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>
44 * dcache->d_inode->i_lock protects:
45 * - i_dentry, d_alias, d_inode of aliases
46 * dcache_hash_bucket lock protects:
47 * - the dcache hash table
48 * s_anon bl list spinlock protects:
49 * - the s_anon list (see __d_drop)
50 * dcache_lru_lock protects:
51 * - the dcache lru lists and counters
58 * - d_parent and d_subdirs
59 * - childrens' d_child and d_parent
63 * dentry->d_inode->i_lock
66 * dcache_hash_bucket lock
69 * If there is an ancestor relationship:
70 * dentry->d_parent->...->d_parent->d_lock
72 * dentry->d_parent->d_lock
75 * If no ancestor relationship:
76 * if (dentry1 < dentry2)
80 int sysctl_vfs_cache_pressure __read_mostly
= 100;
81 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
83 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
84 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
86 EXPORT_SYMBOL(rename_lock
);
88 static struct kmem_cache
*dentry_cache __read_mostly
;
91 * This is the single most critical data structure when it comes
92 * to the dcache: the hashtable for lookups. Somebody should try
93 * to make this good - I've just made it work.
95 * This hash-function tries to avoid losing too many bits of hash
96 * information, yet avoid using a prime hash-size or similar.
98 #define D_HASHBITS d_hash_shift
99 #define D_HASHMASK d_hash_mask
101 static unsigned int d_hash_mask __read_mostly
;
102 static unsigned int d_hash_shift __read_mostly
;
104 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
106 static inline struct hlist_bl_head
*d_hash(struct dentry
*parent
,
109 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
110 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
111 return dentry_hashtable
+ (hash
& D_HASHMASK
);
114 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat
= {
119 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
121 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
122 static int get_nr_dentry(void)
126 for_each_possible_cpu(i
)
127 sum
+= per_cpu(nr_dentry
, i
);
128 return sum
< 0 ? 0 : sum
;
131 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
132 size_t *lenp
, loff_t
*ppos
)
134 dentry_stat
.nr_dentry
= get_nr_dentry();
135 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
139 static void __d_free(struct rcu_head
*head
)
141 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
143 WARN_ON(!list_empty(&dentry
->d_alias
));
144 if (dname_external(dentry
))
145 kfree(dentry
->d_name
.name
);
146 kmem_cache_free(dentry_cache
, dentry
);
152 static void d_free(struct dentry
*dentry
)
154 BUG_ON(dentry
->d_count
);
155 this_cpu_dec(nr_dentry
);
156 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
157 dentry
->d_op
->d_release(dentry
);
159 /* if dentry was never visible to RCU, immediate free is OK */
160 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
161 __d_free(&dentry
->d_u
.d_rcu
);
163 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
167 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
168 * @dentry: the target dentry
169 * After this call, in-progress rcu-walk path lookup will fail. This
170 * should be called after unhashing, and after changing d_inode (if
171 * the dentry has not already been unhashed).
173 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
175 assert_spin_locked(&dentry
->d_lock
);
176 /* Go through a barrier */
177 write_seqcount_barrier(&dentry
->d_seq
);
181 * Release the dentry's inode, using the filesystem
182 * d_iput() operation if defined. Dentry has no refcount
185 static void dentry_iput(struct dentry
* dentry
)
186 __releases(dentry
->d_lock
)
187 __releases(dentry
->d_inode
->i_lock
)
189 struct inode
*inode
= dentry
->d_inode
;
191 dentry
->d_inode
= NULL
;
192 list_del_init(&dentry
->d_alias
);
193 spin_unlock(&dentry
->d_lock
);
194 spin_unlock(&inode
->i_lock
);
196 fsnotify_inoderemove(inode
);
197 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
198 dentry
->d_op
->d_iput(dentry
, inode
);
202 spin_unlock(&dentry
->d_lock
);
207 * Release the dentry's inode, using the filesystem
208 * d_iput() operation if defined. dentry remains in-use.
210 static void dentry_unlink_inode(struct dentry
* dentry
)
211 __releases(dentry
->d_lock
)
212 __releases(dentry
->d_inode
->i_lock
)
214 struct inode
*inode
= dentry
->d_inode
;
215 dentry
->d_inode
= NULL
;
216 list_del_init(&dentry
->d_alias
);
217 dentry_rcuwalk_barrier(dentry
);
218 spin_unlock(&dentry
->d_lock
);
219 spin_unlock(&inode
->i_lock
);
221 fsnotify_inoderemove(inode
);
222 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
223 dentry
->d_op
->d_iput(dentry
, inode
);
229 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
231 static void dentry_lru_add(struct dentry
*dentry
)
233 if (list_empty(&dentry
->d_lru
)) {
234 spin_lock(&dcache_lru_lock
);
235 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
236 dentry
->d_sb
->s_nr_dentry_unused
++;
237 dentry_stat
.nr_unused
++;
238 spin_unlock(&dcache_lru_lock
);
242 static void __dentry_lru_del(struct dentry
*dentry
)
244 list_del_init(&dentry
->d_lru
);
245 dentry
->d_sb
->s_nr_dentry_unused
--;
246 dentry_stat
.nr_unused
--;
250 * Remove a dentry with references from the LRU.
252 static void dentry_lru_del(struct dentry
*dentry
)
254 if (!list_empty(&dentry
->d_lru
)) {
255 spin_lock(&dcache_lru_lock
);
256 __dentry_lru_del(dentry
);
257 spin_unlock(&dcache_lru_lock
);
262 * Remove a dentry that is unreferenced and about to be pruned
263 * (unhashed and destroyed) from the LRU, and inform the file system.
264 * This wrapper should be called _prior_ to unhashing a victim dentry.
266 static void dentry_lru_prune(struct dentry
*dentry
)
268 if (!list_empty(&dentry
->d_lru
)) {
269 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
270 dentry
->d_op
->d_prune(dentry
);
272 spin_lock(&dcache_lru_lock
);
273 __dentry_lru_del(dentry
);
274 spin_unlock(&dcache_lru_lock
);
278 static void dentry_lru_move_tail(struct dentry
*dentry
)
280 spin_lock(&dcache_lru_lock
);
281 if (list_empty(&dentry
->d_lru
)) {
282 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
283 dentry
->d_sb
->s_nr_dentry_unused
++;
284 dentry_stat
.nr_unused
++;
286 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
288 spin_unlock(&dcache_lru_lock
);
292 * d_kill - kill dentry and return parent
293 * @dentry: dentry to kill
294 * @parent: parent dentry
296 * The dentry must already be unhashed and removed from the LRU.
298 * If this is the root of the dentry tree, return NULL.
300 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
303 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
304 __releases(dentry
->d_lock
)
305 __releases(parent
->d_lock
)
306 __releases(dentry
->d_inode
->i_lock
)
308 list_del(&dentry
->d_u
.d_child
);
310 * Inform try_to_ascend() that we are no longer attached to the
313 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
315 spin_unlock(&parent
->d_lock
);
318 * dentry_iput drops the locks, at which point nobody (except
319 * transient RCU lookups) can reach this dentry.
326 * Unhash a dentry without inserting an RCU walk barrier or checking that
327 * dentry->d_lock is locked. The caller must take care of that, if
330 static void __d_shrink(struct dentry
*dentry
)
332 if (!d_unhashed(dentry
)) {
333 struct hlist_bl_head
*b
;
334 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
335 b
= &dentry
->d_sb
->s_anon
;
337 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
340 __hlist_bl_del(&dentry
->d_hash
);
341 dentry
->d_hash
.pprev
= NULL
;
347 * d_drop - drop a dentry
348 * @dentry: dentry to drop
350 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
351 * be found through a VFS lookup any more. Note that this is different from
352 * deleting the dentry - d_delete will try to mark the dentry negative if
353 * possible, giving a successful _negative_ lookup, while d_drop will
354 * just make the cache lookup fail.
356 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
357 * reason (NFS timeouts or autofs deletes).
359 * __d_drop requires dentry->d_lock.
361 void __d_drop(struct dentry
*dentry
)
363 if (!d_unhashed(dentry
)) {
365 dentry_rcuwalk_barrier(dentry
);
368 EXPORT_SYMBOL(__d_drop
);
370 void d_drop(struct dentry
*dentry
)
372 spin_lock(&dentry
->d_lock
);
374 spin_unlock(&dentry
->d_lock
);
376 EXPORT_SYMBOL(d_drop
);
379 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
380 * @dentry: dentry to drop
382 * This is called when we do a lookup on a placeholder dentry that needed to be
383 * looked up. The dentry should have been hashed in order for it to be found by
384 * the lookup code, but now needs to be unhashed while we do the actual lookup
385 * and clear the DCACHE_NEED_LOOKUP flag.
