry
);
1246 /* already taking dcache_lock, so d_add() by hand */
1247 __d_instantiate(dentry
, inode
);
1248 spin_unlock(&dcache_lock
);
1249 security_d_instantiate(dentry
, inode
);
1253 d_add(dentry
, inode
);
1256 EXPORT_SYMBOL(d_splice_alias
);
1259 * d_add_ci - lookup or allocate new dentry with case-exact name
1260 * @inode: the inode case-insensitive lookup has found
1261 * @dentry: the negative dentry that was passed to the parent's lookup func
1262 * @name: the case-exact name to be associated with the returned dentry
1264 * This is to avoid filling the dcache with case-insensitive names to the
1265 * same inode, only the actual correct case is stored in the dcache for
1266 * case-insensitive filesystems.
1268 * For a case-insensitive lookup match and if the the case-exact dentry
1269 * already exists in in the dcache, use it and return it.
1271 * If no entry exists with the exact case name, allocate new dentry with
1272 * the exact case, and return the spliced entry.
1274 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1278 struct dentry
*found
;
1282 * First check if a dentry matching the name already exists,
1283 * if not go ahead and create it now.
1285 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1287 new = d_alloc(dentry
->d_parent
, name
);
1293 found
= d_splice_alias(inode
, new);
1302 * If a matching dentry exists, and it's not negative use it.
1304 * Decrement the reference count to balance the iget() done
1307 if (found
->d_inode
) {
1308 if (unlikely(found
->d_inode
!= inode
)) {
1309 /* This can't happen because bad inodes are unhashed. */
1310 BUG_ON(!is_bad_inode(inode
));
1311 BUG_ON(!is_bad_inode(found
->d_inode
));
1318 * Negative dentry: instantiate it unless the inode is a directory and
1319 * already has a dentry.
1321 spin_lock(&dcache_lock
);
1322 if (!S_ISDIR(inode
->i_mode
) || list_empty(&inode
->i_dentry
)) {
1323 __d_instantiate(found
, inode
);
1324 spin_unlock(&dcache_lock
);
1325 security_d_instantiate(found
, inode
);
1330 * In case a directory already has a (disconnected) entry grab a
1331 * reference to it, move it in place and use it.
1333 new = list_entry(inode
->i_dentry
.next
, struct dentry
, d_alias
);
1335 spin_unlock(&dcache_lock
);
1336 security_d_instantiate(found
, inode
);
1344 return ERR_PTR(error
);
1346 EXPORT_SYMBOL(d_add_ci
);
1349 * d_lookup - search for a dentry
1350 * @parent: parent dentry
1351 * @name: qstr of name we wish to find
1352 * Returns: dentry, or NULL
1354 * d_lookup searches the children of the parent dentry for the name in
1355 * question. If the dentry is found its reference count is incremented and the
1356 * dentry is returned. The caller must use dput to free the entry when it has
1357 * finished using it. %NULL is returned if the dentry does not exist.
1359 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1361 struct dentry
* dentry
= NULL
;
1365 seq
= read_seqbegin(&rename_lock
);
1366 dentry
= __d_lookup(parent
, name
);
1369 } while (read_seqretry(&rename_lock
, seq
));
1372 EXPORT_SYMBOL(d_lookup
);
1375 * __d_lookup - search for a dentry (racy)
1376 * @parent: parent dentry
1377 * @name: qstr of name we wish to find
1378 * Returns: dentry, or NULL
1380 * __d_lookup is like d_lookup, however it may (rarely) return a
1381 * false-negative result due to unrelated rename activity.
1383 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1384 * however it must be used carefully, eg. with a following d_lookup in
1385 * the case of failure.
1387 * __d_lookup callers must be commented.
1389 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1391 unsigned int len
= name
->len
;
1392 unsigned int hash
= name
->hash
;
1393 const unsigned char *str
= name
->name
;
1394 struct hlist_head
*head
= d_hash(parent
,hash
);
1395 struct dentry
*found
= NULL
;
1396 struct hlist_node
*node
;
1397 struct dentry
*dentry
;
1400 * The hash list is protected using RCU.
