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>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
45 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
47 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
51 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
59 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
64 * dentry->d_inode->i_lock
67 * dcache_hash_bucket lock
70 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
73 * dentry->d_parent->d_lock
76 * If no ancestor relationship:
77 * if (dentry1 < dentry2)
81 int sysctl_vfs_cache_pressure __read_mostly
= 100;
82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
84 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
85 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
87 EXPORT_SYMBOL(rename_lock
);
89 static struct kmem_cache
*dentry_cache __read_mostly
;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
99 #define D_HASHBITS d_hash_shift
100 #define D_HASHMASK d_hash_mask
102 static unsigned int d_hash_mask __read_mostly
;
103 static unsigned int d_hash_shift __read_mostly
;
105 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
107 static inline struct hlist_bl_head
*d_hash(const struct dentry
*parent
,
110 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
111 hash
= hash
+ (hash
>> D_HASHBITS
);
112 return dentry_hashtable
+ (hash
& D_HASHMASK
);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat
= {
120 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 static int get_nr_dentry(void)
127 for_each_possible_cpu(i
)
128 sum
+= per_cpu(nr_dentry
, i
);
129 return sum
< 0 ? 0 : sum
;
132 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
133 size_t *lenp
, loff_t
*ppos
)
135 dentry_stat
.nr_dentry
= get_nr_dentry();
136 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
144 static inline int dentry_cmp(const unsigned char *cs
, size_t scount
,
145 const unsigned char *ct
, size_t tcount
)
147 #ifdef CONFIG_DCACHE_WORD_ACCESS
148 unsigned long a
,b
,mask
;
150 if (unlikely(scount
!= tcount
))
154 a
= *(unsigned long *)cs
;
155 b
= *(unsigned long *)ct
;
156 if (tcount
< sizeof(unsigned long))
158 if (unlikely(a
!= b
))
160 cs
+= sizeof(unsigned long);
161 ct
+= sizeof(unsigned long);
162 tcount
-= sizeof(unsigned long);
166 mask
= ~(~0ul << tcount
*8);
167 return unlikely(!!((a
^ b
) & mask
));
169 if (scount
!= tcount
)
183 static void __d_free(struct rcu_head
*head
)
185 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
187 WARN_ON(!list_empty(&dentry
->d_alias
));
188 if (dname_external(dentry
))
189 kfree(dentry
->d_name
.name
);
190 kmem_cache_free(dentry_cache
, dentry
);
196 static void d_free(struct dentry
*dentry
)
198 BUG_ON(dentry
->d_count
);
199 this_cpu_dec(nr_dentry
);
200 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
201 dentry
->d_op
->d_release(dentry
);
203 /* if dentry was never visible to RCU, immediate free is OK */
204 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
205 __d_free(&dentry
->d_u
.d_rcu
);
207 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
211 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
212 * @dentry: the target dentry
213 * After this call, in-progress rcu-walk path lookup will fail. This
214 * should be called after unhashing, and after changing d_inode (if
215 * the dentry has not already been unhashed).
217 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
219 assert_spin_locked(&dentry
->d_lock
);
220 /* Go through a barrier */
221 write_seqcount_barrier(&dentry
->d_seq
);
225 * Release the dentry's inode, using the filesystem
226 * d_iput() operation if defined. Dentry has no refcount
229 static void dentry_iput(struct dentry
* dentry
)
230 __releases(dentry
->d_lock
)
231 __releases(dentry
->d_inode
->i_lock
)
233 struct inode
*inode
= dentry
->d_inode
;
235 dentry
->d_inode
= NULL
;
236 list_del_init(&dentry
->d_alias
);
237 spin_unlock(&dentry
->d_lock
);
238 spin_unlock(&inode
->i_lock
);
240 fsnotify_inoderemove(inode
);
241 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
242 dentry
->d_op
->d_iput(dentry
, inode
);
246 spin_unlock(&dentry
->d_lock
);
251 * Release the dentry's inode, using the filesystem
252 * d_iput() operation if defined. dentry remains in-use.
254 static void dentry_unlink_inode(struct dentry
* dentry
)
255 __releases(dentry
->d_lock
)
256 __releases(dentry
->d_inode
->i_lock
)
258 struct inode
*inode
= dentry
->d_inode
;
259 dentry
->d_inode
= NULL
;
260 list_del_init(&dentry
->d_alias
);
261 dentry_rcuwalk_barrier(dentry
);
262 spin_unlock(&dentry
->d_lock
);
263 spin_unlock(&inode
->i_lock
);
265 fsnotify_inoderemove(inode
);
266 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
267 dentry
->d_op
->d_iput(dentry
, inode
);
273 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
275 static void dentry_lru_add(struct dentry
*dentry
)
277 if (list_empty(&dentry
->d_lru
)) {
278 spin_lock(&dcache_lru_lock
);
279 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
280 dentry
->d_sb
->s_nr_dentry_unused
++;
281 dentry_stat
.nr_unused
++;
282 spin_unlock(&dcache_lru_lock
);
286 static void __dentry_lru_del(struct dentry
*dentry
)
288 list_del_init(&dentry
->d_lru
);
289 dentry
->d_flags
&= ~DCACHE_SHRINK_LIST
;
290 dentry
->d_sb
->s_nr_dentry_unused
--;
291 dentry_stat
.nr_unused
--;
295 * Remove a dentry with references from the LRU.
297 static void dentry_lru_del(struct dentry
*dentry
)
299 if (!list_empty(&dentry
->d_lru
)) {
300 spin_lock(&dcache_lru_lock
);
301 __dentry_lru_del(dentry
);
302 spin_unlock(&dcache_lru_lock
);
307 * Remove a dentry that is unreferenced and about to be pruned
308 * (unhashed and destroyed) from the LRU, and inform the file system.
309 * This wrapper should be called _prior_ to unhashing a victim dentry.
311 static void dentry_lru_prune(struct dentry
*dentry
)
313 if (!list_empty(&dentry
->d_lru
)) {
314 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
315 dentry
->d_op
->d_prune(dentry
);
317 spin_lock(&dcache_lru_lock
);
318 __dentry_lru_del(dentry
);
319 spin_unlock(&dcache_lru_lock
);
323 static void dentry_lru_move_list(struct dentry
*dentry
, struct list_head
*list
)
325 spin_lock(&dcache_lru_lock
);
326 if (list_empty(&dentry
->d_lru
)) {
327 list_add_tail(&dentry
->d_lru
, list
);
328 dentry
->d_sb
->s_nr_dentry_unused
++;
329 dentry_stat
.nr_unused
++;
331 list_move_tail(&dentry
->d_lru
, list
);
333 spin_unlock(&dcache_lru_lock
);
337 * d_kill - kill dentry and return parent
338 * @dentry: dentry to kill
339 * @parent: parent dentry
341 * The dentry must already be unhashed and removed from the LRU.
343 * If this is the root of the dentry tree, return NULL.
345 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
348 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
349 __releases(dentry
->d_lock
)
350 __releases(parent
->d_lock
)
351 __releases(dentry
->d_inode
->i_lock
)
353 list_del(&dentry
->d_u
.d_child
);
355 * Inform try_to_ascend() that we are no longer attached to the
358 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
360 spin_unlock(&parent
->d_lock
);
363 * dentry_iput drops the locks, at which point nobody (except
364 * transient RCU lookups) can reach this dentry.
371 * Unhash a dentry without inserting an RCU walk barrier or checking that
372 * dentry->d_lock is locked. The caller must take care of that, if
375 static void __d_shrink(struct dentry
*dentry
)
377 if (!d_unhashed(dentry
)) {
378 struct hlist_bl_head
*b
;
379 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
380 b
= &dentry
->d_sb
->s_anon
;
382 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
385 __hlist_bl_del(&dentry
->d_hash
);
386 dentry
->d_hash
.pprev
= NULL
;
392 * d_drop - drop a dentry
393 * @dentry: dentry to drop
395 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
396 * be found through a VFS lookup any more. Note that this is different from
397 * deleting the dentry - d_delete will try to mark the dentry negative if
398 * possible, giving a successful _negative_ lookup, while d_drop will
399 * just make the cache lookup fail.
401 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
402 * reason (NFS timeouts or autofs deletes).
404 * __d_drop requires dentry->d_lock.
406 void __d_drop(struct dentry
*dentry
)
408 if (!d_unhashed(dentry
)) {
410 dentry_rcuwalk_barrier(dentry
);
413 EXPORT_SYMBOL(__d_drop
);
415 void d_drop(struct dentry
*dentry
)
417 spin_lock(&dentry
->d_lock
);
419 spin_unlock(&dentry
->d_lock
);
421 EXPORT_SYMBOL(d_drop
);
424 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
425 * @dentry: dentry to drop
427 * This is called when we do a lookup on a placeholder dentry that needed to be
428 * looked up. The dentry should have been hashed in order for it to be found by
429 * the lookup code, but now needs to be unhashed while we do the actual lookup
430 * and clear the DCACHE_NEED_LOOKUP flag.