387 void d_clear_need_lookup(struct dentry
*dentry
)
389 spin_lock(&dentry
->d_lock
);
391 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
392 spin_unlock(&dentry
->d_lock
);
394 EXPORT_SYMBOL(d_clear_need_lookup
);
397 * Finish off a dentry we've decided to kill.
398 * dentry->d_lock must be held, returns with it unlocked.
399 * If ref is non-zero, then decrement the refcount too.
400 * Returns dentry requiring refcount drop, or NULL if we're done.
402 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
403 __releases(dentry
->d_lock
)
406 struct dentry
*parent
;
408 inode
= dentry
->d_inode
;
409 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
411 spin_unlock(&dentry
->d_lock
);
413 return dentry
; /* try again with same dentry */
418 parent
= dentry
->d_parent
;
419 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
421 spin_unlock(&inode
->i_lock
);
428 * if dentry was on the d_lru list delete it from there.
429 * inform the fs via d_prune that this dentry is about to be
430 * unhashed and destroyed.
432 dentry_lru_prune(dentry
);
433 /* if it was on the hash then remove it */
435 return d_kill(dentry
, parent
);
441 * This is complicated by the fact that we do not want to put
442 * dentries that are no longer on any hash chain on the unused
443 * list: we'd much rather just get rid of them immediately.
445 * However, that implies that we have to traverse the dentry
446 * tree upwards to the parents which might _also_ now be
447 * scheduled for deletion (it may have been only waiting for
448 * its last child to go away).
450 * This tail recursion is done by hand as we don't want to depend
451 * on the compiler to always get this right (gcc generally doesn't).
452 * Real recursion would eat up our stack space.
456 * dput - release a dentry
457 * @dentry: dentry to release
459 * Release a dentry. This will drop the usage count and if appropriate
460 * call the dentry unlink method as well as removing it from the queues and
461 * releasing its resources. If the parent dentries were scheduled for release
462 * they too may now get deleted.
464 void dput(struct dentry
*dentry
)
470 if (dentry
->d_count
== 1)
472 spin_lock(&dentry
->d_lock
);
473 BUG_ON(!dentry
->d_count
);
474 if (dentry
->d_count
> 1) {
476 spin_unlock(&dentry
->d_lock
);
480 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
481 if (dentry
->d_op
->d_delete(dentry
))
485 /* Unreachable? Get rid of it */
486 if (d_unhashed(dentry
))
490 * If this dentry needs lookup, don't set the referenced flag so that it
491 * is more likely to be cleaned up by the dcache shrinker in case of
494 if (!d_need_lookup(dentry
))
495 dentry
->d_flags
|= DCACHE_REFERENCED
;
496 dentry_lru_add(dentry
);
499 spin_unlock(&dentry
->d_lock
);
503 dentry
= dentry_kill(dentry
, 1);
510 * d_invalidate - invalidate a dentry
511 * @dentry: dentry to invalidate
513 * Try to invalidate the dentry if it turns out to be
514 * possible. If there are other dentries that can be
515 * reached through this one we can't delete it and we
516 * return -EBUSY. On success we return 0.
521 int d_invalidate(struct dentry
* dentry
)
524 * If it's already been dropped, return OK.
526 spin_lock(&dentry
->d_lock
);
527 if (d_unhashed(dentry
)) {
528 spin_unlock(&dentry
->d_lock
);
532 * Check whether to do a partial shrink_dcache
533 * to get rid of unused child entries.
535 if (!list_empty(&dentry
->d_subdirs
)) {
536 spin_unlock(&dentry
->d_lock
);
537 shrink_dcache_parent(dentry
);
538 spin_lock(&dentry
->d_lock
);
542 * Somebody else still using it?
544 * If it's a directory, we can't drop it
545 * for fear of somebody re-populating it
546 * with children (even though dropping it
547 * would make it unreachable from the root,
548 * we might still populate it if it was a
549 * working directory or similar).
550 * We also need to leave mountpoints alone,
553 if (dentry
->d_count
> 1 && dentry
->d_inode
) {
554 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
555 spin_unlock(&dentry
->d_lock
);
561 spin_unlock(&dentry
->d_lock
);
564 EXPORT_SYMBOL(d_invalidate
);
566 /* This must be called with d_lock held */
567 static inline void __dget_dlock(struct dentry
*dentry
)
572 static inline void __dget(struct dentry
*dentry
)
574 spin_lock(&dentry
->d_lock
);
575 __dget_dlock(dentry
);
576 spin_unlock(&dentry
->d_lock
);
579 struct dentry
*dget_parent(struct dentry
*dentry
)
585 * Don't need rcu_dereference because we re-check it was correct under
589 ret
= dentry
->d_parent
;
590 spin_lock(&ret
->d_lock
);
591 if (unlikely(ret
!= dentry
->d_parent
)) {
592 spin_unlock(&ret
->d_lock
);
597 BUG_ON(!ret
->d_count
);
599 spin_unlock(&ret
->d_lock
);
602 EXPORT_SYMBOL(dget_parent
);
605 * d_find_alias - grab a hashed alias of inode
606 * @inode: inode in question
607 * @want_discon: flag, used by d_splice_alias, to request
608 * that only a DISCONNECTED alias be returned.
610 * If inode has a hashed alias, or is a directory and has any alias,
611 * acquire the reference to alias and return it. Otherwise return NULL.
612 * Notice that if inode is a directory there can be only one alias and
613 * it can be unhashed only if it has no children, or if it is the root
616 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
617 * any other hashed alias over that one unless @want_discon is set,
618 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
620 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
622 struct dentry
*alias
, *discon_alias
;
626 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
627 spin_lock(&alias
->d_lock
);
628 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
629 if (IS_ROOT(alias
) &&
630 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
631 discon_alias
= alias
;
632 } else if (!want_discon
) {
634 spin_unlock(&alias
->d_lock
);
638 spin_unlock(&alias
->d_lock
);
641 alias
= discon_alias
;
642 spin_lock(&alias
->d_lock
);
643 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
644 if (IS_ROOT(alias
) &&
645 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
647 spin_unlock(&alias
->d_lock
);
651 spin_unlock(&alias
->d_lock
);
657 struct dentry
*d_find_alias(struct inode
*inode
)
659 struct dentry
*de
= NULL
;
661 if (!list_empty(&inode
->i_dentry
)) {
662 spin_lock(&inode
->i_lock
);
663 de
= __d_find_alias(inode
, 0);
664 spin_unlock(&inode
->i_lock
);
668 EXPORT_SYMBOL(d_find_alias
);
671 * Try to kill dentries associated with this inode.
672 * WARNING: you must own a reference to inode.
674 void d_prune_aliases(struct inode
*inode
)
676 struct dentry
*dentry
;
678 spin_lock(&inode
->i_lock
);
679 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
680 spin_lock(&dentry
->d_lock
);
681 if (!dentry
->d_count
) {
682 __dget_dlock(dentry
);
684 spin_unlock(&dentry
->d_lock
);
685 spin_unlock(&inode
->i_lock
);
689 spin_unlock(&dentry
->d_lock
);
691 spin_unlock(&inode
->i_lock
);
693 EXPORT_SYMBOL(d_prune_aliases
);
696 * Try to throw away a dentry - free the inode, dput the parent.
697 * Requires dentry->d_lock is held, and dentry->d_count == 0.
698 * Releases dentry->d_lock.
700 * This may fail if locks cannot be acquired no problem, just try again.
702 static void try_prune_one_dentry(struct dentry
*dentry
)
703 __releases(dentry
->d_lock
)
705 struct dentry
*parent
;
707 parent
= dentry_kill(dentry
, 0);
709 * If dentry_kill returns NULL, we have nothing more to do.
710 * if it returns the same dentry, trylocks failed. In either
711 * case, just loop again.
713 * Otherwise, we need to prune ancestors too. This is necessary
714 * to prevent quadratic behavior of shrink_dcache_parent(), but
715 * is also expected to be beneficial in reducing dentry cache
720 if (parent
== dentry
)
723 /* Prune ancestors. */
726 spin_lock(&dentry
->d_lock
);
727 if (dentry
->d_count
> 1) {
729 spin_unlock(&dentry
->d_lock
);
732 dentry
= dentry_kill(dentry
, 1);
736 static void shrink_dentry_list(struct list_head
*list
)
738 struct dentry
*dentry
;
742 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
743 if (&dentry
->d_lru
== list
)
745 spin_lock(&dentry
->d_lock
);
746 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
747 spin_unlock(&dentry
->d_lock
);
752 * We found an inuse dentry which was not removed from
753 * the LRU because of laziness during lookup. Do not free
754 * it - just keep it off the LRU list.
756 if (dentry
->d_count
) {
757 dentry_lru_del(dentry
);
758 spin_unlock(&dentry
->d_lock
);
764 try_prune_one_dentry(dentry
);
772 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
773 * @sb: superblock to shrink dentry LRU.
