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>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
42 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects:
45 * - the dcache hash table
46 * s_anon bl list spinlock protects:
47 * - the s_anon list (see __d_drop)
48 * dcache_lru_lock protects:
49 * - the dcache lru lists and counters
56 * - d_parent and d_subdirs
57 * - childrens' d_child and d_parent
61 * dentry->d_inode->i_lock
64 * dcache_hash_bucket lock
67 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock
70 * dentry->d_parent->d_lock
73 * If no ancestor relationship:
74 * if (dentry1 < dentry2)
78 int sysctl_vfs_cache_pressure __read_mostly
= 100;
79 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
81 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
82 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
84 EXPORT_SYMBOL(rename_lock
);
86 static struct kmem_cache
*dentry_cache __read_mostly
;
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
;
102 struct dcache_hash_bucket
{
103 struct hlist_bl_head head
;
105 static struct dcache_hash_bucket
*dentry_hashtable __read_mostly
;
107 static inline struct dcache_hash_bucket
*d_hash(struct dentry
*parent
,
110 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
111 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
112 return dentry_hashtable
+ (hash
& D_HASHMASK
);
115 static inline void spin_lock_bucket(struct dcache_hash_bucket
*b
)
117 bit_spin_lock(0, (unsigned long *)&b
->head
.first
);
120 static inline void spin_unlock_bucket(struct dcache_hash_bucket
*b
)
122 __bit_spin_unlock(0, (unsigned long *)&b
->head
.first
);
125 /* Statistics gathering. */
126 struct dentry_stat_t dentry_stat
= {
130 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
132 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
133 static int get_nr_dentry(void)
137 for_each_possible_cpu(i
)
138 sum
+= per_cpu(nr_dentry
, i
);
139 return sum
< 0 ? 0 : sum
;
142 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
143 size_t *lenp
, loff_t
*ppos
)
145 dentry_stat
.nr_dentry
= get_nr_dentry();
146 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
150 static void __d_free(struct rcu_head
*head
)
152 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
154 WARN_ON(!list_empty(&dentry
->d_alias
));
155 if (dname_external(dentry
))
156 kfree(dentry
->d_name
.name
);
157 kmem_cache_free(dentry_cache
, dentry
);
163 static void d_free(struct dentry
*dentry
)
165 BUG_ON(dentry
->d_count
);
166 this_cpu_dec(nr_dentry
);
167 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
168 dentry
->d_op
->d_release(dentry
);
170 /* if dentry was never inserted into hash, immediate free is OK */
171 if (hlist_bl_unhashed(&dentry
->d_hash
))
172 __d_free(&dentry
->d_u
.d_rcu
);
174 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
178 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
179 * After this call, in-progress rcu-walk path lookup will fail. This
180 * should be called after unhashing, and after changing d_inode (if
181 * the dentry has not already been unhashed).
183 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
185 assert_spin_locked(&dentry
->d_lock
);
186 /* Go through a barrier */
187 write_seqcount_barrier(&dentry
->d_seq
);
191 * Release the dentry's inode, using the filesystem
192 * d_iput() operation if defined. Dentry has no refcount
195 static void dentry_iput(struct dentry
* dentry
)
196 __releases(dentry
->d_lock
)
197 __releases(dentry
->d_inode
->i_lock
)
199 struct inode
*inode
= dentry
->d_inode
;
201 dentry
->d_inode
= NULL
;
202 list_del_init(&dentry
->d_alias
);
203 spin_unlock(&dentry
->d_lock
);
204 spin_unlock(&inode
->i_lock
);
206 fsnotify_inoderemove(inode
);
207 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
208 dentry
->d_op
->d_iput(dentry
, inode
);
212 spin_unlock(&dentry
->d_lock
);
217 * Release the dentry's inode, using the filesystem
218 * d_iput() operation if defined. dentry remains in-use.
220 static void dentry_unlink_inode(struct dentry
* dentry
)
221 __releases(dentry
->d_lock
)
222 __releases(dentry
->d_inode
->i_lock
)
224 struct inode
*inode
= dentry
->d_inode
;
225 dentry
->d_inode
= NULL
;
226 list_del_init(&dentry
->d_alias
);
227 dentry_rcuwalk_barrier(dentry
);
228 spin_unlock(&dentry
->d_lock
);
229 spin_unlock(&inode
->i_lock
);
231 fsnotify_inoderemove(inode
);
232 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
233 dentry
->d_op
->d_iput(dentry
, inode
);
239 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
241 static void dentry_lru_add(struct dentry
*dentry
)
243 if (list_empty(&dentry
->d_lru
)) {
244 spin_lock(&dcache_lru_lock
);
245 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
246 dentry
->d_sb
->s_nr_dentry_unused
++;
247 dentry_stat
.nr_unused
++;
248 spin_unlock(&dcache_lru_lock
);
252 static void __dentry_lru_del(struct dentry
*dentry
)
254 list_del_init(&dentry
->d_lru
);
255 dentry
->d_sb
->s_nr_dentry_unused
--;
256 dentry_stat
.nr_unused
--;
259 static void dentry_lru_del(struct dentry
*dentry
)
261 if (!list_empty(&dentry
->d_lru
)) {
262 spin_lock(&dcache_lru_lock
);
263 __dentry_lru_del(dentry
);
264 spin_unlock(&dcache_lru_lock
);
268 static void dentry_lru_move_tail(struct dentry
*dentry
)
270 spin_lock(&dcache_lru_lock
);
271 if (list_empty(&dentry
->d_lru
)) {
272 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
273 dentry
->d_sb
->s_nr_dentry_unused
++;
274 dentry_stat
.nr_unused
++;
276 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
278 spin_unlock(&dcache_lru_lock
);
282 * d_kill - kill dentry and return parent
283 * @dentry: dentry to kill
285 * The dentry must already be unhashed and removed from the LRU.
287 * If this is the root of the dentry tree, return NULL.
289 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
292 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
293 __releases(dentry
->d_lock
)
294 __releases(parent
->d_lock
)
295 __releases(dentry
->d_inode
->i_lock
)
297 dentry
->d_parent
= NULL
;
298 list_del(&dentry
->d_u
.d_child
);
300 spin_unlock(&parent
->d_lock
);
303 * dentry_iput drops the locks, at which point nobody (except
304 * transient RCU lookups) can reach this dentry.
311 * d_drop - drop a dentry
312 * @dentry: dentry to drop
314 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
315 * be found through a VFS lookup any more. Note that this is different from
316 * deleting the dentry - d_delete will try to mark the dentry negative if
317 * possible, giving a successful _negative_ lookup, while d_drop will
318 * just make the cache lookup fail.
320 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
321 * reason (NFS timeouts or autofs deletes).
323 * __d_drop requires dentry->d_lock.
325 void __d_drop(struct dentry
*dentry
)
327 if (!(dentry
->d_flags
& DCACHE_UNHASHED
)) {
328 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
)) {
330 (unsigned long *)&dentry
->d_sb
->s_anon
.first
);
331 dentry
->d_flags
|= DCACHE_UNHASHED
;
332 hlist_bl_del_init(&dentry
->d_hash
);
334 (unsigned long *)&dentry
->d_sb
->s_anon
.first
);
336 struct dcache_hash_bucket
*b
;
337 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
340 * We may not actually need to put DCACHE_UNHASHED
341 * manipulations under the hash lock, but follow
342 * the principle of least surprise.
344 dentry
->d_flags
|= DCACHE_UNHASHED
;
345 hlist_bl_del_rcu(&dentry
->d_hash
);
346 spin_unlock_bucket(b
);
347 dentry_rcuwalk_barrier(dentry
);
351 EXPORT_SYMBOL(__d_drop
);
353 void d_drop(struct dentry
*dentry
)
355 spin_lock(&dentry
->d_lock
);
357 spin_unlock(&dentry
->d_lock
);
359 EXPORT_SYMBOL(d_drop
);
362 * Finish off a dentry we've decided to kill.
363 * dentry->d_lock must be held, returns with it unlocked.
364 * If ref is non-zero, then decrement the refcount too.
365 * Returns dentry requiring refcount drop, or NULL if we're done.
367 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
368 __releases(dentry
->d_lock
)
371 struct dentry
*parent
;
373 inode
= dentry
->d_inode
;
374 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
376 spin_unlock(&dentry
->d_lock
);
378 return dentry
; /* try again with same dentry */
383 parent
= dentry
->d_parent
;
384 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
386 spin_unlock(&inode
->i_lock
);
392 /* if dentry was on the d_lru list delete it from there */
393 dentry_lru_del(dentry
);
394 /* if it was on the hash then remove it */
396 return d_kill(dentry
, parent
);
402 * This is complicated by the fact that we do not want to put
403 * dentries that are no longer on any hash chain on the unused
404 * list: we'd much rather just get rid of them immediately.
406 * However, that implies that we have to traverse the dentry
407 * tree upwards to the parents which might _also_ now be
408 * scheduled for deletion (it may have been only waiting for
409 * its last child to go away).
411 * This tail recursion is done by hand as we don't want to depend
412 * on the compiler to always get this right (gcc generally doesn't).