432 void d_clear_need_lookup(struct dentry
*dentry
)
434 spin_lock(&dentry
->d_lock
);
436 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
437 spin_unlock(&dentry
->d_lock
);
439 EXPORT_SYMBOL(d_clear_need_lookup
);
442 * Finish off a dentry we've decided to kill.
443 * dentry->d_lock must be held, returns with it unlocked.
444 * If ref is non-zero, then decrement the refcount too.
445 * Returns dentry requiring refcount drop, or NULL if we're done.
447 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
448 __releases(dentry
->d_lock
)
451 struct dentry
*parent
;
453 inode
= dentry
->d_inode
;
454 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
456 spin_unlock(&dentry
->d_lock
);
458 return dentry
; /* try again with same dentry */
463 parent
= dentry
->d_parent
;
464 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
466 spin_unlock(&inode
->i_lock
);
473 * if dentry was on the d_lru list delete it from there.
474 * inform the fs via d_prune that this dentry is about to be
475 * unhashed and destroyed.
477 dentry_lru_prune(dentry
);
478 /* if it was on the hash then remove it */
480 return d_kill(dentry
, parent
);
486 * This is complicated by the fact that we do not want to put
487 * dentries that are no longer on any hash chain on the unused
488 * list: we'd much rather just get rid of them immediately.
490 * However, that implies that we have to traverse the dentry
491 * tree upwards to the parents which might _also_ now be
492 * scheduled for deletion (it may have been only waiting for
493 * its last child to go away).
495 * This tail recursion is done by hand as we don't want to depend
496 * on the compiler to always get this right (gcc generally doesn't).
497 * Real recursion would eat up our stack space.
501 * dput - release a dentry
502 * @dentry: dentry to release
504 * Release a dentry. This will drop the usage count and if appropriate
505 * call the dentry unlink method as well as removing it from the queues and
506 * releasing its resources. If the parent dentries were scheduled for release
507 * they too may now get deleted.
509 void dput(struct dentry
*dentry
)
515 if (dentry
->d_count
== 1)
517 spin_lock(&dentry
->d_lock
);
518 BUG_ON(!dentry
->d_count
);
519 if (dentry
->d_count
> 1) {
521 spin_unlock(&dentry
->d_lock
);
525 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
526 if (dentry
->d_op
->d_delete(dentry
))
530 /* Unreachable? Get rid of it */
531 if (d_unhashed(dentry
))
535 * If this dentry needs lookup, don't set the referenced flag so that it
536 * is more likely to be cleaned up by the dcache shrinker in case of
539 if (!d_need_lookup(dentry
))
540 dentry
->d_flags
|= DCACHE_REFERENCED
;
541 dentry_lru_add(dentry
);
544 spin_unlock(&dentry
->d_lock
);
548 dentry
= dentry_kill(dentry
, 1);
555 * d_invalidate - invalidate a dentry
556 * @dentry: dentry to invalidate
558 * Try to invalidate the dentry if it turns out to be
559 * possible. If there are other dentries that can be
560 * reached through this one we can't delete it and we
561 * return -EBUSY. On success we return 0.
566 int d_invalidate(struct dentry
* dentry
)
569 * If it's already been dropped, return OK.
571 spin_lock(&dentry
->d_lock
);
572 if (d_unhashed(dentry
)) {
573 spin_unlock(&dentry
->d_lock
);
577 * Check whether to do a partial shrink_dcache
578 * to get rid of unused child entries.
580 if (!list_empty(&dentry
->d_subdirs
)) {
581 spin_unlock(&dentry
->d_lock
);
582 shrink_dcache_parent(dentry
);
583 spin_lock(&dentry
->d_lock
);
587 * Somebody else still using it?
589 * If it's a directory, we can't drop it
590 * for fear of somebody re-populating it
591 * with children (even though dropping it
592 * would make it unreachable from the root,
593 * we might still populate it if it was a
594 * working directory or similar).
595 * We also need to leave mountpoints alone,
598 if (dentry
->d_count
> 1 && dentry
->d_inode
) {
599 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
600 spin_unlock(&dentry
->d_lock
);
606 spin_unlock(&dentry
->d_lock
);
609 EXPORT_SYMBOL(d_invalidate
);
611 /* This must be called with d_lock held */
612 static inline void __dget_dlock(struct dentry
*dentry
)
617 static inline void __dget(struct dentry
*dentry
)
619 spin_lock(&dentry
->d_lock
);
620 __dget_dlock(dentry
);
621 spin_unlock(&dentry
->d_lock
);
624 struct dentry
*dget_parent(struct dentry
*dentry
)
630 * Don't need rcu_dereference because we re-check it was correct under
634 ret
= dentry
->d_parent
;
635 spin_lock(&ret
->d_lock
);
636 if (unlikely(ret
!= dentry
->d_parent
)) {
637 spin_unlock(&ret
->d_lock
);
642 BUG_ON(!ret
->d_count
);
644 spin_unlock(&ret
->d_lock
);
647 EXPORT_SYMBOL(dget_parent
);
650 * d_find_alias - grab a hashed alias of inode
651 * @inode: inode in question
652 * @want_discon: flag, used by d_splice_alias, to request
653 * that only a DISCONNECTED alias be returned.
655 * If inode has a hashed alias, or is a directory and has any alias,
656 * acquire the reference to alias and return it. Otherwise return NULL.
657 * Notice that if inode is a directory there can be only one alias and
658 * it can be unhashed only if it has no children, or if it is the root
661 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
662 * any other hashed alias over that one unless @want_discon is set,
663 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
665 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
667 struct dentry
*alias
, *discon_alias
;
671 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
672 spin_lock(&alias
->d_lock
);
673 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
674 if (IS_ROOT(alias
) &&
675 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
676 discon_alias
= alias
;
677 } else if (!want_discon
) {
679 spin_unlock(&alias
->d_lock
);
683 spin_unlock(&alias
->d_lock
);
686 alias
= discon_alias
;
687 spin_lock(&alias
->d_lock
);
688 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
689 if (IS_ROOT(alias
) &&
690 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
692 spin_unlock(&alias
->d_lock
);
696 spin_unlock(&alias
->d_lock
);
702 struct dentry
*d_find_alias(struct inode
*inode
)
704 struct dentry
*de
= NULL
;
706 if (!list_empty(&inode
->i_dentry
)) {
707 spin_lock(&inode
->i_lock
);
708 de
= __d_find_alias(inode
, 0);
709 spin_unlock(&inode
->i_lock
);
713 EXPORT_SYMBOL(d_find_alias
);
716 * Try to kill dentries associated with this inode.
717 * WARNING: you must own a reference to inode.
719 void d_prune_aliases(struct inode
*inode
)
721 struct dentry
*dentry
;
723 spin_lock(&inode
->i_lock
);
724 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
725 spin_lock(&dentry
->d_lock
);
726 if (!dentry
->d_count
) {
727 __dget_dlock(dentry
);
729 spin_unlock(&dentry
->d_lock
);
730 spin_unlock(&inode
->i_lock
);
734 spin_unlock(&dentry
->d_lock
);
736 spin_unlock(&inode
->i_lock
);
738 EXPORT_SYMBOL(d_prune_aliases
);
741 * Try to throw away a dentry - free the inode, dput the parent.
742 * Requires dentry->d_lock is held, and dentry->d_count == 0.
743 * Releases dentry->d_lock.
745 * This may fail if locks cannot be acquired no problem, just try again.
747 static void try_prune_one_dentry(struct dentry
*dentry
)
748 __releases(dentry
->d_lock
)
750 struct dentry
*parent
;
752 parent
= dentry_kill(dentry
, 0);
754 * If dentry_kill returns NULL, we have nothing more to do.
755 * if it returns the same dentry, trylocks failed. In either
756 * case, just loop again.
758 * Otherwise, we need to prune ancestors too. This is necessary
759 * to prevent quadratic behavior of shrink_dcache_parent(), but
760 * is also expected to be beneficial in reducing dentry cache
765 if (parent
== dentry
)
768 /* Prune ancestors. */
771 spin_lock(&dentry
->d_lock
);
772 if (dentry
->d_count
> 1) {
774 spin_unlock(&dentry
->d_lock
);
777 dentry
= dentry_kill(dentry
, 1);
781 static void shrink_dentry_list(struct list_head
*list
)
783 struct dentry
*dentry
;
787 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
788 if (&dentry
->d_lru
== list
)
790 spin_lock(&dentry
->d_lock
);
791 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
792 spin_unlock(&dentry
->d_lock
);
797 * We found an inuse dentry which was not removed from
798 * the LRU because of laziness during lookup. Do not free
799 * it - just keep it off the LRU list.