774 * @count: number of entries to prune
775 * @flags: flags to control the dentry processing
777 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
779 static void __shrink_dcache_sb(struct super_block
*sb
, int count
, int flags
)
781 struct dentry
*dentry
;
782 LIST_HEAD(referenced
);
786 spin_lock(&dcache_lru_lock
);
787 while (!list_empty(&sb
->s_dentry_lru
)) {
788 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
789 struct dentry
, d_lru
);
790 BUG_ON(dentry
->d_sb
!= sb
);
792 if (!spin_trylock(&dentry
->d_lock
)) {
793 spin_unlock(&dcache_lru_lock
);
799 * If we are honouring the DCACHE_REFERENCED flag and the
800 * dentry has this flag set, don't free it. Clear the flag
801 * and put it back on the LRU.
803 if (flags
& DCACHE_REFERENCED
&&
804 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 spin_unlock(&dentry
->d_lock
);
814 cond_resched_lock(&dcache_lru_lock
);
816 if (!list_empty(&referenced
))
817 list_splice(&referenced
, &sb
->s_dentry_lru
);
818 spin_unlock(&dcache_lru_lock
);
820 shrink_dentry_list(&tmp
);
824 * prune_dcache_sb - shrink the dcache
826 * @nr_to_scan: number of entries to try to free
828 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
829 * done when we need more memory an called from the superblock shrinker
832 * This function may fail to free any resources if all the dentries are in
835 void prune_dcache_sb(struct super_block
*sb
, int nr_to_scan
)
837 __shrink_dcache_sb(sb
, nr_to_scan
, DCACHE_REFERENCED
);
841 * shrink_dcache_sb - shrink dcache for a superblock
844 * Shrink the dcache for the specified super block. This is used to free
845 * the dcache before unmounting a file system.
847 void shrink_dcache_sb(struct super_block
*sb
)
851 spin_lock(&dcache_lru_lock
);
852 while (!list_empty(&sb
->s_dentry_lru
)) {
853 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
854 spin_unlock(&dcache_lru_lock
);
855 shrink_dentry_list(&tmp
);
856 spin_lock(&dcache_lru_lock
);
858 spin_unlock(&dcache_lru_lock
);
860 EXPORT_SYMBOL(shrink_dcache_sb
);
863 * destroy a single subtree of dentries for unmount
864 * - see the comments on shrink_dcache_for_umount() for a description of the
867 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
869 struct dentry
*parent
;
871 BUG_ON(!IS_ROOT(dentry
));
874 /* descend to the first leaf in the current subtree */
875 while (!list_empty(&dentry
->d_subdirs
))
876 dentry
= list_entry(dentry
->d_subdirs
.next
,
877 struct dentry
, d_u
.d_child
);
879 /* consume the dentries from this leaf up through its parents
880 * until we find one with children or run out altogether */
885 * remove the dentry from the lru, and inform
886 * the fs that this dentry is about to be
887 * unhashed and destroyed.
889 dentry_lru_prune(dentry
);
892 if (dentry
->d_count
!= 0) {
894 "BUG: Dentry %p{i=%lx,n=%s}"
896 " [unmount of %s %s]\n",
899 dentry
->d_inode
->i_ino
: 0UL,
902 dentry
->d_sb
->s_type
->name
,
907 if (IS_ROOT(dentry
)) {
909 list_del(&dentry
->d_u
.d_child
);
911 parent
= dentry
->d_parent
;
913 list_del(&dentry
->d_u
.d_child
);
916 inode
= dentry
->d_inode
;
918 dentry
->d_inode
= NULL
;
919 list_del_init(&dentry
->d_alias
);
920 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
921 dentry
->d_op
->d_iput(dentry
, inode
);
928 /* finished when we fall off the top of the tree,
929 * otherwise we ascend to the parent and move to the
930 * next sibling if there is one */
934 } while (list_empty(&dentry
->d_subdirs
));
936 dentry
= list_entry(dentry
->d_subdirs
.next
,
937 struct dentry
, d_u
.d_child
);
942 * destroy the dentries attached to a superblock on unmounting
943 * - we don't need to use dentry->d_lock because:
944 * - the superblock is detached from all mountings and open files, so the
945 * dentry trees will not be rearranged by the VFS
946 * - s_umount is write-locked, so the memory pressure shrinker will ignore
947 * any dentries belonging to this superblock that it comes across
948 * - the filesystem itself is no longer permitted to rearrange the dentries
951 void shrink_dcache_for_umount(struct super_block
*sb
)
953 struct dentry
*dentry
;
955 if (down_read_trylock(&sb
->s_umount
))
961 shrink_dcache_for_umount_subtree(dentry
);
963 while (!hlist_bl_empty(&sb
->s_anon
)) {
964 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
965 shrink_dcache_for_umount_subtree(dentry
);
970 * This tries to ascend one level of parenthood, but
971 * we can race with renaming, so we need to re-check
972 * the parenthood after dropping the lock and check
973 * that the sequence number still matches.
975 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
977 struct dentry
*new = old
->d_parent
;
980 spin_unlock(&old
->d_lock
);
981 spin_lock(&new->d_lock
);
984 * might go back up the wrong parent if we have had a rename
987 if (new != old
->d_parent
||
988 (old
->d_flags
& DCACHE_DISCONNECTED
) ||
989 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
990 spin_unlock(&new->d_lock
);
999 * Search for at least 1 mount point in the dentry's subdirs.
1000 * We descend to the next level whenever the d_subdirs
1001 * list is non-empty and continue searching.
1005 * have_submounts - check for mounts over a dentry
1006 * @parent: dentry to check.
1008 * Return true if the parent or its subdirectories contain
1011 int have_submounts(struct dentry
*parent
)
1013 struct dentry
*this_parent
;
1014 struct list_head
*next
;
1018 seq
= read_seqbegin(&rename_lock
);
1020 this_parent
= parent
;
1022 if (d_mountpoint(parent
))
1024 spin_lock(&this_parent
->d_lock
);
1026 next
= this_parent
->d_subdirs
.next
;
1028 while (next
!= &this_parent
->d_subdirs
) {
1029 struct list_head
*tmp
= next
;
1030 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1033 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1034 /* Have we found a mount point ? */
1035 if (d_mountpoint(dentry
)) {
1036 spin_unlock(&dentry
->d_lock
);
1037 spin_unlock(&this_parent
->d_lock
);
1040 if (!list_empty(&dentry
->d_subdirs
)) {
1041 spin_unlock(&this_parent
->d_lock
);
1042 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1043 this_parent
= dentry
;
1044 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1047 spin_unlock(&dentry
->d_lock
);
1050 * All done at this level ... ascend and resume the search.
1052 if (this_parent
!= parent
) {
1053 struct dentry
*child
= this_parent
;
1054 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1057 next
= child
->d_u
.d_child
.next
;
1060 spin_unlock(&this_parent
->d_lock
);
1061 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1064 write_sequnlock(&rename_lock
);
1065 return 0; /* No mount points found in tree */
1067 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1070 write_sequnlock(&rename_lock
);
1075 write_seqlock(&rename_lock
);
1078 EXPORT_SYMBOL(have_submounts
);
1081 * Search the dentry child list for the specified parent,
1082 * and move any unused dentries to the end of the unused
1083 * list for prune_dcache(). We descend to the next level
1084 * whenever the d_subdirs list is non-empty and continue
1087 * It returns zero iff there are no unused children,
1088 * otherwise it returns the number of children moved to
1089 * the end of the unused list. This may not be the total
1090 * number of unused children, because select_parent can
1091 * drop the lock and return early due to latency
1094 static int select_parent(struct dentry
* parent
)
1096 struct dentry
*this_parent
;
1097 struct list_head
*next
;
1102 seq
= read_seqbegin(&rename_lock
);
1104 this_parent
= parent
;
1105 spin_lock(&this_parent
->d_lock
);
1107 next
= this_parent
->d_subdirs
.next
;
1109 while (next
!= &this_parent
->d_subdirs
) {
1110 struct list_head
*tmp
= next
;
1111 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1114 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1117 * move only zero ref count dentries to the end
1118 * of the unused list for prune_dcache
1120 if (!dentry
->d_count
) {
1121 dentry_lru_move_tail(dentry
);
1124 dentry_lru_del(dentry
);
1128 * We can return to the caller if we have found some (this
1129 * ensures forward progress). We'll be coming back to find
1132 if (found
&& need_resched()) {
1133 spin_unlock(&dentry
->d_lock
);
1138 * Descend a level if the d_subdirs list is non-empty.
1140 if (!list_empty(&dentry
->d_subdirs
)) {
1141 spin_unlock(&this_parent
->d_lock
);
1142 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1143 this_parent
= dentry
;
1144 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1148 spin_unlock(&dentry
->d_lock
);
1151 * All done at this level ... ascend and resume the search.
