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_sb
->s_nr_dentry_unused
--;
247 dentry_stat
.nr_unused
--;
251 * Remove a dentry with references from the LRU.
253 static void dentry_lru_del(struct dentry
*dentry
)
255 if (!list_empty(&dentry
->d_lru
)) {
256 spin_lock(&dcache_lru_lock
);
257 __dentry_lru_del(dentry
);
258 spin_unlock(&dcache_lru_lock
);
263 * Remove a dentry that is unreferenced and about to be pruned
264 * (unhashed and destroyed) from the LRU, and inform the file system.
265 * This wrapper should be called _prior_ to unhashing a victim dentry.
267 static void dentry_lru_prune(struct dentry
*dentry
)
269 if (!list_empty(&dentry
->d_lru
)) {
270 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
271 dentry
->d_op
->d_prune(dentry
);
273 spin_lock(&dcache_lru_lock
);
274 __dentry_lru_del(dentry
);
275 spin_unlock(&dcache_lru_lock
);
279 static void dentry_lru_move_list(struct dentry
*dentry
, struct list_head
*list
)
281 spin_lock(&dcache_lru_lock
);
282 if (list_empty(&dentry
->d_lru
)) {
283 list_add_tail(&dentry
->d_lru
, list
);
284 dentry
->d_sb
->s_nr_dentry_unused
++;
285 dentry_stat
.nr_unused
++;
287 list_move_tail(&dentry
->d_lru
, list
);
289 spin_unlock(&dcache_lru_lock
);
293 * d_kill - kill dentry and return parent
294 * @dentry: dentry to kill
295 * @parent: parent dentry
297 * The dentry must already be unhashed and removed from the LRU.
299 * If this is the root of the dentry tree, return NULL.
301 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
304 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
305 __releases(dentry
->d_lock
)
306 __releases(parent
->d_lock
)
307 __releases(dentry
->d_inode
->i_lock
)
309 list_del(&dentry
->d_u
.d_child
);
311 * Inform try_to_ascend() that we are no longer attached to the
314 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
316 spin_unlock(&parent
->d_lock
);
319 * dentry_iput drops the locks, at which point nobody (except
320 * transient RCU lookups) can reach this dentry.
327 * Unhash a dentry without inserting an RCU walk barrier or checking that
328 * dentry->d_lock is locked. The caller must take care of that, if
331 static void __d_shrink(struct dentry
*dentry
)
333 if (!d_unhashed(dentry
)) {
334 struct hlist_bl_head
*b
;
335 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
336 b
= &dentry
->d_sb
->s_anon
;
338 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
341 __hlist_bl_del(&dentry
->d_hash
);
342 dentry
->d_hash
.pprev
= NULL
;
348 * d_drop - drop a dentry
349 * @dentry: dentry to drop
351 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
352 * be found through a VFS lookup any more. Note that this is different from
353 * deleting the dentry - d_delete will try to mark the dentry negative if
354 * possible, giving a successful _negative_ lookup, while d_drop will
355 * just make the cache lookup fail.
357 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
358 * reason (NFS timeouts or autofs deletes).
360 * __d_drop requires dentry->d_lock.
362 void __d_drop(struct dentry
*dentry
)
364 if (!d_unhashed(dentry
)) {
366 dentry_rcuwalk_barrier(dentry
);
369 EXPORT_SYMBOL(__d_drop
);
371 void d_drop(struct dentry
*dentry
)
373 spin_lock(&dentry
->d_lock
);
375 spin_unlock(&dentry
->d_lock
);
377 EXPORT_SYMBOL(d_drop
);
380 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
381 * @dentry: dentry to drop
383 * This is called when we do a lookup on a placeholder dentry that needed to be
384 * looked up. The dentry should have been hashed in order for it to be found by
385 * the lookup code, but now needs to be unhashed while we do the actual lookup
386 * and clear the DCACHE_NEED_LOOKUP flag.
388 void d_clear_need_lookup(struct dentry
*dentry
)
390 spin_lock(&dentry
->d_lock
);
392 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
393 spin_unlock(&dentry
->d_lock
);
395 EXPORT_SYMBOL(d_clear_need_lookup
);
398 * Finish off a dentry we've decided to kill.
399 * dentry->d_lock must be held, returns with it unlocked.
400 * If ref is non-zero, then decrement the refcount too.
401 * Returns dentry requiring refcount drop, or NULL if we're done.
403 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
404 __releases(dentry
->d_lock
)
407 struct dentry
*parent
;
409 inode
= dentry
->d_inode
;
410 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
412 spin_unlock(&dentry
->d_lock
);
414 return dentry
; /* try again with same dentry */
419 parent
= dentry
->d_parent
;
420 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
422 spin_unlock(&inode
->i_lock
);
429 * if dentry was on the d_lru list delete it from there.
430 * inform the fs via d_prune that this dentry is about to be
431 * unhashed and destroyed.
433 dentry_lru_prune(dentry
);
434 /* if it was on the hash then remove it */
436 return d_kill(dentry
, parent
);
442 * This is complicated by the fact that we do not want to put
443 * dentries that are no longer on any hash chain on the unused
444 * list: we'd much rather just get rid of them immediately.
446 * However, that implies that we have to traverse the dentry
447 * tree upwards to the parents which might _also_ now be
448 * scheduled for deletion (it may have been only waiting for
449 * its last child to go away).
451 * This tail recursion is done by hand as we don't want to depend
452 * on the compiler to always get this right (gcc generally doesn't).
453 * Real recursion would eat up our stack space.
457 * dput - release a dentry
458 * @dentry: dentry to release
460 * Release a dentry. This will drop the usage count and if appropriate
461 * call the dentry unlink method as well as removing it from the queues and
462 * releasing its resources. If the parent dentries were scheduled for release
463 * they too may now get deleted.
465 void dput(struct dentry
*dentry
)
471 if (dentry
->d_count
== 1)
473 spin_lock(&dentry
->d_lock
);
474 BUG_ON(!dentry
->d_count
);
475 if (dentry
->d_count
> 1) {
477 spin_unlock(&dentry
->d_lock
);
481 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
482 if (dentry
->d_op
->d_delete(dentry
))
486 /* Unreachable? Get rid of it */
487 if (d_unhashed(dentry
))
491 * If this dentry needs lookup, don't set the referenced flag so that it
492 * is more likely to be cleaned up by the dcache shrinker in case of
495 if (!d_need_lookup(dentry
))
496 dentry
->d_flags
|= DCACHE_REFERENCED
;
497 dentry_lru_add(dentry
);
500 spin_unlock(&dentry
->d_lock
);
504 dentry
= dentry_kill(dentry
, 1);
511 * d_invalidate - invalidate a dentry
512 * @dentry: dentry to invalidate
514 * Try to invalidate the dentry if it turns out to be
515 * possible. If there are other dentries that can be
516 * reached through this one we can't delete it and we
517 * return -EBUSY. On success we return 0.
522 int d_invalidate(struct dentry
* dentry
)
525 * If it's already been dropped, return OK.
527 spin_lock(&dentry
->d_lock
);
528 if (d_unhashed(dentry
)) {
529 spin_unlock(&dentry
->d_lock
);
533 * Check whether to do a partial shrink_dcache
534 * to get rid of unused child entries.
536 if (!list_empty(&dentry
->d_subdirs
)) {
537 spin_unlock(&dentry
->d_lock
);
538 shrink_dcache_parent(dentry
);
539 spin_lock(&dentry
->d_lock
);
543 * Somebody else still using it?
545 * If it's a directory, we can't drop it
546 * for fear of somebody re-populating it
547 * with children (even though dropping it
548 * would make it unreachable from the root,
549 * we might still populate it if it was a
550 * working directory or similar).
