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(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 /* 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
);
140 static void __d_free(struct rcu_head
*head
)
142 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
144 WARN_ON(!list_empty(&dentry
->d_alias
));
145 if (dname_external(dentry
))
146 kfree(dentry
->d_name
.name
);
147 kmem_cache_free(dentry_cache
, dentry
);
153 static void d_free(struct dentry
*dentry
)
155 BUG_ON(dentry
->d_count
);
156 this_cpu_dec(nr_dentry
);
157 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
158 dentry
->d_op
->d_release(dentry
);
160 /* if dentry was never visible to RCU, immediate free is OK */
161 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
162 __d_free(&dentry
->d_u
.d_rcu
);
164 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
168 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
169 * @dentry: the target dentry
170 * After this call, in-progress rcu-walk path lookup will fail. This
171 * should be called after unhashing, and after changing d_inode (if
172 * the dentry has not already been unhashed).
174 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
176 assert_spin_locked(&dentry
->d_lock
);
177 /* Go through a barrier */
178 write_seqcount_barrier(&dentry
->d_seq
);
182 * Release the dentry's inode, using the filesystem
183 * d_iput() operation if defined. Dentry has no refcount
186 static void dentry_iput(struct dentry
* dentry
)
187 __releases(dentry
->d_lock
)
188 __releases(dentry
->d_inode
->i_lock
)
190 struct inode
*inode
= dentry
->d_inode
;
192 dentry
->d_inode
= NULL
;
193 list_del_init(&dentry
->d_alias
);
194 spin_unlock(&dentry
->d_lock
);
195 spin_unlock(&inode
->i_lock
);
197 fsnotify_inoderemove(inode
);
198 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
199 dentry
->d_op
->d_iput(dentry
, inode
);
203 spin_unlock(&dentry
->d_lock
);
208 * Release the dentry's inode, using the filesystem
209 * d_iput() operation if defined. dentry remains in-use.
211 static void dentry_unlink_inode(struct dentry
* dentry
)
212 __releases(dentry
->d_lock
)
213 __releases(dentry
->d_inode
->i_lock
)
215 struct inode
*inode
= dentry
->d_inode
;
216 dentry
->d_inode
= NULL
;
217 list_del_init(&dentry
->d_alias
);
218 dentry_rcuwalk_barrier(dentry
);
219 spin_unlock(&dentry
->d_lock
);
220 spin_unlock(&inode
->i_lock
);
222 fsnotify_inoderemove(inode
);
223 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
224 dentry
->d_op
->d_iput(dentry
, inode
);
230 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
232 static void dentry_lru_add(struct dentry
*dentry
)
234 if (list_empty(&dentry
->d_lru
)) {
235 spin_lock(&dcache_lru_lock
);
236 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
237 dentry
->d_sb
->s_nr_dentry_unused
++;
238 dentry_stat
.nr_unused
++;
239 spin_unlock(&dcache_lru_lock
);
243 static void __dentry_lru_del(struct dentry
*dentry
)
245 list_del_init(&dentry
->d_lru
);
246 dentry
->d_flags
&= ~DCACHE_SHRINK_LIST
;
247 dentry
->d_sb
->s_nr_dentry_unused
--;
248 dentry_stat
.nr_unused
--;
252 * Remove a dentry with references from the LRU.
254 static void dentry_lru_del(struct dentry
*dentry
)
256 if (!list_empty(&dentry
->d_lru
)) {
257 spin_lock(&dcache_lru_lock
);
258 __dentry_lru_del(dentry
);
259 spin_unlock(&dcache_lru_lock
);
264 * Remove a dentry that is unreferenced and about to be pruned
265 * (unhashed and destroyed) from the LRU, and inform the file system.
266 * This wrapper should be called _prior_ to unhashing a victim dentry.
268 static void dentry_lru_prune(struct dentry
*dentry
)
270 if (!list_empty(&dentry
->d_lru
)) {
271 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
272 dentry
->d_op
->d_prune(dentry
);
274 spin_lock(&dcache_lru_lock
);
275 __dentry_lru_del(dentry
);
276 spin_unlock(&dcache_lru_lock
);
280 static void dentry_lru_move_list(struct dentry
*dentry
, struct list_head
*list
)
282 spin_lock(&dcache_lru_lock
);
283 if (list_empty(&dentry
->d_lru
)) {
284 list_add_tail(&dentry
->d_lru
, list
);
285 dentry
->d_sb
->s_nr_dentry_unused
++;
286 dentry_stat
.nr_unused
++;
288 list_move_tail(&dentry
->d_lru
, list
);
290 spin_unlock(&dcache_lru_lock
);
294 * d_kill - kill dentry and return parent
295 * @dentry: dentry to kill
296 * @parent: parent dentry
298 * The dentry must already be unhashed and removed from the LRU.
300 * If this is the root of the dentry tree, return NULL.
302 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
305 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
306 __releases(dentry
->d_lock
)
307 __releases(parent
->d_lock
)
308 __releases(dentry
->d_inode
->i_lock
)
310 list_del(&dentry
->d_u
.d_child
);
312 * Inform try_to_ascend() that we are no longer attached to the
315 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
317 spin_unlock(&parent
->d_lock
);
320 * dentry_iput drops the locks, at which point nobody (except
321 * transient RCU lookups) can reach this dentry.
328 * Unhash a dentry without inserting an RCU walk barrier or checking that
329 * dentry->d_lock is locked. The caller must take care of that, if
332 static void __d_shrink(struct dentry
*dentry
)
334 if (!d_unhashed(dentry
)) {
335 struct hlist_bl_head
*b
;
336 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
337 b
= &dentry
->d_sb
->s_anon
;
339 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
342 __hlist_bl_del(&dentry
->d_hash
);
343 dentry
->d_hash
.pprev
= NULL
;
349 * d_drop - drop a dentry
350 * @dentry: dentry to drop
352 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
353 * be found through a VFS lookup any more. Note that this is different from
354 * deleting the dentry - d_delete will try to mark the dentry negative if
355 * possible, giving a successful _negative_ lookup, while d_drop will
356 * just make the cache lookup fail.
358 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
359 * reason (NFS timeouts or autofs deletes).
361 * __d_drop requires dentry->d_lock.
363 void __d_drop(struct dentry
*dentry
)
365 if (!d_unhashed(dentry
)) {
367 dentry_rcuwalk_barrier(dentry
);
370 EXPORT_SYMBOL(__d_drop
);
372 void d_drop(struct dentry
*dentry
)
374 spin_lock(&dentry
->d_lock
);
376 spin_unlock(&dentry
->d_lock
);
378 EXPORT_SYMBOL(d_drop
);
381 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
382 * @dentry: dentry to drop
384 * This is called when we do a lookup on a placeholder dentry that needed to be
385 * looked up. The dentry should have been hashed in order for it to be found by
386 * the lookup code, but now needs to be unhashed while we do the actual lookup
387 * and clear the DCACHE_NEED_LOOKUP flag.
389 void d_clear_need_lookup(struct dentry
*dentry
)
391 spin_lock(&dentry
->d_lock
);
393 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
394 spin_unlock(&dentry
->d_lock
);
396 EXPORT_SYMBOL(d_clear_need_lookup
);
399 * Finish off a dentry we've decided to kill.
400 * dentry->d_lock must be held, returns with it unlocked.
401 * If ref is non-zero, then decrement the refcount too.
402 * Returns dentry requiring refcount drop, or NULL if we're done.
404 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
405 __releases(dentry
->d_lock
)
408 struct dentry
*parent
;
410 inode
= dentry
->d_inode
;
411 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
413 spin_unlock(&dentry
->d_lock
);
415 return dentry
; /* try again with same dentry */
420 parent
= dentry
->d_parent
;
421 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
423 spin_unlock(&inode
->i_lock
);
430 * if dentry was on the d_lru list delete it from there.
431 * inform the fs via d_prune that this dentry is about to be
432 * unhashed and destroyed.
434 dentry_lru_prune(dentry
);
435 /* if it was on the hash then remove it */
437 return d_kill(dentry
, parent
);
443 * This is complicated by the fact that we do not want to put
444 * dentries that are no longer on any hash chain on the unused
445 * list: we'd much rather just get rid of them immediately.
447 * However, that implies that we have to traverse the dentry
448 * tree upwards to the parents which might _also_ now be
449 * scheduled for deletion (it may have been only waiting for
450 * its last child to go away).
452 * This tail recursion is done by hand as we don't want to depend
453 * on the compiler to always get this right (gcc generally doesn't).