1153 if (this_parent
!= parent
) {
1154 struct dentry
*child
= this_parent
;
1155 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1158 next
= child
->d_u
.d_child
.next
;
1162 spin_unlock(&this_parent
->d_lock
);
1163 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1166 write_sequnlock(&rename_lock
);
1173 write_seqlock(&rename_lock
);
1178 * shrink_dcache_parent - prune dcache
1179 * @parent: parent of entries to prune
1181 * Prune the dcache to remove unused children of the parent dentry.
1184 void shrink_dcache_parent(struct dentry
* parent
)
1186 struct super_block
*sb
= parent
->d_sb
;
1189 while ((found
= select_parent(parent
)) != 0)
1190 __shrink_dcache_sb(sb
, found
, 0);
1192 EXPORT_SYMBOL(shrink_dcache_parent
);
1195 * __d_alloc - allocate a dcache entry
1196 * @sb: filesystem it will belong to
1197 * @name: qstr of the name
1199 * Allocates a dentry. It returns %NULL if there is insufficient memory
1200 * available. On a success the dentry is returned. The name passed in is
1201 * copied and the copy passed in may be reused after this call.
1204 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1206 struct dentry
*dentry
;
1209 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1213 if (name
->len
> DNAME_INLINE_LEN
-1) {
1214 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1216 kmem_cache_free(dentry_cache
, dentry
);
1220 dname
= dentry
->d_iname
;
1222 dentry
->d_name
.name
= dname
;
1224 dentry
->d_name
.len
= name
->len
;
1225 dentry
->d_name
.hash
= name
->hash
;
1226 memcpy(dname
, name
->name
, name
->len
);
1227 dname
[name
->len
] = 0;
1229 dentry
->d_count
= 1;
1230 dentry
->d_flags
= 0;
1231 spin_lock_init(&dentry
->d_lock
);
1232 seqcount_init(&dentry
->d_seq
);
1233 dentry
->d_inode
= NULL
;
1234 dentry
->d_parent
= dentry
;
1236 dentry
->d_op
= NULL
;
1237 dentry
->d_fsdata
= NULL
;
1238 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1239 INIT_LIST_HEAD(&dentry
->d_lru
);
1240 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1241 INIT_LIST_HEAD(&dentry
->d_alias
);
1242 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1243 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1245 this_cpu_inc(nr_dentry
);
1251 * d_alloc - allocate a dcache entry
1252 * @parent: parent of entry to allocate
1253 * @name: qstr of the name
1255 * Allocates a dentry. It returns %NULL if there is insufficient memory
1256 * available. On a success the dentry is returned. The name passed in is
1257 * copied and the copy passed in may be reused after this call.
1259 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1261 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1265 spin_lock(&parent
->d_lock
);
1267 * don't need child lock because it is not subject
1268 * to concurrency here
1270 __dget_dlock(parent
);
1271 dentry
->d_parent
= parent
;
1272 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1273 spin_unlock(&parent
->d_lock
);
1277 EXPORT_SYMBOL(d_alloc
);
1279 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1281 struct dentry
*dentry
= __d_alloc(sb
, name
);
1283 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1286 EXPORT_SYMBOL(d_alloc_pseudo
);
1288 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1293 q
.len
= strlen(name
);
1294 q
.hash
= full_name_hash(q
.name
, q
.len
);
1295 return d_alloc(parent
, &q
);
1297 EXPORT_SYMBOL(d_alloc_name
);
1299 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1301 WARN_ON_ONCE(dentry
->d_op
);
1302 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1304 DCACHE_OP_REVALIDATE
|
1305 DCACHE_OP_DELETE
));
1310 dentry
->d_flags
|= DCACHE_OP_HASH
;
1312 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1313 if (op
->d_revalidate
)
1314 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1316 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1318 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1321 EXPORT_SYMBOL(d_set_d_op
);
1323 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1325 spin_lock(&dentry
->d_lock
);
1327 if (unlikely(IS_AUTOMOUNT(inode
)))
1328 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1329 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1331 dentry
->d_inode
= inode
;
1332 dentry_rcuwalk_barrier(dentry
);
1333 spin_unlock(&dentry
->d_lock
);
1334 fsnotify_d_instantiate(dentry
, inode
);
1338 * d_instantiate - fill in inode information for a dentry
1339 * @entry: dentry to complete
1340 * @inode: inode to attach to this dentry
1342 * Fill in inode information in the entry.
1344 * This turns negative dentries into productive full members
1347 * NOTE! This assumes that the inode count has been incremented
1348 * (or otherwise set) by the caller to indicate that it is now
1349 * in use by the dcache.
1352 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1354 BUG_ON(!list_empty(&entry
->d_alias
));
1356 spin_lock(&inode
->i_lock
);
1357 __d_instantiate(entry
, inode
);
1359 spin_unlock(&inode
->i_lock
);
1360 security_d_instantiate(entry
, inode
);
1362 EXPORT_SYMBOL(d_instantiate
);
1365 * d_instantiate_unique - instantiate a non-aliased dentry
1366 * @entry: dentry to instantiate
1367 * @inode: inode to attach to this dentry
1369 * Fill in inode information in the entry. On success, it returns NULL.
1370 * If an unhashed alias of "entry" already exists, then we return the
1371 * aliased dentry instead and drop one reference to inode.
1373 * Note that in order to avoid conflicts with rename() etc, the caller
1374 * had better be holding the parent directory semaphore.
1376 * This also assumes that the inode count has been incremented
1377 * (or otherwise set) by the caller to indicate that it is now
1378 * in use by the dcache.
1380 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1381 struct inode
*inode
)
1383 struct dentry
*alias
;
1384 int len
= entry
->d_name
.len
;
1385 const char *name
= entry
->d_name
.name
;
1386 unsigned int hash
= entry
->d_name
.hash
;
1389 __d_instantiate(entry
, NULL
);
1393 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1394 struct qstr
*qstr
= &alias
->d_name
;
1397 * Don't need alias->d_lock here, because aliases with
1398 * d_parent == entry->d_parent are not subject to name or
1399 * parent changes, because the parent inode i_mutex is held.
1401 if (qstr
->hash
!= hash
)
1403 if (alias
->d_parent
!= entry
->d_parent
)
1405 if (dentry_cmp(qstr
->name
, qstr
->len
, name
, len
))
1411 __d_instantiate(entry
, inode
);
1415 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1417 struct dentry
*result
;
1419 BUG_ON(!list_empty(&entry
->d_alias
));
1422 spin_lock(&inode
->i_lock
);
1423 result
= __d_instantiate_unique(entry
, inode
);
1425 spin_unlock(&inode
->i_lock
);
1428 security_d_instantiate(entry
, inode
);
1432 BUG_ON(!d_unhashed(result
));
1437 EXPORT_SYMBOL(d_instantiate_unique
);
1440 * d_alloc_root - allocate root dentry
1441 * @root_inode: inode to allocate the root for
1443 * Allocate a root ("/") dentry for the inode given. The inode is
1444 * instantiated and returned. %NULL is returned if there is insufficient
1445 * memory or the inode passed is %NULL.
1448 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1450 struct dentry
*res
= NULL
;
1453 static const struct qstr name
= { .name
= "/", .len
= 1 };
1455 res
= __d_alloc(root_inode
->i_sb
, &name
);
1457 d_instantiate(res
, root_inode
);
1461 EXPORT_SYMBOL(d_alloc_root
);
1463 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1465 struct dentry
*alias
;
1467 if (list_empty(&inode
->i_dentry
))
1469 alias
= list_first_entry(&inode
->i_dentry
, struct dentry
, d_alias
);
1475 * d_find_any_alias - find any alias for a given inode
1476 * @inode: inode to find an alias for
1478 * If any aliases exist for the given inode, take and return a
1479 * reference for one of them. If no aliases exist, return %NULL.
1481 struct dentry
*d_find_any_alias(struct inode
*inode
)
1485 spin_lock(&inode
->i_lock
);
1486 de
= __d_find_any_alias(inode
);
1487 spin_unlock(&inode
->i_lock
);
1490 EXPORT_SYMBOL(d_find_any_alias
);
1493 * d_obtain_alias - find or allocate a dentry for a given inode
1494 * @inode: inode to allocate the dentry for
1496 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1497 * similar open by handle operations. The returned dentry may be anonymous,
1498 * or may have a full name (if the inode was already in the cache).
1500 * When called on a directory inode, we must ensure that the inode only ever
1501 * has one dentry. If a dentry is found, that is returned instead of
1502 * allocating a new one.
1504 * On successful return, the reference to the inode has been transferred
1505 * to the dentry. In case of an error the reference on the inode is released.