551 * We also need to leave mountpoints alone,
554 if (dentry
->d_count
> 1 && dentry
->d_inode
) {
555 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
556 spin_unlock(&dentry
->d_lock
);
562 spin_unlock(&dentry
->d_lock
);
565 EXPORT_SYMBOL(d_invalidate
);
567 /* This must be called with d_lock held */
568 static inline void __dget_dlock(struct dentry
*dentry
)
573 static inline void __dget(struct dentry
*dentry
)
575 spin_lock(&dentry
->d_lock
);
576 __dget_dlock(dentry
);
577 spin_unlock(&dentry
->d_lock
);
580 struct dentry
*dget_parent(struct dentry
*dentry
)
586 * Don't need rcu_dereference because we re-check it was correct under
590 ret
= dentry
->d_parent
;
591 spin_lock(&ret
->d_lock
);
592 if (unlikely(ret
!= dentry
->d_parent
)) {
593 spin_unlock(&ret
->d_lock
);
598 BUG_ON(!ret
->d_count
);
600 spin_unlock(&ret
->d_lock
);
603 EXPORT_SYMBOL(dget_parent
);
606 * d_find_alias - grab a hashed alias of inode
607 * @inode: inode in question
608 * @want_discon: flag, used by d_splice_alias, to request
609 * that only a DISCONNECTED alias be returned.
611 * If inode has a hashed alias, or is a directory and has any alias,
612 * acquire the reference to alias and return it. Otherwise return NULL.
613 * Notice that if inode is a directory there can be only one alias and
614 * it can be unhashed only if it has no children, or if it is the root
617 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
618 * any other hashed alias over that one unless @want_discon is set,
619 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
621 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
623 struct dentry
*alias
, *discon_alias
;
627 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
628 spin_lock(&alias
->d_lock
);
629 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
630 if (IS_ROOT(alias
) &&
631 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
632 discon_alias
= alias
;
633 } else if (!want_discon
) {
635 spin_unlock(&alias
->d_lock
);
639 spin_unlock(&alias
->d_lock
);
642 alias
= discon_alias
;
643 spin_lock(&alias
->d_lock
);
644 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
645 if (IS_ROOT(alias
) &&
646 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
648 spin_unlock(&alias
->d_lock
);
652 spin_unlock(&alias
->d_lock
);
658 struct dentry
*d_find_alias(struct inode
*inode
)
660 struct dentry
*de
= NULL
;
662 if (!list_empty(&inode
->i_dentry
)) {
663 spin_lock(&inode
->i_lock
);
664 de
= __d_find_alias(inode
, 0);
665 spin_unlock(&inode
->i_lock
);
669 EXPORT_SYMBOL(d_find_alias
);
672 * Try to kill dentries associated with this inode.
673 * WARNING: you must own a reference to inode.
675 void d_prune_aliases(struct inode
*inode
)
677 struct dentry
*dentry
;
679 spin_lock(&inode
->i_lock
);
680 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
681 spin_lock(&dentry
->d_lock
);
682 if (!dentry
->d_count
) {
683 __dget_dlock(dentry
);
685 spin_unlock(&dentry
->d_lock
);
686 spin_unlock(&inode
->i_lock
);
690 spin_unlock(&dentry
->d_lock
);
692 spin_unlock(&inode
->i_lock
);
694 EXPORT_SYMBOL(d_prune_aliases
);
697 * Try to throw away a dentry - free the inode, dput the parent.
698 * Requires dentry->d_lock is held, and dentry->d_count == 0.
699 * Releases dentry->d_lock.
701 * This may fail if locks cannot be acquired no problem, just try again.
703 static void try_prune_one_dentry(struct dentry
*dentry
)
704 __releases(dentry
->d_lock
)
706 struct dentry
*parent
;
708 parent
= dentry_kill(dentry
, 0);
710 * If dentry_kill returns NULL, we have nothing more to do.
711 * if it returns the same dentry, trylocks failed. In either
712 * case, just loop again.
714 * Otherwise, we need to prune ancestors too. This is necessary
715 * to prevent quadratic behavior of shrink_dcache_parent(), but
716 * is also expected to be beneficial in reducing dentry cache
721 if (parent
== dentry
)
724 /* Prune ancestors. */
727 spin_lock(&dentry
->d_lock
);
728 if (dentry
->d_count
> 1) {
730 spin_unlock(&dentry
->d_lock
);
733 dentry
= dentry_kill(dentry
, 1);
737 static void shrink_dentry_list(struct list_head
*list
)
739 struct dentry
*dentry
;
743 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
744 if (&dentry
->d_lru
== list
)
746 spin_lock(&dentry
->d_lock
);
747 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
748 spin_unlock(&dentry
->d_lock
);
753 * We found an inuse dentry which was not removed from
754 * the LRU because of laziness during lookup. Do not free
755 * it - just keep it off the LRU list.
757 if (dentry
->d_count
) {
758 dentry_lru_del(dentry
);
759 spin_unlock(&dentry
->d_lock
);
765 try_prune_one_dentry(dentry
);
773 * prune_dcache_sb - shrink the dcache
775 * @count: number of entries to try to free
777 * Attempt to shrink the superblock dcache LRU by @count entries. This is
778 * done when we need more memory an called from the superblock shrinker
781 * This function may fail to free any resources if all the dentries are in
784 void prune_dcache_sb(struct super_block
*sb
, int count
)
786 struct dentry
*dentry
;
787 LIST_HEAD(referenced
);
791 spin_lock(&dcache_lru_lock
);
792 while (!list_empty(&sb
->s_dentry_lru
)) {
793 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
794 struct dentry
, d_lru
);
795 BUG_ON(dentry
->d_sb
!= sb
);
797 if (!spin_trylock(&dentry
->d_lock
)) {
798 spin_unlock(&dcache_lru_lock
);
803 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
804 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
805 list_move(&dentry
->d_lru
, &referenced
);
806 spin_unlock(&dentry
->d_lock
);
808 list_move_tail(&dentry
->d_lru
, &tmp
);
809 spin_unlock(&dentry
->d_lock
);
813 cond_resched_lock(&dcache_lru_lock
);
815 if (!list_empty(&referenced
))
816 list_splice(&referenced
, &sb
->s_dentry_lru
);
817 spin_unlock(&dcache_lru_lock
);
819 shrink_dentry_list(&tmp
);
823 * shrink_dcache_sb - shrink dcache for a superblock
826 * Shrink the dcache for the specified super block. This is used to free
827 * the dcache before unmounting a file system.
829 void shrink_dcache_sb(struct super_block
*sb
)
833 spin_lock(&dcache_lru_lock
);
834 while (!list_empty(&sb
->s_dentry_lru
)) {
835 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
836 spin_unlock(&dcache_lru_lock
);
837 shrink_dentry_list(&tmp
);
838 spin_lock(&dcache_lru_lock
);
840 spin_unlock(&dcache_lru_lock
);
842 EXPORT_SYMBOL(shrink_dcache_sb
);
845 * destroy a single subtree of dentries for unmount
846 * - see the comments on shrink_dcache_for_umount() for a description of the
849 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
851 struct dentry
*parent
;
853 BUG_ON(!IS_ROOT(dentry
));
856 /* descend to the first leaf in the current subtree */
857 while (!list_empty(&dentry
->d_subdirs
))
858 dentry
= list_entry(dentry
->d_subdirs
.next
,
859 struct dentry
, d_u
.d_child
);
861 /* consume the dentries from this leaf up through its parents
862 * until we find one with children or run out altogether */
867 * remove the dentry from the lru, and inform
868 * the fs that this dentry is about to be
869 * unhashed and destroyed.