454 * Real recursion would eat up our stack space.
458 * dput - release a dentry
459 * @dentry: dentry to release
461 * Release a dentry. This will drop the usage count and if appropriate
462 * call the dentry unlink method as well as removing it from the queues and
463 * releasing its resources. If the parent dentries were scheduled for release
464 * they too may now get deleted.
466 void dput(struct dentry
*dentry
)
472 if (dentry
->d_count
== 1)
474 spin_lock(&dentry
->d_lock
);
475 BUG_ON(!dentry
->d_count
);
476 if (dentry
->d_count
> 1) {
478 spin_unlock(&dentry
->d_lock
);
482 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
483 if (dentry
->d_op
->d_delete(dentry
))
487 /* Unreachable? Get rid of it */
488 if (d_unhashed(dentry
))
492 * If this dentry needs lookup, don't set the referenced flag so that it
493 * is more likely to be cleaned up by the dcache shrinker in case of
496 if (!d_need_lookup(dentry
))
497 dentry
->d_flags
|= DCACHE_REFERENCED
;
498 dentry_lru_add(dentry
);
501 spin_unlock(&dentry
->d_lock
);
505 dentry
= dentry_kill(dentry
, 1);
512 * d_invalidate - invalidate a dentry
513 * @dentry: dentry to invalidate
515 * Try to invalidate the dentry if it turns out to be
516 * possible. If there are other dentries that can be
517 * reached through this one we can't delete it and we
518 * return -EBUSY. On success we return 0.
523 int d_invalidate(struct dentry
* dentry
)
526 * If it's already been dropped, return OK.
528 spin_lock(&dentry
->d_lock
);
529 if (d_unhashed(dentry
)) {
530 spin_unlock(&dentry
->d_lock
);
534 * Check whether to do a partial shrink_dcache
535 * to get rid of unused child entries.
537 if (!list_empty(&dentry
->d_subdirs
)) {
538 spin_unlock(&dentry
->d_lock
);
539 shrink_dcache_parent(dentry
);
540 spin_lock(&dentry
->d_lock
);
544 * Somebody else still using it?
546 * If it's a directory, we can't drop it
547 * for fear of somebody re-populating it
548 * with children (even though dropping it
549 * would make it unreachable from the root,
550 * we might still populate it if it was a
551 * working directory or similar).
552 * We also need to leave mountpoints alone,
555 if (dentry
->d_count
> 1 && dentry
->d_inode
) {
556 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
557 spin_unlock(&dentry
->d_lock
);
563 spin_unlock(&dentry
->d_lock
);
566 EXPORT_SYMBOL(d_invalidate
);
568 /* This must be called with d_lock held */
569 static inline void __dget_dlock(struct dentry
*dentry
)
574 static inline void __dget(struct dentry
*dentry
)
576 spin_lock(&dentry
->d_lock
);
577 __dget_dlock(dentry
);
578 spin_unlock(&dentry
->d_lock
);
581 struct dentry
*dget_parent(struct dentry
*dentry
)
587 * Don't need rcu_dereference because we re-check it was correct under
591 ret
= dentry
->d_parent
;
592 spin_lock(&ret
->d_lock
);
593 if (unlikely(ret
!= dentry
->d_parent
)) {
594 spin_unlock(&ret
->d_lock
);
599 BUG_ON(!ret
->d_count
);
601 spin_unlock(&ret
->d_lock
);
604 EXPORT_SYMBOL(dget_parent
);
607 * d_find_alias - grab a hashed alias of inode
608 * @inode: inode in question
609 * @want_discon: flag, used by d_splice_alias, to request
610 * that only a DISCONNECTED alias be returned.
612 * If inode has a hashed alias, or is a directory and has any alias,
613 * acquire the reference to alias and return it. Otherwise return NULL.
614 * Notice that if inode is a directory there can be only one alias and
615 * it can be unhashed only if it has no children, or if it is the root
618 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
619 * any other hashed alias over that one unless @want_discon is set,
620 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
622 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
624 struct dentry
*alias
, *discon_alias
;
628 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
629 spin_lock(&alias
->d_lock
);
630 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
631 if (IS_ROOT(alias
) &&
632 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
633 discon_alias
= alias
;
634 } else if (!want_discon
) {
636 spin_unlock(&alias
->d_lock
);
640 spin_unlock(&alias
->d_lock
);
643 alias
= discon_alias
;
644 spin_lock(&alias
->d_lock
);
645 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
646 if (IS_ROOT(alias
) &&
647 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
649 spin_unlock(&alias
->d_lock
);
653 spin_unlock(&alias
->d_lock
);
659 struct dentry
*d_find_alias(struct inode
*inode
)
661 struct dentry
*de
= NULL
;
663 if (!list_empty(&inode
->i_dentry
)) {
664 spin_lock(&inode
->i_lock
);
665 de
= __d_find_alias(inode
, 0);
666 spin_unlock(&inode
->i_lock
);
670 EXPORT_SYMBOL(d_find_alias
);
673 * Try to kill dentries associated with this inode.
674 * WARNING: you must own a reference to inode.
676 void d_prune_aliases(struct inode
*inode
)
678 struct dentry
*dentry
;
680 spin_lock(&inode
->i_lock
);
681 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
682 spin_lock(&dentry
->d_lock
);
683 if (!dentry
->d_count
) {
684 __dget_dlock(dentry
);
686 spin_unlock(&dentry
->d_lock
);
687 spin_unlock(&inode
->i_lock
);
691 spin_unlock(&dentry
->d_lock
);
693 spin_unlock(&inode
->i_lock
);
695 EXPORT_SYMBOL(d_prune_aliases
);
698 * Try to throw away a dentry - free the inode, dput the parent.
699 * Requires dentry->d_lock is held, and dentry->d_count == 0.
700 * Releases dentry->d_lock.
702 * This may fail if locks cannot be acquired no problem, just try again.
704 static void try_prune_one_dentry(struct dentry
*dentry
)
705 __releases(dentry
->d_lock
)
707 struct dentry
*parent
;
709 parent
= dentry_kill(dentry
, 0);
711 * If dentry_kill returns NULL, we have nothing more to do.
712 * if it returns the same dentry, trylocks failed. In either
713 * case, just loop again.
715 * Otherwise, we need to prune ancestors too. This is necessary
716 * to prevent quadratic behavior of shrink_dcache_parent(), but
717 * is also expected to be beneficial in reducing dentry cache
722 if (parent
== dentry
)
725 /* Prune ancestors. */
728 spin_lock(&dentry
->d_lock
);
729 if (dentry
->d_count
> 1) {
731 spin_unlock(&dentry
->d_lock
);
734 dentry
= dentry_kill(dentry
, 1);
738 static void shrink_dentry_list(struct list_head
*list
)
740 struct dentry
*dentry
;
744 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
745 if (&dentry
->d_lru
== list
)
747 spin_lock(&dentry
->d_lock
);
748 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
749 spin_unlock(&dentry
->d_lock
);
754 * We found an inuse dentry which was not removed from
755 * the LRU because of laziness during lookup. Do not free
756 * it - just keep it off the LRU list.
758 if (dentry
->d_count
) {
759 dentry_lru_del(dentry
);
760 spin_unlock(&dentry
->d_lock
);
766 try_prune_one_dentry(dentry
);
774 * prune_dcache_sb - shrink the dcache
776 * @count: number of entries to try to free
778 * Attempt to shrink the superblock dcache LRU by @count entries. This is
779 * done when we need more memory an called from the superblock shrinker
782 * This function may fail to free any resources if all the dentries are in
785 void prune_dcache_sb(struct super_block
*sb
, int count
)
787 struct dentry
*dentry
;
788 LIST_HEAD(referenced
);
792 spin_lock(&dcache_lru_lock
);
793 while (!list_empty(&sb
->s_dentry_lru
)) {
794 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
795 struct dentry
, d_lru
);
796 BUG_ON(dentry
->d_sb
!= sb
);
798 if (!spin_trylock(&dentry
->d_lock
)) {
799 spin_unlock(&dcache_lru_lock
);
804 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
805 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
806 list_move(&dentry
->d_lru
, &referenced
);
807 spin_unlock(&dentry
->d_lock
);
809 list_move_tail(&dentry
->d_lru
, &tmp
);
810 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
811 spin_unlock(&dentry
->d_lock
);
815 cond_resched_lock(&dcache_lru_lock
);
817 if (!list_empty(&referenced
))
818 list_splice(&referenced
, &sb
->s_dentry_lru
);
819 spin_unlock(&dcache_lru_lock
);
821 shrink_dentry_list(&tmp
);
825 * shrink_dcache_sb - shrink dcache for a superblock
828 * Shrink the dcache for the specified super block. This is used to free
829 * the dcache before unmounting a file system.