413 * Real recursion would eat up our stack space.
417 * dput - release a dentry
418 * @dentry: dentry to release
420 * Release a dentry. This will drop the usage count and if appropriate
421 * call the dentry unlink method as well as removing it from the queues and
422 * releasing its resources. If the parent dentries were scheduled for release
423 * they too may now get deleted.
425 void dput(struct dentry
*dentry
)
431 if (dentry
->d_count
== 1)
433 spin_lock(&dentry
->d_lock
);
434 BUG_ON(!dentry
->d_count
);
435 if (dentry
->d_count
> 1) {
437 spin_unlock(&dentry
->d_lock
);
441 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
442 if (dentry
->d_op
->d_delete(dentry
))
446 /* Unreachable? Get rid of it */
447 if (d_unhashed(dentry
))
450 /* Otherwise leave it cached and ensure it's on the LRU */
451 dentry
->d_flags
|= DCACHE_REFERENCED
;
452 dentry_lru_add(dentry
);
455 spin_unlock(&dentry
->d_lock
);
459 dentry
= dentry_kill(dentry
, 1);
466 * d_invalidate - invalidate a dentry
467 * @dentry: dentry to invalidate
469 * Try to invalidate the dentry if it turns out to be
470 * possible. If there are other dentries that can be
471 * reached through this one we can't delete it and we
472 * return -EBUSY. On success we return 0.
477 int d_invalidate(struct dentry
* dentry
)
480 * If it's already been dropped, return OK.
482 spin_lock(&dentry
->d_lock
);
483 if (d_unhashed(dentry
)) {
484 spin_unlock(&dentry
->d_lock
);
488 * Check whether to do a partial shrink_dcache
489 * to get rid of unused child entries.
491 if (!list_empty(&dentry
->d_subdirs
)) {
492 spin_unlock(&dentry
->d_lock
);
493 shrink_dcache_parent(dentry
);
494 spin_lock(&dentry
->d_lock
);
498 * Somebody else still using it?
500 * If it's a directory, we can't drop it
501 * for fear of somebody re-populating it
502 * with children (even though dropping it
503 * would make it unreachable from the root,
504 * we might still populate it if it was a
505 * working directory or similar).
507 if (dentry
->d_count
> 1) {
508 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
509 spin_unlock(&dentry
->d_lock
);
515 spin_unlock(&dentry
->d_lock
);
518 EXPORT_SYMBOL(d_invalidate
);
520 /* This must be called with d_lock held */
521 static inline void __dget_dlock(struct dentry
*dentry
)
526 static inline void __dget(struct dentry
*dentry
)
528 spin_lock(&dentry
->d_lock
);
529 __dget_dlock(dentry
);
530 spin_unlock(&dentry
->d_lock
);
533 struct dentry
*dget_parent(struct dentry
*dentry
)
539 * Don't need rcu_dereference because we re-check it was correct under
543 ret
= dentry
->d_parent
;
548 spin_lock(&ret
->d_lock
);
549 if (unlikely(ret
!= dentry
->d_parent
)) {
550 spin_unlock(&ret
->d_lock
);
555 BUG_ON(!ret
->d_count
);
557 spin_unlock(&ret
->d_lock
);
561 EXPORT_SYMBOL(dget_parent
);
564 * d_find_alias - grab a hashed alias of inode
565 * @inode: inode in question
566 * @want_discon: flag, used by d_splice_alias, to request
567 * that only a DISCONNECTED alias be returned.
569 * If inode has a hashed alias, or is a directory and has any alias,
570 * acquire the reference to alias and return it. Otherwise return NULL.
571 * Notice that if inode is a directory there can be only one alias and
572 * it can be unhashed only if it has no children, or if it is the root
575 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
576 * any other hashed alias over that one unless @want_discon is set,
577 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
579 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
581 struct dentry
*alias
, *discon_alias
;
585 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
586 spin_lock(&alias
->d_lock
);
587 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
588 if (IS_ROOT(alias
) &&
589 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
590 discon_alias
= alias
;
591 } else if (!want_discon
) {
593 spin_unlock(&alias
->d_lock
);
597 spin_unlock(&alias
->d_lock
);
600 alias
= discon_alias
;
601 spin_lock(&alias
->d_lock
);
602 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
603 if (IS_ROOT(alias
) &&
604 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
606 spin_unlock(&alias
->d_lock
);
610 spin_unlock(&alias
->d_lock
);
616 struct dentry
*d_find_alias(struct inode
*inode
)
618 struct dentry
*de
= NULL
;
620 if (!list_empty(&inode
->i_dentry
)) {
621 spin_lock(&inode
->i_lock
);
622 de
= __d_find_alias(inode
, 0);
623 spin_unlock(&inode
->i_lock
);
627 EXPORT_SYMBOL(d_find_alias
);
630 * Try to kill dentries associated with this inode.
631 * WARNING: you must own a reference to inode.
633 void d_prune_aliases(struct inode
*inode
)
635 struct dentry
*dentry
;
637 spin_lock(&inode
->i_lock
);
638 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
639 spin_lock(&dentry
->d_lock
);
640 if (!dentry
->d_count
) {
641 __dget_dlock(dentry
);
643 spin_unlock(&dentry
->d_lock
);
644 spin_unlock(&inode
->i_lock
);
648 spin_unlock(&dentry
->d_lock
);
650 spin_unlock(&inode
->i_lock
);
652 EXPORT_SYMBOL(d_prune_aliases
);
655 * Try to throw away a dentry - free the inode, dput the parent.
656 * Requires dentry->d_lock is held, and dentry->d_count == 0.
657 * Releases dentry->d_lock.
659 * This may fail if locks cannot be acquired no problem, just try again.
661 static void try_prune_one_dentry(struct dentry
*dentry
)
662 __releases(dentry
->d_lock
)
664 struct dentry
*parent
;
666 parent
= dentry_kill(dentry
, 0);
668 * If dentry_kill returns NULL, we have nothing more to do.
669 * if it returns the same dentry, trylocks failed. In either
670 * case, just loop again.
672 * Otherwise, we need to prune ancestors too. This is necessary
673 * to prevent quadratic behavior of shrink_dcache_parent(), but
674 * is also expected to be beneficial in reducing dentry cache
679 if (parent
== dentry
)
682 /* Prune ancestors. */
685 spin_lock(&dentry
->d_lock
);
686 if (dentry
->d_count
> 1) {
688 spin_unlock(&dentry
->d_lock
);
691 dentry
= dentry_kill(dentry
, 1);
695 static void shrink_dentry_list(struct list_head
*list
)
697 struct dentry
*dentry
;
701 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
702 if (&dentry
->d_lru
== list
)
704 spin_lock(&dentry
->d_lock
);
705 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
706 spin_unlock(&dentry
->d_lock
);
711 * We found an inuse dentry which was not removed from
712 * the LRU because of laziness during lookup. Do not free
713 * it - just keep it off the LRU list.
715 if (dentry
->d_count
) {
716 dentry_lru_del(dentry
);
717 spin_unlock(&dentry
->d_lock
);
723 try_prune_one_dentry(dentry
);
731 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
732 * @sb: superblock to shrink dentry LRU.
733 * @count: number of entries to prune
734 * @flags: flags to control the dentry processing
736 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
738 static void __shrink_dcache_sb(struct super_block
*sb
, int *count
, int flags
)
740 /* called from prune_dcache() and shrink_dcache_parent() */
741 struct dentry
*dentry
;
742 LIST_HEAD(referenced
);
747 spin_lock(&dcache_lru_lock
);
748 while (!list_empty(&sb
->s_dentry_lru
)) {
749 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
750 struct dentry
, d_lru
);
751 BUG_ON(dentry
->d_sb
!= sb
);
753 if (!spin_trylock(&dentry
->d_lock
)) {
754 spin_unlock(&dcache_lru_lock
);
760 * If we are honouring the DCACHE_REFERENCED flag and the
761 * dentry has this flag set, don't free it. Clear the flag
762 * and put it back on the LRU.
764 if (flags
& DCACHE_REFERENCED
&&
765 dentry
->d_flags
& DCACHE_REFERENCED
) {
766 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
767 list_move(&dentry
->d_lru
, &referenced
);
768 spin_unlock(&dentry
->d_lock
);
770 list_move_tail(&dentry
->d_lru
, &tmp
);
771 spin_unlock(&dentry
->d_lock
);
775 cond_resched_lock(&dcache_lru_lock
);
777 if (!list_empty(&referenced
))
778 list_splice(&referenced
, &sb
->s_dentry_lru
);
779 spin_unlock(&dcache_lru_lock
);
781 shrink_dentry_list(&tmp
);
787 * prune_dcache - shrink the dcache
788 * @count: number of entries to try to free
790 * Shrink the dcache. This is done when we need more memory, or simply when we
791 * need to unmount something (at which point we need to unuse all dentries).
793 * This function may fail to free any resources if all the dentries are in use.
