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/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
45 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
47 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
51 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
59 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
64 * dentry->d_inode->i_lock
67 * dcache_hash_bucket lock
70 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
73 * dentry->d_parent->d_lock
76 * If no ancestor relationship:
77 * if (dentry1 < dentry2)
81 int sysctl_vfs_cache_pressure __read_mostly
= 100;
82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
84 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
85 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
87 EXPORT_SYMBOL(rename_lock
);
89 static struct kmem_cache
*dentry_cache __read_mostly
;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
99 #define D_HASHBITS d_hash_shift
100 #define D_HASHMASK d_hash_mask
102 static unsigned int d_hash_mask __read_mostly
;
103 static unsigned int d_hash_shift __read_mostly
;
105 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
107 static inline struct hlist_bl_head
*d_hash(const struct dentry
*parent
,
110 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
111 hash
= hash
+ (hash
>> D_HASHBITS
);
112 return dentry_hashtable
+ (hash
& D_HASHMASK
);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat
= {
120 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 static int get_nr_dentry(void)
127 for_each_possible_cpu(i
)
128 sum
+= per_cpu(nr_dentry
, i
);
129 return sum
< 0 ? 0 : sum
;
132 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
133 size_t *lenp
, loff_t
*ppos
)
135 dentry_stat
.nr_dentry
= get_nr_dentry();
136 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
144 #ifdef CONFIG_DCACHE_WORD_ACCESS
146 #include <asm/word-at-a-time.h>
148 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149 * aligned allocation for this particular component. We don't
150 * strictly need the load_unaligned_zeropad() safety, but it
151 * doesn't hurt either.
153 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154 * need the careful unaligned handling.
156 static inline int dentry_cmp(const unsigned char *cs
, size_t scount
,
157 const unsigned char *ct
, size_t tcount
)
159 unsigned long a
,b
,mask
;
161 if (unlikely(scount
!= tcount
))
165 a
= load_unaligned_zeropad(cs
);
166 b
= load_unaligned_zeropad(ct
);
167 if (tcount
< sizeof(unsigned long))
169 if (unlikely(a
!= b
))
171 cs
+= sizeof(unsigned long);
172 ct
+= sizeof(unsigned long);
173 tcount
-= sizeof(unsigned long);
177 mask
= ~(~0ul << tcount
*8);
178 return unlikely(!!((a
^ b
) & mask
));
183 static inline int dentry_cmp(const unsigned char *cs
, size_t scount
,
184 const unsigned char *ct
, size_t tcount
)
186 if (scount
!= tcount
)
201 static void __d_free(struct rcu_head
*head
)
203 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
205 WARN_ON(!list_empty(&dentry
->d_alias
));
206 if (dname_external(dentry
))
207 kfree(dentry
->d_name
.name
);
208 kmem_cache_free(dentry_cache
, dentry
);
214 static void d_free(struct dentry
*dentry
)
216 BUG_ON(dentry
->d_count
);
217 this_cpu_dec(nr_dentry
);
218 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
219 dentry
->d_op
->d_release(dentry
);
221 /* if dentry was never visible to RCU, immediate free is OK */
222 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
223 __d_free(&dentry
->d_u
.d_rcu
);
225 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
229 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
230 * @dentry: the target dentry
231 * After this call, in-progress rcu-walk path lookup will fail. This
232 * should be called after unhashing, and after changing d_inode (if
233 * the dentry has not already been unhashed).
235 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
237 assert_spin_locked(&dentry
->d_lock
);
238 /* Go through a barrier */
239 write_seqcount_barrier(&dentry
->d_seq
);
243 * Release the dentry's inode, using the filesystem
244 * d_iput() operation if defined. Dentry has no refcount
247 static void dentry_iput(struct dentry
* dentry
)
248 __releases(dentry
->d_lock
)
249 __releases(dentry
->d_inode
->i_lock
)
251 struct inode
*inode
= dentry
->d_inode
;
253 dentry
->d_inode
= NULL
;
254 list_del_init(&dentry
->d_alias
);
255 spin_unlock(&dentry
->d_lock
);
256 spin_unlock(&inode
->i_lock
);
258 fsnotify_inoderemove(inode
);
259 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
260 dentry
->d_op
->d_iput(dentry
, inode
);
264 spin_unlock(&dentry
->d_lock
);
269 * Release the dentry's inode, using the filesystem
270 * d_iput() operation if defined. dentry remains in-use.
272 static void dentry_unlink_inode(struct dentry
* dentry
)
273 __releases(dentry
->d_lock
)
274 __releases(dentry
->d_inode
->i_lock
)
276 struct inode
*inode
= dentry
->d_inode
;
277 dentry
->d_inode
= NULL
;
278 list_del_init(&dentry
->d_alias
);
279 dentry_rcuwalk_barrier(dentry
);
280 spin_unlock(&dentry
->d_lock
);
281 spin_unlock(&inode
->i_lock
);
283 fsnotify_inoderemove(inode
);
284 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
285 dentry
->d_op
->d_iput(dentry
, inode
);
291 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
293 static void dentry_lru_add(struct dentry
*dentry
)
295 if (list_empty(&dentry
->d_lru
)) {
296 spin_lock(&dcache_lru_lock
);
297 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
298 dentry
->d_sb
->s_nr_dentry_unused
++;
299 dentry_stat
.nr_unused
++;
300 spin_unlock(&dcache_lru_lock
);
304 static void __dentry_lru_del(struct dentry
*dentry
)
306 list_del_init(&dentry
->d_lru
);
307 dentry
->d_flags
&= ~DCACHE_SHRINK_LIST
;
308 dentry
->d_sb
->s_nr_dentry_unused
--;
309 dentry_stat
.nr_unused
--;
313 * Remove a dentry with references from the LRU.
315 static void dentry_lru_del(struct dentry
*dentry
)
317 if (!list_empty(&dentry
->d_lru
)) {
318 spin_lock(&dcache_lru_lock
);
319 __dentry_lru_del(dentry
);
320 spin_unlock(&dcache_lru_lock
);
325 * Remove a dentry that is unreferenced and about to be pruned
326 * (unhashed and destroyed) from the LRU, and inform the file system.
327 * This wrapper should be called _prior_ to unhashing a victim dentry.
329 static void dentry_lru_prune(struct dentry
*dentry
)
331 if (!list_empty(&dentry
->d_lru
)) {
332 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
333 dentry
->d_op
->d_prune(dentry
);
335 spin_lock(&dcache_lru_lock
);
336 __dentry_lru_del(dentry
);
337 spin_unlock(&dcache_lru_lock
);
341 static void dentry_lru_move_list(struct dentry
*dentry
, struct list_head
*list
)
343 spin_lock(&dcache_lru_lock
);
344 if (list_empty(&dentry
->d_lru
)) {
345 list_add_tail(&dentry
->d_lru
, list
);
346 dentry
->d_sb
->s_nr_dentry_unused
++;
347 dentry_stat
.nr_unused
++;
349 list_move_tail(&dentry
->d_lru
, list
);
351 spin_unlock(&dcache_lru_lock
);
355 * d_kill - kill dentry and return parent
356 * @dentry: dentry to kill
357 * @parent: parent dentry
359 * The dentry must already be unhashed and removed from the LRU.
361 * If this is the root of the dentry tree, return NULL.
363 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
366 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
367 __releases(dentry
->d_lock
)
368 __releases(parent
->d_lock
)
369 __releases(dentry
->d_inode
->i_lock
)
371 list_del(&dentry
->d_u
.d_child
);
373 * Inform try_to_ascend() that we are no longer attached to the
376 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
378 spin_unlock(&parent
->d_lock
);
381 * dentry_iput drops the locks, at which point nobody (except
382 * transient RCU lookups) can reach this dentry.
389 * Unhash a dentry without inserting an RCU walk barrier or checking that
390 * dentry->d_lock is locked. The caller must take care of that, if
393 static void __d_shrink(struct dentry
*dentry
)
395 if (!d_unhashed(dentry
)) {
396 struct hlist_bl_head
*b
;
397 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
398 b
= &dentry
->d_sb
->s_anon
;
400 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
403 __hlist_bl_del(&dentry
->d_hash
);
404 dentry
->d_hash
.pprev
= NULL
;
410 * d_drop - drop a dentry
411 * @dentry: dentry to drop
413 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
414 * be found through a VFS lookup any more. Note that this is different from
415 * deleting the dentry - d_delete will try to mark the dentry negative if
416 * possible, giving a successful _negative_ lookup, while d_drop will
417 * just make the cache lookup fail.
419 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
420 * reason (NFS timeouts or autofs deletes).
422 * __d_drop requires dentry->d_lock.