831 void shrink_dcache_sb(struct super_block
*sb
)
835 spin_lock(&dcache_lru_lock
);
836 while (!list_empty(&sb
->s_dentry_lru
)) {
837 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
838 spin_unlock(&dcache_lru_lock
);
839 shrink_dentry_list(&tmp
);
840 spin_lock(&dcache_lru_lock
);
842 spin_unlock(&dcache_lru_lock
);
844 EXPORT_SYMBOL(shrink_dcache_sb
);
847 * destroy a single subtree of dentries for unmount
848 * - see the comments on shrink_dcache_for_umount() for a description of the
851 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
853 struct dentry
*parent
;
855 BUG_ON(!IS_ROOT(dentry
));
858 /* descend to the first leaf in the current subtree */
859 while (!list_empty(&dentry
->d_subdirs
))
860 dentry
= list_entry(dentry
->d_subdirs
.next
,
861 struct dentry
, d_u
.d_child
);
863 /* consume the dentries from this leaf up through its parents
864 * until we find one with children or run out altogether */
869 * remove the dentry from the lru, and inform
870 * the fs that this dentry is about to be
871 * unhashed and destroyed.
873 dentry_lru_prune(dentry
);
876 if (dentry
->d_count
!= 0) {
878 "BUG: Dentry %p{i=%lx,n=%s}"
880 " [unmount of %s %s]\n",
883 dentry
->d_inode
->i_ino
: 0UL,
886 dentry
->d_sb
->s_type
->name
,
891 if (IS_ROOT(dentry
)) {
893 list_del(&dentry
->d_u
.d_child
);
895 parent
= dentry
->d_parent
;
897 list_del(&dentry
->d_u
.d_child
);
900 inode
= dentry
->d_inode
;
902 dentry
->d_inode
= NULL
;
903 list_del_init(&dentry
->d_alias
);
904 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
905 dentry
->d_op
->d_iput(dentry
, inode
);
912 /* finished when we fall off the top of the tree,
913 * otherwise we ascend to the parent and move to the
914 * next sibling if there is one */
918 } while (list_empty(&dentry
->d_subdirs
));
920 dentry
= list_entry(dentry
->d_subdirs
.next
,
921 struct dentry
, d_u
.d_child
);
926 * destroy the dentries attached to a superblock on unmounting
927 * - we don't need to use dentry->d_lock because:
928 * - the superblock is detached from all mountings and open files, so the
929 * dentry trees will not be rearranged by the VFS
930 * - s_umount is write-locked, so the memory pressure shrinker will ignore
931 * any dentries belonging to this superblock that it comes across
932 * - the filesystem itself is no longer permitted to rearrange the dentries
935 void shrink_dcache_for_umount(struct super_block
*sb
)
937 struct dentry
*dentry
;
939 if (down_read_trylock(&sb
->s_umount
))
945 shrink_dcache_for_umount_subtree(dentry
);
947 while (!hlist_bl_empty(&sb
->s_anon
)) {
948 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
949 shrink_dcache_for_umount_subtree(dentry
);
954 * This tries to ascend one level of parenthood, but
955 * we can race with renaming, so we need to re-check
956 * the parenthood after dropping the lock and check
957 * that the sequence number still matches.
959 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
961 struct dentry
*new = old
->d_parent
;
964 spin_unlock(&old
->d_lock
);
965 spin_lock(&new->d_lock
);
968 * might go back up the wrong parent if we have had a rename
971 if (new != old
->d_parent
||
972 (old
->d_flags
& DCACHE_DISCONNECTED
) ||
973 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
974 spin_unlock(&new->d_lock
);
983 * Search for at least 1 mount point in the dentry's subdirs.
984 * We descend to the next level whenever the d_subdirs
985 * list is non-empty and continue searching.
989 * have_submounts - check for mounts over a dentry
990 * @parent: dentry to check.
992 * Return true if the parent or its subdirectories contain
995 int have_submounts(struct dentry
*parent
)
997 struct dentry
*this_parent
;
998 struct list_head
*next
;
1002 seq
= read_seqbegin(&rename_lock
);
1004 this_parent
= parent
;
1006 if (d_mountpoint(parent
))
1008 spin_lock(&this_parent
->d_lock
);
1010 next
= this_parent
->d_subdirs
.next
;
1012 while (next
!= &this_parent
->d_subdirs
) {
1013 struct list_head
*tmp
= next
;
1014 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1017 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1018 /* Have we found a mount point ? */
1019 if (d_mountpoint(dentry
)) {
1020 spin_unlock(&dentry
->d_lock
);
1021 spin_unlock(&this_parent
->d_lock
);
1024 if (!list_empty(&dentry
->d_subdirs
)) {
1025 spin_unlock(&this_parent
->d_lock
);
1026 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1027 this_parent
= dentry
;
1028 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1031 spin_unlock(&dentry
->d_lock
);
1034 * All done at this level ... ascend and resume the search.
1036 if (this_parent
!= parent
) {
1037 struct dentry
*child
= this_parent
;
1038 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1041 next
= child
->d_u
.d_child
.next
;
1044 spin_unlock(&this_parent
->d_lock
);
1045 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1048 write_sequnlock(&rename_lock
);
1049 return 0; /* No mount points found in tree */
1051 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1054 write_sequnlock(&rename_lock
);
1059 write_seqlock(&rename_lock
);
1062 EXPORT_SYMBOL(have_submounts
);
1065 * Search the dentry child list for the specified parent,
1066 * and move any unused dentries to the end of the unused
1067 * list for prune_dcache(). We descend to the next level
1068 * whenever the d_subdirs list is non-empty and continue
1071 * It returns zero iff there are no unused children,
1072 * otherwise it returns the number of children moved to
1073 * the end of the unused list. This may not be the total
1074 * number of unused children, because select_parent can
1075 * drop the lock and return early due to latency
1078 static int select_parent(struct dentry
*parent
, struct list_head
*dispose
)
1080 struct dentry
*this_parent
;
1081 struct list_head
*next
;
1086 seq
= read_seqbegin(&rename_lock
);
1088 this_parent
= parent
;
1089 spin_lock(&this_parent
->d_lock
);
1091 next
= this_parent
->d_subdirs
.next
;
1093 while (next
!= &this_parent
->d_subdirs
) {
1094 struct list_head
*tmp
= next
;
1095 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1098 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1101 * move only zero ref count dentries to the dispose list.
1103 * Those which are presently on the shrink list, being processed
1104 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1105 * loop in shrink_dcache_parent() might not make any progress
1108 if (dentry
->d_count
) {
1109 dentry_lru_del(dentry
);
1110 } else if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
)) {
1111 dentry_lru_move_list(dentry
, dispose
);
1112 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
1116 * We can return to the caller if we have found some (this
1117 * ensures forward progress). We'll be coming back to find
1120 if (found
&& need_resched()) {
1121 spin_unlock(&dentry
->d_lock
);
1126 * Descend a level if the d_subdirs list is non-empty.
1128 if (!list_empty(&dentry
->d_subdirs
)) {
1129 spin_unlock(&this_parent
->d_lock
);
1130 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1131 this_parent
= dentry
;
1132 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1136 spin_unlock(&dentry
->d_lock
);
1139 * All done at this level ... ascend and resume the search.
1141 if (this_parent
!= parent
) {
1142 struct dentry
*child
= this_parent
;
1143 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1146 next
= child
->d_u
.d_child
.next
;
1150 spin_unlock(&this_parent
->d_lock
);
1151 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1154 write_sequnlock(&rename_lock
);
1161 write_seqlock(&rename_lock
);
1166 * shrink_dcache_parent - prune dcache
1167 * @parent: parent of entries to prune
1169 * Prune the dcache to remove unused children of the parent dentry.
1171 void shrink_dcache_parent(struct dentry
* parent
)
1176 while ((found
= select_parent(parent
, &dispose
)) != 0)
1177 shrink_dentry_list(&dispose
);
1179 EXPORT_SYMBOL(shrink_dcache_parent
);
1182 * __d_alloc - allocate a dcache entry
1183 * @sb: filesystem it will belong to
1184 * @name: qstr of the name
1186 * Allocates a dentry. It returns %NULL if there is insufficient memory
1187 * available. On a success the dentry is returned. The name passed in is
1188 * copied and the copy passed in may be reused after this call.
