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>
40 * dcache_inode_lock protects:
41 * - i_dentry, d_alias, d_inode
42 * dcache_hash_lock protects:
43 * - the dcache hash table, s_anon lists
44 * dcache_lru_lock protects:
45 * - the dcache lru lists and counters
52 * - d_parent and d_subdirs
53 * - childrens' d_child and d_parent
62 * If there is an ancestor relationship:
63 * dentry->d_parent->...->d_parent->d_lock
65 * dentry->d_parent->d_lock
68 * If no ancestor relationship:
69 * if (dentry1 < dentry2)
73 int sysctl_vfs_cache_pressure __read_mostly
= 100;
74 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
76 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_inode_lock
);
77 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_hash_lock
);
78 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
79 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
81 EXPORT_SYMBOL(rename_lock
);
82 EXPORT_SYMBOL(dcache_inode_lock
);
84 static struct kmem_cache
*dentry_cache __read_mostly
;
86 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
89 * This is the single most critical data structure when it comes
90 * to the dcache: the hashtable for lookups. Somebody should try
91 * to make this good - I've just made it work.
93 * This hash-function tries to avoid losing too many bits of hash
94 * information, yet avoid using a prime hash-size or similar.
96 #define D_HASHBITS d_hash_shift
97 #define D_HASHMASK d_hash_mask
99 static unsigned int d_hash_mask __read_mostly
;
100 static unsigned int d_hash_shift __read_mostly
;
101 static struct hlist_head
*dentry_hashtable __read_mostly
;
103 /* Statistics gathering. */
104 struct dentry_stat_t dentry_stat
= {
108 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
110 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
111 static int get_nr_dentry(void)
115 for_each_possible_cpu(i
)
116 sum
+= per_cpu(nr_dentry
, i
);
117 return sum
< 0 ? 0 : sum
;
120 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
121 size_t *lenp
, loff_t
*ppos
)
123 dentry_stat
.nr_dentry
= get_nr_dentry();
124 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
128 static void __d_free(struct rcu_head
*head
)
130 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
132 WARN_ON(!list_empty(&dentry
->d_alias
));
133 if (dname_external(dentry
))
134 kfree(dentry
->d_name
.name
);
135 kmem_cache_free(dentry_cache
, dentry
);
141 static void d_free(struct dentry
*dentry
)
143 BUG_ON(dentry
->d_count
);
144 this_cpu_dec(nr_dentry
);
145 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
146 dentry
->d_op
->d_release(dentry
);
148 /* if dentry was never inserted into hash, immediate free is OK */
149 if (hlist_unhashed(&dentry
->d_hash
))
150 __d_free(&dentry
->d_u
.d_rcu
);
152 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
156 * Release the dentry's inode, using the filesystem
157 * d_iput() operation if defined.
159 static void dentry_iput(struct dentry
* dentry
)
160 __releases(dentry
->d_lock
)
161 __releases(dcache_inode_lock
)
163 struct inode
*inode
= dentry
->d_inode
;
165 dentry
->d_inode
= NULL
;
166 list_del_init(&dentry
->d_alias
);
167 spin_unlock(&dentry
->d_lock
);
168 spin_unlock(&dcache_inode_lock
);
170 fsnotify_inoderemove(inode
);
171 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
172 dentry
->d_op
->d_iput(dentry
, inode
);
176 spin_unlock(&dentry
->d_lock
);
177 spin_unlock(&dcache_inode_lock
);
182 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
184 static void dentry_lru_add(struct dentry
*dentry
)
186 if (list_empty(&dentry
->d_lru
)) {
187 spin_lock(&dcache_lru_lock
);
188 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
189 dentry
->d_sb
->s_nr_dentry_unused
++;
190 dentry_stat
.nr_unused
++;
191 spin_unlock(&dcache_lru_lock
);
195 static void __dentry_lru_del(struct dentry
*dentry
)
197 list_del_init(&dentry
->d_lru
);
198 dentry
->d_sb
->s_nr_dentry_unused
--;
199 dentry_stat
.nr_unused
--;
202 static void dentry_lru_del(struct dentry
*dentry
)
204 if (!list_empty(&dentry
->d_lru
)) {
205 spin_lock(&dcache_lru_lock
);
206 __dentry_lru_del(dentry
);
207 spin_unlock(&dcache_lru_lock
);
211 static void dentry_lru_move_tail(struct dentry
*dentry
)
213 spin_lock(&dcache_lru_lock
);
214 if (list_empty(&dentry
->d_lru
)) {
215 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
216 dentry
->d_sb
->s_nr_dentry_unused
++;
217 dentry_stat
.nr_unused
++;
219 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
221 spin_unlock(&dcache_lru_lock
);
225 * d_kill - kill dentry and return parent
226 * @dentry: dentry to kill
228 * The dentry must already be unhashed and removed from the LRU.
230 * If this is the root of the dentry tree, return NULL.
232 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
235 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
236 __releases(dentry
->d_lock
)
237 __releases(parent
->d_lock
)
238 __releases(dcache_inode_lock
)
240 dentry
->d_parent
= NULL
;
241 list_del(&dentry
->d_u
.d_child
);
243 spin_unlock(&parent
->d_lock
);
246 * dentry_iput drops the locks, at which point nobody (except
247 * transient RCU lookups) can reach this dentry.
254 * d_drop - drop a dentry
255 * @dentry: dentry to drop
257 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
258 * be found through a VFS lookup any more. Note that this is different from
259 * deleting the dentry - d_delete will try to mark the dentry negative if
260 * possible, giving a successful _negative_ lookup, while d_drop will
261 * just make the cache lookup fail.
263 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
264 * reason (NFS timeouts or autofs deletes).
266 * __d_drop requires dentry->d_lock.
268 void __d_drop(struct dentry
*dentry
)
270 if (!(dentry
->d_flags
& DCACHE_UNHASHED
)) {
271 dentry
->d_flags
|= DCACHE_UNHASHED
;
272 spin_lock(&dcache_hash_lock
);
273 hlist_del_rcu(&dentry
->d_hash
);
274 spin_unlock(&dcache_hash_lock
);
277 EXPORT_SYMBOL(__d_drop
);
279 void d_drop(struct dentry
*dentry
)
281 spin_lock(&dentry
->d_lock
);
283 spin_unlock(&dentry
->d_lock
);
285 EXPORT_SYMBOL(d_drop
);
290 * This is complicated by the fact that we do not want to put
291 * dentries that are no longer on any hash chain on the unused
292 * list: we'd much rather just get rid of them immediately.
294 * However, that implies that we have to traverse the dentry
295 * tree upwards to the parents which might _also_ now be
296 * scheduled for deletion (it may have been only waiting for
297 * its last child to go away).
299 * This tail recursion is done by hand as we don't want to depend
300 * on the compiler to always get this right (gcc generally doesn't).
301 * Real recursion would eat up our stack space.
305 * dput - release a dentry
306 * @dentry: dentry to release
308 * Release a dentry. This will drop the usage count and if appropriate
309 * call the dentry unlink method as well as removing it from the queues and
310 * releasing its resources. If the parent dentries were scheduled for release
311 * they too may now get deleted.
313 * no dcache lock, please.
316 void dput(struct dentry
*dentry
)
318 struct dentry
*parent
;
323 if (dentry
->d_count
== 1)
325 spin_lock(&dentry
->d_lock
);
326 BUG_ON(!dentry
->d_count
);
327 if (dentry
->d_count
> 1) {
329 spin_unlock(&dentry
->d_lock
);
333 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
334 if (dentry
->d_op
->d_delete(dentry
))
338 /* Unreachable? Get rid of it */
339 if (d_unhashed(dentry
))
342 /* Otherwise leave it cached and ensure it's on the LRU */
343 dentry
->d_flags
|= DCACHE_REFERENCED
;
344 dentry_lru_add(dentry
);
347 spin_unlock(&dentry
->d_lock
);
351 if (!spin_trylock(&dcache_inode_lock
)) {
353 spin_unlock(&dentry
->d_lock
);
360 parent
= dentry
->d_parent
;
361 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
362 spin_unlock(&dcache_inode_lock
);
366 /* if dentry was on the d_lru list delete it from there */
367 dentry_lru_del(dentry
);
368 /* if it was on the hash (d_delete case), then remove it */
370 dentry
= d_kill(dentry
, parent
);
377 * d_invalidate - invalidate a dentry
378 * @dentry: dentry to invalidate
380 * Try to invalidate the dentry if it turns out to be
381 * possible. If there are other dentries that can be
382 * reached through this one we can't delete it and we
383 * return -EBUSY. On success we return 0.