871 dentry_lru_prune(dentry
);
874 if (dentry
->d_count
!= 0) {
876 "BUG: Dentry %p{i=%lx,n=%s}"
878 " [unmount of %s %s]\n",
881 dentry
->d_inode
->i_ino
: 0UL,
884 dentry
->d_sb
->s_type
->name
,
889 if (IS_ROOT(dentry
)) {
891 list_del(&dentry
->d_u
.d_child
);
893 parent
= dentry
->d_parent
;
895 list_del(&dentry
->d_u
.d_child
);
898 inode
= dentry
->d_inode
;
900 dentry
->d_inode
= NULL
;
901 list_del_init(&dentry
->d_alias
);
902 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
903 dentry
->d_op
->d_iput(dentry
, inode
);
910 /* finished when we fall off the top of the tree,
911 * otherwise we ascend to the parent and move to the
912 * next sibling if there is one */
916 } while (list_empty(&dentry
->d_subdirs
));
918 dentry
= list_entry(dentry
->d_subdirs
.next
,
919 struct dentry
, d_u
.d_child
);
924 * destroy the dentries attached to a superblock on unmounting
925 * - we don't need to use dentry->d_lock because:
926 * - the superblock is detached from all mountings and open files, so the
927 * dentry trees will not be rearranged by the VFS
928 * - s_umount is write-locked, so the memory pressure shrinker will ignore
929 * any dentries belonging to this superblock that it comes across
930 * - the filesystem itself is no longer permitted to rearrange the dentries
933 void shrink_dcache_for_umount(struct super_block
*sb
)
935 struct dentry
*dentry
;
937 if (down_read_trylock(&sb
->s_umount
))
943 shrink_dcache_for_umount_subtree(dentry
);
945 while (!hlist_bl_empty(&sb
->s_anon
)) {
946 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
947 shrink_dcache_for_umount_subtree(dentry
);
952 * This tries to ascend one level of parenthood, but
953 * we can race with renaming, so we need to re-check
954 * the parenthood after dropping the lock and check
955 * that the sequence number still matches.
957 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
959 struct dentry
*new = old
->d_parent
;
962 spin_unlock(&old
->d_lock
);
963 spin_lock(&new->d_lock
);
966 * might go back up the wrong parent if we have had a rename
969 if (new != old
->d_parent
||
970 (old
->d_flags
& DCACHE_DISCONNECTED
) ||
971 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
972 spin_unlock(&new->d_lock
);
981 * Search for at least 1 mount point in the dentry's subdirs.
982 * We descend to the next level whenever the d_subdirs
983 * list is non-empty and continue searching.
987 * have_submounts - check for mounts over a dentry
988 * @parent: dentry to check.
990 * Return true if the parent or its subdirectories contain
993 int have_submounts(struct dentry
*parent
)
995 struct dentry
*this_parent
;
996 struct list_head
*next
;
1000 seq
= read_seqbegin(&rename_lock
);
1002 this_parent
= parent
;
1004 if (d_mountpoint(parent
))
1006 spin_lock(&this_parent
->d_lock
);
1008 next
= this_parent
->d_subdirs
.next
;
1010 while (next
!= &this_parent
->d_subdirs
) {
1011 struct list_head
*tmp
= next
;
1012 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1015 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1016 /* Have we found a mount point ? */
1017 if (d_mountpoint(dentry
)) {
1018 spin_unlock(&dentry
->d_lock
);
1019 spin_unlock(&this_parent
->d_lock
);
1022 if (!list_empty(&dentry
->d_subdirs
)) {
1023 spin_unlock(&this_parent
->d_lock
);
1024 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1025 this_parent
= dentry
;
1026 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1029 spin_unlock(&dentry
->d_lock
);
1032 * All done at this level ... ascend and resume the search.
1034 if (this_parent
!= parent
) {
1035 struct dentry
*child
= this_parent
;
1036 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1039 next
= child
->d_u
.d_child
.next
;
1042 spin_unlock(&this_parent
->d_lock
);
1043 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1046 write_sequnlock(&rename_lock
);
1047 return 0; /* No mount points found in tree */
1049 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1052 write_sequnlock(&rename_lock
);
1057 write_seqlock(&rename_lock
);
1060 EXPORT_SYMBOL(have_submounts
);
1063 * Search the dentry child list for the specified parent,
1064 * and move any unused dentries to the end of the unused
1065 * list for prune_dcache(). We descend to the next level
1066 * whenever the d_subdirs list is non-empty and continue
1069 * It returns zero iff there are no unused children,
1070 * otherwise it returns the number of children moved to
1071 * the end of the unused list. This may not be the total
1072 * number of unused children, because select_parent can
1073 * drop the lock and return early due to latency
1076 static int select_parent(struct dentry
*parent
, struct list_head
*dispose
)
1078 struct dentry
*this_parent
;
1079 struct list_head
*next
;
1084 seq
= read_seqbegin(&rename_lock
);
1086 this_parent
= parent
;
1087 spin_lock(&this_parent
->d_lock
);
1089 next
= this_parent
->d_subdirs
.next
;
1091 while (next
!= &this_parent
->d_subdirs
) {
1092 struct list_head
*tmp
= next
;
1093 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1096 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1099 * move only zero ref count dentries to the dispose list.
1101 if (!dentry
->d_count
) {
1102 dentry_lru_move_list(dentry
, dispose
);
1105 dentry_lru_del(dentry
);
1109 * We can return to the caller if we have found some (this
1110 * ensures forward progress). We'll be coming back to find
1113 if (found
&& need_resched()) {
1114 spin_unlock(&dentry
->d_lock
);
1119 * Descend a level if the d_subdirs list is non-empty.
1121 if (!list_empty(&dentry
->d_subdirs
)) {
1122 spin_unlock(&this_parent
->d_lock
);
1123 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1124 this_parent
= dentry
;
1125 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1129 spin_unlock(&dentry
->d_lock
);
1132 * All done at this level ... ascend and resume the search.
1134 if (this_parent
!= parent
) {
1135 struct dentry
*child
= this_parent
;
1136 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1139 next
= child
->d_u
.d_child
.next
;
1143 spin_unlock(&this_parent
->d_lock
);
1144 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1147 write_sequnlock(&rename_lock
);
1154 write_seqlock(&rename_lock
);
1159 * shrink_dcache_parent - prune dcache
1160 * @parent: parent of entries to prune
1162 * Prune the dcache to remove unused children of the parent dentry.
1164 void shrink_dcache_parent(struct dentry
* parent
)
1169 while ((found
= select_parent(parent
, &dispose
)) != 0)
1170 shrink_dentry_list(&dispose
);
1172 EXPORT_SYMBOL(shrink_dcache_parent
);
1175 * __d_alloc - allocate a dcache entry
1176 * @sb: filesystem it will belong to
1177 * @name: qstr of the name
1179 * Allocates a dentry. It returns %NULL if there is insufficient memory
1180 * available. On a success the dentry is returned. The name passed in is
1181 * copied and the copy passed in may be reused after this call.
1184 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1186 struct dentry
*dentry
;
1189 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1193 if (name
->len
> DNAME_INLINE_LEN
-1) {
1194 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1196 kmem_cache_free(dentry_cache
, dentry
);
1200 dname
= dentry
->d_iname
;
1202 dentry
->d_name
.name
= dname
;
1204 dentry
->d_name
.len
= name
->len
;
1205 dentry
->d_name
.hash
= name
->hash
;
1206 memcpy(dname
, name
->name
, name
->len
);
1207 dname
[name
->len
] = 0;
1209 dentry
->d_count
= 1;
1210 dentry
->d_flags
= 0;
1211 spin_lock_init(&dentry
->d_lock
);
1212 seqcount_init(&dentry
->d_seq
);
1213 dentry
->d_inode
= NULL
;
1214 dentry
->d_parent
= dentry
;
1216 dentry
->d_op
= NULL
;
1217 dentry
->d_fsdata
= NULL
;
1218 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1219 INIT_LIST_HEAD(&dentry
->d_lru
);
1220 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1221 INIT_LIST_HEAD(&dentry
->d_alias
);
1222 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1223 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1225 this_cpu_inc(nr_dentry
);
1231 * d_alloc - allocate a dcache entry
1232 * @parent: parent of entry to allocate
1233 * @name: qstr of the name
1235 * Allocates a dentry. It returns %NULL if there is insufficient memory
1236 * available. On a success the dentry is returned. The name passed in is
1237 * copied and the copy passed in may be reused after this call.
