4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
45 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
47 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
51 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
59 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
64 * dentry->d_inode->i_lock
67 * dcache_hash_bucket lock
70 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
73 * dentry->d_parent->d_lock
76 * If no ancestor relationship:
77 * if (dentry1 < dentry2)
81 int sysctl_vfs_cache_pressure __read_mostly
= 100;
82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
84 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
85 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
87 EXPORT_SYMBOL(rename_lock
);
89 static struct kmem_cache
*dentry_cache __read_mostly
;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
99 #define D_HASHBITS d_hash_shift
100 #define D_HASHMASK d_hash_mask
102 static unsigned int d_hash_mask __read_mostly
;
103 static unsigned int d_hash_shift __read_mostly
;
105 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
107 static inline struct hlist_bl_head
*d_hash(struct dentry
*parent
,
110 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
111 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
112 return dentry_hashtable
+ (hash
& D_HASHMASK
);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat
= {
120 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 static int get_nr_dentry(void)
127 for_each_possible_cpu(i
)
128 sum
+= per_cpu(nr_dentry
, i
);
129 return sum
< 0 ? 0 : sum
;
132 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
133 size_t *lenp
, loff_t
*ppos
)
135 dentry_stat
.nr_dentry
= get_nr_dentry();
136 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
140 static void __d_free(struct rcu_head
*head
)
142 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
144 WARN_ON(!list_empty(&dentry
->d_alias
));
145 if (dname_external(dentry
))
146 kfree(dentry
->d_name
.name
);
147 kmem_cache_free(dentry_cache
, dentry
);
153 static void d_free(struct dentry
*dentry
)
155 BUG_ON(dentry
->d_count
);
156 this_cpu_dec(nr_dentry
);
157 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
158 dentry
->d_op
->d_release(dentry
);
160 /* if dentry was never visible to RCU, immediate free is OK */
161 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
162 __d_free(&dentry
->d_u
.d_rcu
);
164 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
168 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
169 * @dentry: the target dentry
170 * After this call, in-progress rcu-walk path lookup will fail. This
171 * should be called after unhashing, and after changing d_inode (if
172 * the dentry has not already been unhashed).
174 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
176 assert_spin_locked(&dentry
->d_lock
);
177 /* Go through a barrier */
178 write_seqcount_barrier(&dentry
->d_seq
);
182 * Release the dentry's inode, using the filesystem
183 * d_iput() operation if defined. Dentry has no refcount
186 static void dentry_iput(struct dentry
* dentry
)
187 __releases(dentry
->d_lock
)
188 __releases(dentry
->d_inode
->i_lock
)
190 struct inode
*inode
= dentry
->d_inode
;
192 dentry
->d_inode
= NULL
;
193 list_del_init(&dentry
->d_alias
);
194 spin_unlock(&dentry
->d_lock
);
195 spin_unlock(&inode
->i_lock
);
197 fsnotify_inoderemove(inode
);
198 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
199 dentry
->d_op
->d_iput(dentry
, inode
);
203 spin_unlock(&dentry
->d_lock
);
208 * Release the dentry's inode, using the filesystem
209 * d_iput() operation if defined. dentry remains in-use.
211 static void dentry_unlink_inode(struct dentry
* dentry
)
212 __releases(dentry
->d_lock
)
213 __releases(dentry
->d_inode
->i_lock
)
215 struct inode
*inode
= dentry
->d_inode
;
216 dentry
->d_inode
= NULL
;
217 list_del_init(&dentry
->d_alias
);
218 dentry_rcuwalk_barrier(dentry
);
219 spin_unlock(&dentry
->d_lock
);
220 spin_unlock(&inode
->i_lock
);
222 fsnotify_inoderemove(inode
);
223 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
224 dentry
->d_op
->d_iput(dentry
, inode
);
230 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
232 static void dentry_lru_add(struct dentry
*dentry
)
234 if (list_empty(&dentry
->d_lru
)) {
235 spin_lock(&dcache_lru_lock
);
236 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
237 dentry
->d_sb
->s_nr_dentry_unused
++;
238 dentry_stat
.nr_unused
++;
239 spin_unlock(&dcache_lru_lock
);
243 static void __dentry_lru_del(struct dentry
*dentry
)
245 list_del_init(&dentry
->d_lru
);
246 dentry
->d_sb
->s_nr_dentry_unused
--;
247 dentry_stat
.nr_unused
--;
251 * Remove a dentry with references from the LRU.
253 static void dentry_lru_del(struct dentry
*dentry
)
255 if (!list_empty(&dentry
->d_lru
)) {
256 spin_lock(&dcache_lru_lock
);
257 __dentry_lru_del(dentry
);
258 spin_unlock(&dcache_lru_lock
);
263 * Remove a dentry that is unreferenced and about to be pruned
264 * (unhashed and destroyed) from the LRU, and inform the file system.
265 * This wrapper should be called _prior_ to unhashing a victim dentry.
267 static void dentry_lru_prune(struct dentry
*dentry
)
269 if (!list_empty(&dentry
->d_lru
)) {
270 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
271 dentry
->d_op
->d_prune(dentry
);
273 spin_lock(&dcache_lru_lock
);
274 __dentry_lru_del(dentry
);
275 spin_unlock(&dcache_lru_lock
);
279 static void dentry_lru_move_tail(struct dentry
*dentry
)
281 spin_lock(&dcache_lru_lock
);
282 if (list_empty(&dentry
->d_lru
)) {
283 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
284 dentry
->d_sb
->s_nr_dentry_unused
++;
285 dentry_stat
.nr_unused
++;
287 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
289 spin_unlock(&dcache_lru_lock
);
293 * d_kill - kill dentry and return parent
294 * @dentry: dentry to kill
295 * @parent: parent dentry
297 * The dentry must already be unhashed and removed from the LRU.
299 * If this is the root of the dentry tree, return NULL.
301 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
304 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
305 __releases(dentry
->d_lock
)
306 __releases(parent
->d_lock
)
307 __releases(dentry
->d_inode
->i_lock
)
309 list_del(&dentry
->d_u
.d_child
);
311 * Inform try_to_ascend() that we are no longer attached to the
314 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
316 spin_unlock(&parent
->d_lock
);
319 * dentry_iput drops the locks, at which point nobody (except
320 * transient RCU lookups) can reach this dentry.
327 * Unhash a dentry without inserting an RCU walk barrier or checking that
328 * dentry->d_lock is locked. The caller must take care of that, if
331 static void __d_shrink(struct dentry
*dentry
)
333 if (!d_unhashed(dentry
)) {
334 struct hlist_bl_head
*b
;
335 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
336 b
= &dentry
->d_sb
->s_anon
;
338 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
341 __hlist_bl_del(&dentry
->d_hash
);
342 dentry
->d_hash
.pprev
= NULL
;
348 * d_drop - drop a dentry
349 * @dentry: dentry to drop
351 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
352 * be found through a VFS lookup any more. Note that this is different from
353 * deleting the dentry - d_delete will try to mark the dentry negative if
354 * possible, giving a successful _negative_ lookup, while d_drop will
355 * just make the cache lookup fail.
357 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
358 * reason (NFS timeouts or autofs deletes).
360 * __d_drop requires dentry->d_lock.
362 void __d_drop(struct dentry
*dentry
)
364 if (!d_unhashed(dentry
)) {
366 dentry_rcuwalk_barrier(dentry
);
369 EXPORT_SYMBOL(__d_drop
);
371 void d_drop(struct dentry
*dentry
)
373 spin_lock(&dentry
->d_lock
);
375 spin_unlock(&dentry
->d_lock
);
377 EXPORT_SYMBOL(d_drop
);
380 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
381 * @dentry: dentry to drop
383 * This is called when we do a lookup on a placeholder dentry that needed to be
384 * looked up. The dentry should have been hashed in order for it to be found by
385 * the lookup code, but now needs to be unhashed while we do the actual lookup
386 * and clear the DCACHE_NEED_LOOKUP flag.
388 void d_clear_need_lookup(struct dentry
*dentry
)
390 spin_lock(&dentry
->d_lock
);
392 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
393 spin_unlock(&dentry
->d_lock
);
395 EXPORT_SYMBOL(d_clear_need_lookup
);
398 * Finish off a dentry we've decided to kill.
399 * dentry->d_lock must be held, returns with it unlocked.
400 * If ref is non-zero, then decrement the refcount too.
401 * Returns dentry requiring refcount drop, or NULL if we're done.
403 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
404 __releases(dentry
->d_lock
)
407 struct dentry
*parent
;
409 inode
= dentry
->d_inode
;
410 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
412 spin_unlock(&dentry
->d_lock
);
414 return dentry
; /* try again with same dentry */
419 parent
= dentry
->d_parent
;
420 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
422 spin_unlock(&inode
->i_lock
);
429 * if dentry was on the d_lru list delete it from there.