1191 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1193 struct dentry
*dentry
;
1196 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1200 if (name
->len
> DNAME_INLINE_LEN
-1) {
1201 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1203 kmem_cache_free(dentry_cache
, dentry
);
1207 dname
= dentry
->d_iname
;
1209 dentry
->d_name
.name
= dname
;
1211 dentry
->d_name
.len
= name
->len
;
1212 dentry
->d_name
.hash
= name
->hash
;
1213 memcpy(dname
, name
->name
, name
->len
);
1214 dname
[name
->len
] = 0;
1216 dentry
->d_count
= 1;
1217 dentry
->d_flags
= 0;
1218 spin_lock_init(&dentry
->d_lock
);
1219 seqcount_init(&dentry
->d_seq
);
1220 dentry
->d_inode
= NULL
;
1221 dentry
->d_parent
= dentry
;
1223 dentry
->d_op
= NULL
;
1224 dentry
->d_fsdata
= NULL
;
1225 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1226 INIT_LIST_HEAD(&dentry
->d_lru
);
1227 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1228 INIT_LIST_HEAD(&dentry
->d_alias
);
1229 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1230 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1232 this_cpu_inc(nr_dentry
);
1238 * d_alloc - allocate a dcache entry
1239 * @parent: parent of entry to allocate
1240 * @name: qstr of the name
1242 * Allocates a dentry. It returns %NULL if there is insufficient memory
1243 * available. On a success the dentry is returned. The name passed in is
1244 * copied and the copy passed in may be reused after this call.
1246 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1248 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1252 spin_lock(&parent
->d_lock
);
1254 * don't need child lock because it is not subject
1255 * to concurrency here
1257 __dget_dlock(parent
);
1258 dentry
->d_parent
= parent
;
1259 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1260 spin_unlock(&parent
->d_lock
);
1264 EXPORT_SYMBOL(d_alloc
);
1266 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1268 struct dentry
*dentry
= __d_alloc(sb
, name
);
1270 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1273 EXPORT_SYMBOL(d_alloc_pseudo
);
1275 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1280 q
.len
= strlen(name
);
1281 q
.hash
= full_name_hash(q
.name
, q
.len
);
1282 return d_alloc(parent
, &q
);
1284 EXPORT_SYMBOL(d_alloc_name
);
1286 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1288 WARN_ON_ONCE(dentry
->d_op
);
1289 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1291 DCACHE_OP_REVALIDATE
|
1292 DCACHE_OP_DELETE
));
1297 dentry
->d_flags
|= DCACHE_OP_HASH
;
1299 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1300 if (op
->d_revalidate
)
1301 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1303 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1305 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1308 EXPORT_SYMBOL(d_set_d_op
);
1310 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1312 spin_lock(&dentry
->d_lock
);
1314 if (unlikely(IS_AUTOMOUNT(inode
)))
1315 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1316 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1318 dentry
->d_inode
= inode
;
1319 dentry_rcuwalk_barrier(dentry
);
1320 spin_unlock(&dentry
->d_lock
);
1321 fsnotify_d_instantiate(dentry
, inode
);
1325 * d_instantiate - fill in inode information for a dentry
1326 * @entry: dentry to complete
1327 * @inode: inode to attach to this dentry
1329 * Fill in inode information in the entry.
1331 * This turns negative dentries into productive full members
1334 * NOTE! This assumes that the inode count has been incremented
1335 * (or otherwise set) by the caller to indicate that it is now
1336 * in use by the dcache.
1339 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1341 BUG_ON(!list_empty(&entry
->d_alias
));
1343 spin_lock(&inode
->i_lock
);
1344 __d_instantiate(entry
, inode
);
1346 spin_unlock(&inode
->i_lock
);
1347 security_d_instantiate(entry
, inode
);
1349 EXPORT_SYMBOL(d_instantiate
);
1352 * d_instantiate_unique - instantiate a non-aliased dentry
1353 * @entry: dentry to instantiate
1354 * @inode: inode to attach to this dentry
1356 * Fill in inode information in the entry. On success, it returns NULL.
1357 * If an unhashed alias of "entry" already exists, then we return the
1358 * aliased dentry instead and drop one reference to inode.
1360 * Note that in order to avoid conflicts with rename() etc, the caller
1361 * had better be holding the parent directory semaphore.
1363 * This also assumes that the inode count has been incremented
1364 * (or otherwise set) by the caller to indicate that it is now
1365 * in use by the dcache.
1367 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1368 struct inode
*inode
)
1370 struct dentry
*alias
;
1371 int len
= entry
->d_name
.len
;
1372 const char *name
= entry
->d_name
.name
;
1373 unsigned int hash
= entry
->d_name
.hash
;
1376 __d_instantiate(entry
, NULL
);
1380 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1381 struct qstr
*qstr
= &alias
->d_name
;
1384 * Don't need alias->d_lock here, because aliases with
1385 * d_parent == entry->d_parent are not subject to name or
1386 * parent changes, because the parent inode i_mutex is held.
1388 if (qstr
->hash
!= hash
)
1390 if (alias
->d_parent
!= entry
->d_parent
)
1392 if (dentry_cmp(qstr
->name
, qstr
->len
, name
, len
))
1398 __d_instantiate(entry
, inode
);
1402 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1404 struct dentry
*result
;
1406 BUG_ON(!list_empty(&entry
->d_alias
));
1409 spin_lock(&inode
->i_lock
);
1410 result
= __d_instantiate_unique(entry
, inode
);
1412 spin_unlock(&inode
->i_lock
);
1415 security_d_instantiate(entry
, inode
);
1419 BUG_ON(!d_unhashed(result
));
1424 EXPORT_SYMBOL(d_instantiate_unique
);
1427 * d_alloc_root - allocate root dentry
1428 * @root_inode: inode to allocate the root for
1430 * Allocate a root ("/") dentry for the inode given. The inode is
1431 * instantiated and returned. %NULL is returned if there is insufficient
1432 * memory or the inode passed is %NULL.
1435 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1437 struct dentry
*res
= NULL
;
1440 static const struct qstr name
= { .name
= "/", .len
= 1 };
1442 res
= __d_alloc(root_inode
->i_sb
, &name
);
1444 d_instantiate(res
, root_inode
);
1448 EXPORT_SYMBOL(d_alloc_root
);
1450 struct dentry
*d_make_root(struct inode
*root_inode
)
1452 struct dentry
*res
= NULL
;
1455 static const struct qstr name
= { .name
= "/", .len
= 1 };
1457 res
= __d_alloc(root_inode
->i_sb
, &name
);
1459 d_instantiate(res
, root_inode
);
1465 EXPORT_SYMBOL(d_make_root
);
1467 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1469 struct dentry
*alias
;
1471 if (list_empty(&inode
->i_dentry
))
1473 alias
= list_first_entry(&inode
->i_dentry
, struct dentry
, d_alias
);
1478 static struct dentry
* d_find_any_alias(struct inode
*inode
)
1482 spin_lock(&inode
->i_lock
);
1483 de
= __d_find_any_alias(inode
);
1484 spin_unlock(&inode
->i_lock
);
1490 * d_obtain_alias - find or allocate a dentry for a given inode
1491 * @inode: inode to allocate the dentry for
1493 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1494 * similar open by handle operations. The returned dentry may be anonymous,
1495 * or may have a full name (if the inode was already in the cache).
1497 * When called on a directory inode, we must ensure that the inode only ever
1498 * has one dentry. If a dentry is found, that is returned instead of
1499 * allocating a new one.
1501 * On successful return, the reference to the inode has been transferred
1502 * to the dentry. In case of an error the reference on the inode is released.
1503 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1504 * be passed in and will be the error will be propagate to the return value,
1505 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1507 struct dentry
*d_obtain_alias(struct inode
*inode
)
1509 static const struct qstr anonstring
= { .name
= "" };
1514 return ERR_PTR(-ESTALE
);
1516 return ERR_CAST(inode
);
1518 res
= d_find_any_alias(inode
);
1522 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1524 res
= ERR_PTR(-ENOMEM
);
1528 spin_lock(&inode
->i_lock
);
1529 res
= __d_find_any_alias(inode
);
1531 spin_unlock(&inode
->i_lock
);
1536 /* attach a disconnected dentry */
1537 spin_lock(&tmp
->d_lock
);
1538 tmp
->d_inode
= inode
;
1539 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1540 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1541 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1542 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1543 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1544 spin_unlock(&tmp
->d_lock
);
1545 spin_unlock(&inode
->i_lock
);
1546 security_d_instantiate(tmp
, inode
);
1551 if (res
&& !IS_ERR(res
))
1552 security_d_instantiate(res
, inode
);
1556 EXPORT_SYMBOL(d_obtain_alias
);
1559 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1560 * @inode: the inode which may have a disconnected dentry
1561 * @dentry: a negative dentry which we want to point to the inode.