424 void __d_drop(struct dentry
*dentry
)
426 if (!d_unhashed(dentry
)) {
428 dentry_rcuwalk_barrier(dentry
);
431 EXPORT_SYMBOL(__d_drop
);
433 void d_drop(struct dentry
*dentry
)
435 spin_lock(&dentry
->d_lock
);
437 spin_unlock(&dentry
->d_lock
);
439 EXPORT_SYMBOL(d_drop
);
442 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
443 * @dentry: dentry to drop
445 * This is called when we do a lookup on a placeholder dentry that needed to be
446 * looked up. The dentry should have been hashed in order for it to be found by
447 * the lookup code, but now needs to be unhashed while we do the actual lookup
448 * and clear the DCACHE_NEED_LOOKUP flag.
450 void d_clear_need_lookup(struct dentry
*dentry
)
452 spin_lock(&dentry
->d_lock
);
454 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
455 spin_unlock(&dentry
->d_lock
);
457 EXPORT_SYMBOL(d_clear_need_lookup
);
460 * Finish off a dentry we've decided to kill.
461 * dentry->d_lock must be held, returns with it unlocked.
462 * If ref is non-zero, then decrement the refcount too.
463 * Returns dentry requiring refcount drop, or NULL if we're done.
465 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
466 __releases(dentry
->d_lock
)
469 struct dentry
*parent
;
471 inode
= dentry
->d_inode
;
472 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
474 spin_unlock(&dentry
->d_lock
);
476 return dentry
; /* try again with same dentry */
481 parent
= dentry
->d_parent
;
482 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
484 spin_unlock(&inode
->i_lock
);
491 * if dentry was on the d_lru list delete it from there.
492 * inform the fs via d_prune that this dentry is about to be
493 * unhashed and destroyed.
495 dentry_lru_prune(dentry
);
496 /* if it was on the hash then remove it */
498 return d_kill(dentry
, parent
);
504 * This is complicated by the fact that we do not want to put
505 * dentries that are no longer on any hash chain on the unused
506 * list: we'd much rather just get rid of them immediately.
508 * However, that implies that we have to traverse the dentry
509 * tree upwards to the parents which might _also_ now be
510 * scheduled for deletion (it may have been only waiting for
511 * its last child to go away).
513 * This tail recursion is done by hand as we don't want to depend
514 * on the compiler to always get this right (gcc generally doesn't).
515 * Real recursion would eat up our stack space.
519 * dput - release a dentry
520 * @dentry: dentry to release
522 * Release a dentry. This will drop the usage count and if appropriate
523 * call the dentry unlink method as well as removing it from the queues and
524 * releasing its resources. If the parent dentries were scheduled for release
525 * they too may now get deleted.
527 void dput(struct dentry
*dentry
)
533 if (dentry
->d_count
== 1)
535 spin_lock(&dentry
->d_lock
);
536 BUG_ON(!dentry
->d_count
);
537 if (dentry
->d_count
> 1) {
539 spin_unlock(&dentry
->d_lock
);
543 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
544 if (dentry
->d_op
->d_delete(dentry
))
548 /* Unreachable? Get rid of it */
549 if (d_unhashed(dentry
))
553 * If this dentry needs lookup, don't set the referenced flag so that it
554 * is more likely to be cleaned up by the dcache shrinker in case of
557 if (!d_need_lookup(dentry
))
558 dentry
->d_flags
|= DCACHE_REFERENCED
;
559 dentry_lru_add(dentry
);
562 spin_unlock(&dentry
->d_lock
);
566 dentry
= dentry_kill(dentry
, 1);
573 * d_invalidate - invalidate a dentry
574 * @dentry: dentry to invalidate
576 * Try to invalidate the dentry if it turns out to be
577 * possible. If there are other dentries that can be
578 * reached through this one we can't delete it and we
579 * return -EBUSY. On success we return 0.
584 int d_invalidate(struct dentry
* dentry
)
587 * If it's already been dropped, return OK.
589 spin_lock(&dentry
->d_lock
);
590 if (d_unhashed(dentry
)) {
591 spin_unlock(&dentry
->d_lock
);
595 * Check whether to do a partial shrink_dcache
596 * to get rid of unused child entries.
598 if (!list_empty(&dentry
->d_subdirs
)) {
599 spin_unlock(&dentry
->d_lock
);
600 shrink_dcache_parent(dentry
);
601 spin_lock(&dentry
->d_lock
);
605 * Somebody else still using it?
607 * If it's a directory, we can't drop it
608 * for fear of somebody re-populating it
609 * with children (even though dropping it
610 * would make it unreachable from the root,
611 * we might still populate it if it was a
612 * working directory or similar).
613 * We also need to leave mountpoints alone,
616 if (dentry
->d_count
> 1 && dentry
->d_inode
) {
617 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
618 spin_unlock(&dentry
->d_lock
);
624 spin_unlock(&dentry
->d_lock
);
627 EXPORT_SYMBOL(d_invalidate
);
629 /* This must be called with d_lock held */
630 static inline void __dget_dlock(struct dentry
*dentry
)
635 static inline void __dget(struct dentry
*dentry
)
637 spin_lock(&dentry
->d_lock
);
638 __dget_dlock(dentry
);
639 spin_unlock(&dentry
->d_lock
);
642 struct dentry
*dget_parent(struct dentry
*dentry
)
648 * Don't need rcu_dereference because we re-check it was correct under
652 ret
= dentry
->d_parent
;
653 spin_lock(&ret
->d_lock
);
654 if (unlikely(ret
!= dentry
->d_parent
)) {
655 spin_unlock(&ret
->d_lock
);
660 BUG_ON(!ret
->d_count
);
662 spin_unlock(&ret
->d_lock
);
665 EXPORT_SYMBOL(dget_parent
);
668 * d_find_alias - grab a hashed alias of inode
669 * @inode: inode in question
670 * @want_discon: flag, used by d_splice_alias, to request
671 * that only a DISCONNECTED alias be returned.
673 * If inode has a hashed alias, or is a directory and has any alias,
674 * acquire the reference to alias and return it. Otherwise return NULL.
675 * Notice that if inode is a directory there can be only one alias and
676 * it can be unhashed only if it has no children, or if it is the root
679 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
680 * any other hashed alias over that one unless @want_discon is set,
681 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
683 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
685 struct dentry
*alias
, *discon_alias
;
689 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
690 spin_lock(&alias
->d_lock
);
691 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
692 if (IS_ROOT(alias
) &&
693 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
694 discon_alias
= alias
;
695 } else if (!want_discon
) {
697 spin_unlock(&alias
->d_lock
);
701 spin_unlock(&alias
->d_lock
);
704 alias
= discon_alias
;
705 spin_lock(&alias
->d_lock
);
706 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
707 if (IS_ROOT(alias
) &&
708 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
710 spin_unlock(&alias
->d_lock
);
714 spin_unlock(&alias
->d_lock
);
720 struct dentry
*d_find_alias(struct inode
*inode
)
722 struct dentry
*de
= NULL
;
724 if (!list_empty(&inode
->i_dentry
)) {
725 spin_lock(&inode
->i_lock
);
726 de
= __d_find_alias(inode
, 0);
727 spin_unlock(&inode
->i_lock
);
731 EXPORT_SYMBOL(d_find_alias
);
734 * Try to kill dentries associated with this inode.
735 * WARNING: you must own a reference to inode.
737 void d_prune_aliases(struct inode
*inode
)
739 struct dentry
*dentry
;
741 spin_lock(&inode
->i_lock
);
742 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
743 spin_lock(&dentry
->d_lock
);
744 if (!dentry
->d_count
) {
745 __dget_dlock(dentry
);
747 spin_unlock(&dentry
->d_lock
);
748 spin_unlock(&inode
->i_lock
);
752 spin_unlock(&dentry
->d_lock
);
754 spin_unlock(&inode
->i_lock
);
756 EXPORT_SYMBOL(d_prune_aliases
);
759 * Try to throw away a dentry - free the inode, dput the parent.
760 * Requires dentry->d_lock is held, and dentry->d_count == 0.
761 * Releases dentry->d_lock.
763 * This may fail if locks cannot be acquired no problem, just try again.
765 static void try_prune_one_dentry(struct dentry
*dentry
)
766 __releases(dentry
->d_lock
)
768 struct dentry
*parent
;
770 parent
= dentry_kill(dentry
, 0);
772 * If dentry_kill returns NULL, we have nothing more to do.
773 * if it returns the same dentry, trylocks failed. In either
774 * case, just loop again.
776 * Otherwise, we need to prune ancestors too. This is necessary
777 * to prevent quadratic behavior of shrink_dcache_parent(), but
778 * is also expected to be beneficial in reducing dentry cache
783 if (parent
== dentry
)
786 /* Prune ancestors. */
789 spin_lock(&dentry
->d_lock
);
790 if (dentry
->d_count
> 1) {
792 spin_unlock(&dentry
->d_lock
);
795 dentry
= dentry_kill(dentry
, 1);
799 static void shrink_dentry_list(struct list_head
*list
)
801 struct dentry
*dentry
;
805 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
806 if (&dentry
->d_lru
== list
)
808 spin_lock(&dentry
->d_lock
);
809 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
810 spin_unlock(&dentry
->d_lock
);
815 * We found an inuse dentry which was not removed from
816 * the LRU because of laziness during lookup. Do not free
817 * it - just keep it off the LRU list.