430 * inform the fs via d_prune that this dentry is about to be
431 * unhashed and destroyed.
433 dentry_lru_prune(dentry
);
434 /* if it was on the hash then remove it */
436 return d_kill(dentry
, parent
);
442 * This is complicated by the fact that we do not want to put
443 * dentries that are no longer on any hash chain on the unused
444 * list: we'd much rather just get rid of them immediately.
446 * However, that implies that we have to traverse the dentry
447 * tree upwards to the parents which might _also_ now be
448 * scheduled for deletion (it may have been only waiting for
449 * its last child to go away).
451 * This tail recursion is done by hand as we don't want to depend
452 * on the compiler to always get this right (gcc generally doesn't).
453 * Real recursion would eat up our stack space.
457 * dput - release a dentry
458 * @dentry: dentry to release
460 * Release a dentry. This will drop the usage count and if appropriate
461 * call the dentry unlink method as well as removing it from the queues and
462 * releasing its resources. If the parent dentries were scheduled for release
463 * they too may now get deleted.
465 void dput(struct dentry
*dentry
)
471 if (dentry
->d_count
== 1)
473 spin_lock(&dentry
->d_lock
);
474 BUG_ON(!dentry
->d_count
);
475 if (dentry
->d_count
> 1) {
477 spin_unlock(&dentry
->d_lock
);
481 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
482 if (dentry
->d_op
->d_delete(dentry
))
486 /* Unreachable? Get rid of it */
487 if (d_unhashed(dentry
))
491 * If this dentry needs lookup, don't set the referenced flag so that it
492 * is more likely to be cleaned up by the dcache shrinker in case of
495 if (!d_need_lookup(dentry
))
496 dentry
->d_flags
|= DCACHE_REFERENCED
;
497 dentry_lru_add(dentry
);
500 spin_unlock(&dentry
->d_lock
);
504 dentry
= dentry_kill(dentry
, 1);
511 * d_invalidate - invalidate a dentry
512 * @dentry: dentry to invalidate
514 * Try to invalidate the dentry if it turns out to be
515 * possible. If there are other dentries that can be
516 * reached through this one we can't delete it and we
517 * return -EBUSY. On success we return 0.
522 int d_invalidate(struct dentry
* dentry
)
525 * If it's already been dropped, return OK.
527 spin_lock(&dentry
->d_lock
);
528 if (d_unhashed(dentry
)) {
529 spin_unlock(&dentry
->d_lock
);
533 * Check whether to do a partial shrink_dcache
534 * to get rid of unused child entries.
536 if (!list_empty(&dentry
->d_subdirs
)) {
537 spin_unlock(&dentry
->d_lock
);
538 shrink_dcache_parent(dentry
);
539 spin_lock(&dentry
->d_lock
);
543 * Somebody else still using it?
545 * If it's a directory, we can't drop it
546 * for fear of somebody re-populating it
547 * with children (even though dropping it
548 * would make it unreachable from the root,
549 * we might still populate it if it was a
550 * working directory or similar).
551 * We also need to leave mountpoints alone,
554 if (dentry
->d_count
> 1 && dentry
->d_inode
) {
555 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
556 spin_unlock(&dentry
->d_lock
);
562 spin_unlock(&dentry
->d_lock
);
565 EXPORT_SYMBOL(d_invalidate
);
567 /* This must be called with d_lock held */
568 static inline void __dget_dlock(struct dentry
*dentry
)
573 static inline void __dget(struct dentry
*dentry
)
575 spin_lock(&dentry
->d_lock
);
576 __dget_dlock(dentry
);
577 spin_unlock(&dentry
->d_lock
);
580 struct dentry
*dget_parent(struct dentry
*dentry
)
586 * Don't need rcu_dereference because we re-check it was correct under
590 ret
= dentry
->d_parent
;
591 spin_lock(&ret
->d_lock
);
592 if (unlikely(ret
!= dentry
->d_parent
)) {
593 spin_unlock(&ret
->d_lock
);
598 BUG_ON(!ret
->d_count
);
600 spin_unlock(&ret
->d_lock
);
603 EXPORT_SYMBOL(dget_parent
);
606 * d_find_alias - grab a hashed alias of inode
607 * @inode: inode in question
608 * @want_discon: flag, used by d_splice_alias, to request
609 * that only a DISCONNECTED alias be returned.
611 * If inode has a hashed alias, or is a directory and has any alias,
612 * acquire the reference to alias and return it. Otherwise return NULL.
613 * Notice that if inode is a directory there can be only one alias and
614 * it can be unhashed only if it has no children, or if it is the root
617 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
618 * any other hashed alias over that one unless @want_discon is set,
619 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
621 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
623 struct dentry
*alias
, *discon_alias
;
627 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
628 spin_lock(&alias
->d_lock
);
629 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
630 if (IS_ROOT(alias
) &&
631 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
632 discon_alias
= alias
;
633 } else if (!want_discon
) {
635 spin_unlock(&alias
->d_lock
);
639 spin_unlock(&alias
->d_lock
);
642 alias
= discon_alias
;
643 spin_lock(&alias
->d_lock
);
644 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
645 if (IS_ROOT(alias
) &&
646 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
648 spin_unlock(&alias
->d_lock
);
652 spin_unlock(&alias
->d_lock
);
658 struct dentry
*d_find_alias(struct inode
*inode
)
660 struct dentry
*de
= NULL
;
662 if (!list_empty(&inode
->i_dentry
)) {
663 spin_lock(&inode
->i_lock
);
664 de
= __d_find_alias(inode
, 0);
665 spin_unlock(&inode
->i_lock
);
669 EXPORT_SYMBOL(d_find_alias
);
672 * Try to kill dentries associated with this inode.
673 * WARNING: you must own a reference to inode.
675 void d_prune_aliases(struct inode
*inode
)
677 struct dentry
*dentry
;
679 spin_lock(&inode
->i_lock
);
680 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
681 spin_lock(&dentry
->d_lock
);
682 if (!dentry
->d_count
) {
683 __dget_dlock(dentry
);
685 spin_unlock(&dentry
->d_lock
);
686 spin_unlock(&inode
->i_lock
);
690 spin_unlock(&dentry
->d_lock
);
692 spin_unlock(&inode
->i_lock
);
694 EXPORT_SYMBOL(d_prune_aliases
);
697 * Try to throw away a dentry - free the inode, dput the parent.
698 * Requires dentry->d_lock is held, and dentry->d_count == 0.
699 * Releases dentry->d_lock.
701 * This may fail if locks cannot be acquired no problem, just try again.
703 static void try_prune_one_dentry(struct dentry
*dentry
)
704 __releases(dentry
->d_lock
)
706 struct dentry
*parent
;
708 parent
= dentry_kill(dentry
, 0);
710 * If dentry_kill returns NULL, we have nothing more to do.
711 * if it returns the same dentry, trylocks failed. In either
712 * case, just loop again.
714 * Otherwise, we need to prune ancestors too. This is necessary
715 * to prevent quadratic behavior of shrink_dcache_parent(), but
716 * is also expected to be beneficial in reducing dentry cache
721 if (parent
== dentry
)
724 /* Prune ancestors. */
727 spin_lock(&dentry
->d_lock
);
728 if (dentry
->d_count
> 1) {
730 spin_unlock(&dentry
->d_lock
);
733 dentry
= dentry_kill(dentry
, 1);
737 static void shrink_dentry_list(struct list_head
*list
)
739 struct dentry
*dentry
;
743 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
744 if (&dentry
->d_lru
== list
)
746 spin_lock(&dentry
->d_lock
);
747 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
748 spin_unlock(&dentry
->d_lock
);
753 * We found an inuse dentry which was not removed from
754 * the LRU because of laziness during lookup. Do not free
755 * it - just keep it off the LRU list.
757 if (dentry
->d_count
) {
758 dentry_lru_del(dentry
);
759 spin_unlock(&dentry
->d_lock
);
765 try_prune_one_dentry(dentry
);
773 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
774 * @sb: superblock to shrink dentry LRU.
775 * @count: number of entries to prune
776 * @flags: flags to control the dentry processing
778 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
780 static void __shrink_dcache_sb(struct super_block
*sb
, int count
, int flags
)
782 struct dentry
*dentry
;
783 LIST_HEAD(referenced
);
787 spin_lock(&dcache_lru_lock
);
788 while (!list_empty(&sb
->s_dentry_lru
)) {
789 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
790 struct dentry
, d_lru
);
791 BUG_ON(dentry
->d_sb
!= sb
);
793 if (!spin_trylock(&dentry
->d_lock
)) {
794 spin_unlock(&dcache_lru_lock
);
800 * If we are honouring the DCACHE_REFERENCED flag and the
801 * dentry has this flag set, don't free it. Clear the flag
802 * and put it back on the LRU.