1402 * Take d_lock when comparing a candidate dentry, to avoid races
1405 * It is possible that concurrent renames can mess up our list
1406 * walk here and result in missing our dentry, resulting in the
1407 * false-negative result. d_lookup() protects against concurrent
1408 * renames using rename_lock seqlock.
1410 * See Documentation/vfs/dcache-locking.txt for more details.
1414 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1417 if (dentry
->d_name
.hash
!= hash
)
1419 if (dentry
->d_parent
!= parent
)
1422 spin_lock(&dentry
->d_lock
);
1425 * Recheck the dentry after taking the lock - d_move may have
1426 * changed things. Don't bother checking the hash because
1427 * we're about to compare the whole name anyway.
1429 if (dentry
->d_parent
!= parent
)
1432 /* non-existing due to RCU? */
1433 if (d_unhashed(dentry
))
1437 * It is safe to compare names since d_move() cannot
1438 * change the qstr (protected by d_lock).
1440 qstr
= &dentry
->d_name
;
1441 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1442 if (parent
->d_op
->d_compare(parent
, qstr
, name
))
1445 if (qstr
->len
!= len
)
1447 if (memcmp(qstr
->name
, str
, len
))
1451 atomic_inc(&dentry
->d_count
);
1453 spin_unlock(&dentry
->d_lock
);
1456 spin_unlock(&dentry
->d_lock
);
1464 * d_hash_and_lookup - hash the qstr then search for a dentry
1465 * @dir: Directory to search in
1466 * @name: qstr of name we wish to find
1468 * On hash failure or on lookup failure NULL is returned.
1470 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1472 struct dentry
*dentry
= NULL
;
1475 * Check for a fs-specific hash function. Note that we must
1476 * calculate the standard hash first, as the d_op->d_hash()
1477 * routine may choose to leave the hash value unchanged.
1479 name
->hash
= full_name_hash(name
->name
, name
->len
);
1480 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1481 if (dir
->d_op
->d_hash(dir
, name
) < 0)
1484 dentry
= d_lookup(dir
, name
);
1490 * d_validate - verify dentry provided from insecure source (deprecated)
1491 * @dentry: The dentry alleged to be valid child of @dparent
1492 * @dparent: The parent dentry (known to be valid)
1494 * An insecure source has sent us a dentry, here we verify it and dget() it.
1495 * This is used by ncpfs in its readdir implementation.
1496 * Zero is returned in the dentry is invalid.
1498 * This function is slow for big directories, and deprecated, do not use it.
1500 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1502 struct dentry
*child
;
1504 spin_lock(&dcache_lock
);
1505 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1506 if (dentry
== child
) {
1507 __dget_locked(dentry
);
1508 spin_unlock(&dcache_lock
);
1512 spin_unlock(&dcache_lock
);
1516 EXPORT_SYMBOL(d_validate
);
1519 * When a file is deleted, we have two options:
1520 * - turn this dentry into a negative dentry
1521 * - unhash this dentry and free it.
1523 * Usually, we want to just turn this into
1524 * a negative dentry, but if anybody else is
1525 * currently using the dentry or the inode
1526 * we can't do that and we fall back on removing
1527 * it from the hash queues and waiting for
1528 * it to be deleted later when it has no users
1532 * d_delete - delete a dentry
1533 * @dentry: The dentry to delete
1535 * Turn the dentry into a negative dentry if possible, otherwise
1536 * remove it from the hash queues so it can be deleted later
1539 void d_delete(struct dentry
* dentry
)
1543 * Are we the only user?
1545 spin_lock(&dcache_lock
);
1546 spin_lock(&dentry
->d_lock
);
1547 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1548 if (atomic_read(&dentry
->d_count
) == 1) {
1549 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1550 dentry_iput(dentry
);
1551 fsnotify_nameremove(dentry
, isdir
);
1555 if (!d_unhashed(dentry
))
1558 spin_unlock(&dentry
->d_lock
);
1559 spin_unlock(&dcache_lock
);
1561 fsnotify_nameremove(dentry
, isdir
);
1563 EXPORT_SYMBOL(d_delete
);
1565 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1568 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1569 hlist_add_head_rcu(&entry
->d_hash
, list
);
1572 static void _d_rehash(struct dentry
* entry
)
1574 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1578 * d_rehash - add an entry back to the hash
1579 * @entry: dentry to add to the hash
1581 * Adds a dentry to the hash according to its name.