1239 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1241 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1245 spin_lock(&parent
->d_lock
);
1247 * don't need child lock because it is not subject
1248 * to concurrency here
1250 __dget_dlock(parent
);
1251 dentry
->d_parent
= parent
;
1252 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1253 spin_unlock(&parent
->d_lock
);
1257 EXPORT_SYMBOL(d_alloc
);
1259 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1261 struct dentry
*dentry
= __d_alloc(sb
, name
);
1263 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1266 EXPORT_SYMBOL(d_alloc_pseudo
);
1268 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1273 q
.len
= strlen(name
);
1274 q
.hash
= full_name_hash(q
.name
, q
.len
);
1275 return d_alloc(parent
, &q
);
1277 EXPORT_SYMBOL(d_alloc_name
);
1279 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1281 WARN_ON_ONCE(dentry
->d_op
);
1282 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1284 DCACHE_OP_REVALIDATE
|
1285 DCACHE_OP_DELETE
));
1290 dentry
->d_flags
|= DCACHE_OP_HASH
;
1292 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1293 if (op
->d_revalidate
)
1294 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1296 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1298 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1301 EXPORT_SYMBOL(d_set_d_op
);
1303 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1305 spin_lock(&dentry
->d_lock
);
1307 if (unlikely(IS_AUTOMOUNT(inode
)))
1308 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1309 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1311 dentry
->d_inode
= inode
;
1312 dentry_rcuwalk_barrier(dentry
);
1313 spin_unlock(&dentry
->d_lock
);
1314 fsnotify_d_instantiate(dentry
, inode
);
1318 * d_instantiate - fill in inode information for a dentry
1319 * @entry: dentry to complete
1320 * @inode: inode to attach to this dentry
1322 * Fill in inode information in the entry.
1324 * This turns negative dentries into productive full members
1327 * NOTE! This assumes that the inode count has been incremented
1328 * (or otherwise set) by the caller to indicate that it is now
1329 * in use by the dcache.
1332 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1334 BUG_ON(!list_empty(&entry
->d_alias
));
1336 spin_lock(&inode
->i_lock
);
1337 __d_instantiate(entry
, inode
);
1339 spin_unlock(&inode
->i_lock
);
1340 security_d_instantiate(entry
, inode
);
1342 EXPORT_SYMBOL(d_instantiate
);
1345 * d_instantiate_unique - instantiate a non-aliased dentry
1346 * @entry: dentry to instantiate
1347 * @inode: inode to attach to this dentry
1349 * Fill in inode information in the entry. On success, it returns NULL.
1350 * If an unhashed alias of "entry" already exists, then we return the
1351 * aliased dentry instead and drop one reference to inode.
1353 * Note that in order to avoid conflicts with rename() etc, the caller
1354 * had better be holding the parent directory semaphore.
1356 * This also assumes that the inode count has been incremented
1357 * (or otherwise set) by the caller to indicate that it is now
1358 * in use by the dcache.
1360 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1361 struct inode
*inode
)
1363 struct dentry
*alias
;
1364 int len
= entry
->d_name
.len
;
1365 const char *name
= entry
->d_name
.name
;
1366 unsigned int hash
= entry
->d_name
.hash
;
1369 __d_instantiate(entry
, NULL
);
1373 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1374 struct qstr
*qstr
= &alias
->d_name
;
1377 * Don't need alias->d_lock here, because aliases with
1378 * d_parent == entry->d_parent are not subject to name or
1379 * parent changes, because the parent inode i_mutex is held.
1381 if (qstr
->hash
!= hash
)
1383 if (alias
->d_parent
!= entry
->d_parent
)
1385 if (dentry_cmp(qstr
->name
, qstr
->len
, name
, len
))
1391 __d_instantiate(entry
, inode
);
1395 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1397 struct dentry
*result
;
1399 BUG_ON(!list_empty(&entry
->d_alias
));
1402 spin_lock(&inode
->i_lock
);
1403 result
= __d_instantiate_unique(entry
, inode
);
1405 spin_unlock(&inode
->i_lock
);
1408 security_d_instantiate(entry
, inode
);
1412 BUG_ON(!d_unhashed(result
));
1417 EXPORT_SYMBOL(d_instantiate_unique
);
1420 * d_alloc_root - allocate root dentry
1421 * @root_inode: inode to allocate the root for
1423 * Allocate a root ("/") dentry for the inode given. The inode is
1424 * instantiated and returned. %NULL is returned if there is insufficient
1425 * memory or the inode passed is %NULL.
1428 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1430 struct dentry
*res
= NULL
;
1433 static const struct qstr name
= { .name
= "/", .len
= 1 };
1435 res
= __d_alloc(root_inode
->i_sb
, &name
);
1437 d_instantiate(res
, root_inode
);
1441 EXPORT_SYMBOL(d_alloc_root
);
1443 struct dentry
*d_make_root(struct inode
*root_inode
)
1445 struct dentry
*res
= NULL
;
1448 static const struct qstr name
= { .name
= "/", .len
= 1 };
1450 res
= __d_alloc(root_inode
->i_sb
, &name
);
1452 d_instantiate(res
, root_inode
);
1458 EXPORT_SYMBOL(d_make_root
);
1460 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1462 struct dentry
*alias
;
1464 if (list_empty(&inode
->i_dentry
))
1466 alias
= list_first_entry(&inode
->i_dentry
, struct dentry
, d_alias
);
1471 static struct dentry
* d_find_any_alias(struct inode
*inode
)
1475 spin_lock(&inode
->i_lock
);
1476 de
= __d_find_any_alias(inode
);
1477 spin_unlock(&inode
->i_lock
);
1483 * d_obtain_alias - find or allocate a dentry for a given inode
1484 * @inode: inode to allocate the dentry for
1486 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1487 * similar open by handle operations. The returned dentry may be anonymous,
1488 * or may have a full name (if the inode was already in the cache).
1490 * When called on a directory inode, we must ensure that the inode only ever
1491 * has one dentry. If a dentry is found, that is returned instead of
1492 * allocating a new one.
1494 * On successful return, the reference to the inode has been transferred
1495 * to the dentry. In case of an error the reference on the inode is released.
1496 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1497 * be passed in and will be the error will be propagate to the return value,
1498 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1500 struct dentry
*d_obtain_alias(struct inode
*inode
)
1502 static const struct qstr anonstring
= { .name
= "" };
1507 return ERR_PTR(-ESTALE
);
1509 return ERR_CAST(inode
);
1511 res
= d_find_any_alias(inode
);
1515 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1517 res
= ERR_PTR(-ENOMEM
);
1521 spin_lock(&inode
->i_lock
);
1522 res
= __d_find_any_alias(inode
);
1524 spin_unlock(&inode
->i_lock
);
1529 /* attach a disconnected dentry */
1530 spin_lock(&tmp
->d_lock
);
1531 tmp
->d_inode
= inode
;
1532 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1533 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1534 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1535 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1536 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1537 spin_unlock(&tmp
->d_lock
);
1538 spin_unlock(&inode
->i_lock
);
1539 security_d_instantiate(tmp
, inode
);
1544 if (res
&& !IS_ERR(res
))
1545 security_d_instantiate(res
, inode
);
1549 EXPORT_SYMBOL(d_obtain_alias
);
1552 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1553 * @inode: the inode which may have a disconnected dentry
1554 * @dentry: a negative dentry which we want to point to the inode.
1556 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1557 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1558 * and return it, else simply d_add the inode to the dentry and return NULL.
1560 * This is needed in the lookup routine of any filesystem that is exportable
1561 * (via knfsd) so that we can build dcache paths to directories effectively.
1563 * If a dentry was found and moved, then it is returned. Otherwise NULL
1564 * is returned. This matches the expected return value of ->lookup.
1567 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1569 struct dentry
*new = NULL
;
1572 return ERR_CAST(inode
);
1574 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1575 spin_lock(&inode
->i_lock
);
1576 new = __d_find_alias(inode
, 1);
1578 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1579 spin_unlock(&inode
->i_lock
);
1580 security_d_instantiate(new, inode
);
1581 d_move(new, dentry
);
1584 /* already taking inode->i_lock, so d_add() by hand */
1585 __d_instantiate(dentry
, inode
);
1586 spin_unlock(&inode
->i_lock
);
1587 security_d_instantiate(dentry
, inode
);
1591 d_add(dentry
, inode
);
1594 EXPORT_SYMBOL(d_splice_alias
);
1597 * d_add_ci - lookup or allocate new dentry with case-exact name
1598 * @inode: the inode case-insensitive lookup has found
1599 * @dentry: the negative dentry that was passed to the parent's lookup func
1600 * @name: the case-exact name to be associated with the returned dentry
1602 * This is to avoid filling the dcache with case-insensitive names to the
1603 * same inode, only the actual correct case is stored in the dcache for
1604 * case-insensitive filesystems.