388 int d_invalidate(struct dentry
* dentry
)
391 * If it's already been dropped, return OK.
393 spin_lock(&dentry
->d_lock
);
394 if (d_unhashed(dentry
)) {
395 spin_unlock(&dentry
->d_lock
);
399 * Check whether to do a partial shrink_dcache
400 * to get rid of unused child entries.
402 if (!list_empty(&dentry
->d_subdirs
)) {
403 spin_unlock(&dentry
->d_lock
);
404 shrink_dcache_parent(dentry
);
405 spin_lock(&dentry
->d_lock
);
409 * Somebody else still using it?
411 * If it's a directory, we can't drop it
412 * for fear of somebody re-populating it
413 * with children (even though dropping it
414 * would make it unreachable from the root,
415 * we might still populate it if it was a
416 * working directory or similar).
418 if (dentry
->d_count
> 1) {
419 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
420 spin_unlock(&dentry
->d_lock
);
426 spin_unlock(&dentry
->d_lock
);
429 EXPORT_SYMBOL(d_invalidate
);
431 /* This must be called with d_lock held */
432 static inline void __dget_dlock(struct dentry
*dentry
)
437 static inline void __dget(struct dentry
*dentry
)
439 spin_lock(&dentry
->d_lock
);
440 __dget_dlock(dentry
);
441 spin_unlock(&dentry
->d_lock
);
444 struct dentry
*dget_parent(struct dentry
*dentry
)
450 * Don't need rcu_dereference because we re-check it was correct under
454 ret
= dentry
->d_parent
;
459 spin_lock(&ret
->d_lock
);
460 if (unlikely(ret
!= dentry
->d_parent
)) {
461 spin_unlock(&ret
->d_lock
);
466 BUG_ON(!ret
->d_count
);
468 spin_unlock(&ret
->d_lock
);
472 EXPORT_SYMBOL(dget_parent
);
475 * d_find_alias - grab a hashed alias of inode
476 * @inode: inode in question
477 * @want_discon: flag, used by d_splice_alias, to request
478 * that only a DISCONNECTED alias be returned.
480 * If inode has a hashed alias, or is a directory and has any alias,
481 * acquire the reference to alias and return it. Otherwise return NULL.
482 * Notice that if inode is a directory there can be only one alias and
483 * it can be unhashed only if it has no children, or if it is the root
486 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
487 * any other hashed alias over that one unless @want_discon is set,
488 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
490 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
492 struct dentry
*alias
, *discon_alias
;
496 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
497 spin_lock(&alias
->d_lock
);
498 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
499 if (IS_ROOT(alias
) &&
500 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
501 discon_alias
= alias
;
502 } else if (!want_discon
) {
504 spin_unlock(&alias
->d_lock
);
508 spin_unlock(&alias
->d_lock
);
511 alias
= discon_alias
;
512 spin_lock(&alias
->d_lock
);
513 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
514 if (IS_ROOT(alias
) &&
515 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
517 spin_unlock(&alias
->d_lock
);
521 spin_unlock(&alias
->d_lock
);
527 struct dentry
*d_find_alias(struct inode
*inode
)
529 struct dentry
*de
= NULL
;
531 if (!list_empty(&inode
->i_dentry
)) {
532 spin_lock(&dcache_inode_lock
);
533 de
= __d_find_alias(inode
, 0);
534 spin_unlock(&dcache_inode_lock
);
538 EXPORT_SYMBOL(d_find_alias
);
541 * Try to kill dentries associated with this inode.
542 * WARNING: you must own a reference to inode.
544 void d_prune_aliases(struct inode
*inode
)
546 struct dentry
*dentry
;
548 spin_lock(&dcache_inode_lock
);
549 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
550 spin_lock(&dentry
->d_lock
);
551 if (!dentry
->d_count
) {
552 __dget_dlock(dentry
);
554 spin_unlock(&dentry
->d_lock
);
555 spin_unlock(&dcache_inode_lock
);
559 spin_unlock(&dentry
->d_lock
);
561 spin_unlock(&dcache_inode_lock
);
563 EXPORT_SYMBOL(d_prune_aliases
);
566 * Throw away a dentry - free the inode, dput the parent. This requires that
567 * the LRU list has already been removed.
569 * Try to prune ancestors as well. This is necessary to prevent
570 * quadratic behavior of shrink_dcache_parent(), but is also expected
571 * to be beneficial in reducing dentry cache fragmentation.
573 static void prune_one_dentry(struct dentry
*dentry
, struct dentry
*parent
)
574 __releases(dentry
->d_lock
)
575 __releases(parent
->d_lock
)
576 __releases(dcache_inode_lock
)
579 dentry
= d_kill(dentry
, parent
);
585 spin_lock(&dcache_inode_lock
);
587 spin_lock(&dentry
->d_lock
);
591 parent
= dentry
->d_parent
;
592 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
593 spin_unlock(&dentry
->d_lock
);
597 if (dentry
->d_count
) {
599 spin_unlock(&parent
->d_lock
);
600 spin_unlock(&dentry
->d_lock
);
601 spin_unlock(&dcache_inode_lock
);
605 dentry_lru_del(dentry
);
607 dentry
= d_kill(dentry
, parent
);
611 static void shrink_dentry_list(struct list_head
*list
)
613 struct dentry
*dentry
;
615 while (!list_empty(list
)) {
616 struct dentry
*parent
;
618 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
620 if (!spin_trylock(&dentry
->d_lock
)) {
622 spin_unlock(&dcache_lru_lock
);
624 spin_lock(&dcache_lru_lock
);
629 * We found an inuse dentry which was not removed from
630 * the LRU because of laziness during lookup. Do not free
631 * it - just keep it off the LRU list.
633 if (dentry
->d_count
) {
634 __dentry_lru_del(dentry
);
635 spin_unlock(&dentry
->d_lock
);
641 parent
= dentry
->d_parent
;
642 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
643 spin_unlock(&dentry
->d_lock
);
646 __dentry_lru_del(dentry
);
647 spin_unlock(&dcache_lru_lock
);
649 prune_one_dentry(dentry
, parent
);
650 /* dcache_inode_lock and dentry->d_lock dropped */
651 spin_lock(&dcache_inode_lock
);
652 spin_lock(&dcache_lru_lock
);
657 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
658 * @sb: superblock to shrink dentry LRU.
659 * @count: number of entries to prune
660 * @flags: flags to control the dentry processing
662 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
664 static void __shrink_dcache_sb(struct super_block
*sb
, int *count
, int flags
)
666 /* called from prune_dcache() and shrink_dcache_parent() */
667 struct dentry
*dentry
;
668 LIST_HEAD(referenced
);
672 spin_lock(&dcache_inode_lock
);
674 spin_lock(&dcache_lru_lock
);
675 while (!list_empty(&sb
->s_dentry_lru
)) {
676 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
677 struct dentry
, d_lru
);
678 BUG_ON(dentry
->d_sb
!= sb
);
680 if (!spin_trylock(&dentry
->d_lock
)) {
681 spin_unlock(&dcache_lru_lock
);
687 * If we are honouring the DCACHE_REFERENCED flag and the
688 * dentry has this flag set, don't free it. Clear the flag
689 * and put it back on the LRU.
691 if (flags
& DCACHE_REFERENCED
&&
692 dentry
->d_flags
& DCACHE_REFERENCED
) {
693 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
694 list_move(&dentry
->d_lru
, &referenced
);
695 spin_unlock(&dentry
->d_lock
);
697 list_move_tail(&dentry
->d_lru
, &tmp
);
698 spin_unlock(&dentry
->d_lock
);
702 /* XXX: re-add cond_resched_lock when dcache_lock goes away */
706 shrink_dentry_list(&tmp
);
708 if (!list_empty(&referenced
))
709 list_splice(&referenced
, &sb
->s_dentry_lru
);
710 spin_unlock(&dcache_lru_lock
);
711 spin_unlock(&dcache_inode_lock
);
715 * prune_dcache - shrink the dcache
716 * @count: number of entries to try to free
718 * Shrink the dcache. This is done when we need more memory, or simply when we
719 * need to unmount something (at which point we need to unuse all dentries).
721 * This function may fail to free any resources if all the dentries are in use.