795 static void prune_dcache(int count
)
797 struct super_block
*sb
, *p
= NULL
;
799 int unused
= dentry_stat
.nr_unused
;
803 if (unused
== 0 || count
== 0)
808 prune_ratio
= unused
/ count
;
810 list_for_each_entry(sb
, &super_blocks
, s_list
) {
811 if (list_empty(&sb
->s_instances
))
813 if (sb
->s_nr_dentry_unused
== 0)
816 /* Now, we reclaim unused dentrins with fairness.
817 * We reclaim them same percentage from each superblock.
818 * We calculate number of dentries to scan on this sb
819 * as follows, but the implementation is arranged to avoid
821 * number of dentries to scan on this sb =
822 * count * (number of dentries on this sb /
823 * number of dentries in the machine)
825 spin_unlock(&sb_lock
);
826 if (prune_ratio
!= 1)
827 w_count
= (sb
->s_nr_dentry_unused
/ prune_ratio
) + 1;
829 w_count
= sb
->s_nr_dentry_unused
;
832 * We need to be sure this filesystem isn't being unmounted,
833 * otherwise we could race with generic_shutdown_super(), and
834 * end up holding a reference to an inode while the filesystem
835 * is unmounted. So we try to get s_umount, and make sure
838 if (down_read_trylock(&sb
->s_umount
)) {
839 if ((sb
->s_root
!= NULL
) &&
840 (!list_empty(&sb
->s_dentry_lru
))) {
841 __shrink_dcache_sb(sb
, &w_count
,
845 up_read(&sb
->s_umount
);
852 /* more work left to do? */
858 spin_unlock(&sb_lock
);
862 * shrink_dcache_sb - shrink dcache for a superblock
865 * Shrink the dcache for the specified super block. This is used to free
866 * the dcache before unmounting a file system.
868 void shrink_dcache_sb(struct super_block
*sb
)
872 spin_lock(&dcache_lru_lock
);
873 while (!list_empty(&sb
->s_dentry_lru
)) {
874 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
875 spin_unlock(&dcache_lru_lock
);
876 shrink_dentry_list(&tmp
);
877 spin_lock(&dcache_lru_lock
);
879 spin_unlock(&dcache_lru_lock
);
881 EXPORT_SYMBOL(shrink_dcache_sb
);
884 * destroy a single subtree of dentries for unmount
885 * - see the comments on shrink_dcache_for_umount() for a description of the
888 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
890 struct dentry
*parent
;
891 unsigned detached
= 0;
893 BUG_ON(!IS_ROOT(dentry
));
895 /* detach this root from the system */
896 spin_lock(&dentry
->d_lock
);
897 dentry_lru_del(dentry
);
899 spin_unlock(&dentry
->d_lock
);
902 /* descend to the first leaf in the current subtree */
903 while (!list_empty(&dentry
->d_subdirs
)) {
906 /* this is a branch with children - detach all of them
907 * from the system in one go */
908 spin_lock(&dentry
->d_lock
);
909 list_for_each_entry(loop
, &dentry
->d_subdirs
,
911 spin_lock_nested(&loop
->d_lock
,
912 DENTRY_D_LOCK_NESTED
);
913 dentry_lru_del(loop
);
915 spin_unlock(&loop
->d_lock
);
917 spin_unlock(&dentry
->d_lock
);
919 /* move to the first child */
920 dentry
= list_entry(dentry
->d_subdirs
.next
,
921 struct dentry
, d_u
.d_child
);
924 /* consume the dentries from this leaf up through its parents
925 * until we find one with children or run out altogether */
929 if (dentry
->d_count
!= 0) {
931 "BUG: Dentry %p{i=%lx,n=%s}"
933 " [unmount of %s %s]\n",
936 dentry
->d_inode
->i_ino
: 0UL,
939 dentry
->d_sb
->s_type
->name
,
944 if (IS_ROOT(dentry
)) {
946 list_del(&dentry
->d_u
.d_child
);
948 parent
= dentry
->d_parent
;
949 spin_lock(&parent
->d_lock
);
951 list_del(&dentry
->d_u
.d_child
);
952 spin_unlock(&parent
->d_lock
);
957 inode
= dentry
->d_inode
;
959 dentry
->d_inode
= NULL
;
960 list_del_init(&dentry
->d_alias
);
961 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
962 dentry
->d_op
->d_iput(dentry
, inode
);
969 /* finished when we fall off the top of the tree,
970 * otherwise we ascend to the parent and move to the
971 * next sibling if there is one */
975 } while (list_empty(&dentry
->d_subdirs
));
977 dentry
= list_entry(dentry
->d_subdirs
.next
,
978 struct dentry
, d_u
.d_child
);
983 * destroy the dentries attached to a superblock on unmounting
984 * - we don't need to use dentry->d_lock because:
985 * - the superblock is detached from all mountings and open files, so the
986 * dentry trees will not be rearranged by the VFS
987 * - s_umount is write-locked, so the memory pressure shrinker will ignore
988 * any dentries belonging to this superblock that it comes across
989 * - the filesystem itself is no longer permitted to rearrange the dentries
992 void shrink_dcache_for_umount(struct super_block
*sb
)
994 struct dentry
*dentry
;
996 if (down_read_trylock(&sb
->s_umount
))
1001 spin_lock(&dentry
->d_lock
);
1003 spin_unlock(&dentry
->d_lock
);
1004 shrink_dcache_for_umount_subtree(dentry
);
1006 while (!hlist_bl_empty(&sb
->s_anon
)) {
1007 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
1008 shrink_dcache_for_umount_subtree(dentry
);
1013 * Search for at least 1 mount point in the dentry's subdirs.
1014 * We descend to the next level whenever the d_subdirs
1015 * list is non-empty and continue searching.
1019 * have_submounts - check for mounts over a dentry
1020 * @parent: dentry to check.
1022 * Return true if the parent or its subdirectories contain
1025 int have_submounts(struct dentry
*parent
)
1027 struct dentry
*this_parent
;
1028 struct list_head
*next
;
1032 seq
= read_seqbegin(&rename_lock
);
1034 this_parent
= parent
;
1036 if (d_mountpoint(parent
))
1038 spin_lock(&this_parent
->d_lock
);
1040 next
= this_parent
->d_subdirs
.next
;
1042 while (next
!= &this_parent
->d_subdirs
) {
1043 struct list_head
*tmp
= next
;
1044 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1047 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1048 /* Have we found a mount point ? */
1049 if (d_mountpoint(dentry
)) {
1050 spin_unlock(&dentry
->d_lock
);
1051 spin_unlock(&this_parent
->d_lock
);
1054 if (!list_empty(&dentry
->d_subdirs
)) {
1055 spin_unlock(&this_parent
->d_lock
);
1056 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1057 this_parent
= dentry
;
1058 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1061 spin_unlock(&dentry
->d_lock
);
1064 * All done at this level ... ascend and resume the search.
1066 if (this_parent
!= parent
) {
1068 struct dentry
*child
;
1070 tmp
= this_parent
->d_parent
;
1072 spin_unlock(&this_parent
->d_lock
);
1073 child
= this_parent
;
1075 spin_lock(&this_parent
->d_lock
);
1076 /* might go back up the wrong parent if we have had a rename
1078 if (this_parent
!= child
->d_parent
||
1079 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1080 spin_unlock(&this_parent
->d_lock
);
1085 next
= child
->d_u
.d_child
.next
;
1088 spin_unlock(&this_parent
->d_lock
);
1089 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1092 write_sequnlock(&rename_lock
);
1093 return 0; /* No mount points found in tree */
1095 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1098 write_sequnlock(&rename_lock
);
1103 write_seqlock(&rename_lock
);
1106 EXPORT_SYMBOL(have_submounts
);
1109 * Search the dentry child list for the specified parent,
1110 * and move any unused dentries to the end of the unused
1111 * list for prune_dcache(). We descend to the next level
1112 * whenever the d_subdirs list is non-empty and continue
1115 * It returns zero iff there are no unused children,
1116 * otherwise it returns the number of children moved to
1117 * the end of the unused list. This may not be the total
1118 * number of unused children, because select_parent can
1119 * drop the lock and return early due to latency
1122 static int select_parent(struct dentry
* parent
)
1124 struct dentry
*this_parent
;
1125 struct list_head
*next
;
1130 seq
= read_seqbegin(&rename_lock
);
1132 this_parent
= parent
;
1133 spin_lock(&this_parent
->d_lock
);
1135 next
= this_parent
->d_subdirs
.next
;
1137 while (next
!= &this_parent
->d_subdirs
) {
1138 struct list_head
*tmp
= next
;
1139 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1142 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1145 * move only zero ref count dentries to the end
1146 * of the unused list for prune_dcache
1148 if (!dentry
->d_count
) {
1149 dentry_lru_move_tail(dentry
);
1152 dentry_lru_del(dentry
);
1156 * We can return to the caller if we have found some (this
1157 * ensures forward progress). We'll be coming back to find
1160 if (found
&& need_resched()) {
1161 spin_unlock(&dentry
->d_lock
);
1166 * Descend a level if the d_subdirs list is non-empty.