804 if (flags
& DCACHE_REFERENCED
&&
805 dentry
->d_flags
& DCACHE_REFERENCED
) {
806 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
807 list_move(&dentry
->d_lru
, &referenced
);
808 spin_unlock(&dentry
->d_lock
);
810 list_move_tail(&dentry
->d_lru
, &tmp
);
811 spin_unlock(&dentry
->d_lock
);
815 cond_resched_lock(&dcache_lru_lock
);
817 if (!list_empty(&referenced
))
818 list_splice(&referenced
, &sb
->s_dentry_lru
);
819 spin_unlock(&dcache_lru_lock
);
821 shrink_dentry_list(&tmp
);
825 * prune_dcache_sb - shrink the dcache
827 * @nr_to_scan: number of entries to try to free
829 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
830 * done when we need more memory an called from the superblock shrinker
833 * This function may fail to free any resources if all the dentries are in
836 void prune_dcache_sb(struct super_block
*sb
, int nr_to_scan
)
838 __shrink_dcache_sb(sb
, nr_to_scan
, DCACHE_REFERENCED
);
842 * shrink_dcache_sb - shrink dcache for a superblock
845 * Shrink the dcache for the specified super block. This is used to free
846 * the dcache before unmounting a file system.
848 void shrink_dcache_sb(struct super_block
*sb
)
852 spin_lock(&dcache_lru_lock
);
853 while (!list_empty(&sb
->s_dentry_lru
)) {
854 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
855 spin_unlock(&dcache_lru_lock
);
856 shrink_dentry_list(&tmp
);
857 spin_lock(&dcache_lru_lock
);
859 spin_unlock(&dcache_lru_lock
);
861 EXPORT_SYMBOL(shrink_dcache_sb
);
864 * destroy a single subtree of dentries for unmount
865 * - see the comments on shrink_dcache_for_umount() for a description of the
868 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
870 struct dentry
*parent
;
872 BUG_ON(!IS_ROOT(dentry
));
875 /* descend to the first leaf in the current subtree */
876 while (!list_empty(&dentry
->d_subdirs
))
877 dentry
= list_entry(dentry
->d_subdirs
.next
,
878 struct dentry
, d_u
.d_child
);
880 /* consume the dentries from this leaf up through its parents
881 * until we find one with children or run out altogether */
886 * remove the dentry from the lru, and inform
887 * the fs that this dentry is about to be
888 * unhashed and destroyed.
890 dentry_lru_prune(dentry
);
893 if (dentry
->d_count
!= 0) {
895 "BUG: Dentry %p{i=%lx,n=%s}"
897 " [unmount of %s %s]\n",
900 dentry
->d_inode
->i_ino
: 0UL,
903 dentry
->d_sb
->s_type
->name
,
908 if (IS_ROOT(dentry
)) {
910 list_del(&dentry
->d_u
.d_child
);
912 parent
= dentry
->d_parent
;
914 list_del(&dentry
->d_u
.d_child
);
917 inode
= dentry
->d_inode
;
919 dentry
->d_inode
= NULL
;
920 list_del_init(&dentry
->d_alias
);
921 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
922 dentry
->d_op
->d_iput(dentry
, inode
);
929 /* finished when we fall off the top of the tree,
930 * otherwise we ascend to the parent and move to the
931 * next sibling if there is one */
935 } while (list_empty(&dentry
->d_subdirs
));
937 dentry
= list_entry(dentry
->d_subdirs
.next
,
938 struct dentry
, d_u
.d_child
);
943 * destroy the dentries attached to a superblock on unmounting
944 * - we don't need to use dentry->d_lock because:
945 * - the superblock is detached from all mountings and open files, so the
946 * dentry trees will not be rearranged by the VFS
947 * - s_umount is write-locked, so the memory pressure shrinker will ignore
948 * any dentries belonging to this superblock that it comes across
949 * - the filesystem itself is no longer permitted to rearrange the dentries
952 void shrink_dcache_for_umount(struct super_block
*sb
)
954 struct dentry
*dentry
;
956 if (down_read_trylock(&sb
->s_umount
))
962 shrink_dcache_for_umount_subtree(dentry
);
964 while (!hlist_bl_empty(&sb
->s_anon
)) {
965 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
966 shrink_dcache_for_umount_subtree(dentry
);
971 * This tries to ascend one level of parenthood, but
972 * we can race with renaming, so we need to re-check
973 * the parenthood after dropping the lock and check
974 * that the sequence number still matches.
976 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
978 struct dentry
*new = old
->d_parent
;
981 spin_unlock(&old
->d_lock
);
982 spin_lock(&new->d_lock
);
985 * might go back up the wrong parent if we have had a rename
988 if (new != old
->d_parent
||
989 (old
->d_flags
& DCACHE_DISCONNECTED
) ||
990 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
991 spin_unlock(&new->d_lock
);
1000 * Search for at least 1 mount point in the dentry's subdirs.
1001 * We descend to the next level whenever the d_subdirs
1002 * list is non-empty and continue searching.
1006 * have_submounts - check for mounts over a dentry
1007 * @parent: dentry to check.
1009 * Return true if the parent or its subdirectories contain
1012 int have_submounts(struct dentry
*parent
)
1014 struct dentry
*this_parent
;
1015 struct list_head
*next
;
1019 seq
= read_seqbegin(&rename_lock
);
1021 this_parent
= parent
;
1023 if (d_mountpoint(parent
))
1025 spin_lock(&this_parent
->d_lock
);
1027 next
= this_parent
->d_subdirs
.next
;
1029 while (next
!= &this_parent
->d_subdirs
) {
1030 struct list_head
*tmp
= next
;
1031 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1034 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1035 /* Have we found a mount point ? */
1036 if (d_mountpoint(dentry
)) {
1037 spin_unlock(&dentry
->d_lock
);
1038 spin_unlock(&this_parent
->d_lock
);
1041 if (!list_empty(&dentry
->d_subdirs
)) {
1042 spin_unlock(&this_parent
->d_lock
);
1043 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1044 this_parent
= dentry
;
1045 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1048 spin_unlock(&dentry
->d_lock
);
1051 * All done at this level ... ascend and resume the search.
1053 if (this_parent
!= parent
) {
1054 struct dentry
*child
= this_parent
;
1055 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1058 next
= child
->d_u
.d_child
.next
;
1061 spin_unlock(&this_parent
->d_lock
);
1062 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1065 write_sequnlock(&rename_lock
);
1066 return 0; /* No mount points found in tree */
1068 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1071 write_sequnlock(&rename_lock
);
1076 write_seqlock(&rename_lock
);
1079 EXPORT_SYMBOL(have_submounts
);
1082 * Search the dentry child list for the specified parent,
1083 * and move any unused dentries to the end of the unused
1084 * list for prune_dcache(). We descend to the next level
1085 * whenever the d_subdirs list is non-empty and continue
1088 * It returns zero iff there are no unused children,
1089 * otherwise it returns the number of children moved to
1090 * the end of the unused list. This may not be the total
1091 * number of unused children, because select_parent can
1092 * drop the lock and return early due to latency
1095 static int select_parent(struct dentry
* parent
)
1097 struct dentry
*this_parent
;
1098 struct list_head
*next
;
1103 seq
= read_seqbegin(&rename_lock
);
1105 this_parent
= parent
;
1106 spin_lock(&this_parent
->d_lock
);
1108 next
= this_parent
->d_subdirs
.next
;
1110 while (next
!= &this_parent
->d_subdirs
) {
1111 struct list_head
*tmp
= next
;
1112 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1115 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1118 * move only zero ref count dentries to the end
1119 * of the unused list for prune_dcache
1121 if (!dentry
->d_count
) {
1122 dentry_lru_move_tail(dentry
);
1125 dentry_lru_del(dentry
);
1129 * We can return to the caller if we have found some (this
1130 * ensures forward progress). We'll be coming back to find
1133 if (found
&& need_resched()) {
1134 spin_unlock(&dentry
->d_lock
);
1139 * Descend a level if the d_subdirs list is non-empty.
1141 if (!list_empty(&dentry
->d_subdirs
)) {
1142 spin_unlock(&this_parent
->d_lock
);
1143 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1144 this_parent
= dentry
;
1145 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1149 spin_unlock(&dentry
->d_lock
);
1152 * All done at this level ... ascend and resume the search.