801 if (dentry
->d_count
) {
802 dentry_lru_del(dentry
);
803 spin_unlock(&dentry
->d_lock
);
809 try_prune_one_dentry(dentry
);
817 * prune_dcache_sb - shrink the dcache
819 * @count: number of entries to try to free
821 * Attempt to shrink the superblock dcache LRU by @count entries. This is
822 * done when we need more memory an called from the superblock shrinker
825 * This function may fail to free any resources if all the dentries are in
828 void prune_dcache_sb(struct super_block
*sb
, int count
)
830 struct dentry
*dentry
;
831 LIST_HEAD(referenced
);
835 spin_lock(&dcache_lru_lock
);
836 while (!list_empty(&sb
->s_dentry_lru
)) {
837 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
838 struct dentry
, d_lru
);
839 BUG_ON(dentry
->d_sb
!= sb
);
841 if (!spin_trylock(&dentry
->d_lock
)) {
842 spin_unlock(&dcache_lru_lock
);
847 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
848 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
849 list_move(&dentry
->d_lru
, &referenced
);
850 spin_unlock(&dentry
->d_lock
);
852 list_move_tail(&dentry
->d_lru
, &tmp
);
853 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
854 spin_unlock(&dentry
->d_lock
);
858 cond_resched_lock(&dcache_lru_lock
);
860 if (!list_empty(&referenced
))
861 list_splice(&referenced
, &sb
->s_dentry_lru
);
862 spin_unlock(&dcache_lru_lock
);
864 shrink_dentry_list(&tmp
);
868 * shrink_dcache_sb - shrink dcache for a superblock
871 * Shrink the dcache for the specified super block. This is used to free
872 * the dcache before unmounting a file system.
874 void shrink_dcache_sb(struct super_block
*sb
)
878 spin_lock(&dcache_lru_lock
);
879 while (!list_empty(&sb
->s_dentry_lru
)) {
880 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
881 spin_unlock(&dcache_lru_lock
);
882 shrink_dentry_list(&tmp
);
883 spin_lock(&dcache_lru_lock
);
885 spin_unlock(&dcache_lru_lock
);
887 EXPORT_SYMBOL(shrink_dcache_sb
);
890 * destroy a single subtree of dentries for unmount
891 * - see the comments on shrink_dcache_for_umount() for a description of the
894 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
896 struct dentry
*parent
;
898 BUG_ON(!IS_ROOT(dentry
));
901 /* descend to the first leaf in the current subtree */
902 while (!list_empty(&dentry
->d_subdirs
))
903 dentry
= list_entry(dentry
->d_subdirs
.next
,
904 struct dentry
, d_u
.d_child
);
906 /* consume the dentries from this leaf up through its parents
907 * until we find one with children or run out altogether */
912 * remove the dentry from the lru, and inform
913 * the fs that this dentry is about to be
914 * unhashed and destroyed.
916 dentry_lru_prune(dentry
);
919 if (dentry
->d_count
!= 0) {
921 "BUG: Dentry %p{i=%lx,n=%s}"
923 " [unmount of %s %s]\n",
926 dentry
->d_inode
->i_ino
: 0UL,
929 dentry
->d_sb
->s_type
->name
,
934 if (IS_ROOT(dentry
)) {
936 list_del(&dentry
->d_u
.d_child
);
938 parent
= dentry
->d_parent
;
940 list_del(&dentry
->d_u
.d_child
);
943 inode
= dentry
->d_inode
;
945 dentry
->d_inode
= NULL
;
946 list_del_init(&dentry
->d_alias
);
947 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
948 dentry
->d_op
->d_iput(dentry
, inode
);
955 /* finished when we fall off the top of the tree,
956 * otherwise we ascend to the parent and move to the
957 * next sibling if there is one */
961 } while (list_empty(&dentry
->d_subdirs
));
963 dentry
= list_entry(dentry
->d_subdirs
.next
,
964 struct dentry
, d_u
.d_child
);
969 * destroy the dentries attached to a superblock on unmounting
970 * - we don't need to use dentry->d_lock because:
971 * - the superblock is detached from all mountings and open files, so the
972 * dentry trees will not be rearranged by the VFS
973 * - s_umount is write-locked, so the memory pressure shrinker will ignore
974 * any dentries belonging to this superblock that it comes across
975 * - the filesystem itself is no longer permitted to rearrange the dentries
978 void shrink_dcache_for_umount(struct super_block
*sb
)
980 struct dentry
*dentry
;
982 if (down_read_trylock(&sb
->s_umount
))
988 shrink_dcache_for_umount_subtree(dentry
);
990 while (!hlist_bl_empty(&sb
->s_anon
)) {
991 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
992 shrink_dcache_for_umount_subtree(dentry
);
997 * This tries to ascend one level of parenthood, but
998 * we can race with renaming, so we need to re-check
999 * the parenthood after dropping the lock and check
1000 * that the sequence number still matches.
1002 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
1004 struct dentry
*new = old
->d_parent
;
1007 spin_unlock(&old
->d_lock
);
1008 spin_lock(&new->d_lock
);
1011 * might go back up the wrong parent if we have had a rename
1014 if (new != old
->d_parent
||
1015 (old
->d_flags
& DCACHE_DISCONNECTED
) ||
1016 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1017 spin_unlock(&new->d_lock
);
1026 * Search for at least 1 mount point in the dentry's subdirs.
1027 * We descend to the next level whenever the d_subdirs
1028 * list is non-empty and continue searching.
1032 * have_submounts - check for mounts over a dentry
1033 * @parent: dentry to check.
1035 * Return true if the parent or its subdirectories contain
1038 int have_submounts(struct dentry
*parent
)
1040 struct dentry
*this_parent
;
1041 struct list_head
*next
;
1045 seq
= read_seqbegin(&rename_lock
);
1047 this_parent
= parent
;
1049 if (d_mountpoint(parent
))
1051 spin_lock(&this_parent
->d_lock
);
1053 next
= this_parent
->d_subdirs
.next
;
1055 while (next
!= &this_parent
->d_subdirs
) {
1056 struct list_head
*tmp
= next
;
1057 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1060 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1061 /* Have we found a mount point ? */
1062 if (d_mountpoint(dentry
)) {
1063 spin_unlock(&dentry
->d_lock
);
1064 spin_unlock(&this_parent
->d_lock
);
1067 if (!list_empty(&dentry
->d_subdirs
)) {
1068 spin_unlock(&this_parent
->d_lock
);
1069 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1070 this_parent
= dentry
;
1071 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1074 spin_unlock(&dentry
->d_lock
);
1077 * All done at this level ... ascend and resume the search.
1079 if (this_parent
!= parent
) {
1080 struct dentry
*child
= this_parent
;
1081 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1084 next
= child
->d_u
.d_child
.next
;
1087 spin_unlock(&this_parent
->d_lock
);
1088 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1091 write_sequnlock(&rename_lock
);
1092 return 0; /* No mount points found in tree */
1094 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1097 write_sequnlock(&rename_lock
);
1102 write_seqlock(&rename_lock
);
1105 EXPORT_SYMBOL(have_submounts
);
1108 * Search the dentry child list for the specified parent,
1109 * and move any unused dentries to the end of the unused
1110 * list for prune_dcache(). We descend to the next level
1111 * whenever the d_subdirs list is non-empty and continue
1114 * It returns zero iff there are no unused children,
1115 * otherwise it returns the number of children moved to
1116 * the end of the unused list. This may not be the total
1117 * number of unused children, because select_parent can
1118 * drop the lock and return early due to latency
1121 static int select_parent(struct dentry
*parent
, struct list_head
*dispose
)
1123 struct dentry
*this_parent
;
1124 struct list_head
*next
;
1129 seq
= read_seqbegin(&rename_lock
);
1131 this_parent
= parent
;
1132 spin_lock(&this_parent
->d_lock
);
1134 next
= this_parent
->d_subdirs
.next
;
1136 while (next
!= &this_parent
->d_subdirs
) {
1137 struct list_head
*tmp
= next
;
1138 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1141 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1144 * move only zero ref count dentries to the dispose list.
1146 * Those which are presently on the shrink list, being processed
1147 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1148 * loop in shrink_dcache_parent() might not make any progress
1151 if (dentry
->d_count
) {
1152 dentry_lru_del(dentry
);
1153 } else if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
)) {
1154 dentry_lru_move_list(dentry
, dispose
);
1155 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
1159 * We can return to the caller if we have found some (this
1160 * ensures forward progress). We'll be coming back to find
1163 if (found
&& need_resched()) {
1164 spin_unlock(&dentry
->d_lock
);
1169 * Descend a level if the d_subdirs list is non-empty.