1563 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1564 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1565 * and return it, else simply d_add the inode to the dentry and return NULL.
1567 * This is needed in the lookup routine of any filesystem that is exportable
1568 * (via knfsd) so that we can build dcache paths to directories effectively.
1570 * If a dentry was found and moved, then it is returned. Otherwise NULL
1571 * is returned. This matches the expected return value of ->lookup.
1574 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1576 struct dentry
*new = NULL
;
1579 return ERR_CAST(inode
);
1581 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1582 spin_lock(&inode
->i_lock
);
1583 new = __d_find_alias(inode
, 1);
1585 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1586 spin_unlock(&inode
->i_lock
);
1587 security_d_instantiate(new, inode
);
1588 d_move(new, dentry
);
1591 /* already taking inode->i_lock, so d_add() by hand */
1592 __d_instantiate(dentry
, inode
);
1593 spin_unlock(&inode
->i_lock
);
1594 security_d_instantiate(dentry
, inode
);
1598 d_add(dentry
, inode
);
1601 EXPORT_SYMBOL(d_splice_alias
);
1604 * d_add_ci - lookup or allocate new dentry with case-exact name
1605 * @inode: the inode case-insensitive lookup has found
1606 * @dentry: the negative dentry that was passed to the parent's lookup func
1607 * @name: the case-exact name to be associated with the returned dentry
1609 * This is to avoid filling the dcache with case-insensitive names to the
1610 * same inode, only the actual correct case is stored in the dcache for
1611 * case-insensitive filesystems.
1613 * For a case-insensitive lookup match and if the the case-exact dentry
1614 * already exists in in the dcache, use it and return it.
1616 * If no entry exists with the exact case name, allocate new dentry with
1617 * the exact case, and return the spliced entry.
1619 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1623 struct dentry
*found
;
1627 * First check if a dentry matching the name already exists,
1628 * if not go ahead and create it now.
1630 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1632 new = d_alloc(dentry
->d_parent
, name
);
1638 found
= d_splice_alias(inode
, new);
1647 * If a matching dentry exists, and it's not negative use it.
1649 * Decrement the reference count to balance the iget() done
1652 if (found
->d_inode
) {
1653 if (unlikely(found
->d_inode
!= inode
)) {
1654 /* This can't happen because bad inodes are unhashed. */
1655 BUG_ON(!is_bad_inode(inode
));
1656 BUG_ON(!is_bad_inode(found
->d_inode
));
1663 * We are going to instantiate this dentry, unhash it and clear the
1664 * lookup flag so we can do that.
1666 if (unlikely(d_need_lookup(found
)))
1667 d_clear_need_lookup(found
);
1670 * Negative dentry: instantiate it unless the inode is a directory and
1671 * already has a dentry.
1673 new = d_splice_alias(inode
, found
);
1682 return ERR_PTR(error
);
1684 EXPORT_SYMBOL(d_add_ci
);
1687 * __d_lookup_rcu - search for a dentry (racy, store-free)
1688 * @parent: parent dentry
1689 * @name: qstr of name we wish to find
1690 * @seq: returns d_seq value at the point where the dentry was found
1691 * @inode: returns dentry->d_inode when the inode was found valid.
1692 * Returns: dentry, or NULL
1694 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1695 * resolution (store-free path walking) design described in
1696 * Documentation/filesystems/path-lookup.txt.
1698 * This is not to be used outside core vfs.
1700 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1701 * held, and rcu_read_lock held. The returned dentry must not be stored into
1702 * without taking d_lock and checking d_seq sequence count against @seq
1705 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1708 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1709 * the returned dentry, so long as its parent's seqlock is checked after the
1710 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1711 * is formed, giving integrity down the path walk.
1713 struct dentry
*__d_lookup_rcu(struct dentry
*parent
, struct qstr
*name
,
1714 unsigned *seq
, struct inode
**inode
)
1716 unsigned int len
= name
->len
;
1717 unsigned int hash
= name
->hash
;
1718 const unsigned char *str
= name
->name
;
1719 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1720 struct hlist_bl_node
*node
;
1721 struct dentry
*dentry
;
1724 * Note: There is significant duplication with __d_lookup_rcu which is
1725 * required to prevent single threaded performance regressions
1726 * especially on architectures where smp_rmb (in seqcounts) are costly.
1727 * Keep the two functions in sync.
1731 * The hash list is protected using RCU.
1733 * Carefully use d_seq when comparing a candidate dentry, to avoid
1734 * races with d_move().
1736 * It is possible that concurrent renames can mess up our list
1737 * walk here and result in missing our dentry, resulting in the
1738 * false-negative result. d_lookup() protects against concurrent
1739 * renames using rename_lock seqlock.
1741 * See Documentation/filesystems/path-lookup.txt for more details.
1743 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1748 if (dentry
->d_name
.hash
!= hash
)
1752 *seq
= read_seqcount_begin(&dentry
->d_seq
);
1753 if (dentry
->d_parent
!= parent
)
1755 if (d_unhashed(dentry
))
1757 tlen
= dentry
->d_name
.len
;
1758 tname
= dentry
->d_name
.name
;
1759 i
= dentry
->d_inode
;
1762 * This seqcount check is required to ensure name and
1763 * len are loaded atomically, so as not to walk off the
1764 * edge of memory when walking. If we could load this
1765 * atomically some other way, we could drop this check.
1767 if (read_seqcount_retry(&dentry
->d_seq
, *seq
))
1769 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1770 if (parent
->d_op
->d_compare(parent
, *inode
,
1775 if (dentry_cmp(tname
, tlen
, str
, len
))
1779 * No extra seqcount check is required after the name
1780 * compare. The caller must perform a seqcount check in
1781 * order to do anything useful with the returned dentry
1791 * d_lookup - search for a dentry
1792 * @parent: parent dentry
1793 * @name: qstr of name we wish to find
1794 * Returns: dentry, or NULL
1796 * d_lookup searches the children of the parent dentry for the name in
1797 * question. If the dentry is found its reference count is incremented and the
1798 * dentry is returned. The caller must use dput to free the entry when it has
1799 * finished using it. %NULL is returned if the dentry does not exist.
1801 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1803 struct dentry
*dentry
;
1807 seq
= read_seqbegin(&rename_lock
);
1808 dentry
= __d_lookup(parent
, name
);
1811 } while (read_seqretry(&rename_lock
, seq
));
1814 EXPORT_SYMBOL(d_lookup
);
1817 * __d_lookup - search for a dentry (racy)
1818 * @parent: parent dentry
1819 * @name: qstr of name we wish to find
1820 * Returns: dentry, or NULL
1822 * __d_lookup is like d_lookup, however it may (rarely) return a
1823 * false-negative result due to unrelated rename activity.
1825 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1826 * however it must be used carefully, eg. with a following d_lookup in
1827 * the case of failure.
1829 * __d_lookup callers must be commented.
1831 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1833 unsigned int len
= name
->len
;
1834 unsigned int hash
= name
->hash
;
1835 const unsigned char *str
= name
->name
;
1836 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1837 struct hlist_bl_node
*node
;
1838 struct dentry
*found
= NULL
;
1839 struct dentry
*dentry
;
1842 * Note: There is significant duplication with __d_lookup_rcu which is
1843 * required to prevent single threaded performance regressions
1844 * especially on architectures where smp_rmb (in seqcounts) are costly.
1845 * Keep the two functions in sync.
1849 * The hash list is protected using RCU.
1851 * Take d_lock when comparing a candidate dentry, to avoid races
1854 * It is possible that concurrent renames can mess up our list
1855 * walk here and result in missing our dentry, resulting in the
1856 * false-negative result. d_lookup() protects against concurrent
1857 * renames using rename_lock seqlock.
1859 * See Documentation/filesystems/path-lookup.txt for more details.
1863 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1867 if (dentry
->d_name
.hash
!= hash
)
1870 spin_lock(&dentry
->d_lock
);
1871 if (dentry
->d_parent
!= parent
)
1873 if (d_unhashed(dentry
))
1877 * It is safe to compare names since d_move() cannot
1878 * change the qstr (protected by d_lock).