723 static void prune_dcache(int count
)
725 struct super_block
*sb
, *p
= NULL
;
727 int unused
= dentry_stat
.nr_unused
;
731 if (unused
== 0 || count
== 0)
736 prune_ratio
= unused
/ count
;
738 list_for_each_entry(sb
, &super_blocks
, s_list
) {
739 if (list_empty(&sb
->s_instances
))
741 if (sb
->s_nr_dentry_unused
== 0)
744 /* Now, we reclaim unused dentrins with fairness.
745 * We reclaim them same percentage from each superblock.
746 * We calculate number of dentries to scan on this sb
747 * as follows, but the implementation is arranged to avoid
749 * number of dentries to scan on this sb =
750 * count * (number of dentries on this sb /
751 * number of dentries in the machine)
753 spin_unlock(&sb_lock
);
754 if (prune_ratio
!= 1)
755 w_count
= (sb
->s_nr_dentry_unused
/ prune_ratio
) + 1;
757 w_count
= sb
->s_nr_dentry_unused
;
760 * We need to be sure this filesystem isn't being unmounted,
761 * otherwise we could race with generic_shutdown_super(), and
762 * end up holding a reference to an inode while the filesystem
763 * is unmounted. So we try to get s_umount, and make sure
766 if (down_read_trylock(&sb
->s_umount
)) {
767 if ((sb
->s_root
!= NULL
) &&
768 (!list_empty(&sb
->s_dentry_lru
))) {
769 __shrink_dcache_sb(sb
, &w_count
,
773 up_read(&sb
->s_umount
);
780 /* more work left to do? */
786 spin_unlock(&sb_lock
);
790 * shrink_dcache_sb - shrink dcache for a superblock
793 * Shrink the dcache for the specified super block. This is used to free
794 * the dcache before unmounting a file system.
796 void shrink_dcache_sb(struct super_block
*sb
)
800 spin_lock(&dcache_inode_lock
);
801 spin_lock(&dcache_lru_lock
);
802 while (!list_empty(&sb
->s_dentry_lru
)) {
803 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
804 shrink_dentry_list(&tmp
);
806 spin_unlock(&dcache_lru_lock
);
807 spin_unlock(&dcache_inode_lock
);
809 EXPORT_SYMBOL(shrink_dcache_sb
);
812 * destroy a single subtree of dentries for unmount
813 * - see the comments on shrink_dcache_for_umount() for a description of the
816 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
818 struct dentry
*parent
;
819 unsigned detached
= 0;
821 BUG_ON(!IS_ROOT(dentry
));
823 /* detach this root from the system */
824 spin_lock(&dentry
->d_lock
);
825 dentry_lru_del(dentry
);
827 spin_unlock(&dentry
->d_lock
);
830 /* descend to the first leaf in the current subtree */
831 while (!list_empty(&dentry
->d_subdirs
)) {
834 /* this is a branch with children - detach all of them
835 * from the system in one go */
836 spin_lock(&dentry
->d_lock
);
837 list_for_each_entry(loop
, &dentry
->d_subdirs
,
839 spin_lock_nested(&loop
->d_lock
,
840 DENTRY_D_LOCK_NESTED
);
841 dentry_lru_del(loop
);
843 spin_unlock(&loop
->d_lock
);
845 spin_unlock(&dentry
->d_lock
);
847 /* move to the first child */
848 dentry
= list_entry(dentry
->d_subdirs
.next
,
849 struct dentry
, d_u
.d_child
);
852 /* consume the dentries from this leaf up through its parents
853 * until we find one with children or run out altogether */
857 if (dentry
->d_count
!= 0) {
859 "BUG: Dentry %p{i=%lx,n=%s}"
861 " [unmount of %s %s]\n",
864 dentry
->d_inode
->i_ino
: 0UL,
867 dentry
->d_sb
->s_type
->name
,
872 if (IS_ROOT(dentry
)) {
874 list_del(&dentry
->d_u
.d_child
);
876 parent
= dentry
->d_parent
;
877 spin_lock(&parent
->d_lock
);
879 list_del(&dentry
->d_u
.d_child
);
880 spin_unlock(&parent
->d_lock
);
885 inode
= dentry
->d_inode
;
887 dentry
->d_inode
= NULL
;
888 list_del_init(&dentry
->d_alias
);
889 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
890 dentry
->d_op
->d_iput(dentry
, inode
);
897 /* finished when we fall off the top of the tree,
898 * otherwise we ascend to the parent and move to the
899 * next sibling if there is one */
903 } while (list_empty(&dentry
->d_subdirs
));
905 dentry
= list_entry(dentry
->d_subdirs
.next
,
906 struct dentry
, d_u
.d_child
);
911 * destroy the dentries attached to a superblock on unmounting
912 * - we don't need to use dentry->d_lock because:
913 * - the superblock is detached from all mountings and open files, so the
914 * dentry trees will not be rearranged by the VFS
915 * - s_umount is write-locked, so the memory pressure shrinker will ignore
916 * any dentries belonging to this superblock that it comes across
917 * - the filesystem itself is no longer permitted to rearrange the dentries
920 void shrink_dcache_for_umount(struct super_block
*sb
)
922 struct dentry
*dentry
;
924 if (down_read_trylock(&sb
->s_umount
))
929 spin_lock(&dentry
->d_lock
);
931 spin_unlock(&dentry
->d_lock
);
932 shrink_dcache_for_umount_subtree(dentry
);
934 while (!hlist_empty(&sb
->s_anon
)) {
935 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
936 shrink_dcache_for_umount_subtree(dentry
);
941 * Search for at least 1 mount point in the dentry's subdirs.
942 * We descend to the next level whenever the d_subdirs
943 * list is non-empty and continue searching.
947 * have_submounts - check for mounts over a dentry
948 * @parent: dentry to check.
950 * Return true if the parent or its subdirectories contain
953 int have_submounts(struct dentry
*parent
)
955 struct dentry
*this_parent
;
956 struct list_head
*next
;
960 seq
= read_seqbegin(&rename_lock
);
962 this_parent
= parent
;
964 if (d_mountpoint(parent
))
966 spin_lock(&this_parent
->d_lock
);
968 next
= this_parent
->d_subdirs
.next
;
970 while (next
!= &this_parent
->d_subdirs
) {
971 struct list_head
*tmp
= next
;
972 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
975 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
976 /* Have we found a mount point ? */
977 if (d_mountpoint(dentry
)) {
978 spin_unlock(&dentry
->d_lock
);
979 spin_unlock(&this_parent
->d_lock
);
982 if (!list_empty(&dentry
->d_subdirs
)) {
983 spin_unlock(&this_parent
->d_lock
);
984 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
985 this_parent
= dentry
;
986 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
989 spin_unlock(&dentry
->d_lock
);
992 * All done at this level ... ascend and resume the search.
994 if (this_parent
!= parent
) {
996 struct dentry
*child
;
998 tmp
= this_parent
->d_parent
;
1000 spin_unlock(&this_parent
->d_lock
);
1001 child
= this_parent
;
1003 spin_lock(&this_parent
->d_lock
);
1004 /* might go back up the wrong parent if we have had a rename
1006 if (this_parent
!= child
->d_parent
||
1007 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1008 spin_unlock(&this_parent
->d_lock
);
1013 next
= child
->d_u
.d_child
.next
;
1016 spin_unlock(&this_parent
->d_lock
);
1017 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1020 write_sequnlock(&rename_lock
);
1021 return 0; /* No mount points found in tree */
1023 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1026 write_sequnlock(&rename_lock
);
1031 write_seqlock(&rename_lock
);
1034 EXPORT_SYMBOL(have_submounts
);
1037 * Search the dentry child list for the specified parent,
1038 * and move any unused dentries to the end of the unused
1039 * list for prune_dcache(). We descend to the next level
1040 * whenever the d_subdirs list is non-empty and continue
1043 * It returns zero iff there are no unused children,
1044 * otherwise it returns the number of children moved to
1045 * the end of the unused list. This may not be the total
1046 * number of unused children, because select_parent can
1047 * drop the lock and return early due to latency
1050 static int select_parent(struct dentry
* parent
)
1052 struct dentry
*this_parent
;
1053 struct list_head
*next
;
1058 seq
= read_seqbegin(&rename_lock
);
1060 this_parent
= parent
;
1061 spin_lock(&this_parent
->d_lock
);
1063 next
= this_parent
->d_subdirs
.next
;
1065 while (next
!= &this_parent
->d_subdirs
) {
1066 struct list_head
*tmp
= next
;
1067 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1070 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1073 * move only zero ref count dentries to the end
1074 * of the unused list for prune_dcache
1076 if (!dentry
->d_count
) {
1077 dentry_lru_move_tail(dentry
);
1080 dentry_lru_del(dentry
);
1084 * We can return to the caller if we have found some (this
1085 * ensures forward progress). We'll be coming back to find
1088 if (found
&& need_resched()) {
1089 spin_unlock(&dentry
->d_lock
);
1094 * Descend a level if the d_subdirs list is non-empty.