1584 void d_rehash(struct dentry
* entry
)
1586 spin_lock(&dcache_lock
);
1587 spin_lock(&entry
->d_lock
);
1589 spin_unlock(&entry
->d_lock
);
1590 spin_unlock(&dcache_lock
);
1592 EXPORT_SYMBOL(d_rehash
);
1595 * When switching names, the actual string doesn't strictly have to
1596 * be preserved in the target - because we're dropping the target
1597 * anyway. As such, we can just do a simple memcpy() to copy over
1598 * the new name before we switch.
1600 * Note that we have to be a lot more careful about getting the hash
1601 * switched - we have to switch the hash value properly even if it
1602 * then no longer matches the actual (corrupted) string of the target.
1603 * The hash value has to match the hash queue that the dentry is on..
1605 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1607 if (dname_external(target
)) {
1608 if (dname_external(dentry
)) {
1610 * Both external: swap the pointers
1612 swap(target
->d_name
.name
, dentry
->d_name
.name
);
1615 * dentry:internal, target:external. Steal target's
1616 * storage and make target internal.
1618 memcpy(target
->d_iname
, dentry
->d_name
.name
,
1619 dentry
->d_name
.len
+ 1);
1620 dentry
->d_name
.name
= target
->d_name
.name
;
1621 target
->d_name
.name
= target
->d_iname
;
1624 if (dname_external(dentry
)) {
1626 * dentry:external, target:internal. Give dentry's
1627 * storage to target and make dentry internal
1629 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1630 target
->d_name
.len
+ 1);
1631 target
->d_name
.name
= dentry
->d_name
.name
;
1632 dentry
->d_name
.name
= dentry
->d_iname
;
1635 * Both are internal. Just copy target to dentry
1637 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1638 target
->d_name
.len
+ 1);
1639 dentry
->d_name
.len
= target
->d_name
.len
;
1643 swap(dentry
->d_name
.len
, target
->d_name
.len
);
1647 * We cannibalize "target" when moving dentry on top of it,
1648 * because it's going to be thrown away anyway. We could be more
1649 * polite about it, though.
1651 * This forceful removal will result in ugly /proc output if
1652 * somebody holds a file open that got deleted due to a rename.
1653 * We could be nicer about the deleted file, and let it show
1654 * up under the name it had before it was deleted rather than
1655 * under the original name of the file that was moved on top of it.
1659 * d_move_locked - move a dentry
1660 * @dentry: entry to move
1661 * @target: new dentry
1663 * Update the dcache to reflect the move of a file name. Negative
1664 * dcache entries should not be moved in this way.
1666 static void d_move_locked(struct dentry
* dentry
, struct dentry
* target
)
1668 struct hlist_head
*list
;
1670 if (!dentry
->d_inode
)
1671 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1673 write_seqlock(&rename_lock
);
1675 * XXXX: do we really need to take target->d_lock?
1677 if (target
< dentry
) {
1678 spin_lock(&target
->d_lock
);
1679 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1681 spin_lock(&dentry
->d_lock
);
1682 spin_lock_nested(&target
->d_lock
, DENTRY_D_LOCK_NESTED
);
1685 /* Move the dentry to the target hash queue, if on different bucket */
1686 if (d_unhashed(dentry
))
1687 goto already_unhashed
;
1689 hlist_del_rcu(&dentry
->d_hash
);
1692 list
= d_hash(target
->d_parent
, target
->d_name
.hash
);
1693 __d_rehash(dentry
, list
);
1695 /* Unhash the target: dput() will then get rid of it */
1698 list_del(&dentry
->d_u
.d_child
);
1699 list_del(&target
->d_u
.d_child
);
1701 /* Switch the names.. */
1702 switch_names(dentry
, target
);
1703 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
1705 /* ... and switch the parents */
1706 if (IS_ROOT(dentry
)) {
1707 dentry
->d_parent
= target
->d_parent
;
1708 target
->d_parent
= target
;
1709 INIT_LIST_HEAD(&target
->d_u
.d_child
);
1711 swap(dentry
->d_parent
, target
->d_parent
);
1713 /* And add them back to the (new) parent lists */
1714 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
1717 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1718 spin_unlock(&target
->d_lock
);
1719 fsnotify_d_move(dentry
);
1720 spin_unlock(&dentry
->d_lock
);
1721 write_sequnlock(&rename_lock
);
1725 * d_move - move a dentry
1726 * @dentry: entry to move
1727 * @target: new dentry
1729 * Update the dcache to reflect the move of a file name. Negative
1730 * dcache entries should not be moved in this way.
