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
^ 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
);
1479 * d_find_any_alias - find any alias for a given inode
1480 * @inode: inode to find an alias for
1482 * If any aliases exist for the given inode, take and return a
1483 * reference for one of them. If no aliases exist, return %NULL.
1485 struct dentry
*d_find_any_alias(struct inode
*inode
)
1489 spin_lock(&inode
->i_lock
);
1490 de
= __d_find_any_alias(inode
);
1491 spin_unlock(&inode
->i_lock
);
1494 EXPORT_SYMBOL(d_find_any_alias
);
1497 * d_obtain_alias - find or allocate a dentry for a given inode
1498 * @inode: inode to allocate the dentry for
1500 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1501 * similar open by handle operations. The returned dentry may be anonymous,
1502 * or may have a full name (if the inode was already in the cache).
1504 * When called on a directory inode, we must ensure that the inode only ever
1505 * has one dentry. If a dentry is found, that is returned instead of
1506 * allocating a new one.
1508 * On successful return, the reference to the inode has been transferred
1509 * to the dentry. In case of an error the reference on the inode is released.
1510 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1511 * be passed in and will be the error will be propagate to the return value,
1512 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1514 struct dentry
*d_obtain_alias(struct inode
*inode
)
1516 static const struct qstr anonstring
= { .name
= "" };
1521 return ERR_PTR(-ESTALE
);
1523 return ERR_CAST(inode
);
1525 res
= d_find_any_alias(inode
);
1529 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1531 res
= ERR_PTR(-ENOMEM
);
1535 spin_lock(&inode
->i_lock
);
1536 res
= __d_find_any_alias(inode
);
1538 spin_unlock(&inode
->i_lock
);
1543 /* attach a disconnected dentry */
1544 spin_lock(&tmp
->d_lock
);
1545 tmp
->d_inode
= inode
;
1546 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1547 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1548 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1549 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1550 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1551 spin_unlock(&tmp
->d_lock
);
1552 spin_unlock(&inode
->i_lock
);
1553 security_d_instantiate(tmp
, inode
);
1558 if (res
&& !IS_ERR(res
))
1559 security_d_instantiate(res
, inode
);
1563 EXPORT_SYMBOL(d_obtain_alias
);
1566 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1567 * @inode: the inode which may have a disconnected dentry
1568 * @dentry: a negative dentry which we want to point to the inode.
1570 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1571 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1572 * and return it, else simply d_add the inode to the dentry and return NULL.
1574 * This is needed in the lookup routine of any filesystem that is exportable
1575 * (via knfsd) so that we can build dcache paths to directories effectively.
1577 * If a dentry was found and moved, then it is returned. Otherwise NULL
1578 * is returned. This matches the expected return value of ->lookup.
1581 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1583 struct dentry
*new = NULL
;
1586 return ERR_CAST(inode
);
1588 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1589 spin_lock(&inode
->i_lock
);
1590 new = __d_find_alias(inode
, 1);
1592 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1593 spin_unlock(&inode
->i_lock
);
1594 security_d_instantiate(new, inode
);
1595 d_move(new, dentry
);
1598 /* already taking inode->i_lock, so d_add() by hand */
1599 __d_instantiate(dentry
, inode
);
1600 spin_unlock(&inode
->i_lock
);
1601 security_d_instantiate(dentry
, inode
);
1605 d_add(dentry
, inode
);
1608 EXPORT_SYMBOL(d_splice_alias
);
1611 * d_add_ci - lookup or allocate new dentry with case-exact name
1612 * @inode: the inode case-insensitive lookup has found
1613 * @dentry: the negative dentry that was passed to the parent's lookup func
1614 * @name: the case-exact name to be associated with the returned dentry
1616 * This is to avoid filling the dcache with case-insensitive names to the
1617 * same inode, only the actual correct case is stored in the dcache for
1618 * case-insensitive filesystems.
1620 * For a case-insensitive lookup match and if the the case-exact dentry
1621 * already exists in in the dcache, use it and return it.
1623 * If no entry exists with the exact case name, allocate new dentry with
1624 * the exact case, and return the spliced entry.
1626 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1630 struct dentry
*found
;
1634 * First check if a dentry matching the name already exists,
1635 * if not go ahead and create it now.
1637 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1639 new = d_alloc(dentry
->d_parent
, name
);
1645 found
= d_splice_alias(inode
, new);
1654 * If a matching dentry exists, and it's not negative use it.