1606 * For a case-insensitive lookup match and if the the case-exact dentry
1607 * already exists in in the dcache, use it and return it.
1609 * If no entry exists with the exact case name, allocate new dentry with
1610 * the exact case, and return the spliced entry.
1612 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1616 struct dentry
*found
;
1620 * First check if a dentry matching the name already exists,
1621 * if not go ahead and create it now.
1623 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1625 new = d_alloc(dentry
->d_parent
, name
);
1631 found
= d_splice_alias(inode
, new);
1640 * If a matching dentry exists, and it's not negative use it.
1642 * Decrement the reference count to balance the iget() done
1645 if (found
->d_inode
) {
1646 if (unlikely(found
->d_inode
!= inode
)) {
1647 /* This can't happen because bad inodes are unhashed. */
1648 BUG_ON(!is_bad_inode(inode
));
1649 BUG_ON(!is_bad_inode(found
->d_inode
));
1656 * We are going to instantiate this dentry, unhash it and clear the
1657 * lookup flag so we can do that.
1659 if (unlikely(d_need_lookup(found
)))
1660 d_clear_need_lookup(found
);
1663 * Negative dentry: instantiate it unless the inode is a directory and
1664 * already has a dentry.
1666 new = d_splice_alias(inode
, found
);
1675 return ERR_PTR(error
);
1677 EXPORT_SYMBOL(d_add_ci
);
1680 * __d_lookup_rcu - search for a dentry (racy, store-free)
1681 * @parent: parent dentry
1682 * @name: qstr of name we wish to find
1683 * @seq: returns d_seq value at the point where the dentry was found
1684 * @inode: returns dentry->d_inode when the inode was found valid.
1685 * Returns: dentry, or NULL
1687 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1688 * resolution (store-free path walking) design described in
1689 * Documentation/filesystems/path-lookup.txt.
1691 * This is not to be used outside core vfs.
1693 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1694 * held, and rcu_read_lock held. The returned dentry must not be stored into
1695 * without taking d_lock and checking d_seq sequence count against @seq
1698 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1701 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1702 * the returned dentry, so long as its parent's seqlock is checked after the
1703 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1704 * is formed, giving integrity down the path walk.
1706 struct dentry
*__d_lookup_rcu(struct dentry
*parent
, struct qstr
*name
,
1707 unsigned *seq
, struct inode
**inode
)
1709 unsigned int len
= name
->len
;
1710 unsigned int hash
= name
->hash
;
1711 const unsigned char *str
= name
->name
;
1712 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1713 struct hlist_bl_node
*node
;
1714 struct dentry
*dentry
;
1717 * Note: There is significant duplication with __d_lookup_rcu which is
1718 * required to prevent single threaded performance regressions
1719 * especially on architectures where smp_rmb (in seqcounts) are costly.
1720 * Keep the two functions in sync.
1724 * The hash list is protected using RCU.
1726 * Carefully use d_seq when comparing a candidate dentry, to avoid
1727 * races with d_move().
1729 * It is possible that concurrent renames can mess up our list
1730 * walk here and result in missing our dentry, resulting in the
1731 * false-negative result. d_lookup() protects against concurrent
1732 * renames using rename_lock seqlock.
1734 * See Documentation/filesystems/path-lookup.txt for more details.
1736 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1741 if (dentry
->d_name
.hash
!= hash
)
1745 *seq
= read_seqcount_begin(&dentry
->d_seq
);
1746 if (dentry
->d_parent
!= parent
)
1748 if (d_unhashed(dentry
))
1750 tlen
= dentry
->d_name
.len
;
1751 tname
= dentry
->d_name
.name
;
1752 i
= dentry
->d_inode
;
1755 * This seqcount check is required to ensure name and
1756 * len are loaded atomically, so as not to walk off the
1757 * edge of memory when walking. If we could load this
1758 * atomically some other way, we could drop this check.
1760 if (read_seqcount_retry(&dentry
->d_seq
, *seq
))
1762 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1763 if (parent
->d_op
->d_compare(parent
, *inode
,
1768 if (dentry_cmp(tname
, tlen
, str
, len
))
1772 * No extra seqcount check is required after the name
1773 * compare. The caller must perform a seqcount check in
1774 * order to do anything useful with the returned dentry
1784 * d_lookup - search for a dentry
1785 * @parent: parent dentry
1786 * @name: qstr of name we wish to find
1787 * Returns: dentry, or NULL
1789 * d_lookup searches the children of the parent dentry for the name in
1790 * question. If the dentry is found its reference count is incremented and the
1791 * dentry is returned. The caller must use dput to free the entry when it has
1792 * finished using it. %NULL is returned if the dentry does not exist.
1794 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1796 struct dentry
*dentry
;
1800 seq
= read_seqbegin(&rename_lock
);
1801 dentry
= __d_lookup(parent
, name
);
1804 } while (read_seqretry(&rename_lock
, seq
));
1807 EXPORT_SYMBOL(d_lookup
);
1810 * __d_lookup - search for a dentry (racy)
1811 * @parent: parent dentry
1812 * @name: qstr of name we wish to find
1813 * Returns: dentry, or NULL
1815 * __d_lookup is like d_lookup, however it may (rarely) return a
1816 * false-negative result due to unrelated rename activity.
1818 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1819 * however it must be used carefully, eg. with a following d_lookup in
1820 * the case of failure.
1822 * __d_lookup callers must be commented.
1824 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1826 unsigned int len
= name
->len
;
1827 unsigned int hash
= name
->hash
;
1828 const unsigned char *str
= name
->name
;
1829 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1830 struct hlist_bl_node
*node
;
1831 struct dentry
*found
= NULL
;
1832 struct dentry
*dentry
;
1835 * Note: There is significant duplication with __d_lookup_rcu which is
1836 * required to prevent single threaded performance regressions
1837 * especially on architectures where smp_rmb (in seqcounts) are costly.
1838 * Keep the two functions in sync.
1842 * The hash list is protected using RCU.
1844 * Take d_lock when comparing a candidate dentry, to avoid races
1847 * It is possible that concurrent renames can mess up our list
1848 * walk here and result in missing our dentry, resulting in the
1849 * false-negative result. d_lookup() protects against concurrent
1850 * renames using rename_lock seqlock.
1852 * See Documentation/filesystems/path-lookup.txt for more details.
1856 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1860 if (dentry
->d_name
.hash
!= hash
)
1863 spin_lock(&dentry
->d_lock
);
1864 if (dentry
->d_parent
!= parent
)
1866 if (d_unhashed(dentry
))
1870 * It is safe to compare names since d_move() cannot
1871 * change the qstr (protected by d_lock).
1873 tlen
= dentry
->d_name
.len
;
1874 tname
= dentry
->d_name
.name
;
1875 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1876 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1877 dentry
, dentry
->d_inode
,
1881 if (dentry_cmp(tname
, tlen
, str
, len
))
1887 spin_unlock(&dentry
->d_lock
);
1890 spin_unlock(&dentry
->d_lock
);
1898 * d_hash_and_lookup - hash the qstr then search for a dentry
1899 * @dir: Directory to search in
1900 * @name: qstr of name we wish to find
1902 * On hash failure or on lookup failure NULL is returned.
1904 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1906 struct dentry
*dentry
= NULL
;
1909 * Check for a fs-specific hash function. Note that we must
1910 * calculate the standard hash first, as the d_op->d_hash()
1911 * routine may choose to leave the hash value unchanged.
1913 name
->hash
= full_name_hash(name
->name
, name
->len
);
1914 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1915 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1918 dentry
= d_lookup(dir
, name
);
1924 * d_validate - verify dentry provided from insecure source (deprecated)
1925 * @dentry: The dentry alleged to be valid child of @dparent
1926 * @dparent: The parent dentry (known to be valid)
1928 * An insecure source has sent us a dentry, here we verify it and dget() it.
1929 * This is used by ncpfs in its readdir implementation.
1930 * Zero is returned in the dentry is invalid.
1932 * This function is slow for big directories, and deprecated, do not use it.