1168 if (!list_empty(&dentry
->d_subdirs
)) {
1169 spin_unlock(&this_parent
->d_lock
);
1170 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1171 this_parent
= dentry
;
1172 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1176 spin_unlock(&dentry
->d_lock
);
1179 * All done at this level ... ascend and resume the search.
1181 if (this_parent
!= parent
) {
1183 struct dentry
*child
;
1185 tmp
= this_parent
->d_parent
;
1187 spin_unlock(&this_parent
->d_lock
);
1188 child
= this_parent
;
1190 spin_lock(&this_parent
->d_lock
);
1191 /* might go back up the wrong parent if we have had a rename
1193 if (this_parent
!= child
->d_parent
||
1194 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1195 spin_unlock(&this_parent
->d_lock
);
1200 next
= child
->d_u
.d_child
.next
;
1204 spin_unlock(&this_parent
->d_lock
);
1205 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1208 write_sequnlock(&rename_lock
);
1215 write_seqlock(&rename_lock
);
1220 * shrink_dcache_parent - prune dcache
1221 * @parent: parent of entries to prune
1223 * Prune the dcache to remove unused children of the parent dentry.
1226 void shrink_dcache_parent(struct dentry
* parent
)
1228 struct super_block
*sb
= parent
->d_sb
;
1231 while ((found
= select_parent(parent
)) != 0)
1232 __shrink_dcache_sb(sb
, &found
, 0);
1234 EXPORT_SYMBOL(shrink_dcache_parent
);
1237 * Scan `nr' dentries and return the number which remain.
1239 * We need to avoid reentering the filesystem if the caller is performing a
1240 * GFP_NOFS allocation attempt. One example deadlock is:
1242 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1243 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1244 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1246 * In this case we return -1 to tell the caller that we baled.
1248 static int shrink_dcache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
1251 if (!(gfp_mask
& __GFP_FS
))
1256 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
1259 static struct shrinker dcache_shrinker
= {
1260 .shrink
= shrink_dcache_memory
,
1261 .seeks
= DEFAULT_SEEKS
,
1265 * d_alloc - allocate a dcache entry
1266 * @parent: parent of entry to allocate
1267 * @name: qstr of the name
1269 * Allocates a dentry. It returns %NULL if there is insufficient memory
1270 * available. On a success the dentry is returned. The name passed in is
1271 * copied and the copy passed in may be reused after this call.
1274 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1276 struct dentry
*dentry
;
1279 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1283 if (name
->len
> DNAME_INLINE_LEN
-1) {
1284 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1286 kmem_cache_free(dentry_cache
, dentry
);
1290 dname
= dentry
->d_iname
;
1292 dentry
->d_name
.name
= dname
;
1294 dentry
->d_name
.len
= name
->len
;
1295 dentry
->d_name
.hash
= name
->hash
;
1296 memcpy(dname
, name
->name
, name
->len
);
1297 dname
[name
->len
] = 0;
1299 dentry
->d_count
= 1;
1300 dentry
->d_flags
= DCACHE_UNHASHED
;
1301 spin_lock_init(&dentry
->d_lock
);
1302 seqcount_init(&dentry
->d_seq
);
1303 dentry
->d_inode
= NULL
;
1304 dentry
->d_parent
= NULL
;
1305 dentry
->d_sb
= NULL
;
1306 dentry
->d_op
= NULL
;
1307 dentry
->d_fsdata
= NULL
;
1308 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1309 INIT_LIST_HEAD(&dentry
->d_lru
);
1310 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1311 INIT_LIST_HEAD(&dentry
->d_alias
);
1312 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1315 spin_lock(&parent
->d_lock
);
1317 * don't need child lock because it is not subject
1318 * to concurrency here
1320 __dget_dlock(parent
);
1321 dentry
->d_parent
= parent
;
1322 dentry
->d_sb
= parent
->d_sb
;
1323 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1324 spin_unlock(&parent
->d_lock
);
1327 this_cpu_inc(nr_dentry
);
1331 EXPORT_SYMBOL(d_alloc
);
1333 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1338 q
.len
= strlen(name
);
1339 q
.hash
= full_name_hash(q
.name
, q
.len
);
1340 return d_alloc(parent
, &q
);
1342 EXPORT_SYMBOL(d_alloc_name
);
1344 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1346 BUG_ON(dentry
->d_op
);
1347 BUG_ON(dentry
->d_flags
& (DCACHE_OP_HASH
|
1349 DCACHE_OP_REVALIDATE
|
1350 DCACHE_OP_DELETE
));
1355 dentry
->d_flags
|= DCACHE_OP_HASH
;
1357 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1358 if (op
->d_revalidate
)
1359 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1361 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1364 EXPORT_SYMBOL(d_set_d_op
);
1366 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1368 spin_lock(&dentry
->d_lock
);
1370 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1371 dentry
->d_inode
= inode
;
1372 dentry_rcuwalk_barrier(dentry
);
1373 spin_unlock(&dentry
->d_lock
);
1374 fsnotify_d_instantiate(dentry
, inode
);
1378 * d_instantiate - fill in inode information for a dentry
1379 * @entry: dentry to complete
1380 * @inode: inode to attach to this dentry
1382 * Fill in inode information in the entry.
1384 * This turns negative dentries into productive full members
1387 * NOTE! This assumes that the inode count has been incremented
1388 * (or otherwise set) by the caller to indicate that it is now
1389 * in use by the dcache.
1392 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1394 BUG_ON(!list_empty(&entry
->d_alias
));
1396 spin_lock(&inode
->i_lock
);
1397 __d_instantiate(entry
, inode
);
1399 spin_unlock(&inode
->i_lock
);
1400 security_d_instantiate(entry
, inode
);
1402 EXPORT_SYMBOL(d_instantiate
);
1405 * d_instantiate_unique - instantiate a non-aliased dentry
1406 * @entry: dentry to instantiate
1407 * @inode: inode to attach to this dentry
1409 * Fill in inode information in the entry. On success, it returns NULL.
1410 * If an unhashed alias of "entry" already exists, then we return the
1411 * aliased dentry instead and drop one reference to inode.
1413 * Note that in order to avoid conflicts with rename() etc, the caller
1414 * had better be holding the parent directory semaphore.
1416 * This also assumes that the inode count has been incremented
1417 * (or otherwise set) by the caller to indicate that it is now
1418 * in use by the dcache.
1420 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1421 struct inode
*inode
)
1423 struct dentry
*alias
;
1424 int len
= entry
->d_name
.len
;
1425 const char *name
= entry
->d_name
.name
;
1426 unsigned int hash
= entry
->d_name
.hash
;
1429 __d_instantiate(entry
, NULL
);
1433 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1434 struct qstr
*qstr
= &alias
->d_name
;
1437 * Don't need alias->d_lock here, because aliases with
1438 * d_parent == entry->d_parent are not subject to name or
1439 * parent changes, because the parent inode i_mutex is held.
1441 if (qstr
->hash
!= hash
)
1443 if (alias
->d_parent
!= entry
->d_parent
)
1445 if (qstr
->len
!= len
)
1447 if (memcmp(qstr
->name
, name
, len
))
1453 __d_instantiate(entry
, inode
);
1457 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1459 struct dentry
*result
;
1461 BUG_ON(!list_empty(&entry
->d_alias
));
1464 spin_lock(&inode
->i_lock
);
1465 result
= __d_instantiate_unique(entry
, inode
);
1467 spin_unlock(&inode
->i_lock
);
1470 security_d_instantiate(entry
, inode
);
1474 BUG_ON(!d_unhashed(result
));
1479 EXPORT_SYMBOL(d_instantiate_unique
);
1482 * d_alloc_root - allocate root dentry
1483 * @root_inode: inode to allocate the root for
1485 * Allocate a root ("/") dentry for the inode given. The inode is
1486 * instantiated and returned. %NULL is returned if there is insufficient
1487 * memory or the inode passed is %NULL.
1490 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1492 struct dentry
*res
= NULL
;
1495 static const struct qstr name
= { .name
= "/", .len
= 1 };
1497 res
= d_alloc(NULL
, &name
);
1499 res
->d_sb
= root_inode
->i_sb
;
1500 res
->d_parent
= res
;
1501 d_instantiate(res
, root_inode
);
1506 EXPORT_SYMBOL(d_alloc_root
);
1509 * d_obtain_alias - find or allocate a dentry for a given inode
1510 * @inode: inode to allocate the dentry for
1512 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1513 * similar open by handle operations. The returned dentry may be anonymous,
1514 * or may have a full name (if the inode was already in the cache).
1516 * When called on a directory inode, we must ensure that the inode only ever
1517 * has one dentry. If a dentry is found, that is returned instead of
1518 * allocating a new one.
1520 * On successful return, the reference to the inode has been transferred
1521 * to the dentry. In case of an error the reference on the inode is released.