1656 * Decrement the reference count to balance the iget() done
1659 if (found
->d_inode
) {
1660 if (unlikely(found
->d_inode
!= inode
)) {
1661 /* This can't happen because bad inodes are unhashed. */
1662 BUG_ON(!is_bad_inode(inode
));
1663 BUG_ON(!is_bad_inode(found
->d_inode
));
1670 * We are going to instantiate this dentry, unhash it and clear the
1671 * lookup flag so we can do that.
1673 if (unlikely(d_need_lookup(found
)))
1674 d_clear_need_lookup(found
);
1677 * Negative dentry: instantiate it unless the inode is a directory and
1678 * already has a dentry.
1680 new = d_splice_alias(inode
, found
);
1689 return ERR_PTR(error
);
1691 EXPORT_SYMBOL(d_add_ci
);
1694 * __d_lookup_rcu - search for a dentry (racy, store-free)
1695 * @parent: parent dentry
1696 * @name: qstr of name we wish to find
1697 * @seq: returns d_seq value at the point where the dentry was found
1698 * @inode: returns dentry->d_inode when the inode was found valid.
1699 * Returns: dentry, or NULL
1701 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1702 * resolution (store-free path walking) design described in
1703 * Documentation/filesystems/path-lookup.txt.
1705 * This is not to be used outside core vfs.
1707 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1708 * held, and rcu_read_lock held. The returned dentry must not be stored into
1709 * without taking d_lock and checking d_seq sequence count against @seq
1712 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1715 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1716 * the returned dentry, so long as its parent's seqlock is checked after the
1717 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1718 * is formed, giving integrity down the path walk.
1720 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
1721 const struct qstr
*name
,
1722 unsigned *seqp
, struct inode
**inode
)
1724 unsigned int len
= name
->len
;
1725 unsigned int hash
= name
->hash
;
1726 const unsigned char *str
= name
->name
;
1727 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1728 struct hlist_bl_node
*node
;
1729 struct dentry
*dentry
;
1732 * Note: There is significant duplication with __d_lookup_rcu which is
1733 * required to prevent single threaded performance regressions
1734 * especially on architectures where smp_rmb (in seqcounts) are costly.
1735 * Keep the two functions in sync.
1739 * The hash list is protected using RCU.
1741 * Carefully use d_seq when comparing a candidate dentry, to avoid
1742 * races with d_move().
1744 * It is possible that concurrent renames can mess up our list
1745 * walk here and result in missing our dentry, resulting in the
1746 * false-negative result. d_lookup() protects against concurrent
1747 * renames using rename_lock seqlock.
1749 * See Documentation/filesystems/path-lookup.txt for more details.
1751 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1757 if (dentry
->d_name
.hash
!= hash
)
1761 seq
= read_seqcount_begin(&dentry
->d_seq
);
1762 if (dentry
->d_parent
!= parent
)
1764 if (d_unhashed(dentry
))
1766 tlen
= dentry
->d_name
.len
;
1767 tname
= dentry
->d_name
.name
;
1768 i
= dentry
->d_inode
;
1771 * This seqcount check is required to ensure name and
1772 * len are loaded atomically, so as not to walk off the
1773 * edge of memory when walking. If we could load this
1774 * atomically some other way, we could drop this check.
1776 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1778 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1779 if (parent
->d_op
->d_compare(parent
, *inode
,
1784 if (dentry_cmp(tname
, tlen
, str
, len
))
1788 * No extra seqcount check is required after the name
1789 * compare. The caller must perform a seqcount check in
1790 * order to do anything useful with the returned dentry
1801 * d_lookup - search for a dentry
1802 * @parent: parent dentry
1803 * @name: qstr of name we wish to find
1804 * Returns: dentry, or NULL
1806 * d_lookup searches the children of the parent dentry for the name in
1807 * question. If the dentry is found its reference count is incremented and the
1808 * dentry is returned. The caller must use dput to free the entry when it has
1809 * finished using it. %NULL is returned if the dentry does not exist.
1811 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1813 struct dentry
*dentry
;
1817 seq
= read_seqbegin(&rename_lock
);
1818 dentry
= __d_lookup(parent
, name
);
1821 } while (read_seqretry(&rename_lock
, seq
));
1824 EXPORT_SYMBOL(d_lookup
);
1827 * __d_lookup - search for a dentry (racy)
1828 * @parent: parent dentry
1829 * @name: qstr of name we wish to find
1830 * Returns: dentry, or NULL
1832 * __d_lookup is like d_lookup, however it may (rarely) return a
1833 * false-negative result due to unrelated rename activity.
1835 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1836 * however it must be used carefully, eg. with a following d_lookup in
1837 * the case of failure.
1839 * __d_lookup callers must be commented.