819 if (dentry
->d_count
) {
820 dentry_lru_del(dentry
);
821 spin_unlock(&dentry
->d_lock
);
827 try_prune_one_dentry(dentry
);
835 * prune_dcache_sb - shrink the dcache
837 * @count: number of entries to try to free
839 * Attempt to shrink the superblock dcache LRU by @count entries. This is
840 * done when we need more memory an called from the superblock shrinker
843 * This function may fail to free any resources if all the dentries are in
846 void prune_dcache_sb(struct super_block
*sb
, int count
)
848 struct dentry
*dentry
;
849 LIST_HEAD(referenced
);
853 spin_lock(&dcache_lru_lock
);
854 while (!list_empty(&sb
->s_dentry_lru
)) {
855 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
856 struct dentry
, d_lru
);
857 BUG_ON(dentry
->d_sb
!= sb
);
859 if (!spin_trylock(&dentry
->d_lock
)) {
860 spin_unlock(&dcache_lru_lock
);
865 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
866 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
867 list_move(&dentry
->d_lru
, &referenced
);
868 spin_unlock(&dentry
->d_lock
);
870 list_move_tail(&dentry
->d_lru
, &tmp
);
871 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
872 spin_unlock(&dentry
->d_lock
);
876 cond_resched_lock(&dcache_lru_lock
);
878 if (!list_empty(&referenced
))
879 list_splice(&referenced
, &sb
->s_dentry_lru
);
880 spin_unlock(&dcache_lru_lock
);
882 shrink_dentry_list(&tmp
);
886 * shrink_dcache_sb - shrink dcache for a superblock
889 * Shrink the dcache for the specified super block. This is used to free
890 * the dcache before unmounting a file system.
892 void shrink_dcache_sb(struct super_block
*sb
)
896 spin_lock(&dcache_lru_lock
);
897 while (!list_empty(&sb
->s_dentry_lru
)) {
898 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
899 spin_unlock(&dcache_lru_lock
);
900 shrink_dentry_list(&tmp
);
901 spin_lock(&dcache_lru_lock
);
903 spin_unlock(&dcache_lru_lock
);
905 EXPORT_SYMBOL(shrink_dcache_sb
);
908 * destroy a single subtree of dentries for unmount
909 * - see the comments on shrink_dcache_for_umount() for a description of the
912 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
914 struct dentry
*parent
;
916 BUG_ON(!IS_ROOT(dentry
));
919 /* descend to the first leaf in the current subtree */
920 while (!list_empty(&dentry
->d_subdirs
))
921 dentry
= list_entry(dentry
->d_subdirs
.next
,
922 struct dentry
, d_u
.d_child
);
924 /* consume the dentries from this leaf up through its parents
925 * until we find one with children or run out altogether */
930 * remove the dentry from the lru, and inform
931 * the fs that this dentry is about to be
932 * unhashed and destroyed.
934 dentry_lru_prune(dentry
);
937 if (dentry
->d_count
!= 0) {
939 "BUG: Dentry %p{i=%lx,n=%s}"
941 " [unmount of %s %s]\n",
944 dentry
->d_inode
->i_ino
: 0UL,
947 dentry
->d_sb
->s_type
->name
,
952 if (IS_ROOT(dentry
)) {
954 list_del(&dentry
->d_u
.d_child
);
956 parent
= dentry
->d_parent
;
958 list_del(&dentry
->d_u
.d_child
);
961 inode
= dentry
->d_inode
;
963 dentry
->d_inode
= NULL
;
964 list_del_init(&dentry
->d_alias
);
965 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
966 dentry
->d_op
->d_iput(dentry
, inode
);
973 /* finished when we fall off the top of the tree,
974 * otherwise we ascend to the parent and move to the
975 * next sibling if there is one */
979 } while (list_empty(&dentry
->d_subdirs
));
981 dentry
= list_entry(dentry
->d_subdirs
.next
,
982 struct dentry
, d_u
.d_child
);
987 * destroy the dentries attached to a superblock on unmounting
988 * - we don't need to use dentry->d_lock because:
989 * - the superblock is detached from all mountings and open files, so the
990 * dentry trees will not be rearranged by the VFS
991 * - s_umount is write-locked, so the memory pressure shrinker will ignore
992 * any dentries belonging to this superblock that it comes across
993 * - the filesystem itself is no longer permitted to rearrange the dentries
996 void shrink_dcache_for_umount(struct super_block
*sb
)
998 struct dentry
*dentry
;
1000 if (down_read_trylock(&sb
->s_umount
))
1003 dentry
= sb
->s_root
;
1006 shrink_dcache_for_umount_subtree(dentry
);
1008 while (!hlist_bl_empty(&sb
->s_anon
)) {
1009 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
1010 shrink_dcache_for_umount_subtree(dentry
);
1015 * This tries to ascend one level of parenthood, but
1016 * we can race with renaming, so we need to re-check
1017 * the parenthood after dropping the lock and check
1018 * that the sequence number still matches.
1020 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
1022 struct dentry
*new = old
->d_parent
;
1025 spin_unlock(&old
->d_lock
);
1026 spin_lock(&new->d_lock
);
1029 * might go back up the wrong parent if we have had a rename
1032 if (new != old
->d_parent
||
1033 (old
->d_flags
& DCACHE_DISCONNECTED
) ||
1034 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1035 spin_unlock(&new->d_lock
);
1044 * Search for at least 1 mount point in the dentry's subdirs.
1045 * We descend to the next level whenever the d_subdirs
1046 * list is non-empty and continue searching.
1050 * have_submounts - check for mounts over a dentry
1051 * @parent: dentry to check.
1053 * Return true if the parent or its subdirectories contain
1056 int have_submounts(struct dentry
*parent
)
1058 struct dentry
*this_parent
;
1059 struct list_head
*next
;
1063 seq
= read_seqbegin(&rename_lock
);
1065 this_parent
= parent
;
1067 if (d_mountpoint(parent
))
1069 spin_lock(&this_parent
->d_lock
);
1071 next
= this_parent
->d_subdirs
.next
;
1073 while (next
!= &this_parent
->d_subdirs
) {
1074 struct list_head
*tmp
= next
;
1075 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1078 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1079 /* Have we found a mount point ? */
1080 if (d_mountpoint(dentry
)) {
1081 spin_unlock(&dentry
->d_lock
);
1082 spin_unlock(&this_parent
->d_lock
);
1085 if (!list_empty(&dentry
->d_subdirs
)) {
1086 spin_unlock(&this_parent
->d_lock
);
1087 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1088 this_parent
= dentry
;
1089 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1092 spin_unlock(&dentry
->d_lock
);
1095 * All done at this level ... ascend and resume the search.
1097 if (this_parent
!= parent
) {
1098 struct dentry
*child
= this_parent
;
1099 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1102 next
= child
->d_u
.d_child
.next
;
1105 spin_unlock(&this_parent
->d_lock
);
1106 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1109 write_sequnlock(&rename_lock
);
1110 return 0; /* No mount points found in tree */
1112 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1115 write_sequnlock(&rename_lock
);
1120 write_seqlock(&rename_lock
);
1123 EXPORT_SYMBOL(have_submounts
);
1126 * Search the dentry child list for the specified parent,
1127 * and move any unused dentries to the end of the unused
1128 * list for prune_dcache(). We descend to the next level
1129 * whenever the d_subdirs list is non-empty and continue
1132 * It returns zero iff there are no unused children,
1133 * otherwise it returns the number of children moved to
1134 * the end of the unused list. This may not be the total
1135 * number of unused children, because select_parent can
1136 * drop the lock and return early due to latency
1139 static int select_parent(struct dentry
*parent
, struct list_head
*dispose
)
1141 struct dentry
*this_parent
;
1142 struct list_head
*next
;
1147 seq
= read_seqbegin(&rename_lock
);
1149 this_parent
= parent
;
1150 spin_lock(&this_parent
->d_lock
);
1152 next
= this_parent
->d_subdirs
.next
;
1154 while (next
!= &this_parent
->d_subdirs
) {
1155 struct list_head
*tmp
= next
;
1156 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1159 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1162 * move only zero ref count dentries to the dispose list.