1096 if (!list_empty(&dentry
->d_subdirs
)) {
1097 spin_unlock(&this_parent
->d_lock
);
1098 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1099 this_parent
= dentry
;
1100 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1104 spin_unlock(&dentry
->d_lock
);
1107 * All done at this level ... ascend and resume the search.
1109 if (this_parent
!= parent
) {
1111 struct dentry
*child
;
1113 tmp
= this_parent
->d_parent
;
1115 spin_unlock(&this_parent
->d_lock
);
1116 child
= this_parent
;
1118 spin_lock(&this_parent
->d_lock
);
1119 /* might go back up the wrong parent if we have had a rename
1121 if (this_parent
!= child
->d_parent
||
1122 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1123 spin_unlock(&this_parent
->d_lock
);
1128 next
= child
->d_u
.d_child
.next
;
1132 spin_unlock(&this_parent
->d_lock
);
1133 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1136 write_sequnlock(&rename_lock
);
1143 write_seqlock(&rename_lock
);
1148 * shrink_dcache_parent - prune dcache
1149 * @parent: parent of entries to prune
1151 * Prune the dcache to remove unused children of the parent dentry.
1154 void shrink_dcache_parent(struct dentry
* parent
)
1156 struct super_block
*sb
= parent
->d_sb
;
1159 while ((found
= select_parent(parent
)) != 0)
1160 __shrink_dcache_sb(sb
, &found
, 0);
1162 EXPORT_SYMBOL(shrink_dcache_parent
);
1165 * Scan `nr' dentries and return the number which remain.
1167 * We need to avoid reentering the filesystem if the caller is performing a
1168 * GFP_NOFS allocation attempt. One example deadlock is:
1170 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1171 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1172 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1174 * In this case we return -1 to tell the caller that we baled.
1176 static int shrink_dcache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
1179 if (!(gfp_mask
& __GFP_FS
))
1184 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
1187 static struct shrinker dcache_shrinker
= {
1188 .shrink
= shrink_dcache_memory
,
1189 .seeks
= DEFAULT_SEEKS
,
1193 * d_alloc - allocate a dcache entry
1194 * @parent: parent of entry to allocate
1195 * @name: qstr of the name
1197 * Allocates a dentry. It returns %NULL if there is insufficient memory
1198 * available. On a success the dentry is returned. The name passed in is
1199 * copied and the copy passed in may be reused after this call.
1202 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1204 struct dentry
*dentry
;
1207 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1211 if (name
->len
> DNAME_INLINE_LEN
-1) {
1212 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1214 kmem_cache_free(dentry_cache
, dentry
);
1218 dname
= dentry
->d_iname
;
1220 dentry
->d_name
.name
= dname
;
1222 dentry
->d_name
.len
= name
->len
;
1223 dentry
->d_name
.hash
= name
->hash
;
1224 memcpy(dname
, name
->name
, name
->len
);
1225 dname
[name
->len
] = 0;
1227 dentry
->d_count
= 1;
1228 dentry
->d_flags
= DCACHE_UNHASHED
;
1229 spin_lock_init(&dentry
->d_lock
);
1230 dentry
->d_inode
= NULL
;
1231 dentry
->d_parent
= NULL
;
1232 dentry
->d_sb
= NULL
;
1233 dentry
->d_op
= NULL
;
1234 dentry
->d_fsdata
= NULL
;
1235 dentry
->d_mounted
= 0;
1236 INIT_HLIST_NODE(&dentry
->d_hash
);
1237 INIT_LIST_HEAD(&dentry
->d_lru
);
1238 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1239 INIT_LIST_HEAD(&dentry
->d_alias
);
1240 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1243 spin_lock(&parent
->d_lock
);
1245 * don't need child lock because it is not subject
1246 * to concurrency here
1248 __dget_dlock(parent
);
1249 dentry
->d_parent
= parent
;
1250 dentry
->d_sb
= parent
->d_sb
;
1251 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1252 spin_unlock(&parent
->d_lock
);
1255 this_cpu_inc(nr_dentry
);
1259 EXPORT_SYMBOL(d_alloc
);
1261 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1266 q
.len
= strlen(name
);
1267 q
.hash
= full_name_hash(q
.name
, q
.len
);
1268 return d_alloc(parent
, &q
);
1270 EXPORT_SYMBOL(d_alloc_name
);
1272 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1274 spin_lock(&dentry
->d_lock
);
1276 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1277 dentry
->d_inode
= inode
;
1278 spin_unlock(&dentry
->d_lock
);
1279 fsnotify_d_instantiate(dentry
, inode
);
1283 * d_instantiate - fill in inode information for a dentry
1284 * @entry: dentry to complete
1285 * @inode: inode to attach to this dentry
1287 * Fill in inode information in the entry.
1289 * This turns negative dentries into productive full members
1292 * NOTE! This assumes that the inode count has been incremented
1293 * (or otherwise set) by the caller to indicate that it is now
1294 * in use by the dcache.
1297 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1299 BUG_ON(!list_empty(&entry
->d_alias
));
1300 spin_lock(&dcache_inode_lock
);
1301 __d_instantiate(entry
, inode
);
1302 spin_unlock(&dcache_inode_lock
);
1303 security_d_instantiate(entry
, inode
);
1305 EXPORT_SYMBOL(d_instantiate
);
1308 * d_instantiate_unique - instantiate a non-aliased dentry
1309 * @entry: dentry to instantiate
1310 * @inode: inode to attach to this dentry
1312 * Fill in inode information in the entry. On success, it returns NULL.
1313 * If an unhashed alias of "entry" already exists, then we return the
1314 * aliased dentry instead and drop one reference to inode.
1316 * Note that in order to avoid conflicts with rename() etc, the caller
1317 * had better be holding the parent directory semaphore.
1319 * This also assumes that the inode count has been incremented
1320 * (or otherwise set) by the caller to indicate that it is now
1321 * in use by the dcache.
1323 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1324 struct inode
*inode
)
1326 struct dentry
*alias
;
1327 int len
= entry
->d_name
.len
;
1328 const char *name
= entry
->d_name
.name
;
1329 unsigned int hash
= entry
->d_name
.hash
;
1332 __d_instantiate(entry
, NULL
);
1336 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1337 struct qstr
*qstr
= &alias
->d_name
;
1340 * Don't need alias->d_lock here, because aliases with
1341 * d_parent == entry->d_parent are not subject to name or
1342 * parent changes, because the parent inode i_mutex is held.
1344 if (qstr
->hash
!= hash
)
1346 if (alias
->d_parent
!= entry
->d_parent
)
1348 if (qstr
->len
!= len
)
1350 if (memcmp(qstr
->name
, name
, len
))
1356 __d_instantiate(entry
, inode
);
1360 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1362 struct dentry
*result
;
1364 BUG_ON(!list_empty(&entry
->d_alias
));
1366 spin_lock(&dcache_inode_lock
);
1367 result
= __d_instantiate_unique(entry
, inode
);
1368 spin_unlock(&dcache_inode_lock
);
1371 security_d_instantiate(entry
, inode
);
1375 BUG_ON(!d_unhashed(result
));
1380 EXPORT_SYMBOL(d_instantiate_unique
);
1383 * d_alloc_root - allocate root dentry
1384 * @root_inode: inode to allocate the root for
1386 * Allocate a root ("/") dentry for the inode given. The inode is
1387 * instantiated and returned. %NULL is returned if there is insufficient
1388 * memory or the inode passed is %NULL.