1841 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1843 unsigned int len
= name
->len
;
1844 unsigned int hash
= name
->hash
;
1845 const unsigned char *str
= name
->name
;
1846 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1847 struct hlist_bl_node
*node
;
1848 struct dentry
*found
= NULL
;
1849 struct dentry
*dentry
;
1852 * Note: There is significant duplication with __d_lookup_rcu which is
1853 * required to prevent single threaded performance regressions
1854 * especially on architectures where smp_rmb (in seqcounts) are costly.
1855 * Keep the two functions in sync.
1859 * The hash list is protected using RCU.
1861 * Take d_lock when comparing a candidate dentry, to avoid races
1864 * It is possible that concurrent renames can mess up our list
1865 * walk here and result in missing our dentry, resulting in the
1866 * false-negative result. d_lookup() protects against concurrent
1867 * renames using rename_lock seqlock.
1869 * See Documentation/filesystems/path-lookup.txt for more details.
1873 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1877 if (dentry
->d_name
.hash
!= hash
)
1880 spin_lock(&dentry
->d_lock
);
1881 if (dentry
->d_parent
!= parent
)
1883 if (d_unhashed(dentry
))
1887 * It is safe to compare names since d_move() cannot
1888 * change the qstr (protected by d_lock).
1890 tlen
= dentry
->d_name
.len
;
1891 tname
= dentry
->d_name
.name
;
1892 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1893 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1894 dentry
, dentry
->d_inode
,
1898 if (dentry_cmp(tname
, tlen
, str
, len
))
1904 spin_unlock(&dentry
->d_lock
);
1907 spin_unlock(&dentry
->d_lock
);
1915 * d_hash_and_lookup - hash the qstr then search for a dentry
1916 * @dir: Directory to search in
1917 * @name: qstr of name we wish to find
1919 * On hash failure or on lookup failure NULL is returned.
1921 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1923 struct dentry
*dentry
= NULL
;
1926 * Check for a fs-specific hash function. Note that we must
1927 * calculate the standard hash first, as the d_op->d_hash()
1928 * routine may choose to leave the hash value unchanged.
1930 name
->hash
= full_name_hash(name
->name
, name
->len
);
1931 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1932 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1935 dentry
= d_lookup(dir
, name
);
1941 * d_validate - verify dentry provided from insecure source (deprecated)
1942 * @dentry: The dentry alleged to be valid child of @dparent
1943 * @dparent: The parent dentry (known to be valid)
1945 * An insecure source has sent us a dentry, here we verify it and dget() it.
1946 * This is used by ncpfs in its readdir implementation.
1947 * Zero is returned in the dentry is invalid.
1949 * This function is slow for big directories, and deprecated, do not use it.
1951 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1953 struct dentry
*child
;
1955 spin_lock(&dparent
->d_lock
);
1956 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1957 if (dentry
== child
) {
1958 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1959 __dget_dlock(dentry
);
1960 spin_unlock(&dentry
->d_lock
);
1961 spin_unlock(&dparent
->d_lock
);
1965 spin_unlock(&dparent
->d_lock
);
1969 EXPORT_SYMBOL(d_validate
);
1972 * When a file is deleted, we have two options:
1973 * - turn this dentry into a negative dentry
1974 * - unhash this dentry and free it.
1976 * Usually, we want to just turn this into
1977 * a negative dentry, but if anybody else is
1978 * currently using the dentry or the inode
1979 * we can't do that and we fall back on removing
1980 * it from the hash queues and waiting for
1981 * it to be deleted later when it has no users
1985 * d_delete - delete a dentry
1986 * @dentry: The dentry to delete
1988 * Turn the dentry into a negative dentry if possible, otherwise
1989 * remove it from the hash queues so it can be deleted later
1992 void d_delete(struct dentry
* dentry
)
1994 struct inode
*inode
;
1997 * Are we the only user?
2000 spin_lock(&dentry
->d_lock
);
2001 inode
= dentry
->d_inode
;
2002 isdir
= S_ISDIR(inode
->i_mode
);
2003 if (dentry
->d_count
== 1) {
2004 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
2005 spin_unlock(&dentry
->d_lock
);
2009 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2010 dentry_unlink_inode(dentry
);
2011 fsnotify_nameremove(dentry
, isdir
);
2015 if (!d_unhashed(dentry
))
2018 spin_unlock(&dentry
->d_lock
);
2020 fsnotify_nameremove(dentry
, isdir
);
2022 EXPORT_SYMBOL(d_delete
);
2024 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2026 BUG_ON(!d_unhashed(entry
));
2028 entry
->d_flags
|= DCACHE_RCUACCESS
;
2029 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2033 static void _d_rehash(struct dentry
* entry
)
2035 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2039 * d_rehash - add an entry back to the hash
2040 * @entry: dentry to add to the hash
2042 * Adds a dentry to the hash according to its name.