1164 * Those which are presently on the shrink list, being processed
1165 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1166 * loop in shrink_dcache_parent() might not make any progress
1169 if (dentry
->d_count
) {
1170 dentry_lru_del(dentry
);
1171 } else if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
)) {
1172 dentry_lru_move_list(dentry
, dispose
);
1173 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
1177 * We can return to the caller if we have found some (this
1178 * ensures forward progress). We'll be coming back to find
1181 if (found
&& need_resched()) {
1182 spin_unlock(&dentry
->d_lock
);
1187 * Descend a level if the d_subdirs list is non-empty.
1189 if (!list_empty(&dentry
->d_subdirs
)) {
1190 spin_unlock(&this_parent
->d_lock
);
1191 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1192 this_parent
= dentry
;
1193 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1197 spin_unlock(&dentry
->d_lock
);
1200 * All done at this level ... ascend and resume the search.
1202 if (this_parent
!= parent
) {
1203 struct dentry
*child
= this_parent
;
1204 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1207 next
= child
->d_u
.d_child
.next
;
1211 spin_unlock(&this_parent
->d_lock
);
1212 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1215 write_sequnlock(&rename_lock
);
1222 write_seqlock(&rename_lock
);
1227 * shrink_dcache_parent - prune dcache
1228 * @parent: parent of entries to prune
1230 * Prune the dcache to remove unused children of the parent dentry.
1232 void shrink_dcache_parent(struct dentry
* parent
)
1237 while ((found
= select_parent(parent
, &dispose
)) != 0)
1238 shrink_dentry_list(&dispose
);
1240 EXPORT_SYMBOL(shrink_dcache_parent
);
1243 * __d_alloc - allocate a dcache entry
1244 * @sb: filesystem it will belong to
1245 * @name: qstr of the name
1247 * Allocates a dentry. It returns %NULL if there is insufficient memory
1248 * available. On a success the dentry is returned. The name passed in is
1249 * copied and the copy passed in may be reused after this call.
1252 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1254 struct dentry
*dentry
;
1257 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1261 if (name
->len
> DNAME_INLINE_LEN
-1) {
1262 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1264 kmem_cache_free(dentry_cache
, dentry
);
1268 dname
= dentry
->d_iname
;
1270 dentry
->d_name
.name
= dname
;
1272 dentry
->d_name
.len
= name
->len
;
1273 dentry
->d_name
.hash
= name
->hash
;
1274 memcpy(dname
, name
->name
, name
->len
);
1275 dname
[name
->len
] = 0;
1277 dentry
->d_count
= 1;
1278 dentry
->d_flags
= 0;
1279 spin_lock_init(&dentry
->d_lock
);
1280 seqcount_init(&dentry
->d_seq
);
1281 dentry
->d_inode
= NULL
;
1282 dentry
->d_parent
= dentry
;
1284 dentry
->d_op
= NULL
;
1285 dentry
->d_fsdata
= NULL
;
1286 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1287 INIT_LIST_HEAD(&dentry
->d_lru
);
1288 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1289 INIT_LIST_HEAD(&dentry
->d_alias
);
1290 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1291 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1293 this_cpu_inc(nr_dentry
);
1299 * d_alloc - allocate a dcache entry
1300 * @parent: parent of entry to allocate
1301 * @name: qstr of the name
1303 * Allocates a dentry. It returns %NULL if there is insufficient memory
1304 * available. On a success the dentry is returned. The name passed in is
1305 * copied and the copy passed in may be reused after this call.
1307 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1309 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1313 spin_lock(&parent
->d_lock
);
1315 * don't need child lock because it is not subject
1316 * to concurrency here
1318 __dget_dlock(parent
);
1319 dentry
->d_parent
= parent
;
1320 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1321 spin_unlock(&parent
->d_lock
);
1325 EXPORT_SYMBOL(d_alloc
);
1327 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1329 struct dentry
*dentry
= __d_alloc(sb
, name
);
1331 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1334 EXPORT_SYMBOL(d_alloc_pseudo
);
1336 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1341 q
.len
= strlen(name
);
1342 q
.hash
= full_name_hash(q
.name
, q
.len
);
1343 return d_alloc(parent
, &q
);
1345 EXPORT_SYMBOL(d_alloc_name
);
1347 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1349 WARN_ON_ONCE(dentry
->d_op
);
1350 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1352 DCACHE_OP_REVALIDATE
|
1353 DCACHE_OP_DELETE
));
1358 dentry
->d_flags
|= DCACHE_OP_HASH
;
1360 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1361 if (op
->d_revalidate
)
1362 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1364 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1366 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1369 EXPORT_SYMBOL(d_set_d_op
);
1371 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1373 spin_lock(&dentry
->d_lock
);
1375 if (unlikely(IS_AUTOMOUNT(inode
)))
1376 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1377 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1379 dentry
->d_inode
= inode
;
1380 dentry_rcuwalk_barrier(dentry
);
1381 spin_unlock(&dentry
->d_lock
);
1382 fsnotify_d_instantiate(dentry
, inode
);
1386 * d_instantiate - fill in inode information for a dentry
1387 * @entry: dentry to complete
1388 * @inode: inode to attach to this dentry
1390 * Fill in inode information in the entry.
1392 * This turns negative dentries into productive full members
1395 * NOTE! This assumes that the inode count has been incremented
1396 * (or otherwise set) by the caller to indicate that it is now
1397 * in use by the dcache.
1400 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1402 BUG_ON(!list_empty(&entry
->d_alias
));
1404 spin_lock(&inode
->i_lock
);
1405 __d_instantiate(entry
, inode
);
1407 spin_unlock(&inode
->i_lock
);
1408 security_d_instantiate(entry
, inode
);
1410 EXPORT_SYMBOL(d_instantiate
);
1413 * d_instantiate_unique - instantiate a non-aliased dentry
1414 * @entry: dentry to instantiate
1415 * @inode: inode to attach to this dentry
1417 * Fill in inode information in the entry. On success, it returns NULL.
1418 * If an unhashed alias of "entry" already exists, then we return the
1419 * aliased dentry instead and drop one reference to inode.
1421 * Note that in order to avoid conflicts with rename() etc, the caller
1422 * had better be holding the parent directory semaphore.
1424 * This also assumes that the inode count has been incremented
1425 * (or otherwise set) by the caller to indicate that it is now
1426 * in use by the dcache.
1428 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1429 struct inode
*inode
)
1431 struct dentry
*alias
;
1432 int len
= entry
->d_name
.len
;
1433 const char *name
= entry
->d_name
.name
;
1434 unsigned int hash
= entry
->d_name
.hash
;
1437 __d_instantiate(entry
, NULL
);
1441 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1442 struct qstr
*qstr
= &alias
->d_name
;
1445 * Don't need alias->d_lock here, because aliases with
1446 * d_parent == entry->d_parent are not subject to name or
1447 * parent changes, because the parent inode i_mutex is held.
1449 if (qstr
->hash
!= hash
)
1451 if (alias
->d_parent
!= entry
->d_parent
)
1453 if (dentry_cmp(qstr
->name
, qstr
->len
, name
, len
))
1459 __d_instantiate(entry
, inode
);
1463 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1465 struct dentry
*result
;
1467 BUG_ON(!list_empty(&entry
->d_alias
));
1470 spin_lock(&inode
->i_lock
);
1471 result
= __d_instantiate_unique(entry
, inode
);
1473 spin_unlock(&inode
->i_lock
);
1476 security_d_instantiate(entry
, inode
);
1480 BUG_ON(!d_unhashed(result
));
1485 EXPORT_SYMBOL(d_instantiate_unique
);
1487 struct dentry
*d_make_root(struct inode
*root_inode
)
1489 struct dentry
*res
= NULL
;
1492 static const struct qstr name
= { .name
= "/", .len
= 1 };
1494 res
= __d_alloc(root_inode
->i_sb
, &name
);
1496 d_instantiate(res
, root_inode
);
1502 EXPORT_SYMBOL(d_make_root
);
1504 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1506 struct dentry
*alias
;
1508 if (list_empty(&inode
->i_dentry
))
1510 alias
= list_first_entry(&inode
->i_dentry
, struct dentry
, d_alias
);
1516 * d_find_any_alias - find any alias for a given inode
1517 * @inode: inode to find an alias for
1519 * If any aliases exist for the given inode, take and return a
1520 * reference for one of them. If no aliases exist, return %NULL.
1522 struct dentry
*d_find_any_alias(struct inode
*inode
)
1526 spin_lock(&inode
->i_lock
);
1527 de
= __d_find_any_alias(inode
);
1528 spin_unlock(&inode
->i_lock
);
1531 EXPORT_SYMBOL(d_find_any_alias
);
1534 * d_obtain_alias - find or allocate a dentry for a given inode
1535 * @inode: inode to allocate the dentry for
1537 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1538 * similar open by handle operations. The returned dentry may be anonymous,
1539 * or may have a full name (if the inode was already in the cache).
1541 * When called on a directory inode, we must ensure that the inode only ever
1542 * has one dentry. If a dentry is found, that is returned instead of
1543 * allocating a new one.