1154 if (this_parent
!= parent
) {
1155 struct dentry
*child
= this_parent
;
1156 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1159 next
= child
->d_u
.d_child
.next
;
1163 spin_unlock(&this_parent
->d_lock
);
1164 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1167 write_sequnlock(&rename_lock
);
1174 write_seqlock(&rename_lock
);
1179 * shrink_dcache_parent - prune dcache
1180 * @parent: parent of entries to prune
1182 * Prune the dcache to remove unused children of the parent dentry.
1185 void shrink_dcache_parent(struct dentry
* parent
)
1187 struct super_block
*sb
= parent
->d_sb
;
1190 while ((found
= select_parent(parent
)) != 0)
1191 __shrink_dcache_sb(sb
, found
, 0);
1193 EXPORT_SYMBOL(shrink_dcache_parent
);
1196 * __d_alloc - allocate a dcache entry
1197 * @sb: filesystem it will belong to
1198 * @name: qstr of the name
1200 * Allocates a dentry. It returns %NULL if there is insufficient memory
1201 * available. On a success the dentry is returned. The name passed in is
1202 * copied and the copy passed in may be reused after this call.
1205 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1207 struct dentry
*dentry
;
1210 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1214 if (name
->len
> DNAME_INLINE_LEN
-1) {
1215 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1217 kmem_cache_free(dentry_cache
, dentry
);
1221 dname
= dentry
->d_iname
;
1223 dentry
->d_name
.name
= dname
;
1225 dentry
->d_name
.len
= name
->len
;
1226 dentry
->d_name
.hash
= name
->hash
;
1227 memcpy(dname
, name
->name
, name
->len
);
1228 dname
[name
->len
] = 0;
1230 dentry
->d_count
= 1;
1231 dentry
->d_flags
= 0;
1232 spin_lock_init(&dentry
->d_lock
);
1233 seqcount_init(&dentry
->d_seq
);
1234 dentry
->d_inode
= NULL
;
1235 dentry
->d_parent
= dentry
;
1237 dentry
->d_op
= NULL
;
1238 dentry
->d_fsdata
= NULL
;
1239 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1240 INIT_LIST_HEAD(&dentry
->d_lru
);
1241 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1242 INIT_LIST_HEAD(&dentry
->d_alias
);
1243 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1244 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1246 this_cpu_inc(nr_dentry
);
1252 * d_alloc - allocate a dcache entry
1253 * @parent: parent of entry to allocate
1254 * @name: qstr of the name
1256 * Allocates a dentry. It returns %NULL if there is insufficient memory
1257 * available. On a success the dentry is returned. The name passed in is
1258 * copied and the copy passed in may be reused after this call.
1260 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1262 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1266 spin_lock(&parent
->d_lock
);
1268 * don't need child lock because it is not subject
1269 * to concurrency here
1271 __dget_dlock(parent
);
1272 dentry
->d_parent
= parent
;
1273 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1274 spin_unlock(&parent
->d_lock
);
1278 EXPORT_SYMBOL(d_alloc
);
1280 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1282 struct dentry
*dentry
= __d_alloc(sb
, name
);
1284 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1287 EXPORT_SYMBOL(d_alloc_pseudo
);
1289 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1294 q
.len
= strlen(name
);
1295 q
.hash
= full_name_hash(q
.name
, q
.len
);
1296 return d_alloc(parent
, &q
);
1298 EXPORT_SYMBOL(d_alloc_name
);
1300 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1302 WARN_ON_ONCE(dentry
->d_op
);
1303 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1305 DCACHE_OP_REVALIDATE
|
1306 DCACHE_OP_DELETE
));
1311 dentry
->d_flags
|= DCACHE_OP_HASH
;
1313 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1314 if (op
->d_revalidate
)
1315 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1317 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1319 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1322 EXPORT_SYMBOL(d_set_d_op
);
1324 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1326 spin_lock(&dentry
->d_lock
);
1328 if (unlikely(IS_AUTOMOUNT(inode
)))
1329 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1330 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1332 dentry
->d_inode
= inode
;
1333 dentry_rcuwalk_barrier(dentry
);
1334 spin_unlock(&dentry
->d_lock
);
1335 fsnotify_d_instantiate(dentry
, inode
);
1339 * d_instantiate - fill in inode information for a dentry
1340 * @entry: dentry to complete
1341 * @inode: inode to attach to this dentry
1343 * Fill in inode information in the entry.
1345 * This turns negative dentries into productive full members
1348 * NOTE! This assumes that the inode count has been incremented
1349 * (or otherwise set) by the caller to indicate that it is now
1350 * in use by the dcache.
1353 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1355 BUG_ON(!list_empty(&entry
->d_alias
));
1357 spin_lock(&inode
->i_lock
);
1358 __d_instantiate(entry
, inode
);
1360 spin_unlock(&inode
->i_lock
);
1361 security_d_instantiate(entry
, inode
);
1363 EXPORT_SYMBOL(d_instantiate
);
1366 * d_instantiate_unique - instantiate a non-aliased dentry
1367 * @entry: dentry to instantiate
1368 * @inode: inode to attach to this dentry
1370 * Fill in inode information in the entry. On success, it returns NULL.
1371 * If an unhashed alias of "entry" already exists, then we return the
1372 * aliased dentry instead and drop one reference to inode.
1374 * Note that in order to avoid conflicts with rename() etc, the caller
1375 * had better be holding the parent directory semaphore.
1377 * This also assumes that the inode count has been incremented
1378 * (or otherwise set) by the caller to indicate that it is now
1379 * in use by the dcache.
1381 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1382 struct inode
*inode
)
1384 struct dentry
*alias
;
1385 int len
= entry
->d_name
.len
;
1386 const char *name
= entry
->d_name
.name
;
1387 unsigned int hash
= entry
->d_name
.hash
;
1390 __d_instantiate(entry
, NULL
);
1394 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1395 struct qstr
*qstr
= &alias
->d_name
;
1398 * Don't need alias->d_lock here, because aliases with
1399 * d_parent == entry->d_parent are not subject to name or
1400 * parent changes, because the parent inode i_mutex is held.
1402 if (qstr
->hash
!= hash
)
1404 if (alias
->d_parent
!= entry
->d_parent
)
1406 if (dentry_cmp(qstr
->name
, qstr
->len
, name
, len
))
1412 __d_instantiate(entry
, inode
);
1416 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1418 struct dentry
*result
;
1420 BUG_ON(!list_empty(&entry
->d_alias
));
1423 spin_lock(&inode
->i_lock
);
1424 result
= __d_instantiate_unique(entry
, inode
);
1426 spin_unlock(&inode
->i_lock
);
1429 security_d_instantiate(entry
, inode
);
1433 BUG_ON(!d_unhashed(result
));
1438 EXPORT_SYMBOL(d_instantiate_unique
);
1441 * d_alloc_root - allocate root dentry
1442 * @root_inode: inode to allocate the root for
1444 * Allocate a root ("/") dentry for the inode given. The inode is
1445 * instantiated and returned. %NULL is returned if there is insufficient
1446 * memory or the inode passed is %NULL.
1449 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1451 struct dentry
*res
= NULL
;
1454 static const struct qstr name
= { .name
= "/", .len
= 1 };
1456 res
= __d_alloc(root_inode
->i_sb
, &name
);
1458 d_instantiate(res
, root_inode
);
1462 EXPORT_SYMBOL(d_alloc_root
);
1464 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1466 struct dentry
*alias
;
1468 if (list_empty(&inode
->i_dentry
))
1470 alias
= list_first_entry(&inode
->i_dentry
, struct dentry
, d_alias
);
1475 static struct dentry
* d_find_any_alias(struct inode
*inode
)
1479 spin_lock(&inode
->i_lock
);
1480 de
= __d_find_any_alias(inode
);
1481 spin_unlock(&inode
->i_lock
);
1487 * d_obtain_alias - find or allocate a dentry for a given inode
1488 * @inode: inode to allocate the dentry for
1490 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1491 * similar open by handle operations. The returned dentry may be anonymous,
1492 * or may have a full name (if the inode was already in the cache).
1494 * When called on a directory inode, we must ensure that the inode only ever
1495 * has one dentry. If a dentry is found, that is returned instead of
1496 * allocating a new one.
1498 * On successful return, the reference to the inode has been transferred
1499 * to the dentry. In case of an error the reference on the inode is released.