1391 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1393 struct dentry
*res
= NULL
;
1396 static const struct qstr name
= { .name
= "/", .len
= 1 };
1398 res
= d_alloc(NULL
, &name
);
1400 res
->d_sb
= root_inode
->i_sb
;
1401 res
->d_parent
= res
;
1402 d_instantiate(res
, root_inode
);
1407 EXPORT_SYMBOL(d_alloc_root
);
1409 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1412 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1413 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1414 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1418 * d_obtain_alias - find or allocate a dentry for a given inode
1419 * @inode: inode to allocate the dentry for
1421 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1422 * similar open by handle operations. The returned dentry may be anonymous,
1423 * or may have a full name (if the inode was already in the cache).
1425 * When called on a directory inode, we must ensure that the inode only ever
1426 * has one dentry. If a dentry is found, that is returned instead of
1427 * allocating a new one.
1429 * On successful return, the reference to the inode has been transferred
1430 * to the dentry. In case of an error the reference on the inode is released.
1431 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1432 * be passed in and will be the error will be propagate to the return value,
1433 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1435 struct dentry
*d_obtain_alias(struct inode
*inode
)
1437 static const struct qstr anonstring
= { .name
= "" };
1442 return ERR_PTR(-ESTALE
);
1444 return ERR_CAST(inode
);
1446 res
= d_find_alias(inode
);
1450 tmp
= d_alloc(NULL
, &anonstring
);
1452 res
= ERR_PTR(-ENOMEM
);
1455 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1458 spin_lock(&dcache_inode_lock
);
1459 res
= __d_find_alias(inode
, 0);
1461 spin_unlock(&dcache_inode_lock
);
1466 /* attach a disconnected dentry */
1467 spin_lock(&tmp
->d_lock
);
1468 tmp
->d_sb
= inode
->i_sb
;
1469 tmp
->d_inode
= inode
;
1470 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1471 tmp
->d_flags
&= ~DCACHE_UNHASHED
;
1472 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1473 spin_lock(&dcache_hash_lock
);
1474 hlist_add_head(&tmp
->d_hash
, &inode
->i_sb
->s_anon
);
1475 spin_unlock(&dcache_hash_lock
);
1476 spin_unlock(&tmp
->d_lock
);
1477 spin_unlock(&dcache_inode_lock
);
1485 EXPORT_SYMBOL(d_obtain_alias
);
1488 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1489 * @inode: the inode which may have a disconnected dentry
1490 * @dentry: a negative dentry which we want to point to the inode.
1492 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1493 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1494 * and return it, else simply d_add the inode to the dentry and return NULL.
1496 * This is needed in the lookup routine of any filesystem that is exportable
1497 * (via knfsd) so that we can build dcache paths to directories effectively.
1499 * If a dentry was found and moved, then it is returned. Otherwise NULL
1500 * is returned. This matches the expected return value of ->lookup.
1503 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1505 struct dentry
*new = NULL
;
1507 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1508 spin_lock(&dcache_inode_lock
);
1509 new = __d_find_alias(inode
, 1);
1511 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1512 spin_unlock(&dcache_inode_lock
);
1513 security_d_instantiate(new, inode
);
1514 d_move(new, dentry
);
1517 /* already taking dcache_inode_lock, so d_add() by hand */
1518 __d_instantiate(dentry
, inode
);
1519 spin_unlock(&dcache_inode_lock
);
1520 security_d_instantiate(dentry
, inode
);
1524 d_add(dentry
, inode
);
1527 EXPORT_SYMBOL(d_splice_alias
);
1530 * d_add_ci - lookup or allocate new dentry with case-exact name
1531 * @inode: the inode case-insensitive lookup has found
1532 * @dentry: the negative dentry that was passed to the parent's lookup func
1533 * @name: the case-exact name to be associated with the returned dentry
1535 * This is to avoid filling the dcache with case-insensitive names to the
1536 * same inode, only the actual correct case is stored in the dcache for
1537 * case-insensitive filesystems.
1539 * For a case-insensitive lookup match and if the the case-exact dentry
1540 * already exists in in the dcache, use it and return it.
1542 * If no entry exists with the exact case name, allocate new dentry with
1543 * the exact case, and return the spliced entry.
1545 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1549 struct dentry
*found
;
1553 * First check if a dentry matching the name already exists,
1554 * if not go ahead and create it now.
1556 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1558 new = d_alloc(dentry
->d_parent
, name
);
1564 found
= d_splice_alias(inode
, new);
1573 * If a matching dentry exists, and it's not negative use it.
1575 * Decrement the reference count to balance the iget() done
1578 if (found
->d_inode
) {
1579 if (unlikely(found
->d_inode
!= inode
)) {
1580 /* This can't happen because bad inodes are unhashed. */
1581 BUG_ON(!is_bad_inode(inode
));
1582 BUG_ON(!is_bad_inode(found
->d_inode
));
1589 * Negative dentry: instantiate it unless the inode is a directory and
1590 * already has a dentry.
1592 spin_lock(&dcache_inode_lock
);
1593 if (!S_ISDIR(inode
->i_mode
) || list_empty(&inode
->i_dentry
)) {
1594 __d_instantiate(found
, inode
);
1595 spin_unlock(&dcache_inode_lock
);
1596 security_d_instantiate(found
, inode
);
1601 * In case a directory already has a (disconnected) entry grab a
1602 * reference to it, move it in place and use it.
1604 new = list_entry(inode
->i_dentry
.next
, struct dentry
, d_alias
);
1606 spin_unlock(&dcache_inode_lock
);
1607 security_d_instantiate(found
, inode
);
1615 return ERR_PTR(error
);
1617 EXPORT_SYMBOL(d_add_ci
);
1620 * d_lookup - search for a dentry
1621 * @parent: parent dentry
1622 * @name: qstr of name we wish to find
1623 * Returns: dentry, or NULL
1625 * d_lookup searches the children of the parent dentry for the name in
1626 * question. If the dentry is found its reference count is incremented and the
1627 * dentry is returned. The caller must use dput to free the entry when it has
1628 * finished using it. %NULL is returned if the dentry does not exist.
1630 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1632 struct dentry
* dentry
= NULL
;
1636 seq
= read_seqbegin(&rename_lock
);
1637 dentry
= __d_lookup(parent
, name
);
1640 } while (read_seqretry(&rename_lock
, seq
));
1643 EXPORT_SYMBOL(d_lookup
);
1646 * __d_lookup - search for a dentry (racy)
1647 * @parent: parent dentry
1648 * @name: qstr of name we wish to find
1649 * Returns: dentry, or NULL
1651 * __d_lookup is like d_lookup, however it may (rarely) return a
1652 * false-negative result due to unrelated rename activity.
1654 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1655 * however it must be used carefully, eg. with a following d_lookup in
1656 * the case of failure.
1658 * __d_lookup callers must be commented.
1660 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1662 unsigned int len
= name
->len
;
1663 unsigned int hash
= name
->hash
;
1664 const unsigned char *str
= name
->name
;
1665 struct hlist_head
*head
= d_hash(parent
,hash
);
1666 struct dentry
*found
= NULL
;
1667 struct hlist_node
*node
;
1668 struct dentry
*dentry
;
1671 * The hash list is protected using RCU.
1673 * Take d_lock when comparing a candidate dentry, to avoid races
1676 * It is possible that concurrent renames can mess up our list
1677 * walk here and result in missing our dentry, resulting in the
1678 * false-negative result. d_lookup() protects against concurrent
1679 * renames using rename_lock seqlock.
1681 * See Documentation/vfs/dcache-locking.txt for more details.
1685 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1688 if (dentry
->d_name
.hash
!= hash
)
1690 if (dentry
->d_parent
!= parent
)
1693 spin_lock(&dentry
->d_lock
);
1696 * Recheck the dentry after taking the lock - d_move may have
1697 * changed things. Don't bother checking the hash because
1698 * we're about to compare the whole name anyway.
1700 if (dentry
->d_parent
!= parent
)
1703 /* non-existing due to RCU? */
1704 if (d_unhashed(dentry
))
1708 * It is safe to compare names since d_move() cannot
1709 * change the qstr (protected by d_lock).
1711 qstr
= &dentry
->d_name
;
1712 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1713 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1714 dentry
, dentry
->d_inode
,
1715 qstr
->len
, qstr
->name
, name
))
1718 if (qstr
->len
!= len
)
1720 if (memcmp(qstr
->name
, str
, len
))
1726 spin_unlock(&dentry
->d_lock
);
1729 spin_unlock(&dentry
->d_lock
);
1737 * d_hash_and_lookup - hash the qstr then search for a dentry
1738 * @dir: Directory to search in
1739 * @name: qstr of name we wish to find
1741 * On hash failure or on lookup failure NULL is returned.