1880 tlen
= dentry
->d_name
.len
;
1881 tname
= dentry
->d_name
.name
;
1882 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1883 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1884 dentry
, dentry
->d_inode
,
1888 if (dentry_cmp(tname
, tlen
, str
, len
))
1894 spin_unlock(&dentry
->d_lock
);
1897 spin_unlock(&dentry
->d_lock
);
1905 * d_hash_and_lookup - hash the qstr then search for a dentry
1906 * @dir: Directory to search in
1907 * @name: qstr of name we wish to find
1909 * On hash failure or on lookup failure NULL is returned.
1911 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1913 struct dentry
*dentry
= NULL
;
1916 * Check for a fs-specific hash function. Note that we must
1917 * calculate the standard hash first, as the d_op->d_hash()
1918 * routine may choose to leave the hash value unchanged.
1920 name
->hash
= full_name_hash(name
->name
, name
->len
);
1921 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1922 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1925 dentry
= d_lookup(dir
, name
);
1931 * d_validate - verify dentry provided from insecure source (deprecated)
1932 * @dentry: The dentry alleged to be valid child of @dparent
1933 * @dparent: The parent dentry (known to be valid)
1935 * An insecure source has sent us a dentry, here we verify it and dget() it.
1936 * This is used by ncpfs in its readdir implementation.
1937 * Zero is returned in the dentry is invalid.
1939 * This function is slow for big directories, and deprecated, do not use it.
1941 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1943 struct dentry
*child
;
1945 spin_lock(&dparent
->d_lock
);
1946 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1947 if (dentry
== child
) {
1948 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1949 __dget_dlock(dentry
);
1950 spin_unlock(&dentry
->d_lock
);
1951 spin_unlock(&dparent
->d_lock
);
1955 spin_unlock(&dparent
->d_lock
);
1959 EXPORT_SYMBOL(d_validate
);
1962 * When a file is deleted, we have two options:
1963 * - turn this dentry into a negative dentry
1964 * - unhash this dentry and free it.
1966 * Usually, we want to just turn this into
1967 * a negative dentry, but if anybody else is
1968 * currently using the dentry or the inode
1969 * we can't do that and we fall back on removing
1970 * it from the hash queues and waiting for
1971 * it to be deleted later when it has no users
1975 * d_delete - delete a dentry
1976 * @dentry: The dentry to delete
1978 * Turn the dentry into a negative dentry if possible, otherwise
1979 * remove it from the hash queues so it can be deleted later
1982 void d_delete(struct dentry
* dentry
)
1984 struct inode
*inode
;
1987 * Are we the only user?
1990 spin_lock(&dentry
->d_lock
);
1991 inode
= dentry
->d_inode
;
1992 isdir
= S_ISDIR(inode
->i_mode
);
1993 if (dentry
->d_count
== 1) {
1994 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
1995 spin_unlock(&dentry
->d_lock
);
1999 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2000 dentry_unlink_inode(dentry
);
2001 fsnotify_nameremove(dentry
, isdir
);
2005 if (!d_unhashed(dentry
))
2008 spin_unlock(&dentry
->d_lock
);
2010 fsnotify_nameremove(dentry
, isdir
);
2012 EXPORT_SYMBOL(d_delete
);
2014 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2016 BUG_ON(!d_unhashed(entry
));
2018 entry
->d_flags
|= DCACHE_RCUACCESS
;
2019 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2023 static void _d_rehash(struct dentry
* entry
)
2025 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2029 * d_rehash - add an entry back to the hash
2030 * @entry: dentry to add to the hash
2032 * Adds a dentry to the hash according to its name.
2035 void d_rehash(struct dentry
* entry
)
2037 spin_lock(&entry
->d_lock
);
2039 spin_unlock(&entry
->d_lock
);
2041 EXPORT_SYMBOL(d_rehash
);
2044 * dentry_update_name_case - update case insensitive dentry with a new name
2045 * @dentry: dentry to be updated
2048 * Update a case insensitive dentry with new case of name.
2050 * dentry must have been returned by d_lookup with name @name. Old and new
2051 * name lengths must match (ie. no d_compare which allows mismatched name
2054 * Parent inode i_mutex must be held over d_lookup and into this call (to
2055 * keep renames and concurrent inserts, and readdir(2) away).
2057 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2059 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2060 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2062 spin_lock(&dentry
->d_lock
);
2063 write_seqcount_begin(&dentry
->d_seq
);
2064 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2065 write_seqcount_end(&dentry
->d_seq
);
2066 spin_unlock(&dentry
->d_lock
);
2068 EXPORT_SYMBOL(dentry_update_name_case
);
2070 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2072 if (dname_external(target
)) {
2073 if (dname_external(dentry
)) {
2075 * Both external: swap the pointers
2077 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2080 * dentry:internal, target:external. Steal target's
2081 * storage and make target internal.
2083 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2084 dentry
->d_name
.len
+ 1);
2085 dentry
->d_name
.name
= target
->d_name
.name
;
2086 target
->d_name
.name
= target
->d_iname
;
2089 if (dname_external(dentry
)) {
2091 * dentry:external, target:internal. Give dentry's
2092 * storage to target and make dentry internal
2094 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2095 target
->d_name
.len
+ 1);
2096 target
->d_name
.name
= dentry
->d_name
.name
;
2097 dentry
->d_name
.name
= dentry
->d_iname
;
2100 * Both are internal. Just copy target to dentry
2102 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2103 target
->d_name
.len
+ 1);
2104 dentry
->d_name
.len
= target
->d_name
.len
;
2108 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2111 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2114 * XXXX: do we really need to take target->d_lock?
2116 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2117 spin_lock(&target
->d_parent
->d_lock
);
2119 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2120 spin_lock(&dentry
->d_parent
->d_lock
);
2121 spin_lock_nested(&target
->d_parent
->d_lock
,
2122 DENTRY_D_LOCK_NESTED
);
2124 spin_lock(&target
->d_parent
->d_lock
);
2125 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2126 DENTRY_D_LOCK_NESTED
);
2129 if (target
< dentry
) {
2130 spin_lock_nested(&target
->d_lock
, 2);
2131 spin_lock_nested(&dentry
->d_lock
, 3);
2133 spin_lock_nested(&dentry
->d_lock
, 2);
2134 spin_lock_nested(&target
->d_lock
, 3);
2138 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2139 struct dentry
*target
)
2141 if (target
->d_parent
!= dentry
->d_parent
)
2142 spin_unlock(&dentry
->d_parent
->d_lock
);
2143 if (target
->d_parent
!= target
)
2144 spin_unlock(&target
->d_parent
->d_lock
);
2148 * When switching names, the actual string doesn't strictly have to
2149 * be preserved in the target - because we're dropping the target
2150 * anyway. As such, we can just do a simple memcpy() to copy over
2151 * the new name before we switch.
2153 * Note that we have to be a lot more careful about getting the hash
2154 * switched - we have to switch the hash value properly even if it
2155 * then no longer matches the actual (corrupted) string of the target.
2156 * The hash value has to match the hash queue that the dentry is on..
2159 * __d_move - move a dentry
2160 * @dentry: entry to move
2161 * @target: new dentry
2163 * Update the dcache to reflect the move of a file name. Negative
2164 * dcache entries should not be moved in this way. Caller must hold
2165 * rename_lock, the i_mutex of the source and target directories,
2166 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2168 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2170 if (!dentry
->d_inode
)
2171 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2173 BUG_ON(d_ancestor(dentry
, target
));
2174 BUG_ON(d_ancestor(target
, dentry
));
2176 dentry_lock_for_move(dentry
, target
);
2178 write_seqcount_begin(&dentry
->d_seq
);
2179 write_seqcount_begin(&target
->d_seq
);
2181 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2184 * Move the dentry to the target hash queue. Don't bother checking
2185 * for the same hash queue because of how unlikely it is.
2188 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2190 /* Unhash the target: dput() will then get rid of it */
2193 list_del(&dentry
->d_u
.d_child
);
2194 list_del(&target
->d_u
.d_child
);
2196 /* Switch the names.. */
2197 switch_names(dentry
, target
);
2198 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2200 /* ... and switch the parents */
2201 if (IS_ROOT(dentry
)) {
2202 dentry
->d_parent
= target
->d_parent
;
2203 target
->d_parent
= target
;
2204 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2206 swap(dentry
->d_parent
, target
->d_parent
);
2208 /* And add them back to the (new) parent lists */
2209 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2212 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2214 write_seqcount_end(&target
->d_seq
);
2215 write_seqcount_end(&dentry
->d_seq
);
2217 dentry_unlock_parents_for_move(dentry
, target
);
2218 spin_unlock(&target
->d_lock
);
2219 fsnotify_d_move(dentry
);
2220 spin_unlock(&dentry
->d_lock
);
2224 * d_move - move a dentry
2225 * @dentry: entry to move
2226 * @target: new dentry
2228 * Update the dcache to reflect the move of a file name. Negative
2229 * dcache entries should not be moved in this way. See the locking
2230 * requirements for __d_move.