1522 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1523 * be passed in and will be the error will be propagate to the return value,
1524 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1526 struct dentry
*d_obtain_alias(struct inode
*inode
)
1528 static const struct qstr anonstring
= { .name
= "" };
1533 return ERR_PTR(-ESTALE
);
1535 return ERR_CAST(inode
);
1537 res
= d_find_alias(inode
);
1541 tmp
= d_alloc(NULL
, &anonstring
);
1543 res
= ERR_PTR(-ENOMEM
);
1546 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1549 spin_lock(&inode
->i_lock
);
1550 res
= __d_find_alias(inode
, 0);
1552 spin_unlock(&inode
->i_lock
);
1557 /* attach a disconnected dentry */
1558 spin_lock(&tmp
->d_lock
);
1559 tmp
->d_sb
= inode
->i_sb
;
1560 tmp
->d_inode
= inode
;
1561 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1562 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1563 bit_spin_lock(0, (unsigned long *)&tmp
->d_sb
->s_anon
.first
);
1564 tmp
->d_flags
&= ~DCACHE_UNHASHED
;
1565 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1566 __bit_spin_unlock(0, (unsigned long *)&tmp
->d_sb
->s_anon
.first
);
1567 spin_unlock(&tmp
->d_lock
);
1568 spin_unlock(&inode
->i_lock
);
1576 EXPORT_SYMBOL(d_obtain_alias
);
1579 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1580 * @inode: the inode which may have a disconnected dentry
1581 * @dentry: a negative dentry which we want to point to the inode.
1583 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1584 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1585 * and return it, else simply d_add the inode to the dentry and return NULL.
1587 * This is needed in the lookup routine of any filesystem that is exportable
1588 * (via knfsd) so that we can build dcache paths to directories effectively.
1590 * If a dentry was found and moved, then it is returned. Otherwise NULL
1591 * is returned. This matches the expected return value of ->lookup.
1594 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1596 struct dentry
*new = NULL
;
1598 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1599 spin_lock(&inode
->i_lock
);
1600 new = __d_find_alias(inode
, 1);
1602 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1603 spin_unlock(&inode
->i_lock
);
1604 security_d_instantiate(new, inode
);
1605 d_move(new, dentry
);
1608 /* already taking inode->i_lock, so d_add() by hand */
1609 __d_instantiate(dentry
, inode
);
1610 spin_unlock(&inode
->i_lock
);
1611 security_d_instantiate(dentry
, inode
);
1615 d_add(dentry
, inode
);
1618 EXPORT_SYMBOL(d_splice_alias
);
1621 * d_add_ci - lookup or allocate new dentry with case-exact name
1622 * @inode: the inode case-insensitive lookup has found
1623 * @dentry: the negative dentry that was passed to the parent's lookup func
1624 * @name: the case-exact name to be associated with the returned dentry
1626 * This is to avoid filling the dcache with case-insensitive names to the
1627 * same inode, only the actual correct case is stored in the dcache for
1628 * case-insensitive filesystems.
1630 * For a case-insensitive lookup match and if the the case-exact dentry
1631 * already exists in in the dcache, use it and return it.
1633 * If no entry exists with the exact case name, allocate new dentry with
1634 * the exact case, and return the spliced entry.
1636 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1640 struct dentry
*found
;
1644 * First check if a dentry matching the name already exists,
1645 * if not go ahead and create it now.
1647 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1649 new = d_alloc(dentry
->d_parent
, name
);
1655 found
= d_splice_alias(inode
, new);
1664 * If a matching dentry exists, and it's not negative use it.
1666 * Decrement the reference count to balance the iget() done
1669 if (found
->d_inode
) {
1670 if (unlikely(found
->d_inode
!= inode
)) {
1671 /* This can't happen because bad inodes are unhashed. */
1672 BUG_ON(!is_bad_inode(inode
));
1673 BUG_ON(!is_bad_inode(found
->d_inode
));
1680 * Negative dentry: instantiate it unless the inode is a directory and
1681 * already has a dentry.
1683 spin_lock(&inode
->i_lock
);
1684 if (!S_ISDIR(inode
->i_mode
) || list_empty(&inode
->i_dentry
)) {
1685 __d_instantiate(found
, inode
);
1686 spin_unlock(&inode
->i_lock
);
1687 security_d_instantiate(found
, inode
);
1692 * In case a directory already has a (disconnected) entry grab a
1693 * reference to it, move it in place and use it.
1695 new = list_entry(inode
->i_dentry
.next
, struct dentry
, d_alias
);
1697 spin_unlock(&inode
->i_lock
);
1698 security_d_instantiate(found
, inode
);
1706 return ERR_PTR(error
);
1708 EXPORT_SYMBOL(d_add_ci
);
1711 * __d_lookup_rcu - search for a dentry (racy, store-free)
1712 * @parent: parent dentry
1713 * @name: qstr of name we wish to find
1714 * @seq: returns d_seq value at the point where the dentry was found
1715 * @inode: returns dentry->d_inode when the inode was found valid.
1716 * Returns: dentry, or NULL
1718 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1719 * resolution (store-free path walking) design described in
1720 * Documentation/filesystems/path-lookup.txt.
1722 * This is not to be used outside core vfs.
1724 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1725 * held, and rcu_read_lock held. The returned dentry must not be stored into
1726 * without taking d_lock and checking d_seq sequence count against @seq
1729 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1732 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1733 * the returned dentry, so long as its parent's seqlock is checked after the
1734 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1735 * is formed, giving integrity down the path walk.
1737 struct dentry
*__d_lookup_rcu(struct dentry
*parent
, struct qstr
*name
,
1738 unsigned *seq
, struct inode
**inode
)
1740 unsigned int len
= name
->len
;
1741 unsigned int hash
= name
->hash
;
1742 const unsigned char *str
= name
->name
;
1743 struct dcache_hash_bucket
*b
= d_hash(parent
, hash
);
1744 struct hlist_bl_node
*node
;
1745 struct dentry
*dentry
;
1748 * Note: There is significant duplication with __d_lookup_rcu which is
1749 * required to prevent single threaded performance regressions
1750 * especially on architectures where smp_rmb (in seqcounts) are costly.
1751 * Keep the two functions in sync.
1755 * The hash list is protected using RCU.
1757 * Carefully use d_seq when comparing a candidate dentry, to avoid
1758 * races with d_move().
1760 * It is possible that concurrent renames can mess up our list
1761 * walk here and result in missing our dentry, resulting in the
1762 * false-negative result. d_lookup() protects against concurrent
1763 * renames using rename_lock seqlock.
1765 * See Documentation/vfs/dcache-locking.txt for more details.
1767 hlist_bl_for_each_entry_rcu(dentry
, node
, &b
->head
, d_hash
) {
1772 if (dentry
->d_name
.hash
!= hash
)
1776 *seq
= read_seqcount_begin(&dentry
->d_seq
);
1777 if (dentry
->d_parent
!= parent
)
1779 if (d_unhashed(dentry
))
1781 tlen
= dentry
->d_name
.len
;
1782 tname
= dentry
->d_name
.name
;
1783 i
= dentry
->d_inode
;
1785 * This seqcount check is required to ensure name and
1786 * len are loaded atomically, so as not to walk off the
1787 * edge of memory when walking. If we could load this
1788 * atomically some other way, we could drop this check.
1790 if (read_seqcount_retry(&dentry
->d_seq
, *seq
))
1792 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1793 if (parent
->d_op
->d_compare(parent
, *inode
,
1800 if (memcmp(tname
, str
, tlen
))
1804 * No extra seqcount check is required after the name
1805 * compare. The caller must perform a seqcount check in
1806 * order to do anything useful with the returned dentry
1816 * d_lookup - search for a dentry
1817 * @parent: parent dentry
1818 * @name: qstr of name we wish to find
1819 * Returns: dentry, or NULL
1821 * d_lookup searches the children of the parent dentry for the name in
1822 * question. If the dentry is found its reference count is incremented and the
1823 * dentry is returned. The caller must use dput to free the entry when it has
1824 * finished using it. %NULL is returned if the dentry does not exist.
1826 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1828 struct dentry
*dentry
;
1832 seq
= read_seqbegin(&rename_lock
);
1833 dentry
= __d_lookup(parent
, name
);
1836 } while (read_seqretry(&rename_lock
, seq
));
1839 EXPORT_SYMBOL(d_lookup
);
1842 * __d_lookup - search for a dentry (racy)
1843 * @parent: parent dentry
1844 * @name: qstr of name we wish to find
1845 * Returns: dentry, or NULL
1847 * __d_lookup is like d_lookup, however it may (rarely) return a
1848 * false-negative result due to unrelated rename activity.
1850 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1851 * however it must be used carefully, eg. with a following d_lookup in
1852 * the case of failure.
1854 * __d_lookup callers must be commented.
1856 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1858 unsigned int len
= name
->len
;
1859 unsigned int hash
= name
->hash
;
1860 const unsigned char *str
= name
->name
;
1861 struct dcache_hash_bucket
*b
= d_hash(parent
, hash
);
1862 struct hlist_bl_node
*node
;
1863 struct dentry
*found
= NULL
;
1864 struct dentry
*dentry
;
1867 * Note: There is significant duplication with __d_lookup_rcu which is
1868 * required to prevent single threaded performance regressions
1869 * especially on architectures where smp_rmb (in seqcounts) are costly.