1171 if (!list_empty(&dentry
->d_subdirs
)) {
1172 spin_unlock(&this_parent
->d_lock
);
1173 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1174 this_parent
= dentry
;
1175 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1179 spin_unlock(&dentry
->d_lock
);
1182 * All done at this level ... ascend and resume the search.
1184 if (this_parent
!= parent
) {
1185 struct dentry
*child
= this_parent
;
1186 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1189 next
= child
->d_u
.d_child
.next
;
1193 spin_unlock(&this_parent
->d_lock
);
1194 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1197 write_sequnlock(&rename_lock
);
1204 write_seqlock(&rename_lock
);
1209 * shrink_dcache_parent - prune dcache
1210 * @parent: parent of entries to prune
1212 * Prune the dcache to remove unused children of the parent dentry.
1214 void shrink_dcache_parent(struct dentry
* parent
)
1219 while ((found
= select_parent(parent
, &dispose
)) != 0)
1220 shrink_dentry_list(&dispose
);
1222 EXPORT_SYMBOL(shrink_dcache_parent
);
1225 * __d_alloc - allocate a dcache entry
1226 * @sb: filesystem it will belong to
1227 * @name: qstr of the name
1229 * Allocates a dentry. It returns %NULL if there is insufficient memory
1230 * available. On a success the dentry is returned. The name passed in is
1231 * copied and the copy passed in may be reused after this call.
1234 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1236 struct dentry
*dentry
;
1239 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1243 if (name
->len
> DNAME_INLINE_LEN
-1) {
1244 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1246 kmem_cache_free(dentry_cache
, dentry
);
1250 dname
= dentry
->d_iname
;
1252 dentry
->d_name
.name
= dname
;
1254 dentry
->d_name
.len
= name
->len
;
1255 dentry
->d_name
.hash
= name
->hash
;
1256 memcpy(dname
, name
->name
, name
->len
);
1257 dname
[name
->len
] = 0;
1259 dentry
->d_count
= 1;
1260 dentry
->d_flags
= 0;
1261 spin_lock_init(&dentry
->d_lock
);
1262 seqcount_init(&dentry
->d_seq
);
1263 dentry
->d_inode
= NULL
;
1264 dentry
->d_parent
= dentry
;
1266 dentry
->d_op
= NULL
;
1267 dentry
->d_fsdata
= NULL
;
1268 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1269 INIT_LIST_HEAD(&dentry
->d_lru
);
1270 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1271 INIT_LIST_HEAD(&dentry
->d_alias
);
1272 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1273 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1275 this_cpu_inc(nr_dentry
);
1281 * d_alloc - allocate a dcache entry
1282 * @parent: parent of entry to allocate
1283 * @name: qstr of the name
1285 * Allocates a dentry. It returns %NULL if there is insufficient memory
1286 * available. On a success the dentry is returned. The name passed in is
1287 * copied and the copy passed in may be reused after this call.
1289 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1291 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1295 spin_lock(&parent
->d_lock
);
1297 * don't need child lock because it is not subject
1298 * to concurrency here
1300 __dget_dlock(parent
);
1301 dentry
->d_parent
= parent
;
1302 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1303 spin_unlock(&parent
->d_lock
);
1307 EXPORT_SYMBOL(d_alloc
);
1309 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1311 struct dentry
*dentry
= __d_alloc(sb
, name
);
1313 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1316 EXPORT_SYMBOL(d_alloc_pseudo
);
1318 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1323 q
.len
= strlen(name
);
1324 q
.hash
= full_name_hash(q
.name
, q
.len
);
1325 return d_alloc(parent
, &q
);
1327 EXPORT_SYMBOL(d_alloc_name
);
1329 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1331 WARN_ON_ONCE(dentry
->d_op
);
1332 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1334 DCACHE_OP_REVALIDATE
|
1335 DCACHE_OP_DELETE
));
1340 dentry
->d_flags
|= DCACHE_OP_HASH
;
1342 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1343 if (op
->d_revalidate
)
1344 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1346 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1348 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1351 EXPORT_SYMBOL(d_set_d_op
);
1353 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1355 spin_lock(&dentry
->d_lock
);
1357 if (unlikely(IS_AUTOMOUNT(inode
)))
1358 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1359 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1361 dentry
->d_inode
= inode
;
1362 dentry_rcuwalk_barrier(dentry
);
1363 spin_unlock(&dentry
->d_lock
);
1364 fsnotify_d_instantiate(dentry
, inode
);
1368 * d_instantiate - fill in inode information for a dentry
1369 * @entry: dentry to complete
1370 * @inode: inode to attach to this dentry
1372 * Fill in inode information in the entry.
1374 * This turns negative dentries into productive full members
1377 * NOTE! This assumes that the inode count has been incremented
1378 * (or otherwise set) by the caller to indicate that it is now
1379 * in use by the dcache.
1382 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1384 BUG_ON(!list_empty(&entry
->d_alias
));
1386 spin_lock(&inode
->i_lock
);
1387 __d_instantiate(entry
, inode
);
1389 spin_unlock(&inode
->i_lock
);
1390 security_d_instantiate(entry
, inode
);
1392 EXPORT_SYMBOL(d_instantiate
);
1395 * d_instantiate_unique - instantiate a non-aliased dentry
1396 * @entry: dentry to instantiate
1397 * @inode: inode to attach to this dentry
1399 * Fill in inode information in the entry. On success, it returns NULL.
1400 * If an unhashed alias of "entry" already exists, then we return the
1401 * aliased dentry instead and drop one reference to inode.
1403 * Note that in order to avoid conflicts with rename() etc, the caller
1404 * had better be holding the parent directory semaphore.
1406 * This also assumes that the inode count has been incremented
1407 * (or otherwise set) by the caller to indicate that it is now
1408 * in use by the dcache.
1410 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1411 struct inode
*inode
)
1413 struct dentry
*alias
;
1414 int len
= entry
->d_name
.len
;
1415 const char *name
= entry
->d_name
.name
;
1416 unsigned int hash
= entry
->d_name
.hash
;
1419 __d_instantiate(entry
, NULL
);
1423 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1424 struct qstr
*qstr
= &alias
->d_name
;
1427 * Don't need alias->d_lock here, because aliases with
1428 * d_parent == entry->d_parent are not subject to name or
1429 * parent changes, because the parent inode i_mutex is held.
1431 if (qstr
->hash
!= hash
)
1433 if (alias
->d_parent
!= entry
->d_parent
)
1435 if (dentry_cmp(qstr
->name
, qstr
->len
, name
, len
))
1441 __d_instantiate(entry
, inode
);
1445 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1447 struct dentry
*result
;
1449 BUG_ON(!list_empty(&entry
->d_alias
));
1452 spin_lock(&inode
->i_lock
);
1453 result
= __d_instantiate_unique(entry
, inode
);
1455 spin_unlock(&inode
->i_lock
);
1458 security_d_instantiate(entry
, inode
);
1462 BUG_ON(!d_unhashed(result
));
1467 EXPORT_SYMBOL(d_instantiate_unique
);
1470 * d_alloc_root - allocate root dentry
1471 * @root_inode: inode to allocate the root for
1473 * Allocate a root ("/") dentry for the inode given. The inode is
1474 * instantiated and returned. %NULL is returned if there is insufficient
1475 * memory or the inode passed is %NULL.
1478 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1480 struct dentry
*res
= NULL
;
1483 static const struct qstr name
= { .name
= "/", .len
= 1 };
1485 res
= __d_alloc(root_inode
->i_sb
, &name
);
1487 d_instantiate(res
, root_inode
);
1491 EXPORT_SYMBOL(d_alloc_root
);
1493 struct dentry
*d_make_root(struct inode
*root_inode
)
1495 struct dentry
*res
= NULL
;
1498 static const struct qstr name
= { .name
= "/", .len
= 1 };
1500 res
= __d_alloc(root_inode
->i_sb
, &name
);
1502 d_instantiate(res
, root_inode
);
1508 EXPORT_SYMBOL(d_make_root
);
1510 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1512 struct dentry
*alias
;
1514 if (list_empty(&inode
->i_dentry
))
1516 alias
= list_first_entry(&inode
->i_dentry
, struct dentry
, d_alias
);
1522 * d_find_any_alias - find any alias for a given inode
1523 * @inode: inode to find an alias for
1525 * If any aliases exist for the given inode, take and return a
1526 * reference for one of them. If no aliases exist, return %NULL.
1528 struct dentry
*d_find_any_alias(struct inode
*inode
)
1532 spin_lock(&inode
->i_lock
);
1533 de
= __d_find_any_alias(inode
);
1534 spin_unlock(&inode
->i_lock
);
1537 EXPORT_SYMBOL(d_find_any_alias
);
1540 * d_obtain_alias - find or allocate a dentry for a given inode
1541 * @inode: inode to allocate the dentry for
1543 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1544 * similar open by handle operations. The returned dentry may be anonymous,
1545 * or may have a full name (if the inode was already in the cache).