2232 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2234 write_seqlock(&rename_lock
);
2235 __d_move(dentry
, target
);
2236 write_sequnlock(&rename_lock
);
2238 EXPORT_SYMBOL(d_move
);
2241 * d_ancestor - search for an ancestor
2242 * @p1: ancestor dentry
2245 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2246 * an ancestor of p2, else NULL.
2248 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2252 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2253 if (p
->d_parent
== p1
)
2260 * This helper attempts to cope with remotely renamed directories
2262 * It assumes that the caller is already holding
2263 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2265 * Note: If ever the locking in lock_rename() changes, then please
2266 * remember to update this too...
2268 static struct dentry
*__d_unalias(struct inode
*inode
,
2269 struct dentry
*dentry
, struct dentry
*alias
)
2271 struct mutex
*m1
= NULL
, *m2
= NULL
;
2274 /* If alias and dentry share a parent, then no extra locks required */
2275 if (alias
->d_parent
== dentry
->d_parent
)
2278 /* See lock_rename() */
2279 ret
= ERR_PTR(-EBUSY
);
2280 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2282 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2283 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2285 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2287 __d_move(alias
, dentry
);
2290 spin_unlock(&inode
->i_lock
);
2299 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2300 * named dentry in place of the dentry to be replaced.
2301 * returns with anon->d_lock held!
2303 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2305 struct dentry
*dparent
, *aparent
;
2307 dentry_lock_for_move(anon
, dentry
);
2309 write_seqcount_begin(&dentry
->d_seq
);
2310 write_seqcount_begin(&anon
->d_seq
);
2312 dparent
= dentry
->d_parent
;
2313 aparent
= anon
->d_parent
;
2315 switch_names(dentry
, anon
);
2316 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2318 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2319 list_del(&dentry
->d_u
.d_child
);
2320 if (!IS_ROOT(dentry
))
2321 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2323 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2325 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2326 list_del(&anon
->d_u
.d_child
);
2328 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2330 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2332 write_seqcount_end(&dentry
->d_seq
);
2333 write_seqcount_end(&anon
->d_seq
);
2335 dentry_unlock_parents_for_move(anon
, dentry
);
2336 spin_unlock(&dentry
->d_lock
);
2338 /* anon->d_lock still locked, returns locked */
2339 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2343 * d_materialise_unique - introduce an inode into the tree
2344 * @dentry: candidate dentry
2345 * @inode: inode to bind to the dentry, to which aliases may be attached
2347 * Introduces an dentry into the tree, substituting an extant disconnected
2348 * root directory alias in its place if there is one. Caller must hold the
2349 * i_mutex of the parent directory.
2351 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2353 struct dentry
*actual
;
2355 BUG_ON(!d_unhashed(dentry
));
2359 __d_instantiate(dentry
, NULL
);
2364 spin_lock(&inode
->i_lock
);
2366 if (S_ISDIR(inode
->i_mode
)) {
2367 struct dentry
*alias
;
2369 /* Does an aliased dentry already exist? */
2370 alias
= __d_find_alias(inode
, 0);
2373 write_seqlock(&rename_lock
);
2375 if (d_ancestor(alias
, dentry
)) {
2376 /* Check for loops */
2377 actual
= ERR_PTR(-ELOOP
);
2378 } else if (IS_ROOT(alias
)) {
2379 /* Is this an anonymous mountpoint that we
2380 * could splice into our tree? */
2381 __d_materialise_dentry(dentry
, alias
);
2382 write_sequnlock(&rename_lock
);
2386 /* Nope, but we must(!) avoid directory
2388 actual
= __d_unalias(inode
, dentry
, alias
);
2390 write_sequnlock(&rename_lock
);
2391 if (IS_ERR(actual
)) {
2392 if (PTR_ERR(actual
) == -ELOOP
)
2393 pr_warn_ratelimited(
2394 "VFS: Lookup of '%s' in %s %s"
2395 " would have caused loop\n",
2396 dentry
->d_name
.name
,
2397 inode
->i_sb
->s_type
->name
,
2405 /* Add a unique reference */
2406 actual
= __d_instantiate_unique(dentry
, inode
);
2410 BUG_ON(!d_unhashed(actual
));
2412 spin_lock(&actual
->d_lock
);
2415 spin_unlock(&actual
->d_lock
);
2416 spin_unlock(&inode
->i_lock
);
2418 if (actual
== dentry
) {
2419 security_d_instantiate(dentry
, inode
);
2426 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2428 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2432 return -ENAMETOOLONG
;
2434 memcpy(*buffer
, str
, namelen
);
2438 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2440 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2444 * prepend_path - Prepend path string to a buffer
2445 * @path: the dentry/vfsmount to report
2446 * @root: root vfsmnt/dentry
2447 * @buffer: pointer to the end of the buffer
2448 * @buflen: pointer to buffer length
2450 * Caller holds the rename_lock.
2452 static int prepend_path(const struct path
*path
,
2453 const struct path
*root
,
2454 char **buffer
, int *buflen
)
2456 struct dentry
*dentry
= path
->dentry
;
2457 struct vfsmount
*vfsmnt
= path
->mnt
;
2458 struct mount
*mnt
= real_mount(vfsmnt
);
2462 br_read_lock(vfsmount_lock
);
2463 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2464 struct dentry
* parent
;
2466 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2468 if (!mnt_has_parent(mnt
))
2470 dentry
= mnt
->mnt_mountpoint
;
2471 mnt
= mnt
->mnt_parent
;
2475 parent
= dentry
->d_parent
;
2477 spin_lock(&dentry
->d_lock
);
2478 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2479 spin_unlock(&dentry
->d_lock
);
2481 error
= prepend(buffer
, buflen
, "/", 1);
2489 if (!error
&& !slash
)
2490 error
= prepend(buffer
, buflen
, "/", 1);
2493 br_read_unlock(vfsmount_lock
);
2498 * Filesystems needing to implement special "root names"
2499 * should do so with ->d_dname()
2501 if (IS_ROOT(dentry
) &&
2502 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2503 WARN(1, "Root dentry has weird name <%.*s>\n",
2504 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2507 error
= prepend(buffer
, buflen
, "/", 1);
2509 error
= real_mount(vfsmnt
)->mnt_ns
? 1 : 2;
2514 * __d_path - return the path of a dentry
2515 * @path: the dentry/vfsmount to report
2516 * @root: root vfsmnt/dentry
2517 * @buf: buffer to return value in
2518 * @buflen: buffer length
2520 * Convert a dentry into an ASCII path name.
2522 * Returns a pointer into the buffer or an error code if the
2523 * path was too long.
2525 * "buflen" should be positive.
2527 * If the path is not reachable from the supplied root, return %NULL.
2529 char *__d_path(const struct path
*path
,
2530 const struct path
*root
,
2531 char *buf
, int buflen
)
2533 char *res
= buf
+ buflen
;
2536 prepend(&res
, &buflen
, "\0", 1);
2537 write_seqlock(&rename_lock
);
2538 error
= prepend_path(path
, root
, &res
, &buflen
);
2539 write_sequnlock(&rename_lock
);
2542 return ERR_PTR(error
);
2548 char *d_absolute_path(const struct path
*path
,
2549 char *buf
, int buflen
)
2551 struct path root
= {};
2552 char *res
= buf
+ buflen
;
2555 prepend(&res
, &buflen
, "\0", 1);
2556 write_seqlock(&rename_lock
);
2557 error
= prepend_path(path
, &root
, &res
, &buflen
);
2558 write_sequnlock(&rename_lock
);
2563 return ERR_PTR(error
);
2568 * same as __d_path but appends "(deleted)" for unlinked files.
2570 static int path_with_deleted(const struct path
*path
,
2571 const struct path
*root
,
2572 char **buf
, int *buflen
)
2574 prepend(buf
, buflen
, "\0", 1);
2575 if (d_unlinked(path
->dentry
)) {
2576 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2581 return prepend_path(path
, root
, buf
, buflen
);
2584 static int prepend_unreachable(char **buffer
, int *buflen
)
2586 return prepend(buffer
, buflen
, "(unreachable)", 13);
2590 * d_path - return the path of a dentry
2591 * @path: path to report
2592 * @buf: buffer to return value in
2593 * @buflen: buffer length
2595 * Convert a dentry into an ASCII path name. If the entry has been deleted
2596 * the string " (deleted)" is appended. Note that this is ambiguous.