1870 * Keep the two functions in sync.
1874 * The hash list is protected using RCU.
1876 * Take d_lock when comparing a candidate dentry, to avoid races
1879 * It is possible that concurrent renames can mess up our list
1880 * walk here and result in missing our dentry, resulting in the
1881 * false-negative result. d_lookup() protects against concurrent
1882 * renames using rename_lock seqlock.
1884 * See Documentation/vfs/dcache-locking.txt for more details.
1888 hlist_bl_for_each_entry_rcu(dentry
, node
, &b
->head
, d_hash
) {
1892 if (dentry
->d_name
.hash
!= hash
)
1895 spin_lock(&dentry
->d_lock
);
1896 if (dentry
->d_parent
!= parent
)
1898 if (d_unhashed(dentry
))
1902 * It is safe to compare names since d_move() cannot
1903 * change the qstr (protected by d_lock).
1905 tlen
= dentry
->d_name
.len
;
1906 tname
= dentry
->d_name
.name
;
1907 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1908 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1909 dentry
, dentry
->d_inode
,
1915 if (memcmp(tname
, str
, tlen
))
1921 spin_unlock(&dentry
->d_lock
);
1924 spin_unlock(&dentry
->d_lock
);
1932 * d_hash_and_lookup - hash the qstr then search for a dentry
1933 * @dir: Directory to search in
1934 * @name: qstr of name we wish to find
1936 * On hash failure or on lookup failure NULL is returned.
1938 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1940 struct dentry
*dentry
= NULL
;
1943 * Check for a fs-specific hash function. Note that we must
1944 * calculate the standard hash first, as the d_op->d_hash()
1945 * routine may choose to leave the hash value unchanged.
1947 name
->hash
= full_name_hash(name
->name
, name
->len
);
1948 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1949 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1952 dentry
= d_lookup(dir
, name
);
1958 * d_validate - verify dentry provided from insecure source (deprecated)
1959 * @dentry: The dentry alleged to be valid child of @dparent
1960 * @dparent: The parent dentry (known to be valid)
1962 * An insecure source has sent us a dentry, here we verify it and dget() it.
1963 * This is used by ncpfs in its readdir implementation.
1964 * Zero is returned in the dentry is invalid.
1966 * This function is slow for big directories, and deprecated, do not use it.
1968 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1970 struct dentry
*child
;
1972 spin_lock(&dparent
->d_lock
);
1973 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1974 if (dentry
== child
) {
1975 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1976 __dget_dlock(dentry
);
1977 spin_unlock(&dentry
->d_lock
);
1978 spin_unlock(&dparent
->d_lock
);
1982 spin_unlock(&dparent
->d_lock
);
1986 EXPORT_SYMBOL(d_validate
);
1989 * When a file is deleted, we have two options:
1990 * - turn this dentry into a negative dentry
1991 * - unhash this dentry and free it.
1993 * Usually, we want to just turn this into
1994 * a negative dentry, but if anybody else is
1995 * currently using the dentry or the inode
1996 * we can't do that and we fall back on removing
1997 * it from the hash queues and waiting for
1998 * it to be deleted later when it has no users
2002 * d_delete - delete a dentry
2003 * @dentry: The dentry to delete
2005 * Turn the dentry into a negative dentry if possible, otherwise
2006 * remove it from the hash queues so it can be deleted later
2009 void d_delete(struct dentry
* dentry
)
2011 struct inode
*inode
;
2014 * Are we the only user?
2017 spin_lock(&dentry
->d_lock
);
2018 inode
= dentry
->d_inode
;
2019 isdir
= S_ISDIR(inode
->i_mode
);
2020 if (dentry
->d_count
== 1) {
2021 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
2022 spin_unlock(&dentry
->d_lock
);
2026 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2027 dentry_unlink_inode(dentry
);
2028 fsnotify_nameremove(dentry
, isdir
);
2032 if (!d_unhashed(dentry
))
2035 spin_unlock(&dentry
->d_lock
);
2037 fsnotify_nameremove(dentry
, isdir
);
2039 EXPORT_SYMBOL(d_delete
);
2041 static void __d_rehash(struct dentry
* entry
, struct dcache_hash_bucket
*b
)
2043 BUG_ON(!d_unhashed(entry
));
2044 spin_lock_bucket(b
);
2045 entry
->d_flags
&= ~DCACHE_UNHASHED
;
2046 hlist_bl_add_head_rcu(&entry
->d_hash
, &b
->head
);
2047 spin_unlock_bucket(b
);
2050 static void _d_rehash(struct dentry
* entry
)
2052 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2056 * d_rehash - add an entry back to the hash
2057 * @entry: dentry to add to the hash
2059 * Adds a dentry to the hash according to its name.
2062 void d_rehash(struct dentry
* entry
)
2064 spin_lock(&entry
->d_lock
);
2066 spin_unlock(&entry
->d_lock
);
2068 EXPORT_SYMBOL(d_rehash
);
2071 * dentry_update_name_case - update case insensitive dentry with a new name
2072 * @dentry: dentry to be updated
2075 * Update a case insensitive dentry with new case of name.
2077 * dentry must have been returned by d_lookup with name @name. Old and new
2078 * name lengths must match (ie. no d_compare which allows mismatched name
2081 * Parent inode i_mutex must be held over d_lookup and into this call (to
2082 * keep renames and concurrent inserts, and readdir(2) away).
2084 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2086 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
2087 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2089 spin_lock(&dentry
->d_lock
);
2090 write_seqcount_begin(&dentry
->d_seq
);
2091 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2092 write_seqcount_end(&dentry
->d_seq
);
2093 spin_unlock(&dentry
->d_lock
);
2095 EXPORT_SYMBOL(dentry_update_name_case
);
2097 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2099 if (dname_external(target
)) {
2100 if (dname_external(dentry
)) {
2102 * Both external: swap the pointers
2104 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2107 * dentry:internal, target:external. Steal target's
2108 * storage and make target internal.
2110 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2111 dentry
->d_name
.len
+ 1);
2112 dentry
->d_name
.name
= target
->d_name
.name
;
2113 target
->d_name
.name
= target
->d_iname
;
2116 if (dname_external(dentry
)) {
2118 * dentry:external, target:internal. Give dentry's
2119 * storage to target and make dentry internal
2121 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2122 target
->d_name
.len
+ 1);
2123 target
->d_name
.name
= dentry
->d_name
.name
;
2124 dentry
->d_name
.name
= dentry
->d_iname
;
2127 * Both are internal. Just copy target to dentry
2129 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2130 target
->d_name
.len
+ 1);
2131 dentry
->d_name
.len
= target
->d_name
.len
;
2135 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2138 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2141 * XXXX: do we really need to take target->d_lock?
2143 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2144 spin_lock(&target
->d_parent
->d_lock
);
2146 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2147 spin_lock(&dentry
->d_parent
->d_lock
);
2148 spin_lock_nested(&target
->d_parent
->d_lock
,
2149 DENTRY_D_LOCK_NESTED
);
2151 spin_lock(&target
->d_parent
->d_lock
);
2152 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2153 DENTRY_D_LOCK_NESTED
);
2156 if (target
< dentry
) {
2157 spin_lock_nested(&target
->d_lock
, 2);
2158 spin_lock_nested(&dentry
->d_lock
, 3);
2160 spin_lock_nested(&dentry
->d_lock
, 2);
2161 spin_lock_nested(&target
->d_lock
, 3);
2165 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2166 struct dentry
*target
)
2168 if (target
->d_parent
!= dentry
->d_parent
)
2169 spin_unlock(&dentry
->d_parent
->d_lock
);
2170 if (target
->d_parent
!= target
)
2171 spin_unlock(&target
->d_parent
->d_lock
);
2175 * When switching names, the actual string doesn't strictly have to
2176 * be preserved in the target - because we're dropping the target
2177 * anyway. As such, we can just do a simple memcpy() to copy over
2178 * the new name before we switch.
2180 * Note that we have to be a lot more careful about getting the hash
2181 * switched - we have to switch the hash value properly even if it
2182 * then no longer matches the actual (corrupted) string of the target.
2183 * The hash value has to match the hash queue that the dentry is on..
2186 * d_move - move a dentry
2187 * @dentry: entry to move
2188 * @target: new dentry
2190 * Update the dcache to reflect the move of a file name. Negative
2191 * dcache entries should not be moved in this way.
2193 void d_move(struct dentry
* dentry
, struct dentry
* target
)
2195 if (!dentry
->d_inode
)
2196 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2198 BUG_ON(d_ancestor(dentry
, target
));
2199 BUG_ON(d_ancestor(target
, dentry
));
2201 write_seqlock(&rename_lock
);
2203 dentry_lock_for_move(dentry
, target
);
2205 write_seqcount_begin(&dentry
->d_seq
);
2206 write_seqcount_begin(&target
->d_seq
);
2208 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2211 * Move the dentry to the target hash queue. Don't bother checking
2212 * for the same hash queue because of how unlikely it is.