1547 * When called on a directory inode, we must ensure that the inode only ever
1548 * has one dentry. If a dentry is found, that is returned instead of
1549 * allocating a new one.
1551 * On successful return, the reference to the inode has been transferred
1552 * to the dentry. In case of an error the reference on the inode is released.
1553 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1554 * be passed in and will be the error will be propagate to the return value,
1555 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1557 struct dentry
*d_obtain_alias(struct inode
*inode
)
1559 static const struct qstr anonstring
= { .name
= "" };
1564 return ERR_PTR(-ESTALE
);
1566 return ERR_CAST(inode
);
1568 res
= d_find_any_alias(inode
);
1572 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1574 res
= ERR_PTR(-ENOMEM
);
1578 spin_lock(&inode
->i_lock
);
1579 res
= __d_find_any_alias(inode
);
1581 spin_unlock(&inode
->i_lock
);
1586 /* attach a disconnected dentry */
1587 spin_lock(&tmp
->d_lock
);
1588 tmp
->d_inode
= inode
;
1589 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1590 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1591 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1592 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1593 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1594 spin_unlock(&tmp
->d_lock
);
1595 spin_unlock(&inode
->i_lock
);
1596 security_d_instantiate(tmp
, inode
);
1601 if (res
&& !IS_ERR(res
))
1602 security_d_instantiate(res
, inode
);
1606 EXPORT_SYMBOL(d_obtain_alias
);
1609 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1610 * @inode: the inode which may have a disconnected dentry
1611 * @dentry: a negative dentry which we want to point to the inode.
1613 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1614 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1615 * and return it, else simply d_add the inode to the dentry and return NULL.
1617 * This is needed in the lookup routine of any filesystem that is exportable
1618 * (via knfsd) so that we can build dcache paths to directories effectively.
1620 * If a dentry was found and moved, then it is returned. Otherwise NULL
1621 * is returned. This matches the expected return value of ->lookup.
1624 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1626 struct dentry
*new = NULL
;
1629 return ERR_CAST(inode
);
1631 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1632 spin_lock(&inode
->i_lock
);
1633 new = __d_find_alias(inode
, 1);
1635 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1636 spin_unlock(&inode
->i_lock
);
1637 security_d_instantiate(new, inode
);
1638 d_move(new, dentry
);
1641 /* already taking inode->i_lock, so d_add() by hand */
1642 __d_instantiate(dentry
, inode
);
1643 spin_unlock(&inode
->i_lock
);
1644 security_d_instantiate(dentry
, inode
);
1648 d_add(dentry
, inode
);
1651 EXPORT_SYMBOL(d_splice_alias
);
1654 * d_add_ci - lookup or allocate new dentry with case-exact name
1655 * @inode: the inode case-insensitive lookup has found
1656 * @dentry: the negative dentry that was passed to the parent's lookup func
1657 * @name: the case-exact name to be associated with the returned dentry
1659 * This is to avoid filling the dcache with case-insensitive names to the
1660 * same inode, only the actual correct case is stored in the dcache for
1661 * case-insensitive filesystems.
1663 * For a case-insensitive lookup match and if the the case-exact dentry
1664 * already exists in in the dcache, use it and return it.
1666 * If no entry exists with the exact case name, allocate new dentry with
1667 * the exact case, and return the spliced entry.
1669 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1673 struct dentry
*found
;
1677 * First check if a dentry matching the name already exists,
1678 * if not go ahead and create it now.
1680 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1682 new = d_alloc(dentry
->d_parent
, name
);
1688 found
= d_splice_alias(inode
, new);
1697 * If a matching dentry exists, and it's not negative use it.
1699 * Decrement the reference count to balance the iget() done
1702 if (found
->d_inode
) {
1703 if (unlikely(found
->d_inode
!= inode
)) {
1704 /* This can't happen because bad inodes are unhashed. */
1705 BUG_ON(!is_bad_inode(inode
));
1706 BUG_ON(!is_bad_inode(found
->d_inode
));
1713 * We are going to instantiate this dentry, unhash it and clear the
1714 * lookup flag so we can do that.
1716 if (unlikely(d_need_lookup(found
)))
1717 d_clear_need_lookup(found
);
1720 * Negative dentry: instantiate it unless the inode is a directory and
1721 * already has a dentry.
1723 new = d_splice_alias(inode
, found
);
1732 return ERR_PTR(error
);
1734 EXPORT_SYMBOL(d_add_ci
);
1737 * __d_lookup_rcu - search for a dentry (racy, store-free)
1738 * @parent: parent dentry
1739 * @name: qstr of name we wish to find
1740 * @seq: returns d_seq value at the point where the dentry was found
1741 * @inode: returns dentry->d_inode when the inode was found valid.
1742 * Returns: dentry, or NULL
1744 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1745 * resolution (store-free path walking) design described in
1746 * Documentation/filesystems/path-lookup.txt.
1748 * This is not to be used outside core vfs.
1750 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1751 * held, and rcu_read_lock held. The returned dentry must not be stored into
1752 * without taking d_lock and checking d_seq sequence count against @seq
1755 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1758 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1759 * the returned dentry, so long as its parent's seqlock is checked after the
1760 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1761 * is formed, giving integrity down the path walk.
1763 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
1764 const struct qstr
*name
,
1765 unsigned *seqp
, struct inode
**inode
)
1767 unsigned int len
= name
->len
;
1768 unsigned int hash
= name
->hash
;
1769 const unsigned char *str
= name
->name
;
1770 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1771 struct hlist_bl_node
*node
;
1772 struct dentry
*dentry
;
1775 * Note: There is significant duplication with __d_lookup_rcu which is
1776 * required to prevent single threaded performance regressions
1777 * especially on architectures where smp_rmb (in seqcounts) are costly.
1778 * Keep the two functions in sync.
1782 * The hash list is protected using RCU.
1784 * Carefully use d_seq when comparing a candidate dentry, to avoid
1785 * races with d_move().
1787 * It is possible that concurrent renames can mess up our list
1788 * walk here and result in missing our dentry, resulting in the
1789 * false-negative result. d_lookup() protects against concurrent
1790 * renames using rename_lock seqlock.
1792 * See Documentation/filesystems/path-lookup.txt for more details.
1794 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1800 if (dentry
->d_name
.hash
!= hash
)
1804 seq
= read_seqcount_begin(&dentry
->d_seq
);
1805 if (dentry
->d_parent
!= parent
)
1807 if (d_unhashed(dentry
))
1809 tlen
= dentry
->d_name
.len
;
1810 tname
= dentry
->d_name
.name
;
1811 i
= dentry
->d_inode
;
1814 * This seqcount check is required to ensure name and
1815 * len are loaded atomically, so as not to walk off the
1816 * edge of memory when walking. If we could load this
1817 * atomically some other way, we could drop this check.
1819 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1821 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1822 if (parent
->d_op
->d_compare(parent
, *inode
,
1827 if (dentry_cmp(tname
, tlen
, str
, len
))
1831 * No extra seqcount check is required after the name
1832 * compare. The caller must perform a seqcount check in
1833 * order to do anything useful with the returned dentry
1844 * d_lookup - search for a dentry
1845 * @parent: parent dentry
1846 * @name: qstr of name we wish to find
1847 * Returns: dentry, or NULL
1849 * d_lookup searches the children of the parent dentry for the name in
1850 * question. If the dentry is found its reference count is incremented and the
1851 * dentry is returned. The caller must use dput to free the entry when it has
1852 * finished using it. %NULL is returned if the dentry does not exist.
1854 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1856 struct dentry
*dentry
;
1860 seq
= read_seqbegin(&rename_lock
);
1861 dentry
= __d_lookup(parent
, name
);
1864 } while (read_seqretry(&rename_lock
, seq
));
1867 EXPORT_SYMBOL(d_lookup
);
1870 * __d_lookup - search for a dentry (racy)
1871 * @parent: parent dentry
1872 * @name: qstr of name we wish to find
1873 * Returns: dentry, or NULL
1875 * __d_lookup is like d_lookup, however it may (rarely) return a
1876 * false-negative result due to unrelated rename activity.
1878 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1879 * however it must be used carefully, eg. with a following d_lookup in
1880 * the case of failure.
1882 * __d_lookup callers must be commented.
1884 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1886 unsigned int len
= name
->len
;
1887 unsigned int hash
= name
->hash
;
1888 const unsigned char *str
= name
->name
;
1889 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1890 struct hlist_bl_node
*node
;
1891 struct dentry
*found
= NULL
;
1892 struct dentry
*dentry
;
1895 * Note: There is significant duplication with __d_lookup_rcu which is
1896 * required to prevent single threaded performance regressions
1897 * especially on architectures where smp_rmb (in seqcounts) are costly.
1898 * Keep the two functions in sync.
1902 * The hash list is protected using RCU.