1545 * On successful return, the reference to the inode has been transferred
1546 * to the dentry. In case of an error the reference on the inode is released.
1547 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1548 * be passed in and will be the error will be propagate to the return value,
1549 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1551 struct dentry
*d_obtain_alias(struct inode
*inode
)
1553 static const struct qstr anonstring
= { .name
= "" };
1558 return ERR_PTR(-ESTALE
);
1560 return ERR_CAST(inode
);
1562 res
= d_find_any_alias(inode
);
1566 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1568 res
= ERR_PTR(-ENOMEM
);
1572 spin_lock(&inode
->i_lock
);
1573 res
= __d_find_any_alias(inode
);
1575 spin_unlock(&inode
->i_lock
);
1580 /* attach a disconnected dentry */
1581 spin_lock(&tmp
->d_lock
);
1582 tmp
->d_inode
= inode
;
1583 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1584 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1585 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1586 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1587 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1588 spin_unlock(&tmp
->d_lock
);
1589 spin_unlock(&inode
->i_lock
);
1590 security_d_instantiate(tmp
, inode
);
1595 if (res
&& !IS_ERR(res
))
1596 security_d_instantiate(res
, inode
);
1600 EXPORT_SYMBOL(d_obtain_alias
);
1603 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1604 * @inode: the inode which may have a disconnected dentry
1605 * @dentry: a negative dentry which we want to point to the inode.
1607 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1608 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1609 * and return it, else simply d_add the inode to the dentry and return NULL.
1611 * This is needed in the lookup routine of any filesystem that is exportable
1612 * (via knfsd) so that we can build dcache paths to directories effectively.
1614 * If a dentry was found and moved, then it is returned. Otherwise NULL
1615 * is returned. This matches the expected return value of ->lookup.
1618 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1620 struct dentry
*new = NULL
;
1623 return ERR_CAST(inode
);
1625 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1626 spin_lock(&inode
->i_lock
);
1627 new = __d_find_alias(inode
, 1);
1629 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1630 spin_unlock(&inode
->i_lock
);
1631 security_d_instantiate(new, inode
);
1632 d_move(new, dentry
);
1635 /* already taking inode->i_lock, so d_add() by hand */
1636 __d_instantiate(dentry
, inode
);
1637 spin_unlock(&inode
->i_lock
);
1638 security_d_instantiate(dentry
, inode
);
1642 d_add(dentry
, inode
);
1645 EXPORT_SYMBOL(d_splice_alias
);
1648 * d_add_ci - lookup or allocate new dentry with case-exact name
1649 * @inode: the inode case-insensitive lookup has found
1650 * @dentry: the negative dentry that was passed to the parent's lookup func
1651 * @name: the case-exact name to be associated with the returned dentry
1653 * This is to avoid filling the dcache with case-insensitive names to the
1654 * same inode, only the actual correct case is stored in the dcache for
1655 * case-insensitive filesystems.
1657 * For a case-insensitive lookup match and if the the case-exact dentry
1658 * already exists in in the dcache, use it and return it.
1660 * If no entry exists with the exact case name, allocate new dentry with
1661 * the exact case, and return the spliced entry.
1663 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1667 struct dentry
*found
;
1671 * First check if a dentry matching the name already exists,
1672 * if not go ahead and create it now.
1674 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1676 new = d_alloc(dentry
->d_parent
, name
);
1682 found
= d_splice_alias(inode
, new);
1691 * If a matching dentry exists, and it's not negative use it.
1693 * Decrement the reference count to balance the iget() done
1696 if (found
->d_inode
) {
1697 if (unlikely(found
->d_inode
!= inode
)) {
1698 /* This can't happen because bad inodes are unhashed. */
1699 BUG_ON(!is_bad_inode(inode
));
1700 BUG_ON(!is_bad_inode(found
->d_inode
));
1707 * We are going to instantiate this dentry, unhash it and clear the
1708 * lookup flag so we can do that.
1710 if (unlikely(d_need_lookup(found
)))
1711 d_clear_need_lookup(found
);
1714 * Negative dentry: instantiate it unless the inode is a directory and
1715 * already has a dentry.
1717 new = d_splice_alias(inode
, found
);
1726 return ERR_PTR(error
);
1728 EXPORT_SYMBOL(d_add_ci
);
1731 * __d_lookup_rcu - search for a dentry (racy, store-free)
1732 * @parent: parent dentry
1733 * @name: qstr of name we wish to find
1734 * @seqp: returns d_seq value at the point where the dentry was found
1735 * @inode: returns dentry->d_inode when the inode was found valid.
1736 * Returns: dentry, or NULL
1738 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1739 * resolution (store-free path walking) design described in
1740 * Documentation/filesystems/path-lookup.txt.
1742 * This is not to be used outside core vfs.
1744 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1745 * held, and rcu_read_lock held. The returned dentry must not be stored into
1746 * without taking d_lock and checking d_seq sequence count against @seq
1749 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1752 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1753 * the returned dentry, so long as its parent's seqlock is checked after the
1754 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1755 * is formed, giving integrity down the path walk.
1757 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
1758 const struct qstr
*name
,
1759 unsigned *seqp
, struct inode
**inode
)
1761 unsigned int len
= name
->len
;
1762 unsigned int hash
= name
->hash
;
1763 const unsigned char *str
= name
->name
;
1764 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1765 struct hlist_bl_node
*node
;
1766 struct dentry
*dentry
;
1769 * Note: There is significant duplication with __d_lookup_rcu which is
1770 * required to prevent single threaded performance regressions
1771 * especially on architectures where smp_rmb (in seqcounts) are costly.
1772 * Keep the two functions in sync.
1776 * The hash list is protected using RCU.
1778 * Carefully use d_seq when comparing a candidate dentry, to avoid
1779 * races with d_move().
1781 * It is possible that concurrent renames can mess up our list
1782 * walk here and result in missing our dentry, resulting in the
1783 * false-negative result. d_lookup() protects against concurrent
1784 * renames using rename_lock seqlock.
1786 * See Documentation/filesystems/path-lookup.txt for more details.
1788 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1794 if (dentry
->d_name
.hash
!= hash
)
1798 seq
= read_seqcount_begin(&dentry
->d_seq
);
1799 if (dentry
->d_parent
!= parent
)
1801 if (d_unhashed(dentry
))
1803 tlen
= dentry
->d_name
.len
;
1804 tname
= dentry
->d_name
.name
;
1805 i
= dentry
->d_inode
;
1808 * This seqcount check is required to ensure name and
1809 * len are loaded atomically, so as not to walk off the
1810 * edge of memory when walking. If we could load this
1811 * atomically some other way, we could drop this check.
1813 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1815 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1816 if (parent
->d_op
->d_compare(parent
, *inode
,
1821 if (dentry_cmp(tname
, tlen
, str
, len
))
1825 * No extra seqcount check is required after the name
1826 * compare. The caller must perform a seqcount check in
1827 * order to do anything useful with the returned dentry
1838 * d_lookup - search for a dentry
1839 * @parent: parent dentry
1840 * @name: qstr of name we wish to find
1841 * Returns: dentry, or NULL
1843 * d_lookup searches the children of the parent dentry for the name in
1844 * question. If the dentry is found its reference count is incremented and the
1845 * dentry is returned. The caller must use dput to free the entry when it has
1846 * finished using it. %NULL is returned if the dentry does not exist.
1848 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1850 struct dentry
*dentry
;
1854 seq
= read_seqbegin(&rename_lock
);
1855 dentry
= __d_lookup(parent
, name
);
1858 } while (read_seqretry(&rename_lock
, seq
));
1861 EXPORT_SYMBOL(d_lookup
);
1864 * __d_lookup - search for a dentry (racy)
1865 * @parent: parent dentry
1866 * @name: qstr of name we wish to find
1867 * Returns: dentry, or NULL
1869 * __d_lookup is like d_lookup, however it may (rarely) return a
1870 * false-negative result due to unrelated rename activity.
1872 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1873 * however it must be used carefully, eg. with a following d_lookup in
1874 * the case of failure.
1876 * __d_lookup callers must be commented.
1878 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1880 unsigned int len
= name
->len
;
1881 unsigned int hash
= name
->hash
;
1882 const unsigned char *str
= name
->name
;
1883 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1884 struct hlist_bl_node
*node
;
1885 struct dentry
*found
= NULL
;
1886 struct dentry
*dentry
;
1889 * Note: There is significant duplication with __d_lookup_rcu which is
1890 * required to prevent single threaded performance regressions
1891 * especially on architectures where smp_rmb (in seqcounts) are costly.
1892 * Keep the two functions in sync.
1896 * The hash list is protected using RCU.