1500 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1501 * be passed in and will be the error will be propagate to the return value,
1502 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1504 struct dentry
*d_obtain_alias(struct inode
*inode
)
1506 static const struct qstr anonstring
= { .name
= "" };
1511 return ERR_PTR(-ESTALE
);
1513 return ERR_CAST(inode
);
1515 res
= d_find_any_alias(inode
);
1519 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1521 res
= ERR_PTR(-ENOMEM
);
1525 spin_lock(&inode
->i_lock
);
1526 res
= __d_find_any_alias(inode
);
1528 spin_unlock(&inode
->i_lock
);
1533 /* attach a disconnected dentry */
1534 spin_lock(&tmp
->d_lock
);
1535 tmp
->d_inode
= inode
;
1536 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1537 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1538 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1539 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1540 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1541 spin_unlock(&tmp
->d_lock
);
1542 spin_unlock(&inode
->i_lock
);
1543 security_d_instantiate(tmp
, inode
);
1548 if (res
&& !IS_ERR(res
))
1549 security_d_instantiate(res
, inode
);
1553 EXPORT_SYMBOL(d_obtain_alias
);
1556 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1557 * @inode: the inode which may have a disconnected dentry
1558 * @dentry: a negative dentry which we want to point to the inode.
1560 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1561 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1562 * and return it, else simply d_add the inode to the dentry and return NULL.
1564 * This is needed in the lookup routine of any filesystem that is exportable
1565 * (via knfsd) so that we can build dcache paths to directories effectively.
1567 * If a dentry was found and moved, then it is returned. Otherwise NULL
1568 * is returned. This matches the expected return value of ->lookup.
1571 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1573 struct dentry
*new = NULL
;
1576 return ERR_CAST(inode
);
1578 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1579 spin_lock(&inode
->i_lock
);
1580 new = __d_find_alias(inode
, 1);
1582 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1583 spin_unlock(&inode
->i_lock
);
1584 security_d_instantiate(new, inode
);
1585 d_move(new, dentry
);
1588 /* already taking inode->i_lock, so d_add() by hand */
1589 __d_instantiate(dentry
, inode
);
1590 spin_unlock(&inode
->i_lock
);
1591 security_d_instantiate(dentry
, inode
);
1595 d_add(dentry
, inode
);
1598 EXPORT_SYMBOL(d_splice_alias
);
1601 * d_add_ci - lookup or allocate new dentry with case-exact name
1602 * @inode: the inode case-insensitive lookup has found
1603 * @dentry: the negative dentry that was passed to the parent's lookup func
1604 * @name: the case-exact name to be associated with the returned dentry
1606 * This is to avoid filling the dcache with case-insensitive names to the
1607 * same inode, only the actual correct case is stored in the dcache for
1608 * case-insensitive filesystems.
1610 * For a case-insensitive lookup match and if the the case-exact dentry
1611 * already exists in in the dcache, use it and return it.
1613 * If no entry exists with the exact case name, allocate new dentry with
1614 * the exact case, and return the spliced entry.
1616 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1620 struct dentry
*found
;
1624 * First check if a dentry matching the name already exists,
1625 * if not go ahead and create it now.
1627 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1629 new = d_alloc(dentry
->d_parent
, name
);
1635 found
= d_splice_alias(inode
, new);
1644 * If a matching dentry exists, and it's not negative use it.
1646 * Decrement the reference count to balance the iget() done
1649 if (found
->d_inode
) {
1650 if (unlikely(found
->d_inode
!= inode
)) {
1651 /* This can't happen because bad inodes are unhashed. */
1652 BUG_ON(!is_bad_inode(inode
));
1653 BUG_ON(!is_bad_inode(found
->d_inode
));
1660 * We are going to instantiate this dentry, unhash it and clear the
1661 * lookup flag so we can do that.
1663 if (unlikely(d_need_lookup(found
)))
1664 d_clear_need_lookup(found
);
1667 * Negative dentry: instantiate it unless the inode is a directory and
1668 * already has a dentry.
1670 new = d_splice_alias(inode
, found
);
1679 return ERR_PTR(error
);
1681 EXPORT_SYMBOL(d_add_ci
);
1684 * __d_lookup_rcu - search for a dentry (racy, store-free)
1685 * @parent: parent dentry
1686 * @name: qstr of name we wish to find
1687 * @seq: returns d_seq value at the point where the dentry was found
1688 * @inode: returns dentry->d_inode when the inode was found valid.
1689 * Returns: dentry, or NULL
1691 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1692 * resolution (store-free path walking) design described in
1693 * Documentation/filesystems/path-lookup.txt.
1695 * This is not to be used outside core vfs.
1697 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1698 * held, and rcu_read_lock held. The returned dentry must not be stored into
1699 * without taking d_lock and checking d_seq sequence count against @seq
1702 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1705 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1706 * the returned dentry, so long as its parent's seqlock is checked after the
1707 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1708 * is formed, giving integrity down the path walk.
1710 struct dentry
*__d_lookup_rcu(struct dentry
*parent
, struct qstr
*name
,
1711 unsigned *seq
, struct inode
**inode
)
1713 unsigned int len
= name
->len
;
1714 unsigned int hash
= name
->hash
;
1715 const unsigned char *str
= name
->name
;
1716 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1717 struct hlist_bl_node
*node
;
1718 struct dentry
*dentry
;
1721 * Note: There is significant duplication with __d_lookup_rcu which is
1722 * required to prevent single threaded performance regressions
1723 * especially on architectures where smp_rmb (in seqcounts) are costly.
1724 * Keep the two functions in sync.
1728 * The hash list is protected using RCU.
1730 * Carefully use d_seq when comparing a candidate dentry, to avoid
1731 * races with d_move().
1733 * It is possible that concurrent renames can mess up our list
1734 * walk here and result in missing our dentry, resulting in the
1735 * false-negative result. d_lookup() protects against concurrent
1736 * renames using rename_lock seqlock.
1738 * See Documentation/filesystems/path-lookup.txt for more details.
1740 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1745 if (dentry
->d_name
.hash
!= hash
)
1749 *seq
= read_seqcount_begin(&dentry
->d_seq
);
1750 if (dentry
->d_parent
!= parent
)
1752 if (d_unhashed(dentry
))
1754 tlen
= dentry
->d_name
.len
;
1755 tname
= dentry
->d_name
.name
;
1756 i
= dentry
->d_inode
;
1759 * This seqcount check is required to ensure name and
1760 * len are loaded atomically, so as not to walk off the
1761 * edge of memory when walking. If we could load this
1762 * atomically some other way, we could drop this check.
1764 if (read_seqcount_retry(&dentry
->d_seq
, *seq
))
1766 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1767 if (parent
->d_op
->d_compare(parent
, *inode
,
1772 if (dentry_cmp(tname
, tlen
, str
, len
))
1776 * No extra seqcount check is required after the name
1777 * compare. The caller must perform a seqcount check in
1778 * order to do anything useful with the returned dentry
1788 * d_lookup - search for a dentry
1789 * @parent: parent dentry
1790 * @name: qstr of name we wish to find
1791 * Returns: dentry, or NULL
1793 * d_lookup searches the children of the parent dentry for the name in
1794 * question. If the dentry is found its reference count is incremented and the
1795 * dentry is returned. The caller must use dput to free the entry when it has
1796 * finished using it. %NULL is returned if the dentry does not exist.
1798 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1800 struct dentry
*dentry
;
1804 seq
= read_seqbegin(&rename_lock
);
1805 dentry
= __d_lookup(parent
, name
);
1808 } while (read_seqretry(&rename_lock
, seq
));
1811 EXPORT_SYMBOL(d_lookup
);
1814 * __d_lookup - search for a dentry (racy)
1815 * @parent: parent dentry
1816 * @name: qstr of name we wish to find
1817 * Returns: dentry, or NULL
1819 * __d_lookup is like d_lookup, however it may (rarely) return a
1820 * false-negative result due to unrelated rename activity.
1822 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1823 * however it must be used carefully, eg. with a following d_lookup in
1824 * the case of failure.
1826 * __d_lookup callers must be commented.
1828 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1830 unsigned int len
= name
->len
;
1831 unsigned int hash
= name
->hash
;
1832 const unsigned char *str
= name
->name
;
1833 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1834 struct hlist_bl_node
*node
;
1835 struct dentry
*found
= NULL
;
1836 struct dentry
*dentry
;
1839 * Note: There is significant duplication with __d_lookup_rcu which is
1840 * required to prevent single threaded performance regressions
1841 * especially on architectures where smp_rmb (in seqcounts) are costly.
1842 * Keep the two functions in sync.
1846 * The hash list is protected using RCU.
1848 * Take d_lock when comparing a candidate dentry, to avoid races
1851 * It is possible that concurrent renames can mess up our list
1852 * walk here and result in missing our dentry, resulting in the
1853 * false-negative result. d_lookup() protects against concurrent
1854 * renames using rename_lock seqlock.
1856 * See Documentation/filesystems/path-lookup.txt for more details.