2215 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2217 /* Unhash the target: dput() will then get rid of it */
2220 list_del(&dentry
->d_u
.d_child
);
2221 list_del(&target
->d_u
.d_child
);
2223 /* Switch the names.. */
2224 switch_names(dentry
, target
);
2225 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2227 /* ... and switch the parents */
2228 if (IS_ROOT(dentry
)) {
2229 dentry
->d_parent
= target
->d_parent
;
2230 target
->d_parent
= target
;
2231 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2233 swap(dentry
->d_parent
, target
->d_parent
);
2235 /* And add them back to the (new) parent lists */
2236 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2239 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2241 write_seqcount_end(&target
->d_seq
);
2242 write_seqcount_end(&dentry
->d_seq
);
2244 dentry_unlock_parents_for_move(dentry
, target
);
2245 spin_unlock(&target
->d_lock
);
2246 fsnotify_d_move(dentry
);
2247 spin_unlock(&dentry
->d_lock
);
2248 write_sequnlock(&rename_lock
);
2250 EXPORT_SYMBOL(d_move
);
2253 * d_ancestor - search for an ancestor
2254 * @p1: ancestor dentry
2257 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2258 * an ancestor of p2, else NULL.
2260 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2264 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2265 if (p
->d_parent
== p1
)
2272 * This helper attempts to cope with remotely renamed directories
2274 * It assumes that the caller is already holding
2275 * dentry->d_parent->d_inode->i_mutex and the inode->i_lock
2277 * Note: If ever the locking in lock_rename() changes, then please
2278 * remember to update this too...
2280 static struct dentry
*__d_unalias(struct inode
*inode
,
2281 struct dentry
*dentry
, struct dentry
*alias
)
2283 struct mutex
*m1
= NULL
, *m2
= NULL
;
2286 /* If alias and dentry share a parent, then no extra locks required */
2287 if (alias
->d_parent
== dentry
->d_parent
)
2290 /* Check for loops */
2291 ret
= ERR_PTR(-ELOOP
);
2292 if (d_ancestor(alias
, dentry
))
2295 /* See lock_rename() */
2296 ret
= ERR_PTR(-EBUSY
);
2297 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2299 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2300 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2302 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2304 d_move(alias
, dentry
);
2307 spin_unlock(&inode
->i_lock
);
2316 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2317 * named dentry in place of the dentry to be replaced.
2318 * returns with anon->d_lock held!
2320 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2322 struct dentry
*dparent
, *aparent
;
2324 dentry_lock_for_move(anon
, dentry
);
2326 write_seqcount_begin(&dentry
->d_seq
);
2327 write_seqcount_begin(&anon
->d_seq
);
2329 dparent
= dentry
->d_parent
;
2330 aparent
= anon
->d_parent
;
2332 switch_names(dentry
, anon
);
2333 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2335 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2336 list_del(&dentry
->d_u
.d_child
);
2337 if (!IS_ROOT(dentry
))
2338 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2340 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2342 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2343 list_del(&anon
->d_u
.d_child
);
2345 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2347 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2349 write_seqcount_end(&dentry
->d_seq
);
2350 write_seqcount_end(&anon
->d_seq
);
2352 dentry_unlock_parents_for_move(anon
, dentry
);
2353 spin_unlock(&dentry
->d_lock
);
2355 /* anon->d_lock still locked, returns locked */
2356 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2360 * d_materialise_unique - introduce an inode into the tree
2361 * @dentry: candidate dentry
2362 * @inode: inode to bind to the dentry, to which aliases may be attached
2364 * Introduces an dentry into the tree, substituting an extant disconnected
2365 * root directory alias in its place if there is one
2367 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2369 struct dentry
*actual
;
2371 BUG_ON(!d_unhashed(dentry
));
2375 __d_instantiate(dentry
, NULL
);
2380 spin_lock(&inode
->i_lock
);
2382 if (S_ISDIR(inode
->i_mode
)) {
2383 struct dentry
*alias
;
2385 /* Does an aliased dentry already exist? */
2386 alias
= __d_find_alias(inode
, 0);
2389 /* Is this an anonymous mountpoint that we could splice
2391 if (IS_ROOT(alias
)) {
2392 __d_materialise_dentry(dentry
, alias
);
2396 /* Nope, but we must(!) avoid directory aliasing */
2397 actual
= __d_unalias(inode
, dentry
, alias
);
2404 /* Add a unique reference */
2405 actual
= __d_instantiate_unique(dentry
, inode
);
2409 BUG_ON(!d_unhashed(actual
));
2411 spin_lock(&actual
->d_lock
);
2414 spin_unlock(&actual
->d_lock
);
2415 spin_unlock(&inode
->i_lock
);
2417 if (actual
== dentry
) {
2418 security_d_instantiate(dentry
, inode
);
2425 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2427 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2431 return -ENAMETOOLONG
;
2433 memcpy(*buffer
, str
, namelen
);
2437 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2439 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2443 * Prepend path string to a buffer
2445 * @path: the dentry/vfsmount to report
2446 * @root: root vfsmnt/dentry (may be modified by this function)
2447 * @buffer: pointer to the end of the buffer
2448 * @buflen: pointer to buffer length
2450 * Caller holds the rename_lock.
2452 * If path is not reachable from the supplied root, then the value of
2453 * root is changed (without modifying refcounts).
2455 static int prepend_path(const struct path
*path
, struct path
*root
,
2456 char **buffer
, int *buflen
)
2458 struct dentry
*dentry
= path
->dentry
;
2459 struct vfsmount
*vfsmnt
= path
->mnt
;
2463 br_read_lock(vfsmount_lock
);
2464 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2465 struct dentry
* parent
;
2467 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2469 if (vfsmnt
->mnt_parent
== vfsmnt
) {
2472 dentry
= vfsmnt
->mnt_mountpoint
;
2473 vfsmnt
= vfsmnt
->mnt_parent
;
2476 parent
= dentry
->d_parent
;
2478 spin_lock(&dentry
->d_lock
);
2479 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2480 spin_unlock(&dentry
->d_lock
);
2482 error
= prepend(buffer
, buflen
, "/", 1);
2491 if (!error
&& !slash
)
2492 error
= prepend(buffer
, buflen
, "/", 1);
2494 br_read_unlock(vfsmount_lock
);
2499 * Filesystems needing to implement special "root names"
2500 * should do so with ->d_dname()
2502 if (IS_ROOT(dentry
) &&
2503 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2504 WARN(1, "Root dentry has weird name <%.*s>\n",
2505 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2508 root
->dentry
= dentry
;
2513 * __d_path - return the path of a dentry
2514 * @path: the dentry/vfsmount to report
2515 * @root: root vfsmnt/dentry (may be modified by this function)
2516 * @buf: buffer to return value in
2517 * @buflen: buffer length
2519 * Convert a dentry into an ASCII path name.
2521 * Returns a pointer into the buffer or an error code if the
2522 * path was too long.
2524 * "buflen" should be positive.
2526 * If path is not reachable from the supplied root, then the value of
2527 * root is changed (without modifying refcounts).
2529 char *__d_path(const struct path
*path
, struct path
*root
,
2530 char *buf
, int buflen
)
2532 char *res
= buf
+ buflen
;
2535 prepend(&res
, &buflen
, "\0", 1);
2536 write_seqlock(&rename_lock
);
2537 error
= prepend_path(path
, root
, &res
, &buflen
);
2538 write_sequnlock(&rename_lock
);
2541 return ERR_PTR(error
);
2546 * same as __d_path but appends "(deleted)" for unlinked files.
2548 static int path_with_deleted(const struct path
*path
, struct path
*root
,
2549 char **buf
, int *buflen
)
2551 prepend(buf
, buflen
, "\0", 1);
2552 if (d_unlinked(path
->dentry
)) {
2553 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2558 return prepend_path(path
, root
, buf
, buflen
);
2561 static int prepend_unreachable(char **buffer
, int *buflen
)
2563 return prepend(buffer
, buflen
, "(unreachable)", 13);
2567 * d_path - return the path of a dentry
2568 * @path: path to report
2569 * @buf: buffer to return value in
2570 * @buflen: buffer length
2572 * Convert a dentry into an ASCII path name. If the entry has been deleted
2573 * the string " (deleted)" is appended. Note that this is ambiguous.
2575 * Returns a pointer into the buffer or an error code if the path was
2576 * too long. Note: Callers should use the returned pointer, not the passed
2577 * in buffer, to use the name! The implementation often starts at an offset
2578 * into the buffer, and may leave 0 bytes at the start.
2580 * "buflen" should be positive.
2582 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2584 char *res
= buf
+ buflen
;
2590 * We have various synthetic filesystems that never get mounted. On
2591 * these filesystems dentries are never used for lookup purposes, and
2592 * thus don't need to be hashed. They also don't need a name until a
2593 * user wants to identify the object in /proc/pid/fd/. The little hack
2594 * below allows us to generate a name for these objects on demand:
2596 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2597 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2599 get_fs_root(current
->fs
, &root
);
2600 write_seqlock(&rename_lock
);
2602 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2604 res
= ERR_PTR(error
);
2605 write_sequnlock(&rename_lock
);
2609 EXPORT_SYMBOL(d_path
);
2612 * d_path_with_unreachable - return the path of a dentry
2613 * @path: path to report
2614 * @buf: buffer to return value in
2615 * @buflen: buffer length
2617 * The difference from d_path() is that this prepends "(unreachable)"
2618 * to paths which are unreachable from the current process' root.