1733 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1735 spin_lock(&dcache_lock
);
1736 d_move_locked(dentry
, target
);
1737 spin_unlock(&dcache_lock
);
1739 EXPORT_SYMBOL(d_move
);
1742 * d_ancestor - search for an ancestor
1743 * @p1: ancestor dentry
1746 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1747 * an ancestor of p2, else NULL.
1749 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
1753 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
1754 if (p
->d_parent
== p1
)
1761 * This helper attempts to cope with remotely renamed directories
1763 * It assumes that the caller is already holding
1764 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1766 * Note: If ever the locking in lock_rename() changes, then please
1767 * remember to update this too...
1769 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
1770 __releases(dcache_lock
)
1772 struct mutex
*m1
= NULL
, *m2
= NULL
;
1775 /* If alias and dentry share a parent, then no extra locks required */
1776 if (alias
->d_parent
== dentry
->d_parent
)
1779 /* Check for loops */
1780 ret
= ERR_PTR(-ELOOP
);
1781 if (d_ancestor(alias
, dentry
))
1784 /* See lock_rename() */
1785 ret
= ERR_PTR(-EBUSY
);
1786 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
1788 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
1789 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
1791 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
1793 d_move_locked(alias
, dentry
);
1796 spin_unlock(&dcache_lock
);
1805 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1806 * named dentry in place of the dentry to be replaced.
1808 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
1810 struct dentry
*dparent
, *aparent
;
1812 switch_names(dentry
, anon
);
1813 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
1815 dparent
= dentry
->d_parent
;
1816 aparent
= anon
->d_parent
;
1818 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
1819 list_del(&dentry
->d_u
.d_child
);
1820 if (!IS_ROOT(dentry
))
1821 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
1823 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1825 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
1826 list_del(&anon
->d_u
.d_child
);
1828 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
1830 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
1832 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
1836 * d_materialise_unique - introduce an inode into the tree
1837 * @dentry: candidate dentry
1838 * @inode: inode to bind to the dentry, to which aliases may be attached
1840 * Introduces an dentry into the tree, substituting an extant disconnected
1841 * root directory alias in its place if there is one
1843 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
1845 struct dentry
*actual
;
1847 BUG_ON(!d_unhashed(dentry
));
1849 spin_lock(&dcache_lock
);
1853 __d_instantiate(dentry
, NULL
);
1857 if (S_ISDIR(inode
->i_mode
)) {
1858 struct dentry
*alias
;
1860 /* Does an aliased dentry already exist? */
1861 alias
= __d_find_alias(inode
, 0);
1864 /* Is this an anonymous mountpoint that we could splice
1866 if (IS_ROOT(alias
)) {
1867 spin_lock(&alias
->d_lock
);
1868 __d_materialise_dentry(dentry
, alias
);
1872 /* Nope, but we must(!) avoid directory aliasing */
1873 actual
= __d_unalias(dentry
, alias
);
1880 /* Add a unique reference */
1881 actual
= __d_instantiate_unique(dentry
, inode
);
1884 else if (unlikely(!d_unhashed(actual
)))
1885 goto shouldnt_be_hashed
;
1888 spin_lock(&actual
->d_lock
);
1891 spin_unlock(&actual
->d_lock
);
1892 spin_unlock(&dcache_lock
);
1894 if (actual
== dentry
) {
1895 security_d_instantiate(dentry
, inode
);
1903 spin_unlock(&dcache_lock
);
1906 EXPORT_SYMBOL_GPL(d_materialise_unique
);
1908 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
1912 return -ENAMETOOLONG
;
1914 memcpy(*buffer
, str
, namelen
);
1918 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
1920 return prepend(buffer
, buflen
, name
->name
, name
->len
);
1924 * Prepend path string to a buffer
1926 * @path: the dentry/vfsmount to report
1927 * @root: root vfsmnt/dentry (may be modified by this function)
1928 * @buffer: pointer to the end of the buffer
1929 * @buflen: pointer to buffer length
1931 * Caller holds the dcache_lock.