1904 * Take d_lock when comparing a candidate dentry, to avoid races
1907 * It is possible that concurrent renames can mess up our list
1908 * walk here and result in missing our dentry, resulting in the
1909 * false-negative result. d_lookup() protects against concurrent
1910 * renames using rename_lock seqlock.
1912 * See Documentation/filesystems/path-lookup.txt for more details.
1916 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1920 if (dentry
->d_name
.hash
!= hash
)
1923 spin_lock(&dentry
->d_lock
);
1924 if (dentry
->d_parent
!= parent
)
1926 if (d_unhashed(dentry
))
1930 * It is safe to compare names since d_move() cannot
1931 * change the qstr (protected by d_lock).
1933 tlen
= dentry
->d_name
.len
;
1934 tname
= dentry
->d_name
.name
;
1935 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1936 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1937 dentry
, dentry
->d_inode
,
1941 if (dentry_cmp(tname
, tlen
, str
, len
))
1947 spin_unlock(&dentry
->d_lock
);
1950 spin_unlock(&dentry
->d_lock
);
1958 * d_hash_and_lookup - hash the qstr then search for a dentry
1959 * @dir: Directory to search in
1960 * @name: qstr of name we wish to find
1962 * On hash failure or on lookup failure NULL is returned.
1964 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1966 struct dentry
*dentry
= NULL
;
1969 * Check for a fs-specific hash function. Note that we must
1970 * calculate the standard hash first, as the d_op->d_hash()
1971 * routine may choose to leave the hash value unchanged.
1973 name
->hash
= full_name_hash(name
->name
, name
->len
);
1974 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1975 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1978 dentry
= d_lookup(dir
, name
);
1984 * d_validate - verify dentry provided from insecure source (deprecated)
1985 * @dentry: The dentry alleged to be valid child of @dparent
1986 * @dparent: The parent dentry (known to be valid)
1988 * An insecure source has sent us a dentry, here we verify it and dget() it.
1989 * This is used by ncpfs in its readdir implementation.
1990 * Zero is returned in the dentry is invalid.
1992 * This function is slow for big directories, and deprecated, do not use it.
1994 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1996 struct dentry
*child
;
1998 spin_lock(&dparent
->d_lock
);
1999 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
2000 if (dentry
== child
) {
2001 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2002 __dget_dlock(dentry
);
2003 spin_unlock(&dentry
->d_lock
);
2004 spin_unlock(&dparent
->d_lock
);
2008 spin_unlock(&dparent
->d_lock
);
2012 EXPORT_SYMBOL(d_validate
);
2015 * When a file is deleted, we have two options:
2016 * - turn this dentry into a negative dentry
2017 * - unhash this dentry and free it.
2019 * Usually, we want to just turn this into
2020 * a negative dentry, but if anybody else is
2021 * currently using the dentry or the inode
2022 * we can't do that and we fall back on removing
2023 * it from the hash queues and waiting for
2024 * it to be deleted later when it has no users
2028 * d_delete - delete a dentry
2029 * @dentry: The dentry to delete
2031 * Turn the dentry into a negative dentry if possible, otherwise
2032 * remove it from the hash queues so it can be deleted later
2035 void d_delete(struct dentry
* dentry
)
2037 struct inode
*inode
;
2040 * Are we the only user?
2043 spin_lock(&dentry
->d_lock
);
2044 inode
= dentry
->d_inode
;
2045 isdir
= S_ISDIR(inode
->i_mode
);
2046 if (dentry
->d_count
== 1) {
2047 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
2048 spin_unlock(&dentry
->d_lock
);
2052 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2053 dentry_unlink_inode(dentry
);
2054 fsnotify_nameremove(dentry
, isdir
);
2058 if (!d_unhashed(dentry
))
2061 spin_unlock(&dentry
->d_lock
);
2063 fsnotify_nameremove(dentry
, isdir
);
2065 EXPORT_SYMBOL(d_delete
);
2067 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2069 BUG_ON(!d_unhashed(entry
));
2071 entry
->d_flags
|= DCACHE_RCUACCESS
;
2072 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2076 static void _d_rehash(struct dentry
* entry
)
2078 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2082 * d_rehash - add an entry back to the hash
2083 * @entry: dentry to add to the hash
2085 * Adds a dentry to the hash according to its name.
2088 void d_rehash(struct dentry
* entry
)
2090 spin_lock(&entry
->d_lock
);
2092 spin_unlock(&entry
->d_lock
);
2094 EXPORT_SYMBOL(d_rehash
);
2097 * dentry_update_name_case - update case insensitive dentry with a new name
2098 * @dentry: dentry to be updated
2101 * Update a case insensitive dentry with new case of name.
2103 * dentry must have been returned by d_lookup with name @name. Old and new
2104 * name lengths must match (ie. no d_compare which allows mismatched name
2107 * Parent inode i_mutex must be held over d_lookup and into this call (to
2108 * keep renames and concurrent inserts, and readdir(2) away).
2110 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2112 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2113 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2115 spin_lock(&dentry
->d_lock
);
2116 write_seqcount_begin(&dentry
->d_seq
);
2117 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2118 write_seqcount_end(&dentry
->d_seq
);
2119 spin_unlock(&dentry
->d_lock
);
2121 EXPORT_SYMBOL(dentry_update_name_case
);
2123 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2125 if (dname_external(target
)) {
2126 if (dname_external(dentry
)) {
2128 * Both external: swap the pointers
2130 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2133 * dentry:internal, target:external. Steal target's
2134 * storage and make target internal.
2136 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2137 dentry
->d_name
.len
+ 1);
2138 dentry
->d_name
.name
= target
->d_name
.name
;
2139 target
->d_name
.name
= target
->d_iname
;
2142 if (dname_external(dentry
)) {
2144 * dentry:external, target:internal. Give dentry's
2145 * storage to target and make dentry internal
2147 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2148 target
->d_name
.len
+ 1);
2149 target
->d_name
.name
= dentry
->d_name
.name
;
2150 dentry
->d_name
.name
= dentry
->d_iname
;
2153 * Both are internal. Just copy target to dentry
2155 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2156 target
->d_name
.len
+ 1);
2157 dentry
->d_name
.len
= target
->d_name
.len
;
2161 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2164 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2167 * XXXX: do we really need to take target->d_lock?
2169 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2170 spin_lock(&target
->d_parent
->d_lock
);
2172 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2173 spin_lock(&dentry
->d_parent
->d_lock
);
2174 spin_lock_nested(&target
->d_parent
->d_lock
,
2175 DENTRY_D_LOCK_NESTED
);
2177 spin_lock(&target
->d_parent
->d_lock
);
2178 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2179 DENTRY_D_LOCK_NESTED
);
2182 if (target
< dentry
) {
2183 spin_lock_nested(&target
->d_lock
, 2);
2184 spin_lock_nested(&dentry
->d_lock
, 3);
2186 spin_lock_nested(&dentry
->d_lock
, 2);
2187 spin_lock_nested(&target
->d_lock
, 3);
2191 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2192 struct dentry
*target
)
2194 if (target
->d_parent
!= dentry
->d_parent
)
2195 spin_unlock(&dentry
->d_parent
->d_lock
);
2196 if (target
->d_parent
!= target
)
2197 spin_unlock(&target
->d_parent
->d_lock
);
2201 * When switching names, the actual string doesn't strictly have to
2202 * be preserved in the target - because we're dropping the target
2203 * anyway. As such, we can just do a simple memcpy() to copy over
2204 * the new name before we switch.
2206 * Note that we have to be a lot more careful about getting the hash
2207 * switched - we have to switch the hash value properly even if it
2208 * then no longer matches the actual (corrupted) string of the target.
2209 * The hash value has to match the hash queue that the dentry is on..
2212 * __d_move - move a dentry
2213 * @dentry: entry to move
2214 * @target: new dentry
2216 * Update the dcache to reflect the move of a file name. Negative
2217 * dcache entries should not be moved in this way. Caller must hold
2218 * rename_lock, the i_mutex of the source and target directories,
2219 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2221 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2223 if (!dentry
->d_inode
)
2224 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2226 BUG_ON(d_ancestor(dentry
, target
));
2227 BUG_ON(d_ancestor(target
, dentry
));
2229 dentry_lock_for_move(dentry
, target
);
2231 write_seqcount_begin(&dentry
->d_seq
);
2232 write_seqcount_begin(&target
->d_seq
);
2234 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2237 * Move the dentry to the target hash queue. Don't bother checking
2238 * for the same hash queue because of how unlikely it is.