2045 void d_rehash(struct dentry
* entry
)
2047 spin_lock(&entry
->d_lock
);
2049 spin_unlock(&entry
->d_lock
);
2051 EXPORT_SYMBOL(d_rehash
);
2054 * dentry_update_name_case - update case insensitive dentry with a new name
2055 * @dentry: dentry to be updated
2058 * Update a case insensitive dentry with new case of name.
2060 * dentry must have been returned by d_lookup with name @name. Old and new
2061 * name lengths must match (ie. no d_compare which allows mismatched name
2064 * Parent inode i_mutex must be held over d_lookup and into this call (to
2065 * keep renames and concurrent inserts, and readdir(2) away).
2067 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2069 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2070 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2072 spin_lock(&dentry
->d_lock
);
2073 write_seqcount_begin(&dentry
->d_seq
);
2074 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2075 write_seqcount_end(&dentry
->d_seq
);
2076 spin_unlock(&dentry
->d_lock
);
2078 EXPORT_SYMBOL(dentry_update_name_case
);
2080 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2082 if (dname_external(target
)) {
2083 if (dname_external(dentry
)) {
2085 * Both external: swap the pointers
2087 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2090 * dentry:internal, target:external. Steal target's
2091 * storage and make target internal.
2093 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2094 dentry
->d_name
.len
+ 1);
2095 dentry
->d_name
.name
= target
->d_name
.name
;
2096 target
->d_name
.name
= target
->d_iname
;
2099 if (dname_external(dentry
)) {
2101 * dentry:external, target:internal. Give dentry's
2102 * storage to target and make dentry internal
2104 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2105 target
->d_name
.len
+ 1);
2106 target
->d_name
.name
= dentry
->d_name
.name
;
2107 dentry
->d_name
.name
= dentry
->d_iname
;
2110 * Both are internal. Just copy target to dentry
2112 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2113 target
->d_name
.len
+ 1);
2114 dentry
->d_name
.len
= target
->d_name
.len
;
2118 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2121 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2124 * XXXX: do we really need to take target->d_lock?
2126 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2127 spin_lock(&target
->d_parent
->d_lock
);
2129 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2130 spin_lock(&dentry
->d_parent
->d_lock
);
2131 spin_lock_nested(&target
->d_parent
->d_lock
,
2132 DENTRY_D_LOCK_NESTED
);
2134 spin_lock(&target
->d_parent
->d_lock
);
2135 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2136 DENTRY_D_LOCK_NESTED
);
2139 if (target
< dentry
) {
2140 spin_lock_nested(&target
->d_lock
, 2);
2141 spin_lock_nested(&dentry
->d_lock
, 3);
2143 spin_lock_nested(&dentry
->d_lock
, 2);
2144 spin_lock_nested(&target
->d_lock
, 3);
2148 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2149 struct dentry
*target
)
2151 if (target
->d_parent
!= dentry
->d_parent
)
2152 spin_unlock(&dentry
->d_parent
->d_lock
);
2153 if (target
->d_parent
!= target
)
2154 spin_unlock(&target
->d_parent
->d_lock
);
2158 * When switching names, the actual string doesn't strictly have to
2159 * be preserved in the target - because we're dropping the target
2160 * anyway. As such, we can just do a simple memcpy() to copy over
2161 * the new name before we switch.
2163 * Note that we have to be a lot more careful about getting the hash
2164 * switched - we have to switch the hash value properly even if it
2165 * then no longer matches the actual (corrupted) string of the target.
2166 * The hash value has to match the hash queue that the dentry is on..
2169 * __d_move - move a dentry
2170 * @dentry: entry to move
2171 * @target: new dentry
2173 * Update the dcache to reflect the move of a file name. Negative
2174 * dcache entries should not be moved in this way. Caller must hold
2175 * rename_lock, the i_mutex of the source and target directories,
2176 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2178 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2180 if (!dentry
->d_inode
)
2181 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2183 BUG_ON(d_ancestor(dentry
, target
));
2184 BUG_ON(d_ancestor(target
, dentry
));
2186 dentry_lock_for_move(dentry
, target
);
2188 write_seqcount_begin(&dentry
->d_seq
);
2189 write_seqcount_begin(&target
->d_seq
);
2191 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2194 * Move the dentry to the target hash queue. Don't bother checking
2195 * for the same hash queue because of how unlikely it is.