1506 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1507 * be passed in and will be the error will be propagate to the return value,
1508 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1510 struct dentry
*d_obtain_alias(struct inode
*inode
)
1512 static const struct qstr anonstring
= { .name
= "" };
1517 return ERR_PTR(-ESTALE
);
1519 return ERR_CAST(inode
);
1521 res
= d_find_any_alias(inode
);
1525 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1527 res
= ERR_PTR(-ENOMEM
);
1531 spin_lock(&inode
->i_lock
);
1532 res
= __d_find_any_alias(inode
);
1534 spin_unlock(&inode
->i_lock
);
1539 /* attach a disconnected dentry */
1540 spin_lock(&tmp
->d_lock
);
1541 tmp
->d_inode
= inode
;
1542 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1543 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1544 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1545 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1546 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1547 spin_unlock(&tmp
->d_lock
);
1548 spin_unlock(&inode
->i_lock
);
1549 security_d_instantiate(tmp
, inode
);
1554 if (res
&& !IS_ERR(res
))
1555 security_d_instantiate(res
, inode
);
1559 EXPORT_SYMBOL(d_obtain_alias
);
1562 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1563 * @inode: the inode which may have a disconnected dentry
1564 * @dentry: a negative dentry which we want to point to the inode.
1566 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1567 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1568 * and return it, else simply d_add the inode to the dentry and return NULL.
1570 * This is needed in the lookup routine of any filesystem that is exportable
1571 * (via knfsd) so that we can build dcache paths to directories effectively.
1573 * If a dentry was found and moved, then it is returned. Otherwise NULL
1574 * is returned. This matches the expected return value of ->lookup.
1577 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1579 struct dentry
*new = NULL
;
1582 return ERR_CAST(inode
);
1584 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1585 spin_lock(&inode
->i_lock
);
1586 new = __d_find_alias(inode
, 1);
1588 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1589 spin_unlock(&inode
->i_lock
);
1590 security_d_instantiate(new, inode
);
1591 d_move(new, dentry
);
1594 /* already taking inode->i_lock, so d_add() by hand */
1595 __d_instantiate(dentry
, inode
);
1596 spin_unlock(&inode
->i_lock
);
1597 security_d_instantiate(dentry
, inode
);
1601 d_add(dentry
, inode
);
1604 EXPORT_SYMBOL(d_splice_alias
);
1607 * d_add_ci - lookup or allocate new dentry with case-exact name
1608 * @inode: the inode case-insensitive lookup has found
1609 * @dentry: the negative dentry that was passed to the parent's lookup func
1610 * @name: the case-exact name to be associated with the returned dentry
1612 * This is to avoid filling the dcache with case-insensitive names to the
1613 * same inode, only the actual correct case is stored in the dcache for
1614 * case-insensitive filesystems.
1616 * For a case-insensitive lookup match and if the the case-exact dentry
1617 * already exists in in the dcache, use it and return it.
1619 * If no entry exists with the exact case name, allocate new dentry with
1620 * the exact case, and return the spliced entry.
1622 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1626 struct dentry
*found
;
1630 * First check if a dentry matching the name already exists,
1631 * if not go ahead and create it now.
1633 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1635 new = d_alloc(dentry
->d_parent
, name
);
1641 found
= d_splice_alias(inode
, new);
1650 * If a matching dentry exists, and it's not negative use it.
1652 * Decrement the reference count to balance the iget() done
1655 if (found
->d_inode
) {
1656 if (unlikely(found
->d_inode
!= inode
)) {
1657 /* This can't happen because bad inodes are unhashed. */
1658 BUG_ON(!is_bad_inode(inode
));
1659 BUG_ON(!is_bad_inode(found
->d_inode
));
1666 * We are going to instantiate this dentry, unhash it and clear the
1667 * lookup flag so we can do that.
1669 if (unlikely(d_need_lookup(found
)))
1670 d_clear_need_lookup(found
);
1673 * Negative dentry: instantiate it unless the inode is a directory and
1674 * already has a dentry.
1676 new = d_splice_alias(inode
, found
);
1685 return ERR_PTR(error
);
1687 EXPORT_SYMBOL(d_add_ci
);
1690 * __d_lookup_rcu - search for a dentry (racy, store-free)
1691 * @parent: parent dentry
1692 * @name: qstr of name we wish to find
1693 * @seq: returns d_seq value at the point where the dentry was found
1694 * @inode: returns dentry->d_inode when the inode was found valid.
1695 * Returns: dentry, or NULL
1697 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1698 * resolution (store-free path walking) design described in
1699 * Documentation/filesystems/path-lookup.txt.
1701 * This is not to be used outside core vfs.
1703 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1704 * held, and rcu_read_lock held. The returned dentry must not be stored into
1705 * without taking d_lock and checking d_seq sequence count against @seq
1708 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1711 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1712 * the returned dentry, so long as its parent's seqlock is checked after the
1713 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1714 * is formed, giving integrity down the path walk.
1716 struct dentry
*__d_lookup_rcu(struct dentry
*parent
, struct qstr
*name
,
1717 unsigned *seq
, struct inode
**inode
)
1719 unsigned int len
= name
->len
;
1720 unsigned int hash
= name
->hash
;
1721 const unsigned char *str
= name
->name
;
1722 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1723 struct hlist_bl_node
*node
;
1724 struct dentry
*dentry
;
1727 * Note: There is significant duplication with __d_lookup_rcu which is
1728 * required to prevent single threaded performance regressions
1729 * especially on architectures where smp_rmb (in seqcounts) are costly.
1730 * Keep the two functions in sync.
1734 * The hash list is protected using RCU.
1736 * Carefully use d_seq when comparing a candidate dentry, to avoid
1737 * races with d_move().
1739 * It is possible that concurrent renames can mess up our list
1740 * walk here and result in missing our dentry, resulting in the
1741 * false-negative result. d_lookup() protects against concurrent
1742 * renames using rename_lock seqlock.
1744 * See Documentation/filesystems/path-lookup.txt for more details.
1746 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1751 if (dentry
->d_name
.hash
!= hash
)
1755 *seq
= read_seqcount_begin(&dentry
->d_seq
);
1756 if (dentry
->d_parent
!= parent
)
1758 if (d_unhashed(dentry
))
1760 tlen
= dentry
->d_name
.len
;
1761 tname
= dentry
->d_name
.name
;
1762 i
= dentry
->d_inode
;
1765 * This seqcount check is required to ensure name and
1766 * len are loaded atomically, so as not to walk off the
1767 * edge of memory when walking. If we could load this
1768 * atomically some other way, we could drop this check.
1770 if (read_seqcount_retry(&dentry
->d_seq
, *seq
))
1772 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1773 if (parent
->d_op
->d_compare(parent
, *inode
,
1778 if (dentry_cmp(tname
, tlen
, str
, len
))
1782 * No extra seqcount check is required after the name
1783 * compare. The caller must perform a seqcount check in
1784 * order to do anything useful with the returned dentry
1794 * d_lookup - search for a dentry
1795 * @parent: parent dentry
1796 * @name: qstr of name we wish to find
1797 * Returns: dentry, or NULL
1799 * d_lookup searches the children of the parent dentry for the name in
1800 * question. If the dentry is found its reference count is incremented and the
1801 * dentry is returned. The caller must use dput to free the entry when it has
1802 * finished using it. %NULL is returned if the dentry does not exist.
1804 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1806 struct dentry
*dentry
;
1810 seq
= read_seqbegin(&rename_lock
);
1811 dentry
= __d_lookup(parent
, name
);
1814 } while (read_seqretry(&rename_lock
, seq
));
1817 EXPORT_SYMBOL(d_lookup
);
1820 * __d_lookup - search for a dentry (racy)
1821 * @parent: parent dentry
1822 * @name: qstr of name we wish to find
1823 * Returns: dentry, or NULL
1825 * __d_lookup is like d_lookup, however it may (rarely) return a
1826 * false-negative result due to unrelated rename activity.
1828 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1829 * however it must be used carefully, eg. with a following d_lookup in
1830 * the case of failure.
1832 * __d_lookup callers must be commented.
1834 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1836 unsigned int len
= name
->len
;
1837 unsigned int hash
= name
->hash
;
1838 const unsigned char *str
= name
->name
;
1839 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1840 struct hlist_bl_node
*node
;
1841 struct dentry
*found
= NULL
;
1842 struct dentry
*dentry
;
1845 * Note: There is significant duplication with __d_lookup_rcu which is
1846 * required to prevent single threaded performance regressions
1847 * especially on architectures where smp_rmb (in seqcounts) are costly.
1848 * Keep the two functions in sync.
1852 * The hash list is protected using RCU.
1854 * Take d_lock when comparing a candidate dentry, to avoid races
1857 * It is possible that concurrent renames can mess up our list
1858 * walk here and result in missing our dentry, resulting in the
1859 * false-negative result. d_lookup() protects against concurrent
1860 * renames using rename_lock seqlock.
1862 * See Documentation/filesystems/path-lookup.txt for more details.
1866 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1870 if (dentry
->d_name
.hash
!= hash
)
1873 spin_lock(&dentry
->d_lock
);
1874 if (dentry
->d_parent
!= parent
)
1876 if (d_unhashed(dentry
))
1880 * It is safe to compare names since d_move() cannot
1881 * change the qstr (protected by d_lock).