1934 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1936 struct dentry
*child
;
1938 spin_lock(&dparent
->d_lock
);
1939 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1940 if (dentry
== child
) {
1941 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1942 __dget_dlock(dentry
);
1943 spin_unlock(&dentry
->d_lock
);
1944 spin_unlock(&dparent
->d_lock
);
1948 spin_unlock(&dparent
->d_lock
);
1952 EXPORT_SYMBOL(d_validate
);
1955 * When a file is deleted, we have two options:
1956 * - turn this dentry into a negative dentry
1957 * - unhash this dentry and free it.
1959 * Usually, we want to just turn this into
1960 * a negative dentry, but if anybody else is
1961 * currently using the dentry or the inode
1962 * we can't do that and we fall back on removing
1963 * it from the hash queues and waiting for
1964 * it to be deleted later when it has no users
1968 * d_delete - delete a dentry
1969 * @dentry: The dentry to delete
1971 * Turn the dentry into a negative dentry if possible, otherwise
1972 * remove it from the hash queues so it can be deleted later
1975 void d_delete(struct dentry
* dentry
)
1977 struct inode
*inode
;
1980 * Are we the only user?
1983 spin_lock(&dentry
->d_lock
);
1984 inode
= dentry
->d_inode
;
1985 isdir
= S_ISDIR(inode
->i_mode
);
1986 if (dentry
->d_count
== 1) {
1987 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
1988 spin_unlock(&dentry
->d_lock
);
1992 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1993 dentry_unlink_inode(dentry
);
1994 fsnotify_nameremove(dentry
, isdir
);
1998 if (!d_unhashed(dentry
))
2001 spin_unlock(&dentry
->d_lock
);
2003 fsnotify_nameremove(dentry
, isdir
);
2005 EXPORT_SYMBOL(d_delete
);
2007 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2009 BUG_ON(!d_unhashed(entry
));
2011 entry
->d_flags
|= DCACHE_RCUACCESS
;
2012 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2016 static void _d_rehash(struct dentry
* entry
)
2018 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2022 * d_rehash - add an entry back to the hash
2023 * @entry: dentry to add to the hash
2025 * Adds a dentry to the hash according to its name.
2028 void d_rehash(struct dentry
* entry
)
2030 spin_lock(&entry
->d_lock
);
2032 spin_unlock(&entry
->d_lock
);
2034 EXPORT_SYMBOL(d_rehash
);
2037 * dentry_update_name_case - update case insensitive dentry with a new name
2038 * @dentry: dentry to be updated
2041 * Update a case insensitive dentry with new case of name.
2043 * dentry must have been returned by d_lookup with name @name. Old and new
2044 * name lengths must match (ie. no d_compare which allows mismatched name
2047 * Parent inode i_mutex must be held over d_lookup and into this call (to
2048 * keep renames and concurrent inserts, and readdir(2) away).
2050 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2052 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2053 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2055 spin_lock(&dentry
->d_lock
);
2056 write_seqcount_begin(&dentry
->d_seq
);
2057 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2058 write_seqcount_end(&dentry
->d_seq
);
2059 spin_unlock(&dentry
->d_lock
);
2061 EXPORT_SYMBOL(dentry_update_name_case
);
2063 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2065 if (dname_external(target
)) {
2066 if (dname_external(dentry
)) {
2068 * Both external: swap the pointers
2070 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2073 * dentry:internal, target:external. Steal target's
2074 * storage and make target internal.
2076 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2077 dentry
->d_name
.len
+ 1);
2078 dentry
->d_name
.name
= target
->d_name
.name
;
2079 target
->d_name
.name
= target
->d_iname
;
2082 if (dname_external(dentry
)) {
2084 * dentry:external, target:internal. Give dentry's
2085 * storage to target and make dentry internal
2087 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2088 target
->d_name
.len
+ 1);
2089 target
->d_name
.name
= dentry
->d_name
.name
;
2090 dentry
->d_name
.name
= dentry
->d_iname
;
2093 * Both are internal. Just copy target to dentry
2095 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2096 target
->d_name
.len
+ 1);
2097 dentry
->d_name
.len
= target
->d_name
.len
;
2101 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2104 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2107 * XXXX: do we really need to take target->d_lock?
2109 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2110 spin_lock(&target
->d_parent
->d_lock
);
2112 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2113 spin_lock(&dentry
->d_parent
->d_lock
);
2114 spin_lock_nested(&target
->d_parent
->d_lock
,
2115 DENTRY_D_LOCK_NESTED
);
2117 spin_lock(&target
->d_parent
->d_lock
);
2118 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2119 DENTRY_D_LOCK_NESTED
);
2122 if (target
< dentry
) {
2123 spin_lock_nested(&target
->d_lock
, 2);
2124 spin_lock_nested(&dentry
->d_lock
, 3);
2126 spin_lock_nested(&dentry
->d_lock
, 2);
2127 spin_lock_nested(&target
->d_lock
, 3);
2131 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2132 struct dentry
*target
)
2134 if (target
->d_parent
!= dentry
->d_parent
)
2135 spin_unlock(&dentry
->d_parent
->d_lock
);
2136 if (target
->d_parent
!= target
)
2137 spin_unlock(&target
->d_parent
->d_lock
);
2141 * When switching names, the actual string doesn't strictly have to
2142 * be preserved in the target - because we're dropping the target
2143 * anyway. As such, we can just do a simple memcpy() to copy over
2144 * the new name before we switch.
2146 * Note that we have to be a lot more careful about getting the hash
2147 * switched - we have to switch the hash value properly even if it
2148 * then no longer matches the actual (corrupted) string of the target.
2149 * The hash value has to match the hash queue that the dentry is on..
2152 * __d_move - move a dentry
2153 * @dentry: entry to move
2154 * @target: new dentry
2156 * Update the dcache to reflect the move of a file name. Negative
2157 * dcache entries should not be moved in this way. Caller must hold
2158 * rename_lock, the i_mutex of the source and target directories,
2159 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2161 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2163 if (!dentry
->d_inode
)
2164 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2166 BUG_ON(d_ancestor(dentry
, target
));
2167 BUG_ON(d_ancestor(target
, dentry
));
2169 dentry_lock_for_move(dentry
, target
);
2171 write_seqcount_begin(&dentry
->d_seq
);
2172 write_seqcount_begin(&target
->d_seq
);
2174 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2177 * Move the dentry to the target hash queue. Don't bother checking
2178 * for the same hash queue because of how unlikely it is.
2181 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2183 /* Unhash the target: dput() will then get rid of it */
2186 list_del(&dentry
->d_u
.d_child
);
2187 list_del(&target
->d_u
.d_child
);
2189 /* Switch the names.. */
2190 switch_names(dentry
, target
);
2191 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2193 /* ... and switch the parents */
2194 if (IS_ROOT(dentry
)) {
2195 dentry
->d_parent
= target
->d_parent
;
2196 target
->d_parent
= target
;
2197 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2199 swap(dentry
->d_parent
, target
->d_parent
);
2201 /* And add them back to the (new) parent lists */
2202 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2205 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2207 write_seqcount_end(&target
->d_seq
);
2208 write_seqcount_end(&dentry
->d_seq
);
2210 dentry_unlock_parents_for_move(dentry
, target
);
2211 spin_unlock(&target
->d_lock
);
2212 fsnotify_d_move(dentry
);
2213 spin_unlock(&dentry
->d_lock
);
2217 * d_move - move a dentry
2218 * @dentry: entry to move
2219 * @target: new dentry
2221 * Update the dcache to reflect the move of a file name. Negative
2222 * dcache entries should not be moved in this way. See the locking
2223 * requirements for __d_move.
2225 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2227 write_seqlock(&rename_lock
);
2228 __d_move(dentry
, target
);
2229 write_sequnlock(&rename_lock
);
2231 EXPORT_SYMBOL(d_move
);
2234 * d_ancestor - search for an ancestor
2235 * @p1: ancestor dentry
2238 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2239 * an ancestor of p2, else NULL.