2620 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2622 char *res
= buf
+ buflen
;
2627 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2628 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2630 get_fs_root(current
->fs
, &root
);
2631 write_seqlock(&rename_lock
);
2633 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2634 if (!error
&& !path_equal(&tmp
, &root
))
2635 error
= prepend_unreachable(&res
, &buflen
);
2636 write_sequnlock(&rename_lock
);
2639 res
= ERR_PTR(error
);
2645 * Helper function for dentry_operations.d_dname() members
2647 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2648 const char *fmt
, ...)
2654 va_start(args
, fmt
);
2655 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2658 if (sz
> sizeof(temp
) || sz
> buflen
)
2659 return ERR_PTR(-ENAMETOOLONG
);
2661 buffer
+= buflen
- sz
;
2662 return memcpy(buffer
, temp
, sz
);
2666 * Write full pathname from the root of the filesystem into the buffer.
2668 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2670 char *end
= buf
+ buflen
;
2673 prepend(&end
, &buflen
, "\0", 1);
2680 while (!IS_ROOT(dentry
)) {
2681 struct dentry
*parent
= dentry
->d_parent
;
2685 spin_lock(&dentry
->d_lock
);
2686 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2687 spin_unlock(&dentry
->d_lock
);
2688 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2696 return ERR_PTR(-ENAMETOOLONG
);
2699 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2703 write_seqlock(&rename_lock
);
2704 retval
= __dentry_path(dentry
, buf
, buflen
);
2705 write_sequnlock(&rename_lock
);
2709 EXPORT_SYMBOL(dentry_path_raw
);
2711 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2716 write_seqlock(&rename_lock
);
2717 if (d_unlinked(dentry
)) {
2719 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2723 retval
= __dentry_path(dentry
, buf
, buflen
);
2724 write_sequnlock(&rename_lock
);
2725 if (!IS_ERR(retval
) && p
)
2726 *p
= '/'; /* restore '/' overriden with '\0' */
2729 return ERR_PTR(-ENAMETOOLONG
);
2733 * NOTE! The user-level library version returns a
2734 * character pointer. The kernel system call just
2735 * returns the length of the buffer filled (which
2736 * includes the ending '\0' character), or a negative
2737 * error value. So libc would do something like
2739 * char *getcwd(char * buf, size_t size)
2743 * retval = sys_getcwd(buf, size);
2750 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2753 struct path pwd
, root
;
2754 char *page
= (char *) __get_free_page(GFP_USER
);
2759 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2762 write_seqlock(&rename_lock
);
2763 if (!d_unlinked(pwd
.dentry
)) {
2765 struct path tmp
= root
;
2766 char *cwd
= page
+ PAGE_SIZE
;
2767 int buflen
= PAGE_SIZE
;
2769 prepend(&cwd
, &buflen
, "\0", 1);
2770 error
= prepend_path(&pwd
, &tmp
, &cwd
, &buflen
);
2771 write_sequnlock(&rename_lock
);
2776 /* Unreachable from current root */
2777 if (!path_equal(&tmp
, &root
)) {
2778 error
= prepend_unreachable(&cwd
, &buflen
);
2784 len
= PAGE_SIZE
+ page
- cwd
;
2787 if (copy_to_user(buf
, cwd
, len
))
2791 write_sequnlock(&rename_lock
);
2797 free_page((unsigned long) page
);
2802 * Test whether new_dentry is a subdirectory of old_dentry.
2804 * Trivially implemented using the dcache structure
2808 * is_subdir - is new dentry a subdirectory of old_dentry
2809 * @new_dentry: new dentry
2810 * @old_dentry: old dentry
2812 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2813 * Returns 0 otherwise.
2814 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2817 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2822 if (new_dentry
== old_dentry
)
2826 /* for restarting inner loop in case of seq retry */
2827 seq
= read_seqbegin(&rename_lock
);
2829 * Need rcu_readlock to protect against the d_parent trashing
2833 if (d_ancestor(old_dentry
, new_dentry
))
2838 } while (read_seqretry(&rename_lock
, seq
));
2843 int path_is_under(struct path
*path1
, struct path
*path2
)
2845 struct vfsmount
*mnt
= path1
->mnt
;
2846 struct dentry
*dentry
= path1
->dentry
;
2849 br_read_lock(vfsmount_lock
);
2850 if (mnt
!= path2
->mnt
) {
2852 if (mnt
->mnt_parent
== mnt
) {
2853 br_read_unlock(vfsmount_lock
);
2856 if (mnt
->mnt_parent
== path2
->mnt
)
2858 mnt
= mnt
->mnt_parent
;
2860 dentry
= mnt
->mnt_mountpoint
;
2862 res
= is_subdir(dentry
, path2
->dentry
);
2863 br_read_unlock(vfsmount_lock
);
2866 EXPORT_SYMBOL(path_is_under
);
2868 void d_genocide(struct dentry
*root
)
2870 struct dentry
*this_parent
;
2871 struct list_head
*next
;
2875 seq
= read_seqbegin(&rename_lock
);
2878 spin_lock(&this_parent
->d_lock
);
2880 next
= this_parent
->d_subdirs
.next
;
2882 while (next
!= &this_parent
->d_subdirs
) {
2883 struct list_head
*tmp
= next
;
2884 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2887 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2888 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2889 spin_unlock(&dentry
->d_lock
);
2892 if (!list_empty(&dentry
->d_subdirs
)) {
2893 spin_unlock(&this_parent
->d_lock
);
2894 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2895 this_parent
= dentry
;
2896 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2899 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2900 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2903 spin_unlock(&dentry
->d_lock
);
2905 if (this_parent
!= root
) {
2907 struct dentry
*child
;
2909 tmp
= this_parent
->d_parent
;
2910 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2911 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2912 this_parent
->d_count
--;
2915 spin_unlock(&this_parent
->d_lock
);
2916 child
= this_parent
;
2918 spin_lock(&this_parent
->d_lock
);
2919 /* might go back up the wrong parent if we have had a rename
2921 if (this_parent
!= child
->d_parent
||
2922 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
2923 spin_unlock(&this_parent
->d_lock
);
2928 next
= child
->d_u
.d_child
.next
;
2931 spin_unlock(&this_parent
->d_lock
);
2932 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2935 write_sequnlock(&rename_lock
);
2940 write_seqlock(&rename_lock
);
2945 * find_inode_number - check for dentry with name
2946 * @dir: directory to check
2947 * @name: Name to find.
2949 * Check whether a dentry already exists for the given name,
2950 * and return the inode number if it has an inode. Otherwise
2953 * This routine is used to post-process directory listings for
2954 * filesystems using synthetic inode numbers, and is necessary
2955 * to keep getcwd() working.
2958 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2960 struct dentry
* dentry
;
2963 dentry
= d_hash_and_lookup(dir
, name
);
2965 if (dentry
->d_inode
)
2966 ino
= dentry
->d_inode
->i_ino
;
2971 EXPORT_SYMBOL(find_inode_number
);
2973 static __initdata
unsigned long dhash_entries
;
2974 static int __init
set_dhash_entries(char *str
)
2978 dhash_entries
= simple_strtoul(str
, &str
, 0);
2981 __setup("dhash_entries=", set_dhash_entries
);
2983 static void __init
dcache_init_early(void)
2987 /* If hashes are distributed across NUMA nodes, defer
2988 * hash allocation until vmalloc space is available.
2994 alloc_large_system_hash("Dentry cache",
2995 sizeof(struct dcache_hash_bucket
),
3003 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
3004 INIT_HLIST_BL_HEAD(&dentry_hashtable
[loop
].head
);
3007 static void __init
dcache_init(void)
3012 * A constructor could be added for stable state like the lists,
3013 * but it is probably not worth it because of the cache nature
3016 dentry_cache
= KMEM_CACHE(dentry
,
3017 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3019 register_shrinker(&dcache_shrinker
);
3021 /* Hash may have been set up in dcache_init_early */
3026 alloc_large_system_hash("Dentry cache",
3027 sizeof(struct dcache_hash_bucket
),
3035 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
3036 INIT_HLIST_BL_HEAD(&dentry_hashtable
[loop
].head
);
3039 /* SLAB cache for __getname() consumers */
3040 struct kmem_cache
*names_cachep __read_mostly
;
3041 EXPORT_SYMBOL(names_cachep
);
3043 EXPORT_SYMBOL(d_genocide
);
3045 void __init
vfs_caches_init_early(void)
3047 dcache_init_early();
3051 void __init
vfs_caches_init(unsigned long mempages
)
3053 unsigned long reserve
;
3055 /* Base hash sizes on available memory, with a reserve equal to
3056 150% of current kernel size */
3058 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3059 mempages
-= reserve
;
3061 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3062 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3066 files_init(mempages
);