2198 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2200 /* Unhash the target: dput() will then get rid of it */
2203 list_del(&dentry
->d_u
.d_child
);
2204 list_del(&target
->d_u
.d_child
);
2206 /* Switch the names.. */
2207 switch_names(dentry
, target
);
2208 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2210 /* ... and switch the parents */
2211 if (IS_ROOT(dentry
)) {
2212 dentry
->d_parent
= target
->d_parent
;
2213 target
->d_parent
= target
;
2214 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2216 swap(dentry
->d_parent
, target
->d_parent
);
2218 /* And add them back to the (new) parent lists */
2219 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2222 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2224 write_seqcount_end(&target
->d_seq
);
2225 write_seqcount_end(&dentry
->d_seq
);
2227 dentry_unlock_parents_for_move(dentry
, target
);
2228 spin_unlock(&target
->d_lock
);
2229 fsnotify_d_move(dentry
);
2230 spin_unlock(&dentry
->d_lock
);
2234 * d_move - move a dentry
2235 * @dentry: entry to move
2236 * @target: new dentry
2238 * Update the dcache to reflect the move of a file name. Negative
2239 * dcache entries should not be moved in this way. See the locking
2240 * requirements for __d_move.
2242 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2244 write_seqlock(&rename_lock
);
2245 __d_move(dentry
, target
);
2246 write_sequnlock(&rename_lock
);
2248 EXPORT_SYMBOL(d_move
);
2251 * d_ancestor - search for an ancestor
2252 * @p1: ancestor dentry
2255 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2256 * an ancestor of p2, else NULL.
2258 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2262 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2263 if (p
->d_parent
== p1
)
2270 * This helper attempts to cope with remotely renamed directories
2272 * It assumes that the caller is already holding
2273 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2275 * Note: If ever the locking in lock_rename() changes, then please
2276 * remember to update this too...
2278 static struct dentry
*__d_unalias(struct inode
*inode
,
2279 struct dentry
*dentry
, struct dentry
*alias
)
2281 struct mutex
*m1
= NULL
, *m2
= NULL
;
2284 /* If alias and dentry share a parent, then no extra locks required */
2285 if (alias
->d_parent
== dentry
->d_parent
)
2288 /* See lock_rename() */
2289 ret
= ERR_PTR(-EBUSY
);
2290 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2292 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2293 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2295 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2297 __d_move(alias
, dentry
);
2300 spin_unlock(&inode
->i_lock
);
2309 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2310 * named dentry in place of the dentry to be replaced.
2311 * returns with anon->d_lock held!
2313 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2315 struct dentry
*dparent
, *aparent
;
2317 dentry_lock_for_move(anon
, dentry
);
2319 write_seqcount_begin(&dentry
->d_seq
);
2320 write_seqcount_begin(&anon
->d_seq
);
2322 dparent
= dentry
->d_parent
;
2323 aparent
= anon
->d_parent
;
2325 switch_names(dentry
, anon
);
2326 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2328 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2329 list_del(&dentry
->d_u
.d_child
);
2330 if (!IS_ROOT(dentry
))
2331 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2333 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2335 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2336 list_del(&anon
->d_u
.d_child
);
2338 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2340 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2342 write_seqcount_end(&dentry
->d_seq
);
2343 write_seqcount_end(&anon
->d_seq
);
2345 dentry_unlock_parents_for_move(anon
, dentry
);
2346 spin_unlock(&dentry
->d_lock
);
2348 /* anon->d_lock still locked, returns locked */
2349 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2353 * d_materialise_unique - introduce an inode into the tree
2354 * @dentry: candidate dentry
2355 * @inode: inode to bind to the dentry, to which aliases may be attached
2357 * Introduces an dentry into the tree, substituting an extant disconnected
2358 * root directory alias in its place if there is one. Caller must hold the
2359 * i_mutex of the parent directory.