1860 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1864 if (dentry
->d_name
.hash
!= hash
)
1867 spin_lock(&dentry
->d_lock
);
1868 if (dentry
->d_parent
!= parent
)
1870 if (d_unhashed(dentry
))
1874 * It is safe to compare names since d_move() cannot
1875 * change the qstr (protected by d_lock).
1877 tlen
= dentry
->d_name
.len
;
1878 tname
= dentry
->d_name
.name
;
1879 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1880 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1881 dentry
, dentry
->d_inode
,
1885 if (dentry_cmp(tname
, tlen
, str
, len
))
1891 spin_unlock(&dentry
->d_lock
);
1894 spin_unlock(&dentry
->d_lock
);
1902 * d_hash_and_lookup - hash the qstr then search for a dentry
1903 * @dir: Directory to search in
1904 * @name: qstr of name we wish to find
1906 * On hash failure or on lookup failure NULL is returned.
1908 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1910 struct dentry
*dentry
= NULL
;
1913 * Check for a fs-specific hash function. Note that we must
1914 * calculate the standard hash first, as the d_op->d_hash()
1915 * routine may choose to leave the hash value unchanged.
1917 name
->hash
= full_name_hash(name
->name
, name
->len
);
1918 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1919 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1922 dentry
= d_lookup(dir
, name
);
1928 * d_validate - verify dentry provided from insecure source (deprecated)
1929 * @dentry: The dentry alleged to be valid child of @dparent
1930 * @dparent: The parent dentry (known to be valid)
1932 * An insecure source has sent us a dentry, here we verify it and dget() it.
1933 * This is used by ncpfs in its readdir implementation.
1934 * Zero is returned in the dentry is invalid.
1936 * This function is slow for big directories, and deprecated, do not use it.
1938 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1940 struct dentry
*child
;
1942 spin_lock(&dparent
->d_lock
);
1943 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1944 if (dentry
== child
) {
1945 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1946 __dget_dlock(dentry
);
1947 spin_unlock(&dentry
->d_lock
);
1948 spin_unlock(&dparent
->d_lock
);
1952 spin_unlock(&dparent
->d_lock
);
1956 EXPORT_SYMBOL(d_validate
);
1959 * When a file is deleted, we have two options:
1960 * - turn this dentry into a negative dentry
1961 * - unhash this dentry and free it.
1963 * Usually, we want to just turn this into
1964 * a negative dentry, but if anybody else is
1965 * currently using the dentry or the inode
1966 * we can't do that and we fall back on removing
1967 * it from the hash queues and waiting for
1968 * it to be deleted later when it has no users
1972 * d_delete - delete a dentry
1973 * @dentry: The dentry to delete
1975 * Turn the dentry into a negative dentry if possible, otherwise
1976 * remove it from the hash queues so it can be deleted later
1979 void d_delete(struct dentry
* dentry
)
1981 struct inode
*inode
;
1984 * Are we the only user?
1987 spin_lock(&dentry
->d_lock
);
1988 inode
= dentry
->d_inode
;
1989 isdir
= S_ISDIR(inode
->i_mode
);
1990 if (dentry
->d_count
== 1) {
1991 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
1992 spin_unlock(&dentry
->d_lock
);
1996 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1997 dentry_unlink_inode(dentry
);
1998 fsnotify_nameremove(dentry
, isdir
);
2002 if (!d_unhashed(dentry
))
2005 spin_unlock(&dentry
->d_lock
);
2007 fsnotify_nameremove(dentry
, isdir
);
2009 EXPORT_SYMBOL(d_delete
);
2011 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2013 BUG_ON(!d_unhashed(entry
));
2015 entry
->d_flags
|= DCACHE_RCUACCESS
;
2016 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2020 static void _d_rehash(struct dentry
* entry
)
2022 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2026 * d_rehash - add an entry back to the hash
2027 * @entry: dentry to add to the hash
2029 * Adds a dentry to the hash according to its name.
2032 void d_rehash(struct dentry
* entry
)
2034 spin_lock(&entry
->d_lock
);
2036 spin_unlock(&entry
->d_lock
);
2038 EXPORT_SYMBOL(d_rehash
);
2041 * dentry_update_name_case - update case insensitive dentry with a new name
2042 * @dentry: dentry to be updated
2045 * Update a case insensitive dentry with new case of name.
2047 * dentry must have been returned by d_lookup with name @name. Old and new
2048 * name lengths must match (ie. no d_compare which allows mismatched name
2051 * Parent inode i_mutex must be held over d_lookup and into this call (to
2052 * keep renames and concurrent inserts, and readdir(2) away).
2054 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2056 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2057 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2059 spin_lock(&dentry
->d_lock
);
2060 write_seqcount_begin(&dentry
->d_seq
);
2061 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2062 write_seqcount_end(&dentry
->d_seq
);
2063 spin_unlock(&dentry
->d_lock
);
2065 EXPORT_SYMBOL(dentry_update_name_case
);
2067 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2069 if (dname_external(target
)) {
2070 if (dname_external(dentry
)) {
2072 * Both external: swap the pointers
2074 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2077 * dentry:internal, target:external. Steal target's
2078 * storage and make target internal.
2080 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2081 dentry
->d_name
.len
+ 1);
2082 dentry
->d_name
.name
= target
->d_name
.name
;
2083 target
->d_name
.name
= target
->d_iname
;
2086 if (dname_external(dentry
)) {
2088 * dentry:external, target:internal. Give dentry's
2089 * storage to target and make dentry internal
2091 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2092 target
->d_name
.len
+ 1);
2093 target
->d_name
.name
= dentry
->d_name
.name
;
2094 dentry
->d_name
.name
= dentry
->d_iname
;
2097 * Both are internal. Just copy target to dentry
2099 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2100 target
->d_name
.len
+ 1);
2101 dentry
->d_name
.len
= target
->d_name
.len
;
2105 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2108 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2111 * XXXX: do we really need to take target->d_lock?
2113 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2114 spin_lock(&target
->d_parent
->d_lock
);
2116 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2117 spin_lock(&dentry
->d_parent
->d_lock
);
2118 spin_lock_nested(&target
->d_parent
->d_lock
,
2119 DENTRY_D_LOCK_NESTED
);
2121 spin_lock(&target
->d_parent
->d_lock
);
2122 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2123 DENTRY_D_LOCK_NESTED
);
2126 if (target
< dentry
) {
2127 spin_lock_nested(&target
->d_lock
, 2);
2128 spin_lock_nested(&dentry
->d_lock
, 3);
2130 spin_lock_nested(&dentry
->d_lock
, 2);
2131 spin_lock_nested(&target
->d_lock
, 3);
2135 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2136 struct dentry
*target
)
2138 if (target
->d_parent
!= dentry
->d_parent
)
2139 spin_unlock(&dentry
->d_parent
->d_lock
);
2140 if (target
->d_parent
!= target
)
2141 spin_unlock(&target
->d_parent
->d_lock
);
2145 * When switching names, the actual string doesn't strictly have to
2146 * be preserved in the target - because we're dropping the target
2147 * anyway. As such, we can just do a simple memcpy() to copy over
2148 * the new name before we switch.
2150 * Note that we have to be a lot more careful about getting the hash
2151 * switched - we have to switch the hash value properly even if it
2152 * then no longer matches the actual (corrupted) string of the target.
2153 * The hash value has to match the hash queue that the dentry is on..
2156 * __d_move - move a dentry
2157 * @dentry: entry to move
2158 * @target: new dentry
2160 * Update the dcache to reflect the move of a file name. Negative
2161 * dcache entries should not be moved in this way. Caller must hold
2162 * rename_lock, the i_mutex of the source and target directories,
2163 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2165 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2167 if (!dentry
->d_inode
)
2168 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2170 BUG_ON(d_ancestor(dentry
, target
));
2171 BUG_ON(d_ancestor(target
, dentry
));
2173 dentry_lock_for_move(dentry
, target
);
2175 write_seqcount_begin(&dentry
->d_seq
);
2176 write_seqcount_begin(&target
->d_seq
);
2178 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2181 * Move the dentry to the target hash queue. Don't bother checking
2182 * for the same hash queue because of how unlikely it is.