2241 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2243 /* Unhash the target: dput() will then get rid of it */
2246 list_del(&dentry
->d_u
.d_child
);
2247 list_del(&target
->d_u
.d_child
);
2249 /* Switch the names.. */
2250 switch_names(dentry
, target
);
2251 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2253 /* ... and switch the parents */
2254 if (IS_ROOT(dentry
)) {
2255 dentry
->d_parent
= target
->d_parent
;
2256 target
->d_parent
= target
;
2257 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2259 swap(dentry
->d_parent
, target
->d_parent
);
2261 /* And add them back to the (new) parent lists */
2262 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2265 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2267 write_seqcount_end(&target
->d_seq
);
2268 write_seqcount_end(&dentry
->d_seq
);
2270 dentry_unlock_parents_for_move(dentry
, target
);
2271 spin_unlock(&target
->d_lock
);
2272 fsnotify_d_move(dentry
);
2273 spin_unlock(&dentry
->d_lock
);
2277 * d_move - move a dentry
2278 * @dentry: entry to move
2279 * @target: new dentry
2281 * Update the dcache to reflect the move of a file name. Negative
2282 * dcache entries should not be moved in this way. See the locking
2283 * requirements for __d_move.
2285 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2287 write_seqlock(&rename_lock
);
2288 __d_move(dentry
, target
);
2289 write_sequnlock(&rename_lock
);
2291 EXPORT_SYMBOL(d_move
);
2294 * d_ancestor - search for an ancestor
2295 * @p1: ancestor dentry
2298 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2299 * an ancestor of p2, else NULL.
2301 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2305 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2306 if (p
->d_parent
== p1
)
2313 * This helper attempts to cope with remotely renamed directories
2315 * It assumes that the caller is already holding
2316 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2318 * Note: If ever the locking in lock_rename() changes, then please
2319 * remember to update this too...
2321 static struct dentry
*__d_unalias(struct inode
*inode
,
2322 struct dentry
*dentry
, struct dentry
*alias
)
2324 struct mutex
*m1
= NULL
, *m2
= NULL
;
2327 /* If alias and dentry share a parent, then no extra locks required */
2328 if (alias
->d_parent
== dentry
->d_parent
)
2331 /* See lock_rename() */
2332 ret
= ERR_PTR(-EBUSY
);
2333 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2335 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2336 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2338 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2340 __d_move(alias
, dentry
);
2343 spin_unlock(&inode
->i_lock
);
2352 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2353 * named dentry in place of the dentry to be replaced.
2354 * returns with anon->d_lock held!
2356 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2358 struct dentry
*dparent
, *aparent
;
2360 dentry_lock_for_move(anon
, dentry
);
2362 write_seqcount_begin(&dentry
->d_seq
);
2363 write_seqcount_begin(&anon
->d_seq
);
2365 dparent
= dentry
->d_parent
;
2366 aparent
= anon
->d_parent
;
2368 switch_names(dentry
, anon
);
2369 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2371 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2372 list_del(&dentry
->d_u
.d_child
);
2373 if (!IS_ROOT(dentry
))
2374 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2376 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2378 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2379 list_del(&anon
->d_u
.d_child
);
2381 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2383 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2385 write_seqcount_end(&dentry
->d_seq
);
2386 write_seqcount_end(&anon
->d_seq
);
2388 dentry_unlock_parents_for_move(anon
, dentry
);
2389 spin_unlock(&dentry
->d_lock
);
2391 /* anon->d_lock still locked, returns locked */
2392 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2396 * d_materialise_unique - introduce an inode into the tree
2397 * @dentry: candidate dentry
2398 * @inode: inode to bind to the dentry, to which aliases may be attached
2400 * Introduces an dentry into the tree, substituting an extant disconnected
2401 * root directory alias in its place if there is one. Caller must hold the
2402 * i_mutex of the parent directory.
2404 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2406 struct dentry
*actual
;
2408 BUG_ON(!d_unhashed(dentry
));
2412 __d_instantiate(dentry
, NULL
);
2417 spin_lock(&inode
->i_lock
);
2419 if (S_ISDIR(inode
->i_mode
)) {
2420 struct dentry
*alias
;
2422 /* Does an aliased dentry already exist? */
2423 alias
= __d_find_alias(inode
, 0);
2426 write_seqlock(&rename_lock
);
2428 if (d_ancestor(alias
, dentry
)) {
2429 /* Check for loops */
2430 actual
= ERR_PTR(-ELOOP
);
2431 } else if (IS_ROOT(alias
)) {
2432 /* Is this an anonymous mountpoint that we
2433 * could splice into our tree? */
2434 __d_materialise_dentry(dentry
, alias
);
2435 write_sequnlock(&rename_lock
);
2439 /* Nope, but we must(!) avoid directory
2441 actual
= __d_unalias(inode
, dentry
, alias
);
2443 write_sequnlock(&rename_lock
);
2444 if (IS_ERR(actual
)) {
2445 if (PTR_ERR(actual
) == -ELOOP
)
2446 pr_warn_ratelimited(
2447 "VFS: Lookup of '%s' in %s %s"
2448 " would have caused loop\n",
2449 dentry
->d_name
.name
,
2450 inode
->i_sb
->s_type
->name
,
2458 /* Add a unique reference */
2459 actual
= __d_instantiate_unique(dentry
, inode
);
2463 BUG_ON(!d_unhashed(actual
));
2465 spin_lock(&actual
->d_lock
);
2468 spin_unlock(&actual
->d_lock
);
2469 spin_unlock(&inode
->i_lock
);
2471 if (actual
== dentry
) {
2472 security_d_instantiate(dentry
, inode
);
2479 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2481 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2485 return -ENAMETOOLONG
;
2487 memcpy(*buffer
, str
, namelen
);
2491 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2493 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2497 * prepend_path - Prepend path string to a buffer
2498 * @path: the dentry/vfsmount to report
2499 * @root: root vfsmnt/dentry
2500 * @buffer: pointer to the end of the buffer
2501 * @buflen: pointer to buffer length
2503 * Caller holds the rename_lock.
2505 static int prepend_path(const struct path
*path
,
2506 const struct path
*root
,
2507 char **buffer
, int *buflen
)
2509 struct dentry
*dentry
= path
->dentry
;
2510 struct vfsmount
*vfsmnt
= path
->mnt
;
2511 struct mount
*mnt
= real_mount(vfsmnt
);
2515 br_read_lock(vfsmount_lock
);
2516 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2517 struct dentry
* parent
;
2519 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2521 if (!mnt_has_parent(mnt
))
2523 dentry
= mnt
->mnt_mountpoint
;
2524 mnt
= mnt
->mnt_parent
;
2528 parent
= dentry
->d_parent
;
2530 spin_lock(&dentry
->d_lock
);
2531 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2532 spin_unlock(&dentry
->d_lock
);
2534 error
= prepend(buffer
, buflen
, "/", 1);
2542 if (!error
&& !slash
)
2543 error
= prepend(buffer
, buflen
, "/", 1);
2546 br_read_unlock(vfsmount_lock
);
2551 * Filesystems needing to implement special "root names"
2552 * should do so with ->d_dname()
2554 if (IS_ROOT(dentry
) &&
2555 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2556 WARN(1, "Root dentry has weird name <%.*s>\n",
2557 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2560 error
= prepend(buffer
, buflen
, "/", 1);
2562 error
= real_mount(vfsmnt
)->mnt_ns
? 1 : 2;
2567 * __d_path - return the path of a dentry
2568 * @path: the dentry/vfsmount to report
2569 * @root: root vfsmnt/dentry
2570 * @buf: buffer to return value in
2571 * @buflen: buffer length
2573 * Convert a dentry into an ASCII path name.
2575 * Returns a pointer into the buffer or an error code if the
2576 * path was too long.
2578 * "buflen" should be positive.
2580 * If the path is not reachable from the supplied root, return %NULL.
2582 char *__d_path(const struct path
*path
,
2583 const struct path
*root
,
2584 char *buf
, int buflen
)
2586 char *res
= buf
+ buflen
;
2589 prepend(&res
, &buflen
, "\0", 1);
2590 write_seqlock(&rename_lock
);
2591 error
= prepend_path(path
, root
, &res
, &buflen
);
2592 write_sequnlock(&rename_lock
);
2595 return ERR_PTR(error
);
2601 char *d_absolute_path(const struct path
*path
,
2602 char *buf
, int buflen
)
2604 struct path root
= {};
2605 char *res
= buf
+ buflen
;
2608 prepend(&res
, &buflen
, "\0", 1);
2609 write_seqlock(&rename_lock
);
2610 error
= prepend_path(path
, &root
, &res
, &buflen
);
2611 write_sequnlock(&rename_lock
);
2616 return ERR_PTR(error
);
2621 * same as __d_path but appends "(deleted)" for unlinked files.