2241 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2245 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2246 if (p
->d_parent
== p1
)
2253 * This helper attempts to cope with remotely renamed directories
2255 * It assumes that the caller is already holding
2256 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2258 * Note: If ever the locking in lock_rename() changes, then please
2259 * remember to update this too...
2261 static struct dentry
*__d_unalias(struct inode
*inode
,
2262 struct dentry
*dentry
, struct dentry
*alias
)
2264 struct mutex
*m1
= NULL
, *m2
= NULL
;
2267 /* If alias and dentry share a parent, then no extra locks required */
2268 if (alias
->d_parent
== dentry
->d_parent
)
2271 /* See lock_rename() */
2272 ret
= ERR_PTR(-EBUSY
);
2273 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2275 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2276 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2278 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2280 __d_move(alias
, dentry
);
2283 spin_unlock(&inode
->i_lock
);
2292 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2293 * named dentry in place of the dentry to be replaced.
2294 * returns with anon->d_lock held!
2296 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2298 struct dentry
*dparent
, *aparent
;
2300 dentry_lock_for_move(anon
, dentry
);
2302 write_seqcount_begin(&dentry
->d_seq
);
2303 write_seqcount_begin(&anon
->d_seq
);
2305 dparent
= dentry
->d_parent
;
2306 aparent
= anon
->d_parent
;
2308 switch_names(dentry
, anon
);
2309 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2311 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2312 list_del(&dentry
->d_u
.d_child
);
2313 if (!IS_ROOT(dentry
))
2314 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2316 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2318 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2319 list_del(&anon
->d_u
.d_child
);
2321 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2323 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2325 write_seqcount_end(&dentry
->d_seq
);
2326 write_seqcount_end(&anon
->d_seq
);
2328 dentry_unlock_parents_for_move(anon
, dentry
);
2329 spin_unlock(&dentry
->d_lock
);
2331 /* anon->d_lock still locked, returns locked */
2332 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2336 * d_materialise_unique - introduce an inode into the tree
2337 * @dentry: candidate dentry
2338 * @inode: inode to bind to the dentry, to which aliases may be attached
2340 * Introduces an dentry into the tree, substituting an extant disconnected
2341 * root directory alias in its place if there is one. Caller must hold the
2342 * i_mutex of the parent directory.
2344 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2346 struct dentry
*actual
;
2348 BUG_ON(!d_unhashed(dentry
));
2352 __d_instantiate(dentry
, NULL
);
2357 spin_lock(&inode
->i_lock
);
2359 if (S_ISDIR(inode
->i_mode
)) {
2360 struct dentry
*alias
;
2362 /* Does an aliased dentry already exist? */
2363 alias
= __d_find_alias(inode
, 0);
2366 write_seqlock(&rename_lock
);
2368 if (d_ancestor(alias
, dentry
)) {
2369 /* Check for loops */
2370 actual
= ERR_PTR(-ELOOP
);
2371 } else if (IS_ROOT(alias
)) {
2372 /* Is this an anonymous mountpoint that we
2373 * could splice into our tree? */
2374 __d_materialise_dentry(dentry
, alias
);
2375 write_sequnlock(&rename_lock
);
2379 /* Nope, but we must(!) avoid directory
2381 actual
= __d_unalias(inode
, dentry
, alias
);
2383 write_sequnlock(&rename_lock
);
2384 if (IS_ERR(actual
)) {
2385 if (PTR_ERR(actual
) == -ELOOP
)
2386 pr_warn_ratelimited(
2387 "VFS: Lookup of '%s' in %s %s"
2388 " would have caused loop\n",
2389 dentry
->d_name
.name
,
2390 inode
->i_sb
->s_type
->name
,
2398 /* Add a unique reference */
2399 actual
= __d_instantiate_unique(dentry
, inode
);
2403 BUG_ON(!d_unhashed(actual
));
2405 spin_lock(&actual
->d_lock
);
2408 spin_unlock(&actual
->d_lock
);
2409 spin_unlock(&inode
->i_lock
);
2411 if (actual
== dentry
) {
2412 security_d_instantiate(dentry
, inode
);
2419 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2421 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2425 return -ENAMETOOLONG
;
2427 memcpy(*buffer
, str
, namelen
);
2431 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2433 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2437 * prepend_path - Prepend path string to a buffer
2438 * @path: the dentry/vfsmount to report
2439 * @root: root vfsmnt/dentry
2440 * @buffer: pointer to the end of the buffer
2441 * @buflen: pointer to buffer length
2443 * Caller holds the rename_lock.
2445 static int prepend_path(const struct path
*path
,
2446 const struct path
*root
,
2447 char **buffer
, int *buflen
)
2449 struct dentry
*dentry
= path
->dentry
;
2450 struct vfsmount
*vfsmnt
= path
->mnt
;
2451 struct mount
*mnt
= real_mount(vfsmnt
);
2455 br_read_lock(vfsmount_lock
);
2456 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2457 struct dentry
* parent
;
2459 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2461 if (!mnt_has_parent(mnt
))
2463 dentry
= mnt
->mnt_mountpoint
;
2464 mnt
= mnt
->mnt_parent
;
2468 parent
= dentry
->d_parent
;
2470 spin_lock(&dentry
->d_lock
);
2471 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2472 spin_unlock(&dentry
->d_lock
);
2474 error
= prepend(buffer
, buflen
, "/", 1);
2482 if (!error
&& !slash
)
2483 error
= prepend(buffer
, buflen
, "/", 1);
2486 br_read_unlock(vfsmount_lock
);
2491 * Filesystems needing to implement special "root names"
2492 * should do so with ->d_dname()
2494 if (IS_ROOT(dentry
) &&
2495 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2496 WARN(1, "Root dentry has weird name <%.*s>\n",
2497 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2500 error
= prepend(buffer
, buflen
, "/", 1);
2502 error
= real_mount(vfsmnt
)->mnt_ns
? 1 : 2;
2507 * __d_path - return the path of a dentry
2508 * @path: the dentry/vfsmount to report
2509 * @root: root vfsmnt/dentry
2510 * @buf: buffer to return value in
2511 * @buflen: buffer length
2513 * Convert a dentry into an ASCII path name.
2515 * Returns a pointer into the buffer or an error code if the
2516 * path was too long.
2518 * "buflen" should be positive.
2520 * If the path is not reachable from the supplied root, return %NULL.
2522 char *__d_path(const struct path
*path
,
2523 const struct path
*root
,
2524 char *buf
, int buflen
)
2526 char *res
= buf
+ buflen
;
2529 prepend(&res
, &buflen
, "\0", 1);
2530 write_seqlock(&rename_lock
);
2531 error
= prepend_path(path
, root
, &res
, &buflen
);
2532 write_sequnlock(&rename_lock
);
2535 return ERR_PTR(error
);
2541 char *d_absolute_path(const struct path
*path
,
2542 char *buf
, int buflen
)
2544 struct path root
= {};
2545 char *res
= buf
+ buflen
;
2548 prepend(&res
, &buflen
, "\0", 1);
2549 write_seqlock(&rename_lock
);
2550 error
= prepend_path(path
, &root
, &res
, &buflen
);
2551 write_sequnlock(&rename_lock
);
2556 return ERR_PTR(error
);
2561 * same as __d_path but appends "(deleted)" for unlinked files.
2563 static int path_with_deleted(const struct path
*path
,
2564 const struct path
*root
,
2565 char **buf
, int *buflen
)
2567 prepend(buf
, buflen
, "\0", 1);
2568 if (d_unlinked(path
->dentry
)) {
2569 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2574 return prepend_path(path
, root
, buf
, buflen
);
2577 static int prepend_unreachable(char **buffer
, int *buflen
)
2579 return prepend(buffer
, buflen
, "(unreachable)", 13);
2583 * d_path - return the path of a dentry
2584 * @path: path to report
2585 * @buf: buffer to return value in
2586 * @buflen: buffer length
2588 * Convert a dentry into an ASCII path name. If the entry has been deleted
2589 * the string " (deleted)" is appended. Note that this is ambiguous.
2591 * Returns a pointer into the buffer or an error code if the path was
2592 * too long. Note: Callers should use the returned pointer, not the passed
2593 * in buffer, to use the name! The implementation often starts at an offset
2594 * into the buffer, and may leave 0 bytes at the start.