1933 * If path is not reachable from the supplied root, then the value of
1934 * root is changed (without modifying refcounts).
1936 static int prepend_path(const struct path
*path
, struct path
*root
,
1937 char **buffer
, int *buflen
)
1939 struct dentry
*dentry
= path
->dentry
;
1940 struct vfsmount
*vfsmnt
= path
->mnt
;
1944 br_read_lock(vfsmount_lock
);
1945 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
1946 struct dentry
* parent
;
1948 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
1950 if (vfsmnt
->mnt_parent
== vfsmnt
) {
1953 dentry
= vfsmnt
->mnt_mountpoint
;
1954 vfsmnt
= vfsmnt
->mnt_parent
;
1957 parent
= dentry
->d_parent
;
1959 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
1961 error
= prepend(buffer
, buflen
, "/", 1);
1970 if (!error
&& !slash
)
1971 error
= prepend(buffer
, buflen
, "/", 1);
1973 br_read_unlock(vfsmount_lock
);
1978 * Filesystems needing to implement special "root names"
1979 * should do so with ->d_dname()
1981 if (IS_ROOT(dentry
) &&
1982 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
1983 WARN(1, "Root dentry has weird name <%.*s>\n",
1984 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
1987 root
->dentry
= dentry
;
1992 * __d_path - return the path of a dentry
1993 * @path: the dentry/vfsmount to report
1994 * @root: root vfsmnt/dentry (may be modified by this function)
1995 * @buf: buffer to return value in
1996 * @buflen: buffer length
1998 * Convert a dentry into an ASCII path name.
2000 * Returns a pointer into the buffer or an error code if the
2001 * path was too long.
2003 * "buflen" should be positive.
2005 * If path is not reachable from the supplied root, then the value of
2006 * root is changed (without modifying refcounts).
2008 char *__d_path(const struct path
*path
, struct path
*root
,
2009 char *buf
, int buflen
)
2011 char *res
= buf
+ buflen
;
2014 prepend(&res
, &buflen
, "\0", 1);
2015 spin_lock(&dcache_lock
);
2016 error
= prepend_path(path
, root
, &res
, &buflen
);
2017 spin_unlock(&dcache_lock
);
2020 return ERR_PTR(error
);
2025 * same as __d_path but appends "(deleted)" for unlinked files.
2027 static int path_with_deleted(const struct path
*path
, struct path
*root
,
2028 char **buf
, int *buflen
)
2030 prepend(buf
, buflen
, "\0", 1);
2031 if (d_unlinked(path
->dentry
)) {
2032 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2037 return prepend_path(path
, root
, buf
, buflen
);
2040 static int prepend_unreachable(char **buffer
, int *buflen
)
2042 return prepend(buffer
, buflen
, "(unreachable)", 13);
2046 * d_path - return the path of a dentry
2047 * @path: path to report
2048 * @buf: buffer to return value in
2049 * @buflen: buffer length
2051 * Convert a dentry into an ASCII path name. If the entry has been deleted
2052 * the string " (deleted)" is appended. Note that this is ambiguous.
2054 * Returns a pointer into the buffer or an error code if the path was
2055 * too long. Note: Callers should use the returned pointer, not the passed
2056 * in buffer, to use the name! The implementation often starts at an offset
2057 * into the buffer, and may leave 0 bytes at the start.
2059 * "buflen" should be positive.
2061 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2063 char *res
= buf
+ buflen
;
2069 * We have various synthetic filesystems that never get mounted. On
2070 * these filesystems dentries are never used for lookup purposes, and
2071 * thus don't need to be hashed. They also don't need a name until a
2072 * user wants to identify the object in /proc/pid/fd/. The little hack
2073 * below allows us to generate a name for these objects on demand:
2075 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2076 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2078 get_fs_root(current
->fs
, &root
);
2079 spin_lock(&dcache_lock
);
2081 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2083 res
= ERR_PTR(error
);
2084 spin_unlock(&dcache_lock
);
2088 EXPORT_SYMBOL(d_path
);
2091 * d_path_with_unreachable - return the path of a dentry
2092 * @path: path to report
2093 * @buf: buffer to return value in
2094 * @buflen: buffer length
2096 * The difference from d_path() is that this prepends "(unreachable)"
2097 * to paths which are unreachable from the current process' root.