1743 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1745 struct dentry
*dentry
= NULL
;
1748 * Check for a fs-specific hash function. Note that we must
1749 * calculate the standard hash first, as the d_op->d_hash()
1750 * routine may choose to leave the hash value unchanged.
1752 name
->hash
= full_name_hash(name
->name
, name
->len
);
1753 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1754 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1757 dentry
= d_lookup(dir
, name
);
1763 * d_validate - verify dentry provided from insecure source (deprecated)
1764 * @dentry: The dentry alleged to be valid child of @dparent
1765 * @dparent: The parent dentry (known to be valid)
1767 * An insecure source has sent us a dentry, here we verify it and dget() it.
1768 * This is used by ncpfs in its readdir implementation.
1769 * Zero is returned in the dentry is invalid.
1771 * This function is slow for big directories, and deprecated, do not use it.
1773 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1775 struct dentry
*child
;
1777 spin_lock(&dparent
->d_lock
);
1778 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1779 if (dentry
== child
) {
1780 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1781 __dget_dlock(dentry
);
1782 spin_unlock(&dentry
->d_lock
);
1783 spin_unlock(&dparent
->d_lock
);
1787 spin_unlock(&dparent
->d_lock
);
1791 EXPORT_SYMBOL(d_validate
);
1794 * When a file is deleted, we have two options:
1795 * - turn this dentry into a negative dentry
1796 * - unhash this dentry and free it.
1798 * Usually, we want to just turn this into
1799 * a negative dentry, but if anybody else is
1800 * currently using the dentry or the inode
1801 * we can't do that and we fall back on removing
1802 * it from the hash queues and waiting for
1803 * it to be deleted later when it has no users
1807 * d_delete - delete a dentry
1808 * @dentry: The dentry to delete
1810 * Turn the dentry into a negative dentry if possible, otherwise
1811 * remove it from the hash queues so it can be deleted later
1814 void d_delete(struct dentry
* dentry
)
1818 * Are we the only user?
1821 spin_lock(&dentry
->d_lock
);
1822 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1823 if (dentry
->d_count
== 1) {
1824 if (!spin_trylock(&dcache_inode_lock
)) {
1825 spin_unlock(&dentry
->d_lock
);
1829 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1830 dentry_iput(dentry
);
1831 fsnotify_nameremove(dentry
, isdir
);
1835 if (!d_unhashed(dentry
))
1838 spin_unlock(&dentry
->d_lock
);
1840 fsnotify_nameremove(dentry
, isdir
);
1842 EXPORT_SYMBOL(d_delete
);
1844 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1847 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1848 hlist_add_head_rcu(&entry
->d_hash
, list
);
1851 static void _d_rehash(struct dentry
* entry
)
1853 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1857 * d_rehash - add an entry back to the hash
1858 * @entry: dentry to add to the hash
1860 * Adds a dentry to the hash according to its name.
1863 void d_rehash(struct dentry
* entry
)
1865 spin_lock(&entry
->d_lock
);
1866 spin_lock(&dcache_hash_lock
);
1868 spin_unlock(&dcache_hash_lock
);
1869 spin_unlock(&entry
->d_lock
);
1871 EXPORT_SYMBOL(d_rehash
);
1874 * dentry_update_name_case - update case insensitive dentry with a new name
1875 * @dentry: dentry to be updated
1878 * Update a case insensitive dentry with new case of name.
1880 * dentry must have been returned by d_lookup with name @name. Old and new
1881 * name lengths must match (ie. no d_compare which allows mismatched name
1884 * Parent inode i_mutex must be held over d_lookup and into this call (to
1885 * keep renames and concurrent inserts, and readdir(2) away).
1887 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
1889 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
1890 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
1892 spin_lock(&dentry
->d_lock
);
1893 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
1894 spin_unlock(&dentry
->d_lock
);
1896 EXPORT_SYMBOL(dentry_update_name_case
);
1898 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1900 if (dname_external(target
)) {
1901 if (dname_external(dentry
)) {
1903 * Both external: swap the pointers
1905 swap(target
->d_name
.name
, dentry
->d_name
.name
);
1908 * dentry:internal, target:external. Steal target's
1909 * storage and make target internal.
1911 memcpy(target
->d_iname
, dentry
->d_name
.name
,
1912 dentry
->d_name
.len
+ 1);
1913 dentry
->d_name
.name
= target
->d_name
.name
;
1914 target
->d_name
.name
= target
->d_iname
;
1917 if (dname_external(dentry
)) {
1919 * dentry:external, target:internal. Give dentry's
1920 * storage to target and make dentry internal
1922 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1923 target
->d_name
.len
+ 1);
1924 target
->d_name
.name
= dentry
->d_name
.name
;
1925 dentry
->d_name
.name
= dentry
->d_iname
;
1928 * Both are internal. Just copy target to dentry
1930 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1931 target
->d_name
.len
+ 1);
1932 dentry
->d_name
.len
= target
->d_name
.len
;
1936 swap(dentry
->d_name
.len
, target
->d_name
.len
);
1939 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
1942 * XXXX: do we really need to take target->d_lock?
1944 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
1945 spin_lock(&target
->d_parent
->d_lock
);
1947 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
1948 spin_lock(&dentry
->d_parent
->d_lock
);
1949 spin_lock_nested(&target
->d_parent
->d_lock
,
1950 DENTRY_D_LOCK_NESTED
);
1952 spin_lock(&target
->d_parent
->d_lock
);
1953 spin_lock_nested(&dentry
->d_parent
->d_lock
,
1954 DENTRY_D_LOCK_NESTED
);
1957 if (target
< dentry
) {
1958 spin_lock_nested(&target
->d_lock
, 2);
1959 spin_lock_nested(&dentry
->d_lock
, 3);
1961 spin_lock_nested(&dentry
->d_lock
, 2);
1962 spin_lock_nested(&target
->d_lock
, 3);
1966 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
1967 struct dentry
*target
)
1969 if (target
->d_parent
!= dentry
->d_parent
)
1970 spin_unlock(&dentry
->d_parent
->d_lock
);
1971 if (target
->d_parent
!= target
)
1972 spin_unlock(&target
->d_parent
->d_lock
);
1976 * When switching names, the actual string doesn't strictly have to
1977 * be preserved in the target - because we're dropping the target
1978 * anyway. As such, we can just do a simple memcpy() to copy over
1979 * the new name before we switch.
1981 * Note that we have to be a lot more careful about getting the hash
1982 * switched - we have to switch the hash value properly even if it
1983 * then no longer matches the actual (corrupted) string of the target.
1984 * The hash value has to match the hash queue that the dentry is on..
1987 * d_move - move a dentry
1988 * @dentry: entry to move
1989 * @target: new dentry
1991 * Update the dcache to reflect the move of a file name. Negative
1992 * dcache entries should not be moved in this way.
1994 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1996 if (!dentry
->d_inode
)
1997 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1999 BUG_ON(d_ancestor(dentry
, target
));
2000 BUG_ON(d_ancestor(target
, dentry
));
2002 write_seqlock(&rename_lock
);
2004 dentry_lock_for_move(dentry
, target
);
2006 /* Move the dentry to the target hash queue, if on different bucket */
2007 spin_lock(&dcache_hash_lock
);
2008 if (!d_unhashed(dentry
))
2009 hlist_del_rcu(&dentry
->d_hash
);
2010 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2011 spin_unlock(&dcache_hash_lock
);
2013 /* Unhash the target: dput() will then get rid of it */
2016 list_del(&dentry
->d_u
.d_child
);
2017 list_del(&target
->d_u
.d_child
);
2019 /* Switch the names.. */
2020 switch_names(dentry
, target
);
2021 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2023 /* ... and switch the parents */
2024 if (IS_ROOT(dentry
)) {
2025 dentry
->d_parent
= target
->d_parent
;
2026 target
->d_parent
= target
;
2027 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2029 swap(dentry
->d_parent
, target
->d_parent
);
2031 /* And add them back to the (new) parent lists */
2032 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2035 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2037 dentry_unlock_parents_for_move(dentry
, target
);
2038 spin_unlock(&target
->d_lock
);
2039 fsnotify_d_move(dentry
);
2040 spin_unlock(&dentry
->d_lock
);
2041 write_sequnlock(&rename_lock
);
2043 EXPORT_SYMBOL(d_move
);
2046 * d_ancestor - search for an ancestor
2047 * @p1: ancestor dentry
2050 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2051 * an ancestor of p2, else NULL.