2596 * "buflen" should be positive.
2598 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2600 char *res
= buf
+ buflen
;
2605 * We have various synthetic filesystems that never get mounted. On
2606 * these filesystems dentries are never used for lookup purposes, and
2607 * thus don't need to be hashed. They also don't need a name until a
2608 * user wants to identify the object in /proc/pid/fd/. The little hack
2609 * below allows us to generate a name for these objects on demand:
2611 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2612 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2614 get_fs_root(current
->fs
, &root
);
2615 write_seqlock(&rename_lock
);
2616 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2618 res
= ERR_PTR(error
);
2619 write_sequnlock(&rename_lock
);
2623 EXPORT_SYMBOL(d_path
);
2626 * d_path_with_unreachable - return the path of a dentry
2627 * @path: path to report
2628 * @buf: buffer to return value in
2629 * @buflen: buffer length
2631 * The difference from d_path() is that this prepends "(unreachable)"
2632 * to paths which are unreachable from the current process' root.
2634 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2636 char *res
= buf
+ buflen
;
2640 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2641 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2643 get_fs_root(current
->fs
, &root
);
2644 write_seqlock(&rename_lock
);
2645 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2647 error
= prepend_unreachable(&res
, &buflen
);
2648 write_sequnlock(&rename_lock
);
2651 res
= ERR_PTR(error
);
2657 * Helper function for dentry_operations.d_dname() members
2659 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2660 const char *fmt
, ...)
2666 va_start(args
, fmt
);
2667 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2670 if (sz
> sizeof(temp
) || sz
> buflen
)
2671 return ERR_PTR(-ENAMETOOLONG
);
2673 buffer
+= buflen
- sz
;
2674 return memcpy(buffer
, temp
, sz
);
2678 * Write full pathname from the root of the filesystem into the buffer.
2680 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2682 char *end
= buf
+ buflen
;
2685 prepend(&end
, &buflen
, "\0", 1);
2692 while (!IS_ROOT(dentry
)) {
2693 struct dentry
*parent
= dentry
->d_parent
;
2697 spin_lock(&dentry
->d_lock
);
2698 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2699 spin_unlock(&dentry
->d_lock
);
2700 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2708 return ERR_PTR(-ENAMETOOLONG
);
2711 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2715 write_seqlock(&rename_lock
);
2716 retval
= __dentry_path(dentry
, buf
, buflen
);
2717 write_sequnlock(&rename_lock
);
2721 EXPORT_SYMBOL(dentry_path_raw
);
2723 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2728 write_seqlock(&rename_lock
);
2729 if (d_unlinked(dentry
)) {
2731 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2735 retval
= __dentry_path(dentry
, buf
, buflen
);
2736 write_sequnlock(&rename_lock
);
2737 if (!IS_ERR(retval
) && p
)
2738 *p
= '/'; /* restore '/' overriden with '\0' */
2741 return ERR_PTR(-ENAMETOOLONG
);
2745 * NOTE! The user-level library version returns a
2746 * character pointer. The kernel system call just
2747 * returns the length of the buffer filled (which
2748 * includes the ending '\0' character), or a negative
2749 * error value. So libc would do something like
2751 * char *getcwd(char * buf, size_t size)
2755 * retval = sys_getcwd(buf, size);
2762 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2765 struct path pwd
, root
;
2766 char *page
= (char *) __get_free_page(GFP_USER
);
2771 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2774 write_seqlock(&rename_lock
);
2775 if (!d_unlinked(pwd
.dentry
)) {
2777 char *cwd
= page
+ PAGE_SIZE
;
2778 int buflen
= PAGE_SIZE
;
2780 prepend(&cwd
, &buflen
, "\0", 1);
2781 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2782 write_sequnlock(&rename_lock
);
2787 /* Unreachable from current root */
2789 error
= prepend_unreachable(&cwd
, &buflen
);
2795 len
= PAGE_SIZE
+ page
- cwd
;
2798 if (copy_to_user(buf
, cwd
, len
))
2802 write_sequnlock(&rename_lock
);
2808 free_page((unsigned long) page
);
2813 * Test whether new_dentry is a subdirectory of old_dentry.
2815 * Trivially implemented using the dcache structure
2819 * is_subdir - is new dentry a subdirectory of old_dentry
2820 * @new_dentry: new dentry
2821 * @old_dentry: old dentry
2823 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2824 * Returns 0 otherwise.
2825 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2828 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2833 if (new_dentry
== old_dentry
)
2837 /* for restarting inner loop in case of seq retry */
2838 seq
= read_seqbegin(&rename_lock
);
2840 * Need rcu_readlock to protect against the d_parent trashing
2844 if (d_ancestor(old_dentry
, new_dentry
))
2849 } while (read_seqretry(&rename_lock
, seq
));
2854 void d_genocide(struct dentry
*root
)
2856 struct dentry
*this_parent
;
2857 struct list_head
*next
;
2861 seq
= read_seqbegin(&rename_lock
);
2864 spin_lock(&this_parent
->d_lock
);
2866 next
= this_parent
->d_subdirs
.next
;
2868 while (next
!= &this_parent
->d_subdirs
) {
2869 struct list_head
*tmp
= next
;
2870 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2873 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2874 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2875 spin_unlock(&dentry
->d_lock
);
2878 if (!list_empty(&dentry
->d_subdirs
)) {
2879 spin_unlock(&this_parent
->d_lock
);
2880 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2881 this_parent
= dentry
;
2882 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2885 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2886 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2889 spin_unlock(&dentry
->d_lock
);
2891 if (this_parent
!= root
) {
2892 struct dentry
*child
= this_parent
;
2893 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2894 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2895 this_parent
->d_count
--;
2897 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2900 next
= child
->d_u
.d_child
.next
;
2903 spin_unlock(&this_parent
->d_lock
);
2904 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2907 write_sequnlock(&rename_lock
);
2912 write_seqlock(&rename_lock
);
2917 * find_inode_number - check for dentry with name
2918 * @dir: directory to check
2919 * @name: Name to find.
2921 * Check whether a dentry already exists for the given name,
2922 * and return the inode number if it has an inode. Otherwise
2925 * This routine is used to post-process directory listings for
2926 * filesystems using synthetic inode numbers, and is necessary
2927 * to keep getcwd() working.
2930 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2932 struct dentry
* dentry
;
2935 dentry
= d_hash_and_lookup(dir
, name
);
2937 if (dentry
->d_inode
)
2938 ino
= dentry
->d_inode
->i_ino
;
2943 EXPORT_SYMBOL(find_inode_number
);
2945 static __initdata
unsigned long dhash_entries
;
2946 static int __init
set_dhash_entries(char *str
)
2950 dhash_entries
= simple_strtoul(str
, &str
, 0);
2953 __setup("dhash_entries=", set_dhash_entries
);
2955 static void __init
dcache_init_early(void)
2959 /* If hashes are distributed across NUMA nodes, defer
2960 * hash allocation until vmalloc space is available.
2966 alloc_large_system_hash("Dentry cache",
2967 sizeof(struct hlist_bl_head
),
2975 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2976 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
2979 static void __init
dcache_init(void)
2984 * A constructor could be added for stable state like the lists,
2985 * but it is probably not worth it because of the cache nature
2988 dentry_cache
= KMEM_CACHE(dentry
,
2989 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2991 /* Hash may have been set up in dcache_init_early */
2996 alloc_large_system_hash("Dentry cache",
2997 sizeof(struct hlist_bl_head
),
3005 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
3006 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3009 /* SLAB cache for __getname() consumers */
3010 struct kmem_cache
*names_cachep __read_mostly
;
3011 EXPORT_SYMBOL(names_cachep
);
3013 EXPORT_SYMBOL(d_genocide
);
3015 void __init
vfs_caches_init_early(void)
3017 dcache_init_early();
3021 void __init
vfs_caches_init(unsigned long mempages
)
3023 unsigned long reserve
;
3025 /* Base hash sizes on available memory, with a reserve equal to
3026 150% of current kernel size */
3028 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3029 mempages
-= reserve
;
3031 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3032 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3036 files_init(mempages
);