1898 * Take d_lock when comparing a candidate dentry, to avoid races
1901 * It is possible that concurrent renames can mess up our list
1902 * walk here and result in missing our dentry, resulting in the
1903 * false-negative result. d_lookup() protects against concurrent
1904 * renames using rename_lock seqlock.
1906 * See Documentation/filesystems/path-lookup.txt for more details.
1910 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1914 if (dentry
->d_name
.hash
!= hash
)
1917 spin_lock(&dentry
->d_lock
);
1918 if (dentry
->d_parent
!= parent
)
1920 if (d_unhashed(dentry
))
1924 * It is safe to compare names since d_move() cannot
1925 * change the qstr (protected by d_lock).
1927 tlen
= dentry
->d_name
.len
;
1928 tname
= dentry
->d_name
.name
;
1929 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1930 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1931 dentry
, dentry
->d_inode
,
1935 if (dentry_cmp(tname
, tlen
, str
, len
))
1941 spin_unlock(&dentry
->d_lock
);
1944 spin_unlock(&dentry
->d_lock
);
1952 * d_hash_and_lookup - hash the qstr then search for a dentry
1953 * @dir: Directory to search in
1954 * @name: qstr of name we wish to find
1956 * On hash failure or on lookup failure NULL is returned.
1958 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1960 struct dentry
*dentry
= NULL
;
1963 * Check for a fs-specific hash function. Note that we must
1964 * calculate the standard hash first, as the d_op->d_hash()
1965 * routine may choose to leave the hash value unchanged.
1967 name
->hash
= full_name_hash(name
->name
, name
->len
);
1968 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1969 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1972 dentry
= d_lookup(dir
, name
);
1978 * d_validate - verify dentry provided from insecure source (deprecated)
1979 * @dentry: The dentry alleged to be valid child of @dparent
1980 * @dparent: The parent dentry (known to be valid)
1982 * An insecure source has sent us a dentry, here we verify it and dget() it.
1983 * This is used by ncpfs in its readdir implementation.
1984 * Zero is returned in the dentry is invalid.
1986 * This function is slow for big directories, and deprecated, do not use it.
1988 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1990 struct dentry
*child
;
1992 spin_lock(&dparent
->d_lock
);
1993 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1994 if (dentry
== child
) {
1995 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1996 __dget_dlock(dentry
);
1997 spin_unlock(&dentry
->d_lock
);
1998 spin_unlock(&dparent
->d_lock
);
2002 spin_unlock(&dparent
->d_lock
);
2006 EXPORT_SYMBOL(d_validate
);
2009 * When a file is deleted, we have two options:
2010 * - turn this dentry into a negative dentry
2011 * - unhash this dentry and free it.
2013 * Usually, we want to just turn this into
2014 * a negative dentry, but if anybody else is
2015 * currently using the dentry or the inode
2016 * we can't do that and we fall back on removing
2017 * it from the hash queues and waiting for
2018 * it to be deleted later when it has no users
2022 * d_delete - delete a dentry
2023 * @dentry: The dentry to delete
2025 * Turn the dentry into a negative dentry if possible, otherwise
2026 * remove it from the hash queues so it can be deleted later
2029 void d_delete(struct dentry
* dentry
)
2031 struct inode
*inode
;
2034 * Are we the only user?
2037 spin_lock(&dentry
->d_lock
);
2038 inode
= dentry
->d_inode
;
2039 isdir
= S_ISDIR(inode
->i_mode
);
2040 if (dentry
->d_count
== 1) {
2041 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
2042 spin_unlock(&dentry
->d_lock
);
2046 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2047 dentry_unlink_inode(dentry
);
2048 fsnotify_nameremove(dentry
, isdir
);
2052 if (!d_unhashed(dentry
))
2055 spin_unlock(&dentry
->d_lock
);
2057 fsnotify_nameremove(dentry
, isdir
);
2059 EXPORT_SYMBOL(d_delete
);
2061 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2063 BUG_ON(!d_unhashed(entry
));
2065 entry
->d_flags
|= DCACHE_RCUACCESS
;
2066 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2070 static void _d_rehash(struct dentry
* entry
)
2072 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2076 * d_rehash - add an entry back to the hash
2077 * @entry: dentry to add to the hash
2079 * Adds a dentry to the hash according to its name.
2082 void d_rehash(struct dentry
* entry
)
2084 spin_lock(&entry
->d_lock
);
2086 spin_unlock(&entry
->d_lock
);
2088 EXPORT_SYMBOL(d_rehash
);
2091 * dentry_update_name_case - update case insensitive dentry with a new name
2092 * @dentry: dentry to be updated
2095 * Update a case insensitive dentry with new case of name.
2097 * dentry must have been returned by d_lookup with name @name. Old and new
2098 * name lengths must match (ie. no d_compare which allows mismatched name
2101 * Parent inode i_mutex must be held over d_lookup and into this call (to
2102 * keep renames and concurrent inserts, and readdir(2) away).
2104 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2106 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2107 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2109 spin_lock(&dentry
->d_lock
);
2110 write_seqcount_begin(&dentry
->d_seq
);
2111 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2112 write_seqcount_end(&dentry
->d_seq
);
2113 spin_unlock(&dentry
->d_lock
);
2115 EXPORT_SYMBOL(dentry_update_name_case
);
2117 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2119 if (dname_external(target
)) {
2120 if (dname_external(dentry
)) {
2122 * Both external: swap the pointers
2124 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2127 * dentry:internal, target:external. Steal target's
2128 * storage and make target internal.
2130 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2131 dentry
->d_name
.len
+ 1);
2132 dentry
->d_name
.name
= target
->d_name
.name
;
2133 target
->d_name
.name
= target
->d_iname
;
2136 if (dname_external(dentry
)) {
2138 * dentry:external, target:internal. Give dentry's
2139 * storage to target and make dentry internal
2141 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2142 target
->d_name
.len
+ 1);
2143 target
->d_name
.name
= dentry
->d_name
.name
;
2144 dentry
->d_name
.name
= dentry
->d_iname
;
2147 * Both are internal. Just copy target to dentry
2149 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2150 target
->d_name
.len
+ 1);
2151 dentry
->d_name
.len
= target
->d_name
.len
;
2155 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2158 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2161 * XXXX: do we really need to take target->d_lock?
2163 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2164 spin_lock(&target
->d_parent
->d_lock
);
2166 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2167 spin_lock(&dentry
->d_parent
->d_lock
);
2168 spin_lock_nested(&target
->d_parent
->d_lock
,
2169 DENTRY_D_LOCK_NESTED
);
2171 spin_lock(&target
->d_parent
->d_lock
);
2172 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2173 DENTRY_D_LOCK_NESTED
);
2176 if (target
< dentry
) {
2177 spin_lock_nested(&target
->d_lock
, 2);
2178 spin_lock_nested(&dentry
->d_lock
, 3);
2180 spin_lock_nested(&dentry
->d_lock
, 2);
2181 spin_lock_nested(&target
->d_lock
, 3);
2185 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2186 struct dentry
*target
)
2188 if (target
->d_parent
!= dentry
->d_parent
)
2189 spin_unlock(&dentry
->d_parent
->d_lock
);
2190 if (target
->d_parent
!= target
)
2191 spin_unlock(&target
->d_parent
->d_lock
);
2195 * When switching names, the actual string doesn't strictly have to
2196 * be preserved in the target - because we're dropping the target
2197 * anyway. As such, we can just do a simple memcpy() to copy over
2198 * the new name before we switch.
2200 * Note that we have to be a lot more careful about getting the hash
2201 * switched - we have to switch the hash value properly even if it
2202 * then no longer matches the actual (corrupted) string of the target.
2203 * The hash value has to match the hash queue that the dentry is on..
2206 * __d_move - move a dentry
2207 * @dentry: entry to move
2208 * @target: new dentry
2210 * Update the dcache to reflect the move of a file name. Negative
2211 * dcache entries should not be moved in this way. Caller must hold
2212 * rename_lock, the i_mutex of the source and target directories,
2213 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2215 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2217 if (!dentry
->d_inode
)
2218 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2220 BUG_ON(d_ancestor(dentry
, target
));
2221 BUG_ON(d_ancestor(target
, dentry
));
2223 dentry_lock_for_move(dentry
, target
);
2225 write_seqcount_begin(&dentry
->d_seq
);
2226 write_seqcount_begin(&target
->d_seq
);
2228 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2231 * Move the dentry to the target hash queue. Don't bother checking
2232 * for the same hash queue because of how unlikely it is.