2053 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2057 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2058 if (p
->d_parent
== p1
)
2065 * This helper attempts to cope with remotely renamed directories
2067 * It assumes that the caller is already holding
2068 * dentry->d_parent->d_inode->i_mutex and the dcache_inode_lock
2070 * Note: If ever the locking in lock_rename() changes, then please
2071 * remember to update this too...
2073 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
2074 __releases(dcache_inode_lock
)
2076 struct mutex
*m1
= NULL
, *m2
= NULL
;
2079 /* If alias and dentry share a parent, then no extra locks required */
2080 if (alias
->d_parent
== dentry
->d_parent
)
2083 /* Check for loops */
2084 ret
= ERR_PTR(-ELOOP
);
2085 if (d_ancestor(alias
, dentry
))
2088 /* See lock_rename() */
2089 ret
= ERR_PTR(-EBUSY
);
2090 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2092 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2093 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2095 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2097 d_move(alias
, dentry
);
2100 spin_unlock(&dcache_inode_lock
);
2109 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2110 * named dentry in place of the dentry to be replaced.
2111 * returns with anon->d_lock held!
2113 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2115 struct dentry
*dparent
, *aparent
;
2117 dentry_lock_for_move(anon
, dentry
);
2119 dparent
= dentry
->d_parent
;
2120 aparent
= anon
->d_parent
;
2122 switch_names(dentry
, anon
);
2123 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2125 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2126 list_del(&dentry
->d_u
.d_child
);
2127 if (!IS_ROOT(dentry
))
2128 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2130 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2132 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2133 list_del(&anon
->d_u
.d_child
);
2135 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2137 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2139 dentry_unlock_parents_for_move(anon
, dentry
);
2140 spin_unlock(&dentry
->d_lock
);
2142 /* anon->d_lock still locked, returns locked */
2143 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2147 * d_materialise_unique - introduce an inode into the tree
2148 * @dentry: candidate dentry
2149 * @inode: inode to bind to the dentry, to which aliases may be attached
2151 * Introduces an dentry into the tree, substituting an extant disconnected
2152 * root directory alias in its place if there is one
2154 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2156 struct dentry
*actual
;
2158 BUG_ON(!d_unhashed(dentry
));
2162 __d_instantiate(dentry
, NULL
);
2167 spin_lock(&dcache_inode_lock
);
2169 if (S_ISDIR(inode
->i_mode
)) {
2170 struct dentry
*alias
;
2172 /* Does an aliased dentry already exist? */
2173 alias
= __d_find_alias(inode
, 0);
2176 /* Is this an anonymous mountpoint that we could splice
2178 if (IS_ROOT(alias
)) {
2179 __d_materialise_dentry(dentry
, alias
);
2183 /* Nope, but we must(!) avoid directory aliasing */
2184 actual
= __d_unalias(dentry
, alias
);
2191 /* Add a unique reference */
2192 actual
= __d_instantiate_unique(dentry
, inode
);
2196 BUG_ON(!d_unhashed(actual
));
2198 spin_lock(&actual
->d_lock
);
2200 spin_lock(&dcache_hash_lock
);
2202 spin_unlock(&dcache_hash_lock
);
2203 spin_unlock(&actual
->d_lock
);
2204 spin_unlock(&dcache_inode_lock
);
2206 if (actual
== dentry
) {
2207 security_d_instantiate(dentry
, inode
);
2214 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2216 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2220 return -ENAMETOOLONG
;
2222 memcpy(*buffer
, str
, namelen
);
2226 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2228 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2232 * Prepend path string to a buffer
2234 * @path: the dentry/vfsmount to report
2235 * @root: root vfsmnt/dentry (may be modified by this function)
2236 * @buffer: pointer to the end of the buffer
2237 * @buflen: pointer to buffer length
2239 * Caller holds the rename_lock.
2241 * If path is not reachable from the supplied root, then the value of
2242 * root is changed (without modifying refcounts).
2244 static int prepend_path(const struct path
*path
, struct path
*root
,
2245 char **buffer
, int *buflen
)
2247 struct dentry
*dentry
= path
->dentry
;
2248 struct vfsmount
*vfsmnt
= path
->mnt
;
2252 br_read_lock(vfsmount_lock
);
2253 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2254 struct dentry
* parent
;
2256 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2258 if (vfsmnt
->mnt_parent
== vfsmnt
) {
2261 dentry
= vfsmnt
->mnt_mountpoint
;
2262 vfsmnt
= vfsmnt
->mnt_parent
;
2265 parent
= dentry
->d_parent
;
2267 spin_lock(&dentry
->d_lock
);
2268 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2269 spin_unlock(&dentry
->d_lock
);
2271 error
= prepend(buffer
, buflen
, "/", 1);
2280 if (!error
&& !slash
)
2281 error
= prepend(buffer
, buflen
, "/", 1);
2283 br_read_unlock(vfsmount_lock
);
2288 * Filesystems needing to implement special "root names"
2289 * should do so with ->d_dname()
2291 if (IS_ROOT(dentry
) &&
2292 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2293 WARN(1, "Root dentry has weird name <%.*s>\n",
2294 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2297 root
->dentry
= dentry
;
2302 * __d_path - return the path of a dentry
2303 * @path: the dentry/vfsmount to report
2304 * @root: root vfsmnt/dentry (may be modified by this function)
2305 * @buf: buffer to return value in
2306 * @buflen: buffer length
2308 * Convert a dentry into an ASCII path name.
2310 * Returns a pointer into the buffer or an error code if the
2311 * path was too long.
2313 * "buflen" should be positive.
2315 * If path is not reachable from the supplied root, then the value of
2316 * root is changed (without modifying refcounts).
2318 char *__d_path(const struct path
*path
, struct path
*root
,
2319 char *buf
, int buflen
)
2321 char *res
= buf
+ buflen
;
2324 prepend(&res
, &buflen
, "\0", 1);
2325 write_seqlock(&rename_lock
);
2326 error
= prepend_path(path
, root
, &res
, &buflen
);
2327 write_sequnlock(&rename_lock
);
2330 return ERR_PTR(error
);
2335 * same as __d_path but appends "(deleted)" for unlinked files.
2337 static int path_with_deleted(const struct path
*path
, struct path
*root
,
2338 char **buf
, int *buflen
)
2340 prepend(buf
, buflen
, "\0", 1);
2341 if (d_unlinked(path
->dentry
)) {
2342 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2347 return prepend_path(path
, root
, buf
, buflen
);
2350 static int prepend_unreachable(char **buffer
, int *buflen
)
2352 return prepend(buffer
, buflen
, "(unreachable)", 13);
2356 * d_path - return the path of a dentry
2357 * @path: path to report
2358 * @buf: buffer to return value in
2359 * @buflen: buffer length
2361 * Convert a dentry into an ASCII path name. If the entry has been deleted
2362 * the string " (deleted)" is appended. Note that this is ambiguous.
2364 * Returns a pointer into the buffer or an error code if the path was
2365 * too long. Note: Callers should use the returned pointer, not the passed
2366 * in buffer, to use the name! The implementation often starts at an offset
2367 * into the buffer, and may leave 0 bytes at the start.
2369 * "buflen" should be positive.
2371 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2373 char *res
= buf
+ buflen
;
2379 * We have various synthetic filesystems that never get mounted. On
2380 * these filesystems dentries are never used for lookup purposes, and
2381 * thus don't need to be hashed. They also don't need a name until a
2382 * user wants to identify the object in /proc/pid/fd/. The little hack
2383 * below allows us to generate a name for these objects on demand:
2385 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2386 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2388 get_fs_root(current
->fs
, &root
);
2389 write_seqlock(&rename_lock
);
2391 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2393 res
= ERR_PTR(error
);
2394 write_sequnlock(&rename_lock
);
2398 EXPORT_SYMBOL(d_path
);
2401 * d_path_with_unreachable - return the path of a dentry
2402 * @path: path to report
2403 * @buf: buffer to return value in
2404 * @buflen: buffer length
2406 * The difference from d_path() is that this prepends "(unreachable)"
2407 * to paths which are unreachable from the current process' root.