2185 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2187 /* Unhash the target: dput() will then get rid of it */
2190 list_del(&dentry
->d_u
.d_child
);
2191 list_del(&target
->d_u
.d_child
);
2193 /* Switch the names.. */
2194 switch_names(dentry
, target
);
2195 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2197 /* ... and switch the parents */
2198 if (IS_ROOT(dentry
)) {
2199 dentry
->d_parent
= target
->d_parent
;
2200 target
->d_parent
= target
;
2201 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2203 swap(dentry
->d_parent
, target
->d_parent
);
2205 /* And add them back to the (new) parent lists */
2206 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2209 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2211 write_seqcount_end(&target
->d_seq
);
2212 write_seqcount_end(&dentry
->d_seq
);
2214 dentry_unlock_parents_for_move(dentry
, target
);
2215 spin_unlock(&target
->d_lock
);
2216 fsnotify_d_move(dentry
);
2217 spin_unlock(&dentry
->d_lock
);
2221 * d_move - move a dentry
2222 * @dentry: entry to move
2223 * @target: new dentry
2225 * Update the dcache to reflect the move of a file name. Negative
2226 * dcache entries should not be moved in this way. See the locking
2227 * requirements for __d_move.
2229 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2231 write_seqlock(&rename_lock
);
2232 __d_move(dentry
, target
);
2233 write_sequnlock(&rename_lock
);
2235 EXPORT_SYMBOL(d_move
);
2238 * d_ancestor - search for an ancestor
2239 * @p1: ancestor dentry
2242 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2243 * an ancestor of p2, else NULL.
2245 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2249 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2250 if (p
->d_parent
== p1
)
2257 * This helper attempts to cope with remotely renamed directories
2259 * It assumes that the caller is already holding
2260 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2262 * Note: If ever the locking in lock_rename() changes, then please
2263 * remember to update this too...
2265 static struct dentry
*__d_unalias(struct inode
*inode
,
2266 struct dentry
*dentry
, struct dentry
*alias
)
2268 struct mutex
*m1
= NULL
, *m2
= NULL
;
2271 /* If alias and dentry share a parent, then no extra locks required */
2272 if (alias
->d_parent
== dentry
->d_parent
)
2275 /* See lock_rename() */
2276 ret
= ERR_PTR(-EBUSY
);
2277 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2279 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2280 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2282 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2284 __d_move(alias
, dentry
);
2287 spin_unlock(&inode
->i_lock
);
2296 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2297 * named dentry in place of the dentry to be replaced.
2298 * returns with anon->d_lock held!
2300 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2302 struct dentry
*dparent
, *aparent
;
2304 dentry_lock_for_move(anon
, dentry
);
2306 write_seqcount_begin(&dentry
->d_seq
);
2307 write_seqcount_begin(&anon
->d_seq
);
2309 dparent
= dentry
->d_parent
;
2310 aparent
= anon
->d_parent
;
2312 switch_names(dentry
, anon
);
2313 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2315 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2316 list_del(&dentry
->d_u
.d_child
);
2317 if (!IS_ROOT(dentry
))
2318 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2320 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2322 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2323 list_del(&anon
->d_u
.d_child
);
2325 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2327 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2329 write_seqcount_end(&dentry
->d_seq
);
2330 write_seqcount_end(&anon
->d_seq
);
2332 dentry_unlock_parents_for_move(anon
, dentry
);
2333 spin_unlock(&dentry
->d_lock
);
2335 /* anon->d_lock still locked, returns locked */
2336 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2340 * d_materialise_unique - introduce an inode into the tree
2341 * @dentry: candidate dentry
2342 * @inode: inode to bind to the dentry, to which aliases may be attached
2344 * Introduces an dentry into the tree, substituting an extant disconnected
2345 * root directory alias in its place if there is one. Caller must hold the
2346 * i_mutex of the parent directory.
2348 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2350 struct dentry
*actual
;
2352 BUG_ON(!d_unhashed(dentry
));
2356 __d_instantiate(dentry
, NULL
);
2361 spin_lock(&inode
->i_lock
);
2363 if (S_ISDIR(inode
->i_mode
)) {
2364 struct dentry
*alias
;
2366 /* Does an aliased dentry already exist? */
2367 alias
= __d_find_alias(inode
, 0);
2370 write_seqlock(&rename_lock
);
2372 if (d_ancestor(alias
, dentry
)) {
2373 /* Check for loops */
2374 actual
= ERR_PTR(-ELOOP
);
2375 } else if (IS_ROOT(alias
)) {
2376 /* Is this an anonymous mountpoint that we
2377 * could splice into our tree? */
2378 __d_materialise_dentry(dentry
, alias
);
2379 write_sequnlock(&rename_lock
);
2383 /* Nope, but we must(!) avoid directory
2385 actual
= __d_unalias(inode
, dentry
, alias
);
2387 write_sequnlock(&rename_lock
);
2388 if (IS_ERR(actual
)) {
2389 if (PTR_ERR(actual
) == -ELOOP
)
2390 pr_warn_ratelimited(
2391 "VFS: Lookup of '%s' in %s %s"
2392 " would have caused loop\n",
2393 dentry
->d_name
.name
,
2394 inode
->i_sb
->s_type
->name
,
2402 /* Add a unique reference */
2403 actual
= __d_instantiate_unique(dentry
, inode
);
2407 BUG_ON(!d_unhashed(actual
));
2409 spin_lock(&actual
->d_lock
);
2412 spin_unlock(&actual
->d_lock
);
2413 spin_unlock(&inode
->i_lock
);
2415 if (actual
== dentry
) {
2416 security_d_instantiate(dentry
, inode
);
2423 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2425 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2429 return -ENAMETOOLONG
;
2431 memcpy(*buffer
, str
, namelen
);
2435 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2437 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2441 * prepend_path - Prepend path string to a buffer
2442 * @path: the dentry/vfsmount to report
2443 * @root: root vfsmnt/dentry
2444 * @buffer: pointer to the end of the buffer
2445 * @buflen: pointer to buffer length
2447 * Caller holds the rename_lock.
2449 static int prepend_path(const struct path
*path
,
2450 const struct path
*root
,
2451 char **buffer
, int *buflen
)
2453 struct dentry
*dentry
= path
->dentry
;
2454 struct vfsmount
*vfsmnt
= path
->mnt
;
2458 br_read_lock(vfsmount_lock
);
2459 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2460 struct dentry
* parent
;
2462 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2464 if (!mnt_has_parent(vfsmnt
))
2466 dentry
= vfsmnt
->mnt_mountpoint
;
2467 vfsmnt
= vfsmnt
->mnt_parent
;
2470 parent
= dentry
->d_parent
;
2472 spin_lock(&dentry
->d_lock
);
2473 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2474 spin_unlock(&dentry
->d_lock
);
2476 error
= prepend(buffer
, buflen
, "/", 1);
2484 if (!error
&& !slash
)
2485 error
= prepend(buffer
, buflen
, "/", 1);
2488 br_read_unlock(vfsmount_lock
);
2493 * Filesystems needing to implement special "root names"
2494 * should do so with ->d_dname()
2496 if (IS_ROOT(dentry
) &&
2497 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2498 WARN(1, "Root dentry has weird name <%.*s>\n",
2499 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2502 error
= prepend(buffer
, buflen
, "/", 1);
2504 error
= vfsmnt
->mnt_ns
? 1 : 2;
2509 * __d_path - return the path of a dentry
2510 * @path: the dentry/vfsmount to report
2511 * @root: root vfsmnt/dentry
2512 * @buf: buffer to return value in
2513 * @buflen: buffer length
2515 * Convert a dentry into an ASCII path name.
2517 * Returns a pointer into the buffer or an error code if the
2518 * path was too long.
2520 * "buflen" should be positive.
2522 * If the path is not reachable from the supplied root, return %NULL.
2524 char *__d_path(const struct path
*path
,
2525 const struct path
*root
,
2526 char *buf
, int buflen
)
2528 char *res
= buf
+ buflen
;
2531 prepend(&res
, &buflen
, "\0", 1);
2532 write_seqlock(&rename_lock
);
2533 error
= prepend_path(path
, root
, &res
, &buflen
);
2534 write_sequnlock(&rename_lock
);
2537 return ERR_PTR(error
);
2543 char *d_absolute_path(const struct path
*path
,
2544 char *buf
, int buflen
)
2546 struct path root
= {};
2547 char *res
= buf
+ buflen
;
2550 prepend(&res
, &buflen
, "\0", 1);
2551 write_seqlock(&rename_lock
);
2552 error
= prepend_path(path
, &root
, &res
, &buflen
);
2553 write_sequnlock(&rename_lock
);
2558 return ERR_PTR(error
);
2563 * same as __d_path but appends "(deleted)" for unlinked files.
2565 static int path_with_deleted(const struct path
*path
,
2566 const struct path
*root
,
2567 char **buf
, int *buflen
)
2569 prepend(buf
, buflen
, "\0", 1);
2570 if (d_unlinked(path
->dentry
)) {
2571 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2576 return prepend_path(path
, root
, buf
, buflen
);
2579 static int prepend_unreachable(char **buffer
, int *buflen
)
2581 return prepend(buffer
, buflen
, "(unreachable)", 13);
2585 * d_path - return the path of a dentry
2586 * @path: path to report
2587 * @buf: buffer to return value in
2588 * @buflen: buffer length
2590 * Convert a dentry into an ASCII path name. If the entry has been deleted
2591 * the string " (deleted)" is appended. Note that this is ambiguous.