1883 tlen
= dentry
->d_name
.len
;
1884 tname
= dentry
->d_name
.name
;
1885 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1886 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1887 dentry
, dentry
->d_inode
,
1891 if (dentry_cmp(tname
, tlen
, str
, len
))
1897 spin_unlock(&dentry
->d_lock
);
1900 spin_unlock(&dentry
->d_lock
);
1908 * d_hash_and_lookup - hash the qstr then search for a dentry
1909 * @dir: Directory to search in
1910 * @name: qstr of name we wish to find
1912 * On hash failure or on lookup failure NULL is returned.
1914 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1916 struct dentry
*dentry
= NULL
;
1919 * Check for a fs-specific hash function. Note that we must
1920 * calculate the standard hash first, as the d_op->d_hash()
1921 * routine may choose to leave the hash value unchanged.
1923 name
->hash
= full_name_hash(name
->name
, name
->len
);
1924 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1925 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1928 dentry
= d_lookup(dir
, name
);
1934 * d_validate - verify dentry provided from insecure source (deprecated)
1935 * @dentry: The dentry alleged to be valid child of @dparent
1936 * @dparent: The parent dentry (known to be valid)
1938 * An insecure source has sent us a dentry, here we verify it and dget() it.
1939 * This is used by ncpfs in its readdir implementation.
1940 * Zero is returned in the dentry is invalid.
1942 * This function is slow for big directories, and deprecated, do not use it.
1944 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1946 struct dentry
*child
;
1948 spin_lock(&dparent
->d_lock
);
1949 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1950 if (dentry
== child
) {
1951 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1952 __dget_dlock(dentry
);
1953 spin_unlock(&dentry
->d_lock
);
1954 spin_unlock(&dparent
->d_lock
);
1958 spin_unlock(&dparent
->d_lock
);
1962 EXPORT_SYMBOL(d_validate
);
1965 * When a file is deleted, we have two options:
1966 * - turn this dentry into a negative dentry
1967 * - unhash this dentry and free it.
1969 * Usually, we want to just turn this into
1970 * a negative dentry, but if anybody else is
1971 * currently using the dentry or the inode
1972 * we can't do that and we fall back on removing
1973 * it from the hash queues and waiting for
1974 * it to be deleted later when it has no users
1978 * d_delete - delete a dentry
1979 * @dentry: The dentry to delete
1981 * Turn the dentry into a negative dentry if possible, otherwise
1982 * remove it from the hash queues so it can be deleted later
1985 void d_delete(struct dentry
* dentry
)
1987 struct inode
*inode
;
1990 * Are we the only user?
1993 spin_lock(&dentry
->d_lock
);
1994 inode
= dentry
->d_inode
;
1995 isdir
= S_ISDIR(inode
->i_mode
);
1996 if (dentry
->d_count
== 1) {
1997 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
1998 spin_unlock(&dentry
->d_lock
);
2002 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2003 dentry_unlink_inode(dentry
);
2004 fsnotify_nameremove(dentry
, isdir
);
2008 if (!d_unhashed(dentry
))
2011 spin_unlock(&dentry
->d_lock
);
2013 fsnotify_nameremove(dentry
, isdir
);
2015 EXPORT_SYMBOL(d_delete
);
2017 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2019 BUG_ON(!d_unhashed(entry
));
2021 entry
->d_flags
|= DCACHE_RCUACCESS
;
2022 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2026 static void _d_rehash(struct dentry
* entry
)
2028 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2032 * d_rehash - add an entry back to the hash
2033 * @entry: dentry to add to the hash
2035 * Adds a dentry to the hash according to its name.
2038 void d_rehash(struct dentry
* entry
)
2040 spin_lock(&entry
->d_lock
);
2042 spin_unlock(&entry
->d_lock
);
2044 EXPORT_SYMBOL(d_rehash
);
2047 * dentry_update_name_case - update case insensitive dentry with a new name
2048 * @dentry: dentry to be updated
2051 * Update a case insensitive dentry with new case of name.
2053 * dentry must have been returned by d_lookup with name @name. Old and new
2054 * name lengths must match (ie. no d_compare which allows mismatched name
2057 * Parent inode i_mutex must be held over d_lookup and into this call (to
2058 * keep renames and concurrent inserts, and readdir(2) away).
2060 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2062 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2063 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2065 spin_lock(&dentry
->d_lock
);
2066 write_seqcount_begin(&dentry
->d_seq
);
2067 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2068 write_seqcount_end(&dentry
->d_seq
);
2069 spin_unlock(&dentry
->d_lock
);
2071 EXPORT_SYMBOL(dentry_update_name_case
);
2073 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2075 if (dname_external(target
)) {
2076 if (dname_external(dentry
)) {
2078 * Both external: swap the pointers
2080 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2083 * dentry:internal, target:external. Steal target's
2084 * storage and make target internal.
2086 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2087 dentry
->d_name
.len
+ 1);
2088 dentry
->d_name
.name
= target
->d_name
.name
;
2089 target
->d_name
.name
= target
->d_iname
;
2092 if (dname_external(dentry
)) {
2094 * dentry:external, target:internal. Give dentry's
2095 * storage to target and make dentry internal
2097 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2098 target
->d_name
.len
+ 1);
2099 target
->d_name
.name
= dentry
->d_name
.name
;
2100 dentry
->d_name
.name
= dentry
->d_iname
;
2103 * Both are internal. Just copy target to dentry
2105 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2106 target
->d_name
.len
+ 1);
2107 dentry
->d_name
.len
= target
->d_name
.len
;
2111 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2114 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2117 * XXXX: do we really need to take target->d_lock?
2119 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2120 spin_lock(&target
->d_parent
->d_lock
);
2122 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2123 spin_lock(&dentry
->d_parent
->d_lock
);
2124 spin_lock_nested(&target
->d_parent
->d_lock
,
2125 DENTRY_D_LOCK_NESTED
);
2127 spin_lock(&target
->d_parent
->d_lock
);
2128 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2129 DENTRY_D_LOCK_NESTED
);
2132 if (target
< dentry
) {
2133 spin_lock_nested(&target
->d_lock
, 2);
2134 spin_lock_nested(&dentry
->d_lock
, 3);
2136 spin_lock_nested(&dentry
->d_lock
, 2);
2137 spin_lock_nested(&target
->d_lock
, 3);
2141 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2142 struct dentry
*target
)
2144 if (target
->d_parent
!= dentry
->d_parent
)
2145 spin_unlock(&dentry
->d_parent
->d_lock
);
2146 if (target
->d_parent
!= target
)
2147 spin_unlock(&target
->d_parent
->d_lock
);
2151 * When switching names, the actual string doesn't strictly have to
2152 * be preserved in the target - because we're dropping the target
2153 * anyway. As such, we can just do a simple memcpy() to copy over
2154 * the new name before we switch.
2156 * Note that we have to be a lot more careful about getting the hash
2157 * switched - we have to switch the hash value properly even if it
2158 * then no longer matches the actual (corrupted) string of the target.
2159 * The hash value has to match the hash queue that the dentry is on..
2162 * __d_move - move a dentry
2163 * @dentry: entry to move
2164 * @target: new dentry
2166 * Update the dcache to reflect the move of a file name. Negative
2167 * dcache entries should not be moved in this way. Caller must hold
2168 * rename_lock, the i_mutex of the source and target directories,
2169 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2171 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2173 if (!dentry
->d_inode
)
2174 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2176 BUG_ON(d_ancestor(dentry
, target
));
2177 BUG_ON(d_ancestor(target
, dentry
));
2179 dentry_lock_for_move(dentry
, target
);
2181 write_seqcount_begin(&dentry
->d_seq
);
2182 write_seqcount_begin(&target
->d_seq
);
2184 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2187 * Move the dentry to the target hash queue. Don't bother checking
2188 * for the same hash queue because of how unlikely it is.
2191 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2193 /* Unhash the target: dput() will then get rid of it */
2196 list_del(&dentry
->d_u
.d_child
);
2197 list_del(&target
->d_u
.d_child
);
2199 /* Switch the names.. */
2200 switch_names(dentry
, target
);
2201 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2203 /* ... and switch the parents */
2204 if (IS_ROOT(dentry
)) {
2205 dentry
->d_parent
= target
->d_parent
;
2206 target
->d_parent
= target
;
2207 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2209 swap(dentry
->d_parent
, target
->d_parent
);
2211 /* And add them back to the (new) parent lists */
2212 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2215 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2217 write_seqcount_end(&target
->d_seq
);
2218 write_seqcount_end(&dentry
->d_seq
);
2220 dentry_unlock_parents_for_move(dentry
, target
);
2221 spin_unlock(&target
->d_lock
);
2222 fsnotify_d_move(dentry
);
2223 spin_unlock(&dentry
->d_lock
);
2227 * d_move - move a dentry
2228 * @dentry: entry to move
2229 * @target: new dentry
2231 * Update the dcache to reflect the move of a file name. Negative
2232 * dcache entries should not be moved in this way. See the locking
2233 * requirements for __d_move.