2409 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2411 char *res
= buf
+ buflen
;
2416 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2417 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2419 get_fs_root(current
->fs
, &root
);
2420 write_seqlock(&rename_lock
);
2422 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2423 if (!error
&& !path_equal(&tmp
, &root
))
2424 error
= prepend_unreachable(&res
, &buflen
);
2425 write_sequnlock(&rename_lock
);
2428 res
= ERR_PTR(error
);
2434 * Helper function for dentry_operations.d_dname() members
2436 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2437 const char *fmt
, ...)
2443 va_start(args
, fmt
);
2444 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2447 if (sz
> sizeof(temp
) || sz
> buflen
)
2448 return ERR_PTR(-ENAMETOOLONG
);
2450 buffer
+= buflen
- sz
;
2451 return memcpy(buffer
, temp
, sz
);
2455 * Write full pathname from the root of the filesystem into the buffer.
2457 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2459 char *end
= buf
+ buflen
;
2462 prepend(&end
, &buflen
, "\0", 1);
2469 while (!IS_ROOT(dentry
)) {
2470 struct dentry
*parent
= dentry
->d_parent
;
2474 spin_lock(&dentry
->d_lock
);
2475 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2476 spin_unlock(&dentry
->d_lock
);
2477 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2485 return ERR_PTR(-ENAMETOOLONG
);
2488 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2492 write_seqlock(&rename_lock
);
2493 retval
= __dentry_path(dentry
, buf
, buflen
);
2494 write_sequnlock(&rename_lock
);
2498 EXPORT_SYMBOL(dentry_path_raw
);
2500 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2505 write_seqlock(&rename_lock
);
2506 if (d_unlinked(dentry
)) {
2508 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2512 retval
= __dentry_path(dentry
, buf
, buflen
);
2513 write_sequnlock(&rename_lock
);
2514 if (!IS_ERR(retval
) && p
)
2515 *p
= '/'; /* restore '/' overriden with '\0' */
2518 return ERR_PTR(-ENAMETOOLONG
);
2522 * NOTE! The user-level library version returns a
2523 * character pointer. The kernel system call just
2524 * returns the length of the buffer filled (which
2525 * includes the ending '\0' character), or a negative
2526 * error value. So libc would do something like
2528 * char *getcwd(char * buf, size_t size)
2532 * retval = sys_getcwd(buf, size);
2539 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2542 struct path pwd
, root
;
2543 char *page
= (char *) __get_free_page(GFP_USER
);
2548 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2551 write_seqlock(&rename_lock
);
2552 if (!d_unlinked(pwd
.dentry
)) {
2554 struct path tmp
= root
;
2555 char *cwd
= page
+ PAGE_SIZE
;
2556 int buflen
= PAGE_SIZE
;
2558 prepend(&cwd
, &buflen
, "\0", 1);
2559 error
= prepend_path(&pwd
, &tmp
, &cwd
, &buflen
);
2560 write_sequnlock(&rename_lock
);
2565 /* Unreachable from current root */
2566 if (!path_equal(&tmp
, &root
)) {
2567 error
= prepend_unreachable(&cwd
, &buflen
);
2573 len
= PAGE_SIZE
+ page
- cwd
;
2576 if (copy_to_user(buf
, cwd
, len
))
2580 write_sequnlock(&rename_lock
);
2586 free_page((unsigned long) page
);
2591 * Test whether new_dentry is a subdirectory of old_dentry.
2593 * Trivially implemented using the dcache structure
2597 * is_subdir - is new dentry a subdirectory of old_dentry
2598 * @new_dentry: new dentry
2599 * @old_dentry: old dentry
2601 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2602 * Returns 0 otherwise.
2603 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2606 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2611 if (new_dentry
== old_dentry
)
2615 /* for restarting inner loop in case of seq retry */
2616 seq
= read_seqbegin(&rename_lock
);
2618 * Need rcu_readlock to protect against the d_parent trashing
2622 if (d_ancestor(old_dentry
, new_dentry
))
2627 } while (read_seqretry(&rename_lock
, seq
));
2632 int path_is_under(struct path
*path1
, struct path
*path2
)
2634 struct vfsmount
*mnt
= path1
->mnt
;
2635 struct dentry
*dentry
= path1
->dentry
;
2638 br_read_lock(vfsmount_lock
);
2639 if (mnt
!= path2
->mnt
) {
2641 if (mnt
->mnt_parent
== mnt
) {
2642 br_read_unlock(vfsmount_lock
);
2645 if (mnt
->mnt_parent
== path2
->mnt
)
2647 mnt
= mnt
->mnt_parent
;
2649 dentry
= mnt
->mnt_mountpoint
;
2651 res
= is_subdir(dentry
, path2
->dentry
);
2652 br_read_unlock(vfsmount_lock
);
2655 EXPORT_SYMBOL(path_is_under
);
2657 void d_genocide(struct dentry
*root
)
2659 struct dentry
*this_parent
;
2660 struct list_head
*next
;
2664 seq
= read_seqbegin(&rename_lock
);
2667 spin_lock(&this_parent
->d_lock
);
2669 next
= this_parent
->d_subdirs
.next
;
2671 while (next
!= &this_parent
->d_subdirs
) {
2672 struct list_head
*tmp
= next
;
2673 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2676 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2677 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2678 spin_unlock(&dentry
->d_lock
);
2681 if (!list_empty(&dentry
->d_subdirs
)) {
2682 spin_unlock(&this_parent
->d_lock
);
2683 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2684 this_parent
= dentry
;
2685 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2688 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2689 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2692 spin_unlock(&dentry
->d_lock
);
2694 if (this_parent
!= root
) {
2696 struct dentry
*child
;
2698 tmp
= this_parent
->d_parent
;
2699 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2700 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2701 this_parent
->d_count
--;
2704 spin_unlock(&this_parent
->d_lock
);
2705 child
= this_parent
;
2707 spin_lock(&this_parent
->d_lock
);
2708 /* might go back up the wrong parent if we have had a rename
2710 if (this_parent
!= child
->d_parent
||
2711 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
2712 spin_unlock(&this_parent
->d_lock
);
2717 next
= child
->d_u
.d_child
.next
;
2720 spin_unlock(&this_parent
->d_lock
);
2721 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2724 write_sequnlock(&rename_lock
);
2729 write_seqlock(&rename_lock
);
2734 * find_inode_number - check for dentry with name
2735 * @dir: directory to check
2736 * @name: Name to find.
2738 * Check whether a dentry already exists for the given name,
2739 * and return the inode number if it has an inode. Otherwise
2742 * This routine is used to post-process directory listings for
2743 * filesystems using synthetic inode numbers, and is necessary
2744 * to keep getcwd() working.
2747 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2749 struct dentry
* dentry
;
2752 dentry
= d_hash_and_lookup(dir
, name
);
2754 if (dentry
->d_inode
)
2755 ino
= dentry
->d_inode
->i_ino
;
2760 EXPORT_SYMBOL(find_inode_number
);
2762 static __initdata
unsigned long dhash_entries
;
2763 static int __init
set_dhash_entries(char *str
)
2767 dhash_entries
= simple_strtoul(str
, &str
, 0);
2770 __setup("dhash_entries=", set_dhash_entries
);
2772 static void __init
dcache_init_early(void)
2776 /* If hashes are distributed across NUMA nodes, defer
2777 * hash allocation until vmalloc space is available.
2783 alloc_large_system_hash("Dentry cache",
2784 sizeof(struct hlist_head
),
2792 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2793 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2796 static void __init
dcache_init(void)
2801 * A constructor could be added for stable state like the lists,
2802 * but it is probably not worth it because of the cache nature
2805 dentry_cache
= KMEM_CACHE(dentry
,
2806 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2808 register_shrinker(&dcache_shrinker
);
2810 /* Hash may have been set up in dcache_init_early */
2815 alloc_large_system_hash("Dentry cache",
2816 sizeof(struct hlist_head
),
2824 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2825 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2828 /* SLAB cache for __getname() consumers */
2829 struct kmem_cache
*names_cachep __read_mostly
;
2830 EXPORT_SYMBOL(names_cachep
);
2832 EXPORT_SYMBOL(d_genocide
);
2834 void __init
vfs_caches_init_early(void)
2836 dcache_init_early();
2840 void __init
vfs_caches_init(unsigned long mempages
)
2842 unsigned long reserve
;
2844 /* Base hash sizes on available memory, with a reserve equal to
2845 150% of current kernel size */
2847 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2848 mempages
-= reserve
;
2850 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2851 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2855 files_init(mempages
);