2361 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2363 struct dentry
*actual
;
2365 BUG_ON(!d_unhashed(dentry
));
2369 __d_instantiate(dentry
, NULL
);
2374 spin_lock(&inode
->i_lock
);
2376 if (S_ISDIR(inode
->i_mode
)) {
2377 struct dentry
*alias
;
2379 /* Does an aliased dentry already exist? */
2380 alias
= __d_find_alias(inode
, 0);
2383 write_seqlock(&rename_lock
);
2385 if (d_ancestor(alias
, dentry
)) {
2386 /* Check for loops */
2387 actual
= ERR_PTR(-ELOOP
);
2388 } else if (IS_ROOT(alias
)) {
2389 /* Is this an anonymous mountpoint that we
2390 * could splice into our tree? */
2391 __d_materialise_dentry(dentry
, alias
);
2392 write_sequnlock(&rename_lock
);
2396 /* Nope, but we must(!) avoid directory
2398 actual
= __d_unalias(inode
, dentry
, alias
);
2400 write_sequnlock(&rename_lock
);
2401 if (IS_ERR(actual
)) {
2402 if (PTR_ERR(actual
) == -ELOOP
)
2403 pr_warn_ratelimited(
2404 "VFS: Lookup of '%s' in %s %s"
2405 " would have caused loop\n",
2406 dentry
->d_name
.name
,
2407 inode
->i_sb
->s_type
->name
,
2415 /* Add a unique reference */
2416 actual
= __d_instantiate_unique(dentry
, inode
);
2420 BUG_ON(!d_unhashed(actual
));
2422 spin_lock(&actual
->d_lock
);
2425 spin_unlock(&actual
->d_lock
);
2426 spin_unlock(&inode
->i_lock
);
2428 if (actual
== dentry
) {
2429 security_d_instantiate(dentry
, inode
);
2436 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2438 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2442 return -ENAMETOOLONG
;
2444 memcpy(*buffer
, str
, namelen
);
2448 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2450 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2454 * prepend_path - Prepend path string to a buffer
2455 * @path: the dentry/vfsmount to report
2456 * @root: root vfsmnt/dentry
2457 * @buffer: pointer to the end of the buffer
2458 * @buflen: pointer to buffer length
2460 * Caller holds the rename_lock.
2462 static int prepend_path(const struct path
*path
,
2463 const struct path
*root
,
2464 char **buffer
, int *buflen
)
2466 struct dentry
*dentry
= path
->dentry
;
2467 struct vfsmount
*vfsmnt
= path
->mnt
;
2468 struct mount
*mnt
= real_mount(vfsmnt
);
2472 br_read_lock(vfsmount_lock
);
2473 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2474 struct dentry
* parent
;
2476 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2478 if (!mnt_has_parent(mnt
))
2480 dentry
= mnt
->mnt_mountpoint
;
2481 mnt
= mnt
->mnt_parent
;
2485 parent
= dentry
->d_parent
;
2487 spin_lock(&dentry
->d_lock
);
2488 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2489 spin_unlock(&dentry
->d_lock
);
2491 error
= prepend(buffer
, buflen
, "/", 1);
2499 if (!error
&& !slash
)
2500 error
= prepend(buffer
, buflen
, "/", 1);
2503 br_read_unlock(vfsmount_lock
);
2508 * Filesystems needing to implement special "root names"
2509 * should do so with ->d_dname()
2511 if (IS_ROOT(dentry
) &&
2512 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2513 WARN(1, "Root dentry has weird name <%.*s>\n",
2514 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2517 error
= prepend(buffer
, buflen
, "/", 1);
2519 error
= real_mount(vfsmnt
)->mnt_ns
? 1 : 2;
2524 * __d_path - return the path of a dentry
2525 * @path: the dentry/vfsmount to report
2526 * @root: root vfsmnt/dentry
2527 * @buf: buffer to return value in
2528 * @buflen: buffer length
2530 * Convert a dentry into an ASCII path name.
2532 * Returns a pointer into the buffer or an error code if the
2533 * path was too long.
2535 * "buflen" should be positive.
2537 * If the path is not reachable from the supplied root, return %NULL.
2539 char *__d_path(const struct path
*path
,
2540 const struct path
*root
,
2541 char *buf
, int buflen
)
2543 char *res
= buf
+ buflen
;
2546 prepend(&res
, &buflen
, "\0", 1);
2547 write_seqlock(&rename_lock
);
2548 error
= prepend_path(path
, root
, &res
, &buflen
);
2549 write_sequnlock(&rename_lock
);
2552 return ERR_PTR(error
);
2558 char *d_absolute_path(const struct path
*path
,
2559 char *buf
, int buflen
)
2561 struct path root
= {};
2562 char *res
= buf
+ buflen
;
2565 prepend(&res
, &buflen
, "\0", 1);
2566 write_seqlock(&rename_lock
);
2567 error
= prepend_path(path
, &root
, &res
, &buflen
);
2568 write_sequnlock(&rename_lock
);
2573 return ERR_PTR(error
);
2578 * same as __d_path but appends "(deleted)" for unlinked files.