2235 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2237 write_seqlock(&rename_lock
);
2238 __d_move(dentry
, target
);
2239 write_sequnlock(&rename_lock
);
2241 EXPORT_SYMBOL(d_move
);
2244 * d_ancestor - search for an ancestor
2245 * @p1: ancestor dentry
2248 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2249 * an ancestor of p2, else NULL.
2251 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2255 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2256 if (p
->d_parent
== p1
)
2263 * This helper attempts to cope with remotely renamed directories
2265 * It assumes that the caller is already holding
2266 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2268 * Note: If ever the locking in lock_rename() changes, then please
2269 * remember to update this too...
2271 static struct dentry
*__d_unalias(struct inode
*inode
,
2272 struct dentry
*dentry
, struct dentry
*alias
)
2274 struct mutex
*m1
= NULL
, *m2
= NULL
;
2277 /* If alias and dentry share a parent, then no extra locks required */
2278 if (alias
->d_parent
== dentry
->d_parent
)
2281 /* See lock_rename() */
2282 ret
= ERR_PTR(-EBUSY
);
2283 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2285 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2286 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2288 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2290 __d_move(alias
, dentry
);
2293 spin_unlock(&inode
->i_lock
);
2302 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2303 * named dentry in place of the dentry to be replaced.
2304 * returns with anon->d_lock held!
2306 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2308 struct dentry
*dparent
, *aparent
;
2310 dentry_lock_for_move(anon
, dentry
);
2312 write_seqcount_begin(&dentry
->d_seq
);
2313 write_seqcount_begin(&anon
->d_seq
);
2315 dparent
= dentry
->d_parent
;
2316 aparent
= anon
->d_parent
;
2318 switch_names(dentry
, anon
);
2319 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2321 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2322 list_del(&dentry
->d_u
.d_child
);
2323 if (!IS_ROOT(dentry
))
2324 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2326 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2328 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2329 list_del(&anon
->d_u
.d_child
);
2331 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2333 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2335 write_seqcount_end(&dentry
->d_seq
);
2336 write_seqcount_end(&anon
->d_seq
);
2338 dentry_unlock_parents_for_move(anon
, dentry
);
2339 spin_unlock(&dentry
->d_lock
);
2341 /* anon->d_lock still locked, returns locked */
2342 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2346 * d_materialise_unique - introduce an inode into the tree
2347 * @dentry: candidate dentry
2348 * @inode: inode to bind to the dentry, to which aliases may be attached
2350 * Introduces an dentry into the tree, substituting an extant disconnected
2351 * root directory alias in its place if there is one. Caller must hold the
2352 * i_mutex of the parent directory.
2354 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2356 struct dentry
*actual
;
2358 BUG_ON(!d_unhashed(dentry
));
2362 __d_instantiate(dentry
, NULL
);
2367 spin_lock(&inode
->i_lock
);
2369 if (S_ISDIR(inode
->i_mode
)) {
2370 struct dentry
*alias
;
2372 /* Does an aliased dentry already exist? */
2373 alias
= __d_find_alias(inode
, 0);
2376 write_seqlock(&rename_lock
);
2378 if (d_ancestor(alias
, dentry
)) {
2379 /* Check for loops */
2380 actual
= ERR_PTR(-ELOOP
);
2381 } else if (IS_ROOT(alias
)) {
2382 /* Is this an anonymous mountpoint that we
2383 * could splice into our tree? */
2384 __d_materialise_dentry(dentry
, alias
);
2385 write_sequnlock(&rename_lock
);
2389 /* Nope, but we must(!) avoid directory
2391 actual
= __d_unalias(inode
, dentry
, alias
);
2393 write_sequnlock(&rename_lock
);
2394 if (IS_ERR(actual
)) {
2395 if (PTR_ERR(actual
) == -ELOOP
)
2396 pr_warn_ratelimited(
2397 "VFS: Lookup of '%s' in %s %s"
2398 " would have caused loop\n",
2399 dentry
->d_name
.name
,
2400 inode
->i_sb
->s_type
->name
,
2408 /* Add a unique reference */
2409 actual
= __d_instantiate_unique(dentry
, inode
);
2413 BUG_ON(!d_unhashed(actual
));
2415 spin_lock(&actual
->d_lock
);
2418 spin_unlock(&actual
->d_lock
);
2419 spin_unlock(&inode
->i_lock
);
2421 if (actual
== dentry
) {
2422 security_d_instantiate(dentry
, inode
);
2429 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2431 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2435 return -ENAMETOOLONG
;
2437 memcpy(*buffer
, str
, namelen
);
2441 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2443 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2447 * prepend_path - Prepend path string to a buffer
2448 * @path: the dentry/vfsmount to report
2449 * @root: root vfsmnt/dentry
2450 * @buffer: pointer to the end of the buffer
2451 * @buflen: pointer to buffer length
2453 * Caller holds the rename_lock.
2455 static int prepend_path(const struct path
*path
,
2456 const struct path
*root
,
2457 char **buffer
, int *buflen
)
2459 struct dentry
*dentry
= path
->dentry
;
2460 struct vfsmount
*vfsmnt
= path
->mnt
;
2464 br_read_lock(vfsmount_lock
);
2465 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2466 struct dentry
* parent
;
2468 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2470 if (vfsmnt
->mnt_parent
== vfsmnt
) {
2473 dentry
= vfsmnt
->mnt_mountpoint
;
2474 vfsmnt
= vfsmnt
->mnt_parent
;
2477 parent
= dentry
->d_parent
;
2479 spin_lock(&dentry
->d_lock
);
2480 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2481 spin_unlock(&dentry
->d_lock
);
2483 error
= prepend(buffer
, buflen
, "/", 1);
2491 if (!error
&& !slash
)
2492 error
= prepend(buffer
, buflen
, "/", 1);
2495 br_read_unlock(vfsmount_lock
);
2500 * Filesystems needing to implement special "root names"
2501 * should do so with ->d_dname()
2503 if (IS_ROOT(dentry
) &&
2504 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2505 WARN(1, "Root dentry has weird name <%.*s>\n",
2506 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2509 error
= prepend(buffer
, buflen
, "/", 1);
2511 error
= vfsmnt
->mnt_ns
? 1 : 2;
2516 * __d_path - return the path of a dentry
2517 * @path: the dentry/vfsmount to report
2518 * @root: root vfsmnt/dentry
2519 * @buf: buffer to return value in
2520 * @buflen: buffer length
2522 * Convert a dentry into an ASCII path name.
2524 * Returns a pointer into the buffer or an error code if the
2525 * path was too long.
2527 * "buflen" should be positive.
2529 * If the path is not reachable from the supplied root, return %NULL.
2531 char *__d_path(const struct path
*path
,
2532 const struct path
*root
,
2533 char *buf
, int buflen
)
2535 char *res
= buf
+ buflen
;
2538 prepend(&res
, &buflen
, "\0", 1);
2539 write_seqlock(&rename_lock
);
2540 error
= prepend_path(path
, root
, &res
, &buflen
);
2541 write_sequnlock(&rename_lock
);
2544 return ERR_PTR(error
);
2550 char *d_absolute_path(const struct path
*path
,
2551 char *buf
, int buflen
)
2553 struct path root
= {};
2554 char *res
= buf
+ buflen
;
2557 prepend(&res
, &buflen
, "\0", 1);
2558 write_seqlock(&rename_lock
);
2559 error
= prepend_path(path
, &root
, &res
, &buflen
);
2560 write_sequnlock(&rename_lock
);
2565 return ERR_PTR(error
);
2570 * same as __d_path but appends "(deleted)" for unlinked files.
2572 static int path_with_deleted(const struct path
*path
,
2573 const struct path
*root
,
2574 char **buf
, int *buflen
)
2576 prepend(buf
, buflen
, "\0", 1);
2577 if (d_unlinked(path
->dentry
)) {
2578 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2583 return prepend_path(path
, root
, buf
, buflen
);
2586 static int prepend_unreachable(char **buffer
, int *buflen
)
2588 return prepend(buffer
, buflen
, "(unreachable)", 13);
2592 * d_path - return the path of a dentry
2593 * @path: path to report
2594 * @buf: buffer to return value in
2595 * @buflen: buffer length
2597 * Convert a dentry into an ASCII path name. If the entry has been deleted
2598 * the string " (deleted)" is appended. Note that this is ambiguous.
2600 * Returns a pointer into the buffer or an error code if the path was
2601 * too long. Note: Callers should use the returned pointer, not the passed
2602 * in buffer, to use the name! The implementation often starts at an offset
2603 * into the buffer, and may leave 0 bytes at the start.