2235 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2237 /* Unhash the target: dput() will then get rid of it */
2240 list_del(&dentry
->d_u
.d_child
);
2241 list_del(&target
->d_u
.d_child
);
2243 /* Switch the names.. */
2244 switch_names(dentry
, target
);
2245 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2247 /* ... and switch the parents */
2248 if (IS_ROOT(dentry
)) {
2249 dentry
->d_parent
= target
->d_parent
;
2250 target
->d_parent
= target
;
2251 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2253 swap(dentry
->d_parent
, target
->d_parent
);
2255 /* And add them back to the (new) parent lists */
2256 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2259 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2261 write_seqcount_end(&target
->d_seq
);
2262 write_seqcount_end(&dentry
->d_seq
);
2264 dentry_unlock_parents_for_move(dentry
, target
);
2265 spin_unlock(&target
->d_lock
);
2266 fsnotify_d_move(dentry
);
2267 spin_unlock(&dentry
->d_lock
);
2271 * d_move - move a dentry
2272 * @dentry: entry to move
2273 * @target: new dentry
2275 * Update the dcache to reflect the move of a file name. Negative
2276 * dcache entries should not be moved in this way. See the locking
2277 * requirements for __d_move.
2279 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2281 write_seqlock(&rename_lock
);
2282 __d_move(dentry
, target
);
2283 write_sequnlock(&rename_lock
);
2285 EXPORT_SYMBOL(d_move
);
2288 * d_ancestor - search for an ancestor
2289 * @p1: ancestor dentry
2292 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2293 * an ancestor of p2, else NULL.
2295 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2299 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2300 if (p
->d_parent
== p1
)
2307 * This helper attempts to cope with remotely renamed directories
2309 * It assumes that the caller is already holding
2310 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2312 * Note: If ever the locking in lock_rename() changes, then please
2313 * remember to update this too...
2315 static struct dentry
*__d_unalias(struct inode
*inode
,
2316 struct dentry
*dentry
, struct dentry
*alias
)
2318 struct mutex
*m1
= NULL
, *m2
= NULL
;
2321 /* If alias and dentry share a parent, then no extra locks required */
2322 if (alias
->d_parent
== dentry
->d_parent
)
2325 /* See lock_rename() */
2326 ret
= ERR_PTR(-EBUSY
);
2327 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2329 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2330 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2332 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2334 __d_move(alias
, dentry
);
2337 spin_unlock(&inode
->i_lock
);
2346 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2347 * named dentry in place of the dentry to be replaced.
2348 * returns with anon->d_lock held!
2350 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2352 struct dentry
*dparent
, *aparent
;
2354 dentry_lock_for_move(anon
, dentry
);
2356 write_seqcount_begin(&dentry
->d_seq
);
2357 write_seqcount_begin(&anon
->d_seq
);
2359 dparent
= dentry
->d_parent
;
2360 aparent
= anon
->d_parent
;
2362 switch_names(dentry
, anon
);
2363 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2365 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2366 list_del(&dentry
->d_u
.d_child
);
2367 if (!IS_ROOT(dentry
))
2368 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2370 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2372 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2373 list_del(&anon
->d_u
.d_child
);
2375 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2377 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2379 write_seqcount_end(&dentry
->d_seq
);
2380 write_seqcount_end(&anon
->d_seq
);
2382 dentry_unlock_parents_for_move(anon
, dentry
);
2383 spin_unlock(&dentry
->d_lock
);
2385 /* anon->d_lock still locked, returns locked */
2386 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2390 * d_materialise_unique - introduce an inode into the tree
2391 * @dentry: candidate dentry
2392 * @inode: inode to bind to the dentry, to which aliases may be attached
2394 * Introduces an dentry into the tree, substituting an extant disconnected
2395 * root directory alias in its place if there is one. Caller must hold the
2396 * i_mutex of the parent directory.
2398 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2400 struct dentry
*actual
;
2402 BUG_ON(!d_unhashed(dentry
));
2406 __d_instantiate(dentry
, NULL
);
2411 spin_lock(&inode
->i_lock
);
2413 if (S_ISDIR(inode
->i_mode
)) {
2414 struct dentry
*alias
;
2416 /* Does an aliased dentry already exist? */
2417 alias
= __d_find_alias(inode
, 0);
2420 write_seqlock(&rename_lock
);
2422 if (d_ancestor(alias
, dentry
)) {
2423 /* Check for loops */
2424 actual
= ERR_PTR(-ELOOP
);
2425 spin_unlock(&inode
->i_lock
);
2426 } else if (IS_ROOT(alias
)) {
2427 /* Is this an anonymous mountpoint that we
2428 * could splice into our tree? */
2429 __d_materialise_dentry(dentry
, alias
);
2430 write_sequnlock(&rename_lock
);
2434 /* Nope, but we must(!) avoid directory
2435 * aliasing. This drops inode->i_lock */
2436 actual
= __d_unalias(inode
, dentry
, alias
);
2438 write_sequnlock(&rename_lock
);
2439 if (IS_ERR(actual
)) {
2440 if (PTR_ERR(actual
) == -ELOOP
)
2441 pr_warn_ratelimited(
2442 "VFS: Lookup of '%s' in %s %s"
2443 " would have caused loop\n",
2444 dentry
->d_name
.name
,
2445 inode
->i_sb
->s_type
->name
,
2453 /* Add a unique reference */
2454 actual
= __d_instantiate_unique(dentry
, inode
);
2458 BUG_ON(!d_unhashed(actual
));
2460 spin_lock(&actual
->d_lock
);
2463 spin_unlock(&actual
->d_lock
);
2464 spin_unlock(&inode
->i_lock
);
2466 if (actual
== dentry
) {
2467 security_d_instantiate(dentry
, inode
);
2474 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2476 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2480 return -ENAMETOOLONG
;
2482 memcpy(*buffer
, str
, namelen
);
2486 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2488 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2492 * prepend_path - Prepend path string to a buffer
2493 * @path: the dentry/vfsmount to report
2494 * @root: root vfsmnt/dentry
2495 * @buffer: pointer to the end of the buffer
2496 * @buflen: pointer to buffer length
2498 * Caller holds the rename_lock.
2500 static int prepend_path(const struct path
*path
,
2501 const struct path
*root
,
2502 char **buffer
, int *buflen
)
2504 struct dentry
*dentry
= path
->dentry
;
2505 struct vfsmount
*vfsmnt
= path
->mnt
;
2506 struct mount
*mnt
= real_mount(vfsmnt
);
2510 br_read_lock(vfsmount_lock
);
2511 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2512 struct dentry
* parent
;
2514 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2516 if (!mnt_has_parent(mnt
))
2518 dentry
= mnt
->mnt_mountpoint
;
2519 mnt
= mnt
->mnt_parent
;
2523 parent
= dentry
->d_parent
;
2525 spin_lock(&dentry
->d_lock
);
2526 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2527 spin_unlock(&dentry
->d_lock
);
2529 error
= prepend(buffer
, buflen
, "/", 1);
2537 if (!error
&& !slash
)
2538 error
= prepend(buffer
, buflen
, "/", 1);
2541 br_read_unlock(vfsmount_lock
);
2546 * Filesystems needing to implement special "root names"
2547 * should do so with ->d_dname()
2549 if (IS_ROOT(dentry
) &&
2550 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2551 WARN(1, "Root dentry has weird name <%.*s>\n",
2552 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2555 error
= prepend(buffer
, buflen
, "/", 1);
2557 error
= real_mount(vfsmnt
)->mnt_ns
? 1 : 2;
2562 * __d_path - return the path of a dentry
2563 * @path: the dentry/vfsmount to report
2564 * @root: root vfsmnt/dentry
2565 * @buf: buffer to return value in
2566 * @buflen: buffer length
2568 * Convert a dentry into an ASCII path name.
2570 * Returns a pointer into the buffer or an error code if the
2571 * path was too long.
2573 * "buflen" should be positive.
2575 * If the path is not reachable from the supplied root, return %NULL.
2577 char *__d_path(const struct path
*path
,
2578 const struct path
*root
,
2579 char *buf
, int buflen
)
2581 char *res
= buf
+ buflen
;
2584 prepend(&res
, &buflen
, "\0", 1);
2585 write_seqlock(&rename_lock
);
2586 error
= prepend_path(path
, root
, &res
, &buflen
);
2587 write_sequnlock(&rename_lock
);
2590 return ERR_PTR(error
);
2596 char *d_absolute_path(const struct path
*path
,
2597 char *buf
, int buflen
)
2599 struct path root
= {};
2600 char *res
= buf
+ buflen
;
2603 prepend(&res
, &buflen
, "\0", 1);
2604 write_seqlock(&rename_lock
);
2605 error
= prepend_path(path
, &root
, &res
, &buflen
);
2606 write_sequnlock(&rename_lock
);
2611 return ERR_PTR(error
);
2616 * same as __d_path but appends "(deleted)" for unlinked files.