2099 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2101 char *res
= buf
+ buflen
;
2106 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2107 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2109 get_fs_root(current
->fs
, &root
);
2110 spin_lock(&dcache_lock
);
2112 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2113 if (!error
&& !path_equal(&tmp
, &root
))
2114 error
= prepend_unreachable(&res
, &buflen
);
2115 spin_unlock(&dcache_lock
);
2118 res
= ERR_PTR(error
);
2124 * Helper function for dentry_operations.d_dname() members
2126 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2127 const char *fmt
, ...)
2133 va_start(args
, fmt
);
2134 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2137 if (sz
> sizeof(temp
) || sz
> buflen
)
2138 return ERR_PTR(-ENAMETOOLONG
);
2140 buffer
+= buflen
- sz
;
2141 return memcpy(buffer
, temp
, sz
);
2145 * Write full pathname from the root of the filesystem into the buffer.
2147 char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2149 char *end
= buf
+ buflen
;
2152 prepend(&end
, &buflen
, "\0", 1);
2159 while (!IS_ROOT(dentry
)) {
2160 struct dentry
*parent
= dentry
->d_parent
;
2163 if ((prepend_name(&end
, &buflen
, &dentry
->d_name
) != 0) ||
2164 (prepend(&end
, &buflen
, "/", 1) != 0))
2172 return ERR_PTR(-ENAMETOOLONG
);
2174 EXPORT_SYMBOL(__dentry_path
);
2176 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2181 spin_lock(&dcache_lock
);
2182 if (d_unlinked(dentry
)) {
2184 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2188 retval
= __dentry_path(dentry
, buf
, buflen
);
2189 spin_unlock(&dcache_lock
);
2190 if (!IS_ERR(retval
) && p
)
2191 *p
= '/'; /* restore '/' overriden with '\0' */
2194 spin_unlock(&dcache_lock
);
2195 return ERR_PTR(-ENAMETOOLONG
);
2199 * NOTE! The user-level library version returns a
2200 * character pointer. The kernel system call just
2201 * returns the length of the buffer filled (which
2202 * includes the ending '\0' character), or a negative
2203 * error value. So libc would do something like
2205 * char *getcwd(char * buf, size_t size)
2209 * retval = sys_getcwd(buf, size);
2216 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2219 struct path pwd
, root
;
2220 char *page
= (char *) __get_free_page(GFP_USER
);
2225 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2228 spin_lock(&dcache_lock
);
2229 if (!d_unlinked(pwd
.dentry
)) {
2231 struct path tmp
= root
;
2232 char *cwd
= page
+ PAGE_SIZE
;
2233 int buflen
= PAGE_SIZE
;
2235 prepend(&cwd
, &buflen
, "\0", 1);
2236 error
= prepend_path(&pwd
, &tmp
, &cwd
, &buflen
);
2237 spin_unlock(&dcache_lock
);
2242 /* Unreachable from current root */
2243 if (!path_equal(&tmp
, &root
)) {
2244 error
= prepend_unreachable(&cwd
, &buflen
);
2250 len
= PAGE_SIZE
+ page
- cwd
;
2253 if (copy_to_user(buf
, cwd
, len
))
2257 spin_unlock(&dcache_lock
);
2262 free_page((unsigned long) page
);
2267 * Test whether new_dentry is a subdirectory of old_dentry.
2269 * Trivially implemented using the dcache structure
2273 * is_subdir - is new dentry a subdirectory of old_dentry
2274 * @new_dentry: new dentry
2275 * @old_dentry: old dentry
2277 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2278 * Returns 0 otherwise.