2580 static int path_with_deleted(const struct path
*path
,
2581 const struct path
*root
,
2582 char **buf
, int *buflen
)
2584 prepend(buf
, buflen
, "\0", 1);
2585 if (d_unlinked(path
->dentry
)) {
2586 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2591 return prepend_path(path
, root
, buf
, buflen
);
2594 static int prepend_unreachable(char **buffer
, int *buflen
)
2596 return prepend(buffer
, buflen
, "(unreachable)", 13);
2600 * d_path - return the path of a dentry
2601 * @path: path to report
2602 * @buf: buffer to return value in
2603 * @buflen: buffer length
2605 * Convert a dentry into an ASCII path name. If the entry has been deleted
2606 * the string " (deleted)" is appended. Note that this is ambiguous.
2608 * Returns a pointer into the buffer or an error code if the path was
2609 * too long. Note: Callers should use the returned pointer, not the passed
2610 * in buffer, to use the name! The implementation often starts at an offset
2611 * into the buffer, and may leave 0 bytes at the start.
2613 * "buflen" should be positive.
2615 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2617 char *res
= buf
+ buflen
;
2622 * We have various synthetic filesystems that never get mounted. On
2623 * these filesystems dentries are never used for lookup purposes, and
2624 * thus don't need to be hashed. They also don't need a name until a
2625 * user wants to identify the object in /proc/pid/fd/. The little hack
2626 * below allows us to generate a name for these objects on demand:
2628 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2629 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2631 get_fs_root(current
->fs
, &root
);
2632 write_seqlock(&rename_lock
);
2633 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2635 res
= ERR_PTR(error
);
2636 write_sequnlock(&rename_lock
);
2640 EXPORT_SYMBOL(d_path
);
2643 * d_path_with_unreachable - return the path of a dentry
2644 * @path: path to report
2645 * @buf: buffer to return value in
2646 * @buflen: buffer length
2648 * The difference from d_path() is that this prepends "(unreachable)"
2649 * to paths which are unreachable from the current process' root.
2651 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2653 char *res
= buf
+ buflen
;
2657 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2658 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2660 get_fs_root(current
->fs
, &root
);
2661 write_seqlock(&rename_lock
);
2662 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2664 error
= prepend_unreachable(&res
, &buflen
);
2665 write_sequnlock(&rename_lock
);
2668 res
= ERR_PTR(error
);
2674 * Helper function for dentry_operations.d_dname() members
2676 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2677 const char *fmt
, ...)
2683 va_start(args
, fmt
);
2684 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2687 if (sz
> sizeof(temp
) || sz
> buflen
)
2688 return ERR_PTR(-ENAMETOOLONG
);
2690 buffer
+= buflen
- sz
;
2691 return memcpy(buffer
, temp
, sz
);
2695 * Write full pathname from the root of the filesystem into the buffer.
2697 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2699 char *end
= buf
+ buflen
;
2702 prepend(&end
, &buflen
, "\0", 1);
2709 while (!IS_ROOT(dentry
)) {
2710 struct dentry
*parent
= dentry
->d_parent
;
2714 spin_lock(&dentry
->d_lock
);
2715 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2716 spin_unlock(&dentry
->d_lock
);
2717 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2725 return ERR_PTR(-ENAMETOOLONG
);
2728 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2732 write_seqlock(&rename_lock
);
2733 retval
= __dentry_path(dentry
, buf
, buflen
);
2734 write_sequnlock(&rename_lock
);
2738 EXPORT_SYMBOL(dentry_path_raw
);
2740 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2745 write_seqlock(&rename_lock
);
2746 if (d_unlinked(dentry
)) {
2748 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2752 retval
= __dentry_path(dentry
, buf
, buflen
);
2753 write_sequnlock(&rename_lock
);
2754 if (!IS_ERR(retval
) && p
)
2755 *p
= '/'; /* restore '/' overriden with '\0' */
2758 return ERR_PTR(-ENAMETOOLONG
);
2762 * NOTE! The user-level library version returns a
2763 * character pointer. The kernel system call just
2764 * returns the length of the buffer filled (which
2765 * includes the ending '\0' character), or a negative
2766 * error value. So libc would do something like
2768 * char *getcwd(char * buf, size_t size)
2772 * retval = sys_getcwd(buf, size);
2779 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2782 struct path pwd
, root
;
2783 char *page
= (char *) __get_free_page(GFP_USER
);
2788 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2791 write_seqlock(&rename_lock
);
2792 if (!d_unlinked(pwd
.dentry
)) {
2794 char *cwd
= page
+ PAGE_SIZE
;
2795 int buflen
= PAGE_SIZE
;
2797 prepend(&cwd
, &buflen
, "\0", 1);
2798 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2799 write_sequnlock(&rename_lock
);
2804 /* Unreachable from current root */
2806 error
= prepend_unreachable(&cwd
, &buflen
);
2812 len
= PAGE_SIZE
+ page
- cwd
;
2815 if (copy_to_user(buf
, cwd
, len
))
2819 write_sequnlock(&rename_lock
);
2825 free_page((unsigned long) page
);
2830 * Test whether new_dentry is a subdirectory of old_dentry.