2623 static int path_with_deleted(const struct path
*path
,
2624 const struct path
*root
,
2625 char **buf
, int *buflen
)
2627 prepend(buf
, buflen
, "\0", 1);
2628 if (d_unlinked(path
->dentry
)) {
2629 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2634 return prepend_path(path
, root
, buf
, buflen
);
2637 static int prepend_unreachable(char **buffer
, int *buflen
)
2639 return prepend(buffer
, buflen
, "(unreachable)", 13);
2643 * d_path - return the path of a dentry
2644 * @path: path to report
2645 * @buf: buffer to return value in
2646 * @buflen: buffer length
2648 * Convert a dentry into an ASCII path name. If the entry has been deleted
2649 * the string " (deleted)" is appended. Note that this is ambiguous.
2651 * Returns a pointer into the buffer or an error code if the path was
2652 * too long. Note: Callers should use the returned pointer, not the passed
2653 * in buffer, to use the name! The implementation often starts at an offset
2654 * into the buffer, and may leave 0 bytes at the start.
2656 * "buflen" should be positive.
2658 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2660 char *res
= buf
+ buflen
;
2665 * We have various synthetic filesystems that never get mounted. On
2666 * these filesystems dentries are never used for lookup purposes, and
2667 * thus don't need to be hashed. They also don't need a name until a
2668 * user wants to identify the object in /proc/pid/fd/. The little hack
2669 * below allows us to generate a name for these objects on demand:
2671 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2672 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2674 get_fs_root(current
->fs
, &root
);
2675 write_seqlock(&rename_lock
);
2676 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2678 res
= ERR_PTR(error
);
2679 write_sequnlock(&rename_lock
);
2683 EXPORT_SYMBOL(d_path
);
2686 * d_path_with_unreachable - return the path of a dentry
2687 * @path: path to report
2688 * @buf: buffer to return value in
2689 * @buflen: buffer length
2691 * The difference from d_path() is that this prepends "(unreachable)"
2692 * to paths which are unreachable from the current process' root.
2694 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2696 char *res
= buf
+ buflen
;
2700 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2701 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2703 get_fs_root(current
->fs
, &root
);
2704 write_seqlock(&rename_lock
);
2705 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2707 error
= prepend_unreachable(&res
, &buflen
);
2708 write_sequnlock(&rename_lock
);
2711 res
= ERR_PTR(error
);
2717 * Helper function for dentry_operations.d_dname() members
2719 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2720 const char *fmt
, ...)
2726 va_start(args
, fmt
);
2727 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2730 if (sz
> sizeof(temp
) || sz
> buflen
)
2731 return ERR_PTR(-ENAMETOOLONG
);
2733 buffer
+= buflen
- sz
;
2734 return memcpy(buffer
, temp
, sz
);
2738 * Write full pathname from the root of the filesystem into the buffer.
2740 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2742 char *end
= buf
+ buflen
;
2745 prepend(&end
, &buflen
, "\0", 1);
2752 while (!IS_ROOT(dentry
)) {
2753 struct dentry
*parent
= dentry
->d_parent
;
2757 spin_lock(&dentry
->d_lock
);
2758 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2759 spin_unlock(&dentry
->d_lock
);
2760 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2768 return ERR_PTR(-ENAMETOOLONG
);
2771 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2775 write_seqlock(&rename_lock
);
2776 retval
= __dentry_path(dentry
, buf
, buflen
);
2777 write_sequnlock(&rename_lock
);
2781 EXPORT_SYMBOL(dentry_path_raw
);
2783 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2788 write_seqlock(&rename_lock
);
2789 if (d_unlinked(dentry
)) {
2791 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2795 retval
= __dentry_path(dentry
, buf
, buflen
);
2796 write_sequnlock(&rename_lock
);
2797 if (!IS_ERR(retval
) && p
)
2798 *p
= '/'; /* restore '/' overriden with '\0' */
2801 return ERR_PTR(-ENAMETOOLONG
);
2805 * NOTE! The user-level library version returns a
2806 * character pointer. The kernel system call just
2807 * returns the length of the buffer filled (which
2808 * includes the ending '\0' character), or a negative
2809 * error value. So libc would do something like
2811 * char *getcwd(char * buf, size_t size)
2815 * retval = sys_getcwd(buf, size);
2822 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2825 struct path pwd
, root
;
2826 char *page
= (char *) __get_free_page(GFP_USER
);
2831 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2834 write_seqlock(&rename_lock
);
2835 if (!d_unlinked(pwd
.dentry
)) {
2837 char *cwd
= page
+ PAGE_SIZE
;
2838 int buflen
= PAGE_SIZE
;
2840 prepend(&cwd
, &buflen
, "\0", 1);
2841 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2842 write_sequnlock(&rename_lock
);
2847 /* Unreachable from current root */
2849 error
= prepend_unreachable(&cwd
, &buflen
);
2855 len
= PAGE_SIZE
+ page
- cwd
;
2858 if (copy_to_user(buf
, cwd
, len
))
2862 write_sequnlock(&rename_lock
);
2868 free_page((unsigned long) page
);
2873 * Test whether new_dentry is a subdirectory of old_dentry.
2875 * Trivially implemented using the dcache structure
2879 * is_subdir - is new dentry a subdirectory of old_dentry
2880 * @new_dentry: new dentry
2881 * @old_dentry: old dentry
2883 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2884 * Returns 0 otherwise.
2885 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2888 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2893 if (new_dentry
== old_dentry
)
2897 /* for restarting inner loop in case of seq retry */
2898 seq
= read_seqbegin(&rename_lock
);
2900 * Need rcu_readlock to protect against the d_parent trashing
2904 if (d_ancestor(old_dentry
, new_dentry
))
2909 } while (read_seqretry(&rename_lock
, seq
));
2914 void d_genocide(struct dentry
*root
)
2916 struct dentry
*this_parent
;
2917 struct list_head
*next
;
2921 seq
= read_seqbegin(&rename_lock
);
2924 spin_lock(&this_parent
->d_lock
);
2926 next
= this_parent
->d_subdirs
.next
;
2928 while (next
!= &this_parent
->d_subdirs
) {
2929 struct list_head
*tmp
= next
;
2930 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2933 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2934 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2935 spin_unlock(&dentry
->d_lock
);
2938 if (!list_empty(&dentry
->d_subdirs
)) {
2939 spin_unlock(&this_parent
->d_lock
);
2940 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2941 this_parent
= dentry
;
2942 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2945 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2946 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2949 spin_unlock(&dentry
->d_lock
);
2951 if (this_parent
!= root
) {
2952 struct dentry
*child
= this_parent
;
2953 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2954 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2955 this_parent
->d_count
--;
2957 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2960 next
= child
->d_u
.d_child
.next
;
2963 spin_unlock(&this_parent
->d_lock
);
2964 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2967 write_sequnlock(&rename_lock
);
2972 write_seqlock(&rename_lock
);
2977 * find_inode_number - check for dentry with name
2978 * @dir: directory to check
2979 * @name: Name to find.
2981 * Check whether a dentry already exists for the given name,
2982 * and return the inode number if it has an inode. Otherwise
2985 * This routine is used to post-process directory listings for
2986 * filesystems using synthetic inode numbers, and is necessary
2987 * to keep getcwd() working.
2990 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2992 struct dentry
* dentry
;
2995 dentry
= d_hash_and_lookup(dir
, name
);
2997 if (dentry
->d_inode
)
2998 ino
= dentry
->d_inode
->i_ino
;
3003 EXPORT_SYMBOL(find_inode_number
);
3005 static __initdata
unsigned long dhash_entries
;
3006 static int __init
set_dhash_entries(char *str
)
3010 dhash_entries
= simple_strtoul(str
, &str
, 0);
3013 __setup("dhash_entries=", set_dhash_entries
);
3015 static void __init
dcache_init_early(void)
3019 /* If hashes are distributed across NUMA nodes, defer
3020 * hash allocation until vmalloc space is available.
3026 alloc_large_system_hash("Dentry cache",
3027 sizeof(struct hlist_bl_head
),
3035 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3036 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3039 static void __init
dcache_init(void)
3044 * A constructor could be added for stable state like the lists,
3045 * but it is probably not worth it because of the cache nature
3048 dentry_cache
= KMEM_CACHE(dentry
,
3049 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3051 /* Hash may have been set up in dcache_init_early */
3056 alloc_large_system_hash("Dentry cache",
3057 sizeof(struct hlist_bl_head
),
3065 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3066 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3069 /* SLAB cache for __getname() consumers */
3070 struct kmem_cache
*names_cachep __read_mostly
;
3071 EXPORT_SYMBOL(names_cachep
);
3073 EXPORT_SYMBOL(d_genocide
);
3075 void __init
vfs_caches_init_early(void)
3077 dcache_init_early();
3081 void __init
vfs_caches_init(unsigned long mempages
)
3083 unsigned long reserve
;
3085 /* Base hash sizes on available memory, with a reserve equal to
3086 150% of current kernel size */
3088 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3089 mempages
-= reserve
;
3091 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3092 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3096 files_init(mempages
);