2593 * Returns a pointer into the buffer or an error code if the path was
2594 * too long. Note: Callers should use the returned pointer, not the passed
2595 * in buffer, to use the name! The implementation often starts at an offset
2596 * into the buffer, and may leave 0 bytes at the start.
2598 * "buflen" should be positive.
2600 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2602 char *res
= buf
+ buflen
;
2607 * We have various synthetic filesystems that never get mounted. On
2608 * these filesystems dentries are never used for lookup purposes, and
2609 * thus don't need to be hashed. They also don't need a name until a
2610 * user wants to identify the object in /proc/pid/fd/. The little hack
2611 * below allows us to generate a name for these objects on demand:
2613 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2614 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2616 get_fs_root(current
->fs
, &root
);
2617 write_seqlock(&rename_lock
);
2618 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2620 res
= ERR_PTR(error
);
2621 write_sequnlock(&rename_lock
);
2625 EXPORT_SYMBOL(d_path
);
2628 * d_path_with_unreachable - return the path of a dentry
2629 * @path: path to report
2630 * @buf: buffer to return value in
2631 * @buflen: buffer length
2633 * The difference from d_path() is that this prepends "(unreachable)"
2634 * to paths which are unreachable from the current process' root.
2636 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2638 char *res
= buf
+ buflen
;
2642 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2643 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2645 get_fs_root(current
->fs
, &root
);
2646 write_seqlock(&rename_lock
);
2647 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2649 error
= prepend_unreachable(&res
, &buflen
);
2650 write_sequnlock(&rename_lock
);
2653 res
= ERR_PTR(error
);
2659 * Helper function for dentry_operations.d_dname() members
2661 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2662 const char *fmt
, ...)
2668 va_start(args
, fmt
);
2669 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2672 if (sz
> sizeof(temp
) || sz
> buflen
)
2673 return ERR_PTR(-ENAMETOOLONG
);
2675 buffer
+= buflen
- sz
;
2676 return memcpy(buffer
, temp
, sz
);
2680 * Write full pathname from the root of the filesystem into the buffer.
2682 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2684 char *end
= buf
+ buflen
;
2687 prepend(&end
, &buflen
, "\0", 1);
2694 while (!IS_ROOT(dentry
)) {
2695 struct dentry
*parent
= dentry
->d_parent
;
2699 spin_lock(&dentry
->d_lock
);
2700 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2701 spin_unlock(&dentry
->d_lock
);
2702 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2710 return ERR_PTR(-ENAMETOOLONG
);
2713 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2717 write_seqlock(&rename_lock
);
2718 retval
= __dentry_path(dentry
, buf
, buflen
);
2719 write_sequnlock(&rename_lock
);
2723 EXPORT_SYMBOL(dentry_path_raw
);
2725 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2730 write_seqlock(&rename_lock
);
2731 if (d_unlinked(dentry
)) {
2733 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2737 retval
= __dentry_path(dentry
, buf
, buflen
);
2738 write_sequnlock(&rename_lock
);
2739 if (!IS_ERR(retval
) && p
)
2740 *p
= '/'; /* restore '/' overriden with '\0' */
2743 return ERR_PTR(-ENAMETOOLONG
);
2747 * NOTE! The user-level library version returns a
2748 * character pointer. The kernel system call just
2749 * returns the length of the buffer filled (which
2750 * includes the ending '\0' character), or a negative
2751 * error value. So libc would do something like
2753 * char *getcwd(char * buf, size_t size)
2757 * retval = sys_getcwd(buf, size);
2764 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2767 struct path pwd
, root
;
2768 char *page
= (char *) __get_free_page(GFP_USER
);
2773 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2776 write_seqlock(&rename_lock
);
2777 if (!d_unlinked(pwd
.dentry
)) {
2779 char *cwd
= page
+ PAGE_SIZE
;
2780 int buflen
= PAGE_SIZE
;
2782 prepend(&cwd
, &buflen
, "\0", 1);
2783 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2784 write_sequnlock(&rename_lock
);
2789 /* Unreachable from current root */
2791 error
= prepend_unreachable(&cwd
, &buflen
);
2797 len
= PAGE_SIZE
+ page
- cwd
;
2800 if (copy_to_user(buf
, cwd
, len
))
2804 write_sequnlock(&rename_lock
);
2810 free_page((unsigned long) page
);
2815 * Test whether new_dentry is a subdirectory of old_dentry.
2817 * Trivially implemented using the dcache structure
2821 * is_subdir - is new dentry a subdirectory of old_dentry
2822 * @new_dentry: new dentry
2823 * @old_dentry: old dentry
2825 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2826 * Returns 0 otherwise.
2827 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2830 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2835 if (new_dentry
== old_dentry
)
2839 /* for restarting inner loop in case of seq retry */
2840 seq
= read_seqbegin(&rename_lock
);
2842 * Need rcu_readlock to protect against the d_parent trashing
2846 if (d_ancestor(old_dentry
, new_dentry
))
2851 } while (read_seqretry(&rename_lock
, seq
));
2856 void d_genocide(struct dentry
*root
)
2858 struct dentry
*this_parent
;
2859 struct list_head
*next
;
2863 seq
= read_seqbegin(&rename_lock
);
2866 spin_lock(&this_parent
->d_lock
);
2868 next
= this_parent
->d_subdirs
.next
;
2870 while (next
!= &this_parent
->d_subdirs
) {
2871 struct list_head
*tmp
= next
;
2872 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2875 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2876 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2877 spin_unlock(&dentry
->d_lock
);
2880 if (!list_empty(&dentry
->d_subdirs
)) {
2881 spin_unlock(&this_parent
->d_lock
);
2882 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2883 this_parent
= dentry
;
2884 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2887 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2888 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2891 spin_unlock(&dentry
->d_lock
);
2893 if (this_parent
!= root
) {
2894 struct dentry
*child
= this_parent
;
2895 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2896 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2897 this_parent
->d_count
--;
2899 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2902 next
= child
->d_u
.d_child
.next
;
2905 spin_unlock(&this_parent
->d_lock
);
2906 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2909 write_sequnlock(&rename_lock
);
2914 write_seqlock(&rename_lock
);
2919 * find_inode_number - check for dentry with name
2920 * @dir: directory to check
2921 * @name: Name to find.
2923 * Check whether a dentry already exists for the given name,
2924 * and return the inode number if it has an inode. Otherwise
2927 * This routine is used to post-process directory listings for
2928 * filesystems using synthetic inode numbers, and is necessary
2929 * to keep getcwd() working.
2932 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2934 struct dentry
* dentry
;
2937 dentry
= d_hash_and_lookup(dir
, name
);
2939 if (dentry
->d_inode
)
2940 ino
= dentry
->d_inode
->i_ino
;
2945 EXPORT_SYMBOL(find_inode_number
);
2947 static __initdata
unsigned long dhash_entries
;
2948 static int __init
set_dhash_entries(char *str
)
2952 dhash_entries
= simple_strtoul(str
, &str
, 0);
2955 __setup("dhash_entries=", set_dhash_entries
);
2957 static void __init
dcache_init_early(void)
2961 /* If hashes are distributed across NUMA nodes, defer
2962 * hash allocation until vmalloc space is available.
2968 alloc_large_system_hash("Dentry cache",
2969 sizeof(struct hlist_bl_head
),
2977 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2978 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
2981 static void __init
dcache_init(void)
2986 * A constructor could be added for stable state like the lists,
2987 * but it is probably not worth it because of the cache nature
2990 dentry_cache
= KMEM_CACHE(dentry
,
2991 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2993 /* Hash may have been set up in dcache_init_early */
2998 alloc_large_system_hash("Dentry cache",
2999 sizeof(struct hlist_bl_head
),
3007 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
3008 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3011 /* SLAB cache for __getname() consumers */
3012 struct kmem_cache
*names_cachep __read_mostly
;
3013 EXPORT_SYMBOL(names_cachep
);
3015 EXPORT_SYMBOL(d_genocide
);
3017 void __init
vfs_caches_init_early(void)
3019 dcache_init_early();
3023 void __init
vfs_caches_init(unsigned long mempages
)
3025 unsigned long reserve
;
3027 /* Base hash sizes on available memory, with a reserve equal to
3028 150% of current kernel size */
3030 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3031 mempages
-= reserve
;
3033 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3034 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3038 files_init(mempages
);