2832 * Trivially implemented using the dcache structure
2836 * is_subdir - is new dentry a subdirectory of old_dentry
2837 * @new_dentry: new dentry
2838 * @old_dentry: old dentry
2840 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2841 * Returns 0 otherwise.
2842 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2845 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2850 if (new_dentry
== old_dentry
)
2854 /* for restarting inner loop in case of seq retry */
2855 seq
= read_seqbegin(&rename_lock
);
2857 * Need rcu_readlock to protect against the d_parent trashing
2861 if (d_ancestor(old_dentry
, new_dentry
))
2866 } while (read_seqretry(&rename_lock
, seq
));
2871 void d_genocide(struct dentry
*root
)
2873 struct dentry
*this_parent
;
2874 struct list_head
*next
;
2878 seq
= read_seqbegin(&rename_lock
);
2881 spin_lock(&this_parent
->d_lock
);
2883 next
= this_parent
->d_subdirs
.next
;
2885 while (next
!= &this_parent
->d_subdirs
) {
2886 struct list_head
*tmp
= next
;
2887 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2890 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2891 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2892 spin_unlock(&dentry
->d_lock
);
2895 if (!list_empty(&dentry
->d_subdirs
)) {
2896 spin_unlock(&this_parent
->d_lock
);
2897 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2898 this_parent
= dentry
;
2899 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2902 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2903 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2906 spin_unlock(&dentry
->d_lock
);
2908 if (this_parent
!= root
) {
2909 struct dentry
*child
= this_parent
;
2910 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2911 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2912 this_parent
->d_count
--;
2914 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2917 next
= child
->d_u
.d_child
.next
;
2920 spin_unlock(&this_parent
->d_lock
);
2921 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2924 write_sequnlock(&rename_lock
);
2929 write_seqlock(&rename_lock
);
2934 * find_inode_number - check for dentry with name
2935 * @dir: directory to check
2936 * @name: Name to find.
2938 * Check whether a dentry already exists for the given name,
2939 * and return the inode number if it has an inode. Otherwise
2942 * This routine is used to post-process directory listings for
2943 * filesystems using synthetic inode numbers, and is necessary
2944 * to keep getcwd() working.
2947 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2949 struct dentry
* dentry
;
2952 dentry
= d_hash_and_lookup(dir
, name
);
2954 if (dentry
->d_inode
)
2955 ino
= dentry
->d_inode
->i_ino
;
2960 EXPORT_SYMBOL(find_inode_number
);
2962 static __initdata
unsigned long dhash_entries
;
2963 static int __init
set_dhash_entries(char *str
)
2967 dhash_entries
= simple_strtoul(str
, &str
, 0);
2970 __setup("dhash_entries=", set_dhash_entries
);
2972 static void __init
dcache_init_early(void)
2976 /* If hashes are distributed across NUMA nodes, defer
2977 * hash allocation until vmalloc space is available.
2983 alloc_large_system_hash("Dentry cache",
2984 sizeof(struct hlist_bl_head
),
2992 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
2993 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
2996 static void __init
dcache_init(void)
3001 * A constructor could be added for stable state like the lists,
3002 * but it is probably not worth it because of the cache nature
3005 dentry_cache
= KMEM_CACHE(dentry
,
3006 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3008 /* Hash may have been set up in dcache_init_early */
3013 alloc_large_system_hash("Dentry cache",
3014 sizeof(struct hlist_bl_head
),
3022 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3023 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3026 /* SLAB cache for __getname() consumers */
3027 struct kmem_cache
*names_cachep __read_mostly
;
3028 EXPORT_SYMBOL(names_cachep
);
3030 EXPORT_SYMBOL(d_genocide
);
3032 void __init
vfs_caches_init_early(void)
3034 dcache_init_early();
3038 void __init
vfs_caches_init(unsigned long mempages
)
3040 unsigned long reserve
;
3042 /* Base hash sizes on available memory, with a reserve equal to
3043 150% of current kernel size */
3045 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3046 mempages
-= reserve
;
3048 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3049 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3053 files_init(mempages
);