2618 static int path_with_deleted(const struct path
*path
,
2619 const struct path
*root
,
2620 char **buf
, int *buflen
)
2622 prepend(buf
, buflen
, "\0", 1);
2623 if (d_unlinked(path
->dentry
)) {
2624 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2629 return prepend_path(path
, root
, buf
, buflen
);
2632 static int prepend_unreachable(char **buffer
, int *buflen
)
2634 return prepend(buffer
, buflen
, "(unreachable)", 13);
2638 * d_path - return the path of a dentry
2639 * @path: path to report
2640 * @buf: buffer to return value in
2641 * @buflen: buffer length
2643 * Convert a dentry into an ASCII path name. If the entry has been deleted
2644 * the string " (deleted)" is appended. Note that this is ambiguous.
2646 * Returns a pointer into the buffer or an error code if the path was
2647 * too long. Note: Callers should use the returned pointer, not the passed
2648 * in buffer, to use the name! The implementation often starts at an offset
2649 * into the buffer, and may leave 0 bytes at the start.
2651 * "buflen" should be positive.
2653 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2655 char *res
= buf
+ buflen
;
2660 * We have various synthetic filesystems that never get mounted. On
2661 * these filesystems dentries are never used for lookup purposes, and
2662 * thus don't need to be hashed. They also don't need a name until a
2663 * user wants to identify the object in /proc/pid/fd/. The little hack
2664 * below allows us to generate a name for these objects on demand:
2666 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2667 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2669 get_fs_root(current
->fs
, &root
);
2670 write_seqlock(&rename_lock
);
2671 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2673 res
= ERR_PTR(error
);
2674 write_sequnlock(&rename_lock
);
2678 EXPORT_SYMBOL(d_path
);
2681 * d_path_with_unreachable - return the path of a dentry
2682 * @path: path to report
2683 * @buf: buffer to return value in
2684 * @buflen: buffer length
2686 * The difference from d_path() is that this prepends "(unreachable)"
2687 * to paths which are unreachable from the current process' root.
2689 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2691 char *res
= buf
+ buflen
;
2695 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2696 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2698 get_fs_root(current
->fs
, &root
);
2699 write_seqlock(&rename_lock
);
2700 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2702 error
= prepend_unreachable(&res
, &buflen
);
2703 write_sequnlock(&rename_lock
);
2706 res
= ERR_PTR(error
);
2712 * Helper function for dentry_operations.d_dname() members
2714 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2715 const char *fmt
, ...)
2721 va_start(args
, fmt
);
2722 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2725 if (sz
> sizeof(temp
) || sz
> buflen
)
2726 return ERR_PTR(-ENAMETOOLONG
);
2728 buffer
+= buflen
- sz
;
2729 return memcpy(buffer
, temp
, sz
);
2733 * Write full pathname from the root of the filesystem into the buffer.
2735 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2737 char *end
= buf
+ buflen
;
2740 prepend(&end
, &buflen
, "\0", 1);
2747 while (!IS_ROOT(dentry
)) {
2748 struct dentry
*parent
= dentry
->d_parent
;
2752 spin_lock(&dentry
->d_lock
);
2753 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2754 spin_unlock(&dentry
->d_lock
);
2755 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2763 return ERR_PTR(-ENAMETOOLONG
);
2766 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2770 write_seqlock(&rename_lock
);
2771 retval
= __dentry_path(dentry
, buf
, buflen
);
2772 write_sequnlock(&rename_lock
);
2776 EXPORT_SYMBOL(dentry_path_raw
);
2778 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2783 write_seqlock(&rename_lock
);
2784 if (d_unlinked(dentry
)) {
2786 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2790 retval
= __dentry_path(dentry
, buf
, buflen
);
2791 write_sequnlock(&rename_lock
);
2792 if (!IS_ERR(retval
) && p
)
2793 *p
= '/'; /* restore '/' overriden with '\0' */
2796 return ERR_PTR(-ENAMETOOLONG
);
2800 * NOTE! The user-level library version returns a
2801 * character pointer. The kernel system call just
2802 * returns the length of the buffer filled (which
2803 * includes the ending '\0' character), or a negative
2804 * error value. So libc would do something like
2806 * char *getcwd(char * buf, size_t size)
2810 * retval = sys_getcwd(buf, size);
2817 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2820 struct path pwd
, root
;
2821 char *page
= (char *) __get_free_page(GFP_USER
);
2826 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2829 write_seqlock(&rename_lock
);
2830 if (!d_unlinked(pwd
.dentry
)) {
2832 char *cwd
= page
+ PAGE_SIZE
;
2833 int buflen
= PAGE_SIZE
;
2835 prepend(&cwd
, &buflen
, "\0", 1);
2836 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2837 write_sequnlock(&rename_lock
);
2842 /* Unreachable from current root */
2844 error
= prepend_unreachable(&cwd
, &buflen
);
2850 len
= PAGE_SIZE
+ page
- cwd
;
2853 if (copy_to_user(buf
, cwd
, len
))
2857 write_sequnlock(&rename_lock
);
2863 free_page((unsigned long) page
);
2868 * Test whether new_dentry is a subdirectory of old_dentry.
2870 * Trivially implemented using the dcache structure
2874 * is_subdir - is new dentry a subdirectory of old_dentry
2875 * @new_dentry: new dentry
2876 * @old_dentry: old dentry
2878 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2879 * Returns 0 otherwise.
2880 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2883 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2888 if (new_dentry
== old_dentry
)
2892 /* for restarting inner loop in case of seq retry */
2893 seq
= read_seqbegin(&rename_lock
);
2895 * Need rcu_readlock to protect against the d_parent trashing
2899 if (d_ancestor(old_dentry
, new_dentry
))
2904 } while (read_seqretry(&rename_lock
, seq
));
2909 void d_genocide(struct dentry
*root
)
2911 struct dentry
*this_parent
;
2912 struct list_head
*next
;
2916 seq
= read_seqbegin(&rename_lock
);
2919 spin_lock(&this_parent
->d_lock
);
2921 next
= this_parent
->d_subdirs
.next
;
2923 while (next
!= &this_parent
->d_subdirs
) {
2924 struct list_head
*tmp
= next
;
2925 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2928 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2929 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2930 spin_unlock(&dentry
->d_lock
);
2933 if (!list_empty(&dentry
->d_subdirs
)) {
2934 spin_unlock(&this_parent
->d_lock
);
2935 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2936 this_parent
= dentry
;
2937 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2940 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2941 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2944 spin_unlock(&dentry
->d_lock
);
2946 if (this_parent
!= root
) {
2947 struct dentry
*child
= this_parent
;
2948 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2949 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2950 this_parent
->d_count
--;
2952 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2955 next
= child
->d_u
.d_child
.next
;
2958 spin_unlock(&this_parent
->d_lock
);
2959 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2962 write_sequnlock(&rename_lock
);
2967 write_seqlock(&rename_lock
);
2972 * find_inode_number - check for dentry with name
2973 * @dir: directory to check
2974 * @name: Name to find.
2976 * Check whether a dentry already exists for the given name,
2977 * and return the inode number if it has an inode. Otherwise
2980 * This routine is used to post-process directory listings for
2981 * filesystems using synthetic inode numbers, and is necessary
2982 * to keep getcwd() working.
2985 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2987 struct dentry
* dentry
;
2990 dentry
= d_hash_and_lookup(dir
, name
);
2992 if (dentry
->d_inode
)
2993 ino
= dentry
->d_inode
->i_ino
;
2998 EXPORT_SYMBOL(find_inode_number
);
3000 static __initdata
unsigned long dhash_entries
;
3001 static int __init
set_dhash_entries(char *str
)
3005 dhash_entries
= simple_strtoul(str
, &str
, 0);
3008 __setup("dhash_entries=", set_dhash_entries
);
3010 static void __init
dcache_init_early(void)
3014 /* If hashes are distributed across NUMA nodes, defer
3015 * hash allocation until vmalloc space is available.
3021 alloc_large_system_hash("Dentry cache",
3022 sizeof(struct hlist_bl_head
),
3030 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3031 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3034 static void __init
dcache_init(void)
3039 * A constructor could be added for stable state like the lists,
3040 * but it is probably not worth it because of the cache nature
3043 dentry_cache
= KMEM_CACHE(dentry
,
3044 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3046 /* Hash may have been set up in dcache_init_early */
3051 alloc_large_system_hash("Dentry cache",
3052 sizeof(struct hlist_bl_head
),
3060 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3061 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3064 /* SLAB cache for __getname() consumers */
3065 struct kmem_cache
*names_cachep __read_mostly
;
3066 EXPORT_SYMBOL(names_cachep
);
3068 EXPORT_SYMBOL(d_genocide
);
3070 void __init
vfs_caches_init_early(void)
3072 dcache_init_early();
3076 void __init
vfs_caches_init(unsigned long mempages
)
3078 unsigned long reserve
;
3080 /* Base hash sizes on available memory, with a reserve equal to
3081 150% of current kernel size */
3083 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3084 mempages
-= reserve
;
3086 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3087 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3091 files_init(mempages
);