2598 * Returns a pointer into the buffer or an error code if the path was
2599 * too long. Note: Callers should use the returned pointer, not the passed
2600 * in buffer, to use the name! The implementation often starts at an offset
2601 * into the buffer, and may leave 0 bytes at the start.
2603 * "buflen" should be positive.
2605 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2607 char *res
= buf
+ buflen
;
2612 * We have various synthetic filesystems that never get mounted. On
2613 * these filesystems dentries are never used for lookup purposes, and
2614 * thus don't need to be hashed. They also don't need a name until a
2615 * user wants to identify the object in /proc/pid/fd/. The little hack
2616 * below allows us to generate a name for these objects on demand:
2618 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2619 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2621 get_fs_root(current
->fs
, &root
);
2622 write_seqlock(&rename_lock
);
2623 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2625 res
= ERR_PTR(error
);
2626 write_sequnlock(&rename_lock
);
2630 EXPORT_SYMBOL(d_path
);
2633 * d_path_with_unreachable - return the path of a dentry
2634 * @path: path to report
2635 * @buf: buffer to return value in
2636 * @buflen: buffer length
2638 * The difference from d_path() is that this prepends "(unreachable)"
2639 * to paths which are unreachable from the current process' root.
2641 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2643 char *res
= buf
+ buflen
;
2647 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2648 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2650 get_fs_root(current
->fs
, &root
);
2651 write_seqlock(&rename_lock
);
2652 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2654 error
= prepend_unreachable(&res
, &buflen
);
2655 write_sequnlock(&rename_lock
);
2658 res
= ERR_PTR(error
);
2664 * Helper function for dentry_operations.d_dname() members
2666 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2667 const char *fmt
, ...)
2673 va_start(args
, fmt
);
2674 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2677 if (sz
> sizeof(temp
) || sz
> buflen
)
2678 return ERR_PTR(-ENAMETOOLONG
);
2680 buffer
+= buflen
- sz
;
2681 return memcpy(buffer
, temp
, sz
);
2685 * Write full pathname from the root of the filesystem into the buffer.
2687 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2689 char *end
= buf
+ buflen
;
2692 prepend(&end
, &buflen
, "\0", 1);
2699 while (!IS_ROOT(dentry
)) {
2700 struct dentry
*parent
= dentry
->d_parent
;
2704 spin_lock(&dentry
->d_lock
);
2705 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2706 spin_unlock(&dentry
->d_lock
);
2707 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2715 return ERR_PTR(-ENAMETOOLONG
);
2718 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2722 write_seqlock(&rename_lock
);
2723 retval
= __dentry_path(dentry
, buf
, buflen
);
2724 write_sequnlock(&rename_lock
);
2728 EXPORT_SYMBOL(dentry_path_raw
);
2730 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2735 write_seqlock(&rename_lock
);
2736 if (d_unlinked(dentry
)) {
2738 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2742 retval
= __dentry_path(dentry
, buf
, buflen
);
2743 write_sequnlock(&rename_lock
);
2744 if (!IS_ERR(retval
) && p
)
2745 *p
= '/'; /* restore '/' overriden with '\0' */
2748 return ERR_PTR(-ENAMETOOLONG
);
2752 * NOTE! The user-level library version returns a
2753 * character pointer. The kernel system call just
2754 * returns the length of the buffer filled (which
2755 * includes the ending '\0' character), or a negative
2756 * error value. So libc would do something like
2758 * char *getcwd(char * buf, size_t size)
2762 * retval = sys_getcwd(buf, size);
2769 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2772 struct path pwd
, root
;
2773 char *page
= (char *) __get_free_page(GFP_USER
);
2778 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2781 write_seqlock(&rename_lock
);
2782 if (!d_unlinked(pwd
.dentry
)) {
2784 char *cwd
= page
+ PAGE_SIZE
;
2785 int buflen
= PAGE_SIZE
;
2787 prepend(&cwd
, &buflen
, "\0", 1);
2788 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2789 write_sequnlock(&rename_lock
);
2794 /* Unreachable from current root */
2796 error
= prepend_unreachable(&cwd
, &buflen
);
2802 len
= PAGE_SIZE
+ page
- cwd
;
2805 if (copy_to_user(buf
, cwd
, len
))
2809 write_sequnlock(&rename_lock
);
2815 free_page((unsigned long) page
);
2820 * Test whether new_dentry is a subdirectory of old_dentry.
2822 * Trivially implemented using the dcache structure
2826 * is_subdir - is new dentry a subdirectory of old_dentry
2827 * @new_dentry: new dentry
2828 * @old_dentry: old dentry
2830 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2831 * Returns 0 otherwise.
2832 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2835 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2840 if (new_dentry
== old_dentry
)
2844 /* for restarting inner loop in case of seq retry */
2845 seq
= read_seqbegin(&rename_lock
);
2847 * Need rcu_readlock to protect against the d_parent trashing
2851 if (d_ancestor(old_dentry
, new_dentry
))
2856 } while (read_seqretry(&rename_lock
, seq
));
2861 void d_genocide(struct dentry
*root
)
2863 struct dentry
*this_parent
;
2864 struct list_head
*next
;
2868 seq
= read_seqbegin(&rename_lock
);
2871 spin_lock(&this_parent
->d_lock
);
2873 next
= this_parent
->d_subdirs
.next
;
2875 while (next
!= &this_parent
->d_subdirs
) {
2876 struct list_head
*tmp
= next
;
2877 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2880 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2881 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2882 spin_unlock(&dentry
->d_lock
);
2885 if (!list_empty(&dentry
->d_subdirs
)) {
2886 spin_unlock(&this_parent
->d_lock
);
2887 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2888 this_parent
= dentry
;
2889 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2892 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2893 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2896 spin_unlock(&dentry
->d_lock
);
2898 if (this_parent
!= root
) {
2899 struct dentry
*child
= this_parent
;
2900 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2901 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2902 this_parent
->d_count
--;
2904 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2907 next
= child
->d_u
.d_child
.next
;
2910 spin_unlock(&this_parent
->d_lock
);
2911 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2914 write_sequnlock(&rename_lock
);
2919 write_seqlock(&rename_lock
);
2924 * find_inode_number - check for dentry with name
2925 * @dir: directory to check
2926 * @name: Name to find.
2928 * Check whether a dentry already exists for the given name,
2929 * and return the inode number if it has an inode. Otherwise
2932 * This routine is used to post-process directory listings for
2933 * filesystems using synthetic inode numbers, and is necessary
2934 * to keep getcwd() working.
2937 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2939 struct dentry
* dentry
;
2942 dentry
= d_hash_and_lookup(dir
, name
);
2944 if (dentry
->d_inode
)
2945 ino
= dentry
->d_inode
->i_ino
;
2950 EXPORT_SYMBOL(find_inode_number
);
2952 static __initdata
unsigned long dhash_entries
;
2953 static int __init
set_dhash_entries(char *str
)
2957 dhash_entries
= simple_strtoul(str
, &str
, 0);
2960 __setup("dhash_entries=", set_dhash_entries
);
2962 static void __init
dcache_init_early(void)
2966 /* If hashes are distributed across NUMA nodes, defer
2967 * hash allocation until vmalloc space is available.
2973 alloc_large_system_hash("Dentry cache",
2974 sizeof(struct hlist_bl_head
),
2982 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2983 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
2986 static void __init
dcache_init(void)
2991 * A constructor could be added for stable state like the lists,
2992 * but it is probably not worth it because of the cache nature
2995 dentry_cache
= KMEM_CACHE(dentry
,
2996 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2998 /* Hash may have been set up in dcache_init_early */
3003 alloc_large_system_hash("Dentry cache",
3004 sizeof(struct hlist_bl_head
),
3012 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
3013 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3016 /* SLAB cache for __getname() consumers */
3017 struct kmem_cache
*names_cachep __read_mostly
;
3018 EXPORT_SYMBOL(names_cachep
);
3020 EXPORT_SYMBOL(d_genocide
);
3022 void __init
vfs_caches_init_early(void)
3024 dcache_init_early();
3028 void __init
vfs_caches_init(unsigned long mempages
)
3030 unsigned long reserve
;
3032 /* Base hash sizes on available memory, with a reserve equal to
3033 150% of current kernel size */
3035 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3036 mempages
-= reserve
;
3038 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3039 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3043 files_init(mempages
);