2605 * "buflen" should be positive.
2607 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2609 char *res
= buf
+ buflen
;
2614 * We have various synthetic filesystems that never get mounted. On
2615 * these filesystems dentries are never used for lookup purposes, and
2616 * thus don't need to be hashed. They also don't need a name until a
2617 * user wants to identify the object in /proc/pid/fd/. The little hack
2618 * below allows us to generate a name for these objects on demand:
2620 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2621 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2623 get_fs_root(current
->fs
, &root
);
2624 write_seqlock(&rename_lock
);
2625 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2627 res
= ERR_PTR(error
);
2628 write_sequnlock(&rename_lock
);
2632 EXPORT_SYMBOL(d_path
);
2635 * d_path_with_unreachable - return the path of a dentry
2636 * @path: path to report
2637 * @buf: buffer to return value in
2638 * @buflen: buffer length
2640 * The difference from d_path() is that this prepends "(unreachable)"
2641 * to paths which are unreachable from the current process' root.
2643 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2645 char *res
= buf
+ buflen
;
2649 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2650 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2652 get_fs_root(current
->fs
, &root
);
2653 write_seqlock(&rename_lock
);
2654 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2656 error
= prepend_unreachable(&res
, &buflen
);
2657 write_sequnlock(&rename_lock
);
2660 res
= ERR_PTR(error
);
2666 * Helper function for dentry_operations.d_dname() members
2668 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2669 const char *fmt
, ...)
2675 va_start(args
, fmt
);
2676 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2679 if (sz
> sizeof(temp
) || sz
> buflen
)
2680 return ERR_PTR(-ENAMETOOLONG
);
2682 buffer
+= buflen
- sz
;
2683 return memcpy(buffer
, temp
, sz
);
2687 * Write full pathname from the root of the filesystem into the buffer.
2689 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2691 char *end
= buf
+ buflen
;
2694 prepend(&end
, &buflen
, "\0", 1);
2701 while (!IS_ROOT(dentry
)) {
2702 struct dentry
*parent
= dentry
->d_parent
;
2706 spin_lock(&dentry
->d_lock
);
2707 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2708 spin_unlock(&dentry
->d_lock
);
2709 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2717 return ERR_PTR(-ENAMETOOLONG
);
2720 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2724 write_seqlock(&rename_lock
);
2725 retval
= __dentry_path(dentry
, buf
, buflen
);
2726 write_sequnlock(&rename_lock
);
2730 EXPORT_SYMBOL(dentry_path_raw
);
2732 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2737 write_seqlock(&rename_lock
);
2738 if (d_unlinked(dentry
)) {
2740 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2744 retval
= __dentry_path(dentry
, buf
, buflen
);
2745 write_sequnlock(&rename_lock
);
2746 if (!IS_ERR(retval
) && p
)
2747 *p
= '/'; /* restore '/' overriden with '\0' */
2750 return ERR_PTR(-ENAMETOOLONG
);
2754 * NOTE! The user-level library version returns a
2755 * character pointer. The kernel system call just
2756 * returns the length of the buffer filled (which
2757 * includes the ending '\0' character), or a negative
2758 * error value. So libc would do something like
2760 * char *getcwd(char * buf, size_t size)
2764 * retval = sys_getcwd(buf, size);
2771 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2774 struct path pwd
, root
;
2775 char *page
= (char *) __get_free_page(GFP_USER
);
2780 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2783 write_seqlock(&rename_lock
);
2784 if (!d_unlinked(pwd
.dentry
)) {
2786 char *cwd
= page
+ PAGE_SIZE
;
2787 int buflen
= PAGE_SIZE
;
2789 prepend(&cwd
, &buflen
, "\0", 1);
2790 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2791 write_sequnlock(&rename_lock
);
2796 /* Unreachable from current root */
2798 error
= prepend_unreachable(&cwd
, &buflen
);
2804 len
= PAGE_SIZE
+ page
- cwd
;
2807 if (copy_to_user(buf
, cwd
, len
))
2811 write_sequnlock(&rename_lock
);
2817 free_page((unsigned long) page
);
2822 * Test whether new_dentry is a subdirectory of old_dentry.
2824 * Trivially implemented using the dcache structure
2828 * is_subdir - is new dentry a subdirectory of old_dentry
2829 * @new_dentry: new dentry
2830 * @old_dentry: old dentry
2832 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2833 * Returns 0 otherwise.
2834 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2837 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2842 if (new_dentry
== old_dentry
)
2846 /* for restarting inner loop in case of seq retry */
2847 seq
= read_seqbegin(&rename_lock
);
2849 * Need rcu_readlock to protect against the d_parent trashing
2853 if (d_ancestor(old_dentry
, new_dentry
))
2858 } while (read_seqretry(&rename_lock
, seq
));
2863 int path_is_under(struct path
*path1
, struct path
*path2
)
2865 struct vfsmount
*mnt
= path1
->mnt
;
2866 struct dentry
*dentry
= path1
->dentry
;
2869 br_read_lock(vfsmount_lock
);
2870 if (mnt
!= path2
->mnt
) {
2872 if (mnt
->mnt_parent
== mnt
) {
2873 br_read_unlock(vfsmount_lock
);
2876 if (mnt
->mnt_parent
== path2
->mnt
)
2878 mnt
= mnt
->mnt_parent
;
2880 dentry
= mnt
->mnt_mountpoint
;
2882 res
= is_subdir(dentry
, path2
->dentry
);
2883 br_read_unlock(vfsmount_lock
);
2886 EXPORT_SYMBOL(path_is_under
);
2888 void d_genocide(struct dentry
*root
)
2890 struct dentry
*this_parent
;
2891 struct list_head
*next
;
2895 seq
= read_seqbegin(&rename_lock
);
2898 spin_lock(&this_parent
->d_lock
);
2900 next
= this_parent
->d_subdirs
.next
;
2902 while (next
!= &this_parent
->d_subdirs
) {
2903 struct list_head
*tmp
= next
;
2904 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2907 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2908 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2909 spin_unlock(&dentry
->d_lock
);
2912 if (!list_empty(&dentry
->d_subdirs
)) {
2913 spin_unlock(&this_parent
->d_lock
);
2914 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2915 this_parent
= dentry
;
2916 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2919 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2920 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2923 spin_unlock(&dentry
->d_lock
);
2925 if (this_parent
!= root
) {
2926 struct dentry
*child
= this_parent
;
2927 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2928 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2929 this_parent
->d_count
--;
2931 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2934 next
= child
->d_u
.d_child
.next
;
2937 spin_unlock(&this_parent
->d_lock
);
2938 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2941 write_sequnlock(&rename_lock
);
2946 write_seqlock(&rename_lock
);
2951 * find_inode_number - check for dentry with name
2952 * @dir: directory to check
2953 * @name: Name to find.
2955 * Check whether a dentry already exists for the given name,
2956 * and return the inode number if it has an inode. Otherwise
2959 * This routine is used to post-process directory listings for
2960 * filesystems using synthetic inode numbers, and is necessary
2961 * to keep getcwd() working.
2964 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2966 struct dentry
* dentry
;
2969 dentry
= d_hash_and_lookup(dir
, name
);
2971 if (dentry
->d_inode
)
2972 ino
= dentry
->d_inode
->i_ino
;
2977 EXPORT_SYMBOL(find_inode_number
);
2979 static __initdata
unsigned long dhash_entries
;
2980 static int __init
set_dhash_entries(char *str
)
2984 dhash_entries
= simple_strtoul(str
, &str
, 0);
2987 __setup("dhash_entries=", set_dhash_entries
);
2989 static void __init
dcache_init_early(void)
2993 /* If hashes are distributed across NUMA nodes, defer
2994 * hash allocation until vmalloc space is available.
3000 alloc_large_system_hash("Dentry cache",
3001 sizeof(struct hlist_bl_head
),
3009 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
3010 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3013 static void __init
dcache_init(void)
3018 * A constructor could be added for stable state like the lists,
3019 * but it is probably not worth it because of the cache nature
3022 dentry_cache
= KMEM_CACHE(dentry
,
3023 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3025 /* Hash may have been set up in dcache_init_early */
3030 alloc_large_system_hash("Dentry cache",
3031 sizeof(struct hlist_bl_head
),
3039 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
3040 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3043 /* SLAB cache for __getname() consumers */
3044 struct kmem_cache
*names_cachep __read_mostly
;
3045 EXPORT_SYMBOL(names_cachep
);
3047 EXPORT_SYMBOL(d_genocide
);
3049 void __init
vfs_caches_init_early(void)
3051 dcache_init_early();
3055 void __init
vfs_caches_init(unsigned long mempages
)
3057 unsigned long reserve
;
3059 /* Base hash sizes on available memory, with a reserve equal to
3060 150% of current kernel size */
3062 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3063 mempages
-= reserve
;
3065 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3066 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3070 files_init(mempages
);