2279 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2282 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2287 if (new_dentry
== old_dentry
)
2291 * Need rcu_readlock to protect against the d_parent trashing
2296 /* for restarting inner loop in case of seq retry */
2297 seq
= read_seqbegin(&rename_lock
);
2298 if (d_ancestor(old_dentry
, new_dentry
))
2302 } while (read_seqretry(&rename_lock
, seq
));
2308 int path_is_under(struct path
*path1
, struct path
*path2
)
2310 struct vfsmount
*mnt
= path1
->mnt
;
2311 struct dentry
*dentry
= path1
->dentry
;
2314 br_read_lock(vfsmount_lock
);
2315 if (mnt
!= path2
->mnt
) {
2317 if (mnt
->mnt_parent
== mnt
) {
2318 br_read_unlock(vfsmount_lock
);
2321 if (mnt
->mnt_parent
== path2
->mnt
)
2323 mnt
= mnt
->mnt_parent
;
2325 dentry
= mnt
->mnt_mountpoint
;
2327 res
= is_subdir(dentry
, path2
->dentry
);
2328 br_read_unlock(vfsmount_lock
);
2331 EXPORT_SYMBOL(path_is_under
);
2333 void d_genocide(struct dentry
*root
)
2335 struct dentry
*this_parent
= root
;
2336 struct list_head
*next
;
2338 spin_lock(&dcache_lock
);
2340 next
= this_parent
->d_subdirs
.next
;
2342 while (next
!= &this_parent
->d_subdirs
) {
2343 struct list_head
*tmp
= next
;
2344 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2346 if (d_unhashed(dentry
)||!dentry
->d_inode
)
2348 if (!list_empty(&dentry
->d_subdirs
)) {
2349 this_parent
= dentry
;
2352 atomic_dec(&dentry
->d_count
);
2354 if (this_parent
!= root
) {
2355 next
= this_parent
->d_u
.d_child
.next
;
2356 atomic_dec(&this_parent
->d_count
);
2357 this_parent
= this_parent
->d_parent
;
2360 spin_unlock(&dcache_lock
);
2364 * find_inode_number - check for dentry with name
2365 * @dir: directory to check
2366 * @name: Name to find.
2368 * Check whether a dentry already exists for the given name,
2369 * and return the inode number if it has an inode. Otherwise
2372 * This routine is used to post-process directory listings for
2373 * filesystems using synthetic inode numbers, and is necessary
2374 * to keep getcwd() working.
2377 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2379 struct dentry
* dentry
;
2382 dentry
= d_hash_and_lookup(dir
, name
);
2384 if (dentry
->d_inode
)
2385 ino
= dentry
->d_inode
->i_ino
;
2390 EXPORT_SYMBOL(find_inode_number
);
2392 static __initdata
unsigned long dhash_entries
;
2393 static int __init
set_dhash_entries(char *str
)
2397 dhash_entries
= simple_strtoul(str
, &str
, 0);
2400 __setup("dhash_entries=", set_dhash_entries
);
2402 static void __init
dcache_init_early(void)
2406 /* If hashes are distributed across NUMA nodes, defer
2407 * hash allocation until vmalloc space is available.
2413 alloc_large_system_hash("Dentry cache",
2414 sizeof(struct hlist_head
),
2422 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2423 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2426 static void __init
dcache_init(void)
2431 * A constructor could be added for stable state like the lists,
2432 * but it is probably not worth it because of the cache nature
2435 dentry_cache
= KMEM_CACHE(dentry
,
2436 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2438 register_shrinker(&dcache_shrinker
);
2440 /* Hash may have been set up in dcache_init_early */
2445 alloc_large_system_hash("Dentry cache",
2446 sizeof(struct hlist_head
),
2454 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2455 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2458 /* SLAB cache for __getname() consumers */
2459 struct kmem_cache
*names_cachep __read_mostly
;
2460 EXPORT_SYMBOL(names_cachep
);
2462 EXPORT_SYMBOL(d_genocide
);
2464 void __init
vfs_caches_init_early(void)
2466 dcache_init_early();
2470 void __init
vfs_caches_init(unsigned long mempages
)
2472 unsigned long reserve
;
2474 /* Base hash sizes on available memory, with a reserve equal to
2475 150% of current kernel size */
2477 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2478 mempages
-= reserve
;
2480 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2481 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2485 files_init(mempages
);